Clean-up: reduce scope of dyanamically-allocated string
[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
1274 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1275 consumer_data->channel_monitor_pipe);
1276 if (ret) {
1277 goto error;
1278 }
1279 /* Discard the socket wrapper as it is no longer needed. */
1280 consumer_destroy_socket(cmd_socket_wrapper);
1281 cmd_socket_wrapper = NULL;
1282
1283 /* The thread is completely initialized, signal that it is ready. */
1284 signal_consumer_condition(consumer_data, 1);
1285
1286 /* Infinite blocking call, waiting for transmission */
1287 restart_poll:
1288 while (1) {
1289 health_code_update();
1290
1291 /* Exit the thread because the thread quit pipe has been triggered. */
1292 if (should_quit) {
1293 /* Not a health error. */
1294 err = 0;
1295 goto exit;
1296 }
1297
1298 health_poll_entry();
1299 ret = lttng_poll_wait(&events, -1);
1300 health_poll_exit();
1301 if (ret < 0) {
1302 /*
1303 * Restart interrupted system call.
1304 */
1305 if (errno == EINTR) {
1306 goto restart_poll;
1307 }
1308 goto error;
1309 }
1310
1311 nb_fd = ret;
1312
1313 for (i = 0; i < nb_fd; i++) {
1314 /* Fetch once the poll data */
1315 revents = LTTNG_POLL_GETEV(&events, i);
1316 pollfd = LTTNG_POLL_GETFD(&events, i);
1317
1318 health_code_update();
1319
1320 if (!revents) {
1321 /* No activity for this FD (poll implementation). */
1322 continue;
1323 }
1324
1325 /*
1326 * Thread quit pipe has been triggered, flag that we should stop
1327 * but continue the current loop to handle potential data from
1328 * consumer.
1329 */
1330 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1331
1332 if (pollfd == sock) {
1333 /* Event on the consumerd socket */
1334 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1335 && !(revents & LPOLLIN)) {
1336 ERR("consumer err socket second poll error");
1337 goto error;
1338 }
1339 health_code_update();
1340 /* Wait for any kconsumerd error */
1341 ret = lttcomm_recv_unix_sock(sock, &code,
1342 sizeof(enum lttcomm_return_code));
1343 if (ret <= 0) {
1344 ERR("consumer closed the command socket");
1345 goto error;
1346 }
1347
1348 ERR("consumer return code : %s",
1349 lttcomm_get_readable_code(-code));
1350
1351 goto exit;
1352 } else if (pollfd == consumer_data->metadata_fd) {
1353 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1354 && !(revents & LPOLLIN)) {
1355 ERR("consumer err metadata socket second poll error");
1356 goto error;
1357 }
1358 /* UST metadata requests */
1359 ret = ust_consumer_metadata_request(
1360 &consumer_data->metadata_sock);
1361 if (ret < 0) {
1362 ERR("Handling metadata request");
1363 goto error;
1364 }
1365 }
1366 /* No need for an else branch all FDs are tested prior. */
1367 }
1368 health_code_update();
1369 }
1370
1371 exit:
1372 error:
1373 /*
1374 * We lock here because we are about to close the sockets and some other
1375 * thread might be using them so get exclusive access which will abort all
1376 * other consumer command by other threads.
1377 */
1378 pthread_mutex_lock(&consumer_data->lock);
1379
1380 /* Immediately set the consumerd state to stopped */
1381 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1382 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1383 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1384 consumer_data->type == LTTNG_CONSUMER32_UST) {
1385 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1386 } else {
1387 /* Code flow error... */
1388 assert(0);
1389 }
1390
1391 if (consumer_data->err_sock >= 0) {
1392 ret = close(consumer_data->err_sock);
1393 if (ret) {
1394 PERROR("close");
1395 }
1396 consumer_data->err_sock = -1;
1397 }
1398 if (consumer_data->cmd_sock >= 0) {
1399 ret = close(consumer_data->cmd_sock);
1400 if (ret) {
1401 PERROR("close");
1402 }
1403 consumer_data->cmd_sock = -1;
1404 }
1405 if (consumer_data->metadata_sock.fd_ptr &&
1406 *consumer_data->metadata_sock.fd_ptr >= 0) {
1407 ret = close(*consumer_data->metadata_sock.fd_ptr);
1408 if (ret) {
1409 PERROR("close");
1410 }
1411 }
1412 if (sock >= 0) {
1413 ret = close(sock);
1414 if (ret) {
1415 PERROR("close");
1416 }
1417 }
1418
1419 unlink(consumer_data->err_unix_sock_path);
1420 unlink(consumer_data->cmd_unix_sock_path);
1421 pthread_mutex_unlock(&consumer_data->lock);
1422
1423 /* Cleanup metadata socket mutex. */
1424 if (consumer_data->metadata_sock.lock) {
1425 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1426 free(consumer_data->metadata_sock.lock);
1427 }
1428 lttng_poll_clean(&events);
1429
1430 if (cmd_socket_wrapper) {
1431 consumer_destroy_socket(cmd_socket_wrapper);
1432 }
1433 error_poll:
1434 if (err) {
1435 health_error();
1436 ERR("Health error occurred in %s", __func__);
1437 }
1438 health_unregister(health_sessiond);
1439 DBG("consumer thread cleanup completed");
1440
1441 rcu_thread_offline();
1442 rcu_unregister_thread();
1443
1444 return NULL;
1445 }
1446
1447 /*
1448 * This thread manage application communication.
1449 */
1450 static void *thread_manage_apps(void *data)
1451 {
1452 int i, ret, pollfd, err = -1;
1453 ssize_t size_ret;
1454 uint32_t revents, nb_fd;
1455 struct lttng_poll_event events;
1456
1457 DBG("[thread] Manage application started");
1458
1459 rcu_register_thread();
1460 rcu_thread_online();
1461
1462 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1463
1464 if (testpoint(sessiond_thread_manage_apps)) {
1465 goto error_testpoint;
1466 }
1467
1468 health_code_update();
1469
1470 ret = sessiond_set_thread_pollset(&events, 2);
1471 if (ret < 0) {
1472 goto error_poll_create;
1473 }
1474
1475 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1476 if (ret < 0) {
1477 goto error;
1478 }
1479
1480 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1481 goto error;
1482 }
1483
1484 health_code_update();
1485
1486 while (1) {
1487 DBG("Apps thread polling");
1488
1489 /* Inifinite blocking call, waiting for transmission */
1490 restart:
1491 health_poll_entry();
1492 ret = lttng_poll_wait(&events, -1);
1493 DBG("Apps thread return from poll on %d fds",
1494 LTTNG_POLL_GETNB(&events));
1495 health_poll_exit();
1496 if (ret < 0) {
1497 /*
1498 * Restart interrupted system call.
1499 */
1500 if (errno == EINTR) {
1501 goto restart;
1502 }
1503 goto error;
1504 }
1505
1506 nb_fd = ret;
1507
1508 for (i = 0; i < nb_fd; i++) {
1509 /* Fetch once the poll data */
1510 revents = LTTNG_POLL_GETEV(&events, i);
1511 pollfd = LTTNG_POLL_GETFD(&events, i);
1512
1513 health_code_update();
1514
1515 if (!revents) {
1516 /* No activity for this FD (poll implementation). */
1517 continue;
1518 }
1519
1520 /* Thread quit pipe has been closed. Killing thread. */
1521 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1522 if (ret) {
1523 err = 0;
1524 goto exit;
1525 }
1526
1527 /* Inspect the apps cmd pipe */
1528 if (pollfd == apps_cmd_pipe[0]) {
1529 if (revents & LPOLLIN) {
1530 int sock;
1531
1532 /* Empty pipe */
1533 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1534 if (size_ret < sizeof(sock)) {
1535 PERROR("read apps cmd pipe");
1536 goto error;
1537 }
1538
1539 health_code_update();
1540
1541 /*
1542 * Since this is a command socket (write then read),
1543 * we only monitor the error events of the socket.
1544 */
1545 ret = lttng_poll_add(&events, sock,
1546 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1547 if (ret < 0) {
1548 goto error;
1549 }
1550
1551 DBG("Apps with sock %d added to poll set", sock);
1552 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1553 ERR("Apps command pipe error");
1554 goto error;
1555 } else {
1556 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1557 goto error;
1558 }
1559 } else {
1560 /*
1561 * At this point, we know that a registered application made
1562 * the event at poll_wait.
1563 */
1564 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1565 /* Removing from the poll set */
1566 ret = lttng_poll_del(&events, pollfd);
1567 if (ret < 0) {
1568 goto error;
1569 }
1570
1571 /* Socket closed on remote end. */
1572 ust_app_unregister(pollfd);
1573 } else {
1574 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1575 goto error;
1576 }
1577 }
1578
1579 health_code_update();
1580 }
1581 }
1582
1583 exit:
1584 error:
1585 lttng_poll_clean(&events);
1586 error_poll_create:
1587 error_testpoint:
1588 utils_close_pipe(apps_cmd_pipe);
1589 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1590
1591 /*
1592 * We don't clean the UST app hash table here since already registered
1593 * applications can still be controlled so let them be until the session
1594 * daemon dies or the applications stop.
1595 */
1596
1597 if (err) {
1598 health_error();
1599 ERR("Health error occurred in %s", __func__);
1600 }
1601 health_unregister(health_sessiond);
1602 DBG("Application communication apps thread cleanup complete");
1603 rcu_thread_offline();
1604 rcu_unregister_thread();
1605 return NULL;
1606 }
1607
1608 /*
1609 * Send a socket to a thread This is called from the dispatch UST registration
1610 * thread once all sockets are set for the application.
1611 *
1612 * The sock value can be invalid, we don't really care, the thread will handle
1613 * it and make the necessary cleanup if so.
1614 *
1615 * On success, return 0 else a negative value being the errno message of the
1616 * write().
1617 */
1618 static int send_socket_to_thread(int fd, int sock)
1619 {
1620 ssize_t ret;
1621
1622 /*
1623 * It's possible that the FD is set as invalid with -1 concurrently just
1624 * before calling this function being a shutdown state of the thread.
1625 */
1626 if (fd < 0) {
1627 ret = -EBADF;
1628 goto error;
1629 }
1630
1631 ret = lttng_write(fd, &sock, sizeof(sock));
1632 if (ret < sizeof(sock)) {
1633 PERROR("write apps pipe %d", fd);
1634 if (ret < 0) {
1635 ret = -errno;
1636 }
1637 goto error;
1638 }
1639
1640 /* All good. Don't send back the write positive ret value. */
1641 ret = 0;
1642 error:
1643 return (int) ret;
1644 }
1645
1646 /*
1647 * Sanitize the wait queue of the dispatch registration thread meaning removing
1648 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1649 * notify socket is never received.
1650 */
1651 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1652 {
1653 int ret, nb_fd = 0, i;
1654 unsigned int fd_added = 0;
1655 struct lttng_poll_event events;
1656 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1657
1658 assert(wait_queue);
1659
1660 lttng_poll_init(&events);
1661
1662 /* Just skip everything for an empty queue. */
1663 if (!wait_queue->count) {
1664 goto end;
1665 }
1666
1667 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1668 if (ret < 0) {
1669 goto error_create;
1670 }
1671
1672 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1673 &wait_queue->head, head) {
1674 assert(wait_node->app);
1675 ret = lttng_poll_add(&events, wait_node->app->sock,
1676 LPOLLHUP | LPOLLERR);
1677 if (ret < 0) {
1678 goto error;
1679 }
1680
1681 fd_added = 1;
1682 }
1683
1684 if (!fd_added) {
1685 goto end;
1686 }
1687
1688 /*
1689 * Poll but don't block so we can quickly identify the faulty events and
1690 * clean them afterwards from the wait queue.
1691 */
1692 ret = lttng_poll_wait(&events, 0);
1693 if (ret < 0) {
1694 goto error;
1695 }
1696 nb_fd = ret;
1697
1698 for (i = 0; i < nb_fd; i++) {
1699 /* Get faulty FD. */
1700 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1701 int pollfd = LTTNG_POLL_GETFD(&events, i);
1702
1703 if (!revents) {
1704 /* No activity for this FD (poll implementation). */
1705 continue;
1706 }
1707
1708 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1709 &wait_queue->head, head) {
1710 if (pollfd == wait_node->app->sock &&
1711 (revents & (LPOLLHUP | LPOLLERR))) {
1712 cds_list_del(&wait_node->head);
1713 wait_queue->count--;
1714 ust_app_destroy(wait_node->app);
1715 free(wait_node);
1716 /*
1717 * Silence warning of use-after-free in
1718 * cds_list_for_each_entry_safe which uses
1719 * __typeof__(*wait_node).
1720 */
1721 wait_node = NULL;
1722 break;
1723 } else {
1724 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1725 goto error;
1726 }
1727 }
1728 }
1729
1730 if (nb_fd > 0) {
1731 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1732 }
1733
1734 end:
1735 lttng_poll_clean(&events);
1736 return;
1737
1738 error:
1739 lttng_poll_clean(&events);
1740 error_create:
1741 ERR("Unable to sanitize wait queue");
1742 return;
1743 }
1744
1745 /*
1746 * Dispatch request from the registration threads to the application
1747 * communication thread.
1748 */
1749 static void *thread_dispatch_ust_registration(void *data)
1750 {
1751 int ret, err = -1;
1752 struct cds_wfcq_node *node;
1753 struct ust_command *ust_cmd = NULL;
1754 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1755 struct ust_reg_wait_queue wait_queue = {
1756 .count = 0,
1757 };
1758
1759 rcu_register_thread();
1760
1761 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1762
1763 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1764 goto error_testpoint;
1765 }
1766
1767 health_code_update();
1768
1769 CDS_INIT_LIST_HEAD(&wait_queue.head);
1770
1771 DBG("[thread] Dispatch UST command started");
1772
1773 for (;;) {
1774 health_code_update();
1775
1776 /* Atomically prepare the queue futex */
1777 futex_nto1_prepare(&ust_cmd_queue.futex);
1778
1779 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1780 break;
1781 }
1782
1783 do {
1784 struct ust_app *app = NULL;
1785 ust_cmd = NULL;
1786
1787 /*
1788 * Make sure we don't have node(s) that have hung up before receiving
1789 * the notify socket. This is to clean the list in order to avoid
1790 * memory leaks from notify socket that are never seen.
1791 */
1792 sanitize_wait_queue(&wait_queue);
1793
1794 health_code_update();
1795 /* Dequeue command for registration */
1796 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1797 if (node == NULL) {
1798 DBG("Woken up but nothing in the UST command queue");
1799 /* Continue thread execution */
1800 break;
1801 }
1802
1803 ust_cmd = caa_container_of(node, struct ust_command, node);
1804
1805 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1806 " gid:%d sock:%d name:%s (version %d.%d)",
1807 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1808 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1809 ust_cmd->sock, ust_cmd->reg_msg.name,
1810 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1811
1812 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1813 wait_node = zmalloc(sizeof(*wait_node));
1814 if (!wait_node) {
1815 PERROR("zmalloc wait_node dispatch");
1816 ret = close(ust_cmd->sock);
1817 if (ret < 0) {
1818 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1819 }
1820 lttng_fd_put(LTTNG_FD_APPS, 1);
1821 free(ust_cmd);
1822 goto error;
1823 }
1824 CDS_INIT_LIST_HEAD(&wait_node->head);
1825
1826 /* Create application object if socket is CMD. */
1827 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1828 ust_cmd->sock);
1829 if (!wait_node->app) {
1830 ret = close(ust_cmd->sock);
1831 if (ret < 0) {
1832 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1833 }
1834 lttng_fd_put(LTTNG_FD_APPS, 1);
1835 free(wait_node);
1836 free(ust_cmd);
1837 continue;
1838 }
1839 /*
1840 * Add application to the wait queue so we can set the notify
1841 * socket before putting this object in the global ht.
1842 */
1843 cds_list_add(&wait_node->head, &wait_queue.head);
1844 wait_queue.count++;
1845
1846 free(ust_cmd);
1847 /*
1848 * We have to continue here since we don't have the notify
1849 * socket and the application MUST be added to the hash table
1850 * only at that moment.
1851 */
1852 continue;
1853 } else {
1854 /*
1855 * Look for the application in the local wait queue and set the
1856 * notify socket if found.
1857 */
1858 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1859 &wait_queue.head, head) {
1860 health_code_update();
1861 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1862 wait_node->app->notify_sock = ust_cmd->sock;
1863 cds_list_del(&wait_node->head);
1864 wait_queue.count--;
1865 app = wait_node->app;
1866 free(wait_node);
1867 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1868 break;
1869 }
1870 }
1871
1872 /*
1873 * With no application at this stage the received socket is
1874 * basically useless so close it before we free the cmd data
1875 * structure for good.
1876 */
1877 if (!app) {
1878 ret = close(ust_cmd->sock);
1879 if (ret < 0) {
1880 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1881 }
1882 lttng_fd_put(LTTNG_FD_APPS, 1);
1883 }
1884 free(ust_cmd);
1885 }
1886
1887 if (app) {
1888 /*
1889 * @session_lock_list
1890 *
1891 * Lock the global session list so from the register up to the
1892 * registration done message, no thread can see the application
1893 * and change its state.
