Fix: put session list lock around the app registration
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/uatomic.h>
36 #include <unistd.h>
37 #include <config.h>
38
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "kernel.h"
55 #include "kernel-consumer.h"
56 #include "modprobe.h"
57 #include "shm.h"
58 #include "ust-ctl.h"
59 #include "ust-consumer.h"
60 #include "utils.h"
61 #include "fd-limit.h"
62 #include "health.h"
63 #include "testpoint.h"
64
65 #define CONSUMERD_FILE "lttng-consumerd"
66
67 /* Const values */
68 const char default_home_dir[] = DEFAULT_HOME_DIR;
69 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
70 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
71 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
72
73 const char *progname;
74 const char *opt_tracing_group;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread;
153 static pthread_t reg_apps_thread;
154 static pthread_t client_thread;
155 static pthread_t kernel_thread;
156 static pthread_t dispatch_thread;
157 static pthread_t health_thread;
158
159 /*
160 * UST registration command queue. This queue is tied with a futex and uses a N
161 * wakers / 1 waiter implemented and detailed in futex.c/.h
162 *
163 * The thread_manage_apps and thread_dispatch_ust_registration interact with
164 * this queue and the wait/wake scheme.
165 */
166 static struct ust_cmd_queue ust_cmd_queue;
167
168 /*
169 * Pointer initialized before thread creation.
170 *
171 * This points to the tracing session list containing the session count and a
172 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
173 * MUST NOT be taken if you call a public function in session.c.
174 *
175 * The lock is nested inside the structure: session_list_ptr->lock. Please use
176 * session_lock_list and session_unlock_list for lock acquisition.
177 */
178 static struct ltt_session_list *session_list_ptr;
179
180 int ust_consumerd64_fd = -1;
181 int ust_consumerd32_fd = -1;
182
183 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
184 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
185 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
186 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
187
188 static const char *module_proc_lttng = "/proc/lttng";
189
190 /*
191 * Consumer daemon state which is changed when spawning it, killing it or in
192 * case of a fatal error.
193 */
194 enum consumerd_state {
195 CONSUMER_STARTED = 1,
196 CONSUMER_STOPPED = 2,
197 CONSUMER_ERROR = 3,
198 };
199
200 /*
201 * This consumer daemon state is used to validate if a client command will be
202 * able to reach the consumer. If not, the client is informed. For instance,
203 * doing a "lttng start" when the consumer state is set to ERROR will return an
204 * error to the client.
205 *
206 * The following example shows a possible race condition of this scheme:
207 *
208 * consumer thread error happens
209 * client cmd arrives
210 * client cmd checks state -> still OK
211 * consumer thread exit, sets error
212 * client cmd try to talk to consumer
213 * ...
214 *
215 * However, since the consumer is a different daemon, we have no way of making
216 * sure the command will reach it safely even with this state flag. This is why
217 * we consider that up to the state validation during command processing, the
218 * command is safe. After that, we can not guarantee the correctness of the
219 * client request vis-a-vis the consumer.
220 */
221 static enum consumerd_state ust_consumerd_state;
222 static enum consumerd_state kernel_consumerd_state;
223
224 /*
225 * Socket timeout for receiving and sending in seconds.
226 */
227 static int app_socket_timeout;
228
229 static
230 void setup_consumerd_path(void)
231 {
232 const char *bin, *libdir;
233
234 /*
235 * Allow INSTALL_BIN_PATH to be used as a target path for the
236 * native architecture size consumer if CONFIG_CONSUMER*_PATH
237 * has not been defined.
238 */
239 #if (CAA_BITS_PER_LONG == 32)
240 if (!consumerd32_bin[0]) {
241 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
242 }
243 if (!consumerd32_libdir[0]) {
244 consumerd32_libdir = INSTALL_LIB_PATH;
245 }
246 #elif (CAA_BITS_PER_LONG == 64)
247 if (!consumerd64_bin[0]) {
248 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd64_libdir[0]) {
251 consumerd64_libdir = INSTALL_LIB_PATH;
252 }
253 #else
254 #error "Unknown bitness"
255 #endif
256
257 /*
258 * runtime env. var. overrides the build default.
259 */
260 bin = getenv("LTTNG_CONSUMERD32_BIN");
261 if (bin) {
262 consumerd32_bin = bin;
263 }
264 bin = getenv("LTTNG_CONSUMERD64_BIN");
265 if (bin) {
266 consumerd64_bin = bin;
267 }
268 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
269 if (libdir) {
270 consumerd32_libdir = libdir;
271 }
272 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
273 if (libdir) {
274 consumerd64_libdir = libdir;
275 }
276 }
277
278 /*
279 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
280 */
281 static int create_thread_poll_set(struct lttng_poll_event *events,
282 unsigned int size)
283 {
284 int ret;
285
286 if (events == NULL || size == 0) {
287 ret = -1;
288 goto error;
289 }
290
291 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
292 if (ret < 0) {
293 goto error;
294 }
295
296 /* Add quit pipe */
297 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
298 if (ret < 0) {
299 goto error;
300 }
301
302 return 0;
303
304 error:
305 return ret;
306 }
307
308 /*
309 * Check if the thread quit pipe was triggered.
310 *
311 * Return 1 if it was triggered else 0;
312 */
313 static int check_thread_quit_pipe(int fd, uint32_t events)
314 {
315 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
316 return 1;
317 }
318
319 return 0;
320 }
321
322 /*
323 * Return group ID of the tracing group or -1 if not found.
324 */
325 static gid_t allowed_group(void)
326 {
327 struct group *grp;
328
329 if (opt_tracing_group) {
330 grp = getgrnam(opt_tracing_group);
331 } else {
332 grp = getgrnam(default_tracing_group);
333 }
334 if (!grp) {
335 return -1;
336 } else {
337 return grp->gr_gid;
338 }
339 }
340
341 /*
342 * Init thread quit pipe.
343 *
344 * Return -1 on error or 0 if all pipes are created.
345 */
346 static int init_thread_quit_pipe(void)
347 {
348 int ret, i;
349
350 ret = pipe(thread_quit_pipe);
351 if (ret < 0) {
352 PERROR("thread quit pipe");
353 goto error;
354 }
355
356 for (i = 0; i < 2; i++) {
357 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
358 if (ret < 0) {
359 PERROR("fcntl");
360 goto error;
361 }
362 }
363
364 error:
365 return ret;
366 }
367
368 /*
369 * Stop all threads by closing the thread quit pipe.
370 */
371 static void stop_threads(void)
372 {
373 int ret;
374
375 /* Stopping all threads */
376 DBG("Terminating all threads");
377 ret = notify_thread_pipe(thread_quit_pipe[1]);
378 if (ret < 0) {
379 ERR("write error on thread quit pipe");
380 }
381
382 /* Dispatch thread */
383 CMM_STORE_SHARED(dispatch_thread_exit, 1);
384 futex_nto1_wake(&ust_cmd_queue.futex);
385 }
386
387 /*
388 * Cleanup the daemon
389 */
390 static void cleanup(void)
391 {
392 int ret;
393 char *cmd = NULL;
394 struct ltt_session *sess, *stmp;
395
396 DBG("Cleaning up");
397
398 /* First thing first, stop all threads */
399 utils_close_pipe(thread_quit_pipe);
400
401 DBG("Removing %s directory", rundir);
402 ret = asprintf(&cmd, "rm -rf %s", rundir);
403 if (ret < 0) {
404 ERR("asprintf failed. Something is really wrong!");
405 }
406
407 /* Remove lttng run directory */
408 ret = system(cmd);
409 if (ret < 0) {
410 ERR("Unable to clean %s", rundir);
411 }
412 free(cmd);
413 free(rundir);
414
415 DBG("Cleaning up all sessions");
416
417 /* Destroy session list mutex */
418 if (session_list_ptr != NULL) {
419 pthread_mutex_destroy(&session_list_ptr->lock);
420
421 /* Cleanup ALL session */
422 cds_list_for_each_entry_safe(sess, stmp,
423 &session_list_ptr->head, list) {
424 cmd_destroy_session(sess, kernel_poll_pipe[1]);
425 }
426 }
427
428 DBG("Closing all UST sockets");
429 ust_app_clean_list();
430
431 if (is_root && !opt_no_kernel) {
432 DBG2("Closing kernel fd");
433 if (kernel_tracer_fd >= 0) {
434 ret = close(kernel_tracer_fd);
435 if (ret) {
436 PERROR("close");
437 }
438 }
439 DBG("Unloading kernel modules");
440 modprobe_remove_lttng_all();
441 }
442
443 /* <fun> */
444 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
445 "Matthew, BEET driven development works!%c[%dm",
446 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
447 /* </fun> */
448 }
449
450 /*
451 * Send data on a unix socket using the liblttsessiondcomm API.
452 *
453 * Return lttcomm error code.
454 */
455 static int send_unix_sock(int sock, void *buf, size_t len)
456 {
457 /* Check valid length */
458 if (len == 0) {
459 return -1;
460 }
461
462 return lttcomm_send_unix_sock(sock, buf, len);
463 }
464
465 /*
466 * Free memory of a command context structure.
467 */
468 static void clean_command_ctx(struct command_ctx **cmd_ctx)
469 {
470 DBG("Clean command context structure");
471 if (*cmd_ctx) {
472 if ((*cmd_ctx)->llm) {
473 free((*cmd_ctx)->llm);
474 }
475 if ((*cmd_ctx)->lsm) {
476 free((*cmd_ctx)->lsm);
477 }
478 free(*cmd_ctx);
479 *cmd_ctx = NULL;
480 }
481 }
482
483 /*
484 * Notify UST applications using the shm mmap futex.
485 */
486 static int notify_ust_apps(int active)
487 {
488 char *wait_shm_mmap;
489
490 DBG("Notifying applications of session daemon state: %d", active);
491
492 /* See shm.c for this call implying mmap, shm and futex calls */
493 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
494 if (wait_shm_mmap == NULL) {
495 goto error;
496 }
497
498 /* Wake waiting process */
499 futex_wait_update((int32_t *) wait_shm_mmap, active);
500
501 /* Apps notified successfully */
502 return 0;
503
504 error:
505 return -1;
506 }
507
508 /*
509 * Setup the outgoing data buffer for the response (llm) by allocating the
510 * right amount of memory and copying the original information from the lsm
511 * structure.
512 *
513 * Return total size of the buffer pointed by buf.
514 */
515 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
516 {
517 int ret, buf_size;
518
519 buf_size = size;
520
521 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
522 if (cmd_ctx->llm == NULL) {
523 PERROR("zmalloc");
524 ret = -ENOMEM;
525 goto error;
526 }
527
528 /* Copy common data */
529 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
530 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
531
532 cmd_ctx->llm->data_size = size;
533 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
534
535 return buf_size;
536
537 error:
538 return ret;
539 }
540
541 /*
542 * Update the kernel poll set of all channel fd available over all tracing
543 * session. Add the wakeup pipe at the end of the set.
544 */
545 static int update_kernel_poll(struct lttng_poll_event *events)
546 {
547 int ret;
548 struct ltt_session *session;
549 struct ltt_kernel_channel *channel;
550
551 DBG("Updating kernel poll set");
552
553 session_lock_list();
554 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
555 session_lock(session);
556 if (session->kernel_session == NULL) {
557 session_unlock(session);
558 continue;
559 }
560
561 cds_list_for_each_entry(channel,
562 &session->kernel_session->channel_list.head, list) {
563 /* Add channel fd to the kernel poll set */
564 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
565 if (ret < 0) {
566 session_unlock(session);
567 goto error;
568 }
569 DBG("Channel fd %d added to kernel set", channel->fd);
570 }
571 session_unlock(session);
572 }
573 session_unlock_list();
574
575 return 0;
576
577 error:
578 session_unlock_list();
579 return -1;
580 }
581
582 /*
583 * Find the channel fd from 'fd' over all tracing session. When found, check
584 * for new channel stream and send those stream fds to the kernel consumer.
585 *
586 * Useful for CPU hotplug feature.
587 */
588 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
589 {
590 int ret = 0;
591 struct ltt_session *session;
592 struct ltt_kernel_session *ksess;
593 struct ltt_kernel_channel *channel;
594
595 DBG("Updating kernel streams for channel fd %d", fd);
596
597 session_lock_list();
598 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
599 session_lock(session);
600 if (session->kernel_session == NULL) {
601 session_unlock(session);
602 continue;
603 }
604 ksess = session->kernel_session;
605
606 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
607 if (channel->fd == fd) {
608 DBG("Channel found, updating kernel streams");
609 ret = kernel_open_channel_stream(channel);
610 if (ret < 0) {
611 goto error;
612 }
613
614 /*
615 * Have we already sent fds to the consumer? If yes, it means
616 * that tracing is started so it is safe to send our updated
617 * stream fds.
618 */
619 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
620 struct lttng_ht_iter iter;
621 struct consumer_socket *socket;
622
623 rcu_read_lock();
624 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
625 &iter.iter, socket, node.node) {
626 /* Code flow error */
627 assert(socket->fd >= 0);
628
629 pthread_mutex_lock(socket->lock);
630 ret = kernel_consumer_send_channel_stream(socket,
631 channel, ksess);
632 pthread_mutex_unlock(socket->lock);
633 if (ret < 0) {
634 rcu_read_unlock();
635 goto error;
636 }
637 }
638 rcu_read_unlock();
639 }
640 goto error;
641 }
642 }
643 session_unlock(session);
644 }
645 session_unlock_list();
646 return ret;
647
648 error:
649 session_unlock(session);
650 session_unlock_list();
651 return ret;
652 }
653
654 /*
655 * For each tracing session, update newly registered apps.
656 */
657 static void update_ust_app(int app_sock)
658 {
659 struct ltt_session *sess, *stmp;
660
661 /* For all tracing session(s) */
662 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
663 session_lock(sess);
664 if (sess->ust_session) {
665 ust_app_global_update(sess->ust_session, app_sock);
666 }
667 session_unlock(sess);
668 }
669 }
670
671 /*
672 * This thread manage event coming from the kernel.
