076d0b7b03ce24295d8d7aeb924c7cdbcb117e16
[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 <paths.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <inttypes.h>
30 #include <sys/mman.h>
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
34 #include <sys/stat.h>
35 #include <sys/types.h>
36 #include <sys/wait.h>
37 #include <urcu/uatomic.h>
38 #include <unistd.h>
39 #include <config.h>
40
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/defaults.h>
44 #include <common/kernel-consumer/kernel-consumer.h>
45 #include <common/futex.h>
46 #include <common/relayd/relayd.h>
47 #include <common/utils.h>
48
49 #include "lttng-sessiond.h"
50 #include "buffer-registry.h"
51 #include "channel.h"
52 #include "cmd.h"
53 #include "consumer.h"
54 #include "context.h"
55 #include "event.h"
56 #include "kernel.h"
57 #include "kernel-consumer.h"
58 #include "modprobe.h"
59 #include "shm.h"
60 #include "ust-ctl.h"
61 #include "ust-consumer.h"
62 #include "utils.h"
63 #include "fd-limit.h"
64 #include "health-sessiond.h"
65 #include "testpoint.h"
66 #include "ust-thread.h"
67 #include "jul-thread.h"
68
69 #define CONSUMERD_FILE "lttng-consumerd"
70
71 const char *progname;
72 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
73 static const char *opt_pidfile;
74 static int opt_sig_parent;
75 static int opt_verbose_consumer;
76 static int opt_daemon;
77 static int opt_no_kernel;
78 static int is_root; /* Set to 1 if the daemon is running as root */
79 static pid_t ppid; /* Parent PID for --sig-parent option */
80 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
81 static char *rundir;
82
83 /* Set to 1 when a SIGUSR1 signal is received. */
84 static int recv_child_signal;
85
86 /*
87 * Consumer daemon specific control data. Every value not initialized here is
88 * set to 0 by the static definition.
89 */
90 static struct consumer_data kconsumer_data = {
91 .type = LTTNG_CONSUMER_KERNEL,
92 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
93 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
94 .err_sock = -1,
95 .cmd_sock = -1,
96 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
97 .lock = PTHREAD_MUTEX_INITIALIZER,
98 .cond = PTHREAD_COND_INITIALIZER,
99 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
100 };
101 static struct consumer_data ustconsumer64_data = {
102 .type = LTTNG_CONSUMER64_UST,
103 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
104 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
105 .err_sock = -1,
106 .cmd_sock = -1,
107 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
108 .lock = PTHREAD_MUTEX_INITIALIZER,
109 .cond = PTHREAD_COND_INITIALIZER,
110 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
111 };
112 static struct consumer_data ustconsumer32_data = {
113 .type = LTTNG_CONSUMER32_UST,
114 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
115 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
116 .err_sock = -1,
117 .cmd_sock = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123
124 /* Shared between threads */
125 static int dispatch_thread_exit;
126
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path[PATH_MAX];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path[PATH_MAX];
131 /* global wait shm path for UST */
132 static char wait_shm_path[PATH_MAX];
133 /* Global health check unix path */
134 static char health_unix_sock_path[PATH_MAX];
135
136 /* Sockets and FDs */
137 static int client_sock = -1;
138 static int apps_sock = -1;
139 int kernel_tracer_fd = -1;
140 static int kernel_poll_pipe[2] = { -1, -1 };
141
142 /*
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
145 */
146 static int thread_quit_pipe[2] = { -1, -1 };
147
148 /*
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
151 */
152 static int apps_cmd_pipe[2] = { -1, -1 };
153
154 int apps_cmd_notify_pipe[2] = { -1, -1 };
155
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread;
158 static pthread_t apps_notify_thread;
159 static pthread_t reg_apps_thread;
160 static pthread_t client_thread;
161 static pthread_t kernel_thread;
162 static pthread_t dispatch_thread;
163 static pthread_t health_thread;
164 static pthread_t ht_cleanup_thread;
165 static pthread_t jul_reg_thread;
166
167 /*
168 * UST registration command queue. This queue is tied with a futex and uses a N
169 * wakers / 1 waiter implemented and detailed in futex.c/.h
170 *
171 * The thread_manage_apps and thread_dispatch_ust_registration interact with
172 * this queue and the wait/wake scheme.
173 */
174 static struct ust_cmd_queue ust_cmd_queue;
175
176 /*
177 * Pointer initialized before thread creation.
178 *
179 * This points to the tracing session list containing the session count and a
180 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
181 * MUST NOT be taken if you call a public function in session.c.
182 *
183 * The lock is nested inside the structure: session_list_ptr->lock. Please use
184 * session_lock_list and session_unlock_list for lock acquisition.
185 */
186 static struct ltt_session_list *session_list_ptr;
187
188 int ust_consumerd64_fd = -1;
189 int ust_consumerd32_fd = -1;
190
191 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
192 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
193 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
194 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
195
196 static const char *module_proc_lttng = "/proc/lttng";
197
198 /*
199 * Consumer daemon state which is changed when spawning it, killing it or in
200 * case of a fatal error.
201 */
202 enum consumerd_state {
203 CONSUMER_STARTED = 1,
204 CONSUMER_STOPPED = 2,
205 CONSUMER_ERROR = 3,
206 };
207
208 /*
209 * This consumer daemon state is used to validate if a client command will be
210 * able to reach the consumer. If not, the client is informed. For instance,
211 * doing a "lttng start" when the consumer state is set to ERROR will return an
212 * error to the client.
213 *
214 * The following example shows a possible race condition of this scheme:
215 *
216 * consumer thread error happens
217 * client cmd arrives
218 * client cmd checks state -> still OK
219 * consumer thread exit, sets error
220 * client cmd try to talk to consumer
221 * ...
222 *
223 * However, since the consumer is a different daemon, we have no way of making
224 * sure the command will reach it safely even with this state flag. This is why
225 * we consider that up to the state validation during command processing, the
226 * command is safe. After that, we can not guarantee the correctness of the
227 * client request vis-a-vis the consumer.
228 */
229 static enum consumerd_state ust_consumerd_state;
230 static enum consumerd_state kernel_consumerd_state;
231
232 /*
233 * Socket timeout for receiving and sending in seconds.
234 */
235 static int app_socket_timeout;
236
237 /* Set in main() with the current page size. */
238 long page_size;
239
240 /* Application health monitoring */
241 struct health_app *health_sessiond;
242
243 /* JUL TCP port for registration. Used by the JUL thread. */
244 unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
245
246 static
247 void setup_consumerd_path(void)
248 {
249 const char *bin, *libdir;
250
251 /*
252 * Allow INSTALL_BIN_PATH to be used as a target path for the
253 * native architecture size consumer if CONFIG_CONSUMER*_PATH
254 * has not been defined.
255 */
256 #if (CAA_BITS_PER_LONG == 32)
257 if (!consumerd32_bin[0]) {
258 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
259 }
260 if (!consumerd32_libdir[0]) {
261 consumerd32_libdir = INSTALL_LIB_PATH;
262 }
263 #elif (CAA_BITS_PER_LONG == 64)
264 if (!consumerd64_bin[0]) {
265 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
266 }
267 if (!consumerd64_libdir[0]) {
268 consumerd64_libdir = INSTALL_LIB_PATH;
269 }
270 #else
271 #error "Unknown bitness"
272 #endif
273
274 /*
275 * runtime env. var. overrides the build default.
276 */
277 bin = getenv("LTTNG_CONSUMERD32_BIN");
278 if (bin) {
279 consumerd32_bin = bin;
280 }
281 bin = getenv("LTTNG_CONSUMERD64_BIN");
282 if (bin) {
283 consumerd64_bin = bin;
284 }
285 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
286 if (libdir) {
287 consumerd32_libdir = libdir;
288 }
289 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
290 if (libdir) {
291 consumerd64_libdir = libdir;
292 }
293 }
294
295 /*
296 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
297 */
298 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
299 {
300 int ret;
301
302 assert(events);
303
304 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
305 if (ret < 0) {
306 goto error;
307 }
308
309 /* Add quit pipe */
310 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
311 if (ret < 0) {
312 goto error;
313 }
314
315 return 0;
316
317 error:
318 return ret;
319 }
320
321 /*
322 * Check if the thread quit pipe was triggered.
323 *
324 * Return 1 if it was triggered else 0;
325 */
326 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
327 {
328 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
329 return 1;
330 }
331
332 return 0;
333 }
334
335 /*
336 * Init thread quit pipe.
337 *
338 * Return -1 on error or 0 if all pipes are created.
339 */
340 static int init_thread_quit_pipe(void)
341 {
342 int ret, i;
343
344 ret = pipe(thread_quit_pipe);
345 if (ret < 0) {
346 PERROR("thread quit pipe");
347 goto error;
348 }
349
350 for (i = 0; i < 2; i++) {
351 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
352 if (ret < 0) {
353 PERROR("fcntl");
354 goto error;
355 }
356 }
357
358 error:
359 return ret;
360 }
361
362 /*
363 * Stop all threads by closing the thread quit pipe.
364 */
365 static void stop_threads(void)
366 {
367 int ret;
368
369 /* Stopping all threads */
370 DBG("Terminating all threads");
371 ret = notify_thread_pipe(thread_quit_pipe[1]);
372 if (ret < 0) {
373 ERR("write error on thread quit pipe");
374 }
375
376 /* Dispatch thread */
377 CMM_STORE_SHARED(dispatch_thread_exit, 1);
378 futex_nto1_wake(&ust_cmd_queue.futex);
379 }
380
381 /*
382 * Close every consumer sockets.
383 */
384 static void close_consumer_sockets(void)
385 {
386 int ret;
387
388 if (kconsumer_data.err_sock >= 0) {
389 ret = close(kconsumer_data.err_sock);
390 if (ret < 0) {
391 PERROR("kernel consumer err_sock close");
392 }
393 }
394 if (ustconsumer32_data.err_sock >= 0) {
395 ret = close(ustconsumer32_data.err_sock);
396 if (ret < 0) {
397 PERROR("UST consumerd32 err_sock close");
398 }
399 }
400 if (ustconsumer64_data.err_sock >= 0) {
401 ret = close(ustconsumer64_data.err_sock);
402 if (ret < 0) {
403 PERROR("UST consumerd64 err_sock close");
404 }
405 }
406 if (kconsumer_data.cmd_sock >= 0) {
407 ret = close(kconsumer_data.cmd_sock);
408 if (ret < 0) {
409 PERROR("kernel consumer cmd_sock close");
410 }
411 }
412 if (ustconsumer32_data.cmd_sock >= 0) {
413 ret = close(ustconsumer32_data.cmd_sock);
414 if (ret < 0) {
415 PERROR("UST consumerd32 cmd_sock close");
416 }
417 }
418 if (ustconsumer64_data.cmd_sock >= 0) {
419 ret = close(ustconsumer64_data.cmd_sock);
420 if (ret < 0) {
421 PERROR("UST consumerd64 cmd_sock close");
422 }
423 }
424 }
425
426 /*
427 * Cleanup the daemon
428 */
429 static void cleanup(void)
430 {
431 int ret;
432 struct ltt_session *sess, *stmp;
433 char path[PATH_MAX];
434
435 DBG("Cleaning up");
436
437 /*
438 * Close the thread quit pipe. It has already done its job,
439 * since we are now called.
440 */
441 utils_close_pipe(thread_quit_pipe);
442
443 /*
444 * If opt_pidfile is undefined, the default file will be wiped when
445 * removing the rundir.
446 */
447 if (opt_pidfile) {
448 ret = remove(opt_pidfile);
449 if (ret < 0) {
450 PERROR("remove pidfile %s", opt_pidfile);
451 }
452 }
453
454 DBG("Removing sessiond and consumerd content of directory %s", rundir);
455
456 /* sessiond */
457 snprintf(path, PATH_MAX,
458 "%s/%s",
459 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
460 DBG("Removing %s", path);
461 (void) unlink(path);
462
463 snprintf(path, PATH_MAX, "%s/%s", rundir,
464 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
465 DBG("Removing %s", path);
466 (void) unlink(path);
467
468 /* kconsumerd */
469 snprintf(path, PATH_MAX,
470 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
471 rundir);
472 DBG("Removing %s", path);
473 (void) unlink(path);
474
475 snprintf(path, PATH_MAX,
476 DEFAULT_KCONSUMERD_PATH,
477 rundir);
478 DBG("Removing directory %s", path);
479 (void) rmdir(path);
480
481 /* ust consumerd 32 */
482 snprintf(path, PATH_MAX,
483 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
484 rundir);
485 DBG("Removing %s", path);
486 (void) unlink(path);
487
488 snprintf(path, PATH_MAX,
489 DEFAULT_USTCONSUMERD32_PATH,
490 rundir);
491 DBG("Removing directory %s", path);
492 (void) rmdir(path);
493
494 /* ust consumerd 64 */
495 snprintf(path, PATH_MAX,
496 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
497 rundir);
498 DBG("Removing %s", path);
499 (void) unlink(path);
500
501 snprintf(path, PATH_MAX,
502 DEFAULT_USTCONSUMERD64_PATH,
503 rundir);
504 DBG("Removing directory %s", path);
505 (void) rmdir(path);
506
507 /*
508 * We do NOT rmdir rundir because there are other processes
509 * using it, for instance lttng-relayd, which can start in
510 * parallel with this teardown.
511 */
512
513 free(rundir);
514
515 DBG("Cleaning up all sessions");
516
517 /* Destroy session list mutex */
518 if (session_list_ptr != NULL) {
519 pthread_mutex_destroy(&session_list_ptr->lock);
520
521 /* Cleanup ALL session */
522 cds_list_for_each_entry_safe(sess, stmp,
523 &session_list_ptr->head, list) {
524 cmd_destroy_session(sess, kernel_poll_pipe[1]);
525 }
526 }
527
528 DBG("Closing all UST sockets");
529 ust_app_clean_list();
530 buffer_reg_destroy_registries();
531
532 if (is_root && !opt_no_kernel) {
533 DBG2("Closing kernel fd");
534 if (kernel_tracer_fd >= 0) {
535 ret = close(kernel_tracer_fd);
536 if (ret) {
537 PERROR("close");
538 }
539 }
540 DBG("Unloading kernel modules");
541 modprobe_remove_lttng_all();
542 }
543
544 close_consumer_sockets();
545
546 /* <fun> */
547 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
548 "Matthew, BEET driven development works!%c[%dm",
549 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
550 /* </fun> */
551 }
552
553 /*
554 * Send data on a unix socket using the liblttsessiondcomm API.
555 *
556 * Return lttcomm error code.
557 */
558 static int send_unix_sock(int sock, void *buf, size_t len)
559 {
560 /* Check valid length */
561 if (len == 0) {
562 return -1;
563 }
564
565 return lttcomm_send_unix_sock(sock, buf, len);
566 }
567
568 /*
569 * Free memory of a command context structure.
570 */
571 static void clean_command_ctx(struct command_ctx **cmd_ctx)
572 {
573 DBG("Clean command context structure");
574 if (*cmd_ctx) {
575 if ((*cmd_ctx)->llm) {
576 free((*cmd_ctx)->llm);
577 }
578 if ((*cmd_ctx)->lsm) {
579 free((*cmd_ctx)->lsm);
580 }
581 free(*cmd_ctx);
582 *cmd_ctx = NULL;
583 }
584 }
585
586 /*
587 * Notify UST applications using the shm mmap futex.
588 */
589 static int notify_ust_apps(int active)
590 {
591 char *wait_shm_mmap;
592
593 DBG("Notifying applications of session daemon state: %d", active);
594
595 /* See shm.c for this call implying mmap, shm and futex calls */
596 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
597 if (wait_shm_mmap == NULL) {
598 goto error;
599 }
600
601 /* Wake waiting process */
602 futex_wait_update((int32_t *) wait_shm_mmap, active);
603
604 /* Apps notified successfully */
605 return 0;
606
607 error:
608 return -1;
609 }
610
611 /*
612 * Setup the outgoing data buffer for the response (llm) by allocating the
613 * right amount of memory and copying the original information from the lsm
614 * structure.
615 *
616 * Return total size of the buffer pointed by buf.
617 */
618 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
619 {
620 int ret, buf_size;
621
622 buf_size = size;
623
624 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
625 if (cmd_ctx->llm == NULL) {
626 PERROR("zmalloc");
627 ret = -ENOMEM;
628 goto error;
629 }
630
631 /* Copy common data */
632 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
633 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
634
635 cmd_ctx->llm->data_size = size;
636 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
637
638 return buf_size;
639
640 error:
641 return ret;
642 }
643
644 /*
645 * Update the kernel poll set of all channel fd available over all tracing
646 * session. Add the wakeup pipe at the end of the set.
647 */
648 static int update_kernel_poll(struct lttng_poll_event *events)
649 {
650 int ret;
651 struct ltt_session *session;
652 struct ltt_kernel_channel *channel;
653
654 DBG("Updating kernel poll set");
655
656 session_lock_list();
657 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
658 session_lock(session);
659 if (session->kernel_session == NULL) {
660 session_unlock(session);
661 continue;
662 }
663
664 cds_list_for_each_entry(channel,
665 &session->kernel_session->channel_list.head, list) {
666 /* Add channel fd to the kernel poll set */
667 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
668 if (ret < 0) {
669 session_unlock(session);
670 goto error;
671 }
672 DBG("Channel fd %d added to kernel set", channel->fd);
673 }
674 session_unlock(session);
675 }
676 session_unlock_list();
677
678 return 0;
679
680 error:
681 session_unlock_list();
682 return -1;
683 }
684
685 /*
686 * Find the channel fd from 'fd' over all tracing session. When found, check
687 * for new channel stream and send those stream fds to the kernel consumer.
