339b20d3ee433be2d17f33d3462b104ab6266448
[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 /* Used for the health monitoring of the session daemon. See health.h */
226 struct health_state health_thread_cmd;
227 struct health_state health_thread_app_manage;
228 struct health_state health_thread_app_reg;
229 struct health_state health_thread_kernel;
230
231 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 static
237 void setup_consumerd_path(void)
238 {
239 const char *bin, *libdir;
240
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
245 */
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
256 }
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
259 }
260 #else
261 #error "Unknown bitness"
262 #endif
263
264 /*
265 * runtime env. var. overrides the build default.
266 */
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
270 }
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
274 }
275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
280 if (libdir) {
281 consumerd64_libdir = libdir;
282 }
283 }
284
285 /*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288 static int create_thread_poll_set(struct lttng_poll_event *events,
289 unsigned int size)
290 {
291 int ret;
292
293 if (events == NULL || size == 0) {
294 ret = -1;
295 goto error;
296 }
297
298 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
299 if (ret < 0) {
300 goto error;
301 }
302
303 /* Add quit pipe */
304 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
305 if (ret < 0) {
306 goto error;
307 }
308
309 return 0;
310
311 error:
312 return ret;
313 }
314
315 /*
316 * Check if the thread quit pipe was triggered.
317 *
318 * Return 1 if it was triggered else 0;
319 */
320 static int check_thread_quit_pipe(int fd, uint32_t events)
321 {
322 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
323 return 1;
324 }
325
326 return 0;
327 }
328
329 /*
330 * Return group ID of the tracing group or -1 if not found.
331 */
332 static gid_t allowed_group(void)
333 {
334 struct group *grp;
335
336 if (opt_tracing_group) {
337 grp = getgrnam(opt_tracing_group);
338 } else {
339 grp = getgrnam(default_tracing_group);
340 }
341 if (!grp) {
342 return -1;
343 } else {
344 return grp->gr_gid;
345 }
346 }
347
348 /*
349 * Init thread quit pipe.
350 *
351 * Return -1 on error or 0 if all pipes are created.
352 */
353 static int init_thread_quit_pipe(void)
354 {
355 int ret, i;
356
357 ret = pipe(thread_quit_pipe);
358 if (ret < 0) {
359 PERROR("thread quit pipe");
360 goto error;
361 }
362
363 for (i = 0; i < 2; i++) {
364 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
365 if (ret < 0) {
366 PERROR("fcntl");
367 goto error;
368 }
369 }
370
371 error:
372 return ret;
373 }
374
375 /*
376 * Stop all threads by closing the thread quit pipe.
377 */
378 static void stop_threads(void)
379 {
380 int ret;
381
382 /* Stopping all threads */
383 DBG("Terminating all threads");
384 ret = notify_thread_pipe(thread_quit_pipe[1]);
385 if (ret < 0) {
386 ERR("write error on thread quit pipe");
387 }
388
389 /* Dispatch thread */
390 CMM_STORE_SHARED(dispatch_thread_exit, 1);
391 futex_nto1_wake(&ust_cmd_queue.futex);
392 }
393
394 /*
395 * Cleanup the daemon
396 */
397 static void cleanup(void)
398 {
399 int ret;
400 char *cmd = NULL;
401 struct ltt_session *sess, *stmp;
402
403 DBG("Cleaning up");
404
405 /* First thing first, stop all threads */
406 utils_close_pipe(thread_quit_pipe);
407
408 /*
409 * If opt_pidfile is undefined, the default file will be wiped when
410 * removing the rundir.
411 */
412 if (opt_pidfile) {
413 ret = remove(opt_pidfile);
414 if (ret < 0) {
415 PERROR("remove pidfile %s", opt_pidfile);
416 }
417 }
418
419 DBG("Removing %s directory", rundir);
420 ret = asprintf(&cmd, "rm -rf %s", rundir);
421 if (ret < 0) {
422 ERR("asprintf failed. Something is really wrong!");
423 }
424
425 /* Remove lttng run directory */
426 ret = system(cmd);
427 if (ret < 0) {
428 ERR("Unable to clean %s", rundir);
429 }
430 free(cmd);
431 free(rundir);
432
433 DBG("Cleaning up all sessions");
434
435 /* Destroy session list mutex */
436 if (session_list_ptr != NULL) {
437 pthread_mutex_destroy(&session_list_ptr->lock);
438
439 /* Cleanup ALL session */
440 cds_list_for_each_entry_safe(sess, stmp,
441 &session_list_ptr->head, list) {
442 cmd_destroy_session(sess, kernel_poll_pipe[1]);
443 }
444 }
445
446 DBG("Closing all UST sockets");
447 ust_app_clean_list();
448
449 if (is_root && !opt_no_kernel) {
450 DBG2("Closing kernel fd");
451 if (kernel_tracer_fd >= 0) {
452 ret = close(kernel_tracer_fd);
453 if (ret) {
454 PERROR("close");
455 }
456 }
457 DBG("Unloading kernel modules");
458 modprobe_remove_lttng_all();
459 }
460
461 /* <fun> */
462 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
463 "Matthew, BEET driven development works!%c[%dm",
464 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
465 /* </fun> */
466 }
467
468 /*
469 * Send data on a unix socket using the liblttsessiondcomm API.
470 *
471 * Return lttcomm error code.
472 */
473 static int send_unix_sock(int sock, void *buf, size_t len)
474 {
475 /* Check valid length */
476 if (len == 0) {
477 return -1;
478 }
479
480 return lttcomm_send_unix_sock(sock, buf, len);
481 }
482
483 /*
484 * Free memory of a command context structure.
485 */
486 static void clean_command_ctx(struct command_ctx **cmd_ctx)
487 {
488 DBG("Clean command context structure");
489 if (*cmd_ctx) {
490 if ((*cmd_ctx)->llm) {
491 free((*cmd_ctx)->llm);
492 }
493 if ((*cmd_ctx)->lsm) {
494 free((*cmd_ctx)->lsm);
495 }
496 free(*cmd_ctx);
497 *cmd_ctx = NULL;
498 }
499 }
500
501 /*
502 * Notify UST applications using the shm mmap futex.
503 */
504 static int notify_ust_apps(int active)
505 {
506 char *wait_shm_mmap;
507
508 DBG("Notifying applications of session daemon state: %d", active);
509
510 /* See shm.c for this call implying mmap, shm and futex calls */
511 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
512 if (wait_shm_mmap == NULL) {
513 goto error;
514 }
515
516 /* Wake waiting process */
517 futex_wait_update((int32_t *) wait_shm_mmap, active);
518
519 /* Apps notified successfully */
520 return 0;
521
522 error:
523 return -1;
524 }
525
526 /*
527 * Setup the outgoing data buffer for the response (llm) by allocating the
528 * right amount of memory and copying the original information from the lsm
529 * structure.
530 *
531 * Return total size of the buffer pointed by buf.
532 */
533 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
534 {
535 int ret, buf_size;
536
537 buf_size = size;
538
539 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
540 if (cmd_ctx->llm == NULL) {
541 PERROR("zmalloc");
542 ret = -ENOMEM;
543 goto error;
544 }
545
546 /* Copy common data */
547 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
548 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
549
550 cmd_ctx->llm->data_size = size;
551 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
552
553 return buf_size;
554
555 error:
556 return ret;
557 }
558
559 /*
560 * Update the kernel poll set of all channel fd available over all tracing
561 * session. Add the wakeup pipe at the end of the set.
562 */
563 static int update_kernel_poll(struct lttng_poll_event *events)
564 {
565 int ret;
566 struct ltt_session *session;
567 struct ltt_kernel_channel *channel;
568
569 DBG("Updating kernel poll set");
570
571 session_lock_list();
572 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
573 session_lock(session);
574 if (session->kernel_session == NULL) {
575 session_unlock(session);
576 continue;
577 }
578
579 cds_list_for_each_entry(channel,
580 &session->kernel_session->channel_list.head, list) {
581 /* Add channel fd to the kernel poll set */
582 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
583 if (ret < 0) {
584 session_unlock(session);
585 goto error;
586 }
587 DBG("Channel fd %d added to kernel set", channel->fd);
588 }
589 session_unlock(session);
590 }
591 session_unlock_list();
592
593 return 0;
594
595 error:
596 session_unlock_list();
597 return -1;
598 }
599
600 /*
601 * Find the channel fd from 'fd' over all tracing session. When found, check
602 * for new channel stream and send those stream fds to the kernel consumer.
603 *
604 * Useful for CPU hotplug feature.
605 */
606 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
607 {
608 int ret = 0;
609 struct ltt_session *session;
610 struct ltt_kernel_session *ksess;
611 struct ltt_kernel_channel *channel;
612
613 DBG("Updating kernel streams for channel fd %d", fd);
614
615 session_lock_list();
616 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
617 session_lock(session);
618 if (session->kernel_session == NULL) {
619 session_unlock(session);
620 continue;
621 }
622 ksess = session->kernel_session;
623
624 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
625 if (channel->fd == fd) {
626 DBG("Channel found, updating kernel streams");
627 ret = kernel_open_channel_stream(channel);
628 if (ret < 0) {
629 goto error;
630 }
631
632 /*
633 * Have we already sent fds to the consumer? If yes, it means
634 * that tracing is started so it is safe to send our updated
635 * stream fds.
