Fix: missing health exit in registration app thread
[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 manage kernel started");
693
694 testpoint(thread_manage_kernel);
695
696 health_code_update(&health_thread_kernel);
697
698 testpoint(thread_manage_kernel_before_loop);
699
700 ret = create_thread_poll_set(&events, 2);
701 if (ret < 0) {
702 goto error_poll_create;
703 }
704
705 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
706 if (ret < 0) {
707 goto error;
708 }
709
710 while (1) {
711 health_code_update(&health_thread_kernel);
712
713 if (update_poll_flag == 1) {
714 /*
715 * Reset number of fd in the poll set. Always 2 since there is the thread
716 * quit pipe and the kernel pipe.
717 */
718 events.nb_fd = 2;
719
720 ret = update_kernel_poll(&events);
721 if (ret < 0) {
722 goto error;
723 }
724 update_poll_flag = 0;
725 }
726
727 nb_fd = LTTNG_POLL_GETNB(&events);
728
729 DBG("Thread kernel polling on %d fds", nb_fd);
730
731 /* Zeroed the poll events */
732 lttng_poll_reset(&events);
733
734 /* Poll infinite value of time */
735 restart:
736 health_poll_update(&health_thread_kernel);
737 ret = lttng_poll_wait(&events, -1);
738 health_poll_update(&health_thread_kernel);
739 if (ret < 0) {
740 /*
741 * Restart interrupted system call.
742 */
743 if (errno == EINTR) {
744 goto restart;
745 }
746 goto error;
747 } else if (ret == 0) {
748 /* Should not happen since timeout is infinite */
749 ERR("Return value of poll is 0 with an infinite timeout.\n"
750 "This should not have happened! Continuing...");
751 continue;
752 }
753
754 for (i = 0; i < nb_fd; i++) {
755 /* Fetch once the poll data */
756 revents = LTTNG_POLL_GETEV(&events, i);
757 pollfd = LTTNG_POLL_GETFD(&events, i);
758
759 health_code_update(&health_thread_kernel);
760
761 /* Thread quit pipe has been closed. Killing thread. */
762 ret = check_thread_quit_pipe(pollfd, revents);
763 if (ret) {
764 err = 0;
765 goto exit;
766 }
767
768 /* Check for data on kernel pipe */
769 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
770 ret = read(kernel_poll_pipe[0], &tmp, 1);
771 update_poll_flag = 1;
772 continue;
773 } else {
774 /*
775 * New CPU detected by the kernel. Adding kernel stream to
776 * kernel session and updating the kernel consumer
777 */
778 if (revents & LPOLLIN) {
779 ret = update_kernel_stream(&kconsumer_data, pollfd);
780 if (ret < 0) {
781 continue;
782 }
783 break;
784 /*
785 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
786 * and unregister kernel stream at this point.
787 */
788 }
789 }
790 }
791 }
792
793 exit:
794 error:
795 lttng_poll_clean(&events);
796 error_poll_create:
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 nb_fd = LTTNG_POLL_GETNB(&events);
885
886 health_code_update(&consumer_data->health);
887
888 /* Inifinite blocking call, waiting for transmission */
889 restart:
890 health_poll_update(&consumer_data->health);
891
892 testpoint(thread_manage_consumer);
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 for (i = 0; i < nb_fd; i++) {
907 /* Fetch once the poll data */
908 revents = LTTNG_POLL_GETEV(&events, i);
909 pollfd = LTTNG_POLL_GETFD(&events, i);
910
911 health_code_update(&consumer_data->health);
912
913 /* Thread quit pipe has been closed. Killing thread. */
914 ret = check_thread_quit_pipe(pollfd, revents);
915 if (ret) {
916 err = 0;
917 goto exit;
918 }
919
920 /* Event on the registration socket */
921 if (pollfd == consumer_data->err_sock) {
922 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
923 ERR("consumer err socket poll error");
924 goto error;
925 }
926 }
927 }
928
929 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
930 if (sock < 0) {
931 goto error;
932 }
933
934 /*
935 * Set the CLOEXEC flag. Return code is useless because either way, the
936 * show must go on.
937 */
938 (void) utils_set_fd_cloexec(sock);
939
940 health_code_update(&consumer_data->health);
941
942 DBG2("Receiving code from consumer err_sock");
943
944 /* Getting status code from kconsumerd */
945 ret = lttcomm_recv_unix_sock(sock, &code,
946 sizeof(enum lttcomm_return_code));
947 if (ret <= 0) {
948 goto error;
949 }
950
951 health_code_update(&consumer_data->health);
952
953 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
954 consumer_data->cmd_sock =
955 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
956 if (consumer_data->cmd_sock < 0) {
957 /* On error, signal condition and quit. */
958 signal_consumer_condition(consumer_data, -1);
959 PERROR("consumer connect");
960 goto error;
961 }
962 signal_consumer_condition(consumer_data, 1);
963 DBG("Consumer command socket ready");
964 } else {
965 ERR("consumer error when waiting for SOCK_READY : %s",
966 lttcomm_get_readable_code(-code));
967 goto error;
968 }
969
970 /* Remove the kconsumerd error sock since we've established a connexion */
971 ret = lttng_poll_del(&events, consumer_data->err_sock);
972 if (ret < 0) {
973 goto error;
974 }
975
976 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
977 if (ret < 0) {
978 goto error;
979 }
980
981 health_code_update(&consumer_data->health);
982
983 /* Update number of fd */
984 nb_fd = LTTNG_POLL_GETNB(&events);
985
986 /* Inifinite blocking call, waiting for transmission */
987 restart_poll:
988 health_poll_update(&consumer_data->health);
989 ret = lttng_poll_wait(&events, -1);
990 health_poll_update(&consumer_data->health);
991 if (ret < 0) {
992 /*
993 * Restart interrupted system call.
994 */
995 if (errno == EINTR) {
996 goto restart_poll;
997 }
998 goto error;
999 }
1000
1001 for (i = 0; i < nb_fd; i++) {
1002 /* Fetch once the poll data */
1003 revents = LTTNG_POLL_GETEV(&events, i);
1004 pollfd = LTTNG_POLL_GETFD(&events, i);
1005
1006 health_code_update(&consumer_data->health);
1007
1008 /* Thread quit pipe has been closed. Killing thread. */
1009 ret = check_thread_quit_pipe(pollfd, revents);
1010 if (ret) {
1011 err = 0;
1012 goto exit;
1013 }
1014
1015 /* Event on the kconsumerd socket */
1016 if (pollfd == sock) {
1017 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1018 ERR("consumer err socket second poll error");
1019 goto error;
1020 }
1021 }
1022 }
1023
1024 health_code_update(&consumer_data->health);
1025
1026 /* Wait for any kconsumerd error */
1027 ret = lttcomm_recv_unix_sock(sock, &code,
1028 sizeof(enum lttcomm_return_code));
1029 if (ret <= 0) {
1030 ERR("consumer closed the command socket");
1031 goto error;
1032 }
1033
1034 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1035
1036 exit:
1037 error:
1038 /* Immediately set the consumerd state to stopped */
1039 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1040 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1041 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1042 consumer_data->type == LTTNG_CONSUMER32_UST) {
1043 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1044 } else {
1045 /* Code flow error... */
1046 assert(0);
1047 }
1048
1049 if (consumer_data->err_sock >= 0) {
1050 ret = close(consumer_data->err_sock);
1051 if (ret) {
1052 PERROR("close");
1053 }
1054 }
1055 if (consumer_data->cmd_sock >= 0) {
1056 ret = close(consumer_data->cmd_sock);
1057 if (ret) {
1058 PERROR("close");
1059 }
1060 }
1061 if (sock >= 0) {
1062 ret = close(sock);
1063 if (ret) {
1064 PERROR("close");
1065 }
1066 }
1067
1068 unlink(consumer_data->err_unix_sock_path);
1069 unlink(consumer_data->cmd_unix_sock_path);
1070 consumer_data->pid = 0;
1071
1072 lttng_poll_clean(&events);
1073 error_poll:
1074 if (err) {
1075 health_error(&consumer_data->health);
1076 ERR("Health error occurred in %s", __func__);
1077 }
1078 health_exit(&consumer_data->health);
1079 DBG("consumer thread cleanup completed");
1080
1081 return NULL;
1082 }
1083
1084 /*
1085 * This thread manage application communication.
