Fix: change health poll update to entry/exit calls
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/uatomic.h>
36 #include <unistd.h>
37 #include <config.h>
38
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "kernel.h"
55 #include "kernel-consumer.h"
56 #include "modprobe.h"
57 #include "shm.h"
58 #include "ust-ctl.h"
59 #include "ust-consumer.h"
60 #include "utils.h"
61 #include "fd-limit.h"
62 #include "health.h"
63 #include "testpoint.h"
64
65 #define CONSUMERD_FILE "lttng-consumerd"
66
67 /* Const values */
68 const char default_home_dir[] = DEFAULT_HOME_DIR;
69 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
70 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
71 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
72
73 const char *progname;
74 const char *opt_tracing_group;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread;
153 static pthread_t reg_apps_thread;
154 static pthread_t client_thread;
155 static pthread_t kernel_thread;
156 static pthread_t dispatch_thread;
157 static pthread_t health_thread;
158
159 /*
160 * UST registration command queue. This queue is tied with a futex and uses a N
161 * wakers / 1 waiter implemented and detailed in futex.c/.h
162 *
163 * The thread_manage_apps and thread_dispatch_ust_registration interact with
164 * this queue and the wait/wake scheme.
165 */
166 static struct ust_cmd_queue ust_cmd_queue;
167
168 /*
169 * Pointer initialized before thread creation.
170 *
171 * This points to the tracing session list containing the session count and a
172 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
173 * MUST NOT be taken if you call a public function in session.c.
174 *
175 * The lock is nested inside the structure: session_list_ptr->lock. Please use
176 * session_lock_list and session_unlock_list for lock acquisition.
177 */
178 static struct ltt_session_list *session_list_ptr;
179
180 int ust_consumerd64_fd = -1;
181 int ust_consumerd32_fd = -1;
182
183 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
184 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
185 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
186 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
187
188 static const char *module_proc_lttng = "/proc/lttng";
189
190 /*
191 * Consumer daemon state which is changed when spawning it, killing it or in
192 * case of a fatal error.
193 */
194 enum consumerd_state {
195 CONSUMER_STARTED = 1,
196 CONSUMER_STOPPED = 2,
197 CONSUMER_ERROR = 3,
198 };
199
200 /*
201 * This consumer daemon state is used to validate if a client command will be
202 * able to reach the consumer. If not, the client is informed. For instance,
203 * doing a "lttng start" when the consumer state is set to ERROR will return an
204 * error to the client.
205 *
206 * The following example shows a possible race condition of this scheme:
207 *
208 * consumer thread error happens
209 * client cmd arrives
210 * client cmd checks state -> still OK
211 * consumer thread exit, sets error
212 * client cmd try to talk to consumer
213 * ...
214 *
215 * However, since the consumer is a different daemon, we have no way of making
216 * sure the command will reach it safely even with this state flag. This is why
217 * we consider that up to the state validation during command processing, the
218 * command is safe. After that, we can not guarantee the correctness of the
219 * client request vis-a-vis the consumer.
220 */
221 static enum consumerd_state ust_consumerd_state;
222 static enum consumerd_state kernel_consumerd_state;
223
224 /*
225 * Socket timeout for receiving and sending in seconds.
226 */
227 static int app_socket_timeout;
228
229 static
230 void setup_consumerd_path(void)
231 {
232 const char *bin, *libdir;
233
234 /*
235 * Allow INSTALL_BIN_PATH to be used as a target path for the
236 * native architecture size consumer if CONFIG_CONSUMER*_PATH
237 * has not been defined.
238 */
239 #if (CAA_BITS_PER_LONG == 32)
240 if (!consumerd32_bin[0]) {
241 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
242 }
243 if (!consumerd32_libdir[0]) {
244 consumerd32_libdir = INSTALL_LIB_PATH;
245 }
246 #elif (CAA_BITS_PER_LONG == 64)
247 if (!consumerd64_bin[0]) {
248 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd64_libdir[0]) {
251 consumerd64_libdir = INSTALL_LIB_PATH;
252 }
253 #else
254 #error "Unknown bitness"
255 #endif
256
257 /*
258 * runtime env. var. overrides the build default.
259 */
260 bin = getenv("LTTNG_CONSUMERD32_BIN");
261 if (bin) {
262 consumerd32_bin = bin;
263 }
264 bin = getenv("LTTNG_CONSUMERD64_BIN");
265 if (bin) {
266 consumerd64_bin = bin;
267 }
268 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
269 if (libdir) {
270 consumerd32_libdir = libdir;
271 }
272 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
273 if (libdir) {
274 consumerd64_libdir = libdir;
275 }
276 }
277
278 /*
279 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
280 */
281 static int create_thread_poll_set(struct lttng_poll_event *events,
282 unsigned int size)
283 {
284 int ret;
285
286 if (events == NULL || size == 0) {
287 ret = -1;
288 goto error;
289 }
290
291 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
292 if (ret < 0) {
293 goto error;
294 }
295
296 /* Add quit pipe */
297 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
298 if (ret < 0) {
299 goto error;
300 }
301
302 return 0;
303
304 error:
305 return ret;
306 }
307
308 /*
309 * Check if the thread quit pipe was triggered.
310 *
311 * Return 1 if it was triggered else 0;
312 */
313 static int check_thread_quit_pipe(int fd, uint32_t events)
314 {
315 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
316 return 1;
317 }
318
319 return 0;
320 }
321
322 /*
323 * Return group ID of the tracing group or -1 if not found.
324 */
325 static gid_t allowed_group(void)
326 {
327 struct group *grp;
328
329 if (opt_tracing_group) {
330 grp = getgrnam(opt_tracing_group);
331 } else {
332 grp = getgrnam(default_tracing_group);
333 }
334 if (!grp) {
335 return -1;
336 } else {
337 return grp->gr_gid;
338 }
339 }
340
341 /*
342 * Init thread quit pipe.
343 *
344 * Return -1 on error or 0 if all pipes are created.
345 */
346 static int init_thread_quit_pipe(void)
347 {
348 int ret, i;
349
350 ret = pipe(thread_quit_pipe);
351 if (ret < 0) {
352 PERROR("thread quit pipe");
353 goto error;
354 }
355
356 for (i = 0; i < 2; i++) {
357 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
358 if (ret < 0) {
359 PERROR("fcntl");
360 goto error;
361 }
362 }
363
364 error:
365 return ret;
366 }
367
368 /*
369 * Stop all threads by closing the thread quit pipe.
370 */
371 static void stop_threads(void)
372 {
373 int ret;
374
375 /* Stopping all threads */
376 DBG("Terminating all threads");
377 ret = notify_thread_pipe(thread_quit_pipe[1]);
378 if (ret < 0) {
379 ERR("write error on thread quit pipe");
380 }
381
382 /* Dispatch thread */
383 CMM_STORE_SHARED(dispatch_thread_exit, 1);
384 futex_nto1_wake(&ust_cmd_queue.futex);
385 }
386
387 /*
388 * Cleanup the daemon
389 */
390 static void cleanup(void)
391 {
392 int ret;
393 char *cmd = NULL;
394 struct ltt_session *sess, *stmp;
395
396 DBG("Cleaning up");
397
398 /* First thing first, stop all threads */
399 utils_close_pipe(thread_quit_pipe);
400
401 DBG("Removing %s directory", rundir);
402 ret = asprintf(&cmd, "rm -rf %s", rundir);
403 if (ret < 0) {
404 ERR("asprintf failed. Something is really wrong!");
405 }
406
407 /* Remove lttng run directory */
408 ret = system(cmd);
409 if (ret < 0) {
410 ERR("Unable to clean %s", rundir);
411 }
412 free(cmd);
413 free(rundir);
414
415 DBG("Cleaning up all sessions");
416
417 /* Destroy session list mutex */
418 if (session_list_ptr != NULL) {
419 pthread_mutex_destroy(&session_list_ptr->lock);
420
421 /* Cleanup ALL session */
422 cds_list_for_each_entry_safe(sess, stmp,
423 &session_list_ptr->head, list) {
424 cmd_destroy_session(sess, kernel_poll_pipe[1]);
425 }
426 }
427
428 DBG("Closing all UST sockets");
429 ust_app_clean_list();
430
431 if (is_root && !opt_no_kernel) {
432 DBG2("Closing kernel fd");
433 if (kernel_tracer_fd >= 0) {
434 ret = close(kernel_tracer_fd);
435 if (ret) {
436 PERROR("close");
437 }
438 }
439 DBG("Unloading kernel modules");
440 modprobe_remove_lttng_all();
441 }
442
443 /* <fun> */
444 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
445 "Matthew, BEET driven development works!%c[%dm",
446 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
447 /* </fun> */
448 }
449
450 /*
451 * Send data on a unix socket using the liblttsessiondcomm API.
452 *
453 * Return lttcomm error code.
454 */
455 static int send_unix_sock(int sock, void *buf, size_t len)
456 {
457 /* Check valid length */
458 if (len == 0) {
459 return -1;
460 }
461
462 return lttcomm_send_unix_sock(sock, buf, len);
463 }
464
465 /*
466 * Free memory of a command context structure.
467 */
468 static void clean_command_ctx(struct command_ctx **cmd_ctx)
469 {
470 DBG("Clean command context structure");
471 if (*cmd_ctx) {
472 if ((*cmd_ctx)->llm) {
473 free((*cmd_ctx)->llm);
474 }
475 if ((*cmd_ctx)->lsm) {
476 free((*cmd_ctx)->lsm);
477 }
478 free(*cmd_ctx);
479 *cmd_ctx = NULL;
480 }
481 }
482
483 /*
484 * Notify UST applications using the shm mmap futex.
485 */
486 static int notify_ust_apps(int active)
487 {
488 char *wait_shm_mmap;
489
490 DBG("Notifying applications of session daemon state: %d", active);
491
492 /* See shm.c for this call implying mmap, shm and futex calls */
493 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
494 if (wait_shm_mmap == NULL) {
495 goto error;
496 }
497
498 /* Wake waiting process */
499 futex_wait_update((int32_t *) wait_shm_mmap, active);
500
501 /* Apps notified successfully */
502 return 0;
503
504 error:
505 return -1;
506 }
507
508 /*
509 * Setup the outgoing data buffer for the response (llm) by allocating the
510 * right amount of memory and copying the original information from the lsm
511 * structure.
512 *
513 * Return total size of the buffer pointed by buf.
514 */
515 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
516 {
517 int ret, buf_size;
518
519 buf_size = size;
520
521 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
522 if (cmd_ctx->llm == NULL) {
523 PERROR("zmalloc");
524 ret = -ENOMEM;
525 goto error;
526 }
527
528 /* Copy common data */
529 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
530 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
531
532 cmd_ctx->llm->data_size = size;
533 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
534
535 return buf_size;
536
537 error:
538 return ret;
539 }
540
541 /*
542 * Update the kernel poll set of all channel fd available over all tracing
543 * session. Add the wakeup pipe at the end of the set.
544 */
545 static int update_kernel_poll(struct lttng_poll_event *events)
546 {
547 int ret;
548 struct ltt_session *session;
549 struct ltt_kernel_channel *channel;
550
551 DBG("Updating kernel poll set");
552
553 session_lock_list();
554 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
555 session_lock(session);
556 if (session->kernel_session == NULL) {
557 session_unlock(session);
558 continue;
559 }
560
561 cds_list_for_each_entry(channel,
562 &session->kernel_session->channel_list.head, list) {
563 /* Add channel fd to the kernel poll set */
564 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
565 if (ret < 0) {
566 session_unlock(session);
567 goto error;
568 }
569 DBG("Channel fd %d added to kernel set", channel->fd);
570 }
571 session_unlock(session);
572 }
573 session_unlock_list();
574
575 return 0;
576
577 error:
578 session_unlock_list();
579 return -1;
580 }
581
582 /*
583 * Find the channel fd from 'fd' over all tracing session. When found, check
584 * for new channel stream and send those stream fds to the kernel consumer.
585 *
586 * Useful for CPU hotplug feature.
587 */
588 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
589 {
590 int ret = 0;
591 struct ltt_session *session;
592 struct ltt_kernel_session *ksess;
593 struct ltt_kernel_channel *channel;
594
595 DBG("Updating kernel streams for channel fd %d", fd);
596
597 session_lock_list();
598 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
599 session_lock(session);
600 if (session->kernel_session == NULL) {
601 session_unlock(session);
602 continue;
603 }
604 ksess = session->kernel_session;
605
606 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
607 if (channel->fd == fd) {
608 DBG("Channel found, updating kernel streams");
609 ret = kernel_open_channel_stream(channel);
610 if (ret < 0) {
611 goto error;
612 }
613
614 /*
615 * Have we already sent fds to the consumer? If yes, it means
616 * that tracing is started so it is safe to send our updated
617 * stream fds.
618 */
619 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
620 struct lttng_ht_iter iter;
621 struct consumer_socket *socket;
622
623 rcu_read_lock();
624 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
625 &iter.iter, socket, node.node) {
626 /* Code flow error */
627 assert(socket->fd >= 0);
628
629 pthread_mutex_lock(socket->lock);
630 ret = kernel_consumer_send_channel_stream(socket,
631 channel, ksess);
632 pthread_mutex_unlock(socket->lock);
633 if (ret < 0) {
634 rcu_read_unlock();
635 goto error;
636 }
637 }
638 rcu_read_unlock();
639 }
640 goto error;
641 }
642 }
643 session_unlock(session);
644 }
645 session_unlock_list();
646 return ret;
647
648 error:
649 session_unlock(session);
650 session_unlock_list();
651 return ret;
652 }
653
654 /*
655 * For each tracing session, update newly registered apps.
656 */
657 static void update_ust_app(int app_sock)
658 {
659 struct ltt_session *sess, *stmp;
660
661 session_lock_list();
662
663 /* For all tracing session(s) */
664 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
665 session_lock(sess);
666 if (sess->ust_session) {
667 ust_app_global_update(sess->ust_session, app_sock);
668 }
669 session_unlock(sess);
670 }
671
672 session_unlock_list();
673 }
674
675 /*
676 * This thread manage event coming from the kernel.
