Add return code to the testpoint mechanism
[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 nb_fd = LTTNG_POLL_GETNB(&events);
732
733 DBG("Thread kernel polling on %d fds", nb_fd);
734
735 /* Zeroed the poll events */
736 lttng_poll_reset(&events);
737
738 /* Poll infinite value of time */
739 restart:
740 health_poll_update(&health_thread_kernel);
741 ret = lttng_poll_wait(&events, -1);
742 health_poll_update(&health_thread_kernel);
743 if (ret < 0) {
744 /*
745 * Restart interrupted system call.
746 */
747 if (errno == EINTR) {
748 goto restart;
749 }
750 goto error;
751 } else if (ret == 0) {
752 /* Should not happen since timeout is infinite */
753 ERR("Return value of poll is 0 with an infinite timeout.\n"
754 "This should not have happened! Continuing...");
755 continue;
756 }
757
758 for (i = 0; i < nb_fd; i++) {
759 /* Fetch once the poll data */
760 revents = LTTNG_POLL_GETEV(&events, i);
761 pollfd = LTTNG_POLL_GETFD(&events, i);
762
763 health_code_update(&health_thread_kernel);
764
765 /* Thread quit pipe has been closed. Killing thread. */
766 ret = check_thread_quit_pipe(pollfd, revents);
767 if (ret) {
768 err = 0;
769 goto exit;
770 }
771
772 /* Check for data on kernel pipe */
773 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
774 ret = read(kernel_poll_pipe[0], &tmp, 1);
775 update_poll_flag = 1;
776 continue;
777 } else {
778 /*
779 * New CPU detected by the kernel. Adding kernel stream to
780 * kernel session and updating the kernel consumer
781 */
782 if (revents & LPOLLIN) {
783 ret = update_kernel_stream(&kconsumer_data, pollfd);
784 if (ret < 0) {
785 continue;
786 }
787 break;
788 /*
789 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
790 * and unregister kernel stream at this point.
791 */
792 }
793 }
794 }
795 }
796
797 exit:
798 error:
799 lttng_poll_clean(&events);
800 error_poll_create:
801 error_testpoint:
802 utils_close_pipe(kernel_poll_pipe);
803 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
804 if (err) {
805 health_error(&health_thread_kernel);
806 ERR("Health error occurred in %s", __func__);
807 WARN("Kernel thread died unexpectedly. "
808 "Kernel tracing can continue but CPU hotplug is disabled.");
809 }
810 health_exit(&health_thread_kernel);
811 DBG("Kernel thread dying");
812 return NULL;
813 }
814
815 /*
816 * Signal pthread condition of the consumer data that the thread.
817 */
818 static void signal_consumer_condition(struct consumer_data *data, int state)
819 {
820 pthread_mutex_lock(&data->cond_mutex);
821
822 /*
823 * The state is set before signaling. It can be any value, it's the waiter
824 * job to correctly interpret this condition variable associated to the
825 * consumer pthread_cond.
826 *
827 * A value of 0 means that the corresponding thread of the consumer data
828 * was not started. 1 indicates that the thread has started and is ready
829 * for action. A negative value means that there was an error during the
830 * thread bootstrap.
831 */
832 data->consumer_thread_is_ready = state;
833 (void) pthread_cond_signal(&data->cond);
834
835 pthread_mutex_unlock(&data->cond_mutex);
836 }
837
838 /*
839 * This thread manage the consumer error sent back to the session daemon.
840 */
841 static void *thread_manage_consumer(void *data)
842 {
843 int sock = -1, i, ret, pollfd, err = -1;
844 uint32_t revents, nb_fd;
845 enum lttcomm_return_code code;
846 struct lttng_poll_event events;
847 struct consumer_data *consumer_data = data;
848
849 DBG("[thread] Manage consumer started");
850
851 /*
852 * Since the consumer thread can be spawned at any moment in time, we init
853 * the health to a poll status (1, which is a valid health over time).
854 * When the thread starts, we update here the health to a "code" path being
855 * an even value so this thread, when reaching a poll wait, does not
856 * trigger an error with an even value.
857 *
858 * Here is the use case we avoid.
859 *
860 * +1: the first poll update during initialization (main())
861 * +2 * x: multiple code update once in this thread.
862 * +1: poll wait in this thread (being a good health state).
863 * == even number which after the wait period shows as a bad health.
864 *
865 * In a nutshell, the following poll update to the health state brings back
866 * the state to an even value meaning a code path.
867 */
868 health_poll_update(&consumer_data->health);
869
870 /*
871 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
872 * Nothing more will be added to this poll set.
873 */
874 ret = create_thread_poll_set(&events, 2);
875 if (ret < 0) {
876 goto error_poll;
877 }
878
879 /*
880 * The error socket here is already in a listening state which was done
881 * just before spawning this thread to avoid a race between the consumer
882 * daemon exec trying to connect and the listen() call.
883 */
884 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
885 if (ret < 0) {
886 goto error;
887 }
888
889 nb_fd = LTTNG_POLL_GETNB(&events);
890
891 health_code_update(&consumer_data->health);
892
893 /* Inifinite blocking call, waiting for transmission */
894 restart:
895 health_poll_update(&consumer_data->health);
896
897 if (testpoint(thread_manage_consumer)) {
898 goto error;
899 }
900
901 ret = lttng_poll_wait(&events, -1);
902 health_poll_update(&consumer_data->health);
903 if (ret < 0) {
904 /*
905 * Restart interrupted system call.
906 */
907 if (errno == EINTR) {
908 goto restart;
909 }
910 goto error;
911 }
912
913 for (i = 0; i < nb_fd; i++) {
914 /* Fetch once the poll data */
915 revents = LTTNG_POLL_GETEV(&events, i);
916 pollfd = LTTNG_POLL_GETFD(&events, i);
917
918 health_code_update(&consumer_data->health);
919
920 /* Thread quit pipe has been closed. Killing thread. */
921 ret = check_thread_quit_pipe(pollfd, revents);
922 if (ret) {
923 err = 0;
924 goto exit;
925 }
926
927 /* Event on the registration socket */
928 if (pollfd == consumer_data->err_sock) {
929 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
930 ERR("consumer err socket poll error");
931 goto error;
932 }
933 }
934 }
935
936 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
937 if (sock < 0) {
938 goto error;
939 }
940
941 /*
942 * Set the CLOEXEC flag. Return code is useless because either way, the
943 * show must go on.
944 */
945 (void) utils_set_fd_cloexec(sock);
946
947 health_code_update(&consumer_data->health);
948
949 DBG2("Receiving code from consumer err_sock");
950
951 /* Getting status code from kconsumerd */
952 ret = lttcomm_recv_unix_sock(sock, &code,
953 sizeof(enum lttcomm_return_code));
954 if (ret <= 0) {
955 goto error;
956 }
957
958 health_code_update(&consumer_data->health);
959
960 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
961 consumer_data->cmd_sock =
962 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
963 if (consumer_data->cmd_sock < 0) {
964 /* On error, signal condition and quit. */
965 signal_consumer_condition(consumer_data, -1);
966 PERROR("consumer connect");
967 goto error;
968 }
969 signal_consumer_condition(consumer_data, 1);
970 DBG("Consumer command socket ready");
971 } else {
972 ERR("consumer error when waiting for SOCK_READY : %s",
973 lttcomm_get_readable_code(-code));
974 goto error;
975 }
976
977 /* Remove the kconsumerd error sock since we've established a connexion */
978 ret = lttng_poll_del(&events, consumer_data->err_sock);
979 if (ret < 0) {
980 goto error;
981 }
982
983 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
984 if (ret < 0) {
985 goto error;
986 }
987
988 health_code_update(&consumer_data->health);
989
990 /* Update number of fd */
991 nb_fd = LTTNG_POLL_GETNB(&events);
992
993 /* Inifinite blocking call, waiting for transmission */
994 restart_poll:
995 health_poll_update(&consumer_data->health);
996 ret = lttng_poll_wait(&events, -1);
997 health_poll_update(&consumer_data->health);
998 if (ret < 0) {
999 /*
1000 * Restart interrupted system call.
1001 */
1002 if (errno == EINTR) {
1003 goto restart_poll;
1004 }
1005 goto error;
1006 }
1007
1008 for (i = 0; i < nb_fd; i++) {
1009 /* Fetch once the poll data */
1010 revents = LTTNG_POLL_GETEV(&events, i);
1011 pollfd = LTTNG_POLL_GETFD(&events, i);
1012
1013 health_code_update(&consumer_data->health);
1014
1015 /* Thread quit pipe has been closed. Killing thread. */
1016 ret = check_thread_quit_pipe(pollfd, revents);
1017 if (ret) {
1018 err = 0;
1019 goto exit;
1020 }
1021
1022 /* Event on the kconsumerd socket */
1023 if (pollfd == sock) {
1024 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1025 ERR("consumer err socket second poll error");
1026 goto error;
1027 }
1028 }
1029 }
1030
1031 health_code_update(&consumer_data->health);
1032
1033 /* Wait for any kconsumerd error */
1034 ret = lttcomm_recv_unix_sock(sock, &code,
1035 sizeof(enum lttcomm_return_code));
1036 if (ret <= 0) {
1037 ERR("consumer closed the command socket");
1038 goto error;
1039 }
1040
1041 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1042
1043 exit:
1044 error:
1045 /* Immediately set the consumerd state to stopped */
1046 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1047 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1048 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1049 consumer_data->type == LTTNG_CONSUMER32_UST) {
1050 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1051 } else {
1052 /* Code flow error... */
1053 assert(0);
1054 }
1055
1056 if (consumer_data->err_sock >= 0) {
1057 ret = close(consumer_data->err_sock);
1058 if (ret) {
1059 PERROR("close");
1060 }
1061 }
1062 if (consumer_data->cmd_sock >= 0) {
1063 ret = close(consumer_data->cmd_sock);
1064 if (ret) {
1065 PERROR("close");
1066 }
1067 }
1068 if (sock >= 0) {
1069 ret = close(sock);
1070 if (ret) {
1071 PERROR("close");
1072 }
1073 }
1074
1075 unlink(consumer_data->err_unix_sock_path);
1076 unlink(consumer_data->cmd_unix_sock_path);
1077 consumer_data->pid = 0;
1078
1079 lttng_poll_clean(&events);
1080 error_poll:
1081 if (err) {
1082 health_error(&consumer_data->health);
1083 ERR("Health error occurred in %s", __func__);
1084 }
1085 health_exit(&consumer_data->health);
1086 DBG("consumer thread cleanup completed");
1087
1088 return NULL;
1089 }
1090
1091 /*
1092 * This thread manage application communication.
1093 */
1094 static void *thread_manage_apps(void *data)
1095 {
1096 int i, ret, pollfd, err = -1;
1097 uint32_t revents, nb_fd;
1098 struct ust_command ust_cmd;
1099 struct lttng_poll_event events;
1100
1101 DBG("[thread] Manage application started");
1102
1103 rcu_register_thread();
1104 rcu_thread_online();
1105
1106 if (testpoint(thread_manage_apps)) {
1107 goto error_testpoint;
1108 }
1109
1110 health_code_update(&health_thread_app_manage);
1111
1112 ret = create_thread_poll_set(&events, 2);
1113 if (ret < 0) {
1114 goto error_poll_create;
1115 }
1116
1117 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1118 if (ret < 0) {
1119 goto error;
1120 }
1121
1122 if (testpoint(thread_manage_apps_before_loop)) {
1123 goto error;
1124 }
1125
1126 health_code_update(&health_thread_app_manage);
1127
1128 while (1) {
1129 /* Zeroed the events structure */
1130 lttng_poll_reset(&events);
1131
1132 nb_fd = LTTNG_POLL_GETNB(&events);
1133
1134 DBG("Apps thread polling on %d fds", nb_fd);
1135
1136 /* Inifinite blocking call, waiting for transmission */
1137 restart:
1138 health_poll_update(&health_thread_app_manage);
1139 ret = lttng_poll_wait(&events, -1);
1140 health_poll_update(&health_thread_app_manage);
1141 if (ret < 0) {
1142 /*
1143 * Restart interrupted system call.
