Cleanup: add missing dot
[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 <inttypes.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <config.h>
39
40 #include <common/common.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 "buffer-registry.h"
50 #include "channel.h"
51 #include "cmd.h"
52 #include "consumer.h"
53 #include "context.h"
54 #include "event.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "modprobe.h"
58 #include "shm.h"
59 #include "ust-ctl.h"
60 #include "ust-consumer.h"
61 #include "utils.h"
62 #include "fd-limit.h"
63 #include "health.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66
67 #define CONSUMERD_FILE "lttng-consumerd"
68
69 /* Const values */
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71
72 const char *progname;
73 const char *opt_tracing_group;
74 static const char *opt_pidfile;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .metadata_sock.fd = -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 .metadata_sock.fd = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer32_data = {
112 .type = LTTNG_CONSUMER32_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .metadata_sock.fd = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123
124 /* Shared between threads */
125 static int dispatch_thread_exit;
126
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path[PATH_MAX];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path[PATH_MAX];
131 /* global wait shm path for UST */
132 static char wait_shm_path[PATH_MAX];
133 /* Global health check unix path */
134 static char health_unix_sock_path[PATH_MAX];
135
136 /* Sockets and FDs */
137 static int client_sock = -1;
138 static int apps_sock = -1;
139 int kernel_tracer_fd = -1;
140 static int kernel_poll_pipe[2] = { -1, -1 };
141
142 /*
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
145 */
146 static int thread_quit_pipe[2] = { -1, -1 };
147
148 /*
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
151 */
152 static int apps_cmd_pipe[2] = { -1, -1 };
153
154 int apps_cmd_notify_pipe[2] = { -1, -1 };
155
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread;
158 static pthread_t apps_notify_thread;
159 static pthread_t reg_apps_thread;
160 static pthread_t client_thread;
161 static pthread_t kernel_thread;
162 static pthread_t dispatch_thread;
163 static pthread_t health_thread;
164 static pthread_t ht_cleanup_thread;
165
166 /*
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
169 *
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
172 */
173 static struct ust_cmd_queue ust_cmd_queue;
174
175 /*
176 * Pointer initialized before thread creation.
177 *
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
181 *
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
184 */
185 static struct ltt_session_list *session_list_ptr;
186
187 int ust_consumerd64_fd = -1;
188 int ust_consumerd32_fd = -1;
189
190 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
191 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
192 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
193 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
194
195 static const char *module_proc_lttng = "/proc/lttng";
196
197 /*
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
200 */
201 enum consumerd_state {
202 CONSUMER_STARTED = 1,
203 CONSUMER_STOPPED = 2,
204 CONSUMER_ERROR = 3,
205 };
206
207 /*
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
212 *
213 * The following example shows a possible race condition of this scheme:
214 *
215 * consumer thread error happens
216 * client cmd arrives
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
220 * ...
221 *
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
227 */
228 static enum consumerd_state ust_consumerd_state;
229 static enum consumerd_state kernel_consumerd_state;
230
231 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 /* Set in main() with the current page size. */
237 long page_size;
238
239 static
240 void setup_consumerd_path(void)
241 {
242 const char *bin, *libdir;
243
244 /*
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
248 */
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin[0]) {
251 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
252 }
253 if (!consumerd32_libdir[0]) {
254 consumerd32_libdir = INSTALL_LIB_PATH;
255 }
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin[0]) {
258 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
259 }
260 if (!consumerd64_libdir[0]) {
261 consumerd64_libdir = INSTALL_LIB_PATH;
262 }
263 #else
264 #error "Unknown bitness"
265 #endif
266
267 /*
268 * runtime env. var. overrides the build default.
269 */
270 bin = getenv("LTTNG_CONSUMERD32_BIN");
271 if (bin) {
272 consumerd32_bin = bin;
273 }
274 bin = getenv("LTTNG_CONSUMERD64_BIN");
275 if (bin) {
276 consumerd64_bin = bin;
277 }
278 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
279 if (libdir) {
280 consumerd32_libdir = libdir;
281 }
282 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
283 if (libdir) {
284 consumerd64_libdir = libdir;
285 }
286 }
287
288 /*
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
290 */
291 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
292 {
293 int ret;
294
295 assert(events);
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310 error:
311 return ret;
312 }
313
314 /*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
319 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326 }
327
328 /*
329 * Return group ID of the tracing group or -1 if not found.
330 */
331 static gid_t allowed_group(void)
332 {
333 struct group *grp;
334
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345 }
346
347 /*
348 * Init thread quit pipe.
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352 static int init_thread_quit_pipe(void)
353 {
354 int ret, i;
355
356 ret = pipe(thread_quit_pipe);
357 if (ret < 0) {
358 PERROR("thread quit pipe");
359 goto error;
360 }
361
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
370 error:
371 return ret;
372 }
373
374 /*
375 * Stop all threads by closing the thread quit pipe.
376 */
377 static void stop_threads(void)
378 {
379 int ret;
380
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret = notify_thread_pipe(thread_quit_pipe[1]);
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
390 futex_nto1_wake(&ust_cmd_queue.futex);
391 }
392
393 /*
394 * Close every consumer sockets.
395 */
396 static void close_consumer_sockets(void)
397 {
398 int ret;
399
400 if (kconsumer_data.err_sock >= 0) {
401 ret = close(kconsumer_data.err_sock);
402 if (ret < 0) {
403 PERROR("kernel consumer err_sock close");
404 }
405 }
406 if (ustconsumer32_data.err_sock >= 0) {
407 ret = close(ustconsumer32_data.err_sock);
408 if (ret < 0) {
409 PERROR("UST consumerd32 err_sock close");
410 }
411 }
412 if (ustconsumer64_data.err_sock >= 0) {
413 ret = close(ustconsumer64_data.err_sock);
414 if (ret < 0) {
415 PERROR("UST consumerd64 err_sock close");
416 }
417 }
418 if (kconsumer_data.cmd_sock >= 0) {
419 ret = close(kconsumer_data.cmd_sock);
420 if (ret < 0) {
421 PERROR("kernel consumer cmd_sock close");
422 }
423 }
424 if (ustconsumer32_data.cmd_sock >= 0) {
425 ret = close(ustconsumer32_data.cmd_sock);
426 if (ret < 0) {
427 PERROR("UST consumerd32 cmd_sock close");
428 }
429 }
430 if (ustconsumer64_data.cmd_sock >= 0) {
431 ret = close(ustconsumer64_data.cmd_sock);
432 if (ret < 0) {
433 PERROR("UST consumerd64 cmd_sock close");
434 }
435 }
436 }
437
438 /*
439 * Cleanup the daemon
440 */
441 static void cleanup(void)
442 {
443 int ret;
444 char *cmd = NULL;
445 struct ltt_session *sess, *stmp;
446
447 DBG("Cleaning up");
448
449 /*
450 * Close the thread quit pipe. It has already done its job,
451 * since we are now called.
452 */
453 utils_close_pipe(thread_quit_pipe);
454
455 /*
456 * If opt_pidfile is undefined, the default file will be wiped when
457 * removing the rundir.
458 */
459 if (opt_pidfile) {
460 ret = remove(opt_pidfile);
461 if (ret < 0) {
462 PERROR("remove pidfile %s", opt_pidfile);
463 }
464 }
465
466 DBG("Removing %s directory", rundir);
467 ret = asprintf(&cmd, "rm -rf %s", rundir);
468 if (ret < 0) {
469 ERR("asprintf failed. Something is really wrong!");
470 }
471
472 /* Remove lttng run directory */
473 ret = system(cmd);
474 if (ret < 0) {
475 ERR("Unable to clean %s", rundir);
476 }
477 free(cmd);
478 free(rundir);
479
480 DBG("Cleaning up all sessions");
481
482 /* Destroy session list mutex */
483 if (session_list_ptr != NULL) {
484 pthread_mutex_destroy(&session_list_ptr->lock);
485
486 /* Cleanup ALL session */
487 cds_list_for_each_entry_safe(sess, stmp,
488 &session_list_ptr->head, list) {
489 cmd_destroy_session(sess, kernel_poll_pipe[1]);
490 }
491 }
492
493 DBG("Closing all UST sockets");
494 ust_app_clean_list();
495 buffer_reg_destroy_registries();
496
497 if (is_root && !opt_no_kernel) {
498 DBG2("Closing kernel fd");
499 if (kernel_tracer_fd >= 0) {
500 ret = close(kernel_tracer_fd);
501 if (ret) {
502 PERROR("close");
503 }
504 }
505 DBG("Unloading kernel modules");
506 modprobe_remove_lttng_all();
507 }
508
509 close_consumer_sockets();
510
511 /* <fun> */
512 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
513 "Matthew, BEET driven development works!%c[%dm",
514 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
515 /* </fun> */
516 }
517
518 /*
519 * Send data on a unix socket using the liblttsessiondcomm API.
520 *
521 * Return lttcomm error code.
522 */
523 static int send_unix_sock(int sock, void *buf, size_t len)
524 {
525 /* Check valid length */
526 if (len == 0) {
527 return -1;
528 }
529
530 return lttcomm_send_unix_sock(sock, buf, len);
531 }
532
533 /*
534 * Free memory of a command context structure.
535 */
536 static void clean_command_ctx(struct command_ctx **cmd_ctx)
537 {
538 DBG("Clean command context structure");
539 if (*cmd_ctx) {
540 if ((*cmd_ctx)->llm) {
541 free((*cmd_ctx)->llm);
542 }
543 if ((*cmd_ctx)->lsm) {
544 free((*cmd_ctx)->lsm);
545 }
546 free(*cmd_ctx);
547 *cmd_ctx = NULL;
548 }
549 }
550
551 /*
552 * Notify UST applications using the shm mmap futex.
553 */
554 static int notify_ust_apps(int active)
555 {
556 char *wait_shm_mmap;
557
558 DBG("Notifying applications of session daemon state: %d", active);
559
560 /* See shm.c for this call implying mmap, shm and futex calls */
561 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
562 if (wait_shm_mmap == NULL) {
563 goto error;
564 }
565
566 /* Wake waiting process */
567 futex_wait_update((int32_t *) wait_shm_mmap, active);
568
569 /* Apps notified successfully */
570 return 0;
571
572 error:
573 return -1;
574 }
575
576 /*
577 * Setup the outgoing data buffer for the response (llm) by allocating the
578 * right amount of memory and copying the original information from the lsm
579 * structure.
580 *
581 * Return total size of the buffer pointed by buf.
582 */
583 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
584 {
585 int ret, buf_size;
586
587 buf_size = size;
588
589 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
590 if (cmd_ctx->llm == NULL) {
591 PERROR("zmalloc");
592 ret = -ENOMEM;
593 goto error;
594 }
595
596 /* Copy common data */
597 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
598 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
599
600 cmd_ctx->llm->data_size = size;
601 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
602
603 return buf_size;
604
605 error:
606 return ret;
607 }
608
609 /*
610 * Update the kernel poll set of all channel fd available over all tracing
611 * session. Add the wakeup pipe at the end of the set.
612 */
613 static int update_kernel_poll(struct lttng_poll_event *events)
614 {
615 int ret;
616 struct ltt_session *session;
617 struct ltt_kernel_channel *channel;
618
619 DBG("Updating kernel poll set");
620
621 session_lock_list();
622 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
623 session_lock(session);
624 if (session->kernel_session == NULL) {
625 session_unlock(session);
626 continue;
627 }
628
629 cds_list_for_each_entry(channel,
630 &session->kernel_session->channel_list.head, list) {
631 /* Add channel fd to the kernel poll set */
632 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
633 if (ret < 0) {
634 session_unlock(session);
635 goto error;
636 }
637 DBG("Channel fd %d added to kernel set", channel->fd);
638 }
639 session_unlock(session);
640 }
641 session_unlock_list();
642
643 return 0;
644
645 error:
646 session_unlock_list();
647 return -1;
648 }
649
650 /*
651 * Find the channel fd from 'fd' over all tracing session. When found, check
652 * for new channel stream and send those stream fds to the kernel consumer.
653 *
654 * Useful for CPU hotplug feature.
655 */
656 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
657 {
658 int ret = 0;
659 struct ltt_session *session;
660 struct ltt_kernel_session *ksess;
661 struct ltt_kernel_channel *channel;
662
663 DBG("Updating kernel streams for channel fd %d", fd);
664
665 session_lock_list();
666 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
667 session_lock(session);
668 if (session->kernel_session == NULL) {
669 session_unlock(session);
670 continue;
671 }
672 ksess = session->kernel_session;
673
674 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
675 if (channel->fd == fd) {
676 DBG("Channel found, updating kernel streams");
677 ret = kernel_open_channel_stream(channel);
678 if (ret < 0) {
679 goto error;
680 }
681 /* Update the stream global counter */
682 ksess->stream_count_global += ret;
683
684 /*
685 * Have we already sent fds to the consumer? If yes, it means
686 * that tracing is started so it is safe to send our updated
687 * stream fds.
688 */
689 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
690 struct lttng_ht_iter iter;
691 struct consumer_socket *socket;
692
693 rcu_read_lock();
694 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
695 &iter.iter, socket, node.node) {
696 /* Code flow error */
697 assert(socket->fd >= 0);
698
699 pthread_mutex_lock(socket->lock);
700 ret = kernel_consumer_send_channel_stream(socket,
701 channel, ksess,
702 session->output_traces ? 1 : 0);
703 pthread_mutex_unlock(socket->lock);
704 if (ret < 0) {
705 rcu_read_unlock();
706 goto error;
707 }
708 }
709 rcu_read_unlock();
710 }
711 goto error;
712 }
713 }
714 session_unlock(session);
715 }
716 session_unlock_list();
717 return ret;
718
719 error:
720 session_unlock(session);
721 session_unlock_list();
722 return ret;
723 }
724
725 /*
726 * For each tracing session, update newly registered apps. The session list
727 * lock MUST be acquired before calling this.
728 */
729 static void update_ust_app(int app_sock)
730 {
731 struct ltt_session *sess, *stmp;
732
733 /* For all tracing session(s) */
734 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
735 session_lock(sess);
736 if (sess->ust_session) {
737 ust_app_global_update(sess->ust_session, app_sock);
738 }
739 session_unlock(sess);
740 }
741 }
742
743 /*
744 * This thread manage event coming from the kernel.
745 *
746 * Features supported in this thread:
747 * -) CPU Hotplug
748 */
749 static void *thread_manage_kernel(void *data)
750 {
751 int ret, i, pollfd, update_poll_flag = 1, err = -1;
752 uint32_t revents, nb_fd;
753 char tmp;
754 struct lttng_poll_event events;
755
756 DBG("[thread] Thread manage kernel started");
757
758 health_register(HEALTH_TYPE_KERNEL);
759
760 /*
761 * This first step of the while is to clean this structure which could free
762 * non NULL pointers so initialize it before the loop.
763 */
764 lttng_poll_init(&events);
765
766 if (testpoint(thread_manage_kernel)) {
767 goto error_testpoint;
768 }
769
770 health_code_update();
771
772 if (testpoint(thread_manage_kernel_before_loop)) {
773 goto error_testpoint;
774 }
775
776 while (1) {
777 health_code_update();
778
779 if (update_poll_flag == 1) {
780 /* Clean events object. We are about to populate it again. */
781 lttng_poll_clean(&events);
782
783 ret = sessiond_set_thread_pollset(&events, 2);
784 if (ret < 0) {
785 goto error_poll_create;
786 }
787
788 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
789 if (ret < 0) {
790 goto error;
791 }
792
793 /* This will add the available kernel channel if any. */
794 ret = update_kernel_poll(&events);
795 if (ret < 0) {
796 goto error;
797 }
798 update_poll_flag = 0;
799 }
800
801 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
802
803 /* Poll infinite value of time */
804 restart:
805 health_poll_entry();
806 ret = lttng_poll_wait(&events, -1);
807 health_poll_exit();
808 if (ret < 0) {
809 /*
810 * Restart interrupted system call.
