91fba2643fc2ae94acf79850fd0e9aa589cfbf90
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/uatomic.h>
36 #include <unistd.h>
37 #include <config.h>
38
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "kernel.h"
55 #include "kernel-consumer.h"
56 #include "modprobe.h"
57 #include "shm.h"
58 #include "ust-ctl.h"
59 #include "ust-consumer.h"
60 #include "utils.h"
61 #include "fd-limit.h"
62 #include "health.h"
63 #include "testpoint.h"
64
65 #define CONSUMERD_FILE "lttng-consumerd"
66
67 /* Const values */
68 const char default_home_dir[] = DEFAULT_HOME_DIR;
69 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
70 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
71 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
72
73 const char *progname;
74 const char *opt_tracing_group;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread;
153 static pthread_t reg_apps_thread;
154 static pthread_t client_thread;
155 static pthread_t kernel_thread;
156 static pthread_t dispatch_thread;
157 static pthread_t health_thread;
158
159 /*
160 * UST registration command queue. This queue is tied with a futex and uses a N
161 * wakers / 1 waiter implemented and detailed in futex.c/.h
162 *
163 * The thread_manage_apps and thread_dispatch_ust_registration interact with
164 * this queue and the wait/wake scheme.
165 */
166 static struct ust_cmd_queue ust_cmd_queue;
167
168 /*
169 * Pointer initialized before thread creation.
170 *
171 * This points to the tracing session list containing the session count and a
172 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
173 * MUST NOT be taken if you call a public function in session.c.
174 *
175 * The lock is nested inside the structure: session_list_ptr->lock. Please use
176 * session_lock_list and session_unlock_list for lock acquisition.
177 */
178 static struct ltt_session_list *session_list_ptr;
179
180 int ust_consumerd64_fd = -1;
181 int ust_consumerd32_fd = -1;
182
183 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
184 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
185 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
186 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
187
188 static const char *module_proc_lttng = "/proc/lttng";
189
190 /*
191 * Consumer daemon state which is changed when spawning it, killing it or in
192 * case of a fatal error.
193 */
194 enum consumerd_state {
195 CONSUMER_STARTED = 1,
196 CONSUMER_STOPPED = 2,
197 CONSUMER_ERROR = 3,
198 };
199
200 /*
201 * This consumer daemon state is used to validate if a client command will be
202 * able to reach the consumer. If not, the client is informed. For instance,
203 * doing a "lttng start" when the consumer state is set to ERROR will return an
204 * error to the client.
205 *
206 * The following example shows a possible race condition of this scheme:
207 *
208 * consumer thread error happens
209 * client cmd arrives
210 * client cmd checks state -> still OK
211 * consumer thread exit, sets error
212 * client cmd try to talk to consumer
213 * ...
214 *
215 * However, since the consumer is a different daemon, we have no way of making
216 * sure the command will reach it safely even with this state flag. This is why
217 * we consider that up to the state validation during command processing, the
218 * command is safe. After that, we can not guarantee the correctness of the
219 * client request vis-a-vis the consumer.
220 */
221 static enum consumerd_state ust_consumerd_state;
222 static enum consumerd_state kernel_consumerd_state;
223
224 /* Used for the health monitoring of the session daemon. See health.h */
225 struct health_state health_thread_cmd;
226 struct health_state health_thread_app_manage;
227 struct health_state health_thread_app_reg;
228 struct health_state health_thread_kernel;
229
230 /*
231 * Socket timeout for receiving and sending in seconds.
232 */
233 static int app_socket_timeout;
234
235 static
236 void setup_consumerd_path(void)
237 {
238 const char *bin, *libdir;
239
240 /*
241 * Allow INSTALL_BIN_PATH to be used as a target path for the
242 * native architecture size consumer if CONFIG_CONSUMER*_PATH
243 * has not been defined.
244 */
245 #if (CAA_BITS_PER_LONG == 32)
246 if (!consumerd32_bin[0]) {
247 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
248 }
249 if (!consumerd32_libdir[0]) {
250 consumerd32_libdir = INSTALL_LIB_PATH;
251 }
252 #elif (CAA_BITS_PER_LONG == 64)
253 if (!consumerd64_bin[0]) {
254 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
255 }
256 if (!consumerd64_libdir[0]) {
257 consumerd64_libdir = INSTALL_LIB_PATH;
258 }
259 #else
260 #error "Unknown bitness"
261 #endif
262
263 /*
264 * runtime env. var. overrides the build default.
265 */
266 bin = getenv("LTTNG_CONSUMERD32_BIN");
267 if (bin) {
268 consumerd32_bin = bin;
269 }
270 bin = getenv("LTTNG_CONSUMERD64_BIN");
271 if (bin) {
272 consumerd64_bin = bin;
273 }
274 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
275 if (libdir) {
276 consumerd32_libdir = libdir;
277 }
278 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
279 if (libdir) {
280 consumerd64_libdir = libdir;
281 }
282 }
283
284 /*
285 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
286 */
287 static int create_thread_poll_set(struct lttng_poll_event *events,
288 unsigned int size)
289 {
290 int ret;
291
292 if (events == NULL || size == 0) {
293 ret = -1;
294 goto error;
295 }
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310 error:
311 return ret;
312 }
313
314 /*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
319 static int check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326 }
327
328 /*
329 * Return group ID of the tracing group or -1 if not found.
330 */
331 static gid_t allowed_group(void)
332 {
333 struct group *grp;
334
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345 }
346
347 /*
348 * Init thread quit pipe.
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352 static int init_thread_quit_pipe(void)
353 {
354 int ret, i;
355
356 ret = pipe(thread_quit_pipe);
357 if (ret < 0) {
358 PERROR("thread quit pipe");
359 goto error;
360 }
361
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
370 error:
371 return ret;
372 }
373
374 /*
375 * Stop all threads by closing the thread quit pipe.
376 */
377 static void stop_threads(void)
378 {
379 int ret;
380
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret = notify_thread_pipe(thread_quit_pipe[1]);
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
390 futex_nto1_wake(&ust_cmd_queue.futex);
391 }
392
393 /*
394 * Cleanup the daemon
395 */
396 static void cleanup(void)
397 {
398 int ret;
399 char *cmd = NULL;
400 struct ltt_session *sess, *stmp;
401
402 DBG("Cleaning up");
403
404 /* First thing first, stop all threads */
405 utils_close_pipe(thread_quit_pipe);
406
407 DBG("Removing %s directory", rundir);
408 ret = asprintf(&cmd, "rm -rf %s", rundir);
409 if (ret < 0) {
410 ERR("asprintf failed. Something is really wrong!");
411 }
412
413 /* Remove lttng run directory */
414 ret = system(cmd);
415 if (ret < 0) {
416 ERR("Unable to clean %s", rundir);
417 }
418 free(cmd);
419 free(rundir);
420
421 DBG("Cleaning up all sessions");
422
423 /* Destroy session list mutex */
424 if (session_list_ptr != NULL) {
425 pthread_mutex_destroy(&session_list_ptr->lock);
426
427 /* Cleanup ALL session */
428 cds_list_for_each_entry_safe(sess, stmp,
429 &session_list_ptr->head, list) {
430 cmd_destroy_session(sess, kernel_poll_pipe[1]);
431 }
432 }
433
434 DBG("Closing all UST sockets");
435 ust_app_clean_list();
436
437 if (is_root && !opt_no_kernel) {
438 DBG2("Closing kernel fd");
439 if (kernel_tracer_fd >= 0) {
440 ret = close(kernel_tracer_fd);
441 if (ret) {
442 PERROR("close");
443 }
444 }
445 DBG("Unloading kernel modules");
446 modprobe_remove_lttng_all();
447 }
448
449 /* <fun> */
450 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
451 "Matthew, BEET driven development works!%c[%dm",
452 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
453 /* </fun> */
454 }
455
456 /*
457 * Send data on a unix socket using the liblttsessiondcomm API.
458 *
459 * Return lttcomm error code.
460 */
461 static int send_unix_sock(int sock, void *buf, size_t len)
462 {
463 /* Check valid length */
464 if (len == 0) {
465 return -1;
466 }
467
468 return lttcomm_send_unix_sock(sock, buf, len);
469 }
470
471 /*
472 * Free memory of a command context structure.
473 */
474 static void clean_command_ctx(struct command_ctx **cmd_ctx)
475 {
476 DBG("Clean command context structure");
477 if (*cmd_ctx) {
478 if ((*cmd_ctx)->llm) {
479 free((*cmd_ctx)->llm);
480 }
481 if ((*cmd_ctx)->lsm) {
482 free((*cmd_ctx)->lsm);
483 }
484 free(*cmd_ctx);
485 *cmd_ctx = NULL;
486 }
487 }
488
489 /*
490 * Notify UST applications using the shm mmap futex.
491 */
492 static int notify_ust_apps(int active)
493 {
494 char *wait_shm_mmap;
495
496 DBG("Notifying applications of session daemon state: %d", active);
497
498 /* See shm.c for this call implying mmap, shm and futex calls */
499 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
500 if (wait_shm_mmap == NULL) {
501 goto error;
502 }
503
504 /* Wake waiting process */
505 futex_wait_update((int32_t *) wait_shm_mmap, active);
506
507 /* Apps notified successfully */
508 return 0;
509
510 error:
511 return -1;
512 }
513
514 /*
515 * Setup the outgoing data buffer for the response (llm) by allocating the
516 * right amount of memory and copying the original information from the lsm
517 * structure.
518 *
519 * Return total size of the buffer pointed by buf.
520 */
521 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
522 {
523 int ret, buf_size;
524
525 buf_size = size;
526
527 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
528 if (cmd_ctx->llm == NULL) {
529 PERROR("zmalloc");
530 ret = -ENOMEM;
531 goto error;
532 }
533
534 /* Copy common data */
535 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
536 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
537
538 cmd_ctx->llm->data_size = size;
539 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
540
541 return buf_size;
542
543 error:
544 return ret;
545 }
546
547 /*
548 * Update the kernel poll set of all channel fd available over all tracing
549 * session. Add the wakeup pipe at the end of the set.
550 */
551 static int update_kernel_poll(struct lttng_poll_event *events)
552 {
553 int ret;
554 struct ltt_session *session;
555 struct ltt_kernel_channel *channel;
556
557 DBG("Updating kernel poll set");
558
559 session_lock_list();
560 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
561 session_lock(session);
562 if (session->kernel_session == NULL) {
563 session_unlock(session);
564 continue;
565 }
566
567 cds_list_for_each_entry(channel,
568 &session->kernel_session->channel_list.head, list) {
569 /* Add channel fd to the kernel poll set */
570 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
571 if (ret < 0) {
572 session_unlock(session);
573 goto error;
574 }
575 DBG("Channel fd %d added to kernel set", channel->fd);
576 }
577 session_unlock(session);
578 }
579 session_unlock_list();
580
581 return 0;
582
583 error:
584 session_unlock_list();
585 return -1;
586 }
587
588 /*
589 * Find the channel fd from 'fd' over all tracing session. When found, check
590 * for new channel stream and send those stream fds to the kernel consumer.
591 *
592 * Useful for CPU hotplug feature.
593 */
594 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
595 {
596 int ret = 0;
597 struct ltt_session *session;
598 struct ltt_kernel_session *ksess;
599 struct ltt_kernel_channel *channel;
600
601 DBG("Updating kernel streams for channel fd %d", fd);
602
603 session_lock_list();
604 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
605 session_lock(session);
606 if (session->kernel_session == NULL) {
607 session_unlock(session);
608 continue;
609 }
610 ksess = session->kernel_session;
611
612 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
613 if (channel->fd == fd) {
614 DBG("Channel found, updating kernel streams");
615 ret = kernel_open_channel_stream(channel);
616 if (ret < 0) {
617 goto error;
618 }
619
620 /*
621 * Have we already sent fds to the consumer? If yes, it means
622 * that tracing is started so it is safe to send our updated
623 * stream fds.
624 */
625 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
626 struct lttng_ht_iter iter;
627 struct consumer_socket *socket;
628
629
630 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
631 &iter.iter, socket, node.node) {
632 /* Code flow error */
633 assert(socket->fd >= 0);
634
635 pthread_mutex_lock(socket->lock);
636 ret = kernel_consumer_send_channel_stream(socket,
637 channel, ksess);
638 pthread_mutex_unlock(socket->lock);
639 if (ret < 0) {
640 goto error;
641 }
642 }
643 }
644 goto error;
645 }
646 }
647 session_unlock(session);
648 }
649 session_unlock_list();
650 return ret;
651
652 error:
653 session_unlock(session);
654 session_unlock_list();
655 return ret;
656 }
657
658 /*
659 * For each tracing session, update newly registered apps.
660 */
661 static void update_ust_app(int app_sock)
662 {
663 struct ltt_session *sess, *stmp;
664
665 session_lock_list();
666
667 /* For all tracing session(s) */
668 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
669 session_lock(sess);
670 if (sess->ust_session) {
671 ust_app_global_update(sess->ust_session, app_sock);
672 }
673 session_unlock(sess);
674 }
675
676 session_unlock_list();
677 }
678
679 /*
680 * This thread manage event coming from the kernel.
681 *
682 * Features supported in this thread:
683 * -) CPU Hotplug
684 */
685 static void *thread_manage_kernel(void *data)
686 {
687 int ret, i, pollfd, update_poll_flag = 1, err = -1;
688 uint32_t revents, nb_fd;
689 char tmp;
690 struct lttng_poll_event events;
691
692 DBG("[thread] Thread manage kernel started");
693
694 if (testpoint(thread_manage_kernel)) {
695 goto error_testpoint;
696 }
697
698 health_code_update(&health_thread_kernel);
699
700 ret = create_thread_poll_set(&events, 2);
701 if (ret < 0) {
702 goto error_poll_create;
703 }
704
705 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
706 if (ret < 0) {
707 goto error;
708 }
709
710 if (testpoint(thread_manage_kernel_before_loop)) {
711 goto error;
712 }
713
714 while (1) {
715 health_code_update(&health_thread_kernel);
716
717 if (update_poll_flag == 1) {
718 /*
719 * Reset number of fd in the poll set. Always 2 since there is the thread
720 * quit pipe and the kernel pipe.