1144 */
1145 if (errno == EINTR) {
1146 goto restart;
1147 }
1148 goto error;
1149 }
1150
1151 for (i = 0; i < nb_fd; i++) {
1152 /* Fetch once the poll data */
1153 revents = LTTNG_POLL_GETEV(&events, i);
1154 pollfd = LTTNG_POLL_GETFD(&events, i);
1155
1156 health_code_update(&health_thread_app_manage);
1157
1158 /* Thread quit pipe has been closed. Killing thread. */
1159 ret = check_thread_quit_pipe(pollfd, revents);
1160 if (ret) {
1161 err = 0;
1162 goto exit;
1163 }
1164
1165 /* Inspect the apps cmd pipe */
1166 if (pollfd == apps_cmd_pipe[0]) {
1167 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1168 ERR("Apps command pipe error");
1169 goto error;
1170 } else if (revents & LPOLLIN) {
1171 /* Empty pipe */
1172 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1173 if (ret < 0 || ret < sizeof(ust_cmd)) {
1174 PERROR("read apps cmd pipe");
1175 goto error;
1176 }
1177
1178 health_code_update(&health_thread_app_manage);
1179
1180 /* Register applicaton to the session daemon */
1181 ret = ust_app_register(&ust_cmd.reg_msg,
1182 ust_cmd.sock);
1183 if (ret == -ENOMEM) {
1184 goto error;
1185 } else if (ret < 0) {
1186 break;
1187 }
1188
1189 health_code_update(&health_thread_app_manage);
1190
1191 /*
1192 * Validate UST version compatibility.
1193 */
1194 ret = ust_app_validate_version(ust_cmd.sock);
1195 if (ret >= 0) {
1196 /*
1197 * Add channel(s) and event(s) to newly registered apps
1198 * from lttng global UST domain.
1199 */
1200 update_ust_app(ust_cmd.sock);
1201 }
1202
1203 health_code_update(&health_thread_app_manage);
1204
1205 ret = ust_app_register_done(ust_cmd.sock);
1206 if (ret < 0) {
1207 /*
1208 * If the registration is not possible, we simply
1209 * unregister the apps and continue
1210 */
1211 ust_app_unregister(ust_cmd.sock);
1212 } else {
1213 /*
1214 * We only monitor the error events of the socket. This
1215 * thread does not handle any incoming data from UST
1216 * (POLLIN).
1217 */
1218 ret = lttng_poll_add(&events, ust_cmd.sock,
1219 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1220 if (ret < 0) {
1221 goto error;
1222 }
1223
1224 /* Set socket timeout for both receiving and ending */
1225 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1226 app_socket_timeout);
1227 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1228 app_socket_timeout);
1229
1230 DBG("Apps with sock %d added to poll set",
1231 ust_cmd.sock);
1232 }
1233
1234 health_code_update(&health_thread_app_manage);
1235
1236 break;
1237 }
1238 } else {
1239 /*
1240 * At this point, we know that a registered application made
1241 * the event at poll_wait.
1242 */
1243 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1244 /* Removing from the poll set */
1245 ret = lttng_poll_del(&events, pollfd);
1246 if (ret < 0) {
1247 goto error;
1248 }
1249
1250 /* Socket closed on remote end. */
1251 ust_app_unregister(pollfd);
1252 break;
1253 }
1254 }
1255
1256 health_code_update(&health_thread_app_manage);
1257 }
1258 }
1259
1260 exit:
1261 error:
1262 lttng_poll_clean(&events);
1263 error_poll_create:
1264 error_testpoint:
1265 utils_close_pipe(apps_cmd_pipe);
1266 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1267
1268 /*
1269 * We don't clean the UST app hash table here since already registered
1270 * applications can still be controlled so let them be until the session
1271 * daemon dies or the applications stop.
1272 */
1273
1274 if (err) {
1275 health_error(&health_thread_app_manage);
1276 ERR("Health error occurred in %s", __func__);
1277 }
1278 health_exit(&health_thread_app_manage);
1279 DBG("Application communication apps thread cleanup complete");
1280 rcu_thread_offline();
1281 rcu_unregister_thread();
1282 return NULL;
1283 }
1284
1285 /*
1286 * Dispatch request from the registration threads to the application
1287 * communication thread.
1288 */
1289 static void *thread_dispatch_ust_registration(void *data)
1290 {
1291 int ret;
1292 struct cds_wfq_node *node;
1293 struct ust_command *ust_cmd = NULL;
1294
1295 DBG("[thread] Dispatch UST command started");
1296
1297 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1298 /* Atomically prepare the queue futex */
1299 futex_nto1_prepare(&ust_cmd_queue.futex);
1300
1301 do {
1302 /* Dequeue command for registration */
1303 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1304 if (node == NULL) {
1305 DBG("Woken up but nothing in the UST command queue");
1306 /* Continue thread execution */
1307 break;
1308 }
1309
1310 ust_cmd = caa_container_of(node, struct ust_command, node);
1311
1312 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1313 " gid:%d sock:%d name:%s (version %d.%d)",
1314 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1315 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1316 ust_cmd->sock, ust_cmd->reg_msg.name,
1317 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1318 /*
1319 * Inform apps thread of the new application registration. This
1320 * call is blocking so we can be assured that the data will be read
1321 * at some point in time or wait to the end of the world :)
1322 */
1323 if (apps_cmd_pipe[1] >= 0) {
1324 ret = write(apps_cmd_pipe[1], ust_cmd,
1325 sizeof(struct ust_command));
1326 if (ret < 0) {
1327 PERROR("write apps cmd pipe");
1328 if (errno == EBADF) {
1329 /*
1330 * We can't inform the application thread to process
1331 * registration. We will exit or else application
1332 * registration will not occur and tracing will never
1333 * start.
1334 */
1335 goto error;
1336 }
1337 }
1338 } else {
1339 /* Application manager thread is not available. */
1340 ret = close(ust_cmd->sock);
1341 if (ret < 0) {
1342 PERROR("close ust_cmd sock");
1343 }
1344 }
1345 free(ust_cmd);
1346 } while (node != NULL);
1347
1348 /* Futex wait on queue. Blocking call on futex() */
1349 futex_nto1_wait(&ust_cmd_queue.futex);
1350 }
1351
1352 error:
1353 DBG("Dispatch thread dying");
1354 return NULL;
1355 }
1356
1357 /*
1358 * This thread manage application registration.
1359 */
1360 static void *thread_registration_apps(void *data)
1361 {
1362 int sock = -1, i, ret, pollfd, err = -1;
1363 uint32_t revents, nb_fd;
1364 struct lttng_poll_event events;
1365 /*
1366 * Get allocated in this thread, enqueued to a global queue, dequeued and
1367 * freed in the manage apps thread.
1368 */
1369 struct ust_command *ust_cmd = NULL;
1370
1371 DBG("[thread] Manage application registration started");
1372
1373 if (testpoint(thread_registration_apps)) {
1374 goto error_testpoint;
1375 }
1376
1377 ret = lttcomm_listen_unix_sock(apps_sock);
1378 if (ret < 0) {
1379 goto error_listen;
1380 }
1381
1382 /*
1383 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1384 * more will be added to this poll set.
1385 */
1386 ret = create_thread_poll_set(&events, 2);
1387 if (ret < 0) {
1388 goto error_create_poll;
1389 }
1390
1391 /* Add the application registration socket */
1392 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1393 if (ret < 0) {
1394 goto error_poll_add;
1395 }
1396
1397 /* Notify all applications to register */
1398 ret = notify_ust_apps(1);
1399 if (ret < 0) {
1400 ERR("Failed to notify applications or create the wait shared memory.\n"
1401 "Execution continues but there might be problem for already\n"
1402 "running applications that wishes to register.");
1403 }
1404
1405 while (1) {
1406 DBG("Accepting application registration");
1407
1408 nb_fd = LTTNG_POLL_GETNB(&events);
1409
1410 /* Inifinite blocking call, waiting for transmission */
1411 restart:
1412 health_poll_update(&health_thread_app_reg);
1413 ret = lttng_poll_wait(&events, -1);
1414 health_poll_update(&health_thread_app_reg);
1415 if (ret < 0) {
1416 /*
1417 * Restart interrupted system call.
1418 */
1419 if (errno == EINTR) {
1420 goto restart;
1421 }
1422 goto error;
1423 }
1424
1425 for (i = 0; i < nb_fd; i++) {
1426 health_code_update(&health_thread_app_reg);
1427
1428 /* Fetch once the poll data */
1429 revents = LTTNG_POLL_GETEV(&events, i);
1430 pollfd = LTTNG_POLL_GETFD(&events, i);
1431
1432 /* Thread quit pipe has been closed. Killing thread. */
1433 ret = check_thread_quit_pipe(pollfd, revents);
1434 if (ret) {
1435 err = 0;
1436 goto exit;
1437 }
1438
1439 /* Event on the registration socket */
1440 if (pollfd == apps_sock) {
1441 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1442 ERR("Register apps socket poll error");
1443 goto error;
1444 } else if (revents & LPOLLIN) {
1445 sock = lttcomm_accept_unix_sock(apps_sock);
1446 if (sock < 0) {
1447 goto error;
1448 }
1449
1450 /*
1451 * Set the CLOEXEC flag. Return code is useless because
1452 * either way, the show must go on.
1453 */
1454 (void) utils_set_fd_cloexec(sock);
1455
1456 /* Create UST registration command for enqueuing */
1457 ust_cmd = zmalloc(sizeof(struct ust_command));
1458 if (ust_cmd == NULL) {
1459 PERROR("ust command zmalloc");
1460 goto error;
1461 }
1462
1463 /*
1464 * Using message-based transmissions to ensure we don't
1465 * have to deal with partially received messages.
1466 */
1467 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1468 if (ret < 0) {
1469 ERR("Exhausted file descriptors allowed for applications.");
1470 free(ust_cmd);
1471 ret = close(sock);
1472 if (ret) {
1473 PERROR("close");
1474 }
1475 sock = -1;
1476 continue;
1477 }
1478 health_code_update(&health_thread_app_reg);
1479 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1480 sizeof(struct ust_register_msg));
1481 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1482 if (ret < 0) {
1483 PERROR("lttcomm_recv_unix_sock register apps");
1484 } else {
1485 ERR("Wrong size received on apps register");
1486 }
1487 free(ust_cmd);
1488 ret = close(sock);
1489 if (ret) {
1490 PERROR("close");
1491 }
1492 lttng_fd_put(LTTNG_FD_APPS, 1);
1493 sock = -1;
1494 continue;
1495 }
1496 health_code_update(&health_thread_app_reg);
1497
1498 ust_cmd->sock = sock;
1499 sock = -1;
1500
1501 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1502 " gid:%d sock:%d name:%s (version %d.%d)",
1503 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1504 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1505 ust_cmd->sock, ust_cmd->reg_msg.name,
1506 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1507
1508 /*
1509 * Lock free enqueue the registration request. The red pill
1510 * has been taken! This apps will be part of the *system*.
1511 */
1512 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1513
1514 /*
1515 * Wake the registration queue futex. Implicit memory
1516 * barrier with the exchange in cds_wfq_enqueue.
1517 */
1518 futex_nto1_wake(&ust_cmd_queue.futex);
1519 }
1520 }
1521 }
1522 }
1523
1524 exit:
1525 error:
1526 if (err) {
1527 health_error(&health_thread_app_reg);
1528 ERR("Health error occurred in %s", __func__);
1529 }
1530
1531 /* Notify that the registration thread is gone */
1532 notify_ust_apps(0);
1533
1534 if (apps_sock >= 0) {
1535 ret = close(apps_sock);
1536 if (ret) {
1537 PERROR("close");
1538 }
1539 }
1540 if (sock >= 0) {
1541 ret = close(sock);
1542 if (ret) {
1543 PERROR("close");
1544 }
1545 lttng_fd_put(LTTNG_FD_APPS, 1);
1546 }
1547 unlink(apps_unix_sock_path);
1548
1549 error_poll_add:
1550 lttng_poll_clean(&events);
1551 error_listen:
1552 error_create_poll:
1553 error_testpoint:
1554 DBG("UST Registration thread cleanup complete");
1555 health_exit(&health_thread_app_reg);
1556
1557 return NULL;
1558 }
1559
1560 /*
1561 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1562 * exec or it will fails.
1563 */
1564 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1565 {
1566 int ret, clock_ret;
1567 struct timespec timeout;
1568
1569 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1570 consumer_data->consumer_thread_is_ready = 0;
1571
1572 /* Setup pthread condition */
1573 ret = pthread_condattr_init(&consumer_data->condattr);
1574 if (ret != 0) {
1575 errno = ret;
1576 PERROR("pthread_condattr_init consumer data");
1577 goto error;
1578 }
1579
1580 /*
1581 * Set the monotonic clock in order to make sure we DO NOT jump in time
1582 * between the clock_gettime() call and the timedwait call. See bug #324
1583 * for a more details and how we noticed it.
