Move UST registry into sessiond and implement notifiers
[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/socket.h>
41 #include <common/defaults.h>
42 #include <common/kernel-consumer/kernel-consumer.h>
43 #include <common/futex.h>
44 #include <common/relayd/relayd.h>
45 #include <common/utils.h>
46
47 #include "lttng-sessiond.h"
48 #include "channel.h"
49 #include "cmd.h"
50 #include "consumer.h"
51 #include "context.h"
52 #include "event.h"
53 #include "kernel.h"
54 #include "kernel-consumer.h"
55 #include "modprobe.h"
56 #include "shm.h"
57 #include "ust-ctl.h"
58 #include "ust-consumer.h"
59 #include "utils.h"
60 #include "fd-limit.h"
61 #include "health.h"
62 #include "testpoint.h"
63 #include "ust-thread.h"
64
65 #define CONSUMERD_FILE "lttng-consumerd"
66
67 /* Const values */
68 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
69
70 const char *progname;
71 const char *opt_tracing_group;
72 static const char *opt_pidfile;
73 static int opt_sig_parent;
74 static int opt_verbose_consumer;
75 static int opt_daemon;
76 static int opt_no_kernel;
77 static int is_root; /* Set to 1 if the daemon is running as root */
78 static pid_t ppid; /* Parent PID for --sig-parent option */
79 static char *rundir;
80
81 /*
82 * Consumer daemon specific control data. Every value not initialized here is
83 * set to 0 by the static definition.
84 */
85 static struct consumer_data kconsumer_data = {
86 .type = LTTNG_CONSUMER_KERNEL,
87 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
88 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
89 .err_sock = -1,
90 .cmd_sock = -1,
91 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
92 .lock = PTHREAD_MUTEX_INITIALIZER,
93 .cond = PTHREAD_COND_INITIALIZER,
94 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
95 };
96 static struct consumer_data ustconsumer64_data = {
97 .type = LTTNG_CONSUMER64_UST,
98 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
99 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
100 .err_sock = -1,
101 .cmd_sock = -1,
102 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
103 .lock = PTHREAD_MUTEX_INITIALIZER,
104 .cond = PTHREAD_COND_INITIALIZER,
105 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
106 };
107 static struct consumer_data ustconsumer32_data = {
108 .type = LTTNG_CONSUMER32_UST,
109 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
110 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
111 .err_sock = -1,
112 .cmd_sock = -1,
113 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
114 .lock = PTHREAD_MUTEX_INITIALIZER,
115 .cond = PTHREAD_COND_INITIALIZER,
116 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
117 };
118
119 /* Shared between threads */
120 static int dispatch_thread_exit;
121
122 /* Global application Unix socket path */
123 static char apps_unix_sock_path[PATH_MAX];
124 /* Global client Unix socket path */
125 static char client_unix_sock_path[PATH_MAX];
126 /* global wait shm path for UST */
127 static char wait_shm_path[PATH_MAX];
128 /* Global health check unix path */
129 static char health_unix_sock_path[PATH_MAX];
130
131 /* Sockets and FDs */
132 static int client_sock = -1;
133 static int apps_sock = -1;
134 int kernel_tracer_fd = -1;
135 static int kernel_poll_pipe[2] = { -1, -1 };
136
137 /*
138 * Quit pipe for all threads. This permits a single cancellation point
139 * for all threads when receiving an event on the pipe.
140 */
141 static int thread_quit_pipe[2] = { -1, -1 };
142
143 /*
144 * This pipe is used to inform the thread managing application communication
145 * that a command is queued and ready to be processed.
146 */
147 static int apps_cmd_pipe[2] = { -1, -1 };
148
149 int apps_cmd_notify_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread;
153 static pthread_t apps_notify_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 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
283 {
284 int ret;
285
286 assert(events);
287
288 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
289 if (ret < 0) {
290 goto error;
291 }
292
293 /* Add quit pipe */
294 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
295 if (ret < 0) {
296 goto error;
297 }
298
299 return 0;
300
301 error:
302 return ret;
303 }
304
305 /*
306 * Check if the thread quit pipe was triggered.
307 *
308 * Return 1 if it was triggered else 0;
309 */
310 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
311 {
312 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
313 return 1;
314 }
315
316 return 0;
317 }
318
319 /*
320 * Return group ID of the tracing group or -1 if not found.
321 */
322 static gid_t allowed_group(void)
323 {
324 struct group *grp;
325
326 if (opt_tracing_group) {
327 grp = getgrnam(opt_tracing_group);
328 } else {
329 grp = getgrnam(default_tracing_group);
330 }
331 if (!grp) {
332 return -1;
333 } else {
334 return grp->gr_gid;
335 }
336 }
337
338 /*
339 * Init thread quit pipe.
340 *
341 * Return -1 on error or 0 if all pipes are created.
342 */
343 static int init_thread_quit_pipe(void)
344 {
345 int ret, i;
346
347 ret = pipe(thread_quit_pipe);
348 if (ret < 0) {
349 PERROR("thread quit pipe");
350 goto error;
351 }
352
353 for (i = 0; i < 2; i++) {
354 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
355 if (ret < 0) {
356 PERROR("fcntl");
357 goto error;
358 }
359 }
360
361 error:
362 return ret;
363 }
364
365 /*
366 * Stop all threads by closing the thread quit pipe.
367 */
368 static void stop_threads(void)
369 {
370 int ret;
371
372 /* Stopping all threads */
373 DBG("Terminating all threads");
374 ret = notify_thread_pipe(thread_quit_pipe[1]);
375 if (ret < 0) {
376 ERR("write error on thread quit pipe");
377 }
378
379 /* Dispatch thread */
380 CMM_STORE_SHARED(dispatch_thread_exit, 1);
381 futex_nto1_wake(&ust_cmd_queue.futex);
382 }
383
384 /*
385 * Cleanup the daemon
386 */
387 static void cleanup(void)
388 {
389 int ret;
390 char *cmd = NULL;
391 struct ltt_session *sess, *stmp;
392
393 DBG("Cleaning up");
394
395 /* First thing first, stop all threads */
396 utils_close_pipe(thread_quit_pipe);
397
398 /*
399 * If opt_pidfile is undefined, the default file will be wiped when
400 * removing the rundir.
401 */
402 if (opt_pidfile) {
403 ret = remove(opt_pidfile);
404 if (ret < 0) {
405 PERROR("remove pidfile %s", opt_pidfile);
406 }
407 }
408
409 DBG("Removing %s directory", rundir);
410 ret = asprintf(&cmd, "rm -rf %s", rundir);
411 if (ret < 0) {
412 ERR("asprintf failed. Something is really wrong!");
413 }
414
415 /* Remove lttng run directory */
416 ret = system(cmd);
417 if (ret < 0) {
418 ERR("Unable to clean %s", rundir);
419 }
420 free(cmd);
421 free(rundir);
422
423 DBG("Cleaning up all sessions");
424
425 /* Destroy session list mutex */
426 if (session_list_ptr != NULL) {
427 pthread_mutex_destroy(&session_list_ptr->lock);
428
429 /* Cleanup ALL session */
430 cds_list_for_each_entry_safe(sess, stmp,
431 &session_list_ptr->head, list) {
432 cmd_destroy_session(sess, kernel_poll_pipe[1]);
433 }
434 }
435
436 DBG("Closing all UST sockets");
437 ust_app_clean_list();
438
439 if (is_root && !opt_no_kernel) {
440 DBG2("Closing kernel fd");
441 if (kernel_tracer_fd >= 0) {
442 ret = close(kernel_tracer_fd);
443 if (ret) {
444 PERROR("close");
445 }
446 }
447 DBG("Unloading kernel modules");
448 modprobe_remove_lttng_all();
449 }
450
451 /* <fun> */
452 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
453 "Matthew, BEET driven development works!%c[%dm",
454 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
455 /* </fun> */
456 }
457
458 /*
459 * Send data on a unix socket using the liblttsessiondcomm API.
460 *
461 * Return lttcomm error code.
462 */
463 static int send_unix_sock(int sock, void *buf, size_t len)
464 {
465 /* Check valid length */
466 if (len == 0) {
467 return -1;
468 }
469
470 return lttcomm_send_unix_sock(sock, buf, len);
471 }
472
473 /*
474 * Free memory of a command context structure.
475 */
476 static void clean_command_ctx(struct command_ctx **cmd_ctx)
477 {
478 DBG("Clean command context structure");
479 if (*cmd_ctx) {
480 if ((*cmd_ctx)->llm) {
481 free((*cmd_ctx)->llm);
482 }
483 if ((*cmd_ctx)->lsm) {
484 free((*cmd_ctx)->lsm);
485 }
486 free(*cmd_ctx);
487 *cmd_ctx = NULL;
488 }
489 }
490
491 /*
492 * Notify UST applications using the shm mmap futex.
493 */
494 static int notify_ust_apps(int active)
495 {
496 char *wait_shm_mmap;
497
498 DBG("Notifying applications of session daemon state: %d", active);
499
500 /* See shm.c for this call implying mmap, shm and futex calls */
501 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
502 if (wait_shm_mmap == NULL) {
503 goto error;
504 }
505
506 /* Wake waiting process */
507 futex_wait_update((int32_t *) wait_shm_mmap, active);
508
509 /* Apps notified successfully */
510 return 0;
511
512 error:
513 return -1;
514 }
515
516 /*
517 * Setup the outgoing data buffer for the response (llm) by allocating the
518 * right amount of memory and copying the original information from the lsm
519 * structure.
520 *
521 * Return total size of the buffer pointed by buf.
522 */
523 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
524 {
525 int ret, buf_size;
526
527 buf_size = size;
528
529 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
530 if (cmd_ctx->llm == NULL) {
531 PERROR("zmalloc");
532 ret = -ENOMEM;
533 goto error;
534 }
535
536 /* Copy common data */
537 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
538 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
539
540 cmd_ctx->llm->data_size = size;
541 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
542
543 return buf_size;
544
545 error:
546 return ret;
547 }
548
549 /*
550 * Update the kernel poll set of all channel fd available over all tracing
551 * session. Add the wakeup pipe at the end of the set.
552 */
553 static int update_kernel_poll(struct lttng_poll_event *events)
554 {
555 int ret;
556 struct ltt_session *session;
557 struct ltt_kernel_channel *channel;
558
559 DBG("Updating kernel poll set");
560
561 session_lock_list();
562 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
563 session_lock(session);
564 if (session->kernel_session == NULL) {
565 session_unlock(session);
566 continue;
567 }
568
569 cds_list_for_each_entry(channel,
570 &session->kernel_session->channel_list.head, list) {
571 /* Add channel fd to the kernel poll set */
572 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
573 if (ret < 0) {
574 session_unlock(session);
575 goto error;
576 }
577 DBG("Channel fd %d added to kernel set", channel->fd);
578 }
579 session_unlock(session);
580 }
581 session_unlock_list();
582
583 return 0;
584
585 error:
586 session_unlock_list();
587 return -1;
588 }
589
590 /*
591 * Find the channel fd from 'fd' over all tracing session. When found, check
592 * for new channel stream and send those stream fds to the kernel consumer.
593 *
594 * Useful for CPU hotplug feature.
595 */
596 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
597 {
598 int ret = 0;
599 struct ltt_session *session;
600 struct ltt_kernel_session *ksess;
601 struct ltt_kernel_channel *channel;
602
603 DBG("Updating kernel streams for channel fd %d", fd);
604
605 session_lock_list();
606 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
607 session_lock(session);
608 if (session->kernel_session == NULL) {
609 session_unlock(session);
610 continue;
611 }
612 ksess = session->kernel_session;
613
614 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
615 if (channel->fd == fd) {
616 DBG("Channel found, updating kernel streams");
617 ret = kernel_open_channel_stream(channel);
618 if (ret < 0) {
619 goto error;
620 }
621
622 /*
623 * Have we already sent fds to the consumer? If yes, it means
624 * that tracing is started so it is safe to send our updated
625 * stream fds.
626 */
627 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
628 struct lttng_ht_iter iter;
629 struct consumer_socket *socket;
630
631 rcu_read_lock();
632 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
633 &iter.iter, socket, node.node) {
634 /* Code flow error */
635 assert(socket->fd >= 0);
636
637 pthread_mutex_lock(socket->lock);
638 ret = kernel_consumer_send_channel_stream(socket,
639 channel, ksess);
640 pthread_mutex_unlock(socket->lock);
641 if (ret < 0) {
642 rcu_read_unlock();
643 goto error;
644 }
645 }
646 rcu_read_unlock();
647 }
648 goto error;
649 }
650 }
651 session_unlock(session);
652 }
653 session_unlock_list();
654 return ret;
655
656 error:
657 session_unlock(session);
658 session_unlock_list();
659 return ret;
660 }
661
662 /*
663 * For each tracing session, update newly registered apps. The session list
664 * lock MUST be acquired before calling this.
665 */
666 static void update_ust_app(int app_sock)
667 {
668 struct ltt_session *sess, *stmp;
669
670 /* For all tracing session(s) */
671 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
672 session_lock(sess);
673 if (sess->ust_session) {
674 ust_app_global_update(sess->ust_session, app_sock);
675 }
676 session_unlock(sess);
677 }
678 }
679
680 /*
681 * This thread manage event coming from the kernel.
682 *
683 * Features supported in this thread:
684 * -) CPU Hotplug
685 */
686 static void *thread_manage_kernel(void *data)
687 {
688 int ret, i, pollfd, update_poll_flag = 1, err = -1;
689 uint32_t revents, nb_fd;
690 char tmp;
691 struct lttng_poll_event events;
692
693 DBG("[thread] Thread manage kernel started");
694
695 health_register(HEALTH_TYPE_KERNEL);
696
697 /*
698 * This first step of the while is to clean this structure which could free
699 * non NULL pointers so zero it before the loop.
700 */
701 memset(&events, 0, sizeof(events));
702
703 if (testpoint(thread_manage_kernel)) {
704 goto error_testpoint;
705 }
706
707 health_code_update();
708
709 if (testpoint(thread_manage_kernel_before_loop)) {
710 goto error_testpoint;
711 }
712
713 while (1) {
714 health_code_update();
715
716 if (update_poll_flag == 1) {
717 /* Clean events object. We are about to populate it again. */
718 lttng_poll_clean(&events);
719
720 ret = sessiond_set_thread_pollset(&events, 2);
721 if (ret < 0) {
722 goto error_poll_create;
723 }
724
725 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
726 if (ret < 0) {
727 goto error;
728 }
729
730 /* This will add the available kernel channel if any. */
731 ret = update_kernel_poll(&events);
732 if (ret < 0) {
733 goto error;
734 }
735 update_poll_flag = 0;
736 }
737
738 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
739
740 /* Poll infinite value of time */
741 restart:
742 health_poll_entry();
743 ret = lttng_poll_wait(&events, -1);
744 health_poll_exit();
745 if (ret < 0) {
746 /*
747 * Restart interrupted system call.
