Fix: poll and epoll fd set reallocation
[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
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 goto error;
641 }
642 }
643 }
644 goto error;
645 }
646 }
647 session_unlock(session);
648 }
649 session_unlock_list();
650 return ret;
651
652 error:
653 session_unlock(session);
654 session_unlock_list();
655 return ret;
656 }
657
658 /*
659 * For each tracing session, update newly registered apps.
660 */
661 static void update_ust_app(int app_sock)
662 {
663 struct ltt_session *sess, *stmp;
664
665 session_lock_list();
666
667 /* For all tracing session(s) */
668 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
669 session_lock(sess);
670 if (sess->ust_session) {
671 ust_app_global_update(sess->ust_session, app_sock);
672 }
673 session_unlock(sess);
674 }
675
676 session_unlock_list();
677 }
678
679 /*
680 * This thread manage event coming from the kernel.
681 *
682 * Features supported in this thread:
683 * -) CPU Hotplug
684 */
685 static void *thread_manage_kernel(void *data)
686 {
687 int ret, i, pollfd, update_poll_flag = 1, err = -1;
688 uint32_t revents, nb_fd;
689 char tmp;
690 struct lttng_poll_event events;
691
692 DBG("[thread] Thread manage kernel started");
693
694 if (testpoint(thread_manage_kernel)) {
695 goto error_testpoint;
696 }
697
698 health_code_update(&health_thread_kernel);
699
700 if (testpoint(thread_manage_kernel_before_loop)) {
701 goto error_testpoint;
702 }
703
704 while (1) {
705 health_code_update(&health_thread_kernel);
706
707 if (update_poll_flag == 1) {
708 /* Clean events object. We are about to populate it again. */
709 lttng_poll_clean(&events);
710
711 ret = create_thread_poll_set(&events, 2);
712 if (ret < 0) {
713 goto error_poll_create;
714 }
715
716 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
717 if (ret < 0) {
718 goto error;
719 }
720
721 /* This will add the available kernel channel if any. */
722 ret = update_kernel_poll(&events);
723 if (ret < 0) {
724 goto error;
725 }
726 update_poll_flag = 0;
727 }
728
729 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
730
731 /* Poll infinite value of time */
732 restart:
733 health_poll_update(&health_thread_kernel);
734 ret = lttng_poll_wait(&events, -1);
735 health_poll_update(&health_thread_kernel);
736 if (ret < 0) {
737 /*
738 * Restart interrupted system call.
739 */
740 if (errno == EINTR) {
741 goto restart;
742 }
743 goto error;
744 } else if (ret == 0) {
745 /* Should not happen since timeout is infinite */
746 ERR("Return value of poll is 0 with an infinite timeout.\n"
747 "This should not have happened! Continuing...");
748 continue;
749 }
750
751 nb_fd = ret;
752
753 for (i = 0; i < nb_fd; i++) {
754 /* Fetch once the poll data */
755 revents = LTTNG_POLL_GETEV(&events, i);
756 pollfd = LTTNG_POLL_GETFD(&events, i);
757
758 health_code_update(&health_thread_kernel);
759
760 /* Thread quit pipe has been closed. Killing thread. */
761 ret = check_thread_quit_pipe(pollfd, revents);
762 if (ret) {
763 err = 0;
764 goto exit;
765 }
766
767 /* Check for data on kernel pipe */
768 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
769 ret = read(kernel_poll_pipe[0], &tmp, 1);
770 update_poll_flag = 1;
771 continue;
772 } else {
773 /*
774 * New CPU detected by the kernel. Adding kernel stream to
775 * kernel session and updating the kernel consumer
776 */
777 if (revents & LPOLLIN) {
778 ret = update_kernel_stream(&kconsumer_data, pollfd);
779 if (ret < 0) {
780 continue;
781 }
782 break;
783 /*
784 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
785 * and unregister kernel stream at this point.
786 */
787 }
788 }
789 }
790 }
791
792 exit:
793 error:
794 lttng_poll_clean(&events);
795 error_poll_create:
796 error_testpoint:
797 utils_close_pipe(kernel_poll_pipe);
798 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
799 if (err) {
800 health_error(&health_thread_kernel);
801 ERR("Health error occurred in %s", __func__);
802 WARN("Kernel thread died unexpectedly. "
803 "Kernel tracing can continue but CPU hotplug is disabled.");
804 }
805 health_exit(&health_thread_kernel);
806 DBG("Kernel thread dying");
807 return NULL;
808 }
809
810 /*
811 * Signal pthread condition of the consumer data that the thread.
812 */
813 static void signal_consumer_condition(struct consumer_data *data, int state)
814 {
815 pthread_mutex_lock(&data->cond_mutex);
816
817 /*
818 * The state is set before signaling. It can be any value, it's the waiter
819 * job to correctly interpret this condition variable associated to the
820 * consumer pthread_cond.
821 *
822 * A value of 0 means that the corresponding thread of the consumer data
823 * was not started. 1 indicates that the thread has started and is ready
824 * for action. A negative value means that there was an error during the
825 * thread bootstrap.
826 */
827 data->consumer_thread_is_ready = state;
828 (void) pthread_cond_signal(&data->cond);
829
830 pthread_mutex_unlock(&data->cond_mutex);
831 }
832
833 /*
834 * This thread manage the consumer error sent back to the session daemon.
835 */
836 static void *thread_manage_consumer(void *data)
837 {
838 int sock = -1, i, ret, pollfd, err = -1;
839 uint32_t revents, nb_fd;
840 enum lttcomm_return_code code;
841 struct lttng_poll_event events;
842 struct consumer_data *consumer_data = data;
843
844 DBG("[thread] Manage consumer started");
845
846 /*
847 * Since the consumer thread can be spawned at any moment in time, we init
848 * the health to a poll status (1, which is a valid health over time).
849 * When the thread starts, we update here the health to a "code" path being
850 * an even value so this thread, when reaching a poll wait, does not
851 * trigger an error with an even value.
852 *
853 * Here is the use case we avoid.
854 *
855 * +1: the first poll update during initialization (main())
856 * +2 * x: multiple code update once in this thread.
857 * +1: poll wait in this thread (being a good health state).
858 * == even number which after the wait period shows as a bad health.
859 *
860 * In a nutshell, the following poll update to the health state brings back
861 * the state to an even value meaning a code path.
862 */
863 health_poll_update(&consumer_data->health);
864
865 /*
866 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
867 * Nothing more will be added to this poll set.
868 */
869 ret = create_thread_poll_set(&events, 2);
870 if (ret < 0) {
871 goto error_poll;
872 }
873
874 /*
875 * The error socket here is already in a listening state which was done
876 * just before spawning this thread to avoid a race between the consumer
877 * daemon exec trying to connect and the listen() call.
878 */
879 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
880 if (ret < 0) {
881 goto error;
882 }
883
884 health_code_update(&consumer_data->health);
885
886 /* Inifinite blocking call, waiting for transmission */
887 restart:
888 health_poll_update(&consumer_data->health);
889
890 if (testpoint(thread_manage_consumer)) {
891 goto error;
892 }
893
894 ret = lttng_poll_wait(&events, -1);
895 health_poll_update(&consumer_data->health);
896 if (ret < 0) {
897 /*
898 * Restart interrupted system call.
899 */
900 if (errno == EINTR) {
901 goto restart;
902 }
903 goto error;
904 }
905
906 nb_fd = ret;
907
908 for (i = 0; i < nb_fd; i++) {
909 /* Fetch once the poll data */
910 revents = LTTNG_POLL_GETEV(&events, i);
911 pollfd = LTTNG_POLL_GETFD(&events, i);
912
913 health_code_update(&consumer_data->health);
914
915 /* Thread quit pipe has been closed. Killing thread. */
916 ret = check_thread_quit_pipe(pollfd, revents);
917 if (ret) {
918 err = 0;
919 goto exit;
920 }
921
922 /* Event on the registration socket */
923 if (pollfd == consumer_data->err_sock) {
924 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
925 ERR("consumer err socket poll error");
926 goto error;
927 }
928 }
929 }
930
931 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
932 if (sock < 0) {
933 goto error;
934 }
935
936 /*
937 * Set the CLOEXEC flag. Return code is useless because either way, the
938 * show must go on.
939 */
940 (void) utils_set_fd_cloexec(sock);
941
942 health_code_update(&consumer_data->health);
943
944 DBG2("Receiving code from consumer err_sock");
945
946 /* Getting status code from kconsumerd */
947 ret = lttcomm_recv_unix_sock(sock, &code,
948 sizeof(enum lttcomm_return_code));
949 if (ret <= 0) {
950 goto error;
951 }
952
953 health_code_update(&consumer_data->health);
954
955 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
956 consumer_data->cmd_sock =
957 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
958 if (consumer_data->cmd_sock < 0) {
959 /* On error, signal condition and quit. */
960 signal_consumer_condition(consumer_data, -1);
961 PERROR("consumer connect");
962 goto error;
963 }
964 signal_consumer_condition(consumer_data, 1);
965 DBG("Consumer command socket ready");
966 } else {
967 ERR("consumer error when waiting for SOCK_READY : %s",
968 lttcomm_get_readable_code(-code));
969 goto error;
970 }
971
972 /* Remove the kconsumerd error sock since we've established a connexion */
973 ret = lttng_poll_del(&events, consumer_data->err_sock);
974 if (ret < 0) {
975 goto error;
976 }
977
978 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
979 if (ret < 0) {
980 goto error;
981 }
982
983 health_code_update(&consumer_data->health);
984
985 /* Inifinite blocking call, waiting for transmission */
986 restart_poll:
987 health_poll_update(&consumer_data->health);
988 ret = lttng_poll_wait(&events, -1);
989 health_poll_update(&consumer_data->health);
990 if (ret < 0) {
991 /*
992 * Restart interrupted system call.
993 */
994 if (errno == EINTR) {
995 goto restart_poll;
996 }
997 goto error;
998 }
999
1000 nb_fd = ret;
1001
1002 for (i = 0; i < nb_fd; i++) {
1003 /* Fetch once the poll data */
1004 revents = LTTNG_POLL_GETEV(&events, i);
1005 pollfd = LTTNG_POLL_GETFD(&events, i);
1006
1007 health_code_update(&consumer_data->health);
1008
1009 /* Thread quit pipe has been closed. Killing thread. */
1010 ret = check_thread_quit_pipe(pollfd, revents);
1011 if (ret) {
1012 err = 0;
1013 goto exit;
1014 }
1015
1016 /* Event on the kconsumerd socket */
1017 if (pollfd == sock) {
1018 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1019 ERR("consumer err socket second poll error");
1020 goto error;
1021 }
1022 }
1023 }
1024
1025 health_code_update(&consumer_data->health);
1026
1027 /* Wait for any kconsumerd error */
1028 ret = lttcomm_recv_unix_sock(sock, &code,
1029 sizeof(enum lttcomm_return_code));
1030 if (ret <= 0) {
1031 ERR("consumer closed the command socket");
1032 goto error;
1033 }
1034
1035 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1036
1037 exit:
1038 error:
1039 /* Immediately set the consumerd state to stopped */
1040 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1041 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1042 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1043 consumer_data->type == LTTNG_CONSUMER32_UST) {
1044 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1045 } else {
1046 /* Code flow error... */
1047 assert(0);
1048 }
1049
1050 if (consumer_data->err_sock >= 0) {
1051 ret = close(consumer_data->err_sock);
1052 if (ret) {
1053 PERROR("close");
1054 }
1055 }
1056 if (consumer_data->cmd_sock >= 0) {
1057 ret = close(consumer_data->cmd_sock);
1058 if (ret) {
1059 PERROR("close");
1060 }
1061 }
1062 if (sock >= 0) {
1063 ret = close(sock);
1064 if (ret) {
1065 PERROR("close");
1066 }
1067 }
1068
1069 unlink(consumer_data->err_unix_sock_path);
1070 unlink(consumer_data->cmd_unix_sock_path);
1071 consumer_data->pid = 0;
1072
1073 lttng_poll_clean(&events);
1074 error_poll:
1075 if (err) {
1076 health_error(&consumer_data->health);
1077 ERR("Health error occurred in %s", __func__);
1078 }
1079 health_exit(&consumer_data->health);
1080 DBG("consumer thread cleanup completed");
1081
1082 return NULL;
1083 }
1084
1085 /*
1086 * This thread manage application communication.
