Fix: close consumer sockets in sessiond cleanup
[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 <inttypes.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <config.h>
39
40 #include <common/common.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 "buffer-registry.h"
50 #include "channel.h"
51 #include "cmd.h"
52 #include "consumer.h"
53 #include "context.h"
54 #include "event.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "modprobe.h"
58 #include "shm.h"
59 #include "ust-ctl.h"
60 #include "ust-consumer.h"
61 #include "utils.h"
62 #include "fd-limit.h"
63 #include "health.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66
67 #define CONSUMERD_FILE "lttng-consumerd"
68
69 /* Const values */
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71
72 const char *progname;
73 const char *opt_tracing_group;
74 static const char *opt_pidfile;
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 .metadata_sock.fd = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .metadata_sock.fd = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer32_data = {
112 .type = LTTNG_CONSUMER32_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .metadata_sock.fd = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123
124 /* Shared between threads */
125 static int dispatch_thread_exit;
126
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path[PATH_MAX];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path[PATH_MAX];
131 /* global wait shm path for UST */
132 static char wait_shm_path[PATH_MAX];
133 /* Global health check unix path */
134 static char health_unix_sock_path[PATH_MAX];
135
136 /* Sockets and FDs */
137 static int client_sock = -1;
138 static int apps_sock = -1;
139 int kernel_tracer_fd = -1;
140 static int kernel_poll_pipe[2] = { -1, -1 };
141
142 /*
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
145 */
146 static int thread_quit_pipe[2] = { -1, -1 };
147
148 /*
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
151 */
152 static int apps_cmd_pipe[2] = { -1, -1 };
153
154 int apps_cmd_notify_pipe[2] = { -1, -1 };
155
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread;
158 static pthread_t apps_notify_thread;
159 static pthread_t reg_apps_thread;
160 static pthread_t client_thread;
161 static pthread_t kernel_thread;
162 static pthread_t dispatch_thread;
163 static pthread_t health_thread;
164 static pthread_t ht_cleanup_thread;
165
166 /*
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
169 *
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
172 */
173 static struct ust_cmd_queue ust_cmd_queue;
174
175 /*
176 * Pointer initialized before thread creation.
177 *
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
181 *
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
184 */
185 static struct ltt_session_list *session_list_ptr;
186
187 int ust_consumerd64_fd = -1;
188 int ust_consumerd32_fd = -1;
189
190 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
191 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
192 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
193 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
194
195 static const char *module_proc_lttng = "/proc/lttng";
196
197 /*
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
200 */
201 enum consumerd_state {
202 CONSUMER_STARTED = 1,
203 CONSUMER_STOPPED = 2,
204 CONSUMER_ERROR = 3,
205 };
206
207 /*
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
212 *
213 * The following example shows a possible race condition of this scheme:
214 *
215 * consumer thread error happens
216 * client cmd arrives
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
220 * ...
221 *
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
227 */
228 static enum consumerd_state ust_consumerd_state;
229 static enum consumerd_state kernel_consumerd_state;
230
231 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 /* Set in main() with the current page size. */
237 long page_size;
238
239 static
240 void setup_consumerd_path(void)
241 {
242 const char *bin, *libdir;
243
244 /*
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
248 */
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin[0]) {
251 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
252 }
253 if (!consumerd32_libdir[0]) {
254 consumerd32_libdir = INSTALL_LIB_PATH;
255 }
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin[0]) {
258 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
259 }
260 if (!consumerd64_libdir[0]) {
261 consumerd64_libdir = INSTALL_LIB_PATH;
262 }
263 #else
264 #error "Unknown bitness"
265 #endif
266
267 /*
268 * runtime env. var. overrides the build default.
269 */
270 bin = getenv("LTTNG_CONSUMERD32_BIN");
271 if (bin) {
272 consumerd32_bin = bin;
273 }
274 bin = getenv("LTTNG_CONSUMERD64_BIN");
275 if (bin) {
276 consumerd64_bin = bin;
277 }
278 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
279 if (libdir) {
280 consumerd32_libdir = libdir;
281 }
282 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
283 if (libdir) {
284 consumerd64_libdir = libdir;
285 }
286 }
287
288 /*
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
290 */
291 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
292 {
293 int ret;
294
295 assert(events);
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310 error:
311 return ret;
312 }
313
314 /*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
319 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326 }
327
328 /*
329 * Return group ID of the tracing group or -1 if not found.
330 */
331 static gid_t allowed_group(void)
332 {
333 struct group *grp;
334
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345 }
346
347 /*
348 * Init thread quit pipe.
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352 static int init_thread_quit_pipe(void)
353 {
354 int ret, i;
355
356 ret = pipe(thread_quit_pipe);
357 if (ret < 0) {
358 PERROR("thread quit pipe");
359 goto error;
360 }
361
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
370 error:
371 return ret;
372 }
373
374 /*
375 * Stop all threads by closing the thread quit pipe.
376 */
377 static void stop_threads(void)
378 {
379 int ret;
380
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret = notify_thread_pipe(thread_quit_pipe[1]);
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
390 futex_nto1_wake(&ust_cmd_queue.futex);
391 }
392
393 /*
394 * Close every consumer sockets.
395 */
396 static void close_consumer_sockets(void)
397 {
398 int ret;
399
400 if (kconsumer_data.err_sock >= 0) {
401 ret = close(kconsumer_data.err_sock);
402 if (ret < 0) {
403 PERROR("kernel consumer err_sock close");
404 }
405 }
406 if (ustconsumer32_data.err_sock >= 0) {
407 ret = close(ustconsumer32_data.err_sock);
408 if (ret < 0) {
409 PERROR("UST consumer32 err_sock close");
410 }
411 }
412 if (ustconsumer64_data.err_sock >= 0) {
413 ret = close(ustconsumer64_data.err_sock);
414 if (ret < 0) {
415 PERROR("UST consumer64 err_sock close");
416 }
417 }
418 if (kconsumer_data.cmd_sock >= 0) {
419 ret = close(kconsumer_data.cmd_sock);
420 if (ret < 0) {
421 PERROR("kernel consumer cmd_sock close");
422 }
423 }
424 if (ustconsumer32_data.cmd_sock >= 0) {
425 ret = close(ustconsumer32_data.cmd_sock);
426 if (ret < 0) {
427 PERROR("UST consumer32 cmd_sock close");
428 }
429 }
430 if (ustconsumer64_data.cmd_sock >= 0) {
431 ret = close(ustconsumer64_data.cmd_sock);
432 if (ret < 0) {
433 PERROR("UST consumer64 cmd_sock close");
434 }
435 }
436 }
437
438 /*
439 * Cleanup the daemon
440 */
441 static void cleanup(void)
442 {
443 int ret;
444 char *cmd = NULL;
445 struct ltt_session *sess, *stmp;
446
447 DBG("Cleaning up");
448
449 /* First thing first, stop all threads */
450 utils_close_pipe(thread_quit_pipe);
451
452 /*
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
455 */
456 if (opt_pidfile) {
457 ret = remove(opt_pidfile);
458 if (ret < 0) {
459 PERROR("remove pidfile %s", opt_pidfile);
460 }
461 }
462
463 DBG("Removing %s directory", rundir);
464 ret = asprintf(&cmd, "rm -rf %s", rundir);
465 if (ret < 0) {
466 ERR("asprintf failed. Something is really wrong!");
467 }
468
469 /* Remove lttng run directory */
470 ret = system(cmd);
471 if (ret < 0) {
472 ERR("Unable to clean %s", rundir);
473 }
474 free(cmd);
475 free(rundir);
476
477 DBG("Cleaning up all sessions");
478
479 /* Destroy session list mutex */
480 if (session_list_ptr != NULL) {
481 pthread_mutex_destroy(&session_list_ptr->lock);
482
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess, stmp,
485 &session_list_ptr->head, list) {
486 cmd_destroy_session(sess, kernel_poll_pipe[1]);
487 }
488 }
489
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
493
494 if (is_root && !opt_no_kernel) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd >= 0) {
497 ret = close(kernel_tracer_fd);
498 if (ret) {
499 PERROR("close");
500 }
501 }
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
504 }
505
506 close_consumer_sockets();
507
508 /* <fun> */
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
512 /* </fun> */
513 }
514
515 /*
516 * Send data on a unix socket using the liblttsessiondcomm API.
517 *
518 * Return lttcomm error code.
519 */
520 static int send_unix_sock(int sock, void *buf, size_t len)
521 {
522 /* Check valid length */
523 if (len == 0) {
524 return -1;
525 }
526
527 return lttcomm_send_unix_sock(sock, buf, len);
528 }
529
530 /*
531 * Free memory of a command context structure.
532 */
533 static void clean_command_ctx(struct command_ctx **cmd_ctx)
534 {
535 DBG("Clean command context structure");
536 if (*cmd_ctx) {
537 if ((*cmd_ctx)->llm) {
538 free((*cmd_ctx)->llm);
539 }
540 if ((*cmd_ctx)->lsm) {
541 free((*cmd_ctx)->lsm);
542 }
543 free(*cmd_ctx);
544 *cmd_ctx = NULL;
545 }
546 }
547
548 /*
549 * Notify UST applications using the shm mmap futex.
550 */
551 static int notify_ust_apps(int active)
552 {
553 char *wait_shm_mmap;
554
555 DBG("Notifying applications of session daemon state: %d", active);
556
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
559 if (wait_shm_mmap == NULL) {
560 goto error;
561 }
562
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap, active);
565
566 /* Apps notified successfully */
567 return 0;
568
569 error:
570 return -1;
571 }
572
573 /*
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
576 * structure.
577 *
578 * Return total size of the buffer pointed by buf.
579 */
580 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
581 {
582 int ret, buf_size;
583
584 buf_size = size;
585
586 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
587 if (cmd_ctx->llm == NULL) {
588 PERROR("zmalloc");
589 ret = -ENOMEM;
590 goto error;
591 }
592
593 /* Copy common data */
594 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
595 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
596
597 cmd_ctx->llm->data_size = size;
598 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
599
600 return buf_size;
601
602 error:
603 return ret;
604 }
605
606 /*
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
609 */
610 static int update_kernel_poll(struct lttng_poll_event *events)
611 {
612 int ret;
613 struct ltt_session *session;
614 struct ltt_kernel_channel *channel;
615
616 DBG("Updating kernel poll set");
617
618 session_lock_list();
619 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
620 session_lock(session);
621 if (session->kernel_session == NULL) {
622 session_unlock(session);
623 continue;
624 }
625
626 cds_list_for_each_entry(channel,
627 &session->kernel_session->channel_list.head, list) {
628 /* Add channel fd to the kernel poll set */
629 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
630 if (ret < 0) {
631 session_unlock(session);
632 goto error;
633 }
634 DBG("Channel fd %d added to kernel set", channel->fd);
635 }
636 session_unlock(session);
637 }
638 session_unlock_list();
639
640 return 0;
641
642 error:
643 session_unlock_list();
644 return -1;
645 }
646
647 /*
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
650 *
651 * Useful for CPU hotplug feature.
652 */
653 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
654 {
655 int ret = 0;
656 struct ltt_session *session;
657 struct ltt_kernel_session *ksess;
658 struct ltt_kernel_channel *channel;
659
660 DBG("Updating kernel streams for channel fd %d", fd);
661
662 session_lock_list();
663 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
664 session_lock(session);
665 if (session->kernel_session == NULL) {
666 session_unlock(session);
667 continue;
668 }
669 ksess = session->kernel_session;
670
671 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
672 if (channel->fd == fd) {
673 DBG("Channel found, updating kernel streams");
674 ret = kernel_open_channel_stream(channel);
675 if (ret < 0) {
676 goto error;
677 }
678
679 /*
680 * Have we already sent fds to the consumer? If yes, it means
681 * that tracing is started so it is safe to send our updated
682 * stream fds.
683 */
684 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
685 struct lttng_ht_iter iter;
686 struct consumer_socket *socket;
687
688 rcu_read_lock();
689 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
690 &iter.iter, socket, node.node) {
691 /* Code flow error */
692 assert(socket->fd >= 0);
693
694 pthread_mutex_lock(socket->lock);
695 ret = kernel_consumer_send_channel_stream(socket,
696 channel, ksess,
697 session->output_traces ? 1 : 0);
698 pthread_mutex_unlock(socket->lock);
699 if (ret < 0) {
700 rcu_read_unlock();
701 goto error;
702 }
703 }
704 rcu_read_unlock();
705 }
706 goto error;
707 }
708 }
709 session_unlock(session);
710 }
711 session_unlock_list();
712 return ret;
713
714 error:
715 session_unlock(session);
716 session_unlock_list();
717 return ret;
718 }
719
720 /*
721 * For each tracing session, update newly registered apps. The session list
722 * lock MUST be acquired before calling this.
723 */
724 static void update_ust_app(int app_sock)
725 {
726 struct ltt_session *sess, *stmp;
727
728 /* For all tracing session(s) */
729 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
730 session_lock(sess);
731 if (sess->ust_session) {
732 ust_app_global_update(sess->ust_session, app_sock);
733 }
734 session_unlock(sess);
735 }
736 }
737
738 /*
739 * This thread manage event coming from the kernel.
740 *
741 * Features supported in this thread:
742 * -) CPU Hotplug
743 */
744 static void *thread_manage_kernel(void *data)
745 {
746 int ret, i, pollfd, update_poll_flag = 1, err = -1;
747 uint32_t revents, nb_fd;
748 char tmp;
749 struct lttng_poll_event events;
750
751 DBG("[thread] Thread manage kernel started");
752
753 health_register(HEALTH_TYPE_KERNEL);
754
755 /*
756 * This first step of the while is to clean this structure which could free
757 * non NULL pointers so initialize it before the loop.
758 */
759 lttng_poll_init(&events);
760
761 if (testpoint(thread_manage_kernel)) {
762 goto error_testpoint;
763 }
764
765 health_code_update();
766
767 if (testpoint(thread_manage_kernel_before_loop)) {
768 goto error_testpoint;
769 }
770
771 while (1) {
772 health_code_update();
773
774 if (update_poll_flag == 1) {
775 /* Clean events object. We are about to populate it again. */
776 lttng_poll_clean(&events);
777
778 ret = sessiond_set_thread_pollset(&events, 2);
779 if (ret < 0) {
780 goto error_poll_create;
781 }
782
783 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
784 if (ret < 0) {
785 goto error;
786 }
787
788 /* This will add the available kernel channel if any. */
789 ret = update_kernel_poll(&events);
790 if (ret < 0) {
791 goto error;
792 }
793 update_poll_flag = 0;
794 }
795
796 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
797
798 /* Poll infinite value of time */
799 restart:
800 health_poll_entry();
801 ret = lttng_poll_wait(&events, -1);
802 health_poll_exit();
803 if (ret < 0) {
804 /*
805 * Restart interrupted system call.
806 */
807 if (errno == EINTR) {
808 goto restart;
809 }
810 goto error;
811 } else if (ret == 0) {
812 /* Should not happen since timeout is infinite */
813 ERR("Return value of poll is 0 with an infinite timeout.\n"
814 "This should not have happened! Continuing...");
815 continue;
816 }
817
818 nb_fd = ret;
819
820 for (i = 0; i < nb_fd; i++) {
821 /* Fetch once the poll data */
822 revents = LTTNG_POLL_GETEV(&events, i);
823 pollfd = LTTNG_POLL_GETFD(&events, i);
824
825 health_code_update();
826
827 /* Thread quit pipe has been closed. Killing thread. */
828 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
829 if (ret) {
830 err = 0;
831 goto exit;
832 }
833
834 /* Check for data on kernel pipe */
835 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
836 do {
837 ret = read(kernel_poll_pipe[0], &tmp, 1);
838 } while (ret < 0 && errno == EINTR);
839 /*
840 * Ret value is useless here, if this pipe gets any actions an
841 * update is required anyway.
