Fix: streamline ret/errno of run_as()
[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 _GNU_SOURCE
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <paths.h>
26 #include <pthread.h>
27 #include <signal.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <string.h>
31 #include <inttypes.h>
32 #include <sys/mman.h>
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
36 #include <sys/stat.h>
37 #include <sys/types.h>
38 #include <sys/wait.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <config.h>
42
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
53
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
56 #include "channel.h"
57 #include "cmd.h"
58 #include "consumer.h"
59 #include "context.h"
60 #include "event.h"
61 #include "kernel.h"
62 #include "kernel-consumer.h"
63 #include "modprobe.h"
64 #include "shm.h"
65 #include "ust-ctl.h"
66 #include "ust-consumer.h"
67 #include "utils.h"
68 #include "fd-limit.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
73 #include "save.h"
74 #include "load-session-thread.h"
75 #include "syscall.h"
76 #include "agent.h"
77
78 #define CONSUMERD_FILE "lttng-consumerd"
79
80 const char *progname;
81 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
82 static int tracing_group_name_override;
83 static char *opt_pidfile;
84 static int opt_sig_parent;
85 static int opt_verbose_consumer;
86 static int opt_daemon, opt_background;
87 static int opt_no_kernel;
88 static char *opt_load_session_path;
89 static pid_t ppid; /* Parent PID for --sig-parent option */
90 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
91 static char *rundir;
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_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
104 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
105 .err_sock = -1,
106 .cmd_sock = -1,
107 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
108 .lock = PTHREAD_MUTEX_INITIALIZER,
109 .cond = PTHREAD_COND_INITIALIZER,
110 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
111 };
112 static struct consumer_data ustconsumer64_data = {
113 .type = LTTNG_CONSUMER64_UST,
114 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
115 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
116 .err_sock = -1,
117 .cmd_sock = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123 static struct consumer_data ustconsumer32_data = {
124 .type = LTTNG_CONSUMER32_UST,
125 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
126 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
127 .err_sock = -1,
128 .cmd_sock = -1,
129 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
130 .lock = PTHREAD_MUTEX_INITIALIZER,
131 .cond = PTHREAD_COND_INITIALIZER,
132 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
133 };
134
135 /* Command line options */
136 static const struct option long_options[] = {
137 { "client-sock", required_argument, 0, 'c' },
138 { "apps-sock", required_argument, 0, 'a' },
139 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
140 { "kconsumerd-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
144 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
145 { "consumerd32-path", required_argument, 0, '\0' },
146 { "consumerd32-libdir", required_argument, 0, '\0' },
147 { "consumerd64-path", required_argument, 0, '\0' },
148 { "consumerd64-libdir", required_argument, 0, '\0' },
149 { "daemonize", no_argument, 0, 'd' },
150 { "background", no_argument, 0, 'b' },
151 { "sig-parent", no_argument, 0, 'S' },
152 { "help", no_argument, 0, 'h' },
153 { "group", required_argument, 0, 'g' },
154 { "version", no_argument, 0, 'V' },
155 { "quiet", no_argument, 0, 'q' },
156 { "verbose", no_argument, 0, 'v' },
157 { "verbose-consumer", no_argument, 0, '\0' },
158 { "no-kernel", no_argument, 0, '\0' },
159 { "pidfile", required_argument, 0, 'p' },
160 { "agent-tcp-port", required_argument, 0, '\0' },
161 { "config", required_argument, 0, 'f' },
162 { "load", required_argument, 0, 'l' },
163 { "kmod-probes", required_argument, 0, '\0' },
164 { "extra-kmod-probes", required_argument, 0, '\0' },
165 { NULL, 0, 0, 0 }
166 };
167
168 /* Command line options to ignore from configuration file */
169 static const char *config_ignore_options[] = { "help", "version", "config" };
170
171 /* Shared between threads */
172 static int dispatch_thread_exit;
173
174 /* Global application Unix socket path */
175 static char apps_unix_sock_path[PATH_MAX];
176 /* Global client Unix socket path */
177 static char client_unix_sock_path[PATH_MAX];
178 /* global wait shm path for UST */
179 static char wait_shm_path[PATH_MAX];
180 /* Global health check unix path */
181 static char health_unix_sock_path[PATH_MAX];
182
183 /* Sockets and FDs */
184 static int client_sock = -1;
185 static int apps_sock = -1;
186 int kernel_tracer_fd = -1;
187 static int kernel_poll_pipe[2] = { -1, -1 };
188
189 /*
190 * Quit pipe for all threads. This permits a single cancellation point
191 * for all threads when receiving an event on the pipe.
192 */
193 static int thread_quit_pipe[2] = { -1, -1 };
194 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
195
196 /*
197 * This pipe is used to inform the thread managing application communication
198 * that a command is queued and ready to be processed.
199 */
200 static int apps_cmd_pipe[2] = { -1, -1 };
201
202 int apps_cmd_notify_pipe[2] = { -1, -1 };
203
204 /* Pthread, Mutexes and Semaphores */
205 static pthread_t apps_thread;
206 static pthread_t apps_notify_thread;
207 static pthread_t reg_apps_thread;
208 static pthread_t client_thread;
209 static pthread_t kernel_thread;
210 static pthread_t dispatch_thread;
211 static pthread_t health_thread;
212 static pthread_t ht_cleanup_thread;
213 static pthread_t agent_reg_thread;
214 static pthread_t load_session_thread;
215
216 /*
217 * UST registration command queue. This queue is tied with a futex and uses a N
218 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 *
220 * The thread_registration_apps and thread_dispatch_ust_registration uses this
221 * queue along with the wait/wake scheme. The thread_manage_apps receives down
222 * the line new application socket and monitors it for any I/O error or clean
223 * close that triggers an unregistration of the application.
224 */
225 static struct ust_cmd_queue ust_cmd_queue;
226
227 /*
228 * Pointer initialized before thread creation.
229 *
230 * This points to the tracing session list containing the session count and a
231 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
232 * MUST NOT be taken if you call a public function in session.c.
233 *
234 * The lock is nested inside the structure: session_list_ptr->lock. Please use
235 * session_lock_list and session_unlock_list for lock acquisition.
236 */
237 static struct ltt_session_list *session_list_ptr;
238
239 int ust_consumerd64_fd = -1;
240 int ust_consumerd32_fd = -1;
241
242 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
243 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
244 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
245 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
246 static int consumerd32_bin_override;
247 static int consumerd64_bin_override;
248 static int consumerd32_libdir_override;
249 static int consumerd64_libdir_override;
250
251 static const char *module_proc_lttng = "/proc/lttng";
252
253 /*
254 * Consumer daemon state which is changed when spawning it, killing it or in
255 * case of a fatal error.
256 */
257 enum consumerd_state {
258 CONSUMER_STARTED = 1,
259 CONSUMER_STOPPED = 2,
260 CONSUMER_ERROR = 3,
261 };
262
263 /*
264 * This consumer daemon state is used to validate if a client command will be
265 * able to reach the consumer. If not, the client is informed. For instance,
266 * doing a "lttng start" when the consumer state is set to ERROR will return an
267 * error to the client.
268 *
269 * The following example shows a possible race condition of this scheme:
270 *
271 * consumer thread error happens
272 * client cmd arrives
273 * client cmd checks state -> still OK
274 * consumer thread exit, sets error
275 * client cmd try to talk to consumer
276 * ...
277 *
278 * However, since the consumer is a different daemon, we have no way of making
279 * sure the command will reach it safely even with this state flag. This is why
280 * we consider that up to the state validation during command processing, the
281 * command is safe. After that, we can not guarantee the correctness of the
282 * client request vis-a-vis the consumer.
283 */
284 static enum consumerd_state ust_consumerd_state;
285 static enum consumerd_state kernel_consumerd_state;
286
287 /*
288 * Socket timeout for receiving and sending in seconds.
289 */
290 static int app_socket_timeout;
291
292 /* Set in main() with the current page size. */
293 long page_size;
294
295 /* Application health monitoring */
296 struct health_app *health_sessiond;
297
298 /* Agent TCP port for registration. Used by the agent thread. */
299 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
300
301 /* Am I root or not. */
302 int is_root; /* Set to 1 if the daemon is running as root */
303
304 const char * const config_section_name = "sessiond";
305
306 /* Load session thread information to operate. */
307 struct load_session_thread_data *load_info;
308
309 /* Global hash tables */
310 struct lttng_ht *agent_apps_ht_by_sock = NULL;
311
312 /*
313 * Whether sessiond is ready for commands/health check requests.
314 * NR_LTTNG_SESSIOND_READY must match the number of calls to
315 * sessiond_notify_ready().
316 */
317 #define NR_LTTNG_SESSIOND_READY 3
318 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
319
320 /* Notify parents that we are ready for cmd and health check */
321 LTTNG_HIDDEN
322 void sessiond_notify_ready(void)
323 {
324 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
325 /*
326 * Notify parent pid that we are ready to accept command
327 * for client side. This ppid is the one from the
328 * external process that spawned us.
329 */
330 if (opt_sig_parent) {
331 kill(ppid, SIGUSR1);
332 }
333
334 /*
335 * Notify the parent of the fork() process that we are
336 * ready.
337 */
338 if (opt_daemon || opt_background) {
339 kill(child_ppid, SIGUSR1);
340 }
341 }
342 }
343
344 static
345 void setup_consumerd_path(void)
346 {
347 const char *bin, *libdir;
348
349 /*
350 * Allow INSTALL_BIN_PATH to be used as a target path for the
351 * native architecture size consumer if CONFIG_CONSUMER*_PATH
352 * has not been defined.
353 */
354 #if (CAA_BITS_PER_LONG == 32)
355 if (!consumerd32_bin[0]) {
356 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
357 }
358 if (!consumerd32_libdir[0]) {
359 consumerd32_libdir = INSTALL_LIB_PATH;
360 }
361 #elif (CAA_BITS_PER_LONG == 64)
362 if (!consumerd64_bin[0]) {
363 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
364 }
365 if (!consumerd64_libdir[0]) {
366 consumerd64_libdir = INSTALL_LIB_PATH;
367 }
368 #else
369 #error "Unknown bitness"
370 #endif
371
372 /*
373 * runtime env. var. overrides the build default.
374 */
375 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
376 if (bin) {
377 consumerd32_bin = bin;
378 }
379 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
380 if (bin) {
381 consumerd64_bin = bin;
382 }
383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
384 if (libdir) {
385 consumerd32_libdir = libdir;
386 }
387 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
388 if (libdir) {
389 consumerd64_libdir = libdir;
390 }
391 }
392
393 static
394 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
395 int *a_pipe)
396 {
397 int ret;
398
399 assert(events);
400
401 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
402 if (ret < 0) {
403 goto error;
404 }
405
406 /* Add quit pipe */
407 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
408 if (ret < 0) {
409 goto error;
410 }
411
412 return 0;
413
414 error:
415 return ret;
416 }
417
418 /*
419 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
420 */
421 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
422 {
423 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
424 }
425
426 /*
427 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
428 */
429 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
430 size_t size)
431 {
432 return __sessiond_set_thread_pollset(events, size,
433 ht_cleanup_quit_pipe);
434 }
435
436 static
437 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
438 {
439 if (fd == a_pipe && (events & LPOLLIN)) {
440 return 1;
441 }
442 return 0;
443 }
444
445 /*
446 * Check if the thread quit pipe was triggered.
447 *
448 * Return 1 if it was triggered else 0;
449 */
450 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
451 {
452 return __sessiond_check_thread_quit_pipe(fd, events,
453 thread_quit_pipe[0]);
454 }
455
456 /*
457 * Check if the ht_cleanup thread quit pipe was triggered.
458 *
459 * Return 1 if it was triggered else 0;
460 */
461 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
462 {
463 return __sessiond_check_thread_quit_pipe(fd, events,
464 ht_cleanup_quit_pipe[0]);
465 }
466
467 /*
468 * Init thread quit pipe.
469 *
470 * Return -1 on error or 0 if all pipes are created.
471 */
472 static int __init_thread_quit_pipe(int *a_pipe)
473 {
474 int ret, i;
475
476 ret = pipe(a_pipe);
477 if (ret < 0) {
478 PERROR("thread quit pipe");
479 goto error;
480 }
481
482 for (i = 0; i < 2; i++) {
483 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
484 if (ret < 0) {
485 PERROR("fcntl");
486 goto error;
487 }
488 }
489
490 error:
491 return ret;
492 }
493
494 static int init_thread_quit_pipe(void)
495 {
496 return __init_thread_quit_pipe(thread_quit_pipe);
497 }
498
499 static int init_ht_cleanup_quit_pipe(void)
500 {
501 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
502 }
503
504 /*
505 * Stop all threads by closing the thread quit pipe.
506 */
507 static void stop_threads(void)
508 {
509 int ret;
510
511 /* Stopping all threads */
512 DBG("Terminating all threads");
513 ret = notify_thread_pipe(thread_quit_pipe[1]);
514 if (ret < 0) {
515 ERR("write error on thread quit pipe");
516 }
517
518 /* Dispatch thread */
519 CMM_STORE_SHARED(dispatch_thread_exit, 1);
520 futex_nto1_wake(&ust_cmd_queue.futex);
521 }
522
523 /*
524 * Close every consumer sockets.
525 */
526 static void close_consumer_sockets(void)
527 {
528 int ret;
529
530 if (kconsumer_data.err_sock >= 0) {
531 ret = close(kconsumer_data.err_sock);
532 if (ret < 0) {
533 PERROR("kernel consumer err_sock close");
534 }
535 }
536 if (ustconsumer32_data.err_sock >= 0) {
537 ret = close(ustconsumer32_data.err_sock);
538 if (ret < 0) {
539 PERROR("UST consumerd32 err_sock close");
540 }
541 }
542 if (ustconsumer64_data.err_sock >= 0) {
543 ret = close(ustconsumer64_data.err_sock);
544 if (ret < 0) {
545 PERROR("UST consumerd64 err_sock close");
546 }
547 }
548 if (kconsumer_data.cmd_sock >= 0) {
549 ret = close(kconsumer_data.cmd_sock);
550 if (ret < 0) {
551 PERROR("kernel consumer cmd_sock close");
552 }
553 }
554 if (ustconsumer32_data.cmd_sock >= 0) {
555 ret = close(ustconsumer32_data.cmd_sock);
556 if (ret < 0) {
557 PERROR("UST consumerd32 cmd_sock close");
558 }
559 }
560 if (ustconsumer64_data.cmd_sock >= 0) {
561 ret = close(ustconsumer64_data.cmd_sock);
562 if (ret < 0) {
563 PERROR("UST consumerd64 cmd_sock close");
564 }
565 }
566 }
567
568 /*
569 * Generate the full lock file path using the rundir.
570 *
571 * Return the snprintf() return value thus a negative value is an error.
572 */
573 static int generate_lock_file_path(char *path, size_t len)
574 {
575 int ret;
576
577 assert(path);
578 assert(rundir);
579
580 /* Build lockfile path from rundir. */
581 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
582 if (ret < 0) {
583 PERROR("snprintf lockfile path");
584 }
585
586 return ret;
587 }
588
589 /*
590 * Cleanup the session daemon's data structures.
591 */
592 static void sessiond_cleanup(void)
593 {
594 int ret;
595 struct ltt_session *sess, *stmp;
596 char path[PATH_MAX];
597
598 DBG("Cleanup sessiond");
599
600 /*
601 * Close the thread quit pipe. It has already done its job,
602 * since we are now called.
603 */
604 utils_close_pipe(thread_quit_pipe);
605
606 /*
607 * If opt_pidfile is undefined, the default file will be wiped when
608 * removing the rundir.
609 */
610 if (opt_pidfile) {
611 ret = remove(opt_pidfile);
612 if (ret < 0) {
613 PERROR("remove pidfile %s", opt_pidfile);
614 }
615 }
616
617 DBG("Removing sessiond and consumerd content of directory %s", rundir);
618
619 /* sessiond */
620 snprintf(path, PATH_MAX,
621 "%s/%s",
622 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
623 DBG("Removing %s", path);
624 (void) unlink(path);
625
626 snprintf(path, PATH_MAX, "%s/%s", rundir,
627 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
628 DBG("Removing %s", path);
629 (void) unlink(path);
630
631 /* kconsumerd */
632 snprintf(path, PATH_MAX,
633 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
634 rundir);
635 DBG("Removing %s", path);
636 (void) unlink(path);
637
638 snprintf(path, PATH_MAX,
639 DEFAULT_KCONSUMERD_PATH,
640 rundir);
641 DBG("Removing directory %s", path);
642 (void) rmdir(path);
643
644 /* ust consumerd 32 */
645 snprintf(path, PATH_MAX,
646 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
647 rundir);
648 DBG("Removing %s", path);
649 (void) unlink(path);
650
651 snprintf(path, PATH_MAX,
652 DEFAULT_USTCONSUMERD32_PATH,
653 rundir);
654 DBG("Removing directory %s", path);
655 (void) rmdir(path);
656
657 /* ust consumerd 64 */
658 snprintf(path, PATH_MAX,
659 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
660 rundir);
661 DBG("Removing %s", path);
662 (void) unlink(path);
663
664 snprintf(path, PATH_MAX,
665 DEFAULT_USTCONSUMERD64_PATH,
666 rundir);
667 DBG("Removing directory %s", path);
668 (void) rmdir(path);
669
670 DBG("Cleaning up all sessions");
671
672 /* Destroy session list mutex */
673 if (session_list_ptr != NULL) {
674 pthread_mutex_destroy(&session_list_ptr->lock);
675
676 /* Cleanup ALL session */
677 cds_list_for_each_entry_safe(sess, stmp,
678 &session_list_ptr->head, list) {
679 cmd_destroy_session(sess, kernel_poll_pipe[1]);
680 }
681 }
682
683 DBG("Cleaning up all agent apps");
684 agent_app_ht_clean();
685
686 DBG("Closing all UST sockets");
687 ust_app_clean_list();
688 buffer_reg_destroy_registries();
689
690 if (is_root && !opt_no_kernel) {
691 DBG2("Closing kernel fd");
692 if (kernel_tracer_fd >= 0) {
693 ret = close(kernel_tracer_fd);
694 if (ret) {
695 PERROR("close");
696 }
697 }
698 DBG("Unloading kernel modules");
699 modprobe_remove_lttng_all();
700 free(syscall_table);
701 }
702
703 close_consumer_sockets();
704
705 if (load_info) {
706 load_session_destroy_data(load_info);
707 free(load_info);
708 }
709
710 /*
711 * Cleanup lock file by deleting it and finaly closing it which will
712 * release the file system lock.
713 */
714 if (lockfile_fd >= 0) {
715 char lockfile_path[PATH_MAX];
716
717 ret = generate_lock_file_path(lockfile_path,
718 sizeof(lockfile_path));
719 if (ret > 0) {
720 ret = remove(lockfile_path);
721 if (ret < 0) {
722 PERROR("remove lock file");
723 }
724 ret = close(lockfile_fd);
725 if (ret < 0) {
726 PERROR("close lock file");
727 }
728 }
729 }
730
731 /*
732 * We do NOT rmdir rundir because there are other processes
733 * using it, for instance lttng-relayd, which can start in
734 * parallel with this teardown.
735 */
736
737 free(rundir);
738 }
739
740 /*
741 * Cleanup the daemon's option data structures.
742 */
743 static void sessiond_cleanup_options(void)
744 {
745 DBG("Cleaning up options");
746
747 /*
748 * If the override option is set, the pointer points to a *non* const
749 * thus freeing it even though the variable type is set to const.
750 */
751 if (tracing_group_name_override) {
752 free((void *) tracing_group_name);
753 }
754 if (consumerd32_bin_override) {
755 free((void *) consumerd32_bin);
756 }
757 if (consumerd64_bin_override) {
758 free((void *) consumerd64_bin);
759 }
760 if (consumerd32_libdir_override) {
761 free((void *) consumerd32_libdir);
762 }
763 if (consumerd64_libdir_override) {
764 free((void *) consumerd64_libdir);
765 }
766
767 free(opt_pidfile);
768 free(opt_load_session_path);
769 free(kmod_probes_list);
770 free(kmod_extra_probes_list);
771
772 /* <fun> */
773 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
774 "Matthew, BEET driven development works!%c[%dm",
775 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
776 /* </fun> */
777 }
778
779 /*
780 * Send data on a unix socket using the liblttsessiondcomm API.
781 *
782 * Return lttcomm error code.
783 */
784 static int send_unix_sock(int sock, void *buf, size_t len)
785 {
786 /* Check valid length */
787 if (len == 0) {
788 return -1;
789 }
790
791 return lttcomm_send_unix_sock(sock, buf, len);
792 }
793
794 /*
795 * Free memory of a command context structure.
796 */
797 static void clean_command_ctx(struct command_ctx **cmd_ctx)
798 {
799 DBG("Clean command context structure");
800 if (*cmd_ctx) {
801 if ((*cmd_ctx)->llm) {
802 free((*cmd_ctx)->llm);
803 }
804 if ((*cmd_ctx)->lsm) {
805 free((*cmd_ctx)->lsm);
806 }
807 free(*cmd_ctx);
808 *cmd_ctx = NULL;
809 }
810 }
811
812 /*
813 * Notify UST applications using the shm mmap futex.
