Fix: quiet option is not set in sessiond-config
[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 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <ctype.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
52
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
55 #include "channel.h"
56 #include "cmd.h"
57 #include "consumer.h"
58 #include "context.h"
59 #include "event.h"
60 #include "kernel.h"
61 #include "kernel-consumer.h"
62 #include "modprobe.h"
63 #include "shm.h"
64 #include "ust-ctl.h"
65 #include "ust-consumer.h"
66 #include "utils.h"
67 #include "fd-limit.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
72 #include "save.h"
73 #include "load-session-thread.h"
74 #include "notification-thread.h"
75 #include "notification-thread-commands.h"
76 #include "syscall.h"
77 #include "agent.h"
78 #include "ht-cleanup.h"
79 #include "sessiond-config.h"
80
81 static const char *help_msg =
82 #ifdef LTTNG_EMBED_HELP
83 #include <lttng-sessiond.8.h>
84 #else
85 NULL
86 #endif
87 ;
88
89 const char *progname;
90 static pid_t ppid; /* Parent PID for --sig-parent option */
91 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
92 static int lockfile_fd = -1;
93
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal;
96
97 /*
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
100 */
101 static struct consumer_data kconsumer_data = {
102 .type = LTTNG_CONSUMER_KERNEL,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .channel_monitor_pipe = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .channel_monitor_pipe = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121 static struct consumer_data ustconsumer32_data = {
122 .type = LTTNG_CONSUMER32_UST,
123 .err_sock = -1,
124 .cmd_sock = -1,
125 .channel_monitor_pipe = -1,
126 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
127 .lock = PTHREAD_MUTEX_INITIALIZER,
128 .cond = PTHREAD_COND_INITIALIZER,
129 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
130 };
131
132 /* Command line options */
133 static const struct option long_options[] = {
134 { "client-sock", required_argument, 0, 'c' },
135 { "apps-sock", required_argument, 0, 'a' },
136 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
137 { "kconsumerd-err-sock", required_argument, 0, '\0' },
138 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
139 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
142 { "consumerd32-path", required_argument, 0, '\0' },
143 { "consumerd32-libdir", required_argument, 0, '\0' },
144 { "consumerd64-path", required_argument, 0, '\0' },
145 { "consumerd64-libdir", required_argument, 0, '\0' },
146 { "daemonize", no_argument, 0, 'd' },
147 { "background", no_argument, 0, 'b' },
148 { "sig-parent", no_argument, 0, 'S' },
149 { "help", no_argument, 0, 'h' },
150 { "group", required_argument, 0, 'g' },
151 { "version", no_argument, 0, 'V' },
152 { "quiet", no_argument, 0, 'q' },
153 { "verbose", no_argument, 0, 'v' },
154 { "verbose-consumer", no_argument, 0, '\0' },
155 { "no-kernel", no_argument, 0, '\0' },
156 { "pidfile", required_argument, 0, 'p' },
157 { "agent-tcp-port", required_argument, 0, '\0' },
158 { "config", required_argument, 0, 'f' },
159 { "load", required_argument, 0, 'l' },
160 { "kmod-probes", required_argument, 0, '\0' },
161 { "extra-kmod-probes", required_argument, 0, '\0' },
162 { NULL, 0, 0, 0 }
163 };
164
165 struct sessiond_config config;
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Sockets and FDs */
174 static int client_sock = -1;
175 static int apps_sock = -1;
176 int kernel_tracer_fd = -1;
177 static int kernel_poll_pipe[2] = { -1, -1 };
178
179 /*
180 * Quit pipe for all threads. This permits a single cancellation point
181 * for all threads when receiving an event on the pipe.
182 */
183 static int thread_quit_pipe[2] = { -1, -1 };
184
185 /*
186 * This pipe is used to inform the thread managing application communication
187 * that a command is queued and ready to be processed.
188 */
189 static int apps_cmd_pipe[2] = { -1, -1 };
190
191 int apps_cmd_notify_pipe[2] = { -1, -1 };
192
193 /* Pthread, Mutexes and Semaphores */
194 static pthread_t apps_thread;
195 static pthread_t apps_notify_thread;
196 static pthread_t reg_apps_thread;
197 static pthread_t client_thread;
198 static pthread_t kernel_thread;
199 static pthread_t dispatch_thread;
200 static pthread_t health_thread;
201 static pthread_t ht_cleanup_thread;
202 static pthread_t agent_reg_thread;
203 static pthread_t load_session_thread;
204 static pthread_t notification_thread;
205
206 /*
207 * UST registration command queue. This queue is tied with a futex and uses a N
208 * wakers / 1 waiter implemented and detailed in futex.c/.h
209 *
210 * The thread_registration_apps and thread_dispatch_ust_registration uses this
211 * queue along with the wait/wake scheme. The thread_manage_apps receives down
212 * the line new application socket and monitors it for any I/O error or clean
213 * close that triggers an unregistration of the application.
214 */
215 static struct ust_cmd_queue ust_cmd_queue;
216
217 /*
218 * Pointer initialized before thread creation.
219 *
220 * This points to the tracing session list containing the session count and a
221 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
222 * MUST NOT be taken if you call a public function in session.c.
223 *
224 * The lock is nested inside the structure: session_list_ptr->lock. Please use
225 * session_lock_list and session_unlock_list for lock acquisition.
226 */
227 static struct ltt_session_list *session_list_ptr;
228
229 int ust_consumerd64_fd = -1;
230 int ust_consumerd32_fd = -1;
231
232 static const char *module_proc_lttng = "/proc/lttng";
233
234 /*
235 * Consumer daemon state which is changed when spawning it, killing it or in
236 * case of a fatal error.
237 */
238 enum consumerd_state {
239 CONSUMER_STARTED = 1,
240 CONSUMER_STOPPED = 2,
241 CONSUMER_ERROR = 3,
242 };
243
244 /*
245 * This consumer daemon state is used to validate if a client command will be
246 * able to reach the consumer. If not, the client is informed. For instance,
247 * doing a "lttng start" when the consumer state is set to ERROR will return an
248 * error to the client.
249 *
250 * The following example shows a possible race condition of this scheme:
251 *
252 * consumer thread error happens
253 * client cmd arrives
254 * client cmd checks state -> still OK
255 * consumer thread exit, sets error
256 * client cmd try to talk to consumer
257 * ...
258 *
259 * However, since the consumer is a different daemon, we have no way of making
260 * sure the command will reach it safely even with this state flag. This is why
261 * we consider that up to the state validation during command processing, the
262 * command is safe. After that, we can not guarantee the correctness of the
263 * client request vis-a-vis the consumer.
264 */
265 static enum consumerd_state ust_consumerd_state;
266 static enum consumerd_state kernel_consumerd_state;
267
268 /* Set in main() with the current page size. */
269 long page_size;
270
271 /* Application health monitoring */
272 struct health_app *health_sessiond;
273
274 /* Am I root or not. */
275 int is_root; /* Set to 1 if the daemon is running as root */
276
277 const char * const config_section_name = "sessiond";
278
279 /* Load session thread information to operate. */
280 struct load_session_thread_data *load_info;
281
282 /* Notification thread handle. */
283 struct notification_thread_handle *notification_thread_handle;
284
285 /* Global hash tables */
286 struct lttng_ht *agent_apps_ht_by_sock = NULL;
287
288 /*
289 * Whether sessiond is ready for commands/notification channel/health check
290 * requests.
291 * NR_LTTNG_SESSIOND_READY must match the number of calls to
292 * sessiond_notify_ready().
293 */
294 #define NR_LTTNG_SESSIOND_READY 4
295 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
296
297 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
298 {
299 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
300 }
301
302 /* Notify parents that we are ready for cmd and health check */
303 LTTNG_HIDDEN
304 void sessiond_notify_ready(void)
305 {
306 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
307 /*
308 * Notify parent pid that we are ready to accept command
309 * for client side. This ppid is the one from the
310 * external process that spawned us.
311 */
312 if (config.sig_parent) {
313 kill(ppid, SIGUSR1);
314 }
315
316 /*
317 * Notify the parent of the fork() process that we are
318 * ready.
319 */
320 if (config.daemonize || config.background) {
321 kill(child_ppid, SIGUSR1);
322 }
323 }
324 }
325
326 static
327 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
328 int *a_pipe)
329 {
330 int ret;
331
332 assert(events);
333
334 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
335 if (ret < 0) {
336 goto error;
337 }
338
339 /* Add quit pipe */
340 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
341 if (ret < 0) {
342 goto error;
343 }
344
345 return 0;
346
347 error:
348 return ret;
349 }
350
351 /*
352 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
353 */
354 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
355 {
356 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
357 }
358
359 /*
360 * Init thread quit pipe.
361 *
362 * Return -1 on error or 0 if all pipes are created.
363 */
364 static int __init_thread_quit_pipe(int *a_pipe)
365 {
366 int ret, i;
367
368 ret = pipe(a_pipe);
369 if (ret < 0) {
370 PERROR("thread quit pipe");
371 goto error;
372 }
373
374 for (i = 0; i < 2; i++) {
375 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
376 if (ret < 0) {
377 PERROR("fcntl");
378 goto error;
379 }
380 }
381
382 error:
383 return ret;
384 }
385
386 static int init_thread_quit_pipe(void)
387 {
388 return __init_thread_quit_pipe(thread_quit_pipe);
389 }
390
391 /*
392 * Stop all threads by closing the thread quit pipe.
393 */
394 static void stop_threads(void)
395 {
396 int ret;
397
398 /* Stopping all threads */
399 DBG("Terminating all threads");
400 ret = notify_thread_pipe(thread_quit_pipe[1]);
401 if (ret < 0) {
402 ERR("write error on thread quit pipe");
403 }
404
405 /* Dispatch thread */
406 CMM_STORE_SHARED(dispatch_thread_exit, 1);
407 futex_nto1_wake(&ust_cmd_queue.futex);
408 }
409
410 /*
411 * Close every consumer sockets.
412 */
413 static void close_consumer_sockets(void)
414 {
415 int ret;
416
417 if (kconsumer_data.err_sock >= 0) {
418 ret = close(kconsumer_data.err_sock);
419 if (ret < 0) {
420 PERROR("kernel consumer err_sock close");
421 }
422 }
423 if (ustconsumer32_data.err_sock >= 0) {
424 ret = close(ustconsumer32_data.err_sock);
425 if (ret < 0) {
426 PERROR("UST consumerd32 err_sock close");
427 }
428 }
429 if (ustconsumer64_data.err_sock >= 0) {
430 ret = close(ustconsumer64_data.err_sock);
431 if (ret < 0) {
432 PERROR("UST consumerd64 err_sock close");
433 }
434 }
435 if (kconsumer_data.cmd_sock >= 0) {
436 ret = close(kconsumer_data.cmd_sock);
437 if (ret < 0) {
438 PERROR("kernel consumer cmd_sock close");
439 }
440 }
441 if (ustconsumer32_data.cmd_sock >= 0) {
442 ret = close(ustconsumer32_data.cmd_sock);
443 if (ret < 0) {
444 PERROR("UST consumerd32 cmd_sock close");
445 }
446 }
447 if (ustconsumer64_data.cmd_sock >= 0) {
448 ret = close(ustconsumer64_data.cmd_sock);
449 if (ret < 0) {
450 PERROR("UST consumerd64 cmd_sock close");
451 }
452 }
453 if (kconsumer_data.channel_monitor_pipe >= 0) {
454 ret = close(kconsumer_data.channel_monitor_pipe);
455 if (ret < 0) {
456 PERROR("kernel consumer channel monitor pipe close");
457 }
458 }
459 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
460 ret = close(ustconsumer32_data.channel_monitor_pipe);
461 if (ret < 0) {
462 PERROR("UST consumerd32 channel monitor pipe close");
463 }
464 }
465 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
466 ret = close(ustconsumer64_data.channel_monitor_pipe);
467 if (ret < 0) {
468 PERROR("UST consumerd64 channel monitor pipe close");
469 }
470 }
471 }
472
473 /*
474 * Wait on consumer process termination.
475 *
476 * Need to be called with the consumer data lock held or from a context
477 * ensuring no concurrent access to data (e.g: cleanup).
478 */
479 static void wait_consumer(struct consumer_data *consumer_data)
480 {
481 pid_t ret;
482 int status;
483
484 if (consumer_data->pid <= 0) {
485 return;
486 }
487
488 DBG("Waiting for complete teardown of consumerd (PID: %d)",
489 consumer_data->pid);
490 ret = waitpid(consumer_data->pid, &status, 0);
491 if (ret == -1) {
492 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
493 } else if (!WIFEXITED(status)) {
494 ERR("consumerd termination with error: %d",
495 WEXITSTATUS(ret));
496 }
497 consumer_data->pid = 0;
498 }
499
500 /*
501 * Cleanup the session daemon's data structures.
502 */
503 static void sessiond_cleanup(void)
504 {
505 int ret;
506 struct ltt_session *sess, *stmp;
507
508 DBG("Cleanup sessiond");
509
510 /*
511 * Close the thread quit pipe. It has already done its job,
512 * since we are now called.
513 */
514 utils_close_pipe(thread_quit_pipe);
515
516 /*
517 * If config.pid_file_path.value is undefined, the default file will be
518 * wiped when removing the rundir.
519 */
520 if (config.pid_file_path.value) {
521 ret = remove(config.pid_file_path.value);
522 if (ret < 0) {
523 PERROR("remove pidfile %s", config.pid_file_path.value);
524 }
525 }
526
527 DBG("Removing sessiond and consumerd content of directory %s",
528 config.rundir.value);
529
530 /* sessiond */
531 DBG("Removing %s", config.pid_file_path.value);
532 (void) unlink(config.pid_file_path.value);
533
534 DBG("Removing %s", config.agent_port_file_path.value);
535 (void) unlink(config.agent_port_file_path.value);
536
537 /* kconsumerd */
538 DBG("Removing %s", kconsumer_data.err_unix_sock_path);
539 (void) unlink(kconsumer_data.err_unix_sock_path);
540
541 DBG("Removing directory %s", config.kconsumerd_path.value);
542 (void) rmdir(config.kconsumerd_path.value);
543
544 /* ust consumerd 32 */
545 DBG("Removing %s", config.consumerd32_err_unix_sock_path.value);
546 (void) unlink(config.consumerd32_err_unix_sock_path.value);
547
548 DBG("Removing directory %s", config.consumerd32_path.value);
549 (void) rmdir(config.consumerd32_path.value);
550
551 /* ust consumerd 64 */
552 DBG("Removing %s", config.consumerd64_err_unix_sock_path.value);
553 (void) unlink(config.consumerd64_err_unix_sock_path.value);
554
555 DBG("Removing directory %s", config.consumerd64_path.value);
556 (void) rmdir(config.consumerd64_path.value);
557
558 DBG("Cleaning up all sessions");
559
560 /* Destroy session list mutex */
561 if (session_list_ptr != NULL) {
562 pthread_mutex_destroy(&session_list_ptr->lock);
563
564 /* Cleanup ALL session */
565 cds_list_for_each_entry_safe(sess, stmp,
566 &session_list_ptr->head, list) {
567 cmd_destroy_session(sess, kernel_poll_pipe[1]);
568 }
569 }
570
571 wait_consumer(&kconsumer_data);
572 wait_consumer(&ustconsumer64_data);
573 wait_consumer(&ustconsumer32_data);
574
575 DBG("Cleaning up all agent apps");
576 agent_app_ht_clean();
577
578 DBG("Closing all UST sockets");
579 ust_app_clean_list();
580 buffer_reg_destroy_registries();
581
582 if (is_root && !config.no_kernel) {
583 DBG2("Closing kernel fd");
584 if (kernel_tracer_fd >= 0) {
585 ret = close(kernel_tracer_fd);
586 if (ret) {
587 PERROR("close");
588 }
589 }
590 DBG("Unloading kernel modules");
591 modprobe_remove_lttng_all();
592 free(syscall_table);
593 }
594
595 close_consumer_sockets();
596
597 if (load_info) {
598 load_session_destroy_data(load_info);
599 free(load_info);
600 }
601
602 /*
603 * Cleanup lock file by deleting it and finaly closing it which will
604 * release the file system lock.
605 */
606 if (lockfile_fd >= 0) {
607 ret = remove(config.lock_file_path.value);
608 if (ret < 0) {
609 PERROR("remove lock file");
610 }
611 ret = close(lockfile_fd);
612 if (ret < 0) {
613 PERROR("close lock file");
614 }
615 }
616
617 /*
618 * We do NOT rmdir rundir because there are other processes
619 * using it, for instance lttng-relayd, which can start in
620 * parallel with this teardown.
621 */
622 }
623
624 /*
625 * Cleanup the daemon's option data structures.
626 */
627 static void sessiond_cleanup_options(void)
628 {
629 DBG("Cleaning up options");
630
631 sessiond_config_fini(&config);
632
633 run_as_destroy_worker();
634 }
635
636 /*
637 * Send data on a unix socket using the liblttsessiondcomm API.
638 *
639 * Return lttcomm error code.
640 */
641 static int send_unix_sock(int sock, void *buf, size_t len)
642 {
643 /* Check valid length */
644 if (len == 0) {
645 return -1;
646 }
647
648 return lttcomm_send_unix_sock(sock, buf, len);
649 }
650
651 /*
652 * Free memory of a command context structure.
653 */
654 static void clean_command_ctx(struct command_ctx **cmd_ctx)
655 {
656 DBG("Clean command context structure");
657 if (*cmd_ctx) {
658 if ((*cmd_ctx)->llm) {
659 free((*cmd_ctx)->llm);
660 }
661 if ((*cmd_ctx)->lsm) {
662 free((*cmd_ctx)->lsm);
663 }
664 free(*cmd_ctx);
665 *cmd_ctx = NULL;
666 }
667 }
668
669 /*
670 * Notify UST applications using the shm mmap futex.
671 */
672 static int notify_ust_apps(int active)
673 {
674 char *wait_shm_mmap;
675
676 DBG("Notifying applications of session daemon state: %d", active);
677
678 /* See shm.c for this call implying mmap, shm and futex calls */
679 wait_shm_mmap = shm_ust_get_mmap(config.wait_shm_path.value, is_root);
680 if (wait_shm_mmap == NULL) {
681 goto error;
682 }
683
684 /* Wake waiting process */
685 futex_wait_update((int32_t *) wait_shm_mmap, active);
686
687 /* Apps notified successfully */
688 return 0;
689
690 error:
691 return -1;
692 }
693
694 /*
695 * Setup the outgoing data buffer for the response (llm) by allocating the
696 * right amount of memory and copying the original information from the lsm
697 * structure.
698 *
699 * Return 0 on success, negative value on error.
700 */
701 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
702 const void *payload_buf, size_t payload_len,
703 const void *cmd_header_buf, size_t cmd_header_len)
704 {
705 int ret = 0;
706 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
707 const size_t cmd_header_offset = header_len;
708 const size_t payload_offset = cmd_header_offset + cmd_header_len;
709 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
710
711 cmd_ctx->llm = zmalloc(total_msg_size);
712
713 if (cmd_ctx->llm == NULL) {
714 PERROR("zmalloc");
715 ret = -ENOMEM;
716 goto end;
717 }
718
719 /* Copy common data */
720 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
721 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
722 cmd_ctx->llm->cmd_header_size = cmd_header_len;
723 cmd_ctx->llm->data_size = payload_len;
724 cmd_ctx->lttng_msg_size = total_msg_size;
725
726 /* Copy command header */
727 if (cmd_header_len) {
728 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
729 cmd_header_len);
730 }
731
732 /* Copy payload */
733 if (payload_len) {
734 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
735 payload_len);
736 }
737
738 end:
739 return ret;
740 }
741
742 /*
743 * Version of setup_lttng_msg() without command header.
744 */
745 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
746 void *payload_buf, size_t payload_len)
747 {
748 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
749 }
750 /*
751 * Update the kernel poll set of all channel fd available over all tracing
752 * session. Add the wakeup pipe at the end of the set.
753 */
754 static int update_kernel_poll(struct lttng_poll_event *events)
755 {
756 int ret;
757 struct ltt_session *session;
758 struct ltt_kernel_channel *channel;
759
760 DBG("Updating kernel poll set");
761
762 session_lock_list();
763 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
764 session_lock(session);
765 if (session->kernel_session == NULL) {
766 session_unlock(session);
767 continue;
768 }
769
770 cds_list_for_each_entry(channel,
771 &session->kernel_session->channel_list.head, list) {
772 /* Add channel fd to the kernel poll set */
773 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
774 if (ret < 0) {
775 session_unlock(session);
776 goto error;
777 }
778 DBG("Channel fd %d added to kernel set", channel->fd);
779 }
780 session_unlock(session);
781 }
782 session_unlock_list();
783
784 return 0;
785
786 error:
787 session_unlock_list();
788 return -1;
789 }
790
791 /*
792 * Find the channel fd from 'fd' over all tracing session. When found, check
793 * for new channel stream and send those stream fds to the kernel consumer.
794 *
795 * Useful for CPU hotplug feature.