1086 */
1087 static void *thread_manage_apps(void *data)
1088 {
1089 int i, ret, pollfd, err = -1;
1090 uint32_t revents, nb_fd;
1091 struct ust_command ust_cmd;
1092 struct lttng_poll_event events;
1093
1094 DBG("[thread] Manage application started");
1095
1096 testpoint(thread_manage_apps);
1097
1098 rcu_register_thread();
1099 rcu_thread_online();
1100
1101 health_code_update(&health_thread_app_manage);
1102
1103 ret = create_thread_poll_set(&events, 2);
1104 if (ret < 0) {
1105 goto error_poll_create;
1106 }
1107
1108 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1109 if (ret < 0) {
1110 goto error;
1111 }
1112
1113 testpoint(thread_manage_apps_before_loop);
1114
1115 health_code_update(&health_thread_app_manage);
1116
1117 while (1) {
1118 /* Zeroed the events structure */
1119 lttng_poll_reset(&events);
1120
1121 nb_fd = LTTNG_POLL_GETNB(&events);
1122
1123 DBG("Apps thread polling on %d fds", nb_fd);
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 for (i = 0; i < nb_fd; i++) {
1141 /* Fetch once the poll data */
1142 revents = LTTNG_POLL_GETEV(&events, i);
1143 pollfd = LTTNG_POLL_GETFD(&events, i);
1144
1145 health_code_update(&health_thread_app_manage);
1146
1147 /* Thread quit pipe has been closed. Killing thread. */
1148 ret = check_thread_quit_pipe(pollfd, revents);
1149 if (ret) {
1150 err = 0;
1151 goto exit;
1152 }
1153
1154 /* Inspect the apps cmd pipe */
1155 if (pollfd == apps_cmd_pipe[0]) {
1156 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1157 ERR("Apps command pipe error");
1158 goto error;
1159 } else if (revents & LPOLLIN) {
1160 /* Empty pipe */
1161 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1162 if (ret < 0 || ret < sizeof(ust_cmd)) {
1163 PERROR("read apps cmd pipe");
1164 goto error;
1165 }
1166
1167 health_code_update(&health_thread_app_manage);
1168
1169 /* Register applicaton to the session daemon */
1170 ret = ust_app_register(&ust_cmd.reg_msg,
1171 ust_cmd.sock);
1172 if (ret == -ENOMEM) {
1173 goto error;
1174 } else if (ret < 0) {
1175 break;
1176 }
1177
1178 health_code_update(&health_thread_app_manage);
1179
1180 /*
1181 * Validate UST version compatibility.
1182 */
1183 ret = ust_app_validate_version(ust_cmd.sock);
1184 if (ret >= 0) {
1185 /*
1186 * Add channel(s) and event(s) to newly registered apps
1187 * from lttng global UST domain.
1188 */
1189 update_ust_app(ust_cmd.sock);
1190 }
1191
1192 health_code_update(&health_thread_app_manage);
1193
1194 ret = ust_app_register_done(ust_cmd.sock);
1195 if (ret < 0) {
1196 /*
1197 * If the registration is not possible, we simply
1198 * unregister the apps and continue
1199 */
1200 ust_app_unregister(ust_cmd.sock);
1201 } else {
1202 /*
1203 * We only monitor the error events of the socket. This
1204 * thread does not handle any incoming data from UST
1205 * (POLLIN).
1206 */
1207 ret = lttng_poll_add(&events, ust_cmd.sock,
1208 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1209 if (ret < 0) {
1210 goto error;
1211 }
1212
1213 /* Set socket timeout for both receiving and ending */
1214 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1215 app_socket_timeout);
1216 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1217 app_socket_timeout);
1218
1219 DBG("Apps with sock %d added to poll set",
1220 ust_cmd.sock);
1221 }
1222
1223 health_code_update(&health_thread_app_manage);
1224
1225 break;
1226 }
1227 } else {
1228 /*
1229 * At this point, we know that a registered application made
1230 * the event at poll_wait.
1231 */
1232 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1233 /* Removing from the poll set */
1234 ret = lttng_poll_del(&events, pollfd);
1235 if (ret < 0) {
1236 goto error;
1237 }
1238
1239 /* Socket closed on remote end. */
1240 ust_app_unregister(pollfd);
1241 break;
1242 }
1243 }
1244
1245 health_code_update(&health_thread_app_manage);
1246 }
1247 }
1248
1249 exit:
1250 error:
1251 lttng_poll_clean(&events);
1252 error_poll_create:
1253 utils_close_pipe(apps_cmd_pipe);
1254 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1255
1256 /*
1257 * We don't clean the UST app hash table here since already registered
1258 * applications can still be controlled so let them be until the session
1259 * daemon dies or the applications stop.
1260 */
1261
1262 if (err) {
1263 health_error(&health_thread_app_manage);
1264 ERR("Health error occurred in %s", __func__);
1265 }
1266 health_exit(&health_thread_app_manage);
1267 DBG("Application communication apps thread cleanup complete");
1268 rcu_thread_offline();
1269 rcu_unregister_thread();
1270 return NULL;
1271 }
1272
1273 /*
1274 * Dispatch request from the registration threads to the application
1275 * communication thread.
1276 */
1277 static void *thread_dispatch_ust_registration(void *data)
1278 {
1279 int ret;
1280 struct cds_wfq_node *node;
1281 struct ust_command *ust_cmd = NULL;
1282
1283 DBG("[thread] Dispatch UST command started");
1284
1285 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1286 /* Atomically prepare the queue futex */
1287 futex_nto1_prepare(&ust_cmd_queue.futex);
1288
1289 do {
1290 /* Dequeue command for registration */
1291 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1292 if (node == NULL) {
1293 DBG("Woken up but nothing in the UST command queue");
1294 /* Continue thread execution */
1295 break;
1296 }
1297
1298 ust_cmd = caa_container_of(node, struct ust_command, node);
1299
1300 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1301 " gid:%d sock:%d name:%s (version %d.%d)",
1302 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1303 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1304 ust_cmd->sock, ust_cmd->reg_msg.name,
1305 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1306 /*
1307 * Inform apps thread of the new application registration. This
1308 * call is blocking so we can be assured that the data will be read
1309 * at some point in time or wait to the end of the world :)
1310 */
1311 if (apps_cmd_pipe[1] >= 0) {
1312 ret = write(apps_cmd_pipe[1], ust_cmd,
1313 sizeof(struct ust_command));
1314 if (ret < 0) {
1315 PERROR("write apps cmd pipe");
1316 if (errno == EBADF) {
1317 /*
1318 * We can't inform the application thread to process
1319 * registration. We will exit or else application
1320 * registration will not occur and tracing will never
1321 * start.
1322 */
1323 goto error;
1324 }
1325 }
1326 } else {
1327 /* Application manager thread is not available. */
1328 ret = close(ust_cmd->sock);
1329 if (ret < 0) {
1330 PERROR("close ust_cmd sock");
1331 }
1332 }
1333 free(ust_cmd);
1334 } while (node != NULL);
1335
1336 /* Futex wait on queue. Blocking call on futex() */
1337 futex_nto1_wait(&ust_cmd_queue.futex);
1338 }
1339
1340 error:
1341 DBG("Dispatch thread dying");
1342 return NULL;
1343 }
1344
1345 /*
1346 * This thread manage application registration.
1347 */
1348 static void *thread_registration_apps(void *data)
1349 {
1350 int sock = -1, i, ret, pollfd, err = -1;
1351 uint32_t revents, nb_fd;
1352 struct lttng_poll_event events;
1353 /*
1354 * Get allocated in this thread, enqueued to a global queue, dequeued and
1355 * freed in the manage apps thread.
1356 */
1357 struct ust_command *ust_cmd = NULL;
1358
1359 DBG("[thread] Manage application registration started");
1360
1361 testpoint(thread_registration_apps);
1362
1363 ret = lttcomm_listen_unix_sock(apps_sock);
1364 if (ret < 0) {
1365 goto error_listen;
1366 }
1367
1368 /*
1369 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1370 * more will be added to this poll set.
1371 */
1372 ret = create_thread_poll_set(&events, 2);
1373 if (ret < 0) {
1374 goto error_create_poll;
1375 }
1376
1377 /* Add the application registration socket */
1378 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1379 if (ret < 0) {
1380 goto error_poll_add;
1381 }
1382
1383 /* Notify all applications to register */
1384 ret = notify_ust_apps(1);
1385 if (ret < 0) {
1386 ERR("Failed to notify applications or create the wait shared memory.\n"
1387 "Execution continues but there might be problem for already\n"
1388 "running applications that wishes to register.");
1389 }
1390
1391 while (1) {
1392 DBG("Accepting application registration");
1393
1394 nb_fd = LTTNG_POLL_GETNB(&events);
1395
1396 /* Inifinite blocking call, waiting for transmission */
1397 restart:
1398 health_poll_update(&health_thread_app_reg);
1399 ret = lttng_poll_wait(&events, -1);
1400 health_poll_update(&health_thread_app_reg);
1401 if (ret < 0) {
1402 /*
1403 * Restart interrupted system call.
