Fix: using putenv() and free()-ing the value is invalid
[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 char *tmpnew = NULL;
2444
2445 if (config.consumerd64_lib_dir.value) {
2446 char *tmp;
2447 size_t tmplen;
2448
2449 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2450 if (!tmp) {
2451 tmp = "";
2452 }
2453 tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2454 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2455 if (!tmpnew) {
2456 ret = -ENOMEM;
2457 goto error;
2458 }
2459 strcat(tmpnew, config.consumerd64_lib_dir.value);
2460 if (tmp[0] != '\0') {
2461 strcat(tmpnew, ":");
2462 strcat(tmpnew, tmp);
2463 }
2464 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2465 if (ret) {
2466 ret = -errno;
2467 free(tmpnew);
2468 goto error;
2469 }
2470 }
2471 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2472 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2473 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2474 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2475 "--group", config.tracing_group_name.value,
2476 NULL);
2477 free(tmpnew);
2478 break;
2479 }
2480 case LTTNG_CONSUMER32_UST:
2481 {
2482 char *tmpnew = NULL;
2483
2484 if (config.consumerd32_lib_dir.value) {
2485 char *tmp;
2486 size_t tmplen;
2487
2488 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2489 if (!tmp) {
2490 tmp = "";
2491 }
2492 tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2493 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2494 if (!tmpnew) {
2495 ret = -ENOMEM;
2496 goto error;
2497 }
2498 strcat(tmpnew, config.consumerd32_lib_dir.value);
2499 if (tmp[0] != '\0') {
2500 strcat(tmpnew, ":");
2501 strcat(tmpnew, tmp);
2502 }
2503 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2504 if (ret) {
2505 ret = -errno;
2506 free(tmpnew);
2507 goto error;
2508 }
2509 }
2510 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2511 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2512 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2513 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2514 "--group", config.tracing_group_name.value,
2515 NULL);
2516 free(tmpnew);
2517 break;
2518 }
2519 default:
2520 ERR("unknown consumer type");
2521 errno = 0;
2522 }
2523 if (errno != 0) {
2524 PERROR("Consumer execl()");
2525 }
2526 /* Reaching this point, we got a failure on our execl(). */
2527 exit(EXIT_FAILURE);
2528 } else if (pid > 0) {
2529 ret = pid;
2530 } else {
2531 PERROR("start consumer fork");
2532 ret = -errno;
2533 }
2534 error:
2535 return ret;
2536 }
2537
2538 /*
2539 * Spawn the consumerd daemon and session daemon thread.
2540 */
2541 static int start_consumerd(struct consumer_data *consumer_data)
2542 {
2543 int ret;
2544
2545 /*
2546 * Set the listen() state on the socket since there is a possible race
2547 * between the exec() of the consumer daemon and this call if place in the
2548 * consumer thread. See bug #366 for more details.
2549 */
2550 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2551 if (ret < 0) {
2552 goto error;
2553 }
2554
2555 pthread_mutex_lock(&consumer_data->pid_mutex);
2556 if (consumer_data->pid != 0) {
2557 pthread_mutex_unlock(&consumer_data->pid_mutex);
2558 goto end;
2559 }
2560
2561 ret = spawn_consumerd(consumer_data);
2562 if (ret < 0) {
2563 ERR("Spawning consumerd failed");
2564 pthread_mutex_unlock(&consumer_data->pid_mutex);
2565 goto error;
2566 }
2567
2568 /* Setting up the consumer_data pid */
2569 consumer_data->pid = ret;
2570 DBG2("Consumer pid %d", consumer_data->pid);
2571 pthread_mutex_unlock(&consumer_data->pid_mutex);
2572
2573 DBG2("Spawning consumer control thread");
2574 ret = spawn_consumer_thread(consumer_data);
2575 if (ret < 0) {
2576 ERR("Fatal error spawning consumer control thread");
2577 goto error;
2578 }
2579
2580 end:
2581 return 0;
2582
2583 error:
2584 /* Cleanup already created sockets on error. */
2585 if (consumer_data->err_sock >= 0) {
2586 int err;
2587
2588 err = close(consumer_data->err_sock);
2589 if (err < 0) {
2590 PERROR("close consumer data error socket");
2591 }
2592 }
2593 return ret;
2594 }
2595
2596 /*
2597 * Setup necessary data for kernel tracer action.
2598 */
2599 static int init_kernel_tracer(void)
2600 {
2601 int ret;
2602
2603 /* Modprobe lttng kernel modules */
2604 ret = modprobe_lttng_control();
2605 if (ret < 0) {
2606 goto error;
2607 }
2608
2609 /* Open debugfs lttng */
2610 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2611 if (kernel_tracer_fd < 0) {
2612 DBG("Failed to open %s", module_proc_lttng);
2613 goto error_open;
2614 }
2615
2616 /* Validate kernel version */
2617 ret = kernel_validate_version(kernel_tracer_fd);
2618 if (ret < 0) {
2619 goto error_version;
2620 }
2621
2622 ret = modprobe_lttng_data();
2623 if (ret < 0) {
2624 goto error_modules;
2625 }
2626
2627 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2628 kernel_tracer_fd);
2629 if (ret < 0) {
2630 goto error_modules;
2631 }
2632
2633 if (ret < 1) {
2634 WARN("Kernel tracer does not support buffer monitoring. "
2635 "The monitoring timer of channels in the kernel domain "
2636 "will be set to 0 (disabled).");
2637 }
2638
2639 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2640 return 0;
2641
2642 error_version:
2643 modprobe_remove_lttng_control();
2644 ret = close(kernel_tracer_fd);
2645 if (ret) {
2646 PERROR("close");
2647 }
2648 kernel_tracer_fd = -1;
2649 return LTTNG_ERR_KERN_VERSION;
2650
2651 error_modules:
2652 ret = close(kernel_tracer_fd);
2653 if (ret) {
2654 PERROR("close");
2655 }
2656
2657 error_open:
2658 modprobe_remove_lttng_control();
2659
2660 error:
2661 WARN("No kernel tracer available");
2662 kernel_tracer_fd = -1;
2663 if (!is_root) {
2664 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2665 } else {
2666 return LTTNG_ERR_KERN_NA;
2667 }
2668 }
2669
2670
2671 /*
2672 * Copy consumer output from the tracing session to the domain session. The
2673 * function also applies the right modification on a per domain basis for the
2674 * trace files destination directory.
2675 *
2676 * Should *NOT* be called with RCU read-side lock held.
2677 */
2678 static int copy_session_consumer(int domain, struct ltt_session *session)
2679 {
2680 int ret;
2681 const char *dir_name;
2682 struct consumer_output *consumer;
2683
2684 assert(session);
2685 assert(session->consumer);
2686
2687 switch (domain) {
2688 case LTTNG_DOMAIN_KERNEL:
2689 DBG3("Copying tracing session consumer output in kernel session");
2690 /*
2691 * XXX: We should audit the session creation and what this function
2692 * does "extra" in order to avoid a destroy since this function is used
2693 * in the domain session creation (kernel and ust) only. Same for UST
2694 * domain.
2695 */
2696 if (session->kernel_session->consumer) {
2697 consumer_output_put(session->kernel_session->consumer);
2698 }
2699 session->kernel_session->consumer =
2700 consumer_copy_output(session->consumer);
2701 /* Ease our life a bit for the next part */
2702 consumer = session->kernel_session->consumer;
2703 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2704 break;
2705 case LTTNG_DOMAIN_JUL:
2706 case LTTNG_DOMAIN_LOG4J:
2707 case LTTNG_DOMAIN_PYTHON:
2708 case LTTNG_DOMAIN_UST:
2709 DBG3("Copying tracing session consumer output in UST session");
2710 if (session->ust_session->consumer) {
2711 consumer_output_put(session->ust_session->consumer);
2712 }
2713 session->ust_session->consumer =
2714 consumer_copy_output(session->consumer);
2715 /* Ease our life a bit for the next part */
2716 consumer = session->ust_session->consumer;
2717 dir_name = DEFAULT_UST_TRACE_DIR;
2718 break;
2719 default:
2720 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2721 goto error;
2722 }
2723
2724 /* Append correct directory to subdir */
2725 strncat(consumer->subdir, dir_name,
2726 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2727 DBG3("Copy session consumer subdir %s", consumer->subdir);
2728
2729 ret = LTTNG_OK;
2730
2731 error:
2732 return ret;
2733 }
2734
2735 /*
2736 * Create an UST session and add it to the session ust list.
2737 *
2738 * Should *NOT* be called with RCU read-side lock held.
2739 */
2740 static int create_ust_session(struct ltt_session *session,
2741 struct lttng_domain *domain)
2742 {
2743 int ret;
2744 struct ltt_ust_session *lus = NULL;
2745
2746 assert(session);
2747 assert(domain);
2748 assert(session->consumer);
2749
2750 switch (domain->type) {
2751 case LTTNG_DOMAIN_JUL:
2752 case LTTNG_DOMAIN_LOG4J:
2753 case LTTNG_DOMAIN_PYTHON:
2754 case LTTNG_DOMAIN_UST:
2755 break;
2756 default:
2757 ERR("Unknown UST domain on create session %d", domain->type);
2758 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2759 goto error;
2760 }
2761
2762 DBG("Creating UST session");
2763
2764 lus = trace_ust_create_session(session->id);
2765 if (lus == NULL) {
2766 ret = LTTNG_ERR_UST_SESS_FAIL;
2767 goto error;
2768 }
2769
2770 lus->uid = session->uid;
2771 lus->gid = session->gid;
2772 lus->output_traces = session->output_traces;
2773 lus->snapshot_mode = session->snapshot_mode;
2774 lus->live_timer_interval = session->live_timer;
2775 session->ust_session = lus;
2776 if (session->shm_path[0]) {
2777 strncpy(lus->root_shm_path, session->shm_path,
2778 sizeof(lus->root_shm_path));
2779 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2780 strncpy(lus->shm_path, session->shm_path,
2781 sizeof(lus->shm_path));
2782 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2783 strncat(lus->shm_path, "/ust",
2784 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2785 }
2786 /* Copy session output to the newly created UST session */
2787 ret = copy_session_consumer(domain->type, session);
2788 if (ret != LTTNG_OK) {
2789 goto error;
2790 }
2791
2792 return LTTNG_OK;
2793
2794 error:
2795 free(lus);
2796 session->ust_session = NULL;
2797 return ret;
2798 }
2799
2800 /*
2801 * Create a kernel tracer session then create the default channel.
2802 */
2803 static int create_kernel_session(struct ltt_session *session)
2804 {
2805 int ret;
2806
2807 DBG("Creating kernel session");
2808
2809 ret = kernel_create_session(session, kernel_tracer_fd);
2810 if (ret < 0) {
2811 ret = LTTNG_ERR_KERN_SESS_FAIL;
2812 goto error;
2813 }
2814
2815 /* Code flow safety */
2816 assert(session->kernel_session);
2817
2818 /* Copy session output to the newly created Kernel session */
2819 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2820 if (ret != LTTNG_OK) {
2821 goto error;
2822 }
2823
2824 /* Create directory(ies) on local filesystem. */
2825 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2826 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2827 ret = run_as_mkdir_recursive(
2828 session->kernel_session->consumer->dst.trace_path,
2829 S_IRWXU | S_IRWXG, session->uid, session->gid);
2830 if (ret < 0) {
2831 if (errno != EEXIST) {
2832 ERR("Trace directory creation error");
2833 goto error;
2834 }
2835 }
2836 }
2837
2838 session->kernel_session->uid = session->uid;
2839 session->kernel_session->gid = session->gid;
2840 session->kernel_session->output_traces = session->output_traces;
2841 session->kernel_session->snapshot_mode = session->snapshot_mode;
2842
2843 return LTTNG_OK;
2844
2845 error:
2846 trace_kernel_destroy_session(session->kernel_session);
2847 session->kernel_session = NULL;
2848 return ret;
2849 }
2850
2851 /*
2852 * Count number of session permitted by uid/gid.
2853 */
2854 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2855 {
2856 unsigned int i = 0;
2857 struct ltt_session *session;
2858
2859 DBG("Counting number of available session for UID %d GID %d",
2860 uid, gid);
2861 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2862 /*
2863 * Only list the sessions the user can control.
2864 */
2865 if (!session_access_ok(session, uid, gid)) {
2866 continue;
2867 }
2868 i++;
2869 }
2870 return i;
2871 }
2872
2873 /*
2874 * Process the command requested by the lttng client within the command
2875 * context structure. This function make sure that the return structure (llm)
2876 * is set and ready for transmission before returning.
2877 *
2878 * Return any error encountered or 0 for success.
2879 *
2880 * "sock" is only used for special-case var. len data.
2881 *
2882 * Should *NOT* be called with RCU read-side lock held.