748 */
749 if (errno == EINTR) {
750 goto restart;
751 }
752 goto error;
753 } else if (ret == 0) {
754 /* Should not happen since timeout is infinite */
755 ERR("Return value of poll is 0 with an infinite timeout.\n"
756 "This should not have happened! Continuing...");
757 continue;
758 }
759
760 nb_fd = ret;
761
762 for (i = 0; i < nb_fd; i++) {
763 /* Fetch once the poll data */
764 revents = LTTNG_POLL_GETEV(&events, i);
765 pollfd = LTTNG_POLL_GETFD(&events, i);
766
767 health_code_update();
768
769 /* Thread quit pipe has been closed. Killing thread. */
770 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
771 if (ret) {
772 err = 0;
773 goto exit;
774 }
775
776 /* Check for data on kernel pipe */
777 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
778 do {
779 ret = read(kernel_poll_pipe[0], &tmp, 1);
780 } while (ret < 0 && errno == EINTR);
781 /*
782 * Ret value is useless here, if this pipe gets any actions an
783 * update is required anyway.
784 */
785 update_poll_flag = 1;
786 continue;
787 } else {
788 /*
789 * New CPU detected by the kernel. Adding kernel stream to
790 * kernel session and updating the kernel consumer
791 */
792 if (revents & LPOLLIN) {
793 ret = update_kernel_stream(&kconsumer_data, pollfd);
794 if (ret < 0) {
795 continue;
796 }
797 break;
798 /*
799 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
800 * and unregister kernel stream at this point.
801 */
802 }
803 }
804 }
805 }
806
807 exit:
808 error:
809 lttng_poll_clean(&events);
810 error_poll_create:
811 error_testpoint:
812 utils_close_pipe(kernel_poll_pipe);
813 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
814 if (err) {
815 health_error();
816 ERR("Health error occurred in %s", __func__);
817 WARN("Kernel thread died unexpectedly. "
818 "Kernel tracing can continue but CPU hotplug is disabled.");
819 }
820 health_unregister();
821 DBG("Kernel thread dying");
822 return NULL;
823 }
824
825 /*
826 * Signal pthread condition of the consumer data that the thread.
827 */
828 static void signal_consumer_condition(struct consumer_data *data, int state)
829 {
830 pthread_mutex_lock(&data->cond_mutex);
831
832 /*
833 * The state is set before signaling. It can be any value, it's the waiter
834 * job to correctly interpret this condition variable associated to the
835 * consumer pthread_cond.
836 *
837 * A value of 0 means that the corresponding thread of the consumer data
838 * was not started. 1 indicates that the thread has started and is ready
839 * for action. A negative value means that there was an error during the
840 * thread bootstrap.
841 */
842 data->consumer_thread_is_ready = state;
843 (void) pthread_cond_signal(&data->cond);
844
845 pthread_mutex_unlock(&data->cond_mutex);
846 }
847
848 /*
849 * This thread manage the consumer error sent back to the session daemon.
850 */
851 static void *thread_manage_consumer(void *data)
852 {
853 int sock = -1, i, ret, pollfd, err = -1;
854 uint32_t revents, nb_fd;
855 enum lttcomm_return_code code;
856 struct lttng_poll_event events;
857 struct consumer_data *consumer_data = data;
858
859 DBG("[thread] Manage consumer started");
860
861 health_register(HEALTH_TYPE_CONSUMER);
862
863 health_code_update();
864
865 /*
866 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
867 * Nothing more will be added to this poll set.
868 */
869 ret = sessiond_set_thread_pollset(&events, 2);
870 if (ret < 0) {
871 goto error_poll;
872 }
873
874 /*
875 * The error socket here is already in a listening state which was done
876 * just before spawning this thread to avoid a race between the consumer
877 * daemon exec trying to connect and the listen() call.
878 */
879 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
880 if (ret < 0) {
881 goto error;
882 }
883
884 health_code_update();
885
886 /* Inifinite blocking call, waiting for transmission */
887 restart:
888 health_poll_entry();
889
890 if (testpoint(thread_manage_consumer)) {
891 goto error;
892 }
893
894 ret = lttng_poll_wait(&events, -1);
895 health_poll_exit();
896 if (ret < 0) {
897 /*
898 * Restart interrupted system call.
899 */
900 if (errno == EINTR) {
901 goto restart;
902 }
903 goto error;
904 }
905
906 nb_fd = ret;
907
908 for (i = 0; i < nb_fd; i++) {
909 /* Fetch once the poll data */
910 revents = LTTNG_POLL_GETEV(&events, i);
911 pollfd = LTTNG_POLL_GETFD(&events, i);
912
913 health_code_update();
914
915 /* Thread quit pipe has been closed. Killing thread. */
916 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
917 if (ret) {
918 err = 0;
919 goto exit;
920 }
921
922 /* Event on the registration socket */
923 if (pollfd == consumer_data->err_sock) {
924 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
925 ERR("consumer err socket poll error");
926 goto error;
927 }
928 }
929 }
930
931 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
932 if (sock < 0) {
933 goto error;
934 }
935
936 /*
937 * Set the CLOEXEC flag. Return code is useless because either way, the
938 * show must go on.
939 */
940 (void) utils_set_fd_cloexec(sock);
941
942 health_code_update();
943
944 DBG2("Receiving code from consumer err_sock");
945
946 /* Getting status code from kconsumerd */
947 ret = lttcomm_recv_unix_sock(sock, &code,
948 sizeof(enum lttcomm_return_code));
949 if (ret <= 0) {
950 goto error;
951 }
952
953 health_code_update();
954
955 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
956 consumer_data->cmd_sock =
957 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
958 if (consumer_data->cmd_sock < 0) {
959 /* On error, signal condition and quit. */
960 signal_consumer_condition(consumer_data, -1);
961 PERROR("consumer connect");
962 goto error;
963 }
964 signal_consumer_condition(consumer_data, 1);
965 DBG("Consumer command socket ready");
966 } else {
967 ERR("consumer error when waiting for SOCK_READY : %s",
968 lttcomm_get_readable_code(-code));
969 goto error;
970 }
971
972 /* Remove the kconsumerd error sock since we've established a connexion */
973 ret = lttng_poll_del(&events, consumer_data->err_sock);
974 if (ret < 0) {
975 goto error;
976 }
977
978 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
979 if (ret < 0) {
980 goto error;
981 }
982
983 health_code_update();
984
985 /* Inifinite blocking call, waiting for transmission */
986 restart_poll:
987 health_poll_entry();
988 ret = lttng_poll_wait(&events, -1);
989 health_poll_exit();
990 if (ret < 0) {
991 /*
992 * Restart interrupted system call.
993 */
994 if (errno == EINTR) {
995 goto restart_poll;
996 }
997 goto error;
998 }
999
1000 nb_fd = ret;
1001
1002 for (i = 0; i < nb_fd; i++) {
1003 /* Fetch once the poll data */
1004 revents = LTTNG_POLL_GETEV(&events, i);
1005 pollfd = LTTNG_POLL_GETFD(&events, i);
1006
1007 health_code_update();
1008
1009 /* Thread quit pipe has been closed. Killing thread. */
1010 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1011 if (ret) {
1012 err = 0;
1013 goto exit;
1014 }
1015
1016 /* Event on the kconsumerd socket */
1017 if (pollfd == sock) {
1018 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1019 ERR("consumer err socket second poll error");
1020 goto error;
1021 }
1022 }
1023 }
1024
1025 health_code_update();
1026
1027 /* Wait for any kconsumerd error */
1028 ret = lttcomm_recv_unix_sock(sock, &code,
1029 sizeof(enum lttcomm_return_code));
1030 if (ret <= 0) {
1031 ERR("consumer closed the command socket");
1032 goto error;
1033 }
1034
1035 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1036
1037 exit:
1038 error:
1039 /* Immediately set the consumerd state to stopped */
1040 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1041 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1042 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1043 consumer_data->type == LTTNG_CONSUMER32_UST) {
1044 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1045 } else {
1046 /* Code flow error... */
1047 assert(0);
1048 }
1049
1050 if (consumer_data->err_sock >= 0) {
1051 ret = close(consumer_data->err_sock);
1052 if (ret) {
1053 PERROR("close");
1054 }
1055 }
1056 if (consumer_data->cmd_sock >= 0) {
1057 ret = close(consumer_data->cmd_sock);
1058 if (ret) {
1059 PERROR("close");
1060 }
1061 }
1062 if (sock >= 0) {
1063 ret = close(sock);
1064 if (ret) {
1065 PERROR("close");
1066 }
1067 }
1068
1069 unlink(consumer_data->err_unix_sock_path);
1070 unlink(consumer_data->cmd_unix_sock_path);
1071 consumer_data->pid = 0;
1072
1073 lttng_poll_clean(&events);
1074 error_poll:
1075 if (err) {
1076 health_error();
1077 ERR("Health error occurred in %s", __func__);
1078 }
1079 health_unregister();
1080 DBG("consumer thread cleanup completed");
1081
1082 return NULL;
1083 }
1084
1085 /*
1086 * This thread manage application communication.
1087 */
1088 static void *thread_manage_apps(void *data)
1089 {
1090 int i, ret, pollfd, err = -1;
1091 uint32_t revents, nb_fd;
1092 struct lttng_poll_event events;
1093
1094 DBG("[thread] Manage application started");
1095
1096 rcu_register_thread();
1097 rcu_thread_online();
1098
1099 health_register(HEALTH_TYPE_APP_MANAGE);
1100
1101 if (testpoint(thread_manage_apps)) {
1102 goto error_testpoint;
1103 }
1104
1105 health_code_update();
1106
1107 ret = sessiond_set_thread_pollset(&events, 2);
1108 if (ret < 0) {
1109 goto error_poll_create;
1110 }
1111
1112 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1113 if (ret < 0) {
1114 goto error;
1115 }
1116
1117 if (testpoint(thread_manage_apps_before_loop)) {
1118 goto error;
1119 }
1120
1121 health_code_update();
1122
1123 while (1) {
1124 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1125
1126 /* Inifinite blocking call, waiting for transmission */
1127 restart:
1128 health_poll_entry();
1129 ret = lttng_poll_wait(&events, -1);
1130 health_poll_exit();
1131 if (ret < 0) {
1132 /*
1133 * Restart interrupted system call.
1134 */
1135 if (errno == EINTR) {
1136 goto restart;
1137 }
1138 goto error;
1139 }
1140
1141 nb_fd = ret;
1142
1143 for (i = 0; i < nb_fd; i++) {
1144 /* Fetch once the poll data */
1145 revents = LTTNG_POLL_GETEV(&events, i);
1146 pollfd = LTTNG_POLL_GETFD(&events, i);
1147
1148 health_code_update();
1149
1150 /* Thread quit pipe has been closed. Killing thread. */
1151 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1152 if (ret) {
1153 err = 0;
1154 goto exit;
1155 }
1156
1157 /* Inspect the apps cmd pipe */
1158 if (pollfd == apps_cmd_pipe[0]) {
1159 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1160 ERR("Apps command pipe error");
1161 goto error;
1162 } else if (revents & LPOLLIN) {
1163 int sock;
1164
1165 /* Empty pipe */
1166 do {
1167 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1168 } while (ret < 0 && errno == EINTR);
1169 if (ret < 0 || ret < sizeof(sock)) {
1170 PERROR("read apps cmd pipe");
1171 goto error;
1172 }
1173
1174 health_code_update();
1175
1176 /*
1177 * We only monitor the error events of the socket. This
1178 * thread does not handle any incoming data from UST
1179 * (POLLIN).
1180 */
1181 ret = lttng_poll_add(&events, sock,
1182 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1183 if (ret < 0) {
1184 goto error;
1185 }
1186
1187 /* Set socket timeout for both receiving and ending */
1188 (void) lttcomm_setsockopt_rcv_timeout(sock,
1189 app_socket_timeout);
1190 (void) lttcomm_setsockopt_snd_timeout(sock,
1191 app_socket_timeout);
1192
1193 DBG("Apps with sock %d added to poll set", sock);
1194
1195 health_code_update();
1196
1197 break;
1198 }
1199 } else {
1200 /*
1201 * At this point, we know that a registered application made
1202 * the event at poll_wait.
1203 */
1204 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1205 /* Removing from the poll set */
1206 ret = lttng_poll_del(&events, pollfd);
1207 if (ret < 0) {
1208 goto error;
1209 }
1210
1211 /* Socket closed on remote end. */
1212 ust_app_unregister(pollfd);
1213 break;
1214 }
1215 }
1216
1217 health_code_update();
1218 }
1219 }
1220
1221 exit:
1222 error:
1223 lttng_poll_clean(&events);
1224 error_poll_create:
1225 error_testpoint:
1226 utils_close_pipe(apps_cmd_pipe);
1227 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1228
1229 /*
1230 * We don't clean the UST app hash table here since already registered
1231 * applications can still be controlled so let them be until the session
1232 * daemon dies or the applications stop.
1233 */
1234
1235 if (err) {
1236 health_error();
1237 ERR("Health error occurred in %s", __func__);
1238 }
1239 health_unregister();
1240 DBG("Application communication apps thread cleanup complete");
1241 rcu_thread_offline();
1242 rcu_unregister_thread();
1243 return NULL;
1244 }
1245
1246 /*
1247 * Send the application sockets (cmd and notify) to the respective threads.
1248 * This is called from the dispatch UST registration thread once all sockets
1249 * are set for the application.
1250 *
1251 * On success, return 0 else a negative value being the errno message of the
1252 * write().
