Fix: consumer signal handling race
[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
77 #define CONSUMERD_FILE "lttng-consumerd"
78
79 const char *progname;
80 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
81 static int tracing_group_name_override;
82 static char *opt_pidfile;
83 static int opt_sig_parent;
84 static int opt_verbose_consumer;
85 static int opt_daemon, opt_background;
86 static int opt_no_kernel;
87 static char *opt_load_session_path;
88 static pid_t ppid; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
90 static char *rundir;
91 static int lockfile_fd = -1;
92
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal;
95
96 /*
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
99 */
100 static struct consumer_data kconsumer_data = {
101 .type = LTTNG_CONSUMER_KERNEL,
102 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
103 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
104 .err_sock = -1,
105 .cmd_sock = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
118 .lock = PTHREAD_MUTEX_INITIALIZER,
119 .cond = PTHREAD_COND_INITIALIZER,
120 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
121 };
122 static struct consumer_data ustconsumer32_data = {
123 .type = LTTNG_CONSUMER32_UST,
124 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
125 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
126 .err_sock = -1,
127 .cmd_sock = -1,
128 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
129 .lock = PTHREAD_MUTEX_INITIALIZER,
130 .cond = PTHREAD_COND_INITIALIZER,
131 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
132 };
133
134 /* Command line options */
135 static const struct option long_options[] = {
136 { "client-sock", required_argument, 0, 'c' },
137 { "apps-sock", required_argument, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
144 { "consumerd32-path", required_argument, 0, '\0' },
145 { "consumerd32-libdir", required_argument, 0, '\0' },
146 { "consumerd64-path", required_argument, 0, '\0' },
147 { "consumerd64-libdir", required_argument, 0, '\0' },
148 { "daemonize", no_argument, 0, 'd' },
149 { "background", no_argument, 0, 'b' },
150 { "sig-parent", no_argument, 0, 'S' },
151 { "help", no_argument, 0, 'h' },
152 { "group", required_argument, 0, 'g' },
153 { "version", no_argument, 0, 'V' },
154 { "quiet", no_argument, 0, 'q' },
155 { "verbose", no_argument, 0, 'v' },
156 { "verbose-consumer", no_argument, 0, '\0' },
157 { "no-kernel", no_argument, 0, '\0' },
158 { "pidfile", required_argument, 0, 'p' },
159 { "agent-tcp-port", required_argument, 0, '\0' },
160 { "config", required_argument, 0, 'f' },
161 { "load", required_argument, 0, 'l' },
162 { "kmod-probes", required_argument, 0, '\0' },
163 { "extra-kmod-probes", required_argument, 0, '\0' },
164 { NULL, 0, 0, 0 }
165 };
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path[PATH_MAX];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path[PATH_MAX];
177 /* global wait shm path for UST */
178 static char wait_shm_path[PATH_MAX];
179 /* Global health check unix path */
180 static char health_unix_sock_path[PATH_MAX];
181
182 /* Sockets and FDs */
183 static int client_sock = -1;
184 static int apps_sock = -1;
185 int kernel_tracer_fd = -1;
186 static int kernel_poll_pipe[2] = { -1, -1 };
187
188 /*
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
191 */
192 static int thread_quit_pipe[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
194
195 /*
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
198 */
199 static int apps_cmd_pipe[2] = { -1, -1 };
200
201 int apps_cmd_notify_pipe[2] = { -1, -1 };
202
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread;
205 static pthread_t apps_notify_thread;
206 static pthread_t reg_apps_thread;
207 static pthread_t client_thread;
208 static pthread_t kernel_thread;
209 static pthread_t dispatch_thread;
210 static pthread_t health_thread;
211 static pthread_t ht_cleanup_thread;
212 static pthread_t agent_reg_thread;
213 static pthread_t load_session_thread;
214
215 /*
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 *
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
223 */
224 static struct ust_cmd_queue ust_cmd_queue;
225
226 /*
227 * Pointer initialized before thread creation.
228 *
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
232 *
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
235 */
236 static struct ltt_session_list *session_list_ptr;
237
238 int ust_consumerd64_fd = -1;
239 int ust_consumerd32_fd = -1;
240
241 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
242 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
243 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
244 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
245 static int consumerd32_bin_override;
246 static int consumerd64_bin_override;
247 static int consumerd32_libdir_override;
248 static int consumerd64_libdir_override;
249
250 static const char *module_proc_lttng = "/proc/lttng";
251
252 /*
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
255 */
256 enum consumerd_state {
257 CONSUMER_STARTED = 1,
258 CONSUMER_STOPPED = 2,
259 CONSUMER_ERROR = 3,
260 };
261
262 /*
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
267 *
268 * The following example shows a possible race condition of this scheme:
269 *
270 * consumer thread error happens
271 * client cmd arrives
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
275 * ...
276 *
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
282 */
283 static enum consumerd_state ust_consumerd_state;
284 static enum consumerd_state kernel_consumerd_state;
285
286 /*
287 * Socket timeout for receiving and sending in seconds.
288 */
289 static int app_socket_timeout;
290
291 /* Set in main() with the current page size. */
292 long page_size;
293
294 /* Application health monitoring */
295 struct health_app *health_sessiond;
296
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
299
300 /* Am I root or not. */
301 int is_root; /* Set to 1 if the daemon is running as root */
302
303 const char * const config_section_name = "sessiond";
304
305 /* Load session thread information to operate. */
306 struct load_session_thread_data *load_info;
307
308 /*
309 * Whether sessiond is ready for commands/health check requests.
310 * NR_LTTNG_SESSIOND_READY must match the number of calls to
311 * sessiond_notify_ready().
312 */
313 #define NR_LTTNG_SESSIOND_READY 3
314 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
315
316 /* Notify parents that we are ready for cmd and health check */
317 LTTNG_HIDDEN
318 void sessiond_notify_ready(void)
319 {
320 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
321 /*
322 * Notify parent pid that we are ready to accept command
323 * for client side. This ppid is the one from the
324 * external process that spawned us.
325 */
326 if (opt_sig_parent) {
327 kill(ppid, SIGUSR1);
328 }
329
330 /*
331 * Notify the parent of the fork() process that we are
332 * ready.
333 */
334 if (opt_daemon || opt_background) {
335 kill(child_ppid, SIGUSR1);
336 }
337 }
338 }
339
340 static
341 void setup_consumerd_path(void)
342 {
343 const char *bin, *libdir;
344
345 /*
346 * Allow INSTALL_BIN_PATH to be used as a target path for the
347 * native architecture size consumer if CONFIG_CONSUMER*_PATH
348 * has not been defined.
349 */
350 #if (CAA_BITS_PER_LONG == 32)
351 if (!consumerd32_bin[0]) {
352 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
353 }
354 if (!consumerd32_libdir[0]) {
355 consumerd32_libdir = INSTALL_LIB_PATH;
356 }
357 #elif (CAA_BITS_PER_LONG == 64)
358 if (!consumerd64_bin[0]) {
359 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
360 }
361 if (!consumerd64_libdir[0]) {
362 consumerd64_libdir = INSTALL_LIB_PATH;
363 }
364 #else
365 #error "Unknown bitness"
366 #endif
367
368 /*
369 * runtime env. var. overrides the build default.
370 */
371 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
372 if (bin) {
373 consumerd32_bin = bin;
374 }
375 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
376 if (bin) {
377 consumerd64_bin = bin;
378 }
379 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
380 if (libdir) {
381 consumerd32_libdir = libdir;
382 }
383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
384 if (libdir) {
385 consumerd64_libdir = libdir;
386 }
387 }
388
389 static
390 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
391 int *a_pipe)
392 {
393 int ret;
394
395 assert(events);
396
397 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
398 if (ret < 0) {
399 goto error;
400 }
401
402 /* Add quit pipe */
403 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
404 if (ret < 0) {
405 goto error;
406 }
407
408 return 0;
409
410 error:
411 return ret;
412 }
413
414 /*
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
416 */
417 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
418 {
419 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
420 }
421
422 /*
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 */
425 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
426 size_t size)
427 {
428 return __sessiond_set_thread_pollset(events, size,
429 ht_cleanup_quit_pipe);
430 }
431
432 static
433 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
434 {
435 if (fd == a_pipe && (events & LPOLLIN)) {
436 return 1;
437 }
438 return 0;
439 }
440
441 /*
442 * Check if the thread quit pipe was triggered.
443 *
444 * Return 1 if it was triggered else 0;
445 */
446 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
447 {
448 return __sessiond_check_thread_quit_pipe(fd, events,
449 thread_quit_pipe[0]);
450 }
451
452 /*
453 * Check if the ht_cleanup thread quit pipe was triggered.
454 *
455 * Return 1 if it was triggered else 0;
456 */
457 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
458 {
459 return __sessiond_check_thread_quit_pipe(fd, events,
460 ht_cleanup_quit_pipe[0]);
461 }
462
463 /*
464 * Init thread quit pipe.
465 *
466 * Return -1 on error or 0 if all pipes are created.
467 */
468 static int __init_thread_quit_pipe(int *a_pipe)
469 {
470 int ret, i;
471
472 ret = pipe(a_pipe);
473 if (ret < 0) {
474 PERROR("thread quit pipe");
475 goto error;
476 }
477
478 for (i = 0; i < 2; i++) {
479 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
480 if (ret < 0) {
481 PERROR("fcntl");
482 goto error;
483 }
484 }
485
486 error:
487 return ret;
488 }
489
490 static int init_thread_quit_pipe(void)
491 {
492 return __init_thread_quit_pipe(thread_quit_pipe);
493 }
494
495 static int init_ht_cleanup_quit_pipe(void)
496 {
497 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
498 }
499
500 /*
501 * Stop all threads by closing the thread quit pipe.
502 */
503 static void stop_threads(void)
504 {
505 int ret;
506
507 /* Stopping all threads */
508 DBG("Terminating all threads");
509 ret = notify_thread_pipe(thread_quit_pipe[1]);
510 if (ret < 0) {
511 ERR("write error on thread quit pipe");
512 }
513
514 /* Dispatch thread */
515 CMM_STORE_SHARED(dispatch_thread_exit, 1);
516 futex_nto1_wake(&ust_cmd_queue.futex);
517 }
518
519 /*
520 * Close every consumer sockets.
521 */
522 static void close_consumer_sockets(void)
523 {
524 int ret;
525
526 if (kconsumer_data.err_sock >= 0) {
527 ret = close(kconsumer_data.err_sock);
528 if (ret < 0) {
529 PERROR("kernel consumer err_sock close");
530 }
531 }
532 if (ustconsumer32_data.err_sock >= 0) {
533 ret = close(ustconsumer32_data.err_sock);
534 if (ret < 0) {
535 PERROR("UST consumerd32 err_sock close");
536 }
537 }
538 if (ustconsumer64_data.err_sock >= 0) {
539 ret = close(ustconsumer64_data.err_sock);
540 if (ret < 0) {
541 PERROR("UST consumerd64 err_sock close");
542 }
543 }
544 if (kconsumer_data.cmd_sock >= 0) {
545 ret = close(kconsumer_data.cmd_sock);
546 if (ret < 0) {
547 PERROR("kernel consumer cmd_sock close");
548 }
549 }
550 if (ustconsumer32_data.cmd_sock >= 0) {
551 ret = close(ustconsumer32_data.cmd_sock);
552 if (ret < 0) {
553 PERROR("UST consumerd32 cmd_sock close");
554 }
555 }
556 if (ustconsumer64_data.cmd_sock >= 0) {
557 ret = close(ustconsumer64_data.cmd_sock);
558 if (ret < 0) {
559 PERROR("UST consumerd64 cmd_sock close");
560 }
561 }
562 }
563
564 /*
565 * Generate the full lock file path using the rundir.
566 *
567 * Return the snprintf() return value thus a negative value is an error.
568 */
569 static int generate_lock_file_path(char *path, size_t len)
570 {
571 int ret;
572
573 assert(path);
574 assert(rundir);
575
576 /* Build lockfile path from rundir. */
577 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
578 if (ret < 0) {
579 PERROR("snprintf lockfile path");
580 }
581
582 return ret;
583 }
584
585 /*
586 * Wait on consumer process termination.
587 *
588 * Need to be called with the consumer data lock held or from a context
589 * ensuring no concurrent access to data (e.g: cleanup).
590 */
591 static void wait_consumer(struct consumer_data *consumer_data)
592 {
593 pid_t ret;
594 int status;
595
596 if (consumer_data->pid <= 0) {
597 return;
598 }
599
600 DBG("Waiting for complete teardown of consumerd (PID: %d)",
601 consumer_data->pid);
602 ret = waitpid(consumer_data->pid, &status, 0);
603 if (ret == -1) {
604 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
605 }
606 if (!WIFEXITED(status)) {
607 ERR("consumerd termination with error: %d",
608 WEXITSTATUS(ret));
609 }
610 consumer_data->pid = 0;
611 }
612
613 /*
614 * Cleanup the session daemon's data structures.
615 */
616 static void sessiond_cleanup(void)
617 {
618 int ret;
619 struct ltt_session *sess, *stmp;
620 char path[PATH_MAX];
621
622 DBG("Cleanup sessiond");
623
624 /*
625 * Close the thread quit pipe. It has already done its job,
626 * since we are now called.
627 */
628 utils_close_pipe(thread_quit_pipe);
629
630 /*
631 * If opt_pidfile is undefined, the default file will be wiped when
632 * removing the rundir.
633 */
634 if (opt_pidfile) {
635 ret = remove(opt_pidfile);
636 if (ret < 0) {
637 PERROR("remove pidfile %s", opt_pidfile);
638 }
639 }
640
641 DBG("Removing sessiond and consumerd content of directory %s", rundir);
642
643 /* sessiond */
644 snprintf(path, PATH_MAX,
645 "%s/%s",
646 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
647 DBG("Removing %s", path);
648 (void) unlink(path);
649
650 snprintf(path, PATH_MAX, "%s/%s", rundir,
651 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
652 DBG("Removing %s", path);
653 (void) unlink(path);
654
655 /* kconsumerd */
656 snprintf(path, PATH_MAX,
657 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
658 rundir);
659 DBG("Removing %s", path);
660 (void) unlink(path);
661
662 snprintf(path, PATH_MAX,
663 DEFAULT_KCONSUMERD_PATH,
664 rundir);
665 DBG("Removing directory %s", path);
666 (void) rmdir(path);
667
668 /* ust consumerd 32 */
669 snprintf(path, PATH_MAX,
670 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
671 rundir);
672 DBG("Removing %s", path);
673 (void) unlink(path);
674
675 snprintf(path, PATH_MAX,
676 DEFAULT_USTCONSUMERD32_PATH,
677 rundir);
678 DBG("Removing directory %s", path);
679 (void) rmdir(path);
680
681 /* ust consumerd 64 */
682 snprintf(path, PATH_MAX,
683 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
684 rundir);
685 DBG("Removing %s", path);
686 (void) unlink(path);
687
688 snprintf(path, PATH_MAX,
689 DEFAULT_USTCONSUMERD64_PATH,
690 rundir);
691 DBG("Removing directory %s", path);
692 (void) rmdir(path);
693
694 DBG("Cleaning up all sessions");
695
696 /* Destroy session list mutex */
697 if (session_list_ptr != NULL) {
698 pthread_mutex_destroy(&session_list_ptr->lock);
699
700 /* Cleanup ALL session */
701 cds_list_for_each_entry_safe(sess, stmp,
702 &session_list_ptr->head, list) {
703 cmd_destroy_session(sess, kernel_poll_pipe[1]);
704 }
705 }
706
707 wait_consumer(&kconsumer_data);
708 wait_consumer(&ustconsumer64_data);
709 wait_consumer(&ustconsumer32_data);
710
711 DBG("Cleaning up all agent apps");
712 agent_app_ht_clean();
713
714 DBG("Closing all UST sockets");
715 ust_app_clean_list();
716 buffer_reg_destroy_registries();
717
718 if (is_root && !opt_no_kernel) {
719 DBG2("Closing kernel fd");
720 if (kernel_tracer_fd >= 0) {
721 ret = close(kernel_tracer_fd);
722 if (ret) {
723 PERROR("close");
724 }
725 }
726 DBG("Unloading kernel modules");
727 modprobe_remove_lttng_all();
728 free(syscall_table);
729 }
730
731 close_consumer_sockets();
732
733 if (load_info) {
734 load_session_destroy_data(load_info);
735 free(load_info);
736 }
737
738 /*
739 * Cleanup lock file by deleting it and finaly closing it which will
740 * release the file system lock.
741 */
742 if (lockfile_fd >= 0) {
743 char lockfile_path[PATH_MAX];
744
745 ret = generate_lock_file_path(lockfile_path,
746 sizeof(lockfile_path));
747 if (ret > 0) {
748 ret = remove(lockfile_path);
749 if (ret < 0) {
750 PERROR("remove lock file");
751 }
752 ret = close(lockfile_fd);
753 if (ret < 0) {
754 PERROR("close lock file");
755 }
756 }
757 }
758
759 /*
760 * We do NOT rmdir rundir because there are other processes
761 * using it, for instance lttng-relayd, which can start in
762 * parallel with this teardown.
763 */
764
765 free(rundir);
766 }
767
768 /*
769 * Cleanup the daemon's option data structures.
770 */
771 static void sessiond_cleanup_options(void)
772 {
773 DBG("Cleaning up options");
774
775 /*
776 * If the override option is set, the pointer points to a *non* const
777 * thus freeing it even though the variable type is set to const.
778 */
779 if (tracing_group_name_override) {
780 free((void *) tracing_group_name);
781 }
782 if (consumerd32_bin_override) {
783 free((void *) consumerd32_bin);
784 }
785 if (consumerd64_bin_override) {
786 free((void *) consumerd64_bin);
787 }
788 if (consumerd32_libdir_override) {
789 free((void *) consumerd32_libdir);
790 }
791 if (consumerd64_libdir_override) {
792 free((void *) consumerd64_libdir);
793 }
794
795 free(opt_pidfile);
796 free(opt_load_session_path);
797 free(kmod_probes_list);
798 free(kmod_extra_probes_list);
799
800 /* <fun> */
801 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
802 "Matthew, BEET driven development works!%c[%dm",
803 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
804 /* </fun> */
805 }
806
807 /*
808 * Send data on a unix socket using the liblttsessiondcomm API.
809 *
810 * Return lttcomm error code.
811 */
812 static int send_unix_sock(int sock, void *buf, size_t len)
813 {
814 /* Check valid length */
815 if (len == 0) {
816 return -1;
817 }
818
819 return lttcomm_send_unix_sock(sock, buf, len);
820 }
821
822 /*
823 * Free memory of a command context structure.
824 */
825 static void clean_command_ctx(struct command_ctx **cmd_ctx)
826 {
827 DBG("Clean command context structure");
828 if (*cmd_ctx) {
829 if ((*cmd_ctx)->llm) {
830 free((*cmd_ctx)->llm);
831 }
832 if ((*cmd_ctx)->lsm) {
833 free((*cmd_ctx)->lsm);
834 }
835 free(*cmd_ctx);
836 *cmd_ctx = NULL;
837 }
838 }
839
840 /*
841 * Notify UST applications using the shm mmap futex.
842 */
843 static int notify_ust_apps(int active)
844 {
845 char *wait_shm_mmap;
846
847 DBG("Notifying applications of session daemon state: %d", active);
848
849 /* See shm.c for this call implying mmap, shm and futex calls */
850 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
851 if (wait_shm_mmap == NULL) {
852 goto error;
853 }
854
855 /* Wake waiting process */
856 futex_wait_update((int32_t *) wait_shm_mmap, active);
857
858 /* Apps notified successfully */
859 return 0;
860
861 error:
862 return -1;
863 }
864
865 /*
866 * Setup the outgoing data buffer for the response (llm) by allocating the
867 * right amount of memory and copying the original information from the lsm
868 * structure.
869 *
870 * Return total size of the buffer pointed by buf.
871 */
872 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
873 {
874 int ret, buf_size;
875
876 buf_size = size;
877
878 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
879 if (cmd_ctx->llm == NULL) {
880 PERROR("zmalloc");
881 ret = -ENOMEM;
882 goto error;
883 }
884
885 /* Copy common data */
886 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
887 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
888
889 cmd_ctx->llm->data_size = size;
890 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
891
892 return buf_size;
893
894 error:
895 return ret;
896 }
897
898 /*
899 * Update the kernel poll set of all channel fd available over all tracing
900 * session. Add the wakeup pipe at the end of the set.
