Fix: add missing rcu read side lock
[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 rcu_read_lock();
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 rcu_read_unlock();
641 goto error;
642 }
643 }
644 rcu_read_unlock();
645 }
646 goto error;
647 }
648 }
649 session_unlock(session);
650 }
651 session_unlock_list();
652 return ret;
653
654 error:
655 session_unlock(session);
656 session_unlock_list();
657 return ret;
658 }
659
660 /*
661 * For each tracing session, update newly registered apps.
662 */
663 static void update_ust_app(int app_sock)
664 {
665 struct ltt_session *sess, *stmp;
666
667 session_lock_list();
668
669 /* For all tracing session(s) */
670 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
671 session_lock(sess);
672 if (sess->ust_session) {
673 ust_app_global_update(sess->ust_session, app_sock);
674 }
675 session_unlock(sess);
676 }
677
678 session_unlock_list();
679 }
680
681 /*
682 * This thread manage event coming from the kernel.
683 *
684 * Features supported in this thread:
685 * -) CPU Hotplug
686 */
687 static void *thread_manage_kernel(void *data)
688 {
689 int ret, i, pollfd, update_poll_flag = 1, err = -1;
690 uint32_t revents, nb_fd;
691 char tmp;
692 struct lttng_poll_event events;
693
694 DBG("[thread] Thread manage kernel started");
695
696 /*
697 * This first step of the while is to clean this structure which could free
698 * non NULL pointers so zero it before the loop.
699 */
700 memset(&events, 0, sizeof(events));
701
702 if (testpoint(thread_manage_kernel)) {
703 goto error_testpoint;
704 }
705
706 health_code_update(&health_thread_kernel);
707
708 if (testpoint(thread_manage_kernel_before_loop)) {
709 goto error_testpoint;
710 }
711
712 while (1) {
713 health_code_update(&health_thread_kernel);
714
715 if (update_poll_flag == 1) {
716 /* Clean events object. We are about to populate it again. */
717 lttng_poll_clean(&events);
718
719 ret = create_thread_poll_set(&events, 2);
720 if (ret < 0) {
721 goto error_poll_create;
722 }
723
724 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
725 if (ret < 0) {
726 goto error;
727 }
728
729 /* This will add the available kernel channel if any. */
730 ret = update_kernel_poll(&events);
731 if (ret < 0) {
732 goto error;
733 }
734 update_poll_flag = 0;
735 }
736
737 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
738
739 /* Poll infinite value of time */
740 restart:
741 health_poll_update(&health_thread_kernel);
742 ret = lttng_poll_wait(&events, -1);
743 health_poll_update(&health_thread_kernel);
744 if (ret < 0) {
745 /*
746 * Restart interrupted system call.
747 */
748 if (errno == EINTR) {
749 goto restart;
750 }
751 goto error;
752 } else if (ret == 0) {
753 /* Should not happen since timeout is infinite */
754 ERR("Return value of poll is 0 with an infinite timeout.\n"
755 "This should not have happened! Continuing...");
756 continue;
757 }
758
759 nb_fd = ret;
760
761 for (i = 0; i < nb_fd; i++) {
762 /* Fetch once the poll data */
763 revents = LTTNG_POLL_GETEV(&events, i);
764 pollfd = LTTNG_POLL_GETFD(&events, i);
765
766 health_code_update(&health_thread_kernel);
767
768 /* Thread quit pipe has been closed. Killing thread. */
769 ret = check_thread_quit_pipe(pollfd, revents);
770 if (ret) {
771 err = 0;
772 goto exit;
773 }
774
775 /* Check for data on kernel pipe */
776 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
777 do {
778 ret = read(kernel_poll_pipe[0], &tmp, 1);
779 } while (ret < 0 && errno == EINTR);
780 /*
781 * Ret value is useless here, if this pipe gets any actions an
782 * update is required anyway.
783 */
784 update_poll_flag = 1;
785 continue;
786 } else {
787 /*
788 * New CPU detected by the kernel. Adding kernel stream to
789 * kernel session and updating the kernel consumer
790 */
791 if (revents & LPOLLIN) {
792 ret = update_kernel_stream(&kconsumer_data, pollfd);
793 if (ret < 0) {
794 continue;
795 }
796 break;
797 /*
798 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
799 * and unregister kernel stream at this point.
800 */
801 }
802 }
803 }
804 }
805
806 exit:
807 error:
808 lttng_poll_clean(&events);
809 error_poll_create:
810 error_testpoint:
811 utils_close_pipe(kernel_poll_pipe);
812 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
813 if (err) {
814 health_error(&health_thread_kernel);
815 ERR("Health error occurred in %s", __func__);
816 WARN("Kernel thread died unexpectedly. "
817 "Kernel tracing can continue but CPU hotplug is disabled.");
818 }
819 health_exit(&health_thread_kernel);
820 DBG("Kernel thread dying");
821 return NULL;
822 }
823
824 /*
825 * Signal pthread condition of the consumer data that the thread.
826 */
827 static void signal_consumer_condition(struct consumer_data *data, int state)
828 {
829 pthread_mutex_lock(&data->cond_mutex);
830
831 /*
832 * The state is set before signaling. It can be any value, it's the waiter
833 * job to correctly interpret this condition variable associated to the
834 * consumer pthread_cond.
835 *
836 * A value of 0 means that the corresponding thread of the consumer data
837 * was not started. 1 indicates that the thread has started and is ready
838 * for action. A negative value means that there was an error during the
839 * thread bootstrap.
840 */
841 data->consumer_thread_is_ready = state;
842 (void) pthread_cond_signal(&data->cond);
843
844 pthread_mutex_unlock(&data->cond_mutex);
845 }
846
847 /*
848 * This thread manage the consumer error sent back to the session daemon.
849 */
850 static void *thread_manage_consumer(void *data)
851 {
852 int sock = -1, i, ret, pollfd, err = -1;
853 uint32_t revents, nb_fd;
854 enum lttcomm_return_code code;
855 struct lttng_poll_event events;
856 struct consumer_data *consumer_data = data;
857
858 DBG("[thread] Manage consumer started");
859
860 /*
861 * Since the consumer thread can be spawned at any moment in time, we init
862 * the health to a poll status (1, which is a valid health over time).
863 * When the thread starts, we update here the health to a "code" path being
864 * an even value so this thread, when reaching a poll wait, does not
865 * trigger an error with an even value.
866 *
867 * Here is the use case we avoid.
868 *
869 * +1: the first poll update during initialization (main())
870 * +2 * x: multiple code update once in this thread.
871 * +1: poll wait in this thread (being a good health state).
872 * == even number which after the wait period shows as a bad health.
873 *
874 * In a nutshell, the following poll update to the health state brings back
875 * the state to an even value meaning a code path.
876 */
877 health_poll_update(&consumer_data->health);
878
879 /*
880 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
881 * Nothing more will be added to this poll set.
882 */
883 ret = create_thread_poll_set(&events, 2);
884 if (ret < 0) {
885 goto error_poll;
886 }
887
888 /*
889 * The error socket here is already in a listening state which was done
890 * just before spawning this thread to avoid a race between the consumer
891 * daemon exec trying to connect and the listen() call.
892 */
893 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
894 if (ret < 0) {
895 goto error;
896 }
897
898 health_code_update(&consumer_data->health);
899
900 /* Inifinite blocking call, waiting for transmission */
901 restart:
902 health_poll_update(&consumer_data->health);
903
904 if (testpoint(thread_manage_consumer)) {
905 goto error;
906 }
907
908 ret = lttng_poll_wait(&events, -1);
909 health_poll_update(&consumer_data->health);
910 if (ret < 0) {
911 /*
912 * Restart interrupted system call.
913 */
914 if (errno == EINTR) {
915 goto restart;
916 }
917 goto error;
918 }
919
920 nb_fd = ret;
921
922 for (i = 0; i < nb_fd; i++) {
923 /* Fetch once the poll data */
924 revents = LTTNG_POLL_GETEV(&events, i);
925 pollfd = LTTNG_POLL_GETFD(&events, i);
926
927 health_code_update(&consumer_data->health);
928
929 /* Thread quit pipe has been closed. Killing thread. */
930 ret = check_thread_quit_pipe(pollfd, revents);
931 if (ret) {
932 err = 0;
933 goto exit;
934 }
935
936 /* Event on the registration socket */
937 if (pollfd == consumer_data->err_sock) {
938 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
939 ERR("consumer err socket poll error");
940 goto error;
941 }
942 }
943 }
944
945 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
946 if (sock < 0) {
947 goto error;
948 }
949
950 /*
951 * Set the CLOEXEC flag. Return code is useless because either way, the
952 * show must go on.
953 */
954 (void) utils_set_fd_cloexec(sock);
955
956 health_code_update(&consumer_data->health);
957
958 DBG2("Receiving code from consumer err_sock");
959
960 /* Getting status code from kconsumerd */
961 ret = lttcomm_recv_unix_sock(sock, &code,
962 sizeof(enum lttcomm_return_code));
963 if (ret <= 0) {
964 goto error;
965 }
966
967 health_code_update(&consumer_data->health);
968
969 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
970 consumer_data->cmd_sock =
971 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
972 if (consumer_data->cmd_sock < 0) {
973 /* On error, signal condition and quit. */
974 signal_consumer_condition(consumer_data, -1);
975 PERROR("consumer connect");
976 goto error;
977 }
978 signal_consumer_condition(consumer_data, 1);
979 DBG("Consumer command socket ready");
980 } else {
981 ERR("consumer error when waiting for SOCK_READY : %s",
982 lttcomm_get_readable_code(-code));
983 goto error;
984 }
985
986 /* Remove the kconsumerd error sock since we've established a connexion */
987 ret = lttng_poll_del(&events, consumer_data->err_sock);
988 if (ret < 0) {
989 goto error;
990 }
991
992 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
993 if (ret < 0) {
994 goto error;
995 }
996
997 health_code_update(&consumer_data->health);
998
999 /* Inifinite blocking call, waiting for transmission */
1000 restart_poll:
1001 health_poll_update(&consumer_data->health);
1002 ret = lttng_poll_wait(&events, -1);
1003 health_poll_update(&consumer_data->health);
1004 if (ret < 0) {
1005 /*
1006 * Restart interrupted system call.
1007 */
1008 if (errno == EINTR) {
1009 goto restart_poll;
1010 }
1011 goto error;
1012 }
1013
1014 nb_fd = ret;
1015
1016 for (i = 0; i < nb_fd; i++) {
1017 /* Fetch once the poll data */
1018 revents = LTTNG_POLL_GETEV(&events, i);
1019 pollfd = LTTNG_POLL_GETFD(&events, i);
1020
1021 health_code_update(&consumer_data->health);
1022
1023 /* Thread quit pipe has been closed. Killing thread. */
1024 ret = check_thread_quit_pipe(pollfd, revents);
1025 if (ret) {
1026 err = 0;
1027 goto exit;
1028 }
1029
1030 /* Event on the kconsumerd socket */
1031 if (pollfd == sock) {
1032 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1033 ERR("consumer err socket second poll error");
1034 goto error;
1035 }
1036 }
1037 }
1038
1039 health_code_update(&consumer_data->health);
1040
1041 /* Wait for any kconsumerd error */
1042 ret = lttcomm_recv_unix_sock(sock, &code,
1043 sizeof(enum lttcomm_return_code));
1044 if (ret <= 0) {
1045 ERR("consumer closed the command socket");
1046 goto error;
1047 }
1048
1049 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1050
1051 exit:
1052 error:
1053 /* Immediately set the consumerd state to stopped */
1054 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1055 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1056 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1057 consumer_data->type == LTTNG_CONSUMER32_UST) {
1058 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1059 } else {
1060 /* Code flow error... */
1061 assert(0);
1062 }
1063
1064 if (consumer_data->err_sock >= 0) {
1065 ret = close(consumer_data->err_sock);
1066 if (ret) {
1067 PERROR("close");
1068 }
1069 }
1070 if (consumer_data->cmd_sock >= 0) {
1071 ret = close(consumer_data->cmd_sock);
1072 if (ret) {
1073 PERROR("close");
1074 }
1075 }
1076 if (sock >= 0) {
1077 ret = close(sock);
1078 if (ret) {
1079 PERROR("close");
1080 }
1081 }
1082
1083 unlink(consumer_data->err_unix_sock_path);
1084 unlink(consumer_data->cmd_unix_sock_path);
1085 consumer_data->pid = 0;
1086
1087 lttng_poll_clean(&events);
1088 error_poll:
1089 if (err) {
1090 health_error(&consumer_data->health);
1091 ERR("Health error occurred in %s", __func__);
1092 }
1093 health_exit(&consumer_data->health);
1094 DBG("consumer thread cleanup completed");
1095
1096 return NULL;
1097 }
1098
1099 /*
1100 * This thread manage application communication.