636 */
637 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
638 struct lttng_ht_iter iter;
639 struct consumer_socket *socket;
640
641 rcu_read_lock();
642 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
643 &iter.iter, socket, node.node) {
644 /* Code flow error */
645 assert(socket->fd >= 0);
646
647 pthread_mutex_lock(socket->lock);
648 ret = kernel_consumer_send_channel_stream(socket,
649 channel, ksess);
650 pthread_mutex_unlock(socket->lock);
651 if (ret < 0) {
652 rcu_read_unlock();
653 goto error;
654 }
655 }
656 rcu_read_unlock();
657 }
658 goto error;
659 }
660 }
661 session_unlock(session);
662 }
663 session_unlock_list();
664 return ret;
665
666 error:
667 session_unlock(session);
668 session_unlock_list();
669 return ret;
670 }
671
672 /*
673 * For each tracing session, update newly registered apps.
674 */
675 static void update_ust_app(int app_sock)
676 {
677 struct ltt_session *sess, *stmp;
678
679 session_lock_list();
680
681 /* For all tracing session(s) */
682 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
683 session_lock(sess);
684 if (sess->ust_session) {
685 ust_app_global_update(sess->ust_session, app_sock);
686 }
687 session_unlock(sess);
688 }
689
690 session_unlock_list();
691 }
692
693 /*
694 * This thread manage event coming from the kernel.
695 *
696 * Features supported in this thread:
697 * -) CPU Hotplug
698 */
699 static void *thread_manage_kernel(void *data)
700 {
701 int ret, i, pollfd, update_poll_flag = 1, err = -1;
702 uint32_t revents, nb_fd;
703 char tmp;
704 struct lttng_poll_event events;
705
706 DBG("[thread] Thread manage kernel started");
707
708 /*
709 * This first step of the while is to clean this structure which could free
710 * non NULL pointers so zero it before the loop.
711 */
712 memset(&events, 0, sizeof(events));
713
714 if (testpoint(thread_manage_kernel)) {
715 goto error_testpoint;
716 }
717
718 health_code_update(&health_thread_kernel);
719
720 if (testpoint(thread_manage_kernel_before_loop)) {
721 goto error_testpoint;
722 }
723
724 while (1) {
725 health_code_update(&health_thread_kernel);
726
727 if (update_poll_flag == 1) {
728 /* Clean events object. We are about to populate it again. */
729 lttng_poll_clean(&events);
730
731 ret = create_thread_poll_set(&events, 2);
732 if (ret < 0) {
733 goto error_poll_create;
734 }
735
736 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
737 if (ret < 0) {
738 goto error;
739 }
740
741 /* This will add the available kernel channel if any. */
742 ret = update_kernel_poll(&events);
743 if (ret < 0) {
744 goto error;
745 }
746 update_poll_flag = 0;
747 }
748
749 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
750
751 /* Poll infinite value of time */
752 restart:
753 health_poll_update(&health_thread_kernel);
754 ret = lttng_poll_wait(&events, -1);
755 health_poll_update(&health_thread_kernel);
756 if (ret < 0) {
757 /*
758 * Restart interrupted system call.
759 */
760 if (errno == EINTR) {
761 goto restart;
762 }
763 goto error;
764 } else if (ret == 0) {
765 /* Should not happen since timeout is infinite */
766 ERR("Return value of poll is 0 with an infinite timeout.\n"
767 "This should not have happened! Continuing...");
768 continue;
769 }
770
771 nb_fd = ret;
772
773 for (i = 0; i < nb_fd; i++) {
774 /* Fetch once the poll data */
775 revents = LTTNG_POLL_GETEV(&events, i);
776 pollfd = LTTNG_POLL_GETFD(&events, i);
777
778 health_code_update(&health_thread_kernel);
779
780 /* Thread quit pipe has been closed. Killing thread. */
781 ret = check_thread_quit_pipe(pollfd, revents);
782 if (ret) {
783 err = 0;
784 goto exit;
785 }
786
787 /* Check for data on kernel pipe */
788 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
789 do {
790 ret = read(kernel_poll_pipe[0], &tmp, 1);
791 } while (ret < 0 && errno == EINTR);
792 /*
793 * Ret value is useless here, if this pipe gets any actions an
794 * update is required anyway.
795 */
796 update_poll_flag = 1;
797 continue;
798 } else {
799 /*
800 * New CPU detected by the kernel. Adding kernel stream to
801 * kernel session and updating the kernel consumer
802 */
803 if (revents & LPOLLIN) {
804 ret = update_kernel_stream(&kconsumer_data, pollfd);
805 if (ret < 0) {
806 continue;
807 }
808 break;
809 /*
810 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
811 * and unregister kernel stream at this point.
812 */
813 }
814 }
815 }
816 }
817
818 exit:
819 error:
820 lttng_poll_clean(&events);
821 error_poll_create:
822 error_testpoint:
823 utils_close_pipe(kernel_poll_pipe);
824 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
825 if (err) {
826 health_error(&health_thread_kernel);
827 ERR("Health error occurred in %s", __func__);
828 WARN("Kernel thread died unexpectedly. "
829 "Kernel tracing can continue but CPU hotplug is disabled.");
830 }
831 health_exit(&health_thread_kernel);
832 DBG("Kernel thread dying");
833 return NULL;
834 }
835
836 /*
837 * Signal pthread condition of the consumer data that the thread.
838 */
839 static void signal_consumer_condition(struct consumer_data *data, int state)
840 {
841 pthread_mutex_lock(&data->cond_mutex);
842
843 /*
844 * The state is set before signaling. It can be any value, it's the waiter
845 * job to correctly interpret this condition variable associated to the
846 * consumer pthread_cond.
847 *
848 * A value of 0 means that the corresponding thread of the consumer data
849 * was not started. 1 indicates that the thread has started and is ready
850 * for action. A negative value means that there was an error during the
851 * thread bootstrap.
852 */
853 data->consumer_thread_is_ready = state;
854 (void) pthread_cond_signal(&data->cond);
855
856 pthread_mutex_unlock(&data->cond_mutex);
857 }
858
859 /*
860 * This thread manage the consumer error sent back to the session daemon.
861 */
862 static void *thread_manage_consumer(void *data)
863 {
864 int sock = -1, i, ret, pollfd, err = -1;
865 uint32_t revents, nb_fd;
866 enum lttcomm_return_code code;
867 struct lttng_poll_event events;
868 struct consumer_data *consumer_data = data;
869
870 DBG("[thread] Manage consumer started");
871
872 /*
873 * Since the consumer thread can be spawned at any moment in time, we init
874 * the health to a poll status (1, which is a valid health over time).
875 * When the thread starts, we update here the health to a "code" path being
876 * an even value so this thread, when reaching a poll wait, does not
877 * trigger an error with an even value.
878 *
879 * Here is the use case we avoid.
880 *
881 * +1: the first poll update during initialization (main())
882 * +2 * x: multiple code update once in this thread.
883 * +1: poll wait in this thread (being a good health state).
884 * == even number which after the wait period shows as a bad health.
885 *
886 * In a nutshell, the following poll update to the health state brings back
887 * the state to an even value meaning a code path.
888 */
889 health_poll_update(&consumer_data->health);
890
891 /*
892 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
893 * Nothing more will be added to this poll set.
894 */
895 ret = create_thread_poll_set(&events, 2);
896 if (ret < 0) {
897 goto error_poll;
898 }
899
900 /*
901 * The error socket here is already in a listening state which was done
902 * just before spawning this thread to avoid a race between the consumer
903 * daemon exec trying to connect and the listen() call.
904 */
905 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
906 if (ret < 0) {
907 goto error;
908 }
909
910 health_code_update(&consumer_data->health);
911
912 /* Inifinite blocking call, waiting for transmission */
913 restart:
914 health_poll_update(&consumer_data->health);
915
916 if (testpoint(thread_manage_consumer)) {
917 goto error;
918 }
919
920 ret = lttng_poll_wait(&events, -1);
921 health_poll_update(&consumer_data->health);
922 if (ret < 0) {
923 /*
924 * Restart interrupted system call.
925 */
926 if (errno == EINTR) {
927 goto restart;
928 }
929 goto error;
930 }
931
932 nb_fd = ret;
933
934 for (i = 0; i < nb_fd; i++) {
935 /* Fetch once the poll data */
936 revents = LTTNG_POLL_GETEV(&events, i);
937 pollfd = LTTNG_POLL_GETFD(&events, i);
938
939 health_code_update(&consumer_data->health);
940
941 /* Thread quit pipe has been closed. Killing thread. */
942 ret = check_thread_quit_pipe(pollfd, revents);
943 if (ret) {
944 err = 0;
945 goto exit;
946 }
947
948 /* Event on the registration socket */
949 if (pollfd == consumer_data->err_sock) {
950 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
951 ERR("consumer err socket poll error");
952 goto error;
953 }
954 }
955 }
956
957 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
958 if (sock < 0) {
959 goto error;
960 }
961
962 /*
963 * Set the CLOEXEC flag. Return code is useless because either way, the
964 * show must go on.