2883 */
2884 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2885 int *sock_error)
2886 {
2887 int ret = LTTNG_OK;
2888 int need_tracing_session = 1;
2889 int need_domain;
2890
2891 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2892
2893 assert(!rcu_read_ongoing());
2894
2895 *sock_error = 0;
2896
2897 switch (cmd_ctx->lsm->cmd_type) {
2898 case LTTNG_CREATE_SESSION:
2899 case LTTNG_CREATE_SESSION_SNAPSHOT:
2900 case LTTNG_CREATE_SESSION_LIVE:
2901 case LTTNG_DESTROY_SESSION:
2902 case LTTNG_LIST_SESSIONS:
2903 case LTTNG_LIST_DOMAINS:
2904 case LTTNG_START_TRACE:
2905 case LTTNG_STOP_TRACE:
2906 case LTTNG_DATA_PENDING:
2907 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2908 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2909 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2910 case LTTNG_SNAPSHOT_RECORD:
2911 case LTTNG_SAVE_SESSION:
2912 case LTTNG_SET_SESSION_SHM_PATH:
2913 case LTTNG_REGENERATE_METADATA:
2914 case LTTNG_REGENERATE_STATEDUMP:
2915 case LTTNG_REGISTER_TRIGGER:
2916 case LTTNG_UNREGISTER_TRIGGER:
2917 need_domain = 0;
2918 break;
2919 default:
2920 need_domain = 1;
2921 }
2922
2923 if (config.no_kernel && need_domain
2924 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2925 if (!is_root) {
2926 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2927 } else {
2928 ret = LTTNG_ERR_KERN_NA;
2929 }
2930 goto error;
2931 }
2932
2933 /* Deny register consumer if we already have a spawned consumer. */
2934 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2935 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2936 if (kconsumer_data.pid > 0) {
2937 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2938 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2939 goto error;
2940 }
2941 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2942 }
2943
2944 /*
2945 * Check for command that don't needs to allocate a returned payload. We do
2946 * this here so we don't have to make the call for no payload at each
2947 * command.
2948 */
2949 switch(cmd_ctx->lsm->cmd_type) {
2950 case LTTNG_LIST_SESSIONS:
2951 case LTTNG_LIST_TRACEPOINTS:
2952 case LTTNG_LIST_TRACEPOINT_FIELDS:
2953 case LTTNG_LIST_DOMAINS:
2954 case LTTNG_LIST_CHANNELS:
2955 case LTTNG_LIST_EVENTS:
2956 case LTTNG_LIST_SYSCALLS:
2957 case LTTNG_LIST_TRACKER_PIDS:
2958 case LTTNG_DATA_PENDING:
2959 break;
2960 default:
2961 /* Setup lttng message with no payload */
2962 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
2963 if (ret < 0) {
2964 /* This label does not try to unlock the session */
2965 goto init_setup_error;
2966 }
2967 }
2968
2969 /* Commands that DO NOT need a session. */
2970 switch (cmd_ctx->lsm->cmd_type) {
2971 case LTTNG_CREATE_SESSION:
2972 case LTTNG_CREATE_SESSION_SNAPSHOT:
2973 case LTTNG_CREATE_SESSION_LIVE:
2974 case LTTNG_LIST_SESSIONS:
2975 case LTTNG_LIST_TRACEPOINTS:
2976 case LTTNG_LIST_SYSCALLS:
2977 case LTTNG_LIST_TRACEPOINT_FIELDS:
2978 case LTTNG_SAVE_SESSION:
2979 case LTTNG_REGISTER_TRIGGER:
2980 case LTTNG_UNREGISTER_TRIGGER:
2981 need_tracing_session = 0;
2982 break;
2983 default:
2984 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2985 /*
2986 * We keep the session list lock across _all_ commands
2987 * for now, because the per-session lock does not
2988 * handle teardown properly.
2989 */
2990 session_lock_list();
2991 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2992 if (cmd_ctx->session == NULL) {
2993 ret = LTTNG_ERR_SESS_NOT_FOUND;
2994 goto error;
2995 } else {
2996 /* Acquire lock for the session */
2997 session_lock(cmd_ctx->session);
2998 }
2999 break;
3000 }
3001
3002 /*
3003 * Commands that need a valid session but should NOT create one if none
3004 * exists. Instead of creating one and destroying it when the command is
3005 * handled, process that right before so we save some round trip in useless
3006 * code path.
3007 */
3008 switch (cmd_ctx->lsm->cmd_type) {
3009 case LTTNG_DISABLE_CHANNEL:
3010 case LTTNG_DISABLE_EVENT:
3011 switch (cmd_ctx->lsm->domain.type) {
3012 case LTTNG_DOMAIN_KERNEL:
3013 if (!cmd_ctx->session->kernel_session) {
3014 ret = LTTNG_ERR_NO_CHANNEL;
3015 goto error;
3016 }
3017 break;
3018 case LTTNG_DOMAIN_JUL:
3019 case LTTNG_DOMAIN_LOG4J:
3020 case LTTNG_DOMAIN_PYTHON:
3021 case LTTNG_DOMAIN_UST:
3022 if (!cmd_ctx->session->ust_session) {
3023 ret = LTTNG_ERR_NO_CHANNEL;
3024 goto error;
3025 }
3026 break;
3027 default:
3028 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3029 goto error;
3030 }
3031 default:
3032 break;
3033 }
3034
3035 if (!need_domain) {
3036 goto skip_domain;
3037 }
3038
3039 /*
3040 * Check domain type for specific "pre-action".
3041 */
3042 switch (cmd_ctx->lsm->domain.type) {
3043 case LTTNG_DOMAIN_KERNEL:
3044 if (!is_root) {
3045 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3046 goto error;
3047 }
3048
3049 /* Kernel tracer check */
3050 if (kernel_tracer_fd == -1) {
3051 /* Basically, load kernel tracer modules */
3052 ret = init_kernel_tracer();
3053 if (ret != 0) {
3054 goto error;
3055 }
3056 }
3057
3058 /* Consumer is in an ERROR state. Report back to client */
3059 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3060 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3061 goto error;
3062 }
3063
3064 /* Need a session for kernel command */
3065 if (need_tracing_session) {
3066 if (cmd_ctx->session->kernel_session == NULL) {
3067 ret = create_kernel_session(cmd_ctx->session);
3068 if (ret < 0) {
3069 ret = LTTNG_ERR_KERN_SESS_FAIL;
3070 goto error;
3071 }
3072 }
3073
3074 /* Start the kernel consumer daemon */
3075 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3076 if (kconsumer_data.pid == 0 &&
3077 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3078 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3079 ret = start_consumerd(&kconsumer_data);
3080 if (ret < 0) {
3081 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3082 goto error;
3083 }
3084 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3085 } else {
3086 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3087 }
3088
3089 /*
3090 * The consumer was just spawned so we need to add the socket to
3091 * the consumer output of the session if exist.
3092 */
3093 ret = consumer_create_socket(&kconsumer_data,
3094 cmd_ctx->session->kernel_session->consumer);
3095 if (ret < 0) {
3096 goto error;
3097 }
3098 }
3099
3100 break;
3101 case LTTNG_DOMAIN_JUL:
3102 case LTTNG_DOMAIN_LOG4J:
3103 case LTTNG_DOMAIN_PYTHON:
3104 case LTTNG_DOMAIN_UST:
3105 {
3106 if (!ust_app_supported()) {
3107 ret = LTTNG_ERR_NO_UST;
3108 goto error;
3109 }
3110 /* Consumer is in an ERROR state. Report back to client */
3111 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3112 ret = LTTNG_ERR_NO_USTCONSUMERD;
3113 goto error;
3114 }
3115
3116 if (need_tracing_session) {
3117 /* Create UST session if none exist. */
3118 if (cmd_ctx->session->ust_session == NULL) {
3119 ret = create_ust_session(cmd_ctx->session,
3120 &cmd_ctx->lsm->domain);
3121 if (ret != LTTNG_OK) {
3122 goto error;
3123 }
3124 }
3125
3126 /* Start the UST consumer daemons */
3127 /* 64-bit */
3128 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3129 if (config.consumerd64_bin_path.value &&
3130 ustconsumer64_data.pid == 0 &&
3131 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3132 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3133 ret = start_consumerd(&ustconsumer64_data);
3134 if (ret < 0) {
3135 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3136 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3137 goto error;
3138 }
3139
3140 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3141 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3142 } else {
3143 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3144 }
3145
3146 /*
3147 * Setup socket for consumer 64 bit. No need for atomic access
3148 * since it was set above and can ONLY be set in this thread.
3149 */
3150 ret = consumer_create_socket(&ustconsumer64_data,
3151 cmd_ctx->session->ust_session->consumer);
3152 if (ret < 0) {
3153 goto error;
3154 }
3155
3156 /* 32-bit */
3157 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3158 if (config.consumerd32_bin_path.value &&
3159 ustconsumer32_data.pid == 0 &&
3160 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3161 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3162 ret = start_consumerd(&ustconsumer32_data);
3163 if (ret < 0) {
3164 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3165 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3166 goto error;
3167 }
3168
3169 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3170 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3171 } else {
3172 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3173 }
3174
3175 /*
3176 * Setup socket for consumer 64 bit. No need for atomic access
3177 * since it was set above and can ONLY be set in this thread.
3178 */
3179 ret = consumer_create_socket(&ustconsumer32_data,
3180 cmd_ctx->session->ust_session->consumer);
3181 if (ret < 0) {
3182 goto error;
3183 }
3184 }
3185 break;
3186 }
3187 default:
3188 break;
3189 }
3190 skip_domain:
3191
3192 /* Validate consumer daemon state when start/stop trace command */
3193 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3194 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3195 switch (cmd_ctx->lsm->domain.type) {
3196 case LTTNG_DOMAIN_NONE:
3197 break;
3198 case LTTNG_DOMAIN_JUL:
3199 case LTTNG_DOMAIN_LOG4J:
3200 case LTTNG_DOMAIN_PYTHON:
3201 case LTTNG_DOMAIN_UST:
3202 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3203 ret = LTTNG_ERR_NO_USTCONSUMERD;
3204 goto error;
3205 }
3206 break;
3207 case LTTNG_DOMAIN_KERNEL:
3208 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3209 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3210 goto error;
3211 }
3212 break;
3213 default:
3214 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3215 goto error;
3216 }
3217 }
3218
3219 /*
3220 * Check that the UID or GID match that of the tracing session.
3221 * The root user can interact with all sessions.
3222 */
3223 if (need_tracing_session) {
3224 if (!session_access_ok(cmd_ctx->session,
3225 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3226 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3227 ret = LTTNG_ERR_EPERM;
3228 goto error;
3229 }
3230 }
3231
3232 /*
3233 * Send relayd information to consumer as soon as we have a domain and a
3234 * session defined.
3235 */
3236 if (cmd_ctx->session && need_domain) {
3237 /*
3238 * Setup relayd if not done yet. If the relayd information was already
3239 * sent to the consumer, this call will gracefully return.
3240 */
3241 ret = cmd_setup_relayd(cmd_ctx->session);
3242 if (ret != LTTNG_OK) {
3243 goto error;
3244 }
3245 }
3246
3247 /* Process by command type */
3248 switch (cmd_ctx->lsm->cmd_type) {
3249 case LTTNG_ADD_CONTEXT:
3250 {
3251 /*
3252 * An LTTNG_ADD_CONTEXT command might have a supplementary
3253 * payload if the context being added is an application context.
3254 */
3255 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3256 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3257 char *provider_name = NULL, *context_name = NULL;
3258 size_t provider_name_len =
3259 cmd_ctx->lsm->u.context.provider_name_len;
3260 size_t context_name_len =
3261 cmd_ctx->lsm->u.context.context_name_len;
3262
3263 if (provider_name_len == 0 || context_name_len == 0) {
3264 /*
3265 * Application provider and context names MUST
3266 * be provided.
3267 */
3268 ret = -LTTNG_ERR_INVALID;
3269 goto error;
3270 }
3271
3272 provider_name = zmalloc(provider_name_len + 1);
3273 if (!provider_name) {
3274 ret = -LTTNG_ERR_NOMEM;
3275 goto error;
3276 }
3277 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3278 provider_name;
3279
3280 context_name = zmalloc(context_name_len + 1);
3281 if (!context_name) {
3282 ret = -LTTNG_ERR_NOMEM;
3283 goto error_add_context;
3284 }
3285 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3286 context_name;
3287
3288 ret = lttcomm_recv_unix_sock(sock, provider_name,
3289 provider_name_len);
3290 if (ret < 0) {
3291 goto error_add_context;
3292 }
3293
3294 ret = lttcomm_recv_unix_sock(sock, context_name,
3295 context_name_len);
3296 if (ret < 0) {
3297 goto error_add_context;
3298 }
3299 }
3300
3301 /*
3302 * cmd_add_context assumes ownership of the provider and context
3303 * names.
3304 */
3305 ret = cmd_add_context(cmd_ctx->session,
3306 cmd_ctx->lsm->domain.type,
3307 cmd_ctx->lsm->u.context.channel_name,
3308 &cmd_ctx->lsm->u.context.ctx,
3309 kernel_poll_pipe[1]);
3310
3311 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3312 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3313 error_add_context:
3314 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3315 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3316 if (ret < 0) {
3317 goto error;
3318 }
3319 break;
3320 }
3321 case LTTNG_DISABLE_CHANNEL:
3322 {
3323 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3324 cmd_ctx->lsm->u.disable.channel_name);
3325 break;
3326 }
3327 case LTTNG_DISABLE_EVENT:
3328 {
3329
3330 /*
3331 * FIXME: handle filter; for now we just receive the filter's
3332 * bytecode along with the filter expression which are sent by
3333 * liblttng-ctl and discard them.