814 */
815 static int notify_ust_apps(int active)
816 {
817 char *wait_shm_mmap;
818
819 DBG("Notifying applications of session daemon state: %d", active);
820
821 /* See shm.c for this call implying mmap, shm and futex calls */
822 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
823 if (wait_shm_mmap == NULL) {
824 goto error;
825 }
826
827 /* Wake waiting process */
828 futex_wait_update((int32_t *) wait_shm_mmap, active);
829
830 /* Apps notified successfully */
831 return 0;
832
833 error:
834 return -1;
835 }
836
837 /*
838 * Setup the outgoing data buffer for the response (llm) by allocating the
839 * right amount of memory and copying the original information from the lsm
840 * structure.
841 *
842 * Return total size of the buffer pointed by buf.
843 */
844 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
845 {
846 int ret, buf_size;
847
848 buf_size = size;
849
850 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
851 if (cmd_ctx->llm == NULL) {
852 PERROR("zmalloc");
853 ret = -ENOMEM;
854 goto error;
855 }
856
857 /* Copy common data */
858 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
859 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
860
861 cmd_ctx->llm->data_size = size;
862 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
863
864 return buf_size;
865
866 error:
867 return ret;
868 }
869
870 /*
871 * Update the kernel poll set of all channel fd available over all tracing
872 * session. Add the wakeup pipe at the end of the set.
873 */
874 static int update_kernel_poll(struct lttng_poll_event *events)
875 {
876 int ret;
877 struct ltt_session *session;
878 struct ltt_kernel_channel *channel;
879
880 DBG("Updating kernel poll set");
881
882 session_lock_list();
883 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
884 session_lock(session);
885 if (session->kernel_session == NULL) {
886 session_unlock(session);
887 continue;
888 }
889
890 cds_list_for_each_entry(channel,
891 &session->kernel_session->channel_list.head, list) {
892 /* Add channel fd to the kernel poll set */
893 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
894 if (ret < 0) {
895 session_unlock(session);
896 goto error;
897 }
898 DBG("Channel fd %d added to kernel set", channel->fd);
899 }
900 session_unlock(session);
901 }
902 session_unlock_list();
903
904 return 0;
905
906 error:
907 session_unlock_list();
908 return -1;
909 }
910
911 /*
912 * Find the channel fd from 'fd' over all tracing session. When found, check
913 * for new channel stream and send those stream fds to the kernel consumer.
914 *
915 * Useful for CPU hotplug feature.
916 */
917 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
918 {
919 int ret = 0;
920 struct ltt_session *session;
921 struct ltt_kernel_session *ksess;
922 struct ltt_kernel_channel *channel;
923
924 DBG("Updating kernel streams for channel fd %d", fd);
925
926 session_lock_list();
927 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
928 session_lock(session);
929 if (session->kernel_session == NULL) {
930 session_unlock(session);
931 continue;
932 }
933 ksess = session->kernel_session;
934
935 cds_list_for_each_entry(channel,
936 &ksess->channel_list.head, list) {
937 struct lttng_ht_iter iter;
938 struct consumer_socket *socket;
939
940 if (channel->fd != fd) {
941 continue;
942 }
943 DBG("Channel found, updating kernel streams");
944 ret = kernel_open_channel_stream(channel);
945 if (ret < 0) {
946 goto error;
947 }
948 /* Update the stream global counter */
949 ksess->stream_count_global += ret;
950
951 /*
952 * Have we already sent fds to the consumer? If yes, it
953 * means that tracing is started so it is safe to send
954 * our updated stream fds.
955 */
956 if (ksess->consumer_fds_sent != 1
957 || ksess->consumer == NULL) {
958 ret = -1;
959 goto error;
960 }
961
962 rcu_read_lock();
963 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
964 &iter.iter, socket, node.node) {
965 pthread_mutex_lock(socket->lock);
966 ret = kernel_consumer_send_channel_stream(socket,
967 channel, ksess,
968 session->output_traces ? 1 : 0);
969 pthread_mutex_unlock(socket->lock);
970 if (ret < 0) {
971 rcu_read_unlock();
972 goto error;
973 }
974 }
975 rcu_read_unlock();
976 }
977 session_unlock(session);
978 }
979 session_unlock_list();
980 return ret;
981
982 error:
983 session_unlock(session);
984 session_unlock_list();
985 return ret;
986 }
987
988 /*
989 * For each tracing session, update newly registered apps. The session list
990 * lock MUST be acquired before calling this.
991 */
992 static void update_ust_app(int app_sock)
993 {
994 struct ltt_session *sess, *stmp;
995
996 /* Consumer is in an ERROR state. Stop any application update. */
997 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
998 /* Stop the update process since the consumer is dead. */
999 return;
1000 }
1001
1002 /* For all tracing session(s) */
1003 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1004 struct ust_app *app;
1005
1006 session_lock(sess);
1007 if (!sess->ust_session) {
1008 goto unlock_session;
1009 }
1010
1011 rcu_read_lock();
1012 assert(app_sock >= 0);
1013 app = ust_app_find_by_sock(app_sock);
1014 if (app == NULL) {
1015 /*
1016 * Application can be unregistered before so
1017 * this is possible hence simply stopping the
1018 * update.
1019 */
1020 DBG3("UST app update failed to find app sock %d",
1021 app_sock);
1022 goto unlock_rcu;
1023 }
1024 ust_app_global_update(sess->ust_session, app);
1025 unlock_rcu:
1026 rcu_read_unlock();
1027 unlock_session:
1028 session_unlock(sess);
1029 }
1030 }
1031
1032 /*
1033 * This thread manage event coming from the kernel.
1034 *
1035 * Features supported in this thread:
1036 * -) CPU Hotplug
1037 */
1038 static void *thread_manage_kernel(void *data)
1039 {
1040 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1041 uint32_t revents, nb_fd;
1042 char tmp;
1043 struct lttng_poll_event events;
1044
1045 DBG("[thread] Thread manage kernel started");
1046
1047 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1048
1049 /*
1050 * This first step of the while is to clean this structure which could free
1051 * non NULL pointers so initialize it before the loop.
1052 */
1053 lttng_poll_init(&events);
1054
1055 if (testpoint(sessiond_thread_manage_kernel)) {
1056 goto error_testpoint;
1057 }
1058
1059 health_code_update();
1060
1061 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1062 goto error_testpoint;
1063 }
1064
1065 while (1) {
1066 health_code_update();
1067
1068 if (update_poll_flag == 1) {
1069 /* Clean events object. We are about to populate it again. */
1070 lttng_poll_clean(&events);
1071
1072 ret = sessiond_set_thread_pollset(&events, 2);
1073 if (ret < 0) {
1074 goto error_poll_create;
1075 }
1076
1077 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1078 if (ret < 0) {
1079 goto error;
1080 }
1081
1082 /* This will add the available kernel channel if any. */
1083 ret = update_kernel_poll(&events);
1084 if (ret < 0) {
1085 goto error;
1086 }
1087 update_poll_flag = 0;
1088 }
1089
1090 DBG("Thread kernel polling");
1091
1092 /* Poll infinite value of time */
1093 restart:
1094 health_poll_entry();
1095 ret = lttng_poll_wait(&events, -1);
1096 DBG("Thread kernel return from poll on %d fds",
1097 LTTNG_POLL_GETNB(&events));
1098 health_poll_exit();
1099 if (ret < 0) {
1100 /*
1101 * Restart interrupted system call.
1102 */
1103 if (errno == EINTR) {
1104 goto restart;
1105 }
1106 goto error;
1107 } else if (ret == 0) {
1108 /* Should not happen since timeout is infinite */
1109 ERR("Return value of poll is 0 with an infinite timeout.\n"
1110 "This should not have happened! Continuing...");
1111 continue;
1112 }
1113
1114 nb_fd = ret;
1115
1116 for (i = 0; i < nb_fd; i++) {
1117 /* Fetch once the poll data */
1118 revents = LTTNG_POLL_GETEV(&events, i);
1119 pollfd = LTTNG_POLL_GETFD(&events, i);
1120
1121 health_code_update();
1122
1123 if (!revents) {
1124 /* No activity for this FD (poll implementation). */
1125 continue;
1126 }
1127
1128 /* Thread quit pipe has been closed. Killing thread. */
1129 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1130 if (ret) {
1131 err = 0;
1132 goto exit;
1133 }
1134
1135 /* Check for data on kernel pipe */
1136 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1137 (void) lttng_read(kernel_poll_pipe[0],
1138 &tmp, 1);
1139 /*
1140 * Ret value is useless here, if this pipe gets any actions an
1141 * update is required anyway.
1142 */
1143 update_poll_flag = 1;
1144 continue;
1145 } else {
1146 /*
1147 * New CPU detected by the kernel. Adding kernel stream to
1148 * kernel session and updating the kernel consumer
1149 */
1150 if (revents & LPOLLIN) {
1151 ret = update_kernel_stream(&kconsumer_data, pollfd);
1152 if (ret < 0) {
1153 continue;
1154 }
1155 break;
1156 /*
1157 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1158 * and unregister kernel stream at this point.
1159 */
1160 }
1161 }
1162 }
1163 }
1164
1165 exit:
1166 error:
1167 lttng_poll_clean(&events);
1168 error_poll_create:
1169 error_testpoint:
1170 utils_close_pipe(kernel_poll_pipe);
1171 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1172 if (err) {
1173 health_error();
1174 ERR("Health error occurred in %s", __func__);
1175 WARN("Kernel thread died unexpectedly. "
1176 "Kernel tracing can continue but CPU hotplug is disabled.");
1177 }
1178 health_unregister(health_sessiond);
1179 DBG("Kernel thread dying");
1180 return NULL;
1181 }
1182
1183 /*
1184 * Signal pthread condition of the consumer data that the thread.
1185 */
1186 static void signal_consumer_condition(struct consumer_data *data, int state)
1187 {
1188 pthread_mutex_lock(&data->cond_mutex);
1189
1190 /*
1191 * The state is set before signaling. It can be any value, it's the waiter
1192 * job to correctly interpret this condition variable associated to the
1193 * consumer pthread_cond.
1194 *
1195 * A value of 0 means that the corresponding thread of the consumer data
1196 * was not started. 1 indicates that the thread has started and is ready
1197 * for action. A negative value means that there was an error during the
1198 * thread bootstrap.
1199 */
1200 data->consumer_thread_is_ready = state;
1201 (void) pthread_cond_signal(&data->cond);
1202
1203 pthread_mutex_unlock(&data->cond_mutex);
1204 }
1205
1206 /*
1207 * This thread manage the consumer error sent back to the session daemon.
1208 */
1209 static void *thread_manage_consumer(void *data)
1210 {
1211 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1212 uint32_t revents, nb_fd;
1213 enum lttcomm_return_code code;
1214 struct lttng_poll_event events;
1215 struct consumer_data *consumer_data = data;
1216
1217 DBG("[thread] Manage consumer started");
1218
1219 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1220
1221 health_code_update();
1222
1223 /*
1224 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1225 * metadata_sock. Nothing more will be added to this poll set.
1226 */
1227 ret = sessiond_set_thread_pollset(&events, 3);
1228 if (ret < 0) {
1229 goto error_poll;
1230 }
1231
1232 /*
1233 * The error socket here is already in a listening state which was done
1234 * just before spawning this thread to avoid a race between the consumer
1235 * daemon exec trying to connect and the listen() call.
1236 */
1237 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1238 if (ret < 0) {
1239 goto error;
1240 }
1241
1242 health_code_update();
1243
1244 /* Infinite blocking call, waiting for transmission */
1245 restart:
1246 health_poll_entry();
1247
1248 if (testpoint(sessiond_thread_manage_consumer)) {
1249 goto error;
1250 }
1251
1252 ret = lttng_poll_wait(&events, -1);
1253 health_poll_exit();
1254 if (ret < 0) {
1255 /*
1256 * Restart interrupted system call.
1257 */
1258 if (errno == EINTR) {
1259 goto restart;
1260 }
1261 goto error;
1262 }
1263
1264 nb_fd = ret;
1265
1266 for (i = 0; i < nb_fd; i++) {
1267 /* Fetch once the poll data */
1268 revents = LTTNG_POLL_GETEV(&events, i);
1269 pollfd = LTTNG_POLL_GETFD(&events, i);
1270
1271 health_code_update();
1272
1273 if (!revents) {
1274 /* No activity for this FD (poll implementation). */
1275 continue;
1276 }
1277
1278 /* Thread quit pipe has been closed. Killing thread. */
1279 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1280 if (ret) {
1281 err = 0;
1282 goto exit;
1283 }
1284
1285 /* Event on the registration socket */
1286 if (pollfd == consumer_data->err_sock) {
1287 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1288 ERR("consumer err socket poll error");
1289 goto error;
1290 }
1291 }
1292 }
1293
1294 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1295 if (sock < 0) {
1296 goto error;
1297 }
1298
1299 /*
1300 * Set the CLOEXEC flag. Return code is useless because either way, the
1301 * show must go on.
1302 */
1303 (void) utils_set_fd_cloexec(sock);
1304
1305 health_code_update();
1306
1307 DBG2("Receiving code from consumer err_sock");
1308
1309 /* Getting status code from kconsumerd */
1310 ret = lttcomm_recv_unix_sock(sock, &code,
1311 sizeof(enum lttcomm_return_code));
1312 if (ret <= 0) {
1313 goto error;
1314 }
1315
1316 health_code_update();
1317 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1318 /* Connect both socket, command and metadata. */
1319 consumer_data->cmd_sock =
1320 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1321 consumer_data->metadata_fd =
1322 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1323 if (consumer_data->cmd_sock < 0
1324 || consumer_data->metadata_fd < 0) {
1325 PERROR("consumer connect cmd socket");
1326 /* On error, signal condition and quit. */
1327 signal_consumer_condition(consumer_data, -1);
1328 goto error;
1329 }
1330 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1331 /* Create metadata socket lock. */
1332 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1333 if (consumer_data->metadata_sock.lock == NULL) {
1334 PERROR("zmalloc pthread mutex");
1335 ret = -1;
1336 goto error;
1337 }
1338 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1339
1340 signal_consumer_condition(consumer_data, 1);
1341 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1342 DBG("Consumer metadata socket ready (fd: %d)",
1343 consumer_data->metadata_fd);
1344 } else {
1345 ERR("consumer error when waiting for SOCK_READY : %s",
1346 lttcomm_get_readable_code(-code));
1347 goto error;
1348 }
1349
1350 /* Remove the consumerd error sock since we've established a connexion */
1351 ret = lttng_poll_del(&events, consumer_data->err_sock);
1352 if (ret < 0) {
1353 goto error;
1354 }
1355
1356 /* Add new accepted error socket. */
1357 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1358 if (ret < 0) {
1359 goto error;
1360 }
1361
1362 /* Add metadata socket that is successfully connected. */
1363 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1364 LPOLLIN | LPOLLRDHUP);
1365 if (ret < 0) {
1366 goto error;
1367 }
1368
1369 health_code_update();
1370
1371 /* Infinite blocking call, waiting for transmission */
1372 restart_poll:
1373 while (1) {
1374 health_code_update();
1375
1376 /* Exit the thread because the thread quit pipe has been triggered. */
1377 if (should_quit) {
1378 /* Not a health error. */
1379 err = 0;
1380 goto exit;
1381 }
1382
1383 health_poll_entry();
1384 ret = lttng_poll_wait(&events, -1);
1385 health_poll_exit();
1386 if (ret < 0) {
1387 /*
1388 * Restart interrupted system call.
1389 */
1390 if (errno == EINTR) {
1391 goto restart_poll;
1392 }
1393 goto error;
1394 }
1395
1396 nb_fd = ret;
1397
1398 for (i = 0; i < nb_fd; i++) {
1399 /* Fetch once the poll data */
1400 revents = LTTNG_POLL_GETEV(&events, i);
1401 pollfd = LTTNG_POLL_GETFD(&events, i);
1402
1403 health_code_update();
1404
1405 if (!revents) {
1406 /* No activity for this FD (poll implementation). */
1407 continue;
1408 }
1409
1410 /*
1411 * Thread quit pipe has been triggered, flag that we should stop
1412 * but continue the current loop to handle potential data from
1413 * consumer.
1414 */
1415 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1416
1417 if (pollfd == sock) {
1418 /* Event on the consumerd socket */
1419 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1420 ERR("consumer err socket second poll error");
1421 goto error;
1422 }
1423 health_code_update();
1424 /* Wait for any kconsumerd error */
1425 ret = lttcomm_recv_unix_sock(sock, &code,
1426 sizeof(enum lttcomm_return_code));
1427 if (ret <= 0) {
1428 ERR("consumer closed the command socket");
1429 goto error;
1430 }
1431
1432 ERR("consumer return code : %s",
1433 lttcomm_get_readable_code(-code));
1434
1435 goto exit;
1436 } else if (pollfd == consumer_data->metadata_fd) {
1437 /* UST metadata requests */
1438 ret = ust_consumer_metadata_request(
1439 &consumer_data->metadata_sock);
1440 if (ret < 0) {
1441 ERR("Handling metadata request");
1442 goto error;
1443 }
1444 }
1445 /* No need for an else branch all FDs are tested prior. */
1446 }
1447 health_code_update();
1448 }
1449
1450 exit:
1451 error:
1452 /*
1453 * We lock here because we are about to close the sockets and some other
1454 * thread might be using them so get exclusive access which will abort all
1455 * other consumer command by other threads.
1456 */
1457 pthread_mutex_lock(&consumer_data->lock);
1458
1459 /* Immediately set the consumerd state to stopped */
1460 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1461 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1462 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1463 consumer_data->type == LTTNG_CONSUMER32_UST) {
1464 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1465 } else {
1466 /* Code flow error... */
1467 assert(0);
1468 }
1469
1470 if (consumer_data->err_sock >= 0) {
1471 ret = close(consumer_data->err_sock);
1472 if (ret) {
1473 PERROR("close");
1474 }
1475 consumer_data->err_sock = -1;
1476 }
1477 if (consumer_data->cmd_sock >= 0) {
1478 ret = close(consumer_data->cmd_sock);
1479 if (ret) {
1480 PERROR("close");
1481 }
1482 consumer_data->cmd_sock = -1;
1483 }
1484 if (consumer_data->metadata_sock.fd_ptr &&
1485 *consumer_data->metadata_sock.fd_ptr >= 0) {
1486 ret = close(*consumer_data->metadata_sock.fd_ptr);
1487 if (ret) {
1488 PERROR("close");
1489 }
1490 }
1491 if (sock >= 0) {
1492 ret = close(sock);
1493 if (ret) {
1494 PERROR("close");
1495 }
1496 }
1497
1498 unlink(consumer_data->err_unix_sock_path);
1499 unlink(consumer_data->cmd_unix_sock_path);
1500 consumer_data->pid = 0;
1501 pthread_mutex_unlock(&consumer_data->lock);
1502
1503 /* Cleanup metadata socket mutex. */
1504 if (consumer_data->metadata_sock.lock) {
1505 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1506 free(consumer_data->metadata_sock.lock);
1507 }
1508 lttng_poll_clean(&events);
1509 error_poll:
1510 if (err) {
1511 health_error();
1512 ERR("Health error occurred in %s", __func__);
1513 }
1514 health_unregister(health_sessiond);
1515 DBG("consumer thread cleanup completed");
1516
1517 return NULL;
1518 }
1519
1520 /*
1521 * This thread manage application communication.
1522 */
1523 static void *thread_manage_apps(void *data)
1524 {
1525 int i, ret, pollfd, err = -1;
1526 ssize_t size_ret;
1527 uint32_t revents, nb_fd;
1528 struct lttng_poll_event events;
1529
1530 DBG("[thread] Manage application started");
1531
1532 rcu_register_thread();
1533 rcu_thread_online();
1534
1535 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1536
1537 if (testpoint(sessiond_thread_manage_apps)) {
1538 goto error_testpoint;
1539 }
1540
1541 health_code_update();
1542
1543 ret = sessiond_set_thread_pollset(&events, 2);
1544 if (ret < 0) {
1545 goto error_poll_create;
1546 }
1547
1548 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1549 if (ret < 0) {
1550 goto error;
1551 }
1552
1553 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1554 goto error;
1555 }
1556
1557 health_code_update();
1558
1559 while (1) {
1560 DBG("Apps thread polling");
1561
1562 /* Inifinite blocking call, waiting for transmission */
1563 restart:
1564 health_poll_entry();
1565 ret = lttng_poll_wait(&events, -1);
1566 DBG("Apps thread return from poll on %d fds",
1567 LTTNG_POLL_GETNB(&events));
1568 health_poll_exit();
1569 if (ret < 0) {
1570 /*
1571 * Restart interrupted system call.