721 */
722 events.nb_fd = 2;
723
724 ret = update_kernel_poll(&events);
725 if (ret < 0) {
726 goto error;
727 }
728 update_poll_flag = 0;
729 }
730
731 DBG("Thread kernel polling on %d fds", events.nb_fd);
732
733 /* Poll infinite value of time */
734 restart:
735 health_poll_update(&health_thread_kernel);
736 ret = lttng_poll_wait(&events, -1);
737 health_poll_update(&health_thread_kernel);
738 if (ret < 0) {
739 /*
740 * Restart interrupted system call.
741 */
742 if (errno == EINTR) {
743 goto restart;
744 }
745 goto error;
746 } else if (ret == 0) {
747 /* Should not happen since timeout is infinite */
748 ERR("Return value of poll is 0 with an infinite timeout.\n"
749 "This should not have happened! Continuing...");
750 continue;
751 }
752
753 nb_fd = ret;
754
755 for (i = 0; i < nb_fd; i++) {
756 /* Fetch once the poll data */
757 revents = LTTNG_POLL_GETEV(&events, i);
758 pollfd = LTTNG_POLL_GETFD(&events, i);
759
760 health_code_update(&health_thread_kernel);
761
762 /* Thread quit pipe has been closed. Killing thread. */
763 ret = check_thread_quit_pipe(pollfd, revents);
764 if (ret) {
765 err = 0;
766 goto exit;
767 }
768
769 /* Check for data on kernel pipe */
770 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
771 ret = read(kernel_poll_pipe[0], &tmp, 1);
772 update_poll_flag = 1;
773 continue;
774 } else {
775 /*
776 * New CPU detected by the kernel. Adding kernel stream to
777 * kernel session and updating the kernel consumer
778 */
779 if (revents & LPOLLIN) {
780 ret = update_kernel_stream(&kconsumer_data, pollfd);
781 if (ret < 0) {
782 continue;
783 }
784 break;
785 /*
786 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
787 * and unregister kernel stream at this point.
788 */
789 }
790 }
791 }
792 }
793
794 exit:
795 error:
796 lttng_poll_clean(&events);
797 error_poll_create:
798 error_testpoint:
799 utils_close_pipe(kernel_poll_pipe);
800 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
801 if (err) {
802 health_error(&health_thread_kernel);
803 ERR("Health error occurred in %s", __func__);
804 WARN("Kernel thread died unexpectedly. "
805 "Kernel tracing can continue but CPU hotplug is disabled.");
806 }
807 health_exit(&health_thread_kernel);
808 DBG("Kernel thread dying");
809 return NULL;
810 }
811
812 /*
813 * Signal pthread condition of the consumer data that the thread.
814 */
815 static void signal_consumer_condition(struct consumer_data *data, int state)
816 {
817 pthread_mutex_lock(&data->cond_mutex);
818
819 /*
820 * The state is set before signaling. It can be any value, it's the waiter
821 * job to correctly interpret this condition variable associated to the
822 * consumer pthread_cond.
823 *
824 * A value of 0 means that the corresponding thread of the consumer data
825 * was not started. 1 indicates that the thread has started and is ready
826 * for action. A negative value means that there was an error during the
827 * thread bootstrap.
828 */
829 data->consumer_thread_is_ready = state;
830 (void) pthread_cond_signal(&data->cond);
831
832 pthread_mutex_unlock(&data->cond_mutex);
833 }
834
835 /*
836 * This thread manage the consumer error sent back to the session daemon.
837 */
838 static void *thread_manage_consumer(void *data)
839 {
840 int sock = -1, i, ret, pollfd, err = -1;
841 uint32_t revents, nb_fd;
842 enum lttcomm_return_code code;
843 struct lttng_poll_event events;
844 struct consumer_data *consumer_data = data;
845
846 DBG("[thread] Manage consumer started");
847
848 /*
849 * Since the consumer thread can be spawned at any moment in time, we init
850 * the health to a poll status (1, which is a valid health over time).
851 * When the thread starts, we update here the health to a "code" path being
852 * an even value so this thread, when reaching a poll wait, does not
853 * trigger an error with an even value.
854 *
855 * Here is the use case we avoid.
856 *
857 * +1: the first poll update during initialization (main())
858 * +2 * x: multiple code update once in this thread.
859 * +1: poll wait in this thread (being a good health state).
860 * == even number which after the wait period shows as a bad health.
861 *
862 * In a nutshell, the following poll update to the health state brings back
863 * the state to an even value meaning a code path.
864 */
865 health_poll_update(&consumer_data->health);
866
867 /*
868 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
869 * Nothing more will be added to this poll set.
870 */
871 ret = create_thread_poll_set(&events, 2);
872 if (ret < 0) {
873 goto error_poll;
874 }
875
876 /*
877 * The error socket here is already in a listening state which was done
878 * just before spawning this thread to avoid a race between the consumer
879 * daemon exec trying to connect and the listen() call.
880 */
881 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
882 if (ret < 0) {
883 goto error;
884 }
885
886 health_code_update(&consumer_data->health);
887
888 /* Inifinite blocking call, waiting for transmission */
889 restart:
890 health_poll_update(&consumer_data->health);
891
892 if (testpoint(thread_manage_consumer)) {
893 goto error;
894 }
895
896 ret = lttng_poll_wait(&events, -1);
897 health_poll_update(&consumer_data->health);
898 if (ret < 0) {
899 /*
900 * Restart interrupted system call.
901 */
902 if (errno == EINTR) {
903 goto restart;
904 }
905 goto error;
906 }
907
908 nb_fd = ret;
909
910 for (i = 0; i < nb_fd; i++) {
911 /* Fetch once the poll data */
912 revents = LTTNG_POLL_GETEV(&events, i);
913 pollfd = LTTNG_POLL_GETFD(&events, i);
914
915 health_code_update(&consumer_data->health);
916
917 /* Thread quit pipe has been closed. Killing thread. */
918 ret = check_thread_quit_pipe(pollfd, revents);
919 if (ret) {
920 err = 0;
921 goto exit;
922 }
923
924 /* Event on the registration socket */
925 if (pollfd == consumer_data->err_sock) {
926 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
927 ERR("consumer err socket poll error");
928 goto error;
929 }
930 }
931 }
932
933 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
934 if (sock < 0) {
935 goto error;
936 }
937
938 /*
939 * Set the CLOEXEC flag. Return code is useless because either way, the
940 * show must go on.
941 */
942 (void) utils_set_fd_cloexec(sock);
943
944 health_code_update(&consumer_data->health);
945
946 DBG2("Receiving code from consumer err_sock");
947
948 /* Getting status code from kconsumerd */
949 ret = lttcomm_recv_unix_sock(sock, &code,
950 sizeof(enum lttcomm_return_code));
951 if (ret <= 0) {
952 goto error;
953 }
954
955 health_code_update(&consumer_data->health);
956
957 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
958 consumer_data->cmd_sock =
959 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
960 if (consumer_data->cmd_sock < 0) {
961 /* On error, signal condition and quit. */
962 signal_consumer_condition(consumer_data, -1);
963 PERROR("consumer connect");
964 goto error;
965 }
966 signal_consumer_condition(consumer_data, 1);
967 DBG("Consumer command socket ready");
968 } else {
969 ERR("consumer error when waiting for SOCK_READY : %s",
970 lttcomm_get_readable_code(-code));
971 goto error;
972 }
973
974 /* Remove the kconsumerd error sock since we've established a connexion */
975 ret = lttng_poll_del(&events, consumer_data->err_sock);
976 if (ret < 0) {
977 goto error;
978 }
979
980 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
981 if (ret < 0) {
982 goto error;
983 }
984
985 health_code_update(&consumer_data->health);
986
987 /* Inifinite blocking call, waiting for transmission */
988 restart_poll:
989 health_poll_update(&consumer_data->health);
990 ret = lttng_poll_wait(&events, -1);
991 health_poll_update(&consumer_data->health);
992 if (ret < 0) {
993 /*
994 * Restart interrupted system call.
995 */
996 if (errno == EINTR) {
997 goto restart_poll;
998 }
999 goto error;
1000 }
1001
1002 nb_fd = ret;
1003
1004 for (i = 0; i < nb_fd; i++) {
1005 /* Fetch once the poll data */
1006 revents = LTTNG_POLL_GETEV(&events, i);
1007 pollfd = LTTNG_POLL_GETFD(&events, i);
1008
1009 health_code_update(&consumer_data->health);
1010
1011 /* Thread quit pipe has been closed. Killing thread. */
1012 ret = check_thread_quit_pipe(pollfd, revents);
1013 if (ret) {
1014 err = 0;
1015 goto exit;
1016 }
1017
1018 /* Event on the kconsumerd socket */
1019 if (pollfd == sock) {
1020 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1021 ERR("consumer err socket second poll error");
1022 goto error;
1023 }
1024 }
1025 }
1026
1027 health_code_update(&consumer_data->health);
1028
1029 /* Wait for any kconsumerd error */
1030 ret = lttcomm_recv_unix_sock(sock, &code,
1031 sizeof(enum lttcomm_return_code));
1032 if (ret <= 0) {
1033 ERR("consumer closed the command socket");
1034 goto error;
1035 }
1036
1037 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1038
1039 exit:
1040 error:
1041 /* Immediately set the consumerd state to stopped */
1042 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1043 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1044 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1045 consumer_data->type == LTTNG_CONSUMER32_UST) {
1046 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1047 } else {
1048 /* Code flow error... */
1049 assert(0);
1050 }
1051
1052 if (consumer_data->err_sock >= 0) {
1053 ret = close(consumer_data->err_sock);
1054 if (ret) {
1055 PERROR("close");
1056 }
1057 }
1058 if (consumer_data->cmd_sock >= 0) {
1059 ret = close(consumer_data->cmd_sock);
1060 if (ret) {
1061 PERROR("close");
1062 }
1063 }
1064 if (sock >= 0) {
1065 ret = close(sock);
1066 if (ret) {
1067 PERROR("close");
1068 }
1069 }
1070
1071 unlink(consumer_data->err_unix_sock_path);
1072 unlink(consumer_data->cmd_unix_sock_path);
1073 consumer_data->pid = 0;
1074
1075 lttng_poll_clean(&events);
1076 error_poll:
1077 if (err) {
1078 health_error(&consumer_data->health);
1079 ERR("Health error occurred in %s", __func__);
1080 }
1081 health_exit(&consumer_data->health);
1082 DBG("consumer thread cleanup completed");
1083
1084 return NULL;
1085 }
1086
1087 /*
1088 * This thread manage application communication.
1089 */
1090 static void *thread_manage_apps(void *data)
1091 {
1092 int i, ret, pollfd, err = -1;
1093 uint32_t revents, nb_fd;
1094 struct ust_command ust_cmd;
1095 struct lttng_poll_event events;
1096
1097 DBG("[thread] Manage application started");
1098
1099 rcu_register_thread();
1100 rcu_thread_online();
1101
1102 if (testpoint(thread_manage_apps)) {
1103 goto error_testpoint;
1104 }
1105
1106 health_code_update(&health_thread_app_manage);
1107
1108 ret = create_thread_poll_set(&events, 2);
1109 if (ret < 0) {
1110 goto error_poll_create;
1111 }
1112
1113 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1114 if (ret < 0) {
1115 goto error;
1116 }
1117
1118 if (testpoint(thread_manage_apps_before_loop)) {
1119 goto error;
1120 }
1121
1122 health_code_update(&health_thread_app_manage);
1123
1124 while (1) {
1125 DBG("Apps thread polling on %d fds", events.nb_fd);
1126
1127 /* Inifinite blocking call, waiting for transmission */
1128 restart:
1129 health_poll_update(&health_thread_app_manage);
1130 ret = lttng_poll_wait(&events, -1);
1131 health_poll_update(&health_thread_app_manage);
1132 if (ret < 0) {
1133 /*
1134 * Restart interrupted system call.
1135 */
1136 if (errno == EINTR) {
1137 goto restart;
1138 }
1139 goto error;
1140 }
1141
1142 nb_fd = ret;
1143
1144 for (i = 0; i < nb_fd; i++) {
1145 /* Fetch once the poll data */
1146 revents = LTTNG_POLL_GETEV(&events, i);
1147 pollfd = LTTNG_POLL_GETFD(&events, i);
1148
1149 health_code_update(&health_thread_app_manage);
1150
1151 /* Thread quit pipe has been closed. Killing thread. */
1152 ret = check_thread_quit_pipe(pollfd, revents);
1153 if (ret) {
1154 err = 0;
1155 goto exit;
1156 }
1157
1158 /* Inspect the apps cmd pipe */
1159 if (pollfd == apps_cmd_pipe[0]) {
1160 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1161 ERR("Apps command pipe error");
1162 goto error;
1163 } else if (revents & LPOLLIN) {
1164 /* Empty pipe */
1165 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1166 if (ret < 0 || ret < sizeof(ust_cmd)) {
1167 PERROR("read apps cmd pipe");
1168 goto error;
1169 }
1170
1171 health_code_update(&health_thread_app_manage);
1172
1173 /* Register applicaton to the session daemon */
1174 ret = ust_app_register(&ust_cmd.reg_msg,
1175 ust_cmd.sock);
1176 if (ret == -ENOMEM) {
1177 goto error;
1178 } else if (ret < 0) {
1179 break;
1180 }
1181
1182 health_code_update(&health_thread_app_manage);
1183
1184 /*
1185 * Validate UST version compatibility.
1186 */
1187 ret = ust_app_validate_version(ust_cmd.sock);
1188 if (ret >= 0) {
1189 /*
1190 * Add channel(s) and event(s) to newly registered apps
1191 * from lttng global UST domain.
1192 */
1193 update_ust_app(ust_cmd.sock);
1194 }
1195
1196 health_code_update(&health_thread_app_manage);
1197
1198 ret = ust_app_register_done(ust_cmd.sock);
1199 if (ret < 0) {
1200 /*
1201 * If the registration is not possible, we simply
1202 * unregister the apps and continue
1203 */
1204 ust_app_unregister(ust_cmd.sock);
1205 } else {
1206 /*
1207 * We only monitor the error events of the socket. This
1208 * thread does not handle any incoming data from UST
1209 * (POLLIN).
1210 */
1211 ret = lttng_poll_add(&events, ust_cmd.sock,
1212 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1213 if (ret < 0) {
1214 goto error;
1215 }
1216
1217 /* Set socket timeout for both receiving and ending */
1218 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1219 app_socket_timeout);
1220 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1221 app_socket_timeout);
1222
1223 DBG("Apps with sock %d added to poll set",
1224 ust_cmd.sock);
1225 }
1226
1227 health_code_update(&health_thread_app_manage);
1228
1229 break;
1230 }
1231 } else {
1232 /*
1233 * At this point, we know that a registered application made
1234 * the event at poll_wait.
