Fix: Teardown of thread_manage_clients on failure of listen/create_poll
[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 "filter.h"
63 #include "health.h"
64 #include "testpoint.h"
65
66 #define CONSUMERD_FILE "lttng-consumerd"
67
68 /* Const values */
69 const char default_home_dir[] = DEFAULT_HOME_DIR;
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
72 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
73
74 const char *progname;
75 const char *opt_tracing_group;
76 static int opt_sig_parent;
77 static int opt_verbose_consumer;
78 static int opt_daemon;
79 static int opt_no_kernel;
80 static int is_root; /* Set to 1 if the daemon is running as root */
81 static pid_t ppid; /* Parent PID for --sig-parent option */
82 static char *rundir;
83
84 /*
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
87 */
88 static struct consumer_data kconsumer_data = {
89 .type = LTTNG_CONSUMER_KERNEL,
90 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
91 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
92 .err_sock = -1,
93 .cmd_sock = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer32_data = {
111 .type = LTTNG_CONSUMER32_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121
122 /* Shared between threads */
123 static int dispatch_thread_exit;
124
125 /* Global application Unix socket path */
126 static char apps_unix_sock_path[PATH_MAX];
127 /* Global client Unix socket path */
128 static char client_unix_sock_path[PATH_MAX];
129 /* global wait shm path for UST */
130 static char wait_shm_path[PATH_MAX];
131 /* Global health check unix path */
132 static char health_unix_sock_path[PATH_MAX];
133
134 /* Sockets and FDs */
135 static int client_sock = -1;
136 static int apps_sock = -1;
137 int kernel_tracer_fd = -1;
138 static int kernel_poll_pipe[2] = { -1, -1 };
139
140 /*
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
143 */
144 static int thread_quit_pipe[2] = { -1, -1 };
145
146 /*
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
149 */
150 static int apps_cmd_pipe[2] = { -1, -1 };
151
152 /* Pthread, Mutexes and Semaphores */
153 static pthread_t apps_thread;
154 static pthread_t reg_apps_thread;
155 static pthread_t client_thread;
156 static pthread_t kernel_thread;
157 static pthread_t dispatch_thread;
158 static pthread_t health_thread;
159
160 /*
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 *
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
166 */
167 static struct ust_cmd_queue ust_cmd_queue;
168
169 /*
170 * Pointer initialized before thread creation.
171 *
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
175 *
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
178 */
179 static struct ltt_session_list *session_list_ptr;
180
181 int ust_consumerd64_fd = -1;
182 int ust_consumerd32_fd = -1;
183
184 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
185 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
186 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
187 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
188
189 static const char *module_proc_lttng = "/proc/lttng";
190
191 /*
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
194 */
195 enum consumerd_state {
196 CONSUMER_STARTED = 1,
197 CONSUMER_STOPPED = 2,
198 CONSUMER_ERROR = 3,
199 };
200
201 /*
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
206 *
207 * The following example shows a possible race condition of this scheme:
208 *
209 * consumer thread error happens
210 * client cmd arrives
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
214 * ...
215 *
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
221 */
222 static enum consumerd_state ust_consumerd_state;
223 static enum consumerd_state kernel_consumerd_state;
224
225 /* Used for the health monitoring of the session daemon. See health.h */
226 struct health_state health_thread_cmd;
227 struct health_state health_thread_app_manage;
228 struct health_state health_thread_app_reg;
229 struct health_state health_thread_kernel;
230
231 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 static
237 void setup_consumerd_path(void)
238 {
239 const char *bin, *libdir;
240
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
245 */
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
256 }
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
259 }
260 #else
261 #error "Unknown bitness"
262 #endif
263
264 /*
265 * runtime env. var. overrides the build default.
266 */
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
270 }
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
274 }
275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
280 if (libdir) {
281 consumerd64_libdir = libdir;
282 }
283 }
284
285 /*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288 static int create_thread_poll_set(struct lttng_poll_event *events,
289 unsigned int size)
290 {
291 int ret;
292
293 if (events == NULL || size == 0) {
294 ret = -1;
295 goto error;
296 }
297
298 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
299 if (ret < 0) {
300 goto error;
301 }
302
303 /* Add quit pipe */
304 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
305 if (ret < 0) {
306 goto error;
307 }
308
309 return 0;
310
311 error:
312 return ret;
313 }
314
315 /*
316 * Check if the thread quit pipe was triggered.
317 *
318 * Return 1 if it was triggered else 0;
319 */
320 static int check_thread_quit_pipe(int fd, uint32_t events)
321 {
322 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
323 return 1;
324 }
325
326 return 0;
327 }
328
329 /*
330 * Return group ID of the tracing group or -1 if not found.
331 */
332 static gid_t allowed_group(void)
333 {
334 struct group *grp;
335
336 if (opt_tracing_group) {
337 grp = getgrnam(opt_tracing_group);
338 } else {
339 grp = getgrnam(default_tracing_group);
340 }
341 if (!grp) {
342 return -1;
343 } else {
344 return grp->gr_gid;
345 }
346 }
347
348 /*
349 * Init thread quit pipe.
350 *
351 * Return -1 on error or 0 if all pipes are created.
352 */
353 static int init_thread_quit_pipe(void)
354 {
355 int ret, i;
356
357 ret = pipe(thread_quit_pipe);
358 if (ret < 0) {
359 PERROR("thread quit pipe");
360 goto error;
361 }
362
363 for (i = 0; i < 2; i++) {
364 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
365 if (ret < 0) {
366 PERROR("fcntl");
367 goto error;
368 }
369 }
370
371 error:
372 return ret;
373 }
374
375 /*
376 * Stop all threads by closing the thread quit pipe.
377 */
378 static void stop_threads(void)
379 {
380 int ret;
381
382 /* Stopping all threads */
383 DBG("Terminating all threads");
384 ret = notify_thread_pipe(thread_quit_pipe[1]);
385 if (ret < 0) {
386 ERR("write error on thread quit pipe");
387 }
388
389 /* Dispatch thread */
390 CMM_STORE_SHARED(dispatch_thread_exit, 1);
391 futex_nto1_wake(&ust_cmd_queue.futex);
392 }
393
394 /*
395 * Cleanup the daemon
396 */
397 static void cleanup(void)
398 {
399 int ret;
400 char *cmd;
401 struct ltt_session *sess, *stmp;
402
403 DBG("Cleaning up");
404
405 /* First thing first, stop all threads */
406 utils_close_pipe(thread_quit_pipe);
407
408 DBG("Removing %s directory", rundir);
409 ret = asprintf(&cmd, "rm -rf %s", rundir);
410 if (ret < 0) {
411 ERR("asprintf failed. Something is really wrong!");
412 }
413
414 /* Remove lttng run directory */
415 ret = system(cmd);
416 if (ret < 0) {
417 ERR("Unable to clean %s", rundir);
418 }
419 free(cmd);
420 free(rundir);
421
422 DBG("Cleaning up all sessions");
423
424 /* Destroy session list mutex */
425 if (session_list_ptr != NULL) {
426 pthread_mutex_destroy(&session_list_ptr->lock);
427
428 /* Cleanup ALL session */
429 cds_list_for_each_entry_safe(sess, stmp,
430 &session_list_ptr->head, list) {
431 cmd_destroy_session(sess, kernel_poll_pipe[1]);
432 }
433 }
434
435 DBG("Closing all UST sockets");
436 ust_app_clean_list();
437
438 if (is_root && !opt_no_kernel) {
439 DBG2("Closing kernel fd");
440 if (kernel_tracer_fd >= 0) {
441 ret = close(kernel_tracer_fd);
442 if (ret) {
443 PERROR("close");
444 }
445 }
446 DBG("Unloading kernel modules");
447 modprobe_remove_lttng_all();
448 }
449
450 utils_close_pipe(kernel_poll_pipe);
451 utils_close_pipe(apps_cmd_pipe);
452
453 /* <fun> */
454 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
455 "Matthew, BEET driven development works!%c[%dm",
456 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
457 /* </fun> */
458 }
459
460 /*
461 * Send data on a unix socket using the liblttsessiondcomm API.
462 *
463 * Return lttcomm error code.
464 */
465 static int send_unix_sock(int sock, void *buf, size_t len)
466 {
467 /* Check valid length */
468 if (len <= 0) {
469 return -1;
470 }
471
472 return lttcomm_send_unix_sock(sock, buf, len);
473 }
474
475 /*
476 * Free memory of a command context structure.
477 */
478 static void clean_command_ctx(struct command_ctx **cmd_ctx)
479 {
480 DBG("Clean command context structure");
481 if (*cmd_ctx) {
482 if ((*cmd_ctx)->llm) {
483 free((*cmd_ctx)->llm);
484 }
485 if ((*cmd_ctx)->lsm) {
486 free((*cmd_ctx)->lsm);
487 }
488 free(*cmd_ctx);
489 *cmd_ctx = NULL;
490 }
491 }
492
493 /*
494 * Notify UST applications using the shm mmap futex.
495 */
496 static int notify_ust_apps(int active)
497 {
498 char *wait_shm_mmap;
499
500 DBG("Notifying applications of session daemon state: %d", active);
501
502 /* See shm.c for this call implying mmap, shm and futex calls */
503 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
504 if (wait_shm_mmap == NULL) {
505 goto error;
506 }
507
508 /* Wake waiting process */
509 futex_wait_update((int32_t *) wait_shm_mmap, active);
510
511 /* Apps notified successfully */
512 return 0;
513
514 error:
515 return -1;
516 }
517
518 /*
519 * Setup the outgoing data buffer for the response (llm) by allocating the
520 * right amount of memory and copying the original information from the lsm
521 * structure.
522 *
523 * Return total size of the buffer pointed by buf.
524 */
525 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
526 {
527 int ret, buf_size;
528
529 buf_size = size;
530
531 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
532 if (cmd_ctx->llm == NULL) {
533 PERROR("zmalloc");
534 ret = -ENOMEM;
535 goto error;
536 }
537
538 /* Copy common data */
539 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
540 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
541
542 cmd_ctx->llm->data_size = size;
543 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
544
545 return buf_size;
546
547 error:
548 return ret;
549 }
550
551 /*
552 * Update the kernel poll set of all channel fd available over all tracing
553 * session. Add the wakeup pipe at the end of the set.
554 */
555 static int update_kernel_poll(struct lttng_poll_event *events)
556 {
557 int ret;
558 struct ltt_session *session;
559 struct ltt_kernel_channel *channel;
560
561 DBG("Updating kernel poll set");
562
563 session_lock_list();
564 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
565 session_lock(session);
566 if (session->kernel_session == NULL) {
567 session_unlock(session);
568 continue;
569 }
570
571 cds_list_for_each_entry(channel,
572 &session->kernel_session->channel_list.head, list) {
573 /* Add channel fd to the kernel poll set */
574 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
575 if (ret < 0) {
576 session_unlock(session);
577 goto error;
578 }
579 DBG("Channel fd %d added to kernel set", channel->fd);
580 }
581 session_unlock(session);
582 }
583 session_unlock_list();
584
585 return 0;
586
587 error:
588 session_unlock_list();
589 return -1;
590 }
591
592 /*
593 * Find the channel fd from 'fd' over all tracing session. When found, check
594 * for new channel stream and send those stream fds to the kernel consumer.
595 *
596 * Useful for CPU hotplug feature.
597 */
598 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
599 {
600 int ret = 0;
601 struct ltt_session *session;
602 struct ltt_kernel_session *ksess;
603 struct ltt_kernel_channel *channel;
604
605 DBG("Updating kernel streams for channel fd %d", fd);
606
607 session_lock_list();
608 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
609 session_lock(session);
610 if (session->kernel_session == NULL) {
611 session_unlock(session);
612 continue;
613 }
614 ksess = session->kernel_session;
615
616 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
617 if (channel->fd == fd) {
618 DBG("Channel found, updating kernel streams");
619 ret = kernel_open_channel_stream(channel);
620 if (ret < 0) {
621 goto error;
622 }
623
624 /*
625 * Have we already sent fds to the consumer? If yes, it means
626 * that tracing is started so it is safe to send our updated
627 * stream fds.
628 */
629 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
630 struct lttng_ht_iter iter;
631 struct consumer_socket *socket;
632
633
634 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
635 &iter.iter, socket, node.node) {
636 /* Code flow error */
637 assert(socket->fd >= 0);
638
639 pthread_mutex_lock(socket->lock);
640 ret = kernel_consumer_send_channel_stream(socket->fd,
641 channel, ksess);
642 pthread_mutex_unlock(socket->lock);
643 if (ret < 0) {
644 goto error;
645 }
646 }
647 }
648 goto error;
649 }
650 }
651 session_unlock(session);
652 }
653 session_unlock_list();
654 return ret;
655
656 error:
657 session_unlock(session);
658 session_unlock_list();
659 return ret;
660 }
661
662 /*
663 * For each tracing session, update newly registered apps.