796 */
797 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
798 {
799 int ret = 0;
800 struct ltt_session *session;
801 struct ltt_kernel_session *ksess;
802 struct ltt_kernel_channel *channel;
803
804 DBG("Updating kernel streams for channel fd %d", fd);
805
806 session_lock_list();
807 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
808 session_lock(session);
809 if (session->kernel_session == NULL) {
810 session_unlock(session);
811 continue;
812 }
813 ksess = session->kernel_session;
814
815 cds_list_for_each_entry(channel,
816 &ksess->channel_list.head, list) {
817 struct lttng_ht_iter iter;
818 struct consumer_socket *socket;
819
820 if (channel->fd != fd) {
821 continue;
822 }
823 DBG("Channel found, updating kernel streams");
824 ret = kernel_open_channel_stream(channel);
825 if (ret < 0) {
826 goto error;
827 }
828 /* Update the stream global counter */
829 ksess->stream_count_global += ret;
830
831 /*
832 * Have we already sent fds to the consumer? If yes, it
833 * means that tracing is started so it is safe to send
834 * our updated stream fds.
835 */
836 if (ksess->consumer_fds_sent != 1
837 || ksess->consumer == NULL) {
838 ret = -1;
839 goto error;
840 }
841
842 rcu_read_lock();
843 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
844 &iter.iter, socket, node.node) {
845 pthread_mutex_lock(socket->lock);
846 ret = kernel_consumer_send_channel_stream(socket,
847 channel, ksess,
848 session->output_traces ? 1 : 0);
849 pthread_mutex_unlock(socket->lock);
850 if (ret < 0) {
851 rcu_read_unlock();
852 goto error;
853 }
854 }
855 rcu_read_unlock();
856 }
857 session_unlock(session);
858 }
859 session_unlock_list();
860 return ret;
861
862 error:
863 session_unlock(session);
864 session_unlock_list();
865 return ret;
866 }
867
868 /*
869 * For each tracing session, update newly registered apps. The session list
870 * lock MUST be acquired before calling this.
871 */
872 static void update_ust_app(int app_sock)
873 {
874 struct ltt_session *sess, *stmp;
875
876 /* Consumer is in an ERROR state. Stop any application update. */
877 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
878 /* Stop the update process since the consumer is dead. */
879 return;
880 }
881
882 /* For all tracing session(s) */
883 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
884 struct ust_app *app;
885
886 session_lock(sess);
887 if (!sess->ust_session) {
888 goto unlock_session;
889 }
890
891 rcu_read_lock();
892 assert(app_sock >= 0);
893 app = ust_app_find_by_sock(app_sock);
894 if (app == NULL) {
895 /*
896 * Application can be unregistered before so
897 * this is possible hence simply stopping the
898 * update.
899 */
900 DBG3("UST app update failed to find app sock %d",
901 app_sock);
902 goto unlock_rcu;
903 }
904 ust_app_global_update(sess->ust_session, app);
905 unlock_rcu:
906 rcu_read_unlock();
907 unlock_session:
908 session_unlock(sess);
909 }
910 }
911
912 /*
913 * This thread manage event coming from the kernel.
914 *
915 * Features supported in this thread:
916 * -) CPU Hotplug
917 */
918 static void *thread_manage_kernel(void *data)
919 {
920 int ret, i, pollfd, update_poll_flag = 1, err = -1;
921 uint32_t revents, nb_fd;
922 char tmp;
923 struct lttng_poll_event events;
924
925 DBG("[thread] Thread manage kernel started");
926
927 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
928
929 /*
930 * This first step of the while is to clean this structure which could free
931 * non NULL pointers so initialize it before the loop.
932 */
933 lttng_poll_init(&events);
934
935 if (testpoint(sessiond_thread_manage_kernel)) {
936 goto error_testpoint;
937 }
938
939 health_code_update();
940
941 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
942 goto error_testpoint;
943 }
944
945 while (1) {
946 health_code_update();
947
948 if (update_poll_flag == 1) {
949 /* Clean events object. We are about to populate it again. */
950 lttng_poll_clean(&events);
951
952 ret = sessiond_set_thread_pollset(&events, 2);
953 if (ret < 0) {
954 goto error_poll_create;
955 }
956
957 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
958 if (ret < 0) {
959 goto error;
960 }
961
962 /* This will add the available kernel channel if any. */
963 ret = update_kernel_poll(&events);
964 if (ret < 0) {
965 goto error;
966 }
967 update_poll_flag = 0;
968 }
969
970 DBG("Thread kernel polling");
971
972 /* Poll infinite value of time */
973 restart:
974 health_poll_entry();
975 ret = lttng_poll_wait(&events, -1);
976 DBG("Thread kernel return from poll on %d fds",
977 LTTNG_POLL_GETNB(&events));
978 health_poll_exit();
979 if (ret < 0) {
980 /*
981 * Restart interrupted system call.
982 */
983 if (errno == EINTR) {
984 goto restart;
985 }
986 goto error;
987 } else if (ret == 0) {
988 /* Should not happen since timeout is infinite */
989 ERR("Return value of poll is 0 with an infinite timeout.\n"
990 "This should not have happened! Continuing...");
991 continue;
992 }
993
994 nb_fd = ret;
995
996 for (i = 0; i < nb_fd; i++) {
997 /* Fetch once the poll data */
998 revents = LTTNG_POLL_GETEV(&events, i);
999 pollfd = LTTNG_POLL_GETFD(&events, i);
1000
1001 health_code_update();
1002
1003 if (!revents) {
1004 /* No activity for this FD (poll implementation). */
1005 continue;
1006 }
1007
1008 /* Thread quit pipe has been closed. Killing thread. */
1009 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1010 if (ret) {
1011 err = 0;
1012 goto exit;
1013 }
1014
1015 /* Check for data on kernel pipe */
1016 if (revents & LPOLLIN) {
1017 if (pollfd == kernel_poll_pipe[0]) {
1018 (void) lttng_read(kernel_poll_pipe[0],
1019 &tmp, 1);
1020 /*
1021 * Ret value is useless here, if this pipe gets any actions an
1022 * update is required anyway.
1023 */
1024 update_poll_flag = 1;
1025 continue;
1026 } else {
1027 /*
1028 * New CPU detected by the kernel. Adding kernel stream to
1029 * kernel session and updating the kernel consumer
1030 */
1031 ret = update_kernel_stream(&kconsumer_data, pollfd);
1032 if (ret < 0) {
1033 continue;
1034 }
1035 break;
1036 }
1037 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1038 update_poll_flag = 1;
1039 continue;
1040 } else {
1041 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1042 goto error;
1043 }
1044 }
1045 }
1046
1047 exit:
1048 error:
1049 lttng_poll_clean(&events);
1050 error_poll_create:
1051 error_testpoint:
1052 utils_close_pipe(kernel_poll_pipe);
1053 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1054 if (err) {
1055 health_error();
1056 ERR("Health error occurred in %s", __func__);
1057 WARN("Kernel thread died unexpectedly. "
1058 "Kernel tracing can continue but CPU hotplug is disabled.");
1059 }
1060 health_unregister(health_sessiond);
1061 DBG("Kernel thread dying");
1062 return NULL;
1063 }
1064
1065 /*
1066 * Signal pthread condition of the consumer data that the thread.
1067 */
1068 static void signal_consumer_condition(struct consumer_data *data, int state)
1069 {
1070 pthread_mutex_lock(&data->cond_mutex);
1071
1072 /*
1073 * The state is set before signaling. It can be any value, it's the waiter
1074 * job to correctly interpret this condition variable associated to the
1075 * consumer pthread_cond.
1076 *
1077 * A value of 0 means that the corresponding thread of the consumer data
1078 * was not started. 1 indicates that the thread has started and is ready
1079 * for action. A negative value means that there was an error during the
1080 * thread bootstrap.
1081 */
1082 data->consumer_thread_is_ready = state;
1083 (void) pthread_cond_signal(&data->cond);
1084
1085 pthread_mutex_unlock(&data->cond_mutex);
1086 }
1087
1088 /*
1089 * This thread manage the consumer error sent back to the session daemon.
1090 */
1091 static void *thread_manage_consumer(void *data)
1092 {
1093 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1094 uint32_t revents, nb_fd;
1095 enum lttcomm_return_code code;
1096 struct lttng_poll_event events;
1097 struct consumer_data *consumer_data = data;
1098 struct consumer_socket *cmd_socket_wrapper = NULL;
1099
1100 DBG("[thread] Manage consumer started");
1101
1102 rcu_register_thread();
1103 rcu_thread_online();
1104
1105 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1106
1107 health_code_update();
1108
1109 /*
1110 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1111 * metadata_sock. Nothing more will be added to this poll set.
1112 */
1113 ret = sessiond_set_thread_pollset(&events, 3);
1114 if (ret < 0) {
1115 goto error_poll;
1116 }
1117
1118 /*
1119 * The error socket here is already in a listening state which was done
1120 * just before spawning this thread to avoid a race between the consumer
1121 * daemon exec trying to connect and the listen() call.
1122 */
1123 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1124 if (ret < 0) {
1125 goto error;
1126 }
1127
1128 health_code_update();
1129
1130 /* Infinite blocking call, waiting for transmission */
1131 restart:
1132 health_poll_entry();
1133
1134 if (testpoint(sessiond_thread_manage_consumer)) {
1135 goto error;
1136 }
1137
1138 ret = lttng_poll_wait(&events, -1);
1139 health_poll_exit();
1140 if (ret < 0) {
1141 /*
1142 * Restart interrupted system call.
1143 */
1144 if (errno == EINTR) {
1145 goto restart;
1146 }
1147 goto error;
1148 }
1149
1150 nb_fd = ret;
1151
1152 for (i = 0; i < nb_fd; i++) {
1153 /* Fetch once the poll data */
1154 revents = LTTNG_POLL_GETEV(&events, i);
1155 pollfd = LTTNG_POLL_GETFD(&events, i);
1156
1157 health_code_update();
1158
1159 if (!revents) {
1160 /* No activity for this FD (poll implementation). */
1161 continue;
1162 }
1163
1164 /* Thread quit pipe has been closed. Killing thread. */
1165 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1166 if (ret) {
1167 err = 0;
1168 goto exit;
1169 }
1170
1171 /* Event on the registration socket */
1172 if (pollfd == consumer_data->err_sock) {
1173 if (revents & LPOLLIN) {
1174 continue;
1175 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1176 ERR("consumer err socket poll error");
1177 goto error;
1178 } else {
1179 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1180 goto error;
1181 }
1182 }
1183 }
1184
1185 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1186 if (sock < 0) {
1187 goto error;
1188 }
1189
1190 /*
1191 * Set the CLOEXEC flag. Return code is useless because either way, the
1192 * show must go on.
1193 */
1194 (void) utils_set_fd_cloexec(sock);
1195
1196 health_code_update();
1197
1198 DBG2("Receiving code from consumer err_sock");
1199
1200 /* Getting status code from kconsumerd */
1201 ret = lttcomm_recv_unix_sock(sock, &code,
1202 sizeof(enum lttcomm_return_code));
1203 if (ret <= 0) {
1204 goto error;
1205 }
1206
1207 health_code_update();
1208 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1209 ERR("consumer error when waiting for SOCK_READY : %s",
1210 lttcomm_get_readable_code(-code));
1211 goto error;
1212 }
1213
1214 /* Connect both command and metadata sockets. */
1215 consumer_data->cmd_sock =
1216 lttcomm_connect_unix_sock(
1217 consumer_data->cmd_unix_sock_path);
1218 consumer_data->metadata_fd =
1219 lttcomm_connect_unix_sock(
1220 consumer_data->cmd_unix_sock_path);
1221 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1222 PERROR("consumer connect cmd socket");
1223 /* On error, signal condition and quit. */
1224 signal_consumer_condition(consumer_data, -1);
1225 goto error;
1226 }
1227
1228 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1229
1230 /* Create metadata socket lock. */
1231 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1232 if (consumer_data->metadata_sock.lock == NULL) {
1233 PERROR("zmalloc pthread mutex");
1234 goto error;
1235 }
1236 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1237
1238 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1239 DBG("Consumer metadata socket ready (fd: %d)",
1240 consumer_data->metadata_fd);
1241
1242 /*
1243 * Remove the consumerd error sock since we've established a connection.
1244 */
1245 ret = lttng_poll_del(&events, consumer_data->err_sock);
1246 if (ret < 0) {
1247 goto error;
1248 }
1249
1250 /* Add new accepted error socket. */
1251 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1252 if (ret < 0) {
1253 goto error;
1254 }
1255
1256 /* Add metadata socket that is successfully connected. */
1257 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1258 LPOLLIN | LPOLLRDHUP);
1259 if (ret < 0) {
1260 goto error;
1261 }
1262
1263 health_code_update();
1264
1265 /*
1266 * Transfer the write-end of the channel monitoring pipe to the
1267 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1268 */
1269 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1270 if (!cmd_socket_wrapper) {
1271 goto error;
1272 }
1273 cmd_socket_wrapper->lock = &consumer_data->lock;
1274
1275 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1276 consumer_data->channel_monitor_pipe);
1277 if (ret) {
1278 goto error;
1279 }
1280 /* Discard the socket wrapper as it is no longer needed. */
1281 consumer_destroy_socket(cmd_socket_wrapper);
1282 cmd_socket_wrapper = NULL;
1283
1284 /* The thread is completely initialized, signal that it is ready. */
1285 signal_consumer_condition(consumer_data, 1);
1286
1287 /* Infinite blocking call, waiting for transmission */
1288 restart_poll:
1289 while (1) {
1290 health_code_update();
1291
1292 /* Exit the thread because the thread quit pipe has been triggered. */
1293 if (should_quit) {
1294 /* Not a health error. */
1295 err = 0;
1296 goto exit;
1297 }
1298
1299 health_poll_entry();
1300 ret = lttng_poll_wait(&events, -1);
1301 health_poll_exit();
1302 if (ret < 0) {
1303 /*
1304 * Restart interrupted system call.
1305 */
1306 if (errno == EINTR) {
1307 goto restart_poll;
1308 }
1309 goto error;
1310 }
1311
1312 nb_fd = ret;
1313
1314 for (i = 0; i < nb_fd; i++) {
1315 /* Fetch once the poll data */
1316 revents = LTTNG_POLL_GETEV(&events, i);
1317 pollfd = LTTNG_POLL_GETFD(&events, i);
1318
1319 health_code_update();
1320
1321 if (!revents) {
1322 /* No activity for this FD (poll implementation). */
1323 continue;
1324 }
1325
1326 /*
1327 * Thread quit pipe has been triggered, flag that we should stop
1328 * but continue the current loop to handle potential data from
1329 * consumer.
1330 */
1331 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1332
1333 if (pollfd == sock) {
1334 /* Event on the consumerd socket */
1335 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1336 && !(revents & LPOLLIN)) {
1337 ERR("consumer err socket second poll error");
1338 goto error;
1339 }
1340 health_code_update();
1341 /* Wait for any kconsumerd error */
1342 ret = lttcomm_recv_unix_sock(sock, &code,
1343 sizeof(enum lttcomm_return_code));
1344 if (ret <= 0) {
1345 ERR("consumer closed the command socket");
1346 goto error;
1347 }
1348
1349 ERR("consumer return code : %s",
1350 lttcomm_get_readable_code(-code));
1351
1352 goto exit;
1353 } else if (pollfd == consumer_data->metadata_fd) {
1354 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1355 && !(revents & LPOLLIN)) {
1356 ERR("consumer err metadata socket second poll error");
1357 goto error;
1358 }
1359 /* UST metadata requests */
1360 ret = ust_consumer_metadata_request(
1361 &consumer_data->metadata_sock);
1362 if (ret < 0) {
1363 ERR("Handling metadata request");
1364 goto error;
1365 }
1366 }
1367 /* No need for an else branch all FDs are tested prior. */
1368 }
1369 health_code_update();
1370 }
1371
1372 exit:
1373 error:
1374 /*
1375 * We lock here because we are about to close the sockets and some other
1376 * thread might be using them so get exclusive access which will abort all
1377 * other consumer command by other threads.
1378 */
1379 pthread_mutex_lock(&consumer_data->lock);
1380
1381 /* Immediately set the consumerd state to stopped */
1382 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1383 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1384 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1385 consumer_data->type == LTTNG_CONSUMER32_UST) {
1386 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1387 } else {
1388 /* Code flow error... */
1389 assert(0);
1390 }
1391
1392 if (consumer_data->err_sock >= 0) {
1393 ret = close(consumer_data->err_sock);
1394 if (ret) {
1395 PERROR("close");
1396 }
1397 consumer_data->err_sock = -1;
1398 }
1399 if (consumer_data->cmd_sock >= 0) {
1400 ret = close(consumer_data->cmd_sock);
1401 if (ret) {
1402 PERROR("close");
1403 }
1404 consumer_data->cmd_sock = -1;
1405 }
1406 if (consumer_data->metadata_sock.fd_ptr &&
1407 *consumer_data->metadata_sock.fd_ptr >= 0) {
1408 ret = close(*consumer_data->metadata_sock.fd_ptr);
1409 if (ret) {
1410 PERROR("close");
1411 }
1412 }
1413 if (sock >= 0) {
1414 ret = close(sock);
1415 if (ret) {
1416 PERROR("close");
1417 }
1418 }
1419
1420 unlink(consumer_data->err_unix_sock_path);
1421 unlink(consumer_data->cmd_unix_sock_path);
1422 pthread_mutex_unlock(&consumer_data->lock);
1423
1424 /* Cleanup metadata socket mutex. */
1425 if (consumer_data->metadata_sock.lock) {
1426 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1427 free(consumer_data->metadata_sock.lock);
1428 }
1429 lttng_poll_clean(&events);
1430
1431 if (cmd_socket_wrapper) {
1432 consumer_destroy_socket(cmd_socket_wrapper);
1433 }
1434 error_poll:
1435 if (err) {
1436 health_error();
1437 ERR("Health error occurred in %s", __func__);
1438 }
1439 health_unregister(health_sessiond);
1440 DBG("consumer thread cleanup completed");
1441
1442 rcu_thread_offline();
1443 rcu_unregister_thread();
1444
1445 return NULL;
1446 }
1447
1448 /*
1449 * This thread manage application communication.
1450 */
1451 static void *thread_manage_apps(void *data)
1452 {
1453 int i, ret, pollfd, err = -1;
1454 ssize_t size_ret;
1455 uint32_t revents, nb_fd;
1456 struct lttng_poll_event events;
1457
1458 DBG("[thread] Manage application started");
1459
1460 rcu_register_thread();
1461 rcu_thread_online();
1462
1463 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1464
1465 if (testpoint(sessiond_thread_manage_apps)) {
1466 goto error_testpoint;
1467 }
1468
1469 health_code_update();
1470
1471 ret = sessiond_set_thread_pollset(&events, 2);
1472 if (ret < 0) {
1473 goto error_poll_create;
1474 }
1475
1476 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1477 if (ret < 0) {
1478 goto error;
1479 }
1480
1481 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1482 goto error;
1483 }
1484
1485 health_code_update();
1486
1487 while (1) {
1488 DBG("Apps thread polling");
1489
1490 /* Inifinite blocking call, waiting for transmission */
1491 restart:
1492 health_poll_entry();
1493 ret = lttng_poll_wait(&events, -1);
1494 DBG("Apps thread return from poll on %d fds",
1495 LTTNG_POLL_GETNB(&events));
1496 health_poll_exit();
1497 if (ret < 0) {
1498 /*
1499 * Restart interrupted system call.
1500 */
1501 if (errno == EINTR) {
1502 goto restart;
1503 }
1504 goto error;
1505 }
1506
1507 nb_fd = ret;
1508
1509 for (i = 0; i < nb_fd; i++) {
1510 /* Fetch once the poll data */
1511 revents = LTTNG_POLL_GETEV(&events, i);
1512 pollfd = LTTNG_POLL_GETFD(&events, i);
1513
1514 health_code_update();
1515
1516 if (!revents) {
1517 /* No activity for this FD (poll implementation). */
1518 continue;
1519 }
1520
1521 /* Thread quit pipe has been closed. Killing thread. */
1522 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1523 if (ret) {
1524 err = 0;
1525 goto exit;
1526 }
1527
1528 /* Inspect the apps cmd pipe */
1529 if (pollfd == apps_cmd_pipe[0]) {
1530 if (revents & LPOLLIN) {
1531 int sock;
1532
1533 /* Empty pipe */
1534 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1535 if (size_ret < sizeof(sock)) {
1536 PERROR("read apps cmd pipe");
1537 goto error;
1538 }
1539
1540 health_code_update();
1541
1542 /*
1543 * Since this is a command socket (write then read),
1544 * we only monitor the error events of the socket.
1545 */
1546 ret = lttng_poll_add(&events, sock,
1547 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1548 if (ret < 0) {
1549 goto error;
1550 }
1551
1552 DBG("Apps with sock %d added to poll set", sock);
1553 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1554 ERR("Apps command pipe error");
1555 goto error;
1556 } else {
1557 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1558 goto error;
1559 }
1560 } else {
1561 /*
1562 * At this point, we know that a registered application made
1563 * the event at poll_wait.
1564 */
1565 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1566 /* Removing from the poll set */
1567 ret = lttng_poll_del(&events, pollfd);
1568 if (ret < 0) {
1569 goto error;
1570 }
1571
1572 /* Socket closed on remote end. */
1573 ust_app_unregister(pollfd);
1574 } else {
1575 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1576 goto error;
1577 }
1578 }
1579
1580 health_code_update();
1581 }
1582 }
1583
1584 exit:
1585 error:
1586 lttng_poll_clean(&events);
1587 error_poll_create:
1588 error_testpoint:
1589 utils_close_pipe(apps_cmd_pipe);
1590 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1591
1592 /*
1593 * We don't clean the UST app hash table here since already registered
1594 * applications can still be controlled so let them be until the session
1595 * daemon dies or the applications stop.