1894 */
1895 session_lock_list();
1896 rcu_read_lock();
1897
1898 /*
1899 * Add application to the global hash table. This needs to be
1900 * done before the update to the UST registry can locate the
1901 * application.
1902 */
1903 ust_app_add(app);
1904
1905 /* Set app version. This call will print an error if needed. */
1906 (void) ust_app_version(app);
1907
1908 /* Send notify socket through the notify pipe. */
1909 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1910 app->notify_sock);
1911 if (ret < 0) {
1912 rcu_read_unlock();
1913 session_unlock_list();
1914 /*
1915 * No notify thread, stop the UST tracing. However, this is
1916 * not an internal error of the this thread thus setting
1917 * the health error code to a normal exit.
1918 */
1919 err = 0;
1920 goto error;
1921 }
1922
1923 /*
1924 * Update newly registered application with the tracing
1925 * registry info already enabled information.
1926 */
1927 update_ust_app(app->sock);
1928
1929 /*
1930 * Don't care about return value. Let the manage apps threads
1931 * handle app unregistration upon socket close.
1932 */
1933 (void) ust_app_register_done(app);
1934
1935 /*
1936 * Even if the application socket has been closed, send the app
1937 * to the thread and unregistration will take place at that
1938 * place.
1939 */
1940 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1941 if (ret < 0) {
1942 rcu_read_unlock();
1943 session_unlock_list();
1944 /*
1945 * No apps. thread, stop the UST tracing. However, this is
1946 * not an internal error of the this thread thus setting
1947 * the health error code to a normal exit.
1948 */
1949 err = 0;
1950 goto error;
1951 }
1952
1953 rcu_read_unlock();
1954 session_unlock_list();
1955 }
1956 } while (node != NULL);
1957
1958 health_poll_entry();
1959 /* Futex wait on queue. Blocking call on futex() */
1960 futex_nto1_wait(&ust_cmd_queue.futex);
1961 health_poll_exit();
1962 }
1963 /* Normal exit, no error */
1964 err = 0;
1965
1966 error:
1967 /* Clean up wait queue. */
1968 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1969 &wait_queue.head, head) {
1970 cds_list_del(&wait_node->head);
1971 wait_queue.count--;
1972 free(wait_node);
1973 }
1974
1975 /* Empty command queue. */
1976 for (;;) {
1977 /* Dequeue command for registration */
1978 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1979 if (node == NULL) {
1980 break;
1981 }
1982 ust_cmd = caa_container_of(node, struct ust_command, node);
1983 ret = close(ust_cmd->sock);
1984 if (ret < 0) {
1985 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
1986 }
1987 lttng_fd_put(LTTNG_FD_APPS, 1);
1988 free(ust_cmd);
1989 }
1990
1991 error_testpoint:
1992 DBG("Dispatch thread dying");
1993 if (err) {
1994 health_error();
1995 ERR("Health error occurred in %s", __func__);
1996 }
1997 health_unregister(health_sessiond);
1998 rcu_unregister_thread();
1999 return NULL;
2000 }
2001
2002 /*
2003 * This thread manage application registration.
2004 */
2005 static void *thread_registration_apps(void *data)
2006 {
2007 int sock = -1, i, ret, pollfd, err = -1;
2008 uint32_t revents, nb_fd;
2009 struct lttng_poll_event events;
2010 /*
2011 * Get allocated in this thread, enqueued to a global queue, dequeued and
2012 * freed in the manage apps thread.
2013 */
2014 struct ust_command *ust_cmd = NULL;
2015
2016 DBG("[thread] Manage application registration started");
2017
2018 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2019
2020 if (testpoint(sessiond_thread_registration_apps)) {
2021 goto error_testpoint;
2022 }
2023
2024 ret = lttcomm_listen_unix_sock(apps_sock);
2025 if (ret < 0) {
2026 goto error_listen;
2027 }
2028
2029 /*
2030 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2031 * more will be added to this poll set.
2032 */
2033 ret = sessiond_set_thread_pollset(&events, 2);
2034 if (ret < 0) {
2035 goto error_create_poll;
2036 }
2037
2038 /* Add the application registration socket */
2039 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2040 if (ret < 0) {
2041 goto error_poll_add;
2042 }
2043
2044 /* Notify all applications to register */
2045 ret = notify_ust_apps(1);
2046 if (ret < 0) {
2047 ERR("Failed to notify applications or create the wait shared memory.\n"
2048 "Execution continues but there might be problem for already\n"
2049 "running applications that wishes to register.");
2050 }
2051
2052 while (1) {
2053 DBG("Accepting application registration");
2054
2055 /* Inifinite blocking call, waiting for transmission */
2056 restart:
2057 health_poll_entry();
2058 ret = lttng_poll_wait(&events, -1);
2059 health_poll_exit();
2060 if (ret < 0) {
2061 /*
2062 * Restart interrupted system call.
2063 */
2064 if (errno == EINTR) {
2065 goto restart;
2066 }
2067 goto error;
2068 }
2069
2070 nb_fd = ret;
2071
2072 for (i = 0; i < nb_fd; i++) {
2073 health_code_update();
2074
2075 /* Fetch once the poll data */
2076 revents = LTTNG_POLL_GETEV(&events, i);
2077 pollfd = LTTNG_POLL_GETFD(&events, i);
2078
2079 if (!revents) {
2080 /* No activity for this FD (poll implementation). */
2081 continue;
2082 }
2083
2084 /* Thread quit pipe has been closed. Killing thread. */
2085 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2086 if (ret) {
2087 err = 0;
2088 goto exit;
2089 }
2090
2091 /* Event on the registration socket */
2092 if (pollfd == apps_sock) {
2093 if (revents & LPOLLIN) {
2094 sock = lttcomm_accept_unix_sock(apps_sock);
2095 if (sock < 0) {
2096 goto error;
2097 }
2098
2099 /*
2100 * Set socket timeout for both receiving and ending.
2101 * app_socket_timeout is in seconds, whereas
2102 * lttcomm_setsockopt_rcv_timeout and
2103 * lttcomm_setsockopt_snd_timeout expect msec as
2104 * parameter.
2105 */
2106 if (config.app_socket_timeout >= 0) {
2107 (void) lttcomm_setsockopt_rcv_timeout(sock,
2108 config.app_socket_timeout * 1000);
2109 (void) lttcomm_setsockopt_snd_timeout(sock,
2110 config.app_socket_timeout * 1000);
2111 }
2112
2113 /*
2114 * Set the CLOEXEC flag. Return code is useless because
2115 * either way, the show must go on.
2116 */
2117 (void) utils_set_fd_cloexec(sock);
2118
2119 /* Create UST registration command for enqueuing */
2120 ust_cmd = zmalloc(sizeof(struct ust_command));
2121 if (ust_cmd == NULL) {
2122 PERROR("ust command zmalloc");
2123 ret = close(sock);
2124 if (ret) {
2125 PERROR("close");
2126 }
2127 goto error;
2128 }
2129
2130 /*
2131 * Using message-based transmissions to ensure we don't
2132 * have to deal with partially received messages.
2133 */
2134 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2135 if (ret < 0) {
2136 ERR("Exhausted file descriptors allowed for applications.");
2137 free(ust_cmd);
2138 ret = close(sock);
2139 if (ret) {
2140 PERROR("close");
2141 }
2142 sock = -1;
2143 continue;
2144 }
2145
2146 health_code_update();
2147 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2148 if (ret < 0) {
2149 free(ust_cmd);
2150 /* Close socket of the application. */
2151 ret = close(sock);
2152 if (ret) {
2153 PERROR("close");
2154 }
2155 lttng_fd_put(LTTNG_FD_APPS, 1);
2156 sock = -1;
2157 continue;
2158 }
2159 health_code_update();
2160
2161 ust_cmd->sock = sock;
2162 sock = -1;
2163
2164 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2165 " gid:%d sock:%d name:%s (version %d.%d)",
2166 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2167 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2168 ust_cmd->sock, ust_cmd->reg_msg.name,
2169 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2170
2171 /*
2172 * Lock free enqueue the registration request. The red pill
2173 * has been taken! This apps will be part of the *system*.
2174 */
2175 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2176
2177 /*
2178 * Wake the registration queue futex. Implicit memory
2179 * barrier with the exchange in cds_wfcq_enqueue.
2180 */
2181 futex_nto1_wake(&ust_cmd_queue.futex);
2182 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2183 ERR("Register apps socket poll error");
2184 goto error;
2185 } else {
2186 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2187 goto error;
2188 }
2189 }
2190 }
2191 }
2192
2193 exit:
2194 error:
2195 /* Notify that the registration thread is gone */
2196 notify_ust_apps(0);
2197
2198 if (apps_sock >= 0) {
2199 ret = close(apps_sock);
2200 if (ret) {
2201 PERROR("close");
2202 }
2203 }
2204 if (sock >= 0) {
2205 ret = close(sock);
2206 if (ret) {
2207 PERROR("close");
2208 }
2209 lttng_fd_put(LTTNG_FD_APPS, 1);
2210 }
2211 unlink(config.apps_unix_sock_path.value);
2212
2213 error_poll_add:
2214 lttng_poll_clean(&events);
2215 error_listen:
2216 error_create_poll:
2217 error_testpoint:
2218 DBG("UST Registration thread cleanup complete");
2219 if (err) {
2220 health_error();
2221 ERR("Health error occurred in %s", __func__);
2222 }
2223 health_unregister(health_sessiond);
2224
2225 return NULL;
2226 }
2227
2228 /*
2229 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2230 * exec or it will fails.
2231 */
2232 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2233 {
2234 int ret, clock_ret;
2235 struct timespec timeout;
2236
2237 /*
2238 * Make sure we set the readiness flag to 0 because we are NOT ready.
2239 * This access to consumer_thread_is_ready does not need to be
2240 * protected by consumer_data.cond_mutex (yet) since the consumer
2241 * management thread has not been started at this point.
2242 */
2243 consumer_data->consumer_thread_is_ready = 0;
2244
2245 /* Setup pthread condition */
2246 ret = pthread_condattr_init(&consumer_data->condattr);
2247 if (ret) {
2248 errno = ret;
2249 PERROR("pthread_condattr_init consumer data");
2250 goto error;
2251 }
2252
2253 /*
2254 * Set the monotonic clock in order to make sure we DO NOT jump in time
2255 * between the clock_gettime() call and the timedwait call. See bug #324
2256 * for a more details and how we noticed it.
2257 */
2258 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2259 if (ret) {
2260 errno = ret;
2261 PERROR("pthread_condattr_setclock consumer data");
2262 goto error;
2263 }
2264
2265 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2266 if (ret) {
2267 errno = ret;
2268 PERROR("pthread_cond_init consumer data");
2269 goto error;
2270 }
2271
2272 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2273 thread_manage_consumer, consumer_data);
2274 if (ret) {
2275 errno = ret;
2276 PERROR("pthread_create consumer");
2277 ret = -1;
2278 goto error;
2279 }
2280
2281 /* We are about to wait on a pthread condition */
2282 pthread_mutex_lock(&consumer_data->cond_mutex);
2283
2284 /* Get time for sem_timedwait absolute timeout */
2285 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2286 /*
2287 * Set the timeout for the condition timed wait even if the clock gettime
2288 * call fails since we might loop on that call and we want to avoid to
2289 * increment the timeout too many times.
2290 */
2291 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2292
2293 /*
2294 * The following loop COULD be skipped in some conditions so this is why we
2295 * set ret to 0 in order to make sure at least one round of the loop is
2296 * done.
2297 */
2298 ret = 0;
2299
2300 /*
2301 * Loop until the condition is reached or when a timeout is reached. Note
2302 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2303 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2304 * possible. This loop does not take any chances and works with both of
2305 * them.
2306 */
2307 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2308 if (clock_ret < 0) {
2309 PERROR("clock_gettime spawn consumer");
2310 /* Infinite wait for the consumerd thread to be ready */
2311 ret = pthread_cond_wait(&consumer_data->cond,
2312 &consumer_data->cond_mutex);
2313 } else {
2314 ret = pthread_cond_timedwait(&consumer_data->cond,
2315 &consumer_data->cond_mutex, &timeout);
2316 }
2317 }
2318
2319 /* Release the pthread condition */
2320 pthread_mutex_unlock(&consumer_data->cond_mutex);
2321
2322 if (ret != 0) {
2323 errno = ret;
2324 if (ret == ETIMEDOUT) {
2325 int pth_ret;
2326
2327 /*
2328 * Call has timed out so we kill the kconsumerd_thread and return
2329 * an error.
2330 */
2331 ERR("Condition timed out. The consumer thread was never ready."
2332 " Killing it");
2333 pth_ret = pthread_cancel(consumer_data->thread);
2334 if (pth_ret < 0) {
2335 PERROR("pthread_cancel consumer thread");
2336 }
2337 } else {
2338 PERROR("pthread_cond_wait failed consumer thread");
2339 }
2340 /* Caller is expecting a negative value on failure. */
2341 ret = -1;
2342 goto error;
2343 }
2344
2345 pthread_mutex_lock(&consumer_data->pid_mutex);
2346 if (consumer_data->pid == 0) {
2347 ERR("Consumerd did not start");
2348 pthread_mutex_unlock(&consumer_data->pid_mutex);
2349 goto error;
2350 }
2351 pthread_mutex_unlock(&consumer_data->pid_mutex);
2352
2353 return 0;
2354
2355 error:
2356 return ret;
2357 }
2358
2359 /*
2360 * Join consumer thread
2361 */
2362 static int join_consumer_thread(struct consumer_data *consumer_data)
2363 {
2364 void *status;
2365
2366 /* Consumer pid must be a real one. */
2367 if (consumer_data->pid > 0) {
2368 int ret;
2369 ret = kill(consumer_data->pid, SIGTERM);
2370 if (ret) {
2371 PERROR("Error killing consumer daemon");
2372 return ret;
2373 }
2374 return pthread_join(consumer_data->thread, &status);
2375 } else {
2376 return 0;
2377 }
2378 }
2379
2380 /*
2381 * Fork and exec a consumer daemon (consumerd).
2382 *
2383 * Return pid if successful else -1.
2384 */
2385 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2386 {
2387 int ret;
2388 pid_t pid;
2389 const char *consumer_to_use;
2390 const char *verbosity;
2391 struct stat st;
2392
2393 DBG("Spawning consumerd");
2394
2395 pid = fork();
2396 if (pid == 0) {
2397 /*
2398 * Exec consumerd.
2399 */
2400 if (config.verbose_consumer) {
2401 verbosity = "--verbose";
2402 } else if (lttng_opt_quiet) {
2403 verbosity = "--quiet";
2404 } else {
2405 verbosity = "";
2406 }
2407
2408 switch (consumer_data->type) {
2409 case LTTNG_CONSUMER_KERNEL:
2410 /*
2411 * Find out which consumerd to execute. We will first try the
2412 * 64-bit path, then the sessiond's installation directory, and
2413 * fallback on the 32-bit one,
2414 */
2415 DBG3("Looking for a kernel consumer at these locations:");
2416 DBG3(" 1) %s", config.consumerd64_bin_path.value ? : "NULL");
2417 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
2418 DBG3(" 3) %s", config.consumerd32_bin_path.value ? : "NULL");
2419 if (stat(config.consumerd64_bin_path.value, &st) == 0) {
2420 DBG3("Found location #1");
2421 consumer_to_use = config.consumerd64_bin_path.value;
2422 } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
2423 DBG3("Found location #2");
2424 consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
2425 } else if (stat(config.consumerd32_bin_path.value, &st) == 0) {
2426 DBG3("Found location #3");
2427 consumer_to_use = config.consumerd32_bin_path.value;
2428 } else {
2429 DBG("Could not find any valid consumerd executable");
2430 ret = -EINVAL;
2431 goto error;
2432 }
2433 DBG("Using kernel consumer at: %s", consumer_to_use);
2434 (void) execl(consumer_to_use,
2435 "lttng-consumerd", verbosity, "-k",
2436 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2437 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2438 "--group", config.tracing_group_name.value,
2439 NULL);
2440 break;
2441 case LTTNG_CONSUMER64_UST:
2442 {
2443 if (config.consumerd64_lib_dir.value) {
2444 char *tmp;
2445 size_t tmplen;
2446 char *tmpnew;
2447
2448 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2449 if (!tmp) {
2450 tmp = "";
2451 }
2452 tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2453 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2454 if (!tmpnew) {
2455 ret = -ENOMEM;
2456 goto error;
2457 }
2458 strcat(tmpnew, config.consumerd64_lib_dir.value);
2459 if (tmp[0] != '\0') {
2460 strcat(tmpnew, ":");
2461 strcat(tmpnew, tmp);
2462 }
2463 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2464 free(tmpnew);
2465 if (ret) {
2466 ret = -errno;
2467 goto error;
2468 }
2469 }
2470 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2471 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2472 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2473 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2474 "--group", config.tracing_group_name.value,
2475 NULL);
2476 break;
2477 }
2478 case LTTNG_CONSUMER32_UST:
2479 {
2480 if (config.consumerd32_lib_dir.value) {
2481 char *tmp;
2482 size_t tmplen;
2483 char *tmpnew;
2484
2485 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2486 if (!tmp) {
2487 tmp = "";
2488 }
2489 tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2490 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2491 if (!tmpnew) {
2492 ret = -ENOMEM;
2493 goto error;
2494 }
2495 strcat(tmpnew, config.consumerd32_lib_dir.value);
2496 if (tmp[0] != '\0') {
2497 strcat(tmpnew, ":");
2498 strcat(tmpnew, tmp);
2499 }
2500 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2501 free(tmpnew);
2502 if (ret) {
2503 ret = -errno;
2504 goto error;
2505 }
2506 }
2507 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2508 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2509 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2510 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2511 "--group", config.tracing_group_name.value,
2512 NULL);
2513 break;
2514 }
2515 default:
2516 ERR("unknown consumer type");
2517 errno = 0;
2518 }
2519 if (errno != 0) {
2520 PERROR("Consumer execl()");
2521 }
2522 /* Reaching this point, we got a failure on our execl(). */
2523 exit(EXIT_FAILURE);
2524 } else if (pid > 0) {
2525 ret = pid;
2526 } else {
2527 PERROR("start consumer fork");
2528 ret = -errno;
2529 }
2530 error:
2531 return ret;
2532 }
2533
2534 /*
2535 * Spawn the consumerd daemon and session daemon thread.