3334 *
3335 * This fixes an issue where the client may block while sending
3336 * the filter payload and encounter an error because the session
3337 * daemon closes the socket without ever handling this data.
3338 */
3339 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3340 cmd_ctx->lsm->u.disable.bytecode_len;
3341
3342 if (count) {
3343 char data[LTTNG_FILTER_MAX_LEN];
3344
3345 DBG("Discarding disable event command payload of size %zu", count);
3346 while (count) {
3347 ret = lttcomm_recv_unix_sock(sock, data,
3348 count > sizeof(data) ? sizeof(data) : count);
3349 if (ret < 0) {
3350 goto error;
3351 }
3352
3353 count -= (size_t) ret;
3354 }
3355 }
3356 /* FIXME: passing packed structure to non-packed pointer */
3357 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3358 cmd_ctx->lsm->u.disable.channel_name,
3359 &cmd_ctx->lsm->u.disable.event);
3360 break;
3361 }
3362 case LTTNG_ENABLE_CHANNEL:
3363 {
3364 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3365 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3366 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3367 &cmd_ctx->lsm->u.channel.chan,
3368 kernel_poll_pipe[1]);
3369 break;
3370 }
3371 case LTTNG_TRACK_PID:
3372 {
3373 ret = cmd_track_pid(cmd_ctx->session,
3374 cmd_ctx->lsm->domain.type,
3375 cmd_ctx->lsm->u.pid_tracker.pid);
3376 break;
3377 }
3378 case LTTNG_UNTRACK_PID:
3379 {
3380 ret = cmd_untrack_pid(cmd_ctx->session,
3381 cmd_ctx->lsm->domain.type,
3382 cmd_ctx->lsm->u.pid_tracker.pid);
3383 break;
3384 }
3385 case LTTNG_ENABLE_EVENT:
3386 {
3387 struct lttng_event_exclusion *exclusion = NULL;
3388 struct lttng_filter_bytecode *bytecode = NULL;
3389 char *filter_expression = NULL;
3390
3391 /* Handle exclusion events and receive it from the client. */
3392 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3393 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3394
3395 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3396 (count * LTTNG_SYMBOL_NAME_LEN));
3397 if (!exclusion) {
3398 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3399 goto error;
3400 }
3401
3402 DBG("Receiving var len exclusion event list from client ...");
3403 exclusion->count = count;
3404 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3405 count * LTTNG_SYMBOL_NAME_LEN);
3406 if (ret <= 0) {
3407 DBG("Nothing recv() from client var len data... continuing");
3408 *sock_error = 1;
3409 free(exclusion);
3410 ret = LTTNG_ERR_EXCLUSION_INVAL;
3411 goto error;
3412 }
3413 }
3414
3415 /* Get filter expression from client. */
3416 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3417 size_t expression_len =
3418 cmd_ctx->lsm->u.enable.expression_len;
3419
3420 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3421 ret = LTTNG_ERR_FILTER_INVAL;
3422 free(exclusion);
3423 goto error;
3424 }
3425
3426 filter_expression = zmalloc(expression_len);
3427 if (!filter_expression) {
3428 free(exclusion);
3429 ret = LTTNG_ERR_FILTER_NOMEM;
3430 goto error;
3431 }
3432
3433 /* Receive var. len. data */
3434 DBG("Receiving var len filter's expression from client ...");
3435 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3436 expression_len);
3437 if (ret <= 0) {
3438 DBG("Nothing recv() from client car len data... continuing");
3439 *sock_error = 1;
3440 free(filter_expression);
3441 free(exclusion);
3442 ret = LTTNG_ERR_FILTER_INVAL;
3443 goto error;
3444 }
3445 }
3446
3447 /* Handle filter and get bytecode from client. */
3448 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3449 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3450
3451 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3452 ret = LTTNG_ERR_FILTER_INVAL;
3453 free(filter_expression);
3454 free(exclusion);
3455 goto error;
3456 }
3457
3458 bytecode = zmalloc(bytecode_len);
3459 if (!bytecode) {
3460 free(filter_expression);
3461 free(exclusion);
3462 ret = LTTNG_ERR_FILTER_NOMEM;
3463 goto error;
3464 }
3465
3466 /* Receive var. len. data */
3467 DBG("Receiving var len filter's bytecode from client ...");
3468 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3469 if (ret <= 0) {
3470 DBG("Nothing recv() from client car len data... continuing");
3471 *sock_error = 1;
3472 free(filter_expression);
3473 free(bytecode);
3474 free(exclusion);
3475 ret = LTTNG_ERR_FILTER_INVAL;
3476 goto error;
3477 }
3478
3479 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3480 free(filter_expression);
3481 free(bytecode);
3482 free(exclusion);
3483 ret = LTTNG_ERR_FILTER_INVAL;
3484 goto error;
3485 }
3486 }
3487
3488 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3489 cmd_ctx->lsm->u.enable.channel_name,
3490 &cmd_ctx->lsm->u.enable.event,
3491 filter_expression, bytecode, exclusion,
3492 kernel_poll_pipe[1]);
3493 break;
3494 }
3495 case LTTNG_LIST_TRACEPOINTS:
3496 {
3497 struct lttng_event *events;
3498 ssize_t nb_events;
3499
3500 session_lock_list();
3501 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3502 session_unlock_list();
3503 if (nb_events < 0) {
3504 /* Return value is a negative lttng_error_code. */
3505 ret = -nb_events;
3506 goto error;
3507 }
3508
3509 /*
3510 * Setup lttng message with payload size set to the event list size in
3511 * bytes and then copy list into the llm payload.
3512 */
3513 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3514 sizeof(struct lttng_event) * nb_events);
3515 free(events);
3516
3517 if (ret < 0) {
3518 goto setup_error;
3519 }
3520
3521 ret = LTTNG_OK;
3522 break;
3523 }
3524 case LTTNG_LIST_TRACEPOINT_FIELDS:
3525 {
3526 struct lttng_event_field *fields;
3527 ssize_t nb_fields;
3528
3529 session_lock_list();
3530 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3531 &fields);
3532 session_unlock_list();
3533 if (nb_fields < 0) {
3534 /* Return value is a negative lttng_error_code. */
3535 ret = -nb_fields;
3536 goto error;
3537 }
3538
3539 /*
3540 * Setup lttng message with payload size set to the event list size in
3541 * bytes and then copy list into the llm payload.
3542 */
3543 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3544 sizeof(struct lttng_event_field) * nb_fields);
3545 free(fields);
3546
3547 if (ret < 0) {
3548 goto setup_error;
3549 }
3550
3551 ret = LTTNG_OK;
3552 break;
3553 }
3554 case LTTNG_LIST_SYSCALLS:
3555 {
3556 struct lttng_event *events;
3557 ssize_t nb_events;
3558
3559 nb_events = cmd_list_syscalls(&events);
3560 if (nb_events < 0) {
3561 /* Return value is a negative lttng_error_code. */
3562 ret = -nb_events;
3563 goto error;
3564 }
3565
3566 /*
3567 * Setup lttng message with payload size set to the event list size in
3568 * bytes and then copy list into the llm payload.
3569 */
3570 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3571 sizeof(struct lttng_event) * nb_events);
3572 free(events);
3573
3574 if (ret < 0) {
3575 goto setup_error;
3576 }
3577
3578 ret = LTTNG_OK;
3579 break;
3580 }
3581 case LTTNG_LIST_TRACKER_PIDS:
3582 {
3583 int32_t *pids = NULL;
3584 ssize_t nr_pids;
3585
3586 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3587 cmd_ctx->lsm->domain.type, &pids);
3588 if (nr_pids < 0) {
3589 /* Return value is a negative lttng_error_code. */
3590 ret = -nr_pids;
3591 goto error;
3592 }
3593
3594 /*
3595 * Setup lttng message with payload size set to the event list size in
3596 * bytes and then copy list into the llm payload.
3597 */
3598 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3599 sizeof(int32_t) * nr_pids);
3600 free(pids);
3601
3602 if (ret < 0) {
3603 goto setup_error;
3604 }
3605
3606 ret = LTTNG_OK;
3607 break;
3608 }
3609 case LTTNG_SET_CONSUMER_URI:
3610 {
3611 size_t nb_uri, len;
3612 struct lttng_uri *uris;
3613
3614 nb_uri = cmd_ctx->lsm->u.uri.size;
3615 len = nb_uri * sizeof(struct lttng_uri);
3616
3617 if (nb_uri == 0) {
3618 ret = LTTNG_ERR_INVALID;
3619 goto error;
3620 }
3621
3622 uris = zmalloc(len);
3623 if (uris == NULL) {
3624 ret = LTTNG_ERR_FATAL;
3625 goto error;
3626 }
3627
3628 /* Receive variable len data */
3629 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3630 ret = lttcomm_recv_unix_sock(sock, uris, len);
3631 if (ret <= 0) {
3632 DBG("No URIs received from client... continuing");
3633 *sock_error = 1;
3634 ret = LTTNG_ERR_SESSION_FAIL;
3635 free(uris);
3636 goto error;
3637 }
3638
3639 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3640 free(uris);
3641 if (ret != LTTNG_OK) {
3642 goto error;
3643 }
3644
3645
3646 break;
3647 }
3648 case LTTNG_START_TRACE:
3649 {
3650 ret = cmd_start_trace(cmd_ctx->session);
3651 break;
3652 }
3653 case LTTNG_STOP_TRACE:
3654 {
3655 ret = cmd_stop_trace(cmd_ctx->session);
3656 break;
3657 }
3658 case LTTNG_CREATE_SESSION:
3659 {
3660 size_t nb_uri, len;
3661 struct lttng_uri *uris = NULL;
3662
3663 nb_uri = cmd_ctx->lsm->u.uri.size;
3664 len = nb_uri * sizeof(struct lttng_uri);
3665
3666 if (nb_uri > 0) {
3667 uris = zmalloc(len);
3668 if (uris == NULL) {
3669 ret = LTTNG_ERR_FATAL;
3670 goto error;
3671 }
3672
3673 /* Receive variable len data */
3674 DBG("Waiting for %zu URIs from client ...", nb_uri);
3675 ret = lttcomm_recv_unix_sock(sock, uris, len);
3676 if (ret <= 0) {
3677 DBG("No URIs received from client... continuing");
3678 *sock_error = 1;
3679 ret = LTTNG_ERR_SESSION_FAIL;
3680 free(uris);
3681 goto error;
3682 }
3683
3684 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3685 DBG("Creating session with ONE network URI is a bad call");
3686 ret = LTTNG_ERR_SESSION_FAIL;
3687 free(uris);
3688 goto error;
3689 }
3690 }
3691
3692 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3693 &cmd_ctx->creds, 0);
3694
3695 free(uris);
3696
3697 break;
3698 }
3699 case LTTNG_DESTROY_SESSION:
3700 {
3701 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3702
3703 /* Set session to NULL so we do not unlock it after free. */
3704 cmd_ctx->session = NULL;
3705 break;
3706 }
3707 case LTTNG_LIST_DOMAINS:
3708 {
3709 ssize_t nb_dom;
3710 struct lttng_domain *domains = NULL;
3711
3712 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3713 if (nb_dom < 0) {
3714 /* Return value is a negative lttng_error_code. */
3715 ret = -nb_dom;
3716 goto error;
3717 }
3718
3719 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3720 nb_dom * sizeof(struct lttng_domain));
3721 free(domains);
3722
3723 if (ret < 0) {
3724 goto setup_error;
3725 }
3726
3727 ret = LTTNG_OK;
3728 break;
3729 }
3730 case LTTNG_LIST_CHANNELS:
3731 {
3732 ssize_t payload_size;
3733 struct lttng_channel *channels = NULL;
3734
3735 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3736 cmd_ctx->session, &channels);
3737 if (payload_size < 0) {
3738 /* Return value is a negative lttng_error_code. */
3739 ret = -payload_size;
3740 goto error;
3741 }
3742
3743 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3744 payload_size);
3745 free(channels);
3746
3747 if (ret < 0) {
3748 goto setup_error;
3749 }
3750
3751 ret = LTTNG_OK;
3752 break;
3753 }
3754 case LTTNG_LIST_EVENTS:
3755 {
3756 ssize_t nb_event;
3757 struct lttng_event *events = NULL;
3758 struct lttcomm_event_command_header cmd_header;
3759 size_t total_size;
3760
3761 memset(&cmd_header, 0, sizeof(cmd_header));
3762 /* Extended infos are included at the end of events */
3763 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3764 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3765 &events, &total_size);
3766
3767 if (nb_event < 0) {
3768 /* Return value is a negative lttng_error_code. */
3769 ret = -nb_event;
3770 goto error;
3771 }
3772
3773 cmd_header.nb_events = nb_event;
3774 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3775 &cmd_header, sizeof(cmd_header));
3776 free(events);
3777
3778 if (ret < 0) {
3779 goto setup_error;
3780 }
3781
3782 ret = LTTNG_OK;
3783 break;
3784 }
3785 case LTTNG_LIST_SESSIONS:
3786 {
3787 unsigned int nr_sessions;
3788 void *sessions_payload;
3789 size_t payload_len;
3790
3791 session_lock_list();
3792 nr_sessions = lttng_sessions_count(
3793 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3794 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3795 payload_len = sizeof(struct lttng_session) * nr_sessions;
3796 sessions_payload = zmalloc(payload_len);
3797
3798 if (!sessions_payload) {
3799 session_unlock_list();
3800 ret = -ENOMEM;
3801 goto setup_error;
3802 }
3803
3804 cmd_list_lttng_sessions(sessions_payload,
3805 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3806 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3807 session_unlock_list();
3808
3809 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3810 payload_len);
3811 free(sessions_payload);
3812
3813 if (ret < 0) {
3814 goto setup_error;
3815 }
3816
3817 ret = LTTNG_OK;
3818 break;
3819 }
3820 case LTTNG_REGISTER_CONSUMER:
3821 {
3822 struct consumer_data *cdata;
3823
3824 switch (cmd_ctx->lsm->domain.type) {
3825 case LTTNG_DOMAIN_KERNEL:
3826 cdata = &kconsumer_data;
3827 break;
3828 default:
3829 ret = LTTNG_ERR_UND;
3830 goto error;
3831 }
3832
3833 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3834 cmd_ctx->lsm->u.reg.path, cdata);
3835 break;
3836 }
3837 case LTTNG_DATA_PENDING:
3838 {
3839 int pending_ret;
3840 uint8_t pending_ret_byte;
3841
3842 pending_ret = cmd_data_pending(cmd_ctx->session);
3843
3844 /*
3845 * FIXME
3846 *
3847 * This function may returns 0 or 1 to indicate whether or not
3848 * there is data pending. In case of error, it should return an
3849 * LTTNG_ERR code. However, some code paths may still return
3850 * a nondescript error code, which we handle by returning an
3851 * "unknown" error.