688 *
689 * Useful for CPU hotplug feature.
690 */
691 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
692 {
693 int ret = 0;
694 struct ltt_session *session;
695 struct ltt_kernel_session *ksess;
696 struct ltt_kernel_channel *channel;
697
698 DBG("Updating kernel streams for channel fd %d", fd);
699
700 session_lock_list();
701 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
702 session_lock(session);
703 if (session->kernel_session == NULL) {
704 session_unlock(session);
705 continue;
706 }
707 ksess = session->kernel_session;
708
709 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
710 if (channel->fd == fd) {
711 DBG("Channel found, updating kernel streams");
712 ret = kernel_open_channel_stream(channel);
713 if (ret < 0) {
714 goto error;
715 }
716 /* Update the stream global counter */
717 ksess->stream_count_global += ret;
718
719 /*
720 * Have we already sent fds to the consumer? If yes, it means
721 * that tracing is started so it is safe to send our updated
722 * stream fds.
723 */
724 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
725 struct lttng_ht_iter iter;
726 struct consumer_socket *socket;
727
728 rcu_read_lock();
729 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
730 &iter.iter, socket, node.node) {
731 pthread_mutex_lock(socket->lock);
732 ret = kernel_consumer_send_channel_stream(socket,
733 channel, ksess,
734 session->output_traces ? 1 : 0);
735 pthread_mutex_unlock(socket->lock);
736 if (ret < 0) {
737 rcu_read_unlock();
738 goto error;
739 }
740 }
741 rcu_read_unlock();
742 }
743 goto error;
744 }
745 }
746 session_unlock(session);
747 }
748 session_unlock_list();
749 return ret;
750
751 error:
752 session_unlock(session);
753 session_unlock_list();
754 return ret;
755 }
756
757 /*
758 * For each tracing session, update newly registered apps. The session list
759 * lock MUST be acquired before calling this.
760 */
761 static void update_ust_app(int app_sock)
762 {
763 struct ltt_session *sess, *stmp;
764
765 /* Consumer is in an ERROR state. Stop any application update. */
766 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
767 /* Stop the update process since the consumer is dead. */
768 return;
769 }
770
771 /* For all tracing session(s) */
772 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
773 session_lock(sess);
774 if (sess->ust_session) {
775 ust_app_global_update(sess->ust_session, app_sock);
776 }
777 session_unlock(sess);
778 }
779 }
780
781 /*
782 * This thread manage event coming from the kernel.
783 *
784 * Features supported in this thread:
785 * -) CPU Hotplug
786 */
787 static void *thread_manage_kernel(void *data)
788 {
789 int ret, i, pollfd, update_poll_flag = 1, err = -1;
790 uint32_t revents, nb_fd;
791 char tmp;
792 struct lttng_poll_event events;
793
794 DBG("[thread] Thread manage kernel started");
795
796 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
797
798 /*
799 * This first step of the while is to clean this structure which could free
800 * non NULL pointers so initialize it before the loop.
801 */
802 lttng_poll_init(&events);
803
804 if (testpoint(thread_manage_kernel)) {
805 goto error_testpoint;
806 }
807
808 health_code_update();
809
810 if (testpoint(thread_manage_kernel_before_loop)) {
811 goto error_testpoint;
812 }
813
814 while (1) {
815 health_code_update();
816
817 if (update_poll_flag == 1) {
818 /* Clean events object. We are about to populate it again. */
819 lttng_poll_clean(&events);
820
821 ret = sessiond_set_thread_pollset(&events, 2);
822 if (ret < 0) {
823 goto error_poll_create;
824 }
825
826 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
827 if (ret < 0) {
828 goto error;
829 }
830
831 /* This will add the available kernel channel if any. */
832 ret = update_kernel_poll(&events);
833 if (ret < 0) {
834 goto error;
835 }
836 update_poll_flag = 0;
837 }
838
839 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
840
841 /* Poll infinite value of time */
842 restart:
843 health_poll_entry();
844 ret = lttng_poll_wait(&events, -1);
845 health_poll_exit();
846 if (ret < 0) {
847 /*
848 * Restart interrupted system call.
849 */
850 if (errno == EINTR) {
851 goto restart;
852 }
853 goto error;
854 } else if (ret == 0) {
855 /* Should not happen since timeout is infinite */
856 ERR("Return value of poll is 0 with an infinite timeout.\n"
857 "This should not have happened! Continuing...");
858 continue;
859 }
860
861 nb_fd = ret;
862
863 for (i = 0; i < nb_fd; i++) {
864 /* Fetch once the poll data */
865 revents = LTTNG_POLL_GETEV(&events, i);
866 pollfd = LTTNG_POLL_GETFD(&events, i);
867
868 health_code_update();
869
870 /* Thread quit pipe has been closed. Killing thread. */
871 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
872 if (ret) {
873 err = 0;
874 goto exit;
875 }
876
877 /* Check for data on kernel pipe */
878 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
879 (void) lttng_read(kernel_poll_pipe[0],
880 &tmp, 1);
881 /*
882 * Ret value is useless here, if this pipe gets any actions an
883 * update is required anyway.
884 */
885 update_poll_flag = 1;
886 continue;
887 } else {
888 /*
889 * New CPU detected by the kernel. Adding kernel stream to
890 * kernel session and updating the kernel consumer
891 */
892 if (revents & LPOLLIN) {
893 ret = update_kernel_stream(&kconsumer_data, pollfd);
894 if (ret < 0) {
895 continue;
896 }
897 break;
898 /*
899 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
900 * and unregister kernel stream at this point.
901 */
902 }
903 }
904 }
905 }
906
907 exit:
908 error:
909 lttng_poll_clean(&events);
910 error_poll_create:
911 error_testpoint:
912 utils_close_pipe(kernel_poll_pipe);
913 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
914 if (err) {
915 health_error();
916 ERR("Health error occurred in %s", __func__);
917 WARN("Kernel thread died unexpectedly. "
918 "Kernel tracing can continue but CPU hotplug is disabled.");
919 }
920 health_unregister(health_sessiond);
921 DBG("Kernel thread dying");
922 return NULL;
923 }
924
925 /*
926 * Signal pthread condition of the consumer data that the thread.
927 */
928 static void signal_consumer_condition(struct consumer_data *data, int state)
929 {
930 pthread_mutex_lock(&data->cond_mutex);
931
932 /*
933 * The state is set before signaling. It can be any value, it's the waiter
934 * job to correctly interpret this condition variable associated to the
935 * consumer pthread_cond.
936 *
937 * A value of 0 means that the corresponding thread of the consumer data
938 * was not started. 1 indicates that the thread has started and is ready
939 * for action. A negative value means that there was an error during the
940 * thread bootstrap.
941 */
942 data->consumer_thread_is_ready = state;
943 (void) pthread_cond_signal(&data->cond);
944
945 pthread_mutex_unlock(&data->cond_mutex);
946 }
947
948 /*
949 * This thread manage the consumer error sent back to the session daemon.
950 */
951 static void *thread_manage_consumer(void *data)
952 {
953 int sock = -1, i, ret, pollfd, err = -1;
954 uint32_t revents, nb_fd;
955 enum lttcomm_return_code code;
956 struct lttng_poll_event events;
957 struct consumer_data *consumer_data = data;
958
959 DBG("[thread] Manage consumer started");
960
961 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
962
963 health_code_update();
964
965 /*
966 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
967 * metadata_sock. Nothing more will be added to this poll set.
968 */
969 ret = sessiond_set_thread_pollset(&events, 3);
970 if (ret < 0) {
971 goto error_poll;
972 }
973
974 /*
975 * The error socket here is already in a listening state which was done
976 * just before spawning this thread to avoid a race between the consumer
977 * daemon exec trying to connect and the listen() call.
978 */
979 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
980 if (ret < 0) {
981 goto error;
982 }
983
984 health_code_update();
985
986 /* Infinite blocking call, waiting for transmission */
987 restart:
988 health_poll_entry();
989
990 if (testpoint(thread_manage_consumer)) {
991 goto error;
992 }
993
994 ret = lttng_poll_wait(&events, -1);
995 health_poll_exit();
996 if (ret < 0) {
997 /*
998 * Restart interrupted system call.
999 */
1000 if (errno == EINTR) {
1001 goto restart;
1002 }
1003 goto error;
1004 }
1005
1006 nb_fd = ret;
1007
1008 for (i = 0; i < nb_fd; i++) {
1009 /* Fetch once the poll data */
1010 revents = LTTNG_POLL_GETEV(&events, i);
1011 pollfd = LTTNG_POLL_GETFD(&events, i);
1012
1013 health_code_update();
1014
1015 /* Thread quit pipe has been closed. Killing thread. */
1016 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1017 if (ret) {
1018 err = 0;
1019 goto exit;
1020 }
1021
1022 /* Event on the registration socket */
1023 if (pollfd == consumer_data->err_sock) {
1024 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1025 ERR("consumer err socket poll error");
1026 goto error;
1027 }
1028 }
1029 }
1030
1031 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1032 if (sock < 0) {
1033 goto error;
1034 }
1035
1036 /*
1037 * Set the CLOEXEC flag. Return code is useless because either way, the
1038 * show must go on.
1039 */
1040 (void) utils_set_fd_cloexec(sock);
1041
1042 health_code_update();
1043
1044 DBG2("Receiving code from consumer err_sock");
1045
1046 /* Getting status code from kconsumerd */
1047 ret = lttcomm_recv_unix_sock(sock, &code,
1048 sizeof(enum lttcomm_return_code));
1049 if (ret <= 0) {
1050 goto error;
1051 }
1052
1053 health_code_update();
1054
1055 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1056 /* Connect both socket, command and metadata. */
1057 consumer_data->cmd_sock =
1058 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1059 consumer_data->metadata_fd =
1060 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1061 if (consumer_data->cmd_sock < 0
1062 || consumer_data->metadata_fd < 0) {
1063 PERROR("consumer connect cmd socket");
1064 /* On error, signal condition and quit. */
1065 signal_consumer_condition(consumer_data, -1);
1066 goto error;
1067 }
1068 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1069 /* Create metadata socket lock. */
1070 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1071 if (consumer_data->metadata_sock.lock == NULL) {
1072 PERROR("zmalloc pthread mutex");
1073 ret = -1;
1074 goto error;
1075 }
1076 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1077
1078 signal_consumer_condition(consumer_data, 1);
1079 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1080 DBG("Consumer metadata socket ready (fd: %d)",
1081 consumer_data->metadata_fd);
1082 } else {
1083 ERR("consumer error when waiting for SOCK_READY : %s",
1084 lttcomm_get_readable_code(-code));
1085 goto error;
1086 }
1087
1088 /* Remove the consumerd error sock since we've established a connexion */
1089 ret = lttng_poll_del(&events, consumer_data->err_sock);
1090 if (ret < 0) {
1091 goto error;
1092 }
1093
1094 /* Add new accepted error socket. */
1095 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1096 if (ret < 0) {
1097 goto error;
1098 }
1099
1100 /* Add metadata socket that is successfully connected. */
1101 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1102 LPOLLIN | LPOLLRDHUP);
1103 if (ret < 0) {
1104 goto error;
1105 }
1106
1107 health_code_update();
1108
1109 /* Infinite blocking call, waiting for transmission */
1110 restart_poll:
1111 while (1) {
1112 health_poll_entry();
1113 ret = lttng_poll_wait(&events, -1);
1114 health_poll_exit();
1115 if (ret < 0) {
1116 /*
1117 * Restart interrupted system call.
1118 */
1119 if (errno == EINTR) {
1120 goto restart_poll;
1121 }
1122 goto error;
1123 }
1124
1125 nb_fd = ret;
1126
1127 for (i = 0; i < nb_fd; i++) {
1128 /* Fetch once the poll data */
1129 revents = LTTNG_POLL_GETEV(&events, i);
1130 pollfd = LTTNG_POLL_GETFD(&events, i);
1131
1132 health_code_update();
1133
1134 /* Thread quit pipe has been closed. Killing thread. */
1135 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1136 if (ret) {
1137 err = 0;
1138 goto exit;
1139 }
1140
1141 if (pollfd == sock) {
1142 /* Event on the consumerd socket */
1143 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1144 ERR("consumer err socket second poll error");
1145 goto error;
1146 }
1147 health_code_update();
1148 /* Wait for any kconsumerd error */
1149 ret = lttcomm_recv_unix_sock(sock, &code,
1150 sizeof(enum lttcomm_return_code));
1151 if (ret <= 0) {
1152 ERR("consumer closed the command socket");
1153 goto error;
1154 }
1155
1156 ERR("consumer return code : %s",
1157 lttcomm_get_readable_code(-code));
1158
1159 goto exit;
1160 } else if (pollfd == consumer_data->metadata_fd) {
1161 /* UST metadata requests */
1162 ret = ust_consumer_metadata_request(
1163 &consumer_data->metadata_sock);
1164 if (ret < 0) {
1165 ERR("Handling metadata request");
1166 goto error;
1167 }
1168 break;
1169 } else {
1170 ERR("Unknown pollfd");
1171 goto error;
1172 }
1173 }
1174 health_code_update();
1175 }
1176
1177 exit:
1178 error:
1179 /*
1180 * We lock here because we are about to close the sockets and some other
1181 * thread might be using them so get exclusive access which will abort all
1182 * other consumer command by other threads.
1183 */
1184 pthread_mutex_lock(&consumer_data->lock);
1185
1186 /* Immediately set the consumerd state to stopped */
1187 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1188 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1189 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1190 consumer_data->type == LTTNG_CONSUMER32_UST) {
1191 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1192 } else {
1193 /* Code flow error... */
1194 assert(0);
1195 }
1196
1197 if (consumer_data->err_sock >= 0) {
1198 ret = close(consumer_data->err_sock);
1199 if (ret) {
1200 PERROR("close");
1201 }
1202 consumer_data->err_sock = -1;
1203 }
1204 if (consumer_data->cmd_sock >= 0) {
1205 ret = close(consumer_data->cmd_sock);
1206 if (ret) {
1207 PERROR("close");
1208 }
1209 consumer_data->cmd_sock = -1;
1210 }
1211 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1212 ret = close(*consumer_data->metadata_sock.fd_ptr);
1213 if (ret) {
1214 PERROR("close");
1215 }
1216 }
1217
1218 if (sock >= 0) {
1219 ret = close(sock);
1220 if (ret) {
1221 PERROR("close");
1222 }
1223 }
1224
1225 unlink(consumer_data->err_unix_sock_path);
1226 unlink(consumer_data->cmd_unix_sock_path);
1227 consumer_data->pid = 0;
1228 pthread_mutex_unlock(&consumer_data->lock);
1229
1230 /* Cleanup metadata socket mutex. */
1231 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1232 free(consumer_data->metadata_sock.lock);
1233
1234 lttng_poll_clean(&events);
1235 error_poll:
1236 if (err) {
1237 health_error();
1238 ERR("Health error occurred in %s", __func__);
1239 }
1240 health_unregister(health_sessiond);
1241 DBG("consumer thread cleanup completed");
1242
1243 return NULL;
1244 }
1245
1246 /*
1247 * This thread manage application communication.
1248 */
1249 static void *thread_manage_apps(void *data)
1250 {
1251 int i, ret, pollfd, err = -1;
1252 ssize_t size_ret;
1253 uint32_t revents, nb_fd;
1254 struct lttng_poll_event events;
1255
1256 DBG("[thread] Manage application started");
1257
1258 rcu_register_thread();
1259 rcu_thread_online();
1260
1261 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1262
1263 if (testpoint(thread_manage_apps)) {
1264 goto error_testpoint;
1265 }
1266
1267 health_code_update();
1268
1269 ret = sessiond_set_thread_pollset(&events, 2);
1270 if (ret < 0) {
1271 goto error_poll_create;
1272 }
1273
1274 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1275 if (ret < 0) {
1276 goto error;
1277 }
1278
1279 if (testpoint(thread_manage_apps_before_loop)) {
1280 goto error;
1281 }
1282
1283 health_code_update();
1284
1285 while (1) {
1286 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1287
1288 /* Inifinite blocking call, waiting for transmission */
1289 restart:
1290 health_poll_entry();
1291 ret = lttng_poll_wait(&events, -1);
1292 health_poll_exit();
1293 if (ret < 0) {
1294 /*
1295 * Restart interrupted system call.
1296 */
1297 if (errno == EINTR) {
1298 goto restart;
1299 }
1300 goto error;
1301 }
1302
1303 nb_fd = ret;
1304
1305 for (i = 0; i < nb_fd; i++) {
1306 /* Fetch once the poll data */
1307 revents = LTTNG_POLL_GETEV(&events, i);
1308 pollfd = LTTNG_POLL_GETFD(&events, i);
1309
1310 health_code_update();
1311
1312 /* Thread quit pipe has been closed. Killing thread. */
1313 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1314 if (ret) {
1315 err = 0;
1316 goto exit;
1317 }
1318
1319 /* Inspect the apps cmd pipe */
1320 if (pollfd == apps_cmd_pipe[0]) {
1321 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1322 ERR("Apps command pipe error");
1323 goto error;
1324 } else if (revents & LPOLLIN) {
1325 int sock;
1326
1327 /* Empty pipe */
1328 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1329 if (size_ret < sizeof(sock)) {
1330 PERROR("read apps cmd pipe");
1331 goto error;
1332 }
1333
1334 health_code_update();
1335
1336 /*
1337 * We only monitor the error events of the socket. This
1338 * thread does not handle any incoming data from UST
1339 * (POLLIN).
