Fix: add asserts and remove unused variable
[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-sessiond.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66 #include "jul-thread.h"
67
68 #define CONSUMERD_FILE "lttng-consumerd"
69
70 const char *progname;
71 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
72 static const char *opt_pidfile;
73 static int opt_sig_parent;
74 static int opt_verbose_consumer;
75 static int opt_daemon;
76 static int opt_no_kernel;
77 static int is_root; /* Set to 1 if the daemon is running as root */
78 static pid_t ppid; /* Parent PID for --sig-parent option */
79 static char *rundir;
80
81 /*
82 * Consumer daemon specific control data. Every value not initialized here is
83 * set to 0 by the static definition.
84 */
85 static struct consumer_data kconsumer_data = {
86 .type = LTTNG_CONSUMER_KERNEL,
87 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
88 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
89 .err_sock = -1,
90 .cmd_sock = -1,
91 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
92 .lock = PTHREAD_MUTEX_INITIALIZER,
93 .cond = PTHREAD_COND_INITIALIZER,
94 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
95 };
96 static struct consumer_data ustconsumer64_data = {
97 .type = LTTNG_CONSUMER64_UST,
98 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
99 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
100 .err_sock = -1,
101 .cmd_sock = -1,
102 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
103 .lock = PTHREAD_MUTEX_INITIALIZER,
104 .cond = PTHREAD_COND_INITIALIZER,
105 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
106 };
107 static struct consumer_data ustconsumer32_data = {
108 .type = LTTNG_CONSUMER32_UST,
109 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
110 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
111 .err_sock = -1,
112 .cmd_sock = -1,
113 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
114 .lock = PTHREAD_MUTEX_INITIALIZER,
115 .cond = PTHREAD_COND_INITIALIZER,
116 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
117 };
118
119 /* Shared between threads */
120 static int dispatch_thread_exit;
121
122 /* Global application Unix socket path */
123 static char apps_unix_sock_path[PATH_MAX];
124 /* Global client Unix socket path */
125 static char client_unix_sock_path[PATH_MAX];
126 /* global wait shm path for UST */
127 static char wait_shm_path[PATH_MAX];
128 /* Global health check unix path */
129 static char health_unix_sock_path[PATH_MAX];
130
131 /* Sockets and FDs */
132 static int client_sock = -1;
133 static int apps_sock = -1;
134 int kernel_tracer_fd = -1;
135 static int kernel_poll_pipe[2] = { -1, -1 };
136
137 /*
138 * Quit pipe for all threads. This permits a single cancellation point
139 * for all threads when receiving an event on the pipe.
140 */
141 static int thread_quit_pipe[2] = { -1, -1 };
142
143 /*
144 * This pipe is used to inform the thread managing application communication
145 * that a command is queued and ready to be processed.
146 */
147 static int apps_cmd_pipe[2] = { -1, -1 };
148
149 int apps_cmd_notify_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread;
153 static pthread_t apps_notify_thread;
154 static pthread_t reg_apps_thread;
155 static pthread_t client_thread;
156 static pthread_t kernel_thread;
157 static pthread_t dispatch_thread;
158 static pthread_t health_thread;
159 static pthread_t ht_cleanup_thread;
160 static pthread_t jul_reg_thread;
161
162 /*
163 * UST registration command queue. This queue is tied with a futex and uses a N
164 * wakers / 1 waiter implemented and detailed in futex.c/.h
165 *
166 * The thread_manage_apps and thread_dispatch_ust_registration interact with
167 * this queue and the wait/wake scheme.
168 */
169 static struct ust_cmd_queue ust_cmd_queue;
170
171 /*
172 * Pointer initialized before thread creation.
173 *
174 * This points to the tracing session list containing the session count and a
175 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
176 * MUST NOT be taken if you call a public function in session.c.
177 *
178 * The lock is nested inside the structure: session_list_ptr->lock. Please use
179 * session_lock_list and session_unlock_list for lock acquisition.
180 */
181 static struct ltt_session_list *session_list_ptr;
182
183 int ust_consumerd64_fd = -1;
184 int ust_consumerd32_fd = -1;
185
186 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
187 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
188 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
189 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
190
191 static const char *module_proc_lttng = "/proc/lttng";
192
193 /*
194 * Consumer daemon state which is changed when spawning it, killing it or in
195 * case of a fatal error.
196 */
197 enum consumerd_state {
198 CONSUMER_STARTED = 1,
199 CONSUMER_STOPPED = 2,
200 CONSUMER_ERROR = 3,
201 };
202
203 /*
204 * This consumer daemon state is used to validate if a client command will be
205 * able to reach the consumer. If not, the client is informed. For instance,
206 * doing a "lttng start" when the consumer state is set to ERROR will return an
207 * error to the client.
208 *
209 * The following example shows a possible race condition of this scheme:
210 *
211 * consumer thread error happens
212 * client cmd arrives
213 * client cmd checks state -> still OK
214 * consumer thread exit, sets error
215 * client cmd try to talk to consumer
216 * ...
217 *
218 * However, since the consumer is a different daemon, we have no way of making
219 * sure the command will reach it safely even with this state flag. This is why
220 * we consider that up to the state validation during command processing, the
221 * command is safe. After that, we can not guarantee the correctness of the
222 * client request vis-a-vis the consumer.
223 */
224 static enum consumerd_state ust_consumerd_state;
225 static enum consumerd_state kernel_consumerd_state;
226
227 /*
228 * Socket timeout for receiving and sending in seconds.
229 */
230 static int app_socket_timeout;
231
232 /* Set in main() with the current page size. */
233 long page_size;
234
235 /* Application health monitoring */
236 struct health_app *health_sessiond;
237
238 /* JUL TCP port for registration. Used by the JUL thread. */
239 unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
240
241 static
242 void setup_consumerd_path(void)
243 {
244 const char *bin, *libdir;
245
246 /*
247 * Allow INSTALL_BIN_PATH to be used as a target path for the
248 * native architecture size consumer if CONFIG_CONSUMER*_PATH
249 * has not been defined.
250 */
251 #if (CAA_BITS_PER_LONG == 32)
252 if (!consumerd32_bin[0]) {
253 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
254 }
255 if (!consumerd32_libdir[0]) {
256 consumerd32_libdir = INSTALL_LIB_PATH;
257 }
258 #elif (CAA_BITS_PER_LONG == 64)
259 if (!consumerd64_bin[0]) {
260 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
261 }
262 if (!consumerd64_libdir[0]) {
263 consumerd64_libdir = INSTALL_LIB_PATH;
264 }
265 #else
266 #error "Unknown bitness"
267 #endif
268
269 /*
270 * runtime env. var. overrides the build default.
271 */
272 bin = getenv("LTTNG_CONSUMERD32_BIN");
273 if (bin) {
274 consumerd32_bin = bin;
275 }
276 bin = getenv("LTTNG_CONSUMERD64_BIN");
277 if (bin) {
278 consumerd64_bin = bin;
279 }
280 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
281 if (libdir) {
282 consumerd32_libdir = libdir;
283 }
284 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
285 if (libdir) {
286 consumerd64_libdir = libdir;
287 }
288 }
289
290 /*
291 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
292 */
293 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
294 {
295 int ret;
296
297 assert(events);
298
299 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
300 if (ret < 0) {
301 goto error;
302 }
303
304 /* Add quit pipe */
305 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
306 if (ret < 0) {
307 goto error;
308 }
309
310 return 0;
311
312 error:
313 return ret;
314 }
315
316 /*
317 * Check if the thread quit pipe was triggered.
318 *
319 * Return 1 if it was triggered else 0;
320 */
321 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
322 {
323 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
324 return 1;
325 }
326
327 return 0;
328 }
329
330 /*
331 * Init thread quit pipe.
332 *
333 * Return -1 on error or 0 if all pipes are created.
334 */
335 static int init_thread_quit_pipe(void)
336 {
337 int ret, i;
338
339 ret = pipe(thread_quit_pipe);
340 if (ret < 0) {
341 PERROR("thread quit pipe");
342 goto error;
343 }
344
345 for (i = 0; i < 2; i++) {
346 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
347 if (ret < 0) {
348 PERROR("fcntl");
349 goto error;
350 }
351 }
352
353 error:
354 return ret;
355 }
356
357 /*
358 * Stop all threads by closing the thread quit pipe.
359 */
360 static void stop_threads(void)
361 {
362 int ret;
363
364 /* Stopping all threads */
365 DBG("Terminating all threads");
366 ret = notify_thread_pipe(thread_quit_pipe[1]);
367 if (ret < 0) {
368 ERR("write error on thread quit pipe");
369 }
370
371 /* Dispatch thread */
372 CMM_STORE_SHARED(dispatch_thread_exit, 1);
373 futex_nto1_wake(&ust_cmd_queue.futex);
374 }
375
376 /*
377 * Close every consumer sockets.
378 */
379 static void close_consumer_sockets(void)
380 {
381 int ret;
382
383 if (kconsumer_data.err_sock >= 0) {
384 ret = close(kconsumer_data.err_sock);
385 if (ret < 0) {
386 PERROR("kernel consumer err_sock close");
387 }
388 }
389 if (ustconsumer32_data.err_sock >= 0) {
390 ret = close(ustconsumer32_data.err_sock);
391 if (ret < 0) {
392 PERROR("UST consumerd32 err_sock close");
393 }
394 }
395 if (ustconsumer64_data.err_sock >= 0) {
396 ret = close(ustconsumer64_data.err_sock);
397 if (ret < 0) {
398 PERROR("UST consumerd64 err_sock close");
399 }
400 }
401 if (kconsumer_data.cmd_sock >= 0) {
402 ret = close(kconsumer_data.cmd_sock);
403 if (ret < 0) {
404 PERROR("kernel consumer cmd_sock close");
405 }
406 }
407 if (ustconsumer32_data.cmd_sock >= 0) {
408 ret = close(ustconsumer32_data.cmd_sock);
409 if (ret < 0) {
410 PERROR("UST consumerd32 cmd_sock close");
411 }
412 }
413 if (ustconsumer64_data.cmd_sock >= 0) {
414 ret = close(ustconsumer64_data.cmd_sock);
415 if (ret < 0) {
416 PERROR("UST consumerd64 cmd_sock close");
417 }
418 }
419 }
420
421 /*
422 * Cleanup the daemon
423 */
424 static void cleanup(void)
425 {
426 int ret;
427 struct ltt_session *sess, *stmp;
428 char path[PATH_MAX];
429
430 DBG("Cleaning up");
431
432 /*
433 * Close the thread quit pipe. It has already done its job,
434 * since we are now called.
435 */
436 utils_close_pipe(thread_quit_pipe);
437
438 /*
439 * If opt_pidfile is undefined, the default file will be wiped when
440 * removing the rundir.
441 */
442 if (opt_pidfile) {
443 ret = remove(opt_pidfile);
444 if (ret < 0) {
445 PERROR("remove pidfile %s", opt_pidfile);
446 }
447 }
448
449 DBG("Removing sessiond and consumerd content of directory %s", rundir);
450
451 /* sessiond */
452 snprintf(path, PATH_MAX,
453 "%s/%s",
454 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
455 DBG("Removing %s", path);
456 (void) unlink(path);
457
458 snprintf(path, PATH_MAX, "%s/%s", rundir,
459 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
460 DBG("Removing %s", path);
461 (void) unlink(path);
462
463 /* kconsumerd */
464 snprintf(path, PATH_MAX,
465 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
466 rundir);
467 DBG("Removing %s", path);
468 (void) unlink(path);
469
470 snprintf(path, PATH_MAX,
471 DEFAULT_KCONSUMERD_PATH,
472 rundir);
473 DBG("Removing directory %s", path);
474 (void) rmdir(path);
475
476 /* ust consumerd 32 */
477 snprintf(path, PATH_MAX,
478 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
479 rundir);
480 DBG("Removing %s", path);
481 (void) unlink(path);
482
483 snprintf(path, PATH_MAX,
484 DEFAULT_USTCONSUMERD32_PATH,
485 rundir);
486 DBG("Removing directory %s", path);
487 (void) rmdir(path);
488
489 /* ust consumerd 64 */
490 snprintf(path, PATH_MAX,
491 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
492 rundir);
493 DBG("Removing %s", path);
494 (void) unlink(path);
495
496 snprintf(path, PATH_MAX,
497 DEFAULT_USTCONSUMERD64_PATH,
498 rundir);
499 DBG("Removing directory %s", path);
500 (void) rmdir(path);
501
502 /*
503 * We do NOT rmdir rundir because there are other processes
504 * using it, for instance lttng-relayd, which can start in
505 * parallel with this teardown.
506 */
507
508 free(rundir);
509
510 DBG("Cleaning up all sessions");
511
512 /* Destroy session list mutex */
513 if (session_list_ptr != NULL) {
514 pthread_mutex_destroy(&session_list_ptr->lock);
515
516 /* Cleanup ALL session */
517 cds_list_for_each_entry_safe(sess, stmp,
518 &session_list_ptr->head, list) {
519 cmd_destroy_session(sess, kernel_poll_pipe[1]);
520 }
521 }
522
523 DBG("Closing all UST sockets");
524 ust_app_clean_list();
525 buffer_reg_destroy_registries();
526
527 if (is_root && !opt_no_kernel) {
528 DBG2("Closing kernel fd");
529 if (kernel_tracer_fd >= 0) {
530 ret = close(kernel_tracer_fd);
531 if (ret) {
532 PERROR("close");
533 }
534 }
535 DBG("Unloading kernel modules");
536 modprobe_remove_lttng_all();
537 }
538
539 close_consumer_sockets();
540
541 /* <fun> */
542 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
543 "Matthew, BEET driven development works!%c[%dm",
544 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
545 /* </fun> */
546 }
547
548 /*
549 * Send data on a unix socket using the liblttsessiondcomm API.
550 *
551 * Return lttcomm error code.
552 */
553 static int send_unix_sock(int sock, void *buf, size_t len)
554 {
555 /* Check valid length */
556 if (len == 0) {
557 return -1;
558 }
559
560 return lttcomm_send_unix_sock(sock, buf, len);
561 }
562
563 /*
564 * Free memory of a command context structure.
565 */
566 static void clean_command_ctx(struct command_ctx **cmd_ctx)
567 {
568 DBG("Clean command context structure");
569 if (*cmd_ctx) {
570 if ((*cmd_ctx)->llm) {
571 free((*cmd_ctx)->llm);
572 }
573 if ((*cmd_ctx)->lsm) {
574 free((*cmd_ctx)->lsm);
575 }
576 free(*cmd_ctx);
577 *cmd_ctx = NULL;
578 }
579 }
580
581 /*
582 * Notify UST applications using the shm mmap futex.
583 */
584 static int notify_ust_apps(int active)
585 {
586 char *wait_shm_mmap;
587
588 DBG("Notifying applications of session daemon state: %d", active);
589
590 /* See shm.c for this call implying mmap, shm and futex calls */
591 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
592 if (wait_shm_mmap == NULL) {
593 goto error;
594 }
595
596 /* Wake waiting process */
597 futex_wait_update((int32_t *) wait_shm_mmap, active);
598
599 /* Apps notified successfully */
600 return 0;
601
602 error:
603 return -1;
604 }
605
606 /*
607 * Setup the outgoing data buffer for the response (llm) by allocating the
608 * right amount of memory and copying the original information from the lsm
609 * structure.
610 *
611 * Return total size of the buffer pointed by buf.
612 */
613 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
614 {
615 int ret, buf_size;
616
617 buf_size = size;
618
619 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
620 if (cmd_ctx->llm == NULL) {
621 PERROR("zmalloc");
622 ret = -ENOMEM;
623 goto error;
624 }
625
626 /* Copy common data */
627 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
628 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
629
630 cmd_ctx->llm->data_size = size;
631 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
632
633 return buf_size;
634
635 error:
636 return ret;
637 }
638
639 /*
640 * Update the kernel poll set of all channel fd available over all tracing
641 * session. Add the wakeup pipe at the end of the set.
642 */
643 static int update_kernel_poll(struct lttng_poll_event *events)
644 {
645 int ret;
646 struct ltt_session *session;
647 struct ltt_kernel_channel *channel;
648
649 DBG("Updating kernel poll set");
650
651 session_lock_list();
652 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
653 session_lock(session);
654 if (session->kernel_session == NULL) {
655 session_unlock(session);
656 continue;
657 }
658
659 cds_list_for_each_entry(channel,
660 &session->kernel_session->channel_list.head, list) {
661 /* Add channel fd to the kernel poll set */
662 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
663 if (ret < 0) {
664 session_unlock(session);
665 goto error;
666 }
667 DBG("Channel fd %d added to kernel set", channel->fd);
668 }
669 session_unlock(session);
670 }
671 session_unlock_list();
672
673 return 0;
674
675 error:
676 session_unlock_list();
677 return -1;
678 }
679
680 /*
681 * Find the channel fd from 'fd' over all tracing session. When found, check
682 * for new channel stream and send those stream fds to the kernel consumer.
683 *
684 * Useful for CPU hotplug feature.