677 *
678 * Features supported in this thread:
679 * -) CPU Hotplug
680 */
681 static void *thread_manage_kernel(void *data)
682 {
683 int ret, i, pollfd, update_poll_flag = 1, err = -1;
684 uint32_t revents, nb_fd;
685 char tmp;
686 struct lttng_poll_event events;
687
688 DBG("[thread] Thread manage kernel started");
689
690 health_register(HEALTH_TYPE_KERNEL);
691
692 /*
693 * This first step of the while is to clean this structure which could free
694 * non NULL pointers so zero it before the loop.
695 */
696 memset(&events, 0, sizeof(events));
697
698 if (testpoint(thread_manage_kernel)) {
699 goto error_testpoint;
700 }
701
702 health_code_update();
703
704 if (testpoint(thread_manage_kernel_before_loop)) {
705 goto error_testpoint;
706 }
707
708 while (1) {
709 health_code_update();
710
711 if (update_poll_flag == 1) {
712 /* Clean events object. We are about to populate it again. */
713 lttng_poll_clean(&events);
714
715 ret = create_thread_poll_set(&events, 2);
716 if (ret < 0) {
717 goto error_poll_create;
718 }
719
720 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
721 if (ret < 0) {
722 goto error;
723 }
724
725 /* This will add the available kernel channel if any. */
726 ret = update_kernel_poll(&events);
727 if (ret < 0) {
728 goto error;
729 }
730 update_poll_flag = 0;
731 }
732
733 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
734
735 /* Poll infinite value of time */
736 restart:
737 health_poll_entry();
738 ret = lttng_poll_wait(&events, -1);
739 health_poll_exit();
740 if (ret < 0) {
741 /*
742 * Restart interrupted system call.
743 */
744 if (errno == EINTR) {
745 goto restart;
746 }
747 goto error;
748 } else if (ret == 0) {
749 /* Should not happen since timeout is infinite */
750 ERR("Return value of poll is 0 with an infinite timeout.\n"
751 "This should not have happened! Continuing...");
752 continue;
753 }
754
755 nb_fd = ret;
756
757 for (i = 0; i < nb_fd; i++) {
758 /* Fetch once the poll data */
759 revents = LTTNG_POLL_GETEV(&events, i);
760 pollfd = LTTNG_POLL_GETFD(&events, i);
761
762 health_code_update();
763
764 /* Thread quit pipe has been closed. Killing thread. */
765 ret = check_thread_quit_pipe(pollfd, revents);
766 if (ret) {
767 err = 0;
768 goto exit;
769 }
770
771 /* Check for data on kernel pipe */
772 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
773 do {
774 ret = read(kernel_poll_pipe[0], &tmp, 1);
775 } while (ret < 0 && errno == EINTR);
776 /*
777 * Ret value is useless here, if this pipe gets any actions an
778 * update is required anyway.
779 */
780 update_poll_flag = 1;
781 continue;
782 } else {
783 /*
784 * New CPU detected by the kernel. Adding kernel stream to
785 * kernel session and updating the kernel consumer
786 */
787 if (revents & LPOLLIN) {
788 ret = update_kernel_stream(&kconsumer_data, pollfd);
789 if (ret < 0) {
790 continue;
791 }
792 break;
793 /*
794 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
795 * and unregister kernel stream at this point.
796 */
797 }
798 }
799 }
800 }
801
802 exit:
803 error:
804 lttng_poll_clean(&events);
805 error_poll_create:
806 error_testpoint:
807 utils_close_pipe(kernel_poll_pipe);
808 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
809 if (err) {
810 health_error();
811 ERR("Health error occurred in %s", __func__);
812 WARN("Kernel thread died unexpectedly. "
813 "Kernel tracing can continue but CPU hotplug is disabled.");
814 }
815 health_unregister();
816 DBG("Kernel thread dying");
817 return NULL;
818 }
819
820 /*
821 * Signal pthread condition of the consumer data that the thread.
822 */
823 static void signal_consumer_condition(struct consumer_data *data, int state)
824 {
825 pthread_mutex_lock(&data->cond_mutex);
826
827 /*
828 * The state is set before signaling. It can be any value, it's the waiter
829 * job to correctly interpret this condition variable associated to the
830 * consumer pthread_cond.
831 *
832 * A value of 0 means that the corresponding thread of the consumer data
833 * was not started. 1 indicates that the thread has started and is ready
834 * for action. A negative value means that there was an error during the
835 * thread bootstrap.
836 */
837 data->consumer_thread_is_ready = state;
838 (void) pthread_cond_signal(&data->cond);
839
840 pthread_mutex_unlock(&data->cond_mutex);
841 }
842
843 /*
844 * This thread manage the consumer error sent back to the session daemon.
845 */
846 static void *thread_manage_consumer(void *data)
847 {
848 int sock = -1, i, ret, pollfd, err = -1;
849 uint32_t revents, nb_fd;
850 enum lttcomm_return_code code;
851 struct lttng_poll_event events;
852 struct consumer_data *consumer_data = data;
853
854 DBG("[thread] Manage consumer started");
855
856 health_register(HEALTH_TYPE_CONSUMER);
857
858 health_code_update();
859
860 /*
861 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
862 * Nothing more will be added to this poll set.
863 */
864 ret = create_thread_poll_set(&events, 2);
865 if (ret < 0) {
866 goto error_poll;
867 }
868
869 /*
870 * The error socket here is already in a listening state which was done
871 * just before spawning this thread to avoid a race between the consumer
872 * daemon exec trying to connect and the listen() call.
873 */
874 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
875 if (ret < 0) {
876 goto error;
877 }
878
879 health_code_update();
880
881 /* Inifinite blocking call, waiting for transmission */
882 restart:
883 health_poll_entry();
884
885 if (testpoint(thread_manage_consumer)) {
886 goto error;
887 }
888
889 ret = lttng_poll_wait(&events, -1);
890 health_poll_exit();
891 if (ret < 0) {
892 /*
893 * Restart interrupted system call.
894 */
895 if (errno == EINTR) {
896 goto restart;
897 }
898 goto error;
899 }
900
901 nb_fd = ret;
902
903 for (i = 0; i < nb_fd; i++) {
904 /* Fetch once the poll data */
905 revents = LTTNG_POLL_GETEV(&events, i);
906 pollfd = LTTNG_POLL_GETFD(&events, i);
907
908 health_code_update();
909
910 /* Thread quit pipe has been closed. Killing thread. */
911 ret = check_thread_quit_pipe(pollfd, revents);
912 if (ret) {
913 err = 0;
914 goto exit;
915 }
916
917 /* Event on the registration socket */
918 if (pollfd == consumer_data->err_sock) {
919 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
920 ERR("consumer err socket poll error");
921 goto error;
922 }
923 }
924 }
925
926 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
927 if (sock < 0) {
928 goto error;
929 }
930
931 /*
932 * Set the CLOEXEC flag. Return code is useless because either way, the
933 * show must go on.
934 */
935 (void) utils_set_fd_cloexec(sock);
936
937 health_code_update();
938
939 DBG2("Receiving code from consumer err_sock");
940
941 /* Getting status code from kconsumerd */
942 ret = lttcomm_recv_unix_sock(sock, &code,
943 sizeof(enum lttcomm_return_code));
944 if (ret <= 0) {
945 goto error;
946 }
947
948 health_code_update();
949
950 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
951 consumer_data->cmd_sock =
952 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
953 if (consumer_data->cmd_sock < 0) {
954 /* On error, signal condition and quit. */
955 signal_consumer_condition(consumer_data, -1);
956 PERROR("consumer connect");
957 goto error;
958 }
959 signal_consumer_condition(consumer_data, 1);
960 DBG("Consumer command socket ready");
961 } else {
962 ERR("consumer error when waiting for SOCK_READY : %s",
963 lttcomm_get_readable_code(-code));
964 goto error;
965 }
966
967 /* Remove the kconsumerd error sock since we've established a connexion */
968 ret = lttng_poll_del(&events, consumer_data->err_sock);
969 if (ret < 0) {
970 goto error;
971 }
972
973 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
974 if (ret < 0) {
975 goto error;
976 }
977
978 health_code_update();
979
980 /* Inifinite blocking call, waiting for transmission */
981 restart_poll:
982 health_poll_entry();
983 ret = lttng_poll_wait(&events, -1);
984 health_poll_exit();
985 if (ret < 0) {
986 /*
987 * Restart interrupted system call.
988 */
989 if (errno == EINTR) {
990 goto restart_poll;
991 }
992 goto error;
993 }
994
995 nb_fd = ret;
996
997 for (i = 0; i < nb_fd; i++) {
998 /* Fetch once the poll data */
999 revents = LTTNG_POLL_GETEV(&events, i);
1000 pollfd = LTTNG_POLL_GETFD(&events, i);
1001
1002 health_code_update();
1003
1004 /* Thread quit pipe has been closed. Killing thread. */
1005 ret = check_thread_quit_pipe(pollfd, revents);
1006 if (ret) {
1007 err = 0;
1008 goto exit;
1009 }
1010
1011 /* Event on the kconsumerd socket */
1012 if (pollfd == sock) {
1013 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1014 ERR("consumer err socket second poll error");
1015 goto error;
1016 }
1017 }
1018 }
1019
1020 health_code_update();
1021
1022 /* Wait for any kconsumerd error */
1023 ret = lttcomm_recv_unix_sock(sock, &code,
1024 sizeof(enum lttcomm_return_code));
1025 if (ret <= 0) {
1026 ERR("consumer closed the command socket");
1027 goto error;
1028 }
1029
1030 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1031
1032 exit:
1033 error:
1034 /* Immediately set the consumerd state to stopped */
1035 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1036 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1037 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1038 consumer_data->type == LTTNG_CONSUMER32_UST) {
1039 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1040 } else {
1041 /* Code flow error... */
1042 assert(0);
1043 }
1044
1045 if (consumer_data->err_sock >= 0) {
1046 ret = close(consumer_data->err_sock);
1047 if (ret) {
1048 PERROR("close");
1049 }
1050 }
1051 if (consumer_data->cmd_sock >= 0) {
1052 ret = close(consumer_data->cmd_sock);
1053 if (ret) {
1054 PERROR("close");
1055 }
1056 }
1057 if (sock >= 0) {
1058 ret = close(sock);
1059 if (ret) {
1060 PERROR("close");
1061 }
1062 }
1063
1064 unlink(consumer_data->err_unix_sock_path);
1065 unlink(consumer_data->cmd_unix_sock_path);
1066 consumer_data->pid = 0;
1067
1068 lttng_poll_clean(&events);
1069 error_poll:
1070 if (err) {
1071 health_error();
1072 ERR("Health error occurred in %s", __func__);
1073 }
1074 health_unregister();
1075 DBG("consumer thread cleanup completed");
1076
1077 return NULL;
1078 }
1079
1080 /*
1081 * This thread manage application communication.
1082 */
1083 static void *thread_manage_apps(void *data)
1084 {
1085 int i, ret, pollfd, err = -1;
1086 uint32_t revents, nb_fd;
1087 struct ust_command ust_cmd;
1088 struct lttng_poll_event events;
1089
1090 DBG("[thread] Manage application started");
1091
1092 rcu_register_thread();
1093 rcu_thread_online();
1094
1095 health_register(HEALTH_TYPE_APP_MANAGE);
1096
1097 if (testpoint(thread_manage_apps)) {
1098 goto error_testpoint;
1099 }
1100
1101 health_code_update();
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 if (testpoint(thread_manage_apps_before_loop)) {
1114 goto error;
1115 }
1116
1117 health_code_update();
1118
1119 while (1) {
1120 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1121
1122 /* Inifinite blocking call, waiting for transmission */
1123 restart:
1124 health_poll_entry();
1125 ret = lttng_poll_wait(&events, -1);
1126 health_poll_exit();
1127 if (ret < 0) {
1128 /*
1129 * Restart interrupted system call.
1130 */
1131 if (errno == EINTR) {
1132 goto restart;
1133 }
1134 goto error;
1135 }
1136
1137 nb_fd = ret;
1138
1139 for (i = 0; i < nb_fd; i++) {
1140 /* Fetch once the poll data */
1141 revents = LTTNG_POLL_GETEV(&events, i);
1142 pollfd = LTTNG_POLL_GETFD(&events, i);
1143
1144 health_code_update();
1145
1146 /* Thread quit pipe has been closed. Killing thread. */
1147 ret = check_thread_quit_pipe(pollfd, revents);
1148 if (ret) {
1149 err = 0;
1150 goto exit;
1151 }
1152
1153 /* Inspect the apps cmd pipe */
1154 if (pollfd == apps_cmd_pipe[0]) {
1155 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1156 ERR("Apps command pipe error");
1157 goto error;
1158 } else if (revents & LPOLLIN) {
1159 /* Empty pipe */
1160 do {
1161 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1162 } while (ret < 0 && errno == EINTR);
1163 if (ret < 0 || ret < sizeof(ust_cmd)) {
1164 PERROR("read apps cmd pipe");
1165 goto error;
1166 }
1167
1168 health_code_update();
1169
1170 /* Register applicaton to the session daemon */
1171 ret = ust_app_register(&ust_cmd.reg_msg,
1172 ust_cmd.sock);
1173 if (ret == -ENOMEM) {
1174 goto error;
1175 } else if (ret < 0) {
1176 break;
1177 }
1178
1179 health_code_update();
1180
1181 /*
1182 * Validate UST version compatibility.
1183 */
1184 ret = ust_app_validate_version(ust_cmd.sock);
1185 if (ret >= 0) {
1186 /*
1187 * Add channel(s) and event(s) to newly registered apps
1188 * from lttng global UST domain.
1189 */
1190 update_ust_app(ust_cmd.sock);
1191 }
1192
1193 health_code_update();
1194
1195 ret = ust_app_register_done(ust_cmd.sock);
1196 if (ret < 0) {
1197 /*
1198 * If the registration is not possible, we simply
1199 * unregister the apps and continue
1200 */
1201 ust_app_unregister(ust_cmd.sock);
1202 } else {
1203 /*
1204 * We only monitor the error events of the socket. This
1205 * thread does not handle any incoming data from UST
1206 * (POLLIN).