1253 */
1254 static int send_app_sockets_to_threads(struct ust_app *app)
1255 {
1256 int ret;
1257
1258 assert(app);
1259 /* Sockets MUST be set or else this should not have been called. */
1260 assert(app->sock >= 0);
1261 assert(app->notify_sock >= 0);
1262 assert(apps_cmd_pipe[1] >= 0);
1263 assert(apps_cmd_notify_pipe[1] >= 0);
1264
1265 do {
1266 ret = write(apps_cmd_pipe[1], &app->sock, sizeof(app->sock));
1267 } while (ret < 0 && errno == EINTR);
1268 if (ret < 0 || ret != sizeof(app->sock)) {
1269 PERROR("write apps cmd pipe %d", apps_cmd_pipe[1]);
1270 if (ret < 0) {
1271 ret = -errno;
1272 }
1273 goto error;
1274 }
1275
1276 do {
1277 ret = write(apps_cmd_notify_pipe[1], &app->notify_sock,
1278 sizeof(app->notify_sock));
1279 } while (ret < 0 && errno == EINTR);
1280 if (ret < 0 || ret != sizeof(app->notify_sock)) {
1281 PERROR("write apps notify cmd pipe %d", apps_cmd_notify_pipe[1]);
1282 if (ret < 0) {
1283 ret = -errno;
1284 }
1285 goto error;
1286 }
1287
1288 /* All good. Don't send back the write positive ret value. */
1289 ret = 0;
1290 error:
1291 return ret;
1292 }
1293
1294 /*
1295 * Dispatch request from the registration threads to the application
1296 * communication thread.
1297 */
1298 static void *thread_dispatch_ust_registration(void *data)
1299 {
1300 int ret;
1301 struct cds_wfq_node *node;
1302 struct ust_command *ust_cmd = NULL;
1303 struct {
1304 struct ust_app *app;
1305 struct cds_list_head head;
1306 } *wait_node = NULL;
1307
1308 CDS_LIST_HEAD(wait_queue);
1309
1310 DBG("[thread] Dispatch UST command started");
1311
1312 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1313 /* Atomically prepare the queue futex */
1314 futex_nto1_prepare(&ust_cmd_queue.futex);
1315
1316 do {
1317 struct ust_app *app = NULL;
1318
1319 /* Dequeue command for registration */
1320 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1321 if (node == NULL) {
1322 DBG("Woken up but nothing in the UST command queue");
1323 /* Continue thread execution */
1324 break;
1325 }
1326
1327 ust_cmd = caa_container_of(node, struct ust_command, node);
1328
1329 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1330 " gid:%d sock:%d name:%s (version %d.%d)",
1331 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1332 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1333 ust_cmd->sock, ust_cmd->reg_msg.name,
1334 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1335
1336 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1337 wait_node = zmalloc(sizeof(*wait_node));
1338 if (!wait_node) {
1339 PERROR("zmalloc wait_node dispatch");
1340 goto error;
1341 }
1342 CDS_INIT_LIST_HEAD(&wait_node->head);
1343
1344 /* Create application object if socket is CMD. */
1345 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1346 ust_cmd->sock);
1347 if (!wait_node->app) {
1348 ret = close(ust_cmd->sock);
1349 if (ret < 0) {
1350 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1351 }
1352 continue;
1353 }
1354 /*
1355 * Add application to the wait queue so we can set the notify
1356 * socket before putting this object in the global ht.
1357 */
1358 cds_list_add(&wait_node->head, &wait_queue);
1359
1360 /*
1361 * We have to continue here since we don't have the notify
1362 * socket and the application MUST be added to the hash table
1363 * only at that moment.
1364 */
1365 continue;
1366 } else {
1367 /*
1368 * Look for the application in the local wait queue and set the
1369 * notify socket if found.
1370 */
1371 cds_list_for_each_entry(wait_node, &wait_queue, head) {
1372 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1373 wait_node->app->notify_sock = ust_cmd->sock;
1374 cds_list_del(&wait_node->head);
1375 app = wait_node->app;
1376 free(wait_node);
1377 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1378 break;
1379 }
1380 }
1381 }
1382
1383 if (app) {
1384 ret = send_app_sockets_to_threads(app);
1385 if (ret < 0) {
1386 goto error;
1387 }
1388 /*
1389 * @session_lock_list
1390 *
1391 * Lock the global session list so from the register up to the
1392 * registration done message, no thread can see the application
1393 * and change its state.
1394 */
1395 session_lock_list();
1396 rcu_read_lock();
1397 /*
1398 * Add application to the global hash table. This needs to be
1399 * done before the update to the UST registry can locate the
1400 * application.
1401 */
1402 ust_app_add(app);
1403 /*
1404 * Get app version.
1405 */
1406 ret = ust_app_version(app);
1407 if (ret) {
1408 ERR("Unable to get app version");
1409 }
1410 /*
1411 * Update newly registered application with the tracing
1412 * registry info already enabled information.
1413 */
1414 update_ust_app(app->sock);
1415 ret = ust_app_register_done(app->sock);
1416 if (ret < 0) {
1417 /* Remove application from the registry. */
1418 ust_app_unregister(app->sock);
1419 }
1420 rcu_read_unlock();
1421 session_unlock_list();
1422 } else {
1423 /* Application manager threads are not available. */
1424 ret = close(ust_cmd->sock);
1425 if (ret < 0) {
1426 PERROR("close ust_cmd sock");
1427 }
1428 }
1429 free(ust_cmd);
1430 } while (node != NULL);
1431
1432 /* Futex wait on queue. Blocking call on futex() */
1433 futex_nto1_wait(&ust_cmd_queue.futex);
1434 }
1435
1436 error:
1437 DBG("Dispatch thread dying");
1438 return NULL;
1439 }
1440
1441 /*
1442 * This thread manage application registration.
1443 */
1444 static void *thread_registration_apps(void *data)
1445 {
1446 int sock = -1, i, ret, pollfd, err = -1;
1447 uint32_t revents, nb_fd;
1448 struct lttng_poll_event events;
1449 /*
1450 * Get allocated in this thread, enqueued to a global queue, dequeued and
1451 * freed in the manage apps thread.
1452 */
1453 struct ust_command *ust_cmd = NULL;
1454
1455 DBG("[thread] Manage application registration started");
1456
1457 health_register(HEALTH_TYPE_APP_REG);
1458
1459 if (testpoint(thread_registration_apps)) {
1460 goto error_testpoint;
1461 }
1462
1463 ret = lttcomm_listen_unix_sock(apps_sock);
1464 if (ret < 0) {
1465 goto error_listen;
1466 }
1467
1468 /*
1469 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1470 * more will be added to this poll set.
1471 */
1472 ret = sessiond_set_thread_pollset(&events, 2);
1473 if (ret < 0) {
1474 goto error_create_poll;
1475 }
1476
1477 /* Add the application registration socket */
1478 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1479 if (ret < 0) {
1480 goto error_poll_add;
1481 }
1482
1483 /* Notify all applications to register */
1484 ret = notify_ust_apps(1);
1485 if (ret < 0) {
1486 ERR("Failed to notify applications or create the wait shared memory.\n"
1487 "Execution continues but there might be problem for already\n"
1488 "running applications that wishes to register.");
1489 }
1490
1491 while (1) {
1492 DBG("Accepting application registration");
1493
1494 /* Inifinite blocking call, waiting for transmission */
1495 restart:
1496 health_poll_entry();
1497 ret = lttng_poll_wait(&events, -1);
1498 health_poll_exit();
1499 if (ret < 0) {
1500 /*
1501 * Restart interrupted system call.
1502 */
1503 if (errno == EINTR) {
1504 goto restart;
1505 }
1506 goto error;
1507 }
1508
1509 nb_fd = ret;
1510
1511 for (i = 0; i < nb_fd; i++) {
1512 health_code_update();
1513
1514 /* Fetch once the poll data */
1515 revents = LTTNG_POLL_GETEV(&events, i);
1516 pollfd = LTTNG_POLL_GETFD(&events, i);
1517
1518 /* Thread quit pipe has been closed. Killing thread. */
1519 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1520 if (ret) {
1521 err = 0;
1522 goto exit;
1523 }
1524
1525 /* Event on the registration socket */
1526 if (pollfd == apps_sock) {
1527 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1528 ERR("Register apps socket poll error");
1529 goto error;
1530 } else if (revents & LPOLLIN) {
1531 sock = lttcomm_accept_unix_sock(apps_sock);
1532 if (sock < 0) {
1533 goto error;
1534 }
1535
1536 /*
1537 * Set the CLOEXEC flag. Return code is useless because
1538 * either way, the show must go on.
1539 */
1540 (void) utils_set_fd_cloexec(sock);
1541
1542 /* Create UST registration command for enqueuing */
1543 ust_cmd = zmalloc(sizeof(struct ust_command));
1544 if (ust_cmd == NULL) {
1545 PERROR("ust command zmalloc");
1546 goto error;
1547 }
1548
1549 /*
1550 * Using message-based transmissions to ensure we don't
1551 * have to deal with partially received messages.
1552 */
1553 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1554 if (ret < 0) {
1555 ERR("Exhausted file descriptors allowed for applications.");
1556 free(ust_cmd);
1557 ret = close(sock);
1558 if (ret) {
1559 PERROR("close");
1560 }
1561 sock = -1;
1562 continue;
1563 }
1564 health_code_update();
1565 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1566 if (ret < 0) {
1567 free(ust_cmd);
1568 /* Close socket of the application. */
1569 ret = close(sock);
1570 if (ret) {
1571 PERROR("close");
1572 }
1573 lttng_fd_put(LTTNG_FD_APPS, 1);
1574 sock = -1;
1575 continue;
1576 }
1577 health_code_update();
1578
1579 ust_cmd->sock = sock;
1580 sock = -1;
1581
1582 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1583 " gid:%d sock:%d name:%s (version %d.%d)",
1584 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1585 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1586 ust_cmd->sock, ust_cmd->reg_msg.name,
1587 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1588
1589 /*
1590 * Lock free enqueue the registration request. The red pill
1591 * has been taken! This apps will be part of the *system*.
1592 */
1593 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1594
1595 /*
1596 * Wake the registration queue futex. Implicit memory
1597 * barrier with the exchange in cds_wfq_enqueue.
1598 */
1599 futex_nto1_wake(&ust_cmd_queue.futex);
1600 }
1601 }
1602 }
1603 }
1604
1605 exit:
1606 error:
1607 if (err) {
1608 health_error();
1609 ERR("Health error occurred in %s", __func__);
1610 }
1611
1612 /* Notify that the registration thread is gone */
1613 notify_ust_apps(0);
1614
1615 if (apps_sock >= 0) {
1616 ret = close(apps_sock);
1617 if (ret) {
1618 PERROR("close");
1619 }
1620 }
1621 if (sock >= 0) {
1622 ret = close(sock);
1623 if (ret) {
1624 PERROR("close");
1625 }
1626 lttng_fd_put(LTTNG_FD_APPS, 1);
1627 }
1628 unlink(apps_unix_sock_path);
1629
1630 error_poll_add:
1631 lttng_poll_clean(&events);
1632 error_listen:
1633 error_create_poll:
1634 error_testpoint:
1635 DBG("UST Registration thread cleanup complete");
1636 health_unregister();
1637
1638 return NULL;
1639 }
1640
1641 /*
1642 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1643 * exec or it will fails.
1644 */
1645 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1646 {
1647 int ret, clock_ret;
1648 struct timespec timeout;
1649
1650 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1651 consumer_data->consumer_thread_is_ready = 0;
1652
1653 /* Setup pthread condition */
1654 ret = pthread_condattr_init(&consumer_data->condattr);
1655 if (ret != 0) {
1656 errno = ret;
1657 PERROR("pthread_condattr_init consumer data");
1658 goto error;
1659 }
1660
1661 /*
1662 * Set the monotonic clock in order to make sure we DO NOT jump in time
1663 * between the clock_gettime() call and the timedwait call. See bug #324
1664 * for a more details and how we noticed it.
1665 */
1666 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1667 if (ret != 0) {
1668 errno = ret;
1669 PERROR("pthread_condattr_setclock consumer data");
1670 goto error;
1671 }
1672
1673 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1674 if (ret != 0) {
1675 errno = ret;
1676 PERROR("pthread_cond_init consumer data");
1677 goto error;
1678 }
1679
1680 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1681 consumer_data);
1682 if (ret != 0) {
1683 PERROR("pthread_create consumer");
1684 ret = -1;
1685 goto error;
1686 }
1687
1688 /* We are about to wait on a pthread condition */
1689 pthread_mutex_lock(&consumer_data->cond_mutex);
1690
1691 /* Get time for sem_timedwait absolute timeout */
1692 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1693 /*
1694 * Set the timeout for the condition timed wait even if the clock gettime
1695 * call fails since we might loop on that call and we want to avoid to
1696 * increment the timeout too many times.
1697 */
1698 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1699
1700 /*
1701 * The following loop COULD be skipped in some conditions so this is why we
1702 * set ret to 0 in order to make sure at least one round of the loop is
1703 * done.
1704 */
1705 ret = 0;
1706
1707 /*
1708 * Loop until the condition is reached or when a timeout is reached. Note
1709 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1710 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1711 * possible. This loop does not take any chances and works with both of
1712 * them.
1713 */
1714 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1715 if (clock_ret < 0) {
1716 PERROR("clock_gettime spawn consumer");
1717 /* Infinite wait for the consumerd thread to be ready */
1718 ret = pthread_cond_wait(&consumer_data->cond,
1719 &consumer_data->cond_mutex);
1720 } else {
1721 ret = pthread_cond_timedwait(&consumer_data->cond,
1722 &consumer_data->cond_mutex, &timeout);
1723 }
1724 }
1725
1726 /* Release the pthread condition */
1727 pthread_mutex_unlock(&consumer_data->cond_mutex);
1728
1729 if (ret != 0) {
1730 errno = ret;
1731 if (ret == ETIMEDOUT) {
1732 /*
1733 * Call has timed out so we kill the kconsumerd_thread and return
1734 * an error.
1735 */
1736 ERR("Condition timed out. The consumer thread was never ready."
1737 " Killing it");
1738 ret = pthread_cancel(consumer_data->thread);
1739 if (ret < 0) {
1740 PERROR("pthread_cancel consumer thread");
1741 }
1742 } else {
1743 PERROR("pthread_cond_wait failed consumer thread");
1744 }
1745 goto error;
1746 }
1747
1748 pthread_mutex_lock(&consumer_data->pid_mutex);
1749 if (consumer_data->pid == 0) {
1750 ERR("Consumerd did not start");
1751 pthread_mutex_unlock(&consumer_data->pid_mutex);
1752 goto error;
1753 }
1754 pthread_mutex_unlock(&consumer_data->pid_mutex);
1755
1756 return 0;
1757
1758 error:
1759 return ret;
1760 }
1761
1762 /*
1763 * Join consumer thread
1764 */
1765 static int join_consumer_thread(struct consumer_data *consumer_data)
1766 {
1767 void *status;
1768
1769 /* Consumer pid must be a real one. */
1770 if (consumer_data->pid > 0) {
1771 int ret;
1772 ret = kill(consumer_data->pid, SIGTERM);
1773 if (ret) {
1774 ERR("Error killing consumer daemon");
1775 return ret;
1776 }
1777 return pthread_join(consumer_data->thread, &status);
1778 } else {
1779 return 0;
1780 }
1781 }
1782
1783 /*
1784 * Fork and exec a consumer daemon (consumerd).