1101 */
1102 static void *thread_manage_apps(void *data)
1103 {
1104 int i, ret, pollfd, err = -1;
1105 uint32_t revents, nb_fd;
1106 struct ust_command ust_cmd;
1107 struct lttng_poll_event events;
1108
1109 DBG("[thread] Manage application started");
1110
1111 rcu_register_thread();
1112 rcu_thread_online();
1113
1114 if (testpoint(thread_manage_apps)) {
1115 goto error_testpoint;
1116 }
1117
1118 health_code_update(&health_thread_app_manage);
1119
1120 ret = create_thread_poll_set(&events, 2);
1121 if (ret < 0) {
1122 goto error_poll_create;
1123 }
1124
1125 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1126 if (ret < 0) {
1127 goto error;
1128 }
1129
1130 if (testpoint(thread_manage_apps_before_loop)) {
1131 goto error;
1132 }
1133
1134 health_code_update(&health_thread_app_manage);
1135
1136 while (1) {
1137 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1138
1139 /* Inifinite blocking call, waiting for transmission */
1140 restart:
1141 health_poll_update(&health_thread_app_manage);
1142 ret = lttng_poll_wait(&events, -1);
1143 health_poll_update(&health_thread_app_manage);
1144 if (ret < 0) {
1145 /*
1146 * Restart interrupted system call.
1147 */
1148 if (errno == EINTR) {
1149 goto restart;
1150 }
1151 goto error;
1152 }
1153
1154 nb_fd = ret;
1155
1156 for (i = 0; i < nb_fd; i++) {
1157 /* Fetch once the poll data */
1158 revents = LTTNG_POLL_GETEV(&events, i);
1159 pollfd = LTTNG_POLL_GETFD(&events, i);
1160
1161 health_code_update(&health_thread_app_manage);
1162
1163 /* Thread quit pipe has been closed. Killing thread. */
1164 ret = check_thread_quit_pipe(pollfd, revents);
1165 if (ret) {
1166 err = 0;
1167 goto exit;
1168 }
1169
1170 /* Inspect the apps cmd pipe */
1171 if (pollfd == apps_cmd_pipe[0]) {
1172 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1173 ERR("Apps command pipe error");
1174 goto error;
1175 } else if (revents & LPOLLIN) {
1176 /* Empty pipe */
1177 do {
1178 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1179 } while (ret < 0 && errno == EINTR);
1180 if (ret < 0 || ret < sizeof(ust_cmd)) {
1181 PERROR("read apps cmd pipe");
1182 goto error;
1183 }
1184
1185 health_code_update(&health_thread_app_manage);
1186
1187 /* Register applicaton to the session daemon */
1188 ret = ust_app_register(&ust_cmd.reg_msg,
1189 ust_cmd.sock);
1190 if (ret == -ENOMEM) {
1191 goto error;
1192 } else if (ret < 0) {
1193 break;
1194 }
1195
1196 health_code_update(&health_thread_app_manage);
1197
1198 /*
1199 * Validate UST version compatibility.
1200 */
1201 ret = ust_app_validate_version(ust_cmd.sock);
1202 if (ret >= 0) {
1203 /*
1204 * Add channel(s) and event(s) to newly registered apps
1205 * from lttng global UST domain.
1206 */
1207 update_ust_app(ust_cmd.sock);
1208 }
1209
1210 health_code_update(&health_thread_app_manage);
1211
1212 ret = ust_app_register_done(ust_cmd.sock);
1213 if (ret < 0) {
1214 /*
1215 * If the registration is not possible, we simply
1216 * unregister the apps and continue
1217 */
1218 ust_app_unregister(ust_cmd.sock);
1219 } else {
1220 /*
1221 * We only monitor the error events of the socket. This
1222 * thread does not handle any incoming data from UST
1223 * (POLLIN).
1224 */
1225 ret = lttng_poll_add(&events, ust_cmd.sock,
1226 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1227 if (ret < 0) {
1228 goto error;
1229 }
1230
1231 /* Set socket timeout for both receiving and ending */
1232 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1233 app_socket_timeout);
1234 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1235 app_socket_timeout);
1236
1237 DBG("Apps with sock %d added to poll set",
1238 ust_cmd.sock);
1239 }
1240
1241 health_code_update(&health_thread_app_manage);
1242
1243 break;
1244 }
1245 } else {
1246 /*
1247 * At this point, we know that a registered application made
1248 * the event at poll_wait.
1249 */
1250 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1251 /* Removing from the poll set */
1252 ret = lttng_poll_del(&events, pollfd);
1253 if (ret < 0) {
1254 goto error;
1255 }
1256
1257 /* Socket closed on remote end. */
1258 ust_app_unregister(pollfd);
1259 break;
1260 }
1261 }
1262
1263 health_code_update(&health_thread_app_manage);
1264 }
1265 }
1266
1267 exit:
1268 error:
1269 lttng_poll_clean(&events);
1270 error_poll_create:
1271 error_testpoint:
1272 utils_close_pipe(apps_cmd_pipe);
1273 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1274
1275 /*
1276 * We don't clean the UST app hash table here since already registered
1277 * applications can still be controlled so let them be until the session
1278 * daemon dies or the applications stop.
1279 */
1280
1281 if (err) {
1282 health_error(&health_thread_app_manage);
1283 ERR("Health error occurred in %s", __func__);
1284 }
1285 health_exit(&health_thread_app_manage);
1286 DBG("Application communication apps thread cleanup complete");
1287 rcu_thread_offline();
1288 rcu_unregister_thread();
1289 return NULL;
1290 }
1291
1292 /*
1293 * Dispatch request from the registration threads to the application
1294 * communication thread.
1295 */
1296 static void *thread_dispatch_ust_registration(void *data)
1297 {
1298 int ret;
1299 struct cds_wfq_node *node;
1300 struct ust_command *ust_cmd = NULL;
1301
1302 DBG("[thread] Dispatch UST command started");
1303
1304 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1305 /* Atomically prepare the queue futex */
1306 futex_nto1_prepare(&ust_cmd_queue.futex);
1307
1308 do {
1309 /* Dequeue command for registration */
1310 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1311 if (node == NULL) {
1312 DBG("Woken up but nothing in the UST command queue");
1313 /* Continue thread execution */
1314 break;
1315 }
1316
1317 ust_cmd = caa_container_of(node, struct ust_command, node);
1318
1319 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1320 " gid:%d sock:%d name:%s (version %d.%d)",
1321 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1322 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1323 ust_cmd->sock, ust_cmd->reg_msg.name,
1324 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1325 /*
1326 * Inform apps thread of the new application registration. This
1327 * call is blocking so we can be assured that the data will be read
1328 * at some point in time or wait to the end of the world :)
1329 */
1330 if (apps_cmd_pipe[1] >= 0) {
1331 do {
1332 ret = write(apps_cmd_pipe[1], ust_cmd,
1333 sizeof(struct ust_command));
1334 } while (ret < 0 && errno == EINTR);
1335 if (ret < 0 || ret != sizeof(struct ust_command)) {
1336 PERROR("write apps cmd pipe");
1337 if (errno == EBADF) {
1338 /*
1339 * We can't inform the application thread to process
1340 * registration. We will exit or else application
1341 * registration will not occur and tracing will never
1342 * start.
1343 */
1344 goto error;
1345 }
1346 }
1347 } else {
1348 /* Application manager thread is not available. */
1349 ret = close(ust_cmd->sock);
1350 if (ret < 0) {
1351 PERROR("close ust_cmd sock");
1352 }
1353 }
1354 free(ust_cmd);
1355 } while (node != NULL);
1356
1357 /* Futex wait on queue. Blocking call on futex() */
1358 futex_nto1_wait(&ust_cmd_queue.futex);
1359 }
1360
1361 error:
1362 DBG("Dispatch thread dying");
1363 return NULL;
1364 }
1365
1366 /*
1367 * This thread manage application registration.
1368 */
1369 static void *thread_registration_apps(void *data)
1370 {
1371 int sock = -1, i, ret, pollfd, err = -1;
1372 uint32_t revents, nb_fd;
1373 struct lttng_poll_event events;
1374 /*
1375 * Get allocated in this thread, enqueued to a global queue, dequeued and
1376 * freed in the manage apps thread.
1377 */
1378 struct ust_command *ust_cmd = NULL;
1379
1380 DBG("[thread] Manage application registration started");
1381
1382 if (testpoint(thread_registration_apps)) {
1383 goto error_testpoint;
1384 }
1385
1386 ret = lttcomm_listen_unix_sock(apps_sock);
1387 if (ret < 0) {
1388 goto error_listen;
1389 }
1390
1391 /*
1392 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1393 * more will be added to this poll set.
1394 */
1395 ret = create_thread_poll_set(&events, 2);
1396 if (ret < 0) {
1397 goto error_create_poll;
1398 }
1399
1400 /* Add the application registration socket */
1401 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1402 if (ret < 0) {
1403 goto error_poll_add;
1404 }
1405
1406 /* Notify all applications to register */
1407 ret = notify_ust_apps(1);
1408 if (ret < 0) {
1409 ERR("Failed to notify applications or create the wait shared memory.\n"
1410 "Execution continues but there might be problem for already\n"
1411 "running applications that wishes to register.");
1412 }
1413
1414 while (1) {
1415 DBG("Accepting application registration");
1416
1417 /* Inifinite blocking call, waiting for transmission */
1418 restart:
1419 health_poll_update(&health_thread_app_reg);
1420 ret = lttng_poll_wait(&events, -1);
1421 health_poll_update(&health_thread_app_reg);
1422 if (ret < 0) {
1423 /*
1424 * Restart interrupted system call.
1425 */
1426 if (errno == EINTR) {
1427 goto restart;
1428 }
1429 goto error;
1430 }
1431
1432 nb_fd = ret;
1433
1434 for (i = 0; i < nb_fd; i++) {
1435 health_code_update(&health_thread_app_reg);
1436
1437 /* Fetch once the poll data */
1438 revents = LTTNG_POLL_GETEV(&events, i);
1439 pollfd = LTTNG_POLL_GETFD(&events, i);
1440
1441 /* Thread quit pipe has been closed. Killing thread. */
1442 ret = check_thread_quit_pipe(pollfd, revents);
1443 if (ret) {
1444 err = 0;
1445 goto exit;
1446 }
1447
1448 /* Event on the registration socket */
1449 if (pollfd == apps_sock) {
1450 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1451 ERR("Register apps socket poll error");
1452 goto error;
1453 } else if (revents & LPOLLIN) {
1454 sock = lttcomm_accept_unix_sock(apps_sock);
1455 if (sock < 0) {
1456 goto error;
1457 }
1458
1459 /*
1460 * Set the CLOEXEC flag. Return code is useless because
1461 * either way, the show must go on.
1462 */
1463 (void) utils_set_fd_cloexec(sock);
1464
1465 /* Create UST registration command for enqueuing */
1466 ust_cmd = zmalloc(sizeof(struct ust_command));
1467 if (ust_cmd == NULL) {
1468 PERROR("ust command zmalloc");
1469 goto error;
1470 }
1471
1472 /*
1473 * Using message-based transmissions to ensure we don't
1474 * have to deal with partially received messages.
1475 */
1476 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1477 if (ret < 0) {
1478 ERR("Exhausted file descriptors allowed for applications.");
1479 free(ust_cmd);
1480 ret = close(sock);
1481 if (ret) {
1482 PERROR("close");
1483 }
1484 sock = -1;
1485 continue;
1486 }
1487 health_code_update(&health_thread_app_reg);
1488 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1489 sizeof(struct ust_register_msg));
1490 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1491 if (ret < 0) {
1492 PERROR("lttcomm_recv_unix_sock register apps");
1493 } else {
1494 ERR("Wrong size received on apps register");
1495 }
1496 free(ust_cmd);
1497 ret = close(sock);
1498 if (ret) {
1499 PERROR("close");
1500 }
1501 lttng_fd_put(LTTNG_FD_APPS, 1);
1502 sock = -1;
1503 continue;
1504 }
1505 health_code_update(&health_thread_app_reg);
1506
1507 ust_cmd->sock = sock;
1508 sock = -1;
1509
1510 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1511 " gid:%d sock:%d name:%s (version %d.%d)",
1512 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1513 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1514 ust_cmd->sock, ust_cmd->reg_msg.name,
1515 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1516
1517 /*
1518 * Lock free enqueue the registration request. The red pill
1519 * has been taken! This apps will be part of the *system*.
1520 */
1521 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1522
1523 /*
1524 * Wake the registration queue futex. Implicit memory
1525 * barrier with the exchange in cds_wfq_enqueue.