1404 */
1405 if (errno == EINTR) {
1406 goto restart;
1407 }
1408 goto error;
1409 }
1410
1411 for (i = 0; i < nb_fd; i++) {
1412 health_code_update(&health_thread_app_reg);
1413
1414 /* Fetch once the poll data */
1415 revents = LTTNG_POLL_GETEV(&events, i);
1416 pollfd = LTTNG_POLL_GETFD(&events, i);
1417
1418 /* Thread quit pipe has been closed. Killing thread. */
1419 ret = check_thread_quit_pipe(pollfd, revents);
1420 if (ret) {
1421 err = 0;
1422 goto exit;
1423 }
1424
1425 /* Event on the registration socket */
1426 if (pollfd == apps_sock) {
1427 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1428 ERR("Register apps socket poll error");
1429 goto error;
1430 } else if (revents & LPOLLIN) {
1431 sock = lttcomm_accept_unix_sock(apps_sock);
1432 if (sock < 0) {
1433 goto error;
1434 }
1435
1436 /*
1437 * Set the CLOEXEC flag. Return code is useless because
1438 * either way, the show must go on.
1439 */
1440 (void) utils_set_fd_cloexec(sock);
1441
1442 /* Create UST registration command for enqueuing */
1443 ust_cmd = zmalloc(sizeof(struct ust_command));
1444 if (ust_cmd == NULL) {
1445 PERROR("ust command zmalloc");
1446 goto error;
1447 }
1448
1449 /*
1450 * Using message-based transmissions to ensure we don't
1451 * have to deal with partially received messages.
1452 */
1453 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1454 if (ret < 0) {
1455 ERR("Exhausted file descriptors allowed for applications.");
1456 free(ust_cmd);
1457 ret = close(sock);
1458 if (ret) {
1459 PERROR("close");
1460 }
1461 sock = -1;
1462 continue;
1463 }
1464 health_code_update(&health_thread_app_reg);
1465 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1466 sizeof(struct ust_register_msg));
1467 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1468 if (ret < 0) {
1469 PERROR("lttcomm_recv_unix_sock register apps");
1470 } else {
1471 ERR("Wrong size received on apps register");
1472 }
1473 free(ust_cmd);
1474 ret = close(sock);
1475 if (ret) {
1476 PERROR("close");
1477 }
1478 lttng_fd_put(LTTNG_FD_APPS, 1);
1479 sock = -1;
1480 continue;
1481 }
1482 health_code_update(&health_thread_app_reg);
1483
1484 ust_cmd->sock = sock;
1485 sock = -1;
1486
1487 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1488 " gid:%d sock:%d name:%s (version %d.%d)",
1489 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1490 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1491 ust_cmd->sock, ust_cmd->reg_msg.name,
1492 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1493
1494 /*
1495 * Lock free enqueue the registration request. The red pill
1496 * has been taken! This apps will be part of the *system*.
1497 */
1498 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1499
1500 /*
1501 * Wake the registration queue futex. Implicit memory
1502 * barrier with the exchange in cds_wfq_enqueue.
1503 */
1504 futex_nto1_wake(&ust_cmd_queue.futex);
1505 }
1506 }
1507 }
1508 }
1509
1510 exit:
1511 error:
1512 if (err) {
1513 health_error(&health_thread_app_reg);
1514 ERR("Health error occurred in %s", __func__);
1515 }
1516
1517 /* Notify that the registration thread is gone */
1518 notify_ust_apps(0);
1519
1520 if (apps_sock >= 0) {
1521 ret = close(apps_sock);
1522 if (ret) {
1523 PERROR("close");
1524 }
1525 }
1526 if (sock >= 0) {
1527 ret = close(sock);
1528 if (ret) {
1529 PERROR("close");
1530 }
1531 lttng_fd_put(LTTNG_FD_APPS, 1);
1532 }
1533 unlink(apps_unix_sock_path);
1534
1535 error_poll_add:
1536 lttng_poll_clean(&events);
1537 error_listen:
1538 error_create_poll:
1539 DBG("UST Registration thread cleanup complete");
1540 health_exit(&health_thread_app_reg);
1541
1542 return NULL;
1543 }
1544
1545 /*
1546 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1547 * exec or it will fails.
1548 */
1549 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1550 {
1551 int ret, clock_ret;
1552 struct timespec timeout;
1553
1554 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1555 consumer_data->consumer_thread_is_ready = 0;
1556
1557 /* Setup pthread condition */
1558 ret = pthread_condattr_init(&consumer_data->condattr);
1559 if (ret != 0) {
1560 errno = ret;
1561 PERROR("pthread_condattr_init consumer data");
1562 goto error;
1563 }
1564
1565 /*
1566 * Set the monotonic clock in order to make sure we DO NOT jump in time
1567 * between the clock_gettime() call and the timedwait call. See bug #324
1568 * for a more details and how we noticed it.
1569 */
1570 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1571 if (ret != 0) {
1572 errno = ret;
1573 PERROR("pthread_condattr_setclock consumer data");
1574 goto error;
1575 }
1576
1577 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1578 if (ret != 0) {
1579 errno = ret;
1580 PERROR("pthread_cond_init consumer data");
1581 goto error;
1582 }
1583
1584 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1585 consumer_data);
1586 if (ret != 0) {
1587 PERROR("pthread_create consumer");
1588 ret = -1;
1589 goto error;
1590 }
1591
1592 /* We are about to wait on a pthread condition */
1593 pthread_mutex_lock(&consumer_data->cond_mutex);
1594
1595 /* Get time for sem_timedwait absolute timeout */
1596 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1597 /*
1598 * Set the timeout for the condition timed wait even if the clock gettime
1599 * call fails since we might loop on that call and we want to avoid to
1600 * increment the timeout too many times.
1601 */
1602 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1603
1604 /*
1605 * The following loop COULD be skipped in some conditions so this is why we
1606 * set ret to 0 in order to make sure at least one round of the loop is
1607 * done.
1608 */
1609 ret = 0;
1610
1611 /*
1612 * Loop until the condition is reached or when a timeout is reached. Note
1613 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1614 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1615 * possible. This loop does not take any chances and works with both of
1616 * them.
1617 */
1618 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1619 if (clock_ret < 0) {
1620 PERROR("clock_gettime spawn consumer");
1621 /* Infinite wait for the consumerd thread to be ready */
1622 ret = pthread_cond_wait(&consumer_data->cond,
1623 &consumer_data->cond_mutex);
1624 } else {
1625 ret = pthread_cond_timedwait(&consumer_data->cond,
1626 &consumer_data->cond_mutex, &timeout);
1627 }
1628 }
1629
1630 /* Release the pthread condition */
1631 pthread_mutex_unlock(&consumer_data->cond_mutex);
1632
1633 if (ret != 0) {
1634 errno = ret;
1635 if (ret == ETIMEDOUT) {
1636 /*
1637 * Call has timed out so we kill the kconsumerd_thread and return
1638 * an error.
1639 */
1640 ERR("Condition timed out. The consumer thread was never ready."
1641 " Killing it");
1642 ret = pthread_cancel(consumer_data->thread);
1643 if (ret < 0) {
1644 PERROR("pthread_cancel consumer thread");
1645 }
1646 } else {
1647 PERROR("pthread_cond_wait failed consumer thread");
1648 }
1649 goto error;
1650 }
1651
1652 pthread_mutex_lock(&consumer_data->pid_mutex);
1653 if (consumer_data->pid == 0) {
1654 ERR("Consumerd did not start");
1655 pthread_mutex_unlock(&consumer_data->pid_mutex);
1656 goto error;
1657 }
1658 pthread_mutex_unlock(&consumer_data->pid_mutex);
1659
1660 return 0;
1661
1662 error:
1663 return ret;
1664 }
1665
1666 /*
1667 * Join consumer thread
1668 */
1669 static int join_consumer_thread(struct consumer_data *consumer_data)
1670 {
1671 void *status;
1672 int ret;
1673
1674 /* Consumer pid must be a real one. */
1675 if (consumer_data->pid > 0) {
1676 ret = kill(consumer_data->pid, SIGTERM);
1677 if (ret) {
1678 ERR("Error killing consumer daemon");
1679 return ret;
1680 }
1681 return pthread_join(consumer_data->thread, &status);
1682 } else {
1683 return 0;
1684 }
1685 }
1686
1687 /*
1688 * Fork and exec a consumer daemon (consumerd).
1689 *
1690 * Return pid if successful else -1.
1691 */
1692 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1693 {
1694 int ret;
1695 pid_t pid;
1696 const char *consumer_to_use;
1697 const char *verbosity;
1698 struct stat st;
1699
1700 DBG("Spawning consumerd");
1701
1702 pid = fork();
1703 if (pid == 0) {
1704 /*
1705 * Exec consumerd.
