Fix: honor negative (unlimited) app socket timeout
[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 run_as_destroy_worker();
801
802 /* <fun> */
803 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
804 "Matthew, BEET driven development works!%c[%dm",
805 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
806 /* </fun> */
807 }
808
809 /*
810 * Send data on a unix socket using the liblttsessiondcomm API.
811 *
812 * Return lttcomm error code.
813 */
814 static int send_unix_sock(int sock, void *buf, size_t len)
815 {
816 /* Check valid length */
817 if (len == 0) {
818 return -1;
819 }
820
821 return lttcomm_send_unix_sock(sock, buf, len);
822 }
823
824 /*
825 * Free memory of a command context structure.
826 */
827 static void clean_command_ctx(struct command_ctx **cmd_ctx)
828 {
829 DBG("Clean command context structure");
830 if (*cmd_ctx) {
831 if ((*cmd_ctx)->llm) {
832 free((*cmd_ctx)->llm);
833 }
834 if ((*cmd_ctx)->lsm) {
835 free((*cmd_ctx)->lsm);
836 }
837 free(*cmd_ctx);
838 *cmd_ctx = NULL;
839 }
840 }
841
842 /*
843 * Notify UST applications using the shm mmap futex.
844 */
845 static int notify_ust_apps(int active)
846 {
847 char *wait_shm_mmap;
848
849 DBG("Notifying applications of session daemon state: %d", active);
850
851 /* See shm.c for this call implying mmap, shm and futex calls */
852 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
853 if (wait_shm_mmap == NULL) {
854 goto error;
855 }
856
857 /* Wake waiting process */
858 futex_wait_update((int32_t *) wait_shm_mmap, active);
859
860 /* Apps notified successfully */
861 return 0;
862
863 error:
864 return -1;
865 }
866
867 /*
868 * Setup the outgoing data buffer for the response (llm) by allocating the
869 * right amount of memory and copying the original information from the lsm
870 * structure.
871 *
872 * Return total size of the buffer pointed by buf.
873 */
874 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
875 {
876 int ret, buf_size;
877
878 buf_size = size;
879
880 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
881 if (cmd_ctx->llm == NULL) {
882 PERROR("zmalloc");
883 ret = -ENOMEM;
884 goto error;
885 }
886
887 /* Copy common data */
888 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
889 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
890
891 cmd_ctx->llm->data_size = size;
892 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
893
894 return buf_size;
895
896 error:
897 return ret;
898 }
899
900 /*
901 * Update the kernel poll set of all channel fd available over all tracing
902 * session. Add the wakeup pipe at the end of the set.
903 */
904 static int update_kernel_poll(struct lttng_poll_event *events)
905 {
906 int ret;
907 struct ltt_session *session;
908 struct ltt_kernel_channel *channel;
909
910 DBG("Updating kernel poll set");
911
912 session_lock_list();
913 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
914 session_lock(session);
915 if (session->kernel_session == NULL) {
916 session_unlock(session);
917 continue;
918 }
919
920 cds_list_for_each_entry(channel,
921 &session->kernel_session->channel_list.head, list) {
922 /* Add channel fd to the kernel poll set */
923 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
924 if (ret < 0) {
925 session_unlock(session);
926 goto error;
927 }
928 DBG("Channel fd %d added to kernel set", channel->fd);
929 }
930 session_unlock(session);
931 }
932 session_unlock_list();
933
934 return 0;
935
936 error:
937 session_unlock_list();
938 return -1;
939 }
940
941 /*
942 * Find the channel fd from 'fd' over all tracing session. When found, check
943 * for new channel stream and send those stream fds to the kernel consumer.
944 *
945 * Useful for CPU hotplug feature.
946 */
947 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
948 {
949 int ret = 0;
950 struct ltt_session *session;
951 struct ltt_kernel_session *ksess;
952 struct ltt_kernel_channel *channel;
953
954 DBG("Updating kernel streams for channel fd %d", fd);
955
956 session_lock_list();
957 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
958 session_lock(session);
959 if (session->kernel_session == NULL) {
960 session_unlock(session);
961 continue;
962 }
963 ksess = session->kernel_session;
964
965 cds_list_for_each_entry(channel,
966 &ksess->channel_list.head, list) {
967 struct lttng_ht_iter iter;
968 struct consumer_socket *socket;
969
970 if (channel->fd != fd) {
971 continue;
972 }
973 DBG("Channel found, updating kernel streams");
974 ret = kernel_open_channel_stream(channel);
975 if (ret < 0) {
976 goto error;
977 }
978 /* Update the stream global counter */
979 ksess->stream_count_global += ret;
980
981 /*
982 * Have we already sent fds to the consumer? If yes, it
983 * means that tracing is started so it is safe to send
984 * our updated stream fds.
985 */
986 if (ksess->consumer_fds_sent != 1
987 || ksess->consumer == NULL) {
988 ret = -1;
989 goto error;
990 }
991
992 rcu_read_lock();
993 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
994 &iter.iter, socket, node.node) {
995 pthread_mutex_lock(socket->lock);
996 ret = kernel_consumer_send_channel_stream(socket,
997 channel, ksess,
998 session->output_traces ? 1 : 0);
999 pthread_mutex_unlock(socket->lock);
1000 if (ret < 0) {
1001 rcu_read_unlock();
1002 goto error;
1003 }
1004 }
1005 rcu_read_unlock();
1006 }
1007 session_unlock(session);
1008 }
1009 session_unlock_list();
1010 return ret;
1011
1012 error:
1013 session_unlock(session);
1014 session_unlock_list();
1015 return ret;
1016 }
1017
1018 /*
1019 * For each tracing session, update newly registered apps. The session list
1020 * lock MUST be acquired before calling this.
1021 */
1022 static void update_ust_app(int app_sock)
1023 {
1024 struct ltt_session *sess, *stmp;
1025
1026 /* Consumer is in an ERROR state. Stop any application update. */
1027 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1028 /* Stop the update process since the consumer is dead. */
1029 return;
1030 }
1031
1032 /* For all tracing session(s) */
1033 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1034 struct ust_app *app;
1035
1036 session_lock(sess);
1037 if (!sess->ust_session) {
1038 goto unlock_session;
1039 }
1040
1041 rcu_read_lock();
1042 assert(app_sock >= 0);
1043 app = ust_app_find_by_sock(app_sock);
1044 if (app == NULL) {
1045 /*
1046 * Application can be unregistered before so
1047 * this is possible hence simply stopping the
1048 * update.
1049 */
1050 DBG3("UST app update failed to find app sock %d",
1051 app_sock);
1052 goto unlock_rcu;
1053 }
1054 ust_app_global_update(sess->ust_session, app);
1055 unlock_rcu:
1056 rcu_read_unlock();
1057 unlock_session:
1058 session_unlock(sess);
1059 }
1060 }
1061
1062 /*
1063 * This thread manage event coming from the kernel.
1064 *
1065 * Features supported in this thread:
1066 * -) CPU Hotplug
1067 */
1068 static void *thread_manage_kernel(void *data)
1069 {
1070 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1071 uint32_t revents, nb_fd;
1072 char tmp;
1073 struct lttng_poll_event events;
1074
1075 DBG("[thread] Thread manage kernel started");
1076
1077 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1078
1079 /*
1080 * This first step of the while is to clean this structure which could free
1081 * non NULL pointers so initialize it before the loop.
1082 */
1083 lttng_poll_init(&events);
1084
1085 if (testpoint(sessiond_thread_manage_kernel)) {
1086 goto error_testpoint;
1087 }
1088
1089 health_code_update();
1090
1091 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1092 goto error_testpoint;
1093 }
1094
1095 while (1) {
1096 health_code_update();
1097
1098 if (update_poll_flag == 1) {
1099 /* Clean events object. We are about to populate it again. */
1100 lttng_poll_clean(&events);
1101
1102 ret = sessiond_set_thread_pollset(&events, 2);
1103 if (ret < 0) {
1104 goto error_poll_create;
1105 }
1106
1107 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1108 if (ret < 0) {
1109 goto error;
1110 }
1111
1112 /* This will add the available kernel channel if any. */
1113 ret = update_kernel_poll(&events);
1114 if (ret < 0) {
1115 goto error;
1116 }
1117 update_poll_flag = 0;
1118 }
1119
1120 DBG("Thread kernel polling");
1121
1122 /* Poll infinite value of time */
1123 restart:
1124 health_poll_entry();
1125 ret = lttng_poll_wait(&events, -1);
1126 DBG("Thread kernel return from poll on %d fds",
1127 LTTNG_POLL_GETNB(&events));
1128 health_poll_exit();
1129 if (ret < 0) {
1130 /*
1131 * Restart interrupted system call.
1132 */
1133 if (errno == EINTR) {
1134 goto restart;
1135 }
1136 goto error;
1137 } else if (ret == 0) {
1138 /* Should not happen since timeout is infinite */
1139 ERR("Return value of poll is 0 with an infinite timeout.\n"
1140 "This should not have happened! Continuing...");
1141 continue;
1142 }
1143
1144 nb_fd = ret;
1145
1146 for (i = 0; i < nb_fd; i++) {
1147 /* Fetch once the poll data */
1148 revents = LTTNG_POLL_GETEV(&events, i);
1149 pollfd = LTTNG_POLL_GETFD(&events, i);
1150
1151 health_code_update();
1152
1153 if (!revents) {
1154 /* No activity for this FD (poll implementation). */
1155 continue;
1156 }
1157
1158 /* Thread quit pipe has been closed. Killing thread. */
1159 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1160 if (ret) {
1161 err = 0;
1162 goto exit;
1163 }
1164
1165 /* Check for data on kernel pipe */
1166 if (revents & LPOLLIN) {
1167 if (pollfd == kernel_poll_pipe[0]) {
1168 (void) lttng_read(kernel_poll_pipe[0],
1169 &tmp, 1);
1170 /*
1171 * Ret value is useless here, if this pipe gets any actions an
1172 * update is required anyway.
1173 */
1174 update_poll_flag = 1;
1175 continue;
1176 } else {
1177 /*
1178 * New CPU detected by the kernel. Adding kernel stream to
1179 * kernel session and updating the kernel consumer
1180 */
1181 ret = update_kernel_stream(&kconsumer_data, pollfd);
1182 if (ret < 0) {
1183 continue;
1184 }
1185 break;
1186 }
1187 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1188 update_poll_flag = 1;
1189 continue;
1190 } else {
1191 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1192 goto error;
1193 }
1194 }
1195 }
1196
1197 exit:
1198 error:
1199 lttng_poll_clean(&events);
1200 error_poll_create:
1201 error_testpoint:
1202 utils_close_pipe(kernel_poll_pipe);
1203 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1204 if (err) {
1205 health_error();
1206 ERR("Health error occurred in %s", __func__);
1207 WARN("Kernel thread died unexpectedly. "
1208 "Kernel tracing can continue but CPU hotplug is disabled.");
1209 }
1210 health_unregister(health_sessiond);
1211 DBG("Kernel thread dying");
1212 return NULL;
1213 }
1214
1215 /*
1216 * Signal pthread condition of the consumer data that the thread.