1526 */
1527 futex_nto1_wake(&ust_cmd_queue.futex);
1528 }
1529 }
1530 }
1531 }
1532
1533 exit:
1534 error:
1535 if (err) {
1536 health_error(&health_thread_app_reg);
1537 ERR("Health error occurred in %s", __func__);
1538 }
1539
1540 /* Notify that the registration thread is gone */
1541 notify_ust_apps(0);
1542
1543 if (apps_sock >= 0) {
1544 ret = close(apps_sock);
1545 if (ret) {
1546 PERROR("close");
1547 }
1548 }
1549 if (sock >= 0) {
1550 ret = close(sock);
1551 if (ret) {
1552 PERROR("close");
1553 }
1554 lttng_fd_put(LTTNG_FD_APPS, 1);
1555 }
1556 unlink(apps_unix_sock_path);
1557
1558 error_poll_add:
1559 lttng_poll_clean(&events);
1560 error_listen:
1561 error_create_poll:
1562 error_testpoint:
1563 DBG("UST Registration thread cleanup complete");
1564 health_exit(&health_thread_app_reg);
1565
1566 return NULL;
1567 }
1568
1569 /*
1570 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1571 * exec or it will fails.
1572 */
1573 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1574 {
1575 int ret, clock_ret;
1576 struct timespec timeout;
1577
1578 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1579 consumer_data->consumer_thread_is_ready = 0;
1580
1581 /* Setup pthread condition */
1582 ret = pthread_condattr_init(&consumer_data->condattr);
1583 if (ret != 0) {
1584 errno = ret;
1585 PERROR("pthread_condattr_init consumer data");
1586 goto error;
1587 }
1588
1589 /*
1590 * Set the monotonic clock in order to make sure we DO NOT jump in time
1591 * between the clock_gettime() call and the timedwait call. See bug #324
1592 * for a more details and how we noticed it.
1593 */
1594 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1595 if (ret != 0) {
1596 errno = ret;
1597 PERROR("pthread_condattr_setclock consumer data");
1598 goto error;
1599 }
1600
1601 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1602 if (ret != 0) {
1603 errno = ret;
1604 PERROR("pthread_cond_init consumer data");
1605 goto error;
1606 }
1607
1608 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1609 consumer_data);
1610 if (ret != 0) {
1611 PERROR("pthread_create consumer");
1612 ret = -1;
1613 goto error;
1614 }
1615
1616 /* We are about to wait on a pthread condition */
1617 pthread_mutex_lock(&consumer_data->cond_mutex);
1618
1619 /* Get time for sem_timedwait absolute timeout */
1620 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1621 /*
1622 * Set the timeout for the condition timed wait even if the clock gettime
1623 * call fails since we might loop on that call and we want to avoid to
1624 * increment the timeout too many times.
1625 */
1626 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1627
1628 /*
1629 * The following loop COULD be skipped in some conditions so this is why we
1630 * set ret to 0 in order to make sure at least one round of the loop is
1631 * done.
1632 */
1633 ret = 0;
1634
1635 /*
1636 * Loop until the condition is reached or when a timeout is reached. Note
1637 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1638 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1639 * possible. This loop does not take any chances and works with both of
1640 * them.
1641 */
1642 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1643 if (clock_ret < 0) {
1644 PERROR("clock_gettime spawn consumer");
1645 /* Infinite wait for the consumerd thread to be ready */
1646 ret = pthread_cond_wait(&consumer_data->cond,
1647 &consumer_data->cond_mutex);
1648 } else {
1649 ret = pthread_cond_timedwait(&consumer_data->cond,
1650 &consumer_data->cond_mutex, &timeout);
1651 }
1652 }
1653
1654 /* Release the pthread condition */
1655 pthread_mutex_unlock(&consumer_data->cond_mutex);
1656
1657 if (ret != 0) {
1658 errno = ret;
1659 if (ret == ETIMEDOUT) {
1660 /*
1661 * Call has timed out so we kill the kconsumerd_thread and return
1662 * an error.
1663 */
1664 ERR("Condition timed out. The consumer thread was never ready."
1665 " Killing it");
1666 ret = pthread_cancel(consumer_data->thread);
1667 if (ret < 0) {
1668 PERROR("pthread_cancel consumer thread");
1669 }
1670 } else {
1671 PERROR("pthread_cond_wait failed consumer thread");
1672 }
1673 goto error;
1674 }
1675
1676 pthread_mutex_lock(&consumer_data->pid_mutex);
1677 if (consumer_data->pid == 0) {
1678 ERR("Consumerd did not start");
1679 pthread_mutex_unlock(&consumer_data->pid_mutex);
1680 goto error;
1681 }
1682 pthread_mutex_unlock(&consumer_data->pid_mutex);
1683
1684 return 0;
1685
1686 error:
1687 return ret;
1688 }
1689
1690 /*
1691 * Join consumer thread
1692 */
1693 static int join_consumer_thread(struct consumer_data *consumer_data)
1694 {
1695 void *status;
1696
1697 /* Consumer pid must be a real one. */
1698 if (consumer_data->pid > 0) {
1699 int ret;
1700 ret = kill(consumer_data->pid, SIGTERM);
1701 if (ret) {
1702 ERR("Error killing consumer daemon");
1703 return ret;
1704 }
1705 return pthread_join(consumer_data->thread, &status);
1706 } else {
1707 return 0;
1708 }
1709 }
1710
1711 /*
1712 * Fork and exec a consumer daemon (consumerd).
1713 *
1714 * Return pid if successful else -1.
1715 */
1716 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1717 {
1718 int ret;
1719 pid_t pid;
1720 const char *consumer_to_use;
1721 const char *verbosity;
1722 struct stat st;
1723
1724 DBG("Spawning consumerd");
1725
1726 pid = fork();
1727 if (pid == 0) {
1728 /*
1729 * Exec consumerd.
1730 */
1731 if (opt_verbose_consumer) {
1732 verbosity = "--verbose";
1733 } else {
1734 verbosity = "--quiet";
1735 }
1736 switch (consumer_data->type) {
1737 case LTTNG_CONSUMER_KERNEL:
1738 /*
1739 * Find out which consumerd to execute. We will first try the
1740 * 64-bit path, then the sessiond's installation directory, and
1741 * fallback on the 32-bit one,
1742 */
1743 DBG3("Looking for a kernel consumer at these locations:");
1744 DBG3(" 1) %s", consumerd64_bin);
1745 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1746 DBG3(" 3) %s", consumerd32_bin);
1747 if (stat(consumerd64_bin, &st) == 0) {
1748 DBG3("Found location #1");
1749 consumer_to_use = consumerd64_bin;
1750 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1751 DBG3("Found location #2");
1752 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1753 } else if (stat(consumerd32_bin, &st) == 0) {
1754 DBG3("Found location #3");
1755 consumer_to_use = consumerd32_bin;
1756 } else {
1757 DBG("Could not find any valid consumerd executable");
1758 break;
1759 }
1760 DBG("Using kernel consumer at: %s", consumer_to_use);
1761 execl(consumer_to_use,
1762 "lttng-consumerd", verbosity, "-k",
1763 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1764 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1765 NULL);
1766 break;
1767 case LTTNG_CONSUMER64_UST:
1768 {
1769 char *tmpnew = NULL;
1770
1771 if (consumerd64_libdir[0] != '\0') {
1772 char *tmp;
1773 size_t tmplen;
1774
1775 tmp = getenv("LD_LIBRARY_PATH");
1776 if (!tmp) {
1777 tmp = "";
1778 }
1779 tmplen = strlen("LD_LIBRARY_PATH=")
1780 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1781 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1782 if (!tmpnew) {
1783 ret = -ENOMEM;
1784 goto error;
1785 }
1786 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1787 strcat(tmpnew, consumerd64_libdir);
1788 if (tmp[0] != '\0') {
1789 strcat(tmpnew, ":");
1790 strcat(tmpnew, tmp);
1791 }
1792 ret = putenv(tmpnew);
1793 if (ret) {
1794 ret = -errno;
1795 goto error;
1796 }
1797 }
1798 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1799 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1800 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1801 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1802 NULL);
1803 if (consumerd64_libdir[0] != '\0') {
1804 free(tmpnew);
1805 }
1806 if (ret) {
1807 goto error;
1808 }
1809 break;
1810 }
1811 case LTTNG_CONSUMER32_UST:
1812 {
1813 char *tmpnew = NULL;
1814
1815 if (consumerd32_libdir[0] != '\0') {
1816 char *tmp;
1817 size_t tmplen;
1818
1819 tmp = getenv("LD_LIBRARY_PATH");
1820 if (!tmp) {
1821 tmp = "";
1822 }
1823 tmplen = strlen("LD_LIBRARY_PATH=")
1824 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1825 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1826 if (!tmpnew) {
1827 ret = -ENOMEM;
1828 goto error;
1829 }
1830 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1831 strcat(tmpnew, consumerd32_libdir);
1832 if (tmp[0] != '\0') {
1833 strcat(tmpnew, ":");
1834 strcat(tmpnew, tmp);
1835 }
1836 ret = putenv(tmpnew);
1837 if (ret) {
1838 ret = -errno;
1839 goto error;
1840 }
1841 }
1842 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1843 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1844 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1845 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1846 NULL);
1847 if (consumerd32_libdir[0] != '\0') {
1848 free(tmpnew);
1849 }
1850 if (ret) {
1851 goto error;
1852 }
1853 break;
1854 }
1855 default:
1856 PERROR("unknown consumer type");
1857 exit(EXIT_FAILURE);
1858 }
1859 if (errno != 0) {
1860 PERROR("kernel start consumer exec");
1861 }
1862 exit(EXIT_FAILURE);
1863 } else if (pid > 0) {
1864 ret = pid;
1865 } else {
1866 PERROR("start consumer fork");
1867 ret = -errno;
1868 }
1869 error:
1870 return ret;
1871 }
1872
1873 /*
1874 * Spawn the consumerd daemon and session daemon thread.
1875 */
1876 static int start_consumerd(struct consumer_data *consumer_data)
1877 {
1878 int ret;
1879
1880 /*
1881 * Set the listen() state on the socket since there is a possible race
1882 * between the exec() of the consumer daemon and this call if place in the
1883 * consumer thread. See bug #366 for more details.
1884 */
1885 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1886 if (ret < 0) {
1887 goto error;
1888 }
1889
1890 pthread_mutex_lock(&consumer_data->pid_mutex);
1891 if (consumer_data->pid != 0) {
1892 pthread_mutex_unlock(&consumer_data->pid_mutex);
1893 goto end;
1894 }
1895
1896 ret = spawn_consumerd(consumer_data);
1897 if (ret < 0) {
1898 ERR("Spawning consumerd failed");
1899 pthread_mutex_unlock(&consumer_data->pid_mutex);
1900 goto error;
1901 }
1902
1903 /* Setting up the consumer_data pid */
1904 consumer_data->pid = ret;
1905 DBG2("Consumer pid %d", consumer_data->pid);
1906 pthread_mutex_unlock(&consumer_data->pid_mutex);
1907
1908 DBG2("Spawning consumer control thread");
1909 ret = spawn_consumer_thread(consumer_data);
1910 if (ret < 0) {
1911 ERR("Fatal error spawning consumer control thread");
1912 goto error;
1913 }
1914
1915 end:
1916 return 0;
1917
1918 error:
1919 /* Cleanup already created socket on error. */
1920 if (consumer_data->err_sock >= 0) {
1921 int err;
1922
1923 err = close(consumer_data->err_sock);
1924 if (err < 0) {
1925 PERROR("close consumer data error socket");
1926 }
1927 }
1928 return ret;
1929 }
1930
1931 /*
1932 * Compute health status of each consumer. If one of them is zero (bad
1933 * state), we return 0.
1934 */
1935 static int check_consumer_health(void)
1936 {
1937 int ret;
1938
1939 ret = health_check_state(&kconsumer_data.health) &&
1940 health_check_state(&ustconsumer32_data.health) &&
1941 health_check_state(&ustconsumer64_data.health);
1942
1943 DBG3("Health consumer check %d", ret);
1944
1945 return ret;
1946 }
1947
1948 /*
1949 * Setup necessary data for kernel tracer action.
1950 */
1951 static int init_kernel_tracer(void)
1952 {
1953 int ret;
1954
1955 /* Modprobe lttng kernel modules */
1956 ret = modprobe_lttng_control();
1957 if (ret < 0) {
1958 goto error;
1959 }
1960
1961 /* Open debugfs lttng */
1962 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1963 if (kernel_tracer_fd < 0) {
1964 DBG("Failed to open %s", module_proc_lttng);
1965 ret = -1;
1966 goto error_open;
1967 }
1968
1969 /* Validate kernel version */
1970 ret = kernel_validate_version(kernel_tracer_fd);
1971 if (ret < 0) {
1972 goto error_version;
1973 }
1974
1975 ret = modprobe_lttng_data();
1976 if (ret < 0) {
1977 goto error_modules;
1978 }
1979
1980 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1981 return 0;
1982
1983 error_version:
1984 modprobe_remove_lttng_control();
1985 ret = close(kernel_tracer_fd);
1986 if (ret) {
1987 PERROR("close");
1988 }
1989 kernel_tracer_fd = -1;
1990 return LTTNG_ERR_KERN_VERSION;
1991
1992 error_modules:
1993 ret = close(kernel_tracer_fd);
1994 if (ret) {
1995 PERROR("close");
1996 }
1997
1998 error_open:
1999 modprobe_remove_lttng_control();
2000
2001 error:
2002 WARN("No kernel tracer available");
2003 kernel_tracer_fd = -1;
2004 if (!is_root) {
2005 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2006 } else {
2007 return LTTNG_ERR_KERN_NA;
2008 }
2009 }
2010
2011
2012 /*
2013 * Copy consumer output from the tracing session to the domain session. The
2014 * function also applies the right modification on a per domain basis for the
2015 * trace files destination directory.
