58333c22df73fed08c3c64df0f6dac056dfd283e
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/uatomic.h>
36 #include <unistd.h>
37 #include <config.h>
38
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "kernel.h"
55 #include "kernel-consumer.h"
56 #include "modprobe.h"
57 #include "shm.h"
58 #include "ust-ctl.h"
59 #include "ust-consumer.h"
60 #include "utils.h"
61 #include "fd-limit.h"
62 #include "health.h"
63 #include "testpoint.h"
64
65 #define CONSUMERD_FILE "lttng-consumerd"
66
67 /* Const values */
68 const char default_home_dir[] = DEFAULT_HOME_DIR;
69 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
70 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
71 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
72
73 const char *progname;
74 const char *opt_tracing_group;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread;
153 static pthread_t reg_apps_thread;
154 static pthread_t client_thread;
155 static pthread_t kernel_thread;
156 static pthread_t dispatch_thread;
157 static pthread_t health_thread;
158
159 /*
160 * UST registration command queue. This queue is tied with a futex and uses a N
161 * wakers / 1 waiter implemented and detailed in futex.c/.h
162 *
163 * The thread_manage_apps and thread_dispatch_ust_registration interact with
164 * this queue and the wait/wake scheme.
165 */
166 static struct ust_cmd_queue ust_cmd_queue;
167
168 /*
169 * Pointer initialized before thread creation.
170 *
171 * This points to the tracing session list containing the session count and a
172 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
173 * MUST NOT be taken if you call a public function in session.c.
174 *
175 * The lock is nested inside the structure: session_list_ptr->lock. Please use
176 * session_lock_list and session_unlock_list for lock acquisition.
177 */
178 static struct ltt_session_list *session_list_ptr;
179
180 int ust_consumerd64_fd = -1;
181 int ust_consumerd32_fd = -1;
182
183 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
184 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
185 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
186 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
187
188 static const char *module_proc_lttng = "/proc/lttng";
189
190 /*
191 * Consumer daemon state which is changed when spawning it, killing it or in
192 * case of a fatal error.
193 */
194 enum consumerd_state {
195 CONSUMER_STARTED = 1,
196 CONSUMER_STOPPED = 2,
197 CONSUMER_ERROR = 3,
198 };
199
200 /*
201 * This consumer daemon state is used to validate if a client command will be
202 * able to reach the consumer. If not, the client is informed. For instance,
203 * doing a "lttng start" when the consumer state is set to ERROR will return an
204 * error to the client.
205 *
206 * The following example shows a possible race condition of this scheme:
207 *
208 * consumer thread error happens
209 * client cmd arrives
210 * client cmd checks state -> still OK
211 * consumer thread exit, sets error
212 * client cmd try to talk to consumer
213 * ...
214 *
215 * However, since the consumer is a different daemon, we have no way of making
216 * sure the command will reach it safely even with this state flag. This is why
217 * we consider that up to the state validation during command processing, the
218 * command is safe. After that, we can not guarantee the correctness of the
219 * client request vis-a-vis the consumer.
220 */
221 static enum consumerd_state ust_consumerd_state;
222 static enum consumerd_state kernel_consumerd_state;
223
224 /* Used for the health monitoring of the session daemon. See health.h */
225 struct health_state health_thread_cmd;
226 struct health_state health_thread_app_manage;
227 struct health_state health_thread_app_reg;
228 struct health_state health_thread_kernel;
229
230 /*
231 * Socket timeout for receiving and sending in seconds.
232 */
233 static int app_socket_timeout;
234
235 static
236 void setup_consumerd_path(void)
237 {
238 const char *bin, *libdir;
239
240 /*
241 * Allow INSTALL_BIN_PATH to be used as a target path for the
242 * native architecture size consumer if CONFIG_CONSUMER*_PATH
243 * has not been defined.
244 */
245 #if (CAA_BITS_PER_LONG == 32)
246 if (!consumerd32_bin[0]) {
247 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
248 }
249 if (!consumerd32_libdir[0]) {
250 consumerd32_libdir = INSTALL_LIB_PATH;
251 }
252 #elif (CAA_BITS_PER_LONG == 64)
253 if (!consumerd64_bin[0]) {
254 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
255 }
256 if (!consumerd64_libdir[0]) {
257 consumerd64_libdir = INSTALL_LIB_PATH;
258 }
259 #else
260 #error "Unknown bitness"
261 #endif
262
263 /*
264 * runtime env. var. overrides the build default.
265 */
266 bin = getenv("LTTNG_CONSUMERD32_BIN");
267 if (bin) {
268 consumerd32_bin = bin;
269 }
270 bin = getenv("LTTNG_CONSUMERD64_BIN");
271 if (bin) {
272 consumerd64_bin = bin;
273 }
274 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
275 if (libdir) {
276 consumerd32_libdir = libdir;
277 }
278 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
279 if (libdir) {
280 consumerd64_libdir = libdir;
281 }
282 }
283
284 /*
285 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
286 */
287 static int create_thread_poll_set(struct lttng_poll_event *events,
288 unsigned int size)
289 {
290 int ret;
291
292 if (events == NULL || size == 0) {
293 ret = -1;
294 goto error;
295 }
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310 error:
311 return ret;
312 }
313
314 /*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
319 static int check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326 }
327
328 /*
329 * Return group ID of the tracing group or -1 if not found.
330 */
331 static gid_t allowed_group(void)
332 {
333 struct group *grp;
334
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345 }
346
347 /*
348 * Init thread quit pipe.
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352 static int init_thread_quit_pipe(void)
353 {
354 int ret, i;
355
356 ret = pipe(thread_quit_pipe);
357 if (ret < 0) {
358 PERROR("thread quit pipe");
359 goto error;
360 }
361
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
370 error:
371 return ret;
372 }
373
374 /*
375 * Stop all threads by closing the thread quit pipe.
376 */
377 static void stop_threads(void)
378 {
379 int ret;
380
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret = notify_thread_pipe(thread_quit_pipe[1]);
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
390 futex_nto1_wake(&ust_cmd_queue.futex);
391 }
392
393 /*
394 * Cleanup the daemon
395 */
396 static void cleanup(void)
397 {
398 int ret;
399 char *cmd = NULL;
400 struct ltt_session *sess, *stmp;
401
402 DBG("Cleaning up");
403
404 /* First thing first, stop all threads */
405 utils_close_pipe(thread_quit_pipe);
406
407 DBG("Removing %s directory", rundir);
408 ret = asprintf(&cmd, "rm -rf %s", rundir);
409 if (ret < 0) {
410 ERR("asprintf failed. Something is really wrong!");
411 }
412
413 /* Remove lttng run directory */
414 ret = system(cmd);
415 if (ret < 0) {
416 ERR("Unable to clean %s", rundir);
417 }
418 free(cmd);
419 free(rundir);
420
421 DBG("Cleaning up all sessions");
422
423 /* Destroy session list mutex */
424 if (session_list_ptr != NULL) {
425 pthread_mutex_destroy(&session_list_ptr->lock);
426
427 /* Cleanup ALL session */
428 cds_list_for_each_entry_safe(sess, stmp,
429 &session_list_ptr->head, list) {
430 cmd_destroy_session(sess, kernel_poll_pipe[1]);
431 }
432 }
433
434 DBG("Closing all UST sockets");
435 ust_app_clean_list();
436
437 if (is_root && !opt_no_kernel) {
438 DBG2("Closing kernel fd");
439 if (kernel_tracer_fd >= 0) {
440 ret = close(kernel_tracer_fd);
441 if (ret) {
442 PERROR("close");
443 }
444 }
445 DBG("Unloading kernel modules");
446 modprobe_remove_lttng_all();
447 }
448
449 /* <fun> */
450 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
451 "Matthew, BEET driven development works!%c[%dm",
452 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
453 /* </fun> */
454 }
455
456 /*
457 * Send data on a unix socket using the liblttsessiondcomm API.
458 *
459 * Return lttcomm error code.
460 */
461 static int send_unix_sock(int sock, void *buf, size_t len)
462 {
463 /* Check valid length */
464 if (len == 0) {
465 return -1;
466 }
467
468 return lttcomm_send_unix_sock(sock, buf, len);
469 }
470
471 /*
472 * Free memory of a command context structure.
473 */
474 static void clean_command_ctx(struct command_ctx **cmd_ctx)
475 {
476 DBG("Clean command context structure");
477 if (*cmd_ctx) {
478 if ((*cmd_ctx)->llm) {
479 free((*cmd_ctx)->llm);
480 }
481 if ((*cmd_ctx)->lsm) {
482 free((*cmd_ctx)->lsm);
483 }
484 free(*cmd_ctx);
485 *cmd_ctx = NULL;
486 }
487 }
488
489 /*
490 * Notify UST applications using the shm mmap futex.
491 */
492 static int notify_ust_apps(int active)
493 {
494 char *wait_shm_mmap;
495
496 DBG("Notifying applications of session daemon state: %d", active);
497
498 /* See shm.c for this call implying mmap, shm and futex calls */
499 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
500 if (wait_shm_mmap == NULL) {
501 goto error;
502 }
503
504 /* Wake waiting process */
505 futex_wait_update((int32_t *) wait_shm_mmap, active);
506
507 /* Apps notified successfully */
508 return 0;
509
510 error:
511 return -1;
512 }
513
514 /*
515 * Setup the outgoing data buffer for the response (llm) by allocating the
516 * right amount of memory and copying the original information from the lsm
517 * structure.
518 *
519 * Return total size of the buffer pointed by buf.
520 */
521 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
522 {
523 int ret, buf_size;
524
525 buf_size = size;
526
527 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
528 if (cmd_ctx->llm == NULL) {
529 PERROR("zmalloc");
530 ret = -ENOMEM;
531 goto error;
532 }
533
534 /* Copy common data */
535 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
536 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
537
538 cmd_ctx->llm->data_size = size;
539 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
540
541 return buf_size;
542
543 error:
544 return ret;
545 }
546
547 /*
548 * Update the kernel poll set of all channel fd available over all tracing
549 * session. Add the wakeup pipe at the end of the set.
550 */
551 static int update_kernel_poll(struct lttng_poll_event *events)
552 {
553 int ret;
554 struct ltt_session *session;
555 struct ltt_kernel_channel *channel;
556
557 DBG("Updating kernel poll set");
558
559 session_lock_list();
560 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
561 session_lock(session);
562 if (session->kernel_session == NULL) {
563 session_unlock(session);
564 continue;
565 }
566
567 cds_list_for_each_entry(channel,
568 &session->kernel_session->channel_list.head, list) {
569 /* Add channel fd to the kernel poll set */
570 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
571 if (ret < 0) {
572 session_unlock(session);
573 goto error;
574 }
575 DBG("Channel fd %d added to kernel set", channel->fd);
576 }
577 session_unlock(session);
578 }
579 session_unlock_list();
580
581 return 0;
582
583 error:
584 session_unlock_list();
585 return -1;
586 }
587
588 /*
589 * Find the channel fd from 'fd' over all tracing session. When found, check
590 * for new channel stream and send those stream fds to the kernel consumer.
591 *
592 * Useful for CPU hotplug feature.
593 */
594 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
595 {
596 int ret = 0;
597 struct ltt_session *session;
598 struct ltt_kernel_session *ksess;
599 struct ltt_kernel_channel *channel;
600
601 DBG("Updating kernel streams for channel fd %d", fd);
602
603 session_lock_list();
604 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
605 session_lock(session);
606 if (session->kernel_session == NULL) {
607 session_unlock(session);
608 continue;
609 }
610 ksess = session->kernel_session;
611
612 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
613 if (channel->fd == fd) {
614 DBG("Channel found, updating kernel streams");
615 ret = kernel_open_channel_stream(channel);
616 if (ret < 0) {
617 goto error;
618 }
619
620 /*
621 * Have we already sent fds to the consumer? If yes, it means
622 * that tracing is started so it is safe to send our updated
623 * stream fds.
624 */
625 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
626 struct lttng_ht_iter iter;
627 struct consumer_socket *socket;
628
629 rcu_read_lock();
630 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
631 &iter.iter, socket, node.node) {
632 /* Code flow error */
633 assert(socket->fd >= 0);
634
635 pthread_mutex_lock(socket->lock);
636 ret = kernel_consumer_send_channel_stream(socket,
637 channel, ksess);
638 pthread_mutex_unlock(socket->lock);
639 if (ret < 0) {
640 rcu_read_unlock();
641 goto error;
642 }
643 }
644 rcu_read_unlock();
645 }
646 goto error;
647 }
648 }
649 session_unlock(session);
650 }
651 session_unlock_list();
652 return ret;
653
654 error:
655 session_unlock(session);
656 session_unlock_list();
657 return ret;
658 }
659
660 /*
661 * For each tracing session, update newly registered apps.
662 */
663 static void update_ust_app(int app_sock)
664 {
665 struct ltt_session *sess, *stmp;
666
667 session_lock_list();
668
669 /* For all tracing session(s) */
670 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
671 session_lock(sess);
672 if (sess->ust_session) {
673 ust_app_global_update(sess->ust_session, app_sock);
674 }
675 session_unlock(sess);
676 }
677
678 session_unlock_list();
679 }
680
681 /*
682 * This thread manage event coming from the kernel.
683 *
684 * Features supported in this thread:
685 * -) CPU Hotplug
686 */
687 static void *thread_manage_kernel(void *data)
688 {
689 int ret, i, pollfd, update_poll_flag = 1, err = -1;
690 uint32_t revents, nb_fd;
691 char tmp;
692 struct lttng_poll_event events;
693
694 DBG("[thread] Thread manage kernel started");
695
696 health_register(HEALTH_TYPE_KERNEL);
697
698 /*
699 * This first step of the while is to clean this structure which could free
700 * non NULL pointers so zero it before the loop.