673 *
674 * Features supported in this thread:
675 * -) CPU Hotplug
676 */
677 static void *thread_manage_kernel(void *data)
678 {
679 int ret, i, pollfd, update_poll_flag = 1, err = -1;
680 uint32_t revents, nb_fd;
681 char tmp;
682 struct lttng_poll_event events;
683
684 DBG("[thread] Thread manage kernel started");
685
686 health_register(HEALTH_TYPE_KERNEL);
687
688 /*
689 * This first step of the while is to clean this structure which could free
690 * non NULL pointers so zero it before the loop.
691 */
692 memset(&events, 0, sizeof(events));
693
694 if (testpoint(thread_manage_kernel)) {
695 goto error_testpoint;
696 }
697
698 health_code_update();
699
700 if (testpoint(thread_manage_kernel_before_loop)) {
701 goto error_testpoint;
702 }
703
704 while (1) {
705 health_code_update();
706
707 if (update_poll_flag == 1) {
708 /* Clean events object. We are about to populate it again. */
709 lttng_poll_clean(&events);
710
711 ret = create_thread_poll_set(&events, 2);
712 if (ret < 0) {
713 goto error_poll_create;
714 }
715
716 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
717 if (ret < 0) {
718 goto error;
719 }
720
721 /* This will add the available kernel channel if any. */
722 ret = update_kernel_poll(&events);
723 if (ret < 0) {
724 goto error;
725 }
726 update_poll_flag = 0;
727 }
728
729 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
730
731 /* Poll infinite value of time */
732 restart:
733 health_poll_entry();
734 ret = lttng_poll_wait(&events, -1);
735 health_poll_exit();
736 if (ret < 0) {
737 /*
738 * Restart interrupted system call.
739 */
740 if (errno == EINTR) {
741 goto restart;
742 }
743 goto error;
744 } else if (ret == 0) {
745 /* Should not happen since timeout is infinite */
746 ERR("Return value of poll is 0 with an infinite timeout.\n"
747 "This should not have happened! Continuing...");
748 continue;
749 }
750
751 nb_fd = ret;
752
753 for (i = 0; i < nb_fd; i++) {
754 /* Fetch once the poll data */
755 revents = LTTNG_POLL_GETEV(&events, i);
756 pollfd = LTTNG_POLL_GETFD(&events, i);
757
758 health_code_update();
759
760 /* Thread quit pipe has been closed. Killing thread. */
761 ret = check_thread_quit_pipe(pollfd, revents);
762 if (ret) {
763 err = 0;
764 goto exit;
765 }
766
767 /* Check for data on kernel pipe */
768 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
769 do {
770 ret = read(kernel_poll_pipe[0], &tmp, 1);
771 } while (ret < 0 && errno == EINTR);
772 /*
773 * Ret value is useless here, if this pipe gets any actions an
774 * update is required anyway.
775 */
776 update_poll_flag = 1;
777 continue;
778 } else {
779 /*
780 * New CPU detected by the kernel. Adding kernel stream to
781 * kernel session and updating the kernel consumer
782 */
783 if (revents & LPOLLIN) {
784 ret = update_kernel_stream(&kconsumer_data, pollfd);
785 if (ret < 0) {
786 continue;
787 }
788 break;
789 /*
790 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
791 * and unregister kernel stream at this point.
792 */
793 }
794 }
795 }
796 }
797
798 exit:
799 error:
800 lttng_poll_clean(&events);
801 error_poll_create:
802 error_testpoint:
803 utils_close_pipe(kernel_poll_pipe);
804 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
805 if (err) {
806 health_error();
807 ERR("Health error occurred in %s", __func__);
808 WARN("Kernel thread died unexpectedly. "
809 "Kernel tracing can continue but CPU hotplug is disabled.");
810 }
811 health_unregister();
812 DBG("Kernel thread dying");
813 return NULL;
814 }
815
816 /*
817 * Signal pthread condition of the consumer data that the thread.
818 */
819 static void signal_consumer_condition(struct consumer_data *data, int state)
820 {
821 pthread_mutex_lock(&data->cond_mutex);
822
823 /*
824 * The state is set before signaling. It can be any value, it's the waiter
825 * job to correctly interpret this condition variable associated to the
826 * consumer pthread_cond.
827 *
828 * A value of 0 means that the corresponding thread of the consumer data
829 * was not started. 1 indicates that the thread has started and is ready
830 * for action. A negative value means that there was an error during the
831 * thread bootstrap.
832 */
833 data->consumer_thread_is_ready = state;
834 (void) pthread_cond_signal(&data->cond);
835
836 pthread_mutex_unlock(&data->cond_mutex);
837 }
838
839 /*
840 * This thread manage the consumer error sent back to the session daemon.
841 */
842 static void *thread_manage_consumer(void *data)
843 {
844 int sock = -1, i, ret, pollfd, err = -1;
845 uint32_t revents, nb_fd;
846 enum lttcomm_return_code code;
847 struct lttng_poll_event events;
848 struct consumer_data *consumer_data = data;
849
850 DBG("[thread] Manage consumer started");
851
852 health_register(HEALTH_TYPE_CONSUMER);
853
854 health_code_update();
855
856 /*
857 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
858 * Nothing more will be added to this poll set.
859 */
860 ret = create_thread_poll_set(&events, 2);
861 if (ret < 0) {
862 goto error_poll;
863 }
864
865 /*
866 * The error socket here is already in a listening state which was done
867 * just before spawning this thread to avoid a race between the consumer
868 * daemon exec trying to connect and the listen() call.
869 */
870 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
871 if (ret < 0) {
872 goto error;
873 }
874
875 health_code_update();
876
877 /* Inifinite blocking call, waiting for transmission */
878 restart:
879 health_poll_entry();
880
881 if (testpoint(thread_manage_consumer)) {
882 goto error;
883 }
884
885 ret = lttng_poll_wait(&events, -1);
886 health_poll_exit();
887 if (ret < 0) {
888 /*
889 * Restart interrupted system call.
890 */
891 if (errno == EINTR) {
892 goto restart;
893 }
894 goto error;
895 }
896
897 nb_fd = ret;
898
899 for (i = 0; i < nb_fd; i++) {
900 /* Fetch once the poll data */
901 revents = LTTNG_POLL_GETEV(&events, i);
902 pollfd = LTTNG_POLL_GETFD(&events, i);
903
904 health_code_update();
905
906 /* Thread quit pipe has been closed. Killing thread. */
907 ret = check_thread_quit_pipe(pollfd, revents);
908 if (ret) {
909 err = 0;
910 goto exit;
911 }
912
913 /* Event on the registration socket */
914 if (pollfd == consumer_data->err_sock) {
915 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
916 ERR("consumer err socket poll error");
917 goto error;
918 }
919 }
920 }
921
922 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
923 if (sock < 0) {
924 goto error;
925 }
926
927 /*
928 * Set the CLOEXEC flag. Return code is useless because either way, the
929 * show must go on.
930 */
931 (void) utils_set_fd_cloexec(sock);
932
933 health_code_update();
934
935 DBG2("Receiving code from consumer err_sock");
936
937 /* Getting status code from kconsumerd */
938 ret = lttcomm_recv_unix_sock(sock, &code,
939 sizeof(enum lttcomm_return_code));
940 if (ret <= 0) {
941 goto error;
942 }
943
944 health_code_update();
945
946 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
947 consumer_data->cmd_sock =
948 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
949 if (consumer_data->cmd_sock < 0) {
950 /* On error, signal condition and quit. */
951 signal_consumer_condition(consumer_data, -1);
952 PERROR("consumer connect");
953 goto error;
954 }
955 signal_consumer_condition(consumer_data, 1);
956 DBG("Consumer command socket ready");
957 } else {
958 ERR("consumer error when waiting for SOCK_READY : %s",
959 lttcomm_get_readable_code(-code));
960 goto error;
961 }
962
963 /* Remove the kconsumerd error sock since we've established a connexion */
964 ret = lttng_poll_del(&events, consumer_data->err_sock);
965 if (ret < 0) {
966 goto error;
967 }
968
969 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
970 if (ret < 0) {
971 goto error;
972 }
973
974 health_code_update();
975
976 /* Inifinite blocking call, waiting for transmission */
977 restart_poll:
978 health_poll_entry();
979 ret = lttng_poll_wait(&events, -1);
980 health_poll_exit();
981 if (ret < 0) {
982 /*
983 * Restart interrupted system call.
984 */
985 if (errno == EINTR) {
986 goto restart_poll;
987 }
988 goto error;
989 }
990
991 nb_fd = ret;
992
993 for (i = 0; i < nb_fd; i++) {
994 /* Fetch once the poll data */
995 revents = LTTNG_POLL_GETEV(&events, i);
996 pollfd = LTTNG_POLL_GETFD(&events, i);
997
998 health_code_update();
999
1000 /* Thread quit pipe has been closed. Killing thread. */
1001 ret = check_thread_quit_pipe(pollfd, revents);
1002 if (ret) {
1003 err = 0;
1004 goto exit;
1005 }
1006
1007 /* Event on the kconsumerd socket */
1008 if (pollfd == sock) {
1009 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1010 ERR("consumer err socket second poll error");
1011 goto error;
1012 }
1013 }
1014 }
1015
1016 health_code_update();
1017
1018 /* Wait for any kconsumerd error */
1019 ret = lttcomm_recv_unix_sock(sock, &code,
1020 sizeof(enum lttcomm_return_code));
1021 if (ret <= 0) {
1022 ERR("consumer closed the command socket");
1023 goto error;
1024 }
1025
1026 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1027
1028 exit:
1029 error:
1030 /* Immediately set the consumerd state to stopped */
1031 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1032 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1033 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1034 consumer_data->type == LTTNG_CONSUMER32_UST) {
1035 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1036 } else {
1037 /* Code flow error... */
1038 assert(0);
1039 }
1040
1041 if (consumer_data->err_sock >= 0) {
1042 ret = close(consumer_data->err_sock);
1043 if (ret) {
1044 PERROR("close");
1045 }
1046 }
1047 if (consumer_data->cmd_sock >= 0) {
1048 ret = close(consumer_data->cmd_sock);
1049 if (ret) {
1050 PERROR("close");
1051 }
1052 }
1053 if (sock >= 0) {
1054 ret = close(sock);
1055 if (ret) {
1056 PERROR("close");
1057 }
1058 }
1059
1060 unlink(consumer_data->err_unix_sock_path);
1061 unlink(consumer_data->cmd_unix_sock_path);
1062 consumer_data->pid = 0;
1063
1064 lttng_poll_clean(&events);
1065 error_poll:
1066 if (err) {
1067 health_error();
1068 ERR("Health error occurred in %s", __func__);
1069 }
1070 health_unregister();
1071 DBG("consumer thread cleanup completed");
1072
1073 return NULL;
1074 }
1075
1076 /*
1077 * This thread manage application communication.
1078 */
1079 static void *thread_manage_apps(void *data)
1080 {
1081 int i, ret, pollfd, err = -1;
1082 uint32_t revents, nb_fd;
1083 struct ust_command ust_cmd;
1084 struct lttng_poll_event events;
1085
1086 DBG("[thread] Manage application started");
1087
1088 rcu_register_thread();
1089 rcu_thread_online();
1090
1091 health_register(HEALTH_TYPE_APP_MANAGE);
1092
1093 if (testpoint(thread_manage_apps)) {
1094 goto error_testpoint;
1095 }
1096
1097 health_code_update();
1098
1099 ret = create_thread_poll_set(&events, 2);
1100 if (ret < 0) {
1101 goto error_poll_create;
1102 }
1103
1104 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1105 if (ret < 0) {
1106 goto error;
1107 }
1108
1109 if (testpoint(thread_manage_apps_before_loop)) {
1110 goto error;
1111 }
1112
1113 health_code_update();
1114
1115 while (1) {
1116 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1117
1118 /* Inifinite blocking call, waiting for transmission */
1119 restart:
1120 health_poll_entry();
1121 ret = lttng_poll_wait(&events, -1);
1122 health_poll_exit();
1123 if (ret < 0) {
1124 /*
1125 * Restart interrupted system call.
1126 */
1127 if (errno == EINTR) {
1128 goto restart;
1129 }
1130 goto error;
1131 }
1132
1133 nb_fd = ret;
1134
1135 for (i = 0; i < nb_fd; i++) {
1136 /* Fetch once the poll data */
1137 revents = LTTNG_POLL_GETEV(&events, i);
1138 pollfd = LTTNG_POLL_GETFD(&events, i);
1139
1140 health_code_update();
1141
1142 /* Thread quit pipe has been closed. Killing thread. */
1143 ret = check_thread_quit_pipe(pollfd, revents);
1144 if (ret) {
1145 err = 0;
1146 goto exit;
1147 }
1148
1149 /* Inspect the apps cmd pipe */
1150 if (pollfd == apps_cmd_pipe[0]) {
1151 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1152 ERR("Apps command pipe error");
1153 goto error;
1154 } else if (revents & LPOLLIN) {
1155 /* Empty pipe */
1156 do {
1157 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1158 } while (ret < 0 && errno == EINTR);
1159 if (ret < 0 || ret < sizeof(ust_cmd)) {
1160 PERROR("read apps cmd pipe");
1161 goto error;
1162 }
1163
1164 health_code_update();
1165
1166 /*
1167 * @session_lock
1168 * Lock the global session list so from the register up to
1169 * the registration done message, no thread can see the
1170 * application and change its state.
1171 */
1172 session_lock_list();
1173
1174 /* Register applicaton to the session daemon */
1175 ret = ust_app_register(&ust_cmd.reg_msg,
1176 ust_cmd.sock);
1177 if (ret == -ENOMEM) {
1178 session_unlock_list();
1179 goto error;
1180 } else if (ret < 0) {
1181 session_unlock_list();
1182 break;
1183 }
1184
1185 health_code_update();
1186
1187 /*
1188 * Validate UST version compatibility.
1189 */
1190 ret = ust_app_validate_version(ust_cmd.sock);
1191 if (ret >= 0) {
1192 /*
1193 * Add channel(s) and event(s) to newly registered apps
1194 * from lttng global UST domain.