685 */
686 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
687 {
688 int ret = 0;
689 struct ltt_session *session;
690 struct ltt_kernel_session *ksess;
691 struct ltt_kernel_channel *channel;
692
693 DBG("Updating kernel streams for channel fd %d", fd);
694
695 session_lock_list();
696 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
697 session_lock(session);
698 if (session->kernel_session == NULL) {
699 session_unlock(session);
700 continue;
701 }
702 ksess = session->kernel_session;
703
704 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
705 if (channel->fd == fd) {
706 DBG("Channel found, updating kernel streams");
707 ret = kernel_open_channel_stream(channel);
708 if (ret < 0) {
709 goto error;
710 }
711 /* Update the stream global counter */
712 ksess->stream_count_global += ret;
713
714 /*
715 * Have we already sent fds to the consumer? If yes, it means
716 * that tracing is started so it is safe to send our updated
717 * stream fds.
718 */
719 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
720 struct lttng_ht_iter iter;
721 struct consumer_socket *socket;
722
723 rcu_read_lock();
724 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
725 &iter.iter, socket, node.node) {
726 pthread_mutex_lock(socket->lock);
727 ret = kernel_consumer_send_channel_stream(socket,
728 channel, ksess,
729 session->output_traces ? 1 : 0);
730 pthread_mutex_unlock(socket->lock);
731 if (ret < 0) {
732 rcu_read_unlock();
733 goto error;
734 }
735 }
736 rcu_read_unlock();
737 }
738 goto error;
739 }
740 }
741 session_unlock(session);
742 }
743 session_unlock_list();
744 return ret;
745
746 error:
747 session_unlock(session);
748 session_unlock_list();
749 return ret;
750 }
751
752 /*
753 * For each tracing session, update newly registered apps. The session list
754 * lock MUST be acquired before calling this.
755 */
756 static void update_ust_app(int app_sock)
757 {
758 struct ltt_session *sess, *stmp;
759
760 /* Consumer is in an ERROR state. Stop any application update. */
761 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
762 /* Stop the update process since the consumer is dead. */
763 return;
764 }
765
766 /* For all tracing session(s) */
767 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
768 session_lock(sess);
769 if (sess->ust_session) {
770 ust_app_global_update(sess->ust_session, app_sock);
771 }
772 session_unlock(sess);
773 }
774 }
775
776 /*
777 * This thread manage event coming from the kernel.
778 *
779 * Features supported in this thread:
780 * -) CPU Hotplug
781 */
782 static void *thread_manage_kernel(void *data)
783 {
784 int ret, i, pollfd, update_poll_flag = 1, err = -1;
785 uint32_t revents, nb_fd;
786 char tmp;
787 struct lttng_poll_event events;
788
789 DBG("[thread] Thread manage kernel started");
790
791 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
792
793 /*
794 * This first step of the while is to clean this structure which could free
795 * non NULL pointers so initialize it before the loop.
796 */
797 lttng_poll_init(&events);
798
799 if (testpoint(thread_manage_kernel)) {
800 goto error_testpoint;
801 }
802
803 health_code_update();
804
805 if (testpoint(thread_manage_kernel_before_loop)) {
806 goto error_testpoint;
807 }
808
809 while (1) {
810 health_code_update();
811
812 if (update_poll_flag == 1) {
813 /* Clean events object. We are about to populate it again. */
814 lttng_poll_clean(&events);
815
816 ret = sessiond_set_thread_pollset(&events, 2);
817 if (ret < 0) {
818 goto error_poll_create;
819 }
820
821 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
822 if (ret < 0) {
823 goto error;
824 }
825
826 /* This will add the available kernel channel if any. */
827 ret = update_kernel_poll(&events);
828 if (ret < 0) {
829 goto error;
830 }
831 update_poll_flag = 0;
832 }
833
834 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
835
836 /* Poll infinite value of time */
837 restart:
838 health_poll_entry();
839 ret = lttng_poll_wait(&events, -1);
840 health_poll_exit();
841 if (ret < 0) {
842 /*
843 * Restart interrupted system call.
844 */
845 if (errno == EINTR) {
846 goto restart;
847 }
848 goto error;
849 } else if (ret == 0) {
850 /* Should not happen since timeout is infinite */
851 ERR("Return value of poll is 0 with an infinite timeout.\n"
852 "This should not have happened! Continuing...");
853 continue;
854 }
855
856 nb_fd = ret;
857
858 for (i = 0; i < nb_fd; i++) {
859 /* Fetch once the poll data */
860 revents = LTTNG_POLL_GETEV(&events, i);
861 pollfd = LTTNG_POLL_GETFD(&events, i);
862
863 health_code_update();
864
865 /* Thread quit pipe has been closed. Killing thread. */
866 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
867 if (ret) {
868 err = 0;
869 goto exit;
870 }
871
872 /* Check for data on kernel pipe */
873 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
874 (void) lttng_read(kernel_poll_pipe[0],
875 &tmp, 1);
876 /*
877 * Ret value is useless here, if this pipe gets any actions an
878 * update is required anyway.
879 */
880 update_poll_flag = 1;
881 continue;
882 } else {
883 /*
884 * New CPU detected by the kernel. Adding kernel stream to
885 * kernel session and updating the kernel consumer
886 */
887 if (revents & LPOLLIN) {
888 ret = update_kernel_stream(&kconsumer_data, pollfd);
889 if (ret < 0) {
890 continue;
891 }
892 break;
893 /*
894 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
895 * and unregister kernel stream at this point.
896 */
897 }
898 }
899 }
900 }
901
902 exit:
903 error:
904 lttng_poll_clean(&events);
905 error_poll_create:
906 error_testpoint:
907 utils_close_pipe(kernel_poll_pipe);
908 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
909 if (err) {
910 health_error();
911 ERR("Health error occurred in %s", __func__);
912 WARN("Kernel thread died unexpectedly. "
913 "Kernel tracing can continue but CPU hotplug is disabled.");
914 }
915 health_unregister(health_sessiond);
916 DBG("Kernel thread dying");
917 return NULL;
918 }
919
920 /*
921 * Signal pthread condition of the consumer data that the thread.
922 */
923 static void signal_consumer_condition(struct consumer_data *data, int state)
924 {
925 pthread_mutex_lock(&data->cond_mutex);
926
927 /*
928 * The state is set before signaling. It can be any value, it's the waiter
929 * job to correctly interpret this condition variable associated to the
930 * consumer pthread_cond.
931 *
932 * A value of 0 means that the corresponding thread of the consumer data
933 * was not started. 1 indicates that the thread has started and is ready
934 * for action. A negative value means that there was an error during the
935 * thread bootstrap.
936 */
937 data->consumer_thread_is_ready = state;
938 (void) pthread_cond_signal(&data->cond);
939
940 pthread_mutex_unlock(&data->cond_mutex);
941 }
942
943 /*
944 * This thread manage the consumer error sent back to the session daemon.
945 */
946 static void *thread_manage_consumer(void *data)
947 {
948 int sock = -1, i, ret, pollfd, err = -1;
949 uint32_t revents, nb_fd;
950 enum lttcomm_return_code code;
951 struct lttng_poll_event events;
952 struct consumer_data *consumer_data = data;
953
954 DBG("[thread] Manage consumer started");
955
956 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
957
958 health_code_update();
959
960 /*
961 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
962 * metadata_sock. Nothing more will be added to this poll set.
963 */
964 ret = sessiond_set_thread_pollset(&events, 3);
965 if (ret < 0) {
966 goto error_poll;
967 }
968
969 /*
970 * The error socket here is already in a listening state which was done
971 * just before spawning this thread to avoid a race between the consumer
972 * daemon exec trying to connect and the listen() call.
973 */
974 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
975 if (ret < 0) {
976 goto error;
977 }
978
979 health_code_update();
980
981 /* Infinite blocking call, waiting for transmission */
982 restart:
983 health_poll_entry();
984
985 if (testpoint(thread_manage_consumer)) {
986 goto error;
987 }
988
989 ret = lttng_poll_wait(&events, -1);
990 health_poll_exit();
991 if (ret < 0) {
992 /*
993 * Restart interrupted system call.
994 */
995 if (errno == EINTR) {
996 goto restart;
997 }
998 goto error;
999 }
1000
1001 nb_fd = ret;
1002
1003 for (i = 0; i < nb_fd; i++) {
1004 /* Fetch once the poll data */
1005 revents = LTTNG_POLL_GETEV(&events, i);
1006 pollfd = LTTNG_POLL_GETFD(&events, i);
1007
1008 health_code_update();
1009
1010 /* Thread quit pipe has been closed. Killing thread. */
1011 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1012 if (ret) {
1013 err = 0;
1014 goto exit;
1015 }
1016
1017 /* Event on the registration socket */
1018 if (pollfd == consumer_data->err_sock) {
1019 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1020 ERR("consumer err socket poll error");
1021 goto error;
1022 }
1023 }
1024 }
1025
1026 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1027 if (sock < 0) {
1028 goto error;
1029 }
1030
1031 /*
1032 * Set the CLOEXEC flag. Return code is useless because either way, the
1033 * show must go on.
1034 */
1035 (void) utils_set_fd_cloexec(sock);
1036
1037 health_code_update();
1038
1039 DBG2("Receiving code from consumer err_sock");
1040
1041 /* Getting status code from kconsumerd */
1042 ret = lttcomm_recv_unix_sock(sock, &code,
1043 sizeof(enum lttcomm_return_code));
1044 if (ret <= 0) {
1045 goto error;
1046 }
1047
1048 health_code_update();
1049
1050 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1051 /* Connect both socket, command and metadata. */
1052 consumer_data->cmd_sock =
1053 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1054 consumer_data->metadata_fd =
1055 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1056 if (consumer_data->cmd_sock < 0
1057 || consumer_data->metadata_fd < 0) {
1058 PERROR("consumer connect cmd socket");
1059 /* On error, signal condition and quit. */
1060 signal_consumer_condition(consumer_data, -1);
1061 goto error;
1062 }
1063 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1064 /* Create metadata socket lock. */
1065 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1066 if (consumer_data->metadata_sock.lock == NULL) {
1067 PERROR("zmalloc pthread mutex");
1068 ret = -1;
1069 goto error;
1070 }
1071 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1072
1073 signal_consumer_condition(consumer_data, 1);
1074 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1075 DBG("Consumer metadata socket ready (fd: %d)",
1076 consumer_data->metadata_fd);
1077 } else {
1078 ERR("consumer error when waiting for SOCK_READY : %s",
1079 lttcomm_get_readable_code(-code));
1080 goto error;
1081 }
1082
1083 /* Remove the consumerd error sock since we've established a connexion */
1084 ret = lttng_poll_del(&events, consumer_data->err_sock);
1085 if (ret < 0) {
1086 goto error;
1087 }
1088
1089 /* Add new accepted error socket. */
1090 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1091 if (ret < 0) {
1092 goto error;
1093 }
1094
1095 /* Add metadata socket that is successfully connected. */
1096 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1097 LPOLLIN | LPOLLRDHUP);
1098 if (ret < 0) {
1099 goto error;
1100 }
1101
1102 health_code_update();
1103
1104 /* Infinite blocking call, waiting for transmission */
1105 restart_poll:
1106 while (1) {
1107 health_poll_entry();
1108 ret = lttng_poll_wait(&events, -1);
1109 health_poll_exit();
1110 if (ret < 0) {
1111 /*
1112 * Restart interrupted system call.
1113 */
1114 if (errno == EINTR) {
1115 goto restart_poll;
1116 }
1117 goto error;
1118 }
1119
1120 nb_fd = ret;
1121
1122 for (i = 0; i < nb_fd; i++) {
1123 /* Fetch once the poll data */
1124 revents = LTTNG_POLL_GETEV(&events, i);
1125 pollfd = LTTNG_POLL_GETFD(&events, i);
1126
1127 health_code_update();
1128
1129 /* Thread quit pipe has been closed. Killing thread. */
1130 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1131 if (ret) {
1132 err = 0;
1133 goto exit;
1134 }
1135
1136 if (pollfd == sock) {
1137 /* Event on the consumerd socket */
1138 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1139 ERR("consumer err socket second poll error");
1140 goto error;
1141 }
1142 health_code_update();
1143 /* Wait for any kconsumerd error */
1144 ret = lttcomm_recv_unix_sock(sock, &code,
1145 sizeof(enum lttcomm_return_code));
1146 if (ret <= 0) {
1147 ERR("consumer closed the command socket");
1148 goto error;
1149 }
1150
1151 ERR("consumer return code : %s",
1152 lttcomm_get_readable_code(-code));
1153
1154 goto exit;
1155 } else if (pollfd == consumer_data->metadata_fd) {
1156 /* UST metadata requests */
1157 ret = ust_consumer_metadata_request(
1158 &consumer_data->metadata_sock);
1159 if (ret < 0) {
1160 ERR("Handling metadata request");
1161 goto error;
1162 }
1163 break;
1164 } else {
1165 ERR("Unknown pollfd");
1166 goto error;
1167 }
1168 }
1169 health_code_update();
1170 }
1171
1172 exit:
1173 error:
1174 /*
1175 * We lock here because we are about to close the sockets and some other
1176 * thread might be using them so get exclusive access which will abort all
1177 * other consumer command by other threads.
1178 */
1179 pthread_mutex_lock(&consumer_data->lock);
1180
1181 /* Immediately set the consumerd state to stopped */
1182 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1183 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1184 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1185 consumer_data->type == LTTNG_CONSUMER32_UST) {
1186 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1187 } else {
1188 /* Code flow error... */
1189 assert(0);
1190 }
1191
1192 if (consumer_data->err_sock >= 0) {
1193 ret = close(consumer_data->err_sock);
1194 if (ret) {
1195 PERROR("close");
1196 }
1197 consumer_data->err_sock = -1;
1198 }
1199 if (consumer_data->cmd_sock >= 0) {
1200 ret = close(consumer_data->cmd_sock);
1201 if (ret) {
1202 PERROR("close");
1203 }
1204 consumer_data->cmd_sock = -1;
1205 }
1206 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1207 ret = close(*consumer_data->metadata_sock.fd_ptr);
1208 if (ret) {
1209 PERROR("close");
1210 }
1211 }
1212
1213 if (sock >= 0) {
1214 ret = close(sock);
1215 if (ret) {
1216 PERROR("close");
1217 }
1218 }
1219
1220 unlink(consumer_data->err_unix_sock_path);
1221 unlink(consumer_data->cmd_unix_sock_path);
1222 consumer_data->pid = 0;
1223 pthread_mutex_unlock(&consumer_data->lock);
1224
1225 /* Cleanup metadata socket mutex. */
1226 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1227 free(consumer_data->metadata_sock.lock);
1228
1229 lttng_poll_clean(&events);
1230 error_poll:
1231 if (err) {
1232 health_error();
1233 ERR("Health error occurred in %s", __func__);
1234 }
1235 health_unregister(health_sessiond);
1236 DBG("consumer thread cleanup completed");
1237
1238 return NULL;
1239 }
1240
1241 /*
1242 * This thread manage application communication.
1243 */
1244 static void *thread_manage_apps(void *data)
1245 {
1246 int i, ret, pollfd, err = -1;
1247 ssize_t size_ret;
1248 uint32_t revents, nb_fd;
1249 struct lttng_poll_event events;
1250
1251 DBG("[thread] Manage application started");
1252
1253 rcu_register_thread();
1254 rcu_thread_online();
1255
1256 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1257
1258 if (testpoint(thread_manage_apps)) {
1259 goto error_testpoint;
1260 }
1261
1262 health_code_update();
1263
1264 ret = sessiond_set_thread_pollset(&events, 2);
1265 if (ret < 0) {
1266 goto error_poll_create;
1267 }
1268
1269 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1270 if (ret < 0) {
1271 goto error;
1272 }
1273
1274 if (testpoint(thread_manage_apps_before_loop)) {
1275 goto error;
1276 }
1277
1278 health_code_update();
1279
1280 while (1) {
1281 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1282
1283 /* Inifinite blocking call, waiting for transmission */
1284 restart:
1285 health_poll_entry();
1286 ret = lttng_poll_wait(&events, -1);
1287 health_poll_exit();
1288 if (ret < 0) {
1289 /*
1290 * Restart interrupted system call.
1291 */
1292 if (errno == EINTR) {
1293 goto restart;
1294 }
1295 goto error;
1296 }
1297
1298 nb_fd = ret;
1299
1300 for (i = 0; i < nb_fd; i++) {
1301 /* Fetch once the poll data */
1302 revents = LTTNG_POLL_GETEV(&events, i);
1303 pollfd = LTTNG_POLL_GETFD(&events, i);
1304
1305 health_code_update();
1306
1307 /* Thread quit pipe has been closed. Killing thread. */
1308 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1309 if (ret) {
1310 err = 0;
1311 goto exit;
1312 }
1313
1314 /* Inspect the apps cmd pipe */
1315 if (pollfd == apps_cmd_pipe[0]) {
1316 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1317 ERR("Apps command pipe error");
1318 goto error;
1319 } else if (revents & LPOLLIN) {
1320 int sock;
1321
1322 /* Empty pipe */
1323 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1324 if (size_ret < sizeof(sock)) {
1325 PERROR("read apps cmd pipe");
1326 goto error;
1327 }
1328
1329 health_code_update();
1330
1331 /*
1332 * We only monitor the error events of the socket. This
1333 * thread does not handle any incoming data from UST
1334 * (POLLIN).