1596 */
1597
1598 if (err) {
1599 health_error();
1600 ERR("Health error occurred in %s", __func__);
1601 }
1602 health_unregister(health_sessiond);
1603 DBG("Application communication apps thread cleanup complete");
1604 rcu_thread_offline();
1605 rcu_unregister_thread();
1606 return NULL;
1607 }
1608
1609 /*
1610 * Send a socket to a thread This is called from the dispatch UST registration
1611 * thread once all sockets are set for the application.
1612 *
1613 * The sock value can be invalid, we don't really care, the thread will handle
1614 * it and make the necessary cleanup if so.
1615 *
1616 * On success, return 0 else a negative value being the errno message of the
1617 * write().
1618 */
1619 static int send_socket_to_thread(int fd, int sock)
1620 {
1621 ssize_t ret;
1622
1623 /*
1624 * It's possible that the FD is set as invalid with -1 concurrently just
1625 * before calling this function being a shutdown state of the thread.
1626 */
1627 if (fd < 0) {
1628 ret = -EBADF;
1629 goto error;
1630 }
1631
1632 ret = lttng_write(fd, &sock, sizeof(sock));
1633 if (ret < sizeof(sock)) {
1634 PERROR("write apps pipe %d", fd);
1635 if (ret < 0) {
1636 ret = -errno;
1637 }
1638 goto error;
1639 }
1640
1641 /* All good. Don't send back the write positive ret value. */
1642 ret = 0;
1643 error:
1644 return (int) ret;
1645 }
1646
1647 /*
1648 * Sanitize the wait queue of the dispatch registration thread meaning removing
1649 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1650 * notify socket is never received.
1651 */
1652 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1653 {
1654 int ret, nb_fd = 0, i;
1655 unsigned int fd_added = 0;
1656 struct lttng_poll_event events;
1657 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1658
1659 assert(wait_queue);
1660
1661 lttng_poll_init(&events);
1662
1663 /* Just skip everything for an empty queue. */
1664 if (!wait_queue->count) {
1665 goto end;
1666 }
1667
1668 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1669 if (ret < 0) {
1670 goto error_create;
1671 }
1672
1673 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1674 &wait_queue->head, head) {
1675 assert(wait_node->app);
1676 ret = lttng_poll_add(&events, wait_node->app->sock,
1677 LPOLLHUP | LPOLLERR);
1678 if (ret < 0) {
1679 goto error;
1680 }
1681
1682 fd_added = 1;
1683 }
1684
1685 if (!fd_added) {
1686 goto end;
1687 }
1688
1689 /*
1690 * Poll but don't block so we can quickly identify the faulty events and
1691 * clean them afterwards from the wait queue.
1692 */
1693 ret = lttng_poll_wait(&events, 0);
1694 if (ret < 0) {
1695 goto error;
1696 }
1697 nb_fd = ret;
1698
1699 for (i = 0; i < nb_fd; i++) {
1700 /* Get faulty FD. */
1701 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1702 int pollfd = LTTNG_POLL_GETFD(&events, i);
1703
1704 if (!revents) {
1705 /* No activity for this FD (poll implementation). */
1706 continue;
1707 }
1708
1709 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1710 &wait_queue->head, head) {
1711 if (pollfd == wait_node->app->sock &&
1712 (revents & (LPOLLHUP | LPOLLERR))) {
1713 cds_list_del(&wait_node->head);
1714 wait_queue->count--;
1715 ust_app_destroy(wait_node->app);
1716 free(wait_node);
1717 /*
1718 * Silence warning of use-after-free in
1719 * cds_list_for_each_entry_safe which uses
1720 * __typeof__(*wait_node).
1721 */
1722 wait_node = NULL;
1723 break;
1724 } else {
1725 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1726 goto error;
1727 }
1728 }
1729 }
1730
1731 if (nb_fd > 0) {
1732 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1733 }
1734
1735 end:
1736 lttng_poll_clean(&events);
1737 return;
1738
1739 error:
1740 lttng_poll_clean(&events);
1741 error_create:
1742 ERR("Unable to sanitize wait queue");
1743 return;
1744 }
1745
1746 /*
1747 * Dispatch request from the registration threads to the application
1748 * communication thread.
1749 */
1750 static void *thread_dispatch_ust_registration(void *data)
1751 {
1752 int ret, err = -1;
1753 struct cds_wfcq_node *node;
1754 struct ust_command *ust_cmd = NULL;
1755 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1756 struct ust_reg_wait_queue wait_queue = {
1757 .count = 0,
1758 };
1759
1760 rcu_register_thread();
1761
1762 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1763
1764 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1765 goto error_testpoint;
1766 }
1767
1768 health_code_update();
1769
1770 CDS_INIT_LIST_HEAD(&wait_queue.head);
1771
1772 DBG("[thread] Dispatch UST command started");
1773
1774 for (;;) {
1775 health_code_update();
1776
1777 /* Atomically prepare the queue futex */
1778 futex_nto1_prepare(&ust_cmd_queue.futex);
1779
1780 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1781 break;
1782 }
1783
1784 do {
1785 struct ust_app *app = NULL;
1786 ust_cmd = NULL;
1787
1788 /*
1789 * Make sure we don't have node(s) that have hung up before receiving
1790 * the notify socket. This is to clean the list in order to avoid
1791 * memory leaks from notify socket that are never seen.
1792 */
1793 sanitize_wait_queue(&wait_queue);
1794
1795 health_code_update();
1796 /* Dequeue command for registration */
1797 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1798 if (node == NULL) {
1799 DBG("Woken up but nothing in the UST command queue");
1800 /* Continue thread execution */
1801 break;
1802 }
1803
1804 ust_cmd = caa_container_of(node, struct ust_command, node);
1805
1806 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1807 " gid:%d sock:%d name:%s (version %d.%d)",
1808 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1809 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1810 ust_cmd->sock, ust_cmd->reg_msg.name,
1811 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1812
1813 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1814 wait_node = zmalloc(sizeof(*wait_node));
1815 if (!wait_node) {
1816 PERROR("zmalloc wait_node dispatch");
1817 ret = close(ust_cmd->sock);
1818 if (ret < 0) {
1819 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1820 }
1821 lttng_fd_put(LTTNG_FD_APPS, 1);
1822 free(ust_cmd);
1823 goto error;
1824 }
1825 CDS_INIT_LIST_HEAD(&wait_node->head);
1826
1827 /* Create application object if socket is CMD. */
1828 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1829 ust_cmd->sock);
1830 if (!wait_node->app) {
1831 ret = close(ust_cmd->sock);
1832 if (ret < 0) {
1833 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1834 }
1835 lttng_fd_put(LTTNG_FD_APPS, 1);
1836 free(wait_node);
1837 free(ust_cmd);
1838 continue;
1839 }
1840 /*
1841 * Add application to the wait queue so we can set the notify
1842 * socket before putting this object in the global ht.
1843 */
1844 cds_list_add(&wait_node->head, &wait_queue.head);
1845 wait_queue.count++;
1846
1847 free(ust_cmd);
1848 /*
1849 * We have to continue here since we don't have the notify
1850 * socket and the application MUST be added to the hash table
1851 * only at that moment.
1852 */
1853 continue;
1854 } else {
1855 /*
1856 * Look for the application in the local wait queue and set the
1857 * notify socket if found.
1858 */
1859 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1860 &wait_queue.head, head) {
1861 health_code_update();
1862 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1863 wait_node->app->notify_sock = ust_cmd->sock;
1864 cds_list_del(&wait_node->head);
1865 wait_queue.count--;
1866 app = wait_node->app;
1867 free(wait_node);
1868 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1869 break;
1870 }
1871 }
1872
1873 /*
1874 * With no application at this stage the received socket is
1875 * basically useless so close it before we free the cmd data
1876 * structure for good.
1877 */
1878 if (!app) {
1879 ret = close(ust_cmd->sock);
1880 if (ret < 0) {
1881 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1882 }
1883 lttng_fd_put(LTTNG_FD_APPS, 1);
1884 }
1885 free(ust_cmd);
1886 }
1887
1888 if (app) {
1889 /*
1890 * @session_lock_list
1891 *
1892 * Lock the global session list so from the register up to the
1893 * registration done message, no thread can see the application
1894 * and change its state.
1895 */
1896 session_lock_list();
1897 rcu_read_lock();
1898
1899 /*
1900 * Add application to the global hash table. This needs to be
1901 * done before the update to the UST registry can locate the
1902 * application.
1903 */
1904 ust_app_add(app);
1905
1906 /* Set app version. This call will print an error if needed. */
1907 (void) ust_app_version(app);
1908
1909 /* Send notify socket through the notify pipe. */
1910 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1911 app->notify_sock);
1912 if (ret < 0) {
1913 rcu_read_unlock();
1914 session_unlock_list();
1915 /*
1916 * No notify thread, stop the UST tracing. However, this is
1917 * not an internal error of the this thread thus setting
1918 * the health error code to a normal exit.
1919 */
1920 err = 0;
1921 goto error;
1922 }
1923
1924 /*
1925 * Update newly registered application with the tracing
1926 * registry info already enabled information.
1927 */
1928 update_ust_app(app->sock);
1929
1930 /*
1931 * Don't care about return value. Let the manage apps threads
1932 * handle app unregistration upon socket close.
1933 */
1934 (void) ust_app_register_done(app);
1935
1936 /*
1937 * Even if the application socket has been closed, send the app
1938 * to the thread and unregistration will take place at that
1939 * place.
1940 */
1941 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1942 if (ret < 0) {
1943 rcu_read_unlock();
1944 session_unlock_list();
1945 /*
1946 * No apps. thread, stop the UST tracing. However, this is
1947 * not an internal error of the this thread thus setting
1948 * the health error code to a normal exit.
1949 */
1950 err = 0;
1951 goto error;
1952 }
1953
1954 rcu_read_unlock();
1955 session_unlock_list();
1956 }
1957 } while (node != NULL);
1958
1959 health_poll_entry();
1960 /* Futex wait on queue. Blocking call on futex() */
1961 futex_nto1_wait(&ust_cmd_queue.futex);
1962 health_poll_exit();
1963 }
1964 /* Normal exit, no error */
1965 err = 0;
1966
1967 error:
1968 /* Clean up wait queue. */
1969 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1970 &wait_queue.head, head) {
1971 cds_list_del(&wait_node->head);
1972 wait_queue.count--;
1973 free(wait_node);
1974 }
1975
1976 /* Empty command queue. */
1977 for (;;) {
1978 /* Dequeue command for registration */
1979 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1980 if (node == NULL) {
1981 break;
1982 }
1983 ust_cmd = caa_container_of(node, struct ust_command, node);
1984 ret = close(ust_cmd->sock);
1985 if (ret < 0) {
1986 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
1987 }
1988 lttng_fd_put(LTTNG_FD_APPS, 1);
1989 free(ust_cmd);
1990 }
1991
1992 error_testpoint:
1993 DBG("Dispatch thread dying");
1994 if (err) {
1995 health_error();
1996 ERR("Health error occurred in %s", __func__);
1997 }
1998 health_unregister(health_sessiond);
1999 rcu_unregister_thread();
2000 return NULL;
2001 }
2002
2003 /*
2004 * This thread manage application registration.
2005 */
2006 static void *thread_registration_apps(void *data)
2007 {
2008 int sock = -1, i, ret, pollfd, err = -1;
2009 uint32_t revents, nb_fd;
2010 struct lttng_poll_event events;
2011 /*
2012 * Get allocated in this thread, enqueued to a global queue, dequeued and
2013 * freed in the manage apps thread.
2014 */
2015 struct ust_command *ust_cmd = NULL;
2016
2017 DBG("[thread] Manage application registration started");
2018
2019 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2020
2021 if (testpoint(sessiond_thread_registration_apps)) {
2022 goto error_testpoint;
2023 }
2024
2025 ret = lttcomm_listen_unix_sock(apps_sock);
2026 if (ret < 0) {
2027 goto error_listen;
2028 }
2029
2030 /*
2031 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2032 * more will be added to this poll set.
2033 */
2034 ret = sessiond_set_thread_pollset(&events, 2);
2035 if (ret < 0) {
2036 goto error_create_poll;
2037 }
2038
2039 /* Add the application registration socket */
2040 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2041 if (ret < 0) {
2042 goto error_poll_add;
2043 }
2044
2045 /* Notify all applications to register */
2046 ret = notify_ust_apps(1);
2047 if (ret < 0) {
2048 ERR("Failed to notify applications or create the wait shared memory.\n"
2049 "Execution continues but there might be problem for already\n"
2050 "running applications that wishes to register.");
2051 }
2052
2053 while (1) {
2054 DBG("Accepting application registration");
2055
2056 /* Inifinite blocking call, waiting for transmission */
2057 restart:
2058 health_poll_entry();
2059 ret = lttng_poll_wait(&events, -1);
2060 health_poll_exit();
2061 if (ret < 0) {
2062 /*
2063 * Restart interrupted system call.
2064 */
2065 if (errno == EINTR) {
2066 goto restart;
2067 }
2068 goto error;
2069 }
2070
2071 nb_fd = ret;
2072
2073 for (i = 0; i < nb_fd; i++) {
2074 health_code_update();
2075
2076 /* Fetch once the poll data */
2077 revents = LTTNG_POLL_GETEV(&events, i);
2078 pollfd = LTTNG_POLL_GETFD(&events, i);
2079
2080 if (!revents) {
2081 /* No activity for this FD (poll implementation). */
2082 continue;
2083 }
2084
2085 /* Thread quit pipe has been closed. Killing thread. */
2086 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2087 if (ret) {
2088 err = 0;
2089 goto exit;
2090 }
2091
2092 /* Event on the registration socket */
2093 if (pollfd == apps_sock) {
2094 if (revents & LPOLLIN) {
2095 sock = lttcomm_accept_unix_sock(apps_sock);
2096 if (sock < 0) {
2097 goto error;
2098 }
2099
2100 /*
2101 * Set socket timeout for both receiving and ending.
2102 * app_socket_timeout is in seconds, whereas
2103 * lttcomm_setsockopt_rcv_timeout and
2104 * lttcomm_setsockopt_snd_timeout expect msec as
2105 * parameter.
2106 */
2107 if (config.app_socket_timeout >= 0) {
2108 (void) lttcomm_setsockopt_rcv_timeout(sock,
2109 config.app_socket_timeout * 1000);
2110 (void) lttcomm_setsockopt_snd_timeout(sock,
2111 config.app_socket_timeout * 1000);
2112 }
2113
2114 /*
2115 * Set the CLOEXEC flag. Return code is useless because
2116 * either way, the show must go on.
2117 */
2118 (void) utils_set_fd_cloexec(sock);
2119
2120 /* Create UST registration command for enqueuing */
2121 ust_cmd = zmalloc(sizeof(struct ust_command));
2122 if (ust_cmd == NULL) {
2123 PERROR("ust command zmalloc");
2124 ret = close(sock);
2125 if (ret) {
2126 PERROR("close");
2127 }
2128 goto error;
2129 }
2130
2131 /*
2132 * Using message-based transmissions to ensure we don't
2133 * have to deal with partially received messages.
2134 */
2135 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2136 if (ret < 0) {
2137 ERR("Exhausted file descriptors allowed for applications.");
2138 free(ust_cmd);
2139 ret = close(sock);
2140 if (ret) {
2141 PERROR("close");
2142 }
2143 sock = -1;
2144 continue;
2145 }
2146
2147 health_code_update();
2148 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2149 if (ret < 0) {
2150 free(ust_cmd);
2151 /* Close socket of the application. */
2152 ret = close(sock);
2153 if (ret) {
2154 PERROR("close");
2155 }
2156 lttng_fd_put(LTTNG_FD_APPS, 1);
2157 sock = -1;
2158 continue;
2159 }
2160 health_code_update();
2161
2162 ust_cmd->sock = sock;
2163 sock = -1;
2164
2165 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2166 " gid:%d sock:%d name:%s (version %d.%d)",
2167 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2168 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2169 ust_cmd->sock, ust_cmd->reg_msg.name,
2170 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2171
2172 /*
2173 * Lock free enqueue the registration request. The red pill
2174 * has been taken! This apps will be part of the *system*.
2175 */
2176 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2177
2178 /*
2179 * Wake the registration queue futex. Implicit memory
2180 * barrier with the exchange in cds_wfcq_enqueue.
2181 */
2182 futex_nto1_wake(&ust_cmd_queue.futex);
2183 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2184 ERR("Register apps socket poll error");
2185 goto error;
2186 } else {
2187 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2188 goto error;
2189 }
2190 }
2191 }
2192 }
2193
2194 exit:
2195 error:
2196 /* Notify that the registration thread is gone */
2197 notify_ust_apps(0);
2198
2199 if (apps_sock >= 0) {
2200 ret = close(apps_sock);
2201 if (ret) {
2202 PERROR("close");
2203 }
2204 }
2205 if (sock >= 0) {
2206 ret = close(sock);
2207 if (ret) {
2208 PERROR("close");
2209 }
2210 lttng_fd_put(LTTNG_FD_APPS, 1);
2211 }
2212 unlink(config.apps_unix_sock_path.value);
2213
2214 error_poll_add:
2215 lttng_poll_clean(&events);
2216 error_listen:
2217 error_create_poll:
2218 error_testpoint:
2219 DBG("UST Registration thread cleanup complete");
2220 if (err) {
2221 health_error();
2222 ERR("Health error occurred in %s", __func__);
2223 }
2224 health_unregister(health_sessiond);
2225
2226 return NULL;
2227 }
2228
2229 /*
2230 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2231 * exec or it will fails.
2232 */
2233 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2234 {
2235 int ret, clock_ret;
2236 struct timespec timeout;
2237
2238 /*
2239 * Make sure we set the readiness flag to 0 because we are NOT ready.
2240 * This access to consumer_thread_is_ready does not need to be
2241 * protected by consumer_data.cond_mutex (yet) since the consumer
2242 * management thread has not been started at this point.
2243 */
2244 consumer_data->consumer_thread_is_ready = 0;
2245
2246 /* Setup pthread condition */
2247 ret = pthread_condattr_init(&consumer_data->condattr);
2248 if (ret) {
2249 errno = ret;
2250 PERROR("pthread_condattr_init consumer data");
2251 goto error;
2252 }
2253
2254 /*
2255 * Set the monotonic clock in order to make sure we DO NOT jump in time
2256 * between the clock_gettime() call and the timedwait call. See bug #324
2257 * for a more details and how we noticed it.
2258 */
2259 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2260 if (ret) {
2261 errno = ret;
2262 PERROR("pthread_condattr_setclock consumer data");
2263 goto error;
2264 }
2265
2266 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2267 if (ret) {
2268 errno = ret;
2269 PERROR("pthread_cond_init consumer data");
2270 goto error;
2271 }
2272
2273 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2274 thread_manage_consumer, consumer_data);
2275 if (ret) {
2276 errno = ret;
2277 PERROR("pthread_create consumer");
2278 ret = -1;
2279 goto error;
2280 }
2281
2282 /* We are about to wait on a pthread condition */
2283 pthread_mutex_lock(&consumer_data->cond_mutex);
2284
2285 /* Get time for sem_timedwait absolute timeout */
2286 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2287 /*
2288 * Set the timeout for the condition timed wait even if the clock gettime
2289 * call fails since we might loop on that call and we want to avoid to
2290 * increment the timeout too many times.
2291 */
2292 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2293
2294 /*
2295 * The following loop COULD be skipped in some conditions so this is why we
2296 * set ret to 0 in order to make sure at least one round of the loop is
2297 * done.
2298 */
2299 ret = 0;
2300
2301 /*
2302 * Loop until the condition is reached or when a timeout is reached. Note
2303 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2304 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2305 * possible. This loop does not take any chances and works with both of
2306 * them.
2307 */
2308 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2309 if (clock_ret < 0) {
2310 PERROR("clock_gettime spawn consumer");
2311 /* Infinite wait for the consumerd thread to be ready */
2312 ret = pthread_cond_wait(&consumer_data->cond,
2313 &consumer_data->cond_mutex);
2314 } else {
2315 ret = pthread_cond_timedwait(&consumer_data->cond,
2316 &consumer_data->cond_mutex, &timeout);
2317 }
2318 }
2319
2320 /* Release the pthread condition */
2321 pthread_mutex_unlock(&consumer_data->cond_mutex);
2322
2323 if (ret != 0) {
2324 errno = ret;
2325 if (ret == ETIMEDOUT) {
2326 int pth_ret;
2327
2328 /*
2329 * Call has timed out so we kill the kconsumerd_thread and return
2330 * an error.