965 */
966 (void) utils_set_fd_cloexec(sock);
967
968 health_code_update(&consumer_data->health);
969
970 DBG2("Receiving code from consumer err_sock");
971
972 /* Getting status code from kconsumerd */
973 ret = lttcomm_recv_unix_sock(sock, &code,
974 sizeof(enum lttcomm_return_code));
975 if (ret <= 0) {
976 goto error;
977 }
978
979 health_code_update(&consumer_data->health);
980
981 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
982 consumer_data->cmd_sock =
983 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
984 if (consumer_data->cmd_sock < 0) {
985 /* On error, signal condition and quit. */
986 signal_consumer_condition(consumer_data, -1);
987 PERROR("consumer connect");
988 goto error;
989 }
990 signal_consumer_condition(consumer_data, 1);
991 DBG("Consumer command socket ready");
992 } else {
993 ERR("consumer error when waiting for SOCK_READY : %s",
994 lttcomm_get_readable_code(-code));
995 goto error;
996 }
997
998 /* Remove the kconsumerd error sock since we've established a connexion */
999 ret = lttng_poll_del(&events, consumer_data->err_sock);
1000 if (ret < 0) {
1001 goto error;
1002 }
1003
1004 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1005 if (ret < 0) {
1006 goto error;
1007 }
1008
1009 health_code_update(&consumer_data->health);
1010
1011 /* Inifinite blocking call, waiting for transmission */
1012 restart_poll:
1013 health_poll_update(&consumer_data->health);
1014 ret = lttng_poll_wait(&events, -1);
1015 health_poll_update(&consumer_data->health);
1016 if (ret < 0) {
1017 /*
1018 * Restart interrupted system call.
1019 */
1020 if (errno == EINTR) {
1021 goto restart_poll;
1022 }
1023 goto error;
1024 }
1025
1026 nb_fd = ret;
1027
1028 for (i = 0; i < nb_fd; i++) {
1029 /* Fetch once the poll data */
1030 revents = LTTNG_POLL_GETEV(&events, i);
1031 pollfd = LTTNG_POLL_GETFD(&events, i);
1032
1033 health_code_update(&consumer_data->health);
1034
1035 /* Thread quit pipe has been closed. Killing thread. */
1036 ret = check_thread_quit_pipe(pollfd, revents);
1037 if (ret) {
1038 err = 0;
1039 goto exit;
1040 }
1041
1042 /* Event on the kconsumerd socket */
1043 if (pollfd == sock) {
1044 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1045 ERR("consumer err socket second poll error");
1046 goto error;
1047 }
1048 }
1049 }
1050
1051 health_code_update(&consumer_data->health);
1052
1053 /* Wait for any kconsumerd error */
1054 ret = lttcomm_recv_unix_sock(sock, &code,
1055 sizeof(enum lttcomm_return_code));
1056 if (ret <= 0) {
1057 ERR("consumer closed the command socket");
1058 goto error;
1059 }
1060
1061 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1062
1063 exit:
1064 error:
1065 /* Immediately set the consumerd state to stopped */
1066 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1067 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1068 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1069 consumer_data->type == LTTNG_CONSUMER32_UST) {
1070 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1071 } else {
1072 /* Code flow error... */
1073 assert(0);
1074 }
1075
1076 if (consumer_data->err_sock >= 0) {
1077 ret = close(consumer_data->err_sock);
1078 if (ret) {
1079 PERROR("close");
1080 }
1081 }
1082 if (consumer_data->cmd_sock >= 0) {
1083 ret = close(consumer_data->cmd_sock);
1084 if (ret) {
1085 PERROR("close");
1086 }
1087 }
1088 if (sock >= 0) {
1089 ret = close(sock);
1090 if (ret) {
1091 PERROR("close");
1092 }
1093 }
1094
1095 unlink(consumer_data->err_unix_sock_path);
1096 unlink(consumer_data->cmd_unix_sock_path);
1097 consumer_data->pid = 0;
1098
1099 lttng_poll_clean(&events);
1100 error_poll:
1101 if (err) {
1102 health_error(&consumer_data->health);
1103 ERR("Health error occurred in %s", __func__);
1104 }
1105 health_exit(&consumer_data->health);
1106 DBG("consumer thread cleanup completed");
1107
1108 return NULL;
1109 }
1110
1111 /*
1112 * This thread manage application communication.
1113 */
1114 static void *thread_manage_apps(void *data)
1115 {
1116 int i, ret, pollfd, err = -1;
1117 uint32_t revents, nb_fd;
1118 struct ust_command ust_cmd;
1119 struct lttng_poll_event events;
1120
1121 DBG("[thread] Manage application started");
1122
1123 rcu_register_thread();
1124 rcu_thread_online();
1125
1126 if (testpoint(thread_manage_apps)) {
1127 goto error_testpoint;
1128 }
1129
1130 health_code_update(&health_thread_app_manage);
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(&health_thread_app_manage);
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(&health_thread_app_manage);
1154 ret = lttng_poll_wait(&events, -1);
1155 health_poll_update(&health_thread_app_manage);
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(&health_thread_app_manage);
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(&health_thread_app_manage);
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(&health_thread_app_manage);
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(&health_thread_app_manage);
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(&health_thread_app_manage);
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(&health_thread_app_manage);
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(&health_thread_app_manage);
1295 ERR("Health error occurred in %s", __func__);
1296 }
1297 health_exit(&health_thread_app_manage);
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 if (testpoint(thread_registration_apps)) {
1395 goto error_testpoint;
1396 }
1397
1398 ret = lttcomm_listen_unix_sock(apps_sock);
1399 if (ret < 0) {
1400 goto error_listen;
1401 }
1402
1403 /*
1404 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1405 * more will be added to this poll set.
1406 */
1407 ret = create_thread_poll_set(&events, 2);
1408 if (ret < 0) {
1409 goto error_create_poll;
1410 }
1411
1412 /* Add the application registration socket */
1413 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1414 if (ret < 0) {
1415 goto error_poll_add;
1416 }
1417
1418 /* Notify all applications to register */
1419 ret = notify_ust_apps(1);
1420 if (ret < 0) {
1421 ERR("Failed to notify applications or create the wait shared memory.\n"
1422 "Execution continues but there might be problem for already\n"
1423 "running applications that wishes to register.");
1424 }
1425
1426 while (1) {
1427 DBG("Accepting application registration");
1428
1429 /* Inifinite blocking call, waiting for transmission */
1430 restart:
1431 health_poll_update(&health_thread_app_reg);
1432 ret = lttng_poll_wait(&events, -1);
1433 health_poll_update(&health_thread_app_reg);
1434 if (ret < 0) {
1435 /*
1436 * Restart interrupted system call.
1437 */
1438 if (errno == EINTR) {
1439 goto restart;
1440 }
1441 goto error;
1442 }
1443
1444 nb_fd = ret;
1445
1446 for (i = 0; i < nb_fd; i++) {
1447 health_code_update(&health_thread_app_reg);
1448
1449 /* Fetch once the poll data */
1450 revents = LTTNG_POLL_GETEV(&events, i);
1451 pollfd = LTTNG_POLL_GETFD(&events, i);
1452
1453 /* Thread quit pipe has been closed. Killing thread. */
1454 ret = check_thread_quit_pipe(pollfd, revents);
1455 if (ret) {
1456 err = 0;
1457 goto exit;
1458 }
1459
1460 /* Event on the registration socket */
1461 if (pollfd == apps_sock) {
1462 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1463 ERR("Register apps socket poll error");
1464 goto error;
1465 } else if (revents & LPOLLIN) {
1466 sock = lttcomm_accept_unix_sock(apps_sock);
1467 if (sock < 0) {
1468 goto error;
1469 }
1470
1471 /*
1472 * Set the CLOEXEC flag. Return code is useless because
1473 * either way, the show must go on.
1474 */
1475 (void) utils_set_fd_cloexec(sock);
1476
1477 /* Create UST registration command for enqueuing */
1478 ust_cmd = zmalloc(sizeof(struct ust_command));
1479 if (ust_cmd == NULL) {
1480 PERROR("ust command zmalloc");
1481 goto error;
1482 }
1483
1484 /*
1485 * Using message-based transmissions to ensure we don't
1486 * have to deal with partially received messages.
1487 */
1488 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1489 if (ret < 0) {
1490 ERR("Exhausted file descriptors allowed for applications.");
1491 free(ust_cmd);
1492 ret = close(sock);
1493 if (ret) {
1494 PERROR("close");
1495 }
1496 sock = -1;
1497 continue;
1498 }
1499 health_code_update(&health_thread_app_reg);
1500 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1501 sizeof(struct ust_register_msg));
1502 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1503 if (ret < 0) {
1504 PERROR("lttcomm_recv_unix_sock register apps");
1505 } else {
1506 ERR("Wrong size received on apps register");
1507 }
1508 free(ust_cmd);
1509 ret = close(sock);
1510 if (ret) {
1511 PERROR("close");
1512 }
1513 lttng_fd_put(LTTNG_FD_APPS, 1);
1514 sock = -1;
1515 continue;
1516 }
1517 health_code_update(&health_thread_app_reg);
1518
1519 ust_cmd->sock = sock;
1520 sock = -1;
1521
1522 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1523 " gid:%d sock:%d name:%s (version %d.%d)",
1524 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1525 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1526 ust_cmd->sock, ust_cmd->reg_msg.name,
1527 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1528
1529 /*
1530 * Lock free enqueue the registration request. The red pill
1531 * has been taken! This apps will be part of the *system*.
