859223a11216b1d342b0899ebdd4e66bc657d657
[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 const char *opt_pidfile;
76 static int opt_sig_parent;
77 static int opt_verbose_consumer;
78 static int opt_daemon;
79 static int opt_no_kernel;
80 static int is_root; /* Set to 1 if the daemon is running as root */
81 static pid_t ppid; /* Parent PID for --sig-parent option */
82 static char *rundir;
83
84 /*
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
87 */
88 static struct consumer_data kconsumer_data = {
89 .type = LTTNG_CONSUMER_KERNEL,
90 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
91 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
92 .err_sock = -1,
93 .cmd_sock = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer32_data = {
111 .type = LTTNG_CONSUMER32_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121
122 /* Shared between threads */
123 static int dispatch_thread_exit;
124
125 /* Global application Unix socket path */
126 static char apps_unix_sock_path[PATH_MAX];
127 /* Global client Unix socket path */
128 static char client_unix_sock_path[PATH_MAX];
129 /* global wait shm path for UST */
130 static char wait_shm_path[PATH_MAX];
131 /* Global health check unix path */
132 static char health_unix_sock_path[PATH_MAX];
133
134 /* Sockets and FDs */
135 static int client_sock = -1;
136 static int apps_sock = -1;
137 int kernel_tracer_fd = -1;
138 static int kernel_poll_pipe[2] = { -1, -1 };
139
140 /*
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
143 */
144 static int thread_quit_pipe[2] = { -1, -1 };
145
146 /*
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
149 */
150 static int apps_cmd_pipe[2] = { -1, -1 };
151
152 /* Pthread, Mutexes and Semaphores */
153 static pthread_t apps_thread;
154 static pthread_t reg_apps_thread;
155 static pthread_t client_thread;
156 static pthread_t kernel_thread;
157 static pthread_t dispatch_thread;
158 static pthread_t health_thread;
159
160 /*
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 *
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
166 */
167 static struct ust_cmd_queue ust_cmd_queue;
168
169 /*
170 * Pointer initialized before thread creation.
171 *
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
175 *
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
178 */
179 static struct ltt_session_list *session_list_ptr;
180
181 int ust_consumerd64_fd = -1;
182 int ust_consumerd32_fd = -1;
183
184 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
185 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
186 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
187 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
188
189 static const char *module_proc_lttng = "/proc/lttng";
190
191 /*
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
194 */
195 enum consumerd_state {
196 CONSUMER_STARTED = 1,
197 CONSUMER_STOPPED = 2,
198 CONSUMER_ERROR = 3,
199 };
200
201 /*
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
206 *
207 * The following example shows a possible race condition of this scheme:
208 *
209 * consumer thread error happens
210 * client cmd arrives
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
214 * ...
215 *
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
221 */
222 static enum consumerd_state ust_consumerd_state;
223 static enum consumerd_state kernel_consumerd_state;
224
225 /*
226 * Socket timeout for receiving and sending in seconds.
227 */
228 static int app_socket_timeout;
229
230 static
231 void setup_consumerd_path(void)
232 {
233 const char *bin, *libdir;
234
235 /*
236 * Allow INSTALL_BIN_PATH to be used as a target path for the
237 * native architecture size consumer if CONFIG_CONSUMER*_PATH
238 * has not been defined.
239 */
240 #if (CAA_BITS_PER_LONG == 32)
241 if (!consumerd32_bin[0]) {
242 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
243 }
244 if (!consumerd32_libdir[0]) {
245 consumerd32_libdir = INSTALL_LIB_PATH;
246 }
247 #elif (CAA_BITS_PER_LONG == 64)
248 if (!consumerd64_bin[0]) {
249 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
250 }
251 if (!consumerd64_libdir[0]) {
252 consumerd64_libdir = INSTALL_LIB_PATH;
253 }
254 #else
255 #error "Unknown bitness"
256 #endif
257
258 /*
259 * runtime env. var. overrides the build default.
260 */
261 bin = getenv("LTTNG_CONSUMERD32_BIN");
262 if (bin) {
263 consumerd32_bin = bin;
264 }
265 bin = getenv("LTTNG_CONSUMERD64_BIN");
266 if (bin) {
267 consumerd64_bin = bin;
268 }
269 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
270 if (libdir) {
271 consumerd32_libdir = libdir;
272 }
273 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
274 if (libdir) {
275 consumerd64_libdir = libdir;
276 }
277 }
278
279 /*
280 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
281 */
282 static int create_thread_poll_set(struct lttng_poll_event *events,
283 unsigned int size)
284 {
285 int ret;
286
287 if (events == NULL || size == 0) {
288 ret = -1;
289 goto error;
290 }
291
292 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
293 if (ret < 0) {
294 goto error;
295 }
296
297 /* Add quit pipe */
298 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
299 if (ret < 0) {
300 goto error;
301 }
302
303 return 0;
304
305 error:
306 return ret;
307 }
308
309 /*
310 * Check if the thread quit pipe was triggered.
311 *
312 * Return 1 if it was triggered else 0;
313 */
314 static int check_thread_quit_pipe(int fd, uint32_t events)
315 {
316 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
317 return 1;
318 }
319
320 return 0;
321 }
322
323 /*
324 * Return group ID of the tracing group or -1 if not found.
325 */
326 static gid_t allowed_group(void)
327 {
328 struct group *grp;
329
330 if (opt_tracing_group) {
331 grp = getgrnam(opt_tracing_group);
332 } else {
333 grp = getgrnam(default_tracing_group);
334 }
335 if (!grp) {
336 return -1;
337 } else {
338 return grp->gr_gid;
339 }
340 }
341
342 /*
343 * Init thread quit pipe.
344 *
345 * Return -1 on error or 0 if all pipes are created.
346 */
347 static int init_thread_quit_pipe(void)
348 {
349 int ret, i;
350
351 ret = pipe(thread_quit_pipe);
352 if (ret < 0) {
353 PERROR("thread quit pipe");
354 goto error;
355 }
356
357 for (i = 0; i < 2; i++) {
358 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
359 if (ret < 0) {
360 PERROR("fcntl");
361 goto error;
362 }
363 }
364
365 error:
366 return ret;
367 }
368
369 /*
370 * Stop all threads by closing the thread quit pipe.
371 */
372 static void stop_threads(void)
373 {
374 int ret;
375
376 /* Stopping all threads */
377 DBG("Terminating all threads");
378 ret = notify_thread_pipe(thread_quit_pipe[1]);
379 if (ret < 0) {
380 ERR("write error on thread quit pipe");
381 }
382
383 /* Dispatch thread */
384 CMM_STORE_SHARED(dispatch_thread_exit, 1);
385 futex_nto1_wake(&ust_cmd_queue.futex);
386 }
387
388 /*
389 * Cleanup the daemon
390 */
391 static void cleanup(void)
392 {
393 int ret;
394 char *cmd = NULL;
395 struct ltt_session *sess, *stmp;
396
397 DBG("Cleaning up");
398
399 /* First thing first, stop all threads */
400 utils_close_pipe(thread_quit_pipe);
401
402 /*
403 * If opt_pidfile is undefined, the default file will be wiped when
404 * removing the rundir.
405 */
406 if (opt_pidfile) {
407 ret = remove(opt_pidfile);
408 if (ret < 0) {
409 PERROR("remove pidfile %s", opt_pidfile);
410 }
411 }
412
413 DBG("Removing %s directory", rundir);
414 ret = asprintf(&cmd, "rm -rf %s", rundir);
415 if (ret < 0) {
416 ERR("asprintf failed. Something is really wrong!");
417 }
418
419 /* Remove lttng run directory */
420 ret = system(cmd);
421 if (ret < 0) {
422 ERR("Unable to clean %s", rundir);
423 }
424 free(cmd);
425 free(rundir);
426
427 DBG("Cleaning up all sessions");
428
429 /* Destroy session list mutex */
430 if (session_list_ptr != NULL) {
431 pthread_mutex_destroy(&session_list_ptr->lock);
432
433 /* Cleanup ALL session */
434 cds_list_for_each_entry_safe(sess, stmp,
435 &session_list_ptr->head, list) {
436 cmd_destroy_session(sess, kernel_poll_pipe[1]);
437 }
438 }
439
440 DBG("Closing all UST sockets");
441 ust_app_clean_list();
442
443 if (is_root && !opt_no_kernel) {
444 DBG2("Closing kernel fd");
445 if (kernel_tracer_fd >= 0) {
446 ret = close(kernel_tracer_fd);
447 if (ret) {
448 PERROR("close");
449 }
450 }
451 DBG("Unloading kernel modules");
452 modprobe_remove_lttng_all();
453 }
454
455 /* <fun> */
456 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
457 "Matthew, BEET driven development works!%c[%dm",
458 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
459 /* </fun> */
460 }
461
462 /*
463 * Send data on a unix socket using the liblttsessiondcomm API.
464 *
465 * Return lttcomm error code.
466 */
467 static int send_unix_sock(int sock, void *buf, size_t len)
468 {
469 /* Check valid length */
470 if (len == 0) {
471 return -1;
472 }
473
474 return lttcomm_send_unix_sock(sock, buf, len);
475 }
476
477 /*
478 * Free memory of a command context structure.
479 */
480 static void clean_command_ctx(struct command_ctx **cmd_ctx)
481 {
482 DBG("Clean command context structure");
483 if (*cmd_ctx) {
484 if ((*cmd_ctx)->llm) {
485 free((*cmd_ctx)->llm);
486 }
487 if ((*cmd_ctx)->lsm) {
488 free((*cmd_ctx)->lsm);
489 }
490 free(*cmd_ctx);
491 *cmd_ctx = NULL;
492 }
493 }
494
495 /*
496 * Notify UST applications using the shm mmap futex.
497 */
498 static int notify_ust_apps(int active)
499 {
500 char *wait_shm_mmap;
501
502 DBG("Notifying applications of session daemon state: %d", active);
503
504 /* See shm.c for this call implying mmap, shm and futex calls */
505 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
506 if (wait_shm_mmap == NULL) {
507 goto error;
508 }
509
510 /* Wake waiting process */
511 futex_wait_update((int32_t *) wait_shm_mmap, active);
512
513 /* Apps notified successfully */
514 return 0;
515
516 error:
517 return -1;
518 }
519
520 /*
521 * Setup the outgoing data buffer for the response (llm) by allocating the
522 * right amount of memory and copying the original information from the lsm
523 * structure.
524 *
525 * Return total size of the buffer pointed by buf.
526 */
527 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
528 {
529 int ret, buf_size;
530
531 buf_size = size;
532
533 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
534 if (cmd_ctx->llm == NULL) {
535 PERROR("zmalloc");
536 ret = -ENOMEM;
537 goto error;
538 }
539
540 /* Copy common data */
541 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
542 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
543
544 cmd_ctx->llm->data_size = size;
545 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
546
547 return buf_size;
548
549 error:
550 return ret;
551 }
552
553 /*
554 * Update the kernel poll set of all channel fd available over all tracing
555 * session. Add the wakeup pipe at the end of the set.
556 */
557 static int update_kernel_poll(struct lttng_poll_event *events)
558 {
559 int ret;
560 struct ltt_session *session;
561 struct ltt_kernel_channel *channel;
562
563 DBG("Updating kernel poll set");
564
565 session_lock_list();
566 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
567 session_lock(session);
568 if (session->kernel_session == NULL) {
569 session_unlock(session);
570 continue;
571 }
572
573 cds_list_for_each_entry(channel,
574 &session->kernel_session->channel_list.head, list) {
575 /* Add channel fd to the kernel poll set */
576 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
577 if (ret < 0) {
578 session_unlock(session);
579 goto error;
580 }
581 DBG("Channel fd %d added to kernel set", channel->fd);
582 }
583 session_unlock(session);
584 }
585 session_unlock_list();
586
587 return 0;
588
589 error:
590 session_unlock_list();
591 return -1;
592 }
593
594 /*
595 * Find the channel fd from 'fd' over all tracing session. When found, check
596 * for new channel stream and send those stream fds to the kernel consumer.
597 *
598 * Useful for CPU hotplug feature.
599 */
600 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
601 {
602 int ret = 0;
603 struct ltt_session *session;
604 struct ltt_kernel_session *ksess;
605 struct ltt_kernel_channel *channel;
606
607 DBG("Updating kernel streams for channel fd %d", fd);
608
609 session_lock_list();
610 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
611 session_lock(session);
612 if (session->kernel_session == NULL) {
613 session_unlock(session);
614 continue;
615 }
616 ksess = session->kernel_session;
617
618 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
619 if (channel->fd == fd) {
620 DBG("Channel found, updating kernel streams");
621 ret = kernel_open_channel_stream(channel);
622 if (ret < 0) {
623 goto error;
624 }
625
626 /*
627 * Have we already sent fds to the consumer? If yes, it means
628 * that tracing is started so it is safe to send our updated
629 * stream fds.
630 */
631 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
632 struct lttng_ht_iter iter;
633 struct consumer_socket *socket;
634
635 rcu_read_lock();
636 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
637 &iter.iter, socket, node.node) {
638 /* Code flow error */
639 assert(socket->fd >= 0);
640
641 pthread_mutex_lock(socket->lock);
642 ret = kernel_consumer_send_channel_stream(socket,
643 channel, ksess);
644 pthread_mutex_unlock(socket->lock);
645 if (ret < 0) {
646 rcu_read_unlock();
647 goto error;
648 }
649 }
650 rcu_read_unlock();
651 }
652 goto error;
653 }
654 }
655 session_unlock(session);
656 }
657 session_unlock_list();
658 return ret;
659
660 error:
661 session_unlock(session);
662 session_unlock_list();
663 return ret;
664 }
665
666 /*
667 * For each tracing session, update newly registered apps.
668 */
669 static void update_ust_app(int app_sock)
670 {
671 struct ltt_session *sess, *stmp;
672
673 session_lock_list();
674
675 /* For all tracing session(s) */
676 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
677 session_lock(sess);
678 if (sess->ust_session) {
679 ust_app_global_update(sess->ust_session, app_sock);
680 }
681 session_unlock(sess);
682 }
683
684 session_unlock_list();
685 }
686
687 /*
688 * This thread manage event coming from the kernel.