1207 */
1208 ret = lttng_poll_add(&events, ust_cmd.sock,
1209 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1210 if (ret < 0) {
1211 goto error;
1212 }
1213
1214 /* Set socket timeout for both receiving and ending */
1215 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1216 app_socket_timeout);
1217 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1218 app_socket_timeout);
1219
1220 DBG("Apps with sock %d added to poll set",
1221 ust_cmd.sock);
1222 }
1223
1224 health_code_update();
1225
1226 break;
1227 }
1228 } else {
1229 /*
1230 * At this point, we know that a registered application made
1231 * the event at poll_wait.
1232 */
1233 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1234 /* Removing from the poll set */
1235 ret = lttng_poll_del(&events, pollfd);
1236 if (ret < 0) {
1237 goto error;
1238 }
1239
1240 /* Socket closed on remote end. */
1241 ust_app_unregister(pollfd);
1242 break;
1243 }
1244 }
1245
1246 health_code_update();
1247 }
1248 }
1249
1250 exit:
1251 error:
1252 lttng_poll_clean(&events);
1253 error_poll_create:
1254 error_testpoint:
1255 utils_close_pipe(apps_cmd_pipe);
1256 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1257
1258 /*
1259 * We don't clean the UST app hash table here since already registered
1260 * applications can still be controlled so let them be until the session
1261 * daemon dies or the applications stop.
1262 */
1263
1264 if (err) {
1265 health_error();
1266 ERR("Health error occurred in %s", __func__);
1267 }
1268 health_unregister();
1269 DBG("Application communication apps thread cleanup complete");
1270 rcu_thread_offline();
1271 rcu_unregister_thread();
1272 return NULL;
1273 }
1274
1275 /*
1276 * Dispatch request from the registration threads to the application
1277 * communication thread.
1278 */
1279 static void *thread_dispatch_ust_registration(void *data)
1280 {
1281 int ret;
1282 struct cds_wfq_node *node;
1283 struct ust_command *ust_cmd = NULL;
1284
1285 DBG("[thread] Dispatch UST command started");
1286
1287 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1288 /* Atomically prepare the queue futex */
1289 futex_nto1_prepare(&ust_cmd_queue.futex);
1290
1291 do {
1292 /* Dequeue command for registration */
1293 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1294 if (node == NULL) {
1295 DBG("Woken up but nothing in the UST command queue");
1296 /* Continue thread execution */
1297 break;
1298 }
1299
1300 ust_cmd = caa_container_of(node, struct ust_command, node);
1301
1302 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1303 " gid:%d sock:%d name:%s (version %d.%d)",
1304 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1305 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1306 ust_cmd->sock, ust_cmd->reg_msg.name,
1307 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1308 /*
1309 * Inform apps thread of the new application registration. This
1310 * call is blocking so we can be assured that the data will be read
1311 * at some point in time or wait to the end of the world :)
1312 */
1313 if (apps_cmd_pipe[1] >= 0) {
1314 do {
1315 ret = write(apps_cmd_pipe[1], ust_cmd,
1316 sizeof(struct ust_command));
1317 } while (ret < 0 && errno == EINTR);
1318 if (ret < 0 || ret != sizeof(struct ust_command)) {
1319 PERROR("write apps cmd pipe");
1320 if (errno == EBADF) {
1321 /*
1322 * We can't inform the application thread to process
1323 * registration. We will exit or else application
1324 * registration will not occur and tracing will never
1325 * start.
1326 */
1327 goto error;
1328 }
1329 }
1330 } else {
1331 /* Application manager thread is not available. */
1332 ret = close(ust_cmd->sock);
1333 if (ret < 0) {
1334 PERROR("close ust_cmd sock");
1335 }
1336 }
1337 free(ust_cmd);
1338 } while (node != NULL);
1339
1340 /* Futex wait on queue. Blocking call on futex() */
1341 futex_nto1_wait(&ust_cmd_queue.futex);
1342 }
1343
1344 error:
1345 DBG("Dispatch thread dying");
1346 return NULL;
1347 }
1348
1349 /*
1350 * This thread manage application registration.
1351 */
1352 static void *thread_registration_apps(void *data)
1353 {
1354 int sock = -1, i, ret, pollfd, err = -1;
1355 uint32_t revents, nb_fd;
1356 struct lttng_poll_event events;
1357 /*
1358 * Get allocated in this thread, enqueued to a global queue, dequeued and
1359 * freed in the manage apps thread.
1360 */
1361 struct ust_command *ust_cmd = NULL;
1362
1363 DBG("[thread] Manage application registration started");
1364
1365 health_register(HEALTH_TYPE_APP_REG);
1366
1367 if (testpoint(thread_registration_apps)) {
1368 goto error_testpoint;
1369 }
1370
1371 ret = lttcomm_listen_unix_sock(apps_sock);
1372 if (ret < 0) {
1373 goto error_listen;
1374 }
1375
1376 /*
1377 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1378 * more will be added to this poll set.
1379 */
1380 ret = create_thread_poll_set(&events, 2);
1381 if (ret < 0) {
1382 goto error_create_poll;
1383 }
1384
1385 /* Add the application registration socket */
1386 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1387 if (ret < 0) {
1388 goto error_poll_add;
1389 }
1390
1391 /* Notify all applications to register */
1392 ret = notify_ust_apps(1);
1393 if (ret < 0) {
1394 ERR("Failed to notify applications or create the wait shared memory.\n"
1395 "Execution continues but there might be problem for already\n"
1396 "running applications that wishes to register.");
1397 }
1398
1399 while (1) {
1400 DBG("Accepting application registration");
1401
1402 /* Inifinite blocking call, waiting for transmission */
1403 restart:
1404 health_poll_entry();
1405 ret = lttng_poll_wait(&events, -1);
1406 health_poll_exit();
1407 if (ret < 0) {
1408 /*
1409 * Restart interrupted system call.
1410 */
1411 if (errno == EINTR) {
1412 goto restart;
1413 }
1414 goto error;
1415 }
1416
1417 nb_fd = ret;
1418
1419 for (i = 0; i < nb_fd; i++) {
1420 health_code_update();
1421
1422 /* Fetch once the poll data */
1423 revents = LTTNG_POLL_GETEV(&events, i);
1424 pollfd = LTTNG_POLL_GETFD(&events, i);
1425
1426 /* Thread quit pipe has been closed. Killing thread. */
1427 ret = check_thread_quit_pipe(pollfd, revents);
1428 if (ret) {
1429 err = 0;
1430 goto exit;
1431 }
1432
1433 /* Event on the registration socket */
1434 if (pollfd == apps_sock) {
1435 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1436 ERR("Register apps socket poll error");
1437 goto error;
1438 } else if (revents & LPOLLIN) {
1439 sock = lttcomm_accept_unix_sock(apps_sock);
1440 if (sock < 0) {
1441 goto error;
1442 }
1443
1444 /*
1445 * Set the CLOEXEC flag. Return code is useless because
1446 * either way, the show must go on.
1447 */
1448 (void) utils_set_fd_cloexec(sock);
1449
1450 /* Create UST registration command for enqueuing */
1451 ust_cmd = zmalloc(sizeof(struct ust_command));
1452 if (ust_cmd == NULL) {
1453 PERROR("ust command zmalloc");
1454 goto error;
1455 }
1456
1457 /*
1458 * Using message-based transmissions to ensure we don't
1459 * have to deal with partially received messages.
1460 */
1461 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1462 if (ret < 0) {
1463 ERR("Exhausted file descriptors allowed for applications.");
1464 free(ust_cmd);
1465 ret = close(sock);
1466 if (ret) {
1467 PERROR("close");
1468 }
1469 sock = -1;
1470 continue;
1471 }
1472 health_code_update();
1473 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1474 sizeof(struct ust_register_msg));
1475 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1476 if (ret < 0) {
1477 PERROR("lttcomm_recv_unix_sock register apps");
1478 } else {
1479 ERR("Wrong size received on apps register");
1480 }
1481 free(ust_cmd);
1482 ret = close(sock);
1483 if (ret) {
1484 PERROR("close");
1485 }
1486 lttng_fd_put(LTTNG_FD_APPS, 1);
1487 sock = -1;
1488 continue;
1489 }
1490 health_code_update();
1491
1492 ust_cmd->sock = sock;
1493 sock = -1;
1494
1495 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1496 " gid:%d sock:%d name:%s (version %d.%d)",
1497 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1498 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1499 ust_cmd->sock, ust_cmd->reg_msg.name,
1500 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1501
1502 /*
1503 * Lock free enqueue the registration request. The red pill
1504 * has been taken! This apps will be part of the *system*.
1505 */
1506 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1507
1508 /*
1509 * Wake the registration queue futex. Implicit memory
1510 * barrier with the exchange in cds_wfq_enqueue.
1511 */
1512 futex_nto1_wake(&ust_cmd_queue.futex);
1513 }
1514 }
1515 }
1516 }
1517
1518 exit:
1519 error:
1520 if (err) {
1521 health_error();
1522 ERR("Health error occurred in %s", __func__);
1523 }
1524
1525 /* Notify that the registration thread is gone */
1526 notify_ust_apps(0);
1527
1528 if (apps_sock >= 0) {
1529 ret = close(apps_sock);
1530 if (ret) {
1531 PERROR("close");
1532 }
1533 }
1534 if (sock >= 0) {
1535 ret = close(sock);
1536 if (ret) {
1537 PERROR("close");
1538 }
1539 lttng_fd_put(LTTNG_FD_APPS, 1);
1540 }
1541 unlink(apps_unix_sock_path);
1542
1543 error_poll_add:
1544 lttng_poll_clean(&events);
1545 error_listen:
1546 error_create_poll:
1547 error_testpoint:
1548 DBG("UST Registration thread cleanup complete");
1549 health_unregister();
1550
1551 return NULL;
1552 }
1553
1554 /*
1555 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1556 * exec or it will fails.
1557 */
1558 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1559 {
1560 int ret, clock_ret;
1561 struct timespec timeout;
1562
1563 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1564 consumer_data->consumer_thread_is_ready = 0;
1565
1566 /* Setup pthread condition */
1567 ret = pthread_condattr_init(&consumer_data->condattr);
1568 if (ret != 0) {
1569 errno = ret;
1570 PERROR("pthread_condattr_init consumer data");
1571 goto error;
1572 }
1573
1574 /*
1575 * Set the monotonic clock in order to make sure we DO NOT jump in time
1576 * between the clock_gettime() call and the timedwait call. See bug #324
1577 * for a more details and how we noticed it.
1578 */
1579 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1580 if (ret != 0) {
1581 errno = ret;
1582 PERROR("pthread_condattr_setclock consumer data");
1583 goto error;
1584 }
1585
1586 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1587 if (ret != 0) {
1588 errno = ret;
1589 PERROR("pthread_cond_init consumer data");
1590 goto error;
1591 }
1592
1593 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1594 consumer_data);
1595 if (ret != 0) {
1596 PERROR("pthread_create consumer");
1597 ret = -1;
1598 goto error;
1599 }
1600
1601 /* We are about to wait on a pthread condition */
1602 pthread_mutex_lock(&consumer_data->cond_mutex);
1603
1604 /* Get time for sem_timedwait absolute timeout */
1605 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1606 /*
1607 * Set the timeout for the condition timed wait even if the clock gettime
1608 * call fails since we might loop on that call and we want to avoid to
1609 * increment the timeout too many times.
1610 */
1611 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1612
1613 /*
1614 * The following loop COULD be skipped in some conditions so this is why we
1615 * set ret to 0 in order to make sure at least one round of the loop is
1616 * done.
1617 */
1618 ret = 0;
1619
1620 /*
1621 * Loop until the condition is reached or when a timeout is reached. Note
1622 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1623 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1624 * possible. This loop does not take any chances and works with both of
1625 * them.
1626 */
1627 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1628 if (clock_ret < 0) {
1629 PERROR("clock_gettime spawn consumer");
1630 /* Infinite wait for the consumerd thread to be ready */
1631 ret = pthread_cond_wait(&consumer_data->cond,
1632 &consumer_data->cond_mutex);
1633 } else {
1634 ret = pthread_cond_timedwait(&consumer_data->cond,
1635 &consumer_data->cond_mutex, &timeout);
1636 }
1637 }
1638
1639 /* Release the pthread condition */
1640 pthread_mutex_unlock(&consumer_data->cond_mutex);
1641
1642 if (ret != 0) {
1643 errno = ret;
1644 if (ret == ETIMEDOUT) {
1645 /*
1646 * Call has timed out so we kill the kconsumerd_thread and return
1647 * an error.
1648 */
1649 ERR("Condition timed out. The consumer thread was never ready."
1650 " Killing it");
1651 ret = pthread_cancel(consumer_data->thread);
1652 if (ret < 0) {
1653 PERROR("pthread_cancel consumer thread");
1654 }
1655 } else {
1656 PERROR("pthread_cond_wait failed consumer thread");
1657 }
1658 goto error;
1659 }
1660
1661 pthread_mutex_lock(&consumer_data->pid_mutex);
1662 if (consumer_data->pid == 0) {
1663 ERR("Consumerd did not start");
1664 pthread_mutex_unlock(&consumer_data->pid_mutex);
1665 goto error;
1666 }
1667 pthread_mutex_unlock(&consumer_data->pid_mutex);
1668
1669 return 0;
1670
1671 error:
1672 return ret;
1673 }
1674
1675 /*
1676 * Join consumer thread
1677 */
1678 static int join_consumer_thread(struct consumer_data *consumer_data)
1679 {
1680 void *status;
1681
1682 /* Consumer pid must be a real one. */
1683 if (consumer_data->pid > 0) {
1684 int ret;
1685 ret = kill(consumer_data->pid, SIGTERM);
1686 if (ret) {
1687 ERR("Error killing consumer daemon");
1688 return ret;
1689 }
1690 return pthread_join(consumer_data->thread, &status);
1691 } else {
1692 return 0;
1693 }
1694 }
1695
1696 /*
1697 * Fork and exec a consumer daemon (consumerd).
1698 *
1699 * Return pid if successful else -1.