901 */
902 static int update_kernel_poll(struct lttng_poll_event *events)
903 {
904 int ret;
905 struct ltt_session *session;
906 struct ltt_kernel_channel *channel;
907
908 DBG("Updating kernel poll set");
909
910 session_lock_list();
911 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
912 session_lock(session);
913 if (session->kernel_session == NULL) {
914 session_unlock(session);
915 continue;
916 }
917
918 cds_list_for_each_entry(channel,
919 &session->kernel_session->channel_list.head, list) {
920 /* Add channel fd to the kernel poll set */
921 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
922 if (ret < 0) {
923 session_unlock(session);
924 goto error;
925 }
926 DBG("Channel fd %d added to kernel set", channel->fd);
927 }
928 session_unlock(session);
929 }
930 session_unlock_list();
931
932 return 0;
933
934 error:
935 session_unlock_list();
936 return -1;
937 }
938
939 /*
940 * Find the channel fd from 'fd' over all tracing session. When found, check
941 * for new channel stream and send those stream fds to the kernel consumer.
942 *
943 * Useful for CPU hotplug feature.
944 */
945 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
946 {
947 int ret = 0;
948 struct ltt_session *session;
949 struct ltt_kernel_session *ksess;
950 struct ltt_kernel_channel *channel;
951
952 DBG("Updating kernel streams for channel fd %d", fd);
953
954 session_lock_list();
955 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
956 session_lock(session);
957 if (session->kernel_session == NULL) {
958 session_unlock(session);
959 continue;
960 }
961 ksess = session->kernel_session;
962
963 cds_list_for_each_entry(channel,
964 &ksess->channel_list.head, list) {
965 struct lttng_ht_iter iter;
966 struct consumer_socket *socket;
967
968 if (channel->fd != fd) {
969 continue;
970 }
971 DBG("Channel found, updating kernel streams");
972 ret = kernel_open_channel_stream(channel);
973 if (ret < 0) {
974 goto error;
975 }
976 /* Update the stream global counter */
977 ksess->stream_count_global += ret;
978
979 /*
980 * Have we already sent fds to the consumer? If yes, it
981 * means that tracing is started so it is safe to send
982 * our updated stream fds.
983 */
984 if (ksess->consumer_fds_sent != 1
985 || ksess->consumer == NULL) {
986 ret = -1;
987 goto error;
988 }
989
990 rcu_read_lock();
991 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
992 &iter.iter, socket, node.node) {
993 pthread_mutex_lock(socket->lock);
994 ret = kernel_consumer_send_channel_stream(socket,
995 channel, ksess,
996 session->output_traces ? 1 : 0);
997 pthread_mutex_unlock(socket->lock);
998 if (ret < 0) {
999 rcu_read_unlock();
1000 goto error;
1001 }
1002 }
1003 rcu_read_unlock();
1004 }
1005 session_unlock(session);
1006 }
1007 session_unlock_list();
1008 return ret;
1009
1010 error:
1011 session_unlock(session);
1012 session_unlock_list();
1013 return ret;
1014 }
1015
1016 /*
1017 * For each tracing session, update newly registered apps. The session list
1018 * lock MUST be acquired before calling this.
1019 */
1020 static void update_ust_app(int app_sock)
1021 {
1022 struct ltt_session *sess, *stmp;
1023
1024 /* Consumer is in an ERROR state. Stop any application update. */
1025 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1026 /* Stop the update process since the consumer is dead. */
1027 return;
1028 }
1029
1030 /* For all tracing session(s) */
1031 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1032 struct ust_app *app;
1033
1034 session_lock(sess);
1035 if (!sess->ust_session) {
1036 goto unlock_session;
1037 }
1038
1039 rcu_read_lock();
1040 assert(app_sock >= 0);
1041 app = ust_app_find_by_sock(app_sock);
1042 if (app == NULL) {
1043 /*
1044 * Application can be unregistered before so
1045 * this is possible hence simply stopping the
1046 * update.
1047 */
1048 DBG3("UST app update failed to find app sock %d",
1049 app_sock);
1050 goto unlock_rcu;
1051 }
1052 ust_app_global_update(sess->ust_session, app);
1053 unlock_rcu:
1054 rcu_read_unlock();
1055 unlock_session:
1056 session_unlock(sess);
1057 }
1058 }
1059
1060 /*
1061 * This thread manage event coming from the kernel.
1062 *
1063 * Features supported in this thread:
1064 * -) CPU Hotplug
1065 */
1066 static void *thread_manage_kernel(void *data)
1067 {
1068 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1069 uint32_t revents, nb_fd;
1070 char tmp;
1071 struct lttng_poll_event events;
1072
1073 DBG("[thread] Thread manage kernel started");
1074
1075 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1076
1077 /*
1078 * This first step of the while is to clean this structure which could free
1079 * non NULL pointers so initialize it before the loop.
1080 */
1081 lttng_poll_init(&events);
1082
1083 if (testpoint(sessiond_thread_manage_kernel)) {
1084 goto error_testpoint;
1085 }
1086
1087 health_code_update();
1088
1089 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1090 goto error_testpoint;
1091 }
1092
1093 while (1) {
1094 health_code_update();
1095
1096 if (update_poll_flag == 1) {
1097 /* Clean events object. We are about to populate it again. */
1098 lttng_poll_clean(&events);
1099
1100 ret = sessiond_set_thread_pollset(&events, 2);
1101 if (ret < 0) {
1102 goto error_poll_create;
1103 }
1104
1105 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1106 if (ret < 0) {
1107 goto error;
1108 }
1109
1110 /* This will add the available kernel channel if any. */
1111 ret = update_kernel_poll(&events);
1112 if (ret < 0) {
1113 goto error;
1114 }
1115 update_poll_flag = 0;
1116 }
1117
1118 DBG("Thread kernel polling");
1119
1120 /* Poll infinite value of time */
1121 restart:
1122 health_poll_entry();
1123 ret = lttng_poll_wait(&events, -1);
1124 DBG("Thread kernel return from poll on %d fds",
1125 LTTNG_POLL_GETNB(&events));
1126 health_poll_exit();
1127 if (ret < 0) {
1128 /*
1129 * Restart interrupted system call.
1130 */
1131 if (errno == EINTR) {
1132 goto restart;
1133 }
1134 goto error;
1135 } else if (ret == 0) {
1136 /* Should not happen since timeout is infinite */
1137 ERR("Return value of poll is 0 with an infinite timeout.\n"
1138 "This should not have happened! Continuing...");
1139 continue;
1140 }
1141
1142 nb_fd = ret;
1143
1144 for (i = 0; i < nb_fd; i++) {
1145 /* Fetch once the poll data */
1146 revents = LTTNG_POLL_GETEV(&events, i);
1147 pollfd = LTTNG_POLL_GETFD(&events, i);
1148
1149 health_code_update();
1150
1151 if (!revents) {
1152 /* No activity for this FD (poll implementation). */
1153 continue;
1154 }
1155
1156 /* Thread quit pipe has been closed. Killing thread. */
1157 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1158 if (ret) {
1159 err = 0;
1160 goto exit;
1161 }
1162
1163 /* Check for data on kernel pipe */
1164 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1165 (void) lttng_read(kernel_poll_pipe[0],
1166 &tmp, 1);
1167 /*
1168 * Ret value is useless here, if this pipe gets any actions an
1169 * update is required anyway.
1170 */
1171 update_poll_flag = 1;
1172 continue;
1173 } else {
1174 /*
1175 * New CPU detected by the kernel. Adding kernel stream to
1176 * kernel session and updating the kernel consumer
1177 */
1178 if (revents & LPOLLIN) {
1179 ret = update_kernel_stream(&kconsumer_data, pollfd);
1180 if (ret < 0) {
1181 continue;
1182 }
1183 break;
1184 /*
1185 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1186 * and unregister kernel stream at this point.
1187 */
1188 }
1189 }
1190 }
1191 }
1192
1193 exit:
1194 error:
1195 lttng_poll_clean(&events);
1196 error_poll_create:
1197 error_testpoint:
1198 utils_close_pipe(kernel_poll_pipe);
1199 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1200 if (err) {
1201 health_error();
1202 ERR("Health error occurred in %s", __func__);
1203 WARN("Kernel thread died unexpectedly. "
1204 "Kernel tracing can continue but CPU hotplug is disabled.");
1205 }
1206 health_unregister(health_sessiond);
1207 DBG("Kernel thread dying");
1208 return NULL;
1209 }
1210
1211 /*
1212 * Signal pthread condition of the consumer data that the thread.
1213 */
1214 static void signal_consumer_condition(struct consumer_data *data, int state)
1215 {
1216 pthread_mutex_lock(&data->cond_mutex);
1217
1218 /*
1219 * The state is set before signaling. It can be any value, it's the waiter
1220 * job to correctly interpret this condition variable associated to the
1221 * consumer pthread_cond.
1222 *
1223 * A value of 0 means that the corresponding thread of the consumer data
1224 * was not started. 1 indicates that the thread has started and is ready
1225 * for action. A negative value means that there was an error during the
1226 * thread bootstrap.
1227 */
1228 data->consumer_thread_is_ready = state;
1229 (void) pthread_cond_signal(&data->cond);
1230
1231 pthread_mutex_unlock(&data->cond_mutex);
1232 }
1233
1234 /*
1235 * This thread manage the consumer error sent back to the session daemon.
1236 */
1237 static void *thread_manage_consumer(void *data)
1238 {
1239 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1240 uint32_t revents, nb_fd;
1241 enum lttcomm_return_code code;
1242 struct lttng_poll_event events;
1243 struct consumer_data *consumer_data = data;
1244
1245 DBG("[thread] Manage consumer started");
1246
1247 rcu_register_thread();
1248 rcu_thread_online();
1249
1250 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1251
1252 health_code_update();
1253
1254 /*
1255 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1256 * metadata_sock. Nothing more will be added to this poll set.
1257 */
1258 ret = sessiond_set_thread_pollset(&events, 3);
1259 if (ret < 0) {
1260 goto error_poll;
1261 }
1262
1263 /*
1264 * The error socket here is already in a listening state which was done
1265 * just before spawning this thread to avoid a race between the consumer
1266 * daemon exec trying to connect and the listen() call.
1267 */
1268 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1269 if (ret < 0) {
1270 goto error;
1271 }
1272
1273 health_code_update();
1274
1275 /* Infinite blocking call, waiting for transmission */
1276 restart:
1277 health_poll_entry();
1278
1279 if (testpoint(sessiond_thread_manage_consumer)) {
1280 goto error;
1281 }
1282
1283 ret = lttng_poll_wait(&events, -1);
1284 health_poll_exit();
1285 if (ret < 0) {
1286 /*
1287 * Restart interrupted system call.
1288 */
1289 if (errno == EINTR) {
1290 goto restart;
1291 }
1292 goto error;
1293 }
1294
1295 nb_fd = ret;
1296
1297 for (i = 0; i < nb_fd; i++) {
1298 /* Fetch once the poll data */
1299 revents = LTTNG_POLL_GETEV(&events, i);
1300 pollfd = LTTNG_POLL_GETFD(&events, i);
1301
1302 health_code_update();
1303
1304 if (!revents) {
1305 /* No activity for this FD (poll implementation). */
1306 continue;
1307 }
1308
1309 /* Thread quit pipe has been closed. Killing thread. */
1310 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1311 if (ret) {
1312 err = 0;
1313 goto exit;
1314 }
1315
1316 /* Event on the registration socket */
1317 if (pollfd == consumer_data->err_sock) {
1318 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1319 ERR("consumer err socket poll error");
1320 goto error;
1321 }
1322 }
1323 }
1324
1325 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1326 if (sock < 0) {
1327 goto error;
1328 }
1329
1330 /*
1331 * Set the CLOEXEC flag. Return code is useless because either way, the
1332 * show must go on.
1333 */
1334 (void) utils_set_fd_cloexec(sock);
1335
1336 health_code_update();
1337
1338 DBG2("Receiving code from consumer err_sock");
1339
1340 /* Getting status code from kconsumerd */
1341 ret = lttcomm_recv_unix_sock(sock, &code,
1342 sizeof(enum lttcomm_return_code));
1343 if (ret <= 0) {
1344 goto error;
1345 }
1346
1347 health_code_update();
1348 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1349 /* Connect both socket, command and metadata. */
1350 consumer_data->cmd_sock =
1351 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1352 consumer_data->metadata_fd =
1353 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1354 if (consumer_data->cmd_sock < 0
1355 || consumer_data->metadata_fd < 0) {
1356 PERROR("consumer connect cmd socket");
1357 /* On error, signal condition and quit. */
1358 signal_consumer_condition(consumer_data, -1);
1359 goto error;
1360 }
1361 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1362 /* Create metadata socket lock. */
1363 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1364 if (consumer_data->metadata_sock.lock == NULL) {
1365 PERROR("zmalloc pthread mutex");
1366 ret = -1;
1367 goto error;
1368 }
1369 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1370
1371 signal_consumer_condition(consumer_data, 1);
1372 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1373 DBG("Consumer metadata socket ready (fd: %d)",
1374 consumer_data->metadata_fd);
1375 } else {
1376 ERR("consumer error when waiting for SOCK_READY : %s",
1377 lttcomm_get_readable_code(-code));
1378 goto error;
1379 }
1380
1381 /* Remove the consumerd error sock since we've established a connexion */
1382 ret = lttng_poll_del(&events, consumer_data->err_sock);
1383 if (ret < 0) {
1384 goto error;
1385 }
1386
1387 /* Add new accepted error socket. */
1388 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1389 if (ret < 0) {
1390 goto error;
1391 }
1392
1393 /* Add metadata socket that is successfully connected. */
1394 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1395 LPOLLIN | LPOLLRDHUP);
1396 if (ret < 0) {
1397 goto error;
1398 }
1399
1400 health_code_update();
1401
1402 /* Infinite blocking call, waiting for transmission */
1403 restart_poll:
1404 while (1) {
1405 health_code_update();
1406
1407 /* Exit the thread because the thread quit pipe has been triggered. */
1408 if (should_quit) {
1409 /* Not a health error. */
1410 err = 0;
1411 goto exit;
1412 }
1413
1414 health_poll_entry();
1415 ret = lttng_poll_wait(&events, -1);
1416 health_poll_exit();
1417 if (ret < 0) {
1418 /*
1419 * Restart interrupted system call.
1420 */
1421 if (errno == EINTR) {
1422 goto restart_poll;
1423 }
1424 goto error;
1425 }
1426
1427 nb_fd = ret;
1428
1429 for (i = 0; i < nb_fd; i++) {
1430 /* Fetch once the poll data */
1431 revents = LTTNG_POLL_GETEV(&events, i);
1432 pollfd = LTTNG_POLL_GETFD(&events, i);
1433
1434 health_code_update();
1435
1436 if (!revents) {
1437 /* No activity for this FD (poll implementation). */
1438 continue;
1439 }
1440
1441 /*
1442 * Thread quit pipe has been triggered, flag that we should stop
1443 * but continue the current loop to handle potential data from
1444 * consumer.
1445 */
1446 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1447
1448 if (pollfd == sock) {
1449 /* Event on the consumerd socket */
1450 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1451 ERR("consumer err socket second poll error");
1452 goto error;
1453 }
1454 health_code_update();
1455 /* Wait for any kconsumerd error */
1456 ret = lttcomm_recv_unix_sock(sock, &code,
1457 sizeof(enum lttcomm_return_code));
1458 if (ret <= 0) {
1459 ERR("consumer closed the command socket");
1460 goto error;
1461 }
1462
1463 ERR("consumer return code : %s",
1464 lttcomm_get_readable_code(-code));
1465
1466 goto exit;
1467 } else if (pollfd == consumer_data->metadata_fd) {
1468 /* UST metadata requests */
1469 ret = ust_consumer_metadata_request(
1470 &consumer_data->metadata_sock);
1471 if (ret < 0) {
1472 ERR("Handling metadata request");
1473 goto error;
1474 }
1475 }
1476 /* No need for an else branch all FDs are tested prior. */
1477 }
1478 health_code_update();
1479 }
1480
1481 exit:
1482 error:
1483 /*
1484 * We lock here because we are about to close the sockets and some other
1485 * thread might be using them so get exclusive access which will abort all
1486 * other consumer command by other threads.
1487 */
1488 pthread_mutex_lock(&consumer_data->lock);
1489
1490 /* Immediately set the consumerd state to stopped */
1491 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1492 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1493 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1494 consumer_data->type == LTTNG_CONSUMER32_UST) {
1495 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1496 } else {
1497 /* Code flow error... */
1498 assert(0);
1499 }
1500
1501 if (consumer_data->err_sock >= 0) {
1502 ret = close(consumer_data->err_sock);
1503 if (ret) {
1504 PERROR("close");
1505 }
1506 consumer_data->err_sock = -1;
1507 }
1508 if (consumer_data->cmd_sock >= 0) {
1509 ret = close(consumer_data->cmd_sock);
1510 if (ret) {
1511 PERROR("close");
1512 }
1513 consumer_data->cmd_sock = -1;
1514 }
1515 if (consumer_data->metadata_sock.fd_ptr &&
1516 *consumer_data->metadata_sock.fd_ptr >= 0) {
1517 ret = close(*consumer_data->metadata_sock.fd_ptr);
1518 if (ret) {
1519 PERROR("close");
1520 }
1521 }
1522 if (sock >= 0) {
1523 ret = close(sock);
1524 if (ret) {
1525 PERROR("close");
1526 }
1527 }
1528
1529 unlink(consumer_data->err_unix_sock_path);
1530 unlink(consumer_data->cmd_unix_sock_path);
1531 pthread_mutex_unlock(&consumer_data->lock);
1532
1533 /* Cleanup metadata socket mutex. */
1534 if (consumer_data->metadata_sock.lock) {
1535 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1536 free(consumer_data->metadata_sock.lock);
1537 }
1538 lttng_poll_clean(&events);
1539 error_poll:
1540 if (err) {
1541 health_error();
1542 ERR("Health error occurred in %s", __func__);
1543 }
1544 health_unregister(health_sessiond);
1545 DBG("consumer thread cleanup completed");
1546
1547 rcu_thread_offline();
1548 rcu_unregister_thread();
1549
1550 return NULL;
1551 }
1552
1553 /*
1554 * This thread manage application communication.
1555 */
1556 static void *thread_manage_apps(void *data)
1557 {
1558 int i, ret, pollfd, err = -1;
1559 ssize_t size_ret;
1560 uint32_t revents, nb_fd;
1561 struct lttng_poll_event events;
1562
1563 DBG("[thread] Manage application started");
1564
1565 rcu_register_thread();
1566 rcu_thread_online();
1567
1568 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1569
1570 if (testpoint(sessiond_thread_manage_apps)) {
1571 goto error_testpoint;
1572 }
1573
1574 health_code_update();
1575
1576 ret = sessiond_set_thread_pollset(&events, 2);
1577 if (ret < 0) {
1578 goto error_poll_create;
1579 }
1580
1581 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1582 if (ret < 0) {
1583 goto error;
1584 }
1585
1586 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1587 goto error;
1588 }
1589
1590 health_code_update();
1591
1592 while (1) {
1593 DBG("Apps thread polling");
1594
1595 /* Inifinite blocking call, waiting for transmission */
1596 restart:
1597 health_poll_entry();
1598 ret = lttng_poll_wait(&events, -1);
1599 DBG("Apps thread return from poll on %d fds",
1600 LTTNG_POLL_GETNB(&events));
1601 health_poll_exit();
1602 if (ret < 0) {
1603 /*
1604 * Restart interrupted system call.
1605 */
1606 if (errno == EINTR) {
1607 goto restart;
1608 }
1609 goto error;
1610 }
1611
1612 nb_fd = ret;
1613
1614 for (i = 0; i < nb_fd; i++) {
1615 /* Fetch once the poll data */
1616 revents = LTTNG_POLL_GETEV(&events, i);
1617 pollfd = LTTNG_POLL_GETFD(&events, i);
1618
1619 health_code_update();
1620
1621 if (!revents) {
1622 /* No activity for this FD (poll implementation). */
1623 continue;
1624 }
1625
1626 /* Thread quit pipe has been closed. Killing thread. */
1627 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1628 if (ret) {
1629 err = 0;
1630 goto exit;
1631 }
1632
1633 /* Inspect the apps cmd pipe */
1634 if (pollfd == apps_cmd_pipe[0]) {
1635 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1636 ERR("Apps command pipe error");
1637 goto error;
1638 } else if (revents & LPOLLIN) {
1639 int sock;
1640
1641 /* Empty pipe */
1642 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1643 if (size_ret < sizeof(sock)) {
1644 PERROR("read apps cmd pipe");
1645 goto error;
1646 }
1647
1648 health_code_update();
1649
1650 /*
1651 * We only monitor the error events of the socket. This
1652 * thread does not handle any incoming data from UST
1653 * (POLLIN).
1654 */
1655 ret = lttng_poll_add(&events, sock,
1656 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1657 if (ret < 0) {
1658 goto error;
1659 }
1660
1661 DBG("Apps with sock %d added to poll set", sock);
1662 }
1663 } else {
1664 /*
1665 * At this point, we know that a registered application made
1666 * the event at poll_wait.