689 *
690 * Features supported in this thread:
691 * -) CPU Hotplug
692 */
693 static void *thread_manage_kernel(void *data)
694 {
695 int ret, i, pollfd, update_poll_flag = 1, err = -1;
696 uint32_t revents, nb_fd;
697 char tmp;
698 struct lttng_poll_event events;
699
700 DBG("[thread] Thread manage kernel started");
701
702 health_register(HEALTH_TYPE_KERNEL);
703
704 /*
705 * This first step of the while is to clean this structure which could free
706 * non NULL pointers so zero it before the loop.
707 */
708 memset(&events, 0, sizeof(events));
709
710 if (testpoint(thread_manage_kernel)) {
711 goto error_testpoint;
712 }
713
714 health_code_update();
715
716 if (testpoint(thread_manage_kernel_before_loop)) {
717 goto error_testpoint;
718 }
719
720 while (1) {
721 health_code_update();
722
723 if (update_poll_flag == 1) {
724 /* Clean events object. We are about to populate it again. */
725 lttng_poll_clean(&events);
726
727 ret = create_thread_poll_set(&events, 2);
728 if (ret < 0) {
729 goto error_poll_create;
730 }
731
732 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
733 if (ret < 0) {
734 goto error;
735 }
736
737 /* This will add the available kernel channel if any. */
738 ret = update_kernel_poll(&events);
739 if (ret < 0) {
740 goto error;
741 }
742 update_poll_flag = 0;
743 }
744
745 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
746
747 /* Poll infinite value of time */
748 restart:
749 health_poll_update();
750 ret = lttng_poll_wait(&events, -1);
751 health_poll_update();
752 if (ret < 0) {
753 /*
754 * Restart interrupted system call.
755 */
756 if (errno == EINTR) {
757 goto restart;
758 }
759 goto error;
760 } else if (ret == 0) {
761 /* Should not happen since timeout is infinite */
762 ERR("Return value of poll is 0 with an infinite timeout.\n"
763 "This should not have happened! Continuing...");
764 continue;
765 }
766
767 nb_fd = ret;
768
769 for (i = 0; i < nb_fd; i++) {
770 /* Fetch once the poll data */
771 revents = LTTNG_POLL_GETEV(&events, i);
772 pollfd = LTTNG_POLL_GETFD(&events, i);
773
774 health_code_update();
775
776 /* Thread quit pipe has been closed. Killing thread. */
777 ret = check_thread_quit_pipe(pollfd, revents);
778 if (ret) {
779 err = 0;
780 goto exit;
781 }
782
783 /* Check for data on kernel pipe */
784 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
785 do {
786 ret = read(kernel_poll_pipe[0], &tmp, 1);
787 } while (ret < 0 && errno == EINTR);
788 /*
789 * Ret value is useless here, if this pipe gets any actions an
790 * update is required anyway.
791 */
792 update_poll_flag = 1;
793 continue;
794 } else {
795 /*
796 * New CPU detected by the kernel. Adding kernel stream to
797 * kernel session and updating the kernel consumer
798 */
799 if (revents & LPOLLIN) {
800 ret = update_kernel_stream(&kconsumer_data, pollfd);
801 if (ret < 0) {
802 continue;
803 }
804 break;
805 /*
806 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
807 * and unregister kernel stream at this point.
808 */
809 }
810 }
811 }
812 }
813
814 exit:
815 error:
816 lttng_poll_clean(&events);
817 error_poll_create:
818 error_testpoint:
819 utils_close_pipe(kernel_poll_pipe);
820 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
821 if (err) {
822 health_error();
823 ERR("Health error occurred in %s", __func__);
824 WARN("Kernel thread died unexpectedly. "
825 "Kernel tracing can continue but CPU hotplug is disabled.");
826 }
827 health_unregister();
828 DBG("Kernel thread dying");
829 return NULL;
830 }
831
832 /*
833 * Signal pthread condition of the consumer data that the thread.
834 */
835 static void signal_consumer_condition(struct consumer_data *data, int state)
836 {
837 pthread_mutex_lock(&data->cond_mutex);
838
839 /*
840 * The state is set before signaling. It can be any value, it's the waiter
841 * job to correctly interpret this condition variable associated to the
842 * consumer pthread_cond.
843 *
844 * A value of 0 means that the corresponding thread of the consumer data
845 * was not started. 1 indicates that the thread has started and is ready
846 * for action. A negative value means that there was an error during the
847 * thread bootstrap.
848 */
849 data->consumer_thread_is_ready = state;
850 (void) pthread_cond_signal(&data->cond);
851
852 pthread_mutex_unlock(&data->cond_mutex);
853 }
854
855 /*
856 * This thread manage the consumer error sent back to the session daemon.
857 */
858 static void *thread_manage_consumer(void *data)
859 {
860 int sock = -1, i, ret, pollfd, err = -1;
861 uint32_t revents, nb_fd;
862 enum lttcomm_return_code code;
863 struct lttng_poll_event events;
864 struct consumer_data *consumer_data = data;
865
866 DBG("[thread] Manage consumer started");
867
868 health_register(HEALTH_TYPE_CONSUMER);
869
870 /*
871 * Since the consumer thread can be spawned at any moment in time, we init
872 * the health to a poll status (1, which is a valid health over time).
873 * When the thread starts, we update here the health to a "code" path being
874 * an even value so this thread, when reaching a poll wait, does not
875 * trigger an error with an even value.
876 *
877 * Here is the use case we avoid.
878 *
879 * +1: the first poll update during initialization (main())
880 * +2 * x: multiple code update once in this thread.
881 * +1: poll wait in this thread (being a good health state).
882 * == even number which after the wait period shows as a bad health.
883 *
884 * In a nutshell, the following poll update to the health state brings back
885 * the state to an even value meaning a code path.
886 */
887 health_poll_update();
888
889 /*
890 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
891 * Nothing more will be added to this poll set.
892 */
893 ret = create_thread_poll_set(&events, 2);
894 if (ret < 0) {
895 goto error_poll;
896 }
897
898 /*
899 * The error socket here is already in a listening state which was done
900 * just before spawning this thread to avoid a race between the consumer
901 * daemon exec trying to connect and the listen() call.
902 */
903 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
904 if (ret < 0) {
905 goto error;
906 }
907
908 health_code_update();
909
910 /* Inifinite blocking call, waiting for transmission */
911 restart:
912 health_poll_update();
913
914 if (testpoint(thread_manage_consumer)) {
915 goto error;
916 }
917
918 ret = lttng_poll_wait(&events, -1);
919 health_poll_update();
920 if (ret < 0) {
921 /*
922 * Restart interrupted system call.
923 */
924 if (errno == EINTR) {
925 goto restart;
926 }
927 goto error;
928 }
929
930 nb_fd = ret;
931
932 for (i = 0; i < nb_fd; i++) {
933 /* Fetch once the poll data */
934 revents = LTTNG_POLL_GETEV(&events, i);
935 pollfd = LTTNG_POLL_GETFD(&events, i);
936
937 health_code_update();
938
939 /* Thread quit pipe has been closed. Killing thread. */
940 ret = check_thread_quit_pipe(pollfd, revents);
941 if (ret) {
942 err = 0;
943 goto exit;
944 }
945
946 /* Event on the registration socket */
947 if (pollfd == consumer_data->err_sock) {
948 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
949 ERR("consumer err socket poll error");
950 goto error;
951 }
952 }
953 }
954
955 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
956 if (sock < 0) {
957 goto error;
958 }
959
960 /*
961 * Set the CLOEXEC flag. Return code is useless because either way, the
962 * show must go on.
963 */
964 (void) utils_set_fd_cloexec(sock);
965
966 health_code_update();
967
968 DBG2("Receiving code from consumer err_sock");
969
970 /* Getting status code from kconsumerd */
971 ret = lttcomm_recv_unix_sock(sock, &code,
972 sizeof(enum lttcomm_return_code));
973 if (ret <= 0) {
974 goto error;
975 }
976
977 health_code_update();
978
979 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
980 consumer_data->cmd_sock =
981 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
982 if (consumer_data->cmd_sock < 0) {
983 /* On error, signal condition and quit. */
984 signal_consumer_condition(consumer_data, -1);
985 PERROR("consumer connect");
986 goto error;
987 }
988 signal_consumer_condition(consumer_data, 1);
989 DBG("Consumer command socket ready");
990 } else {
991 ERR("consumer error when waiting for SOCK_READY : %s",
992 lttcomm_get_readable_code(-code));
993 goto error;
994 }
995
996 /* Remove the kconsumerd error sock since we've established a connexion */
997 ret = lttng_poll_del(&events, consumer_data->err_sock);
998 if (ret < 0) {
999 goto error;
1000 }
1001
1002 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1003 if (ret < 0) {
1004 goto error;
1005 }
1006
1007 health_code_update();
1008
1009 /* Inifinite blocking call, waiting for transmission */
1010 restart_poll:
1011 health_poll_update();
1012 ret = lttng_poll_wait(&events, -1);
1013 health_poll_update();
1014 if (ret < 0) {
1015 /*
1016 * Restart interrupted system call.
1017 */
1018 if (errno == EINTR) {
1019 goto restart_poll;
1020 }
1021 goto error;
1022 }
1023
1024 nb_fd = ret;
1025
1026 for (i = 0; i < nb_fd; i++) {
1027 /* Fetch once the poll data */
1028 revents = LTTNG_POLL_GETEV(&events, i);
1029 pollfd = LTTNG_POLL_GETFD(&events, i);
1030
1031 health_code_update();
1032
1033 /* Thread quit pipe has been closed. Killing thread. */
1034 ret = check_thread_quit_pipe(pollfd, revents);
1035 if (ret) {
1036 err = 0;
1037 goto exit;
1038 }
1039
1040 /* Event on the kconsumerd socket */
1041 if (pollfd == sock) {
1042 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1043 ERR("consumer err socket second poll error");
1044 goto error;
1045 }
1046 }
1047 }
1048
1049 health_code_update();
1050
1051 /* Wait for any kconsumerd error */
1052 ret = lttcomm_recv_unix_sock(sock, &code,
1053 sizeof(enum lttcomm_return_code));
1054 if (ret <= 0) {
1055 ERR("consumer closed the command socket");
1056 goto error;
1057 }
1058
1059 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1060
1061 exit:
1062 error:
1063 /* Immediately set the consumerd state to stopped */
1064 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1065 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1066 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1067 consumer_data->type == LTTNG_CONSUMER32_UST) {
1068 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1069 } else {
1070 /* Code flow error... */
1071 assert(0);
1072 }
1073
1074 if (consumer_data->err_sock >= 0) {
1075 ret = close(consumer_data->err_sock);
1076 if (ret) {
1077 PERROR("close");
1078 }
1079 }
1080 if (consumer_data->cmd_sock >= 0) {
1081 ret = close(consumer_data->cmd_sock);
1082 if (ret) {
1083 PERROR("close");
1084 }
1085 }
1086 if (sock >= 0) {
1087 ret = close(sock);
1088 if (ret) {
1089 PERROR("close");
1090 }
1091 }
1092
1093 unlink(consumer_data->err_unix_sock_path);
1094 unlink(consumer_data->cmd_unix_sock_path);
1095 consumer_data->pid = 0;
1096
1097 lttng_poll_clean(&events);
1098 error_poll:
1099 if (err) {
1100 health_error();
1101 ERR("Health error occurred in %s", __func__);
1102 }
1103 health_unregister();
1104 DBG("consumer thread cleanup completed");
1105
1106 return NULL;
1107 }
1108
1109 /*
1110 * This thread manage application communication.
1111 */
1112 static void *thread_manage_apps(void *data)
1113 {
1114 int i, ret, pollfd, err = -1;
1115 uint32_t revents, nb_fd;
1116 struct ust_command ust_cmd;
1117 struct lttng_poll_event events;
1118
1119 DBG("[thread] Manage application started");
1120
1121 rcu_register_thread();
1122 rcu_thread_online();
1123
1124 health_register(HEALTH_TYPE_APP_MANAGE);
1125
1126 if (testpoint(thread_manage_apps)) {
1127 goto error_testpoint;
1128 }
1129
1130 health_code_update();
1131
1132 ret = create_thread_poll_set(&events, 2);
1133 if (ret < 0) {
1134 goto error_poll_create;
1135 }
1136
1137 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1138 if (ret < 0) {
1139 goto error;
1140 }
1141
1142 if (testpoint(thread_manage_apps_before_loop)) {
1143 goto error;
1144 }
1145
1146 health_code_update();
1147
1148 while (1) {
1149 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1150
1151 /* Inifinite blocking call, waiting for transmission */
1152 restart:
1153 health_poll_update();
1154 ret = lttng_poll_wait(&events, -1);
1155 health_poll_update();
1156 if (ret < 0) {
1157 /*
1158 * Restart interrupted system call.
1159 */
1160 if (errno == EINTR) {
1161 goto restart;
1162 }
1163 goto error;
1164 }
1165
1166 nb_fd = ret;
1167
1168 for (i = 0; i < nb_fd; i++) {
1169 /* Fetch once the poll data */
1170 revents = LTTNG_POLL_GETEV(&events, i);
1171 pollfd = LTTNG_POLL_GETFD(&events, i);
1172
1173 health_code_update();
1174
1175 /* Thread quit pipe has been closed. Killing thread. */
1176 ret = check_thread_quit_pipe(pollfd, revents);
1177 if (ret) {
1178 err = 0;
1179 goto exit;
1180 }
1181
1182 /* Inspect the apps cmd pipe */
1183 if (pollfd == apps_cmd_pipe[0]) {
1184 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1185 ERR("Apps command pipe error");
1186 goto error;
1187 } else if (revents & LPOLLIN) {
1188 /* Empty pipe */
1189 do {
1190 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1191 } while (ret < 0 && errno == EINTR);
1192 if (ret < 0 || ret < sizeof(ust_cmd)) {
1193 PERROR("read apps cmd pipe");
1194 goto error;
1195 }
1196
1197 health_code_update();
1198
1199 /* Register applicaton to the session daemon */
1200 ret = ust_app_register(&ust_cmd.reg_msg,
1201 ust_cmd.sock);
1202 if (ret == -ENOMEM) {
1203 goto error;
1204 } else if (ret < 0) {
1205 break;
1206 }
1207
1208 health_code_update();
1209
1210 /*
1211 * Validate UST version compatibility.
1212 */
1213 ret = ust_app_validate_version(ust_cmd.sock);
1214 if (ret >= 0) {
1215 /*
1216 * Add channel(s) and event(s) to newly registered apps
1217 * from lttng global UST domain.