2016 */
2017 static int copy_session_consumer(int domain, struct ltt_session *session)
2018 {
2019 int ret;
2020 const char *dir_name;
2021 struct consumer_output *consumer;
2022
2023 assert(session);
2024 assert(session->consumer);
2025
2026 switch (domain) {
2027 case LTTNG_DOMAIN_KERNEL:
2028 DBG3("Copying tracing session consumer output in kernel session");
2029 /*
2030 * XXX: We should audit the session creation and what this function
2031 * does "extra" in order to avoid a destroy since this function is used
2032 * in the domain session creation (kernel and ust) only. Same for UST
2033 * domain.
2034 */
2035 if (session->kernel_session->consumer) {
2036 consumer_destroy_output(session->kernel_session->consumer);
2037 }
2038 session->kernel_session->consumer =
2039 consumer_copy_output(session->consumer);
2040 /* Ease our life a bit for the next part */
2041 consumer = session->kernel_session->consumer;
2042 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2043 break;
2044 case LTTNG_DOMAIN_UST:
2045 DBG3("Copying tracing session consumer output in UST session");
2046 if (session->ust_session->consumer) {
2047 consumer_destroy_output(session->ust_session->consumer);
2048 }
2049 session->ust_session->consumer =
2050 consumer_copy_output(session->consumer);
2051 /* Ease our life a bit for the next part */
2052 consumer = session->ust_session->consumer;
2053 dir_name = DEFAULT_UST_TRACE_DIR;
2054 break;
2055 default:
2056 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2057 goto error;
2058 }
2059
2060 /* Append correct directory to subdir */
2061 strncat(consumer->subdir, dir_name,
2062 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2063 DBG3("Copy session consumer subdir %s", consumer->subdir);
2064
2065 ret = LTTNG_OK;
2066
2067 error:
2068 return ret;
2069 }
2070
2071 /*
2072 * Create an UST session and add it to the session ust list.
2073 */
2074 static int create_ust_session(struct ltt_session *session,
2075 struct lttng_domain *domain)
2076 {
2077 int ret;
2078 struct ltt_ust_session *lus = NULL;
2079
2080 assert(session);
2081 assert(domain);
2082 assert(session->consumer);
2083
2084 switch (domain->type) {
2085 case LTTNG_DOMAIN_UST:
2086 break;
2087 default:
2088 ERR("Unknown UST domain on create session %d", domain->type);
2089 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2090 goto error;
2091 }
2092
2093 DBG("Creating UST session");
2094
2095 lus = trace_ust_create_session(session->path, session->id, domain);
2096 if (lus == NULL) {
2097 ret = LTTNG_ERR_UST_SESS_FAIL;
2098 goto error;
2099 }
2100
2101 lus->uid = session->uid;
2102 lus->gid = session->gid;
2103 session->ust_session = lus;
2104
2105 /* Copy session output to the newly created UST session */
2106 ret = copy_session_consumer(domain->type, session);
2107 if (ret != LTTNG_OK) {
2108 goto error;
2109 }
2110
2111 return LTTNG_OK;
2112
2113 error:
2114 free(lus);
2115 session->ust_session = NULL;
2116 return ret;
2117 }
2118
2119 /*
2120 * Create a kernel tracer session then create the default channel.
2121 */
2122 static int create_kernel_session(struct ltt_session *session)
2123 {
2124 int ret;
2125
2126 DBG("Creating kernel session");
2127
2128 ret = kernel_create_session(session, kernel_tracer_fd);
2129 if (ret < 0) {
2130 ret = LTTNG_ERR_KERN_SESS_FAIL;
2131 goto error;
2132 }
2133
2134 /* Code flow safety */
2135 assert(session->kernel_session);
2136
2137 /* Copy session output to the newly created Kernel session */
2138 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2139 if (ret != LTTNG_OK) {
2140 goto error;
2141 }
2142
2143 /* Create directory(ies) on local filesystem. */
2144 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2145 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2146 ret = run_as_mkdir_recursive(
2147 session->kernel_session->consumer->dst.trace_path,
2148 S_IRWXU | S_IRWXG, session->uid, session->gid);
2149 if (ret < 0) {
2150 if (ret != -EEXIST) {
2151 ERR("Trace directory creation error");
2152 goto error;
2153 }
2154 }
2155 }
2156
2157 session->kernel_session->uid = session->uid;
2158 session->kernel_session->gid = session->gid;
2159
2160 return LTTNG_OK;
2161
2162 error:
2163 trace_kernel_destroy_session(session->kernel_session);
2164 session->kernel_session = NULL;
2165 return ret;
2166 }
2167
2168 /*
2169 * Count number of session permitted by uid/gid.
2170 */
2171 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2172 {
2173 unsigned int i = 0;
2174 struct ltt_session *session;
2175
2176 DBG("Counting number of available session for UID %d GID %d",
2177 uid, gid);
2178 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2179 /*
2180 * Only list the sessions the user can control.
2181 */
2182 if (!session_access_ok(session, uid, gid)) {
2183 continue;
2184 }
2185 i++;
2186 }
2187 return i;
2188 }
2189
2190 /*
2191 * Process the command requested by the lttng client within the command
2192 * context structure. This function make sure that the return structure (llm)
2193 * is set and ready for transmission before returning.
2194 *
2195 * Return any error encountered or 0 for success.
2196 *
2197 * "sock" is only used for special-case var. len data.
2198 */
2199 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2200 int *sock_error)
2201 {
2202 int ret = LTTNG_OK;
2203 int need_tracing_session = 1;
2204 int need_domain;
2205
2206 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2207
2208 *sock_error = 0;
2209
2210 switch (cmd_ctx->lsm->cmd_type) {
2211 case LTTNG_CREATE_SESSION:
2212 case LTTNG_DESTROY_SESSION:
2213 case LTTNG_LIST_SESSIONS:
2214 case LTTNG_LIST_DOMAINS:
2215 case LTTNG_START_TRACE:
2216 case LTTNG_STOP_TRACE:
2217 case LTTNG_DATA_PENDING:
2218 need_domain = 0;
2219 break;
2220 default:
2221 need_domain = 1;
2222 }
2223
2224 if (opt_no_kernel && need_domain
2225 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2226 if (!is_root) {
2227 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2228 } else {
2229 ret = LTTNG_ERR_KERN_NA;
2230 }
2231 goto error;
2232 }
2233
2234 /* Deny register consumer if we already have a spawned consumer. */
2235 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2236 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2237 if (kconsumer_data.pid > 0) {
2238 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2239 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2240 goto error;
2241 }
2242 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2243 }
2244
2245 /*
2246 * Check for command that don't needs to allocate a returned payload. We do
2247 * this here so we don't have to make the call for no payload at each
2248 * command.
2249 */
2250 switch(cmd_ctx->lsm->cmd_type) {
2251 case LTTNG_LIST_SESSIONS:
2252 case LTTNG_LIST_TRACEPOINTS:
2253 case LTTNG_LIST_TRACEPOINT_FIELDS:
2254 case LTTNG_LIST_DOMAINS:
2255 case LTTNG_LIST_CHANNELS:
2256 case LTTNG_LIST_EVENTS:
2257 break;
2258 default:
2259 /* Setup lttng message with no payload */
2260 ret = setup_lttng_msg(cmd_ctx, 0);
2261 if (ret < 0) {
2262 /* This label does not try to unlock the session */
2263 goto init_setup_error;
2264 }
2265 }
2266
2267 /* Commands that DO NOT need a session. */
2268 switch (cmd_ctx->lsm->cmd_type) {
2269 case LTTNG_CREATE_SESSION:
2270 case LTTNG_CALIBRATE:
2271 case LTTNG_LIST_SESSIONS:
2272 case LTTNG_LIST_TRACEPOINTS:
2273 case LTTNG_LIST_TRACEPOINT_FIELDS:
2274 need_tracing_session = 0;
2275 break;
2276 default:
2277 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2278 /*
2279 * We keep the session list lock across _all_ commands
2280 * for now, because the per-session lock does not
2281 * handle teardown properly.
2282 */
2283 session_lock_list();
2284 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2285 if (cmd_ctx->session == NULL) {
2286 if (cmd_ctx->lsm->session.name != NULL) {
2287 ret = LTTNG_ERR_SESS_NOT_FOUND;
2288 } else {
2289 /* If no session name specified */
2290 ret = LTTNG_ERR_SELECT_SESS;
2291 }
2292 goto error;
2293 } else {
2294 /* Acquire lock for the session */
2295 session_lock(cmd_ctx->session);
2296 }
2297 break;
2298 }
2299
2300 if (!need_domain) {
2301 goto skip_domain;
2302 }
2303
2304 /*
2305 * Check domain type for specific "pre-action".
2306 */
2307 switch (cmd_ctx->lsm->domain.type) {
2308 case LTTNG_DOMAIN_KERNEL:
2309 if (!is_root) {
2310 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2311 goto error;
2312 }
2313
2314 /* Kernel tracer check */
2315 if (kernel_tracer_fd == -1) {
2316 /* Basically, load kernel tracer modules */
2317 ret = init_kernel_tracer();
2318 if (ret != 0) {
2319 goto error;
2320 }
2321 }
2322
2323 /* Consumer is in an ERROR state. Report back to client */
2324 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2325 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2326 goto error;
2327 }
2328
2329 /* Need a session for kernel command */
2330 if (need_tracing_session) {
2331 if (cmd_ctx->session->kernel_session == NULL) {
2332 ret = create_kernel_session(cmd_ctx->session);
2333 if (ret < 0) {
2334 ret = LTTNG_ERR_KERN_SESS_FAIL;
2335 goto error;
2336 }
2337 }
2338
2339 /* Start the kernel consumer daemon */
2340 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2341 if (kconsumer_data.pid == 0 &&
2342 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2343 cmd_ctx->session->start_consumer) {
2344 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2345 ret = start_consumerd(&kconsumer_data);
2346 if (ret < 0) {
2347 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2348 goto error;
2349 }
2350 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2351 } else {
2352 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2353 }
2354
2355 /*
2356 * The consumer was just spawned so we need to add the socket to
2357 * the consumer output of the session if exist.
2358 */
2359 ret = consumer_create_socket(&kconsumer_data,
2360 cmd_ctx->session->kernel_session->consumer);
2361 if (ret < 0) {
2362 goto error;
2363 }
2364 }
2365
2366 break;
2367 case LTTNG_DOMAIN_UST:
2368 {
2369 /* Consumer is in an ERROR state. Report back to client */
2370 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2371 ret = LTTNG_ERR_NO_USTCONSUMERD;
2372 goto error;
2373 }
2374
2375 if (need_tracing_session) {
2376 /* Create UST session if none exist. */
2377 if (cmd_ctx->session->ust_session == NULL) {
2378 ret = create_ust_session(cmd_ctx->session,
2379 &cmd_ctx->lsm->domain);
2380 if (ret != LTTNG_OK) {
2381 goto error;
2382 }
2383 }
2384
2385 /* Start the UST consumer daemons */
2386 /* 64-bit */
2387 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2388 if (consumerd64_bin[0] != '\0' &&
2389 ustconsumer64_data.pid == 0 &&
2390 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2391 cmd_ctx->session->start_consumer) {
2392 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2393 ret = start_consumerd(&ustconsumer64_data);
2394 if (ret < 0) {
2395 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2396 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2397 goto error;
2398 }
2399
2400 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2401 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2402 } else {
2403 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2404 }
2405
2406 /*
2407 * Setup socket for consumer 64 bit. No need for atomic access
2408 * since it was set above and can ONLY be set in this thread.
2409 */
2410 ret = consumer_create_socket(&ustconsumer64_data,
2411 cmd_ctx->session->ust_session->consumer);
2412 if (ret < 0) {
2413 goto error;
2414 }
2415
2416 /* 32-bit */
2417 if (consumerd32_bin[0] != '\0' &&
2418 ustconsumer32_data.pid == 0 &&
2419 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2420 cmd_ctx->session->start_consumer) {
2421 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2422 ret = start_consumerd(&ustconsumer32_data);
2423 if (ret < 0) {
2424 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2425 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2426 goto error;
2427 }
2428
2429 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2430 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2431 } else {
2432 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2433 }
2434
2435 /*
2436 * Setup socket for consumer 64 bit. No need for atomic access
2437 * since it was set above and can ONLY be set in this thread.