811 */
812 if (errno == EINTR) {
813 goto restart;
814 }
815 goto error;
816 } else if (ret == 0) {
817 /* Should not happen since timeout is infinite */
818 ERR("Return value of poll is 0 with an infinite timeout.\n"
819 "This should not have happened! Continuing...");
820 continue;
821 }
822
823 nb_fd = ret;
824
825 for (i = 0; i < nb_fd; i++) {
826 /* Fetch once the poll data */
827 revents = LTTNG_POLL_GETEV(&events, i);
828 pollfd = LTTNG_POLL_GETFD(&events, i);
829
830 health_code_update();
831
832 /* Thread quit pipe has been closed. Killing thread. */
833 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
834 if (ret) {
835 err = 0;
836 goto exit;
837 }
838
839 /* Check for data on kernel pipe */
840 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
841 do {
842 ret = read(kernel_poll_pipe[0], &tmp, 1);
843 } while (ret < 0 && errno == EINTR);
844 /*
845 * Ret value is useless here, if this pipe gets any actions an
846 * update is required anyway.
847 */
848 update_poll_flag = 1;
849 continue;
850 } else {
851 /*
852 * New CPU detected by the kernel. Adding kernel stream to
853 * kernel session and updating the kernel consumer
854 */
855 if (revents & LPOLLIN) {
856 ret = update_kernel_stream(&kconsumer_data, pollfd);
857 if (ret < 0) {
858 continue;
859 }
860 break;
861 /*
862 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
863 * and unregister kernel stream at this point.
864 */
865 }
866 }
867 }
868 }
869
870 exit:
871 error:
872 lttng_poll_clean(&events);
873 error_poll_create:
874 error_testpoint:
875 utils_close_pipe(kernel_poll_pipe);
876 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
877 if (err) {
878 health_error();
879 ERR("Health error occurred in %s", __func__);
880 WARN("Kernel thread died unexpectedly. "
881 "Kernel tracing can continue but CPU hotplug is disabled.");
882 }
883 health_unregister();
884 DBG("Kernel thread dying");
885 return NULL;
886 }
887
888 /*
889 * Signal pthread condition of the consumer data that the thread.
890 */
891 static void signal_consumer_condition(struct consumer_data *data, int state)
892 {
893 pthread_mutex_lock(&data->cond_mutex);
894
895 /*
896 * The state is set before signaling. It can be any value, it's the waiter
897 * job to correctly interpret this condition variable associated to the
898 * consumer pthread_cond.
899 *
900 * A value of 0 means that the corresponding thread of the consumer data
901 * was not started. 1 indicates that the thread has started and is ready
902 * for action. A negative value means that there was an error during the
903 * thread bootstrap.
904 */
905 data->consumer_thread_is_ready = state;
906 (void) pthread_cond_signal(&data->cond);
907
908 pthread_mutex_unlock(&data->cond_mutex);
909 }
910
911 /*
912 * This thread manage the consumer error sent back to the session daemon.
913 */
914 static void *thread_manage_consumer(void *data)
915 {
916 int sock = -1, i, ret, pollfd, err = -1;
917 uint32_t revents, nb_fd;
918 enum lttcomm_return_code code;
919 struct lttng_poll_event events;
920 struct consumer_data *consumer_data = data;
921
922 DBG("[thread] Manage consumer started");
923
924 health_register(HEALTH_TYPE_CONSUMER);
925
926 health_code_update();
927
928 /*
929 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
930 * metadata_sock. Nothing more will be added to this poll set.
931 */
932 ret = sessiond_set_thread_pollset(&events, 3);
933 if (ret < 0) {
934 goto error_poll;
935 }
936
937 /*
938 * The error socket here is already in a listening state which was done
939 * just before spawning this thread to avoid a race between the consumer
940 * daemon exec trying to connect and the listen() call.
941 */
942 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
943 if (ret < 0) {
944 goto error;
945 }
946
947 health_code_update();
948
949 /* Infinite blocking call, waiting for transmission */
950 restart:
951 health_poll_entry();
952
953 if (testpoint(thread_manage_consumer)) {
954 goto error;
955 }
956
957 ret = lttng_poll_wait(&events, -1);
958 health_poll_exit();
959 if (ret < 0) {
960 /*
961 * Restart interrupted system call.
962 */
963 if (errno == EINTR) {
964 goto restart;
965 }
966 goto error;
967 }
968
969 nb_fd = ret;
970
971 for (i = 0; i < nb_fd; i++) {
972 /* Fetch once the poll data */
973 revents = LTTNG_POLL_GETEV(&events, i);
974 pollfd = LTTNG_POLL_GETFD(&events, i);
975
976 health_code_update();
977
978 /* Thread quit pipe has been closed. Killing thread. */
979 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
980 if (ret) {
981 err = 0;
982 goto exit;
983 }
984
985 /* Event on the registration socket */
986 if (pollfd == consumer_data->err_sock) {
987 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
988 ERR("consumer err socket poll error");
989 goto error;
990 }
991 }
992 }
993
994 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
995 if (sock < 0) {
996 goto error;
997 }
998
999 /*
1000 * Set the CLOEXEC flag. Return code is useless because either way, the
1001 * show must go on.
1002 */
1003 (void) utils_set_fd_cloexec(sock);
1004
1005 health_code_update();
1006
1007 DBG2("Receiving code from consumer err_sock");
1008
1009 /* Getting status code from kconsumerd */
1010 ret = lttcomm_recv_unix_sock(sock, &code,
1011 sizeof(enum lttcomm_return_code));
1012 if (ret <= 0) {
1013 goto error;
1014 }
1015
1016 health_code_update();
1017
1018 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1019 /* Connect both socket, command and metadata. */
1020 consumer_data->cmd_sock =
1021 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1022 consumer_data->metadata_sock.fd =
1023 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1024 if (consumer_data->cmd_sock < 0 ||
1025 consumer_data->metadata_sock.fd < 0) {
1026 PERROR("consumer connect cmd socket");
1027 /* On error, signal condition and quit. */
1028 signal_consumer_condition(consumer_data, -1);
1029 goto error;
1030 }
1031 /* Create metadata socket lock. */
1032 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1033 if (consumer_data->metadata_sock.lock == NULL) {
1034 PERROR("zmalloc pthread mutex");
1035 ret = -1;
1036 goto error;
1037 }
1038 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1039
1040 signal_consumer_condition(consumer_data, 1);
1041 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1042 DBG("Consumer metadata socket ready (fd: %d)",
1043 consumer_data->metadata_sock.fd);
1044 } else {
1045 ERR("consumer error when waiting for SOCK_READY : %s",
1046 lttcomm_get_readable_code(-code));
1047 goto error;
1048 }
1049
1050 /* Remove the consumerd error sock since we've established a connexion */
1051 ret = lttng_poll_del(&events, consumer_data->err_sock);
1052 if (ret < 0) {
1053 goto error;
1054 }
1055
1056 /* Add new accepted error socket. */
1057 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1058 if (ret < 0) {
1059 goto error;
1060 }
1061
1062 /* Add metadata socket that is successfully connected. */
1063 ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
1064 LPOLLIN | LPOLLRDHUP);
1065 if (ret < 0) {
1066 goto error;
1067 }
1068
1069 health_code_update();
1070
1071 /* Infinite blocking call, waiting for transmission */
1072 restart_poll:
1073 while (1) {
1074 health_poll_entry();
1075 ret = lttng_poll_wait(&events, -1);
1076 health_poll_exit();
1077 if (ret < 0) {
1078 /*
1079 * Restart interrupted system call.
1080 */
1081 if (errno == EINTR) {
1082 goto restart_poll;
1083 }
1084 goto error;
1085 }
1086
1087 nb_fd = ret;
1088
1089 for (i = 0; i < nb_fd; i++) {
1090 /* Fetch once the poll data */
1091 revents = LTTNG_POLL_GETEV(&events, i);
1092 pollfd = LTTNG_POLL_GETFD(&events, i);
1093
1094 health_code_update();
1095
1096 /* Thread quit pipe has been closed. Killing thread. */
1097 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1098 if (ret) {
1099 err = 0;
1100 goto exit;
1101 }
1102
1103 if (pollfd == sock) {
1104 /* Event on the consumerd socket */
1105 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1106 ERR("consumer err socket second poll error");
1107 goto error;
1108 }
1109 health_code_update();
1110 /* Wait for any kconsumerd error */
1111 ret = lttcomm_recv_unix_sock(sock, &code,
1112 sizeof(enum lttcomm_return_code));
1113 if (ret <= 0) {
1114 ERR("consumer closed the command socket");
1115 goto error;
1116 }
1117
1118 ERR("consumer return code : %s",
1119 lttcomm_get_readable_code(-code));
1120
1121 goto exit;
1122 } else if (pollfd == consumer_data->metadata_sock.fd) {
1123 /* UST metadata requests */
1124 ret = ust_consumer_metadata_request(
1125 &consumer_data->metadata_sock);
1126 if (ret < 0) {
1127 ERR("Handling metadata request");
1128 goto error;
1129 }
1130 break;
1131 } else {
1132 ERR("Unknown pollfd");
1133 goto error;
1134 }
1135 }
1136 health_code_update();
1137 }
1138
1139 exit:
1140 error:
1141 /* Immediately set the consumerd state to stopped */
1142 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1143 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1144 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1145 consumer_data->type == LTTNG_CONSUMER32_UST) {
1146 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1147 } else {
1148 /* Code flow error... */
1149 assert(0);
1150 }
1151
1152 if (consumer_data->err_sock >= 0) {
1153 ret = close(consumer_data->err_sock);
1154 if (ret) {
1155 PERROR("close");
1156 }
1157 consumer_data->err_sock = -1;
1158 }
1159 if (consumer_data->cmd_sock >= 0) {
1160 ret = close(consumer_data->cmd_sock);
1161 if (ret) {
1162 PERROR("close");
1163 }
1164 consumer_data->cmd_sock = -1;
1165 }
1166 if (consumer_data->metadata_sock.fd >= 0) {
1167 ret = close(consumer_data->metadata_sock.fd);
1168 if (ret) {
1169 PERROR("close");
1170 }
1171 }
1172 /* Cleanup metadata socket mutex. */
1173 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1174 free(consumer_data->metadata_sock.lock);
1175
1176 if (sock >= 0) {
1177 ret = close(sock);
1178 if (ret) {
1179 PERROR("close");
1180 }
1181 }
1182
1183 unlink(consumer_data->err_unix_sock_path);
1184 unlink(consumer_data->cmd_unix_sock_path);
1185 consumer_data->pid = 0;
1186
1187 lttng_poll_clean(&events);
1188 error_poll:
1189 if (err) {
1190 health_error();
1191 ERR("Health error occurred in %s", __func__);
1192 }
1193 health_unregister();
1194 DBG("consumer thread cleanup completed");
1195
1196 return NULL;
1197 }
1198
1199 /*
1200 * This thread manage application communication.
1201 */
1202 static void *thread_manage_apps(void *data)
1203 {
1204 int i, ret, pollfd, err = -1;
1205 uint32_t revents, nb_fd;
1206 struct lttng_poll_event events;
1207
1208 DBG("[thread] Manage application started");
1209
1210 rcu_register_thread();
1211 rcu_thread_online();
1212
1213 health_register(HEALTH_TYPE_APP_MANAGE);
1214
1215 if (testpoint(thread_manage_apps)) {
1216 goto error_testpoint;
1217 }
1218
1219 health_code_update();
1220
1221 ret = sessiond_set_thread_pollset(&events, 2);
1222 if (ret < 0) {
1223 goto error_poll_create;
1224 }
1225
1226 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1227 if (ret < 0) {
1228 goto error;
1229 }
1230
1231 if (testpoint(thread_manage_apps_before_loop)) {
1232 goto error;
1233 }
1234
1235 health_code_update();
1236
1237 while (1) {
1238 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1239
1240 /* Inifinite blocking call, waiting for transmission */
1241 restart:
1242 health_poll_entry();
1243 ret = lttng_poll_wait(&events, -1);
1244 health_poll_exit();
1245 if (ret < 0) {
1246 /*
1247 * Restart interrupted system call.
1248 */
1249 if (errno == EINTR) {
1250 goto restart;
1251 }
1252 goto error;
1253 }
1254
1255 nb_fd = ret;
1256
1257 for (i = 0; i < nb_fd; i++) {
1258 /* Fetch once the poll data */
1259 revents = LTTNG_POLL_GETEV(&events, i);
1260 pollfd = LTTNG_POLL_GETFD(&events, i);
1261
1262 health_code_update();
1263
1264 /* Thread quit pipe has been closed. Killing thread. */
1265 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1266 if (ret) {
1267 err = 0;
1268 goto exit;
1269 }
1270
1271 /* Inspect the apps cmd pipe */
1272 if (pollfd == apps_cmd_pipe[0]) {
1273 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1274 ERR("Apps command pipe error");
1275 goto error;
1276 } else if (revents & LPOLLIN) {
1277 int sock;
1278
1279 /* Empty pipe */
1280 do {
1281 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1282 } while (ret < 0 && errno == EINTR);
1283 if (ret < 0 || ret < sizeof(sock)) {
1284 PERROR("read apps cmd pipe");
1285 goto error;
1286 }
1287
1288 health_code_update();
1289
1290 /*
1291 * We only monitor the error events of the socket. This
1292 * thread does not handle any incoming data from UST
1293 * (POLLIN).
1294 */
1295 ret = lttng_poll_add(&events, sock,
1296 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1297 if (ret < 0) {
1298 goto error;
1299 }
1300
1301 /* Set socket timeout for both receiving and ending */
1302 (void) lttcomm_setsockopt_rcv_timeout(sock,
1303 app_socket_timeout);
1304 (void) lttcomm_setsockopt_snd_timeout(sock,
1305 app_socket_timeout);
1306
1307 DBG("Apps with sock %d added to poll set", sock);
1308
1309 health_code_update();
1310
1311 break;
1312 }
1313 } else {
1314 /*
1315 * At this point, we know that a registered application made
1316 * the event at poll_wait.
1317 */
1318 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1319 /* Removing from the poll set */
1320 ret = lttng_poll_del(&events, pollfd);
1321 if (ret < 0) {
1322 goto error;
1323 }
1324
1325 /* Socket closed on remote end. */
1326 ust_app_unregister(pollfd);
1327 break;
1328 }
1329 }
1330
1331 health_code_update();
1332 }
1333 }
1334
1335 exit:
1336 error:
1337 lttng_poll_clean(&events);
1338 error_poll_create:
1339 error_testpoint:
1340 utils_close_pipe(apps_cmd_pipe);
1341 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1342
1343 /*
1344 * We don't clean the UST app hash table here since already registered
1345 * applications can still be controlled so let them be until the session
1346 * daemon dies or the applications stop.
1347 */
1348
1349 if (err) {
1350 health_error();
1351 ERR("Health error occurred in %s", __func__);
1352 }
1353 health_unregister();
1354 DBG("Application communication apps thread cleanup complete");
1355 rcu_thread_offline();
1356 rcu_unregister_thread();
1357 return NULL;
1358 }
1359
1360 /*
1361 * Send a socket to a thread This is called from the dispatch UST registration
1362 * thread once all sockets are set for the application.
1363 *
1364 * On success, return 0 else a negative value being the errno message of the
1365 * write().
1366 */
1367 static int send_socket_to_thread(int fd, int sock)
1368 {
1369 int ret;
1370
1371 /* Sockets MUST be set or else this should not have been called. */
1372 assert(fd >= 0);
1373 assert(sock >= 0);
1374
1375 do {
1376 ret = write(fd, &sock, sizeof(sock));
1377 } while (ret < 0 && errno == EINTR);
1378 if (ret < 0 || ret != sizeof(sock)) {
1379 PERROR("write apps pipe %d", fd);
1380 if (ret < 0) {
1381 ret = -errno;
1382 }
1383 goto error;
1384 }
1385
1386 /* All good. Don't send back the write positive ret value. */
1387 ret = 0;
1388 error:
1389 return ret;
1390 }
1391
1392 /*
1393 * Sanitize the wait queue of the dispatch registration thread meaning removing
1394 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1395 * notify socket is never received.