1218 */
1219 update_ust_app(ust_cmd.sock);
1220 }
1221
1222 health_code_update();
1223
1224 ret = ust_app_register_done(ust_cmd.sock);
1225 if (ret < 0) {
1226 /*
1227 * If the registration is not possible, we simply
1228 * unregister the apps and continue
1229 */
1230 ust_app_unregister(ust_cmd.sock);
1231 } else {
1232 /*
1233 * We only monitor the error events of the socket. This
1234 * thread does not handle any incoming data from UST
1235 * (POLLIN).
1236 */
1237 ret = lttng_poll_add(&events, ust_cmd.sock,
1238 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1239 if (ret < 0) {
1240 goto error;
1241 }
1242
1243 /* Set socket timeout for both receiving and ending */
1244 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1245 app_socket_timeout);
1246 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1247 app_socket_timeout);
1248
1249 DBG("Apps with sock %d added to poll set",
1250 ust_cmd.sock);
1251 }
1252
1253 health_code_update();
1254
1255 break;
1256 }
1257 } else {
1258 /*
1259 * At this point, we know that a registered application made
1260 * the event at poll_wait.
1261 */
1262 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1263 /* Removing from the poll set */
1264 ret = lttng_poll_del(&events, pollfd);
1265 if (ret < 0) {
1266 goto error;
1267 }
1268
1269 /* Socket closed on remote end. */
1270 ust_app_unregister(pollfd);
1271 break;
1272 }
1273 }
1274
1275 health_code_update();
1276 }
1277 }
1278
1279 exit:
1280 error:
1281 lttng_poll_clean(&events);
1282 error_poll_create:
1283 error_testpoint:
1284 utils_close_pipe(apps_cmd_pipe);
1285 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1286
1287 /*
1288 * We don't clean the UST app hash table here since already registered
1289 * applications can still be controlled so let them be until the session
1290 * daemon dies or the applications stop.
1291 */
1292
1293 if (err) {
1294 health_error();
1295 ERR("Health error occurred in %s", __func__);
1296 }
1297 health_unregister();
1298 DBG("Application communication apps thread cleanup complete");
1299 rcu_thread_offline();
1300 rcu_unregister_thread();
1301 return NULL;
1302 }
1303
1304 /*
1305 * Dispatch request from the registration threads to the application
1306 * communication thread.
1307 */
1308 static void *thread_dispatch_ust_registration(void *data)
1309 {
1310 int ret;
1311 struct cds_wfq_node *node;
1312 struct ust_command *ust_cmd = NULL;
1313
1314 DBG("[thread] Dispatch UST command started");
1315
1316 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1317 /* Atomically prepare the queue futex */
1318 futex_nto1_prepare(&ust_cmd_queue.futex);
1319
1320 do {
1321 /* Dequeue command for registration */
1322 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1323 if (node == NULL) {
1324 DBG("Woken up but nothing in the UST command queue");
1325 /* Continue thread execution */
1326 break;
1327 }
1328
1329 ust_cmd = caa_container_of(node, struct ust_command, node);
1330
1331 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1332 " gid:%d sock:%d name:%s (version %d.%d)",
1333 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1334 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1335 ust_cmd->sock, ust_cmd->reg_msg.name,
1336 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1337 /*
1338 * Inform apps thread of the new application registration. This
1339 * call is blocking so we can be assured that the data will be read
1340 * at some point in time or wait to the end of the world :)
1341 */
1342 if (apps_cmd_pipe[1] >= 0) {
1343 do {
1344 ret = write(apps_cmd_pipe[1], ust_cmd,
1345 sizeof(struct ust_command));
1346 } while (ret < 0 && errno == EINTR);
1347 if (ret < 0 || ret != sizeof(struct ust_command)) {
1348 PERROR("write apps cmd pipe");
1349 if (errno == EBADF) {
1350 /*
1351 * We can't inform the application thread to process
1352 * registration. We will exit or else application
1353 * registration will not occur and tracing will never
1354 * start.
1355 */
1356 goto error;
1357 }
1358 }
1359 } else {
1360 /* Application manager thread is not available. */
1361 ret = close(ust_cmd->sock);
1362 if (ret < 0) {
1363 PERROR("close ust_cmd sock");
1364 }
1365 }
1366 free(ust_cmd);
1367 } while (node != NULL);
1368
1369 /* Futex wait on queue. Blocking call on futex() */
1370 futex_nto1_wait(&ust_cmd_queue.futex);
1371 }
1372
1373 error:
1374 DBG("Dispatch thread dying");
1375 return NULL;
1376 }
1377
1378 /*
1379 * This thread manage application registration.
1380 */
1381 static void *thread_registration_apps(void *data)
1382 {
1383 int sock = -1, i, ret, pollfd, err = -1;
1384 uint32_t revents, nb_fd;
1385 struct lttng_poll_event events;
1386 /*
1387 * Get allocated in this thread, enqueued to a global queue, dequeued and
1388 * freed in the manage apps thread.
1389 */
1390 struct ust_command *ust_cmd = NULL;
1391
1392 DBG("[thread] Manage application registration started");
1393
1394 health_register(HEALTH_TYPE_APP_REG);
1395
1396 if (testpoint(thread_registration_apps)) {
1397 goto error_testpoint;
1398 }
1399
1400 ret = lttcomm_listen_unix_sock(apps_sock);
1401 if (ret < 0) {
1402 goto error_listen;
1403 }
1404
1405 /*
1406 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1407 * more will be added to this poll set.
1408 */
1409 ret = create_thread_poll_set(&events, 2);
1410 if (ret < 0) {
1411 goto error_create_poll;
1412 }
1413
1414 /* Add the application registration socket */
1415 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1416 if (ret < 0) {
1417 goto error_poll_add;
1418 }
1419
1420 /* Notify all applications to register */
1421 ret = notify_ust_apps(1);
1422 if (ret < 0) {
1423 ERR("Failed to notify applications or create the wait shared memory.\n"
1424 "Execution continues but there might be problem for already\n"
1425 "running applications that wishes to register.");
1426 }
1427
1428 while (1) {
1429 DBG("Accepting application registration");
1430
1431 /* Inifinite blocking call, waiting for transmission */
1432 restart:
1433 health_poll_update();
1434 ret = lttng_poll_wait(&events, -1);
1435 health_poll_update();
1436 if (ret < 0) {
1437 /*
1438 * Restart interrupted system call.
1439 */
1440 if (errno == EINTR) {
1441 goto restart;
1442 }
1443 goto error;
1444 }
1445
1446 nb_fd = ret;
1447
1448 for (i = 0; i < nb_fd; i++) {
1449 health_code_update();
1450
1451 /* Fetch once the poll data */
1452 revents = LTTNG_POLL_GETEV(&events, i);
1453 pollfd = LTTNG_POLL_GETFD(&events, i);
1454
1455 /* Thread quit pipe has been closed. Killing thread. */
1456 ret = check_thread_quit_pipe(pollfd, revents);
1457 if (ret) {
1458 err = 0;
1459 goto exit;
1460 }
1461
1462 /* Event on the registration socket */
1463 if (pollfd == apps_sock) {
1464 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1465 ERR("Register apps socket poll error");
1466 goto error;
1467 } else if (revents & LPOLLIN) {
1468 sock = lttcomm_accept_unix_sock(apps_sock);
1469 if (sock < 0) {
1470 goto error;
1471 }
1472
1473 /*
1474 * Set the CLOEXEC flag. Return code is useless because
1475 * either way, the show must go on.
1476 */
1477 (void) utils_set_fd_cloexec(sock);
1478
1479 /* Create UST registration command for enqueuing */
1480 ust_cmd = zmalloc(sizeof(struct ust_command));
1481 if (ust_cmd == NULL) {
1482 PERROR("ust command zmalloc");
1483 goto error;
1484 }
1485
1486 /*
1487 * Using message-based transmissions to ensure we don't
1488 * have to deal with partially received messages.
1489 */
1490 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1491 if (ret < 0) {
1492 ERR("Exhausted file descriptors allowed for applications.");
1493 free(ust_cmd);
1494 ret = close(sock);
1495 if (ret) {
1496 PERROR("close");
1497 }
1498 sock = -1;
1499 continue;
1500 }
1501 health_code_update();
1502 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1503 sizeof(struct ust_register_msg));
1504 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1505 if (ret < 0) {
1506 PERROR("lttcomm_recv_unix_sock register apps");
1507 } else {
1508 ERR("Wrong size received on apps register");
1509 }
1510 free(ust_cmd);
1511 ret = close(sock);
1512 if (ret) {
1513 PERROR("close");
1514 }
1515 lttng_fd_put(LTTNG_FD_APPS, 1);
1516 sock = -1;
1517 continue;
1518 }
1519 health_code_update();
1520
1521 ust_cmd->sock = sock;
1522 sock = -1;
1523
1524 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1525 " gid:%d sock:%d name:%s (version %d.%d)",
1526 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1527 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1528 ust_cmd->sock, ust_cmd->reg_msg.name,
1529 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1530
1531 /*
1532 * Lock free enqueue the registration request. The red pill
1533 * has been taken! This apps will be part of the *system*.
1534 */
1535 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1536
1537 /*
1538 * Wake the registration queue futex. Implicit memory
1539 * barrier with the exchange in cds_wfq_enqueue.
1540 */
1541 futex_nto1_wake(&ust_cmd_queue.futex);
1542 }
1543 }
1544 }
1545 }
1546
1547 exit:
1548 error:
1549 if (err) {
1550 health_error();
1551 ERR("Health error occurred in %s", __func__);
1552 }
1553
1554 /* Notify that the registration thread is gone */
1555 notify_ust_apps(0);
1556
1557 if (apps_sock >= 0) {
1558 ret = close(apps_sock);
1559 if (ret) {
1560 PERROR("close");
1561 }
1562 }
1563 if (sock >= 0) {
1564 ret = close(sock);
1565 if (ret) {
1566 PERROR("close");
1567 }
1568 lttng_fd_put(LTTNG_FD_APPS, 1);
1569 }
1570 unlink(apps_unix_sock_path);
1571
1572 error_poll_add:
1573 lttng_poll_clean(&events);
1574 error_listen:
1575 error_create_poll:
1576 error_testpoint:
1577 DBG("UST Registration thread cleanup complete");
1578 health_unregister();
1579
1580 return NULL;
1581 }
1582
1583 /*
1584 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1585 * exec or it will fails.
1586 */
1587 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1588 {
1589 int ret, clock_ret;
1590 struct timespec timeout;
1591
1592 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1593 consumer_data->consumer_thread_is_ready = 0;
1594
1595 /* Setup pthread condition */
1596 ret = pthread_condattr_init(&consumer_data->condattr);
1597 if (ret != 0) {
1598 errno = ret;
1599 PERROR("pthread_condattr_init consumer data");
1600 goto error;
1601 }
1602
1603 /*
1604 * Set the monotonic clock in order to make sure we DO NOT jump in time
1605 * between the clock_gettime() call and the timedwait call. See bug #324
1606 * for a more details and how we noticed it.
1607 */
1608 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1609 if (ret != 0) {
1610 errno = ret;
1611 PERROR("pthread_condattr_setclock consumer data");
1612 goto error;
1613 }
1614
1615 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1616 if (ret != 0) {
1617 errno = ret;
1618 PERROR("pthread_cond_init consumer data");
1619 goto error;
1620 }
1621
1622 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1623 consumer_data);
1624 if (ret != 0) {
1625 PERROR("pthread_create consumer");
1626 ret = -1;
1627 goto error;
1628 }
1629
1630 /* We are about to wait on a pthread condition */
1631 pthread_mutex_lock(&consumer_data->cond_mutex);
1632
1633 /* Get time for sem_timedwait absolute timeout */
1634 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1635 /*
1636 * Set the timeout for the condition timed wait even if the clock gettime
1637 * call fails since we might loop on that call and we want to avoid to
1638 * increment the timeout too many times.
1639 */
1640 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1641
1642 /*
1643 * The following loop COULD be skipped in some conditions so this is why we
1644 * set ret to 0 in order to make sure at least one round of the loop is
1645 * done.
1646 */
1647 ret = 0;
1648
1649 /*
1650 * Loop until the condition is reached or when a timeout is reached. Note
1651 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1652 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1653 * possible. This loop does not take any chances and works with both of
1654 * them.
1655 */
1656 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1657 if (clock_ret < 0) {
1658 PERROR("clock_gettime spawn consumer");
1659 /* Infinite wait for the consumerd thread to be ready */
1660 ret = pthread_cond_wait(&consumer_data->cond,
1661 &consumer_data->cond_mutex);
1662 } else {
1663 ret = pthread_cond_timedwait(&consumer_data->cond,
1664 &consumer_data->cond_mutex, &timeout);
1665 }
1666 }
1667
1668 /* Release the pthread condition */
1669 pthread_mutex_unlock(&consumer_data->cond_mutex);
1670
1671 if (ret != 0) {
1672 errno = ret;
1673 if (ret == ETIMEDOUT) {
1674 /*
1675 * Call has timed out so we kill the kconsumerd_thread and return
1676 * an error.
1677 */
1678 ERR("Condition timed out. The consumer thread was never ready."
1679 " Killing it");
1680 ret = pthread_cancel(consumer_data->thread);
1681 if (ret < 0) {
1682 PERROR("pthread_cancel consumer thread");
1683 }
1684 } else {
1685 PERROR("pthread_cond_wait failed consumer thread");
1686 }
1687 goto error;
1688 }
1689
1690 pthread_mutex_lock(&consumer_data->pid_mutex);
1691 if (consumer_data->pid == 0) {
1692 ERR("Consumerd did not start");
1693 pthread_mutex_unlock(&consumer_data->pid_mutex);
1694 goto error;
1695 }
1696 pthread_mutex_unlock(&consumer_data->pid_mutex);
1697
1698 return 0;
1699
1700 error:
1701 return ret;
1702 }
1703
1704 /*
1705 * Join consumer thread
1706 */
1707 static int join_consumer_thread(struct consumer_data *consumer_data)
1708 {
1709 void *status;
1710
1711 /* Consumer pid must be a real one. */
1712 if (consumer_data->pid > 0) {
1713 int ret;
1714 ret = kill(consumer_data->pid, SIGTERM);
1715 if (ret) {
1716 ERR("Error killing consumer daemon");
1717 return ret;
1718 }
1719 return pthread_join(consumer_data->thread, &status);
1720 } else {
1721 return 0;
1722 }
1723 }
1724
1725 /*
1726 * Fork and exec a consumer daemon (consumerd).