842 */
843 update_poll_flag = 1;
844 continue;
845 } else {
846 /*
847 * New CPU detected by the kernel. Adding kernel stream to
848 * kernel session and updating the kernel consumer
849 */
850 if (revents & LPOLLIN) {
851 ret = update_kernel_stream(&kconsumer_data, pollfd);
852 if (ret < 0) {
853 continue;
854 }
855 break;
856 /*
857 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
858 * and unregister kernel stream at this point.
859 */
860 }
861 }
862 }
863 }
864
865 exit:
866 error:
867 lttng_poll_clean(&events);
868 error_poll_create:
869 error_testpoint:
870 utils_close_pipe(kernel_poll_pipe);
871 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
872 if (err) {
873 health_error();
874 ERR("Health error occurred in %s", __func__);
875 WARN("Kernel thread died unexpectedly. "
876 "Kernel tracing can continue but CPU hotplug is disabled.");
877 }
878 health_unregister();
879 DBG("Kernel thread dying");
880 return NULL;
881 }
882
883 /*
884 * Signal pthread condition of the consumer data that the thread.
885 */
886 static void signal_consumer_condition(struct consumer_data *data, int state)
887 {
888 pthread_mutex_lock(&data->cond_mutex);
889
890 /*
891 * The state is set before signaling. It can be any value, it's the waiter
892 * job to correctly interpret this condition variable associated to the
893 * consumer pthread_cond.
894 *
895 * A value of 0 means that the corresponding thread of the consumer data
896 * was not started. 1 indicates that the thread has started and is ready
897 * for action. A negative value means that there was an error during the
898 * thread bootstrap.
899 */
900 data->consumer_thread_is_ready = state;
901 (void) pthread_cond_signal(&data->cond);
902
903 pthread_mutex_unlock(&data->cond_mutex);
904 }
905
906 /*
907 * This thread manage the consumer error sent back to the session daemon.
908 */
909 static void *thread_manage_consumer(void *data)
910 {
911 int sock = -1, i, ret, pollfd, err = -1;
912 uint32_t revents, nb_fd;
913 enum lttcomm_return_code code;
914 struct lttng_poll_event events;
915 struct consumer_data *consumer_data = data;
916
917 DBG("[thread] Manage consumer started");
918
919 health_register(HEALTH_TYPE_CONSUMER);
920
921 health_code_update();
922
923 /*
924 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
925 * metadata_sock. Nothing more will be added to this poll set.
926 */
927 ret = sessiond_set_thread_pollset(&events, 3);
928 if (ret < 0) {
929 goto error_poll;
930 }
931
932 /*
933 * The error socket here is already in a listening state which was done
934 * just before spawning this thread to avoid a race between the consumer
935 * daemon exec trying to connect and the listen() call.
936 */
937 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
938 if (ret < 0) {
939 goto error;
940 }
941
942 health_code_update();
943
944 /* Infinite blocking call, waiting for transmission */
945 restart:
946 health_poll_entry();
947
948 if (testpoint(thread_manage_consumer)) {
949 goto error;
950 }
951
952 ret = lttng_poll_wait(&events, -1);
953 health_poll_exit();
954 if (ret < 0) {
955 /*
956 * Restart interrupted system call.
957 */
958 if (errno == EINTR) {
959 goto restart;
960 }
961 goto error;
962 }
963
964 nb_fd = ret;
965
966 for (i = 0; i < nb_fd; i++) {
967 /* Fetch once the poll data */
968 revents = LTTNG_POLL_GETEV(&events, i);
969 pollfd = LTTNG_POLL_GETFD(&events, i);
970
971 health_code_update();
972
973 /* Thread quit pipe has been closed. Killing thread. */
974 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
975 if (ret) {
976 err = 0;
977 goto exit;
978 }
979
980 /* Event on the registration socket */
981 if (pollfd == consumer_data->err_sock) {
982 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
983 ERR("consumer err socket poll error");
984 goto error;
985 }
986 }
987 }
988
989 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
990 if (sock < 0) {
991 goto error;
992 }
993
994 /*
995 * Set the CLOEXEC flag. Return code is useless because either way, the
996 * show must go on.
997 */
998 (void) utils_set_fd_cloexec(sock);
999
1000 health_code_update();
1001
1002 DBG2("Receiving code from consumer err_sock");
1003
1004 /* Getting status code from kconsumerd */
1005 ret = lttcomm_recv_unix_sock(sock, &code,
1006 sizeof(enum lttcomm_return_code));
1007 if (ret <= 0) {
1008 goto error;
1009 }
1010
1011 health_code_update();
1012
1013 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1014 /* Connect both socket, command and metadata. */
1015 consumer_data->cmd_sock =
1016 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1017 consumer_data->metadata_sock.fd =
1018 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1019 if (consumer_data->cmd_sock < 0 ||
1020 consumer_data->metadata_sock.fd < 0) {
1021 PERROR("consumer connect cmd socket");
1022 /* On error, signal condition and quit. */
1023 signal_consumer_condition(consumer_data, -1);
1024 goto error;
1025 }
1026 /* Create metadata socket lock. */
1027 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1028 if (consumer_data->metadata_sock.lock == NULL) {
1029 PERROR("zmalloc pthread mutex");
1030 ret = -1;
1031 goto error;
1032 }
1033 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1034
1035 signal_consumer_condition(consumer_data, 1);
1036 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1037 DBG("Consumer metadata socket ready (fd: %d)",
1038 consumer_data->metadata_sock.fd);
1039 } else {
1040 ERR("consumer error when waiting for SOCK_READY : %s",
1041 lttcomm_get_readable_code(-code));
1042 goto error;
1043 }
1044
1045 /* Remove the consumerd error sock since we've established a connexion */
1046 ret = lttng_poll_del(&events, consumer_data->err_sock);
1047 if (ret < 0) {
1048 goto error;
1049 }
1050
1051 /* Add new accepted error socket. */
1052 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1053 if (ret < 0) {
1054 goto error;
1055 }
1056
1057 /* Add metadata socket that is successfully connected. */
1058 ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
1059 LPOLLIN | LPOLLRDHUP);
1060 if (ret < 0) {
1061 goto error;
1062 }
1063
1064 health_code_update();
1065
1066 /* Infinite blocking call, waiting for transmission */
1067 restart_poll:
1068 while (1) {
1069 health_poll_entry();
1070 ret = lttng_poll_wait(&events, -1);
1071 health_poll_exit();
1072 if (ret < 0) {
1073 /*
1074 * Restart interrupted system call.
1075 */
1076 if (errno == EINTR) {
1077 goto restart_poll;
1078 }
1079 goto error;
1080 }
1081
1082 nb_fd = ret;
1083
1084 for (i = 0; i < nb_fd; i++) {
1085 /* Fetch once the poll data */
1086 revents = LTTNG_POLL_GETEV(&events, i);
1087 pollfd = LTTNG_POLL_GETFD(&events, i);
1088
1089 health_code_update();
1090
1091 /* Thread quit pipe has been closed. Killing thread. */
1092 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1093 if (ret) {
1094 err = 0;
1095 goto exit;
1096 }
1097
1098 if (pollfd == sock) {
1099 /* Event on the consumerd socket */
1100 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1101 ERR("consumer err socket second poll error");
1102 goto error;
1103 }
1104 health_code_update();
1105 /* Wait for any kconsumerd error */
1106 ret = lttcomm_recv_unix_sock(sock, &code,
1107 sizeof(enum lttcomm_return_code));
1108 if (ret <= 0) {
1109 ERR("consumer closed the command socket");
1110 goto error;
1111 }
1112
1113 ERR("consumer return code : %s",
1114 lttcomm_get_readable_code(-code));
1115
1116 goto exit;
1117 } else if (pollfd == consumer_data->metadata_sock.fd) {
1118 /* UST metadata requests */
1119 ret = ust_consumer_metadata_request(
1120 &consumer_data->metadata_sock);
1121 if (ret < 0) {
1122 ERR("Handling metadata request");
1123 goto error;
1124 }
1125 break;
1126 } else {
1127 ERR("Unknown pollfd");
1128 goto error;
1129 }
1130 }
1131 health_code_update();
1132 }
1133
1134 exit:
1135 error:
1136 /* Immediately set the consumerd state to stopped */
1137 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1138 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1139 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1140 consumer_data->type == LTTNG_CONSUMER32_UST) {
1141 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1142 } else {
1143 /* Code flow error... */
1144 assert(0);
1145 }
1146
1147 if (consumer_data->err_sock >= 0) {
1148 ret = close(consumer_data->err_sock);
1149 if (ret) {
1150 PERROR("close");
1151 }
1152 }
1153 if (consumer_data->cmd_sock >= 0) {
1154 ret = close(consumer_data->cmd_sock);
1155 if (ret) {
1156 PERROR("close");
1157 }
1158 }
1159 if (consumer_data->metadata_sock.fd >= 0) {
1160 ret = close(consumer_data->metadata_sock.fd);
1161 if (ret) {
1162 PERROR("close");
1163 }
1164 }
1165 /* Cleanup metadata socket mutex. */
1166 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1167 free(consumer_data->metadata_sock.lock);
1168
1169 if (sock >= 0) {
1170 ret = close(sock);
1171 if (ret) {
1172 PERROR("close");
1173 }
1174 }
1175
1176 unlink(consumer_data->err_unix_sock_path);
1177 unlink(consumer_data->cmd_unix_sock_path);
1178 consumer_data->pid = 0;
1179
1180 lttng_poll_clean(&events);
1181 error_poll:
1182 if (err) {
1183 health_error();
1184 ERR("Health error occurred in %s", __func__);
1185 }
1186 health_unregister();
1187 DBG("consumer thread cleanup completed");
1188
1189 return NULL;
1190 }
1191
1192 /*
1193 * This thread manage application communication.
1194 */
1195 static void *thread_manage_apps(void *data)
1196 {
1197 int i, ret, pollfd, err = -1;
1198 uint32_t revents, nb_fd;
1199 struct lttng_poll_event events;
1200
1201 DBG("[thread] Manage application started");
1202
1203 rcu_register_thread();
1204 rcu_thread_online();
1205
1206 health_register(HEALTH_TYPE_APP_MANAGE);
1207
1208 if (testpoint(thread_manage_apps)) {
1209 goto error_testpoint;
1210 }
1211
1212 health_code_update();
1213
1214 ret = sessiond_set_thread_pollset(&events, 2);
1215 if (ret < 0) {
1216 goto error_poll_create;
1217 }
1218
1219 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1220 if (ret < 0) {
1221 goto error;
1222 }
1223
1224 if (testpoint(thread_manage_apps_before_loop)) {
1225 goto error;
1226 }
1227
1228 health_code_update();
1229
1230 while (1) {
1231 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1232
1233 /* Inifinite blocking call, waiting for transmission */
1234 restart:
1235 health_poll_entry();
1236 ret = lttng_poll_wait(&events, -1);
1237 health_poll_exit();
1238 if (ret < 0) {
1239 /*
1240 * Restart interrupted system call.
1241 */
1242 if (errno == EINTR) {
1243 goto restart;
1244 }
1245 goto error;
1246 }
1247
1248 nb_fd = ret;
1249
1250 for (i = 0; i < nb_fd; i++) {
1251 /* Fetch once the poll data */
1252 revents = LTTNG_POLL_GETEV(&events, i);
1253 pollfd = LTTNG_POLL_GETFD(&events, i);
1254
1255 health_code_update();
1256
1257 /* Thread quit pipe has been closed. Killing thread. */
1258 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1259 if (ret) {
1260 err = 0;
1261 goto exit;
1262 }
1263
1264 /* Inspect the apps cmd pipe */
1265 if (pollfd == apps_cmd_pipe[0]) {
1266 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1267 ERR("Apps command pipe error");
1268 goto error;
1269 } else if (revents & LPOLLIN) {
1270 int sock;
1271
1272 /* Empty pipe */
1273 do {
1274 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1275 } while (ret < 0 && errno == EINTR);
1276 if (ret < 0 || ret < sizeof(sock)) {
1277 PERROR("read apps cmd pipe");
1278 goto error;
1279 }
1280
1281 health_code_update();
1282
1283 /*
1284 * We only monitor the error events of the socket. This
1285 * thread does not handle any incoming data from UST
1286 * (POLLIN).
1287 */
1288 ret = lttng_poll_add(&events, sock,
1289 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1290 if (ret < 0) {
1291 goto error;
1292 }
1293
1294 /* Set socket timeout for both receiving and ending */
1295 (void) lttcomm_setsockopt_rcv_timeout(sock,
1296 app_socket_timeout);
1297 (void) lttcomm_setsockopt_snd_timeout(sock,
1298 app_socket_timeout);
1299
1300 DBG("Apps with sock %d added to poll set", sock);
1301
1302 health_code_update();
1303
1304 break;
1305 }
1306 } else {
1307 /*
1308 * At this point, we know that a registered application made
1309 * the event at poll_wait.
1310 */
1311 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1312 /* Removing from the poll set */
1313 ret = lttng_poll_del(&events, pollfd);
1314 if (ret < 0) {
1315 goto error;
1316 }
1317
1318 /* Socket closed on remote end. */
1319 ust_app_unregister(pollfd);
1320 break;
1321 }
1322 }
1323
1324 health_code_update();
1325 }
1326 }
1327
1328 exit:
1329 error:
1330 lttng_poll_clean(&events);
1331 error_poll_create:
1332 error_testpoint:
1333 utils_close_pipe(apps_cmd_pipe);
1334 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1335
1336 /*
1337 * We don't clean the UST app hash table here since already registered
1338 * applications can still be controlled so let them be until the session
1339 * daemon dies or the applications stop.
1340 */
1341
1342 if (err) {
1343 health_error();
1344 ERR("Health error occurred in %s", __func__);
1345 }
1346 health_unregister();
1347 DBG("Application communication apps thread cleanup complete");
1348 rcu_thread_offline();
1349 rcu_unregister_thread();
1350 return NULL;
1351 }
1352
1353 /*
1354 * Send a socket to a thread This is called from the dispatch UST registration
1355 * thread once all sockets are set for the application.
1356 *
1357 * On success, return 0 else a negative value being the errno message of the
1358 * write().
1359 */
1360 static int send_socket_to_thread(int fd, int sock)
1361 {
1362 int ret;
1363
1364 /* Sockets MUST be set or else this should not have been called. */
1365 assert(fd >= 0);
1366 assert(sock >= 0);
1367
1368 do {
1369 ret = write(fd, &sock, sizeof(sock));
1370 } while (ret < 0 && errno == EINTR);
1371 if (ret < 0 || ret != sizeof(sock)) {
1372 PERROR("write apps pipe %d", fd);
1373 if (ret < 0) {
1374 ret = -errno;
1375 }
1376 goto error;
1377 }
1378
1379 /* All good. Don't send back the write positive ret value. */
1380 ret = 0;
1381 error:
1382 return ret;
1383 }
1384
1385 /*
1386 * Sanitize the wait queue of the dispatch registration thread meaning removing
1387 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1388 * notify socket is never received.