1532 */
1533 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1534
1535 /*
1536 * Wake the registration queue futex. Implicit memory
1537 * barrier with the exchange in cds_wfq_enqueue.
1538 */
1539 futex_nto1_wake(&ust_cmd_queue.futex);
1540 }
1541 }
1542 }
1543 }
1544
1545 exit:
1546 error:
1547 if (err) {
1548 health_error(&health_thread_app_reg);
1549 ERR("Health error occurred in %s", __func__);
1550 }
1551
1552 /* Notify that the registration thread is gone */
1553 notify_ust_apps(0);
1554
1555 if (apps_sock >= 0) {
1556 ret = close(apps_sock);
1557 if (ret) {
1558 PERROR("close");
1559 }
1560 }
1561 if (sock >= 0) {
1562 ret = close(sock);
1563 if (ret) {
1564 PERROR("close");
1565 }
1566 lttng_fd_put(LTTNG_FD_APPS, 1);
1567 }
1568 unlink(apps_unix_sock_path);
1569
1570 error_poll_add:
1571 lttng_poll_clean(&events);
1572 error_listen:
1573 error_create_poll:
1574 error_testpoint:
1575 DBG("UST Registration thread cleanup complete");
1576 health_exit(&health_thread_app_reg);
1577
1578 return NULL;
1579 }
1580
1581 /*
1582 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1583 * exec or it will fails.
1584 */
1585 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1586 {
1587 int ret, clock_ret;
1588 struct timespec timeout;
1589
1590 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1591 consumer_data->consumer_thread_is_ready = 0;
1592
1593 /* Setup pthread condition */
1594 ret = pthread_condattr_init(&consumer_data->condattr);
1595 if (ret != 0) {
1596 errno = ret;
1597 PERROR("pthread_condattr_init consumer data");
1598 goto error;
1599 }
1600
1601 /*
1602 * Set the monotonic clock in order to make sure we DO NOT jump in time
1603 * between the clock_gettime() call and the timedwait call. See bug #324
1604 * for a more details and how we noticed it.
1605 */
1606 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1607 if (ret != 0) {
1608 errno = ret;
1609 PERROR("pthread_condattr_setclock consumer data");
1610 goto error;
1611 }
1612
1613 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1614 if (ret != 0) {
1615 errno = ret;
1616 PERROR("pthread_cond_init consumer data");
1617 goto error;
1618 }
1619
1620 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1621 consumer_data);
1622 if (ret != 0) {
1623 PERROR("pthread_create consumer");
1624 ret = -1;
1625 goto error;
1626 }
1627
1628 /* We are about to wait on a pthread condition */
1629 pthread_mutex_lock(&consumer_data->cond_mutex);
1630
1631 /* Get time for sem_timedwait absolute timeout */
1632 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1633 /*
1634 * Set the timeout for the condition timed wait even if the clock gettime
1635 * call fails since we might loop on that call and we want to avoid to
1636 * increment the timeout too many times.
1637 */
1638 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1639
1640 /*
1641 * The following loop COULD be skipped in some conditions so this is why we
1642 * set ret to 0 in order to make sure at least one round of the loop is
1643 * done.
1644 */
1645 ret = 0;
1646
1647 /*
1648 * Loop until the condition is reached or when a timeout is reached. Note
1649 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1650 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1651 * possible. This loop does not take any chances and works with both of
1652 * them.
1653 */
1654 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1655 if (clock_ret < 0) {
1656 PERROR("clock_gettime spawn consumer");
1657 /* Infinite wait for the consumerd thread to be ready */
1658 ret = pthread_cond_wait(&consumer_data->cond,
1659 &consumer_data->cond_mutex);
1660 } else {
1661 ret = pthread_cond_timedwait(&consumer_data->cond,
1662 &consumer_data->cond_mutex, &timeout);
1663 }
1664 }
1665
1666 /* Release the pthread condition */
1667 pthread_mutex_unlock(&consumer_data->cond_mutex);
1668
1669 if (ret != 0) {
1670 errno = ret;
1671 if (ret == ETIMEDOUT) {
1672 /*
1673 * Call has timed out so we kill the kconsumerd_thread and return
1674 * an error.
1675 */
1676 ERR("Condition timed out. The consumer thread was never ready."
1677 " Killing it");
1678 ret = pthread_cancel(consumer_data->thread);
1679 if (ret < 0) {
1680 PERROR("pthread_cancel consumer thread");
1681 }
1682 } else {
1683 PERROR("pthread_cond_wait failed consumer thread");
1684 }
1685 goto error;
1686 }
1687
1688 pthread_mutex_lock(&consumer_data->pid_mutex);
1689 if (consumer_data->pid == 0) {
1690 ERR("Consumerd did not start");
1691 pthread_mutex_unlock(&consumer_data->pid_mutex);
1692 goto error;
1693 }
1694 pthread_mutex_unlock(&consumer_data->pid_mutex);
1695
1696 return 0;
1697
1698 error:
1699 return ret;
1700 }
1701
1702 /*
1703 * Join consumer thread
1704 */
1705 static int join_consumer_thread(struct consumer_data *consumer_data)
1706 {
1707 void *status;
1708
1709 /* Consumer pid must be a real one. */
1710 if (consumer_data->pid > 0) {
1711 int ret;
1712 ret = kill(consumer_data->pid, SIGTERM);
1713 if (ret) {
1714 ERR("Error killing consumer daemon");
1715 return ret;
1716 }
1717 return pthread_join(consumer_data->thread, &status);
1718 } else {
1719 return 0;
1720 }
1721 }
1722
1723 /*
1724 * Fork and exec a consumer daemon (consumerd).
1725 *
1726 * Return pid if successful else -1.
1727 */
1728 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1729 {
1730 int ret;
1731 pid_t pid;
1732 const char *consumer_to_use;
1733 const char *verbosity;
1734 struct stat st;
1735
1736 DBG("Spawning consumerd");
1737
1738 pid = fork();
1739 if (pid == 0) {
1740 /*
1741 * Exec consumerd.
1742 */
1743 if (opt_verbose_consumer) {
1744 verbosity = "--verbose";
1745 } else {
1746 verbosity = "--quiet";
1747 }
1748 switch (consumer_data->type) {
1749 case LTTNG_CONSUMER_KERNEL:
1750 /*
1751 * Find out which consumerd to execute. We will first try the
1752 * 64-bit path, then the sessiond's installation directory, and
1753 * fallback on the 32-bit one,
1754 */
1755 DBG3("Looking for a kernel consumer at these locations:");
1756 DBG3(" 1) %s", consumerd64_bin);
1757 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1758 DBG3(" 3) %s", consumerd32_bin);
1759 if (stat(consumerd64_bin, &st) == 0) {
1760 DBG3("Found location #1");
1761 consumer_to_use = consumerd64_bin;
1762 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1763 DBG3("Found location #2");
1764 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1765 } else if (stat(consumerd32_bin, &st) == 0) {
1766 DBG3("Found location #3");
1767 consumer_to_use = consumerd32_bin;
1768 } else {
1769 DBG("Could not find any valid consumerd executable");
1770 break;
1771 }
1772 DBG("Using kernel consumer at: %s", consumer_to_use);
1773 execl(consumer_to_use,
1774 "lttng-consumerd", verbosity, "-k",
1775 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1776 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1777 NULL);
1778 break;
1779 case LTTNG_CONSUMER64_UST:
1780 {
1781 char *tmpnew = NULL;
1782
1783 if (consumerd64_libdir[0] != '\0') {
1784 char *tmp;
1785 size_t tmplen;
1786
1787 tmp = getenv("LD_LIBRARY_PATH");
1788 if (!tmp) {
1789 tmp = "";
1790 }
1791 tmplen = strlen("LD_LIBRARY_PATH=")
1792 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1793 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1794 if (!tmpnew) {
1795 ret = -ENOMEM;
1796 goto error;
1797 }
1798 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1799 strcat(tmpnew, consumerd64_libdir);
1800 if (tmp[0] != '\0') {
1801 strcat(tmpnew, ":");
1802 strcat(tmpnew, tmp);
1803 }
1804 ret = putenv(tmpnew);
1805 if (ret) {
1806 ret = -errno;
1807 goto error;
1808 }
1809 }
1810 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1811 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1812 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1813 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1814 NULL);
1815 if (consumerd64_libdir[0] != '\0') {
1816 free(tmpnew);
1817 }
1818 if (ret) {
1819 goto error;
1820 }
1821 break;
1822 }
1823 case LTTNG_CONSUMER32_UST:
1824 {
1825 char *tmpnew = NULL;
1826
1827 if (consumerd32_libdir[0] != '\0') {
1828 char *tmp;
1829 size_t tmplen;
1830
1831 tmp = getenv("LD_LIBRARY_PATH");
1832 if (!tmp) {
1833 tmp = "";
1834 }
1835 tmplen = strlen("LD_LIBRARY_PATH=")
1836 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1837 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1838 if (!tmpnew) {
1839 ret = -ENOMEM;
1840 goto error;
1841 }
1842 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1843 strcat(tmpnew, consumerd32_libdir);
1844 if (tmp[0] != '\0') {
1845 strcat(tmpnew, ":");
1846 strcat(tmpnew, tmp);
1847 }
1848 ret = putenv(tmpnew);
1849 if (ret) {
1850 ret = -errno;
1851 goto error;
1852 }
1853 }
1854 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1855 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1856 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1857 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1858 NULL);
1859 if (consumerd32_libdir[0] != '\0') {
1860 free(tmpnew);
1861 }
1862 if (ret) {
1863 goto error;
1864 }
1865 break;
1866 }
1867 default:
1868 PERROR("unknown consumer type");
1869 exit(EXIT_FAILURE);
1870 }
1871 if (errno != 0) {
1872 PERROR("kernel start consumer exec");
1873 }
1874 exit(EXIT_FAILURE);
1875 } else if (pid > 0) {
1876 ret = pid;
1877 } else {
1878 PERROR("start consumer fork");
1879 ret = -errno;
1880 }
1881 error:
1882 return ret;
1883 }
1884
1885 /*
1886 * Spawn the consumerd daemon and session daemon thread.