1584 */
1585 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1586 if (ret != 0) {
1587 errno = ret;
1588 PERROR("pthread_condattr_setclock consumer data");
1589 goto error;
1590 }
1591
1592 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1593 if (ret != 0) {
1594 errno = ret;
1595 PERROR("pthread_cond_init consumer data");
1596 goto error;
1597 }
1598
1599 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1600 consumer_data);
1601 if (ret != 0) {
1602 PERROR("pthread_create consumer");
1603 ret = -1;
1604 goto error;
1605 }
1606
1607 /* We are about to wait on a pthread condition */
1608 pthread_mutex_lock(&consumer_data->cond_mutex);
1609
1610 /* Get time for sem_timedwait absolute timeout */
1611 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1612 /*
1613 * Set the timeout for the condition timed wait even if the clock gettime
1614 * call fails since we might loop on that call and we want to avoid to
1615 * increment the timeout too many times.
1616 */
1617 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1618
1619 /*
1620 * The following loop COULD be skipped in some conditions so this is why we
1621 * set ret to 0 in order to make sure at least one round of the loop is
1622 * done.
1623 */
1624 ret = 0;
1625
1626 /*
1627 * Loop until the condition is reached or when a timeout is reached. Note
1628 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1629 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1630 * possible. This loop does not take any chances and works with both of
1631 * them.
1632 */
1633 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1634 if (clock_ret < 0) {
1635 PERROR("clock_gettime spawn consumer");
1636 /* Infinite wait for the consumerd thread to be ready */
1637 ret = pthread_cond_wait(&consumer_data->cond,
1638 &consumer_data->cond_mutex);
1639 } else {
1640 ret = pthread_cond_timedwait(&consumer_data->cond,
1641 &consumer_data->cond_mutex, &timeout);
1642 }
1643 }
1644
1645 /* Release the pthread condition */
1646 pthread_mutex_unlock(&consumer_data->cond_mutex);
1647
1648 if (ret != 0) {
1649 errno = ret;
1650 if (ret == ETIMEDOUT) {
1651 /*
1652 * Call has timed out so we kill the kconsumerd_thread and return
1653 * an error.
1654 */
1655 ERR("Condition timed out. The consumer thread was never ready."
1656 " Killing it");
1657 ret = pthread_cancel(consumer_data->thread);
1658 if (ret < 0) {
1659 PERROR("pthread_cancel consumer thread");
1660 }
1661 } else {
1662 PERROR("pthread_cond_wait failed consumer thread");
1663 }
1664 goto error;
1665 }
1666
1667 pthread_mutex_lock(&consumer_data->pid_mutex);
1668 if (consumer_data->pid == 0) {
1669 ERR("Consumerd did not start");
1670 pthread_mutex_unlock(&consumer_data->pid_mutex);
1671 goto error;
1672 }
1673 pthread_mutex_unlock(&consumer_data->pid_mutex);
1674
1675 return 0;
1676
1677 error:
1678 return ret;
1679 }
1680
1681 /*
1682 * Join consumer thread
1683 */
1684 static int join_consumer_thread(struct consumer_data *consumer_data)
1685 {
1686 void *status;
1687 int ret;
1688
1689 /* Consumer pid must be a real one. */
1690 if (consumer_data->pid > 0) {
1691 ret = kill(consumer_data->pid, SIGTERM);
1692 if (ret) {
1693 ERR("Error killing consumer daemon");
1694 return ret;
1695 }
1696 return pthread_join(consumer_data->thread, &status);
1697 } else {
1698 return 0;
1699 }
1700 }
1701
1702 /*
1703 * Fork and exec a consumer daemon (consumerd).
1704 *
1705 * Return pid if successful else -1.
1706 */
1707 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1708 {
1709 int ret;
1710 pid_t pid;
1711 const char *consumer_to_use;
1712 const char *verbosity;
1713 struct stat st;
1714
1715 DBG("Spawning consumerd");
1716
1717 pid = fork();
1718 if (pid == 0) {
1719 /*
1720 * Exec consumerd.
1721 */
1722 if (opt_verbose_consumer) {
1723 verbosity = "--verbose";
1724 } else {
1725 verbosity = "--quiet";
1726 }
1727 switch (consumer_data->type) {
1728 case LTTNG_CONSUMER_KERNEL:
1729 /*
1730 * Find out which consumerd to execute. We will first try the
1731 * 64-bit path, then the sessiond's installation directory, and
1732 * fallback on the 32-bit one,
1733 */
1734 DBG3("Looking for a kernel consumer at these locations:");
1735 DBG3(" 1) %s", consumerd64_bin);
1736 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1737 DBG3(" 3) %s", consumerd32_bin);
1738 if (stat(consumerd64_bin, &st) == 0) {
1739 DBG3("Found location #1");
1740 consumer_to_use = consumerd64_bin;
1741 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1742 DBG3("Found location #2");
1743 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1744 } else if (stat(consumerd32_bin, &st) == 0) {
1745 DBG3("Found location #3");
1746 consumer_to_use = consumerd32_bin;
1747 } else {
1748 DBG("Could not find any valid consumerd executable");
1749 break;
1750 }
1751 DBG("Using kernel consumer at: %s", consumer_to_use);
1752 execl(consumer_to_use,
1753 "lttng-consumerd", verbosity, "-k",
1754 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1755 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1756 NULL);
1757 break;
1758 case LTTNG_CONSUMER64_UST:
1759 {
1760 char *tmpnew = NULL;
1761
1762 if (consumerd64_libdir[0] != '\0') {
1763 char *tmp;
1764 size_t tmplen;
1765
1766 tmp = getenv("LD_LIBRARY_PATH");
1767 if (!tmp) {
1768 tmp = "";
1769 }
1770 tmplen = strlen("LD_LIBRARY_PATH=")
1771 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1772 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1773 if (!tmpnew) {
1774 ret = -ENOMEM;
1775 goto error;
1776 }
1777 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1778 strcat(tmpnew, consumerd64_libdir);
1779 if (tmp[0] != '\0') {
1780 strcat(tmpnew, ":");
1781 strcat(tmpnew, tmp);
1782 }
1783 ret = putenv(tmpnew);
1784 if (ret) {
1785 ret = -errno;
1786 goto error;
1787 }
1788 }
1789 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1790 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1791 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1792 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1793 NULL);
1794 if (consumerd64_libdir[0] != '\0') {
1795 free(tmpnew);
1796 }
1797 if (ret) {
1798 goto error;
1799 }
1800 break;
1801 }
1802 case LTTNG_CONSUMER32_UST:
1803 {
1804 char *tmpnew = NULL;
1805
1806 if (consumerd32_libdir[0] != '\0') {
1807 char *tmp;
1808 size_t tmplen;
1809
1810 tmp = getenv("LD_LIBRARY_PATH");
1811 if (!tmp) {
1812 tmp = "";
1813 }
1814 tmplen = strlen("LD_LIBRARY_PATH=")
1815 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1816 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1817 if (!tmpnew) {
1818 ret = -ENOMEM;
1819 goto error;
1820 }
1821 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1822 strcat(tmpnew, consumerd32_libdir);
1823 if (tmp[0] != '\0') {
1824 strcat(tmpnew, ":");
1825 strcat(tmpnew, tmp);
1826 }
1827 ret = putenv(tmpnew);
1828 if (ret) {
1829 ret = -errno;
1830 goto error;
1831 }
1832 }
1833 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1834 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1835 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1836 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1837 NULL);
1838 if (consumerd32_libdir[0] != '\0') {
1839 free(tmpnew);
1840 }
1841 if (ret) {
1842 goto error;
1843 }
1844 break;
1845 }
1846 default:
1847 PERROR("unknown consumer type");
1848 exit(EXIT_FAILURE);
1849 }
1850 if (errno != 0) {
1851 PERROR("kernel start consumer exec");
1852 }
1853 exit(EXIT_FAILURE);
1854 } else if (pid > 0) {
1855 ret = pid;
1856 } else {
1857 PERROR("start consumer fork");
1858 ret = -errno;
1859 }
1860 error:
1861 return ret;
1862 }
1863
1864 /*
1865 * Spawn the consumerd daemon and session daemon thread.
1866 */
1867 static int start_consumerd(struct consumer_data *consumer_data)
1868 {
1869 int ret, err;
1870
1871 /*
1872 * Set the listen() state on the socket since there is a possible race
1873 * between the exec() of the consumer daemon and this call if place in the
1874 * consumer thread. See bug #366 for more details.
1875 */
1876 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1877 if (ret < 0) {
1878 goto error;
1879 }
1880
1881 pthread_mutex_lock(&consumer_data->pid_mutex);
1882 if (consumer_data->pid != 0) {
1883 pthread_mutex_unlock(&consumer_data->pid_mutex);
1884 goto end;
1885 }
1886
1887 ret = spawn_consumerd(consumer_data);
1888 if (ret < 0) {
1889 ERR("Spawning consumerd failed");
1890 pthread_mutex_unlock(&consumer_data->pid_mutex);
1891 goto error;
1892 }
1893
1894 /* Setting up the consumer_data pid */
1895 consumer_data->pid = ret;
1896 DBG2("Consumer pid %d", consumer_data->pid);
1897 pthread_mutex_unlock(&consumer_data->pid_mutex);
1898
1899 DBG2("Spawning consumer control thread");
1900 ret = spawn_consumer_thread(consumer_data);
1901 if (ret < 0) {
1902 ERR("Fatal error spawning consumer control thread");
1903 goto error;
1904 }
1905
1906 end:
1907 return 0;
1908
1909 error:
1910 /* Cleanup already created socket on error. */
1911 if (consumer_data->err_sock >= 0) {
1912 err = close(consumer_data->err_sock);
1913 if (err < 0) {
1914 PERROR("close consumer data error socket");
1915 }
1916 }
1917 return ret;
1918 }
1919
1920 /*
1921 * Compute health status of each consumer. If one of them is zero (bad
1922 * state), we return 0.
1923 */
1924 static int check_consumer_health(void)
1925 {
1926 int ret;
1927
1928 ret = health_check_state(&kconsumer_data.health) &&
1929 health_check_state(&ustconsumer32_data.health) &&
1930 health_check_state(&ustconsumer64_data.health);
1931
1932 DBG3("Health consumer check %d", ret);
1933
1934 return ret;
1935 }
1936
1937 /*
1938 * Setup necessary data for kernel tracer action.
1939 */
1940 static int init_kernel_tracer(void)
1941 {
1942 int ret;
1943
1944 /* Modprobe lttng kernel modules */
1945 ret = modprobe_lttng_control();
1946 if (ret < 0) {
1947 goto error;
1948 }
1949
1950 /* Open debugfs lttng */
1951 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1952 if (kernel_tracer_fd < 0) {
1953 DBG("Failed to open %s", module_proc_lttng);
1954 ret = -1;
1955 goto error_open;
1956 }
1957
1958 /* Validate kernel version */
1959 ret = kernel_validate_version(kernel_tracer_fd);
1960 if (ret < 0) {
1961 goto error_version;
1962 }
1963
1964 ret = modprobe_lttng_data();
1965 if (ret < 0) {
1966 goto error_modules;
1967 }
1968
1969 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1970 return 0;
1971
1972 error_version:
1973 modprobe_remove_lttng_control();
1974 ret = close(kernel_tracer_fd);
1975 if (ret) {
1976 PERROR("close");
1977 }
1978 kernel_tracer_fd = -1;
1979 return LTTNG_ERR_KERN_VERSION;
1980
1981 error_modules:
1982 ret = close(kernel_tracer_fd);
1983 if (ret) {
1984 PERROR("close");
1985 }
1986
1987 error_open:
1988 modprobe_remove_lttng_control();
1989
1990 error:
1991 WARN("No kernel tracer available");
1992 kernel_tracer_fd = -1;
1993 if (!is_root) {
1994 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1995 } else {
1996 return LTTNG_ERR_KERN_NA;
1997 }
1998 }
1999
2000
2001 /*
2002 * Copy consumer output from the tracing session to the domain session. The
2003 * function also applies the right modification on a per domain basis for the
2004 * trace files destination directory.
2005 */
2006 static int copy_session_consumer(int domain, struct ltt_session *session)
2007 {
2008 int ret;
2009 const char *dir_name;
2010 struct consumer_output *consumer;
2011
2012 assert(session);
2013 assert(session->consumer);
2014
2015 switch (domain) {
2016 case LTTNG_DOMAIN_KERNEL:
2017 DBG3("Copying tracing session consumer output in kernel session");
2018 /*
2019 * XXX: We should audit the session creation and what this function
2020 * does "extra" in order to avoid a destroy since this function is used
2021 * in the domain session creation (kernel and ust) only. Same for UST
2022 * domain.