1389 */
1390 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1391 {
1392 int ret, nb_fd = 0, i;
1393 unsigned int fd_added = 0;
1394 struct lttng_poll_event events;
1395 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1396
1397 assert(wait_queue);
1398
1399 lttng_poll_init(&events);
1400
1401 /* Just skip everything for an empty queue. */
1402 if (!wait_queue->count) {
1403 goto end;
1404 }
1405
1406 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1407 if (ret < 0) {
1408 goto error_create;
1409 }
1410
1411 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1412 &wait_queue->head, head) {
1413 assert(wait_node->app);
1414 ret = lttng_poll_add(&events, wait_node->app->sock,
1415 LPOLLHUP | LPOLLERR);
1416 if (ret < 0) {
1417 goto error;
1418 }
1419
1420 fd_added = 1;
1421 }
1422
1423 if (!fd_added) {
1424 goto end;
1425 }
1426
1427 /*
1428 * Poll but don't block so we can quickly identify the faulty events and
1429 * clean them afterwards from the wait queue.
1430 */
1431 ret = lttng_poll_wait(&events, 0);
1432 if (ret < 0) {
1433 goto error;
1434 }
1435 nb_fd = ret;
1436
1437 for (i = 0; i < nb_fd; i++) {
1438 /* Get faulty FD. */
1439 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1440 int pollfd = LTTNG_POLL_GETFD(&events, i);
1441
1442 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1443 &wait_queue->head, head) {
1444 if (pollfd == wait_node->app->sock &&
1445 (revents & (LPOLLHUP | LPOLLERR))) {
1446 cds_list_del(&wait_node->head);
1447 wait_queue->count--;
1448 ust_app_destroy(wait_node->app);
1449 free(wait_node);
1450 break;
1451 }
1452 }
1453 }
1454
1455 if (nb_fd > 0) {
1456 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1457 }
1458
1459 end:
1460 lttng_poll_clean(&events);
1461 return;
1462
1463 error:
1464 lttng_poll_clean(&events);
1465 error_create:
1466 ERR("Unable to sanitize wait queue");
1467 return;
1468 }
1469
1470 /*
1471 * Dispatch request from the registration threads to the application
1472 * communication thread.
1473 */
1474 static void *thread_dispatch_ust_registration(void *data)
1475 {
1476 int ret, err = -1;
1477 struct cds_wfq_node *node;
1478 struct ust_command *ust_cmd = NULL;
1479 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1480 struct ust_reg_wait_queue wait_queue = {
1481 .count = 0,
1482 };
1483
1484 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1485
1486 health_code_update();
1487
1488 CDS_INIT_LIST_HEAD(&wait_queue.head);
1489
1490 DBG("[thread] Dispatch UST command started");
1491
1492 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1493 health_code_update();
1494
1495 /* Atomically prepare the queue futex */
1496 futex_nto1_prepare(&ust_cmd_queue.futex);
1497
1498 do {
1499 struct ust_app *app = NULL;
1500 ust_cmd = NULL;
1501
1502 /*
1503 * Make sure we don't have node(s) that have hung up before receiving
1504 * the notify socket. This is to clean the list in order to avoid
1505 * memory leaks from notify socket that are never seen.
1506 */
1507 sanitize_wait_queue(&wait_queue);
1508
1509 health_code_update();
1510 /* Dequeue command for registration */
1511 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1512 if (node == NULL) {
1513 DBG("Woken up but nothing in the UST command queue");
1514 /* Continue thread execution */
1515 break;
1516 }
1517
1518 ust_cmd = caa_container_of(node, struct ust_command, node);
1519
1520 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1521 " gid:%d sock:%d name:%s (version %d.%d)",
1522 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1523 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1524 ust_cmd->sock, ust_cmd->reg_msg.name,
1525 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1526
1527 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1528 wait_node = zmalloc(sizeof(*wait_node));
1529 if (!wait_node) {
1530 PERROR("zmalloc wait_node dispatch");
1531 ret = close(ust_cmd->sock);
1532 if (ret < 0) {
1533 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1534 }
1535 lttng_fd_put(1, LTTNG_FD_APPS);
1536 free(ust_cmd);
1537 goto error;
1538 }
1539 CDS_INIT_LIST_HEAD(&wait_node->head);
1540
1541 /* Create application object if socket is CMD. */
1542 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1543 ust_cmd->sock);
1544 if (!wait_node->app) {
1545 ret = close(ust_cmd->sock);
1546 if (ret < 0) {
1547 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1548 }
1549 lttng_fd_put(1, LTTNG_FD_APPS);
1550 free(wait_node);
1551 free(ust_cmd);
1552 continue;
1553 }
1554 /*
1555 * Add application to the wait queue so we can set the notify
1556 * socket before putting this object in the global ht.
1557 */
1558 cds_list_add(&wait_node->head, &wait_queue.head);
1559 wait_queue.count++;
1560
1561 free(ust_cmd);
1562 /*
1563 * We have to continue here since we don't have the notify
1564 * socket and the application MUST be added to the hash table
1565 * only at that moment.
1566 */
1567 continue;
1568 } else {
1569 /*
1570 * Look for the application in the local wait queue and set the
1571 * notify socket if found.
1572 */
1573 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1574 &wait_queue.head, head) {
1575 health_code_update();
1576 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1577 wait_node->app->notify_sock = ust_cmd->sock;
1578 cds_list_del(&wait_node->head);
1579 wait_queue.count--;
1580 app = wait_node->app;
1581 free(wait_node);
1582 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1583 break;
1584 }
1585 }
1586
1587 /*
1588 * With no application at this stage the received socket is
1589 * basically useless so close it before we free the cmd data
1590 * structure for good.
1591 */
1592 if (!app) {
1593 ret = close(ust_cmd->sock);
1594 if (ret < 0) {
1595 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1596 }
1597 lttng_fd_put(1, LTTNG_FD_APPS);
1598 }
1599 free(ust_cmd);
1600 }
1601
1602 if (app) {
1603 /*
1604 * @session_lock_list
1605 *
1606 * Lock the global session list so from the register up to the
1607 * registration done message, no thread can see the application
1608 * and change its state.
1609 */
1610 session_lock_list();
1611 rcu_read_lock();
1612
1613 /*
1614 * Add application to the global hash table. This needs to be
1615 * done before the update to the UST registry can locate the
1616 * application.
1617 */
1618 ust_app_add(app);
1619
1620 /* Set app version. This call will print an error if needed. */
1621 (void) ust_app_version(app);
1622
1623 /* Send notify socket through the notify pipe. */
1624 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1625 app->notify_sock);
1626 if (ret < 0) {
1627 rcu_read_unlock();
1628 session_unlock_list();
1629 /* No notify thread, stop the UST tracing. */
1630 goto error;
1631 }
1632
1633 /*
1634 * Update newly registered application with the tracing
1635 * registry info already enabled information.
1636 */
1637 update_ust_app(app->sock);
1638
1639 /*
1640 * Don't care about return value. Let the manage apps threads
1641 * handle app unregistration upon socket close.
1642 */
1643 (void) ust_app_register_done(app->sock);
1644
1645 /*
1646 * Even if the application socket has been closed, send the app
1647 * to the thread and unregistration will take place at that
1648 * place.
1649 */
1650 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1651 if (ret < 0) {
1652 rcu_read_unlock();
1653 session_unlock_list();
1654 /* No apps. thread, stop the UST tracing. */
1655 goto error;
1656 }
1657
1658 rcu_read_unlock();
1659 session_unlock_list();
1660 }
1661 } while (node != NULL);
1662
1663 health_poll_entry();
1664 /* Futex wait on queue. Blocking call on futex() */
1665 futex_nto1_wait(&ust_cmd_queue.futex);
1666 health_poll_exit();
1667 }
1668 /* Normal exit, no error */
1669 err = 0;
1670
1671 error:
1672 /* Clean up wait queue. */
1673 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1674 &wait_queue.head, head) {
1675 cds_list_del(&wait_node->head);
1676 wait_queue.count--;
1677 free(wait_node);
1678 }
1679
1680 DBG("Dispatch thread dying");
1681 if (err) {
1682 health_error();
1683 ERR("Health error occurred in %s", __func__);
1684 }
1685 health_unregister();
1686 return NULL;
1687 }
1688
1689 /*
1690 * This thread manage application registration.
1691 */
1692 static void *thread_registration_apps(void *data)
1693 {
1694 int sock = -1, i, ret, pollfd, err = -1;
1695 uint32_t revents, nb_fd;
1696 struct lttng_poll_event events;
1697 /*
1698 * Get allocated in this thread, enqueued to a global queue, dequeued and
1699 * freed in the manage apps thread.
1700 */
1701 struct ust_command *ust_cmd = NULL;
1702
1703 DBG("[thread] Manage application registration started");
1704
1705 health_register(HEALTH_TYPE_APP_REG);
1706
1707 if (testpoint(thread_registration_apps)) {
1708 goto error_testpoint;
1709 }
1710
1711 ret = lttcomm_listen_unix_sock(apps_sock);
1712 if (ret < 0) {
1713 goto error_listen;
1714 }
1715
1716 /*
1717 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1718 * more will be added to this poll set.
1719 */
1720 ret = sessiond_set_thread_pollset(&events, 2);
1721 if (ret < 0) {
1722 goto error_create_poll;
1723 }
1724
1725 /* Add the application registration socket */
1726 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1727 if (ret < 0) {
1728 goto error_poll_add;
1729 }
1730
1731 /* Notify all applications to register */
1732 ret = notify_ust_apps(1);
1733 if (ret < 0) {
1734 ERR("Failed to notify applications or create the wait shared memory.\n"
1735 "Execution continues but there might be problem for already\n"
1736 "running applications that wishes to register.");
1737 }
1738
1739 while (1) {
1740 DBG("Accepting application registration");
1741
1742 /* Inifinite blocking call, waiting for transmission */
1743 restart:
1744 health_poll_entry();
1745 ret = lttng_poll_wait(&events, -1);
1746 health_poll_exit();
1747 if (ret < 0) {
1748 /*
1749 * Restart interrupted system call.
1750 */
1751 if (errno == EINTR) {
1752 goto restart;
1753 }
1754 goto error;
1755 }
1756
1757 nb_fd = ret;
1758
1759 for (i = 0; i < nb_fd; i++) {
1760 health_code_update();
1761
1762 /* Fetch once the poll data */
1763 revents = LTTNG_POLL_GETEV(&events, i);
1764 pollfd = LTTNG_POLL_GETFD(&events, i);
1765
1766 /* Thread quit pipe has been closed. Killing thread. */
1767 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1768 if (ret) {
1769 err = 0;
1770 goto exit;
1771 }
1772
1773 /* Event on the registration socket */
1774 if (pollfd == apps_sock) {
1775 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1776 ERR("Register apps socket poll error");
1777 goto error;
1778 } else if (revents & LPOLLIN) {
1779 sock = lttcomm_accept_unix_sock(apps_sock);
1780 if (sock < 0) {
1781 goto error;
1782 }
1783
1784 /*
1785 * Set the CLOEXEC flag. Return code is useless because
1786 * either way, the show must go on.
1787 */
1788 (void) utils_set_fd_cloexec(sock);
1789
1790 /* Create UST registration command for enqueuing */
1791 ust_cmd = zmalloc(sizeof(struct ust_command));
1792 if (ust_cmd == NULL) {
1793 PERROR("ust command zmalloc");
1794 goto error;
1795 }
1796
1797 /*
1798 * Using message-based transmissions to ensure we don't
1799 * have to deal with partially received messages.
1800 */
1801 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1802 if (ret < 0) {
1803 ERR("Exhausted file descriptors allowed for applications.");
1804 free(ust_cmd);
1805 ret = close(sock);
1806 if (ret) {
1807 PERROR("close");
1808 }
1809 sock = -1;
1810 continue;
1811 }
1812
1813 health_code_update();
1814 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1815 if (ret < 0) {
1816 free(ust_cmd);
1817 /* Close socket of the application. */
1818 ret = close(sock);
1819 if (ret) {
1820 PERROR("close");
1821 }
1822 lttng_fd_put(LTTNG_FD_APPS, 1);
1823 sock = -1;
1824 continue;
1825 }
1826 health_code_update();
1827
1828 ust_cmd->sock = sock;
1829 sock = -1;
1830
1831 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1832 " gid:%d sock:%d name:%s (version %d.%d)",
1833 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1834 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1835 ust_cmd->sock, ust_cmd->reg_msg.name,
1836 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1837
1838 /*
1839 * Lock free enqueue the registration request. The red pill
1840 * has been taken! This apps will be part of the *system*.
1841 */
1842 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1843
1844 /*
1845 * Wake the registration queue futex. Implicit memory
1846 * barrier with the exchange in cds_wfq_enqueue.
1847 */
1848 futex_nto1_wake(&ust_cmd_queue.futex);
1849 }
1850 }
1851 }
1852 }
1853
1854 exit:
1855 error:
1856 if (err) {
1857 health_error();
1858 ERR("Health error occurred in %s", __func__);
1859 }
1860
1861 /* Notify that the registration thread is gone */
1862 notify_ust_apps(0);
1863
1864 if (apps_sock >= 0) {
1865 ret = close(apps_sock);
1866 if (ret) {
1867 PERROR("close");
1868 }
1869 }
1870 if (sock >= 0) {
1871 ret = close(sock);
1872 if (ret) {
1873 PERROR("close");
1874 }
1875 lttng_fd_put(LTTNG_FD_APPS, 1);
1876 }
1877 unlink(apps_unix_sock_path);
1878
1879 error_poll_add:
1880 lttng_poll_clean(&events);
1881 error_listen:
1882 error_create_poll:
1883 error_testpoint:
1884 DBG("UST Registration thread cleanup complete");
1885 health_unregister();
1886
1887 return NULL;
1888 }
1889
1890 /*
1891 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1892 * exec or it will fails.
1893 */
1894 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1895 {
1896 int ret, clock_ret;
1897 struct timespec timeout;
1898
1899 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1900 consumer_data->consumer_thread_is_ready = 0;
1901
1902 /* Setup pthread condition */
1903 ret = pthread_condattr_init(&consumer_data->condattr);
1904 if (ret != 0) {
1905 errno = ret;
1906 PERROR("pthread_condattr_init consumer data");
1907 goto error;
1908 }
1909
1910 /*
1911 * Set the monotonic clock in order to make sure we DO NOT jump in time
1912 * between the clock_gettime() call and the timedwait call. See bug #324
1913 * for a more details and how we noticed it.
1914 */
1915 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1916 if (ret != 0) {
1917 errno = ret;
1918 PERROR("pthread_condattr_setclock consumer data");
1919 goto error;
1920 }
1921
1922 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1923 if (ret != 0) {
1924 errno = ret;
1925 PERROR("pthread_cond_init consumer data");
1926 goto error;
1927 }
1928
1929 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1930 consumer_data);
1931 if (ret != 0) {
1932 PERROR("pthread_create consumer");
1933 ret = -1;
1934 goto error;
1935 }
1936
1937 /* We are about to wait on a pthread condition */
1938 pthread_mutex_lock(&consumer_data->cond_mutex);
1939
1940 /* Get time for sem_timedwait absolute timeout */
1941 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1942 /*
1943 * Set the timeout for the condition timed wait even if the clock gettime
1944 * call fails since we might loop on that call and we want to avoid to
1945 * increment the timeout too many times.
1946 */
1947 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1948
1949 /*
1950 * The following loop COULD be skipped in some conditions so this is why we
1951 * set ret to 0 in order to make sure at least one round of the loop is
1952 * done.