701 */
702 memset(&events, 0, sizeof(events));
703
704 if (testpoint(thread_manage_kernel)) {
705 goto error_testpoint;
706 }
707
708 health_code_update(&health_thread_kernel);
709
710 if (testpoint(thread_manage_kernel_before_loop)) {
711 goto error_testpoint;
712 }
713
714 while (1) {
715 health_code_update(&health_thread_kernel);
716
717 if (update_poll_flag == 1) {
718 /* Clean events object. We are about to populate it again. */
719 lttng_poll_clean(&events);
720
721 ret = create_thread_poll_set(&events, 2);
722 if (ret < 0) {
723 goto error_poll_create;
724 }
725
726 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
727 if (ret < 0) {
728 goto error;
729 }
730
731 /* This will add the available kernel channel if any. */
732 ret = update_kernel_poll(&events);
733 if (ret < 0) {
734 goto error;
735 }
736 update_poll_flag = 0;
737 }
738
739 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
740
741 /* Poll infinite value of time */
742 restart:
743 health_poll_update(&health_thread_kernel);
744 ret = lttng_poll_wait(&events, -1);
745 health_poll_update(&health_thread_kernel);
746 if (ret < 0) {
747 /*
748 * Restart interrupted system call.
749 */
750 if (errno == EINTR) {
751 goto restart;
752 }
753 goto error;
754 } else if (ret == 0) {
755 /* Should not happen since timeout is infinite */
756 ERR("Return value of poll is 0 with an infinite timeout.\n"
757 "This should not have happened! Continuing...");
758 continue;
759 }
760
761 nb_fd = ret;
762
763 for (i = 0; i < nb_fd; i++) {
764 /* Fetch once the poll data */
765 revents = LTTNG_POLL_GETEV(&events, i);
766 pollfd = LTTNG_POLL_GETFD(&events, i);
767
768 health_code_update(&health_thread_kernel);
769
770 /* Thread quit pipe has been closed. Killing thread. */
771 ret = check_thread_quit_pipe(pollfd, revents);
772 if (ret) {
773 err = 0;
774 goto exit;
775 }
776
777 /* Check for data on kernel pipe */
778 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
779 do {
780 ret = read(kernel_poll_pipe[0], &tmp, 1);
781 } while (ret < 0 && errno == EINTR);
782 /*
783 * Ret value is useless here, if this pipe gets any actions an
784 * update is required anyway.
785 */
786 update_poll_flag = 1;
787 continue;
788 } else {
789 /*
790 * New CPU detected by the kernel. Adding kernel stream to
791 * kernel session and updating the kernel consumer
792 */
793 if (revents & LPOLLIN) {
794 ret = update_kernel_stream(&kconsumer_data, pollfd);
795 if (ret < 0) {
796 continue;
797 }
798 break;
799 /*
800 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
801 * and unregister kernel stream at this point.
802 */
803 }
804 }
805 }
806 }
807
808 exit:
809 error:
810 lttng_poll_clean(&events);
811 error_poll_create:
812 error_testpoint:
813 utils_close_pipe(kernel_poll_pipe);
814 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
815 if (err) {
816 health_error(&health_thread_kernel);
817 ERR("Health error occurred in %s", __func__);
818 WARN("Kernel thread died unexpectedly. "
819 "Kernel tracing can continue but CPU hotplug is disabled.");
820 }
821 health_unregister();
822 DBG("Kernel thread dying");
823 return NULL;
824 }
825
826 /*
827 * Signal pthread condition of the consumer data that the thread.
828 */
829 static void signal_consumer_condition(struct consumer_data *data, int state)
830 {
831 pthread_mutex_lock(&data->cond_mutex);
832
833 /*
834 * The state is set before signaling. It can be any value, it's the waiter
835 * job to correctly interpret this condition variable associated to the
836 * consumer pthread_cond.
837 *
838 * A value of 0 means that the corresponding thread of the consumer data
839 * was not started. 1 indicates that the thread has started and is ready
840 * for action. A negative value means that there was an error during the
841 * thread bootstrap.
842 */
843 data->consumer_thread_is_ready = state;
844 (void) pthread_cond_signal(&data->cond);
845
846 pthread_mutex_unlock(&data->cond_mutex);
847 }
848
849 /*
850 * This thread manage the consumer error sent back to the session daemon.
851 */
852 static void *thread_manage_consumer(void *data)
853 {
854 int sock = -1, i, ret, pollfd, err = -1;
855 uint32_t revents, nb_fd;
856 enum lttcomm_return_code code;
857 struct lttng_poll_event events;
858 struct consumer_data *consumer_data = data;
859
860 DBG("[thread] Manage consumer started");
861
862 health_register(HEALTH_TYPE_CONSUMER);
863
864 /*
865 * Since the consumer thread can be spawned at any moment in time, we init
866 * the health to a poll status (1, which is a valid health over time).
867 * When the thread starts, we update here the health to a "code" path being
868 * an even value so this thread, when reaching a poll wait, does not
869 * trigger an error with an even value.
870 *
871 * Here is the use case we avoid.
872 *
873 * +1: the first poll update during initialization (main())
874 * +2 * x: multiple code update once in this thread.
875 * +1: poll wait in this thread (being a good health state).
876 * == even number which after the wait period shows as a bad health.
877 *
878 * In a nutshell, the following poll update to the health state brings back
879 * the state to an even value meaning a code path.
880 */
881 health_poll_update(&consumer_data->health);
882
883 /*
884 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
885 * Nothing more will be added to this poll set.
886 */
887 ret = create_thread_poll_set(&events, 2);
888 if (ret < 0) {
889 goto error_poll;
890 }
891
892 /*
893 * The error socket here is already in a listening state which was done
894 * just before spawning this thread to avoid a race between the consumer
895 * daemon exec trying to connect and the listen() call.
896 */
897 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
898 if (ret < 0) {
899 goto error;
900 }
901
902 health_code_update(&consumer_data->health);
903
904 /* Inifinite blocking call, waiting for transmission */
905 restart:
906 health_poll_update(&consumer_data->health);
907
908 if (testpoint(thread_manage_consumer)) {
909 goto error;
910 }
911
912 ret = lttng_poll_wait(&events, -1);
913 health_poll_update(&consumer_data->health);
914 if (ret < 0) {
915 /*
916 * Restart interrupted system call.
917 */
918 if (errno == EINTR) {
919 goto restart;
920 }
921 goto error;
922 }
923
924 nb_fd = ret;
925
926 for (i = 0; i < nb_fd; i++) {
927 /* Fetch once the poll data */
928 revents = LTTNG_POLL_GETEV(&events, i);
929 pollfd = LTTNG_POLL_GETFD(&events, i);
930
931 health_code_update(&consumer_data->health);
932
933 /* Thread quit pipe has been closed. Killing thread. */
934 ret = check_thread_quit_pipe(pollfd, revents);
935 if (ret) {
936 err = 0;
937 goto exit;
938 }
939
940 /* Event on the registration socket */
941 if (pollfd == consumer_data->err_sock) {
942 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
943 ERR("consumer err socket poll error");
944 goto error;
945 }
946 }
947 }
948
949 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
950 if (sock < 0) {
951 goto error;
952 }
953
954 /*
955 * Set the CLOEXEC flag. Return code is useless because either way, the
956 * show must go on.
957 */
958 (void) utils_set_fd_cloexec(sock);
959
960 health_code_update(&consumer_data->health);
961
962 DBG2("Receiving code from consumer err_sock");
963
964 /* Getting status code from kconsumerd */
965 ret = lttcomm_recv_unix_sock(sock, &code,
966 sizeof(enum lttcomm_return_code));
967 if (ret <= 0) {
968 goto error;
969 }
970
971 health_code_update(&consumer_data->health);
972
973 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
974 consumer_data->cmd_sock =
975 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
976 if (consumer_data->cmd_sock < 0) {
977 /* On error, signal condition and quit. */
978 signal_consumer_condition(consumer_data, -1);
979 PERROR("consumer connect");
980 goto error;
981 }
982 signal_consumer_condition(consumer_data, 1);
983 DBG("Consumer command socket ready");
984 } else {
985 ERR("consumer error when waiting for SOCK_READY : %s",
986 lttcomm_get_readable_code(-code));
987 goto error;
988 }
989
990 /* Remove the kconsumerd error sock since we've established a connexion */
991 ret = lttng_poll_del(&events, consumer_data->err_sock);
992 if (ret < 0) {
993 goto error;
994 }
995
996 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
997 if (ret < 0) {
998 goto error;
999 }
1000
1001 health_code_update(&consumer_data->health);
1002
1003 /* Inifinite blocking call, waiting for transmission */
1004 restart_poll:
1005 health_poll_update(&consumer_data->health);
1006 ret = lttng_poll_wait(&events, -1);
1007 health_poll_update(&consumer_data->health);
1008 if (ret < 0) {
1009 /*
1010 * Restart interrupted system call.
1011 */
1012 if (errno == EINTR) {
1013 goto restart_poll;
1014 }
1015 goto error;
1016 }
1017
1018 nb_fd = ret;
1019
1020 for (i = 0; i < nb_fd; i++) {
1021 /* Fetch once the poll data */
1022 revents = LTTNG_POLL_GETEV(&events, i);
1023 pollfd = LTTNG_POLL_GETFD(&events, i);
1024
1025 health_code_update(&consumer_data->health);
1026
1027 /* Thread quit pipe has been closed. Killing thread. */
1028 ret = check_thread_quit_pipe(pollfd, revents);
1029 if (ret) {
1030 err = 0;
1031 goto exit;
1032 }
1033
1034 /* Event on the kconsumerd socket */
1035 if (pollfd == sock) {
1036 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1037 ERR("consumer err socket second poll error");
1038 goto error;
1039 }
1040 }
1041 }
1042
1043 health_code_update(&consumer_data->health);
1044
1045 /* Wait for any kconsumerd error */
1046 ret = lttcomm_recv_unix_sock(sock, &code,
1047 sizeof(enum lttcomm_return_code));
1048 if (ret <= 0) {
1049 ERR("consumer closed the command socket");
1050 goto error;
1051 }
1052
1053 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1054
1055 exit:
1056 error:
1057 /* Immediately set the consumerd state to stopped */
1058 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1059 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1060 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1061 consumer_data->type == LTTNG_CONSUMER32_UST) {
1062 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1063 } else {
1064 /* Code flow error... */
1065 assert(0);
1066 }
1067
1068 if (consumer_data->err_sock >= 0) {
1069 ret = close(consumer_data->err_sock);
1070 if (ret) {
1071 PERROR("close");
1072 }
1073 }
1074 if (consumer_data->cmd_sock >= 0) {
1075 ret = close(consumer_data->cmd_sock);
1076 if (ret) {
1077 PERROR("close");
1078 }
1079 }
1080 if (sock >= 0) {
1081 ret = close(sock);
1082 if (ret) {
1083 PERROR("close");
1084 }
1085 }
1086
1087 unlink(consumer_data->err_unix_sock_path);
1088 unlink(consumer_data->cmd_unix_sock_path);
1089 consumer_data->pid = 0;
1090
1091 lttng_poll_clean(&events);
1092 error_poll:
1093 if (err) {
1094 health_error(&consumer_data->health);
1095 ERR("Health error occurred in %s", __func__);
1096 }
1097 health_unregister();
1098 DBG("consumer thread cleanup completed");
1099
1100 return NULL;
1101 }
1102
1103 /*
1104 * This thread manage application communication.
1105 */
1106 static void *thread_manage_apps(void *data)
1107 {
1108 int i, ret, pollfd, err = -1;
1109 uint32_t revents, nb_fd;
1110 struct ust_command ust_cmd;
1111 struct lttng_poll_event events;
1112
1113 DBG("[thread] Manage application started");
1114
1115 rcu_register_thread();
1116 rcu_thread_online();
1117
1118 health_register(HEALTH_TYPE_APP_MANAGE);
1119
1120 if (testpoint(thread_manage_apps)) {
1121 goto error_testpoint;
1122 }
1123
1124 health_code_update(&health_thread_app_manage);
1125
1126 ret = create_thread_poll_set(&events, 2);
1127 if (ret < 0) {
1128 goto error_poll_create;
1129 }
1130
1131 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1132 if (ret < 0) {
1133 goto error;
1134 }
1135
1136 if (testpoint(thread_manage_apps_before_loop)) {
1137 goto error;
1138 }
1139
1140 health_code_update(&health_thread_app_manage);
1141
1142 while (1) {
1143 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1144
1145 /* Inifinite blocking call, waiting for transmission */
1146 restart:
1147 health_poll_update(&health_thread_app_manage);
1148 ret = lttng_poll_wait(&events, -1);
1149 health_poll_update(&health_thread_app_manage);
1150 if (ret < 0) {
1151 /*
1152 * Restart interrupted system call.
1153 */
1154 if (errno == EINTR) {
1155 goto restart;
1156 }
1157 goto error;
1158 }
1159
1160 nb_fd = ret;
1161
1162 for (i = 0; i < nb_fd; i++) {
1163 /* Fetch once the poll data */
1164 revents = LTTNG_POLL_GETEV(&events, i);
1165 pollfd = LTTNG_POLL_GETFD(&events, i);
1166
1167 health_code_update(&health_thread_app_manage);
1168
1169 /* Thread quit pipe has been closed. Killing thread. */
1170 ret = check_thread_quit_pipe(pollfd, revents);
1171 if (ret) {
1172 err = 0;
1173 goto exit;
1174 }
1175
1176 /* Inspect the apps cmd pipe */
1177 if (pollfd == apps_cmd_pipe[0]) {
1178 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1179 ERR("Apps command pipe error");
1180 goto error;
1181 } else if (revents & LPOLLIN) {
1182 /* Empty pipe */
1183 do {
1184 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1185 } while (ret < 0 && errno == EINTR);
1186 if (ret < 0 || ret < sizeof(ust_cmd)) {
1187 PERROR("read apps cmd pipe");
1188 goto error;
1189 }
1190
1191 health_code_update(&health_thread_app_manage);
1192
1193 /* Register applicaton to the session daemon */
1194 ret = ust_app_register(&ust_cmd.reg_msg,
1195 ust_cmd.sock);
1196 if (ret == -ENOMEM) {
1197 goto error;
1198 } else if (ret < 0) {
1199 break;
1200 }
1201
1202 health_code_update(&health_thread_app_manage);
1203
1204 /*
1205 * Validate UST version compatibility.
1206 */
1207 ret = ust_app_validate_version(ust_cmd.sock);
1208 if (ret >= 0) {
1209 /*
1210 * Add channel(s) and event(s) to newly registered apps
1211 * from lttng global UST domain.
