Deprecate enable/disable-consumer
[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 const char *opt_pidfile;
76 static int opt_sig_parent;
77 static int opt_verbose_consumer;
78 static int opt_daemon;
79 static int opt_no_kernel;
80 static int is_root; /* Set to 1 if the daemon is running as root */
81 static pid_t ppid; /* Parent PID for --sig-parent option */
82 static char *rundir;
83
84 /*
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
87 */
88 static struct consumer_data kconsumer_data = {
89 .type = LTTNG_CONSUMER_KERNEL,
90 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
91 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
92 .err_sock = -1,
93 .cmd_sock = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer32_data = {
111 .type = LTTNG_CONSUMER32_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121
122 /* Shared between threads */
123 static int dispatch_thread_exit;
124
125 /* Global application Unix socket path */
126 static char apps_unix_sock_path[PATH_MAX];
127 /* Global client Unix socket path */
128 static char client_unix_sock_path[PATH_MAX];
129 /* global wait shm path for UST */
130 static char wait_shm_path[PATH_MAX];
131 /* Global health check unix path */
132 static char health_unix_sock_path[PATH_MAX];
133
134 /* Sockets and FDs */
135 static int client_sock = -1;
136 static int apps_sock = -1;
137 int kernel_tracer_fd = -1;
138 static int kernel_poll_pipe[2] = { -1, -1 };
139
140 /*
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
143 */
144 static int thread_quit_pipe[2] = { -1, -1 };
145
146 /*
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
149 */
150 static int apps_cmd_pipe[2] = { -1, -1 };
151
152 /* Pthread, Mutexes and Semaphores */
153 static pthread_t apps_thread;
154 static pthread_t reg_apps_thread;
155 static pthread_t client_thread;
156 static pthread_t kernel_thread;
157 static pthread_t dispatch_thread;
158 static pthread_t health_thread;
159
160 /*
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 *
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
166 */
167 static struct ust_cmd_queue ust_cmd_queue;
168
169 /*
170 * Pointer initialized before thread creation.
171 *
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
175 *
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
178 */
179 static struct ltt_session_list *session_list_ptr;
180
181 int ust_consumerd64_fd = -1;
182 int ust_consumerd32_fd = -1;
183
184 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
185 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
186 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
187 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
188
189 static const char *module_proc_lttng = "/proc/lttng";
190
191 /*
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
194 */
195 enum consumerd_state {
196 CONSUMER_STARTED = 1,
197 CONSUMER_STOPPED = 2,
198 CONSUMER_ERROR = 3,
199 };
200
201 /*
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
206 *
207 * The following example shows a possible race condition of this scheme:
208 *
209 * consumer thread error happens
210 * client cmd arrives
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
214 * ...
215 *
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
221 */
222 static enum consumerd_state ust_consumerd_state;
223 static enum consumerd_state kernel_consumerd_state;
224
225 /*
226 * Socket timeout for receiving and sending in seconds.
227 */
228 static int app_socket_timeout;
229
230 static
231 void setup_consumerd_path(void)
232 {
233 const char *bin, *libdir;
234
235 /*
236 * Allow INSTALL_BIN_PATH to be used as a target path for the
237 * native architecture size consumer if CONFIG_CONSUMER*_PATH
238 * has not been defined.
239 */
240 #if (CAA_BITS_PER_LONG == 32)
241 if (!consumerd32_bin[0]) {
242 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
243 }
244 if (!consumerd32_libdir[0]) {
245 consumerd32_libdir = INSTALL_LIB_PATH;
246 }
247 #elif (CAA_BITS_PER_LONG == 64)
248 if (!consumerd64_bin[0]) {
249 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
250 }
251 if (!consumerd64_libdir[0]) {
252 consumerd64_libdir = INSTALL_LIB_PATH;
253 }
254 #else
255 #error "Unknown bitness"
256 #endif
257
258 /*
259 * runtime env. var. overrides the build default.
260 */
261 bin = getenv("LTTNG_CONSUMERD32_BIN");
262 if (bin) {
263 consumerd32_bin = bin;
264 }
265 bin = getenv("LTTNG_CONSUMERD64_BIN");
266 if (bin) {
267 consumerd64_bin = bin;
268 }
269 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
270 if (libdir) {
271 consumerd32_libdir = libdir;
272 }
273 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
274 if (libdir) {
275 consumerd64_libdir = libdir;
276 }
277 }
278
279 /*
280 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
281 */
282 static int create_thread_poll_set(struct lttng_poll_event *events,
283 unsigned int size)
284 {
285 int ret;
286
287 if (events == NULL || size == 0) {
288 ret = -1;
289 goto error;
290 }
291
292 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
293 if (ret < 0) {
294 goto error;
295 }
296
297 /* Add quit pipe */
298 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
299 if (ret < 0) {
300 goto error;
301 }
302
303 return 0;
304
305 error:
306 return ret;
307 }
308
309 /*
310 * Check if the thread quit pipe was triggered.
311 *
312 * Return 1 if it was triggered else 0;
313 */
314 static int check_thread_quit_pipe(int fd, uint32_t events)
315 {
316 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
317 return 1;
318 }
319
320 return 0;
321 }
322
323 /*
324 * Return group ID of the tracing group or -1 if not found.
325 */
326 static gid_t allowed_group(void)
327 {
328 struct group *grp;
329
330 if (opt_tracing_group) {
331 grp = getgrnam(opt_tracing_group);
332 } else {
333 grp = getgrnam(default_tracing_group);
334 }
335 if (!grp) {
336 return -1;
337 } else {
338 return grp->gr_gid;
339 }
340 }
341
342 /*
343 * Init thread quit pipe.
344 *
345 * Return -1 on error or 0 if all pipes are created.
346 */
347 static int init_thread_quit_pipe(void)
348 {
349 int ret, i;
350
351 ret = pipe(thread_quit_pipe);
352 if (ret < 0) {
353 PERROR("thread quit pipe");
354 goto error;
355 }
356
357 for (i = 0; i < 2; i++) {
358 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
359 if (ret < 0) {
360 PERROR("fcntl");
361 goto error;
362 }
363 }
364
365 error:
366 return ret;
367 }
368
369 /*
370 * Stop all threads by closing the thread quit pipe.
371 */
372 static void stop_threads(void)
373 {
374 int ret;
375
376 /* Stopping all threads */
377 DBG("Terminating all threads");
378 ret = notify_thread_pipe(thread_quit_pipe[1]);
379 if (ret < 0) {
380 ERR("write error on thread quit pipe");
381 }
382
383 /* Dispatch thread */
384 CMM_STORE_SHARED(dispatch_thread_exit, 1);
385 futex_nto1_wake(&ust_cmd_queue.futex);
386 }
387
388 /*
389 * Cleanup the daemon
390 */
391 static void cleanup(void)
392 {
393 int ret;
394 char *cmd = NULL;
395 struct ltt_session *sess, *stmp;
396
397 DBG("Cleaning up");
398
399 /* First thing first, stop all threads */
400 utils_close_pipe(thread_quit_pipe);
401
402 /*
403 * If opt_pidfile is undefined, the default file will be wiped when
404 * removing the rundir.
405 */
406 if (opt_pidfile) {
407 ret = remove(opt_pidfile);
408 if (ret < 0) {
409 PERROR("remove pidfile %s", opt_pidfile);
410 }
411 }
412
413 DBG("Removing %s directory", rundir);
414 ret = asprintf(&cmd, "rm -rf %s", rundir);
415 if (ret < 0) {
416 ERR("asprintf failed. Something is really wrong!");
417 }
418
419 /* Remove lttng run directory */
420 ret = system(cmd);
421 if (ret < 0) {
422 ERR("Unable to clean %s", rundir);
423 }
424 free(cmd);
425 free(rundir);
426
427 DBG("Cleaning up all sessions");
428
429 /* Destroy session list mutex */
430 if (session_list_ptr != NULL) {
431 pthread_mutex_destroy(&session_list_ptr->lock);
432
433 /* Cleanup ALL session */
434 cds_list_for_each_entry_safe(sess, stmp,
435 &session_list_ptr->head, list) {
436 cmd_destroy_session(sess, kernel_poll_pipe[1]);
437 }
438 }
439
440 DBG("Closing all UST sockets");
441 ust_app_clean_list();
442
443 if (is_root && !opt_no_kernel) {
444 DBG2("Closing kernel fd");
445 if (kernel_tracer_fd >= 0) {
446 ret = close(kernel_tracer_fd);
447 if (ret) {
448 PERROR("close");
449 }
450 }
451 DBG("Unloading kernel modules");
452 modprobe_remove_lttng_all();
453 }
454
455 /* <fun> */
456 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
457 "Matthew, BEET driven development works!%c[%dm",
458 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
459 /* </fun> */
460 }
461
462 /*
463 * Send data on a unix socket using the liblttsessiondcomm API.
464 *
465 * Return lttcomm error code.
466 */
467 static int send_unix_sock(int sock, void *buf, size_t len)
468 {
469 /* Check valid length */
470 if (len == 0) {
471 return -1;
472 }
473
474 return lttcomm_send_unix_sock(sock, buf, len);
475 }
476
477 /*
478 * Free memory of a command context structure.
479 */
480 static void clean_command_ctx(struct command_ctx **cmd_ctx)
481 {
482 DBG("Clean command context structure");
483 if (*cmd_ctx) {
484 if ((*cmd_ctx)->llm) {
485 free((*cmd_ctx)->llm);
486 }
487 if ((*cmd_ctx)->lsm) {
488 free((*cmd_ctx)->lsm);
489 }
490 free(*cmd_ctx);
491 *cmd_ctx = NULL;
492 }
493 }
494
495 /*
496 * Notify UST applications using the shm mmap futex.
497 */
498 static int notify_ust_apps(int active)
499 {
500 char *wait_shm_mmap;
501
502 DBG("Notifying applications of session daemon state: %d", active);
503
504 /* See shm.c for this call implying mmap, shm and futex calls */
505 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
506 if (wait_shm_mmap == NULL) {
507 goto error;
508 }
509
510 /* Wake waiting process */
511 futex_wait_update((int32_t *) wait_shm_mmap, active);
512
513 /* Apps notified successfully */
514 return 0;
515
516 error:
517 return -1;
518 }
519
520 /*
521 * Setup the outgoing data buffer for the response (llm) by allocating the
522 * right amount of memory and copying the original information from the lsm
523 * structure.
524 *
525 * Return total size of the buffer pointed by buf.
526 */
527 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
528 {
529 int ret, buf_size;
530
531 buf_size = size;
532
533 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
534 if (cmd_ctx->llm == NULL) {
535 PERROR("zmalloc");
536 ret = -ENOMEM;
537 goto error;
538 }
539
540 /* Copy common data */
541 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
542 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
543
544 cmd_ctx->llm->data_size = size;
545 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
546
547 return buf_size;
548
549 error:
550 return ret;
551 }
552
553 /*
554 * Update the kernel poll set of all channel fd available over all tracing
555 * session. Add the wakeup pipe at the end of the set.
556 */
557 static int update_kernel_poll(struct lttng_poll_event *events)
558 {
559 int ret;
560 struct ltt_session *session;
561 struct ltt_kernel_channel *channel;
562
563 DBG("Updating kernel poll set");
564
565 session_lock_list();
566 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
567 session_lock(session);
568 if (session->kernel_session == NULL) {
569 session_unlock(session);
570 continue;
571 }
572
573 cds_list_for_each_entry(channel,
574 &session->kernel_session->channel_list.head, list) {
575 /* Add channel fd to the kernel poll set */
576 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
577 if (ret < 0) {
578 session_unlock(session);
579 goto error;
580 }
581 DBG("Channel fd %d added to kernel set", channel->fd);
582 }
583 session_unlock(session);
584 }
585 session_unlock_list();
586
587 return 0;
588
589 error:
590 session_unlock_list();
591 return -1;
592 }
593
594 /*
595 * Find the channel fd from 'fd' over all tracing session. When found, check
596 * for new channel stream and send those stream fds to the kernel consumer.
597 *
598 * Useful for CPU hotplug feature.
599 */
600 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
601 {
602 int ret = 0;
603 struct ltt_session *session;
604 struct ltt_kernel_session *ksess;
605 struct ltt_kernel_channel *channel;
606
607 DBG("Updating kernel streams for channel fd %d", fd);
608
609 session_lock_list();
610 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
611 session_lock(session);
612 if (session->kernel_session == NULL) {
613 session_unlock(session);
614 continue;
615 }
616 ksess = session->kernel_session;
617
618 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
619 if (channel->fd == fd) {
620 DBG("Channel found, updating kernel streams");
621 ret = kernel_open_channel_stream(channel);
622 if (ret < 0) {
623 goto error;
624 }
625
626 /*
627 * Have we already sent fds to the consumer? If yes, it means
628 * that tracing is started so it is safe to send our updated
629 * stream fds.
630 */
631 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
632 struct lttng_ht_iter iter;
633 struct consumer_socket *socket;
634
635 rcu_read_lock();
636 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
637 &iter.iter, socket, node.node) {
638 /* Code flow error */
639 assert(socket->fd >= 0);
640
641 pthread_mutex_lock(socket->lock);
642 ret = kernel_consumer_send_channel_stream(socket,
643 channel, ksess);
644 pthread_mutex_unlock(socket->lock);
645 if (ret < 0) {
646 rcu_read_unlock();
647 goto error;
648 }
649 }
650 rcu_read_unlock();
651 }
652 goto error;
653 }
654 }
655 session_unlock(session);
656 }
657 session_unlock_list();
658 return ret;
659
660 error:
661 session_unlock(session);
662 session_unlock_list();
663 return ret;
664 }
665
666 /*
667 * For each tracing session, update newly registered apps. The session list
668 * lock MUST be acquired before calling this.
669 */
670 static void update_ust_app(int app_sock)
671 {
672 struct ltt_session *sess, *stmp;
673
674 /* For all tracing session(s) */
675 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
676 session_lock(sess);
677 if (sess->ust_session) {
678 ust_app_global_update(sess->ust_session, app_sock);
679 }
680 session_unlock(sess);
681 }
682 }
683
684 /*
685 * This thread manage event coming from the kernel.