1700 */
1701 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1702 {
1703 int ret;
1704 pid_t pid;
1705 const char *consumer_to_use;
1706 const char *verbosity;
1707 struct stat st;
1708
1709 DBG("Spawning consumerd");
1710
1711 pid = fork();
1712 if (pid == 0) {
1713 /*
1714 * Exec consumerd.
1715 */
1716 if (opt_verbose_consumer) {
1717 verbosity = "--verbose";
1718 } else {
1719 verbosity = "--quiet";
1720 }
1721 switch (consumer_data->type) {
1722 case LTTNG_CONSUMER_KERNEL:
1723 /*
1724 * Find out which consumerd to execute. We will first try the
1725 * 64-bit path, then the sessiond's installation directory, and
1726 * fallback on the 32-bit one,
1727 */
1728 DBG3("Looking for a kernel consumer at these locations:");
1729 DBG3(" 1) %s", consumerd64_bin);
1730 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1731 DBG3(" 3) %s", consumerd32_bin);
1732 if (stat(consumerd64_bin, &st) == 0) {
1733 DBG3("Found location #1");
1734 consumer_to_use = consumerd64_bin;
1735 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1736 DBG3("Found location #2");
1737 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1738 } else if (stat(consumerd32_bin, &st) == 0) {
1739 DBG3("Found location #3");
1740 consumer_to_use = consumerd32_bin;
1741 } else {
1742 DBG("Could not find any valid consumerd executable");
1743 break;
1744 }
1745 DBG("Using kernel consumer at: %s", consumer_to_use);
1746 execl(consumer_to_use,
1747 "lttng-consumerd", verbosity, "-k",
1748 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1749 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1750 NULL);
1751 break;
1752 case LTTNG_CONSUMER64_UST:
1753 {
1754 char *tmpnew = NULL;
1755
1756 if (consumerd64_libdir[0] != '\0') {
1757 char *tmp;
1758 size_t tmplen;
1759
1760 tmp = getenv("LD_LIBRARY_PATH");
1761 if (!tmp) {
1762 tmp = "";
1763 }
1764 tmplen = strlen("LD_LIBRARY_PATH=")
1765 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1766 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1767 if (!tmpnew) {
1768 ret = -ENOMEM;
1769 goto error;
1770 }
1771 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1772 strcat(tmpnew, consumerd64_libdir);
1773 if (tmp[0] != '\0') {
1774 strcat(tmpnew, ":");
1775 strcat(tmpnew, tmp);
1776 }
1777 ret = putenv(tmpnew);
1778 if (ret) {
1779 ret = -errno;
1780 goto error;
1781 }
1782 }
1783 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1784 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1785 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1786 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1787 NULL);
1788 if (consumerd64_libdir[0] != '\0') {
1789 free(tmpnew);
1790 }
1791 if (ret) {
1792 goto error;
1793 }
1794 break;
1795 }
1796 case LTTNG_CONSUMER32_UST:
1797 {
1798 char *tmpnew = NULL;
1799
1800 if (consumerd32_libdir[0] != '\0') {
1801 char *tmp;
1802 size_t tmplen;
1803
1804 tmp = getenv("LD_LIBRARY_PATH");
1805 if (!tmp) {
1806 tmp = "";
1807 }
1808 tmplen = strlen("LD_LIBRARY_PATH=")
1809 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1810 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1811 if (!tmpnew) {
1812 ret = -ENOMEM;
1813 goto error;
1814 }
1815 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1816 strcat(tmpnew, consumerd32_libdir);
1817 if (tmp[0] != '\0') {
1818 strcat(tmpnew, ":");
1819 strcat(tmpnew, tmp);
1820 }
1821 ret = putenv(tmpnew);
1822 if (ret) {
1823 ret = -errno;
1824 goto error;
1825 }
1826 }
1827 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1828 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1829 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1830 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1831 NULL);
1832 if (consumerd32_libdir[0] != '\0') {
1833 free(tmpnew);
1834 }
1835 if (ret) {
1836 goto error;
1837 }
1838 break;
1839 }
1840 default:
1841 PERROR("unknown consumer type");
1842 exit(EXIT_FAILURE);
1843 }
1844 if (errno != 0) {
1845 PERROR("kernel start consumer exec");
1846 }
1847 exit(EXIT_FAILURE);
1848 } else if (pid > 0) {
1849 ret = pid;
1850 } else {
1851 PERROR("start consumer fork");
1852 ret = -errno;
1853 }
1854 error:
1855 return ret;
1856 }
1857
1858 /*
1859 * Spawn the consumerd daemon and session daemon thread.
1860 */
1861 static int start_consumerd(struct consumer_data *consumer_data)
1862 {
1863 int ret;
1864
1865 /*
1866 * Set the listen() state on the socket since there is a possible race
1867 * between the exec() of the consumer daemon and this call if place in the
1868 * consumer thread. See bug #366 for more details.
1869 */
1870 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1871 if (ret < 0) {
1872 goto error;
1873 }
1874
1875 pthread_mutex_lock(&consumer_data->pid_mutex);
1876 if (consumer_data->pid != 0) {
1877 pthread_mutex_unlock(&consumer_data->pid_mutex);
1878 goto end;
1879 }
1880
1881 ret = spawn_consumerd(consumer_data);
1882 if (ret < 0) {
1883 ERR("Spawning consumerd failed");
1884 pthread_mutex_unlock(&consumer_data->pid_mutex);
1885 goto error;
1886 }
1887
1888 /* Setting up the consumer_data pid */
1889 consumer_data->pid = ret;
1890 DBG2("Consumer pid %d", consumer_data->pid);
1891 pthread_mutex_unlock(&consumer_data->pid_mutex);
1892
1893 DBG2("Spawning consumer control thread");
1894 ret = spawn_consumer_thread(consumer_data);
1895 if (ret < 0) {
1896 ERR("Fatal error spawning consumer control thread");
1897 goto error;
1898 }
1899
1900 end:
1901 return 0;
1902
1903 error:
1904 /* Cleanup already created socket on error. */
1905 if (consumer_data->err_sock >= 0) {
1906 int err;
1907
1908 err = close(consumer_data->err_sock);
1909 if (err < 0) {
1910 PERROR("close consumer data error socket");
1911 }
1912 }
1913 return ret;
1914 }
1915
1916 /*
1917 * Compute health status of each consumer. If one of them is zero (bad
1918 * state), we return 0.
1919 */
1920 static int check_consumer_health(void)
1921 {
1922 int ret;
1923
1924 ret = health_check_state(HEALTH_TYPE_CONSUMER);
1925
1926 DBG3("Health consumer check %d", ret);
1927
1928 return ret;
1929 }
1930
1931 /*
1932 * Setup necessary data for kernel tracer action.
1933 */
1934 static int init_kernel_tracer(void)
1935 {
1936 int ret;
1937
1938 /* Modprobe lttng kernel modules */
1939 ret = modprobe_lttng_control();
1940 if (ret < 0) {
1941 goto error;
1942 }
1943
1944 /* Open debugfs lttng */
1945 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1946 if (kernel_tracer_fd < 0) {
1947 DBG("Failed to open %s", module_proc_lttng);
1948 ret = -1;
1949 goto error_open;
1950 }
1951
1952 /* Validate kernel version */
1953 ret = kernel_validate_version(kernel_tracer_fd);
1954 if (ret < 0) {
1955 goto error_version;
1956 }
1957
1958 ret = modprobe_lttng_data();
1959 if (ret < 0) {
1960 goto error_modules;
1961 }
1962
1963 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1964 return 0;
1965
1966 error_version:
1967 modprobe_remove_lttng_control();
1968 ret = close(kernel_tracer_fd);
1969 if (ret) {
1970 PERROR("close");
1971 }
1972 kernel_tracer_fd = -1;
1973 return LTTNG_ERR_KERN_VERSION;
1974
1975 error_modules:
1976 ret = close(kernel_tracer_fd);
1977 if (ret) {
1978 PERROR("close");
1979 }
1980
1981 error_open:
1982 modprobe_remove_lttng_control();
1983
1984 error:
1985 WARN("No kernel tracer available");
1986 kernel_tracer_fd = -1;
1987 if (!is_root) {
1988 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1989 } else {
1990 return LTTNG_ERR_KERN_NA;
1991 }
1992 }
1993
1994
1995 /*
1996 * Copy consumer output from the tracing session to the domain session. The
1997 * function also applies the right modification on a per domain basis for the
1998 * trace files destination directory.
1999 */
2000 static int copy_session_consumer(int domain, struct ltt_session *session)
2001 {
2002 int ret;
2003 const char *dir_name;
2004 struct consumer_output *consumer;
2005
2006 assert(session);
2007 assert(session->consumer);
2008
2009 switch (domain) {
2010 case LTTNG_DOMAIN_KERNEL:
2011 DBG3("Copying tracing session consumer output in kernel session");
2012 /*
2013 * XXX: We should audit the session creation and what this function
2014 * does "extra" in order to avoid a destroy since this function is used
2015 * in the domain session creation (kernel and ust) only. Same for UST
2016 * domain.
2017 */
2018 if (session->kernel_session->consumer) {
2019 consumer_destroy_output(session->kernel_session->consumer);
2020 }
2021 session->kernel_session->consumer =
2022 consumer_copy_output(session->consumer);
2023 /* Ease our life a bit for the next part */
2024 consumer = session->kernel_session->consumer;
2025 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2026 break;
2027 case LTTNG_DOMAIN_UST:
2028 DBG3("Copying tracing session consumer output in UST session");
2029 if (session->ust_session->consumer) {
2030 consumer_destroy_output(session->ust_session->consumer);
2031 }
2032 session->ust_session->consumer =
2033 consumer_copy_output(session->consumer);
2034 /* Ease our life a bit for the next part */
2035 consumer = session->ust_session->consumer;
2036 dir_name = DEFAULT_UST_TRACE_DIR;
2037 break;
2038 default:
2039 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2040 goto error;
2041 }
2042
2043 /* Append correct directory to subdir */
2044 strncat(consumer->subdir, dir_name,
2045 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2046 DBG3("Copy session consumer subdir %s", consumer->subdir);
2047
2048 ret = LTTNG_OK;
2049
2050 error:
2051 return ret;
2052 }
2053
2054 /*
2055 * Create an UST session and add it to the session ust list.
2056 */
2057 static int create_ust_session(struct ltt_session *session,
2058 struct lttng_domain *domain)
2059 {
2060 int ret;
2061 struct ltt_ust_session *lus = NULL;
2062
2063 assert(session);
2064 assert(domain);
2065 assert(session->consumer);
2066
2067 switch (domain->type) {
2068 case LTTNG_DOMAIN_UST:
2069 break;
2070 default:
2071 ERR("Unknown UST domain on create session %d", domain->type);
2072 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2073 goto error;
2074 }
2075
2076 DBG("Creating UST session");
2077
2078 lus = trace_ust_create_session(session->path, session->id, domain);
2079 if (lus == NULL) {
2080 ret = LTTNG_ERR_UST_SESS_FAIL;
2081 goto error;
2082 }
2083
2084 lus->uid = session->uid;
2085 lus->gid = session->gid;
2086 session->ust_session = lus;
2087
2088 /* Copy session output to the newly created UST session */
2089 ret = copy_session_consumer(domain->type, session);
2090 if (ret != LTTNG_OK) {
2091 goto error;
2092 }
2093
2094 return LTTNG_OK;
2095
2096 error:
2097 free(lus);
2098 session->ust_session = NULL;
2099 return ret;
2100 }
2101
2102 /*
2103 * Create a kernel tracer session then create the default channel.
2104 */
2105 static int create_kernel_session(struct ltt_session *session)
2106 {
2107 int ret;
2108
2109 DBG("Creating kernel session");
2110
2111 ret = kernel_create_session(session, kernel_tracer_fd);
2112 if (ret < 0) {
2113 ret = LTTNG_ERR_KERN_SESS_FAIL;
2114 goto error;
2115 }
2116
2117 /* Code flow safety */
2118 assert(session->kernel_session);
2119
2120 /* Copy session output to the newly created Kernel session */
2121 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2122 if (ret != LTTNG_OK) {
2123 goto error;
2124 }
2125
2126 /* Create directory(ies) on local filesystem. */
2127 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2128 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2129 ret = run_as_mkdir_recursive(
2130 session->kernel_session->consumer->dst.trace_path,
2131 S_IRWXU | S_IRWXG, session->uid, session->gid);
2132 if (ret < 0) {
2133 if (ret != -EEXIST) {
2134 ERR("Trace directory creation error");
2135 goto error;
2136 }
2137 }
2138 }
2139
2140 session->kernel_session->uid = session->uid;
2141 session->kernel_session->gid = session->gid;
2142
2143 return LTTNG_OK;
2144
2145 error:
2146 trace_kernel_destroy_session(session->kernel_session);
2147 session->kernel_session = NULL;
2148 return ret;
2149 }
2150
2151 /*
2152 * Count number of session permitted by uid/gid.
2153 */
2154 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2155 {
2156 unsigned int i = 0;
2157 struct ltt_session *session;
2158
2159 DBG("Counting number of available session for UID %d GID %d",
2160 uid, gid);
2161 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2162 /*
2163 * Only list the sessions the user can control.
2164 */
2165 if (!session_access_ok(session, uid, gid)) {
2166 continue;
2167 }
2168 i++;
2169 }
2170 return i;
2171 }
2172
2173 /*
2174 * Process the command requested by the lttng client within the command
2175 * context structure. This function make sure that the return structure (llm)
2176 * is set and ready for transmission before returning.
2177 *
2178 * Return any error encountered or 0 for success.
2179 *
2180 * "sock" is only used for special-case var. len data.