1396 */
1397 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1398 {
1399 int ret, nb_fd = 0, i;
1400 unsigned int fd_added = 0;
1401 struct lttng_poll_event events;
1402 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1403
1404 assert(wait_queue);
1405
1406 lttng_poll_init(&events);
1407
1408 /* Just skip everything for an empty queue. */
1409 if (!wait_queue->count) {
1410 goto end;
1411 }
1412
1413 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1414 if (ret < 0) {
1415 goto error_create;
1416 }
1417
1418 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1419 &wait_queue->head, head) {
1420 assert(wait_node->app);
1421 ret = lttng_poll_add(&events, wait_node->app->sock,
1422 LPOLLHUP | LPOLLERR);
1423 if (ret < 0) {
1424 goto error;
1425 }
1426
1427 fd_added = 1;
1428 }
1429
1430 if (!fd_added) {
1431 goto end;
1432 }
1433
1434 /*
1435 * Poll but don't block so we can quickly identify the faulty events and
1436 * clean them afterwards from the wait queue.
1437 */
1438 ret = lttng_poll_wait(&events, 0);
1439 if (ret < 0) {
1440 goto error;
1441 }
1442 nb_fd = ret;
1443
1444 for (i = 0; i < nb_fd; i++) {
1445 /* Get faulty FD. */
1446 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1447 int pollfd = LTTNG_POLL_GETFD(&events, i);
1448
1449 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1450 &wait_queue->head, head) {
1451 if (pollfd == wait_node->app->sock &&
1452 (revents & (LPOLLHUP | LPOLLERR))) {
1453 cds_list_del(&wait_node->head);
1454 wait_queue->count--;
1455 ust_app_destroy(wait_node->app);
1456 free(wait_node);
1457 break;
1458 }
1459 }
1460 }
1461
1462 if (nb_fd > 0) {
1463 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1464 }
1465
1466 end:
1467 lttng_poll_clean(&events);
1468 return;
1469
1470 error:
1471 lttng_poll_clean(&events);
1472 error_create:
1473 ERR("Unable to sanitize wait queue");
1474 return;
1475 }
1476
1477 /*
1478 * Dispatch request from the registration threads to the application
1479 * communication thread.
1480 */
1481 static void *thread_dispatch_ust_registration(void *data)
1482 {
1483 int ret, err = -1;
1484 struct cds_wfq_node *node;
1485 struct ust_command *ust_cmd = NULL;
1486 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1487 struct ust_reg_wait_queue wait_queue = {
1488 .count = 0,
1489 };
1490
1491 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1492
1493 health_code_update();
1494
1495 CDS_INIT_LIST_HEAD(&wait_queue.head);
1496
1497 DBG("[thread] Dispatch UST command started");
1498
1499 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1500 health_code_update();
1501
1502 /* Atomically prepare the queue futex */
1503 futex_nto1_prepare(&ust_cmd_queue.futex);
1504
1505 do {
1506 struct ust_app *app = NULL;
1507 ust_cmd = NULL;
1508
1509 /*
1510 * Make sure we don't have node(s) that have hung up before receiving
1511 * the notify socket. This is to clean the list in order to avoid
1512 * memory leaks from notify socket that are never seen.
1513 */
1514 sanitize_wait_queue(&wait_queue);
1515
1516 health_code_update();
1517 /* Dequeue command for registration */
1518 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1519 if (node == NULL) {
1520 DBG("Woken up but nothing in the UST command queue");
1521 /* Continue thread execution */
1522 break;
1523 }
1524
1525 ust_cmd = caa_container_of(node, struct ust_command, node);
1526
1527 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1528 " gid:%d sock:%d name:%s (version %d.%d)",
1529 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1530 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1531 ust_cmd->sock, ust_cmd->reg_msg.name,
1532 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1533
1534 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1535 wait_node = zmalloc(sizeof(*wait_node));
1536 if (!wait_node) {
1537 PERROR("zmalloc wait_node dispatch");
1538 ret = close(ust_cmd->sock);
1539 if (ret < 0) {
1540 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1541 }
1542 lttng_fd_put(1, LTTNG_FD_APPS);
1543 free(ust_cmd);
1544 goto error;
1545 }
1546 CDS_INIT_LIST_HEAD(&wait_node->head);
1547
1548 /* Create application object if socket is CMD. */
1549 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1550 ust_cmd->sock);
1551 if (!wait_node->app) {
1552 ret = close(ust_cmd->sock);
1553 if (ret < 0) {
1554 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1555 }
1556 lttng_fd_put(1, LTTNG_FD_APPS);
1557 free(wait_node);
1558 free(ust_cmd);
1559 continue;
1560 }
1561 /*
1562 * Add application to the wait queue so we can set the notify
1563 * socket before putting this object in the global ht.
1564 */
1565 cds_list_add(&wait_node->head, &wait_queue.head);
1566 wait_queue.count++;
1567
1568 free(ust_cmd);
1569 /*
1570 * We have to continue here since we don't have the notify
1571 * socket and the application MUST be added to the hash table
1572 * only at that moment.
1573 */
1574 continue;
1575 } else {
1576 /*
1577 * Look for the application in the local wait queue and set the
1578 * notify socket if found.
1579 */
1580 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1581 &wait_queue.head, head) {
1582 health_code_update();
1583 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1584 wait_node->app->notify_sock = ust_cmd->sock;
1585 cds_list_del(&wait_node->head);
1586 wait_queue.count--;
1587 app = wait_node->app;
1588 free(wait_node);
1589 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1590 break;
1591 }
1592 }
1593
1594 /*
1595 * With no application at this stage the received socket is
1596 * basically useless so close it before we free the cmd data
1597 * structure for good.
1598 */
1599 if (!app) {
1600 ret = close(ust_cmd->sock);
1601 if (ret < 0) {
1602 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1603 }
1604 lttng_fd_put(1, LTTNG_FD_APPS);
1605 }
1606 free(ust_cmd);
1607 }
1608
1609 if (app) {
1610 /*
1611 * @session_lock_list
1612 *
1613 * Lock the global session list so from the register up to the
1614 * registration done message, no thread can see the application
1615 * and change its state.
1616 */
1617 session_lock_list();
1618 rcu_read_lock();
1619
1620 /*
1621 * Add application to the global hash table. This needs to be
1622 * done before the update to the UST registry can locate the
1623 * application.
1624 */
1625 ust_app_add(app);
1626
1627 /* Set app version. This call will print an error if needed. */
1628 (void) ust_app_version(app);
1629
1630 /* Send notify socket through the notify pipe. */
1631 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1632 app->notify_sock);
1633 if (ret < 0) {
1634 rcu_read_unlock();
1635 session_unlock_list();
1636 /* No notify thread, stop the UST tracing. */
1637 goto error;
1638 }
1639
1640 /*
1641 * Update newly registered application with the tracing
1642 * registry info already enabled information.
1643 */
1644 update_ust_app(app->sock);
1645
1646 /*
1647 * Don't care about return value. Let the manage apps threads
1648 * handle app unregistration upon socket close.
1649 */
1650 (void) ust_app_register_done(app->sock);
1651
1652 /*
1653 * Even if the application socket has been closed, send the app
1654 * to the thread and unregistration will take place at that
1655 * place.
1656 */
1657 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1658 if (ret < 0) {
1659 rcu_read_unlock();
1660 session_unlock_list();
1661 /* No apps. thread, stop the UST tracing. */
1662 goto error;
1663 }
1664
1665 rcu_read_unlock();
1666 session_unlock_list();
1667 }
1668 } while (node != NULL);
1669
1670 health_poll_entry();
1671 /* Futex wait on queue. Blocking call on futex() */
1672 futex_nto1_wait(&ust_cmd_queue.futex);
1673 health_poll_exit();
1674 }
1675 /* Normal exit, no error */
1676 err = 0;
1677
1678 error:
1679 /* Clean up wait queue. */
1680 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1681 &wait_queue.head, head) {
1682 cds_list_del(&wait_node->head);
1683 wait_queue.count--;
1684 free(wait_node);
1685 }
1686
1687 DBG("Dispatch thread dying");
1688 if (err) {
1689 health_error();
1690 ERR("Health error occurred in %s", __func__);
1691 }
1692 health_unregister();
1693 return NULL;
1694 }
1695
1696 /*
1697 * This thread manage application registration.
1698 */
1699 static void *thread_registration_apps(void *data)
1700 {
1701 int sock = -1, i, ret, pollfd, err = -1;
1702 uint32_t revents, nb_fd;
1703 struct lttng_poll_event events;
1704 /*
1705 * Get allocated in this thread, enqueued to a global queue, dequeued and
1706 * freed in the manage apps thread.
1707 */
1708 struct ust_command *ust_cmd = NULL;
1709
1710 DBG("[thread] Manage application registration started");
1711
1712 health_register(HEALTH_TYPE_APP_REG);
1713
1714 if (testpoint(thread_registration_apps)) {
1715 goto error_testpoint;
1716 }
1717
1718 ret = lttcomm_listen_unix_sock(apps_sock);
1719 if (ret < 0) {
1720 goto error_listen;
1721 }
1722
1723 /*
1724 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1725 * more will be added to this poll set.
1726 */
1727 ret = sessiond_set_thread_pollset(&events, 2);
1728 if (ret < 0) {
1729 goto error_create_poll;
1730 }
1731
1732 /* Add the application registration socket */
1733 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1734 if (ret < 0) {
1735 goto error_poll_add;
1736 }
1737
1738 /* Notify all applications to register */
1739 ret = notify_ust_apps(1);
1740 if (ret < 0) {
1741 ERR("Failed to notify applications or create the wait shared memory.\n"
1742 "Execution continues but there might be problem for already\n"
1743 "running applications that wishes to register.");
1744 }
1745
1746 while (1) {
1747 DBG("Accepting application registration");
1748
1749 /* Inifinite blocking call, waiting for transmission */
1750 restart:
1751 health_poll_entry();
1752 ret = lttng_poll_wait(&events, -1);
1753 health_poll_exit();
1754 if (ret < 0) {
1755 /*
1756 * Restart interrupted system call.
1757 */
1758 if (errno == EINTR) {
1759 goto restart;
1760 }
1761 goto error;
1762 }
1763
1764 nb_fd = ret;
1765
1766 for (i = 0; i < nb_fd; i++) {
1767 health_code_update();
1768
1769 /* Fetch once the poll data */
1770 revents = LTTNG_POLL_GETEV(&events, i);
1771 pollfd = LTTNG_POLL_GETFD(&events, i);
1772
1773 /* Thread quit pipe has been closed. Killing thread. */
1774 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1775 if (ret) {
1776 err = 0;
1777 goto exit;
1778 }
1779
1780 /* Event on the registration socket */
1781 if (pollfd == apps_sock) {
1782 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1783 ERR("Register apps socket poll error");
1784 goto error;
1785 } else if (revents & LPOLLIN) {
1786 sock = lttcomm_accept_unix_sock(apps_sock);
1787 if (sock < 0) {
1788 goto error;
1789 }
1790
1791 /*
1792 * Set the CLOEXEC flag. Return code is useless because
1793 * either way, the show must go on.
1794 */
1795 (void) utils_set_fd_cloexec(sock);
1796
1797 /* Create UST registration command for enqueuing */
1798 ust_cmd = zmalloc(sizeof(struct ust_command));
1799 if (ust_cmd == NULL) {
1800 PERROR("ust command zmalloc");
1801 goto error;
1802 }
1803
1804 /*
1805 * Using message-based transmissions to ensure we don't
1806 * have to deal with partially received messages.
1807 */
1808 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1809 if (ret < 0) {
1810 ERR("Exhausted file descriptors allowed for applications.");
1811 free(ust_cmd);
1812 ret = close(sock);
1813 if (ret) {
1814 PERROR("close");
1815 }
1816 sock = -1;
1817 continue;
1818 }
1819
1820 health_code_update();
1821 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1822 if (ret < 0) {
1823 free(ust_cmd);
1824 /* Close socket of the application. */
1825 ret = close(sock);
1826 if (ret) {
1827 PERROR("close");
1828 }
1829 lttng_fd_put(LTTNG_FD_APPS, 1);
1830 sock = -1;
1831 continue;
1832 }
1833 health_code_update();
1834
1835 ust_cmd->sock = sock;
1836 sock = -1;
1837
1838 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1839 " gid:%d sock:%d name:%s (version %d.%d)",
1840 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1841 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1842 ust_cmd->sock, ust_cmd->reg_msg.name,
1843 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1844
1845 /*
1846 * Lock free enqueue the registration request. The red pill
1847 * has been taken! This apps will be part of the *system*.
1848 */
1849 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1850
1851 /*
1852 * Wake the registration queue futex. Implicit memory
1853 * barrier with the exchange in cds_wfq_enqueue.
1854 */
1855 futex_nto1_wake(&ust_cmd_queue.futex);
1856 }
1857 }
1858 }
1859 }
1860
1861 exit:
1862 error:
1863 if (err) {
1864 health_error();
1865 ERR("Health error occurred in %s", __func__);
1866 }
1867
1868 /* Notify that the registration thread is gone */
1869 notify_ust_apps(0);
1870
1871 if (apps_sock >= 0) {
1872 ret = close(apps_sock);
1873 if (ret) {
1874 PERROR("close");
1875 }
1876 }
1877 if (sock >= 0) {
1878 ret = close(sock);
1879 if (ret) {
1880 PERROR("close");
1881 }
1882 lttng_fd_put(LTTNG_FD_APPS, 1);
1883 }
1884 unlink(apps_unix_sock_path);
1885
1886 error_poll_add:
1887 lttng_poll_clean(&events);
1888 error_listen:
1889 error_create_poll:
1890 error_testpoint:
1891 DBG("UST Registration thread cleanup complete");
1892 health_unregister();
1893
1894 return NULL;
1895 }
1896
1897 /*
1898 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1899 * exec or it will fails.
1900 */
1901 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1902 {
1903 int ret, clock_ret;
1904 struct timespec timeout;
1905
1906 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1907 consumer_data->consumer_thread_is_ready = 0;
1908
1909 /* Setup pthread condition */
1910 ret = pthread_condattr_init(&consumer_data->condattr);
1911 if (ret != 0) {
1912 errno = ret;
1913 PERROR("pthread_condattr_init consumer data");
1914 goto error;
1915 }
1916
1917 /*
1918 * Set the monotonic clock in order to make sure we DO NOT jump in time
1919 * between the clock_gettime() call and the timedwait call. See bug #324
1920 * for a more details and how we noticed it.
1921 */
1922 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1923 if (ret != 0) {
1924 errno = ret;
1925 PERROR("pthread_condattr_setclock consumer data");
1926 goto error;
1927 }
1928
1929 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1930 if (ret != 0) {
1931 errno = ret;
1932 PERROR("pthread_cond_init consumer data");
1933 goto error;
1934 }
1935
1936 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1937 consumer_data);
1938 if (ret != 0) {
1939 PERROR("pthread_create consumer");
1940 ret = -1;
1941 goto error;
1942 }
1943
1944 /* We are about to wait on a pthread condition */
1945 pthread_mutex_lock(&consumer_data->cond_mutex);
1946
1947 /* Get time for sem_timedwait absolute timeout */
1948 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1949 /*
1950 * Set the timeout for the condition timed wait even if the clock gettime
1951 * call fails since we might loop on that call and we want to avoid to
1952 * increment the timeout too many times.