686 *
687 * Features supported in this thread:
688 * -) CPU Hotplug
689 */
690 static void *thread_manage_kernel(void *data)
691 {
692 int ret, i, pollfd, update_poll_flag = 1, err = -1;
693 uint32_t revents, nb_fd;
694 char tmp;
695 struct lttng_poll_event events;
696
697 DBG("[thread] Thread manage kernel started");
698
699 health_register(HEALTH_TYPE_KERNEL);
700
701 /*
702 * This first step of the while is to clean this structure which could free
703 * non NULL pointers so zero it before the loop.
704 */
705 memset(&events, 0, sizeof(events));
706
707 if (testpoint(thread_manage_kernel)) {
708 goto error_testpoint;
709 }
710
711 health_code_update();
712
713 if (testpoint(thread_manage_kernel_before_loop)) {
714 goto error_testpoint;
715 }
716
717 while (1) {
718 health_code_update();
719
720 if (update_poll_flag == 1) {
721 /* Clean events object. We are about to populate it again. */
722 lttng_poll_clean(&events);
723
724 ret = create_thread_poll_set(&events, 2);
725 if (ret < 0) {
726 goto error_poll_create;
727 }
728
729 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
730 if (ret < 0) {
731 goto error;
732 }
733
734 /* This will add the available kernel channel if any. */
735 ret = update_kernel_poll(&events);
736 if (ret < 0) {
737 goto error;
738 }
739 update_poll_flag = 0;
740 }
741
742 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
743
744 /* Poll infinite value of time */
745 restart:
746 health_poll_entry();
747 ret = lttng_poll_wait(&events, -1);
748 health_poll_exit();
749 if (ret < 0) {
750 /*
751 * Restart interrupted system call.
752 */
753 if (errno == EINTR) {
754 goto restart;
755 }
756 goto error;
757 } else if (ret == 0) {
758 /* Should not happen since timeout is infinite */
759 ERR("Return value of poll is 0 with an infinite timeout.\n"
760 "This should not have happened! Continuing...");
761 continue;
762 }
763
764 nb_fd = ret;
765
766 for (i = 0; i < nb_fd; i++) {
767 /* Fetch once the poll data */
768 revents = LTTNG_POLL_GETEV(&events, i);
769 pollfd = LTTNG_POLL_GETFD(&events, i);
770
771 health_code_update();
772
773 /* Thread quit pipe has been closed. Killing thread. */
774 ret = check_thread_quit_pipe(pollfd, revents);
775 if (ret) {
776 err = 0;
777 goto exit;
778 }
779
780 /* Check for data on kernel pipe */
781 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
782 do {
783 ret = read(kernel_poll_pipe[0], &tmp, 1);
784 } while (ret < 0 && errno == EINTR);
785 /*
786 * Ret value is useless here, if this pipe gets any actions an
787 * update is required anyway.
788 */
789 update_poll_flag = 1;
790 continue;
791 } else {
792 /*
793 * New CPU detected by the kernel. Adding kernel stream to
794 * kernel session and updating the kernel consumer
795 */
796 if (revents & LPOLLIN) {
797 ret = update_kernel_stream(&kconsumer_data, pollfd);
798 if (ret < 0) {
799 continue;
800 }
801 break;
802 /*
803 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
804 * and unregister kernel stream at this point.
805 */
806 }
807 }
808 }
809 }
810
811 exit:
812 error:
813 lttng_poll_clean(&events);
814 error_poll_create:
815 error_testpoint:
816 utils_close_pipe(kernel_poll_pipe);
817 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
818 if (err) {
819 health_error();
820 ERR("Health error occurred in %s", __func__);
821 WARN("Kernel thread died unexpectedly. "
822 "Kernel tracing can continue but CPU hotplug is disabled.");
823 }
824 health_unregister();
825 DBG("Kernel thread dying");
826 return NULL;
827 }
828
829 /*
830 * Signal pthread condition of the consumer data that the thread.
831 */
832 static void signal_consumer_condition(struct consumer_data *data, int state)
833 {
834 pthread_mutex_lock(&data->cond_mutex);
835
836 /*
837 * The state is set before signaling. It can be any value, it's the waiter
838 * job to correctly interpret this condition variable associated to the
839 * consumer pthread_cond.
840 *
841 * A value of 0 means that the corresponding thread of the consumer data
842 * was not started. 1 indicates that the thread has started and is ready
843 * for action. A negative value means that there was an error during the
844 * thread bootstrap.
845 */
846 data->consumer_thread_is_ready = state;
847 (void) pthread_cond_signal(&data->cond);
848
849 pthread_mutex_unlock(&data->cond_mutex);
850 }
851
852 /*
853 * This thread manage the consumer error sent back to the session daemon.
854 */
855 static void *thread_manage_consumer(void *data)
856 {
857 int sock = -1, i, ret, pollfd, err = -1;
858 uint32_t revents, nb_fd;
859 enum lttcomm_return_code code;
860 struct lttng_poll_event events;
861 struct consumer_data *consumer_data = data;
862
863 DBG("[thread] Manage consumer started");
864
865 health_register(HEALTH_TYPE_CONSUMER);
866
867 health_code_update();
868
869 /*
870 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
871 * Nothing more will be added to this poll set.
872 */
873 ret = create_thread_poll_set(&events, 2);
874 if (ret < 0) {
875 goto error_poll;
876 }
877
878 /*
879 * The error socket here is already in a listening state which was done
880 * just before spawning this thread to avoid a race between the consumer
881 * daemon exec trying to connect and the listen() call.
882 */
883 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
884 if (ret < 0) {
885 goto error;
886 }
887
888 health_code_update();
889
890 /* Inifinite blocking call, waiting for transmission */
891 restart:
892 health_poll_entry();
893
894 if (testpoint(thread_manage_consumer)) {
895 goto error;
896 }
897
898 ret = lttng_poll_wait(&events, -1);
899 health_poll_exit();
900 if (ret < 0) {
901 /*
902 * Restart interrupted system call.
903 */
904 if (errno == EINTR) {
905 goto restart;
906 }
907 goto error;
908 }
909
910 nb_fd = ret;
911
912 for (i = 0; i < nb_fd; i++) {
913 /* Fetch once the poll data */
914 revents = LTTNG_POLL_GETEV(&events, i);
915 pollfd = LTTNG_POLL_GETFD(&events, i);
916
917 health_code_update();
918
919 /* Thread quit pipe has been closed. Killing thread. */
920 ret = check_thread_quit_pipe(pollfd, revents);
921 if (ret) {
922 err = 0;
923 goto exit;
924 }
925
926 /* Event on the registration socket */
927 if (pollfd == consumer_data->err_sock) {
928 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
929 ERR("consumer err socket poll error");
930 goto error;
931 }
932 }
933 }
934
935 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
936 if (sock < 0) {
937 goto error;
938 }
939
940 /*
941 * Set the CLOEXEC flag. Return code is useless because either way, the
942 * show must go on.
943 */
944 (void) utils_set_fd_cloexec(sock);
945
946 health_code_update();
947
948 DBG2("Receiving code from consumer err_sock");
949
950 /* Getting status code from kconsumerd */
951 ret = lttcomm_recv_unix_sock(sock, &code,
952 sizeof(enum lttcomm_return_code));
953 if (ret <= 0) {
954 goto error;
955 }
956
957 health_code_update();
958
959 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
960 consumer_data->cmd_sock =
961 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
962 if (consumer_data->cmd_sock < 0) {
963 /* On error, signal condition and quit. */
964 signal_consumer_condition(consumer_data, -1);
965 PERROR("consumer connect");
966 goto error;
967 }
968 signal_consumer_condition(consumer_data, 1);
969 DBG("Consumer command socket ready");
970 } else {
971 ERR("consumer error when waiting for SOCK_READY : %s",
972 lttcomm_get_readable_code(-code));
973 goto error;
974 }
975
976 /* Remove the kconsumerd error sock since we've established a connexion */
977 ret = lttng_poll_del(&events, consumer_data->err_sock);
978 if (ret < 0) {
979 goto error;
980 }
981
982 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
983 if (ret < 0) {
984 goto error;
985 }
986
987 health_code_update();
988
989 /* Inifinite blocking call, waiting for transmission */
990 restart_poll:
991 health_poll_entry();
992 ret = lttng_poll_wait(&events, -1);
993 health_poll_exit();
994 if (ret < 0) {
995 /*
996 * Restart interrupted system call.
997 */
998 if (errno == EINTR) {
999 goto restart_poll;
1000 }
1001 goto error;
1002 }
1003
1004 nb_fd = ret;
1005
1006 for (i = 0; i < nb_fd; i++) {
1007 /* Fetch once the poll data */
1008 revents = LTTNG_POLL_GETEV(&events, i);
1009 pollfd = LTTNG_POLL_GETFD(&events, i);
1010
1011 health_code_update();
1012
1013 /* Thread quit pipe has been closed. Killing thread. */
1014 ret = check_thread_quit_pipe(pollfd, revents);
1015 if (ret) {
1016 err = 0;
1017 goto exit;
1018 }
1019
1020 /* Event on the kconsumerd socket */
1021 if (pollfd == sock) {
1022 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1023 ERR("consumer err socket second poll error");
1024 goto error;
1025 }
1026 }
1027 }
1028
1029 health_code_update();
1030
1031 /* Wait for any kconsumerd error */
1032 ret = lttcomm_recv_unix_sock(sock, &code,
1033 sizeof(enum lttcomm_return_code));
1034 if (ret <= 0) {
1035 ERR("consumer closed the command socket");
1036 goto error;
1037 }
1038
1039 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1040
1041 exit:
1042 error:
1043 /* Immediately set the consumerd state to stopped */
1044 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1045 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1046 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1047 consumer_data->type == LTTNG_CONSUMER32_UST) {
1048 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1049 } else {
1050 /* Code flow error... */
1051 assert(0);
1052 }
1053
1054 if (consumer_data->err_sock >= 0) {
1055 ret = close(consumer_data->err_sock);
1056 if (ret) {
1057 PERROR("close");
1058 }
1059 }
1060 if (consumer_data->cmd_sock >= 0) {
1061 ret = close(consumer_data->cmd_sock);
1062 if (ret) {
1063 PERROR("close");
1064 }
1065 }
1066 if (sock >= 0) {
1067 ret = close(sock);
1068 if (ret) {
1069 PERROR("close");
1070 }
1071 }
1072
1073 unlink(consumer_data->err_unix_sock_path);
1074 unlink(consumer_data->cmd_unix_sock_path);
1075 consumer_data->pid = 0;
1076
1077 lttng_poll_clean(&events);
1078 error_poll:
1079 if (err) {
1080 health_error();
1081 ERR("Health error occurred in %s", __func__);
1082 }
1083 health_unregister();
1084 DBG("consumer thread cleanup completed");
1085
1086 return NULL;
1087 }
1088
1089 /*
1090 * This thread manage application communication.
1091 */
1092 static void *thread_manage_apps(void *data)
1093 {
1094 int i, ret, pollfd, err = -1;
1095 uint32_t revents, nb_fd;
1096 struct ust_command ust_cmd;
1097 struct lttng_poll_event events;
1098
1099 DBG("[thread] Manage application started");
1100
1101 rcu_register_thread();
1102 rcu_thread_online();
1103
1104 health_register(HEALTH_TYPE_APP_MANAGE);
1105
1106 if (testpoint(thread_manage_apps)) {
1107 goto error_testpoint;
1108 }
1109
1110 health_code_update();
1111
1112 ret = create_thread_poll_set(&events, 2);
1113 if (ret < 0) {
1114 goto error_poll_create;
1115 }
1116
1117 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1118 if (ret < 0) {
1119 goto error;
1120 }
1121
1122 if (testpoint(thread_manage_apps_before_loop)) {
1123 goto error;
1124 }
1125
1126 health_code_update();
1127
1128 while (1) {
1129 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1130
1131 /* Inifinite blocking call, waiting for transmission */
1132 restart:
1133 health_poll_entry();
1134 ret = lttng_poll_wait(&events, -1);
1135 health_poll_exit();
1136 if (ret < 0) {
1137 /*
1138 * Restart interrupted system call.
1139 */
1140 if (errno == EINTR) {
1141 goto restart;
1142 }
1143 goto error;
1144 }
1145
1146 nb_fd = ret;
1147
1148 for (i = 0; i < nb_fd; i++) {
1149 /* Fetch once the poll data */
1150 revents = LTTNG_POLL_GETEV(&events, i);
1151 pollfd = LTTNG_POLL_GETFD(&events, i);
1152
1153 health_code_update();
1154
1155 /* Thread quit pipe has been closed. Killing thread. */
1156 ret = check_thread_quit_pipe(pollfd, revents);
1157 if (ret) {
1158 err = 0;
1159 goto exit;
1160 }
1161
1162 /* Inspect the apps cmd pipe */
1163 if (pollfd == apps_cmd_pipe[0]) {
1164 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1165 ERR("Apps command pipe error");
1166 goto error;
1167 } else if (revents & LPOLLIN) {
1168 /* Empty pipe */
1169 do {
1170 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1171 } while (ret < 0 && errno == EINTR);
1172 if (ret < 0 || ret < sizeof(ust_cmd)) {
1173 PERROR("read apps cmd pipe");
1174 goto error;
1175 }
1176
1177 health_code_update();
1178
1179 /*
1180 * @session_lock
1181 * Lock the global session list so from the register up to
1182 * the registration done message, no thread can see the
1183 * application and change its state.
1184 */
1185 session_lock_list();
1186
1187 /* Register applicaton to the session daemon */
1188 ret = ust_app_register(&ust_cmd.reg_msg,
1189 ust_cmd.sock);
1190 if (ret == -ENOMEM) {
1191 session_unlock_list();
1192 goto error;
1193 } else if (ret < 0) {
1194 session_unlock_list();
1195 break;
1196 }
1197
1198 health_code_update();
1199
1200 /*
1201 * Validate UST version compatibility.
1202 */
1203 ret = ust_app_validate_version(ust_cmd.sock);
1204 if (ret >= 0) {
1205 /*
1206 * Add channel(s) and event(s) to newly registered apps
1207 * from lttng global UST domain.