3852 */
3853 if (pending_ret == 0 || pending_ret == 1) {
3854 /*
3855 * ret will be set to LTTNG_OK at the end of
3856 * this function.
3857 */
3858 } else if (pending_ret < 0) {
3859 ret = LTTNG_ERR_UNK;
3860 goto setup_error;
3861 } else {
3862 ret = pending_ret;
3863 goto setup_error;
3864 }
3865
3866 pending_ret_byte = (uint8_t) pending_ret;
3867
3868 /* 1 byte to return whether or not data is pending */
3869 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3870 &pending_ret_byte, 1);
3871
3872 if (ret < 0) {
3873 goto setup_error;
3874 }
3875
3876 ret = LTTNG_OK;
3877 break;
3878 }
3879 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3880 {
3881 struct lttcomm_lttng_output_id reply;
3882
3883 ret = cmd_snapshot_add_output(cmd_ctx->session,
3884 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3885 if (ret != LTTNG_OK) {
3886 goto error;
3887 }
3888
3889 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
3890 sizeof(reply));
3891 if (ret < 0) {
3892 goto setup_error;
3893 }
3894
3895 /* Copy output list into message payload */
3896 ret = LTTNG_OK;
3897 break;
3898 }
3899 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3900 {
3901 ret = cmd_snapshot_del_output(cmd_ctx->session,
3902 &cmd_ctx->lsm->u.snapshot_output.output);
3903 break;
3904 }
3905 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3906 {
3907 ssize_t nb_output;
3908 struct lttng_snapshot_output *outputs = NULL;
3909
3910 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3911 if (nb_output < 0) {
3912 ret = -nb_output;
3913 goto error;
3914 }
3915
3916 assert((nb_output > 0 && outputs) || nb_output == 0);
3917 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
3918 nb_output * sizeof(struct lttng_snapshot_output));
3919 free(outputs);
3920
3921 if (ret < 0) {
3922 goto setup_error;
3923 }
3924
3925 ret = LTTNG_OK;
3926 break;
3927 }
3928 case LTTNG_SNAPSHOT_RECORD:
3929 {
3930 ret = cmd_snapshot_record(cmd_ctx->session,
3931 &cmd_ctx->lsm->u.snapshot_record.output,
3932 cmd_ctx->lsm->u.snapshot_record.wait);
3933 break;
3934 }
3935 case LTTNG_CREATE_SESSION_SNAPSHOT:
3936 {
3937 size_t nb_uri, len;
3938 struct lttng_uri *uris = NULL;
3939
3940 nb_uri = cmd_ctx->lsm->u.uri.size;
3941 len = nb_uri * sizeof(struct lttng_uri);
3942
3943 if (nb_uri > 0) {
3944 uris = zmalloc(len);
3945 if (uris == NULL) {
3946 ret = LTTNG_ERR_FATAL;
3947 goto error;
3948 }
3949
3950 /* Receive variable len data */
3951 DBG("Waiting for %zu URIs from client ...", nb_uri);
3952 ret = lttcomm_recv_unix_sock(sock, uris, len);
3953 if (ret <= 0) {
3954 DBG("No URIs received from client... continuing");
3955 *sock_error = 1;
3956 ret = LTTNG_ERR_SESSION_FAIL;
3957 free(uris);
3958 goto error;
3959 }
3960
3961 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3962 DBG("Creating session with ONE network URI is a bad call");
3963 ret = LTTNG_ERR_SESSION_FAIL;
3964 free(uris);
3965 goto error;
3966 }
3967 }
3968
3969 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3970 nb_uri, &cmd_ctx->creds);
3971 free(uris);
3972 break;
3973 }
3974 case LTTNG_CREATE_SESSION_LIVE:
3975 {
3976 size_t nb_uri, len;
3977 struct lttng_uri *uris = NULL;
3978
3979 nb_uri = cmd_ctx->lsm->u.uri.size;
3980 len = nb_uri * sizeof(struct lttng_uri);
3981
3982 if (nb_uri > 0) {
3983 uris = zmalloc(len);
3984 if (uris == NULL) {
3985 ret = LTTNG_ERR_FATAL;
3986 goto error;
3987 }
3988
3989 /* Receive variable len data */
3990 DBG("Waiting for %zu URIs from client ...", nb_uri);
3991 ret = lttcomm_recv_unix_sock(sock, uris, len);
3992 if (ret <= 0) {
3993 DBG("No URIs received from client... continuing");
3994 *sock_error = 1;
3995 ret = LTTNG_ERR_SESSION_FAIL;
3996 free(uris);
3997 goto error;
3998 }
3999
4000 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4001 DBG("Creating session with ONE network URI is a bad call");
4002 ret = LTTNG_ERR_SESSION_FAIL;
4003 free(uris);
4004 goto error;
4005 }
4006 }
4007
4008 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4009 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4010 free(uris);
4011 break;
4012 }
4013 case LTTNG_SAVE_SESSION:
4014 {
4015 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4016 &cmd_ctx->creds);
4017 break;
4018 }
4019 case LTTNG_SET_SESSION_SHM_PATH:
4020 {
4021 ret = cmd_set_session_shm_path(cmd_ctx->session,
4022 cmd_ctx->lsm->u.set_shm_path.shm_path);
4023 break;
4024 }
4025 case LTTNG_REGENERATE_METADATA:
4026 {
4027 ret = cmd_regenerate_metadata(cmd_ctx->session);
4028 break;
4029 }
4030 case LTTNG_REGENERATE_STATEDUMP:
4031 {
4032 ret = cmd_regenerate_statedump(cmd_ctx->session);
4033 break;
4034 }
4035 case LTTNG_REGISTER_TRIGGER:
4036 {
4037 ret = cmd_register_trigger(cmd_ctx, sock,
4038 notification_thread_handle);
4039 break;
4040 }
4041 case LTTNG_UNREGISTER_TRIGGER:
4042 {
4043 ret = cmd_unregister_trigger(cmd_ctx, sock,
4044 notification_thread_handle);
4045 break;
4046 }
4047 default:
4048 ret = LTTNG_ERR_UND;
4049 break;
4050 }
4051
4052 error:
4053 if (cmd_ctx->llm == NULL) {
4054 DBG("Missing llm structure. Allocating one.");
4055 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4056 goto setup_error;
4057 }
4058 }
4059 /* Set return code */
4060 cmd_ctx->llm->ret_code = ret;
4061 setup_error:
4062 if (cmd_ctx->session) {
4063 session_unlock(cmd_ctx->session);
4064 }
4065 if (need_tracing_session) {
4066 session_unlock_list();
4067 }
4068 init_setup_error:
4069 assert(!rcu_read_ongoing());
4070 return ret;
4071 }
4072
4073 /*
4074 * Thread managing health check socket.
4075 */
4076 static void *thread_manage_health(void *data)
4077 {
4078 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4079 uint32_t revents, nb_fd;
4080 struct lttng_poll_event events;
4081 struct health_comm_msg msg;
4082 struct health_comm_reply reply;
4083
4084 DBG("[thread] Manage health check started");
4085
4086 rcu_register_thread();
4087
4088 /* We might hit an error path before this is created. */
4089 lttng_poll_init(&events);
4090
4091 /* Create unix socket */
4092 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4093 if (sock < 0) {
4094 ERR("Unable to create health check Unix socket");
4095 goto error;
4096 }
4097
4098 if (is_root) {
4099 /* lttng health client socket path permissions */
4100 ret = chown(config.health_unix_sock_path.value, 0,
4101 utils_get_group_id(config.tracing_group_name.value));
4102 if (ret < 0) {
4103 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4104 PERROR("chown");
4105 goto error;
4106 }
4107
4108 ret = chmod(config.health_unix_sock_path.value,
4109 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4110 if (ret < 0) {
4111 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4112 PERROR("chmod");
4113 goto error;
4114 }
4115 }
4116
4117 /*
4118 * Set the CLOEXEC flag. Return code is useless because either way, the
4119 * show must go on.
4120 */
4121 (void) utils_set_fd_cloexec(sock);
4122
4123 ret = lttcomm_listen_unix_sock(sock);
4124 if (ret < 0) {
4125 goto error;
4126 }
4127
4128 /*
4129 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4130 * more will be added to this poll set.
4131 */
4132 ret = sessiond_set_thread_pollset(&events, 2);
4133 if (ret < 0) {
4134 goto error;
4135 }
4136
4137 /* Add the application registration socket */
4138 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4139 if (ret < 0) {
4140 goto error;
4141 }
4142
4143 sessiond_notify_ready();
4144
4145 while (1) {
4146 DBG("Health check ready");
4147
4148 /* Inifinite blocking call, waiting for transmission */
4149 restart:
4150 ret = lttng_poll_wait(&events, -1);
4151 if (ret < 0) {
4152 /*
4153 * Restart interrupted system call.
4154 */
4155 if (errno == EINTR) {
4156 goto restart;
4157 }
4158 goto error;
4159 }
4160
4161 nb_fd = ret;
4162
4163 for (i = 0; i < nb_fd; i++) {
4164 /* Fetch once the poll data */
4165 revents = LTTNG_POLL_GETEV(&events, i);
4166 pollfd = LTTNG_POLL_GETFD(&events, i);
4167
4168 if (!revents) {
4169 /* No activity for this FD (poll implementation). */
4170 continue;
4171 }
4172
4173 /* Thread quit pipe has been closed. Killing thread. */
4174 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4175 if (ret) {
4176 err = 0;
4177 goto exit;
4178 }
4179
4180 /* Event on the registration socket */
4181 if (pollfd == sock) {
4182 if (revents & LPOLLIN) {
4183 continue;
4184 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4185 ERR("Health socket poll error");
4186 goto error;
4187 } else {
4188 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4189 goto error;
4190 }
4191 }
4192 }
4193
4194 new_sock = lttcomm_accept_unix_sock(sock);
4195 if (new_sock < 0) {
4196 goto error;
4197 }
4198
4199 /*
4200 * Set the CLOEXEC flag. Return code is useless because either way, the
4201 * show must go on.
4202 */
4203 (void) utils_set_fd_cloexec(new_sock);
4204
4205 DBG("Receiving data from client for health...");
4206 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4207 if (ret <= 0) {
4208 DBG("Nothing recv() from client... continuing");
4209 ret = close(new_sock);
4210 if (ret) {
4211 PERROR("close");
4212 }
4213 continue;
4214 }
4215
4216 rcu_thread_online();
4217
4218 memset(&reply, 0, sizeof(reply));
4219 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4220 /*
4221 * health_check_state returns 0 if health is
4222 * bad.
4223 */
4224 if (!health_check_state(health_sessiond, i)) {
4225 reply.ret_code |= 1ULL << i;
4226 }
4227 }
4228
4229 DBG2("Health check return value %" PRIx64, reply.ret_code);
4230
4231 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4232 if (ret < 0) {
4233 ERR("Failed to send health data back to client");
4234 }
4235
4236 /* End of transmission */
4237 ret = close(new_sock);
4238 if (ret) {
4239 PERROR("close");
4240 }
4241 }
4242
4243 exit:
4244 error:
4245 if (err) {
4246 ERR("Health error occurred in %s", __func__);
4247 }
4248 DBG("Health check thread dying");
4249 unlink(config.health_unix_sock_path.value);
4250 if (sock >= 0) {
4251 ret = close(sock);
4252 if (ret) {
4253 PERROR("close");
4254 }
4255 }
4256
4257 lttng_poll_clean(&events);
4258 stop_threads();
4259 rcu_unregister_thread();
4260 return NULL;
4261 }
4262
4263 /*
4264 * This thread manage all clients request using the unix client socket for
4265 * communication.