1212 */
1213 update_ust_app(ust_cmd.sock);
1214 }
1215
1216 health_code_update(&health_thread_app_manage);
1217
1218 ret = ust_app_register_done(ust_cmd.sock);
1219 if (ret < 0) {
1220 /*
1221 * If the registration is not possible, we simply
1222 * unregister the apps and continue
1223 */
1224 ust_app_unregister(ust_cmd.sock);
1225 } else {
1226 /*
1227 * We only monitor the error events of the socket. This
1228 * thread does not handle any incoming data from UST
1229 * (POLLIN).
1230 */
1231 ret = lttng_poll_add(&events, ust_cmd.sock,
1232 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1233 if (ret < 0) {
1234 goto error;
1235 }
1236
1237 /* Set socket timeout for both receiving and ending */
1238 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1239 app_socket_timeout);
1240 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1241 app_socket_timeout);
1242
1243 DBG("Apps with sock %d added to poll set",
1244 ust_cmd.sock);
1245 }
1246
1247 health_code_update(&health_thread_app_manage);
1248
1249 break;
1250 }
1251 } else {
1252 /*
1253 * At this point, we know that a registered application made
1254 * the event at poll_wait.
1255 */
1256 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1257 /* Removing from the poll set */
1258 ret = lttng_poll_del(&events, pollfd);
1259 if (ret < 0) {
1260 goto error;
1261 }
1262
1263 /* Socket closed on remote end. */
1264 ust_app_unregister(pollfd);
1265 break;
1266 }
1267 }
1268
1269 health_code_update(&health_thread_app_manage);
1270 }
1271 }
1272
1273 exit:
1274 error:
1275 lttng_poll_clean(&events);
1276 error_poll_create:
1277 error_testpoint:
1278 utils_close_pipe(apps_cmd_pipe);
1279 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1280
1281 /*
1282 * We don't clean the UST app hash table here since already registered
1283 * applications can still be controlled so let them be until the session
1284 * daemon dies or the applications stop.
1285 */
1286
1287 if (err) {
1288 health_error(&health_thread_app_manage);
1289 ERR("Health error occurred in %s", __func__);
1290 }
1291 health_unregister();
1292 DBG("Application communication apps thread cleanup complete");
1293 rcu_thread_offline();
1294 rcu_unregister_thread();
1295 return NULL;
1296 }
1297
1298 /*
1299 * Dispatch request from the registration threads to the application
1300 * communication thread.
1301 */
1302 static void *thread_dispatch_ust_registration(void *data)
1303 {
1304 int ret;
1305 struct cds_wfq_node *node;
1306 struct ust_command *ust_cmd = NULL;
1307
1308 DBG("[thread] Dispatch UST command started");
1309
1310 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1311 /* Atomically prepare the queue futex */
1312 futex_nto1_prepare(&ust_cmd_queue.futex);
1313
1314 do {
1315 /* Dequeue command for registration */
1316 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1317 if (node == NULL) {
1318 DBG("Woken up but nothing in the UST command queue");
1319 /* Continue thread execution */
1320 break;
1321 }
1322
1323 ust_cmd = caa_container_of(node, struct ust_command, node);
1324
1325 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1326 " gid:%d sock:%d name:%s (version %d.%d)",
1327 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1328 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1329 ust_cmd->sock, ust_cmd->reg_msg.name,
1330 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1331 /*
1332 * Inform apps thread of the new application registration. This
1333 * call is blocking so we can be assured that the data will be read
1334 * at some point in time or wait to the end of the world :)
1335 */
1336 if (apps_cmd_pipe[1] >= 0) {
1337 do {
1338 ret = write(apps_cmd_pipe[1], ust_cmd,
1339 sizeof(struct ust_command));
1340 } while (ret < 0 && errno == EINTR);
1341 if (ret < 0 || ret != sizeof(struct ust_command)) {
1342 PERROR("write apps cmd pipe");
1343 if (errno == EBADF) {
1344 /*
1345 * We can't inform the application thread to process
1346 * registration. We will exit or else application
1347 * registration will not occur and tracing will never
1348 * start.
1349 */
1350 goto error;
1351 }
1352 }
1353 } else {
1354 /* Application manager thread is not available. */
1355 ret = close(ust_cmd->sock);
1356 if (ret < 0) {
1357 PERROR("close ust_cmd sock");
1358 }
1359 }
1360 free(ust_cmd);
1361 } while (node != NULL);
1362
1363 /* Futex wait on queue. Blocking call on futex() */
1364 futex_nto1_wait(&ust_cmd_queue.futex);
1365 }
1366
1367 error:
1368 DBG("Dispatch thread dying");
1369 return NULL;
1370 }
1371
1372 /*
1373 * This thread manage application registration.
1374 */
1375 static void *thread_registration_apps(void *data)
1376 {
1377 int sock = -1, i, ret, pollfd, err = -1;
1378 uint32_t revents, nb_fd;
1379 struct lttng_poll_event events;
1380 /*
1381 * Get allocated in this thread, enqueued to a global queue, dequeued and
1382 * freed in the manage apps thread.
1383 */
1384 struct ust_command *ust_cmd = NULL;
1385
1386 DBG("[thread] Manage application registration started");
1387
1388 health_register(HEALTH_TYPE_APP_REG);
1389
1390 if (testpoint(thread_registration_apps)) {
1391 goto error_testpoint;
1392 }
1393
1394 ret = lttcomm_listen_unix_sock(apps_sock);
1395 if (ret < 0) {
1396 goto error_listen;
1397 }
1398
1399 /*
1400 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1401 * more will be added to this poll set.
1402 */
1403 ret = create_thread_poll_set(&events, 2);
1404 if (ret < 0) {
1405 goto error_create_poll;
1406 }
1407
1408 /* Add the application registration socket */
1409 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1410 if (ret < 0) {
1411 goto error_poll_add;
1412 }
1413
1414 /* Notify all applications to register */
1415 ret = notify_ust_apps(1);
1416 if (ret < 0) {
1417 ERR("Failed to notify applications or create the wait shared memory.\n"
1418 "Execution continues but there might be problem for already\n"
1419 "running applications that wishes to register.");
1420 }
1421
1422 while (1) {
1423 DBG("Accepting application registration");
1424
1425 /* Inifinite blocking call, waiting for transmission */
1426 restart:
1427 health_poll_update(&health_thread_app_reg);
1428 ret = lttng_poll_wait(&events, -1);
1429 health_poll_update(&health_thread_app_reg);
1430 if (ret < 0) {
1431 /*
1432 * Restart interrupted system call.
1433 */
1434 if (errno == EINTR) {
1435 goto restart;
1436 }
1437 goto error;
1438 }
1439
1440 nb_fd = ret;
1441
1442 for (i = 0; i < nb_fd; i++) {
1443 health_code_update(&health_thread_app_reg);
1444
1445 /* Fetch once the poll data */
1446 revents = LTTNG_POLL_GETEV(&events, i);
1447 pollfd = LTTNG_POLL_GETFD(&events, i);
1448
1449 /* Thread quit pipe has been closed. Killing thread. */
1450 ret = check_thread_quit_pipe(pollfd, revents);
1451 if (ret) {
1452 err = 0;
1453 goto exit;
1454 }
1455
1456 /* Event on the registration socket */
1457 if (pollfd == apps_sock) {
1458 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1459 ERR("Register apps socket poll error");
1460 goto error;
1461 } else if (revents & LPOLLIN) {
1462 sock = lttcomm_accept_unix_sock(apps_sock);
1463 if (sock < 0) {
1464 goto error;
1465 }
1466
1467 /*
1468 * Set the CLOEXEC flag. Return code is useless because
1469 * either way, the show must go on.
1470 */
1471 (void) utils_set_fd_cloexec(sock);
1472
1473 /* Create UST registration command for enqueuing */
1474 ust_cmd = zmalloc(sizeof(struct ust_command));
1475 if (ust_cmd == NULL) {
1476 PERROR("ust command zmalloc");
1477 goto error;
1478 }
1479
1480 /*
1481 * Using message-based transmissions to ensure we don't
1482 * have to deal with partially received messages.
1483 */
1484 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1485 if (ret < 0) {
1486 ERR("Exhausted file descriptors allowed for applications.");
1487 free(ust_cmd);
1488 ret = close(sock);
1489 if (ret) {
1490 PERROR("close");
1491 }
1492 sock = -1;
1493 continue;
1494 }
1495 health_code_update(&health_thread_app_reg);
1496 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1497 sizeof(struct ust_register_msg));
1498 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1499 if (ret < 0) {
1500 PERROR("lttcomm_recv_unix_sock register apps");
1501 } else {
1502 ERR("Wrong size received on apps register");
1503 }
1504 free(ust_cmd);
1505 ret = close(sock);
1506 if (ret) {
1507 PERROR("close");
1508 }
1509 lttng_fd_put(LTTNG_FD_APPS, 1);
1510 sock = -1;
1511 continue;
1512 }
1513 health_code_update(&health_thread_app_reg);
1514
1515 ust_cmd->sock = sock;
1516 sock = -1;
1517
1518 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1519 " gid:%d sock:%d name:%s (version %d.%d)",
1520 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1521 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1522 ust_cmd->sock, ust_cmd->reg_msg.name,
1523 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1524
1525 /*
1526 * Lock free enqueue the registration request. The red pill
1527 * has been taken! This apps will be part of the *system*.
1528 */
1529 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1530
1531 /*
1532 * Wake the registration queue futex. Implicit memory
1533 * barrier with the exchange in cds_wfq_enqueue.
1534 */
1535 futex_nto1_wake(&ust_cmd_queue.futex);
1536 }
1537 }
1538 }
1539 }
1540
1541 exit:
1542 error:
1543 if (err) {
1544 health_error(&health_thread_app_reg);
1545 ERR("Health error occurred in %s", __func__);
1546 }
1547
1548 /* Notify that the registration thread is gone */
1549 notify_ust_apps(0);
1550
1551 if (apps_sock >= 0) {
1552 ret = close(apps_sock);
1553 if (ret) {
1554 PERROR("close");
1555 }
1556 }
1557 if (sock >= 0) {
1558 ret = close(sock);
1559 if (ret) {
1560 PERROR("close");
1561 }
1562 lttng_fd_put(LTTNG_FD_APPS, 1);
1563 }
1564 unlink(apps_unix_sock_path);
1565
1566 error_poll_add:
1567 lttng_poll_clean(&events);
1568 error_listen:
1569 error_create_poll:
1570 error_testpoint:
1571 DBG("UST Registration thread cleanup complete");
1572 health_unregister();
1573
1574 return NULL;
1575 }
1576
1577 /*
1578 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1579 * exec or it will fails.
1580 */
1581 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1582 {
1583 int ret, clock_ret;
1584 struct timespec timeout;
1585
1586 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1587 consumer_data->consumer_thread_is_ready = 0;
1588
1589 /* Setup pthread condition */
1590 ret = pthread_condattr_init(&consumer_data->condattr);
1591 if (ret != 0) {
1592 errno = ret;
1593 PERROR("pthread_condattr_init consumer data");
1594 goto error;
1595 }
1596
1597 /*
1598 * Set the monotonic clock in order to make sure we DO NOT jump in time
1599 * between the clock_gettime() call and the timedwait call. See bug #324
1600 * for a more details and how we noticed it.
1601 */
1602 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1603 if (ret != 0) {
1604 errno = ret;
1605 PERROR("pthread_condattr_setclock consumer data");
1606 goto error;
1607 }
1608
1609 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1610 if (ret != 0) {
1611 errno = ret;
1612 PERROR("pthread_cond_init consumer data");
1613 goto error;
1614 }
1615
1616 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1617 consumer_data);
1618 if (ret != 0) {
1619 PERROR("pthread_create consumer");
1620 ret = -1;
1621 goto error;
1622 }
1623
1624 /* We are about to wait on a pthread condition */
1625 pthread_mutex_lock(&consumer_data->cond_mutex);
1626
1627 /* Get time for sem_timedwait absolute timeout */
1628 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1629 /*
1630 * Set the timeout for the condition timed wait even if the clock gettime
1631 * call fails since we might loop on that call and we want to avoid to
1632 * increment the timeout too many times.
1633 */
1634 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1635
1636 /*
1637 * The following loop COULD be skipped in some conditions so this is why we
1638 * set ret to 0 in order to make sure at least one round of the loop is
1639 * done.
1640 */
1641 ret = 0;
1642
1643 /*
1644 * Loop until the condition is reached or when a timeout is reached. Note
1645 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1646 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1647 * possible. This loop does not take any chances and works with both of
1648 * them.
1649 */
1650 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1651 if (clock_ret < 0) {
1652 PERROR("clock_gettime spawn consumer");
1653 /* Infinite wait for the consumerd thread to be ready */
1654 ret = pthread_cond_wait(&consumer_data->cond,
1655 &consumer_data->cond_mutex);
1656 } else {
1657 ret = pthread_cond_timedwait(&consumer_data->cond,
1658 &consumer_data->cond_mutex, &timeout);
1659 }
1660 }
1661
1662 /* Release the pthread condition */
1663 pthread_mutex_unlock(&consumer_data->cond_mutex);
1664
1665 if (ret != 0) {
1666 errno = ret;
1667 if (ret == ETIMEDOUT) {
1668 /*
1669 * Call has timed out so we kill the kconsumerd_thread and return
1670 * an error.
1671 */
1672 ERR("Condition timed out. The consumer thread was never ready."
1673 " Killing it");
1674 ret = pthread_cancel(consumer_data->thread);
1675 if (ret < 0) {
1676 PERROR("pthread_cancel consumer thread");
1677 }
1678 } else {
1679 PERROR("pthread_cond_wait failed consumer thread");
1680 }
1681 goto error;
1682 }
1683
1684 pthread_mutex_lock(&consumer_data->pid_mutex);
1685 if (consumer_data->pid == 0) {
1686 ERR("Consumerd did not start");
1687 pthread_mutex_unlock(&consumer_data->pid_mutex);
1688 goto error;
1689 }
1690 pthread_mutex_unlock(&consumer_data->pid_mutex);
1691
1692 return 0;
1693
1694 error:
1695 return ret;
1696 }
1697
1698 /*
1699 * Join consumer thread
1700 */
1701 static int join_consumer_thread(struct consumer_data *consumer_data)
1702 {
1703 void *status;
1704
1705 /* Consumer pid must be a real one. */
1706 if (consumer_data->pid > 0) {
1707 int ret;
1708 ret = kill(consumer_data->pid, SIGTERM);
1709 if (ret) {
1710 ERR("Error killing consumer daemon");
1711 return ret;
1712 }
1713 return pthread_join(consumer_data->thread, &status);
1714 } else {
1715 return 0;
1716 }
1717 }
1718
1719 /*
1720 * Fork and exec a consumer daemon (consumerd).