1335 */
1336 ret = lttng_poll_add(&events, sock,
1337 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1338 if (ret < 0) {
1339 goto error;
1340 }
1341
1342 DBG("Apps with sock %d added to poll set", sock);
1343
1344 health_code_update();
1345
1346 break;
1347 }
1348 } else {
1349 /*
1350 * At this point, we know that a registered application made
1351 * the event at poll_wait.
1352 */
1353 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1354 /* Removing from the poll set */
1355 ret = lttng_poll_del(&events, pollfd);
1356 if (ret < 0) {
1357 goto error;
1358 }
1359
1360 /* Socket closed on remote end. */
1361 ust_app_unregister(pollfd);
1362 break;
1363 }
1364 }
1365
1366 health_code_update();
1367 }
1368 }
1369
1370 exit:
1371 error:
1372 lttng_poll_clean(&events);
1373 error_poll_create:
1374 error_testpoint:
1375 utils_close_pipe(apps_cmd_pipe);
1376 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1377
1378 /*
1379 * We don't clean the UST app hash table here since already registered
1380 * applications can still be controlled so let them be until the session
1381 * daemon dies or the applications stop.
1382 */
1383
1384 if (err) {
1385 health_error();
1386 ERR("Health error occurred in %s", __func__);
1387 }
1388 health_unregister(health_sessiond);
1389 DBG("Application communication apps thread cleanup complete");
1390 rcu_thread_offline();
1391 rcu_unregister_thread();
1392 return NULL;
1393 }
1394
1395 /*
1396 * Send a socket to a thread This is called from the dispatch UST registration
1397 * thread once all sockets are set for the application.
1398 *
1399 * The sock value can be invalid, we don't really care, the thread will handle
1400 * it and make the necessary cleanup if so.
1401 *
1402 * On success, return 0 else a negative value being the errno message of the
1403 * write().
1404 */
1405 static int send_socket_to_thread(int fd, int sock)
1406 {
1407 ssize_t ret;
1408
1409 /*
1410 * It's possible that the FD is set as invalid with -1 concurrently just
1411 * before calling this function being a shutdown state of the thread.
1412 */
1413 if (fd < 0) {
1414 ret = -EBADF;
1415 goto error;
1416 }
1417
1418 ret = lttng_write(fd, &sock, sizeof(sock));
1419 if (ret < sizeof(sock)) {
1420 PERROR("write apps pipe %d", fd);
1421 if (ret < 0) {
1422 ret = -errno;
1423 }
1424 goto error;
1425 }
1426
1427 /* All good. Don't send back the write positive ret value. */
1428 ret = 0;
1429 error:
1430 return (int) ret;
1431 }
1432
1433 /*
1434 * Sanitize the wait queue of the dispatch registration thread meaning removing
1435 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1436 * notify socket is never received.
1437 */
1438 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1439 {
1440 int ret, nb_fd = 0, i;
1441 unsigned int fd_added = 0;
1442 struct lttng_poll_event events;
1443 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1444
1445 assert(wait_queue);
1446
1447 lttng_poll_init(&events);
1448
1449 /* Just skip everything for an empty queue. */
1450 if (!wait_queue->count) {
1451 goto end;
1452 }
1453
1454 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1455 if (ret < 0) {
1456 goto error_create;
1457 }
1458
1459 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1460 &wait_queue->head, head) {
1461 assert(wait_node->app);
1462 ret = lttng_poll_add(&events, wait_node->app->sock,
1463 LPOLLHUP | LPOLLERR);
1464 if (ret < 0) {
1465 goto error;
1466 }
1467
1468 fd_added = 1;
1469 }
1470
1471 if (!fd_added) {
1472 goto end;
1473 }
1474
1475 /*
1476 * Poll but don't block so we can quickly identify the faulty events and
1477 * clean them afterwards from the wait queue.
1478 */
1479 ret = lttng_poll_wait(&events, 0);
1480 if (ret < 0) {
1481 goto error;
1482 }
1483 nb_fd = ret;
1484
1485 for (i = 0; i < nb_fd; i++) {
1486 /* Get faulty FD. */
1487 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1488 int pollfd = LTTNG_POLL_GETFD(&events, i);
1489
1490 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1491 &wait_queue->head, head) {
1492 if (pollfd == wait_node->app->sock &&
1493 (revents & (LPOLLHUP | LPOLLERR))) {
1494 cds_list_del(&wait_node->head);
1495 wait_queue->count--;
1496 ust_app_destroy(wait_node->app);
1497 free(wait_node);
1498 break;
1499 }
1500 }
1501 }
1502
1503 if (nb_fd > 0) {
1504 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1505 }
1506
1507 end:
1508 lttng_poll_clean(&events);
1509 return;
1510
1511 error:
1512 lttng_poll_clean(&events);
1513 error_create:
1514 ERR("Unable to sanitize wait queue");
1515 return;
1516 }
1517
1518 /*
1519 * Dispatch request from the registration threads to the application
1520 * communication thread.
1521 */
1522 static void *thread_dispatch_ust_registration(void *data)
1523 {
1524 int ret, err = -1;
1525 struct cds_wfq_node *node;
1526 struct ust_command *ust_cmd = NULL;
1527 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1528 struct ust_reg_wait_queue wait_queue = {
1529 .count = 0,
1530 };
1531
1532 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1533
1534 health_code_update();
1535
1536 CDS_INIT_LIST_HEAD(&wait_queue.head);
1537
1538 DBG("[thread] Dispatch UST command started");
1539
1540 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1541 health_code_update();
1542
1543 /* Atomically prepare the queue futex */
1544 futex_nto1_prepare(&ust_cmd_queue.futex);
1545
1546 do {
1547 struct ust_app *app = NULL;
1548 ust_cmd = NULL;
1549
1550 /*
1551 * Make sure we don't have node(s) that have hung up before receiving
1552 * the notify socket. This is to clean the list in order to avoid
1553 * memory leaks from notify socket that are never seen.
1554 */
1555 sanitize_wait_queue(&wait_queue);
1556
1557 health_code_update();
1558 /* Dequeue command for registration */
1559 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1560 if (node == NULL) {
1561 DBG("Woken up but nothing in the UST command queue");
1562 /* Continue thread execution */
1563 break;
1564 }
1565
1566 ust_cmd = caa_container_of(node, struct ust_command, node);
1567
1568 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1569 " gid:%d sock:%d name:%s (version %d.%d)",
1570 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1571 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1572 ust_cmd->sock, ust_cmd->reg_msg.name,
1573 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1574
1575 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1576 wait_node = zmalloc(sizeof(*wait_node));
1577 if (!wait_node) {
1578 PERROR("zmalloc wait_node dispatch");
1579 ret = close(ust_cmd->sock);
1580 if (ret < 0) {
1581 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1582 }
1583 lttng_fd_put(LTTNG_FD_APPS, 1);
1584 free(ust_cmd);
1585 goto error;
1586 }
1587 CDS_INIT_LIST_HEAD(&wait_node->head);
1588
1589 /* Create application object if socket is CMD. */
1590 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1591 ust_cmd->sock);
1592 if (!wait_node->app) {
1593 ret = close(ust_cmd->sock);
1594 if (ret < 0) {
1595 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1596 }
1597 lttng_fd_put(LTTNG_FD_APPS, 1);
1598 free(wait_node);
1599 free(ust_cmd);
1600 continue;
1601 }
1602 /*
1603 * Add application to the wait queue so we can set the notify
1604 * socket before putting this object in the global ht.
1605 */
1606 cds_list_add(&wait_node->head, &wait_queue.head);
1607 wait_queue.count++;
1608
1609 free(ust_cmd);
1610 /*
1611 * We have to continue here since we don't have the notify
1612 * socket and the application MUST be added to the hash table
1613 * only at that moment.
1614 */
1615 continue;
1616 } else {
1617 /*
1618 * Look for the application in the local wait queue and set the
1619 * notify socket if found.
1620 */
1621 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1622 &wait_queue.head, head) {
1623 health_code_update();
1624 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1625 wait_node->app->notify_sock = ust_cmd->sock;
1626 cds_list_del(&wait_node->head);
1627 wait_queue.count--;
1628 app = wait_node->app;
1629 free(wait_node);
1630 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1631 break;
1632 }
1633 }
1634
1635 /*
1636 * With no application at this stage the received socket is
1637 * basically useless so close it before we free the cmd data
1638 * structure for good.
1639 */
1640 if (!app) {
1641 ret = close(ust_cmd->sock);
1642 if (ret < 0) {
1643 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1644 }
1645 lttng_fd_put(LTTNG_FD_APPS, 1);
1646 }
1647 free(ust_cmd);
1648 }
1649
1650 if (app) {
1651 /*
1652 * @session_lock_list
1653 *
1654 * Lock the global session list so from the register up to the
1655 * registration done message, no thread can see the application
1656 * and change its state.
1657 */
1658 session_lock_list();
1659 rcu_read_lock();
1660
1661 /*
1662 * Add application to the global hash table. This needs to be
1663 * done before the update to the UST registry can locate the
1664 * application.
1665 */
1666 ust_app_add(app);
1667
1668 /* Set app version. This call will print an error if needed. */
1669 (void) ust_app_version(app);
1670
1671 /* Send notify socket through the notify pipe. */
1672 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1673 app->notify_sock);
1674 if (ret < 0) {
1675 rcu_read_unlock();
1676 session_unlock_list();
1677 /*
1678 * No notify thread, stop the UST tracing. However, this is
1679 * not an internal error of the this thread thus setting
1680 * the health error code to a normal exit.
1681 */
1682 err = 0;
1683 goto error;
1684 }
1685
1686 /*
1687 * Update newly registered application with the tracing
1688 * registry info already enabled information.
1689 */
1690 update_ust_app(app->sock);
1691
1692 /*
1693 * Don't care about return value. Let the manage apps threads
1694 * handle app unregistration upon socket close.
1695 */
1696 (void) ust_app_register_done(app->sock);
1697
1698 /*
1699 * Even if the application socket has been closed, send the app
1700 * to the thread and unregistration will take place at that
1701 * place.
1702 */
1703 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1704 if (ret < 0) {
1705 rcu_read_unlock();
1706 session_unlock_list();
1707 /*
1708 * No apps. thread, stop the UST tracing. However, this is
1709 * not an internal error of the this thread thus setting
1710 * the health error code to a normal exit.
1711 */
1712 err = 0;
1713 goto error;
1714 }
1715
1716 rcu_read_unlock();
1717 session_unlock_list();
1718 }
1719 } while (node != NULL);
1720
1721 health_poll_entry();
1722 /* Futex wait on queue. Blocking call on futex() */
1723 futex_nto1_wait(&ust_cmd_queue.futex);
1724 health_poll_exit();
1725 }
1726 /* Normal exit, no error */
1727 err = 0;
1728
1729 error:
1730 /* Clean up wait queue. */
1731 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1732 &wait_queue.head, head) {
1733 cds_list_del(&wait_node->head);
1734 wait_queue.count--;
1735 free(wait_node);
1736 }
1737
1738 DBG("Dispatch thread dying");
1739 if (err) {
1740 health_error();
1741 ERR("Health error occurred in %s", __func__);
1742 }
1743 health_unregister(health_sessiond);
1744 return NULL;
1745 }
1746
1747 /*
1748 * This thread manage application registration.
1749 */
1750 static void *thread_registration_apps(void *data)
1751 {
1752 int sock = -1, i, ret, pollfd, err = -1;
1753 uint32_t revents, nb_fd;
1754 struct lttng_poll_event events;
1755 /*
1756 * Get allocated in this thread, enqueued to a global queue, dequeued and
1757 * freed in the manage apps thread.
1758 */
1759 struct ust_command *ust_cmd = NULL;
1760
1761 DBG("[thread] Manage application registration started");
1762
1763 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1764
1765 if (testpoint(thread_registration_apps)) {
1766 goto error_testpoint;
1767 }
1768
1769 ret = lttcomm_listen_unix_sock(apps_sock);
1770 if (ret < 0) {
1771 goto error_listen;
1772 }
1773
1774 /*
1775 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1776 * more will be added to this poll set.
1777 */
1778 ret = sessiond_set_thread_pollset(&events, 2);
1779 if (ret < 0) {
1780 goto error_create_poll;
1781 }
1782
1783 /* Add the application registration socket */
1784 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1785 if (ret < 0) {
1786 goto error_poll_add;
1787 }
1788
1789 /* Notify all applications to register */
1790 ret = notify_ust_apps(1);
1791 if (ret < 0) {
1792 ERR("Failed to notify applications or create the wait shared memory.\n"
1793 "Execution continues but there might be problem for already\n"
1794 "running applications that wishes to register.");
1795 }
1796
1797 while (1) {
1798 DBG("Accepting application registration");
1799
1800 /* Inifinite blocking call, waiting for transmission */
1801 restart:
1802 health_poll_entry();
1803 ret = lttng_poll_wait(&events, -1);
1804 health_poll_exit();
1805 if (ret < 0) {
1806 /*
1807 * Restart interrupted system call.
1808 */
1809 if (errno == EINTR) {
1810 goto restart;
1811 }
1812 goto error;
1813 }
1814
1815 nb_fd = ret;
1816
1817 for (i = 0; i < nb_fd; i++) {
1818 health_code_update();
1819
1820 /* Fetch once the poll data */
1821 revents = LTTNG_POLL_GETEV(&events, i);
1822 pollfd = LTTNG_POLL_GETFD(&events, i);
1823
1824 /* Thread quit pipe has been closed. Killing thread. */
1825 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1826 if (ret) {
1827 err = 0;
1828 goto exit;
1829 }
1830
1831 /* Event on the registration socket */
1832 if (pollfd == apps_sock) {
1833 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1834 ERR("Register apps socket poll error");
1835 goto error;
1836 } else if (revents & LPOLLIN) {
1837 sock = lttcomm_accept_unix_sock(apps_sock);
1838 if (sock < 0) {
1839 goto error;
1840 }
1841
1842 /*
1843 * Set socket timeout for both receiving and ending.
1844 * app_socket_timeout is in seconds, whereas
1845 * lttcomm_setsockopt_rcv_timeout and
1846 * lttcomm_setsockopt_snd_timeout expect msec as
1847 * parameter.
1848 */
1849 (void) lttcomm_setsockopt_rcv_timeout(sock,
1850 app_socket_timeout * 1000);
1851 (void) lttcomm_setsockopt_snd_timeout(sock,
1852 app_socket_timeout * 1000);
1853
1854 /*
1855 * Set the CLOEXEC flag. Return code is useless because
1856 * either way, the show must go on.
1857 */
1858 (void) utils_set_fd_cloexec(sock);
1859
1860 /* Create UST registration command for enqueuing */
1861 ust_cmd = zmalloc(sizeof(struct ust_command));
1862 if (ust_cmd == NULL) {
1863 PERROR("ust command zmalloc");
1864 goto error;
1865 }
1866
1867 /*
1868 * Using message-based transmissions to ensure we don't
1869 * have to deal with partially received messages.
1870 */
1871 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1872 if (ret < 0) {
1873 ERR("Exhausted file descriptors allowed for applications.");
1874 free(ust_cmd);
1875 ret = close(sock);
1876 if (ret) {
1877 PERROR("close");
1878 }
1879 sock = -1;
1880 continue;
1881 }
1882
1883 health_code_update();
1884 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1885 if (ret < 0) {
1886 free(ust_cmd);
1887 /* Close socket of the application. */
1888 ret = close(sock);
1889 if (ret) {
1890 PERROR("close");
1891 }
1892 lttng_fd_put(LTTNG_FD_APPS, 1);
1893 sock = -1;
1894 continue;
1895 }
1896 health_code_update();
1897
1898 ust_cmd->sock = sock;
1899 sock = -1;
1900
1901 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1902 " gid:%d sock:%d name:%s (version %d.%d)",
1903 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1904 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1905 ust_cmd->sock, ust_cmd->reg_msg.name,
1906 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1907
1908 /*
1909 * Lock free enqueue the registration request. The red pill
1910 * has been taken! This apps will be part of the *system*.
1911 */
1912 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1913
1914 /*
1915 * Wake the registration queue futex. Implicit memory
1916 * barrier with the exchange in cds_wfq_enqueue.