1706 */
1707 if (opt_verbose_consumer) {
1708 verbosity = "--verbose";
1709 } else {
1710 verbosity = "--quiet";
1711 }
1712 switch (consumer_data->type) {
1713 case LTTNG_CONSUMER_KERNEL:
1714 /*
1715 * Find out which consumerd to execute. We will first try the
1716 * 64-bit path, then the sessiond's installation directory, and
1717 * fallback on the 32-bit one,
1718 */
1719 DBG3("Looking for a kernel consumer at these locations:");
1720 DBG3(" 1) %s", consumerd64_bin);
1721 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1722 DBG3(" 3) %s", consumerd32_bin);
1723 if (stat(consumerd64_bin, &st) == 0) {
1724 DBG3("Found location #1");
1725 consumer_to_use = consumerd64_bin;
1726 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1727 DBG3("Found location #2");
1728 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1729 } else if (stat(consumerd32_bin, &st) == 0) {
1730 DBG3("Found location #3");
1731 consumer_to_use = consumerd32_bin;
1732 } else {
1733 DBG("Could not find any valid consumerd executable");
1734 break;
1735 }
1736 DBG("Using kernel consumer at: %s", consumer_to_use);
1737 execl(consumer_to_use,
1738 "lttng-consumerd", verbosity, "-k",
1739 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1740 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1741 NULL);
1742 break;
1743 case LTTNG_CONSUMER64_UST:
1744 {
1745 char *tmpnew = NULL;
1746
1747 if (consumerd64_libdir[0] != '\0') {
1748 char *tmp;
1749 size_t tmplen;
1750
1751 tmp = getenv("LD_LIBRARY_PATH");
1752 if (!tmp) {
1753 tmp = "";
1754 }
1755 tmplen = strlen("LD_LIBRARY_PATH=")
1756 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1757 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1758 if (!tmpnew) {
1759 ret = -ENOMEM;
1760 goto error;
1761 }
1762 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1763 strcat(tmpnew, consumerd64_libdir);
1764 if (tmp[0] != '\0') {
1765 strcat(tmpnew, ":");
1766 strcat(tmpnew, tmp);
1767 }
1768 ret = putenv(tmpnew);
1769 if (ret) {
1770 ret = -errno;
1771 goto error;
1772 }
1773 }
1774 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1775 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1776 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1777 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1778 NULL);
1779 if (consumerd64_libdir[0] != '\0') {
1780 free(tmpnew);
1781 }
1782 if (ret) {
1783 goto error;
1784 }
1785 break;
1786 }
1787 case LTTNG_CONSUMER32_UST:
1788 {
1789 char *tmpnew = NULL;
1790
1791 if (consumerd32_libdir[0] != '\0') {
1792 char *tmp;
1793 size_t tmplen;
1794
1795 tmp = getenv("LD_LIBRARY_PATH");
1796 if (!tmp) {
1797 tmp = "";
1798 }
1799 tmplen = strlen("LD_LIBRARY_PATH=")
1800 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1801 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1802 if (!tmpnew) {
1803 ret = -ENOMEM;
1804 goto error;
1805 }
1806 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1807 strcat(tmpnew, consumerd32_libdir);
1808 if (tmp[0] != '\0') {
1809 strcat(tmpnew, ":");
1810 strcat(tmpnew, tmp);
1811 }
1812 ret = putenv(tmpnew);
1813 if (ret) {
1814 ret = -errno;
1815 goto error;
1816 }
1817 }
1818 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1819 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1820 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1821 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1822 NULL);
1823 if (consumerd32_libdir[0] != '\0') {
1824 free(tmpnew);
1825 }
1826 if (ret) {
1827 goto error;
1828 }
1829 break;
1830 }
1831 default:
1832 PERROR("unknown consumer type");
1833 exit(EXIT_FAILURE);
1834 }
1835 if (errno != 0) {
1836 PERROR("kernel start consumer exec");
1837 }
1838 exit(EXIT_FAILURE);
1839 } else if (pid > 0) {
1840 ret = pid;
1841 } else {
1842 PERROR("start consumer fork");
1843 ret = -errno;
1844 }
1845 error:
1846 return ret;
1847 }
1848
1849 /*
1850 * Spawn the consumerd daemon and session daemon thread.
1851 */
1852 static int start_consumerd(struct consumer_data *consumer_data)
1853 {
1854 int ret, err;
1855
1856 /*
1857 * Set the listen() state on the socket since there is a possible race
1858 * between the exec() of the consumer daemon and this call if place in the
1859 * consumer thread. See bug #366 for more details.
1860 */
1861 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1862 if (ret < 0) {
1863 goto error;
1864 }
1865
1866 pthread_mutex_lock(&consumer_data->pid_mutex);
1867 if (consumer_data->pid != 0) {
1868 pthread_mutex_unlock(&consumer_data->pid_mutex);
1869 goto end;
1870 }
1871
1872 ret = spawn_consumerd(consumer_data);
1873 if (ret < 0) {
1874 ERR("Spawning consumerd failed");
1875 pthread_mutex_unlock(&consumer_data->pid_mutex);
1876 goto error;
1877 }
1878
1879 /* Setting up the consumer_data pid */
1880 consumer_data->pid = ret;
1881 DBG2("Consumer pid %d", consumer_data->pid);
1882 pthread_mutex_unlock(&consumer_data->pid_mutex);
1883
1884 DBG2("Spawning consumer control thread");
1885 ret = spawn_consumer_thread(consumer_data);
1886 if (ret < 0) {
1887 ERR("Fatal error spawning consumer control thread");
1888 goto error;
1889 }
1890
1891 end:
1892 return 0;
1893
1894 error:
1895 /* Cleanup already created socket on error. */
1896 if (consumer_data->err_sock >= 0) {
1897 err = close(consumer_data->err_sock);
1898 if (err < 0) {
1899 PERROR("close consumer data error socket");
1900 }
1901 }
1902 return ret;
1903 }
1904
1905 /*
1906 * Compute health status of each consumer. If one of them is zero (bad
1907 * state), we return 0.
1908 */
1909 static int check_consumer_health(void)
1910 {
1911 int ret;
1912
1913 ret = health_check_state(&kconsumer_data.health) &&
1914 health_check_state(&ustconsumer32_data.health) &&
1915 health_check_state(&ustconsumer64_data.health);
1916
1917 DBG3("Health consumer check %d", ret);
1918
1919 return ret;
1920 }
1921
1922 /*
1923 * Setup necessary data for kernel tracer action.
1924 */
1925 static int init_kernel_tracer(void)
1926 {
1927 int ret;
1928
1929 /* Modprobe lttng kernel modules */
1930 ret = modprobe_lttng_control();
1931 if (ret < 0) {
1932 goto error;
1933 }
1934
1935 /* Open debugfs lttng */
1936 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1937 if (kernel_tracer_fd < 0) {
1938 DBG("Failed to open %s", module_proc_lttng);
1939 ret = -1;
1940 goto error_open;
1941 }
1942
1943 /* Validate kernel version */
1944 ret = kernel_validate_version(kernel_tracer_fd);
1945 if (ret < 0) {
1946 goto error_version;
1947 }
1948
1949 ret = modprobe_lttng_data();
1950 if (ret < 0) {
1951 goto error_modules;
1952 }
1953
1954 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1955 return 0;
1956
1957 error_version:
1958 modprobe_remove_lttng_control();
1959 ret = close(kernel_tracer_fd);
1960 if (ret) {
1961 PERROR("close");
1962 }
1963 kernel_tracer_fd = -1;
1964 return LTTNG_ERR_KERN_VERSION;
1965
1966 error_modules:
1967 ret = close(kernel_tracer_fd);
1968 if (ret) {
1969 PERROR("close");
1970 }
1971
1972 error_open:
1973 modprobe_remove_lttng_control();
1974
1975 error:
1976 WARN("No kernel tracer available");
1977 kernel_tracer_fd = -1;
1978 if (!is_root) {
1979 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1980 } else {
1981 return LTTNG_ERR_KERN_NA;
1982 }
1983 }
1984
1985
1986 /*
1987 * Copy consumer output from the tracing session to the domain session. The
1988 * function also applies the right modification on a per domain basis for the
1989 * trace files destination directory.
1990 */
1991 static int copy_session_consumer(int domain, struct ltt_session *session)
1992 {
1993 int ret;
1994 const char *dir_name;
1995 struct consumer_output *consumer;
1996
1997 assert(session);
1998 assert(session->consumer);
1999
2000 switch (domain) {
2001 case LTTNG_DOMAIN_KERNEL:
2002 DBG3("Copying tracing session consumer output in kernel session");
2003 /*
2004 * XXX: We should audit the session creation and what this function
2005 * does "extra" in order to avoid a destroy since this function is used
2006 * in the domain session creation (kernel and ust) only. Same for UST
2007 * domain.