1572 */
1573 if (errno == EINTR) {
1574 goto restart;
1575 }
1576 goto error;
1577 }
1578
1579 nb_fd = ret;
1580
1581 for (i = 0; i < nb_fd; i++) {
1582 /* Fetch once the poll data */
1583 revents = LTTNG_POLL_GETEV(&events, i);
1584 pollfd = LTTNG_POLL_GETFD(&events, i);
1585
1586 health_code_update();
1587
1588 if (!revents) {
1589 /* No activity for this FD (poll implementation). */
1590 continue;
1591 }
1592
1593 /* Thread quit pipe has been closed. Killing thread. */
1594 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1595 if (ret) {
1596 err = 0;
1597 goto exit;
1598 }
1599
1600 /* Inspect the apps cmd pipe */
1601 if (pollfd == apps_cmd_pipe[0]) {
1602 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1603 ERR("Apps command pipe error");
1604 goto error;
1605 } else if (revents & LPOLLIN) {
1606 int sock;
1607
1608 /* Empty pipe */
1609 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1610 if (size_ret < sizeof(sock)) {
1611 PERROR("read apps cmd pipe");
1612 goto error;
1613 }
1614
1615 health_code_update();
1616
1617 /*
1618 * We only monitor the error events of the socket. This
1619 * thread does not handle any incoming data from UST
1620 * (POLLIN).
1621 */
1622 ret = lttng_poll_add(&events, sock,
1623 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1624 if (ret < 0) {
1625 goto error;
1626 }
1627
1628 DBG("Apps with sock %d added to poll set", sock);
1629 }
1630 } else {
1631 /*
1632 * At this point, we know that a registered application made
1633 * the event at poll_wait.
1634 */
1635 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1636 /* Removing from the poll set */
1637 ret = lttng_poll_del(&events, pollfd);
1638 if (ret < 0) {
1639 goto error;
1640 }
1641
1642 /* Socket closed on remote end. */
1643 ust_app_unregister(pollfd);
1644 }
1645 }
1646
1647 health_code_update();
1648 }
1649 }
1650
1651 exit:
1652 error:
1653 lttng_poll_clean(&events);
1654 error_poll_create:
1655 error_testpoint:
1656 utils_close_pipe(apps_cmd_pipe);
1657 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1658
1659 /*
1660 * We don't clean the UST app hash table here since already registered
1661 * applications can still be controlled so let them be until the session
1662 * daemon dies or the applications stop.
1663 */
1664
1665 if (err) {
1666 health_error();
1667 ERR("Health error occurred in %s", __func__);
1668 }
1669 health_unregister(health_sessiond);
1670 DBG("Application communication apps thread cleanup complete");
1671 rcu_thread_offline();
1672 rcu_unregister_thread();
1673 return NULL;
1674 }
1675
1676 /*
1677 * Send a socket to a thread This is called from the dispatch UST registration
1678 * thread once all sockets are set for the application.
1679 *
1680 * The sock value can be invalid, we don't really care, the thread will handle
1681 * it and make the necessary cleanup if so.
1682 *
1683 * On success, return 0 else a negative value being the errno message of the
1684 * write().
1685 */
1686 static int send_socket_to_thread(int fd, int sock)
1687 {
1688 ssize_t ret;
1689
1690 /*
1691 * It's possible that the FD is set as invalid with -1 concurrently just
1692 * before calling this function being a shutdown state of the thread.
1693 */
1694 if (fd < 0) {
1695 ret = -EBADF;
1696 goto error;
1697 }
1698
1699 ret = lttng_write(fd, &sock, sizeof(sock));
1700 if (ret < sizeof(sock)) {
1701 PERROR("write apps pipe %d", fd);
1702 if (ret < 0) {
1703 ret = -errno;
1704 }
1705 goto error;
1706 }
1707
1708 /* All good. Don't send back the write positive ret value. */
1709 ret = 0;
1710 error:
1711 return (int) ret;
1712 }
1713
1714 /*
1715 * Sanitize the wait queue of the dispatch registration thread meaning removing
1716 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1717 * notify socket is never received.
1718 */
1719 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1720 {
1721 int ret, nb_fd = 0, i;
1722 unsigned int fd_added = 0;
1723 struct lttng_poll_event events;
1724 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1725
1726 assert(wait_queue);
1727
1728 lttng_poll_init(&events);
1729
1730 /* Just skip everything for an empty queue. */
1731 if (!wait_queue->count) {
1732 goto end;
1733 }
1734
1735 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1736 if (ret < 0) {
1737 goto error_create;
1738 }
1739
1740 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1741 &wait_queue->head, head) {
1742 assert(wait_node->app);
1743 ret = lttng_poll_add(&events, wait_node->app->sock,
1744 LPOLLHUP | LPOLLERR);
1745 if (ret < 0) {
1746 goto error;
1747 }
1748
1749 fd_added = 1;
1750 }
1751
1752 if (!fd_added) {
1753 goto end;
1754 }
1755
1756 /*
1757 * Poll but don't block so we can quickly identify the faulty events and
1758 * clean them afterwards from the wait queue.
1759 */
1760 ret = lttng_poll_wait(&events, 0);
1761 if (ret < 0) {
1762 goto error;
1763 }
1764 nb_fd = ret;
1765
1766 for (i = 0; i < nb_fd; i++) {
1767 /* Get faulty FD. */
1768 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1769 int pollfd = LTTNG_POLL_GETFD(&events, i);
1770
1771 if (!revents) {
1772 /* No activity for this FD (poll implementation). */
1773 continue;
1774 }
1775
1776 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1777 &wait_queue->head, head) {
1778 if (pollfd == wait_node->app->sock &&
1779 (revents & (LPOLLHUP | LPOLLERR))) {
1780 cds_list_del(&wait_node->head);
1781 wait_queue->count--;
1782 ust_app_destroy(wait_node->app);
1783 free(wait_node);
1784 break;
1785 }
1786 }
1787 }
1788
1789 if (nb_fd > 0) {
1790 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1791 }
1792
1793 end:
1794 lttng_poll_clean(&events);
1795 return;
1796
1797 error:
1798 lttng_poll_clean(&events);
1799 error_create:
1800 ERR("Unable to sanitize wait queue");
1801 return;
1802 }
1803
1804 /*
1805 * Dispatch request from the registration threads to the application
1806 * communication thread.
1807 */
1808 static void *thread_dispatch_ust_registration(void *data)
1809 {
1810 int ret, err = -1;
1811 struct cds_wfcq_node *node;
1812 struct ust_command *ust_cmd = NULL;
1813 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1814 struct ust_reg_wait_queue wait_queue = {
1815 .count = 0,
1816 };
1817
1818 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1819
1820 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1821 goto error_testpoint;
1822 }
1823
1824 health_code_update();
1825
1826 CDS_INIT_LIST_HEAD(&wait_queue.head);
1827
1828 DBG("[thread] Dispatch UST command started");
1829
1830 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1831 health_code_update();
1832
1833 /* Atomically prepare the queue futex */
1834 futex_nto1_prepare(&ust_cmd_queue.futex);
1835
1836 do {
1837 struct ust_app *app = NULL;
1838 ust_cmd = NULL;
1839
1840 /*
1841 * Make sure we don't have node(s) that have hung up before receiving
1842 * the notify socket. This is to clean the list in order to avoid
1843 * memory leaks from notify socket that are never seen.
1844 */
1845 sanitize_wait_queue(&wait_queue);
1846
1847 health_code_update();
1848 /* Dequeue command for registration */
1849 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1850 if (node == NULL) {
1851 DBG("Woken up but nothing in the UST command queue");
1852 /* Continue thread execution */
1853 break;
1854 }
1855
1856 ust_cmd = caa_container_of(node, struct ust_command, node);
1857
1858 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1859 " gid:%d sock:%d name:%s (version %d.%d)",
1860 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1861 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1862 ust_cmd->sock, ust_cmd->reg_msg.name,
1863 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1864
1865 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1866 wait_node = zmalloc(sizeof(*wait_node));
1867 if (!wait_node) {
1868 PERROR("zmalloc wait_node dispatch");
1869 ret = close(ust_cmd->sock);
1870 if (ret < 0) {
1871 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1872 }
1873 lttng_fd_put(LTTNG_FD_APPS, 1);
1874 free(ust_cmd);
1875 goto error;
1876 }
1877 CDS_INIT_LIST_HEAD(&wait_node->head);
1878
1879 /* Create application object if socket is CMD. */
1880 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1881 ust_cmd->sock);
1882 if (!wait_node->app) {
1883 ret = close(ust_cmd->sock);
1884 if (ret < 0) {
1885 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1886 }
1887 lttng_fd_put(LTTNG_FD_APPS, 1);
1888 free(wait_node);
1889 free(ust_cmd);
1890 continue;
1891 }
1892 /*
1893 * Add application to the wait queue so we can set the notify
1894 * socket before putting this object in the global ht.
1895 */
1896 cds_list_add(&wait_node->head, &wait_queue.head);
1897 wait_queue.count++;
1898
1899 free(ust_cmd);
1900 /*
1901 * We have to continue here since we don't have the notify
1902 * socket and the application MUST be added to the hash table
1903 * only at that moment.
1904 */
1905 continue;
1906 } else {
1907 /*
1908 * Look for the application in the local wait queue and set the
1909 * notify socket if found.
1910 */
1911 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1912 &wait_queue.head, head) {
1913 health_code_update();
1914 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1915 wait_node->app->notify_sock = ust_cmd->sock;
1916 cds_list_del(&wait_node->head);
1917 wait_queue.count--;
1918 app = wait_node->app;
1919 free(wait_node);
1920 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1921 break;
1922 }
1923 }
1924
1925 /*
1926 * With no application at this stage the received socket is
1927 * basically useless so close it before we free the cmd data
1928 * structure for good.
1929 */
1930 if (!app) {
1931 ret = close(ust_cmd->sock);
1932 if (ret < 0) {
1933 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1934 }
1935 lttng_fd_put(LTTNG_FD_APPS, 1);
1936 }
1937 free(ust_cmd);
1938 }
1939
1940 if (app) {
1941 /*
1942 * @session_lock_list
1943 *
1944 * Lock the global session list so from the register up to the
1945 * registration done message, no thread can see the application
1946 * and change its state.
1947 */
1948 session_lock_list();
1949 rcu_read_lock();
1950
1951 /*
1952 * Add application to the global hash table. This needs to be
1953 * done before the update to the UST registry can locate the
1954 * application.
1955 */
1956 ust_app_add(app);
1957
1958 /* Set app version. This call will print an error if needed. */
1959 (void) ust_app_version(app);
1960
1961 /* Send notify socket through the notify pipe. */
1962 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1963 app->notify_sock);
1964 if (ret < 0) {
1965 rcu_read_unlock();
1966 session_unlock_list();
1967 /*
1968 * No notify thread, stop the UST tracing. However, this is
1969 * not an internal error of the this thread thus setting
1970 * the health error code to a normal exit.
1971 */
1972 err = 0;
1973 goto error;
1974 }
1975
1976 /*
1977 * Update newly registered application with the tracing
1978 * registry info already enabled information.
1979 */
1980 update_ust_app(app->sock);
1981
1982 /*
1983 * Don't care about return value. Let the manage apps threads
1984 * handle app unregistration upon socket close.
1985 */
1986 (void) ust_app_register_done(app->sock);
1987
1988 /*
1989 * Even if the application socket has been closed, send the app
1990 * to the thread and unregistration will take place at that
1991 * place.
1992 */
1993 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1994 if (ret < 0) {
1995 rcu_read_unlock();
1996 session_unlock_list();
1997 /*
1998 * No apps. thread, stop the UST tracing. However, this is
1999 * not an internal error of the this thread thus setting
2000 * the health error code to a normal exit.
2001 */
2002 err = 0;
2003 goto error;
2004 }
2005
2006 rcu_read_unlock();
2007 session_unlock_list();
2008 }
2009 } while (node != NULL);
2010
2011 health_poll_entry();
2012 /* Futex wait on queue. Blocking call on futex() */
2013 futex_nto1_wait(&ust_cmd_queue.futex);
2014 health_poll_exit();
2015 }
2016 /* Normal exit, no error */
2017 err = 0;
2018
2019 error:
2020 /* Clean up wait queue. */
2021 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2022 &wait_queue.head, head) {
2023 cds_list_del(&wait_node->head);
2024 wait_queue.count--;
2025 free(wait_node);
2026 }
2027
2028 error_testpoint:
2029 DBG("Dispatch thread dying");
2030 if (err) {
2031 health_error();
2032 ERR("Health error occurred in %s", __func__);
2033 }
2034 health_unregister(health_sessiond);
2035 return NULL;
2036 }
2037
2038 /*
2039 * This thread manage application registration.
2040 */
2041 static void *thread_registration_apps(void *data)
2042 {
2043 int sock = -1, i, ret, pollfd, err = -1;
2044 uint32_t revents, nb_fd;
2045 struct lttng_poll_event events;
2046 /*
2047 * Get allocated in this thread, enqueued to a global queue, dequeued and
2048 * freed in the manage apps thread.
2049 */
2050 struct ust_command *ust_cmd = NULL;
2051
2052 DBG("[thread] Manage application registration started");
2053
2054 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2055
2056 if (testpoint(sessiond_thread_registration_apps)) {
2057 goto error_testpoint;
2058 }
2059
2060 ret = lttcomm_listen_unix_sock(apps_sock);
2061 if (ret < 0) {
2062 goto error_listen;
2063 }
2064
2065 /*
2066 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2067 * more will be added to this poll set.
2068 */
2069 ret = sessiond_set_thread_pollset(&events, 2);
2070 if (ret < 0) {
2071 goto error_create_poll;
2072 }
2073
2074 /* Add the application registration socket */
2075 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2076 if (ret < 0) {
2077 goto error_poll_add;
2078 }
2079
2080 /* Notify all applications to register */
2081 ret = notify_ust_apps(1);
2082 if (ret < 0) {
2083 ERR("Failed to notify applications or create the wait shared memory.\n"
2084 "Execution continues but there might be problem for already\n"
2085 "running applications that wishes to register.");
2086 }
2087
2088 while (1) {
2089 DBG("Accepting application registration");
2090
2091 /* Inifinite blocking call, waiting for transmission */
2092 restart:
2093 health_poll_entry();
2094 ret = lttng_poll_wait(&events, -1);
2095 health_poll_exit();
2096 if (ret < 0) {
2097 /*
2098 * Restart interrupted system call.
2099 */
2100 if (errno == EINTR) {
2101 goto restart;
2102 }
2103 goto error;
2104 }
2105
2106 nb_fd = ret;
2107
2108 for (i = 0; i < nb_fd; i++) {
2109 health_code_update();
2110
2111 /* Fetch once the poll data */
2112 revents = LTTNG_POLL_GETEV(&events, i);
2113 pollfd = LTTNG_POLL_GETFD(&events, i);
2114
2115 if (!revents) {
2116 /* No activity for this FD (poll implementation). */
2117 continue;
2118 }
2119
2120 /* Thread quit pipe has been closed. Killing thread. */
2121 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2122 if (ret) {
2123 err = 0;
2124 goto exit;
2125 }
2126
2127 /* Event on the registration socket */
2128 if (pollfd == apps_sock) {
2129 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2130 ERR("Register apps socket poll error");
2131 goto error;
2132 } else if (revents & LPOLLIN) {
2133 sock = lttcomm_accept_unix_sock(apps_sock);
2134 if (sock < 0) {
2135 goto error;
2136 }
2137
2138 /*
2139 * Set socket timeout for both receiving and ending.
2140 * app_socket_timeout is in seconds, whereas
2141 * lttcomm_setsockopt_rcv_timeout and
2142 * lttcomm_setsockopt_snd_timeout expect msec as
2143 * parameter.
2144 */
2145 (void) lttcomm_setsockopt_rcv_timeout(sock,
2146 app_socket_timeout * 1000);
2147 (void) lttcomm_setsockopt_snd_timeout(sock,
2148 app_socket_timeout * 1000);
2149
2150 /*
2151 * Set the CLOEXEC flag. Return code is useless because
2152 * either way, the show must go on.
2153 */
2154 (void) utils_set_fd_cloexec(sock);
2155
2156 /* Create UST registration command for enqueuing */
2157 ust_cmd = zmalloc(sizeof(struct ust_command));
2158 if (ust_cmd == NULL) {
2159 PERROR("ust command zmalloc");
2160 ret = close(sock);
2161 if (ret) {
2162 PERROR("close");
2163 }
2164 goto error;
2165 }
2166
2167 /*
2168 * Using message-based transmissions to ensure we don't
2169 * have to deal with partially received messages.
2170 */
2171 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2172 if (ret < 0) {
2173 ERR("Exhausted file descriptors allowed for applications.");
2174 free(ust_cmd);
2175 ret = close(sock);
2176 if (ret) {
2177 PERROR("close");
2178 }
2179 sock = -1;
2180 continue;
2181 }
2182
2183 health_code_update();
2184 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2185 if (ret < 0) {
2186 free(ust_cmd);
2187 /* Close socket of the application. */
2188 ret = close(sock);
2189 if (ret) {
2190 PERROR("close");
2191 }
2192 lttng_fd_put(LTTNG_FD_APPS, 1);
2193 sock = -1;
2194 continue;
2195 }
2196 health_code_update();
2197
2198 ust_cmd->sock = sock;
2199 sock = -1;
2200
2201 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2202 " gid:%d sock:%d name:%s (version %d.%d)",
2203 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2204 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2205 ust_cmd->sock, ust_cmd->reg_msg.name,
2206 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2207
2208 /*
2209 * Lock free enqueue the registration request. The red pill
2210 * has been taken! This apps will be part of the *system*.
2211 */
2212 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2213
2214 /*
2215 * Wake the registration queue futex. Implicit memory
2216 * barrier with the exchange in cds_wfcq_enqueue.
2217 */
2218 futex_nto1_wake(&ust_cmd_queue.futex);
2219 }
2220 }
2221 }
2222 }
2223
2224 exit:
2225 error:
2226 /* Notify that the registration thread is gone */
2227 notify_ust_apps(0);
2228
2229 if (apps_sock >= 0) {
2230 ret = close(apps_sock);
2231 if (ret) {
2232 PERROR("close");
2233 }
2234 }
2235 if (sock >= 0) {
2236 ret = close(sock);
2237 if (ret) {
2238 PERROR("close");
2239 }
2240 lttng_fd_put(LTTNG_FD_APPS, 1);
2241 }
2242 unlink(apps_unix_sock_path);
2243
2244 error_poll_add:
2245 lttng_poll_clean(&events);
2246 error_listen:
2247 error_create_poll:
2248 error_testpoint:
2249 DBG("UST Registration thread cleanup complete");
2250 if (err) {
2251 health_error();
2252 ERR("Health error occurred in %s", __func__);
2253 }
2254 health_unregister(health_sessiond);
2255
2256 return NULL;
2257 }
2258
2259 /*
2260 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2261 * exec or it will fails.
2262 */
2263 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2264 {
2265 int ret, clock_ret;
2266 struct timespec timeout;
2267
2268 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2269 consumer_data->consumer_thread_is_ready = 0;
2270
2271 /* Setup pthread condition */
2272 ret = pthread_condattr_init(&consumer_data->condattr);
2273 if (ret) {
2274 errno = ret;
2275 PERROR("pthread_condattr_init consumer data");
2276 goto error;
2277 }
2278
2279 /*
2280 * Set the monotonic clock in order to make sure we DO NOT jump in time
2281 * between the clock_gettime() call and the timedwait call. See bug #324
2282 * for a more details and how we noticed it.
2283 */
2284 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2285 if (ret) {
2286 errno = ret;
2287 PERROR("pthread_condattr_setclock consumer data");
2288 goto error;
2289 }
2290
2291 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2292 if (ret) {
2293 errno = ret;
2294 PERROR("pthread_cond_init consumer data");
2295 goto error;
2296 }
2297
2298 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2299 consumer_data);
2300 if (ret) {
2301 errno = ret;
2302 PERROR("pthread_create consumer");
2303 ret = -1;
2304 goto error;
2305 }
2306
2307 /* We are about to wait on a pthread condition */
2308 pthread_mutex_lock(&consumer_data->cond_mutex);
2309
2310 /* Get time for sem_timedwait absolute timeout */
2311 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2312 /*
2313 * Set the timeout for the condition timed wait even if the clock gettime
2314 * call fails since we might loop on that call and we want to avoid to
2315 * increment the timeout too many times.
2316 */
2317 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2318
2319 /*
2320 * The following loop COULD be skipped in some conditions so this is why we
2321 * set ret to 0 in order to make sure at least one round of the loop is
2322 * done.
2323 */
2324 ret = 0;
2325
2326 /*
2327 * Loop until the condition is reached or when a timeout is reached. Note
2328 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2329 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2330 * possible. This loop does not take any chances and works with both of
2331 * them.