1195 */
1196 update_ust_app(ust_cmd.sock);
1197 }
1198
1199 health_code_update();
1200
1201 ret = ust_app_register_done(ust_cmd.sock);
1202 if (ret < 0) {
1203 /*
1204 * If the registration is not possible, we simply
1205 * unregister the apps and continue
1206 */
1207 ust_app_unregister(ust_cmd.sock);
1208 } else {
1209 /*
1210 * We only monitor the error events of the socket. This
1211 * thread does not handle any incoming data from UST
1212 * (POLLIN).
1213 */
1214 ret = lttng_poll_add(&events, ust_cmd.sock,
1215 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1216 if (ret < 0) {
1217 session_unlock_list();
1218 goto error;
1219 }
1220
1221 /* Set socket timeout for both receiving and ending */
1222 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1223 app_socket_timeout);
1224 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1225 app_socket_timeout);
1226
1227 DBG("Apps with sock %d added to poll set",
1228 ust_cmd.sock);
1229 }
1230 session_unlock_list();
1231
1232 health_code_update();
1233
1234 break;
1235 }
1236 } else {
1237 /*
1238 * At this point, we know that a registered application made
1239 * the event at poll_wait.
1240 */
1241 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1242 /* Removing from the poll set */
1243 ret = lttng_poll_del(&events, pollfd);
1244 if (ret < 0) {
1245 goto error;
1246 }
1247
1248 /* Socket closed on remote end. */
1249 ust_app_unregister(pollfd);
1250 break;
1251 }
1252 }
1253
1254 health_code_update();
1255 }
1256 }
1257
1258 exit:
1259 error:
1260 lttng_poll_clean(&events);
1261 error_poll_create:
1262 error_testpoint:
1263 utils_close_pipe(apps_cmd_pipe);
1264 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1265
1266 /*
1267 * We don't clean the UST app hash table here since already registered
1268 * applications can still be controlled so let them be until the session
1269 * daemon dies or the applications stop.
1270 */
1271
1272 if (err) {
1273 health_error();
1274 ERR("Health error occurred in %s", __func__);
1275 }
1276 health_unregister();
1277 DBG("Application communication apps thread cleanup complete");
1278 rcu_thread_offline();
1279 rcu_unregister_thread();
1280 return NULL;
1281 }
1282
1283 /*
1284 * Dispatch request from the registration threads to the application
1285 * communication thread.
1286 */
1287 static void *thread_dispatch_ust_registration(void *data)
1288 {
1289 int ret;
1290 struct cds_wfq_node *node;
1291 struct ust_command *ust_cmd = NULL;
1292
1293 DBG("[thread] Dispatch UST command started");
1294
1295 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1296 /* Atomically prepare the queue futex */
1297 futex_nto1_prepare(&ust_cmd_queue.futex);
1298
1299 do {
1300 /* Dequeue command for registration */
1301 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1302 if (node == NULL) {
1303 DBG("Woken up but nothing in the UST command queue");
1304 /* Continue thread execution */
1305 break;
1306 }
1307
1308 ust_cmd = caa_container_of(node, struct ust_command, node);
1309
1310 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1311 " gid:%d sock:%d name:%s (version %d.%d)",
1312 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1313 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1314 ust_cmd->sock, ust_cmd->reg_msg.name,
1315 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1316 /*
1317 * Inform apps thread of the new application registration. This
1318 * call is blocking so we can be assured that the data will be read
1319 * at some point in time or wait to the end of the world :)
1320 */
1321 if (apps_cmd_pipe[1] >= 0) {
1322 do {
1323 ret = write(apps_cmd_pipe[1], ust_cmd,
1324 sizeof(struct ust_command));
1325 } while (ret < 0 && errno == EINTR);
1326 if (ret < 0 || ret != sizeof(struct ust_command)) {
1327 PERROR("write apps cmd pipe");
1328 if (errno == EBADF) {
1329 /*
1330 * We can't inform the application thread to process
1331 * registration. We will exit or else application
1332 * registration will not occur and tracing will never
1333 * start.
1334 */
1335 goto error;
1336 }
1337 }
1338 } else {
1339 /* Application manager thread is not available. */
1340 ret = close(ust_cmd->sock);
1341 if (ret < 0) {
1342 PERROR("close ust_cmd sock");
1343 }
1344 }
1345 free(ust_cmd);
1346 } while (node != NULL);
1347
1348 /* Futex wait on queue. Blocking call on futex() */
1349 futex_nto1_wait(&ust_cmd_queue.futex);
1350 }
1351
1352 error:
1353 DBG("Dispatch thread dying");
1354 return NULL;
1355 }
1356
1357 /*
1358 * This thread manage application registration.
1359 */
1360 static void *thread_registration_apps(void *data)
1361 {
1362 int sock = -1, i, ret, pollfd, err = -1;
1363 uint32_t revents, nb_fd;
1364 struct lttng_poll_event events;
1365 /*
1366 * Get allocated in this thread, enqueued to a global queue, dequeued and
1367 * freed in the manage apps thread.
1368 */
1369 struct ust_command *ust_cmd = NULL;
1370
1371 DBG("[thread] Manage application registration started");
1372
1373 health_register(HEALTH_TYPE_APP_REG);
1374
1375 if (testpoint(thread_registration_apps)) {
1376 goto error_testpoint;
1377 }
1378
1379 ret = lttcomm_listen_unix_sock(apps_sock);
1380 if (ret < 0) {
1381 goto error_listen;
1382 }
1383
1384 /*
1385 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1386 * more will be added to this poll set.
1387 */
1388 ret = create_thread_poll_set(&events, 2);
1389 if (ret < 0) {
1390 goto error_create_poll;
1391 }
1392
1393 /* Add the application registration socket */
1394 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1395 if (ret < 0) {
1396 goto error_poll_add;
1397 }
1398
1399 /* Notify all applications to register */
1400 ret = notify_ust_apps(1);
1401 if (ret < 0) {
1402 ERR("Failed to notify applications or create the wait shared memory.\n"
1403 "Execution continues but there might be problem for already\n"
1404 "running applications that wishes to register.");
1405 }
1406
1407 while (1) {
1408 DBG("Accepting application registration");
1409
1410 /* Inifinite blocking call, waiting for transmission */
1411 restart:
1412 health_poll_entry();
1413 ret = lttng_poll_wait(&events, -1);
1414 health_poll_exit();
1415 if (ret < 0) {
1416 /*
1417 * Restart interrupted system call.
1418 */
1419 if (errno == EINTR) {
1420 goto restart;
1421 }
1422 goto error;
1423 }
1424
1425 nb_fd = ret;
1426
1427 for (i = 0; i < nb_fd; i++) {
1428 health_code_update();
1429
1430 /* Fetch once the poll data */
1431 revents = LTTNG_POLL_GETEV(&events, i);
1432 pollfd = LTTNG_POLL_GETFD(&events, i);
1433
1434 /* Thread quit pipe has been closed. Killing thread. */
1435 ret = check_thread_quit_pipe(pollfd, revents);
1436 if (ret) {
1437 err = 0;
1438 goto exit;
1439 }
1440
1441 /* Event on the registration socket */
1442 if (pollfd == apps_sock) {
1443 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1444 ERR("Register apps socket poll error");
1445 goto error;
1446 } else if (revents & LPOLLIN) {
1447 sock = lttcomm_accept_unix_sock(apps_sock);
1448 if (sock < 0) {
1449 goto error;
1450 }
1451
1452 /*
1453 * Set the CLOEXEC flag. Return code is useless because
1454 * either way, the show must go on.
1455 */
1456 (void) utils_set_fd_cloexec(sock);
1457
1458 /* Create UST registration command for enqueuing */
1459 ust_cmd = zmalloc(sizeof(struct ust_command));
1460 if (ust_cmd == NULL) {
1461 PERROR("ust command zmalloc");
1462 goto error;
1463 }
1464
1465 /*
1466 * Using message-based transmissions to ensure we don't
1467 * have to deal with partially received messages.
1468 */
1469 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1470 if (ret < 0) {
1471 ERR("Exhausted file descriptors allowed for applications.");
1472 free(ust_cmd);
1473 ret = close(sock);
1474 if (ret) {
1475 PERROR("close");
1476 }
1477 sock = -1;
1478 continue;
1479 }
1480 health_code_update();
1481 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1482 sizeof(struct ust_register_msg));
1483 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1484 if (ret < 0) {
1485 PERROR("lttcomm_recv_unix_sock register apps");
1486 } else {
1487 ERR("Wrong size received on apps register");
1488 }
1489 free(ust_cmd);
1490 ret = close(sock);
1491 if (ret) {
1492 PERROR("close");
1493 }
1494 lttng_fd_put(LTTNG_FD_APPS, 1);
1495 sock = -1;
1496 continue;
1497 }
1498 health_code_update();
1499
1500 ust_cmd->sock = sock;
1501 sock = -1;
1502
1503 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1504 " gid:%d sock:%d name:%s (version %d.%d)",
1505 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1506 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1507 ust_cmd->sock, ust_cmd->reg_msg.name,
1508 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1509
1510 /*
1511 * Lock free enqueue the registration request. The red pill
1512 * has been taken! This apps will be part of the *system*.
1513 */
1514 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1515
1516 /*
1517 * Wake the registration queue futex. Implicit memory
1518 * barrier with the exchange in cds_wfq_enqueue.
1519 */
1520 futex_nto1_wake(&ust_cmd_queue.futex);
1521 }
1522 }
1523 }
1524 }
1525
1526 exit:
1527 error:
1528 if (err) {
1529 health_error();
1530 ERR("Health error occurred in %s", __func__);
1531 }
1532
1533 /* Notify that the registration thread is gone */
1534 notify_ust_apps(0);
1535
1536 if (apps_sock >= 0) {
1537 ret = close(apps_sock);
1538 if (ret) {
1539 PERROR("close");
1540 }
1541 }
1542 if (sock >= 0) {
1543 ret = close(sock);
1544 if (ret) {
1545 PERROR("close");
1546 }
1547 lttng_fd_put(LTTNG_FD_APPS, 1);
1548 }
1549 unlink(apps_unix_sock_path);
1550
1551 error_poll_add:
1552 lttng_poll_clean(&events);
1553 error_listen:
1554 error_create_poll:
1555 error_testpoint:
1556 DBG("UST Registration thread cleanup complete");
1557 health_unregister();
1558
1559 return NULL;
1560 }
1561
1562 /*
1563 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1564 * exec or it will fails.
1565 */
1566 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1567 {
1568 int ret, clock_ret;
1569 struct timespec timeout;
1570
1571 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1572 consumer_data->consumer_thread_is_ready = 0;
1573
1574 /* Setup pthread condition */
1575 ret = pthread_condattr_init(&consumer_data->condattr);
1576 if (ret != 0) {
1577 errno = ret;
1578 PERROR("pthread_condattr_init consumer data");
1579 goto error;
1580 }
1581
1582 /*
1583 * Set the monotonic clock in order to make sure we DO NOT jump in time
1584 * between the clock_gettime() call and the timedwait call. See bug #324
1585 * for a more details and how we noticed it.
1586 */
1587 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1588 if (ret != 0) {
1589 errno = ret;
1590 PERROR("pthread_condattr_setclock consumer data");
1591 goto error;
1592 }
1593
1594 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1595 if (ret != 0) {
1596 errno = ret;
1597 PERROR("pthread_cond_init consumer data");
1598 goto error;
1599 }
1600
1601 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1602 consumer_data);
1603 if (ret != 0) {
1604 PERROR("pthread_create consumer");
1605 ret = -1;
1606 goto error;
1607 }
1608
1609 /* We are about to wait on a pthread condition */
1610 pthread_mutex_lock(&consumer_data->cond_mutex);
1611
1612 /* Get time for sem_timedwait absolute timeout */
1613 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1614 /*
1615 * Set the timeout for the condition timed wait even if the clock gettime
1616 * call fails since we might loop on that call and we want to avoid to
1617 * increment the timeout too many times.
1618 */
1619 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1620
1621 /*
1622 * The following loop COULD be skipped in some conditions so this is why we
1623 * set ret to 0 in order to make sure at least one round of the loop is
1624 * done.
1625 */
1626 ret = 0;
1627
1628 /*
1629 * Loop until the condition is reached or when a timeout is reached. Note
1630 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1631 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1632 * possible. This loop does not take any chances and works with both of
1633 * them.
1634 */
1635 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1636 if (clock_ret < 0) {
1637 PERROR("clock_gettime spawn consumer");
1638 /* Infinite wait for the consumerd thread to be ready */
1639 ret = pthread_cond_wait(&consumer_data->cond,
1640 &consumer_data->cond_mutex);
1641 } else {
1642 ret = pthread_cond_timedwait(&consumer_data->cond,
1643 &consumer_data->cond_mutex, &timeout);
1644 }
1645 }
1646
1647 /* Release the pthread condition */
1648 pthread_mutex_unlock(&consumer_data->cond_mutex);
1649
1650 if (ret != 0) {
1651 errno = ret;
1652 if (ret == ETIMEDOUT) {
1653 /*
1654 * Call has timed out so we kill the kconsumerd_thread and return
1655 * an error.
1656 */
1657 ERR("Condition timed out. The consumer thread was never ready."
1658 " Killing it");
1659 ret = pthread_cancel(consumer_data->thread);
1660 if (ret < 0) {
1661 PERROR("pthread_cancel consumer thread");
1662 }
1663 } else {
1664 PERROR("pthread_cond_wait failed consumer thread");
1665 }
1666 goto error;
1667 }
1668
1669 pthread_mutex_lock(&consumer_data->pid_mutex);
1670 if (consumer_data->pid == 0) {
1671 ERR("Consumerd did not start");
1672 pthread_mutex_unlock(&consumer_data->pid_mutex);
1673 goto error;
1674 }
1675 pthread_mutex_unlock(&consumer_data->pid_mutex);
1676
1677 return 0;
1678
1679 error:
1680 return ret;
1681 }
1682
1683 /*
1684 * Join consumer thread
1685 */
1686 static int join_consumer_thread(struct consumer_data *consumer_data)
1687 {
1688 void *status;
1689
1690 /* Consumer pid must be a real one. */
1691 if (consumer_data->pid > 0) {
1692 int ret;
1693 ret = kill(consumer_data->pid, SIGTERM);
1694 if (ret) {
1695 ERR("Error killing consumer daemon");
1696 return ret;
1697 }
1698 return pthread_join(consumer_data->thread, &status);
1699 } else {
1700 return 0;
1701 }
1702 }
1703
1704 /*
1705 * Fork and exec a consumer daemon (consumerd).