1785 *
1786 * Return pid if successful else -1.
1787 */
1788 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1789 {
1790 int ret;
1791 pid_t pid;
1792 const char *consumer_to_use;
1793 const char *verbosity;
1794 struct stat st;
1795
1796 DBG("Spawning consumerd");
1797
1798 pid = fork();
1799 if (pid == 0) {
1800 /*
1801 * Exec consumerd.
1802 */
1803 if (opt_verbose_consumer) {
1804 verbosity = "--verbose";
1805 } else {
1806 verbosity = "--quiet";
1807 }
1808 switch (consumer_data->type) {
1809 case LTTNG_CONSUMER_KERNEL:
1810 /*
1811 * Find out which consumerd to execute. We will first try the
1812 * 64-bit path, then the sessiond's installation directory, and
1813 * fallback on the 32-bit one,
1814 */
1815 DBG3("Looking for a kernel consumer at these locations:");
1816 DBG3(" 1) %s", consumerd64_bin);
1817 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1818 DBG3(" 3) %s", consumerd32_bin);
1819 if (stat(consumerd64_bin, &st) == 0) {
1820 DBG3("Found location #1");
1821 consumer_to_use = consumerd64_bin;
1822 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1823 DBG3("Found location #2");
1824 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1825 } else if (stat(consumerd32_bin, &st) == 0) {
1826 DBG3("Found location #3");
1827 consumer_to_use = consumerd32_bin;
1828 } else {
1829 DBG("Could not find any valid consumerd executable");
1830 break;
1831 }
1832 DBG("Using kernel consumer at: %s", consumer_to_use);
1833 execl(consumer_to_use,
1834 "lttng-consumerd", verbosity, "-k",
1835 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1836 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1837 NULL);
1838 break;
1839 case LTTNG_CONSUMER64_UST:
1840 {
1841 char *tmpnew = NULL;
1842
1843 if (consumerd64_libdir[0] != '\0') {
1844 char *tmp;
1845 size_t tmplen;
1846
1847 tmp = getenv("LD_LIBRARY_PATH");
1848 if (!tmp) {
1849 tmp = "";
1850 }
1851 tmplen = strlen("LD_LIBRARY_PATH=")
1852 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1853 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1854 if (!tmpnew) {
1855 ret = -ENOMEM;
1856 goto error;
1857 }
1858 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1859 strcat(tmpnew, consumerd64_libdir);
1860 if (tmp[0] != '\0') {
1861 strcat(tmpnew, ":");
1862 strcat(tmpnew, tmp);
1863 }
1864 ret = putenv(tmpnew);
1865 if (ret) {
1866 ret = -errno;
1867 goto error;
1868 }
1869 }
1870 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1871 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1872 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1873 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1874 NULL);
1875 if (consumerd64_libdir[0] != '\0') {
1876 free(tmpnew);
1877 }
1878 if (ret) {
1879 goto error;
1880 }
1881 break;
1882 }
1883 case LTTNG_CONSUMER32_UST:
1884 {
1885 char *tmpnew = NULL;
1886
1887 if (consumerd32_libdir[0] != '\0') {
1888 char *tmp;
1889 size_t tmplen;
1890
1891 tmp = getenv("LD_LIBRARY_PATH");
1892 if (!tmp) {
1893 tmp = "";
1894 }
1895 tmplen = strlen("LD_LIBRARY_PATH=")
1896 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1897 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1898 if (!tmpnew) {
1899 ret = -ENOMEM;
1900 goto error;
1901 }
1902 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1903 strcat(tmpnew, consumerd32_libdir);
1904 if (tmp[0] != '\0') {
1905 strcat(tmpnew, ":");
1906 strcat(tmpnew, tmp);
1907 }
1908 ret = putenv(tmpnew);
1909 if (ret) {
1910 ret = -errno;
1911 goto error;
1912 }
1913 }
1914 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1915 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1916 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1917 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1918 NULL);
1919 if (consumerd32_libdir[0] != '\0') {
1920 free(tmpnew);
1921 }
1922 if (ret) {
1923 goto error;
1924 }
1925 break;
1926 }
1927 default:
1928 PERROR("unknown consumer type");
1929 exit(EXIT_FAILURE);
1930 }
1931 if (errno != 0) {
1932 PERROR("kernel start consumer exec");
1933 }
1934 exit(EXIT_FAILURE);
1935 } else if (pid > 0) {
1936 ret = pid;
1937 } else {
1938 PERROR("start consumer fork");
1939 ret = -errno;
1940 }
1941 error:
1942 return ret;
1943 }
1944
1945 /*
1946 * Spawn the consumerd daemon and session daemon thread.
1947 */
1948 static int start_consumerd(struct consumer_data *consumer_data)
1949 {
1950 int ret;
1951
1952 /*
1953 * Set the listen() state on the socket since there is a possible race
1954 * between the exec() of the consumer daemon and this call if place in the
1955 * consumer thread. See bug #366 for more details.
1956 */
1957 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1958 if (ret < 0) {
1959 goto error;
1960 }
1961
1962 pthread_mutex_lock(&consumer_data->pid_mutex);
1963 if (consumer_data->pid != 0) {
1964 pthread_mutex_unlock(&consumer_data->pid_mutex);
1965 goto end;
1966 }
1967
1968 ret = spawn_consumerd(consumer_data);
1969 if (ret < 0) {
1970 ERR("Spawning consumerd failed");
1971 pthread_mutex_unlock(&consumer_data->pid_mutex);
1972 goto error;
1973 }
1974
1975 /* Setting up the consumer_data pid */
1976 consumer_data->pid = ret;
1977 DBG2("Consumer pid %d", consumer_data->pid);
1978 pthread_mutex_unlock(&consumer_data->pid_mutex);
1979
1980 DBG2("Spawning consumer control thread");
1981 ret = spawn_consumer_thread(consumer_data);
1982 if (ret < 0) {
1983 ERR("Fatal error spawning consumer control thread");
1984 goto error;
1985 }
1986
1987 end:
1988 return 0;
1989
1990 error:
1991 /* Cleanup already created socket on error. */
1992 if (consumer_data->err_sock >= 0) {
1993 int err;
1994
1995 err = close(consumer_data->err_sock);
1996 if (err < 0) {
1997 PERROR("close consumer data error socket");
1998 }
1999 }
2000 return ret;
2001 }
2002
2003 /*
2004 * Compute health status of each consumer. If one of them is zero (bad
2005 * state), we return 0.
2006 */
2007 static int check_consumer_health(void)
2008 {
2009 int ret;
2010
2011 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2012
2013 DBG3("Health consumer check %d", ret);
2014
2015 return ret;
2016 }
2017
2018 /*
2019 * Setup necessary data for kernel tracer action.
2020 */
2021 static int init_kernel_tracer(void)
2022 {
2023 int ret;
2024
2025 /* Modprobe lttng kernel modules */
2026 ret = modprobe_lttng_control();
2027 if (ret < 0) {
2028 goto error;
2029 }
2030
2031 /* Open debugfs lttng */
2032 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2033 if (kernel_tracer_fd < 0) {
2034 DBG("Failed to open %s", module_proc_lttng);
2035 ret = -1;
2036 goto error_open;
2037 }
2038
2039 /* Validate kernel version */
2040 ret = kernel_validate_version(kernel_tracer_fd);
2041 if (ret < 0) {
2042 goto error_version;
2043 }
2044
2045 ret = modprobe_lttng_data();
2046 if (ret < 0) {
2047 goto error_modules;
2048 }
2049
2050 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2051 return 0;
2052
2053 error_version:
2054 modprobe_remove_lttng_control();
2055 ret = close(kernel_tracer_fd);
2056 if (ret) {
2057 PERROR("close");
2058 }
2059 kernel_tracer_fd = -1;
2060 return LTTNG_ERR_KERN_VERSION;
2061
2062 error_modules:
2063 ret = close(kernel_tracer_fd);
2064 if (ret) {
2065 PERROR("close");
2066 }
2067
2068 error_open:
2069 modprobe_remove_lttng_control();
2070
2071 error:
2072 WARN("No kernel tracer available");
2073 kernel_tracer_fd = -1;
2074 if (!is_root) {
2075 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2076 } else {
2077 return LTTNG_ERR_KERN_NA;
2078 }
2079 }
2080
2081
2082 /*
2083 * Copy consumer output from the tracing session to the domain session. The
2084 * function also applies the right modification on a per domain basis for the
2085 * trace files destination directory.
2086 */
2087 static int copy_session_consumer(int domain, struct ltt_session *session)
2088 {
2089 int ret;
2090 const char *dir_name;
2091 struct consumer_output *consumer;
2092
2093 assert(session);
2094 assert(session->consumer);
2095
2096 switch (domain) {
2097 case LTTNG_DOMAIN_KERNEL:
2098 DBG3("Copying tracing session consumer output in kernel session");
2099 /*
2100 * XXX: We should audit the session creation and what this function
2101 * does "extra" in order to avoid a destroy since this function is used
2102 * in the domain session creation (kernel and ust) only. Same for UST
2103 * domain.
2104 */
2105 if (session->kernel_session->consumer) {
2106 consumer_destroy_output(session->kernel_session->consumer);
2107 }
2108 session->kernel_session->consumer =
2109 consumer_copy_output(session->consumer);
2110 /* Ease our life a bit for the next part */
2111 consumer = session->kernel_session->consumer;
2112 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2113 break;
2114 case LTTNG_DOMAIN_UST:
2115 DBG3("Copying tracing session consumer output in UST session");
2116 if (session->ust_session->consumer) {
2117 consumer_destroy_output(session->ust_session->consumer);
2118 }
2119 session->ust_session->consumer =
2120 consumer_copy_output(session->consumer);
2121 /* Ease our life a bit for the next part */
2122 consumer = session->ust_session->consumer;
2123 dir_name = DEFAULT_UST_TRACE_DIR;
2124 break;
2125 default:
2126 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2127 goto error;
2128 }
2129
2130 /* Append correct directory to subdir */
2131 strncat(consumer->subdir, dir_name,
2132 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2133 DBG3("Copy session consumer subdir %s", consumer->subdir);
2134
2135 ret = LTTNG_OK;
2136
2137 error:
2138 return ret;
2139 }
2140
2141 /*
2142 * Create an UST session and add it to the session ust list.
2143 */
2144 static int create_ust_session(struct ltt_session *session,
2145 struct lttng_domain *domain)
2146 {
2147 int ret;
2148 struct ltt_ust_session *lus = NULL;
2149
2150 assert(session);
2151 assert(domain);
2152 assert(session->consumer);
2153
2154 switch (domain->type) {
2155 case LTTNG_DOMAIN_UST:
2156 break;
2157 default:
2158 ERR("Unknown UST domain on create session %d", domain->type);
2159 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2160 goto error;
2161 }
2162
2163 DBG("Creating UST session");
2164
2165 lus = trace_ust_create_session(session->path, session->id);
2166 if (lus == NULL) {
2167 ret = LTTNG_ERR_UST_SESS_FAIL;
2168 goto error;
2169 }
2170
2171 lus->uid = session->uid;
2172 lus->gid = session->gid;
2173 session->ust_session = lus;
2174
2175 /* Copy session output to the newly created UST session */
2176 ret = copy_session_consumer(domain->type, session);
2177 if (ret != LTTNG_OK) {
2178 goto error;
2179 }
2180
2181 return LTTNG_OK;
2182
2183 error:
2184 free(lus);
2185 session->ust_session = NULL;
2186 return ret;
2187 }
2188
2189 /*
2190 * Create a kernel tracer session then create the default channel.
2191 */
2192 static int create_kernel_session(struct ltt_session *session)
2193 {
2194 int ret;
2195
2196 DBG("Creating kernel session");
2197
2198 ret = kernel_create_session(session, kernel_tracer_fd);
2199 if (ret < 0) {
2200 ret = LTTNG_ERR_KERN_SESS_FAIL;
2201 goto error;
2202 }
2203
2204 /* Code flow safety */
2205 assert(session->kernel_session);
2206
2207 /* Copy session output to the newly created Kernel session */
2208 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2209 if (ret != LTTNG_OK) {
2210 goto error;
2211 }
2212
2213 /* Create directory(ies) on local filesystem. */
2214 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2215 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2216 ret = run_as_mkdir_recursive(
2217 session->kernel_session->consumer->dst.trace_path,
2218 S_IRWXU | S_IRWXG, session->uid, session->gid);
2219 if (ret < 0) {
2220 if (ret != -EEXIST) {
2221 ERR("Trace directory creation error");
2222 goto error;
2223 }
2224 }
2225 }
2226
2227 session->kernel_session->uid = session->uid;
2228 session->kernel_session->gid = session->gid;
2229
2230 return LTTNG_OK;
2231
2232 error:
2233 trace_kernel_destroy_session(session->kernel_session);
2234 session->kernel_session = NULL;
2235 return ret;
2236 }
2237
2238 /*
2239 * Count number of session permitted by uid/gid.
2240 */
2241 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2242 {
2243 unsigned int i = 0;
2244 struct ltt_session *session;
2245
2246 DBG("Counting number of available session for UID %d GID %d",
2247 uid, gid);
2248 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2249 /*
2250 * Only list the sessions the user can control.
2251 */
2252 if (!session_access_ok(session, uid, gid)) {
2253 continue;
2254 }
2255 i++;
2256 }
2257 return i;
2258 }
2259
2260 /*
2261 * Process the command requested by the lttng client within the command
2262 * context structure. This function make sure that the return structure (llm)
2263 * is set and ready for transmission before returning.
2264 *
2265 * Return any error encountered or 0 for success.
2266 *
2267 * "sock" is only used for special-case var. len data.