1953 */
1954 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1955
1956 /*
1957 * The following loop COULD be skipped in some conditions so this is why we
1958 * set ret to 0 in order to make sure at least one round of the loop is
1959 * done.
1960 */
1961 ret = 0;
1962
1963 /*
1964 * Loop until the condition is reached or when a timeout is reached. Note
1965 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1966 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1967 * possible. This loop does not take any chances and works with both of
1968 * them.
1969 */
1970 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1971 if (clock_ret < 0) {
1972 PERROR("clock_gettime spawn consumer");
1973 /* Infinite wait for the consumerd thread to be ready */
1974 ret = pthread_cond_wait(&consumer_data->cond,
1975 &consumer_data->cond_mutex);
1976 } else {
1977 ret = pthread_cond_timedwait(&consumer_data->cond,
1978 &consumer_data->cond_mutex, &timeout);
1979 }
1980 }
1981
1982 /* Release the pthread condition */
1983 pthread_mutex_unlock(&consumer_data->cond_mutex);
1984
1985 if (ret != 0) {
1986 errno = ret;
1987 if (ret == ETIMEDOUT) {
1988 /*
1989 * Call has timed out so we kill the kconsumerd_thread and return
1990 * an error.
1991 */
1992 ERR("Condition timed out. The consumer thread was never ready."
1993 " Killing it");
1994 ret = pthread_cancel(consumer_data->thread);
1995 if (ret < 0) {
1996 PERROR("pthread_cancel consumer thread");
1997 }
1998 } else {
1999 PERROR("pthread_cond_wait failed consumer thread");
2000 }
2001 goto error;
2002 }
2003
2004 pthread_mutex_lock(&consumer_data->pid_mutex);
2005 if (consumer_data->pid == 0) {
2006 ERR("Consumerd did not start");
2007 pthread_mutex_unlock(&consumer_data->pid_mutex);
2008 goto error;
2009 }
2010 pthread_mutex_unlock(&consumer_data->pid_mutex);
2011
2012 return 0;
2013
2014 error:
2015 return ret;
2016 }
2017
2018 /*
2019 * Join consumer thread
2020 */
2021 static int join_consumer_thread(struct consumer_data *consumer_data)
2022 {
2023 void *status;
2024
2025 /* Consumer pid must be a real one. */
2026 if (consumer_data->pid > 0) {
2027 int ret;
2028 ret = kill(consumer_data->pid, SIGTERM);
2029 if (ret) {
2030 ERR("Error killing consumer daemon");
2031 return ret;
2032 }
2033 return pthread_join(consumer_data->thread, &status);
2034 } else {
2035 return 0;
2036 }
2037 }
2038
2039 /*
2040 * Fork and exec a consumer daemon (consumerd).
2041 *
2042 * Return pid if successful else -1.
2043 */
2044 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2045 {
2046 int ret;
2047 pid_t pid;
2048 const char *consumer_to_use;
2049 const char *verbosity;
2050 struct stat st;
2051
2052 DBG("Spawning consumerd");
2053
2054 pid = fork();
2055 if (pid == 0) {
2056 /*
2057 * Exec consumerd.
2058 */
2059 if (opt_verbose_consumer) {
2060 verbosity = "--verbose";
2061 } else {
2062 verbosity = "--quiet";
2063 }
2064 switch (consumer_data->type) {
2065 case LTTNG_CONSUMER_KERNEL:
2066 /*
2067 * Find out which consumerd to execute. We will first try the
2068 * 64-bit path, then the sessiond's installation directory, and
2069 * fallback on the 32-bit one,
2070 */
2071 DBG3("Looking for a kernel consumer at these locations:");
2072 DBG3(" 1) %s", consumerd64_bin);
2073 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2074 DBG3(" 3) %s", consumerd32_bin);
2075 if (stat(consumerd64_bin, &st) == 0) {
2076 DBG3("Found location #1");
2077 consumer_to_use = consumerd64_bin;
2078 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2079 DBG3("Found location #2");
2080 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2081 } else if (stat(consumerd32_bin, &st) == 0) {
2082 DBG3("Found location #3");
2083 consumer_to_use = consumerd32_bin;
2084 } else {
2085 DBG("Could not find any valid consumerd executable");
2086 break;
2087 }
2088 DBG("Using kernel consumer at: %s", consumer_to_use);
2089 execl(consumer_to_use,
2090 "lttng-consumerd", verbosity, "-k",
2091 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2092 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2093 NULL);
2094 break;
2095 case LTTNG_CONSUMER64_UST:
2096 {
2097 char *tmpnew = NULL;
2098
2099 if (consumerd64_libdir[0] != '\0') {
2100 char *tmp;
2101 size_t tmplen;
2102
2103 tmp = getenv("LD_LIBRARY_PATH");
2104 if (!tmp) {
2105 tmp = "";
2106 }
2107 tmplen = strlen("LD_LIBRARY_PATH=")
2108 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2109 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2110 if (!tmpnew) {
2111 ret = -ENOMEM;
2112 goto error;
2113 }
2114 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2115 strcat(tmpnew, consumerd64_libdir);
2116 if (tmp[0] != '\0') {
2117 strcat(tmpnew, ":");
2118 strcat(tmpnew, tmp);
2119 }
2120 ret = putenv(tmpnew);
2121 if (ret) {
2122 ret = -errno;
2123 free(tmpnew);
2124 goto error;
2125 }
2126 }
2127 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2128 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2129 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2130 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2131 NULL);
2132 if (consumerd64_libdir[0] != '\0') {
2133 free(tmpnew);
2134 }
2135 if (ret) {
2136 goto error;
2137 }
2138 break;
2139 }
2140 case LTTNG_CONSUMER32_UST:
2141 {
2142 char *tmpnew = NULL;
2143
2144 if (consumerd32_libdir[0] != '\0') {
2145 char *tmp;
2146 size_t tmplen;
2147
2148 tmp = getenv("LD_LIBRARY_PATH");
2149 if (!tmp) {
2150 tmp = "";
2151 }
2152 tmplen = strlen("LD_LIBRARY_PATH=")
2153 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2154 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2155 if (!tmpnew) {
2156 ret = -ENOMEM;
2157 goto error;
2158 }
2159 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2160 strcat(tmpnew, consumerd32_libdir);
2161 if (tmp[0] != '\0') {
2162 strcat(tmpnew, ":");
2163 strcat(tmpnew, tmp);
2164 }
2165 ret = putenv(tmpnew);
2166 if (ret) {
2167 ret = -errno;
2168 free(tmpnew);
2169 goto error;
2170 }
2171 }
2172 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2173 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2174 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2175 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2176 NULL);
2177 if (consumerd32_libdir[0] != '\0') {
2178 free(tmpnew);
2179 }
2180 if (ret) {
2181 goto error;
2182 }
2183 break;
2184 }
2185 default:
2186 PERROR("unknown consumer type");
2187 exit(EXIT_FAILURE);
2188 }
2189 if (errno != 0) {
2190 PERROR("kernel start consumer exec");
2191 }
2192 exit(EXIT_FAILURE);
2193 } else if (pid > 0) {
2194 ret = pid;
2195 } else {
2196 PERROR("start consumer fork");
2197 ret = -errno;
2198 }
2199 error:
2200 return ret;
2201 }
2202
2203 /*
2204 * Spawn the consumerd daemon and session daemon thread.
2205 */
2206 static int start_consumerd(struct consumer_data *consumer_data)
2207 {
2208 int ret;
2209
2210 /*
2211 * Set the listen() state on the socket since there is a possible race
2212 * between the exec() of the consumer daemon and this call if place in the
2213 * consumer thread. See bug #366 for more details.
2214 */
2215 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2216 if (ret < 0) {
2217 goto error;
2218 }
2219
2220 pthread_mutex_lock(&consumer_data->pid_mutex);
2221 if (consumer_data->pid != 0) {
2222 pthread_mutex_unlock(&consumer_data->pid_mutex);
2223 goto end;
2224 }
2225
2226 ret = spawn_consumerd(consumer_data);
2227 if (ret < 0) {
2228 ERR("Spawning consumerd failed");
2229 pthread_mutex_unlock(&consumer_data->pid_mutex);
2230 goto error;
2231 }
2232
2233 /* Setting up the consumer_data pid */
2234 consumer_data->pid = ret;
2235 DBG2("Consumer pid %d", consumer_data->pid);
2236 pthread_mutex_unlock(&consumer_data->pid_mutex);
2237
2238 DBG2("Spawning consumer control thread");
2239 ret = spawn_consumer_thread(consumer_data);
2240 if (ret < 0) {
2241 ERR("Fatal error spawning consumer control thread");
2242 goto error;
2243 }
2244
2245 end:
2246 return 0;
2247
2248 error:
2249 /* Cleanup already created sockets on error. */
2250 if (consumer_data->err_sock >= 0) {
2251 int err;
2252
2253 err = close(consumer_data->err_sock);
2254 if (err < 0) {
2255 PERROR("close consumer data error socket");
2256 }
2257 }
2258 return ret;
2259 }
2260
2261 /*
2262 * Compute health status of each consumer. If one of them is zero (bad
2263 * state), we return 0.
2264 */
2265 static int check_consumer_health(void)
2266 {
2267 int ret;
2268
2269 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2270
2271 DBG3("Health consumer check %d", ret);
2272
2273 return ret;
2274 }
2275
2276 /*
2277 * Setup necessary data for kernel tracer action.
2278 */
2279 static int init_kernel_tracer(void)
2280 {
2281 int ret;
2282
2283 /* Modprobe lttng kernel modules */
2284 ret = modprobe_lttng_control();
2285 if (ret < 0) {
2286 goto error;
2287 }
2288
2289 /* Open debugfs lttng */
2290 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2291 if (kernel_tracer_fd < 0) {
2292 DBG("Failed to open %s", module_proc_lttng);
2293 ret = -1;
2294 goto error_open;
2295 }
2296
2297 /* Validate kernel version */
2298 ret = kernel_validate_version(kernel_tracer_fd);
2299 if (ret < 0) {
2300 goto error_version;
2301 }
2302
2303 ret = modprobe_lttng_data();
2304 if (ret < 0) {
2305 goto error_modules;
2306 }
2307
2308 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2309 return 0;
2310
2311 error_version:
2312 modprobe_remove_lttng_control();
2313 ret = close(kernel_tracer_fd);
2314 if (ret) {
2315 PERROR("close");
2316 }
2317 kernel_tracer_fd = -1;
2318 return LTTNG_ERR_KERN_VERSION;
2319
2320 error_modules:
2321 ret = close(kernel_tracer_fd);
2322 if (ret) {
2323 PERROR("close");
2324 }
2325
2326 error_open:
2327 modprobe_remove_lttng_control();
2328
2329 error:
2330 WARN("No kernel tracer available");
2331 kernel_tracer_fd = -1;
2332 if (!is_root) {
2333 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2334 } else {
2335 return LTTNG_ERR_KERN_NA;
2336 }
2337 }
2338
2339
2340 /*
2341 * Copy consumer output from the tracing session to the domain session. The
2342 * function also applies the right modification on a per domain basis for the
2343 * trace files destination directory.
2344 *
2345 * Should *NOT* be called with RCU read-side lock held.
2346 */
2347 static int copy_session_consumer(int domain, struct ltt_session *session)
2348 {
2349 int ret;
2350 const char *dir_name;
2351 struct consumer_output *consumer;
2352
2353 assert(session);
2354 assert(session->consumer);
2355
2356 switch (domain) {
2357 case LTTNG_DOMAIN_KERNEL:
2358 DBG3("Copying tracing session consumer output in kernel session");
2359 /*
2360 * XXX: We should audit the session creation and what this function
2361 * does "extra" in order to avoid a destroy since this function is used
2362 * in the domain session creation (kernel and ust) only. Same for UST
2363 * domain.
2364 */
2365 if (session->kernel_session->consumer) {
2366 consumer_destroy_output(session->kernel_session->consumer);
2367 }
2368 session->kernel_session->consumer =
2369 consumer_copy_output(session->consumer);
2370 /* Ease our life a bit for the next part */
2371 consumer = session->kernel_session->consumer;
2372 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2373 break;
2374 case LTTNG_DOMAIN_UST:
2375 DBG3("Copying tracing session consumer output in UST session");
2376 if (session->ust_session->consumer) {
2377 consumer_destroy_output(session->ust_session->consumer);
2378 }
2379 session->ust_session->consumer =
2380 consumer_copy_output(session->consumer);
2381 /* Ease our life a bit for the next part */
2382 consumer = session->ust_session->consumer;
2383 dir_name = DEFAULT_UST_TRACE_DIR;
2384 break;
2385 default:
2386 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2387 goto error;
2388 }
2389
2390 /* Append correct directory to subdir */
2391 strncat(consumer->subdir, dir_name,
2392 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2393 DBG3("Copy session consumer subdir %s", consumer->subdir);
2394
2395 ret = LTTNG_OK;
2396
2397 error:
2398 return ret;
2399 }
2400
2401 /*
2402 * Create an UST session and add it to the session ust list.
2403 *
2404 * Should *NOT* be called with RCU read-side lock held.
2405 */
2406 static int create_ust_session(struct ltt_session *session,
2407 struct lttng_domain *domain)
2408 {
2409 int ret;
2410 struct ltt_ust_session *lus = NULL;
2411
2412 assert(session);
2413 assert(domain);
2414 assert(session->consumer);
2415
2416 switch (domain->type) {
2417 case LTTNG_DOMAIN_UST:
2418 break;
2419 default:
2420 ERR("Unknown UST domain on create session %d", domain->type);
2421 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2422 goto error;
2423 }
2424
2425 DBG("Creating UST session");
2426
2427 lus = trace_ust_create_session(session->id);
2428 if (lus == NULL) {
2429 ret = LTTNG_ERR_UST_SESS_FAIL;
2430 goto error;
2431 }
2432
2433 lus->uid = session->uid;
2434 lus->gid = session->gid;
2435 lus->output_traces = session->output_traces;
2436 lus->snapshot_mode = session->snapshot_mode;
2437 session->ust_session = lus;
2438
2439 /* Copy session output to the newly created UST session */
2440 ret = copy_session_consumer(domain->type, session);
2441 if (ret != LTTNG_OK) {
2442 goto error;
2443 }
2444
2445 return LTTNG_OK;
2446
2447 error:
2448 free(lus);
2449 session->ust_session = NULL;
2450 return ret;
2451 }
2452
2453 /*
2454 * Create a kernel tracer session then create the default channel.
2455 */
2456 static int create_kernel_session(struct ltt_session *session)
2457 {
2458 int ret;
2459
2460 DBG("Creating kernel session");
2461
2462 ret = kernel_create_session(session, kernel_tracer_fd);
2463 if (ret < 0) {
2464 ret = LTTNG_ERR_KERN_SESS_FAIL;
2465 goto error;
2466 }
2467
2468 /* Code flow safety */
2469 assert(session->kernel_session);
2470
2471 /* Copy session output to the newly created Kernel session */
2472 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2473 if (ret != LTTNG_OK) {
2474 goto error;
2475 }
2476
2477 /* Create directory(ies) on local filesystem. */
2478 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2479 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2480 ret = run_as_mkdir_recursive(
2481 session->kernel_session->consumer->dst.trace_path,
2482 S_IRWXU | S_IRWXG, session->uid, session->gid);
2483 if (ret < 0) {
2484 if (ret != -EEXIST) {
2485 ERR("Trace directory creation error");
2486 goto error;
2487 }
2488 }
2489 }
2490
2491 session->kernel_session->uid = session->uid;
2492 session->kernel_session->gid = session->gid;
2493 session->kernel_session->output_traces = session->output_traces;
2494 session->kernel_session->snapshot_mode = session->snapshot_mode;
2495
2496 return LTTNG_OK;
2497
2498 error:
2499 trace_kernel_destroy_session(session->kernel_session);
2500 session->kernel_session = NULL;
2501 return ret;
2502 }
2503
2504 /*
2505 * Count number of session permitted by uid/gid.