2023 */
2024 if (session->kernel_session->consumer) {
2025 consumer_destroy_output(session->kernel_session->consumer);
2026 }
2027 session->kernel_session->consumer =
2028 consumer_copy_output(session->consumer);
2029 /* Ease our life a bit for the next part */
2030 consumer = session->kernel_session->consumer;
2031 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2032 break;
2033 case LTTNG_DOMAIN_UST:
2034 DBG3("Copying tracing session consumer output in UST session");
2035 if (session->ust_session->consumer) {
2036 consumer_destroy_output(session->ust_session->consumer);
2037 }
2038 session->ust_session->consumer =
2039 consumer_copy_output(session->consumer);
2040 /* Ease our life a bit for the next part */
2041 consumer = session->ust_session->consumer;
2042 dir_name = DEFAULT_UST_TRACE_DIR;
2043 break;
2044 default:
2045 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2046 goto error;
2047 }
2048
2049 /* Append correct directory to subdir */
2050 strncat(consumer->subdir, dir_name,
2051 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2052 DBG3("Copy session consumer subdir %s", consumer->subdir);
2053
2054 ret = LTTNG_OK;
2055
2056 error:
2057 return ret;
2058 }
2059
2060 /*
2061 * Create an UST session and add it to the session ust list.
2062 */
2063 static int create_ust_session(struct ltt_session *session,
2064 struct lttng_domain *domain)
2065 {
2066 int ret;
2067 struct ltt_ust_session *lus = NULL;
2068
2069 assert(session);
2070 assert(domain);
2071 assert(session->consumer);
2072
2073 switch (domain->type) {
2074 case LTTNG_DOMAIN_UST:
2075 break;
2076 default:
2077 ERR("Unknown UST domain on create session %d", domain->type);
2078 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2079 goto error;
2080 }
2081
2082 DBG("Creating UST session");
2083
2084 lus = trace_ust_create_session(session->path, session->id, domain);
2085 if (lus == NULL) {
2086 ret = LTTNG_ERR_UST_SESS_FAIL;
2087 goto error;
2088 }
2089
2090 lus->uid = session->uid;
2091 lus->gid = session->gid;
2092 session->ust_session = lus;
2093
2094 /* Copy session output to the newly created UST session */
2095 ret = copy_session_consumer(domain->type, session);
2096 if (ret != LTTNG_OK) {
2097 goto error;
2098 }
2099
2100 return LTTNG_OK;
2101
2102 error:
2103 free(lus);
2104 session->ust_session = NULL;
2105 return ret;
2106 }
2107
2108 /*
2109 * Create a kernel tracer session then create the default channel.
2110 */
2111 static int create_kernel_session(struct ltt_session *session)
2112 {
2113 int ret;
2114
2115 DBG("Creating kernel session");
2116
2117 ret = kernel_create_session(session, kernel_tracer_fd);
2118 if (ret < 0) {
2119 ret = LTTNG_ERR_KERN_SESS_FAIL;
2120 goto error;
2121 }
2122
2123 /* Code flow safety */
2124 assert(session->kernel_session);
2125
2126 /* Copy session output to the newly created Kernel session */
2127 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2128 if (ret != LTTNG_OK) {
2129 goto error;
2130 }
2131
2132 /* Create directory(ies) on local filesystem. */
2133 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2134 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2135 ret = run_as_mkdir_recursive(
2136 session->kernel_session->consumer->dst.trace_path,
2137 S_IRWXU | S_IRWXG, session->uid, session->gid);
2138 if (ret < 0) {
2139 if (ret != -EEXIST) {
2140 ERR("Trace directory creation error");
2141 goto error;
2142 }
2143 }
2144 }
2145
2146 session->kernel_session->uid = session->uid;
2147 session->kernel_session->gid = session->gid;
2148
2149 return LTTNG_OK;
2150
2151 error:
2152 trace_kernel_destroy_session(session->kernel_session);
2153 session->kernel_session = NULL;
2154 return ret;
2155 }
2156
2157 /*
2158 * Count number of session permitted by uid/gid.
2159 */
2160 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2161 {
2162 unsigned int i = 0;
2163 struct ltt_session *session;
2164
2165 DBG("Counting number of available session for UID %d GID %d",
2166 uid, gid);
2167 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2168 /*
2169 * Only list the sessions the user can control.
2170 */
2171 if (!session_access_ok(session, uid, gid)) {
2172 continue;
2173 }
2174 i++;
2175 }
2176 return i;
2177 }
2178
2179 /*
2180 * Process the command requested by the lttng client within the command
2181 * context structure. This function make sure that the return structure (llm)
2182 * is set and ready for transmission before returning.
2183 *
2184 * Return any error encountered or 0 for success.
2185 *
2186 * "sock" is only used for special-case var. len data.
2187 */
2188 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2189 int *sock_error)
2190 {
2191 int ret = LTTNG_OK;
2192 int need_tracing_session = 1;
2193 int need_domain;
2194
2195 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2196
2197 *sock_error = 0;
2198
2199 switch (cmd_ctx->lsm->cmd_type) {
2200 case LTTNG_CREATE_SESSION:
2201 case LTTNG_DESTROY_SESSION:
2202 case LTTNG_LIST_SESSIONS:
2203 case LTTNG_LIST_DOMAINS:
2204 case LTTNG_START_TRACE:
2205 case LTTNG_STOP_TRACE:
2206 case LTTNG_DATA_PENDING:
2207 need_domain = 0;
2208 break;
2209 default:
2210 need_domain = 1;
2211 }
2212
2213 if (opt_no_kernel && need_domain
2214 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2215 if (!is_root) {
2216 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2217 } else {
2218 ret = LTTNG_ERR_KERN_NA;
2219 }
2220 goto error;
2221 }
2222
2223 /* Deny register consumer if we already have a spawned consumer. */
2224 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2225 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2226 if (kconsumer_data.pid > 0) {
2227 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2228 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2229 goto error;
2230 }
2231 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2232 }
2233
2234 /*
2235 * Check for command that don't needs to allocate a returned payload. We do
2236 * this here so we don't have to make the call for no payload at each
2237 * command.
2238 */
2239 switch(cmd_ctx->lsm->cmd_type) {
2240 case LTTNG_LIST_SESSIONS:
2241 case LTTNG_LIST_TRACEPOINTS:
2242 case LTTNG_LIST_TRACEPOINT_FIELDS:
2243 case LTTNG_LIST_DOMAINS:
2244 case LTTNG_LIST_CHANNELS:
2245 case LTTNG_LIST_EVENTS:
2246 break;
2247 default:
2248 /* Setup lttng message with no payload */
2249 ret = setup_lttng_msg(cmd_ctx, 0);
2250 if (ret < 0) {
2251 /* This label does not try to unlock the session */
2252 goto init_setup_error;
2253 }
2254 }
2255
2256 /* Commands that DO NOT need a session. */
2257 switch (cmd_ctx->lsm->cmd_type) {
2258 case LTTNG_CREATE_SESSION:
2259 case LTTNG_CALIBRATE:
2260 case LTTNG_LIST_SESSIONS:
2261 case LTTNG_LIST_TRACEPOINTS:
2262 case LTTNG_LIST_TRACEPOINT_FIELDS:
2263 need_tracing_session = 0;
2264 break;
2265 default:
2266 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2267 /*
2268 * We keep the session list lock across _all_ commands
2269 * for now, because the per-session lock does not
2270 * handle teardown properly.
2271 */
2272 session_lock_list();
2273 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2274 if (cmd_ctx->session == NULL) {
2275 if (cmd_ctx->lsm->session.name != NULL) {
2276 ret = LTTNG_ERR_SESS_NOT_FOUND;
2277 } else {
2278 /* If no session name specified */
2279 ret = LTTNG_ERR_SELECT_SESS;
2280 }
2281 goto error;
2282 } else {
2283 /* Acquire lock for the session */
2284 session_lock(cmd_ctx->session);
2285 }
2286 break;
2287 }
2288
2289 if (!need_domain) {
2290 goto skip_domain;
2291 }
2292
2293 /*
2294 * Check domain type for specific "pre-action".
2295 */
2296 switch (cmd_ctx->lsm->domain.type) {
2297 case LTTNG_DOMAIN_KERNEL:
2298 if (!is_root) {
2299 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2300 goto error;
2301 }
2302
2303 /* Kernel tracer check */
2304 if (kernel_tracer_fd == -1) {
2305 /* Basically, load kernel tracer modules */
2306 ret = init_kernel_tracer();
2307 if (ret != 0) {
2308 goto error;
2309 }
2310 }
2311
2312 /* Consumer is in an ERROR state. Report back to client */
2313 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2314 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2315 goto error;
2316 }
2317
2318 /* Need a session for kernel command */
2319 if (need_tracing_session) {
2320 if (cmd_ctx->session->kernel_session == NULL) {
2321 ret = create_kernel_session(cmd_ctx->session);
2322 if (ret < 0) {
2323 ret = LTTNG_ERR_KERN_SESS_FAIL;
2324 goto error;
2325 }
2326 }
2327
2328 /* Start the kernel consumer daemon */
2329 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2330 if (kconsumer_data.pid == 0 &&
2331 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2332 cmd_ctx->session->start_consumer) {
2333 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2334 ret = start_consumerd(&kconsumer_data);
2335 if (ret < 0) {
2336 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2337 goto error;
2338 }
2339 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2340 } else {
2341 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2342 }
2343
2344 /*
2345 * The consumer was just spawned so we need to add the socket to
2346 * the consumer output of the session if exist.
2347 */
2348 ret = consumer_create_socket(&kconsumer_data,
2349 cmd_ctx->session->kernel_session->consumer);
2350 if (ret < 0) {
2351 goto error;
2352 }
2353 }
2354
2355 break;
2356 case LTTNG_DOMAIN_UST:
2357 {
2358 /* Consumer is in an ERROR state. Report back to client */
2359 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2360 ret = LTTNG_ERR_NO_USTCONSUMERD;
2361 goto error;
2362 }
2363
2364 if (need_tracing_session) {
2365 /* Create UST session if none exist. */
2366 if (cmd_ctx->session->ust_session == NULL) {
2367 ret = create_ust_session(cmd_ctx->session,
2368 &cmd_ctx->lsm->domain);
2369 if (ret != LTTNG_OK) {
2370 goto error;
2371 }
2372 }
2373
2374 /* Start the UST consumer daemons */
2375 /* 64-bit */
2376 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2377 if (consumerd64_bin[0] != '\0' &&
2378 ustconsumer64_data.pid == 0 &&
2379 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2380 cmd_ctx->session->start_consumer) {
2381 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2382 ret = start_consumerd(&ustconsumer64_data);
2383 if (ret < 0) {
2384 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2385 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2386 goto error;
2387 }
2388
2389 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2390 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2391 } else {
2392 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2393 }
2394
2395 /*
2396 * Setup socket for consumer 64 bit. No need for atomic access
2397 * since it was set above and can ONLY be set in this thread.
2398 */
2399 ret = consumer_create_socket(&ustconsumer64_data,
2400 cmd_ctx->session->ust_session->consumer);
2401 if (ret < 0) {
2402 goto error;
2403 }
2404
2405 /* 32-bit */
2406 if (consumerd32_bin[0] != '\0' &&
2407 ustconsumer32_data.pid == 0 &&
2408 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2409 cmd_ctx->session->start_consumer) {
2410 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2411 ret = start_consumerd(&ustconsumer32_data);
2412 if (ret < 0) {
2413 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2414 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2415 goto error;
2416 }
2417
2418 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2419 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2420 } else {
2421 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2422 }
2423
2424 /*
2425 * Setup socket for consumer 64 bit. No need for atomic access
2426 * since it was set above and can ONLY be set in this thread.
2427 */
2428 ret = consumer_create_socket(&ustconsumer32_data,
2429 cmd_ctx->session->ust_session->consumer);
2430 if (ret < 0) {
2431 goto error;
2432 }
2433 }
2434 break;
2435 }
2436 default:
2437 break;
2438 }
2439 skip_domain:
2440
2441 /* Validate consumer daemon state when start/stop trace command */
2442 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2443 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2444 switch (cmd_ctx->lsm->domain.type) {
2445 case LTTNG_DOMAIN_UST:
2446 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2447 ret = LTTNG_ERR_NO_USTCONSUMERD;
2448 goto error;
2449 }
2450 break;
2451 case LTTNG_DOMAIN_KERNEL:
2452 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2453 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2454 goto error;
2455 }
2456 break;
2457 }
2458 }
2459
2460 /*
2461 * Check that the UID or GID match that of the tracing session.
2462 * The root user can interact with all sessions.