1340 */
1341 ret = lttng_poll_add(&events, sock,
1342 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1343 if (ret < 0) {
1344 goto error;
1345 }
1346
1347 DBG("Apps with sock %d added to poll set", sock);
1348
1349 health_code_update();
1350
1351 break;
1352 }
1353 } else {
1354 /*
1355 * At this point, we know that a registered application made
1356 * the event at poll_wait.
1357 */
1358 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1359 /* Removing from the poll set */
1360 ret = lttng_poll_del(&events, pollfd);
1361 if (ret < 0) {
1362 goto error;
1363 }
1364
1365 /* Socket closed on remote end. */
1366 ust_app_unregister(pollfd);
1367 break;
1368 }
1369 }
1370
1371 health_code_update();
1372 }
1373 }
1374
1375 exit:
1376 error:
1377 lttng_poll_clean(&events);
1378 error_poll_create:
1379 error_testpoint:
1380 utils_close_pipe(apps_cmd_pipe);
1381 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1382
1383 /*
1384 * We don't clean the UST app hash table here since already registered
1385 * applications can still be controlled so let them be until the session
1386 * daemon dies or the applications stop.
1387 */
1388
1389 if (err) {
1390 health_error();
1391 ERR("Health error occurred in %s", __func__);
1392 }
1393 health_unregister(health_sessiond);
1394 DBG("Application communication apps thread cleanup complete");
1395 rcu_thread_offline();
1396 rcu_unregister_thread();
1397 return NULL;
1398 }
1399
1400 /*
1401 * Send a socket to a thread This is called from the dispatch UST registration
1402 * thread once all sockets are set for the application.
1403 *
1404 * The sock value can be invalid, we don't really care, the thread will handle
1405 * it and make the necessary cleanup if so.
1406 *
1407 * On success, return 0 else a negative value being the errno message of the
1408 * write().
1409 */
1410 static int send_socket_to_thread(int fd, int sock)
1411 {
1412 ssize_t ret;
1413
1414 /*
1415 * It's possible that the FD is set as invalid with -1 concurrently just
1416 * before calling this function being a shutdown state of the thread.
1417 */
1418 if (fd < 0) {
1419 ret = -EBADF;
1420 goto error;
1421 }
1422
1423 ret = lttng_write(fd, &sock, sizeof(sock));
1424 if (ret < sizeof(sock)) {
1425 PERROR("write apps pipe %d", fd);
1426 if (ret < 0) {
1427 ret = -errno;
1428 }
1429 goto error;
1430 }
1431
1432 /* All good. Don't send back the write positive ret value. */
1433 ret = 0;
1434 error:
1435 return (int) ret;
1436 }
1437
1438 /*
1439 * Sanitize the wait queue of the dispatch registration thread meaning removing
1440 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1441 * notify socket is never received.
1442 */
1443 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1444 {
1445 int ret, nb_fd = 0, i;
1446 unsigned int fd_added = 0;
1447 struct lttng_poll_event events;
1448 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1449
1450 assert(wait_queue);
1451
1452 lttng_poll_init(&events);
1453
1454 /* Just skip everything for an empty queue. */
1455 if (!wait_queue->count) {
1456 goto end;
1457 }
1458
1459 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1460 if (ret < 0) {
1461 goto error_create;
1462 }
1463
1464 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1465 &wait_queue->head, head) {
1466 assert(wait_node->app);
1467 ret = lttng_poll_add(&events, wait_node->app->sock,
1468 LPOLLHUP | LPOLLERR);
1469 if (ret < 0) {
1470 goto error;
1471 }
1472
1473 fd_added = 1;
1474 }
1475
1476 if (!fd_added) {
1477 goto end;
1478 }
1479
1480 /*
1481 * Poll but don't block so we can quickly identify the faulty events and
1482 * clean them afterwards from the wait queue.
1483 */
1484 ret = lttng_poll_wait(&events, 0);
1485 if (ret < 0) {
1486 goto error;
1487 }
1488 nb_fd = ret;
1489
1490 for (i = 0; i < nb_fd; i++) {
1491 /* Get faulty FD. */
1492 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1493 int pollfd = LTTNG_POLL_GETFD(&events, i);
1494
1495 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1496 &wait_queue->head, head) {
1497 if (pollfd == wait_node->app->sock &&
1498 (revents & (LPOLLHUP | LPOLLERR))) {
1499 cds_list_del(&wait_node->head);
1500 wait_queue->count--;
1501 ust_app_destroy(wait_node->app);
1502 free(wait_node);
1503 break;
1504 }
1505 }
1506 }
1507
1508 if (nb_fd > 0) {
1509 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1510 }
1511
1512 end:
1513 lttng_poll_clean(&events);
1514 return;
1515
1516 error:
1517 lttng_poll_clean(&events);
1518 error_create:
1519 ERR("Unable to sanitize wait queue");
1520 return;
1521 }
1522
1523 /*
1524 * Dispatch request from the registration threads to the application
1525 * communication thread.
1526 */
1527 static void *thread_dispatch_ust_registration(void *data)
1528 {
1529 int ret, err = -1;
1530 struct cds_wfq_node *node;
1531 struct ust_command *ust_cmd = NULL;
1532 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1533 struct ust_reg_wait_queue wait_queue = {
1534 .count = 0,
1535 };
1536
1537 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1538
1539 health_code_update();
1540
1541 CDS_INIT_LIST_HEAD(&wait_queue.head);
1542
1543 DBG("[thread] Dispatch UST command started");
1544
1545 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1546 health_code_update();
1547
1548 /* Atomically prepare the queue futex */
1549 futex_nto1_prepare(&ust_cmd_queue.futex);
1550
1551 do {
1552 struct ust_app *app = NULL;
1553 ust_cmd = NULL;
1554
1555 /*
1556 * Make sure we don't have node(s) that have hung up before receiving
1557 * the notify socket. This is to clean the list in order to avoid
1558 * memory leaks from notify socket that are never seen.
1559 */
1560 sanitize_wait_queue(&wait_queue);
1561
1562 health_code_update();
1563 /* Dequeue command for registration */
1564 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1565 if (node == NULL) {
1566 DBG("Woken up but nothing in the UST command queue");
1567 /* Continue thread execution */
1568 break;
1569 }
1570
1571 ust_cmd = caa_container_of(node, struct ust_command, node);
1572
1573 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1574 " gid:%d sock:%d name:%s (version %d.%d)",
1575 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1576 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1577 ust_cmd->sock, ust_cmd->reg_msg.name,
1578 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1579
1580 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1581 wait_node = zmalloc(sizeof(*wait_node));
1582 if (!wait_node) {
1583 PERROR("zmalloc wait_node dispatch");
1584 ret = close(ust_cmd->sock);
1585 if (ret < 0) {
1586 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1587 }
1588 lttng_fd_put(LTTNG_FD_APPS, 1);
1589 free(ust_cmd);
1590 goto error;
1591 }
1592 CDS_INIT_LIST_HEAD(&wait_node->head);
1593
1594 /* Create application object if socket is CMD. */
1595 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1596 ust_cmd->sock);
1597 if (!wait_node->app) {
1598 ret = close(ust_cmd->sock);
1599 if (ret < 0) {
1600 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1601 }
1602 lttng_fd_put(LTTNG_FD_APPS, 1);
1603 free(wait_node);
1604 free(ust_cmd);
1605 continue;
1606 }
1607 /*
1608 * Add application to the wait queue so we can set the notify
1609 * socket before putting this object in the global ht.
1610 */
1611 cds_list_add(&wait_node->head, &wait_queue.head);
1612 wait_queue.count++;
1613
1614 free(ust_cmd);
1615 /*
1616 * We have to continue here since we don't have the notify
1617 * socket and the application MUST be added to the hash table
1618 * only at that moment.
1619 */
1620 continue;
1621 } else {
1622 /*
1623 * Look for the application in the local wait queue and set the
1624 * notify socket if found.
1625 */
1626 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1627 &wait_queue.head, head) {
1628 health_code_update();
1629 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1630 wait_node->app->notify_sock = ust_cmd->sock;
1631 cds_list_del(&wait_node->head);
1632 wait_queue.count--;
1633 app = wait_node->app;
1634 free(wait_node);
1635 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1636 break;
1637 }
1638 }
1639
1640 /*
1641 * With no application at this stage the received socket is
1642 * basically useless so close it before we free the cmd data
1643 * structure for good.
1644 */
1645 if (!app) {
1646 ret = close(ust_cmd->sock);
1647 if (ret < 0) {
1648 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1649 }
1650 lttng_fd_put(LTTNG_FD_APPS, 1);
1651 }
1652 free(ust_cmd);
1653 }
1654
1655 if (app) {
1656 /*
1657 * @session_lock_list
1658 *
1659 * Lock the global session list so from the register up to the
1660 * registration done message, no thread can see the application
1661 * and change its state.
1662 */
1663 session_lock_list();
1664 rcu_read_lock();
1665
1666 /*
1667 * Add application to the global hash table. This needs to be
1668 * done before the update to the UST registry can locate the
1669 * application.
1670 */
1671 ust_app_add(app);
1672
1673 /* Set app version. This call will print an error if needed. */
1674 (void) ust_app_version(app);
1675
1676 /* Send notify socket through the notify pipe. */
1677 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1678 app->notify_sock);
1679 if (ret < 0) {
1680 rcu_read_unlock();
1681 session_unlock_list();
1682 /*
1683 * No notify thread, stop the UST tracing. However, this is
1684 * not an internal error of the this thread thus setting
1685 * the health error code to a normal exit.
1686 */
1687 err = 0;
1688 goto error;
1689 }
1690
1691 /*
1692 * Update newly registered application with the tracing
1693 * registry info already enabled information.
1694 */
1695 update_ust_app(app->sock);
1696
1697 /*
1698 * Don't care about return value. Let the manage apps threads
1699 * handle app unregistration upon socket close.
1700 */
1701 (void) ust_app_register_done(app->sock);
1702
1703 /*
1704 * Even if the application socket has been closed, send the app
1705 * to the thread and unregistration will take place at that
1706 * place.
1707 */
1708 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1709 if (ret < 0) {
1710 rcu_read_unlock();
1711 session_unlock_list();
1712 /*
1713 * No apps. thread, stop the UST tracing. However, this is
1714 * not an internal error of the this thread thus setting
1715 * the health error code to a normal exit.
1716 */
1717 err = 0;
1718 goto error;
1719 }
1720
1721 rcu_read_unlock();
1722 session_unlock_list();
1723 }
1724 } while (node != NULL);
1725
1726 health_poll_entry();
1727 /* Futex wait on queue. Blocking call on futex() */
1728 futex_nto1_wait(&ust_cmd_queue.futex);
1729 health_poll_exit();
1730 }
1731 /* Normal exit, no error */
1732 err = 0;
1733
1734 error:
1735 /* Clean up wait queue. */
1736 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1737 &wait_queue.head, head) {
1738 cds_list_del(&wait_node->head);
1739 wait_queue.count--;
1740 free(wait_node);
1741 }
1742
1743 DBG("Dispatch thread dying");
1744 if (err) {
1745 health_error();
1746 ERR("Health error occurred in %s", __func__);
1747 }
1748 health_unregister(health_sessiond);
1749 return NULL;
1750 }
1751
1752 /*
1753 * This thread manage application registration.
1754 */
1755 static void *thread_registration_apps(void *data)
1756 {
1757 int sock = -1, i, ret, pollfd, err = -1;
1758 uint32_t revents, nb_fd;
1759 struct lttng_poll_event events;
1760 /*
1761 * Get allocated in this thread, enqueued to a global queue, dequeued and
1762 * freed in the manage apps thread.
1763 */
1764 struct ust_command *ust_cmd = NULL;
1765
1766 DBG("[thread] Manage application registration started");
1767
1768 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1769
1770 if (testpoint(thread_registration_apps)) {
1771 goto error_testpoint;
1772 }
1773
1774 ret = lttcomm_listen_unix_sock(apps_sock);
1775 if (ret < 0) {
1776 goto error_listen;
1777 }
1778
1779 /*
1780 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1781 * more will be added to this poll set.
1782 */
1783 ret = sessiond_set_thread_pollset(&events, 2);
1784 if (ret < 0) {
1785 goto error_create_poll;
1786 }
1787
1788 /* Add the application registration socket */
1789 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1790 if (ret < 0) {
1791 goto error_poll_add;
1792 }
1793
1794 /* Notify all applications to register */
1795 ret = notify_ust_apps(1);
1796 if (ret < 0) {
1797 ERR("Failed to notify applications or create the wait shared memory.\n"
1798 "Execution continues but there might be problem for already\n"
1799 "running applications that wishes to register.");
1800 }
1801
1802 while (1) {
1803 DBG("Accepting application registration");
1804
1805 /* Inifinite blocking call, waiting for transmission */
1806 restart:
1807 health_poll_entry();
1808 ret = lttng_poll_wait(&events, -1);
1809 health_poll_exit();
1810 if (ret < 0) {
1811 /*
1812 * Restart interrupted system call.
1813 */
1814 if (errno == EINTR) {
1815 goto restart;
1816 }
1817 goto error;
1818 }
1819
1820 nb_fd = ret;
1821
1822 for (i = 0; i < nb_fd; i++) {
1823 health_code_update();
1824
1825 /* Fetch once the poll data */
1826 revents = LTTNG_POLL_GETEV(&events, i);
1827 pollfd = LTTNG_POLL_GETFD(&events, i);
1828
1829 /* Thread quit pipe has been closed. Killing thread. */
1830 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1831 if (ret) {
1832 err = 0;
1833 goto exit;
1834 }
1835
1836 /* Event on the registration socket */
1837 if (pollfd == apps_sock) {
1838 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1839 ERR("Register apps socket poll error");
1840 goto error;
1841 } else if (revents & LPOLLIN) {
1842 sock = lttcomm_accept_unix_sock(apps_sock);
1843 if (sock < 0) {
1844 goto error;
1845 }
1846
1847 /*
1848 * Set socket timeout for both receiving and ending.
1849 * app_socket_timeout is in seconds, whereas
1850 * lttcomm_setsockopt_rcv_timeout and
1851 * lttcomm_setsockopt_snd_timeout expect msec as
1852 * parameter.
1853 */
1854 (void) lttcomm_setsockopt_rcv_timeout(sock,
1855 app_socket_timeout * 1000);
1856 (void) lttcomm_setsockopt_snd_timeout(sock,
1857 app_socket_timeout * 1000);
1858
1859 /*
1860 * Set the CLOEXEC flag. Return code is useless because
1861 * either way, the show must go on.
1862 */
1863 (void) utils_set_fd_cloexec(sock);
1864
1865 /* Create UST registration command for enqueuing */
1866 ust_cmd = zmalloc(sizeof(struct ust_command));
1867 if (ust_cmd == NULL) {
1868 PERROR("ust command zmalloc");
1869 goto error;
1870 }
1871
1872 /*
1873 * Using message-based transmissions to ensure we don't
1874 * have to deal with partially received messages.
1875 */
1876 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1877 if (ret < 0) {
1878 ERR("Exhausted file descriptors allowed for applications.");
1879 free(ust_cmd);
1880 ret = close(sock);
1881 if (ret) {
1882 PERROR("close");
1883 }
1884 sock = -1;
1885 continue;
1886 }
1887
1888 health_code_update();
1889 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1890 if (ret < 0) {
1891 free(ust_cmd);
1892 /* Close socket of the application. */
1893 ret = close(sock);
1894 if (ret) {
1895 PERROR("close");
1896 }
1897 lttng_fd_put(LTTNG_FD_APPS, 1);
1898 sock = -1;
1899 continue;
1900 }
1901 health_code_update();
1902
1903 ust_cmd->sock = sock;
1904 sock = -1;
1905
1906 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1907 " gid:%d sock:%d name:%s (version %d.%d)",
1908 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1909 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1910 ust_cmd->sock, ust_cmd->reg_msg.name,
1911 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1912
1913 /*
1914 * Lock free enqueue the registration request. The red pill
1915 * has been taken! This apps will be part of the *system*.
1916 */
1917 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1918
1919 /*
1920 * Wake the registration queue futex. Implicit memory
1921 * barrier with the exchange in cds_wfq_enqueue.
1922 */
1923 futex_nto1_wake(&ust_cmd_queue.futex);
1924 }
1925 }
1926 }
1927 }
1928
1929 exit:
1930 error:
1931 if (err) {
1932 health_error();
1933 ERR("Health error occurred in %s", __func__);
1934 }
1935
1936 /* Notify that the registration thread is gone */
1937 notify_ust_apps(0);
1938
1939 if (apps_sock >= 0) {
1940 ret = close(apps_sock);
1941 if (ret) {
1942 PERROR("close");
1943 }
1944 }
1945 if (sock >= 0) {
1946 ret = close(sock);
1947 if (ret) {
1948 PERROR("close");
1949 }
1950 lttng_fd_put(LTTNG_FD_APPS, 1);
1951 }
1952 unlink(apps_unix_sock_path);
1953
1954 error_poll_add:
1955 lttng_poll_clean(&events);
1956 error_listen:
1957 error_create_poll:
1958 error_testpoint:
1959 DBG("UST Registration thread cleanup complete");
1960 health_unregister(health_sessiond);
1961
1962 return NULL;
1963 }
1964
1965 /*
1966 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1967 * exec or it will fails.