1706 *
1707 * Return pid if successful else -1.
1708 */
1709 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1710 {
1711 int ret;
1712 pid_t pid;
1713 const char *consumer_to_use;
1714 const char *verbosity;
1715 struct stat st;
1716
1717 DBG("Spawning consumerd");
1718
1719 pid = fork();
1720 if (pid == 0) {
1721 /*
1722 * Exec consumerd.
1723 */
1724 if (opt_verbose_consumer) {
1725 verbosity = "--verbose";
1726 } else {
1727 verbosity = "--quiet";
1728 }
1729 switch (consumer_data->type) {
1730 case LTTNG_CONSUMER_KERNEL:
1731 /*
1732 * Find out which consumerd to execute. We will first try the
1733 * 64-bit path, then the sessiond's installation directory, and
1734 * fallback on the 32-bit one,
1735 */
1736 DBG3("Looking for a kernel consumer at these locations:");
1737 DBG3(" 1) %s", consumerd64_bin);
1738 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1739 DBG3(" 3) %s", consumerd32_bin);
1740 if (stat(consumerd64_bin, &st) == 0) {
1741 DBG3("Found location #1");
1742 consumer_to_use = consumerd64_bin;
1743 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1744 DBG3("Found location #2");
1745 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1746 } else if (stat(consumerd32_bin, &st) == 0) {
1747 DBG3("Found location #3");
1748 consumer_to_use = consumerd32_bin;
1749 } else {
1750 DBG("Could not find any valid consumerd executable");
1751 break;
1752 }
1753 DBG("Using kernel consumer at: %s", consumer_to_use);
1754 execl(consumer_to_use,
1755 "lttng-consumerd", verbosity, "-k",
1756 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1757 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1758 NULL);
1759 break;
1760 case LTTNG_CONSUMER64_UST:
1761 {
1762 char *tmpnew = NULL;
1763
1764 if (consumerd64_libdir[0] != '\0') {
1765 char *tmp;
1766 size_t tmplen;
1767
1768 tmp = getenv("LD_LIBRARY_PATH");
1769 if (!tmp) {
1770 tmp = "";
1771 }
1772 tmplen = strlen("LD_LIBRARY_PATH=")
1773 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1774 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1775 if (!tmpnew) {
1776 ret = -ENOMEM;
1777 goto error;
1778 }
1779 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1780 strcat(tmpnew, consumerd64_libdir);
1781 if (tmp[0] != '\0') {
1782 strcat(tmpnew, ":");
1783 strcat(tmpnew, tmp);
1784 }
1785 ret = putenv(tmpnew);
1786 if (ret) {
1787 ret = -errno;
1788 goto error;
1789 }
1790 }
1791 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1792 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1793 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1794 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1795 NULL);
1796 if (consumerd64_libdir[0] != '\0') {
1797 free(tmpnew);
1798 }
1799 if (ret) {
1800 goto error;
1801 }
1802 break;
1803 }
1804 case LTTNG_CONSUMER32_UST:
1805 {
1806 char *tmpnew = NULL;
1807
1808 if (consumerd32_libdir[0] != '\0') {
1809 char *tmp;
1810 size_t tmplen;
1811
1812 tmp = getenv("LD_LIBRARY_PATH");
1813 if (!tmp) {
1814 tmp = "";
1815 }
1816 tmplen = strlen("LD_LIBRARY_PATH=")
1817 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1818 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1819 if (!tmpnew) {
1820 ret = -ENOMEM;
1821 goto error;
1822 }
1823 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1824 strcat(tmpnew, consumerd32_libdir);
1825 if (tmp[0] != '\0') {
1826 strcat(tmpnew, ":");
1827 strcat(tmpnew, tmp);
1828 }
1829 ret = putenv(tmpnew);
1830 if (ret) {
1831 ret = -errno;
1832 goto error;
1833 }
1834 }
1835 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1836 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1837 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1838 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1839 NULL);
1840 if (consumerd32_libdir[0] != '\0') {
1841 free(tmpnew);
1842 }
1843 if (ret) {
1844 goto error;
1845 }
1846 break;
1847 }
1848 default:
1849 PERROR("unknown consumer type");
1850 exit(EXIT_FAILURE);
1851 }
1852 if (errno != 0) {
1853 PERROR("kernel start consumer exec");
1854 }
1855 exit(EXIT_FAILURE);
1856 } else if (pid > 0) {
1857 ret = pid;
1858 } else {
1859 PERROR("start consumer fork");
1860 ret = -errno;
1861 }
1862 error:
1863 return ret;
1864 }
1865
1866 /*
1867 * Spawn the consumerd daemon and session daemon thread.
1868 */
1869 static int start_consumerd(struct consumer_data *consumer_data)
1870 {
1871 int ret;
1872
1873 /*
1874 * Set the listen() state on the socket since there is a possible race
1875 * between the exec() of the consumer daemon and this call if place in the
1876 * consumer thread. See bug #366 for more details.
1877 */
1878 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1879 if (ret < 0) {
1880 goto error;
1881 }
1882
1883 pthread_mutex_lock(&consumer_data->pid_mutex);
1884 if (consumer_data->pid != 0) {
1885 pthread_mutex_unlock(&consumer_data->pid_mutex);
1886 goto end;
1887 }
1888
1889 ret = spawn_consumerd(consumer_data);
1890 if (ret < 0) {
1891 ERR("Spawning consumerd failed");
1892 pthread_mutex_unlock(&consumer_data->pid_mutex);
1893 goto error;
1894 }
1895
1896 /* Setting up the consumer_data pid */
1897 consumer_data->pid = ret;
1898 DBG2("Consumer pid %d", consumer_data->pid);
1899 pthread_mutex_unlock(&consumer_data->pid_mutex);
1900
1901 DBG2("Spawning consumer control thread");
1902 ret = spawn_consumer_thread(consumer_data);
1903 if (ret < 0) {
1904 ERR("Fatal error spawning consumer control thread");
1905 goto error;
1906 }
1907
1908 end:
1909 return 0;
1910
1911 error:
1912 /* Cleanup already created socket on error. */
1913 if (consumer_data->err_sock >= 0) {
1914 int err;
1915
1916 err = close(consumer_data->err_sock);
1917 if (err < 0) {
1918 PERROR("close consumer data error socket");
1919 }
1920 }
1921 return ret;
1922 }
1923
1924 /*
1925 * Compute health status of each consumer. If one of them is zero (bad
1926 * state), we return 0.
1927 */
1928 static int check_consumer_health(void)
1929 {
1930 int ret;
1931
1932 ret = health_check_state(HEALTH_TYPE_CONSUMER);
1933
1934 DBG3("Health consumer check %d", ret);
1935
1936 return ret;
1937 }
1938
1939 /*
1940 * Setup necessary data for kernel tracer action.
1941 */
1942 static int init_kernel_tracer(void)
1943 {
1944 int ret;
1945
1946 /* Modprobe lttng kernel modules */
1947 ret = modprobe_lttng_control();
1948 if (ret < 0) {
1949 goto error;
1950 }
1951
1952 /* Open debugfs lttng */
1953 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1954 if (kernel_tracer_fd < 0) {
1955 DBG("Failed to open %s", module_proc_lttng);
1956 ret = -1;
1957 goto error_open;
1958 }
1959
1960 /* Validate kernel version */
1961 ret = kernel_validate_version(kernel_tracer_fd);
1962 if (ret < 0) {
1963 goto error_version;
1964 }
1965
1966 ret = modprobe_lttng_data();
1967 if (ret < 0) {
1968 goto error_modules;
1969 }
1970
1971 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1972 return 0;
1973
1974 error_version:
1975 modprobe_remove_lttng_control();
1976 ret = close(kernel_tracer_fd);
1977 if (ret) {
1978 PERROR("close");
1979 }
1980 kernel_tracer_fd = -1;
1981 return LTTNG_ERR_KERN_VERSION;
1982
1983 error_modules:
1984 ret = close(kernel_tracer_fd);
1985 if (ret) {
1986 PERROR("close");
1987 }
1988
1989 error_open:
1990 modprobe_remove_lttng_control();
1991
1992 error:
1993 WARN("No kernel tracer available");
1994 kernel_tracer_fd = -1;
1995 if (!is_root) {
1996 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1997 } else {
1998 return LTTNG_ERR_KERN_NA;
1999 }
2000 }
2001
2002
2003 /*
2004 * Copy consumer output from the tracing session to the domain session. The
2005 * function also applies the right modification on a per domain basis for the
2006 * trace files destination directory.
2007 */
2008 static int copy_session_consumer(int domain, struct ltt_session *session)
2009 {
2010 int ret;
2011 const char *dir_name;
2012 struct consumer_output *consumer;
2013
2014 assert(session);
2015 assert(session->consumer);
2016
2017 switch (domain) {
2018 case LTTNG_DOMAIN_KERNEL:
2019 DBG3("Copying tracing session consumer output in kernel session");
2020 /*
2021 * XXX: We should audit the session creation and what this function
2022 * does "extra" in order to avoid a destroy since this function is used
2023 * in the domain session creation (kernel and ust) only. Same for UST
2024 * domain.
2025 */
2026 if (session->kernel_session->consumer) {
2027 consumer_destroy_output(session->kernel_session->consumer);
2028 }
2029 session->kernel_session->consumer =
2030 consumer_copy_output(session->consumer);
2031 /* Ease our life a bit for the next part */
2032 consumer = session->kernel_session->consumer;
2033 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2034 break;
2035 case LTTNG_DOMAIN_UST:
2036 DBG3("Copying tracing session consumer output in UST session");
2037 if (session->ust_session->consumer) {
2038 consumer_destroy_output(session->ust_session->consumer);
2039 }
2040 session->ust_session->consumer =
2041 consumer_copy_output(session->consumer);
2042 /* Ease our life a bit for the next part */
2043 consumer = session->ust_session->consumer;
2044 dir_name = DEFAULT_UST_TRACE_DIR;
2045 break;
2046 default:
2047 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2048 goto error;
2049 }
2050
2051 /* Append correct directory to subdir */
2052 strncat(consumer->subdir, dir_name,
2053 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2054 DBG3("Copy session consumer subdir %s", consumer->subdir);
2055
2056 ret = LTTNG_OK;
2057
2058 error:
2059 return ret;
2060 }
2061
2062 /*
2063 * Create an UST session and add it to the session ust list.
2064 */
2065 static int create_ust_session(struct ltt_session *session,
2066 struct lttng_domain *domain)
2067 {
2068 int ret;
2069 struct ltt_ust_session *lus = NULL;
2070
2071 assert(session);
2072 assert(domain);
2073 assert(session->consumer);
2074
2075 switch (domain->type) {
2076 case LTTNG_DOMAIN_UST:
2077 break;
2078 default:
2079 ERR("Unknown UST domain on create session %d", domain->type);
2080 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2081 goto error;
2082 }
2083
2084 DBG("Creating UST session");
2085
2086 lus = trace_ust_create_session(session->path, session->id, domain);
2087 if (lus == NULL) {
2088 ret = LTTNG_ERR_UST_SESS_FAIL;
2089 goto error;
2090 }
2091
2092 lus->uid = session->uid;
2093 lus->gid = session->gid;
2094 session->ust_session = lus;
2095
2096 /* Copy session output to the newly created UST session */
2097 ret = copy_session_consumer(domain->type, session);
2098 if (ret != LTTNG_OK) {
2099 goto error;
2100 }
2101
2102 return LTTNG_OK;
2103
2104 error:
2105 free(lus);
2106 session->ust_session = NULL;
2107 return ret;
2108 }
2109
2110 /*
2111 * Create a kernel tracer session then create the default channel.
2112 */
2113 static int create_kernel_session(struct ltt_session *session)
2114 {
2115 int ret;
2116
2117 DBG("Creating kernel session");
2118
2119 ret = kernel_create_session(session, kernel_tracer_fd);
2120 if (ret < 0) {
2121 ret = LTTNG_ERR_KERN_SESS_FAIL;
2122 goto error;
2123 }
2124
2125 /* Code flow safety */
2126 assert(session->kernel_session);
2127
2128 /* Copy session output to the newly created Kernel session */
2129 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2130 if (ret != LTTNG_OK) {
2131 goto error;
2132 }
2133
2134 /* Create directory(ies) on local filesystem. */
2135 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2136 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2137 ret = run_as_mkdir_recursive(
2138 session->kernel_session->consumer->dst.trace_path,
2139 S_IRWXU | S_IRWXG, session->uid, session->gid);
2140 if (ret < 0) {
2141 if (ret != -EEXIST) {
2142 ERR("Trace directory creation error");
2143 goto error;
2144 }
2145 }
2146 }
2147
2148 session->kernel_session->uid = session->uid;
2149 session->kernel_session->gid = session->gid;
2150
2151 return LTTNG_OK;
2152
2153 error:
2154 trace_kernel_destroy_session(session->kernel_session);
2155 session->kernel_session = NULL;
2156 return ret;
2157 }
2158
2159 /*
2160 * Count number of session permitted by uid/gid.
2161 */
2162 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2163 {
2164 unsigned int i = 0;
2165 struct ltt_session *session;
2166
2167 DBG("Counting number of available session for UID %d GID %d",
2168 uid, gid);
2169 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2170 /*
2171 * Only list the sessions the user can control.
2172 */
2173 if (!session_access_ok(session, uid, gid)) {
2174 continue;
2175 }
2176 i++;
2177 }
2178 return i;
2179 }
2180
2181 /*
2182 * Process the command requested by the lttng client within the command
2183 * context structure. This function make sure that the return structure (llm)
2184 * is set and ready for transmission before returning.
2185 *
2186 * Return any error encountered or 0 for success.
2187 *
2188 * "sock" is only used for special-case var. len data.