1667 */
1668 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1669 /* Removing from the poll set */
1670 ret = lttng_poll_del(&events, pollfd);
1671 if (ret < 0) {
1672 goto error;
1673 }
1674
1675 /* Socket closed on remote end. */
1676 ust_app_unregister(pollfd);
1677 }
1678 }
1679
1680 health_code_update();
1681 }
1682 }
1683
1684 exit:
1685 error:
1686 lttng_poll_clean(&events);
1687 error_poll_create:
1688 error_testpoint:
1689 utils_close_pipe(apps_cmd_pipe);
1690 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1691
1692 /*
1693 * We don't clean the UST app hash table here since already registered
1694 * applications can still be controlled so let them be until the session
1695 * daemon dies or the applications stop.
1696 */
1697
1698 if (err) {
1699 health_error();
1700 ERR("Health error occurred in %s", __func__);
1701 }
1702 health_unregister(health_sessiond);
1703 DBG("Application communication apps thread cleanup complete");
1704 rcu_thread_offline();
1705 rcu_unregister_thread();
1706 return NULL;
1707 }
1708
1709 /*
1710 * Send a socket to a thread This is called from the dispatch UST registration
1711 * thread once all sockets are set for the application.
1712 *
1713 * The sock value can be invalid, we don't really care, the thread will handle
1714 * it and make the necessary cleanup if so.
1715 *
1716 * On success, return 0 else a negative value being the errno message of the
1717 * write().
1718 */
1719 static int send_socket_to_thread(int fd, int sock)
1720 {
1721 ssize_t ret;
1722
1723 /*
1724 * It's possible that the FD is set as invalid with -1 concurrently just
1725 * before calling this function being a shutdown state of the thread.
1726 */
1727 if (fd < 0) {
1728 ret = -EBADF;
1729 goto error;
1730 }
1731
1732 ret = lttng_write(fd, &sock, sizeof(sock));
1733 if (ret < sizeof(sock)) {
1734 PERROR("write apps pipe %d", fd);
1735 if (ret < 0) {
1736 ret = -errno;
1737 }
1738 goto error;
1739 }
1740
1741 /* All good. Don't send back the write positive ret value. */
1742 ret = 0;
1743 error:
1744 return (int) ret;
1745 }
1746
1747 /*
1748 * Sanitize the wait queue of the dispatch registration thread meaning removing
1749 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1750 * notify socket is never received.
1751 */
1752 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1753 {
1754 int ret, nb_fd = 0, i;
1755 unsigned int fd_added = 0;
1756 struct lttng_poll_event events;
1757 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1758
1759 assert(wait_queue);
1760
1761 lttng_poll_init(&events);
1762
1763 /* Just skip everything for an empty queue. */
1764 if (!wait_queue->count) {
1765 goto end;
1766 }
1767
1768 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1769 if (ret < 0) {
1770 goto error_create;
1771 }
1772
1773 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1774 &wait_queue->head, head) {
1775 assert(wait_node->app);
1776 ret = lttng_poll_add(&events, wait_node->app->sock,
1777 LPOLLHUP | LPOLLERR);
1778 if (ret < 0) {
1779 goto error;
1780 }
1781
1782 fd_added = 1;
1783 }
1784
1785 if (!fd_added) {
1786 goto end;
1787 }
1788
1789 /*
1790 * Poll but don't block so we can quickly identify the faulty events and
1791 * clean them afterwards from the wait queue.
1792 */
1793 ret = lttng_poll_wait(&events, 0);
1794 if (ret < 0) {
1795 goto error;
1796 }
1797 nb_fd = ret;
1798
1799 for (i = 0; i < nb_fd; i++) {
1800 /* Get faulty FD. */
1801 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1802 int pollfd = LTTNG_POLL_GETFD(&events, i);
1803
1804 if (!revents) {
1805 /* No activity for this FD (poll implementation). */
1806 continue;
1807 }
1808
1809 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1810 &wait_queue->head, head) {
1811 if (pollfd == wait_node->app->sock &&
1812 (revents & (LPOLLHUP | LPOLLERR))) {
1813 cds_list_del(&wait_node->head);
1814 wait_queue->count--;
1815 ust_app_destroy(wait_node->app);
1816 free(wait_node);
1817 break;
1818 }
1819 }
1820 }
1821
1822 if (nb_fd > 0) {
1823 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1824 }
1825
1826 end:
1827 lttng_poll_clean(&events);
1828 return;
1829
1830 error:
1831 lttng_poll_clean(&events);
1832 error_create:
1833 ERR("Unable to sanitize wait queue");
1834 return;
1835 }
1836
1837 /*
1838 * Dispatch request from the registration threads to the application
1839 * communication thread.
1840 */
1841 static void *thread_dispatch_ust_registration(void *data)
1842 {
1843 int ret, err = -1;
1844 struct cds_wfcq_node *node;
1845 struct ust_command *ust_cmd = NULL;
1846 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1847 struct ust_reg_wait_queue wait_queue = {
1848 .count = 0,
1849 };
1850
1851 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1852
1853 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1854 goto error_testpoint;
1855 }
1856
1857 health_code_update();
1858
1859 CDS_INIT_LIST_HEAD(&wait_queue.head);
1860
1861 DBG("[thread] Dispatch UST command started");
1862
1863 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1864 health_code_update();
1865
1866 /* Atomically prepare the queue futex */
1867 futex_nto1_prepare(&ust_cmd_queue.futex);
1868
1869 do {
1870 struct ust_app *app = NULL;
1871 ust_cmd = NULL;
1872
1873 /*
1874 * Make sure we don't have node(s) that have hung up before receiving
1875 * the notify socket. This is to clean the list in order to avoid
1876 * memory leaks from notify socket that are never seen.
1877 */
1878 sanitize_wait_queue(&wait_queue);
1879
1880 health_code_update();
1881 /* Dequeue command for registration */
1882 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1883 if (node == NULL) {
1884 DBG("Woken up but nothing in the UST command queue");
1885 /* Continue thread execution */
1886 break;
1887 }
1888
1889 ust_cmd = caa_container_of(node, struct ust_command, node);
1890
1891 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1892 " gid:%d sock:%d name:%s (version %d.%d)",
1893 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1894 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1895 ust_cmd->sock, ust_cmd->reg_msg.name,
1896 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1897
1898 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1899 wait_node = zmalloc(sizeof(*wait_node));
1900 if (!wait_node) {
1901 PERROR("zmalloc wait_node dispatch");
1902 ret = close(ust_cmd->sock);
1903 if (ret < 0) {
1904 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1905 }
1906 lttng_fd_put(LTTNG_FD_APPS, 1);
1907 free(ust_cmd);
1908 goto error;
1909 }
1910 CDS_INIT_LIST_HEAD(&wait_node->head);
1911
1912 /* Create application object if socket is CMD. */
1913 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1914 ust_cmd->sock);
1915 if (!wait_node->app) {
1916 ret = close(ust_cmd->sock);
1917 if (ret < 0) {
1918 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1919 }
1920 lttng_fd_put(LTTNG_FD_APPS, 1);
1921 free(wait_node);
1922 free(ust_cmd);
1923 continue;
1924 }
1925 /*
1926 * Add application to the wait queue so we can set the notify
1927 * socket before putting this object in the global ht.
1928 */
1929 cds_list_add(&wait_node->head, &wait_queue.head);
1930 wait_queue.count++;
1931
1932 free(ust_cmd);
1933 /*
1934 * We have to continue here since we don't have the notify
1935 * socket and the application MUST be added to the hash table
1936 * only at that moment.
1937 */
1938 continue;
1939 } else {
1940 /*
1941 * Look for the application in the local wait queue and set the
1942 * notify socket if found.
1943 */
1944 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1945 &wait_queue.head, head) {
1946 health_code_update();
1947 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1948 wait_node->app->notify_sock = ust_cmd->sock;
1949 cds_list_del(&wait_node->head);
1950 wait_queue.count--;
1951 app = wait_node->app;
1952 free(wait_node);
1953 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1954 break;
1955 }
1956 }
1957
1958 /*
1959 * With no application at this stage the received socket is
1960 * basically useless so close it before we free the cmd data
1961 * structure for good.
1962 */
1963 if (!app) {
1964 ret = close(ust_cmd->sock);
1965 if (ret < 0) {
1966 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1967 }
1968 lttng_fd_put(LTTNG_FD_APPS, 1);
1969 }
1970 free(ust_cmd);
1971 }
1972
1973 if (app) {
1974 /*
1975 * @session_lock_list
1976 *
1977 * Lock the global session list so from the register up to the
1978 * registration done message, no thread can see the application
1979 * and change its state.
1980 */
1981 session_lock_list();
1982 rcu_read_lock();
1983
1984 /*
1985 * Add application to the global hash table. This needs to be
1986 * done before the update to the UST registry can locate the
1987 * application.
1988 */
1989 ust_app_add(app);
1990
1991 /* Set app version. This call will print an error if needed. */
1992 (void) ust_app_version(app);
1993
1994 /* Send notify socket through the notify pipe. */
1995 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1996 app->notify_sock);
1997 if (ret < 0) {
1998 rcu_read_unlock();
1999 session_unlock_list();
2000 /*
2001 * No notify thread, stop the UST tracing. However, this is
2002 * not an internal error of the this thread thus setting
2003 * the health error code to a normal exit.
2004 */
2005 err = 0;
2006 goto error;
2007 }
2008
2009 /*
2010 * Update newly registered application with the tracing
2011 * registry info already enabled information.
2012 */
2013 update_ust_app(app->sock);
2014
2015 /*
2016 * Don't care about return value. Let the manage apps threads
2017 * handle app unregistration upon socket close.
2018 */
2019 (void) ust_app_register_done(app);
2020
2021 /*
2022 * Even if the application socket has been closed, send the app
2023 * to the thread and unregistration will take place at that
2024 * place.
2025 */
2026 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2027 if (ret < 0) {
2028 rcu_read_unlock();
2029 session_unlock_list();
2030 /*
2031 * No apps. thread, stop the UST tracing. However, this is
2032 * not an internal error of the this thread thus setting
2033 * the health error code to a normal exit.
2034 */
2035 err = 0;
2036 goto error;
2037 }
2038
2039 rcu_read_unlock();
2040 session_unlock_list();
2041 }
2042 } while (node != NULL);
2043
2044 health_poll_entry();
2045 /* Futex wait on queue. Blocking call on futex() */
2046 futex_nto1_wait(&ust_cmd_queue.futex);
2047 health_poll_exit();
2048 }
2049 /* Normal exit, no error */
2050 err = 0;
2051
2052 error:
2053 /* Clean up wait queue. */
2054 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2055 &wait_queue.head, head) {
2056 cds_list_del(&wait_node->head);
2057 wait_queue.count--;
2058 free(wait_node);
2059 }
2060
2061 /* Empty command queue. */
2062 for (;;) {
2063 /* Dequeue command for registration */
2064 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2065 if (node == NULL) {
2066 break;
2067 }
2068 ust_cmd = caa_container_of(node, struct ust_command, node);
2069 ret = close(ust_cmd->sock);
2070 if (ret < 0) {
2071 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2072 }
2073 lttng_fd_put(LTTNG_FD_APPS, 1);
2074 free(ust_cmd);
2075 }
2076
2077 error_testpoint:
2078 DBG("Dispatch thread dying");
2079 if (err) {
2080 health_error();
2081 ERR("Health error occurred in %s", __func__);
2082 }
2083 health_unregister(health_sessiond);
2084 return NULL;
2085 }
2086
2087 /*
2088 * This thread manage application registration.
2089 */
2090 static void *thread_registration_apps(void *data)
2091 {
2092 int sock = -1, i, ret, pollfd, err = -1;
2093 uint32_t revents, nb_fd;
2094 struct lttng_poll_event events;
2095 /*
2096 * Get allocated in this thread, enqueued to a global queue, dequeued and
2097 * freed in the manage apps thread.
2098 */
2099 struct ust_command *ust_cmd = NULL;
2100
2101 DBG("[thread] Manage application registration started");
2102
2103 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2104
2105 if (testpoint(sessiond_thread_registration_apps)) {
2106 goto error_testpoint;
2107 }
2108
2109 ret = lttcomm_listen_unix_sock(apps_sock);
2110 if (ret < 0) {
2111 goto error_listen;
2112 }
2113
2114 /*
2115 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2116 * more will be added to this poll set.
2117 */
2118 ret = sessiond_set_thread_pollset(&events, 2);
2119 if (ret < 0) {
2120 goto error_create_poll;
2121 }
2122
2123 /* Add the application registration socket */
2124 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2125 if (ret < 0) {
2126 goto error_poll_add;
2127 }
2128
2129 /* Notify all applications to register */
2130 ret = notify_ust_apps(1);
2131 if (ret < 0) {
2132 ERR("Failed to notify applications or create the wait shared memory.\n"
2133 "Execution continues but there might be problem for already\n"
2134 "running applications that wishes to register.");
2135 }
2136
2137 while (1) {
2138 DBG("Accepting application registration");
2139
2140 /* Inifinite blocking call, waiting for transmission */
2141 restart:
2142 health_poll_entry();
2143 ret = lttng_poll_wait(&events, -1);
2144 health_poll_exit();
2145 if (ret < 0) {
2146 /*
2147 * Restart interrupted system call.
2148 */
2149 if (errno == EINTR) {
2150 goto restart;
2151 }
2152 goto error;
2153 }
2154
2155 nb_fd = ret;
2156
2157 for (i = 0; i < nb_fd; i++) {
2158 health_code_update();
2159
2160 /* Fetch once the poll data */
2161 revents = LTTNG_POLL_GETEV(&events, i);
2162 pollfd = LTTNG_POLL_GETFD(&events, i);
2163
2164 if (!revents) {
2165 /* No activity for this FD (poll implementation). */
2166 continue;
2167 }
2168
2169 /* Thread quit pipe has been closed. Killing thread. */
2170 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2171 if (ret) {
2172 err = 0;
2173 goto exit;
2174 }
2175
2176 /* Event on the registration socket */
2177 if (pollfd == apps_sock) {
2178 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2179 ERR("Register apps socket poll error");
2180 goto error;
2181 } else if (revents & LPOLLIN) {
2182 sock = lttcomm_accept_unix_sock(apps_sock);
2183 if (sock < 0) {
2184 goto error;
2185 }
2186
2187 /*
2188 * Set socket timeout for both receiving and ending.
2189 * app_socket_timeout is in seconds, whereas
2190 * lttcomm_setsockopt_rcv_timeout and
2191 * lttcomm_setsockopt_snd_timeout expect msec as
2192 * parameter.
2193 */
2194 (void) lttcomm_setsockopt_rcv_timeout(sock,
2195 app_socket_timeout * 1000);
2196 (void) lttcomm_setsockopt_snd_timeout(sock,
2197 app_socket_timeout * 1000);
2198
2199 /*
2200 * Set the CLOEXEC flag. Return code is useless because
2201 * either way, the show must go on.
2202 */
2203 (void) utils_set_fd_cloexec(sock);
2204
2205 /* Create UST registration command for enqueuing */
2206 ust_cmd = zmalloc(sizeof(struct ust_command));
2207 if (ust_cmd == NULL) {
2208 PERROR("ust command zmalloc");
2209 ret = close(sock);
2210 if (ret) {
2211 PERROR("close");
2212 }
2213 goto error;
2214 }
2215
2216 /*
2217 * Using message-based transmissions to ensure we don't
2218 * have to deal with partially received messages.
2219 */
2220 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2221 if (ret < 0) {
2222 ERR("Exhausted file descriptors allowed for applications.");
2223 free(ust_cmd);
2224 ret = close(sock);
2225 if (ret) {
2226 PERROR("close");
2227 }
2228 sock = -1;
2229 continue;
2230 }
2231
2232 health_code_update();
2233 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2234 if (ret < 0) {
2235 free(ust_cmd);
2236 /* Close socket of the application. */
2237 ret = close(sock);
2238 if (ret) {
2239 PERROR("close");
2240 }
2241 lttng_fd_put(LTTNG_FD_APPS, 1);
2242 sock = -1;
2243 continue;
2244 }
2245 health_code_update();
2246
2247 ust_cmd->sock = sock;
2248 sock = -1;
2249
2250 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2251 " gid:%d sock:%d name:%s (version %d.%d)",
2252 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2253 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2254 ust_cmd->sock, ust_cmd->reg_msg.name,
2255 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2256
2257 /*
2258 * Lock free enqueue the registration request. The red pill
2259 * has been taken! This apps will be part of the *system*.
2260 */
2261 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2262
2263 /*
2264 * Wake the registration queue futex. Implicit memory
2265 * barrier with the exchange in cds_wfcq_enqueue.
2266 */
2267 futex_nto1_wake(&ust_cmd_queue.futex);
2268 }
2269 }
2270 }
2271 }
2272
2273 exit:
2274 error:
2275 /* Notify that the registration thread is gone */
2276 notify_ust_apps(0);
2277
2278 if (apps_sock >= 0) {
2279 ret = close(apps_sock);
2280 if (ret) {
2281 PERROR("close");
2282 }
2283 }
2284 if (sock >= 0) {
2285 ret = close(sock);
2286 if (ret) {
2287 PERROR("close");
2288 }
2289 lttng_fd_put(LTTNG_FD_APPS, 1);
2290 }
2291 unlink(apps_unix_sock_path);
2292
2293 error_poll_add:
2294 lttng_poll_clean(&events);
2295 error_listen:
2296 error_create_poll:
2297 error_testpoint:
2298 DBG("UST Registration thread cleanup complete");
2299 if (err) {
2300 health_error();
2301 ERR("Health error occurred in %s", __func__);
2302 }
2303 health_unregister(health_sessiond);
2304
2305 return NULL;
2306 }
2307
2308 /*
2309 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2310 * exec or it will fails.
2311 */
2312 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2313 {
2314 int ret, clock_ret;
2315 struct timespec timeout;
2316
2317 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2318 consumer_data->consumer_thread_is_ready = 0;
2319
2320 /* Setup pthread condition */
2321 ret = pthread_condattr_init(&consumer_data->condattr);
2322 if (ret) {
2323 errno = ret;
2324 PERROR("pthread_condattr_init consumer data");
2325 goto error;
2326 }
2327
2328 /*
2329 * Set the monotonic clock in order to make sure we DO NOT jump in time
2330 * between the clock_gettime() call and the timedwait call. See bug #324
2331 * for a more details and how we noticed it.
2332 */
2333 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2334 if (ret) {
2335 errno = ret;
2336 PERROR("pthread_condattr_setclock consumer data");
2337 goto error;
2338 }
2339
2340 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2341 if (ret) {
2342 errno = ret;
2343 PERROR("pthread_cond_init consumer data");
2344 goto error;
2345 }
2346
2347 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2348 consumer_data);
2349 if (ret) {
2350 errno = ret;
2351 PERROR("pthread_create consumer");
2352 ret = -1;
2353 goto error;
2354 }
2355
2356 /* We are about to wait on a pthread condition */
2357 pthread_mutex_lock(&consumer_data->cond_mutex);
2358
2359 /* Get time for sem_timedwait absolute timeout */
2360 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2361 /*
2362 * Set the timeout for the condition timed wait even if the clock gettime
2363 * call fails since we might loop on that call and we want to avoid to
2364 * increment the timeout too many times.
2365 */
2366 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2367
2368 /*
2369 * The following loop COULD be skipped in some conditions so this is why we
2370 * set ret to 0 in order to make sure at least one round of the loop is
2371 * done.
2372 */
2373 ret = 0;
2374
2375 /*
2376 * Loop until the condition is reached or when a timeout is reached. Note
2377 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2378 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2379 * possible. This loop does not take any chances and works with both of
2380 * them.
2381 */
2382 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2383 if (clock_ret < 0) {
2384 PERROR("clock_gettime spawn consumer");
2385 /* Infinite wait for the consumerd thread to be ready */
2386 ret = pthread_cond_wait(&consumer_data->cond,
2387 &consumer_data->cond_mutex);
2388 } else {
2389 ret = pthread_cond_timedwait(&consumer_data->cond,
2390 &consumer_data->cond_mutex, &timeout);
2391 }
2392 }
2393
2394 /* Release the pthread condition */
2395 pthread_mutex_unlock(&consumer_data->cond_mutex);
2396
2397 if (ret != 0) {
2398 errno = ret;
2399 if (ret == ETIMEDOUT) {
2400 int pth_ret;
2401
2402 /*
2403 * Call has timed out so we kill the kconsumerd_thread and return
2404 * an error.