2438 */
2439 ret = consumer_create_socket(&ustconsumer32_data,
2440 cmd_ctx->session->ust_session->consumer);
2441 if (ret < 0) {
2442 goto error;
2443 }
2444 }
2445 break;
2446 }
2447 default:
2448 break;
2449 }
2450 skip_domain:
2451
2452 /* Validate consumer daemon state when start/stop trace command */
2453 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2454 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2455 switch (cmd_ctx->lsm->domain.type) {
2456 case LTTNG_DOMAIN_UST:
2457 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2458 ret = LTTNG_ERR_NO_USTCONSUMERD;
2459 goto error;
2460 }
2461 break;
2462 case LTTNG_DOMAIN_KERNEL:
2463 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2464 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2465 goto error;
2466 }
2467 break;
2468 }
2469 }
2470
2471 /*
2472 * Check that the UID or GID match that of the tracing session.
2473 * The root user can interact with all sessions.
2474 */
2475 if (need_tracing_session) {
2476 if (!session_access_ok(cmd_ctx->session,
2477 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2478 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2479 ret = LTTNG_ERR_EPERM;
2480 goto error;
2481 }
2482 }
2483
2484 /* Process by command type */
2485 switch (cmd_ctx->lsm->cmd_type) {
2486 case LTTNG_ADD_CONTEXT:
2487 {
2488 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2489 cmd_ctx->lsm->u.context.channel_name,
2490 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2491 break;
2492 }
2493 case LTTNG_DISABLE_CHANNEL:
2494 {
2495 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2496 cmd_ctx->lsm->u.disable.channel_name);
2497 break;
2498 }
2499 case LTTNG_DISABLE_EVENT:
2500 {
2501 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2502 cmd_ctx->lsm->u.disable.channel_name,
2503 cmd_ctx->lsm->u.disable.name);
2504 break;
2505 }
2506 case LTTNG_DISABLE_ALL_EVENT:
2507 {
2508 DBG("Disabling all events");
2509
2510 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2511 cmd_ctx->lsm->u.disable.channel_name);
2512 break;
2513 }
2514 case LTTNG_DISABLE_CONSUMER:
2515 {
2516 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2517 break;
2518 }
2519 case LTTNG_ENABLE_CHANNEL:
2520 {
2521 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2522 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2523 break;
2524 }
2525 case LTTNG_ENABLE_CONSUMER:
2526 {
2527 /*
2528 * XXX: 0 means that this URI should be applied on the session. Should
2529 * be a DOMAIN enuam.
2530 */
2531 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2532 if (ret != LTTNG_OK) {
2533 goto error;
2534 }
2535
2536 if (cmd_ctx->lsm->domain.type == 0) {
2537 /* Add the URI for the UST session if a consumer is present. */
2538 if (cmd_ctx->session->ust_session &&
2539 cmd_ctx->session->ust_session->consumer) {
2540 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2541 } else if (cmd_ctx->session->kernel_session &&
2542 cmd_ctx->session->kernel_session->consumer) {
2543 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2544 cmd_ctx->session);
2545 }
2546 }
2547 break;
2548 }
2549 case LTTNG_ENABLE_EVENT:
2550 {
2551 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2552 cmd_ctx->lsm->u.enable.channel_name,
2553 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2554 break;
2555 }
2556 case LTTNG_ENABLE_ALL_EVENT:
2557 {
2558 DBG("Enabling all events");
2559
2560 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2561 cmd_ctx->lsm->u.enable.channel_name,
2562 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2563 break;
2564 }
2565 case LTTNG_LIST_TRACEPOINTS:
2566 {
2567 struct lttng_event *events;
2568 ssize_t nb_events;
2569
2570 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2571 if (nb_events < 0) {
2572 /* Return value is a negative lttng_error_code. */
2573 ret = -nb_events;
2574 goto error;
2575 }
2576
2577 /*
2578 * Setup lttng message with payload size set to the event list size in
2579 * bytes and then copy list into the llm payload.
2580 */
2581 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2582 if (ret < 0) {
2583 free(events);
2584 goto setup_error;
2585 }
2586
2587 /* Copy event list into message payload */
2588 memcpy(cmd_ctx->llm->payload, events,
2589 sizeof(struct lttng_event) * nb_events);
2590
2591 free(events);
2592
2593 ret = LTTNG_OK;
2594 break;
2595 }
2596 case LTTNG_LIST_TRACEPOINT_FIELDS:
2597 {
2598 struct lttng_event_field *fields;
2599 ssize_t nb_fields;
2600
2601 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2602 &fields);
2603 if (nb_fields < 0) {
2604 /* Return value is a negative lttng_error_code. */
2605 ret = -nb_fields;
2606 goto error;
2607 }
2608
2609 /*
2610 * Setup lttng message with payload size set to the event list size in
2611 * bytes and then copy list into the llm payload.
2612 */
2613 ret = setup_lttng_msg(cmd_ctx,
2614 sizeof(struct lttng_event_field) * nb_fields);
2615 if (ret < 0) {
2616 free(fields);
2617 goto setup_error;
2618 }
2619
2620 /* Copy event list into message payload */
2621 memcpy(cmd_ctx->llm->payload, fields,
2622 sizeof(struct lttng_event_field) * nb_fields);
2623
2624 free(fields);
2625
2626 ret = LTTNG_OK;
2627 break;
2628 }
2629 case LTTNG_SET_CONSUMER_URI:
2630 {
2631 size_t nb_uri, len;
2632 struct lttng_uri *uris;
2633
2634 nb_uri = cmd_ctx->lsm->u.uri.size;
2635 len = nb_uri * sizeof(struct lttng_uri);
2636
2637 if (nb_uri == 0) {
2638 ret = LTTNG_ERR_INVALID;
2639 goto error;
2640 }
2641
2642 uris = zmalloc(len);
2643 if (uris == NULL) {
2644 ret = LTTNG_ERR_FATAL;
2645 goto error;
2646 }
2647
2648 /* Receive variable len data */
2649 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2650 ret = lttcomm_recv_unix_sock(sock, uris, len);
2651 if (ret <= 0) {
2652 DBG("No URIs received from client... continuing");
2653 *sock_error = 1;
2654 ret = LTTNG_ERR_SESSION_FAIL;
2655 free(uris);
2656 goto error;
2657 }
2658
2659 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2660 nb_uri, uris);
2661 if (ret != LTTNG_OK) {
2662 free(uris);
2663 goto error;
2664 }
2665
2666 /*
2667 * XXX: 0 means that this URI should be applied on the session. Should
2668 * be a DOMAIN enuam.
2669 */
2670 if (cmd_ctx->lsm->domain.type == 0) {
2671 /* Add the URI for the UST session if a consumer is present. */
2672 if (cmd_ctx->session->ust_session &&
2673 cmd_ctx->session->ust_session->consumer) {
2674 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2675 nb_uri, uris);
2676 } else if (cmd_ctx->session->kernel_session &&
2677 cmd_ctx->session->kernel_session->consumer) {
2678 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2679 cmd_ctx->session, nb_uri, uris);
2680 }
2681 }
2682
2683 free(uris);
2684
2685 break;
2686 }
2687 case LTTNG_START_TRACE:
2688 {
2689 ret = cmd_start_trace(cmd_ctx->session);
2690 break;
2691 }
2692 case LTTNG_STOP_TRACE:
2693 {
2694 ret = cmd_stop_trace(cmd_ctx->session);
2695 break;
2696 }
2697 case LTTNG_CREATE_SESSION:
2698 {
2699 size_t nb_uri, len;
2700 struct lttng_uri *uris = NULL;
2701
2702 nb_uri = cmd_ctx->lsm->u.uri.size;
2703 len = nb_uri * sizeof(struct lttng_uri);
2704
2705 if (nb_uri > 0) {
2706 uris = zmalloc(len);
2707 if (uris == NULL) {
2708 ret = LTTNG_ERR_FATAL;
2709 goto error;
2710 }
2711
2712 /* Receive variable len data */
2713 DBG("Waiting for %zu URIs from client ...", nb_uri);
2714 ret = lttcomm_recv_unix_sock(sock, uris, len);
2715 if (ret <= 0) {
2716 DBG("No URIs received from client... continuing");
2717 *sock_error = 1;
2718 ret = LTTNG_ERR_SESSION_FAIL;
2719 free(uris);
2720 goto error;
2721 }
2722
2723 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2724 DBG("Creating session with ONE network URI is a bad call");
2725 ret = LTTNG_ERR_SESSION_FAIL;
2726 free(uris);
2727 goto error;
2728 }
2729 }
2730
2731 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2732 &cmd_ctx->creds);
2733
2734 free(uris);
2735
2736 break;
2737 }
2738 case LTTNG_DESTROY_SESSION:
2739 {
2740 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2741
2742 /* Set session to NULL so we do not unlock it after free. */
2743 cmd_ctx->session = NULL;
2744 break;
2745 }
2746 case LTTNG_LIST_DOMAINS:
2747 {
2748 ssize_t nb_dom;
2749 struct lttng_domain *domains;
2750
2751 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2752 if (nb_dom < 0) {
2753 /* Return value is a negative lttng_error_code. */
2754 ret = -nb_dom;
2755 goto error;
2756 }
2757
2758 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2759 if (ret < 0) {
2760 goto setup_error;
2761 }
2762
2763 /* Copy event list into message payload */
2764 memcpy(cmd_ctx->llm->payload, domains,
2765 nb_dom * sizeof(struct lttng_domain));
2766
2767 free(domains);
2768
2769 ret = LTTNG_OK;
2770 break;
2771 }
2772 case LTTNG_LIST_CHANNELS:
2773 {
2774 int nb_chan;
2775 struct lttng_channel *channels;
2776
2777 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2778 cmd_ctx->session, &channels);
2779 if (nb_chan < 0) {
2780 /* Return value is a negative lttng_error_code. */
2781 ret = -nb_chan;
2782 goto error;
2783 }
2784
2785 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2786 if (ret < 0) {
2787 goto setup_error;
2788 }
2789
2790 /* Copy event list into message payload */
2791 memcpy(cmd_ctx->llm->payload, channels,
2792 nb_chan * sizeof(struct lttng_channel));
2793
2794 free(channels);
2795
2796 ret = LTTNG_OK;
2797 break;
2798 }
2799 case LTTNG_LIST_EVENTS:
2800 {
2801 ssize_t nb_event;
2802 struct lttng_event *events = NULL;
2803
2804 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2805 cmd_ctx->lsm->u.list.channel_name, &events);
2806 if (nb_event < 0) {
2807 /* Return value is a negative lttng_error_code. */
2808 ret = -nb_event;
2809 goto error;
2810 }
2811
2812 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2813 if (ret < 0) {
2814 goto setup_error;
2815 }
2816
2817 /* Copy event list into message payload */
2818 memcpy(cmd_ctx->llm->payload, events,
2819 nb_event * sizeof(struct lttng_event));
2820
2821 free(events);
2822
2823 ret = LTTNG_OK;
2824 break;
2825 }
2826 case LTTNG_LIST_SESSIONS:
2827 {
2828 unsigned int nr_sessions;
2829
2830 session_lock_list();
2831 nr_sessions = lttng_sessions_count(
2832 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2833 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2834
2835 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2836 if (ret < 0) {
2837 session_unlock_list();
2838 goto setup_error;
2839 }
2840
2841 /* Filled the session array */
2842 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2843 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2844 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2845
2846 session_unlock_list();
2847
2848 ret = LTTNG_OK;
2849 break;
2850 }
2851 case LTTNG_CALIBRATE:
2852 {
2853 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2854 &cmd_ctx->lsm->u.calibrate);
2855 break;
2856 }
2857 case LTTNG_REGISTER_CONSUMER:
2858 {
2859 struct consumer_data *cdata;
2860
2861 switch (cmd_ctx->lsm->domain.type) {
2862 case LTTNG_DOMAIN_KERNEL:
2863 cdata = &kconsumer_data;
2864 break;
2865 default:
2866 ret = LTTNG_ERR_UND;
2867 goto error;
2868 }
2869
2870 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2871 cmd_ctx->lsm->u.reg.path, cdata);
2872 break;
2873 }
2874 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2875 {
2876 struct lttng_filter_bytecode *bytecode;
2877
2878 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2879 ret = LTTNG_ERR_FILTER_INVAL;
2880 goto error;
2881 }
2882 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2883 ret = LTTNG_ERR_FILTER_INVAL;
2884 goto error;
2885 }
2886 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2887 if (!bytecode) {
2888 ret = LTTNG_ERR_FILTER_NOMEM;
2889 goto error;
2890 }
2891 /* Receive var. len. data */
2892 DBG("Receiving var len data from client ...");
2893 ret = lttcomm_recv_unix_sock(sock, bytecode,
2894 cmd_ctx->lsm->u.enable.bytecode_len);
2895 if (ret <= 0) {
2896 DBG("Nothing recv() from client var len data... continuing");
2897 *sock_error = 1;
2898 ret = LTTNG_ERR_FILTER_INVAL;
2899 goto error;
2900 }
2901
2902 if (bytecode->len + sizeof(*bytecode)
2903 != cmd_ctx->lsm->u.enable.bytecode_len) {
2904 free(bytecode);
2905 ret = LTTNG_ERR_FILTER_INVAL;
2906 goto error;
2907 }
2908
2909 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2910 cmd_ctx->lsm->u.enable.channel_name,
2911 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2912 break;
2913 }
2914 case LTTNG_DATA_PENDING:
2915 {
2916 ret = cmd_data_pending(cmd_ctx->session);
2917 break;
2918 }
2919 default:
2920 ret = LTTNG_ERR_UND;
2921 break;
2922 }
2923
2924 error:
2925 if (cmd_ctx->llm == NULL) {
2926 DBG("Missing llm structure. Allocating one.");
2927 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2928 goto setup_error;
2929 }
2930 }
2931 /* Set return code */
2932 cmd_ctx->llm->ret_code = ret;
2933 setup_error:
2934 if (cmd_ctx->session) {
2935 session_unlock(cmd_ctx->session);
2936 }
2937 if (need_tracing_session) {
2938 session_unlock_list();
2939 }
2940 init_setup_error:
2941 return ret;
2942 }
2943
2944 /*
2945 * Thread managing health check socket.