1968 */
1969 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1970 {
1971 int ret, clock_ret;
1972 struct timespec timeout;
1973
1974 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1975 consumer_data->consumer_thread_is_ready = 0;
1976
1977 /* Setup pthread condition */
1978 ret = pthread_condattr_init(&consumer_data->condattr);
1979 if (ret != 0) {
1980 errno = ret;
1981 PERROR("pthread_condattr_init consumer data");
1982 goto error;
1983 }
1984
1985 /*
1986 * Set the monotonic clock in order to make sure we DO NOT jump in time
1987 * between the clock_gettime() call and the timedwait call. See bug #324
1988 * for a more details and how we noticed it.
1989 */
1990 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1991 if (ret != 0) {
1992 errno = ret;
1993 PERROR("pthread_condattr_setclock consumer data");
1994 goto error;
1995 }
1996
1997 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1998 if (ret != 0) {
1999 errno = ret;
2000 PERROR("pthread_cond_init consumer data");
2001 goto error;
2002 }
2003
2004 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2005 consumer_data);
2006 if (ret != 0) {
2007 PERROR("pthread_create consumer");
2008 ret = -1;
2009 goto error;
2010 }
2011
2012 /* We are about to wait on a pthread condition */
2013 pthread_mutex_lock(&consumer_data->cond_mutex);
2014
2015 /* Get time for sem_timedwait absolute timeout */
2016 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2017 /*
2018 * Set the timeout for the condition timed wait even if the clock gettime
2019 * call fails since we might loop on that call and we want to avoid to
2020 * increment the timeout too many times.
2021 */
2022 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2023
2024 /*
2025 * The following loop COULD be skipped in some conditions so this is why we
2026 * set ret to 0 in order to make sure at least one round of the loop is
2027 * done.
2028 */
2029 ret = 0;
2030
2031 /*
2032 * Loop until the condition is reached or when a timeout is reached. Note
2033 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2034 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2035 * possible. This loop does not take any chances and works with both of
2036 * them.
2037 */
2038 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2039 if (clock_ret < 0) {
2040 PERROR("clock_gettime spawn consumer");
2041 /* Infinite wait for the consumerd thread to be ready */
2042 ret = pthread_cond_wait(&consumer_data->cond,
2043 &consumer_data->cond_mutex);
2044 } else {
2045 ret = pthread_cond_timedwait(&consumer_data->cond,
2046 &consumer_data->cond_mutex, &timeout);
2047 }
2048 }
2049
2050 /* Release the pthread condition */
2051 pthread_mutex_unlock(&consumer_data->cond_mutex);
2052
2053 if (ret != 0) {
2054 errno = ret;
2055 if (ret == ETIMEDOUT) {
2056 /*
2057 * Call has timed out so we kill the kconsumerd_thread and return
2058 * an error.
2059 */
2060 ERR("Condition timed out. The consumer thread was never ready."
2061 " Killing it");
2062 ret = pthread_cancel(consumer_data->thread);
2063 if (ret < 0) {
2064 PERROR("pthread_cancel consumer thread");
2065 }
2066 } else {
2067 PERROR("pthread_cond_wait failed consumer thread");
2068 }
2069 goto error;
2070 }
2071
2072 pthread_mutex_lock(&consumer_data->pid_mutex);
2073 if (consumer_data->pid == 0) {
2074 ERR("Consumerd did not start");
2075 pthread_mutex_unlock(&consumer_data->pid_mutex);
2076 goto error;
2077 }
2078 pthread_mutex_unlock(&consumer_data->pid_mutex);
2079
2080 return 0;
2081
2082 error:
2083 return ret;
2084 }
2085
2086 /*
2087 * Join consumer thread
2088 */
2089 static int join_consumer_thread(struct consumer_data *consumer_data)
2090 {
2091 void *status;
2092
2093 /* Consumer pid must be a real one. */
2094 if (consumer_data->pid > 0) {
2095 int ret;
2096 ret = kill(consumer_data->pid, SIGTERM);
2097 if (ret) {
2098 ERR("Error killing consumer daemon");
2099 return ret;
2100 }
2101 return pthread_join(consumer_data->thread, &status);
2102 } else {
2103 return 0;
2104 }
2105 }
2106
2107 /*
2108 * Fork and exec a consumer daemon (consumerd).
2109 *
2110 * Return pid if successful else -1.
2111 */
2112 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2113 {
2114 int ret;
2115 pid_t pid;
2116 const char *consumer_to_use;
2117 const char *verbosity;
2118 struct stat st;
2119
2120 DBG("Spawning consumerd");
2121
2122 pid = fork();
2123 if (pid == 0) {
2124 /*
2125 * Exec consumerd.
2126 */
2127 if (opt_verbose_consumer) {
2128 verbosity = "--verbose";
2129 } else {
2130 verbosity = "--quiet";
2131 }
2132 switch (consumer_data->type) {
2133 case LTTNG_CONSUMER_KERNEL:
2134 /*
2135 * Find out which consumerd to execute. We will first try the
2136 * 64-bit path, then the sessiond's installation directory, and
2137 * fallback on the 32-bit one,
2138 */
2139 DBG3("Looking for a kernel consumer at these locations:");
2140 DBG3(" 1) %s", consumerd64_bin);
2141 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2142 DBG3(" 3) %s", consumerd32_bin);
2143 if (stat(consumerd64_bin, &st) == 0) {
2144 DBG3("Found location #1");
2145 consumer_to_use = consumerd64_bin;
2146 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2147 DBG3("Found location #2");
2148 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2149 } else if (stat(consumerd32_bin, &st) == 0) {
2150 DBG3("Found location #3");
2151 consumer_to_use = consumerd32_bin;
2152 } else {
2153 DBG("Could not find any valid consumerd executable");
2154 break;
2155 }
2156 DBG("Using kernel consumer at: %s", consumer_to_use);
2157 execl(consumer_to_use,
2158 "lttng-consumerd", verbosity, "-k",
2159 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2160 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2161 "--group", tracing_group_name,
2162 NULL);
2163 break;
2164 case LTTNG_CONSUMER64_UST:
2165 {
2166 char *tmpnew = NULL;
2167
2168 if (consumerd64_libdir[0] != '\0') {
2169 char *tmp;
2170 size_t tmplen;
2171
2172 tmp = getenv("LD_LIBRARY_PATH");
2173 if (!tmp) {
2174 tmp = "";
2175 }
2176 tmplen = strlen("LD_LIBRARY_PATH=")
2177 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2178 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2179 if (!tmpnew) {
2180 ret = -ENOMEM;
2181 goto error;
2182 }
2183 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2184 strcat(tmpnew, consumerd64_libdir);
2185 if (tmp[0] != '\0') {
2186 strcat(tmpnew, ":");
2187 strcat(tmpnew, tmp);
2188 }
2189 ret = putenv(tmpnew);
2190 if (ret) {
2191 ret = -errno;
2192 free(tmpnew);
2193 goto error;
2194 }
2195 }
2196 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2197 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2198 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2199 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2200 "--group", tracing_group_name,
2201 NULL);
2202 if (consumerd64_libdir[0] != '\0') {
2203 free(tmpnew);
2204 }
2205 if (ret) {
2206 goto error;
2207 }
2208 break;
2209 }
2210 case LTTNG_CONSUMER32_UST:
2211 {
2212 char *tmpnew = NULL;
2213
2214 if (consumerd32_libdir[0] != '\0') {
2215 char *tmp;
2216 size_t tmplen;
2217
2218 tmp = getenv("LD_LIBRARY_PATH");
2219 if (!tmp) {
2220 tmp = "";
2221 }
2222 tmplen = strlen("LD_LIBRARY_PATH=")
2223 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2224 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2225 if (!tmpnew) {
2226 ret = -ENOMEM;
2227 goto error;
2228 }
2229 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2230 strcat(tmpnew, consumerd32_libdir);
2231 if (tmp[0] != '\0') {
2232 strcat(tmpnew, ":");
2233 strcat(tmpnew, tmp);
2234 }
2235 ret = putenv(tmpnew);
2236 if (ret) {
2237 ret = -errno;
2238 free(tmpnew);
2239 goto error;
2240 }
2241 }
2242 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2243 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2244 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2245 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2246 "--group", tracing_group_name,
2247 NULL);
2248 if (consumerd32_libdir[0] != '\0') {
2249 free(tmpnew);
2250 }
2251 if (ret) {
2252 goto error;
2253 }
2254 break;
2255 }
2256 default:
2257 PERROR("unknown consumer type");
2258 exit(EXIT_FAILURE);
2259 }
2260 if (errno != 0) {
2261 PERROR("kernel start consumer exec");
2262 }
2263 exit(EXIT_FAILURE);
2264 } else if (pid > 0) {
2265 ret = pid;
2266 } else {
2267 PERROR("start consumer fork");
2268 ret = -errno;
2269 }
2270 error:
2271 return ret;
2272 }
2273
2274 /*
2275 * Spawn the consumerd daemon and session daemon thread.
2276 */
2277 static int start_consumerd(struct consumer_data *consumer_data)
2278 {
2279 int ret;
2280
2281 /*
2282 * Set the listen() state on the socket since there is a possible race
2283 * between the exec() of the consumer daemon and this call if place in the
2284 * consumer thread. See bug #366 for more details.
2285 */
2286 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2287 if (ret < 0) {
2288 goto error;
2289 }
2290
2291 pthread_mutex_lock(&consumer_data->pid_mutex);
2292 if (consumer_data->pid != 0) {
2293 pthread_mutex_unlock(&consumer_data->pid_mutex);
2294 goto end;
2295 }
2296
2297 ret = spawn_consumerd(consumer_data);
2298 if (ret < 0) {
2299 ERR("Spawning consumerd failed");
2300 pthread_mutex_unlock(&consumer_data->pid_mutex);
2301 goto error;
2302 }
2303
2304 /* Setting up the consumer_data pid */
2305 consumer_data->pid = ret;
2306 DBG2("Consumer pid %d", consumer_data->pid);
2307 pthread_mutex_unlock(&consumer_data->pid_mutex);
2308
2309 DBG2("Spawning consumer control thread");
2310 ret = spawn_consumer_thread(consumer_data);
2311 if (ret < 0) {
2312 ERR("Fatal error spawning consumer control thread");
2313 goto error;
2314 }
2315
2316 end:
2317 return 0;
2318
2319 error:
2320 /* Cleanup already created sockets on error. */
2321 if (consumer_data->err_sock >= 0) {
2322 int err;
2323
2324 err = close(consumer_data->err_sock);
2325 if (err < 0) {
2326 PERROR("close consumer data error socket");
2327 }
2328 }
2329 return ret;
2330 }
2331
2332 /*
2333 * Setup necessary data for kernel tracer action.
2334 */
2335 static int init_kernel_tracer(void)
2336 {
2337 int ret;
2338
2339 /* Modprobe lttng kernel modules */
2340 ret = modprobe_lttng_control();
2341 if (ret < 0) {
2342 goto error;
2343 }
2344
2345 /* Open debugfs lttng */
2346 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2347 if (kernel_tracer_fd < 0) {
2348 DBG("Failed to open %s", module_proc_lttng);
2349 ret = -1;
2350 goto error_open;
2351 }
2352
2353 /* Validate kernel version */
2354 ret = kernel_validate_version(kernel_tracer_fd);
2355 if (ret < 0) {
2356 goto error_version;
2357 }
2358
2359 ret = modprobe_lttng_data();
2360 if (ret < 0) {
2361 goto error_modules;
2362 }
2363
2364 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2365 return 0;
2366
2367 error_version:
2368 modprobe_remove_lttng_control();
2369 ret = close(kernel_tracer_fd);
2370 if (ret) {
2371 PERROR("close");
2372 }
2373 kernel_tracer_fd = -1;
2374 return LTTNG_ERR_KERN_VERSION;
2375
2376 error_modules:
2377 ret = close(kernel_tracer_fd);
2378 if (ret) {
2379 PERROR("close");
2380 }
2381
2382 error_open:
2383 modprobe_remove_lttng_control();
2384
2385 error:
2386 WARN("No kernel tracer available");
2387 kernel_tracer_fd = -1;
2388 if (!is_root) {
2389 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2390 } else {
2391 return LTTNG_ERR_KERN_NA;
2392 }
2393 }
2394
2395
2396 /*
2397 * Copy consumer output from the tracing session to the domain session. The
2398 * function also applies the right modification on a per domain basis for the
2399 * trace files destination directory.
2400 *
2401 * Should *NOT* be called with RCU read-side lock held.
2402 */
2403 static int copy_session_consumer(int domain, struct ltt_session *session)
2404 {
2405 int ret;
2406 const char *dir_name;
2407 struct consumer_output *consumer;
2408
2409 assert(session);
2410 assert(session->consumer);
2411
2412 switch (domain) {
2413 case LTTNG_DOMAIN_KERNEL:
2414 DBG3("Copying tracing session consumer output in kernel session");
2415 /*
2416 * XXX: We should audit the session creation and what this function
2417 * does "extra" in order to avoid a destroy since this function is used
2418 * in the domain session creation (kernel and ust) only. Same for UST
2419 * domain.
2420 */
2421 if (session->kernel_session->consumer) {
2422 consumer_destroy_output(session->kernel_session->consumer);
2423 }
2424 session->kernel_session->consumer =
2425 consumer_copy_output(session->consumer);
2426 /* Ease our life a bit for the next part */
2427 consumer = session->kernel_session->consumer;
2428 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2429 break;
2430 case LTTNG_DOMAIN_JUL:
2431 case LTTNG_DOMAIN_UST:
2432 DBG3("Copying tracing session consumer output in UST session");
2433 if (session->ust_session->consumer) {
2434 consumer_destroy_output(session->ust_session->consumer);
2435 }
2436 session->ust_session->consumer =
2437 consumer_copy_output(session->consumer);
2438 /* Ease our life a bit for the next part */
2439 consumer = session->ust_session->consumer;
2440 dir_name = DEFAULT_UST_TRACE_DIR;
2441 break;
2442 default:
2443 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2444 goto error;
2445 }
2446
2447 /* Append correct directory to subdir */
2448 strncat(consumer->subdir, dir_name,
2449 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2450 DBG3("Copy session consumer subdir %s", consumer->subdir);
2451
2452 ret = LTTNG_OK;
2453
2454 error:
2455 return ret;
2456 }
2457
2458 /*
2459 * Create an UST session and add it to the session ust list.
2460 *
2461 * Should *NOT* be called with RCU read-side lock held.
2462 */
2463 static int create_ust_session(struct ltt_session *session,
2464 struct lttng_domain *domain)
2465 {
2466 int ret;
2467 struct ltt_ust_session *lus = NULL;
2468
2469 assert(session);
2470 assert(domain);
2471 assert(session->consumer);
2472
2473 switch (domain->type) {
2474 case LTTNG_DOMAIN_JUL:
2475 case LTTNG_DOMAIN_UST:
2476 break;
2477 default:
2478 ERR("Unknown UST domain on create session %d", domain->type);
2479 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2480 goto error;
2481 }
2482
2483 DBG("Creating UST session");
2484
2485 lus = trace_ust_create_session(session->id);
2486 if (lus == NULL) {
2487 ret = LTTNG_ERR_UST_SESS_FAIL;
2488 goto error;
2489 }
2490
2491 lus->uid = session->uid;
2492 lus->gid = session->gid;
2493 lus->output_traces = session->output_traces;
2494 lus->snapshot_mode = session->snapshot_mode;
2495 lus->live_timer_interval = session->live_timer;
2496 session->ust_session = lus;
2497
2498 /* Copy session output to the newly created UST session */
2499 ret = copy_session_consumer(domain->type, session);
2500 if (ret != LTTNG_OK) {
2501 goto error;
2502 }
2503
2504 return LTTNG_OK;
2505
2506 error:
2507 free(lus);
2508 session->ust_session = NULL;
2509 return ret;
2510 }
2511
2512 /*
2513 * Create a kernel tracer session then create the default channel.
2514 */
2515 static int create_kernel_session(struct ltt_session *session)
2516 {
2517 int ret;
2518
2519 DBG("Creating kernel session");
2520
2521 ret = kernel_create_session(session, kernel_tracer_fd);
2522 if (ret < 0) {
2523 ret = LTTNG_ERR_KERN_SESS_FAIL;
2524 goto error;
2525 }
2526
2527 /* Code flow safety */
2528 assert(session->kernel_session);
2529
2530 /* Copy session output to the newly created Kernel session */
2531 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2532 if (ret != LTTNG_OK) {
2533 goto error;
2534 }
2535
2536 /* Create directory(ies) on local filesystem. */
2537 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2538 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2539 ret = run_as_mkdir_recursive(
2540 session->kernel_session->consumer->dst.trace_path,
2541 S_IRWXU | S_IRWXG, session->uid, session->gid);
2542 if (ret < 0) {
2543 if (ret != -EEXIST) {
2544 ERR("Trace directory creation error");
2545 goto error;
2546 }
2547 }
2548 }
2549
2550 session->kernel_session->uid = session->uid;
2551 session->kernel_session->gid = session->gid;
2552 session->kernel_session->output_traces = session->output_traces;
2553 session->kernel_session->snapshot_mode = session->snapshot_mode;
2554
2555 return LTTNG_OK;
2556
2557 error:
2558 trace_kernel_destroy_session(session->kernel_session);
2559 session->kernel_session = NULL;
2560 return ret;
2561 }
2562
2563 /*
2564 * Count number of session permitted by uid/gid.