2405 */
2406 ERR("Condition timed out. The consumer thread was never ready."
2407 " Killing it");
2408 pth_ret = pthread_cancel(consumer_data->thread);
2409 if (pth_ret < 0) {
2410 PERROR("pthread_cancel consumer thread");
2411 }
2412 } else {
2413 PERROR("pthread_cond_wait failed consumer thread");
2414 }
2415 /* Caller is expecting a negative value on failure. */
2416 ret = -1;
2417 goto error;
2418 }
2419
2420 pthread_mutex_lock(&consumer_data->pid_mutex);
2421 if (consumer_data->pid == 0) {
2422 ERR("Consumerd did not start");
2423 pthread_mutex_unlock(&consumer_data->pid_mutex);
2424 goto error;
2425 }
2426 pthread_mutex_unlock(&consumer_data->pid_mutex);
2427
2428 return 0;
2429
2430 error:
2431 return ret;
2432 }
2433
2434 /*
2435 * Join consumer thread
2436 */
2437 static int join_consumer_thread(struct consumer_data *consumer_data)
2438 {
2439 void *status;
2440
2441 /* Consumer pid must be a real one. */
2442 if (consumer_data->pid > 0) {
2443 int ret;
2444 ret = kill(consumer_data->pid, SIGTERM);
2445 if (ret) {
2446 PERROR("Error killing consumer daemon");
2447 return ret;
2448 }
2449 return pthread_join(consumer_data->thread, &status);
2450 } else {
2451 return 0;
2452 }
2453 }
2454
2455 /*
2456 * Fork and exec a consumer daemon (consumerd).
2457 *
2458 * Return pid if successful else -1.
2459 */
2460 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2461 {
2462 int ret;
2463 pid_t pid;
2464 const char *consumer_to_use;
2465 const char *verbosity;
2466 struct stat st;
2467
2468 DBG("Spawning consumerd");
2469
2470 pid = fork();
2471 if (pid == 0) {
2472 /*
2473 * Exec consumerd.
2474 */
2475 if (opt_verbose_consumer) {
2476 verbosity = "--verbose";
2477 } else if (lttng_opt_quiet) {
2478 verbosity = "--quiet";
2479 } else {
2480 verbosity = "";
2481 }
2482
2483 switch (consumer_data->type) {
2484 case LTTNG_CONSUMER_KERNEL:
2485 /*
2486 * Find out which consumerd to execute. We will first try the
2487 * 64-bit path, then the sessiond's installation directory, and
2488 * fallback on the 32-bit one,
2489 */
2490 DBG3("Looking for a kernel consumer at these locations:");
2491 DBG3(" 1) %s", consumerd64_bin);
2492 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2493 DBG3(" 3) %s", consumerd32_bin);
2494 if (stat(consumerd64_bin, &st) == 0) {
2495 DBG3("Found location #1");
2496 consumer_to_use = consumerd64_bin;
2497 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2498 DBG3("Found location #2");
2499 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2500 } else if (stat(consumerd32_bin, &st) == 0) {
2501 DBG3("Found location #3");
2502 consumer_to_use = consumerd32_bin;
2503 } else {
2504 DBG("Could not find any valid consumerd executable");
2505 ret = -EINVAL;
2506 break;
2507 }
2508 DBG("Using kernel consumer at: %s", consumer_to_use);
2509 ret = execl(consumer_to_use,
2510 "lttng-consumerd", verbosity, "-k",
2511 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2512 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2513 "--group", tracing_group_name,
2514 NULL);
2515 break;
2516 case LTTNG_CONSUMER64_UST:
2517 {
2518 char *tmpnew = NULL;
2519
2520 if (consumerd64_libdir[0] != '\0') {
2521 char *tmp;
2522 size_t tmplen;
2523
2524 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2525 if (!tmp) {
2526 tmp = "";
2527 }
2528 tmplen = strlen("LD_LIBRARY_PATH=")
2529 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2530 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2531 if (!tmpnew) {
2532 ret = -ENOMEM;
2533 goto error;
2534 }
2535 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2536 strcat(tmpnew, consumerd64_libdir);
2537 if (tmp[0] != '\0') {
2538 strcat(tmpnew, ":");
2539 strcat(tmpnew, tmp);
2540 }
2541 ret = putenv(tmpnew);
2542 if (ret) {
2543 ret = -errno;
2544 free(tmpnew);
2545 goto error;
2546 }
2547 }
2548 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2549 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2550 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2551 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2552 "--group", tracing_group_name,
2553 NULL);
2554 if (consumerd64_libdir[0] != '\0') {
2555 free(tmpnew);
2556 }
2557 break;
2558 }
2559 case LTTNG_CONSUMER32_UST:
2560 {
2561 char *tmpnew = NULL;
2562
2563 if (consumerd32_libdir[0] != '\0') {
2564 char *tmp;
2565 size_t tmplen;
2566
2567 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2568 if (!tmp) {
2569 tmp = "";
2570 }
2571 tmplen = strlen("LD_LIBRARY_PATH=")
2572 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2573 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2574 if (!tmpnew) {
2575 ret = -ENOMEM;
2576 goto error;
2577 }
2578 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2579 strcat(tmpnew, consumerd32_libdir);
2580 if (tmp[0] != '\0') {
2581 strcat(tmpnew, ":");
2582 strcat(tmpnew, tmp);
2583 }
2584 ret = putenv(tmpnew);
2585 if (ret) {
2586 ret = -errno;
2587 free(tmpnew);
2588 goto error;
2589 }
2590 }
2591 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2592 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2593 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2594 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2595 "--group", tracing_group_name,
2596 NULL);
2597 if (consumerd32_libdir[0] != '\0') {
2598 free(tmpnew);
2599 }
2600 break;
2601 }
2602 default:
2603 PERROR("unknown consumer type");
2604 exit(EXIT_FAILURE);
2605 }
2606 if (errno != 0) {
2607 PERROR("Consumer execl()");
2608 }
2609 /* Reaching this point, we got a failure on our execl(). */
2610 exit(EXIT_FAILURE);
2611 } else if (pid > 0) {
2612 ret = pid;
2613 } else {
2614 PERROR("start consumer fork");
2615 ret = -errno;
2616 }
2617 error:
2618 return ret;
2619 }
2620
2621 /*
2622 * Spawn the consumerd daemon and session daemon thread.
2623 */
2624 static int start_consumerd(struct consumer_data *consumer_data)
2625 {
2626 int ret;
2627
2628 /*
2629 * Set the listen() state on the socket since there is a possible race
2630 * between the exec() of the consumer daemon and this call if place in the
2631 * consumer thread. See bug #366 for more details.
2632 */
2633 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2634 if (ret < 0) {
2635 goto error;
2636 }
2637
2638 pthread_mutex_lock(&consumer_data->pid_mutex);
2639 if (consumer_data->pid != 0) {
2640 pthread_mutex_unlock(&consumer_data->pid_mutex);
2641 goto end;
2642 }
2643
2644 ret = spawn_consumerd(consumer_data);
2645 if (ret < 0) {
2646 ERR("Spawning consumerd failed");
2647 pthread_mutex_unlock(&consumer_data->pid_mutex);
2648 goto error;
2649 }
2650
2651 /* Setting up the consumer_data pid */
2652 consumer_data->pid = ret;
2653 DBG2("Consumer pid %d", consumer_data->pid);
2654 pthread_mutex_unlock(&consumer_data->pid_mutex);
2655
2656 DBG2("Spawning consumer control thread");
2657 ret = spawn_consumer_thread(consumer_data);
2658 if (ret < 0) {
2659 ERR("Fatal error spawning consumer control thread");
2660 goto error;
2661 }
2662
2663 end:
2664 return 0;
2665
2666 error:
2667 /* Cleanup already created sockets on error. */
2668 if (consumer_data->err_sock >= 0) {
2669 int err;
2670
2671 err = close(consumer_data->err_sock);
2672 if (err < 0) {
2673 PERROR("close consumer data error socket");
2674 }
2675 }
2676 return ret;
2677 }
2678
2679 /*
2680 * Setup necessary data for kernel tracer action.
2681 */
2682 static int init_kernel_tracer(void)
2683 {
2684 int ret;
2685
2686 /* Modprobe lttng kernel modules */
2687 ret = modprobe_lttng_control();
2688 if (ret < 0) {
2689 goto error;
2690 }
2691
2692 /* Open debugfs lttng */
2693 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2694 if (kernel_tracer_fd < 0) {
2695 DBG("Failed to open %s", module_proc_lttng);
2696 ret = -1;
2697 goto error_open;
2698 }
2699
2700 /* Validate kernel version */
2701 ret = kernel_validate_version(kernel_tracer_fd);
2702 if (ret < 0) {
2703 goto error_version;
2704 }
2705
2706 ret = modprobe_lttng_data();
2707 if (ret < 0) {
2708 goto error_modules;
2709 }
2710
2711 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2712 return 0;
2713
2714 error_version:
2715 modprobe_remove_lttng_control();
2716 ret = close(kernel_tracer_fd);
2717 if (ret) {
2718 PERROR("close");
2719 }
2720 kernel_tracer_fd = -1;
2721 return LTTNG_ERR_KERN_VERSION;
2722
2723 error_modules:
2724 ret = close(kernel_tracer_fd);
2725 if (ret) {
2726 PERROR("close");
2727 }
2728
2729 error_open:
2730 modprobe_remove_lttng_control();
2731
2732 error:
2733 WARN("No kernel tracer available");
2734 kernel_tracer_fd = -1;
2735 if (!is_root) {
2736 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2737 } else {
2738 return LTTNG_ERR_KERN_NA;
2739 }
2740 }
2741
2742
2743 /*
2744 * Copy consumer output from the tracing session to the domain session. The
2745 * function also applies the right modification on a per domain basis for the
2746 * trace files destination directory.
2747 *
2748 * Should *NOT* be called with RCU read-side lock held.
2749 */
2750 static int copy_session_consumer(int domain, struct ltt_session *session)
2751 {
2752 int ret;
2753 const char *dir_name;
2754 struct consumer_output *consumer;
2755
2756 assert(session);
2757 assert(session->consumer);
2758
2759 switch (domain) {
2760 case LTTNG_DOMAIN_KERNEL:
2761 DBG3("Copying tracing session consumer output in kernel session");
2762 /*
2763 * XXX: We should audit the session creation and what this function
2764 * does "extra" in order to avoid a destroy since this function is used
2765 * in the domain session creation (kernel and ust) only. Same for UST
2766 * domain.
2767 */
2768 if (session->kernel_session->consumer) {
2769 consumer_output_put(session->kernel_session->consumer);
2770 }
2771 session->kernel_session->consumer =
2772 consumer_copy_output(session->consumer);
2773 /* Ease our life a bit for the next part */
2774 consumer = session->kernel_session->consumer;
2775 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2776 break;
2777 case LTTNG_DOMAIN_JUL:
2778 case LTTNG_DOMAIN_LOG4J:
2779 case LTTNG_DOMAIN_PYTHON:
2780 case LTTNG_DOMAIN_UST:
2781 DBG3("Copying tracing session consumer output in UST session");
2782 if (session->ust_session->consumer) {
2783 consumer_output_put(session->ust_session->consumer);
2784 }
2785 session->ust_session->consumer =
2786 consumer_copy_output(session->consumer);
2787 /* Ease our life a bit for the next part */
2788 consumer = session->ust_session->consumer;
2789 dir_name = DEFAULT_UST_TRACE_DIR;
2790 break;
2791 default:
2792 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2793 goto error;
2794 }
2795
2796 /* Append correct directory to subdir */
2797 strncat(consumer->subdir, dir_name,
2798 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2799 DBG3("Copy session consumer subdir %s", consumer->subdir);
2800
2801 ret = LTTNG_OK;
2802
2803 error:
2804 return ret;
2805 }
2806
2807 /*
2808 * Create an UST session and add it to the session ust list.
2809 *
2810 * Should *NOT* be called with RCU read-side lock held.
2811 */
2812 static int create_ust_session(struct ltt_session *session,
2813 struct lttng_domain *domain)
2814 {
2815 int ret;
2816 struct ltt_ust_session *lus = NULL;
2817
2818 assert(session);
2819 assert(domain);
2820 assert(session->consumer);
2821
2822 switch (domain->type) {
2823 case LTTNG_DOMAIN_JUL:
2824 case LTTNG_DOMAIN_LOG4J:
2825 case LTTNG_DOMAIN_PYTHON:
2826 case LTTNG_DOMAIN_UST:
2827 break;
2828 default:
2829 ERR("Unknown UST domain on create session %d", domain->type);
2830 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2831 goto error;
2832 }
2833
2834 DBG("Creating UST session");
2835
2836 lus = trace_ust_create_session(session->id);
2837 if (lus == NULL) {
2838 ret = LTTNG_ERR_UST_SESS_FAIL;
2839 goto error;
2840 }
2841
2842 lus->uid = session->uid;
2843 lus->gid = session->gid;
2844 lus->output_traces = session->output_traces;
2845 lus->snapshot_mode = session->snapshot_mode;
2846 lus->live_timer_interval = session->live_timer;
2847 session->ust_session = lus;
2848 if (session->shm_path[0]) {
2849 strncpy(lus->root_shm_path, session->shm_path,
2850 sizeof(lus->root_shm_path));
2851 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2852 strncpy(lus->shm_path, session->shm_path,
2853 sizeof(lus->shm_path));
2854 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2855 strncat(lus->shm_path, "/ust",
2856 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2857 }
2858 /* Copy session output to the newly created UST session */
2859 ret = copy_session_consumer(domain->type, session);
2860 if (ret != LTTNG_OK) {
2861 goto error;
2862 }
2863
2864 return LTTNG_OK;
2865
2866 error:
2867 free(lus);
2868 session->ust_session = NULL;
2869 return ret;
2870 }
2871
2872 /*
2873 * Create a kernel tracer session then create the default channel.
2874 */
2875 static int create_kernel_session(struct ltt_session *session)
2876 {
2877 int ret;
2878
2879 DBG("Creating kernel session");
2880
2881 ret = kernel_create_session(session, kernel_tracer_fd);
2882 if (ret < 0) {
2883 ret = LTTNG_ERR_KERN_SESS_FAIL;
2884 goto error;
2885 }
2886
2887 /* Code flow safety */
2888 assert(session->kernel_session);
2889
2890 /* Copy session output to the newly created Kernel session */
2891 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2892 if (ret != LTTNG_OK) {
2893 goto error;
2894 }
2895
2896 /* Create directory(ies) on local filesystem. */
2897 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2898 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2899 ret = run_as_mkdir_recursive(
2900 session->kernel_session->consumer->dst.trace_path,
2901 S_IRWXU | S_IRWXG, session->uid, session->gid);
2902 if (ret < 0) {
2903 if (errno != EEXIST) {
2904 ERR("Trace directory creation error");
2905 goto error;
2906 }
2907 }
2908 }
2909
2910 session->kernel_session->uid = session->uid;
2911 session->kernel_session->gid = session->gid;
2912 session->kernel_session->output_traces = session->output_traces;
2913 session->kernel_session->snapshot_mode = session->snapshot_mode;
2914
2915 return LTTNG_OK;
2916
2917 error:
2918 trace_kernel_destroy_session(session->kernel_session);
2919 session->kernel_session = NULL;
2920 return ret;
2921 }
2922
2923 /*
2924 * Count number of session permitted by uid/gid.
2925 */
2926 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2927 {
2928 unsigned int i = 0;
2929 struct ltt_session *session;
2930
2931 DBG("Counting number of available session for UID %d GID %d",
2932 uid, gid);
2933 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2934 /*
2935 * Only list the sessions the user can control.
2936 */
2937 if (!session_access_ok(session, uid, gid)) {
2938 continue;
2939 }
2940 i++;
2941 }
2942 return i;
2943 }
2944
2945 /*
2946 * Process the command requested by the lttng client within the command
2947 * context structure. This function make sure that the return structure (llm)
2948 * is set and ready for transmission before returning.
2949 *
2950 * Return any error encountered or 0 for success.
2951 *
2952 * "sock" is only used for special-case var. len data.
2953 *
2954 * Should *NOT* be called with RCU read-side lock held.
2955 */
2956 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2957 int *sock_error)
2958 {
2959 int ret = LTTNG_OK;
2960 int need_tracing_session = 1;
2961 int need_domain;
2962
2963 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2964
2965 *sock_error = 0;
2966
2967 switch (cmd_ctx->lsm->cmd_type) {
2968 case LTTNG_CREATE_SESSION:
2969 case LTTNG_CREATE_SESSION_SNAPSHOT:
2970 case LTTNG_CREATE_SESSION_LIVE:
2971 case LTTNG_DESTROY_SESSION:
2972 case LTTNG_LIST_SESSIONS:
2973 case LTTNG_LIST_DOMAINS:
2974 case LTTNG_START_TRACE:
2975 case LTTNG_STOP_TRACE:
2976 case LTTNG_DATA_PENDING:
2977 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2978 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2979 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2980 case LTTNG_SNAPSHOT_RECORD:
2981 case LTTNG_SAVE_SESSION:
2982 case LTTNG_SET_SESSION_SHM_PATH:
2983 need_domain = 0;
2984 break;
2985 default:
2986 need_domain = 1;
2987 }
2988
2989 if (opt_no_kernel && need_domain
2990 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2991 if (!is_root) {
2992 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2993 } else {
2994 ret = LTTNG_ERR_KERN_NA;
2995 }
2996 goto error;
2997 }
2998
2999 /* Deny register consumer if we already have a spawned consumer. */
3000 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3001 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3002 if (kconsumer_data.pid > 0) {
3003 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3004 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3005 goto error;
3006 }
3007 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3008 }
3009
3010 /*
3011 * Check for command that don't needs to allocate a returned payload. We do
3012 * this here so we don't have to make the call for no payload at each
3013 * command.
3014 */
3015 switch(cmd_ctx->lsm->cmd_type) {
3016 case LTTNG_LIST_SESSIONS:
3017 case LTTNG_LIST_TRACEPOINTS:
3018 case LTTNG_LIST_TRACEPOINT_FIELDS:
3019 case LTTNG_LIST_DOMAINS:
3020 case LTTNG_LIST_CHANNELS:
3021 case LTTNG_LIST_EVENTS:
3022 case LTTNG_LIST_SYSCALLS:
3023 case LTTNG_LIST_TRACKER_PIDS:
3024 break;
3025 default:
3026 /* Setup lttng message with no payload */
3027 ret = setup_lttng_msg(cmd_ctx, 0);
3028 if (ret < 0) {
3029 /* This label does not try to unlock the session */
3030 goto init_setup_error;
3031 }
3032 }
3033
3034 /* Commands that DO NOT need a session. */
3035 switch (cmd_ctx->lsm->cmd_type) {
3036 case LTTNG_CREATE_SESSION:
3037 case LTTNG_CREATE_SESSION_SNAPSHOT:
3038 case LTTNG_CREATE_SESSION_LIVE:
3039 case LTTNG_CALIBRATE:
3040 case LTTNG_LIST_SESSIONS:
3041 case LTTNG_LIST_TRACEPOINTS:
3042 case LTTNG_LIST_SYSCALLS:
3043 case LTTNG_LIST_TRACEPOINT_FIELDS:
3044 case LTTNG_SAVE_SESSION:
3045 need_tracing_session = 0;
3046 break;
3047 default:
3048 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3049 /*
3050 * We keep the session list lock across _all_ commands
3051 * for now, because the per-session lock does not
3052 * handle teardown properly.
3053 */
3054 session_lock_list();
3055 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3056 if (cmd_ctx->session == NULL) {
3057 ret = LTTNG_ERR_SESS_NOT_FOUND;
3058 goto error;
3059 } else {
3060 /* Acquire lock for the session */
3061 session_lock(cmd_ctx->session);
3062 }
3063 break;
3064 }
3065
3066 /*
3067 * Commands that need a valid session but should NOT create one if none
3068 * exists. Instead of creating one and destroying it when the command is
3069 * handled, process that right before so we save some round trip in useless
3070 * code path.
3071 */
3072 switch (cmd_ctx->lsm->cmd_type) {
3073 case LTTNG_DISABLE_CHANNEL:
3074 case LTTNG_DISABLE_EVENT:
3075 switch (cmd_ctx->lsm->domain.type) {
3076 case LTTNG_DOMAIN_KERNEL:
3077 if (!cmd_ctx->session->kernel_session) {
3078 ret = LTTNG_ERR_NO_CHANNEL;
3079 goto error;
3080 }
3081 break;
3082 case LTTNG_DOMAIN_JUL:
3083 case LTTNG_DOMAIN_LOG4J:
3084 case LTTNG_DOMAIN_PYTHON:
3085 case LTTNG_DOMAIN_UST:
3086 if (!cmd_ctx->session->ust_session) {
3087 ret = LTTNG_ERR_NO_CHANNEL;
3088 goto error;
3089 }
3090 break;
3091 default:
3092 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3093 goto error;
3094 }
3095 default:
3096 break;
3097 }
3098
3099 if (!need_domain) {
3100 goto skip_domain;
3101 }
3102
3103 /*
3104 * Check domain type for specific "pre-action".