1087 */
1088 static void *thread_manage_apps(void *data)
1089 {
1090 int i, ret, pollfd, err = -1;
1091 uint32_t revents, nb_fd;
1092 struct ust_command ust_cmd;
1093 struct lttng_poll_event events;
1094
1095 DBG("[thread] Manage application started");
1096
1097 rcu_register_thread();
1098 rcu_thread_online();
1099
1100 if (testpoint(thread_manage_apps)) {
1101 goto error_testpoint;
1102 }
1103
1104 health_code_update(&health_thread_app_manage);
1105
1106 ret = create_thread_poll_set(&events, 2);
1107 if (ret < 0) {
1108 goto error_poll_create;
1109 }
1110
1111 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1112 if (ret < 0) {
1113 goto error;
1114 }
1115
1116 if (testpoint(thread_manage_apps_before_loop)) {
1117 goto error;
1118 }
1119
1120 health_code_update(&health_thread_app_manage);
1121
1122 while (1) {
1123 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1124
1125 /* Inifinite blocking call, waiting for transmission */
1126 restart:
1127 health_poll_update(&health_thread_app_manage);
1128 ret = lttng_poll_wait(&events, -1);
1129 health_poll_update(&health_thread_app_manage);
1130 if (ret < 0) {
1131 /*
1132 * Restart interrupted system call.
1133 */
1134 if (errno == EINTR) {
1135 goto restart;
1136 }
1137 goto error;
1138 }
1139
1140 nb_fd = ret;
1141
1142 for (i = 0; i < nb_fd; i++) {
1143 /* Fetch once the poll data */
1144 revents = LTTNG_POLL_GETEV(&events, i);
1145 pollfd = LTTNG_POLL_GETFD(&events, i);
1146
1147 health_code_update(&health_thread_app_manage);
1148
1149 /* Thread quit pipe has been closed. Killing thread. */
1150 ret = check_thread_quit_pipe(pollfd, revents);
1151 if (ret) {
1152 err = 0;
1153 goto exit;
1154 }
1155
1156 /* Inspect the apps cmd pipe */
1157 if (pollfd == apps_cmd_pipe[0]) {
1158 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1159 ERR("Apps command pipe error");
1160 goto error;
1161 } else if (revents & LPOLLIN) {
1162 /* Empty pipe */
1163 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1164 if (ret < 0 || ret < sizeof(ust_cmd)) {
1165 PERROR("read apps cmd pipe");
1166 goto error;
1167 }
1168
1169 health_code_update(&health_thread_app_manage);
1170
1171 /* Register applicaton to the session daemon */
1172 ret = ust_app_register(&ust_cmd.reg_msg,
1173 ust_cmd.sock);
1174 if (ret == -ENOMEM) {
1175 goto error;
1176 } else if (ret < 0) {
1177 break;
1178 }
1179
1180 health_code_update(&health_thread_app_manage);
1181
1182 /*
1183 * Validate UST version compatibility.
1184 */
1185 ret = ust_app_validate_version(ust_cmd.sock);
1186 if (ret >= 0) {
1187 /*
1188 * Add channel(s) and event(s) to newly registered apps
1189 * from lttng global UST domain.
1190 */
1191 update_ust_app(ust_cmd.sock);
1192 }
1193
1194 health_code_update(&health_thread_app_manage);
1195
1196 ret = ust_app_register_done(ust_cmd.sock);
1197 if (ret < 0) {
1198 /*
1199 * If the registration is not possible, we simply
1200 * unregister the apps and continue
1201 */
1202 ust_app_unregister(ust_cmd.sock);
1203 } else {
1204 /*
1205 * We only monitor the error events of the socket. This
1206 * thread does not handle any incoming data from UST
1207 * (POLLIN).
1208 */
1209 ret = lttng_poll_add(&events, ust_cmd.sock,
1210 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1211 if (ret < 0) {
1212 goto error;
1213 }
1214
1215 /* Set socket timeout for both receiving and ending */
1216 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1217 app_socket_timeout);
1218 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1219 app_socket_timeout);
1220
1221 DBG("Apps with sock %d added to poll set",
1222 ust_cmd.sock);
1223 }
1224
1225 health_code_update(&health_thread_app_manage);
1226
1227 break;
1228 }
1229 } else {
1230 /*
1231 * At this point, we know that a registered application made
1232 * the event at poll_wait.
1233 */
1234 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1235 /* Removing from the poll set */
1236 ret = lttng_poll_del(&events, pollfd);
1237 if (ret < 0) {
1238 goto error;
1239 }
1240
1241 /* Socket closed on remote end. */
1242 ust_app_unregister(pollfd);
1243 break;
1244 }
1245 }
1246
1247 health_code_update(&health_thread_app_manage);
1248 }
1249 }
1250
1251 exit:
1252 error:
1253 lttng_poll_clean(&events);
1254 error_poll_create:
1255 error_testpoint:
1256 utils_close_pipe(apps_cmd_pipe);
1257 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1258
1259 /*
1260 * We don't clean the UST app hash table here since already registered
1261 * applications can still be controlled so let them be until the session
1262 * daemon dies or the applications stop.
1263 */
1264
1265 if (err) {
1266 health_error(&health_thread_app_manage);
1267 ERR("Health error occurred in %s", __func__);
1268 }
1269 health_exit(&health_thread_app_manage);
1270 DBG("Application communication apps thread cleanup complete");
1271 rcu_thread_offline();
1272 rcu_unregister_thread();
1273 return NULL;
1274 }
1275
1276 /*
1277 * Dispatch request from the registration threads to the application
1278 * communication thread.
1279 */
1280 static void *thread_dispatch_ust_registration(void *data)
1281 {
1282 int ret;
1283 struct cds_wfq_node *node;
1284 struct ust_command *ust_cmd = NULL;
1285
1286 DBG("[thread] Dispatch UST command started");
1287
1288 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1289 /* Atomically prepare the queue futex */
1290 futex_nto1_prepare(&ust_cmd_queue.futex);
1291
1292 do {
1293 /* Dequeue command for registration */
1294 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1295 if (node == NULL) {
1296 DBG("Woken up but nothing in the UST command queue");
1297 /* Continue thread execution */
1298 break;
1299 }
1300
1301 ust_cmd = caa_container_of(node, struct ust_command, node);
1302
1303 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1304 " gid:%d sock:%d name:%s (version %d.%d)",
1305 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1306 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1307 ust_cmd->sock, ust_cmd->reg_msg.name,
1308 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1309 /*
1310 * Inform apps thread of the new application registration. This
1311 * call is blocking so we can be assured that the data will be read
1312 * at some point in time or wait to the end of the world :)
1313 */
1314 if (apps_cmd_pipe[1] >= 0) {
1315 ret = write(apps_cmd_pipe[1], ust_cmd,
1316 sizeof(struct ust_command));
1317 if (ret < 0) {
1318 PERROR("write apps cmd pipe");
1319 if (errno == EBADF) {
1320 /*
1321 * We can't inform the application thread to process
1322 * registration. We will exit or else application
1323 * registration will not occur and tracing will never
1324 * start.
1325 */
1326 goto error;
1327 }
1328 }
1329 } else {
1330 /* Application manager thread is not available. */
1331 ret = close(ust_cmd->sock);
1332 if (ret < 0) {
1333 PERROR("close ust_cmd sock");
1334 }
1335 }
1336 free(ust_cmd);
1337 } while (node != NULL);
1338
1339 /* Futex wait on queue. Blocking call on futex() */
1340 futex_nto1_wait(&ust_cmd_queue.futex);
1341 }
1342
1343 error:
1344 DBG("Dispatch thread dying");
1345 return NULL;
1346 }
1347
1348 /*
1349 * This thread manage application registration.
1350 */
1351 static void *thread_registration_apps(void *data)
1352 {
1353 int sock = -1, i, ret, pollfd, err = -1;
1354 uint32_t revents, nb_fd;
1355 struct lttng_poll_event events;
1356 /*
1357 * Get allocated in this thread, enqueued to a global queue, dequeued and
1358 * freed in the manage apps thread.
1359 */
1360 struct ust_command *ust_cmd = NULL;
1361
1362 DBG("[thread] Manage application registration started");
1363
1364 if (testpoint(thread_registration_apps)) {
1365 goto error_testpoint;
1366 }
1367
1368 ret = lttcomm_listen_unix_sock(apps_sock);
1369 if (ret < 0) {
1370 goto error_listen;
1371 }
1372
1373 /*
1374 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1375 * more will be added to this poll set.
1376 */
1377 ret = create_thread_poll_set(&events, 2);
1378 if (ret < 0) {
1379 goto error_create_poll;
1380 }
1381
1382 /* Add the application registration socket */
1383 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1384 if (ret < 0) {
1385 goto error_poll_add;
1386 }
1387
1388 /* Notify all applications to register */
1389 ret = notify_ust_apps(1);
1390 if (ret < 0) {
1391 ERR("Failed to notify applications or create the wait shared memory.\n"
1392 "Execution continues but there might be problem for already\n"
1393 "running applications that wishes to register.");
1394 }
1395
1396 while (1) {
1397 DBG("Accepting application registration");
1398
1399 /* Inifinite blocking call, waiting for transmission */
1400 restart:
1401 health_poll_update(&health_thread_app_reg);
1402 ret = lttng_poll_wait(&events, -1);
1403 health_poll_update(&health_thread_app_reg);
1404 if (ret < 0) {
1405 /*
1406 * Restart interrupted system call.
1407 */
1408 if (errno == EINTR) {
1409 goto restart;
1410 }
1411 goto error;
1412 }
1413
1414 nb_fd = ret;
1415
1416 for (i = 0; i < nb_fd; i++) {
1417 health_code_update(&health_thread_app_reg);
1418
1419 /* Fetch once the poll data */
1420 revents = LTTNG_POLL_GETEV(&events, i);
1421 pollfd = LTTNG_POLL_GETFD(&events, i);
1422
1423 /* Thread quit pipe has been closed. Killing thread. */
1424 ret = check_thread_quit_pipe(pollfd, revents);
1425 if (ret) {
1426 err = 0;
1427 goto exit;
1428 }
1429
1430 /* Event on the registration socket */
1431 if (pollfd == apps_sock) {
1432 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1433 ERR("Register apps socket poll error");
1434 goto error;
1435 } else if (revents & LPOLLIN) {
1436 sock = lttcomm_accept_unix_sock(apps_sock);
1437 if (sock < 0) {
1438 goto error;
1439 }
1440
1441 /*
1442 * Set the CLOEXEC flag. Return code is useless because
1443 * either way, the show must go on.
1444 */
1445 (void) utils_set_fd_cloexec(sock);
1446
1447 /* Create UST registration command for enqueuing */
1448 ust_cmd = zmalloc(sizeof(struct ust_command));
1449 if (ust_cmd == NULL) {
1450 PERROR("ust command zmalloc");
1451 goto error;
1452 }
1453
1454 /*
1455 * Using message-based transmissions to ensure we don't
1456 * have to deal with partially received messages.
1457 */
1458 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1459 if (ret < 0) {
1460 ERR("Exhausted file descriptors allowed for applications.");
1461 free(ust_cmd);
1462 ret = close(sock);
1463 if (ret) {
1464 PERROR("close");
1465 }
1466 sock = -1;
1467 continue;
1468 }
1469 health_code_update(&health_thread_app_reg);
1470 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1471 sizeof(struct ust_register_msg));
1472 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1473 if (ret < 0) {
1474 PERROR("lttcomm_recv_unix_sock register apps");
1475 } else {
1476 ERR("Wrong size received on apps register");
1477 }
1478 free(ust_cmd);
1479 ret = close(sock);
1480 if (ret) {
1481 PERROR("close");
1482 }
1483 lttng_fd_put(LTTNG_FD_APPS, 1);
1484 sock = -1;
1485 continue;
1486 }
1487 health_code_update(&health_thread_app_reg);
1488
1489 ust_cmd->sock = sock;
1490 sock = -1;
1491
1492 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1493 " gid:%d sock:%d name:%s (version %d.%d)",
1494 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1495 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1496 ust_cmd->sock, ust_cmd->reg_msg.name,
1497 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1498
1499 /*
1500 * Lock free enqueue the registration request. The red pill
1501 * has been taken! This apps will be part of the *system*.
1502 */
1503 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1504
1505 /*
1506 * Wake the registration queue futex. Implicit memory
1507 * barrier with the exchange in cds_wfq_enqueue.
1508 */
1509 futex_nto1_wake(&ust_cmd_queue.futex);
1510 }
1511 }
1512 }
1513 }
1514
1515 exit:
1516 error:
1517 if (err) {
1518 health_error(&health_thread_app_reg);
1519 ERR("Health error occurred in %s", __func__);
1520 }
1521
1522 /* Notify that the registration thread is gone */
1523 notify_ust_apps(0);
1524
1525 if (apps_sock >= 0) {
1526 ret = close(apps_sock);
1527 if (ret) {
1528 PERROR("close");
1529 }
1530 }
1531 if (sock >= 0) {
1532 ret = close(sock);
1533 if (ret) {
1534 PERROR("close");
1535 }
1536 lttng_fd_put(LTTNG_FD_APPS, 1);
1537 }
1538 unlink(apps_unix_sock_path);
1539
1540 error_poll_add:
1541 lttng_poll_clean(&events);
1542 error_listen:
1543 error_create_poll:
1544 error_testpoint:
1545 DBG("UST Registration thread cleanup complete");
1546 health_exit(&health_thread_app_reg);
1547
1548 return NULL;
1549 }
1550
1551 /*
1552 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1553 * exec or it will fails.