2506 */
2507 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2508 {
2509 unsigned int i = 0;
2510 struct ltt_session *session;
2511
2512 DBG("Counting number of available session for UID %d GID %d",
2513 uid, gid);
2514 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2515 /*
2516 * Only list the sessions the user can control.
2517 */
2518 if (!session_access_ok(session, uid, gid)) {
2519 continue;
2520 }
2521 i++;
2522 }
2523 return i;
2524 }
2525
2526 /*
2527 * Process the command requested by the lttng client within the command
2528 * context structure. This function make sure that the return structure (llm)
2529 * is set and ready for transmission before returning.
2530 *
2531 * Return any error encountered or 0 for success.
2532 *
2533 * "sock" is only used for special-case var. len data.
2534 *
2535 * Should *NOT* be called with RCU read-side lock held.
2536 */
2537 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2538 int *sock_error)
2539 {
2540 int ret = LTTNG_OK;
2541 int need_tracing_session = 1;
2542 int need_domain;
2543
2544 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2545
2546 *sock_error = 0;
2547
2548 switch (cmd_ctx->lsm->cmd_type) {
2549 case LTTNG_CREATE_SESSION:
2550 case LTTNG_CREATE_SESSION_SNAPSHOT:
2551 case LTTNG_DESTROY_SESSION:
2552 case LTTNG_LIST_SESSIONS:
2553 case LTTNG_LIST_DOMAINS:
2554 case LTTNG_START_TRACE:
2555 case LTTNG_STOP_TRACE:
2556 case LTTNG_DATA_PENDING:
2557 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2558 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2559 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2560 case LTTNG_SNAPSHOT_RECORD:
2561 need_domain = 0;
2562 break;
2563 default:
2564 need_domain = 1;
2565 }
2566
2567 if (opt_no_kernel && need_domain
2568 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2569 if (!is_root) {
2570 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2571 } else {
2572 ret = LTTNG_ERR_KERN_NA;
2573 }
2574 goto error;
2575 }
2576
2577 /* Deny register consumer if we already have a spawned consumer. */
2578 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2579 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2580 if (kconsumer_data.pid > 0) {
2581 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2582 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2583 goto error;
2584 }
2585 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2586 }
2587
2588 /*
2589 * Check for command that don't needs to allocate a returned payload. We do
2590 * this here so we don't have to make the call for no payload at each
2591 * command.
2592 */
2593 switch(cmd_ctx->lsm->cmd_type) {
2594 case LTTNG_LIST_SESSIONS:
2595 case LTTNG_LIST_TRACEPOINTS:
2596 case LTTNG_LIST_TRACEPOINT_FIELDS:
2597 case LTTNG_LIST_DOMAINS:
2598 case LTTNG_LIST_CHANNELS:
2599 case LTTNG_LIST_EVENTS:
2600 break;
2601 default:
2602 /* Setup lttng message with no payload */
2603 ret = setup_lttng_msg(cmd_ctx, 0);
2604 if (ret < 0) {
2605 /* This label does not try to unlock the session */
2606 goto init_setup_error;
2607 }
2608 }
2609
2610 /* Commands that DO NOT need a session. */
2611 switch (cmd_ctx->lsm->cmd_type) {
2612 case LTTNG_CREATE_SESSION:
2613 case LTTNG_CREATE_SESSION_SNAPSHOT:
2614 case LTTNG_CALIBRATE:
2615 case LTTNG_LIST_SESSIONS:
2616 case LTTNG_LIST_TRACEPOINTS:
2617 case LTTNG_LIST_TRACEPOINT_FIELDS:
2618 need_tracing_session = 0;
2619 break;
2620 default:
2621 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2622 /*
2623 * We keep the session list lock across _all_ commands
2624 * for now, because the per-session lock does not
2625 * handle teardown properly.
2626 */
2627 session_lock_list();
2628 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2629 if (cmd_ctx->session == NULL) {
2630 ret = LTTNG_ERR_SESS_NOT_FOUND;
2631 goto error;
2632 } else {
2633 /* Acquire lock for the session */
2634 session_lock(cmd_ctx->session);
2635 }
2636 break;
2637 }
2638
2639 if (!need_domain) {
2640 goto skip_domain;
2641 }
2642
2643 /*
2644 * Check domain type for specific "pre-action".
2645 */
2646 switch (cmd_ctx->lsm->domain.type) {
2647 case LTTNG_DOMAIN_KERNEL:
2648 if (!is_root) {
2649 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2650 goto error;
2651 }
2652
2653 /* Kernel tracer check */
2654 if (kernel_tracer_fd == -1) {
2655 /* Basically, load kernel tracer modules */
2656 ret = init_kernel_tracer();
2657 if (ret != 0) {
2658 goto error;
2659 }
2660 }
2661
2662 /* Consumer is in an ERROR state. Report back to client */
2663 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2664 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2665 goto error;
2666 }
2667
2668 /* Need a session for kernel command */
2669 if (need_tracing_session) {
2670 if (cmd_ctx->session->kernel_session == NULL) {
2671 ret = create_kernel_session(cmd_ctx->session);
2672 if (ret < 0) {
2673 ret = LTTNG_ERR_KERN_SESS_FAIL;
2674 goto error;
2675 }
2676 }
2677
2678 /* Start the kernel consumer daemon */
2679 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2680 if (kconsumer_data.pid == 0 &&
2681 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2682 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2683 ret = start_consumerd(&kconsumer_data);
2684 if (ret < 0) {
2685 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2686 goto error;
2687 }
2688 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2689 } else {
2690 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2691 }
2692
2693 /*
2694 * The consumer was just spawned so we need to add the socket to
2695 * the consumer output of the session if exist.
2696 */
2697 ret = consumer_create_socket(&kconsumer_data,
2698 cmd_ctx->session->kernel_session->consumer);
2699 if (ret < 0) {
2700 goto error;
2701 }
2702 }
2703
2704 break;
2705 case LTTNG_DOMAIN_UST:
2706 {
2707 /* Consumer is in an ERROR state. Report back to client */
2708 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2709 ret = LTTNG_ERR_NO_USTCONSUMERD;
2710 goto error;
2711 }
2712
2713 if (need_tracing_session) {
2714 /* Create UST session if none exist. */
2715 if (cmd_ctx->session->ust_session == NULL) {
2716 ret = create_ust_session(cmd_ctx->session,
2717 &cmd_ctx->lsm->domain);
2718 if (ret != LTTNG_OK) {
2719 goto error;
2720 }
2721 }
2722
2723 /* Start the UST consumer daemons */
2724 /* 64-bit */
2725 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2726 if (consumerd64_bin[0] != '\0' &&
2727 ustconsumer64_data.pid == 0 &&
2728 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2729 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2730 ret = start_consumerd(&ustconsumer64_data);
2731 if (ret < 0) {
2732 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2733 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2734 goto error;
2735 }
2736
2737 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2738 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2739 } else {
2740 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2741 }
2742
2743 /*
2744 * Setup socket for consumer 64 bit. No need for atomic access
2745 * since it was set above and can ONLY be set in this thread.
2746 */
2747 ret = consumer_create_socket(&ustconsumer64_data,
2748 cmd_ctx->session->ust_session->consumer);
2749 if (ret < 0) {
2750 goto error;
2751 }
2752
2753 /* 32-bit */
2754 if (consumerd32_bin[0] != '\0' &&
2755 ustconsumer32_data.pid == 0 &&
2756 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2757 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2758 ret = start_consumerd(&ustconsumer32_data);
2759 if (ret < 0) {
2760 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2761 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2762 goto error;
2763 }
2764
2765 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2766 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2767 } else {
2768 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2769 }
2770
2771 /*
2772 * Setup socket for consumer 64 bit. No need for atomic access
2773 * since it was set above and can ONLY be set in this thread.
2774 */
2775 ret = consumer_create_socket(&ustconsumer32_data,
2776 cmd_ctx->session->ust_session->consumer);
2777 if (ret < 0) {
2778 goto error;
2779 }
2780 }
2781 break;
2782 }
2783 default:
2784 break;
2785 }
2786 skip_domain:
2787
2788 /* Validate consumer daemon state when start/stop trace command */
2789 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2790 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2791 switch (cmd_ctx->lsm->domain.type) {
2792 case LTTNG_DOMAIN_UST:
2793 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2794 ret = LTTNG_ERR_NO_USTCONSUMERD;
2795 goto error;
2796 }
2797 break;
2798 case LTTNG_DOMAIN_KERNEL:
2799 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2800 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2801 goto error;
2802 }
2803 break;
2804 }
2805 }
2806
2807 /*
2808 * Check that the UID or GID match that of the tracing session.
2809 * The root user can interact with all sessions.
2810 */
2811 if (need_tracing_session) {
2812 if (!session_access_ok(cmd_ctx->session,
2813 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2814 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2815 ret = LTTNG_ERR_EPERM;
2816 goto error;
2817 }
2818 }
2819
2820 /*
2821 * Send relayd information to consumer as soon as we have a domain and a
2822 * session defined.
2823 */
2824 if (cmd_ctx->session && need_domain) {
2825 /*
2826 * Setup relayd if not done yet. If the relayd information was already
2827 * sent to the consumer, this call will gracefully return.
2828 */
2829 ret = cmd_setup_relayd(cmd_ctx->session);
2830 if (ret != LTTNG_OK) {
2831 goto error;
2832 }
2833 }
2834
2835 /* Process by command type */
2836 switch (cmd_ctx->lsm->cmd_type) {
2837 case LTTNG_ADD_CONTEXT:
2838 {
2839 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2840 cmd_ctx->lsm->u.context.channel_name,
2841 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2842 break;
2843 }
2844 case LTTNG_DISABLE_CHANNEL:
2845 {
2846 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2847 cmd_ctx->lsm->u.disable.channel_name);
2848 break;
2849 }
2850 case LTTNG_DISABLE_EVENT:
2851 {
2852 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2853 cmd_ctx->lsm->u.disable.channel_name,
2854 cmd_ctx->lsm->u.disable.name);
2855 break;
2856 }
2857 case LTTNG_DISABLE_ALL_EVENT:
2858 {
2859 DBG("Disabling all events");
2860
2861 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2862 cmd_ctx->lsm->u.disable.channel_name);
2863 break;
2864 }
2865 case LTTNG_ENABLE_CHANNEL:
2866 {
2867 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2868 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2869 break;
2870 }
2871 case LTTNG_ENABLE_EVENT:
2872 {
2873 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2874 cmd_ctx->lsm->u.enable.channel_name,
2875 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2876 break;
2877 }
2878 case LTTNG_ENABLE_ALL_EVENT:
2879 {
2880 DBG("Enabling all events");
2881
2882 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2883 cmd_ctx->lsm->u.enable.channel_name,
2884 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2885 break;
2886 }
2887 case LTTNG_LIST_TRACEPOINTS:
2888 {
2889 struct lttng_event *events;
2890 ssize_t nb_events;
2891
2892 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2893 if (nb_events < 0) {
2894 /* Return value is a negative lttng_error_code. */
2895 ret = -nb_events;
2896 goto error;
2897 }
2898
2899 /*
2900 * Setup lttng message with payload size set to the event list size in
2901 * bytes and then copy list into the llm payload.
2902 */
2903 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2904 if (ret < 0) {
2905 free(events);
2906 goto setup_error;
2907 }
2908
2909 /* Copy event list into message payload */
2910 memcpy(cmd_ctx->llm->payload, events,
2911 sizeof(struct lttng_event) * nb_events);
2912
2913 free(events);
2914
2915 ret = LTTNG_OK;
2916 break;
2917 }
2918 case LTTNG_LIST_TRACEPOINT_FIELDS:
2919 {
2920 struct lttng_event_field *fields;
2921 ssize_t nb_fields;
2922
2923 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2924 &fields);
2925 if (nb_fields < 0) {
2926 /* Return value is a negative lttng_error_code. */
2927 ret = -nb_fields;
2928 goto error;
2929 }
2930
2931 /*
2932 * Setup lttng message with payload size set to the event list size in
2933 * bytes and then copy list into the llm payload.
2934 */
2935 ret = setup_lttng_msg(cmd_ctx,
2936 sizeof(struct lttng_event_field) * nb_fields);
2937 if (ret < 0) {
2938 free(fields);
2939 goto setup_error;
2940 }
2941
2942 /* Copy event list into message payload */
2943 memcpy(cmd_ctx->llm->payload, fields,
2944 sizeof(struct lttng_event_field) * nb_fields);
2945
2946 free(fields);
2947
2948 ret = LTTNG_OK;
2949 break;
2950 }
2951 case LTTNG_SET_CONSUMER_URI:
2952 {
2953 size_t nb_uri, len;
2954 struct lttng_uri *uris;
2955
2956 nb_uri = cmd_ctx->lsm->u.uri.size;
2957 len = nb_uri * sizeof(struct lttng_uri);
2958
2959 if (nb_uri == 0) {
2960 ret = LTTNG_ERR_INVALID;
2961 goto error;
2962 }
2963
2964 uris = zmalloc(len);
2965 if (uris == NULL) {
2966 ret = LTTNG_ERR_FATAL;
2967 goto error;
2968 }
2969
2970 /* Receive variable len data */
2971 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2972 ret = lttcomm_recv_unix_sock(sock, uris, len);
2973 if (ret <= 0) {
2974 DBG("No URIs received from client... continuing");
2975 *sock_error = 1;
2976 ret = LTTNG_ERR_SESSION_FAIL;
2977 free(uris);
2978 goto error;
2979 }
2980
2981 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2982 nb_uri, uris);
2983 if (ret != LTTNG_OK) {
2984 free(uris);
2985 goto error;
2986 }
2987
2988 /*
2989 * XXX: 0 means that this URI should be applied on the session. Should
2990 * be a DOMAIN enuam.