2946 */
2947 static void *thread_manage_health(void *data)
2948 {
2949 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2950 uint32_t revents, nb_fd;
2951 struct lttng_poll_event events;
2952 struct lttcomm_health_msg msg;
2953 struct lttcomm_health_data reply;
2954
2955 DBG("[thread] Manage health check started");
2956
2957 rcu_register_thread();
2958
2959 /* Create unix socket */
2960 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2961 if (sock < 0) {
2962 ERR("Unable to create health check Unix socket");
2963 ret = -1;
2964 goto error;
2965 }
2966
2967 /*
2968 * Set the CLOEXEC flag. Return code is useless because either way, the
2969 * show must go on.
2970 */
2971 (void) utils_set_fd_cloexec(sock);
2972
2973 ret = lttcomm_listen_unix_sock(sock);
2974 if (ret < 0) {
2975 goto error;
2976 }
2977
2978 /*
2979 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2980 * more will be added to this poll set.
2981 */
2982 ret = create_thread_poll_set(&events, 2);
2983 if (ret < 0) {
2984 goto error;
2985 }
2986
2987 /* Add the application registration socket */
2988 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2989 if (ret < 0) {
2990 goto error;
2991 }
2992
2993 while (1) {
2994 DBG("Health check ready");
2995
2996 /* Inifinite blocking call, waiting for transmission */
2997 restart:
2998 ret = lttng_poll_wait(&events, -1);
2999 if (ret < 0) {
3000 /*
3001 * Restart interrupted system call.
3002 */
3003 if (errno == EINTR) {
3004 goto restart;
3005 }
3006 goto error;
3007 }
3008
3009 nb_fd = ret;
3010
3011 for (i = 0; i < nb_fd; i++) {
3012 /* Fetch once the poll data */
3013 revents = LTTNG_POLL_GETEV(&events, i);
3014 pollfd = LTTNG_POLL_GETFD(&events, i);
3015
3016 /* Thread quit pipe has been closed. Killing thread. */
3017 ret = check_thread_quit_pipe(pollfd, revents);
3018 if (ret) {
3019 err = 0;
3020 goto exit;
3021 }
3022
3023 /* Event on the registration socket */
3024 if (pollfd == sock) {
3025 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3026 ERR("Health socket poll error");
3027 goto error;
3028 }
3029 }
3030 }
3031
3032 new_sock = lttcomm_accept_unix_sock(sock);
3033 if (new_sock < 0) {
3034 goto error;
3035 }
3036
3037 /*
3038 * Set the CLOEXEC flag. Return code is useless because either way, the
3039 * show must go on.
3040 */
3041 (void) utils_set_fd_cloexec(new_sock);
3042
3043 DBG("Receiving data from client for health...");
3044 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3045 if (ret <= 0) {
3046 DBG("Nothing recv() from client... continuing");
3047 ret = close(new_sock);
3048 if (ret) {
3049 PERROR("close");
3050 }
3051 new_sock = -1;
3052 continue;
3053 }
3054
3055 rcu_thread_online();
3056
3057 switch (msg.component) {
3058 case LTTNG_HEALTH_CMD:
3059 reply.ret_code = health_check_state(&health_thread_cmd);
3060 break;
3061 case LTTNG_HEALTH_APP_MANAGE:
3062 reply.ret_code = health_check_state(&health_thread_app_manage);
3063 break;
3064 case LTTNG_HEALTH_APP_REG:
3065 reply.ret_code = health_check_state(&health_thread_app_reg);
3066 break;
3067 case LTTNG_HEALTH_KERNEL:
3068 reply.ret_code = health_check_state(&health_thread_kernel);
3069 break;
3070 case LTTNG_HEALTH_CONSUMER:
3071 reply.ret_code = check_consumer_health();
3072 break;
3073 case LTTNG_HEALTH_ALL:
3074 reply.ret_code =
3075 health_check_state(&health_thread_app_manage) &&
3076 health_check_state(&health_thread_app_reg) &&
3077 health_check_state(&health_thread_cmd) &&
3078 health_check_state(&health_thread_kernel) &&
3079 check_consumer_health();
3080 break;
3081 default:
3082 reply.ret_code = LTTNG_ERR_UND;
3083 break;
3084 }
3085
3086 /*
3087 * Flip ret value since 0 is a success and 1 indicates a bad health for
3088 * the client where in the sessiond it is the opposite. Again, this is
3089 * just to make things easier for us poor developer which enjoy a lot
3090 * lazyness.
3091 */
3092 if (reply.ret_code == 0 || reply.ret_code == 1) {
3093 reply.ret_code = !reply.ret_code;
3094 }
3095
3096 DBG2("Health check return value %d", reply.ret_code);
3097
3098 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3099 if (ret < 0) {
3100 ERR("Failed to send health data back to client");
3101 }
3102
3103 /* End of transmission */
3104 ret = close(new_sock);
3105 if (ret) {
3106 PERROR("close");
3107 }
3108 new_sock = -1;
3109 }
3110
3111 exit:
3112 error:
3113 if (err) {
3114 ERR("Health error occurred in %s", __func__);
3115 }
3116 DBG("Health check thread dying");
3117 unlink(health_unix_sock_path);
3118 if (sock >= 0) {
3119 ret = close(sock);
3120 if (ret) {
3121 PERROR("close");
3122 }
3123 }
3124 if (new_sock >= 0) {
3125 ret = close(new_sock);
3126 if (ret) {
3127 PERROR("close");
3128 }
3129 }
3130
3131 lttng_poll_clean(&events);
3132
3133 rcu_unregister_thread();
3134 return NULL;
3135 }
3136
3137 /*
3138 * This thread manage all clients request using the unix client socket for
3139 * communication.
3140 */
3141 static void *thread_manage_clients(void *data)
3142 {
3143 int sock = -1, ret, i, pollfd, err = -1;
3144 int sock_error;
3145 uint32_t revents, nb_fd;
3146 struct command_ctx *cmd_ctx = NULL;
3147 struct lttng_poll_event events;
3148
3149 DBG("[thread] Manage client started");
3150
3151 rcu_register_thread();
3152
3153 if (testpoint(thread_manage_clients)) {
3154 goto error_testpoint;
3155 }
3156
3157 health_code_update(&health_thread_cmd);
3158
3159 ret = lttcomm_listen_unix_sock(client_sock);
3160 if (ret < 0) {
3161 goto error_listen;
3162 }
3163
3164 /*
3165 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3166 * more will be added to this poll set.
3167 */
3168 ret = create_thread_poll_set(&events, 2);
3169 if (ret < 0) {
3170 goto error_create_poll;
3171 }
3172
3173 /* Add the application registration socket */
3174 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3175 if (ret < 0) {
3176 goto error;
3177 }
3178
3179 /*
3180 * Notify parent pid that we are ready to accept command for client side.
3181 */
3182 if (opt_sig_parent) {
3183 kill(ppid, SIGUSR1);
3184 }
3185
3186 if (testpoint(thread_manage_clients_before_loop)) {
3187 goto error;
3188 }
3189
3190 health_code_update(&health_thread_cmd);
3191
3192 while (1) {
3193 DBG("Accepting client command ...");
3194
3195 /* Inifinite blocking call, waiting for transmission */
3196 restart:
3197 health_poll_update(&health_thread_cmd);
3198 ret = lttng_poll_wait(&events, -1);
3199 health_poll_update(&health_thread_cmd);
3200 if (ret < 0) {
3201 /*
3202 * Restart interrupted system call.
3203 */
3204 if (errno == EINTR) {
3205 goto restart;
3206 }
3207 goto error;
3208 }
3209
3210 nb_fd = ret;
3211
3212 for (i = 0; i < nb_fd; i++) {
3213 /* Fetch once the poll data */
3214 revents = LTTNG_POLL_GETEV(&events, i);
3215 pollfd = LTTNG_POLL_GETFD(&events, i);
3216
3217 health_code_update(&health_thread_cmd);
3218
3219 /* Thread quit pipe has been closed. Killing thread. */
3220 ret = check_thread_quit_pipe(pollfd, revents);
3221 if (ret) {
3222 err = 0;
3223 goto exit;
3224 }
3225
3226 /* Event on the registration socket */
3227 if (pollfd == client_sock) {
3228 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3229 ERR("Client socket poll error");
3230 goto error;
3231 }
3232 }
3233 }
3234
3235 DBG("Wait for client response");
3236
3237 health_code_update(&health_thread_cmd);
3238
3239 sock = lttcomm_accept_unix_sock(client_sock);
3240 if (sock < 0) {
3241 goto error;
3242 }
3243
3244 /*
3245 * Set the CLOEXEC flag. Return code is useless because either way, the
3246 * show must go on.
3247 */
3248 (void) utils_set_fd_cloexec(sock);
3249
3250 /* Set socket option for credentials retrieval */
3251 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3252 if (ret < 0) {
3253 goto error;
3254 }
3255
3256 /* Allocate context command to process the client request */
3257 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3258 if (cmd_ctx == NULL) {
3259 PERROR("zmalloc cmd_ctx");
3260 goto error;
3261 }
3262
3263 /* Allocate data buffer for reception */
3264 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3265 if (cmd_ctx->lsm == NULL) {
3266 PERROR("zmalloc cmd_ctx->lsm");
3267 goto error;
3268 }
3269
3270 cmd_ctx->llm = NULL;
3271 cmd_ctx->session = NULL;
3272
3273 health_code_update(&health_thread_cmd);
3274
3275 /*
3276 * Data is received from the lttng client. The struct
3277 * lttcomm_session_msg (lsm) contains the command and data request of
3278 * the client.
3279 */
3280 DBG("Receiving data from client ...");
3281 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3282 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3283 if (ret <= 0) {
3284 DBG("Nothing recv() from client... continuing");
3285 ret = close(sock);
3286 if (ret) {
3287 PERROR("close");
3288 }
3289 sock = -1;
3290 clean_command_ctx(&cmd_ctx);
3291 continue;
3292 }
3293
3294 health_code_update(&health_thread_cmd);
3295
3296 // TODO: Validate cmd_ctx including sanity check for
3297 // security purpose.
3298
3299 rcu_thread_online();
3300 /*
3301 * This function dispatch the work to the kernel or userspace tracer
3302 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3303 * informations for the client. The command context struct contains
3304 * everything this function may needs.
3305 */
3306 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3307 rcu_thread_offline();
3308 if (ret < 0) {
3309 if (sock_error) {
3310 ret = close(sock);
3311 if (ret) {
3312 PERROR("close");
3313 }
3314 sock = -1;
3315 }
3316 /*
3317 * TODO: Inform client somehow of the fatal error. At
3318 * this point, ret < 0 means that a zmalloc failed
3319 * (ENOMEM). Error detected but still accept
3320 * command, unless a socket error has been
3321 * detected.