4266 */
4267 static void *thread_manage_clients(void *data)
4268 {
4269 int sock = -1, ret, i, pollfd, err = -1;
4270 int sock_error;
4271 uint32_t revents, nb_fd;
4272 struct command_ctx *cmd_ctx = NULL;
4273 struct lttng_poll_event events;
4274
4275 DBG("[thread] Manage client started");
4276
4277 rcu_register_thread();
4278
4279 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4280
4281 health_code_update();
4282
4283 ret = lttcomm_listen_unix_sock(client_sock);
4284 if (ret < 0) {
4285 goto error_listen;
4286 }
4287
4288 /*
4289 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4290 * more will be added to this poll set.
4291 */
4292 ret = sessiond_set_thread_pollset(&events, 2);
4293 if (ret < 0) {
4294 goto error_create_poll;
4295 }
4296
4297 /* Add the application registration socket */
4298 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4299 if (ret < 0) {
4300 goto error;
4301 }
4302
4303 sessiond_notify_ready();
4304 ret = sem_post(&load_info->message_thread_ready);
4305 if (ret) {
4306 PERROR("sem_post message_thread_ready");
4307 goto error;
4308 }
4309
4310 /* This testpoint is after we signal readiness to the parent. */
4311 if (testpoint(sessiond_thread_manage_clients)) {
4312 goto error;
4313 }
4314
4315 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4316 goto error;
4317 }
4318
4319 health_code_update();
4320
4321 while (1) {
4322 DBG("Accepting client command ...");
4323
4324 /* Inifinite blocking call, waiting for transmission */
4325 restart:
4326 health_poll_entry();
4327 ret = lttng_poll_wait(&events, -1);
4328 health_poll_exit();
4329 if (ret < 0) {
4330 /*
4331 * Restart interrupted system call.
4332 */
4333 if (errno == EINTR) {
4334 goto restart;
4335 }
4336 goto error;
4337 }
4338
4339 nb_fd = ret;
4340
4341 for (i = 0; i < nb_fd; i++) {
4342 /* Fetch once the poll data */
4343 revents = LTTNG_POLL_GETEV(&events, i);
4344 pollfd = LTTNG_POLL_GETFD(&events, i);
4345
4346 health_code_update();
4347
4348 if (!revents) {
4349 /* No activity for this FD (poll implementation). */
4350 continue;
4351 }
4352
4353 /* Thread quit pipe has been closed. Killing thread. */
4354 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4355 if (ret) {
4356 err = 0;
4357 goto exit;
4358 }
4359
4360 /* Event on the registration socket */
4361 if (pollfd == client_sock) {
4362 if (revents & LPOLLIN) {
4363 continue;
4364 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4365 ERR("Client socket poll error");
4366 goto error;
4367 } else {
4368 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4369 goto error;
4370 }
4371 }
4372 }
4373
4374 DBG("Wait for client response");
4375
4376 health_code_update();
4377
4378 sock = lttcomm_accept_unix_sock(client_sock);
4379 if (sock < 0) {
4380 goto error;
4381 }
4382
4383 /*
4384 * Set the CLOEXEC flag. Return code is useless because either way, the
4385 * show must go on.
4386 */
4387 (void) utils_set_fd_cloexec(sock);
4388
4389 /* Set socket option for credentials retrieval */
4390 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4391 if (ret < 0) {
4392 goto error;
4393 }
4394
4395 /* Allocate context command to process the client request */
4396 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4397 if (cmd_ctx == NULL) {
4398 PERROR("zmalloc cmd_ctx");
4399 goto error;
4400 }
4401
4402 /* Allocate data buffer for reception */
4403 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4404 if (cmd_ctx->lsm == NULL) {
4405 PERROR("zmalloc cmd_ctx->lsm");
4406 goto error;
4407 }
4408
4409 cmd_ctx->llm = NULL;
4410 cmd_ctx->session = NULL;
4411
4412 health_code_update();
4413
4414 /*
4415 * Data is received from the lttng client. The struct
4416 * lttcomm_session_msg (lsm) contains the command and data request of
4417 * the client.
4418 */
4419 DBG("Receiving data from client ...");
4420 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4421 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4422 if (ret <= 0) {
4423 DBG("Nothing recv() from client... continuing");
4424 ret = close(sock);
4425 if (ret) {
4426 PERROR("close");
4427 }
4428 sock = -1;
4429 clean_command_ctx(&cmd_ctx);
4430 continue;
4431 }
4432
4433 health_code_update();
4434
4435 // TODO: Validate cmd_ctx including sanity check for
4436 // security purpose.
4437
4438 rcu_thread_online();
4439 /*
4440 * This function dispatch the work to the kernel or userspace tracer
4441 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4442 * informations for the client. The command context struct contains
4443 * everything this function may needs.
4444 */
4445 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4446 rcu_thread_offline();
4447 if (ret < 0) {
4448 ret = close(sock);
4449 if (ret) {
4450 PERROR("close");
4451 }
4452 sock = -1;
4453 /*
4454 * TODO: Inform client somehow of the fatal error. At
4455 * this point, ret < 0 means that a zmalloc failed
4456 * (ENOMEM). Error detected but still accept
4457 * command, unless a socket error has been
4458 * detected.
4459 */
4460 clean_command_ctx(&cmd_ctx);
4461 continue;
4462 }
4463
4464 health_code_update();
4465
4466 DBG("Sending response (size: %d, retcode: %s (%d))",
4467 cmd_ctx->lttng_msg_size,
4468 lttng_strerror(-cmd_ctx->llm->ret_code),
4469 cmd_ctx->llm->ret_code);
4470 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4471 if (ret < 0) {
4472 ERR("Failed to send data back to client");
4473 }
4474
4475 /* End of transmission */
4476 ret = close(sock);
4477 if (ret) {
4478 PERROR("close");
4479 }
4480 sock = -1;
4481
4482 clean_command_ctx(&cmd_ctx);
4483
4484 health_code_update();
4485 }
4486
4487 exit:
4488 error:
4489 if (sock >= 0) {
4490 ret = close(sock);
4491 if (ret) {
4492 PERROR("close");
4493 }
4494 }
4495
4496 lttng_poll_clean(&events);
4497 clean_command_ctx(&cmd_ctx);
4498
4499 error_listen:
4500 error_create_poll:
4501 unlink(config.client_unix_sock_path.value);
4502 if (client_sock >= 0) {
4503 ret = close(client_sock);
4504 if (ret) {
4505 PERROR("close");
4506 }
4507 }
4508
4509 if (err) {
4510 health_error();
4511 ERR("Health error occurred in %s", __func__);
4512 }
4513
4514 health_unregister(health_sessiond);
4515
4516 DBG("Client thread dying");
4517
4518 rcu_unregister_thread();
4519
4520 /*
4521 * Since we are creating the consumer threads, we own them, so we need
4522 * to join them before our thread exits.
4523 */
4524 ret = join_consumer_thread(&kconsumer_data);
4525 if (ret) {
4526 errno = ret;
4527 PERROR("join_consumer");
4528 }
4529
4530 ret = join_consumer_thread(&ustconsumer32_data);
4531 if (ret) {
4532 errno = ret;
4533 PERROR("join_consumer ust32");
4534 }
4535
4536 ret = join_consumer_thread(&ustconsumer64_data);
4537 if (ret) {
4538 errno = ret;
4539 PERROR("join_consumer ust64");
4540 }
4541 return NULL;
4542 }
4543
4544 static int string_match(const char *str1, const char *str2)
4545 {
4546 return (str1 && str2) && !strcmp(str1, str2);
4547 }
4548
4549 /*
4550 * Take an option from the getopt output and set it in the right variable to be
4551 * used later.
4552 *
4553 * Return 0 on success else a negative value.
4554 */
4555 static int set_option(int opt, const char *arg, const char *optname)
4556 {
4557 int ret = 0;
4558
4559 if (string_match(optname, "client-sock") || opt == 'c') {
4560 if (!arg || *arg == '\0') {
4561 ret = -EINVAL;
4562 goto end;
4563 }
4564 if (lttng_is_setuid_setgid()) {
4565 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4566 "-c, --client-sock");
4567 } else {
4568 config_string_set(&config.client_unix_sock_path,
4569 strdup(arg));
4570 if (!config.client_unix_sock_path.value) {
4571 ret = -ENOMEM;
4572 PERROR("strdup");
4573 }
4574 }
4575 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4576 if (!arg || *arg == '\0') {
4577 ret = -EINVAL;
4578 goto end;
4579 }
4580 if (lttng_is_setuid_setgid()) {
4581 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4582 "-a, --apps-sock");
4583 } else {
4584 config_string_set(&config.apps_unix_sock_path,
4585 strdup(arg));
4586 if (!config.apps_unix_sock_path.value) {
4587 ret = -ENOMEM;
4588 PERROR("strdup");
4589 }
4590 }
4591 } else if (string_match(optname, "daemonize") || opt == 'd') {
4592 config.daemonize = true;
4593 } else if (string_match(optname, "background") || opt == 'b') {
4594 config.background = true;
4595 } else if (string_match(optname, "group") || opt == 'g') {
4596 if (!arg || *arg == '\0') {
4597 ret = -EINVAL;
4598 goto end;
4599 }
4600 if (lttng_is_setuid_setgid()) {
4601 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4602 "-g, --group");
4603 } else {
4604 config_string_set(&config.tracing_group_name,
4605 strdup(arg));
4606 if (!config.tracing_group_name.value) {
4607 ret = -ENOMEM;
4608 PERROR("strdup");
4609 }
4610 }
4611 } else if (string_match(optname, "help") || opt == 'h') {
4612 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4613 if (ret) {
4614 ERR("Cannot show --help for `lttng-sessiond`");
4615 perror("exec");
4616 }
4617 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4618 } else if (string_match(optname, "version") || opt == 'V') {
4619 fprintf(stdout, "%s\n", VERSION);
4620 exit(EXIT_SUCCESS);
4621 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4622 config.sig_parent = true;
4623 } else if (string_match(optname, "kconsumerd-err-sock")) {
4624 if (!arg || *arg == '\0') {
4625 ret = -EINVAL;
4626 goto end;
4627 }
4628 if (lttng_is_setuid_setgid()) {
4629 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4630 "--kconsumerd-err-sock");
4631 } else {
4632 config_string_set(&config.kconsumerd_err_unix_sock_path,
4633 strdup(arg));
4634 if (!config.kconsumerd_err_unix_sock_path.value) {
4635 ret = -ENOMEM;
4636 PERROR("strdup");
4637 }
4638 }
4639 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4640 if (!arg || *arg == '\0') {
4641 ret = -EINVAL;
4642 goto end;
4643 }
4644 if (lttng_is_setuid_setgid()) {
4645 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4646 "--kconsumerd-cmd-sock");
4647 } else {
4648 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
4649 strdup(arg));
4650 if (!config.kconsumerd_cmd_unix_sock_path.value) {
4651 ret = -ENOMEM;
4652 PERROR("strdup");
4653 }
4654 }
4655 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4656 if (!arg || *arg == '\0') {
4657 ret = -EINVAL;
4658 goto end;
4659 }
4660 if (lttng_is_setuid_setgid()) {
4661 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4662 "--ustconsumerd64-err-sock");
4663 } else {
4664 config_string_set(&config.consumerd64_err_unix_sock_path,
4665 strdup(arg));
4666 if (!config.consumerd64_err_unix_sock_path.value) {
4667 ret = -ENOMEM;
4668 PERROR("strdup");
4669 }
4670 }
4671 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4672 if (!arg || *arg == '\0') {
4673 ret = -EINVAL;
4674 goto end;
4675 }
4676 if (lttng_is_setuid_setgid()) {
4677 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4678 "--ustconsumerd64-cmd-sock");
4679 } else {
4680 config_string_set(&config.consumerd64_cmd_unix_sock_path,
4681 strdup(arg));
4682 if (!config.consumerd64_cmd_unix_sock_path.value) {
4683 ret = -ENOMEM;
4684 PERROR("strdup");
4685 }
4686 }
4687 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4688 if (!arg || *arg == '\0') {
4689 ret = -EINVAL;
4690 goto end;
4691 }
4692 if (lttng_is_setuid_setgid()) {
4693 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4694 "--ustconsumerd32-err-sock");
4695 } else {
4696 config_string_set(&config.consumerd32_err_unix_sock_path,
4697 strdup(arg));
4698 if (!config.consumerd32_err_unix_sock_path.value) {
4699 ret = -ENOMEM;
4700 PERROR("strdup");
4701 }
4702 }
4703 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4704 if (!arg || *arg == '\0') {
4705 ret = -EINVAL;
4706 goto end;
4707 }
4708 if (lttng_is_setuid_setgid()) {
4709 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4710 "--ustconsumerd32-cmd-sock");
4711 } else {
4712 config_string_set(&config.consumerd32_cmd_unix_sock_path,