2463 */
2464 if (need_tracing_session) {
2465 if (!session_access_ok(cmd_ctx->session,
2466 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2467 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2468 ret = LTTNG_ERR_EPERM;
2469 goto error;
2470 }
2471 }
2472
2473 /* Process by command type */
2474 switch (cmd_ctx->lsm->cmd_type) {
2475 case LTTNG_ADD_CONTEXT:
2476 {
2477 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2478 cmd_ctx->lsm->u.context.channel_name,
2479 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2480 break;
2481 }
2482 case LTTNG_DISABLE_CHANNEL:
2483 {
2484 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2485 cmd_ctx->lsm->u.disable.channel_name);
2486 break;
2487 }
2488 case LTTNG_DISABLE_EVENT:
2489 {
2490 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2491 cmd_ctx->lsm->u.disable.channel_name,
2492 cmd_ctx->lsm->u.disable.name);
2493 break;
2494 }
2495 case LTTNG_DISABLE_ALL_EVENT:
2496 {
2497 DBG("Disabling all events");
2498
2499 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2500 cmd_ctx->lsm->u.disable.channel_name);
2501 break;
2502 }
2503 case LTTNG_DISABLE_CONSUMER:
2504 {
2505 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2506 break;
2507 }
2508 case LTTNG_ENABLE_CHANNEL:
2509 {
2510 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2511 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2512 break;
2513 }
2514 case LTTNG_ENABLE_CONSUMER:
2515 {
2516 /*
2517 * XXX: 0 means that this URI should be applied on the session. Should
2518 * be a DOMAIN enuam.
2519 */
2520 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2521 if (ret != LTTNG_OK) {
2522 goto error;
2523 }
2524
2525 if (cmd_ctx->lsm->domain.type == 0) {
2526 /* Add the URI for the UST session if a consumer is present. */
2527 if (cmd_ctx->session->ust_session &&
2528 cmd_ctx->session->ust_session->consumer) {
2529 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2530 } else if (cmd_ctx->session->kernel_session &&
2531 cmd_ctx->session->kernel_session->consumer) {
2532 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2533 cmd_ctx->session);
2534 }
2535 }
2536 break;
2537 }
2538 case LTTNG_ENABLE_EVENT:
2539 {
2540 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2541 cmd_ctx->lsm->u.enable.channel_name,
2542 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2543 break;
2544 }
2545 case LTTNG_ENABLE_ALL_EVENT:
2546 {
2547 DBG("Enabling all events");
2548
2549 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2550 cmd_ctx->lsm->u.enable.channel_name,
2551 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2552 break;
2553 }
2554 case LTTNG_LIST_TRACEPOINTS:
2555 {
2556 struct lttng_event *events;
2557 ssize_t nb_events;
2558
2559 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2560 if (nb_events < 0) {
2561 /* Return value is a negative lttng_error_code. */
2562 ret = -nb_events;
2563 goto error;
2564 }
2565
2566 /*
2567 * Setup lttng message with payload size set to the event list size in
2568 * bytes and then copy list into the llm payload.
2569 */
2570 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2571 if (ret < 0) {
2572 free(events);
2573 goto setup_error;
2574 }
2575
2576 /* Copy event list into message payload */
2577 memcpy(cmd_ctx->llm->payload, events,
2578 sizeof(struct lttng_event) * nb_events);
2579
2580 free(events);
2581
2582 ret = LTTNG_OK;
2583 break;
2584 }
2585 case LTTNG_LIST_TRACEPOINT_FIELDS:
2586 {
2587 struct lttng_event_field *fields;
2588 ssize_t nb_fields;
2589
2590 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2591 &fields);
2592 if (nb_fields < 0) {
2593 /* Return value is a negative lttng_error_code. */
2594 ret = -nb_fields;
2595 goto error;
2596 }
2597
2598 /*
2599 * Setup lttng message with payload size set to the event list size in
2600 * bytes and then copy list into the llm payload.
2601 */
2602 ret = setup_lttng_msg(cmd_ctx,
2603 sizeof(struct lttng_event_field) * nb_fields);
2604 if (ret < 0) {
2605 free(fields);
2606 goto setup_error;
2607 }
2608
2609 /* Copy event list into message payload */
2610 memcpy(cmd_ctx->llm->payload, fields,
2611 sizeof(struct lttng_event_field) * nb_fields);
2612
2613 free(fields);
2614
2615 ret = LTTNG_OK;
2616 break;
2617 }
2618 case LTTNG_SET_CONSUMER_URI:
2619 {
2620 size_t nb_uri, len;
2621 struct lttng_uri *uris;
2622
2623 nb_uri = cmd_ctx->lsm->u.uri.size;
2624 len = nb_uri * sizeof(struct lttng_uri);
2625
2626 if (nb_uri == 0) {
2627 ret = LTTNG_ERR_INVALID;
2628 goto error;
2629 }
2630
2631 uris = zmalloc(len);
2632 if (uris == NULL) {
2633 ret = LTTNG_ERR_FATAL;
2634 goto error;
2635 }
2636
2637 /* Receive variable len data */
2638 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2639 ret = lttcomm_recv_unix_sock(sock, uris, len);
2640 if (ret <= 0) {
2641 DBG("No URIs received from client... continuing");
2642 *sock_error = 1;
2643 ret = LTTNG_ERR_SESSION_FAIL;
2644 free(uris);
2645 goto error;
2646 }
2647
2648 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2649 nb_uri, uris);
2650 if (ret != LTTNG_OK) {
2651 free(uris);
2652 goto error;
2653 }
2654
2655 /*
2656 * XXX: 0 means that this URI should be applied on the session. Should
2657 * be a DOMAIN enuam.
2658 */
2659 if (cmd_ctx->lsm->domain.type == 0) {
2660 /* Add the URI for the UST session if a consumer is present. */
2661 if (cmd_ctx->session->ust_session &&
2662 cmd_ctx->session->ust_session->consumer) {
2663 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2664 nb_uri, uris);
2665 } else if (cmd_ctx->session->kernel_session &&
2666 cmd_ctx->session->kernel_session->consumer) {
2667 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2668 cmd_ctx->session, nb_uri, uris);
2669 }
2670 }
2671
2672 free(uris);
2673
2674 break;
2675 }
2676 case LTTNG_START_TRACE:
2677 {
2678 ret = cmd_start_trace(cmd_ctx->session);
2679 break;
2680 }
2681 case LTTNG_STOP_TRACE:
2682 {
2683 ret = cmd_stop_trace(cmd_ctx->session);
2684 break;
2685 }
2686 case LTTNG_CREATE_SESSION:
2687 {
2688 size_t nb_uri, len;
2689 struct lttng_uri *uris = NULL;
2690
2691 nb_uri = cmd_ctx->lsm->u.uri.size;
2692 len = nb_uri * sizeof(struct lttng_uri);
2693
2694 if (nb_uri > 0) {
2695 uris = zmalloc(len);
2696 if (uris == NULL) {
2697 ret = LTTNG_ERR_FATAL;
2698 goto error;
2699 }
2700
2701 /* Receive variable len data */
2702 DBG("Waiting for %zu URIs from client ...", nb_uri);
2703 ret = lttcomm_recv_unix_sock(sock, uris, len);
2704 if (ret <= 0) {
2705 DBG("No URIs received from client... continuing");
2706 *sock_error = 1;
2707 ret = LTTNG_ERR_SESSION_FAIL;
2708 free(uris);
2709 goto error;
2710 }
2711
2712 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2713 DBG("Creating session with ONE network URI is a bad call");
2714 ret = LTTNG_ERR_SESSION_FAIL;
2715 free(uris);
2716 goto error;
2717 }
2718 }
2719
2720 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2721 &cmd_ctx->creds);
2722
2723 free(uris);
2724
2725 break;
2726 }
2727 case LTTNG_DESTROY_SESSION:
2728 {
2729 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2730
2731 /* Set session to NULL so we do not unlock it after free. */
2732 cmd_ctx->session = NULL;
2733 break;
2734 }
2735 case LTTNG_LIST_DOMAINS:
2736 {
2737 ssize_t nb_dom;
2738 struct lttng_domain *domains;
2739
2740 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2741 if (nb_dom < 0) {
2742 /* Return value is a negative lttng_error_code. */
2743 ret = -nb_dom;
2744 goto error;
2745 }
2746
2747 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2748 if (ret < 0) {
2749 goto setup_error;
2750 }
2751
2752 /* Copy event list into message payload */
2753 memcpy(cmd_ctx->llm->payload, domains,
2754 nb_dom * sizeof(struct lttng_domain));
2755
2756 free(domains);
2757
2758 ret = LTTNG_OK;
2759 break;
2760 }
2761 case LTTNG_LIST_CHANNELS:
2762 {
2763 int nb_chan;
2764 struct lttng_channel *channels;
2765
2766 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2767 cmd_ctx->session, &channels);
2768 if (nb_chan < 0) {
2769 /* Return value is a negative lttng_error_code. */
2770 ret = -nb_chan;
2771 goto error;
2772 }
2773
2774 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2775 if (ret < 0) {
2776 goto setup_error;
2777 }
2778
2779 /* Copy event list into message payload */
2780 memcpy(cmd_ctx->llm->payload, channels,
2781 nb_chan * sizeof(struct lttng_channel));
2782
2783 free(channels);
2784
2785 ret = LTTNG_OK;
2786 break;
2787 }
2788 case LTTNG_LIST_EVENTS:
2789 {
2790 ssize_t nb_event;
2791 struct lttng_event *events = NULL;
2792
2793 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2794 cmd_ctx->lsm->u.list.channel_name, &events);
2795 if (nb_event < 0) {
2796 /* Return value is a negative lttng_error_code. */
2797 ret = -nb_event;
2798 goto error;
2799 }
2800
2801 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2802 if (ret < 0) {
2803 goto setup_error;
2804 }
2805
2806 /* Copy event list into message payload */
2807 memcpy(cmd_ctx->llm->payload, events,
2808 nb_event * sizeof(struct lttng_event));
2809
2810 free(events);
2811
2812 ret = LTTNG_OK;
2813 break;
2814 }
2815 case LTTNG_LIST_SESSIONS:
2816 {
2817 unsigned int nr_sessions;
2818
2819 session_lock_list();
2820 nr_sessions = lttng_sessions_count(
2821 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2822 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2823
2824 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2825 if (ret < 0) {
2826 session_unlock_list();
2827 goto setup_error;
2828 }
2829
2830 /* Filled the session array */
2831 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2832 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2833 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2834
2835 session_unlock_list();
2836
2837 ret = LTTNG_OK;
2838 break;
2839 }
2840 case LTTNG_CALIBRATE:
2841 {
2842 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2843 &cmd_ctx->lsm->u.calibrate);
2844 break;
2845 }
2846 case LTTNG_REGISTER_CONSUMER:
2847 {
2848 struct consumer_data *cdata;
2849
2850 switch (cmd_ctx->lsm->domain.type) {
2851 case LTTNG_DOMAIN_KERNEL:
2852 cdata = &kconsumer_data;
2853 break;
2854 default:
2855 ret = LTTNG_ERR_UND;
2856 goto error;
2857 }
2858
2859 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2860 cmd_ctx->lsm->u.reg.path, cdata);
2861 break;
2862 }
2863 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2864 {
2865 struct lttng_filter_bytecode *bytecode;
2866
2867 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2868 ret = LTTNG_ERR_FILTER_INVAL;
2869 goto error;
2870 }
2871 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2872 ret = LTTNG_ERR_FILTER_INVAL;
2873 goto error;
2874 }
2875 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2876 if (!bytecode) {
2877 ret = LTTNG_ERR_FILTER_NOMEM;
2878 goto error;
2879 }
2880 /* Receive var. len. data */
2881 DBG("Receiving var len data from client ...");
2882 ret = lttcomm_recv_unix_sock(sock, bytecode,
2883 cmd_ctx->lsm->u.enable.bytecode_len);
2884 if (ret <= 0) {
2885 DBG("Nothing recv() from client var len data... continuing");
2886 *sock_error = 1;
2887 ret = LTTNG_ERR_FILTER_INVAL;
2888 goto error;
2889 }
2890
2891 if (bytecode->len + sizeof(*bytecode)
2892 != cmd_ctx->lsm->u.enable.bytecode_len) {
2893 free(bytecode);
2894 ret = LTTNG_ERR_FILTER_INVAL;
2895 goto error;
2896 }
2897
2898 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2899 cmd_ctx->lsm->u.enable.channel_name,
2900 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2901 break;
2902 }
2903 case LTTNG_DATA_PENDING:
2904 {
2905 ret = cmd_data_pending(cmd_ctx->session);
2906 break;
2907 }
2908 default:
2909 ret = LTTNG_ERR_UND;
2910 break;
2911 }
2912
2913 error:
2914 if (cmd_ctx->llm == NULL) {
2915 DBG("Missing llm structure. Allocating one.");
2916 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2917 goto setup_error;
2918 }
2919 }
2920 /* Set return code */
2921 cmd_ctx->llm->ret_code = ret;
2922 setup_error:
2923 if (cmd_ctx->session) {
2924 session_unlock(cmd_ctx->session);
2925 }
2926 if (need_tracing_session) {
2927 session_unlock_list();
2928 }
2929 init_setup_error:
2930 return ret;
2931 }
2932
2933 /*
2934 * Thread managing health check socket.