2008 */
2009 if (session->kernel_session->consumer) {
2010 consumer_destroy_output(session->kernel_session->consumer);
2011 }
2012 session->kernel_session->consumer =
2013 consumer_copy_output(session->consumer);
2014 /* Ease our life a bit for the next part */
2015 consumer = session->kernel_session->consumer;
2016 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2017 break;
2018 case LTTNG_DOMAIN_UST:
2019 DBG3("Copying tracing session consumer output in UST session");
2020 if (session->ust_session->consumer) {
2021 consumer_destroy_output(session->ust_session->consumer);
2022 }
2023 session->ust_session->consumer =
2024 consumer_copy_output(session->consumer);
2025 /* Ease our life a bit for the next part */
2026 consumer = session->ust_session->consumer;
2027 dir_name = DEFAULT_UST_TRACE_DIR;
2028 break;
2029 default:
2030 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2031 goto error;
2032 }
2033
2034 /* Append correct directory to subdir */
2035 strncat(consumer->subdir, dir_name,
2036 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2037 DBG3("Copy session consumer subdir %s", consumer->subdir);
2038
2039 ret = LTTNG_OK;
2040
2041 error:
2042 return ret;
2043 }
2044
2045 /*
2046 * Create an UST session and add it to the session ust list.
2047 */
2048 static int create_ust_session(struct ltt_session *session,
2049 struct lttng_domain *domain)
2050 {
2051 int ret;
2052 struct ltt_ust_session *lus = NULL;
2053
2054 assert(session);
2055 assert(domain);
2056 assert(session->consumer);
2057
2058 switch (domain->type) {
2059 case LTTNG_DOMAIN_UST:
2060 break;
2061 default:
2062 ERR("Unknown UST domain on create session %d", domain->type);
2063 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2064 goto error;
2065 }
2066
2067 DBG("Creating UST session");
2068
2069 lus = trace_ust_create_session(session->path, session->id, domain);
2070 if (lus == NULL) {
2071 ret = LTTNG_ERR_UST_SESS_FAIL;
2072 goto error;
2073 }
2074
2075 lus->uid = session->uid;
2076 lus->gid = session->gid;
2077 session->ust_session = lus;
2078
2079 /* Copy session output to the newly created UST session */
2080 ret = copy_session_consumer(domain->type, session);
2081 if (ret != LTTNG_OK) {
2082 goto error;
2083 }
2084
2085 return LTTNG_OK;
2086
2087 error:
2088 free(lus);
2089 session->ust_session = NULL;
2090 return ret;
2091 }
2092
2093 /*
2094 * Create a kernel tracer session then create the default channel.
2095 */
2096 static int create_kernel_session(struct ltt_session *session)
2097 {
2098 int ret;
2099
2100 DBG("Creating kernel session");
2101
2102 ret = kernel_create_session(session, kernel_tracer_fd);
2103 if (ret < 0) {
2104 ret = LTTNG_ERR_KERN_SESS_FAIL;
2105 goto error;
2106 }
2107
2108 /* Code flow safety */
2109 assert(session->kernel_session);
2110
2111 /* Copy session output to the newly created Kernel session */
2112 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2113 if (ret != LTTNG_OK) {
2114 goto error;
2115 }
2116
2117 /* Create directory(ies) on local filesystem. */
2118 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2119 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2120 ret = run_as_mkdir_recursive(
2121 session->kernel_session->consumer->dst.trace_path,
2122 S_IRWXU | S_IRWXG, session->uid, session->gid);
2123 if (ret < 0) {
2124 if (ret != -EEXIST) {
2125 ERR("Trace directory creation error");
2126 goto error;
2127 }
2128 }
2129 }
2130
2131 session->kernel_session->uid = session->uid;
2132 session->kernel_session->gid = session->gid;
2133
2134 return LTTNG_OK;
2135
2136 error:
2137 trace_kernel_destroy_session(session->kernel_session);
2138 session->kernel_session = NULL;
2139 return ret;
2140 }
2141
2142 /*
2143 * Count number of session permitted by uid/gid.
2144 */
2145 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2146 {
2147 unsigned int i = 0;
2148 struct ltt_session *session;
2149
2150 DBG("Counting number of available session for UID %d GID %d",
2151 uid, gid);
2152 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2153 /*
2154 * Only list the sessions the user can control.
2155 */
2156 if (!session_access_ok(session, uid, gid)) {
2157 continue;
2158 }
2159 i++;
2160 }
2161 return i;
2162 }
2163
2164 /*
2165 * Process the command requested by the lttng client within the command
2166 * context structure. This function make sure that the return structure (llm)
2167 * is set and ready for transmission before returning.
2168 *
2169 * Return any error encountered or 0 for success.
2170 *
2171 * "sock" is only used for special-case var. len data.
2172 */
2173 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2174 int *sock_error)
2175 {
2176 int ret = LTTNG_OK;
2177 int need_tracing_session = 1;
2178 int need_domain;
2179
2180 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2181
2182 *sock_error = 0;
2183
2184 switch (cmd_ctx->lsm->cmd_type) {
2185 case LTTNG_CREATE_SESSION:
2186 case LTTNG_DESTROY_SESSION:
2187 case LTTNG_LIST_SESSIONS:
2188 case LTTNG_LIST_DOMAINS:
2189 case LTTNG_START_TRACE:
2190 case LTTNG_STOP_TRACE:
2191 case LTTNG_DATA_PENDING:
2192 need_domain = 0;
2193 break;
2194 default:
2195 need_domain = 1;
2196 }
2197
2198 if (opt_no_kernel && need_domain
2199 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2200 if (!is_root) {
2201 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2202 } else {
2203 ret = LTTNG_ERR_KERN_NA;
2204 }
2205 goto error;
2206 }
2207
2208 /* Deny register consumer if we already have a spawned consumer. */
2209 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2210 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2211 if (kconsumer_data.pid > 0) {
2212 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2213 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2214 goto error;
2215 }
2216 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2217 }
2218
2219 /*
2220 * Check for command that don't needs to allocate a returned payload. We do
2221 * this here so we don't have to make the call for no payload at each
2222 * command.
2223 */
2224 switch(cmd_ctx->lsm->cmd_type) {
2225 case LTTNG_LIST_SESSIONS:
2226 case LTTNG_LIST_TRACEPOINTS:
2227 case LTTNG_LIST_TRACEPOINT_FIELDS:
2228 case LTTNG_LIST_DOMAINS:
2229 case LTTNG_LIST_CHANNELS:
2230 case LTTNG_LIST_EVENTS:
2231 break;
2232 default:
2233 /* Setup lttng message with no payload */
2234 ret = setup_lttng_msg(cmd_ctx, 0);
2235 if (ret < 0) {
2236 /* This label does not try to unlock the session */
2237 goto init_setup_error;
2238 }
2239 }
2240
2241 /* Commands that DO NOT need a session. */
2242 switch (cmd_ctx->lsm->cmd_type) {
2243 case LTTNG_CREATE_SESSION:
2244 case LTTNG_CALIBRATE:
2245 case LTTNG_LIST_SESSIONS:
2246 case LTTNG_LIST_TRACEPOINTS:
2247 case LTTNG_LIST_TRACEPOINT_FIELDS:
2248 need_tracing_session = 0;
2249 break;
2250 default:
2251 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2252 /*
2253 * We keep the session list lock across _all_ commands
2254 * for now, because the per-session lock does not
2255 * handle teardown properly.
2256 */
2257 session_lock_list();
2258 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2259 if (cmd_ctx->session == NULL) {
2260 if (cmd_ctx->lsm->session.name != NULL) {
2261 ret = LTTNG_ERR_SESS_NOT_FOUND;
2262 } else {
2263 /* If no session name specified */
2264 ret = LTTNG_ERR_SELECT_SESS;
2265 }
2266 goto error;
2267 } else {
2268 /* Acquire lock for the session */
2269 session_lock(cmd_ctx->session);
2270 }
2271 break;
2272 }
2273
2274 if (!need_domain) {
2275 goto skip_domain;
2276 }
2277
2278 /*
2279 * Check domain type for specific "pre-action".
2280 */
2281 switch (cmd_ctx->lsm->domain.type) {
2282 case LTTNG_DOMAIN_KERNEL:
2283 if (!is_root) {
2284 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2285 goto error;
2286 }
2287
2288 /* Kernel tracer check */
2289 if (kernel_tracer_fd == -1) {
2290 /* Basically, load kernel tracer modules */
2291 ret = init_kernel_tracer();
2292 if (ret != 0) {
2293 goto error;
2294 }
2295 }
2296
2297 /* Consumer is in an ERROR state. Report back to client */
2298 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2299 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2300 goto error;
2301 }
2302
2303 /* Need a session for kernel command */
2304 if (need_tracing_session) {
2305 if (cmd_ctx->session->kernel_session == NULL) {
2306 ret = create_kernel_session(cmd_ctx->session);
2307 if (ret < 0) {
2308 ret = LTTNG_ERR_KERN_SESS_FAIL;
2309 goto error;
2310 }
2311 }
2312
2313 /* Start the kernel consumer daemon */
2314 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2315 if (kconsumer_data.pid == 0 &&
2316 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2317 cmd_ctx->session->start_consumer) {
2318 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2319 ret = start_consumerd(&kconsumer_data);
2320 if (ret < 0) {
2321 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2322 goto error;
2323 }
2324 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2325 } else {
2326 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2327 }
2328
2329 /*
2330 * The consumer was just spawned so we need to add the socket to
2331 * the consumer output of the session if exist.