2536 */
2537 static int start_consumerd(struct consumer_data *consumer_data)
2538 {
2539 int ret;
2540
2541 /*
2542 * Set the listen() state on the socket since there is a possible race
2543 * between the exec() of the consumer daemon and this call if place in the
2544 * consumer thread. See bug #366 for more details.
2545 */
2546 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2547 if (ret < 0) {
2548 goto error;
2549 }
2550
2551 pthread_mutex_lock(&consumer_data->pid_mutex);
2552 if (consumer_data->pid != 0) {
2553 pthread_mutex_unlock(&consumer_data->pid_mutex);
2554 goto end;
2555 }
2556
2557 ret = spawn_consumerd(consumer_data);
2558 if (ret < 0) {
2559 ERR("Spawning consumerd failed");
2560 pthread_mutex_unlock(&consumer_data->pid_mutex);
2561 goto error;
2562 }
2563
2564 /* Setting up the consumer_data pid */
2565 consumer_data->pid = ret;
2566 DBG2("Consumer pid %d", consumer_data->pid);
2567 pthread_mutex_unlock(&consumer_data->pid_mutex);
2568
2569 DBG2("Spawning consumer control thread");
2570 ret = spawn_consumer_thread(consumer_data);
2571 if (ret < 0) {
2572 ERR("Fatal error spawning consumer control thread");
2573 goto error;
2574 }
2575
2576 end:
2577 return 0;
2578
2579 error:
2580 /* Cleanup already created sockets on error. */
2581 if (consumer_data->err_sock >= 0) {
2582 int err;
2583
2584 err = close(consumer_data->err_sock);
2585 if (err < 0) {
2586 PERROR("close consumer data error socket");
2587 }
2588 }
2589 return ret;
2590 }
2591
2592 /*
2593 * Setup necessary data for kernel tracer action.
2594 */
2595 static int init_kernel_tracer(void)
2596 {
2597 int ret;
2598
2599 /* Modprobe lttng kernel modules */
2600 ret = modprobe_lttng_control();
2601 if (ret < 0) {
2602 goto error;
2603 }
2604
2605 /* Open debugfs lttng */
2606 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2607 if (kernel_tracer_fd < 0) {
2608 DBG("Failed to open %s", module_proc_lttng);
2609 goto error_open;
2610 }
2611
2612 /* Validate kernel version */
2613 ret = kernel_validate_version(kernel_tracer_fd);
2614 if (ret < 0) {
2615 goto error_version;
2616 }
2617
2618 ret = modprobe_lttng_data();
2619 if (ret < 0) {
2620 goto error_modules;
2621 }
2622
2623 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2624 kernel_tracer_fd);
2625 if (ret < 0) {
2626 goto error_modules;
2627 }
2628
2629 if (ret < 1) {
2630 WARN("Kernel tracer does not support buffer monitoring. "
2631 "The monitoring timer of channels in the kernel domain "
2632 "will be set to 0 (disabled).");
2633 }
2634
2635 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2636 return 0;
2637
2638 error_version:
2639 modprobe_remove_lttng_control();
2640 ret = close(kernel_tracer_fd);
2641 if (ret) {
2642 PERROR("close");
2643 }
2644 kernel_tracer_fd = -1;
2645 return LTTNG_ERR_KERN_VERSION;
2646
2647 error_modules:
2648 ret = close(kernel_tracer_fd);
2649 if (ret) {
2650 PERROR("close");
2651 }
2652
2653 error_open:
2654 modprobe_remove_lttng_control();
2655
2656 error:
2657 WARN("No kernel tracer available");
2658 kernel_tracer_fd = -1;
2659 if (!is_root) {
2660 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2661 } else {
2662 return LTTNG_ERR_KERN_NA;
2663 }
2664 }
2665
2666
2667 /*
2668 * Copy consumer output from the tracing session to the domain session. The
2669 * function also applies the right modification on a per domain basis for the
2670 * trace files destination directory.
2671 *
2672 * Should *NOT* be called with RCU read-side lock held.
2673 */
2674 static int copy_session_consumer(int domain, struct ltt_session *session)
2675 {
2676 int ret;
2677 const char *dir_name;
2678 struct consumer_output *consumer;
2679
2680 assert(session);
2681 assert(session->consumer);
2682
2683 switch (domain) {
2684 case LTTNG_DOMAIN_KERNEL:
2685 DBG3("Copying tracing session consumer output in kernel session");
2686 /*
2687 * XXX: We should audit the session creation and what this function
2688 * does "extra" in order to avoid a destroy since this function is used
2689 * in the domain session creation (kernel and ust) only. Same for UST
2690 * domain.
2691 */
2692 if (session->kernel_session->consumer) {
2693 consumer_output_put(session->kernel_session->consumer);
2694 }
2695 session->kernel_session->consumer =
2696 consumer_copy_output(session->consumer);
2697 /* Ease our life a bit for the next part */
2698 consumer = session->kernel_session->consumer;
2699 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2700 break;
2701 case LTTNG_DOMAIN_JUL:
2702 case LTTNG_DOMAIN_LOG4J:
2703 case LTTNG_DOMAIN_PYTHON:
2704 case LTTNG_DOMAIN_UST:
2705 DBG3("Copying tracing session consumer output in UST session");
2706 if (session->ust_session->consumer) {
2707 consumer_output_put(session->ust_session->consumer);
2708 }
2709 session->ust_session->consumer =
2710 consumer_copy_output(session->consumer);
2711 /* Ease our life a bit for the next part */
2712 consumer = session->ust_session->consumer;
2713 dir_name = DEFAULT_UST_TRACE_DIR;
2714 break;
2715 default:
2716 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2717 goto error;
2718 }
2719
2720 /* Append correct directory to subdir */
2721 strncat(consumer->subdir, dir_name,
2722 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2723 DBG3("Copy session consumer subdir %s", consumer->subdir);
2724
2725 ret = LTTNG_OK;
2726
2727 error:
2728 return ret;
2729 }
2730
2731 /*
2732 * Create an UST session and add it to the session ust list.
2733 *
2734 * Should *NOT* be called with RCU read-side lock held.
2735 */
2736 static int create_ust_session(struct ltt_session *session,
2737 struct lttng_domain *domain)
2738 {
2739 int ret;
2740 struct ltt_ust_session *lus = NULL;
2741
2742 assert(session);
2743 assert(domain);
2744 assert(session->consumer);
2745
2746 switch (domain->type) {
2747 case LTTNG_DOMAIN_JUL:
2748 case LTTNG_DOMAIN_LOG4J:
2749 case LTTNG_DOMAIN_PYTHON:
2750 case LTTNG_DOMAIN_UST:
2751 break;
2752 default:
2753 ERR("Unknown UST domain on create session %d", domain->type);
2754 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2755 goto error;
2756 }
2757
2758 DBG("Creating UST session");
2759
2760 lus = trace_ust_create_session(session->id);
2761 if (lus == NULL) {
2762 ret = LTTNG_ERR_UST_SESS_FAIL;
2763 goto error;
2764 }
2765
2766 lus->uid = session->uid;
2767 lus->gid = session->gid;
2768 lus->output_traces = session->output_traces;
2769 lus->snapshot_mode = session->snapshot_mode;
2770 lus->live_timer_interval = session->live_timer;
2771 session->ust_session = lus;
2772 if (session->shm_path[0]) {
2773 strncpy(lus->root_shm_path, session->shm_path,
2774 sizeof(lus->root_shm_path));
2775 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2776 strncpy(lus->shm_path, session->shm_path,
2777 sizeof(lus->shm_path));
2778 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2779 strncat(lus->shm_path, "/ust",
2780 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2781 }
2782 /* Copy session output to the newly created UST session */
2783 ret = copy_session_consumer(domain->type, session);
2784 if (ret != LTTNG_OK) {
2785 goto error;
2786 }
2787
2788 return LTTNG_OK;
2789
2790 error:
2791 free(lus);
2792 session->ust_session = NULL;
2793 return ret;
2794 }
2795
2796 /*
2797 * Create a kernel tracer session then create the default channel.
2798 */
2799 static int create_kernel_session(struct ltt_session *session)
2800 {
2801 int ret;
2802
2803 DBG("Creating kernel session");
2804
2805 ret = kernel_create_session(session, kernel_tracer_fd);
2806 if (ret < 0) {
2807 ret = LTTNG_ERR_KERN_SESS_FAIL;
2808 goto error;
2809 }
2810
2811 /* Code flow safety */
2812 assert(session->kernel_session);
2813
2814 /* Copy session output to the newly created Kernel session */
2815 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2816 if (ret != LTTNG_OK) {
2817 goto error;
2818 }
2819
2820 /* Create directory(ies) on local filesystem. */
2821 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2822 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2823 ret = run_as_mkdir_recursive(
2824 session->kernel_session->consumer->dst.trace_path,
2825 S_IRWXU | S_IRWXG, session->uid, session->gid);
2826 if (ret < 0) {
2827 if (errno != EEXIST) {
2828 ERR("Trace directory creation error");
2829 goto error;
2830 }
2831 }
2832 }
2833
2834 session->kernel_session->uid = session->uid;
2835 session->kernel_session->gid = session->gid;
2836 session->kernel_session->output_traces = session->output_traces;
2837 session->kernel_session->snapshot_mode = session->snapshot_mode;
2838
2839 return LTTNG_OK;
2840
2841 error:
2842 trace_kernel_destroy_session(session->kernel_session);
2843 session->kernel_session = NULL;
2844 return ret;
2845 }
2846
2847 /*
2848 * Count number of session permitted by uid/gid.
2849 */
2850 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2851 {
2852 unsigned int i = 0;
2853 struct ltt_session *session;
2854
2855 DBG("Counting number of available session for UID %d GID %d",
2856 uid, gid);
2857 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2858 /*
2859 * Only list the sessions the user can control.
2860 */
2861 if (!session_access_ok(session, uid, gid)) {
2862 continue;
2863 }
2864 i++;
2865 }
2866 return i;
2867 }
2868
2869 /*
2870 * Process the command requested by the lttng client within the command
2871 * context structure. This function make sure that the return structure (llm)
2872 * is set and ready for transmission before returning.
2873 *
2874 * Return any error encountered or 0 for success.
2875 *
2876 * "sock" is only used for special-case var. len data.
2877 *
2878 * Should *NOT* be called with RCU read-side lock held.
2879 */
2880 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2881 int *sock_error)
2882 {
2883 int ret = LTTNG_OK;
2884 int need_tracing_session = 1;
2885 int need_domain;
2886
2887 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2888
2889 assert(!rcu_read_ongoing());
2890
2891 *sock_error = 0;
2892
2893 switch (cmd_ctx->lsm->cmd_type) {
2894 case LTTNG_CREATE_SESSION:
2895 case LTTNG_CREATE_SESSION_SNAPSHOT:
2896 case LTTNG_CREATE_SESSION_LIVE:
2897 case LTTNG_DESTROY_SESSION:
2898 case LTTNG_LIST_SESSIONS:
2899 case LTTNG_LIST_DOMAINS:
2900 case LTTNG_START_TRACE:
2901 case LTTNG_STOP_TRACE:
2902 case LTTNG_DATA_PENDING:
2903 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2904 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2905 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2906 case LTTNG_SNAPSHOT_RECORD:
2907 case LTTNG_SAVE_SESSION:
2908 case LTTNG_SET_SESSION_SHM_PATH:
2909 case LTTNG_REGENERATE_METADATA:
2910 case LTTNG_REGENERATE_STATEDUMP:
2911 case LTTNG_REGISTER_TRIGGER:
2912 case LTTNG_UNREGISTER_TRIGGER:
2913 need_domain = 0;
2914 break;
2915 default:
2916 need_domain = 1;
2917 }
2918
2919 if (config.no_kernel && need_domain
2920 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2921 if (!is_root) {
2922 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2923 } else {
2924 ret = LTTNG_ERR_KERN_NA;
2925 }
2926 goto error;
2927 }
2928
2929 /* Deny register consumer if we already have a spawned consumer. */
2930 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2931 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2932 if (kconsumer_data.pid > 0) {
2933 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2934 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2935 goto error;
2936 }
2937 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2938 }
2939
2940 /*
2941 * Check for command that don't needs to allocate a returned payload. We do
2942 * this here so we don't have to make the call for no payload at each
2943 * command.
2944 */
2945 switch(cmd_ctx->lsm->cmd_type) {
2946 case LTTNG_LIST_SESSIONS:
2947 case LTTNG_LIST_TRACEPOINTS:
2948 case LTTNG_LIST_TRACEPOINT_FIELDS:
2949 case LTTNG_LIST_DOMAINS:
2950 case LTTNG_LIST_CHANNELS:
2951 case LTTNG_LIST_EVENTS:
2952 case LTTNG_LIST_SYSCALLS:
2953 case LTTNG_LIST_TRACKER_PIDS:
2954 case LTTNG_DATA_PENDING:
2955 break;
2956 default:
2957 /* Setup lttng message with no payload */
2958 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
2959 if (ret < 0) {
2960 /* This label does not try to unlock the session */
2961 goto init_setup_error;
2962 }
2963 }
2964
2965 /* Commands that DO NOT need a session. */
2966 switch (cmd_ctx->lsm->cmd_type) {
2967 case LTTNG_CREATE_SESSION:
2968 case LTTNG_CREATE_SESSION_SNAPSHOT:
2969 case LTTNG_CREATE_SESSION_LIVE:
2970 case LTTNG_LIST_SESSIONS:
2971 case LTTNG_LIST_TRACEPOINTS:
2972 case LTTNG_LIST_SYSCALLS:
2973 case LTTNG_LIST_TRACEPOINT_FIELDS:
2974 case LTTNG_SAVE_SESSION:
2975 case LTTNG_REGISTER_TRIGGER:
2976 case LTTNG_UNREGISTER_TRIGGER:
2977 need_tracing_session = 0;
2978 break;
2979 default:
2980 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2981 /*
2982 * We keep the session list lock across _all_ commands
2983 * for now, because the per-session lock does not
2984 * handle teardown properly.
2985 */
2986 session_lock_list();
2987 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2988 if (cmd_ctx->session == NULL) {
2989 ret = LTTNG_ERR_SESS_NOT_FOUND;
2990 goto error;
2991 } else {
2992 /* Acquire lock for the session */
2993 session_lock(cmd_ctx->session);
2994 }
2995 break;
2996 }
2997
2998 /*
2999 * Commands that need a valid session but should NOT create one if none
3000 * exists. Instead of creating one and destroying it when the command is
3001 * handled, process that right before so we save some round trip in useless
3002 * code path.
3003 */
3004 switch (cmd_ctx->lsm->cmd_type) {
3005 case LTTNG_DISABLE_CHANNEL:
3006 case LTTNG_DISABLE_EVENT:
3007 switch (cmd_ctx->lsm->domain.type) {
3008 case LTTNG_DOMAIN_KERNEL:
3009 if (!cmd_ctx->session->kernel_session) {
3010 ret = LTTNG_ERR_NO_CHANNEL;
3011 goto error;
3012 }
3013 break;
3014 case LTTNG_DOMAIN_JUL:
3015 case LTTNG_DOMAIN_LOG4J:
3016 case LTTNG_DOMAIN_PYTHON:
3017 case LTTNG_DOMAIN_UST:
3018 if (!cmd_ctx->session->ust_session) {
3019 ret = LTTNG_ERR_NO_CHANNEL;
3020 goto error;
3021 }
3022 break;
3023 default:
3024 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3025 goto error;
3026 }
3027 default:
3028 break;
3029 }
3030
3031 if (!need_domain) {
3032 goto skip_domain;
3033 }
3034
3035 /*
3036 * Check domain type for specific "pre-action".