1217 */
1218 static void signal_consumer_condition(struct consumer_data *data, int state)
1219 {
1220 pthread_mutex_lock(&data->cond_mutex);
1221
1222 /*
1223 * The state is set before signaling. It can be any value, it's the waiter
1224 * job to correctly interpret this condition variable associated to the
1225 * consumer pthread_cond.
1226 *
1227 * A value of 0 means that the corresponding thread of the consumer data
1228 * was not started. 1 indicates that the thread has started and is ready
1229 * for action. A negative value means that there was an error during the
1230 * thread bootstrap.
1231 */
1232 data->consumer_thread_is_ready = state;
1233 (void) pthread_cond_signal(&data->cond);
1234
1235 pthread_mutex_unlock(&data->cond_mutex);
1236 }
1237
1238 /*
1239 * This thread manage the consumer error sent back to the session daemon.
1240 */
1241 static void *thread_manage_consumer(void *data)
1242 {
1243 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1244 uint32_t revents, nb_fd;
1245 enum lttcomm_return_code code;
1246 struct lttng_poll_event events;
1247 struct consumer_data *consumer_data = data;
1248
1249 DBG("[thread] Manage consumer started");
1250
1251 rcu_register_thread();
1252 rcu_thread_online();
1253
1254 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1255
1256 health_code_update();
1257
1258 /*
1259 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1260 * metadata_sock. Nothing more will be added to this poll set.
1261 */
1262 ret = sessiond_set_thread_pollset(&events, 3);
1263 if (ret < 0) {
1264 goto error_poll;
1265 }
1266
1267 /*
1268 * The error socket here is already in a listening state which was done
1269 * just before spawning this thread to avoid a race between the consumer
1270 * daemon exec trying to connect and the listen() call.
1271 */
1272 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1273 if (ret < 0) {
1274 goto error;
1275 }
1276
1277 health_code_update();
1278
1279 /* Infinite blocking call, waiting for transmission */
1280 restart:
1281 health_poll_entry();
1282
1283 if (testpoint(sessiond_thread_manage_consumer)) {
1284 goto error;
1285 }
1286
1287 ret = lttng_poll_wait(&events, -1);
1288 health_poll_exit();
1289 if (ret < 0) {
1290 /*
1291 * Restart interrupted system call.
1292 */
1293 if (errno == EINTR) {
1294 goto restart;
1295 }
1296 goto error;
1297 }
1298
1299 nb_fd = ret;
1300
1301 for (i = 0; i < nb_fd; i++) {
1302 /* Fetch once the poll data */
1303 revents = LTTNG_POLL_GETEV(&events, i);
1304 pollfd = LTTNG_POLL_GETFD(&events, i);
1305
1306 health_code_update();
1307
1308 if (!revents) {
1309 /* No activity for this FD (poll implementation). */
1310 continue;
1311 }
1312
1313 /* Thread quit pipe has been closed. Killing thread. */
1314 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1315 if (ret) {
1316 err = 0;
1317 goto exit;
1318 }
1319
1320 /* Event on the registration socket */
1321 if (pollfd == consumer_data->err_sock) {
1322 if (revents & LPOLLIN) {
1323 continue;
1324 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1325 ERR("consumer err socket poll error");
1326 goto error;
1327 } else {
1328 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1329 goto error;
1330 }
1331 }
1332 }
1333
1334 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1335 if (sock < 0) {
1336 goto error;
1337 }
1338
1339 /*
1340 * Set the CLOEXEC flag. Return code is useless because either way, the
1341 * show must go on.
1342 */
1343 (void) utils_set_fd_cloexec(sock);
1344
1345 health_code_update();
1346
1347 DBG2("Receiving code from consumer err_sock");
1348
1349 /* Getting status code from kconsumerd */
1350 ret = lttcomm_recv_unix_sock(sock, &code,
1351 sizeof(enum lttcomm_return_code));
1352 if (ret <= 0) {
1353 goto error;
1354 }
1355
1356 health_code_update();
1357 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1358 /* Connect both socket, command and metadata. */
1359 consumer_data->cmd_sock =
1360 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1361 consumer_data->metadata_fd =
1362 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1363 if (consumer_data->cmd_sock < 0
1364 || consumer_data->metadata_fd < 0) {
1365 PERROR("consumer connect cmd socket");
1366 /* On error, signal condition and quit. */
1367 signal_consumer_condition(consumer_data, -1);
1368 goto error;
1369 }
1370 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1371 /* Create metadata socket lock. */
1372 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1373 if (consumer_data->metadata_sock.lock == NULL) {
1374 PERROR("zmalloc pthread mutex");
1375 ret = -1;
1376 goto error;
1377 }
1378 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1379
1380 signal_consumer_condition(consumer_data, 1);
1381 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1382 DBG("Consumer metadata socket ready (fd: %d)",
1383 consumer_data->metadata_fd);
1384 } else {
1385 ERR("consumer error when waiting for SOCK_READY : %s",
1386 lttcomm_get_readable_code(-code));
1387 goto error;
1388 }
1389
1390 /* Remove the consumerd error sock since we've established a connexion */
1391 ret = lttng_poll_del(&events, consumer_data->err_sock);
1392 if (ret < 0) {
1393 goto error;
1394 }
1395
1396 /* Add new accepted error socket. */
1397 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1398 if (ret < 0) {
1399 goto error;
1400 }
1401
1402 /* Add metadata socket that is successfully connected. */
1403 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1404 LPOLLIN | LPOLLRDHUP);
1405 if (ret < 0) {
1406 goto error;
1407 }
1408
1409 health_code_update();
1410
1411 /* Infinite blocking call, waiting for transmission */
1412 restart_poll:
1413 while (1) {
1414 health_code_update();
1415
1416 /* Exit the thread because the thread quit pipe has been triggered. */
1417 if (should_quit) {
1418 /* Not a health error. */
1419 err = 0;
1420 goto exit;
1421 }
1422
1423 health_poll_entry();
1424 ret = lttng_poll_wait(&events, -1);
1425 health_poll_exit();
1426 if (ret < 0) {
1427 /*
1428 * Restart interrupted system call.
1429 */
1430 if (errno == EINTR) {
1431 goto restart_poll;
1432 }
1433 goto error;
1434 }
1435
1436 nb_fd = ret;
1437
1438 for (i = 0; i < nb_fd; i++) {
1439 /* Fetch once the poll data */
1440 revents = LTTNG_POLL_GETEV(&events, i);
1441 pollfd = LTTNG_POLL_GETFD(&events, i);
1442
1443 health_code_update();
1444
1445 if (!revents) {
1446 /* No activity for this FD (poll implementation). */
1447 continue;
1448 }
1449
1450 /*
1451 * Thread quit pipe has been triggered, flag that we should stop
1452 * but continue the current loop to handle potential data from
1453 * consumer.
1454 */
1455 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1456
1457 if (pollfd == sock) {
1458 /* Event on the consumerd socket */
1459 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1460 && !(revents & LPOLLIN)) {
1461 ERR("consumer err socket second poll error");
1462 goto error;
1463 }
1464 health_code_update();
1465 /* Wait for any kconsumerd error */
1466 ret = lttcomm_recv_unix_sock(sock, &code,
1467 sizeof(enum lttcomm_return_code));
1468 if (ret <= 0) {
1469 ERR("consumer closed the command socket");
1470 goto error;
1471 }
1472
1473 ERR("consumer return code : %s",
1474 lttcomm_get_readable_code(-code));
1475
1476 goto exit;
1477 } else if (pollfd == consumer_data->metadata_fd) {
1478 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1479 && !(revents & LPOLLIN)) {
1480 ERR("consumer err metadata socket second poll error");
1481 goto error;
1482 }
1483 /* UST metadata requests */
1484 ret = ust_consumer_metadata_request(
1485 &consumer_data->metadata_sock);
1486 if (ret < 0) {
1487 ERR("Handling metadata request");
1488 goto error;
1489 }
1490 }
1491 /* No need for an else branch all FDs are tested prior. */
1492 }
1493 health_code_update();
1494 }
1495
1496 exit:
1497 error:
1498 /*
1499 * We lock here because we are about to close the sockets and some other
1500 * thread might be using them so get exclusive access which will abort all
1501 * other consumer command by other threads.
1502 */
1503 pthread_mutex_lock(&consumer_data->lock);
1504
1505 /* Immediately set the consumerd state to stopped */
1506 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1507 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1508 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1509 consumer_data->type == LTTNG_CONSUMER32_UST) {
1510 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1511 } else {
1512 /* Code flow error... */
1513 assert(0);
1514 }
1515
1516 if (consumer_data->err_sock >= 0) {
1517 ret = close(consumer_data->err_sock);
1518 if (ret) {
1519 PERROR("close");
1520 }
1521 consumer_data->err_sock = -1;
1522 }
1523 if (consumer_data->cmd_sock >= 0) {
1524 ret = close(consumer_data->cmd_sock);
1525 if (ret) {
1526 PERROR("close");
1527 }
1528 consumer_data->cmd_sock = -1;
1529 }
1530 if (consumer_data->metadata_sock.fd_ptr &&
1531 *consumer_data->metadata_sock.fd_ptr >= 0) {
1532 ret = close(*consumer_data->metadata_sock.fd_ptr);
1533 if (ret) {
1534 PERROR("close");
1535 }
1536 }
1537 if (sock >= 0) {
1538 ret = close(sock);
1539 if (ret) {
1540 PERROR("close");
1541 }
1542 }
1543
1544 unlink(consumer_data->err_unix_sock_path);
1545 unlink(consumer_data->cmd_unix_sock_path);
1546 pthread_mutex_unlock(&consumer_data->lock);
1547
1548 /* Cleanup metadata socket mutex. */
1549 if (consumer_data->metadata_sock.lock) {
1550 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1551 free(consumer_data->metadata_sock.lock);
1552 }
1553 lttng_poll_clean(&events);
1554 error_poll:
1555 if (err) {
1556 health_error();
1557 ERR("Health error occurred in %s", __func__);
1558 }
1559 health_unregister(health_sessiond);
1560 DBG("consumer thread cleanup completed");
1561
1562 rcu_thread_offline();
1563 rcu_unregister_thread();
1564
1565 return NULL;
1566 }
1567
1568 /*
1569 * This thread manage application communication.