2268 */
2269 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2270 int *sock_error)
2271 {
2272 int ret = LTTNG_OK;
2273 int need_tracing_session = 1;
2274 int need_domain;
2275
2276 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2277
2278 *sock_error = 0;
2279
2280 switch (cmd_ctx->lsm->cmd_type) {
2281 case LTTNG_CREATE_SESSION:
2282 case LTTNG_DESTROY_SESSION:
2283 case LTTNG_LIST_SESSIONS:
2284 case LTTNG_LIST_DOMAINS:
2285 case LTTNG_START_TRACE:
2286 case LTTNG_STOP_TRACE:
2287 case LTTNG_DATA_PENDING:
2288 need_domain = 0;
2289 break;
2290 default:
2291 need_domain = 1;
2292 }
2293
2294 if (opt_no_kernel && need_domain
2295 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2296 if (!is_root) {
2297 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2298 } else {
2299 ret = LTTNG_ERR_KERN_NA;
2300 }
2301 goto error;
2302 }
2303
2304 /* Deny register consumer if we already have a spawned consumer. */
2305 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2306 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2307 if (kconsumer_data.pid > 0) {
2308 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2309 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2310 goto error;
2311 }
2312 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2313 }
2314
2315 /*
2316 * Check for command that don't needs to allocate a returned payload. We do
2317 * this here so we don't have to make the call for no payload at each
2318 * command.
2319 */
2320 switch(cmd_ctx->lsm->cmd_type) {
2321 case LTTNG_LIST_SESSIONS:
2322 case LTTNG_LIST_TRACEPOINTS:
2323 case LTTNG_LIST_TRACEPOINT_FIELDS:
2324 case LTTNG_LIST_DOMAINS:
2325 case LTTNG_LIST_CHANNELS:
2326 case LTTNG_LIST_EVENTS:
2327 break;
2328 default:
2329 /* Setup lttng message with no payload */
2330 ret = setup_lttng_msg(cmd_ctx, 0);
2331 if (ret < 0) {
2332 /* This label does not try to unlock the session */
2333 goto init_setup_error;
2334 }
2335 }
2336
2337 /* Commands that DO NOT need a session. */
2338 switch (cmd_ctx->lsm->cmd_type) {
2339 case LTTNG_CREATE_SESSION:
2340 case LTTNG_CALIBRATE:
2341 case LTTNG_LIST_SESSIONS:
2342 case LTTNG_LIST_TRACEPOINTS:
2343 case LTTNG_LIST_TRACEPOINT_FIELDS:
2344 need_tracing_session = 0;
2345 break;
2346 default:
2347 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2348 /*
2349 * We keep the session list lock across _all_ commands
2350 * for now, because the per-session lock does not
2351 * handle teardown properly.
2352 */
2353 session_lock_list();
2354 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2355 if (cmd_ctx->session == NULL) {
2356 if (cmd_ctx->lsm->session.name != NULL) {
2357 ret = LTTNG_ERR_SESS_NOT_FOUND;
2358 } else {
2359 /* If no session name specified */
2360 ret = LTTNG_ERR_SELECT_SESS;
2361 }
2362 goto error;
2363 } else {
2364 /* Acquire lock for the session */
2365 session_lock(cmd_ctx->session);
2366 }
2367 break;
2368 }
2369
2370 if (!need_domain) {
2371 goto skip_domain;
2372 }
2373
2374 /*
2375 * Check domain type for specific "pre-action".
2376 */
2377 switch (cmd_ctx->lsm->domain.type) {
2378 case LTTNG_DOMAIN_KERNEL:
2379 if (!is_root) {
2380 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2381 goto error;
2382 }
2383
2384 /* Kernel tracer check */
2385 if (kernel_tracer_fd == -1) {
2386 /* Basically, load kernel tracer modules */
2387 ret = init_kernel_tracer();
2388 if (ret != 0) {
2389 goto error;
2390 }
2391 }
2392
2393 /* Consumer is in an ERROR state. Report back to client */
2394 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2395 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2396 goto error;
2397 }
2398
2399 /* Need a session for kernel command */
2400 if (need_tracing_session) {
2401 if (cmd_ctx->session->kernel_session == NULL) {
2402 ret = create_kernel_session(cmd_ctx->session);
2403 if (ret < 0) {
2404 ret = LTTNG_ERR_KERN_SESS_FAIL;
2405 goto error;
2406 }
2407 }
2408
2409 /* Start the kernel consumer daemon */
2410 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2411 if (kconsumer_data.pid == 0 &&
2412 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2413 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2414 ret = start_consumerd(&kconsumer_data);
2415 if (ret < 0) {
2416 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2417 goto error;
2418 }
2419 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2420 } else {
2421 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2422 }
2423
2424 /*
2425 * The consumer was just spawned so we need to add the socket to
2426 * the consumer output of the session if exist.
2427 */
2428 ret = consumer_create_socket(&kconsumer_data,
2429 cmd_ctx->session->kernel_session->consumer);
2430 if (ret < 0) {
2431 goto error;
2432 }
2433 }
2434
2435 break;
2436 case LTTNG_DOMAIN_UST:
2437 {
2438 /* Consumer is in an ERROR state. Report back to client */
2439 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2440 ret = LTTNG_ERR_NO_USTCONSUMERD;
2441 goto error;
2442 }
2443
2444 if (need_tracing_session) {
2445 /* Create UST session if none exist. */
2446 if (cmd_ctx->session->ust_session == NULL) {
2447 ret = create_ust_session(cmd_ctx->session,
2448 &cmd_ctx->lsm->domain);
2449 if (ret != LTTNG_OK) {
2450 goto error;
2451 }
2452 }
2453
2454 /* Start the UST consumer daemons */
2455 /* 64-bit */
2456 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2457 if (consumerd64_bin[0] != '\0' &&
2458 ustconsumer64_data.pid == 0 &&
2459 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2460 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2461 ret = start_consumerd(&ustconsumer64_data);
2462 if (ret < 0) {
2463 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2464 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2465 goto error;
2466 }
2467
2468 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2469 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2470 } else {
2471 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2472 }
2473
2474 /*
2475 * Setup socket for consumer 64 bit. No need for atomic access
2476 * since it was set above and can ONLY be set in this thread.
2477 */
2478 ret = consumer_create_socket(&ustconsumer64_data,
2479 cmd_ctx->session->ust_session->consumer);
2480 if (ret < 0) {
2481 goto error;
2482 }
2483
2484 /* 32-bit */
2485 if (consumerd32_bin[0] != '\0' &&
2486 ustconsumer32_data.pid == 0 &&
2487 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2488 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2489 ret = start_consumerd(&ustconsumer32_data);
2490 if (ret < 0) {
2491 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2492 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2493 goto error;
2494 }
2495
2496 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2497 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2498 } else {
2499 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2500 }
2501
2502 /*
2503 * Setup socket for consumer 64 bit. No need for atomic access
2504 * since it was set above and can ONLY be set in this thread.
2505 */
2506 ret = consumer_create_socket(&ustconsumer32_data,
2507 cmd_ctx->session->ust_session->consumer);
2508 if (ret < 0) {
2509 goto error;
2510 }
2511 }
2512 break;
2513 }
2514 default:
2515 break;
2516 }
2517 skip_domain:
2518
2519 /* Validate consumer daemon state when start/stop trace command */
2520 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2521 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2522 switch (cmd_ctx->lsm->domain.type) {
2523 case LTTNG_DOMAIN_UST:
2524 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2525 ret = LTTNG_ERR_NO_USTCONSUMERD;
2526 goto error;
2527 }
2528 break;
2529 case LTTNG_DOMAIN_KERNEL:
2530 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2531 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2532 goto error;
2533 }
2534 break;
2535 }
2536 }
2537
2538 /*
2539 * Check that the UID or GID match that of the tracing session.
2540 * The root user can interact with all sessions.
2541 */
2542 if (need_tracing_session) {
2543 if (!session_access_ok(cmd_ctx->session,
2544 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2545 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2546 ret = LTTNG_ERR_EPERM;
2547 goto error;
2548 }
2549 }
2550
2551 /*
2552 * Send relayd information to consumer as soon as we have a domain and a
2553 * session defined.
2554 */
2555 if (cmd_ctx->session && need_domain) {
2556 /*
2557 * Setup relayd if not done yet. If the relayd information was already
2558 * sent to the consumer, this call will gracefully return.
2559 */
2560 ret = cmd_setup_relayd(cmd_ctx->session);
2561 if (ret != LTTNG_OK) {
2562 goto error;
2563 }
2564 }
2565
2566 /* Process by command type */
2567 switch (cmd_ctx->lsm->cmd_type) {
2568 case LTTNG_ADD_CONTEXT:
2569 {
2570 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2571 cmd_ctx->lsm->u.context.channel_name,
2572 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2573 break;
2574 }
2575 case LTTNG_DISABLE_CHANNEL:
2576 {
2577 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2578 cmd_ctx->lsm->u.disable.channel_name);
2579 break;
2580 }
2581 case LTTNG_DISABLE_EVENT:
2582 {
2583 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2584 cmd_ctx->lsm->u.disable.channel_name,
2585 cmd_ctx->lsm->u.disable.name);
2586 break;
2587 }
2588 case LTTNG_DISABLE_ALL_EVENT:
2589 {
2590 DBG("Disabling all events");
2591
2592 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2593 cmd_ctx->lsm->u.disable.channel_name);
2594 break;
2595 }
2596 case LTTNG_ENABLE_CHANNEL:
2597 {
2598 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2599 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2600 break;
2601 }
2602 case LTTNG_ENABLE_EVENT:
2603 {
2604 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2605 cmd_ctx->lsm->u.enable.channel_name,
2606 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2607 break;
2608 }
2609 case LTTNG_ENABLE_ALL_EVENT:
2610 {
2611 DBG("Enabling all events");
2612
2613 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2614 cmd_ctx->lsm->u.enable.channel_name,
2615 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2616 break;
2617 }
2618 case LTTNG_LIST_TRACEPOINTS:
2619 {
2620 struct lttng_event *events;
2621 ssize_t nb_events;
2622
2623 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2624 if (nb_events < 0) {
2625 /* Return value is a negative lttng_error_code. */
2626 ret = -nb_events;
2627 goto error;
2628 }
2629
2630 /*
2631 * Setup lttng message with payload size set to the event list size in
2632 * bytes and then copy list into the llm payload.
2633 */
2634 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2635 if (ret < 0) {
2636 free(events);
2637 goto setup_error;
2638 }
2639
2640 /* Copy event list into message payload */
2641 memcpy(cmd_ctx->llm->payload, events,
2642 sizeof(struct lttng_event) * nb_events);
2643
2644 free(events);
2645
2646 ret = LTTNG_OK;
2647 break;
2648 }
2649 case LTTNG_LIST_TRACEPOINT_FIELDS:
2650 {
2651 struct lttng_event_field *fields;
2652 ssize_t nb_fields;
2653
2654 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2655 &fields);
2656 if (nb_fields < 0) {
2657 /* Return value is a negative lttng_error_code. */
2658 ret = -nb_fields;
2659 goto error;
2660 }
2661
2662 /*
2663 * Setup lttng message with payload size set to the event list size in
2664 * bytes and then copy list into the llm payload.
2665 */
2666 ret = setup_lttng_msg(cmd_ctx,
2667 sizeof(struct lttng_event_field) * nb_fields);
2668 if (ret < 0) {
2669 free(fields);
2670 goto setup_error;
2671 }
2672
2673 /* Copy event list into message payload */
2674 memcpy(cmd_ctx->llm->payload, fields,
2675 sizeof(struct lttng_event_field) * nb_fields);
2676
2677 free(fields);
2678
2679 ret = LTTNG_OK;
2680 break;
2681 }
2682 case LTTNG_SET_CONSUMER_URI:
2683 {
2684 size_t nb_uri, len;
2685 struct lttng_uri *uris;
2686
2687 nb_uri = cmd_ctx->lsm->u.uri.size;
2688 len = nb_uri * sizeof(struct lttng_uri);
2689
2690 if (nb_uri == 0) {
2691 ret = LTTNG_ERR_INVALID;
2692 goto error;
2693 }
2694
2695 uris = zmalloc(len);
2696 if (uris == NULL) {
2697 ret = LTTNG_ERR_FATAL;
2698 goto error;
2699 }
2700
2701 /* Receive variable len data */
2702 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2703 ret = lttcomm_recv_unix_sock(sock, uris, len);
2704 if (ret <= 0) {
2705 DBG("No URIs received from client... continuing");
2706 *sock_error = 1;
2707 ret = LTTNG_ERR_SESSION_FAIL;
2708 free(uris);
2709 goto error;
2710 }
2711
2712 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2713 nb_uri, uris);
2714 if (ret != LTTNG_OK) {
2715 free(uris);
2716 goto error;
2717 }
2718
2719 /*
2720 * XXX: 0 means that this URI should be applied on the session. Should
2721 * be a DOMAIN enuam.