2332 */
2333 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2334 if (clock_ret < 0) {
2335 PERROR("clock_gettime spawn consumer");
2336 /* Infinite wait for the consumerd thread to be ready */
2337 ret = pthread_cond_wait(&consumer_data->cond,
2338 &consumer_data->cond_mutex);
2339 } else {
2340 ret = pthread_cond_timedwait(&consumer_data->cond,
2341 &consumer_data->cond_mutex, &timeout);
2342 }
2343 }
2344
2345 /* Release the pthread condition */
2346 pthread_mutex_unlock(&consumer_data->cond_mutex);
2347
2348 if (ret != 0) {
2349 errno = ret;
2350 if (ret == ETIMEDOUT) {
2351 int pth_ret;
2352
2353 /*
2354 * Call has timed out so we kill the kconsumerd_thread and return
2355 * an error.
2356 */
2357 ERR("Condition timed out. The consumer thread was never ready."
2358 " Killing it");
2359 pth_ret = pthread_cancel(consumer_data->thread);
2360 if (pth_ret < 0) {
2361 PERROR("pthread_cancel consumer thread");
2362 }
2363 } else {
2364 PERROR("pthread_cond_wait failed consumer thread");
2365 }
2366 /* Caller is expecting a negative value on failure. */
2367 ret = -1;
2368 goto error;
2369 }
2370
2371 pthread_mutex_lock(&consumer_data->pid_mutex);
2372 if (consumer_data->pid == 0) {
2373 ERR("Consumerd did not start");
2374 pthread_mutex_unlock(&consumer_data->pid_mutex);
2375 goto error;
2376 }
2377 pthread_mutex_unlock(&consumer_data->pid_mutex);
2378
2379 return 0;
2380
2381 error:
2382 return ret;
2383 }
2384
2385 /*
2386 * Join consumer thread
2387 */
2388 static int join_consumer_thread(struct consumer_data *consumer_data)
2389 {
2390 void *status;
2391
2392 /* Consumer pid must be a real one. */
2393 if (consumer_data->pid > 0) {
2394 int ret;
2395 ret = kill(consumer_data->pid, SIGTERM);
2396 if (ret) {
2397 PERROR("Error killing consumer daemon");
2398 return ret;
2399 }
2400 return pthread_join(consumer_data->thread, &status);
2401 } else {
2402 return 0;
2403 }
2404 }
2405
2406 /*
2407 * Fork and exec a consumer daemon (consumerd).
2408 *
2409 * Return pid if successful else -1.
2410 */
2411 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2412 {
2413 int ret;
2414 pid_t pid;
2415 const char *consumer_to_use;
2416 const char *verbosity;
2417 struct stat st;
2418
2419 DBG("Spawning consumerd");
2420
2421 pid = fork();
2422 if (pid == 0) {
2423 /*
2424 * Exec consumerd.
2425 */
2426 if (opt_verbose_consumer) {
2427 verbosity = "--verbose";
2428 } else if (lttng_opt_quiet) {
2429 verbosity = "--quiet";
2430 } else {
2431 verbosity = "";
2432 }
2433
2434 switch (consumer_data->type) {
2435 case LTTNG_CONSUMER_KERNEL:
2436 /*
2437 * Find out which consumerd to execute. We will first try the
2438 * 64-bit path, then the sessiond's installation directory, and
2439 * fallback on the 32-bit one,
2440 */
2441 DBG3("Looking for a kernel consumer at these locations:");
2442 DBG3(" 1) %s", consumerd64_bin);
2443 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2444 DBG3(" 3) %s", consumerd32_bin);
2445 if (stat(consumerd64_bin, &st) == 0) {
2446 DBG3("Found location #1");
2447 consumer_to_use = consumerd64_bin;
2448 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2449 DBG3("Found location #2");
2450 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2451 } else if (stat(consumerd32_bin, &st) == 0) {
2452 DBG3("Found location #3");
2453 consumer_to_use = consumerd32_bin;
2454 } else {
2455 DBG("Could not find any valid consumerd executable");
2456 ret = -EINVAL;
2457 break;
2458 }
2459 DBG("Using kernel consumer at: %s", consumer_to_use);
2460 ret = execl(consumer_to_use,
2461 "lttng-consumerd", verbosity, "-k",
2462 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2463 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2464 "--group", tracing_group_name,
2465 NULL);
2466 break;
2467 case LTTNG_CONSUMER64_UST:
2468 {
2469 char *tmpnew = NULL;
2470
2471 if (consumerd64_libdir[0] != '\0') {
2472 char *tmp;
2473 size_t tmplen;
2474
2475 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2476 if (!tmp) {
2477 tmp = "";
2478 }
2479 tmplen = strlen("LD_LIBRARY_PATH=")
2480 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2481 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2482 if (!tmpnew) {
2483 ret = -ENOMEM;
2484 goto error;
2485 }
2486 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2487 strcat(tmpnew, consumerd64_libdir);
2488 if (tmp[0] != '\0') {
2489 strcat(tmpnew, ":");
2490 strcat(tmpnew, tmp);
2491 }
2492 ret = putenv(tmpnew);
2493 if (ret) {
2494 ret = -errno;
2495 free(tmpnew);
2496 goto error;
2497 }
2498 }
2499 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2500 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2501 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2502 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2503 "--group", tracing_group_name,
2504 NULL);
2505 if (consumerd64_libdir[0] != '\0') {
2506 free(tmpnew);
2507 }
2508 break;
2509 }
2510 case LTTNG_CONSUMER32_UST:
2511 {
2512 char *tmpnew = NULL;
2513
2514 if (consumerd32_libdir[0] != '\0') {
2515 char *tmp;
2516 size_t tmplen;
2517
2518 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2519 if (!tmp) {
2520 tmp = "";
2521 }
2522 tmplen = strlen("LD_LIBRARY_PATH=")
2523 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2524 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2525 if (!tmpnew) {
2526 ret = -ENOMEM;
2527 goto error;
2528 }
2529 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2530 strcat(tmpnew, consumerd32_libdir);
2531 if (tmp[0] != '\0') {
2532 strcat(tmpnew, ":");
2533 strcat(tmpnew, tmp);
2534 }
2535 ret = putenv(tmpnew);
2536 if (ret) {
2537 ret = -errno;
2538 free(tmpnew);
2539 goto error;
2540 }
2541 }
2542 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2543 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2544 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2545 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2546 "--group", tracing_group_name,
2547 NULL);
2548 if (consumerd32_libdir[0] != '\0') {
2549 free(tmpnew);
2550 }
2551 break;
2552 }
2553 default:
2554 PERROR("unknown consumer type");
2555 exit(EXIT_FAILURE);
2556 }
2557 if (errno != 0) {
2558 PERROR("Consumer execl()");
2559 }
2560 /* Reaching this point, we got a failure on our execl(). */
2561 exit(EXIT_FAILURE);
2562 } else if (pid > 0) {
2563 ret = pid;
2564 } else {
2565 PERROR("start consumer fork");
2566 ret = -errno;
2567 }
2568 error:
2569 return ret;
2570 }
2571
2572 /*
2573 * Spawn the consumerd daemon and session daemon thread.
2574 */
2575 static int start_consumerd(struct consumer_data *consumer_data)
2576 {
2577 int ret;
2578
2579 /*
2580 * Set the listen() state on the socket since there is a possible race
2581 * between the exec() of the consumer daemon and this call if place in the
2582 * consumer thread. See bug #366 for more details.
2583 */
2584 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2585 if (ret < 0) {
2586 goto error;
2587 }
2588
2589 pthread_mutex_lock(&consumer_data->pid_mutex);
2590 if (consumer_data->pid != 0) {
2591 pthread_mutex_unlock(&consumer_data->pid_mutex);
2592 goto end;
2593 }
2594
2595 ret = spawn_consumerd(consumer_data);
2596 if (ret < 0) {
2597 ERR("Spawning consumerd failed");
2598 pthread_mutex_unlock(&consumer_data->pid_mutex);
2599 goto error;
2600 }
2601
2602 /* Setting up the consumer_data pid */
2603 consumer_data->pid = ret;
2604 DBG2("Consumer pid %d", consumer_data->pid);
2605 pthread_mutex_unlock(&consumer_data->pid_mutex);
2606
2607 DBG2("Spawning consumer control thread");
2608 ret = spawn_consumer_thread(consumer_data);
2609 if (ret < 0) {
2610 ERR("Fatal error spawning consumer control thread");
2611 goto error;
2612 }
2613
2614 end:
2615 return 0;
2616
2617 error:
2618 /* Cleanup already created sockets on error. */
2619 if (consumer_data->err_sock >= 0) {
2620 int err;
2621
2622 err = close(consumer_data->err_sock);
2623 if (err < 0) {
2624 PERROR("close consumer data error socket");
2625 }
2626 }
2627 return ret;
2628 }
2629
2630 /*
2631 * Setup necessary data for kernel tracer action.
2632 */
2633 static int init_kernel_tracer(void)
2634 {
2635 int ret;
2636
2637 /* Modprobe lttng kernel modules */
2638 ret = modprobe_lttng_control();
2639 if (ret < 0) {
2640 goto error;
2641 }
2642
2643 /* Open debugfs lttng */
2644 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2645 if (kernel_tracer_fd < 0) {
2646 DBG("Failed to open %s", module_proc_lttng);
2647 ret = -1;
2648 goto error_open;
2649 }
2650
2651 /* Validate kernel version */
2652 ret = kernel_validate_version(kernel_tracer_fd);
2653 if (ret < 0) {
2654 goto error_version;
2655 }
2656
2657 ret = modprobe_lttng_data();
2658 if (ret < 0) {
2659 goto error_modules;
2660 }
2661
2662 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2663 return 0;
2664
2665 error_version:
2666 modprobe_remove_lttng_control();
2667 ret = close(kernel_tracer_fd);
2668 if (ret) {
2669 PERROR("close");
2670 }
2671 kernel_tracer_fd = -1;
2672 return LTTNG_ERR_KERN_VERSION;
2673
2674 error_modules:
2675 ret = close(kernel_tracer_fd);
2676 if (ret) {
2677 PERROR("close");
2678 }
2679
2680 error_open:
2681 modprobe_remove_lttng_control();
2682
2683 error:
2684 WARN("No kernel tracer available");
2685 kernel_tracer_fd = -1;
2686 if (!is_root) {
2687 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2688 } else {
2689 return LTTNG_ERR_KERN_NA;
2690 }
2691 }
2692
2693
2694 /*
2695 * Copy consumer output from the tracing session to the domain session. The
2696 * function also applies the right modification on a per domain basis for the
2697 * trace files destination directory.
2698 *
2699 * Should *NOT* be called with RCU read-side lock held.
2700 */
2701 static int copy_session_consumer(int domain, struct ltt_session *session)
2702 {
2703 int ret;
2704 const char *dir_name;
2705 struct consumer_output *consumer;
2706
2707 assert(session);
2708 assert(session->consumer);
2709
2710 switch (domain) {
2711 case LTTNG_DOMAIN_KERNEL:
2712 DBG3("Copying tracing session consumer output in kernel session");
2713 /*
2714 * XXX: We should audit the session creation and what this function
2715 * does "extra" in order to avoid a destroy since this function is used
2716 * in the domain session creation (kernel and ust) only. Same for UST
2717 * domain.
2718 */
2719 if (session->kernel_session->consumer) {
2720 consumer_output_put(session->kernel_session->consumer);
2721 }
2722 session->kernel_session->consumer =
2723 consumer_copy_output(session->consumer);
2724 /* Ease our life a bit for the next part */
2725 consumer = session->kernel_session->consumer;
2726 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2727 break;
2728 case LTTNG_DOMAIN_JUL:
2729 case LTTNG_DOMAIN_LOG4J:
2730 case LTTNG_DOMAIN_PYTHON:
2731 case LTTNG_DOMAIN_UST:
2732 DBG3("Copying tracing session consumer output in UST session");
2733 if (session->ust_session->consumer) {
2734 consumer_output_put(session->ust_session->consumer);
2735 }
2736 session->ust_session->consumer =
2737 consumer_copy_output(session->consumer);
2738 /* Ease our life a bit for the next part */
2739 consumer = session->ust_session->consumer;
2740 dir_name = DEFAULT_UST_TRACE_DIR;
2741 break;
2742 default:
2743 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2744 goto error;
2745 }
2746
2747 /* Append correct directory to subdir */
2748 strncat(consumer->subdir, dir_name,
2749 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2750 DBG3("Copy session consumer subdir %s", consumer->subdir);
2751
2752 ret = LTTNG_OK;
2753
2754 error:
2755 return ret;
2756 }
2757
2758 /*
2759 * Create an UST session and add it to the session ust list.
2760 *
2761 * Should *NOT* be called with RCU read-side lock held.
2762 */
2763 static int create_ust_session(struct ltt_session *session,
2764 struct lttng_domain *domain)
2765 {
2766 int ret;
2767 struct ltt_ust_session *lus = NULL;
2768
2769 assert(session);
2770 assert(domain);
2771 assert(session->consumer);
2772
2773 switch (domain->type) {
2774 case LTTNG_DOMAIN_JUL:
2775 case LTTNG_DOMAIN_LOG4J:
2776 case LTTNG_DOMAIN_PYTHON:
2777 case LTTNG_DOMAIN_UST:
2778 break;
2779 default:
2780 ERR("Unknown UST domain on create session %d", domain->type);
2781 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2782 goto error;
2783 }
2784
2785 DBG("Creating UST session");
2786
2787 lus = trace_ust_create_session(session->id);
2788 if (lus == NULL) {
2789 ret = LTTNG_ERR_UST_SESS_FAIL;
2790 goto error;
2791 }
2792
2793 lus->uid = session->uid;
2794 lus->gid = session->gid;
2795 lus->output_traces = session->output_traces;
2796 lus->snapshot_mode = session->snapshot_mode;
2797 lus->live_timer_interval = session->live_timer;
2798 session->ust_session = lus;
2799 if (session->shm_path[0]) {
2800 strncpy(lus->root_shm_path, session->shm_path,
2801 sizeof(lus->root_shm_path));
2802 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2803 strncpy(lus->shm_path, session->shm_path,
2804 sizeof(lus->shm_path));
2805 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2806 strncat(lus->shm_path, "/ust",
2807 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2808 }
2809 /* Copy session output to the newly created UST session */
2810 ret = copy_session_consumer(domain->type, session);
2811 if (ret != LTTNG_OK) {
2812 goto error;
2813 }
2814
2815 return LTTNG_OK;
2816
2817 error:
2818 free(lus);
2819 session->ust_session = NULL;
2820 return ret;
2821 }
2822
2823 /*
2824 * Create a kernel tracer session then create the default channel.
2825 */
2826 static int create_kernel_session(struct ltt_session *session)
2827 {
2828 int ret;
2829
2830 DBG("Creating kernel session");
2831
2832 ret = kernel_create_session(session, kernel_tracer_fd);
2833 if (ret < 0) {
2834 ret = LTTNG_ERR_KERN_SESS_FAIL;
2835 goto error;
2836 }
2837
2838 /* Code flow safety */
2839 assert(session->kernel_session);
2840
2841 /* Copy session output to the newly created Kernel session */
2842 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2843 if (ret != LTTNG_OK) {
2844 goto error;
2845 }
2846
2847 /* Create directory(ies) on local filesystem. */
2848 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2849 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2850 ret = run_as_mkdir_recursive(
2851 session->kernel_session->consumer->dst.trace_path,
2852 S_IRWXU | S_IRWXG, session->uid, session->gid);
2853 if (ret < 0) {
2854 if (errno != EEXIST) {
2855 ERR("Trace directory creation error");
2856 goto error;
2857 }
2858 }
2859 }
2860
2861 session->kernel_session->uid = session->uid;
2862 session->kernel_session->gid = session->gid;
2863 session->kernel_session->output_traces = session->output_traces;
2864 session->kernel_session->snapshot_mode = session->snapshot_mode;
2865
2866 return LTTNG_OK;
2867
2868 error:
2869 trace_kernel_destroy_session(session->kernel_session);
2870 session->kernel_session = NULL;
2871 return ret;
2872 }
2873
2874 /*
2875 * Count number of session permitted by uid/gid.
2876 */
2877 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2878 {
2879 unsigned int i = 0;
2880 struct ltt_session *session;
2881
2882 DBG("Counting number of available session for UID %d GID %d",
2883 uid, gid);
2884 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2885 /*
2886 * Only list the sessions the user can control.
2887 */
2888 if (!session_access_ok(session, uid, gid)) {
2889 continue;
2890 }
2891 i++;
2892 }
2893 return i;
2894 }
2895
2896 /*
2897 * Process the command requested by the lttng client within the command
2898 * context structure. This function make sure that the return structure (llm)
2899 * is set and ready for transmission before returning.
2900 *
2901 * Return any error encountered or 0 for success.
2902 *
2903 * "sock" is only used for special-case var. len data.
2904 *
2905 * Should *NOT* be called with RCU read-side lock held.
2906 */
2907 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2908 int *sock_error)
2909 {
2910 int ret = LTTNG_OK;
2911 int need_tracing_session = 1;
2912 int need_domain;
2913
2914 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2915
2916 *sock_error = 0;
2917
2918 switch (cmd_ctx->lsm->cmd_type) {
2919 case LTTNG_CREATE_SESSION:
2920 case LTTNG_CREATE_SESSION_SNAPSHOT:
2921 case LTTNG_CREATE_SESSION_LIVE:
2922 case LTTNG_DESTROY_SESSION:
2923 case LTTNG_LIST_SESSIONS:
2924 case LTTNG_LIST_DOMAINS:
2925 case LTTNG_START_TRACE:
2926 case LTTNG_STOP_TRACE:
2927 case LTTNG_DATA_PENDING:
2928 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2929 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2930 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2931 case LTTNG_SNAPSHOT_RECORD:
2932 case LTTNG_SAVE_SESSION:
2933 case LTTNG_SET_SESSION_SHM_PATH:
2934 need_domain = 0;
2935 break;
2936 default:
2937 need_domain = 1;
2938 }
2939
2940 if (opt_no_kernel && need_domain
2941 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2942 if (!is_root) {
2943 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2944 } else {
2945 ret = LTTNG_ERR_KERN_NA;
2946 }
2947 goto error;
2948 }
2949
2950 /* Deny register consumer if we already have a spawned consumer. */
2951 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2952 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2953 if (kconsumer_data.pid > 0) {
2954 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2955 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2956 goto error;
2957 }
2958 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2959 }
2960
2961 /*
2962 * Check for command that don't needs to allocate a returned payload. We do
2963 * this here so we don't have to make the call for no payload at each
2964 * command.
2965 */
2966 switch(cmd_ctx->lsm->cmd_type) {
2967 case LTTNG_LIST_SESSIONS:
2968 case LTTNG_LIST_TRACEPOINTS:
2969 case LTTNG_LIST_TRACEPOINT_FIELDS:
2970 case LTTNG_LIST_DOMAINS:
2971 case LTTNG_LIST_CHANNELS:
2972 case LTTNG_LIST_EVENTS:
2973 case LTTNG_LIST_SYSCALLS:
2974 case LTTNG_LIST_TRACKER_PIDS:
2975 break;
2976 default:
2977 /* Setup lttng message with no payload */
2978 ret = setup_lttng_msg(cmd_ctx, 0);
2979 if (ret < 0) {
2980 /* This label does not try to unlock the session */
2981 goto init_setup_error;
2982 }
2983 }
2984
2985 /* Commands that DO NOT need a session. */
2986 switch (cmd_ctx->lsm->cmd_type) {
2987 case LTTNG_CREATE_SESSION:
2988 case LTTNG_CREATE_SESSION_SNAPSHOT:
2989 case LTTNG_CREATE_SESSION_LIVE:
2990 case LTTNG_CALIBRATE:
2991 case LTTNG_LIST_SESSIONS:
2992 case LTTNG_LIST_TRACEPOINTS:
2993 case LTTNG_LIST_SYSCALLS:
2994 case LTTNG_LIST_TRACEPOINT_FIELDS:
2995 case LTTNG_SAVE_SESSION:
2996 need_tracing_session = 0;
2997 break;
2998 default:
2999 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3000 /*
3001 * We keep the session list lock across _all_ commands
3002 * for now, because the per-session lock does not
3003 * handle teardown properly.
3004 */
3005 session_lock_list();
3006 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3007 if (cmd_ctx->session == NULL) {
3008 ret = LTTNG_ERR_SESS_NOT_FOUND;
3009 goto error;
3010 } else {
3011 /* Acquire lock for the session */
3012 session_lock(cmd_ctx->session);
3013 }
3014 break;
3015 }
3016
3017 /*
3018 * Commands that need a valid session but should NOT create one if none
3019 * exists. Instead of creating one and destroying it when the command is
3020 * handled, process that right before so we save some round trip in useless
3021 * code path.