2565 */
2566 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2567 {
2568 unsigned int i = 0;
2569 struct ltt_session *session;
2570
2571 DBG("Counting number of available session for UID %d GID %d",
2572 uid, gid);
2573 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2574 /*
2575 * Only list the sessions the user can control.
2576 */
2577 if (!session_access_ok(session, uid, gid)) {
2578 continue;
2579 }
2580 i++;
2581 }
2582 return i;
2583 }
2584
2585 /*
2586 * Process the command requested by the lttng client within the command
2587 * context structure. This function make sure that the return structure (llm)
2588 * is set and ready for transmission before returning.
2589 *
2590 * Return any error encountered or 0 for success.
2591 *
2592 * "sock" is only used for special-case var. len data.
2593 *
2594 * Should *NOT* be called with RCU read-side lock held.
2595 */
2596 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2597 int *sock_error)
2598 {
2599 int ret = LTTNG_OK;
2600 int need_tracing_session = 1;
2601 int need_domain;
2602
2603 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2604
2605 *sock_error = 0;
2606
2607 switch (cmd_ctx->lsm->cmd_type) {
2608 case LTTNG_CREATE_SESSION:
2609 case LTTNG_CREATE_SESSION_SNAPSHOT:
2610 case LTTNG_CREATE_SESSION_LIVE:
2611 case LTTNG_DESTROY_SESSION:
2612 case LTTNG_LIST_SESSIONS:
2613 case LTTNG_LIST_DOMAINS:
2614 case LTTNG_START_TRACE:
2615 case LTTNG_STOP_TRACE:
2616 case LTTNG_DATA_PENDING:
2617 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2618 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2619 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2620 case LTTNG_SNAPSHOT_RECORD:
2621 need_domain = 0;
2622 break;
2623 default:
2624 need_domain = 1;
2625 }
2626
2627 if (opt_no_kernel && need_domain
2628 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2629 if (!is_root) {
2630 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2631 } else {
2632 ret = LTTNG_ERR_KERN_NA;
2633 }
2634 goto error;
2635 }
2636
2637 /* Deny register consumer if we already have a spawned consumer. */
2638 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2639 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2640 if (kconsumer_data.pid > 0) {
2641 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2642 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2643 goto error;
2644 }
2645 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2646 }
2647
2648 /*
2649 * Check for command that don't needs to allocate a returned payload. We do
2650 * this here so we don't have to make the call for no payload at each
2651 * command.
2652 */
2653 switch(cmd_ctx->lsm->cmd_type) {
2654 case LTTNG_LIST_SESSIONS:
2655 case LTTNG_LIST_TRACEPOINTS:
2656 case LTTNG_LIST_TRACEPOINT_FIELDS:
2657 case LTTNG_LIST_DOMAINS:
2658 case LTTNG_LIST_CHANNELS:
2659 case LTTNG_LIST_EVENTS:
2660 break;
2661 default:
2662 /* Setup lttng message with no payload */
2663 ret = setup_lttng_msg(cmd_ctx, 0);
2664 if (ret < 0) {
2665 /* This label does not try to unlock the session */
2666 goto init_setup_error;
2667 }
2668 }
2669
2670 /* Commands that DO NOT need a session. */
2671 switch (cmd_ctx->lsm->cmd_type) {
2672 case LTTNG_CREATE_SESSION:
2673 case LTTNG_CREATE_SESSION_SNAPSHOT:
2674 case LTTNG_CREATE_SESSION_LIVE:
2675 case LTTNG_CALIBRATE:
2676 case LTTNG_LIST_SESSIONS:
2677 case LTTNG_LIST_TRACEPOINTS:
2678 case LTTNG_LIST_TRACEPOINT_FIELDS:
2679 need_tracing_session = 0;
2680 break;
2681 default:
2682 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2683 /*
2684 * We keep the session list lock across _all_ commands
2685 * for now, because the per-session lock does not
2686 * handle teardown properly.
2687 */
2688 session_lock_list();
2689 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2690 if (cmd_ctx->session == NULL) {
2691 ret = LTTNG_ERR_SESS_NOT_FOUND;
2692 goto error;
2693 } else {
2694 /* Acquire lock for the session */
2695 session_lock(cmd_ctx->session);
2696 }
2697 break;
2698 }
2699
2700 if (!need_domain) {
2701 goto skip_domain;
2702 }
2703
2704 /*
2705 * Check domain type for specific "pre-action".
2706 */
2707 switch (cmd_ctx->lsm->domain.type) {
2708 case LTTNG_DOMAIN_KERNEL:
2709 if (!is_root) {
2710 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2711 goto error;
2712 }
2713
2714 /* Kernel tracer check */
2715 if (kernel_tracer_fd == -1) {
2716 /* Basically, load kernel tracer modules */
2717 ret = init_kernel_tracer();
2718 if (ret != 0) {
2719 goto error;
2720 }
2721 }
2722
2723 /* Consumer is in an ERROR state. Report back to client */
2724 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2725 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2726 goto error;
2727 }
2728
2729 /* Need a session for kernel command */
2730 if (need_tracing_session) {
2731 if (cmd_ctx->session->kernel_session == NULL) {
2732 ret = create_kernel_session(cmd_ctx->session);
2733 if (ret < 0) {
2734 ret = LTTNG_ERR_KERN_SESS_FAIL;
2735 goto error;
2736 }
2737 }
2738
2739 /* Start the kernel consumer daemon */
2740 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2741 if (kconsumer_data.pid == 0 &&
2742 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2743 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2744 ret = start_consumerd(&kconsumer_data);
2745 if (ret < 0) {
2746 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2747 goto error;
2748 }
2749 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2750 } else {
2751 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2752 }
2753
2754 /*
2755 * The consumer was just spawned so we need to add the socket to
2756 * the consumer output of the session if exist.
2757 */
2758 ret = consumer_create_socket(&kconsumer_data,
2759 cmd_ctx->session->kernel_session->consumer);
2760 if (ret < 0) {
2761 goto error;
2762 }
2763 }
2764
2765 break;
2766 case LTTNG_DOMAIN_JUL:
2767 case LTTNG_DOMAIN_UST:
2768 {
2769 if (!ust_app_supported()) {
2770 ret = LTTNG_ERR_NO_UST;
2771 goto error;
2772 }
2773 /* Consumer is in an ERROR state. Report back to client */
2774 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2775 ret = LTTNG_ERR_NO_USTCONSUMERD;
2776 goto error;
2777 }
2778
2779 if (need_tracing_session) {
2780 /* Create UST session if none exist. */
2781 if (cmd_ctx->session->ust_session == NULL) {
2782 ret = create_ust_session(cmd_ctx->session,
2783 &cmd_ctx->lsm->domain);
2784 if (ret != LTTNG_OK) {
2785 goto error;
2786 }
2787 }
2788
2789 /* Start the UST consumer daemons */
2790 /* 64-bit */
2791 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2792 if (consumerd64_bin[0] != '\0' &&
2793 ustconsumer64_data.pid == 0 &&
2794 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2795 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2796 ret = start_consumerd(&ustconsumer64_data);
2797 if (ret < 0) {
2798 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2799 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2800 goto error;
2801 }
2802
2803 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2804 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2805 } else {
2806 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2807 }
2808
2809 /*
2810 * Setup socket for consumer 64 bit. No need for atomic access
2811 * since it was set above and can ONLY be set in this thread.
2812 */
2813 ret = consumer_create_socket(&ustconsumer64_data,
2814 cmd_ctx->session->ust_session->consumer);
2815 if (ret < 0) {
2816 goto error;
2817 }
2818
2819 /* 32-bit */
2820 if (consumerd32_bin[0] != '\0' &&
2821 ustconsumer32_data.pid == 0 &&
2822 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2823 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2824 ret = start_consumerd(&ustconsumer32_data);
2825 if (ret < 0) {
2826 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2827 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2828 goto error;
2829 }
2830
2831 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2832 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2833 } else {
2834 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2835 }
2836
2837 /*
2838 * Setup socket for consumer 64 bit. No need for atomic access
2839 * since it was set above and can ONLY be set in this thread.
2840 */
2841 ret = consumer_create_socket(&ustconsumer32_data,
2842 cmd_ctx->session->ust_session->consumer);
2843 if (ret < 0) {
2844 goto error;
2845 }
2846 }
2847 break;
2848 }
2849 default:
2850 break;
2851 }
2852 skip_domain:
2853
2854 /* Validate consumer daemon state when start/stop trace command */
2855 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2856 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2857 switch (cmd_ctx->lsm->domain.type) {
2858 case LTTNG_DOMAIN_JUL:
2859 case LTTNG_DOMAIN_UST:
2860 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2861 ret = LTTNG_ERR_NO_USTCONSUMERD;
2862 goto error;
2863 }
2864 break;
2865 case LTTNG_DOMAIN_KERNEL:
2866 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2867 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2868 goto error;
2869 }
2870 break;
2871 }
2872 }
2873
2874 /*
2875 * Check that the UID or GID match that of the tracing session.
2876 * The root user can interact with all sessions.
2877 */
2878 if (need_tracing_session) {
2879 if (!session_access_ok(cmd_ctx->session,
2880 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2881 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2882 ret = LTTNG_ERR_EPERM;
2883 goto error;
2884 }
2885 }
2886
2887 /*
2888 * Send relayd information to consumer as soon as we have a domain and a
2889 * session defined.
2890 */
2891 if (cmd_ctx->session && need_domain) {
2892 /*
2893 * Setup relayd if not done yet. If the relayd information was already
2894 * sent to the consumer, this call will gracefully return.
2895 */
2896 ret = cmd_setup_relayd(cmd_ctx->session);
2897 if (ret != LTTNG_OK) {
2898 goto error;
2899 }
2900 }
2901
2902 /* Process by command type */
2903 switch (cmd_ctx->lsm->cmd_type) {
2904 case LTTNG_ADD_CONTEXT:
2905 {
2906 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2907 cmd_ctx->lsm->u.context.channel_name,
2908 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2909 break;
2910 }
2911 case LTTNG_DISABLE_CHANNEL:
2912 {
2913 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2914 cmd_ctx->lsm->u.disable.channel_name);
2915 break;
2916 }
2917 case LTTNG_DISABLE_EVENT:
2918 {
2919 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2920 cmd_ctx->lsm->u.disable.channel_name,
2921 cmd_ctx->lsm->u.disable.name);
2922 break;
2923 }
2924 case LTTNG_DISABLE_ALL_EVENT:
2925 {
2926 DBG("Disabling all events");
2927
2928 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2929 cmd_ctx->lsm->u.disable.channel_name);
2930 break;
2931 }
2932 case LTTNG_ENABLE_CHANNEL:
2933 {
2934 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2935 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2936 break;
2937 }
2938 case LTTNG_ENABLE_EVENT:
2939 {
2940 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2941 cmd_ctx->lsm->u.enable.channel_name,
2942 &cmd_ctx->lsm->u.enable.event, NULL, NULL, kernel_poll_pipe[1]);
2943 break;
2944 }
2945 case LTTNG_ENABLE_ALL_EVENT:
2946 {
2947 DBG("Enabling all events");
2948
2949 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2950 cmd_ctx->lsm->u.enable.channel_name,
2951 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2952 break;
2953 }
2954 case LTTNG_LIST_TRACEPOINTS:
2955 {
2956 struct lttng_event *events;
2957 ssize_t nb_events;
2958
2959 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2960 if (nb_events < 0) {
2961 /* Return value is a negative lttng_error_code. */
2962 ret = -nb_events;
2963 goto error;
2964 }
2965
2966 /*
2967 * Setup lttng message with payload size set to the event list size in
2968 * bytes and then copy list into the llm payload.
2969 */
2970 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2971 if (ret < 0) {
2972 free(events);
2973 goto setup_error;
2974 }
2975
2976 /* Copy event list into message payload */
2977 memcpy(cmd_ctx->llm->payload, events,
2978 sizeof(struct lttng_event) * nb_events);
2979
2980 free(events);
2981
2982 ret = LTTNG_OK;
2983 break;
2984 }
2985 case LTTNG_LIST_TRACEPOINT_FIELDS:
2986 {
2987 struct lttng_event_field *fields;
2988 ssize_t nb_fields;
2989
2990 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2991 &fields);
2992 if (nb_fields < 0) {
2993 /* Return value is a negative lttng_error_code. */
2994 ret = -nb_fields;
2995 goto error;
2996 }
2997
2998 /*
2999 * Setup lttng message with payload size set to the event list size in
3000 * bytes and then copy list into the llm payload.
3001 */
3002 ret = setup_lttng_msg(cmd_ctx,
3003 sizeof(struct lttng_event_field) * nb_fields);
3004 if (ret < 0) {
3005 free(fields);
3006 goto setup_error;
3007 }
3008
3009 /* Copy event list into message payload */
3010 memcpy(cmd_ctx->llm->payload, fields,
3011 sizeof(struct lttng_event_field) * nb_fields);
3012
3013 free(fields);
3014
3015 ret = LTTNG_OK;
3016 break;
3017 }
3018 case LTTNG_SET_CONSUMER_URI:
3019 {
3020 size_t nb_uri, len;
3021 struct lttng_uri *uris;
3022
3023 nb_uri = cmd_ctx->lsm->u.uri.size;
3024 len = nb_uri * sizeof(struct lttng_uri);
3025
3026 if (nb_uri == 0) {
3027 ret = LTTNG_ERR_INVALID;
3028 goto error;
3029 }
3030
3031 uris = zmalloc(len);
3032 if (uris == NULL) {
3033 ret = LTTNG_ERR_FATAL;
3034 goto error;
3035 }
3036
3037 /* Receive variable len data */
3038 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3039 ret = lttcomm_recv_unix_sock(sock, uris, len);
3040 if (ret <= 0) {
3041 DBG("No URIs received from client... continuing");
3042 *sock_error = 1;
3043 ret = LTTNG_ERR_SESSION_FAIL;
3044 free(uris);
3045 goto error;
3046 }
3047
3048 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3049 nb_uri, uris);
3050 if (ret != LTTNG_OK) {
3051 free(uris);
3052 goto error;
3053 }
3054
3055 /*
3056 * XXX: 0 means that this URI should be applied on the session. Should
3057 * be a DOMAIN enuam.