2722 */
2723 if (cmd_ctx->lsm->domain.type == 0) {
2724 /* Add the URI for the UST session if a consumer is present. */
2725 if (cmd_ctx->session->ust_session &&
2726 cmd_ctx->session->ust_session->consumer) {
2727 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2728 nb_uri, uris);
2729 } else if (cmd_ctx->session->kernel_session &&
2730 cmd_ctx->session->kernel_session->consumer) {
2731 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2732 cmd_ctx->session, nb_uri, uris);
2733 }
2734 }
2735
2736 free(uris);
2737
2738 break;
2739 }
2740 case LTTNG_START_TRACE:
2741 {
2742 ret = cmd_start_trace(cmd_ctx->session);
2743 break;
2744 }
2745 case LTTNG_STOP_TRACE:
2746 {
2747 ret = cmd_stop_trace(cmd_ctx->session);
2748 break;
2749 }
2750 case LTTNG_CREATE_SESSION:
2751 {
2752 size_t nb_uri, len;
2753 struct lttng_uri *uris = NULL;
2754
2755 nb_uri = cmd_ctx->lsm->u.uri.size;
2756 len = nb_uri * sizeof(struct lttng_uri);
2757
2758 if (nb_uri > 0) {
2759 uris = zmalloc(len);
2760 if (uris == NULL) {
2761 ret = LTTNG_ERR_FATAL;
2762 goto error;
2763 }
2764
2765 /* Receive variable len data */
2766 DBG("Waiting for %zu URIs from client ...", nb_uri);
2767 ret = lttcomm_recv_unix_sock(sock, uris, len);
2768 if (ret <= 0) {
2769 DBG("No URIs received from client... continuing");
2770 *sock_error = 1;
2771 ret = LTTNG_ERR_SESSION_FAIL;
2772 free(uris);
2773 goto error;
2774 }
2775
2776 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2777 DBG("Creating session with ONE network URI is a bad call");
2778 ret = LTTNG_ERR_SESSION_FAIL;
2779 free(uris);
2780 goto error;
2781 }
2782 }
2783
2784 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2785 &cmd_ctx->creds);
2786
2787 free(uris);
2788
2789 break;
2790 }
2791 case LTTNG_DESTROY_SESSION:
2792 {
2793 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2794
2795 /* Set session to NULL so we do not unlock it after free. */
2796 cmd_ctx->session = NULL;
2797 break;
2798 }
2799 case LTTNG_LIST_DOMAINS:
2800 {
2801 ssize_t nb_dom;
2802 struct lttng_domain *domains;
2803
2804 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2805 if (nb_dom < 0) {
2806 /* Return value is a negative lttng_error_code. */
2807 ret = -nb_dom;
2808 goto error;
2809 }
2810
2811 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2812 if (ret < 0) {
2813 goto setup_error;
2814 }
2815
2816 /* Copy event list into message payload */
2817 memcpy(cmd_ctx->llm->payload, domains,
2818 nb_dom * sizeof(struct lttng_domain));
2819
2820 free(domains);
2821
2822 ret = LTTNG_OK;
2823 break;
2824 }
2825 case LTTNG_LIST_CHANNELS:
2826 {
2827 int nb_chan;
2828 struct lttng_channel *channels;
2829
2830 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2831 cmd_ctx->session, &channels);
2832 if (nb_chan < 0) {
2833 /* Return value is a negative lttng_error_code. */
2834 ret = -nb_chan;
2835 goto error;
2836 }
2837
2838 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2839 if (ret < 0) {
2840 goto setup_error;
2841 }
2842
2843 /* Copy event list into message payload */
2844 memcpy(cmd_ctx->llm->payload, channels,
2845 nb_chan * sizeof(struct lttng_channel));
2846
2847 free(channels);
2848
2849 ret = LTTNG_OK;
2850 break;
2851 }
2852 case LTTNG_LIST_EVENTS:
2853 {
2854 ssize_t nb_event;
2855 struct lttng_event *events = NULL;
2856
2857 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2858 cmd_ctx->lsm->u.list.channel_name, &events);
2859 if (nb_event < 0) {
2860 /* Return value is a negative lttng_error_code. */
2861 ret = -nb_event;
2862 goto error;
2863 }
2864
2865 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2866 if (ret < 0) {
2867 goto setup_error;
2868 }
2869
2870 /* Copy event list into message payload */
2871 memcpy(cmd_ctx->llm->payload, events,
2872 nb_event * sizeof(struct lttng_event));
2873
2874 free(events);
2875
2876 ret = LTTNG_OK;
2877 break;
2878 }
2879 case LTTNG_LIST_SESSIONS:
2880 {
2881 unsigned int nr_sessions;
2882
2883 session_lock_list();
2884 nr_sessions = lttng_sessions_count(
2885 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2886 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2887
2888 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2889 if (ret < 0) {
2890 session_unlock_list();
2891 goto setup_error;
2892 }
2893
2894 /* Filled the session array */
2895 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2896 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2897 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2898
2899 session_unlock_list();
2900
2901 ret = LTTNG_OK;
2902 break;
2903 }
2904 case LTTNG_CALIBRATE:
2905 {
2906 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2907 &cmd_ctx->lsm->u.calibrate);
2908 break;
2909 }
2910 case LTTNG_REGISTER_CONSUMER:
2911 {
2912 struct consumer_data *cdata;
2913
2914 switch (cmd_ctx->lsm->domain.type) {
2915 case LTTNG_DOMAIN_KERNEL:
2916 cdata = &kconsumer_data;
2917 break;
2918 default:
2919 ret = LTTNG_ERR_UND;
2920 goto error;
2921 }
2922
2923 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2924 cmd_ctx->lsm->u.reg.path, cdata);
2925 break;
2926 }
2927 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2928 {
2929 struct lttng_filter_bytecode *bytecode;
2930
2931 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2932 ret = LTTNG_ERR_FILTER_INVAL;
2933 goto error;
2934 }
2935 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2936 ret = LTTNG_ERR_FILTER_INVAL;
2937 goto error;
2938 }
2939 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2940 if (!bytecode) {
2941 ret = LTTNG_ERR_FILTER_NOMEM;
2942 goto error;
2943 }
2944 /* Receive var. len. data */
2945 DBG("Receiving var len data from client ...");
2946 ret = lttcomm_recv_unix_sock(sock, bytecode,
2947 cmd_ctx->lsm->u.enable.bytecode_len);
2948 if (ret <= 0) {
2949 DBG("Nothing recv() from client var len data... continuing");
2950 *sock_error = 1;
2951 ret = LTTNG_ERR_FILTER_INVAL;
2952 goto error;
2953 }
2954
2955 if (bytecode->len + sizeof(*bytecode)
2956 != cmd_ctx->lsm->u.enable.bytecode_len) {
2957 free(bytecode);
2958 ret = LTTNG_ERR_FILTER_INVAL;
2959 goto error;
2960 }
2961
2962 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2963 cmd_ctx->lsm->u.enable.channel_name,
2964 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2965 break;
2966 }
2967 case LTTNG_DATA_PENDING:
2968 {
2969 ret = cmd_data_pending(cmd_ctx->session);
2970 break;
2971 }
2972 default:
2973 ret = LTTNG_ERR_UND;
2974 break;
2975 }
2976
2977 error:
2978 if (cmd_ctx->llm == NULL) {
2979 DBG("Missing llm structure. Allocating one.");
2980 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2981 goto setup_error;
2982 }
2983 }
2984 /* Set return code */
2985 cmd_ctx->llm->ret_code = ret;
2986 setup_error:
2987 if (cmd_ctx->session) {
2988 session_unlock(cmd_ctx->session);
2989 }
2990 if (need_tracing_session) {
2991 session_unlock_list();
2992 }
2993 init_setup_error:
2994 return ret;
2995 }
2996
2997 /*
2998 * Thread managing health check socket.
2999 */
3000 static void *thread_manage_health(void *data)
3001 {
3002 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3003 uint32_t revents, nb_fd;
3004 struct lttng_poll_event events;
3005 struct lttcomm_health_msg msg;
3006 struct lttcomm_health_data reply;
3007
3008 DBG("[thread] Manage health check started");
3009
3010 rcu_register_thread();
3011
3012 /* Create unix socket */
3013 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3014 if (sock < 0) {
3015 ERR("Unable to create health check Unix socket");
3016 ret = -1;
3017 goto error;
3018 }
3019
3020 /*
3021 * Set the CLOEXEC flag. Return code is useless because either way, the
3022 * show must go on.
3023 */
3024 (void) utils_set_fd_cloexec(sock);
3025
3026 ret = lttcomm_listen_unix_sock(sock);
3027 if (ret < 0) {
3028 goto error;
3029 }
3030
3031 /*
3032 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3033 * more will be added to this poll set.
3034 */
3035 ret = sessiond_set_thread_pollset(&events, 2);
3036 if (ret < 0) {
3037 goto error;
3038 }
3039
3040 /* Add the application registration socket */
3041 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3042 if (ret < 0) {
3043 goto error;
3044 }
3045
3046 while (1) {
3047 DBG("Health check ready");
3048
3049 /* Inifinite blocking call, waiting for transmission */
3050 restart:
3051 ret = lttng_poll_wait(&events, -1);
3052 if (ret < 0) {
3053 /*
3054 * Restart interrupted system call.
3055 */
3056 if (errno == EINTR) {
3057 goto restart;
3058 }
3059 goto error;
3060 }
3061
3062 nb_fd = ret;
3063
3064 for (i = 0; i < nb_fd; i++) {
3065 /* Fetch once the poll data */
3066 revents = LTTNG_POLL_GETEV(&events, i);
3067 pollfd = LTTNG_POLL_GETFD(&events, i);
3068
3069 /* Thread quit pipe has been closed. Killing thread. */
3070 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3071 if (ret) {
3072 err = 0;
3073 goto exit;
3074 }
3075
3076 /* Event on the registration socket */
3077 if (pollfd == sock) {
3078 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3079 ERR("Health socket poll error");
3080 goto error;
3081 }
3082 }
3083 }
3084
3085 new_sock = lttcomm_accept_unix_sock(sock);
3086 if (new_sock < 0) {
3087 goto error;
3088 }
3089
3090 /*
3091 * Set the CLOEXEC flag. Return code is useless because either way, the
3092 * show must go on.
3093 */
3094 (void) utils_set_fd_cloexec(new_sock);
3095
3096 DBG("Receiving data from client for health...");
3097 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3098 if (ret <= 0) {
3099 DBG("Nothing recv() from client... continuing");
3100 ret = close(new_sock);
3101 if (ret) {
3102 PERROR("close");
3103 }
3104 new_sock = -1;
3105 continue;
3106 }
3107
3108 rcu_thread_online();
3109
3110 switch (msg.component) {
3111 case LTTNG_HEALTH_CMD:
3112 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3113 break;
3114 case LTTNG_HEALTH_APP_MANAGE:
3115 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3116 break;
3117 case LTTNG_HEALTH_APP_REG:
3118 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3119 break;
3120 case LTTNG_HEALTH_KERNEL:
3121 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3122 break;
3123 case LTTNG_HEALTH_CONSUMER:
3124 reply.ret_code = check_consumer_health();
3125 break;
3126 case LTTNG_HEALTH_ALL:
3127 reply.ret_code =
3128 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3129 health_check_state(HEALTH_TYPE_APP_REG) &&
3130 health_check_state(HEALTH_TYPE_CMD) &&
3131 health_check_state(HEALTH_TYPE_KERNEL) &&
3132 check_consumer_health();
3133 break;
3134 default:
3135 reply.ret_code = LTTNG_ERR_UND;
3136 break;
3137 }
3138
3139 /*
3140 * Flip ret value since 0 is a success and 1 indicates a bad health for
3141 * the client where in the sessiond it is the opposite. Again, this is
3142 * just to make things easier for us poor developer which enjoy a lot
3143 * lazyness.
3144 */
3145 if (reply.ret_code == 0 || reply.ret_code == 1) {
3146 reply.ret_code = !reply.ret_code;
3147 }
3148
3149 DBG2("Health check return value %d", reply.ret_code);
3150
3151 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3152 if (ret < 0) {
3153 ERR("Failed to send health data back to client");
3154 }
3155
3156 /* End of transmission */
3157 ret = close(new_sock);
3158 if (ret) {
3159 PERROR("close");
3160 }
3161 new_sock = -1;
3162 }
3163
3164 exit:
3165 error:
3166 if (err) {
3167 ERR("Health error occurred in %s", __func__);
3168 }
3169 DBG("Health check thread dying");
3170 unlink(health_unix_sock_path);
3171 if (sock >= 0) {
3172 ret = close(sock);
3173 if (ret) {
3174 PERROR("close");
3175 }
3176 }
3177 if (new_sock >= 0) {
3178 ret = close(new_sock);
3179 if (ret) {
3180 PERROR("close");
3181 }
3182 }
3183
3184 lttng_poll_clean(&events);
3185
3186 rcu_unregister_thread();
3187 return NULL;
3188 }
3189
3190 /*
3191 * This thread manage all clients request using the unix client socket for
3192 * communication.
3193 */
3194 static void *thread_manage_clients(void *data)
3195 {
3196 int sock = -1, ret, i, pollfd, err = -1;
3197 int sock_error;
3198 uint32_t revents, nb_fd;
3199 struct command_ctx *cmd_ctx = NULL;
3200 struct lttng_poll_event events;
3201
3202 DBG("[thread] Manage client started");
3203
3204 rcu_register_thread();
3205
3206 health_register(HEALTH_TYPE_CMD);
3207
3208 if (testpoint(thread_manage_clients)) {
3209 goto error_testpoint;
3210 }
3211
3212 health_code_update();
3213
3214 ret = lttcomm_listen_unix_sock(client_sock);
3215 if (ret < 0) {
3216 goto error_listen;
3217 }
3218
3219 /*
3220 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3221 * more will be added to this poll set.
3222 */
3223 ret = sessiond_set_thread_pollset(&events, 2);
3224 if (ret < 0) {
3225 goto error_create_poll;
3226 }
3227
3228 /* Add the application registration socket */
3229 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3230 if (ret < 0) {
3231 goto error;
3232 }
3233
3234 /*
3235 * Notify parent pid that we are ready to accept command for client side.
3236 */
3237 if (opt_sig_parent) {
3238 kill(ppid, SIGUSR1);
3239 }
3240
3241 if (testpoint(thread_manage_clients_before_loop)) {
3242 goto error;
3243 }
3244
3245 health_code_update();
3246
3247 while (1) {
3248 DBG("Accepting client command ...");
3249
3250 /* Inifinite blocking call, waiting for transmission */
3251 restart:
3252 health_poll_entry();
3253 ret = lttng_poll_wait(&events, -1);
3254 health_poll_exit();
3255 if (ret < 0) {
3256 /*
3257 * Restart interrupted system call.
3258 */
3259 if (errno == EINTR) {
3260 goto restart;
3261 }
3262 goto error;
3263 }
3264
3265 nb_fd = ret;
3266
3267 for (i = 0; i < nb_fd; i++) {
3268 /* Fetch once the poll data */
3269 revents = LTTNG_POLL_GETEV(&events, i);
3270 pollfd = LTTNG_POLL_GETFD(&events, i);
3271
3272 health_code_update();
3273
3274 /* Thread quit pipe has been closed. Killing thread. */
3275 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3276 if (ret) {
3277 err = 0;
3278 goto exit;
3279 }
3280
3281 /* Event on the registration socket */
3282 if (pollfd == client_sock) {
3283 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3284 ERR("Client socket poll error");
3285 goto error;
3286 }
3287 }
3288 }
3289
3290 DBG("Wait for client response");
3291
3292 health_code_update();
3293
3294 sock = lttcomm_accept_unix_sock(client_sock);
3295 if (sock < 0) {
3296 goto error;
3297 }
3298
3299 /*
3300 * Set the CLOEXEC flag. Return code is useless because either way, the
3301 * show must go on.