2332 */
2333 ret = consumer_create_socket(&kconsumer_data,
2334 cmd_ctx->session->kernel_session->consumer);
2335 if (ret < 0) {
2336 goto error;
2337 }
2338 }
2339
2340 break;
2341 case LTTNG_DOMAIN_UST:
2342 {
2343 /* Consumer is in an ERROR state. Report back to client */
2344 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2345 ret = LTTNG_ERR_NO_USTCONSUMERD;
2346 goto error;
2347 }
2348
2349 if (need_tracing_session) {
2350 /* Create UST session if none exist. */
2351 if (cmd_ctx->session->ust_session == NULL) {
2352 ret = create_ust_session(cmd_ctx->session,
2353 &cmd_ctx->lsm->domain);
2354 if (ret != LTTNG_OK) {
2355 goto error;
2356 }
2357 }
2358
2359 /* Start the UST consumer daemons */
2360 /* 64-bit */
2361 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2362 if (consumerd64_bin[0] != '\0' &&
2363 ustconsumer64_data.pid == 0 &&
2364 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2365 cmd_ctx->session->start_consumer) {
2366 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2367 ret = start_consumerd(&ustconsumer64_data);
2368 if (ret < 0) {
2369 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2370 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2371 goto error;
2372 }
2373
2374 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2375 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2376 } else {
2377 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2378 }
2379
2380 /*
2381 * Setup socket for consumer 64 bit. No need for atomic access
2382 * since it was set above and can ONLY be set in this thread.
2383 */
2384 ret = consumer_create_socket(&ustconsumer64_data,
2385 cmd_ctx->session->ust_session->consumer);
2386 if (ret < 0) {
2387 goto error;
2388 }
2389
2390 /* 32-bit */
2391 if (consumerd32_bin[0] != '\0' &&
2392 ustconsumer32_data.pid == 0 &&
2393 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2394 cmd_ctx->session->start_consumer) {
2395 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2396 ret = start_consumerd(&ustconsumer32_data);
2397 if (ret < 0) {
2398 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2399 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2400 goto error;
2401 }
2402
2403 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2404 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2405 } else {
2406 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2407 }
2408
2409 /*
2410 * Setup socket for consumer 64 bit. No need for atomic access
2411 * since it was set above and can ONLY be set in this thread.
2412 */
2413 ret = consumer_create_socket(&ustconsumer32_data,
2414 cmd_ctx->session->ust_session->consumer);
2415 if (ret < 0) {
2416 goto error;
2417 }
2418 }
2419 break;
2420 }
2421 default:
2422 break;
2423 }
2424 skip_domain:
2425
2426 /* Validate consumer daemon state when start/stop trace command */
2427 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2428 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2429 switch (cmd_ctx->lsm->domain.type) {
2430 case LTTNG_DOMAIN_UST:
2431 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2432 ret = LTTNG_ERR_NO_USTCONSUMERD;
2433 goto error;
2434 }
2435 break;
2436 case LTTNG_DOMAIN_KERNEL:
2437 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2438 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2439 goto error;
2440 }
2441 break;
2442 }
2443 }
2444
2445 /*
2446 * Check that the UID or GID match that of the tracing session.
2447 * The root user can interact with all sessions.
2448 */
2449 if (need_tracing_session) {
2450 if (!session_access_ok(cmd_ctx->session,
2451 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2452 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2453 ret = LTTNG_ERR_EPERM;
2454 goto error;
2455 }
2456 }
2457
2458 /* Process by command type */
2459 switch (cmd_ctx->lsm->cmd_type) {
2460 case LTTNG_ADD_CONTEXT:
2461 {
2462 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2463 cmd_ctx->lsm->u.context.channel_name,
2464 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2465 break;
2466 }
2467 case LTTNG_DISABLE_CHANNEL:
2468 {
2469 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2470 cmd_ctx->lsm->u.disable.channel_name);
2471 break;
2472 }
2473 case LTTNG_DISABLE_EVENT:
2474 {
2475 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2476 cmd_ctx->lsm->u.disable.channel_name,
2477 cmd_ctx->lsm->u.disable.name);
2478 break;
2479 }
2480 case LTTNG_DISABLE_ALL_EVENT:
2481 {
2482 DBG("Disabling all events");
2483
2484 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2485 cmd_ctx->lsm->u.disable.channel_name);
2486 break;
2487 }
2488 case LTTNG_DISABLE_CONSUMER:
2489 {
2490 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2491 break;
2492 }
2493 case LTTNG_ENABLE_CHANNEL:
2494 {
2495 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2496 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2497 break;
2498 }
2499 case LTTNG_ENABLE_CONSUMER:
2500 {
2501 /*
2502 * XXX: 0 means that this URI should be applied on the session. Should
2503 * be a DOMAIN enuam.
2504 */
2505 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2506 if (ret != LTTNG_OK) {
2507 goto error;
2508 }
2509
2510 if (cmd_ctx->lsm->domain.type == 0) {
2511 /* Add the URI for the UST session if a consumer is present. */
2512 if (cmd_ctx->session->ust_session &&
2513 cmd_ctx->session->ust_session->consumer) {
2514 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2515 } else if (cmd_ctx->session->kernel_session &&
2516 cmd_ctx->session->kernel_session->consumer) {
2517 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2518 cmd_ctx->session);
2519 }
2520 }
2521 break;
2522 }
2523 case LTTNG_ENABLE_EVENT:
2524 {
2525 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2526 cmd_ctx->lsm->u.enable.channel_name,
2527 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2528 break;
2529 }
2530 case LTTNG_ENABLE_ALL_EVENT:
2531 {
2532 DBG("Enabling all events");
2533
2534 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2535 cmd_ctx->lsm->u.enable.channel_name,
2536 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2537 break;
2538 }
2539 case LTTNG_LIST_TRACEPOINTS:
2540 {
2541 struct lttng_event *events;
2542 ssize_t nb_events;
2543
2544 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2545 if (nb_events < 0) {
2546 /* Return value is a negative lttng_error_code. */
2547 ret = -nb_events;
2548 goto error;
2549 }
2550
2551 /*
2552 * Setup lttng message with payload size set to the event list size in
2553 * bytes and then copy list into the llm payload.
2554 */
2555 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2556 if (ret < 0) {
2557 free(events);
2558 goto setup_error;
2559 }
2560
2561 /* Copy event list into message payload */
2562 memcpy(cmd_ctx->llm->payload, events,
2563 sizeof(struct lttng_event) * nb_events);
2564
2565 free(events);
2566
2567 ret = LTTNG_OK;
2568 break;
2569 }
2570 case LTTNG_LIST_TRACEPOINT_FIELDS:
2571 {
2572 struct lttng_event_field *fields;
2573 ssize_t nb_fields;
2574
2575 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2576 &fields);
2577 if (nb_fields < 0) {
2578 /* Return value is a negative lttng_error_code. */
2579 ret = -nb_fields;
2580 goto error;
2581 }
2582
2583 /*
2584 * Setup lttng message with payload size set to the event list size in
2585 * bytes and then copy list into the llm payload.
2586 */
2587 ret = setup_lttng_msg(cmd_ctx,
2588 sizeof(struct lttng_event_field) * nb_fields);
2589 if (ret < 0) {
2590 free(fields);
2591 goto setup_error;
2592 }
2593
2594 /* Copy event list into message payload */
2595 memcpy(cmd_ctx->llm->payload, fields,
2596 sizeof(struct lttng_event_field) * nb_fields);
2597
2598 free(fields);
2599
2600 ret = LTTNG_OK;
2601 break;
2602 }
2603 case LTTNG_SET_CONSUMER_URI:
2604 {
2605 size_t nb_uri, len;
2606 struct lttng_uri *uris;
2607
2608 nb_uri = cmd_ctx->lsm->u.uri.size;
2609 len = nb_uri * sizeof(struct lttng_uri);
2610
2611 if (nb_uri == 0) {
2612 ret = LTTNG_ERR_INVALID;
2613 goto error;
2614 }
2615
2616 uris = zmalloc(len);
2617 if (uris == NULL) {
2618 ret = LTTNG_ERR_FATAL;
2619 goto error;
2620 }
2621
2622 /* Receive variable len data */
2623 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2624 ret = lttcomm_recv_unix_sock(sock, uris, len);
2625 if (ret <= 0) {
2626 DBG("No URIs received from client... continuing");
2627 *sock_error = 1;
2628 ret = LTTNG_ERR_SESSION_FAIL;
2629 free(uris);
2630 goto error;
2631 }
2632
2633 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2634 nb_uri, uris);
2635 if (ret != LTTNG_OK) {
2636 free(uris);
2637 goto error;
2638 }
2639
2640 /*
2641 * XXX: 0 means that this URI should be applied on the session. Should
2642 * be a DOMAIN enuam.