1235 */
1236 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1237 /* Removing from the poll set */
1238 ret = lttng_poll_del(&events, pollfd);
1239 if (ret < 0) {
1240 goto error;
1241 }
1242
1243 /* Socket closed on remote end. */
1244 ust_app_unregister(pollfd);
1245 break;
1246 }
1247 }
1248
1249 health_code_update(&health_thread_app_manage);
1250 }
1251 }
1252
1253 exit:
1254 error:
1255 lttng_poll_clean(&events);
1256 error_poll_create:
1257 error_testpoint:
1258 utils_close_pipe(apps_cmd_pipe);
1259 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1260
1261 /*
1262 * We don't clean the UST app hash table here since already registered
1263 * applications can still be controlled so let them be until the session
1264 * daemon dies or the applications stop.
1265 */
1266
1267 if (err) {
1268 health_error(&health_thread_app_manage);
1269 ERR("Health error occurred in %s", __func__);
1270 }
1271 health_exit(&health_thread_app_manage);
1272 DBG("Application communication apps thread cleanup complete");
1273 rcu_thread_offline();
1274 rcu_unregister_thread();
1275 return NULL;
1276 }
1277
1278 /*
1279 * Dispatch request from the registration threads to the application
1280 * communication thread.
1281 */
1282 static void *thread_dispatch_ust_registration(void *data)
1283 {
1284 int ret;
1285 struct cds_wfq_node *node;
1286 struct ust_command *ust_cmd = NULL;
1287
1288 DBG("[thread] Dispatch UST command started");
1289
1290 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1291 /* Atomically prepare the queue futex */
1292 futex_nto1_prepare(&ust_cmd_queue.futex);
1293
1294 do {
1295 /* Dequeue command for registration */
1296 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1297 if (node == NULL) {
1298 DBG("Woken up but nothing in the UST command queue");
1299 /* Continue thread execution */
1300 break;
1301 }
1302
1303 ust_cmd = caa_container_of(node, struct ust_command, node);
1304
1305 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1306 " gid:%d sock:%d name:%s (version %d.%d)",
1307 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1308 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1309 ust_cmd->sock, ust_cmd->reg_msg.name,
1310 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1311 /*
1312 * Inform apps thread of the new application registration. This
1313 * call is blocking so we can be assured that the data will be read
1314 * at some point in time or wait to the end of the world :)
1315 */
1316 if (apps_cmd_pipe[1] >= 0) {
1317 ret = write(apps_cmd_pipe[1], ust_cmd,
1318 sizeof(struct ust_command));
1319 if (ret < 0) {
1320 PERROR("write apps cmd pipe");
1321 if (errno == EBADF) {
1322 /*
1323 * We can't inform the application thread to process
1324 * registration. We will exit or else application
1325 * registration will not occur and tracing will never
1326 * start.
1327 */
1328 goto error;
1329 }
1330 }
1331 } else {
1332 /* Application manager thread is not available. */
1333 ret = close(ust_cmd->sock);
1334 if (ret < 0) {
1335 PERROR("close ust_cmd sock");
1336 }
1337 }
1338 free(ust_cmd);
1339 } while (node != NULL);
1340
1341 /* Futex wait on queue. Blocking call on futex() */
1342 futex_nto1_wait(&ust_cmd_queue.futex);
1343 }
1344
1345 error:
1346 DBG("Dispatch thread dying");
1347 return NULL;
1348 }
1349
1350 /*
1351 * This thread manage application registration.
1352 */
1353 static void *thread_registration_apps(void *data)
1354 {
1355 int sock = -1, i, ret, pollfd, err = -1;
1356 uint32_t revents, nb_fd;
1357 struct lttng_poll_event events;
1358 /*
1359 * Get allocated in this thread, enqueued to a global queue, dequeued and
1360 * freed in the manage apps thread.
1361 */
1362 struct ust_command *ust_cmd = NULL;
1363
1364 DBG("[thread] Manage application registration started");
1365
1366 if (testpoint(thread_registration_apps)) {
1367 goto error_testpoint;
1368 }
1369
1370 ret = lttcomm_listen_unix_sock(apps_sock);
1371 if (ret < 0) {
1372 goto error_listen;
1373 }
1374
1375 /*
1376 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1377 * more will be added to this poll set.
1378 */
1379 ret = create_thread_poll_set(&events, 2);
1380 if (ret < 0) {
1381 goto error_create_poll;
1382 }
1383
1384 /* Add the application registration socket */
1385 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1386 if (ret < 0) {
1387 goto error_poll_add;
1388 }
1389
1390 /* Notify all applications to register */
1391 ret = notify_ust_apps(1);
1392 if (ret < 0) {
1393 ERR("Failed to notify applications or create the wait shared memory.\n"
1394 "Execution continues but there might be problem for already\n"
1395 "running applications that wishes to register.");
1396 }
1397
1398 while (1) {
1399 DBG("Accepting application registration");
1400
1401 /* Inifinite blocking call, waiting for transmission */
1402 restart:
1403 health_poll_update(&health_thread_app_reg);
1404 ret = lttng_poll_wait(&events, -1);
1405 health_poll_update(&health_thread_app_reg);
1406 if (ret < 0) {
1407 /*
1408 * Restart interrupted system call.
1409 */
1410 if (errno == EINTR) {
1411 goto restart;
1412 }
1413 goto error;
1414 }
1415
1416 nb_fd = ret;
1417
1418 for (i = 0; i < nb_fd; i++) {
1419 health_code_update(&health_thread_app_reg);
1420
1421 /* Fetch once the poll data */
1422 revents = LTTNG_POLL_GETEV(&events, i);
1423 pollfd = LTTNG_POLL_GETFD(&events, i);
1424
1425 /* Thread quit pipe has been closed. Killing thread. */
1426 ret = check_thread_quit_pipe(pollfd, revents);
1427 if (ret) {
1428 err = 0;
1429 goto exit;
1430 }
1431
1432 /* Event on the registration socket */
1433 if (pollfd == apps_sock) {
1434 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1435 ERR("Register apps socket poll error");
1436 goto error;
1437 } else if (revents & LPOLLIN) {
1438 sock = lttcomm_accept_unix_sock(apps_sock);
1439 if (sock < 0) {
1440 goto error;
1441 }
1442
1443 /*
1444 * Set the CLOEXEC flag. Return code is useless because
1445 * either way, the show must go on.
1446 */
1447 (void) utils_set_fd_cloexec(sock);
1448
1449 /* Create UST registration command for enqueuing */
1450 ust_cmd = zmalloc(sizeof(struct ust_command));
1451 if (ust_cmd == NULL) {
1452 PERROR("ust command zmalloc");
1453 goto error;
1454 }
1455
1456 /*
1457 * Using message-based transmissions to ensure we don't
1458 * have to deal with partially received messages.
1459 */
1460 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1461 if (ret < 0) {
1462 ERR("Exhausted file descriptors allowed for applications.");
1463 free(ust_cmd);
1464 ret = close(sock);
1465 if (ret) {
1466 PERROR("close");
1467 }
1468 sock = -1;
1469 continue;
1470 }
1471 health_code_update(&health_thread_app_reg);
1472 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1473 sizeof(struct ust_register_msg));
1474 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1475 if (ret < 0) {
1476 PERROR("lttcomm_recv_unix_sock register apps");
1477 } else {
1478 ERR("Wrong size received on apps register");
1479 }
1480 free(ust_cmd);
1481 ret = close(sock);
1482 if (ret) {
1483 PERROR("close");
1484 }
1485 lttng_fd_put(LTTNG_FD_APPS, 1);
1486 sock = -1;
1487 continue;
1488 }
1489 health_code_update(&health_thread_app_reg);
1490
1491 ust_cmd->sock = sock;
1492 sock = -1;
1493
1494 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1495 " gid:%d sock:%d name:%s (version %d.%d)",
1496 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1497 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1498 ust_cmd->sock, ust_cmd->reg_msg.name,
1499 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1500
1501 /*
1502 * Lock free enqueue the registration request. The red pill
1503 * has been taken! This apps will be part of the *system*.
1504 */
1505 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1506
1507 /*
1508 * Wake the registration queue futex. Implicit memory
1509 * barrier with the exchange in cds_wfq_enqueue.
1510 */
1511 futex_nto1_wake(&ust_cmd_queue.futex);
1512 }
1513 }
1514 }
1515 }
1516
1517 exit:
1518 error:
1519 if (err) {
1520 health_error(&health_thread_app_reg);
1521 ERR("Health error occurred in %s", __func__);
1522 }
1523
1524 /* Notify that the registration thread is gone */
1525 notify_ust_apps(0);
1526
1527 if (apps_sock >= 0) {
1528 ret = close(apps_sock);
1529 if (ret) {
1530 PERROR("close");
1531 }
1532 }
1533 if (sock >= 0) {
1534 ret = close(sock);
1535 if (ret) {
1536 PERROR("close");
1537 }
1538 lttng_fd_put(LTTNG_FD_APPS, 1);
1539 }
1540 unlink(apps_unix_sock_path);
1541
1542 error_poll_add:
1543 lttng_poll_clean(&events);
1544 error_listen:
1545 error_create_poll:
1546 error_testpoint:
1547 DBG("UST Registration thread cleanup complete");
1548 health_exit(&health_thread_app_reg);
1549
1550 return NULL;
1551 }
1552
1553 /*
1554 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1555 * exec or it will fails.
1556 */
1557 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1558 {
1559 int ret, clock_ret;
1560 struct timespec timeout;
1561
1562 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1563 consumer_data->consumer_thread_is_ready = 0;
1564
1565 /* Setup pthread condition */
1566 ret = pthread_condattr_init(&consumer_data->condattr);
1567 if (ret != 0) {
1568 errno = ret;
1569 PERROR("pthread_condattr_init consumer data");
1570 goto error;
1571 }
1572
1573 /*
1574 * Set the monotonic clock in order to make sure we DO NOT jump in time
1575 * between the clock_gettime() call and the timedwait call. See bug #324
1576 * for a more details and how we noticed it.
1577 */
1578 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1579 if (ret != 0) {
1580 errno = ret;
1581 PERROR("pthread_condattr_setclock consumer data");
1582 goto error;
1583 }
1584
1585 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1586 if (ret != 0) {
1587 errno = ret;
1588 PERROR("pthread_cond_init consumer data");
1589 goto error;
1590 }
1591
1592 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1593 consumer_data);
1594 if (ret != 0) {
1595 PERROR("pthread_create consumer");
1596 ret = -1;
1597 goto error;
1598 }
1599
1600 /* We are about to wait on a pthread condition */
1601 pthread_mutex_lock(&consumer_data->cond_mutex);
1602
1603 /* Get time for sem_timedwait absolute timeout */
1604 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1605 /*
1606 * Set the timeout for the condition timed wait even if the clock gettime
1607 * call fails since we might loop on that call and we want to avoid to
1608 * increment the timeout too many times.
1609 */
1610 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1611
1612 /*
1613 * The following loop COULD be skipped in some conditions so this is why we
1614 * set ret to 0 in order to make sure at least one round of the loop is
1615 * done.
1616 */
1617 ret = 0;
1618
1619 /*
1620 * Loop until the condition is reached or when a timeout is reached. Note
1621 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1622 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1623 * possible. This loop does not take any chances and works with both of
1624 * them.
1625 */
1626 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1627 if (clock_ret < 0) {
1628 PERROR("clock_gettime spawn consumer");
1629 /* Infinite wait for the consumerd thread to be ready */
1630 ret = pthread_cond_wait(&consumer_data->cond,
1631 &consumer_data->cond_mutex);
1632 } else {
1633 ret = pthread_cond_timedwait(&consumer_data->cond,
1634 &consumer_data->cond_mutex, &timeout);
1635 }
1636 }
1637
1638 /* Release the pthread condition */
1639 pthread_mutex_unlock(&consumer_data->cond_mutex);
1640
1641 if (ret != 0) {
1642 errno = ret;
1643 if (ret == ETIMEDOUT) {
1644 /*
1645 * Call has timed out so we kill the kconsumerd_thread and return
1646 * an error.
1647 */
1648 ERR("Condition timed out. The consumer thread was never ready."
1649 " Killing it");
1650 ret = pthread_cancel(consumer_data->thread);
1651 if (ret < 0) {
1652 PERROR("pthread_cancel consumer thread");
1653 }
1654 } else {
1655 PERROR("pthread_cond_wait failed consumer thread");
1656 }
1657 goto error;
1658 }
1659
1660 pthread_mutex_lock(&consumer_data->pid_mutex);
1661 if (consumer_data->pid == 0) {
1662 ERR("Consumerd did not start");
1663 pthread_mutex_unlock(&consumer_data->pid_mutex);
1664 goto error;
1665 }
1666 pthread_mutex_unlock(&consumer_data->pid_mutex);
1667
1668 return 0;
1669
1670 error:
1671 return ret;
1672 }
1673
1674 /*
1675 * Join consumer thread
1676 */
1677 static int join_consumer_thread(struct consumer_data *consumer_data)
1678 {
1679 void *status;
1680
1681 /* Consumer pid must be a real one. */
1682 if (consumer_data->pid > 0) {
1683 int ret;
1684 ret = kill(consumer_data->pid, SIGTERM);
1685 if (ret) {
1686 ERR("Error killing consumer daemon");
1687 return ret;
1688 }
1689 return pthread_join(consumer_data->thread, &status);
1690 } else {
1691 return 0;
1692 }
1693 }
1694
1695 /*
1696 * Fork and exec a consumer daemon (consumerd).
1697 *
1698 * Return pid if successful else -1.
1699 */
1700 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1701 {
1702 int ret;
1703 pid_t pid;
1704 const char *consumer_to_use;
1705 const char *verbosity;
1706 struct stat st;
1707
1708 DBG("Spawning consumerd");
1709
1710 pid = fork();
1711 if (pid == 0) {
1712 /*
1713 * Exec consumerd.