3302 */
3303 (void) utils_set_fd_cloexec(sock);
3304
3305 /* Set socket option for credentials retrieval */
3306 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3307 if (ret < 0) {
3308 goto error;
3309 }
3310
3311 /* Allocate context command to process the client request */
3312 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3313 if (cmd_ctx == NULL) {
3314 PERROR("zmalloc cmd_ctx");
3315 goto error;
3316 }
3317
3318 /* Allocate data buffer for reception */
3319 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3320 if (cmd_ctx->lsm == NULL) {
3321 PERROR("zmalloc cmd_ctx->lsm");
3322 goto error;
3323 }
3324
3325 cmd_ctx->llm = NULL;
3326 cmd_ctx->session = NULL;
3327
3328 health_code_update();
3329
3330 /*
3331 * Data is received from the lttng client. The struct
3332 * lttcomm_session_msg (lsm) contains the command and data request of
3333 * the client.
3334 */
3335 DBG("Receiving data from client ...");
3336 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3337 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3338 if (ret <= 0) {
3339 DBG("Nothing recv() from client... continuing");
3340 ret = close(sock);
3341 if (ret) {
3342 PERROR("close");
3343 }
3344 sock = -1;
3345 clean_command_ctx(&cmd_ctx);
3346 continue;
3347 }
3348
3349 health_code_update();
3350
3351 // TODO: Validate cmd_ctx including sanity check for
3352 // security purpose.
3353
3354 rcu_thread_online();
3355 /*
3356 * This function dispatch the work to the kernel or userspace tracer
3357 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3358 * informations for the client. The command context struct contains
3359 * everything this function may needs.
3360 */
3361 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3362 rcu_thread_offline();
3363 if (ret < 0) {
3364 if (sock_error) {
3365 ret = close(sock);
3366 if (ret) {
3367 PERROR("close");
3368 }
3369 sock = -1;
3370 }
3371 /*
3372 * TODO: Inform client somehow of the fatal error. At
3373 * this point, ret < 0 means that a zmalloc failed
3374 * (ENOMEM). Error detected but still accept
3375 * command, unless a socket error has been
3376 * detected.
3377 */
3378 clean_command_ctx(&cmd_ctx);
3379 continue;
3380 }
3381
3382 health_code_update();
3383
3384 DBG("Sending response (size: %d, retcode: %s)",
3385 cmd_ctx->lttng_msg_size,
3386 lttng_strerror(-cmd_ctx->llm->ret_code));
3387 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3388 if (ret < 0) {
3389 ERR("Failed to send data back to client");
3390 }
3391
3392 /* End of transmission */
3393 ret = close(sock);
3394 if (ret) {
3395 PERROR("close");
3396 }
3397 sock = -1;
3398
3399 clean_command_ctx(&cmd_ctx);
3400
3401 health_code_update();
3402 }
3403
3404 exit:
3405 error:
3406 if (sock >= 0) {
3407 ret = close(sock);
3408 if (ret) {
3409 PERROR("close");
3410 }
3411 }
3412
3413 lttng_poll_clean(&events);
3414 clean_command_ctx(&cmd_ctx);
3415
3416 error_listen:
3417 error_create_poll:
3418 error_testpoint:
3419 unlink(client_unix_sock_path);
3420 if (client_sock >= 0) {
3421 ret = close(client_sock);
3422 if (ret) {
3423 PERROR("close");
3424 }
3425 }
3426
3427 if (err) {
3428 health_error();
3429 ERR("Health error occurred in %s", __func__);
3430 }
3431
3432 health_unregister();
3433
3434 DBG("Client thread dying");
3435
3436 rcu_unregister_thread();
3437 return NULL;
3438 }
3439
3440
3441 /*
3442 * usage function on stderr
3443 */
3444 static void usage(void)
3445 {
3446 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3447 fprintf(stderr, " -h, --help Display this usage.\n");
3448 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3449 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3450 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3451 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3452 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3453 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3454 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3455 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3456 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3457 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3458 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3459 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3460 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3461 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3462 fprintf(stderr, " -V, --version Show version number.\n");
3463 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3464 fprintf(stderr, " -q, --quiet No output at all.\n");
3465 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3466 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3467 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3468 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3469 }
3470
3471 /*
3472 * daemon argument parsing
3473 */
3474 static int parse_args(int argc, char **argv)
3475 {
3476 int c;
3477
3478 static struct option long_options[] = {
3479 { "client-sock", 1, 0, 'c' },
3480 { "apps-sock", 1, 0, 'a' },
3481 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3482 { "kconsumerd-err-sock", 1, 0, 'E' },
3483 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3484 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3485 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3486 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3487 { "consumerd32-path", 1, 0, 'u' },
3488 { "consumerd32-libdir", 1, 0, 'U' },
3489 { "consumerd64-path", 1, 0, 't' },
3490 { "consumerd64-libdir", 1, 0, 'T' },
3491 { "daemonize", 0, 0, 'd' },
3492 { "sig-parent", 0, 0, 'S' },
3493 { "help", 0, 0, 'h' },
3494 { "group", 1, 0, 'g' },
3495 { "version", 0, 0, 'V' },
3496 { "quiet", 0, 0, 'q' },
3497 { "verbose", 0, 0, 'v' },
3498 { "verbose-consumer", 0, 0, 'Z' },
3499 { "no-kernel", 0, 0, 'N' },
3500 { "pidfile", 1, 0, 'p' },
3501 { NULL, 0, 0, 0 }
3502 };
3503
3504 while (1) {
3505 int option_index = 0;
3506 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3507 long_options, &option_index);
3508 if (c == -1) {
3509 break;
3510 }
3511
3512 switch (c) {
3513 case 0:
3514 fprintf(stderr, "option %s", long_options[option_index].name);
3515 if (optarg) {
3516 fprintf(stderr, " with arg %s\n", optarg);
3517 }
3518 break;
3519 case 'c':
3520 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3521 break;
3522 case 'a':
3523 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3524 break;
3525 case 'd':
3526 opt_daemon = 1;
3527 break;
3528 case 'g':
3529 opt_tracing_group = optarg;
3530 break;
3531 case 'h':
3532 usage();
3533 exit(EXIT_FAILURE);
3534 case 'V':
3535 fprintf(stdout, "%s\n", VERSION);
3536 exit(EXIT_SUCCESS);
3537 case 'S':
3538 opt_sig_parent = 1;
3539 break;
3540 case 'E':
3541 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3542 break;
3543 case 'C':
3544 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3545 break;
3546 case 'F':
3547 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3548 break;
3549 case 'D':
3550 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3551 break;
3552 case 'H':
3553 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3554 break;
3555 case 'G':
3556 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3557 break;
3558 case 'N':
3559 opt_no_kernel = 1;
3560 break;
3561 case 'q':
3562 lttng_opt_quiet = 1;
3563 break;
3564 case 'v':
3565 /* Verbose level can increase using multiple -v */
3566 lttng_opt_verbose += 1;
3567 break;
3568 case 'Z':
3569 opt_verbose_consumer += 1;
3570 break;
3571 case 'u':
3572 consumerd32_bin= optarg;
3573 break;
3574 case 'U':
3575 consumerd32_libdir = optarg;
3576 break;
3577 case 't':
3578 consumerd64_bin = optarg;
3579 break;
3580 case 'T':
3581 consumerd64_libdir = optarg;
3582 break;
3583 case 'p':
3584 opt_pidfile = optarg;
3585 break;
3586 default:
3587 /* Unknown option or other error.
3588 * Error is printed by getopt, just return */
3589 return -1;
3590 }
3591 }
3592
3593 return 0;
3594 }
3595
3596 /*
3597 * Creates the two needed socket by the daemon.
3598 * apps_sock - The communication socket for all UST apps.
3599 * client_sock - The communication of the cli tool (lttng).
3600 */
3601 static int init_daemon_socket(void)
3602 {
3603 int ret = 0;
3604 mode_t old_umask;
3605
3606 old_umask = umask(0);
3607
3608 /* Create client tool unix socket */
3609 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3610 if (client_sock < 0) {
3611 ERR("Create unix sock failed: %s", client_unix_sock_path);
3612 ret = -1;
3613 goto end;
3614 }
3615
3616 /* Set the cloexec flag */
3617 ret = utils_set_fd_cloexec(client_sock);
3618 if (ret < 0) {
3619 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3620 "Continuing but note that the consumer daemon will have a "
3621 "reference to this socket on exec()", client_sock);
3622 }
3623
3624 /* File permission MUST be 660 */
3625 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3626 if (ret < 0) {
3627 ERR("Set file permissions failed: %s", client_unix_sock_path);
3628 PERROR("chmod");
3629 goto end;
3630 }
3631
3632 /* Create the application unix socket */
3633 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3634 if (apps_sock < 0) {
3635 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3636 ret = -1;
3637 goto end;
3638 }
3639
3640 /* Set the cloexec flag */
3641 ret = utils_set_fd_cloexec(apps_sock);
3642 if (ret < 0) {
3643 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3644 "Continuing but note that the consumer daemon will have a "
3645 "reference to this socket on exec()", apps_sock);
3646 }
3647
3648 /* File permission MUST be 666 */
3649 ret = chmod(apps_unix_sock_path,
3650 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3651 if (ret < 0) {
3652 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3653 PERROR("chmod");
3654 goto end;
3655 }
3656
3657 DBG3("Session daemon client socket %d and application socket %d created",
3658 client_sock, apps_sock);
3659
3660 end:
3661 umask(old_umask);
3662 return ret;
3663 }
3664
3665 /*
3666 * Check if the global socket is available, and if a daemon is answering at the
3667 * other side. If yes, error is returned.
3668 */
3669 static int check_existing_daemon(void)
3670 {
3671 /* Is there anybody out there ? */
3672 if (lttng_session_daemon_alive()) {
3673 return -EEXIST;
3674 }
3675
3676 return 0;
3677 }
3678
3679 /*
3680 * Set the tracing group gid onto the client socket.
3681 *
3682 * Race window between mkdir and chown is OK because we are going from more
3683 * permissive (root.root) to less permissive (root.tracing).
3684 */
3685 static int set_permissions(char *rundir)
3686 {
3687 int ret;
3688 gid_t gid;
3689
3690 ret = allowed_group();
3691 if (ret < 0) {
3692 WARN("No tracing group detected");
3693 ret = 0;
3694 goto end;
3695 }
3696
3697 gid = ret;
3698
3699 /* Set lttng run dir */
3700 ret = chown(rundir, 0, gid);
3701 if (ret < 0) {
3702 ERR("Unable to set group on %s", rundir);
3703 PERROR("chown");
3704 }
3705
3706 /* Ensure tracing group can search the run dir */
3707 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3708 if (ret < 0) {
3709 ERR("Unable to set permissions on %s", rundir);
3710 PERROR("chmod");
3711 }
3712
3713 /* lttng client socket path */
3714 ret = chown(client_unix_sock_path, 0, gid);
3715 if (ret < 0) {
3716 ERR("Unable to set group on %s", client_unix_sock_path);
3717 PERROR("chown");
3718 }
3719
3720 /* kconsumer error socket path */
3721 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3722 if (ret < 0) {
3723 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3724 PERROR("chown");
3725 }
3726
3727 /* 64-bit ustconsumer error socket path */
3728 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3729 if (ret < 0) {
3730 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3731 PERROR("chown");
3732 }
3733
3734 /* 32-bit ustconsumer compat32 error socket path */
3735 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3736 if (ret < 0) {
3737 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3738 PERROR("chown");
3739 }
3740
3741 DBG("All permissions are set");
3742
3743 end:
3744 return ret;
3745 }
3746
3747 /*
3748 * Create the lttng run directory needed for all global sockets and pipe.
3749 */
3750 static int create_lttng_rundir(const char *rundir)
3751 {
3752 int ret;
3753
3754 DBG3("Creating LTTng run directory: %s", rundir);
3755
3756 ret = mkdir(rundir, S_IRWXU);
3757 if (ret < 0) {
3758 if (errno != EEXIST) {
3759 ERR("Unable to create %s", rundir);
3760 goto error;
3761 } else {
3762 ret = 0;
3763 }
3764 }
3765
3766 error:
3767 return ret;
3768 }
3769
3770 /*
3771 * Setup sockets and directory needed by the kconsumerd communication with the
3772 * session daemon.
3773 */
3774 static int set_consumer_sockets(struct consumer_data *consumer_data,
3775 const char *rundir)
3776 {
3777 int ret;
3778 char path[PATH_MAX];
3779
3780 switch (consumer_data->type) {
3781 case LTTNG_CONSUMER_KERNEL:
3782 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3783 break;
3784 case LTTNG_CONSUMER64_UST:
3785 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3786 break;
3787 case LTTNG_CONSUMER32_UST:
3788 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3789 break;
3790 default:
3791 ERR("Consumer type unknown");
3792 ret = -EINVAL;
3793 goto error;
3794 }
3795
3796 DBG2("Creating consumer directory: %s", path);
3797
3798 ret = mkdir(path, S_IRWXU);
3799 if (ret < 0) {
3800 if (errno != EEXIST) {
3801 PERROR("mkdir");
3802 ERR("Failed to create %s", path);
3803 goto error;
3804 }
3805 ret = -1;
3806 }
3807
3808 /* Create the kconsumerd error unix socket */
3809 consumer_data->err_sock =
3810 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3811 if (consumer_data->err_sock < 0) {
3812 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3813 ret = -1;
3814 goto error;
3815 }
3816
3817 /* File permission MUST be 660 */
3818 ret = chmod(consumer_data->err_unix_sock_path,
3819 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3820 if (ret < 0) {
3821 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3822 PERROR("chmod");
3823 goto error;
3824 }
3825
3826 error:
3827 return ret;
3828 }
3829
3830 /*
3831 * Signal handler for the daemon
3832 *
3833 * Simply stop all worker threads, leaving main() return gracefully after
3834 * joining all threads and calling cleanup().
3835 */
3836 static void sighandler(int sig)
3837 {
3838 switch (sig) {
3839 case SIGPIPE:
3840 DBG("SIGPIPE caught");
3841 return;
3842 case SIGINT:
3843 DBG("SIGINT caught");
3844 stop_threads();
3845 break;
3846 case SIGTERM:
3847 DBG("SIGTERM caught");
3848 stop_threads();
3849 break;
3850 default:
3851 break;
3852 }
3853 }
3854
3855 /*
3856 * Setup signal handler for :
3857 * SIGINT, SIGTERM, SIGPIPE
3858 */
3859 static int set_signal_handler(void)
3860 {
3861 int ret = 0;
3862 struct sigaction sa;
3863 sigset_t sigset;
3864
3865 if ((ret = sigemptyset(&sigset)) < 0) {
3866 PERROR("sigemptyset");
3867 return ret;
3868 }
3869
3870 sa.sa_handler = sighandler;
3871 sa.sa_mask = sigset;
3872 sa.sa_flags = 0;
3873 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3874 PERROR("sigaction");
3875 return ret;
3876 }
3877
3878 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3879 PERROR("sigaction");
3880 return ret;
3881 }
3882
3883 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3884 PERROR("sigaction");
3885 return ret;
3886 }
3887
3888 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3889
3890 return ret;
3891 }
3892
3893 /*
3894 * Set open files limit to unlimited. This daemon can open a large number of
3895 * file descriptors in order to consumer multiple kernel traces.