1917 */
1918 futex_nto1_wake(&ust_cmd_queue.futex);
1919 }
1920 }
1921 }
1922 }
1923
1924 exit:
1925 error:
1926 if (err) {
1927 health_error();
1928 ERR("Health error occurred in %s", __func__);
1929 }
1930
1931 /* Notify that the registration thread is gone */
1932 notify_ust_apps(0);
1933
1934 if (apps_sock >= 0) {
1935 ret = close(apps_sock);
1936 if (ret) {
1937 PERROR("close");
1938 }
1939 }
1940 if (sock >= 0) {
1941 ret = close(sock);
1942 if (ret) {
1943 PERROR("close");
1944 }
1945 lttng_fd_put(LTTNG_FD_APPS, 1);
1946 }
1947 unlink(apps_unix_sock_path);
1948
1949 error_poll_add:
1950 lttng_poll_clean(&events);
1951 error_listen:
1952 error_create_poll:
1953 error_testpoint:
1954 DBG("UST Registration thread cleanup complete");
1955 health_unregister(health_sessiond);
1956
1957 return NULL;
1958 }
1959
1960 /*
1961 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1962 * exec or it will fails.
1963 */
1964 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1965 {
1966 int ret, clock_ret;
1967 struct timespec timeout;
1968
1969 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1970 consumer_data->consumer_thread_is_ready = 0;
1971
1972 /* Setup pthread condition */
1973 ret = pthread_condattr_init(&consumer_data->condattr);
1974 if (ret != 0) {
1975 errno = ret;
1976 PERROR("pthread_condattr_init consumer data");
1977 goto error;
1978 }
1979
1980 /*
1981 * Set the monotonic clock in order to make sure we DO NOT jump in time
1982 * between the clock_gettime() call and the timedwait call. See bug #324
1983 * for a more details and how we noticed it.
1984 */
1985 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1986 if (ret != 0) {
1987 errno = ret;
1988 PERROR("pthread_condattr_setclock consumer data");
1989 goto error;
1990 }
1991
1992 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1993 if (ret != 0) {
1994 errno = ret;
1995 PERROR("pthread_cond_init consumer data");
1996 goto error;
1997 }
1998
1999 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2000 consumer_data);
2001 if (ret != 0) {
2002 PERROR("pthread_create consumer");
2003 ret = -1;
2004 goto error;
2005 }
2006
2007 /* We are about to wait on a pthread condition */
2008 pthread_mutex_lock(&consumer_data->cond_mutex);
2009
2010 /* Get time for sem_timedwait absolute timeout */
2011 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2012 /*
2013 * Set the timeout for the condition timed wait even if the clock gettime
2014 * call fails since we might loop on that call and we want to avoid to
2015 * increment the timeout too many times.
2016 */
2017 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2018
2019 /*
2020 * The following loop COULD be skipped in some conditions so this is why we
2021 * set ret to 0 in order to make sure at least one round of the loop is
2022 * done.
2023 */
2024 ret = 0;
2025
2026 /*
2027 * Loop until the condition is reached or when a timeout is reached. Note
2028 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2029 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2030 * possible. This loop does not take any chances and works with both of
2031 * them.
2032 */
2033 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2034 if (clock_ret < 0) {
2035 PERROR("clock_gettime spawn consumer");
2036 /* Infinite wait for the consumerd thread to be ready */
2037 ret = pthread_cond_wait(&consumer_data->cond,
2038 &consumer_data->cond_mutex);
2039 } else {
2040 ret = pthread_cond_timedwait(&consumer_data->cond,
2041 &consumer_data->cond_mutex, &timeout);
2042 }
2043 }
2044
2045 /* Release the pthread condition */
2046 pthread_mutex_unlock(&consumer_data->cond_mutex);
2047
2048 if (ret != 0) {
2049 errno = ret;
2050 if (ret == ETIMEDOUT) {
2051 /*
2052 * Call has timed out so we kill the kconsumerd_thread and return
2053 * an error.
2054 */
2055 ERR("Condition timed out. The consumer thread was never ready."
2056 " Killing it");
2057 ret = pthread_cancel(consumer_data->thread);
2058 if (ret < 0) {
2059 PERROR("pthread_cancel consumer thread");
2060 }
2061 } else {
2062 PERROR("pthread_cond_wait failed consumer thread");
2063 }
2064 goto error;
2065 }
2066
2067 pthread_mutex_lock(&consumer_data->pid_mutex);
2068 if (consumer_data->pid == 0) {
2069 ERR("Consumerd did not start");
2070 pthread_mutex_unlock(&consumer_data->pid_mutex);
2071 goto error;
2072 }
2073 pthread_mutex_unlock(&consumer_data->pid_mutex);
2074
2075 return 0;
2076
2077 error:
2078 return ret;
2079 }
2080
2081 /*
2082 * Join consumer thread
2083 */
2084 static int join_consumer_thread(struct consumer_data *consumer_data)
2085 {
2086 void *status;
2087
2088 /* Consumer pid must be a real one. */
2089 if (consumer_data->pid > 0) {
2090 int ret;
2091 ret = kill(consumer_data->pid, SIGTERM);
2092 if (ret) {
2093 ERR("Error killing consumer daemon");
2094 return ret;
2095 }
2096 return pthread_join(consumer_data->thread, &status);
2097 } else {
2098 return 0;
2099 }
2100 }
2101
2102 /*
2103 * Fork and exec a consumer daemon (consumerd).
2104 *
2105 * Return pid if successful else -1.
2106 */
2107 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2108 {
2109 int ret;
2110 pid_t pid;
2111 const char *consumer_to_use;
2112 const char *verbosity;
2113 struct stat st;
2114
2115 DBG("Spawning consumerd");
2116
2117 pid = fork();
2118 if (pid == 0) {
2119 /*
2120 * Exec consumerd.
2121 */
2122 if (opt_verbose_consumer) {
2123 verbosity = "--verbose";
2124 } else {
2125 verbosity = "--quiet";
2126 }
2127 switch (consumer_data->type) {
2128 case LTTNG_CONSUMER_KERNEL:
2129 /*
2130 * Find out which consumerd to execute. We will first try the
2131 * 64-bit path, then the sessiond's installation directory, and
2132 * fallback on the 32-bit one,
2133 */
2134 DBG3("Looking for a kernel consumer at these locations:");
2135 DBG3(" 1) %s", consumerd64_bin);
2136 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2137 DBG3(" 3) %s", consumerd32_bin);
2138 if (stat(consumerd64_bin, &st) == 0) {
2139 DBG3("Found location #1");
2140 consumer_to_use = consumerd64_bin;
2141 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2142 DBG3("Found location #2");
2143 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2144 } else if (stat(consumerd32_bin, &st) == 0) {
2145 DBG3("Found location #3");
2146 consumer_to_use = consumerd32_bin;
2147 } else {
2148 DBG("Could not find any valid consumerd executable");
2149 break;
2150 }
2151 DBG("Using kernel consumer at: %s", consumer_to_use);
2152 execl(consumer_to_use,
2153 "lttng-consumerd", verbosity, "-k",
2154 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2155 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2156 "--group", tracing_group_name,
2157 NULL);
2158 break;
2159 case LTTNG_CONSUMER64_UST:
2160 {
2161 char *tmpnew = NULL;
2162
2163 if (consumerd64_libdir[0] != '\0') {
2164 char *tmp;
2165 size_t tmplen;
2166
2167 tmp = getenv("LD_LIBRARY_PATH");
2168 if (!tmp) {
2169 tmp = "";
2170 }
2171 tmplen = strlen("LD_LIBRARY_PATH=")
2172 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2173 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2174 if (!tmpnew) {
2175 ret = -ENOMEM;
2176 goto error;
2177 }
2178 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2179 strcat(tmpnew, consumerd64_libdir);
2180 if (tmp[0] != '\0') {
2181 strcat(tmpnew, ":");
2182 strcat(tmpnew, tmp);
2183 }
2184 ret = putenv(tmpnew);
2185 if (ret) {
2186 ret = -errno;
2187 free(tmpnew);
2188 goto error;
2189 }
2190 }
2191 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2192 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2193 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2194 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2195 "--group", tracing_group_name,
2196 NULL);
2197 if (consumerd64_libdir[0] != '\0') {
2198 free(tmpnew);
2199 }
2200 if (ret) {
2201 goto error;
2202 }
2203 break;
2204 }
2205 case LTTNG_CONSUMER32_UST:
2206 {
2207 char *tmpnew = NULL;
2208
2209 if (consumerd32_libdir[0] != '\0') {
2210 char *tmp;
2211 size_t tmplen;
2212
2213 tmp = getenv("LD_LIBRARY_PATH");
2214 if (!tmp) {
2215 tmp = "";
2216 }
2217 tmplen = strlen("LD_LIBRARY_PATH=")
2218 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2219 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2220 if (!tmpnew) {
2221 ret = -ENOMEM;
2222 goto error;
2223 }
2224 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2225 strcat(tmpnew, consumerd32_libdir);
2226 if (tmp[0] != '\0') {
2227 strcat(tmpnew, ":");
2228 strcat(tmpnew, tmp);
2229 }
2230 ret = putenv(tmpnew);
2231 if (ret) {
2232 ret = -errno;
2233 free(tmpnew);
2234 goto error;
2235 }
2236 }
2237 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2238 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2239 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2240 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2241 "--group", tracing_group_name,
2242 NULL);
2243 if (consumerd32_libdir[0] != '\0') {
2244 free(tmpnew);
2245 }
2246 if (ret) {
2247 goto error;
2248 }
2249 break;
2250 }
2251 default:
2252 PERROR("unknown consumer type");
2253 exit(EXIT_FAILURE);
2254 }
2255 if (errno != 0) {
2256 PERROR("kernel start consumer exec");
2257 }
2258 exit(EXIT_FAILURE);
2259 } else if (pid > 0) {
2260 ret = pid;
2261 } else {
2262 PERROR("start consumer fork");
2263 ret = -errno;
2264 }
2265 error:
2266 return ret;
2267 }
2268
2269 /*
2270 * Spawn the consumerd daemon and session daemon thread.
2271 */
2272 static int start_consumerd(struct consumer_data *consumer_data)
2273 {
2274 int ret;
2275
2276 /*
2277 * Set the listen() state on the socket since there is a possible race
2278 * between the exec() of the consumer daemon and this call if place in the
2279 * consumer thread. See bug #366 for more details.
2280 */
2281 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2282 if (ret < 0) {
2283 goto error;
2284 }
2285
2286 pthread_mutex_lock(&consumer_data->pid_mutex);
2287 if (consumer_data->pid != 0) {
2288 pthread_mutex_unlock(&consumer_data->pid_mutex);
2289 goto end;
2290 }
2291
2292 ret = spawn_consumerd(consumer_data);
2293 if (ret < 0) {
2294 ERR("Spawning consumerd failed");
2295 pthread_mutex_unlock(&consumer_data->pid_mutex);
2296 goto error;
2297 }
2298
2299 /* Setting up the consumer_data pid */
2300 consumer_data->pid = ret;
2301 DBG2("Consumer pid %d", consumer_data->pid);
2302 pthread_mutex_unlock(&consumer_data->pid_mutex);
2303
2304 DBG2("Spawning consumer control thread");
2305 ret = spawn_consumer_thread(consumer_data);
2306 if (ret < 0) {
2307 ERR("Fatal error spawning consumer control thread");
2308 goto error;
2309 }
2310
2311 end:
2312 return 0;
2313
2314 error:
2315 /* Cleanup already created sockets on error. */
2316 if (consumer_data->err_sock >= 0) {
2317 int err;
2318
2319 err = close(consumer_data->err_sock);
2320 if (err < 0) {
2321 PERROR("close consumer data error socket");
2322 }
2323 }
2324 return ret;
2325 }
2326
2327 /*
2328 * Setup necessary data for kernel tracer action.
2329 */
2330 static int init_kernel_tracer(void)
2331 {
2332 int ret;
2333
2334 /* Modprobe lttng kernel modules */
2335 ret = modprobe_lttng_control();
2336 if (ret < 0) {
2337 goto error;
2338 }
2339
2340 /* Open debugfs lttng */
2341 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2342 if (kernel_tracer_fd < 0) {
2343 DBG("Failed to open %s", module_proc_lttng);
2344 ret = -1;
2345 goto error_open;
2346 }
2347
2348 /* Validate kernel version */
2349 ret = kernel_validate_version(kernel_tracer_fd);
2350 if (ret < 0) {
2351 goto error_version;
2352 }
2353
2354 ret = modprobe_lttng_data();
2355 if (ret < 0) {
2356 goto error_modules;
2357 }
2358
2359 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2360 return 0;
2361
2362 error_version:
2363 modprobe_remove_lttng_control();
2364 ret = close(kernel_tracer_fd);
2365 if (ret) {
2366 PERROR("close");
2367 }
2368 kernel_tracer_fd = -1;
2369 return LTTNG_ERR_KERN_VERSION;
2370
2371 error_modules:
2372 ret = close(kernel_tracer_fd);
2373 if (ret) {
2374 PERROR("close");
2375 }
2376
2377 error_open:
2378 modprobe_remove_lttng_control();
2379
2380 error:
2381 WARN("No kernel tracer available");
2382 kernel_tracer_fd = -1;
2383 if (!is_root) {
2384 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2385 } else {
2386 return LTTNG_ERR_KERN_NA;
2387 }
2388 }
2389
2390
2391 /*
2392 * Copy consumer output from the tracing session to the domain session. The
2393 * function also applies the right modification on a per domain basis for the
2394 * trace files destination directory.
2395 *
2396 * Should *NOT* be called with RCU read-side lock held.
2397 */
2398 static int copy_session_consumer(int domain, struct ltt_session *session)
2399 {
2400 int ret;
2401 const char *dir_name;
2402 struct consumer_output *consumer;
2403
2404 assert(session);
2405 assert(session->consumer);
2406
2407 switch (domain) {
2408 case LTTNG_DOMAIN_KERNEL:
2409 DBG3("Copying tracing session consumer output in kernel session");
2410 /*
2411 * XXX: We should audit the session creation and what this function
2412 * does "extra" in order to avoid a destroy since this function is used
2413 * in the domain session creation (kernel and ust) only. Same for UST
2414 * domain.
2415 */
2416 if (session->kernel_session->consumer) {
2417 consumer_destroy_output(session->kernel_session->consumer);
2418 }
2419 session->kernel_session->consumer =
2420 consumer_copy_output(session->consumer);
2421 /* Ease our life a bit for the next part */
2422 consumer = session->kernel_session->consumer;
2423 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2424 break;
2425 case LTTNG_DOMAIN_JUL:
2426 case LTTNG_DOMAIN_UST:
2427 DBG3("Copying tracing session consumer output in UST session");
2428 if (session->ust_session->consumer) {
2429 consumer_destroy_output(session->ust_session->consumer);
2430 }
2431 session->ust_session->consumer =
2432 consumer_copy_output(session->consumer);
2433 /* Ease our life a bit for the next part */
2434 consumer = session->ust_session->consumer;
2435 dir_name = DEFAULT_UST_TRACE_DIR;
2436 break;
2437 default:
2438 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2439 goto error;
2440 }
2441
2442 /* Append correct directory to subdir */
2443 strncat(consumer->subdir, dir_name,
2444 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2445 DBG3("Copy session consumer subdir %s", consumer->subdir);
2446
2447 ret = LTTNG_OK;
2448
2449 error:
2450 return ret;
2451 }
2452
2453 /*
2454 * Create an UST session and add it to the session ust list.
2455 *
2456 * Should *NOT* be called with RCU read-side lock held.
2457 */
2458 static int create_ust_session(struct ltt_session *session,
2459 struct lttng_domain *domain)
2460 {
2461 int ret;
2462 struct ltt_ust_session *lus = NULL;
2463
2464 assert(session);
2465 assert(domain);
2466 assert(session->consumer);
2467
2468 switch (domain->type) {
2469 case LTTNG_DOMAIN_JUL:
2470 case LTTNG_DOMAIN_UST:
2471 break;
2472 default:
2473 ERR("Unknown UST domain on create session %d", domain->type);
2474 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2475 goto error;
2476 }
2477
2478 DBG("Creating UST session");
2479
2480 lus = trace_ust_create_session(session->id);
2481 if (lus == NULL) {
2482 ret = LTTNG_ERR_UST_SESS_FAIL;
2483 goto error;
2484 }
2485
2486 lus->uid = session->uid;
2487 lus->gid = session->gid;
2488 lus->output_traces = session->output_traces;
2489 lus->snapshot_mode = session->snapshot_mode;
2490 lus->live_timer_interval = session->live_timer;
2491 session->ust_session = lus;
2492
2493 /* Copy session output to the newly created UST session */
2494 ret = copy_session_consumer(domain->type, session);
2495 if (ret != LTTNG_OK) {
2496 goto error;
2497 }
2498
2499 return LTTNG_OK;
2500
2501 error:
2502 free(lus);
2503 session->ust_session = NULL;
2504 return ret;
2505 }
2506
2507 /*
2508 * Create a kernel tracer session then create the default channel.