1208 */
1209 update_ust_app(ust_cmd.sock);
1210 }
1211
1212 health_code_update();
1213
1214 ret = ust_app_register_done(ust_cmd.sock);
1215 if (ret < 0) {
1216 /*
1217 * If the registration is not possible, we simply
1218 * unregister the apps and continue
1219 */
1220 ust_app_unregister(ust_cmd.sock);
1221 } else {
1222 /*
1223 * We only monitor the error events of the socket. This
1224 * thread does not handle any incoming data from UST
1225 * (POLLIN).
1226 */
1227 ret = lttng_poll_add(&events, ust_cmd.sock,
1228 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1229 if (ret < 0) {
1230 session_unlock_list();
1231 goto error;
1232 }
1233
1234 /* Set socket timeout for both receiving and ending */
1235 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1236 app_socket_timeout);
1237 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1238 app_socket_timeout);
1239
1240 DBG("Apps with sock %d added to poll set",
1241 ust_cmd.sock);
1242 }
1243 session_unlock_list();
1244
1245 health_code_update();
1246
1247 break;
1248 }
1249 } else {
1250 /*
1251 * At this point, we know that a registered application made
1252 * the event at poll_wait.
1253 */
1254 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1255 /* Removing from the poll set */
1256 ret = lttng_poll_del(&events, pollfd);
1257 if (ret < 0) {
1258 goto error;
1259 }
1260
1261 /* Socket closed on remote end. */
1262 ust_app_unregister(pollfd);
1263 break;
1264 }
1265 }
1266
1267 health_code_update();
1268 }
1269 }
1270
1271 exit:
1272 error:
1273 lttng_poll_clean(&events);
1274 error_poll_create:
1275 error_testpoint:
1276 utils_close_pipe(apps_cmd_pipe);
1277 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1278
1279 /*
1280 * We don't clean the UST app hash table here since already registered
1281 * applications can still be controlled so let them be until the session
1282 * daemon dies or the applications stop.
1283 */
1284
1285 if (err) {
1286 health_error();
1287 ERR("Health error occurred in %s", __func__);
1288 }
1289 health_unregister();
1290 DBG("Application communication apps thread cleanup complete");
1291 rcu_thread_offline();
1292 rcu_unregister_thread();
1293 return NULL;
1294 }
1295
1296 /*
1297 * Dispatch request from the registration threads to the application
1298 * communication thread.
1299 */
1300 static void *thread_dispatch_ust_registration(void *data)
1301 {
1302 int ret;
1303 struct cds_wfq_node *node;
1304 struct ust_command *ust_cmd = NULL;
1305
1306 DBG("[thread] Dispatch UST command started");
1307
1308 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1309 /* Atomically prepare the queue futex */
1310 futex_nto1_prepare(&ust_cmd_queue.futex);
1311
1312 do {
1313 /* Dequeue command for registration */
1314 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1315 if (node == NULL) {
1316 DBG("Woken up but nothing in the UST command queue");
1317 /* Continue thread execution */
1318 break;
1319 }
1320
1321 ust_cmd = caa_container_of(node, struct ust_command, node);
1322
1323 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1324 " gid:%d sock:%d name:%s (version %d.%d)",
1325 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1326 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1327 ust_cmd->sock, ust_cmd->reg_msg.name,
1328 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1329 /*
1330 * Inform apps thread of the new application registration. This
1331 * call is blocking so we can be assured that the data will be read
1332 * at some point in time or wait to the end of the world :)
1333 */
1334 if (apps_cmd_pipe[1] >= 0) {
1335 do {
1336 ret = write(apps_cmd_pipe[1], ust_cmd,
1337 sizeof(struct ust_command));
1338 } while (ret < 0 && errno == EINTR);
1339 if (ret < 0 || ret != sizeof(struct ust_command)) {
1340 PERROR("write apps cmd pipe");
1341 if (errno == EBADF) {
1342 /*
1343 * We can't inform the application thread to process
1344 * registration. We will exit or else application
1345 * registration will not occur and tracing will never
1346 * start.
1347 */
1348 goto error;
1349 }
1350 }
1351 } else {
1352 /* Application manager thread is not available. */
1353 ret = close(ust_cmd->sock);
1354 if (ret < 0) {
1355 PERROR("close ust_cmd sock");
1356 }
1357 }
1358 free(ust_cmd);
1359 } while (node != NULL);
1360
1361 /* Futex wait on queue. Blocking call on futex() */
1362 futex_nto1_wait(&ust_cmd_queue.futex);
1363 }
1364
1365 error:
1366 DBG("Dispatch thread dying");
1367 return NULL;
1368 }
1369
1370 /*
1371 * This thread manage application registration.
1372 */
1373 static void *thread_registration_apps(void *data)
1374 {
1375 int sock = -1, i, ret, pollfd, err = -1;
1376 uint32_t revents, nb_fd;
1377 struct lttng_poll_event events;
1378 /*
1379 * Get allocated in this thread, enqueued to a global queue, dequeued and
1380 * freed in the manage apps thread.
1381 */
1382 struct ust_command *ust_cmd = NULL;
1383
1384 DBG("[thread] Manage application registration started");
1385
1386 health_register(HEALTH_TYPE_APP_REG);
1387
1388 if (testpoint(thread_registration_apps)) {
1389 goto error_testpoint;
1390 }
1391
1392 ret = lttcomm_listen_unix_sock(apps_sock);
1393 if (ret < 0) {
1394 goto error_listen;
1395 }
1396
1397 /*
1398 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1399 * more will be added to this poll set.
1400 */
1401 ret = create_thread_poll_set(&events, 2);
1402 if (ret < 0) {
1403 goto error_create_poll;
1404 }
1405
1406 /* Add the application registration socket */
1407 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1408 if (ret < 0) {
1409 goto error_poll_add;
1410 }
1411
1412 /* Notify all applications to register */
1413 ret = notify_ust_apps(1);
1414 if (ret < 0) {
1415 ERR("Failed to notify applications or create the wait shared memory.\n"
1416 "Execution continues but there might be problem for already\n"
1417 "running applications that wishes to register.");
1418 }
1419
1420 while (1) {
1421 DBG("Accepting application registration");
1422
1423 /* Inifinite blocking call, waiting for transmission */
1424 restart:
1425 health_poll_entry();
1426 ret = lttng_poll_wait(&events, -1);
1427 health_poll_exit();
1428 if (ret < 0) {
1429 /*
1430 * Restart interrupted system call.
1431 */
1432 if (errno == EINTR) {
1433 goto restart;
1434 }
1435 goto error;
1436 }
1437
1438 nb_fd = ret;
1439
1440 for (i = 0; i < nb_fd; i++) {
1441 health_code_update();
1442
1443 /* Fetch once the poll data */
1444 revents = LTTNG_POLL_GETEV(&events, i);
1445 pollfd = LTTNG_POLL_GETFD(&events, i);
1446
1447 /* Thread quit pipe has been closed. Killing thread. */
1448 ret = check_thread_quit_pipe(pollfd, revents);
1449 if (ret) {
1450 err = 0;
1451 goto exit;
1452 }
1453
1454 /* Event on the registration socket */
1455 if (pollfd == apps_sock) {
1456 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1457 ERR("Register apps socket poll error");
1458 goto error;
1459 } else if (revents & LPOLLIN) {
1460 sock = lttcomm_accept_unix_sock(apps_sock);
1461 if (sock < 0) {
1462 goto error;
1463 }
1464
1465 /*
1466 * Set the CLOEXEC flag. Return code is useless because
1467 * either way, the show must go on.
1468 */
1469 (void) utils_set_fd_cloexec(sock);
1470
1471 /* Create UST registration command for enqueuing */
1472 ust_cmd = zmalloc(sizeof(struct ust_command));
1473 if (ust_cmd == NULL) {
1474 PERROR("ust command zmalloc");
1475 goto error;
1476 }
1477
1478 /*
1479 * Using message-based transmissions to ensure we don't
1480 * have to deal with partially received messages.
1481 */
1482 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1483 if (ret < 0) {
1484 ERR("Exhausted file descriptors allowed for applications.");
1485 free(ust_cmd);
1486 ret = close(sock);
1487 if (ret) {
1488 PERROR("close");
1489 }
1490 sock = -1;
1491 continue;
1492 }
1493 health_code_update();
1494 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1495 sizeof(struct ust_register_msg));
1496 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1497 if (ret < 0) {
1498 PERROR("lttcomm_recv_unix_sock register apps");
1499 } else {
1500 ERR("Wrong size received on apps register");
1501 }
1502 free(ust_cmd);
1503 ret = close(sock);
1504 if (ret) {
1505 PERROR("close");
1506 }
1507 lttng_fd_put(LTTNG_FD_APPS, 1);
1508 sock = -1;
1509 continue;
1510 }
1511 health_code_update();
1512
1513 ust_cmd->sock = sock;
1514 sock = -1;
1515
1516 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1517 " gid:%d sock:%d name:%s (version %d.%d)",
1518 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1519 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1520 ust_cmd->sock, ust_cmd->reg_msg.name,
1521 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1522
1523 /*
1524 * Lock free enqueue the registration request. The red pill
1525 * has been taken! This apps will be part of the *system*.
1526 */
1527 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1528
1529 /*
1530 * Wake the registration queue futex. Implicit memory
1531 * barrier with the exchange in cds_wfq_enqueue.
1532 */
1533 futex_nto1_wake(&ust_cmd_queue.futex);
1534 }
1535 }
1536 }
1537 }
1538
1539 exit:
1540 error:
1541 if (err) {
1542 health_error();
1543 ERR("Health error occurred in %s", __func__);
1544 }
1545
1546 /* Notify that the registration thread is gone */
1547 notify_ust_apps(0);
1548
1549 if (apps_sock >= 0) {
1550 ret = close(apps_sock);
1551 if (ret) {
1552 PERROR("close");
1553 }
1554 }
1555 if (sock >= 0) {
1556 ret = close(sock);
1557 if (ret) {
1558 PERROR("close");
1559 }
1560 lttng_fd_put(LTTNG_FD_APPS, 1);
1561 }
1562 unlink(apps_unix_sock_path);
1563
1564 error_poll_add:
1565 lttng_poll_clean(&events);
1566 error_listen:
1567 error_create_poll:
1568 error_testpoint:
1569 DBG("UST Registration thread cleanup complete");
1570 health_unregister();
1571
1572 return NULL;
1573 }
1574
1575 /*
1576 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1577 * exec or it will fails.
1578 */
1579 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1580 {
1581 int ret, clock_ret;
1582 struct timespec timeout;
1583
1584 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1585 consumer_data->consumer_thread_is_ready = 0;
1586
1587 /* Setup pthread condition */
1588 ret = pthread_condattr_init(&consumer_data->condattr);
1589 if (ret != 0) {
1590 errno = ret;
1591 PERROR("pthread_condattr_init consumer data");
1592 goto error;
1593 }
1594
1595 /*
1596 * Set the monotonic clock in order to make sure we DO NOT jump in time
1597 * between the clock_gettime() call and the timedwait call. See bug #324
1598 * for a more details and how we noticed it.
1599 */
1600 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1601 if (ret != 0) {
1602 errno = ret;
1603 PERROR("pthread_condattr_setclock consumer data");
1604 goto error;
1605 }
1606
1607 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1608 if (ret != 0) {
1609 errno = ret;
1610 PERROR("pthread_cond_init consumer data");
1611 goto error;
1612 }
1613
1614 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1615 consumer_data);
1616 if (ret != 0) {
1617 PERROR("pthread_create consumer");
1618 ret = -1;
1619 goto error;
1620 }
1621
1622 /* We are about to wait on a pthread condition */
1623 pthread_mutex_lock(&consumer_data->cond_mutex);
1624
1625 /* Get time for sem_timedwait absolute timeout */
1626 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1627 /*
1628 * Set the timeout for the condition timed wait even if the clock gettime
1629 * call fails since we might loop on that call and we want to avoid to
1630 * increment the timeout too many times.
1631 */
1632 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1633
1634 /*
1635 * The following loop COULD be skipped in some conditions so this is why we
1636 * set ret to 0 in order to make sure at least one round of the loop is
1637 * done.
1638 */
1639 ret = 0;
1640
1641 /*
1642 * Loop until the condition is reached or when a timeout is reached. Note
1643 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1644 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1645 * possible. This loop does not take any chances and works with both of
1646 * them.
1647 */
1648 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1649 if (clock_ret < 0) {
1650 PERROR("clock_gettime spawn consumer");
1651 /* Infinite wait for the consumerd thread to be ready */
1652 ret = pthread_cond_wait(&consumer_data->cond,
1653 &consumer_data->cond_mutex);
1654 } else {
1655 ret = pthread_cond_timedwait(&consumer_data->cond,
1656 &consumer_data->cond_mutex, &timeout);
1657 }
1658 }
1659
1660 /* Release the pthread condition */
1661 pthread_mutex_unlock(&consumer_data->cond_mutex);
1662
1663 if (ret != 0) {
1664 errno = ret;
1665 if (ret == ETIMEDOUT) {
1666 /*
1667 * Call has timed out so we kill the kconsumerd_thread and return
1668 * an error.
1669 */
1670 ERR("Condition timed out. The consumer thread was never ready."
1671 " Killing it");
1672 ret = pthread_cancel(consumer_data->thread);
1673 if (ret < 0) {
1674 PERROR("pthread_cancel consumer thread");
1675 }
1676 } else {
1677 PERROR("pthread_cond_wait failed consumer thread");
1678 }
1679 goto error;
1680 }
1681
1682 pthread_mutex_lock(&consumer_data->pid_mutex);
1683 if (consumer_data->pid == 0) {
1684 ERR("Consumerd did not start");
1685 pthread_mutex_unlock(&consumer_data->pid_mutex);
1686 goto error;
1687 }
1688 pthread_mutex_unlock(&consumer_data->pid_mutex);
1689
1690 return 0;
1691
1692 error:
1693 return ret;
1694 }
1695
1696 /*
1697 * Join consumer thread
1698 */
1699 static int join_consumer_thread(struct consumer_data *consumer_data)
1700 {
1701 void *status;
1702
1703 /* Consumer pid must be a real one. */
1704 if (consumer_data->pid > 0) {
1705 int ret;
1706 ret = kill(consumer_data->pid, SIGTERM);
1707 if (ret) {
1708 ERR("Error killing consumer daemon");
1709 return ret;
1710 }
1711 return pthread_join(consumer_data->thread, &status);
1712 } else {
1713 return 0;
1714 }
1715 }
1716
1717 /*
1718 * Fork and exec a consumer daemon (consumerd).