3058 */
3059 if (cmd_ctx->lsm->domain.type == 0) {
3060 /* Add the URI for the UST session if a consumer is present. */
3061 if (cmd_ctx->session->ust_session &&
3062 cmd_ctx->session->ust_session->consumer) {
3063 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3064 nb_uri, uris);
3065 } else if (cmd_ctx->session->kernel_session &&
3066 cmd_ctx->session->kernel_session->consumer) {
3067 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3068 cmd_ctx->session, nb_uri, uris);
3069 }
3070 }
3071
3072 free(uris);
3073
3074 break;
3075 }
3076 case LTTNG_START_TRACE:
3077 {
3078 ret = cmd_start_trace(cmd_ctx->session);
3079 break;
3080 }
3081 case LTTNG_STOP_TRACE:
3082 {
3083 ret = cmd_stop_trace(cmd_ctx->session);
3084 break;
3085 }
3086 case LTTNG_CREATE_SESSION:
3087 {
3088 size_t nb_uri, len;
3089 struct lttng_uri *uris = NULL;
3090
3091 nb_uri = cmd_ctx->lsm->u.uri.size;
3092 len = nb_uri * sizeof(struct lttng_uri);
3093
3094 if (nb_uri > 0) {
3095 uris = zmalloc(len);
3096 if (uris == NULL) {
3097 ret = LTTNG_ERR_FATAL;
3098 goto error;
3099 }
3100
3101 /* Receive variable len data */
3102 DBG("Waiting for %zu URIs from client ...", nb_uri);
3103 ret = lttcomm_recv_unix_sock(sock, uris, len);
3104 if (ret <= 0) {
3105 DBG("No URIs received from client... continuing");
3106 *sock_error = 1;
3107 ret = LTTNG_ERR_SESSION_FAIL;
3108 free(uris);
3109 goto error;
3110 }
3111
3112 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3113 DBG("Creating session with ONE network URI is a bad call");
3114 ret = LTTNG_ERR_SESSION_FAIL;
3115 free(uris);
3116 goto error;
3117 }
3118 }
3119
3120 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3121 &cmd_ctx->creds, 0);
3122
3123 free(uris);
3124
3125 break;
3126 }
3127 case LTTNG_DESTROY_SESSION:
3128 {
3129 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3130
3131 /* Set session to NULL so we do not unlock it after free. */
3132 cmd_ctx->session = NULL;
3133 break;
3134 }
3135 case LTTNG_LIST_DOMAINS:
3136 {
3137 ssize_t nb_dom;
3138 struct lttng_domain *domains;
3139
3140 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3141 if (nb_dom < 0) {
3142 /* Return value is a negative lttng_error_code. */
3143 ret = -nb_dom;
3144 goto error;
3145 }
3146
3147 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3148 if (ret < 0) {
3149 free(domains);
3150 goto setup_error;
3151 }
3152
3153 /* Copy event list into message payload */
3154 memcpy(cmd_ctx->llm->payload, domains,
3155 nb_dom * sizeof(struct lttng_domain));
3156
3157 free(domains);
3158
3159 ret = LTTNG_OK;
3160 break;
3161 }
3162 case LTTNG_LIST_CHANNELS:
3163 {
3164 int nb_chan;
3165 struct lttng_channel *channels;
3166
3167 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3168 cmd_ctx->session, &channels);
3169 if (nb_chan < 0) {
3170 /* Return value is a negative lttng_error_code. */
3171 ret = -nb_chan;
3172 goto error;
3173 }
3174
3175 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3176 if (ret < 0) {
3177 free(channels);
3178 goto setup_error;
3179 }
3180
3181 /* Copy event list into message payload */
3182 memcpy(cmd_ctx->llm->payload, channels,
3183 nb_chan * sizeof(struct lttng_channel));
3184
3185 free(channels);
3186
3187 ret = LTTNG_OK;
3188 break;
3189 }
3190 case LTTNG_LIST_EVENTS:
3191 {
3192 ssize_t nb_event;
3193 struct lttng_event *events = NULL;
3194
3195 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3196 cmd_ctx->lsm->u.list.channel_name, &events);
3197 if (nb_event < 0) {
3198 /* Return value is a negative lttng_error_code. */
3199 ret = -nb_event;
3200 goto error;
3201 }
3202
3203 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3204 if (ret < 0) {
3205 free(events);
3206 goto setup_error;
3207 }
3208
3209 /* Copy event list into message payload */
3210 memcpy(cmd_ctx->llm->payload, events,
3211 nb_event * sizeof(struct lttng_event));
3212
3213 free(events);
3214
3215 ret = LTTNG_OK;
3216 break;
3217 }
3218 case LTTNG_LIST_SESSIONS:
3219 {
3220 unsigned int nr_sessions;
3221
3222 session_lock_list();
3223 nr_sessions = lttng_sessions_count(
3224 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3225 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3226
3227 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3228 if (ret < 0) {
3229 session_unlock_list();
3230 goto setup_error;
3231 }
3232
3233 /* Filled the session array */
3234 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3235 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3236 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3237
3238 session_unlock_list();
3239
3240 ret = LTTNG_OK;
3241 break;
3242 }
3243 case LTTNG_CALIBRATE:
3244 {
3245 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3246 &cmd_ctx->lsm->u.calibrate);
3247 break;
3248 }
3249 case LTTNG_REGISTER_CONSUMER:
3250 {
3251 struct consumer_data *cdata;
3252
3253 switch (cmd_ctx->lsm->domain.type) {
3254 case LTTNG_DOMAIN_KERNEL:
3255 cdata = &kconsumer_data;
3256 break;
3257 default:
3258 ret = LTTNG_ERR_UND;
3259 goto error;
3260 }
3261
3262 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3263 cmd_ctx->lsm->u.reg.path, cdata);
3264 break;
3265 }
3266 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3267 {
3268 struct lttng_filter_bytecode *bytecode;
3269
3270 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3271 ret = LTTNG_ERR_FILTER_INVAL;
3272 goto error;
3273 }
3274 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3275 ret = LTTNG_ERR_FILTER_INVAL;
3276 goto error;
3277 }
3278 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3279 if (!bytecode) {
3280 ret = LTTNG_ERR_FILTER_NOMEM;
3281 goto error;
3282 }
3283 /* Receive var. len. data */
3284 DBG("Receiving var len data from client ...");
3285 ret = lttcomm_recv_unix_sock(sock, bytecode,
3286 cmd_ctx->lsm->u.enable.bytecode_len);
3287 if (ret <= 0) {
3288 DBG("Nothing recv() from client var len data... continuing");
3289 *sock_error = 1;
3290 ret = LTTNG_ERR_FILTER_INVAL;
3291 goto error;
3292 }
3293
3294 if (bytecode->len + sizeof(*bytecode)
3295 != cmd_ctx->lsm->u.enable.bytecode_len) {
3296 free(bytecode);
3297 ret = LTTNG_ERR_FILTER_INVAL;
3298 goto error;
3299 }
3300
3301 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3302 cmd_ctx->lsm->u.enable.channel_name,
3303 &cmd_ctx->lsm->u.enable.event, bytecode, NULL, kernel_poll_pipe[1]);
3304 break;
3305 }
3306 case LTTNG_DATA_PENDING:
3307 {
3308 ret = cmd_data_pending(cmd_ctx->session);
3309 break;
3310 }
3311 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3312 {
3313 struct lttcomm_lttng_output_id reply;
3314
3315 ret = cmd_snapshot_add_output(cmd_ctx->session,
3316 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3317 if (ret != LTTNG_OK) {
3318 goto error;
3319 }
3320
3321 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3322 if (ret < 0) {
3323 goto setup_error;
3324 }
3325
3326 /* Copy output list into message payload */
3327 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3328 ret = LTTNG_OK;
3329 break;
3330 }
3331 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3332 {
3333 ret = cmd_snapshot_del_output(cmd_ctx->session,
3334 &cmd_ctx->lsm->u.snapshot_output.output);
3335 break;
3336 }
3337 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3338 {
3339 ssize_t nb_output;
3340 struct lttng_snapshot_output *outputs = NULL;
3341
3342 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3343 if (nb_output < 0) {
3344 ret = -nb_output;
3345 goto error;
3346 }
3347
3348 ret = setup_lttng_msg(cmd_ctx,
3349 nb_output * sizeof(struct lttng_snapshot_output));
3350 if (ret < 0) {
3351 free(outputs);
3352 goto setup_error;
3353 }
3354
3355 if (outputs) {
3356 /* Copy output list into message payload */
3357 memcpy(cmd_ctx->llm->payload, outputs,
3358 nb_output * sizeof(struct lttng_snapshot_output));
3359 free(outputs);
3360 }
3361
3362 ret = LTTNG_OK;
3363 break;
3364 }
3365 case LTTNG_SNAPSHOT_RECORD:
3366 {
3367 ret = cmd_snapshot_record(cmd_ctx->session,
3368 &cmd_ctx->lsm->u.snapshot_record.output,
3369 cmd_ctx->lsm->u.snapshot_record.wait);
3370 break;
3371 }
3372 case LTTNG_CREATE_SESSION_SNAPSHOT:
3373 {
3374 size_t nb_uri, len;
3375 struct lttng_uri *uris = NULL;
3376
3377 nb_uri = cmd_ctx->lsm->u.uri.size;
3378 len = nb_uri * sizeof(struct lttng_uri);
3379
3380 if (nb_uri > 0) {
3381 uris = zmalloc(len);
3382 if (uris == NULL) {
3383 ret = LTTNG_ERR_FATAL;
3384 goto error;
3385 }
3386
3387 /* Receive variable len data */
3388 DBG("Waiting for %zu URIs from client ...", nb_uri);
3389 ret = lttcomm_recv_unix_sock(sock, uris, len);
3390 if (ret <= 0) {
3391 DBG("No URIs received from client... continuing");
3392 *sock_error = 1;
3393 ret = LTTNG_ERR_SESSION_FAIL;
3394 free(uris);
3395 goto error;
3396 }
3397
3398 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3399 DBG("Creating session with ONE network URI is a bad call");
3400 ret = LTTNG_ERR_SESSION_FAIL;
3401 free(uris);
3402 goto error;
3403 }
3404 }
3405
3406 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3407 nb_uri, &cmd_ctx->creds);
3408 free(uris);
3409 break;
3410 }
3411 case LTTNG_CREATE_SESSION_LIVE:
3412 {
3413 size_t nb_uri, len;
3414 struct lttng_uri *uris = NULL;
3415
3416 nb_uri = cmd_ctx->lsm->u.uri.size;
3417 len = nb_uri * sizeof(struct lttng_uri);
3418
3419 if (nb_uri > 0) {
3420 uris = zmalloc(len);
3421 if (uris == NULL) {
3422 ret = LTTNG_ERR_FATAL;
3423 goto error;
3424 }
3425
3426 /* Receive variable len data */
3427 DBG("Waiting for %zu URIs from client ...", nb_uri);
3428 ret = lttcomm_recv_unix_sock(sock, uris, len);
3429 if (ret <= 0) {
3430 DBG("No URIs received from client... continuing");
3431 *sock_error = 1;
3432 ret = LTTNG_ERR_SESSION_FAIL;
3433 free(uris);
3434 goto error;
3435 }
3436
3437 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3438 DBG("Creating session with ONE network URI is a bad call");
3439 ret = LTTNG_ERR_SESSION_FAIL;
3440 free(uris);
3441 goto error;
3442 }
3443 }
3444
3445 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3446 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3447 free(uris);
3448 break;
3449 }
3450 default:
3451 ret = LTTNG_ERR_UND;
3452 break;
3453 }
3454
3455 error:
3456 if (cmd_ctx->llm == NULL) {
3457 DBG("Missing llm structure. Allocating one.");
3458 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3459 goto setup_error;
3460 }
3461 }
3462 /* Set return code */
3463 cmd_ctx->llm->ret_code = ret;
3464 setup_error:
3465 if (cmd_ctx->session) {
3466 session_unlock(cmd_ctx->session);
3467 }
3468 if (need_tracing_session) {
3469 session_unlock_list();
3470 }
3471 init_setup_error:
3472 return ret;
3473 }
3474
3475 /*
3476 * Thread managing health check socket.
3477 */
3478 static void *thread_manage_health(void *data)
3479 {
3480 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3481 uint32_t revents, nb_fd;
3482 struct lttng_poll_event events;
3483 struct health_comm_msg msg;
3484 struct health_comm_reply reply;
3485
3486 DBG("[thread] Manage health check started");
3487
3488 rcu_register_thread();
3489
3490 /* We might hit an error path before this is created. */
3491 lttng_poll_init(&events);
3492
3493 /* Create unix socket */
3494 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3495 if (sock < 0) {
3496 ERR("Unable to create health check Unix socket");
3497 ret = -1;
3498 goto error;
3499 }
3500
3501 if (is_root) {
3502 /* lttng health client socket path permissions */
3503 ret = chown(health_unix_sock_path, 0,
3504 utils_get_group_id(tracing_group_name));
3505 if (ret < 0) {
3506 ERR("Unable to set group on %s", health_unix_sock_path);
3507 PERROR("chown");
3508 ret = -1;
3509 goto error;
3510 }
3511
3512 ret = chmod(health_unix_sock_path,
3513 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3514 if (ret < 0) {
3515 ERR("Unable to set permissions on %s", health_unix_sock_path);
3516 PERROR("chmod");
3517 ret = -1;
3518 goto error;
3519 }
3520 }
3521
3522 /*
3523 * Set the CLOEXEC flag. Return code is useless because either way, the
3524 * show must go on.
3525 */
3526 (void) utils_set_fd_cloexec(sock);
3527
3528 ret = lttcomm_listen_unix_sock(sock);
3529 if (ret < 0) {
3530 goto error;
3531 }
3532
3533 /*
3534 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3535 * more will be added to this poll set.
3536 */
3537 ret = sessiond_set_thread_pollset(&events, 2);
3538 if (ret < 0) {
3539 goto error;
3540 }
3541
3542 /* Add the application registration socket */
3543 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3544 if (ret < 0) {
3545 goto error;
3546 }
3547
3548 while (1) {
3549 DBG("Health check ready");
3550
3551 /* Inifinite blocking call, waiting for transmission */
3552 restart:
3553 ret = lttng_poll_wait(&events, -1);
3554 if (ret < 0) {
3555 /*
3556 * Restart interrupted system call.
3557 */
3558 if (errno == EINTR) {
3559 goto restart;
3560 }
3561 goto error;
3562 }
3563
3564 nb_fd = ret;
3565
3566 for (i = 0; i < nb_fd; i++) {
3567 /* Fetch once the poll data */
3568 revents = LTTNG_POLL_GETEV(&events, i);
3569 pollfd = LTTNG_POLL_GETFD(&events, i);
3570
3571 /* Thread quit pipe has been closed. Killing thread. */
3572 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3573 if (ret) {
3574 err = 0;
3575 goto exit;
3576 }
3577
3578 /* Event on the registration socket */
3579 if (pollfd == sock) {
3580 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3581 ERR("Health socket poll error");
3582 goto error;
3583 }
3584 }
3585 }
3586
3587 new_sock = lttcomm_accept_unix_sock(sock);
3588 if (new_sock < 0) {
3589 goto error;
3590 }
3591
3592 /*
3593 * Set the CLOEXEC flag. Return code is useless because either way, the
3594 * show must go on.
3595 */
3596 (void) utils_set_fd_cloexec(new_sock);
3597
3598 DBG("Receiving data from client for health...");
3599 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3600 if (ret <= 0) {
3601 DBG("Nothing recv() from client... continuing");
3602 ret = close(new_sock);
3603 if (ret) {
3604 PERROR("close");
3605 }
3606 new_sock = -1;
3607 continue;
3608 }
3609
3610 rcu_thread_online();
3611
3612 reply.ret_code = 0;
3613 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3614 /*
3615 * health_check_state returns 0 if health is
3616 * bad.
3617 */
3618 if (!health_check_state(health_sessiond, i)) {
3619 reply.ret_code |= 1ULL << i;
3620 }
3621 }
3622
3623 DBG2("Health check return value %" PRIx64, reply.ret_code);
3624
3625 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3626 if (ret < 0) {
3627 ERR("Failed to send health data back to client");
3628 }
3629
3630 /* End of transmission */
3631 ret = close(new_sock);
3632 if (ret) {
3633 PERROR("close");
3634 }
3635 new_sock = -1;
3636 }
3637
3638 exit:
3639 error:
3640 if (err) {
3641 ERR("Health error occurred in %s", __func__);
3642 }
3643 DBG("Health check thread dying");
3644 unlink(health_unix_sock_path);
3645 if (sock >= 0) {
3646 ret = close(sock);
3647 if (ret) {
3648 PERROR("close");
3649 }
3650 }
3651
3652 lttng_poll_clean(&events);
3653
3654 rcu_unregister_thread();
3655 return NULL;
3656 }
3657
3658 /*
3659 * This thread manage all clients request using the unix client socket for
3660 * communication.
3661 */
3662 static void *thread_manage_clients(void *data)
3663 {
3664 int sock = -1, ret, i, pollfd, err = -1;
3665 int sock_error;
3666 uint32_t revents, nb_fd;
3667 struct command_ctx *cmd_ctx = NULL;
3668 struct lttng_poll_event events;
3669
3670 DBG("[thread] Manage client started");
3671
3672 rcu_register_thread();
3673
3674 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3675
3676 if (testpoint(thread_manage_clients)) {
3677 goto error_testpoint;
3678 }
3679
3680 health_code_update();
3681
3682 ret = lttcomm_listen_unix_sock(client_sock);
3683 if (ret < 0) {
3684 goto error_listen;
3685 }
3686
3687 /*
3688 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3689 * more will be added to this poll set.
3690 */
3691 ret = sessiond_set_thread_pollset(&events, 2);
3692 if (ret < 0) {
3693 goto error_create_poll;
3694 }
3695
3696 /* Add the application registration socket */
3697 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3698 if (ret < 0) {
3699 goto error;
3700 }
3701
3702 /*
3703 * Notify parent pid that we are ready to accept command for client side.
3704 * This ppid is the one from the external process that spawned us.
3705 */
3706 if (opt_sig_parent) {
3707 kill(ppid, SIGUSR1);
3708 }
3709
3710 /* Notify the parent of the fork() process that we are ready. */
3711 if (opt_daemon) {
3712 kill(child_ppid, SIGUSR1);
3713 }
3714
3715 if (testpoint(thread_manage_clients_before_loop)) {
3716 goto error;
3717 }
3718
3719 health_code_update();
3720
3721 while (1) {
3722 DBG("Accepting client command ...");
3723
3724 /* Inifinite blocking call, waiting for transmission */
3725 restart:
3726 health_poll_entry();
3727 ret = lttng_poll_wait(&events, -1);
3728 health_poll_exit();
3729 if (ret < 0) {
3730 /*
3731 * Restart interrupted system call.
3732 */
3733 if (errno == EINTR) {
3734 goto restart;
3735 }
3736 goto error;
3737 }
3738
3739 nb_fd = ret;
3740
3741 for (i = 0; i < nb_fd; i++) {
3742 /* Fetch once the poll data */
3743 revents = LTTNG_POLL_GETEV(&events, i);
3744 pollfd = LTTNG_POLL_GETFD(&events, i);
3745
3746 health_code_update();
3747
3748 /* Thread quit pipe has been closed. Killing thread. */
3749 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3750 if (ret) {
3751 err = 0;
3752 goto exit;
3753 }
3754
3755 /* Event on the registration socket */
3756 if (pollfd == client_sock) {
3757 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3758 ERR("Client socket poll error");
3759 goto error;
3760 }
3761 }
3762 }
3763
3764 DBG("Wait for client response");
3765
3766 health_code_update();
3767
3768 sock = lttcomm_accept_unix_sock(client_sock);
3769 if (sock < 0) {
3770 goto error;
3771 }
3772
3773 /*
3774 * Set the CLOEXEC flag. Return code is useless because either way, the
3775 * show must go on.
3776 */
3777 (void) utils_set_fd_cloexec(sock);
3778
3779 /* Set socket option for credentials retrieval */
3780 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3781 if (ret < 0) {
3782 goto error;
3783 }
3784
3785 /* Allocate context command to process the client request */
3786 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3787 if (cmd_ctx == NULL) {
3788 PERROR("zmalloc cmd_ctx");
3789 goto error;
3790 }
3791
3792 /* Allocate data buffer for reception */
3793 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3794 if (cmd_ctx->lsm == NULL) {
3795 PERROR("zmalloc cmd_ctx->lsm");
3796 goto error;
3797 }
3798
3799 cmd_ctx->llm = NULL;
3800 cmd_ctx->session = NULL;
3801
3802 health_code_update();
3803
3804 /*
3805 * Data is received from the lttng client. The struct
3806 * lttcomm_session_msg (lsm) contains the command and data request of
3807 * the client.