1554 */
1555 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1556 {
1557 int ret, clock_ret;
1558 struct timespec timeout;
1559
1560 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1561 consumer_data->consumer_thread_is_ready = 0;
1562
1563 /* Setup pthread condition */
1564 ret = pthread_condattr_init(&consumer_data->condattr);
1565 if (ret != 0) {
1566 errno = ret;
1567 PERROR("pthread_condattr_init consumer data");
1568 goto error;
1569 }
1570
1571 /*
1572 * Set the monotonic clock in order to make sure we DO NOT jump in time
1573 * between the clock_gettime() call and the timedwait call. See bug #324
1574 * for a more details and how we noticed it.
1575 */
1576 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1577 if (ret != 0) {
1578 errno = ret;
1579 PERROR("pthread_condattr_setclock consumer data");
1580 goto error;
1581 }
1582
1583 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1584 if (ret != 0) {
1585 errno = ret;
1586 PERROR("pthread_cond_init consumer data");
1587 goto error;
1588 }
1589
1590 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1591 consumer_data);
1592 if (ret != 0) {
1593 PERROR("pthread_create consumer");
1594 ret = -1;
1595 goto error;
1596 }
1597
1598 /* We are about to wait on a pthread condition */
1599 pthread_mutex_lock(&consumer_data->cond_mutex);
1600
1601 /* Get time for sem_timedwait absolute timeout */
1602 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1603 /*
1604 * Set the timeout for the condition timed wait even if the clock gettime
1605 * call fails since we might loop on that call and we want to avoid to
1606 * increment the timeout too many times.
1607 */
1608 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1609
1610 /*
1611 * The following loop COULD be skipped in some conditions so this is why we
1612 * set ret to 0 in order to make sure at least one round of the loop is
1613 * done.
1614 */
1615 ret = 0;
1616
1617 /*
1618 * Loop until the condition is reached or when a timeout is reached. Note
1619 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1620 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1621 * possible. This loop does not take any chances and works with both of
1622 * them.
1623 */
1624 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1625 if (clock_ret < 0) {
1626 PERROR("clock_gettime spawn consumer");
1627 /* Infinite wait for the consumerd thread to be ready */
1628 ret = pthread_cond_wait(&consumer_data->cond,
1629 &consumer_data->cond_mutex);
1630 } else {
1631 ret = pthread_cond_timedwait(&consumer_data->cond,
1632 &consumer_data->cond_mutex, &timeout);
1633 }
1634 }
1635
1636 /* Release the pthread condition */
1637 pthread_mutex_unlock(&consumer_data->cond_mutex);
1638
1639 if (ret != 0) {
1640 errno = ret;
1641 if (ret == ETIMEDOUT) {
1642 /*
1643 * Call has timed out so we kill the kconsumerd_thread and return
1644 * an error.
1645 */
1646 ERR("Condition timed out. The consumer thread was never ready."
1647 " Killing it");
1648 ret = pthread_cancel(consumer_data->thread);
1649 if (ret < 0) {
1650 PERROR("pthread_cancel consumer thread");
1651 }
1652 } else {
1653 PERROR("pthread_cond_wait failed consumer thread");
1654 }
1655 goto error;
1656 }
1657
1658 pthread_mutex_lock(&consumer_data->pid_mutex);
1659 if (consumer_data->pid == 0) {
1660 ERR("Consumerd did not start");
1661 pthread_mutex_unlock(&consumer_data->pid_mutex);
1662 goto error;
1663 }
1664 pthread_mutex_unlock(&consumer_data->pid_mutex);
1665
1666 return 0;
1667
1668 error:
1669 return ret;
1670 }
1671
1672 /*
1673 * Join consumer thread
1674 */
1675 static int join_consumer_thread(struct consumer_data *consumer_data)
1676 {
1677 void *status;
1678
1679 /* Consumer pid must be a real one. */
1680 if (consumer_data->pid > 0) {
1681 int ret;
1682 ret = kill(consumer_data->pid, SIGTERM);
1683 if (ret) {
1684 ERR("Error killing consumer daemon");
1685 return ret;
1686 }
1687 return pthread_join(consumer_data->thread, &status);
1688 } else {
1689 return 0;
1690 }
1691 }
1692
1693 /*
1694 * Fork and exec a consumer daemon (consumerd).
1695 *
1696 * Return pid if successful else -1.
1697 */
1698 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1699 {
1700 int ret;
1701 pid_t pid;
1702 const char *consumer_to_use;
1703 const char *verbosity;
1704 struct stat st;
1705
1706 DBG("Spawning consumerd");
1707
1708 pid = fork();
1709 if (pid == 0) {
1710 /*
1711 * Exec consumerd.
1712 */
1713 if (opt_verbose_consumer) {
1714 verbosity = "--verbose";
1715 } else {
1716 verbosity = "--quiet";
1717 }
1718 switch (consumer_data->type) {
1719 case LTTNG_CONSUMER_KERNEL:
1720 /*
1721 * Find out which consumerd to execute. We will first try the
1722 * 64-bit path, then the sessiond's installation directory, and
1723 * fallback on the 32-bit one,
1724 */
1725 DBG3("Looking for a kernel consumer at these locations:");
1726 DBG3(" 1) %s", consumerd64_bin);
1727 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1728 DBG3(" 3) %s", consumerd32_bin);
1729 if (stat(consumerd64_bin, &st) == 0) {
1730 DBG3("Found location #1");
1731 consumer_to_use = consumerd64_bin;
1732 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1733 DBG3("Found location #2");
1734 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1735 } else if (stat(consumerd32_bin, &st) == 0) {
1736 DBG3("Found location #3");
1737 consumer_to_use = consumerd32_bin;
1738 } else {
1739 DBG("Could not find any valid consumerd executable");
1740 break;
1741 }
1742 DBG("Using kernel consumer at: %s", consumer_to_use);
1743 execl(consumer_to_use,
1744 "lttng-consumerd", verbosity, "-k",
1745 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1746 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1747 NULL);
1748 break;
1749 case LTTNG_CONSUMER64_UST:
1750 {
1751 char *tmpnew = NULL;
1752
1753 if (consumerd64_libdir[0] != '\0') {
1754 char *tmp;
1755 size_t tmplen;
1756
1757 tmp = getenv("LD_LIBRARY_PATH");
1758 if (!tmp) {
1759 tmp = "";
1760 }
1761 tmplen = strlen("LD_LIBRARY_PATH=")
1762 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1763 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1764 if (!tmpnew) {
1765 ret = -ENOMEM;
1766 goto error;
1767 }
1768 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1769 strcat(tmpnew, consumerd64_libdir);
1770 if (tmp[0] != '\0') {
1771 strcat(tmpnew, ":");
1772 strcat(tmpnew, tmp);
1773 }
1774 ret = putenv(tmpnew);
1775 if (ret) {
1776 ret = -errno;
1777 goto error;
1778 }
1779 }
1780 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1781 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1782 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1783 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1784 NULL);
1785 if (consumerd64_libdir[0] != '\0') {
1786 free(tmpnew);
1787 }
1788 if (ret) {
1789 goto error;
1790 }
1791 break;
1792 }
1793 case LTTNG_CONSUMER32_UST:
1794 {
1795 char *tmpnew = NULL;
1796
1797 if (consumerd32_libdir[0] != '\0') {
1798 char *tmp;
1799 size_t tmplen;
1800
1801 tmp = getenv("LD_LIBRARY_PATH");
1802 if (!tmp) {
1803 tmp = "";
1804 }
1805 tmplen = strlen("LD_LIBRARY_PATH=")
1806 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1807 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1808 if (!tmpnew) {
1809 ret = -ENOMEM;
1810 goto error;
1811 }
1812 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1813 strcat(tmpnew, consumerd32_libdir);
1814 if (tmp[0] != '\0') {
1815 strcat(tmpnew, ":");
1816 strcat(tmpnew, tmp);
1817 }
1818 ret = putenv(tmpnew);
1819 if (ret) {
1820 ret = -errno;
1821 goto error;
1822 }
1823 }
1824 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1825 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1826 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1827 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1828 NULL);
1829 if (consumerd32_libdir[0] != '\0') {
1830 free(tmpnew);
1831 }
1832 if (ret) {
1833 goto error;
1834 }
1835 break;
1836 }
1837 default:
1838 PERROR("unknown consumer type");
1839 exit(EXIT_FAILURE);
1840 }
1841 if (errno != 0) {
1842 PERROR("kernel start consumer exec");
1843 }
1844 exit(EXIT_FAILURE);
1845 } else if (pid > 0) {
1846 ret = pid;
1847 } else {
1848 PERROR("start consumer fork");
1849 ret = -errno;
1850 }
1851 error:
1852 return ret;
1853 }
1854
1855 /*
1856 * Spawn the consumerd daemon and session daemon thread.
1857 */
1858 static int start_consumerd(struct consumer_data *consumer_data)
1859 {
1860 int ret;
1861
1862 /*
1863 * Set the listen() state on the socket since there is a possible race
1864 * between the exec() of the consumer daemon and this call if place in the
1865 * consumer thread. See bug #366 for more details.
1866 */
1867 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1868 if (ret < 0) {
1869 goto error;
1870 }
1871
1872 pthread_mutex_lock(&consumer_data->pid_mutex);
1873 if (consumer_data->pid != 0) {
1874 pthread_mutex_unlock(&consumer_data->pid_mutex);
1875 goto end;
1876 }
1877
1878 ret = spawn_consumerd(consumer_data);
1879 if (ret < 0) {
1880 ERR("Spawning consumerd failed");
1881 pthread_mutex_unlock(&consumer_data->pid_mutex);
1882 goto error;
1883 }
1884
1885 /* Setting up the consumer_data pid */
1886 consumer_data->pid = ret;
1887 DBG2("Consumer pid %d", consumer_data->pid);
1888 pthread_mutex_unlock(&consumer_data->pid_mutex);
1889
1890 DBG2("Spawning consumer control thread");
1891 ret = spawn_consumer_thread(consumer_data);
1892 if (ret < 0) {
1893 ERR("Fatal error spawning consumer control thread");
1894 goto error;
1895 }
1896
1897 end:
1898 return 0;
1899
1900 error:
1901 /* Cleanup already created socket on error. */
1902 if (consumer_data->err_sock >= 0) {
1903 int err;
1904
1905 err = close(consumer_data->err_sock);
1906 if (err < 0) {
1907 PERROR("close consumer data error socket");
1908 }
1909 }
1910 return ret;
1911 }
1912
1913 /*
1914 * Compute health status of each consumer. If one of them is zero (bad
1915 * state), we return 0.
1916 */
1917 static int check_consumer_health(void)
1918 {
1919 int ret;
1920
1921 ret = health_check_state(&kconsumer_data.health) &&
1922 health_check_state(&ustconsumer32_data.health) &&
1923 health_check_state(&ustconsumer64_data.health);
1924
1925 DBG3("Health consumer check %d", ret);
1926
1927 return ret;
1928 }
1929
1930 /*
1931 * Setup necessary data for kernel tracer action.
1932 */
1933 static int init_kernel_tracer(void)
1934 {
1935 int ret;
1936
1937 /* Modprobe lttng kernel modules */
1938 ret = modprobe_lttng_control();
1939 if (ret < 0) {
1940 goto error;
1941 }
1942
1943 /* Open debugfs lttng */
1944 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1945 if (kernel_tracer_fd < 0) {
1946 DBG("Failed to open %s", module_proc_lttng);
1947 ret = -1;
1948 goto error_open;
1949 }
1950
1951 /* Validate kernel version */
1952 ret = kernel_validate_version(kernel_tracer_fd);
1953 if (ret < 0) {
1954 goto error_version;
1955 }
1956
1957 ret = modprobe_lttng_data();
1958 if (ret < 0) {
1959 goto error_modules;
1960 }
1961
1962 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1963 return 0;
1964
1965 error_version:
1966 modprobe_remove_lttng_control();
1967 ret = close(kernel_tracer_fd);
1968 if (ret) {
1969 PERROR("close");
1970 }
1971 kernel_tracer_fd = -1;
1972 return LTTNG_ERR_KERN_VERSION;
1973
1974 error_modules:
1975 ret = close(kernel_tracer_fd);
1976 if (ret) {
1977 PERROR("close");
1978 }
1979
1980 error_open:
1981 modprobe_remove_lttng_control();
1982
1983 error:
1984 WARN("No kernel tracer available");
1985 kernel_tracer_fd = -1;
1986 if (!is_root) {
1987 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1988 } else {
1989 return LTTNG_ERR_KERN_NA;
1990 }
1991 }
1992
1993
1994 /*
1995 * Copy consumer output from the tracing session to the domain session. The
1996 * function also applies the right modification on a per domain basis for the
1997 * trace files destination directory.