1719 *
1720 * Return pid if successful else -1.
1721 */
1722 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1723 {
1724 int ret;
1725 pid_t pid;
1726 const char *consumer_to_use;
1727 const char *verbosity;
1728 struct stat st;
1729
1730 DBG("Spawning consumerd");
1731
1732 pid = fork();
1733 if (pid == 0) {
1734 /*
1735 * Exec consumerd.
1736 */
1737 if (opt_verbose_consumer) {
1738 verbosity = "--verbose";
1739 } else {
1740 verbosity = "--quiet";
1741 }
1742 switch (consumer_data->type) {
1743 case LTTNG_CONSUMER_KERNEL:
1744 /*
1745 * Find out which consumerd to execute. We will first try the
1746 * 64-bit path, then the sessiond's installation directory, and
1747 * fallback on the 32-bit one,
1748 */
1749 DBG3("Looking for a kernel consumer at these locations:");
1750 DBG3(" 1) %s", consumerd64_bin);
1751 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1752 DBG3(" 3) %s", consumerd32_bin);
1753 if (stat(consumerd64_bin, &st) == 0) {
1754 DBG3("Found location #1");
1755 consumer_to_use = consumerd64_bin;
1756 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1757 DBG3("Found location #2");
1758 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1759 } else if (stat(consumerd32_bin, &st) == 0) {
1760 DBG3("Found location #3");
1761 consumer_to_use = consumerd32_bin;
1762 } else {
1763 DBG("Could not find any valid consumerd executable");
1764 break;
1765 }
1766 DBG("Using kernel consumer at: %s", consumer_to_use);
1767 execl(consumer_to_use,
1768 "lttng-consumerd", verbosity, "-k",
1769 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1770 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1771 NULL);
1772 break;
1773 case LTTNG_CONSUMER64_UST:
1774 {
1775 char *tmpnew = NULL;
1776
1777 if (consumerd64_libdir[0] != '\0') {
1778 char *tmp;
1779 size_t tmplen;
1780
1781 tmp = getenv("LD_LIBRARY_PATH");
1782 if (!tmp) {
1783 tmp = "";
1784 }
1785 tmplen = strlen("LD_LIBRARY_PATH=")
1786 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1787 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1788 if (!tmpnew) {
1789 ret = -ENOMEM;
1790 goto error;
1791 }
1792 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1793 strcat(tmpnew, consumerd64_libdir);
1794 if (tmp[0] != '\0') {
1795 strcat(tmpnew, ":");
1796 strcat(tmpnew, tmp);
1797 }
1798 ret = putenv(tmpnew);
1799 if (ret) {
1800 ret = -errno;
1801 goto error;
1802 }
1803 }
1804 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1805 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1806 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1807 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1808 NULL);
1809 if (consumerd64_libdir[0] != '\0') {
1810 free(tmpnew);
1811 }
1812 if (ret) {
1813 goto error;
1814 }
1815 break;
1816 }
1817 case LTTNG_CONSUMER32_UST:
1818 {
1819 char *tmpnew = NULL;
1820
1821 if (consumerd32_libdir[0] != '\0') {
1822 char *tmp;
1823 size_t tmplen;
1824
1825 tmp = getenv("LD_LIBRARY_PATH");
1826 if (!tmp) {
1827 tmp = "";
1828 }
1829 tmplen = strlen("LD_LIBRARY_PATH=")
1830 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1831 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1832 if (!tmpnew) {
1833 ret = -ENOMEM;
1834 goto error;
1835 }
1836 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1837 strcat(tmpnew, consumerd32_libdir);
1838 if (tmp[0] != '\0') {
1839 strcat(tmpnew, ":");
1840 strcat(tmpnew, tmp);
1841 }
1842 ret = putenv(tmpnew);
1843 if (ret) {
1844 ret = -errno;
1845 goto error;
1846 }
1847 }
1848 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1849 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1850 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1851 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1852 NULL);
1853 if (consumerd32_libdir[0] != '\0') {
1854 free(tmpnew);
1855 }
1856 if (ret) {
1857 goto error;
1858 }
1859 break;
1860 }
1861 default:
1862 PERROR("unknown consumer type");
1863 exit(EXIT_FAILURE);
1864 }
1865 if (errno != 0) {
1866 PERROR("kernel start consumer exec");
1867 }
1868 exit(EXIT_FAILURE);
1869 } else if (pid > 0) {
1870 ret = pid;
1871 } else {
1872 PERROR("start consumer fork");
1873 ret = -errno;
1874 }
1875 error:
1876 return ret;
1877 }
1878
1879 /*
1880 * Spawn the consumerd daemon and session daemon thread.
1881 */
1882 static int start_consumerd(struct consumer_data *consumer_data)
1883 {
1884 int ret;
1885
1886 /*
1887 * Set the listen() state on the socket since there is a possible race
1888 * between the exec() of the consumer daemon and this call if place in the
1889 * consumer thread. See bug #366 for more details.
1890 */
1891 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1892 if (ret < 0) {
1893 goto error;
1894 }
1895
1896 pthread_mutex_lock(&consumer_data->pid_mutex);
1897 if (consumer_data->pid != 0) {
1898 pthread_mutex_unlock(&consumer_data->pid_mutex);
1899 goto end;
1900 }
1901
1902 ret = spawn_consumerd(consumer_data);
1903 if (ret < 0) {
1904 ERR("Spawning consumerd failed");
1905 pthread_mutex_unlock(&consumer_data->pid_mutex);
1906 goto error;
1907 }
1908
1909 /* Setting up the consumer_data pid */
1910 consumer_data->pid = ret;
1911 DBG2("Consumer pid %d", consumer_data->pid);
1912 pthread_mutex_unlock(&consumer_data->pid_mutex);
1913
1914 DBG2("Spawning consumer control thread");
1915 ret = spawn_consumer_thread(consumer_data);
1916 if (ret < 0) {
1917 ERR("Fatal error spawning consumer control thread");
1918 goto error;
1919 }
1920
1921 end:
1922 return 0;
1923
1924 error:
1925 /* Cleanup already created socket on error. */
1926 if (consumer_data->err_sock >= 0) {
1927 int err;
1928
1929 err = close(consumer_data->err_sock);
1930 if (err < 0) {
1931 PERROR("close consumer data error socket");
1932 }
1933 }
1934 return ret;
1935 }
1936
1937 /*
1938 * Compute health status of each consumer. If one of them is zero (bad
1939 * state), we return 0.
1940 */
1941 static int check_consumer_health(void)
1942 {
1943 int ret;
1944
1945 ret = health_check_state(HEALTH_TYPE_CONSUMER);
1946
1947 DBG3("Health consumer check %d", ret);
1948
1949 return ret;
1950 }
1951
1952 /*
1953 * Setup necessary data for kernel tracer action.
1954 */
1955 static int init_kernel_tracer(void)
1956 {
1957 int ret;
1958
1959 /* Modprobe lttng kernel modules */
1960 ret = modprobe_lttng_control();
1961 if (ret < 0) {
1962 goto error;
1963 }
1964
1965 /* Open debugfs lttng */
1966 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1967 if (kernel_tracer_fd < 0) {
1968 DBG("Failed to open %s", module_proc_lttng);
1969 ret = -1;
1970 goto error_open;
1971 }
1972
1973 /* Validate kernel version */
1974 ret = kernel_validate_version(kernel_tracer_fd);
1975 if (ret < 0) {
1976 goto error_version;
1977 }
1978
1979 ret = modprobe_lttng_data();
1980 if (ret < 0) {
1981 goto error_modules;
1982 }
1983
1984 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1985 return 0;
1986
1987 error_version:
1988 modprobe_remove_lttng_control();
1989 ret = close(kernel_tracer_fd);
1990 if (ret) {
1991 PERROR("close");
1992 }
1993 kernel_tracer_fd = -1;
1994 return LTTNG_ERR_KERN_VERSION;
1995
1996 error_modules:
1997 ret = close(kernel_tracer_fd);
1998 if (ret) {
1999 PERROR("close");
2000 }
2001
2002 error_open:
2003 modprobe_remove_lttng_control();
2004
2005 error:
2006 WARN("No kernel tracer available");
2007 kernel_tracer_fd = -1;
2008 if (!is_root) {
2009 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2010 } else {
2011 return LTTNG_ERR_KERN_NA;
2012 }
2013 }
2014
2015
2016 /*
2017 * Copy consumer output from the tracing session to the domain session. The
2018 * function also applies the right modification on a per domain basis for the
2019 * trace files destination directory.
2020 */
2021 static int copy_session_consumer(int domain, struct ltt_session *session)
2022 {
2023 int ret;
2024 const char *dir_name;
2025 struct consumer_output *consumer;
2026
2027 assert(session);
2028 assert(session->consumer);
2029
2030 switch (domain) {
2031 case LTTNG_DOMAIN_KERNEL:
2032 DBG3("Copying tracing session consumer output in kernel session");
2033 /*
2034 * XXX: We should audit the session creation and what this function
2035 * does "extra" in order to avoid a destroy since this function is used
2036 * in the domain session creation (kernel and ust) only. Same for UST
2037 * domain.
2038 */
2039 if (session->kernel_session->consumer) {
2040 consumer_destroy_output(session->kernel_session->consumer);
2041 }
2042 session->kernel_session->consumer =
2043 consumer_copy_output(session->consumer);
2044 /* Ease our life a bit for the next part */
2045 consumer = session->kernel_session->consumer;
2046 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2047 break;
2048 case LTTNG_DOMAIN_UST:
2049 DBG3("Copying tracing session consumer output in UST session");
2050 if (session->ust_session->consumer) {
2051 consumer_destroy_output(session->ust_session->consumer);
2052 }
2053 session->ust_session->consumer =
2054 consumer_copy_output(session->consumer);
2055 /* Ease our life a bit for the next part */
2056 consumer = session->ust_session->consumer;
2057 dir_name = DEFAULT_UST_TRACE_DIR;
2058 break;
2059 default:
2060 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2061 goto error;
2062 }
2063
2064 /* Append correct directory to subdir */
2065 strncat(consumer->subdir, dir_name,
2066 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2067 DBG3("Copy session consumer subdir %s", consumer->subdir);
2068
2069 ret = LTTNG_OK;
2070
2071 error:
2072 return ret;
2073 }
2074
2075 /*
2076 * Create an UST session and add it to the session ust list.
2077 */
2078 static int create_ust_session(struct ltt_session *session,
2079 struct lttng_domain *domain)
2080 {
2081 int ret;
2082 struct ltt_ust_session *lus = NULL;
2083
2084 assert(session);
2085 assert(domain);
2086 assert(session->consumer);
2087
2088 switch (domain->type) {
2089 case LTTNG_DOMAIN_UST:
2090 break;
2091 default:
2092 ERR("Unknown UST domain on create session %d", domain->type);
2093 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2094 goto error;
2095 }
2096
2097 DBG("Creating UST session");
2098
2099 lus = trace_ust_create_session(session->path, session->id);
2100 if (lus == NULL) {
2101 ret = LTTNG_ERR_UST_SESS_FAIL;
2102 goto error;
2103 }
2104
2105 lus->uid = session->uid;
2106 lus->gid = session->gid;
2107 session->ust_session = lus;
2108
2109 /* Copy session output to the newly created UST session */
2110 ret = copy_session_consumer(domain->type, session);
2111 if (ret != LTTNG_OK) {
2112 goto error;
2113 }
2114
2115 return LTTNG_OK;
2116
2117 error:
2118 free(lus);
2119 session->ust_session = NULL;
2120 return ret;
2121 }
2122
2123 /*
2124 * Create a kernel tracer session then create the default channel.
2125 */
2126 static int create_kernel_session(struct ltt_session *session)
2127 {
2128 int ret;
2129
2130 DBG("Creating kernel session");
2131
2132 ret = kernel_create_session(session, kernel_tracer_fd);
2133 if (ret < 0) {
2134 ret = LTTNG_ERR_KERN_SESS_FAIL;
2135 goto error;
2136 }
2137
2138 /* Code flow safety */
2139 assert(session->kernel_session);
2140
2141 /* Copy session output to the newly created Kernel session */
2142 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2143 if (ret != LTTNG_OK) {
2144 goto error;
2145 }
2146
2147 /* Create directory(ies) on local filesystem. */
2148 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2149 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2150 ret = run_as_mkdir_recursive(
2151 session->kernel_session->consumer->dst.trace_path,
2152 S_IRWXU | S_IRWXG, session->uid, session->gid);
2153 if (ret < 0) {
2154 if (ret != -EEXIST) {
2155 ERR("Trace directory creation error");
2156 goto error;
2157 }
2158 }
2159 }
2160
2161 session->kernel_session->uid = session->uid;
2162 session->kernel_session->gid = session->gid;
2163
2164 return LTTNG_OK;
2165
2166 error:
2167 trace_kernel_destroy_session(session->kernel_session);
2168 session->kernel_session = NULL;
2169 return ret;
2170 }
2171
2172 /*
2173 * Count number of session permitted by uid/gid.
2174 */
2175 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2176 {
2177 unsigned int i = 0;
2178 struct ltt_session *session;
2179
2180 DBG("Counting number of available session for UID %d GID %d",
2181 uid, gid);
2182 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2183 /*
2184 * Only list the sessions the user can control.
2185 */
2186 if (!session_access_ok(session, uid, gid)) {
2187 continue;
2188 }
2189 i++;
2190 }
2191 return i;
2192 }
2193
2194 /*
2195 * Process the command requested by the lttng client within the command
2196 * context structure. This function make sure that the return structure (llm)
2197 * is set and ready for transmission before returning.
2198 *
2199 * Return any error encountered or 0 for success.
2200 *
2201 * "sock" is only used for special-case var. len data.