3105 */
3106 switch (cmd_ctx->lsm->domain.type) {
3107 case LTTNG_DOMAIN_KERNEL:
3108 if (!is_root) {
3109 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3110 goto error;
3111 }
3112
3113 /* Kernel tracer check */
3114 if (kernel_tracer_fd == -1) {
3115 /* Basically, load kernel tracer modules */
3116 ret = init_kernel_tracer();
3117 if (ret != 0) {
3118 goto error;
3119 }
3120 }
3121
3122 /* Consumer is in an ERROR state. Report back to client */
3123 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3124 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3125 goto error;
3126 }
3127
3128 /* Need a session for kernel command */
3129 if (need_tracing_session) {
3130 if (cmd_ctx->session->kernel_session == NULL) {
3131 ret = create_kernel_session(cmd_ctx->session);
3132 if (ret < 0) {
3133 ret = LTTNG_ERR_KERN_SESS_FAIL;
3134 goto error;
3135 }
3136 }
3137
3138 /* Start the kernel consumer daemon */
3139 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3140 if (kconsumer_data.pid == 0 &&
3141 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3142 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3143 ret = start_consumerd(&kconsumer_data);
3144 if (ret < 0) {
3145 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3146 goto error;
3147 }
3148 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3149 } else {
3150 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3151 }
3152
3153 /*
3154 * The consumer was just spawned so we need to add the socket to
3155 * the consumer output of the session if exist.
3156 */
3157 ret = consumer_create_socket(&kconsumer_data,
3158 cmd_ctx->session->kernel_session->consumer);
3159 if (ret < 0) {
3160 goto error;
3161 }
3162 }
3163
3164 break;
3165 case LTTNG_DOMAIN_JUL:
3166 case LTTNG_DOMAIN_LOG4J:
3167 case LTTNG_DOMAIN_PYTHON:
3168 case LTTNG_DOMAIN_UST:
3169 {
3170 if (!ust_app_supported()) {
3171 ret = LTTNG_ERR_NO_UST;
3172 goto error;
3173 }
3174 /* Consumer is in an ERROR state. Report back to client */
3175 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3176 ret = LTTNG_ERR_NO_USTCONSUMERD;
3177 goto error;
3178 }
3179
3180 if (need_tracing_session) {
3181 /* Create UST session if none exist. */
3182 if (cmd_ctx->session->ust_session == NULL) {
3183 ret = create_ust_session(cmd_ctx->session,
3184 &cmd_ctx->lsm->domain);
3185 if (ret != LTTNG_OK) {
3186 goto error;
3187 }
3188 }
3189
3190 /* Start the UST consumer daemons */
3191 /* 64-bit */
3192 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3193 if (consumerd64_bin[0] != '\0' &&
3194 ustconsumer64_data.pid == 0 &&
3195 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3196 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3197 ret = start_consumerd(&ustconsumer64_data);
3198 if (ret < 0) {
3199 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3200 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3201 goto error;
3202 }
3203
3204 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3205 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3206 } else {
3207 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3208 }
3209
3210 /*
3211 * Setup socket for consumer 64 bit. No need for atomic access
3212 * since it was set above and can ONLY be set in this thread.
3213 */
3214 ret = consumer_create_socket(&ustconsumer64_data,
3215 cmd_ctx->session->ust_session->consumer);
3216 if (ret < 0) {
3217 goto error;
3218 }
3219
3220 /* 32-bit */
3221 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3222 if (consumerd32_bin[0] != '\0' &&
3223 ustconsumer32_data.pid == 0 &&
3224 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3225 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3226 ret = start_consumerd(&ustconsumer32_data);
3227 if (ret < 0) {
3228 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3229 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3230 goto error;
3231 }
3232
3233 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3234 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3235 } else {
3236 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3237 }
3238
3239 /*
3240 * Setup socket for consumer 64 bit. No need for atomic access
3241 * since it was set above and can ONLY be set in this thread.
3242 */
3243 ret = consumer_create_socket(&ustconsumer32_data,
3244 cmd_ctx->session->ust_session->consumer);
3245 if (ret < 0) {
3246 goto error;
3247 }
3248 }
3249 break;
3250 }
3251 default:
3252 break;
3253 }
3254 skip_domain:
3255
3256 /* Validate consumer daemon state when start/stop trace command */
3257 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3258 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3259 switch (cmd_ctx->lsm->domain.type) {
3260 case LTTNG_DOMAIN_JUL:
3261 case LTTNG_DOMAIN_LOG4J:
3262 case LTTNG_DOMAIN_PYTHON:
3263 case LTTNG_DOMAIN_UST:
3264 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3265 ret = LTTNG_ERR_NO_USTCONSUMERD;
3266 goto error;
3267 }
3268 break;
3269 case LTTNG_DOMAIN_KERNEL:
3270 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3271 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3272 goto error;
3273 }
3274 break;
3275 }
3276 }
3277
3278 /*
3279 * Check that the UID or GID match that of the tracing session.
3280 * The root user can interact with all sessions.
3281 */
3282 if (need_tracing_session) {
3283 if (!session_access_ok(cmd_ctx->session,
3284 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3285 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3286 ret = LTTNG_ERR_EPERM;
3287 goto error;
3288 }
3289 }
3290
3291 /*
3292 * Send relayd information to consumer as soon as we have a domain and a
3293 * session defined.
3294 */
3295 if (cmd_ctx->session && need_domain) {
3296 /*
3297 * Setup relayd if not done yet. If the relayd information was already
3298 * sent to the consumer, this call will gracefully return.
3299 */
3300 ret = cmd_setup_relayd(cmd_ctx->session);
3301 if (ret != LTTNG_OK) {
3302 goto error;
3303 }
3304 }
3305
3306 /* Process by command type */
3307 switch (cmd_ctx->lsm->cmd_type) {
3308 case LTTNG_ADD_CONTEXT:
3309 {
3310 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3311 cmd_ctx->lsm->u.context.channel_name,
3312 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3313 break;
3314 }
3315 case LTTNG_DISABLE_CHANNEL:
3316 {
3317 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3318 cmd_ctx->lsm->u.disable.channel_name);
3319 break;
3320 }
3321 case LTTNG_DISABLE_EVENT:
3322 {
3323
3324 /*
3325 * FIXME: handle filter; for now we just receive the filter's
3326 * bytecode along with the filter expression which are sent by
3327 * liblttng-ctl and discard them.
3328 *
3329 * This fixes an issue where the client may block while sending
3330 * the filter payload and encounter an error because the session
3331 * daemon closes the socket without ever handling this data.
3332 */
3333 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3334 cmd_ctx->lsm->u.disable.bytecode_len;
3335
3336 if (count) {
3337 char data[LTTNG_FILTER_MAX_LEN];
3338
3339 DBG("Discarding disable event command payload of size %zu", count);
3340 while (count) {
3341 ret = lttcomm_recv_unix_sock(sock, data,
3342 count > sizeof(data) ? sizeof(data) : count);
3343 if (ret < 0) {
3344 goto error;
3345 }
3346
3347 count -= (size_t) ret;
3348 }
3349 }
3350 /* FIXME: passing packed structure to non-packed pointer */
3351 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3352 cmd_ctx->lsm->u.disable.channel_name,
3353 &cmd_ctx->lsm->u.disable.event);
3354 break;
3355 }
3356 case LTTNG_ENABLE_CHANNEL:
3357 {
3358 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3359 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3360 break;
3361 }
3362 case LTTNG_TRACK_PID:
3363 {
3364 ret = cmd_track_pid(cmd_ctx->session,
3365 cmd_ctx->lsm->domain.type,
3366 cmd_ctx->lsm->u.pid_tracker.pid);
3367 break;
3368 }
3369 case LTTNG_UNTRACK_PID:
3370 {
3371 ret = cmd_untrack_pid(cmd_ctx->session,
3372 cmd_ctx->lsm->domain.type,
3373 cmd_ctx->lsm->u.pid_tracker.pid);
3374 break;
3375 }
3376 case LTTNG_ENABLE_EVENT:
3377 {
3378 struct lttng_event_exclusion *exclusion = NULL;
3379 struct lttng_filter_bytecode *bytecode = NULL;
3380 char *filter_expression = NULL;
3381
3382 /* Handle exclusion events and receive it from the client. */
3383 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3384 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3385
3386 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3387 (count * LTTNG_SYMBOL_NAME_LEN));
3388 if (!exclusion) {
3389 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3390 goto error;
3391 }
3392
3393 DBG("Receiving var len exclusion event list from client ...");
3394 exclusion->count = count;
3395 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3396 count * LTTNG_SYMBOL_NAME_LEN);
3397 if (ret <= 0) {
3398 DBG("Nothing recv() from client var len data... continuing");
3399 *sock_error = 1;
3400 free(exclusion);
3401 ret = LTTNG_ERR_EXCLUSION_INVAL;
3402 goto error;
3403 }
3404 }
3405
3406 /* Get filter expression from client. */
3407 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3408 size_t expression_len =
3409 cmd_ctx->lsm->u.enable.expression_len;
3410
3411 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3412 ret = LTTNG_ERR_FILTER_INVAL;
3413 free(exclusion);
3414 goto error;
3415 }
3416
3417 filter_expression = zmalloc(expression_len);
3418 if (!filter_expression) {
3419 free(exclusion);
3420 ret = LTTNG_ERR_FILTER_NOMEM;
3421 goto error;
3422 }
3423
3424 /* Receive var. len. data */
3425 DBG("Receiving var len filter's expression from client ...");
3426 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3427 expression_len);
3428 if (ret <= 0) {
3429 DBG("Nothing recv() from client car len data... continuing");
3430 *sock_error = 1;
3431 free(filter_expression);
3432 free(exclusion);
3433 ret = LTTNG_ERR_FILTER_INVAL;
3434 goto error;
3435 }
3436 }
3437
3438 /* Handle filter and get bytecode from client. */
3439 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3440 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3441
3442 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3443 ret = LTTNG_ERR_FILTER_INVAL;
3444 free(filter_expression);
3445 free(exclusion);
3446 goto error;
3447 }
3448
3449 bytecode = zmalloc(bytecode_len);
3450 if (!bytecode) {
3451 free(filter_expression);
3452 free(exclusion);
3453 ret = LTTNG_ERR_FILTER_NOMEM;
3454 goto error;
3455 }
3456
3457 /* Receive var. len. data */
3458 DBG("Receiving var len filter's bytecode from client ...");
3459 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3460 if (ret <= 0) {
3461 DBG("Nothing recv() from client car len data... continuing");
3462 *sock_error = 1;
3463 free(filter_expression);
3464 free(bytecode);
3465 free(exclusion);
3466 ret = LTTNG_ERR_FILTER_INVAL;
3467 goto error;
3468 }
3469
3470 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3471 free(filter_expression);
3472 free(bytecode);
3473 free(exclusion);
3474 ret = LTTNG_ERR_FILTER_INVAL;
3475 goto error;
3476 }
3477 }
3478
3479 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3480 cmd_ctx->lsm->u.enable.channel_name,
3481 &cmd_ctx->lsm->u.enable.event,
3482 filter_expression, bytecode, exclusion,
3483 kernel_poll_pipe[1]);
3484 break;
3485 }
3486 case LTTNG_LIST_TRACEPOINTS:
3487 {
3488 struct lttng_event *events;
3489 ssize_t nb_events;
3490
3491 session_lock_list();
3492 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3493 session_unlock_list();
3494 if (nb_events < 0) {
3495 /* Return value is a negative lttng_error_code. */
3496 ret = -nb_events;
3497 goto error;
3498 }
3499
3500 /*
3501 * Setup lttng message with payload size set to the event list size in
3502 * bytes and then copy list into the llm payload.
3503 */
3504 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3505 if (ret < 0) {
3506 free(events);
3507 goto setup_error;
3508 }
3509
3510 /* Copy event list into message payload */
3511 memcpy(cmd_ctx->llm->payload, events,
3512 sizeof(struct lttng_event) * nb_events);
3513
3514 free(events);
3515
3516 ret = LTTNG_OK;
3517 break;
3518 }
3519 case LTTNG_LIST_TRACEPOINT_FIELDS:
3520 {
3521 struct lttng_event_field *fields;
3522 ssize_t nb_fields;
3523
3524 session_lock_list();
3525 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3526 &fields);
3527 session_unlock_list();
3528 if (nb_fields < 0) {
3529 /* Return value is a negative lttng_error_code. */
3530 ret = -nb_fields;
3531 goto error;
3532 }
3533
3534 /*
3535 * Setup lttng message with payload size set to the event list size in
3536 * bytes and then copy list into the llm payload.
3537 */
3538 ret = setup_lttng_msg(cmd_ctx,
3539 sizeof(struct lttng_event_field) * nb_fields);
3540 if (ret < 0) {
3541 free(fields);
3542 goto setup_error;
3543 }
3544
3545 /* Copy event list into message payload */
3546 memcpy(cmd_ctx->llm->payload, fields,
3547 sizeof(struct lttng_event_field) * nb_fields);
3548
3549 free(fields);
3550
3551 ret = LTTNG_OK;
3552 break;
3553 }
3554 case LTTNG_LIST_SYSCALLS:
3555 {
3556 struct lttng_event *events;
3557 ssize_t nb_events;
3558
3559 nb_events = cmd_list_syscalls(&events);
3560 if (nb_events < 0) {
3561 /* Return value is a negative lttng_error_code. */
3562 ret = -nb_events;
3563 goto error;
3564 }
3565
3566 /*
3567 * Setup lttng message with payload size set to the event list size in
3568 * bytes and then copy list into the llm payload.
3569 */
3570 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3571 if (ret < 0) {
3572 free(events);
3573 goto setup_error;
3574 }
3575
3576 /* Copy event list into message payload */
3577 memcpy(cmd_ctx->llm->payload, events,
3578 sizeof(struct lttng_event) * nb_events);
3579
3580 free(events);
3581
3582 ret = LTTNG_OK;
3583 break;
3584 }
3585 case LTTNG_LIST_TRACKER_PIDS:
3586 {
3587 int32_t *pids = NULL;
3588 ssize_t nr_pids;
3589
3590 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3591 cmd_ctx->lsm->domain.type, &pids);
3592 if (nr_pids < 0) {
3593 /* Return value is a negative lttng_error_code. */
3594 ret = -nr_pids;
3595 goto error;
3596 }
3597
3598 /*
3599 * Setup lttng message with payload size set to the event list size in
3600 * bytes and then copy list into the llm payload.
3601 */
3602 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3603 if (ret < 0) {
3604 free(pids);
3605 goto setup_error;
3606 }
3607
3608 /* Copy event list into message payload */
3609 memcpy(cmd_ctx->llm->payload, pids,
3610 sizeof(int) * nr_pids);
3611
3612 free(pids);
3613
3614 ret = LTTNG_OK;
3615 break;
3616 }
3617 case LTTNG_SET_CONSUMER_URI:
3618 {
3619 size_t nb_uri, len;
3620 struct lttng_uri *uris;
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 ret = LTTNG_ERR_INVALID;
3627 goto error;
3628 }
3629
3630 uris = zmalloc(len);
3631 if (uris == NULL) {
3632 ret = LTTNG_ERR_FATAL;
3633 goto error;
3634 }
3635
3636 /* Receive variable len data */
3637 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3638 ret = lttcomm_recv_unix_sock(sock, uris, len);
3639 if (ret <= 0) {
3640 DBG("No URIs received from client... continuing");
3641 *sock_error = 1;
3642 ret = LTTNG_ERR_SESSION_FAIL;
3643 free(uris);
3644 goto error;
3645 }
3646
3647 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3648 free(uris);
3649 if (ret != LTTNG_OK) {
3650 goto error;
3651 }
3652
3653
3654 break;
3655 }
3656 case LTTNG_START_TRACE:
3657 {
3658 ret = cmd_start_trace(cmd_ctx->session);
3659 break;
3660 }
3661 case LTTNG_STOP_TRACE:
3662 {
3663 ret = cmd_stop_trace(cmd_ctx->session);
3664 break;
3665 }
3666 case LTTNG_CREATE_SESSION:
3667 {
3668 size_t nb_uri, len;
3669 struct lttng_uri *uris = NULL;
3670
3671 nb_uri = cmd_ctx->lsm->u.uri.size;
3672 len = nb_uri * sizeof(struct lttng_uri);
3673
3674 if (nb_uri > 0) {
3675 uris = zmalloc(len);
3676 if (uris == NULL) {
3677 ret = LTTNG_ERR_FATAL;
3678 goto error;
3679 }
3680
3681 /* Receive variable len data */
3682 DBG("Waiting for %zu URIs from client ...", nb_uri);
3683 ret = lttcomm_recv_unix_sock(sock, uris, len);
3684 if (ret <= 0) {
3685 DBG("No URIs received from client... continuing");
3686 *sock_error = 1;
3687 ret = LTTNG_ERR_SESSION_FAIL;
3688 free(uris);
3689 goto error;
3690 }
3691
3692 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3693 DBG("Creating session with ONE network URI is a bad call");
3694 ret = LTTNG_ERR_SESSION_FAIL;
3695 free(uris);
3696 goto error;
3697 }
3698 }
3699
3700 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3701 &cmd_ctx->creds, 0);
3702
3703 free(uris);
3704
3705 break;
3706 }
3707 case LTTNG_DESTROY_SESSION:
3708 {
3709 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3710
3711 /* Set session to NULL so we do not unlock it after free. */
3712 cmd_ctx->session = NULL;
3713 break;
3714 }
3715 case LTTNG_LIST_DOMAINS:
3716 {
3717 ssize_t nb_dom;
3718 struct lttng_domain *domains = NULL;
3719
3720 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3721 if (nb_dom < 0) {
3722 /* Return value is a negative lttng_error_code. */
3723 ret = -nb_dom;
3724 goto error;
3725 }
3726
3727 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3728 if (ret < 0) {
3729 free(domains);
3730 goto setup_error;
3731 }
3732
3733 /* Copy event list into message payload */
3734 memcpy(cmd_ctx->llm->payload, domains,
3735 nb_dom * sizeof(struct lttng_domain));
3736
3737 free(domains);
3738
3739 ret = LTTNG_OK;
3740 break;
3741 }
3742 case LTTNG_LIST_CHANNELS:
3743 {
3744 int nb_chan;
3745 struct lttng_channel *channels = NULL;
3746
3747 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3748 cmd_ctx->session, &channels);
3749 if (nb_chan < 0) {
3750 /* Return value is a negative lttng_error_code. */
3751 ret = -nb_chan;
3752 goto error;
3753 }
3754
3755 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3756 if (ret < 0) {
3757 free(channels);
3758 goto setup_error;
3759 }
3760
3761 /* Copy event list into message payload */
3762 memcpy(cmd_ctx->llm->payload, channels,
3763 nb_chan * sizeof(struct lttng_channel));
3764
3765 free(channels);
3766
3767 ret = LTTNG_OK;
3768 break;
3769 }
3770 case LTTNG_LIST_EVENTS:
3771 {
3772 ssize_t nb_event;
3773 struct lttng_event *events = NULL;
3774
3775 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3776 cmd_ctx->lsm->u.list.channel_name, &events);
3777 if (nb_event < 0) {
3778 /* Return value is a negative lttng_error_code. */
3779 ret = -nb_event;
3780 goto error;
3781 }
3782
3783 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3784 if (ret < 0) {
3785 free(events);
3786 goto setup_error;
3787 }
3788
3789 /* Copy event list into message payload */
3790 memcpy(cmd_ctx->llm->payload, events,
3791 nb_event * sizeof(struct lttng_event));
3792
3793 free(events);
3794
3795 ret = LTTNG_OK;
3796 break;
3797 }
3798 case LTTNG_LIST_SESSIONS:
3799 {
3800 unsigned int nr_sessions;
3801
3802 session_lock_list();
3803 nr_sessions = lttng_sessions_count(
3804 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3805 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3806
3807 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3808 if (ret < 0) {
3809 session_unlock_list();
3810 goto setup_error;
3811 }
3812
3813 /* Filled the session array */
3814 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3815 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3816 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3817
3818 session_unlock_list();
3819
3820 ret = LTTNG_OK;
3821 break;
3822 }
3823 case LTTNG_CALIBRATE:
3824 {
3825 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3826 &cmd_ctx->lsm->u.calibrate);
3827 break;
3828 }
3829 case LTTNG_REGISTER_CONSUMER:
3830 {
3831 struct consumer_data *cdata;
3832
3833 switch (cmd_ctx->lsm->domain.type) {
3834 case LTTNG_DOMAIN_KERNEL:
3835 cdata = &kconsumer_data;
3836 break;
3837 default:
3838 ret = LTTNG_ERR_UND;
3839 goto error;
3840 }
3841
3842 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3843 cmd_ctx->lsm->u.reg.path, cdata);
3844 break;
3845 }
3846 case LTTNG_DATA_PENDING:
3847 {
3848 ret = cmd_data_pending(cmd_ctx->session);
3849 break;
3850 }
3851 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3852 {
3853 struct lttcomm_lttng_output_id reply;
3854
3855 ret = cmd_snapshot_add_output(cmd_ctx->session,
3856 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3857 if (ret != LTTNG_OK) {
3858 goto error;
3859 }
3860
3861 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3862 if (ret < 0) {
3863 goto setup_error;
3864 }
3865
3866 /* Copy output list into message payload */
3867 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3868 ret = LTTNG_OK;
3869 break;
3870 }
3871 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3872 {
3873 ret = cmd_snapshot_del_output(cmd_ctx->session,
3874 &cmd_ctx->lsm->u.snapshot_output.output);
3875 break;
3876 }
3877 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3878 {
3879 ssize_t nb_output;
3880 struct lttng_snapshot_output *outputs = NULL;
3881
3882 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3883 if (nb_output < 0) {
3884 ret = -nb_output;
3885 goto error;
3886 }
3887
3888 ret = setup_lttng_msg(cmd_ctx,
3889 nb_output * sizeof(struct lttng_snapshot_output));
3890 if (ret < 0) {
3891 free(outputs);
3892 goto setup_error;
3893 }
3894
3895 if (outputs) {
3896 /* Copy output list into message payload */
3897 memcpy(cmd_ctx->llm->payload, outputs,
3898 nb_output * sizeof(struct lttng_snapshot_output));
3899 free(outputs);
3900 }
3901
3902 ret = LTTNG_OK;
3903 break;
3904 }
3905 case LTTNG_SNAPSHOT_RECORD:
3906 {
3907 ret = cmd_snapshot_record(cmd_ctx->session,
3908 &cmd_ctx->lsm->u.snapshot_record.output,
3909 cmd_ctx->lsm->u.snapshot_record.wait);
3910 break;
3911 }
3912 case LTTNG_CREATE_SESSION_SNAPSHOT:
3913 {
3914 size_t nb_uri, len;
3915 struct lttng_uri *uris = NULL;
3916
3917 nb_uri = cmd_ctx->lsm->u.uri.size;
3918 len = nb_uri * sizeof(struct lttng_uri);
3919
3920 if (nb_uri > 0) {
3921 uris = zmalloc(len);
3922 if (uris == NULL) {
3923 ret = LTTNG_ERR_FATAL;
3924 goto error;
3925 }
3926
3927 /* Receive variable len data */
3928 DBG("Waiting for %zu URIs from client ...", nb_uri);
3929 ret = lttcomm_recv_unix_sock(sock, uris, len);
3930 if (ret <= 0) {
3931 DBG("No URIs received from client... continuing");
3932 *sock_error = 1;
3933 ret = LTTNG_ERR_SESSION_FAIL;
3934 free(uris);
3935 goto error;
3936 }
3937
3938 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3939 DBG("Creating session with ONE network URI is a bad call");
3940 ret = LTTNG_ERR_SESSION_FAIL;
3941 free(uris);
3942 goto error;
3943 }
3944 }
3945
3946 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3947 nb_uri, &cmd_ctx->creds);
3948 free(uris);
3949 break;
3950 }
3951 case LTTNG_CREATE_SESSION_LIVE:
3952 {
3953 size_t nb_uri, len;
3954 struct lttng_uri *uris = NULL;
3955
3956 nb_uri = cmd_ctx->lsm->u.uri.size;
3957 len = nb_uri * sizeof(struct lttng_uri);
3958
3959 if (nb_uri > 0) {
3960 uris = zmalloc(len);
3961 if (uris == NULL) {
3962 ret = LTTNG_ERR_FATAL;
3963 goto error;
3964 }
3965
3966 /* Receive variable len data */
3967 DBG("Waiting for %zu URIs from client ...", nb_uri);
3968 ret = lttcomm_recv_unix_sock(sock, uris, len);
3969 if (ret <= 0) {
3970 DBG("No URIs received from client... continuing");
3971 *sock_error = 1;
3972 ret = LTTNG_ERR_SESSION_FAIL;
3973 free(uris);
3974 goto error;
3975 }
3976
3977 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3978 DBG("Creating session with ONE network URI is a bad call");
3979 ret = LTTNG_ERR_SESSION_FAIL;
3980 free(uris);
3981 goto error;
3982 }
3983 }
3984
3985 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3986 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3987 free(uris);
3988 break;
3989 }
3990 case LTTNG_SAVE_SESSION:
3991 {
3992 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3993 &cmd_ctx->creds);
3994 break;
3995 }
3996 case LTTNG_SET_SESSION_SHM_PATH:
3997 {
3998 ret = cmd_set_session_shm_path(cmd_ctx->session,
3999 cmd_ctx->lsm->u.set_shm_path.shm_path);
4000 break;
4001 }
4002 default:
4003 ret = LTTNG_ERR_UND;
4004 break;
4005 }
4006
4007 error:
4008 if (cmd_ctx->llm == NULL) {
4009 DBG("Missing llm structure. Allocating one.");
4010 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
4011 goto setup_error;
4012 }
4013 }
4014 /* Set return code */
4015 cmd_ctx->llm->ret_code = ret;
4016 setup_error:
4017 if (cmd_ctx->session) {
4018 session_unlock(cmd_ctx->session);
4019 }
4020 if (need_tracing_session) {
4021 session_unlock_list();
4022 }
4023 init_setup_error:
4024 return ret;
4025 }
4026
4027 /*
4028 * Thread managing health check socket.