3037 */
3038 switch (cmd_ctx->lsm->domain.type) {
3039 case LTTNG_DOMAIN_KERNEL:
3040 if (!is_root) {
3041 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3042 goto error;
3043 }
3044
3045 /* Kernel tracer check */
3046 if (kernel_tracer_fd == -1) {
3047 /* Basically, load kernel tracer modules */
3048 ret = init_kernel_tracer();
3049 if (ret != 0) {
3050 goto error;
3051 }
3052 }
3053
3054 /* Consumer is in an ERROR state. Report back to client */
3055 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3056 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3057 goto error;
3058 }
3059
3060 /* Need a session for kernel command */
3061 if (need_tracing_session) {
3062 if (cmd_ctx->session->kernel_session == NULL) {
3063 ret = create_kernel_session(cmd_ctx->session);
3064 if (ret < 0) {
3065 ret = LTTNG_ERR_KERN_SESS_FAIL;
3066 goto error;
3067 }
3068 }
3069
3070 /* Start the kernel consumer daemon */
3071 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3072 if (kconsumer_data.pid == 0 &&
3073 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3074 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3075 ret = start_consumerd(&kconsumer_data);
3076 if (ret < 0) {
3077 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3078 goto error;
3079 }
3080 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3081 } else {
3082 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3083 }
3084
3085 /*
3086 * The consumer was just spawned so we need to add the socket to
3087 * the consumer output of the session if exist.
3088 */
3089 ret = consumer_create_socket(&kconsumer_data,
3090 cmd_ctx->session->kernel_session->consumer);
3091 if (ret < 0) {
3092 goto error;
3093 }
3094 }
3095
3096 break;
3097 case LTTNG_DOMAIN_JUL:
3098 case LTTNG_DOMAIN_LOG4J:
3099 case LTTNG_DOMAIN_PYTHON:
3100 case LTTNG_DOMAIN_UST:
3101 {
3102 if (!ust_app_supported()) {
3103 ret = LTTNG_ERR_NO_UST;
3104 goto error;
3105 }
3106 /* Consumer is in an ERROR state. Report back to client */
3107 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3108 ret = LTTNG_ERR_NO_USTCONSUMERD;
3109 goto error;
3110 }
3111
3112 if (need_tracing_session) {
3113 /* Create UST session if none exist. */
3114 if (cmd_ctx->session->ust_session == NULL) {
3115 ret = create_ust_session(cmd_ctx->session,
3116 &cmd_ctx->lsm->domain);
3117 if (ret != LTTNG_OK) {
3118 goto error;
3119 }
3120 }
3121
3122 /* Start the UST consumer daemons */
3123 /* 64-bit */
3124 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3125 if (config.consumerd64_bin_path.value &&
3126 ustconsumer64_data.pid == 0 &&
3127 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3128 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3129 ret = start_consumerd(&ustconsumer64_data);
3130 if (ret < 0) {
3131 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3132 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3133 goto error;
3134 }
3135
3136 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3137 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3138 } else {
3139 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3140 }
3141
3142 /*
3143 * Setup socket for consumer 64 bit. No need for atomic access
3144 * since it was set above and can ONLY be set in this thread.
3145 */
3146 ret = consumer_create_socket(&ustconsumer64_data,
3147 cmd_ctx->session->ust_session->consumer);
3148 if (ret < 0) {
3149 goto error;
3150 }
3151
3152 /* 32-bit */
3153 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3154 if (config.consumerd32_bin_path.value &&
3155 ustconsumer32_data.pid == 0 &&
3156 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3157 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3158 ret = start_consumerd(&ustconsumer32_data);
3159 if (ret < 0) {
3160 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3161 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3162 goto error;
3163 }
3164
3165 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3166 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3167 } else {
3168 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3169 }
3170
3171 /*
3172 * Setup socket for consumer 64 bit. No need for atomic access
3173 * since it was set above and can ONLY be set in this thread.
3174 */
3175 ret = consumer_create_socket(&ustconsumer32_data,
3176 cmd_ctx->session->ust_session->consumer);
3177 if (ret < 0) {
3178 goto error;
3179 }
3180 }
3181 break;
3182 }
3183 default:
3184 break;
3185 }
3186 skip_domain:
3187
3188 /* Validate consumer daemon state when start/stop trace command */
3189 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3190 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3191 switch (cmd_ctx->lsm->domain.type) {
3192 case LTTNG_DOMAIN_NONE:
3193 break;
3194 case LTTNG_DOMAIN_JUL:
3195 case LTTNG_DOMAIN_LOG4J:
3196 case LTTNG_DOMAIN_PYTHON:
3197 case LTTNG_DOMAIN_UST:
3198 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3199 ret = LTTNG_ERR_NO_USTCONSUMERD;
3200 goto error;
3201 }
3202 break;
3203 case LTTNG_DOMAIN_KERNEL:
3204 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3205 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3206 goto error;
3207 }
3208 break;
3209 default:
3210 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3211 goto error;
3212 }
3213 }
3214
3215 /*
3216 * Check that the UID or GID match that of the tracing session.
3217 * The root user can interact with all sessions.
3218 */
3219 if (need_tracing_session) {
3220 if (!session_access_ok(cmd_ctx->session,
3221 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3222 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3223 ret = LTTNG_ERR_EPERM;
3224 goto error;
3225 }
3226 }
3227
3228 /*
3229 * Send relayd information to consumer as soon as we have a domain and a
3230 * session defined.
3231 */
3232 if (cmd_ctx->session && need_domain) {
3233 /*
3234 * Setup relayd if not done yet. If the relayd information was already
3235 * sent to the consumer, this call will gracefully return.
3236 */
3237 ret = cmd_setup_relayd(cmd_ctx->session);
3238 if (ret != LTTNG_OK) {
3239 goto error;
3240 }
3241 }
3242
3243 /* Process by command type */
3244 switch (cmd_ctx->lsm->cmd_type) {
3245 case LTTNG_ADD_CONTEXT:
3246 {
3247 /*
3248 * An LTTNG_ADD_CONTEXT command might have a supplementary
3249 * payload if the context being added is an application context.
3250 */
3251 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3252 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3253 char *provider_name = NULL, *context_name = NULL;
3254 size_t provider_name_len =
3255 cmd_ctx->lsm->u.context.provider_name_len;
3256 size_t context_name_len =
3257 cmd_ctx->lsm->u.context.context_name_len;
3258
3259 if (provider_name_len == 0 || context_name_len == 0) {
3260 /*
3261 * Application provider and context names MUST
3262 * be provided.
3263 */
3264 ret = -LTTNG_ERR_INVALID;
3265 goto error;
3266 }
3267
3268 provider_name = zmalloc(provider_name_len + 1);
3269 if (!provider_name) {
3270 ret = -LTTNG_ERR_NOMEM;
3271 goto error;
3272 }
3273 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3274 provider_name;
3275
3276 context_name = zmalloc(context_name_len + 1);
3277 if (!context_name) {
3278 ret = -LTTNG_ERR_NOMEM;
3279 goto error_add_context;
3280 }
3281 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3282 context_name;
3283
3284 ret = lttcomm_recv_unix_sock(sock, provider_name,
3285 provider_name_len);
3286 if (ret < 0) {
3287 goto error_add_context;
3288 }
3289
3290 ret = lttcomm_recv_unix_sock(sock, context_name,
3291 context_name_len);
3292 if (ret < 0) {
3293 goto error_add_context;
3294 }
3295 }
3296
3297 /*
3298 * cmd_add_context assumes ownership of the provider and context
3299 * names.
3300 */
3301 ret = cmd_add_context(cmd_ctx->session,
3302 cmd_ctx->lsm->domain.type,
3303 cmd_ctx->lsm->u.context.channel_name,
3304 &cmd_ctx->lsm->u.context.ctx,
3305 kernel_poll_pipe[1]);
3306
3307 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3308 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3309 error_add_context:
3310 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3311 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3312 if (ret < 0) {
3313 goto error;
3314 }
3315 break;
3316 }
3317 case LTTNG_DISABLE_CHANNEL:
3318 {
3319 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3320 cmd_ctx->lsm->u.disable.channel_name);
3321 break;
3322 }
3323 case LTTNG_DISABLE_EVENT:
3324 {
3325
3326 /*
3327 * FIXME: handle filter; for now we just receive the filter's
3328 * bytecode along with the filter expression which are sent by
3329 * liblttng-ctl and discard them.
3330 *
3331 * This fixes an issue where the client may block while sending
3332 * the filter payload and encounter an error because the session
3333 * daemon closes the socket without ever handling this data.
3334 */
3335 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3336 cmd_ctx->lsm->u.disable.bytecode_len;
3337
3338 if (count) {
3339 char data[LTTNG_FILTER_MAX_LEN];
3340
3341 DBG("Discarding disable event command payload of size %zu", count);
3342 while (count) {
3343 ret = lttcomm_recv_unix_sock(sock, data,
3344 count > sizeof(data) ? sizeof(data) : count);
3345 if (ret < 0) {
3346 goto error;
3347 }
3348
3349 count -= (size_t) ret;
3350 }
3351 }
3352 /* FIXME: passing packed structure to non-packed pointer */
3353 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3354 cmd_ctx->lsm->u.disable.channel_name,
3355 &cmd_ctx->lsm->u.disable.event);
3356 break;
3357 }
3358 case LTTNG_ENABLE_CHANNEL:
3359 {
3360 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3361 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3362 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3363 &cmd_ctx->lsm->u.channel.chan,
3364 kernel_poll_pipe[1]);
3365 break;
3366 }
3367 case LTTNG_TRACK_PID:
3368 {
3369 ret = cmd_track_pid(cmd_ctx->session,
3370 cmd_ctx->lsm->domain.type,
3371 cmd_ctx->lsm->u.pid_tracker.pid);
3372 break;
3373 }
3374 case LTTNG_UNTRACK_PID:
3375 {
3376 ret = cmd_untrack_pid(cmd_ctx->session,
3377 cmd_ctx->lsm->domain.type,
3378 cmd_ctx->lsm->u.pid_tracker.pid);
3379 break;
3380 }
3381 case LTTNG_ENABLE_EVENT:
3382 {
3383 struct lttng_event_exclusion *exclusion = NULL;
3384 struct lttng_filter_bytecode *bytecode = NULL;
3385 char *filter_expression = NULL;
3386
3387 /* Handle exclusion events and receive it from the client. */
3388 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3389 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3390
3391 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3392 (count * LTTNG_SYMBOL_NAME_LEN));
3393 if (!exclusion) {
3394 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3395 goto error;
3396 }
3397
3398 DBG("Receiving var len exclusion event list from client ...");
3399 exclusion->count = count;
3400 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3401 count * LTTNG_SYMBOL_NAME_LEN);
3402 if (ret <= 0) {
3403 DBG("Nothing recv() from client var len data... continuing");
3404 *sock_error = 1;
3405 free(exclusion);
3406 ret = LTTNG_ERR_EXCLUSION_INVAL;
3407 goto error;
3408 }
3409 }
3410
3411 /* Get filter expression from client. */
3412 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3413 size_t expression_len =
3414 cmd_ctx->lsm->u.enable.expression_len;
3415
3416 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3417 ret = LTTNG_ERR_FILTER_INVAL;
3418 free(exclusion);
3419 goto error;
3420 }
3421
3422 filter_expression = zmalloc(expression_len);
3423 if (!filter_expression) {
3424 free(exclusion);
3425 ret = LTTNG_ERR_FILTER_NOMEM;
3426 goto error;
3427 }
3428
3429 /* Receive var. len. data */
3430 DBG("Receiving var len filter's expression from client ...");
3431 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3432 expression_len);
3433 if (ret <= 0) {
3434 DBG("Nothing recv() from client car len data... continuing");
3435 *sock_error = 1;
3436 free(filter_expression);
3437 free(exclusion);
3438 ret = LTTNG_ERR_FILTER_INVAL;
3439 goto error;
3440 }
3441 }
3442
3443 /* Handle filter and get bytecode from client. */
3444 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3445 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3446
3447 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3448 ret = LTTNG_ERR_FILTER_INVAL;
3449 free(filter_expression);
3450 free(exclusion);
3451 goto error;
3452 }
3453
3454 bytecode = zmalloc(bytecode_len);
3455 if (!bytecode) {
3456 free(filter_expression);
3457 free(exclusion);
3458 ret = LTTNG_ERR_FILTER_NOMEM;
3459 goto error;
3460 }
3461
3462 /* Receive var. len. data */
3463 DBG("Receiving var len filter's bytecode from client ...");
3464 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3465 if (ret <= 0) {
3466 DBG("Nothing recv() from client car len data... continuing");
3467 *sock_error = 1;
3468 free(filter_expression);
3469 free(bytecode);
3470 free(exclusion);
3471 ret = LTTNG_ERR_FILTER_INVAL;
3472 goto error;
3473 }
3474
3475 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3476 free(filter_expression);
3477 free(bytecode);
3478 free(exclusion);
3479 ret = LTTNG_ERR_FILTER_INVAL;
3480 goto error;
3481 }
3482 }
3483
3484 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3485 cmd_ctx->lsm->u.enable.channel_name,
3486 &cmd_ctx->lsm->u.enable.event,
3487 filter_expression, bytecode, exclusion,
3488 kernel_poll_pipe[1]);
3489 break;
3490 }
3491 case LTTNG_LIST_TRACEPOINTS:
3492 {
3493 struct lttng_event *events;
3494 ssize_t nb_events;
3495
3496 session_lock_list();
3497 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3498 session_unlock_list();
3499 if (nb_events < 0) {
3500 /* Return value is a negative lttng_error_code. */
3501 ret = -nb_events;
3502 goto error;
3503 }
3504
3505 /*
3506 * Setup lttng message with payload size set to the event list size in
3507 * bytes and then copy list into the llm payload.
3508 */
3509 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3510 sizeof(struct lttng_event) * nb_events);
3511 free(events);
3512
3513 if (ret < 0) {
3514 goto setup_error;
3515 }
3516
3517 ret = LTTNG_OK;
3518 break;
3519 }
3520 case LTTNG_LIST_TRACEPOINT_FIELDS:
3521 {
3522 struct lttng_event_field *fields;
3523 ssize_t nb_fields;
3524
3525 session_lock_list();
3526 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3527 &fields);
3528 session_unlock_list();
3529 if (nb_fields < 0) {
3530 /* Return value is a negative lttng_error_code. */
3531 ret = -nb_fields;
3532 goto error;
3533 }
3534
3535 /*
3536 * Setup lttng message with payload size set to the event list size in
3537 * bytes and then copy list into the llm payload.
3538 */
3539 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3540 sizeof(struct lttng_event_field) * nb_fields);
3541 free(fields);
3542
3543 if (ret < 0) {
3544 goto setup_error;
3545 }
3546
3547 ret = LTTNG_OK;
3548 break;
3549 }
3550 case LTTNG_LIST_SYSCALLS:
3551 {
3552 struct lttng_event *events;
3553 ssize_t nb_events;
3554
3555 nb_events = cmd_list_syscalls(&events);
3556 if (nb_events < 0) {
3557 /* Return value is a negative lttng_error_code. */
3558 ret = -nb_events;
3559 goto error;
3560 }
3561
3562 /*
3563 * Setup lttng message with payload size set to the event list size in
3564 * bytes and then copy list into the llm payload.
3565 */
3566 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3567 sizeof(struct lttng_event) * nb_events);
3568 free(events);
3569
3570 if (ret < 0) {
3571 goto setup_error;
3572 }
3573
3574 ret = LTTNG_OK;
3575 break;
3576 }
3577 case LTTNG_LIST_TRACKER_PIDS:
3578 {
3579 int32_t *pids = NULL;
3580 ssize_t nr_pids;
3581
3582 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3583 cmd_ctx->lsm->domain.type, &pids);
3584 if (nr_pids < 0) {
3585 /* Return value is a negative lttng_error_code. */
3586 ret = -nr_pids;
3587 goto error;
3588 }
3589
3590 /*
3591 * Setup lttng message with payload size set to the event list size in
3592 * bytes and then copy list into the llm payload.