1953 */
1954 ret = 0;
1955
1956 /*
1957 * Loop until the condition is reached or when a timeout is reached. Note
1958 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1959 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1960 * possible. This loop does not take any chances and works with both of
1961 * them.
1962 */
1963 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1964 if (clock_ret < 0) {
1965 PERROR("clock_gettime spawn consumer");
1966 /* Infinite wait for the consumerd thread to be ready */
1967 ret = pthread_cond_wait(&consumer_data->cond,
1968 &consumer_data->cond_mutex);
1969 } else {
1970 ret = pthread_cond_timedwait(&consumer_data->cond,
1971 &consumer_data->cond_mutex, &timeout);
1972 }
1973 }
1974
1975 /* Release the pthread condition */
1976 pthread_mutex_unlock(&consumer_data->cond_mutex);
1977
1978 if (ret != 0) {
1979 errno = ret;
1980 if (ret == ETIMEDOUT) {
1981 /*
1982 * Call has timed out so we kill the kconsumerd_thread and return
1983 * an error.
1984 */
1985 ERR("Condition timed out. The consumer thread was never ready."
1986 " Killing it");
1987 ret = pthread_cancel(consumer_data->thread);
1988 if (ret < 0) {
1989 PERROR("pthread_cancel consumer thread");
1990 }
1991 } else {
1992 PERROR("pthread_cond_wait failed consumer thread");
1993 }
1994 goto error;
1995 }
1996
1997 pthread_mutex_lock(&consumer_data->pid_mutex);
1998 if (consumer_data->pid == 0) {
1999 ERR("Consumerd did not start");
2000 pthread_mutex_unlock(&consumer_data->pid_mutex);
2001 goto error;
2002 }
2003 pthread_mutex_unlock(&consumer_data->pid_mutex);
2004
2005 return 0;
2006
2007 error:
2008 return ret;
2009 }
2010
2011 /*
2012 * Join consumer thread
2013 */
2014 static int join_consumer_thread(struct consumer_data *consumer_data)
2015 {
2016 void *status;
2017
2018 /* Consumer pid must be a real one. */
2019 if (consumer_data->pid > 0) {
2020 int ret;
2021 ret = kill(consumer_data->pid, SIGTERM);
2022 if (ret) {
2023 ERR("Error killing consumer daemon");
2024 return ret;
2025 }
2026 return pthread_join(consumer_data->thread, &status);
2027 } else {
2028 return 0;
2029 }
2030 }
2031
2032 /*
2033 * Fork and exec a consumer daemon (consumerd).
2034 *
2035 * Return pid if successful else -1.
2036 */
2037 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2038 {
2039 int ret;
2040 pid_t pid;
2041 const char *consumer_to_use;
2042 const char *verbosity;
2043 struct stat st;
2044
2045 DBG("Spawning consumerd");
2046
2047 pid = fork();
2048 if (pid == 0) {
2049 /*
2050 * Exec consumerd.
2051 */
2052 if (opt_verbose_consumer) {
2053 verbosity = "--verbose";
2054 } else {
2055 verbosity = "--quiet";
2056 }
2057 switch (consumer_data->type) {
2058 case LTTNG_CONSUMER_KERNEL:
2059 /*
2060 * Find out which consumerd to execute. We will first try the
2061 * 64-bit path, then the sessiond's installation directory, and
2062 * fallback on the 32-bit one,
2063 */
2064 DBG3("Looking for a kernel consumer at these locations:");
2065 DBG3(" 1) %s", consumerd64_bin);
2066 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2067 DBG3(" 3) %s", consumerd32_bin);
2068 if (stat(consumerd64_bin, &st) == 0) {
2069 DBG3("Found location #1");
2070 consumer_to_use = consumerd64_bin;
2071 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2072 DBG3("Found location #2");
2073 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2074 } else if (stat(consumerd32_bin, &st) == 0) {
2075 DBG3("Found location #3");
2076 consumer_to_use = consumerd32_bin;
2077 } else {
2078 DBG("Could not find any valid consumerd executable");
2079 break;
2080 }
2081 DBG("Using kernel consumer at: %s", consumer_to_use);
2082 execl(consumer_to_use,
2083 "lttng-consumerd", verbosity, "-k",
2084 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2085 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2086 NULL);
2087 break;
2088 case LTTNG_CONSUMER64_UST:
2089 {
2090 char *tmpnew = NULL;
2091
2092 if (consumerd64_libdir[0] != '\0') {
2093 char *tmp;
2094 size_t tmplen;
2095
2096 tmp = getenv("LD_LIBRARY_PATH");
2097 if (!tmp) {
2098 tmp = "";
2099 }
2100 tmplen = strlen("LD_LIBRARY_PATH=")
2101 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2102 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2103 if (!tmpnew) {
2104 ret = -ENOMEM;
2105 goto error;
2106 }
2107 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2108 strcat(tmpnew, consumerd64_libdir);
2109 if (tmp[0] != '\0') {
2110 strcat(tmpnew, ":");
2111 strcat(tmpnew, tmp);
2112 }
2113 ret = putenv(tmpnew);
2114 if (ret) {
2115 ret = -errno;
2116 free(tmpnew);
2117 goto error;
2118 }
2119 }
2120 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2121 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2122 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2123 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2124 NULL);
2125 if (consumerd64_libdir[0] != '\0') {
2126 free(tmpnew);
2127 }
2128 if (ret) {
2129 goto error;
2130 }
2131 break;
2132 }
2133 case LTTNG_CONSUMER32_UST:
2134 {
2135 char *tmpnew = NULL;
2136
2137 if (consumerd32_libdir[0] != '\0') {
2138 char *tmp;
2139 size_t tmplen;
2140
2141 tmp = getenv("LD_LIBRARY_PATH");
2142 if (!tmp) {
2143 tmp = "";
2144 }
2145 tmplen = strlen("LD_LIBRARY_PATH=")
2146 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2147 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2148 if (!tmpnew) {
2149 ret = -ENOMEM;
2150 goto error;
2151 }
2152 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2153 strcat(tmpnew, consumerd32_libdir);
2154 if (tmp[0] != '\0') {
2155 strcat(tmpnew, ":");
2156 strcat(tmpnew, tmp);
2157 }
2158 ret = putenv(tmpnew);
2159 if (ret) {
2160 ret = -errno;
2161 free(tmpnew);
2162 goto error;
2163 }
2164 }
2165 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2166 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2167 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2168 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2169 NULL);
2170 if (consumerd32_libdir[0] != '\0') {
2171 free(tmpnew);
2172 }
2173 if (ret) {
2174 goto error;
2175 }
2176 break;
2177 }
2178 default:
2179 PERROR("unknown consumer type");
2180 exit(EXIT_FAILURE);
2181 }
2182 if (errno != 0) {
2183 PERROR("kernel start consumer exec");
2184 }
2185 exit(EXIT_FAILURE);
2186 } else if (pid > 0) {
2187 ret = pid;
2188 } else {
2189 PERROR("start consumer fork");
2190 ret = -errno;
2191 }
2192 error:
2193 return ret;
2194 }
2195
2196 /*
2197 * Spawn the consumerd daemon and session daemon thread.
2198 */
2199 static int start_consumerd(struct consumer_data *consumer_data)
2200 {
2201 int ret;
2202
2203 /*
2204 * Set the listen() state on the socket since there is a possible race
2205 * between the exec() of the consumer daemon and this call if place in the
2206 * consumer thread. See bug #366 for more details.
2207 */
2208 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2209 if (ret < 0) {
2210 goto error;
2211 }
2212
2213 pthread_mutex_lock(&consumer_data->pid_mutex);
2214 if (consumer_data->pid != 0) {
2215 pthread_mutex_unlock(&consumer_data->pid_mutex);
2216 goto end;
2217 }
2218
2219 ret = spawn_consumerd(consumer_data);
2220 if (ret < 0) {
2221 ERR("Spawning consumerd failed");
2222 pthread_mutex_unlock(&consumer_data->pid_mutex);
2223 goto error;
2224 }
2225
2226 /* Setting up the consumer_data pid */
2227 consumer_data->pid = ret;
2228 DBG2("Consumer pid %d", consumer_data->pid);
2229 pthread_mutex_unlock(&consumer_data->pid_mutex);
2230
2231 DBG2("Spawning consumer control thread");
2232 ret = spawn_consumer_thread(consumer_data);
2233 if (ret < 0) {
2234 ERR("Fatal error spawning consumer control thread");
2235 goto error;
2236 }
2237
2238 end:
2239 return 0;
2240
2241 error:
2242 /* Cleanup already created sockets on error. */
2243 if (consumer_data->err_sock >= 0) {
2244 int err;
2245
2246 err = close(consumer_data->err_sock);
2247 if (err < 0) {
2248 PERROR("close consumer data error socket");
2249 }
2250 }
2251 return ret;
2252 }
2253
2254 /*
2255 * Compute health status of each consumer. If one of them is zero (bad
2256 * state), we return 0.
2257 */
2258 static int check_consumer_health(void)
2259 {
2260 int ret;
2261
2262 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2263
2264 DBG3("Health consumer check %d", ret);
2265
2266 return ret;
2267 }
2268
2269 /*
2270 * Setup necessary data for kernel tracer action.
2271 */
2272 static int init_kernel_tracer(void)
2273 {
2274 int ret;
2275
2276 /* Modprobe lttng kernel modules */
2277 ret = modprobe_lttng_control();
2278 if (ret < 0) {
2279 goto error;
2280 }
2281
2282 /* Open debugfs lttng */
2283 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2284 if (kernel_tracer_fd < 0) {
2285 DBG("Failed to open %s", module_proc_lttng);
2286 ret = -1;
2287 goto error_open;
2288 }
2289
2290 /* Validate kernel version */
2291 ret = kernel_validate_version(kernel_tracer_fd);
2292 if (ret < 0) {
2293 goto error_version;
2294 }
2295
2296 ret = modprobe_lttng_data();
2297 if (ret < 0) {
2298 goto error_modules;
2299 }
2300
2301 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2302 return 0;
2303
2304 error_version:
2305 modprobe_remove_lttng_control();
2306 ret = close(kernel_tracer_fd);
2307 if (ret) {
2308 PERROR("close");
2309 }
2310 kernel_tracer_fd = -1;
2311 return LTTNG_ERR_KERN_VERSION;
2312
2313 error_modules:
2314 ret = close(kernel_tracer_fd);
2315 if (ret) {
2316 PERROR("close");
2317 }
2318
2319 error_open:
2320 modprobe_remove_lttng_control();
2321
2322 error:
2323 WARN("No kernel tracer available");
2324 kernel_tracer_fd = -1;
2325 if (!is_root) {
2326 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2327 } else {
2328 return LTTNG_ERR_KERN_NA;
2329 }
2330 }
2331
2332
2333 /*
2334 * Copy consumer output from the tracing session to the domain session. The
2335 * function also applies the right modification on a per domain basis for the
2336 * trace files destination directory.
2337 *
2338 * Should *NOT* be called with RCU read-side lock held.
2339 */
2340 static int copy_session_consumer(int domain, struct ltt_session *session)
2341 {
2342 int ret;
2343 const char *dir_name;
2344 struct consumer_output *consumer;
2345
2346 assert(session);
2347 assert(session->consumer);
2348
2349 switch (domain) {
2350 case LTTNG_DOMAIN_KERNEL:
2351 DBG3("Copying tracing session consumer output in kernel session");
2352 /*
2353 * XXX: We should audit the session creation and what this function
2354 * does "extra" in order to avoid a destroy since this function is used
2355 * in the domain session creation (kernel and ust) only. Same for UST
2356 * domain.
2357 */
2358 if (session->kernel_session->consumer) {
2359 consumer_destroy_output(session->kernel_session->consumer);
2360 }
2361 session->kernel_session->consumer =
2362 consumer_copy_output(session->consumer);
2363 /* Ease our life a bit for the next part */
2364 consumer = session->kernel_session->consumer;
2365 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2366 break;
2367 case LTTNG_DOMAIN_UST:
2368 DBG3("Copying tracing session consumer output in UST session");
2369 if (session->ust_session->consumer) {
2370 consumer_destroy_output(session->ust_session->consumer);
2371 }
2372 session->ust_session->consumer =
2373 consumer_copy_output(session->consumer);
2374 /* Ease our life a bit for the next part */
2375 consumer = session->ust_session->consumer;
2376 dir_name = DEFAULT_UST_TRACE_DIR;
2377 break;
2378 default:
2379 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2380 goto error;
2381 }
2382
2383 /* Append correct directory to subdir */
2384 strncat(consumer->subdir, dir_name,
2385 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2386 DBG3("Copy session consumer subdir %s", consumer->subdir);
2387
2388 ret = LTTNG_OK;
2389
2390 error:
2391 return ret;
2392 }
2393
2394 /*
2395 * Create an UST session and add it to the session ust list.
2396 *
2397 * Should *NOT* be called with RCU read-side lock held.
2398 */
2399 static int create_ust_session(struct ltt_session *session,
2400 struct lttng_domain *domain)
2401 {
2402 int ret;
2403 struct ltt_ust_session *lus = NULL;
2404
2405 assert(session);
2406 assert(domain);
2407 assert(session->consumer);
2408
2409 switch (domain->type) {
2410 case LTTNG_DOMAIN_UST:
2411 break;
2412 default:
2413 ERR("Unknown UST domain on create session %d", domain->type);
2414 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2415 goto error;
2416 }
2417
2418 DBG("Creating UST session");
2419
2420 lus = trace_ust_create_session(session->id);
2421 if (lus == NULL) {
2422 ret = LTTNG_ERR_UST_SESS_FAIL;
2423 goto error;
2424 }
2425
2426 lus->uid = session->uid;
2427 lus->gid = session->gid;
2428 lus->output_traces = session->output_traces;
2429 session->ust_session = lus;
2430
2431 /* Copy session output to the newly created UST session */
2432 ret = copy_session_consumer(domain->type, session);
2433 if (ret != LTTNG_OK) {
2434 goto error;
2435 }
2436
2437 return LTTNG_OK;
2438
2439 error:
2440 free(lus);
2441 session->ust_session = NULL;
2442 return ret;
2443 }
2444
2445 /*
2446 * Create a kernel tracer session then create the default channel.
2447 */
2448 static int create_kernel_session(struct ltt_session *session)
2449 {
2450 int ret;
2451
2452 DBG("Creating kernel session");
2453
2454 ret = kernel_create_session(session, kernel_tracer_fd);
2455 if (ret < 0) {
2456 ret = LTTNG_ERR_KERN_SESS_FAIL;
2457 goto error;
2458 }
2459
2460 /* Code flow safety */
2461 assert(session->kernel_session);
2462
2463 /* Copy session output to the newly created Kernel session */
2464 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2465 if (ret != LTTNG_OK) {
2466 goto error;
2467 }
2468
2469 /* Create directory(ies) on local filesystem. */
2470 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2471 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2472 ret = run_as_mkdir_recursive(
2473 session->kernel_session->consumer->dst.trace_path,
2474 S_IRWXU | S_IRWXG, session->uid, session->gid);
2475 if (ret < 0) {
2476 if (ret != -EEXIST) {
2477 ERR("Trace directory creation error");
2478 goto error;
2479 }
2480 }
2481 }
2482
2483 session->kernel_session->uid = session->uid;
2484 session->kernel_session->gid = session->gid;
2485 session->kernel_session->output_traces = session->output_traces;
2486
2487 return LTTNG_OK;
2488
2489 error:
2490 trace_kernel_destroy_session(session->kernel_session);
2491 session->kernel_session = NULL;
2492 return ret;
2493 }
2494
2495 /*
2496 * Count number of session permitted by uid/gid.