4713 strdup(arg));
4714 if (!config.consumerd32_cmd_unix_sock_path.value) {
4715 ret = -ENOMEM;
4716 PERROR("strdup");
4717 }
4718 }
4719 } else if (string_match(optname, "no-kernel")) {
4720 config.no_kernel = true;
4721 } else if (string_match(optname, "quiet") || opt == 'q') {
4722 lttng_opt_quiet = true;
4723 } else if (string_match(optname, "verbose") || opt == 'v') {
4724 /* Verbose level can increase using multiple -v */
4725 if (arg) {
4726 /* Value obtained from config file */
4727 config.verbose = config_parse_value(arg);
4728 } else {
4729 /* -v used on command line */
4730 config.verbose++;
4731 }
4732 /* Clamp value to [0, 3] */
4733 config.verbose = config.verbose < 0 ? 0 :
4734 (config.verbose <= 3 ? config.verbose : 3);
4735 } else if (string_match(optname, "verbose-consumer")) {
4736 if (arg) {
4737 config.verbose_consumer = config_parse_value(arg);
4738 } else {
4739 config.verbose_consumer++;
4740 }
4741 } else if (string_match(optname, "consumerd32-path")) {
4742 if (!arg || *arg == '\0') {
4743 ret = -EINVAL;
4744 goto end;
4745 }
4746 if (lttng_is_setuid_setgid()) {
4747 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4748 "--consumerd32-path");
4749 } else {
4750 config_string_set(&config.consumerd32_bin_path,
4751 strdup(arg));
4752 if (!config.consumerd32_bin_path.value) {
4753 PERROR("strdup");
4754 ret = -ENOMEM;
4755 }
4756 }
4757 } else if (string_match(optname, "consumerd32-libdir")) {
4758 if (!arg || *arg == '\0') {
4759 ret = -EINVAL;
4760 goto end;
4761 }
4762 if (lttng_is_setuid_setgid()) {
4763 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4764 "--consumerd32-libdir");
4765 } else {
4766 config_string_set(&config.consumerd32_lib_dir,
4767 strdup(arg));
4768 if (!config.consumerd32_lib_dir.value) {
4769 PERROR("strdup");
4770 ret = -ENOMEM;
4771 }
4772 }
4773 } else if (string_match(optname, "consumerd64-path")) {
4774 if (!arg || *arg == '\0') {
4775 ret = -EINVAL;
4776 goto end;
4777 }
4778 if (lttng_is_setuid_setgid()) {
4779 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4780 "--consumerd64-path");
4781 } else {
4782 config_string_set(&config.consumerd64_bin_path,
4783 strdup(arg));
4784 if (!config.consumerd64_bin_path.value) {
4785 PERROR("strdup");
4786 ret = -ENOMEM;
4787 }
4788 }
4789 } else if (string_match(optname, "consumerd64-libdir")) {
4790 if (!arg || *arg == '\0') {
4791 ret = -EINVAL;
4792 goto end;
4793 }
4794 if (lttng_is_setuid_setgid()) {
4795 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4796 "--consumerd64-libdir");
4797 } else {
4798 config_string_set(&config.consumerd64_lib_dir,
4799 strdup(arg));
4800 if (!config.consumerd64_lib_dir.value) {
4801 PERROR("strdup");
4802 ret = -ENOMEM;
4803 }
4804 }
4805 } else if (string_match(optname, "pidfile") || opt == 'p') {
4806 if (!arg || *arg == '\0') {
4807 ret = -EINVAL;
4808 goto end;
4809 }
4810 if (lttng_is_setuid_setgid()) {
4811 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4812 "-p, --pidfile");
4813 } else {
4814 config_string_set(&config.pid_file_path, strdup(arg));
4815 if (!config.pid_file_path.value) {
4816 PERROR("strdup");
4817 ret = -ENOMEM;
4818 }
4819 }
4820 } else if (string_match(optname, "agent-tcp-port")) {
4821 if (!arg || *arg == '\0') {
4822 ret = -EINVAL;
4823 goto end;
4824 }
4825 if (lttng_is_setuid_setgid()) {
4826 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4827 "--agent-tcp-port");
4828 } else {
4829 unsigned long v;
4830
4831 errno = 0;
4832 v = strtoul(arg, NULL, 0);
4833 if (errno != 0 || !isdigit(arg[0])) {
4834 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4835 return -1;
4836 }
4837 if (v == 0 || v >= 65535) {
4838 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4839 return -1;
4840 }
4841 config.agent_tcp_port = (uint32_t) v;
4842 DBG3("Agent TCP port set to non default: %u", config.agent_tcp_port);
4843 }
4844 } else if (string_match(optname, "load") || opt == 'l') {
4845 if (!arg || *arg == '\0') {
4846 ret = -EINVAL;
4847 goto end;
4848 }
4849 if (lttng_is_setuid_setgid()) {
4850 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4851 "-l, --load");
4852 } else {
4853 config_string_set(&config.load_session_path, strdup(arg));
4854 if (!config.load_session_path.value) {
4855 PERROR("strdup");
4856 ret = -ENOMEM;
4857 }
4858 }
4859 } else if (string_match(optname, "kmod-probes")) {
4860 if (!arg || *arg == '\0') {
4861 ret = -EINVAL;
4862 goto end;
4863 }
4864 if (lttng_is_setuid_setgid()) {
4865 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4866 "--kmod-probes");
4867 } else {
4868 config_string_set(&config.kmod_probes_list, strdup(arg));
4869 if (!config.kmod_probes_list.value) {
4870 PERROR("strdup");
4871 ret = -ENOMEM;
4872 }
4873 }
4874 } else if (string_match(optname, "extra-kmod-probes")) {
4875 if (!arg || *arg == '\0') {
4876 ret = -EINVAL;
4877 goto end;
4878 }
4879 if (lttng_is_setuid_setgid()) {
4880 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4881 "--extra-kmod-probes");
4882 } else {
4883 config_string_set(&config.kmod_extra_probes_list,
4884 strdup(arg));
4885 if (!config.kmod_extra_probes_list.value) {
4886 PERROR("strdup");
4887 ret = -ENOMEM;
4888 }
4889 }
4890 } else if (string_match(optname, "config") || opt == 'f') {
4891 /* This is handled in set_options() thus silent skip. */
4892 goto end;
4893 } else {
4894 /* Unknown option or other error.
4895 * Error is printed by getopt, just return */
4896 ret = -1;
4897 }
4898
4899 end:
4900 if (ret == -EINVAL) {
4901 const char *opt_name = "unknown";
4902 int i;
4903
4904 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4905 i++) {
4906 if (opt == long_options[i].val) {
4907 opt_name = long_options[i].name;
4908 break;
4909 }
4910 }
4911
4912 WARN("Invalid argument provided for option \"%s\", using default value.",
4913 opt_name);
4914 }
4915
4916 return ret;
4917 }
4918
4919 /*
4920 * config_entry_handler_cb used to handle options read from a config file.
4921 * See config_entry_handler_cb comment in common/config/session-config.h for the
4922 * return value conventions.
4923 */
4924 static int config_entry_handler(const struct config_entry *entry, void *unused)
4925 {
4926 int ret = 0, i;
4927
4928 if (!entry || !entry->name || !entry->value) {
4929 ret = -EINVAL;
4930 goto end;
4931 }
4932
4933 /* Check if the option is to be ignored */
4934 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4935 if (!strcmp(entry->name, config_ignore_options[i])) {
4936 goto end;
4937 }
4938 }
4939
4940 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4941 i++) {
4942
4943 /* Ignore if not fully matched. */
4944 if (strcmp(entry->name, long_options[i].name)) {
4945 continue;
4946 }
4947
4948 /*
4949 * If the option takes no argument on the command line, we have to
4950 * check if the value is "true". We support non-zero numeric values,
4951 * true, on and yes.
4952 */
4953 if (!long_options[i].has_arg) {
4954 ret = config_parse_value(entry->value);
4955 if (ret <= 0) {
4956 if (ret) {
4957 WARN("Invalid configuration value \"%s\" for option %s",
4958 entry->value, entry->name);
4959 }
4960 /* False, skip boolean config option. */
4961 goto end;
4962 }
4963 }
4964
4965 ret = set_option(long_options[i].val, entry->value, entry->name);
4966 goto end;
4967 }
4968
4969 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4970
4971 end:
4972 return ret;
4973 }
4974
4975 /*
4976 * daemon configuration loading and argument parsing
4977 */
4978 static int set_options(int argc, char **argv)
4979 {
4980 int ret = 0, c = 0, option_index = 0;
4981 int orig_optopt = optopt, orig_optind = optind;
4982 char *optstring;
4983 const char *config_path = NULL;
4984
4985 optstring = utils_generate_optstring(long_options,
4986 sizeof(long_options) / sizeof(struct option));
4987 if (!optstring) {
4988 ret = -ENOMEM;
4989 goto end;
4990 }
4991
4992 /* Check for the --config option */
4993 while ((c = getopt_long(argc, argv, optstring, long_options,
4994 &option_index)) != -1) {
4995 if (c == '?') {
4996 ret = -EINVAL;
4997 goto end;
4998 } else if (c != 'f') {
4999 /* if not equal to --config option. */
5000 continue;
5001 }
5002
5003 if (lttng_is_setuid_setgid()) {
5004 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5005 "-f, --config");
5006 } else {
5007 config_path = utils_expand_path(optarg);
5008 if (!config_path) {
5009 ERR("Failed to resolve path: %s", optarg);
5010 }
5011 }
5012 }
5013
5014 ret = config_get_section_entries(config_path, config_section_name,
5015 config_entry_handler, NULL);
5016 if (ret) {
5017 if (ret > 0) {
5018 ERR("Invalid configuration option at line %i", ret);
5019 ret = -1;
5020 }
5021 goto end;
5022 }
5023
5024 /* Reset getopt's global state */
5025 optopt = orig_optopt;
5026 optind = orig_optind;
5027 while (1) {
5028 option_index = -1;
5029 /*
5030 * getopt_long() will not set option_index if it encounters a
5031 * short option.
5032 */
5033 c = getopt_long(argc, argv, optstring, long_options,
5034 &option_index);
5035 if (c == -1) {
5036 break;
5037 }
5038
5039 /*
5040 * Pass NULL as the long option name if popt left the index
5041 * unset.
5042 */
5043 ret = set_option(c, optarg,
5044 option_index < 0 ? NULL :
5045 long_options[option_index].name);
5046 if (ret < 0) {
5047 break;
5048 }
5049 }
5050
5051 end:
5052 free(optstring);
5053 return ret;
5054 }
5055
5056 /*
5057 * Creates the two needed socket by the daemon.
5058 * apps_sock - The communication socket for all UST apps.
5059 * client_sock - The communication of the cli tool (lttng).
5060 */
5061 static int init_daemon_socket(void)
5062 {
5063 int ret = 0;
5064 mode_t old_umask;
5065
5066 old_umask = umask(0);
5067
5068 /* Create client tool unix socket */
5069 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5070 if (client_sock < 0) {
5071 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5072 ret = -1;
5073 goto end;
5074 }
5075
5076 /* Set the cloexec flag */
5077 ret = utils_set_fd_cloexec(client_sock);
5078 if (ret < 0) {
5079 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5080 "Continuing but note that the consumer daemon will have a "
5081 "reference to this socket on exec()", client_sock);
5082 }
5083
5084 /* File permission MUST be 660 */
5085 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5086 if (ret < 0) {
5087 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5088 PERROR("chmod");
5089 goto end;
5090 }
5091
5092 /* Create the application unix socket */
5093 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5094 if (apps_sock < 0) {
5095 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5096 ret = -1;
5097 goto end;
5098 }
5099
5100 /* Set the cloexec flag */
5101 ret = utils_set_fd_cloexec(apps_sock);
5102 if (ret < 0) {
5103 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5104 "Continuing but note that the consumer daemon will have a "
5105 "reference to this socket on exec()", apps_sock);
5106 }
5107
5108 /* File permission MUST be 666 */
5109 ret = chmod(config.apps_unix_sock_path.value,
5110 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5111 if (ret < 0) {
5112 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5113 PERROR("chmod");
5114 goto end;
5115 }
5116
5117 DBG3("Session daemon client socket %d and application socket %d created",
5118 client_sock, apps_sock);
5119
5120 end:
5121 umask(old_umask);
5122 return ret;
5123 }
5124
5125 /*
5126 * Check if the global socket is available, and if a daemon is answering at the
5127 * other side. If yes, error is returned.
5128 */
5129 static int check_existing_daemon(void)
5130 {
5131 /* Is there anybody out there ? */
5132 if (lttng_session_daemon_alive()) {
5133 return -EEXIST;
5134 }
5135
5136 return 0;
5137 }
5138
5139 /*
5140 * Set the tracing group gid onto the client socket.
5141 *
5142 * Race window between mkdir and chown is OK because we are going from more
5143 * permissive (root.root) to less permissive (root.tracing).