2509 */
2510 static int create_kernel_session(struct ltt_session *session)
2511 {
2512 int ret;
2513
2514 DBG("Creating kernel session");
2515
2516 ret = kernel_create_session(session, kernel_tracer_fd);
2517 if (ret < 0) {
2518 ret = LTTNG_ERR_KERN_SESS_FAIL;
2519 goto error;
2520 }
2521
2522 /* Code flow safety */
2523 assert(session->kernel_session);
2524
2525 /* Copy session output to the newly created Kernel session */
2526 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2527 if (ret != LTTNG_OK) {
2528 goto error;
2529 }
2530
2531 /* Create directory(ies) on local filesystem. */
2532 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2533 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2534 ret = run_as_mkdir_recursive(
2535 session->kernel_session->consumer->dst.trace_path,
2536 S_IRWXU | S_IRWXG, session->uid, session->gid);
2537 if (ret < 0) {
2538 if (ret != -EEXIST) {
2539 ERR("Trace directory creation error");
2540 goto error;
2541 }
2542 }
2543 }
2544
2545 session->kernel_session->uid = session->uid;
2546 session->kernel_session->gid = session->gid;
2547 session->kernel_session->output_traces = session->output_traces;
2548 session->kernel_session->snapshot_mode = session->snapshot_mode;
2549
2550 return LTTNG_OK;
2551
2552 error:
2553 trace_kernel_destroy_session(session->kernel_session);
2554 session->kernel_session = NULL;
2555 return ret;
2556 }
2557
2558 /*
2559 * Count number of session permitted by uid/gid.
2560 */
2561 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2562 {
2563 unsigned int i = 0;
2564 struct ltt_session *session;
2565
2566 DBG("Counting number of available session for UID %d GID %d",
2567 uid, gid);
2568 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2569 /*
2570 * Only list the sessions the user can control.
2571 */
2572 if (!session_access_ok(session, uid, gid)) {
2573 continue;
2574 }
2575 i++;
2576 }
2577 return i;
2578 }
2579
2580 /*
2581 * Process the command requested by the lttng client within the command
2582 * context structure. This function make sure that the return structure (llm)
2583 * is set and ready for transmission before returning.
2584 *
2585 * Return any error encountered or 0 for success.
2586 *
2587 * "sock" is only used for special-case var. len data.
2588 *
2589 * Should *NOT* be called with RCU read-side lock held.
2590 */
2591 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2592 int *sock_error)
2593 {
2594 int ret = LTTNG_OK;
2595 int need_tracing_session = 1;
2596 int need_domain;
2597
2598 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2599
2600 *sock_error = 0;
2601
2602 switch (cmd_ctx->lsm->cmd_type) {
2603 case LTTNG_CREATE_SESSION:
2604 case LTTNG_CREATE_SESSION_SNAPSHOT:
2605 case LTTNG_CREATE_SESSION_LIVE:
2606 case LTTNG_DESTROY_SESSION:
2607 case LTTNG_LIST_SESSIONS:
2608 case LTTNG_LIST_DOMAINS:
2609 case LTTNG_START_TRACE:
2610 case LTTNG_STOP_TRACE:
2611 case LTTNG_DATA_PENDING:
2612 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2613 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2614 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2615 case LTTNG_SNAPSHOT_RECORD:
2616 need_domain = 0;
2617 break;
2618 default:
2619 need_domain = 1;
2620 }
2621
2622 if (opt_no_kernel && need_domain
2623 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2624 if (!is_root) {
2625 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2626 } else {
2627 ret = LTTNG_ERR_KERN_NA;
2628 }
2629 goto error;
2630 }
2631
2632 /* Deny register consumer if we already have a spawned consumer. */
2633 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2634 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2635 if (kconsumer_data.pid > 0) {
2636 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2637 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2638 goto error;
2639 }
2640 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2641 }
2642
2643 /*
2644 * Check for command that don't needs to allocate a returned payload. We do
2645 * this here so we don't have to make the call for no payload at each
2646 * command.
2647 */
2648 switch(cmd_ctx->lsm->cmd_type) {
2649 case LTTNG_LIST_SESSIONS:
2650 case LTTNG_LIST_TRACEPOINTS:
2651 case LTTNG_LIST_TRACEPOINT_FIELDS:
2652 case LTTNG_LIST_DOMAINS:
2653 case LTTNG_LIST_CHANNELS:
2654 case LTTNG_LIST_EVENTS:
2655 break;
2656 default:
2657 /* Setup lttng message with no payload */
2658 ret = setup_lttng_msg(cmd_ctx, 0);
2659 if (ret < 0) {
2660 /* This label does not try to unlock the session */
2661 goto init_setup_error;
2662 }
2663 }
2664
2665 /* Commands that DO NOT need a session. */
2666 switch (cmd_ctx->lsm->cmd_type) {
2667 case LTTNG_CREATE_SESSION:
2668 case LTTNG_CREATE_SESSION_SNAPSHOT:
2669 case LTTNG_CREATE_SESSION_LIVE:
2670 case LTTNG_CALIBRATE:
2671 case LTTNG_LIST_SESSIONS:
2672 case LTTNG_LIST_TRACEPOINTS:
2673 case LTTNG_LIST_TRACEPOINT_FIELDS:
2674 need_tracing_session = 0;
2675 break;
2676 default:
2677 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2678 /*
2679 * We keep the session list lock across _all_ commands
2680 * for now, because the per-session lock does not
2681 * handle teardown properly.
2682 */
2683 session_lock_list();
2684 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2685 if (cmd_ctx->session == NULL) {
2686 ret = LTTNG_ERR_SESS_NOT_FOUND;
2687 goto error;
2688 } else {
2689 /* Acquire lock for the session */
2690 session_lock(cmd_ctx->session);
2691 }
2692 break;
2693 }
2694
2695 if (!need_domain) {
2696 goto skip_domain;
2697 }
2698
2699 /*
2700 * Check domain type for specific "pre-action".
2701 */
2702 switch (cmd_ctx->lsm->domain.type) {
2703 case LTTNG_DOMAIN_KERNEL:
2704 if (!is_root) {
2705 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2706 goto error;
2707 }
2708
2709 /* Kernel tracer check */
2710 if (kernel_tracer_fd == -1) {
2711 /* Basically, load kernel tracer modules */
2712 ret = init_kernel_tracer();
2713 if (ret != 0) {
2714 goto error;
2715 }
2716 }
2717
2718 /* Consumer is in an ERROR state. Report back to client */
2719 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2720 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2721 goto error;
2722 }
2723
2724 /* Need a session for kernel command */
2725 if (need_tracing_session) {
2726 if (cmd_ctx->session->kernel_session == NULL) {
2727 ret = create_kernel_session(cmd_ctx->session);
2728 if (ret < 0) {
2729 ret = LTTNG_ERR_KERN_SESS_FAIL;
2730 goto error;
2731 }
2732 }
2733
2734 /* Start the kernel consumer daemon */
2735 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2736 if (kconsumer_data.pid == 0 &&
2737 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2738 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2739 ret = start_consumerd(&kconsumer_data);
2740 if (ret < 0) {
2741 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2742 goto error;
2743 }
2744 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2745 } else {
2746 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2747 }
2748
2749 /*
2750 * The consumer was just spawned so we need to add the socket to
2751 * the consumer output of the session if exist.
2752 */
2753 ret = consumer_create_socket(&kconsumer_data,
2754 cmd_ctx->session->kernel_session->consumer);
2755 if (ret < 0) {
2756 goto error;
2757 }
2758 }
2759
2760 break;
2761 case LTTNG_DOMAIN_JUL:
2762 case LTTNG_DOMAIN_UST:
2763 {
2764 if (!ust_app_supported()) {
2765 ret = LTTNG_ERR_NO_UST;
2766 goto error;
2767 }
2768 /* Consumer is in an ERROR state. Report back to client */
2769 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2770 ret = LTTNG_ERR_NO_USTCONSUMERD;
2771 goto error;
2772 }
2773
2774 if (need_tracing_session) {
2775 /* Create UST session if none exist. */
2776 if (cmd_ctx->session->ust_session == NULL) {
2777 ret = create_ust_session(cmd_ctx->session,
2778 &cmd_ctx->lsm->domain);
2779 if (ret != LTTNG_OK) {
2780 goto error;
2781 }
2782 }
2783
2784 /* Start the UST consumer daemons */
2785 /* 64-bit */
2786 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2787 if (consumerd64_bin[0] != '\0' &&
2788 ustconsumer64_data.pid == 0 &&
2789 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2790 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2791 ret = start_consumerd(&ustconsumer64_data);
2792 if (ret < 0) {
2793 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2794 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2795 goto error;
2796 }
2797
2798 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2799 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2800 } else {
2801 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2802 }
2803
2804 /*
2805 * Setup socket for consumer 64 bit. No need for atomic access
2806 * since it was set above and can ONLY be set in this thread.
2807 */
2808 ret = consumer_create_socket(&ustconsumer64_data,
2809 cmd_ctx->session->ust_session->consumer);
2810 if (ret < 0) {
2811 goto error;
2812 }
2813
2814 /* 32-bit */
2815 if (consumerd32_bin[0] != '\0' &&
2816 ustconsumer32_data.pid == 0 &&
2817 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2818 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2819 ret = start_consumerd(&ustconsumer32_data);
2820 if (ret < 0) {
2821 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2822 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2823 goto error;
2824 }
2825
2826 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2827 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2828 } else {
2829 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2830 }
2831
2832 /*
2833 * Setup socket for consumer 64 bit. No need for atomic access
2834 * since it was set above and can ONLY be set in this thread.
2835 */
2836 ret = consumer_create_socket(&ustconsumer32_data,
2837 cmd_ctx->session->ust_session->consumer);
2838 if (ret < 0) {
2839 goto error;
2840 }
2841 }
2842 break;
2843 }
2844 default:
2845 break;
2846 }
2847 skip_domain:
2848
2849 /* Validate consumer daemon state when start/stop trace command */
2850 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2851 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2852 switch (cmd_ctx->lsm->domain.type) {
2853 case LTTNG_DOMAIN_JUL:
2854 case LTTNG_DOMAIN_UST:
2855 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2856 ret = LTTNG_ERR_NO_USTCONSUMERD;
2857 goto error;
2858 }
2859 break;
2860 case LTTNG_DOMAIN_KERNEL:
2861 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2862 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2863 goto error;
2864 }
2865 break;
2866 }
2867 }
2868
2869 /*
2870 * Check that the UID or GID match that of the tracing session.
2871 * The root user can interact with all sessions.
2872 */
2873 if (need_tracing_session) {
2874 if (!session_access_ok(cmd_ctx->session,
2875 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2876 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2877 ret = LTTNG_ERR_EPERM;
2878 goto error;
2879 }
2880 }
2881
2882 /*
2883 * Send relayd information to consumer as soon as we have a domain and a
2884 * session defined.
2885 */
2886 if (cmd_ctx->session && need_domain) {
2887 /*
2888 * Setup relayd if not done yet. If the relayd information was already
2889 * sent to the consumer, this call will gracefully return.
2890 */
2891 ret = cmd_setup_relayd(cmd_ctx->session);
2892 if (ret != LTTNG_OK) {
2893 goto error;
2894 }
2895 }
2896
2897 /* Process by command type */
2898 switch (cmd_ctx->lsm->cmd_type) {
2899 case LTTNG_ADD_CONTEXT:
2900 {
2901 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2902 cmd_ctx->lsm->u.context.channel_name,
2903 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2904 break;
2905 }
2906 case LTTNG_DISABLE_CHANNEL:
2907 {
2908 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2909 cmd_ctx->lsm->u.disable.channel_name);
2910 break;
2911 }
2912 case LTTNG_DISABLE_EVENT:
2913 {
2914 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2915 cmd_ctx->lsm->u.disable.channel_name,
2916 cmd_ctx->lsm->u.disable.name);
2917 break;
2918 }
2919 case LTTNG_DISABLE_ALL_EVENT:
2920 {
2921 DBG("Disabling all events");
2922
2923 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2924 cmd_ctx->lsm->u.disable.channel_name);
2925 break;
2926 }
2927 case LTTNG_ENABLE_CHANNEL:
2928 {
2929 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2930 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2931 break;
2932 }
2933 case LTTNG_ENABLE_EVENT:
2934 {
2935 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2936 cmd_ctx->lsm->u.enable.channel_name,
2937 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2938 break;
2939 }
2940 case LTTNG_ENABLE_ALL_EVENT:
2941 {
2942 DBG("Enabling all events");
2943
2944 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2945 cmd_ctx->lsm->u.enable.channel_name,
2946 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2947 break;
2948 }
2949 case LTTNG_LIST_TRACEPOINTS:
2950 {
2951 struct lttng_event *events;
2952 ssize_t nb_events;
2953
2954 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2955 if (nb_events < 0) {
2956 /* Return value is a negative lttng_error_code. */
2957 ret = -nb_events;
2958 goto error;
2959 }
2960
2961 /*
2962 * Setup lttng message with payload size set to the event list size in
2963 * bytes and then copy list into the llm payload.
2964 */
2965 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2966 if (ret < 0) {
2967 free(events);
2968 goto setup_error;
2969 }
2970
2971 /* Copy event list into message payload */
2972 memcpy(cmd_ctx->llm->payload, events,
2973 sizeof(struct lttng_event) * nb_events);
2974
2975 free(events);
2976
2977 ret = LTTNG_OK;
2978 break;
2979 }
2980 case LTTNG_LIST_TRACEPOINT_FIELDS:
2981 {
2982 struct lttng_event_field *fields;
2983 ssize_t nb_fields;
2984
2985 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2986 &fields);
2987 if (nb_fields < 0) {
2988 /* Return value is a negative lttng_error_code. */
2989 ret = -nb_fields;
2990 goto error;
2991 }
2992
2993 /*
2994 * Setup lttng message with payload size set to the event list size in
2995 * bytes and then copy list into the llm payload.
2996 */
2997 ret = setup_lttng_msg(cmd_ctx,
2998 sizeof(struct lttng_event_field) * nb_fields);
2999 if (ret < 0) {
3000 free(fields);
3001 goto setup_error;
3002 }
3003
3004 /* Copy event list into message payload */
3005 memcpy(cmd_ctx->llm->payload, fields,
3006 sizeof(struct lttng_event_field) * nb_fields);
3007
3008 free(fields);
3009
3010 ret = LTTNG_OK;
3011 break;
3012 }
3013 case LTTNG_SET_CONSUMER_URI:
3014 {
3015 size_t nb_uri, len;
3016 struct lttng_uri *uris;
3017
3018 nb_uri = cmd_ctx->lsm->u.uri.size;
3019 len = nb_uri * sizeof(struct lttng_uri);
3020
3021 if (nb_uri == 0) {
3022 ret = LTTNG_ERR_INVALID;
3023 goto error;
3024 }
3025
3026 uris = zmalloc(len);
3027 if (uris == NULL) {
3028 ret = LTTNG_ERR_FATAL;
3029 goto error;
3030 }
3031
3032 /* Receive variable len data */
3033 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3034 ret = lttcomm_recv_unix_sock(sock, uris, len);
3035 if (ret <= 0) {
3036 DBG("No URIs received from client... continuing");
3037 *sock_error = 1;
3038 ret = LTTNG_ERR_SESSION_FAIL;
3039 free(uris);
3040 goto error;
3041 }
3042
3043 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3044 nb_uri, uris);
3045 if (ret != LTTNG_OK) {
3046 free(uris);
3047 goto error;
3048 }
3049
3050 /*
3051 * XXX: 0 means that this URI should be applied on the session. Should
3052 * be a DOMAIN enuam.