2331 */
2332 ERR("Condition timed out. The consumer thread was never ready."
2333 " Killing it");
2334 pth_ret = pthread_cancel(consumer_data->thread);
2335 if (pth_ret < 0) {
2336 PERROR("pthread_cancel consumer thread");
2337 }
2338 } else {
2339 PERROR("pthread_cond_wait failed consumer thread");
2340 }
2341 /* Caller is expecting a negative value on failure. */
2342 ret = -1;
2343 goto error;
2344 }
2345
2346 pthread_mutex_lock(&consumer_data->pid_mutex);
2347 if (consumer_data->pid == 0) {
2348 ERR("Consumerd did not start");
2349 pthread_mutex_unlock(&consumer_data->pid_mutex);
2350 goto error;
2351 }
2352 pthread_mutex_unlock(&consumer_data->pid_mutex);
2353
2354 return 0;
2355
2356 error:
2357 return ret;
2358 }
2359
2360 /*
2361 * Join consumer thread
2362 */
2363 static int join_consumer_thread(struct consumer_data *consumer_data)
2364 {
2365 void *status;
2366
2367 /* Consumer pid must be a real one. */
2368 if (consumer_data->pid > 0) {
2369 int ret;
2370 ret = kill(consumer_data->pid, SIGTERM);
2371 if (ret) {
2372 PERROR("Error killing consumer daemon");
2373 return ret;
2374 }
2375 return pthread_join(consumer_data->thread, &status);
2376 } else {
2377 return 0;
2378 }
2379 }
2380
2381 /*
2382 * Fork and exec a consumer daemon (consumerd).
2383 *
2384 * Return pid if successful else -1.
2385 */
2386 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2387 {
2388 int ret;
2389 pid_t pid;
2390 const char *consumer_to_use;
2391 const char *verbosity;
2392 struct stat st;
2393
2394 DBG("Spawning consumerd");
2395
2396 pid = fork();
2397 if (pid == 0) {
2398 /*
2399 * Exec consumerd.
2400 */
2401 if (config.verbose_consumer) {
2402 verbosity = "--verbose";
2403 } else if (lttng_opt_quiet) {
2404 verbosity = "--quiet";
2405 } else {
2406 verbosity = "";
2407 }
2408
2409 switch (consumer_data->type) {
2410 case LTTNG_CONSUMER_KERNEL:
2411 /*
2412 * Find out which consumerd to execute. We will first try the
2413 * 64-bit path, then the sessiond's installation directory, and
2414 * fallback on the 32-bit one,
2415 */
2416 DBG3("Looking for a kernel consumer at these locations:");
2417 DBG3(" 1) %s", config.consumerd64_bin_path.value ? : "NULL");
2418 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
2419 DBG3(" 3) %s", config.consumerd32_bin_path.value ? : "NULL");
2420 if (stat(config.consumerd64_bin_path.value, &st) == 0) {
2421 DBG3("Found location #1");
2422 consumer_to_use = config.consumerd64_bin_path.value;
2423 } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
2424 DBG3("Found location #2");
2425 consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
2426 } else if (stat(config.consumerd32_bin_path.value, &st) == 0) {
2427 DBG3("Found location #3");
2428 consumer_to_use = config.consumerd32_bin_path.value;
2429 } else {
2430 DBG("Could not find any valid consumerd executable");
2431 ret = -EINVAL;
2432 goto error;
2433 }
2434 DBG("Using kernel consumer at: %s", consumer_to_use);
2435 (void) execl(consumer_to_use,
2436 "lttng-consumerd", verbosity, "-k",
2437 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2438 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2439 "--group", config.tracing_group_name.value,
2440 NULL);
2441 break;
2442 case LTTNG_CONSUMER64_UST:
2443 {
2444 char *tmpnew = NULL;
2445
2446 if (config.consumerd64_lib_dir.value) {
2447 char *tmp;
2448 size_t tmplen;
2449
2450 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2451 if (!tmp) {
2452 tmp = "";
2453 }
2454 tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2455 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2456 if (!tmpnew) {
2457 ret = -ENOMEM;
2458 goto error;
2459 }
2460 strcat(tmpnew, config.consumerd64_lib_dir.value);
2461 if (tmp[0] != '\0') {
2462 strcat(tmpnew, ":");
2463 strcat(tmpnew, tmp);
2464 }
2465 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2466 if (ret) {
2467 ret = -errno;
2468 free(tmpnew);
2469 goto error;
2470 }
2471 }
2472 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2473 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2474 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2475 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2476 "--group", config.tracing_group_name.value,
2477 NULL);
2478 free(tmpnew);
2479 break;
2480 }
2481 case LTTNG_CONSUMER32_UST:
2482 {
2483 char *tmpnew = NULL;
2484
2485 if (config.consumerd32_lib_dir.value) {
2486 char *tmp;
2487 size_t tmplen;
2488
2489 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2490 if (!tmp) {
2491 tmp = "";
2492 }
2493 tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2494 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2495 if (!tmpnew) {
2496 ret = -ENOMEM;
2497 goto error;
2498 }
2499 strcat(tmpnew, config.consumerd32_lib_dir.value);
2500 if (tmp[0] != '\0') {
2501 strcat(tmpnew, ":");
2502 strcat(tmpnew, tmp);
2503 }
2504 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2505 if (ret) {
2506 ret = -errno;
2507 free(tmpnew);
2508 goto error;
2509 }
2510 }
2511 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2512 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2513 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2514 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2515 "--group", config.tracing_group_name.value,
2516 NULL);
2517 free(tmpnew);
2518 break;
2519 }
2520 default:
2521 ERR("unknown consumer type");
2522 exit(EXIT_FAILURE);
2523 }
2524 if (errno != 0) {
2525 PERROR("Consumer execl()");
2526 }
2527 /* Reaching this point, we got a failure on our execl(). */
2528 exit(EXIT_FAILURE);
2529 } else if (pid > 0) {
2530 ret = pid;
2531 } else {
2532 PERROR("start consumer fork");
2533 ret = -errno;
2534 }
2535 error:
2536 return ret;
2537 }
2538
2539 /*
2540 * Spawn the consumerd daemon and session daemon thread.
2541 */
2542 static int start_consumerd(struct consumer_data *consumer_data)
2543 {
2544 int ret;
2545
2546 /*
2547 * Set the listen() state on the socket since there is a possible race
2548 * between the exec() of the consumer daemon and this call if place in the
2549 * consumer thread. See bug #366 for more details.
2550 */
2551 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2552 if (ret < 0) {
2553 goto error;
2554 }
2555
2556 pthread_mutex_lock(&consumer_data->pid_mutex);
2557 if (consumer_data->pid != 0) {
2558 pthread_mutex_unlock(&consumer_data->pid_mutex);
2559 goto end;
2560 }
2561
2562 ret = spawn_consumerd(consumer_data);
2563 if (ret < 0) {
2564 ERR("Spawning consumerd failed");
2565 pthread_mutex_unlock(&consumer_data->pid_mutex);
2566 goto error;
2567 }
2568
2569 /* Setting up the consumer_data pid */
2570 consumer_data->pid = ret;
2571 DBG2("Consumer pid %d", consumer_data->pid);
2572 pthread_mutex_unlock(&consumer_data->pid_mutex);
2573
2574 DBG2("Spawning consumer control thread");
2575 ret = spawn_consumer_thread(consumer_data);
2576 if (ret < 0) {
2577 ERR("Fatal error spawning consumer control thread");
2578 goto error;
2579 }
2580
2581 end:
2582 return 0;
2583
2584 error:
2585 /* Cleanup already created sockets on error. */
2586 if (consumer_data->err_sock >= 0) {
2587 int err;
2588
2589 err = close(consumer_data->err_sock);
2590 if (err < 0) {
2591 PERROR("close consumer data error socket");
2592 }
2593 }
2594 return ret;
2595 }
2596
2597 /*
2598 * Setup necessary data for kernel tracer action.
2599 */
2600 static int init_kernel_tracer(void)
2601 {
2602 int ret;
2603
2604 /* Modprobe lttng kernel modules */
2605 ret = modprobe_lttng_control();
2606 if (ret < 0) {
2607 goto error;
2608 }
2609
2610 /* Open debugfs lttng */
2611 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2612 if (kernel_tracer_fd < 0) {
2613 DBG("Failed to open %s", module_proc_lttng);
2614 goto error_open;
2615 }
2616
2617 /* Validate kernel version */
2618 ret = kernel_validate_version(kernel_tracer_fd);
2619 if (ret < 0) {
2620 goto error_version;
2621 }
2622
2623 ret = modprobe_lttng_data();
2624 if (ret < 0) {
2625 goto error_modules;
2626 }
2627
2628 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2629 kernel_tracer_fd);
2630 if (ret < 0) {
2631 goto error_modules;
2632 }
2633
2634 if (ret < 1) {
2635 WARN("Kernel tracer does not support buffer monitoring. "
2636 "The monitoring timer of channels in the kernel domain "
2637 "will be set to 0 (disabled).");
2638 }
2639
2640 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2641 return 0;
2642
2643 error_version:
2644 modprobe_remove_lttng_control();
2645 ret = close(kernel_tracer_fd);
2646 if (ret) {
2647 PERROR("close");
2648 }
2649 kernel_tracer_fd = -1;
2650 return LTTNG_ERR_KERN_VERSION;
2651
2652 error_modules:
2653 ret = close(kernel_tracer_fd);
2654 if (ret) {
2655 PERROR("close");
2656 }
2657
2658 error_open:
2659 modprobe_remove_lttng_control();
2660
2661 error:
2662 WARN("No kernel tracer available");
2663 kernel_tracer_fd = -1;
2664 if (!is_root) {
2665 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2666 } else {
2667 return LTTNG_ERR_KERN_NA;
2668 }
2669 }
2670
2671
2672 /*
2673 * Copy consumer output from the tracing session to the domain session. The
2674 * function also applies the right modification on a per domain basis for the
2675 * trace files destination directory.
2676 *
2677 * Should *NOT* be called with RCU read-side lock held.
2678 */
2679 static int copy_session_consumer(int domain, struct ltt_session *session)
2680 {
2681 int ret;
2682 const char *dir_name;
2683 struct consumer_output *consumer;
2684
2685 assert(session);
2686 assert(session->consumer);
2687
2688 switch (domain) {
2689 case LTTNG_DOMAIN_KERNEL:
2690 DBG3("Copying tracing session consumer output in kernel session");
2691 /*
2692 * XXX: We should audit the session creation and what this function
2693 * does "extra" in order to avoid a destroy since this function is used
2694 * in the domain session creation (kernel and ust) only. Same for UST
2695 * domain.
2696 */
2697 if (session->kernel_session->consumer) {
2698 consumer_output_put(session->kernel_session->consumer);
2699 }
2700 session->kernel_session->consumer =
2701 consumer_copy_output(session->consumer);
2702 /* Ease our life a bit for the next part */
2703 consumer = session->kernel_session->consumer;
2704 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2705 break;
2706 case LTTNG_DOMAIN_JUL:
2707 case LTTNG_DOMAIN_LOG4J:
2708 case LTTNG_DOMAIN_PYTHON:
2709 case LTTNG_DOMAIN_UST:
2710 DBG3("Copying tracing session consumer output in UST session");
2711 if (session->ust_session->consumer) {
2712 consumer_output_put(session->ust_session->consumer);
2713 }
2714 session->ust_session->consumer =
2715 consumer_copy_output(session->consumer);
2716 /* Ease our life a bit for the next part */
2717 consumer = session->ust_session->consumer;
2718 dir_name = DEFAULT_UST_TRACE_DIR;
2719 break;
2720 default:
2721 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2722 goto error;
2723 }
2724
2725 /* Append correct directory to subdir */
2726 strncat(consumer->subdir, dir_name,
2727 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2728 DBG3("Copy session consumer subdir %s", consumer->subdir);
2729
2730 ret = LTTNG_OK;
2731
2732 error:
2733 return ret;
2734 }
2735
2736 /*
2737 * Create an UST session and add it to the session ust list.
2738 *
2739 * Should *NOT* be called with RCU read-side lock held.
2740 */
2741 static int create_ust_session(struct ltt_session *session,
2742 struct lttng_domain *domain)
2743 {
2744 int ret;
2745 struct ltt_ust_session *lus = NULL;
2746
2747 assert(session);
2748 assert(domain);
2749 assert(session->consumer);
2750
2751 switch (domain->type) {
2752 case LTTNG_DOMAIN_JUL:
2753 case LTTNG_DOMAIN_LOG4J:
2754 case LTTNG_DOMAIN_PYTHON:
2755 case LTTNG_DOMAIN_UST:
2756 break;
2757 default:
2758 ERR("Unknown UST domain on create session %d", domain->type);
2759 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2760 goto error;
2761 }
2762
2763 DBG("Creating UST session");
2764
2765 lus = trace_ust_create_session(session->id);
2766 if (lus == NULL) {
2767 ret = LTTNG_ERR_UST_SESS_FAIL;
2768 goto error;
2769 }
2770
2771 lus->uid = session->uid;
2772 lus->gid = session->gid;
2773 lus->output_traces = session->output_traces;
2774 lus->snapshot_mode = session->snapshot_mode;
2775 lus->live_timer_interval = session->live_timer;
2776 session->ust_session = lus;
2777 if (session->shm_path[0]) {
2778 strncpy(lus->root_shm_path, session->shm_path,
2779 sizeof(lus->root_shm_path));
2780 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2781 strncpy(lus->shm_path, session->shm_path,
2782 sizeof(lus->shm_path));
2783 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2784 strncat(lus->shm_path, "/ust",
2785 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2786 }
2787 /* Copy session output to the newly created UST session */
2788 ret = copy_session_consumer(domain->type, session);
2789 if (ret != LTTNG_OK) {
2790 goto error;
2791 }
2792
2793 return LTTNG_OK;
2794
2795 error:
2796 free(lus);
2797 session->ust_session = NULL;
2798 return ret;
2799 }
2800
2801 /*
2802 * Create a kernel tracer session then create the default channel.
2803 */
2804 static int create_kernel_session(struct ltt_session *session)
2805 {
2806 int ret;
2807
2808 DBG("Creating kernel session");
2809
2810 ret = kernel_create_session(session, kernel_tracer_fd);
2811 if (ret < 0) {
2812 ret = LTTNG_ERR_KERN_SESS_FAIL;
2813 goto error;
2814 }
2815
2816 /* Code flow safety */
2817 assert(session->kernel_session);
2818
2819 /* Copy session output to the newly created Kernel session */
2820 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2821 if (ret != LTTNG_OK) {
2822 goto error;
2823 }
2824
2825 /* Create directory(ies) on local filesystem. */
2826 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2827 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2828 ret = run_as_mkdir_recursive(
2829 session->kernel_session->consumer->dst.trace_path,
2830 S_IRWXU | S_IRWXG, session->uid, session->gid);
2831 if (ret < 0) {
2832 if (errno != EEXIST) {
2833 ERR("Trace directory creation error");
2834 goto error;
2835 }
2836 }
2837 }
2838
2839 session->kernel_session->uid = session->uid;
2840 session->kernel_session->gid = session->gid;
2841 session->kernel_session->output_traces = session->output_traces;
2842 session->kernel_session->snapshot_mode = session->snapshot_mode;
2843
2844 return LTTNG_OK;
2845
2846 error:
2847 trace_kernel_destroy_session(session->kernel_session);
2848 session->kernel_session = NULL;
2849 return ret;
2850 }
2851
2852 /*
2853 * Count number of session permitted by uid/gid.
2854 */
2855 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2856 {
2857 unsigned int i = 0;
2858 struct ltt_session *session;
2859
2860 DBG("Counting number of available session for UID %d GID %d",
2861 uid, gid);
2862 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2863 /*
2864 * Only list the sessions the user can control.
2865 */
2866 if (!session_access_ok(session, uid, gid)) {
2867 continue;
2868 }
2869 i++;
2870 }
2871 return i;
2872 }
2873
2874 /*
2875 * Process the command requested by the lttng client within the command
2876 * context structure. This function make sure that the return structure (llm)
2877 * is set and ready for transmission before returning.
2878 *
2879 * Return any error encountered or 0 for success.
2880 *
2881 * "sock" is only used for special-case var. len data.
2882 *
2883 * Should *NOT* be called with RCU read-side lock held.
2884 */
2885 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2886 int *sock_error)
2887 {
2888 int ret = LTTNG_OK;
2889 int need_tracing_session = 1;
2890 int need_domain;
2891
2892 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2893
2894 assert(!rcu_read_ongoing());
2895
2896 *sock_error = 0;
2897
2898 switch (cmd_ctx->lsm->cmd_type) {
2899 case LTTNG_CREATE_SESSION:
2900 case LTTNG_CREATE_SESSION_SNAPSHOT:
2901 case LTTNG_CREATE_SESSION_LIVE:
2902 case LTTNG_DESTROY_SESSION:
2903 case LTTNG_LIST_SESSIONS:
2904 case LTTNG_LIST_DOMAINS:
2905 case LTTNG_START_TRACE:
2906 case LTTNG_STOP_TRACE:
2907 case LTTNG_DATA_PENDING:
2908 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2909 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2910 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2911 case LTTNG_SNAPSHOT_RECORD:
2912 case LTTNG_SAVE_SESSION:
2913 case LTTNG_SET_SESSION_SHM_PATH:
2914 case LTTNG_REGENERATE_METADATA:
2915 case LTTNG_REGENERATE_STATEDUMP:
2916 case LTTNG_REGISTER_TRIGGER:
2917 case LTTNG_UNREGISTER_TRIGGER:
2918 need_domain = 0;
2919 break;
2920 default:
2921 need_domain = 1;
2922 }
2923
2924 if (config.no_kernel && need_domain
2925 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2926 if (!is_root) {
2927 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2928 } else {
2929 ret = LTTNG_ERR_KERN_NA;
2930 }
2931 goto error;
2932 }
2933
2934 /* Deny register consumer if we already have a spawned consumer. */
2935 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2936 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2937 if (kconsumer_data.pid > 0) {
2938 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2939 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2940 goto error;
2941 }
2942 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2943 }
2944
2945 /*
2946 * Check for command that don't needs to allocate a returned payload. We do
2947 * this here so we don't have to make the call for no payload at each
2948 * command.
2949 */
2950 switch(cmd_ctx->lsm->cmd_type) {
2951 case LTTNG_LIST_SESSIONS:
2952 case LTTNG_LIST_TRACEPOINTS:
2953 case LTTNG_LIST_TRACEPOINT_FIELDS:
2954 case LTTNG_LIST_DOMAINS:
2955 case LTTNG_LIST_CHANNELS:
2956 case LTTNG_LIST_EVENTS:
2957 case LTTNG_LIST_SYSCALLS:
2958 case LTTNG_LIST_TRACKER_PIDS:
2959 case LTTNG_DATA_PENDING:
2960 break;
2961 default:
2962 /* Setup lttng message with no payload */
2963 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
2964 if (ret < 0) {
2965 /* This label does not try to unlock the session */
2966 goto init_setup_error;
2967 }
2968 }
2969
2970 /* Commands that DO NOT need a session. */
2971 switch (cmd_ctx->lsm->cmd_type) {
2972 case LTTNG_CREATE_SESSION:
2973 case LTTNG_CREATE_SESSION_SNAPSHOT:
2974 case LTTNG_CREATE_SESSION_LIVE:
2975 case LTTNG_LIST_SESSIONS:
2976 case LTTNG_LIST_TRACEPOINTS:
2977 case LTTNG_LIST_SYSCALLS:
2978 case LTTNG_LIST_TRACEPOINT_FIELDS:
2979 case LTTNG_SAVE_SESSION:
2980 case LTTNG_REGISTER_TRIGGER:
2981 case LTTNG_UNREGISTER_TRIGGER:
2982 need_tracing_session = 0;
2983 break;
2984 default:
2985 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2986 /*
2987 * We keep the session list lock across _all_ commands
2988 * for now, because the per-session lock does not
2989 * handle teardown properly.
2990 */
2991 session_lock_list();
2992 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2993 if (cmd_ctx->session == NULL) {
2994 ret = LTTNG_ERR_SESS_NOT_FOUND;
2995 goto error;
2996 } else {
2997 /* Acquire lock for the session */
2998 session_lock(cmd_ctx->session);
2999 }
3000 break;
3001 }
3002
3003 /*
3004 * Commands that need a valid session but should NOT create one if none
3005 * exists. Instead of creating one and destroying it when the command is
3006 * handled, process that right before so we save some round trip in useless
3007 * code path.
3008 */
3009 switch (cmd_ctx->lsm->cmd_type) {
3010 case LTTNG_DISABLE_CHANNEL:
3011 case LTTNG_DISABLE_EVENT:
3012 switch (cmd_ctx->lsm->domain.type) {
3013 case LTTNG_DOMAIN_KERNEL:
3014 if (!cmd_ctx->session->kernel_session) {
3015 ret = LTTNG_ERR_NO_CHANNEL;
3016 goto error;
3017 }
3018 break;
3019 case LTTNG_DOMAIN_JUL:
3020 case LTTNG_DOMAIN_LOG4J:
3021 case LTTNG_DOMAIN_PYTHON:
3022 case LTTNG_DOMAIN_UST:
3023 if (!cmd_ctx->session->ust_session) {
3024 ret = LTTNG_ERR_NO_CHANNEL;
3025 goto error;
3026 }
3027 break;
3028 default:
3029 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3030 goto error;
3031 }
3032 default:
3033 break;
3034 }
3035
3036 if (!need_domain) {
3037 goto skip_domain;
3038 }
3039
3040 /*
3041 * Check domain type for specific "pre-action".