2181 */
2182 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2183 int *sock_error)
2184 {
2185 int ret = LTTNG_OK;
2186 int need_tracing_session = 1;
2187 int need_domain;
2188
2189 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2190
2191 *sock_error = 0;
2192
2193 switch (cmd_ctx->lsm->cmd_type) {
2194 case LTTNG_CREATE_SESSION:
2195 case LTTNG_DESTROY_SESSION:
2196 case LTTNG_LIST_SESSIONS:
2197 case LTTNG_LIST_DOMAINS:
2198 case LTTNG_START_TRACE:
2199 case LTTNG_STOP_TRACE:
2200 case LTTNG_DATA_PENDING:
2201 need_domain = 0;
2202 break;
2203 default:
2204 need_domain = 1;
2205 }
2206
2207 if (opt_no_kernel && need_domain
2208 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2209 if (!is_root) {
2210 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2211 } else {
2212 ret = LTTNG_ERR_KERN_NA;
2213 }
2214 goto error;
2215 }
2216
2217 /* Deny register consumer if we already have a spawned consumer. */
2218 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2219 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2220 if (kconsumer_data.pid > 0) {
2221 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2222 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2223 goto error;
2224 }
2225 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2226 }
2227
2228 /*
2229 * Check for command that don't needs to allocate a returned payload. We do
2230 * this here so we don't have to make the call for no payload at each
2231 * command.
2232 */
2233 switch(cmd_ctx->lsm->cmd_type) {
2234 case LTTNG_LIST_SESSIONS:
2235 case LTTNG_LIST_TRACEPOINTS:
2236 case LTTNG_LIST_TRACEPOINT_FIELDS:
2237 case LTTNG_LIST_DOMAINS:
2238 case LTTNG_LIST_CHANNELS:
2239 case LTTNG_LIST_EVENTS:
2240 break;
2241 default:
2242 /* Setup lttng message with no payload */
2243 ret = setup_lttng_msg(cmd_ctx, 0);
2244 if (ret < 0) {
2245 /* This label does not try to unlock the session */
2246 goto init_setup_error;
2247 }
2248 }
2249
2250 /* Commands that DO NOT need a session. */
2251 switch (cmd_ctx->lsm->cmd_type) {
2252 case LTTNG_CREATE_SESSION:
2253 case LTTNG_CALIBRATE:
2254 case LTTNG_LIST_SESSIONS:
2255 case LTTNG_LIST_TRACEPOINTS:
2256 case LTTNG_LIST_TRACEPOINT_FIELDS:
2257 need_tracing_session = 0;
2258 break;
2259 default:
2260 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2261 /*
2262 * We keep the session list lock across _all_ commands
2263 * for now, because the per-session lock does not
2264 * handle teardown properly.
2265 */
2266 session_lock_list();
2267 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2268 if (cmd_ctx->session == NULL) {
2269 if (cmd_ctx->lsm->session.name != NULL) {
2270 ret = LTTNG_ERR_SESS_NOT_FOUND;
2271 } else {
2272 /* If no session name specified */
2273 ret = LTTNG_ERR_SELECT_SESS;
2274 }
2275 goto error;
2276 } else {
2277 /* Acquire lock for the session */
2278 session_lock(cmd_ctx->session);
2279 }
2280 break;
2281 }
2282
2283 if (!need_domain) {
2284 goto skip_domain;
2285 }
2286
2287 /*
2288 * Check domain type for specific "pre-action".
2289 */
2290 switch (cmd_ctx->lsm->domain.type) {
2291 case LTTNG_DOMAIN_KERNEL:
2292 if (!is_root) {
2293 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2294 goto error;
2295 }
2296
2297 /* Kernel tracer check */
2298 if (kernel_tracer_fd == -1) {
2299 /* Basically, load kernel tracer modules */
2300 ret = init_kernel_tracer();
2301 if (ret != 0) {
2302 goto error;
2303 }
2304 }
2305
2306 /* Consumer is in an ERROR state. Report back to client */
2307 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2308 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2309 goto error;
2310 }
2311
2312 /* Need a session for kernel command */
2313 if (need_tracing_session) {
2314 if (cmd_ctx->session->kernel_session == NULL) {
2315 ret = create_kernel_session(cmd_ctx->session);
2316 if (ret < 0) {
2317 ret = LTTNG_ERR_KERN_SESS_FAIL;
2318 goto error;
2319 }
2320 }
2321
2322 /* Start the kernel consumer daemon */
2323 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2324 if (kconsumer_data.pid == 0 &&
2325 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2326 cmd_ctx->session->start_consumer) {
2327 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2328 ret = start_consumerd(&kconsumer_data);
2329 if (ret < 0) {
2330 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2331 goto error;
2332 }
2333 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2334 } else {
2335 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2336 }
2337
2338 /*
2339 * The consumer was just spawned so we need to add the socket to
2340 * the consumer output of the session if exist.
2341 */
2342 ret = consumer_create_socket(&kconsumer_data,
2343 cmd_ctx->session->kernel_session->consumer);
2344 if (ret < 0) {
2345 goto error;
2346 }
2347 }
2348
2349 break;
2350 case LTTNG_DOMAIN_UST:
2351 {
2352 /* Consumer is in an ERROR state. Report back to client */
2353 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2354 ret = LTTNG_ERR_NO_USTCONSUMERD;
2355 goto error;
2356 }
2357
2358 if (need_tracing_session) {
2359 /* Create UST session if none exist. */
2360 if (cmd_ctx->session->ust_session == NULL) {
2361 ret = create_ust_session(cmd_ctx->session,
2362 &cmd_ctx->lsm->domain);
2363 if (ret != LTTNG_OK) {
2364 goto error;
2365 }
2366 }
2367
2368 /* Start the UST consumer daemons */
2369 /* 64-bit */
2370 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2371 if (consumerd64_bin[0] != '\0' &&
2372 ustconsumer64_data.pid == 0 &&
2373 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2374 cmd_ctx->session->start_consumer) {
2375 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2376 ret = start_consumerd(&ustconsumer64_data);
2377 if (ret < 0) {
2378 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2379 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2380 goto error;
2381 }
2382
2383 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2384 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2385 } else {
2386 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2387 }
2388
2389 /*
2390 * Setup socket for consumer 64 bit. No need for atomic access
2391 * since it was set above and can ONLY be set in this thread.
2392 */
2393 ret = consumer_create_socket(&ustconsumer64_data,
2394 cmd_ctx->session->ust_session->consumer);
2395 if (ret < 0) {
2396 goto error;
2397 }
2398
2399 /* 32-bit */
2400 if (consumerd32_bin[0] != '\0' &&
2401 ustconsumer32_data.pid == 0 &&
2402 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2403 cmd_ctx->session->start_consumer) {
2404 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2405 ret = start_consumerd(&ustconsumer32_data);
2406 if (ret < 0) {
2407 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2408 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2409 goto error;
2410 }
2411
2412 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2413 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2414 } else {
2415 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2416 }
2417
2418 /*
2419 * Setup socket for consumer 64 bit. No need for atomic access
2420 * since it was set above and can ONLY be set in this thread.
2421 */
2422 ret = consumer_create_socket(&ustconsumer32_data,
2423 cmd_ctx->session->ust_session->consumer);
2424 if (ret < 0) {
2425 goto error;
2426 }
2427 }
2428 break;
2429 }
2430 default:
2431 break;
2432 }
2433 skip_domain:
2434
2435 /* Validate consumer daemon state when start/stop trace command */
2436 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2437 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2438 switch (cmd_ctx->lsm->domain.type) {
2439 case LTTNG_DOMAIN_UST:
2440 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2441 ret = LTTNG_ERR_NO_USTCONSUMERD;
2442 goto error;
2443 }
2444 break;
2445 case LTTNG_DOMAIN_KERNEL:
2446 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2447 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2448 goto error;
2449 }
2450 break;
2451 }
2452 }
2453
2454 /*
2455 * Check that the UID or GID match that of the tracing session.
2456 * The root user can interact with all sessions.
2457 */
2458 if (need_tracing_session) {
2459 if (!session_access_ok(cmd_ctx->session,
2460 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2461 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2462 ret = LTTNG_ERR_EPERM;
2463 goto error;
2464 }
2465 }
2466
2467 /* Process by command type */
2468 switch (cmd_ctx->lsm->cmd_type) {
2469 case LTTNG_ADD_CONTEXT:
2470 {
2471 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2472 cmd_ctx->lsm->u.context.channel_name,
2473 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2474 break;
2475 }
2476 case LTTNG_DISABLE_CHANNEL:
2477 {
2478 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2479 cmd_ctx->lsm->u.disable.channel_name);
2480 break;
2481 }
2482 case LTTNG_DISABLE_EVENT:
2483 {
2484 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2485 cmd_ctx->lsm->u.disable.channel_name,
2486 cmd_ctx->lsm->u.disable.name);
2487 break;
2488 }
2489 case LTTNG_DISABLE_ALL_EVENT:
2490 {
2491 DBG("Disabling all events");
2492
2493 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2494 cmd_ctx->lsm->u.disable.channel_name);
2495 break;
2496 }
2497 case LTTNG_DISABLE_CONSUMER:
2498 {
2499 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2500 break;
2501 }
2502 case LTTNG_ENABLE_CHANNEL:
2503 {
2504 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2505 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2506 break;
2507 }
2508 case LTTNG_ENABLE_CONSUMER:
2509 {
2510 /*
2511 * XXX: 0 means that this URI should be applied on the session. Should
2512 * be a DOMAIN enuam.
2513 */
2514 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2515 if (ret != LTTNG_OK) {
2516 goto error;
2517 }
2518
2519 if (cmd_ctx->lsm->domain.type == 0) {
2520 /* Add the URI for the UST session if a consumer is present. */
2521 if (cmd_ctx->session->ust_session &&
2522 cmd_ctx->session->ust_session->consumer) {
2523 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2524 } else if (cmd_ctx->session->kernel_session &&
2525 cmd_ctx->session->kernel_session->consumer) {
2526 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2527 cmd_ctx->session);
2528 }
2529 }
2530 break;
2531 }
2532 case LTTNG_ENABLE_EVENT:
2533 {
2534 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2535 cmd_ctx->lsm->u.enable.channel_name,
2536 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2537 break;
2538 }
2539 case LTTNG_ENABLE_ALL_EVENT:
2540 {
2541 DBG("Enabling all events");
2542
2543 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2544 cmd_ctx->lsm->u.enable.channel_name,
2545 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2546 break;
2547 }
2548 case LTTNG_LIST_TRACEPOINTS:
2549 {
2550 struct lttng_event *events;
2551 ssize_t nb_events;
2552
2553 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2554 if (nb_events < 0) {
2555 /* Return value is a negative lttng_error_code. */
2556 ret = -nb_events;
2557 goto error;
2558 }
2559
2560 /*
2561 * Setup lttng message with payload size set to the event list size in
2562 * bytes and then copy list into the llm payload.
2563 */
2564 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2565 if (ret < 0) {
2566 free(events);
2567 goto setup_error;
2568 }
2569
2570 /* Copy event list into message payload */
2571 memcpy(cmd_ctx->llm->payload, events,
2572 sizeof(struct lttng_event) * nb_events);
2573
2574 free(events);
2575
2576 ret = LTTNG_OK;
2577 break;
2578 }
2579 case LTTNG_LIST_TRACEPOINT_FIELDS:
2580 {
2581 struct lttng_event_field *fields;
2582 ssize_t nb_fields;
2583
2584 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2585 &fields);
2586 if (nb_fields < 0) {
2587 /* Return value is a negative lttng_error_code. */
2588 ret = -nb_fields;
2589 goto error;
2590 }
2591
2592 /*
2593 * Setup lttng message with payload size set to the event list size in
2594 * bytes and then copy list into the llm payload.
2595 */
2596 ret = setup_lttng_msg(cmd_ctx,
2597 sizeof(struct lttng_event_field) * nb_fields);
2598 if (ret < 0) {
2599 free(fields);
2600 goto setup_error;
2601 }
2602
2603 /* Copy event list into message payload */
2604 memcpy(cmd_ctx->llm->payload, fields,
2605 sizeof(struct lttng_event_field) * nb_fields);
2606
2607 free(fields);
2608
2609 ret = LTTNG_OK;
2610 break;
2611 }
2612 case LTTNG_SET_CONSUMER_URI:
2613 {
2614 size_t nb_uri, len;
2615 struct lttng_uri *uris;
2616
2617 nb_uri = cmd_ctx->lsm->u.uri.size;
2618 len = nb_uri * sizeof(struct lttng_uri);
2619
2620 if (nb_uri == 0) {
2621 ret = LTTNG_ERR_INVALID;
2622 goto error;
2623 }
2624
2625 uris = zmalloc(len);
2626 if (uris == NULL) {
2627 ret = LTTNG_ERR_FATAL;
2628 goto error;
2629 }
2630
2631 /* Receive variable len data */
2632 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2633 ret = lttcomm_recv_unix_sock(sock, uris, len);
2634 if (ret <= 0) {
2635 DBG("No URIs received from client... continuing");
2636 *sock_error = 1;
2637 ret = LTTNG_ERR_SESSION_FAIL;
2638 free(uris);
2639 goto error;
2640 }
2641
2642 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2643 nb_uri, uris);
2644 if (ret != LTTNG_OK) {
2645 free(uris);
2646 goto error;
2647 }
2648
2649 /*
2650 * XXX: 0 means that this URI should be applied on the session. Should
2651 * be a DOMAIN enuam.