4029 */
4030 static void *thread_manage_health(void *data)
4031 {
4032 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4033 uint32_t revents, nb_fd;
4034 struct lttng_poll_event events;
4035 struct health_comm_msg msg;
4036 struct health_comm_reply reply;
4037
4038 DBG("[thread] Manage health check started");
4039
4040 rcu_register_thread();
4041
4042 /* We might hit an error path before this is created. */
4043 lttng_poll_init(&events);
4044
4045 /* Create unix socket */
4046 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4047 if (sock < 0) {
4048 ERR("Unable to create health check Unix socket");
4049 ret = -1;
4050 goto error;
4051 }
4052
4053 if (is_root) {
4054 /* lttng health client socket path permissions */
4055 ret = chown(health_unix_sock_path, 0,
4056 utils_get_group_id(tracing_group_name));
4057 if (ret < 0) {
4058 ERR("Unable to set group on %s", health_unix_sock_path);
4059 PERROR("chown");
4060 ret = -1;
4061 goto error;
4062 }
4063
4064 ret = chmod(health_unix_sock_path,
4065 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4066 if (ret < 0) {
4067 ERR("Unable to set permissions on %s", health_unix_sock_path);
4068 PERROR("chmod");
4069 ret = -1;
4070 goto error;
4071 }
4072 }
4073
4074 /*
4075 * Set the CLOEXEC flag. Return code is useless because either way, the
4076 * show must go on.
4077 */
4078 (void) utils_set_fd_cloexec(sock);
4079
4080 ret = lttcomm_listen_unix_sock(sock);
4081 if (ret < 0) {
4082 goto error;
4083 }
4084
4085 /*
4086 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4087 * more will be added to this poll set.
4088 */
4089 ret = sessiond_set_thread_pollset(&events, 2);
4090 if (ret < 0) {
4091 goto error;
4092 }
4093
4094 /* Add the application registration socket */
4095 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4096 if (ret < 0) {
4097 goto error;
4098 }
4099
4100 sessiond_notify_ready();
4101
4102 while (1) {
4103 DBG("Health check ready");
4104
4105 /* Inifinite blocking call, waiting for transmission */
4106 restart:
4107 ret = lttng_poll_wait(&events, -1);
4108 if (ret < 0) {
4109 /*
4110 * Restart interrupted system call.
4111 */
4112 if (errno == EINTR) {
4113 goto restart;
4114 }
4115 goto error;
4116 }
4117
4118 nb_fd = ret;
4119
4120 for (i = 0; i < nb_fd; i++) {
4121 /* Fetch once the poll data */
4122 revents = LTTNG_POLL_GETEV(&events, i);
4123 pollfd = LTTNG_POLL_GETFD(&events, i);
4124
4125 if (!revents) {
4126 /* No activity for this FD (poll implementation). */
4127 continue;
4128 }
4129
4130 /* Thread quit pipe has been closed. Killing thread. */
4131 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4132 if (ret) {
4133 err = 0;
4134 goto exit;
4135 }
4136
4137 /* Event on the registration socket */
4138 if (pollfd == sock) {
4139 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4140 ERR("Health socket poll error");
4141 goto error;
4142 }
4143 }
4144 }
4145
4146 new_sock = lttcomm_accept_unix_sock(sock);
4147 if (new_sock < 0) {
4148 goto error;
4149 }
4150
4151 /*
4152 * Set the CLOEXEC flag. Return code is useless because either way, the
4153 * show must go on.
4154 */
4155 (void) utils_set_fd_cloexec(new_sock);
4156
4157 DBG("Receiving data from client for health...");
4158 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4159 if (ret <= 0) {
4160 DBG("Nothing recv() from client... continuing");
4161 ret = close(new_sock);
4162 if (ret) {
4163 PERROR("close");
4164 }
4165 new_sock = -1;
4166 continue;
4167 }
4168
4169 rcu_thread_online();
4170
4171 memset(&reply, 0, sizeof(reply));
4172 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4173 /*
4174 * health_check_state returns 0 if health is
4175 * bad.
4176 */
4177 if (!health_check_state(health_sessiond, i)) {
4178 reply.ret_code |= 1ULL << i;
4179 }
4180 }
4181
4182 DBG2("Health check return value %" PRIx64, reply.ret_code);
4183
4184 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4185 if (ret < 0) {
4186 ERR("Failed to send health data back to client");
4187 }
4188
4189 /* End of transmission */
4190 ret = close(new_sock);
4191 if (ret) {
4192 PERROR("close");
4193 }
4194 new_sock = -1;
4195 }
4196
4197 exit:
4198 error:
4199 if (err) {
4200 ERR("Health error occurred in %s", __func__);
4201 }
4202 DBG("Health check thread dying");
4203 unlink(health_unix_sock_path);
4204 if (sock >= 0) {
4205 ret = close(sock);
4206 if (ret) {
4207 PERROR("close");
4208 }
4209 }
4210
4211 lttng_poll_clean(&events);
4212
4213 rcu_unregister_thread();
4214 return NULL;
4215 }
4216
4217 /*
4218 * This thread manage all clients request using the unix client socket for
4219 * communication.
4220 */
4221 static void *thread_manage_clients(void *data)
4222 {
4223 int sock = -1, ret, i, pollfd, err = -1;
4224 int sock_error;
4225 uint32_t revents, nb_fd;
4226 struct command_ctx *cmd_ctx = NULL;
4227 struct lttng_poll_event events;
4228
4229 DBG("[thread] Manage client started");
4230
4231 rcu_register_thread();
4232
4233 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4234
4235 health_code_update();
4236
4237 ret = lttcomm_listen_unix_sock(client_sock);
4238 if (ret < 0) {
4239 goto error_listen;
4240 }
4241
4242 /*
4243 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4244 * more will be added to this poll set.
4245 */
4246 ret = sessiond_set_thread_pollset(&events, 2);
4247 if (ret < 0) {
4248 goto error_create_poll;
4249 }
4250
4251 /* Add the application registration socket */
4252 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4253 if (ret < 0) {
4254 goto error;
4255 }
4256
4257 sessiond_notify_ready();
4258 ret = sem_post(&load_info->message_thread_ready);
4259 if (ret) {
4260 PERROR("sem_post message_thread_ready");
4261 goto error;
4262 }
4263
4264 /* This testpoint is after we signal readiness to the parent. */
4265 if (testpoint(sessiond_thread_manage_clients)) {
4266 goto error;
4267 }
4268
4269 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4270 goto error;
4271 }
4272
4273 health_code_update();
4274
4275 while (1) {
4276 DBG("Accepting client command ...");
4277
4278 /* Inifinite blocking call, waiting for transmission */
4279 restart:
4280 health_poll_entry();
4281 ret = lttng_poll_wait(&events, -1);
4282 health_poll_exit();
4283 if (ret < 0) {
4284 /*
4285 * Restart interrupted system call.
4286 */
4287 if (errno == EINTR) {
4288 goto restart;
4289 }
4290 goto error;
4291 }
4292
4293 nb_fd = ret;
4294
4295 for (i = 0; i < nb_fd; i++) {
4296 /* Fetch once the poll data */
4297 revents = LTTNG_POLL_GETEV(&events, i);
4298 pollfd = LTTNG_POLL_GETFD(&events, i);
4299
4300 health_code_update();
4301
4302 if (!revents) {
4303 /* No activity for this FD (poll implementation). */
4304 continue;
4305 }
4306
4307 /* Thread quit pipe has been closed. Killing thread. */
4308 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4309 if (ret) {
4310 err = 0;
4311 goto exit;
4312 }
4313
4314 /* Event on the registration socket */
4315 if (pollfd == client_sock) {
4316 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4317 ERR("Client socket poll error");
4318 goto error;
4319 }
4320 }
4321 }
4322
4323 DBG("Wait for client response");
4324
4325 health_code_update();
4326
4327 sock = lttcomm_accept_unix_sock(client_sock);
4328 if (sock < 0) {
4329 goto error;
4330 }
4331
4332 /*
4333 * Set the CLOEXEC flag. Return code is useless because either way, the
4334 * show must go on.
4335 */
4336 (void) utils_set_fd_cloexec(sock);
4337
4338 /* Set socket option for credentials retrieval */
4339 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4340 if (ret < 0) {
4341 goto error;
4342 }
4343
4344 /* Allocate context command to process the client request */
4345 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4346 if (cmd_ctx == NULL) {
4347 PERROR("zmalloc cmd_ctx");
4348 goto error;
4349 }
4350
4351 /* Allocate data buffer for reception */
4352 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4353 if (cmd_ctx->lsm == NULL) {
4354 PERROR("zmalloc cmd_ctx->lsm");
4355 goto error;
4356 }
4357
4358 cmd_ctx->llm = NULL;
4359 cmd_ctx->session = NULL;
4360
4361 health_code_update();
4362
4363 /*
4364 * Data is received from the lttng client. The struct
4365 * lttcomm_session_msg (lsm) contains the command and data request of
4366 * the client.
4367 */
4368 DBG("Receiving data from client ...");
4369 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4370 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4371 if (ret <= 0) {
4372 DBG("Nothing recv() from client... continuing");
4373 ret = close(sock);
4374 if (ret) {
4375 PERROR("close");
4376 }
4377 sock = -1;
4378 clean_command_ctx(&cmd_ctx);
4379 continue;
4380 }
4381
4382 health_code_update();
4383
4384 // TODO: Validate cmd_ctx including sanity check for
4385 // security purpose.
4386
4387 rcu_thread_online();
4388 /*
4389 * This function dispatch the work to the kernel or userspace tracer
4390 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4391 * informations for the client. The command context struct contains
4392 * everything this function may needs.
4393 */
4394 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4395 rcu_thread_offline();
4396 if (ret < 0) {
4397 ret = close(sock);
4398 if (ret) {
4399 PERROR("close");
4400 }
4401 sock = -1;
4402 /*
4403 * TODO: Inform client somehow of the fatal error. At
4404 * this point, ret < 0 means that a zmalloc failed
4405 * (ENOMEM). Error detected but still accept
4406 * command, unless a socket error has been
4407 * detected.
4408 */
4409 clean_command_ctx(&cmd_ctx);
4410 continue;
4411 }
4412
4413 health_code_update();
4414
4415 DBG("Sending response (size: %d, retcode: %s)",
4416 cmd_ctx->lttng_msg_size,
4417 lttng_strerror(-cmd_ctx->llm->ret_code));
4418 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4419 if (ret < 0) {
4420 ERR("Failed to send data back to client");
4421 }
4422
4423 /* End of transmission */
4424 ret = close(sock);
4425 if (ret) {
4426 PERROR("close");
4427 }
4428 sock = -1;
4429
4430 clean_command_ctx(&cmd_ctx);
4431
4432 health_code_update();
4433 }
4434
4435 exit:
4436 error:
4437 if (sock >= 0) {
4438 ret = close(sock);
4439 if (ret) {
4440 PERROR("close");
4441 }
4442 }
4443
4444 lttng_poll_clean(&events);
4445 clean_command_ctx(&cmd_ctx);
4446
4447 error_listen:
4448 error_create_poll:
4449 unlink(client_unix_sock_path);
4450 if (client_sock >= 0) {
4451 ret = close(client_sock);
4452 if (ret) {
4453 PERROR("close");
4454 }
4455 }
4456
4457 if (err) {
4458 health_error();
4459 ERR("Health error occurred in %s", __func__);
4460 }
4461
4462 health_unregister(health_sessiond);
4463
4464 DBG("Client thread dying");
4465
4466 rcu_unregister_thread();
4467
4468 /*
4469 * Since we are creating the consumer threads, we own them, so we need
4470 * to join them before our thread exits.
4471 */
4472 ret = join_consumer_thread(&kconsumer_data);
4473 if (ret) {
4474 errno = ret;
4475 PERROR("join_consumer");
4476 }
4477
4478 ret = join_consumer_thread(&ustconsumer32_data);
4479 if (ret) {
4480 errno = ret;
4481 PERROR("join_consumer ust32");
4482 }
4483
4484 ret = join_consumer_thread(&ustconsumer64_data);
4485 if (ret) {
4486 errno = ret;
4487 PERROR("join_consumer ust64");
4488 }
4489 return NULL;
4490 }
4491
4492
4493 /*
4494 * usage function on stderr
4495 */
4496 static void usage(void)
4497 {
4498 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4499 fprintf(stderr, " -h, --help Display this usage.\n");
4500 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4501 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4502 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4503 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4504 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4505 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4506 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4507 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4508 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4509 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4510 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4511 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4512 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4513 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4514 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4515 fprintf(stderr, " -V, --version Show version number.\n");
4516 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4517 fprintf(stderr, " -q, --quiet No output at all.\n");
4518 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4519 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4520 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4521 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4522 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4523 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4524 fprintf(stderr, " -l --load PATH Load session configuration\n");
4525 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4526 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4527 }
4528
4529 static int string_match(const char *str1, const char *str2)
4530 {
4531 return (str1 && str2) && !strcmp(str1, str2);
4532 }
4533
4534 /*
4535 * Take an option from the getopt output and set it in the right variable to be
4536 * used later.
4537 *
4538 * Return 0 on success else a negative value.
4539 */
4540 static int set_option(int opt, const char *arg, const char *optname)
4541 {
4542 int ret = 0;
4543
4544 if (arg && arg[0] == '\0') {
4545 /*
4546 * This only happens if the value is read from daemon config
4547 * file. This means the option requires an argument and the
4548 * configuration file contains a line such as:
4549 * my_option =
4550 */
4551 ret = -EINVAL;
4552 goto end;
4553 }
4554
4555 if (string_match(optname, "client-sock") || opt == 'c') {
4556 if (lttng_is_setuid_setgid()) {
4557 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4558 "-c, --client-sock");
4559 } else {
4560 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4561 }
4562 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4563 if (lttng_is_setuid_setgid()) {
4564 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4565 "-a, --apps-sock");
4566 } else {
4567 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4568 }
4569 } else if (string_match(optname, "daemonize") || opt == 'd') {
4570 opt_daemon = 1;
4571 } else if (string_match(optname, "background") || opt == 'b') {
4572 opt_background = 1;
4573 } else if (string_match(optname, "group") || opt == 'g') {
4574 if (lttng_is_setuid_setgid()) {
4575 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4576 "-g, --group");
4577 } else {
4578 /*
4579 * If the override option is set, the pointer points to a
4580 * *non* const thus freeing it even though the variable type is
4581 * set to const.