3042 */
3043 switch (cmd_ctx->lsm->domain.type) {
3044 case LTTNG_DOMAIN_KERNEL:
3045 if (!is_root) {
3046 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3047 goto error;
3048 }
3049
3050 /* Kernel tracer check */
3051 if (kernel_tracer_fd == -1) {
3052 /* Basically, load kernel tracer modules */
3053 ret = init_kernel_tracer();
3054 if (ret != 0) {
3055 goto error;
3056 }
3057 }
3058
3059 /* Consumer is in an ERROR state. Report back to client */
3060 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3061 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3062 goto error;
3063 }
3064
3065 /* Need a session for kernel command */
3066 if (need_tracing_session) {
3067 if (cmd_ctx->session->kernel_session == NULL) {
3068 ret = create_kernel_session(cmd_ctx->session);
3069 if (ret < 0) {
3070 ret = LTTNG_ERR_KERN_SESS_FAIL;
3071 goto error;
3072 }
3073 }
3074
3075 /* Start the kernel consumer daemon */
3076 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3077 if (kconsumer_data.pid == 0 &&
3078 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3079 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3080 ret = start_consumerd(&kconsumer_data);
3081 if (ret < 0) {
3082 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3083 goto error;
3084 }
3085 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3086 } else {
3087 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3088 }
3089
3090 /*
3091 * The consumer was just spawned so we need to add the socket to
3092 * the consumer output of the session if exist.
3093 */
3094 ret = consumer_create_socket(&kconsumer_data,
3095 cmd_ctx->session->kernel_session->consumer);
3096 if (ret < 0) {
3097 goto error;
3098 }
3099 }
3100
3101 break;
3102 case LTTNG_DOMAIN_JUL:
3103 case LTTNG_DOMAIN_LOG4J:
3104 case LTTNG_DOMAIN_PYTHON:
3105 case LTTNG_DOMAIN_UST:
3106 {
3107 if (!ust_app_supported()) {
3108 ret = LTTNG_ERR_NO_UST;
3109 goto error;
3110 }
3111 /* Consumer is in an ERROR state. Report back to client */
3112 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3113 ret = LTTNG_ERR_NO_USTCONSUMERD;
3114 goto error;
3115 }
3116
3117 if (need_tracing_session) {
3118 /* Create UST session if none exist. */
3119 if (cmd_ctx->session->ust_session == NULL) {
3120 ret = create_ust_session(cmd_ctx->session,
3121 &cmd_ctx->lsm->domain);
3122 if (ret != LTTNG_OK) {
3123 goto error;
3124 }
3125 }
3126
3127 /* Start the UST consumer daemons */
3128 /* 64-bit */
3129 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3130 if (config.consumerd64_bin_path.value &&
3131 ustconsumer64_data.pid == 0 &&
3132 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3133 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3134 ret = start_consumerd(&ustconsumer64_data);
3135 if (ret < 0) {
3136 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3137 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3138 goto error;
3139 }
3140
3141 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3142 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3143 } else {
3144 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3145 }
3146
3147 /*
3148 * Setup socket for consumer 64 bit. No need for atomic access
3149 * since it was set above and can ONLY be set in this thread.
3150 */
3151 ret = consumer_create_socket(&ustconsumer64_data,
3152 cmd_ctx->session->ust_session->consumer);
3153 if (ret < 0) {
3154 goto error;
3155 }
3156
3157 /* 32-bit */
3158 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3159 if (config.consumerd32_bin_path.value &&
3160 ustconsumer32_data.pid == 0 &&
3161 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3162 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3163 ret = start_consumerd(&ustconsumer32_data);
3164 if (ret < 0) {
3165 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3166 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3167 goto error;
3168 }
3169
3170 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3171 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3172 } else {
3173 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3174 }
3175
3176 /*
3177 * Setup socket for consumer 64 bit. No need for atomic access
3178 * since it was set above and can ONLY be set in this thread.
3179 */
3180 ret = consumer_create_socket(&ustconsumer32_data,
3181 cmd_ctx->session->ust_session->consumer);
3182 if (ret < 0) {
3183 goto error;
3184 }
3185 }
3186 break;
3187 }
3188 default:
3189 break;
3190 }
3191 skip_domain:
3192
3193 /* Validate consumer daemon state when start/stop trace command */
3194 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3195 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3196 switch (cmd_ctx->lsm->domain.type) {
3197 case LTTNG_DOMAIN_NONE:
3198 break;
3199 case LTTNG_DOMAIN_JUL:
3200 case LTTNG_DOMAIN_LOG4J:
3201 case LTTNG_DOMAIN_PYTHON:
3202 case LTTNG_DOMAIN_UST:
3203 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3204 ret = LTTNG_ERR_NO_USTCONSUMERD;
3205 goto error;
3206 }
3207 break;
3208 case LTTNG_DOMAIN_KERNEL:
3209 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3210 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3211 goto error;
3212 }
3213 break;
3214 default:
3215 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3216 goto error;
3217 }
3218 }
3219
3220 /*
3221 * Check that the UID or GID match that of the tracing session.
3222 * The root user can interact with all sessions.
3223 */
3224 if (need_tracing_session) {
3225 if (!session_access_ok(cmd_ctx->session,
3226 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3227 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3228 ret = LTTNG_ERR_EPERM;
3229 goto error;
3230 }
3231 }
3232
3233 /*
3234 * Send relayd information to consumer as soon as we have a domain and a
3235 * session defined.
3236 */
3237 if (cmd_ctx->session && need_domain) {
3238 /*
3239 * Setup relayd if not done yet. If the relayd information was already
3240 * sent to the consumer, this call will gracefully return.
3241 */
3242 ret = cmd_setup_relayd(cmd_ctx->session);
3243 if (ret != LTTNG_OK) {
3244 goto error;
3245 }
3246 }
3247
3248 /* Process by command type */
3249 switch (cmd_ctx->lsm->cmd_type) {
3250 case LTTNG_ADD_CONTEXT:
3251 {
3252 /*
3253 * An LTTNG_ADD_CONTEXT command might have a supplementary
3254 * payload if the context being added is an application context.
3255 */
3256 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3257 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3258 char *provider_name = NULL, *context_name = NULL;
3259 size_t provider_name_len =
3260 cmd_ctx->lsm->u.context.provider_name_len;
3261 size_t context_name_len =
3262 cmd_ctx->lsm->u.context.context_name_len;
3263
3264 if (provider_name_len == 0 || context_name_len == 0) {
3265 /*
3266 * Application provider and context names MUST
3267 * be provided.
3268 */
3269 ret = -LTTNG_ERR_INVALID;
3270 goto error;
3271 }
3272
3273 provider_name = zmalloc(provider_name_len + 1);
3274 if (!provider_name) {
3275 ret = -LTTNG_ERR_NOMEM;
3276 goto error;
3277 }
3278 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3279 provider_name;
3280
3281 context_name = zmalloc(context_name_len + 1);
3282 if (!context_name) {
3283 ret = -LTTNG_ERR_NOMEM;
3284 goto error_add_context;
3285 }
3286 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3287 context_name;
3288
3289 ret = lttcomm_recv_unix_sock(sock, provider_name,
3290 provider_name_len);
3291 if (ret < 0) {
3292 goto error_add_context;
3293 }
3294
3295 ret = lttcomm_recv_unix_sock(sock, context_name,
3296 context_name_len);
3297 if (ret < 0) {
3298 goto error_add_context;
3299 }
3300 }
3301
3302 /*
3303 * cmd_add_context assumes ownership of the provider and context
3304 * names.
3305 */
3306 ret = cmd_add_context(cmd_ctx->session,
3307 cmd_ctx->lsm->domain.type,
3308 cmd_ctx->lsm->u.context.channel_name,
3309 &cmd_ctx->lsm->u.context.ctx,
3310 kernel_poll_pipe[1]);
3311
3312 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3313 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3314 error_add_context:
3315 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3316 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3317 if (ret < 0) {
3318 goto error;
3319 }
3320 break;
3321 }
3322 case LTTNG_DISABLE_CHANNEL:
3323 {
3324 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3325 cmd_ctx->lsm->u.disable.channel_name);
3326 break;
3327 }
3328 case LTTNG_DISABLE_EVENT:
3329 {
3330
3331 /*
3332 * FIXME: handle filter; for now we just receive the filter's
3333 * bytecode along with the filter expression which are sent by
3334 * liblttng-ctl and discard them.
3335 *
3336 * This fixes an issue where the client may block while sending
3337 * the filter payload and encounter an error because the session
3338 * daemon closes the socket without ever handling this data.
3339 */
3340 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3341 cmd_ctx->lsm->u.disable.bytecode_len;
3342
3343 if (count) {
3344 char data[LTTNG_FILTER_MAX_LEN];
3345
3346 DBG("Discarding disable event command payload of size %zu", count);
3347 while (count) {
3348 ret = lttcomm_recv_unix_sock(sock, data,
3349 count > sizeof(data) ? sizeof(data) : count);
3350 if (ret < 0) {
3351 goto error;
3352 }
3353
3354 count -= (size_t) ret;
3355 }
3356 }
3357 /* FIXME: passing packed structure to non-packed pointer */
3358 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3359 cmd_ctx->lsm->u.disable.channel_name,
3360 &cmd_ctx->lsm->u.disable.event);
3361 break;
3362 }
3363 case LTTNG_ENABLE_CHANNEL:
3364 {
3365 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3366 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3367 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3368 &cmd_ctx->lsm->u.channel.chan,
3369 kernel_poll_pipe[1]);
3370 break;
3371 }
3372 case LTTNG_TRACK_PID:
3373 {
3374 ret = cmd_track_pid(cmd_ctx->session,
3375 cmd_ctx->lsm->domain.type,
3376 cmd_ctx->lsm->u.pid_tracker.pid);
3377 break;
3378 }
3379 case LTTNG_UNTRACK_PID:
3380 {
3381 ret = cmd_untrack_pid(cmd_ctx->session,
3382 cmd_ctx->lsm->domain.type,
3383 cmd_ctx->lsm->u.pid_tracker.pid);
3384 break;
3385 }
3386 case LTTNG_ENABLE_EVENT:
3387 {
3388 struct lttng_event_exclusion *exclusion = NULL;
3389 struct lttng_filter_bytecode *bytecode = NULL;
3390 char *filter_expression = NULL;
3391
3392 /* Handle exclusion events and receive it from the client. */
3393 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3394 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3395
3396 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3397 (count * LTTNG_SYMBOL_NAME_LEN));
3398 if (!exclusion) {
3399 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3400 goto error;
3401 }
3402
3403 DBG("Receiving var len exclusion event list from client ...");
3404 exclusion->count = count;
3405 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3406 count * LTTNG_SYMBOL_NAME_LEN);
3407 if (ret <= 0) {
3408 DBG("Nothing recv() from client var len data... continuing");
3409 *sock_error = 1;
3410 free(exclusion);
3411 ret = LTTNG_ERR_EXCLUSION_INVAL;
3412 goto error;
3413 }
3414 }
3415
3416 /* Get filter expression from client. */
3417 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3418 size_t expression_len =
3419 cmd_ctx->lsm->u.enable.expression_len;
3420
3421 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3422 ret = LTTNG_ERR_FILTER_INVAL;
3423 free(exclusion);
3424 goto error;
3425 }
3426
3427 filter_expression = zmalloc(expression_len);
3428 if (!filter_expression) {
3429 free(exclusion);
3430 ret = LTTNG_ERR_FILTER_NOMEM;
3431 goto error;
3432 }
3433
3434 /* Receive var. len. data */
3435 DBG("Receiving var len filter's expression from client ...");
3436 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3437 expression_len);
3438 if (ret <= 0) {
3439 DBG("Nothing recv() from client car len data... continuing");
3440 *sock_error = 1;
3441 free(filter_expression);
3442 free(exclusion);
3443 ret = LTTNG_ERR_FILTER_INVAL;
3444 goto error;
3445 }
3446 }
3447
3448 /* Handle filter and get bytecode from client. */
3449 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3450 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3451
3452 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3453 ret = LTTNG_ERR_FILTER_INVAL;
3454 free(filter_expression);
3455 free(exclusion);
3456 goto error;
3457 }
3458
3459 bytecode = zmalloc(bytecode_len);
3460 if (!bytecode) {
3461 free(filter_expression);
3462 free(exclusion);
3463 ret = LTTNG_ERR_FILTER_NOMEM;
3464 goto error;
3465 }
3466
3467 /* Receive var. len. data */
3468 DBG("Receiving var len filter's bytecode from client ...");
3469 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3470 if (ret <= 0) {
3471 DBG("Nothing recv() from client car len data... continuing");
3472 *sock_error = 1;
3473 free(filter_expression);
3474 free(bytecode);
3475 free(exclusion);
3476 ret = LTTNG_ERR_FILTER_INVAL;
3477 goto error;
3478 }
3479
3480 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3481 free(filter_expression);
3482 free(bytecode);
3483 free(exclusion);
3484 ret = LTTNG_ERR_FILTER_INVAL;
3485 goto error;
3486 }
3487 }
3488
3489 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3490 cmd_ctx->lsm->u.enable.channel_name,
3491 &cmd_ctx->lsm->u.enable.event,
3492 filter_expression, bytecode, exclusion,
3493 kernel_poll_pipe[1]);
3494 break;
3495 }
3496 case LTTNG_LIST_TRACEPOINTS:
3497 {
3498 struct lttng_event *events;
3499 ssize_t nb_events;
3500
3501 session_lock_list();
3502 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3503 session_unlock_list();
3504 if (nb_events < 0) {
3505 /* Return value is a negative lttng_error_code. */
3506 ret = -nb_events;
3507 goto error;
3508 }
3509
3510 /*
3511 * Setup lttng message with payload size set to the event list size in
3512 * bytes and then copy list into the llm payload.
3513 */
3514 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3515 sizeof(struct lttng_event) * nb_events);
3516 free(events);
3517
3518 if (ret < 0) {
3519 goto setup_error;
3520 }
3521
3522 ret = LTTNG_OK;
3523 break;
3524 }
3525 case LTTNG_LIST_TRACEPOINT_FIELDS:
3526 {
3527 struct lttng_event_field *fields;
3528 ssize_t nb_fields;
3529
3530 session_lock_list();
3531 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3532 &fields);
3533 session_unlock_list();
3534 if (nb_fields < 0) {
3535 /* Return value is a negative lttng_error_code. */
3536 ret = -nb_fields;
3537 goto error;
3538 }
3539
3540 /*
3541 * Setup lttng message with payload size set to the event list size in
3542 * bytes and then copy list into the llm payload.
3543 */
3544 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3545 sizeof(struct lttng_event_field) * nb_fields);
3546 free(fields);
3547
3548 if (ret < 0) {
3549 goto setup_error;
3550 }
3551
3552 ret = LTTNG_OK;
3553 break;
3554 }
3555 case LTTNG_LIST_SYSCALLS:
3556 {
3557 struct lttng_event *events;
3558 ssize_t nb_events;
3559
3560 nb_events = cmd_list_syscalls(&events);
3561 if (nb_events < 0) {
3562 /* Return value is a negative lttng_error_code. */
3563 ret = -nb_events;
3564 goto error;
3565 }
3566
3567 /*
3568 * Setup lttng message with payload size set to the event list size in
3569 * bytes and then copy list into the llm payload.
3570 */
3571 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3572 sizeof(struct lttng_event) * nb_events);
3573 free(events);
3574
3575 if (ret < 0) {
3576 goto setup_error;
3577 }
3578
3579 ret = LTTNG_OK;
3580 break;
3581 }
3582 case LTTNG_LIST_TRACKER_PIDS:
3583 {
3584 int32_t *pids = NULL;
3585 ssize_t nr_pids;
3586
3587 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3588 cmd_ctx->lsm->domain.type, &pids);
3589 if (nr_pids < 0) {
3590 /* Return value is a negative lttng_error_code. */
3591 ret = -nr_pids;
3592 goto error;
3593 }
3594
3595 /*
3596 * Setup lttng message with payload size set to the event list size in
3597 * bytes and then copy list into the llm payload.
3598 */
3599 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3600 sizeof(int32_t) * nr_pids);
3601 free(pids);
3602
3603 if (ret < 0) {
3604 goto setup_error;
3605 }
3606
3607 ret = LTTNG_OK;
3608 break;
3609 }
3610 case LTTNG_SET_CONSUMER_URI:
3611 {
3612 size_t nb_uri, len;
3613 struct lttng_uri *uris;
3614
3615 nb_uri = cmd_ctx->lsm->u.uri.size;
3616 len = nb_uri * sizeof(struct lttng_uri);
3617
3618 if (nb_uri == 0) {
3619 ret = LTTNG_ERR_INVALID;
3620 goto error;
3621 }
3622
3623 uris = zmalloc(len);
3624 if (uris == NULL) {
3625 ret = LTTNG_ERR_FATAL;
3626 goto error;
3627 }
3628
3629 /* Receive variable len data */
3630 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3631 ret = lttcomm_recv_unix_sock(sock, uris, len);
3632 if (ret <= 0) {
3633 DBG("No URIs received from client... continuing");
3634 *sock_error = 1;
3635 ret = LTTNG_ERR_SESSION_FAIL;
3636 free(uris);
3637 goto error;
3638 }
3639
3640 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3641 free(uris);
3642 if (ret != LTTNG_OK) {
3643 goto error;
3644 }
3645
3646
3647 break;
3648 }
3649 case LTTNG_START_TRACE:
3650 {
3651 ret = cmd_start_trace(cmd_ctx->session);
3652 break;
3653 }
3654 case LTTNG_STOP_TRACE:
3655 {
3656 ret = cmd_stop_trace(cmd_ctx->session);
3657 break;
3658 }
3659 case LTTNG_CREATE_SESSION:
3660 {
3661 size_t nb_uri, len;
3662 struct lttng_uri *uris = NULL;
3663
3664 nb_uri = cmd_ctx->lsm->u.uri.size;
3665 len = nb_uri * sizeof(struct lttng_uri);
3666
3667 if (nb_uri > 0) {
3668 uris = zmalloc(len);
3669 if (uris == NULL) {
3670 ret = LTTNG_ERR_FATAL;
3671 goto error;
3672 }
3673
3674 /* Receive variable len data */
3675 DBG("Waiting for %zu URIs from client ...", nb_uri);
3676 ret = lttcomm_recv_unix_sock(sock, uris, len);
3677 if (ret <= 0) {
3678 DBG("No URIs received from client... continuing");
3679 *sock_error = 1;
3680 ret = LTTNG_ERR_SESSION_FAIL;
3681 free(uris);
3682 goto error;
3683 }
3684
3685 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3686 DBG("Creating session with ONE network URI is a bad call");
3687 ret = LTTNG_ERR_SESSION_FAIL;
3688 free(uris);
3689 goto error;
3690 }
3691 }
3692
3693 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3694 &cmd_ctx->creds, 0);
3695
3696 free(uris);
3697
3698 break;
3699 }
3700 case LTTNG_DESTROY_SESSION:
3701 {
3702 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3703
3704 /* Set session to NULL so we do not unlock it after free. */
3705 cmd_ctx->session = NULL;
3706 break;
3707 }
3708 case LTTNG_LIST_DOMAINS:
3709 {
3710 ssize_t nb_dom;
3711 struct lttng_domain *domains = NULL;
3712
3713 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3714 if (nb_dom < 0) {
3715 /* Return value is a negative lttng_error_code. */
3716 ret = -nb_dom;
3717 goto error;
3718 }
3719
3720 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3721 nb_dom * sizeof(struct lttng_domain));
3722 free(domains);
3723
3724 if (ret < 0) {
3725 goto setup_error;
3726 }
3727
3728 ret = LTTNG_OK;
3729 break;
3730 }
3731 case LTTNG_LIST_CHANNELS:
3732 {
3733 ssize_t payload_size;
3734 struct lttng_channel *channels = NULL;
3735
3736 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3737 cmd_ctx->session, &channels);
3738 if (payload_size < 0) {
3739 /* Return value is a negative lttng_error_code. */
3740 ret = -payload_size;
3741 goto error;
3742 }
3743
3744 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3745 payload_size);
3746 free(channels);
3747
3748 if (ret < 0) {
3749 goto setup_error;
3750 }
3751
3752 ret = LTTNG_OK;
3753 break;
3754 }
3755 case LTTNG_LIST_EVENTS:
3756 {
3757 ssize_t nb_event;
3758 struct lttng_event *events = NULL;
3759 struct lttcomm_event_command_header cmd_header;
3760 size_t total_size;
3761
3762 memset(&cmd_header, 0, sizeof(cmd_header));
3763 /* Extended infos are included at the end of events */
3764 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3765 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3766 &events, &total_size);
3767
3768 if (nb_event < 0) {
3769 /* Return value is a negative lttng_error_code. */
3770 ret = -nb_event;
3771 goto error;
3772 }
3773
3774 cmd_header.nb_events = nb_event;
3775 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3776 &cmd_header, sizeof(cmd_header));
3777 free(events);
3778
3779 if (ret < 0) {
3780 goto setup_error;
3781 }
3782
3783 ret = LTTNG_OK;
3784 break;
3785 }
3786 case LTTNG_LIST_SESSIONS:
3787 {
3788 unsigned int nr_sessions;
3789 void *sessions_payload;
3790 size_t payload_len;
3791
3792 session_lock_list();
3793 nr_sessions = lttng_sessions_count(
3794 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3795 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3796 payload_len = sizeof(struct lttng_session) * nr_sessions;
3797 sessions_payload = zmalloc(payload_len);
3798
3799 if (!sessions_payload) {
3800 session_unlock_list();
3801 ret = -ENOMEM;
3802 goto setup_error;
3803 }
3804
3805 cmd_list_lttng_sessions(sessions_payload,
3806 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3807 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3808 session_unlock_list();
3809
3810 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3811 payload_len);
3812 free(sessions_payload);
3813
3814 if (ret < 0) {
3815 goto setup_error;
3816 }
3817
3818 ret = LTTNG_OK;
3819 break;
3820 }
3821 case LTTNG_REGISTER_CONSUMER:
3822 {
3823 struct consumer_data *cdata;
3824
3825 switch (cmd_ctx->lsm->domain.type) {
3826 case LTTNG_DOMAIN_KERNEL:
3827 cdata = &kconsumer_data;
3828 break;
3829 default:
3830 ret = LTTNG_ERR_UND;
3831 goto error;
3832 }
3833
3834 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3835 cmd_ctx->lsm->u.reg.path, cdata);
3836 break;
3837 }
3838 case LTTNG_DATA_PENDING:
3839 {
3840 int pending_ret;
3841 uint8_t pending_ret_byte;
3842
3843 pending_ret = cmd_data_pending(cmd_ctx->session);
3844
3845 /*
3846 * FIXME
3847 *
3848 * This function may returns 0 or 1 to indicate whether or not
3849 * there is data pending. In case of error, it should return an
3850 * LTTNG_ERR code. However, some code paths may still return
3851 * a nondescript error code, which we handle by returning an
3852 * "unknown" error.