1887 */
1888 static int start_consumerd(struct consumer_data *consumer_data)
1889 {
1890 int ret;
1891
1892 /*
1893 * Set the listen() state on the socket since there is a possible race
1894 * between the exec() of the consumer daemon and this call if place in the
1895 * consumer thread. See bug #366 for more details.
1896 */
1897 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1898 if (ret < 0) {
1899 goto error;
1900 }
1901
1902 pthread_mutex_lock(&consumer_data->pid_mutex);
1903 if (consumer_data->pid != 0) {
1904 pthread_mutex_unlock(&consumer_data->pid_mutex);
1905 goto end;
1906 }
1907
1908 ret = spawn_consumerd(consumer_data);
1909 if (ret < 0) {
1910 ERR("Spawning consumerd failed");
1911 pthread_mutex_unlock(&consumer_data->pid_mutex);
1912 goto error;
1913 }
1914
1915 /* Setting up the consumer_data pid */
1916 consumer_data->pid = ret;
1917 DBG2("Consumer pid %d", consumer_data->pid);
1918 pthread_mutex_unlock(&consumer_data->pid_mutex);
1919
1920 DBG2("Spawning consumer control thread");
1921 ret = spawn_consumer_thread(consumer_data);
1922 if (ret < 0) {
1923 ERR("Fatal error spawning consumer control thread");
1924 goto error;
1925 }
1926
1927 end:
1928 return 0;
1929
1930 error:
1931 /* Cleanup already created socket on error. */
1932 if (consumer_data->err_sock >= 0) {
1933 int err;
1934
1935 err = close(consumer_data->err_sock);
1936 if (err < 0) {
1937 PERROR("close consumer data error socket");
1938 }
1939 }
1940 return ret;
1941 }
1942
1943 /*
1944 * Compute health status of each consumer. If one of them is zero (bad
1945 * state), we return 0.
1946 */
1947 static int check_consumer_health(void)
1948 {
1949 int ret;
1950
1951 ret = health_check_state(&kconsumer_data.health) &&
1952 health_check_state(&ustconsumer32_data.health) &&
1953 health_check_state(&ustconsumer64_data.health);
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, domain);
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_thread_cmd);
3072 break;
3073 case LTTNG_HEALTH_APP_MANAGE:
3074 reply.ret_code = health_check_state(&health_thread_app_manage);
3075 break;
3076 case LTTNG_HEALTH_APP_REG:
3077 reply.ret_code = health_check_state(&health_thread_app_reg);
3078 break;
3079 case LTTNG_HEALTH_KERNEL:
3080 reply.ret_code = health_check_state(&health_thread_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_thread_app_manage) &&
3088 health_check_state(&health_thread_app_reg) &&
3089 health_check_state(&health_thread_cmd) &&
3090 health_check_state(&health_thread_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 if (testpoint(thread_manage_clients)) {
3166 goto error_testpoint;
3167 }
3168
3169 health_code_update(&health_thread_cmd);
3170
3171 ret = lttcomm_listen_unix_sock(client_sock);
3172 if (ret < 0) {
3173 goto error_listen;
3174 }
3175
3176 /*
3177 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3178 * more will be added to this poll set.
3179 */
3180 ret = create_thread_poll_set(&events, 2);
3181 if (ret < 0) {
3182 goto error_create_poll;
3183 }
3184
3185 /* Add the application registration socket */
3186 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3187 if (ret < 0) {
3188 goto error;
3189 }
3190
3191 /*
3192 * Notify parent pid that we are ready to accept command for client side.
3193 */
3194 if (opt_sig_parent) {
3195 kill(ppid, SIGUSR1);
3196 }
3197
3198 if (testpoint(thread_manage_clients_before_loop)) {
3199 goto error;
3200 }
3201
3202 health_code_update(&health_thread_cmd);
3203
3204 while (1) {
3205 DBG("Accepting client command ...");
3206
3207 /* Inifinite blocking call, waiting for transmission */
3208 restart:
3209 health_poll_update(&health_thread_cmd);
3210 ret = lttng_poll_wait(&events, -1);
3211 health_poll_update(&health_thread_cmd);
3212 if (ret < 0) {
3213 /*
3214 * Restart interrupted system call.
3215 */
3216 if (errno == EINTR) {
3217 goto restart;
3218 }
3219 goto error;
3220 }
3221
3222 nb_fd = ret;
3223
3224 for (i = 0; i < nb_fd; i++) {
3225 /* Fetch once the poll data */
3226 revents = LTTNG_POLL_GETEV(&events, i);
3227 pollfd = LTTNG_POLL_GETFD(&events, i);
3228
3229 health_code_update(&health_thread_cmd);
3230
3231 /* Thread quit pipe has been closed. Killing thread. */
3232 ret = check_thread_quit_pipe(pollfd, revents);
3233 if (ret) {
3234 err = 0;
3235 goto exit;
3236 }
3237
3238 /* Event on the registration socket */
3239 if (pollfd == client_sock) {
3240 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3241 ERR("Client socket poll error");
3242 goto error;
3243 }
3244 }
3245 }
3246
3247 DBG("Wait for client response");
3248
3249 health_code_update(&health_thread_cmd);
3250
3251 sock = lttcomm_accept_unix_sock(client_sock);
3252 if (sock < 0) {
3253 goto error;
3254 }
3255
3256 /*
3257 * Set the CLOEXEC flag. Return code is useless because either way, the
3258 * show must go on.
3259 */
3260 (void) utils_set_fd_cloexec(sock);
3261
3262 /* Set socket option for credentials retrieval */
3263 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3264 if (ret < 0) {
3265 goto error;
3266 }
3267
3268 /* Allocate context command to process the client request */
3269 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3270 if (cmd_ctx == NULL) {
3271 PERROR("zmalloc cmd_ctx");
3272 goto error;
3273 }
3274
3275 /* Allocate data buffer for reception */
3276 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3277 if (cmd_ctx->lsm == NULL) {
3278 PERROR("zmalloc cmd_ctx->lsm");
3279 goto error;
3280 }
3281
3282 cmd_ctx->llm = NULL;
3283 cmd_ctx->session = NULL;
3284
3285 health_code_update(&health_thread_cmd);
3286
3287 /*
3288 * Data is received from the lttng client. The struct
3289 * lttcomm_session_msg (lsm) contains the command and data request of
3290 * the client.
3291 */
3292 DBG("Receiving data from client ...");
3293 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3294 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3295 if (ret <= 0) {
3296 DBG("Nothing recv() from client... continuing");
3297 ret = close(sock);
3298 if (ret) {
3299 PERROR("close");
3300 }
3301 sock = -1;
3302 clean_command_ctx(&cmd_ctx);
3303 continue;
3304 }
3305
3306 health_code_update(&health_thread_cmd);
3307
3308 // TODO: Validate cmd_ctx including sanity check for
3309 // security purpose.
3310
3311 rcu_thread_online();
3312 /*
3313 * This function dispatch the work to the kernel or userspace tracer
3314 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3315 * informations for the client. The command context struct contains
3316 * everything this function may needs.
3317 */
3318 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3319 rcu_thread_offline();
3320 if (ret < 0) {
3321 if (sock_error) {
3322 ret = close(sock);
3323 if (ret) {
3324 PERROR("close");
3325 }
3326 sock = -1;
3327 }
3328 /*
3329 * TODO: Inform client somehow of the fatal error. At
3330 * this point, ret < 0 means that a zmalloc failed
3331 * (ENOMEM). Error detected but still accept
3332 * command, unless a socket error has been
3333 * detected.