5144 */
5145 static int set_permissions(char *rundir)
5146 {
5147 int ret;
5148 gid_t gid;
5149
5150 gid = utils_get_group_id(config.tracing_group_name.value);
5151
5152 /* Set lttng run dir */
5153 ret = chown(rundir, 0, gid);
5154 if (ret < 0) {
5155 ERR("Unable to set group on %s", rundir);
5156 PERROR("chown");
5157 }
5158
5159 /*
5160 * Ensure all applications and tracing group can search the run
5161 * dir. Allow everyone to read the directory, since it does not
5162 * buy us anything to hide its content.
5163 */
5164 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5165 if (ret < 0) {
5166 ERR("Unable to set permissions on %s", rundir);
5167 PERROR("chmod");
5168 }
5169
5170 /* lttng client socket path */
5171 ret = chown(config.client_unix_sock_path.value, 0, gid);
5172 if (ret < 0) {
5173 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5174 PERROR("chown");
5175 }
5176
5177 /* kconsumer error socket path */
5178 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5179 if (ret < 0) {
5180 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5181 PERROR("chown");
5182 }
5183
5184 /* 64-bit ustconsumer error socket path */
5185 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5186 if (ret < 0) {
5187 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5188 PERROR("chown");
5189 }
5190
5191 /* 32-bit ustconsumer compat32 error socket path */
5192 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5193 if (ret < 0) {
5194 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5195 PERROR("chown");
5196 }
5197
5198 DBG("All permissions are set");
5199
5200 return ret;
5201 }
5202
5203 /*
5204 * Create the lttng run directory needed for all global sockets and pipe.
5205 */
5206 static int create_lttng_rundir(void)
5207 {
5208 int ret;
5209
5210 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5211
5212 ret = mkdir(config.rundir.value, S_IRWXU);
5213 if (ret < 0) {
5214 if (errno != EEXIST) {
5215 ERR("Unable to create %s", config.rundir.value);
5216 goto error;
5217 } else {
5218 ret = 0;
5219 }
5220 }
5221
5222 error:
5223 return ret;
5224 }
5225
5226 /*
5227 * Setup sockets and directory needed by the consumerds' communication with the
5228 * session daemon.
5229 */
5230 static int set_consumer_sockets(struct consumer_data *consumer_data)
5231 {
5232 int ret;
5233 char *path = NULL;
5234
5235 switch (consumer_data->type) {
5236 case LTTNG_CONSUMER_KERNEL:
5237 path = config.kconsumerd_path.value;
5238 break;
5239 case LTTNG_CONSUMER64_UST:
5240 path = config.consumerd64_path.value;
5241 break;
5242 case LTTNG_CONSUMER32_UST:
5243 path = config.consumerd32_path.value;
5244 break;
5245 default:
5246 ERR("Consumer type unknown");
5247 ret = -EINVAL;
5248 goto error;
5249 }
5250 assert(path);
5251
5252 DBG2("Creating consumer directory: %s", path);
5253
5254 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5255 if (ret < 0 && errno != EEXIST) {
5256 PERROR("mkdir");
5257 ERR("Failed to create %s", path);
5258 goto error;
5259 }
5260 if (is_root) {
5261 ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value));
5262 if (ret < 0) {
5263 ERR("Unable to set group on %s", path);
5264 PERROR("chown");
5265 goto error;
5266 }
5267 }
5268
5269 /* Create the consumerd error unix socket */
5270 consumer_data->err_sock =
5271 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5272 if (consumer_data->err_sock < 0) {
5273 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5274 ret = -1;
5275 goto error;
5276 }
5277
5278 /*
5279 * Set the CLOEXEC flag. Return code is useless because either way, the
5280 * show must go on.
5281 */
5282 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5283 if (ret < 0) {
5284 PERROR("utils_set_fd_cloexec");
5285 /* continue anyway */
5286 }
5287
5288 /* File permission MUST be 660 */
5289 ret = chmod(consumer_data->err_unix_sock_path,
5290 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5291 if (ret < 0) {
5292 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5293 PERROR("chmod");
5294 goto error;
5295 }
5296
5297 error:
5298 return ret;
5299 }
5300
5301 /*
5302 * Signal handler for the daemon
5303 *
5304 * Simply stop all worker threads, leaving main() return gracefully after
5305 * joining all threads and calling cleanup().
5306 */
5307 static void sighandler(int sig)
5308 {
5309 switch (sig) {
5310 case SIGINT:
5311 DBG("SIGINT caught");
5312 stop_threads();
5313 break;
5314 case SIGTERM:
5315 DBG("SIGTERM caught");
5316 stop_threads();
5317 break;
5318 case SIGUSR1:
5319 CMM_STORE_SHARED(recv_child_signal, 1);
5320 break;
5321 default:
5322 break;
5323 }
5324 }
5325
5326 /*
5327 * Setup signal handler for :
5328 * SIGINT, SIGTERM, SIGPIPE
5329 */
5330 static int set_signal_handler(void)
5331 {
5332 int ret = 0;
5333 struct sigaction sa;
5334 sigset_t sigset;
5335
5336 if ((ret = sigemptyset(&sigset)) < 0) {
5337 PERROR("sigemptyset");
5338 return ret;
5339 }
5340
5341 sa.sa_mask = sigset;
5342 sa.sa_flags = 0;
5343
5344 sa.sa_handler = sighandler;
5345 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5346 PERROR("sigaction");
5347 return ret;
5348 }
5349
5350 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5351 PERROR("sigaction");
5352 return ret;
5353 }
5354
5355 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5356 PERROR("sigaction");
5357 return ret;
5358 }
5359
5360 sa.sa_handler = SIG_IGN;
5361 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5362 PERROR("sigaction");
5363 return ret;
5364 }
5365
5366 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5367
5368 return ret;
5369 }
5370
5371 /*
5372 * Set open files limit to unlimited. This daemon can open a large number of
5373 * file descriptors in order to consume multiple kernel traces.
5374 */
5375 static void set_ulimit(void)
5376 {
5377 int ret;
5378 struct rlimit lim;
5379
5380 /* The kernel does not allow an infinite limit for open files */
5381 lim.rlim_cur = 65535;
5382 lim.rlim_max = 65535;
5383
5384 ret = setrlimit(RLIMIT_NOFILE, &lim);
5385 if (ret < 0) {
5386 PERROR("failed to set open files limit");
5387 }
5388 }
5389
5390 static int write_pidfile(void)
5391 {
5392 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5393 }
5394
5395 /*
5396 * Create lockfile using the rundir and return its fd.
5397 */
5398 static int create_lockfile(void)
5399 {
5400 return utils_create_lock_file(config.lock_file_path.value);
5401 }
5402
5403 /*
5404 * Write agent TCP port using the rundir.
5405 */
5406 static int write_agent_port(void)
5407 {
5408 return utils_create_pid_file(config.agent_tcp_port,
5409 config.agent_port_file_path.value);
5410 }
5411
5412 static int set_clock_plugin_env(void)
5413 {
5414 int ret = 0;
5415 char *env_value = NULL;
5416
5417 if (!config.lttng_ust_clock_plugin.value) {
5418 goto end;
5419 }
5420
5421 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5422 config.lttng_ust_clock_plugin.value);
5423 if (ret < 0) {
5424 PERROR("asprintf");
5425 goto end;
5426 }
5427
5428 ret = putenv(env_value);
5429 if (ret) {
5430 free(env_value);
5431 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5432 goto end;
5433 }
5434
5435 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5436 config.lttng_ust_clock_plugin.value);
5437 end:
5438 return ret;
5439 }
5440
5441 /*
5442 * main
5443 */
5444 int main(int argc, char **argv)
5445 {
5446 int ret = 0, retval = 0;
5447 void *status;
5448 const char *env_app_timeout;
5449 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5450 *ust64_channel_monitor_pipe = NULL,
5451 *kernel_channel_monitor_pipe = NULL;
5452 bool notification_thread_running = false;
5453
5454 init_kernel_workarounds();
5455
5456 rcu_register_thread();
5457
5458 if (set_signal_handler()) {
5459 retval = -1;
5460 goto exit_set_signal_handler;
5461 }
5462
5463 page_size = sysconf(_SC_PAGESIZE);
5464 if (page_size < 0) {
5465 PERROR("sysconf _SC_PAGESIZE");
5466 page_size = LONG_MAX;
5467 WARN("Fallback page size to %ld", page_size);
5468 }
5469
5470 ret = sessiond_config_init(&config);
5471 if (ret) {
5472 retval = -1;
5473 goto exit_set_signal_handler;
5474 }
5475
5476 /*
5477 * Parse arguments and load the daemon configuration file.
5478 *
5479 * We have an exit_options exit path to free memory reserved by
5480 * set_options. This is needed because the rest of sessiond_cleanup()
5481 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5482 * depends on set_options.
5483 */
5484 progname = argv[0];
5485 if (set_options(argc, argv)) {
5486 retval = -1;
5487 goto exit_options;
5488 }
5489
5490 /* Init config from environment variables. */
5491 sessiond_config_apply_env_config(&config);
5492
5493 /*
5494 * Resolve all paths received as arguments, configuration option, or
5495 * through environment variable as absolute paths. This is necessary
5496 * since daemonizing causes the sessiond's current working directory
5497 * to '/'.
5498 */
5499 ret = sessiond_config_resolve_paths(&config);
5500 if (ret) {
5501 goto exit_options;
5502 }
5503
5504 /* Apply config. */
5505 lttng_opt_verbose = config.verbose;
5506 lttng_opt_quiet = config.quiet;
5507 kconsumer_data.err_unix_sock_path =
5508 config.kconsumerd_err_unix_sock_path.value;
5509 kconsumer_data.cmd_unix_sock_path =
5510 config.kconsumerd_cmd_unix_sock_path.value;
5511 ustconsumer32_data.err_unix_sock_path =
5512 config.consumerd32_err_unix_sock_path.value;
5513 ustconsumer32_data.cmd_unix_sock_path =
5514 config.consumerd32_cmd_unix_sock_path.value;
5515 ustconsumer64_data.err_unix_sock_path =
5516 config.consumerd64_err_unix_sock_path.value;
5517 ustconsumer64_data.cmd_unix_sock_path =
5518 config.consumerd64_cmd_unix_sock_path.value;
5519 set_clock_plugin_env();
5520
5521 sessiond_config_log(&config);
5522
5523 /* Daemonize */
5524 if (config.daemonize || config.background) {
5525 int i;
5526
5527 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5528 !config.background);
5529 if (ret < 0) {
5530 retval = -1;
5531 goto exit_options;
5532 }
5533
5534 /*
5535 * We are in the child. Make sure all other file descriptors are
5536 * closed, in case we are called with more opened file
5537 * descriptors than the standard ones.
5538 */
5539 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5540 (void) close(i);
5541 }
5542 }
5543
5544 if (run_as_create_worker(argv[0]) < 0) {
5545 goto exit_create_run_as_worker_cleanup;
5546 }
5547
5548 /*
5549 * Starting from here, we can create threads. This needs to be after
5550 * lttng_daemonize due to RCU.
5551 */
5552
5553 /*
5554 * Initialize the health check subsystem. This call should set the
5555 * appropriate time values.
5556 */
5557 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5558 if (!health_sessiond) {
5559 PERROR("health_app_create error");
5560 retval = -1;
5561 goto exit_health_sessiond_cleanup;
5562 }
5563
5564 /* Create thread to clean up RCU hash tables */
5565 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5566 retval = -1;
5567 goto exit_ht_cleanup;
5568 }
5569
5570 /* Create thread quit pipe */
5571 if (init_thread_quit_pipe()) {
5572 retval = -1;
5573 goto exit_init_data;
5574 }
5575
5576 /* Check if daemon is UID = 0 */
5577 is_root = !getuid();
5578
5579 if (create_lttng_rundir()) {
5580 retval = -1;
5581 goto exit_init_data;
5582 }
5583
5584 if (is_root) {
5585 /* Create global run dir with root access */
5586
5587 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5588 if (!kernel_channel_monitor_pipe) {
5589 ERR("Failed to create kernel consumer channel monitor pipe");
5590 retval = -1;
5591 goto exit_init_data;
5592 }
5593 kconsumer_data.channel_monitor_pipe =
5594 lttng_pipe_release_writefd(
5595 kernel_channel_monitor_pipe);
5596 if (kconsumer_data.channel_monitor_pipe < 0) {
5597 retval = -1;
5598 goto exit_init_data;
5599 }
5600 }
5601
5602 lockfile_fd = create_lockfile();
5603 if (lockfile_fd < 0) {
5604 retval = -1;
5605 goto exit_init_data;
5606 }
5607
5608 /* Set consumer initial state */
5609 kernel_consumerd_state = CONSUMER_STOPPED;
5610 ust_consumerd_state = CONSUMER_STOPPED;
5611
5612 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5613 if (!ust32_channel_monitor_pipe) {
5614 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5615 retval = -1;
5616 goto exit_init_data;
5617 }
5618 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5619 ust32_channel_monitor_pipe);
5620 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5621 retval = -1;
5622 goto exit_init_data;
5623 }
5624
5625 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5626 if (!ust64_channel_monitor_pipe) {
5627 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5628 retval = -1;
5629 goto exit_init_data;
5630 }
5631 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5632 ust64_channel_monitor_pipe);
5633 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5634 retval = -1;
5635 goto exit_init_data;
5636 }
5637
5638 /*
5639 * See if daemon already exist.