3853 */
3854 if (pending_ret == 0 || pending_ret == 1) {
3855 /*
3856 * ret will be set to LTTNG_OK at the end of
3857 * this function.
3858 */
3859 } else if (pending_ret < 0) {
3860 ret = LTTNG_ERR_UNK;
3861 goto setup_error;
3862 } else {
3863 ret = pending_ret;
3864 goto setup_error;
3865 }
3866
3867 pending_ret_byte = (uint8_t) pending_ret;
3868
3869 /* 1 byte to return whether or not data is pending */
3870 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3871 &pending_ret_byte, 1);
3872
3873 if (ret < 0) {
3874 goto setup_error;
3875 }
3876
3877 ret = LTTNG_OK;
3878 break;
3879 }
3880 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3881 {
3882 struct lttcomm_lttng_output_id reply;
3883
3884 ret = cmd_snapshot_add_output(cmd_ctx->session,
3885 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3886 if (ret != LTTNG_OK) {
3887 goto error;
3888 }
3889
3890 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
3891 sizeof(reply));
3892 if (ret < 0) {
3893 goto setup_error;
3894 }
3895
3896 /* Copy output list into message payload */
3897 ret = LTTNG_OK;
3898 break;
3899 }
3900 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3901 {
3902 ret = cmd_snapshot_del_output(cmd_ctx->session,
3903 &cmd_ctx->lsm->u.snapshot_output.output);
3904 break;
3905 }
3906 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3907 {
3908 ssize_t nb_output;
3909 struct lttng_snapshot_output *outputs = NULL;
3910
3911 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3912 if (nb_output < 0) {
3913 ret = -nb_output;
3914 goto error;
3915 }
3916
3917 assert((nb_output > 0 && outputs) || nb_output == 0);
3918 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
3919 nb_output * sizeof(struct lttng_snapshot_output));
3920 free(outputs);
3921
3922 if (ret < 0) {
3923 goto setup_error;
3924 }
3925
3926 ret = LTTNG_OK;
3927 break;
3928 }
3929 case LTTNG_SNAPSHOT_RECORD:
3930 {
3931 ret = cmd_snapshot_record(cmd_ctx->session,
3932 &cmd_ctx->lsm->u.snapshot_record.output,
3933 cmd_ctx->lsm->u.snapshot_record.wait);
3934 break;
3935 }
3936 case LTTNG_CREATE_SESSION_SNAPSHOT:
3937 {
3938 size_t nb_uri, len;
3939 struct lttng_uri *uris = NULL;
3940
3941 nb_uri = cmd_ctx->lsm->u.uri.size;
3942 len = nb_uri * sizeof(struct lttng_uri);
3943
3944 if (nb_uri > 0) {
3945 uris = zmalloc(len);
3946 if (uris == NULL) {
3947 ret = LTTNG_ERR_FATAL;
3948 goto error;
3949 }
3950
3951 /* Receive variable len data */
3952 DBG("Waiting for %zu URIs from client ...", nb_uri);
3953 ret = lttcomm_recv_unix_sock(sock, uris, len);
3954 if (ret <= 0) {
3955 DBG("No URIs received from client... continuing");
3956 *sock_error = 1;
3957 ret = LTTNG_ERR_SESSION_FAIL;
3958 free(uris);
3959 goto error;
3960 }
3961
3962 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3963 DBG("Creating session with ONE network URI is a bad call");
3964 ret = LTTNG_ERR_SESSION_FAIL;
3965 free(uris);
3966 goto error;
3967 }
3968 }
3969
3970 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3971 nb_uri, &cmd_ctx->creds);
3972 free(uris);
3973 break;
3974 }
3975 case LTTNG_CREATE_SESSION_LIVE:
3976 {
3977 size_t nb_uri, len;
3978 struct lttng_uri *uris = NULL;
3979
3980 nb_uri = cmd_ctx->lsm->u.uri.size;
3981 len = nb_uri * sizeof(struct lttng_uri);
3982
3983 if (nb_uri > 0) {
3984 uris = zmalloc(len);
3985 if (uris == NULL) {
3986 ret = LTTNG_ERR_FATAL;
3987 goto error;
3988 }
3989
3990 /* Receive variable len data */
3991 DBG("Waiting for %zu URIs from client ...", nb_uri);
3992 ret = lttcomm_recv_unix_sock(sock, uris, len);
3993 if (ret <= 0) {
3994 DBG("No URIs received from client... continuing");
3995 *sock_error = 1;
3996 ret = LTTNG_ERR_SESSION_FAIL;
3997 free(uris);
3998 goto error;
3999 }
4000
4001 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4002 DBG("Creating session with ONE network URI is a bad call");
4003 ret = LTTNG_ERR_SESSION_FAIL;
4004 free(uris);
4005 goto error;
4006 }
4007 }
4008
4009 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4010 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4011 free(uris);
4012 break;
4013 }
4014 case LTTNG_SAVE_SESSION:
4015 {
4016 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4017 &cmd_ctx->creds);
4018 break;
4019 }
4020 case LTTNG_SET_SESSION_SHM_PATH:
4021 {
4022 ret = cmd_set_session_shm_path(cmd_ctx->session,
4023 cmd_ctx->lsm->u.set_shm_path.shm_path);
4024 break;
4025 }
4026 case LTTNG_REGENERATE_METADATA:
4027 {
4028 ret = cmd_regenerate_metadata(cmd_ctx->session);
4029 break;
4030 }
4031 case LTTNG_REGENERATE_STATEDUMP:
4032 {
4033 ret = cmd_regenerate_statedump(cmd_ctx->session);
4034 break;
4035 }
4036 case LTTNG_REGISTER_TRIGGER:
4037 {
4038 ret = cmd_register_trigger(cmd_ctx, sock,
4039 notification_thread_handle);
4040 break;
4041 }
4042 case LTTNG_UNREGISTER_TRIGGER:
4043 {
4044 ret = cmd_unregister_trigger(cmd_ctx, sock,
4045 notification_thread_handle);
4046 break;
4047 }
4048 default:
4049 ret = LTTNG_ERR_UND;
4050 break;
4051 }
4052
4053 error:
4054 if (cmd_ctx->llm == NULL) {
4055 DBG("Missing llm structure. Allocating one.");
4056 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4057 goto setup_error;
4058 }
4059 }
4060 /* Set return code */
4061 cmd_ctx->llm->ret_code = ret;
4062 setup_error:
4063 if (cmd_ctx->session) {
4064 session_unlock(cmd_ctx->session);
4065 }
4066 if (need_tracing_session) {
4067 session_unlock_list();
4068 }
4069 init_setup_error:
4070 assert(!rcu_read_ongoing());
4071 return ret;
4072 }
4073
4074 /*
4075 * Thread managing health check socket.
4076 */
4077 static void *thread_manage_health(void *data)
4078 {
4079 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4080 uint32_t revents, nb_fd;
4081 struct lttng_poll_event events;
4082 struct health_comm_msg msg;
4083 struct health_comm_reply reply;
4084
4085 DBG("[thread] Manage health check started");
4086
4087 rcu_register_thread();
4088
4089 /* We might hit an error path before this is created. */
4090 lttng_poll_init(&events);
4091
4092 /* Create unix socket */
4093 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4094 if (sock < 0) {
4095 ERR("Unable to create health check Unix socket");
4096 goto error;
4097 }
4098
4099 if (is_root) {
4100 /* lttng health client socket path permissions */
4101 ret = chown(config.health_unix_sock_path.value, 0,
4102 utils_get_group_id(config.tracing_group_name.value));
4103 if (ret < 0) {
4104 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4105 PERROR("chown");
4106 goto error;
4107 }
4108
4109 ret = chmod(config.health_unix_sock_path.value,
4110 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4111 if (ret < 0) {
4112 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4113 PERROR("chmod");
4114 goto error;
4115 }
4116 }
4117
4118 /*
4119 * Set the CLOEXEC flag. Return code is useless because either way, the
4120 * show must go on.
4121 */
4122 (void) utils_set_fd_cloexec(sock);
4123
4124 ret = lttcomm_listen_unix_sock(sock);
4125 if (ret < 0) {
4126 goto error;
4127 }
4128
4129 /*
4130 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4131 * more will be added to this poll set.
4132 */
4133 ret = sessiond_set_thread_pollset(&events, 2);
4134 if (ret < 0) {
4135 goto error;
4136 }
4137
4138 /* Add the application registration socket */
4139 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4140 if (ret < 0) {
4141 goto error;
4142 }
4143
4144 sessiond_notify_ready();
4145
4146 while (1) {
4147 DBG("Health check ready");
4148
4149 /* Inifinite blocking call, waiting for transmission */
4150 restart:
4151 ret = lttng_poll_wait(&events, -1);
4152 if (ret < 0) {
4153 /*
4154 * Restart interrupted system call.
4155 */
4156 if (errno == EINTR) {
4157 goto restart;
4158 }
4159 goto error;
4160 }
4161
4162 nb_fd = ret;
4163
4164 for (i = 0; i < nb_fd; i++) {
4165 /* Fetch once the poll data */
4166 revents = LTTNG_POLL_GETEV(&events, i);
4167 pollfd = LTTNG_POLL_GETFD(&events, i);
4168
4169 if (!revents) {
4170 /* No activity for this FD (poll implementation). */
4171 continue;
4172 }
4173
4174 /* Thread quit pipe has been closed. Killing thread. */
4175 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4176 if (ret) {
4177 err = 0;
4178 goto exit;
4179 }
4180
4181 /* Event on the registration socket */
4182 if (pollfd == sock) {
4183 if (revents & LPOLLIN) {
4184 continue;
4185 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4186 ERR("Health socket poll error");
4187 goto error;
4188 } else {
4189 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4190 goto error;
4191 }
4192 }
4193 }
4194
4195 new_sock = lttcomm_accept_unix_sock(sock);
4196 if (new_sock < 0) {
4197 goto error;
4198 }
4199
4200 /*
4201 * Set the CLOEXEC flag. Return code is useless because either way, the
4202 * show must go on.
4203 */
4204 (void) utils_set_fd_cloexec(new_sock);
4205
4206 DBG("Receiving data from client for health...");
4207 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4208 if (ret <= 0) {
4209 DBG("Nothing recv() from client... continuing");
4210 ret = close(new_sock);
4211 if (ret) {
4212 PERROR("close");
4213 }
4214 continue;
4215 }
4216
4217 rcu_thread_online();
4218
4219 memset(&reply, 0, sizeof(reply));
4220 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4221 /*
4222 * health_check_state returns 0 if health is
4223 * bad.
4224 */
4225 if (!health_check_state(health_sessiond, i)) {
4226 reply.ret_code |= 1ULL << i;
4227 }
4228 }
4229
4230 DBG2("Health check return value %" PRIx64, reply.ret_code);
4231
4232 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4233 if (ret < 0) {
4234 ERR("Failed to send health data back to client");
4235 }
4236
4237 /* End of transmission */
4238 ret = close(new_sock);
4239 if (ret) {
4240 PERROR("close");
4241 }
4242 }
4243
4244 exit:
4245 error:
4246 if (err) {
4247 ERR("Health error occurred in %s", __func__);
4248 }
4249 DBG("Health check thread dying");
4250 unlink(config.health_unix_sock_path.value);
4251 if (sock >= 0) {
4252 ret = close(sock);
4253 if (ret) {
4254 PERROR("close");
4255 }
4256 }
4257
4258 lttng_poll_clean(&events);
4259 stop_threads();
4260 rcu_unregister_thread();
4261 return NULL;
4262 }
4263
4264 /*
4265 * This thread manage all clients request using the unix client socket for
4266 * communication.
4267 */
4268 static void *thread_manage_clients(void *data)
4269 {
4270 int sock = -1, ret, i, pollfd, err = -1;
4271 int sock_error;
4272 uint32_t revents, nb_fd;
4273 struct command_ctx *cmd_ctx = NULL;
4274 struct lttng_poll_event events;
4275
4276 DBG("[thread] Manage client started");
4277
4278 rcu_register_thread();
4279
4280 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4281
4282 health_code_update();
4283
4284 ret = lttcomm_listen_unix_sock(client_sock);
4285 if (ret < 0) {
4286 goto error_listen;
4287 }
4288
4289 /*
4290 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4291 * more will be added to this poll set.
4292 */
4293 ret = sessiond_set_thread_pollset(&events, 2);
4294 if (ret < 0) {
4295 goto error_create_poll;
4296 }
4297
4298 /* Add the application registration socket */
4299 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4300 if (ret < 0) {
4301 goto error;
4302 }
4303
4304 sessiond_notify_ready();
4305 ret = sem_post(&load_info->message_thread_ready);
4306 if (ret) {
4307 PERROR("sem_post message_thread_ready");
4308 goto error;
4309 }
4310
4311 /* This testpoint is after we signal readiness to the parent. */
4312 if (testpoint(sessiond_thread_manage_clients)) {
4313 goto error;
4314 }
4315
4316 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4317 goto error;
4318 }
4319
4320 health_code_update();
4321
4322 while (1) {
4323 DBG("Accepting client command ...");
4324
4325 /* Inifinite blocking call, waiting for transmission */
4326 restart:
4327 health_poll_entry();
4328 ret = lttng_poll_wait(&events, -1);
4329 health_poll_exit();
4330 if (ret < 0) {
4331 /*
4332 * Restart interrupted system call.
4333 */
4334 if (errno == EINTR) {
4335 goto restart;
4336 }
4337 goto error;
4338 }
4339
4340 nb_fd = ret;
4341
4342 for (i = 0; i < nb_fd; i++) {
4343 /* Fetch once the poll data */
4344 revents = LTTNG_POLL_GETEV(&events, i);
4345 pollfd = LTTNG_POLL_GETFD(&events, i);
4346
4347 health_code_update();
4348
4349 if (!revents) {
4350 /* No activity for this FD (poll implementation). */
4351 continue;
4352 }
4353
4354 /* Thread quit pipe has been closed. Killing thread. */
4355 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4356 if (ret) {
4357 err = 0;
4358 goto exit;
4359 }
4360
4361 /* Event on the registration socket */
4362 if (pollfd == client_sock) {
4363 if (revents & LPOLLIN) {
4364 continue;
4365 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4366 ERR("Client socket poll error");
4367 goto error;
4368 } else {
4369 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4370 goto error;
4371 }
4372 }
4373 }
4374
4375 DBG("Wait for client response");
4376
4377 health_code_update();
4378
4379 sock = lttcomm_accept_unix_sock(client_sock);
4380 if (sock < 0) {
4381 goto error;
4382 }
4383
4384 /*
4385 * Set the CLOEXEC flag. Return code is useless because either way, the
4386 * show must go on.
4387 */
4388 (void) utils_set_fd_cloexec(sock);
4389
4390 /* Set socket option for credentials retrieval */
4391 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4392 if (ret < 0) {
4393 goto error;
4394 }
4395
4396 /* Allocate context command to process the client request */
4397 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4398 if (cmd_ctx == NULL) {
4399 PERROR("zmalloc cmd_ctx");
4400 goto error;
4401 }
4402
4403 /* Allocate data buffer for reception */
4404 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4405 if (cmd_ctx->lsm == NULL) {
4406 PERROR("zmalloc cmd_ctx->lsm");
4407 goto error;
4408 }
4409
4410 cmd_ctx->llm = NULL;
4411 cmd_ctx->session = NULL;
4412
4413 health_code_update();
4414
4415 /*
4416 * Data is received from the lttng client. The struct
4417 * lttcomm_session_msg (lsm) contains the command and data request of
4418 * the client.
4419 */
4420 DBG("Receiving data from client ...");
4421 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4422 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4423 if (ret <= 0) {
4424 DBG("Nothing recv() from client... continuing");
4425 ret = close(sock);
4426 if (ret) {
4427 PERROR("close");
4428 }
4429 sock = -1;
4430 clean_command_ctx(&cmd_ctx);
4431 continue;
4432 }
4433
4434 health_code_update();
4435
4436 // TODO: Validate cmd_ctx including sanity check for
4437 // security purpose.
4438
4439 rcu_thread_online();
4440 /*
4441 * This function dispatch the work to the kernel or userspace tracer
4442 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4443 * informations for the client. The command context struct contains
4444 * everything this function may needs.
4445 */
4446 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4447 rcu_thread_offline();
4448 if (ret < 0) {
4449 ret = close(sock);
4450 if (ret) {
4451 PERROR("close");
4452 }
4453 sock = -1;
4454 /*
4455 * TODO: Inform client somehow of the fatal error. At
4456 * this point, ret < 0 means that a zmalloc failed
4457 * (ENOMEM). Error detected but still accept
4458 * command, unless a socket error has been
4459 * detected.
4460 */
4461 clean_command_ctx(&cmd_ctx);
4462 continue;
4463 }
4464
4465 health_code_update();
4466
4467 DBG("Sending response (size: %d, retcode: %s (%d))",
4468 cmd_ctx->lttng_msg_size,
4469 lttng_strerror(-cmd_ctx->llm->ret_code),
4470 cmd_ctx->llm->ret_code);
4471 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4472 if (ret < 0) {
4473 ERR("Failed to send data back to client");
4474 }
4475
4476 /* End of transmission */
4477 ret = close(sock);
4478 if (ret) {
4479 PERROR("close");
4480 }
4481 sock = -1;
4482
4483 clean_command_ctx(&cmd_ctx);
4484
4485 health_code_update();
4486 }
4487
4488 exit:
4489 error:
4490 if (sock >= 0) {
4491 ret = close(sock);
4492 if (ret) {
4493 PERROR("close");
4494 }
4495 }
4496
4497 lttng_poll_clean(&events);
4498 clean_command_ctx(&cmd_ctx);
4499
4500 error_listen:
4501 error_create_poll:
4502 unlink(config.client_unix_sock_path.value);
4503 if (client_sock >= 0) {
4504 ret = close(client_sock);
4505 if (ret) {
4506 PERROR("close");
4507 }
4508 }
4509
4510 if (err) {
4511 health_error();
4512 ERR("Health error occurred in %s", __func__);
4513 }
4514
4515 health_unregister(health_sessiond);
4516
4517 DBG("Client thread dying");
4518
4519 rcu_unregister_thread();
4520
4521 /*
4522 * Since we are creating the consumer threads, we own them, so we need
4523 * to join them before our thread exits.
4524 */
4525 ret = join_consumer_thread(&kconsumer_data);
4526 if (ret) {
4527 errno = ret;
4528 PERROR("join_consumer");
4529 }
4530
4531 ret = join_consumer_thread(&ustconsumer32_data);
4532 if (ret) {
4533 errno = ret;
4534 PERROR("join_consumer ust32");
4535 }
4536
4537 ret = join_consumer_thread(&ustconsumer64_data);
4538 if (ret) {
4539 errno = ret;
4540 PERROR("join_consumer ust64");
4541 }
4542 return NULL;
4543 }
4544
4545 static int string_match(const char *str1, const char *str2)
4546 {
4547 return (str1 && str2) && !strcmp(str1, str2);
4548 }
4549
4550 /*
4551 * Take an option from the getopt output and set it in the right variable to be
4552 * used later.
4553 *
4554 * Return 0 on success else a negative value.