1721 *
1722 * Return pid if successful else -1.
1723 */
1724 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1725 {
1726 int ret;
1727 pid_t pid;
1728 const char *consumer_to_use;
1729 const char *verbosity;
1730 struct stat st;
1731
1732 DBG("Spawning consumerd");
1733
1734 pid = fork();
1735 if (pid == 0) {
1736 /*
1737 * Exec consumerd.
1738 */
1739 if (opt_verbose_consumer) {
1740 verbosity = "--verbose";
1741 } else {
1742 verbosity = "--quiet";
1743 }
1744 switch (consumer_data->type) {
1745 case LTTNG_CONSUMER_KERNEL:
1746 /*
1747 * Find out which consumerd to execute. We will first try the
1748 * 64-bit path, then the sessiond's installation directory, and
1749 * fallback on the 32-bit one,
1750 */
1751 DBG3("Looking for a kernel consumer at these locations:");
1752 DBG3(" 1) %s", consumerd64_bin);
1753 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1754 DBG3(" 3) %s", consumerd32_bin);
1755 if (stat(consumerd64_bin, &st) == 0) {
1756 DBG3("Found location #1");
1757 consumer_to_use = consumerd64_bin;
1758 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1759 DBG3("Found location #2");
1760 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1761 } else if (stat(consumerd32_bin, &st) == 0) {
1762 DBG3("Found location #3");
1763 consumer_to_use = consumerd32_bin;
1764 } else {
1765 DBG("Could not find any valid consumerd executable");
1766 break;
1767 }
1768 DBG("Using kernel consumer at: %s", consumer_to_use);
1769 execl(consumer_to_use,
1770 "lttng-consumerd", verbosity, "-k",
1771 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1772 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1773 NULL);
1774 break;
1775 case LTTNG_CONSUMER64_UST:
1776 {
1777 char *tmpnew = NULL;
1778
1779 if (consumerd64_libdir[0] != '\0') {
1780 char *tmp;
1781 size_t tmplen;
1782
1783 tmp = getenv("LD_LIBRARY_PATH");
1784 if (!tmp) {
1785 tmp = "";
1786 }
1787 tmplen = strlen("LD_LIBRARY_PATH=")
1788 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1789 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1790 if (!tmpnew) {
1791 ret = -ENOMEM;
1792 goto error;
1793 }
1794 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1795 strcat(tmpnew, consumerd64_libdir);
1796 if (tmp[0] != '\0') {
1797 strcat(tmpnew, ":");
1798 strcat(tmpnew, tmp);
1799 }
1800 ret = putenv(tmpnew);
1801 if (ret) {
1802 ret = -errno;
1803 goto error;
1804 }
1805 }
1806 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1807 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1808 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1809 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1810 NULL);
1811 if (consumerd64_libdir[0] != '\0') {
1812 free(tmpnew);
1813 }
1814 if (ret) {
1815 goto error;
1816 }
1817 break;
1818 }
1819 case LTTNG_CONSUMER32_UST:
1820 {
1821 char *tmpnew = NULL;
1822
1823 if (consumerd32_libdir[0] != '\0') {
1824 char *tmp;
1825 size_t tmplen;
1826
1827 tmp = getenv("LD_LIBRARY_PATH");
1828 if (!tmp) {
1829 tmp = "";
1830 }
1831 tmplen = strlen("LD_LIBRARY_PATH=")
1832 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1833 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1834 if (!tmpnew) {
1835 ret = -ENOMEM;
1836 goto error;
1837 }
1838 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1839 strcat(tmpnew, consumerd32_libdir);
1840 if (tmp[0] != '\0') {
1841 strcat(tmpnew, ":");
1842 strcat(tmpnew, tmp);
1843 }
1844 ret = putenv(tmpnew);
1845 if (ret) {
1846 ret = -errno;
1847 goto error;
1848 }
1849 }
1850 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1851 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1852 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1853 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1854 NULL);
1855 if (consumerd32_libdir[0] != '\0') {
1856 free(tmpnew);
1857 }
1858 if (ret) {
1859 goto error;
1860 }
1861 break;
1862 }
1863 default:
1864 PERROR("unknown consumer type");
1865 exit(EXIT_FAILURE);
1866 }
1867 if (errno != 0) {
1868 PERROR("kernel start consumer exec");
1869 }
1870 exit(EXIT_FAILURE);
1871 } else if (pid > 0) {
1872 ret = pid;
1873 } else {
1874 PERROR("start consumer fork");
1875 ret = -errno;
1876 }
1877 error:
1878 return ret;
1879 }
1880
1881 /*
1882 * Spawn the consumerd daemon and session daemon thread.
1883 */
1884 static int start_consumerd(struct consumer_data *consumer_data)
1885 {
1886 int ret;
1887
1888 /*
1889 * Set the listen() state on the socket since there is a possible race
1890 * between the exec() of the consumer daemon and this call if place in the
1891 * consumer thread. See bug #366 for more details.
1892 */
1893 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1894 if (ret < 0) {
1895 goto error;
1896 }
1897
1898 pthread_mutex_lock(&consumer_data->pid_mutex);
1899 if (consumer_data->pid != 0) {
1900 pthread_mutex_unlock(&consumer_data->pid_mutex);
1901 goto end;
1902 }
1903
1904 ret = spawn_consumerd(consumer_data);
1905 if (ret < 0) {
1906 ERR("Spawning consumerd failed");
1907 pthread_mutex_unlock(&consumer_data->pid_mutex);
1908 goto error;
1909 }
1910
1911 /* Setting up the consumer_data pid */
1912 consumer_data->pid = ret;
1913 DBG2("Consumer pid %d", consumer_data->pid);
1914 pthread_mutex_unlock(&consumer_data->pid_mutex);
1915
1916 DBG2("Spawning consumer control thread");
1917 ret = spawn_consumer_thread(consumer_data);
1918 if (ret < 0) {
1919 ERR("Fatal error spawning consumer control thread");
1920 goto error;
1921 }
1922
1923 end:
1924 return 0;
1925
1926 error:
1927 /* Cleanup already created socket on error. */
1928 if (consumer_data->err_sock >= 0) {
1929 int err;
1930
1931 err = close(consumer_data->err_sock);
1932 if (err < 0) {
1933 PERROR("close consumer data error socket");
1934 }
1935 }
1936 return ret;
1937 }
1938
1939 /*
1940 * Compute health status of each consumer. If one of them is zero (bad
1941 * state), we return 0.
1942 */
1943 static int check_consumer_health(void)
1944 {
1945 int ret;
1946
1947 ret = health_check_state(HEALTH_TYPE_CONSUMER);
1948
1949 DBG3("Health consumer check %d", ret);
1950
1951 return ret;
1952 }
1953
1954 /*
1955 * Setup necessary data for kernel tracer action.
1956 */
1957 static int init_kernel_tracer(void)
1958 {
1959 int ret;
1960
1961 /* Modprobe lttng kernel modules */
1962 ret = modprobe_lttng_control();
1963 if (ret < 0) {
1964 goto error;
1965 }
1966
1967 /* Open debugfs lttng */
1968 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1969 if (kernel_tracer_fd < 0) {
1970 DBG("Failed to open %s", module_proc_lttng);
1971 ret = -1;
1972 goto error_open;
1973 }
1974
1975 /* Validate kernel version */
1976 ret = kernel_validate_version(kernel_tracer_fd);
1977 if (ret < 0) {
1978 goto error_version;
1979 }
1980
1981 ret = modprobe_lttng_data();
1982 if (ret < 0) {
1983 goto error_modules;
1984 }
1985
1986 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1987 return 0;
1988
1989 error_version:
1990 modprobe_remove_lttng_control();
1991 ret = close(kernel_tracer_fd);
1992 if (ret) {
1993 PERROR("close");
1994 }
1995 kernel_tracer_fd = -1;
1996 return LTTNG_ERR_KERN_VERSION;
1997
1998 error_modules:
1999 ret = close(kernel_tracer_fd);
2000 if (ret) {
2001 PERROR("close");
2002 }
2003
2004 error_open:
2005 modprobe_remove_lttng_control();
2006
2007 error:
2008 WARN("No kernel tracer available");
2009 kernel_tracer_fd = -1;
2010 if (!is_root) {
2011 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2012 } else {
2013 return LTTNG_ERR_KERN_NA;
2014 }
2015 }
2016
2017
2018 /*
2019 * Copy consumer output from the tracing session to the domain session. The
2020 * function also applies the right modification on a per domain basis for the
2021 * trace files destination directory.
2022 */
2023 static int copy_session_consumer(int domain, struct ltt_session *session)
2024 {
2025 int ret;
2026 const char *dir_name;
2027 struct consumer_output *consumer;
2028
2029 assert(session);
2030 assert(session->consumer);
2031
2032 switch (domain) {
2033 case LTTNG_DOMAIN_KERNEL:
2034 DBG3("Copying tracing session consumer output in kernel session");
2035 /*
2036 * XXX: We should audit the session creation and what this function
2037 * does "extra" in order to avoid a destroy since this function is used
2038 * in the domain session creation (kernel and ust) only. Same for UST
2039 * domain.
2040 */
2041 if (session->kernel_session->consumer) {
2042 consumer_destroy_output(session->kernel_session->consumer);
2043 }
2044 session->kernel_session->consumer =
2045 consumer_copy_output(session->consumer);
2046 /* Ease our life a bit for the next part */
2047 consumer = session->kernel_session->consumer;
2048 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2049 break;
2050 case LTTNG_DOMAIN_UST:
2051 DBG3("Copying tracing session consumer output in UST session");
2052 if (session->ust_session->consumer) {
2053 consumer_destroy_output(session->ust_session->consumer);
2054 }
2055 session->ust_session->consumer =
2056 consumer_copy_output(session->consumer);
2057 /* Ease our life a bit for the next part */
2058 consumer = session->ust_session->consumer;
2059 dir_name = DEFAULT_UST_TRACE_DIR;
2060 break;
2061 default:
2062 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2063 goto error;
2064 }
2065
2066 /* Append correct directory to subdir */
2067 strncat(consumer->subdir, dir_name,
2068 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2069 DBG3("Copy session consumer subdir %s", consumer->subdir);
2070
2071 ret = LTTNG_OK;
2072
2073 error:
2074 return ret;
2075 }
2076
2077 /*
2078 * Create an UST session and add it to the session ust list.
2079 */
2080 static int create_ust_session(struct ltt_session *session,
2081 struct lttng_domain *domain)
2082 {
2083 int ret;
2084 struct ltt_ust_session *lus = NULL;
2085
2086 assert(session);
2087 assert(domain);
2088 assert(session->consumer);
2089
2090 switch (domain->type) {
2091 case LTTNG_DOMAIN_UST:
2092 break;
2093 default:
2094 ERR("Unknown UST domain on create session %d", domain->type);
2095 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2096 goto error;
2097 }
2098
2099 DBG("Creating UST session");
2100
2101 lus = trace_ust_create_session(session->path, session->id, domain);
2102 if (lus == NULL) {
2103 ret = LTTNG_ERR_UST_SESS_FAIL;
2104 goto error;
2105 }
2106
2107 lus->uid = session->uid;
2108 lus->gid = session->gid;
2109 session->ust_session = lus;
2110
2111 /* Copy session output to the newly created UST session */
2112 ret = copy_session_consumer(domain->type, session);
2113 if (ret != LTTNG_OK) {
2114 goto error;
2115 }
2116
2117 return LTTNG_OK;
2118
2119 error:
2120 free(lus);
2121 session->ust_session = NULL;
2122 return ret;
2123 }
2124
2125 /*
2126 * Create a kernel tracer session then create the default channel.
2127 */
2128 static int create_kernel_session(struct ltt_session *session)
2129 {
2130 int ret;
2131
2132 DBG("Creating kernel session");
2133
2134 ret = kernel_create_session(session, kernel_tracer_fd);
2135 if (ret < 0) {
2136 ret = LTTNG_ERR_KERN_SESS_FAIL;
2137 goto error;
2138 }
2139
2140 /* Code flow safety */
2141 assert(session->kernel_session);
2142
2143 /* Copy session output to the newly created Kernel session */
2144 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2145 if (ret != LTTNG_OK) {
2146 goto error;
2147 }
2148
2149 /* Create directory(ies) on local filesystem. */
2150 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2151 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2152 ret = run_as_mkdir_recursive(
2153 session->kernel_session->consumer->dst.trace_path,
2154 S_IRWXU | S_IRWXG, session->uid, session->gid);
2155 if (ret < 0) {
2156 if (ret != -EEXIST) {
2157 ERR("Trace directory creation error");
2158 goto error;
2159 }
2160 }
2161 }
2162
2163 session->kernel_session->uid = session->uid;
2164 session->kernel_session->gid = session->gid;
2165
2166 return LTTNG_OK;
2167
2168 error:
2169 trace_kernel_destroy_session(session->kernel_session);
2170 session->kernel_session = NULL;
2171 return ret;
2172 }
2173
2174 /*
2175 * Count number of session permitted by uid/gid.
2176 */
2177 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2178 {
2179 unsigned int i = 0;
2180 struct ltt_session *session;
2181
2182 DBG("Counting number of available session for UID %d GID %d",
2183 uid, gid);
2184 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2185 /*
2186 * Only list the sessions the user can control.
2187 */
2188 if (!session_access_ok(session, uid, gid)) {
2189 continue;
2190 }
2191 i++;
2192 }
2193 return i;
2194 }
2195
2196 /*
2197 * Process the command requested by the lttng client within the command
2198 * context structure. This function make sure that the return structure (llm)
2199 * is set and ready for transmission before returning.
2200 *
2201 * Return any error encountered or 0 for success.
2202 *
2203 * "sock" is only used for special-case var. len data.