3053 */
3054 if (cmd_ctx->lsm->domain.type == 0) {
3055 /* Add the URI for the UST session if a consumer is present. */
3056 if (cmd_ctx->session->ust_session &&
3057 cmd_ctx->session->ust_session->consumer) {
3058 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3059 nb_uri, uris);
3060 } else if (cmd_ctx->session->kernel_session &&
3061 cmd_ctx->session->kernel_session->consumer) {
3062 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3063 cmd_ctx->session, nb_uri, uris);
3064 }
3065 }
3066
3067 free(uris);
3068
3069 break;
3070 }
3071 case LTTNG_START_TRACE:
3072 {
3073 ret = cmd_start_trace(cmd_ctx->session);
3074 break;
3075 }
3076 case LTTNG_STOP_TRACE:
3077 {
3078 ret = cmd_stop_trace(cmd_ctx->session);
3079 break;
3080 }
3081 case LTTNG_CREATE_SESSION:
3082 {
3083 size_t nb_uri, len;
3084 struct lttng_uri *uris = NULL;
3085
3086 nb_uri = cmd_ctx->lsm->u.uri.size;
3087 len = nb_uri * sizeof(struct lttng_uri);
3088
3089 if (nb_uri > 0) {
3090 uris = zmalloc(len);
3091 if (uris == NULL) {
3092 ret = LTTNG_ERR_FATAL;
3093 goto error;
3094 }
3095
3096 /* Receive variable len data */
3097 DBG("Waiting for %zu URIs from client ...", nb_uri);
3098 ret = lttcomm_recv_unix_sock(sock, uris, len);
3099 if (ret <= 0) {
3100 DBG("No URIs received from client... continuing");
3101 *sock_error = 1;
3102 ret = LTTNG_ERR_SESSION_FAIL;
3103 free(uris);
3104 goto error;
3105 }
3106
3107 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3108 DBG("Creating session with ONE network URI is a bad call");
3109 ret = LTTNG_ERR_SESSION_FAIL;
3110 free(uris);
3111 goto error;
3112 }
3113 }
3114
3115 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3116 &cmd_ctx->creds, 0);
3117
3118 free(uris);
3119
3120 break;
3121 }
3122 case LTTNG_DESTROY_SESSION:
3123 {
3124 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3125
3126 /* Set session to NULL so we do not unlock it after free. */
3127 cmd_ctx->session = NULL;
3128 break;
3129 }
3130 case LTTNG_LIST_DOMAINS:
3131 {
3132 ssize_t nb_dom;
3133 struct lttng_domain *domains;
3134
3135 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3136 if (nb_dom < 0) {
3137 /* Return value is a negative lttng_error_code. */
3138 ret = -nb_dom;
3139 goto error;
3140 }
3141
3142 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3143 if (ret < 0) {
3144 free(domains);
3145 goto setup_error;
3146 }
3147
3148 /* Copy event list into message payload */
3149 memcpy(cmd_ctx->llm->payload, domains,
3150 nb_dom * sizeof(struct lttng_domain));
3151
3152 free(domains);
3153
3154 ret = LTTNG_OK;
3155 break;
3156 }
3157 case LTTNG_LIST_CHANNELS:
3158 {
3159 int nb_chan;
3160 struct lttng_channel *channels;
3161
3162 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3163 cmd_ctx->session, &channels);
3164 if (nb_chan < 0) {
3165 /* Return value is a negative lttng_error_code. */
3166 ret = -nb_chan;
3167 goto error;
3168 }
3169
3170 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3171 if (ret < 0) {
3172 free(channels);
3173 goto setup_error;
3174 }
3175
3176 /* Copy event list into message payload */
3177 memcpy(cmd_ctx->llm->payload, channels,
3178 nb_chan * sizeof(struct lttng_channel));
3179
3180 free(channels);
3181
3182 ret = LTTNG_OK;
3183 break;
3184 }
3185 case LTTNG_LIST_EVENTS:
3186 {
3187 ssize_t nb_event;
3188 struct lttng_event *events = NULL;
3189
3190 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3191 cmd_ctx->lsm->u.list.channel_name, &events);
3192 if (nb_event < 0) {
3193 /* Return value is a negative lttng_error_code. */
3194 ret = -nb_event;
3195 goto error;
3196 }
3197
3198 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3199 if (ret < 0) {
3200 free(events);
3201 goto setup_error;
3202 }
3203
3204 /* Copy event list into message payload */
3205 memcpy(cmd_ctx->llm->payload, events,
3206 nb_event * sizeof(struct lttng_event));
3207
3208 free(events);
3209
3210 ret = LTTNG_OK;
3211 break;
3212 }
3213 case LTTNG_LIST_SESSIONS:
3214 {
3215 unsigned int nr_sessions;
3216
3217 session_lock_list();
3218 nr_sessions = lttng_sessions_count(
3219 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3220 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3221
3222 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3223 if (ret < 0) {
3224 session_unlock_list();
3225 goto setup_error;
3226 }
3227
3228 /* Filled the session array */
3229 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3230 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3231 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3232
3233 session_unlock_list();
3234
3235 ret = LTTNG_OK;
3236 break;
3237 }
3238 case LTTNG_CALIBRATE:
3239 {
3240 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3241 &cmd_ctx->lsm->u.calibrate);
3242 break;
3243 }
3244 case LTTNG_REGISTER_CONSUMER:
3245 {
3246 struct consumer_data *cdata;
3247
3248 switch (cmd_ctx->lsm->domain.type) {
3249 case LTTNG_DOMAIN_KERNEL:
3250 cdata = &kconsumer_data;
3251 break;
3252 default:
3253 ret = LTTNG_ERR_UND;
3254 goto error;
3255 }
3256
3257 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3258 cmd_ctx->lsm->u.reg.path, cdata);
3259 break;
3260 }
3261 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3262 {
3263 struct lttng_filter_bytecode *bytecode;
3264
3265 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3266 ret = LTTNG_ERR_FILTER_INVAL;
3267 goto error;
3268 }
3269 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3270 ret = LTTNG_ERR_FILTER_INVAL;
3271 goto error;
3272 }
3273 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3274 if (!bytecode) {
3275 ret = LTTNG_ERR_FILTER_NOMEM;
3276 goto error;
3277 }
3278 /* Receive var. len. data */
3279 DBG("Receiving var len data from client ...");
3280 ret = lttcomm_recv_unix_sock(sock, bytecode,
3281 cmd_ctx->lsm->u.enable.bytecode_len);
3282 if (ret <= 0) {
3283 DBG("Nothing recv() from client var len data... continuing");
3284 *sock_error = 1;
3285 ret = LTTNG_ERR_FILTER_INVAL;
3286 goto error;
3287 }
3288
3289 if (bytecode->len + sizeof(*bytecode)
3290 != cmd_ctx->lsm->u.enable.bytecode_len) {
3291 free(bytecode);
3292 ret = LTTNG_ERR_FILTER_INVAL;
3293 goto error;
3294 }
3295
3296 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3297 cmd_ctx->lsm->u.enable.channel_name,
3298 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3299 break;
3300 }
3301 case LTTNG_DATA_PENDING:
3302 {
3303 ret = cmd_data_pending(cmd_ctx->session);
3304 break;
3305 }
3306 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3307 {
3308 struct lttcomm_lttng_output_id reply;
3309
3310 ret = cmd_snapshot_add_output(cmd_ctx->session,
3311 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3312 if (ret != LTTNG_OK) {
3313 goto error;
3314 }
3315
3316 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3317 if (ret < 0) {
3318 goto setup_error;
3319 }
3320
3321 /* Copy output list into message payload */
3322 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3323 ret = LTTNG_OK;
3324 break;
3325 }
3326 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3327 {
3328 ret = cmd_snapshot_del_output(cmd_ctx->session,
3329 &cmd_ctx->lsm->u.snapshot_output.output);
3330 break;
3331 }
3332 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3333 {
3334 ssize_t nb_output;
3335 struct lttng_snapshot_output *outputs = NULL;
3336
3337 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3338 if (nb_output < 0) {
3339 ret = -nb_output;
3340 goto error;
3341 }
3342
3343 ret = setup_lttng_msg(cmd_ctx,
3344 nb_output * sizeof(struct lttng_snapshot_output));
3345 if (ret < 0) {
3346 free(outputs);
3347 goto setup_error;
3348 }
3349
3350 if (outputs) {
3351 /* Copy output list into message payload */
3352 memcpy(cmd_ctx->llm->payload, outputs,
3353 nb_output * sizeof(struct lttng_snapshot_output));
3354 free(outputs);
3355 }
3356
3357 ret = LTTNG_OK;
3358 break;
3359 }
3360 case LTTNG_SNAPSHOT_RECORD:
3361 {
3362 ret = cmd_snapshot_record(cmd_ctx->session,
3363 &cmd_ctx->lsm->u.snapshot_record.output,
3364 cmd_ctx->lsm->u.snapshot_record.wait);
3365 break;
3366 }
3367 case LTTNG_CREATE_SESSION_SNAPSHOT:
3368 {
3369 size_t nb_uri, len;
3370 struct lttng_uri *uris = NULL;
3371
3372 nb_uri = cmd_ctx->lsm->u.uri.size;
3373 len = nb_uri * sizeof(struct lttng_uri);
3374
3375 if (nb_uri > 0) {
3376 uris = zmalloc(len);
3377 if (uris == NULL) {
3378 ret = LTTNG_ERR_FATAL;
3379 goto error;
3380 }
3381
3382 /* Receive variable len data */
3383 DBG("Waiting for %zu URIs from client ...", nb_uri);
3384 ret = lttcomm_recv_unix_sock(sock, uris, len);
3385 if (ret <= 0) {
3386 DBG("No URIs received from client... continuing");
3387 *sock_error = 1;
3388 ret = LTTNG_ERR_SESSION_FAIL;
3389 free(uris);
3390 goto error;
3391 }
3392
3393 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3394 DBG("Creating session with ONE network URI is a bad call");
3395 ret = LTTNG_ERR_SESSION_FAIL;
3396 free(uris);
3397 goto error;
3398 }
3399 }
3400
3401 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3402 nb_uri, &cmd_ctx->creds);
3403 free(uris);
3404 break;
3405 }
3406 case LTTNG_CREATE_SESSION_LIVE:
3407 {
3408 size_t nb_uri, len;
3409 struct lttng_uri *uris = NULL;
3410
3411 nb_uri = cmd_ctx->lsm->u.uri.size;
3412 len = nb_uri * sizeof(struct lttng_uri);
3413
3414 if (nb_uri > 0) {
3415 uris = zmalloc(len);
3416 if (uris == NULL) {
3417 ret = LTTNG_ERR_FATAL;
3418 goto error;
3419 }
3420
3421 /* Receive variable len data */
3422 DBG("Waiting for %zu URIs from client ...", nb_uri);
3423 ret = lttcomm_recv_unix_sock(sock, uris, len);
3424 if (ret <= 0) {
3425 DBG("No URIs received from client... continuing");
3426 *sock_error = 1;
3427 ret = LTTNG_ERR_SESSION_FAIL;
3428 free(uris);
3429 goto error;
3430 }
3431
3432 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3433 DBG("Creating session with ONE network URI is a bad call");
3434 ret = LTTNG_ERR_SESSION_FAIL;
3435 free(uris);
3436 goto error;
3437 }
3438 }
3439
3440 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3441 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3442 free(uris);
3443 break;
3444 }
3445 default:
3446 ret = LTTNG_ERR_UND;
3447 break;
3448 }
3449
3450 error:
3451 if (cmd_ctx->llm == NULL) {
3452 DBG("Missing llm structure. Allocating one.");
3453 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3454 goto setup_error;
3455 }
3456 }
3457 /* Set return code */
3458 cmd_ctx->llm->ret_code = ret;
3459 setup_error:
3460 if (cmd_ctx->session) {
3461 session_unlock(cmd_ctx->session);
3462 }
3463 if (need_tracing_session) {
3464 session_unlock_list();
3465 }
3466 init_setup_error:
3467 return ret;
3468 }
3469
3470 /*
3471 * Thread managing health check socket.
3472 */
3473 static void *thread_manage_health(void *data)
3474 {
3475 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3476 uint32_t revents, nb_fd;
3477 struct lttng_poll_event events;
3478 struct health_comm_msg msg;
3479 struct health_comm_reply reply;
3480
3481 DBG("[thread] Manage health check started");
3482
3483 rcu_register_thread();
3484
3485 /* We might hit an error path before this is created. */
3486 lttng_poll_init(&events);
3487
3488 /* Create unix socket */
3489 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3490 if (sock < 0) {
3491 ERR("Unable to create health check Unix socket");
3492 ret = -1;
3493 goto error;
3494 }
3495
3496 if (is_root) {
3497 /* lttng health client socket path permissions */
3498 ret = chown(health_unix_sock_path, 0,
3499 utils_get_group_id(tracing_group_name));
3500 if (ret < 0) {
3501 ERR("Unable to set group on %s", health_unix_sock_path);
3502 PERROR("chown");
3503 ret = -1;
3504 goto error;
3505 }
3506
3507 ret = chmod(health_unix_sock_path,
3508 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3509 if (ret < 0) {
3510 ERR("Unable to set permissions on %s", health_unix_sock_path);
3511 PERROR("chmod");
3512 ret = -1;
3513 goto error;
3514 }
3515 }
3516
3517 /*
3518 * Set the CLOEXEC flag. Return code is useless because either way, the
3519 * show must go on.
3520 */
3521 (void) utils_set_fd_cloexec(sock);
3522
3523 ret = lttcomm_listen_unix_sock(sock);
3524 if (ret < 0) {
3525 goto error;
3526 }
3527
3528 /*
3529 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3530 * more will be added to this poll set.
3531 */
3532 ret = sessiond_set_thread_pollset(&events, 2);
3533 if (ret < 0) {
3534 goto error;
3535 }
3536
3537 /* Add the application registration socket */
3538 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3539 if (ret < 0) {
3540 goto error;
3541 }
3542
3543 while (1) {
3544 DBG("Health check ready");
3545
3546 /* Inifinite blocking call, waiting for transmission */
3547 restart:
3548 ret = lttng_poll_wait(&events, -1);
3549 if (ret < 0) {
3550 /*
3551 * Restart interrupted system call.
3552 */
3553 if (errno == EINTR) {
3554 goto restart;
3555 }
3556 goto error;
3557 }
3558
3559 nb_fd = ret;
3560
3561 for (i = 0; i < nb_fd; i++) {
3562 /* Fetch once the poll data */
3563 revents = LTTNG_POLL_GETEV(&events, i);
3564 pollfd = LTTNG_POLL_GETFD(&events, i);
3565
3566 /* Thread quit pipe has been closed. Killing thread. */
3567 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3568 if (ret) {
3569 err = 0;
3570 goto exit;
3571 }
3572
3573 /* Event on the registration socket */
3574 if (pollfd == sock) {
3575 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3576 ERR("Health socket poll error");
3577 goto error;
3578 }
3579 }
3580 }
3581
3582 new_sock = lttcomm_accept_unix_sock(sock);
3583 if (new_sock < 0) {
3584 goto error;
3585 }
3586
3587 /*
3588 * Set the CLOEXEC flag. Return code is useless because either way, the
3589 * show must go on.
3590 */
3591 (void) utils_set_fd_cloexec(new_sock);
3592
3593 DBG("Receiving data from client for health...");
3594 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3595 if (ret <= 0) {
3596 DBG("Nothing recv() from client... continuing");
3597 ret = close(new_sock);
3598 if (ret) {
3599 PERROR("close");
3600 }
3601 new_sock = -1;
3602 continue;
3603 }
3604
3605 rcu_thread_online();
3606
3607 reply.ret_code = 0;
3608 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3609 /*
3610 * health_check_state returns 0 if health is
3611 * bad.
3612 */
3613 if (!health_check_state(health_sessiond, i)) {
3614 reply.ret_code |= 1ULL << i;
3615 }
3616 }
3617
3618 DBG2("Health check return value %" PRIx64, reply.ret_code);
3619
3620 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3621 if (ret < 0) {
3622 ERR("Failed to send health data back to client");
3623 }
3624
3625 /* End of transmission */
3626 ret = close(new_sock);
3627 if (ret) {
3628 PERROR("close");
3629 }
3630 new_sock = -1;
3631 }
3632
3633 exit:
3634 error:
3635 if (err) {
3636 ERR("Health error occurred in %s", __func__);
3637 }
3638 DBG("Health check thread dying");
3639 unlink(health_unix_sock_path);
3640 if (sock >= 0) {
3641 ret = close(sock);
3642 if (ret) {
3643 PERROR("close");
3644 }
3645 }
3646
3647 lttng_poll_clean(&events);
3648
3649 rcu_unregister_thread();
3650 return NULL;
3651 }
3652
3653 /*
3654 * This thread manage all clients request using the unix client socket for
3655 * communication.
3656 */
3657 static void *thread_manage_clients(void *data)
3658 {
3659 int sock = -1, ret, i, pollfd, err = -1;
3660 int sock_error;
3661 uint32_t revents, nb_fd;
3662 struct command_ctx *cmd_ctx = NULL;
3663 struct lttng_poll_event events;
3664
3665 DBG("[thread] Manage client started");
3666
3667 rcu_register_thread();
3668
3669 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3670
3671 if (testpoint(thread_manage_clients)) {
3672 goto error_testpoint;
3673 }
3674
3675 health_code_update();
3676
3677 ret = lttcomm_listen_unix_sock(client_sock);
3678 if (ret < 0) {
3679 goto error_listen;
3680 }
3681
3682 /*
3683 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3684 * more will be added to this poll set.
3685 */
3686 ret = sessiond_set_thread_pollset(&events, 2);
3687 if (ret < 0) {
3688 goto error_create_poll;
3689 }
3690
3691 /* Add the application registration socket */
3692 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3693 if (ret < 0) {
3694 goto error;
3695 }
3696
3697 /*
3698 * Notify parent pid that we are ready to accept command for client side.
3699 */
3700 if (opt_sig_parent) {
3701 kill(ppid, SIGUSR1);
3702 }
3703
3704 if (testpoint(thread_manage_clients_before_loop)) {
3705 goto error;
3706 }
3707
3708 health_code_update();
3709
3710 while (1) {
3711 DBG("Accepting client command ...");
3712
3713 /* Inifinite blocking call, waiting for transmission */
3714 restart:
3715 health_poll_entry();
3716 ret = lttng_poll_wait(&events, -1);
3717 health_poll_exit();
3718 if (ret < 0) {
3719 /*
3720 * Restart interrupted system call.