3593 */
3594 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3595 sizeof(int32_t) * nr_pids);
3596 free(pids);
3597
3598 if (ret < 0) {
3599 goto setup_error;
3600 }
3601
3602 ret = LTTNG_OK;
3603 break;
3604 }
3605 case LTTNG_SET_CONSUMER_URI:
3606 {
3607 size_t nb_uri, len;
3608 struct lttng_uri *uris;
3609
3610 nb_uri = cmd_ctx->lsm->u.uri.size;
3611 len = nb_uri * sizeof(struct lttng_uri);
3612
3613 if (nb_uri == 0) {
3614 ret = LTTNG_ERR_INVALID;
3615 goto error;
3616 }
3617
3618 uris = zmalloc(len);
3619 if (uris == NULL) {
3620 ret = LTTNG_ERR_FATAL;
3621 goto error;
3622 }
3623
3624 /* Receive variable len data */
3625 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3626 ret = lttcomm_recv_unix_sock(sock, uris, len);
3627 if (ret <= 0) {
3628 DBG("No URIs received from client... continuing");
3629 *sock_error = 1;
3630 ret = LTTNG_ERR_SESSION_FAIL;
3631 free(uris);
3632 goto error;
3633 }
3634
3635 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3636 free(uris);
3637 if (ret != LTTNG_OK) {
3638 goto error;
3639 }
3640
3641
3642 break;
3643 }
3644 case LTTNG_START_TRACE:
3645 {
3646 ret = cmd_start_trace(cmd_ctx->session);
3647 break;
3648 }
3649 case LTTNG_STOP_TRACE:
3650 {
3651 ret = cmd_stop_trace(cmd_ctx->session);
3652 break;
3653 }
3654 case LTTNG_CREATE_SESSION:
3655 {
3656 size_t nb_uri, len;
3657 struct lttng_uri *uris = NULL;
3658
3659 nb_uri = cmd_ctx->lsm->u.uri.size;
3660 len = nb_uri * sizeof(struct lttng_uri);
3661
3662 if (nb_uri > 0) {
3663 uris = zmalloc(len);
3664 if (uris == NULL) {
3665 ret = LTTNG_ERR_FATAL;
3666 goto error;
3667 }
3668
3669 /* Receive variable len data */
3670 DBG("Waiting for %zu URIs from client ...", nb_uri);
3671 ret = lttcomm_recv_unix_sock(sock, uris, len);
3672 if (ret <= 0) {
3673 DBG("No URIs received from client... continuing");
3674 *sock_error = 1;
3675 ret = LTTNG_ERR_SESSION_FAIL;
3676 free(uris);
3677 goto error;
3678 }
3679
3680 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3681 DBG("Creating session with ONE network URI is a bad call");
3682 ret = LTTNG_ERR_SESSION_FAIL;
3683 free(uris);
3684 goto error;
3685 }
3686 }
3687
3688 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3689 &cmd_ctx->creds, 0);
3690
3691 free(uris);
3692
3693 break;
3694 }
3695 case LTTNG_DESTROY_SESSION:
3696 {
3697 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3698
3699 /* Set session to NULL so we do not unlock it after free. */
3700 cmd_ctx->session = NULL;
3701 break;
3702 }
3703 case LTTNG_LIST_DOMAINS:
3704 {
3705 ssize_t nb_dom;
3706 struct lttng_domain *domains = NULL;
3707
3708 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3709 if (nb_dom < 0) {
3710 /* Return value is a negative lttng_error_code. */
3711 ret = -nb_dom;
3712 goto error;
3713 }
3714
3715 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3716 nb_dom * sizeof(struct lttng_domain));
3717 free(domains);
3718
3719 if (ret < 0) {
3720 goto setup_error;
3721 }
3722
3723 ret = LTTNG_OK;
3724 break;
3725 }
3726 case LTTNG_LIST_CHANNELS:
3727 {
3728 ssize_t payload_size;
3729 struct lttng_channel *channels = NULL;
3730
3731 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3732 cmd_ctx->session, &channels);
3733 if (payload_size < 0) {
3734 /* Return value is a negative lttng_error_code. */
3735 ret = -payload_size;
3736 goto error;
3737 }
3738
3739 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3740 payload_size);
3741 free(channels);
3742
3743 if (ret < 0) {
3744 goto setup_error;
3745 }
3746
3747 ret = LTTNG_OK;
3748 break;
3749 }
3750 case LTTNG_LIST_EVENTS:
3751 {
3752 ssize_t nb_event;
3753 struct lttng_event *events = NULL;
3754 struct lttcomm_event_command_header cmd_header;
3755 size_t total_size;
3756
3757 memset(&cmd_header, 0, sizeof(cmd_header));
3758 /* Extended infos are included at the end of events */
3759 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3760 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3761 &events, &total_size);
3762
3763 if (nb_event < 0) {
3764 /* Return value is a negative lttng_error_code. */
3765 ret = -nb_event;
3766 goto error;
3767 }
3768
3769 cmd_header.nb_events = nb_event;
3770 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3771 &cmd_header, sizeof(cmd_header));
3772 free(events);
3773
3774 if (ret < 0) {
3775 goto setup_error;
3776 }
3777
3778 ret = LTTNG_OK;
3779 break;
3780 }
3781 case LTTNG_LIST_SESSIONS:
3782 {
3783 unsigned int nr_sessions;
3784 void *sessions_payload;
3785 size_t payload_len;
3786
3787 session_lock_list();
3788 nr_sessions = lttng_sessions_count(
3789 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3790 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3791 payload_len = sizeof(struct lttng_session) * nr_sessions;
3792 sessions_payload = zmalloc(payload_len);
3793
3794 if (!sessions_payload) {
3795 session_unlock_list();
3796 ret = -ENOMEM;
3797 goto setup_error;
3798 }
3799
3800 cmd_list_lttng_sessions(sessions_payload,
3801 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3802 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3803 session_unlock_list();
3804
3805 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3806 payload_len);
3807 free(sessions_payload);
3808
3809 if (ret < 0) {
3810 goto setup_error;
3811 }
3812
3813 ret = LTTNG_OK;
3814 break;
3815 }
3816 case LTTNG_REGISTER_CONSUMER:
3817 {
3818 struct consumer_data *cdata;
3819
3820 switch (cmd_ctx->lsm->domain.type) {
3821 case LTTNG_DOMAIN_KERNEL:
3822 cdata = &kconsumer_data;
3823 break;
3824 default:
3825 ret = LTTNG_ERR_UND;
3826 goto error;
3827 }
3828
3829 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3830 cmd_ctx->lsm->u.reg.path, cdata);
3831 break;
3832 }
3833 case LTTNG_DATA_PENDING:
3834 {
3835 int pending_ret;
3836 uint8_t pending_ret_byte;
3837
3838 pending_ret = cmd_data_pending(cmd_ctx->session);
3839
3840 /*
3841 * FIXME
3842 *
3843 * This function may returns 0 or 1 to indicate whether or not
3844 * there is data pending. In case of error, it should return an
3845 * LTTNG_ERR code. However, some code paths may still return
3846 * a nondescript error code, which we handle by returning an
3847 * "unknown" error.
3848 */
3849 if (pending_ret == 0 || pending_ret == 1) {
3850 /*
3851 * ret will be set to LTTNG_OK at the end of
3852 * this function.
3853 */
3854 } else if (pending_ret < 0) {
3855 ret = LTTNG_ERR_UNK;
3856 goto setup_error;
3857 } else {
3858 ret = pending_ret;
3859 goto setup_error;
3860 }
3861
3862 pending_ret_byte = (uint8_t) pending_ret;
3863
3864 /* 1 byte to return whether or not data is pending */
3865 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3866 &pending_ret_byte, 1);
3867
3868 if (ret < 0) {
3869 goto setup_error;
3870 }
3871
3872 ret = LTTNG_OK;
3873 break;
3874 }
3875 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3876 {
3877 struct lttcomm_lttng_output_id reply;
3878
3879 ret = cmd_snapshot_add_output(cmd_ctx->session,
3880 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3881 if (ret != LTTNG_OK) {
3882 goto error;
3883 }
3884
3885 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
3886 sizeof(reply));
3887 if (ret < 0) {
3888 goto setup_error;
3889 }
3890
3891 /* Copy output list into message payload */
3892 ret = LTTNG_OK;
3893 break;
3894 }
3895 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3896 {
3897 ret = cmd_snapshot_del_output(cmd_ctx->session,
3898 &cmd_ctx->lsm->u.snapshot_output.output);
3899 break;
3900 }
3901 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3902 {
3903 ssize_t nb_output;
3904 struct lttng_snapshot_output *outputs = NULL;
3905
3906 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3907 if (nb_output < 0) {
3908 ret = -nb_output;
3909 goto error;
3910 }
3911
3912 assert((nb_output > 0 && outputs) || nb_output == 0);
3913 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
3914 nb_output * sizeof(struct lttng_snapshot_output));
3915 free(outputs);
3916
3917 if (ret < 0) {
3918 goto setup_error;
3919 }
3920
3921 ret = LTTNG_OK;
3922 break;
3923 }
3924 case LTTNG_SNAPSHOT_RECORD:
3925 {
3926 ret = cmd_snapshot_record(cmd_ctx->session,
3927 &cmd_ctx->lsm->u.snapshot_record.output,
3928 cmd_ctx->lsm->u.snapshot_record.wait);
3929 break;
3930 }
3931 case LTTNG_CREATE_SESSION_SNAPSHOT:
3932 {
3933 size_t nb_uri, len;
3934 struct lttng_uri *uris = NULL;
3935
3936 nb_uri = cmd_ctx->lsm->u.uri.size;
3937 len = nb_uri * sizeof(struct lttng_uri);
3938
3939 if (nb_uri > 0) {
3940 uris = zmalloc(len);
3941 if (uris == NULL) {
3942 ret = LTTNG_ERR_FATAL;
3943 goto error;
3944 }
3945
3946 /* Receive variable len data */
3947 DBG("Waiting for %zu URIs from client ...", nb_uri);
3948 ret = lttcomm_recv_unix_sock(sock, uris, len);
3949 if (ret <= 0) {
3950 DBG("No URIs received from client... continuing");
3951 *sock_error = 1;
3952 ret = LTTNG_ERR_SESSION_FAIL;
3953 free(uris);
3954 goto error;
3955 }
3956
3957 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3958 DBG("Creating session with ONE network URI is a bad call");
3959 ret = LTTNG_ERR_SESSION_FAIL;
3960 free(uris);
3961 goto error;
3962 }
3963 }
3964
3965 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3966 nb_uri, &cmd_ctx->creds);
3967 free(uris);
3968 break;
3969 }
3970 case LTTNG_CREATE_SESSION_LIVE:
3971 {
3972 size_t nb_uri, len;
3973 struct lttng_uri *uris = NULL;
3974
3975 nb_uri = cmd_ctx->lsm->u.uri.size;
3976 len = nb_uri * sizeof(struct lttng_uri);
3977
3978 if (nb_uri > 0) {
3979 uris = zmalloc(len);
3980 if (uris == NULL) {
3981 ret = LTTNG_ERR_FATAL;
3982 goto error;
3983 }
3984
3985 /* Receive variable len data */
3986 DBG("Waiting for %zu URIs from client ...", nb_uri);
3987 ret = lttcomm_recv_unix_sock(sock, uris, len);
3988 if (ret <= 0) {
3989 DBG("No URIs received from client... continuing");
3990 *sock_error = 1;
3991 ret = LTTNG_ERR_SESSION_FAIL;
3992 free(uris);
3993 goto error;
3994 }
3995
3996 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3997 DBG("Creating session with ONE network URI is a bad call");
3998 ret = LTTNG_ERR_SESSION_FAIL;
3999 free(uris);
4000 goto error;
4001 }
4002 }
4003
4004 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4005 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4006 free(uris);
4007 break;
4008 }
4009 case LTTNG_SAVE_SESSION:
4010 {
4011 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4012 &cmd_ctx->creds);
4013 break;
4014 }
4015 case LTTNG_SET_SESSION_SHM_PATH:
4016 {
4017 ret = cmd_set_session_shm_path(cmd_ctx->session,
4018 cmd_ctx->lsm->u.set_shm_path.shm_path);
4019 break;
4020 }
4021 case LTTNG_REGENERATE_METADATA:
4022 {
4023 ret = cmd_regenerate_metadata(cmd_ctx->session);
4024 break;
4025 }
4026 case LTTNG_REGENERATE_STATEDUMP:
4027 {
4028 ret = cmd_regenerate_statedump(cmd_ctx->session);
4029 break;
4030 }
4031 case LTTNG_REGISTER_TRIGGER:
4032 {
4033 ret = cmd_register_trigger(cmd_ctx, sock,
4034 notification_thread_handle);
4035 break;
4036 }
4037 case LTTNG_UNREGISTER_TRIGGER:
4038 {
4039 ret = cmd_unregister_trigger(cmd_ctx, sock,
4040 notification_thread_handle);
4041 break;
4042 }
4043 default:
4044 ret = LTTNG_ERR_UND;
4045 break;
4046 }
4047
4048 error:
4049 if (cmd_ctx->llm == NULL) {
4050 DBG("Missing llm structure. Allocating one.");
4051 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4052 goto setup_error;
4053 }
4054 }
4055 /* Set return code */
4056 cmd_ctx->llm->ret_code = ret;
4057 setup_error:
4058 if (cmd_ctx->session) {
4059 session_unlock(cmd_ctx->session);
4060 }
4061 if (need_tracing_session) {
4062 session_unlock_list();
4063 }
4064 init_setup_error:
4065 assert(!rcu_read_ongoing());
4066 return ret;
4067 }
4068
4069 /*
4070 * Thread managing health check socket.
4071 */
4072 static void *thread_manage_health(void *data)
4073 {
4074 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4075 uint32_t revents, nb_fd;
4076 struct lttng_poll_event events;
4077 struct health_comm_msg msg;
4078 struct health_comm_reply reply;
4079
4080 DBG("[thread] Manage health check started");
4081
4082 rcu_register_thread();
4083
4084 /* We might hit an error path before this is created. */
4085 lttng_poll_init(&events);
4086
4087 /* Create unix socket */
4088 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4089 if (sock < 0) {
4090 ERR("Unable to create health check Unix socket");
4091 goto error;
4092 }
4093
4094 if (is_root) {
4095 /* lttng health client socket path permissions */
4096 ret = chown(config.health_unix_sock_path.value, 0,
4097 utils_get_group_id(config.tracing_group_name.value));
4098 if (ret < 0) {
4099 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4100 PERROR("chown");
4101 goto error;
4102 }
4103
4104 ret = chmod(config.health_unix_sock_path.value,
4105 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4106 if (ret < 0) {
4107 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4108 PERROR("chmod");
4109 goto error;
4110 }
4111 }
4112
4113 /*
4114 * Set the CLOEXEC flag. Return code is useless because either way, the
4115 * show must go on.
4116 */
4117 (void) utils_set_fd_cloexec(sock);
4118
4119 ret = lttcomm_listen_unix_sock(sock);
4120 if (ret < 0) {
4121 goto error;
4122 }
4123
4124 /*
4125 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4126 * more will be added to this poll set.
4127 */
4128 ret = sessiond_set_thread_pollset(&events, 2);
4129 if (ret < 0) {
4130 goto error;
4131 }
4132
4133 /* Add the application registration socket */
4134 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4135 if (ret < 0) {
4136 goto error;
4137 }
4138
4139 sessiond_notify_ready();
4140
4141 while (1) {
4142 DBG("Health check ready");
4143
4144 /* Inifinite blocking call, waiting for transmission */
4145 restart:
4146 ret = lttng_poll_wait(&events, -1);
4147 if (ret < 0) {
4148 /*
4149 * Restart interrupted system call.
4150 */
4151 if (errno == EINTR) {
4152 goto restart;
4153 }
4154 goto error;
4155 }
4156
4157 nb_fd = ret;
4158
4159 for (i = 0; i < nb_fd; i++) {
4160 /* Fetch once the poll data */
4161 revents = LTTNG_POLL_GETEV(&events, i);
4162 pollfd = LTTNG_POLL_GETFD(&events, i);
4163
4164 if (!revents) {
4165 /* No activity for this FD (poll implementation). */
4166 continue;
4167 }
4168
4169 /* Thread quit pipe has been closed. Killing thread. */
4170 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4171 if (ret) {
4172 err = 0;
4173 goto exit;
4174 }
4175
4176 /* Event on the registration socket */
4177 if (pollfd == sock) {
4178 if (revents & LPOLLIN) {
4179 continue;
4180 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4181 ERR("Health socket poll error");
4182 goto error;
4183 } else {
4184 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4185 goto error;
4186 }
4187 }
4188 }
4189
4190 new_sock = lttcomm_accept_unix_sock(sock);
4191 if (new_sock < 0) {
4192 goto error;
4193 }
4194
4195 /*
4196 * Set the CLOEXEC flag. Return code is useless because either way, the
4197 * show must go on.
4198 */
4199 (void) utils_set_fd_cloexec(new_sock);
4200
4201 DBG("Receiving data from client for health...");
4202 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4203 if (ret <= 0) {
4204 DBG("Nothing recv() from client... continuing");
4205 ret = close(new_sock);
4206 if (ret) {
4207 PERROR("close");
4208 }
4209 continue;
4210 }
4211
4212 rcu_thread_online();
4213
4214 memset(&reply, 0, sizeof(reply));
4215 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4216 /*
4217 * health_check_state returns 0 if health is
4218 * bad.
4219 */
4220 if (!health_check_state(health_sessiond, i)) {
4221 reply.ret_code |= 1ULL << i;
4222 }
4223 }
4224
4225 DBG2("Health check return value %" PRIx64, reply.ret_code);
4226
4227 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4228 if (ret < 0) {
4229 ERR("Failed to send health data back to client");
4230 }
4231
4232 /* End of transmission */
4233 ret = close(new_sock);
4234 if (ret) {
4235 PERROR("close");
4236 }
4237 }
4238
4239 exit:
4240 error:
4241 if (err) {
4242 ERR("Health error occurred in %s", __func__);
4243 }
4244 DBG("Health check thread dying");
4245 unlink(config.health_unix_sock_path.value);
4246 if (sock >= 0) {
4247 ret = close(sock);
4248 if (ret) {
4249 PERROR("close");
4250 }
4251 }
4252
4253 lttng_poll_clean(&events);
4254 stop_threads();
4255 rcu_unregister_thread();
4256 return NULL;
4257 }
4258
4259 /*
4260 * This thread manage all clients request using the unix client socket for
4261 * communication.