1998 */
1999 static int copy_session_consumer(int domain, struct ltt_session *session)
2000 {
2001 int ret;
2002 const char *dir_name;
2003 struct consumer_output *consumer;
2004
2005 assert(session);
2006 assert(session->consumer);
2007
2008 switch (domain) {
2009 case LTTNG_DOMAIN_KERNEL:
2010 DBG3("Copying tracing session consumer output in kernel session");
2011 /*
2012 * XXX: We should audit the session creation and what this function
2013 * does "extra" in order to avoid a destroy since this function is used
2014 * in the domain session creation (kernel and ust) only. Same for UST
2015 * domain.
2016 */
2017 if (session->kernel_session->consumer) {
2018 consumer_destroy_output(session->kernel_session->consumer);
2019 }
2020 session->kernel_session->consumer =
2021 consumer_copy_output(session->consumer);
2022 /* Ease our life a bit for the next part */
2023 consumer = session->kernel_session->consumer;
2024 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2025 break;
2026 case LTTNG_DOMAIN_UST:
2027 DBG3("Copying tracing session consumer output in UST session");
2028 if (session->ust_session->consumer) {
2029 consumer_destroy_output(session->ust_session->consumer);
2030 }
2031 session->ust_session->consumer =
2032 consumer_copy_output(session->consumer);
2033 /* Ease our life a bit for the next part */
2034 consumer = session->ust_session->consumer;
2035 dir_name = DEFAULT_UST_TRACE_DIR;
2036 break;
2037 default:
2038 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2039 goto error;
2040 }
2041
2042 /* Append correct directory to subdir */
2043 strncat(consumer->subdir, dir_name,
2044 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2045 DBG3("Copy session consumer subdir %s", consumer->subdir);
2046
2047 ret = LTTNG_OK;
2048
2049 error:
2050 return ret;
2051 }
2052
2053 /*
2054 * Create an UST session and add it to the session ust list.
2055 */
2056 static int create_ust_session(struct ltt_session *session,
2057 struct lttng_domain *domain)
2058 {
2059 int ret;
2060 struct ltt_ust_session *lus = NULL;
2061
2062 assert(session);
2063 assert(domain);
2064 assert(session->consumer);
2065
2066 switch (domain->type) {
2067 case LTTNG_DOMAIN_UST:
2068 break;
2069 default:
2070 ERR("Unknown UST domain on create session %d", domain->type);
2071 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2072 goto error;
2073 }
2074
2075 DBG("Creating UST session");
2076
2077 lus = trace_ust_create_session(session->path, session->id, domain);
2078 if (lus == NULL) {
2079 ret = LTTNG_ERR_UST_SESS_FAIL;
2080 goto error;
2081 }
2082
2083 lus->uid = session->uid;
2084 lus->gid = session->gid;
2085 session->ust_session = lus;
2086
2087 /* Copy session output to the newly created UST session */
2088 ret = copy_session_consumer(domain->type, session);
2089 if (ret != LTTNG_OK) {
2090 goto error;
2091 }
2092
2093 return LTTNG_OK;
2094
2095 error:
2096 free(lus);
2097 session->ust_session = NULL;
2098 return ret;
2099 }
2100
2101 /*
2102 * Create a kernel tracer session then create the default channel.
2103 */
2104 static int create_kernel_session(struct ltt_session *session)
2105 {
2106 int ret;
2107
2108 DBG("Creating kernel session");
2109
2110 ret = kernel_create_session(session, kernel_tracer_fd);
2111 if (ret < 0) {
2112 ret = LTTNG_ERR_KERN_SESS_FAIL;
2113 goto error;
2114 }
2115
2116 /* Code flow safety */
2117 assert(session->kernel_session);
2118
2119 /* Copy session output to the newly created Kernel session */
2120 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2121 if (ret != LTTNG_OK) {
2122 goto error;
2123 }
2124
2125 /* Create directory(ies) on local filesystem. */
2126 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2127 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2128 ret = run_as_mkdir_recursive(
2129 session->kernel_session->consumer->dst.trace_path,
2130 S_IRWXU | S_IRWXG, session->uid, session->gid);
2131 if (ret < 0) {
2132 if (ret != -EEXIST) {
2133 ERR("Trace directory creation error");
2134 goto error;
2135 }
2136 }
2137 }
2138
2139 session->kernel_session->uid = session->uid;
2140 session->kernel_session->gid = session->gid;
2141
2142 return LTTNG_OK;
2143
2144 error:
2145 trace_kernel_destroy_session(session->kernel_session);
2146 session->kernel_session = NULL;
2147 return ret;
2148 }
2149
2150 /*
2151 * Count number of session permitted by uid/gid.
2152 */
2153 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2154 {
2155 unsigned int i = 0;
2156 struct ltt_session *session;
2157
2158 DBG("Counting number of available session for UID %d GID %d",
2159 uid, gid);
2160 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2161 /*
2162 * Only list the sessions the user can control.
2163 */
2164 if (!session_access_ok(session, uid, gid)) {
2165 continue;
2166 }
2167 i++;
2168 }
2169 return i;
2170 }
2171
2172 /*
2173 * Process the command requested by the lttng client within the command
2174 * context structure. This function make sure that the return structure (llm)
2175 * is set and ready for transmission before returning.
2176 *
2177 * Return any error encountered or 0 for success.
2178 *
2179 * "sock" is only used for special-case var. len data.
2180 */
2181 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2182 int *sock_error)
2183 {
2184 int ret = LTTNG_OK;
2185 int need_tracing_session = 1;
2186 int need_domain;
2187
2188 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2189
2190 *sock_error = 0;
2191
2192 switch (cmd_ctx->lsm->cmd_type) {
2193 case LTTNG_CREATE_SESSION:
2194 case LTTNG_DESTROY_SESSION:
2195 case LTTNG_LIST_SESSIONS:
2196 case LTTNG_LIST_DOMAINS:
2197 case LTTNG_START_TRACE:
2198 case LTTNG_STOP_TRACE:
2199 case LTTNG_DATA_PENDING:
2200 need_domain = 0;
2201 break;
2202 default:
2203 need_domain = 1;
2204 }
2205
2206 if (opt_no_kernel && need_domain
2207 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2208 if (!is_root) {
2209 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2210 } else {
2211 ret = LTTNG_ERR_KERN_NA;
2212 }
2213 goto error;
2214 }
2215
2216 /* Deny register consumer if we already have a spawned consumer. */
2217 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2218 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2219 if (kconsumer_data.pid > 0) {
2220 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2221 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2222 goto error;
2223 }
2224 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2225 }
2226
2227 /*
2228 * Check for command that don't needs to allocate a returned payload. We do
2229 * this here so we don't have to make the call for no payload at each
2230 * command.
2231 */
2232 switch(cmd_ctx->lsm->cmd_type) {
2233 case LTTNG_LIST_SESSIONS:
2234 case LTTNG_LIST_TRACEPOINTS:
2235 case LTTNG_LIST_TRACEPOINT_FIELDS:
2236 case LTTNG_LIST_DOMAINS:
2237 case LTTNG_LIST_CHANNELS:
2238 case LTTNG_LIST_EVENTS:
2239 break;
2240 default:
2241 /* Setup lttng message with no payload */
2242 ret = setup_lttng_msg(cmd_ctx, 0);
2243 if (ret < 0) {
2244 /* This label does not try to unlock the session */
2245 goto init_setup_error;
2246 }
2247 }
2248
2249 /* Commands that DO NOT need a session. */
2250 switch (cmd_ctx->lsm->cmd_type) {
2251 case LTTNG_CREATE_SESSION:
2252 case LTTNG_CALIBRATE:
2253 case LTTNG_LIST_SESSIONS:
2254 case LTTNG_LIST_TRACEPOINTS:
2255 case LTTNG_LIST_TRACEPOINT_FIELDS:
2256 need_tracing_session = 0;
2257 break;
2258 default:
2259 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2260 /*
2261 * We keep the session list lock across _all_ commands
2262 * for now, because the per-session lock does not
2263 * handle teardown properly.
2264 */
2265 session_lock_list();
2266 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2267 if (cmd_ctx->session == NULL) {
2268 if (cmd_ctx->lsm->session.name != NULL) {
2269 ret = LTTNG_ERR_SESS_NOT_FOUND;
2270 } else {
2271 /* If no session name specified */
2272 ret = LTTNG_ERR_SELECT_SESS;
2273 }
2274 goto error;
2275 } else {
2276 /* Acquire lock for the session */
2277 session_lock(cmd_ctx->session);
2278 }
2279 break;
2280 }
2281
2282 if (!need_domain) {
2283 goto skip_domain;
2284 }
2285
2286 /*
2287 * Check domain type for specific "pre-action".
2288 */
2289 switch (cmd_ctx->lsm->domain.type) {
2290 case LTTNG_DOMAIN_KERNEL:
2291 if (!is_root) {
2292 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2293 goto error;
2294 }
2295
2296 /* Kernel tracer check */
2297 if (kernel_tracer_fd == -1) {
2298 /* Basically, load kernel tracer modules */
2299 ret = init_kernel_tracer();
2300 if (ret != 0) {
2301 goto error;
2302 }
2303 }
2304
2305 /* Consumer is in an ERROR state. Report back to client */
2306 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2307 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2308 goto error;
2309 }
2310
2311 /* Need a session for kernel command */
2312 if (need_tracing_session) {
2313 if (cmd_ctx->session->kernel_session == NULL) {
2314 ret = create_kernel_session(cmd_ctx->session);
2315 if (ret < 0) {
2316 ret = LTTNG_ERR_KERN_SESS_FAIL;
2317 goto error;
2318 }
2319 }
2320
2321 /* Start the kernel consumer daemon */
2322 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2323 if (kconsumer_data.pid == 0 &&
2324 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2325 cmd_ctx->session->start_consumer) {
2326 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2327 ret = start_consumerd(&kconsumer_data);
2328 if (ret < 0) {
2329 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2330 goto error;
2331 }
2332 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2333 } else {
2334 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2335 }
2336
2337 /*
2338 * The consumer was just spawned so we need to add the socket to
2339 * the consumer output of the session if exist.
2340 */
2341 ret = consumer_create_socket(&kconsumer_data,
2342 cmd_ctx->session->kernel_session->consumer);
2343 if (ret < 0) {
2344 goto error;
2345 }
2346 }
2347
2348 break;
2349 case LTTNG_DOMAIN_UST:
2350 {
2351 /* Consumer is in an ERROR state. Report back to client */
2352 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2353 ret = LTTNG_ERR_NO_USTCONSUMERD;
2354 goto error;
2355 }
2356
2357 if (need_tracing_session) {
2358 /* Create UST session if none exist. */
2359 if (cmd_ctx->session->ust_session == NULL) {
2360 ret = create_ust_session(cmd_ctx->session,
2361 &cmd_ctx->lsm->domain);
2362 if (ret != LTTNG_OK) {
2363 goto error;
2364 }
2365 }
2366
2367 /* Start the UST consumer daemons */
2368 /* 64-bit */
2369 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2370 if (consumerd64_bin[0] != '\0' &&
2371 ustconsumer64_data.pid == 0 &&
2372 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2373 cmd_ctx->session->start_consumer) {
2374 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2375 ret = start_consumerd(&ustconsumer64_data);
2376 if (ret < 0) {
2377 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2378 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2379 goto error;
2380 }
2381
2382 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2383 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2384 } else {
2385 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2386 }
2387
2388 /*
2389 * Setup socket for consumer 64 bit. No need for atomic access
2390 * since it was set above and can ONLY be set in this thread.
2391 */
2392 ret = consumer_create_socket(&ustconsumer64_data,
2393 cmd_ctx->session->ust_session->consumer);
2394 if (ret < 0) {
2395 goto error;
2396 }
2397
2398 /* 32-bit */
2399 if (consumerd32_bin[0] != '\0' &&
2400 ustconsumer32_data.pid == 0 &&
2401 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2402 cmd_ctx->session->start_consumer) {
2403 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2404 ret = start_consumerd(&ustconsumer32_data);
2405 if (ret < 0) {
2406 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2407 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2408 goto error;
2409 }
2410
2411 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2412 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2413 } else {
2414 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2415 }
2416
2417 /*
2418 * Setup socket for consumer 64 bit. No need for atomic access
2419 * since it was set above and can ONLY be set in this thread.