2652 */
2653 if (cmd_ctx->lsm->domain.type == 0) {
2654 /* Add the URI for the UST session if a consumer is present. */
2655 if (cmd_ctx->session->ust_session &&
2656 cmd_ctx->session->ust_session->consumer) {
2657 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2658 nb_uri, uris);
2659 } else if (cmd_ctx->session->kernel_session &&
2660 cmd_ctx->session->kernel_session->consumer) {
2661 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2662 cmd_ctx->session, nb_uri, uris);
2663 }
2664 }
2665
2666 free(uris);
2667
2668 break;
2669 }
2670 case LTTNG_START_TRACE:
2671 {
2672 ret = cmd_start_trace(cmd_ctx->session);
2673 break;
2674 }
2675 case LTTNG_STOP_TRACE:
2676 {
2677 ret = cmd_stop_trace(cmd_ctx->session);
2678 break;
2679 }
2680 case LTTNG_CREATE_SESSION:
2681 {
2682 size_t nb_uri, len;
2683 struct lttng_uri *uris = NULL;
2684
2685 nb_uri = cmd_ctx->lsm->u.uri.size;
2686 len = nb_uri * sizeof(struct lttng_uri);
2687
2688 if (nb_uri > 0) {
2689 uris = zmalloc(len);
2690 if (uris == NULL) {
2691 ret = LTTNG_ERR_FATAL;
2692 goto error;
2693 }
2694
2695 /* Receive variable len data */
2696 DBG("Waiting for %zu URIs from client ...", nb_uri);
2697 ret = lttcomm_recv_unix_sock(sock, uris, len);
2698 if (ret <= 0) {
2699 DBG("No URIs received from client... continuing");
2700 *sock_error = 1;
2701 ret = LTTNG_ERR_SESSION_FAIL;
2702 free(uris);
2703 goto error;
2704 }
2705
2706 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2707 DBG("Creating session with ONE network URI is a bad call");
2708 ret = LTTNG_ERR_SESSION_FAIL;
2709 free(uris);
2710 goto error;
2711 }
2712 }
2713
2714 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2715 &cmd_ctx->creds);
2716
2717 free(uris);
2718
2719 break;
2720 }
2721 case LTTNG_DESTROY_SESSION:
2722 {
2723 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2724
2725 /* Set session to NULL so we do not unlock it after free. */
2726 cmd_ctx->session = NULL;
2727 break;
2728 }
2729 case LTTNG_LIST_DOMAINS:
2730 {
2731 ssize_t nb_dom;
2732 struct lttng_domain *domains;
2733
2734 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2735 if (nb_dom < 0) {
2736 /* Return value is a negative lttng_error_code. */
2737 ret = -nb_dom;
2738 goto error;
2739 }
2740
2741 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2742 if (ret < 0) {
2743 goto setup_error;
2744 }
2745
2746 /* Copy event list into message payload */
2747 memcpy(cmd_ctx->llm->payload, domains,
2748 nb_dom * sizeof(struct lttng_domain));
2749
2750 free(domains);
2751
2752 ret = LTTNG_OK;
2753 break;
2754 }
2755 case LTTNG_LIST_CHANNELS:
2756 {
2757 int nb_chan;
2758 struct lttng_channel *channels;
2759
2760 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2761 cmd_ctx->session, &channels);
2762 if (nb_chan < 0) {
2763 /* Return value is a negative lttng_error_code. */
2764 ret = -nb_chan;
2765 goto error;
2766 }
2767
2768 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2769 if (ret < 0) {
2770 goto setup_error;
2771 }
2772
2773 /* Copy event list into message payload */
2774 memcpy(cmd_ctx->llm->payload, channels,
2775 nb_chan * sizeof(struct lttng_channel));
2776
2777 free(channels);
2778
2779 ret = LTTNG_OK;
2780 break;
2781 }
2782 case LTTNG_LIST_EVENTS:
2783 {
2784 ssize_t nb_event;
2785 struct lttng_event *events = NULL;
2786
2787 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2788 cmd_ctx->lsm->u.list.channel_name, &events);
2789 if (nb_event < 0) {
2790 /* Return value is a negative lttng_error_code. */
2791 ret = -nb_event;
2792 goto error;
2793 }
2794
2795 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2796 if (ret < 0) {
2797 goto setup_error;
2798 }
2799
2800 /* Copy event list into message payload */
2801 memcpy(cmd_ctx->llm->payload, events,
2802 nb_event * sizeof(struct lttng_event));
2803
2804 free(events);
2805
2806 ret = LTTNG_OK;
2807 break;
2808 }
2809 case LTTNG_LIST_SESSIONS:
2810 {
2811 unsigned int nr_sessions;
2812
2813 session_lock_list();
2814 nr_sessions = lttng_sessions_count(
2815 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2816 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2817
2818 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2819 if (ret < 0) {
2820 session_unlock_list();
2821 goto setup_error;
2822 }
2823
2824 /* Filled the session array */
2825 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2826 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2827 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2828
2829 session_unlock_list();
2830
2831 ret = LTTNG_OK;
2832 break;
2833 }
2834 case LTTNG_CALIBRATE:
2835 {
2836 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2837 &cmd_ctx->lsm->u.calibrate);
2838 break;
2839 }
2840 case LTTNG_REGISTER_CONSUMER:
2841 {
2842 struct consumer_data *cdata;
2843
2844 switch (cmd_ctx->lsm->domain.type) {
2845 case LTTNG_DOMAIN_KERNEL:
2846 cdata = &kconsumer_data;
2847 break;
2848 default:
2849 ret = LTTNG_ERR_UND;
2850 goto error;
2851 }
2852
2853 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2854 cmd_ctx->lsm->u.reg.path, cdata);
2855 break;
2856 }
2857 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2858 {
2859 struct lttng_filter_bytecode *bytecode;
2860
2861 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2862 ret = LTTNG_ERR_FILTER_INVAL;
2863 goto error;
2864 }
2865 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2866 ret = LTTNG_ERR_FILTER_INVAL;
2867 goto error;
2868 }
2869 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2870 if (!bytecode) {
2871 ret = LTTNG_ERR_FILTER_NOMEM;
2872 goto error;
2873 }
2874 /* Receive var. len. data */
2875 DBG("Receiving var len data from client ...");
2876 ret = lttcomm_recv_unix_sock(sock, bytecode,
2877 cmd_ctx->lsm->u.enable.bytecode_len);
2878 if (ret <= 0) {
2879 DBG("Nothing recv() from client var len data... continuing");
2880 *sock_error = 1;
2881 ret = LTTNG_ERR_FILTER_INVAL;
2882 goto error;
2883 }
2884
2885 if (bytecode->len + sizeof(*bytecode)
2886 != cmd_ctx->lsm->u.enable.bytecode_len) {
2887 free(bytecode);
2888 ret = LTTNG_ERR_FILTER_INVAL;
2889 goto error;
2890 }
2891
2892 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2893 cmd_ctx->lsm->u.enable.channel_name,
2894 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2895 break;
2896 }
2897 case LTTNG_DATA_PENDING:
2898 {
2899 ret = cmd_data_pending(cmd_ctx->session);
2900 break;
2901 }
2902 default:
2903 ret = LTTNG_ERR_UND;
2904 break;
2905 }
2906
2907 error:
2908 if (cmd_ctx->llm == NULL) {
2909 DBG("Missing llm structure. Allocating one.");
2910 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2911 goto setup_error;
2912 }
2913 }
2914 /* Set return code */
2915 cmd_ctx->llm->ret_code = ret;
2916 setup_error:
2917 if (cmd_ctx->session) {
2918 session_unlock(cmd_ctx->session);
2919 }
2920 if (need_tracing_session) {
2921 session_unlock_list();
2922 }
2923 init_setup_error:
2924 return ret;
2925 }
2926
2927 /*
2928 * Thread managing health check socket.
2929 */
2930 static void *thread_manage_health(void *data)
2931 {
2932 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2933 uint32_t revents, nb_fd;
2934 struct lttng_poll_event events;
2935 struct lttcomm_health_msg msg;
2936 struct lttcomm_health_data reply;
2937
2938 DBG("[thread] Manage health check started");
2939
2940 rcu_register_thread();
2941
2942 /* Create unix socket */
2943 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2944 if (sock < 0) {
2945 ERR("Unable to create health check Unix socket");
2946 ret = -1;
2947 goto error;
2948 }
2949
2950 /*
2951 * Set the CLOEXEC flag. Return code is useless because either way, the
2952 * show must go on.
2953 */
2954 (void) utils_set_fd_cloexec(sock);
2955
2956 ret = lttcomm_listen_unix_sock(sock);
2957 if (ret < 0) {
2958 goto error;
2959 }
2960
2961 /*
2962 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2963 * more will be added to this poll set.
2964 */
2965 ret = create_thread_poll_set(&events, 2);
2966 if (ret < 0) {
2967 goto error;
2968 }
2969
2970 /* Add the application registration socket */
2971 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2972 if (ret < 0) {
2973 goto error;
2974 }
2975
2976 while (1) {
2977 DBG("Health check ready");
2978
2979 /* Inifinite blocking call, waiting for transmission */
2980 restart:
2981 ret = lttng_poll_wait(&events, -1);
2982 if (ret < 0) {
2983 /*
2984 * Restart interrupted system call.
2985 */
2986 if (errno == EINTR) {
2987 goto restart;
2988 }
2989 goto error;
2990 }
2991
2992 nb_fd = ret;
2993
2994 for (i = 0; i < nb_fd; i++) {
2995 /* Fetch once the poll data */
2996 revents = LTTNG_POLL_GETEV(&events, i);
2997 pollfd = LTTNG_POLL_GETFD(&events, i);
2998
2999 /* Thread quit pipe has been closed. Killing thread. */
3000 ret = check_thread_quit_pipe(pollfd, revents);
3001 if (ret) {
3002 err = 0;
3003 goto exit;
3004 }
3005
3006 /* Event on the registration socket */
3007 if (pollfd == sock) {
3008 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3009 ERR("Health socket poll error");
3010 goto error;
3011 }
3012 }
3013 }
3014
3015 new_sock = lttcomm_accept_unix_sock(sock);
3016 if (new_sock < 0) {
3017 goto error;
3018 }
3019
3020 /*
3021 * Set the CLOEXEC flag. Return code is useless because either way, the
3022 * show must go on.
3023 */
3024 (void) utils_set_fd_cloexec(new_sock);
3025
3026 DBG("Receiving data from client for health...");
3027 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3028 if (ret <= 0) {
3029 DBG("Nothing recv() from client... continuing");
3030 ret = close(new_sock);
3031 if (ret) {
3032 PERROR("close");
3033 }
3034 new_sock = -1;
3035 continue;
3036 }
3037
3038 rcu_thread_online();
3039
3040 switch (msg.component) {
3041 case LTTNG_HEALTH_CMD:
3042 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3043 break;
3044 case LTTNG_HEALTH_APP_MANAGE:
3045 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3046 break;
3047 case LTTNG_HEALTH_APP_REG:
3048 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3049 break;
3050 case LTTNG_HEALTH_KERNEL:
3051 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3052 break;
3053 case LTTNG_HEALTH_CONSUMER:
3054 reply.ret_code = check_consumer_health();
3055 break;
3056 case LTTNG_HEALTH_ALL:
3057 reply.ret_code =
3058 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3059 health_check_state(HEALTH_TYPE_APP_REG) &&
3060 health_check_state(HEALTH_TYPE_CMD) &&
3061 health_check_state(HEALTH_TYPE_KERNEL) &&
3062 check_consumer_health();
3063 break;
3064 default:
3065 reply.ret_code = LTTNG_ERR_UND;
3066 break;
3067 }
3068
3069 /*
3070 * Flip ret value since 0 is a success and 1 indicates a bad health for
3071 * the client where in the sessiond it is the opposite. Again, this is
3072 * just to make things easier for us poor developer which enjoy a lot
3073 * lazyness.
3074 */
3075 if (reply.ret_code == 0 || reply.ret_code == 1) {
3076 reply.ret_code = !reply.ret_code;
3077 }
3078
3079 DBG2("Health check return value %d", reply.ret_code);
3080
3081 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3082 if (ret < 0) {
3083 ERR("Failed to send health data back to client");
3084 }
3085
3086 /* End of transmission */
3087 ret = close(new_sock);
3088 if (ret) {
3089 PERROR("close");
3090 }
3091 new_sock = -1;
3092 }
3093
3094 exit:
3095 error:
3096 if (err) {
3097 ERR("Health error occurred in %s", __func__);
3098 }
3099 DBG("Health check thread dying");
3100 unlink(health_unix_sock_path);
3101 if (sock >= 0) {
3102 ret = close(sock);
3103 if (ret) {
3104 PERROR("close");
3105 }
3106 }
3107 if (new_sock >= 0) {
3108 ret = close(new_sock);
3109 if (ret) {
3110 PERROR("close");
3111 }
3112 }
3113
3114 lttng_poll_clean(&events);
3115
3116 rcu_unregister_thread();
3117 return NULL;
3118 }
3119
3120 /*
3121 * This thread manage all clients request using the unix client socket for
3122 * communication.
3123 */
3124 static void *thread_manage_clients(void *data)
3125 {
3126 int sock = -1, ret, i, pollfd, err = -1;
3127 int sock_error;
3128 uint32_t revents, nb_fd;
3129 struct command_ctx *cmd_ctx = NULL;
3130 struct lttng_poll_event events;
3131
3132 DBG("[thread] Manage client started");
3133
3134 rcu_register_thread();
3135
3136 health_register(HEALTH_TYPE_CMD);
3137
3138 if (testpoint(thread_manage_clients)) {
3139 goto error_testpoint;
3140 }
3141
3142 health_code_update();
3143
3144 ret = lttcomm_listen_unix_sock(client_sock);
3145 if (ret < 0) {
3146 goto error_listen;
3147 }
3148
3149 /*
3150 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3151 * more will be added to this poll set.
3152 */
3153 ret = create_thread_poll_set(&events, 2);
3154 if (ret < 0) {
3155 goto error_create_poll;
3156 }
3157
3158 /* Add the application registration socket */
3159 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3160 if (ret < 0) {
3161 goto error;
3162 }
3163
3164 /*
3165 * Notify parent pid that we are ready to accept command for client side.
3166 */
3167 if (opt_sig_parent) {
3168 kill(ppid, SIGUSR1);
3169 }
3170
3171 if (testpoint(thread_manage_clients_before_loop)) {
3172 goto error;
3173 }
3174
3175 health_code_update();
3176
3177 while (1) {
3178 DBG("Accepting client command ...");
3179
3180 /* Inifinite blocking call, waiting for transmission */
3181 restart:
3182 health_poll_entry();
3183 ret = lttng_poll_wait(&events, -1);
3184 health_poll_exit();
3185 if (ret < 0) {
3186 /*
3187 * Restart interrupted system call.