2497 */
2498 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2499 {
2500 unsigned int i = 0;
2501 struct ltt_session *session;
2502
2503 DBG("Counting number of available session for UID %d GID %d",
2504 uid, gid);
2505 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2506 /*
2507 * Only list the sessions the user can control.
2508 */
2509 if (!session_access_ok(session, uid, gid)) {
2510 continue;
2511 }
2512 i++;
2513 }
2514 return i;
2515 }
2516
2517 /*
2518 * Process the command requested by the lttng client within the command
2519 * context structure. This function make sure that the return structure (llm)
2520 * is set and ready for transmission before returning.
2521 *
2522 * Return any error encountered or 0 for success.
2523 *
2524 * "sock" is only used for special-case var. len data.
2525 *
2526 * Should *NOT* be called with RCU read-side lock held.
2527 */
2528 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2529 int *sock_error)
2530 {
2531 int ret = LTTNG_OK;
2532 int need_tracing_session = 1;
2533 int need_domain;
2534
2535 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2536
2537 *sock_error = 0;
2538
2539 switch (cmd_ctx->lsm->cmd_type) {
2540 case LTTNG_CREATE_SESSION:
2541 case LTTNG_DESTROY_SESSION:
2542 case LTTNG_LIST_SESSIONS:
2543 case LTTNG_LIST_DOMAINS:
2544 case LTTNG_START_TRACE:
2545 case LTTNG_STOP_TRACE:
2546 case LTTNG_DATA_PENDING:
2547 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2548 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2549 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2550 case LTTNG_SNAPSHOT_RECORD:
2551 need_domain = 0;
2552 break;
2553 default:
2554 need_domain = 1;
2555 }
2556
2557 if (opt_no_kernel && need_domain
2558 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2559 if (!is_root) {
2560 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2561 } else {
2562 ret = LTTNG_ERR_KERN_NA;
2563 }
2564 goto error;
2565 }
2566
2567 /* Deny register consumer if we already have a spawned consumer. */
2568 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2569 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2570 if (kconsumer_data.pid > 0) {
2571 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2572 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2573 goto error;
2574 }
2575 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2576 }
2577
2578 /*
2579 * Check for command that don't needs to allocate a returned payload. We do
2580 * this here so we don't have to make the call for no payload at each
2581 * command.
2582 */
2583 switch(cmd_ctx->lsm->cmd_type) {
2584 case LTTNG_LIST_SESSIONS:
2585 case LTTNG_LIST_TRACEPOINTS:
2586 case LTTNG_LIST_TRACEPOINT_FIELDS:
2587 case LTTNG_LIST_DOMAINS:
2588 case LTTNG_LIST_CHANNELS:
2589 case LTTNG_LIST_EVENTS:
2590 break;
2591 default:
2592 /* Setup lttng message with no payload */
2593 ret = setup_lttng_msg(cmd_ctx, 0);
2594 if (ret < 0) {
2595 /* This label does not try to unlock the session */
2596 goto init_setup_error;
2597 }
2598 }
2599
2600 /* Commands that DO NOT need a session. */
2601 switch (cmd_ctx->lsm->cmd_type) {
2602 case LTTNG_CREATE_SESSION:
2603 case LTTNG_CALIBRATE:
2604 case LTTNG_LIST_SESSIONS:
2605 case LTTNG_LIST_TRACEPOINTS:
2606 case LTTNG_LIST_TRACEPOINT_FIELDS:
2607 need_tracing_session = 0;
2608 break;
2609 default:
2610 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2611 /*
2612 * We keep the session list lock across _all_ commands
2613 * for now, because the per-session lock does not
2614 * handle teardown properly.
2615 */
2616 session_lock_list();
2617 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2618 if (cmd_ctx->session == NULL) {
2619 ret = LTTNG_ERR_SESS_NOT_FOUND;
2620 goto error;
2621 } else {
2622 /* Acquire lock for the session */
2623 session_lock(cmd_ctx->session);
2624 }
2625 break;
2626 }
2627
2628 if (!need_domain) {
2629 goto skip_domain;
2630 }
2631
2632 /*
2633 * Check domain type for specific "pre-action".
2634 */
2635 switch (cmd_ctx->lsm->domain.type) {
2636 case LTTNG_DOMAIN_KERNEL:
2637 if (!is_root) {
2638 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2639 goto error;
2640 }
2641
2642 /* Kernel tracer check */
2643 if (kernel_tracer_fd == -1) {
2644 /* Basically, load kernel tracer modules */
2645 ret = init_kernel_tracer();
2646 if (ret != 0) {
2647 goto error;
2648 }
2649 }
2650
2651 /* Consumer is in an ERROR state. Report back to client */
2652 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2653 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2654 goto error;
2655 }
2656
2657 /* Need a session for kernel command */
2658 if (need_tracing_session) {
2659 if (cmd_ctx->session->kernel_session == NULL) {
2660 ret = create_kernel_session(cmd_ctx->session);
2661 if (ret < 0) {
2662 ret = LTTNG_ERR_KERN_SESS_FAIL;
2663 goto error;
2664 }
2665 }
2666
2667 /* Start the kernel consumer daemon */
2668 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2669 if (kconsumer_data.pid == 0 &&
2670 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2671 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2672 ret = start_consumerd(&kconsumer_data);
2673 if (ret < 0) {
2674 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2675 goto error;
2676 }
2677 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2678 } else {
2679 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2680 }
2681
2682 /*
2683 * The consumer was just spawned so we need to add the socket to
2684 * the consumer output of the session if exist.
2685 */
2686 ret = consumer_create_socket(&kconsumer_data,
2687 cmd_ctx->session->kernel_session->consumer);
2688 if (ret < 0) {
2689 goto error;
2690 }
2691 }
2692
2693 break;
2694 case LTTNG_DOMAIN_UST:
2695 {
2696 /* Consumer is in an ERROR state. Report back to client */
2697 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2698 ret = LTTNG_ERR_NO_USTCONSUMERD;
2699 goto error;
2700 }
2701
2702 if (need_tracing_session) {
2703 /* Create UST session if none exist. */
2704 if (cmd_ctx->session->ust_session == NULL) {
2705 ret = create_ust_session(cmd_ctx->session,
2706 &cmd_ctx->lsm->domain);
2707 if (ret != LTTNG_OK) {
2708 goto error;
2709 }
2710 }
2711
2712 /* Start the UST consumer daemons */
2713 /* 64-bit */
2714 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2715 if (consumerd64_bin[0] != '\0' &&
2716 ustconsumer64_data.pid == 0 &&
2717 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2718 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2719 ret = start_consumerd(&ustconsumer64_data);
2720 if (ret < 0) {
2721 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2722 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2723 goto error;
2724 }
2725
2726 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2727 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2728 } else {
2729 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2730 }
2731
2732 /*
2733 * Setup socket for consumer 64 bit. No need for atomic access
2734 * since it was set above and can ONLY be set in this thread.
2735 */
2736 ret = consumer_create_socket(&ustconsumer64_data,
2737 cmd_ctx->session->ust_session->consumer);
2738 if (ret < 0) {
2739 goto error;
2740 }
2741
2742 /* 32-bit */
2743 if (consumerd32_bin[0] != '\0' &&
2744 ustconsumer32_data.pid == 0 &&
2745 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2746 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2747 ret = start_consumerd(&ustconsumer32_data);
2748 if (ret < 0) {
2749 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2750 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2751 goto error;
2752 }
2753
2754 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2755 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2756 } else {
2757 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2758 }
2759
2760 /*
2761 * Setup socket for consumer 64 bit. No need for atomic access
2762 * since it was set above and can ONLY be set in this thread.
2763 */
2764 ret = consumer_create_socket(&ustconsumer32_data,
2765 cmd_ctx->session->ust_session->consumer);
2766 if (ret < 0) {
2767 goto error;
2768 }
2769 }
2770 break;
2771 }
2772 default:
2773 break;
2774 }
2775 skip_domain:
2776
2777 /* Validate consumer daemon state when start/stop trace command */
2778 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2779 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2780 switch (cmd_ctx->lsm->domain.type) {
2781 case LTTNG_DOMAIN_UST:
2782 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2783 ret = LTTNG_ERR_NO_USTCONSUMERD;
2784 goto error;
2785 }
2786 break;
2787 case LTTNG_DOMAIN_KERNEL:
2788 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2789 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2790 goto error;
2791 }
2792 break;
2793 }
2794 }
2795
2796 /*
2797 * Check that the UID or GID match that of the tracing session.
2798 * The root user can interact with all sessions.
2799 */
2800 if (need_tracing_session) {
2801 if (!session_access_ok(cmd_ctx->session,
2802 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2803 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2804 ret = LTTNG_ERR_EPERM;
2805 goto error;
2806 }
2807 }
2808
2809 /*
2810 * Send relayd information to consumer as soon as we have a domain and a
2811 * session defined.
2812 */
2813 if (cmd_ctx->session && need_domain) {
2814 /*
2815 * Setup relayd if not done yet. If the relayd information was already
2816 * sent to the consumer, this call will gracefully return.
2817 */
2818 ret = cmd_setup_relayd(cmd_ctx->session);
2819 if (ret != LTTNG_OK) {
2820 goto error;
2821 }
2822 }
2823
2824 /* Process by command type */
2825 switch (cmd_ctx->lsm->cmd_type) {
2826 case LTTNG_ADD_CONTEXT:
2827 {
2828 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2829 cmd_ctx->lsm->u.context.channel_name,
2830 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2831 break;
2832 }
2833 case LTTNG_DISABLE_CHANNEL:
2834 {
2835 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2836 cmd_ctx->lsm->u.disable.channel_name);
2837 break;
2838 }
2839 case LTTNG_DISABLE_EVENT:
2840 {
2841 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2842 cmd_ctx->lsm->u.disable.channel_name,
2843 cmd_ctx->lsm->u.disable.name);
2844 break;
2845 }
2846 case LTTNG_DISABLE_ALL_EVENT:
2847 {
2848 DBG("Disabling all events");
2849
2850 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2851 cmd_ctx->lsm->u.disable.channel_name);
2852 break;
2853 }
2854 case LTTNG_ENABLE_CHANNEL:
2855 {
2856 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2857 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2858 break;
2859 }
2860 case LTTNG_ENABLE_EVENT:
2861 {
2862 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2863 cmd_ctx->lsm->u.enable.channel_name,
2864 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2865 break;
2866 }
2867 case LTTNG_ENABLE_ALL_EVENT:
2868 {
2869 DBG("Enabling all events");
2870
2871 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2872 cmd_ctx->lsm->u.enable.channel_name,
2873 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2874 break;
2875 }
2876 case LTTNG_LIST_TRACEPOINTS:
2877 {
2878 struct lttng_event *events;
2879 ssize_t nb_events;
2880
2881 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2882 if (nb_events < 0) {
2883 /* Return value is a negative lttng_error_code. */
2884 ret = -nb_events;
2885 goto error;
2886 }
2887
2888 /*
2889 * Setup lttng message with payload size set to the event list size in
2890 * bytes and then copy list into the llm payload.
2891 */
2892 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2893 if (ret < 0) {
2894 free(events);
2895 goto setup_error;
2896 }
2897
2898 /* Copy event list into message payload */
2899 memcpy(cmd_ctx->llm->payload, events,
2900 sizeof(struct lttng_event) * nb_events);
2901
2902 free(events);
2903
2904 ret = LTTNG_OK;
2905 break;
2906 }
2907 case LTTNG_LIST_TRACEPOINT_FIELDS:
2908 {
2909 struct lttng_event_field *fields;
2910 ssize_t nb_fields;
2911
2912 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2913 &fields);
2914 if (nb_fields < 0) {
2915 /* Return value is a negative lttng_error_code. */
2916 ret = -nb_fields;
2917 goto error;
2918 }
2919
2920 /*
2921 * Setup lttng message with payload size set to the event list size in
2922 * bytes and then copy list into the llm payload.
2923 */
2924 ret = setup_lttng_msg(cmd_ctx,
2925 sizeof(struct lttng_event_field) * nb_fields);
2926 if (ret < 0) {
2927 free(fields);
2928 goto setup_error;
2929 }
2930
2931 /* Copy event list into message payload */
2932 memcpy(cmd_ctx->llm->payload, fields,
2933 sizeof(struct lttng_event_field) * nb_fields);
2934
2935 free(fields);
2936
2937 ret = LTTNG_OK;
2938 break;
2939 }
2940 case LTTNG_SET_CONSUMER_URI:
2941 {
2942 size_t nb_uri, len;
2943 struct lttng_uri *uris;
2944
2945 nb_uri = cmd_ctx->lsm->u.uri.size;
2946 len = nb_uri * sizeof(struct lttng_uri);
2947
2948 if (nb_uri == 0) {
2949 ret = LTTNG_ERR_INVALID;
2950 goto error;
2951 }
2952
2953 uris = zmalloc(len);
2954 if (uris == NULL) {
2955 ret = LTTNG_ERR_FATAL;
2956 goto error;
2957 }
2958
2959 /* Receive variable len data */
2960 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2961 ret = lttcomm_recv_unix_sock(sock, uris, len);
2962 if (ret <= 0) {
2963 DBG("No URIs received from client... continuing");
2964 *sock_error = 1;
2965 ret = LTTNG_ERR_SESSION_FAIL;
2966 free(uris);
2967 goto error;
2968 }
2969
2970 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2971 nb_uri, uris);
2972 if (ret != LTTNG_OK) {
2973 free(uris);
2974 goto error;
2975 }
2976
2977 /*
2978 * XXX: 0 means that this URI should be applied on the session. Should
2979 * be a DOMAIN enuam.