664 */
665 static void update_ust_app(int app_sock)
666 {
667 struct ltt_session *sess, *stmp;
668
669 session_lock_list();
670
671 /* For all tracing session(s) */
672 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
673 session_lock(sess);
674 if (sess->ust_session) {
675 ust_app_global_update(sess->ust_session, app_sock);
676 }
677 session_unlock(sess);
678 }
679
680 session_unlock_list();
681 }
682
683 /*
684 * This thread manage event coming from the kernel.
685 *
686 * Features supported in this thread:
687 * -) CPU Hotplug
688 */
689 static void *thread_manage_kernel(void *data)
690 {
691 int ret, i, pollfd, update_poll_flag = 1, err = -1;
692 uint32_t revents, nb_fd;
693 char tmp;
694 struct lttng_poll_event events;
695
696 DBG("Thread manage kernel started");
697
698 testpoint(thread_manage_kernel);
699
700 health_code_update(&health_thread_kernel);
701
702 testpoint(thread_manage_kernel_before_loop);
703
704 ret = create_thread_poll_set(&events, 2);
705 if (ret < 0) {
706 goto error_poll_create;
707 }
708
709 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
710 if (ret < 0) {
711 goto error;
712 }
713
714 while (1) {
715 health_code_update(&health_thread_kernel);
716
717 if (update_poll_flag == 1) {
718 /*
719 * Reset number of fd in the poll set. Always 2 since there is the thread
720 * quit pipe and the kernel pipe.
721 */
722 events.nb_fd = 2;
723
724 ret = update_kernel_poll(&events);
725 if (ret < 0) {
726 goto error;
727 }
728 update_poll_flag = 0;
729 }
730
731 nb_fd = LTTNG_POLL_GETNB(&events);
732
733 DBG("Thread kernel polling on %d fds", nb_fd);
734
735 /* Zeroed the poll events */
736 lttng_poll_reset(&events);
737
738 /* Poll infinite value of time */
739 restart:
740 health_poll_update(&health_thread_kernel);
741 ret = lttng_poll_wait(&events, -1);
742 health_poll_update(&health_thread_kernel);
743 if (ret < 0) {
744 /*
745 * Restart interrupted system call.
746 */
747 if (errno == EINTR) {
748 goto restart;
749 }
750 goto error;
751 } else if (ret == 0) {
752 /* Should not happen since timeout is infinite */
753 ERR("Return value of poll is 0 with an infinite timeout.\n"
754 "This should not have happened! Continuing...");
755 continue;
756 }
757
758 for (i = 0; i < nb_fd; i++) {
759 /* Fetch once the poll data */
760 revents = LTTNG_POLL_GETEV(&events, i);
761 pollfd = LTTNG_POLL_GETFD(&events, i);
762
763 health_code_update(&health_thread_kernel);
764
765 /* Thread quit pipe has been closed. Killing thread. */
766 ret = check_thread_quit_pipe(pollfd, revents);
767 if (ret) {
768 err = 0;
769 goto exit;
770 }
771
772 /* Check for data on kernel pipe */
773 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
774 ret = read(kernel_poll_pipe[0], &tmp, 1);
775 update_poll_flag = 1;
776 continue;
777 } else {
778 /*
779 * New CPU detected by the kernel. Adding kernel stream to
780 * kernel session and updating the kernel consumer
781 */
782 if (revents & LPOLLIN) {
783 ret = update_kernel_stream(&kconsumer_data, pollfd);
784 if (ret < 0) {
785 continue;
786 }
787 break;
788 /*
789 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
790 * and unregister kernel stream at this point.
791 */
792 }
793 }
794 }
795 }
796
797 exit:
798 error:
799 lttng_poll_clean(&events);
800 error_poll_create:
801 if (err) {
802 health_error(&health_thread_kernel);
803 ERR("Health error occurred in %s", __func__);
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 if (err) {
1254 health_error(&health_thread_app_manage);
1255 ERR("Health error occurred in %s", __func__);
1256 }
1257 health_exit(&health_thread_app_manage);
1258 DBG("Application communication apps thread cleanup complete");
1259 rcu_thread_offline();
1260 rcu_unregister_thread();
1261 return NULL;
1262 }
1263
1264 /*
1265 * Dispatch request from the registration threads to the application
1266 * communication thread.
1267 */
1268 static void *thread_dispatch_ust_registration(void *data)
1269 {
1270 int ret;
1271 struct cds_wfq_node *node;
1272 struct ust_command *ust_cmd = NULL;
1273
1274 DBG("[thread] Dispatch UST command started");
1275
1276 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1277 /* Atomically prepare the queue futex */
1278 futex_nto1_prepare(&ust_cmd_queue.futex);
1279
1280 do {
1281 /* Dequeue command for registration */
1282 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1283 if (node == NULL) {
1284 DBG("Woken up but nothing in the UST command queue");
1285 /* Continue thread execution */
1286 break;
1287 }
1288
1289 ust_cmd = caa_container_of(node, struct ust_command, node);
1290
1291 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1292 " gid:%d sock:%d name:%s (version %d.%d)",
1293 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1294 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1295 ust_cmd->sock, ust_cmd->reg_msg.name,
1296 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1297 /*
1298 * Inform apps thread of the new application registration. This
1299 * call is blocking so we can be assured that the data will be read
1300 * at some point in time or wait to the end of the world :)
1301 */
1302 ret = write(apps_cmd_pipe[1], ust_cmd,
1303 sizeof(struct ust_command));
1304 if (ret < 0) {
1305 PERROR("write apps cmd pipe");
1306 if (errno == EBADF) {
1307 /*
1308 * We can't inform the application thread to process
1309 * registration. We will exit or else application
1310 * registration will not occur and tracing will never
1311 * start.
1312 */
1313 goto error;
1314 }
1315 }
1316 free(ust_cmd);
1317 } while (node != NULL);
1318
1319 /* Futex wait on queue. Blocking call on futex() */
1320 futex_nto1_wait(&ust_cmd_queue.futex);
1321 }
1322
1323 error:
1324 DBG("Dispatch thread dying");
1325 return NULL;
1326 }
1327
1328 /*
1329 * This thread manage application registration.
1330 */
1331 static void *thread_registration_apps(void *data)
1332 {
1333 int sock = -1, i, ret, pollfd, err = -1;
1334 uint32_t revents, nb_fd;
1335 struct lttng_poll_event events;
1336 /*
1337 * Get allocated in this thread, enqueued to a global queue, dequeued and
1338 * freed in the manage apps thread.
1339 */
1340 struct ust_command *ust_cmd = NULL;
1341
1342 DBG("[thread] Manage application registration started");
1343
1344 testpoint(thread_registration_apps);
1345
1346 ret = lttcomm_listen_unix_sock(apps_sock);
1347 if (ret < 0) {
1348 goto error_listen;
1349 }
1350
1351 /*
1352 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1353 * more will be added to this poll set.
1354 */
1355 ret = create_thread_poll_set(&events, 2);
1356 if (ret < 0) {
1357 goto error_create_poll;
1358 }
1359
1360 /* Add the application registration socket */
1361 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1362 if (ret < 0) {
1363 goto error_poll_add;
1364 }
1365
1366 /* Notify all applications to register */
1367 ret = notify_ust_apps(1);
1368 if (ret < 0) {
1369 ERR("Failed to notify applications or create the wait shared memory.\n"
1370 "Execution continues but there might be problem for already\n"
1371 "running applications that wishes to register.");
1372 }
1373
1374 while (1) {
1375 DBG("Accepting application registration");
1376
1377 nb_fd = LTTNG_POLL_GETNB(&events);
1378
1379 /* Inifinite blocking call, waiting for transmission */
1380 restart:
1381 health_poll_update(&health_thread_app_reg);
1382 ret = lttng_poll_wait(&events, -1);
1383 health_poll_update(&health_thread_app_reg);
1384 if (ret < 0) {
1385 /*
1386 * Restart interrupted system call.
1387 */
1388 if (errno == EINTR) {
1389 goto restart;
1390 }
1391 goto error;
1392 }
1393
1394 for (i = 0; i < nb_fd; i++) {
1395 health_code_update(&health_thread_app_reg);
1396
1397 /* Fetch once the poll data */
1398 revents = LTTNG_POLL_GETEV(&events, i);
1399 pollfd = LTTNG_POLL_GETFD(&events, i);
1400
1401 /* Thread quit pipe has been closed. Killing thread. */
1402 ret = check_thread_quit_pipe(pollfd, revents);
1403 if (ret) {
1404 err = 0;
1405 goto exit;
1406 }
1407
1408 /* Event on the registration socket */
1409 if (pollfd == apps_sock) {
1410 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1411 ERR("Register apps socket poll error");
1412 goto error;
1413 } else if (revents & LPOLLIN) {
1414 sock = lttcomm_accept_unix_sock(apps_sock);
1415 if (sock < 0) {
1416 goto error;
1417 }
1418
1419 /*
1420 * Set the CLOEXEC flag. Return code is useless because
1421 * either way, the show must go on.
1422 */
1423 (void) utils_set_fd_cloexec(sock);
1424
1425 /* Create UST registration command for enqueuing */
1426 ust_cmd = zmalloc(sizeof(struct ust_command));
1427 if (ust_cmd == NULL) {
1428 PERROR("ust command zmalloc");
1429 goto error;
1430 }
1431
1432 /*
1433 * Using message-based transmissions to ensure we don't
1434 * have to deal with partially received messages.
1435 */
1436 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1437 if (ret < 0) {
1438 ERR("Exhausted file descriptors allowed for applications.");
1439 free(ust_cmd);
1440 ret = close(sock);
1441 if (ret) {
1442 PERROR("close");
1443 }
1444 sock = -1;
1445 continue;
1446 }
1447 health_code_update(&health_thread_app_reg);
1448 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1449 sizeof(struct ust_register_msg));
1450 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1451 if (ret < 0) {
1452 PERROR("lttcomm_recv_unix_sock register apps");
1453 } else {
1454 ERR("Wrong size received on apps register");
1455 }
1456 free(ust_cmd);
1457 ret = close(sock);
1458 if (ret) {
1459 PERROR("close");
1460 }
1461 lttng_fd_put(LTTNG_FD_APPS, 1);
1462 sock = -1;
1463 continue;
1464 }
1465 health_code_update(&health_thread_app_reg);
1466
1467 ust_cmd->sock = sock;
1468 sock = -1;
1469
1470 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1471 " gid:%d sock:%d name:%s (version %d.%d)",
1472 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1473 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1474 ust_cmd->sock, ust_cmd->reg_msg.name,
1475 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1476
1477 /*
1478 * Lock free enqueue the registration request. The red pill
1479 * has been taken! This apps will be part of the *system*.
1480 */
1481 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1482
1483 /*
1484 * Wake the registration queue futex. Implicit memory
1485 * barrier with the exchange in cds_wfq_enqueue.
1486 */
1487 futex_nto1_wake(&ust_cmd_queue.futex);
1488 }
1489 }
1490 }
1491 }
1492
1493 exit:
1494 error:
1495 if (err) {
1496 health_error(&health_thread_app_reg);
1497 ERR("Health error occurred in %s", __func__);
1498 }
1499 health_exit(&health_thread_app_reg);
1500
1501 /* Notify that the registration thread is gone */
1502 notify_ust_apps(0);
1503
1504 if (apps_sock >= 0) {
1505 ret = close(apps_sock);
1506 if (ret) {
1507 PERROR("close");
1508 }
1509 }
1510 if (sock >= 0) {
1511 ret = close(sock);
1512 if (ret) {
1513 PERROR("close");
1514 }
1515 lttng_fd_put(LTTNG_FD_APPS, 1);
1516 }
1517 unlink(apps_unix_sock_path);
1518
1519 error_poll_add:
1520 lttng_poll_clean(&events);
1521 error_listen:
1522 error_create_poll:
1523 DBG("UST Registration thread cleanup complete");
1524
1525 return NULL;
1526 }
1527
1528 /*
1529 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1530 * exec or it will fails.
1531 */
1532 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1533 {
1534 int ret, clock_ret;
1535 struct timespec timeout;
1536
1537 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1538 consumer_data->consumer_thread_is_ready = 0;
1539
1540 /* Setup pthread condition */
1541 ret = pthread_condattr_init(&consumer_data->condattr);
1542 if (ret != 0) {
1543 errno = ret;
1544 PERROR("pthread_condattr_init consumer data");
1545 goto error;
1546 }
1547
1548 /*
1549 * Set the monotonic clock in order to make sure we DO NOT jump in time
1550 * between the clock_gettime() call and the timedwait call. See bug #324
1551 * for a more details and how we noticed it.