3188 */
3189 if (errno == EINTR) {
3190 goto restart;
3191 }
3192 goto error;
3193 }
3194
3195 nb_fd = ret;
3196
3197 for (i = 0; i < nb_fd; i++) {
3198 /* Fetch once the poll data */
3199 revents = LTTNG_POLL_GETEV(&events, i);
3200 pollfd = LTTNG_POLL_GETFD(&events, i);
3201
3202 health_code_update();
3203
3204 /* Thread quit pipe has been closed. Killing thread. */
3205 ret = check_thread_quit_pipe(pollfd, revents);
3206 if (ret) {
3207 err = 0;
3208 goto exit;
3209 }
3210
3211 /* Event on the registration socket */
3212 if (pollfd == client_sock) {
3213 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3214 ERR("Client socket poll error");
3215 goto error;
3216 }
3217 }
3218 }
3219
3220 DBG("Wait for client response");
3221
3222 health_code_update();
3223
3224 sock = lttcomm_accept_unix_sock(client_sock);
3225 if (sock < 0) {
3226 goto error;
3227 }
3228
3229 /*
3230 * Set the CLOEXEC flag. Return code is useless because either way, the
3231 * show must go on.
3232 */
3233 (void) utils_set_fd_cloexec(sock);
3234
3235 /* Set socket option for credentials retrieval */
3236 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3237 if (ret < 0) {
3238 goto error;
3239 }
3240
3241 /* Allocate context command to process the client request */
3242 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3243 if (cmd_ctx == NULL) {
3244 PERROR("zmalloc cmd_ctx");
3245 goto error;
3246 }
3247
3248 /* Allocate data buffer for reception */
3249 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3250 if (cmd_ctx->lsm == NULL) {
3251 PERROR("zmalloc cmd_ctx->lsm");
3252 goto error;
3253 }
3254
3255 cmd_ctx->llm = NULL;
3256 cmd_ctx->session = NULL;
3257
3258 health_code_update();
3259
3260 /*
3261 * Data is received from the lttng client. The struct
3262 * lttcomm_session_msg (lsm) contains the command and data request of
3263 * the client.
3264 */
3265 DBG("Receiving data from client ...");
3266 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3267 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3268 if (ret <= 0) {
3269 DBG("Nothing recv() from client... continuing");
3270 ret = close(sock);
3271 if (ret) {
3272 PERROR("close");
3273 }
3274 sock = -1;
3275 clean_command_ctx(&cmd_ctx);
3276 continue;
3277 }
3278
3279 health_code_update();
3280
3281 // TODO: Validate cmd_ctx including sanity check for
3282 // security purpose.
3283
3284 rcu_thread_online();
3285 /*
3286 * This function dispatch the work to the kernel or userspace tracer
3287 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3288 * informations for the client. The command context struct contains
3289 * everything this function may needs.
3290 */
3291 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3292 rcu_thread_offline();
3293 if (ret < 0) {
3294 if (sock_error) {
3295 ret = close(sock);
3296 if (ret) {
3297 PERROR("close");
3298 }
3299 sock = -1;
3300 }
3301 /*
3302 * TODO: Inform client somehow of the fatal error. At
3303 * this point, ret < 0 means that a zmalloc failed
3304 * (ENOMEM). Error detected but still accept
3305 * command, unless a socket error has been
3306 * detected.
3307 */
3308 clean_command_ctx(&cmd_ctx);
3309 continue;
3310 }
3311
3312 health_code_update();
3313
3314 DBG("Sending response (size: %d, retcode: %s)",
3315 cmd_ctx->lttng_msg_size,
3316 lttng_strerror(-cmd_ctx->llm->ret_code));
3317 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3318 if (ret < 0) {
3319 ERR("Failed to send data back to client");
3320 }
3321
3322 /* End of transmission */
3323 ret = close(sock);
3324 if (ret) {
3325 PERROR("close");
3326 }
3327 sock = -1;
3328
3329 clean_command_ctx(&cmd_ctx);
3330
3331 health_code_update();
3332 }
3333
3334 exit:
3335 error:
3336 if (sock >= 0) {
3337 ret = close(sock);
3338 if (ret) {
3339 PERROR("close");
3340 }
3341 }
3342
3343 lttng_poll_clean(&events);
3344 clean_command_ctx(&cmd_ctx);
3345
3346 error_listen:
3347 error_create_poll:
3348 error_testpoint:
3349 unlink(client_unix_sock_path);
3350 if (client_sock >= 0) {
3351 ret = close(client_sock);
3352 if (ret) {
3353 PERROR("close");
3354 }
3355 }
3356
3357 if (err) {
3358 health_error();
3359 ERR("Health error occurred in %s", __func__);
3360 }
3361
3362 health_unregister();
3363
3364 DBG("Client thread dying");
3365
3366 rcu_unregister_thread();
3367 return NULL;
3368 }
3369
3370
3371 /*
3372 * usage function on stderr
3373 */
3374 static void usage(void)
3375 {
3376 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3377 fprintf(stderr, " -h, --help Display this usage.\n");
3378 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3379 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3380 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3381 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3382 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3383 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3384 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3385 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3386 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3387 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3388 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3389 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3390 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3391 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3392 fprintf(stderr, " -V, --version Show version number.\n");
3393 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3394 fprintf(stderr, " -q, --quiet No output at all.\n");
3395 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3396 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3397 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3398 }
3399
3400 /*
3401 * daemon argument parsing
3402 */
3403 static int parse_args(int argc, char **argv)
3404 {
3405 int c;
3406
3407 static struct option long_options[] = {
3408 { "client-sock", 1, 0, 'c' },
3409 { "apps-sock", 1, 0, 'a' },
3410 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3411 { "kconsumerd-err-sock", 1, 0, 'E' },
3412 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3413 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3414 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3415 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3416 { "consumerd32-path", 1, 0, 'u' },
3417 { "consumerd32-libdir", 1, 0, 'U' },
3418 { "consumerd64-path", 1, 0, 't' },
3419 { "consumerd64-libdir", 1, 0, 'T' },
3420 { "daemonize", 0, 0, 'd' },
3421 { "sig-parent", 0, 0, 'S' },
3422 { "help", 0, 0, 'h' },
3423 { "group", 1, 0, 'g' },
3424 { "version", 0, 0, 'V' },
3425 { "quiet", 0, 0, 'q' },
3426 { "verbose", 0, 0, 'v' },
3427 { "verbose-consumer", 0, 0, 'Z' },
3428 { "no-kernel", 0, 0, 'N' },
3429 { NULL, 0, 0, 0 }
3430 };
3431
3432 while (1) {
3433 int option_index = 0;
3434 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3435 long_options, &option_index);
3436 if (c == -1) {
3437 break;
3438 }
3439
3440 switch (c) {
3441 case 0:
3442 fprintf(stderr, "option %s", long_options[option_index].name);
3443 if (optarg) {
3444 fprintf(stderr, " with arg %s\n", optarg);
3445 }
3446 break;
3447 case 'c':
3448 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3449 break;
3450 case 'a':
3451 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3452 break;
3453 case 'd':
3454 opt_daemon = 1;
3455 break;
3456 case 'g':
3457 opt_tracing_group = optarg;
3458 break;
3459 case 'h':
3460 usage();
3461 exit(EXIT_FAILURE);
3462 case 'V':
3463 fprintf(stdout, "%s\n", VERSION);
3464 exit(EXIT_SUCCESS);
3465 case 'S':
3466 opt_sig_parent = 1;
3467 break;
3468 case 'E':
3469 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3470 break;
3471 case 'C':
3472 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3473 break;
3474 case 'F':
3475 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3476 break;
3477 case 'D':
3478 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3479 break;
3480 case 'H':
3481 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3482 break;
3483 case 'G':
3484 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3485 break;
3486 case 'N':
3487 opt_no_kernel = 1;
3488 break;
3489 case 'q':
3490 lttng_opt_quiet = 1;
3491 break;
3492 case 'v':
3493 /* Verbose level can increase using multiple -v */
3494 lttng_opt_verbose += 1;
3495 break;
3496 case 'Z':
3497 opt_verbose_consumer += 1;
3498 break;
3499 case 'u':
3500 consumerd32_bin= optarg;
3501 break;
3502 case 'U':
3503 consumerd32_libdir = optarg;
3504 break;
3505 case 't':
3506 consumerd64_bin = optarg;
3507 break;
3508 case 'T':
3509 consumerd64_libdir = optarg;
3510 break;
3511 default:
3512 /* Unknown option or other error.
3513 * Error is printed by getopt, just return */
3514 return -1;
3515 }
3516 }
3517
3518 return 0;
3519 }
3520
3521 /*
3522 * Creates the two needed socket by the daemon.
3523 * apps_sock - The communication socket for all UST apps.
3524 * client_sock - The communication of the cli tool (lttng).
3525 */
3526 static int init_daemon_socket(void)
3527 {
3528 int ret = 0;
3529 mode_t old_umask;
3530
3531 old_umask = umask(0);
3532
3533 /* Create client tool unix socket */
3534 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3535 if (client_sock < 0) {
3536 ERR("Create unix sock failed: %s", client_unix_sock_path);
3537 ret = -1;
3538 goto end;
3539 }
3540
3541 /* Set the cloexec flag */
3542 ret = utils_set_fd_cloexec(client_sock);
3543 if (ret < 0) {
3544 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3545 "Continuing but note that the consumer daemon will have a "
3546 "reference to this socket on exec()", client_sock);
3547 }
3548
3549 /* File permission MUST be 660 */
3550 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3551 if (ret < 0) {
3552 ERR("Set file permissions failed: %s", client_unix_sock_path);
3553 PERROR("chmod");
3554 goto end;
3555 }
3556
3557 /* Create the application unix socket */
3558 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3559 if (apps_sock < 0) {
3560 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3561 ret = -1;
3562 goto end;
3563 }
3564
3565 /* Set the cloexec flag */
3566 ret = utils_set_fd_cloexec(apps_sock);
3567 if (ret < 0) {
3568 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3569 "Continuing but note that the consumer daemon will have a "
3570 "reference to this socket on exec()", apps_sock);
3571 }
3572
3573 /* File permission MUST be 666 */
3574 ret = chmod(apps_unix_sock_path,
3575 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3576 if (ret < 0) {
3577 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3578 PERROR("chmod");
3579 goto end;
3580 }
3581
3582 DBG3("Session daemon client socket %d and application socket %d created",
3583 client_sock, apps_sock);
3584
3585 end:
3586 umask(old_umask);
3587 return ret;
3588 }
3589
3590 /*
3591 * Check if the global socket is available, and if a daemon is answering at the
3592 * other side. If yes, error is returned.
3593 */
3594 static int check_existing_daemon(void)
3595 {
3596 /* Is there anybody out there ? */
3597 if (lttng_session_daemon_alive()) {
3598 return -EEXIST;
3599 }
3600
3601 return 0;
3602 }
3603
3604 /*
3605 * Set the tracing group gid onto the client socket.
3606 *
3607 * Race window between mkdir and chown is OK because we are going from more
3608 * permissive (root.root) to less permissive (root.tracing).
3609 */
3610 static int set_permissions(char *rundir)
3611 {
3612 int ret;
3613 gid_t gid;
3614
3615 ret = allowed_group();
3616 if (ret < 0) {
3617 WARN("No tracing group detected");
3618 ret = 0;
3619 goto end;
3620 }
3621
3622 gid = ret;
3623
3624 /* Set lttng run dir */
3625 ret = chown(rundir, 0, gid);
3626 if (ret < 0) {
3627 ERR("Unable to set group on %s", rundir);
3628 PERROR("chown");
3629 }
3630
3631 /* Ensure tracing group can search the run dir */
3632 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3633 if (ret < 0) {
3634 ERR("Unable to set permissions on %s", rundir);
3635 PERROR("chmod");
3636 }
3637
3638 /* lttng client socket path */
3639 ret = chown(client_unix_sock_path, 0, gid);
3640 if (ret < 0) {
3641 ERR("Unable to set group on %s", client_unix_sock_path);
3642 PERROR("chown");
3643 }
3644
3645 /* kconsumer error socket path */
3646 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3647 if (ret < 0) {
3648 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3649 PERROR("chown");
3650 }
3651
3652 /* 64-bit ustconsumer error socket path */
3653 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3654 if (ret < 0) {
3655 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3656 PERROR("chown");
3657 }
3658
3659 /* 32-bit ustconsumer compat32 error socket path */
3660 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3661 if (ret < 0) {
3662 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3663 PERROR("chown");
3664 }
3665
3666 DBG("All permissions are set");
3667
3668 end:
3669 return ret;
3670 }
3671
3672 /*
3673 * Create the lttng run directory needed for all global sockets and pipe.
3674 */
3675 static int create_lttng_rundir(const char *rundir)
3676 {
3677 int ret;
3678
3679 DBG3("Creating LTTng run directory: %s", rundir);
3680
3681 ret = mkdir(rundir, S_IRWXU);
3682 if (ret < 0) {
3683 if (errno != EEXIST) {
3684 ERR("Unable to create %s", rundir);
3685 goto error;
3686 } else {
3687 ret = 0;
3688 }
3689 }
3690
3691 error:
3692 return ret;
3693 }
3694
3695 /*
3696 * Setup sockets and directory needed by the kconsumerd communication with the
3697 * session daemon.
3698 */
3699 static int set_consumer_sockets(struct consumer_data *consumer_data,
3700 const char *rundir)
3701 {
3702 int ret;
3703 char path[PATH_MAX];
3704
3705 switch (consumer_data->type) {
3706 case LTTNG_CONSUMER_KERNEL:
3707 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3708 break;
3709 case LTTNG_CONSUMER64_UST:
3710 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3711 break;
3712 case LTTNG_CONSUMER32_UST:
3713 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3714 break;
3715 default:
3716 ERR("Consumer type unknown");
3717 ret = -EINVAL;
3718 goto error;
3719 }
3720
3721 DBG2("Creating consumer directory: %s", path);
3722
3723 ret = mkdir(path, S_IRWXU);
3724 if (ret < 0) {
3725 if (errno != EEXIST) {
3726 PERROR("mkdir");
3727 ERR("Failed to create %s", path);
3728 goto error;
3729 }
3730 ret = -1;
3731 }
3732
3733 /* Create the kconsumerd error unix socket */
3734 consumer_data->err_sock =
3735 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3736 if (consumer_data->err_sock < 0) {
3737 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3738 ret = -1;
3739 goto error;
3740 }
3741
3742 /* File permission MUST be 660 */
3743 ret = chmod(consumer_data->err_unix_sock_path,
3744 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3745 if (ret < 0) {
3746 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3747 PERROR("chmod");
3748 goto error;
3749 }
3750
3751 error:
3752 return ret;
3753 }
3754
3755 /*
3756 * Signal handler for the daemon
3757 *
3758 * Simply stop all worker threads, leaving main() return gracefully after
3759 * joining all threads and calling cleanup().