4262 */
4263 static void *thread_manage_clients(void *data)
4264 {
4265 int sock = -1, ret, i, pollfd, err = -1;
4266 int sock_error;
4267 uint32_t revents, nb_fd;
4268 struct command_ctx *cmd_ctx = NULL;
4269 struct lttng_poll_event events;
4270
4271 DBG("[thread] Manage client started");
4272
4273 rcu_register_thread();
4274
4275 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4276
4277 health_code_update();
4278
4279 ret = lttcomm_listen_unix_sock(client_sock);
4280 if (ret < 0) {
4281 goto error_listen;
4282 }
4283
4284 /*
4285 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4286 * more will be added to this poll set.
4287 */
4288 ret = sessiond_set_thread_pollset(&events, 2);
4289 if (ret < 0) {
4290 goto error_create_poll;
4291 }
4292
4293 /* Add the application registration socket */
4294 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4295 if (ret < 0) {
4296 goto error;
4297 }
4298
4299 sessiond_notify_ready();
4300 ret = sem_post(&load_info->message_thread_ready);
4301 if (ret) {
4302 PERROR("sem_post message_thread_ready");
4303 goto error;
4304 }
4305
4306 /* This testpoint is after we signal readiness to the parent. */
4307 if (testpoint(sessiond_thread_manage_clients)) {
4308 goto error;
4309 }
4310
4311 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4312 goto error;
4313 }
4314
4315 health_code_update();
4316
4317 while (1) {
4318 DBG("Accepting client command ...");
4319
4320 /* Inifinite blocking call, waiting for transmission */
4321 restart:
4322 health_poll_entry();
4323 ret = lttng_poll_wait(&events, -1);
4324 health_poll_exit();
4325 if (ret < 0) {
4326 /*
4327 * Restart interrupted system call.
4328 */
4329 if (errno == EINTR) {
4330 goto restart;
4331 }
4332 goto error;
4333 }
4334
4335 nb_fd = ret;
4336
4337 for (i = 0; i < nb_fd; i++) {
4338 /* Fetch once the poll data */
4339 revents = LTTNG_POLL_GETEV(&events, i);
4340 pollfd = LTTNG_POLL_GETFD(&events, i);
4341
4342 health_code_update();
4343
4344 if (!revents) {
4345 /* No activity for this FD (poll implementation). */
4346 continue;
4347 }
4348
4349 /* Thread quit pipe has been closed. Killing thread. */
4350 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4351 if (ret) {
4352 err = 0;
4353 goto exit;
4354 }
4355
4356 /* Event on the registration socket */
4357 if (pollfd == client_sock) {
4358 if (revents & LPOLLIN) {
4359 continue;
4360 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4361 ERR("Client socket poll error");
4362 goto error;
4363 } else {
4364 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4365 goto error;
4366 }
4367 }
4368 }
4369
4370 DBG("Wait for client response");
4371
4372 health_code_update();
4373
4374 sock = lttcomm_accept_unix_sock(client_sock);
4375 if (sock < 0) {
4376 goto error;
4377 }
4378
4379 /*
4380 * Set the CLOEXEC flag. Return code is useless because either way, the
4381 * show must go on.
4382 */
4383 (void) utils_set_fd_cloexec(sock);
4384
4385 /* Set socket option for credentials retrieval */
4386 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4387 if (ret < 0) {
4388 goto error;
4389 }
4390
4391 /* Allocate context command to process the client request */
4392 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4393 if (cmd_ctx == NULL) {
4394 PERROR("zmalloc cmd_ctx");
4395 goto error;
4396 }
4397
4398 /* Allocate data buffer for reception */
4399 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4400 if (cmd_ctx->lsm == NULL) {
4401 PERROR("zmalloc cmd_ctx->lsm");
4402 goto error;
4403 }
4404
4405 cmd_ctx->llm = NULL;
4406 cmd_ctx->session = NULL;
4407
4408 health_code_update();
4409
4410 /*
4411 * Data is received from the lttng client. The struct
4412 * lttcomm_session_msg (lsm) contains the command and data request of
4413 * the client.
4414 */
4415 DBG("Receiving data from client ...");
4416 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4417 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4418 if (ret <= 0) {
4419 DBG("Nothing recv() from client... continuing");
4420 ret = close(sock);
4421 if (ret) {
4422 PERROR("close");
4423 }
4424 sock = -1;
4425 clean_command_ctx(&cmd_ctx);
4426 continue;
4427 }
4428
4429 health_code_update();
4430
4431 // TODO: Validate cmd_ctx including sanity check for
4432 // security purpose.
4433
4434 rcu_thread_online();
4435 /*
4436 * This function dispatch the work to the kernel or userspace tracer
4437 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4438 * informations for the client. The command context struct contains
4439 * everything this function may needs.
4440 */
4441 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4442 rcu_thread_offline();
4443 if (ret < 0) {
4444 ret = close(sock);
4445 if (ret) {
4446 PERROR("close");
4447 }
4448 sock = -1;
4449 /*
4450 * TODO: Inform client somehow of the fatal error. At
4451 * this point, ret < 0 means that a zmalloc failed
4452 * (ENOMEM). Error detected but still accept
4453 * command, unless a socket error has been
4454 * detected.
4455 */
4456 clean_command_ctx(&cmd_ctx);
4457 continue;
4458 }
4459
4460 health_code_update();
4461
4462 DBG("Sending response (size: %d, retcode: %s (%d))",
4463 cmd_ctx->lttng_msg_size,
4464 lttng_strerror(-cmd_ctx->llm->ret_code),
4465 cmd_ctx->llm->ret_code);
4466 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4467 if (ret < 0) {
4468 ERR("Failed to send data back to client");
4469 }
4470
4471 /* End of transmission */
4472 ret = close(sock);
4473 if (ret) {
4474 PERROR("close");
4475 }
4476 sock = -1;
4477
4478 clean_command_ctx(&cmd_ctx);
4479
4480 health_code_update();
4481 }
4482
4483 exit:
4484 error:
4485 if (sock >= 0) {
4486 ret = close(sock);
4487 if (ret) {
4488 PERROR("close");
4489 }
4490 }
4491
4492 lttng_poll_clean(&events);
4493 clean_command_ctx(&cmd_ctx);
4494
4495 error_listen:
4496 error_create_poll:
4497 unlink(config.client_unix_sock_path.value);
4498 if (client_sock >= 0) {
4499 ret = close(client_sock);
4500 if (ret) {
4501 PERROR("close");
4502 }
4503 }
4504
4505 if (err) {
4506 health_error();
4507 ERR("Health error occurred in %s", __func__);
4508 }
4509
4510 health_unregister(health_sessiond);
4511
4512 DBG("Client thread dying");
4513
4514 rcu_unregister_thread();
4515
4516 /*
4517 * Since we are creating the consumer threads, we own them, so we need
4518 * to join them before our thread exits.
4519 */
4520 ret = join_consumer_thread(&kconsumer_data);
4521 if (ret) {
4522 errno = ret;
4523 PERROR("join_consumer");
4524 }
4525
4526 ret = join_consumer_thread(&ustconsumer32_data);
4527 if (ret) {
4528 errno = ret;
4529 PERROR("join_consumer ust32");
4530 }
4531
4532 ret = join_consumer_thread(&ustconsumer64_data);
4533 if (ret) {
4534 errno = ret;
4535 PERROR("join_consumer ust64");
4536 }
4537 return NULL;
4538 }
4539
4540 static int string_match(const char *str1, const char *str2)
4541 {
4542 return (str1 && str2) && !strcmp(str1, str2);
4543 }
4544
4545 /*
4546 * Take an option from the getopt output and set it in the right variable to be
4547 * used later.
4548 *
4549 * Return 0 on success else a negative value.
4550 */
4551 static int set_option(int opt, const char *arg, const char *optname)
4552 {
4553 int ret = 0;
4554
4555 if (string_match(optname, "client-sock") || opt == 'c') {
4556 if (!arg || *arg == '\0') {
4557 ret = -EINVAL;
4558 goto end;
4559 }
4560 if (lttng_is_setuid_setgid()) {
4561 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4562 "-c, --client-sock");
4563 } else {
4564 config_string_set(&config.client_unix_sock_path,
4565 strdup(arg));
4566 if (!config.client_unix_sock_path.value) {
4567 ret = -ENOMEM;
4568 PERROR("strdup");
4569 }
4570 }
4571 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4572 if (!arg || *arg == '\0') {
4573 ret = -EINVAL;
4574 goto end;
4575 }
4576 if (lttng_is_setuid_setgid()) {
4577 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4578 "-a, --apps-sock");
4579 } else {
4580 config_string_set(&config.apps_unix_sock_path,
4581 strdup(arg));
4582 if (!config.apps_unix_sock_path.value) {
4583 ret = -ENOMEM;
4584 PERROR("strdup");
4585 }
4586 }
4587 } else if (string_match(optname, "daemonize") || opt == 'd') {
4588 config.daemonize = true;
4589 } else if (string_match(optname, "background") || opt == 'b') {
4590 config.background = true;
4591 } else if (string_match(optname, "group") || opt == 'g') {
4592 if (!arg || *arg == '\0') {
4593 ret = -EINVAL;
4594 goto end;
4595 }
4596 if (lttng_is_setuid_setgid()) {
4597 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4598 "-g, --group");
4599 } else {
4600 config_string_set(&config.tracing_group_name,
4601 strdup(arg));
4602 if (!config.tracing_group_name.value) {
4603 ret = -ENOMEM;
4604 PERROR("strdup");
4605 }
4606 }
4607 } else if (string_match(optname, "help") || opt == 'h') {
4608 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4609 if (ret) {
4610 ERR("Cannot show --help for `lttng-sessiond`");
4611 perror("exec");
4612 }
4613 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4614 } else if (string_match(optname, "version") || opt == 'V') {
4615 fprintf(stdout, "%s\n", VERSION);
4616 exit(EXIT_SUCCESS);
4617 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4618 config.sig_parent = true;
4619 } else if (string_match(optname, "kconsumerd-err-sock")) {
4620 if (!arg || *arg == '\0') {
4621 ret = -EINVAL;
4622 goto end;
4623 }
4624 if (lttng_is_setuid_setgid()) {
4625 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4626 "--kconsumerd-err-sock");
4627 } else {
4628 config_string_set(&config.kconsumerd_err_unix_sock_path,
4629 strdup(arg));
4630 if (!config.kconsumerd_err_unix_sock_path.value) {
4631 ret = -ENOMEM;
4632 PERROR("strdup");
4633 }
4634 }
4635 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4636 if (!arg || *arg == '\0') {
4637 ret = -EINVAL;
4638 goto end;
4639 }
4640 if (lttng_is_setuid_setgid()) {
4641 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4642 "--kconsumerd-cmd-sock");
4643 } else {
4644 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
4645 strdup(arg));
4646 if (!config.kconsumerd_cmd_unix_sock_path.value) {
4647 ret = -ENOMEM;
4648 PERROR("strdup");
4649 }
4650 }
4651 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4652 if (!arg || *arg == '\0') {
4653 ret = -EINVAL;
4654 goto end;
4655 }
4656 if (lttng_is_setuid_setgid()) {
4657 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4658 "--ustconsumerd64-err-sock");
4659 } else {
4660 config_string_set(&config.consumerd64_err_unix_sock_path,
4661 strdup(arg));
4662 if (!config.consumerd64_err_unix_sock_path.value) {
4663 ret = -ENOMEM;
4664 PERROR("strdup");
4665 }
4666 }
4667 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4668 if (!arg || *arg == '\0') {
4669 ret = -EINVAL;
4670 goto end;
4671 }
4672 if (lttng_is_setuid_setgid()) {
4673 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4674 "--ustconsumerd64-cmd-sock");
4675 } else {
4676 config_string_set(&config.consumerd64_cmd_unix_sock_path,
4677 strdup(arg));
4678 if (!config.consumerd64_cmd_unix_sock_path.value) {
4679 ret = -ENOMEM;
4680 PERROR("strdup");
4681 }
4682 }
4683 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4684 if (!arg || *arg == '\0') {
4685 ret = -EINVAL;
4686 goto end;
4687 }
4688 if (lttng_is_setuid_setgid()) {
4689 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4690 "--ustconsumerd32-err-sock");
4691 } else {
4692 config_string_set(&config.consumerd32_err_unix_sock_path,
4693 strdup(arg));
4694 if (!config.consumerd32_err_unix_sock_path.value) {
4695 ret = -ENOMEM;
4696 PERROR("strdup");
4697 }
4698 }
4699 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4700 if (!arg || *arg == '\0') {
4701 ret = -EINVAL;
4702 goto end;
4703 }
4704 if (lttng_is_setuid_setgid()) {
4705 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4706 "--ustconsumerd32-cmd-sock");
4707 } else {
4708 config_string_set(&config.consumerd32_cmd_unix_sock_path,
4709 strdup(arg));
4710 if (!config.consumerd32_cmd_unix_sock_path.value) {
4711 ret = -ENOMEM;
4712 PERROR("strdup");
4713 }
4714 }
4715 } else if (string_match(optname, "no-kernel")) {
4716 config.no_kernel = true;
4717 } else if (string_match(optname, "quiet") || opt == 'q') {
4718 lttng_opt_quiet = true;
4719 } else if (string_match(optname, "verbose") || opt == 'v') {
4720 /* Verbose level can increase using multiple -v */
4721 if (arg) {
4722 /* Value obtained from config file */
4723 config.verbose = config_parse_value(arg);
4724 } else {
4725 /* -v used on command line */
4726 config.verbose++;
4727 }
4728 /* Clamp value to [0, 3] */
4729 config.verbose = config.verbose < 0 ? 0 :
4730 (config.verbose <= 3 ? config.verbose : 3);
4731 } else if (string_match(optname, "verbose-consumer")) {
4732 if (arg) {
4733 config.verbose_consumer = config_parse_value(arg);
4734 } else {
4735 config.verbose_consumer++;
4736 }
4737 } else if (string_match(optname, "consumerd32-path")) {
4738 if (!arg || *arg == '\0') {
4739 ret = -EINVAL;
4740 goto end;
4741 }
4742 if (lttng_is_setuid_setgid()) {
4743 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4744 "--consumerd32-path");
4745 } else {
4746 config_string_set(&config.consumerd32_bin_path,
4747 strdup(arg));
4748 if (!config.consumerd32_bin_path.value) {
4749 PERROR("strdup");
4750 ret = -ENOMEM;
4751 }
4752 }
4753 } else if (string_match(optname, "consumerd32-libdir")) {
4754 if (!arg || *arg == '\0') {
4755 ret = -EINVAL;
4756 goto end;
4757 }
4758 if (lttng_is_setuid_setgid()) {
4759 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4760 "--consumerd32-libdir");
4761 } else {
4762 config_string_set(&config.consumerd32_lib_dir,
4763 strdup(arg));
4764 if (!config.consumerd32_lib_dir.value) {
4765 PERROR("strdup");
4766 ret = -ENOMEM;
4767 }
4768 }
4769 } else if (string_match(optname, "consumerd64-path")) {
4770 if (!arg || *arg == '\0') {
4771 ret = -EINVAL;
4772 goto end;
4773 }
4774 if (lttng_is_setuid_setgid()) {
4775 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4776 "--consumerd64-path");
4777 } else {
4778 config_string_set(&config.consumerd64_bin_path,
4779 strdup(arg));
4780 if (!config.consumerd64_bin_path.value) {
4781 PERROR("strdup");
4782 ret = -ENOMEM;
4783 }
4784 }
4785 } else if (string_match(optname, "consumerd64-libdir")) {
4786 if (!arg || *arg == '\0') {
4787 ret = -EINVAL;
4788 goto end;
4789 }
4790 if (lttng_is_setuid_setgid()) {
4791 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4792 "--consumerd64-libdir");
4793 } else {
4794 config_string_set(&config.consumerd64_lib_dir,
4795 strdup(arg));
4796 if (!config.consumerd64_lib_dir.value) {
4797 PERROR("strdup");
4798 ret = -ENOMEM;
4799 }
4800 }
4801 } else if (string_match(optname, "pidfile") || opt == 'p') {
4802 if (!arg || *arg == '\0') {
4803 ret = -EINVAL;
4804 goto end;
4805 }
4806 if (lttng_is_setuid_setgid()) {
4807 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4808 "-p, --pidfile");
4809 } else {
4810 config_string_set(&config.pid_file_path, strdup(arg));
4811 if (!config.pid_file_path.value) {
4812 PERROR("strdup");
4813 ret = -ENOMEM;
4814 }
4815 }
4816 } else if (string_match(optname, "agent-tcp-port")) {
4817 if (!arg || *arg == '\0') {
4818 ret = -EINVAL;
4819 goto end;
4820 }
4821 if (lttng_is_setuid_setgid()) {
4822 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4823 "--agent-tcp-port");
4824 } else {
4825 unsigned long v;
4826
4827 errno = 0;
4828 v = strtoul(arg, NULL, 0);
4829 if (errno != 0 || !isdigit(arg[0])) {
4830 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4831 return -1;
4832 }
4833 if (v == 0 || v >= 65535) {
4834 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4835 return -1;
4836 }
4837 config.agent_tcp_port = (uint32_t) v;
4838 DBG3("Agent TCP port set to non default: %u", config.agent_tcp_port);
4839 }
4840 } else if (string_match(optname, "load") || opt == 'l') {
4841 if (!arg || *arg == '\0') {
4842 ret = -EINVAL;
4843 goto end;
4844 }
4845 if (lttng_is_setuid_setgid()) {
4846 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4847 "-l, --load");
4848 } else {
4849 config_string_set(&config.load_session_path, strdup(arg));
4850 if (!config.load_session_path.value) {
4851 PERROR("strdup");
4852 ret = -ENOMEM;
4853 }
4854 }
4855 } else if (string_match(optname, "kmod-probes")) {
4856 if (!arg || *arg == '\0') {
4857 ret = -EINVAL;
4858 goto end;
4859 }
4860 if (lttng_is_setuid_setgid()) {
4861 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4862 "--kmod-probes");
4863 } else {
4864 config_string_set(&config.kmod_probes_list, strdup(arg));
4865 if (!config.kmod_probes_list.value) {
4866 PERROR("strdup");
4867 ret = -ENOMEM;
4868 }
4869 }
4870 } else if (string_match(optname, "extra-kmod-probes")) {
4871 if (!arg || *arg == '\0') {
4872 ret = -EINVAL;
4873 goto end;
4874 }
4875 if (lttng_is_setuid_setgid()) {
4876 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4877 "--extra-kmod-probes");
4878 } else {
4879 config_string_set(&config.kmod_extra_probes_list,
4880 strdup(arg));
4881 if (!config.kmod_extra_probes_list.value) {
4882 PERROR("strdup");
4883 ret = -ENOMEM;
4884 }
4885 }
4886 } else if (string_match(optname, "config") || opt == 'f') {
4887 /* This is handled in set_options() thus silent skip. */
4888 goto end;
4889 } else {
4890 /* Unknown option or other error.