1714 */
1715 if (opt_verbose_consumer) {
1716 verbosity = "--verbose";
1717 } else {
1718 verbosity = "--quiet";
1719 }
1720 switch (consumer_data->type) {
1721 case LTTNG_CONSUMER_KERNEL:
1722 /*
1723 * Find out which consumerd to execute. We will first try the
1724 * 64-bit path, then the sessiond's installation directory, and
1725 * fallback on the 32-bit one,
1726 */
1727 DBG3("Looking for a kernel consumer at these locations:");
1728 DBG3(" 1) %s", consumerd64_bin);
1729 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1730 DBG3(" 3) %s", consumerd32_bin);
1731 if (stat(consumerd64_bin, &st) == 0) {
1732 DBG3("Found location #1");
1733 consumer_to_use = consumerd64_bin;
1734 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1735 DBG3("Found location #2");
1736 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1737 } else if (stat(consumerd32_bin, &st) == 0) {
1738 DBG3("Found location #3");
1739 consumer_to_use = consumerd32_bin;
1740 } else {
1741 DBG("Could not find any valid consumerd executable");
1742 break;
1743 }
1744 DBG("Using kernel consumer at: %s", consumer_to_use);
1745 execl(consumer_to_use,
1746 "lttng-consumerd", verbosity, "-k",
1747 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1748 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1749 NULL);
1750 break;
1751 case LTTNG_CONSUMER64_UST:
1752 {
1753 char *tmpnew = NULL;
1754
1755 if (consumerd64_libdir[0] != '\0') {
1756 char *tmp;
1757 size_t tmplen;
1758
1759 tmp = getenv("LD_LIBRARY_PATH");
1760 if (!tmp) {
1761 tmp = "";
1762 }
1763 tmplen = strlen("LD_LIBRARY_PATH=")
1764 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1765 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1766 if (!tmpnew) {
1767 ret = -ENOMEM;
1768 goto error;
1769 }
1770 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1771 strcat(tmpnew, consumerd64_libdir);
1772 if (tmp[0] != '\0') {
1773 strcat(tmpnew, ":");
1774 strcat(tmpnew, tmp);
1775 }
1776 ret = putenv(tmpnew);
1777 if (ret) {
1778 ret = -errno;
1779 goto error;
1780 }
1781 }
1782 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1783 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1784 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1785 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1786 NULL);
1787 if (consumerd64_libdir[0] != '\0') {
1788 free(tmpnew);
1789 }
1790 if (ret) {
1791 goto error;
1792 }
1793 break;
1794 }
1795 case LTTNG_CONSUMER32_UST:
1796 {
1797 char *tmpnew = NULL;
1798
1799 if (consumerd32_libdir[0] != '\0') {
1800 char *tmp;
1801 size_t tmplen;
1802
1803 tmp = getenv("LD_LIBRARY_PATH");
1804 if (!tmp) {
1805 tmp = "";
1806 }
1807 tmplen = strlen("LD_LIBRARY_PATH=")
1808 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1809 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1810 if (!tmpnew) {
1811 ret = -ENOMEM;
1812 goto error;
1813 }
1814 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1815 strcat(tmpnew, consumerd32_libdir);
1816 if (tmp[0] != '\0') {
1817 strcat(tmpnew, ":");
1818 strcat(tmpnew, tmp);
1819 }
1820 ret = putenv(tmpnew);
1821 if (ret) {
1822 ret = -errno;
1823 goto error;
1824 }
1825 }
1826 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1827 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1828 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1829 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1830 NULL);
1831 if (consumerd32_libdir[0] != '\0') {
1832 free(tmpnew);
1833 }
1834 if (ret) {
1835 goto error;
1836 }
1837 break;
1838 }
1839 default:
1840 PERROR("unknown consumer type");
1841 exit(EXIT_FAILURE);
1842 }
1843 if (errno != 0) {
1844 PERROR("kernel start consumer exec");
1845 }
1846 exit(EXIT_FAILURE);
1847 } else if (pid > 0) {
1848 ret = pid;
1849 } else {
1850 PERROR("start consumer fork");
1851 ret = -errno;
1852 }
1853 error:
1854 return ret;
1855 }
1856
1857 /*
1858 * Spawn the consumerd daemon and session daemon thread.
1859 */
1860 static int start_consumerd(struct consumer_data *consumer_data)
1861 {
1862 int ret;
1863
1864 /*
1865 * Set the listen() state on the socket since there is a possible race
1866 * between the exec() of the consumer daemon and this call if place in the
1867 * consumer thread. See bug #366 for more details.
1868 */
1869 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1870 if (ret < 0) {
1871 goto error;
1872 }
1873
1874 pthread_mutex_lock(&consumer_data->pid_mutex);
1875 if (consumer_data->pid != 0) {
1876 pthread_mutex_unlock(&consumer_data->pid_mutex);
1877 goto end;
1878 }
1879
1880 ret = spawn_consumerd(consumer_data);
1881 if (ret < 0) {
1882 ERR("Spawning consumerd failed");
1883 pthread_mutex_unlock(&consumer_data->pid_mutex);
1884 goto error;
1885 }
1886
1887 /* Setting up the consumer_data pid */
1888 consumer_data->pid = ret;
1889 DBG2("Consumer pid %d", consumer_data->pid);
1890 pthread_mutex_unlock(&consumer_data->pid_mutex);
1891
1892 DBG2("Spawning consumer control thread");
1893 ret = spawn_consumer_thread(consumer_data);
1894 if (ret < 0) {
1895 ERR("Fatal error spawning consumer control thread");
1896 goto error;
1897 }
1898
1899 end:
1900 return 0;
1901
1902 error:
1903 /* Cleanup already created socket on error. */
1904 if (consumer_data->err_sock >= 0) {
1905 int err;
1906
1907 err = close(consumer_data->err_sock);
1908 if (err < 0) {
1909 PERROR("close consumer data error socket");
1910 }
1911 }
1912 return ret;
1913 }
1914
1915 /*
1916 * Compute health status of each consumer. If one of them is zero (bad
1917 * state), we return 0.
1918 */
1919 static int check_consumer_health(void)
1920 {
1921 int ret;
1922
1923 ret = health_check_state(&kconsumer_data.health) &&
1924 health_check_state(&ustconsumer32_data.health) &&
1925 health_check_state(&ustconsumer64_data.health);
1926
1927 DBG3("Health consumer check %d", ret);
1928
1929 return ret;
1930 }
1931
1932 /*
1933 * Setup necessary data for kernel tracer action.
1934 */
1935 static int init_kernel_tracer(void)
1936 {
1937 int ret;
1938
1939 /* Modprobe lttng kernel modules */
1940 ret = modprobe_lttng_control();
1941 if (ret < 0) {
1942 goto error;
1943 }
1944
1945 /* Open debugfs lttng */
1946 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1947 if (kernel_tracer_fd < 0) {
1948 DBG("Failed to open %s", module_proc_lttng);
1949 ret = -1;
1950 goto error_open;
1951 }
1952
1953 /* Validate kernel version */
1954 ret = kernel_validate_version(kernel_tracer_fd);
1955 if (ret < 0) {
1956 goto error_version;
1957 }
1958
1959 ret = modprobe_lttng_data();
1960 if (ret < 0) {
1961 goto error_modules;
1962 }
1963
1964 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1965 return 0;
1966
1967 error_version:
1968 modprobe_remove_lttng_control();
1969 ret = close(kernel_tracer_fd);
1970 if (ret) {
1971 PERROR("close");
1972 }
1973 kernel_tracer_fd = -1;
1974 return LTTNG_ERR_KERN_VERSION;
1975
1976 error_modules:
1977 ret = close(kernel_tracer_fd);
1978 if (ret) {
1979 PERROR("close");
1980 }
1981
1982 error_open:
1983 modprobe_remove_lttng_control();
1984
1985 error:
1986 WARN("No kernel tracer available");
1987 kernel_tracer_fd = -1;
1988 if (!is_root) {
1989 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1990 } else {
1991 return LTTNG_ERR_KERN_NA;
1992 }
1993 }
1994
1995
1996 /*
1997 * Copy consumer output from the tracing session to the domain session. The
1998 * function also applies the right modification on a per domain basis for the
1999 * trace files destination directory.
2000 */
2001 static int copy_session_consumer(int domain, struct ltt_session *session)
2002 {
2003 int ret;
2004 const char *dir_name;
2005 struct consumer_output *consumer;
2006
2007 assert(session);
2008 assert(session->consumer);
2009
2010 switch (domain) {
2011 case LTTNG_DOMAIN_KERNEL:
2012 DBG3("Copying tracing session consumer output in kernel session");
2013 /*
2014 * XXX: We should audit the session creation and what this function
2015 * does "extra" in order to avoid a destroy since this function is used
2016 * in the domain session creation (kernel and ust) only. Same for UST
2017 * domain.
2018 */
2019 if (session->kernel_session->consumer) {
2020 consumer_destroy_output(session->kernel_session->consumer);
2021 }
2022 session->kernel_session->consumer =
2023 consumer_copy_output(session->consumer);
2024 /* Ease our life a bit for the next part */
2025 consumer = session->kernel_session->consumer;
2026 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2027 break;
2028 case LTTNG_DOMAIN_UST:
2029 DBG3("Copying tracing session consumer output in UST session");
2030 if (session->ust_session->consumer) {
2031 consumer_destroy_output(session->ust_session->consumer);
2032 }
2033 session->ust_session->consumer =
2034 consumer_copy_output(session->consumer);
2035 /* Ease our life a bit for the next part */
2036 consumer = session->ust_session->consumer;
2037 dir_name = DEFAULT_UST_TRACE_DIR;
2038 break;
2039 default:
2040 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2041 goto error;
2042 }
2043
2044 /* Append correct directory to subdir */
2045 strncat(consumer->subdir, dir_name,
2046 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2047 DBG3("Copy session consumer subdir %s", consumer->subdir);
2048
2049 ret = LTTNG_OK;
2050
2051 error:
2052 return ret;
2053 }
2054
2055 /*
2056 * Create an UST session and add it to the session ust list.
2057 */
2058 static int create_ust_session(struct ltt_session *session,
2059 struct lttng_domain *domain)
2060 {
2061 int ret;
2062 struct ltt_ust_session *lus = NULL;
2063
2064 assert(session);
2065 assert(domain);
2066 assert(session->consumer);
2067
2068 switch (domain->type) {
2069 case LTTNG_DOMAIN_UST:
2070 break;
2071 default:
2072 ERR("Unknown UST domain on create session %d", domain->type);
2073 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2074 goto error;
2075 }
2076
2077 DBG("Creating UST session");
2078
2079 lus = trace_ust_create_session(session->path, session->id, domain);
2080 if (lus == NULL) {
2081 ret = LTTNG_ERR_UST_SESS_FAIL;
2082 goto error;
2083 }
2084
2085 lus->uid = session->uid;
2086 lus->gid = session->gid;
2087 session->ust_session = lus;
2088
2089 /* Copy session output to the newly created UST session */
2090 ret = copy_session_consumer(domain->type, session);
2091 if (ret != LTTNG_OK) {
2092 goto error;
2093 }
2094
2095 return LTTNG_OK;
2096
2097 error:
2098 free(lus);
2099 session->ust_session = NULL;
2100 return ret;
2101 }
2102
2103 /*
2104 * Create a kernel tracer session then create the default channel.
2105 */
2106 static int create_kernel_session(struct ltt_session *session)
2107 {
2108 int ret;
2109
2110 DBG("Creating kernel session");
2111
2112 ret = kernel_create_session(session, kernel_tracer_fd);
2113 if (ret < 0) {
2114 ret = LTTNG_ERR_KERN_SESS_FAIL;
2115 goto error;
2116 }
2117
2118 /* Code flow safety */
2119 assert(session->kernel_session);
2120
2121 /* Copy session output to the newly created Kernel session */
2122 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2123 if (ret != LTTNG_OK) {
2124 goto error;
2125 }
2126
2127 /* Create directory(ies) on local filesystem. */
2128 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2129 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2130 ret = run_as_mkdir_recursive(
2131 session->kernel_session->consumer->dst.trace_path,
2132 S_IRWXU | S_IRWXG, session->uid, session->gid);
2133 if (ret < 0) {
2134 if (ret != -EEXIST) {
2135 ERR("Trace directory creation error");
2136 goto error;
2137 }
2138 }
2139 }
2140
2141 session->kernel_session->uid = session->uid;
2142 session->kernel_session->gid = session->gid;
2143
2144 return LTTNG_OK;
2145
2146 error:
2147 trace_kernel_destroy_session(session->kernel_session);
2148 session->kernel_session = NULL;
2149 return ret;
2150 }
2151
2152 /*
2153 * Count number of session permitted by uid/gid.
2154 */
2155 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2156 {
2157 unsigned int i = 0;
2158 struct ltt_session *session;
2159
2160 DBG("Counting number of available session for UID %d GID %d",
2161 uid, gid);
2162 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2163 /*
2164 * Only list the sessions the user can control.
2165 */
2166 if (!session_access_ok(session, uid, gid)) {
2167 continue;
2168 }
2169 i++;
2170 }
2171 return i;
2172 }
2173
2174 /*
2175 * Process the command requested by the lttng client within the command
2176 * context structure. This function make sure that the return structure (llm)
2177 * is set and ready for transmission before returning.
2178 *
2179 * Return any error encountered or 0 for success.
2180 *
2181 * "sock" is only used for special-case var. len data.
2182 */
2183 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2184 int *sock_error)
2185 {
2186 int ret = LTTNG_OK;
2187 int need_tracing_session = 1;
2188 int need_domain;
2189
2190 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2191
2192 *sock_error = 0;
2193
2194 switch (cmd_ctx->lsm->cmd_type) {
2195 case LTTNG_CREATE_SESSION:
2196 case LTTNG_DESTROY_SESSION:
2197 case LTTNG_LIST_SESSIONS:
2198 case LTTNG_LIST_DOMAINS:
2199 case LTTNG_START_TRACE:
2200 case LTTNG_STOP_TRACE:
2201 case LTTNG_DATA_PENDING:
2202 need_domain = 0;
2203 break;
2204 default:
2205 need_domain = 1;
2206 }
2207
2208 if (opt_no_kernel && need_domain
2209 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2210 if (!is_root) {
2211 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2212 } else {
2213 ret = LTTNG_ERR_KERN_NA;
2214 }
2215 goto error;
2216 }
2217
2218 /* Deny register consumer if we already have a spawned consumer. */
2219 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2220 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2221 if (kconsumer_data.pid > 0) {
2222 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2223 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2224 goto error;
2225 }
2226 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2227 }
2228
2229 /*
2230 * Check for command that don't needs to allocate a returned payload. We do
2231 * this here so we don't have to make the call for no payload at each
2232 * command.