1552 */
1553 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1554 if (ret != 0) {
1555 errno = ret;
1556 PERROR("pthread_condattr_setclock consumer data");
1557 goto error;
1558 }
1559
1560 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1561 if (ret != 0) {
1562 errno = ret;
1563 PERROR("pthread_cond_init consumer data");
1564 goto error;
1565 }
1566
1567 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1568 consumer_data);
1569 if (ret != 0) {
1570 PERROR("pthread_create consumer");
1571 ret = -1;
1572 goto error;
1573 }
1574
1575 /* We are about to wait on a pthread condition */
1576 pthread_mutex_lock(&consumer_data->cond_mutex);
1577
1578 /* Get time for sem_timedwait absolute timeout */
1579 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1580 /*
1581 * Set the timeout for the condition timed wait even if the clock gettime
1582 * call fails since we might loop on that call and we want to avoid to
1583 * increment the timeout too many times.
1584 */
1585 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1586
1587 /*
1588 * The following loop COULD be skipped in some conditions so this is why we
1589 * set ret to 0 in order to make sure at least one round of the loop is
1590 * done.
1591 */
1592 ret = 0;
1593
1594 /*
1595 * Loop until the condition is reached or when a timeout is reached. Note
1596 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1597 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1598 * possible. This loop does not take any chances and works with both of
1599 * them.
1600 */
1601 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1602 if (clock_ret < 0) {
1603 PERROR("clock_gettime spawn consumer");
1604 /* Infinite wait for the consumerd thread to be ready */
1605 ret = pthread_cond_wait(&consumer_data->cond,
1606 &consumer_data->cond_mutex);
1607 } else {
1608 ret = pthread_cond_timedwait(&consumer_data->cond,
1609 &consumer_data->cond_mutex, &timeout);
1610 }
1611 }
1612
1613 /* Release the pthread condition */
1614 pthread_mutex_unlock(&consumer_data->cond_mutex);
1615
1616 if (ret != 0) {
1617 errno = ret;
1618 if (ret == ETIMEDOUT) {
1619 /*
1620 * Call has timed out so we kill the kconsumerd_thread and return
1621 * an error.
1622 */
1623 ERR("Condition timed out. The consumer thread was never ready."
1624 " Killing it");
1625 ret = pthread_cancel(consumer_data->thread);
1626 if (ret < 0) {
1627 PERROR("pthread_cancel consumer thread");
1628 }
1629 } else {
1630 PERROR("pthread_cond_wait failed consumer thread");
1631 }
1632 goto error;
1633 }
1634
1635 pthread_mutex_lock(&consumer_data->pid_mutex);
1636 if (consumer_data->pid == 0) {
1637 ERR("Consumerd did not start");
1638 pthread_mutex_unlock(&consumer_data->pid_mutex);
1639 goto error;
1640 }
1641 pthread_mutex_unlock(&consumer_data->pid_mutex);
1642
1643 return 0;
1644
1645 error:
1646 return ret;
1647 }
1648
1649 /*
1650 * Join consumer thread
1651 */
1652 static int join_consumer_thread(struct consumer_data *consumer_data)
1653 {
1654 void *status;
1655 int ret;
1656
1657 /* Consumer pid must be a real one. */
1658 if (consumer_data->pid > 0) {
1659 ret = kill(consumer_data->pid, SIGTERM);
1660 if (ret) {
1661 ERR("Error killing consumer daemon");
1662 return ret;
1663 }
1664 return pthread_join(consumer_data->thread, &status);
1665 } else {
1666 return 0;
1667 }
1668 }
1669
1670 /*
1671 * Fork and exec a consumer daemon (consumerd).
1672 *
1673 * Return pid if successful else -1.
1674 */
1675 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1676 {
1677 int ret;
1678 pid_t pid;
1679 const char *consumer_to_use;
1680 const char *verbosity;
1681 struct stat st;
1682
1683 DBG("Spawning consumerd");
1684
1685 pid = fork();
1686 if (pid == 0) {
1687 /*
1688 * Exec consumerd.
1689 */
1690 if (opt_verbose_consumer) {
1691 verbosity = "--verbose";
1692 } else {
1693 verbosity = "--quiet";
1694 }
1695 switch (consumer_data->type) {
1696 case LTTNG_CONSUMER_KERNEL:
1697 /*
1698 * Find out which consumerd to execute. We will first try the
1699 * 64-bit path, then the sessiond's installation directory, and
1700 * fallback on the 32-bit one,
1701 */
1702 DBG3("Looking for a kernel consumer at these locations:");
1703 DBG3(" 1) %s", consumerd64_bin);
1704 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1705 DBG3(" 3) %s", consumerd32_bin);
1706 if (stat(consumerd64_bin, &st) == 0) {
1707 DBG3("Found location #1");
1708 consumer_to_use = consumerd64_bin;
1709 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1710 DBG3("Found location #2");
1711 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1712 } else if (stat(consumerd32_bin, &st) == 0) {
1713 DBG3("Found location #3");
1714 consumer_to_use = consumerd32_bin;
1715 } else {
1716 DBG("Could not find any valid consumerd executable");
1717 break;
1718 }
1719 DBG("Using kernel consumer at: %s", consumer_to_use);
1720 execl(consumer_to_use,
1721 "lttng-consumerd", verbosity, "-k",
1722 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1723 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1724 NULL);
1725 break;
1726 case LTTNG_CONSUMER64_UST:
1727 {
1728 char *tmpnew = NULL;
1729
1730 if (consumerd64_libdir[0] != '\0') {
1731 char *tmp;
1732 size_t tmplen;
1733
1734 tmp = getenv("LD_LIBRARY_PATH");
1735 if (!tmp) {
1736 tmp = "";
1737 }
1738 tmplen = strlen("LD_LIBRARY_PATH=")
1739 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1740 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1741 if (!tmpnew) {
1742 ret = -ENOMEM;
1743 goto error;
1744 }
1745 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1746 strcat(tmpnew, consumerd64_libdir);
1747 if (tmp[0] != '\0') {
1748 strcat(tmpnew, ":");
1749 strcat(tmpnew, tmp);
1750 }
1751 ret = putenv(tmpnew);
1752 if (ret) {
1753 ret = -errno;
1754 goto error;
1755 }
1756 }
1757 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1758 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1759 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1760 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1761 NULL);
1762 if (consumerd64_libdir[0] != '\0') {
1763 free(tmpnew);
1764 }
1765 if (ret) {
1766 goto error;
1767 }
1768 break;
1769 }
1770 case LTTNG_CONSUMER32_UST:
1771 {
1772 char *tmpnew = NULL;
1773
1774 if (consumerd32_libdir[0] != '\0') {
1775 char *tmp;
1776 size_t tmplen;
1777
1778 tmp = getenv("LD_LIBRARY_PATH");
1779 if (!tmp) {
1780 tmp = "";
1781 }
1782 tmplen = strlen("LD_LIBRARY_PATH=")
1783 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1784 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1785 if (!tmpnew) {
1786 ret = -ENOMEM;
1787 goto error;
1788 }
1789 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1790 strcat(tmpnew, consumerd32_libdir);
1791 if (tmp[0] != '\0') {
1792 strcat(tmpnew, ":");
1793 strcat(tmpnew, tmp);
1794 }
1795 ret = putenv(tmpnew);
1796 if (ret) {
1797 ret = -errno;
1798 goto error;
1799 }
1800 }
1801 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1802 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1803 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1804 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1805 NULL);
1806 if (consumerd32_libdir[0] != '\0') {
1807 free(tmpnew);
1808 }
1809 if (ret) {
1810 goto error;
1811 }
1812 break;
1813 }
1814 default:
1815 PERROR("unknown consumer type");
1816 exit(EXIT_FAILURE);
1817 }
1818 if (errno != 0) {
1819 PERROR("kernel start consumer exec");
1820 }
1821 exit(EXIT_FAILURE);
1822 } else if (pid > 0) {
1823 ret = pid;
1824 } else {
1825 PERROR("start consumer fork");
1826 ret = -errno;
1827 }
1828 error:
1829 return ret;
1830 }
1831
1832 /*
1833 * Spawn the consumerd daemon and session daemon thread.
1834 */
1835 static int start_consumerd(struct consumer_data *consumer_data)
1836 {
1837 int ret, err;
1838
1839 /*
1840 * Set the listen() state on the socket since there is a possible race
1841 * between the exec() of the consumer daemon and this call if place in the
1842 * consumer thread. See bug #366 for more details.
1843 */
1844 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1845 if (ret < 0) {
1846 goto error;
1847 }
1848
1849 pthread_mutex_lock(&consumer_data->pid_mutex);
1850 if (consumer_data->pid != 0) {
1851 pthread_mutex_unlock(&consumer_data->pid_mutex);
1852 goto end;
1853 }
1854
1855 ret = spawn_consumerd(consumer_data);
1856 if (ret < 0) {
1857 ERR("Spawning consumerd failed");
1858 pthread_mutex_unlock(&consumer_data->pid_mutex);
1859 goto error;
1860 }
1861
1862 /* Setting up the consumer_data pid */
1863 consumer_data->pid = ret;
1864 DBG2("Consumer pid %d", consumer_data->pid);
1865 pthread_mutex_unlock(&consumer_data->pid_mutex);
1866
1867 DBG2("Spawning consumer control thread");
1868 ret = spawn_consumer_thread(consumer_data);
1869 if (ret < 0) {
1870 ERR("Fatal error spawning consumer control thread");
1871 goto error;
1872 }
1873
1874 end:
1875 return 0;
1876
1877 error:
1878 /* Cleanup already created socket on error. */
1879 if (consumer_data->err_sock >= 0) {
1880 err = close(consumer_data->err_sock);
1881 if (err < 0) {
1882 PERROR("close consumer data error socket");
1883 }
1884 }
1885 return ret;
1886 }
1887
1888 /*
1889 * Compute health status of each consumer. If one of them is zero (bad
1890 * state), we return 0.
1891 */
1892 static int check_consumer_health(void)
1893 {
1894 int ret;
1895
1896 ret = health_check_state(&kconsumer_data.health) &&
1897 health_check_state(&ustconsumer32_data.health) &&
1898 health_check_state(&ustconsumer64_data.health);
1899
1900 DBG3("Health consumer check %d", ret);
1901
1902 return ret;
1903 }
1904
1905 /*
1906 * Setup necessary data for kernel tracer action.
1907 */
1908 static int init_kernel_tracer(void)
1909 {
1910 int ret;
1911
1912 /* Modprobe lttng kernel modules */
1913 ret = modprobe_lttng_control();
1914 if (ret < 0) {
1915 goto error;
1916 }
1917
1918 /* Open debugfs lttng */
1919 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1920 if (kernel_tracer_fd < 0) {
1921 DBG("Failed to open %s", module_proc_lttng);
1922 ret = -1;
1923 goto error_open;
1924 }
1925
1926 /* Validate kernel version */
1927 ret = kernel_validate_version(kernel_tracer_fd);
1928 if (ret < 0) {
1929 goto error_version;
1930 }
1931
1932 ret = modprobe_lttng_data();
1933 if (ret < 0) {
1934 goto error_modules;
1935 }
1936
1937 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1938 return 0;
1939
1940 error_version:
1941 modprobe_remove_lttng_control();
1942 ret = close(kernel_tracer_fd);
1943 if (ret) {
1944 PERROR("close");
1945 }
1946 kernel_tracer_fd = -1;
1947 return LTTNG_ERR_KERN_VERSION;
1948
1949 error_modules:
1950 ret = close(kernel_tracer_fd);
1951 if (ret) {
1952 PERROR("close");
1953 }
1954
1955 error_open:
1956 modprobe_remove_lttng_control();
1957
1958 error:
1959 WARN("No kernel tracer available");
1960 kernel_tracer_fd = -1;
1961 if (!is_root) {
1962 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1963 } else {
1964 return LTTNG_ERR_KERN_NA;
1965 }
1966 }
1967
1968
1969 /*
1970 * Copy consumer output from the tracing session to the domain session. The
1971 * function also applies the right modification on a per domain basis for the
1972 * trace files destination directory.
1973 */
1974 static int copy_session_consumer(int domain, struct ltt_session *session)
1975 {
1976 int ret;
1977 const char *dir_name;
1978 struct consumer_output *consumer;
1979
1980 assert(session);
1981 assert(session->consumer);
1982
1983 switch (domain) {
1984 case LTTNG_DOMAIN_KERNEL:
1985 DBG3("Copying tracing session consumer output in kernel session");
1986 /*
1987 * XXX: We should audit the session creation and what this function
1988 * does "extra" in order to avoid a destroy since this function is used
1989 * in the domain session creation (kernel and ust) only. Same for UST
1990 * domain.