4555 */
4556 static int set_option(int opt, const char *arg, const char *optname)
4557 {
4558 int ret = 0;
4559
4560 if (string_match(optname, "client-sock") || opt == 'c') {
4561 if (!arg || *arg == '\0') {
4562 ret = -EINVAL;
4563 goto end;
4564 }
4565 if (lttng_is_setuid_setgid()) {
4566 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4567 "-c, --client-sock");
4568 } else {
4569 config_string_set(&config.client_unix_sock_path,
4570 strdup(arg));
4571 if (!config.client_unix_sock_path.value) {
4572 ret = -ENOMEM;
4573 PERROR("strdup");
4574 }
4575 }
4576 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4577 if (!arg || *arg == '\0') {
4578 ret = -EINVAL;
4579 goto end;
4580 }
4581 if (lttng_is_setuid_setgid()) {
4582 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4583 "-a, --apps-sock");
4584 } else {
4585 config_string_set(&config.apps_unix_sock_path,
4586 strdup(arg));
4587 if (!config.apps_unix_sock_path.value) {
4588 ret = -ENOMEM;
4589 PERROR("strdup");
4590 }
4591 }
4592 } else if (string_match(optname, "daemonize") || opt == 'd') {
4593 config.daemonize = true;
4594 } else if (string_match(optname, "background") || opt == 'b') {
4595 config.background = true;
4596 } else if (string_match(optname, "group") || opt == 'g') {
4597 if (!arg || *arg == '\0') {
4598 ret = -EINVAL;
4599 goto end;
4600 }
4601 if (lttng_is_setuid_setgid()) {
4602 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4603 "-g, --group");
4604 } else {
4605 config_string_set(&config.tracing_group_name,
4606 strdup(arg));
4607 if (!config.tracing_group_name.value) {
4608 ret = -ENOMEM;
4609 PERROR("strdup");
4610 }
4611 }
4612 } else if (string_match(optname, "help") || opt == 'h') {
4613 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4614 if (ret) {
4615 ERR("Cannot show --help for `lttng-sessiond`");
4616 perror("exec");
4617 }
4618 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4619 } else if (string_match(optname, "version") || opt == 'V') {
4620 fprintf(stdout, "%s\n", VERSION);
4621 exit(EXIT_SUCCESS);
4622 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4623 config.sig_parent = true;
4624 } else if (string_match(optname, "kconsumerd-err-sock")) {
4625 if (!arg || *arg == '\0') {
4626 ret = -EINVAL;
4627 goto end;
4628 }
4629 if (lttng_is_setuid_setgid()) {
4630 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4631 "--kconsumerd-err-sock");
4632 } else {
4633 config_string_set(&config.kconsumerd_err_unix_sock_path,
4634 strdup(arg));
4635 if (!config.kconsumerd_err_unix_sock_path.value) {
4636 ret = -ENOMEM;
4637 PERROR("strdup");
4638 }
4639 }
4640 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4641 if (!arg || *arg == '\0') {
4642 ret = -EINVAL;
4643 goto end;
4644 }
4645 if (lttng_is_setuid_setgid()) {
4646 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4647 "--kconsumerd-cmd-sock");
4648 } else {
4649 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
4650 strdup(arg));
4651 if (!config.kconsumerd_cmd_unix_sock_path.value) {
4652 ret = -ENOMEM;
4653 PERROR("strdup");
4654 }
4655 }
4656 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4657 if (!arg || *arg == '\0') {
4658 ret = -EINVAL;
4659 goto end;
4660 }
4661 if (lttng_is_setuid_setgid()) {
4662 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4663 "--ustconsumerd64-err-sock");
4664 } else {
4665 config_string_set(&config.consumerd64_err_unix_sock_path,
4666 strdup(arg));
4667 if (!config.consumerd64_err_unix_sock_path.value) {
4668 ret = -ENOMEM;
4669 PERROR("strdup");
4670 }
4671 }
4672 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4673 if (!arg || *arg == '\0') {
4674 ret = -EINVAL;
4675 goto end;
4676 }
4677 if (lttng_is_setuid_setgid()) {
4678 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4679 "--ustconsumerd64-cmd-sock");
4680 } else {
4681 config_string_set(&config.consumerd64_cmd_unix_sock_path,
4682 strdup(arg));
4683 if (!config.consumerd64_cmd_unix_sock_path.value) {
4684 ret = -ENOMEM;
4685 PERROR("strdup");
4686 }
4687 }
4688 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4689 if (!arg || *arg == '\0') {
4690 ret = -EINVAL;
4691 goto end;
4692 }
4693 if (lttng_is_setuid_setgid()) {
4694 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4695 "--ustconsumerd32-err-sock");
4696 } else {
4697 config_string_set(&config.consumerd32_err_unix_sock_path,
4698 strdup(arg));
4699 if (!config.consumerd32_err_unix_sock_path.value) {
4700 ret = -ENOMEM;
4701 PERROR("strdup");
4702 }
4703 }
4704 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4705 if (!arg || *arg == '\0') {
4706 ret = -EINVAL;
4707 goto end;
4708 }
4709 if (lttng_is_setuid_setgid()) {
4710 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4711 "--ustconsumerd32-cmd-sock");
4712 } else {
4713 config_string_set(&config.consumerd32_cmd_unix_sock_path,
4714 strdup(arg));
4715 if (!config.consumerd32_cmd_unix_sock_path.value) {
4716 ret = -ENOMEM;
4717 PERROR("strdup");
4718 }
4719 }
4720 } else if (string_match(optname, "no-kernel")) {
4721 config.no_kernel = true;
4722 } else if (string_match(optname, "quiet") || opt == 'q') {
4723 config.quiet = true;
4724 } else if (string_match(optname, "verbose") || opt == 'v') {
4725 /* Verbose level can increase using multiple -v */
4726 if (arg) {
4727 /* Value obtained from config file */
4728 config.verbose = config_parse_value(arg);
4729 } else {
4730 /* -v used on command line */
4731 config.verbose++;
4732 }
4733 /* Clamp value to [0, 3] */
4734 config.verbose = config.verbose < 0 ? 0 :
4735 (config.verbose <= 3 ? config.verbose : 3);
4736 } else if (string_match(optname, "verbose-consumer")) {
4737 if (arg) {
4738 config.verbose_consumer = config_parse_value(arg);
4739 } else {
4740 config.verbose_consumer++;
4741 }
4742 } else if (string_match(optname, "consumerd32-path")) {
4743 if (!arg || *arg == '\0') {
4744 ret = -EINVAL;
4745 goto end;
4746 }
4747 if (lttng_is_setuid_setgid()) {
4748 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4749 "--consumerd32-path");
4750 } else {
4751 config_string_set(&config.consumerd32_bin_path,
4752 strdup(arg));
4753 if (!config.consumerd32_bin_path.value) {
4754 PERROR("strdup");
4755 ret = -ENOMEM;
4756 }
4757 }
4758 } else if (string_match(optname, "consumerd32-libdir")) {
4759 if (!arg || *arg == '\0') {
4760 ret = -EINVAL;
4761 goto end;
4762 }
4763 if (lttng_is_setuid_setgid()) {
4764 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4765 "--consumerd32-libdir");
4766 } else {
4767 config_string_set(&config.consumerd32_lib_dir,
4768 strdup(arg));
4769 if (!config.consumerd32_lib_dir.value) {
4770 PERROR("strdup");
4771 ret = -ENOMEM;
4772 }
4773 }
4774 } else if (string_match(optname, "consumerd64-path")) {
4775 if (!arg || *arg == '\0') {
4776 ret = -EINVAL;
4777 goto end;
4778 }
4779 if (lttng_is_setuid_setgid()) {
4780 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4781 "--consumerd64-path");
4782 } else {
4783 config_string_set(&config.consumerd64_bin_path,
4784 strdup(arg));
4785 if (!config.consumerd64_bin_path.value) {
4786 PERROR("strdup");
4787 ret = -ENOMEM;
4788 }
4789 }
4790 } else if (string_match(optname, "consumerd64-libdir")) {
4791 if (!arg || *arg == '\0') {
4792 ret = -EINVAL;
4793 goto end;
4794 }
4795 if (lttng_is_setuid_setgid()) {
4796 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4797 "--consumerd64-libdir");
4798 } else {
4799 config_string_set(&config.consumerd64_lib_dir,
4800 strdup(arg));
4801 if (!config.consumerd64_lib_dir.value) {
4802 PERROR("strdup");
4803 ret = -ENOMEM;
4804 }
4805 }
4806 } else if (string_match(optname, "pidfile") || opt == 'p') {
4807 if (!arg || *arg == '\0') {
4808 ret = -EINVAL;
4809 goto end;
4810 }
4811 if (lttng_is_setuid_setgid()) {
4812 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4813 "-p, --pidfile");
4814 } else {
4815 config_string_set(&config.pid_file_path, strdup(arg));
4816 if (!config.pid_file_path.value) {
4817 PERROR("strdup");
4818 ret = -ENOMEM;
4819 }
4820 }
4821 } else if (string_match(optname, "agent-tcp-port")) {
4822 if (!arg || *arg == '\0') {
4823 ret = -EINVAL;
4824 goto end;
4825 }
4826 if (lttng_is_setuid_setgid()) {
4827 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4828 "--agent-tcp-port");
4829 } else {
4830 unsigned long v;
4831
4832 errno = 0;
4833 v = strtoul(arg, NULL, 0);
4834 if (errno != 0 || !isdigit(arg[0])) {
4835 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4836 return -1;
4837 }
4838 if (v == 0 || v >= 65535) {
4839 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4840 return -1;
4841 }
4842 config.agent_tcp_port = (uint32_t) v;
4843 DBG3("Agent TCP port set to non default: %u", config.agent_tcp_port);
4844 }
4845 } else if (string_match(optname, "load") || opt == 'l') {
4846 if (!arg || *arg == '\0') {
4847 ret = -EINVAL;
4848 goto end;
4849 }
4850 if (lttng_is_setuid_setgid()) {
4851 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4852 "-l, --load");
4853 } else {
4854 config_string_set(&config.load_session_path, strdup(arg));
4855 if (!config.load_session_path.value) {
4856 PERROR("strdup");
4857 ret = -ENOMEM;
4858 }
4859 }
4860 } else if (string_match(optname, "kmod-probes")) {
4861 if (!arg || *arg == '\0') {
4862 ret = -EINVAL;
4863 goto end;
4864 }
4865 if (lttng_is_setuid_setgid()) {
4866 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4867 "--kmod-probes");
4868 } else {
4869 config_string_set(&config.kmod_probes_list, strdup(arg));
4870 if (!config.kmod_probes_list.value) {
4871 PERROR("strdup");
4872 ret = -ENOMEM;
4873 }
4874 }
4875 } else if (string_match(optname, "extra-kmod-probes")) {
4876 if (!arg || *arg == '\0') {
4877 ret = -EINVAL;
4878 goto end;
4879 }
4880 if (lttng_is_setuid_setgid()) {
4881 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4882 "--extra-kmod-probes");
4883 } else {
4884 config_string_set(&config.kmod_extra_probes_list,
4885 strdup(arg));
4886 if (!config.kmod_extra_probes_list.value) {
4887 PERROR("strdup");
4888 ret = -ENOMEM;
4889 }
4890 }
4891 } else if (string_match(optname, "config") || opt == 'f') {
4892 /* This is handled in set_options() thus silent skip. */
4893 goto end;
4894 } else {
4895 /* Unknown option or other error.
4896 * Error is printed by getopt, just return */
4897 ret = -1;
4898 }
4899
4900 end:
4901 if (ret == -EINVAL) {
4902 const char *opt_name = "unknown";
4903 int i;
4904
4905 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4906 i++) {
4907 if (opt == long_options[i].val) {
4908 opt_name = long_options[i].name;
4909 break;
4910 }
4911 }
4912
4913 WARN("Invalid argument provided for option \"%s\", using default value.",
4914 opt_name);
4915 }
4916
4917 return ret;
4918 }
4919
4920 /*
4921 * config_entry_handler_cb used to handle options read from a config file.
4922 * See config_entry_handler_cb comment in common/config/session-config.h for the
4923 * return value conventions.
4924 */
4925 static int config_entry_handler(const struct config_entry *entry, void *unused)
4926 {
4927 int ret = 0, i;
4928
4929 if (!entry || !entry->name || !entry->value) {
4930 ret = -EINVAL;
4931 goto end;
4932 }
4933
4934 /* Check if the option is to be ignored */
4935 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4936 if (!strcmp(entry->name, config_ignore_options[i])) {
4937 goto end;
4938 }
4939 }
4940
4941 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4942 i++) {
4943
4944 /* Ignore if not fully matched. */
4945 if (strcmp(entry->name, long_options[i].name)) {
4946 continue;
4947 }
4948
4949 /*
4950 * If the option takes no argument on the command line, we have to
4951 * check if the value is "true". We support non-zero numeric values,
4952 * true, on and yes.
4953 */
4954 if (!long_options[i].has_arg) {
4955 ret = config_parse_value(entry->value);
4956 if (ret <= 0) {
4957 if (ret) {
4958 WARN("Invalid configuration value \"%s\" for option %s",
4959 entry->value, entry->name);
4960 }
4961 /* False, skip boolean config option. */
4962 goto end;
4963 }
4964 }
4965
4966 ret = set_option(long_options[i].val, entry->value, entry->name);
4967 goto end;
4968 }
4969
4970 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4971
4972 end:
4973 return ret;
4974 }
4975
4976 /*
4977 * daemon configuration loading and argument parsing
4978 */
4979 static int set_options(int argc, char **argv)
4980 {
4981 int ret = 0, c = 0, option_index = 0;
4982 int orig_optopt = optopt, orig_optind = optind;
4983 char *optstring;
4984 const char *config_path = NULL;
4985
4986 optstring = utils_generate_optstring(long_options,
4987 sizeof(long_options) / sizeof(struct option));
4988 if (!optstring) {
4989 ret = -ENOMEM;
4990 goto end;
4991 }
4992
4993 /* Check for the --config option */
4994 while ((c = getopt_long(argc, argv, optstring, long_options,
4995 &option_index)) != -1) {
4996 if (c == '?') {
4997 ret = -EINVAL;
4998 goto end;
4999 } else if (c != 'f') {
5000 /* if not equal to --config option. */
5001 continue;
5002 }
5003
5004 if (lttng_is_setuid_setgid()) {
5005 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5006 "-f, --config");
5007 } else {
5008 config_path = utils_expand_path(optarg);
5009 if (!config_path) {
5010 ERR("Failed to resolve path: %s", optarg);
5011 }
5012 }
5013 }
5014
5015 ret = config_get_section_entries(config_path, config_section_name,
5016 config_entry_handler, NULL);
5017 if (ret) {
5018 if (ret > 0) {
5019 ERR("Invalid configuration option at line %i", ret);
5020 ret = -1;
5021 }
5022 goto end;
5023 }
5024
5025 /* Reset getopt's global state */
5026 optopt = orig_optopt;
5027 optind = orig_optind;
5028 while (1) {
5029 option_index = -1;
5030 /*
5031 * getopt_long() will not set option_index if it encounters a
5032 * short option.
5033 */
5034 c = getopt_long(argc, argv, optstring, long_options,
5035 &option_index);
5036 if (c == -1) {
5037 break;
5038 }
5039
5040 /*
5041 * Pass NULL as the long option name if popt left the index
5042 * unset.
5043 */
5044 ret = set_option(c, optarg,
5045 option_index < 0 ? NULL :
5046 long_options[option_index].name);
5047 if (ret < 0) {
5048 break;
5049 }
5050 }
5051
5052 end:
5053 free(optstring);
5054 return ret;
5055 }
5056
5057 /*
5058 * Creates the two needed socket by the daemon.
5059 * apps_sock - The communication socket for all UST apps.
5060 * client_sock - The communication of the cli tool (lttng).
5061 */
5062 static int init_daemon_socket(void)
5063 {
5064 int ret = 0;
5065 mode_t old_umask;
5066
5067 old_umask = umask(0);
5068
5069 /* Create client tool unix socket */
5070 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5071 if (client_sock < 0) {
5072 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5073 ret = -1;
5074 goto end;
5075 }
5076
5077 /* Set the cloexec flag */
5078 ret = utils_set_fd_cloexec(client_sock);
5079 if (ret < 0) {
5080 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5081 "Continuing but note that the consumer daemon will have a "
5082 "reference to this socket on exec()", client_sock);
5083 }
5084
5085 /* File permission MUST be 660 */
5086 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5087 if (ret < 0) {
5088 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5089 PERROR("chmod");
5090 goto end;
5091 }
5092
5093 /* Create the application unix socket */
5094 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5095 if (apps_sock < 0) {
5096 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5097 ret = -1;
5098 goto end;
5099 }
5100
5101 /* Set the cloexec flag */
5102 ret = utils_set_fd_cloexec(apps_sock);
5103 if (ret < 0) {
5104 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5105 "Continuing but note that the consumer daemon will have a "
5106 "reference to this socket on exec()", apps_sock);
5107 }
5108
5109 /* File permission MUST be 666 */
5110 ret = chmod(config.apps_unix_sock_path.value,
5111 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5112 if (ret < 0) {
5113 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5114 PERROR("chmod");
5115 goto end;
5116 }
5117
5118 DBG3("Session daemon client socket %d and application socket %d created",
5119 client_sock, apps_sock);
5120
5121 end:
5122 umask(old_umask);
5123 return ret;
5124 }
5125
5126 /*
5127 * Check if the global socket is available, and if a daemon is answering at the
5128 * other side. If yes, error is returned.
5129 */
5130 static int check_existing_daemon(void)
5131 {
5132 /* Is there anybody out there ? */
5133 if (lttng_session_daemon_alive()) {
5134 return -EEXIST;
5135 }
5136
5137 return 0;
5138 }
5139
5140 /*
5141 * Set the tracing group gid onto the client socket.
5142 *
5143 * Race window between mkdir and chown is OK because we are going from more
5144 * permissive (root.root) to less permissive (root.tracing).
5145 */
5146 static int set_permissions(char *rundir)
5147 {
5148 int ret;
5149 gid_t gid;
5150
5151 gid = utils_get_group_id(config.tracing_group_name.value);
5152
5153 /* Set lttng run dir */
5154 ret = chown(rundir, 0, gid);
5155 if (ret < 0) {
5156 ERR("Unable to set group on %s", rundir);
5157 PERROR("chown");
5158 }
5159
5160 /*
5161 * Ensure all applications and tracing group can search the run
5162 * dir. Allow everyone to read the directory, since it does not
5163 * buy us anything to hide its content.
5164 */
5165 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5166 if (ret < 0) {
5167 ERR("Unable to set permissions on %s", rundir);
5168 PERROR("chmod");
5169 }
5170
5171 /* lttng client socket path */
5172 ret = chown(config.client_unix_sock_path.value, 0, gid);
5173 if (ret < 0) {
5174 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5175 PERROR("chown");
5176 }
5177
5178 /* kconsumer error socket path */
5179 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5180 if (ret < 0) {
5181 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5182 PERROR("chown");
5183 }
5184
5185 /* 64-bit ustconsumer error socket path */
5186 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5187 if (ret < 0) {
5188 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5189 PERROR("chown");
5190 }
5191
5192 /* 32-bit ustconsumer compat32 error socket path */
5193 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5194 if (ret < 0) {
5195 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5196 PERROR("chown");
5197 }
5198
5199 DBG("All permissions are set");
5200
5201 return ret;
5202 }
5203
5204 /*
5205 * Create the lttng run directory needed for all global sockets and pipe.
5206 */
5207 static int create_lttng_rundir(void)
5208 {
5209 int ret;
5210
5211 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5212
5213 ret = mkdir(config.rundir.value, S_IRWXU);
5214 if (ret < 0) {
5215 if (errno != EEXIST) {
5216 ERR("Unable to create %s", config.rundir.value);
5217 goto error;
5218 } else {
5219 ret = 0;
5220 }
5221 }
5222
5223 error:
5224 return ret;
5225 }
5226
5227 /*
5228 * Setup sockets and directory needed by the consumerds' communication with the
5229 * session daemon.
5230 */
5231 static int set_consumer_sockets(struct consumer_data *consumer_data)
5232 {
5233 int ret;
5234 char *path = NULL;
5235
5236 switch (consumer_data->type) {
5237 case LTTNG_CONSUMER_KERNEL:
5238 path = config.kconsumerd_path.value;
5239 break;
5240 case LTTNG_CONSUMER64_UST:
5241 path = config.consumerd64_path.value;
5242 break;
5243 case LTTNG_CONSUMER32_UST:
5244 path = config.consumerd32_path.value;
5245 break;
5246 default:
5247 ERR("Consumer type unknown");
5248 ret = -EINVAL;
5249 goto error;
5250 }
5251 assert(path);
5252
5253 DBG2("Creating consumer directory: %s", path);
5254
5255 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5256 if (ret < 0 && errno != EEXIST) {
5257 PERROR("mkdir");
5258 ERR("Failed to create %s", path);
5259 goto error;
5260 }
5261 if (is_root) {
5262 ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value));
5263 if (ret < 0) {
5264 ERR("Unable to set group on %s", path);
5265 PERROR("chown");
5266 goto error;
5267 }
5268 }
5269
5270 /* Create the consumerd error unix socket */
5271 consumer_data->err_sock =
5272 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5273 if (consumer_data->err_sock < 0) {
5274 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5275 ret = -1;
5276 goto error;
5277 }
5278
5279 /*
5280 * Set the CLOEXEC flag. Return code is useless because either way, the
5281 * show must go on.
5282 */
5283 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5284 if (ret < 0) {
5285 PERROR("utils_set_fd_cloexec");
5286 /* continue anyway */
5287 }
5288
5289 /* File permission MUST be 660 */
5290 ret = chmod(consumer_data->err_unix_sock_path,
5291 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5292 if (ret < 0) {
5293 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5294 PERROR("chmod");
5295 goto error;
5296 }
5297
5298 error:
5299 return ret;
5300 }
5301
5302 /*
5303 * Signal handler for the daemon
5304 *
5305 * Simply stop all worker threads, leaving main() return gracefully after
5306 * joining all threads and calling cleanup().