2233 */
2234 switch(cmd_ctx->lsm->cmd_type) {
2235 case LTTNG_LIST_SESSIONS:
2236 case LTTNG_LIST_TRACEPOINTS:
2237 case LTTNG_LIST_TRACEPOINT_FIELDS:
2238 case LTTNG_LIST_DOMAINS:
2239 case LTTNG_LIST_CHANNELS:
2240 case LTTNG_LIST_EVENTS:
2241 break;
2242 default:
2243 /* Setup lttng message with no payload */
2244 ret = setup_lttng_msg(cmd_ctx, 0);
2245 if (ret < 0) {
2246 /* This label does not try to unlock the session */
2247 goto init_setup_error;
2248 }
2249 }
2250
2251 /* Commands that DO NOT need a session. */
2252 switch (cmd_ctx->lsm->cmd_type) {
2253 case LTTNG_CREATE_SESSION:
2254 case LTTNG_CALIBRATE:
2255 case LTTNG_LIST_SESSIONS:
2256 case LTTNG_LIST_TRACEPOINTS:
2257 case LTTNG_LIST_TRACEPOINT_FIELDS:
2258 need_tracing_session = 0;
2259 break;
2260 default:
2261 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2262 /*
2263 * We keep the session list lock across _all_ commands
2264 * for now, because the per-session lock does not
2265 * handle teardown properly.
2266 */
2267 session_lock_list();
2268 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2269 if (cmd_ctx->session == NULL) {
2270 if (cmd_ctx->lsm->session.name != NULL) {
2271 ret = LTTNG_ERR_SESS_NOT_FOUND;
2272 } else {
2273 /* If no session name specified */
2274 ret = LTTNG_ERR_SELECT_SESS;
2275 }
2276 goto error;
2277 } else {
2278 /* Acquire lock for the session */
2279 session_lock(cmd_ctx->session);
2280 }
2281 break;
2282 }
2283
2284 if (!need_domain) {
2285 goto skip_domain;
2286 }
2287
2288 /*
2289 * Check domain type for specific "pre-action".
2290 */
2291 switch (cmd_ctx->lsm->domain.type) {
2292 case LTTNG_DOMAIN_KERNEL:
2293 if (!is_root) {
2294 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2295 goto error;
2296 }
2297
2298 /* Kernel tracer check */
2299 if (kernel_tracer_fd == -1) {
2300 /* Basically, load kernel tracer modules */
2301 ret = init_kernel_tracer();
2302 if (ret != 0) {
2303 goto error;
2304 }
2305 }
2306
2307 /* Consumer is in an ERROR state. Report back to client */
2308 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2309 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2310 goto error;
2311 }
2312
2313 /* Need a session for kernel command */
2314 if (need_tracing_session) {
2315 if (cmd_ctx->session->kernel_session == NULL) {
2316 ret = create_kernel_session(cmd_ctx->session);
2317 if (ret < 0) {
2318 ret = LTTNG_ERR_KERN_SESS_FAIL;
2319 goto error;
2320 }
2321 }
2322
2323 /* Start the kernel consumer daemon */
2324 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2325 if (kconsumer_data.pid == 0 &&
2326 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2327 cmd_ctx->session->start_consumer) {
2328 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2329 ret = start_consumerd(&kconsumer_data);
2330 if (ret < 0) {
2331 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2332 goto error;
2333 }
2334 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2335 } else {
2336 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2337 }
2338
2339 /*
2340 * The consumer was just spawned so we need to add the socket to
2341 * the consumer output of the session if exist.
2342 */
2343 ret = consumer_create_socket(&kconsumer_data,
2344 cmd_ctx->session->kernel_session->consumer);
2345 if (ret < 0) {
2346 goto error;
2347 }
2348 }
2349
2350 break;
2351 case LTTNG_DOMAIN_UST:
2352 {
2353 /* Consumer is in an ERROR state. Report back to client */
2354 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2355 ret = LTTNG_ERR_NO_USTCONSUMERD;
2356 goto error;
2357 }
2358
2359 if (need_tracing_session) {
2360 /* Create UST session if none exist. */
2361 if (cmd_ctx->session->ust_session == NULL) {
2362 ret = create_ust_session(cmd_ctx->session,
2363 &cmd_ctx->lsm->domain);
2364 if (ret != LTTNG_OK) {
2365 goto error;
2366 }
2367 }
2368
2369 /* Start the UST consumer daemons */
2370 /* 64-bit */
2371 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2372 if (consumerd64_bin[0] != '\0' &&
2373 ustconsumer64_data.pid == 0 &&
2374 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2375 cmd_ctx->session->start_consumer) {
2376 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2377 ret = start_consumerd(&ustconsumer64_data);
2378 if (ret < 0) {
2379 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2380 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2381 goto error;
2382 }
2383
2384 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2385 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2386 } else {
2387 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2388 }
2389
2390 /*
2391 * Setup socket for consumer 64 bit. No need for atomic access
2392 * since it was set above and can ONLY be set in this thread.
2393 */
2394 ret = consumer_create_socket(&ustconsumer64_data,
2395 cmd_ctx->session->ust_session->consumer);
2396 if (ret < 0) {
2397 goto error;
2398 }
2399
2400 /* 32-bit */
2401 if (consumerd32_bin[0] != '\0' &&
2402 ustconsumer32_data.pid == 0 &&
2403 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2404 cmd_ctx->session->start_consumer) {
2405 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2406 ret = start_consumerd(&ustconsumer32_data);
2407 if (ret < 0) {
2408 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2409 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2410 goto error;
2411 }
2412
2413 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2414 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2415 } else {
2416 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2417 }
2418
2419 /*
2420 * Setup socket for consumer 64 bit. No need for atomic access
2421 * since it was set above and can ONLY be set in this thread.
2422 */
2423 ret = consumer_create_socket(&ustconsumer32_data,
2424 cmd_ctx->session->ust_session->consumer);
2425 if (ret < 0) {
2426 goto error;
2427 }
2428 }
2429 break;
2430 }
2431 default:
2432 break;
2433 }
2434 skip_domain:
2435
2436 /* Validate consumer daemon state when start/stop trace command */
2437 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2438 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2439 switch (cmd_ctx->lsm->domain.type) {
2440 case LTTNG_DOMAIN_UST:
2441 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2442 ret = LTTNG_ERR_NO_USTCONSUMERD;
2443 goto error;
2444 }
2445 break;
2446 case LTTNG_DOMAIN_KERNEL:
2447 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2448 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2449 goto error;
2450 }
2451 break;
2452 }
2453 }
2454
2455 /*
2456 * Check that the UID or GID match that of the tracing session.
2457 * The root user can interact with all sessions.
2458 */
2459 if (need_tracing_session) {
2460 if (!session_access_ok(cmd_ctx->session,
2461 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2462 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2463 ret = LTTNG_ERR_EPERM;
2464 goto error;
2465 }
2466 }
2467
2468 /* Process by command type */
2469 switch (cmd_ctx->lsm->cmd_type) {
2470 case LTTNG_ADD_CONTEXT:
2471 {
2472 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2473 cmd_ctx->lsm->u.context.channel_name,
2474 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2475 break;
2476 }
2477 case LTTNG_DISABLE_CHANNEL:
2478 {
2479 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2480 cmd_ctx->lsm->u.disable.channel_name);
2481 break;
2482 }
2483 case LTTNG_DISABLE_EVENT:
2484 {
2485 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2486 cmd_ctx->lsm->u.disable.channel_name,
2487 cmd_ctx->lsm->u.disable.name);
2488 break;
2489 }
2490 case LTTNG_DISABLE_ALL_EVENT:
2491 {
2492 DBG("Disabling all events");
2493
2494 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2495 cmd_ctx->lsm->u.disable.channel_name);
2496 break;
2497 }
2498 case LTTNG_DISABLE_CONSUMER:
2499 {
2500 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2501 break;
2502 }
2503 case LTTNG_ENABLE_CHANNEL:
2504 {
2505 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2506 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2507 break;
2508 }
2509 case LTTNG_ENABLE_CONSUMER:
2510 {
2511 /*
2512 * XXX: 0 means that this URI should be applied on the session. Should
2513 * be a DOMAIN enuam.
2514 */
2515 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2516 if (ret != LTTNG_OK) {
2517 goto error;
2518 }
2519
2520 if (cmd_ctx->lsm->domain.type == 0) {
2521 /* Add the URI for the UST session if a consumer is present. */
2522 if (cmd_ctx->session->ust_session &&
2523 cmd_ctx->session->ust_session->consumer) {
2524 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2525 } else if (cmd_ctx->session->kernel_session &&
2526 cmd_ctx->session->kernel_session->consumer) {
2527 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2528 cmd_ctx->session);
2529 }
2530 }
2531 break;
2532 }
2533 case LTTNG_ENABLE_EVENT:
2534 {
2535 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2536 cmd_ctx->lsm->u.enable.channel_name,
2537 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2538 break;
2539 }
2540 case LTTNG_ENABLE_ALL_EVENT:
2541 {
2542 DBG("Enabling all events");
2543
2544 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2545 cmd_ctx->lsm->u.enable.channel_name,
2546 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2547 break;
2548 }
2549 case LTTNG_LIST_TRACEPOINTS:
2550 {
2551 struct lttng_event *events;
2552 ssize_t nb_events;
2553
2554 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2555 if (nb_events < 0) {
2556 /* Return value is a negative lttng_error_code. */
2557 ret = -nb_events;
2558 goto error;
2559 }
2560
2561 /*
2562 * Setup lttng message with payload size set to the event list size in
2563 * bytes and then copy list into the llm payload.
2564 */
2565 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2566 if (ret < 0) {
2567 free(events);
2568 goto setup_error;
2569 }
2570
2571 /* Copy event list into message payload */
2572 memcpy(cmd_ctx->llm->payload, events,
2573 sizeof(struct lttng_event) * nb_events);
2574
2575 free(events);
2576
2577 ret = LTTNG_OK;
2578 break;
2579 }
2580 case LTTNG_LIST_TRACEPOINT_FIELDS:
2581 {
2582 struct lttng_event_field *fields;
2583 ssize_t nb_fields;
2584
2585 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2586 &fields);
2587 if (nb_fields < 0) {
2588 /* Return value is a negative lttng_error_code. */
2589 ret = -nb_fields;
2590 goto error;
2591 }
2592
2593 /*
2594 * Setup lttng message with payload size set to the event list size in
2595 * bytes and then copy list into the llm payload.
2596 */
2597 ret = setup_lttng_msg(cmd_ctx,
2598 sizeof(struct lttng_event_field) * nb_fields);
2599 if (ret < 0) {
2600 free(fields);
2601 goto setup_error;
2602 }
2603
2604 /* Copy event list into message payload */
2605 memcpy(cmd_ctx->llm->payload, fields,
2606 sizeof(struct lttng_event_field) * nb_fields);
2607
2608 free(fields);
2609
2610 ret = LTTNG_OK;
2611 break;
2612 }
2613 case LTTNG_SET_CONSUMER_URI:
2614 {
2615 size_t nb_uri, len;
2616 struct lttng_uri *uris;
2617
2618 nb_uri = cmd_ctx->lsm->u.uri.size;
2619 len = nb_uri * sizeof(struct lttng_uri);
2620
2621 if (nb_uri == 0) {
2622 ret = LTTNG_ERR_INVALID;
2623 goto error;
2624 }
2625
2626 uris = zmalloc(len);
2627 if (uris == NULL) {
2628 ret = LTTNG_ERR_FATAL;
2629 goto error;
2630 }
2631
2632 /* Receive variable len data */
2633 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2634 ret = lttcomm_recv_unix_sock(sock, uris, len);
2635 if (ret <= 0) {
2636 DBG("No URIs received from client... continuing");
2637 *sock_error = 1;
2638 ret = LTTNG_ERR_SESSION_FAIL;
2639 free(uris);
2640 goto error;
2641 }
2642
2643 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2644 nb_uri, uris);
2645 if (ret != LTTNG_OK) {
2646 free(uris);
2647 goto error;
2648 }
2649
2650 /*
2651 * XXX: 0 means that this URI should be applied on the session. Should
2652 * be a DOMAIN enuam.