2420 */
2421 ret = consumer_create_socket(&ustconsumer32_data,
2422 cmd_ctx->session->ust_session->consumer);
2423 if (ret < 0) {
2424 goto error;
2425 }
2426 }
2427 break;
2428 }
2429 default:
2430 break;
2431 }
2432 skip_domain:
2433
2434 /* Validate consumer daemon state when start/stop trace command */
2435 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2436 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2437 switch (cmd_ctx->lsm->domain.type) {
2438 case LTTNG_DOMAIN_UST:
2439 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2440 ret = LTTNG_ERR_NO_USTCONSUMERD;
2441 goto error;
2442 }
2443 break;
2444 case LTTNG_DOMAIN_KERNEL:
2445 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2446 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2447 goto error;
2448 }
2449 break;
2450 }
2451 }
2452
2453 /*
2454 * Check that the UID or GID match that of the tracing session.
2455 * The root user can interact with all sessions.
2456 */
2457 if (need_tracing_session) {
2458 if (!session_access_ok(cmd_ctx->session,
2459 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2460 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2461 ret = LTTNG_ERR_EPERM;
2462 goto error;
2463 }
2464 }
2465
2466 /* Process by command type */
2467 switch (cmd_ctx->lsm->cmd_type) {
2468 case LTTNG_ADD_CONTEXT:
2469 {
2470 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2471 cmd_ctx->lsm->u.context.channel_name,
2472 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2473 break;
2474 }
2475 case LTTNG_DISABLE_CHANNEL:
2476 {
2477 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2478 cmd_ctx->lsm->u.disable.channel_name);
2479 break;
2480 }
2481 case LTTNG_DISABLE_EVENT:
2482 {
2483 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2484 cmd_ctx->lsm->u.disable.channel_name,
2485 cmd_ctx->lsm->u.disable.name);
2486 break;
2487 }
2488 case LTTNG_DISABLE_ALL_EVENT:
2489 {
2490 DBG("Disabling all events");
2491
2492 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2493 cmd_ctx->lsm->u.disable.channel_name);
2494 break;
2495 }
2496 case LTTNG_DISABLE_CONSUMER:
2497 {
2498 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2499 break;
2500 }
2501 case LTTNG_ENABLE_CHANNEL:
2502 {
2503 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2504 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2505 break;
2506 }
2507 case LTTNG_ENABLE_CONSUMER:
2508 {
2509 /*
2510 * XXX: 0 means that this URI should be applied on the session. Should
2511 * be a DOMAIN enuam.
2512 */
2513 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2514 if (ret != LTTNG_OK) {
2515 goto error;
2516 }
2517
2518 if (cmd_ctx->lsm->domain.type == 0) {
2519 /* Add the URI for the UST session if a consumer is present. */
2520 if (cmd_ctx->session->ust_session &&
2521 cmd_ctx->session->ust_session->consumer) {
2522 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2523 } else if (cmd_ctx->session->kernel_session &&
2524 cmd_ctx->session->kernel_session->consumer) {
2525 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2526 cmd_ctx->session);
2527 }
2528 }
2529 break;
2530 }
2531 case LTTNG_ENABLE_EVENT:
2532 {
2533 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2534 cmd_ctx->lsm->u.enable.channel_name,
2535 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2536 break;
2537 }
2538 case LTTNG_ENABLE_ALL_EVENT:
2539 {
2540 DBG("Enabling all events");
2541
2542 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2543 cmd_ctx->lsm->u.enable.channel_name,
2544 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2545 break;
2546 }
2547 case LTTNG_LIST_TRACEPOINTS:
2548 {
2549 struct lttng_event *events;
2550 ssize_t nb_events;
2551
2552 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2553 if (nb_events < 0) {
2554 /* Return value is a negative lttng_error_code. */
2555 ret = -nb_events;
2556 goto error;
2557 }
2558
2559 /*
2560 * Setup lttng message with payload size set to the event list size in
2561 * bytes and then copy list into the llm payload.
2562 */
2563 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2564 if (ret < 0) {
2565 free(events);
2566 goto setup_error;
2567 }
2568
2569 /* Copy event list into message payload */
2570 memcpy(cmd_ctx->llm->payload, events,
2571 sizeof(struct lttng_event) * nb_events);
2572
2573 free(events);
2574
2575 ret = LTTNG_OK;
2576 break;
2577 }
2578 case LTTNG_LIST_TRACEPOINT_FIELDS:
2579 {
2580 struct lttng_event_field *fields;
2581 ssize_t nb_fields;
2582
2583 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2584 &fields);
2585 if (nb_fields < 0) {
2586 /* Return value is a negative lttng_error_code. */
2587 ret = -nb_fields;
2588 goto error;
2589 }
2590
2591 /*
2592 * Setup lttng message with payload size set to the event list size in
2593 * bytes and then copy list into the llm payload.
2594 */
2595 ret = setup_lttng_msg(cmd_ctx,
2596 sizeof(struct lttng_event_field) * nb_fields);
2597 if (ret < 0) {
2598 free(fields);
2599 goto setup_error;
2600 }
2601
2602 /* Copy event list into message payload */
2603 memcpy(cmd_ctx->llm->payload, fields,
2604 sizeof(struct lttng_event_field) * nb_fields);
2605
2606 free(fields);
2607
2608 ret = LTTNG_OK;
2609 break;
2610 }
2611 case LTTNG_SET_CONSUMER_URI:
2612 {
2613 size_t nb_uri, len;
2614 struct lttng_uri *uris;
2615
2616 nb_uri = cmd_ctx->lsm->u.uri.size;
2617 len = nb_uri * sizeof(struct lttng_uri);
2618
2619 if (nb_uri == 0) {
2620 ret = LTTNG_ERR_INVALID;
2621 goto error;
2622 }
2623
2624 uris = zmalloc(len);
2625 if (uris == NULL) {
2626 ret = LTTNG_ERR_FATAL;
2627 goto error;
2628 }
2629
2630 /* Receive variable len data */
2631 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2632 ret = lttcomm_recv_unix_sock(sock, uris, len);
2633 if (ret <= 0) {
2634 DBG("No URIs received from client... continuing");
2635 *sock_error = 1;
2636 ret = LTTNG_ERR_SESSION_FAIL;
2637 free(uris);
2638 goto error;
2639 }
2640
2641 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2642 nb_uri, uris);
2643 if (ret != LTTNG_OK) {
2644 free(uris);
2645 goto error;
2646 }
2647
2648 /*
2649 * XXX: 0 means that this URI should be applied on the session. Should
2650 * be a DOMAIN enuam.
2651 */
2652 if (cmd_ctx->lsm->domain.type == 0) {
2653 /* Add the URI for the UST session if a consumer is present. */
2654 if (cmd_ctx->session->ust_session &&
2655 cmd_ctx->session->ust_session->consumer) {
2656 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2657 nb_uri, uris);
2658 } else if (cmd_ctx->session->kernel_session &&
2659 cmd_ctx->session->kernel_session->consumer) {
2660 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2661 cmd_ctx->session, nb_uri, uris);
2662 }
2663 }
2664
2665 free(uris);
2666
2667 break;
2668 }
2669 case LTTNG_START_TRACE:
2670 {
2671 ret = cmd_start_trace(cmd_ctx->session);
2672 break;
2673 }
2674 case LTTNG_STOP_TRACE:
2675 {
2676 ret = cmd_stop_trace(cmd_ctx->session);
2677 break;
2678 }
2679 case LTTNG_CREATE_SESSION:
2680 {
2681 size_t nb_uri, len;
2682 struct lttng_uri *uris = NULL;
2683
2684 nb_uri = cmd_ctx->lsm->u.uri.size;
2685 len = nb_uri * sizeof(struct lttng_uri);
2686
2687 if (nb_uri > 0) {
2688 uris = zmalloc(len);
2689 if (uris == NULL) {
2690 ret = LTTNG_ERR_FATAL;
2691 goto error;
2692 }
2693
2694 /* Receive variable len data */
2695 DBG("Waiting for %zu URIs from client ...", nb_uri);
2696 ret = lttcomm_recv_unix_sock(sock, uris, len);
2697 if (ret <= 0) {
2698 DBG("No URIs received from client... continuing");
2699 *sock_error = 1;
2700 ret = LTTNG_ERR_SESSION_FAIL;
2701 free(uris);
2702 goto error;
2703 }
2704
2705 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2706 DBG("Creating session with ONE network URI is a bad call");
2707 ret = LTTNG_ERR_SESSION_FAIL;
2708 free(uris);
2709 goto error;
2710 }
2711 }
2712
2713 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2714 &cmd_ctx->creds);
2715
2716 free(uris);
2717
2718 break;
2719 }
2720 case LTTNG_DESTROY_SESSION:
2721 {
2722 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2723
2724 /* Set session to NULL so we do not unlock it after free. */
2725 cmd_ctx->session = NULL;
2726 break;
2727 }
2728 case LTTNG_LIST_DOMAINS:
2729 {
2730 ssize_t nb_dom;
2731 struct lttng_domain *domains;
2732
2733 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2734 if (nb_dom < 0) {
2735 /* Return value is a negative lttng_error_code. */
2736 ret = -nb_dom;
2737 goto error;
2738 }
2739
2740 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2741 if (ret < 0) {
2742 goto setup_error;
2743 }
2744
2745 /* Copy event list into message payload */
2746 memcpy(cmd_ctx->llm->payload, domains,
2747 nb_dom * sizeof(struct lttng_domain));
2748
2749 free(domains);
2750
2751 ret = LTTNG_OK;
2752 break;
2753 }
2754 case LTTNG_LIST_CHANNELS:
2755 {
2756 int nb_chan;
2757 struct lttng_channel *channels;
2758
2759 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2760 cmd_ctx->session, &channels);
2761 if (nb_chan < 0) {
2762 /* Return value is a negative lttng_error_code. */
2763 ret = -nb_chan;
2764 goto error;
2765 }
2766
2767 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2768 if (ret < 0) {
2769 goto setup_error;
2770 }
2771
2772 /* Copy event list into message payload */
2773 memcpy(cmd_ctx->llm->payload, channels,
2774 nb_chan * sizeof(struct lttng_channel));
2775
2776 free(channels);
2777
2778 ret = LTTNG_OK;
2779 break;
2780 }
2781 case LTTNG_LIST_EVENTS:
2782 {
2783 ssize_t nb_event;
2784 struct lttng_event *events = NULL;
2785
2786 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2787 cmd_ctx->lsm->u.list.channel_name, &events);
2788 if (nb_event < 0) {
2789 /* Return value is a negative lttng_error_code. */
2790 ret = -nb_event;
2791 goto error;
2792 }
2793
2794 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2795 if (ret < 0) {
2796 goto setup_error;
2797 }
2798
2799 /* Copy event list into message payload */
2800 memcpy(cmd_ctx->llm->payload, events,
2801 nb_event * sizeof(struct lttng_event));
2802
2803 free(events);
2804
2805 ret = LTTNG_OK;
2806 break;
2807 }
2808 case LTTNG_LIST_SESSIONS:
2809 {
2810 unsigned int nr_sessions;
2811
2812 session_lock_list();
2813 nr_sessions = lttng_sessions_count(
2814 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2815 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2816
2817 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2818 if (ret < 0) {
2819 session_unlock_list();
2820 goto setup_error;
2821 }
2822
2823 /* Filled the session array */
2824 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2825 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2826 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2827
2828 session_unlock_list();
2829
2830 ret = LTTNG_OK;
2831 break;
2832 }
2833 case LTTNG_CALIBRATE:
2834 {
2835 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2836 &cmd_ctx->lsm->u.calibrate);
2837 break;
2838 }
2839 case LTTNG_REGISTER_CONSUMER:
2840 {
2841 struct consumer_data *cdata;
2842
2843 switch (cmd_ctx->lsm->domain.type) {
2844 case LTTNG_DOMAIN_KERNEL:
2845 cdata = &kconsumer_data;
2846 break;
2847 default:
2848 ret = LTTNG_ERR_UND;
2849 goto error;
2850 }
2851
2852 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2853 cmd_ctx->lsm->u.reg.path, cdata);
2854 break;
2855 }
2856 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2857 {
2858 struct lttng_filter_bytecode *bytecode;
2859
2860 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2861 ret = LTTNG_ERR_FILTER_INVAL;
2862 goto error;
2863 }
2864 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2865 ret = LTTNG_ERR_FILTER_INVAL;
2866 goto error;
2867 }
2868 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2869 if (!bytecode) {
2870 ret = LTTNG_ERR_FILTER_NOMEM;
2871 goto error;
2872 }
2873 /* Receive var. len. data */
2874 DBG("Receiving var len data from client ...");
2875 ret = lttcomm_recv_unix_sock(sock, bytecode,
2876 cmd_ctx->lsm->u.enable.bytecode_len);
2877 if (ret <= 0) {
2878 DBG("Nothing recv() from client var len data... continuing");
2879 *sock_error = 1;
2880 ret = LTTNG_ERR_FILTER_INVAL;
2881 goto error;
2882 }
2883
2884 if (bytecode->len + sizeof(*bytecode)
2885 != cmd_ctx->lsm->u.enable.bytecode_len) {
2886 free(bytecode);
2887 ret = LTTNG_ERR_FILTER_INVAL;
2888 goto error;
2889 }
2890
2891 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2892 cmd_ctx->lsm->u.enable.channel_name,
2893 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2894 break;
2895 }
2896 case LTTNG_DATA_PENDING:
2897 {
2898 ret = cmd_data_pending(cmd_ctx->session);
2899 break;
2900 }
2901 default:
2902 ret = LTTNG_ERR_UND;
2903 break;
2904 }
2905
2906 error:
2907 if (cmd_ctx->llm == NULL) {
2908 DBG("Missing llm structure. Allocating one.");
2909 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2910 goto setup_error;
2911 }
2912 }
2913 /* Set return code */
2914 cmd_ctx->llm->ret_code = ret;
2915 setup_error:
2916 if (cmd_ctx->session) {
2917 session_unlock(cmd_ctx->session);
2918 }
2919 if (need_tracing_session) {
2920 session_unlock_list();
2921 }
2922 init_setup_error:
2923 return ret;
2924 }
2925
2926 /*
2927 * Thread managing health check socket.