1991 */
1992 if (session->kernel_session->consumer) {
1993 consumer_destroy_output(session->kernel_session->consumer);
1994 }
1995 session->kernel_session->consumer =
1996 consumer_copy_output(session->consumer);
1997 /* Ease our life a bit for the next part */
1998 consumer = session->kernel_session->consumer;
1999 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2000 break;
2001 case LTTNG_DOMAIN_UST:
2002 DBG3("Copying tracing session consumer output in UST session");
2003 if (session->ust_session->consumer) {
2004 consumer_destroy_output(session->ust_session->consumer);
2005 }
2006 session->ust_session->consumer =
2007 consumer_copy_output(session->consumer);
2008 /* Ease our life a bit for the next part */
2009 consumer = session->ust_session->consumer;
2010 dir_name = DEFAULT_UST_TRACE_DIR;
2011 break;
2012 default:
2013 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2014 goto error;
2015 }
2016
2017 /* Append correct directory to subdir */
2018 strncat(consumer->subdir, dir_name,
2019 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2020 DBG3("Copy session consumer subdir %s", consumer->subdir);
2021
2022 ret = LTTNG_OK;
2023
2024 error:
2025 return ret;
2026 }
2027
2028 /*
2029 * Create an UST session and add it to the session ust list.
2030 */
2031 static int create_ust_session(struct ltt_session *session,
2032 struct lttng_domain *domain)
2033 {
2034 int ret;
2035 struct ltt_ust_session *lus = NULL;
2036
2037 assert(session);
2038 assert(domain);
2039 assert(session->consumer);
2040
2041 switch (domain->type) {
2042 case LTTNG_DOMAIN_UST:
2043 break;
2044 default:
2045 ERR("Unknown UST domain on create session %d", domain->type);
2046 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2047 goto error;
2048 }
2049
2050 DBG("Creating UST session");
2051
2052 lus = trace_ust_create_session(session->path, session->id, domain);
2053 if (lus == NULL) {
2054 ret = LTTNG_ERR_UST_SESS_FAIL;
2055 goto error;
2056 }
2057
2058 lus->uid = session->uid;
2059 lus->gid = session->gid;
2060 session->ust_session = lus;
2061
2062 /* Copy session output to the newly created UST session */
2063 ret = copy_session_consumer(domain->type, session);
2064 if (ret != LTTNG_OK) {
2065 goto error;
2066 }
2067
2068 return LTTNG_OK;
2069
2070 error:
2071 free(lus);
2072 session->ust_session = NULL;
2073 return ret;
2074 }
2075
2076 /*
2077 * Create a kernel tracer session then create the default channel.
2078 */
2079 static int create_kernel_session(struct ltt_session *session)
2080 {
2081 int ret;
2082
2083 DBG("Creating kernel session");
2084
2085 ret = kernel_create_session(session, kernel_tracer_fd);
2086 if (ret < 0) {
2087 ret = LTTNG_ERR_KERN_SESS_FAIL;
2088 goto error;
2089 }
2090
2091 /* Code flow safety */
2092 assert(session->kernel_session);
2093
2094 /* Copy session output to the newly created Kernel session */
2095 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2096 if (ret != LTTNG_OK) {
2097 goto error;
2098 }
2099
2100 /* Create directory(ies) on local filesystem. */
2101 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2102 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2103 ret = run_as_mkdir_recursive(
2104 session->kernel_session->consumer->dst.trace_path,
2105 S_IRWXU | S_IRWXG, session->uid, session->gid);
2106 if (ret < 0) {
2107 if (ret != -EEXIST) {
2108 ERR("Trace directory creation error");
2109 goto error;
2110 }
2111 }
2112 }
2113
2114 session->kernel_session->uid = session->uid;
2115 session->kernel_session->gid = session->gid;
2116
2117 return LTTNG_OK;
2118
2119 error:
2120 trace_kernel_destroy_session(session->kernel_session);
2121 session->kernel_session = NULL;
2122 return ret;
2123 }
2124
2125 /*
2126 * Count number of session permitted by uid/gid.
2127 */
2128 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2129 {
2130 unsigned int i = 0;
2131 struct ltt_session *session;
2132
2133 DBG("Counting number of available session for UID %d GID %d",
2134 uid, gid);
2135 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2136 /*
2137 * Only list the sessions the user can control.
2138 */
2139 if (!session_access_ok(session, uid, gid)) {
2140 continue;
2141 }
2142 i++;
2143 }
2144 return i;
2145 }
2146
2147 /*
2148 * Process the command requested by the lttng client within the command
2149 * context structure. This function make sure that the return structure (llm)
2150 * is set and ready for transmission before returning.
2151 *
2152 * Return any error encountered or 0 for success.
2153 *
2154 * "sock" is only used for special-case var. len data.
2155 */
2156 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2157 int *sock_error)
2158 {
2159 int ret = LTTNG_OK;
2160 int need_tracing_session = 1;
2161 int need_domain;
2162
2163 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2164
2165 *sock_error = 0;
2166
2167 switch (cmd_ctx->lsm->cmd_type) {
2168 case LTTNG_CREATE_SESSION:
2169 case LTTNG_DESTROY_SESSION:
2170 case LTTNG_LIST_SESSIONS:
2171 case LTTNG_LIST_DOMAINS:
2172 case LTTNG_START_TRACE:
2173 case LTTNG_STOP_TRACE:
2174 case LTTNG_DATA_PENDING:
2175 need_domain = 0;
2176 break;
2177 default:
2178 need_domain = 1;
2179 }
2180
2181 if (opt_no_kernel && need_domain
2182 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2183 if (!is_root) {
2184 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2185 } else {
2186 ret = LTTNG_ERR_KERN_NA;
2187 }
2188 goto error;
2189 }
2190
2191 /* Deny register consumer if we already have a spawned consumer. */
2192 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2193 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2194 if (kconsumer_data.pid > 0) {
2195 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2196 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2197 goto error;
2198 }
2199 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2200 }
2201
2202 /*
2203 * Check for command that don't needs to allocate a returned payload. We do
2204 * this here so we don't have to make the call for no payload at each
2205 * command.
2206 */
2207 switch(cmd_ctx->lsm->cmd_type) {
2208 case LTTNG_LIST_SESSIONS:
2209 case LTTNG_LIST_TRACEPOINTS:
2210 case LTTNG_LIST_TRACEPOINT_FIELDS:
2211 case LTTNG_LIST_DOMAINS:
2212 case LTTNG_LIST_CHANNELS:
2213 case LTTNG_LIST_EVENTS:
2214 break;
2215 default:
2216 /* Setup lttng message with no payload */
2217 ret = setup_lttng_msg(cmd_ctx, 0);
2218 if (ret < 0) {
2219 /* This label does not try to unlock the session */
2220 goto init_setup_error;
2221 }
2222 }
2223
2224 /* Commands that DO NOT need a session. */
2225 switch (cmd_ctx->lsm->cmd_type) {
2226 case LTTNG_CREATE_SESSION:
2227 case LTTNG_CALIBRATE:
2228 case LTTNG_LIST_SESSIONS:
2229 case LTTNG_LIST_TRACEPOINTS:
2230 case LTTNG_LIST_TRACEPOINT_FIELDS:
2231 need_tracing_session = 0;
2232 break;
2233 default:
2234 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2235 /*
2236 * We keep the session list lock across _all_ commands
2237 * for now, because the per-session lock does not
2238 * handle teardown properly.
2239 */
2240 session_lock_list();
2241 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2242 if (cmd_ctx->session == NULL) {
2243 if (cmd_ctx->lsm->session.name != NULL) {
2244 ret = LTTNG_ERR_SESS_NOT_FOUND;
2245 } else {
2246 /* If no session name specified */
2247 ret = LTTNG_ERR_SELECT_SESS;
2248 }
2249 goto error;
2250 } else {
2251 /* Acquire lock for the session */
2252 session_lock(cmd_ctx->session);
2253 }
2254 break;
2255 }
2256
2257 if (!need_domain) {
2258 goto skip_domain;
2259 }
2260
2261 /*
2262 * Check domain type for specific "pre-action".
2263 */
2264 switch (cmd_ctx->lsm->domain.type) {
2265 case LTTNG_DOMAIN_KERNEL:
2266 if (!is_root) {
2267 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2268 goto error;
2269 }
2270
2271 /* Kernel tracer check */
2272 if (kernel_tracer_fd == -1) {
2273 /* Basically, load kernel tracer modules */
2274 ret = init_kernel_tracer();
2275 if (ret != 0) {
2276 goto error;
2277 }
2278 }
2279
2280 /* Consumer is in an ERROR state. Report back to client */
2281 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2282 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2283 goto error;
2284 }
2285
2286 /* Need a session for kernel command */
2287 if (need_tracing_session) {
2288 if (cmd_ctx->session->kernel_session == NULL) {
2289 ret = create_kernel_session(cmd_ctx->session);
2290 if (ret < 0) {
2291 ret = LTTNG_ERR_KERN_SESS_FAIL;
2292 goto error;
2293 }
2294 }
2295
2296 /* Start the kernel consumer daemon */
2297 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2298 if (kconsumer_data.pid == 0 &&
2299 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2300 cmd_ctx->session->start_consumer) {
2301 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2302 ret = start_consumerd(&kconsumer_data);
2303 if (ret < 0) {
2304 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2305 goto error;
2306 }
2307 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2308 } else {
2309 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2310 }
2311
2312 /*
2313 * The consumer was just spawned so we need to add the socket to
2314 * the consumer output of the session if exist.
2315 */
2316 ret = consumer_create_socket(&kconsumer_data,
2317 cmd_ctx->session->kernel_session->consumer);
2318 if (ret < 0) {
2319 goto error;
2320 }
2321 }
2322
2323 break;
2324 case LTTNG_DOMAIN_UST:
2325 {
2326 /* Consumer is in an ERROR state. Report back to client */
2327 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2328 ret = LTTNG_ERR_NO_USTCONSUMERD;
2329 goto error;
2330 }
2331
2332 if (need_tracing_session) {
2333 /* Create UST session if none exist. */
2334 if (cmd_ctx->session->ust_session == NULL) {
2335 ret = create_ust_session(cmd_ctx->session,
2336 &cmd_ctx->lsm->domain);
2337 if (ret != LTTNG_OK) {
2338 goto error;
2339 }
2340 }
2341
2342 /* Start the UST consumer daemons */
2343 /* 64-bit */
2344 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2345 if (consumerd64_bin[0] != '\0' &&
2346 ustconsumer64_data.pid == 0 &&
2347 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2348 cmd_ctx->session->start_consumer) {
2349 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2350 ret = start_consumerd(&ustconsumer64_data);
2351 if (ret < 0) {
2352 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2353 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2354 goto error;
2355 }
2356
2357 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2358 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2359 } else {
2360 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2361 }
2362
2363 /*
2364 * Setup socket for consumer 64 bit. No need for atomic access
2365 * since it was set above and can ONLY be set in this thread.
2366 */
2367 ret = consumer_create_socket(&ustconsumer64_data,
2368 cmd_ctx->session->ust_session->consumer);
2369 if (ret < 0) {
2370 goto error;
2371 }
2372
2373 /* 32-bit */
2374 if (consumerd32_bin[0] != '\0' &&
2375 ustconsumer32_data.pid == 0 &&
2376 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2377 cmd_ctx->session->start_consumer) {
2378 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2379 ret = start_consumerd(&ustconsumer32_data);
2380 if (ret < 0) {
2381 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2382 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2383 goto error;
2384 }
2385
2386 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2387 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2388 } else {
2389 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2390 }
2391
2392 /*
2393 * Setup socket for consumer 64 bit. No need for atomic access
2394 * since it was set above and can ONLY be set in this thread.
2395 */
2396 ret = consumer_create_socket(&ustconsumer32_data,
2397 cmd_ctx->session->ust_session->consumer);
2398 if (ret < 0) {
2399 goto error;
2400 }
2401 }
2402 break;
2403 }
2404 default:
2405 break;
2406 }
2407 skip_domain:
2408
2409 /* Validate consumer daemon state when start/stop trace command */
2410 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2411 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2412 switch (cmd_ctx->lsm->domain.type) {
2413 case LTTNG_DOMAIN_UST:
2414 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2415 ret = LTTNG_ERR_NO_USTCONSUMERD;
2416 goto error;
2417 }
2418 break;
2419 case LTTNG_DOMAIN_KERNEL:
2420 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2421 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2422 goto error;
2423 }
2424 break;
2425 }
2426 }
2427
2428 /*
2429 * Check that the UID or GID match that of the tracing session.