1727 *
1728 * Return pid if successful else -1.
1729 */
1730 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1731 {
1732 int ret;
1733 pid_t pid;
1734 const char *consumer_to_use;
1735 const char *verbosity;
1736 struct stat st;
1737
1738 DBG("Spawning consumerd");
1739
1740 pid = fork();
1741 if (pid == 0) {
1742 /*
1743 * Exec consumerd.
1744 */
1745 if (opt_verbose_consumer) {
1746 verbosity = "--verbose";
1747 } else {
1748 verbosity = "--quiet";
1749 }
1750 switch (consumer_data->type) {
1751 case LTTNG_CONSUMER_KERNEL:
1752 /*
1753 * Find out which consumerd to execute. We will first try the
1754 * 64-bit path, then the sessiond's installation directory, and
1755 * fallback on the 32-bit one,
1756 */
1757 DBG3("Looking for a kernel consumer at these locations:");
1758 DBG3(" 1) %s", consumerd64_bin);
1759 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1760 DBG3(" 3) %s", consumerd32_bin);
1761 if (stat(consumerd64_bin, &st) == 0) {
1762 DBG3("Found location #1");
1763 consumer_to_use = consumerd64_bin;
1764 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1765 DBG3("Found location #2");
1766 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1767 } else if (stat(consumerd32_bin, &st) == 0) {
1768 DBG3("Found location #3");
1769 consumer_to_use = consumerd32_bin;
1770 } else {
1771 DBG("Could not find any valid consumerd executable");
1772 break;
1773 }
1774 DBG("Using kernel consumer at: %s", consumer_to_use);
1775 execl(consumer_to_use,
1776 "lttng-consumerd", verbosity, "-k",
1777 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1778 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1779 NULL);
1780 break;
1781 case LTTNG_CONSUMER64_UST:
1782 {
1783 char *tmpnew = NULL;
1784
1785 if (consumerd64_libdir[0] != '\0') {
1786 char *tmp;
1787 size_t tmplen;
1788
1789 tmp = getenv("LD_LIBRARY_PATH");
1790 if (!tmp) {
1791 tmp = "";
1792 }
1793 tmplen = strlen("LD_LIBRARY_PATH=")
1794 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1795 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1796 if (!tmpnew) {
1797 ret = -ENOMEM;
1798 goto error;
1799 }
1800 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1801 strcat(tmpnew, consumerd64_libdir);
1802 if (tmp[0] != '\0') {
1803 strcat(tmpnew, ":");
1804 strcat(tmpnew, tmp);
1805 }
1806 ret = putenv(tmpnew);
1807 if (ret) {
1808 ret = -errno;
1809 goto error;
1810 }
1811 }
1812 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1813 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1814 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1815 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1816 NULL);
1817 if (consumerd64_libdir[0] != '\0') {
1818 free(tmpnew);
1819 }
1820 if (ret) {
1821 goto error;
1822 }
1823 break;
1824 }
1825 case LTTNG_CONSUMER32_UST:
1826 {
1827 char *tmpnew = NULL;
1828
1829 if (consumerd32_libdir[0] != '\0') {
1830 char *tmp;
1831 size_t tmplen;
1832
1833 tmp = getenv("LD_LIBRARY_PATH");
1834 if (!tmp) {
1835 tmp = "";
1836 }
1837 tmplen = strlen("LD_LIBRARY_PATH=")
1838 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1839 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1840 if (!tmpnew) {
1841 ret = -ENOMEM;
1842 goto error;
1843 }
1844 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1845 strcat(tmpnew, consumerd32_libdir);
1846 if (tmp[0] != '\0') {
1847 strcat(tmpnew, ":");
1848 strcat(tmpnew, tmp);
1849 }
1850 ret = putenv(tmpnew);
1851 if (ret) {
1852 ret = -errno;
1853 goto error;
1854 }
1855 }
1856 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1857 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1858 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1859 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1860 NULL);
1861 if (consumerd32_libdir[0] != '\0') {
1862 free(tmpnew);
1863 }
1864 if (ret) {
1865 goto error;
1866 }
1867 break;
1868 }
1869 default:
1870 PERROR("unknown consumer type");
1871 exit(EXIT_FAILURE);
1872 }
1873 if (errno != 0) {
1874 PERROR("kernel start consumer exec");
1875 }
1876 exit(EXIT_FAILURE);
1877 } else if (pid > 0) {
1878 ret = pid;
1879 } else {
1880 PERROR("start consumer fork");
1881 ret = -errno;
1882 }
1883 error:
1884 return ret;
1885 }
1886
1887 /*
1888 * Spawn the consumerd daemon and session daemon thread.
1889 */
1890 static int start_consumerd(struct consumer_data *consumer_data)
1891 {
1892 int ret;
1893
1894 /*
1895 * Set the listen() state on the socket since there is a possible race
1896 * between the exec() of the consumer daemon and this call if place in the
1897 * consumer thread. See bug #366 for more details.
1898 */
1899 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1900 if (ret < 0) {
1901 goto error;
1902 }
1903
1904 pthread_mutex_lock(&consumer_data->pid_mutex);
1905 if (consumer_data->pid != 0) {
1906 pthread_mutex_unlock(&consumer_data->pid_mutex);
1907 goto end;
1908 }
1909
1910 ret = spawn_consumerd(consumer_data);
1911 if (ret < 0) {
1912 ERR("Spawning consumerd failed");
1913 pthread_mutex_unlock(&consumer_data->pid_mutex);
1914 goto error;
1915 }
1916
1917 /* Setting up the consumer_data pid */
1918 consumer_data->pid = ret;
1919 DBG2("Consumer pid %d", consumer_data->pid);
1920 pthread_mutex_unlock(&consumer_data->pid_mutex);
1921
1922 DBG2("Spawning consumer control thread");
1923 ret = spawn_consumer_thread(consumer_data);
1924 if (ret < 0) {
1925 ERR("Fatal error spawning consumer control thread");
1926 goto error;
1927 }
1928
1929 end:
1930 return 0;
1931
1932 error:
1933 /* Cleanup already created socket on error. */
1934 if (consumer_data->err_sock >= 0) {
1935 int err;
1936
1937 err = close(consumer_data->err_sock);
1938 if (err < 0) {
1939 PERROR("close consumer data error socket");
1940 }
1941 }
1942 return ret;
1943 }
1944
1945 /*
1946 * Compute health status of each consumer. If one of them is zero (bad
1947 * state), we return 0.
1948 */
1949 static int check_consumer_health(void)
1950 {
1951 int ret;
1952
1953 ret = health_check_state(HEALTH_TYPE_CONSUMER);
1954
1955 DBG3("Health consumer check %d", ret);
1956
1957 return ret;
1958 }
1959
1960 /*
1961 * Setup necessary data for kernel tracer action.
1962 */
1963 static int init_kernel_tracer(void)
1964 {
1965 int ret;
1966
1967 /* Modprobe lttng kernel modules */
1968 ret = modprobe_lttng_control();
1969 if (ret < 0) {
1970 goto error;
1971 }
1972
1973 /* Open debugfs lttng */
1974 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1975 if (kernel_tracer_fd < 0) {
1976 DBG("Failed to open %s", module_proc_lttng);
1977 ret = -1;
1978 goto error_open;
1979 }
1980
1981 /* Validate kernel version */
1982 ret = kernel_validate_version(kernel_tracer_fd);
1983 if (ret < 0) {
1984 goto error_version;
1985 }
1986
1987 ret = modprobe_lttng_data();
1988 if (ret < 0) {
1989 goto error_modules;
1990 }
1991
1992 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1993 return 0;
1994
1995 error_version:
1996 modprobe_remove_lttng_control();
1997 ret = close(kernel_tracer_fd);
1998 if (ret) {
1999 PERROR("close");
2000 }
2001 kernel_tracer_fd = -1;
2002 return LTTNG_ERR_KERN_VERSION;
2003
2004 error_modules:
2005 ret = close(kernel_tracer_fd);
2006 if (ret) {
2007 PERROR("close");
2008 }
2009
2010 error_open:
2011 modprobe_remove_lttng_control();
2012
2013 error:
2014 WARN("No kernel tracer available");
2015 kernel_tracer_fd = -1;
2016 if (!is_root) {
2017 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2018 } else {
2019 return LTTNG_ERR_KERN_NA;
2020 }
2021 }
2022
2023
2024 /*
2025 * Copy consumer output from the tracing session to the domain session. The
2026 * function also applies the right modification on a per domain basis for the
2027 * trace files destination directory.
2028 */
2029 static int copy_session_consumer(int domain, struct ltt_session *session)
2030 {
2031 int ret;
2032 const char *dir_name;
2033 struct consumer_output *consumer;
2034
2035 assert(session);
2036 assert(session->consumer);
2037
2038 switch (domain) {
2039 case LTTNG_DOMAIN_KERNEL:
2040 DBG3("Copying tracing session consumer output in kernel session");
2041 /*
2042 * XXX: We should audit the session creation and what this function
2043 * does "extra" in order to avoid a destroy since this function is used
2044 * in the domain session creation (kernel and ust) only. Same for UST
2045 * domain.
2046 */
2047 if (session->kernel_session->consumer) {
2048 consumer_destroy_output(session->kernel_session->consumer);
2049 }
2050 session->kernel_session->consumer =
2051 consumer_copy_output(session->consumer);
2052 /* Ease our life a bit for the next part */
2053 consumer = session->kernel_session->consumer;
2054 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2055 break;
2056 case LTTNG_DOMAIN_UST:
2057 DBG3("Copying tracing session consumer output in UST session");
2058 if (session->ust_session->consumer) {
2059 consumer_destroy_output(session->ust_session->consumer);
2060 }
2061 session->ust_session->consumer =
2062 consumer_copy_output(session->consumer);
2063 /* Ease our life a bit for the next part */
2064 consumer = session->ust_session->consumer;
2065 dir_name = DEFAULT_UST_TRACE_DIR;
2066 break;
2067 default:
2068 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2069 goto error;
2070 }
2071
2072 /* Append correct directory to subdir */
2073 strncat(consumer->subdir, dir_name,
2074 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2075 DBG3("Copy session consumer subdir %s", consumer->subdir);
2076
2077 ret = LTTNG_OK;
2078
2079 error:
2080 return ret;
2081 }
2082
2083 /*
2084 * Create an UST session and add it to the session ust list.
2085 */
2086 static int create_ust_session(struct ltt_session *session,
2087 struct lttng_domain *domain)
2088 {
2089 int ret;
2090 struct ltt_ust_session *lus = NULL;
2091
2092 assert(session);
2093 assert(domain);
2094 assert(session->consumer);
2095
2096 switch (domain->type) {
2097 case LTTNG_DOMAIN_UST:
2098 break;
2099 default:
2100 ERR("Unknown UST domain on create session %d", domain->type);
2101 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2102 goto error;
2103 }
2104
2105 DBG("Creating UST session");
2106
2107 lus = trace_ust_create_session(session->path, session->id);
2108 if (lus == NULL) {
2109 ret = LTTNG_ERR_UST_SESS_FAIL;
2110 goto error;
2111 }
2112
2113 lus->uid = session->uid;
2114 lus->gid = session->gid;
2115 session->ust_session = lus;
2116
2117 /* Copy session output to the newly created UST session */
2118 ret = copy_session_consumer(domain->type, session);
2119 if (ret != LTTNG_OK) {
2120 goto error;
2121 }
2122
2123 return LTTNG_OK;
2124
2125 error:
2126 free(lus);
2127 session->ust_session = NULL;
2128 return ret;
2129 }
2130
2131 /*
2132 * Create a kernel tracer session then create the default channel.
2133 */
2134 static int create_kernel_session(struct ltt_session *session)
2135 {
2136 int ret;
2137
2138 DBG("Creating kernel session");
2139
2140 ret = kernel_create_session(session, kernel_tracer_fd);
2141 if (ret < 0) {
2142 ret = LTTNG_ERR_KERN_SESS_FAIL;
2143 goto error;
2144 }
2145
2146 /* Code flow safety */
2147 assert(session->kernel_session);
2148
2149 /* Copy session output to the newly created Kernel session */
2150 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2151 if (ret != LTTNG_OK) {
2152 goto error;
2153 }
2154
2155 /* Create directory(ies) on local filesystem. */
2156 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2157 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2158 ret = run_as_mkdir_recursive(
2159 session->kernel_session->consumer->dst.trace_path,
2160 S_IRWXU | S_IRWXG, session->uid, session->gid);
2161 if (ret < 0) {
2162 if (ret != -EEXIST) {
2163 ERR("Trace directory creation error");
2164 goto error;
2165 }
2166 }
2167 }
2168
2169 session->kernel_session->uid = session->uid;
2170 session->kernel_session->gid = session->gid;
2171
2172 return LTTNG_OK;
2173
2174 error:
2175 trace_kernel_destroy_session(session->kernel_session);
2176 session->kernel_session = NULL;
2177 return ret;
2178 }
2179
2180 /*
2181 * Count number of session permitted by uid/gid.
2182 */
2183 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2184 {
2185 unsigned int i = 0;
2186 struct ltt_session *session;
2187
2188 DBG("Counting number of available session for UID %d GID %d",
2189 uid, gid);
2190 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2191 /*
2192 * Only list the sessions the user can control.
2193 */
2194 if (!session_access_ok(session, uid, gid)) {
2195 continue;
2196 }
2197 i++;
2198 }
2199 return i;
2200 }
2201
2202 /*
2203 * Process the command requested by the lttng client within the command
2204 * context structure. This function make sure that the return structure (llm)
2205 * is set and ready for transmission before returning.
2206 *
2207 * Return any error encountered or 0 for success.
2208 *
2209 * "sock" is only used for special-case var. len data.