3760 */
3761 static void sighandler(int sig)
3762 {
3763 switch (sig) {
3764 case SIGPIPE:
3765 DBG("SIGPIPE caught");
3766 return;
3767 case SIGINT:
3768 DBG("SIGINT caught");
3769 stop_threads();
3770 break;
3771 case SIGTERM:
3772 DBG("SIGTERM caught");
3773 stop_threads();
3774 break;
3775 default:
3776 break;
3777 }
3778 }
3779
3780 /*
3781 * Setup signal handler for :
3782 * SIGINT, SIGTERM, SIGPIPE
3783 */
3784 static int set_signal_handler(void)
3785 {
3786 int ret = 0;
3787 struct sigaction sa;
3788 sigset_t sigset;
3789
3790 if ((ret = sigemptyset(&sigset)) < 0) {
3791 PERROR("sigemptyset");
3792 return ret;
3793 }
3794
3795 sa.sa_handler = sighandler;
3796 sa.sa_mask = sigset;
3797 sa.sa_flags = 0;
3798 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3799 PERROR("sigaction");
3800 return ret;
3801 }
3802
3803 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3804 PERROR("sigaction");
3805 return ret;
3806 }
3807
3808 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3809 PERROR("sigaction");
3810 return ret;
3811 }
3812
3813 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3814
3815 return ret;
3816 }
3817
3818 /*
3819 * Set open files limit to unlimited. This daemon can open a large number of
3820 * file descriptors in order to consumer multiple kernel traces.
3821 */
3822 static void set_ulimit(void)
3823 {
3824 int ret;
3825 struct rlimit lim;
3826
3827 /* The kernel does not allowed an infinite limit for open files */
3828 lim.rlim_cur = 65535;
3829 lim.rlim_max = 65535;
3830
3831 ret = setrlimit(RLIMIT_NOFILE, &lim);
3832 if (ret < 0) {
3833 PERROR("failed to set open files limit");
3834 }
3835 }
3836
3837 /*
3838 * main
3839 */
3840 int main(int argc, char **argv)
3841 {
3842 int ret = 0;
3843 void *status;
3844 const char *home_path, *env_app_timeout;
3845
3846 init_kernel_workarounds();
3847
3848 rcu_register_thread();
3849
3850 setup_consumerd_path();
3851
3852 /* Parse arguments */
3853 progname = argv[0];
3854 if ((ret = parse_args(argc, argv)) < 0) {
3855 goto error;
3856 }
3857
3858 /* Daemonize */
3859 if (opt_daemon) {
3860 int i;
3861
3862 /*
3863 * fork
3864 * child: setsid, close FD 0, 1, 2, chdir /
3865 * parent: exit (if fork is successful)
3866 */
3867 ret = daemon(0, 0);
3868 if (ret < 0) {
3869 PERROR("daemon");
3870 goto error;
3871 }
3872 /*
3873 * We are in the child. Make sure all other file
3874 * descriptors are closed, in case we are called with
3875 * more opened file descriptors than the standard ones.
3876 */
3877 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3878 (void) close(i);
3879 }
3880 }
3881
3882 /* Create thread quit pipe */
3883 if ((ret = init_thread_quit_pipe()) < 0) {
3884 goto error;
3885 }
3886
3887 /* Check if daemon is UID = 0 */
3888 is_root = !getuid();
3889
3890 if (is_root) {
3891 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3892
3893 /* Create global run dir with root access */
3894 ret = create_lttng_rundir(rundir);
3895 if (ret < 0) {
3896 goto error;
3897 }
3898
3899 if (strlen(apps_unix_sock_path) == 0) {
3900 snprintf(apps_unix_sock_path, PATH_MAX,
3901 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3902 }
3903
3904 if (strlen(client_unix_sock_path) == 0) {
3905 snprintf(client_unix_sock_path, PATH_MAX,
3906 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3907 }
3908
3909 /* Set global SHM for ust */
3910 if (strlen(wait_shm_path) == 0) {
3911 snprintf(wait_shm_path, PATH_MAX,
3912 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3913 }
3914
3915 if (strlen(health_unix_sock_path) == 0) {
3916 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3917 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3918 }
3919
3920 /* Setup kernel consumerd path */
3921 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3922 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3923 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3924 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3925
3926 DBG2("Kernel consumer err path: %s",
3927 kconsumer_data.err_unix_sock_path);
3928 DBG2("Kernel consumer cmd path: %s",
3929 kconsumer_data.cmd_unix_sock_path);
3930 } else {
3931 home_path = get_home_dir();
3932 if (home_path == NULL) {
3933 /* TODO: Add --socket PATH option */
3934 ERR("Can't get HOME directory for sockets creation.");
3935 ret = -EPERM;
3936 goto error;
3937 }
3938
3939 /*
3940 * Create rundir from home path. This will create something like
3941 * $HOME/.lttng
3942 */
3943 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3944 if (ret < 0) {
3945 ret = -ENOMEM;
3946 goto error;
3947 }
3948
3949 ret = create_lttng_rundir(rundir);
3950 if (ret < 0) {
3951 goto error;
3952 }
3953
3954 if (strlen(apps_unix_sock_path) == 0) {
3955 snprintf(apps_unix_sock_path, PATH_MAX,
3956 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3957 }
3958
3959 /* Set the cli tool unix socket path */
3960 if (strlen(client_unix_sock_path) == 0) {
3961 snprintf(client_unix_sock_path, PATH_MAX,
3962 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3963 }
3964
3965 /* Set global SHM for ust */
3966 if (strlen(wait_shm_path) == 0) {
3967 snprintf(wait_shm_path, PATH_MAX,
3968 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3969 }
3970
3971 /* Set health check Unix path */
3972 if (strlen(health_unix_sock_path) == 0) {
3973 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3974 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3975 }
3976 }
3977
3978 /* Set consumer initial state */
3979 kernel_consumerd_state = CONSUMER_STOPPED;
3980 ust_consumerd_state = CONSUMER_STOPPED;
3981
3982 DBG("Client socket path %s", client_unix_sock_path);
3983 DBG("Application socket path %s", apps_unix_sock_path);
3984 DBG("LTTng run directory path: %s", rundir);
3985
3986 /* 32 bits consumerd path setup */
3987 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3988 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3989 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3990 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3991
3992 DBG2("UST consumer 32 bits err path: %s",
3993 ustconsumer32_data.err_unix_sock_path);
3994 DBG2("UST consumer 32 bits cmd path: %s",
3995 ustconsumer32_data.cmd_unix_sock_path);
3996
3997 /* 64 bits consumerd path setup */
3998 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3999 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4000 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4001 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4002
4003 DBG2("UST consumer 64 bits err path: %s",
4004 ustconsumer64_data.err_unix_sock_path);
4005 DBG2("UST consumer 64 bits cmd path: %s",
4006 ustconsumer64_data.cmd_unix_sock_path);
4007
4008 /*
4009 * See if daemon already exist.
4010 */
4011 if ((ret = check_existing_daemon()) < 0) {
4012 ERR("Already running daemon.\n");
4013 /*
4014 * We do not goto exit because we must not cleanup()
4015 * because a daemon is already running.
4016 */
4017 goto error;
4018 }
4019
4020 /*
4021 * Init UST app hash table. Alloc hash table before this point since
4022 * cleanup() can get called after that point.
4023 */
4024 ust_app_ht_alloc();
4025
4026 /* After this point, we can safely call cleanup() with "goto exit" */
4027
4028 /*
4029 * These actions must be executed as root. We do that *after* setting up
4030 * the sockets path because we MUST make the check for another daemon using
4031 * those paths *before* trying to set the kernel consumer sockets and init
4032 * kernel tracer.
4033 */
4034 if (is_root) {
4035 ret = set_consumer_sockets(&kconsumer_data, rundir);
4036 if (ret < 0) {
4037 goto exit;
4038 }
4039
4040 /* Setup kernel tracer */
4041 if (!opt_no_kernel) {
4042 init_kernel_tracer();
4043 }
4044
4045 /* Set ulimit for open files */
4046 set_ulimit();
4047 }
4048 /* init lttng_fd tracking must be done after set_ulimit. */
4049 lttng_fd_init();
4050
4051 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4052 if (ret < 0) {
4053 goto exit;
4054 }
4055
4056 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4057 if (ret < 0) {
4058 goto exit;
4059 }
4060
4061 if ((ret = set_signal_handler()) < 0) {
4062 goto exit;
4063 }
4064
4065 /* Setup the needed unix socket */
4066 if ((ret = init_daemon_socket()) < 0) {
4067 goto exit;
4068 }
4069
4070 /* Set credentials to socket */
4071 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4072 goto exit;
4073 }
4074
4075 /* Get parent pid if -S, --sig-parent is specified. */
4076 if (opt_sig_parent) {
4077 ppid = getppid();
4078 }
4079
4080 /* Setup the kernel pipe for waking up the kernel thread */
4081 if (is_root && !opt_no_kernel) {
4082 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4083 goto exit;
4084 }
4085 }
4086
4087 /* Setup the thread apps communication pipe. */
4088 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4089 goto exit;
4090 }
4091
4092 /* Init UST command queue. */
4093 cds_wfq_init(&ust_cmd_queue.queue);
4094
4095 /*
4096 * Get session list pointer. This pointer MUST NOT be free(). This list is
4097 * statically declared in session.c
4098 */
4099 session_list_ptr = session_get_list();
4100
4101 /* Set up max poll set size */
4102 lttng_poll_set_max_size();
4103
4104 cmd_init();
4105
4106 /* Check for the application socket timeout env variable. */
4107 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4108 if (env_app_timeout) {
4109 app_socket_timeout = atoi(env_app_timeout);
4110 } else {
4111 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4112 }
4113
4114 /* Create thread to manage the client socket */
4115 ret = pthread_create(&health_thread, NULL,
4116 thread_manage_health, (void *) NULL);
4117 if (ret != 0) {
4118 PERROR("pthread_create health");
4119 goto exit_health;
4120 }
4121
4122 /* Create thread to manage the client socket */
4123 ret = pthread_create(&client_thread, NULL,
4124 thread_manage_clients, (void *) NULL);
4125 if (ret != 0) {
4126 PERROR("pthread_create clients");
4127 goto exit_client;
4128 }
4129
4130 /* Create thread to dispatch registration */
4131 ret = pthread_create(&dispatch_thread, NULL,
4132 thread_dispatch_ust_registration, (void *) NULL);
4133 if (ret != 0) {
4134 PERROR("pthread_create dispatch");
4135 goto exit_dispatch;
4136 }
4137
4138 /* Create thread to manage application registration. */
4139 ret = pthread_create(&reg_apps_thread, NULL,
4140 thread_registration_apps, (void *) NULL);
4141 if (ret != 0) {
4142 PERROR("pthread_create registration");
4143 goto exit_reg_apps;
4144 }
4145
4146 /* Create thread to manage application socket */
4147 ret = pthread_create(&apps_thread, NULL,
4148 thread_manage_apps, (void *) NULL);
4149 if (ret != 0) {
4150 PERROR("pthread_create apps");
4151 goto exit_apps;
4152 }
4153
4154 /* Don't start this thread if kernel tracing is not requested nor root */
4155 if (is_root && !opt_no_kernel) {
4156 /* Create kernel thread to manage kernel event */
4157 ret = pthread_create(&kernel_thread, NULL,
4158 thread_manage_kernel, (void *) NULL);
4159 if (ret != 0) {
4160 PERROR("pthread_create kernel");
4161 goto exit_kernel;
4162 }
4163
4164 ret = pthread_join(kernel_thread, &status);
4165 if (ret != 0) {
4166 PERROR("pthread_join");
4167 goto error; /* join error, exit without cleanup */
4168 }
4169 }
4170
4171 exit_kernel:
4172 ret = pthread_join(apps_thread, &status);
4173 if (ret != 0) {
4174 PERROR("pthread_join");
4175 goto error; /* join error, exit without cleanup */
4176 }
4177
4178 exit_apps:
4179 ret = pthread_join(reg_apps_thread, &status);
4180 if (ret != 0) {
4181 PERROR("pthread_join");
4182 goto error; /* join error, exit without cleanup */
4183 }
4184
4185 exit_reg_apps:
4186 ret = pthread_join(dispatch_thread, &status);
4187 if (ret != 0) {
4188 PERROR("pthread_join");
4189 goto error; /* join error, exit without cleanup */
4190 }
4191
4192 exit_dispatch:
4193 ret = pthread_join(client_thread, &status);
4194 if (ret != 0) {
4195 PERROR("pthread_join");
4196 goto error; /* join error, exit without cleanup */
4197 }
4198
4199 ret = join_consumer_thread(&kconsumer_data);
4200 if (ret != 0) {
4201 PERROR("join_consumer");
4202 goto error; /* join error, exit without cleanup */
4203 }
4204
4205 ret = join_consumer_thread(&ustconsumer32_data);
4206 if (ret != 0) {
4207 PERROR("join_consumer ust32");
4208 goto error; /* join error, exit without cleanup */
4209 }
4210
4211 ret = join_consumer_thread(&ustconsumer64_data);
4212 if (ret != 0) {
4213 PERROR("join_consumer ust64");
4214 goto error; /* join error, exit without cleanup */
4215 }
4216
4217 exit_client:
4218 ret = pthread_join(health_thread, &status);
4219 if (ret != 0) {
4220 PERROR("pthread_join health thread");
4221 goto error; /* join error, exit without cleanup */
4222 }
4223
4224 exit_health:
4225 exit:
4226 /*
4227 * cleanup() is called when no other thread is running.
4228 */
4229 rcu_thread_online();
4230 cleanup();
4231 rcu_thread_offline();
4232 rcu_unregister_thread();
4233 if (!ret) {
4234 exit(EXIT_SUCCESS);
4235 }
4236 error:
4237 exit(EXIT_FAILURE);
4238 }
This page took 0.143795 seconds and 4 git commands to generate.