3808 */
3809 DBG("Receiving data from client ...");
3810 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3811 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3812 if (ret <= 0) {
3813 DBG("Nothing recv() from client... continuing");
3814 ret = close(sock);
3815 if (ret) {
3816 PERROR("close");
3817 }
3818 sock = -1;
3819 clean_command_ctx(&cmd_ctx);
3820 continue;
3821 }
3822
3823 health_code_update();
3824
3825 // TODO: Validate cmd_ctx including sanity check for
3826 // security purpose.
3827
3828 rcu_thread_online();
3829 /*
3830 * This function dispatch the work to the kernel or userspace tracer
3831 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3832 * informations for the client. The command context struct contains
3833 * everything this function may needs.
3834 */
3835 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3836 rcu_thread_offline();
3837 if (ret < 0) {
3838 ret = close(sock);
3839 if (ret) {
3840 PERROR("close");
3841 }
3842 sock = -1;
3843 /*
3844 * TODO: Inform client somehow of the fatal error. At
3845 * this point, ret < 0 means that a zmalloc failed
3846 * (ENOMEM). Error detected but still accept
3847 * command, unless a socket error has been
3848 * detected.
3849 */
3850 clean_command_ctx(&cmd_ctx);
3851 continue;
3852 }
3853
3854 health_code_update();
3855
3856 DBG("Sending response (size: %d, retcode: %s)",
3857 cmd_ctx->lttng_msg_size,
3858 lttng_strerror(-cmd_ctx->llm->ret_code));
3859 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3860 if (ret < 0) {
3861 ERR("Failed to send data back to client");
3862 }
3863
3864 /* End of transmission */
3865 ret = close(sock);
3866 if (ret) {
3867 PERROR("close");
3868 }
3869 sock = -1;
3870
3871 clean_command_ctx(&cmd_ctx);
3872
3873 health_code_update();
3874 }
3875
3876 exit:
3877 error:
3878 if (sock >= 0) {
3879 ret = close(sock);
3880 if (ret) {
3881 PERROR("close");
3882 }
3883 }
3884
3885 lttng_poll_clean(&events);
3886 clean_command_ctx(&cmd_ctx);
3887
3888 error_listen:
3889 error_create_poll:
3890 error_testpoint:
3891 unlink(client_unix_sock_path);
3892 if (client_sock >= 0) {
3893 ret = close(client_sock);
3894 if (ret) {
3895 PERROR("close");
3896 }
3897 }
3898
3899 if (err) {
3900 health_error();
3901 ERR("Health error occurred in %s", __func__);
3902 }
3903
3904 health_unregister(health_sessiond);
3905
3906 DBG("Client thread dying");
3907
3908 rcu_unregister_thread();
3909 return NULL;
3910 }
3911
3912
3913 /*
3914 * usage function on stderr
3915 */
3916 static void usage(void)
3917 {
3918 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3919 fprintf(stderr, " -h, --help Display this usage.\n");
3920 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3921 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3922 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3923 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3924 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3925 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3926 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3927 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3928 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3929 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3930 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3931 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3932 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3933 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3934 fprintf(stderr, " -V, --version Show version number.\n");
3935 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
3936 fprintf(stderr, " -q, --quiet No output at all.\n");
3937 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3938 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3939 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3940 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3941 fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
3942 }
3943
3944 /*
3945 * daemon argument parsing
3946 */
3947 static int parse_args(int argc, char **argv)
3948 {
3949 int c;
3950
3951 static struct option long_options[] = {
3952 { "client-sock", 1, 0, 'c' },
3953 { "apps-sock", 1, 0, 'a' },
3954 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3955 { "kconsumerd-err-sock", 1, 0, 'E' },
3956 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3957 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3958 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3959 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3960 { "consumerd32-path", 1, 0, 'u' },
3961 { "consumerd32-libdir", 1, 0, 'U' },
3962 { "consumerd64-path", 1, 0, 't' },
3963 { "consumerd64-libdir", 1, 0, 'T' },
3964 { "daemonize", 0, 0, 'd' },
3965 { "sig-parent", 0, 0, 'S' },
3966 { "help", 0, 0, 'h' },
3967 { "group", 1, 0, 'g' },
3968 { "version", 0, 0, 'V' },
3969 { "quiet", 0, 0, 'q' },
3970 { "verbose", 0, 0, 'v' },
3971 { "verbose-consumer", 0, 0, 'Z' },
3972 { "no-kernel", 0, 0, 'N' },
3973 { "pidfile", 1, 0, 'p' },
3974 { "jul-tcp-port", 1, 0, 'J' },
3975 { NULL, 0, 0, 0 }
3976 };
3977
3978 while (1) {
3979 int option_index = 0;
3980 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
3981 long_options, &option_index);
3982 if (c == -1) {
3983 break;
3984 }
3985
3986 switch (c) {
3987 case 0:
3988 fprintf(stderr, "option %s", long_options[option_index].name);
3989 if (optarg) {
3990 fprintf(stderr, " with arg %s\n", optarg);
3991 }
3992 break;
3993 case 'c':
3994 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3995 break;
3996 case 'a':
3997 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3998 break;
3999 case 'd':
4000 opt_daemon = 1;
4001 break;
4002 case 'g':
4003 tracing_group_name = optarg;
4004 break;
4005 case 'h':
4006 usage();
4007 exit(EXIT_FAILURE);
4008 case 'V':
4009 fprintf(stdout, "%s\n", VERSION);
4010 exit(EXIT_SUCCESS);
4011 case 'S':
4012 opt_sig_parent = 1;
4013 break;
4014 case 'E':
4015 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4016 break;
4017 case 'C':
4018 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4019 break;
4020 case 'F':
4021 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4022 break;
4023 case 'D':
4024 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4025 break;
4026 case 'H':
4027 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4028 break;
4029 case 'G':
4030 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4031 break;
4032 case 'N':
4033 opt_no_kernel = 1;
4034 break;
4035 case 'q':
4036 lttng_opt_quiet = 1;
4037 break;
4038 case 'v':
4039 /* Verbose level can increase using multiple -v */
4040 lttng_opt_verbose += 1;
4041 break;
4042 case 'Z':
4043 opt_verbose_consumer += 1;
4044 break;
4045 case 'u':
4046 consumerd32_bin= optarg;
4047 break;
4048 case 'U':
4049 consumerd32_libdir = optarg;
4050 break;
4051 case 't':
4052 consumerd64_bin = optarg;
4053 break;
4054 case 'T':
4055 consumerd64_libdir = optarg;
4056 break;
4057 case 'p':
4058 opt_pidfile = optarg;
4059 break;
4060 case 'J': /* JUL TCP port. */
4061 {
4062 unsigned long v;
4063
4064 errno = 0;
4065 v = strtoul(optarg, NULL, 0);
4066 if (errno != 0 || !isdigit(optarg[0])) {
4067 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg);
4068 return -1;
4069 }
4070 if (v == 0 || v >= 65535) {
4071 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg);
4072 return -1;
4073 }
4074 jul_tcp_port = (uint32_t) v;
4075 DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
4076 break;
4077 }
4078 default:
4079 /* Unknown option or other error.
4080 * Error is printed by getopt, just return */
4081 return -1;
4082 }
4083 }
4084
4085 return 0;
4086 }
4087
4088 /*
4089 * Creates the two needed socket by the daemon.
4090 * apps_sock - The communication socket for all UST apps.
4091 * client_sock - The communication of the cli tool (lttng).
4092 */
4093 static int init_daemon_socket(void)
4094 {
4095 int ret = 0;
4096 mode_t old_umask;
4097
4098 old_umask = umask(0);
4099
4100 /* Create client tool unix socket */
4101 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4102 if (client_sock < 0) {
4103 ERR("Create unix sock failed: %s", client_unix_sock_path);
4104 ret = -1;
4105 goto end;
4106 }
4107
4108 /* Set the cloexec flag */
4109 ret = utils_set_fd_cloexec(client_sock);
4110 if (ret < 0) {
4111 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4112 "Continuing but note that the consumer daemon will have a "
4113 "reference to this socket on exec()", client_sock);
4114 }
4115
4116 /* File permission MUST be 660 */
4117 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4118 if (ret < 0) {
4119 ERR("Set file permissions failed: %s", client_unix_sock_path);
4120 PERROR("chmod");
4121 goto end;
4122 }
4123
4124 /* Create the application unix socket */
4125 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4126 if (apps_sock < 0) {
4127 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4128 ret = -1;
4129 goto end;
4130 }
4131
4132 /* Set the cloexec flag */
4133 ret = utils_set_fd_cloexec(apps_sock);
4134 if (ret < 0) {
4135 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4136 "Continuing but note that the consumer daemon will have a "
4137 "reference to this socket on exec()", apps_sock);
4138 }
4139
4140 /* File permission MUST be 666 */
4141 ret = chmod(apps_unix_sock_path,
4142 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4143 if (ret < 0) {
4144 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4145 PERROR("chmod");
4146 goto end;
4147 }
4148
4149 DBG3("Session daemon client socket %d and application socket %d created",
4150 client_sock, apps_sock);
4151
4152 end:
4153 umask(old_umask);
4154 return ret;
4155 }
4156
4157 /*
4158 * Check if the global socket is available, and if a daemon is answering at the
4159 * other side. If yes, error is returned.
4160 */
4161 static int check_existing_daemon(void)
4162 {
4163 /* Is there anybody out there ? */
4164 if (lttng_session_daemon_alive()) {
4165 return -EEXIST;
4166 }
4167
4168 return 0;
4169 }
4170
4171 /*
4172 * Set the tracing group gid onto the client socket.
4173 *
4174 * Race window between mkdir and chown is OK because we are going from more
4175 * permissive (root.root) to less permissive (root.tracing).
4176 */
4177 static int set_permissions(char *rundir)
4178 {
4179 int ret;
4180 gid_t gid;
4181
4182 gid = utils_get_group_id(tracing_group_name);
4183
4184 /* Set lttng run dir */
4185 ret = chown(rundir, 0, gid);
4186 if (ret < 0) {
4187 ERR("Unable to set group on %s", rundir);
4188 PERROR("chown");
4189 }
4190
4191 /*
4192 * Ensure all applications and tracing group can search the run
4193 * dir. Allow everyone to read the directory, since it does not
4194 * buy us anything to hide its content.
4195 */
4196 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4197 if (ret < 0) {
4198 ERR("Unable to set permissions on %s", rundir);
4199 PERROR("chmod");
4200 }
4201
4202 /* lttng client socket path */
4203 ret = chown(client_unix_sock_path, 0, gid);
4204 if (ret < 0) {
4205 ERR("Unable to set group on %s", client_unix_sock_path);
4206 PERROR("chown");
4207 }
4208
4209 /* kconsumer error socket path */
4210 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4211 if (ret < 0) {
4212 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4213 PERROR("chown");
4214 }
4215
4216 /* 64-bit ustconsumer error socket path */
4217 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4218 if (ret < 0) {
4219 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4220 PERROR("chown");
4221 }
4222
4223 /* 32-bit ustconsumer compat32 error socket path */
4224 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4225 if (ret < 0) {
4226 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4227 PERROR("chown");
4228 }
4229
4230 DBG("All permissions are set");
4231
4232 return ret;
4233 }
4234
4235 /*
4236 * Create the lttng run directory needed for all global sockets and pipe.
4237 */
4238 static int create_lttng_rundir(const char *rundir)
4239 {
4240 int ret;
4241
4242 DBG3("Creating LTTng run directory: %s", rundir);
4243
4244 ret = mkdir(rundir, S_IRWXU);
4245 if (ret < 0) {
4246 if (errno != EEXIST) {
4247 ERR("Unable to create %s", rundir);
4248 goto error;
4249 } else {
4250 ret = 0;
4251 }
4252 }
4253
4254 error:
4255 return ret;
4256 }
4257
4258 /*
4259 * Setup sockets and directory needed by the kconsumerd communication with the
4260 * session daemon.
4261 */
4262 static int set_consumer_sockets(struct consumer_data *consumer_data,
4263 const char *rundir)
4264 {
4265 int ret;
4266 char path[PATH_MAX];
4267
4268 switch (consumer_data->type) {
4269 case LTTNG_CONSUMER_KERNEL:
4270 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4271 break;
4272 case LTTNG_CONSUMER64_UST:
4273 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4274 break;
4275 case LTTNG_CONSUMER32_UST:
4276 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4277 break;
4278 default:
4279 ERR("Consumer type unknown");
4280 ret = -EINVAL;
4281 goto error;
4282 }
4283
4284 DBG2("Creating consumer directory: %s", path);
4285
4286 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4287 if (ret < 0) {
4288 if (errno != EEXIST) {
4289 PERROR("mkdir");
4290 ERR("Failed to create %s", path);
4291 goto error;
4292 }
4293 ret = -1;
4294 }
4295 if (is_root) {
4296 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4297 if (ret < 0) {
4298 ERR("Unable to set group on %s", path);
4299 PERROR("chown");
4300 goto error;
4301 }
4302 }
4303
4304 /* Create the kconsumerd error unix socket */
4305 consumer_data->err_sock =
4306 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4307 if (consumer_data->err_sock < 0) {
4308 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4309 ret = -1;
4310 goto error;
4311 }
4312
4313 /*
4314 * Set the CLOEXEC flag. Return code is useless because either way, the
4315 * show must go on.
4316 */
4317 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4318 if (ret < 0) {
4319 PERROR("utils_set_fd_cloexec");
4320 /* continue anyway */
4321 }
4322
4323 /* File permission MUST be 660 */
4324 ret = chmod(consumer_data->err_unix_sock_path,
4325 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4326 if (ret < 0) {
4327 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4328 PERROR("chmod");
4329 goto error;
4330 }
4331
4332 error:
4333 return ret;
4334 }
4335
4336 /*
4337 * Signal handler for the daemon
4338 *
4339 * Simply stop all worker threads, leaving main() return gracefully after
4340 * joining all threads and calling cleanup().
4341 */
4342 static void sighandler(int sig)
4343 {
4344 switch (sig) {
4345 case SIGPIPE:
4346 DBG("SIGPIPE caught");
4347 return;
4348 case SIGINT:
4349 DBG("SIGINT caught");
4350 stop_threads();
4351 break;
4352 case SIGTERM:
4353 DBG("SIGTERM caught");
4354 stop_threads();
4355 break;
4356 case SIGUSR1:
4357 CMM_STORE_SHARED(recv_child_signal, 1);
4358 break;
4359 default:
4360 break;
4361 }
4362 }
4363
4364 /*
4365 * Setup signal handler for :
4366 * SIGINT, SIGTERM, SIGPIPE
4367 */
4368 static int set_signal_handler(void)
4369 {
4370 int ret = 0;
4371 struct sigaction sa;
4372 sigset_t sigset;
4373
4374 if ((ret = sigemptyset(&sigset)) < 0) {
4375 PERROR("sigemptyset");
4376 return ret;
4377 }
4378
4379 sa.sa_handler = sighandler;
4380 sa.sa_mask = sigset;
4381 sa.sa_flags = 0;
4382 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4383 PERROR("sigaction");
4384 return ret;
4385 }
4386
4387 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4388 PERROR("sigaction");
4389 return ret;
4390 }
4391
4392 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4393 PERROR("sigaction");
4394 return ret;
4395 }
4396
4397 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4398 PERROR("sigaction");
4399 return ret;
4400 }
4401
4402 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4403
4404 return ret;
4405 }
4406
4407 /*
4408 * Set open files limit to unlimited. This daemon can open a large number of
4409 * file descriptors in order to consumer multiple kernel traces.
4410 */
4411 static void set_ulimit(void)
4412 {
4413 int ret;
4414 struct rlimit lim;
4415
4416 /* The kernel does not allowed an infinite limit for open files */
4417 lim.rlim_cur = 65535;
4418 lim.rlim_max = 65535;
4419
4420 ret = setrlimit(RLIMIT_NOFILE, &lim);
4421 if (ret < 0) {
4422 PERROR("failed to set open files limit");
4423 }
4424 }
4425
4426 /*
4427 * Write pidfile using the rundir and opt_pidfile.
4428 */
4429 static void write_pidfile(void)
4430 {
4431 int ret;
4432 char pidfile_path[PATH_MAX];
4433
4434 assert(rundir);
4435
4436 if (opt_pidfile) {
4437 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4438 } else {
4439 /* Build pidfile path from rundir and opt_pidfile. */
4440 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4441 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4442 if (ret < 0) {
4443 PERROR("snprintf pidfile path");
4444 goto error;
4445 }
4446 }
4447
4448 /*
4449 * Create pid file in rundir. Return value is of no importance. The
4450 * execution will continue even though we are not able to write the file.
4451 */
4452 (void) utils_create_pid_file(getpid(), pidfile_path);
4453
4454 error:
4455 return;
4456 }
4457
4458 /*
4459 * Write JUL TCP port using the rundir.
4460 */
4461 static void write_julport(void)
4462 {
4463 int ret;
4464 char path[PATH_MAX];
4465
4466 assert(rundir);
4467
4468 ret = snprintf(path, sizeof(path), "%s/"
4469 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
4470 if (ret < 0) {
4471 PERROR("snprintf julport path");
4472 goto error;
4473 }
4474
4475 /*
4476 * Create TCP JUL port file in rundir. Return value is of no importance.
4477 * The execution will continue even though we are not able to write the
4478 * file.