1570 */
1571 static void *thread_manage_apps(void *data)
1572 {
1573 int i, ret, pollfd, err = -1;
1574 ssize_t size_ret;
1575 uint32_t revents, nb_fd;
1576 struct lttng_poll_event events;
1577
1578 DBG("[thread] Manage application started");
1579
1580 rcu_register_thread();
1581 rcu_thread_online();
1582
1583 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1584
1585 if (testpoint(sessiond_thread_manage_apps)) {
1586 goto error_testpoint;
1587 }
1588
1589 health_code_update();
1590
1591 ret = sessiond_set_thread_pollset(&events, 2);
1592 if (ret < 0) {
1593 goto error_poll_create;
1594 }
1595
1596 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1597 if (ret < 0) {
1598 goto error;
1599 }
1600
1601 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1602 goto error;
1603 }
1604
1605 health_code_update();
1606
1607 while (1) {
1608 DBG("Apps thread polling");
1609
1610 /* Inifinite blocking call, waiting for transmission */
1611 restart:
1612 health_poll_entry();
1613 ret = lttng_poll_wait(&events, -1);
1614 DBG("Apps thread return from poll on %d fds",
1615 LTTNG_POLL_GETNB(&events));
1616 health_poll_exit();
1617 if (ret < 0) {
1618 /*
1619 * Restart interrupted system call.
1620 */
1621 if (errno == EINTR) {
1622 goto restart;
1623 }
1624 goto error;
1625 }
1626
1627 nb_fd = ret;
1628
1629 for (i = 0; i < nb_fd; i++) {
1630 /* Fetch once the poll data */
1631 revents = LTTNG_POLL_GETEV(&events, i);
1632 pollfd = LTTNG_POLL_GETFD(&events, i);
1633
1634 health_code_update();
1635
1636 if (!revents) {
1637 /* No activity for this FD (poll implementation). */
1638 continue;
1639 }
1640
1641 /* Thread quit pipe has been closed. Killing thread. */
1642 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1643 if (ret) {
1644 err = 0;
1645 goto exit;
1646 }
1647
1648 /* Inspect the apps cmd pipe */
1649 if (pollfd == apps_cmd_pipe[0]) {
1650 if (revents & LPOLLIN) {
1651 int sock;
1652
1653 /* Empty pipe */
1654 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1655 if (size_ret < sizeof(sock)) {
1656 PERROR("read apps cmd pipe");
1657 goto error;
1658 }
1659
1660 health_code_update();
1661
1662 /*
1663 * Since this is a command socket (write then read),
1664 * we only monitor the error events of the socket.
1665 */
1666 ret = lttng_poll_add(&events, sock,
1667 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1668 if (ret < 0) {
1669 goto error;
1670 }
1671
1672 DBG("Apps with sock %d added to poll set", sock);
1673 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1674 ERR("Apps command pipe error");
1675 goto error;
1676 } else {
1677 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1678 goto error;
1679 }
1680 } else {
1681 /*
1682 * At this point, we know that a registered application made
1683 * the event at poll_wait.
1684 */
1685 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1686 /* Removing from the poll set */
1687 ret = lttng_poll_del(&events, pollfd);
1688 if (ret < 0) {
1689 goto error;
1690 }
1691
1692 /* Socket closed on remote end. */
1693 ust_app_unregister(pollfd);
1694 } else {
1695 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1696 goto error;
1697 }
1698 }
1699
1700 health_code_update();
1701 }
1702 }
1703
1704 exit:
1705 error:
1706 lttng_poll_clean(&events);
1707 error_poll_create:
1708 error_testpoint:
1709 utils_close_pipe(apps_cmd_pipe);
1710 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1711
1712 /*
1713 * We don't clean the UST app hash table here since already registered
1714 * applications can still be controlled so let them be until the session
1715 * daemon dies or the applications stop.
1716 */
1717
1718 if (err) {
1719 health_error();
1720 ERR("Health error occurred in %s", __func__);
1721 }
1722 health_unregister(health_sessiond);
1723 DBG("Application communication apps thread cleanup complete");
1724 rcu_thread_offline();
1725 rcu_unregister_thread();
1726 return NULL;
1727 }
1728
1729 /*
1730 * Send a socket to a thread This is called from the dispatch UST registration
1731 * thread once all sockets are set for the application.
1732 *
1733 * The sock value can be invalid, we don't really care, the thread will handle
1734 * it and make the necessary cleanup if so.
1735 *
1736 * On success, return 0 else a negative value being the errno message of the
1737 * write().
1738 */
1739 static int send_socket_to_thread(int fd, int sock)
1740 {
1741 ssize_t ret;
1742
1743 /*
1744 * It's possible that the FD is set as invalid with -1 concurrently just
1745 * before calling this function being a shutdown state of the thread.
1746 */
1747 if (fd < 0) {
1748 ret = -EBADF;
1749 goto error;
1750 }
1751
1752 ret = lttng_write(fd, &sock, sizeof(sock));
1753 if (ret < sizeof(sock)) {
1754 PERROR("write apps pipe %d", fd);
1755 if (ret < 0) {
1756 ret = -errno;
1757 }
1758 goto error;
1759 }
1760
1761 /* All good. Don't send back the write positive ret value. */
1762 ret = 0;
1763 error:
1764 return (int) ret;
1765 }
1766
1767 /*
1768 * Sanitize the wait queue of the dispatch registration thread meaning removing
1769 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1770 * notify socket is never received.
1771 */
1772 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1773 {
1774 int ret, nb_fd = 0, i;
1775 unsigned int fd_added = 0;
1776 struct lttng_poll_event events;
1777 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1778
1779 assert(wait_queue);
1780
1781 lttng_poll_init(&events);
1782
1783 /* Just skip everything for an empty queue. */
1784 if (!wait_queue->count) {
1785 goto end;
1786 }
1787
1788 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1789 if (ret < 0) {
1790 goto error_create;
1791 }
1792
1793 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1794 &wait_queue->head, head) {
1795 assert(wait_node->app);
1796 ret = lttng_poll_add(&events, wait_node->app->sock,
1797 LPOLLHUP | LPOLLERR);
1798 if (ret < 0) {
1799 goto error;
1800 }
1801
1802 fd_added = 1;
1803 }
1804
1805 if (!fd_added) {
1806 goto end;
1807 }
1808
1809 /*
1810 * Poll but don't block so we can quickly identify the faulty events and
1811 * clean them afterwards from the wait queue.
1812 */
1813 ret = lttng_poll_wait(&events, 0);
1814 if (ret < 0) {
1815 goto error;
1816 }
1817 nb_fd = ret;
1818
1819 for (i = 0; i < nb_fd; i++) {
1820 /* Get faulty FD. */
1821 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1822 int pollfd = LTTNG_POLL_GETFD(&events, i);
1823
1824 if (!revents) {
1825 /* No activity for this FD (poll implementation). */
1826 continue;
1827 }
1828
1829 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1830 &wait_queue->head, head) {
1831 if (pollfd == wait_node->app->sock &&
1832 (revents & (LPOLLHUP | LPOLLERR))) {
1833 cds_list_del(&wait_node->head);
1834 wait_queue->count--;
1835 ust_app_destroy(wait_node->app);
1836 free(wait_node);
1837 break;
1838 } else {
1839 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1840 goto error;
1841 }
1842 }
1843 }
1844
1845 if (nb_fd > 0) {
1846 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1847 }
1848
1849 end:
1850 lttng_poll_clean(&events);
1851 return;
1852
1853 error:
1854 lttng_poll_clean(&events);
1855 error_create:
1856 ERR("Unable to sanitize wait queue");
1857 return;
1858 }
1859
1860 /*
1861 * Dispatch request from the registration threads to the application
1862 * communication thread.
1863 */
1864 static void *thread_dispatch_ust_registration(void *data)
1865 {
1866 int ret, err = -1;
1867 struct cds_wfcq_node *node;
1868 struct ust_command *ust_cmd = NULL;
1869 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1870 struct ust_reg_wait_queue wait_queue = {
1871 .count = 0,
1872 };
1873
1874 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1875
1876 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1877 goto error_testpoint;
1878 }
1879
1880 health_code_update();
1881
1882 CDS_INIT_LIST_HEAD(&wait_queue.head);
1883
1884 DBG("[thread] Dispatch UST command started");
1885
1886 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1887 health_code_update();
1888
1889 /* Atomically prepare the queue futex */
1890 futex_nto1_prepare(&ust_cmd_queue.futex);
1891
1892 do {
1893 struct ust_app *app = NULL;
1894 ust_cmd = NULL;
1895
1896 /*
1897 * Make sure we don't have node(s) that have hung up before receiving
1898 * the notify socket. This is to clean the list in order to avoid
1899 * memory leaks from notify socket that are never seen.
1900 */
1901 sanitize_wait_queue(&wait_queue);
1902
1903 health_code_update();
1904 /* Dequeue command for registration */
1905 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1906 if (node == NULL) {
1907 DBG("Woken up but nothing in the UST command queue");
1908 /* Continue thread execution */
1909 break;
1910 }
1911
1912 ust_cmd = caa_container_of(node, struct ust_command, node);
1913
1914 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1915 " gid:%d sock:%d name:%s (version %d.%d)",
1916 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1917 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1918 ust_cmd->sock, ust_cmd->reg_msg.name,
1919 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1920
1921 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1922 wait_node = zmalloc(sizeof(*wait_node));
1923 if (!wait_node) {
1924 PERROR("zmalloc wait_node dispatch");
1925 ret = close(ust_cmd->sock);
1926 if (ret < 0) {
1927 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1928 }
1929 lttng_fd_put(LTTNG_FD_APPS, 1);
1930 free(ust_cmd);
1931 goto error;
1932 }
1933 CDS_INIT_LIST_HEAD(&wait_node->head);
1934
1935 /* Create application object if socket is CMD. */
1936 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1937 ust_cmd->sock);
1938 if (!wait_node->app) {
1939 ret = close(ust_cmd->sock);
1940 if (ret < 0) {
1941 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1942 }
1943 lttng_fd_put(LTTNG_FD_APPS, 1);
1944 free(wait_node);
1945 free(ust_cmd);
1946 continue;
1947 }
1948 /*
1949 * Add application to the wait queue so we can set the notify
1950 * socket before putting this object in the global ht.
1951 */
1952 cds_list_add(&wait_node->head, &wait_queue.head);
1953 wait_queue.count++;
1954
1955 free(ust_cmd);
1956 /*
1957 * We have to continue here since we don't have the notify
1958 * socket and the application MUST be added to the hash table
1959 * only at that moment.
1960 */
1961 continue;
1962 } else {
1963 /*
1964 * Look for the application in the local wait queue and set the
1965 * notify socket if found.
1966 */
1967 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1968 &wait_queue.head, head) {
1969 health_code_update();
1970 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1971 wait_node->app->notify_sock = ust_cmd->sock;
1972 cds_list_del(&wait_node->head);
1973 wait_queue.count--;
1974 app = wait_node->app;
1975 free(wait_node);
1976 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1977 break;
1978 }
1979 }
1980
1981 /*
1982 * With no application at this stage the received socket is
1983 * basically useless so close it before we free the cmd data
1984 * structure for good.
1985 */
1986 if (!app) {
1987 ret = close(ust_cmd->sock);
1988 if (ret < 0) {
1989 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1990 }
1991 lttng_fd_put(LTTNG_FD_APPS, 1);
1992 }
1993 free(ust_cmd);
1994 }
1995
1996 if (app) {
1997 /*
1998 * @session_lock_list
1999 *
2000 * Lock the global session list so from the register up to the
2001 * registration done message, no thread can see the application
2002 * and change its state.