3721 */
3722 if (errno == EINTR) {
3723 goto restart;
3724 }
3725 goto error;
3726 }
3727
3728 nb_fd = ret;
3729
3730 for (i = 0; i < nb_fd; i++) {
3731 /* Fetch once the poll data */
3732 revents = LTTNG_POLL_GETEV(&events, i);
3733 pollfd = LTTNG_POLL_GETFD(&events, i);
3734
3735 health_code_update();
3736
3737 /* Thread quit pipe has been closed. Killing thread. */
3738 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3739 if (ret) {
3740 err = 0;
3741 goto exit;
3742 }
3743
3744 /* Event on the registration socket */
3745 if (pollfd == client_sock) {
3746 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3747 ERR("Client socket poll error");
3748 goto error;
3749 }
3750 }
3751 }
3752
3753 DBG("Wait for client response");
3754
3755 health_code_update();
3756
3757 sock = lttcomm_accept_unix_sock(client_sock);
3758 if (sock < 0) {
3759 goto error;
3760 }
3761
3762 /*
3763 * Set the CLOEXEC flag. Return code is useless because either way, the
3764 * show must go on.
3765 */
3766 (void) utils_set_fd_cloexec(sock);
3767
3768 /* Set socket option for credentials retrieval */
3769 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3770 if (ret < 0) {
3771 goto error;
3772 }
3773
3774 /* Allocate context command to process the client request */
3775 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3776 if (cmd_ctx == NULL) {
3777 PERROR("zmalloc cmd_ctx");
3778 goto error;
3779 }
3780
3781 /* Allocate data buffer for reception */
3782 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3783 if (cmd_ctx->lsm == NULL) {
3784 PERROR("zmalloc cmd_ctx->lsm");
3785 goto error;
3786 }
3787
3788 cmd_ctx->llm = NULL;
3789 cmd_ctx->session = NULL;
3790
3791 health_code_update();
3792
3793 /*
3794 * Data is received from the lttng client. The struct
3795 * lttcomm_session_msg (lsm) contains the command and data request of
3796 * the client.
3797 */
3798 DBG("Receiving data from client ...");
3799 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3800 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3801 if (ret <= 0) {
3802 DBG("Nothing recv() from client... continuing");
3803 ret = close(sock);
3804 if (ret) {
3805 PERROR("close");
3806 }
3807 sock = -1;
3808 clean_command_ctx(&cmd_ctx);
3809 continue;
3810 }
3811
3812 health_code_update();
3813
3814 // TODO: Validate cmd_ctx including sanity check for
3815 // security purpose.
3816
3817 rcu_thread_online();
3818 /*
3819 * This function dispatch the work to the kernel or userspace tracer
3820 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3821 * informations for the client. The command context struct contains
3822 * everything this function may needs.
3823 */
3824 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3825 rcu_thread_offline();
3826 if (ret < 0) {
3827 ret = close(sock);
3828 if (ret) {
3829 PERROR("close");
3830 }
3831 sock = -1;
3832 /*
3833 * TODO: Inform client somehow of the fatal error. At
3834 * this point, ret < 0 means that a zmalloc failed
3835 * (ENOMEM). Error detected but still accept
3836 * command, unless a socket error has been
3837 * detected.
3838 */
3839 clean_command_ctx(&cmd_ctx);
3840 continue;
3841 }
3842
3843 health_code_update();
3844
3845 DBG("Sending response (size: %d, retcode: %s)",
3846 cmd_ctx->lttng_msg_size,
3847 lttng_strerror(-cmd_ctx->llm->ret_code));
3848 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3849 if (ret < 0) {
3850 ERR("Failed to send data back to client");
3851 }
3852
3853 /* End of transmission */
3854 ret = close(sock);
3855 if (ret) {
3856 PERROR("close");
3857 }
3858 sock = -1;
3859
3860 clean_command_ctx(&cmd_ctx);
3861
3862 health_code_update();
3863 }
3864
3865 exit:
3866 error:
3867 if (sock >= 0) {
3868 ret = close(sock);
3869 if (ret) {
3870 PERROR("close");
3871 }
3872 }
3873
3874 lttng_poll_clean(&events);
3875 clean_command_ctx(&cmd_ctx);
3876
3877 error_listen:
3878 error_create_poll:
3879 error_testpoint:
3880 unlink(client_unix_sock_path);
3881 if (client_sock >= 0) {
3882 ret = close(client_sock);
3883 if (ret) {
3884 PERROR("close");
3885 }
3886 }
3887
3888 if (err) {
3889 health_error();
3890 ERR("Health error occurred in %s", __func__);
3891 }
3892
3893 health_unregister(health_sessiond);
3894
3895 DBG("Client thread dying");
3896
3897 rcu_unregister_thread();
3898 return NULL;
3899 }
3900
3901
3902 /*
3903 * usage function on stderr
3904 */
3905 static void usage(void)
3906 {
3907 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3908 fprintf(stderr, " -h, --help Display this usage.\n");
3909 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3910 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3911 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3912 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3913 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3914 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3915 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3916 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3917 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3918 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3919 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3920 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3921 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3922 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3923 fprintf(stderr, " -V, --version Show version number.\n");
3924 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3925 fprintf(stderr, " -q, --quiet No output at all.\n");
3926 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3927 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3928 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3929 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3930 fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
3931 }
3932
3933 /*
3934 * daemon argument parsing
3935 */
3936 static int parse_args(int argc, char **argv)
3937 {
3938 int c;
3939
3940 static struct option long_options[] = {
3941 { "client-sock", 1, 0, 'c' },
3942 { "apps-sock", 1, 0, 'a' },
3943 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3944 { "kconsumerd-err-sock", 1, 0, 'E' },
3945 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3946 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3947 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3948 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3949 { "consumerd32-path", 1, 0, 'u' },
3950 { "consumerd32-libdir", 1, 0, 'U' },
3951 { "consumerd64-path", 1, 0, 't' },
3952 { "consumerd64-libdir", 1, 0, 'T' },
3953 { "daemonize", 0, 0, 'd' },
3954 { "sig-parent", 0, 0, 'S' },
3955 { "help", 0, 0, 'h' },
3956 { "group", 1, 0, 'g' },
3957 { "version", 0, 0, 'V' },
3958 { "quiet", 0, 0, 'q' },
3959 { "verbose", 0, 0, 'v' },
3960 { "verbose-consumer", 0, 0, 'Z' },
3961 { "no-kernel", 0, 0, 'N' },
3962 { "pidfile", 1, 0, 'p' },
3963 { "jul-tcp-port", 1, 0, 'J' },
3964 { NULL, 0, 0, 0 }
3965 };
3966
3967 while (1) {
3968 int option_index = 0;
3969 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
3970 long_options, &option_index);
3971 if (c == -1) {
3972 break;
3973 }
3974
3975 switch (c) {
3976 case 0:
3977 fprintf(stderr, "option %s", long_options[option_index].name);
3978 if (optarg) {
3979 fprintf(stderr, " with arg %s\n", optarg);
3980 }
3981 break;
3982 case 'c':
3983 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3984 break;
3985 case 'a':
3986 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3987 break;
3988 case 'd':
3989 opt_daemon = 1;
3990 break;
3991 case 'g':
3992 tracing_group_name = optarg;
3993 break;
3994 case 'h':
3995 usage();
3996 exit(EXIT_FAILURE);
3997 case 'V':
3998 fprintf(stdout, "%s\n", VERSION);
3999 exit(EXIT_SUCCESS);
4000 case 'S':
4001 opt_sig_parent = 1;
4002 break;
4003 case 'E':
4004 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4005 break;
4006 case 'C':
4007 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4008 break;
4009 case 'F':
4010 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4011 break;
4012 case 'D':
4013 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4014 break;
4015 case 'H':
4016 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4017 break;
4018 case 'G':
4019 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4020 break;
4021 case 'N':
4022 opt_no_kernel = 1;
4023 break;
4024 case 'q':
4025 lttng_opt_quiet = 1;
4026 break;
4027 case 'v':
4028 /* Verbose level can increase using multiple -v */
4029 lttng_opt_verbose += 1;
4030 break;
4031 case 'Z':
4032 opt_verbose_consumer += 1;
4033 break;
4034 case 'u':
4035 consumerd32_bin= optarg;
4036 break;
4037 case 'U':
4038 consumerd32_libdir = optarg;
4039 break;
4040 case 't':
4041 consumerd64_bin = optarg;
4042 break;
4043 case 'T':
4044 consumerd64_libdir = optarg;
4045 break;
4046 case 'p':
4047 opt_pidfile = optarg;
4048 break;
4049 case 'J': /* JUL TCP port. */
4050 {
4051 unsigned long v;
4052
4053 errno = 0;
4054 v = strtoul(optarg, NULL, 0);
4055 if (errno != 0 || !isdigit(optarg[0])) {
4056 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg);
4057 return -1;
4058 }
4059 if (v == 0 || v >= 65535) {
4060 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg);
4061 return -1;
4062 }
4063 jul_tcp_port = (uint32_t) v;
4064 DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
4065 break;
4066 }
4067 default:
4068 /* Unknown option or other error.
4069 * Error is printed by getopt, just return */
4070 return -1;
4071 }
4072 }
4073
4074 return 0;
4075 }
4076
4077 /*
4078 * Creates the two needed socket by the daemon.
4079 * apps_sock - The communication socket for all UST apps.
4080 * client_sock - The communication of the cli tool (lttng).
4081 */
4082 static int init_daemon_socket(void)
4083 {
4084 int ret = 0;
4085 mode_t old_umask;
4086
4087 old_umask = umask(0);
4088
4089 /* Create client tool unix socket */
4090 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4091 if (client_sock < 0) {
4092 ERR("Create unix sock failed: %s", client_unix_sock_path);
4093 ret = -1;
4094 goto end;
4095 }
4096
4097 /* Set the cloexec flag */
4098 ret = utils_set_fd_cloexec(client_sock);
4099 if (ret < 0) {
4100 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4101 "Continuing but note that the consumer daemon will have a "
4102 "reference to this socket on exec()", client_sock);
4103 }
4104
4105 /* File permission MUST be 660 */
4106 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4107 if (ret < 0) {
4108 ERR("Set file permissions failed: %s", client_unix_sock_path);
4109 PERROR("chmod");
4110 goto end;
4111 }
4112
4113 /* Create the application unix socket */
4114 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4115 if (apps_sock < 0) {
4116 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4117 ret = -1;
4118 goto end;
4119 }
4120
4121 /* Set the cloexec flag */
4122 ret = utils_set_fd_cloexec(apps_sock);
4123 if (ret < 0) {
4124 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4125 "Continuing but note that the consumer daemon will have a "
4126 "reference to this socket on exec()", apps_sock);
4127 }
4128
4129 /* File permission MUST be 666 */
4130 ret = chmod(apps_unix_sock_path,
4131 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4132 if (ret < 0) {
4133 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4134 PERROR("chmod");
4135 goto end;
4136 }
4137
4138 DBG3("Session daemon client socket %d and application socket %d created",
4139 client_sock, apps_sock);
4140
4141 end:
4142 umask(old_umask);
4143 return ret;
4144 }
4145
4146 /*
4147 * Check if the global socket is available, and if a daemon is answering at the
4148 * other side. If yes, error is returned.
4149 */
4150 static int check_existing_daemon(void)
4151 {
4152 /* Is there anybody out there ? */
4153 if (lttng_session_daemon_alive()) {
4154 return -EEXIST;
4155 }
4156
4157 return 0;
4158 }
4159
4160 /*
4161 * Set the tracing group gid onto the client socket.
4162 *
4163 * Race window between mkdir and chown is OK because we are going from more
4164 * permissive (root.root) to less permissive (root.tracing).
4165 */
4166 static int set_permissions(char *rundir)
4167 {
4168 int ret;
4169 gid_t gid;
4170
4171 gid = utils_get_group_id(tracing_group_name);
4172
4173 /* Set lttng run dir */
4174 ret = chown(rundir, 0, gid);
4175 if (ret < 0) {
4176 ERR("Unable to set group on %s", rundir);
4177 PERROR("chown");
4178 }
4179
4180 /*
4181 * Ensure all applications and tracing group can search the run
4182 * dir. Allow everyone to read the directory, since it does not
4183 * buy us anything to hide its content.
4184 */
4185 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4186 if (ret < 0) {
4187 ERR("Unable to set permissions on %s", rundir);
4188 PERROR("chmod");
4189 }
4190
4191 /* lttng client socket path */
4192 ret = chown(client_unix_sock_path, 0, gid);
4193 if (ret < 0) {
4194 ERR("Unable to set group on %s", client_unix_sock_path);
4195 PERROR("chown");
4196 }
4197
4198 /* kconsumer error socket path */
4199 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4200 if (ret < 0) {
4201 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4202 PERROR("chown");
4203 }
4204
4205 /* 64-bit ustconsumer error socket path */
4206 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4207 if (ret < 0) {
4208 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4209 PERROR("chown");
4210 }
4211
4212 /* 32-bit ustconsumer compat32 error socket path */
4213 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4214 if (ret < 0) {
4215 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4216 PERROR("chown");
4217 }
4218
4219 DBG("All permissions are set");
4220
4221 return ret;
4222 }
4223
4224 /*
4225 * Create the lttng run directory needed for all global sockets and pipe.
4226 */
4227 static int create_lttng_rundir(const char *rundir)
4228 {
4229 int ret;
4230
4231 DBG3("Creating LTTng run directory: %s", rundir);
4232
4233 ret = mkdir(rundir, S_IRWXU);
4234 if (ret < 0) {
4235 if (errno != EEXIST) {
4236 ERR("Unable to create %s", rundir);
4237 goto error;
4238 } else {
4239 ret = 0;
4240 }
4241 }
4242
4243 error:
4244 return ret;
4245 }
4246
4247 /*
4248 * Setup sockets and directory needed by the kconsumerd communication with the
4249 * session daemon.
4250 */
4251 static int set_consumer_sockets(struct consumer_data *consumer_data,
4252 const char *rundir)
4253 {
4254 int ret;
4255 char path[PATH_MAX];
4256
4257 switch (consumer_data->type) {
4258 case LTTNG_CONSUMER_KERNEL:
4259 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4260 break;
4261 case LTTNG_CONSUMER64_UST:
4262 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4263 break;
4264 case LTTNG_CONSUMER32_UST:
4265 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4266 break;
4267 default:
4268 ERR("Consumer type unknown");
4269 ret = -EINVAL;
4270 goto error;
4271 }
4272
4273 DBG2("Creating consumer directory: %s", path);
4274
4275 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4276 if (ret < 0) {
4277 if (errno != EEXIST) {
4278 PERROR("mkdir");
4279 ERR("Failed to create %s", path);
4280 goto error;
4281 }
4282 ret = -1;
4283 }
4284 if (is_root) {
4285 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4286 if (ret < 0) {
4287 ERR("Unable to set group on %s", path);
4288 PERROR("chown");
4289 goto error;
4290 }
4291 }
4292
4293 /* Create the kconsumerd error unix socket */
4294 consumer_data->err_sock =
4295 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4296 if (consumer_data->err_sock < 0) {
4297 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4298 ret = -1;
4299 goto error;
4300 }
4301
4302 /*
4303 * Set the CLOEXEC flag. Return code is useless because either way, the
4304 * show must go on.
4305 */
4306 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4307 if (ret < 0) {
4308 PERROR("utils_set_fd_cloexec");
4309 /* continue anyway */
4310 }
4311
4312 /* File permission MUST be 660 */
4313 ret = chmod(consumer_data->err_unix_sock_path,
4314 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4315 if (ret < 0) {
4316 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4317 PERROR("chmod");
4318 goto error;
4319 }
4320
4321 error:
4322 return ret;
4323 }
4324
4325 /*
4326 * Signal handler for the daemon
4327 *
4328 * Simply stop all worker threads, leaving main() return gracefully after
4329 * joining all threads and calling cleanup().
4330 */
4331 static void sighandler(int sig)
4332 {
4333 switch (sig) {
4334 case SIGPIPE:
4335 DBG("SIGPIPE caught");
4336 return;
4337 case SIGINT:
4338 DBG("SIGINT caught");
4339 stop_threads();
4340 break;
4341 case SIGTERM:
4342 DBG("SIGTERM caught");
4343 stop_threads();
4344 break;
4345 default:
4346 break;
4347 }
4348 }
4349
4350 /*
4351 * Setup signal handler for :
4352 * SIGINT, SIGTERM, SIGPIPE
4353 */
4354 static int set_signal_handler(void)
4355 {
4356 int ret = 0;
4357 struct sigaction sa;
4358 sigset_t sigset;
4359
4360 if ((ret = sigemptyset(&sigset)) < 0) {
4361 PERROR("sigemptyset");
4362 return ret;
4363 }
4364
4365 sa.sa_handler = sighandler;
4366 sa.sa_mask = sigset;
4367 sa.sa_flags = 0;
4368 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4369 PERROR("sigaction");
4370 return ret;
4371 }
4372
4373 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4374 PERROR("sigaction");
4375 return ret;
4376 }
4377
4378 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4379 PERROR("sigaction");
4380 return ret;
4381 }
4382
4383 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4384
4385 return ret;
4386 }
4387
4388 /*
4389 * Set open files limit to unlimited. This daemon can open a large number of
4390 * file descriptors in order to consumer multiple kernel traces.