2653 */
2654 if (cmd_ctx->lsm->domain.type == 0) {
2655 /* Add the URI for the UST session if a consumer is present. */
2656 if (cmd_ctx->session->ust_session &&
2657 cmd_ctx->session->ust_session->consumer) {
2658 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2659 nb_uri, uris);
2660 } else if (cmd_ctx->session->kernel_session &&
2661 cmd_ctx->session->kernel_session->consumer) {
2662 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2663 cmd_ctx->session, nb_uri, uris);
2664 }
2665 }
2666
2667 free(uris);
2668
2669 break;
2670 }
2671 case LTTNG_START_TRACE:
2672 {
2673 ret = cmd_start_trace(cmd_ctx->session);
2674 break;
2675 }
2676 case LTTNG_STOP_TRACE:
2677 {
2678 ret = cmd_stop_trace(cmd_ctx->session);
2679 break;
2680 }
2681 case LTTNG_CREATE_SESSION:
2682 {
2683 size_t nb_uri, len;
2684 struct lttng_uri *uris = NULL;
2685
2686 nb_uri = cmd_ctx->lsm->u.uri.size;
2687 len = nb_uri * sizeof(struct lttng_uri);
2688
2689 if (nb_uri > 0) {
2690 uris = zmalloc(len);
2691 if (uris == NULL) {
2692 ret = LTTNG_ERR_FATAL;
2693 goto error;
2694 }
2695
2696 /* Receive variable len data */
2697 DBG("Waiting for %zu URIs from client ...", nb_uri);
2698 ret = lttcomm_recv_unix_sock(sock, uris, len);
2699 if (ret <= 0) {
2700 DBG("No URIs received from client... continuing");
2701 *sock_error = 1;
2702 ret = LTTNG_ERR_SESSION_FAIL;
2703 free(uris);
2704 goto error;
2705 }
2706
2707 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2708 DBG("Creating session with ONE network URI is a bad call");
2709 ret = LTTNG_ERR_SESSION_FAIL;
2710 free(uris);
2711 goto error;
2712 }
2713 }
2714
2715 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2716 &cmd_ctx->creds);
2717
2718 free(uris);
2719
2720 break;
2721 }
2722 case LTTNG_DESTROY_SESSION:
2723 {
2724 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2725
2726 /* Set session to NULL so we do not unlock it after free. */
2727 cmd_ctx->session = NULL;
2728 break;
2729 }
2730 case LTTNG_LIST_DOMAINS:
2731 {
2732 ssize_t nb_dom;
2733 struct lttng_domain *domains;
2734
2735 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2736 if (nb_dom < 0) {
2737 /* Return value is a negative lttng_error_code. */
2738 ret = -nb_dom;
2739 goto error;
2740 }
2741
2742 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2743 if (ret < 0) {
2744 goto setup_error;
2745 }
2746
2747 /* Copy event list into message payload */
2748 memcpy(cmd_ctx->llm->payload, domains,
2749 nb_dom * sizeof(struct lttng_domain));
2750
2751 free(domains);
2752
2753 ret = LTTNG_OK;
2754 break;
2755 }
2756 case LTTNG_LIST_CHANNELS:
2757 {
2758 int nb_chan;
2759 struct lttng_channel *channels;
2760
2761 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2762 cmd_ctx->session, &channels);
2763 if (nb_chan < 0) {
2764 /* Return value is a negative lttng_error_code. */
2765 ret = -nb_chan;
2766 goto error;
2767 }
2768
2769 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2770 if (ret < 0) {
2771 goto setup_error;
2772 }
2773
2774 /* Copy event list into message payload */
2775 memcpy(cmd_ctx->llm->payload, channels,
2776 nb_chan * sizeof(struct lttng_channel));
2777
2778 free(channels);
2779
2780 ret = LTTNG_OK;
2781 break;
2782 }
2783 case LTTNG_LIST_EVENTS:
2784 {
2785 ssize_t nb_event;
2786 struct lttng_event *events = NULL;
2787
2788 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2789 cmd_ctx->lsm->u.list.channel_name, &events);
2790 if (nb_event < 0) {
2791 /* Return value is a negative lttng_error_code. */
2792 ret = -nb_event;
2793 goto error;
2794 }
2795
2796 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2797 if (ret < 0) {
2798 goto setup_error;
2799 }
2800
2801 /* Copy event list into message payload */
2802 memcpy(cmd_ctx->llm->payload, events,
2803 nb_event * sizeof(struct lttng_event));
2804
2805 free(events);
2806
2807 ret = LTTNG_OK;
2808 break;
2809 }
2810 case LTTNG_LIST_SESSIONS:
2811 {
2812 unsigned int nr_sessions;
2813
2814 session_lock_list();
2815 nr_sessions = lttng_sessions_count(
2816 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2817 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2818
2819 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2820 if (ret < 0) {
2821 session_unlock_list();
2822 goto setup_error;
2823 }
2824
2825 /* Filled the session array */
2826 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2827 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2828 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2829
2830 session_unlock_list();
2831
2832 ret = LTTNG_OK;
2833 break;
2834 }
2835 case LTTNG_CALIBRATE:
2836 {
2837 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2838 &cmd_ctx->lsm->u.calibrate);
2839 break;
2840 }
2841 case LTTNG_REGISTER_CONSUMER:
2842 {
2843 struct consumer_data *cdata;
2844
2845 switch (cmd_ctx->lsm->domain.type) {
2846 case LTTNG_DOMAIN_KERNEL:
2847 cdata = &kconsumer_data;
2848 break;
2849 default:
2850 ret = LTTNG_ERR_UND;
2851 goto error;
2852 }
2853
2854 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2855 cmd_ctx->lsm->u.reg.path, cdata);
2856 break;
2857 }
2858 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2859 {
2860 struct lttng_filter_bytecode *bytecode;
2861
2862 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2863 ret = LTTNG_ERR_FILTER_INVAL;
2864 goto error;
2865 }
2866 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2867 ret = LTTNG_ERR_FILTER_INVAL;
2868 goto error;
2869 }
2870 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2871 if (!bytecode) {
2872 ret = LTTNG_ERR_FILTER_NOMEM;
2873 goto error;
2874 }
2875 /* Receive var. len. data */
2876 DBG("Receiving var len data from client ...");
2877 ret = lttcomm_recv_unix_sock(sock, bytecode,
2878 cmd_ctx->lsm->u.enable.bytecode_len);
2879 if (ret <= 0) {
2880 DBG("Nothing recv() from client var len data... continuing");
2881 *sock_error = 1;
2882 ret = LTTNG_ERR_FILTER_INVAL;
2883 goto error;
2884 }
2885
2886 if (bytecode->len + sizeof(*bytecode)
2887 != cmd_ctx->lsm->u.enable.bytecode_len) {
2888 free(bytecode);
2889 ret = LTTNG_ERR_FILTER_INVAL;
2890 goto error;
2891 }
2892
2893 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2894 cmd_ctx->lsm->u.enable.channel_name,
2895 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2896 break;
2897 }
2898 case LTTNG_DATA_PENDING:
2899 {
2900 ret = cmd_data_pending(cmd_ctx->session);
2901 break;
2902 }
2903 default:
2904 ret = LTTNG_ERR_UND;
2905 break;
2906 }
2907
2908 error:
2909 if (cmd_ctx->llm == NULL) {
2910 DBG("Missing llm structure. Allocating one.");
2911 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2912 goto setup_error;
2913 }
2914 }
2915 /* Set return code */
2916 cmd_ctx->llm->ret_code = ret;
2917 setup_error:
2918 if (cmd_ctx->session) {
2919 session_unlock(cmd_ctx->session);
2920 }
2921 if (need_tracing_session) {
2922 session_unlock_list();
2923 }
2924 init_setup_error:
2925 return ret;
2926 }
2927
2928 /*
2929 * Thread managing health check socket.
2930 */
2931 static void *thread_manage_health(void *data)
2932 {
2933 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2934 uint32_t revents, nb_fd;
2935 struct lttng_poll_event events;
2936 struct lttcomm_health_msg msg;
2937 struct lttcomm_health_data reply;
2938
2939 DBG("[thread] Manage health check started");
2940
2941 rcu_register_thread();
2942
2943 /* Create unix socket */
2944 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2945 if (sock < 0) {
2946 ERR("Unable to create health check Unix socket");
2947 ret = -1;
2948 goto error;
2949 }
2950
2951 /*
2952 * Set the CLOEXEC flag. Return code is useless because either way, the
2953 * show must go on.
2954 */
2955 (void) utils_set_fd_cloexec(sock);
2956
2957 ret = lttcomm_listen_unix_sock(sock);
2958 if (ret < 0) {
2959 goto error;
2960 }
2961
2962 /*
2963 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2964 * more will be added to this poll set.
2965 */
2966 ret = create_thread_poll_set(&events, 2);
2967 if (ret < 0) {
2968 goto error;
2969 }
2970
2971 /* Add the application registration socket */
2972 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2973 if (ret < 0) {
2974 goto error;
2975 }
2976
2977 while (1) {
2978 DBG("Health check ready");
2979
2980 /* Inifinite blocking call, waiting for transmission */
2981 restart:
2982 ret = lttng_poll_wait(&events, -1);
2983 if (ret < 0) {
2984 /*
2985 * Restart interrupted system call.
2986 */
2987 if (errno == EINTR) {
2988 goto restart;
2989 }
2990 goto error;
2991 }
2992
2993 nb_fd = ret;
2994
2995 for (i = 0; i < nb_fd; i++) {
2996 /* Fetch once the poll data */
2997 revents = LTTNG_POLL_GETEV(&events, i);
2998 pollfd = LTTNG_POLL_GETFD(&events, i);
2999
3000 /* Thread quit pipe has been closed. Killing thread. */
3001 ret = check_thread_quit_pipe(pollfd, revents);
3002 if (ret) {
3003 err = 0;
3004 goto exit;
3005 }
3006
3007 /* Event on the registration socket */
3008 if (pollfd == sock) {
3009 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3010 ERR("Health socket poll error");
3011 goto error;
3012 }
3013 }
3014 }
3015
3016 new_sock = lttcomm_accept_unix_sock(sock);
3017 if (new_sock < 0) {
3018 goto error;
3019 }
3020
3021 /*
3022 * Set the CLOEXEC flag. Return code is useless because either way, the
3023 * show must go on.
3024 */
3025 (void) utils_set_fd_cloexec(new_sock);
3026
3027 DBG("Receiving data from client for health...");
3028 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3029 if (ret <= 0) {
3030 DBG("Nothing recv() from client... continuing");
3031 ret = close(new_sock);
3032 if (ret) {
3033 PERROR("close");
3034 }
3035 new_sock = -1;
3036 continue;
3037 }
3038
3039 rcu_thread_online();
3040
3041 switch (msg.component) {
3042 case LTTNG_HEALTH_CMD:
3043 reply.ret_code = health_check_state(&health_thread_cmd);
3044 break;
3045 case LTTNG_HEALTH_APP_MANAGE:
3046 reply.ret_code = health_check_state(&health_thread_app_manage);
3047 break;
3048 case LTTNG_HEALTH_APP_REG:
3049 reply.ret_code = health_check_state(&health_thread_app_reg);
3050 break;
3051 case LTTNG_HEALTH_KERNEL:
3052 reply.ret_code = health_check_state(&health_thread_kernel);
3053 break;
3054 case LTTNG_HEALTH_CONSUMER:
3055 reply.ret_code = check_consumer_health();
3056 break;
3057 case LTTNG_HEALTH_ALL:
3058 reply.ret_code =
3059 health_check_state(&health_thread_app_manage) &&
3060 health_check_state(&health_thread_app_reg) &&
3061 health_check_state(&health_thread_cmd) &&
3062 health_check_state(&health_thread_kernel) &&
3063 check_consumer_health();
3064 break;
3065 default:
3066 reply.ret_code = LTTNG_ERR_UND;
3067 break;
3068 }
3069
3070 /*
3071 * Flip ret value since 0 is a success and 1 indicates a bad health for
3072 * the client where in the sessiond it is the opposite. Again, this is
3073 * just to make things easier for us poor developer which enjoy a lot
3074 * lazyness.
3075 */
3076 if (reply.ret_code == 0 || reply.ret_code == 1) {
3077 reply.ret_code = !reply.ret_code;
3078 }
3079
3080 DBG2("Health check return value %d", reply.ret_code);
3081
3082 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3083 if (ret < 0) {
3084 ERR("Failed to send health data back to client");
3085 }
3086
3087 /* End of transmission */
3088 ret = close(new_sock);
3089 if (ret) {
3090 PERROR("close");
3091 }
3092 new_sock = -1;
3093 }
3094
3095 exit:
3096 error:
3097 if (err) {
3098 ERR("Health error occurred in %s", __func__);
3099 }
3100 DBG("Health check thread dying");
3101 unlink(health_unix_sock_path);
3102 if (sock >= 0) {
3103 ret = close(sock);
3104 if (ret) {
3105 PERROR("close");
3106 }
3107 }
3108 if (new_sock >= 0) {
3109 ret = close(new_sock);
3110 if (ret) {
3111 PERROR("close");
3112 }
3113 }
3114
3115 lttng_poll_clean(&events);
3116
3117 rcu_unregister_thread();
3118 return NULL;
3119 }
3120
3121 /*
3122 * This thread manage all clients request using the unix client socket for
3123 * communication.
3124 */
3125 static void *thread_manage_clients(void *data)
3126 {
3127 int sock = -1, ret, i, pollfd, err = -1;
3128 int sock_error;
3129 uint32_t revents, nb_fd;
3130 struct command_ctx *cmd_ctx = NULL;
3131 struct lttng_poll_event events;
3132
3133 DBG("[thread] Manage client started");
3134
3135 rcu_register_thread();
3136
3137 if (testpoint(thread_manage_clients)) {
3138 goto error_testpoint;
3139 }
3140
3141 health_code_update(&health_thread_cmd);
3142
3143 ret = lttcomm_listen_unix_sock(client_sock);
3144 if (ret < 0) {
3145 goto error_listen;
3146 }
3147
3148 /*
3149 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3150 * more will be added to this poll set.
3151 */
3152 ret = create_thread_poll_set(&events, 2);
3153 if (ret < 0) {
3154 goto error_create_poll;
3155 }
3156
3157 /* Add the application registration socket */
3158 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3159 if (ret < 0) {
3160 goto error;
3161 }
3162
3163 /*
3164 * Notify parent pid that we are ready to accept command for client side.
3165 */
3166 if (opt_sig_parent) {
3167 kill(ppid, SIGUSR1);
3168 }
3169
3170 if (testpoint(thread_manage_clients_before_loop)) {
3171 goto error;
3172 }
3173
3174 health_code_update(&health_thread_cmd);
3175
3176 while (1) {
3177 DBG("Accepting client command ...");
3178
3179 /* Inifinite blocking call, waiting for transmission */
3180 restart:
3181 health_poll_update(&health_thread_cmd);
3182 ret = lttng_poll_wait(&events, -1);
3183 health_poll_update(&health_thread_cmd);
3184 if (ret < 0) {
3185 /*
3186 * Restart interrupted system call.
3187 */
3188 if (errno == EINTR) {
3189 goto restart;
3190 }
3191 goto error;
3192 }
3193
3194 nb_fd = ret;
3195
3196 for (i = 0; i < nb_fd; i++) {
3197 /* Fetch once the poll data */
3198 revents = LTTNG_POLL_GETEV(&events, i);
3199 pollfd = LTTNG_POLL_GETFD(&events, i);
3200
3201 health_code_update(&health_thread_cmd);
3202
3203 /* Thread quit pipe has been closed. Killing thread. */
3204 ret = check_thread_quit_pipe(pollfd, revents);
3205 if (ret) {
3206 err = 0;
3207 goto exit;
3208 }
3209
3210 /* Event on the registration socket */
3211 if (pollfd == client_sock) {
3212 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3213 ERR("Client socket poll error");
3214 goto error;
3215 }
3216 }
3217 }
3218
3219 DBG("Wait for client response");
3220
3221 health_code_update(&health_thread_cmd);
3222
3223 sock = lttcomm_accept_unix_sock(client_sock);
3224 if (sock < 0) {
3225 goto error;
3226 }
3227
3228 /*
3229 * Set the CLOEXEC flag. Return code is useless because either way, the
3230 * show must go on.