4582 */
4583 if (tracing_group_name_override) {
4584 free((void *) tracing_group_name);
4585 }
4586 tracing_group_name = strdup(arg);
4587 if (!tracing_group_name) {
4588 PERROR("strdup");
4589 ret = -ENOMEM;
4590 }
4591 tracing_group_name_override = 1;
4592 }
4593 } else if (string_match(optname, "help") || opt == 'h') {
4594 usage();
4595 exit(EXIT_SUCCESS);
4596 } else if (string_match(optname, "version") || opt == 'V') {
4597 fprintf(stdout, "%s\n", VERSION);
4598 exit(EXIT_SUCCESS);
4599 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4600 opt_sig_parent = 1;
4601 } else if (string_match(optname, "kconsumerd-err-sock")) {
4602 if (lttng_is_setuid_setgid()) {
4603 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4604 "--kconsumerd-err-sock");
4605 } else {
4606 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4607 }
4608 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4609 if (lttng_is_setuid_setgid()) {
4610 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4611 "--kconsumerd-cmd-sock");
4612 } else {
4613 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4614 }
4615 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4616 if (lttng_is_setuid_setgid()) {
4617 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4618 "--ustconsumerd64-err-sock");
4619 } else {
4620 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4621 }
4622 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4623 if (lttng_is_setuid_setgid()) {
4624 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4625 "--ustconsumerd64-cmd-sock");
4626 } else {
4627 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4628 }
4629 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4630 if (lttng_is_setuid_setgid()) {
4631 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4632 "--ustconsumerd32-err-sock");
4633 } else {
4634 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4635 }
4636 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4637 if (lttng_is_setuid_setgid()) {
4638 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4639 "--ustconsumerd32-cmd-sock");
4640 } else {
4641 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4642 }
4643 } else if (string_match(optname, "no-kernel")) {
4644 opt_no_kernel = 1;
4645 } else if (string_match(optname, "quiet") || opt == 'q') {
4646 lttng_opt_quiet = 1;
4647 } else if (string_match(optname, "verbose") || opt == 'v') {
4648 /* Verbose level can increase using multiple -v */
4649 if (arg) {
4650 /* Value obtained from config file */
4651 lttng_opt_verbose = config_parse_value(arg);
4652 } else {
4653 /* -v used on command line */
4654 lttng_opt_verbose++;
4655 }
4656 /* Clamp value to [0, 3] */
4657 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4658 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4659 } else if (string_match(optname, "verbose-consumer")) {
4660 if (arg) {
4661 opt_verbose_consumer = config_parse_value(arg);
4662 } else {
4663 opt_verbose_consumer += 1;
4664 }
4665 } else if (string_match(optname, "consumerd32-path")) {
4666 if (lttng_is_setuid_setgid()) {
4667 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4668 "--consumerd32-path");
4669 } else {
4670 if (consumerd32_bin_override) {
4671 free((void *) consumerd32_bin);
4672 }
4673 consumerd32_bin = strdup(arg);
4674 if (!consumerd32_bin) {
4675 PERROR("strdup");
4676 ret = -ENOMEM;
4677 }
4678 consumerd32_bin_override = 1;
4679 }
4680 } else if (string_match(optname, "consumerd32-libdir")) {
4681 if (lttng_is_setuid_setgid()) {
4682 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4683 "--consumerd32-libdir");
4684 } else {
4685 if (consumerd32_libdir_override) {
4686 free((void *) consumerd32_libdir);
4687 }
4688 consumerd32_libdir = strdup(arg);
4689 if (!consumerd32_libdir) {
4690 PERROR("strdup");
4691 ret = -ENOMEM;
4692 }
4693 consumerd32_libdir_override = 1;
4694 }
4695 } else if (string_match(optname, "consumerd64-path")) {
4696 if (lttng_is_setuid_setgid()) {
4697 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4698 "--consumerd64-path");
4699 } else {
4700 if (consumerd64_bin_override) {
4701 free((void *) consumerd64_bin);
4702 }
4703 consumerd64_bin = strdup(arg);
4704 if (!consumerd64_bin) {
4705 PERROR("strdup");
4706 ret = -ENOMEM;
4707 }
4708 consumerd64_bin_override = 1;
4709 }
4710 } else if (string_match(optname, "consumerd64-libdir")) {
4711 if (lttng_is_setuid_setgid()) {
4712 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4713 "--consumerd64-libdir");
4714 } else {
4715 if (consumerd64_libdir_override) {
4716 free((void *) consumerd64_libdir);
4717 }
4718 consumerd64_libdir = strdup(arg);
4719 if (!consumerd64_libdir) {
4720 PERROR("strdup");
4721 ret = -ENOMEM;
4722 }
4723 consumerd64_libdir_override = 1;
4724 }
4725 } else if (string_match(optname, "pidfile") || opt == 'p') {
4726 if (lttng_is_setuid_setgid()) {
4727 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4728 "-p, --pidfile");
4729 } else {
4730 free(opt_pidfile);
4731 opt_pidfile = strdup(arg);
4732 if (!opt_pidfile) {
4733 PERROR("strdup");
4734 ret = -ENOMEM;
4735 }
4736 }
4737 } else if (string_match(optname, "agent-tcp-port")) {
4738 if (lttng_is_setuid_setgid()) {
4739 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4740 "--agent-tcp-port");
4741 } else {
4742 unsigned long v;
4743
4744 if (!arg) {
4745 ret = -EINVAL;
4746 goto end;
4747 }
4748 errno = 0;
4749 v = strtoul(arg, NULL, 0);
4750 if (errno != 0 || !isdigit(arg[0])) {
4751 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4752 return -1;
4753 }
4754 if (v == 0 || v >= 65535) {
4755 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4756 return -1;
4757 }
4758 agent_tcp_port = (uint32_t) v;
4759 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4760 }
4761 } else if (string_match(optname, "load") || opt == 'l') {
4762 if (lttng_is_setuid_setgid()) {
4763 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4764 "-l, --load");
4765 } else {
4766 free(opt_load_session_path);
4767 opt_load_session_path = strdup(arg);
4768 if (!opt_load_session_path) {
4769 PERROR("strdup");
4770 ret = -ENOMEM;
4771 }
4772 }
4773 } else if (string_match(optname, "kmod-probes")) {
4774 if (lttng_is_setuid_setgid()) {
4775 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4776 "--kmod-probes");
4777 } else {
4778 free(kmod_probes_list);
4779 kmod_probes_list = strdup(arg);
4780 if (!kmod_probes_list) {
4781 PERROR("strdup");
4782 ret = -ENOMEM;
4783 }
4784 }
4785 } else if (string_match(optname, "extra-kmod-probes")) {
4786 if (lttng_is_setuid_setgid()) {
4787 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4788 "--extra-kmod-probes");
4789 } else {
4790 free(kmod_extra_probes_list);
4791 kmod_extra_probes_list = strdup(arg);
4792 if (!kmod_extra_probes_list) {
4793 PERROR("strdup");
4794 ret = -ENOMEM;
4795 }
4796 }
4797 } else if (string_match(optname, "config") || opt == 'f') {
4798 /* This is handled in set_options() thus silent skip. */
4799 goto end;
4800 } else {
4801 /* Unknown option or other error.
4802 * Error is printed by getopt, just return */
4803 ret = -1;
4804 }
4805
4806 end:
4807 if (ret == -EINVAL) {
4808 const char *opt_name = "unknown";
4809 int i;
4810
4811 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4812 i++) {
4813 if (opt == long_options[i].val) {
4814 opt_name = long_options[i].name;
4815 break;
4816 }
4817 }
4818
4819 WARN("Invalid argument provided for option \"%s\", using default value.",
4820 opt_name);
4821 }
4822
4823 return ret;
4824 }
4825
4826 /*
4827 * config_entry_handler_cb used to handle options read from a config file.
4828 * See config_entry_handler_cb comment in common/config/config.h for the
4829 * return value conventions.
4830 */
4831 static int config_entry_handler(const struct config_entry *entry, void *unused)
4832 {
4833 int ret = 0, i;
4834
4835 if (!entry || !entry->name || !entry->value) {
4836 ret = -EINVAL;
4837 goto end;
4838 }
4839
4840 /* Check if the option is to be ignored */
4841 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4842 if (!strcmp(entry->name, config_ignore_options[i])) {
4843 goto end;
4844 }
4845 }
4846
4847 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4848 i++) {
4849
4850 /* Ignore if not fully matched. */
4851 if (strcmp(entry->name, long_options[i].name)) {
4852 continue;
4853 }
4854
4855 /*
4856 * If the option takes no argument on the command line, we have to
4857 * check if the value is "true". We support non-zero numeric values,
4858 * true, on and yes.
4859 */
4860 if (!long_options[i].has_arg) {
4861 ret = config_parse_value(entry->value);
4862 if (ret <= 0) {
4863 if (ret) {
4864 WARN("Invalid configuration value \"%s\" for option %s",
4865 entry->value, entry->name);
4866 }
4867 /* False, skip boolean config option. */
4868 goto end;
4869 }
4870 }
4871
4872 ret = set_option(long_options[i].val, entry->value, entry->name);
4873 goto end;
4874 }
4875
4876 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4877
4878 end:
4879 return ret;
4880 }
4881
4882 /*
4883 * daemon configuration loading and argument parsing
4884 */
4885 static int set_options(int argc, char **argv)
4886 {
4887 int ret = 0, c = 0, option_index = 0;
4888 int orig_optopt = optopt, orig_optind = optind;
4889 char *optstring;
4890 const char *config_path = NULL;
4891
4892 optstring = utils_generate_optstring(long_options,
4893 sizeof(long_options) / sizeof(struct option));
4894 if (!optstring) {
4895 ret = -ENOMEM;
4896 goto end;
4897 }
4898
4899 /* Check for the --config option */
4900 while ((c = getopt_long(argc, argv, optstring, long_options,
4901 &option_index)) != -1) {
4902 if (c == '?') {
4903 ret = -EINVAL;
4904 goto end;
4905 } else if (c != 'f') {
4906 /* if not equal to --config option. */
4907 continue;
4908 }
4909
4910 if (lttng_is_setuid_setgid()) {
4911 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4912 "-f, --config");
4913 } else {
4914 config_path = utils_expand_path(optarg);
4915 if (!config_path) {
4916 ERR("Failed to resolve path: %s", optarg);
4917 }
4918 }
4919 }
4920
4921 ret = config_get_section_entries(config_path, config_section_name,
4922 config_entry_handler, NULL);
4923 if (ret) {
4924 if (ret > 0) {
4925 ERR("Invalid configuration option at line %i", ret);
4926 ret = -1;
4927 }
4928 goto end;
4929 }
4930
4931 /* Reset getopt's global state */
4932 optopt = orig_optopt;
4933 optind = orig_optind;
4934 while (1) {
4935 option_index = -1;
4936 /*
4937 * getopt_long() will not set option_index if it encounters a
4938 * short option.
4939 */
4940 c = getopt_long(argc, argv, optstring, long_options,
4941 &option_index);
4942 if (c == -1) {
4943 break;
4944 }
4945
4946 /*
4947 * Pass NULL as the long option name if popt left the index
4948 * unset.
4949 */
4950 ret = set_option(c, optarg,
4951 option_index < 0 ? NULL :
4952 long_options[option_index].name);
4953 if (ret < 0) {
4954 break;
4955 }
4956 }
4957
4958 end:
4959 free(optstring);
4960 return ret;
4961 }
4962
4963 /*
4964 * Creates the two needed socket by the daemon.
4965 * apps_sock - The communication socket for all UST apps.
4966 * client_sock - The communication of the cli tool (lttng).
4967 */
4968 static int init_daemon_socket(void)
4969 {
4970 int ret = 0;
4971 mode_t old_umask;
4972
4973 old_umask = umask(0);
4974
4975 /* Create client tool unix socket */
4976 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4977 if (client_sock < 0) {
4978 ERR("Create unix sock failed: %s", client_unix_sock_path);
4979 ret = -1;
4980 goto end;
4981 }
4982
4983 /* Set the cloexec flag */
4984 ret = utils_set_fd_cloexec(client_sock);
4985 if (ret < 0) {
4986 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4987 "Continuing but note that the consumer daemon will have a "
4988 "reference to this socket on exec()", client_sock);
4989 }
4990
4991 /* File permission MUST be 660 */
4992 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4993 if (ret < 0) {
4994 ERR("Set file permissions failed: %s", client_unix_sock_path);
4995 PERROR("chmod");
4996 goto end;
4997 }
4998
4999 /* Create the application unix socket */
5000 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5001 if (apps_sock < 0) {
5002 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5003 ret = -1;
5004 goto end;
5005 }
5006
5007 /* Set the cloexec flag */
5008 ret = utils_set_fd_cloexec(apps_sock);
5009 if (ret < 0) {
5010 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5011 "Continuing but note that the consumer daemon will have a "
5012 "reference to this socket on exec()", apps_sock);
5013 }
5014
5015 /* File permission MUST be 666 */
5016 ret = chmod(apps_unix_sock_path,
5017 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5018 if (ret < 0) {
5019 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5020 PERROR("chmod");
5021 goto end;
5022 }
5023
5024 DBG3("Session daemon client socket %d and application socket %d created",
5025 client_sock, apps_sock);
5026
5027 end:
5028 umask(old_umask);
5029 return ret;
5030 }
5031
5032 /*
5033 * Check if the global socket is available, and if a daemon is answering at the
5034 * other side. If yes, error is returned.
5035 */
5036 static int check_existing_daemon(void)
5037 {
5038 /* Is there anybody out there ? */
5039 if (lttng_session_daemon_alive()) {
5040 return -EEXIST;
5041 }
5042
5043 return 0;
5044 }
5045
5046 /*
5047 * Set the tracing group gid onto the client socket.
5048 *
5049 * Race window between mkdir and chown is OK because we are going from more
5050 * permissive (root.root) to less permissive (root.tracing).
5051 */
5052 static int set_permissions(char *rundir)
5053 {
5054 int ret;
5055 gid_t gid;
5056
5057 gid = utils_get_group_id(tracing_group_name);
5058
5059 /* Set lttng run dir */
5060 ret = chown(rundir, 0, gid);
5061 if (ret < 0) {
5062 ERR("Unable to set group on %s", rundir);
5063 PERROR("chown");
5064 }
5065
5066 /*
5067 * Ensure all applications and tracing group can search the run
5068 * dir. Allow everyone to read the directory, since it does not
5069 * buy us anything to hide its content.
5070 */
5071 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5072 if (ret < 0) {
5073 ERR("Unable to set permissions on %s", rundir);
5074 PERROR("chmod");
5075 }
5076
5077 /* lttng client socket path */
5078 ret = chown(client_unix_sock_path, 0, gid);
5079 if (ret < 0) {
5080 ERR("Unable to set group on %s", client_unix_sock_path);
5081 PERROR("chown");
5082 }
5083
5084 /* kconsumer error socket path */
5085 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5086 if (ret < 0) {
5087 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5088 PERROR("chown");
5089 }
5090
5091 /* 64-bit ustconsumer error socket path */
5092 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5093 if (ret < 0) {
5094 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5095 PERROR("chown");
5096 }
5097
5098 /* 32-bit ustconsumer compat32 error socket path */
5099 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5100 if (ret < 0) {
5101 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5102 PERROR("chown");
5103 }
5104
5105 DBG("All permissions are set");
5106
5107 return ret;
5108 }
5109
5110 /*
5111 * Create the lttng run directory needed for all global sockets and pipe.
5112 */
5113 static int create_lttng_rundir(const char *rundir)
5114 {
5115 int ret;
5116
5117 DBG3("Creating LTTng run directory: %s", rundir);
5118
5119 ret = mkdir(rundir, S_IRWXU);
5120 if (ret < 0) {
5121 if (errno != EEXIST) {
5122 ERR("Unable to create %s", rundir);
5123 goto error;
5124 } else {
5125 ret = 0;
5126 }
5127 }
5128
5129 error:
5130 return ret;
5131 }
5132
5133 /*
5134 * Setup sockets and directory needed by the kconsumerd communication with the
5135 * session daemon.
5136 */
5137 static int set_consumer_sockets(struct consumer_data *consumer_data,
5138 const char *rundir)
5139 {
5140 int ret;
5141 char path[PATH_MAX];
5142
5143 switch (consumer_data->type) {
5144 case LTTNG_CONSUMER_KERNEL:
5145 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5146 break;
5147 case LTTNG_CONSUMER64_UST:
5148 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5149 break;
5150 case LTTNG_CONSUMER32_UST:
5151 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5152 break;
5153 default:
5154 ERR("Consumer type unknown");
5155 ret = -EINVAL;
5156 goto error;
5157 }
5158
5159 DBG2("Creating consumer directory: %s", path);
5160
5161 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5162 if (ret < 0) {
5163 if (errno != EEXIST) {
5164 PERROR("mkdir");
5165 ERR("Failed to create %s", path);
5166 goto error;
5167 }
5168 ret = -1;
5169 }
5170 if (is_root) {
5171 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5172 if (ret < 0) {
5173 ERR("Unable to set group on %s", path);
5174 PERROR("chown");
5175 goto error;
5176 }
5177 }
5178
5179 /* Create the kconsumerd error unix socket */
5180 consumer_data->err_sock =
5181 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5182 if (consumer_data->err_sock < 0) {
5183 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5184 ret = -1;
5185 goto error;
5186 }
5187
5188 /*
5189 * Set the CLOEXEC flag. Return code is useless because either way, the
5190 * show must go on.
5191 */
5192 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5193 if (ret < 0) {
5194 PERROR("utils_set_fd_cloexec");
5195 /* continue anyway */
5196 }
5197
5198 /* File permission MUST be 660 */
5199 ret = chmod(consumer_data->err_unix_sock_path,
5200 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5201 if (ret < 0) {
5202 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5203 PERROR("chmod");
5204 goto error;
5205 }
5206
5207 error:
5208 return ret;
5209 }
5210
5211 /*
5212 * Signal handler for the daemon
5213 *
5214 * Simply stop all worker threads, leaving main() return gracefully after
5215 * joining all threads and calling cleanup().
5216 */
5217 static void sighandler(int sig)
5218 {
5219 switch (sig) {
5220 case SIGPIPE:
5221 DBG("SIGPIPE caught");
5222 return;
5223 case SIGINT:
5224 DBG("SIGINT caught");
5225 stop_threads();
5226 break;
5227 case SIGTERM:
5228 DBG("SIGTERM caught");
5229 stop_threads();
5230 break;
5231 case SIGUSR1:
5232 CMM_STORE_SHARED(recv_child_signal, 1);
5233 break;
5234 default:
5235 break;
5236 }
5237 }
5238
5239 /*
5240 * Setup signal handler for :
5241 * SIGINT, SIGTERM, SIGPIPE
5242 */
5243 static int set_signal_handler(void)
5244 {
5245 int ret = 0;
5246 struct sigaction sa;
5247 sigset_t sigset;
5248
5249 if ((ret = sigemptyset(&sigset)) < 0) {
5250 PERROR("sigemptyset");
5251 return ret;
5252 }
5253
5254 sa.sa_handler = sighandler;
5255 sa.sa_mask = sigset;
5256 sa.sa_flags = 0;
5257 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5258 PERROR("sigaction");
5259 return ret;
5260 }
5261
5262 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5263 PERROR("sigaction");
5264 return ret;
5265 }
5266
5267 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5268 PERROR("sigaction");
5269 return ret;
5270 }
5271
5272 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5273 PERROR("sigaction");
5274 return ret;
5275 }
5276
5277 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5278
5279 return ret;
5280 }
5281
5282 /*
5283 * Set open files limit to unlimited. This daemon can open a large number of
5284 * file descriptors in order to consumer multiple kernel traces.
5285 */
5286 static void set_ulimit(void)
5287 {
5288 int ret;
5289 struct rlimit lim;
5290
5291 /* The kernel does not allowed an infinite limit for open files */
5292 lim.rlim_cur = 65535;
5293 lim.rlim_max = 65535;
5294
5295 ret = setrlimit(RLIMIT_NOFILE, &lim);
5296 if (ret < 0) {
5297 PERROR("failed to set open files limit");
5298 }
5299 }
5300
5301 /*
5302 * Write pidfile using the rundir and opt_pidfile.
5303 */
5304 static int write_pidfile(void)
5305 {
5306 int ret;
5307 char pidfile_path[PATH_MAX];
5308
5309 assert(rundir);
5310
5311 if (opt_pidfile) {
5312 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5313 } else {
5314 /* Build pidfile path from rundir and opt_pidfile. */
5315 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5316 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5317 if (ret < 0) {
5318 PERROR("snprintf pidfile path");
5319 goto error;
5320 }
5321 }
5322
5323 /*
5324 * Create pid file in rundir.
5325 */
5326 ret = utils_create_pid_file(getpid(), pidfile_path);
5327 error:
5328 return ret;
5329 }
5330
5331 /*
5332 * Create lockfile using the rundir and return its fd.
5333 */
5334 static int create_lockfile(void)
5335 {
5336 int ret;
5337 char lockfile_path[PATH_MAX];
5338
5339 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5340 if (ret < 0) {
5341 goto error;
5342 }
5343
5344 ret = utils_create_lock_file(lockfile_path);
5345 error:
5346 return ret;
5347 }
5348
5349 /*
5350 * Write agent TCP port using the rundir.