2935 */
2936 static void *thread_manage_health(void *data)
2937 {
2938 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2939 uint32_t revents, nb_fd;
2940 struct lttng_poll_event events;
2941 struct lttcomm_health_msg msg;
2942 struct lttcomm_health_data reply;
2943
2944 DBG("[thread] Manage health check started");
2945
2946 rcu_register_thread();
2947
2948 /* Create unix socket */
2949 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2950 if (sock < 0) {
2951 ERR("Unable to create health check Unix socket");
2952 ret = -1;
2953 goto error;
2954 }
2955
2956 /*
2957 * Set the CLOEXEC flag. Return code is useless because either way, the
2958 * show must go on.
2959 */
2960 (void) utils_set_fd_cloexec(sock);
2961
2962 ret = lttcomm_listen_unix_sock(sock);
2963 if (ret < 0) {
2964 goto error;
2965 }
2966
2967 /*
2968 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2969 * more will be added to this poll set.
2970 */
2971 ret = create_thread_poll_set(&events, 2);
2972 if (ret < 0) {
2973 goto error;
2974 }
2975
2976 /* Add the application registration socket */
2977 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2978 if (ret < 0) {
2979 goto error;
2980 }
2981
2982 while (1) {
2983 DBG("Health check ready");
2984
2985 nb_fd = LTTNG_POLL_GETNB(&events);
2986
2987 /* Inifinite blocking call, waiting for transmission */
2988 restart:
2989 ret = lttng_poll_wait(&events, -1);
2990 if (ret < 0) {
2991 /*
2992 * Restart interrupted system call.
2993 */
2994 if (errno == EINTR) {
2995 goto restart;
2996 }
2997 goto error;
2998 }
2999
3000 for (i = 0; i < nb_fd; i++) {
3001 /* Fetch once the poll data */
3002 revents = LTTNG_POLL_GETEV(&events, i);
3003 pollfd = LTTNG_POLL_GETFD(&events, i);
3004
3005 /* Thread quit pipe has been closed. Killing thread. */
3006 ret = check_thread_quit_pipe(pollfd, revents);
3007 if (ret) {
3008 err = 0;
3009 goto exit;
3010 }
3011
3012 /* Event on the registration socket */
3013 if (pollfd == sock) {
3014 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3015 ERR("Health socket poll error");
3016 goto error;
3017 }
3018 }
3019 }
3020
3021 new_sock = lttcomm_accept_unix_sock(sock);
3022 if (new_sock < 0) {
3023 goto error;
3024 }
3025
3026 /*
3027 * Set the CLOEXEC flag. Return code is useless because either way, the
3028 * show must go on.
3029 */
3030 (void) utils_set_fd_cloexec(new_sock);
3031
3032 DBG("Receiving data from client for health...");
3033 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3034 if (ret <= 0) {
3035 DBG("Nothing recv() from client... continuing");
3036 ret = close(new_sock);
3037 if (ret) {
3038 PERROR("close");
3039 }
3040 new_sock = -1;
3041 continue;
3042 }
3043
3044 rcu_thread_online();
3045
3046 switch (msg.component) {
3047 case LTTNG_HEALTH_CMD:
3048 reply.ret_code = health_check_state(&health_thread_cmd);
3049 break;
3050 case LTTNG_HEALTH_APP_MANAGE:
3051 reply.ret_code = health_check_state(&health_thread_app_manage);
3052 break;
3053 case LTTNG_HEALTH_APP_REG:
3054 reply.ret_code = health_check_state(&health_thread_app_reg);
3055 break;
3056 case LTTNG_HEALTH_KERNEL:
3057 reply.ret_code = health_check_state(&health_thread_kernel);
3058 break;
3059 case LTTNG_HEALTH_CONSUMER:
3060 reply.ret_code = check_consumer_health();
3061 break;
3062 case LTTNG_HEALTH_ALL:
3063 reply.ret_code =
3064 health_check_state(&health_thread_app_manage) &&
3065 health_check_state(&health_thread_app_reg) &&
3066 health_check_state(&health_thread_cmd) &&
3067 health_check_state(&health_thread_kernel) &&
3068 check_consumer_health();
3069 break;
3070 default:
3071 reply.ret_code = LTTNG_ERR_UND;
3072 break;
3073 }
3074
3075 /*
3076 * Flip ret value since 0 is a success and 1 indicates a bad health for
3077 * the client where in the sessiond it is the opposite. Again, this is
3078 * just to make things easier for us poor developer which enjoy a lot
3079 * lazyness.
3080 */
3081 if (reply.ret_code == 0 || reply.ret_code == 1) {
3082 reply.ret_code = !reply.ret_code;
3083 }
3084
3085 DBG2("Health check return value %d", reply.ret_code);
3086
3087 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3088 if (ret < 0) {
3089 ERR("Failed to send health data back to client");
3090 }
3091
3092 /* End of transmission */
3093 ret = close(new_sock);
3094 if (ret) {
3095 PERROR("close");
3096 }
3097 new_sock = -1;
3098 }
3099
3100 exit:
3101 error:
3102 if (err) {
3103 ERR("Health error occurred in %s", __func__);
3104 }
3105 DBG("Health check thread dying");
3106 unlink(health_unix_sock_path);
3107 if (sock >= 0) {
3108 ret = close(sock);
3109 if (ret) {
3110 PERROR("close");
3111 }
3112 }
3113 if (new_sock >= 0) {
3114 ret = close(new_sock);
3115 if (ret) {
3116 PERROR("close");
3117 }
3118 }
3119
3120 lttng_poll_clean(&events);
3121
3122 rcu_unregister_thread();
3123 return NULL;
3124 }
3125
3126 /*
3127 * This thread manage all clients request using the unix client socket for
3128 * communication.
3129 */
3130 static void *thread_manage_clients(void *data)
3131 {
3132 int sock = -1, ret, i, pollfd, err = -1;
3133 int sock_error;
3134 uint32_t revents, nb_fd;
3135 struct command_ctx *cmd_ctx = NULL;
3136 struct lttng_poll_event events;
3137
3138 DBG("[thread] Manage client started");
3139
3140 rcu_register_thread();
3141
3142 if (testpoint(thread_manage_clients)) {
3143 goto error_testpoint;
3144 }
3145
3146 health_code_update(&health_thread_cmd);
3147
3148 ret = lttcomm_listen_unix_sock(client_sock);
3149 if (ret < 0) {
3150 goto error_listen;
3151 }
3152
3153 /*
3154 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3155 * more will be added to this poll set.
3156 */
3157 ret = create_thread_poll_set(&events, 2);
3158 if (ret < 0) {
3159 goto error_create_poll;
3160 }
3161
3162 /* Add the application registration socket */
3163 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3164 if (ret < 0) {
3165 goto error;
3166 }
3167
3168 /*
3169 * Notify parent pid that we are ready to accept command for client side.
3170 */
3171 if (opt_sig_parent) {
3172 kill(ppid, SIGUSR1);
3173 }
3174
3175 if (testpoint(thread_manage_clients_before_loop)) {
3176 goto error;
3177 }
3178
3179 health_code_update(&health_thread_cmd);
3180
3181 while (1) {
3182 DBG("Accepting client command ...");
3183
3184 nb_fd = LTTNG_POLL_GETNB(&events);
3185
3186 /* Inifinite blocking call, waiting for transmission */
3187 restart:
3188 health_poll_update(&health_thread_cmd);
3189 ret = lttng_poll_wait(&events, -1);
3190 health_poll_update(&health_thread_cmd);
3191 if (ret < 0) {
3192 /*
3193 * Restart interrupted system call.
3194 */
3195 if (errno == EINTR) {
3196 goto restart;
3197 }
3198 goto error;
3199 }
3200
3201 for (i = 0; i < nb_fd; i++) {
3202 /* Fetch once the poll data */
3203 revents = LTTNG_POLL_GETEV(&events, i);
3204 pollfd = LTTNG_POLL_GETFD(&events, i);
3205
3206 health_code_update(&health_thread_cmd);
3207
3208 /* Thread quit pipe has been closed. Killing thread. */
3209 ret = check_thread_quit_pipe(pollfd, revents);
3210 if (ret) {
3211 err = 0;
3212 goto exit;
3213 }
3214
3215 /* Event on the registration socket */
3216 if (pollfd == client_sock) {
3217 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3218 ERR("Client socket poll error");
3219 goto error;
3220 }
3221 }
3222 }
3223
3224 DBG("Wait for client response");
3225
3226 health_code_update(&health_thread_cmd);
3227
3228 sock = lttcomm_accept_unix_sock(client_sock);
3229 if (sock < 0) {
3230 goto error;
3231 }
3232
3233 /*
3234 * Set the CLOEXEC flag. Return code is useless because either way, the
3235 * show must go on.
3236 */
3237 (void) utils_set_fd_cloexec(sock);
3238
3239 /* Set socket option for credentials retrieval */
3240 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3241 if (ret < 0) {
3242 goto error;
3243 }
3244
3245 /* Allocate context command to process the client request */
3246 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3247 if (cmd_ctx == NULL) {
3248 PERROR("zmalloc cmd_ctx");
3249 goto error;
3250 }
3251
3252 /* Allocate data buffer for reception */
3253 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3254 if (cmd_ctx->lsm == NULL) {
3255 PERROR("zmalloc cmd_ctx->lsm");
3256 goto error;
3257 }
3258
3259 cmd_ctx->llm = NULL;
3260 cmd_ctx->session = NULL;
3261
3262 health_code_update(&health_thread_cmd);
3263
3264 /*
3265 * Data is received from the lttng client. The struct
3266 * lttcomm_session_msg (lsm) contains the command and data request of
3267 * the client.
3268 */
3269 DBG("Receiving data from client ...");
3270 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3271 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3272 if (ret <= 0) {
3273 DBG("Nothing recv() from client... continuing");
3274 ret = close(sock);
3275 if (ret) {
3276 PERROR("close");
3277 }
3278 sock = -1;
3279 clean_command_ctx(&cmd_ctx);
3280 continue;
3281 }
3282
3283 health_code_update(&health_thread_cmd);
3284
3285 // TODO: Validate cmd_ctx including sanity check for
3286 // security purpose.
3287
3288 rcu_thread_online();
3289 /*
3290 * This function dispatch the work to the kernel or userspace tracer
3291 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3292 * informations for the client. The command context struct contains
3293 * everything this function may needs.
3294 */
3295 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3296 rcu_thread_offline();
3297 if (ret < 0) {
3298 if (sock_error) {
3299 ret = close(sock);
3300 if (ret) {
3301 PERROR("close");
3302 }
3303 sock = -1;
3304 }
3305 /*
3306 * TODO: Inform client somehow of the fatal error. At
3307 * this point, ret < 0 means that a zmalloc failed
3308 * (ENOMEM). Error detected but still accept
3309 * command, unless a socket error has been
3310 * detected.