4479 */
4480 (void) utils_create_pid_file(jul_tcp_port, path);
4481
4482 error:
4483 return;
4484 }
4485
4486 /*
4487 * Daemonize this process by forking and making the parent wait for the child
4488 * to signal it indicating readiness. Once received, the parent successfully
4489 * quits.
4490 *
4491 * The child process undergoes the same action that daemon(3) does meaning
4492 * setsid, chdir, and dup /dev/null into 0, 1 and 2.
4493 *
4494 * Return 0 on success else -1 on error.
4495 */
4496 static int daemonize(void)
4497 {
4498 int ret;
4499 pid_t pid;
4500
4501 /* Get parent pid of this process. */
4502 child_ppid = getppid();
4503
4504 pid = fork();
4505 if (pid < 0) {
4506 PERROR("fork");
4507 goto error;
4508 } else if (pid == 0) {
4509 int fd;
4510 pid_t sid;
4511
4512 /* Child */
4513
4514 /*
4515 * Get the newly created parent pid so we can signal that process when
4516 * we are ready to operate.
4517 */
4518 child_ppid = getppid();
4519
4520 sid = setsid();
4521 if (sid < 0) {
4522 PERROR("setsid");
4523 goto error;
4524 }
4525
4526 /* Try to change directory to /. If we can't well at least notify. */
4527 ret = chdir("/");
4528 if (ret < 0) {
4529 PERROR("chdir");
4530 }
4531
4532 fd = open(_PATH_DEVNULL, O_RDWR, 0);
4533 if (fd < 0) {
4534 PERROR("open %s", _PATH_DEVNULL);
4535 /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */
4536 } else {
4537 (void) dup2(fd, STDIN_FILENO);
4538 (void) dup2(fd, STDOUT_FILENO);
4539 (void) dup2(fd, STDERR_FILENO);
4540 if (fd > 2) {
4541 ret = close(fd);
4542 if (ret < 0) {
4543 PERROR("close");
4544 }
4545 }
4546 }
4547 goto end;
4548 } else {
4549 /* Parent */
4550
4551 /*
4552 * Waiting for child to notify this parent that it can exit. Note that
4553 * sleep() is interrupted before the 1 second delay as soon as the
4554 * signal is received, so it will not cause visible delay for the
4555 * user.
4556 */
4557 while (!CMM_LOAD_SHARED(recv_child_signal)) {
4558 sleep(1);
4559 }
4560
4561 /*
4562 * From this point on, the parent can exit and the child is now an
4563 * operationnal session daemon ready to serve clients and applications.
4564 */
4565 exit(EXIT_SUCCESS);
4566 }
4567
4568 end:
4569 return 0;
4570
4571 error:
4572 return -1;
4573 }
4574
4575 /*
4576 * main
4577 */
4578 int main(int argc, char **argv)
4579 {
4580 int ret = 0;
4581 void *status;
4582 const char *home_path, *env_app_timeout;
4583
4584 init_kernel_workarounds();
4585
4586 rcu_register_thread();
4587
4588 if ((ret = set_signal_handler()) < 0) {
4589 goto error;
4590 }
4591
4592 setup_consumerd_path();
4593
4594 page_size = sysconf(_SC_PAGESIZE);
4595 if (page_size < 0) {
4596 PERROR("sysconf _SC_PAGESIZE");
4597 page_size = LONG_MAX;
4598 WARN("Fallback page size to %ld", page_size);
4599 }
4600
4601 /* Parse arguments */
4602 progname = argv[0];
4603 if ((ret = parse_args(argc, argv)) < 0) {
4604 goto error;
4605 }
4606
4607 /* Daemonize */
4608 if (opt_daemon) {
4609 int i;
4610
4611 ret = daemonize();
4612 if (ret < 0) {
4613 goto error;
4614 }
4615
4616 /*
4617 * We are in the child. Make sure all other file descriptors are
4618 * closed, in case we are called with more opened file descriptors than
4619 * the standard ones.
4620 */
4621 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4622 (void) close(i);
4623 }
4624 }
4625
4626 /* Create thread quit pipe */
4627 if ((ret = init_thread_quit_pipe()) < 0) {
4628 goto error;
4629 }
4630
4631 /* Check if daemon is UID = 0 */
4632 is_root = !getuid();
4633
4634 if (is_root) {
4635 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4636
4637 /* Create global run dir with root access */
4638 ret = create_lttng_rundir(rundir);
4639 if (ret < 0) {
4640 goto error;
4641 }
4642
4643 if (strlen(apps_unix_sock_path) == 0) {
4644 snprintf(apps_unix_sock_path, PATH_MAX,
4645 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4646 }
4647
4648 if (strlen(client_unix_sock_path) == 0) {
4649 snprintf(client_unix_sock_path, PATH_MAX,
4650 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4651 }
4652
4653 /* Set global SHM for ust */
4654 if (strlen(wait_shm_path) == 0) {
4655 snprintf(wait_shm_path, PATH_MAX,
4656 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4657 }
4658
4659 if (strlen(health_unix_sock_path) == 0) {
4660 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4661 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4662 }
4663
4664 /* Setup kernel consumerd path */
4665 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4666 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4667 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4668 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4669
4670 DBG2("Kernel consumer err path: %s",
4671 kconsumer_data.err_unix_sock_path);
4672 DBG2("Kernel consumer cmd path: %s",
4673 kconsumer_data.cmd_unix_sock_path);
4674 } else {
4675 home_path = utils_get_home_dir();
4676 if (home_path == NULL) {
4677 /* TODO: Add --socket PATH option */
4678 ERR("Can't get HOME directory for sockets creation.");
4679 ret = -EPERM;
4680 goto error;
4681 }
4682
4683 /*
4684 * Create rundir from home path. This will create something like
4685 * $HOME/.lttng
4686 */
4687 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4688 if (ret < 0) {
4689 ret = -ENOMEM;
4690 goto error;
4691 }
4692
4693 ret = create_lttng_rundir(rundir);
4694 if (ret < 0) {
4695 goto error;
4696 }
4697
4698 if (strlen(apps_unix_sock_path) == 0) {
4699 snprintf(apps_unix_sock_path, PATH_MAX,
4700 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4701 }
4702
4703 /* Set the cli tool unix socket path */
4704 if (strlen(client_unix_sock_path) == 0) {
4705 snprintf(client_unix_sock_path, PATH_MAX,
4706 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4707 }
4708
4709 /* Set global SHM for ust */
4710 if (strlen(wait_shm_path) == 0) {
4711 snprintf(wait_shm_path, PATH_MAX,
4712 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4713 }
4714
4715 /* Set health check Unix path */
4716 if (strlen(health_unix_sock_path) == 0) {
4717 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4718 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4719 }
4720 }
4721
4722 /* Set consumer initial state */
4723 kernel_consumerd_state = CONSUMER_STOPPED;
4724 ust_consumerd_state = CONSUMER_STOPPED;
4725
4726 DBG("Client socket path %s", client_unix_sock_path);
4727 DBG("Application socket path %s", apps_unix_sock_path);
4728 DBG("Application wait path %s", wait_shm_path);
4729 DBG("LTTng run directory path: %s", rundir);
4730
4731 /* 32 bits consumerd path setup */
4732 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4733 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4734 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4735 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4736
4737 DBG2("UST consumer 32 bits err path: %s",
4738 ustconsumer32_data.err_unix_sock_path);
4739 DBG2("UST consumer 32 bits cmd path: %s",
4740 ustconsumer32_data.cmd_unix_sock_path);
4741
4742 /* 64 bits consumerd path setup */
4743 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4744 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4745 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4746 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4747
4748 DBG2("UST consumer 64 bits err path: %s",
4749 ustconsumer64_data.err_unix_sock_path);
4750 DBG2("UST consumer 64 bits cmd path: %s",
4751 ustconsumer64_data.cmd_unix_sock_path);
4752
4753 /*
4754 * See if daemon already exist.
4755 */
4756 if ((ret = check_existing_daemon()) < 0) {
4757 ERR("Already running daemon.\n");
4758 /*
4759 * We do not goto exit because we must not cleanup()
4760 * because a daemon is already running.
4761 */
4762 goto error;
4763 }
4764
4765 /*
4766 * Init UST app hash table. Alloc hash table before this point since
4767 * cleanup() can get called after that point.
4768 */
4769 ust_app_ht_alloc();
4770
4771 /* Initialize JUL domain subsystem. */
4772 if ((ret = jul_init()) < 0) {
4773 /* ENOMEM at this point. */
4774 goto error;
4775 }
4776
4777 /* After this point, we can safely call cleanup() with "goto exit" */
4778
4779 /*
4780 * These actions must be executed as root. We do that *after* setting up
4781 * the sockets path because we MUST make the check for another daemon using
4782 * those paths *before* trying to set the kernel consumer sockets and init
4783 * kernel tracer.
4784 */
4785 if (is_root) {
4786 ret = set_consumer_sockets(&kconsumer_data, rundir);
4787 if (ret < 0) {
4788 goto exit;
4789 }
4790
4791 /* Setup kernel tracer */
4792 if (!opt_no_kernel) {
4793 init_kernel_tracer();
4794 }
4795
4796 /* Set ulimit for open files */
4797 set_ulimit();
4798 }
4799 /* init lttng_fd tracking must be done after set_ulimit. */
4800 lttng_fd_init();
4801
4802 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4803 if (ret < 0) {
4804 goto exit;
4805 }
4806
4807 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4808 if (ret < 0) {
4809 goto exit;
4810 }
4811
4812 /* Setup the needed unix socket */
4813 if ((ret = init_daemon_socket()) < 0) {
4814 goto exit;
4815 }
4816
4817 /* Set credentials to socket */
4818 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4819 goto exit;
4820 }
4821
4822 /* Get parent pid if -S, --sig-parent is specified. */
4823 if (opt_sig_parent) {
4824 ppid = getppid();
4825 }
4826
4827 /* Setup the kernel pipe for waking up the kernel thread */
4828 if (is_root && !opt_no_kernel) {
4829 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4830 goto exit;
4831 }
4832 }
4833
4834 /* Setup the thread ht_cleanup communication pipe. */
4835 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4836 goto exit;
4837 }
4838
4839 /* Setup the thread apps communication pipe. */
4840 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4841 goto exit;
4842 }
4843
4844 /* Setup the thread apps notify communication pipe. */
4845 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4846 goto exit;
4847 }
4848
4849 /* Initialize global buffer per UID and PID registry. */
4850 buffer_reg_init_uid_registry();
4851 buffer_reg_init_pid_registry();
4852
4853 /* Init UST command queue. */
4854 cds_wfq_init(&ust_cmd_queue.queue);
4855
4856 /*
4857 * Get session list pointer. This pointer MUST NOT be free(). This list is
4858 * statically declared in session.c
4859 */
4860 session_list_ptr = session_get_list();
4861
4862 /* Set up max poll set size */
4863 lttng_poll_set_max_size();
4864
4865 cmd_init();
4866
4867 /* Check for the application socket timeout env variable. */
4868 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4869 if (env_app_timeout) {
4870 app_socket_timeout = atoi(env_app_timeout);
4871 } else {
4872 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4873 }
4874
4875 write_pidfile();
4876 write_julport();
4877
4878 /* Initialize communication library */
4879 lttcomm_init();
4880 /* This is to get the TCP timeout value. */
4881 lttcomm_inet_init();
4882
4883 /*
4884 * Initialize the health check subsystem. This call should set the
4885 * appropriate time values.
4886 */
4887 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
4888 if (!health_sessiond) {
4889 PERROR("health_app_create error");
4890 goto exit_health_sessiond_cleanup;
4891 }
4892
4893 /* Create thread to clean up RCU hash tables */
4894 ret = pthread_create(&ht_cleanup_thread, NULL,
4895 thread_ht_cleanup, (void *) NULL);
4896 if (ret != 0) {
4897 PERROR("pthread_create ht_cleanup");
4898 goto exit_ht_cleanup;
4899 }
4900
4901 /* Create health-check thread */
4902 ret = pthread_create(&health_thread, NULL,
4903 thread_manage_health, (void *) NULL);
4904 if (ret != 0) {
4905 PERROR("pthread_create health");
4906 goto exit_health;
4907 }
4908
4909 /* Create thread to manage the client socket */
4910 ret = pthread_create(&client_thread, NULL,
4911 thread_manage_clients, (void *) NULL);
4912 if (ret != 0) {
4913 PERROR("pthread_create clients");
4914 goto exit_client;
4915 }
4916
4917 /* Create thread to dispatch registration */
4918 ret = pthread_create(&dispatch_thread, NULL,
4919 thread_dispatch_ust_registration, (void *) NULL);
4920 if (ret != 0) {
4921 PERROR("pthread_create dispatch");
4922 goto exit_dispatch;
4923 }
4924
4925 /* Create thread to manage application registration. */
4926 ret = pthread_create(&reg_apps_thread, NULL,
4927 thread_registration_apps, (void *) NULL);
4928 if (ret != 0) {
4929 PERROR("pthread_create registration");
4930 goto exit_reg_apps;
4931 }
4932
4933 /* Create thread to manage application socket */
4934 ret = pthread_create(&apps_thread, NULL,
4935 thread_manage_apps, (void *) NULL);
4936 if (ret != 0) {
4937 PERROR("pthread_create apps");
4938 goto exit_apps;
4939 }
4940
4941 /* Create thread to manage application notify socket */
4942 ret = pthread_create(&apps_notify_thread, NULL,
4943 ust_thread_manage_notify, (void *) NULL);
4944 if (ret != 0) {
4945 PERROR("pthread_create apps");
4946 goto exit_apps_notify;
4947 }
4948
4949 /* Create JUL registration thread. */
4950 ret = pthread_create(&jul_reg_thread, NULL,
4951 jul_thread_manage_registration, (void *) NULL);
4952 if (ret != 0) {
4953 PERROR("pthread_create apps");
4954 goto exit_jul_reg;
4955 }
4956
4957 /* Don't start this thread if kernel tracing is not requested nor root */
4958 if (is_root && !opt_no_kernel) {
4959 /* Create kernel thread to manage kernel event */
4960 ret = pthread_create(&kernel_thread, NULL,
4961 thread_manage_kernel, (void *) NULL);
4962 if (ret != 0) {
4963 PERROR("pthread_create kernel");
4964 goto exit_kernel;
4965 }
4966
4967 ret = pthread_join(kernel_thread, &status);
4968 if (ret != 0) {
4969 PERROR("pthread_join");
4970 goto error; /* join error, exit without cleanup */
4971 }
4972 }
4973
4974 exit_kernel:
4975 ret = pthread_join(jul_reg_thread, &status);
4976 if (ret != 0) {
4977 PERROR("pthread_join JUL");
4978 goto error; /* join error, exit without cleanup */
4979 }
4980
4981 exit_jul_reg:
4982 ret = pthread_join(apps_notify_thread, &status);
4983 if (ret != 0) {
4984 PERROR("pthread_join apps notify");
4985 goto error; /* join error, exit without cleanup */
4986 }
4987
4988 exit_apps_notify:
4989 ret = pthread_join(apps_thread, &status);
4990 if (ret != 0) {
4991 PERROR("pthread_join apps");
4992 goto error; /* join error, exit without cleanup */
4993 }
4994
4995
4996 exit_apps:
4997 ret = pthread_join(reg_apps_thread, &status);
4998 if (ret != 0) {
4999 PERROR("pthread_join");
5000 goto error; /* join error, exit without cleanup */
5001 }
5002
5003 exit_reg_apps:
5004 ret = pthread_join(dispatch_thread, &status);
5005 if (ret != 0) {
5006 PERROR("pthread_join");
5007 goto error; /* join error, exit without cleanup */
5008 }
5009
5010 exit_dispatch:
5011 ret = pthread_join(client_thread, &status);
5012 if (ret != 0) {
5013 PERROR("pthread_join");
5014 goto error; /* join error, exit without cleanup */
5015 }
5016
5017 ret = join_consumer_thread(&kconsumer_data);
5018 if (ret != 0) {
5019 PERROR("join_consumer");
5020 goto error; /* join error, exit without cleanup */
5021 }
5022
5023 ret = join_consumer_thread(&ustconsumer32_data);
5024 if (ret != 0) {
5025 PERROR("join_consumer ust32");
5026 goto error; /* join error, exit without cleanup */
5027 }
5028
5029 ret = join_consumer_thread(&ustconsumer64_data);
5030 if (ret != 0) {
5031 PERROR("join_consumer ust64");
5032 goto error; /* join error, exit without cleanup */
5033 }
5034
5035 exit_client:
5036 ret = pthread_join(health_thread, &status);
5037 if (ret != 0) {
5038 PERROR("pthread_join health thread");
5039 goto error; /* join error, exit without cleanup */
5040 }
5041
5042 exit_health:
5043 ret = pthread_join(ht_cleanup_thread, &status);
5044 if (ret != 0) {
5045 PERROR("pthread_join ht cleanup thread");
5046 goto error; /* join error, exit without cleanup */
5047 }
5048 exit_ht_cleanup:
5049 health_app_destroy(health_sessiond);
5050 exit_health_sessiond_cleanup:
5051 exit:
5052 /*
5053 * cleanup() is called when no other thread is running.
5054 */
5055 rcu_thread_online();
5056 cleanup();
5057 rcu_thread_offline();
5058 rcu_unregister_thread();
5059 if (!ret) {
5060 exit(EXIT_SUCCESS);
5061 }
5062 error:
5063 exit(EXIT_FAILURE);
5064 }
This page took 0.186891 seconds and 3 git commands to generate.