2189 */
2190 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2191 int *sock_error)
2192 {
2193 int ret = LTTNG_OK;
2194 int need_tracing_session = 1;
2195 int need_domain;
2196
2197 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2198
2199 *sock_error = 0;
2200
2201 switch (cmd_ctx->lsm->cmd_type) {
2202 case LTTNG_CREATE_SESSION:
2203 case LTTNG_DESTROY_SESSION:
2204 case LTTNG_LIST_SESSIONS:
2205 case LTTNG_LIST_DOMAINS:
2206 case LTTNG_START_TRACE:
2207 case LTTNG_STOP_TRACE:
2208 case LTTNG_DATA_PENDING:
2209 need_domain = 0;
2210 break;
2211 default:
2212 need_domain = 1;
2213 }
2214
2215 if (opt_no_kernel && need_domain
2216 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2217 if (!is_root) {
2218 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2219 } else {
2220 ret = LTTNG_ERR_KERN_NA;
2221 }
2222 goto error;
2223 }
2224
2225 /* Deny register consumer if we already have a spawned consumer. */
2226 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2227 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2228 if (kconsumer_data.pid > 0) {
2229 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2230 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2231 goto error;
2232 }
2233 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2234 }
2235
2236 /*
2237 * Check for command that don't needs to allocate a returned payload. We do
2238 * this here so we don't have to make the call for no payload at each
2239 * command.
2240 */
2241 switch(cmd_ctx->lsm->cmd_type) {
2242 case LTTNG_LIST_SESSIONS:
2243 case LTTNG_LIST_TRACEPOINTS:
2244 case LTTNG_LIST_TRACEPOINT_FIELDS:
2245 case LTTNG_LIST_DOMAINS:
2246 case LTTNG_LIST_CHANNELS:
2247 case LTTNG_LIST_EVENTS:
2248 break;
2249 default:
2250 /* Setup lttng message with no payload */
2251 ret = setup_lttng_msg(cmd_ctx, 0);
2252 if (ret < 0) {
2253 /* This label does not try to unlock the session */
2254 goto init_setup_error;
2255 }
2256 }
2257
2258 /* Commands that DO NOT need a session. */
2259 switch (cmd_ctx->lsm->cmd_type) {
2260 case LTTNG_CREATE_SESSION:
2261 case LTTNG_CALIBRATE:
2262 case LTTNG_LIST_SESSIONS:
2263 case LTTNG_LIST_TRACEPOINTS:
2264 case LTTNG_LIST_TRACEPOINT_FIELDS:
2265 need_tracing_session = 0;
2266 break;
2267 default:
2268 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2269 /*
2270 * We keep the session list lock across _all_ commands
2271 * for now, because the per-session lock does not
2272 * handle teardown properly.
2273 */
2274 session_lock_list();
2275 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2276 if (cmd_ctx->session == NULL) {
2277 if (cmd_ctx->lsm->session.name != NULL) {
2278 ret = LTTNG_ERR_SESS_NOT_FOUND;
2279 } else {
2280 /* If no session name specified */
2281 ret = LTTNG_ERR_SELECT_SESS;
2282 }
2283 goto error;
2284 } else {
2285 /* Acquire lock for the session */
2286 session_lock(cmd_ctx->session);
2287 }
2288 break;
2289 }
2290
2291 if (!need_domain) {
2292 goto skip_domain;
2293 }
2294
2295 /*
2296 * Check domain type for specific "pre-action".
2297 */
2298 switch (cmd_ctx->lsm->domain.type) {
2299 case LTTNG_DOMAIN_KERNEL:
2300 if (!is_root) {
2301 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2302 goto error;
2303 }
2304
2305 /* Kernel tracer check */
2306 if (kernel_tracer_fd == -1) {
2307 /* Basically, load kernel tracer modules */
2308 ret = init_kernel_tracer();
2309 if (ret != 0) {
2310 goto error;
2311 }
2312 }
2313
2314 /* Consumer is in an ERROR state. Report back to client */
2315 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2316 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2317 goto error;
2318 }
2319
2320 /* Need a session for kernel command */
2321 if (need_tracing_session) {
2322 if (cmd_ctx->session->kernel_session == NULL) {
2323 ret = create_kernel_session(cmd_ctx->session);
2324 if (ret < 0) {
2325 ret = LTTNG_ERR_KERN_SESS_FAIL;
2326 goto error;
2327 }
2328 }
2329
2330 /* Start the kernel consumer daemon */
2331 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2332 if (kconsumer_data.pid == 0 &&
2333 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2334 cmd_ctx->session->start_consumer) {
2335 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2336 ret = start_consumerd(&kconsumer_data);
2337 if (ret < 0) {
2338 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2339 goto error;
2340 }
2341 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2342 } else {
2343 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2344 }
2345
2346 /*
2347 * The consumer was just spawned so we need to add the socket to
2348 * the consumer output of the session if exist.
2349 */
2350 ret = consumer_create_socket(&kconsumer_data,
2351 cmd_ctx->session->kernel_session->consumer);
2352 if (ret < 0) {
2353 goto error;
2354 }
2355 }
2356
2357 break;
2358 case LTTNG_DOMAIN_UST:
2359 {
2360 /* Consumer is in an ERROR state. Report back to client */
2361 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2362 ret = LTTNG_ERR_NO_USTCONSUMERD;
2363 goto error;
2364 }
2365
2366 if (need_tracing_session) {
2367 /* Create UST session if none exist. */
2368 if (cmd_ctx->session->ust_session == NULL) {
2369 ret = create_ust_session(cmd_ctx->session,
2370 &cmd_ctx->lsm->domain);
2371 if (ret != LTTNG_OK) {
2372 goto error;
2373 }
2374 }
2375
2376 /* Start the UST consumer daemons */
2377 /* 64-bit */
2378 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2379 if (consumerd64_bin[0] != '\0' &&
2380 ustconsumer64_data.pid == 0 &&
2381 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2382 cmd_ctx->session->start_consumer) {
2383 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2384 ret = start_consumerd(&ustconsumer64_data);
2385 if (ret < 0) {
2386 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2387 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2388 goto error;
2389 }
2390
2391 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2392 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2393 } else {
2394 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2395 }
2396
2397 /*
2398 * Setup socket for consumer 64 bit. No need for atomic access
2399 * since it was set above and can ONLY be set in this thread.
2400 */
2401 ret = consumer_create_socket(&ustconsumer64_data,
2402 cmd_ctx->session->ust_session->consumer);
2403 if (ret < 0) {
2404 goto error;
2405 }
2406
2407 /* 32-bit */
2408 if (consumerd32_bin[0] != '\0' &&
2409 ustconsumer32_data.pid == 0 &&
2410 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2411 cmd_ctx->session->start_consumer) {
2412 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2413 ret = start_consumerd(&ustconsumer32_data);
2414 if (ret < 0) {
2415 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2416 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2417 goto error;
2418 }
2419
2420 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2421 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2422 } else {
2423 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2424 }
2425
2426 /*
2427 * Setup socket for consumer 64 bit. No need for atomic access
2428 * since it was set above and can ONLY be set in this thread.
2429 */
2430 ret = consumer_create_socket(&ustconsumer32_data,
2431 cmd_ctx->session->ust_session->consumer);
2432 if (ret < 0) {
2433 goto error;
2434 }
2435 }
2436 break;
2437 }
2438 default:
2439 break;
2440 }
2441 skip_domain:
2442
2443 /* Validate consumer daemon state when start/stop trace command */
2444 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2445 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2446 switch (cmd_ctx->lsm->domain.type) {
2447 case LTTNG_DOMAIN_UST:
2448 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2449 ret = LTTNG_ERR_NO_USTCONSUMERD;
2450 goto error;
2451 }
2452 break;
2453 case LTTNG_DOMAIN_KERNEL:
2454 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2455 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2456 goto error;
2457 }
2458 break;
2459 }
2460 }
2461
2462 /*
2463 * Check that the UID or GID match that of the tracing session.
2464 * The root user can interact with all sessions.
2465 */
2466 if (need_tracing_session) {
2467 if (!session_access_ok(cmd_ctx->session,
2468 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2469 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2470 ret = LTTNG_ERR_EPERM;
2471 goto error;
2472 }
2473 }
2474
2475 /* Process by command type */
2476 switch (cmd_ctx->lsm->cmd_type) {
2477 case LTTNG_ADD_CONTEXT:
2478 {
2479 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2480 cmd_ctx->lsm->u.context.channel_name,
2481 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2482 break;
2483 }
2484 case LTTNG_DISABLE_CHANNEL:
2485 {
2486 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2487 cmd_ctx->lsm->u.disable.channel_name);
2488 break;
2489 }
2490 case LTTNG_DISABLE_EVENT:
2491 {
2492 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2493 cmd_ctx->lsm->u.disable.channel_name,
2494 cmd_ctx->lsm->u.disable.name);
2495 break;
2496 }
2497 case LTTNG_DISABLE_ALL_EVENT:
2498 {
2499 DBG("Disabling all events");
2500
2501 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2502 cmd_ctx->lsm->u.disable.channel_name);
2503 break;
2504 }
2505 case LTTNG_DISABLE_CONSUMER:
2506 {
2507 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2508 break;
2509 }
2510 case LTTNG_ENABLE_CHANNEL:
2511 {
2512 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2513 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2514 break;
2515 }
2516 case LTTNG_ENABLE_CONSUMER:
2517 {
2518 /*
2519 * XXX: 0 means that this URI should be applied on the session. Should
2520 * be a DOMAIN enuam.
2521 */
2522 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2523 if (ret != LTTNG_OK) {
2524 goto error;
2525 }
2526
2527 if (cmd_ctx->lsm->domain.type == 0) {
2528 /* Add the URI for the UST session if a consumer is present. */
2529 if (cmd_ctx->session->ust_session &&
2530 cmd_ctx->session->ust_session->consumer) {
2531 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2532 } else if (cmd_ctx->session->kernel_session &&
2533 cmd_ctx->session->kernel_session->consumer) {
2534 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2535 cmd_ctx->session);
2536 }
2537 }
2538 break;
2539 }
2540 case LTTNG_ENABLE_EVENT:
2541 {
2542 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2543 cmd_ctx->lsm->u.enable.channel_name,
2544 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2545 break;
2546 }
2547 case LTTNG_ENABLE_ALL_EVENT:
2548 {
2549 DBG("Enabling all events");
2550
2551 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2552 cmd_ctx->lsm->u.enable.channel_name,
2553 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2554 break;
2555 }
2556 case LTTNG_LIST_TRACEPOINTS:
2557 {
2558 struct lttng_event *events;
2559 ssize_t nb_events;
2560
2561 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2562 if (nb_events < 0) {
2563 /* Return value is a negative lttng_error_code. */
2564 ret = -nb_events;
2565 goto error;
2566 }
2567
2568 /*
2569 * Setup lttng message with payload size set to the event list size in
2570 * bytes and then copy list into the llm payload.
2571 */
2572 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2573 if (ret < 0) {
2574 free(events);
2575 goto setup_error;
2576 }
2577
2578 /* Copy event list into message payload */
2579 memcpy(cmd_ctx->llm->payload, events,
2580 sizeof(struct lttng_event) * nb_events);
2581
2582 free(events);
2583
2584 ret = LTTNG_OK;
2585 break;
2586 }
2587 case LTTNG_LIST_TRACEPOINT_FIELDS:
2588 {
2589 struct lttng_event_field *fields;
2590 ssize_t nb_fields;
2591
2592 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2593 &fields);
2594 if (nb_fields < 0) {
2595 /* Return value is a negative lttng_error_code. */
2596 ret = -nb_fields;
2597 goto error;
2598 }
2599
2600 /*
2601 * Setup lttng message with payload size set to the event list size in
2602 * bytes and then copy list into the llm payload.
2603 */
2604 ret = setup_lttng_msg(cmd_ctx,
2605 sizeof(struct lttng_event_field) * nb_fields);
2606 if (ret < 0) {
2607 free(fields);
2608 goto setup_error;
2609 }
2610
2611 /* Copy event list into message payload */
2612 memcpy(cmd_ctx->llm->payload, fields,
2613 sizeof(struct lttng_event_field) * nb_fields);
2614
2615 free(fields);
2616
2617 ret = LTTNG_OK;
2618 break;
2619 }
2620 case LTTNG_SET_CONSUMER_URI:
2621 {
2622 size_t nb_uri, len;
2623 struct lttng_uri *uris;
2624
2625 nb_uri = cmd_ctx->lsm->u.uri.size;
2626 len = nb_uri * sizeof(struct lttng_uri);
2627
2628 if (nb_uri == 0) {
2629 ret = LTTNG_ERR_INVALID;
2630 goto error;
2631 }
2632
2633 uris = zmalloc(len);
2634 if (uris == NULL) {
2635 ret = LTTNG_ERR_FATAL;
2636 goto error;
2637 }
2638
2639 /* Receive variable len data */
2640 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2641 ret = lttcomm_recv_unix_sock(sock, uris, len);
2642 if (ret <= 0) {
2643 DBG("No URIs received from client... continuing");
2644 *sock_error = 1;
2645 ret = LTTNG_ERR_SESSION_FAIL;
2646 free(uris);
2647 goto error;
2648 }
2649
2650 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2651 nb_uri, uris);
2652 if (ret != LTTNG_OK) {
2653 free(uris);
2654 goto error;
2655 }
2656
2657 /*
2658 * XXX: 0 means that this URI should be applied on the session. Should
2659 * be a DOMAIN enuam.