2202 */
2203 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2204 int *sock_error)
2205 {
2206 int ret = LTTNG_OK;
2207 int need_tracing_session = 1;
2208 int need_domain;
2209
2210 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2211
2212 *sock_error = 0;
2213
2214 switch (cmd_ctx->lsm->cmd_type) {
2215 case LTTNG_CREATE_SESSION:
2216 case LTTNG_DESTROY_SESSION:
2217 case LTTNG_LIST_SESSIONS:
2218 case LTTNG_LIST_DOMAINS:
2219 case LTTNG_START_TRACE:
2220 case LTTNG_STOP_TRACE:
2221 case LTTNG_DATA_PENDING:
2222 need_domain = 0;
2223 break;
2224 default:
2225 need_domain = 1;
2226 }
2227
2228 if (opt_no_kernel && need_domain
2229 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2230 if (!is_root) {
2231 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2232 } else {
2233 ret = LTTNG_ERR_KERN_NA;
2234 }
2235 goto error;
2236 }
2237
2238 /* Deny register consumer if we already have a spawned consumer. */
2239 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2240 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2241 if (kconsumer_data.pid > 0) {
2242 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2243 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2244 goto error;
2245 }
2246 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2247 }
2248
2249 /*
2250 * Check for command that don't needs to allocate a returned payload. We do
2251 * this here so we don't have to make the call for no payload at each
2252 * command.
2253 */
2254 switch(cmd_ctx->lsm->cmd_type) {
2255 case LTTNG_LIST_SESSIONS:
2256 case LTTNG_LIST_TRACEPOINTS:
2257 case LTTNG_LIST_TRACEPOINT_FIELDS:
2258 case LTTNG_LIST_DOMAINS:
2259 case LTTNG_LIST_CHANNELS:
2260 case LTTNG_LIST_EVENTS:
2261 break;
2262 default:
2263 /* Setup lttng message with no payload */
2264 ret = setup_lttng_msg(cmd_ctx, 0);
2265 if (ret < 0) {
2266 /* This label does not try to unlock the session */
2267 goto init_setup_error;
2268 }
2269 }
2270
2271 /* Commands that DO NOT need a session. */
2272 switch (cmd_ctx->lsm->cmd_type) {
2273 case LTTNG_CREATE_SESSION:
2274 case LTTNG_CALIBRATE:
2275 case LTTNG_LIST_SESSIONS:
2276 case LTTNG_LIST_TRACEPOINTS:
2277 case LTTNG_LIST_TRACEPOINT_FIELDS:
2278 need_tracing_session = 0;
2279 break;
2280 default:
2281 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2282 /*
2283 * We keep the session list lock across _all_ commands
2284 * for now, because the per-session lock does not
2285 * handle teardown properly.
2286 */
2287 session_lock_list();
2288 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2289 if (cmd_ctx->session == NULL) {
2290 if (cmd_ctx->lsm->session.name != NULL) {
2291 ret = LTTNG_ERR_SESS_NOT_FOUND;
2292 } else {
2293 /* If no session name specified */
2294 ret = LTTNG_ERR_SELECT_SESS;
2295 }
2296 goto error;
2297 } else {
2298 /* Acquire lock for the session */
2299 session_lock(cmd_ctx->session);
2300 }
2301 break;
2302 }
2303
2304 if (!need_domain) {
2305 goto skip_domain;
2306 }
2307
2308 /*
2309 * Check domain type for specific "pre-action".
2310 */
2311 switch (cmd_ctx->lsm->domain.type) {
2312 case LTTNG_DOMAIN_KERNEL:
2313 if (!is_root) {
2314 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2315 goto error;
2316 }
2317
2318 /* Kernel tracer check */
2319 if (kernel_tracer_fd == -1) {
2320 /* Basically, load kernel tracer modules */
2321 ret = init_kernel_tracer();
2322 if (ret != 0) {
2323 goto error;
2324 }
2325 }
2326
2327 /* Consumer is in an ERROR state. Report back to client */
2328 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2329 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2330 goto error;
2331 }
2332
2333 /* Need a session for kernel command */
2334 if (need_tracing_session) {
2335 if (cmd_ctx->session->kernel_session == NULL) {
2336 ret = create_kernel_session(cmd_ctx->session);
2337 if (ret < 0) {
2338 ret = LTTNG_ERR_KERN_SESS_FAIL;
2339 goto error;
2340 }
2341 }
2342
2343 /* Start the kernel consumer daemon */
2344 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2345 if (kconsumer_data.pid == 0 &&
2346 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2347 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2348 ret = start_consumerd(&kconsumer_data);
2349 if (ret < 0) {
2350 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2351 goto error;
2352 }
2353 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2354 } else {
2355 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2356 }
2357
2358 /*
2359 * The consumer was just spawned so we need to add the socket to
2360 * the consumer output of the session if exist.
2361 */
2362 ret = consumer_create_socket(&kconsumer_data,
2363 cmd_ctx->session->kernel_session->consumer);
2364 if (ret < 0) {
2365 goto error;
2366 }
2367 }
2368
2369 break;
2370 case LTTNG_DOMAIN_UST:
2371 {
2372 /* Consumer is in an ERROR state. Report back to client */
2373 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2374 ret = LTTNG_ERR_NO_USTCONSUMERD;
2375 goto error;
2376 }
2377
2378 if (need_tracing_session) {
2379 /* Create UST session if none exist. */
2380 if (cmd_ctx->session->ust_session == NULL) {
2381 ret = create_ust_session(cmd_ctx->session,
2382 &cmd_ctx->lsm->domain);
2383 if (ret != LTTNG_OK) {
2384 goto error;
2385 }
2386 }
2387
2388 /* Start the UST consumer daemons */
2389 /* 64-bit */
2390 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2391 if (consumerd64_bin[0] != '\0' &&
2392 ustconsumer64_data.pid == 0 &&
2393 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2394 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2395 ret = start_consumerd(&ustconsumer64_data);
2396 if (ret < 0) {
2397 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2398 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2399 goto error;
2400 }
2401
2402 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2403 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2404 } else {
2405 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2406 }
2407
2408 /*
2409 * Setup socket for consumer 64 bit. No need for atomic access
2410 * since it was set above and can ONLY be set in this thread.
2411 */
2412 ret = consumer_create_socket(&ustconsumer64_data,
2413 cmd_ctx->session->ust_session->consumer);
2414 if (ret < 0) {
2415 goto error;
2416 }
2417
2418 /* 32-bit */
2419 if (consumerd32_bin[0] != '\0' &&
2420 ustconsumer32_data.pid == 0 &&
2421 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2422 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2423 ret = start_consumerd(&ustconsumer32_data);
2424 if (ret < 0) {
2425 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2426 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2427 goto error;
2428 }
2429
2430 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2431 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2432 } else {
2433 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2434 }
2435
2436 /*
2437 * Setup socket for consumer 64 bit. No need for atomic access
2438 * since it was set above and can ONLY be set in this thread.
2439 */
2440 ret = consumer_create_socket(&ustconsumer32_data,
2441 cmd_ctx->session->ust_session->consumer);
2442 if (ret < 0) {
2443 goto error;
2444 }
2445 }
2446 break;
2447 }
2448 default:
2449 break;
2450 }
2451 skip_domain:
2452
2453 /* Validate consumer daemon state when start/stop trace command */
2454 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2455 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2456 switch (cmd_ctx->lsm->domain.type) {
2457 case LTTNG_DOMAIN_UST:
2458 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2459 ret = LTTNG_ERR_NO_USTCONSUMERD;
2460 goto error;
2461 }
2462 break;
2463 case LTTNG_DOMAIN_KERNEL:
2464 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2465 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2466 goto error;
2467 }
2468 break;
2469 }
2470 }
2471
2472 /*
2473 * Check that the UID or GID match that of the tracing session.
2474 * The root user can interact with all sessions.
2475 */
2476 if (need_tracing_session) {
2477 if (!session_access_ok(cmd_ctx->session,
2478 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2479 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2480 ret = LTTNG_ERR_EPERM;
2481 goto error;
2482 }
2483 }
2484
2485 /*
2486 * Send relayd information to consumer as soon as we have a domain and a
2487 * session defined.
2488 */
2489 if (cmd_ctx->session && need_domain) {
2490 /*
2491 * Setup relayd if not done yet. If the relayd information was already
2492 * sent to the consumer, this call will gracefully return.
2493 */
2494 ret = cmd_setup_relayd(cmd_ctx->session);
2495 if (ret != LTTNG_OK) {
2496 goto error;
2497 }
2498 }
2499
2500 /* Process by command type */
2501 switch (cmd_ctx->lsm->cmd_type) {
2502 case LTTNG_ADD_CONTEXT:
2503 {
2504 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2505 cmd_ctx->lsm->u.context.channel_name,
2506 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2507 break;
2508 }
2509 case LTTNG_DISABLE_CHANNEL:
2510 {
2511 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2512 cmd_ctx->lsm->u.disable.channel_name);
2513 break;
2514 }
2515 case LTTNG_DISABLE_EVENT:
2516 {
2517 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2518 cmd_ctx->lsm->u.disable.channel_name,
2519 cmd_ctx->lsm->u.disable.name);
2520 break;
2521 }
2522 case LTTNG_DISABLE_ALL_EVENT:
2523 {
2524 DBG("Disabling all events");
2525
2526 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2527 cmd_ctx->lsm->u.disable.channel_name);
2528 break;
2529 }
2530 case LTTNG_ENABLE_CHANNEL:
2531 {
2532 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2533 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2534 break;
2535 }
2536 case LTTNG_ENABLE_EVENT:
2537 {
2538 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2539 cmd_ctx->lsm->u.enable.channel_name,
2540 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2541 break;
2542 }
2543 case LTTNG_ENABLE_ALL_EVENT:
2544 {
2545 DBG("Enabling all events");
2546
2547 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2548 cmd_ctx->lsm->u.enable.channel_name,
2549 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2550 break;
2551 }
2552 case LTTNG_LIST_TRACEPOINTS:
2553 {
2554 struct lttng_event *events;
2555 ssize_t nb_events;
2556
2557 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2558 if (nb_events < 0) {
2559 /* Return value is a negative lttng_error_code. */
2560 ret = -nb_events;
2561 goto error;
2562 }
2563
2564 /*
2565 * Setup lttng message with payload size set to the event list size in
2566 * bytes and then copy list into the llm payload.
2567 */
2568 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2569 if (ret < 0) {
2570 free(events);
2571 goto setup_error;
2572 }
2573
2574 /* Copy event list into message payload */
2575 memcpy(cmd_ctx->llm->payload, events,
2576 sizeof(struct lttng_event) * nb_events);
2577
2578 free(events);
2579
2580 ret = LTTNG_OK;
2581 break;
2582 }
2583 case LTTNG_LIST_TRACEPOINT_FIELDS:
2584 {
2585 struct lttng_event_field *fields;
2586 ssize_t nb_fields;
2587
2588 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2589 &fields);
2590 if (nb_fields < 0) {
2591 /* Return value is a negative lttng_error_code. */
2592 ret = -nb_fields;
2593 goto error;
2594 }
2595
2596 /*
2597 * Setup lttng message with payload size set to the event list size in
2598 * bytes and then copy list into the llm payload.
2599 */
2600 ret = setup_lttng_msg(cmd_ctx,
2601 sizeof(struct lttng_event_field) * nb_fields);
2602 if (ret < 0) {
2603 free(fields);
2604 goto setup_error;
2605 }
2606
2607 /* Copy event list into message payload */
2608 memcpy(cmd_ctx->llm->payload, fields,
2609 sizeof(struct lttng_event_field) * nb_fields);
2610
2611 free(fields);
2612
2613 ret = LTTNG_OK;
2614 break;
2615 }
2616 case LTTNG_SET_CONSUMER_URI:
2617 {
2618 size_t nb_uri, len;
2619 struct lttng_uri *uris;
2620
2621 nb_uri = cmd_ctx->lsm->u.uri.size;
2622 len = nb_uri * sizeof(struct lttng_uri);
2623
2624 if (nb_uri == 0) {
2625 ret = LTTNG_ERR_INVALID;
2626 goto error;
2627 }
2628
2629 uris = zmalloc(len);
2630 if (uris == NULL) {
2631 ret = LTTNG_ERR_FATAL;
2632 goto error;
2633 }
2634
2635 /* Receive variable len data */
2636 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2637 ret = lttcomm_recv_unix_sock(sock, uris, len);
2638 if (ret <= 0) {
2639 DBG("No URIs received from client... continuing");
2640 *sock_error = 1;
2641 ret = LTTNG_ERR_SESSION_FAIL;
2642 free(uris);
2643 goto error;
2644 }
2645
2646 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2647 nb_uri, uris);
2648 if (ret != LTTNG_OK) {
2649 free(uris);
2650 goto error;
2651 }
2652
2653 /*
2654 * XXX: 0 means that this URI should be applied on the session. Should
2655 * be a DOMAIN enuam.