2210 */
2211 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2212 int *sock_error)
2213 {
2214 int ret = LTTNG_OK;
2215 int need_tracing_session = 1;
2216 int need_domain;
2217
2218 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2219
2220 *sock_error = 0;
2221
2222 switch (cmd_ctx->lsm->cmd_type) {
2223 case LTTNG_CREATE_SESSION:
2224 case LTTNG_DESTROY_SESSION:
2225 case LTTNG_LIST_SESSIONS:
2226 case LTTNG_LIST_DOMAINS:
2227 case LTTNG_START_TRACE:
2228 case LTTNG_STOP_TRACE:
2229 case LTTNG_DATA_PENDING:
2230 need_domain = 0;
2231 break;
2232 default:
2233 need_domain = 1;
2234 }
2235
2236 if (opt_no_kernel && need_domain
2237 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2238 if (!is_root) {
2239 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2240 } else {
2241 ret = LTTNG_ERR_KERN_NA;
2242 }
2243 goto error;
2244 }
2245
2246 /* Deny register consumer if we already have a spawned consumer. */
2247 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2248 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2249 if (kconsumer_data.pid > 0) {
2250 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2251 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2252 goto error;
2253 }
2254 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2255 }
2256
2257 /*
2258 * Check for command that don't needs to allocate a returned payload. We do
2259 * this here so we don't have to make the call for no payload at each
2260 * command.
2261 */
2262 switch(cmd_ctx->lsm->cmd_type) {
2263 case LTTNG_LIST_SESSIONS:
2264 case LTTNG_LIST_TRACEPOINTS:
2265 case LTTNG_LIST_TRACEPOINT_FIELDS:
2266 case LTTNG_LIST_DOMAINS:
2267 case LTTNG_LIST_CHANNELS:
2268 case LTTNG_LIST_EVENTS:
2269 break;
2270 default:
2271 /* Setup lttng message with no payload */
2272 ret = setup_lttng_msg(cmd_ctx, 0);
2273 if (ret < 0) {
2274 /* This label does not try to unlock the session */
2275 goto init_setup_error;
2276 }
2277 }
2278
2279 /* Commands that DO NOT need a session. */
2280 switch (cmd_ctx->lsm->cmd_type) {
2281 case LTTNG_CREATE_SESSION:
2282 case LTTNG_CALIBRATE:
2283 case LTTNG_LIST_SESSIONS:
2284 case LTTNG_LIST_TRACEPOINTS:
2285 case LTTNG_LIST_TRACEPOINT_FIELDS:
2286 need_tracing_session = 0;
2287 break;
2288 default:
2289 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2290 /*
2291 * We keep the session list lock across _all_ commands
2292 * for now, because the per-session lock does not
2293 * handle teardown properly.
2294 */
2295 session_lock_list();
2296 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2297 if (cmd_ctx->session == NULL) {
2298 if (cmd_ctx->lsm->session.name != NULL) {
2299 ret = LTTNG_ERR_SESS_NOT_FOUND;
2300 } else {
2301 /* If no session name specified */
2302 ret = LTTNG_ERR_SELECT_SESS;
2303 }
2304 goto error;
2305 } else {
2306 /* Acquire lock for the session */
2307 session_lock(cmd_ctx->session);
2308 }
2309 break;
2310 }
2311
2312 if (!need_domain) {
2313 goto skip_domain;
2314 }
2315
2316 /*
2317 * Check domain type for specific "pre-action".
2318 */
2319 switch (cmd_ctx->lsm->domain.type) {
2320 case LTTNG_DOMAIN_KERNEL:
2321 if (!is_root) {
2322 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2323 goto error;
2324 }
2325
2326 /* Kernel tracer check */
2327 if (kernel_tracer_fd == -1) {
2328 /* Basically, load kernel tracer modules */
2329 ret = init_kernel_tracer();
2330 if (ret != 0) {
2331 goto error;
2332 }
2333 }
2334
2335 /* Consumer is in an ERROR state. Report back to client */
2336 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2337 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2338 goto error;
2339 }
2340
2341 /* Need a session for kernel command */
2342 if (need_tracing_session) {
2343 if (cmd_ctx->session->kernel_session == NULL) {
2344 ret = create_kernel_session(cmd_ctx->session);
2345 if (ret < 0) {
2346 ret = LTTNG_ERR_KERN_SESS_FAIL;
2347 goto error;
2348 }
2349 }
2350
2351 /* Start the kernel consumer daemon */
2352 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2353 if (kconsumer_data.pid == 0 &&
2354 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2355 cmd_ctx->session->start_consumer) {
2356 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2357 ret = start_consumerd(&kconsumer_data);
2358 if (ret < 0) {
2359 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2360 goto error;
2361 }
2362 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2363 } else {
2364 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2365 }
2366
2367 /*
2368 * The consumer was just spawned so we need to add the socket to
2369 * the consumer output of the session if exist.
2370 */
2371 ret = consumer_create_socket(&kconsumer_data,
2372 cmd_ctx->session->kernel_session->consumer);
2373 if (ret < 0) {
2374 goto error;
2375 }
2376 }
2377
2378 break;
2379 case LTTNG_DOMAIN_UST:
2380 {
2381 /* Consumer is in an ERROR state. Report back to client */
2382 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2383 ret = LTTNG_ERR_NO_USTCONSUMERD;
2384 goto error;
2385 }
2386
2387 if (need_tracing_session) {
2388 /* Create UST session if none exist. */
2389 if (cmd_ctx->session->ust_session == NULL) {
2390 ret = create_ust_session(cmd_ctx->session,
2391 &cmd_ctx->lsm->domain);
2392 if (ret != LTTNG_OK) {
2393 goto error;
2394 }
2395 }
2396
2397 /* Start the UST consumer daemons */
2398 /* 64-bit */
2399 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2400 if (consumerd64_bin[0] != '\0' &&
2401 ustconsumer64_data.pid == 0 &&
2402 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2403 cmd_ctx->session->start_consumer) {
2404 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2405 ret = start_consumerd(&ustconsumer64_data);
2406 if (ret < 0) {
2407 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2408 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2409 goto error;
2410 }
2411
2412 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2413 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2414 } else {
2415 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2416 }
2417
2418 /*
2419 * Setup socket for consumer 64 bit. No need for atomic access
2420 * since it was set above and can ONLY be set in this thread.
2421 */
2422 ret = consumer_create_socket(&ustconsumer64_data,
2423 cmd_ctx->session->ust_session->consumer);
2424 if (ret < 0) {
2425 goto error;
2426 }
2427
2428 /* 32-bit */
2429 if (consumerd32_bin[0] != '\0' &&
2430 ustconsumer32_data.pid == 0 &&
2431 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2432 cmd_ctx->session->start_consumer) {
2433 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2434 ret = start_consumerd(&ustconsumer32_data);
2435 if (ret < 0) {
2436 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2437 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2438 goto error;
2439 }
2440
2441 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2442 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2443 } else {
2444 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2445 }
2446
2447 /*
2448 * Setup socket for consumer 64 bit. No need for atomic access
2449 * since it was set above and can ONLY be set in this thread.
2450 */
2451 ret = consumer_create_socket(&ustconsumer32_data,
2452 cmd_ctx->session->ust_session->consumer);
2453 if (ret < 0) {
2454 goto error;
2455 }
2456 }
2457 break;
2458 }
2459 default:
2460 break;
2461 }
2462 skip_domain:
2463
2464 /* Validate consumer daemon state when start/stop trace command */
2465 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2466 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2467 switch (cmd_ctx->lsm->domain.type) {
2468 case LTTNG_DOMAIN_UST:
2469 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2470 ret = LTTNG_ERR_NO_USTCONSUMERD;
2471 goto error;
2472 }
2473 break;
2474 case LTTNG_DOMAIN_KERNEL:
2475 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2476 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2477 goto error;
2478 }
2479 break;
2480 }
2481 }
2482
2483 /*
2484 * Check that the UID or GID match that of the tracing session.
2485 * The root user can interact with all sessions.
2486 */
2487 if (need_tracing_session) {
2488 if (!session_access_ok(cmd_ctx->session,
2489 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2490 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2491 ret = LTTNG_ERR_EPERM;
2492 goto error;
2493 }
2494 }
2495
2496 /* Process by command type */
2497 switch (cmd_ctx->lsm->cmd_type) {
2498 case LTTNG_ADD_CONTEXT:
2499 {
2500 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2501 cmd_ctx->lsm->u.context.channel_name,
2502 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2503 break;
2504 }
2505 case LTTNG_DISABLE_CHANNEL:
2506 {
2507 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2508 cmd_ctx->lsm->u.disable.channel_name);
2509 break;
2510 }
2511 case LTTNG_DISABLE_EVENT:
2512 {
2513 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2514 cmd_ctx->lsm->u.disable.channel_name,
2515 cmd_ctx->lsm->u.disable.name);
2516 break;
2517 }
2518 case LTTNG_DISABLE_ALL_EVENT:
2519 {
2520 DBG("Disabling all events");
2521
2522 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2523 cmd_ctx->lsm->u.disable.channel_name);
2524 break;
2525 }
2526 case LTTNG_DISABLE_CONSUMER:
2527 {
2528 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2529 break;
2530 }
2531 case LTTNG_ENABLE_CHANNEL:
2532 {
2533 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2534 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2535 break;
2536 }
2537 case LTTNG_ENABLE_CONSUMER:
2538 {
2539 /*
2540 * XXX: 0 means that this URI should be applied on the session. Should
2541 * be a DOMAIN enuam.
2542 */
2543 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2544 if (ret != LTTNG_OK) {
2545 goto error;
2546 }
2547
2548 if (cmd_ctx->lsm->domain.type == 0) {
2549 /* Add the URI for the UST session if a consumer is present. */
2550 if (cmd_ctx->session->ust_session &&
2551 cmd_ctx->session->ust_session->consumer) {
2552 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2553 } else if (cmd_ctx->session->kernel_session &&
2554 cmd_ctx->session->kernel_session->consumer) {
2555 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2556 cmd_ctx->session);
2557 }
2558 }
2559 break;
2560 }
2561 case LTTNG_ENABLE_EVENT:
2562 {
2563 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2564 cmd_ctx->lsm->u.enable.channel_name,
2565 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2566 break;
2567 }
2568 case LTTNG_ENABLE_ALL_EVENT:
2569 {
2570 DBG("Enabling all events");
2571
2572 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2573 cmd_ctx->lsm->u.enable.channel_name,
2574 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2575 break;
2576 }
2577 case LTTNG_LIST_TRACEPOINTS:
2578 {
2579 struct lttng_event *events;
2580 ssize_t nb_events;
2581
2582 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2583 if (nb_events < 0) {
2584 /* Return value is a negative lttng_error_code. */
2585 ret = -nb_events;
2586 goto error;
2587 }
2588
2589 /*
2590 * Setup lttng message with payload size set to the event list size in
2591 * bytes and then copy list into the llm payload.
2592 */
2593 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2594 if (ret < 0) {
2595 free(events);
2596 goto setup_error;
2597 }
2598
2599 /* Copy event list into message payload */
2600 memcpy(cmd_ctx->llm->payload, events,
2601 sizeof(struct lttng_event) * nb_events);
2602
2603 free(events);
2604
2605 ret = LTTNG_OK;
2606 break;
2607 }
2608 case LTTNG_LIST_TRACEPOINT_FIELDS:
2609 {
2610 struct lttng_event_field *fields;
2611 ssize_t nb_fields;
2612
2613 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2614 &fields);
2615 if (nb_fields < 0) {
2616 /* Return value is a negative lttng_error_code. */
2617 ret = -nb_fields;
2618 goto error;
2619 }
2620
2621 /*
2622 * Setup lttng message with payload size set to the event list size in
2623 * bytes and then copy list into the llm payload.
2624 */
2625 ret = setup_lttng_msg(cmd_ctx,
2626 sizeof(struct lttng_event_field) * nb_fields);
2627 if (ret < 0) {
2628 free(fields);
2629 goto setup_error;
2630 }
2631
2632 /* Copy event list into message payload */
2633 memcpy(cmd_ctx->llm->payload, fields,
2634 sizeof(struct lttng_event_field) * nb_fields);
2635
2636 free(fields);
2637
2638 ret = LTTNG_OK;
2639 break;
2640 }
2641 case LTTNG_SET_CONSUMER_URI:
2642 {
2643 size_t nb_uri, len;
2644 struct lttng_uri *uris;
2645
2646 nb_uri = cmd_ctx->lsm->u.uri.size;
2647 len = nb_uri * sizeof(struct lttng_uri);
2648
2649 if (nb_uri == 0) {
2650 ret = LTTNG_ERR_INVALID;
2651 goto error;
2652 }
2653
2654 uris = zmalloc(len);
2655 if (uris == NULL) {
2656 ret = LTTNG_ERR_FATAL;
2657 goto error;
2658 }
2659
2660 /* Receive variable len data */
2661 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2662 ret = lttcomm_recv_unix_sock(sock, uris, len);
2663 if (ret <= 0) {
2664 DBG("No URIs received from client... continuing");
2665 *sock_error = 1;
2666 ret = LTTNG_ERR_SESSION_FAIL;
2667 free(uris);
2668 goto error;
2669 }
2670
2671 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2672 nb_uri, uris);
2673 if (ret != LTTNG_OK) {
2674 free(uris);
2675 goto error;
2676 }
2677
2678 /*
2679 * XXX: 0 means that this URI should be applied on the session. Should
2680 * be a DOMAIN enuam.