2928 */
2929 static void *thread_manage_health(void *data)
2930 {
2931 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2932 uint32_t revents, nb_fd;
2933 struct lttng_poll_event events;
2934 struct lttcomm_health_msg msg;
2935 struct lttcomm_health_data reply;
2936
2937 DBG("[thread] Manage health check started");
2938
2939 rcu_register_thread();
2940
2941 /* Create unix socket */
2942 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2943 if (sock < 0) {
2944 ERR("Unable to create health check Unix socket");
2945 ret = -1;
2946 goto error;
2947 }
2948
2949 /*
2950 * Set the CLOEXEC flag. Return code is useless because either way, the
2951 * show must go on.
2952 */
2953 (void) utils_set_fd_cloexec(sock);
2954
2955 ret = lttcomm_listen_unix_sock(sock);
2956 if (ret < 0) {
2957 goto error;
2958 }
2959
2960 /*
2961 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2962 * more will be added to this poll set.
2963 */
2964 ret = create_thread_poll_set(&events, 2);
2965 if (ret < 0) {
2966 goto error;
2967 }
2968
2969 /* Add the application registration socket */
2970 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2971 if (ret < 0) {
2972 goto error;
2973 }
2974
2975 while (1) {
2976 DBG("Health check ready");
2977
2978 /* Inifinite blocking call, waiting for transmission */
2979 restart:
2980 ret = lttng_poll_wait(&events, -1);
2981 if (ret < 0) {
2982 /*
2983 * Restart interrupted system call.
2984 */
2985 if (errno == EINTR) {
2986 goto restart;
2987 }
2988 goto error;
2989 }
2990
2991 nb_fd = ret;
2992
2993 for (i = 0; i < nb_fd; i++) {
2994 /* Fetch once the poll data */
2995 revents = LTTNG_POLL_GETEV(&events, i);
2996 pollfd = LTTNG_POLL_GETFD(&events, i);
2997
2998 /* Thread quit pipe has been closed. Killing thread. */
2999 ret = check_thread_quit_pipe(pollfd, revents);
3000 if (ret) {
3001 err = 0;
3002 goto exit;
3003 }
3004
3005 /* Event on the registration socket */
3006 if (pollfd == sock) {
3007 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3008 ERR("Health socket poll error");
3009 goto error;
3010 }
3011 }
3012 }
3013
3014 new_sock = lttcomm_accept_unix_sock(sock);
3015 if (new_sock < 0) {
3016 goto error;
3017 }
3018
3019 /*
3020 * Set the CLOEXEC flag. Return code is useless because either way, the
3021 * show must go on.
3022 */
3023 (void) utils_set_fd_cloexec(new_sock);
3024
3025 DBG("Receiving data from client for health...");
3026 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3027 if (ret <= 0) {
3028 DBG("Nothing recv() from client... continuing");
3029 ret = close(new_sock);
3030 if (ret) {
3031 PERROR("close");
3032 }
3033 new_sock = -1;
3034 continue;
3035 }
3036
3037 rcu_thread_online();
3038
3039 switch (msg.component) {
3040 case LTTNG_HEALTH_CMD:
3041 reply.ret_code = health_check_state(&health_thread_cmd);
3042 break;
3043 case LTTNG_HEALTH_APP_MANAGE:
3044 reply.ret_code = health_check_state(&health_thread_app_manage);
3045 break;
3046 case LTTNG_HEALTH_APP_REG:
3047 reply.ret_code = health_check_state(&health_thread_app_reg);
3048 break;
3049 case LTTNG_HEALTH_KERNEL:
3050 reply.ret_code = health_check_state(&health_thread_kernel);
3051 break;
3052 case LTTNG_HEALTH_CONSUMER:
3053 reply.ret_code = check_consumer_health();
3054 break;
3055 case LTTNG_HEALTH_ALL:
3056 reply.ret_code =
3057 health_check_state(&health_thread_app_manage) &&
3058 health_check_state(&health_thread_app_reg) &&
3059 health_check_state(&health_thread_cmd) &&
3060 health_check_state(&health_thread_kernel) &&
3061 check_consumer_health();
3062 break;
3063 default:
3064 reply.ret_code = LTTNG_ERR_UND;
3065 break;
3066 }
3067
3068 /*
3069 * Flip ret value since 0 is a success and 1 indicates a bad health for
3070 * the client where in the sessiond it is the opposite. Again, this is
3071 * just to make things easier for us poor developer which enjoy a lot
3072 * lazyness.
3073 */
3074 if (reply.ret_code == 0 || reply.ret_code == 1) {
3075 reply.ret_code = !reply.ret_code;
3076 }
3077
3078 DBG2("Health check return value %d", reply.ret_code);
3079
3080 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3081 if (ret < 0) {
3082 ERR("Failed to send health data back to client");
3083 }
3084
3085 /* End of transmission */
3086 ret = close(new_sock);
3087 if (ret) {
3088 PERROR("close");
3089 }
3090 new_sock = -1;
3091 }
3092
3093 exit:
3094 error:
3095 if (err) {
3096 ERR("Health error occurred in %s", __func__);
3097 }
3098 DBG("Health check thread dying");
3099 unlink(health_unix_sock_path);
3100 if (sock >= 0) {
3101 ret = close(sock);
3102 if (ret) {
3103 PERROR("close");
3104 }
3105 }
3106 if (new_sock >= 0) {
3107 ret = close(new_sock);
3108 if (ret) {
3109 PERROR("close");
3110 }
3111 }
3112
3113 lttng_poll_clean(&events);
3114
3115 rcu_unregister_thread();
3116 return NULL;
3117 }
3118
3119 /*
3120 * This thread manage all clients request using the unix client socket for
3121 * communication.
3122 */
3123 static void *thread_manage_clients(void *data)
3124 {
3125 int sock = -1, ret, i, pollfd, err = -1;
3126 int sock_error;
3127 uint32_t revents, nb_fd;
3128 struct command_ctx *cmd_ctx = NULL;
3129 struct lttng_poll_event events;
3130
3131 DBG("[thread] Manage client started");
3132
3133 rcu_register_thread();
3134
3135 if (testpoint(thread_manage_clients)) {
3136 goto error_testpoint;
3137 }
3138
3139 health_code_update(&health_thread_cmd);
3140
3141 ret = lttcomm_listen_unix_sock(client_sock);
3142 if (ret < 0) {
3143 goto error_listen;
3144 }
3145
3146 /*
3147 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3148 * more will be added to this poll set.
3149 */
3150 ret = create_thread_poll_set(&events, 2);
3151 if (ret < 0) {
3152 goto error_create_poll;
3153 }
3154
3155 /* Add the application registration socket */
3156 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3157 if (ret < 0) {
3158 goto error;
3159 }
3160
3161 /*
3162 * Notify parent pid that we are ready to accept command for client side.
3163 */
3164 if (opt_sig_parent) {
3165 kill(ppid, SIGUSR1);
3166 }
3167
3168 if (testpoint(thread_manage_clients_before_loop)) {
3169 goto error;
3170 }
3171
3172 health_code_update(&health_thread_cmd);
3173
3174 while (1) {
3175 DBG("Accepting client command ...");
3176
3177 /* Inifinite blocking call, waiting for transmission */
3178 restart:
3179 health_poll_update(&health_thread_cmd);
3180 ret = lttng_poll_wait(&events, -1);
3181 health_poll_update(&health_thread_cmd);
3182 if (ret < 0) {
3183 /*
3184 * Restart interrupted system call.
3185 */
3186 if (errno == EINTR) {
3187 goto restart;
3188 }
3189 goto error;
3190 }
3191
3192 nb_fd = ret;
3193
3194 for (i = 0; i < nb_fd; i++) {
3195 /* Fetch once the poll data */
3196 revents = LTTNG_POLL_GETEV(&events, i);
3197 pollfd = LTTNG_POLL_GETFD(&events, i);
3198
3199 health_code_update(&health_thread_cmd);
3200
3201 /* Thread quit pipe has been closed. Killing thread. */
3202 ret = check_thread_quit_pipe(pollfd, revents);
3203 if (ret) {
3204 err = 0;
3205 goto exit;
3206 }
3207
3208 /* Event on the registration socket */
3209 if (pollfd == client_sock) {
3210 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3211 ERR("Client socket poll error");
3212 goto error;
3213 }
3214 }
3215 }
3216
3217 DBG("Wait for client response");
3218
3219 health_code_update(&health_thread_cmd);
3220
3221 sock = lttcomm_accept_unix_sock(client_sock);
3222 if (sock < 0) {
3223 goto error;
3224 }
3225
3226 /*
3227 * Set the CLOEXEC flag. Return code is useless because either way, the
3228 * show must go on.
3229 */
3230 (void) utils_set_fd_cloexec(sock);
3231
3232 /* Set socket option for credentials retrieval */
3233 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3234 if (ret < 0) {
3235 goto error;
3236 }
3237
3238 /* Allocate context command to process the client request */
3239 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3240 if (cmd_ctx == NULL) {
3241 PERROR("zmalloc cmd_ctx");
3242 goto error;
3243 }
3244
3245 /* Allocate data buffer for reception */
3246 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3247 if (cmd_ctx->lsm == NULL) {
3248 PERROR("zmalloc cmd_ctx->lsm");
3249 goto error;
3250 }
3251
3252 cmd_ctx->llm = NULL;
3253 cmd_ctx->session = NULL;
3254
3255 health_code_update(&health_thread_cmd);
3256
3257 /*
3258 * Data is received from the lttng client. The struct
3259 * lttcomm_session_msg (lsm) contains the command and data request of
3260 * the client.
3261 */
3262 DBG("Receiving data from client ...");
3263 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3264 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3265 if (ret <= 0) {
3266 DBG("Nothing recv() from client... continuing");
3267 ret = close(sock);
3268 if (ret) {
3269 PERROR("close");
3270 }
3271 sock = -1;
3272 clean_command_ctx(&cmd_ctx);
3273 continue;
3274 }
3275
3276 health_code_update(&health_thread_cmd);
3277
3278 // TODO: Validate cmd_ctx including sanity check for
3279 // security purpose.
3280
3281 rcu_thread_online();
3282 /*
3283 * This function dispatch the work to the kernel or userspace tracer
3284 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3285 * informations for the client. The command context struct contains
3286 * everything this function may needs.
3287 */
3288 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3289 rcu_thread_offline();
3290 if (ret < 0) {
3291 if (sock_error) {
3292 ret = close(sock);
3293 if (ret) {
3294 PERROR("close");
3295 }
3296 sock = -1;
3297 }
3298 /*
3299 * TODO: Inform client somehow of the fatal error. At
3300 * this point, ret < 0 means that a zmalloc failed
3301 * (ENOMEM). Error detected but still accept
3302 * command, unless a socket error has been
3303 * detected.