2204 */
2205 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2206 int *sock_error)
2207 {
2208 int ret = LTTNG_OK;
2209 int need_tracing_session = 1;
2210 int need_domain;
2211
2212 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2213
2214 *sock_error = 0;
2215
2216 switch (cmd_ctx->lsm->cmd_type) {
2217 case LTTNG_CREATE_SESSION:
2218 case LTTNG_DESTROY_SESSION:
2219 case LTTNG_LIST_SESSIONS:
2220 case LTTNG_LIST_DOMAINS:
2221 case LTTNG_START_TRACE:
2222 case LTTNG_STOP_TRACE:
2223 case LTTNG_DATA_PENDING:
2224 need_domain = 0;
2225 break;
2226 default:
2227 need_domain = 1;
2228 }
2229
2230 if (opt_no_kernel && need_domain
2231 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2232 if (!is_root) {
2233 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2234 } else {
2235 ret = LTTNG_ERR_KERN_NA;
2236 }
2237 goto error;
2238 }
2239
2240 /* Deny register consumer if we already have a spawned consumer. */
2241 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2242 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2243 if (kconsumer_data.pid > 0) {
2244 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2245 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2246 goto error;
2247 }
2248 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2249 }
2250
2251 /*
2252 * Check for command that don't needs to allocate a returned payload. We do
2253 * this here so we don't have to make the call for no payload at each
2254 * command.
2255 */
2256 switch(cmd_ctx->lsm->cmd_type) {
2257 case LTTNG_LIST_SESSIONS:
2258 case LTTNG_LIST_TRACEPOINTS:
2259 case LTTNG_LIST_TRACEPOINT_FIELDS:
2260 case LTTNG_LIST_DOMAINS:
2261 case LTTNG_LIST_CHANNELS:
2262 case LTTNG_LIST_EVENTS:
2263 break;
2264 default:
2265 /* Setup lttng message with no payload */
2266 ret = setup_lttng_msg(cmd_ctx, 0);
2267 if (ret < 0) {
2268 /* This label does not try to unlock the session */
2269 goto init_setup_error;
2270 }
2271 }
2272
2273 /* Commands that DO NOT need a session. */
2274 switch (cmd_ctx->lsm->cmd_type) {
2275 case LTTNG_CREATE_SESSION:
2276 case LTTNG_CALIBRATE:
2277 case LTTNG_LIST_SESSIONS:
2278 case LTTNG_LIST_TRACEPOINTS:
2279 case LTTNG_LIST_TRACEPOINT_FIELDS:
2280 need_tracing_session = 0;
2281 break;
2282 default:
2283 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2284 /*
2285 * We keep the session list lock across _all_ commands
2286 * for now, because the per-session lock does not
2287 * handle teardown properly.
2288 */
2289 session_lock_list();
2290 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2291 if (cmd_ctx->session == NULL) {
2292 if (cmd_ctx->lsm->session.name != NULL) {
2293 ret = LTTNG_ERR_SESS_NOT_FOUND;
2294 } else {
2295 /* If no session name specified */
2296 ret = LTTNG_ERR_SELECT_SESS;
2297 }
2298 goto error;
2299 } else {
2300 /* Acquire lock for the session */
2301 session_lock(cmd_ctx->session);
2302 }
2303 break;
2304 }
2305
2306 if (!need_domain) {
2307 goto skip_domain;
2308 }
2309
2310 /*
2311 * Check domain type for specific "pre-action".
2312 */
2313 switch (cmd_ctx->lsm->domain.type) {
2314 case LTTNG_DOMAIN_KERNEL:
2315 if (!is_root) {
2316 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2317 goto error;
2318 }
2319
2320 /* Kernel tracer check */
2321 if (kernel_tracer_fd == -1) {
2322 /* Basically, load kernel tracer modules */
2323 ret = init_kernel_tracer();
2324 if (ret != 0) {
2325 goto error;
2326 }
2327 }
2328
2329 /* Consumer is in an ERROR state. Report back to client */
2330 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2331 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2332 goto error;
2333 }
2334
2335 /* Need a session for kernel command */
2336 if (need_tracing_session) {
2337 if (cmd_ctx->session->kernel_session == NULL) {
2338 ret = create_kernel_session(cmd_ctx->session);
2339 if (ret < 0) {
2340 ret = LTTNG_ERR_KERN_SESS_FAIL;
2341 goto error;
2342 }
2343 }
2344
2345 /* Start the kernel consumer daemon */
2346 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2347 if (kconsumer_data.pid == 0 &&
2348 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2349 cmd_ctx->session->start_consumer) {
2350 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2351 ret = start_consumerd(&kconsumer_data);
2352 if (ret < 0) {
2353 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2354 goto error;
2355 }
2356 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2357 } else {
2358 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2359 }
2360
2361 /*
2362 * The consumer was just spawned so we need to add the socket to
2363 * the consumer output of the session if exist.
2364 */
2365 ret = consumer_create_socket(&kconsumer_data,
2366 cmd_ctx->session->kernel_session->consumer);
2367 if (ret < 0) {
2368 goto error;
2369 }
2370 }
2371
2372 break;
2373 case LTTNG_DOMAIN_UST:
2374 {
2375 /* Consumer is in an ERROR state. Report back to client */
2376 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2377 ret = LTTNG_ERR_NO_USTCONSUMERD;
2378 goto error;
2379 }
2380
2381 if (need_tracing_session) {
2382 /* Create UST session if none exist. */
2383 if (cmd_ctx->session->ust_session == NULL) {
2384 ret = create_ust_session(cmd_ctx->session,
2385 &cmd_ctx->lsm->domain);
2386 if (ret != LTTNG_OK) {
2387 goto error;
2388 }
2389 }
2390
2391 /* Start the UST consumer daemons */
2392 /* 64-bit */
2393 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2394 if (consumerd64_bin[0] != '\0' &&
2395 ustconsumer64_data.pid == 0 &&
2396 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2397 cmd_ctx->session->start_consumer) {
2398 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2399 ret = start_consumerd(&ustconsumer64_data);
2400 if (ret < 0) {
2401 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2402 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2403 goto error;
2404 }
2405
2406 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2407 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2408 } else {
2409 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2410 }
2411
2412 /*
2413 * Setup socket for consumer 64 bit. No need for atomic access
2414 * since it was set above and can ONLY be set in this thread.
2415 */
2416 ret = consumer_create_socket(&ustconsumer64_data,
2417 cmd_ctx->session->ust_session->consumer);
2418 if (ret < 0) {
2419 goto error;
2420 }
2421
2422 /* 32-bit */
2423 if (consumerd32_bin[0] != '\0' &&
2424 ustconsumer32_data.pid == 0 &&
2425 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2426 cmd_ctx->session->start_consumer) {
2427 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2428 ret = start_consumerd(&ustconsumer32_data);
2429 if (ret < 0) {
2430 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2431 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2432 goto error;
2433 }
2434
2435 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2436 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2437 } else {
2438 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2439 }
2440
2441 /*
2442 * Setup socket for consumer 64 bit. No need for atomic access
2443 * since it was set above and can ONLY be set in this thread.
2444 */
2445 ret = consumer_create_socket(&ustconsumer32_data,
2446 cmd_ctx->session->ust_session->consumer);
2447 if (ret < 0) {
2448 goto error;
2449 }
2450 }
2451 break;
2452 }
2453 default:
2454 break;
2455 }
2456 skip_domain:
2457
2458 /* Validate consumer daemon state when start/stop trace command */
2459 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2460 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2461 switch (cmd_ctx->lsm->domain.type) {
2462 case LTTNG_DOMAIN_UST:
2463 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2464 ret = LTTNG_ERR_NO_USTCONSUMERD;
2465 goto error;
2466 }
2467 break;
2468 case LTTNG_DOMAIN_KERNEL:
2469 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2470 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2471 goto error;
2472 }
2473 break;
2474 }
2475 }
2476
2477 /*
2478 * Check that the UID or GID match that of the tracing session.
2479 * The root user can interact with all sessions.
2480 */
2481 if (need_tracing_session) {
2482 if (!session_access_ok(cmd_ctx->session,
2483 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2484 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2485 ret = LTTNG_ERR_EPERM;
2486 goto error;
2487 }
2488 }
2489
2490 /* Process by command type */
2491 switch (cmd_ctx->lsm->cmd_type) {
2492 case LTTNG_ADD_CONTEXT:
2493 {
2494 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2495 cmd_ctx->lsm->u.context.channel_name,
2496 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2497 break;
2498 }
2499 case LTTNG_DISABLE_CHANNEL:
2500 {
2501 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2502 cmd_ctx->lsm->u.disable.channel_name);
2503 break;
2504 }
2505 case LTTNG_DISABLE_EVENT:
2506 {
2507 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2508 cmd_ctx->lsm->u.disable.channel_name,
2509 cmd_ctx->lsm->u.disable.name);
2510 break;
2511 }
2512 case LTTNG_DISABLE_ALL_EVENT:
2513 {
2514 DBG("Disabling all events");
2515
2516 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2517 cmd_ctx->lsm->u.disable.channel_name);
2518 break;
2519 }
2520 case LTTNG_DISABLE_CONSUMER:
2521 {
2522 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2523 break;
2524 }
2525 case LTTNG_ENABLE_CHANNEL:
2526 {
2527 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2528 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2529 break;
2530 }
2531 case LTTNG_ENABLE_CONSUMER:
2532 {
2533 /*
2534 * XXX: 0 means that this URI should be applied on the session. Should
2535 * be a DOMAIN enuam.
2536 */
2537 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2538 if (ret != LTTNG_OK) {
2539 goto error;
2540 }
2541
2542 if (cmd_ctx->lsm->domain.type == 0) {
2543 /* Add the URI for the UST session if a consumer is present. */
2544 if (cmd_ctx->session->ust_session &&
2545 cmd_ctx->session->ust_session->consumer) {
2546 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2547 } else if (cmd_ctx->session->kernel_session &&
2548 cmd_ctx->session->kernel_session->consumer) {
2549 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2550 cmd_ctx->session);
2551 }
2552 }
2553 break;
2554 }
2555 case LTTNG_ENABLE_EVENT:
2556 {
2557 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2558 cmd_ctx->lsm->u.enable.channel_name,
2559 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2560 break;
2561 }
2562 case LTTNG_ENABLE_ALL_EVENT:
2563 {
2564 DBG("Enabling all events");
2565
2566 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2567 cmd_ctx->lsm->u.enable.channel_name,
2568 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2569 break;
2570 }
2571 case LTTNG_LIST_TRACEPOINTS:
2572 {
2573 struct lttng_event *events;
2574 ssize_t nb_events;
2575
2576 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2577 if (nb_events < 0) {
2578 /* Return value is a negative lttng_error_code. */
2579 ret = -nb_events;
2580 goto error;
2581 }
2582
2583 /*
2584 * Setup lttng message with payload size set to the event list size in
2585 * bytes and then copy list into the llm payload.
2586 */
2587 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2588 if (ret < 0) {
2589 free(events);
2590 goto setup_error;
2591 }
2592
2593 /* Copy event list into message payload */
2594 memcpy(cmd_ctx->llm->payload, events,
2595 sizeof(struct lttng_event) * nb_events);
2596
2597 free(events);
2598
2599 ret = LTTNG_OK;
2600 break;
2601 }
2602 case LTTNG_LIST_TRACEPOINT_FIELDS:
2603 {
2604 struct lttng_event_field *fields;
2605 ssize_t nb_fields;
2606
2607 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2608 &fields);
2609 if (nb_fields < 0) {
2610 /* Return value is a negative lttng_error_code. */
2611 ret = -nb_fields;
2612 goto error;
2613 }
2614
2615 /*
2616 * Setup lttng message with payload size set to the event list size in
2617 * bytes and then copy list into the llm payload.
2618 */
2619 ret = setup_lttng_msg(cmd_ctx,
2620 sizeof(struct lttng_event_field) * nb_fields);
2621 if (ret < 0) {
2622 free(fields);
2623 goto setup_error;
2624 }
2625
2626 /* Copy event list into message payload */
2627 memcpy(cmd_ctx->llm->payload, fields,
2628 sizeof(struct lttng_event_field) * nb_fields);
2629
2630 free(fields);
2631
2632 ret = LTTNG_OK;
2633 break;
2634 }
2635 case LTTNG_SET_CONSUMER_URI:
2636 {
2637 size_t nb_uri, len;
2638 struct lttng_uri *uris;
2639
2640 nb_uri = cmd_ctx->lsm->u.uri.size;
2641 len = nb_uri * sizeof(struct lttng_uri);
2642
2643 if (nb_uri == 0) {
2644 ret = LTTNG_ERR_INVALID;
2645 goto error;
2646 }
2647
2648 uris = zmalloc(len);
2649 if (uris == NULL) {
2650 ret = LTTNG_ERR_FATAL;
2651 goto error;
2652 }
2653
2654 /* Receive variable len data */
2655 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2656 ret = lttcomm_recv_unix_sock(sock, uris, len);
2657 if (ret <= 0) {
2658 DBG("No URIs received from client... continuing");
2659 *sock_error = 1;
2660 ret = LTTNG_ERR_SESSION_FAIL;
2661 free(uris);
2662 goto error;
2663 }
2664
2665 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2666 nb_uri, uris);
2667 if (ret != LTTNG_OK) {
2668 free(uris);
2669 goto error;
2670 }
2671
2672 /*
2673 * XXX: 0 means that this URI should be applied on the session. Should
2674 * be a DOMAIN enuam.