2681 */
2682 if (cmd_ctx->lsm->domain.type == 0) {
2683 /* Add the URI for the UST session if a consumer is present. */
2684 if (cmd_ctx->session->ust_session &&
2685 cmd_ctx->session->ust_session->consumer) {
2686 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2687 nb_uri, uris);
2688 } else if (cmd_ctx->session->kernel_session &&
2689 cmd_ctx->session->kernel_session->consumer) {
2690 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2691 cmd_ctx->session, nb_uri, uris);
2692 }
2693 }
2694
2695 free(uris);
2696
2697 break;
2698 }
2699 case LTTNG_START_TRACE:
2700 {
2701 ret = cmd_start_trace(cmd_ctx->session);
2702 break;
2703 }
2704 case LTTNG_STOP_TRACE:
2705 {
2706 ret = cmd_stop_trace(cmd_ctx->session);
2707 break;
2708 }
2709 case LTTNG_CREATE_SESSION:
2710 {
2711 size_t nb_uri, len;
2712 struct lttng_uri *uris = NULL;
2713
2714 nb_uri = cmd_ctx->lsm->u.uri.size;
2715 len = nb_uri * sizeof(struct lttng_uri);
2716
2717 if (nb_uri > 0) {
2718 uris = zmalloc(len);
2719 if (uris == NULL) {
2720 ret = LTTNG_ERR_FATAL;
2721 goto error;
2722 }
2723
2724 /* Receive variable len data */
2725 DBG("Waiting for %zu URIs from client ...", nb_uri);
2726 ret = lttcomm_recv_unix_sock(sock, uris, len);
2727 if (ret <= 0) {
2728 DBG("No URIs received from client... continuing");
2729 *sock_error = 1;
2730 ret = LTTNG_ERR_SESSION_FAIL;
2731 free(uris);
2732 goto error;
2733 }
2734
2735 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2736 DBG("Creating session with ONE network URI is a bad call");
2737 ret = LTTNG_ERR_SESSION_FAIL;
2738 free(uris);
2739 goto error;
2740 }
2741 }
2742
2743 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2744 &cmd_ctx->creds);
2745
2746 free(uris);
2747
2748 break;
2749 }
2750 case LTTNG_DESTROY_SESSION:
2751 {
2752 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2753
2754 /* Set session to NULL so we do not unlock it after free. */
2755 cmd_ctx->session = NULL;
2756 break;
2757 }
2758 case LTTNG_LIST_DOMAINS:
2759 {
2760 ssize_t nb_dom;
2761 struct lttng_domain *domains;
2762
2763 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2764 if (nb_dom < 0) {
2765 /* Return value is a negative lttng_error_code. */
2766 ret = -nb_dom;
2767 goto error;
2768 }
2769
2770 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2771 if (ret < 0) {
2772 goto setup_error;
2773 }
2774
2775 /* Copy event list into message payload */
2776 memcpy(cmd_ctx->llm->payload, domains,
2777 nb_dom * sizeof(struct lttng_domain));
2778
2779 free(domains);
2780
2781 ret = LTTNG_OK;
2782 break;
2783 }
2784 case LTTNG_LIST_CHANNELS:
2785 {
2786 int nb_chan;
2787 struct lttng_channel *channels;
2788
2789 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2790 cmd_ctx->session, &channels);
2791 if (nb_chan < 0) {
2792 /* Return value is a negative lttng_error_code. */
2793 ret = -nb_chan;
2794 goto error;
2795 }
2796
2797 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2798 if (ret < 0) {
2799 goto setup_error;
2800 }
2801
2802 /* Copy event list into message payload */
2803 memcpy(cmd_ctx->llm->payload, channels,
2804 nb_chan * sizeof(struct lttng_channel));
2805
2806 free(channels);
2807
2808 ret = LTTNG_OK;
2809 break;
2810 }
2811 case LTTNG_LIST_EVENTS:
2812 {
2813 ssize_t nb_event;
2814 struct lttng_event *events = NULL;
2815
2816 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2817 cmd_ctx->lsm->u.list.channel_name, &events);
2818 if (nb_event < 0) {
2819 /* Return value is a negative lttng_error_code. */
2820 ret = -nb_event;
2821 goto error;
2822 }
2823
2824 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2825 if (ret < 0) {
2826 goto setup_error;
2827 }
2828
2829 /* Copy event list into message payload */
2830 memcpy(cmd_ctx->llm->payload, events,
2831 nb_event * sizeof(struct lttng_event));
2832
2833 free(events);
2834
2835 ret = LTTNG_OK;
2836 break;
2837 }
2838 case LTTNG_LIST_SESSIONS:
2839 {
2840 unsigned int nr_sessions;
2841
2842 session_lock_list();
2843 nr_sessions = lttng_sessions_count(
2844 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2845 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2846
2847 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2848 if (ret < 0) {
2849 session_unlock_list();
2850 goto setup_error;
2851 }
2852
2853 /* Filled the session array */
2854 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2855 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2856 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2857
2858 session_unlock_list();
2859
2860 ret = LTTNG_OK;
2861 break;
2862 }
2863 case LTTNG_CALIBRATE:
2864 {
2865 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2866 &cmd_ctx->lsm->u.calibrate);
2867 break;
2868 }
2869 case LTTNG_REGISTER_CONSUMER:
2870 {
2871 struct consumer_data *cdata;
2872
2873 switch (cmd_ctx->lsm->domain.type) {
2874 case LTTNG_DOMAIN_KERNEL:
2875 cdata = &kconsumer_data;
2876 break;
2877 default:
2878 ret = LTTNG_ERR_UND;
2879 goto error;
2880 }
2881
2882 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2883 cmd_ctx->lsm->u.reg.path, cdata);
2884 break;
2885 }
2886 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2887 {
2888 struct lttng_filter_bytecode *bytecode;
2889
2890 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2891 ret = LTTNG_ERR_FILTER_INVAL;
2892 goto error;
2893 }
2894 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2895 ret = LTTNG_ERR_FILTER_INVAL;
2896 goto error;
2897 }
2898 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2899 if (!bytecode) {
2900 ret = LTTNG_ERR_FILTER_NOMEM;
2901 goto error;
2902 }
2903 /* Receive var. len. data */
2904 DBG("Receiving var len data from client ...");
2905 ret = lttcomm_recv_unix_sock(sock, bytecode,
2906 cmd_ctx->lsm->u.enable.bytecode_len);
2907 if (ret <= 0) {
2908 DBG("Nothing recv() from client var len data... continuing");
2909 *sock_error = 1;
2910 ret = LTTNG_ERR_FILTER_INVAL;
2911 goto error;
2912 }
2913
2914 if (bytecode->len + sizeof(*bytecode)
2915 != cmd_ctx->lsm->u.enable.bytecode_len) {
2916 free(bytecode);
2917 ret = LTTNG_ERR_FILTER_INVAL;
2918 goto error;
2919 }
2920
2921 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2922 cmd_ctx->lsm->u.enable.channel_name,
2923 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2924 break;
2925 }
2926 case LTTNG_DATA_PENDING:
2927 {
2928 ret = cmd_data_pending(cmd_ctx->session);
2929 break;
2930 }
2931 default:
2932 ret = LTTNG_ERR_UND;
2933 break;
2934 }
2935
2936 error:
2937 if (cmd_ctx->llm == NULL) {
2938 DBG("Missing llm structure. Allocating one.");
2939 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2940 goto setup_error;
2941 }
2942 }
2943 /* Set return code */
2944 cmd_ctx->llm->ret_code = ret;
2945 setup_error:
2946 if (cmd_ctx->session) {
2947 session_unlock(cmd_ctx->session);
2948 }
2949 if (need_tracing_session) {
2950 session_unlock_list();
2951 }
2952 init_setup_error:
2953 return ret;
2954 }
2955
2956 /*
2957 * Thread managing health check socket.
2958 */
2959 static void *thread_manage_health(void *data)
2960 {
2961 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2962 uint32_t revents, nb_fd;
2963 struct lttng_poll_event events;
2964 struct lttcomm_health_msg msg;
2965 struct lttcomm_health_data reply;
2966
2967 DBG("[thread] Manage health check started");
2968
2969 rcu_register_thread();
2970
2971 /* Create unix socket */
2972 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2973 if (sock < 0) {
2974 ERR("Unable to create health check Unix socket");
2975 ret = -1;
2976 goto error;
2977 }
2978
2979 /*
2980 * Set the CLOEXEC flag. Return code is useless because either way, the
2981 * show must go on.
2982 */
2983 (void) utils_set_fd_cloexec(sock);
2984
2985 ret = lttcomm_listen_unix_sock(sock);
2986 if (ret < 0) {
2987 goto error;
2988 }
2989
2990 /*
2991 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2992 * more will be added to this poll set.
2993 */
2994 ret = create_thread_poll_set(&events, 2);
2995 if (ret < 0) {
2996 goto error;
2997 }
2998
2999 /* Add the application registration socket */
3000 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3001 if (ret < 0) {
3002 goto error;
3003 }
3004
3005 while (1) {
3006 DBG("Health check ready");
3007
3008 /* Inifinite blocking call, waiting for transmission */
3009 restart:
3010 ret = lttng_poll_wait(&events, -1);
3011 if (ret < 0) {
3012 /*
3013 * Restart interrupted system call.
3014 */
3015 if (errno == EINTR) {
3016 goto restart;
3017 }
3018 goto error;
3019 }
3020
3021 nb_fd = ret;
3022
3023 for (i = 0; i < nb_fd; i++) {
3024 /* Fetch once the poll data */
3025 revents = LTTNG_POLL_GETEV(&events, i);
3026 pollfd = LTTNG_POLL_GETFD(&events, i);
3027
3028 /* Thread quit pipe has been closed. Killing thread. */
3029 ret = check_thread_quit_pipe(pollfd, revents);
3030 if (ret) {
3031 err = 0;
3032 goto exit;
3033 }
3034
3035 /* Event on the registration socket */
3036 if (pollfd == sock) {
3037 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3038 ERR("Health socket poll error");
3039 goto error;
3040 }
3041 }
3042 }
3043
3044 new_sock = lttcomm_accept_unix_sock(sock);
3045 if (new_sock < 0) {
3046 goto error;
3047 }
3048
3049 /*
3050 * Set the CLOEXEC flag. Return code is useless because either way, the
3051 * show must go on.
3052 */
3053 (void) utils_set_fd_cloexec(new_sock);
3054
3055 DBG("Receiving data from client for health...");
3056 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3057 if (ret <= 0) {
3058 DBG("Nothing recv() from client... continuing");
3059 ret = close(new_sock);
3060 if (ret) {
3061 PERROR("close");
3062 }
3063 new_sock = -1;
3064 continue;
3065 }
3066
3067 rcu_thread_online();
3068
3069 switch (msg.component) {
3070 case LTTNG_HEALTH_CMD:
3071 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3072 break;
3073 case LTTNG_HEALTH_APP_MANAGE:
3074 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3075 break;
3076 case LTTNG_HEALTH_APP_REG:
3077 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3078 break;
3079 case LTTNG_HEALTH_KERNEL:
3080 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3081 break;
3082 case LTTNG_HEALTH_CONSUMER:
3083 reply.ret_code = check_consumer_health();
3084 break;
3085 case LTTNG_HEALTH_ALL:
3086 reply.ret_code =
3087 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3088 health_check_state(HEALTH_TYPE_APP_REG) &&
3089 health_check_state(HEALTH_TYPE_CMD) &&
3090 health_check_state(HEALTH_TYPE_KERNEL) &&
3091 check_consumer_health();
3092 break;
3093 default:
3094 reply.ret_code = LTTNG_ERR_UND;
3095 break;
3096 }
3097
3098 /*
3099 * Flip ret value since 0 is a success and 1 indicates a bad health for
3100 * the client where in the sessiond it is the opposite. Again, this is
3101 * just to make things easier for us poor developer which enjoy a lot
3102 * lazyness.
3103 */
3104 if (reply.ret_code == 0 || reply.ret_code == 1) {
3105 reply.ret_code = !reply.ret_code;
3106 }
3107
3108 DBG2("Health check return value %d", reply.ret_code);
3109
3110 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3111 if (ret < 0) {
3112 ERR("Failed to send health data back to client");
3113 }
3114
3115 /* End of transmission */
3116 ret = close(new_sock);
3117 if (ret) {
3118 PERROR("close");
3119 }
3120 new_sock = -1;
3121 }
3122
3123 exit:
3124 error:
3125 if (err) {
3126 ERR("Health error occurred in %s", __func__);
3127 }
3128 DBG("Health check thread dying");
3129 unlink(health_unix_sock_path);
3130 if (sock >= 0) {
3131 ret = close(sock);
3132 if (ret) {
3133 PERROR("close");
3134 }
3135 }
3136 if (new_sock >= 0) {
3137 ret = close(new_sock);
3138 if (ret) {
3139 PERROR("close");
3140 }
3141 }
3142
3143 lttng_poll_clean(&events);
3144
3145 rcu_unregister_thread();
3146 return NULL;
3147 }
3148
3149 /*
3150 * This thread manage all clients request using the unix client socket for
3151 * communication.
3152 */
3153 static void *thread_manage_clients(void *data)
3154 {
3155 int sock = -1, ret, i, pollfd, err = -1;
3156 int sock_error;
3157 uint32_t revents, nb_fd;
3158 struct command_ctx *cmd_ctx = NULL;
3159 struct lttng_poll_event events;
3160
3161 DBG("[thread] Manage client started");
3162
3163 rcu_register_thread();
3164
3165 health_register(HEALTH_TYPE_CMD);
3166
3167 if (testpoint(thread_manage_clients)) {
3168 goto error_testpoint;
3169 }
3170
3171 health_code_update();
3172
3173 ret = lttcomm_listen_unix_sock(client_sock);
3174 if (ret < 0) {
3175 goto error_listen;
3176 }
3177
3178 /*
3179 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3180 * more will be added to this poll set.
3181 */
3182 ret = create_thread_poll_set(&events, 2);
3183 if (ret < 0) {
3184 goto error_create_poll;
3185 }
3186
3187 /* Add the application registration socket */
3188 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3189 if (ret < 0) {
3190 goto error;
3191 }
3192
3193 /*
3194 * Notify parent pid that we are ready to accept command for client side.
3195 */
3196 if (opt_sig_parent) {
3197 kill(ppid, SIGUSR1);
3198 }
3199
3200 if (testpoint(thread_manage_clients_before_loop)) {
3201 goto error;
3202 }
3203
3204 health_code_update();
3205
3206 while (1) {
3207 DBG("Accepting client command ...");
3208
3209 /* Inifinite blocking call, waiting for transmission */
3210 restart:
3211 health_poll_update();
3212 ret = lttng_poll_wait(&events, -1);
3213 health_poll_update();
3214 if (ret < 0) {
3215 /*
3216 * Restart interrupted system call.
3217 */
3218 if (errno == EINTR) {
3219 goto restart;
3220 }
3221 goto error;
3222 }
3223
3224 nb_fd = ret;
3225
3226 for (i = 0; i < nb_fd; i++) {
3227 /* Fetch once the poll data */
3228 revents = LTTNG_POLL_GETEV(&events, i);
3229 pollfd = LTTNG_POLL_GETFD(&events, i);
3230
3231 health_code_update();
3232
3233 /* Thread quit pipe has been closed. Killing thread. */
3234 ret = check_thread_quit_pipe(pollfd, revents);
3235 if (ret) {
3236 err = 0;
3237 goto exit;
3238 }
3239
3240 /* Event on the registration socket */
3241 if (pollfd == client_sock) {
3242 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3243 ERR("Client socket poll error");
3244 goto error;
3245 }
3246 }
3247 }
3248
3249 DBG("Wait for client response");
3250
3251 health_code_update();
3252
3253 sock = lttcomm_accept_unix_sock(client_sock);
3254 if (sock < 0) {
3255 goto error;
3256 }
3257
3258 /*
3259 * Set the CLOEXEC flag. Return code is useless because either way, the
3260 * show must go on.