2643 */
2644 if (cmd_ctx->lsm->domain.type == 0) {
2645 /* Add the URI for the UST session if a consumer is present. */
2646 if (cmd_ctx->session->ust_session &&
2647 cmd_ctx->session->ust_session->consumer) {
2648 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2649 nb_uri, uris);
2650 } else if (cmd_ctx->session->kernel_session &&
2651 cmd_ctx->session->kernel_session->consumer) {
2652 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2653 cmd_ctx->session, nb_uri, uris);
2654 }
2655 }
2656
2657 free(uris);
2658
2659 break;
2660 }
2661 case LTTNG_START_TRACE:
2662 {
2663 ret = cmd_start_trace(cmd_ctx->session);
2664 break;
2665 }
2666 case LTTNG_STOP_TRACE:
2667 {
2668 ret = cmd_stop_trace(cmd_ctx->session);
2669 break;
2670 }
2671 case LTTNG_CREATE_SESSION:
2672 {
2673 size_t nb_uri, len;
2674 struct lttng_uri *uris = NULL;
2675
2676 nb_uri = cmd_ctx->lsm->u.uri.size;
2677 len = nb_uri * sizeof(struct lttng_uri);
2678
2679 if (nb_uri > 0) {
2680 uris = zmalloc(len);
2681 if (uris == NULL) {
2682 ret = LTTNG_ERR_FATAL;
2683 goto error;
2684 }
2685
2686 /* Receive variable len data */
2687 DBG("Waiting for %zu URIs from client ...", nb_uri);
2688 ret = lttcomm_recv_unix_sock(sock, uris, len);
2689 if (ret <= 0) {
2690 DBG("No URIs received from client... continuing");
2691 *sock_error = 1;
2692 ret = LTTNG_ERR_SESSION_FAIL;
2693 free(uris);
2694 goto error;
2695 }
2696
2697 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2698 DBG("Creating session with ONE network URI is a bad call");
2699 ret = LTTNG_ERR_SESSION_FAIL;
2700 free(uris);
2701 goto error;
2702 }
2703 }
2704
2705 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2706 &cmd_ctx->creds);
2707
2708 free(uris);
2709
2710 break;
2711 }
2712 case LTTNG_DESTROY_SESSION:
2713 {
2714 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2715
2716 /* Set session to NULL so we do not unlock it after free. */
2717 cmd_ctx->session = NULL;
2718 break;
2719 }
2720 case LTTNG_LIST_DOMAINS:
2721 {
2722 ssize_t nb_dom;
2723 struct lttng_domain *domains;
2724
2725 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2726 if (nb_dom < 0) {
2727 /* Return value is a negative lttng_error_code. */
2728 ret = -nb_dom;
2729 goto error;
2730 }
2731
2732 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2733 if (ret < 0) {
2734 goto setup_error;
2735 }
2736
2737 /* Copy event list into message payload */
2738 memcpy(cmd_ctx->llm->payload, domains,
2739 nb_dom * sizeof(struct lttng_domain));
2740
2741 free(domains);
2742
2743 ret = LTTNG_OK;
2744 break;
2745 }
2746 case LTTNG_LIST_CHANNELS:
2747 {
2748 int nb_chan;
2749 struct lttng_channel *channels;
2750
2751 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2752 cmd_ctx->session, &channels);
2753 if (nb_chan < 0) {
2754 /* Return value is a negative lttng_error_code. */
2755 ret = -nb_chan;
2756 goto error;
2757 }
2758
2759 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2760 if (ret < 0) {
2761 goto setup_error;
2762 }
2763
2764 /* Copy event list into message payload */
2765 memcpy(cmd_ctx->llm->payload, channels,
2766 nb_chan * sizeof(struct lttng_channel));
2767
2768 free(channels);
2769
2770 ret = LTTNG_OK;
2771 break;
2772 }
2773 case LTTNG_LIST_EVENTS:
2774 {
2775 ssize_t nb_event;
2776 struct lttng_event *events = NULL;
2777
2778 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2779 cmd_ctx->lsm->u.list.channel_name, &events);
2780 if (nb_event < 0) {
2781 /* Return value is a negative lttng_error_code. */
2782 ret = -nb_event;
2783 goto error;
2784 }
2785
2786 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2787 if (ret < 0) {
2788 goto setup_error;
2789 }
2790
2791 /* Copy event list into message payload */
2792 memcpy(cmd_ctx->llm->payload, events,
2793 nb_event * sizeof(struct lttng_event));
2794
2795 free(events);
2796
2797 ret = LTTNG_OK;
2798 break;
2799 }
2800 case LTTNG_LIST_SESSIONS:
2801 {
2802 unsigned int nr_sessions;
2803
2804 session_lock_list();
2805 nr_sessions = lttng_sessions_count(
2806 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2807 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2808
2809 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2810 if (ret < 0) {
2811 session_unlock_list();
2812 goto setup_error;
2813 }
2814
2815 /* Filled the session array */
2816 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2817 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2818 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2819
2820 session_unlock_list();
2821
2822 ret = LTTNG_OK;
2823 break;
2824 }
2825 case LTTNG_CALIBRATE:
2826 {
2827 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2828 &cmd_ctx->lsm->u.calibrate);
2829 break;
2830 }
2831 case LTTNG_REGISTER_CONSUMER:
2832 {
2833 struct consumer_data *cdata;
2834
2835 switch (cmd_ctx->lsm->domain.type) {
2836 case LTTNG_DOMAIN_KERNEL:
2837 cdata = &kconsumer_data;
2838 break;
2839 default:
2840 ret = LTTNG_ERR_UND;
2841 goto error;
2842 }
2843
2844 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2845 cmd_ctx->lsm->u.reg.path, cdata);
2846 break;
2847 }
2848 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2849 {
2850 struct lttng_filter_bytecode *bytecode;
2851
2852 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2853 ret = LTTNG_ERR_FILTER_INVAL;
2854 goto error;
2855 }
2856 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2857 ret = LTTNG_ERR_FILTER_INVAL;
2858 goto error;
2859 }
2860 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2861 if (!bytecode) {
2862 ret = LTTNG_ERR_FILTER_NOMEM;
2863 goto error;
2864 }
2865 /* Receive var. len. data */
2866 DBG("Receiving var len data from client ...");
2867 ret = lttcomm_recv_unix_sock(sock, bytecode,
2868 cmd_ctx->lsm->u.enable.bytecode_len);
2869 if (ret <= 0) {
2870 DBG("Nothing recv() from client var len data... continuing");
2871 *sock_error = 1;
2872 ret = LTTNG_ERR_FILTER_INVAL;
2873 goto error;
2874 }
2875
2876 if (bytecode->len + sizeof(*bytecode)
2877 != cmd_ctx->lsm->u.enable.bytecode_len) {
2878 free(bytecode);
2879 ret = LTTNG_ERR_FILTER_INVAL;
2880 goto error;
2881 }
2882
2883 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2884 cmd_ctx->lsm->u.enable.channel_name,
2885 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2886 break;
2887 }
2888 case LTTNG_DATA_PENDING:
2889 {
2890 ret = cmd_data_pending(cmd_ctx->session);
2891 break;
2892 }
2893 default:
2894 ret = LTTNG_ERR_UND;
2895 break;
2896 }
2897
2898 error:
2899 if (cmd_ctx->llm == NULL) {
2900 DBG("Missing llm structure. Allocating one.");
2901 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2902 goto setup_error;
2903 }
2904 }
2905 /* Set return code */
2906 cmd_ctx->llm->ret_code = ret;
2907 setup_error:
2908 if (cmd_ctx->session) {
2909 session_unlock(cmd_ctx->session);
2910 }
2911 if (need_tracing_session) {
2912 session_unlock_list();
2913 }
2914 init_setup_error:
2915 return ret;
2916 }
2917
2918 /*
2919 * Thread managing health check socket.