5640 */
5641 if (check_existing_daemon()) {
5642 ERR("Already running daemon.\n");
5643 /*
5644 * We do not goto exit because we must not cleanup()
5645 * because a daemon is already running.
5646 */
5647 retval = -1;
5648 goto exit_init_data;
5649 }
5650
5651 /*
5652 * Init UST app hash table. Alloc hash table before this point since
5653 * cleanup() can get called after that point.
5654 */
5655 if (ust_app_ht_alloc()) {
5656 ERR("Failed to allocate UST app hash table");
5657 retval = -1;
5658 goto exit_init_data;
5659 }
5660
5661 /*
5662 * Initialize agent app hash table. We allocate the hash table here
5663 * since cleanup() can get called after this point.
5664 */
5665 if (agent_app_ht_alloc()) {
5666 ERR("Failed to allocate Agent app hash table");
5667 retval = -1;
5668 goto exit_init_data;
5669 }
5670
5671 /*
5672 * These actions must be executed as root. We do that *after* setting up
5673 * the sockets path because we MUST make the check for another daemon using
5674 * those paths *before* trying to set the kernel consumer sockets and init
5675 * kernel tracer.
5676 */
5677 if (is_root) {
5678 if (set_consumer_sockets(&kconsumer_data)) {
5679 retval = -1;
5680 goto exit_init_data;
5681 }
5682
5683 /* Setup kernel tracer */
5684 if (!config.no_kernel) {
5685 init_kernel_tracer();
5686 if (kernel_tracer_fd >= 0) {
5687 ret = syscall_init_table();
5688 if (ret < 0) {
5689 ERR("Unable to populate syscall table. "
5690 "Syscall tracing won't work "
5691 "for this session daemon.");
5692 }
5693 }
5694 }
5695
5696 /* Set ulimit for open files */
5697 set_ulimit();
5698 }
5699 /* init lttng_fd tracking must be done after set_ulimit. */
5700 lttng_fd_init();
5701
5702 if (set_consumer_sockets(&ustconsumer64_data)) {
5703 retval = -1;
5704 goto exit_init_data;
5705 }
5706
5707 if (set_consumer_sockets(&ustconsumer32_data)) {
5708 retval = -1;
5709 goto exit_init_data;
5710 }
5711
5712 /* Setup the needed unix socket */
5713 if (init_daemon_socket()) {
5714 retval = -1;
5715 goto exit_init_data;
5716 }
5717
5718 /* Set credentials to socket */
5719 if (is_root && set_permissions(config.rundir.value)) {
5720 retval = -1;
5721 goto exit_init_data;
5722 }
5723
5724 /* Get parent pid if -S, --sig-parent is specified. */
5725 if (config.sig_parent) {
5726 ppid = getppid();
5727 }
5728
5729 /* Setup the kernel pipe for waking up the kernel thread */
5730 if (is_root && !config.no_kernel) {
5731 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5732 retval = -1;
5733 goto exit_init_data;
5734 }
5735 }
5736
5737 /* Setup the thread apps communication pipe. */
5738 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5739 retval = -1;
5740 goto exit_init_data;
5741 }
5742
5743 /* Setup the thread apps notify communication pipe. */
5744 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5745 retval = -1;
5746 goto exit_init_data;
5747 }
5748
5749 /* Initialize global buffer per UID and PID registry. */
5750 buffer_reg_init_uid_registry();
5751 buffer_reg_init_pid_registry();
5752
5753 /* Init UST command queue. */
5754 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5755
5756 /*
5757 * Get session list pointer. This pointer MUST NOT be free'd. This list
5758 * is statically declared in session.c
5759 */
5760 session_list_ptr = session_get_list();
5761
5762 cmd_init();
5763
5764 /* Check for the application socket timeout env variable. */
5765 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5766 if (env_app_timeout) {
5767 config.app_socket_timeout = atoi(env_app_timeout);
5768 } else {
5769 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5770 }
5771
5772 ret = write_pidfile();
5773 if (ret) {
5774 ERR("Error in write_pidfile");
5775 retval = -1;
5776 goto exit_init_data;
5777 }
5778 ret = write_agent_port();
5779 if (ret) {
5780 ERR("Error in write_agent_port");
5781 retval = -1;
5782 goto exit_init_data;
5783 }
5784
5785 /* Initialize communication library */
5786 lttcomm_init();
5787 /* Initialize TCP timeout values */
5788 lttcomm_inet_init();
5789
5790 if (load_session_init_data(&load_info) < 0) {
5791 retval = -1;
5792 goto exit_init_data;
5793 }
5794 load_info->path = config.load_session_path.value;
5795
5796 /* Create health-check thread. */
5797 ret = pthread_create(&health_thread, default_pthread_attr(),
5798 thread_manage_health, (void *) NULL);
5799 if (ret) {
5800 errno = ret;
5801 PERROR("pthread_create health");
5802 retval = -1;
5803 goto exit_health;
5804 }
5805
5806 /* notification_thread_data acquires the pipes' read side. */
5807 notification_thread_handle = notification_thread_handle_create(
5808 ust32_channel_monitor_pipe,
5809 ust64_channel_monitor_pipe,
5810 kernel_channel_monitor_pipe);
5811 if (!notification_thread_handle) {
5812 retval = -1;
5813 ERR("Failed to create notification thread shared data");
5814 stop_threads();
5815 goto exit_notification;
5816 }
5817
5818 /* Create notification thread. */
5819 ret = pthread_create(&notification_thread, default_pthread_attr(),
5820 thread_notification, notification_thread_handle);
5821 if (ret) {
5822 errno = ret;
5823 PERROR("pthread_create notification");
5824 retval = -1;
5825 stop_threads();
5826 goto exit_notification;
5827 }
5828 notification_thread_running = true;
5829
5830 /* Create thread to manage the client socket */
5831 ret = pthread_create(&client_thread, default_pthread_attr(),
5832 thread_manage_clients, (void *) NULL);
5833 if (ret) {
5834 errno = ret;
5835 PERROR("pthread_create clients");
5836 retval = -1;
5837 stop_threads();
5838 goto exit_client;
5839 }
5840
5841 /* Create thread to dispatch registration */
5842 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
5843 thread_dispatch_ust_registration, (void *) NULL);
5844 if (ret) {
5845 errno = ret;
5846 PERROR("pthread_create dispatch");
5847 retval = -1;
5848 stop_threads();
5849 goto exit_dispatch;
5850 }
5851
5852 /* Create thread to manage application registration. */
5853 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
5854 thread_registration_apps, (void *) NULL);
5855 if (ret) {
5856 errno = ret;
5857 PERROR("pthread_create registration");
5858 retval = -1;
5859 stop_threads();
5860 goto exit_reg_apps;
5861 }
5862
5863 /* Create thread to manage application socket */
5864 ret = pthread_create(&apps_thread, default_pthread_attr(),
5865 thread_manage_apps, (void *) NULL);
5866 if (ret) {
5867 errno = ret;
5868 PERROR("pthread_create apps");
5869 retval = -1;
5870 stop_threads();
5871 goto exit_apps;
5872 }
5873
5874 /* Create thread to manage application notify socket */
5875 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
5876 ust_thread_manage_notify, (void *) NULL);
5877 if (ret) {
5878 errno = ret;
5879 PERROR("pthread_create notify");
5880 retval = -1;
5881 stop_threads();
5882 goto exit_apps_notify;
5883 }
5884
5885 /* Create agent registration thread. */
5886 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
5887 agent_thread_manage_registration, (void *) NULL);
5888 if (ret) {
5889 errno = ret;
5890 PERROR("pthread_create agent");
5891 retval = -1;
5892 stop_threads();
5893 goto exit_agent_reg;
5894 }
5895
5896 /* Don't start this thread if kernel tracing is not requested nor root */
5897 if (is_root && !config.no_kernel) {
5898 /* Create kernel thread to manage kernel event */
5899 ret = pthread_create(&kernel_thread, default_pthread_attr(),
5900 thread_manage_kernel, (void *) NULL);
5901 if (ret) {
5902 errno = ret;
5903 PERROR("pthread_create kernel");
5904 retval = -1;
5905 stop_threads();
5906 goto exit_kernel;
5907 }
5908 }
5909
5910 /* Create session loading thread. */
5911 ret = pthread_create(&load_session_thread, default_pthread_attr(),
5912 thread_load_session, load_info);
5913 if (ret) {
5914 errno = ret;
5915 PERROR("pthread_create load_session_thread");
5916 retval = -1;
5917 stop_threads();
5918 goto exit_load_session;
5919 }
5920
5921 /*
5922 * This is where we start awaiting program completion (e.g. through
5923 * signal that asks threads to teardown).
5924 */
5925
5926 ret = pthread_join(load_session_thread, &status);
5927 if (ret) {
5928 errno = ret;
5929 PERROR("pthread_join load_session_thread");
5930 retval = -1;
5931 }
5932 exit_load_session:
5933
5934 if (is_root && !config.no_kernel) {
5935 ret = pthread_join(kernel_thread, &status);
5936 if (ret) {
5937 errno = ret;
5938 PERROR("pthread_join");
5939 retval = -1;
5940 }
5941 }
5942 exit_kernel:
5943
5944 ret = pthread_join(agent_reg_thread, &status);
5945 if (ret) {
5946 errno = ret;
5947 PERROR("pthread_join agent");
5948 retval = -1;
5949 }
5950 exit_agent_reg:
5951
5952 ret = pthread_join(apps_notify_thread, &status);
5953 if (ret) {
5954 errno = ret;
5955 PERROR("pthread_join apps notify");
5956 retval = -1;
5957 }
5958 exit_apps_notify:
5959
5960 ret = pthread_join(apps_thread, &status);
5961 if (ret) {
5962 errno = ret;
5963 PERROR("pthread_join apps");
5964 retval = -1;
5965 }
5966 exit_apps:
5967
5968 ret = pthread_join(reg_apps_thread, &status);
5969 if (ret) {
5970 errno = ret;
5971 PERROR("pthread_join");
5972 retval = -1;
5973 }
5974 exit_reg_apps:
5975
5976 /*
5977 * Join dispatch thread after joining reg_apps_thread to ensure
5978 * we don't leak applications in the queue.
5979 */
5980 ret = pthread_join(dispatch_thread, &status);
5981 if (ret) {
5982 errno = ret;
5983 PERROR("pthread_join");
5984 retval = -1;
5985 }
5986 exit_dispatch:
5987
5988 ret = pthread_join(client_thread, &status);
5989 if (ret) {
5990 errno = ret;
5991 PERROR("pthread_join");
5992 retval = -1;
5993 }
5994
5995 exit_client:
5996 exit_notification:
5997 ret = pthread_join(health_thread, &status);
5998 if (ret) {
5999 errno = ret;
6000 PERROR("pthread_join health thread");
6001 retval = -1;
6002 }
6003
6004 exit_health:
6005 exit_init_data:
6006 /*
6007 * Wait for all pending call_rcu work to complete before tearing
6008 * down data structures. call_rcu worker may be trying to
6009 * perform lookups in those structures.
6010 */
6011 rcu_barrier();
6012 /*
6013 * sessiond_cleanup() is called when no other thread is running, except
6014 * the ht_cleanup thread, which is needed to destroy the hash tables.
6015 */
6016 rcu_thread_online();
6017 sessiond_cleanup();
6018
6019 /*
6020 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6021 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6022 * the queue is empty before shutting down the clean-up thread.
6023 */
6024 rcu_barrier();
6025
6026 /*
6027 * The teardown of the notification system is performed after the
6028 * session daemon's teardown in order to allow it to be notified
6029 * of the active session and channels at the moment of the teardown.
6030 */
6031 if (notification_thread_handle) {
6032 if (notification_thread_running) {
6033 notification_thread_command_quit(
6034 notification_thread_handle);
6035 ret = pthread_join(notification_thread, &status);
6036 if (ret) {
6037 errno = ret;
6038 PERROR("pthread_join notification thread");
6039 retval = -1;
6040 }
6041 }
6042 notification_thread_handle_destroy(notification_thread_handle);
6043 }
6044
6045 rcu_thread_offline();
6046 rcu_unregister_thread();
6047
6048 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6049 if (ret) {
6050 retval = -1;
6051 }
6052 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6053 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6054 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6055 exit_ht_cleanup:
6056
6057 health_app_destroy(health_sessiond);
6058 exit_health_sessiond_cleanup:
6059 exit_create_run_as_worker_cleanup:
6060
6061 exit_options:
6062 sessiond_cleanup_options();
6063
6064 exit_set_signal_handler:
6065 if (!retval) {
6066 exit(EXIT_SUCCESS);
6067 } else {
6068 exit(EXIT_FAILURE);
6069 }
6070 }
This page took 0.17939 seconds and 5 git commands to generate.