2660 */
2661 if (cmd_ctx->lsm->domain.type == 0) {
2662 /* Add the URI for the UST session if a consumer is present. */
2663 if (cmd_ctx->session->ust_session &&
2664 cmd_ctx->session->ust_session->consumer) {
2665 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2666 nb_uri, uris);
2667 } else if (cmd_ctx->session->kernel_session &&
2668 cmd_ctx->session->kernel_session->consumer) {
2669 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2670 cmd_ctx->session, nb_uri, uris);
2671 }
2672 }
2673
2674 free(uris);
2675
2676 break;
2677 }
2678 case LTTNG_START_TRACE:
2679 {
2680 ret = cmd_start_trace(cmd_ctx->session);
2681 break;
2682 }
2683 case LTTNG_STOP_TRACE:
2684 {
2685 ret = cmd_stop_trace(cmd_ctx->session);
2686 break;
2687 }
2688 case LTTNG_CREATE_SESSION:
2689 {
2690 size_t nb_uri, len;
2691 struct lttng_uri *uris = NULL;
2692
2693 nb_uri = cmd_ctx->lsm->u.uri.size;
2694 len = nb_uri * sizeof(struct lttng_uri);
2695
2696 if (nb_uri > 0) {
2697 uris = zmalloc(len);
2698 if (uris == NULL) {
2699 ret = LTTNG_ERR_FATAL;
2700 goto error;
2701 }
2702
2703 /* Receive variable len data */
2704 DBG("Waiting for %zu URIs from client ...", nb_uri);
2705 ret = lttcomm_recv_unix_sock(sock, uris, len);
2706 if (ret <= 0) {
2707 DBG("No URIs received from client... continuing");
2708 *sock_error = 1;
2709 ret = LTTNG_ERR_SESSION_FAIL;
2710 free(uris);
2711 goto error;
2712 }
2713
2714 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2715 DBG("Creating session with ONE network URI is a bad call");
2716 ret = LTTNG_ERR_SESSION_FAIL;
2717 free(uris);
2718 goto error;
2719 }
2720 }
2721
2722 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2723 &cmd_ctx->creds);
2724
2725 free(uris);
2726
2727 break;
2728 }
2729 case LTTNG_DESTROY_SESSION:
2730 {
2731 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2732
2733 /* Set session to NULL so we do not unlock it after free. */
2734 cmd_ctx->session = NULL;
2735 break;
2736 }
2737 case LTTNG_LIST_DOMAINS:
2738 {
2739 ssize_t nb_dom;
2740 struct lttng_domain *domains;
2741
2742 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2743 if (nb_dom < 0) {
2744 /* Return value is a negative lttng_error_code. */
2745 ret = -nb_dom;
2746 goto error;
2747 }
2748
2749 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2750 if (ret < 0) {
2751 goto setup_error;
2752 }
2753
2754 /* Copy event list into message payload */
2755 memcpy(cmd_ctx->llm->payload, domains,
2756 nb_dom * sizeof(struct lttng_domain));
2757
2758 free(domains);
2759
2760 ret = LTTNG_OK;
2761 break;
2762 }
2763 case LTTNG_LIST_CHANNELS:
2764 {
2765 int nb_chan;
2766 struct lttng_channel *channels;
2767
2768 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2769 cmd_ctx->session, &channels);
2770 if (nb_chan < 0) {
2771 /* Return value is a negative lttng_error_code. */
2772 ret = -nb_chan;
2773 goto error;
2774 }
2775
2776 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2777 if (ret < 0) {
2778 goto setup_error;
2779 }
2780
2781 /* Copy event list into message payload */
2782 memcpy(cmd_ctx->llm->payload, channels,
2783 nb_chan * sizeof(struct lttng_channel));
2784
2785 free(channels);
2786
2787 ret = LTTNG_OK;
2788 break;
2789 }
2790 case LTTNG_LIST_EVENTS:
2791 {
2792 ssize_t nb_event;
2793 struct lttng_event *events = NULL;
2794
2795 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2796 cmd_ctx->lsm->u.list.channel_name, &events);
2797 if (nb_event < 0) {
2798 /* Return value is a negative lttng_error_code. */
2799 ret = -nb_event;
2800 goto error;
2801 }
2802
2803 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2804 if (ret < 0) {
2805 goto setup_error;
2806 }
2807
2808 /* Copy event list into message payload */
2809 memcpy(cmd_ctx->llm->payload, events,
2810 nb_event * sizeof(struct lttng_event));
2811
2812 free(events);
2813
2814 ret = LTTNG_OK;
2815 break;
2816 }
2817 case LTTNG_LIST_SESSIONS:
2818 {
2819 unsigned int nr_sessions;
2820
2821 session_lock_list();
2822 nr_sessions = lttng_sessions_count(
2823 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2824 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2825
2826 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2827 if (ret < 0) {
2828 session_unlock_list();
2829 goto setup_error;
2830 }
2831
2832 /* Filled the session array */
2833 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2834 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2835 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2836
2837 session_unlock_list();
2838
2839 ret = LTTNG_OK;
2840 break;
2841 }
2842 case LTTNG_CALIBRATE:
2843 {
2844 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2845 &cmd_ctx->lsm->u.calibrate);
2846 break;
2847 }
2848 case LTTNG_REGISTER_CONSUMER:
2849 {
2850 struct consumer_data *cdata;
2851
2852 switch (cmd_ctx->lsm->domain.type) {
2853 case LTTNG_DOMAIN_KERNEL:
2854 cdata = &kconsumer_data;
2855 break;
2856 default:
2857 ret = LTTNG_ERR_UND;
2858 goto error;
2859 }
2860
2861 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2862 cmd_ctx->lsm->u.reg.path, cdata);
2863 break;
2864 }
2865 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2866 {
2867 struct lttng_filter_bytecode *bytecode;
2868
2869 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2870 ret = LTTNG_ERR_FILTER_INVAL;
2871 goto error;
2872 }
2873 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2874 ret = LTTNG_ERR_FILTER_INVAL;
2875 goto error;
2876 }
2877 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2878 if (!bytecode) {
2879 ret = LTTNG_ERR_FILTER_NOMEM;
2880 goto error;
2881 }
2882 /* Receive var. len. data */
2883 DBG("Receiving var len data from client ...");
2884 ret = lttcomm_recv_unix_sock(sock, bytecode,
2885 cmd_ctx->lsm->u.enable.bytecode_len);
2886 if (ret <= 0) {
2887 DBG("Nothing recv() from client var len data... continuing");
2888 *sock_error = 1;
2889 ret = LTTNG_ERR_FILTER_INVAL;
2890 goto error;
2891 }
2892
2893 if (bytecode->len + sizeof(*bytecode)
2894 != cmd_ctx->lsm->u.enable.bytecode_len) {
2895 free(bytecode);
2896 ret = LTTNG_ERR_FILTER_INVAL;
2897 goto error;
2898 }
2899
2900 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2901 cmd_ctx->lsm->u.enable.channel_name,
2902 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2903 break;
2904 }
2905 case LTTNG_DATA_PENDING:
2906 {
2907 ret = cmd_data_pending(cmd_ctx->session);
2908 break;
2909 }
2910 default:
2911 ret = LTTNG_ERR_UND;
2912 break;
2913 }
2914
2915 error:
2916 if (cmd_ctx->llm == NULL) {
2917 DBG("Missing llm structure. Allocating one.");
2918 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2919 goto setup_error;
2920 }
2921 }
2922 /* Set return code */
2923 cmd_ctx->llm->ret_code = ret;
2924 setup_error:
2925 if (cmd_ctx->session) {
2926 session_unlock(cmd_ctx->session);
2927 }
2928 if (need_tracing_session) {
2929 session_unlock_list();
2930 }
2931 init_setup_error:
2932 return ret;
2933 }
2934
2935 /*
2936 * Thread managing health check socket.
2937 */
2938 static void *thread_manage_health(void *data)
2939 {
2940 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2941 uint32_t revents, nb_fd;
2942 struct lttng_poll_event events;
2943 struct lttcomm_health_msg msg;
2944 struct lttcomm_health_data reply;
2945
2946 DBG("[thread] Manage health check started");
2947
2948 rcu_register_thread();
2949
2950 /* Create unix socket */
2951 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2952 if (sock < 0) {
2953 ERR("Unable to create health check Unix socket");
2954 ret = -1;
2955 goto error;
2956 }
2957
2958 /*
2959 * Set the CLOEXEC flag. Return code is useless because either way, the
2960 * show must go on.
2961 */
2962 (void) utils_set_fd_cloexec(sock);
2963
2964 ret = lttcomm_listen_unix_sock(sock);
2965 if (ret < 0) {
2966 goto error;
2967 }
2968
2969 /*
2970 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2971 * more will be added to this poll set.
2972 */
2973 ret = create_thread_poll_set(&events, 2);
2974 if (ret < 0) {
2975 goto error;
2976 }
2977
2978 /* Add the application registration socket */
2979 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2980 if (ret < 0) {
2981 goto error;
2982 }
2983
2984 while (1) {
2985 DBG("Health check ready");
2986
2987 /* Inifinite blocking call, waiting for transmission */
2988 restart:
2989 ret = lttng_poll_wait(&events, -1);
2990 if (ret < 0) {
2991 /*
2992 * Restart interrupted system call.
2993 */
2994 if (errno == EINTR) {
2995 goto restart;
2996 }
2997 goto error;
2998 }
2999
3000 nb_fd = ret;
3001
3002 for (i = 0; i < nb_fd; i++) {
3003 /* Fetch once the poll data */
3004 revents = LTTNG_POLL_GETEV(&events, i);
3005 pollfd = LTTNG_POLL_GETFD(&events, i);
3006
3007 /* Thread quit pipe has been closed. Killing thread. */
3008 ret = check_thread_quit_pipe(pollfd, revents);
3009 if (ret) {
3010 err = 0;
3011 goto exit;
3012 }
3013
3014 /* Event on the registration socket */
3015 if (pollfd == sock) {
3016 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3017 ERR("Health socket poll error");
3018 goto error;
3019 }
3020 }
3021 }
3022
3023 new_sock = lttcomm_accept_unix_sock(sock);
3024 if (new_sock < 0) {
3025 goto error;
3026 }
3027
3028 /*
3029 * Set the CLOEXEC flag. Return code is useless because either way, the
3030 * show must go on.
3031 */
3032 (void) utils_set_fd_cloexec(new_sock);
3033
3034 DBG("Receiving data from client for health...");
3035 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3036 if (ret <= 0) {
3037 DBG("Nothing recv() from client... continuing");
3038 ret = close(new_sock);
3039 if (ret) {
3040 PERROR("close");
3041 }
3042 new_sock = -1;
3043 continue;
3044 }
3045
3046 rcu_thread_online();
3047
3048 switch (msg.component) {
3049 case LTTNG_HEALTH_CMD:
3050 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3051 break;
3052 case LTTNG_HEALTH_APP_MANAGE:
3053 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3054 break;
3055 case LTTNG_HEALTH_APP_REG:
3056 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3057 break;
3058 case LTTNG_HEALTH_KERNEL:
3059 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3060 break;
3061 case LTTNG_HEALTH_CONSUMER:
3062 reply.ret_code = check_consumer_health();
3063 break;
3064 case LTTNG_HEALTH_ALL:
3065 reply.ret_code =
3066 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3067 health_check_state(HEALTH_TYPE_APP_REG) &&
3068 health_check_state(HEALTH_TYPE_CMD) &&
3069 health_check_state(HEALTH_TYPE_KERNEL) &&
3070 check_consumer_health();
3071 break;
3072 default:
3073 reply.ret_code = LTTNG_ERR_UND;
3074 break;
3075 }
3076
3077 /*
3078 * Flip ret value since 0 is a success and 1 indicates a bad health for
3079 * the client where in the sessiond it is the opposite. Again, this is
3080 * just to make things easier for us poor developer which enjoy a lot
3081 * lazyness.
3082 */
3083 if (reply.ret_code == 0 || reply.ret_code == 1) {
3084 reply.ret_code = !reply.ret_code;
3085 }
3086
3087 DBG2("Health check return value %d", reply.ret_code);
3088
3089 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3090 if (ret < 0) {
3091 ERR("Failed to send health data back to client");
3092 }
3093
3094 /* End of transmission */
3095 ret = close(new_sock);
3096 if (ret) {
3097 PERROR("close");
3098 }
3099 new_sock = -1;
3100 }
3101
3102 exit:
3103 error:
3104 if (err) {
3105 ERR("Health error occurred in %s", __func__);
3106 }
3107 DBG("Health check thread dying");
3108 unlink(health_unix_sock_path);
3109 if (sock >= 0) {
3110 ret = close(sock);
3111 if (ret) {
3112 PERROR("close");
3113 }
3114 }
3115 if (new_sock >= 0) {
3116 ret = close(new_sock);
3117 if (ret) {
3118 PERROR("close");
3119 }
3120 }
3121
3122 lttng_poll_clean(&events);
3123
3124 rcu_unregister_thread();
3125 return NULL;
3126 }
3127
3128 /*
3129 * This thread manage all clients request using the unix client socket for
3130 * communication.
3131 */
3132 static void *thread_manage_clients(void *data)
3133 {
3134 int sock = -1, ret, i, pollfd, err = -1;
3135 int sock_error;
3136 uint32_t revents, nb_fd;
3137 struct command_ctx *cmd_ctx = NULL;
3138 struct lttng_poll_event events;
3139
3140 DBG("[thread] Manage client started");
3141
3142 rcu_register_thread();
3143
3144 health_register(HEALTH_TYPE_CMD);
3145
3146 if (testpoint(thread_manage_clients)) {
3147 goto error_testpoint;
3148 }
3149
3150 health_code_update();
3151
3152 ret = lttcomm_listen_unix_sock(client_sock);
3153 if (ret < 0) {
3154 goto error_listen;
3155 }
3156
3157 /*
3158 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3159 * more will be added to this poll set.
3160 */
3161 ret = create_thread_poll_set(&events, 2);
3162 if (ret < 0) {
3163 goto error_create_poll;
3164 }
3165
3166 /* Add the application registration socket */
3167 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3168 if (ret < 0) {
3169 goto error;
3170 }
3171
3172 /*
3173 * Notify parent pid that we are ready to accept command for client side.
3174 */
3175 if (opt_sig_parent) {
3176 kill(ppid, SIGUSR1);
3177 }
3178
3179 if (testpoint(thread_manage_clients_before_loop)) {
3180 goto error;
3181 }
3182
3183 health_code_update();
3184
3185 while (1) {
3186 DBG("Accepting client command ...");
3187
3188 /* Inifinite blocking call, waiting for transmission */
3189 restart:
3190 health_poll_entry();
3191 ret = lttng_poll_wait(&events, -1);
3192 health_poll_exit();
3193 if (ret < 0) {
3194 /*
3195 * Restart interrupted system call.