3231 */
3232 (void) utils_set_fd_cloexec(sock);
3233
3234 /* Set socket option for credentials retrieval */
3235 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3236 if (ret < 0) {
3237 goto error;
3238 }
3239
3240 /* Allocate context command to process the client request */
3241 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3242 if (cmd_ctx == NULL) {
3243 PERROR("zmalloc cmd_ctx");
3244 goto error;
3245 }
3246
3247 /* Allocate data buffer for reception */
3248 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3249 if (cmd_ctx->lsm == NULL) {
3250 PERROR("zmalloc cmd_ctx->lsm");
3251 goto error;
3252 }
3253
3254 cmd_ctx->llm = NULL;
3255 cmd_ctx->session = NULL;
3256
3257 health_code_update(&health_thread_cmd);
3258
3259 /*
3260 * Data is received from the lttng client. The struct
3261 * lttcomm_session_msg (lsm) contains the command and data request of
3262 * the client.
3263 */
3264 DBG("Receiving data from client ...");
3265 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3266 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3267 if (ret <= 0) {
3268 DBG("Nothing recv() from client... continuing");
3269 ret = close(sock);
3270 if (ret) {
3271 PERROR("close");
3272 }
3273 sock = -1;
3274 clean_command_ctx(&cmd_ctx);
3275 continue;
3276 }
3277
3278 health_code_update(&health_thread_cmd);
3279
3280 // TODO: Validate cmd_ctx including sanity check for
3281 // security purpose.
3282
3283 rcu_thread_online();
3284 /*
3285 * This function dispatch the work to the kernel or userspace tracer
3286 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3287 * informations for the client. The command context struct contains
3288 * everything this function may needs.
3289 */
3290 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3291 rcu_thread_offline();
3292 if (ret < 0) {
3293 if (sock_error) {
3294 ret = close(sock);
3295 if (ret) {
3296 PERROR("close");
3297 }
3298 sock = -1;
3299 }
3300 /*
3301 * TODO: Inform client somehow of the fatal error. At
3302 * this point, ret < 0 means that a zmalloc failed
3303 * (ENOMEM). Error detected but still accept
3304 * command, unless a socket error has been
3305 * detected.
3306 */
3307 clean_command_ctx(&cmd_ctx);
3308 continue;
3309 }
3310
3311 health_code_update(&health_thread_cmd);
3312
3313 DBG("Sending response (size: %d, retcode: %s)",
3314 cmd_ctx->lttng_msg_size,
3315 lttng_strerror(-cmd_ctx->llm->ret_code));
3316 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3317 if (ret < 0) {
3318 ERR("Failed to send data back to client");
3319 }
3320
3321 /* End of transmission */
3322 ret = close(sock);
3323 if (ret) {
3324 PERROR("close");
3325 }
3326 sock = -1;
3327
3328 clean_command_ctx(&cmd_ctx);
3329
3330 health_code_update(&health_thread_cmd);
3331 }
3332
3333 exit:
3334 error:
3335 if (sock >= 0) {
3336 ret = close(sock);
3337 if (ret) {
3338 PERROR("close");
3339 }
3340 }
3341
3342 lttng_poll_clean(&events);
3343 clean_command_ctx(&cmd_ctx);
3344
3345 error_listen:
3346 error_create_poll:
3347 error_testpoint:
3348 unlink(client_unix_sock_path);
3349 if (client_sock >= 0) {
3350 ret = close(client_sock);
3351 if (ret) {
3352 PERROR("close");
3353 }
3354 }
3355
3356 if (err) {
3357 health_error(&health_thread_cmd);
3358 ERR("Health error occurred in %s", __func__);
3359 }
3360
3361 health_exit(&health_thread_cmd);
3362
3363 DBG("Client thread dying");
3364
3365 rcu_unregister_thread();
3366 return NULL;
3367 }
3368
3369
3370 /*
3371 * usage function on stderr
3372 */
3373 static void usage(void)
3374 {
3375 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3376 fprintf(stderr, " -h, --help Display this usage.\n");
3377 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3378 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3379 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3380 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3381 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3382 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3383 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3384 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3385 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3386 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3387 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3388 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3389 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3390 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3391 fprintf(stderr, " -V, --version Show version number.\n");
3392 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3393 fprintf(stderr, " -q, --quiet No output at all.\n");
3394 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3395 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3396 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3397 }
3398
3399 /*
3400 * daemon argument parsing
3401 */
3402 static int parse_args(int argc, char **argv)
3403 {
3404 int c;
3405
3406 static struct option long_options[] = {
3407 { "client-sock", 1, 0, 'c' },
3408 { "apps-sock", 1, 0, 'a' },
3409 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3410 { "kconsumerd-err-sock", 1, 0, 'E' },
3411 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3412 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3413 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3414 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3415 { "consumerd32-path", 1, 0, 'u' },
3416 { "consumerd32-libdir", 1, 0, 'U' },
3417 { "consumerd64-path", 1, 0, 't' },
3418 { "consumerd64-libdir", 1, 0, 'T' },
3419 { "daemonize", 0, 0, 'd' },
3420 { "sig-parent", 0, 0, 'S' },
3421 { "help", 0, 0, 'h' },
3422 { "group", 1, 0, 'g' },
3423 { "version", 0, 0, 'V' },
3424 { "quiet", 0, 0, 'q' },
3425 { "verbose", 0, 0, 'v' },
3426 { "verbose-consumer", 0, 0, 'Z' },
3427 { "no-kernel", 0, 0, 'N' },
3428 { NULL, 0, 0, 0 }
3429 };
3430
3431 while (1) {
3432 int option_index = 0;
3433 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3434 long_options, &option_index);
3435 if (c == -1) {
3436 break;
3437 }
3438
3439 switch (c) {
3440 case 0:
3441 fprintf(stderr, "option %s", long_options[option_index].name);
3442 if (optarg) {
3443 fprintf(stderr, " with arg %s\n", optarg);
3444 }
3445 break;
3446 case 'c':
3447 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3448 break;
3449 case 'a':
3450 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3451 break;
3452 case 'd':
3453 opt_daemon = 1;
3454 break;
3455 case 'g':
3456 opt_tracing_group = optarg;
3457 break;
3458 case 'h':
3459 usage();
3460 exit(EXIT_FAILURE);
3461 case 'V':
3462 fprintf(stdout, "%s\n", VERSION);
3463 exit(EXIT_SUCCESS);
3464 case 'S':
3465 opt_sig_parent = 1;
3466 break;
3467 case 'E':
3468 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3469 break;
3470 case 'C':
3471 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3472 break;
3473 case 'F':
3474 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3475 break;
3476 case 'D':
3477 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3478 break;
3479 case 'H':
3480 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3481 break;
3482 case 'G':
3483 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3484 break;
3485 case 'N':
3486 opt_no_kernel = 1;
3487 break;
3488 case 'q':
3489 lttng_opt_quiet = 1;
3490 break;
3491 case 'v':
3492 /* Verbose level can increase using multiple -v */
3493 lttng_opt_verbose += 1;
3494 break;
3495 case 'Z':
3496 opt_verbose_consumer += 1;
3497 break;
3498 case 'u':
3499 consumerd32_bin= optarg;
3500 break;
3501 case 'U':
3502 consumerd32_libdir = optarg;
3503 break;
3504 case 't':
3505 consumerd64_bin = optarg;
3506 break;
3507 case 'T':
3508 consumerd64_libdir = optarg;
3509 break;
3510 default:
3511 /* Unknown option or other error.
3512 * Error is printed by getopt, just return */
3513 return -1;
3514 }
3515 }
3516
3517 return 0;
3518 }
3519
3520 /*
3521 * Creates the two needed socket by the daemon.
3522 * apps_sock - The communication socket for all UST apps.
3523 * client_sock - The communication of the cli tool (lttng).
3524 */
3525 static int init_daemon_socket(void)
3526 {
3527 int ret = 0;
3528 mode_t old_umask;
3529
3530 old_umask = umask(0);
3531
3532 /* Create client tool unix socket */
3533 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3534 if (client_sock < 0) {
3535 ERR("Create unix sock failed: %s", client_unix_sock_path);
3536 ret = -1;
3537 goto end;
3538 }
3539
3540 /* Set the cloexec flag */
3541 ret = utils_set_fd_cloexec(client_sock);
3542 if (ret < 0) {
3543 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3544 "Continuing but note that the consumer daemon will have a "
3545 "reference to this socket on exec()", client_sock);
3546 }
3547
3548 /* File permission MUST be 660 */
3549 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3550 if (ret < 0) {
3551 ERR("Set file permissions failed: %s", client_unix_sock_path);
3552 PERROR("chmod");
3553 goto end;
3554 }
3555
3556 /* Create the application unix socket */
3557 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3558 if (apps_sock < 0) {
3559 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3560 ret = -1;
3561 goto end;
3562 }
3563
3564 /* Set the cloexec flag */
3565 ret = utils_set_fd_cloexec(apps_sock);
3566 if (ret < 0) {
3567 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3568 "Continuing but note that the consumer daemon will have a "
3569 "reference to this socket on exec()", apps_sock);
3570 }
3571
3572 /* File permission MUST be 666 */
3573 ret = chmod(apps_unix_sock_path,
3574 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3575 if (ret < 0) {
3576 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3577 PERROR("chmod");
3578 goto end;
3579 }
3580
3581 DBG3("Session daemon client socket %d and application socket %d created",
3582 client_sock, apps_sock);
3583
3584 end:
3585 umask(old_umask);
3586 return ret;
3587 }
3588
3589 /*
3590 * Check if the global socket is available, and if a daemon is answering at the
3591 * other side. If yes, error is returned.
3592 */
3593 static int check_existing_daemon(void)
3594 {
3595 /* Is there anybody out there ? */
3596 if (lttng_session_daemon_alive()) {
3597 return -EEXIST;
3598 }
3599
3600 return 0;
3601 }
3602
3603 /*
3604 * Set the tracing group gid onto the client socket.
3605 *
3606 * Race window between mkdir and chown is OK because we are going from more
3607 * permissive (root.root) to less permissive (root.tracing).
3608 */
3609 static int set_permissions(char *rundir)
3610 {
3611 int ret;
3612 gid_t gid;
3613
3614 ret = allowed_group();
3615 if (ret < 0) {
3616 WARN("No tracing group detected");
3617 ret = 0;
3618 goto end;
3619 }
3620
3621 gid = ret;
3622
3623 /* Set lttng run dir */
3624 ret = chown(rundir, 0, gid);
3625 if (ret < 0) {
3626 ERR("Unable to set group on %s", rundir);
3627 PERROR("chown");
3628 }
3629
3630 /* Ensure tracing group can search the run dir */
3631 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3632 if (ret < 0) {
3633 ERR("Unable to set permissions on %s", rundir);
3634 PERROR("chmod");
3635 }
3636
3637 /* lttng client socket path */
3638 ret = chown(client_unix_sock_path, 0, gid);
3639 if (ret < 0) {
3640 ERR("Unable to set group on %s", client_unix_sock_path);
3641 PERROR("chown");
3642 }
3643
3644 /* kconsumer error socket path */
3645 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3646 if (ret < 0) {
3647 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3648 PERROR("chown");
3649 }
3650
3651 /* 64-bit ustconsumer error socket path */
3652 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3653 if (ret < 0) {
3654 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3655 PERROR("chown");
3656 }
3657
3658 /* 32-bit ustconsumer compat32 error socket path */
3659 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3660 if (ret < 0) {
3661 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3662 PERROR("chown");
3663 }
3664
3665 DBG("All permissions are set");
3666
3667 end:
3668 return ret;
3669 }
3670
3671 /*
3672 * Create the lttng run directory needed for all global sockets and pipe.
3673 */
3674 static int create_lttng_rundir(const char *rundir)
3675 {
3676 int ret;
3677
3678 DBG3("Creating LTTng run directory: %s", rundir);
3679
3680 ret = mkdir(rundir, S_IRWXU);
3681 if (ret < 0) {
3682 if (errno != EEXIST) {
3683 ERR("Unable to create %s", rundir);
3684 goto error;
3685 } else {
3686 ret = 0;
3687 }
3688 }
3689
3690 error:
3691 return ret;
3692 }
3693
3694 /*
3695 * Setup sockets and directory needed by the kconsumerd communication with the
3696 * session daemon.
3697 */
3698 static int set_consumer_sockets(struct consumer_data *consumer_data,
3699 const char *rundir)
3700 {
3701 int ret;
3702 char path[PATH_MAX];
3703
3704 switch (consumer_data->type) {
3705 case LTTNG_CONSUMER_KERNEL:
3706 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3707 break;
3708 case LTTNG_CONSUMER64_UST:
3709 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3710 break;
3711 case LTTNG_CONSUMER32_UST:
3712 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3713 break;
3714 default:
3715 ERR("Consumer type unknown");
3716 ret = -EINVAL;
3717 goto error;
3718 }
3719
3720 DBG2("Creating consumer directory: %s", path);
3721
3722 ret = mkdir(path, S_IRWXU);
3723 if (ret < 0) {
3724 if (errno != EEXIST) {
3725 PERROR("mkdir");
3726 ERR("Failed to create %s", path);
3727 goto error;
3728 }
3729 ret = -1;
3730 }
3731
3732 /* Create the kconsumerd error unix socket */
3733 consumer_data->err_sock =
3734 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3735 if (consumer_data->err_sock < 0) {
3736 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3737 ret = -1;
3738 goto error;
3739 }
3740
3741 /* File permission MUST be 660 */
3742 ret = chmod(consumer_data->err_unix_sock_path,
3743 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3744 if (ret < 0) {
3745 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3746 PERROR("chmod");
3747 goto error;
3748 }
3749
3750 error:
3751 return ret;
3752 }
3753
3754 /*
3755 * Signal handler for the daemon
3756 *
3757 * Simply stop all worker threads, leaving main() return gracefully after
3758 * joining all threads and calling cleanup().