2920 */
2921 static void *thread_manage_health(void *data)
2922 {
2923 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2924 uint32_t revents, nb_fd;
2925 struct lttng_poll_event events;
2926 struct lttcomm_health_msg msg;
2927 struct lttcomm_health_data reply;
2928
2929 DBG("[thread] Manage health check started");
2930
2931 rcu_register_thread();
2932
2933 /* Create unix socket */
2934 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2935 if (sock < 0) {
2936 ERR("Unable to create health check Unix socket");
2937 ret = -1;
2938 goto error;
2939 }
2940
2941 /*
2942 * Set the CLOEXEC flag. Return code is useless because either way, the
2943 * show must go on.
2944 */
2945 (void) utils_set_fd_cloexec(sock);
2946
2947 ret = lttcomm_listen_unix_sock(sock);
2948 if (ret < 0) {
2949 goto error;
2950 }
2951
2952 /*
2953 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2954 * more will be added to this poll set.
2955 */
2956 ret = create_thread_poll_set(&events, 2);
2957 if (ret < 0) {
2958 goto error;
2959 }
2960
2961 /* Add the application registration socket */
2962 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2963 if (ret < 0) {
2964 goto error;
2965 }
2966
2967 while (1) {
2968 DBG("Health check ready");
2969
2970 nb_fd = LTTNG_POLL_GETNB(&events);
2971
2972 /* Inifinite blocking call, waiting for transmission */
2973 restart:
2974 ret = lttng_poll_wait(&events, -1);
2975 if (ret < 0) {
2976 /*
2977 * Restart interrupted system call.
2978 */
2979 if (errno == EINTR) {
2980 goto restart;
2981 }
2982 goto error;
2983 }
2984
2985 for (i = 0; i < nb_fd; i++) {
2986 /* Fetch once the poll data */
2987 revents = LTTNG_POLL_GETEV(&events, i);
2988 pollfd = LTTNG_POLL_GETFD(&events, i);
2989
2990 /* Thread quit pipe has been closed. Killing thread. */
2991 ret = check_thread_quit_pipe(pollfd, revents);
2992 if (ret) {
2993 err = 0;
2994 goto exit;
2995 }
2996
2997 /* Event on the registration socket */
2998 if (pollfd == sock) {
2999 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3000 ERR("Health socket poll error");
3001 goto error;
3002 }
3003 }
3004 }
3005
3006 new_sock = lttcomm_accept_unix_sock(sock);
3007 if (new_sock < 0) {
3008 goto error;
3009 }
3010
3011 /*
3012 * Set the CLOEXEC flag. Return code is useless because either way, the
3013 * show must go on.
3014 */
3015 (void) utils_set_fd_cloexec(new_sock);
3016
3017 DBG("Receiving data from client for health...");
3018 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3019 if (ret <= 0) {
3020 DBG("Nothing recv() from client... continuing");
3021 ret = close(new_sock);
3022 if (ret) {
3023 PERROR("close");
3024 }
3025 new_sock = -1;
3026 continue;
3027 }
3028
3029 rcu_thread_online();
3030
3031 switch (msg.component) {
3032 case LTTNG_HEALTH_CMD:
3033 reply.ret_code = health_check_state(&health_thread_cmd);
3034 break;
3035 case LTTNG_HEALTH_APP_MANAGE:
3036 reply.ret_code = health_check_state(&health_thread_app_manage);
3037 break;
3038 case LTTNG_HEALTH_APP_REG:
3039 reply.ret_code = health_check_state(&health_thread_app_reg);
3040 break;
3041 case LTTNG_HEALTH_KERNEL:
3042 reply.ret_code = health_check_state(&health_thread_kernel);
3043 break;
3044 case LTTNG_HEALTH_CONSUMER:
3045 reply.ret_code = check_consumer_health();
3046 break;
3047 case LTTNG_HEALTH_ALL:
3048 reply.ret_code =
3049 health_check_state(&health_thread_app_manage) &&
3050 health_check_state(&health_thread_app_reg) &&
3051 health_check_state(&health_thread_cmd) &&
3052 health_check_state(&health_thread_kernel) &&
3053 check_consumer_health();
3054 break;
3055 default:
3056 reply.ret_code = LTTNG_ERR_UND;
3057 break;
3058 }
3059
3060 /*
3061 * Flip ret value since 0 is a success and 1 indicates a bad health for
3062 * the client where in the sessiond it is the opposite. Again, this is
3063 * just to make things easier for us poor developer which enjoy a lot
3064 * lazyness.
3065 */
3066 if (reply.ret_code == 0 || reply.ret_code == 1) {
3067 reply.ret_code = !reply.ret_code;
3068 }
3069
3070 DBG2("Health check return value %d", reply.ret_code);
3071
3072 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3073 if (ret < 0) {
3074 ERR("Failed to send health data back to client");
3075 }
3076
3077 /* End of transmission */
3078 ret = close(new_sock);
3079 if (ret) {
3080 PERROR("close");
3081 }
3082 new_sock = -1;
3083 }
3084
3085 exit:
3086 error:
3087 if (err) {
3088 ERR("Health error occurred in %s", __func__);
3089 }
3090 DBG("Health check thread dying");
3091 unlink(health_unix_sock_path);
3092 if (sock >= 0) {
3093 ret = close(sock);
3094 if (ret) {
3095 PERROR("close");
3096 }
3097 }
3098 if (new_sock >= 0) {
3099 ret = close(new_sock);
3100 if (ret) {
3101 PERROR("close");
3102 }
3103 }
3104
3105 lttng_poll_clean(&events);
3106
3107 rcu_unregister_thread();
3108 return NULL;
3109 }
3110
3111 /*
3112 * This thread manage all clients request using the unix client socket for
3113 * communication.
3114 */
3115 static void *thread_manage_clients(void *data)
3116 {
3117 int sock = -1, ret, i, pollfd, err = -1;
3118 int sock_error;
3119 uint32_t revents, nb_fd;
3120 struct command_ctx *cmd_ctx = NULL;
3121 struct lttng_poll_event events;
3122
3123 DBG("[thread] Manage client started");
3124
3125 testpoint(thread_manage_clients);
3126
3127 rcu_register_thread();
3128
3129 health_code_update(&health_thread_cmd);
3130
3131 ret = lttcomm_listen_unix_sock(client_sock);
3132 if (ret < 0) {
3133 goto error_listen;
3134 }
3135
3136 /*
3137 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3138 * more will be added to this poll set.
3139 */
3140 ret = create_thread_poll_set(&events, 2);
3141 if (ret < 0) {
3142 goto error_create_poll;
3143 }
3144
3145 /* Add the application registration socket */
3146 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3147 if (ret < 0) {
3148 goto error;
3149 }
3150
3151 /*
3152 * Notify parent pid that we are ready to accept command for client side.
3153 */
3154 if (opt_sig_parent) {
3155 kill(ppid, SIGUSR1);
3156 }
3157
3158 testpoint(thread_manage_clients_before_loop);
3159
3160 health_code_update(&health_thread_cmd);
3161
3162 while (1) {
3163 DBG("Accepting client command ...");
3164
3165 nb_fd = LTTNG_POLL_GETNB(&events);
3166
3167 /* Inifinite blocking call, waiting for transmission */
3168 restart:
3169 health_poll_update(&health_thread_cmd);
3170 ret = lttng_poll_wait(&events, -1);
3171 health_poll_update(&health_thread_cmd);
3172 if (ret < 0) {
3173 /*
3174 * Restart interrupted system call.
3175 */
3176 if (errno == EINTR) {
3177 goto restart;
3178 }
3179 goto error;
3180 }
3181
3182 for (i = 0; i < nb_fd; i++) {
3183 /* Fetch once the poll data */
3184 revents = LTTNG_POLL_GETEV(&events, i);
3185 pollfd = LTTNG_POLL_GETFD(&events, i);
3186
3187 health_code_update(&health_thread_cmd);
3188
3189 /* Thread quit pipe has been closed. Killing thread. */
3190 ret = check_thread_quit_pipe(pollfd, revents);
3191 if (ret) {
3192 err = 0;
3193 goto exit;
3194 }
3195
3196 /* Event on the registration socket */
3197 if (pollfd == client_sock) {
3198 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3199 ERR("Client socket poll error");
3200 goto error;
3201 }
3202 }
3203 }
3204
3205 DBG("Wait for client response");
3206
3207 health_code_update(&health_thread_cmd);
3208
3209 sock = lttcomm_accept_unix_sock(client_sock);
3210 if (sock < 0) {
3211 goto error;
3212 }
3213
3214 /*
3215 * Set the CLOEXEC flag. Return code is useless because either way, the
3216 * show must go on.
3217 */
3218 (void) utils_set_fd_cloexec(sock);
3219
3220 /* Set socket option for credentials retrieval */
3221 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3222 if (ret < 0) {
3223 goto error;
3224 }
3225
3226 /* Allocate context command to process the client request */
3227 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3228 if (cmd_ctx == NULL) {
3229 PERROR("zmalloc cmd_ctx");