3261 */
3262 (void) utils_set_fd_cloexec(sock);
3263
3264 /* Set socket option for credentials retrieval */
3265 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3266 if (ret < 0) {
3267 goto error;
3268 }
3269
3270 /* Allocate context command to process the client request */
3271 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3272 if (cmd_ctx == NULL) {
3273 PERROR("zmalloc cmd_ctx");
3274 goto error;
3275 }
3276
3277 /* Allocate data buffer for reception */
3278 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3279 if (cmd_ctx->lsm == NULL) {
3280 PERROR("zmalloc cmd_ctx->lsm");
3281 goto error;
3282 }
3283
3284 cmd_ctx->llm = NULL;
3285 cmd_ctx->session = NULL;
3286
3287 health_code_update();
3288
3289 /*
3290 * Data is received from the lttng client. The struct
3291 * lttcomm_session_msg (lsm) contains the command and data request of
3292 * the client.
3293 */
3294 DBG("Receiving data from client ...");
3295 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3296 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3297 if (ret <= 0) {
3298 DBG("Nothing recv() from client... continuing");
3299 ret = close(sock);
3300 if (ret) {
3301 PERROR("close");
3302 }
3303 sock = -1;
3304 clean_command_ctx(&cmd_ctx);
3305 continue;
3306 }
3307
3308 health_code_update();
3309
3310 // TODO: Validate cmd_ctx including sanity check for
3311 // security purpose.
3312
3313 rcu_thread_online();
3314 /*
3315 * This function dispatch the work to the kernel or userspace tracer
3316 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3317 * informations for the client. The command context struct contains
3318 * everything this function may needs.
3319 */
3320 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3321 rcu_thread_offline();
3322 if (ret < 0) {
3323 if (sock_error) {
3324 ret = close(sock);
3325 if (ret) {
3326 PERROR("close");
3327 }
3328 sock = -1;
3329 }
3330 /*
3331 * TODO: Inform client somehow of the fatal error. At
3332 * this point, ret < 0 means that a zmalloc failed
3333 * (ENOMEM). Error detected but still accept
3334 * command, unless a socket error has been
3335 * detected.
3336 */
3337 clean_command_ctx(&cmd_ctx);
3338 continue;
3339 }
3340
3341 health_code_update();
3342
3343 DBG("Sending response (size: %d, retcode: %s)",
3344 cmd_ctx->lttng_msg_size,
3345 lttng_strerror(-cmd_ctx->llm->ret_code));
3346 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3347 if (ret < 0) {
3348 ERR("Failed to send data back to client");
3349 }
3350
3351 /* End of transmission */
3352 ret = close(sock);
3353 if (ret) {
3354 PERROR("close");
3355 }
3356 sock = -1;
3357
3358 clean_command_ctx(&cmd_ctx);
3359
3360 health_code_update();
3361 }
3362
3363 exit:
3364 error:
3365 if (sock >= 0) {
3366 ret = close(sock);
3367 if (ret) {
3368 PERROR("close");
3369 }
3370 }
3371
3372 lttng_poll_clean(&events);
3373 clean_command_ctx(&cmd_ctx);
3374
3375 error_listen:
3376 error_create_poll:
3377 error_testpoint:
3378 unlink(client_unix_sock_path);
3379 if (client_sock >= 0) {
3380 ret = close(client_sock);
3381 if (ret) {
3382 PERROR("close");
3383 }
3384 }
3385
3386 if (err) {
3387 health_error();
3388 ERR("Health error occurred in %s", __func__);
3389 }
3390
3391 health_unregister();
3392
3393 DBG("Client thread dying");
3394
3395 rcu_unregister_thread();
3396 return NULL;
3397 }
3398
3399
3400 /*
3401 * usage function on stderr
3402 */
3403 static void usage(void)
3404 {
3405 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3406 fprintf(stderr, " -h, --help Display this usage.\n");
3407 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3408 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3409 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3410 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3411 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3412 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3413 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3414 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3415 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3416 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3417 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3418 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3419 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3420 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3421 fprintf(stderr, " -V, --version Show version number.\n");
3422 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3423 fprintf(stderr, " -q, --quiet No output at all.\n");
3424 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3425 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3426 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3427 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3428 }
3429
3430 /*
3431 * daemon argument parsing
3432 */
3433 static int parse_args(int argc, char **argv)
3434 {
3435 int c;
3436
3437 static struct option long_options[] = {
3438 { "client-sock", 1, 0, 'c' },
3439 { "apps-sock", 1, 0, 'a' },
3440 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3441 { "kconsumerd-err-sock", 1, 0, 'E' },
3442 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3443 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3444 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3445 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3446 { "consumerd32-path", 1, 0, 'u' },
3447 { "consumerd32-libdir", 1, 0, 'U' },
3448 { "consumerd64-path", 1, 0, 't' },
3449 { "consumerd64-libdir", 1, 0, 'T' },
3450 { "daemonize", 0, 0, 'd' },
3451 { "sig-parent", 0, 0, 'S' },
3452 { "help", 0, 0, 'h' },
3453 { "group", 1, 0, 'g' },
3454 { "version", 0, 0, 'V' },
3455 { "quiet", 0, 0, 'q' },
3456 { "verbose", 0, 0, 'v' },
3457 { "verbose-consumer", 0, 0, 'Z' },
3458 { "no-kernel", 0, 0, 'N' },
3459 { "pidfile", 1, 0, 'p' },
3460 { NULL, 0, 0, 0 }
3461 };
3462
3463 while (1) {
3464 int option_index = 0;
3465 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3466 long_options, &option_index);
3467 if (c == -1) {
3468 break;
3469 }
3470
3471 switch (c) {
3472 case 0:
3473 fprintf(stderr, "option %s", long_options[option_index].name);
3474 if (optarg) {
3475 fprintf(stderr, " with arg %s\n", optarg);
3476 }
3477 break;
3478 case 'c':
3479 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3480 break;
3481 case 'a':
3482 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3483 break;
3484 case 'd':
3485 opt_daemon = 1;
3486 break;
3487 case 'g':
3488 opt_tracing_group = optarg;
3489 break;
3490 case 'h':
3491 usage();
3492 exit(EXIT_FAILURE);
3493 case 'V':
3494 fprintf(stdout, "%s\n", VERSION);
3495 exit(EXIT_SUCCESS);
3496 case 'S':
3497 opt_sig_parent = 1;
3498 break;
3499 case 'E':
3500 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3501 break;
3502 case 'C':
3503 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3504 break;
3505 case 'F':
3506 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3507 break;
3508 case 'D':
3509 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3510 break;
3511 case 'H':
3512 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3513 break;
3514 case 'G':
3515 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3516 break;
3517 case 'N':
3518 opt_no_kernel = 1;
3519 break;
3520 case 'q':
3521 lttng_opt_quiet = 1;
3522 break;
3523 case 'v':
3524 /* Verbose level can increase using multiple -v */
3525 lttng_opt_verbose += 1;
3526 break;
3527 case 'Z':
3528 opt_verbose_consumer += 1;
3529 break;
3530 case 'u':
3531 consumerd32_bin= optarg;
3532 break;
3533 case 'U':
3534 consumerd32_libdir = optarg;
3535 break;
3536 case 't':
3537 consumerd64_bin = optarg;
3538 break;
3539 case 'T':
3540 consumerd64_libdir = optarg;
3541 break;
3542 case 'p':
3543 opt_pidfile = optarg;
3544 break;
3545 default:
3546 /* Unknown option or other error.
3547 * Error is printed by getopt, just return */
3548 return -1;
3549 }
3550 }
3551
3552 return 0;
3553 }
3554
3555 /*
3556 * Creates the two needed socket by the daemon.
3557 * apps_sock - The communication socket for all UST apps.
3558 * client_sock - The communication of the cli tool (lttng).
3559 */
3560 static int init_daemon_socket(void)
3561 {
3562 int ret = 0;
3563 mode_t old_umask;
3564
3565 old_umask = umask(0);
3566
3567 /* Create client tool unix socket */
3568 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3569 if (client_sock < 0) {
3570 ERR("Create unix sock failed: %s", client_unix_sock_path);
3571 ret = -1;
3572 goto end;
3573 }
3574
3575 /* Set the cloexec flag */
3576 ret = utils_set_fd_cloexec(client_sock);
3577 if (ret < 0) {
3578 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3579 "Continuing but note that the consumer daemon will have a "
3580 "reference to this socket on exec()", client_sock);
3581 }
3582
3583 /* File permission MUST be 660 */
3584 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3585 if (ret < 0) {
3586 ERR("Set file permissions failed: %s", client_unix_sock_path);
3587 PERROR("chmod");
3588 goto end;
3589 }
3590
3591 /* Create the application unix socket */
3592 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3593 if (apps_sock < 0) {
3594 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3595 ret = -1;
3596 goto end;
3597 }
3598
3599 /* Set the cloexec flag */
3600 ret = utils_set_fd_cloexec(apps_sock);
3601 if (ret < 0) {
3602 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3603 "Continuing but note that the consumer daemon will have a "
3604 "reference to this socket on exec()", apps_sock);
3605 }
3606
3607 /* File permission MUST be 666 */
3608 ret = chmod(apps_unix_sock_path,
3609 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3610 if (ret < 0) {
3611 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3612 PERROR("chmod");
3613 goto end;
3614 }
3615
3616 DBG3("Session daemon client socket %d and application socket %d created",
3617 client_sock, apps_sock);
3618
3619 end:
3620 umask(old_umask);
3621 return ret;
3622 }
3623
3624 /*
3625 * Check if the global socket is available, and if a daemon is answering at the
3626 * other side. If yes, error is returned.
3627 */
3628 static int check_existing_daemon(void)
3629 {
3630 /* Is there anybody out there ? */
3631 if (lttng_session_daemon_alive()) {
3632 return -EEXIST;
3633 }
3634
3635 return 0;
3636 }
3637
3638 /*
3639 * Set the tracing group gid onto the client socket.
3640 *
3641 * Race window between mkdir and chown is OK because we are going from more
3642 * permissive (root.root) to less permissive (root.tracing).
3643 */
3644 static int set_permissions(char *rundir)
3645 {
3646 int ret;
3647 gid_t gid;
3648
3649 ret = allowed_group();
3650 if (ret < 0) {
3651 WARN("No tracing group detected");
3652 ret = 0;
3653 goto end;
3654 }
3655
3656 gid = ret;
3657
3658 /* Set lttng run dir */
3659 ret = chown(rundir, 0, gid);
3660 if (ret < 0) {
3661 ERR("Unable to set group on %s", rundir);
3662 PERROR("chown");
3663 }
3664
3665 /* Ensure tracing group can search the run dir */
3666 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3667 if (ret < 0) {
3668 ERR("Unable to set permissions on %s", rundir);
3669 PERROR("chmod");
3670 }
3671
3672 /* lttng client socket path */
3673 ret = chown(client_unix_sock_path, 0, gid);
3674 if (ret < 0) {
3675 ERR("Unable to set group on %s", client_unix_sock_path);
3676 PERROR("chown");
3677 }
3678
3679 /* kconsumer error socket path */
3680 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3681 if (ret < 0) {
3682 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3683 PERROR("chown");
3684 }
3685
3686 /* 64-bit ustconsumer error socket path */
3687 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3688 if (ret < 0) {
3689 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3690 PERROR("chown");
3691 }
3692
3693 /* 32-bit ustconsumer compat32 error socket path */
3694 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3695 if (ret < 0) {
3696 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3697 PERROR("chown");
3698 }
3699
3700 DBG("All permissions are set");
3701
3702 end:
3703 return ret;
3704 }
3705
3706 /*
3707 * Create the lttng run directory needed for all global sockets and pipe.
3708 */
3709 static int create_lttng_rundir(const char *rundir)
3710 {
3711 int ret;
3712
3713 DBG3("Creating LTTng run directory: %s", rundir);
3714
3715 ret = mkdir(rundir, S_IRWXU);
3716 if (ret < 0) {
3717 if (errno != EEXIST) {
3718 ERR("Unable to create %s", rundir);
3719 goto error;
3720 } else {
3721 ret = 0;
3722 }
3723 }
3724
3725 error:
3726 return ret;
3727 }
3728
3729 /*
3730 * Setup sockets and directory needed by the kconsumerd communication with the
3731 * session daemon.
3732 */
3733 static int set_consumer_sockets(struct consumer_data *consumer_data,
3734 const char *rundir)
3735 {
3736 int ret;
3737 char path[PATH_MAX];
3738
3739 switch (consumer_data->type) {
3740 case LTTNG_CONSUMER_KERNEL:
3741 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3742 break;
3743 case LTTNG_CONSUMER64_UST:
3744 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3745 break;
3746 case LTTNG_CONSUMER32_UST:
3747 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3748 break;
3749 default:
3750 ERR("Consumer type unknown");
3751 ret = -EINVAL;
3752 goto error;
3753 }
3754
3755 DBG2("Creating consumer directory: %s", path);
3756
3757 ret = mkdir(path, S_IRWXU);
3758 if (ret < 0) {
3759 if (errno != EEXIST) {
3760 PERROR("mkdir");
3761 ERR("Failed to create %s", path);
3762 goto error;
3763 }
3764 ret = -1;
3765 }
3766
3767 /* Create the kconsumerd error unix socket */
3768 consumer_data->err_sock =
3769 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3770 if (consumer_data->err_sock < 0) {
3771 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3772 ret = -1;
3773 goto error;
3774 }
3775
3776 /* File permission MUST be 660 */
3777 ret = chmod(consumer_data->err_unix_sock_path,
3778 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3779 if (ret < 0) {
3780 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3781 PERROR("chmod");
3782 goto error;
3783 }
3784
3785 error:
3786 return ret;
3787 }
3788
3789 /*
3790 * Signal handler for the daemon
3791 *
3792 * Simply stop all worker threads, leaving main() return gracefully after
3793 * joining all threads and calling cleanup().