4391 */
4392 static void set_ulimit(void)
4393 {
4394 int ret;
4395 struct rlimit lim;
4396
4397 /* The kernel does not allowed an infinite limit for open files */
4398 lim.rlim_cur = 65535;
4399 lim.rlim_max = 65535;
4400
4401 ret = setrlimit(RLIMIT_NOFILE, &lim);
4402 if (ret < 0) {
4403 PERROR("failed to set open files limit");
4404 }
4405 }
4406
4407 /*
4408 * Write pidfile using the rundir and opt_pidfile.
4409 */
4410 static void write_pidfile(void)
4411 {
4412 int ret;
4413 char pidfile_path[PATH_MAX];
4414
4415 assert(rundir);
4416
4417 if (opt_pidfile) {
4418 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4419 } else {
4420 /* Build pidfile path from rundir and opt_pidfile. */
4421 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4422 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4423 if (ret < 0) {
4424 PERROR("snprintf pidfile path");
4425 goto error;
4426 }
4427 }
4428
4429 /*
4430 * Create pid file in rundir. Return value is of no importance. The
4431 * execution will continue even though we are not able to write the file.
4432 */
4433 (void) utils_create_pid_file(getpid(), pidfile_path);
4434
4435 error:
4436 return;
4437 }
4438
4439 /*
4440 * Write JUL TCP port using the rundir.
4441 */
4442 static void write_julport(void)
4443 {
4444 int ret;
4445 char path[PATH_MAX];
4446
4447 assert(rundir);
4448
4449 ret = snprintf(path, sizeof(path), "%s/"
4450 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
4451 if (ret < 0) {
4452 PERROR("snprintf julport path");
4453 goto error;
4454 }
4455
4456 /*
4457 * Create TCP JUL port file in rundir. Return value is of no importance.
4458 * The execution will continue even though we are not able to write the
4459 * file.
4460 */
4461 (void) utils_create_pid_file(jul_tcp_port, path);
4462
4463 error:
4464 return;
4465 }
4466
4467 /*
4468 * main
4469 */
4470 int main(int argc, char **argv)
4471 {
4472 int ret = 0;
4473 void *status;
4474 const char *home_path, *env_app_timeout;
4475
4476 init_kernel_workarounds();
4477
4478 rcu_register_thread();
4479
4480 setup_consumerd_path();
4481
4482 page_size = sysconf(_SC_PAGESIZE);
4483 if (page_size < 0) {
4484 PERROR("sysconf _SC_PAGESIZE");
4485 page_size = LONG_MAX;
4486 WARN("Fallback page size to %ld", page_size);
4487 }
4488
4489 /* Parse arguments */
4490 progname = argv[0];
4491 if ((ret = parse_args(argc, argv)) < 0) {
4492 goto error;
4493 }
4494
4495 /* Daemonize */
4496 if (opt_daemon) {
4497 int i;
4498
4499 /*
4500 * fork
4501 * child: setsid, close FD 0, 1, 2, chdir /
4502 * parent: exit (if fork is successful)
4503 */
4504 ret = daemon(0, 0);
4505 if (ret < 0) {
4506 PERROR("daemon");
4507 goto error;
4508 }
4509 /*
4510 * We are in the child. Make sure all other file
4511 * descriptors are closed, in case we are called with
4512 * more opened file descriptors than the standard ones.
4513 */
4514 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4515 (void) close(i);
4516 }
4517 }
4518
4519 /* Create thread quit pipe */
4520 if ((ret = init_thread_quit_pipe()) < 0) {
4521 goto error;
4522 }
4523
4524 /* Check if daemon is UID = 0 */
4525 is_root = !getuid();
4526
4527 if (is_root) {
4528 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4529
4530 /* Create global run dir with root access */
4531 ret = create_lttng_rundir(rundir);
4532 if (ret < 0) {
4533 goto error;
4534 }
4535
4536 if (strlen(apps_unix_sock_path) == 0) {
4537 snprintf(apps_unix_sock_path, PATH_MAX,
4538 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4539 }
4540
4541 if (strlen(client_unix_sock_path) == 0) {
4542 snprintf(client_unix_sock_path, PATH_MAX,
4543 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4544 }
4545
4546 /* Set global SHM for ust */
4547 if (strlen(wait_shm_path) == 0) {
4548 snprintf(wait_shm_path, PATH_MAX,
4549 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4550 }
4551
4552 if (strlen(health_unix_sock_path) == 0) {
4553 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4554 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4555 }
4556
4557 /* Setup kernel consumerd path */
4558 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4559 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4560 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4561 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4562
4563 DBG2("Kernel consumer err path: %s",
4564 kconsumer_data.err_unix_sock_path);
4565 DBG2("Kernel consumer cmd path: %s",
4566 kconsumer_data.cmd_unix_sock_path);
4567 } else {
4568 home_path = utils_get_home_dir();
4569 if (home_path == NULL) {
4570 /* TODO: Add --socket PATH option */
4571 ERR("Can't get HOME directory for sockets creation.");
4572 ret = -EPERM;
4573 goto error;
4574 }
4575
4576 /*
4577 * Create rundir from home path. This will create something like
4578 * $HOME/.lttng
4579 */
4580 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4581 if (ret < 0) {
4582 ret = -ENOMEM;
4583 goto error;
4584 }
4585
4586 ret = create_lttng_rundir(rundir);
4587 if (ret < 0) {
4588 goto error;
4589 }
4590
4591 if (strlen(apps_unix_sock_path) == 0) {
4592 snprintf(apps_unix_sock_path, PATH_MAX,
4593 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4594 }
4595
4596 /* Set the cli tool unix socket path */
4597 if (strlen(client_unix_sock_path) == 0) {
4598 snprintf(client_unix_sock_path, PATH_MAX,
4599 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4600 }
4601
4602 /* Set global SHM for ust */
4603 if (strlen(wait_shm_path) == 0) {
4604 snprintf(wait_shm_path, PATH_MAX,
4605 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4606 }
4607
4608 /* Set health check Unix path */
4609 if (strlen(health_unix_sock_path) == 0) {
4610 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4611 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4612 }
4613 }
4614
4615 /* Set consumer initial state */
4616 kernel_consumerd_state = CONSUMER_STOPPED;
4617 ust_consumerd_state = CONSUMER_STOPPED;
4618
4619 DBG("Client socket path %s", client_unix_sock_path);
4620 DBG("Application socket path %s", apps_unix_sock_path);
4621 DBG("Application wait path %s", wait_shm_path);
4622 DBG("LTTng run directory path: %s", rundir);
4623
4624 /* 32 bits consumerd path setup */
4625 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4626 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4627 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4628 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4629
4630 DBG2("UST consumer 32 bits err path: %s",
4631 ustconsumer32_data.err_unix_sock_path);
4632 DBG2("UST consumer 32 bits cmd path: %s",
4633 ustconsumer32_data.cmd_unix_sock_path);
4634
4635 /* 64 bits consumerd path setup */
4636 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4637 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4638 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4639 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4640
4641 DBG2("UST consumer 64 bits err path: %s",
4642 ustconsumer64_data.err_unix_sock_path);
4643 DBG2("UST consumer 64 bits cmd path: %s",
4644 ustconsumer64_data.cmd_unix_sock_path);
4645
4646 /*
4647 * See if daemon already exist.
4648 */
4649 if ((ret = check_existing_daemon()) < 0) {
4650 ERR("Already running daemon.\n");
4651 /*
4652 * We do not goto exit because we must not cleanup()
4653 * because a daemon is already running.
4654 */
4655 goto error;
4656 }
4657
4658 /*
4659 * Init UST app hash table. Alloc hash table before this point since
4660 * cleanup() can get called after that point.
4661 */
4662 ust_app_ht_alloc();
4663
4664 /* Initialize JUL domain subsystem. */
4665 if ((ret = jul_init()) < 0) {
4666 /* ENOMEM at this point. */
4667 goto error;
4668 }
4669
4670 /* After this point, we can safely call cleanup() with "goto exit" */
4671
4672 /*
4673 * These actions must be executed as root. We do that *after* setting up
4674 * the sockets path because we MUST make the check for another daemon using
4675 * those paths *before* trying to set the kernel consumer sockets and init
4676 * kernel tracer.
4677 */
4678 if (is_root) {
4679 ret = set_consumer_sockets(&kconsumer_data, rundir);
4680 if (ret < 0) {
4681 goto exit;
4682 }
4683
4684 /* Setup kernel tracer */
4685 if (!opt_no_kernel) {
4686 init_kernel_tracer();
4687 }
4688
4689 /* Set ulimit for open files */
4690 set_ulimit();
4691 }
4692 /* init lttng_fd tracking must be done after set_ulimit. */
4693 lttng_fd_init();
4694
4695 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4696 if (ret < 0) {
4697 goto exit;
4698 }
4699
4700 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4701 if (ret < 0) {
4702 goto exit;
4703 }
4704
4705 if ((ret = set_signal_handler()) < 0) {
4706 goto exit;
4707 }
4708
4709 /* Setup the needed unix socket */
4710 if ((ret = init_daemon_socket()) < 0) {
4711 goto exit;
4712 }
4713
4714 /* Set credentials to socket */
4715 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4716 goto exit;
4717 }
4718
4719 /* Get parent pid if -S, --sig-parent is specified. */
4720 if (opt_sig_parent) {
4721 ppid = getppid();
4722 }
4723
4724 /* Setup the kernel pipe for waking up the kernel thread */
4725 if (is_root && !opt_no_kernel) {
4726 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4727 goto exit;
4728 }
4729 }
4730
4731 /* Setup the thread ht_cleanup communication pipe. */
4732 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4733 goto exit;
4734 }
4735
4736 /* Setup the thread apps communication pipe. */
4737 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4738 goto exit;
4739 }
4740
4741 /* Setup the thread apps notify communication pipe. */
4742 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4743 goto exit;
4744 }
4745
4746 /* Initialize global buffer per UID and PID registry. */
4747 buffer_reg_init_uid_registry();
4748 buffer_reg_init_pid_registry();
4749
4750 /* Init UST command queue. */
4751 cds_wfq_init(&ust_cmd_queue.queue);
4752
4753 /*
4754 * Get session list pointer. This pointer MUST NOT be free(). This list is
4755 * statically declared in session.c
4756 */
4757 session_list_ptr = session_get_list();
4758
4759 /* Set up max poll set size */
4760 lttng_poll_set_max_size();
4761
4762 cmd_init();
4763
4764 /* Check for the application socket timeout env variable. */
4765 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4766 if (env_app_timeout) {
4767 app_socket_timeout = atoi(env_app_timeout);
4768 } else {
4769 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4770 }
4771
4772 write_pidfile();
4773 write_julport();
4774
4775 /* Initialize communication library */
4776 lttcomm_init();
4777 /* This is to get the TCP timeout value. */
4778 lttcomm_inet_init();
4779
4780 /*
4781 * Initialize the health check subsystem. This call should set the
4782 * appropriate time values.
4783 */
4784 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
4785 if (!health_sessiond) {
4786 PERROR("health_app_create error");
4787 goto exit_health_sessiond_cleanup;
4788 }
4789
4790 /* Create thread to manage the client socket */
4791 ret = pthread_create(&ht_cleanup_thread, NULL,
4792 thread_ht_cleanup, (void *) NULL);
4793 if (ret != 0) {
4794 PERROR("pthread_create ht_cleanup");
4795 goto exit_ht_cleanup;
4796 }
4797
4798 /* Create thread to manage the client socket */
4799 ret = pthread_create(&health_thread, NULL,
4800 thread_manage_health, (void *) NULL);
4801 if (ret != 0) {
4802 PERROR("pthread_create health");
4803 goto exit_health;
4804 }
4805
4806 /* Create thread to manage the client socket */
4807 ret = pthread_create(&client_thread, NULL,
4808 thread_manage_clients, (void *) NULL);
4809 if (ret != 0) {
4810 PERROR("pthread_create clients");
4811 goto exit_client;
4812 }
4813
4814 /* Create thread to dispatch registration */
4815 ret = pthread_create(&dispatch_thread, NULL,
4816 thread_dispatch_ust_registration, (void *) NULL);
4817 if (ret != 0) {
4818 PERROR("pthread_create dispatch");
4819 goto exit_dispatch;
4820 }
4821
4822 /* Create thread to manage application registration. */
4823 ret = pthread_create(&reg_apps_thread, NULL,
4824 thread_registration_apps, (void *) NULL);
4825 if (ret != 0) {
4826 PERROR("pthread_create registration");
4827 goto exit_reg_apps;
4828 }
4829
4830 /* Create thread to manage application socket */
4831 ret = pthread_create(&apps_thread, NULL,
4832 thread_manage_apps, (void *) NULL);
4833 if (ret != 0) {
4834 PERROR("pthread_create apps");
4835 goto exit_apps;
4836 }
4837
4838 /* Create thread to manage application notify socket */
4839 ret = pthread_create(&apps_notify_thread, NULL,
4840 ust_thread_manage_notify, (void *) NULL);
4841 if (ret != 0) {
4842 PERROR("pthread_create apps");
4843 goto exit_apps_notify;
4844 }
4845
4846 /* Create JUL registration thread. */
4847 ret = pthread_create(&jul_reg_thread, NULL,
4848 jul_thread_manage_registration, (void *) NULL);
4849 if (ret != 0) {
4850 PERROR("pthread_create apps");
4851 goto exit_jul_reg;
4852 }
4853
4854 /* Don't start this thread if kernel tracing is not requested nor root */
4855 if (is_root && !opt_no_kernel) {
4856 /* Create kernel thread to manage kernel event */
4857 ret = pthread_create(&kernel_thread, NULL,
4858 thread_manage_kernel, (void *) NULL);
4859 if (ret != 0) {
4860 PERROR("pthread_create kernel");
4861 goto exit_kernel;
4862 }
4863
4864 ret = pthread_join(kernel_thread, &status);
4865 if (ret != 0) {
4866 PERROR("pthread_join");
4867 goto error; /* join error, exit without cleanup */
4868 }
4869 }
4870
4871 exit_kernel:
4872 ret = pthread_join(jul_reg_thread, &status);
4873 if (ret != 0) {
4874 PERROR("pthread_join JUL");
4875 goto error; /* join error, exit without cleanup */
4876 }
4877
4878 exit_jul_reg:
4879 ret = pthread_join(apps_notify_thread, &status);
4880 if (ret != 0) {
4881 PERROR("pthread_join apps notify");
4882 goto error; /* join error, exit without cleanup */
4883 }
4884
4885 exit_apps_notify:
4886 ret = pthread_join(apps_thread, &status);
4887 if (ret != 0) {
4888 PERROR("pthread_join apps");
4889 goto error; /* join error, exit without cleanup */
4890 }
4891
4892
4893 exit_apps:
4894 ret = pthread_join(reg_apps_thread, &status);
4895 if (ret != 0) {
4896 PERROR("pthread_join");
4897 goto error; /* join error, exit without cleanup */
4898 }
4899
4900 exit_reg_apps:
4901 ret = pthread_join(dispatch_thread, &status);
4902 if (ret != 0) {
4903 PERROR("pthread_join");
4904 goto error; /* join error, exit without cleanup */
4905 }
4906
4907 exit_dispatch:
4908 ret = pthread_join(client_thread, &status);
4909 if (ret != 0) {
4910 PERROR("pthread_join");
4911 goto error; /* join error, exit without cleanup */
4912 }
4913
4914 ret = join_consumer_thread(&kconsumer_data);
4915 if (ret != 0) {
4916 PERROR("join_consumer");
4917 goto error; /* join error, exit without cleanup */
4918 }
4919
4920 ret = join_consumer_thread(&ustconsumer32_data);
4921 if (ret != 0) {
4922 PERROR("join_consumer ust32");
4923 goto error; /* join error, exit without cleanup */
4924 }
4925
4926 ret = join_consumer_thread(&ustconsumer64_data);
4927 if (ret != 0) {
4928 PERROR("join_consumer ust64");
4929 goto error; /* join error, exit without cleanup */
4930 }
4931
4932 exit_client:
4933 ret = pthread_join(health_thread, &status);
4934 if (ret != 0) {
4935 PERROR("pthread_join health thread");
4936 goto error; /* join error, exit without cleanup */
4937 }
4938
4939 exit_health:
4940 ret = pthread_join(ht_cleanup_thread, &status);
4941 if (ret != 0) {
4942 PERROR("pthread_join ht cleanup thread");
4943 goto error; /* join error, exit without cleanup */
4944 }
4945 exit_ht_cleanup:
4946 health_app_destroy(health_sessiond);
4947 exit_health_sessiond_cleanup:
4948 exit:
4949 /*
4950 * cleanup() is called when no other thread is running.
4951 */
4952 rcu_thread_online();
4953 cleanup();
4954 rcu_thread_offline();
4955 rcu_unregister_thread();
4956 if (!ret) {
4957 exit(EXIT_SUCCESS);
4958 }
4959 error:
4960 exit(EXIT_FAILURE);
4961 }
This page took 0.162815 seconds and 4 git commands to generate.