2991 */
2992 if (cmd_ctx->lsm->domain.type == 0) {
2993 /* Add the URI for the UST session if a consumer is present. */
2994 if (cmd_ctx->session->ust_session &&
2995 cmd_ctx->session->ust_session->consumer) {
2996 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2997 nb_uri, uris);
2998 } else if (cmd_ctx->session->kernel_session &&
2999 cmd_ctx->session->kernel_session->consumer) {
3000 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3001 cmd_ctx->session, nb_uri, uris);
3002 }
3003 }
3004
3005 free(uris);
3006
3007 break;
3008 }
3009 case LTTNG_START_TRACE:
3010 {
3011 ret = cmd_start_trace(cmd_ctx->session);
3012 break;
3013 }
3014 case LTTNG_STOP_TRACE:
3015 {
3016 ret = cmd_stop_trace(cmd_ctx->session);
3017 break;
3018 }
3019 case LTTNG_CREATE_SESSION:
3020 {
3021 size_t nb_uri, len;
3022 struct lttng_uri *uris = NULL;
3023
3024 nb_uri = cmd_ctx->lsm->u.uri.size;
3025 len = nb_uri * sizeof(struct lttng_uri);
3026
3027 if (nb_uri > 0) {
3028 uris = zmalloc(len);
3029 if (uris == NULL) {
3030 ret = LTTNG_ERR_FATAL;
3031 goto error;
3032 }
3033
3034 /* Receive variable len data */
3035 DBG("Waiting for %zu URIs from client ...", nb_uri);
3036 ret = lttcomm_recv_unix_sock(sock, uris, len);
3037 if (ret <= 0) {
3038 DBG("No URIs received from client... continuing");
3039 *sock_error = 1;
3040 ret = LTTNG_ERR_SESSION_FAIL;
3041 free(uris);
3042 goto error;
3043 }
3044
3045 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3046 DBG("Creating session with ONE network URI is a bad call");
3047 ret = LTTNG_ERR_SESSION_FAIL;
3048 free(uris);
3049 goto error;
3050 }
3051 }
3052
3053 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3054 &cmd_ctx->creds);
3055
3056 free(uris);
3057
3058 break;
3059 }
3060 case LTTNG_DESTROY_SESSION:
3061 {
3062 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3063
3064 /* Set session to NULL so we do not unlock it after free. */
3065 cmd_ctx->session = NULL;
3066 break;
3067 }
3068 case LTTNG_LIST_DOMAINS:
3069 {
3070 ssize_t nb_dom;
3071 struct lttng_domain *domains;
3072
3073 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3074 if (nb_dom < 0) {
3075 /* Return value is a negative lttng_error_code. */
3076 ret = -nb_dom;
3077 goto error;
3078 }
3079
3080 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3081 if (ret < 0) {
3082 free(domains);
3083 goto setup_error;
3084 }
3085
3086 /* Copy event list into message payload */
3087 memcpy(cmd_ctx->llm->payload, domains,
3088 nb_dom * sizeof(struct lttng_domain));
3089
3090 free(domains);
3091
3092 ret = LTTNG_OK;
3093 break;
3094 }
3095 case LTTNG_LIST_CHANNELS:
3096 {
3097 int nb_chan;
3098 struct lttng_channel *channels;
3099
3100 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3101 cmd_ctx->session, &channels);
3102 if (nb_chan < 0) {
3103 /* Return value is a negative lttng_error_code. */
3104 ret = -nb_chan;
3105 goto error;
3106 }
3107
3108 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3109 if (ret < 0) {
3110 free(channels);
3111 goto setup_error;
3112 }
3113
3114 /* Copy event list into message payload */
3115 memcpy(cmd_ctx->llm->payload, channels,
3116 nb_chan * sizeof(struct lttng_channel));
3117
3118 free(channels);
3119
3120 ret = LTTNG_OK;
3121 break;
3122 }
3123 case LTTNG_LIST_EVENTS:
3124 {
3125 ssize_t nb_event;
3126 struct lttng_event *events = NULL;
3127
3128 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3129 cmd_ctx->lsm->u.list.channel_name, &events);
3130 if (nb_event < 0) {
3131 /* Return value is a negative lttng_error_code. */
3132 ret = -nb_event;
3133 goto error;
3134 }
3135
3136 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3137 if (ret < 0) {
3138 free(events);
3139 goto setup_error;
3140 }
3141
3142 /* Copy event list into message payload */
3143 memcpy(cmd_ctx->llm->payload, events,
3144 nb_event * sizeof(struct lttng_event));
3145
3146 free(events);
3147
3148 ret = LTTNG_OK;
3149 break;
3150 }
3151 case LTTNG_LIST_SESSIONS:
3152 {
3153 unsigned int nr_sessions;
3154
3155 session_lock_list();
3156 nr_sessions = lttng_sessions_count(
3157 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3158 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3159
3160 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3161 if (ret < 0) {
3162 session_unlock_list();
3163 goto setup_error;
3164 }
3165
3166 /* Filled the session array */
3167 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3168 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3169 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3170
3171 session_unlock_list();
3172
3173 ret = LTTNG_OK;
3174 break;
3175 }
3176 case LTTNG_CALIBRATE:
3177 {
3178 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3179 &cmd_ctx->lsm->u.calibrate);
3180 break;
3181 }
3182 case LTTNG_REGISTER_CONSUMER:
3183 {
3184 struct consumer_data *cdata;
3185
3186 switch (cmd_ctx->lsm->domain.type) {
3187 case LTTNG_DOMAIN_KERNEL:
3188 cdata = &kconsumer_data;
3189 break;
3190 default:
3191 ret = LTTNG_ERR_UND;
3192 goto error;
3193 }
3194
3195 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3196 cmd_ctx->lsm->u.reg.path, cdata);
3197 break;
3198 }
3199 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3200 {
3201 struct lttng_filter_bytecode *bytecode;
3202
3203 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3204 ret = LTTNG_ERR_FILTER_INVAL;
3205 goto error;
3206 }
3207 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3208 ret = LTTNG_ERR_FILTER_INVAL;
3209 goto error;
3210 }
3211 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3212 if (!bytecode) {
3213 ret = LTTNG_ERR_FILTER_NOMEM;
3214 goto error;
3215 }
3216 /* Receive var. len. data */
3217 DBG("Receiving var len data from client ...");
3218 ret = lttcomm_recv_unix_sock(sock, bytecode,
3219 cmd_ctx->lsm->u.enable.bytecode_len);
3220 if (ret <= 0) {
3221 DBG("Nothing recv() from client var len data... continuing");
3222 *sock_error = 1;
3223 ret = LTTNG_ERR_FILTER_INVAL;
3224 goto error;
3225 }
3226
3227 if (bytecode->len + sizeof(*bytecode)
3228 != cmd_ctx->lsm->u.enable.bytecode_len) {
3229 free(bytecode);
3230 ret = LTTNG_ERR_FILTER_INVAL;
3231 goto error;
3232 }
3233
3234 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3235 cmd_ctx->lsm->u.enable.channel_name,
3236 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3237 break;
3238 }
3239 case LTTNG_DATA_PENDING:
3240 {
3241 ret = cmd_data_pending(cmd_ctx->session);
3242 break;
3243 }
3244 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3245 {
3246 struct lttcomm_lttng_output_id reply;
3247
3248 ret = cmd_snapshot_add_output(cmd_ctx->session,
3249 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3250 if (ret != LTTNG_OK) {
3251 goto error;
3252 }
3253
3254 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3255 if (ret < 0) {
3256 goto setup_error;
3257 }
3258
3259 /* Copy output list into message payload */
3260 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3261 ret = LTTNG_OK;
3262 break;
3263 }
3264 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3265 {
3266 ret = cmd_snapshot_del_output(cmd_ctx->session,
3267 &cmd_ctx->lsm->u.snapshot_output.output);
3268 break;
3269 }
3270 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3271 {
3272 ssize_t nb_output;
3273 struct lttng_snapshot_output *outputs = NULL;
3274
3275 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3276 if (nb_output < 0) {
3277 ret = -nb_output;
3278 goto error;
3279 }
3280
3281 ret = setup_lttng_msg(cmd_ctx,
3282 nb_output * sizeof(struct lttng_snapshot_output));
3283 if (ret < 0) {
3284 free(outputs);
3285 goto setup_error;
3286 }
3287
3288 if (outputs) {
3289 /* Copy output list into message payload */
3290 memcpy(cmd_ctx->llm->payload, outputs,
3291 nb_output * sizeof(struct lttng_snapshot_output));
3292 free(outputs);
3293 }
3294
3295 ret = LTTNG_OK;
3296 break;
3297 }
3298 case LTTNG_SNAPSHOT_RECORD:
3299 {
3300 ret = cmd_snapshot_record(cmd_ctx->session,
3301 &cmd_ctx->lsm->u.snapshot_record.output,
3302 cmd_ctx->lsm->u.snapshot_record.wait);
3303 break;
3304 }
3305 case LTTNG_CREATE_SESSION_SNAPSHOT:
3306 {
3307 size_t nb_uri, len;
3308 struct lttng_uri *uris = NULL;
3309
3310 nb_uri = cmd_ctx->lsm->u.uri.size;
3311 len = nb_uri * sizeof(struct lttng_uri);
3312
3313 if (nb_uri > 0) {
3314 uris = zmalloc(len);
3315 if (uris == NULL) {
3316 ret = LTTNG_ERR_FATAL;
3317 goto error;
3318 }
3319
3320 /* Receive variable len data */
3321 DBG("Waiting for %zu URIs from client ...", nb_uri);
3322 ret = lttcomm_recv_unix_sock(sock, uris, len);
3323 if (ret <= 0) {
3324 DBG("No URIs received from client... continuing");
3325 *sock_error = 1;
3326 ret = LTTNG_ERR_SESSION_FAIL;
3327 free(uris);
3328 goto error;
3329 }
3330
3331 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3332 DBG("Creating session with ONE network URI is a bad call");
3333 ret = LTTNG_ERR_SESSION_FAIL;
3334 free(uris);
3335 goto error;
3336 }
3337 }
3338
3339 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3340 nb_uri, &cmd_ctx->creds);
3341 free(uris);
3342 break;
3343 }
3344 default:
3345 ret = LTTNG_ERR_UND;
3346 break;
3347 }
3348
3349 error:
3350 if (cmd_ctx->llm == NULL) {
3351 DBG("Missing llm structure. Allocating one.");
3352 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3353 goto setup_error;
3354 }
3355 }
3356 /* Set return code */
3357 cmd_ctx->llm->ret_code = ret;
3358 setup_error:
3359 if (cmd_ctx->session) {
3360 session_unlock(cmd_ctx->session);
3361 }
3362 if (need_tracing_session) {
3363 session_unlock_list();
3364 }
3365 init_setup_error:
3366 return ret;
3367 }
3368
3369 /*
3370 * Thread managing health check socket.
3371 */
3372 static void *thread_manage_health(void *data)
3373 {
3374 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3375 uint32_t revents, nb_fd;
3376 struct lttng_poll_event events;
3377 struct lttcomm_health_msg msg;
3378 struct lttcomm_health_data reply;
3379
3380 DBG("[thread] Manage health check started");
3381
3382 rcu_register_thread();
3383
3384 /* We might hit an error path before this is created. */
3385 lttng_poll_init(&events);
3386
3387 /* Create unix socket */
3388 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3389 if (sock < 0) {
3390 ERR("Unable to create health check Unix socket");
3391 ret = -1;
3392 goto error;
3393 }
3394
3395 /*
3396 * Set the CLOEXEC flag. Return code is useless because either way, the
3397 * show must go on.
3398 */
3399 (void) utils_set_fd_cloexec(sock);
3400
3401 ret = lttcomm_listen_unix_sock(sock);
3402 if (ret < 0) {
3403 goto error;
3404 }
3405
3406 /*
3407 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3408 * more will be added to this poll set.
3409 */
3410 ret = sessiond_set_thread_pollset(&events, 2);
3411 if (ret < 0) {
3412 goto error;
3413 }
3414
3415 /* Add the application registration socket */
3416 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3417 if (ret < 0) {
3418 goto error;
3419 }
3420
3421 while (1) {
3422 DBG("Health check ready");
3423
3424 /* Inifinite blocking call, waiting for transmission */
3425 restart:
3426 ret = lttng_poll_wait(&events, -1);
3427 if (ret < 0) {
3428 /*
3429 * Restart interrupted system call.
3430 */
3431 if (errno == EINTR) {
3432 goto restart;
3433 }
3434 goto error;
3435 }
3436
3437 nb_fd = ret;
3438
3439 for (i = 0; i < nb_fd; i++) {
3440 /* Fetch once the poll data */
3441 revents = LTTNG_POLL_GETEV(&events, i);
3442 pollfd = LTTNG_POLL_GETFD(&events, i);
3443
3444 /* Thread quit pipe has been closed. Killing thread. */
3445 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3446 if (ret) {
3447 err = 0;
3448 goto exit;
3449 }
3450
3451 /* Event on the registration socket */
3452 if (pollfd == sock) {
3453 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3454 ERR("Health socket poll error");
3455 goto error;
3456 }
3457 }
3458 }
3459
3460 new_sock = lttcomm_accept_unix_sock(sock);
3461 if (new_sock < 0) {
3462 goto error;
3463 }
3464
3465 /*
3466 * Set the CLOEXEC flag. Return code is useless because either way, the
3467 * show must go on.
3468 */
3469 (void) utils_set_fd_cloexec(new_sock);
3470
3471 DBG("Receiving data from client for health...");
3472 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3473 if (ret <= 0) {
3474 DBG("Nothing recv() from client... continuing");
3475 ret = close(new_sock);
3476 if (ret) {
3477 PERROR("close");
3478 }
3479 new_sock = -1;
3480 continue;
3481 }
3482
3483 rcu_thread_online();
3484
3485 switch (msg.component) {
3486 case LTTNG_HEALTH_CMD:
3487 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3488 break;
3489 case LTTNG_HEALTH_APP_MANAGE:
3490 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3491 break;
3492 case LTTNG_HEALTH_APP_REG:
3493 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3494 break;
3495 case LTTNG_HEALTH_KERNEL:
3496 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3497 break;
3498 case LTTNG_HEALTH_CONSUMER:
3499 reply.ret_code = check_consumer_health();
3500 break;
3501 case LTTNG_HEALTH_HT_CLEANUP:
3502 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3503 break;
3504 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3505 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3506 break;
3507 case LTTNG_HEALTH_APP_REG_DISPATCH:
3508 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3509 break;
3510 case LTTNG_HEALTH_ALL:
3511 reply.ret_code =
3512 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3513 health_check_state(HEALTH_TYPE_APP_REG) &&
3514 health_check_state(HEALTH_TYPE_CMD) &&
3515 health_check_state(HEALTH_TYPE_KERNEL) &&
3516 check_consumer_health() &&
3517 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3518 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3519 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3520 break;
3521 default:
3522 reply.ret_code = LTTNG_ERR_UND;
3523 break;
3524 }
3525
3526 /*
3527 * Flip ret value since 0 is a success and 1 indicates a bad health for
3528 * the client where in the sessiond it is the opposite. Again, this is
3529 * just to make things easier for us poor developer which enjoy a lot
3530 * lazyness.
3531 */
3532 if (reply.ret_code == 0 || reply.ret_code == 1) {
3533 reply.ret_code = !reply.ret_code;
3534 }
3535
3536 DBG2("Health check return value %d", reply.ret_code);
3537
3538 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3539 if (ret < 0) {
3540 ERR("Failed to send health data back to client");
3541 }
3542
3543 /* End of transmission */
3544 ret = close(new_sock);
3545 if (ret) {
3546 PERROR("close");
3547 }
3548 new_sock = -1;
3549 }
3550
3551 exit:
3552 error:
3553 if (err) {
3554 ERR("Health error occurred in %s", __func__);
3555 }
3556 DBG("Health check thread dying");
3557 unlink(health_unix_sock_path);
3558 if (sock >= 0) {
3559 ret = close(sock);
3560 if (ret) {
3561 PERROR("close");
3562 }
3563 }
3564
3565 lttng_poll_clean(&events);
3566
3567 rcu_unregister_thread();
3568 return NULL;
3569 }
3570
3571 /*
3572 * This thread manage all clients request using the unix client socket for
3573 * communication.
3574 */
3575 static void *thread_manage_clients(void *data)
3576 {
3577 int sock = -1, ret, i, pollfd, err = -1;
3578 int sock_error;
3579 uint32_t revents, nb_fd;
3580 struct command_ctx *cmd_ctx = NULL;
3581 struct lttng_poll_event events;
3582
3583 DBG("[thread] Manage client started");
3584
3585 rcu_register_thread();
3586
3587 health_register(HEALTH_TYPE_CMD);
3588
3589 if (testpoint(thread_manage_clients)) {
3590 goto error_testpoint;
3591 }
3592
3593 health_code_update();
3594
3595 ret = lttcomm_listen_unix_sock(client_sock);
3596 if (ret < 0) {
3597 goto error_listen;
3598 }
3599
3600 /*
3601 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3602 * more will be added to this poll set.
3603 */
3604 ret = sessiond_set_thread_pollset(&events, 2);
3605 if (ret < 0) {
3606 goto error_create_poll;
3607 }
3608
3609 /* Add the application registration socket */
3610 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3611 if (ret < 0) {
3612 goto error;
3613 }
3614
3615 /*
3616 * Notify parent pid that we are ready to accept command for client side.
3617 */
3618 if (opt_sig_parent) {
3619 kill(ppid, SIGUSR1);
3620 }
3621
3622 if (testpoint(thread_manage_clients_before_loop)) {
3623 goto error;
3624 }
3625
3626 health_code_update();
3627
3628 while (1) {
3629 DBG("Accepting client command ...");
3630
3631 /* Inifinite blocking call, waiting for transmission */
3632 restart:
3633 health_poll_entry();
3634 ret = lttng_poll_wait(&events, -1);
3635 health_poll_exit();
3636 if (ret < 0) {
3637 /*
3638 * Restart interrupted system call.