3334 */
3335 clean_command_ctx(&cmd_ctx);
3336 continue;
3337 }
3338
3339 health_code_update(&health_thread_cmd);
3340
3341 DBG("Sending response (size: %d, retcode: %s)",
3342 cmd_ctx->lttng_msg_size,
3343 lttng_strerror(-cmd_ctx->llm->ret_code));
3344 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3345 if (ret < 0) {
3346 ERR("Failed to send data back to client");
3347 }
3348
3349 /* End of transmission */
3350 ret = close(sock);
3351 if (ret) {
3352 PERROR("close");
3353 }
3354 sock = -1;
3355
3356 clean_command_ctx(&cmd_ctx);
3357
3358 health_code_update(&health_thread_cmd);
3359 }
3360
3361 exit:
3362 error:
3363 if (sock >= 0) {
3364 ret = close(sock);
3365 if (ret) {
3366 PERROR("close");
3367 }
3368 }
3369
3370 lttng_poll_clean(&events);
3371 clean_command_ctx(&cmd_ctx);
3372
3373 error_listen:
3374 error_create_poll:
3375 error_testpoint:
3376 unlink(client_unix_sock_path);
3377 if (client_sock >= 0) {
3378 ret = close(client_sock);
3379 if (ret) {
3380 PERROR("close");
3381 }
3382 }
3383
3384 if (err) {
3385 health_error(&health_thread_cmd);
3386 ERR("Health error occurred in %s", __func__);
3387 }
3388
3389 health_exit(&health_thread_cmd);
3390
3391 DBG("Client thread dying");
3392
3393 rcu_unregister_thread();
3394 return NULL;
3395 }
3396
3397
3398 /*
3399 * usage function on stderr
3400 */
3401 static void usage(void)
3402 {
3403 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3404 fprintf(stderr, " -h, --help Display this usage.\n");
3405 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3406 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3407 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3408 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3409 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3410 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3411 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3412 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3413 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3414 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3415 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3416 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3417 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3418 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3419 fprintf(stderr, " -V, --version Show version number.\n");
3420 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3421 fprintf(stderr, " -q, --quiet No output at all.\n");
3422 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3423 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3424 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3425 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3426 }
3427
3428 /*
3429 * daemon argument parsing
3430 */
3431 static int parse_args(int argc, char **argv)
3432 {
3433 int c;
3434
3435 static struct option long_options[] = {
3436 { "client-sock", 1, 0, 'c' },
3437 { "apps-sock", 1, 0, 'a' },
3438 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3439 { "kconsumerd-err-sock", 1, 0, 'E' },
3440 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3441 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3442 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3443 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3444 { "consumerd32-path", 1, 0, 'u' },
3445 { "consumerd32-libdir", 1, 0, 'U' },
3446 { "consumerd64-path", 1, 0, 't' },
3447 { "consumerd64-libdir", 1, 0, 'T' },
3448 { "daemonize", 0, 0, 'd' },
3449 { "sig-parent", 0, 0, 'S' },
3450 { "help", 0, 0, 'h' },
3451 { "group", 1, 0, 'g' },
3452 { "version", 0, 0, 'V' },
3453 { "quiet", 0, 0, 'q' },
3454 { "verbose", 0, 0, 'v' },
3455 { "verbose-consumer", 0, 0, 'Z' },
3456 { "no-kernel", 0, 0, 'N' },
3457 { "pidfile", 1, 0, 'p' },
3458 { NULL, 0, 0, 0 }
3459 };
3460
3461 while (1) {
3462 int option_index = 0;
3463 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3464 long_options, &option_index);
3465 if (c == -1) {
3466 break;
3467 }
3468
3469 switch (c) {
3470 case 0:
3471 fprintf(stderr, "option %s", long_options[option_index].name);
3472 if (optarg) {
3473 fprintf(stderr, " with arg %s\n", optarg);
3474 }
3475 break;
3476 case 'c':
3477 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3478 break;
3479 case 'a':
3480 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3481 break;
3482 case 'd':
3483 opt_daemon = 1;
3484 break;
3485 case 'g':
3486 opt_tracing_group = optarg;
3487 break;
3488 case 'h':
3489 usage();
3490 exit(EXIT_FAILURE);
3491 case 'V':
3492 fprintf(stdout, "%s\n", VERSION);
3493 exit(EXIT_SUCCESS);
3494 case 'S':
3495 opt_sig_parent = 1;
3496 break;
3497 case 'E':
3498 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3499 break;
3500 case 'C':
3501 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3502 break;
3503 case 'F':
3504 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3505 break;
3506 case 'D':
3507 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3508 break;
3509 case 'H':
3510 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3511 break;
3512 case 'G':
3513 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3514 break;
3515 case 'N':
3516 opt_no_kernel = 1;
3517 break;
3518 case 'q':
3519 lttng_opt_quiet = 1;
3520 break;
3521 case 'v':
3522 /* Verbose level can increase using multiple -v */
3523 lttng_opt_verbose += 1;
3524 break;
3525 case 'Z':
3526 opt_verbose_consumer += 1;
3527 break;
3528 case 'u':
3529 consumerd32_bin= optarg;
3530 break;
3531 case 'U':
3532 consumerd32_libdir = optarg;
3533 break;
3534 case 't':
3535 consumerd64_bin = optarg;
3536 break;
3537 case 'T':
3538 consumerd64_libdir = optarg;
3539 break;
3540 case 'p':
3541 opt_pidfile = optarg;
3542 break;
3543 default:
3544 /* Unknown option or other error.
3545 * Error is printed by getopt, just return */
3546 return -1;
3547 }
3548 }
3549
3550 return 0;
3551 }
3552
3553 /*
3554 * Creates the two needed socket by the daemon.
3555 * apps_sock - The communication socket for all UST apps.
3556 * client_sock - The communication of the cli tool (lttng).
3557 */
3558 static int init_daemon_socket(void)
3559 {
3560 int ret = 0;
3561 mode_t old_umask;
3562
3563 old_umask = umask(0);
3564
3565 /* Create client tool unix socket */
3566 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3567 if (client_sock < 0) {
3568 ERR("Create unix sock failed: %s", client_unix_sock_path);
3569 ret = -1;
3570 goto end;
3571 }
3572
3573 /* Set the cloexec flag */
3574 ret = utils_set_fd_cloexec(client_sock);
3575 if (ret < 0) {
3576 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3577 "Continuing but note that the consumer daemon will have a "
3578 "reference to this socket on exec()", client_sock);
3579 }
3580
3581 /* File permission MUST be 660 */
3582 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3583 if (ret < 0) {
3584 ERR("Set file permissions failed: %s", client_unix_sock_path);
3585 PERROR("chmod");
3586 goto end;
3587 }
3588
3589 /* Create the application unix socket */
3590 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3591 if (apps_sock < 0) {
3592 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3593 ret = -1;
3594 goto end;
3595 }
3596
3597 /* Set the cloexec flag */
3598 ret = utils_set_fd_cloexec(apps_sock);
3599 if (ret < 0) {
3600 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3601 "Continuing but note that the consumer daemon will have a "
3602 "reference to this socket on exec()", apps_sock);
3603 }
3604
3605 /* File permission MUST be 666 */
3606 ret = chmod(apps_unix_sock_path,
3607 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3608 if (ret < 0) {
3609 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3610 PERROR("chmod");
3611 goto end;
3612 }
3613
3614 DBG3("Session daemon client socket %d and application socket %d created",
3615 client_sock, apps_sock);
3616
3617 end:
3618 umask(old_umask);
3619 return ret;
3620 }
3621
3622 /*
3623 * Check if the global socket is available, and if a daemon is answering at the
3624 * other side. If yes, error is returned.
3625 */
3626 static int check_existing_daemon(void)
3627 {
3628 /* Is there anybody out there ? */
3629 if (lttng_session_daemon_alive()) {
3630 return -EEXIST;
3631 }
3632
3633 return 0;
3634 }
3635
3636 /*
3637 * Set the tracing group gid onto the client socket.
3638 *
3639 * Race window between mkdir and chown is OK because we are going from more
3640 * permissive (root.root) to less permissive (root.tracing).
3641 */
3642 static int set_permissions(char *rundir)
3643 {
3644 int ret;
3645 gid_t gid;
3646
3647 ret = allowed_group();
3648 if (ret < 0) {
3649 WARN("No tracing group detected");
3650 ret = 0;
3651 goto end;
3652 }
3653
3654 gid = ret;
3655
3656 /* Set lttng run dir */
3657 ret = chown(rundir, 0, gid);
3658 if (ret < 0) {
3659 ERR("Unable to set group on %s", rundir);
3660 PERROR("chown");
3661 }
3662
3663 /* Ensure tracing group can search the run dir */
3664 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3665 if (ret < 0) {
3666 ERR("Unable to set permissions on %s", rundir);
3667 PERROR("chmod");
3668 }
3669
3670 /* lttng client socket path */
3671 ret = chown(client_unix_sock_path, 0, gid);
3672 if (ret < 0) {
3673 ERR("Unable to set group on %s", client_unix_sock_path);
3674 PERROR("chown");
3675 }
3676
3677 /* kconsumer error socket path */
3678 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3679 if (ret < 0) {
3680 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3681 PERROR("chown");
3682 }
3683
3684 /* 64-bit ustconsumer error socket path */
3685 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3686 if (ret < 0) {
3687 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3688 PERROR("chown");
3689 }
3690
3691 /* 32-bit ustconsumer compat32 error socket path */
3692 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3693 if (ret < 0) {
3694 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3695 PERROR("chown");
3696 }
3697
3698 DBG("All permissions are set");
3699
3700 end:
3701 return ret;
3702 }
3703
3704 /*
3705 * Create the lttng run directory needed for all global sockets and pipe.