3322 */
3323 clean_command_ctx(&cmd_ctx);
3324 continue;
3325 }
3326
3327 health_code_update(&health_thread_cmd);
3328
3329 DBG("Sending response (size: %d, retcode: %s)",
3330 cmd_ctx->lttng_msg_size,
3331 lttng_strerror(-cmd_ctx->llm->ret_code));
3332 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3333 if (ret < 0) {
3334 ERR("Failed to send data back to client");
3335 }
3336
3337 /* End of transmission */
3338 ret = close(sock);
3339 if (ret) {
3340 PERROR("close");
3341 }
3342 sock = -1;
3343
3344 clean_command_ctx(&cmd_ctx);
3345
3346 health_code_update(&health_thread_cmd);
3347 }
3348
3349 exit:
3350 error:
3351 if (sock >= 0) {
3352 ret = close(sock);
3353 if (ret) {
3354 PERROR("close");
3355 }
3356 }
3357
3358 lttng_poll_clean(&events);
3359 clean_command_ctx(&cmd_ctx);
3360
3361 error_listen:
3362 error_create_poll:
3363 error_testpoint:
3364 unlink(client_unix_sock_path);
3365 if (client_sock >= 0) {
3366 ret = close(client_sock);
3367 if (ret) {
3368 PERROR("close");
3369 }
3370 }
3371
3372 if (err) {
3373 health_error(&health_thread_cmd);
3374 ERR("Health error occurred in %s", __func__);
3375 }
3376
3377 health_exit(&health_thread_cmd);
3378
3379 DBG("Client thread dying");
3380
3381 rcu_unregister_thread();
3382 return NULL;
3383 }
3384
3385
3386 /*
3387 * usage function on stderr
3388 */
3389 static void usage(void)
3390 {
3391 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3392 fprintf(stderr, " -h, --help Display this usage.\n");
3393 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3394 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3395 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3396 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3397 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3398 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3399 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3400 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3401 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3402 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3403 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3404 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3405 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3406 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3407 fprintf(stderr, " -V, --version Show version number.\n");
3408 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3409 fprintf(stderr, " -q, --quiet No output at all.\n");
3410 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3411 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3412 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3413 }
3414
3415 /*
3416 * daemon argument parsing
3417 */
3418 static int parse_args(int argc, char **argv)
3419 {
3420 int c;
3421
3422 static struct option long_options[] = {
3423 { "client-sock", 1, 0, 'c' },
3424 { "apps-sock", 1, 0, 'a' },
3425 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3426 { "kconsumerd-err-sock", 1, 0, 'E' },
3427 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3428 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3429 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3430 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3431 { "consumerd32-path", 1, 0, 'u' },
3432 { "consumerd32-libdir", 1, 0, 'U' },
3433 { "consumerd64-path", 1, 0, 't' },
3434 { "consumerd64-libdir", 1, 0, 'T' },
3435 { "daemonize", 0, 0, 'd' },
3436 { "sig-parent", 0, 0, 'S' },
3437 { "help", 0, 0, 'h' },
3438 { "group", 1, 0, 'g' },
3439 { "version", 0, 0, 'V' },
3440 { "quiet", 0, 0, 'q' },
3441 { "verbose", 0, 0, 'v' },
3442 { "verbose-consumer", 0, 0, 'Z' },
3443 { "no-kernel", 0, 0, 'N' },
3444 { NULL, 0, 0, 0 }
3445 };
3446
3447 while (1) {
3448 int option_index = 0;
3449 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3450 long_options, &option_index);
3451 if (c == -1) {
3452 break;
3453 }
3454
3455 switch (c) {
3456 case 0:
3457 fprintf(stderr, "option %s", long_options[option_index].name);
3458 if (optarg) {
3459 fprintf(stderr, " with arg %s\n", optarg);
3460 }
3461 break;
3462 case 'c':
3463 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3464 break;
3465 case 'a':
3466 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3467 break;
3468 case 'd':
3469 opt_daemon = 1;
3470 break;
3471 case 'g':
3472 opt_tracing_group = optarg;
3473 break;
3474 case 'h':
3475 usage();
3476 exit(EXIT_FAILURE);
3477 case 'V':
3478 fprintf(stdout, "%s\n", VERSION);
3479 exit(EXIT_SUCCESS);
3480 case 'S':
3481 opt_sig_parent = 1;
3482 break;
3483 case 'E':
3484 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3485 break;
3486 case 'C':
3487 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3488 break;
3489 case 'F':
3490 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3491 break;
3492 case 'D':
3493 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3494 break;
3495 case 'H':
3496 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3497 break;
3498 case 'G':
3499 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3500 break;
3501 case 'N':
3502 opt_no_kernel = 1;
3503 break;
3504 case 'q':
3505 lttng_opt_quiet = 1;
3506 break;
3507 case 'v':
3508 /* Verbose level can increase using multiple -v */
3509 lttng_opt_verbose += 1;
3510 break;
3511 case 'Z':
3512 opt_verbose_consumer += 1;
3513 break;
3514 case 'u':
3515 consumerd32_bin= optarg;
3516 break;
3517 case 'U':
3518 consumerd32_libdir = optarg;
3519 break;
3520 case 't':
3521 consumerd64_bin = optarg;
3522 break;
3523 case 'T':
3524 consumerd64_libdir = optarg;
3525 break;
3526 default:
3527 /* Unknown option or other error.
3528 * Error is printed by getopt, just return */
3529 return -1;
3530 }
3531 }
3532
3533 return 0;
3534 }
3535
3536 /*
3537 * Creates the two needed socket by the daemon.
3538 * apps_sock - The communication socket for all UST apps.
3539 * client_sock - The communication of the cli tool (lttng).
3540 */
3541 static int init_daemon_socket(void)
3542 {
3543 int ret = 0;
3544 mode_t old_umask;
3545
3546 old_umask = umask(0);
3547
3548 /* Create client tool unix socket */
3549 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3550 if (client_sock < 0) {
3551 ERR("Create unix sock failed: %s", client_unix_sock_path);
3552 ret = -1;
3553 goto end;
3554 }
3555
3556 /* Set the cloexec flag */
3557 ret = utils_set_fd_cloexec(client_sock);
3558 if (ret < 0) {
3559 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3560 "Continuing but note that the consumer daemon will have a "
3561 "reference to this socket on exec()", client_sock);
3562 }
3563
3564 /* File permission MUST be 660 */
3565 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3566 if (ret < 0) {
3567 ERR("Set file permissions failed: %s", client_unix_sock_path);
3568 PERROR("chmod");
3569 goto end;
3570 }
3571
3572 /* Create the application unix socket */
3573 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3574 if (apps_sock < 0) {
3575 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3576 ret = -1;
3577 goto end;
3578 }
3579
3580 /* Set the cloexec flag */
3581 ret = utils_set_fd_cloexec(apps_sock);
3582 if (ret < 0) {
3583 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3584 "Continuing but note that the consumer daemon will have a "
3585 "reference to this socket on exec()", apps_sock);
3586 }
3587
3588 /* File permission MUST be 666 */
3589 ret = chmod(apps_unix_sock_path,
3590 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3591 if (ret < 0) {
3592 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3593 PERROR("chmod");
3594 goto end;
3595 }
3596
3597 DBG3("Session daemon client socket %d and application socket %d created",
3598 client_sock, apps_sock);
3599
3600 end:
3601 umask(old_umask);
3602 return ret;
3603 }
3604
3605 /*
3606 * Check if the global socket is available, and if a daemon is answering at the
3607 * other side. If yes, error is returned.
3608 */
3609 static int check_existing_daemon(void)
3610 {
3611 /* Is there anybody out there ? */
3612 if (lttng_session_daemon_alive()) {
3613 return -EEXIST;
3614 }
3615
3616 return 0;
3617 }
3618
3619 /*
3620 * Set the tracing group gid onto the client socket.
3621 *
3622 * Race window between mkdir and chown is OK because we are going from more
3623 * permissive (root.root) to less permissive (root.tracing).
3624 */
3625 static int set_permissions(char *rundir)
3626 {
3627 int ret;
3628 gid_t gid;
3629
3630 ret = allowed_group();
3631 if (ret < 0) {
3632 WARN("No tracing group detected");
3633 ret = 0;
3634 goto end;
3635 }
3636
3637 gid = ret;
3638
3639 /* Set lttng run dir */
3640 ret = chown(rundir, 0, gid);
3641 if (ret < 0) {
3642 ERR("Unable to set group on %s", rundir);
3643 PERROR("chown");
3644 }
3645
3646 /* Ensure tracing group can search the run dir */
3647 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3648 if (ret < 0) {
3649 ERR("Unable to set permissions on %s", rundir);
3650 PERROR("chmod");
3651 }
3652
3653 /* lttng client socket path */
3654 ret = chown(client_unix_sock_path, 0, gid);
3655 if (ret < 0) {
3656 ERR("Unable to set group on %s", client_unix_sock_path);
3657 PERROR("chown");
3658 }
3659
3660 /* kconsumer error socket path */
3661 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3662 if (ret < 0) {
3663 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3664 PERROR("chown");
3665 }
3666
3667 /* 64-bit ustconsumer error socket path */
3668 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3669 if (ret < 0) {
3670 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3671 PERROR("chown");
3672 }
3673
3674 /* 32-bit ustconsumer compat32 error socket path */
3675 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3676 if (ret < 0) {
3677 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3678 PERROR("chown");
3679 }
3680
3681 DBG("All permissions are set");
3682
3683 end:
3684 return ret;
3685 }
3686
3687 /*
3688 * Create the lttng run directory needed for all global sockets and pipe.
3689 */
3690 static int create_lttng_rundir(const char *rundir)
3691 {
3692 int ret;
3693
3694 DBG3("Creating LTTng run directory: %s", rundir);
3695
3696 ret = mkdir(rundir, S_IRWXU);
3697 if (ret < 0) {
3698 if (errno != EEXIST) {
3699 ERR("Unable to create %s", rundir);
3700 goto error;
3701 } else {
3702 ret = 0;
3703 }
3704 }
3705
3706 error:
3707 return ret;
3708 }
3709
3710 /*
3711 * Setup sockets and directory needed by the kconsumerd communication with the
3712 * session daemon.
3713 */
3714 static int set_consumer_sockets(struct consumer_data *consumer_data,
3715 const char *rundir)
3716 {
3717 int ret;
3718 char path[PATH_MAX];
3719
3720 switch (consumer_data->type) {
3721 case LTTNG_CONSUMER_KERNEL:
3722 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3723 break;
3724 case LTTNG_CONSUMER64_UST:
3725 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3726 break;
3727 case LTTNG_CONSUMER32_UST:
3728 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3729 break;
3730 default:
3731 ERR("Consumer type unknown");
3732 ret = -EINVAL;
3733 goto error;
3734 }
3735
3736 DBG2("Creating consumer directory: %s", path);
3737
3738 ret = mkdir(path, S_IRWXU);
3739 if (ret < 0) {
3740 if (errno != EEXIST) {
3741 PERROR("mkdir");
3742 ERR("Failed to create %s", path);
3743 goto error;
3744 }
3745 ret = -1;
3746 }
3747
3748 /* Create the kconsumerd error unix socket */
3749 consumer_data->err_sock =
3750 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3751 if (consumer_data->err_sock < 0) {
3752 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3753 ret = -1;
3754 goto error;
3755 }
3756
3757 /* File permission MUST be 660 */
3758 ret = chmod(consumer_data->err_unix_sock_path,
3759 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3760 if (ret < 0) {
3761 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3762 PERROR("chmod");
3763 goto error;
3764 }
3765
3766 error:
3767 return ret;
3768 }
3769
3770 /*
3771 * Signal handler for the daemon
3772 *
3773 * Simply stop all worker threads, leaving main() return gracefully after
3774 * joining all threads and calling cleanup().
3775 */
3776 static void sighandler(int sig)
3777 {
3778 switch (sig) {
3779 case SIGPIPE:
3780 DBG("SIGPIPE caught");
3781 return;
3782 case SIGINT:
3783 DBG("SIGINT caught");
3784 stop_threads();
3785 break;
3786 case SIGTERM:
3787 DBG("SIGTERM caught");
3788 stop_threads();
3789 break;
3790 default:
3791 break;
3792 }
3793 }
3794
3795 /*
3796 * Setup signal handler for :
3797 * SIGINT, SIGTERM, SIGPIPE
3798 */
3799 static int set_signal_handler(void)
3800 {
3801 int ret = 0;
3802 struct sigaction sa;
3803 sigset_t sigset;
3804
3805 if ((ret = sigemptyset(&sigset)) < 0) {
3806 PERROR("sigemptyset");
3807 return ret;
3808 }
3809
3810 sa.sa_handler = sighandler;
3811 sa.sa_mask = sigset;
3812 sa.sa_flags = 0;
3813 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3814 PERROR("sigaction");
3815 return ret;
3816 }
3817
3818 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3819 PERROR("sigaction");
3820 return ret;
3821 }
3822
3823 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3824 PERROR("sigaction");
3825 return ret;
3826 }
3827
3828 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3829
3830 return ret;
3831 }
3832
3833 /*
3834 * Set open files limit to unlimited. This daemon can open a large number of
3835 * file descriptors in order to consumer multiple kernel traces.