3759 */
3760 static void sighandler(int sig)
3761 {
3762 switch (sig) {
3763 case SIGPIPE:
3764 DBG("SIGPIPE caught");
3765 return;
3766 case SIGINT:
3767 DBG("SIGINT caught");
3768 stop_threads();
3769 break;
3770 case SIGTERM:
3771 DBG("SIGTERM caught");
3772 stop_threads();
3773 break;
3774 default:
3775 break;
3776 }
3777 }
3778
3779 /*
3780 * Setup signal handler for :
3781 * SIGINT, SIGTERM, SIGPIPE
3782 */
3783 static int set_signal_handler(void)
3784 {
3785 int ret = 0;
3786 struct sigaction sa;
3787 sigset_t sigset;
3788
3789 if ((ret = sigemptyset(&sigset)) < 0) {
3790 PERROR("sigemptyset");
3791 return ret;
3792 }
3793
3794 sa.sa_handler = sighandler;
3795 sa.sa_mask = sigset;
3796 sa.sa_flags = 0;
3797 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3798 PERROR("sigaction");
3799 return ret;
3800 }
3801
3802 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3803 PERROR("sigaction");
3804 return ret;
3805 }
3806
3807 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3808 PERROR("sigaction");
3809 return ret;
3810 }
3811
3812 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3813
3814 return ret;
3815 }
3816
3817 /*
3818 * Set open files limit to unlimited. This daemon can open a large number of
3819 * file descriptors in order to consumer multiple kernel traces.
3820 */
3821 static void set_ulimit(void)
3822 {
3823 int ret;
3824 struct rlimit lim;
3825
3826 /* The kernel does not allowed an infinite limit for open files */
3827 lim.rlim_cur = 65535;
3828 lim.rlim_max = 65535;
3829
3830 ret = setrlimit(RLIMIT_NOFILE, &lim);
3831 if (ret < 0) {
3832 PERROR("failed to set open files limit");
3833 }
3834 }
3835
3836 /*
3837 * main
3838 */
3839 int main(int argc, char **argv)
3840 {
3841 int ret = 0;
3842 void *status;
3843 const char *home_path, *env_app_timeout;
3844
3845 init_kernel_workarounds();
3846
3847 rcu_register_thread();
3848
3849 setup_consumerd_path();
3850
3851 /* Parse arguments */
3852 progname = argv[0];
3853 if ((ret = parse_args(argc, argv)) < 0) {
3854 goto error;
3855 }
3856
3857 /* Daemonize */
3858 if (opt_daemon) {
3859 int i;
3860
3861 /*
3862 * fork
3863 * child: setsid, close FD 0, 1, 2, chdir /
3864 * parent: exit (if fork is successful)
3865 */
3866 ret = daemon(0, 0);
3867 if (ret < 0) {
3868 PERROR("daemon");
3869 goto error;
3870 }
3871 /*
3872 * We are in the child. Make sure all other file
3873 * descriptors are closed, in case we are called with
3874 * more opened file descriptors than the standard ones.
3875 */
3876 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3877 (void) close(i);
3878 }
3879 }
3880
3881 /* Create thread quit pipe */
3882 if ((ret = init_thread_quit_pipe()) < 0) {
3883 goto error;
3884 }
3885
3886 /* Check if daemon is UID = 0 */
3887 is_root = !getuid();
3888
3889 if (is_root) {
3890 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3891
3892 /* Create global run dir with root access */
3893 ret = create_lttng_rundir(rundir);
3894 if (ret < 0) {
3895 goto error;
3896 }
3897
3898 if (strlen(apps_unix_sock_path) == 0) {
3899 snprintf(apps_unix_sock_path, PATH_MAX,
3900 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3901 }
3902
3903 if (strlen(client_unix_sock_path) == 0) {
3904 snprintf(client_unix_sock_path, PATH_MAX,
3905 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3906 }
3907
3908 /* Set global SHM for ust */
3909 if (strlen(wait_shm_path) == 0) {
3910 snprintf(wait_shm_path, PATH_MAX,
3911 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3912 }
3913
3914 if (strlen(health_unix_sock_path) == 0) {
3915 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3916 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3917 }
3918
3919 /* Setup kernel consumerd path */
3920 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3921 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3922 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3923 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3924
3925 DBG2("Kernel consumer err path: %s",
3926 kconsumer_data.err_unix_sock_path);
3927 DBG2("Kernel consumer cmd path: %s",
3928 kconsumer_data.cmd_unix_sock_path);
3929 } else {
3930 home_path = get_home_dir();
3931 if (home_path == NULL) {
3932 /* TODO: Add --socket PATH option */
3933 ERR("Can't get HOME directory for sockets creation.");
3934 ret = -EPERM;
3935 goto error;
3936 }
3937
3938 /*
3939 * Create rundir from home path. This will create something like
3940 * $HOME/.lttng
3941 */
3942 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3943 if (ret < 0) {
3944 ret = -ENOMEM;
3945 goto error;
3946 }
3947
3948 ret = create_lttng_rundir(rundir);
3949 if (ret < 0) {
3950 goto error;
3951 }
3952
3953 if (strlen(apps_unix_sock_path) == 0) {
3954 snprintf(apps_unix_sock_path, PATH_MAX,
3955 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3956 }
3957
3958 /* Set the cli tool unix socket path */
3959 if (strlen(client_unix_sock_path) == 0) {
3960 snprintf(client_unix_sock_path, PATH_MAX,
3961 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3962 }
3963
3964 /* Set global SHM for ust */
3965 if (strlen(wait_shm_path) == 0) {
3966 snprintf(wait_shm_path, PATH_MAX,
3967 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3968 }
3969
3970 /* Set health check Unix path */
3971 if (strlen(health_unix_sock_path) == 0) {
3972 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3973 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3974 }
3975 }
3976
3977 /* Set consumer initial state */
3978 kernel_consumerd_state = CONSUMER_STOPPED;
3979 ust_consumerd_state = CONSUMER_STOPPED;
3980
3981 DBG("Client socket path %s", client_unix_sock_path);
3982 DBG("Application socket path %s", apps_unix_sock_path);
3983 DBG("LTTng run directory path: %s", rundir);
3984
3985 /* 32 bits consumerd path setup */
3986 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3987 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3988 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3989 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3990
3991 DBG2("UST consumer 32 bits err path: %s",
3992 ustconsumer32_data.err_unix_sock_path);
3993 DBG2("UST consumer 32 bits cmd path: %s",
3994 ustconsumer32_data.cmd_unix_sock_path);
3995
3996 /* 64 bits consumerd path setup */
3997 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
3998 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
3999 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4000 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4001
4002 DBG2("UST consumer 64 bits err path: %s",
4003 ustconsumer64_data.err_unix_sock_path);
4004 DBG2("UST consumer 64 bits cmd path: %s",
4005 ustconsumer64_data.cmd_unix_sock_path);
4006
4007 /*
4008 * See if daemon already exist.
4009 */
4010 if ((ret = check_existing_daemon()) < 0) {
4011 ERR("Already running daemon.\n");
4012 /*
4013 * We do not goto exit because we must not cleanup()
4014 * because a daemon is already running.
4015 */
4016 goto error;
4017 }
4018
4019 /*
4020 * Init UST app hash table. Alloc hash table before this point since
4021 * cleanup() can get called after that point.
4022 */
4023 ust_app_ht_alloc();
4024
4025 /* After this point, we can safely call cleanup() with "goto exit" */
4026
4027 /*
4028 * These actions must be executed as root. We do that *after* setting up
4029 * the sockets path because we MUST make the check for another daemon using
4030 * those paths *before* trying to set the kernel consumer sockets and init
4031 * kernel tracer.
4032 */
4033 if (is_root) {
4034 ret = set_consumer_sockets(&kconsumer_data, rundir);
4035 if (ret < 0) {
4036 goto exit;
4037 }
4038
4039 /* Setup kernel tracer */
4040 if (!opt_no_kernel) {
4041 init_kernel_tracer();
4042 }
4043
4044 /* Set ulimit for open files */
4045 set_ulimit();
4046 }
4047 /* init lttng_fd tracking must be done after set_ulimit. */
4048 lttng_fd_init();
4049
4050 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4051 if (ret < 0) {
4052 goto exit;
4053 }
4054
4055 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4056 if (ret < 0) {
4057 goto exit;
4058 }
4059
4060 if ((ret = set_signal_handler()) < 0) {
4061 goto exit;
4062 }
4063
4064 /* Setup the needed unix socket */
4065 if ((ret = init_daemon_socket()) < 0) {
4066 goto exit;
4067 }
4068
4069 /* Set credentials to socket */
4070 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4071 goto exit;
4072 }
4073
4074 /* Get parent pid if -S, --sig-parent is specified. */
4075 if (opt_sig_parent) {
4076 ppid = getppid();
4077 }
4078
4079 /* Setup the kernel pipe for waking up the kernel thread */
4080 if (is_root && !opt_no_kernel) {
4081 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4082 goto exit;
4083 }
4084 }
4085
4086 /* Setup the thread apps communication pipe. */
4087 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4088 goto exit;
4089 }
4090
4091 /* Init UST command queue. */
4092 cds_wfq_init(&ust_cmd_queue.queue);
4093
4094 /*
4095 * Get session list pointer. This pointer MUST NOT be free(). This list is
4096 * statically declared in session.c
4097 */
4098 session_list_ptr = session_get_list();
4099
4100 /* Set up max poll set size */
4101 lttng_poll_set_max_size();
4102
4103 cmd_init();
4104
4105 /* Init all health thread counters. */
4106 health_init(&health_thread_cmd);
4107 health_init(&health_thread_kernel);
4108 health_init(&health_thread_app_manage);
4109 health_init(&health_thread_app_reg);
4110
4111 /*
4112 * Init health counters of the consumer thread. We do a quick hack here to
4113 * the state of the consumer health is fine even if the thread is not
4114 * started. Once the thread starts, the health state is updated with a poll
4115 * value to set a health code path. This is simply to ease our life and has
4116 * no cost what so ever.
4117 */
4118 health_init(&kconsumer_data.health);
4119 health_poll_update(&kconsumer_data.health);
4120 health_init(&ustconsumer32_data.health);
4121 health_poll_update(&ustconsumer32_data.health);
4122 health_init(&ustconsumer64_data.health);
4123 health_poll_update(&ustconsumer64_data.health);
4124
4125 /* Check for the application socket timeout env variable. */
4126 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4127 if (env_app_timeout) {
4128 app_socket_timeout = atoi(env_app_timeout);
4129 } else {
4130 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4131 }
4132
4133 /* Create thread to manage the client socket */
4134 ret = pthread_create(&health_thread, NULL,
4135 thread_manage_health, (void *) NULL);
4136 if (ret != 0) {
4137 PERROR("pthread_create health");
4138 goto exit_health;
4139 }
4140
4141 /* Create thread to manage the client socket */
4142 ret = pthread_create(&client_thread, NULL,
4143 thread_manage_clients, (void *) NULL);
4144 if (ret != 0) {
4145 PERROR("pthread_create clients");
4146 goto exit_client;
4147 }
4148
4149 /* Create thread to dispatch registration */
4150 ret = pthread_create(&dispatch_thread, NULL,
4151 thread_dispatch_ust_registration, (void *) NULL);
4152 if (ret != 0) {
4153 PERROR("pthread_create dispatch");
4154 goto exit_dispatch;
4155 }
4156
4157 /* Create thread to manage application registration. */
4158 ret = pthread_create(&reg_apps_thread, NULL,
4159 thread_registration_apps, (void *) NULL);
4160 if (ret != 0) {
4161 PERROR("pthread_create registration");
4162 goto exit_reg_apps;
4163 }
4164
4165 /* Create thread to manage application socket */
4166 ret = pthread_create(&apps_thread, NULL,
4167 thread_manage_apps, (void *) NULL);
4168 if (ret != 0) {
4169 PERROR("pthread_create apps");
4170 goto exit_apps;
4171 }
4172
4173 /* Don't start this thread if kernel tracing is not requested nor root */
4174 if (is_root && !opt_no_kernel) {
4175 /* Create kernel thread to manage kernel event */
4176 ret = pthread_create(&kernel_thread, NULL,
4177 thread_manage_kernel, (void *) NULL);
4178 if (ret != 0) {
4179 PERROR("pthread_create kernel");
4180 goto exit_kernel;
4181 }
4182
4183 ret = pthread_join(kernel_thread, &status);
4184 if (ret != 0) {
4185 PERROR("pthread_join");
4186 goto error; /* join error, exit without cleanup */
4187 }
4188 }
4189
4190 exit_kernel:
4191 ret = pthread_join(apps_thread, &status);
4192 if (ret != 0) {
4193 PERROR("pthread_join");
4194 goto error; /* join error, exit without cleanup */
4195 }
4196
4197 exit_apps:
4198 ret = pthread_join(reg_apps_thread, &status);
4199 if (ret != 0) {
4200 PERROR("pthread_join");
4201 goto error; /* join error, exit without cleanup */
4202 }
4203
4204 exit_reg_apps:
4205 ret = pthread_join(dispatch_thread, &status);
4206 if (ret != 0) {
4207 PERROR("pthread_join");
4208 goto error; /* join error, exit without cleanup */
4209 }
4210
4211 exit_dispatch:
4212 ret = pthread_join(client_thread, &status);
4213 if (ret != 0) {
4214 PERROR("pthread_join");
4215 goto error; /* join error, exit without cleanup */
4216 }
4217
4218 ret = join_consumer_thread(&kconsumer_data);
4219 if (ret != 0) {
4220 PERROR("join_consumer");
4221 goto error; /* join error, exit without cleanup */
4222 }
4223
4224 ret = join_consumer_thread(&ustconsumer32_data);
4225 if (ret != 0) {
4226 PERROR("join_consumer ust32");
4227 goto error; /* join error, exit without cleanup */
4228 }
4229
4230 ret = join_consumer_thread(&ustconsumer64_data);
4231 if (ret != 0) {
4232 PERROR("join_consumer ust64");
4233 goto error; /* join error, exit without cleanup */
4234 }
4235
4236 exit_client:
4237 ret = pthread_join(health_thread, &status);
4238 if (ret != 0) {
4239 PERROR("pthread_join health thread");
4240 goto error; /* join error, exit without cleanup */
4241 }
4242
4243 exit_health:
4244 exit:
4245 /*
4246 * cleanup() is called when no other thread is running.
4247 */
4248 rcu_thread_online();
4249 cleanup();
4250 rcu_thread_offline();
4251 rcu_unregister_thread();
4252 if (!ret) {
4253 exit(EXIT_SUCCESS);
4254 }
4255 error:
4256 exit(EXIT_FAILURE);
4257 }
This page took 0.157295 seconds and 3 git commands to generate.