3706 */
3707 static int create_lttng_rundir(const char *rundir)
3708 {
3709 int ret;
3710
3711 DBG3("Creating LTTng run directory: %s", rundir);
3712
3713 ret = mkdir(rundir, S_IRWXU);
3714 if (ret < 0) {
3715 if (errno != EEXIST) {
3716 ERR("Unable to create %s", rundir);
3717 goto error;
3718 } else {
3719 ret = 0;
3720 }
3721 }
3722
3723 error:
3724 return ret;
3725 }
3726
3727 /*
3728 * Setup sockets and directory needed by the kconsumerd communication with the
3729 * session daemon.
3730 */
3731 static int set_consumer_sockets(struct consumer_data *consumer_data,
3732 const char *rundir)
3733 {
3734 int ret;
3735 char path[PATH_MAX];
3736
3737 switch (consumer_data->type) {
3738 case LTTNG_CONSUMER_KERNEL:
3739 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3740 break;
3741 case LTTNG_CONSUMER64_UST:
3742 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3743 break;
3744 case LTTNG_CONSUMER32_UST:
3745 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3746 break;
3747 default:
3748 ERR("Consumer type unknown");
3749 ret = -EINVAL;
3750 goto error;
3751 }
3752
3753 DBG2("Creating consumer directory: %s", path);
3754
3755 ret = mkdir(path, S_IRWXU);
3756 if (ret < 0) {
3757 if (errno != EEXIST) {
3758 PERROR("mkdir");
3759 ERR("Failed to create %s", path);
3760 goto error;
3761 }
3762 ret = -1;
3763 }
3764
3765 /* Create the kconsumerd error unix socket */
3766 consumer_data->err_sock =
3767 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3768 if (consumer_data->err_sock < 0) {
3769 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3770 ret = -1;
3771 goto error;
3772 }
3773
3774 /* File permission MUST be 660 */
3775 ret = chmod(consumer_data->err_unix_sock_path,
3776 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3777 if (ret < 0) {
3778 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3779 PERROR("chmod");
3780 goto error;
3781 }
3782
3783 error:
3784 return ret;
3785 }
3786
3787 /*
3788 * Signal handler for the daemon
3789 *
3790 * Simply stop all worker threads, leaving main() return gracefully after
3791 * joining all threads and calling cleanup().
3792 */
3793 static void sighandler(int sig)
3794 {
3795 switch (sig) {
3796 case SIGPIPE:
3797 DBG("SIGPIPE caught");
3798 return;
3799 case SIGINT:
3800 DBG("SIGINT caught");
3801 stop_threads();
3802 break;
3803 case SIGTERM:
3804 DBG("SIGTERM caught");
3805 stop_threads();
3806 break;
3807 default:
3808 break;
3809 }
3810 }
3811
3812 /*
3813 * Setup signal handler for :
3814 * SIGINT, SIGTERM, SIGPIPE
3815 */
3816 static int set_signal_handler(void)
3817 {
3818 int ret = 0;
3819 struct sigaction sa;
3820 sigset_t sigset;
3821
3822 if ((ret = sigemptyset(&sigset)) < 0) {
3823 PERROR("sigemptyset");
3824 return ret;
3825 }
3826
3827 sa.sa_handler = sighandler;
3828 sa.sa_mask = sigset;
3829 sa.sa_flags = 0;
3830 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3831 PERROR("sigaction");
3832 return ret;
3833 }
3834
3835 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3836 PERROR("sigaction");
3837 return ret;
3838 }
3839
3840 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3841 PERROR("sigaction");
3842 return ret;
3843 }
3844
3845 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3846
3847 return ret;
3848 }
3849
3850 /*
3851 * Set open files limit to unlimited. This daemon can open a large number of
3852 * file descriptors in order to consumer multiple kernel traces.
3853 */
3854 static void set_ulimit(void)
3855 {
3856 int ret;
3857 struct rlimit lim;
3858
3859 /* The kernel does not allowed an infinite limit for open files */
3860 lim.rlim_cur = 65535;
3861 lim.rlim_max = 65535;
3862
3863 ret = setrlimit(RLIMIT_NOFILE, &lim);
3864 if (ret < 0) {
3865 PERROR("failed to set open files limit");
3866 }
3867 }
3868
3869 /*
3870 * Write pidfile using the rundir and opt_pidfile.
3871 */
3872 static void write_pidfile(void)
3873 {
3874 int ret;
3875 char pidfile_path[PATH_MAX];
3876
3877 assert(rundir);
3878
3879 if (opt_pidfile) {
3880 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
3881 } else {
3882 /* Build pidfile path from rundir and opt_pidfile. */
3883 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
3884 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
3885 if (ret < 0) {
3886 PERROR("snprintf pidfile path");
3887 goto error;
3888 }
3889 }
3890
3891 /*
3892 * Create pid file in rundir. Return value is of no importance. The
3893 * execution will continue even though we are not able to write the file.
3894 */
3895 (void) utils_create_pid_file(getpid(), pidfile_path);
3896
3897 error:
3898 return;
3899 }
3900
3901 /*
3902 * main
3903 */
3904 int main(int argc, char **argv)
3905 {
3906 int ret = 0;
3907 void *status;
3908 const char *home_path, *env_app_timeout;
3909
3910 init_kernel_workarounds();
3911
3912 rcu_register_thread();
3913
3914 setup_consumerd_path();
3915
3916 /* Parse arguments */
3917 progname = argv[0];
3918 if ((ret = parse_args(argc, argv)) < 0) {
3919 goto error;
3920 }
3921
3922 /* Daemonize */
3923 if (opt_daemon) {
3924 int i;
3925
3926 /*
3927 * fork
3928 * child: setsid, close FD 0, 1, 2, chdir /
3929 * parent: exit (if fork is successful)
3930 */
3931 ret = daemon(0, 0);
3932 if (ret < 0) {
3933 PERROR("daemon");
3934 goto error;
3935 }
3936 /*
3937 * We are in the child. Make sure all other file
3938 * descriptors are closed, in case we are called with
3939 * more opened file descriptors than the standard ones.
3940 */
3941 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3942 (void) close(i);
3943 }
3944 }
3945
3946 /* Create thread quit pipe */
3947 if ((ret = init_thread_quit_pipe()) < 0) {
3948 goto error;
3949 }
3950
3951 /* Check if daemon is UID = 0 */
3952 is_root = !getuid();
3953
3954 if (is_root) {
3955 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3956
3957 /* Create global run dir with root access */
3958 ret = create_lttng_rundir(rundir);
3959 if (ret < 0) {
3960 goto error;
3961 }
3962
3963 if (strlen(apps_unix_sock_path) == 0) {
3964 snprintf(apps_unix_sock_path, PATH_MAX,
3965 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3966 }
3967
3968 if (strlen(client_unix_sock_path) == 0) {
3969 snprintf(client_unix_sock_path, PATH_MAX,
3970 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3971 }
3972
3973 /* Set global SHM for ust */
3974 if (strlen(wait_shm_path) == 0) {
3975 snprintf(wait_shm_path, PATH_MAX,
3976 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3977 }
3978
3979 if (strlen(health_unix_sock_path) == 0) {
3980 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3981 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3982 }
3983
3984 /* Setup kernel consumerd path */
3985 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3986 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3987 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3988 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3989
3990 DBG2("Kernel consumer err path: %s",
3991 kconsumer_data.err_unix_sock_path);
3992 DBG2("Kernel consumer cmd path: %s",
3993 kconsumer_data.cmd_unix_sock_path);
3994 } else {
3995 home_path = get_home_dir();
3996 if (home_path == NULL) {
3997 /* TODO: Add --socket PATH option */
3998 ERR("Can't get HOME directory for sockets creation.");
3999 ret = -EPERM;
4000 goto error;
4001 }
4002
4003 /*
4004 * Create rundir from home path. This will create something like
4005 * $HOME/.lttng
4006 */
4007 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4008 if (ret < 0) {
4009 ret = -ENOMEM;
4010 goto error;
4011 }
4012
4013 ret = create_lttng_rundir(rundir);
4014 if (ret < 0) {
4015 goto error;
4016 }
4017
4018 if (strlen(apps_unix_sock_path) == 0) {
4019 snprintf(apps_unix_sock_path, PATH_MAX,
4020 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4021 }
4022
4023 /* Set the cli tool unix socket path */
4024 if (strlen(client_unix_sock_path) == 0) {
4025 snprintf(client_unix_sock_path, PATH_MAX,
4026 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4027 }
4028
4029 /* Set global SHM for ust */
4030 if (strlen(wait_shm_path) == 0) {
4031 snprintf(wait_shm_path, PATH_MAX,
4032 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
4033 }
4034
4035 /* Set health check Unix path */
4036 if (strlen(health_unix_sock_path) == 0) {
4037 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4038 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4039 }
4040 }
4041
4042 /* Set consumer initial state */
4043 kernel_consumerd_state = CONSUMER_STOPPED;
4044 ust_consumerd_state = CONSUMER_STOPPED;
4045
4046 DBG("Client socket path %s", client_unix_sock_path);
4047 DBG("Application socket path %s", apps_unix_sock_path);
4048 DBG("LTTng run directory path: %s", rundir);
4049
4050 /* 32 bits consumerd path setup */
4051 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4052 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4053 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4054 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4055
4056 DBG2("UST consumer 32 bits err path: %s",
4057 ustconsumer32_data.err_unix_sock_path);
4058 DBG2("UST consumer 32 bits cmd path: %s",
4059 ustconsumer32_data.cmd_unix_sock_path);
4060
4061 /* 64 bits consumerd path setup */
4062 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4063 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4064 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4065 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4066
4067 DBG2("UST consumer 64 bits err path: %s",
4068 ustconsumer64_data.err_unix_sock_path);
4069 DBG2("UST consumer 64 bits cmd path: %s",
4070 ustconsumer64_data.cmd_unix_sock_path);
4071
4072 /*
4073 * See if daemon already exist.