2430 * The root user can interact with all sessions.
2431 */
2432 if (need_tracing_session) {
2433 if (!session_access_ok(cmd_ctx->session,
2434 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2435 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2436 ret = LTTNG_ERR_EPERM;
2437 goto error;
2438 }
2439 }
2440
2441 /* Process by command type */
2442 switch (cmd_ctx->lsm->cmd_type) {
2443 case LTTNG_ADD_CONTEXT:
2444 {
2445 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2446 cmd_ctx->lsm->u.context.channel_name,
2447 cmd_ctx->lsm->u.context.event_name,
2448 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2449 break;
2450 }
2451 case LTTNG_DISABLE_CHANNEL:
2452 {
2453 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2454 cmd_ctx->lsm->u.disable.channel_name);
2455 break;
2456 }
2457 case LTTNG_DISABLE_EVENT:
2458 {
2459 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2460 cmd_ctx->lsm->u.disable.channel_name,
2461 cmd_ctx->lsm->u.disable.name);
2462 break;
2463 }
2464 case LTTNG_DISABLE_ALL_EVENT:
2465 {
2466 DBG("Disabling all events");
2467
2468 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2469 cmd_ctx->lsm->u.disable.channel_name);
2470 break;
2471 }
2472 case LTTNG_DISABLE_CONSUMER:
2473 {
2474 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2475 break;
2476 }
2477 case LTTNG_ENABLE_CHANNEL:
2478 {
2479 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2480 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2481 break;
2482 }
2483 case LTTNG_ENABLE_CONSUMER:
2484 {
2485 /*
2486 * XXX: 0 means that this URI should be applied on the session. Should
2487 * be a DOMAIN enuam.
2488 */
2489 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2490 if (ret != LTTNG_OK) {
2491 goto error;
2492 }
2493
2494 if (cmd_ctx->lsm->domain.type == 0) {
2495 /* Add the URI for the UST session if a consumer is present. */
2496 if (cmd_ctx->session->ust_session &&
2497 cmd_ctx->session->ust_session->consumer) {
2498 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2499 } else if (cmd_ctx->session->kernel_session &&
2500 cmd_ctx->session->kernel_session->consumer) {
2501 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2502 cmd_ctx->session);
2503 }
2504 }
2505 break;
2506 }
2507 case LTTNG_ENABLE_EVENT:
2508 {
2509 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2510 cmd_ctx->lsm->u.enable.channel_name,
2511 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
2512 break;
2513 }
2514 case LTTNG_ENABLE_ALL_EVENT:
2515 {
2516 DBG("Enabling all events");
2517
2518 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2519 cmd_ctx->lsm->u.enable.channel_name,
2520 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
2521 break;
2522 }
2523 case LTTNG_LIST_TRACEPOINTS:
2524 {
2525 struct lttng_event *events;
2526 ssize_t nb_events;
2527
2528 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2529 if (nb_events < 0) {
2530 /* Return value is a negative lttng_error_code. */
2531 ret = -nb_events;
2532 goto error;
2533 }
2534
2535 /*
2536 * Setup lttng message with payload size set to the event list size in
2537 * bytes and then copy list into the llm payload.
2538 */
2539 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2540 if (ret < 0) {
2541 free(events);
2542 goto setup_error;
2543 }
2544
2545 /* Copy event list into message payload */
2546 memcpy(cmd_ctx->llm->payload, events,
2547 sizeof(struct lttng_event) * nb_events);
2548
2549 free(events);
2550
2551 ret = LTTNG_OK;
2552 break;
2553 }
2554 case LTTNG_LIST_TRACEPOINT_FIELDS:
2555 {
2556 struct lttng_event_field *fields;
2557 ssize_t nb_fields;
2558
2559 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2560 &fields);
2561 if (nb_fields < 0) {
2562 /* Return value is a negative lttng_error_code. */
2563 ret = -nb_fields;
2564 goto error;
2565 }
2566
2567 /*
2568 * Setup lttng message with payload size set to the event list size in
2569 * bytes and then copy list into the llm payload.
2570 */
2571 ret = setup_lttng_msg(cmd_ctx,
2572 sizeof(struct lttng_event_field) * nb_fields);
2573 if (ret < 0) {
2574 free(fields);
2575 goto setup_error;
2576 }
2577
2578 /* Copy event list into message payload */
2579 memcpy(cmd_ctx->llm->payload, fields,
2580 sizeof(struct lttng_event_field) * nb_fields);
2581
2582 free(fields);
2583
2584 ret = LTTNG_OK;
2585 break;
2586 }
2587 case LTTNG_SET_CONSUMER_URI:
2588 {
2589 size_t nb_uri, len;
2590 struct lttng_uri *uris;
2591
2592 nb_uri = cmd_ctx->lsm->u.uri.size;
2593 len = nb_uri * sizeof(struct lttng_uri);
2594
2595 if (nb_uri == 0) {
2596 ret = LTTNG_ERR_INVALID;
2597 goto error;
2598 }
2599
2600 uris = zmalloc(len);
2601 if (uris == NULL) {
2602 ret = LTTNG_ERR_FATAL;
2603 goto error;
2604 }
2605
2606 /* Receive variable len data */
2607 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2608 ret = lttcomm_recv_unix_sock(sock, uris, len);
2609 if (ret <= 0) {
2610 DBG("No URIs received from client... continuing");
2611 *sock_error = 1;
2612 ret = LTTNG_ERR_SESSION_FAIL;
2613 free(uris);
2614 goto error;
2615 }
2616
2617 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2618 nb_uri, uris);
2619 if (ret != LTTNG_OK) {
2620 free(uris);
2621 goto error;
2622 }
2623
2624 /*
2625 * XXX: 0 means that this URI should be applied on the session. Should
2626 * be a DOMAIN enuam.
2627 */
2628 if (cmd_ctx->lsm->domain.type == 0) {
2629 /* Add the URI for the UST session if a consumer is present. */
2630 if (cmd_ctx->session->ust_session &&
2631 cmd_ctx->session->ust_session->consumer) {
2632 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2633 nb_uri, uris);
2634 } else if (cmd_ctx->session->kernel_session &&
2635 cmd_ctx->session->kernel_session->consumer) {
2636 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2637 cmd_ctx->session, nb_uri, uris);
2638 }
2639 }
2640
2641 free(uris);
2642
2643 break;
2644 }
2645 case LTTNG_START_TRACE:
2646 {
2647 ret = cmd_start_trace(cmd_ctx->session);
2648 break;
2649 }
2650 case LTTNG_STOP_TRACE:
2651 {
2652 ret = cmd_stop_trace(cmd_ctx->session);
2653 break;
2654 }
2655 case LTTNG_CREATE_SESSION:
2656 {
2657 size_t nb_uri, len;
2658 struct lttng_uri *uris = NULL;
2659
2660 nb_uri = cmd_ctx->lsm->u.uri.size;
2661 len = nb_uri * sizeof(struct lttng_uri);
2662
2663 if (nb_uri > 0) {
2664 uris = zmalloc(len);
2665 if (uris == NULL) {
2666 ret = LTTNG_ERR_FATAL;
2667 goto error;
2668 }
2669
2670 /* Receive variable len data */
2671 DBG("Waiting for %zu URIs from client ...", nb_uri);
2672 ret = lttcomm_recv_unix_sock(sock, uris, len);
2673 if (ret <= 0) {
2674 DBG("No URIs received from client... continuing");
2675 *sock_error = 1;
2676 ret = LTTNG_ERR_SESSION_FAIL;
2677 free(uris);
2678 goto error;
2679 }
2680
2681 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2682 DBG("Creating session with ONE network URI is a bad call");
2683 ret = LTTNG_ERR_SESSION_FAIL;
2684 free(uris);
2685 goto error;
2686 }
2687 }
2688
2689 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2690 &cmd_ctx->creds);
2691
2692 free(uris);
2693
2694 break;
2695 }
2696 case LTTNG_DESTROY_SESSION:
2697 {
2698 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2699
2700 /* Set session to NULL so we do not unlock it after free. */
2701 cmd_ctx->session = NULL;
2702 break;
2703 }
2704 case LTTNG_LIST_DOMAINS:
2705 {
2706 ssize_t nb_dom;
2707 struct lttng_domain *domains;
2708
2709 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2710 if (nb_dom < 0) {
2711 /* Return value is a negative lttng_error_code. */
2712 ret = -nb_dom;
2713 goto error;
2714 }
2715
2716 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2717 if (ret < 0) {
2718 goto setup_error;
2719 }
2720
2721 /* Copy event list into message payload */
2722 memcpy(cmd_ctx->llm->payload, domains,
2723 nb_dom * sizeof(struct lttng_domain));
2724
2725 free(domains);
2726
2727 ret = LTTNG_OK;
2728 break;
2729 }
2730 case LTTNG_LIST_CHANNELS:
2731 {
2732 int nb_chan;
2733 struct lttng_channel *channels;
2734
2735 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2736 cmd_ctx->session, &channels);
2737 if (nb_chan < 0) {
2738 /* Return value is a negative lttng_error_code. */
2739 ret = -nb_chan;
2740 goto error;
2741 }
2742
2743 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2744 if (ret < 0) {
2745 goto setup_error;
2746 }
2747
2748 /* Copy event list into message payload */
2749 memcpy(cmd_ctx->llm->payload, channels,
2750 nb_chan * sizeof(struct lttng_channel));
2751
2752 free(channels);
2753
2754 ret = LTTNG_OK;
2755 break;
2756 }
2757 case LTTNG_LIST_EVENTS:
2758 {
2759 ssize_t nb_event;
2760 struct lttng_event *events = NULL;
2761
2762 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2763 cmd_ctx->lsm->u.list.channel_name, &events);
2764 if (nb_event < 0) {
2765 /* Return value is a negative lttng_error_code. */
2766 ret = -nb_event;
2767 goto error;
2768 }
2769
2770 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2771 if (ret < 0) {
2772 goto setup_error;
2773 }
2774
2775 /* Copy event list into message payload */
2776 memcpy(cmd_ctx->llm->payload, events,
2777 nb_event * sizeof(struct lttng_event));
2778
2779 free(events);
2780
2781 ret = LTTNG_OK;
2782 break;
2783 }
2784 case LTTNG_LIST_SESSIONS:
2785 {
2786 unsigned int nr_sessions;
2787
2788 session_lock_list();
2789 nr_sessions = lttng_sessions_count(
2790 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2791 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2792
2793 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2794 if (ret < 0) {
2795 session_unlock_list();
2796 goto setup_error;
2797 }
2798
2799 /* Filled the session array */
2800 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2801 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2802 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2803
2804 session_unlock_list();
2805
2806 ret = LTTNG_OK;
2807 break;
2808 }
2809 case LTTNG_CALIBRATE:
2810 {
2811 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2812 &cmd_ctx->lsm->u.calibrate);
2813 break;
2814 }
2815 case LTTNG_REGISTER_CONSUMER:
2816 {
2817 struct consumer_data *cdata;
2818
2819 switch (cmd_ctx->lsm->domain.type) {
2820 case LTTNG_DOMAIN_KERNEL:
2821 cdata = &kconsumer_data;
2822 break;
2823 default:
2824 ret = LTTNG_ERR_UND;
2825 goto error;
2826 }
2827
2828 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2829 cmd_ctx->lsm->u.reg.path, cdata);
2830 break;
2831 }
2832 case LTTNG_SET_FILTER:
2833 {
2834 struct lttng_filter_bytecode *bytecode;
2835
2836 if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2837 ret = LTTNG_ERR_FILTER_INVAL;
2838 goto error;
2839 }
2840 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2841 if (!bytecode) {
2842 ret = LTTNG_ERR_FILTER_NOMEM;
2843 goto error;
2844 }
2845 /* Receive var. len. data */
2846 DBG("Receiving var len data from client ...");
2847 ret = lttcomm_recv_unix_sock(sock, bytecode,
2848 cmd_ctx->lsm->u.filter.bytecode_len);
2849 if (ret <= 0) {
2850 DBG("Nothing recv() from client var len data... continuing");
2851 *sock_error = 1;
2852 ret = LTTNG_ERR_FILTER_INVAL;
2853 goto error;
2854 }
2855
2856 if (bytecode->len + sizeof(*bytecode)
2857 != cmd_ctx->lsm->u.filter.bytecode_len) {
2858 free(bytecode);
2859 ret = LTTNG_ERR_FILTER_INVAL;
2860 goto error;
2861 }
2862
2863 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2864 cmd_ctx->lsm->u.filter.channel_name,
2865 cmd_ctx->lsm->u.filter.event_name,
2866 bytecode);
2867 break;
2868 }
2869 case LTTNG_DATA_PENDING:
2870 {
2871 ret = cmd_data_pending(cmd_ctx->session);
2872 break;
2873 }
2874 default:
2875 ret = LTTNG_ERR_UND;
2876 break;
2877 }
2878
2879 error:
2880 if (cmd_ctx->llm == NULL) {
2881 DBG("Missing llm structure. Allocating one.");
2882 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2883 goto setup_error;
2884 }
2885 }
2886 /* Set return code */
2887 cmd_ctx->llm->ret_code = ret;
2888 setup_error:
2889 if (cmd_ctx->session) {
2890 session_unlock(cmd_ctx->session);
2891 }
2892 if (need_tracing_session) {
2893 session_unlock_list();
2894 }
2895 init_setup_error:
2896 return ret;
2897 }
2898
2899 /*
2900 * Thread managing health check socket.