2003 */
2004 session_lock_list();
2005 rcu_read_lock();
2006
2007 /*
2008 * Add application to the global hash table. This needs to be
2009 * done before the update to the UST registry can locate the
2010 * application.
2011 */
2012 ust_app_add(app);
2013
2014 /* Set app version. This call will print an error if needed. */
2015 (void) ust_app_version(app);
2016
2017 /* Send notify socket through the notify pipe. */
2018 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2019 app->notify_sock);
2020 if (ret < 0) {
2021 rcu_read_unlock();
2022 session_unlock_list();
2023 /*
2024 * No notify thread, stop the UST tracing. However, this is
2025 * not an internal error of the this thread thus setting
2026 * the health error code to a normal exit.
2027 */
2028 err = 0;
2029 goto error;
2030 }
2031
2032 /*
2033 * Update newly registered application with the tracing
2034 * registry info already enabled information.
2035 */
2036 update_ust_app(app->sock);
2037
2038 /*
2039 * Don't care about return value. Let the manage apps threads
2040 * handle app unregistration upon socket close.
2041 */
2042 (void) ust_app_register_done(app);
2043
2044 /*
2045 * Even if the application socket has been closed, send the app
2046 * to the thread and unregistration will take place at that
2047 * place.
2048 */
2049 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2050 if (ret < 0) {
2051 rcu_read_unlock();
2052 session_unlock_list();
2053 /*
2054 * No apps. thread, stop the UST tracing. However, this is
2055 * not an internal error of the this thread thus setting
2056 * the health error code to a normal exit.
2057 */
2058 err = 0;
2059 goto error;
2060 }
2061
2062 rcu_read_unlock();
2063 session_unlock_list();
2064 }
2065 } while (node != NULL);
2066
2067 health_poll_entry();
2068 /* Futex wait on queue. Blocking call on futex() */
2069 futex_nto1_wait(&ust_cmd_queue.futex);
2070 health_poll_exit();
2071 }
2072 /* Normal exit, no error */
2073 err = 0;
2074
2075 error:
2076 /* Clean up wait queue. */
2077 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2078 &wait_queue.head, head) {
2079 cds_list_del(&wait_node->head);
2080 wait_queue.count--;
2081 free(wait_node);
2082 }
2083
2084 /* Empty command queue. */
2085 for (;;) {
2086 /* Dequeue command for registration */
2087 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2088 if (node == NULL) {
2089 break;
2090 }
2091 ust_cmd = caa_container_of(node, struct ust_command, node);
2092 ret = close(ust_cmd->sock);
2093 if (ret < 0) {
2094 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2095 }
2096 lttng_fd_put(LTTNG_FD_APPS, 1);
2097 free(ust_cmd);
2098 }
2099
2100 error_testpoint:
2101 DBG("Dispatch thread dying");
2102 if (err) {
2103 health_error();
2104 ERR("Health error occurred in %s", __func__);
2105 }
2106 health_unregister(health_sessiond);
2107 return NULL;
2108 }
2109
2110 /*
2111 * This thread manage application registration.
2112 */
2113 static void *thread_registration_apps(void *data)
2114 {
2115 int sock = -1, i, ret, pollfd, err = -1;
2116 uint32_t revents, nb_fd;
2117 struct lttng_poll_event events;
2118 /*
2119 * Get allocated in this thread, enqueued to a global queue, dequeued and
2120 * freed in the manage apps thread.
2121 */
2122 struct ust_command *ust_cmd = NULL;
2123
2124 DBG("[thread] Manage application registration started");
2125
2126 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2127
2128 if (testpoint(sessiond_thread_registration_apps)) {
2129 goto error_testpoint;
2130 }
2131
2132 ret = lttcomm_listen_unix_sock(apps_sock);
2133 if (ret < 0) {
2134 goto error_listen;
2135 }
2136
2137 /*
2138 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2139 * more will be added to this poll set.
2140 */
2141 ret = sessiond_set_thread_pollset(&events, 2);
2142 if (ret < 0) {
2143 goto error_create_poll;
2144 }
2145
2146 /* Add the application registration socket */
2147 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2148 if (ret < 0) {
2149 goto error_poll_add;
2150 }
2151
2152 /* Notify all applications to register */
2153 ret = notify_ust_apps(1);
2154 if (ret < 0) {
2155 ERR("Failed to notify applications or create the wait shared memory.\n"
2156 "Execution continues but there might be problem for already\n"
2157 "running applications that wishes to register.");
2158 }
2159
2160 while (1) {
2161 DBG("Accepting application registration");
2162
2163 /* Inifinite blocking call, waiting for transmission */
2164 restart:
2165 health_poll_entry();
2166 ret = lttng_poll_wait(&events, -1);
2167 health_poll_exit();
2168 if (ret < 0) {
2169 /*
2170 * Restart interrupted system call.
2171 */
2172 if (errno == EINTR) {
2173 goto restart;
2174 }
2175 goto error;
2176 }
2177
2178 nb_fd = ret;
2179
2180 for (i = 0; i < nb_fd; i++) {
2181 health_code_update();
2182
2183 /* Fetch once the poll data */
2184 revents = LTTNG_POLL_GETEV(&events, i);
2185 pollfd = LTTNG_POLL_GETFD(&events, i);
2186
2187 if (!revents) {
2188 /* No activity for this FD (poll implementation). */
2189 continue;
2190 }
2191
2192 /* Thread quit pipe has been closed. Killing thread. */
2193 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2194 if (ret) {
2195 err = 0;
2196 goto exit;
2197 }
2198
2199 /* Event on the registration socket */
2200 if (pollfd == apps_sock) {
2201 if (revents & LPOLLIN) {
2202 sock = lttcomm_accept_unix_sock(apps_sock);
2203 if (sock < 0) {
2204 goto error;
2205 }
2206
2207 /*
2208 * Set socket timeout for both receiving and ending.
2209 * app_socket_timeout is in seconds, whereas
2210 * lttcomm_setsockopt_rcv_timeout and
2211 * lttcomm_setsockopt_snd_timeout expect msec as
2212 * parameter.
2213 */
2214 if (app_socket_timeout >= 0) {
2215 (void) lttcomm_setsockopt_rcv_timeout(sock,
2216 app_socket_timeout * 1000);
2217 (void) lttcomm_setsockopt_snd_timeout(sock,
2218 app_socket_timeout * 1000);
2219 }
2220
2221 /*
2222 * Set the CLOEXEC flag. Return code is useless because
2223 * either way, the show must go on.
2224 */
2225 (void) utils_set_fd_cloexec(sock);
2226
2227 /* Create UST registration command for enqueuing */
2228 ust_cmd = zmalloc(sizeof(struct ust_command));
2229 if (ust_cmd == NULL) {
2230 PERROR("ust command zmalloc");
2231 ret = close(sock);
2232 if (ret) {
2233 PERROR("close");
2234 }
2235 goto error;
2236 }
2237
2238 /*
2239 * Using message-based transmissions to ensure we don't
2240 * have to deal with partially received messages.
2241 */
2242 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2243 if (ret < 0) {
2244 ERR("Exhausted file descriptors allowed for applications.");
2245 free(ust_cmd);
2246 ret = close(sock);
2247 if (ret) {
2248 PERROR("close");
2249 }
2250 sock = -1;
2251 continue;
2252 }
2253
2254 health_code_update();
2255 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2256 if (ret < 0) {
2257 free(ust_cmd);
2258 /* Close socket of the application. */
2259 ret = close(sock);
2260 if (ret) {
2261 PERROR("close");
2262 }
2263 lttng_fd_put(LTTNG_FD_APPS, 1);
2264 sock = -1;
2265 continue;
2266 }
2267 health_code_update();
2268
2269 ust_cmd->sock = sock;
2270 sock = -1;
2271
2272 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2273 " gid:%d sock:%d name:%s (version %d.%d)",
2274 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2275 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2276 ust_cmd->sock, ust_cmd->reg_msg.name,
2277 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2278
2279 /*
2280 * Lock free enqueue the registration request. The red pill
2281 * has been taken! This apps will be part of the *system*.
2282 */
2283 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2284
2285 /*
2286 * Wake the registration queue futex. Implicit memory
2287 * barrier with the exchange in cds_wfcq_enqueue.
2288 */
2289 futex_nto1_wake(&ust_cmd_queue.futex);
2290 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2291 ERR("Register apps socket poll error");
2292 goto error;
2293 } else {
2294 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2295 goto error;
2296 }
2297 }
2298 }
2299 }
2300
2301 exit:
2302 error:
2303 /* Notify that the registration thread is gone */
2304 notify_ust_apps(0);
2305
2306 if (apps_sock >= 0) {
2307 ret = close(apps_sock);
2308 if (ret) {
2309 PERROR("close");
2310 }
2311 }
2312 if (sock >= 0) {
2313 ret = close(sock);
2314 if (ret) {
2315 PERROR("close");
2316 }
2317 lttng_fd_put(LTTNG_FD_APPS, 1);
2318 }
2319 unlink(apps_unix_sock_path);
2320
2321 error_poll_add:
2322 lttng_poll_clean(&events);
2323 error_listen:
2324 error_create_poll:
2325 error_testpoint:
2326 DBG("UST Registration thread cleanup complete");
2327 if (err) {
2328 health_error();
2329 ERR("Health error occurred in %s", __func__);
2330 }
2331 health_unregister(health_sessiond);
2332
2333 return NULL;
2334 }
2335
2336 /*
2337 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2338 * exec or it will fails.
2339 */
2340 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2341 {
2342 int ret, clock_ret;
2343 struct timespec timeout;
2344
2345 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2346 consumer_data->consumer_thread_is_ready = 0;
2347
2348 /* Setup pthread condition */
2349 ret = pthread_condattr_init(&consumer_data->condattr);
2350 if (ret) {
2351 errno = ret;
2352 PERROR("pthread_condattr_init consumer data");
2353 goto error;
2354 }
2355
2356 /*
2357 * Set the monotonic clock in order to make sure we DO NOT jump in time
2358 * between the clock_gettime() call and the timedwait call. See bug #324
2359 * for a more details and how we noticed it.
2360 */
2361 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2362 if (ret) {
2363 errno = ret;
2364 PERROR("pthread_condattr_setclock consumer data");
2365 goto error;
2366 }
2367
2368 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2369 if (ret) {
2370 errno = ret;
2371 PERROR("pthread_cond_init consumer data");
2372 goto error;
2373 }
2374
2375 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2376 consumer_data);
2377 if (ret) {
2378 errno = ret;
2379 PERROR("pthread_create consumer");
2380 ret = -1;
2381 goto error;
2382 }
2383
2384 /* We are about to wait on a pthread condition */
2385 pthread_mutex_lock(&consumer_data->cond_mutex);
2386
2387 /* Get time for sem_timedwait absolute timeout */
2388 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2389 /*
2390 * Set the timeout for the condition timed wait even if the clock gettime
2391 * call fails since we might loop on that call and we want to avoid to
2392 * increment the timeout too many times.