3196 */
3197 if (errno == EINTR) {
3198 goto restart;
3199 }
3200 goto error;
3201 }
3202
3203 nb_fd = ret;
3204
3205 for (i = 0; i < nb_fd; i++) {
3206 /* Fetch once the poll data */
3207 revents = LTTNG_POLL_GETEV(&events, i);
3208 pollfd = LTTNG_POLL_GETFD(&events, i);
3209
3210 health_code_update();
3211
3212 /* Thread quit pipe has been closed. Killing thread. */
3213 ret = check_thread_quit_pipe(pollfd, revents);
3214 if (ret) {
3215 err = 0;
3216 goto exit;
3217 }
3218
3219 /* Event on the registration socket */
3220 if (pollfd == client_sock) {
3221 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3222 ERR("Client socket poll error");
3223 goto error;
3224 }
3225 }
3226 }
3227
3228 DBG("Wait for client response");
3229
3230 health_code_update();
3231
3232 sock = lttcomm_accept_unix_sock(client_sock);
3233 if (sock < 0) {
3234 goto error;
3235 }
3236
3237 /*
3238 * Set the CLOEXEC flag. Return code is useless because either way, the
3239 * show must go on.
3240 */
3241 (void) utils_set_fd_cloexec(sock);
3242
3243 /* Set socket option for credentials retrieval */
3244 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3245 if (ret < 0) {
3246 goto error;
3247 }
3248
3249 /* Allocate context command to process the client request */
3250 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3251 if (cmd_ctx == NULL) {
3252 PERROR("zmalloc cmd_ctx");
3253 goto error;
3254 }
3255
3256 /* Allocate data buffer for reception */
3257 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3258 if (cmd_ctx->lsm == NULL) {
3259 PERROR("zmalloc cmd_ctx->lsm");
3260 goto error;
3261 }
3262
3263 cmd_ctx->llm = NULL;
3264 cmd_ctx->session = NULL;
3265
3266 health_code_update();
3267
3268 /*
3269 * Data is received from the lttng client. The struct
3270 * lttcomm_session_msg (lsm) contains the command and data request of
3271 * the client.
3272 */
3273 DBG("Receiving data from client ...");
3274 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3275 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3276 if (ret <= 0) {
3277 DBG("Nothing recv() from client... continuing");
3278 ret = close(sock);
3279 if (ret) {
3280 PERROR("close");
3281 }
3282 sock = -1;
3283 clean_command_ctx(&cmd_ctx);
3284 continue;
3285 }
3286
3287 health_code_update();
3288
3289 // TODO: Validate cmd_ctx including sanity check for
3290 // security purpose.
3291
3292 rcu_thread_online();
3293 /*
3294 * This function dispatch the work to the kernel or userspace tracer
3295 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3296 * informations for the client. The command context struct contains
3297 * everything this function may needs.
3298 */
3299 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3300 rcu_thread_offline();
3301 if (ret < 0) {
3302 if (sock_error) {
3303 ret = close(sock);
3304 if (ret) {
3305 PERROR("close");
3306 }
3307 sock = -1;
3308 }
3309 /*
3310 * TODO: Inform client somehow of the fatal error. At
3311 * this point, ret < 0 means that a zmalloc failed
3312 * (ENOMEM). Error detected but still accept
3313 * command, unless a socket error has been
3314 * detected.
3315 */
3316 clean_command_ctx(&cmd_ctx);
3317 continue;
3318 }
3319
3320 health_code_update();
3321
3322 DBG("Sending response (size: %d, retcode: %s)",
3323 cmd_ctx->lttng_msg_size,
3324 lttng_strerror(-cmd_ctx->llm->ret_code));
3325 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3326 if (ret < 0) {
3327 ERR("Failed to send data back to client");
3328 }
3329
3330 /* End of transmission */
3331 ret = close(sock);
3332 if (ret) {
3333 PERROR("close");
3334 }
3335 sock = -1;
3336
3337 clean_command_ctx(&cmd_ctx);
3338
3339 health_code_update();
3340 }
3341
3342 exit:
3343 error:
3344 if (sock >= 0) {
3345 ret = close(sock);
3346 if (ret) {
3347 PERROR("close");
3348 }
3349 }
3350
3351 lttng_poll_clean(&events);
3352 clean_command_ctx(&cmd_ctx);
3353
3354 error_listen:
3355 error_create_poll:
3356 error_testpoint:
3357 unlink(client_unix_sock_path);
3358 if (client_sock >= 0) {
3359 ret = close(client_sock);
3360 if (ret) {
3361 PERROR("close");
3362 }
3363 }
3364
3365 if (err) {
3366 health_error();
3367 ERR("Health error occurred in %s", __func__);
3368 }
3369
3370 health_unregister();
3371
3372 DBG("Client thread dying");
3373
3374 rcu_unregister_thread();
3375 return NULL;
3376 }
3377
3378
3379 /*
3380 * usage function on stderr
3381 */
3382 static void usage(void)
3383 {
3384 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3385 fprintf(stderr, " -h, --help Display this usage.\n");
3386 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3387 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3388 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3389 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3390 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3391 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3392 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3393 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3394 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3395 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3396 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3397 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3398 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3399 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3400 fprintf(stderr, " -V, --version Show version number.\n");
3401 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3402 fprintf(stderr, " -q, --quiet No output at all.\n");
3403 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3404 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3405 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3406 }
3407
3408 /*
3409 * daemon argument parsing
3410 */
3411 static int parse_args(int argc, char **argv)
3412 {
3413 int c;
3414
3415 static struct option long_options[] = {
3416 { "client-sock", 1, 0, 'c' },
3417 { "apps-sock", 1, 0, 'a' },
3418 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3419 { "kconsumerd-err-sock", 1, 0, 'E' },
3420 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3421 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3422 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3423 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3424 { "consumerd32-path", 1, 0, 'u' },
3425 { "consumerd32-libdir", 1, 0, 'U' },
3426 { "consumerd64-path", 1, 0, 't' },
3427 { "consumerd64-libdir", 1, 0, 'T' },
3428 { "daemonize", 0, 0, 'd' },
3429 { "sig-parent", 0, 0, 'S' },
3430 { "help", 0, 0, 'h' },
3431 { "group", 1, 0, 'g' },
3432 { "version", 0, 0, 'V' },
3433 { "quiet", 0, 0, 'q' },
3434 { "verbose", 0, 0, 'v' },
3435 { "verbose-consumer", 0, 0, 'Z' },
3436 { "no-kernel", 0, 0, 'N' },
3437 { NULL, 0, 0, 0 }
3438 };
3439
3440 while (1) {
3441 int option_index = 0;
3442 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3443 long_options, &option_index);
3444 if (c == -1) {
3445 break;
3446 }
3447
3448 switch (c) {
3449 case 0:
3450 fprintf(stderr, "option %s", long_options[option_index].name);
3451 if (optarg) {
3452 fprintf(stderr, " with arg %s\n", optarg);
3453 }
3454 break;
3455 case 'c':
3456 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3457 break;
3458 case 'a':
3459 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3460 break;
3461 case 'd':
3462 opt_daemon = 1;
3463 break;
3464 case 'g':
3465 opt_tracing_group = optarg;
3466 break;
3467 case 'h':
3468 usage();
3469 exit(EXIT_FAILURE);
3470 case 'V':
3471 fprintf(stdout, "%s\n", VERSION);
3472 exit(EXIT_SUCCESS);
3473 case 'S':
3474 opt_sig_parent = 1;
3475 break;
3476 case 'E':
3477 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3478 break;
3479 case 'C':
3480 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3481 break;
3482 case 'F':
3483 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3484 break;
3485 case 'D':
3486 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3487 break;
3488 case 'H':
3489 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3490 break;
3491 case 'G':
3492 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3493 break;
3494 case 'N':
3495 opt_no_kernel = 1;
3496 break;
3497 case 'q':
3498 lttng_opt_quiet = 1;
3499 break;
3500 case 'v':
3501 /* Verbose level can increase using multiple -v */
3502 lttng_opt_verbose += 1;
3503 break;
3504 case 'Z':
3505 opt_verbose_consumer += 1;
3506 break;
3507 case 'u':
3508 consumerd32_bin= optarg;
3509 break;
3510 case 'U':
3511 consumerd32_libdir = optarg;
3512 break;
3513 case 't':
3514 consumerd64_bin = optarg;
3515 break;
3516 case 'T':
3517 consumerd64_libdir = optarg;
3518 break;
3519 default:
3520 /* Unknown option or other error.
3521 * Error is printed by getopt, just return */
3522 return -1;
3523 }
3524 }
3525
3526 return 0;
3527 }
3528
3529 /*
3530 * Creates the two needed socket by the daemon.
3531 * apps_sock - The communication socket for all UST apps.
3532 * client_sock - The communication of the cli tool (lttng).
3533 */
3534 static int init_daemon_socket(void)
3535 {
3536 int ret = 0;
3537 mode_t old_umask;
3538
3539 old_umask = umask(0);
3540
3541 /* Create client tool unix socket */
3542 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3543 if (client_sock < 0) {
3544 ERR("Create unix sock failed: %s", client_unix_sock_path);
3545 ret = -1;
3546 goto end;
3547 }
3548
3549 /* Set the cloexec flag */
3550 ret = utils_set_fd_cloexec(client_sock);
3551 if (ret < 0) {
3552 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3553 "Continuing but note that the consumer daemon will have a "
3554 "reference to this socket on exec()", client_sock);
3555 }
3556
3557 /* File permission MUST be 660 */
3558 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3559 if (ret < 0) {
3560 ERR("Set file permissions failed: %s", client_unix_sock_path);
3561 PERROR("chmod");
3562 goto end;
3563 }
3564
3565 /* Create the application unix socket */
3566 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3567 if (apps_sock < 0) {
3568 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3569 ret = -1;
3570 goto end;
3571 }
3572
3573 /* Set the cloexec flag */
3574 ret = utils_set_fd_cloexec(apps_sock);
3575 if (ret < 0) {
3576 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3577 "Continuing but note that the consumer daemon will have a "
3578 "reference to this socket on exec()", apps_sock);
3579 }
3580
3581 /* File permission MUST be 666 */
3582 ret = chmod(apps_unix_sock_path,
3583 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3584 if (ret < 0) {
3585 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3586 PERROR("chmod");
3587 goto end;
3588 }
3589
3590 DBG3("Session daemon client socket %d and application socket %d created",
3591 client_sock, apps_sock);
3592
3593 end:
3594 umask(old_umask);
3595 return ret;
3596 }
3597
3598 /*
3599 * Check if the global socket is available, and if a daemon is answering at the
3600 * other side. If yes, error is returned.
3601 */
3602 static int check_existing_daemon(void)
3603 {
3604 /* Is there anybody out there ? */
3605 if (lttng_session_daemon_alive()) {
3606 return -EEXIST;
3607 }
3608
3609 return 0;
3610 }
3611
3612 /*
3613 * Set the tracing group gid onto the client socket.
3614 *
3615 * Race window between mkdir and chown is OK because we are going from more
3616 * permissive (root.root) to less permissive (root.tracing).
3617 */
3618 static int set_permissions(char *rundir)
3619 {
3620 int ret;
3621 gid_t gid;
3622
3623 ret = allowed_group();
3624 if (ret < 0) {
3625 WARN("No tracing group detected");
3626 ret = 0;
3627 goto end;
3628 }
3629
3630 gid = ret;
3631
3632 /* Set lttng run dir */
3633 ret = chown(rundir, 0, gid);
3634 if (ret < 0) {
3635 ERR("Unable to set group on %s", rundir);
3636 PERROR("chown");
3637 }
3638
3639 /* Ensure tracing group can search the run dir */
3640 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3641 if (ret < 0) {
3642 ERR("Unable to set permissions on %s", rundir);
3643 PERROR("chmod");
3644 }
3645
3646 /* lttng client socket path */
3647 ret = chown(client_unix_sock_path, 0, gid);
3648 if (ret < 0) {
3649 ERR("Unable to set group on %s", client_unix_sock_path);
3650 PERROR("chown");
3651 }
3652
3653 /* kconsumer error socket path */
3654 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3655 if (ret < 0) {
3656 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3657 PERROR("chown");
3658 }
3659
3660 /* 64-bit ustconsumer error socket path */
3661 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3662 if (ret < 0) {
3663 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3664 PERROR("chown");
3665 }
3666
3667 /* 32-bit ustconsumer compat32 error socket path */
3668 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3669 if (ret < 0) {
3670 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3671 PERROR("chown");
3672 }
3673
3674 DBG("All permissions are set");
3675
3676 end:
3677 return ret;
3678 }
3679
3680 /*
3681 * Create the lttng run directory needed for all global sockets and pipe.
3682 */
3683 static int create_lttng_rundir(const char *rundir)
3684 {
3685 int ret;
3686
3687 DBG3("Creating LTTng run directory: %s", rundir);
3688
3689 ret = mkdir(rundir, S_IRWXU);
3690 if (ret < 0) {
3691 if (errno != EEXIST) {
3692 ERR("Unable to create %s", rundir);
3693 goto error;
3694 } else {
3695 ret = 0;
3696 }
3697 }
3698
3699 error:
3700 return ret;
3701 }
3702
3703 /*
3704 * Setup sockets and directory needed by the kconsumerd communication with the
3705 * session daemon.
3706 */
3707 static int set_consumer_sockets(struct consumer_data *consumer_data,
3708 const char *rundir)
3709 {
3710 int ret;
3711 char path[PATH_MAX];
3712
3713 switch (consumer_data->type) {
3714 case LTTNG_CONSUMER_KERNEL:
3715 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3716 break;
3717 case LTTNG_CONSUMER64_UST:
3718 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3719 break;
3720 case LTTNG_CONSUMER32_UST:
3721 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3722 break;
3723 default:
3724 ERR("Consumer type unknown");
3725 ret = -EINVAL;
3726 goto error;
3727 }
3728
3729 DBG2("Creating consumer directory: %s", path);
3730
3731 ret = mkdir(path, S_IRWXU);
3732 if (ret < 0) {
3733 if (errno != EEXIST) {
3734 PERROR("mkdir");
3735 ERR("Failed to create %s", path);
3736 goto error;
3737 }
3738 ret = -1;
3739 }
3740
3741 /* Create the kconsumerd error unix socket */
3742 consumer_data->err_sock =
3743 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3744 if (consumer_data->err_sock < 0) {
3745 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3746 ret = -1;
3747 goto error;
3748 }
3749
3750 /* File permission MUST be 660 */
3751 ret = chmod(consumer_data->err_unix_sock_path,
3752 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3753 if (ret < 0) {
3754 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3755 PERROR("chmod");
3756 goto error;
3757 }
3758
3759 error:
3760 return ret;
3761 }
3762
3763 /*
3764 * Signal handler for the daemon
3765 *
3766 * Simply stop all worker threads, leaving main() return gracefully after
3767 * joining all threads and calling cleanup().