3311 */
3312 clean_command_ctx(&cmd_ctx);
3313 continue;
3314 }
3315
3316 health_code_update(&health_thread_cmd);
3317
3318 DBG("Sending response (size: %d, retcode: %s)",
3319 cmd_ctx->lttng_msg_size,
3320 lttng_strerror(-cmd_ctx->llm->ret_code));
3321 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3322 if (ret < 0) {
3323 ERR("Failed to send data back to client");
3324 }
3325
3326 /* End of transmission */
3327 ret = close(sock);
3328 if (ret) {
3329 PERROR("close");
3330 }
3331 sock = -1;
3332
3333 clean_command_ctx(&cmd_ctx);
3334
3335 health_code_update(&health_thread_cmd);
3336 }
3337
3338 exit:
3339 error:
3340 if (sock >= 0) {
3341 ret = close(sock);
3342 if (ret) {
3343 PERROR("close");
3344 }
3345 }
3346
3347 lttng_poll_clean(&events);
3348 clean_command_ctx(&cmd_ctx);
3349
3350 error_listen:
3351 error_create_poll:
3352 error_testpoint:
3353 unlink(client_unix_sock_path);
3354 if (client_sock >= 0) {
3355 ret = close(client_sock);
3356 if (ret) {
3357 PERROR("close");
3358 }
3359 }
3360
3361 if (err) {
3362 health_error(&health_thread_cmd);
3363 ERR("Health error occurred in %s", __func__);
3364 }
3365
3366 health_exit(&health_thread_cmd);
3367
3368 DBG("Client thread dying");
3369
3370 rcu_unregister_thread();
3371 return NULL;
3372 }
3373
3374
3375 /*
3376 * usage function on stderr
3377 */
3378 static void usage(void)
3379 {
3380 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3381 fprintf(stderr, " -h, --help Display this usage.\n");
3382 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3383 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3384 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3385 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3386 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3387 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3388 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3389 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3390 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3391 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3392 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3393 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3394 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3395 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3396 fprintf(stderr, " -V, --version Show version number.\n");
3397 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3398 fprintf(stderr, " -q, --quiet No output at all.\n");
3399 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3400 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3401 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3402 }
3403
3404 /*
3405 * daemon argument parsing
3406 */
3407 static int parse_args(int argc, char **argv)
3408 {
3409 int c;
3410
3411 static struct option long_options[] = {
3412 { "client-sock", 1, 0, 'c' },
3413 { "apps-sock", 1, 0, 'a' },
3414 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3415 { "kconsumerd-err-sock", 1, 0, 'E' },
3416 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3417 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3418 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3419 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3420 { "consumerd32-path", 1, 0, 'u' },
3421 { "consumerd32-libdir", 1, 0, 'U' },
3422 { "consumerd64-path", 1, 0, 't' },
3423 { "consumerd64-libdir", 1, 0, 'T' },
3424 { "daemonize", 0, 0, 'd' },
3425 { "sig-parent", 0, 0, 'S' },
3426 { "help", 0, 0, 'h' },
3427 { "group", 1, 0, 'g' },
3428 { "version", 0, 0, 'V' },
3429 { "quiet", 0, 0, 'q' },
3430 { "verbose", 0, 0, 'v' },
3431 { "verbose-consumer", 0, 0, 'Z' },
3432 { "no-kernel", 0, 0, 'N' },
3433 { NULL, 0, 0, 0 }
3434 };
3435
3436 while (1) {
3437 int option_index = 0;
3438 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3439 long_options, &option_index);
3440 if (c == -1) {
3441 break;
3442 }
3443
3444 switch (c) {
3445 case 0:
3446 fprintf(stderr, "option %s", long_options[option_index].name);
3447 if (optarg) {
3448 fprintf(stderr, " with arg %s\n", optarg);
3449 }
3450 break;
3451 case 'c':
3452 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3453 break;
3454 case 'a':
3455 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3456 break;
3457 case 'd':
3458 opt_daemon = 1;
3459 break;
3460 case 'g':
3461 opt_tracing_group = optarg;
3462 break;
3463 case 'h':
3464 usage();
3465 exit(EXIT_FAILURE);
3466 case 'V':
3467 fprintf(stdout, "%s\n", VERSION);
3468 exit(EXIT_SUCCESS);
3469 case 'S':
3470 opt_sig_parent = 1;
3471 break;
3472 case 'E':
3473 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3474 break;
3475 case 'C':
3476 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3477 break;
3478 case 'F':
3479 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3480 break;
3481 case 'D':
3482 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3483 break;
3484 case 'H':
3485 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3486 break;
3487 case 'G':
3488 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3489 break;
3490 case 'N':
3491 opt_no_kernel = 1;
3492 break;
3493 case 'q':
3494 lttng_opt_quiet = 1;
3495 break;
3496 case 'v':
3497 /* Verbose level can increase using multiple -v */
3498 lttng_opt_verbose += 1;
3499 break;
3500 case 'Z':
3501 opt_verbose_consumer += 1;
3502 break;
3503 case 'u':
3504 consumerd32_bin= optarg;
3505 break;
3506 case 'U':
3507 consumerd32_libdir = optarg;
3508 break;
3509 case 't':
3510 consumerd64_bin = optarg;
3511 break;
3512 case 'T':
3513 consumerd64_libdir = optarg;
3514 break;
3515 default:
3516 /* Unknown option or other error.
3517 * Error is printed by getopt, just return */
3518 return -1;
3519 }
3520 }
3521
3522 return 0;
3523 }
3524
3525 /*
3526 * Creates the two needed socket by the daemon.
3527 * apps_sock - The communication socket for all UST apps.
3528 * client_sock - The communication of the cli tool (lttng).
3529 */
3530 static int init_daemon_socket(void)
3531 {
3532 int ret = 0;
3533 mode_t old_umask;
3534
3535 old_umask = umask(0);
3536
3537 /* Create client tool unix socket */
3538 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3539 if (client_sock < 0) {
3540 ERR("Create unix sock failed: %s", client_unix_sock_path);
3541 ret = -1;
3542 goto end;
3543 }
3544
3545 /* Set the cloexec flag */
3546 ret = utils_set_fd_cloexec(client_sock);
3547 if (ret < 0) {
3548 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3549 "Continuing but note that the consumer daemon will have a "
3550 "reference to this socket on exec()", client_sock);
3551 }
3552
3553 /* File permission MUST be 660 */
3554 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3555 if (ret < 0) {
3556 ERR("Set file permissions failed: %s", client_unix_sock_path);
3557 PERROR("chmod");
3558 goto end;
3559 }
3560
3561 /* Create the application unix socket */
3562 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3563 if (apps_sock < 0) {
3564 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3565 ret = -1;
3566 goto end;
3567 }
3568
3569 /* Set the cloexec flag */
3570 ret = utils_set_fd_cloexec(apps_sock);
3571 if (ret < 0) {
3572 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3573 "Continuing but note that the consumer daemon will have a "
3574 "reference to this socket on exec()", apps_sock);
3575 }
3576
3577 /* File permission MUST be 666 */
3578 ret = chmod(apps_unix_sock_path,
3579 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3580 if (ret < 0) {
3581 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3582 PERROR("chmod");
3583 goto end;
3584 }
3585
3586 DBG3("Session daemon client socket %d and application socket %d created",
3587 client_sock, apps_sock);
3588
3589 end:
3590 umask(old_umask);
3591 return ret;
3592 }
3593
3594 /*
3595 * Check if the global socket is available, and if a daemon is answering at the
3596 * other side. If yes, error is returned.
3597 */
3598 static int check_existing_daemon(void)
3599 {
3600 /* Is there anybody out there ? */
3601 if (lttng_session_daemon_alive()) {
3602 return -EEXIST;
3603 }
3604
3605 return 0;
3606 }
3607
3608 /*
3609 * Set the tracing group gid onto the client socket.
3610 *
3611 * Race window between mkdir and chown is OK because we are going from more
3612 * permissive (root.root) to less permissive (root.tracing).
3613 */
3614 static int set_permissions(char *rundir)
3615 {
3616 int ret;
3617 gid_t gid;
3618
3619 ret = allowed_group();
3620 if (ret < 0) {
3621 WARN("No tracing group detected");
3622 ret = 0;
3623 goto end;
3624 }
3625
3626 gid = ret;
3627
3628 /* Set lttng run dir */
3629 ret = chown(rundir, 0, gid);
3630 if (ret < 0) {
3631 ERR("Unable to set group on %s", rundir);
3632 PERROR("chown");
3633 }
3634
3635 /* Ensure tracing group can search the run dir */
3636 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3637 if (ret < 0) {
3638 ERR("Unable to set permissions on %s", rundir);
3639 PERROR("chmod");
3640 }
3641
3642 /* lttng client socket path */
3643 ret = chown(client_unix_sock_path, 0, gid);
3644 if (ret < 0) {
3645 ERR("Unable to set group on %s", client_unix_sock_path);
3646 PERROR("chown");
3647 }
3648
3649 /* kconsumer error socket path */
3650 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3651 if (ret < 0) {
3652 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3653 PERROR("chown");
3654 }
3655
3656 /* 64-bit ustconsumer error socket path */
3657 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3658 if (ret < 0) {
3659 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3660 PERROR("chown");
3661 }
3662
3663 /* 32-bit ustconsumer compat32 error socket path */
3664 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3665 if (ret < 0) {
3666 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3667 PERROR("chown");
3668 }
3669
3670 DBG("All permissions are set");
3671
3672 end:
3673 return ret;
3674 }
3675
3676 /*
3677 * Create the lttng run directory needed for all global sockets and pipe.
3678 */
3679 static int create_lttng_rundir(const char *rundir)
3680 {
3681 int ret;
3682
3683 DBG3("Creating LTTng run directory: %s", rundir);
3684
3685 ret = mkdir(rundir, S_IRWXU);
3686 if (ret < 0) {
3687 if (errno != EEXIST) {
3688 ERR("Unable to create %s", rundir);
3689 goto error;
3690 } else {
3691 ret = 0;
3692 }
3693 }
3694
3695 error:
3696 return ret;
3697 }
3698
3699 /*
3700 * Setup sockets and directory needed by the kconsumerd communication with the
3701 * session daemon.
3702 */
3703 static int set_consumer_sockets(struct consumer_data *consumer_data,
3704 const char *rundir)
3705 {
3706 int ret;
3707 char path[PATH_MAX];
3708
3709 switch (consumer_data->type) {
3710 case LTTNG_CONSUMER_KERNEL:
3711 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3712 break;
3713 case LTTNG_CONSUMER64_UST:
3714 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3715 break;
3716 case LTTNG_CONSUMER32_UST:
3717 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3718 break;
3719 default:
3720 ERR("Consumer type unknown");
3721 ret = -EINVAL;
3722 goto error;
3723 }
3724
3725 DBG2("Creating consumer directory: %s", path);
3726
3727 ret = mkdir(path, S_IRWXU);
3728 if (ret < 0) {
3729 if (errno != EEXIST) {
3730 PERROR("mkdir");
3731 ERR("Failed to create %s", path);
3732 goto error;
3733 }
3734 ret = -1;
3735 }
3736
3737 /* Create the kconsumerd error unix socket */
3738 consumer_data->err_sock =
3739 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3740 if (consumer_data->err_sock < 0) {
3741 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3742 ret = -1;
3743 goto error;
3744 }
3745
3746 /* File permission MUST be 660 */
3747 ret = chmod(consumer_data->err_unix_sock_path,
3748 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3749 if (ret < 0) {
3750 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3751 PERROR("chmod");
3752 goto error;
3753 }
3754
3755 error:
3756 return ret;
3757 }
3758
3759 /*
3760 * Signal handler for the daemon
3761 *
3762 * Simply stop all worker threads, leaving main() return gracefully after
3763 * joining all threads and calling cleanup().
3764 */
3765 static void sighandler(int sig)
3766 {
3767 switch (sig) {
3768 case SIGPIPE:
3769 DBG("SIGPIPE caught");
3770 return;
3771 case SIGINT:
3772 DBG("SIGINT caught");
3773 stop_threads();
3774 break;
3775 case SIGTERM:
3776 DBG("SIGTERM caught");
3777 stop_threads();
3778 break;
3779 default:
3780 break;
3781 }
3782 }
3783
3784 /*
3785 * Setup signal handler for :
3786 * SIGINT, SIGTERM, SIGPIPE
3787 */
3788 static int set_signal_handler(void)
3789 {
3790 int ret = 0;
3791 struct sigaction sa;
3792 sigset_t sigset;
3793
3794 if ((ret = sigemptyset(&sigset)) < 0) {
3795 PERROR("sigemptyset");
3796 return ret;
3797 }
3798
3799 sa.sa_handler = sighandler;
3800 sa.sa_mask = sigset;
3801 sa.sa_flags = 0;
3802 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3803 PERROR("sigaction");
3804 return ret;
3805 }
3806
3807 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3808 PERROR("sigaction");
3809 return ret;
3810 }
3811
3812 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3813 PERROR("sigaction");
3814 return ret;
3815 }
3816
3817 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3818
3819 return ret;
3820 }
3821
3822 /*
3823 * Set open files limit to unlimited. This daemon can open a large number of
3824 * file descriptors in order to consumer multiple kernel traces.