3639 */
3640 if (errno == EINTR) {
3641 goto restart;
3642 }
3643 goto error;
3644 }
3645
3646 nb_fd = ret;
3647
3648 for (i = 0; i < nb_fd; i++) {
3649 /* Fetch once the poll data */
3650 revents = LTTNG_POLL_GETEV(&events, i);
3651 pollfd = LTTNG_POLL_GETFD(&events, i);
3652
3653 health_code_update();
3654
3655 /* Thread quit pipe has been closed. Killing thread. */
3656 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3657 if (ret) {
3658 err = 0;
3659 goto exit;
3660 }
3661
3662 /* Event on the registration socket */
3663 if (pollfd == client_sock) {
3664 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3665 ERR("Client socket poll error");
3666 goto error;
3667 }
3668 }
3669 }
3670
3671 DBG("Wait for client response");
3672
3673 health_code_update();
3674
3675 sock = lttcomm_accept_unix_sock(client_sock);
3676 if (sock < 0) {
3677 goto error;
3678 }
3679
3680 /*
3681 * Set the CLOEXEC flag. Return code is useless because either way, the
3682 * show must go on.
3683 */
3684 (void) utils_set_fd_cloexec(sock);
3685
3686 /* Set socket option for credentials retrieval */
3687 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3688 if (ret < 0) {
3689 goto error;
3690 }
3691
3692 /* Allocate context command to process the client request */
3693 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3694 if (cmd_ctx == NULL) {
3695 PERROR("zmalloc cmd_ctx");
3696 goto error;
3697 }
3698
3699 /* Allocate data buffer for reception */
3700 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3701 if (cmd_ctx->lsm == NULL) {
3702 PERROR("zmalloc cmd_ctx->lsm");
3703 goto error;
3704 }
3705
3706 cmd_ctx->llm = NULL;
3707 cmd_ctx->session = NULL;
3708
3709 health_code_update();
3710
3711 /*
3712 * Data is received from the lttng client. The struct
3713 * lttcomm_session_msg (lsm) contains the command and data request of
3714 * the client.
3715 */
3716 DBG("Receiving data from client ...");
3717 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3718 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3719 if (ret <= 0) {
3720 DBG("Nothing recv() from client... continuing");
3721 ret = close(sock);
3722 if (ret) {
3723 PERROR("close");
3724 }
3725 sock = -1;
3726 clean_command_ctx(&cmd_ctx);
3727 continue;
3728 }
3729
3730 health_code_update();
3731
3732 // TODO: Validate cmd_ctx including sanity check for
3733 // security purpose.
3734
3735 rcu_thread_online();
3736 /*
3737 * This function dispatch the work to the kernel or userspace tracer
3738 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3739 * informations for the client. The command context struct contains
3740 * everything this function may needs.
3741 */
3742 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3743 rcu_thread_offline();
3744 if (ret < 0) {
3745 ret = close(sock);
3746 if (ret) {
3747 PERROR("close");
3748 }
3749 sock = -1;
3750 /*
3751 * TODO: Inform client somehow of the fatal error. At
3752 * this point, ret < 0 means that a zmalloc failed
3753 * (ENOMEM). Error detected but still accept
3754 * command, unless a socket error has been
3755 * detected.
3756 */
3757 clean_command_ctx(&cmd_ctx);
3758 continue;
3759 }
3760
3761 health_code_update();
3762
3763 DBG("Sending response (size: %d, retcode: %s)",
3764 cmd_ctx->lttng_msg_size,
3765 lttng_strerror(-cmd_ctx->llm->ret_code));
3766 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3767 if (ret < 0) {
3768 ERR("Failed to send data back to client");
3769 }
3770
3771 /* End of transmission */
3772 ret = close(sock);
3773 if (ret) {
3774 PERROR("close");
3775 }
3776 sock = -1;
3777
3778 clean_command_ctx(&cmd_ctx);
3779
3780 health_code_update();
3781 }
3782
3783 exit:
3784 error:
3785 if (sock >= 0) {
3786 ret = close(sock);
3787 if (ret) {
3788 PERROR("close");
3789 }
3790 }
3791
3792 lttng_poll_clean(&events);
3793 clean_command_ctx(&cmd_ctx);
3794
3795 error_listen:
3796 error_create_poll:
3797 error_testpoint:
3798 unlink(client_unix_sock_path);
3799 if (client_sock >= 0) {
3800 ret = close(client_sock);
3801 if (ret) {
3802 PERROR("close");
3803 }
3804 }
3805
3806 if (err) {
3807 health_error();
3808 ERR("Health error occurred in %s", __func__);
3809 }
3810
3811 health_unregister();
3812
3813 DBG("Client thread dying");
3814
3815 rcu_unregister_thread();
3816 return NULL;
3817 }
3818
3819
3820 /*
3821 * usage function on stderr
3822 */
3823 static void usage(void)
3824 {
3825 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3826 fprintf(stderr, " -h, --help Display this usage.\n");
3827 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3828 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3829 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3830 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3831 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3832 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3833 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3834 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3835 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3836 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3837 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3838 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3839 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3840 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3841 fprintf(stderr, " -V, --version Show version number.\n");
3842 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3843 fprintf(stderr, " -q, --quiet No output at all.\n");
3844 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3845 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3846 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3847 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3848 }
3849
3850 /*
3851 * daemon argument parsing
3852 */
3853 static int parse_args(int argc, char **argv)
3854 {
3855 int c;
3856
3857 static struct option long_options[] = {
3858 { "client-sock", 1, 0, 'c' },
3859 { "apps-sock", 1, 0, 'a' },
3860 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3861 { "kconsumerd-err-sock", 1, 0, 'E' },
3862 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3863 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3864 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3865 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3866 { "consumerd32-path", 1, 0, 'u' },
3867 { "consumerd32-libdir", 1, 0, 'U' },
3868 { "consumerd64-path", 1, 0, 't' },
3869 { "consumerd64-libdir", 1, 0, 'T' },
3870 { "daemonize", 0, 0, 'd' },
3871 { "sig-parent", 0, 0, 'S' },
3872 { "help", 0, 0, 'h' },
3873 { "group", 1, 0, 'g' },
3874 { "version", 0, 0, 'V' },
3875 { "quiet", 0, 0, 'q' },
3876 { "verbose", 0, 0, 'v' },
3877 { "verbose-consumer", 0, 0, 'Z' },
3878 { "no-kernel", 0, 0, 'N' },
3879 { "pidfile", 1, 0, 'p' },
3880 { NULL, 0, 0, 0 }
3881 };
3882
3883 while (1) {
3884 int option_index = 0;
3885 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3886 long_options, &option_index);
3887 if (c == -1) {
3888 break;
3889 }
3890
3891 switch (c) {
3892 case 0:
3893 fprintf(stderr, "option %s", long_options[option_index].name);
3894 if (optarg) {
3895 fprintf(stderr, " with arg %s\n", optarg);
3896 }
3897 break;
3898 case 'c':
3899 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3900 break;
3901 case 'a':
3902 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3903 break;
3904 case 'd':
3905 opt_daemon = 1;
3906 break;
3907 case 'g':
3908 opt_tracing_group = optarg;
3909 break;
3910 case 'h':
3911 usage();
3912 exit(EXIT_FAILURE);
3913 case 'V':
3914 fprintf(stdout, "%s\n", VERSION);
3915 exit(EXIT_SUCCESS);
3916 case 'S':
3917 opt_sig_parent = 1;
3918 break;
3919 case 'E':
3920 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3921 break;
3922 case 'C':
3923 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3924 break;
3925 case 'F':
3926 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3927 break;
3928 case 'D':
3929 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3930 break;
3931 case 'H':
3932 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3933 break;
3934 case 'G':
3935 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3936 break;
3937 case 'N':
3938 opt_no_kernel = 1;
3939 break;
3940 case 'q':
3941 lttng_opt_quiet = 1;
3942 break;
3943 case 'v':
3944 /* Verbose level can increase using multiple -v */
3945 lttng_opt_verbose += 1;
3946 break;
3947 case 'Z':
3948 opt_verbose_consumer += 1;
3949 break;
3950 case 'u':
3951 consumerd32_bin= optarg;
3952 break;
3953 case 'U':
3954 consumerd32_libdir = optarg;
3955 break;
3956 case 't':
3957 consumerd64_bin = optarg;
3958 break;
3959 case 'T':
3960 consumerd64_libdir = optarg;
3961 break;
3962 case 'p':
3963 opt_pidfile = optarg;
3964 break;
3965 default:
3966 /* Unknown option or other error.
3967 * Error is printed by getopt, just return */
3968 return -1;
3969 }
3970 }
3971
3972 return 0;
3973 }
3974
3975 /*
3976 * Creates the two needed socket by the daemon.
3977 * apps_sock - The communication socket for all UST apps.
3978 * client_sock - The communication of the cli tool (lttng).
3979 */
3980 static int init_daemon_socket(void)
3981 {
3982 int ret = 0;
3983 mode_t old_umask;
3984
3985 old_umask = umask(0);
3986
3987 /* Create client tool unix socket */
3988 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3989 if (client_sock < 0) {
3990 ERR("Create unix sock failed: %s", client_unix_sock_path);
3991 ret = -1;
3992 goto end;
3993 }
3994
3995 /* Set the cloexec flag */
3996 ret = utils_set_fd_cloexec(client_sock);
3997 if (ret < 0) {
3998 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3999 "Continuing but note that the consumer daemon will have a "
4000 "reference to this socket on exec()", client_sock);
4001 }
4002
4003 /* File permission MUST be 660 */
4004 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4005 if (ret < 0) {
4006 ERR("Set file permissions failed: %s", client_unix_sock_path);
4007 PERROR("chmod");
4008 goto end;
4009 }
4010
4011 /* Create the application unix socket */
4012 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4013 if (apps_sock < 0) {
4014 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4015 ret = -1;
4016 goto end;
4017 }
4018
4019 /* Set the cloexec flag */
4020 ret = utils_set_fd_cloexec(apps_sock);
4021 if (ret < 0) {
4022 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4023 "Continuing but note that the consumer daemon will have a "
4024 "reference to this socket on exec()", apps_sock);
4025 }
4026
4027 /* File permission MUST be 666 */
4028 ret = chmod(apps_unix_sock_path,
4029 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4030 if (ret < 0) {
4031 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4032 PERROR("chmod");
4033 goto end;
4034 }
4035
4036 DBG3("Session daemon client socket %d and application socket %d created",
4037 client_sock, apps_sock);
4038
4039 end:
4040 umask(old_umask);
4041 return ret;
4042 }
4043
4044 /*
4045 * Check if the global socket is available, and if a daemon is answering at the
4046 * other side. If yes, error is returned.
4047 */
4048 static int check_existing_daemon(void)
4049 {
4050 /* Is there anybody out there ? */
4051 if (lttng_session_daemon_alive()) {
4052 return -EEXIST;
4053 }
4054
4055 return 0;
4056 }
4057
4058 /*
4059 * Set the tracing group gid onto the client socket.
4060 *
4061 * Race window between mkdir and chown is OK because we are going from more
4062 * permissive (root.root) to less permissive (root.tracing).
4063 */
4064 static int set_permissions(char *rundir)
4065 {
4066 int ret;
4067 gid_t gid;
4068
4069 ret = allowed_group();
4070 if (ret < 0) {
4071 WARN("No tracing group detected");
4072 ret = 0;
4073 goto end;
4074 }
4075
4076 gid = ret;
4077
4078 /* Set lttng run dir */
4079 ret = chown(rundir, 0, gid);
4080 if (ret < 0) {
4081 ERR("Unable to set group on %s", rundir);
4082 PERROR("chown");
4083 }
4084
4085 /* Ensure tracing group can search the run dir */
4086 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
4087 if (ret < 0) {
4088 ERR("Unable to set permissions on %s", rundir);
4089 PERROR("chmod");
4090 }
4091
4092 /* lttng client socket path */
4093 ret = chown(client_unix_sock_path, 0, gid);
4094 if (ret < 0) {
4095 ERR("Unable to set group on %s", client_unix_sock_path);
4096 PERROR("chown");
4097 }
4098
4099 /* kconsumer error socket path */
4100 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4101 if (ret < 0) {
4102 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4103 PERROR("chown");
4104 }
4105
4106 /* 64-bit ustconsumer error socket path */
4107 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4108 if (ret < 0) {
4109 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4110 PERROR("chown");
4111 }
4112
4113 /* 32-bit ustconsumer compat32 error socket path */
4114 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4115 if (ret < 0) {
4116 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4117 PERROR("chown");
4118 }
4119
4120 DBG("All permissions are set");
4121
4122 end:
4123 return ret;
4124 }
4125
4126 /*
4127 * Create the lttng run directory needed for all global sockets and pipe.
4128 */
4129 static int create_lttng_rundir(const char *rundir)
4130 {
4131 int ret;
4132
4133 DBG3("Creating LTTng run directory: %s", rundir);
4134
4135 ret = mkdir(rundir, S_IRWXU);
4136 if (ret < 0) {
4137 if (errno != EEXIST) {
4138 ERR("Unable to create %s", rundir);
4139 goto error;
4140 } else {
4141 ret = 0;
4142 }
4143 }
4144
4145 error:
4146 return ret;
4147 }
4148
4149 /*
4150 * Setup sockets and directory needed by the kconsumerd communication with the
4151 * session daemon.
4152 */
4153 static int set_consumer_sockets(struct consumer_data *consumer_data,
4154 const char *rundir)
4155 {
4156 int ret;
4157 char path[PATH_MAX];
4158
4159 switch (consumer_data->type) {
4160 case LTTNG_CONSUMER_KERNEL:
4161 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4162 break;
4163 case LTTNG_CONSUMER64_UST:
4164 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4165 break;
4166 case LTTNG_CONSUMER32_UST:
4167 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4168 break;
4169 default:
4170 ERR("Consumer type unknown");
4171 ret = -EINVAL;
4172 goto error;
4173 }
4174
4175 DBG2("Creating consumer directory: %s", path);
4176
4177 ret = mkdir(path, S_IRWXU);
4178 if (ret < 0) {
4179 if (errno != EEXIST) {
4180 PERROR("mkdir");
4181 ERR("Failed to create %s", path);
4182 goto error;
4183 }
4184 ret = -1;
4185 }
4186
4187 /* Create the kconsumerd error unix socket */
4188 consumer_data->err_sock =
4189 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4190 if (consumer_data->err_sock < 0) {
4191 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4192 ret = -1;
4193 goto error;
4194 }
4195
4196 /*
4197 * Set the CLOEXEC flag. Return code is useless because either way, the
4198 * show must go on.
4199 */
4200 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4201 if (ret < 0) {
4202 PERROR("utils_set_fd_cloexec");
4203 /* continue anyway */
4204 }
4205
4206 /* File permission MUST be 660 */
4207 ret = chmod(consumer_data->err_unix_sock_path,
4208 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4209 if (ret < 0) {
4210 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4211 PERROR("chmod");
4212 goto error;
4213 }
4214
4215 error:
4216 return ret;
4217 }
4218
4219 /*
4220 * Signal handler for the daemon
4221 *
4222 * Simply stop all worker threads, leaving main() return gracefully after
4223 * joining all threads and calling cleanup().