3768 */
3769 static void sighandler(int sig)
3770 {
3771 switch (sig) {
3772 case SIGPIPE:
3773 DBG("SIGPIPE caught");
3774 return;
3775 case SIGINT:
3776 DBG("SIGINT caught");
3777 stop_threads();
3778 break;
3779 case SIGTERM:
3780 DBG("SIGTERM caught");
3781 stop_threads();
3782 break;
3783 default:
3784 break;
3785 }
3786 }
3787
3788 /*
3789 * Setup signal handler for :
3790 * SIGINT, SIGTERM, SIGPIPE
3791 */
3792 static int set_signal_handler(void)
3793 {
3794 int ret = 0;
3795 struct sigaction sa;
3796 sigset_t sigset;
3797
3798 if ((ret = sigemptyset(&sigset)) < 0) {
3799 PERROR("sigemptyset");
3800 return ret;
3801 }
3802
3803 sa.sa_handler = sighandler;
3804 sa.sa_mask = sigset;
3805 sa.sa_flags = 0;
3806 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3807 PERROR("sigaction");
3808 return ret;
3809 }
3810
3811 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3812 PERROR("sigaction");
3813 return ret;
3814 }
3815
3816 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3817 PERROR("sigaction");
3818 return ret;
3819 }
3820
3821 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3822
3823 return ret;
3824 }
3825
3826 /*
3827 * Set open files limit to unlimited. This daemon can open a large number of
3828 * file descriptors in order to consumer multiple kernel traces.
3829 */
3830 static void set_ulimit(void)
3831 {
3832 int ret;
3833 struct rlimit lim;
3834
3835 /* The kernel does not allowed an infinite limit for open files */
3836 lim.rlim_cur = 65535;
3837 lim.rlim_max = 65535;
3838
3839 ret = setrlimit(RLIMIT_NOFILE, &lim);
3840 if (ret < 0) {
3841 PERROR("failed to set open files limit");
3842 }
3843 }
3844
3845 /*
3846 * main
3847 */
3848 int main(int argc, char **argv)
3849 {
3850 int ret = 0;
3851 void *status;
3852 const char *home_path, *env_app_timeout;
3853
3854 init_kernel_workarounds();
3855
3856 rcu_register_thread();
3857
3858 setup_consumerd_path();
3859
3860 /* Parse arguments */
3861 progname = argv[0];
3862 if ((ret = parse_args(argc, argv)) < 0) {
3863 goto error;
3864 }
3865
3866 /* Daemonize */
3867 if (opt_daemon) {
3868 int i;
3869
3870 /*
3871 * fork
3872 * child: setsid, close FD 0, 1, 2, chdir /
3873 * parent: exit (if fork is successful)
3874 */
3875 ret = daemon(0, 0);
3876 if (ret < 0) {
3877 PERROR("daemon");
3878 goto error;
3879 }
3880 /*
3881 * We are in the child. Make sure all other file
3882 * descriptors are closed, in case we are called with
3883 * more opened file descriptors than the standard ones.
3884 */
3885 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3886 (void) close(i);
3887 }
3888 }
3889
3890 /* Create thread quit pipe */
3891 if ((ret = init_thread_quit_pipe()) < 0) {
3892 goto error;
3893 }
3894
3895 /* Check if daemon is UID = 0 */
3896 is_root = !getuid();
3897
3898 if (is_root) {
3899 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3900
3901 /* Create global run dir with root access */
3902 ret = create_lttng_rundir(rundir);
3903 if (ret < 0) {
3904 goto error;
3905 }
3906
3907 if (strlen(apps_unix_sock_path) == 0) {
3908 snprintf(apps_unix_sock_path, PATH_MAX,
3909 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3910 }
3911
3912 if (strlen(client_unix_sock_path) == 0) {
3913 snprintf(client_unix_sock_path, PATH_MAX,
3914 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3915 }
3916
3917 /* Set global SHM for ust */
3918 if (strlen(wait_shm_path) == 0) {
3919 snprintf(wait_shm_path, PATH_MAX,
3920 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3921 }
3922
3923 if (strlen(health_unix_sock_path) == 0) {
3924 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3925 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3926 }
3927
3928 /* Setup kernel consumerd path */
3929 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3930 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3931 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3932 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3933
3934 DBG2("Kernel consumer err path: %s",
3935 kconsumer_data.err_unix_sock_path);
3936 DBG2("Kernel consumer cmd path: %s",
3937 kconsumer_data.cmd_unix_sock_path);
3938 } else {
3939 home_path = get_home_dir();
3940 if (home_path == NULL) {
3941 /* TODO: Add --socket PATH option */
3942 ERR("Can't get HOME directory for sockets creation.");
3943 ret = -EPERM;
3944 goto error;
3945 }
3946
3947 /*
3948 * Create rundir from home path. This will create something like
3949 * $HOME/.lttng
3950 */
3951 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3952 if (ret < 0) {
3953 ret = -ENOMEM;
3954 goto error;
3955 }
3956
3957 ret = create_lttng_rundir(rundir);
3958 if (ret < 0) {
3959 goto error;
3960 }
3961
3962 if (strlen(apps_unix_sock_path) == 0) {
3963 snprintf(apps_unix_sock_path, PATH_MAX,
3964 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3965 }
3966
3967 /* Set the cli tool unix socket path */
3968 if (strlen(client_unix_sock_path) == 0) {
3969 snprintf(client_unix_sock_path, PATH_MAX,
3970 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3971 }
3972
3973 /* Set global SHM for ust */
3974 if (strlen(wait_shm_path) == 0) {
3975 snprintf(wait_shm_path, PATH_MAX,
3976 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3977 }
3978
3979 /* Set health check Unix path */
3980 if (strlen(health_unix_sock_path) == 0) {
3981 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3982 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3983 }
3984 }
3985
3986 /* Set consumer initial state */
3987 kernel_consumerd_state = CONSUMER_STOPPED;
3988 ust_consumerd_state = CONSUMER_STOPPED;
3989
3990 DBG("Client socket path %s", client_unix_sock_path);
3991 DBG("Application socket path %s", apps_unix_sock_path);
3992 DBG("LTTng run directory path: %s", rundir);
3993
3994 /* 32 bits consumerd path setup */
3995 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3996 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3997 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3998 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3999
4000 DBG2("UST consumer 32 bits err path: %s",
4001 ustconsumer32_data.err_unix_sock_path);
4002 DBG2("UST consumer 32 bits cmd path: %s",
4003 ustconsumer32_data.cmd_unix_sock_path);
4004
4005 /* 64 bits consumerd path setup */
4006 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4007 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4008 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4009 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4010
4011 DBG2("UST consumer 64 bits err path: %s",
4012 ustconsumer64_data.err_unix_sock_path);
4013 DBG2("UST consumer 64 bits cmd path: %s",
4014 ustconsumer64_data.cmd_unix_sock_path);
4015
4016 /*
4017 * See if daemon already exist.
4018 */
4019 if ((ret = check_existing_daemon()) < 0) {
4020 ERR("Already running daemon.\n");
4021 /*
4022 * We do not goto exit because we must not cleanup()
4023 * because a daemon is already running.
4024 */
4025 goto error;
4026 }
4027
4028 /*
4029 * Init UST app hash table. Alloc hash table before this point since
4030 * cleanup() can get called after that point.
4031 */
4032 ust_app_ht_alloc();
4033
4034 /* After this point, we can safely call cleanup() with "goto exit" */
4035
4036 /*
4037 * These actions must be executed as root. We do that *after* setting up
4038 * the sockets path because we MUST make the check for another daemon using
4039 * those paths *before* trying to set the kernel consumer sockets and init
4040 * kernel tracer.
4041 */
4042 if (is_root) {
4043 ret = set_consumer_sockets(&kconsumer_data, rundir);
4044 if (ret < 0) {
4045 goto exit;
4046 }
4047
4048 /* Setup kernel tracer */
4049 if (!opt_no_kernel) {
4050 init_kernel_tracer();
4051 }
4052
4053 /* Set ulimit for open files */
4054 set_ulimit();
4055 }
4056 /* init lttng_fd tracking must be done after set_ulimit. */
4057 lttng_fd_init();
4058
4059 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4060 if (ret < 0) {
4061 goto exit;
4062 }
4063
4064 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4065 if (ret < 0) {
4066 goto exit;
4067 }
4068
4069 if ((ret = set_signal_handler()) < 0) {
4070 goto exit;
4071 }
4072
4073 /* Setup the needed unix socket */
4074 if ((ret = init_daemon_socket()) < 0) {
4075 goto exit;
4076 }
4077
4078 /* Set credentials to socket */
4079 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4080 goto exit;
4081 }
4082
4083 /* Get parent pid if -S, --sig-parent is specified. */
4084 if (opt_sig_parent) {
4085 ppid = getppid();
4086 }
4087
4088 /* Setup the kernel pipe for waking up the kernel thread */
4089 if (is_root && !opt_no_kernel) {
4090 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4091 goto exit;
4092 }
4093 }
4094
4095 /* Setup the thread apps communication pipe. */
4096 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4097 goto exit;
4098 }
4099
4100 /* Init UST command queue. */
4101 cds_wfq_init(&ust_cmd_queue.queue);
4102
4103 /*
4104 * Get session list pointer. This pointer MUST NOT be free(). This list is
4105 * statically declared in session.c
4106 */
4107 session_list_ptr = session_get_list();
4108
4109 /* Set up max poll set size */
4110 lttng_poll_set_max_size();
4111
4112 cmd_init();
4113
4114 /* Check for the application socket timeout env variable. */
4115 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4116 if (env_app_timeout) {
4117 app_socket_timeout = atoi(env_app_timeout);
4118 } else {
4119 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4120 }
4121
4122 /* Create thread to manage the client socket */
4123 ret = pthread_create(&health_thread, NULL,
4124 thread_manage_health, (void *) NULL);
4125 if (ret != 0) {
4126 PERROR("pthread_create health");
4127 goto exit_health;
4128 }
4129
4130 /* Create thread to manage the client socket */
4131 ret = pthread_create(&client_thread, NULL,
4132 thread_manage_clients, (void *) NULL);
4133 if (ret != 0) {
4134 PERROR("pthread_create clients");
4135 goto exit_client;
4136 }
4137
4138 /* Create thread to dispatch registration */
4139 ret = pthread_create(&dispatch_thread, NULL,
4140 thread_dispatch_ust_registration, (void *) NULL);
4141 if (ret != 0) {
4142 PERROR("pthread_create dispatch");
4143 goto exit_dispatch;
4144 }
4145
4146 /* Create thread to manage application registration. */
4147 ret = pthread_create(&reg_apps_thread, NULL,
4148 thread_registration_apps, (void *) NULL);
4149 if (ret != 0) {
4150 PERROR("pthread_create registration");
4151 goto exit_reg_apps;
4152 }
4153
4154 /* Create thread to manage application socket */
4155 ret = pthread_create(&apps_thread, NULL,
4156 thread_manage_apps, (void *) NULL);
4157 if (ret != 0) {
4158 PERROR("pthread_create apps");
4159 goto exit_apps;
4160 }
4161
4162 /* Don't start this thread if kernel tracing is not requested nor root */
4163 if (is_root && !opt_no_kernel) {
4164 /* Create kernel thread to manage kernel event */
4165 ret = pthread_create(&kernel_thread, NULL,
4166 thread_manage_kernel, (void *) NULL);
4167 if (ret != 0) {
4168 PERROR("pthread_create kernel");
4169 goto exit_kernel;
4170 }
4171
4172 ret = pthread_join(kernel_thread, &status);
4173 if (ret != 0) {
4174 PERROR("pthread_join");
4175 goto error; /* join error, exit without cleanup */
4176 }
4177 }
4178
4179 exit_kernel:
4180 ret = pthread_join(apps_thread, &status);
4181 if (ret != 0) {
4182 PERROR("pthread_join");
4183 goto error; /* join error, exit without cleanup */
4184 }
4185
4186 exit_apps:
4187 ret = pthread_join(reg_apps_thread, &status);
4188 if (ret != 0) {
4189 PERROR("pthread_join");
4190 goto error; /* join error, exit without cleanup */
4191 }
4192
4193 exit_reg_apps:
4194 ret = pthread_join(dispatch_thread, &status);
4195 if (ret != 0) {
4196 PERROR("pthread_join");
4197 goto error; /* join error, exit without cleanup */
4198 }
4199
4200 exit_dispatch:
4201 ret = pthread_join(client_thread, &status);
4202 if (ret != 0) {
4203 PERROR("pthread_join");
4204 goto error; /* join error, exit without cleanup */
4205 }
4206
4207 ret = join_consumer_thread(&kconsumer_data);
4208 if (ret != 0) {
4209 PERROR("join_consumer");
4210 goto error; /* join error, exit without cleanup */
4211 }
4212
4213 ret = join_consumer_thread(&ustconsumer32_data);
4214 if (ret != 0) {
4215 PERROR("join_consumer ust32");
4216 goto error; /* join error, exit without cleanup */
4217 }
4218
4219 ret = join_consumer_thread(&ustconsumer64_data);
4220 if (ret != 0) {
4221 PERROR("join_consumer ust64");
4222 goto error; /* join error, exit without cleanup */
4223 }
4224
4225 exit_client:
4226 ret = pthread_join(health_thread, &status);
4227 if (ret != 0) {
4228 PERROR("pthread_join health thread");
4229 goto error; /* join error, exit without cleanup */
4230 }
4231
4232 exit_health:
4233 exit:
4234 /*
4235 * cleanup() is called when no other thread is running.
4236 */
4237 rcu_thread_online();
4238 cleanup();
4239 rcu_thread_offline();
4240 rcu_unregister_thread();
4241 if (!ret) {
4242 exit(EXIT_SUCCESS);
4243 }
4244 error:
4245 exit(EXIT_FAILURE);
4246 }
This page took 0.118828 seconds and 4 git commands to generate.