3794 */
3795 static void sighandler(int sig)
3796 {
3797 switch (sig) {
3798 case SIGPIPE:
3799 DBG("SIGPIPE caught");
3800 return;
3801 case SIGINT:
3802 DBG("SIGINT caught");
3803 stop_threads();
3804 break;
3805 case SIGTERM:
3806 DBG("SIGTERM caught");
3807 stop_threads();
3808 break;
3809 default:
3810 break;
3811 }
3812 }
3813
3814 /*
3815 * Setup signal handler for :
3816 * SIGINT, SIGTERM, SIGPIPE
3817 */
3818 static int set_signal_handler(void)
3819 {
3820 int ret = 0;
3821 struct sigaction sa;
3822 sigset_t sigset;
3823
3824 if ((ret = sigemptyset(&sigset)) < 0) {
3825 PERROR("sigemptyset");
3826 return ret;
3827 }
3828
3829 sa.sa_handler = sighandler;
3830 sa.sa_mask = sigset;
3831 sa.sa_flags = 0;
3832 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3833 PERROR("sigaction");
3834 return ret;
3835 }
3836
3837 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3838 PERROR("sigaction");
3839 return ret;
3840 }
3841
3842 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3843 PERROR("sigaction");
3844 return ret;
3845 }
3846
3847 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3848
3849 return ret;
3850 }
3851
3852 /*
3853 * Set open files limit to unlimited. This daemon can open a large number of
3854 * file descriptors in order to consumer multiple kernel traces.
3855 */
3856 static void set_ulimit(void)
3857 {
3858 int ret;
3859 struct rlimit lim;
3860
3861 /* The kernel does not allowed an infinite limit for open files */
3862 lim.rlim_cur = 65535;
3863 lim.rlim_max = 65535;
3864
3865 ret = setrlimit(RLIMIT_NOFILE, &lim);
3866 if (ret < 0) {
3867 PERROR("failed to set open files limit");
3868 }
3869 }
3870
3871 /*
3872 * Write pidfile using the rundir and opt_pidfile.
3873 */
3874 static void write_pidfile(void)
3875 {
3876 int ret;
3877 char pidfile_path[PATH_MAX];
3878
3879 assert(rundir);
3880
3881 if (opt_pidfile) {
3882 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
3883 } else {
3884 /* Build pidfile path from rundir and opt_pidfile. */
3885 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
3886 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
3887 if (ret < 0) {
3888 PERROR("snprintf pidfile path");
3889 goto error;
3890 }
3891 }
3892
3893 /*
3894 * Create pid file in rundir. Return value is of no importance. The
3895 * execution will continue even though we are not able to write the file.
3896 */
3897 (void) utils_create_pid_file(getpid(), pidfile_path);
3898
3899 error:
3900 return;
3901 }
3902
3903 /*
3904 * main
3905 */
3906 int main(int argc, char **argv)
3907 {
3908 int ret = 0;
3909 void *status;
3910 const char *home_path, *env_app_timeout;
3911
3912 init_kernel_workarounds();
3913
3914 rcu_register_thread();
3915
3916 setup_consumerd_path();
3917
3918 /* Parse arguments */
3919 progname = argv[0];
3920 if ((ret = parse_args(argc, argv)) < 0) {
3921 goto error;
3922 }
3923
3924 /* Daemonize */
3925 if (opt_daemon) {
3926 int i;
3927
3928 /*
3929 * fork
3930 * child: setsid, close FD 0, 1, 2, chdir /
3931 * parent: exit (if fork is successful)
3932 */
3933 ret = daemon(0, 0);
3934 if (ret < 0) {
3935 PERROR("daemon");
3936 goto error;
3937 }
3938 /*
3939 * We are in the child. Make sure all other file
3940 * descriptors are closed, in case we are called with
3941 * more opened file descriptors than the standard ones.
3942 */
3943 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3944 (void) close(i);
3945 }
3946 }
3947
3948 /* Create thread quit pipe */
3949 if ((ret = init_thread_quit_pipe()) < 0) {
3950 goto error;
3951 }
3952
3953 /* Check if daemon is UID = 0 */
3954 is_root = !getuid();
3955
3956 if (is_root) {
3957 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3958
3959 /* Create global run dir with root access */
3960 ret = create_lttng_rundir(rundir);
3961 if (ret < 0) {
3962 goto error;
3963 }
3964
3965 if (strlen(apps_unix_sock_path) == 0) {
3966 snprintf(apps_unix_sock_path, PATH_MAX,
3967 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3968 }
3969
3970 if (strlen(client_unix_sock_path) == 0) {
3971 snprintf(client_unix_sock_path, PATH_MAX,
3972 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3973 }
3974
3975 /* Set global SHM for ust */
3976 if (strlen(wait_shm_path) == 0) {
3977 snprintf(wait_shm_path, PATH_MAX,
3978 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3979 }
3980
3981 if (strlen(health_unix_sock_path) == 0) {
3982 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3983 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3984 }
3985
3986 /* Setup kernel consumerd path */
3987 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3988 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3989 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3990 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3991
3992 DBG2("Kernel consumer err path: %s",
3993 kconsumer_data.err_unix_sock_path);
3994 DBG2("Kernel consumer cmd path: %s",
3995 kconsumer_data.cmd_unix_sock_path);
3996 } else {
3997 home_path = get_home_dir();
3998 if (home_path == NULL) {
3999 /* TODO: Add --socket PATH option */
4000 ERR("Can't get HOME directory for sockets creation.");
4001 ret = -EPERM;
4002 goto error;
4003 }
4004
4005 /*
4006 * Create rundir from home path. This will create something like
4007 * $HOME/.lttng
4008 */
4009 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4010 if (ret < 0) {
4011 ret = -ENOMEM;
4012 goto error;
4013 }
4014
4015 ret = create_lttng_rundir(rundir);
4016 if (ret < 0) {
4017 goto error;
4018 }
4019
4020 if (strlen(apps_unix_sock_path) == 0) {
4021 snprintf(apps_unix_sock_path, PATH_MAX,
4022 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4023 }
4024
4025 /* Set the cli tool unix socket path */
4026 if (strlen(client_unix_sock_path) == 0) {
4027 snprintf(client_unix_sock_path, PATH_MAX,
4028 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4029 }
4030
4031 /* Set global SHM for ust */
4032 if (strlen(wait_shm_path) == 0) {
4033 snprintf(wait_shm_path, PATH_MAX,
4034 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
4035 }
4036
4037 /* Set health check Unix path */
4038 if (strlen(health_unix_sock_path) == 0) {
4039 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4040 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4041 }
4042 }
4043
4044 /* Set consumer initial state */
4045 kernel_consumerd_state = CONSUMER_STOPPED;
4046 ust_consumerd_state = CONSUMER_STOPPED;
4047
4048 DBG("Client socket path %s", client_unix_sock_path);
4049 DBG("Application socket path %s", apps_unix_sock_path);
4050 DBG("LTTng run directory path: %s", rundir);
4051
4052 /* 32 bits consumerd path setup */
4053 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4054 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4055 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4056 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4057
4058 DBG2("UST consumer 32 bits err path: %s",
4059 ustconsumer32_data.err_unix_sock_path);
4060 DBG2("UST consumer 32 bits cmd path: %s",
4061 ustconsumer32_data.cmd_unix_sock_path);
4062
4063 /* 64 bits consumerd path setup */
4064 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4065 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4066 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4067 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4068
4069 DBG2("UST consumer 64 bits err path: %s",
4070 ustconsumer64_data.err_unix_sock_path);
4071 DBG2("UST consumer 64 bits cmd path: %s",
4072 ustconsumer64_data.cmd_unix_sock_path);
4073
4074 /*
4075 * See if daemon already exist.
4076 */
4077 if ((ret = check_existing_daemon()) < 0) {
4078 ERR("Already running daemon.\n");
4079 /*
4080 * We do not goto exit because we must not cleanup()
4081 * because a daemon is already running.
4082 */
4083 goto error;
4084 }
4085
4086 /*
4087 * Init UST app hash table. Alloc hash table before this point since
4088 * cleanup() can get called after that point.
4089 */
4090 ust_app_ht_alloc();
4091
4092 /* After this point, we can safely call cleanup() with "goto exit" */
4093
4094 /*
4095 * These actions must be executed as root. We do that *after* setting up
4096 * the sockets path because we MUST make the check for another daemon using
4097 * those paths *before* trying to set the kernel consumer sockets and init
4098 * kernel tracer.
4099 */
4100 if (is_root) {
4101 ret = set_consumer_sockets(&kconsumer_data, rundir);
4102 if (ret < 0) {
4103 goto exit;
4104 }
4105
4106 /* Setup kernel tracer */
4107 if (!opt_no_kernel) {
4108 init_kernel_tracer();
4109 }
4110
4111 /* Set ulimit for open files */
4112 set_ulimit();
4113 }
4114 /* init lttng_fd tracking must be done after set_ulimit. */
4115 lttng_fd_init();
4116
4117 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4118 if (ret < 0) {
4119 goto exit;
4120 }
4121
4122 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4123 if (ret < 0) {
4124 goto exit;
4125 }
4126
4127 if ((ret = set_signal_handler()) < 0) {
4128 goto exit;
4129 }
4130
4131 /* Setup the needed unix socket */
4132 if ((ret = init_daemon_socket()) < 0) {
4133 goto exit;
4134 }
4135
4136 /* Set credentials to socket */
4137 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4138 goto exit;
4139 }
4140
4141 /* Get parent pid if -S, --sig-parent is specified. */
4142 if (opt_sig_parent) {
4143 ppid = getppid();
4144 }
4145
4146 /* Setup the kernel pipe for waking up the kernel thread */
4147 if (is_root && !opt_no_kernel) {
4148 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4149 goto exit;
4150 }
4151 }
4152
4153 /* Setup the thread apps communication pipe. */
4154 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4155 goto exit;
4156 }
4157
4158 /* Init UST command queue. */
4159 cds_wfq_init(&ust_cmd_queue.queue);
4160
4161 /*
4162 * Get session list pointer. This pointer MUST NOT be free(). This list is
4163 * statically declared in session.c
4164 */
4165 session_list_ptr = session_get_list();
4166
4167 /* Set up max poll set size */
4168 lttng_poll_set_max_size();
4169
4170 cmd_init();
4171
4172 /* Check for the application socket timeout env variable. */
4173 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4174 if (env_app_timeout) {
4175 app_socket_timeout = atoi(env_app_timeout);
4176 } else {
4177 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4178 }
4179
4180 write_pidfile();
4181
4182 /* Create thread to manage the client socket */
4183 ret = pthread_create(&health_thread, NULL,
4184 thread_manage_health, (void *) NULL);
4185 if (ret != 0) {
4186 PERROR("pthread_create health");
4187 goto exit_health;
4188 }
4189
4190 /* Create thread to manage the client socket */
4191 ret = pthread_create(&client_thread, NULL,
4192 thread_manage_clients, (void *) NULL);
4193 if (ret != 0) {
4194 PERROR("pthread_create clients");
4195 goto exit_client;
4196 }
4197
4198 /* Create thread to dispatch registration */
4199 ret = pthread_create(&dispatch_thread, NULL,
4200 thread_dispatch_ust_registration, (void *) NULL);
4201 if (ret != 0) {
4202 PERROR("pthread_create dispatch");
4203 goto exit_dispatch;
4204 }
4205
4206 /* Create thread to manage application registration. */
4207 ret = pthread_create(&reg_apps_thread, NULL,
4208 thread_registration_apps, (void *) NULL);
4209 if (ret != 0) {
4210 PERROR("pthread_create registration");
4211 goto exit_reg_apps;
4212 }
4213
4214 /* Create thread to manage application socket */
4215 ret = pthread_create(&apps_thread, NULL,
4216 thread_manage_apps, (void *) NULL);
4217 if (ret != 0) {
4218 PERROR("pthread_create apps");
4219 goto exit_apps;
4220 }
4221
4222 /* Don't start this thread if kernel tracing is not requested nor root */
4223 if (is_root && !opt_no_kernel) {
4224 /* Create kernel thread to manage kernel event */
4225 ret = pthread_create(&kernel_thread, NULL,
4226 thread_manage_kernel, (void *) NULL);
4227 if (ret != 0) {
4228 PERROR("pthread_create kernel");
4229 goto exit_kernel;
4230 }
4231
4232 ret = pthread_join(kernel_thread, &status);
4233 if (ret != 0) {
4234 PERROR("pthread_join");
4235 goto error; /* join error, exit without cleanup */
4236 }
4237 }
4238
4239 exit_kernel:
4240 ret = pthread_join(apps_thread, &status);
4241 if (ret != 0) {
4242 PERROR("pthread_join");
4243 goto error; /* join error, exit without cleanup */
4244 }
4245
4246 exit_apps:
4247 ret = pthread_join(reg_apps_thread, &status);
4248 if (ret != 0) {
4249 PERROR("pthread_join");
4250 goto error; /* join error, exit without cleanup */
4251 }
4252
4253 exit_reg_apps:
4254 ret = pthread_join(dispatch_thread, &status);
4255 if (ret != 0) {
4256 PERROR("pthread_join");
4257 goto error; /* join error, exit without cleanup */
4258 }
4259
4260 exit_dispatch:
4261 ret = pthread_join(client_thread, &status);
4262 if (ret != 0) {
4263 PERROR("pthread_join");
4264 goto error; /* join error, exit without cleanup */
4265 }
4266
4267 ret = join_consumer_thread(&kconsumer_data);
4268 if (ret != 0) {
4269 PERROR("join_consumer");
4270 goto error; /* join error, exit without cleanup */
4271 }
4272
4273 ret = join_consumer_thread(&ustconsumer32_data);
4274 if (ret != 0) {
4275 PERROR("join_consumer ust32");
4276 goto error; /* join error, exit without cleanup */
4277 }
4278
4279 ret = join_consumer_thread(&ustconsumer64_data);
4280 if (ret != 0) {
4281 PERROR("join_consumer ust64");
4282 goto error; /* join error, exit without cleanup */
4283 }
4284
4285 exit_client:
4286 ret = pthread_join(health_thread, &status);
4287 if (ret != 0) {
4288 PERROR("pthread_join health thread");
4289 goto error; /* join error, exit without cleanup */
4290 }
4291
4292 exit_health:
4293 exit:
4294 /*
4295 * cleanup() is called when no other thread is running.
4296 */
4297 rcu_thread_online();
4298 cleanup();
4299 rcu_thread_offline();
4300 rcu_unregister_thread();
4301 if (!ret) {
4302 exit(EXIT_SUCCESS);
4303 }
4304 error:
4305 exit(EXIT_FAILURE);
4306 }
This page took 0.133507 seconds and 4 git commands to generate.