3825 */
3826 static void set_ulimit(void)
3827 {
3828 int ret;
3829 struct rlimit lim;
3830
3831 /* The kernel does not allowed an infinite limit for open files */
3832 lim.rlim_cur = 65535;
3833 lim.rlim_max = 65535;
3834
3835 ret = setrlimit(RLIMIT_NOFILE, &lim);
3836 if (ret < 0) {
3837 PERROR("failed to set open files limit");
3838 }
3839 }
3840
3841 /*
3842 * main
3843 */
3844 int main(int argc, char **argv)
3845 {
3846 int ret = 0;
3847 void *status;
3848 const char *home_path, *env_app_timeout;
3849
3850 init_kernel_workarounds();
3851
3852 rcu_register_thread();
3853
3854 setup_consumerd_path();
3855
3856 /* Parse arguments */
3857 progname = argv[0];
3858 if ((ret = parse_args(argc, argv) < 0)) {
3859 goto error;
3860 }
3861
3862 /* Daemonize */
3863 if (opt_daemon) {
3864 int i;
3865
3866 /*
3867 * fork
3868 * child: setsid, close FD 0, 1, 2, chdir /
3869 * parent: exit (if fork is successful)
3870 */
3871 ret = daemon(0, 0);
3872 if (ret < 0) {
3873 PERROR("daemon");
3874 goto error;
3875 }
3876 /*
3877 * We are in the child. Make sure all other file
3878 * descriptors are closed, in case we are called with
3879 * more opened file descriptors than the standard ones.
3880 */
3881 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3882 (void) close(i);
3883 }
3884 }
3885
3886 /* Create thread quit pipe */
3887 if ((ret = init_thread_quit_pipe()) < 0) {
3888 goto error;
3889 }
3890
3891 /* Check if daemon is UID = 0 */
3892 is_root = !getuid();
3893
3894 if (is_root) {
3895 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3896
3897 /* Create global run dir with root access */
3898 ret = create_lttng_rundir(rundir);
3899 if (ret < 0) {
3900 goto error;
3901 }
3902
3903 if (strlen(apps_unix_sock_path) == 0) {
3904 snprintf(apps_unix_sock_path, PATH_MAX,
3905 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3906 }
3907
3908 if (strlen(client_unix_sock_path) == 0) {
3909 snprintf(client_unix_sock_path, PATH_MAX,
3910 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3911 }
3912
3913 /* Set global SHM for ust */
3914 if (strlen(wait_shm_path) == 0) {
3915 snprintf(wait_shm_path, PATH_MAX,
3916 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3917 }
3918
3919 if (strlen(health_unix_sock_path) == 0) {
3920 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3921 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3922 }
3923
3924 /* Setup kernel consumerd path */
3925 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3926 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3927 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3928 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3929
3930 DBG2("Kernel consumer err path: %s",
3931 kconsumer_data.err_unix_sock_path);
3932 DBG2("Kernel consumer cmd path: %s",
3933 kconsumer_data.cmd_unix_sock_path);
3934 } else {
3935 home_path = get_home_dir();
3936 if (home_path == NULL) {
3937 /* TODO: Add --socket PATH option */
3938 ERR("Can't get HOME directory for sockets creation.");
3939 ret = -EPERM;
3940 goto error;
3941 }
3942
3943 /*
3944 * Create rundir from home path. This will create something like
3945 * $HOME/.lttng
3946 */
3947 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3948 if (ret < 0) {
3949 ret = -ENOMEM;
3950 goto error;
3951 }
3952
3953 ret = create_lttng_rundir(rundir);
3954 if (ret < 0) {
3955 goto error;
3956 }
3957
3958 if (strlen(apps_unix_sock_path) == 0) {
3959 snprintf(apps_unix_sock_path, PATH_MAX,
3960 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3961 }
3962
3963 /* Set the cli tool unix socket path */
3964 if (strlen(client_unix_sock_path) == 0) {
3965 snprintf(client_unix_sock_path, PATH_MAX,
3966 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3967 }
3968
3969 /* Set global SHM for ust */
3970 if (strlen(wait_shm_path) == 0) {
3971 snprintf(wait_shm_path, PATH_MAX,
3972 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3973 }
3974
3975 /* Set health check Unix path */
3976 if (strlen(health_unix_sock_path) == 0) {
3977 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3978 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3979 }
3980 }
3981
3982 /* Set consumer initial state */
3983 kernel_consumerd_state = CONSUMER_STOPPED;
3984 ust_consumerd_state = CONSUMER_STOPPED;
3985
3986 DBG("Client socket path %s", client_unix_sock_path);
3987 DBG("Application socket path %s", apps_unix_sock_path);
3988 DBG("LTTng run directory path: %s", rundir);
3989
3990 /* 32 bits consumerd path setup */
3991 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
3992 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
3993 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
3994 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
3995
3996 DBG2("UST consumer 32 bits err path: %s",
3997 ustconsumer32_data.err_unix_sock_path);
3998 DBG2("UST consumer 32 bits cmd path: %s",
3999 ustconsumer32_data.cmd_unix_sock_path);
4000
4001 /* 64 bits consumerd path setup */
4002 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4003 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4004 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4005 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4006
4007 DBG2("UST consumer 64 bits err path: %s",
4008 ustconsumer64_data.err_unix_sock_path);
4009 DBG2("UST consumer 64 bits cmd path: %s",
4010 ustconsumer64_data.cmd_unix_sock_path);
4011
4012 /*
4013 * See if daemon already exist.
4014 */
4015 if ((ret = check_existing_daemon()) < 0) {
4016 ERR("Already running daemon.\n");
4017 /*
4018 * We do not goto exit because we must not cleanup()
4019 * because a daemon is already running.
4020 */
4021 goto error;
4022 }
4023
4024 /*
4025 * Init UST app hash table. Alloc hash table before this point since
4026 * cleanup() can get called after that point.
4027 */
4028 ust_app_ht_alloc();
4029
4030 /* After this point, we can safely call cleanup() with "goto exit" */
4031
4032 /*
4033 * These actions must be executed as root. We do that *after* setting up
4034 * the sockets path because we MUST make the check for another daemon using
4035 * those paths *before* trying to set the kernel consumer sockets and init
4036 * kernel tracer.
4037 */
4038 if (is_root) {
4039 ret = set_consumer_sockets(&kconsumer_data, rundir);
4040 if (ret < 0) {
4041 goto exit;
4042 }
4043
4044 /* Setup kernel tracer */
4045 if (!opt_no_kernel) {
4046 init_kernel_tracer();
4047 }
4048
4049 /* Set ulimit for open files */
4050 set_ulimit();
4051 }
4052 /* init lttng_fd tracking must be done after set_ulimit. */
4053 lttng_fd_init();
4054
4055 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4056 if (ret < 0) {
4057 goto exit;
4058 }
4059
4060 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4061 if (ret < 0) {
4062 goto exit;
4063 }
4064
4065 if ((ret = set_signal_handler()) < 0) {
4066 goto exit;
4067 }
4068
4069 /* Setup the needed unix socket */
4070 if ((ret = init_daemon_socket()) < 0) {
4071 goto exit;
4072 }
4073
4074 /* Set credentials to socket */
4075 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4076 goto exit;
4077 }
4078
4079 /* Get parent pid if -S, --sig-parent is specified. */
4080 if (opt_sig_parent) {
4081 ppid = getppid();
4082 }
4083
4084 /* Setup the kernel pipe for waking up the kernel thread */
4085 if (is_root && !opt_no_kernel) {
4086 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4087 goto exit;
4088 }
4089 }
4090
4091 /* Setup the thread apps communication pipe. */
4092 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4093 goto exit;
4094 }
4095
4096 /* Init UST command queue. */
4097 cds_wfq_init(&ust_cmd_queue.queue);
4098
4099 /*
4100 * Get session list pointer. This pointer MUST NOT be free(). This list is
4101 * statically declared in session.c
4102 */
4103 session_list_ptr = session_get_list();
4104
4105 /* Set up max poll set size */
4106 lttng_poll_set_max_size();
4107
4108 cmd_init();
4109
4110 /* Init all health thread counters. */
4111 health_init(&health_thread_cmd);
4112 health_init(&health_thread_kernel);
4113 health_init(&health_thread_app_manage);
4114 health_init(&health_thread_app_reg);
4115
4116 /*
4117 * Init health counters of the consumer thread. We do a quick hack here to
4118 * the state of the consumer health is fine even if the thread is not
4119 * started. Once the thread starts, the health state is updated with a poll
4120 * value to set a health code path. This is simply to ease our life and has
4121 * no cost what so ever.
4122 */
4123 health_init(&kconsumer_data.health);
4124 health_poll_update(&kconsumer_data.health);
4125 health_init(&ustconsumer32_data.health);
4126 health_poll_update(&ustconsumer32_data.health);
4127 health_init(&ustconsumer64_data.health);
4128 health_poll_update(&ustconsumer64_data.health);
4129
4130 /* Check for the application socket timeout env variable. */
4131 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4132 if (env_app_timeout) {
4133 app_socket_timeout = atoi(env_app_timeout);
4134 } else {
4135 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4136 }
4137
4138 /* Create thread to manage the client socket */
4139 ret = pthread_create(&health_thread, NULL,
4140 thread_manage_health, (void *) NULL);
4141 if (ret != 0) {
4142 PERROR("pthread_create health");
4143 goto exit_health;
4144 }
4145
4146 /* Create thread to manage the client socket */
4147 ret = pthread_create(&client_thread, NULL,
4148 thread_manage_clients, (void *) NULL);
4149 if (ret != 0) {
4150 PERROR("pthread_create clients");
4151 goto exit_client;
4152 }
4153
4154 /* Create thread to dispatch registration */
4155 ret = pthread_create(&dispatch_thread, NULL,
4156 thread_dispatch_ust_registration, (void *) NULL);
4157 if (ret != 0) {
4158 PERROR("pthread_create dispatch");
4159 goto exit_dispatch;
4160 }
4161
4162 /* Create thread to manage application registration. */
4163 ret = pthread_create(&reg_apps_thread, NULL,
4164 thread_registration_apps, (void *) NULL);
4165 if (ret != 0) {
4166 PERROR("pthread_create registration");
4167 goto exit_reg_apps;
4168 }
4169
4170 /* Create thread to manage application socket */
4171 ret = pthread_create(&apps_thread, NULL,
4172 thread_manage_apps, (void *) NULL);
4173 if (ret != 0) {
4174 PERROR("pthread_create apps");
4175 goto exit_apps;
4176 }
4177
4178 /* Don't start this thread if kernel tracing is not requested nor root */
4179 if (is_root && !opt_no_kernel) {
4180 /* Create kernel thread to manage kernel event */
4181 ret = pthread_create(&kernel_thread, NULL,
4182 thread_manage_kernel, (void *) NULL);
4183 if (ret != 0) {
4184 PERROR("pthread_create kernel");
4185 goto exit_kernel;
4186 }
4187
4188 ret = pthread_join(kernel_thread, &status);
4189 if (ret != 0) {
4190 PERROR("pthread_join");
4191 goto error; /* join error, exit without cleanup */
4192 }
4193 }
4194
4195 exit_kernel:
4196 ret = pthread_join(apps_thread, &status);
4197 if (ret != 0) {
4198 PERROR("pthread_join");
4199 goto error; /* join error, exit without cleanup */
4200 }
4201
4202 exit_apps:
4203 ret = pthread_join(reg_apps_thread, &status);
4204 if (ret != 0) {
4205 PERROR("pthread_join");
4206 goto error; /* join error, exit without cleanup */
4207 }
4208
4209 exit_reg_apps:
4210 ret = pthread_join(dispatch_thread, &status);
4211 if (ret != 0) {
4212 PERROR("pthread_join");
4213 goto error; /* join error, exit without cleanup */
4214 }
4215
4216 exit_dispatch:
4217 ret = pthread_join(client_thread, &status);
4218 if (ret != 0) {
4219 PERROR("pthread_join");
4220 goto error; /* join error, exit without cleanup */
4221 }
4222
4223 ret = join_consumer_thread(&kconsumer_data);
4224 if (ret != 0) {
4225 PERROR("join_consumer");
4226 goto error; /* join error, exit without cleanup */
4227 }
4228
4229 ret = join_consumer_thread(&ustconsumer32_data);
4230 if (ret != 0) {
4231 PERROR("join_consumer ust32");
4232 goto error; /* join error, exit without cleanup */
4233 }
4234
4235 ret = join_consumer_thread(&ustconsumer64_data);
4236 if (ret != 0) {
4237 PERROR("join_consumer ust64");
4238 goto error; /* join error, exit without cleanup */
4239 }
4240
4241 exit_client:
4242 ret = pthread_join(health_thread, &status);
4243 if (ret != 0) {
4244 PERROR("pthread_join health thread");
4245 goto error; /* join error, exit without cleanup */
4246 }
4247
4248 exit_health:
4249 exit:
4250 /*
4251 * cleanup() is called when no other thread is running.
4252 */
4253 rcu_thread_online();
4254 cleanup();
4255 rcu_thread_offline();
4256 rcu_unregister_thread();
4257 if (!ret) {
4258 exit(EXIT_SUCCESS);
4259 }
4260 error:
4261 exit(EXIT_FAILURE);
4262 }
This page took 0.209322 seconds and 6 git commands to generate.