2656 */
2657 if (cmd_ctx->lsm->domain.type == 0) {
2658 /* Add the URI for the UST session if a consumer is present. */
2659 if (cmd_ctx->session->ust_session &&
2660 cmd_ctx->session->ust_session->consumer) {
2661 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2662 nb_uri, uris);
2663 } else if (cmd_ctx->session->kernel_session &&
2664 cmd_ctx->session->kernel_session->consumer) {
2665 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2666 cmd_ctx->session, nb_uri, uris);
2667 }
2668 }
2669
2670 free(uris);
2671
2672 break;
2673 }
2674 case LTTNG_START_TRACE:
2675 {
2676 ret = cmd_start_trace(cmd_ctx->session);
2677 break;
2678 }
2679 case LTTNG_STOP_TRACE:
2680 {
2681 ret = cmd_stop_trace(cmd_ctx->session);
2682 break;
2683 }
2684 case LTTNG_CREATE_SESSION:
2685 {
2686 size_t nb_uri, len;
2687 struct lttng_uri *uris = NULL;
2688
2689 nb_uri = cmd_ctx->lsm->u.uri.size;
2690 len = nb_uri * sizeof(struct lttng_uri);
2691
2692 if (nb_uri > 0) {
2693 uris = zmalloc(len);
2694 if (uris == NULL) {
2695 ret = LTTNG_ERR_FATAL;
2696 goto error;
2697 }
2698
2699 /* Receive variable len data */
2700 DBG("Waiting for %zu URIs from client ...", nb_uri);
2701 ret = lttcomm_recv_unix_sock(sock, uris, len);
2702 if (ret <= 0) {
2703 DBG("No URIs received from client... continuing");
2704 *sock_error = 1;
2705 ret = LTTNG_ERR_SESSION_FAIL;
2706 free(uris);
2707 goto error;
2708 }
2709
2710 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2711 DBG("Creating session with ONE network URI is a bad call");
2712 ret = LTTNG_ERR_SESSION_FAIL;
2713 free(uris);
2714 goto error;
2715 }
2716 }
2717
2718 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2719 &cmd_ctx->creds);
2720
2721 free(uris);
2722
2723 break;
2724 }
2725 case LTTNG_DESTROY_SESSION:
2726 {
2727 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2728
2729 /* Set session to NULL so we do not unlock it after free. */
2730 cmd_ctx->session = NULL;
2731 break;
2732 }
2733 case LTTNG_LIST_DOMAINS:
2734 {
2735 ssize_t nb_dom;
2736 struct lttng_domain *domains;
2737
2738 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2739 if (nb_dom < 0) {
2740 /* Return value is a negative lttng_error_code. */
2741 ret = -nb_dom;
2742 goto error;
2743 }
2744
2745 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2746 if (ret < 0) {
2747 goto setup_error;
2748 }
2749
2750 /* Copy event list into message payload */
2751 memcpy(cmd_ctx->llm->payload, domains,
2752 nb_dom * sizeof(struct lttng_domain));
2753
2754 free(domains);
2755
2756 ret = LTTNG_OK;
2757 break;
2758 }
2759 case LTTNG_LIST_CHANNELS:
2760 {
2761 int nb_chan;
2762 struct lttng_channel *channels;
2763
2764 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2765 cmd_ctx->session, &channels);
2766 if (nb_chan < 0) {
2767 /* Return value is a negative lttng_error_code. */
2768 ret = -nb_chan;
2769 goto error;
2770 }
2771
2772 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2773 if (ret < 0) {
2774 goto setup_error;
2775 }
2776
2777 /* Copy event list into message payload */
2778 memcpy(cmd_ctx->llm->payload, channels,
2779 nb_chan * sizeof(struct lttng_channel));
2780
2781 free(channels);
2782
2783 ret = LTTNG_OK;
2784 break;
2785 }
2786 case LTTNG_LIST_EVENTS:
2787 {
2788 ssize_t nb_event;
2789 struct lttng_event *events = NULL;
2790
2791 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2792 cmd_ctx->lsm->u.list.channel_name, &events);
2793 if (nb_event < 0) {
2794 /* Return value is a negative lttng_error_code. */
2795 ret = -nb_event;
2796 goto error;
2797 }
2798
2799 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2800 if (ret < 0) {
2801 goto setup_error;
2802 }
2803
2804 /* Copy event list into message payload */
2805 memcpy(cmd_ctx->llm->payload, events,
2806 nb_event * sizeof(struct lttng_event));
2807
2808 free(events);
2809
2810 ret = LTTNG_OK;
2811 break;
2812 }
2813 case LTTNG_LIST_SESSIONS:
2814 {
2815 unsigned int nr_sessions;
2816
2817 session_lock_list();
2818 nr_sessions = lttng_sessions_count(
2819 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2820 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2821
2822 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2823 if (ret < 0) {
2824 session_unlock_list();
2825 goto setup_error;
2826 }
2827
2828 /* Filled the session array */
2829 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2830 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2831 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2832
2833 session_unlock_list();
2834
2835 ret = LTTNG_OK;
2836 break;
2837 }
2838 case LTTNG_CALIBRATE:
2839 {
2840 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2841 &cmd_ctx->lsm->u.calibrate);
2842 break;
2843 }
2844 case LTTNG_REGISTER_CONSUMER:
2845 {
2846 struct consumer_data *cdata;
2847
2848 switch (cmd_ctx->lsm->domain.type) {
2849 case LTTNG_DOMAIN_KERNEL:
2850 cdata = &kconsumer_data;
2851 break;
2852 default:
2853 ret = LTTNG_ERR_UND;
2854 goto error;
2855 }
2856
2857 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2858 cmd_ctx->lsm->u.reg.path, cdata);
2859 break;
2860 }
2861 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2862 {
2863 struct lttng_filter_bytecode *bytecode;
2864
2865 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2866 ret = LTTNG_ERR_FILTER_INVAL;
2867 goto error;
2868 }
2869 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2870 ret = LTTNG_ERR_FILTER_INVAL;
2871 goto error;
2872 }
2873 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2874 if (!bytecode) {
2875 ret = LTTNG_ERR_FILTER_NOMEM;
2876 goto error;
2877 }
2878 /* Receive var. len. data */
2879 DBG("Receiving var len data from client ...");
2880 ret = lttcomm_recv_unix_sock(sock, bytecode,
2881 cmd_ctx->lsm->u.enable.bytecode_len);
2882 if (ret <= 0) {
2883 DBG("Nothing recv() from client var len data... continuing");
2884 *sock_error = 1;
2885 ret = LTTNG_ERR_FILTER_INVAL;
2886 goto error;
2887 }
2888
2889 if (bytecode->len + sizeof(*bytecode)
2890 != cmd_ctx->lsm->u.enable.bytecode_len) {
2891 free(bytecode);
2892 ret = LTTNG_ERR_FILTER_INVAL;
2893 goto error;
2894 }
2895
2896 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2897 cmd_ctx->lsm->u.enable.channel_name,
2898 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2899 break;
2900 }
2901 case LTTNG_DATA_PENDING:
2902 {
2903 ret = cmd_data_pending(cmd_ctx->session);
2904 break;
2905 }
2906 default:
2907 ret = LTTNG_ERR_UND;
2908 break;
2909 }
2910
2911 error:
2912 if (cmd_ctx->llm == NULL) {
2913 DBG("Missing llm structure. Allocating one.");
2914 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2915 goto setup_error;
2916 }
2917 }
2918 /* Set return code */
2919 cmd_ctx->llm->ret_code = ret;
2920 setup_error:
2921 if (cmd_ctx->session) {
2922 session_unlock(cmd_ctx->session);
2923 }
2924 if (need_tracing_session) {
2925 session_unlock_list();
2926 }
2927 init_setup_error:
2928 return ret;
2929 }
2930
2931 /*
2932 * Thread managing health check socket.
2933 */
2934 static void *thread_manage_health(void *data)
2935 {
2936 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2937 uint32_t revents, nb_fd;
2938 struct lttng_poll_event events;
2939 struct lttcomm_health_msg msg;
2940 struct lttcomm_health_data reply;
2941
2942 DBG("[thread] Manage health check started");
2943
2944 rcu_register_thread();
2945
2946 /* Create unix socket */
2947 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2948 if (sock < 0) {
2949 ERR("Unable to create health check Unix socket");
2950 ret = -1;
2951 goto error;
2952 }
2953
2954 /*
2955 * Set the CLOEXEC flag. Return code is useless because either way, the
2956 * show must go on.
2957 */
2958 (void) utils_set_fd_cloexec(sock);
2959
2960 ret = lttcomm_listen_unix_sock(sock);
2961 if (ret < 0) {
2962 goto error;
2963 }
2964
2965 /*
2966 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2967 * more will be added to this poll set.
2968 */
2969 ret = create_thread_poll_set(&events, 2);
2970 if (ret < 0) {
2971 goto error;
2972 }
2973
2974 /* Add the application registration socket */
2975 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2976 if (ret < 0) {
2977 goto error;
2978 }
2979
2980 while (1) {
2981 DBG("Health check ready");
2982
2983 /* Inifinite blocking call, waiting for transmission */
2984 restart:
2985 ret = lttng_poll_wait(&events, -1);
2986 if (ret < 0) {
2987 /*
2988 * Restart interrupted system call.
2989 */
2990 if (errno == EINTR) {
2991 goto restart;
2992 }
2993 goto error;
2994 }
2995
2996 nb_fd = ret;
2997
2998 for (i = 0; i < nb_fd; i++) {
2999 /* Fetch once the poll data */
3000 revents = LTTNG_POLL_GETEV(&events, i);
3001 pollfd = LTTNG_POLL_GETFD(&events, i);
3002
3003 /* Thread quit pipe has been closed. Killing thread. */
3004 ret = check_thread_quit_pipe(pollfd, revents);
3005 if (ret) {
3006 err = 0;
3007 goto exit;
3008 }
3009
3010 /* Event on the registration socket */
3011 if (pollfd == sock) {
3012 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3013 ERR("Health socket poll error");
3014 goto error;
3015 }
3016 }
3017 }
3018
3019 new_sock = lttcomm_accept_unix_sock(sock);
3020 if (new_sock < 0) {
3021 goto error;
3022 }
3023
3024 /*
3025 * Set the CLOEXEC flag. Return code is useless because either way, the
3026 * show must go on.
3027 */
3028 (void) utils_set_fd_cloexec(new_sock);
3029
3030 DBG("Receiving data from client for health...");
3031 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3032 if (ret <= 0) {
3033 DBG("Nothing recv() from client... continuing");
3034 ret = close(new_sock);
3035 if (ret) {
3036 PERROR("close");
3037 }
3038 new_sock = -1;
3039 continue;
3040 }
3041
3042 rcu_thread_online();
3043
3044 switch (msg.component) {
3045 case LTTNG_HEALTH_CMD:
3046 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3047 break;
3048 case LTTNG_HEALTH_APP_MANAGE:
3049 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3050 break;
3051 case LTTNG_HEALTH_APP_REG:
3052 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3053 break;
3054 case LTTNG_HEALTH_KERNEL:
3055 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3056 break;
3057 case LTTNG_HEALTH_CONSUMER:
3058 reply.ret_code = check_consumer_health();
3059 break;
3060 case LTTNG_HEALTH_ALL:
3061 reply.ret_code =
3062 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3063 health_check_state(HEALTH_TYPE_APP_REG) &&
3064 health_check_state(HEALTH_TYPE_CMD) &&
3065 health_check_state(HEALTH_TYPE_KERNEL) &&
3066 check_consumer_health();
3067 break;
3068 default:
3069 reply.ret_code = LTTNG_ERR_UND;
3070 break;
3071 }
3072
3073 /*
3074 * Flip ret value since 0 is a success and 1 indicates a bad health for
3075 * the client where in the sessiond it is the opposite. Again, this is
3076 * just to make things easier for us poor developer which enjoy a lot
3077 * lazyness.
3078 */
3079 if (reply.ret_code == 0 || reply.ret_code == 1) {
3080 reply.ret_code = !reply.ret_code;
3081 }
3082
3083 DBG2("Health check return value %d", reply.ret_code);
3084
3085 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3086 if (ret < 0) {
3087 ERR("Failed to send health data back to client");
3088 }
3089
3090 /* End of transmission */
3091 ret = close(new_sock);
3092 if (ret) {
3093 PERROR("close");
3094 }
3095 new_sock = -1;
3096 }
3097
3098 exit:
3099 error:
3100 if (err) {
3101 ERR("Health error occurred in %s", __func__);
3102 }
3103 DBG("Health check thread dying");
3104 unlink(health_unix_sock_path);
3105 if (sock >= 0) {
3106 ret = close(sock);
3107 if (ret) {
3108 PERROR("close");
3109 }
3110 }
3111 if (new_sock >= 0) {
3112 ret = close(new_sock);
3113 if (ret) {
3114 PERROR("close");
3115 }
3116 }
3117
3118 lttng_poll_clean(&events);
3119
3120 rcu_unregister_thread();
3121 return NULL;
3122 }
3123
3124 /*
3125 * This thread manage all clients request using the unix client socket for
3126 * communication.
3127 */
3128 static void *thread_manage_clients(void *data)
3129 {
3130 int sock = -1, ret, i, pollfd, err = -1;
3131 int sock_error;
3132 uint32_t revents, nb_fd;
3133 struct command_ctx *cmd_ctx = NULL;
3134 struct lttng_poll_event events;
3135
3136 DBG("[thread] Manage client started");
3137
3138 rcu_register_thread();
3139
3140 health_register(HEALTH_TYPE_CMD);
3141
3142 if (testpoint(thread_manage_clients)) {
3143 goto error_testpoint;
3144 }
3145
3146 health_code_update();
3147
3148 ret = lttcomm_listen_unix_sock(client_sock);
3149 if (ret < 0) {
3150 goto error_listen;
3151 }
3152
3153 /*
3154 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3155 * more will be added to this poll set.
3156 */
3157 ret = create_thread_poll_set(&events, 2);
3158 if (ret < 0) {
3159 goto error_create_poll;
3160 }
3161
3162 /* Add the application registration socket */
3163 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3164 if (ret < 0) {
3165 goto error;
3166 }
3167
3168 /*
3169 * Notify parent pid that we are ready to accept command for client side.
3170 */
3171 if (opt_sig_parent) {
3172 kill(ppid, SIGUSR1);
3173 }
3174
3175 if (testpoint(thread_manage_clients_before_loop)) {
3176 goto error;
3177 }
3178
3179 health_code_update();
3180
3181 while (1) {
3182 DBG("Accepting client command ...");
3183
3184 /* Inifinite blocking call, waiting for transmission */
3185 restart:
3186 health_poll_entry();
3187 ret = lttng_poll_wait(&events, -1);
3188 health_poll_exit();
3189 if (ret < 0) {
3190 /*
3191 * Restart interrupted system call.
3192 */
3193 if (errno == EINTR) {
3194 goto restart;
3195 }
3196 goto error;
3197 }
3198
3199 nb_fd = ret;
3200
3201 for (i = 0; i < nb_fd; i++) {
3202 /* Fetch once the poll data */
3203 revents = LTTNG_POLL_GETEV(&events, i);
3204 pollfd = LTTNG_POLL_GETFD(&events, i);
3205
3206 health_code_update();
3207
3208 /* Thread quit pipe has been closed. Killing thread. */
3209 ret = check_thread_quit_pipe(pollfd, revents);
3210 if (ret) {
3211 err = 0;
3212 goto exit;
3213 }
3214
3215 /* Event on the registration socket */
3216 if (pollfd == client_sock) {
3217 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3218 ERR("Client socket poll error");
3219 goto error;
3220 }
3221 }
3222 }
3223
3224 DBG("Wait for client response");
3225
3226 health_code_update();
3227
3228 sock = lttcomm_accept_unix_sock(client_sock);
3229 if (sock < 0) {
3230 goto error;
3231 }
3232
3233 /*
3234 * Set the CLOEXEC flag. Return code is useless because either way, the
3235 * show must go on.