3022 */
3023 switch (cmd_ctx->lsm->cmd_type) {
3024 case LTTNG_DISABLE_CHANNEL:
3025 case LTTNG_DISABLE_EVENT:
3026 switch (cmd_ctx->lsm->domain.type) {
3027 case LTTNG_DOMAIN_KERNEL:
3028 if (!cmd_ctx->session->kernel_session) {
3029 ret = LTTNG_ERR_NO_CHANNEL;
3030 goto error;
3031 }
3032 break;
3033 case LTTNG_DOMAIN_JUL:
3034 case LTTNG_DOMAIN_LOG4J:
3035 case LTTNG_DOMAIN_PYTHON:
3036 case LTTNG_DOMAIN_UST:
3037 if (!cmd_ctx->session->ust_session) {
3038 ret = LTTNG_ERR_NO_CHANNEL;
3039 goto error;
3040 }
3041 break;
3042 default:
3043 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3044 goto error;
3045 }
3046 default:
3047 break;
3048 }
3049
3050 if (!need_domain) {
3051 goto skip_domain;
3052 }
3053
3054 /*
3055 * Check domain type for specific "pre-action".
3056 */
3057 switch (cmd_ctx->lsm->domain.type) {
3058 case LTTNG_DOMAIN_KERNEL:
3059 if (!is_root) {
3060 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3061 goto error;
3062 }
3063
3064 /* Kernel tracer check */
3065 if (kernel_tracer_fd == -1) {
3066 /* Basically, load kernel tracer modules */
3067 ret = init_kernel_tracer();
3068 if (ret != 0) {
3069 goto error;
3070 }
3071 }
3072
3073 /* Consumer is in an ERROR state. Report back to client */
3074 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3075 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3076 goto error;
3077 }
3078
3079 /* Need a session for kernel command */
3080 if (need_tracing_session) {
3081 if (cmd_ctx->session->kernel_session == NULL) {
3082 ret = create_kernel_session(cmd_ctx->session);
3083 if (ret < 0) {
3084 ret = LTTNG_ERR_KERN_SESS_FAIL;
3085 goto error;
3086 }
3087 }
3088
3089 /* Start the kernel consumer daemon */
3090 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3091 if (kconsumer_data.pid == 0 &&
3092 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3093 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3094 ret = start_consumerd(&kconsumer_data);
3095 if (ret < 0) {
3096 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3097 goto error;
3098 }
3099 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3100 } else {
3101 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3102 }
3103
3104 /*
3105 * The consumer was just spawned so we need to add the socket to
3106 * the consumer output of the session if exist.
3107 */
3108 ret = consumer_create_socket(&kconsumer_data,
3109 cmd_ctx->session->kernel_session->consumer);
3110 if (ret < 0) {
3111 goto error;
3112 }
3113 }
3114
3115 break;
3116 case LTTNG_DOMAIN_JUL:
3117 case LTTNG_DOMAIN_LOG4J:
3118 case LTTNG_DOMAIN_PYTHON:
3119 case LTTNG_DOMAIN_UST:
3120 {
3121 if (!ust_app_supported()) {
3122 ret = LTTNG_ERR_NO_UST;
3123 goto error;
3124 }
3125 /* Consumer is in an ERROR state. Report back to client */
3126 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3127 ret = LTTNG_ERR_NO_USTCONSUMERD;
3128 goto error;
3129 }
3130
3131 if (need_tracing_session) {
3132 /* Create UST session if none exist. */
3133 if (cmd_ctx->session->ust_session == NULL) {
3134 ret = create_ust_session(cmd_ctx->session,
3135 &cmd_ctx->lsm->domain);
3136 if (ret != LTTNG_OK) {
3137 goto error;
3138 }
3139 }
3140
3141 /* Start the UST consumer daemons */
3142 /* 64-bit */
3143 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3144 if (consumerd64_bin[0] != '\0' &&
3145 ustconsumer64_data.pid == 0 &&
3146 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3147 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3148 ret = start_consumerd(&ustconsumer64_data);
3149 if (ret < 0) {
3150 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3151 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3152 goto error;
3153 }
3154
3155 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3156 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3157 } else {
3158 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3159 }
3160
3161 /*
3162 * Setup socket for consumer 64 bit. No need for atomic access
3163 * since it was set above and can ONLY be set in this thread.
3164 */
3165 ret = consumer_create_socket(&ustconsumer64_data,
3166 cmd_ctx->session->ust_session->consumer);
3167 if (ret < 0) {
3168 goto error;
3169 }
3170
3171 /* 32-bit */
3172 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3173 if (consumerd32_bin[0] != '\0' &&
3174 ustconsumer32_data.pid == 0 &&
3175 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3176 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3177 ret = start_consumerd(&ustconsumer32_data);
3178 if (ret < 0) {
3179 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3180 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3181 goto error;
3182 }
3183
3184 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3185 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3186 } else {
3187 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3188 }
3189
3190 /*
3191 * Setup socket for consumer 64 bit. No need for atomic access
3192 * since it was set above and can ONLY be set in this thread.
3193 */
3194 ret = consumer_create_socket(&ustconsumer32_data,
3195 cmd_ctx->session->ust_session->consumer);
3196 if (ret < 0) {
3197 goto error;
3198 }
3199 }
3200 break;
3201 }
3202 default:
3203 break;
3204 }
3205 skip_domain:
3206
3207 /* Validate consumer daemon state when start/stop trace command */
3208 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3209 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3210 switch (cmd_ctx->lsm->domain.type) {
3211 case LTTNG_DOMAIN_JUL:
3212 case LTTNG_DOMAIN_LOG4J:
3213 case LTTNG_DOMAIN_PYTHON:
3214 case LTTNG_DOMAIN_UST:
3215 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3216 ret = LTTNG_ERR_NO_USTCONSUMERD;
3217 goto error;
3218 }
3219 break;
3220 case LTTNG_DOMAIN_KERNEL:
3221 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3222 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3223 goto error;
3224 }
3225 break;
3226 }
3227 }
3228
3229 /*
3230 * Check that the UID or GID match that of the tracing session.
3231 * The root user can interact with all sessions.
3232 */
3233 if (need_tracing_session) {
3234 if (!session_access_ok(cmd_ctx->session,
3235 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3236 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3237 ret = LTTNG_ERR_EPERM;
3238 goto error;
3239 }
3240 }
3241
3242 /*
3243 * Send relayd information to consumer as soon as we have a domain and a
3244 * session defined.
3245 */
3246 if (cmd_ctx->session && need_domain) {
3247 /*
3248 * Setup relayd if not done yet. If the relayd information was already
3249 * sent to the consumer, this call will gracefully return.
3250 */
3251 ret = cmd_setup_relayd(cmd_ctx->session);
3252 if (ret != LTTNG_OK) {
3253 goto error;
3254 }
3255 }
3256
3257 /* Process by command type */
3258 switch (cmd_ctx->lsm->cmd_type) {
3259 case LTTNG_ADD_CONTEXT:
3260 {
3261 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3262 cmd_ctx->lsm->u.context.channel_name,
3263 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3264 break;
3265 }
3266 case LTTNG_DISABLE_CHANNEL:
3267 {
3268 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3269 cmd_ctx->lsm->u.disable.channel_name);
3270 break;
3271 }
3272 case LTTNG_DISABLE_EVENT:
3273 {
3274
3275 /*
3276 * FIXME: handle filter; for now we just receive the filter's
3277 * bytecode along with the filter expression which are sent by
3278 * liblttng-ctl and discard them.
3279 *
3280 * This fixes an issue where the client may block while sending
3281 * the filter payload and encounter an error because the session
3282 * daemon closes the socket without ever handling this data.
3283 */
3284 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3285 cmd_ctx->lsm->u.disable.bytecode_len;
3286
3287 if (count) {
3288 char data[LTTNG_FILTER_MAX_LEN];
3289
3290 DBG("Discarding disable event command payload of size %zu", count);
3291 while (count) {
3292 ret = lttcomm_recv_unix_sock(sock, data,
3293 count > sizeof(data) ? sizeof(data) : count);
3294 if (ret < 0) {
3295 goto error;
3296 }
3297
3298 count -= (size_t) ret;
3299 }
3300 }
3301 /* FIXME: passing packed structure to non-packed pointer */
3302 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3303 cmd_ctx->lsm->u.disable.channel_name,
3304 &cmd_ctx->lsm->u.disable.event);
3305 break;
3306 }
3307 case LTTNG_ENABLE_CHANNEL:
3308 {
3309 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3310 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3311 break;
3312 }
3313 case LTTNG_TRACK_PID:
3314 {
3315 ret = cmd_track_pid(cmd_ctx->session,
3316 cmd_ctx->lsm->domain.type,
3317 cmd_ctx->lsm->u.pid_tracker.pid);
3318 break;
3319 }
3320 case LTTNG_UNTRACK_PID:
3321 {
3322 ret = cmd_untrack_pid(cmd_ctx->session,
3323 cmd_ctx->lsm->domain.type,
3324 cmd_ctx->lsm->u.pid_tracker.pid);
3325 break;
3326 }
3327 case LTTNG_ENABLE_EVENT:
3328 {
3329 struct lttng_event_exclusion *exclusion = NULL;
3330 struct lttng_filter_bytecode *bytecode = NULL;
3331 char *filter_expression = NULL;
3332
3333 /* Handle exclusion events and receive it from the client. */
3334 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3335 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3336
3337 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3338 (count * LTTNG_SYMBOL_NAME_LEN));
3339 if (!exclusion) {
3340 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3341 goto error;
3342 }
3343
3344 DBG("Receiving var len exclusion event list from client ...");
3345 exclusion->count = count;
3346 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3347 count * LTTNG_SYMBOL_NAME_LEN);
3348 if (ret <= 0) {
3349 DBG("Nothing recv() from client var len data... continuing");
3350 *sock_error = 1;
3351 free(exclusion);
3352 ret = LTTNG_ERR_EXCLUSION_INVAL;
3353 goto error;
3354 }
3355 }
3356
3357 /* Get filter expression from client. */
3358 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3359 size_t expression_len =
3360 cmd_ctx->lsm->u.enable.expression_len;
3361
3362 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3363 ret = LTTNG_ERR_FILTER_INVAL;
3364 free(exclusion);
3365 goto error;
3366 }
3367
3368 filter_expression = zmalloc(expression_len);
3369 if (!filter_expression) {
3370 free(exclusion);
3371 ret = LTTNG_ERR_FILTER_NOMEM;
3372 goto error;
3373 }
3374
3375 /* Receive var. len. data */
3376 DBG("Receiving var len filter's expression from client ...");
3377 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3378 expression_len);
3379 if (ret <= 0) {
3380 DBG("Nothing recv() from client car len data... continuing");
3381 *sock_error = 1;
3382 free(filter_expression);
3383 free(exclusion);
3384 ret = LTTNG_ERR_FILTER_INVAL;
3385 goto error;
3386 }
3387 }
3388
3389 /* Handle filter and get bytecode from client. */
3390 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3391 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3392
3393 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3394 ret = LTTNG_ERR_FILTER_INVAL;
3395 free(filter_expression);
3396 free(exclusion);
3397 goto error;
3398 }
3399
3400 bytecode = zmalloc(bytecode_len);
3401 if (!bytecode) {
3402 free(filter_expression);
3403 free(exclusion);
3404 ret = LTTNG_ERR_FILTER_NOMEM;
3405 goto error;
3406 }
3407
3408 /* Receive var. len. data */
3409 DBG("Receiving var len filter's bytecode from client ...");
3410 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3411 if (ret <= 0) {
3412 DBG("Nothing recv() from client car len data... continuing");
3413 *sock_error = 1;
3414 free(filter_expression);
3415 free(bytecode);
3416 free(exclusion);
3417 ret = LTTNG_ERR_FILTER_INVAL;
3418 goto error;
3419 }
3420
3421 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3422 free(filter_expression);
3423 free(bytecode);
3424 free(exclusion);
3425 ret = LTTNG_ERR_FILTER_INVAL;
3426 goto error;
3427 }
3428 }
3429
3430 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3431 cmd_ctx->lsm->u.enable.channel_name,
3432 &cmd_ctx->lsm->u.enable.event,
3433 filter_expression, bytecode, exclusion,
3434 kernel_poll_pipe[1]);
3435 break;
3436 }
3437 case LTTNG_LIST_TRACEPOINTS:
3438 {
3439 struct lttng_event *events;
3440 ssize_t nb_events;
3441
3442 session_lock_list();
3443 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3444 session_unlock_list();
3445 if (nb_events < 0) {
3446 /* Return value is a negative lttng_error_code. */
3447 ret = -nb_events;
3448 goto error;
3449 }
3450
3451 /*
3452 * Setup lttng message with payload size set to the event list size in
3453 * bytes and then copy list into the llm payload.
3454 */
3455 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3456 if (ret < 0) {
3457 free(events);
3458 goto setup_error;
3459 }
3460
3461 /* Copy event list into message payload */
3462 memcpy(cmd_ctx->llm->payload, events,
3463 sizeof(struct lttng_event) * nb_events);
3464
3465 free(events);
3466
3467 ret = LTTNG_OK;
3468 break;
3469 }
3470 case LTTNG_LIST_TRACEPOINT_FIELDS:
3471 {
3472 struct lttng_event_field *fields;
3473 ssize_t nb_fields;
3474
3475 session_lock_list();
3476 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3477 &fields);
3478 session_unlock_list();
3479 if (nb_fields < 0) {
3480 /* Return value is a negative lttng_error_code. */
3481 ret = -nb_fields;
3482 goto error;
3483 }
3484
3485 /*
3486 * Setup lttng message with payload size set to the event list size in
3487 * bytes and then copy list into the llm payload.
3488 */
3489 ret = setup_lttng_msg(cmd_ctx,
3490 sizeof(struct lttng_event_field) * nb_fields);
3491 if (ret < 0) {
3492 free(fields);
3493 goto setup_error;
3494 }
3495
3496 /* Copy event list into message payload */
3497 memcpy(cmd_ctx->llm->payload, fields,
3498 sizeof(struct lttng_event_field) * nb_fields);
3499
3500 free(fields);
3501
3502 ret = LTTNG_OK;
3503 break;
3504 }
3505 case LTTNG_LIST_SYSCALLS:
3506 {
3507 struct lttng_event *events;
3508 ssize_t nb_events;
3509
3510 nb_events = cmd_list_syscalls(&events);
3511 if (nb_events < 0) {
3512 /* Return value is a negative lttng_error_code. */
3513 ret = -nb_events;
3514 goto error;
3515 }
3516
3517 /*
3518 * Setup lttng message with payload size set to the event list size in
3519 * bytes and then copy list into the llm payload.
3520 */
3521 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3522 if (ret < 0) {
3523 free(events);
3524 goto setup_error;
3525 }
3526
3527 /* Copy event list into message payload */
3528 memcpy(cmd_ctx->llm->payload, events,
3529 sizeof(struct lttng_event) * nb_events);
3530
3531 free(events);
3532
3533 ret = LTTNG_OK;
3534 break;
3535 }
3536 case LTTNG_LIST_TRACKER_PIDS:
3537 {
3538 int32_t *pids = NULL;
3539 ssize_t nr_pids;
3540
3541 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3542 cmd_ctx->lsm->domain.type, &pids);
3543 if (nr_pids < 0) {
3544 /* Return value is a negative lttng_error_code. */
3545 ret = -nr_pids;
3546 goto error;
3547 }
3548
3549 /*
3550 * Setup lttng message with payload size set to the event list size in
3551 * bytes and then copy list into the llm payload.
3552 */
3553 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3554 if (ret < 0) {
3555 free(pids);
3556 goto setup_error;
3557 }
3558
3559 /* Copy event list into message payload */
3560 memcpy(cmd_ctx->llm->payload, pids,
3561 sizeof(int) * nr_pids);
3562
3563 free(pids);
3564
3565 ret = LTTNG_OK;
3566 break;
3567 }
3568 case LTTNG_SET_CONSUMER_URI:
3569 {
3570 size_t nb_uri, len;
3571 struct lttng_uri *uris;
3572
3573 nb_uri = cmd_ctx->lsm->u.uri.size;
3574 len = nb_uri * sizeof(struct lttng_uri);
3575
3576 if (nb_uri == 0) {
3577 ret = LTTNG_ERR_INVALID;
3578 goto error;
3579 }
3580
3581 uris = zmalloc(len);
3582 if (uris == NULL) {
3583 ret = LTTNG_ERR_FATAL;
3584 goto error;
3585 }
3586
3587 /* Receive variable len data */
3588 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3589 ret = lttcomm_recv_unix_sock(sock, uris, len);
3590 if (ret <= 0) {
3591 DBG("No URIs received from client... continuing");
3592 *sock_error = 1;
3593 ret = LTTNG_ERR_SESSION_FAIL;
3594 free(uris);
3595 goto error;
3596 }
3597
3598 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3599 free(uris);
3600 if (ret != LTTNG_OK) {
3601 goto error;
3602 }
3603
3604
3605 break;
3606 }
3607 case LTTNG_START_TRACE:
3608 {
3609 ret = cmd_start_trace(cmd_ctx->session);
3610 break;
3611 }
3612 case LTTNG_STOP_TRACE:
3613 {
3614 ret = cmd_stop_trace(cmd_ctx->session);
3615 break;
3616 }
3617 case LTTNG_CREATE_SESSION:
3618 {
3619 size_t nb_uri, len;
3620 struct lttng_uri *uris = NULL;
3621
3622 nb_uri = cmd_ctx->lsm->u.uri.size;
3623 len = nb_uri * sizeof(struct lttng_uri);
3624
3625 if (nb_uri > 0) {
3626 uris = zmalloc(len);
3627 if (uris == NULL) {
3628 ret = LTTNG_ERR_FATAL;
3629 goto error;
3630 }
3631
3632 /* Receive variable len data */
3633 DBG("Waiting for %zu URIs from client ...", nb_uri);
3634 ret = lttcomm_recv_unix_sock(sock, uris, len);
3635 if (ret <= 0) {
3636 DBG("No URIs received from client... continuing");
3637 *sock_error = 1;
3638 ret = LTTNG_ERR_SESSION_FAIL;
3639 free(uris);
3640 goto error;
3641 }
3642
3643 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3644 DBG("Creating session with ONE network URI is a bad call");
3645 ret = LTTNG_ERR_SESSION_FAIL;
3646 free(uris);
3647 goto error;
3648 }
3649 }
3650
3651 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3652 &cmd_ctx->creds, 0);
3653
3654 free(uris);
3655
3656 break;
3657 }
3658 case LTTNG_DESTROY_SESSION:
3659 {
3660 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3661
3662 /* Set session to NULL so we do not unlock it after free. */
3663 cmd_ctx->session = NULL;
3664 break;
3665 }
3666 case LTTNG_LIST_DOMAINS:
3667 {
3668 ssize_t nb_dom;
3669 struct lttng_domain *domains = NULL;
3670
3671 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3672 if (nb_dom < 0) {
3673 /* Return value is a negative lttng_error_code. */
3674 ret = -nb_dom;
3675 goto error;
3676 }
3677
3678 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3679 if (ret < 0) {
3680 free(domains);
3681 goto setup_error;
3682 }
3683
3684 /* Copy event list into message payload */
3685 memcpy(cmd_ctx->llm->payload, domains,
3686 nb_dom * sizeof(struct lttng_domain));
3687
3688 free(domains);
3689
3690 ret = LTTNG_OK;
3691 break;
3692 }
3693 case LTTNG_LIST_CHANNELS:
3694 {
3695 int nb_chan;
3696 struct lttng_channel *channels = NULL;
3697
3698 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3699 cmd_ctx->session, &channels);
3700 if (nb_chan < 0) {
3701 /* Return value is a negative lttng_error_code. */
3702 ret = -nb_chan;
3703 goto error;
3704 }
3705
3706 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3707 if (ret < 0) {
3708 free(channels);
3709 goto setup_error;
3710 }
3711
3712 /* Copy event list into message payload */
3713 memcpy(cmd_ctx->llm->payload, channels,
3714 nb_chan * sizeof(struct lttng_channel));
3715
3716 free(channels);
3717
3718 ret = LTTNG_OK;
3719 break;
3720 }
3721 case LTTNG_LIST_EVENTS:
3722 {
3723 ssize_t nb_event;
3724 struct lttng_event *events = NULL;
3725
3726 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3727 cmd_ctx->lsm->u.list.channel_name, &events);
3728 if (nb_event < 0) {
3729 /* Return value is a negative lttng_error_code. */
3730 ret = -nb_event;
3731 goto error;
3732 }
3733
3734 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3735 if (ret < 0) {
3736 free(events);
3737 goto setup_error;
3738 }
3739
3740 /* Copy event list into message payload */
3741 memcpy(cmd_ctx->llm->payload, events,
3742 nb_event * sizeof(struct lttng_event));
3743
3744 free(events);
3745
3746 ret = LTTNG_OK;
3747 break;
3748 }
3749 case LTTNG_LIST_SESSIONS:
3750 {
3751 unsigned int nr_sessions;
3752
3753 session_lock_list();
3754 nr_sessions = lttng_sessions_count(
3755 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3756 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3757
3758 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3759 if (ret < 0) {
3760 session_unlock_list();
3761 goto setup_error;
3762 }
3763
3764 /* Filled the session array */
3765 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3766 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3767 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3768
3769 session_unlock_list();
3770
3771 ret = LTTNG_OK;
3772 break;
3773 }
3774 case LTTNG_CALIBRATE:
3775 {
3776 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3777 &cmd_ctx->lsm->u.calibrate);
3778 break;
3779 }
3780 case LTTNG_REGISTER_CONSUMER:
3781 {
3782 struct consumer_data *cdata;
3783
3784 switch (cmd_ctx->lsm->domain.type) {
3785 case LTTNG_DOMAIN_KERNEL:
3786 cdata = &kconsumer_data;
3787 break;
3788 default:
3789 ret = LTTNG_ERR_UND;
3790 goto error;
3791 }
3792
3793 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3794 cmd_ctx->lsm->u.reg.path, cdata);
3795 break;
3796 }
3797 case LTTNG_DATA_PENDING:
3798 {
3799 int pending_ret;
3800
3801 /* 1 byte to return whether or not data is pending */
3802 ret = setup_lttng_msg(cmd_ctx, 1);
3803 if (ret < 0) {
3804 goto setup_error;
3805 }
3806
3807 pending_ret = cmd_data_pending(cmd_ctx->session);
3808 /*
3809 * FIXME
3810 *
3811 * This function may returns 0 or 1 to indicate whether or not
3812 * there is data pending. In case of error, it should return an
3813 * LTTNG_ERR code. However, some code paths may still return
3814 * a nondescript error code, which we handle by returning an
3815 * "unknown" error.