2901 */
2902 static void *thread_manage_health(void *data)
2903 {
2904 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2905 uint32_t revents, nb_fd;
2906 struct lttng_poll_event events;
2907 struct lttcomm_health_msg msg;
2908 struct lttcomm_health_data reply;
2909
2910 DBG("[thread] Manage health check started");
2911
2912 rcu_register_thread();
2913
2914 /* Create unix socket */
2915 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2916 if (sock < 0) {
2917 ERR("Unable to create health check Unix socket");
2918 ret = -1;
2919 goto error;
2920 }
2921
2922 /*
2923 * Set the CLOEXEC flag. Return code is useless because either way, the
2924 * show must go on.
2925 */
2926 (void) utils_set_fd_cloexec(sock);
2927
2928 ret = lttcomm_listen_unix_sock(sock);
2929 if (ret < 0) {
2930 goto error;
2931 }
2932
2933 /*
2934 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2935 * more will be added to this poll set.
2936 */
2937 ret = create_thread_poll_set(&events, 2);
2938 if (ret < 0) {
2939 goto error;
2940 }
2941
2942 /* Add the application registration socket */
2943 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2944 if (ret < 0) {
2945 goto error;
2946 }
2947
2948 while (1) {
2949 DBG("Health check ready");
2950
2951 nb_fd = LTTNG_POLL_GETNB(&events);
2952
2953 /* Inifinite blocking call, waiting for transmission */
2954 restart:
2955 ret = lttng_poll_wait(&events, -1);
2956 if (ret < 0) {
2957 /*
2958 * Restart interrupted system call.
2959 */
2960 if (errno == EINTR) {
2961 goto restart;
2962 }
2963 goto error;
2964 }
2965
2966 for (i = 0; i < nb_fd; i++) {
2967 /* Fetch once the poll data */
2968 revents = LTTNG_POLL_GETEV(&events, i);
2969 pollfd = LTTNG_POLL_GETFD(&events, i);
2970
2971 /* Thread quit pipe has been closed. Killing thread. */
2972 ret = check_thread_quit_pipe(pollfd, revents);
2973 if (ret) {
2974 err = 0;
2975 goto exit;
2976 }
2977
2978 /* Event on the registration socket */
2979 if (pollfd == sock) {
2980 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2981 ERR("Health socket poll error");
2982 goto error;
2983 }
2984 }
2985 }
2986
2987 new_sock = lttcomm_accept_unix_sock(sock);
2988 if (new_sock < 0) {
2989 goto error;
2990 }
2991
2992 /*
2993 * Set the CLOEXEC flag. Return code is useless because either way, the
2994 * show must go on.
2995 */
2996 (void) utils_set_fd_cloexec(new_sock);
2997
2998 DBG("Receiving data from client for health...");
2999 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3000 if (ret <= 0) {
3001 DBG("Nothing recv() from client... continuing");
3002 ret = close(new_sock);
3003 if (ret) {
3004 PERROR("close");
3005 }
3006 new_sock = -1;
3007 continue;
3008 }
3009
3010 rcu_thread_online();
3011
3012 switch (msg.component) {
3013 case LTTNG_HEALTH_CMD:
3014 reply.ret_code = health_check_state(&health_thread_cmd);
3015 break;
3016 case LTTNG_HEALTH_APP_MANAGE:
3017 reply.ret_code = health_check_state(&health_thread_app_manage);
3018 break;
3019 case LTTNG_HEALTH_APP_REG:
3020 reply.ret_code = health_check_state(&health_thread_app_reg);
3021 break;
3022 case LTTNG_HEALTH_KERNEL:
3023 reply.ret_code = health_check_state(&health_thread_kernel);
3024 break;
3025 case LTTNG_HEALTH_CONSUMER:
3026 reply.ret_code = check_consumer_health();
3027 break;
3028 case LTTNG_HEALTH_ALL:
3029 reply.ret_code =
3030 health_check_state(&health_thread_app_manage) &&
3031 health_check_state(&health_thread_app_reg) &&
3032 health_check_state(&health_thread_cmd) &&
3033 health_check_state(&health_thread_kernel) &&
3034 check_consumer_health();
3035 break;
3036 default:
3037 reply.ret_code = LTTNG_ERR_UND;
3038 break;
3039 }
3040
3041 /*
3042 * Flip ret value since 0 is a success and 1 indicates a bad health for
3043 * the client where in the sessiond it is the opposite. Again, this is
3044 * just to make things easier for us poor developer which enjoy a lot
3045 * lazyness.
3046 */
3047 if (reply.ret_code == 0 || reply.ret_code == 1) {
3048 reply.ret_code = !reply.ret_code;
3049 }
3050
3051 DBG2("Health check return value %d", reply.ret_code);
3052
3053 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3054 if (ret < 0) {
3055 ERR("Failed to send health data back to client");
3056 }
3057
3058 /* End of transmission */
3059 ret = close(new_sock);
3060 if (ret) {
3061 PERROR("close");
3062 }
3063 new_sock = -1;
3064 }
3065
3066 exit:
3067 error:
3068 if (err) {
3069 ERR("Health error occurred in %s", __func__);
3070 }
3071 DBG("Health check thread dying");
3072 unlink(health_unix_sock_path);
3073 if (sock >= 0) {
3074 ret = close(sock);
3075 if (ret) {
3076 PERROR("close");
3077 }
3078 }
3079 if (new_sock >= 0) {
3080 ret = close(new_sock);
3081 if (ret) {
3082 PERROR("close");
3083 }
3084 }
3085
3086 lttng_poll_clean(&events);
3087
3088 rcu_unregister_thread();
3089 return NULL;
3090 }
3091
3092 /*
3093 * This thread manage all clients request using the unix client socket for
3094 * communication.
3095 */
3096 static void *thread_manage_clients(void *data)
3097 {
3098 int sock = -1, ret, i, pollfd, err = -1;
3099 int sock_error;
3100 uint32_t revents, nb_fd;
3101 struct command_ctx *cmd_ctx = NULL;
3102 struct lttng_poll_event events;
3103
3104 DBG("[thread] Manage client started");
3105
3106 testpoint(thread_manage_clients);
3107
3108 rcu_register_thread();
3109
3110 health_code_update(&health_thread_cmd);
3111
3112 ret = lttcomm_listen_unix_sock(client_sock);
3113 if (ret < 0) {
3114 goto error_listen;
3115 }
3116
3117 /*
3118 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3119 * more will be added to this poll set.
3120 */
3121 ret = create_thread_poll_set(&events, 2);
3122 if (ret < 0) {
3123 goto error_create_poll;
3124 }
3125
3126 /* Add the application registration socket */
3127 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3128 if (ret < 0) {
3129 goto error;
3130 }
3131
3132 /*
3133 * Notify parent pid that we are ready to accept command for client side.
3134 */
3135 if (opt_sig_parent) {
3136 kill(ppid, SIGUSR1);
3137 }
3138
3139 testpoint(thread_manage_clients_before_loop);
3140
3141 health_code_update(&health_thread_cmd);
3142
3143 while (1) {
3144 DBG("Accepting client command ...");
3145
3146 nb_fd = LTTNG_POLL_GETNB(&events);
3147
3148 /* Inifinite blocking call, waiting for transmission */
3149 restart:
3150 health_poll_update(&health_thread_cmd);
3151 ret = lttng_poll_wait(&events, -1);
3152 health_poll_update(&health_thread_cmd);
3153 if (ret < 0) {
3154 /*
3155 * Restart interrupted system call.
3156 */
3157 if (errno == EINTR) {
3158 goto restart;
3159 }
3160 goto error;
3161 }
3162
3163 for (i = 0; i < nb_fd; i++) {
3164 /* Fetch once the poll data */
3165 revents = LTTNG_POLL_GETEV(&events, i);
3166 pollfd = LTTNG_POLL_GETFD(&events, i);
3167
3168 health_code_update(&health_thread_cmd);
3169
3170 /* Thread quit pipe has been closed. Killing thread. */
3171 ret = check_thread_quit_pipe(pollfd, revents);
3172 if (ret) {
3173 err = 0;
3174 goto exit;
3175 }
3176
3177 /* Event on the registration socket */
3178 if (pollfd == client_sock) {
3179 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3180 ERR("Client socket poll error");
3181 goto error;
3182 }
3183 }
3184 }
3185
3186 DBG("Wait for client response");
3187
3188 health_code_update(&health_thread_cmd);
3189
3190 sock = lttcomm_accept_unix_sock(client_sock);
3191 if (sock < 0) {
3192 goto error;
3193 }
3194
3195 /*
3196 * Set the CLOEXEC flag. Return code is useless because either way, the
3197 * show must go on.
3198 */
3199 (void) utils_set_fd_cloexec(sock);
3200
3201 /* Set socket option for credentials retrieval */
3202 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3203 if (ret < 0) {
3204 goto error;
3205 }
3206
3207 /* Allocate context command to process the client request */
3208 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3209 if (cmd_ctx == NULL) {
3210 PERROR("zmalloc cmd_ctx");
3211 goto error;
3212 }
3213
3214 /* Allocate data buffer for reception */
3215 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3216 if (cmd_ctx->lsm == NULL) {
3217 PERROR("zmalloc cmd_ctx->lsm");
3218 goto error;
3219 }
3220
3221 cmd_ctx->llm = NULL;
3222 cmd_ctx->session = NULL;
3223
3224 health_code_update(&health_thread_cmd);
3225
3226 /*
3227 * Data is received from the lttng client. The struct
3228 * lttcomm_session_msg (lsm) contains the command and data request of
3229 * the client.
3230 */
3231 DBG("Receiving data from client ...");
3232 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3233 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3234 if (ret <= 0) {
3235 DBG("Nothing recv() from client... continuing");
3236 ret = close(sock);
3237 if (ret) {
3238 PERROR("close");
3239 }
3240 sock = -1;
3241 clean_command_ctx(&cmd_ctx);
3242 continue;
3243 }
3244
3245 health_code_update(&health_thread_cmd);
3246
3247 // TODO: Validate cmd_ctx including sanity check for
3248 // security purpose.
3249
3250 rcu_thread_online();
3251 /*
3252 * This function dispatch the work to the kernel or userspace tracer
3253 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3254 * informations for the client. The command context struct contains
3255 * everything this function may needs.
3256 */
3257 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3258 rcu_thread_offline();
3259 if (ret < 0) {
3260 if (sock_error) {
3261 ret = close(sock);
3262 if (ret) {
3263 PERROR("close");
3264 }
3265 sock = -1;
3266 }
3267 /*
3268 * TODO: Inform client somehow of the fatal error. At
3269 * this point, ret < 0 means that a zmalloc failed
3270 * (ENOMEM). Error detected but still accept
3271 * command, unless a socket error has been
3272 * detected.