3236 */
3237 (void) utils_set_fd_cloexec(sock);
3238
3239 /* Set socket option for credentials retrieval */
3240 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3241 if (ret < 0) {
3242 goto error;
3243 }
3244
3245 /* Allocate context command to process the client request */
3246 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3247 if (cmd_ctx == NULL) {
3248 PERROR("zmalloc cmd_ctx");
3249 goto error;
3250 }
3251
3252 /* Allocate data buffer for reception */
3253 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3254 if (cmd_ctx->lsm == NULL) {
3255 PERROR("zmalloc cmd_ctx->lsm");
3256 goto error;
3257 }
3258
3259 cmd_ctx->llm = NULL;
3260 cmd_ctx->session = NULL;
3261
3262 health_code_update();
3263
3264 /*
3265 * Data is received from the lttng client. The struct
3266 * lttcomm_session_msg (lsm) contains the command and data request of
3267 * the client.
3268 */
3269 DBG("Receiving data from client ...");
3270 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3271 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3272 if (ret <= 0) {
3273 DBG("Nothing recv() from client... continuing");
3274 ret = close(sock);
3275 if (ret) {
3276 PERROR("close");
3277 }
3278 sock = -1;
3279 clean_command_ctx(&cmd_ctx);
3280 continue;
3281 }
3282
3283 health_code_update();
3284
3285 // TODO: Validate cmd_ctx including sanity check for
3286 // security purpose.
3287
3288 rcu_thread_online();
3289 /*
3290 * This function dispatch the work to the kernel or userspace tracer
3291 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3292 * informations for the client. The command context struct contains
3293 * everything this function may needs.
3294 */
3295 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3296 rcu_thread_offline();
3297 if (ret < 0) {
3298 if (sock_error) {
3299 ret = close(sock);
3300 if (ret) {
3301 PERROR("close");
3302 }
3303 sock = -1;
3304 }
3305 /*
3306 * TODO: Inform client somehow of the fatal error. At
3307 * this point, ret < 0 means that a zmalloc failed
3308 * (ENOMEM). Error detected but still accept
3309 * command, unless a socket error has been
3310 * detected.
3311 */
3312 clean_command_ctx(&cmd_ctx);
3313 continue;
3314 }
3315
3316 health_code_update();
3317
3318 DBG("Sending response (size: %d, retcode: %s)",
3319 cmd_ctx->lttng_msg_size,
3320 lttng_strerror(-cmd_ctx->llm->ret_code));
3321 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3322 if (ret < 0) {
3323 ERR("Failed to send data back to client");
3324 }
3325
3326 /* End of transmission */
3327 ret = close(sock);
3328 if (ret) {
3329 PERROR("close");
3330 }
3331 sock = -1;
3332
3333 clean_command_ctx(&cmd_ctx);
3334
3335 health_code_update();
3336 }
3337
3338 exit:
3339 error:
3340 if (sock >= 0) {
3341 ret = close(sock);
3342 if (ret) {
3343 PERROR("close");
3344 }
3345 }
3346
3347 lttng_poll_clean(&events);
3348 clean_command_ctx(&cmd_ctx);
3349
3350 error_listen:
3351 error_create_poll:
3352 error_testpoint:
3353 unlink(client_unix_sock_path);
3354 if (client_sock >= 0) {
3355 ret = close(client_sock);
3356 if (ret) {
3357 PERROR("close");
3358 }
3359 }
3360
3361 if (err) {
3362 health_error();
3363 ERR("Health error occurred in %s", __func__);
3364 }
3365
3366 health_unregister();
3367
3368 DBG("Client thread dying");
3369
3370 rcu_unregister_thread();
3371 return NULL;
3372 }
3373
3374
3375 /*
3376 * usage function on stderr
3377 */
3378 static void usage(void)
3379 {
3380 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3381 fprintf(stderr, " -h, --help Display this usage.\n");
3382 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3383 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3384 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3385 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3386 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3387 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3388 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3389 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3390 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3391 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3392 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3393 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3394 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3395 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3396 fprintf(stderr, " -V, --version Show version number.\n");
3397 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3398 fprintf(stderr, " -q, --quiet No output at all.\n");
3399 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3400 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3401 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3402 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3403 }
3404
3405 /*
3406 * daemon argument parsing
3407 */
3408 static int parse_args(int argc, char **argv)
3409 {
3410 int c;
3411
3412 static struct option long_options[] = {
3413 { "client-sock", 1, 0, 'c' },
3414 { "apps-sock", 1, 0, 'a' },
3415 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3416 { "kconsumerd-err-sock", 1, 0, 'E' },
3417 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3418 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3419 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3420 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3421 { "consumerd32-path", 1, 0, 'u' },
3422 { "consumerd32-libdir", 1, 0, 'U' },
3423 { "consumerd64-path", 1, 0, 't' },
3424 { "consumerd64-libdir", 1, 0, 'T' },
3425 { "daemonize", 0, 0, 'd' },
3426 { "sig-parent", 0, 0, 'S' },
3427 { "help", 0, 0, 'h' },
3428 { "group", 1, 0, 'g' },
3429 { "version", 0, 0, 'V' },
3430 { "quiet", 0, 0, 'q' },
3431 { "verbose", 0, 0, 'v' },
3432 { "verbose-consumer", 0, 0, 'Z' },
3433 { "no-kernel", 0, 0, 'N' },
3434 { "pidfile", 1, 0, 'p' },
3435 { NULL, 0, 0, 0 }
3436 };
3437
3438 while (1) {
3439 int option_index = 0;
3440 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3441 long_options, &option_index);
3442 if (c == -1) {
3443 break;
3444 }
3445
3446 switch (c) {
3447 case 0:
3448 fprintf(stderr, "option %s", long_options[option_index].name);
3449 if (optarg) {
3450 fprintf(stderr, " with arg %s\n", optarg);
3451 }
3452 break;
3453 case 'c':
3454 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3455 break;
3456 case 'a':
3457 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3458 break;
3459 case 'd':
3460 opt_daemon = 1;
3461 break;
3462 case 'g':
3463 opt_tracing_group = optarg;
3464 break;
3465 case 'h':
3466 usage();
3467 exit(EXIT_FAILURE);
3468 case 'V':
3469 fprintf(stdout, "%s\n", VERSION);
3470 exit(EXIT_SUCCESS);
3471 case 'S':
3472 opt_sig_parent = 1;
3473 break;
3474 case 'E':
3475 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3476 break;
3477 case 'C':
3478 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3479 break;
3480 case 'F':
3481 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3482 break;
3483 case 'D':
3484 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3485 break;
3486 case 'H':
3487 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3488 break;
3489 case 'G':
3490 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3491 break;
3492 case 'N':
3493 opt_no_kernel = 1;
3494 break;
3495 case 'q':
3496 lttng_opt_quiet = 1;
3497 break;
3498 case 'v':
3499 /* Verbose level can increase using multiple -v */
3500 lttng_opt_verbose += 1;
3501 break;
3502 case 'Z':
3503 opt_verbose_consumer += 1;
3504 break;
3505 case 'u':
3506 consumerd32_bin= optarg;
3507 break;
3508 case 'U':
3509 consumerd32_libdir = optarg;
3510 break;
3511 case 't':
3512 consumerd64_bin = optarg;
3513 break;
3514 case 'T':
3515 consumerd64_libdir = optarg;
3516 break;
3517 case 'p':
3518 opt_pidfile = optarg;
3519 break;
3520 default:
3521 /* Unknown option or other error.
3522 * Error is printed by getopt, just return */
3523 return -1;
3524 }
3525 }
3526
3527 return 0;
3528 }
3529
3530 /*
3531 * Creates the two needed socket by the daemon.
3532 * apps_sock - The communication socket for all UST apps.
3533 * client_sock - The communication of the cli tool (lttng).
3534 */
3535 static int init_daemon_socket(void)
3536 {
3537 int ret = 0;
3538 mode_t old_umask;
3539
3540 old_umask = umask(0);
3541
3542 /* Create client tool unix socket */
3543 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3544 if (client_sock < 0) {
3545 ERR("Create unix sock failed: %s", client_unix_sock_path);
3546 ret = -1;
3547 goto end;
3548 }
3549
3550 /* Set the cloexec flag */
3551 ret = utils_set_fd_cloexec(client_sock);
3552 if (ret < 0) {
3553 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3554 "Continuing but note that the consumer daemon will have a "
3555 "reference to this socket on exec()", client_sock);
3556 }
3557
3558 /* File permission MUST be 660 */
3559 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3560 if (ret < 0) {
3561 ERR("Set file permissions failed: %s", client_unix_sock_path);
3562 PERROR("chmod");
3563 goto end;
3564 }
3565
3566 /* Create the application unix socket */
3567 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3568 if (apps_sock < 0) {
3569 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3570 ret = -1;
3571 goto end;
3572 }
3573
3574 /* Set the cloexec flag */
3575 ret = utils_set_fd_cloexec(apps_sock);
3576 if (ret < 0) {
3577 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3578 "Continuing but note that the consumer daemon will have a "
3579 "reference to this socket on exec()", apps_sock);
3580 }
3581
3582 /* File permission MUST be 666 */
3583 ret = chmod(apps_unix_sock_path,
3584 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3585 if (ret < 0) {
3586 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3587 PERROR("chmod");
3588 goto end;
3589 }
3590
3591 DBG3("Session daemon client socket %d and application socket %d created",
3592 client_sock, apps_sock);
3593
3594 end:
3595 umask(old_umask);
3596 return ret;
3597 }
3598
3599 /*
3600 * Check if the global socket is available, and if a daemon is answering at the
3601 * other side. If yes, error is returned.
3602 */
3603 static int check_existing_daemon(void)
3604 {
3605 /* Is there anybody out there ? */
3606 if (lttng_session_daemon_alive()) {
3607 return -EEXIST;
3608 }
3609
3610 return 0;
3611 }
3612
3613 /*
3614 * Set the tracing group gid onto the client socket.
3615 *
3616 * Race window between mkdir and chown is OK because we are going from more
3617 * permissive (root.root) to less permissive (root.tracing).
3618 */
3619 static int set_permissions(char *rundir)
3620 {
3621 int ret;
3622 gid_t gid;
3623
3624 ret = allowed_group();
3625 if (ret < 0) {
3626 WARN("No tracing group detected");
3627 ret = 0;
3628 goto end;
3629 }
3630
3631 gid = ret;
3632
3633 /* Set lttng run dir */
3634 ret = chown(rundir, 0, gid);
3635 if (ret < 0) {
3636 ERR("Unable to set group on %s", rundir);
3637 PERROR("chown");
3638 }
3639
3640 /* Ensure tracing group can search the run dir */
3641 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3642 if (ret < 0) {
3643 ERR("Unable to set permissions on %s", rundir);
3644 PERROR("chmod");
3645 }
3646
3647 /* lttng client socket path */
3648 ret = chown(client_unix_sock_path, 0, gid);
3649 if (ret < 0) {
3650 ERR("Unable to set group on %s", client_unix_sock_path);
3651 PERROR("chown");
3652 }
3653
3654 /* kconsumer error socket path */
3655 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3656 if (ret < 0) {
3657 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3658 PERROR("chown");
3659 }
3660
3661 /* 64-bit ustconsumer error socket path */
3662 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3663 if (ret < 0) {
3664 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3665 PERROR("chown");
3666 }
3667
3668 /* 32-bit ustconsumer compat32 error socket path */
3669 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3670 if (ret < 0) {
3671 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3672 PERROR("chown");
3673 }
3674
3675 DBG("All permissions are set");
3676
3677 end:
3678 return ret;
3679 }
3680
3681 /*
3682 * Create the lttng run directory needed for all global sockets and pipe.
3683 */
3684 static int create_lttng_rundir(const char *rundir)
3685 {
3686 int ret;
3687
3688 DBG3("Creating LTTng run directory: %s", rundir);
3689
3690 ret = mkdir(rundir, S_IRWXU);
3691 if (ret < 0) {
3692 if (errno != EEXIST) {
3693 ERR("Unable to create %s", rundir);
3694 goto error;
3695 } else {
3696 ret = 0;
3697 }
3698 }
3699
3700 error:
3701 return ret;
3702 }
3703
3704 /*
3705 * Setup sockets and directory needed by the kconsumerd communication with the
3706 * session daemon.
3707 */
3708 static int set_consumer_sockets(struct consumer_data *consumer_data,
3709 const char *rundir)
3710 {
3711 int ret;
3712 char path[PATH_MAX];
3713
3714 switch (consumer_data->type) {
3715 case LTTNG_CONSUMER_KERNEL:
3716 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3717 break;
3718 case LTTNG_CONSUMER64_UST:
3719 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3720 break;
3721 case LTTNG_CONSUMER32_UST:
3722 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3723 break;
3724 default:
3725 ERR("Consumer type unknown");
3726 ret = -EINVAL;
3727 goto error;
3728 }
3729
3730 DBG2("Creating consumer directory: %s", path);
3731
3732 ret = mkdir(path, S_IRWXU);
3733 if (ret < 0) {
3734 if (errno != EEXIST) {
3735 PERROR("mkdir");
3736 ERR("Failed to create %s", path);
3737 goto error;
3738 }
3739 ret = -1;
3740 }
3741
3742 /* Create the kconsumerd error unix socket */
3743 consumer_data->err_sock =
3744 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3745 if (consumer_data->err_sock < 0) {
3746 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3747 ret = -1;
3748 goto error;
3749 }
3750
3751 /* File permission MUST be 660 */
3752 ret = chmod(consumer_data->err_unix_sock_path,
3753 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3754 if (ret < 0) {
3755 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3756 PERROR("chmod");
3757 goto error;
3758 }
3759
3760 error:
3761 return ret;
3762 }
3763
3764 /*
3765 * Signal handler for the daemon
3766 *
3767 * Simply stop all worker threads, leaving main() return gracefully after
3768 * joining all threads and calling cleanup().