3816 */
3817 if (pending_ret == 0 || pending_ret == 1) {
3818 ret = LTTNG_OK;
3819 } else if (pending_ret < 0) {
3820 ret = LTTNG_ERR_UNK;
3821 goto setup_error;
3822 } else {
3823 ret = pending_ret;
3824 goto setup_error;
3825 }
3826
3827 *cmd_ctx->llm->payload = (uint8_t) pending_ret;
3828 break;
3829 }
3830 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3831 {
3832 struct lttcomm_lttng_output_id reply;
3833
3834 ret = cmd_snapshot_add_output(cmd_ctx->session,
3835 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3836 if (ret != LTTNG_OK) {
3837 goto error;
3838 }
3839
3840 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3841 if (ret < 0) {
3842 goto setup_error;
3843 }
3844
3845 /* Copy output list into message payload */
3846 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3847 ret = LTTNG_OK;
3848 break;
3849 }
3850 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3851 {
3852 ret = cmd_snapshot_del_output(cmd_ctx->session,
3853 &cmd_ctx->lsm->u.snapshot_output.output);
3854 break;
3855 }
3856 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3857 {
3858 ssize_t nb_output;
3859 struct lttng_snapshot_output *outputs = NULL;
3860
3861 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3862 if (nb_output < 0) {
3863 ret = -nb_output;
3864 goto error;
3865 }
3866
3867 ret = setup_lttng_msg(cmd_ctx,
3868 nb_output * sizeof(struct lttng_snapshot_output));
3869 if (ret < 0) {
3870 free(outputs);
3871 goto setup_error;
3872 }
3873
3874 if (outputs) {
3875 /* Copy output list into message payload */
3876 memcpy(cmd_ctx->llm->payload, outputs,
3877 nb_output * sizeof(struct lttng_snapshot_output));
3878 free(outputs);
3879 }
3880
3881 ret = LTTNG_OK;
3882 break;
3883 }
3884 case LTTNG_SNAPSHOT_RECORD:
3885 {
3886 ret = cmd_snapshot_record(cmd_ctx->session,
3887 &cmd_ctx->lsm->u.snapshot_record.output,
3888 cmd_ctx->lsm->u.snapshot_record.wait);
3889 break;
3890 }
3891 case LTTNG_CREATE_SESSION_SNAPSHOT:
3892 {
3893 size_t nb_uri, len;
3894 struct lttng_uri *uris = NULL;
3895
3896 nb_uri = cmd_ctx->lsm->u.uri.size;
3897 len = nb_uri * sizeof(struct lttng_uri);
3898
3899 if (nb_uri > 0) {
3900 uris = zmalloc(len);
3901 if (uris == NULL) {
3902 ret = LTTNG_ERR_FATAL;
3903 goto error;
3904 }
3905
3906 /* Receive variable len data */
3907 DBG("Waiting for %zu URIs from client ...", nb_uri);
3908 ret = lttcomm_recv_unix_sock(sock, uris, len);
3909 if (ret <= 0) {
3910 DBG("No URIs received from client... continuing");
3911 *sock_error = 1;
3912 ret = LTTNG_ERR_SESSION_FAIL;
3913 free(uris);
3914 goto error;
3915 }
3916
3917 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3918 DBG("Creating session with ONE network URI is a bad call");
3919 ret = LTTNG_ERR_SESSION_FAIL;
3920 free(uris);
3921 goto error;
3922 }
3923 }
3924
3925 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3926 nb_uri, &cmd_ctx->creds);
3927 free(uris);
3928 break;
3929 }
3930 case LTTNG_CREATE_SESSION_LIVE:
3931 {
3932 size_t nb_uri, len;
3933 struct lttng_uri *uris = NULL;
3934
3935 nb_uri = cmd_ctx->lsm->u.uri.size;
3936 len = nb_uri * sizeof(struct lttng_uri);
3937
3938 if (nb_uri > 0) {
3939 uris = zmalloc(len);
3940 if (uris == NULL) {
3941 ret = LTTNG_ERR_FATAL;
3942 goto error;
3943 }
3944
3945 /* Receive variable len data */
3946 DBG("Waiting for %zu URIs from client ...", nb_uri);
3947 ret = lttcomm_recv_unix_sock(sock, uris, len);
3948 if (ret <= 0) {
3949 DBG("No URIs received from client... continuing");
3950 *sock_error = 1;
3951 ret = LTTNG_ERR_SESSION_FAIL;
3952 free(uris);
3953 goto error;
3954 }
3955
3956 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3957 DBG("Creating session with ONE network URI is a bad call");
3958 ret = LTTNG_ERR_SESSION_FAIL;
3959 free(uris);
3960 goto error;
3961 }
3962 }
3963
3964 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3965 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3966 free(uris);
3967 break;
3968 }
3969 case LTTNG_SAVE_SESSION:
3970 {
3971 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3972 &cmd_ctx->creds);
3973 break;
3974 }
3975 case LTTNG_SET_SESSION_SHM_PATH:
3976 {
3977 ret = cmd_set_session_shm_path(cmd_ctx->session,
3978 cmd_ctx->lsm->u.set_shm_path.shm_path);
3979 break;
3980 }
3981 default:
3982 ret = LTTNG_ERR_UND;
3983 break;
3984 }
3985
3986 error:
3987 if (cmd_ctx->llm == NULL) {
3988 DBG("Missing llm structure. Allocating one.");
3989 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3990 goto setup_error;
3991 }
3992 }
3993 /* Set return code */
3994 cmd_ctx->llm->ret_code = ret;
3995 setup_error:
3996 if (cmd_ctx->session) {
3997 session_unlock(cmd_ctx->session);
3998 }
3999 if (need_tracing_session) {
4000 session_unlock_list();
4001 }
4002 init_setup_error:
4003 return ret;
4004 }
4005
4006 /*
4007 * Thread managing health check socket.
4008 */
4009 static void *thread_manage_health(void *data)
4010 {
4011 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4012 uint32_t revents, nb_fd;
4013 struct lttng_poll_event events;
4014 struct health_comm_msg msg;
4015 struct health_comm_reply reply;
4016
4017 DBG("[thread] Manage health check started");
4018
4019 rcu_register_thread();
4020
4021 /* We might hit an error path before this is created. */
4022 lttng_poll_init(&events);
4023
4024 /* Create unix socket */
4025 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4026 if (sock < 0) {
4027 ERR("Unable to create health check Unix socket");
4028 ret = -1;
4029 goto error;
4030 }
4031
4032 if (is_root) {
4033 /* lttng health client socket path permissions */
4034 ret = chown(health_unix_sock_path, 0,
4035 utils_get_group_id(tracing_group_name));
4036 if (ret < 0) {
4037 ERR("Unable to set group on %s", health_unix_sock_path);
4038 PERROR("chown");
4039 ret = -1;
4040 goto error;
4041 }
4042
4043 ret = chmod(health_unix_sock_path,
4044 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4045 if (ret < 0) {
4046 ERR("Unable to set permissions on %s", health_unix_sock_path);
4047 PERROR("chmod");
4048 ret = -1;
4049 goto error;
4050 }
4051 }
4052
4053 /*
4054 * Set the CLOEXEC flag. Return code is useless because either way, the
4055 * show must go on.
4056 */
4057 (void) utils_set_fd_cloexec(sock);
4058
4059 ret = lttcomm_listen_unix_sock(sock);
4060 if (ret < 0) {
4061 goto error;
4062 }
4063
4064 /*
4065 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4066 * more will be added to this poll set.
4067 */
4068 ret = sessiond_set_thread_pollset(&events, 2);
4069 if (ret < 0) {
4070 goto error;
4071 }
4072
4073 /* Add the application registration socket */
4074 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4075 if (ret < 0) {
4076 goto error;
4077 }
4078
4079 sessiond_notify_ready();
4080
4081 while (1) {
4082 DBG("Health check ready");
4083
4084 /* Inifinite blocking call, waiting for transmission */
4085 restart:
4086 ret = lttng_poll_wait(&events, -1);
4087 if (ret < 0) {
4088 /*
4089 * Restart interrupted system call.
4090 */
4091 if (errno == EINTR) {
4092 goto restart;
4093 }
4094 goto error;
4095 }
4096
4097 nb_fd = ret;
4098
4099 for (i = 0; i < nb_fd; i++) {
4100 /* Fetch once the poll data */
4101 revents = LTTNG_POLL_GETEV(&events, i);
4102 pollfd = LTTNG_POLL_GETFD(&events, i);
4103
4104 if (!revents) {
4105 /* No activity for this FD (poll implementation). */
4106 continue;
4107 }
4108
4109 /* Thread quit pipe has been closed. Killing thread. */
4110 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4111 if (ret) {
4112 err = 0;
4113 goto exit;
4114 }
4115
4116 /* Event on the registration socket */
4117 if (pollfd == sock) {
4118 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4119 ERR("Health socket poll error");
4120 goto error;
4121 }
4122 }
4123 }
4124
4125 new_sock = lttcomm_accept_unix_sock(sock);
4126 if (new_sock < 0) {
4127 goto error;
4128 }
4129
4130 /*
4131 * Set the CLOEXEC flag. Return code is useless because either way, the
4132 * show must go on.
4133 */
4134 (void) utils_set_fd_cloexec(new_sock);
4135
4136 DBG("Receiving data from client for health...");
4137 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4138 if (ret <= 0) {
4139 DBG("Nothing recv() from client... continuing");
4140 ret = close(new_sock);
4141 if (ret) {
4142 PERROR("close");
4143 }
4144 new_sock = -1;
4145 continue;
4146 }
4147
4148 rcu_thread_online();
4149
4150 memset(&reply, 0, sizeof(reply));
4151 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4152 /*
4153 * health_check_state returns 0 if health is
4154 * bad.
4155 */
4156 if (!health_check_state(health_sessiond, i)) {
4157 reply.ret_code |= 1ULL << i;
4158 }
4159 }
4160
4161 DBG2("Health check return value %" PRIx64, reply.ret_code);
4162
4163 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4164 if (ret < 0) {
4165 ERR("Failed to send health data back to client");
4166 }
4167
4168 /* End of transmission */
4169 ret = close(new_sock);
4170 if (ret) {
4171 PERROR("close");
4172 }
4173 new_sock = -1;
4174 }
4175
4176 exit:
4177 error:
4178 if (err) {
4179 ERR("Health error occurred in %s", __func__);
4180 }
4181 DBG("Health check thread dying");
4182 unlink(health_unix_sock_path);
4183 if (sock >= 0) {
4184 ret = close(sock);
4185 if (ret) {
4186 PERROR("close");
4187 }
4188 }
4189
4190 lttng_poll_clean(&events);
4191
4192 rcu_unregister_thread();
4193 return NULL;
4194 }
4195
4196 /*
4197 * This thread manage all clients request using the unix client socket for
4198 * communication.
4199 */
4200 static void *thread_manage_clients(void *data)
4201 {
4202 int sock = -1, ret, i, pollfd, err = -1;
4203 int sock_error;
4204 uint32_t revents, nb_fd;
4205 struct command_ctx *cmd_ctx = NULL;
4206 struct lttng_poll_event events;
4207
4208 DBG("[thread] Manage client started");
4209
4210 rcu_register_thread();
4211
4212 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4213
4214 health_code_update();
4215
4216 ret = lttcomm_listen_unix_sock(client_sock);
4217 if (ret < 0) {
4218 goto error_listen;
4219 }
4220
4221 /*
4222 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4223 * more will be added to this poll set.
4224 */
4225 ret = sessiond_set_thread_pollset(&events, 2);
4226 if (ret < 0) {
4227 goto error_create_poll;
4228 }
4229
4230 /* Add the application registration socket */
4231 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4232 if (ret < 0) {
4233 goto error;
4234 }
4235
4236 sessiond_notify_ready();
4237 ret = sem_post(&load_info->message_thread_ready);
4238 if (ret) {
4239 PERROR("sem_post message_thread_ready");
4240 goto error;
4241 }
4242
4243 /* This testpoint is after we signal readiness to the parent. */
4244 if (testpoint(sessiond_thread_manage_clients)) {
4245 goto error;
4246 }
4247
4248 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4249 goto error;
4250 }
4251
4252 health_code_update();
4253
4254 while (1) {
4255 DBG("Accepting client command ...");
4256
4257 /* Inifinite blocking call, waiting for transmission */
4258 restart:
4259 health_poll_entry();
4260 ret = lttng_poll_wait(&events, -1);
4261 health_poll_exit();
4262 if (ret < 0) {
4263 /*
4264 * Restart interrupted system call.
4265 */
4266 if (errno == EINTR) {
4267 goto restart;
4268 }
4269 goto error;
4270 }
4271
4272 nb_fd = ret;
4273
4274 for (i = 0; i < nb_fd; i++) {
4275 /* Fetch once the poll data */
4276 revents = LTTNG_POLL_GETEV(&events, i);
4277 pollfd = LTTNG_POLL_GETFD(&events, i);
4278
4279 health_code_update();
4280
4281 if (!revents) {
4282 /* No activity for this FD (poll implementation). */
4283 continue;
4284 }
4285
4286 /* Thread quit pipe has been closed. Killing thread. */
4287 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4288 if (ret) {
4289 err = 0;
4290 goto exit;
4291 }
4292
4293 /* Event on the registration socket */
4294 if (pollfd == client_sock) {
4295 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4296 ERR("Client socket poll error");
4297 goto error;
4298 }
4299 }
4300 }
4301
4302 DBG("Wait for client response");
4303
4304 health_code_update();
4305
4306 sock = lttcomm_accept_unix_sock(client_sock);
4307 if (sock < 0) {
4308 goto error;
4309 }
4310
4311 /*
4312 * Set the CLOEXEC flag. Return code is useless because either way, the
4313 * show must go on.
4314 */
4315 (void) utils_set_fd_cloexec(sock);
4316
4317 /* Set socket option for credentials retrieval */
4318 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4319 if (ret < 0) {
4320 goto error;
4321 }
4322
4323 /* Allocate context command to process the client request */
4324 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4325 if (cmd_ctx == NULL) {
4326 PERROR("zmalloc cmd_ctx");
4327 goto error;
4328 }
4329
4330 /* Allocate data buffer for reception */
4331 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4332 if (cmd_ctx->lsm == NULL) {
4333 PERROR("zmalloc cmd_ctx->lsm");
4334 goto error;
4335 }
4336
4337 cmd_ctx->llm = NULL;
4338 cmd_ctx->session = NULL;
4339
4340 health_code_update();
4341
4342 /*
4343 * Data is received from the lttng client. The struct
4344 * lttcomm_session_msg (lsm) contains the command and data request of
4345 * the client.
4346 */
4347 DBG("Receiving data from client ...");
4348 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4349 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4350 if (ret <= 0) {
4351 DBG("Nothing recv() from client... continuing");
4352 ret = close(sock);
4353 if (ret) {
4354 PERROR("close");
4355 }
4356 sock = -1;
4357 clean_command_ctx(&cmd_ctx);
4358 continue;
4359 }
4360
4361 health_code_update();
4362
4363 // TODO: Validate cmd_ctx including sanity check for
4364 // security purpose.
4365
4366 rcu_thread_online();
4367 /*
4368 * This function dispatch the work to the kernel or userspace tracer
4369 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4370 * informations for the client. The command context struct contains
4371 * everything this function may needs.
4372 */
4373 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4374 rcu_thread_offline();
4375 if (ret < 0) {
4376 ret = close(sock);
4377 if (ret) {
4378 PERROR("close");
4379 }
4380 sock = -1;
4381 /*
4382 * TODO: Inform client somehow of the fatal error. At
4383 * this point, ret < 0 means that a zmalloc failed
4384 * (ENOMEM). Error detected but still accept
4385 * command, unless a socket error has been
4386 * detected.
4387 */
4388 clean_command_ctx(&cmd_ctx);
4389 continue;
4390 }
4391
4392 health_code_update();
4393
4394 DBG("Sending response (size: %d, retcode: %s (%d))",
4395 cmd_ctx->lttng_msg_size,
4396 lttng_strerror(-cmd_ctx->llm->ret_code),
4397 cmd_ctx->llm->ret_code);
4398 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4399 if (ret < 0) {
4400 ERR("Failed to send data back to client");
4401 }
4402
4403 /* End of transmission */
4404 ret = close(sock);
4405 if (ret) {
4406 PERROR("close");
4407 }
4408 sock = -1;
4409
4410 clean_command_ctx(&cmd_ctx);
4411
4412 health_code_update();
4413 }
4414
4415 exit:
4416 error:
4417 if (sock >= 0) {
4418 ret = close(sock);
4419 if (ret) {
4420 PERROR("close");
4421 }
4422 }
4423
4424 lttng_poll_clean(&events);
4425 clean_command_ctx(&cmd_ctx);
4426
4427 error_listen:
4428 error_create_poll:
4429 unlink(client_unix_sock_path);
4430 if (client_sock >= 0) {
4431 ret = close(client_sock);
4432 if (ret) {
4433 PERROR("close");
4434 }
4435 }
4436
4437 if (err) {
4438 health_error();
4439 ERR("Health error occurred in %s", __func__);
4440 }
4441
4442 health_unregister(health_sessiond);
4443
4444 DBG("Client thread dying");
4445
4446 rcu_unregister_thread();
4447
4448 /*
4449 * Since we are creating the consumer threads, we own them, so we need
4450 * to join them before our thread exits.
4451 */
4452 ret = join_consumer_thread(&kconsumer_data);
4453 if (ret) {
4454 errno = ret;
4455 PERROR("join_consumer");
4456 }
4457
4458 ret = join_consumer_thread(&ustconsumer32_data);
4459 if (ret) {
4460 errno = ret;
4461 PERROR("join_consumer ust32");
4462 }
4463
4464 ret = join_consumer_thread(&ustconsumer64_data);
4465 if (ret) {
4466 errno = ret;
4467 PERROR("join_consumer ust64");
4468 }
4469 return NULL;
4470 }
4471
4472
4473 /*
4474 * usage function on stderr
4475 */
4476 static void usage(void)
4477 {
4478 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4479 fprintf(stderr, " -h, --help Display this usage.\n");
4480 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4481 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4482 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4483 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4484 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4485 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4486 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4487 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4488 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4489 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4490 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4491 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4492 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4493 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4494 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4495 fprintf(stderr, " -V, --version Show version number.\n");
4496 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4497 fprintf(stderr, " -q, --quiet No output at all.\n");
4498 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4499 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4500 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4501 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4502 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4503 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4504 fprintf(stderr, " -l --load PATH Load session configuration\n");
4505 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4506 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4507 }
4508
4509 static int string_match(const char *str1, const char *str2)
4510 {
4511 return (str1 && str2) && !strcmp(str1, str2);
4512 }
4513
4514 /*
4515 * Take an option from the getopt output and set it in the right variable to be
4516 * used later.
4517 *
4518 * Return 0 on success else a negative value.
4519 */
4520 static int set_option(int opt, const char *arg, const char *optname)
4521 {
4522 int ret = 0;
4523
4524 if (arg && arg[0] == '\0') {
4525 /*
4526 * This only happens if the value is read from daemon config
4527 * file. This means the option requires an argument and the
4528 * configuration file contains a line such as:
4529 * my_option =
4530 */
4531 ret = -EINVAL;
4532 goto end;
4533 }
4534
4535 if (string_match(optname, "client-sock") || opt == 'c') {
4536 if (lttng_is_setuid_setgid()) {
4537 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4538 "-c, --client-sock");
4539 } else {
4540 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4541 }
4542 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4543 if (lttng_is_setuid_setgid()) {
4544 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4545 "-a, --apps-sock");
4546 } else {
4547 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4548 }
4549 } else if (string_match(optname, "daemonize") || opt == 'd') {
4550 opt_daemon = 1;
4551 } else if (string_match(optname, "background") || opt == 'b') {
4552 opt_background = 1;
4553 } else if (string_match(optname, "group") || opt == 'g') {
4554 if (lttng_is_setuid_setgid()) {
4555 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4556 "-g, --group");
4557 } else {
4558 /*
4559 * If the override option is set, the pointer points to a
4560 * *non* const thus freeing it even though the variable type is
4561 * set to const.