3896 */
3897 static void set_ulimit(void)
3898 {
3899 int ret;
3900 struct rlimit lim;
3901
3902 /* The kernel does not allowed an infinite limit for open files */
3903 lim.rlim_cur = 65535;
3904 lim.rlim_max = 65535;
3905
3906 ret = setrlimit(RLIMIT_NOFILE, &lim);
3907 if (ret < 0) {
3908 PERROR("failed to set open files limit");
3909 }
3910 }
3911
3912 /*
3913 * Write pidfile using the rundir and opt_pidfile.
3914 */
3915 static void write_pidfile(void)
3916 {
3917 int ret;
3918 char pidfile_path[PATH_MAX];
3919
3920 assert(rundir);
3921
3922 if (opt_pidfile) {
3923 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
3924 } else {
3925 /* Build pidfile path from rundir and opt_pidfile. */
3926 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
3927 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
3928 if (ret < 0) {
3929 PERROR("snprintf pidfile path");
3930 goto error;
3931 }
3932 }
3933
3934 /*
3935 * Create pid file in rundir. Return value is of no importance. The
3936 * execution will continue even though we are not able to write the file.
3937 */
3938 (void) utils_create_pid_file(getpid(), pidfile_path);
3939
3940 error:
3941 return;
3942 }
3943
3944 /*
3945 * main
3946 */
3947 int main(int argc, char **argv)
3948 {
3949 int ret = 0;
3950 void *status;
3951 const char *home_path, *env_app_timeout;
3952
3953 init_kernel_workarounds();
3954
3955 rcu_register_thread();
3956
3957 setup_consumerd_path();
3958
3959 /* Parse arguments */
3960 progname = argv[0];
3961 if ((ret = parse_args(argc, argv)) < 0) {
3962 goto error;
3963 }
3964
3965 /* Daemonize */
3966 if (opt_daemon) {
3967 int i;
3968
3969 /*
3970 * fork
3971 * child: setsid, close FD 0, 1, 2, chdir /
3972 * parent: exit (if fork is successful)
3973 */
3974 ret = daemon(0, 0);
3975 if (ret < 0) {
3976 PERROR("daemon");
3977 goto error;
3978 }
3979 /*
3980 * We are in the child. Make sure all other file
3981 * descriptors are closed, in case we are called with
3982 * more opened file descriptors than the standard ones.
3983 */
3984 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3985 (void) close(i);
3986 }
3987 }
3988
3989 /* Create thread quit pipe */
3990 if ((ret = init_thread_quit_pipe()) < 0) {
3991 goto error;
3992 }
3993
3994 /* Check if daemon is UID = 0 */
3995 is_root = !getuid();
3996
3997 if (is_root) {
3998 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3999
4000 /* Create global run dir with root access */
4001 ret = create_lttng_rundir(rundir);
4002 if (ret < 0) {
4003 goto error;
4004 }
4005
4006 if (strlen(apps_unix_sock_path) == 0) {
4007 snprintf(apps_unix_sock_path, PATH_MAX,
4008 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4009 }
4010
4011 if (strlen(client_unix_sock_path) == 0) {
4012 snprintf(client_unix_sock_path, PATH_MAX,
4013 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4014 }
4015
4016 /* Set global SHM for ust */
4017 if (strlen(wait_shm_path) == 0) {
4018 snprintf(wait_shm_path, PATH_MAX,
4019 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4020 }
4021
4022 if (strlen(health_unix_sock_path) == 0) {
4023 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4024 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4025 }
4026
4027 /* Setup kernel consumerd path */
4028 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4029 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4030 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4031 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4032
4033 DBG2("Kernel consumer err path: %s",
4034 kconsumer_data.err_unix_sock_path);
4035 DBG2("Kernel consumer cmd path: %s",
4036 kconsumer_data.cmd_unix_sock_path);
4037 } else {
4038 home_path = get_home_dir();
4039 if (home_path == NULL) {
4040 /* TODO: Add --socket PATH option */
4041 ERR("Can't get HOME directory for sockets creation.");
4042 ret = -EPERM;
4043 goto error;
4044 }
4045
4046 /*
4047 * Create rundir from home path. This will create something like
4048 * $HOME/.lttng
4049 */
4050 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4051 if (ret < 0) {
4052 ret = -ENOMEM;
4053 goto error;
4054 }
4055
4056 ret = create_lttng_rundir(rundir);
4057 if (ret < 0) {
4058 goto error;
4059 }
4060
4061 if (strlen(apps_unix_sock_path) == 0) {
4062 snprintf(apps_unix_sock_path, PATH_MAX,
4063 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4064 }
4065
4066 /* Set the cli tool unix socket path */
4067 if (strlen(client_unix_sock_path) == 0) {
4068 snprintf(client_unix_sock_path, PATH_MAX,
4069 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4070 }
4071
4072 /* Set global SHM for ust */
4073 if (strlen(wait_shm_path) == 0) {
4074 snprintf(wait_shm_path, PATH_MAX,
4075 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4076 }
4077
4078 /* Set health check Unix path */
4079 if (strlen(health_unix_sock_path) == 0) {
4080 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4081 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4082 }
4083 }
4084
4085 /* Set consumer initial state */
4086 kernel_consumerd_state = CONSUMER_STOPPED;
4087 ust_consumerd_state = CONSUMER_STOPPED;
4088
4089 DBG("Client socket path %s", client_unix_sock_path);
4090 DBG("Application socket path %s", apps_unix_sock_path);
4091 DBG("Application wait path %s", wait_shm_path);
4092 DBG("LTTng run directory path: %s", rundir);
4093
4094 /* 32 bits consumerd path setup */
4095 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4096 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4097 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4098 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4099
4100 DBG2("UST consumer 32 bits err path: %s",
4101 ustconsumer32_data.err_unix_sock_path);
4102 DBG2("UST consumer 32 bits cmd path: %s",
4103 ustconsumer32_data.cmd_unix_sock_path);
4104
4105 /* 64 bits consumerd path setup */
4106 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4107 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4108 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4109 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4110
4111 DBG2("UST consumer 64 bits err path: %s",
4112 ustconsumer64_data.err_unix_sock_path);
4113 DBG2("UST consumer 64 bits cmd path: %s",
4114 ustconsumer64_data.cmd_unix_sock_path);
4115
4116 /*
4117 * See if daemon already exist.
4118 */
4119 if ((ret = check_existing_daemon()) < 0) {
4120 ERR("Already running daemon.\n");
4121 /*
4122 * We do not goto exit because we must not cleanup()
4123 * because a daemon is already running.
4124 */
4125 goto error;
4126 }
4127
4128 /*
4129 * Init UST app hash table. Alloc hash table before this point since
4130 * cleanup() can get called after that point.
4131 */
4132 ust_app_ht_alloc();
4133
4134 /* After this point, we can safely call cleanup() with "goto exit" */
4135
4136 /*
4137 * These actions must be executed as root. We do that *after* setting up
4138 * the sockets path because we MUST make the check for another daemon using
4139 * those paths *before* trying to set the kernel consumer sockets and init
4140 * kernel tracer.
4141 */
4142 if (is_root) {
4143 ret = set_consumer_sockets(&kconsumer_data, rundir);
4144 if (ret < 0) {
4145 goto exit;
4146 }
4147
4148 /* Setup kernel tracer */
4149 if (!opt_no_kernel) {
4150 init_kernel_tracer();
4151 }
4152
4153 /* Set ulimit for open files */
4154 set_ulimit();
4155 }
4156 /* init lttng_fd tracking must be done after set_ulimit. */
4157 lttng_fd_init();
4158
4159 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4160 if (ret < 0) {
4161 goto exit;
4162 }
4163
4164 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4165 if (ret < 0) {
4166 goto exit;
4167 }
4168
4169 if ((ret = set_signal_handler()) < 0) {
4170 goto exit;
4171 }
4172
4173 /* Setup the needed unix socket */
4174 if ((ret = init_daemon_socket()) < 0) {
4175 goto exit;
4176 }
4177
4178 /* Set credentials to socket */
4179 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4180 goto exit;
4181 }
4182
4183 /* Get parent pid if -S, --sig-parent is specified. */
4184 if (opt_sig_parent) {
4185 ppid = getppid();
4186 }
4187
4188 /* Setup the kernel pipe for waking up the kernel thread */
4189 if (is_root && !opt_no_kernel) {
4190 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4191 goto exit;
4192 }
4193 }
4194
4195 /* Setup the thread apps communication pipe. */
4196 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4197 goto exit;
4198 }
4199
4200 /* Setup the thread apps notify communication pipe. */
4201 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4202 goto exit;
4203 }
4204
4205 /* Init UST command queue. */
4206 cds_wfq_init(&ust_cmd_queue.queue);
4207
4208 /*
4209 * Get session list pointer. This pointer MUST NOT be free(). This list is
4210 * statically declared in session.c
4211 */
4212 session_list_ptr = session_get_list();
4213
4214 /* Set up max poll set size */
4215 lttng_poll_set_max_size();
4216
4217 cmd_init();
4218
4219 /* Check for the application socket timeout env variable. */
4220 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4221 if (env_app_timeout) {
4222 app_socket_timeout = atoi(env_app_timeout);
4223 } else {
4224 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4225 }
4226
4227 write_pidfile();
4228
4229 /* Create thread to manage the client socket */
4230 ret = pthread_create(&health_thread, NULL,
4231 thread_manage_health, (void *) NULL);
4232 if (ret != 0) {
4233 PERROR("pthread_create health");
4234 goto exit_health;
4235 }
4236
4237 /* Create thread to manage the client socket */
4238 ret = pthread_create(&client_thread, NULL,
4239 thread_manage_clients, (void *) NULL);
4240 if (ret != 0) {
4241 PERROR("pthread_create clients");
4242 goto exit_client;
4243 }
4244
4245 /* Create thread to dispatch registration */
4246 ret = pthread_create(&dispatch_thread, NULL,
4247 thread_dispatch_ust_registration, (void *) NULL);
4248 if (ret != 0) {
4249 PERROR("pthread_create dispatch");
4250 goto exit_dispatch;
4251 }
4252
4253 /* Create thread to manage application registration. */
4254 ret = pthread_create(&reg_apps_thread, NULL,
4255 thread_registration_apps, (void *) NULL);
4256 if (ret != 0) {
4257 PERROR("pthread_create registration");
4258 goto exit_reg_apps;
4259 }
4260
4261 /* Create thread to manage application socket */
4262 ret = pthread_create(&apps_thread, NULL,
4263 thread_manage_apps, (void *) NULL);
4264 if (ret != 0) {
4265 PERROR("pthread_create apps");
4266 goto exit_apps;
4267 }
4268
4269 /* Create thread to manage application notify socket */
4270 ret = pthread_create(&apps_notify_thread, NULL,
4271 ust_thread_manage_notify, (void *) NULL);
4272 if (ret != 0) {
4273 PERROR("pthread_create apps");
4274 goto exit_apps;
4275 }
4276
4277 /* Don't start this thread if kernel tracing is not requested nor root */
4278 if (is_root && !opt_no_kernel) {
4279 /* Create kernel thread to manage kernel event */
4280 ret = pthread_create(&kernel_thread, NULL,
4281 thread_manage_kernel, (void *) NULL);
4282 if (ret != 0) {
4283 PERROR("pthread_create kernel");
4284 goto exit_kernel;
4285 }
4286
4287 ret = pthread_join(kernel_thread, &status);
4288 if (ret != 0) {
4289 PERROR("pthread_join");
4290 goto error; /* join error, exit without cleanup */
4291 }
4292 }
4293
4294 exit_kernel:
4295 ret = pthread_join(apps_thread, &status);
4296 if (ret != 0) {
4297 PERROR("pthread_join");
4298 goto error; /* join error, exit without cleanup */
4299 }
4300
4301 exit_apps:
4302 ret = pthread_join(reg_apps_thread, &status);
4303 if (ret != 0) {
4304 PERROR("pthread_join");
4305 goto error; /* join error, exit without cleanup */
4306 }
4307
4308 exit_reg_apps:
4309 ret = pthread_join(dispatch_thread, &status);
4310 if (ret != 0) {
4311 PERROR("pthread_join");
4312 goto error; /* join error, exit without cleanup */
4313 }
4314
4315 exit_dispatch:
4316 ret = pthread_join(client_thread, &status);
4317 if (ret != 0) {
4318 PERROR("pthread_join");
4319 goto error; /* join error, exit without cleanup */
4320 }
4321
4322 ret = join_consumer_thread(&kconsumer_data);
4323 if (ret != 0) {
4324 PERROR("join_consumer");
4325 goto error; /* join error, exit without cleanup */
4326 }
4327
4328 ret = join_consumer_thread(&ustconsumer32_data);
4329 if (ret != 0) {
4330 PERROR("join_consumer ust32");
4331 goto error; /* join error, exit without cleanup */
4332 }
4333
4334 ret = join_consumer_thread(&ustconsumer64_data);
4335 if (ret != 0) {
4336 PERROR("join_consumer ust64");
4337 goto error; /* join error, exit without cleanup */
4338 }
4339
4340 exit_client:
4341 ret = pthread_join(health_thread, &status);
4342 if (ret != 0) {
4343 PERROR("pthread_join health thread");
4344 goto error; /* join error, exit without cleanup */
4345 }
4346
4347 exit_health:
4348 exit:
4349 /*
4350 * cleanup() is called when no other thread is running.
4351 */
4352 rcu_thread_online();
4353 cleanup();
4354 rcu_thread_offline();
4355 rcu_unregister_thread();
4356 if (!ret) {
4357 exit(EXIT_SUCCESS);
4358 }
4359 error:
4360 exit(EXIT_FAILURE);
4361 }
This page took 0.16875 seconds and 5 git commands to generate.