3304 */
3305 clean_command_ctx(&cmd_ctx);
3306 continue;
3307 }
3308
3309 health_code_update(&health_thread_cmd);
3310
3311 DBG("Sending response (size: %d, retcode: %s)",
3312 cmd_ctx->lttng_msg_size,
3313 lttng_strerror(-cmd_ctx->llm->ret_code));
3314 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3315 if (ret < 0) {
3316 ERR("Failed to send data back to client");
3317 }
3318
3319 /* End of transmission */
3320 ret = close(sock);
3321 if (ret) {
3322 PERROR("close");
3323 }
3324 sock = -1;
3325
3326 clean_command_ctx(&cmd_ctx);
3327
3328 health_code_update(&health_thread_cmd);
3329 }
3330
3331 exit:
3332 error:
3333 if (sock >= 0) {
3334 ret = close(sock);
3335 if (ret) {
3336 PERROR("close");
3337 }
3338 }
3339
3340 lttng_poll_clean(&events);
3341 clean_command_ctx(&cmd_ctx);
3342
3343 error_listen:
3344 error_create_poll:
3345 error_testpoint:
3346 unlink(client_unix_sock_path);
3347 if (client_sock >= 0) {
3348 ret = close(client_sock);
3349 if (ret) {
3350 PERROR("close");
3351 }
3352 }
3353
3354 if (err) {
3355 health_error(&health_thread_cmd);
3356 ERR("Health error occurred in %s", __func__);
3357 }
3358
3359 health_exit(&health_thread_cmd);
3360
3361 DBG("Client thread dying");
3362
3363 rcu_unregister_thread();
3364 return NULL;
3365 }
3366
3367
3368 /*
3369 * usage function on stderr
3370 */
3371 static void usage(void)
3372 {
3373 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3374 fprintf(stderr, " -h, --help Display this usage.\n");
3375 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3376 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3377 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3378 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3379 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3380 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3381 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3382 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3383 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3384 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3385 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3386 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3387 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3388 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3389 fprintf(stderr, " -V, --version Show version number.\n");
3390 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3391 fprintf(stderr, " -q, --quiet No output at all.\n");
3392 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3393 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3394 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3395 }
3396
3397 /*
3398 * daemon argument parsing
3399 */
3400 static int parse_args(int argc, char **argv)
3401 {
3402 int c;
3403
3404 static struct option long_options[] = {
3405 { "client-sock", 1, 0, 'c' },
3406 { "apps-sock", 1, 0, 'a' },
3407 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3408 { "kconsumerd-err-sock", 1, 0, 'E' },
3409 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3410 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3411 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3412 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3413 { "consumerd32-path", 1, 0, 'u' },
3414 { "consumerd32-libdir", 1, 0, 'U' },
3415 { "consumerd64-path", 1, 0, 't' },
3416 { "consumerd64-libdir", 1, 0, 'T' },
3417 { "daemonize", 0, 0, 'd' },
3418 { "sig-parent", 0, 0, 'S' },
3419 { "help", 0, 0, 'h' },
3420 { "group", 1, 0, 'g' },
3421 { "version", 0, 0, 'V' },
3422 { "quiet", 0, 0, 'q' },
3423 { "verbose", 0, 0, 'v' },
3424 { "verbose-consumer", 0, 0, 'Z' },
3425 { "no-kernel", 0, 0, 'N' },
3426 { NULL, 0, 0, 0 }
3427 };
3428
3429 while (1) {
3430 int option_index = 0;
3431 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3432 long_options, &option_index);
3433 if (c == -1) {
3434 break;
3435 }
3436
3437 switch (c) {
3438 case 0:
3439 fprintf(stderr, "option %s", long_options[option_index].name);
3440 if (optarg) {
3441 fprintf(stderr, " with arg %s\n", optarg);
3442 }
3443 break;
3444 case 'c':
3445 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3446 break;
3447 case 'a':
3448 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3449 break;
3450 case 'd':
3451 opt_daemon = 1;
3452 break;
3453 case 'g':
3454 opt_tracing_group = optarg;
3455 break;
3456 case 'h':
3457 usage();
3458 exit(EXIT_FAILURE);
3459 case 'V':
3460 fprintf(stdout, "%s\n", VERSION);
3461 exit(EXIT_SUCCESS);
3462 case 'S':
3463 opt_sig_parent = 1;
3464 break;
3465 case 'E':
3466 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3467 break;
3468 case 'C':
3469 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3470 break;
3471 case 'F':
3472 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3473 break;
3474 case 'D':
3475 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3476 break;
3477 case 'H':
3478 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3479 break;
3480 case 'G':
3481 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3482 break;
3483 case 'N':
3484 opt_no_kernel = 1;
3485 break;
3486 case 'q':
3487 lttng_opt_quiet = 1;
3488 break;
3489 case 'v':
3490 /* Verbose level can increase using multiple -v */
3491 lttng_opt_verbose += 1;
3492 break;
3493 case 'Z':
3494 opt_verbose_consumer += 1;
3495 break;
3496 case 'u':
3497 consumerd32_bin= optarg;
3498 break;
3499 case 'U':
3500 consumerd32_libdir = optarg;
3501 break;
3502 case 't':
3503 consumerd64_bin = optarg;
3504 break;
3505 case 'T':
3506 consumerd64_libdir = optarg;
3507 break;
3508 default:
3509 /* Unknown option or other error.
3510 * Error is printed by getopt, just return */
3511 return -1;
3512 }
3513 }
3514
3515 return 0;
3516 }
3517
3518 /*
3519 * Creates the two needed socket by the daemon.
3520 * apps_sock - The communication socket for all UST apps.
3521 * client_sock - The communication of the cli tool (lttng).
3522 */
3523 static int init_daemon_socket(void)
3524 {
3525 int ret = 0;
3526 mode_t old_umask;
3527
3528 old_umask = umask(0);
3529
3530 /* Create client tool unix socket */
3531 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3532 if (client_sock < 0) {
3533 ERR("Create unix sock failed: %s", client_unix_sock_path);
3534 ret = -1;
3535 goto end;
3536 }
3537
3538 /* Set the cloexec flag */
3539 ret = utils_set_fd_cloexec(client_sock);
3540 if (ret < 0) {
3541 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3542 "Continuing but note that the consumer daemon will have a "
3543 "reference to this socket on exec()", client_sock);
3544 }
3545
3546 /* File permission MUST be 660 */
3547 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3548 if (ret < 0) {
3549 ERR("Set file permissions failed: %s", client_unix_sock_path);
3550 PERROR("chmod");
3551 goto end;
3552 }
3553
3554 /* Create the application unix socket */
3555 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3556 if (apps_sock < 0) {
3557 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3558 ret = -1;
3559 goto end;
3560 }
3561
3562 /* Set the cloexec flag */
3563 ret = utils_set_fd_cloexec(apps_sock);
3564 if (ret < 0) {
3565 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3566 "Continuing but note that the consumer daemon will have a "
3567 "reference to this socket on exec()", apps_sock);
3568 }
3569
3570 /* File permission MUST be 666 */
3571 ret = chmod(apps_unix_sock_path,
3572 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3573 if (ret < 0) {
3574 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3575 PERROR("chmod");
3576 goto end;
3577 }
3578
3579 DBG3("Session daemon client socket %d and application socket %d created",
3580 client_sock, apps_sock);
3581
3582 end:
3583 umask(old_umask);
3584 return ret;
3585 }
3586
3587 /*
3588 * Check if the global socket is available, and if a daemon is answering at the
3589 * other side. If yes, error is returned.
3590 */
3591 static int check_existing_daemon(void)
3592 {
3593 /* Is there anybody out there ? */
3594 if (lttng_session_daemon_alive()) {
3595 return -EEXIST;
3596 }
3597
3598 return 0;
3599 }
3600
3601 /*
3602 * Set the tracing group gid onto the client socket.
3603 *
3604 * Race window between mkdir and chown is OK because we are going from more
3605 * permissive (root.root) to less permissive (root.tracing).
3606 */
3607 static int set_permissions(char *rundir)
3608 {
3609 int ret;
3610 gid_t gid;
3611
3612 ret = allowed_group();
3613 if (ret < 0) {
3614 WARN("No tracing group detected");
3615 ret = 0;
3616 goto end;
3617 }
3618
3619 gid = ret;
3620
3621 /* Set lttng run dir */
3622 ret = chown(rundir, 0, gid);
3623 if (ret < 0) {
3624 ERR("Unable to set group on %s", rundir);
3625 PERROR("chown");
3626 }
3627
3628 /* Ensure tracing group can search the run dir */
3629 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3630 if (ret < 0) {
3631 ERR("Unable to set permissions on %s", rundir);
3632 PERROR("chmod");
3633 }
3634
3635 /* lttng client socket path */
3636 ret = chown(client_unix_sock_path, 0, gid);
3637 if (ret < 0) {
3638 ERR("Unable to set group on %s", client_unix_sock_path);
3639 PERROR("chown");
3640 }
3641
3642 /* kconsumer error socket path */
3643 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3644 if (ret < 0) {
3645 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3646 PERROR("chown");
3647 }
3648
3649 /* 64-bit ustconsumer error socket path */
3650 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3651 if (ret < 0) {
3652 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3653 PERROR("chown");
3654 }
3655
3656 /* 32-bit ustconsumer compat32 error socket path */
3657 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3658 if (ret < 0) {
3659 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3660 PERROR("chown");
3661 }
3662
3663 DBG("All permissions are set");
3664
3665 end:
3666 return ret;
3667 }
3668
3669 /*
3670 * Create the lttng run directory needed for all global sockets and pipe.
3671 */
3672 static int create_lttng_rundir(const char *rundir)
3673 {
3674 int ret;
3675
3676 DBG3("Creating LTTng run directory: %s", rundir);
3677
3678 ret = mkdir(rundir, S_IRWXU);
3679 if (ret < 0) {
3680 if (errno != EEXIST) {
3681 ERR("Unable to create %s", rundir);
3682 goto error;
3683 } else {
3684 ret = 0;
3685 }
3686 }
3687
3688 error:
3689 return ret;
3690 }
3691
3692 /*
3693 * Setup sockets and directory needed by the kconsumerd communication with the
3694 * session daemon.
3695 */
3696 static int set_consumer_sockets(struct consumer_data *consumer_data,
3697 const char *rundir)
3698 {
3699 int ret;
3700 char path[PATH_MAX];
3701
3702 switch (consumer_data->type) {
3703 case LTTNG_CONSUMER_KERNEL:
3704 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3705 break;
3706 case LTTNG_CONSUMER64_UST:
3707 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3708 break;
3709 case LTTNG_CONSUMER32_UST:
3710 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3711 break;
3712 default:
3713 ERR("Consumer type unknown");
3714 ret = -EINVAL;
3715 goto error;
3716 }
3717
3718 DBG2("Creating consumer directory: %s", path);
3719
3720 ret = mkdir(path, S_IRWXU);
3721 if (ret < 0) {
3722 if (errno != EEXIST) {
3723 PERROR("mkdir");
3724 ERR("Failed to create %s", path);
3725 goto error;
3726 }
3727 ret = -1;
3728 }
3729
3730 /* Create the kconsumerd error unix socket */
3731 consumer_data->err_sock =
3732 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3733 if (consumer_data->err_sock < 0) {
3734 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3735 ret = -1;
3736 goto error;
3737 }
3738
3739 /* File permission MUST be 660 */
3740 ret = chmod(consumer_data->err_unix_sock_path,
3741 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3742 if (ret < 0) {
3743 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3744 PERROR("chmod");
3745 goto error;
3746 }
3747
3748 error:
3749 return ret;
3750 }
3751
3752 /*
3753 * Signal handler for the daemon
3754 *
3755 * Simply stop all worker threads, leaving main() return gracefully after
3756 * joining all threads and calling cleanup().
3757 */
3758 static void sighandler(int sig)
3759 {
3760 switch (sig) {
3761 case SIGPIPE:
3762 DBG("SIGPIPE caught");
3763 return;
3764 case SIGINT:
3765 DBG("SIGINT caught");
3766 stop_threads();
3767 break;
3768 case SIGTERM:
3769 DBG("SIGTERM caught");
3770 stop_threads();
3771 break;
3772 default:
3773 break;
3774 }
3775 }
3776
3777 /*
3778 * Setup signal handler for :
3779 * SIGINT, SIGTERM, SIGPIPE
3780 */
3781 static int set_signal_handler(void)
3782 {
3783 int ret = 0;
3784 struct sigaction sa;
3785 sigset_t sigset;
3786
3787 if ((ret = sigemptyset(&sigset)) < 0) {
3788 PERROR("sigemptyset");
3789 return ret;
3790 }
3791
3792 sa.sa_handler = sighandler;
3793 sa.sa_mask = sigset;
3794 sa.sa_flags = 0;
3795 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3796 PERROR("sigaction");
3797 return ret;
3798 }
3799
3800 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3801 PERROR("sigaction");
3802 return ret;
3803 }
3804
3805 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3806 PERROR("sigaction");
3807 return ret;
3808 }
3809
3810 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3811
3812 return ret;
3813 }
3814
3815 /*
3816 * Set open files limit to unlimited. This daemon can open a large number of
3817 * file descriptors in order to consumer multiple kernel traces.