3273 */
3274 clean_command_ctx(&cmd_ctx);
3275 continue;
3276 }
3277
3278 health_code_update(&health_thread_cmd);
3279
3280 DBG("Sending response (size: %d, retcode: %s)",
3281 cmd_ctx->lttng_msg_size,
3282 lttng_strerror(-cmd_ctx->llm->ret_code));
3283 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3284 if (ret < 0) {
3285 ERR("Failed to send data back to client");
3286 }
3287
3288 /* End of transmission */
3289 ret = close(sock);
3290 if (ret) {
3291 PERROR("close");
3292 }
3293 sock = -1;
3294
3295 clean_command_ctx(&cmd_ctx);
3296
3297 health_code_update(&health_thread_cmd);
3298 }
3299
3300 exit:
3301 error:
3302 if (sock >= 0) {
3303 ret = close(sock);
3304 if (ret) {
3305 PERROR("close");
3306 }
3307 }
3308
3309 lttng_poll_clean(&events);
3310 clean_command_ctx(&cmd_ctx);
3311
3312 error_listen:
3313 error_create_poll:
3314 unlink(client_unix_sock_path);
3315 if (client_sock >= 0) {
3316 ret = close(client_sock);
3317 if (ret) {
3318 PERROR("close");
3319 }
3320 }
3321
3322 if (err) {
3323 health_error(&health_thread_cmd);
3324 ERR("Health error occurred in %s", __func__);
3325 }
3326
3327 health_exit(&health_thread_cmd);
3328
3329 DBG("Client thread dying");
3330
3331 rcu_unregister_thread();
3332 return NULL;
3333 }
3334
3335
3336 /*
3337 * usage function on stderr
3338 */
3339 static void usage(void)
3340 {
3341 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3342 fprintf(stderr, " -h, --help Display this usage.\n");
3343 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3344 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3345 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3346 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3347 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3348 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3349 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3350 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3351 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3352 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3353 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3354 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3355 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3356 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3357 fprintf(stderr, " -V, --version Show version number.\n");
3358 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3359 fprintf(stderr, " -q, --quiet No output at all.\n");
3360 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3361 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3362 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3363 }
3364
3365 /*
3366 * daemon argument parsing
3367 */
3368 static int parse_args(int argc, char **argv)
3369 {
3370 int c;
3371
3372 static struct option long_options[] = {
3373 { "client-sock", 1, 0, 'c' },
3374 { "apps-sock", 1, 0, 'a' },
3375 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3376 { "kconsumerd-err-sock", 1, 0, 'E' },
3377 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3378 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3379 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3380 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3381 { "consumerd32-path", 1, 0, 'u' },
3382 { "consumerd32-libdir", 1, 0, 'U' },
3383 { "consumerd64-path", 1, 0, 't' },
3384 { "consumerd64-libdir", 1, 0, 'T' },
3385 { "daemonize", 0, 0, 'd' },
3386 { "sig-parent", 0, 0, 'S' },
3387 { "help", 0, 0, 'h' },
3388 { "group", 1, 0, 'g' },
3389 { "version", 0, 0, 'V' },
3390 { "quiet", 0, 0, 'q' },
3391 { "verbose", 0, 0, 'v' },
3392 { "verbose-consumer", 0, 0, 'Z' },
3393 { "no-kernel", 0, 0, 'N' },
3394 { NULL, 0, 0, 0 }
3395 };
3396
3397 while (1) {
3398 int option_index = 0;
3399 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3400 long_options, &option_index);
3401 if (c == -1) {
3402 break;
3403 }
3404
3405 switch (c) {
3406 case 0:
3407 fprintf(stderr, "option %s", long_options[option_index].name);
3408 if (optarg) {
3409 fprintf(stderr, " with arg %s\n", optarg);
3410 }
3411 break;
3412 case 'c':
3413 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3414 break;
3415 case 'a':
3416 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3417 break;
3418 case 'd':
3419 opt_daemon = 1;
3420 break;
3421 case 'g':
3422 opt_tracing_group = optarg;
3423 break;
3424 case 'h':
3425 usage();
3426 exit(EXIT_FAILURE);
3427 case 'V':
3428 fprintf(stdout, "%s\n", VERSION);
3429 exit(EXIT_SUCCESS);
3430 case 'S':
3431 opt_sig_parent = 1;
3432 break;
3433 case 'E':
3434 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3435 break;
3436 case 'C':
3437 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3438 break;
3439 case 'F':
3440 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3441 break;
3442 case 'D':
3443 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3444 break;
3445 case 'H':
3446 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3447 break;
3448 case 'G':
3449 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3450 break;
3451 case 'N':
3452 opt_no_kernel = 1;
3453 break;
3454 case 'q':
3455 lttng_opt_quiet = 1;
3456 break;
3457 case 'v':
3458 /* Verbose level can increase using multiple -v */
3459 lttng_opt_verbose += 1;
3460 break;
3461 case 'Z':
3462 opt_verbose_consumer += 1;
3463 break;
3464 case 'u':
3465 consumerd32_bin= optarg;
3466 break;
3467 case 'U':
3468 consumerd32_libdir = optarg;
3469 break;
3470 case 't':
3471 consumerd64_bin = optarg;
3472 break;
3473 case 'T':
3474 consumerd64_libdir = optarg;
3475 break;
3476 default:
3477 /* Unknown option or other error.
3478 * Error is printed by getopt, just return */
3479 return -1;
3480 }
3481 }
3482
3483 return 0;
3484 }
3485
3486 /*
3487 * Creates the two needed socket by the daemon.
3488 * apps_sock - The communication socket for all UST apps.
3489 * client_sock - The communication of the cli tool (lttng).
3490 */
3491 static int init_daemon_socket(void)
3492 {
3493 int ret = 0;
3494 mode_t old_umask;
3495
3496 old_umask = umask(0);
3497
3498 /* Create client tool unix socket */
3499 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3500 if (client_sock < 0) {
3501 ERR("Create unix sock failed: %s", client_unix_sock_path);
3502 ret = -1;
3503 goto end;
3504 }
3505
3506 /* Set the cloexec flag */
3507 ret = utils_set_fd_cloexec(client_sock);
3508 if (ret < 0) {
3509 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3510 "Continuing but note that the consumer daemon will have a "
3511 "reference to this socket on exec()", client_sock);
3512 }
3513
3514 /* File permission MUST be 660 */
3515 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3516 if (ret < 0) {
3517 ERR("Set file permissions failed: %s", client_unix_sock_path);
3518 PERROR("chmod");
3519 goto end;
3520 }
3521
3522 /* Create the application unix socket */
3523 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3524 if (apps_sock < 0) {
3525 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3526 ret = -1;
3527 goto end;
3528 }
3529
3530 /* Set the cloexec flag */
3531 ret = utils_set_fd_cloexec(apps_sock);
3532 if (ret < 0) {
3533 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3534 "Continuing but note that the consumer daemon will have a "
3535 "reference to this socket on exec()", apps_sock);
3536 }
3537
3538 /* File permission MUST be 666 */
3539 ret = chmod(apps_unix_sock_path,
3540 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3541 if (ret < 0) {
3542 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3543 PERROR("chmod");
3544 goto end;
3545 }
3546
3547 DBG3("Session daemon client socket %d and application socket %d created",
3548 client_sock, apps_sock);
3549
3550 end:
3551 umask(old_umask);
3552 return ret;
3553 }
3554
3555 /*
3556 * Check if the global socket is available, and if a daemon is answering at the
3557 * other side. If yes, error is returned.
3558 */
3559 static int check_existing_daemon(void)
3560 {
3561 /* Is there anybody out there ? */
3562 if (lttng_session_daemon_alive()) {
3563 return -EEXIST;
3564 }
3565
3566 return 0;
3567 }
3568
3569 /*
3570 * Set the tracing group gid onto the client socket.
3571 *
3572 * Race window between mkdir and chown is OK because we are going from more
3573 * permissive (root.root) to less permissive (root.tracing).
3574 */
3575 static int set_permissions(char *rundir)
3576 {
3577 int ret;
3578 gid_t gid;
3579
3580 ret = allowed_group();
3581 if (ret < 0) {
3582 WARN("No tracing group detected");
3583 ret = 0;
3584 goto end;
3585 }
3586
3587 gid = ret;
3588
3589 /* Set lttng run dir */
3590 ret = chown(rundir, 0, gid);
3591 if (ret < 0) {
3592 ERR("Unable to set group on %s", rundir);
3593 PERROR("chown");
3594 }
3595
3596 /* Ensure tracing group can search the run dir */
3597 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3598 if (ret < 0) {
3599 ERR("Unable to set permissions on %s", rundir);
3600 PERROR("chmod");
3601 }
3602
3603 /* lttng client socket path */
3604 ret = chown(client_unix_sock_path, 0, gid);
3605 if (ret < 0) {
3606 ERR("Unable to set group on %s", client_unix_sock_path);
3607 PERROR("chown");
3608 }
3609
3610 /* kconsumer error socket path */
3611 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3612 if (ret < 0) {
3613 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3614 PERROR("chown");
3615 }
3616
3617 /* 64-bit ustconsumer error socket path */
3618 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3619 if (ret < 0) {
3620 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3621 PERROR("chown");
3622 }
3623
3624 /* 32-bit ustconsumer compat32 error socket path */
3625 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3626 if (ret < 0) {
3627 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3628 PERROR("chown");
3629 }
3630
3631 DBG("All permissions are set");
3632
3633 end:
3634 return ret;
3635 }
3636
3637 /*
3638 * Create the lttng run directory needed for all global sockets and pipe.
3639 */
3640 static int create_lttng_rundir(const char *rundir)
3641 {
3642 int ret;
3643
3644 DBG3("Creating LTTng run directory: %s", rundir);
3645
3646 ret = mkdir(rundir, S_IRWXU);
3647 if (ret < 0) {
3648 if (errno != EEXIST) {
3649 ERR("Unable to create %s", rundir);
3650 goto error;
3651 } else {
3652 ret = 0;
3653 }
3654 }
3655
3656 error:
3657 return ret;
3658 }
3659
3660 /*
3661 * Setup sockets and directory needed by the kconsumerd communication with the
3662 * session daemon.
3663 */
3664 static int set_consumer_sockets(struct consumer_data *consumer_data,
3665 const char *rundir)
3666 {
3667 int ret;
3668 char path[PATH_MAX];
3669
3670 switch (consumer_data->type) {
3671 case LTTNG_CONSUMER_KERNEL:
3672 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3673 break;
3674 case LTTNG_CONSUMER64_UST:
3675 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3676 break;
3677 case LTTNG_CONSUMER32_UST:
3678 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3679 break;
3680 default:
3681 ERR("Consumer type unknown");
3682 ret = -EINVAL;
3683 goto error;
3684 }
3685
3686 DBG2("Creating consumer directory: %s", path);
3687
3688 ret = mkdir(path, S_IRWXU);
3689 if (ret < 0) {
3690 if (errno != EEXIST) {
3691 PERROR("mkdir");
3692 ERR("Failed to create %s", path);
3693 goto error;
3694 }
3695 ret = -1;
3696 }
3697
3698 /* Create the kconsumerd error unix socket */
3699 consumer_data->err_sock =
3700 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3701 if (consumer_data->err_sock < 0) {
3702 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3703 ret = -1;
3704 goto error;
3705 }
3706
3707 /* File permission MUST be 660 */
3708 ret = chmod(consumer_data->err_unix_sock_path,
3709 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3710 if (ret < 0) {
3711 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3712 PERROR("chmod");
3713 goto error;
3714 }
3715
3716 error:
3717 return ret;
3718 }
3719
3720 /*
3721 * Signal handler for the daemon
3722 *
3723 * Simply stop all worker threads, leaving main() return gracefully after
3724 * joining all threads and calling cleanup().
3725 */
3726 static void sighandler(int sig)
3727 {
3728 switch (sig) {
3729 case SIGPIPE:
3730 DBG("SIGPIPE caught");
3731 return;
3732 case SIGINT:
3733 DBG("SIGINT caught");
3734 stop_threads();
3735 break;
3736 case SIGTERM:
3737 DBG("SIGTERM caught");
3738 stop_threads();
3739 break;
3740 default:
3741 break;
3742 }
3743 }
3744
3745 /*
3746 * Setup signal handler for :
3747 * SIGINT, SIGTERM, SIGPIPE
3748 */
3749 static int set_signal_handler(void)
3750 {
3751 int ret = 0;
3752 struct sigaction sa;
3753 sigset_t sigset;
3754
3755 if ((ret = sigemptyset(&sigset)) < 0) {
3756 PERROR("sigemptyset");
3757 return ret;
3758 }
3759
3760 sa.sa_handler = sighandler;
3761 sa.sa_mask = sigset;
3762 sa.sa_flags = 0;
3763 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3764 PERROR("sigaction");
3765 return ret;
3766 }
3767
3768 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3769 PERROR("sigaction");
3770 return ret;
3771 }
3772
3773 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3774 PERROR("sigaction");
3775 return ret;
3776 }
3777
3778 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3779
3780 return ret;
3781 }
3782
3783 /*
3784 * Set open files limit to unlimited. This daemon can open a large number of
3785 * file descriptors in order to consumer multiple kernel traces.