4891 * Error is printed by getopt, just return */
4892 ret = -1;
4893 }
4894
4895 end:
4896 if (ret == -EINVAL) {
4897 const char *opt_name = "unknown";
4898 int i;
4899
4900 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4901 i++) {
4902 if (opt == long_options[i].val) {
4903 opt_name = long_options[i].name;
4904 break;
4905 }
4906 }
4907
4908 WARN("Invalid argument provided for option \"%s\", using default value.",
4909 opt_name);
4910 }
4911
4912 return ret;
4913 }
4914
4915 /*
4916 * config_entry_handler_cb used to handle options read from a config file.
4917 * See config_entry_handler_cb comment in common/config/session-config.h for the
4918 * return value conventions.
4919 */
4920 static int config_entry_handler(const struct config_entry *entry, void *unused)
4921 {
4922 int ret = 0, i;
4923
4924 if (!entry || !entry->name || !entry->value) {
4925 ret = -EINVAL;
4926 goto end;
4927 }
4928
4929 /* Check if the option is to be ignored */
4930 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4931 if (!strcmp(entry->name, config_ignore_options[i])) {
4932 goto end;
4933 }
4934 }
4935
4936 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4937 i++) {
4938
4939 /* Ignore if not fully matched. */
4940 if (strcmp(entry->name, long_options[i].name)) {
4941 continue;
4942 }
4943
4944 /*
4945 * If the option takes no argument on the command line, we have to
4946 * check if the value is "true". We support non-zero numeric values,
4947 * true, on and yes.
4948 */
4949 if (!long_options[i].has_arg) {
4950 ret = config_parse_value(entry->value);
4951 if (ret <= 0) {
4952 if (ret) {
4953 WARN("Invalid configuration value \"%s\" for option %s",
4954 entry->value, entry->name);
4955 }
4956 /* False, skip boolean config option. */
4957 goto end;
4958 }
4959 }
4960
4961 ret = set_option(long_options[i].val, entry->value, entry->name);
4962 goto end;
4963 }
4964
4965 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4966
4967 end:
4968 return ret;
4969 }
4970
4971 /*
4972 * daemon configuration loading and argument parsing
4973 */
4974 static int set_options(int argc, char **argv)
4975 {
4976 int ret = 0, c = 0, option_index = 0;
4977 int orig_optopt = optopt, orig_optind = optind;
4978 char *optstring;
4979 const char *config_path = NULL;
4980
4981 optstring = utils_generate_optstring(long_options,
4982 sizeof(long_options) / sizeof(struct option));
4983 if (!optstring) {
4984 ret = -ENOMEM;
4985 goto end;
4986 }
4987
4988 /* Check for the --config option */
4989 while ((c = getopt_long(argc, argv, optstring, long_options,
4990 &option_index)) != -1) {
4991 if (c == '?') {
4992 ret = -EINVAL;
4993 goto end;
4994 } else if (c != 'f') {
4995 /* if not equal to --config option. */
4996 continue;
4997 }
4998
4999 if (lttng_is_setuid_setgid()) {
5000 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5001 "-f, --config");
5002 } else {
5003 config_path = utils_expand_path(optarg);
5004 if (!config_path) {
5005 ERR("Failed to resolve path: %s", optarg);
5006 }
5007 }
5008 }
5009
5010 ret = config_get_section_entries(config_path, config_section_name,
5011 config_entry_handler, NULL);
5012 if (ret) {
5013 if (ret > 0) {
5014 ERR("Invalid configuration option at line %i", ret);
5015 ret = -1;
5016 }
5017 goto end;
5018 }
5019
5020 /* Reset getopt's global state */
5021 optopt = orig_optopt;
5022 optind = orig_optind;
5023 while (1) {
5024 option_index = -1;
5025 /*
5026 * getopt_long() will not set option_index if it encounters a
5027 * short option.
5028 */
5029 c = getopt_long(argc, argv, optstring, long_options,
5030 &option_index);
5031 if (c == -1) {
5032 break;
5033 }
5034
5035 /*
5036 * Pass NULL as the long option name if popt left the index
5037 * unset.
5038 */
5039 ret = set_option(c, optarg,
5040 option_index < 0 ? NULL :
5041 long_options[option_index].name);
5042 if (ret < 0) {
5043 break;
5044 }
5045 }
5046
5047 end:
5048 free(optstring);
5049 return ret;
5050 }
5051
5052 /*
5053 * Creates the two needed socket by the daemon.
5054 * apps_sock - The communication socket for all UST apps.
5055 * client_sock - The communication of the cli tool (lttng).
5056 */
5057 static int init_daemon_socket(void)
5058 {
5059 int ret = 0;
5060 mode_t old_umask;
5061
5062 old_umask = umask(0);
5063
5064 /* Create client tool unix socket */
5065 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5066 if (client_sock < 0) {
5067 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5068 ret = -1;
5069 goto end;
5070 }
5071
5072 /* Set the cloexec flag */
5073 ret = utils_set_fd_cloexec(client_sock);
5074 if (ret < 0) {
5075 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5076 "Continuing but note that the consumer daemon will have a "
5077 "reference to this socket on exec()", client_sock);
5078 }
5079
5080 /* File permission MUST be 660 */
5081 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5082 if (ret < 0) {
5083 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5084 PERROR("chmod");
5085 goto end;
5086 }
5087
5088 /* Create the application unix socket */
5089 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5090 if (apps_sock < 0) {
5091 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5092 ret = -1;
5093 goto end;
5094 }
5095
5096 /* Set the cloexec flag */
5097 ret = utils_set_fd_cloexec(apps_sock);
5098 if (ret < 0) {
5099 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5100 "Continuing but note that the consumer daemon will have a "
5101 "reference to this socket on exec()", apps_sock);
5102 }
5103
5104 /* File permission MUST be 666 */
5105 ret = chmod(config.apps_unix_sock_path.value,
5106 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5107 if (ret < 0) {
5108 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5109 PERROR("chmod");
5110 goto end;
5111 }
5112
5113 DBG3("Session daemon client socket %d and application socket %d created",
5114 client_sock, apps_sock);
5115
5116 end:
5117 umask(old_umask);
5118 return ret;
5119 }
5120
5121 /*
5122 * Check if the global socket is available, and if a daemon is answering at the
5123 * other side. If yes, error is returned.
5124 */
5125 static int check_existing_daemon(void)
5126 {
5127 /* Is there anybody out there ? */
5128 if (lttng_session_daemon_alive()) {
5129 return -EEXIST;
5130 }
5131
5132 return 0;
5133 }
5134
5135 /*
5136 * Set the tracing group gid onto the client socket.
5137 *
5138 * Race window between mkdir and chown is OK because we are going from more
5139 * permissive (root.root) to less permissive (root.tracing).
5140 */
5141 static int set_permissions(char *rundir)
5142 {
5143 int ret;
5144 gid_t gid;
5145
5146 gid = utils_get_group_id(config.tracing_group_name.value);
5147
5148 /* Set lttng run dir */
5149 ret = chown(rundir, 0, gid);
5150 if (ret < 0) {
5151 ERR("Unable to set group on %s", rundir);
5152 PERROR("chown");
5153 }
5154
5155 /*
5156 * Ensure all applications and tracing group can search the run
5157 * dir. Allow everyone to read the directory, since it does not
5158 * buy us anything to hide its content.
5159 */
5160 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5161 if (ret < 0) {
5162 ERR("Unable to set permissions on %s", rundir);
5163 PERROR("chmod");
5164 }
5165
5166 /* lttng client socket path */
5167 ret = chown(config.client_unix_sock_path.value, 0, gid);
5168 if (ret < 0) {
5169 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5170 PERROR("chown");
5171 }
5172
5173 /* kconsumer error socket path */
5174 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5175 if (ret < 0) {
5176 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5177 PERROR("chown");
5178 }
5179
5180 /* 64-bit ustconsumer error socket path */
5181 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5182 if (ret < 0) {
5183 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5184 PERROR("chown");
5185 }
5186
5187 /* 32-bit ustconsumer compat32 error socket path */
5188 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5189 if (ret < 0) {
5190 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5191 PERROR("chown");
5192 }
5193
5194 DBG("All permissions are set");
5195
5196 return ret;
5197 }
5198
5199 /*
5200 * Create the lttng run directory needed for all global sockets and pipe.
5201 */
5202 static int create_lttng_rundir(void)
5203 {
5204 int ret;
5205
5206 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5207
5208 ret = mkdir(config.rundir.value, S_IRWXU);
5209 if (ret < 0) {
5210 if (errno != EEXIST) {
5211 ERR("Unable to create %s", config.rundir.value);
5212 goto error;
5213 } else {
5214 ret = 0;
5215 }
5216 }
5217
5218 error:
5219 return ret;
5220 }
5221
5222 /*
5223 * Setup sockets and directory needed by the consumerds' communication with the
5224 * session daemon.
5225 */
5226 static int set_consumer_sockets(struct consumer_data *consumer_data)
5227 {
5228 int ret;
5229 char *path = NULL;
5230
5231 switch (consumer_data->type) {
5232 case LTTNG_CONSUMER_KERNEL:
5233 path = config.kconsumerd_path.value;
5234 break;
5235 case LTTNG_CONSUMER64_UST:
5236 path = config.consumerd64_path.value;
5237 break;
5238 case LTTNG_CONSUMER32_UST:
5239 path = config.consumerd32_path.value;
5240 break;
5241 default:
5242 ERR("Consumer type unknown");
5243 ret = -EINVAL;
5244 goto error;
5245 }
5246 assert(path);
5247
5248 DBG2("Creating consumer directory: %s", path);
5249
5250 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5251 if (ret < 0 && errno != EEXIST) {
5252 PERROR("mkdir");
5253 ERR("Failed to create %s", path);
5254 goto error;
5255 }
5256 if (is_root) {
5257 ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value));
5258 if (ret < 0) {
5259 ERR("Unable to set group on %s", path);
5260 PERROR("chown");
5261 goto error;
5262 }
5263 }
5264
5265 /* Create the consumerd error unix socket */
5266 consumer_data->err_sock =
5267 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5268 if (consumer_data->err_sock < 0) {
5269 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5270 ret = -1;
5271 goto error;
5272 }
5273
5274 /*
5275 * Set the CLOEXEC flag. Return code is useless because either way, the
5276 * show must go on.
5277 */
5278 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5279 if (ret < 0) {
5280 PERROR("utils_set_fd_cloexec");
5281 /* continue anyway */
5282 }
5283
5284 /* File permission MUST be 660 */
5285 ret = chmod(consumer_data->err_unix_sock_path,
5286 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5287 if (ret < 0) {
5288 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5289 PERROR("chmod");
5290 goto error;
5291 }
5292
5293 error:
5294 return ret;
5295 }
5296
5297 /*
5298 * Signal handler for the daemon
5299 *
5300 * Simply stop all worker threads, leaving main() return gracefully after
5301 * joining all threads and calling cleanup().
5302 */
5303 static void sighandler(int sig)
5304 {
5305 switch (sig) {
5306 case SIGINT:
5307 DBG("SIGINT caught");
5308 stop_threads();
5309 break;
5310 case SIGTERM:
5311 DBG("SIGTERM caught");
5312 stop_threads();
5313 break;
5314 case SIGUSR1:
5315 CMM_STORE_SHARED(recv_child_signal, 1);
5316 break;
5317 default:
5318 break;
5319 }
5320 }
5321
5322 /*
5323 * Setup signal handler for :
5324 * SIGINT, SIGTERM, SIGPIPE
5325 */
5326 static int set_signal_handler(void)
5327 {
5328 int ret = 0;
5329 struct sigaction sa;
5330 sigset_t sigset;
5331
5332 if ((ret = sigemptyset(&sigset)) < 0) {
5333 PERROR("sigemptyset");
5334 return ret;
5335 }
5336
5337 sa.sa_mask = sigset;
5338 sa.sa_flags = 0;
5339
5340 sa.sa_handler = sighandler;
5341 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5342 PERROR("sigaction");
5343 return ret;
5344 }
5345
5346 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5347 PERROR("sigaction");
5348 return ret;
5349 }
5350
5351 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5352 PERROR("sigaction");
5353 return ret;
5354 }
5355
5356 sa.sa_handler = SIG_IGN;
5357 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5358 PERROR("sigaction");
5359 return ret;
5360 }
5361
5362 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5363
5364 return ret;
5365 }
5366
5367 /*
5368 * Set open files limit to unlimited. This daemon can open a large number of
5369 * file descriptors in order to consume multiple kernel traces.
5370 */
5371 static void set_ulimit(void)
5372 {
5373 int ret;
5374 struct rlimit lim;
5375
5376 /* The kernel does not allow an infinite limit for open files */
5377 lim.rlim_cur = 65535;
5378 lim.rlim_max = 65535;
5379
5380 ret = setrlimit(RLIMIT_NOFILE, &lim);
5381 if (ret < 0) {
5382 PERROR("failed to set open files limit");
5383 }
5384 }
5385
5386 static int write_pidfile(void)
5387 {
5388 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5389 }
5390
5391 /*
5392 * Create lockfile using the rundir and return its fd.
5393 */
5394 static int create_lockfile(void)
5395 {
5396 return utils_create_lock_file(config.lock_file_path.value);
5397 }
5398
5399 /*
5400 * Write agent TCP port using the rundir.
5401 */
5402 static int write_agent_port(void)
5403 {
5404 return utils_create_pid_file(config.agent_tcp_port,
5405 config.agent_port_file_path.value);
5406 }
5407
5408 static int set_clock_plugin_env(void)
5409 {
5410 int ret = 0;
5411 char *env_value = NULL;
5412
5413 if (!config.lttng_ust_clock_plugin.value) {
5414 goto end;
5415 }
5416
5417 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5418 config.lttng_ust_clock_plugin.value);
5419 if (ret < 0) {
5420 PERROR("asprintf");
5421 goto end;
5422 }
5423
5424 ret = putenv(env_value);
5425 if (ret) {
5426 free(env_value);
5427 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5428 goto end;
5429 }
5430
5431 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5432 config.lttng_ust_clock_plugin.value);
5433 end:
5434 return ret;
5435 }
5436
5437 /*
5438 * main
5439 */
5440 int main(int argc, char **argv)
5441 {
5442 int ret = 0, retval = 0;
5443 void *status;
5444 const char *env_app_timeout;
5445 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5446 *ust64_channel_monitor_pipe = NULL,
5447 *kernel_channel_monitor_pipe = NULL;
5448 bool notification_thread_running = false;
5449
5450 init_kernel_workarounds();
5451
5452 rcu_register_thread();
5453
5454 if (set_signal_handler()) {
5455 retval = -1;
5456 goto exit_set_signal_handler;
5457 }
5458
5459 page_size = sysconf(_SC_PAGESIZE);
5460 if (page_size < 0) {
5461 PERROR("sysconf _SC_PAGESIZE");
5462 page_size = LONG_MAX;
5463 WARN("Fallback page size to %ld", page_size);
5464 }
5465
5466 ret = sessiond_config_init(&config);
5467 if (ret) {
5468 retval = -1;
5469 goto exit_set_signal_handler;
5470 }
5471
5472 /*
5473 * Parse arguments and load the daemon configuration file.
5474 *
5475 * We have an exit_options exit path to free memory reserved by
5476 * set_options. This is needed because the rest of sessiond_cleanup()
5477 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5478 * depends on set_options.