3818 */
3819 static void set_ulimit(void)
3820 {
3821 int ret;
3822 struct rlimit lim;
3823
3824 /* The kernel does not allowed an infinite limit for open files */
3825 lim.rlim_cur = 65535;
3826 lim.rlim_max = 65535;
3827
3828 ret = setrlimit(RLIMIT_NOFILE, &lim);
3829 if (ret < 0) {
3830 PERROR("failed to set open files limit");
3831 }
3832 }
3833
3834 /*
3835 * main
3836 */
3837 int main(int argc, char **argv)
3838 {
3839 int ret = 0;
3840 void *status;
3841 const char *home_path, *env_app_timeout;
3842
3843 init_kernel_workarounds();
3844
3845 rcu_register_thread();
3846
3847 setup_consumerd_path();
3848
3849 /* Parse arguments */
3850 progname = argv[0];
3851 if ((ret = parse_args(argc, argv)) < 0) {
3852 goto error;
3853 }
3854
3855 /* Daemonize */
3856 if (opt_daemon) {
3857 int i;
3858
3859 /*
3860 * fork
3861 * child: setsid, close FD 0, 1, 2, chdir /
3862 * parent: exit (if fork is successful)
3863 */
3864 ret = daemon(0, 0);
3865 if (ret < 0) {
3866 PERROR("daemon");
3867 goto error;
3868 }
3869 /*
3870 * We are in the child. Make sure all other file
3871 * descriptors are closed, in case we are called with
3872 * more opened file descriptors than the standard ones.
3873 */
3874 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3875 (void) close(i);
3876 }
3877 }
3878
3879 /* Create thread quit pipe */
3880 if ((ret = init_thread_quit_pipe()) < 0) {
3881 goto error;
3882 }
3883
3884 /* Check if daemon is UID = 0 */
3885 is_root = !getuid();
3886
3887 if (is_root) {
3888 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3889
3890 /* Create global run dir with root access */
3891 ret = create_lttng_rundir(rundir);
3892 if (ret < 0) {
3893 goto error;
3894 }
3895
3896 if (strlen(apps_unix_sock_path) == 0) {
3897 snprintf(apps_unix_sock_path, PATH_MAX,
3898 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3899 }
3900
3901 if (strlen(client_unix_sock_path) == 0) {
3902 snprintf(client_unix_sock_path, PATH_MAX,
3903 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3904 }
3905
3906 /* Set global SHM for ust */
3907 if (strlen(wait_shm_path) == 0) {
3908 snprintf(wait_shm_path, PATH_MAX,
3909 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3910 }
3911
3912 if (strlen(health_unix_sock_path) == 0) {
3913 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3914 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3915 }
3916
3917 /* Setup kernel consumerd path */
3918 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3919 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3920 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3921 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3922
3923 DBG2("Kernel consumer err path: %s",
3924 kconsumer_data.err_unix_sock_path);
3925 DBG2("Kernel consumer cmd path: %s",
3926 kconsumer_data.cmd_unix_sock_path);
3927 } else {
3928 home_path = get_home_dir();
3929 if (home_path == NULL) {
3930 /* TODO: Add --socket PATH option */
3931 ERR("Can't get HOME directory for sockets creation.");
3932 ret = -EPERM;
3933 goto error;
3934 }
3935
3936 /*
3937 * Create rundir from home path. This will create something like
3938 * $HOME/.lttng
3939 */
3940 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3941 if (ret < 0) {
3942 ret = -ENOMEM;
3943 goto error;
3944 }
3945
3946 ret = create_lttng_rundir(rundir);
3947 if (ret < 0) {
3948 goto error;
3949 }
3950
3951 if (strlen(apps_unix_sock_path) == 0) {
3952 snprintf(apps_unix_sock_path, PATH_MAX,
3953 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3954 }
3955
3956 /* Set the cli tool unix socket path */
3957 if (strlen(client_unix_sock_path) == 0) {
3958 snprintf(client_unix_sock_path, PATH_MAX,
3959 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3960 }
3961
3962 /* Set global SHM for ust */
3963 if (strlen(wait_shm_path) == 0) {
3964 snprintf(wait_shm_path, PATH_MAX,
3965 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3966 }
3967
3968 /* Set health check Unix path */
3969 if (strlen(health_unix_sock_path) == 0) {
3970 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3971 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3972 }
3973 }
3974
3975 /* Set consumer initial state */
3976 kernel_consumerd_state = CONSUMER_STOPPED;
3977 ust_consumerd_state = CONSUMER_STOPPED;
3978
3979 DBG("Client socket path %s", client_unix_sock_path);
3980 DBG("Application socket path %s", apps_unix_sock_path);
3981 DBG("LTTng run directory path: %s", rundir);
3982
3983 /* 32 bits consumerd path setup */
3984 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3985 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3986 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3987 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3988
3989 DBG2("UST consumer 32 bits err path: %s",
3990 ustconsumer32_data.err_unix_sock_path);
3991 DBG2("UST consumer 32 bits cmd path: %s",
3992 ustconsumer32_data.cmd_unix_sock_path);
3993
3994 /* 64 bits consumerd path setup */
3995 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3996 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
3997 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
3998 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
3999
4000 DBG2("UST consumer 64 bits err path: %s",
4001 ustconsumer64_data.err_unix_sock_path);
4002 DBG2("UST consumer 64 bits cmd path: %s",
4003 ustconsumer64_data.cmd_unix_sock_path);
4004
4005 /*
4006 * See if daemon already exist.
4007 */
4008 if ((ret = check_existing_daemon()) < 0) {
4009 ERR("Already running daemon.\n");
4010 /*
4011 * We do not goto exit because we must not cleanup()
4012 * because a daemon is already running.
4013 */
4014 goto error;
4015 }
4016
4017 /*
4018 * Init UST app hash table. Alloc hash table before this point since
4019 * cleanup() can get called after that point.
4020 */
4021 ust_app_ht_alloc();
4022
4023 /* After this point, we can safely call cleanup() with "goto exit" */
4024
4025 /*
4026 * These actions must be executed as root. We do that *after* setting up
4027 * the sockets path because we MUST make the check for another daemon using
4028 * those paths *before* trying to set the kernel consumer sockets and init
4029 * kernel tracer.
4030 */
4031 if (is_root) {
4032 ret = set_consumer_sockets(&kconsumer_data, rundir);
4033 if (ret < 0) {
4034 goto exit;
4035 }
4036
4037 /* Setup kernel tracer */
4038 if (!opt_no_kernel) {
4039 init_kernel_tracer();
4040 }
4041
4042 /* Set ulimit for open files */
4043 set_ulimit();
4044 }
4045 /* init lttng_fd tracking must be done after set_ulimit. */
4046 lttng_fd_init();
4047
4048 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4049 if (ret < 0) {
4050 goto exit;
4051 }
4052
4053 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4054 if (ret < 0) {
4055 goto exit;
4056 }
4057
4058 if ((ret = set_signal_handler()) < 0) {
4059 goto exit;
4060 }
4061
4062 /* Setup the needed unix socket */
4063 if ((ret = init_daemon_socket()) < 0) {
4064 goto exit;
4065 }
4066
4067 /* Set credentials to socket */
4068 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4069 goto exit;
4070 }
4071
4072 /* Get parent pid if -S, --sig-parent is specified. */
4073 if (opt_sig_parent) {
4074 ppid = getppid();
4075 }
4076
4077 /* Setup the kernel pipe for waking up the kernel thread */
4078 if (is_root && !opt_no_kernel) {
4079 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4080 goto exit;
4081 }
4082 }
4083
4084 /* Setup the thread apps communication pipe. */
4085 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4086 goto exit;
4087 }
4088
4089 /* Init UST command queue. */
4090 cds_wfq_init(&ust_cmd_queue.queue);
4091
4092 /*
4093 * Get session list pointer. This pointer MUST NOT be free(). This list is
4094 * statically declared in session.c
4095 */
4096 session_list_ptr = session_get_list();
4097
4098 /* Set up max poll set size */
4099 lttng_poll_set_max_size();
4100
4101 cmd_init();
4102
4103 /* Init all health thread counters. */
4104 health_init(&health_thread_cmd);
4105 health_init(&health_thread_kernel);
4106 health_init(&health_thread_app_manage);
4107 health_init(&health_thread_app_reg);
4108
4109 /*
4110 * Init health counters of the consumer thread. We do a quick hack here to
4111 * the state of the consumer health is fine even if the thread is not
4112 * started. Once the thread starts, the health state is updated with a poll
4113 * value to set a health code path. This is simply to ease our life and has
4114 * no cost what so ever.
4115 */
4116 health_init(&kconsumer_data.health);
4117 health_poll_update(&kconsumer_data.health);
4118 health_init(&ustconsumer32_data.health);
4119 health_poll_update(&ustconsumer32_data.health);
4120 health_init(&ustconsumer64_data.health);
4121 health_poll_update(&ustconsumer64_data.health);
4122
4123 /* Check for the application socket timeout env variable. */
4124 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4125 if (env_app_timeout) {
4126 app_socket_timeout = atoi(env_app_timeout);
4127 } else {
4128 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4129 }
4130
4131 /* Create thread to manage the client socket */
4132 ret = pthread_create(&health_thread, NULL,
4133 thread_manage_health, (void *) NULL);
4134 if (ret != 0) {
4135 PERROR("pthread_create health");
4136 goto exit_health;
4137 }
4138
4139 /* Create thread to manage the client socket */
4140 ret = pthread_create(&client_thread, NULL,
4141 thread_manage_clients, (void *) NULL);
4142 if (ret != 0) {
4143 PERROR("pthread_create clients");
4144 goto exit_client;
4145 }
4146
4147 /* Create thread to dispatch registration */
4148 ret = pthread_create(&dispatch_thread, NULL,
4149 thread_dispatch_ust_registration, (void *) NULL);
4150 if (ret != 0) {
4151 PERROR("pthread_create dispatch");
4152 goto exit_dispatch;
4153 }
4154
4155 /* Create thread to manage application registration. */
4156 ret = pthread_create(&reg_apps_thread, NULL,
4157 thread_registration_apps, (void *) NULL);
4158 if (ret != 0) {
4159 PERROR("pthread_create registration");
4160 goto exit_reg_apps;
4161 }
4162
4163 /* Create thread to manage application socket */
4164 ret = pthread_create(&apps_thread, NULL,
4165 thread_manage_apps, (void *) NULL);
4166 if (ret != 0) {
4167 PERROR("pthread_create apps");
4168 goto exit_apps;
4169 }
4170
4171 /* Don't start this thread if kernel tracing is not requested nor root */
4172 if (is_root && !opt_no_kernel) {
4173 /* Create kernel thread to manage kernel event */
4174 ret = pthread_create(&kernel_thread, NULL,
4175 thread_manage_kernel, (void *) NULL);
4176 if (ret != 0) {
4177 PERROR("pthread_create kernel");
4178 goto exit_kernel;
4179 }
4180
4181 ret = pthread_join(kernel_thread, &status);
4182 if (ret != 0) {
4183 PERROR("pthread_join");
4184 goto error; /* join error, exit without cleanup */
4185 }
4186 }
4187
4188 exit_kernel:
4189 ret = pthread_join(apps_thread, &status);
4190 if (ret != 0) {
4191 PERROR("pthread_join");
4192 goto error; /* join error, exit without cleanup */
4193 }
4194
4195 exit_apps:
4196 ret = pthread_join(reg_apps_thread, &status);
4197 if (ret != 0) {
4198 PERROR("pthread_join");
4199 goto error; /* join error, exit without cleanup */
4200 }
4201
4202 exit_reg_apps:
4203 ret = pthread_join(dispatch_thread, &status);
4204 if (ret != 0) {
4205 PERROR("pthread_join");
4206 goto error; /* join error, exit without cleanup */
4207 }
4208
4209 exit_dispatch:
4210 ret = pthread_join(client_thread, &status);
4211 if (ret != 0) {
4212 PERROR("pthread_join");
4213 goto error; /* join error, exit without cleanup */
4214 }
4215
4216 ret = join_consumer_thread(&kconsumer_data);
4217 if (ret != 0) {
4218 PERROR("join_consumer");
4219 goto error; /* join error, exit without cleanup */
4220 }
4221
4222 ret = join_consumer_thread(&ustconsumer32_data);
4223 if (ret != 0) {
4224 PERROR("join_consumer ust32");
4225 goto error; /* join error, exit without cleanup */
4226 }
4227
4228 ret = join_consumer_thread(&ustconsumer64_data);
4229 if (ret != 0) {
4230 PERROR("join_consumer ust64");
4231 goto error; /* join error, exit without cleanup */
4232 }
4233
4234 exit_client:
4235 ret = pthread_join(health_thread, &status);
4236 if (ret != 0) {
4237 PERROR("pthread_join health thread");
4238 goto error; /* join error, exit without cleanup */
4239 }
4240
4241 exit_health:
4242 exit:
4243 /*
4244 * cleanup() is called when no other thread is running.
4245 */
4246 rcu_thread_online();
4247 cleanup();
4248 rcu_thread_offline();
4249 rcu_unregister_thread();
4250 if (!ret) {
4251 exit(EXIT_SUCCESS);
4252 }
4253 error:
4254 exit(EXIT_FAILURE);
4255 }
This page took 0.163854 seconds and 5 git commands to generate.