3836 */
3837 static void set_ulimit(void)
3838 {
3839 int ret;
3840 struct rlimit lim;
3841
3842 /* The kernel does not allowed an infinite limit for open files */
3843 lim.rlim_cur = 65535;
3844 lim.rlim_max = 65535;
3845
3846 ret = setrlimit(RLIMIT_NOFILE, &lim);
3847 if (ret < 0) {
3848 PERROR("failed to set open files limit");
3849 }
3850 }
3851
3852 /*
3853 * main
3854 */
3855 int main(int argc, char **argv)
3856 {
3857 int ret = 0;
3858 void *status;
3859 const char *home_path, *env_app_timeout;
3860
3861 init_kernel_workarounds();
3862
3863 rcu_register_thread();
3864
3865 setup_consumerd_path();
3866
3867 /* Parse arguments */
3868 progname = argv[0];
3869 if ((ret = parse_args(argc, argv)) < 0) {
3870 goto error;
3871 }
3872
3873 /* Daemonize */
3874 if (opt_daemon) {
3875 int i;
3876
3877 /*
3878 * fork
3879 * child: setsid, close FD 0, 1, 2, chdir /
3880 * parent: exit (if fork is successful)
3881 */
3882 ret = daemon(0, 0);
3883 if (ret < 0) {
3884 PERROR("daemon");
3885 goto error;
3886 }
3887 /*
3888 * We are in the child. Make sure all other file
3889 * descriptors are closed, in case we are called with
3890 * more opened file descriptors than the standard ones.
3891 */
3892 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3893 (void) close(i);
3894 }
3895 }
3896
3897 /* Create thread quit pipe */
3898 if ((ret = init_thread_quit_pipe()) < 0) {
3899 goto error;
3900 }
3901
3902 /* Check if daemon is UID = 0 */
3903 is_root = !getuid();
3904
3905 if (is_root) {
3906 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3907
3908 /* Create global run dir with root access */
3909 ret = create_lttng_rundir(rundir);
3910 if (ret < 0) {
3911 goto error;
3912 }
3913
3914 if (strlen(apps_unix_sock_path) == 0) {
3915 snprintf(apps_unix_sock_path, PATH_MAX,
3916 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3917 }
3918
3919 if (strlen(client_unix_sock_path) == 0) {
3920 snprintf(client_unix_sock_path, PATH_MAX,
3921 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3922 }
3923
3924 /* Set global SHM for ust */
3925 if (strlen(wait_shm_path) == 0) {
3926 snprintf(wait_shm_path, PATH_MAX,
3927 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3928 }
3929
3930 if (strlen(health_unix_sock_path) == 0) {
3931 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3932 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3933 }
3934
3935 /* Setup kernel consumerd path */
3936 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3937 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3938 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3939 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3940
3941 DBG2("Kernel consumer err path: %s",
3942 kconsumer_data.err_unix_sock_path);
3943 DBG2("Kernel consumer cmd path: %s",
3944 kconsumer_data.cmd_unix_sock_path);
3945 } else {
3946 home_path = get_home_dir();
3947 if (home_path == NULL) {
3948 /* TODO: Add --socket PATH option */
3949 ERR("Can't get HOME directory for sockets creation.");
3950 ret = -EPERM;
3951 goto error;
3952 }
3953
3954 /*
3955 * Create rundir from home path. This will create something like
3956 * $HOME/.lttng
3957 */
3958 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3959 if (ret < 0) {
3960 ret = -ENOMEM;
3961 goto error;
3962 }
3963
3964 ret = create_lttng_rundir(rundir);
3965 if (ret < 0) {
3966 goto error;
3967 }
3968
3969 if (strlen(apps_unix_sock_path) == 0) {
3970 snprintf(apps_unix_sock_path, PATH_MAX,
3971 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3972 }
3973
3974 /* Set the cli tool unix socket path */
3975 if (strlen(client_unix_sock_path) == 0) {
3976 snprintf(client_unix_sock_path, PATH_MAX,
3977 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3978 }
3979
3980 /* Set global SHM for ust */
3981 if (strlen(wait_shm_path) == 0) {
3982 snprintf(wait_shm_path, PATH_MAX,
3983 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3984 }
3985
3986 /* Set health check Unix path */
3987 if (strlen(health_unix_sock_path) == 0) {
3988 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3989 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3990 }
3991 }
3992
3993 /* Set consumer initial state */
3994 kernel_consumerd_state = CONSUMER_STOPPED;
3995 ust_consumerd_state = CONSUMER_STOPPED;
3996
3997 DBG("Client socket path %s", client_unix_sock_path);
3998 DBG("Application socket path %s", apps_unix_sock_path);
3999 DBG("LTTng run directory path: %s", rundir);
4000
4001 /* 32 bits consumerd path setup */
4002 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4003 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4004 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4005 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4006
4007 DBG2("UST consumer 32 bits err path: %s",
4008 ustconsumer32_data.err_unix_sock_path);
4009 DBG2("UST consumer 32 bits cmd path: %s",
4010 ustconsumer32_data.cmd_unix_sock_path);
4011
4012 /* 64 bits consumerd path setup */
4013 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4014 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4015 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4016 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4017
4018 DBG2("UST consumer 64 bits err path: %s",
4019 ustconsumer64_data.err_unix_sock_path);
4020 DBG2("UST consumer 64 bits cmd path: %s",
4021 ustconsumer64_data.cmd_unix_sock_path);
4022
4023 /*
4024 * See if daemon already exist.
4025 */
4026 if ((ret = check_existing_daemon()) < 0) {
4027 ERR("Already running daemon.\n");
4028 /*
4029 * We do not goto exit because we must not cleanup()
4030 * because a daemon is already running.
4031 */
4032 goto error;
4033 }
4034
4035 /*
4036 * Init UST app hash table. Alloc hash table before this point since
4037 * cleanup() can get called after that point.
4038 */
4039 ust_app_ht_alloc();
4040
4041 /* After this point, we can safely call cleanup() with "goto exit" */
4042
4043 /*
4044 * These actions must be executed as root. We do that *after* setting up
4045 * the sockets path because we MUST make the check for another daemon using
4046 * those paths *before* trying to set the kernel consumer sockets and init
4047 * kernel tracer.
4048 */
4049 if (is_root) {
4050 ret = set_consumer_sockets(&kconsumer_data, rundir);
4051 if (ret < 0) {
4052 goto exit;
4053 }
4054
4055 /* Setup kernel tracer */
4056 if (!opt_no_kernel) {
4057 init_kernel_tracer();
4058 }
4059
4060 /* Set ulimit for open files */
4061 set_ulimit();
4062 }
4063 /* init lttng_fd tracking must be done after set_ulimit. */
4064 lttng_fd_init();
4065
4066 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4067 if (ret < 0) {
4068 goto exit;
4069 }
4070
4071 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4072 if (ret < 0) {
4073 goto exit;
4074 }
4075
4076 if ((ret = set_signal_handler()) < 0) {
4077 goto exit;
4078 }
4079
4080 /* Setup the needed unix socket */
4081 if ((ret = init_daemon_socket()) < 0) {
4082 goto exit;
4083 }
4084
4085 /* Set credentials to socket */
4086 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4087 goto exit;
4088 }
4089
4090 /* Get parent pid if -S, --sig-parent is specified. */
4091 if (opt_sig_parent) {
4092 ppid = getppid();
4093 }
4094
4095 /* Setup the kernel pipe for waking up the kernel thread */
4096 if (is_root && !opt_no_kernel) {
4097 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4098 goto exit;
4099 }
4100 }
4101
4102 /* Setup the thread apps communication pipe. */
4103 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4104 goto exit;
4105 }
4106
4107 /* Init UST command queue. */
4108 cds_wfq_init(&ust_cmd_queue.queue);
4109
4110 /*
4111 * Get session list pointer. This pointer MUST NOT be free(). This list is
4112 * statically declared in session.c
4113 */
4114 session_list_ptr = session_get_list();
4115
4116 /* Set up max poll set size */
4117 lttng_poll_set_max_size();
4118
4119 cmd_init();
4120
4121 /* Init all health thread counters. */
4122 health_init(&health_thread_cmd);
4123 health_init(&health_thread_kernel);
4124 health_init(&health_thread_app_manage);
4125 health_init(&health_thread_app_reg);
4126
4127 /*
4128 * Init health counters of the consumer thread. We do a quick hack here to
4129 * the state of the consumer health is fine even if the thread is not
4130 * started. Once the thread starts, the health state is updated with a poll
4131 * value to set a health code path. This is simply to ease our life and has
4132 * no cost what so ever.
4133 */
4134 health_init(&kconsumer_data.health);
4135 health_poll_update(&kconsumer_data.health);
4136 health_init(&ustconsumer32_data.health);
4137 health_poll_update(&ustconsumer32_data.health);
4138 health_init(&ustconsumer64_data.health);
4139 health_poll_update(&ustconsumer64_data.health);
4140
4141 /* Check for the application socket timeout env variable. */
4142 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4143 if (env_app_timeout) {
4144 app_socket_timeout = atoi(env_app_timeout);
4145 } else {
4146 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4147 }
4148
4149 /* Create thread to manage the client socket */
4150 ret = pthread_create(&health_thread, NULL,
4151 thread_manage_health, (void *) NULL);
4152 if (ret != 0) {
4153 PERROR("pthread_create health");
4154 goto exit_health;
4155 }
4156
4157 /* Create thread to manage the client socket */
4158 ret = pthread_create(&client_thread, NULL,
4159 thread_manage_clients, (void *) NULL);
4160 if (ret != 0) {
4161 PERROR("pthread_create clients");
4162 goto exit_client;
4163 }
4164
4165 /* Create thread to dispatch registration */
4166 ret = pthread_create(&dispatch_thread, NULL,
4167 thread_dispatch_ust_registration, (void *) NULL);
4168 if (ret != 0) {
4169 PERROR("pthread_create dispatch");
4170 goto exit_dispatch;
4171 }
4172
4173 /* Create thread to manage application registration. */
4174 ret = pthread_create(&reg_apps_thread, NULL,
4175 thread_registration_apps, (void *) NULL);
4176 if (ret != 0) {
4177 PERROR("pthread_create registration");
4178 goto exit_reg_apps;
4179 }
4180
4181 /* Create thread to manage application socket */
4182 ret = pthread_create(&apps_thread, NULL,
4183 thread_manage_apps, (void *) NULL);
4184 if (ret != 0) {
4185 PERROR("pthread_create apps");
4186 goto exit_apps;
4187 }
4188
4189 /* Don't start this thread if kernel tracing is not requested nor root */
4190 if (is_root && !opt_no_kernel) {
4191 /* Create kernel thread to manage kernel event */
4192 ret = pthread_create(&kernel_thread, NULL,
4193 thread_manage_kernel, (void *) NULL);
4194 if (ret != 0) {
4195 PERROR("pthread_create kernel");
4196 goto exit_kernel;
4197 }
4198
4199 ret = pthread_join(kernel_thread, &status);
4200 if (ret != 0) {
4201 PERROR("pthread_join");
4202 goto error; /* join error, exit without cleanup */
4203 }
4204 }
4205
4206 exit_kernel:
4207 ret = pthread_join(apps_thread, &status);
4208 if (ret != 0) {
4209 PERROR("pthread_join");
4210 goto error; /* join error, exit without cleanup */
4211 }
4212
4213 exit_apps:
4214 ret = pthread_join(reg_apps_thread, &status);
4215 if (ret != 0) {
4216 PERROR("pthread_join");
4217 goto error; /* join error, exit without cleanup */
4218 }
4219
4220 exit_reg_apps:
4221 ret = pthread_join(dispatch_thread, &status);
4222 if (ret != 0) {
4223 PERROR("pthread_join");
4224 goto error; /* join error, exit without cleanup */
4225 }
4226
4227 exit_dispatch:
4228 ret = pthread_join(client_thread, &status);
4229 if (ret != 0) {
4230 PERROR("pthread_join");
4231 goto error; /* join error, exit without cleanup */
4232 }
4233
4234 ret = join_consumer_thread(&kconsumer_data);
4235 if (ret != 0) {
4236 PERROR("join_consumer");
4237 goto error; /* join error, exit without cleanup */
4238 }
4239
4240 ret = join_consumer_thread(&ustconsumer32_data);
4241 if (ret != 0) {
4242 PERROR("join_consumer ust32");
4243 goto error; /* join error, exit without cleanup */
4244 }
4245
4246 ret = join_consumer_thread(&ustconsumer64_data);
4247 if (ret != 0) {
4248 PERROR("join_consumer ust64");
4249 goto error; /* join error, exit without cleanup */
4250 }
4251
4252 exit_client:
4253 ret = pthread_join(health_thread, &status);
4254 if (ret != 0) {
4255 PERROR("pthread_join health thread");
4256 goto error; /* join error, exit without cleanup */
4257 }
4258
4259 exit_health:
4260 exit:
4261 /*
4262 * cleanup() is called when no other thread is running.
4263 */
4264 rcu_thread_online();
4265 cleanup();
4266 rcu_thread_offline();
4267 rcu_unregister_thread();
4268 if (!ret) {
4269 exit(EXIT_SUCCESS);
4270 }
4271 error:
4272 exit(EXIT_FAILURE);
4273 }
This page took 0.118431 seconds and 4 git commands to generate.