2675 */
2676 if (cmd_ctx->lsm->domain.type == 0) {
2677 /* Add the URI for the UST session if a consumer is present. */
2678 if (cmd_ctx->session->ust_session &&
2679 cmd_ctx->session->ust_session->consumer) {
2680 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2681 nb_uri, uris);
2682 } else if (cmd_ctx->session->kernel_session &&
2683 cmd_ctx->session->kernel_session->consumer) {
2684 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2685 cmd_ctx->session, nb_uri, uris);
2686 }
2687 }
2688
2689 free(uris);
2690
2691 break;
2692 }
2693 case LTTNG_START_TRACE:
2694 {
2695 ret = cmd_start_trace(cmd_ctx->session);
2696 break;
2697 }
2698 case LTTNG_STOP_TRACE:
2699 {
2700 ret = cmd_stop_trace(cmd_ctx->session);
2701 break;
2702 }
2703 case LTTNG_CREATE_SESSION:
2704 {
2705 size_t nb_uri, len;
2706 struct lttng_uri *uris = NULL;
2707
2708 nb_uri = cmd_ctx->lsm->u.uri.size;
2709 len = nb_uri * sizeof(struct lttng_uri);
2710
2711 if (nb_uri > 0) {
2712 uris = zmalloc(len);
2713 if (uris == NULL) {
2714 ret = LTTNG_ERR_FATAL;
2715 goto error;
2716 }
2717
2718 /* Receive variable len data */
2719 DBG("Waiting for %zu URIs from client ...", nb_uri);
2720 ret = lttcomm_recv_unix_sock(sock, uris, len);
2721 if (ret <= 0) {
2722 DBG("No URIs received from client... continuing");
2723 *sock_error = 1;
2724 ret = LTTNG_ERR_SESSION_FAIL;
2725 free(uris);
2726 goto error;
2727 }
2728
2729 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2730 DBG("Creating session with ONE network URI is a bad call");
2731 ret = LTTNG_ERR_SESSION_FAIL;
2732 free(uris);
2733 goto error;
2734 }
2735 }
2736
2737 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2738 &cmd_ctx->creds);
2739
2740 free(uris);
2741
2742 break;
2743 }
2744 case LTTNG_DESTROY_SESSION:
2745 {
2746 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2747
2748 /* Set session to NULL so we do not unlock it after free. */
2749 cmd_ctx->session = NULL;
2750 break;
2751 }
2752 case LTTNG_LIST_DOMAINS:
2753 {
2754 ssize_t nb_dom;
2755 struct lttng_domain *domains;
2756
2757 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2758 if (nb_dom < 0) {
2759 /* Return value is a negative lttng_error_code. */
2760 ret = -nb_dom;
2761 goto error;
2762 }
2763
2764 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2765 if (ret < 0) {
2766 goto setup_error;
2767 }
2768
2769 /* Copy event list into message payload */
2770 memcpy(cmd_ctx->llm->payload, domains,
2771 nb_dom * sizeof(struct lttng_domain));
2772
2773 free(domains);
2774
2775 ret = LTTNG_OK;
2776 break;
2777 }
2778 case LTTNG_LIST_CHANNELS:
2779 {
2780 int nb_chan;
2781 struct lttng_channel *channels;
2782
2783 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2784 cmd_ctx->session, &channels);
2785 if (nb_chan < 0) {
2786 /* Return value is a negative lttng_error_code. */
2787 ret = -nb_chan;
2788 goto error;
2789 }
2790
2791 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2792 if (ret < 0) {
2793 goto setup_error;
2794 }
2795
2796 /* Copy event list into message payload */
2797 memcpy(cmd_ctx->llm->payload, channels,
2798 nb_chan * sizeof(struct lttng_channel));
2799
2800 free(channels);
2801
2802 ret = LTTNG_OK;
2803 break;
2804 }
2805 case LTTNG_LIST_EVENTS:
2806 {
2807 ssize_t nb_event;
2808 struct lttng_event *events = NULL;
2809
2810 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2811 cmd_ctx->lsm->u.list.channel_name, &events);
2812 if (nb_event < 0) {
2813 /* Return value is a negative lttng_error_code. */
2814 ret = -nb_event;
2815 goto error;
2816 }
2817
2818 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2819 if (ret < 0) {
2820 goto setup_error;
2821 }
2822
2823 /* Copy event list into message payload */
2824 memcpy(cmd_ctx->llm->payload, events,
2825 nb_event * sizeof(struct lttng_event));
2826
2827 free(events);
2828
2829 ret = LTTNG_OK;
2830 break;
2831 }
2832 case LTTNG_LIST_SESSIONS:
2833 {
2834 unsigned int nr_sessions;
2835
2836 session_lock_list();
2837 nr_sessions = lttng_sessions_count(
2838 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2839 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2840
2841 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2842 if (ret < 0) {
2843 session_unlock_list();
2844 goto setup_error;
2845 }
2846
2847 /* Filled the session array */
2848 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2849 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2850 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2851
2852 session_unlock_list();
2853
2854 ret = LTTNG_OK;
2855 break;
2856 }
2857 case LTTNG_CALIBRATE:
2858 {
2859 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2860 &cmd_ctx->lsm->u.calibrate);
2861 break;
2862 }
2863 case LTTNG_REGISTER_CONSUMER:
2864 {
2865 struct consumer_data *cdata;
2866
2867 switch (cmd_ctx->lsm->domain.type) {
2868 case LTTNG_DOMAIN_KERNEL:
2869 cdata = &kconsumer_data;
2870 break;
2871 default:
2872 ret = LTTNG_ERR_UND;
2873 goto error;
2874 }
2875
2876 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2877 cmd_ctx->lsm->u.reg.path, cdata);
2878 break;
2879 }
2880 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2881 {
2882 struct lttng_filter_bytecode *bytecode;
2883
2884 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2885 ret = LTTNG_ERR_FILTER_INVAL;
2886 goto error;
2887 }
2888 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2889 ret = LTTNG_ERR_FILTER_INVAL;
2890 goto error;
2891 }
2892 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2893 if (!bytecode) {
2894 ret = LTTNG_ERR_FILTER_NOMEM;
2895 goto error;
2896 }
2897 /* Receive var. len. data */
2898 DBG("Receiving var len data from client ...");
2899 ret = lttcomm_recv_unix_sock(sock, bytecode,
2900 cmd_ctx->lsm->u.enable.bytecode_len);
2901 if (ret <= 0) {
2902 DBG("Nothing recv() from client var len data... continuing");
2903 *sock_error = 1;
2904 ret = LTTNG_ERR_FILTER_INVAL;
2905 goto error;
2906 }
2907
2908 if (bytecode->len + sizeof(*bytecode)
2909 != cmd_ctx->lsm->u.enable.bytecode_len) {
2910 free(bytecode);
2911 ret = LTTNG_ERR_FILTER_INVAL;
2912 goto error;
2913 }
2914
2915 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2916 cmd_ctx->lsm->u.enable.channel_name,
2917 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2918 break;
2919 }
2920 case LTTNG_DATA_PENDING:
2921 {
2922 ret = cmd_data_pending(cmd_ctx->session);
2923 break;
2924 }
2925 default:
2926 ret = LTTNG_ERR_UND;
2927 break;
2928 }
2929
2930 error:
2931 if (cmd_ctx->llm == NULL) {
2932 DBG("Missing llm structure. Allocating one.");
2933 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2934 goto setup_error;
2935 }
2936 }
2937 /* Set return code */
2938 cmd_ctx->llm->ret_code = ret;
2939 setup_error:
2940 if (cmd_ctx->session) {
2941 session_unlock(cmd_ctx->session);
2942 }
2943 if (need_tracing_session) {
2944 session_unlock_list();
2945 }
2946 init_setup_error:
2947 return ret;
2948 }
2949
2950 /*
2951 * Thread managing health check socket.
2952 */
2953 static void *thread_manage_health(void *data)
2954 {
2955 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2956 uint32_t revents, nb_fd;
2957 struct lttng_poll_event events;
2958 struct lttcomm_health_msg msg;
2959 struct lttcomm_health_data reply;
2960
2961 DBG("[thread] Manage health check started");
2962
2963 rcu_register_thread();
2964
2965 /* Create unix socket */
2966 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2967 if (sock < 0) {
2968 ERR("Unable to create health check Unix socket");
2969 ret = -1;
2970 goto error;
2971 }
2972
2973 /*
2974 * Set the CLOEXEC flag. Return code is useless because either way, the
2975 * show must go on.
2976 */
2977 (void) utils_set_fd_cloexec(sock);
2978
2979 ret = lttcomm_listen_unix_sock(sock);
2980 if (ret < 0) {
2981 goto error;
2982 }
2983
2984 /*
2985 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2986 * more will be added to this poll set.
2987 */
2988 ret = create_thread_poll_set(&events, 2);
2989 if (ret < 0) {
2990 goto error;
2991 }
2992
2993 /* Add the application registration socket */
2994 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2995 if (ret < 0) {
2996 goto error;
2997 }
2998
2999 while (1) {
3000 DBG("Health check ready");
3001
3002 /* Inifinite blocking call, waiting for transmission */
3003 restart:
3004 ret = lttng_poll_wait(&events, -1);
3005 if (ret < 0) {
3006 /*
3007 * Restart interrupted system call.
3008 */
3009 if (errno == EINTR) {
3010 goto restart;
3011 }
3012 goto error;
3013 }
3014
3015 nb_fd = ret;
3016
3017 for (i = 0; i < nb_fd; i++) {
3018 /* Fetch once the poll data */
3019 revents = LTTNG_POLL_GETEV(&events, i);
3020 pollfd = LTTNG_POLL_GETFD(&events, i);
3021
3022 /* Thread quit pipe has been closed. Killing thread. */
3023 ret = check_thread_quit_pipe(pollfd, revents);
3024 if (ret) {
3025 err = 0;
3026 goto exit;
3027 }
3028
3029 /* Event on the registration socket */
3030 if (pollfd == sock) {
3031 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3032 ERR("Health socket poll error");
3033 goto error;
3034 }
3035 }
3036 }
3037
3038 new_sock = lttcomm_accept_unix_sock(sock);
3039 if (new_sock < 0) {
3040 goto error;
3041 }
3042
3043 /*
3044 * Set the CLOEXEC flag. Return code is useless because either way, the
3045 * show must go on.
3046 */
3047 (void) utils_set_fd_cloexec(new_sock);
3048
3049 DBG("Receiving data from client for health...");
3050 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3051 if (ret <= 0) {
3052 DBG("Nothing recv() from client... continuing");
3053 ret = close(new_sock);
3054 if (ret) {
3055 PERROR("close");
3056 }
3057 new_sock = -1;
3058 continue;
3059 }
3060
3061 rcu_thread_online();
3062
3063 switch (msg.component) {
3064 case LTTNG_HEALTH_CMD:
3065 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3066 break;
3067 case LTTNG_HEALTH_APP_MANAGE:
3068 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3069 break;
3070 case LTTNG_HEALTH_APP_REG:
3071 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3072 break;
3073 case LTTNG_HEALTH_KERNEL:
3074 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3075 break;
3076 case LTTNG_HEALTH_CONSUMER:
3077 reply.ret_code = check_consumer_health();
3078 break;
3079 case LTTNG_HEALTH_ALL:
3080 reply.ret_code =
3081 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3082 health_check_state(HEALTH_TYPE_APP_REG) &&
3083 health_check_state(HEALTH_TYPE_CMD) &&
3084 health_check_state(HEALTH_TYPE_KERNEL) &&
3085 check_consumer_health();
3086 break;
3087 default:
3088 reply.ret_code = LTTNG_ERR_UND;
3089 break;
3090 }
3091
3092 /*
3093 * Flip ret value since 0 is a success and 1 indicates a bad health for
3094 * the client where in the sessiond it is the opposite. Again, this is
3095 * just to make things easier for us poor developer which enjoy a lot
3096 * lazyness.
3097 */
3098 if (reply.ret_code == 0 || reply.ret_code == 1) {
3099 reply.ret_code = !reply.ret_code;
3100 }
3101
3102 DBG2("Health check return value %d", reply.ret_code);
3103
3104 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3105 if (ret < 0) {
3106 ERR("Failed to send health data back to client");
3107 }
3108
3109 /* End of transmission */
3110 ret = close(new_sock);
3111 if (ret) {
3112 PERROR("close");
3113 }
3114 new_sock = -1;
3115 }
3116
3117 exit:
3118 error:
3119 if (err) {
3120 ERR("Health error occurred in %s", __func__);
3121 }
3122 DBG("Health check thread dying");
3123 unlink(health_unix_sock_path);
3124 if (sock >= 0) {
3125 ret = close(sock);
3126 if (ret) {
3127 PERROR("close");
3128 }
3129 }
3130 if (new_sock >= 0) {
3131 ret = close(new_sock);
3132 if (ret) {
3133 PERROR("close");
3134 }
3135 }
3136
3137 lttng_poll_clean(&events);
3138
3139 rcu_unregister_thread();
3140 return NULL;
3141 }
3142
3143 /*
3144 * This thread manage all clients request using the unix client socket for
3145 * communication.
3146 */
3147 static void *thread_manage_clients(void *data)
3148 {
3149 int sock = -1, ret, i, pollfd, err = -1;
3150 int sock_error;
3151 uint32_t revents, nb_fd;
3152 struct command_ctx *cmd_ctx = NULL;
3153 struct lttng_poll_event events;
3154
3155 DBG("[thread] Manage client started");
3156
3157 rcu_register_thread();
3158
3159 health_register(HEALTH_TYPE_CMD);
3160
3161 if (testpoint(thread_manage_clients)) {
3162 goto error_testpoint;
3163 }
3164
3165 health_code_update(&health_thread_cmd);
3166
3167 ret = lttcomm_listen_unix_sock(client_sock);
3168 if (ret < 0) {
3169 goto error_listen;
3170 }
3171
3172 /*
3173 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3174 * more will be added to this poll set.
3175 */
3176 ret = create_thread_poll_set(&events, 2);
3177 if (ret < 0) {
3178 goto error_create_poll;
3179 }
3180
3181 /* Add the application registration socket */
3182 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3183 if (ret < 0) {
3184 goto error;
3185 }
3186
3187 /*
3188 * Notify parent pid that we are ready to accept command for client side.
3189 */
3190 if (opt_sig_parent) {
3191 kill(ppid, SIGUSR1);
3192 }
3193
3194 if (testpoint(thread_manage_clients_before_loop)) {
3195 goto error;
3196 }
3197
3198 health_code_update(&health_thread_cmd);
3199
3200 while (1) {
3201 DBG("Accepting client command ...");
3202
3203 /* Inifinite blocking call, waiting for transmission */
3204 restart:
3205 health_poll_update(&health_thread_cmd);
3206 ret = lttng_poll_wait(&events, -1);
3207 health_poll_update(&health_thread_cmd);
3208 if (ret < 0) {
3209 /*
3210 * Restart interrupted system call.