5307 */
5308 static void sighandler(int sig)
5309 {
5310 switch (sig) {
5311 case SIGINT:
5312 DBG("SIGINT caught");
5313 stop_threads();
5314 break;
5315 case SIGTERM:
5316 DBG("SIGTERM caught");
5317 stop_threads();
5318 break;
5319 case SIGUSR1:
5320 CMM_STORE_SHARED(recv_child_signal, 1);
5321 break;
5322 default:
5323 break;
5324 }
5325 }
5326
5327 /*
5328 * Setup signal handler for :
5329 * SIGINT, SIGTERM, SIGPIPE
5330 */
5331 static int set_signal_handler(void)
5332 {
5333 int ret = 0;
5334 struct sigaction sa;
5335 sigset_t sigset;
5336
5337 if ((ret = sigemptyset(&sigset)) < 0) {
5338 PERROR("sigemptyset");
5339 return ret;
5340 }
5341
5342 sa.sa_mask = sigset;
5343 sa.sa_flags = 0;
5344
5345 sa.sa_handler = sighandler;
5346 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5347 PERROR("sigaction");
5348 return ret;
5349 }
5350
5351 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5352 PERROR("sigaction");
5353 return ret;
5354 }
5355
5356 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5357 PERROR("sigaction");
5358 return ret;
5359 }
5360
5361 sa.sa_handler = SIG_IGN;
5362 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5363 PERROR("sigaction");
5364 return ret;
5365 }
5366
5367 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5368
5369 return ret;
5370 }
5371
5372 /*
5373 * Set open files limit to unlimited. This daemon can open a large number of
5374 * file descriptors in order to consume multiple kernel traces.
5375 */
5376 static void set_ulimit(void)
5377 {
5378 int ret;
5379 struct rlimit lim;
5380
5381 /* The kernel does not allow an infinite limit for open files */
5382 lim.rlim_cur = 65535;
5383 lim.rlim_max = 65535;
5384
5385 ret = setrlimit(RLIMIT_NOFILE, &lim);
5386 if (ret < 0) {
5387 PERROR("failed to set open files limit");
5388 }
5389 }
5390
5391 static int write_pidfile(void)
5392 {
5393 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5394 }
5395
5396 /*
5397 * Create lockfile using the rundir and return its fd.
5398 */
5399 static int create_lockfile(void)
5400 {
5401 return utils_create_lock_file(config.lock_file_path.value);
5402 }
5403
5404 /*
5405 * Write agent TCP port using the rundir.
5406 */
5407 static int write_agent_port(void)
5408 {
5409 return utils_create_pid_file(config.agent_tcp_port,
5410 config.agent_port_file_path.value);
5411 }
5412
5413 static int set_clock_plugin_env(void)
5414 {
5415 int ret = 0;
5416 char *env_value = NULL;
5417
5418 if (!config.lttng_ust_clock_plugin.value) {
5419 goto end;
5420 }
5421
5422 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5423 config.lttng_ust_clock_plugin.value);
5424 if (ret < 0) {
5425 PERROR("asprintf");
5426 goto end;
5427 }
5428
5429 ret = putenv(env_value);
5430 if (ret) {
5431 free(env_value);
5432 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5433 goto end;
5434 }
5435
5436 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5437 config.lttng_ust_clock_plugin.value);
5438 end:
5439 return ret;
5440 }
5441
5442 /*
5443 * main
5444 */
5445 int main(int argc, char **argv)
5446 {
5447 int ret = 0, retval = 0;
5448 void *status;
5449 const char *env_app_timeout;
5450 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5451 *ust64_channel_monitor_pipe = NULL,
5452 *kernel_channel_monitor_pipe = NULL;
5453 bool notification_thread_running = false;
5454
5455 init_kernel_workarounds();
5456
5457 rcu_register_thread();
5458
5459 if (set_signal_handler()) {
5460 retval = -1;
5461 goto exit_set_signal_handler;
5462 }
5463
5464 page_size = sysconf(_SC_PAGESIZE);
5465 if (page_size < 0) {
5466 PERROR("sysconf _SC_PAGESIZE");
5467 page_size = LONG_MAX;
5468 WARN("Fallback page size to %ld", page_size);
5469 }
5470
5471 ret = sessiond_config_init(&config);
5472 if (ret) {
5473 retval = -1;
5474 goto exit_set_signal_handler;
5475 }
5476
5477 /*
5478 * Parse arguments and load the daemon configuration file.
5479 *
5480 * We have an exit_options exit path to free memory reserved by
5481 * set_options. This is needed because the rest of sessiond_cleanup()
5482 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5483 * depends on set_options.
5484 */
5485 progname = argv[0];
5486 if (set_options(argc, argv)) {
5487 retval = -1;
5488 goto exit_options;
5489 }
5490
5491 /* Init config from environment variables. */
5492 sessiond_config_apply_env_config(&config);
5493
5494 /*
5495 * Resolve all paths received as arguments, configuration option, or
5496 * through environment variable as absolute paths. This is necessary
5497 * since daemonizing causes the sessiond's current working directory
5498 * to '/'.
5499 */
5500 ret = sessiond_config_resolve_paths(&config);
5501 if (ret) {
5502 goto exit_options;
5503 }
5504
5505 /* Apply config. */
5506 lttng_opt_verbose = config.verbose;
5507 lttng_opt_quiet = config.quiet;
5508 kconsumer_data.err_unix_sock_path =
5509 config.kconsumerd_err_unix_sock_path.value;
5510 kconsumer_data.cmd_unix_sock_path =
5511 config.kconsumerd_cmd_unix_sock_path.value;
5512 ustconsumer32_data.err_unix_sock_path =
5513 config.consumerd32_err_unix_sock_path.value;
5514 ustconsumer32_data.cmd_unix_sock_path =
5515 config.consumerd32_cmd_unix_sock_path.value;
5516 ustconsumer64_data.err_unix_sock_path =
5517 config.consumerd64_err_unix_sock_path.value;
5518 ustconsumer64_data.cmd_unix_sock_path =
5519 config.consumerd64_cmd_unix_sock_path.value;
5520 set_clock_plugin_env();
5521
5522 sessiond_config_log(&config);
5523
5524 /* Daemonize */
5525 if (config.daemonize || config.background) {
5526 int i;
5527
5528 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5529 !config.background);
5530 if (ret < 0) {
5531 retval = -1;
5532 goto exit_options;
5533 }
5534
5535 /*
5536 * We are in the child. Make sure all other file descriptors are
5537 * closed, in case we are called with more opened file
5538 * descriptors than the standard ones.
5539 */
5540 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5541 (void) close(i);
5542 }
5543 }
5544
5545 if (run_as_create_worker(argv[0]) < 0) {
5546 goto exit_create_run_as_worker_cleanup;
5547 }
5548
5549 /*
5550 * Starting from here, we can create threads. This needs to be after
5551 * lttng_daemonize due to RCU.
5552 */
5553
5554 /*
5555 * Initialize the health check subsystem. This call should set the
5556 * appropriate time values.
5557 */
5558 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5559 if (!health_sessiond) {
5560 PERROR("health_app_create error");
5561 retval = -1;
5562 goto exit_health_sessiond_cleanup;
5563 }
5564
5565 /* Create thread to clean up RCU hash tables */
5566 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5567 retval = -1;
5568 goto exit_ht_cleanup;
5569 }
5570
5571 /* Create thread quit pipe */
5572 if (init_thread_quit_pipe()) {
5573 retval = -1;
5574 goto exit_init_data;
5575 }
5576
5577 /* Check if daemon is UID = 0 */
5578 is_root = !getuid();
5579
5580 if (create_lttng_rundir()) {
5581 retval = -1;
5582 goto exit_init_data;
5583 }
5584
5585 if (is_root) {
5586 /* Create global run dir with root access */
5587
5588 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5589 if (!kernel_channel_monitor_pipe) {
5590 ERR("Failed to create kernel consumer channel monitor pipe");
5591 retval = -1;
5592 goto exit_init_data;
5593 }
5594 kconsumer_data.channel_monitor_pipe =
5595 lttng_pipe_release_writefd(
5596 kernel_channel_monitor_pipe);
5597 if (kconsumer_data.channel_monitor_pipe < 0) {
5598 retval = -1;
5599 goto exit_init_data;
5600 }
5601 }
5602
5603 lockfile_fd = create_lockfile();
5604 if (lockfile_fd < 0) {
5605 retval = -1;
5606 goto exit_init_data;
5607 }
5608
5609 /* Set consumer initial state */
5610 kernel_consumerd_state = CONSUMER_STOPPED;
5611 ust_consumerd_state = CONSUMER_STOPPED;
5612
5613 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5614 if (!ust32_channel_monitor_pipe) {
5615 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5616 retval = -1;
5617 goto exit_init_data;
5618 }
5619 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5620 ust32_channel_monitor_pipe);
5621 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5622 retval = -1;
5623 goto exit_init_data;
5624 }
5625
5626 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5627 if (!ust64_channel_monitor_pipe) {
5628 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5629 retval = -1;
5630 goto exit_init_data;
5631 }
5632 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5633 ust64_channel_monitor_pipe);
5634 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5635 retval = -1;
5636 goto exit_init_data;
5637 }
5638
5639 /*
5640 * See if daemon already exist.
5641 */
5642 if (check_existing_daemon()) {
5643 ERR("Already running daemon.\n");
5644 /*
5645 * We do not goto exit because we must not cleanup()
5646 * because a daemon is already running.
5647 */
5648 retval = -1;
5649 goto exit_init_data;
5650 }
5651
5652 /*
5653 * Init UST app hash table. Alloc hash table before this point since
5654 * cleanup() can get called after that point.
5655 */
5656 if (ust_app_ht_alloc()) {
5657 ERR("Failed to allocate UST app hash table");
5658 retval = -1;
5659 goto exit_init_data;
5660 }
5661
5662 /*
5663 * Initialize agent app hash table. We allocate the hash table here
5664 * since cleanup() can get called after this point.
5665 */
5666 if (agent_app_ht_alloc()) {
5667 ERR("Failed to allocate Agent app hash table");
5668 retval = -1;
5669 goto exit_init_data;
5670 }
5671
5672 /*
5673 * These actions must be executed as root. We do that *after* setting up
5674 * the sockets path because we MUST make the check for another daemon using
5675 * those paths *before* trying to set the kernel consumer sockets and init
5676 * kernel tracer.
5677 */
5678 if (is_root) {
5679 if (set_consumer_sockets(&kconsumer_data)) {
5680 retval = -1;
5681 goto exit_init_data;
5682 }
5683
5684 /* Setup kernel tracer */
5685 if (!config.no_kernel) {
5686 init_kernel_tracer();
5687 if (kernel_tracer_fd >= 0) {
5688 ret = syscall_init_table();
5689 if (ret < 0) {
5690 ERR("Unable to populate syscall table. "
5691 "Syscall tracing won't work "
5692 "for this session daemon.");
5693 }
5694 }
5695 }
5696
5697 /* Set ulimit for open files */
5698 set_ulimit();
5699 }
5700 /* init lttng_fd tracking must be done after set_ulimit. */
5701 lttng_fd_init();
5702
5703 if (set_consumer_sockets(&ustconsumer64_data)) {
5704 retval = -1;
5705 goto exit_init_data;
5706 }
5707
5708 if (set_consumer_sockets(&ustconsumer32_data)) {
5709 retval = -1;
5710 goto exit_init_data;
5711 }
5712
5713 /* Setup the needed unix socket */
5714 if (init_daemon_socket()) {
5715 retval = -1;
5716 goto exit_init_data;
5717 }
5718
5719 /* Set credentials to socket */
5720 if (is_root && set_permissions(config.rundir.value)) {
5721 retval = -1;
5722 goto exit_init_data;
5723 }
5724
5725 /* Get parent pid if -S, --sig-parent is specified. */
5726 if (config.sig_parent) {
5727 ppid = getppid();
5728 }
5729
5730 /* Setup the kernel pipe for waking up the kernel thread */
5731 if (is_root && !config.no_kernel) {
5732 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5733 retval = -1;
5734 goto exit_init_data;
5735 }
5736 }
5737
5738 /* Setup the thread apps communication pipe. */
5739 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5740 retval = -1;
5741 goto exit_init_data;
5742 }
5743
5744 /* Setup the thread apps notify communication pipe. */
5745 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5746 retval = -1;
5747 goto exit_init_data;
5748 }
5749
5750 /* Initialize global buffer per UID and PID registry. */
5751 buffer_reg_init_uid_registry();
5752 buffer_reg_init_pid_registry();
5753
5754 /* Init UST command queue. */
5755 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5756
5757 /*
5758 * Get session list pointer. This pointer MUST NOT be free'd. This list
5759 * is statically declared in session.c
5760 */
5761 session_list_ptr = session_get_list();
5762
5763 cmd_init();
5764
5765 /* Check for the application socket timeout env variable. */
5766 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5767 if (env_app_timeout) {
5768 config.app_socket_timeout = atoi(env_app_timeout);
5769 } else {
5770 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5771 }
5772
5773 ret = write_pidfile();
5774 if (ret) {
5775 ERR("Error in write_pidfile");
5776 retval = -1;
5777 goto exit_init_data;
5778 }
5779 ret = write_agent_port();
5780 if (ret) {
5781 ERR("Error in write_agent_port");
5782 retval = -1;
5783 goto exit_init_data;
5784 }
5785
5786 /* Initialize communication library */
5787 lttcomm_init();
5788 /* Initialize TCP timeout values */
5789 lttcomm_inet_init();
5790
5791 if (load_session_init_data(&load_info) < 0) {
5792 retval = -1;
5793 goto exit_init_data;
5794 }
5795 load_info->path = config.load_session_path.value;
5796
5797 /* Create health-check thread. */
5798 ret = pthread_create(&health_thread, default_pthread_attr(),
5799 thread_manage_health, (void *) NULL);
5800 if (ret) {
5801 errno = ret;
5802 PERROR("pthread_create health");
5803 retval = -1;
5804 goto exit_health;
5805 }
5806
5807 /* notification_thread_data acquires the pipes' read side. */
5808 notification_thread_handle = notification_thread_handle_create(
5809 ust32_channel_monitor_pipe,
5810 ust64_channel_monitor_pipe,
5811 kernel_channel_monitor_pipe);
5812 if (!notification_thread_handle) {
5813 retval = -1;
5814 ERR("Failed to create notification thread shared data");
5815 stop_threads();
5816 goto exit_notification;
5817 }
5818
5819 /* Create notification thread. */
5820 ret = pthread_create(&notification_thread, default_pthread_attr(),
5821 thread_notification, notification_thread_handle);
5822 if (ret) {
5823 errno = ret;
5824 PERROR("pthread_create notification");
5825 retval = -1;
5826 stop_threads();
5827 goto exit_notification;
5828 }
5829 notification_thread_running = true;
5830
5831 /* Create thread to manage the client socket */
5832 ret = pthread_create(&client_thread, default_pthread_attr(),
5833 thread_manage_clients, (void *) NULL);
5834 if (ret) {
5835 errno = ret;
5836 PERROR("pthread_create clients");
5837 retval = -1;
5838 stop_threads();
5839 goto exit_client;
5840 }
5841
5842 /* Create thread to dispatch registration */
5843 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
5844 thread_dispatch_ust_registration, (void *) NULL);
5845 if (ret) {
5846 errno = ret;
5847 PERROR("pthread_create dispatch");
5848 retval = -1;
5849 stop_threads();
5850 goto exit_dispatch;
5851 }
5852
5853 /* Create thread to manage application registration. */
5854 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
5855 thread_registration_apps, (void *) NULL);
5856 if (ret) {
5857 errno = ret;
5858 PERROR("pthread_create registration");
5859 retval = -1;
5860 stop_threads();
5861 goto exit_reg_apps;
5862 }
5863
5864 /* Create thread to manage application socket */
5865 ret = pthread_create(&apps_thread, default_pthread_attr(),
5866 thread_manage_apps, (void *) NULL);
5867 if (ret) {
5868 errno = ret;
5869 PERROR("pthread_create apps");
5870 retval = -1;
5871 stop_threads();
5872 goto exit_apps;
5873 }
5874
5875 /* Create thread to manage application notify socket */
5876 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
5877 ust_thread_manage_notify, (void *) NULL);
5878 if (ret) {
5879 errno = ret;
5880 PERROR("pthread_create notify");
5881 retval = -1;
5882 stop_threads();
5883 goto exit_apps_notify;
5884 }
5885
5886 /* Create agent registration thread. */
5887 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
5888 agent_thread_manage_registration, (void *) NULL);
5889 if (ret) {
5890 errno = ret;
5891 PERROR("pthread_create agent");
5892 retval = -1;
5893 stop_threads();
5894 goto exit_agent_reg;
5895 }
5896
5897 /* Don't start this thread if kernel tracing is not requested nor root */
5898 if (is_root && !config.no_kernel) {
5899 /* Create kernel thread to manage kernel event */
5900 ret = pthread_create(&kernel_thread, default_pthread_attr(),
5901 thread_manage_kernel, (void *) NULL);
5902 if (ret) {
5903 errno = ret;
5904 PERROR("pthread_create kernel");
5905 retval = -1;
5906 stop_threads();
5907 goto exit_kernel;
5908 }
5909 }
5910
5911 /* Create session loading thread. */
5912 ret = pthread_create(&load_session_thread, default_pthread_attr(),
5913 thread_load_session, load_info);
5914 if (ret) {
5915 errno = ret;
5916 PERROR("pthread_create load_session_thread");
5917 retval = -1;
5918 stop_threads();
5919 goto exit_load_session;
5920 }
5921
5922 /*
5923 * This is where we start awaiting program completion (e.g. through
5924 * signal that asks threads to teardown).
5925 */
5926
5927 ret = pthread_join(load_session_thread, &status);
5928 if (ret) {
5929 errno = ret;
5930 PERROR("pthread_join load_session_thread");
5931 retval = -1;
5932 }
5933 exit_load_session:
5934
5935 if (is_root && !config.no_kernel) {
5936 ret = pthread_join(kernel_thread, &status);
5937 if (ret) {
5938 errno = ret;
5939 PERROR("pthread_join");
5940 retval = -1;
5941 }
5942 }
5943 exit_kernel:
5944
5945 ret = pthread_join(agent_reg_thread, &status);
5946 if (ret) {
5947 errno = ret;
5948 PERROR("pthread_join agent");
5949 retval = -1;
5950 }
5951 exit_agent_reg:
5952
5953 ret = pthread_join(apps_notify_thread, &status);
5954 if (ret) {
5955 errno = ret;
5956 PERROR("pthread_join apps notify");
5957 retval = -1;
5958 }
5959 exit_apps_notify:
5960
5961 ret = pthread_join(apps_thread, &status);
5962 if (ret) {
5963 errno = ret;
5964 PERROR("pthread_join apps");
5965 retval = -1;
5966 }
5967 exit_apps:
5968
5969 ret = pthread_join(reg_apps_thread, &status);
5970 if (ret) {
5971 errno = ret;
5972 PERROR("pthread_join");
5973 retval = -1;
5974 }
5975 exit_reg_apps:
5976
5977 /*
5978 * Join dispatch thread after joining reg_apps_thread to ensure
5979 * we don't leak applications in the queue.
5980 */
5981 ret = pthread_join(dispatch_thread, &status);
5982 if (ret) {
5983 errno = ret;
5984 PERROR("pthread_join");
5985 retval = -1;
5986 }
5987 exit_dispatch:
5988
5989 ret = pthread_join(client_thread, &status);
5990 if (ret) {
5991 errno = ret;
5992 PERROR("pthread_join");
5993 retval = -1;
5994 }
5995
5996 exit_client:
5997 exit_notification:
5998 ret = pthread_join(health_thread, &status);
5999 if (ret) {
6000 errno = ret;
6001 PERROR("pthread_join health thread");
6002 retval = -1;
6003 }
6004
6005 exit_health:
6006 exit_init_data:
6007 /*
6008 * Wait for all pending call_rcu work to complete before tearing
6009 * down data structures. call_rcu worker may be trying to
6010 * perform lookups in those structures.
6011 */
6012 rcu_barrier();
6013 /*
6014 * sessiond_cleanup() is called when no other thread is running, except
6015 * the ht_cleanup thread, which is needed to destroy the hash tables.
6016 */
6017 rcu_thread_online();
6018 sessiond_cleanup();
6019
6020 /*
6021 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6022 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6023 * the queue is empty before shutting down the clean-up thread.
6024 */
6025 rcu_barrier();
6026
6027 /*
6028 * The teardown of the notification system is performed after the
6029 * session daemon's teardown in order to allow it to be notified
6030 * of the active session and channels at the moment of the teardown.
6031 */
6032 if (notification_thread_handle) {
6033 if (notification_thread_running) {
6034 notification_thread_command_quit(
6035 notification_thread_handle);
6036 ret = pthread_join(notification_thread, &status);
6037 if (ret) {
6038 errno = ret;
6039 PERROR("pthread_join notification thread");
6040 retval = -1;
6041 }
6042 }
6043 notification_thread_handle_destroy(notification_thread_handle);
6044 }
6045
6046 rcu_thread_offline();
6047 rcu_unregister_thread();
6048
6049 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6050 if (ret) {
6051 retval = -1;
6052 }
6053 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6054 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6055 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6056 exit_ht_cleanup:
6057
6058 health_app_destroy(health_sessiond);
6059 exit_health_sessiond_cleanup:
6060 exit_create_run_as_worker_cleanup:
6061
6062 exit_options:
6063 sessiond_cleanup_options();
6064
6065 exit_set_signal_handler:
6066 if (!retval) {
6067 exit(EXIT_SUCCESS);
6068 } else {
6069 exit(EXIT_FAILURE);
6070 }
6071 }
This page took 0.201865 seconds and 6 git commands to generate.