5479 */
5480 progname = argv[0];
5481 if (set_options(argc, argv)) {
5482 retval = -1;
5483 goto exit_options;
5484 }
5485
5486 /* Init config from environment variables. */
5487 sessiond_config_apply_env_config(&config);
5488
5489 /*
5490 * Resolve all paths received as arguments, configuration option, or
5491 * through environment variable as absolute paths. This is necessary
5492 * since daemonizing causes the sessiond's current working directory
5493 * to '/'.
5494 */
5495 ret = sessiond_config_resolve_paths(&config);
5496 if (ret) {
5497 goto exit_options;
5498 }
5499
5500 /* Apply config. */
5501 lttng_opt_verbose = config.verbose;
5502 lttng_opt_quiet = config.quiet;
5503 kconsumer_data.err_unix_sock_path =
5504 config.kconsumerd_err_unix_sock_path.value;
5505 kconsumer_data.cmd_unix_sock_path =
5506 config.kconsumerd_cmd_unix_sock_path.value;
5507 ustconsumer32_data.err_unix_sock_path =
5508 config.consumerd32_err_unix_sock_path.value;
5509 ustconsumer32_data.cmd_unix_sock_path =
5510 config.consumerd32_cmd_unix_sock_path.value;
5511 ustconsumer64_data.err_unix_sock_path =
5512 config.consumerd64_err_unix_sock_path.value;
5513 ustconsumer64_data.cmd_unix_sock_path =
5514 config.consumerd64_cmd_unix_sock_path.value;
5515 set_clock_plugin_env();
5516
5517 sessiond_config_log(&config);
5518
5519 /* Daemonize */
5520 if (config.daemonize || config.background) {
5521 int i;
5522
5523 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5524 !config.background);
5525 if (ret < 0) {
5526 retval = -1;
5527 goto exit_options;
5528 }
5529
5530 /*
5531 * We are in the child. Make sure all other file descriptors are
5532 * closed, in case we are called with more opened file
5533 * descriptors than the standard ones.
5534 */
5535 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5536 (void) close(i);
5537 }
5538 }
5539
5540 if (run_as_create_worker(argv[0]) < 0) {
5541 goto exit_create_run_as_worker_cleanup;
5542 }
5543
5544 /*
5545 * Starting from here, we can create threads. This needs to be after
5546 * lttng_daemonize due to RCU.
5547 */
5548
5549 /*
5550 * Initialize the health check subsystem. This call should set the
5551 * appropriate time values.
5552 */
5553 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5554 if (!health_sessiond) {
5555 PERROR("health_app_create error");
5556 retval = -1;
5557 goto exit_health_sessiond_cleanup;
5558 }
5559
5560 /* Create thread to clean up RCU hash tables */
5561 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5562 retval = -1;
5563 goto exit_ht_cleanup;
5564 }
5565
5566 /* Create thread quit pipe */
5567 if (init_thread_quit_pipe()) {
5568 retval = -1;
5569 goto exit_init_data;
5570 }
5571
5572 /* Check if daemon is UID = 0 */
5573 is_root = !getuid();
5574
5575 if (create_lttng_rundir()) {
5576 retval = -1;
5577 goto exit_init_data;
5578 }
5579
5580 if (is_root) {
5581 /* Create global run dir with root access */
5582
5583 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5584 if (!kernel_channel_monitor_pipe) {
5585 ERR("Failed to create kernel consumer channel monitor pipe");
5586 retval = -1;
5587 goto exit_init_data;
5588 }
5589 kconsumer_data.channel_monitor_pipe =
5590 lttng_pipe_release_writefd(
5591 kernel_channel_monitor_pipe);
5592 if (kconsumer_data.channel_monitor_pipe < 0) {
5593 retval = -1;
5594 goto exit_init_data;
5595 }
5596 }
5597
5598 lockfile_fd = create_lockfile();
5599 if (lockfile_fd < 0) {
5600 retval = -1;
5601 goto exit_init_data;
5602 }
5603
5604 /* Set consumer initial state */
5605 kernel_consumerd_state = CONSUMER_STOPPED;
5606 ust_consumerd_state = CONSUMER_STOPPED;
5607
5608 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5609 if (!ust32_channel_monitor_pipe) {
5610 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5611 retval = -1;
5612 goto exit_init_data;
5613 }
5614 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5615 ust32_channel_monitor_pipe);
5616 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5617 retval = -1;
5618 goto exit_init_data;
5619 }
5620
5621 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5622 if (!ust64_channel_monitor_pipe) {
5623 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5624 retval = -1;
5625 goto exit_init_data;
5626 }
5627 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5628 ust64_channel_monitor_pipe);
5629 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5630 retval = -1;
5631 goto exit_init_data;
5632 }
5633
5634 /*
5635 * See if daemon already exist.
5636 */
5637 if (check_existing_daemon()) {
5638 ERR("Already running daemon.\n");
5639 /*
5640 * We do not goto exit because we must not cleanup()
5641 * because a daemon is already running.
5642 */
5643 retval = -1;
5644 goto exit_init_data;
5645 }
5646
5647 /*
5648 * Init UST app hash table. Alloc hash table before this point since
5649 * cleanup() can get called after that point.
5650 */
5651 if (ust_app_ht_alloc()) {
5652 ERR("Failed to allocate UST app hash table");
5653 retval = -1;
5654 goto exit_init_data;
5655 }
5656
5657 /*
5658 * Initialize agent app hash table. We allocate the hash table here
5659 * since cleanup() can get called after this point.
5660 */
5661 if (agent_app_ht_alloc()) {
5662 ERR("Failed to allocate Agent app hash table");
5663 retval = -1;
5664 goto exit_init_data;
5665 }
5666
5667 /*
5668 * These actions must be executed as root. We do that *after* setting up
5669 * the sockets path because we MUST make the check for another daemon using
5670 * those paths *before* trying to set the kernel consumer sockets and init
5671 * kernel tracer.
5672 */
5673 if (is_root) {
5674 if (set_consumer_sockets(&kconsumer_data)) {
5675 retval = -1;
5676 goto exit_init_data;
5677 }
5678
5679 /* Setup kernel tracer */
5680 if (!config.no_kernel) {
5681 init_kernel_tracer();
5682 if (kernel_tracer_fd >= 0) {
5683 ret = syscall_init_table();
5684 if (ret < 0) {
5685 ERR("Unable to populate syscall table. "
5686 "Syscall tracing won't work "
5687 "for this session daemon.");
5688 }
5689 }
5690 }
5691
5692 /* Set ulimit for open files */
5693 set_ulimit();
5694 }
5695 /* init lttng_fd tracking must be done after set_ulimit. */
5696 lttng_fd_init();
5697
5698 if (set_consumer_sockets(&ustconsumer64_data)) {
5699 retval = -1;
5700 goto exit_init_data;
5701 }
5702
5703 if (set_consumer_sockets(&ustconsumer32_data)) {
5704 retval = -1;
5705 goto exit_init_data;
5706 }
5707
5708 /* Setup the needed unix socket */
5709 if (init_daemon_socket()) {
5710 retval = -1;
5711 goto exit_init_data;
5712 }
5713
5714 /* Set credentials to socket */
5715 if (is_root && set_permissions(config.rundir.value)) {
5716 retval = -1;
5717 goto exit_init_data;
5718 }
5719
5720 /* Get parent pid if -S, --sig-parent is specified. */
5721 if (config.sig_parent) {
5722 ppid = getppid();
5723 }
5724
5725 /* Setup the kernel pipe for waking up the kernel thread */
5726 if (is_root && !config.no_kernel) {
5727 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5728 retval = -1;
5729 goto exit_init_data;
5730 }
5731 }
5732
5733 /* Setup the thread apps communication pipe. */
5734 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5735 retval = -1;
5736 goto exit_init_data;
5737 }
5738
5739 /* Setup the thread apps notify communication pipe. */
5740 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5741 retval = -1;
5742 goto exit_init_data;
5743 }
5744
5745 /* Initialize global buffer per UID and PID registry. */
5746 buffer_reg_init_uid_registry();
5747 buffer_reg_init_pid_registry();
5748
5749 /* Init UST command queue. */
5750 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5751
5752 /*
5753 * Get session list pointer. This pointer MUST NOT be free'd. This list
5754 * is statically declared in session.c
5755 */
5756 session_list_ptr = session_get_list();
5757
5758 cmd_init();
5759
5760 /* Check for the application socket timeout env variable. */
5761 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5762 if (env_app_timeout) {
5763 config.app_socket_timeout = atoi(env_app_timeout);
5764 } else {
5765 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5766 }
5767
5768 ret = write_pidfile();
5769 if (ret) {
5770 ERR("Error in write_pidfile");
5771 retval = -1;
5772 goto exit_init_data;
5773 }
5774 ret = write_agent_port();
5775 if (ret) {
5776 ERR("Error in write_agent_port");
5777 retval = -1;
5778 goto exit_init_data;
5779 }
5780
5781 /* Initialize communication library */
5782 lttcomm_init();
5783 /* Initialize TCP timeout values */
5784 lttcomm_inet_init();
5785
5786 if (load_session_init_data(&load_info) < 0) {
5787 retval = -1;
5788 goto exit_init_data;
5789 }
5790 load_info->path = config.load_session_path.value;
5791
5792 /* Create health-check thread. */
5793 ret = pthread_create(&health_thread, default_pthread_attr(),
5794 thread_manage_health, (void *) NULL);
5795 if (ret) {
5796 errno = ret;
5797 PERROR("pthread_create health");
5798 retval = -1;
5799 goto exit_health;
5800 }
5801
5802 /* notification_thread_data acquires the pipes' read side. */
5803 notification_thread_handle = notification_thread_handle_create(
5804 ust32_channel_monitor_pipe,
5805 ust64_channel_monitor_pipe,
5806 kernel_channel_monitor_pipe);
5807 if (!notification_thread_handle) {
5808 retval = -1;
5809 ERR("Failed to create notification thread shared data");
5810 stop_threads();
5811 goto exit_notification;
5812 }
5813
5814 /* Create notification thread. */
5815 ret = pthread_create(&notification_thread, default_pthread_attr(),
5816 thread_notification, notification_thread_handle);
5817 if (ret) {
5818 errno = ret;
5819 PERROR("pthread_create notification");
5820 retval = -1;
5821 stop_threads();
5822 goto exit_notification;
5823 }
5824 notification_thread_running = true;
5825
5826 /* Create thread to manage the client socket */
5827 ret = pthread_create(&client_thread, default_pthread_attr(),
5828 thread_manage_clients, (void *) NULL);
5829 if (ret) {
5830 errno = ret;
5831 PERROR("pthread_create clients");
5832 retval = -1;
5833 stop_threads();
5834 goto exit_client;
5835 }
5836
5837 /* Create thread to dispatch registration */
5838 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
5839 thread_dispatch_ust_registration, (void *) NULL);
5840 if (ret) {
5841 errno = ret;
5842 PERROR("pthread_create dispatch");
5843 retval = -1;
5844 stop_threads();
5845 goto exit_dispatch;
5846 }
5847
5848 /* Create thread to manage application registration. */
5849 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
5850 thread_registration_apps, (void *) NULL);
5851 if (ret) {
5852 errno = ret;
5853 PERROR("pthread_create registration");
5854 retval = -1;
5855 stop_threads();
5856 goto exit_reg_apps;
5857 }
5858
5859 /* Create thread to manage application socket */
5860 ret = pthread_create(&apps_thread, default_pthread_attr(),
5861 thread_manage_apps, (void *) NULL);
5862 if (ret) {
5863 errno = ret;
5864 PERROR("pthread_create apps");
5865 retval = -1;
5866 stop_threads();
5867 goto exit_apps;
5868 }
5869
5870 /* Create thread to manage application notify socket */
5871 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
5872 ust_thread_manage_notify, (void *) NULL);
5873 if (ret) {
5874 errno = ret;
5875 PERROR("pthread_create notify");
5876 retval = -1;
5877 stop_threads();
5878 goto exit_apps_notify;
5879 }
5880
5881 /* Create agent registration thread. */
5882 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
5883 agent_thread_manage_registration, (void *) NULL);
5884 if (ret) {
5885 errno = ret;
5886 PERROR("pthread_create agent");
5887 retval = -1;
5888 stop_threads();
5889 goto exit_agent_reg;
5890 }
5891
5892 /* Don't start this thread if kernel tracing is not requested nor root */
5893 if (is_root && !config.no_kernel) {
5894 /* Create kernel thread to manage kernel event */
5895 ret = pthread_create(&kernel_thread, default_pthread_attr(),
5896 thread_manage_kernel, (void *) NULL);
5897 if (ret) {
5898 errno = ret;
5899 PERROR("pthread_create kernel");
5900 retval = -1;
5901 stop_threads();
5902 goto exit_kernel;
5903 }
5904 }
5905
5906 /* Create session loading thread. */
5907 ret = pthread_create(&load_session_thread, default_pthread_attr(),
5908 thread_load_session, load_info);
5909 if (ret) {
5910 errno = ret;
5911 PERROR("pthread_create load_session_thread");
5912 retval = -1;
5913 stop_threads();
5914 goto exit_load_session;
5915 }
5916
5917 /*
5918 * This is where we start awaiting program completion (e.g. through
5919 * signal that asks threads to teardown).
5920 */
5921
5922 ret = pthread_join(load_session_thread, &status);
5923 if (ret) {
5924 errno = ret;
5925 PERROR("pthread_join load_session_thread");
5926 retval = -1;
5927 }
5928 exit_load_session:
5929
5930 if (is_root && !config.no_kernel) {
5931 ret = pthread_join(kernel_thread, &status);
5932 if (ret) {
5933 errno = ret;
5934 PERROR("pthread_join");
5935 retval = -1;
5936 }
5937 }
5938 exit_kernel:
5939
5940 ret = pthread_join(agent_reg_thread, &status);
5941 if (ret) {
5942 errno = ret;
5943 PERROR("pthread_join agent");
5944 retval = -1;
5945 }
5946 exit_agent_reg:
5947
5948 ret = pthread_join(apps_notify_thread, &status);
5949 if (ret) {
5950 errno = ret;
5951 PERROR("pthread_join apps notify");
5952 retval = -1;
5953 }
5954 exit_apps_notify:
5955
5956 ret = pthread_join(apps_thread, &status);
5957 if (ret) {
5958 errno = ret;
5959 PERROR("pthread_join apps");
5960 retval = -1;
5961 }
5962 exit_apps:
5963
5964 ret = pthread_join(reg_apps_thread, &status);
5965 if (ret) {
5966 errno = ret;
5967 PERROR("pthread_join");
5968 retval = -1;
5969 }
5970 exit_reg_apps:
5971
5972 /*
5973 * Join dispatch thread after joining reg_apps_thread to ensure
5974 * we don't leak applications in the queue.
5975 */
5976 ret = pthread_join(dispatch_thread, &status);
5977 if (ret) {
5978 errno = ret;
5979 PERROR("pthread_join");
5980 retval = -1;
5981 }
5982 exit_dispatch:
5983
5984 ret = pthread_join(client_thread, &status);
5985 if (ret) {
5986 errno = ret;
5987 PERROR("pthread_join");
5988 retval = -1;
5989 }
5990
5991 exit_client:
5992 exit_notification:
5993 ret = pthread_join(health_thread, &status);
5994 if (ret) {
5995 errno = ret;
5996 PERROR("pthread_join health thread");
5997 retval = -1;
5998 }
5999
6000 exit_health:
6001 exit_init_data:
6002 /*
6003 * Wait for all pending call_rcu work to complete before tearing
6004 * down data structures. call_rcu worker may be trying to
6005 * perform lookups in those structures.
6006 */
6007 rcu_barrier();
6008 /*
6009 * sessiond_cleanup() is called when no other thread is running, except
6010 * the ht_cleanup thread, which is needed to destroy the hash tables.
6011 */
6012 rcu_thread_online();
6013 sessiond_cleanup();
6014
6015 /*
6016 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6017 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6018 * the queue is empty before shutting down the clean-up thread.
6019 */
6020 rcu_barrier();
6021
6022 /*
6023 * The teardown of the notification system is performed after the
6024 * session daemon's teardown in order to allow it to be notified
6025 * of the active session and channels at the moment of the teardown.
6026 */
6027 if (notification_thread_handle) {
6028 if (notification_thread_running) {
6029 notification_thread_command_quit(
6030 notification_thread_handle);
6031 ret = pthread_join(notification_thread, &status);
6032 if (ret) {
6033 errno = ret;
6034 PERROR("pthread_join notification thread");
6035 retval = -1;
6036 }
6037 }
6038 notification_thread_handle_destroy(notification_thread_handle);
6039 }
6040
6041 rcu_thread_offline();
6042 rcu_unregister_thread();
6043
6044 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6045 if (ret) {
6046 retval = -1;
6047 }
6048 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6049 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6050 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6051 exit_ht_cleanup:
6052
6053 health_app_destroy(health_sessiond);
6054 exit_health_sessiond_cleanup:
6055 exit_create_run_as_worker_cleanup:
6056
6057 exit_options:
6058 sessiond_cleanup_options();
6059
6060 exit_set_signal_handler:
6061 if (!retval) {
6062 exit(EXIT_SUCCESS);
6063 } else {
6064 exit(EXIT_FAILURE);
6065 }
6066 }
This page took 0.248408 seconds and 5 git commands to generate.