2980 */
2981 if (cmd_ctx->lsm->domain.type == 0) {
2982 /* Add the URI for the UST session if a consumer is present. */
2983 if (cmd_ctx->session->ust_session &&
2984 cmd_ctx->session->ust_session->consumer) {
2985 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2986 nb_uri, uris);
2987 } else if (cmd_ctx->session->kernel_session &&
2988 cmd_ctx->session->kernel_session->consumer) {
2989 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2990 cmd_ctx->session, nb_uri, uris);
2991 }
2992 }
2993
2994 free(uris);
2995
2996 break;
2997 }
2998 case LTTNG_START_TRACE:
2999 {
3000 ret = cmd_start_trace(cmd_ctx->session);
3001 break;
3002 }
3003 case LTTNG_STOP_TRACE:
3004 {
3005 ret = cmd_stop_trace(cmd_ctx->session);
3006 break;
3007 }
3008 case LTTNG_CREATE_SESSION:
3009 {
3010 size_t nb_uri, len;
3011 struct lttng_uri *uris = NULL;
3012
3013 nb_uri = cmd_ctx->lsm->u.uri.size;
3014 len = nb_uri * sizeof(struct lttng_uri);
3015
3016 if (nb_uri > 0) {
3017 uris = zmalloc(len);
3018 if (uris == NULL) {
3019 ret = LTTNG_ERR_FATAL;
3020 goto error;
3021 }
3022
3023 /* Receive variable len data */
3024 DBG("Waiting for %zu URIs from client ...", nb_uri);
3025 ret = lttcomm_recv_unix_sock(sock, uris, len);
3026 if (ret <= 0) {
3027 DBG("No URIs received from client... continuing");
3028 *sock_error = 1;
3029 ret = LTTNG_ERR_SESSION_FAIL;
3030 free(uris);
3031 goto error;
3032 }
3033
3034 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3035 DBG("Creating session with ONE network URI is a bad call");
3036 ret = LTTNG_ERR_SESSION_FAIL;
3037 free(uris);
3038 goto error;
3039 }
3040 }
3041
3042 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3043 &cmd_ctx->creds);
3044
3045 free(uris);
3046
3047 break;
3048 }
3049 case LTTNG_DESTROY_SESSION:
3050 {
3051 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3052
3053 /* Set session to NULL so we do not unlock it after free. */
3054 cmd_ctx->session = NULL;
3055 break;
3056 }
3057 case LTTNG_LIST_DOMAINS:
3058 {
3059 ssize_t nb_dom;
3060 struct lttng_domain *domains;
3061
3062 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3063 if (nb_dom < 0) {
3064 /* Return value is a negative lttng_error_code. */
3065 ret = -nb_dom;
3066 goto error;
3067 }
3068
3069 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3070 if (ret < 0) {
3071 free(domains);
3072 goto setup_error;
3073 }
3074
3075 /* Copy event list into message payload */
3076 memcpy(cmd_ctx->llm->payload, domains,
3077 nb_dom * sizeof(struct lttng_domain));
3078
3079 free(domains);
3080
3081 ret = LTTNG_OK;
3082 break;
3083 }
3084 case LTTNG_LIST_CHANNELS:
3085 {
3086 int nb_chan;
3087 struct lttng_channel *channels;
3088
3089 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3090 cmd_ctx->session, &channels);
3091 if (nb_chan < 0) {
3092 /* Return value is a negative lttng_error_code. */
3093 ret = -nb_chan;
3094 goto error;
3095 }
3096
3097 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3098 if (ret < 0) {
3099 free(channels);
3100 goto setup_error;
3101 }
3102
3103 /* Copy event list into message payload */
3104 memcpy(cmd_ctx->llm->payload, channels,
3105 nb_chan * sizeof(struct lttng_channel));
3106
3107 free(channels);
3108
3109 ret = LTTNG_OK;
3110 break;
3111 }
3112 case LTTNG_LIST_EVENTS:
3113 {
3114 ssize_t nb_event;
3115 struct lttng_event *events = NULL;
3116
3117 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3118 cmd_ctx->lsm->u.list.channel_name, &events);
3119 if (nb_event < 0) {
3120 /* Return value is a negative lttng_error_code. */
3121 ret = -nb_event;
3122 goto error;
3123 }
3124
3125 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3126 if (ret < 0) {
3127 free(events);
3128 goto setup_error;
3129 }
3130
3131 /* Copy event list into message payload */
3132 memcpy(cmd_ctx->llm->payload, events,
3133 nb_event * sizeof(struct lttng_event));
3134
3135 free(events);
3136
3137 ret = LTTNG_OK;
3138 break;
3139 }
3140 case LTTNG_LIST_SESSIONS:
3141 {
3142 unsigned int nr_sessions;
3143
3144 session_lock_list();
3145 nr_sessions = lttng_sessions_count(
3146 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3147 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3148
3149 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3150 if (ret < 0) {
3151 session_unlock_list();
3152 goto setup_error;
3153 }
3154
3155 /* Filled the session array */
3156 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3157 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3158 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3159
3160 session_unlock_list();
3161
3162 ret = LTTNG_OK;
3163 break;
3164 }
3165 case LTTNG_CALIBRATE:
3166 {
3167 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3168 &cmd_ctx->lsm->u.calibrate);
3169 break;
3170 }
3171 case LTTNG_REGISTER_CONSUMER:
3172 {
3173 struct consumer_data *cdata;
3174
3175 switch (cmd_ctx->lsm->domain.type) {
3176 case LTTNG_DOMAIN_KERNEL:
3177 cdata = &kconsumer_data;
3178 break;
3179 default:
3180 ret = LTTNG_ERR_UND;
3181 goto error;
3182 }
3183
3184 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3185 cmd_ctx->lsm->u.reg.path, cdata);
3186 break;
3187 }
3188 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3189 {
3190 struct lttng_filter_bytecode *bytecode;
3191
3192 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3193 ret = LTTNG_ERR_FILTER_INVAL;
3194 goto error;
3195 }
3196 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3197 ret = LTTNG_ERR_FILTER_INVAL;
3198 goto error;
3199 }
3200 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3201 if (!bytecode) {
3202 ret = LTTNG_ERR_FILTER_NOMEM;
3203 goto error;
3204 }
3205 /* Receive var. len. data */
3206 DBG("Receiving var len data from client ...");
3207 ret = lttcomm_recv_unix_sock(sock, bytecode,
3208 cmd_ctx->lsm->u.enable.bytecode_len);
3209 if (ret <= 0) {
3210 DBG("Nothing recv() from client var len data... continuing");
3211 *sock_error = 1;
3212 ret = LTTNG_ERR_FILTER_INVAL;
3213 goto error;
3214 }
3215
3216 if (bytecode->len + sizeof(*bytecode)
3217 != cmd_ctx->lsm->u.enable.bytecode_len) {
3218 free(bytecode);
3219 ret = LTTNG_ERR_FILTER_INVAL;
3220 goto error;
3221 }
3222
3223 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3224 cmd_ctx->lsm->u.enable.channel_name,
3225 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3226 break;
3227 }
3228 case LTTNG_DATA_PENDING:
3229 {
3230 ret = cmd_data_pending(cmd_ctx->session);
3231 break;
3232 }
3233 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3234 {
3235 struct lttcomm_lttng_output_id reply;
3236
3237 ret = cmd_snapshot_add_output(cmd_ctx->session,
3238 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3239 if (ret != LTTNG_OK) {
3240 goto error;
3241 }
3242
3243 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3244 if (ret < 0) {
3245 goto setup_error;
3246 }
3247
3248 /* Copy output list into message payload */
3249 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3250 ret = LTTNG_OK;
3251 break;
3252 }
3253 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3254 {
3255 ret = cmd_snapshot_del_output(cmd_ctx->session,
3256 &cmd_ctx->lsm->u.snapshot_output.output);
3257 break;
3258 }
3259 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3260 {
3261 ssize_t nb_output;
3262 struct lttng_snapshot_output *outputs = NULL;
3263
3264 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3265 if (nb_output < 0) {
3266 ret = -nb_output;
3267 goto error;
3268 }
3269
3270 ret = setup_lttng_msg(cmd_ctx,
3271 nb_output * sizeof(struct lttng_snapshot_output));
3272 if (ret < 0) {
3273 free(outputs);
3274 goto setup_error;
3275 }
3276
3277 if (outputs) {
3278 /* Copy output list into message payload */
3279 memcpy(cmd_ctx->llm->payload, outputs,
3280 nb_output * sizeof(struct lttng_snapshot_output));
3281 free(outputs);
3282 }
3283
3284 ret = LTTNG_OK;
3285 break;
3286 }
3287 case LTTNG_SNAPSHOT_RECORD:
3288 {
3289 ret = cmd_snapshot_record(cmd_ctx->session,
3290 &cmd_ctx->lsm->u.snapshot_record.output,
3291 cmd_ctx->lsm->u.snapshot_record.wait);
3292 break;
3293 }
3294 default:
3295 ret = LTTNG_ERR_UND;
3296 break;
3297 }
3298
3299 error:
3300 if (cmd_ctx->llm == NULL) {
3301 DBG("Missing llm structure. Allocating one.");
3302 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3303 goto setup_error;
3304 }
3305 }
3306 /* Set return code */
3307 cmd_ctx->llm->ret_code = ret;
3308 setup_error:
3309 if (cmd_ctx->session) {
3310 session_unlock(cmd_ctx->session);
3311 }
3312 if (need_tracing_session) {
3313 session_unlock_list();
3314 }
3315 init_setup_error:
3316 return ret;
3317 }
3318
3319 /*
3320 * Thread managing health check socket.
3321 */
3322 static void *thread_manage_health(void *data)
3323 {
3324 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3325 uint32_t revents, nb_fd;
3326 struct lttng_poll_event events;
3327 struct lttcomm_health_msg msg;
3328 struct lttcomm_health_data reply;
3329
3330 DBG("[thread] Manage health check started");
3331
3332 rcu_register_thread();
3333
3334 /* We might hit an error path before this is created. */
3335 lttng_poll_init(&events);
3336
3337 /* Create unix socket */
3338 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3339 if (sock < 0) {
3340 ERR("Unable to create health check Unix socket");
3341 ret = -1;
3342 goto error;
3343 }
3344
3345 /*
3346 * Set the CLOEXEC flag. Return code is useless because either way, the
3347 * show must go on.
3348 */
3349 (void) utils_set_fd_cloexec(sock);
3350
3351 ret = lttcomm_listen_unix_sock(sock);
3352 if (ret < 0) {
3353 goto error;
3354 }
3355
3356 /*
3357 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3358 * more will be added to this poll set.
3359 */
3360 ret = sessiond_set_thread_pollset(&events, 2);
3361 if (ret < 0) {
3362 goto error;
3363 }
3364
3365 /* Add the application registration socket */
3366 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3367 if (ret < 0) {
3368 goto error;
3369 }
3370
3371 while (1) {
3372 DBG("Health check ready");
3373
3374 /* Inifinite blocking call, waiting for transmission */
3375 restart:
3376 ret = lttng_poll_wait(&events, -1);
3377 if (ret < 0) {
3378 /*
3379 * Restart interrupted system call.
3380 */
3381 if (errno == EINTR) {
3382 goto restart;
3383 }
3384 goto error;
3385 }
3386
3387 nb_fd = ret;
3388
3389 for (i = 0; i < nb_fd; i++) {
3390 /* Fetch once the poll data */
3391 revents = LTTNG_POLL_GETEV(&events, i);
3392 pollfd = LTTNG_POLL_GETFD(&events, i);
3393
3394 /* Thread quit pipe has been closed. Killing thread. */
3395 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3396 if (ret) {
3397 err = 0;
3398 goto exit;
3399 }
3400
3401 /* Event on the registration socket */
3402 if (pollfd == sock) {
3403 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3404 ERR("Health socket poll error");
3405 goto error;
3406 }
3407 }
3408 }
3409
3410 new_sock = lttcomm_accept_unix_sock(sock);
3411 if (new_sock < 0) {
3412 goto error;
3413 }
3414
3415 /*
3416 * Set the CLOEXEC flag. Return code is useless because either way, the
3417 * show must go on.
3418 */
3419 (void) utils_set_fd_cloexec(new_sock);
3420
3421 DBG("Receiving data from client for health...");
3422 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3423 if (ret <= 0) {
3424 DBG("Nothing recv() from client... continuing");
3425 ret = close(new_sock);
3426 if (ret) {
3427 PERROR("close");
3428 }
3429 new_sock = -1;
3430 continue;
3431 }
3432
3433 rcu_thread_online();
3434
3435 switch (msg.component) {
3436 case LTTNG_HEALTH_CMD:
3437 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3438 break;
3439 case LTTNG_HEALTH_APP_MANAGE:
3440 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3441 break;
3442 case LTTNG_HEALTH_APP_REG:
3443 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3444 break;
3445 case LTTNG_HEALTH_KERNEL:
3446 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3447 break;
3448 case LTTNG_HEALTH_CONSUMER:
3449 reply.ret_code = check_consumer_health();
3450 break;
3451 case LTTNG_HEALTH_HT_CLEANUP:
3452 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3453 break;
3454 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3455 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3456 break;
3457 case LTTNG_HEALTH_APP_REG_DISPATCH:
3458 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3459 break;
3460 case LTTNG_HEALTH_ALL:
3461 reply.ret_code =
3462 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3463 health_check_state(HEALTH_TYPE_APP_REG) &&
3464 health_check_state(HEALTH_TYPE_CMD) &&
3465 health_check_state(HEALTH_TYPE_KERNEL) &&
3466 check_consumer_health() &&
3467 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3468 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3469 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3470 break;
3471 default:
3472 reply.ret_code = LTTNG_ERR_UND;
3473 break;
3474 }
3475
3476 /*
3477 * Flip ret value since 0 is a success and 1 indicates a bad health for
3478 * the client where in the sessiond it is the opposite. Again, this is
3479 * just to make things easier for us poor developer which enjoy a lot
3480 * lazyness.
3481 */
3482 if (reply.ret_code == 0 || reply.ret_code == 1) {
3483 reply.ret_code = !reply.ret_code;
3484 }
3485
3486 DBG2("Health check return value %d", reply.ret_code);
3487
3488 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3489 if (ret < 0) {
3490 ERR("Failed to send health data back to client");
3491 }
3492
3493 /* End of transmission */
3494 ret = close(new_sock);
3495 if (ret) {
3496 PERROR("close");
3497 }
3498 new_sock = -1;
3499 }
3500
3501 exit:
3502 error:
3503 if (err) {
3504 ERR("Health error occurred in %s", __func__);
3505 }
3506 DBG("Health check thread dying");
3507 unlink(health_unix_sock_path);
3508 if (sock >= 0) {
3509 ret = close(sock);
3510 if (ret) {
3511 PERROR("close");
3512 }
3513 }
3514
3515 lttng_poll_clean(&events);
3516
3517 rcu_unregister_thread();
3518 return NULL;
3519 }
3520
3521 /*
3522 * This thread manage all clients request using the unix client socket for
3523 * communication.
3524 */
3525 static void *thread_manage_clients(void *data)
3526 {
3527 int sock = -1, ret, i, pollfd, err = -1;
3528 int sock_error;
3529 uint32_t revents, nb_fd;
3530 struct command_ctx *cmd_ctx = NULL;
3531 struct lttng_poll_event events;
3532
3533 DBG("[thread] Manage client started");
3534
3535 rcu_register_thread();
3536
3537 health_register(HEALTH_TYPE_CMD);
3538
3539 if (testpoint(thread_manage_clients)) {
3540 goto error_testpoint;
3541 }
3542
3543 health_code_update();
3544
3545 ret = lttcomm_listen_unix_sock(client_sock);
3546 if (ret < 0) {
3547 goto error_listen;
3548 }
3549
3550 /*
3551 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3552 * more will be added to this poll set.
3553 */
3554 ret = sessiond_set_thread_pollset(&events, 2);
3555 if (ret < 0) {
3556 goto error_create_poll;
3557 }
3558
3559 /* Add the application registration socket */
3560 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3561 if (ret < 0) {
3562 goto error;
3563 }
3564
3565 /*
3566 * Notify parent pid that we are ready to accept command for client side.
3567 */
3568 if (opt_sig_parent) {
3569 kill(ppid, SIGUSR1);
3570 }
3571
3572 if (testpoint(thread_manage_clients_before_loop)) {
3573 goto error;
3574 }
3575
3576 health_code_update();
3577
3578 while (1) {
3579 DBG("Accepting client command ...");
3580
3581 /* Inifinite blocking call, waiting for transmission */
3582 restart:
3583 health_poll_entry();
3584 ret = lttng_poll_wait(&events, -1);
3585 health_poll_exit();
3586 if (ret < 0) {
3587 /*
3588 * Restart interrupted system call.