4224 */
4225 static void sighandler(int sig)
4226 {
4227 switch (sig) {
4228 case SIGPIPE:
4229 DBG("SIGPIPE caught");
4230 return;
4231 case SIGINT:
4232 DBG("SIGINT caught");
4233 stop_threads();
4234 break;
4235 case SIGTERM:
4236 DBG("SIGTERM caught");
4237 stop_threads();
4238 break;
4239 default:
4240 break;
4241 }
4242 }
4243
4244 /*
4245 * Setup signal handler for :
4246 * SIGINT, SIGTERM, SIGPIPE
4247 */
4248 static int set_signal_handler(void)
4249 {
4250 int ret = 0;
4251 struct sigaction sa;
4252 sigset_t sigset;
4253
4254 if ((ret = sigemptyset(&sigset)) < 0) {
4255 PERROR("sigemptyset");
4256 return ret;
4257 }
4258
4259 sa.sa_handler = sighandler;
4260 sa.sa_mask = sigset;
4261 sa.sa_flags = 0;
4262 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4263 PERROR("sigaction");
4264 return ret;
4265 }
4266
4267 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4268 PERROR("sigaction");
4269 return ret;
4270 }
4271
4272 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4273 PERROR("sigaction");
4274 return ret;
4275 }
4276
4277 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4278
4279 return ret;
4280 }
4281
4282 /*
4283 * Set open files limit to unlimited. This daemon can open a large number of
4284 * file descriptors in order to consumer multiple kernel traces.
4285 */
4286 static void set_ulimit(void)
4287 {
4288 int ret;
4289 struct rlimit lim;
4290
4291 /* The kernel does not allowed an infinite limit for open files */
4292 lim.rlim_cur = 65535;
4293 lim.rlim_max = 65535;
4294
4295 ret = setrlimit(RLIMIT_NOFILE, &lim);
4296 if (ret < 0) {
4297 PERROR("failed to set open files limit");
4298 }
4299 }
4300
4301 /*
4302 * Write pidfile using the rundir and opt_pidfile.
4303 */
4304 static void write_pidfile(void)
4305 {
4306 int ret;
4307 char pidfile_path[PATH_MAX];
4308
4309 assert(rundir);
4310
4311 if (opt_pidfile) {
4312 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4313 } else {
4314 /* Build pidfile path from rundir and opt_pidfile. */
4315 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4316 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4317 if (ret < 0) {
4318 PERROR("snprintf pidfile path");
4319 goto error;
4320 }
4321 }
4322
4323 /*
4324 * Create pid file in rundir. Return value is of no importance. The
4325 * execution will continue even though we are not able to write the file.
4326 */
4327 (void) utils_create_pid_file(getpid(), pidfile_path);
4328
4329 error:
4330 return;
4331 }
4332
4333 /*
4334 * main
4335 */
4336 int main(int argc, char **argv)
4337 {
4338 int ret = 0;
4339 void *status;
4340 const char *home_path, *env_app_timeout;
4341
4342 init_kernel_workarounds();
4343
4344 rcu_register_thread();
4345
4346 setup_consumerd_path();
4347
4348 page_size = sysconf(_SC_PAGESIZE);
4349 if (page_size < 0) {
4350 PERROR("sysconf _SC_PAGESIZE");
4351 page_size = LONG_MAX;
4352 WARN("Fallback page size to %ld", page_size);
4353 }
4354
4355 /* Parse arguments */
4356 progname = argv[0];
4357 if ((ret = parse_args(argc, argv)) < 0) {
4358 goto error;
4359 }
4360
4361 /* Daemonize */
4362 if (opt_daemon) {
4363 int i;
4364
4365 /*
4366 * fork
4367 * child: setsid, close FD 0, 1, 2, chdir /
4368 * parent: exit (if fork is successful)
4369 */
4370 ret = daemon(0, 0);
4371 if (ret < 0) {
4372 PERROR("daemon");
4373 goto error;
4374 }
4375 /*
4376 * We are in the child. Make sure all other file
4377 * descriptors are closed, in case we are called with
4378 * more opened file descriptors than the standard ones.
4379 */
4380 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4381 (void) close(i);
4382 }
4383 }
4384
4385 /* Create thread quit pipe */
4386 if ((ret = init_thread_quit_pipe()) < 0) {
4387 goto error;
4388 }
4389
4390 /* Check if daemon is UID = 0 */
4391 is_root = !getuid();
4392
4393 if (is_root) {
4394 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4395
4396 /* Create global run dir with root access */
4397 ret = create_lttng_rundir(rundir);
4398 if (ret < 0) {
4399 goto error;
4400 }
4401
4402 if (strlen(apps_unix_sock_path) == 0) {
4403 snprintf(apps_unix_sock_path, PATH_MAX,
4404 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4405 }
4406
4407 if (strlen(client_unix_sock_path) == 0) {
4408 snprintf(client_unix_sock_path, PATH_MAX,
4409 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4410 }
4411
4412 /* Set global SHM for ust */
4413 if (strlen(wait_shm_path) == 0) {
4414 snprintf(wait_shm_path, PATH_MAX,
4415 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4416 }
4417
4418 if (strlen(health_unix_sock_path) == 0) {
4419 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4420 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4421 }
4422
4423 /* Setup kernel consumerd path */
4424 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4425 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4426 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4427 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4428
4429 DBG2("Kernel consumer err path: %s",
4430 kconsumer_data.err_unix_sock_path);
4431 DBG2("Kernel consumer cmd path: %s",
4432 kconsumer_data.cmd_unix_sock_path);
4433 } else {
4434 home_path = utils_get_home_dir();
4435 if (home_path == NULL) {
4436 /* TODO: Add --socket PATH option */
4437 ERR("Can't get HOME directory for sockets creation.");
4438 ret = -EPERM;
4439 goto error;
4440 }
4441
4442 /*
4443 * Create rundir from home path. This will create something like
4444 * $HOME/.lttng
4445 */
4446 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4447 if (ret < 0) {
4448 ret = -ENOMEM;
4449 goto error;
4450 }
4451
4452 ret = create_lttng_rundir(rundir);
4453 if (ret < 0) {
4454 goto error;
4455 }
4456
4457 if (strlen(apps_unix_sock_path) == 0) {
4458 snprintf(apps_unix_sock_path, PATH_MAX,
4459 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4460 }
4461
4462 /* Set the cli tool unix socket path */
4463 if (strlen(client_unix_sock_path) == 0) {
4464 snprintf(client_unix_sock_path, PATH_MAX,
4465 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4466 }
4467
4468 /* Set global SHM for ust */
4469 if (strlen(wait_shm_path) == 0) {
4470 snprintf(wait_shm_path, PATH_MAX,
4471 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4472 }
4473
4474 /* Set health check Unix path */
4475 if (strlen(health_unix_sock_path) == 0) {
4476 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4477 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4478 }
4479 }
4480
4481 /* Set consumer initial state */
4482 kernel_consumerd_state = CONSUMER_STOPPED;
4483 ust_consumerd_state = CONSUMER_STOPPED;
4484
4485 DBG("Client socket path %s", client_unix_sock_path);
4486 DBG("Application socket path %s", apps_unix_sock_path);
4487 DBG("Application wait path %s", wait_shm_path);
4488 DBG("LTTng run directory path: %s", rundir);
4489
4490 /* 32 bits consumerd path setup */
4491 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4492 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4493 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4494 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4495
4496 DBG2("UST consumer 32 bits err path: %s",
4497 ustconsumer32_data.err_unix_sock_path);
4498 DBG2("UST consumer 32 bits cmd path: %s",
4499 ustconsumer32_data.cmd_unix_sock_path);
4500
4501 /* 64 bits consumerd path setup */
4502 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4503 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4504 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4505 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4506
4507 DBG2("UST consumer 64 bits err path: %s",
4508 ustconsumer64_data.err_unix_sock_path);
4509 DBG2("UST consumer 64 bits cmd path: %s",
4510 ustconsumer64_data.cmd_unix_sock_path);
4511
4512 /*
4513 * See if daemon already exist.
4514 */
4515 if ((ret = check_existing_daemon()) < 0) {
4516 ERR("Already running daemon.\n");
4517 /*
4518 * We do not goto exit because we must not cleanup()
4519 * because a daemon is already running.
4520 */
4521 goto error;
4522 }
4523
4524 /*
4525 * Init UST app hash table. Alloc hash table before this point since
4526 * cleanup() can get called after that point.
4527 */
4528 ust_app_ht_alloc();
4529
4530 /* After this point, we can safely call cleanup() with "goto exit" */
4531
4532 /*
4533 * These actions must be executed as root. We do that *after* setting up
4534 * the sockets path because we MUST make the check for another daemon using
4535 * those paths *before* trying to set the kernel consumer sockets and init
4536 * kernel tracer.
4537 */
4538 if (is_root) {
4539 ret = set_consumer_sockets(&kconsumer_data, rundir);
4540 if (ret < 0) {
4541 goto exit;
4542 }
4543
4544 /* Setup kernel tracer */
4545 if (!opt_no_kernel) {
4546 init_kernel_tracer();
4547 }
4548
4549 /* Set ulimit for open files */
4550 set_ulimit();
4551 }
4552 /* init lttng_fd tracking must be done after set_ulimit. */
4553 lttng_fd_init();
4554
4555 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4556 if (ret < 0) {
4557 goto exit;
4558 }
4559
4560 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4561 if (ret < 0) {
4562 goto exit;
4563 }
4564
4565 if ((ret = set_signal_handler()) < 0) {
4566 goto exit;
4567 }
4568
4569 /* Setup the needed unix socket */
4570 if ((ret = init_daemon_socket()) < 0) {
4571 goto exit;
4572 }
4573
4574 /* Set credentials to socket */
4575 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4576 goto exit;
4577 }
4578
4579 /* Get parent pid if -S, --sig-parent is specified. */
4580 if (opt_sig_parent) {
4581 ppid = getppid();
4582 }
4583
4584 /* Setup the kernel pipe for waking up the kernel thread */
4585 if (is_root && !opt_no_kernel) {
4586 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4587 goto exit;
4588 }
4589 }
4590
4591 /* Setup the thread ht_cleanup communication pipe. */
4592 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4593 goto exit;
4594 }
4595
4596 /* Setup the thread apps communication pipe. */
4597 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4598 goto exit;
4599 }
4600
4601 /* Setup the thread apps notify communication pipe. */
4602 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4603 goto exit;
4604 }
4605
4606 /* Initialize global buffer per UID and PID registry. */
4607 buffer_reg_init_uid_registry();
4608 buffer_reg_init_pid_registry();
4609
4610 /* Init UST command queue. */
4611 cds_wfq_init(&ust_cmd_queue.queue);
4612
4613 /*
4614 * Get session list pointer. This pointer MUST NOT be free(). This list is
4615 * statically declared in session.c
4616 */
4617 session_list_ptr = session_get_list();
4618
4619 /* Set up max poll set size */
4620 lttng_poll_set_max_size();
4621
4622 cmd_init();
4623
4624 /* Check for the application socket timeout env variable. */
4625 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4626 if (env_app_timeout) {
4627 app_socket_timeout = atoi(env_app_timeout);
4628 } else {
4629 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4630 }
4631
4632 write_pidfile();
4633
4634 /* Create thread to manage the client socket */
4635 ret = pthread_create(&ht_cleanup_thread, NULL,
4636 thread_ht_cleanup, (void *) NULL);
4637 if (ret != 0) {
4638 PERROR("pthread_create ht_cleanup");
4639 goto exit_ht_cleanup;
4640 }
4641
4642 /* Create thread to manage the client socket */
4643 ret = pthread_create(&health_thread, NULL,
4644 thread_manage_health, (void *) NULL);
4645 if (ret != 0) {
4646 PERROR("pthread_create health");
4647 goto exit_health;
4648 }
4649
4650 /* Create thread to manage the client socket */
4651 ret = pthread_create(&client_thread, NULL,
4652 thread_manage_clients, (void *) NULL);
4653 if (ret != 0) {
4654 PERROR("pthread_create clients");
4655 goto exit_client;
4656 }
4657
4658 /* Create thread to dispatch registration */
4659 ret = pthread_create(&dispatch_thread, NULL,
4660 thread_dispatch_ust_registration, (void *) NULL);
4661 if (ret != 0) {
4662 PERROR("pthread_create dispatch");
4663 goto exit_dispatch;
4664 }
4665
4666 /* Create thread to manage application registration. */
4667 ret = pthread_create(&reg_apps_thread, NULL,
4668 thread_registration_apps, (void *) NULL);
4669 if (ret != 0) {
4670 PERROR("pthread_create registration");
4671 goto exit_reg_apps;
4672 }
4673
4674 /* Create thread to manage application socket */
4675 ret = pthread_create(&apps_thread, NULL,
4676 thread_manage_apps, (void *) NULL);
4677 if (ret != 0) {
4678 PERROR("pthread_create apps");
4679 goto exit_apps;
4680 }
4681
4682 /* Create thread to manage application notify socket */
4683 ret = pthread_create(&apps_notify_thread, NULL,
4684 ust_thread_manage_notify, (void *) NULL);
4685 if (ret != 0) {
4686 PERROR("pthread_create apps");
4687 goto exit_apps;
4688 }
4689
4690 /* Don't start this thread if kernel tracing is not requested nor root */
4691 if (is_root && !opt_no_kernel) {
4692 /* Create kernel thread to manage kernel event */
4693 ret = pthread_create(&kernel_thread, NULL,
4694 thread_manage_kernel, (void *) NULL);
4695 if (ret != 0) {
4696 PERROR("pthread_create kernel");
4697 goto exit_kernel;
4698 }
4699
4700 ret = pthread_join(kernel_thread, &status);
4701 if (ret != 0) {
4702 PERROR("pthread_join");
4703 goto error; /* join error, exit without cleanup */
4704 }
4705 }
4706
4707 exit_kernel:
4708 ret = pthread_join(apps_thread, &status);
4709 if (ret != 0) {
4710 PERROR("pthread_join");
4711 goto error; /* join error, exit without cleanup */
4712 }
4713
4714 exit_apps:
4715 ret = pthread_join(reg_apps_thread, &status);
4716 if (ret != 0) {
4717 PERROR("pthread_join");
4718 goto error; /* join error, exit without cleanup */
4719 }
4720
4721 exit_reg_apps:
4722 ret = pthread_join(dispatch_thread, &status);
4723 if (ret != 0) {
4724 PERROR("pthread_join");
4725 goto error; /* join error, exit without cleanup */
4726 }
4727
4728 exit_dispatch:
4729 ret = pthread_join(client_thread, &status);
4730 if (ret != 0) {
4731 PERROR("pthread_join");
4732 goto error; /* join error, exit without cleanup */
4733 }
4734
4735 ret = join_consumer_thread(&kconsumer_data);
4736 if (ret != 0) {
4737 PERROR("join_consumer");
4738 goto error; /* join error, exit without cleanup */
4739 }
4740
4741 ret = join_consumer_thread(&ustconsumer32_data);
4742 if (ret != 0) {
4743 PERROR("join_consumer ust32");
4744 goto error; /* join error, exit without cleanup */
4745 }
4746
4747 ret = join_consumer_thread(&ustconsumer64_data);
4748 if (ret != 0) {
4749 PERROR("join_consumer ust64");
4750 goto error; /* join error, exit without cleanup */
4751 }
4752
4753 exit_client:
4754 ret = pthread_join(health_thread, &status);
4755 if (ret != 0) {
4756 PERROR("pthread_join health thread");
4757 goto error; /* join error, exit without cleanup */
4758 }
4759
4760 exit_health:
4761 ret = pthread_join(ht_cleanup_thread, &status);
4762 if (ret != 0) {
4763 PERROR("pthread_join ht cleanup thread");
4764 goto error; /* join error, exit without cleanup */
4765 }
4766 exit_ht_cleanup:
4767 exit:
4768 /*
4769 * cleanup() is called when no other thread is running.
4770 */
4771 rcu_thread_online();
4772 cleanup();
4773 rcu_thread_offline();
4774 rcu_unregister_thread();
4775 if (!ret) {
4776 exit(EXIT_SUCCESS);
4777 }
4778 error:
4779 exit(EXIT_FAILURE);
4780 }
This page took 0.127925 seconds and 4 git commands to generate.