2393 */
2394 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2395
2396 /*
2397 * The following loop COULD be skipped in some conditions so this is why we
2398 * set ret to 0 in order to make sure at least one round of the loop is
2399 * done.
2400 */
2401 ret = 0;
2402
2403 /*
2404 * Loop until the condition is reached or when a timeout is reached. Note
2405 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2406 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2407 * possible. This loop does not take any chances and works with both of
2408 * them.
2409 */
2410 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2411 if (clock_ret < 0) {
2412 PERROR("clock_gettime spawn consumer");
2413 /* Infinite wait for the consumerd thread to be ready */
2414 ret = pthread_cond_wait(&consumer_data->cond,
2415 &consumer_data->cond_mutex);
2416 } else {
2417 ret = pthread_cond_timedwait(&consumer_data->cond,
2418 &consumer_data->cond_mutex, &timeout);
2419 }
2420 }
2421
2422 /* Release the pthread condition */
2423 pthread_mutex_unlock(&consumer_data->cond_mutex);
2424
2425 if (ret != 0) {
2426 errno = ret;
2427 if (ret == ETIMEDOUT) {
2428 int pth_ret;
2429
2430 /*
2431 * Call has timed out so we kill the kconsumerd_thread and return
2432 * an error.
2433 */
2434 ERR("Condition timed out. The consumer thread was never ready."
2435 " Killing it");
2436 pth_ret = pthread_cancel(consumer_data->thread);
2437 if (pth_ret < 0) {
2438 PERROR("pthread_cancel consumer thread");
2439 }
2440 } else {
2441 PERROR("pthread_cond_wait failed consumer thread");
2442 }
2443 /* Caller is expecting a negative value on failure. */
2444 ret = -1;
2445 goto error;
2446 }
2447
2448 pthread_mutex_lock(&consumer_data->pid_mutex);
2449 if (consumer_data->pid == 0) {
2450 ERR("Consumerd did not start");
2451 pthread_mutex_unlock(&consumer_data->pid_mutex);
2452 goto error;
2453 }
2454 pthread_mutex_unlock(&consumer_data->pid_mutex);
2455
2456 return 0;
2457
2458 error:
2459 return ret;
2460 }
2461
2462 /*
2463 * Join consumer thread
2464 */
2465 static int join_consumer_thread(struct consumer_data *consumer_data)
2466 {
2467 void *status;
2468
2469 /* Consumer pid must be a real one. */
2470 if (consumer_data->pid > 0) {
2471 int ret;
2472 ret = kill(consumer_data->pid, SIGTERM);
2473 if (ret) {
2474 PERROR("Error killing consumer daemon");
2475 return ret;
2476 }
2477 return pthread_join(consumer_data->thread, &status);
2478 } else {
2479 return 0;
2480 }
2481 }
2482
2483 /*
2484 * Fork and exec a consumer daemon (consumerd).
2485 *
2486 * Return pid if successful else -1.
2487 */
2488 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2489 {
2490 int ret;
2491 pid_t pid;
2492 const char *consumer_to_use;
2493 const char *verbosity;
2494 struct stat st;
2495
2496 DBG("Spawning consumerd");
2497
2498 pid = fork();
2499 if (pid == 0) {
2500 /*
2501 * Exec consumerd.
2502 */
2503 if (opt_verbose_consumer) {
2504 verbosity = "--verbose";
2505 } else if (lttng_opt_quiet) {
2506 verbosity = "--quiet";
2507 } else {
2508 verbosity = "";
2509 }
2510
2511 switch (consumer_data->type) {
2512 case LTTNG_CONSUMER_KERNEL:
2513 /*
2514 * Find out which consumerd to execute. We will first try the
2515 * 64-bit path, then the sessiond's installation directory, and
2516 * fallback on the 32-bit one,
2517 */
2518 DBG3("Looking for a kernel consumer at these locations:");
2519 DBG3(" 1) %s", consumerd64_bin);
2520 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2521 DBG3(" 3) %s", consumerd32_bin);
2522 if (stat(consumerd64_bin, &st) == 0) {
2523 DBG3("Found location #1");
2524 consumer_to_use = consumerd64_bin;
2525 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2526 DBG3("Found location #2");
2527 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2528 } else if (stat(consumerd32_bin, &st) == 0) {
2529 DBG3("Found location #3");
2530 consumer_to_use = consumerd32_bin;
2531 } else {
2532 DBG("Could not find any valid consumerd executable");
2533 ret = -EINVAL;
2534 break;
2535 }
2536 DBG("Using kernel consumer at: %s", consumer_to_use);
2537 ret = execl(consumer_to_use,
2538 "lttng-consumerd", verbosity, "-k",
2539 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2540 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2541 "--group", tracing_group_name,
2542 NULL);
2543 break;
2544 case LTTNG_CONSUMER64_UST:
2545 {
2546 char *tmpnew = NULL;
2547
2548 if (consumerd64_libdir[0] != '\0') {
2549 char *tmp;
2550 size_t tmplen;
2551
2552 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2553 if (!tmp) {
2554 tmp = "";
2555 }
2556 tmplen = strlen("LD_LIBRARY_PATH=")
2557 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2558 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2559 if (!tmpnew) {
2560 ret = -ENOMEM;
2561 goto error;
2562 }
2563 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2564 strcat(tmpnew, consumerd64_libdir);
2565 if (tmp[0] != '\0') {
2566 strcat(tmpnew, ":");
2567 strcat(tmpnew, tmp);
2568 }
2569 ret = putenv(tmpnew);
2570 if (ret) {
2571 ret = -errno;
2572 free(tmpnew);
2573 goto error;
2574 }
2575 }
2576 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2577 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2578 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2579 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2580 "--group", tracing_group_name,
2581 NULL);
2582 if (consumerd64_libdir[0] != '\0') {
2583 free(tmpnew);
2584 }
2585 break;
2586 }
2587 case LTTNG_CONSUMER32_UST:
2588 {
2589 char *tmpnew = NULL;
2590
2591 if (consumerd32_libdir[0] != '\0') {
2592 char *tmp;
2593 size_t tmplen;
2594
2595 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2596 if (!tmp) {
2597 tmp = "";
2598 }
2599 tmplen = strlen("LD_LIBRARY_PATH=")
2600 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2601 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2602 if (!tmpnew) {
2603 ret = -ENOMEM;
2604 goto error;
2605 }
2606 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2607 strcat(tmpnew, consumerd32_libdir);
2608 if (tmp[0] != '\0') {
2609 strcat(tmpnew, ":");
2610 strcat(tmpnew, tmp);
2611 }
2612 ret = putenv(tmpnew);
2613 if (ret) {
2614 ret = -errno;
2615 free(tmpnew);
2616 goto error;
2617 }
2618 }
2619 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2620 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2621 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2622 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2623 "--group", tracing_group_name,
2624 NULL);
2625 if (consumerd32_libdir[0] != '\0') {
2626 free(tmpnew);
2627 }
2628 break;
2629 }
2630 default:
2631 PERROR("unknown consumer type");
2632 exit(EXIT_FAILURE);
2633 }
2634 if (errno != 0) {
2635 PERROR("Consumer execl()");
2636 }
2637 /* Reaching this point, we got a failure on our execl(). */
2638 exit(EXIT_FAILURE);
2639 } else if (pid > 0) {
2640 ret = pid;
2641 } else {
2642 PERROR("start consumer fork");
2643 ret = -errno;
2644 }
2645 error:
2646 return ret;
2647 }
2648
2649 /*
2650 * Spawn the consumerd daemon and session daemon thread.
2651 */
2652 static int start_consumerd(struct consumer_data *consumer_data)
2653 {
2654 int ret;
2655
2656 /*
2657 * Set the listen() state on the socket since there is a possible race
2658 * between the exec() of the consumer daemon and this call if place in the
2659 * consumer thread. See bug #366 for more details.
2660 */
2661 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2662 if (ret < 0) {
2663 goto error;
2664 }
2665
2666 pthread_mutex_lock(&consumer_data->pid_mutex);
2667 if (consumer_data->pid != 0) {
2668 pthread_mutex_unlock(&consumer_data->pid_mutex);
2669 goto end;
2670 }
2671
2672 ret = spawn_consumerd(consumer_data);
2673 if (ret < 0) {
2674 ERR("Spawning consumerd failed");
2675 pthread_mutex_unlock(&consumer_data->pid_mutex);
2676 goto error;
2677 }
2678
2679 /* Setting up the consumer_data pid */
2680 consumer_data->pid = ret;
2681 DBG2("Consumer pid %d", consumer_data->pid);
2682 pthread_mutex_unlock(&consumer_data->pid_mutex);
2683
2684 DBG2("Spawning consumer control thread");
2685 ret = spawn_consumer_thread(consumer_data);
2686 if (ret < 0) {
2687 ERR("Fatal error spawning consumer control thread");
2688 goto error;
2689 }
2690
2691 end:
2692 return 0;
2693
2694 error:
2695 /* Cleanup already created sockets on error. */
2696 if (consumer_data->err_sock >= 0) {
2697 int err;
2698
2699 err = close(consumer_data->err_sock);
2700 if (err < 0) {
2701 PERROR("close consumer data error socket");
2702 }
2703 }
2704 return ret;
2705 }
2706
2707 /*
2708 * Setup necessary data for kernel tracer action.
2709 */
2710 static int init_kernel_tracer(void)
2711 {
2712 int ret;
2713
2714 /* Modprobe lttng kernel modules */
2715 ret = modprobe_lttng_control();
2716 if (ret < 0) {
2717 goto error;
2718 }
2719
2720 /* Open debugfs lttng */
2721 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2722 if (kernel_tracer_fd < 0) {
2723 DBG("Failed to open %s", module_proc_lttng);
2724 ret = -1;
2725 goto error_open;
2726 }
2727
2728 /* Validate kernel version */
2729 ret = kernel_validate_version(kernel_tracer_fd);
2730 if (ret < 0) {
2731 goto error_version;
2732 }
2733
2734 ret = modprobe_lttng_data();
2735 if (ret < 0) {
2736 goto error_modules;
2737 }
2738
2739 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2740 return 0;
2741
2742 error_version:
2743 modprobe_remove_lttng_control();
2744 ret = close(kernel_tracer_fd);
2745 if (ret) {
2746 PERROR("close");
2747 }
2748 kernel_tracer_fd = -1;
2749 return LTTNG_ERR_KERN_VERSION;
2750
2751 error_modules:
2752 ret = close(kernel_tracer_fd);
2753 if (ret) {
2754 PERROR("close");
2755 }
2756
2757 error_open:
2758 modprobe_remove_lttng_control();
2759
2760 error:
2761 WARN("No kernel tracer available");
2762 kernel_tracer_fd = -1;
2763 if (!is_root) {
2764 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2765 } else {
2766 return LTTNG_ERR_KERN_NA;
2767 }
2768 }
2769
2770
2771 /*
2772 * Copy consumer output from the tracing session to the domain session. The
2773 * function also applies the right modification on a per domain basis for the
2774 * trace files destination directory.