3786 */
3787 static void set_ulimit(void)
3788 {
3789 int ret;
3790 struct rlimit lim;
3791
3792 /* The kernel does not allowed an infinite limit for open files */
3793 lim.rlim_cur = 65535;
3794 lim.rlim_max = 65535;
3795
3796 ret = setrlimit(RLIMIT_NOFILE, &lim);
3797 if (ret < 0) {
3798 PERROR("failed to set open files limit");
3799 }
3800 }
3801
3802 /*
3803 * main
3804 */
3805 int main(int argc, char **argv)
3806 {
3807 int ret = 0;
3808 void *status;
3809 const char *home_path, *env_app_timeout;
3810
3811 init_kernel_workarounds();
3812
3813 rcu_register_thread();
3814
3815 setup_consumerd_path();
3816
3817 /* Parse arguments */
3818 progname = argv[0];
3819 if ((ret = parse_args(argc, argv) < 0)) {
3820 goto error;
3821 }
3822
3823 /* Daemonize */
3824 if (opt_daemon) {
3825 int i;
3826
3827 /*
3828 * fork
3829 * child: setsid, close FD 0, 1, 2, chdir /
3830 * parent: exit (if fork is successful)
3831 */
3832 ret = daemon(0, 0);
3833 if (ret < 0) {
3834 PERROR("daemon");
3835 goto error;
3836 }
3837 /*
3838 * We are in the child. Make sure all other file
3839 * descriptors are closed, in case we are called with
3840 * more opened file descriptors than the standard ones.
3841 */
3842 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3843 (void) close(i);
3844 }
3845 }
3846
3847 /* Create thread quit pipe */
3848 if ((ret = init_thread_quit_pipe()) < 0) {
3849 goto error;
3850 }
3851
3852 /* Check if daemon is UID = 0 */
3853 is_root = !getuid();
3854
3855 if (is_root) {
3856 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3857
3858 /* Create global run dir with root access */
3859 ret = create_lttng_rundir(rundir);
3860 if (ret < 0) {
3861 goto error;
3862 }
3863
3864 if (strlen(apps_unix_sock_path) == 0) {
3865 snprintf(apps_unix_sock_path, PATH_MAX,
3866 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3867 }
3868
3869 if (strlen(client_unix_sock_path) == 0) {
3870 snprintf(client_unix_sock_path, PATH_MAX,
3871 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3872 }
3873
3874 /* Set global SHM for ust */
3875 if (strlen(wait_shm_path) == 0) {
3876 snprintf(wait_shm_path, PATH_MAX,
3877 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3878 }
3879
3880 if (strlen(health_unix_sock_path) == 0) {
3881 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3882 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3883 }
3884
3885 /* Setup kernel consumerd path */
3886 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3887 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3888 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3889 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3890
3891 DBG2("Kernel consumer err path: %s",
3892 kconsumer_data.err_unix_sock_path);
3893 DBG2("Kernel consumer cmd path: %s",
3894 kconsumer_data.cmd_unix_sock_path);
3895 } else {
3896 home_path = get_home_dir();
3897 if (home_path == NULL) {
3898 /* TODO: Add --socket PATH option */
3899 ERR("Can't get HOME directory for sockets creation.");
3900 ret = -EPERM;
3901 goto error;
3902 }
3903
3904 /*
3905 * Create rundir from home path. This will create something like
3906 * $HOME/.lttng
3907 */
3908 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3909 if (ret < 0) {
3910 ret = -ENOMEM;
3911 goto error;
3912 }
3913
3914 ret = create_lttng_rundir(rundir);
3915 if (ret < 0) {
3916 goto error;
3917 }
3918
3919 if (strlen(apps_unix_sock_path) == 0) {
3920 snprintf(apps_unix_sock_path, PATH_MAX,
3921 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3922 }
3923
3924 /* Set the cli tool unix socket path */
3925 if (strlen(client_unix_sock_path) == 0) {
3926 snprintf(client_unix_sock_path, PATH_MAX,
3927 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3928 }
3929
3930 /* Set global SHM for ust */
3931 if (strlen(wait_shm_path) == 0) {
3932 snprintf(wait_shm_path, PATH_MAX,
3933 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3934 }
3935
3936 /* Set health check Unix path */
3937 if (strlen(health_unix_sock_path) == 0) {
3938 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3939 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3940 }
3941 }
3942
3943 /* Set consumer initial state */
3944 kernel_consumerd_state = CONSUMER_STOPPED;
3945 ust_consumerd_state = CONSUMER_STOPPED;
3946
3947 DBG("Client socket path %s", client_unix_sock_path);
3948 DBG("Application socket path %s", apps_unix_sock_path);
3949 DBG("LTTng run directory path: %s", rundir);
3950
3951 /* 32 bits consumerd path setup */
3952 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3953 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3954 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3955 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3956
3957 DBG2("UST consumer 32 bits err path: %s",
3958 ustconsumer32_data.err_unix_sock_path);
3959 DBG2("UST consumer 32 bits cmd path: %s",
3960 ustconsumer32_data.cmd_unix_sock_path);
3961
3962 /* 64 bits consumerd path setup */
3963 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3964 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
3965 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
3966 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
3967
3968 DBG2("UST consumer 64 bits err path: %s",
3969 ustconsumer64_data.err_unix_sock_path);
3970 DBG2("UST consumer 64 bits cmd path: %s",
3971 ustconsumer64_data.cmd_unix_sock_path);
3972
3973 /*
3974 * See if daemon already exist.
3975 */
3976 if ((ret = check_existing_daemon()) < 0) {
3977 ERR("Already running daemon.\n");
3978 /*
3979 * We do not goto exit because we must not cleanup()
3980 * because a daemon is already running.
3981 */
3982 goto error;
3983 }
3984
3985 /*
3986 * Init UST app hash table. Alloc hash table before this point since
3987 * cleanup() can get called after that point.
3988 */
3989 ust_app_ht_alloc();
3990
3991 /* After this point, we can safely call cleanup() with "goto exit" */
3992
3993 /*
3994 * These actions must be executed as root. We do that *after* setting up
3995 * the sockets path because we MUST make the check for another daemon using
3996 * those paths *before* trying to set the kernel consumer sockets and init
3997 * kernel tracer.
3998 */
3999 if (is_root) {
4000 ret = set_consumer_sockets(&kconsumer_data, rundir);
4001 if (ret < 0) {
4002 goto exit;
4003 }
4004
4005 /* Setup kernel tracer */
4006 if (!opt_no_kernel) {
4007 init_kernel_tracer();
4008 }
4009
4010 /* Set ulimit for open files */
4011 set_ulimit();
4012 }
4013 /* init lttng_fd tracking must be done after set_ulimit. */
4014 lttng_fd_init();
4015
4016 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4017 if (ret < 0) {
4018 goto exit;
4019 }
4020
4021 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4022 if (ret < 0) {
4023 goto exit;
4024 }
4025
4026 if ((ret = set_signal_handler()) < 0) {
4027 goto exit;
4028 }
4029
4030 /* Setup the needed unix socket */
4031 if ((ret = init_daemon_socket()) < 0) {
4032 goto exit;
4033 }
4034
4035 /* Set credentials to socket */
4036 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4037 goto exit;
4038 }
4039
4040 /* Get parent pid if -S, --sig-parent is specified. */
4041 if (opt_sig_parent) {
4042 ppid = getppid();
4043 }
4044
4045 /* Setup the kernel pipe for waking up the kernel thread */
4046 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4047 goto exit;
4048 }
4049
4050 /* Setup the thread apps communication pipe. */
4051 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4052 goto exit;
4053 }
4054
4055 /* Init UST command queue. */
4056 cds_wfq_init(&ust_cmd_queue.queue);
4057
4058 /*
4059 * Get session list pointer. This pointer MUST NOT be free(). This list is
4060 * statically declared in session.c
4061 */
4062 session_list_ptr = session_get_list();
4063
4064 /* Set up max poll set size */
4065 lttng_poll_set_max_size();
4066
4067 cmd_init();
4068
4069 /* Init all health thread counters. */
4070 health_init(&health_thread_cmd);
4071 health_init(&health_thread_kernel);
4072 health_init(&health_thread_app_manage);
4073 health_init(&health_thread_app_reg);
4074
4075 /*
4076 * Init health counters of the consumer thread. We do a quick hack here to
4077 * the state of the consumer health is fine even if the thread is not
4078 * started. Once the thread starts, the health state is updated with a poll
4079 * value to set a health code path. This is simply to ease our life and has
4080 * no cost what so ever.
4081 */
4082 health_init(&kconsumer_data.health);
4083 health_poll_update(&kconsumer_data.health);
4084 health_init(&ustconsumer32_data.health);
4085 health_poll_update(&ustconsumer32_data.health);
4086 health_init(&ustconsumer64_data.health);
4087 health_poll_update(&ustconsumer64_data.health);
4088
4089 /* Check for the application socket timeout env variable. */
4090 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4091 if (env_app_timeout) {
4092 app_socket_timeout = atoi(env_app_timeout);
4093 } else {
4094 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4095 }
4096
4097 /* Create thread to manage the client socket */
4098 ret = pthread_create(&health_thread, NULL,
4099 thread_manage_health, (void *) NULL);
4100 if (ret != 0) {
4101 PERROR("pthread_create health");
4102 goto exit_health;
4103 }
4104
4105 /* Create thread to manage the client socket */
4106 ret = pthread_create(&client_thread, NULL,
4107 thread_manage_clients, (void *) NULL);
4108 if (ret != 0) {
4109 PERROR("pthread_create clients");
4110 goto exit_client;
4111 }
4112
4113 /* Create thread to dispatch registration */
4114 ret = pthread_create(&dispatch_thread, NULL,
4115 thread_dispatch_ust_registration, (void *) NULL);
4116 if (ret != 0) {
4117 PERROR("pthread_create dispatch");
4118 goto exit_dispatch;
4119 }
4120
4121 /* Create thread to manage application registration. */
4122 ret = pthread_create(&reg_apps_thread, NULL,
4123 thread_registration_apps, (void *) NULL);
4124 if (ret != 0) {
4125 PERROR("pthread_create registration");
4126 goto exit_reg_apps;
4127 }
4128
4129 /* Create thread to manage application socket */
4130 ret = pthread_create(&apps_thread, NULL,
4131 thread_manage_apps, (void *) NULL);
4132 if (ret != 0) {
4133 PERROR("pthread_create apps");
4134 goto exit_apps;
4135 }
4136
4137 /* Create kernel thread to manage kernel event */
4138 ret = pthread_create(&kernel_thread, NULL,
4139 thread_manage_kernel, (void *) NULL);
4140 if (ret != 0) {
4141 PERROR("pthread_create kernel");
4142 goto exit_kernel;
4143 }
4144
4145 ret = pthread_join(kernel_thread, &status);
4146 if (ret != 0) {
4147 PERROR("pthread_join");
4148 goto error; /* join error, exit without cleanup */
4149 }
4150
4151 exit_kernel:
4152 ret = pthread_join(apps_thread, &status);
4153 if (ret != 0) {
4154 PERROR("pthread_join");
4155 goto error; /* join error, exit without cleanup */
4156 }
4157
4158 exit_apps:
4159 ret = pthread_join(reg_apps_thread, &status);
4160 if (ret != 0) {
4161 PERROR("pthread_join");
4162 goto error; /* join error, exit without cleanup */
4163 }
4164
4165 exit_reg_apps:
4166 ret = pthread_join(dispatch_thread, &status);
4167 if (ret != 0) {
4168 PERROR("pthread_join");
4169 goto error; /* join error, exit without cleanup */
4170 }
4171
4172 exit_dispatch:
4173 ret = pthread_join(client_thread, &status);
4174 if (ret != 0) {
4175 PERROR("pthread_join");
4176 goto error; /* join error, exit without cleanup */
4177 }
4178
4179 ret = join_consumer_thread(&kconsumer_data);
4180 if (ret != 0) {
4181 PERROR("join_consumer");
4182 goto error; /* join error, exit without cleanup */
4183 }
4184
4185 ret = join_consumer_thread(&ustconsumer32_data);
4186 if (ret != 0) {
4187 PERROR("join_consumer ust32");
4188 goto error; /* join error, exit without cleanup */
4189 }
4190
4191 ret = join_consumer_thread(&ustconsumer64_data);
4192 if (ret != 0) {
4193 PERROR("join_consumer ust64");
4194 goto error; /* join error, exit without cleanup */
4195 }
4196
4197 exit_client:
4198 ret = pthread_join(health_thread, &status);
4199 if (ret != 0) {
4200 PERROR("pthread_join health thread");
4201 goto error; /* join error, exit without cleanup */
4202 }
4203
4204 exit_health:
4205 exit:
4206 /*
4207 * cleanup() is called when no other thread is running.
4208 */
4209 rcu_thread_online();
4210 cleanup();
4211 rcu_thread_offline();
4212 rcu_unregister_thread();
4213 if (!ret) {
4214 exit(EXIT_SUCCESS);
4215 }
4216 error:
4217 exit(EXIT_FAILURE);
4218 }
This page took 0.198661 seconds and 6 git commands to generate.