4562 */
4563 if (tracing_group_name_override) {
4564 free((void *) tracing_group_name);
4565 }
4566 tracing_group_name = strdup(arg);
4567 if (!tracing_group_name) {
4568 PERROR("strdup");
4569 ret = -ENOMEM;
4570 }
4571 tracing_group_name_override = 1;
4572 }
4573 } else if (string_match(optname, "help") || opt == 'h') {
4574 usage();
4575 exit(EXIT_SUCCESS);
4576 } else if (string_match(optname, "version") || opt == 'V') {
4577 fprintf(stdout, "%s\n", VERSION);
4578 exit(EXIT_SUCCESS);
4579 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4580 opt_sig_parent = 1;
4581 } else if (string_match(optname, "kconsumerd-err-sock")) {
4582 if (lttng_is_setuid_setgid()) {
4583 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4584 "--kconsumerd-err-sock");
4585 } else {
4586 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4587 }
4588 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4589 if (lttng_is_setuid_setgid()) {
4590 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4591 "--kconsumerd-cmd-sock");
4592 } else {
4593 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4594 }
4595 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4596 if (lttng_is_setuid_setgid()) {
4597 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4598 "--ustconsumerd64-err-sock");
4599 } else {
4600 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4601 }
4602 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4603 if (lttng_is_setuid_setgid()) {
4604 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4605 "--ustconsumerd64-cmd-sock");
4606 } else {
4607 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4608 }
4609 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4610 if (lttng_is_setuid_setgid()) {
4611 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4612 "--ustconsumerd32-err-sock");
4613 } else {
4614 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4615 }
4616 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4617 if (lttng_is_setuid_setgid()) {
4618 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4619 "--ustconsumerd32-cmd-sock");
4620 } else {
4621 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4622 }
4623 } else if (string_match(optname, "no-kernel")) {
4624 opt_no_kernel = 1;
4625 } else if (string_match(optname, "quiet") || opt == 'q') {
4626 lttng_opt_quiet = 1;
4627 } else if (string_match(optname, "verbose") || opt == 'v') {
4628 /* Verbose level can increase using multiple -v */
4629 if (arg) {
4630 /* Value obtained from config file */
4631 lttng_opt_verbose = config_parse_value(arg);
4632 } else {
4633 /* -v used on command line */
4634 lttng_opt_verbose++;
4635 }
4636 /* Clamp value to [0, 3] */
4637 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4638 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4639 } else if (string_match(optname, "verbose-consumer")) {
4640 if (arg) {
4641 opt_verbose_consumer = config_parse_value(arg);
4642 } else {
4643 opt_verbose_consumer += 1;
4644 }
4645 } else if (string_match(optname, "consumerd32-path")) {
4646 if (lttng_is_setuid_setgid()) {
4647 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4648 "--consumerd32-path");
4649 } else {
4650 if (consumerd32_bin_override) {
4651 free((void *) consumerd32_bin);
4652 }
4653 consumerd32_bin = strdup(arg);
4654 if (!consumerd32_bin) {
4655 PERROR("strdup");
4656 ret = -ENOMEM;
4657 }
4658 consumerd32_bin_override = 1;
4659 }
4660 } else if (string_match(optname, "consumerd32-libdir")) {
4661 if (lttng_is_setuid_setgid()) {
4662 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4663 "--consumerd32-libdir");
4664 } else {
4665 if (consumerd32_libdir_override) {
4666 free((void *) consumerd32_libdir);
4667 }
4668 consumerd32_libdir = strdup(arg);
4669 if (!consumerd32_libdir) {
4670 PERROR("strdup");
4671 ret = -ENOMEM;
4672 }
4673 consumerd32_libdir_override = 1;
4674 }
4675 } else if (string_match(optname, "consumerd64-path")) {
4676 if (lttng_is_setuid_setgid()) {
4677 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4678 "--consumerd64-path");
4679 } else {
4680 if (consumerd64_bin_override) {
4681 free((void *) consumerd64_bin);
4682 }
4683 consumerd64_bin = strdup(arg);
4684 if (!consumerd64_bin) {
4685 PERROR("strdup");
4686 ret = -ENOMEM;
4687 }
4688 consumerd64_bin_override = 1;
4689 }
4690 } else if (string_match(optname, "consumerd64-libdir")) {
4691 if (lttng_is_setuid_setgid()) {
4692 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4693 "--consumerd64-libdir");
4694 } else {
4695 if (consumerd64_libdir_override) {
4696 free((void *) consumerd64_libdir);
4697 }
4698 consumerd64_libdir = strdup(arg);
4699 if (!consumerd64_libdir) {
4700 PERROR("strdup");
4701 ret = -ENOMEM;
4702 }
4703 consumerd64_libdir_override = 1;
4704 }
4705 } else if (string_match(optname, "pidfile") || opt == 'p') {
4706 if (lttng_is_setuid_setgid()) {
4707 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4708 "-p, --pidfile");
4709 } else {
4710 free(opt_pidfile);
4711 opt_pidfile = strdup(arg);
4712 if (!opt_pidfile) {
4713 PERROR("strdup");
4714 ret = -ENOMEM;
4715 }
4716 }
4717 } else if (string_match(optname, "agent-tcp-port")) {
4718 if (lttng_is_setuid_setgid()) {
4719 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4720 "--agent-tcp-port");
4721 } else {
4722 unsigned long v;
4723
4724 if (!arg) {
4725 ret = -EINVAL;
4726 goto end;
4727 }
4728 errno = 0;
4729 v = strtoul(arg, NULL, 0);
4730 if (errno != 0 || !isdigit(arg[0])) {
4731 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4732 return -1;
4733 }
4734 if (v == 0 || v >= 65535) {
4735 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4736 return -1;
4737 }
4738 agent_tcp_port = (uint32_t) v;
4739 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4740 }
4741 } else if (string_match(optname, "load") || opt == 'l') {
4742 if (lttng_is_setuid_setgid()) {
4743 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4744 "-l, --load");
4745 } else {
4746 free(opt_load_session_path);
4747 opt_load_session_path = strdup(arg);
4748 if (!opt_load_session_path) {
4749 PERROR("strdup");
4750 ret = -ENOMEM;
4751 }
4752 }
4753 } else if (string_match(optname, "kmod-probes")) {
4754 if (lttng_is_setuid_setgid()) {
4755 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4756 "--kmod-probes");
4757 } else {
4758 free(kmod_probes_list);
4759 kmod_probes_list = strdup(arg);
4760 if (!kmod_probes_list) {
4761 PERROR("strdup");
4762 ret = -ENOMEM;
4763 }
4764 }
4765 } else if (string_match(optname, "extra-kmod-probes")) {
4766 if (lttng_is_setuid_setgid()) {
4767 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4768 "--extra-kmod-probes");
4769 } else {
4770 free(kmod_extra_probes_list);
4771 kmod_extra_probes_list = strdup(arg);
4772 if (!kmod_extra_probes_list) {
4773 PERROR("strdup");
4774 ret = -ENOMEM;
4775 }
4776 }
4777 } else if (string_match(optname, "config") || opt == 'f') {
4778 /* This is handled in set_options() thus silent skip. */
4779 goto end;
4780 } else {
4781 /* Unknown option or other error.
4782 * Error is printed by getopt, just return */
4783 ret = -1;
4784 }
4785
4786 end:
4787 if (ret == -EINVAL) {
4788 const char *opt_name = "unknown";
4789 int i;
4790
4791 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4792 i++) {
4793 if (opt == long_options[i].val) {
4794 opt_name = long_options[i].name;
4795 break;
4796 }
4797 }
4798
4799 WARN("Invalid argument provided for option \"%s\", using default value.",
4800 opt_name);
4801 }
4802
4803 return ret;
4804 }
4805
4806 /*
4807 * config_entry_handler_cb used to handle options read from a config file.
4808 * See config_entry_handler_cb comment in common/config/config.h for the
4809 * return value conventions.
4810 */
4811 static int config_entry_handler(const struct config_entry *entry, void *unused)
4812 {
4813 int ret = 0, i;
4814
4815 if (!entry || !entry->name || !entry->value) {
4816 ret = -EINVAL;
4817 goto end;
4818 }
4819
4820 /* Check if the option is to be ignored */
4821 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4822 if (!strcmp(entry->name, config_ignore_options[i])) {
4823 goto end;
4824 }
4825 }
4826
4827 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4828 i++) {
4829
4830 /* Ignore if not fully matched. */
4831 if (strcmp(entry->name, long_options[i].name)) {
4832 continue;
4833 }
4834
4835 /*
4836 * If the option takes no argument on the command line, we have to
4837 * check if the value is "true". We support non-zero numeric values,
4838 * true, on and yes.
4839 */
4840 if (!long_options[i].has_arg) {
4841 ret = config_parse_value(entry->value);
4842 if (ret <= 0) {
4843 if (ret) {
4844 WARN("Invalid configuration value \"%s\" for option %s",
4845 entry->value, entry->name);
4846 }
4847 /* False, skip boolean config option. */
4848 goto end;
4849 }
4850 }
4851
4852 ret = set_option(long_options[i].val, entry->value, entry->name);
4853 goto end;
4854 }
4855
4856 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4857
4858 end:
4859 return ret;
4860 }
4861
4862 /*
4863 * daemon configuration loading and argument parsing
4864 */
4865 static int set_options(int argc, char **argv)
4866 {
4867 int ret = 0, c = 0, option_index = 0;
4868 int orig_optopt = optopt, orig_optind = optind;
4869 char *optstring;
4870 const char *config_path = NULL;
4871
4872 optstring = utils_generate_optstring(long_options,
4873 sizeof(long_options) / sizeof(struct option));
4874 if (!optstring) {
4875 ret = -ENOMEM;
4876 goto end;
4877 }
4878
4879 /* Check for the --config option */
4880 while ((c = getopt_long(argc, argv, optstring, long_options,
4881 &option_index)) != -1) {
4882 if (c == '?') {
4883 ret = -EINVAL;
4884 goto end;
4885 } else if (c != 'f') {
4886 /* if not equal to --config option. */
4887 continue;
4888 }
4889
4890 if (lttng_is_setuid_setgid()) {
4891 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4892 "-f, --config");
4893 } else {
4894 config_path = utils_expand_path(optarg);
4895 if (!config_path) {
4896 ERR("Failed to resolve path: %s", optarg);
4897 }
4898 }
4899 }
4900
4901 ret = config_get_section_entries(config_path, config_section_name,
4902 config_entry_handler, NULL);
4903 if (ret) {
4904 if (ret > 0) {
4905 ERR("Invalid configuration option at line %i", ret);
4906 ret = -1;
4907 }
4908 goto end;
4909 }
4910
4911 /* Reset getopt's global state */
4912 optopt = orig_optopt;
4913 optind = orig_optind;
4914 while (1) {
4915 option_index = -1;
4916 /*
4917 * getopt_long() will not set option_index if it encounters a
4918 * short option.
4919 */
4920 c = getopt_long(argc, argv, optstring, long_options,
4921 &option_index);
4922 if (c == -1) {
4923 break;
4924 }
4925
4926 /*
4927 * Pass NULL as the long option name if popt left the index
4928 * unset.
4929 */
4930 ret = set_option(c, optarg,
4931 option_index < 0 ? NULL :
4932 long_options[option_index].name);
4933 if (ret < 0) {
4934 break;
4935 }
4936 }
4937
4938 end:
4939 free(optstring);
4940 return ret;
4941 }
4942
4943 /*
4944 * Creates the two needed socket by the daemon.
4945 * apps_sock - The communication socket for all UST apps.
4946 * client_sock - The communication of the cli tool (lttng).
4947 */
4948 static int init_daemon_socket(void)
4949 {
4950 int ret = 0;
4951 mode_t old_umask;
4952
4953 old_umask = umask(0);
4954
4955 /* Create client tool unix socket */
4956 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4957 if (client_sock < 0) {
4958 ERR("Create unix sock failed: %s", client_unix_sock_path);
4959 ret = -1;
4960 goto end;
4961 }
4962
4963 /* Set the cloexec flag */
4964 ret = utils_set_fd_cloexec(client_sock);
4965 if (ret < 0) {
4966 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4967 "Continuing but note that the consumer daemon will have a "
4968 "reference to this socket on exec()", client_sock);
4969 }
4970
4971 /* File permission MUST be 660 */
4972 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4973 if (ret < 0) {
4974 ERR("Set file permissions failed: %s", client_unix_sock_path);
4975 PERROR("chmod");
4976 goto end;
4977 }
4978
4979 /* Create the application unix socket */
4980 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4981 if (apps_sock < 0) {
4982 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4983 ret = -1;
4984 goto end;
4985 }
4986
4987 /* Set the cloexec flag */
4988 ret = utils_set_fd_cloexec(apps_sock);
4989 if (ret < 0) {
4990 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4991 "Continuing but note that the consumer daemon will have a "
4992 "reference to this socket on exec()", apps_sock);
4993 }
4994
4995 /* File permission MUST be 666 */
4996 ret = chmod(apps_unix_sock_path,
4997 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4998 if (ret < 0) {
4999 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5000 PERROR("chmod");
5001 goto end;
5002 }
5003
5004 DBG3("Session daemon client socket %d and application socket %d created",
5005 client_sock, apps_sock);
5006
5007 end:
5008 umask(old_umask);
5009 return ret;
5010 }
5011
5012 /*
5013 * Check if the global socket is available, and if a daemon is answering at the
5014 * other side. If yes, error is returned.
5015 */
5016 static int check_existing_daemon(void)
5017 {
5018 /* Is there anybody out there ? */
5019 if (lttng_session_daemon_alive()) {
5020 return -EEXIST;
5021 }
5022
5023 return 0;
5024 }
5025
5026 /*
5027 * Set the tracing group gid onto the client socket.
5028 *
5029 * Race window between mkdir and chown is OK because we are going from more
5030 * permissive (root.root) to less permissive (root.tracing).
5031 */
5032 static int set_permissions(char *rundir)
5033 {
5034 int ret;
5035 gid_t gid;
5036
5037 gid = utils_get_group_id(tracing_group_name);
5038
5039 /* Set lttng run dir */
5040 ret = chown(rundir, 0, gid);
5041 if (ret < 0) {
5042 ERR("Unable to set group on %s", rundir);
5043 PERROR("chown");
5044 }
5045
5046 /*
5047 * Ensure all applications and tracing group can search the run
5048 * dir. Allow everyone to read the directory, since it does not
5049 * buy us anything to hide its content.
5050 */
5051 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5052 if (ret < 0) {
5053 ERR("Unable to set permissions on %s", rundir);
5054 PERROR("chmod");
5055 }
5056
5057 /* lttng client socket path */
5058 ret = chown(client_unix_sock_path, 0, gid);
5059 if (ret < 0) {
5060 ERR("Unable to set group on %s", client_unix_sock_path);
5061 PERROR("chown");
5062 }
5063
5064 /* kconsumer error socket path */
5065 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5066 if (ret < 0) {
5067 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5068 PERROR("chown");
5069 }
5070
5071 /* 64-bit ustconsumer error socket path */
5072 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5073 if (ret < 0) {
5074 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5075 PERROR("chown");
5076 }
5077
5078 /* 32-bit ustconsumer compat32 error socket path */
5079 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5080 if (ret < 0) {
5081 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5082 PERROR("chown");
5083 }
5084
5085 DBG("All permissions are set");
5086
5087 return ret;
5088 }
5089
5090 /*
5091 * Create the lttng run directory needed for all global sockets and pipe.
5092 */
5093 static int create_lttng_rundir(const char *rundir)
5094 {
5095 int ret;
5096
5097 DBG3("Creating LTTng run directory: %s", rundir);
5098
5099 ret = mkdir(rundir, S_IRWXU);
5100 if (ret < 0) {
5101 if (errno != EEXIST) {
5102 ERR("Unable to create %s", rundir);
5103 goto error;
5104 } else {
5105 ret = 0;
5106 }
5107 }
5108
5109 error:
5110 return ret;
5111 }
5112
5113 /*
5114 * Setup sockets and directory needed by the kconsumerd communication with the
5115 * session daemon.
5116 */
5117 static int set_consumer_sockets(struct consumer_data *consumer_data,
5118 const char *rundir)
5119 {
5120 int ret;
5121 char path[PATH_MAX];
5122
5123 switch (consumer_data->type) {
5124 case LTTNG_CONSUMER_KERNEL:
5125 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5126 break;
5127 case LTTNG_CONSUMER64_UST:
5128 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5129 break;
5130 case LTTNG_CONSUMER32_UST:
5131 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5132 break;
5133 default:
5134 ERR("Consumer type unknown");
5135 ret = -EINVAL;
5136 goto error;
5137 }
5138
5139 DBG2("Creating consumer directory: %s", path);
5140
5141 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5142 if (ret < 0) {
5143 if (errno != EEXIST) {
5144 PERROR("mkdir");
5145 ERR("Failed to create %s", path);
5146 goto error;
5147 }
5148 ret = -1;
5149 }
5150 if (is_root) {
5151 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5152 if (ret < 0) {
5153 ERR("Unable to set group on %s", path);
5154 PERROR("chown");
5155 goto error;
5156 }
5157 }
5158
5159 /* Create the kconsumerd error unix socket */
5160 consumer_data->err_sock =
5161 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5162 if (consumer_data->err_sock < 0) {
5163 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5164 ret = -1;
5165 goto error;
5166 }
5167
5168 /*
5169 * Set the CLOEXEC flag. Return code is useless because either way, the
5170 * show must go on.
5171 */
5172 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5173 if (ret < 0) {
5174 PERROR("utils_set_fd_cloexec");
5175 /* continue anyway */
5176 }
5177
5178 /* File permission MUST be 660 */
5179 ret = chmod(consumer_data->err_unix_sock_path,
5180 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5181 if (ret < 0) {
5182 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5183 PERROR("chmod");
5184 goto error;
5185 }
5186
5187 error:
5188 return ret;
5189 }
5190
5191 /*
5192 * Signal handler for the daemon
5193 *
5194 * Simply stop all worker threads, leaving main() return gracefully after
5195 * joining all threads and calling cleanup().
5196 */
5197 static void sighandler(int sig)
5198 {
5199 switch (sig) {
5200 case SIGPIPE:
5201 DBG("SIGPIPE caught");
5202 return;
5203 case SIGINT:
5204 DBG("SIGINT caught");
5205 stop_threads();
5206 break;
5207 case SIGTERM:
5208 DBG("SIGTERM caught");
5209 stop_threads();
5210 break;
5211 case SIGUSR1:
5212 CMM_STORE_SHARED(recv_child_signal, 1);
5213 break;
5214 default:
5215 break;
5216 }
5217 }
5218
5219 /*
5220 * Setup signal handler for :
5221 * SIGINT, SIGTERM, SIGPIPE
5222 */
5223 static int set_signal_handler(void)
5224 {
5225 int ret = 0;
5226 struct sigaction sa;
5227 sigset_t sigset;
5228
5229 if ((ret = sigemptyset(&sigset)) < 0) {
5230 PERROR("sigemptyset");
5231 return ret;
5232 }
5233
5234 sa.sa_handler = sighandler;
5235 sa.sa_mask = sigset;
5236 sa.sa_flags = 0;
5237 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5238 PERROR("sigaction");
5239 return ret;
5240 }
5241
5242 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5243 PERROR("sigaction");
5244 return ret;
5245 }
5246
5247 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5248 PERROR("sigaction");
5249 return ret;
5250 }
5251
5252 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5253 PERROR("sigaction");
5254 return ret;
5255 }
5256
5257 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5258
5259 return ret;
5260 }
5261
5262 /*
5263 * Set open files limit to unlimited. This daemon can open a large number of
5264 * file descriptors in order to consumer multiple kernel traces.
5265 */
5266 static void set_ulimit(void)
5267 {
5268 int ret;
5269 struct rlimit lim;
5270
5271 /* The kernel does not allowed an infinite limit for open files */
5272 lim.rlim_cur = 65535;
5273 lim.rlim_max = 65535;
5274
5275 ret = setrlimit(RLIMIT_NOFILE, &lim);
5276 if (ret < 0) {
5277 PERROR("failed to set open files limit");
5278 }
5279 }
5280
5281 /*
5282 * Write pidfile using the rundir and opt_pidfile.
5283 */
5284 static int write_pidfile(void)
5285 {
5286 int ret;
5287 char pidfile_path[PATH_MAX];
5288
5289 assert(rundir);
5290
5291 if (opt_pidfile) {
5292 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5293 } else {
5294 /* Build pidfile path from rundir and opt_pidfile. */
5295 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5296 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5297 if (ret < 0) {
5298 PERROR("snprintf pidfile path");
5299 goto error;
5300 }
5301 }
5302
5303 /*
5304 * Create pid file in rundir.