2775 *
2776 * Should *NOT* be called with RCU read-side lock held.
2777 */
2778 static int copy_session_consumer(int domain, struct ltt_session *session)
2779 {
2780 int ret;
2781 const char *dir_name;
2782 struct consumer_output *consumer;
2783
2784 assert(session);
2785 assert(session->consumer);
2786
2787 switch (domain) {
2788 case LTTNG_DOMAIN_KERNEL:
2789 DBG3("Copying tracing session consumer output in kernel session");
2790 /*
2791 * XXX: We should audit the session creation and what this function
2792 * does "extra" in order to avoid a destroy since this function is used
2793 * in the domain session creation (kernel and ust) only. Same for UST
2794 * domain.
2795 */
2796 if (session->kernel_session->consumer) {
2797 consumer_output_put(session->kernel_session->consumer);
2798 }
2799 session->kernel_session->consumer =
2800 consumer_copy_output(session->consumer);
2801 /* Ease our life a bit for the next part */
2802 consumer = session->kernel_session->consumer;
2803 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2804 break;
2805 case LTTNG_DOMAIN_JUL:
2806 case LTTNG_DOMAIN_LOG4J:
2807 case LTTNG_DOMAIN_PYTHON:
2808 case LTTNG_DOMAIN_UST:
2809 DBG3("Copying tracing session consumer output in UST session");
2810 if (session->ust_session->consumer) {
2811 consumer_output_put(session->ust_session->consumer);
2812 }
2813 session->ust_session->consumer =
2814 consumer_copy_output(session->consumer);
2815 /* Ease our life a bit for the next part */
2816 consumer = session->ust_session->consumer;
2817 dir_name = DEFAULT_UST_TRACE_DIR;
2818 break;
2819 default:
2820 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2821 goto error;
2822 }
2823
2824 /* Append correct directory to subdir */
2825 strncat(consumer->subdir, dir_name,
2826 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2827 DBG3("Copy session consumer subdir %s", consumer->subdir);
2828
2829 ret = LTTNG_OK;
2830
2831 error:
2832 return ret;
2833 }
2834
2835 /*
2836 * Create an UST session and add it to the session ust list.
2837 *
2838 * Should *NOT* be called with RCU read-side lock held.
2839 */
2840 static int create_ust_session(struct ltt_session *session,
2841 struct lttng_domain *domain)
2842 {
2843 int ret;
2844 struct ltt_ust_session *lus = NULL;
2845
2846 assert(session);
2847 assert(domain);
2848 assert(session->consumer);
2849
2850 switch (domain->type) {
2851 case LTTNG_DOMAIN_JUL:
2852 case LTTNG_DOMAIN_LOG4J:
2853 case LTTNG_DOMAIN_PYTHON:
2854 case LTTNG_DOMAIN_UST:
2855 break;
2856 default:
2857 ERR("Unknown UST domain on create session %d", domain->type);
2858 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2859 goto error;
2860 }
2861
2862 DBG("Creating UST session");
2863
2864 lus = trace_ust_create_session(session->id);
2865 if (lus == NULL) {
2866 ret = LTTNG_ERR_UST_SESS_FAIL;
2867 goto error;
2868 }
2869
2870 lus->uid = session->uid;
2871 lus->gid = session->gid;
2872 lus->output_traces = session->output_traces;
2873 lus->snapshot_mode = session->snapshot_mode;
2874 lus->live_timer_interval = session->live_timer;
2875 session->ust_session = lus;
2876 if (session->shm_path[0]) {
2877 strncpy(lus->root_shm_path, session->shm_path,
2878 sizeof(lus->root_shm_path));
2879 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2880 strncpy(lus->shm_path, session->shm_path,
2881 sizeof(lus->shm_path));
2882 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2883 strncat(lus->shm_path, "/ust",
2884 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2885 }
2886 /* Copy session output to the newly created UST session */
2887 ret = copy_session_consumer(domain->type, session);
2888 if (ret != LTTNG_OK) {
2889 goto error;
2890 }
2891
2892 return LTTNG_OK;
2893
2894 error:
2895 free(lus);
2896 session->ust_session = NULL;
2897 return ret;
2898 }
2899
2900 /*
2901 * Create a kernel tracer session then create the default channel.
2902 */
2903 static int create_kernel_session(struct ltt_session *session)
2904 {
2905 int ret;
2906
2907 DBG("Creating kernel session");
2908
2909 ret = kernel_create_session(session, kernel_tracer_fd);
2910 if (ret < 0) {
2911 ret = LTTNG_ERR_KERN_SESS_FAIL;
2912 goto error;
2913 }
2914
2915 /* Code flow safety */
2916 assert(session->kernel_session);
2917
2918 /* Copy session output to the newly created Kernel session */
2919 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2920 if (ret != LTTNG_OK) {
2921 goto error;
2922 }
2923
2924 /* Create directory(ies) on local filesystem. */
2925 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2926 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2927 ret = run_as_mkdir_recursive(
2928 session->kernel_session->consumer->dst.trace_path,
2929 S_IRWXU | S_IRWXG, session->uid, session->gid);
2930 if (ret < 0) {
2931 if (errno != EEXIST) {
2932 ERR("Trace directory creation error");
2933 goto error;
2934 }
2935 }
2936 }
2937
2938 session->kernel_session->uid = session->uid;
2939 session->kernel_session->gid = session->gid;
2940 session->kernel_session->output_traces = session->output_traces;
2941 session->kernel_session->snapshot_mode = session->snapshot_mode;
2942
2943 return LTTNG_OK;
2944
2945 error:
2946 trace_kernel_destroy_session(session->kernel_session);
2947 session->kernel_session = NULL;
2948 return ret;
2949 }
2950
2951 /*
2952 * Count number of session permitted by uid/gid.
2953 */
2954 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2955 {
2956 unsigned int i = 0;
2957 struct ltt_session *session;
2958
2959 DBG("Counting number of available session for UID %d GID %d",
2960 uid, gid);
2961 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2962 /*
2963 * Only list the sessions the user can control.
2964 */
2965 if (!session_access_ok(session, uid, gid)) {
2966 continue;
2967 }
2968 i++;
2969 }
2970 return i;
2971 }
2972
2973 /*
2974 * Process the command requested by the lttng client within the command
2975 * context structure. This function make sure that the return structure (llm)
2976 * is set and ready for transmission before returning.
2977 *
2978 * Return any error encountered or 0 for success.
2979 *
2980 * "sock" is only used for special-case var. len data.
2981 *
2982 * Should *NOT* be called with RCU read-side lock held.
2983 */
2984 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2985 int *sock_error)
2986 {
2987 int ret = LTTNG_OK;
2988 int need_tracing_session = 1;
2989 int need_domain;
2990
2991 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2992
2993 assert(!rcu_read_ongoing());
2994
2995 *sock_error = 0;
2996
2997 switch (cmd_ctx->lsm->cmd_type) {
2998 case LTTNG_CREATE_SESSION:
2999 case LTTNG_CREATE_SESSION_SNAPSHOT:
3000 case LTTNG_CREATE_SESSION_LIVE:
3001 case LTTNG_DESTROY_SESSION:
3002 case LTTNG_LIST_SESSIONS:
3003 case LTTNG_LIST_DOMAINS:
3004 case LTTNG_START_TRACE:
3005 case LTTNG_STOP_TRACE:
3006 case LTTNG_DATA_PENDING:
3007 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3008 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3009 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3010 case LTTNG_SNAPSHOT_RECORD:
3011 case LTTNG_SAVE_SESSION:
3012 case LTTNG_SET_SESSION_SHM_PATH:
3013 need_domain = 0;
3014 break;
3015 default:
3016 need_domain = 1;
3017 }
3018
3019 if (opt_no_kernel && need_domain
3020 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3021 if (!is_root) {
3022 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3023 } else {
3024 ret = LTTNG_ERR_KERN_NA;
3025 }
3026 goto error;
3027 }
3028
3029 /* Deny register consumer if we already have a spawned consumer. */
3030 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3031 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3032 if (kconsumer_data.pid > 0) {
3033 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3034 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3035 goto error;
3036 }
3037 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3038 }
3039
3040 /*
3041 * Check for command that don't needs to allocate a returned payload. We do
3042 * this here so we don't have to make the call for no payload at each
3043 * command.
3044 */
3045 switch(cmd_ctx->lsm->cmd_type) {
3046 case LTTNG_LIST_SESSIONS:
3047 case LTTNG_LIST_TRACEPOINTS:
3048 case LTTNG_LIST_TRACEPOINT_FIELDS:
3049 case LTTNG_LIST_DOMAINS:
3050 case LTTNG_LIST_CHANNELS:
3051 case LTTNG_LIST_EVENTS:
3052 case LTTNG_LIST_SYSCALLS:
3053 case LTTNG_LIST_TRACKER_PIDS:
3054 case LTTNG_DATA_PENDING:
3055 break;
3056 default:
3057 /* Setup lttng message with no payload */
3058 ret = setup_lttng_msg(cmd_ctx, 0);
3059 if (ret < 0) {
3060 /* This label does not try to unlock the session */
3061 goto init_setup_error;
3062 }
3063 }
3064
3065 /* Commands that DO NOT need a session. */
3066 switch (cmd_ctx->lsm->cmd_type) {
3067 case LTTNG_CREATE_SESSION:
3068 case LTTNG_CREATE_SESSION_SNAPSHOT:
3069 case LTTNG_CREATE_SESSION_LIVE:
3070 case LTTNG_CALIBRATE:
3071 case LTTNG_LIST_SESSIONS:
3072 case LTTNG_LIST_TRACEPOINTS:
3073 case LTTNG_LIST_SYSCALLS:
3074 case LTTNG_LIST_TRACEPOINT_FIELDS:
3075 case LTTNG_SAVE_SESSION:
3076 need_tracing_session = 0;
3077 break;
3078 default:
3079 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3080 /*
3081 * We keep the session list lock across _all_ commands
3082 * for now, because the per-session lock does not
3083 * handle teardown properly.
3084 */
3085 session_lock_list();
3086 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3087 if (cmd_ctx->session == NULL) {
3088 ret = LTTNG_ERR_SESS_NOT_FOUND;
3089 goto error;
3090 } else {
3091 /* Acquire lock for the session */
3092 session_lock(cmd_ctx->session);
3093 }
3094 break;
3095 }
3096
3097 /*
3098 * Commands that need a valid session but should NOT create one if none
3099 * exists. Instead of creating one and destroying it when the command is
3100 * handled, process that right before so we save some round trip in useless
3101 * code path.