3211 */
3212 if (errno == EINTR) {
3213 goto restart;
3214 }
3215 goto error;
3216 }
3217
3218 nb_fd = ret;
3219
3220 for (i = 0; i < nb_fd; i++) {
3221 /* Fetch once the poll data */
3222 revents = LTTNG_POLL_GETEV(&events, i);
3223 pollfd = LTTNG_POLL_GETFD(&events, i);
3224
3225 health_code_update(&health_thread_cmd);
3226
3227 /* Thread quit pipe has been closed. Killing thread. */
3228 ret = check_thread_quit_pipe(pollfd, revents);
3229 if (ret) {
3230 err = 0;
3231 goto exit;
3232 }
3233
3234 /* Event on the registration socket */
3235 if (pollfd == client_sock) {
3236 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3237 ERR("Client socket poll error");
3238 goto error;
3239 }
3240 }
3241 }
3242
3243 DBG("Wait for client response");
3244
3245 health_code_update(&health_thread_cmd);
3246
3247 sock = lttcomm_accept_unix_sock(client_sock);
3248 if (sock < 0) {
3249 goto error;
3250 }
3251
3252 /*
3253 * Set the CLOEXEC flag. Return code is useless because either way, the
3254 * show must go on.
3255 */
3256 (void) utils_set_fd_cloexec(sock);
3257
3258 /* Set socket option for credentials retrieval */
3259 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3260 if (ret < 0) {
3261 goto error;
3262 }
3263
3264 /* Allocate context command to process the client request */
3265 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3266 if (cmd_ctx == NULL) {
3267 PERROR("zmalloc cmd_ctx");
3268 goto error;
3269 }
3270
3271 /* Allocate data buffer for reception */
3272 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3273 if (cmd_ctx->lsm == NULL) {
3274 PERROR("zmalloc cmd_ctx->lsm");
3275 goto error;
3276 }
3277
3278 cmd_ctx->llm = NULL;
3279 cmd_ctx->session = NULL;
3280
3281 health_code_update(&health_thread_cmd);
3282
3283 /*
3284 * Data is received from the lttng client. The struct
3285 * lttcomm_session_msg (lsm) contains the command and data request of
3286 * the client.
3287 */
3288 DBG("Receiving data from client ...");
3289 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3290 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3291 if (ret <= 0) {
3292 DBG("Nothing recv() from client... continuing");
3293 ret = close(sock);
3294 if (ret) {
3295 PERROR("close");
3296 }
3297 sock = -1;
3298 clean_command_ctx(&cmd_ctx);
3299 continue;
3300 }
3301
3302 health_code_update(&health_thread_cmd);
3303
3304 // TODO: Validate cmd_ctx including sanity check for
3305 // security purpose.
3306
3307 rcu_thread_online();
3308 /*
3309 * This function dispatch the work to the kernel or userspace tracer
3310 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3311 * informations for the client. The command context struct contains
3312 * everything this function may needs.
3313 */
3314 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3315 rcu_thread_offline();
3316 if (ret < 0) {
3317 if (sock_error) {
3318 ret = close(sock);
3319 if (ret) {
3320 PERROR("close");
3321 }
3322 sock = -1;
3323 }
3324 /*
3325 * TODO: Inform client somehow of the fatal error. At
3326 * this point, ret < 0 means that a zmalloc failed
3327 * (ENOMEM). Error detected but still accept
3328 * command, unless a socket error has been
3329 * detected.
3330 */
3331 clean_command_ctx(&cmd_ctx);
3332 continue;
3333 }
3334
3335 health_code_update(&health_thread_cmd);
3336
3337 DBG("Sending response (size: %d, retcode: %s)",
3338 cmd_ctx->lttng_msg_size,
3339 lttng_strerror(-cmd_ctx->llm->ret_code));
3340 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3341 if (ret < 0) {
3342 ERR("Failed to send data back to client");
3343 }
3344
3345 /* End of transmission */
3346 ret = close(sock);
3347 if (ret) {
3348 PERROR("close");
3349 }
3350 sock = -1;
3351
3352 clean_command_ctx(&cmd_ctx);
3353
3354 health_code_update(&health_thread_cmd);
3355 }
3356
3357 exit:
3358 error:
3359 if (sock >= 0) {
3360 ret = close(sock);
3361 if (ret) {
3362 PERROR("close");
3363 }
3364 }
3365
3366 lttng_poll_clean(&events);
3367 clean_command_ctx(&cmd_ctx);
3368
3369 error_listen:
3370 error_create_poll:
3371 error_testpoint:
3372 unlink(client_unix_sock_path);
3373 if (client_sock >= 0) {
3374 ret = close(client_sock);
3375 if (ret) {
3376 PERROR("close");
3377 }
3378 }
3379
3380 if (err) {
3381 health_error(&health_thread_cmd);
3382 ERR("Health error occurred in %s", __func__);
3383 }
3384
3385 health_unregister();
3386
3387 DBG("Client thread dying");
3388
3389 rcu_unregister_thread();
3390 return NULL;
3391 }
3392
3393
3394 /*
3395 * usage function on stderr
3396 */
3397 static void usage(void)
3398 {
3399 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3400 fprintf(stderr, " -h, --help Display this usage.\n");
3401 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3402 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3403 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3404 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3405 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3406 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3407 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3408 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3409 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3410 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3411 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3412 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3413 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3414 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3415 fprintf(stderr, " -V, --version Show version number.\n");
3416 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3417 fprintf(stderr, " -q, --quiet No output at all.\n");
3418 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3419 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3420 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3421 }
3422
3423 /*
3424 * daemon argument parsing
3425 */
3426 static int parse_args(int argc, char **argv)
3427 {
3428 int c;
3429
3430 static struct option long_options[] = {
3431 { "client-sock", 1, 0, 'c' },
3432 { "apps-sock", 1, 0, 'a' },
3433 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3434 { "kconsumerd-err-sock", 1, 0, 'E' },
3435 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3436 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3437 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3438 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3439 { "consumerd32-path", 1, 0, 'u' },
3440 { "consumerd32-libdir", 1, 0, 'U' },
3441 { "consumerd64-path", 1, 0, 't' },
3442 { "consumerd64-libdir", 1, 0, 'T' },
3443 { "daemonize", 0, 0, 'd' },
3444 { "sig-parent", 0, 0, 'S' },
3445 { "help", 0, 0, 'h' },
3446 { "group", 1, 0, 'g' },
3447 { "version", 0, 0, 'V' },
3448 { "quiet", 0, 0, 'q' },
3449 { "verbose", 0, 0, 'v' },
3450 { "verbose-consumer", 0, 0, 'Z' },
3451 { "no-kernel", 0, 0, 'N' },
3452 { NULL, 0, 0, 0 }
3453 };
3454
3455 while (1) {
3456 int option_index = 0;
3457 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3458 long_options, &option_index);
3459 if (c == -1) {
3460 break;
3461 }
3462
3463 switch (c) {
3464 case 0:
3465 fprintf(stderr, "option %s", long_options[option_index].name);
3466 if (optarg) {
3467 fprintf(stderr, " with arg %s\n", optarg);
3468 }
3469 break;
3470 case 'c':
3471 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3472 break;
3473 case 'a':
3474 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3475 break;
3476 case 'd':
3477 opt_daemon = 1;
3478 break;
3479 case 'g':
3480 opt_tracing_group = optarg;
3481 break;
3482 case 'h':
3483 usage();
3484 exit(EXIT_FAILURE);
3485 case 'V':
3486 fprintf(stdout, "%s\n", VERSION);
3487 exit(EXIT_SUCCESS);
3488 case 'S':
3489 opt_sig_parent = 1;
3490 break;
3491 case 'E':
3492 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3493 break;
3494 case 'C':
3495 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3496 break;
3497 case 'F':
3498 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3499 break;
3500 case 'D':
3501 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3502 break;
3503 case 'H':
3504 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3505 break;
3506 case 'G':
3507 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3508 break;
3509 case 'N':
3510 opt_no_kernel = 1;
3511 break;
3512 case 'q':
3513 lttng_opt_quiet = 1;
3514 break;
3515 case 'v':
3516 /* Verbose level can increase using multiple -v */
3517 lttng_opt_verbose += 1;
3518 break;
3519 case 'Z':
3520 opt_verbose_consumer += 1;
3521 break;
3522 case 'u':
3523 consumerd32_bin= optarg;
3524 break;
3525 case 'U':
3526 consumerd32_libdir = optarg;
3527 break;
3528 case 't':
3529 consumerd64_bin = optarg;
3530 break;
3531 case 'T':
3532 consumerd64_libdir = optarg;
3533 break;
3534 default:
3535 /* Unknown option or other error.
3536 * Error is printed by getopt, just return */
3537 return -1;
3538 }
3539 }
3540
3541 return 0;
3542 }
3543
3544 /*
3545 * Creates the two needed socket by the daemon.
3546 * apps_sock - The communication socket for all UST apps.
3547 * client_sock - The communication of the cli tool (lttng).
3548 */
3549 static int init_daemon_socket(void)
3550 {
3551 int ret = 0;
3552 mode_t old_umask;
3553
3554 old_umask = umask(0);
3555
3556 /* Create client tool unix socket */
3557 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3558 if (client_sock < 0) {
3559 ERR("Create unix sock failed: %s", client_unix_sock_path);
3560 ret = -1;
3561 goto end;
3562 }
3563
3564 /* Set the cloexec flag */
3565 ret = utils_set_fd_cloexec(client_sock);
3566 if (ret < 0) {
3567 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3568 "Continuing but note that the consumer daemon will have a "
3569 "reference to this socket on exec()", client_sock);
3570 }
3571
3572 /* File permission MUST be 660 */
3573 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3574 if (ret < 0) {
3575 ERR("Set file permissions failed: %s", client_unix_sock_path);
3576 PERROR("chmod");
3577 goto end;
3578 }
3579
3580 /* Create the application unix socket */
3581 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3582 if (apps_sock < 0) {
3583 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3584 ret = -1;
3585 goto end;
3586 }
3587
3588 /* Set the cloexec flag */
3589 ret = utils_set_fd_cloexec(apps_sock);
3590 if (ret < 0) {
3591 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3592 "Continuing but note that the consumer daemon will have a "
3593 "reference to this socket on exec()", apps_sock);
3594 }
3595
3596 /* File permission MUST be 666 */
3597 ret = chmod(apps_unix_sock_path,
3598 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3599 if (ret < 0) {
3600 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3601 PERROR("chmod");
3602 goto end;
3603 }
3604
3605 DBG3("Session daemon client socket %d and application socket %d created",
3606 client_sock, apps_sock);
3607
3608 end:
3609 umask(old_umask);
3610 return ret;
3611 }
3612
3613 /*
3614 * Check if the global socket is available, and if a daemon is answering at the
3615 * other side. If yes, error is returned.
3616 */
3617 static int check_existing_daemon(void)
3618 {
3619 /* Is there anybody out there ? */
3620 if (lttng_session_daemon_alive()) {
3621 return -EEXIST;
3622 }
3623
3624 return 0;
3625 }
3626
3627 /*
3628 * Set the tracing group gid onto the client socket.
3629 *
3630 * Race window between mkdir and chown is OK because we are going from more
3631 * permissive (root.root) to less permissive (root.tracing).
3632 */
3633 static int set_permissions(char *rundir)
3634 {
3635 int ret;
3636 gid_t gid;
3637
3638 ret = allowed_group();
3639 if (ret < 0) {
3640 WARN("No tracing group detected");
3641 ret = 0;
3642 goto end;
3643 }
3644
3645 gid = ret;
3646
3647 /* Set lttng run dir */
3648 ret = chown(rundir, 0, gid);
3649 if (ret < 0) {
3650 ERR("Unable to set group on %s", rundir);
3651 PERROR("chown");
3652 }
3653
3654 /* Ensure tracing group can search the run dir */
3655 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3656 if (ret < 0) {
3657 ERR("Unable to set permissions on %s", rundir);
3658 PERROR("chmod");
3659 }
3660
3661 /* lttng client socket path */
3662 ret = chown(client_unix_sock_path, 0, gid);
3663 if (ret < 0) {
3664 ERR("Unable to set group on %s", client_unix_sock_path);
3665 PERROR("chown");
3666 }
3667
3668 /* kconsumer error socket path */
3669 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3670 if (ret < 0) {
3671 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3672 PERROR("chown");
3673 }
3674
3675 /* 64-bit ustconsumer error socket path */
3676 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3677 if (ret < 0) {
3678 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3679 PERROR("chown");
3680 }
3681
3682 /* 32-bit ustconsumer compat32 error socket path */
3683 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3684 if (ret < 0) {
3685 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3686 PERROR("chown");
3687 }
3688
3689 DBG("All permissions are set");
3690
3691 end:
3692 return ret;
3693 }
3694
3695 /*
3696 * Create the lttng run directory needed for all global sockets and pipe.
3697 */
3698 static int create_lttng_rundir(const char *rundir)
3699 {
3700 int ret;
3701
3702 DBG3("Creating LTTng run directory: %s", rundir);
3703
3704 ret = mkdir(rundir, S_IRWXU);
3705 if (ret < 0) {
3706 if (errno != EEXIST) {
3707 ERR("Unable to create %s", rundir);
3708 goto error;
3709 } else {
3710 ret = 0;
3711 }
3712 }
3713
3714 error:
3715 return ret;
3716 }
3717
3718 /*
3719 * Setup sockets and directory needed by the kconsumerd communication with the
3720 * session daemon.
3721 */
3722 static int set_consumer_sockets(struct consumer_data *consumer_data,
3723 const char *rundir)
3724 {
3725 int ret;
3726 char path[PATH_MAX];
3727
3728 switch (consumer_data->type) {
3729 case LTTNG_CONSUMER_KERNEL:
3730 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3731 break;
3732 case LTTNG_CONSUMER64_UST:
3733 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3734 break;
3735 case LTTNG_CONSUMER32_UST:
3736 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3737 break;
3738 default:
3739 ERR("Consumer type unknown");
3740 ret = -EINVAL;
3741 goto error;
3742 }
3743
3744 DBG2("Creating consumer directory: %s", path);
3745
3746 ret = mkdir(path, S_IRWXU);
3747 if (ret < 0) {
3748 if (errno != EEXIST) {
3749 PERROR("mkdir");
3750 ERR("Failed to create %s", path);
3751 goto error;
3752 }
3753 ret = -1;
3754 }
3755
3756 /* Create the kconsumerd error unix socket */
3757 consumer_data->err_sock =
3758 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3759 if (consumer_data->err_sock < 0) {
3760 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3761 ret = -1;
3762 goto error;
3763 }
3764
3765 /* File permission MUST be 660 */
3766 ret = chmod(consumer_data->err_unix_sock_path,
3767 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3768 if (ret < 0) {
3769 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3770 PERROR("chmod");
3771 goto error;
3772 }
3773
3774 error:
3775 return ret;
3776 }
3777
3778 /*
3779 * Signal handler for the daemon
3780 *
3781 * Simply stop all worker threads, leaving main() return gracefully after
3782 * joining all threads and calling cleanup().