3769 */
3770 static void sighandler(int sig)
3771 {
3772 switch (sig) {
3773 case SIGPIPE:
3774 DBG("SIGPIPE caught");
3775 return;
3776 case SIGINT:
3777 DBG("SIGINT caught");
3778 stop_threads();
3779 break;
3780 case SIGTERM:
3781 DBG("SIGTERM caught");
3782 stop_threads();
3783 break;
3784 default:
3785 break;
3786 }
3787 }
3788
3789 /*
3790 * Setup signal handler for :
3791 * SIGINT, SIGTERM, SIGPIPE
3792 */
3793 static int set_signal_handler(void)
3794 {
3795 int ret = 0;
3796 struct sigaction sa;
3797 sigset_t sigset;
3798
3799 if ((ret = sigemptyset(&sigset)) < 0) {
3800 PERROR("sigemptyset");
3801 return ret;
3802 }
3803
3804 sa.sa_handler = sighandler;
3805 sa.sa_mask = sigset;
3806 sa.sa_flags = 0;
3807 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3808 PERROR("sigaction");
3809 return ret;
3810 }
3811
3812 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3813 PERROR("sigaction");
3814 return ret;
3815 }
3816
3817 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3818 PERROR("sigaction");
3819 return ret;
3820 }
3821
3822 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3823
3824 return ret;
3825 }
3826
3827 /*
3828 * Set open files limit to unlimited. This daemon can open a large number of
3829 * file descriptors in order to consumer multiple kernel traces.
3830 */
3831 static void set_ulimit(void)
3832 {
3833 int ret;
3834 struct rlimit lim;
3835
3836 /* The kernel does not allowed an infinite limit for open files */
3837 lim.rlim_cur = 65535;
3838 lim.rlim_max = 65535;
3839
3840 ret = setrlimit(RLIMIT_NOFILE, &lim);
3841 if (ret < 0) {
3842 PERROR("failed to set open files limit");
3843 }
3844 }
3845
3846 /*
3847 * Write pidfile using the rundir and opt_pidfile.
3848 */
3849 static void write_pidfile(void)
3850 {
3851 int ret;
3852 char pidfile_path[PATH_MAX];
3853
3854 assert(rundir);
3855
3856 if (opt_pidfile) {
3857 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
3858 } else {
3859 /* Build pidfile path from rundir and opt_pidfile. */
3860 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
3861 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
3862 if (ret < 0) {
3863 PERROR("snprintf pidfile path");
3864 goto error;
3865 }
3866 }
3867
3868 /*
3869 * Create pid file in rundir. Return value is of no importance. The
3870 * execution will continue even though we are not able to write the file.
3871 */
3872 (void) utils_create_pid_file(getpid(), pidfile_path);
3873
3874 error:
3875 return;
3876 }
3877
3878 /*
3879 * main
3880 */
3881 int main(int argc, char **argv)
3882 {
3883 int ret = 0;
3884 void *status;
3885 const char *home_path, *env_app_timeout;
3886
3887 init_kernel_workarounds();
3888
3889 rcu_register_thread();
3890
3891 setup_consumerd_path();
3892
3893 /* Parse arguments */
3894 progname = argv[0];
3895 if ((ret = parse_args(argc, argv)) < 0) {
3896 goto error;
3897 }
3898
3899 /* Daemonize */
3900 if (opt_daemon) {
3901 int i;
3902
3903 /*
3904 * fork
3905 * child: setsid, close FD 0, 1, 2, chdir /
3906 * parent: exit (if fork is successful)
3907 */
3908 ret = daemon(0, 0);
3909 if (ret < 0) {
3910 PERROR("daemon");
3911 goto error;
3912 }
3913 /*
3914 * We are in the child. Make sure all other file
3915 * descriptors are closed, in case we are called with
3916 * more opened file descriptors than the standard ones.
3917 */
3918 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3919 (void) close(i);
3920 }
3921 }
3922
3923 /* Create thread quit pipe */
3924 if ((ret = init_thread_quit_pipe()) < 0) {
3925 goto error;
3926 }
3927
3928 /* Check if daemon is UID = 0 */
3929 is_root = !getuid();
3930
3931 if (is_root) {
3932 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3933
3934 /* Create global run dir with root access */
3935 ret = create_lttng_rundir(rundir);
3936 if (ret < 0) {
3937 goto error;
3938 }
3939
3940 if (strlen(apps_unix_sock_path) == 0) {
3941 snprintf(apps_unix_sock_path, PATH_MAX,
3942 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3943 }
3944
3945 if (strlen(client_unix_sock_path) == 0) {
3946 snprintf(client_unix_sock_path, PATH_MAX,
3947 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3948 }
3949
3950 /* Set global SHM for ust */
3951 if (strlen(wait_shm_path) == 0) {
3952 snprintf(wait_shm_path, PATH_MAX,
3953 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3954 }
3955
3956 if (strlen(health_unix_sock_path) == 0) {
3957 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3958 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3959 }
3960
3961 /* Setup kernel consumerd path */
3962 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3963 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3964 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3965 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3966
3967 DBG2("Kernel consumer err path: %s",
3968 kconsumer_data.err_unix_sock_path);
3969 DBG2("Kernel consumer cmd path: %s",
3970 kconsumer_data.cmd_unix_sock_path);
3971 } else {
3972 home_path = get_home_dir();
3973 if (home_path == NULL) {
3974 /* TODO: Add --socket PATH option */
3975 ERR("Can't get HOME directory for sockets creation.");
3976 ret = -EPERM;
3977 goto error;
3978 }
3979
3980 /*
3981 * Create rundir from home path. This will create something like
3982 * $HOME/.lttng
3983 */
3984 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3985 if (ret < 0) {
3986 ret = -ENOMEM;
3987 goto error;
3988 }
3989
3990 ret = create_lttng_rundir(rundir);
3991 if (ret < 0) {
3992 goto error;
3993 }
3994
3995 if (strlen(apps_unix_sock_path) == 0) {
3996 snprintf(apps_unix_sock_path, PATH_MAX,
3997 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3998 }
3999
4000 /* Set the cli tool unix socket path */
4001 if (strlen(client_unix_sock_path) == 0) {
4002 snprintf(client_unix_sock_path, PATH_MAX,
4003 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4004 }
4005
4006 /* Set global SHM for ust */
4007 if (strlen(wait_shm_path) == 0) {
4008 snprintf(wait_shm_path, PATH_MAX,
4009 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
4010 }
4011
4012 /* Set health check Unix path */
4013 if (strlen(health_unix_sock_path) == 0) {
4014 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4015 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4016 }
4017 }
4018
4019 /* Set consumer initial state */
4020 kernel_consumerd_state = CONSUMER_STOPPED;
4021 ust_consumerd_state = CONSUMER_STOPPED;
4022
4023 DBG("Client socket path %s", client_unix_sock_path);
4024 DBG("Application socket path %s", apps_unix_sock_path);
4025 DBG("LTTng run directory path: %s", rundir);
4026
4027 /* 32 bits consumerd path setup */
4028 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4029 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4030 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4031 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4032
4033 DBG2("UST consumer 32 bits err path: %s",
4034 ustconsumer32_data.err_unix_sock_path);
4035 DBG2("UST consumer 32 bits cmd path: %s",
4036 ustconsumer32_data.cmd_unix_sock_path);
4037
4038 /* 64 bits consumerd path setup */
4039 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4040 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4041 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4042 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4043
4044 DBG2("UST consumer 64 bits err path: %s",
4045 ustconsumer64_data.err_unix_sock_path);
4046 DBG2("UST consumer 64 bits cmd path: %s",
4047 ustconsumer64_data.cmd_unix_sock_path);
4048
4049 /*
4050 * See if daemon already exist.
4051 */
4052 if ((ret = check_existing_daemon()) < 0) {
4053 ERR("Already running daemon.\n");
4054 /*
4055 * We do not goto exit because we must not cleanup()
4056 * because a daemon is already running.
4057 */
4058 goto error;
4059 }
4060
4061 /*
4062 * Init UST app hash table. Alloc hash table before this point since
4063 * cleanup() can get called after that point.
4064 */
4065 ust_app_ht_alloc();
4066
4067 /* After this point, we can safely call cleanup() with "goto exit" */
4068
4069 /*
4070 * These actions must be executed as root. We do that *after* setting up
4071 * the sockets path because we MUST make the check for another daemon using
4072 * those paths *before* trying to set the kernel consumer sockets and init
4073 * kernel tracer.
4074 */
4075 if (is_root) {
4076 ret = set_consumer_sockets(&kconsumer_data, rundir);
4077 if (ret < 0) {
4078 goto exit;
4079 }
4080
4081 /* Setup kernel tracer */
4082 if (!opt_no_kernel) {
4083 init_kernel_tracer();
4084 }
4085
4086 /* Set ulimit for open files */
4087 set_ulimit();
4088 }
4089 /* init lttng_fd tracking must be done after set_ulimit. */
4090 lttng_fd_init();
4091
4092 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4093 if (ret < 0) {
4094 goto exit;
4095 }
4096
4097 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4098 if (ret < 0) {
4099 goto exit;
4100 }
4101
4102 if ((ret = set_signal_handler()) < 0) {
4103 goto exit;
4104 }
4105
4106 /* Setup the needed unix socket */
4107 if ((ret = init_daemon_socket()) < 0) {
4108 goto exit;
4109 }
4110
4111 /* Set credentials to socket */
4112 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4113 goto exit;
4114 }
4115
4116 /* Get parent pid if -S, --sig-parent is specified. */
4117 if (opt_sig_parent) {
4118 ppid = getppid();
4119 }
4120
4121 /* Setup the kernel pipe for waking up the kernel thread */
4122 if (is_root && !opt_no_kernel) {
4123 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4124 goto exit;
4125 }
4126 }
4127
4128 /* Setup the thread apps communication pipe. */
4129 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4130 goto exit;
4131 }
4132
4133 /* Init UST command queue. */
4134 cds_wfq_init(&ust_cmd_queue.queue);
4135
4136 /*
4137 * Get session list pointer. This pointer MUST NOT be free(). This list is
4138 * statically declared in session.c
4139 */
4140 session_list_ptr = session_get_list();
4141
4142 /* Set up max poll set size */
4143 lttng_poll_set_max_size();
4144
4145 cmd_init();
4146
4147 /* Check for the application socket timeout env variable. */
4148 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4149 if (env_app_timeout) {
4150 app_socket_timeout = atoi(env_app_timeout);
4151 } else {
4152 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4153 }
4154
4155 write_pidfile();
4156
4157 /* Create thread to manage the client socket */
4158 ret = pthread_create(&health_thread, NULL,
4159 thread_manage_health, (void *) NULL);
4160 if (ret != 0) {
4161 PERROR("pthread_create health");
4162 goto exit_health;
4163 }
4164
4165 /* Create thread to manage the client socket */
4166 ret = pthread_create(&client_thread, NULL,
4167 thread_manage_clients, (void *) NULL);
4168 if (ret != 0) {
4169 PERROR("pthread_create clients");
4170 goto exit_client;
4171 }
4172
4173 /* Create thread to dispatch registration */
4174 ret = pthread_create(&dispatch_thread, NULL,
4175 thread_dispatch_ust_registration, (void *) NULL);
4176 if (ret != 0) {
4177 PERROR("pthread_create dispatch");
4178 goto exit_dispatch;
4179 }
4180
4181 /* Create thread to manage application registration. */
4182 ret = pthread_create(&reg_apps_thread, NULL,
4183 thread_registration_apps, (void *) NULL);
4184 if (ret != 0) {
4185 PERROR("pthread_create registration");
4186 goto exit_reg_apps;
4187 }
4188
4189 /* Create thread to manage application socket */
4190 ret = pthread_create(&apps_thread, NULL,
4191 thread_manage_apps, (void *) NULL);
4192 if (ret != 0) {
4193 PERROR("pthread_create apps");
4194 goto exit_apps;
4195 }
4196
4197 /* Don't start this thread if kernel tracing is not requested nor root */
4198 if (is_root && !opt_no_kernel) {
4199 /* Create kernel thread to manage kernel event */
4200 ret = pthread_create(&kernel_thread, NULL,
4201 thread_manage_kernel, (void *) NULL);
4202 if (ret != 0) {
4203 PERROR("pthread_create kernel");
4204 goto exit_kernel;
4205 }
4206
4207 ret = pthread_join(kernel_thread, &status);
4208 if (ret != 0) {
4209 PERROR("pthread_join");
4210 goto error; /* join error, exit without cleanup */
4211 }
4212 }
4213
4214 exit_kernel:
4215 ret = pthread_join(apps_thread, &status);
4216 if (ret != 0) {
4217 PERROR("pthread_join");
4218 goto error; /* join error, exit without cleanup */
4219 }
4220
4221 exit_apps:
4222 ret = pthread_join(reg_apps_thread, &status);
4223 if (ret != 0) {
4224 PERROR("pthread_join");
4225 goto error; /* join error, exit without cleanup */
4226 }
4227
4228 exit_reg_apps:
4229 ret = pthread_join(dispatch_thread, &status);
4230 if (ret != 0) {
4231 PERROR("pthread_join");
4232 goto error; /* join error, exit without cleanup */
4233 }
4234
4235 exit_dispatch:
4236 ret = pthread_join(client_thread, &status);
4237 if (ret != 0) {
4238 PERROR("pthread_join");
4239 goto error; /* join error, exit without cleanup */
4240 }
4241
4242 ret = join_consumer_thread(&kconsumer_data);
4243 if (ret != 0) {
4244 PERROR("join_consumer");
4245 goto error; /* join error, exit without cleanup */
4246 }
4247
4248 ret = join_consumer_thread(&ustconsumer32_data);
4249 if (ret != 0) {
4250 PERROR("join_consumer ust32");
4251 goto error; /* join error, exit without cleanup */
4252 }
4253
4254 ret = join_consumer_thread(&ustconsumer64_data);
4255 if (ret != 0) {
4256 PERROR("join_consumer ust64");
4257 goto error; /* join error, exit without cleanup */
4258 }
4259
4260 exit_client:
4261 ret = pthread_join(health_thread, &status);
4262 if (ret != 0) {
4263 PERROR("pthread_join health thread");
4264 goto error; /* join error, exit without cleanup */
4265 }
4266
4267 exit_health:
4268 exit:
4269 /*
4270 * cleanup() is called when no other thread is running.
4271 */
4272 rcu_thread_online();
4273 cleanup();
4274 rcu_thread_offline();
4275 rcu_unregister_thread();
4276 if (!ret) {
4277 exit(EXIT_SUCCESS);
4278 }
4279 error:
4280 exit(EXIT_FAILURE);
4281 }
This page took 0.161574 seconds and 6 git commands to generate.