3102 */
3103 switch (cmd_ctx->lsm->cmd_type) {
3104 case LTTNG_DISABLE_CHANNEL:
3105 case LTTNG_DISABLE_EVENT:
3106 switch (cmd_ctx->lsm->domain.type) {
3107 case LTTNG_DOMAIN_KERNEL:
3108 if (!cmd_ctx->session->kernel_session) {
3109 ret = LTTNG_ERR_NO_CHANNEL;
3110 goto error;
3111 }
3112 break;
3113 case LTTNG_DOMAIN_JUL:
3114 case LTTNG_DOMAIN_LOG4J:
3115 case LTTNG_DOMAIN_PYTHON:
3116 case LTTNG_DOMAIN_UST:
3117 if (!cmd_ctx->session->ust_session) {
3118 ret = LTTNG_ERR_NO_CHANNEL;
3119 goto error;
3120 }
3121 break;
3122 default:
3123 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3124 goto error;
3125 }
3126 default:
3127 break;
3128 }
3129
3130 if (!need_domain) {
3131 goto skip_domain;
3132 }
3133
3134 /*
3135 * Check domain type for specific "pre-action".
3136 */
3137 switch (cmd_ctx->lsm->domain.type) {
3138 case LTTNG_DOMAIN_KERNEL:
3139 if (!is_root) {
3140 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3141 goto error;
3142 }
3143
3144 /* Kernel tracer check */
3145 if (kernel_tracer_fd == -1) {
3146 /* Basically, load kernel tracer modules */
3147 ret = init_kernel_tracer();
3148 if (ret != 0) {
3149 goto error;
3150 }
3151 }
3152
3153 /* Consumer is in an ERROR state. Report back to client */
3154 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3155 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3156 goto error;
3157 }
3158
3159 /* Need a session for kernel command */
3160 if (need_tracing_session) {
3161 if (cmd_ctx->session->kernel_session == NULL) {
3162 ret = create_kernel_session(cmd_ctx->session);
3163 if (ret < 0) {
3164 ret = LTTNG_ERR_KERN_SESS_FAIL;
3165 goto error;
3166 }
3167 }
3168
3169 /* Start the kernel consumer daemon */
3170 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3171 if (kconsumer_data.pid == 0 &&
3172 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3173 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3174 ret = start_consumerd(&kconsumer_data);
3175 if (ret < 0) {
3176 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3177 goto error;
3178 }
3179 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3180 } else {
3181 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3182 }
3183
3184 /*
3185 * The consumer was just spawned so we need to add the socket to
3186 * the consumer output of the session if exist.
3187 */
3188 ret = consumer_create_socket(&kconsumer_data,
3189 cmd_ctx->session->kernel_session->consumer);
3190 if (ret < 0) {
3191 goto error;
3192 }
3193 }
3194
3195 break;
3196 case LTTNG_DOMAIN_JUL:
3197 case LTTNG_DOMAIN_LOG4J:
3198 case LTTNG_DOMAIN_PYTHON:
3199 case LTTNG_DOMAIN_UST:
3200 {
3201 if (!ust_app_supported()) {
3202 ret = LTTNG_ERR_NO_UST;
3203 goto error;
3204 }
3205 /* Consumer is in an ERROR state. Report back to client */
3206 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3207 ret = LTTNG_ERR_NO_USTCONSUMERD;
3208 goto error;
3209 }
3210
3211 if (need_tracing_session) {
3212 /* Create UST session if none exist. */
3213 if (cmd_ctx->session->ust_session == NULL) {
3214 ret = create_ust_session(cmd_ctx->session,
3215 &cmd_ctx->lsm->domain);
3216 if (ret != LTTNG_OK) {
3217 goto error;
3218 }
3219 }
3220
3221 /* Start the UST consumer daemons */
3222 /* 64-bit */
3223 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3224 if (consumerd64_bin[0] != '\0' &&
3225 ustconsumer64_data.pid == 0 &&
3226 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3227 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3228 ret = start_consumerd(&ustconsumer64_data);
3229 if (ret < 0) {
3230 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3231 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3232 goto error;
3233 }
3234
3235 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3236 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3237 } else {
3238 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3239 }
3240
3241 /*
3242 * Setup socket for consumer 64 bit. No need for atomic access
3243 * since it was set above and can ONLY be set in this thread.
3244 */
3245 ret = consumer_create_socket(&ustconsumer64_data,
3246 cmd_ctx->session->ust_session->consumer);
3247 if (ret < 0) {
3248 goto error;
3249 }
3250
3251 /* 32-bit */
3252 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3253 if (consumerd32_bin[0] != '\0' &&
3254 ustconsumer32_data.pid == 0 &&
3255 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3256 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3257 ret = start_consumerd(&ustconsumer32_data);
3258 if (ret < 0) {
3259 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3260 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3261 goto error;
3262 }
3263
3264 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3265 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3266 } else {
3267 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3268 }
3269
3270 /*
3271 * Setup socket for consumer 64 bit. No need for atomic access
3272 * since it was set above and can ONLY be set in this thread.
3273 */
3274 ret = consumer_create_socket(&ustconsumer32_data,
3275 cmd_ctx->session->ust_session->consumer);
3276 if (ret < 0) {
3277 goto error;
3278 }
3279 }
3280 break;
3281 }
3282 default:
3283 break;
3284 }
3285 skip_domain:
3286
3287 /* Validate consumer daemon state when start/stop trace command */
3288 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3289 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3290 switch (cmd_ctx->lsm->domain.type) {
3291 case LTTNG_DOMAIN_NONE:
3292 break;
3293 case LTTNG_DOMAIN_JUL:
3294 case LTTNG_DOMAIN_LOG4J:
3295 case LTTNG_DOMAIN_PYTHON:
3296 case LTTNG_DOMAIN_UST:
3297 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3298 ret = LTTNG_ERR_NO_USTCONSUMERD;
3299 goto error;
3300 }
3301 break;
3302 case LTTNG_DOMAIN_KERNEL:
3303 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3304 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3305 goto error;
3306 }
3307 break;
3308 default:
3309 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3310 goto error;
3311 }
3312 }
3313
3314 /*
3315 * Check that the UID or GID match that of the tracing session.
3316 * The root user can interact with all sessions.
3317 */
3318 if (need_tracing_session) {
3319 if (!session_access_ok(cmd_ctx->session,
3320 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3321 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3322 ret = LTTNG_ERR_EPERM;
3323 goto error;
3324 }
3325 }
3326
3327 /*
3328 * Send relayd information to consumer as soon as we have a domain and a
3329 * session defined.
3330 */
3331 if (cmd_ctx->session && need_domain) {
3332 /*
3333 * Setup relayd if not done yet. If the relayd information was already
3334 * sent to the consumer, this call will gracefully return.
3335 */
3336 ret = cmd_setup_relayd(cmd_ctx->session);
3337 if (ret != LTTNG_OK) {
3338 goto error;
3339 }
3340 }
3341
3342 /* Process by command type */
3343 switch (cmd_ctx->lsm->cmd_type) {
3344 case LTTNG_ADD_CONTEXT:
3345 {
3346 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3347 cmd_ctx->lsm->u.context.channel_name,
3348 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3349 break;
3350 }
3351 case LTTNG_DISABLE_CHANNEL:
3352 {
3353 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3354 cmd_ctx->lsm->u.disable.channel_name);
3355 break;
3356 }
3357 case LTTNG_DISABLE_EVENT:
3358 {
3359
3360 /*
3361 * FIXME: handle filter; for now we just receive the filter's
3362 * bytecode along with the filter expression which are sent by
3363 * liblttng-ctl and discard them.
3364 *
3365 * This fixes an issue where the client may block while sending
3366 * the filter payload and encounter an error because the session
3367 * daemon closes the socket without ever handling this data.
3368 */
3369 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3370 cmd_ctx->lsm->u.disable.bytecode_len;
3371
3372 if (count) {
3373 char data[LTTNG_FILTER_MAX_LEN];
3374
3375 DBG("Discarding disable event command payload of size %zu", count);
3376 while (count) {
3377 ret = lttcomm_recv_unix_sock(sock, data,
3378 count > sizeof(data) ? sizeof(data) : count);
3379 if (ret < 0) {
3380 goto error;
3381 }
3382
3383 count -= (size_t) ret;
3384 }
3385 }
3386 /* FIXME: passing packed structure to non-packed pointer */
3387 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3388 cmd_ctx->lsm->u.disable.channel_name,
3389 &cmd_ctx->lsm->u.disable.event);
3390 break;
3391 }
3392 case LTTNG_ENABLE_CHANNEL:
3393 {
3394 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3395 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3396 break;
3397 }
3398 case LTTNG_TRACK_PID:
3399 {
3400 ret = cmd_track_pid(cmd_ctx->session,
3401 cmd_ctx->lsm->domain.type,
3402 cmd_ctx->lsm->u.pid_tracker.pid);
3403 break;
3404 }
3405 case LTTNG_UNTRACK_PID:
3406 {
3407 ret = cmd_untrack_pid(cmd_ctx->session,
3408 cmd_ctx->lsm->domain.type,
3409 cmd_ctx->lsm->u.pid_tracker.pid);
3410 break;
3411 }
3412 case LTTNG_ENABLE_EVENT:
3413 {
3414 struct lttng_event_exclusion *exclusion = NULL;
3415 struct lttng_filter_bytecode *bytecode = NULL;
3416 char *filter_expression = NULL;
3417
3418 /* Handle exclusion events and receive it from the client. */
3419 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3420 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3421
3422 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3423 (count * LTTNG_SYMBOL_NAME_LEN));
3424 if (!exclusion) {
3425 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3426 goto error;
3427 }
3428
3429 DBG("Receiving var len exclusion event list from client ...");
3430 exclusion->count = count;
3431 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3432 count * LTTNG_SYMBOL_NAME_LEN);
3433 if (ret <= 0) {
3434 DBG("Nothing recv() from client var len data... continuing");
3435 *sock_error = 1;
3436 free(exclusion);
3437 ret = LTTNG_ERR_EXCLUSION_INVAL;
3438 goto error;
3439 }
3440 }
3441
3442 /* Get filter expression from client. */
3443 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3444 size_t expression_len =
3445 cmd_ctx->lsm->u.enable.expression_len;
3446
3447 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3448 ret = LTTNG_ERR_FILTER_INVAL;
3449 free(exclusion);
3450 goto error;
3451 }
3452
3453 filter_expression = zmalloc(expression_len);
3454 if (!filter_expression) {
3455 free(exclusion);
3456 ret = LTTNG_ERR_FILTER_NOMEM;
3457 goto error;
3458 }
3459
3460 /* Receive var. len. data */
3461 DBG("Receiving var len filter's expression from client ...");
3462 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3463 expression_len);
3464 if (ret <= 0) {
3465 DBG("Nothing recv() from client car len data... continuing");
3466 *sock_error = 1;
3467 free(filter_expression);
3468 free(exclusion);
3469 ret = LTTNG_ERR_FILTER_INVAL;
3470 goto error;
3471 }
3472 }
3473
3474 /* Handle filter and get bytecode from client. */
3475 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3476 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3477
3478 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3479 ret = LTTNG_ERR_FILTER_INVAL;
3480 free(filter_expression);
3481 free(exclusion);
3482 goto error;
3483 }
3484
3485 bytecode = zmalloc(bytecode_len);
3486 if (!bytecode) {
3487 free(filter_expression);
3488 free(exclusion);
3489 ret = LTTNG_ERR_FILTER_NOMEM;
3490 goto error;
3491 }
3492
3493 /* Receive var. len. data */
3494 DBG("Receiving var len filter's bytecode from client ...");
3495 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3496 if (ret <= 0) {
3497 DBG("Nothing recv() from client car len data... continuing");
3498 *sock_error = 1;
3499 free(filter_expression);
3500 free(bytecode);
3501 free(exclusion);
3502 ret = LTTNG_ERR_FILTER_INVAL;
3503 goto error;
3504 }
3505
3506 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3507 free(filter_expression);
3508 free(bytecode);
3509 free(exclusion);
3510 ret = LTTNG_ERR_FILTER_INVAL;
3511 goto error;
3512 }
3513 }
3514
3515 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3516 cmd_ctx->lsm->u.enable.channel_name,
3517 &cmd_ctx->lsm->u.enable.event,
3518 filter_expression, bytecode, exclusion,
3519 kernel_poll_pipe[1]);
3520 break;
3521 }
3522 case LTTNG_LIST_TRACEPOINTS:
3523 {
3524 struct lttng_event *events;
3525 ssize_t nb_events;
3526
3527 session_lock_list();
3528 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3529 session_unlock_list();
3530 if (nb_events < 0) {
3531 /* Return value is a negative lttng_error_code. */
3532 ret = -nb_events;
3533 goto error;
3534 }
3535
3536 /*
3537 * Setup lttng message with payload size set to the event list size in
3538 * bytes and then copy list into the llm payload.