3783 */
3784 static void sighandler(int sig)
3785 {
3786 switch (sig) {
3787 case SIGPIPE:
3788 DBG("SIGPIPE caught");
3789 return;
3790 case SIGINT:
3791 DBG("SIGINT caught");
3792 stop_threads();
3793 break;
3794 case SIGTERM:
3795 DBG("SIGTERM caught");
3796 stop_threads();
3797 break;
3798 default:
3799 break;
3800 }
3801 }
3802
3803 /*
3804 * Setup signal handler for :
3805 * SIGINT, SIGTERM, SIGPIPE
3806 */
3807 static int set_signal_handler(void)
3808 {
3809 int ret = 0;
3810 struct sigaction sa;
3811 sigset_t sigset;
3812
3813 if ((ret = sigemptyset(&sigset)) < 0) {
3814 PERROR("sigemptyset");
3815 return ret;
3816 }
3817
3818 sa.sa_handler = sighandler;
3819 sa.sa_mask = sigset;
3820 sa.sa_flags = 0;
3821 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3822 PERROR("sigaction");
3823 return ret;
3824 }
3825
3826 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3827 PERROR("sigaction");
3828 return ret;
3829 }
3830
3831 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3832 PERROR("sigaction");
3833 return ret;
3834 }
3835
3836 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3837
3838 return ret;
3839 }
3840
3841 /*
3842 * Set open files limit to unlimited. This daemon can open a large number of
3843 * file descriptors in order to consumer multiple kernel traces.
3844 */
3845 static void set_ulimit(void)
3846 {
3847 int ret;
3848 struct rlimit lim;
3849
3850 /* The kernel does not allowed an infinite limit for open files */
3851 lim.rlim_cur = 65535;
3852 lim.rlim_max = 65535;
3853
3854 ret = setrlimit(RLIMIT_NOFILE, &lim);
3855 if (ret < 0) {
3856 PERROR("failed to set open files limit");
3857 }
3858 }
3859
3860 /*
3861 * main
3862 */
3863 int main(int argc, char **argv)
3864 {
3865 int ret = 0;
3866 void *status;
3867 const char *home_path, *env_app_timeout;
3868
3869 init_kernel_workarounds();
3870
3871 rcu_register_thread();
3872
3873 setup_consumerd_path();
3874
3875 /* Parse arguments */
3876 progname = argv[0];
3877 if ((ret = parse_args(argc, argv)) < 0) {
3878 goto error;
3879 }
3880
3881 /* Daemonize */
3882 if (opt_daemon) {
3883 int i;
3884
3885 /*
3886 * fork
3887 * child: setsid, close FD 0, 1, 2, chdir /
3888 * parent: exit (if fork is successful)
3889 */
3890 ret = daemon(0, 0);
3891 if (ret < 0) {
3892 PERROR("daemon");
3893 goto error;
3894 }
3895 /*
3896 * We are in the child. Make sure all other file
3897 * descriptors are closed, in case we are called with
3898 * more opened file descriptors than the standard ones.
3899 */
3900 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3901 (void) close(i);
3902 }
3903 }
3904
3905 /* Create thread quit pipe */
3906 if ((ret = init_thread_quit_pipe()) < 0) {
3907 goto error;
3908 }
3909
3910 /* Check if daemon is UID = 0 */
3911 is_root = !getuid();
3912
3913 if (is_root) {
3914 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3915
3916 /* Create global run dir with root access */
3917 ret = create_lttng_rundir(rundir);
3918 if (ret < 0) {
3919 goto error;
3920 }
3921
3922 if (strlen(apps_unix_sock_path) == 0) {
3923 snprintf(apps_unix_sock_path, PATH_MAX,
3924 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3925 }
3926
3927 if (strlen(client_unix_sock_path) == 0) {
3928 snprintf(client_unix_sock_path, PATH_MAX,
3929 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3930 }
3931
3932 /* Set global SHM for ust */
3933 if (strlen(wait_shm_path) == 0) {
3934 snprintf(wait_shm_path, PATH_MAX,
3935 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3936 }
3937
3938 if (strlen(health_unix_sock_path) == 0) {
3939 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3940 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3941 }
3942
3943 /* Setup kernel consumerd path */
3944 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3945 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3946 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3947 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3948
3949 DBG2("Kernel consumer err path: %s",
3950 kconsumer_data.err_unix_sock_path);
3951 DBG2("Kernel consumer cmd path: %s",
3952 kconsumer_data.cmd_unix_sock_path);
3953 } else {
3954 home_path = get_home_dir();
3955 if (home_path == NULL) {
3956 /* TODO: Add --socket PATH option */
3957 ERR("Can't get HOME directory for sockets creation.");
3958 ret = -EPERM;
3959 goto error;
3960 }
3961
3962 /*
3963 * Create rundir from home path. This will create something like
3964 * $HOME/.lttng
3965 */
3966 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3967 if (ret < 0) {
3968 ret = -ENOMEM;
3969 goto error;
3970 }
3971
3972 ret = create_lttng_rundir(rundir);
3973 if (ret < 0) {
3974 goto error;
3975 }
3976
3977 if (strlen(apps_unix_sock_path) == 0) {
3978 snprintf(apps_unix_sock_path, PATH_MAX,
3979 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3980 }
3981
3982 /* Set the cli tool unix socket path */
3983 if (strlen(client_unix_sock_path) == 0) {
3984 snprintf(client_unix_sock_path, PATH_MAX,
3985 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3986 }
3987
3988 /* Set global SHM for ust */
3989 if (strlen(wait_shm_path) == 0) {
3990 snprintf(wait_shm_path, PATH_MAX,
3991 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3992 }
3993
3994 /* Set health check Unix path */
3995 if (strlen(health_unix_sock_path) == 0) {
3996 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3997 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3998 }
3999 }
4000
4001 /* Set consumer initial state */
4002 kernel_consumerd_state = CONSUMER_STOPPED;
4003 ust_consumerd_state = CONSUMER_STOPPED;
4004
4005 DBG("Client socket path %s", client_unix_sock_path);
4006 DBG("Application socket path %s", apps_unix_sock_path);
4007 DBG("LTTng run directory path: %s", rundir);
4008
4009 /* 32 bits consumerd path setup */
4010 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4011 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4012 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4013 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4014
4015 DBG2("UST consumer 32 bits err path: %s",
4016 ustconsumer32_data.err_unix_sock_path);
4017 DBG2("UST consumer 32 bits cmd path: %s",
4018 ustconsumer32_data.cmd_unix_sock_path);
4019
4020 /* 64 bits consumerd path setup */
4021 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4022 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4023 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4024 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4025
4026 DBG2("UST consumer 64 bits err path: %s",
4027 ustconsumer64_data.err_unix_sock_path);
4028 DBG2("UST consumer 64 bits cmd path: %s",
4029 ustconsumer64_data.cmd_unix_sock_path);
4030
4031 /*
4032 * See if daemon already exist.
4033 */
4034 if ((ret = check_existing_daemon()) < 0) {
4035 ERR("Already running daemon.\n");
4036 /*
4037 * We do not goto exit because we must not cleanup()
4038 * because a daemon is already running.
4039 */
4040 goto error;
4041 }
4042
4043 /*
4044 * Init UST app hash table. Alloc hash table before this point since
4045 * cleanup() can get called after that point.
4046 */
4047 ust_app_ht_alloc();
4048
4049 /* After this point, we can safely call cleanup() with "goto exit" */
4050
4051 /*
4052 * These actions must be executed as root. We do that *after* setting up
4053 * the sockets path because we MUST make the check for another daemon using
4054 * those paths *before* trying to set the kernel consumer sockets and init
4055 * kernel tracer.
4056 */
4057 if (is_root) {
4058 ret = set_consumer_sockets(&kconsumer_data, rundir);
4059 if (ret < 0) {
4060 goto exit;
4061 }
4062
4063 /* Setup kernel tracer */
4064 if (!opt_no_kernel) {
4065 init_kernel_tracer();
4066 }
4067
4068 /* Set ulimit for open files */
4069 set_ulimit();
4070 }
4071 /* init lttng_fd tracking must be done after set_ulimit. */
4072 lttng_fd_init();
4073
4074 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4075 if (ret < 0) {
4076 goto exit;
4077 }
4078
4079 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4080 if (ret < 0) {
4081 goto exit;
4082 }
4083
4084 if ((ret = set_signal_handler()) < 0) {
4085 goto exit;
4086 }
4087
4088 /* Setup the needed unix socket */
4089 if ((ret = init_daemon_socket()) < 0) {
4090 goto exit;
4091 }
4092
4093 /* Set credentials to socket */
4094 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4095 goto exit;
4096 }
4097
4098 /* Get parent pid if -S, --sig-parent is specified. */
4099 if (opt_sig_parent) {
4100 ppid = getppid();
4101 }
4102
4103 /* Setup the kernel pipe for waking up the kernel thread */
4104 if (is_root && !opt_no_kernel) {
4105 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4106 goto exit;
4107 }
4108 }
4109
4110 /* Setup the thread apps communication pipe. */
4111 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4112 goto exit;
4113 }
4114
4115 /* Init UST command queue. */
4116 cds_wfq_init(&ust_cmd_queue.queue);
4117
4118 /*
4119 * Get session list pointer. This pointer MUST NOT be free(). This list is
4120 * statically declared in session.c
4121 */
4122 session_list_ptr = session_get_list();
4123
4124 /* Set up max poll set size */
4125 lttng_poll_set_max_size();
4126
4127 cmd_init();
4128
4129 /* Check for the application socket timeout env variable. */
4130 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4131 if (env_app_timeout) {
4132 app_socket_timeout = atoi(env_app_timeout);
4133 } else {
4134 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4135 }
4136
4137 /* Create thread to manage the client socket */
4138 ret = pthread_create(&health_thread, NULL,
4139 thread_manage_health, (void *) NULL);
4140 if (ret != 0) {
4141 PERROR("pthread_create health");
4142 goto exit_health;
4143 }
4144
4145 /* Create thread to manage the client socket */
4146 ret = pthread_create(&client_thread, NULL,
4147 thread_manage_clients, (void *) NULL);
4148 if (ret != 0) {
4149 PERROR("pthread_create clients");
4150 goto exit_client;
4151 }
4152
4153 /* Create thread to dispatch registration */
4154 ret = pthread_create(&dispatch_thread, NULL,
4155 thread_dispatch_ust_registration, (void *) NULL);
4156 if (ret != 0) {
4157 PERROR("pthread_create dispatch");
4158 goto exit_dispatch;
4159 }
4160
4161 /* Create thread to manage application registration. */
4162 ret = pthread_create(&reg_apps_thread, NULL,
4163 thread_registration_apps, (void *) NULL);
4164 if (ret != 0) {
4165 PERROR("pthread_create registration");
4166 goto exit_reg_apps;
4167 }
4168
4169 /* Create thread to manage application socket */
4170 ret = pthread_create(&apps_thread, NULL,
4171 thread_manage_apps, (void *) NULL);
4172 if (ret != 0) {
4173 PERROR("pthread_create apps");
4174 goto exit_apps;
4175 }
4176
4177 /* Don't start this thread if kernel tracing is not requested nor root */
4178 if (is_root && !opt_no_kernel) {
4179 /* Create kernel thread to manage kernel event */
4180 ret = pthread_create(&kernel_thread, NULL,
4181 thread_manage_kernel, (void *) NULL);
4182 if (ret != 0) {
4183 PERROR("pthread_create kernel");
4184 goto exit_kernel;
4185 }
4186
4187 ret = pthread_join(kernel_thread, &status);
4188 if (ret != 0) {
4189 PERROR("pthread_join");
4190 goto error; /* join error, exit without cleanup */
4191 }
4192 }
4193
4194 exit_kernel:
4195 ret = pthread_join(apps_thread, &status);
4196 if (ret != 0) {
4197 PERROR("pthread_join");
4198 goto error; /* join error, exit without cleanup */
4199 }
4200
4201 exit_apps:
4202 ret = pthread_join(reg_apps_thread, &status);
4203 if (ret != 0) {
4204 PERROR("pthread_join");
4205 goto error; /* join error, exit without cleanup */
4206 }
4207
4208 exit_reg_apps:
4209 ret = pthread_join(dispatch_thread, &status);
4210 if (ret != 0) {
4211 PERROR("pthread_join");
4212 goto error; /* join error, exit without cleanup */
4213 }
4214
4215 exit_dispatch:
4216 ret = pthread_join(client_thread, &status);
4217 if (ret != 0) {
4218 PERROR("pthread_join");
4219 goto error; /* join error, exit without cleanup */
4220 }
4221
4222 ret = join_consumer_thread(&kconsumer_data);
4223 if (ret != 0) {
4224 PERROR("join_consumer");
4225 goto error; /* join error, exit without cleanup */
4226 }
4227
4228 ret = join_consumer_thread(&ustconsumer32_data);
4229 if (ret != 0) {
4230 PERROR("join_consumer ust32");
4231 goto error; /* join error, exit without cleanup */
4232 }
4233
4234 ret = join_consumer_thread(&ustconsumer64_data);
4235 if (ret != 0) {
4236 PERROR("join_consumer ust64");
4237 goto error; /* join error, exit without cleanup */
4238 }
4239
4240 exit_client:
4241 ret = pthread_join(health_thread, &status);
4242 if (ret != 0) {
4243 PERROR("pthread_join health thread");
4244 goto error; /* join error, exit without cleanup */
4245 }
4246
4247 exit_health:
4248 exit:
4249 /*
4250 * cleanup() is called when no other thread is running.
4251 */
4252 rcu_thread_online();
4253 cleanup();
4254 rcu_thread_offline();
4255 rcu_unregister_thread();
4256 if (!ret) {
4257 exit(EXIT_SUCCESS);
4258 }
4259 error:
4260 exit(EXIT_FAILURE);
4261 }
This page took 0.109585 seconds and 3 git commands to generate.