4074 */
4075 if ((ret = check_existing_daemon()) < 0) {
4076 ERR("Already running daemon.\n");
4077 /*
4078 * We do not goto exit because we must not cleanup()
4079 * because a daemon is already running.
4080 */
4081 goto error;
4082 }
4083
4084 /*
4085 * Init UST app hash table. Alloc hash table before this point since
4086 * cleanup() can get called after that point.
4087 */
4088 ust_app_ht_alloc();
4089
4090 /* After this point, we can safely call cleanup() with "goto exit" */
4091
4092 /*
4093 * These actions must be executed as root. We do that *after* setting up
4094 * the sockets path because we MUST make the check for another daemon using
4095 * those paths *before* trying to set the kernel consumer sockets and init
4096 * kernel tracer.
4097 */
4098 if (is_root) {
4099 ret = set_consumer_sockets(&kconsumer_data, rundir);
4100 if (ret < 0) {
4101 goto exit;
4102 }
4103
4104 /* Setup kernel tracer */
4105 if (!opt_no_kernel) {
4106 init_kernel_tracer();
4107 }
4108
4109 /* Set ulimit for open files */
4110 set_ulimit();
4111 }
4112 /* init lttng_fd tracking must be done after set_ulimit. */
4113 lttng_fd_init();
4114
4115 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4116 if (ret < 0) {
4117 goto exit;
4118 }
4119
4120 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4121 if (ret < 0) {
4122 goto exit;
4123 }
4124
4125 if ((ret = set_signal_handler()) < 0) {
4126 goto exit;
4127 }
4128
4129 /* Setup the needed unix socket */
4130 if ((ret = init_daemon_socket()) < 0) {
4131 goto exit;
4132 }
4133
4134 /* Set credentials to socket */
4135 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4136 goto exit;
4137 }
4138
4139 /* Get parent pid if -S, --sig-parent is specified. */
4140 if (opt_sig_parent) {
4141 ppid = getppid();
4142 }
4143
4144 /* Setup the kernel pipe for waking up the kernel thread */
4145 if (is_root && !opt_no_kernel) {
4146 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4147 goto exit;
4148 }
4149 }
4150
4151 /* Setup the thread apps communication pipe. */
4152 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4153 goto exit;
4154 }
4155
4156 /* Init UST command queue. */
4157 cds_wfq_init(&ust_cmd_queue.queue);
4158
4159 /*
4160 * Get session list pointer. This pointer MUST NOT be free(). This list is
4161 * statically declared in session.c
4162 */
4163 session_list_ptr = session_get_list();
4164
4165 /* Set up max poll set size */
4166 lttng_poll_set_max_size();
4167
4168 cmd_init();
4169
4170 /* Init all health thread counters. */
4171 health_init(&health_thread_cmd);
4172 health_init(&health_thread_kernel);
4173 health_init(&health_thread_app_manage);
4174 health_init(&health_thread_app_reg);
4175
4176 /*
4177 * Init health counters of the consumer thread. We do a quick hack here to
4178 * the state of the consumer health is fine even if the thread is not
4179 * started. Once the thread starts, the health state is updated with a poll
4180 * value to set a health code path. This is simply to ease our life and has
4181 * no cost what so ever.
4182 */
4183 health_init(&kconsumer_data.health);
4184 health_poll_update(&kconsumer_data.health);
4185 health_init(&ustconsumer32_data.health);
4186 health_poll_update(&ustconsumer32_data.health);
4187 health_init(&ustconsumer64_data.health);
4188 health_poll_update(&ustconsumer64_data.health);
4189
4190 /* Check for the application socket timeout env variable. */
4191 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4192 if (env_app_timeout) {
4193 app_socket_timeout = atoi(env_app_timeout);
4194 } else {
4195 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4196 }
4197
4198 write_pidfile();
4199
4200 /* Create thread to manage the client socket */
4201 ret = pthread_create(&health_thread, NULL,
4202 thread_manage_health, (void *) NULL);
4203 if (ret != 0) {
4204 PERROR("pthread_create health");
4205 goto exit_health;
4206 }
4207
4208 /* Create thread to manage the client socket */
4209 ret = pthread_create(&client_thread, NULL,
4210 thread_manage_clients, (void *) NULL);
4211 if (ret != 0) {
4212 PERROR("pthread_create clients");
4213 goto exit_client;
4214 }
4215
4216 /* Create thread to dispatch registration */
4217 ret = pthread_create(&dispatch_thread, NULL,
4218 thread_dispatch_ust_registration, (void *) NULL);
4219 if (ret != 0) {
4220 PERROR("pthread_create dispatch");
4221 goto exit_dispatch;
4222 }
4223
4224 /* Create thread to manage application registration. */
4225 ret = pthread_create(&reg_apps_thread, NULL,
4226 thread_registration_apps, (void *) NULL);
4227 if (ret != 0) {
4228 PERROR("pthread_create registration");
4229 goto exit_reg_apps;
4230 }
4231
4232 /* Create thread to manage application socket */
4233 ret = pthread_create(&apps_thread, NULL,
4234 thread_manage_apps, (void *) NULL);
4235 if (ret != 0) {
4236 PERROR("pthread_create apps");
4237 goto exit_apps;
4238 }
4239
4240 /* Don't start this thread if kernel tracing is not requested nor root */
4241 if (is_root && !opt_no_kernel) {
4242 /* Create kernel thread to manage kernel event */
4243 ret = pthread_create(&kernel_thread, NULL,
4244 thread_manage_kernel, (void *) NULL);
4245 if (ret != 0) {
4246 PERROR("pthread_create kernel");
4247 goto exit_kernel;
4248 }
4249
4250 ret = pthread_join(kernel_thread, &status);
4251 if (ret != 0) {
4252 PERROR("pthread_join");
4253 goto error; /* join error, exit without cleanup */
4254 }
4255 }
4256
4257 exit_kernel:
4258 ret = pthread_join(apps_thread, &status);
4259 if (ret != 0) {
4260 PERROR("pthread_join");
4261 goto error; /* join error, exit without cleanup */
4262 }
4263
4264 exit_apps:
4265 ret = pthread_join(reg_apps_thread, &status);
4266 if (ret != 0) {
4267 PERROR("pthread_join");
4268 goto error; /* join error, exit without cleanup */
4269 }
4270
4271 exit_reg_apps:
4272 ret = pthread_join(dispatch_thread, &status);
4273 if (ret != 0) {
4274 PERROR("pthread_join");
4275 goto error; /* join error, exit without cleanup */
4276 }
4277
4278 exit_dispatch:
4279 ret = pthread_join(client_thread, &status);
4280 if (ret != 0) {
4281 PERROR("pthread_join");
4282 goto error; /* join error, exit without cleanup */
4283 }
4284
4285 ret = join_consumer_thread(&kconsumer_data);
4286 if (ret != 0) {
4287 PERROR("join_consumer");
4288 goto error; /* join error, exit without cleanup */
4289 }
4290
4291 ret = join_consumer_thread(&ustconsumer32_data);
4292 if (ret != 0) {
4293 PERROR("join_consumer ust32");
4294 goto error; /* join error, exit without cleanup */
4295 }
4296
4297 ret = join_consumer_thread(&ustconsumer64_data);
4298 if (ret != 0) {
4299 PERROR("join_consumer ust64");
4300 goto error; /* join error, exit without cleanup */
4301 }
4302
4303 exit_client:
4304 ret = pthread_join(health_thread, &status);
4305 if (ret != 0) {
4306 PERROR("pthread_join health thread");
4307 goto error; /* join error, exit without cleanup */
4308 }
4309
4310 exit_health:
4311 exit:
4312 /*
4313 * cleanup() is called when no other thread is running.
4314 */
4315 rcu_thread_online();
4316 cleanup();
4317 rcu_thread_offline();
4318 rcu_unregister_thread();
4319 if (!ret) {
4320 exit(EXIT_SUCCESS);
4321 }
4322 error:
4323 exit(EXIT_FAILURE);
4324 }
This page took 0.175158 seconds and 3 git commands to generate.