3589 */
3590 if (errno == EINTR) {
3591 goto restart;
3592 }
3593 goto error;
3594 }
3595
3596 nb_fd = ret;
3597
3598 for (i = 0; i < nb_fd; i++) {
3599 /* Fetch once the poll data */
3600 revents = LTTNG_POLL_GETEV(&events, i);
3601 pollfd = LTTNG_POLL_GETFD(&events, i);
3602
3603 health_code_update();
3604
3605 /* Thread quit pipe has been closed. Killing thread. */
3606 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3607 if (ret) {
3608 err = 0;
3609 goto exit;
3610 }
3611
3612 /* Event on the registration socket */
3613 if (pollfd == client_sock) {
3614 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3615 ERR("Client socket poll error");
3616 goto error;
3617 }
3618 }
3619 }
3620
3621 DBG("Wait for client response");
3622
3623 health_code_update();
3624
3625 sock = lttcomm_accept_unix_sock(client_sock);
3626 if (sock < 0) {
3627 goto error;
3628 }
3629
3630 /*
3631 * Set the CLOEXEC flag. Return code is useless because either way, the
3632 * show must go on.
3633 */
3634 (void) utils_set_fd_cloexec(sock);
3635
3636 /* Set socket option for credentials retrieval */
3637 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3638 if (ret < 0) {
3639 goto error;
3640 }
3641
3642 /* Allocate context command to process the client request */
3643 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3644 if (cmd_ctx == NULL) {
3645 PERROR("zmalloc cmd_ctx");
3646 goto error;
3647 }
3648
3649 /* Allocate data buffer for reception */
3650 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3651 if (cmd_ctx->lsm == NULL) {
3652 PERROR("zmalloc cmd_ctx->lsm");
3653 goto error;
3654 }
3655
3656 cmd_ctx->llm = NULL;
3657 cmd_ctx->session = NULL;
3658
3659 health_code_update();
3660
3661 /*
3662 * Data is received from the lttng client. The struct
3663 * lttcomm_session_msg (lsm) contains the command and data request of
3664 * the client.
3665 */
3666 DBG("Receiving data from client ...");
3667 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3668 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3669 if (ret <= 0) {
3670 DBG("Nothing recv() from client... continuing");
3671 ret = close(sock);
3672 if (ret) {
3673 PERROR("close");
3674 }
3675 sock = -1;
3676 clean_command_ctx(&cmd_ctx);
3677 continue;
3678 }
3679
3680 health_code_update();
3681
3682 // TODO: Validate cmd_ctx including sanity check for
3683 // security purpose.
3684
3685 rcu_thread_online();
3686 /*
3687 * This function dispatch the work to the kernel or userspace tracer
3688 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3689 * informations for the client. The command context struct contains
3690 * everything this function may needs.
3691 */
3692 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3693 rcu_thread_offline();
3694 if (ret < 0) {
3695 ret = close(sock);
3696 if (ret) {
3697 PERROR("close");
3698 }
3699 sock = -1;
3700 /*
3701 * TODO: Inform client somehow of the fatal error. At
3702 * this point, ret < 0 means that a zmalloc failed
3703 * (ENOMEM). Error detected but still accept
3704 * command, unless a socket error has been
3705 * detected.
3706 */
3707 clean_command_ctx(&cmd_ctx);
3708 continue;
3709 }
3710
3711 health_code_update();
3712
3713 DBG("Sending response (size: %d, retcode: %s)",
3714 cmd_ctx->lttng_msg_size,
3715 lttng_strerror(-cmd_ctx->llm->ret_code));
3716 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3717 if (ret < 0) {
3718 ERR("Failed to send data back to client");
3719 }
3720
3721 /* End of transmission */
3722 ret = close(sock);
3723 if (ret) {
3724 PERROR("close");
3725 }
3726 sock = -1;
3727
3728 clean_command_ctx(&cmd_ctx);
3729
3730 health_code_update();
3731 }
3732
3733 exit:
3734 error:
3735 if (sock >= 0) {
3736 ret = close(sock);
3737 if (ret) {
3738 PERROR("close");
3739 }
3740 }
3741
3742 lttng_poll_clean(&events);
3743 clean_command_ctx(&cmd_ctx);
3744
3745 error_listen:
3746 error_create_poll:
3747 error_testpoint:
3748 unlink(client_unix_sock_path);
3749 if (client_sock >= 0) {
3750 ret = close(client_sock);
3751 if (ret) {
3752 PERROR("close");
3753 }
3754 }
3755
3756 if (err) {
3757 health_error();
3758 ERR("Health error occurred in %s", __func__);
3759 }
3760
3761 health_unregister();
3762
3763 DBG("Client thread dying");
3764
3765 rcu_unregister_thread();
3766 return NULL;
3767 }
3768
3769
3770 /*
3771 * usage function on stderr
3772 */
3773 static void usage(void)
3774 {
3775 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3776 fprintf(stderr, " -h, --help Display this usage.\n");
3777 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3778 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3779 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3780 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3781 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3782 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3783 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3784 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3785 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3786 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3787 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3788 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3789 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3790 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3791 fprintf(stderr, " -V, --version Show version number.\n");
3792 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3793 fprintf(stderr, " -q, --quiet No output at all.\n");
3794 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3795 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3796 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3797 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3798 }
3799
3800 /*
3801 * daemon argument parsing
3802 */
3803 static int parse_args(int argc, char **argv)
3804 {
3805 int c;
3806
3807 static struct option long_options[] = {
3808 { "client-sock", 1, 0, 'c' },
3809 { "apps-sock", 1, 0, 'a' },
3810 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3811 { "kconsumerd-err-sock", 1, 0, 'E' },
3812 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3813 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3814 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3815 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3816 { "consumerd32-path", 1, 0, 'u' },
3817 { "consumerd32-libdir", 1, 0, 'U' },
3818 { "consumerd64-path", 1, 0, 't' },
3819 { "consumerd64-libdir", 1, 0, 'T' },
3820 { "daemonize", 0, 0, 'd' },
3821 { "sig-parent", 0, 0, 'S' },
3822 { "help", 0, 0, 'h' },
3823 { "group", 1, 0, 'g' },
3824 { "version", 0, 0, 'V' },
3825 { "quiet", 0, 0, 'q' },
3826 { "verbose", 0, 0, 'v' },
3827 { "verbose-consumer", 0, 0, 'Z' },
3828 { "no-kernel", 0, 0, 'N' },
3829 { "pidfile", 1, 0, 'p' },
3830 { NULL, 0, 0, 0 }
3831 };
3832
3833 while (1) {
3834 int option_index = 0;
3835 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3836 long_options, &option_index);
3837 if (c == -1) {
3838 break;
3839 }
3840
3841 switch (c) {
3842 case 0:
3843 fprintf(stderr, "option %s", long_options[option_index].name);
3844 if (optarg) {
3845 fprintf(stderr, " with arg %s\n", optarg);
3846 }
3847 break;
3848 case 'c':
3849 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3850 break;
3851 case 'a':
3852 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3853 break;
3854 case 'd':
3855 opt_daemon = 1;
3856 break;
3857 case 'g':
3858 opt_tracing_group = optarg;
3859 break;
3860 case 'h':
3861 usage();
3862 exit(EXIT_FAILURE);
3863 case 'V':
3864 fprintf(stdout, "%s\n", VERSION);
3865 exit(EXIT_SUCCESS);
3866 case 'S':
3867 opt_sig_parent = 1;
3868 break;
3869 case 'E':
3870 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3871 break;
3872 case 'C':
3873 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3874 break;
3875 case 'F':
3876 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3877 break;
3878 case 'D':
3879 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3880 break;
3881 case 'H':
3882 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3883 break;
3884 case 'G':
3885 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3886 break;
3887 case 'N':
3888 opt_no_kernel = 1;
3889 break;
3890 case 'q':
3891 lttng_opt_quiet = 1;
3892 break;
3893 case 'v':
3894 /* Verbose level can increase using multiple -v */
3895 lttng_opt_verbose += 1;
3896 break;
3897 case 'Z':
3898 opt_verbose_consumer += 1;
3899 break;
3900 case 'u':
3901 consumerd32_bin= optarg;
3902 break;
3903 case 'U':
3904 consumerd32_libdir = optarg;
3905 break;
3906 case 't':
3907 consumerd64_bin = optarg;
3908 break;
3909 case 'T':
3910 consumerd64_libdir = optarg;
3911 break;
3912 case 'p':
3913 opt_pidfile = optarg;
3914 break;
3915 default:
3916 /* Unknown option or other error.
3917 * Error is printed by getopt, just return */
3918 return -1;
3919 }
3920 }
3921
3922 return 0;
3923 }
3924
3925 /*
3926 * Creates the two needed socket by the daemon.
3927 * apps_sock - The communication socket for all UST apps.
3928 * client_sock - The communication of the cli tool (lttng).
3929 */
3930 static int init_daemon_socket(void)
3931 {
3932 int ret = 0;
3933 mode_t old_umask;
3934
3935 old_umask = umask(0);
3936
3937 /* Create client tool unix socket */
3938 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3939 if (client_sock < 0) {
3940 ERR("Create unix sock failed: %s", client_unix_sock_path);
3941 ret = -1;
3942 goto end;
3943 }
3944
3945 /* Set the cloexec flag */
3946 ret = utils_set_fd_cloexec(client_sock);
3947 if (ret < 0) {
3948 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3949 "Continuing but note that the consumer daemon will have a "
3950 "reference to this socket on exec()", client_sock);
3951 }
3952
3953 /* File permission MUST be 660 */
3954 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3955 if (ret < 0) {
3956 ERR("Set file permissions failed: %s", client_unix_sock_path);
3957 PERROR("chmod");
3958 goto end;
3959 }
3960
3961 /* Create the application unix socket */
3962 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3963 if (apps_sock < 0) {
3964 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3965 ret = -1;
3966 goto end;
3967 }
3968
3969 /* Set the cloexec flag */
3970 ret = utils_set_fd_cloexec(apps_sock);
3971 if (ret < 0) {
3972 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3973 "Continuing but note that the consumer daemon will have a "
3974 "reference to this socket on exec()", apps_sock);
3975 }
3976
3977 /* File permission MUST be 666 */
3978 ret = chmod(apps_unix_sock_path,
3979 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3980 if (ret < 0) {
3981 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3982 PERROR("chmod");
3983 goto end;
3984 }
3985
3986 DBG3("Session daemon client socket %d and application socket %d created",
3987 client_sock, apps_sock);
3988
3989 end:
3990 umask(old_umask);
3991 return ret;
3992 }
3993
3994 /*
3995 * Check if the global socket is available, and if a daemon is answering at the
3996 * other side. If yes, error is returned.
3997 */
3998 static int check_existing_daemon(void)
3999 {
4000 /* Is there anybody out there ? */
4001 if (lttng_session_daemon_alive()) {
4002 return -EEXIST;
4003 }
4004
4005 return 0;
4006 }
4007
4008 /*
4009 * Set the tracing group gid onto the client socket.
4010 *
4011 * Race window between mkdir and chown is OK because we are going from more
4012 * permissive (root.root) to less permissive (root.tracing).
4013 */
4014 static int set_permissions(char *rundir)
4015 {
4016 int ret;
4017 gid_t gid;
4018
4019 ret = allowed_group();
4020 if (ret < 0) {
4021 WARN("No tracing group detected");
4022 ret = 0;
4023 goto end;
4024 }
4025
4026 gid = ret;
4027
4028 /* Set lttng run dir */
4029 ret = chown(rundir, 0, gid);
4030 if (ret < 0) {
4031 ERR("Unable to set group on %s", rundir);
4032 PERROR("chown");
4033 }
4034
4035 /* Ensure tracing group can search the run dir */
4036 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
4037 if (ret < 0) {
4038 ERR("Unable to set permissions on %s", rundir);
4039 PERROR("chmod");
4040 }
4041
4042 /* lttng client socket path */
4043 ret = chown(client_unix_sock_path, 0, gid);
4044 if (ret < 0) {
4045 ERR("Unable to set group on %s", client_unix_sock_path);
4046 PERROR("chown");
4047 }
4048
4049 /* kconsumer error socket path */
4050 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4051 if (ret < 0) {
4052 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4053 PERROR("chown");
4054 }
4055
4056 /* 64-bit ustconsumer error socket path */
4057 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4058 if (ret < 0) {
4059 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4060 PERROR("chown");
4061 }
4062
4063 /* 32-bit ustconsumer compat32 error socket path */
4064 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4065 if (ret < 0) {
4066 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4067 PERROR("chown");
4068 }
4069
4070 DBG("All permissions are set");
4071
4072 end:
4073 return ret;
4074 }
4075
4076 /*
4077 * Create the lttng run directory needed for all global sockets and pipe.
4078 */
4079 static int create_lttng_rundir(const char *rundir)
4080 {
4081 int ret;
4082
4083 DBG3("Creating LTTng run directory: %s", rundir);
4084
4085 ret = mkdir(rundir, S_IRWXU);
4086 if (ret < 0) {
4087 if (errno != EEXIST) {
4088 ERR("Unable to create %s", rundir);
4089 goto error;
4090 } else {
4091 ret = 0;
4092 }
4093 }
4094
4095 error:
4096 return ret;
4097 }
4098
4099 /*
4100 * Setup sockets and directory needed by the kconsumerd communication with the
4101 * session daemon.
4102 */
4103 static int set_consumer_sockets(struct consumer_data *consumer_data,
4104 const char *rundir)
4105 {
4106 int ret;
4107 char path[PATH_MAX];
4108
4109 switch (consumer_data->type) {
4110 case LTTNG_CONSUMER_KERNEL:
4111 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4112 break;
4113 case LTTNG_CONSUMER64_UST:
4114 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4115 break;
4116 case LTTNG_CONSUMER32_UST:
4117 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4118 break;
4119 default:
4120 ERR("Consumer type unknown");
4121 ret = -EINVAL;
4122 goto error;
4123 }
4124
4125 DBG2("Creating consumer directory: %s", path);
4126
4127 ret = mkdir(path, S_IRWXU);
4128 if (ret < 0) {
4129 if (errno != EEXIST) {
4130 PERROR("mkdir");
4131 ERR("Failed to create %s", path);
4132 goto error;
4133 }
4134 ret = -1;
4135 }
4136
4137 /* Create the kconsumerd error unix socket */
4138 consumer_data->err_sock =
4139 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4140 if (consumer_data->err_sock < 0) {
4141 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4142 ret = -1;
4143 goto error;
4144 }
4145
4146 /* File permission MUST be 660 */
4147 ret = chmod(consumer_data->err_unix_sock_path,
4148 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4149 if (ret < 0) {
4150 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4151 PERROR("chmod");
4152 goto error;
4153 }
4154
4155 error:
4156 return ret;
4157 }
4158
4159 /*
4160 * Signal handler for the daemon
4161 *
4162 * Simply stop all worker threads, leaving main() return gracefully after
4163 * joining all threads and calling cleanup().
4164 */
4165 static void sighandler(int sig)
4166 {
4167 switch (sig) {
4168 case SIGPIPE:
4169 DBG("SIGPIPE caught");
4170 return;
4171 case SIGINT:
4172 DBG("SIGINT caught");
4173 stop_threads();
4174 break;
4175 case SIGTERM:
4176 DBG("SIGTERM caught");
4177 stop_threads();
4178 break;
4179 default:
4180 break;
4181 }
4182 }
4183
4184 /*
4185 * Setup signal handler for :
4186 * SIGINT, SIGTERM, SIGPIPE
4187 */
4188 static int set_signal_handler(void)
4189 {
4190 int ret = 0;
4191 struct sigaction sa;
4192 sigset_t sigset;
4193
4194 if ((ret = sigemptyset(&sigset)) < 0) {
4195 PERROR("sigemptyset");
4196 return ret;
4197 }
4198
4199 sa.sa_handler = sighandler;
4200 sa.sa_mask = sigset;
4201 sa.sa_flags = 0;
4202 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4203 PERROR("sigaction");
4204 return ret;
4205 }
4206
4207 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4208 PERROR("sigaction");
4209 return ret;
4210 }
4211
4212 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4213 PERROR("sigaction");
4214 return ret;
4215 }
4216
4217 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4218
4219 return ret;
4220 }
4221
4222 /*
4223 * Set open files limit to unlimited. This daemon can open a large number of
4224 * file descriptors in order to consumer multiple kernel traces.