5305 */
5306 ret = utils_create_pid_file(getpid(), pidfile_path);
5307 error:
5308 return ret;
5309 }
5310
5311 /*
5312 * Create lockfile using the rundir and return its fd.
5313 */
5314 static int create_lockfile(void)
5315 {
5316 int ret;
5317 char lockfile_path[PATH_MAX];
5318
5319 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5320 if (ret < 0) {
5321 goto error;
5322 }
5323
5324 ret = utils_create_lock_file(lockfile_path);
5325 error:
5326 return ret;
5327 }
5328
5329 /*
5330 * Write agent TCP port using the rundir.
5331 */
5332 static int write_agent_port(void)
5333 {
5334 int ret;
5335 char path[PATH_MAX];
5336
5337 assert(rundir);
5338
5339 ret = snprintf(path, sizeof(path), "%s/"
5340 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5341 if (ret < 0) {
5342 PERROR("snprintf agent port path");
5343 goto error;
5344 }
5345
5346 /*
5347 * Create TCP agent port file in rundir.
5348 */
5349 ret = utils_create_pid_file(agent_tcp_port, path);
5350
5351 error:
5352 return ret;
5353 }
5354
5355 /*
5356 * main
5357 */
5358 int main(int argc, char **argv)
5359 {
5360 int ret = 0, retval = 0;
5361 void *status;
5362 const char *home_path, *env_app_timeout;
5363
5364 init_kernel_workarounds();
5365
5366 rcu_register_thread();
5367
5368 if (set_signal_handler()) {
5369 retval = -1;
5370 goto exit_set_signal_handler;
5371 }
5372
5373 setup_consumerd_path();
5374
5375 page_size = sysconf(_SC_PAGESIZE);
5376 if (page_size < 0) {
5377 PERROR("sysconf _SC_PAGESIZE");
5378 page_size = LONG_MAX;
5379 WARN("Fallback page size to %ld", page_size);
5380 }
5381
5382 /*
5383 * Parse arguments and load the daemon configuration file.
5384 *
5385 * We have an exit_options exit path to free memory reserved by
5386 * set_options. This is needed because the rest of sessiond_cleanup()
5387 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5388 * depends on set_options.
5389 */
5390 progname = argv[0];
5391 if (set_options(argc, argv)) {
5392 retval = -1;
5393 goto exit_options;
5394 }
5395
5396 /* Daemonize */
5397 if (opt_daemon || opt_background) {
5398 int i;
5399
5400 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5401 !opt_background);
5402 if (ret < 0) {
5403 retval = -1;
5404 goto exit_options;
5405 }
5406
5407 /*
5408 * We are in the child. Make sure all other file descriptors are
5409 * closed, in case we are called with more opened file
5410 * descriptors than the standard ones.
5411 */
5412 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5413 (void) close(i);
5414 }
5415 }
5416
5417 /*
5418 * Starting from here, we can create threads. This needs to be after
5419 * lttng_daemonize due to RCU.
5420 */
5421
5422 /*
5423 * Initialize the health check subsystem. This call should set the
5424 * appropriate time values.
5425 */
5426 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5427 if (!health_sessiond) {
5428 PERROR("health_app_create error");
5429 retval = -1;
5430 goto exit_health_sessiond_cleanup;
5431 }
5432
5433 if (init_ht_cleanup_quit_pipe()) {
5434 retval = -1;
5435 goto exit_ht_cleanup_quit_pipe;
5436 }
5437
5438 /* Setup the thread ht_cleanup communication pipe. */
5439 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5440 retval = -1;
5441 goto exit_ht_cleanup_pipe;
5442 }
5443
5444 /* Set up max poll set size */
5445 if (lttng_poll_set_max_size()) {
5446 retval = -1;
5447 goto exit_set_max_size;
5448 }
5449
5450 /* Create thread to clean up RCU hash tables */
5451 ret = pthread_create(&ht_cleanup_thread, NULL,
5452 thread_ht_cleanup, (void *) NULL);
5453 if (ret) {
5454 errno = ret;
5455 PERROR("pthread_create ht_cleanup");
5456 retval = -1;
5457 goto exit_ht_cleanup;
5458 }
5459
5460 /* Create thread quit pipe */
5461 if (init_thread_quit_pipe()) {
5462 retval = -1;
5463 goto exit_init_data;
5464 }
5465
5466 /* Check if daemon is UID = 0 */
5467 is_root = !getuid();
5468
5469 if (is_root) {
5470 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5471 if (!rundir) {
5472 retval = -1;
5473 goto exit_init_data;
5474 }
5475
5476 /* Create global run dir with root access */
5477 if (create_lttng_rundir(rundir)) {
5478 retval = -1;
5479 goto exit_init_data;
5480 }
5481
5482 if (strlen(apps_unix_sock_path) == 0) {
5483 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5484 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5485 if (ret < 0) {
5486 retval = -1;
5487 goto exit_init_data;
5488 }
5489 }
5490
5491 if (strlen(client_unix_sock_path) == 0) {
5492 ret = snprintf(client_unix_sock_path, PATH_MAX,
5493 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5494 if (ret < 0) {
5495 retval = -1;
5496 goto exit_init_data;
5497 }
5498 }
5499
5500 /* Set global SHM for ust */
5501 if (strlen(wait_shm_path) == 0) {
5502 ret = snprintf(wait_shm_path, PATH_MAX,
5503 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5504 if (ret < 0) {
5505 retval = -1;
5506 goto exit_init_data;
5507 }
5508 }
5509
5510 if (strlen(health_unix_sock_path) == 0) {
5511 ret = snprintf(health_unix_sock_path,
5512 sizeof(health_unix_sock_path),
5513 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5514 if (ret < 0) {
5515 retval = -1;
5516 goto exit_init_data;
5517 }
5518 }
5519
5520 /* Setup kernel consumerd path */
5521 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5522 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5523 if (ret < 0) {
5524 retval = -1;
5525 goto exit_init_data;
5526 }
5527 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5528 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5529 if (ret < 0) {
5530 retval = -1;
5531 goto exit_init_data;
5532 }
5533
5534 DBG2("Kernel consumer err path: %s",
5535 kconsumer_data.err_unix_sock_path);
5536 DBG2("Kernel consumer cmd path: %s",
5537 kconsumer_data.cmd_unix_sock_path);
5538 } else {
5539 home_path = utils_get_home_dir();
5540 if (home_path == NULL) {
5541 /* TODO: Add --socket PATH option */
5542 ERR("Can't get HOME directory for sockets creation.");
5543 retval = -1;
5544 goto exit_init_data;
5545 }
5546
5547 /*
5548 * Create rundir from home path. This will create something like
5549 * $HOME/.lttng
5550 */
5551 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5552 if (ret < 0) {
5553 retval = -1;
5554 goto exit_init_data;
5555 }
5556
5557 if (create_lttng_rundir(rundir)) {
5558 retval = -1;
5559 goto exit_init_data;
5560 }
5561
5562 if (strlen(apps_unix_sock_path) == 0) {
5563 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5564 DEFAULT_HOME_APPS_UNIX_SOCK,
5565 home_path);
5566 if (ret < 0) {
5567 retval = -1;
5568 goto exit_init_data;
5569 }
5570 }
5571
5572 /* Set the cli tool unix socket path */
5573 if (strlen(client_unix_sock_path) == 0) {
5574 ret = snprintf(client_unix_sock_path, PATH_MAX,
5575 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5576 home_path);
5577 if (ret < 0) {
5578 retval = -1;
5579 goto exit_init_data;
5580 }
5581 }
5582
5583 /* Set global SHM for ust */
5584 if (strlen(wait_shm_path) == 0) {
5585 ret = snprintf(wait_shm_path, PATH_MAX,
5586 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5587 getuid());
5588 if (ret < 0) {
5589 retval = -1;
5590 goto exit_init_data;
5591 }
5592 }
5593
5594 /* Set health check Unix path */
5595 if (strlen(health_unix_sock_path) == 0) {
5596 ret = snprintf(health_unix_sock_path,
5597 sizeof(health_unix_sock_path),
5598 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5599 home_path);
5600 if (ret < 0) {
5601 retval = -1;
5602 goto exit_init_data;
5603 }
5604 }
5605 }
5606
5607 lockfile_fd = create_lockfile();
5608 if (lockfile_fd < 0) {
5609 retval = -1;
5610 goto exit_init_data;
5611 }
5612
5613 /* Set consumer initial state */
5614 kernel_consumerd_state = CONSUMER_STOPPED;
5615 ust_consumerd_state = CONSUMER_STOPPED;
5616
5617 DBG("Client socket path %s", client_unix_sock_path);
5618 DBG("Application socket path %s", apps_unix_sock_path);
5619 DBG("Application wait path %s", wait_shm_path);
5620 DBG("LTTng run directory path: %s", rundir);
5621
5622 /* 32 bits consumerd path setup */
5623 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5624 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5625 if (ret < 0) {
5626 PERROR("snprintf 32-bit consumer error socket path");
5627 retval = -1;
5628 goto exit_init_data;
5629 }
5630 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5631 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5632 if (ret < 0) {
5633 PERROR("snprintf 32-bit consumer command socket path");
5634 retval = -1;
5635 goto exit_init_data;
5636 }
5637
5638 DBG2("UST consumer 32 bits err path: %s",
5639 ustconsumer32_data.err_unix_sock_path);
5640 DBG2("UST consumer 32 bits cmd path: %s",
5641 ustconsumer32_data.cmd_unix_sock_path);
5642
5643 /* 64 bits consumerd path setup */
5644 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5645 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5646 if (ret < 0) {
5647 PERROR("snprintf 64-bit consumer error socket path");
5648 retval = -1;
5649 goto exit_init_data;
5650 }
5651 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5652 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5653 if (ret < 0) {
5654 PERROR("snprintf 64-bit consumer command socket path");
5655 retval = -1;
5656 goto exit_init_data;
5657 }
5658
5659 DBG2("UST consumer 64 bits err path: %s",
5660 ustconsumer64_data.err_unix_sock_path);
5661 DBG2("UST consumer 64 bits cmd path: %s",
5662 ustconsumer64_data.cmd_unix_sock_path);
5663
5664 /*
5665 * See if daemon already exist.
5666 */
5667 if (check_existing_daemon()) {
5668 ERR("Already running daemon.\n");
5669 /*
5670 * We do not goto exit because we must not cleanup()
5671 * because a daemon is already running.
5672 */
5673 retval = -1;
5674 goto exit_init_data;
5675 }
5676
5677 /*
5678 * Init UST app hash table. Alloc hash table before this point since
5679 * cleanup() can get called after that point.
5680 */
5681 if (ust_app_ht_alloc()) {
5682 ERR("Failed to allocate UST app hash table");
5683 retval = -1;
5684 goto exit_init_data;
5685 }
5686
5687 /*
5688 * Initialize agent app hash table. We allocate the hash table here
5689 * since cleanup() can get called after this point.
5690 */
5691 if (agent_app_ht_alloc()) {
5692 ERR("Failed to allocate Agent app hash table");
5693 retval = -1;
5694 goto exit_init_data;
5695 }
5696
5697 /*
5698 * These actions must be executed as root. We do that *after* setting up
5699 * the sockets path because we MUST make the check for another daemon using
5700 * those paths *before* trying to set the kernel consumer sockets and init
5701 * kernel tracer.
5702 */
5703 if (is_root) {
5704 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5705 retval = -1;
5706 goto exit_init_data;
5707 }
5708
5709 /* Setup kernel tracer */
5710 if (!opt_no_kernel) {
5711 init_kernel_tracer();
5712 if (kernel_tracer_fd >= 0) {
5713 ret = syscall_init_table();
5714 if (ret < 0) {
5715 ERR("Unable to populate syscall table. "
5716 "Syscall tracing won't work "
5717 "for this session daemon.");
5718 }
5719 }
5720 }
5721
5722 /* Set ulimit for open files */
5723 set_ulimit();
5724 }
5725 /* init lttng_fd tracking must be done after set_ulimit. */
5726 lttng_fd_init();
5727
5728 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5729 retval = -1;
5730 goto exit_init_data;
5731 }
5732
5733 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5734 retval = -1;
5735 goto exit_init_data;
5736 }
5737
5738 /* Setup the needed unix socket */
5739 if (init_daemon_socket()) {
5740 retval = -1;
5741 goto exit_init_data;
5742 }
5743
5744 /* Set credentials to socket */
5745 if (is_root && set_permissions(rundir)) {
5746 retval = -1;
5747 goto exit_init_data;
5748 }
5749
5750 /* Get parent pid if -S, --sig-parent is specified. */
5751 if (opt_sig_parent) {
5752 ppid = getppid();
5753 }
5754
5755 /* Setup the kernel pipe for waking up the kernel thread */
5756 if (is_root && !opt_no_kernel) {
5757 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5758 retval = -1;
5759 goto exit_init_data;
5760 }
5761 }
5762
5763 /* Setup the thread apps communication pipe. */
5764 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5765 retval = -1;
5766 goto exit_init_data;
5767 }
5768
5769 /* Setup the thread apps notify communication pipe. */
5770 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5771 retval = -1;
5772 goto exit_init_data;
5773 }
5774
5775 /* Initialize global buffer per UID and PID registry. */
5776 buffer_reg_init_uid_registry();
5777 buffer_reg_init_pid_registry();
5778
5779 /* Init UST command queue. */
5780 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5781
5782 /*
5783 * Get session list pointer. This pointer MUST NOT be free'd. This list
5784 * is statically declared in session.c
5785 */
5786 session_list_ptr = session_get_list();
5787
5788 cmd_init();
5789
5790 /* Check for the application socket timeout env variable. */
5791 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5792 if (env_app_timeout) {
5793 app_socket_timeout = atoi(env_app_timeout);
5794 } else {
5795 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5796 }
5797
5798 ret = write_pidfile();
5799 if (ret) {
5800 ERR("Error in write_pidfile");
5801 retval = -1;
5802 goto exit_init_data;
5803 }
5804 ret = write_agent_port();
5805 if (ret) {
5806 ERR("Error in write_agent_port");
5807 retval = -1;
5808 goto exit_init_data;
5809 }
5810
5811 /* Initialize communication library */
5812 lttcomm_init();
5813 /* Initialize TCP timeout values */
5814 lttcomm_inet_init();
5815
5816 if (load_session_init_data(&load_info) < 0) {
5817 retval = -1;
5818 goto exit_init_data;
5819 }
5820 load_info->path = opt_load_session_path;
5821
5822 /* Create health-check thread */
5823 ret = pthread_create(&health_thread, NULL,
5824 thread_manage_health, (void *) NULL);
5825 if (ret) {
5826 errno = ret;
5827 PERROR("pthread_create health");
5828 retval = -1;
5829 goto exit_health;
5830 }
5831
5832 /* Create thread to manage the client socket */
5833 ret = pthread_create(&client_thread, NULL,
5834 thread_manage_clients, (void *) NULL);
5835 if (ret) {
5836 errno = ret;
5837 PERROR("pthread_create clients");
5838 retval = -1;
5839 goto exit_client;
5840 }
5841
5842 /* Create thread to dispatch registration */
5843 ret = pthread_create(&dispatch_thread, NULL,
5844 thread_dispatch_ust_registration, (void *) NULL);
5845 if (ret) {
5846 errno = ret;
5847 PERROR("pthread_create dispatch");
5848 retval = -1;
5849 goto exit_dispatch;
5850 }
5851
5852 /* Create thread to manage application registration. */
5853 ret = pthread_create(&reg_apps_thread, NULL,
5854 thread_registration_apps, (void *) NULL);
5855 if (ret) {
5856 errno = ret;
5857 PERROR("pthread_create registration");
5858 retval = -1;
5859 goto exit_reg_apps;
5860 }
5861
5862 /* Create thread to manage application socket */
5863 ret = pthread_create(&apps_thread, NULL,
5864 thread_manage_apps, (void *) NULL);
5865 if (ret) {
5866 errno = ret;
5867 PERROR("pthread_create apps");
5868 retval = -1;
5869 goto exit_apps;
5870 }
5871
5872 /* Create thread to manage application notify socket */
5873 ret = pthread_create(&apps_notify_thread, NULL,
5874 ust_thread_manage_notify, (void *) NULL);
5875 if (ret) {
5876 errno = ret;
5877 PERROR("pthread_create notify");
5878 retval = -1;
5879 goto exit_apps_notify;
5880 }
5881
5882 /* Create agent registration thread. */
5883 ret = pthread_create(&agent_reg_thread, NULL,
5884 agent_thread_manage_registration, (void *) NULL);
5885 if (ret) {
5886 errno = ret;
5887 PERROR("pthread_create agent");
5888 retval = -1;
5889 goto exit_agent_reg;
5890 }
5891
5892 /* Don't start this thread if kernel tracing is not requested nor root */
5893 if (is_root && !opt_no_kernel) {
5894 /* Create kernel thread to manage kernel event */
5895 ret = pthread_create(&kernel_thread, NULL,
5896 thread_manage_kernel, (void *) NULL);
5897 if (ret) {
5898 errno = ret;
5899 PERROR("pthread_create kernel");
5900 retval = -1;
5901 goto exit_kernel;
5902 }
5903 }
5904
5905 /* Create session loading thread. */
5906 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5907 load_info);
5908 if (ret) {
5909 errno = ret;
5910 PERROR("pthread_create load_session_thread");
5911 retval = -1;
5912 goto exit_load_session;
5913 }
5914
5915 /*
5916 * This is where we start awaiting program completion (e.g. through
5917 * signal that asks threads to teardown).
5918 */
5919
5920 ret = pthread_join(load_session_thread, &status);
5921 if (ret) {
5922 errno = ret;
5923 PERROR("pthread_join load_session_thread");
5924 retval = -1;
5925 }
5926 exit_load_session:
5927
5928 if (is_root && !opt_no_kernel) {
5929 ret = pthread_join(kernel_thread, &status);
5930 if (ret) {
5931 errno = ret;
5932 PERROR("pthread_join");
5933 retval = -1;
5934 }
5935 }
5936 exit_kernel:
5937
5938 ret = pthread_join(agent_reg_thread, &status);
5939 if (ret) {
5940 errno = ret;
5941 PERROR("pthread_join agent");
5942 retval = -1;
5943 }
5944 exit_agent_reg:
5945
5946 ret = pthread_join(apps_notify_thread, &status);
5947 if (ret) {
5948 errno = ret;
5949 PERROR("pthread_join apps notify");
5950 retval = -1;
5951 }
5952 exit_apps_notify:
5953
5954 ret = pthread_join(apps_thread, &status);
5955 if (ret) {
5956 errno = ret;
5957 PERROR("pthread_join apps");
5958 retval = -1;
5959 }
5960 exit_apps:
5961
5962 ret = pthread_join(reg_apps_thread, &status);
5963 if (ret) {
5964 errno = ret;
5965 PERROR("pthread_join");
5966 retval = -1;
5967 }
5968 exit_reg_apps:
5969
5970 ret = pthread_join(dispatch_thread, &status);
5971 if (ret) {
5972 errno = ret;
5973 PERROR("pthread_join");
5974 retval = -1;
5975 }
5976 exit_dispatch:
5977
5978 ret = pthread_join(client_thread, &status);
5979 if (ret) {
5980 errno = ret;
5981 PERROR("pthread_join");
5982 retval = -1;
5983 }
5984 exit_client:
5985
5986 ret = pthread_join(health_thread, &status);
5987 if (ret) {
5988 errno = ret;
5989 PERROR("pthread_join health thread");
5990 retval = -1;
5991 }
5992 exit_health:
5993
5994 exit_init_data:
5995 /*
5996 * sessiond_cleanup() is called when no other thread is running, except
5997 * the ht_cleanup thread, which is needed to destroy the hash tables.
5998 */
5999 rcu_thread_online();
6000 sessiond_cleanup();
6001 rcu_thread_offline();
6002 rcu_unregister_thread();
6003
6004 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6005 if (ret < 0) {
6006 ERR("write error on ht_cleanup quit pipe");
6007 retval = -1;
6008 }
6009
6010 ret = pthread_join(ht_cleanup_thread, &status);
6011 if (ret) {
6012 errno = ret;
6013 PERROR("pthread_join ht cleanup thread");
6014 retval = -1;
6015 }
6016 exit_ht_cleanup:
6017 exit_set_max_size:
6018
6019 utils_close_pipe(ht_cleanup_pipe);
6020 exit_ht_cleanup_pipe:
6021
6022 /*
6023 * Close the ht_cleanup quit pipe.
6024 */
6025 utils_close_pipe(ht_cleanup_quit_pipe);
6026 exit_ht_cleanup_quit_pipe:
6027
6028 health_app_destroy(health_sessiond);
6029 exit_health_sessiond_cleanup:
6030
6031 exit_options:
6032 sessiond_cleanup_options();
6033
6034 exit_set_signal_handler:
6035 if (!retval) {
6036 exit(EXIT_SUCCESS);
6037 } else {
6038 exit(EXIT_FAILURE);
6039 }
6040 }
This page took 0.151951 seconds and 4 git commands to generate.