5351 */
5352 static int write_agent_port(void)
5353 {
5354 int ret;
5355 char path[PATH_MAX];
5356
5357 assert(rundir);
5358
5359 ret = snprintf(path, sizeof(path), "%s/"
5360 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5361 if (ret < 0) {
5362 PERROR("snprintf agent port path");
5363 goto error;
5364 }
5365
5366 /*
5367 * Create TCP agent port file in rundir.
5368 */
5369 ret = utils_create_pid_file(agent_tcp_port, path);
5370
5371 error:
5372 return ret;
5373 }
5374
5375 /*
5376 * main
5377 */
5378 int main(int argc, char **argv)
5379 {
5380 int ret = 0, retval = 0;
5381 void *status;
5382 const char *home_path, *env_app_timeout;
5383
5384 /* Initialize agent apps ht global variable */
5385 agent_apps_ht_by_sock = NULL;
5386
5387 init_kernel_workarounds();
5388
5389 rcu_register_thread();
5390
5391 if (set_signal_handler()) {
5392 retval = -1;
5393 goto exit_set_signal_handler;
5394 }
5395
5396 setup_consumerd_path();
5397
5398 page_size = sysconf(_SC_PAGESIZE);
5399 if (page_size < 0) {
5400 PERROR("sysconf _SC_PAGESIZE");
5401 page_size = LONG_MAX;
5402 WARN("Fallback page size to %ld", page_size);
5403 }
5404
5405 /*
5406 * Parse arguments and load the daemon configuration file.
5407 *
5408 * We have an exit_options exit path to free memory reserved by
5409 * set_options. This is needed because the rest of sessiond_cleanup()
5410 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5411 * depends on set_options.
5412 */
5413 progname = argv[0];
5414 if (set_options(argc, argv)) {
5415 retval = -1;
5416 goto exit_options;
5417 }
5418
5419 /* Daemonize */
5420 if (opt_daemon || opt_background) {
5421 int i;
5422
5423 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5424 !opt_background);
5425 if (ret < 0) {
5426 retval = -1;
5427 goto exit_options;
5428 }
5429
5430 /*
5431 * We are in the child. Make sure all other file descriptors are
5432 * closed, in case we are called with more opened file
5433 * descriptors than the standard ones.
5434 */
5435 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5436 (void) close(i);
5437 }
5438 }
5439
5440 /*
5441 * Starting from here, we can create threads. This needs to be after
5442 * lttng_daemonize due to RCU.
5443 */
5444
5445 /*
5446 * Initialize the health check subsystem. This call should set the
5447 * appropriate time values.
5448 */
5449 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5450 if (!health_sessiond) {
5451 PERROR("health_app_create error");
5452 retval = -1;
5453 goto exit_health_sessiond_cleanup;
5454 }
5455
5456 if (init_ht_cleanup_quit_pipe()) {
5457 retval = -1;
5458 goto exit_ht_cleanup_quit_pipe;
5459 }
5460
5461 /* Setup the thread ht_cleanup communication pipe. */
5462 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5463 retval = -1;
5464 goto exit_ht_cleanup_pipe;
5465 }
5466
5467 /* Set up max poll set size */
5468 if (lttng_poll_set_max_size()) {
5469 retval = -1;
5470 goto exit_set_max_size;
5471 }
5472
5473 /* Create thread to clean up RCU hash tables */
5474 ret = pthread_create(&ht_cleanup_thread, NULL,
5475 thread_ht_cleanup, (void *) NULL);
5476 if (ret) {
5477 errno = ret;
5478 PERROR("pthread_create ht_cleanup");
5479 retval = -1;
5480 goto exit_ht_cleanup;
5481 }
5482
5483 /* Create thread quit pipe */
5484 if (init_thread_quit_pipe()) {
5485 retval = -1;
5486 goto exit_init_data;
5487 }
5488
5489 /* Check if daemon is UID = 0 */
5490 is_root = !getuid();
5491
5492 if (is_root) {
5493 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5494 if (!rundir) {
5495 retval = -1;
5496 goto exit_init_data;
5497 }
5498
5499 /* Create global run dir with root access */
5500 if (create_lttng_rundir(rundir)) {
5501 retval = -1;
5502 goto exit_init_data;
5503 }
5504
5505 if (strlen(apps_unix_sock_path) == 0) {
5506 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5507 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5508 if (ret < 0) {
5509 retval = -1;
5510 goto exit_init_data;
5511 }
5512 }
5513
5514 if (strlen(client_unix_sock_path) == 0) {
5515 ret = snprintf(client_unix_sock_path, PATH_MAX,
5516 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5517 if (ret < 0) {
5518 retval = -1;
5519 goto exit_init_data;
5520 }
5521 }
5522
5523 /* Set global SHM for ust */
5524 if (strlen(wait_shm_path) == 0) {
5525 ret = snprintf(wait_shm_path, PATH_MAX,
5526 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5527 if (ret < 0) {
5528 retval = -1;
5529 goto exit_init_data;
5530 }
5531 }
5532
5533 if (strlen(health_unix_sock_path) == 0) {
5534 ret = snprintf(health_unix_sock_path,
5535 sizeof(health_unix_sock_path),
5536 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5537 if (ret < 0) {
5538 retval = -1;
5539 goto exit_init_data;
5540 }
5541 }
5542
5543 /* Setup kernel consumerd path */
5544 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5545 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5546 if (ret < 0) {
5547 retval = -1;
5548 goto exit_init_data;
5549 }
5550 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5551 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5552 if (ret < 0) {
5553 retval = -1;
5554 goto exit_init_data;
5555 }
5556
5557 DBG2("Kernel consumer err path: %s",
5558 kconsumer_data.err_unix_sock_path);
5559 DBG2("Kernel consumer cmd path: %s",
5560 kconsumer_data.cmd_unix_sock_path);
5561 } else {
5562 home_path = utils_get_home_dir();
5563 if (home_path == NULL) {
5564 /* TODO: Add --socket PATH option */
5565 ERR("Can't get HOME directory for sockets creation.");
5566 retval = -1;
5567 goto exit_init_data;
5568 }
5569
5570 /*
5571 * Create rundir from home path. This will create something like
5572 * $HOME/.lttng
5573 */
5574 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5575 if (ret < 0) {
5576 retval = -1;
5577 goto exit_init_data;
5578 }
5579
5580 if (create_lttng_rundir(rundir)) {
5581 retval = -1;
5582 goto exit_init_data;
5583 }
5584
5585 if (strlen(apps_unix_sock_path) == 0) {
5586 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5587 DEFAULT_HOME_APPS_UNIX_SOCK,
5588 home_path);
5589 if (ret < 0) {
5590 retval = -1;
5591 goto exit_init_data;
5592 }
5593 }
5594
5595 /* Set the cli tool unix socket path */
5596 if (strlen(client_unix_sock_path) == 0) {
5597 ret = snprintf(client_unix_sock_path, PATH_MAX,
5598 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5599 home_path);
5600 if (ret < 0) {
5601 retval = -1;
5602 goto exit_init_data;
5603 }
5604 }
5605
5606 /* Set global SHM for ust */
5607 if (strlen(wait_shm_path) == 0) {
5608 ret = snprintf(wait_shm_path, PATH_MAX,
5609 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5610 getuid());
5611 if (ret < 0) {
5612 retval = -1;
5613 goto exit_init_data;
5614 }
5615 }
5616
5617 /* Set health check Unix path */
5618 if (strlen(health_unix_sock_path) == 0) {
5619 ret = snprintf(health_unix_sock_path,
5620 sizeof(health_unix_sock_path),
5621 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5622 home_path);
5623 if (ret < 0) {
5624 retval = -1;
5625 goto exit_init_data;
5626 }
5627 }
5628 }
5629
5630 lockfile_fd = create_lockfile();
5631 if (lockfile_fd < 0) {
5632 retval = -1;
5633 goto exit_init_data;
5634 }
5635
5636 /* Set consumer initial state */
5637 kernel_consumerd_state = CONSUMER_STOPPED;
5638 ust_consumerd_state = CONSUMER_STOPPED;
5639
5640 DBG("Client socket path %s", client_unix_sock_path);
5641 DBG("Application socket path %s", apps_unix_sock_path);
5642 DBG("Application wait path %s", wait_shm_path);
5643 DBG("LTTng run directory path: %s", rundir);
5644
5645 /* 32 bits consumerd path setup */
5646 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5647 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5648 if (ret < 0) {
5649 PERROR("snprintf 32-bit consumer error socket path");
5650 retval = -1;
5651 goto exit_init_data;
5652 }
5653 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5654 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5655 if (ret < 0) {
5656 PERROR("snprintf 32-bit consumer command socket path");
5657 retval = -1;
5658 goto exit_init_data;
5659 }
5660
5661 DBG2("UST consumer 32 bits err path: %s",
5662 ustconsumer32_data.err_unix_sock_path);
5663 DBG2("UST consumer 32 bits cmd path: %s",
5664 ustconsumer32_data.cmd_unix_sock_path);
5665
5666 /* 64 bits consumerd path setup */
5667 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5668 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5669 if (ret < 0) {
5670 PERROR("snprintf 64-bit consumer error socket path");
5671 retval = -1;
5672 goto exit_init_data;
5673 }
5674 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5675 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5676 if (ret < 0) {
5677 PERROR("snprintf 64-bit consumer command socket path");
5678 retval = -1;
5679 goto exit_init_data;
5680 }
5681
5682 DBG2("UST consumer 64 bits err path: %s",
5683 ustconsumer64_data.err_unix_sock_path);
5684 DBG2("UST consumer 64 bits cmd path: %s",
5685 ustconsumer64_data.cmd_unix_sock_path);
5686
5687 /*
5688 * See if daemon already exist.
5689 */
5690 if (check_existing_daemon()) {
5691 ERR("Already running daemon.\n");
5692 /*
5693 * We do not goto exit because we must not cleanup()
5694 * because a daemon is already running.
5695 */
5696 retval = -1;
5697 goto exit_init_data;
5698 }
5699
5700 /*
5701 * Init UST app hash table. Alloc hash table before this point since
5702 * cleanup() can get called after that point.
5703 */
5704 if (ust_app_ht_alloc()) {
5705 ERR("Failed to allocate UST app hash table");
5706 retval = -1;
5707 goto exit_init_data;
5708 }
5709
5710 /*
5711 * Initialize agent app hash table. We allocate the hash table here
5712 * since cleanup() can get called after this point.
5713 */
5714 if (agent_app_ht_alloc()) {
5715 ERR("Failed to allocate Agent app hash table");
5716 retval = -1;
5717 goto exit_init_data;
5718 }
5719
5720 /*
5721 * These actions must be executed as root. We do that *after* setting up
5722 * the sockets path because we MUST make the check for another daemon using
5723 * those paths *before* trying to set the kernel consumer sockets and init
5724 * kernel tracer.
5725 */
5726 if (is_root) {
5727 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5728 retval = -1;
5729 goto exit_init_data;
5730 }
5731
5732 /* Setup kernel tracer */
5733 if (!opt_no_kernel) {
5734 init_kernel_tracer();
5735 if (kernel_tracer_fd >= 0) {
5736 ret = syscall_init_table();
5737 if (ret < 0) {
5738 ERR("Unable to populate syscall table. "
5739 "Syscall tracing won't work "
5740 "for this session daemon.");
5741 }
5742 }
5743 }
5744
5745 /* Set ulimit for open files */
5746 set_ulimit();
5747 }
5748 /* init lttng_fd tracking must be done after set_ulimit. */
5749 lttng_fd_init();
5750
5751 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5752 retval = -1;
5753 goto exit_init_data;
5754 }
5755
5756 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5757 retval = -1;
5758 goto exit_init_data;
5759 }
5760
5761 /* Setup the needed unix socket */
5762 if (init_daemon_socket()) {
5763 retval = -1;
5764 goto exit_init_data;
5765 }
5766
5767 /* Set credentials to socket */
5768 if (is_root && set_permissions(rundir)) {
5769 retval = -1;
5770 goto exit_init_data;
5771 }
5772
5773 /* Get parent pid if -S, --sig-parent is specified. */
5774 if (opt_sig_parent) {
5775 ppid = getppid();
5776 }
5777
5778 /* Setup the kernel pipe for waking up the kernel thread */
5779 if (is_root && !opt_no_kernel) {
5780 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5781 retval = -1;
5782 goto exit_init_data;
5783 }
5784 }
5785
5786 /* Setup the thread apps communication pipe. */
5787 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5788 retval = -1;
5789 goto exit_init_data;
5790 }
5791
5792 /* Setup the thread apps notify communication pipe. */
5793 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5794 retval = -1;
5795 goto exit_init_data;
5796 }
5797
5798 /* Initialize global buffer per UID and PID registry. */
5799 buffer_reg_init_uid_registry();
5800 buffer_reg_init_pid_registry();
5801
5802 /* Init UST command queue. */
5803 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5804
5805 /*
5806 * Get session list pointer. This pointer MUST NOT be free'd. This list
5807 * is statically declared in session.c
5808 */
5809 session_list_ptr = session_get_list();
5810
5811 cmd_init();
5812
5813 /* Check for the application socket timeout env variable. */
5814 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5815 if (env_app_timeout) {
5816 app_socket_timeout = atoi(env_app_timeout);
5817 } else {
5818 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5819 }
5820
5821 ret = write_pidfile();
5822 if (ret) {
5823 ERR("Error in write_pidfile");
5824 retval = -1;
5825 goto exit_init_data;
5826 }
5827 ret = write_agent_port();
5828 if (ret) {
5829 ERR("Error in write_agent_port");
5830 retval = -1;
5831 goto exit_init_data;
5832 }
5833
5834 /* Initialize communication library */
5835 lttcomm_init();
5836 /* Initialize TCP timeout values */
5837 lttcomm_inet_init();
5838
5839 if (load_session_init_data(&load_info) < 0) {
5840 retval = -1;
5841 goto exit_init_data;
5842 }
5843 load_info->path = opt_load_session_path;
5844
5845 /* Create health-check thread */
5846 ret = pthread_create(&health_thread, NULL,
5847 thread_manage_health, (void *) NULL);
5848 if (ret) {
5849 errno = ret;
5850 PERROR("pthread_create health");
5851 retval = -1;
5852 goto exit_health;
5853 }
5854
5855 /* Create thread to manage the client socket */
5856 ret = pthread_create(&client_thread, NULL,
5857 thread_manage_clients, (void *) NULL);
5858 if (ret) {
5859 errno = ret;
5860 PERROR("pthread_create clients");
5861 retval = -1;
5862 goto exit_client;
5863 }
5864
5865 /* Create thread to dispatch registration */
5866 ret = pthread_create(&dispatch_thread, NULL,
5867 thread_dispatch_ust_registration, (void *) NULL);
5868 if (ret) {
5869 errno = ret;
5870 PERROR("pthread_create dispatch");
5871 retval = -1;
5872 goto exit_dispatch;
5873 }
5874
5875 /* Create thread to manage application registration. */
5876 ret = pthread_create(&reg_apps_thread, NULL,
5877 thread_registration_apps, (void *) NULL);
5878 if (ret) {
5879 errno = ret;
5880 PERROR("pthread_create registration");
5881 retval = -1;
5882 goto exit_reg_apps;
5883 }
5884
5885 /* Create thread to manage application socket */
5886 ret = pthread_create(&apps_thread, NULL,
5887 thread_manage_apps, (void *) NULL);
5888 if (ret) {
5889 errno = ret;
5890 PERROR("pthread_create apps");
5891 retval = -1;
5892 goto exit_apps;
5893 }
5894
5895 /* Create thread to manage application notify socket */
5896 ret = pthread_create(&apps_notify_thread, NULL,
5897 ust_thread_manage_notify, (void *) NULL);
5898 if (ret) {
5899 errno = ret;
5900 PERROR("pthread_create notify");
5901 retval = -1;
5902 goto exit_apps_notify;
5903 }
5904
5905 /* Create agent registration thread. */
5906 ret = pthread_create(&agent_reg_thread, NULL,
5907 agent_thread_manage_registration, (void *) NULL);
5908 if (ret) {
5909 errno = ret;
5910 PERROR("pthread_create agent");
5911 retval = -1;
5912 goto exit_agent_reg;
5913 }
5914
5915 /* Don't start this thread if kernel tracing is not requested nor root */
5916 if (is_root && !opt_no_kernel) {
5917 /* Create kernel thread to manage kernel event */
5918 ret = pthread_create(&kernel_thread, NULL,
5919 thread_manage_kernel, (void *) NULL);
5920 if (ret) {
5921 errno = ret;
5922 PERROR("pthread_create kernel");
5923 retval = -1;
5924 goto exit_kernel;
5925 }
5926 }
5927
5928 /* Create session loading thread. */
5929 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5930 load_info);
5931 if (ret) {
5932 errno = ret;
5933 PERROR("pthread_create load_session_thread");
5934 retval = -1;
5935 goto exit_load_session;
5936 }
5937
5938 /*
5939 * This is where we start awaiting program completion (e.g. through
5940 * signal that asks threads to teardown).
5941 */
5942
5943 ret = pthread_join(load_session_thread, &status);
5944 if (ret) {
5945 errno = ret;
5946 PERROR("pthread_join load_session_thread");
5947 retval = -1;
5948 }
5949 exit_load_session:
5950
5951 if (is_root && !opt_no_kernel) {
5952 ret = pthread_join(kernel_thread, &status);
5953 if (ret) {
5954 errno = ret;
5955 PERROR("pthread_join");
5956 retval = -1;
5957 }
5958 }
5959 exit_kernel:
5960
5961 ret = pthread_join(agent_reg_thread, &status);
5962 if (ret) {
5963 errno = ret;
5964 PERROR("pthread_join agent");
5965 retval = -1;
5966 }
5967 exit_agent_reg:
5968
5969 ret = pthread_join(apps_notify_thread, &status);
5970 if (ret) {
5971 errno = ret;
5972 PERROR("pthread_join apps notify");
5973 retval = -1;
5974 }
5975 exit_apps_notify:
5976
5977 ret = pthread_join(apps_thread, &status);
5978 if (ret) {
5979 errno = ret;
5980 PERROR("pthread_join apps");
5981 retval = -1;
5982 }
5983 exit_apps:
5984
5985 ret = pthread_join(reg_apps_thread, &status);
5986 if (ret) {
5987 errno = ret;
5988 PERROR("pthread_join");
5989 retval = -1;
5990 }
5991 exit_reg_apps:
5992
5993 /*
5994 * Join dispatch thread after joining reg_apps_thread to ensure
5995 * we don't leak applications in the queue.
5996 */
5997 ret = pthread_join(dispatch_thread, &status);
5998 if (ret) {
5999 errno = ret;
6000 PERROR("pthread_join");
6001 retval = -1;
6002 }
6003 exit_dispatch:
6004
6005 ret = pthread_join(client_thread, &status);
6006 if (ret) {
6007 errno = ret;
6008 PERROR("pthread_join");
6009 retval = -1;
6010 }
6011 exit_client:
6012
6013 ret = pthread_join(health_thread, &status);
6014 if (ret) {
6015 errno = ret;
6016 PERROR("pthread_join health thread");
6017 retval = -1;
6018 }
6019 exit_health:
6020
6021 exit_init_data:
6022 /*
6023 * sessiond_cleanup() is called when no other thread is running, except
6024 * the ht_cleanup thread, which is needed to destroy the hash tables.
6025 */
6026 rcu_thread_online();
6027 sessiond_cleanup();
6028 rcu_thread_offline();
6029 rcu_unregister_thread();
6030
6031 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6032 if (ret < 0) {
6033 ERR("write error on ht_cleanup quit pipe");
6034 retval = -1;
6035 }
6036
6037 ret = pthread_join(ht_cleanup_thread, &status);
6038 if (ret) {
6039 errno = ret;
6040 PERROR("pthread_join ht cleanup thread");
6041 retval = -1;
6042 }
6043 exit_ht_cleanup:
6044 exit_set_max_size:
6045
6046 utils_close_pipe(ht_cleanup_pipe);
6047 exit_ht_cleanup_pipe:
6048
6049 /*
6050 * Close the ht_cleanup quit pipe.
6051 */
6052 utils_close_pipe(ht_cleanup_quit_pipe);
6053 exit_ht_cleanup_quit_pipe:
6054
6055 health_app_destroy(health_sessiond);
6056 exit_health_sessiond_cleanup:
6057
6058 exit_options:
6059 sessiond_cleanup_options();
6060
6061 exit_set_signal_handler:
6062 /* Ensure all prior call_rcu are done. */
6063 rcu_barrier();
6064
6065 if (!retval) {
6066 exit(EXIT_SUCCESS);
6067 } else {
6068 exit(EXIT_FAILURE);
6069 }
6070 }
This page took 0.152585 seconds and 5 git commands to generate.