4225 */
4226 static void set_ulimit(void)
4227 {
4228 int ret;
4229 struct rlimit lim;
4230
4231 /* The kernel does not allowed an infinite limit for open files */
4232 lim.rlim_cur = 65535;
4233 lim.rlim_max = 65535;
4234
4235 ret = setrlimit(RLIMIT_NOFILE, &lim);
4236 if (ret < 0) {
4237 PERROR("failed to set open files limit");
4238 }
4239 }
4240
4241 /*
4242 * Write pidfile using the rundir and opt_pidfile.
4243 */
4244 static void write_pidfile(void)
4245 {
4246 int ret;
4247 char pidfile_path[PATH_MAX];
4248
4249 assert(rundir);
4250
4251 if (opt_pidfile) {
4252 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4253 } else {
4254 /* Build pidfile path from rundir and opt_pidfile. */
4255 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4256 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4257 if (ret < 0) {
4258 PERROR("snprintf pidfile path");
4259 goto error;
4260 }
4261 }
4262
4263 /*
4264 * Create pid file in rundir. Return value is of no importance. The
4265 * execution will continue even though we are not able to write the file.
4266 */
4267 (void) utils_create_pid_file(getpid(), pidfile_path);
4268
4269 error:
4270 return;
4271 }
4272
4273 /*
4274 * main
4275 */
4276 int main(int argc, char **argv)
4277 {
4278 int ret = 0;
4279 void *status;
4280 const char *home_path, *env_app_timeout;
4281
4282 init_kernel_workarounds();
4283
4284 rcu_register_thread();
4285
4286 setup_consumerd_path();
4287
4288 page_size = sysconf(_SC_PAGESIZE);
4289 if (page_size < 0) {
4290 PERROR("sysconf _SC_PAGESIZE");
4291 page_size = LONG_MAX;
4292 WARN("Fallback page size to %ld", page_size);
4293 }
4294
4295 /* Parse arguments */
4296 progname = argv[0];
4297 if ((ret = parse_args(argc, argv)) < 0) {
4298 goto error;
4299 }
4300
4301 /* Daemonize */
4302 if (opt_daemon) {
4303 int i;
4304
4305 /*
4306 * fork
4307 * child: setsid, close FD 0, 1, 2, chdir /
4308 * parent: exit (if fork is successful)
4309 */
4310 ret = daemon(0, 0);
4311 if (ret < 0) {
4312 PERROR("daemon");
4313 goto error;
4314 }
4315 /*
4316 * We are in the child. Make sure all other file
4317 * descriptors are closed, in case we are called with
4318 * more opened file descriptors than the standard ones.
4319 */
4320 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4321 (void) close(i);
4322 }
4323 }
4324
4325 /* Create thread quit pipe */
4326 if ((ret = init_thread_quit_pipe()) < 0) {
4327 goto error;
4328 }
4329
4330 /* Check if daemon is UID = 0 */
4331 is_root = !getuid();
4332
4333 if (is_root) {
4334 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4335
4336 /* Create global run dir with root access */
4337 ret = create_lttng_rundir(rundir);
4338 if (ret < 0) {
4339 goto error;
4340 }
4341
4342 if (strlen(apps_unix_sock_path) == 0) {
4343 snprintf(apps_unix_sock_path, PATH_MAX,
4344 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4345 }
4346
4347 if (strlen(client_unix_sock_path) == 0) {
4348 snprintf(client_unix_sock_path, PATH_MAX,
4349 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4350 }
4351
4352 /* Set global SHM for ust */
4353 if (strlen(wait_shm_path) == 0) {
4354 snprintf(wait_shm_path, PATH_MAX,
4355 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4356 }
4357
4358 if (strlen(health_unix_sock_path) == 0) {
4359 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4360 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4361 }
4362
4363 /* Setup kernel consumerd path */
4364 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4365 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4366 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4367 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4368
4369 DBG2("Kernel consumer err path: %s",
4370 kconsumer_data.err_unix_sock_path);
4371 DBG2("Kernel consumer cmd path: %s",
4372 kconsumer_data.cmd_unix_sock_path);
4373 } else {
4374 home_path = utils_get_home_dir();
4375 if (home_path == NULL) {
4376 /* TODO: Add --socket PATH option */
4377 ERR("Can't get HOME directory for sockets creation.");
4378 ret = -EPERM;
4379 goto error;
4380 }
4381
4382 /*
4383 * Create rundir from home path. This will create something like
4384 * $HOME/.lttng
4385 */
4386 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4387 if (ret < 0) {
4388 ret = -ENOMEM;
4389 goto error;
4390 }
4391
4392 ret = create_lttng_rundir(rundir);
4393 if (ret < 0) {
4394 goto error;
4395 }
4396
4397 if (strlen(apps_unix_sock_path) == 0) {
4398 snprintf(apps_unix_sock_path, PATH_MAX,
4399 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4400 }
4401
4402 /* Set the cli tool unix socket path */
4403 if (strlen(client_unix_sock_path) == 0) {
4404 snprintf(client_unix_sock_path, PATH_MAX,
4405 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4406 }
4407
4408 /* Set global SHM for ust */
4409 if (strlen(wait_shm_path) == 0) {
4410 snprintf(wait_shm_path, PATH_MAX,
4411 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4412 }
4413
4414 /* Set health check Unix path */
4415 if (strlen(health_unix_sock_path) == 0) {
4416 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4417 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4418 }
4419 }
4420
4421 /* Set consumer initial state */
4422 kernel_consumerd_state = CONSUMER_STOPPED;
4423 ust_consumerd_state = CONSUMER_STOPPED;
4424
4425 DBG("Client socket path %s", client_unix_sock_path);
4426 DBG("Application socket path %s", apps_unix_sock_path);
4427 DBG("Application wait path %s", wait_shm_path);
4428 DBG("LTTng run directory path: %s", rundir);
4429
4430 /* 32 bits consumerd path setup */
4431 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4432 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4433 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4434 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4435
4436 DBG2("UST consumer 32 bits err path: %s",
4437 ustconsumer32_data.err_unix_sock_path);
4438 DBG2("UST consumer 32 bits cmd path: %s",
4439 ustconsumer32_data.cmd_unix_sock_path);
4440
4441 /* 64 bits consumerd path setup */
4442 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4443 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4444 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4445 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4446
4447 DBG2("UST consumer 64 bits err path: %s",
4448 ustconsumer64_data.err_unix_sock_path);
4449 DBG2("UST consumer 64 bits cmd path: %s",
4450 ustconsumer64_data.cmd_unix_sock_path);
4451
4452 /*
4453 * See if daemon already exist.
4454 */
4455 if ((ret = check_existing_daemon()) < 0) {
4456 ERR("Already running daemon.\n");
4457 /*
4458 * We do not goto exit because we must not cleanup()
4459 * because a daemon is already running.
4460 */
4461 goto error;
4462 }
4463
4464 /*
4465 * Init UST app hash table. Alloc hash table before this point since
4466 * cleanup() can get called after that point.
4467 */
4468 ust_app_ht_alloc();
4469
4470 /* After this point, we can safely call cleanup() with "goto exit" */
4471
4472 /*
4473 * These actions must be executed as root. We do that *after* setting up
4474 * the sockets path because we MUST make the check for another daemon using
4475 * those paths *before* trying to set the kernel consumer sockets and init
4476 * kernel tracer.
4477 */
4478 if (is_root) {
4479 ret = set_consumer_sockets(&kconsumer_data, rundir);
4480 if (ret < 0) {
4481 goto exit;
4482 }
4483
4484 /* Setup kernel tracer */
4485 if (!opt_no_kernel) {
4486 init_kernel_tracer();
4487 }
4488
4489 /* Set ulimit for open files */
4490 set_ulimit();
4491 }
4492 /* init lttng_fd tracking must be done after set_ulimit. */
4493 lttng_fd_init();
4494
4495 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4496 if (ret < 0) {
4497 goto exit;
4498 }
4499
4500 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4501 if (ret < 0) {
4502 goto exit;
4503 }
4504
4505 if ((ret = set_signal_handler()) < 0) {
4506 goto exit;
4507 }
4508
4509 /* Setup the needed unix socket */
4510 if ((ret = init_daemon_socket()) < 0) {
4511 goto exit;
4512 }
4513
4514 /* Set credentials to socket */
4515 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4516 goto exit;
4517 }
4518
4519 /* Get parent pid if -S, --sig-parent is specified. */
4520 if (opt_sig_parent) {
4521 ppid = getppid();
4522 }
4523
4524 /* Setup the kernel pipe for waking up the kernel thread */
4525 if (is_root && !opt_no_kernel) {
4526 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4527 goto exit;
4528 }
4529 }
4530
4531 /* Setup the thread ht_cleanup communication pipe. */
4532 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4533 goto exit;
4534 }
4535
4536 /* Setup the thread apps communication pipe. */
4537 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4538 goto exit;
4539 }
4540
4541 /* Setup the thread apps notify communication pipe. */
4542 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4543 goto exit;
4544 }
4545
4546 /* Initialize global buffer per UID and PID registry. */
4547 buffer_reg_init_uid_registry();
4548 buffer_reg_init_pid_registry();
4549
4550 /* Init UST command queue. */
4551 cds_wfq_init(&ust_cmd_queue.queue);
4552
4553 /*
4554 * Get session list pointer. This pointer MUST NOT be free(). This list is
4555 * statically declared in session.c
4556 */
4557 session_list_ptr = session_get_list();
4558
4559 /* Set up max poll set size */
4560 lttng_poll_set_max_size();
4561
4562 cmd_init();
4563
4564 /* Check for the application socket timeout env variable. */
4565 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4566 if (env_app_timeout) {
4567 app_socket_timeout = atoi(env_app_timeout);
4568 } else {
4569 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4570 }
4571
4572 write_pidfile();
4573
4574 /* Create thread to manage the client socket */
4575 ret = pthread_create(&ht_cleanup_thread, NULL,
4576 thread_ht_cleanup, (void *) NULL);
4577 if (ret != 0) {
4578 PERROR("pthread_create ht_cleanup");
4579 goto exit_ht_cleanup;
4580 }
4581
4582 /* Create thread to manage the client socket */
4583 ret = pthread_create(&health_thread, NULL,
4584 thread_manage_health, (void *) NULL);
4585 if (ret != 0) {
4586 PERROR("pthread_create health");
4587 goto exit_health;
4588 }
4589
4590 /* Create thread to manage the client socket */
4591 ret = pthread_create(&client_thread, NULL,
4592 thread_manage_clients, (void *) NULL);
4593 if (ret != 0) {
4594 PERROR("pthread_create clients");
4595 goto exit_client;
4596 }
4597
4598 /* Create thread to dispatch registration */
4599 ret = pthread_create(&dispatch_thread, NULL,
4600 thread_dispatch_ust_registration, (void *) NULL);
4601 if (ret != 0) {
4602 PERROR("pthread_create dispatch");
4603 goto exit_dispatch;
4604 }
4605
4606 /* Create thread to manage application registration. */
4607 ret = pthread_create(&reg_apps_thread, NULL,
4608 thread_registration_apps, (void *) NULL);
4609 if (ret != 0) {
4610 PERROR("pthread_create registration");
4611 goto exit_reg_apps;
4612 }
4613
4614 /* Create thread to manage application socket */
4615 ret = pthread_create(&apps_thread, NULL,
4616 thread_manage_apps, (void *) NULL);
4617 if (ret != 0) {
4618 PERROR("pthread_create apps");
4619 goto exit_apps;
4620 }
4621
4622 /* Create thread to manage application notify socket */
4623 ret = pthread_create(&apps_notify_thread, NULL,
4624 ust_thread_manage_notify, (void *) NULL);
4625 if (ret != 0) {
4626 PERROR("pthread_create apps");
4627 goto exit_apps;
4628 }
4629
4630 /* Don't start this thread if kernel tracing is not requested nor root */
4631 if (is_root && !opt_no_kernel) {
4632 /* Create kernel thread to manage kernel event */
4633 ret = pthread_create(&kernel_thread, NULL,
4634 thread_manage_kernel, (void *) NULL);
4635 if (ret != 0) {
4636 PERROR("pthread_create kernel");
4637 goto exit_kernel;
4638 }
4639
4640 ret = pthread_join(kernel_thread, &status);
4641 if (ret != 0) {
4642 PERROR("pthread_join");
4643 goto error; /* join error, exit without cleanup */
4644 }
4645 }
4646
4647 exit_kernel:
4648 ret = pthread_join(apps_thread, &status);
4649 if (ret != 0) {
4650 PERROR("pthread_join");
4651 goto error; /* join error, exit without cleanup */
4652 }
4653
4654 exit_apps:
4655 ret = pthread_join(reg_apps_thread, &status);
4656 if (ret != 0) {
4657 PERROR("pthread_join");
4658 goto error; /* join error, exit without cleanup */
4659 }
4660
4661 exit_reg_apps:
4662 ret = pthread_join(dispatch_thread, &status);
4663 if (ret != 0) {
4664 PERROR("pthread_join");
4665 goto error; /* join error, exit without cleanup */
4666 }
4667
4668 exit_dispatch:
4669 ret = pthread_join(client_thread, &status);
4670 if (ret != 0) {
4671 PERROR("pthread_join");
4672 goto error; /* join error, exit without cleanup */
4673 }
4674
4675 ret = join_consumer_thread(&kconsumer_data);
4676 if (ret != 0) {
4677 PERROR("join_consumer");
4678 goto error; /* join error, exit without cleanup */
4679 }
4680
4681 ret = join_consumer_thread(&ustconsumer32_data);
4682 if (ret != 0) {
4683 PERROR("join_consumer ust32");
4684 goto error; /* join error, exit without cleanup */
4685 }
4686
4687 ret = join_consumer_thread(&ustconsumer64_data);
4688 if (ret != 0) {
4689 PERROR("join_consumer ust64");
4690 goto error; /* join error, exit without cleanup */
4691 }
4692
4693 exit_client:
4694 ret = pthread_join(health_thread, &status);
4695 if (ret != 0) {
4696 PERROR("pthread_join health thread");
4697 goto error; /* join error, exit without cleanup */
4698 }
4699
4700 exit_health:
4701 ret = pthread_join(ht_cleanup_thread, &status);
4702 if (ret != 0) {
4703 PERROR("pthread_join ht cleanup thread");
4704 goto error; /* join error, exit without cleanup */
4705 }
4706 exit_ht_cleanup:
4707 exit:
4708 /*
4709 * cleanup() is called when no other thread is running.
4710 */
4711 rcu_thread_online();
4712 cleanup();
4713 rcu_thread_offline();
4714 rcu_unregister_thread();
4715 if (!ret) {
4716 exit(EXIT_SUCCESS);
4717 }
4718 error:
4719 exit(EXIT_FAILURE);
4720 }
This page took 0.166287 seconds and 6 git commands to generate.