3539 */
3540 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3541 if (ret < 0) {
3542 free(events);
3543 goto setup_error;
3544 }
3545
3546 /* Copy event list into message payload */
3547 memcpy(cmd_ctx->llm->payload, events,
3548 sizeof(struct lttng_event) * nb_events);
3549
3550 free(events);
3551
3552 ret = LTTNG_OK;
3553 break;
3554 }
3555 case LTTNG_LIST_TRACEPOINT_FIELDS:
3556 {
3557 struct lttng_event_field *fields;
3558 ssize_t nb_fields;
3559
3560 session_lock_list();
3561 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3562 &fields);
3563 session_unlock_list();
3564 if (nb_fields < 0) {
3565 /* Return value is a negative lttng_error_code. */
3566 ret = -nb_fields;
3567 goto error;
3568 }
3569
3570 /*
3571 * Setup lttng message with payload size set to the event list size in
3572 * bytes and then copy list into the llm payload.
3573 */
3574 ret = setup_lttng_msg(cmd_ctx,
3575 sizeof(struct lttng_event_field) * nb_fields);
3576 if (ret < 0) {
3577 free(fields);
3578 goto setup_error;
3579 }
3580
3581 /* Copy event list into message payload */
3582 memcpy(cmd_ctx->llm->payload, fields,
3583 sizeof(struct lttng_event_field) * nb_fields);
3584
3585 free(fields);
3586
3587 ret = LTTNG_OK;
3588 break;
3589 }
3590 case LTTNG_LIST_SYSCALLS:
3591 {
3592 struct lttng_event *events;
3593 ssize_t nb_events;
3594
3595 nb_events = cmd_list_syscalls(&events);
3596 if (nb_events < 0) {
3597 /* Return value is a negative lttng_error_code. */
3598 ret = -nb_events;
3599 goto error;
3600 }
3601
3602 /*
3603 * Setup lttng message with payload size set to the event list size in
3604 * bytes and then copy list into the llm payload.
3605 */
3606 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3607 if (ret < 0) {
3608 free(events);
3609 goto setup_error;
3610 }
3611
3612 /* Copy event list into message payload */
3613 memcpy(cmd_ctx->llm->payload, events,
3614 sizeof(struct lttng_event) * nb_events);
3615
3616 free(events);
3617
3618 ret = LTTNG_OK;
3619 break;
3620 }
3621 case LTTNG_LIST_TRACKER_PIDS:
3622 {
3623 int32_t *pids = NULL;
3624 ssize_t nr_pids;
3625
3626 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3627 cmd_ctx->lsm->domain.type, &pids);
3628 if (nr_pids < 0) {
3629 /* Return value is a negative lttng_error_code. */
3630 ret = -nr_pids;
3631 goto error;
3632 }
3633
3634 /*
3635 * Setup lttng message with payload size set to the event list size in
3636 * bytes and then copy list into the llm payload.
3637 */
3638 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3639 if (ret < 0) {
3640 free(pids);
3641 goto setup_error;
3642 }
3643
3644 /* Copy event list into message payload */
3645 memcpy(cmd_ctx->llm->payload, pids,
3646 sizeof(int) * nr_pids);
3647
3648 free(pids);
3649
3650 ret = LTTNG_OK;
3651 break;
3652 }
3653 case LTTNG_SET_CONSUMER_URI:
3654 {
3655 size_t nb_uri, len;
3656 struct lttng_uri *uris;
3657
3658 nb_uri = cmd_ctx->lsm->u.uri.size;
3659 len = nb_uri * sizeof(struct lttng_uri);
3660
3661 if (nb_uri == 0) {
3662 ret = LTTNG_ERR_INVALID;
3663 goto error;
3664 }
3665
3666 uris = zmalloc(len);
3667 if (uris == NULL) {
3668 ret = LTTNG_ERR_FATAL;
3669 goto error;
3670 }
3671
3672 /* Receive variable len data */
3673 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3674 ret = lttcomm_recv_unix_sock(sock, uris, len);
3675 if (ret <= 0) {
3676 DBG("No URIs received from client... continuing");
3677 *sock_error = 1;
3678 ret = LTTNG_ERR_SESSION_FAIL;
3679 free(uris);
3680 goto error;
3681 }
3682
3683 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3684 free(uris);
3685 if (ret != LTTNG_OK) {
3686 goto error;
3687 }
3688
3689
3690 break;
3691 }
3692 case LTTNG_START_TRACE:
3693 {
3694 ret = cmd_start_trace(cmd_ctx->session);
3695 break;
3696 }
3697 case LTTNG_STOP_TRACE:
3698 {
3699 ret = cmd_stop_trace(cmd_ctx->session);
3700 break;
3701 }
3702 case LTTNG_CREATE_SESSION:
3703 {
3704 size_t nb_uri, len;
3705 struct lttng_uri *uris = NULL;
3706
3707 nb_uri = cmd_ctx->lsm->u.uri.size;
3708 len = nb_uri * sizeof(struct lttng_uri);
3709
3710 if (nb_uri > 0) {
3711 uris = zmalloc(len);
3712 if (uris == NULL) {
3713 ret = LTTNG_ERR_FATAL;
3714 goto error;
3715 }
3716
3717 /* Receive variable len data */
3718 DBG("Waiting for %zu URIs from client ...", nb_uri);
3719 ret = lttcomm_recv_unix_sock(sock, uris, len);
3720 if (ret <= 0) {
3721 DBG("No URIs received from client... continuing");
3722 *sock_error = 1;
3723 ret = LTTNG_ERR_SESSION_FAIL;
3724 free(uris);
3725 goto error;
3726 }
3727
3728 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3729 DBG("Creating session with ONE network URI is a bad call");
3730 ret = LTTNG_ERR_SESSION_FAIL;
3731 free(uris);
3732 goto error;
3733 }
3734 }
3735
3736 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3737 &cmd_ctx->creds, 0);
3738
3739 free(uris);
3740
3741 break;
3742 }
3743 case LTTNG_DESTROY_SESSION:
3744 {
3745 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3746
3747 /* Set session to NULL so we do not unlock it after free. */
3748 cmd_ctx->session = NULL;
3749 break;
3750 }
3751 case LTTNG_LIST_DOMAINS:
3752 {
3753 ssize_t nb_dom;
3754 struct lttng_domain *domains = NULL;
3755
3756 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3757 if (nb_dom < 0) {
3758 /* Return value is a negative lttng_error_code. */
3759 ret = -nb_dom;
3760 goto error;
3761 }
3762
3763 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3764 if (ret < 0) {
3765 free(domains);
3766 goto setup_error;
3767 }
3768
3769 /* Copy event list into message payload */
3770 memcpy(cmd_ctx->llm->payload, domains,
3771 nb_dom * sizeof(struct lttng_domain));
3772
3773 free(domains);
3774
3775 ret = LTTNG_OK;
3776 break;
3777 }
3778 case LTTNG_LIST_CHANNELS:
3779 {
3780 int nb_chan;
3781 struct lttng_channel *channels = NULL;
3782
3783 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3784 cmd_ctx->session, &channels);
3785 if (nb_chan < 0) {
3786 /* Return value is a negative lttng_error_code. */
3787 ret = -nb_chan;
3788 goto error;
3789 }
3790
3791 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3792 if (ret < 0) {
3793 free(channels);
3794 goto setup_error;
3795 }
3796
3797 /* Copy event list into message payload */
3798 memcpy(cmd_ctx->llm->payload, channels,
3799 nb_chan * sizeof(struct lttng_channel));
3800
3801 free(channels);
3802
3803 ret = LTTNG_OK;
3804 break;
3805 }
3806 case LTTNG_LIST_EVENTS:
3807 {
3808 ssize_t nb_event;
3809 struct lttng_event *events = NULL;
3810
3811 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3812 cmd_ctx->lsm->u.list.channel_name, &events);
3813 if (nb_event < 0) {
3814 /* Return value is a negative lttng_error_code. */
3815 ret = -nb_event;
3816 goto error;
3817 }
3818
3819 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3820 if (ret < 0) {
3821 free(events);
3822 goto setup_error;
3823 }
3824
3825 /* Copy event list into message payload */
3826 memcpy(cmd_ctx->llm->payload, events,
3827 nb_event * sizeof(struct lttng_event));
3828
3829 free(events);
3830
3831 ret = LTTNG_OK;
3832 break;
3833 }
3834 case LTTNG_LIST_SESSIONS:
3835 {
3836 unsigned int nr_sessions;
3837
3838 session_lock_list();
3839 nr_sessions = lttng_sessions_count(
3840 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3841 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3842
3843 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3844 if (ret < 0) {
3845 session_unlock_list();
3846 goto setup_error;
3847 }
3848
3849 /* Filled the session array */
3850 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3851 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3852 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3853
3854 session_unlock_list();
3855
3856 ret = LTTNG_OK;
3857 break;
3858 }
3859 case LTTNG_CALIBRATE:
3860 {
3861 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3862 &cmd_ctx->lsm->u.calibrate);
3863 break;
3864 }
3865 case LTTNG_REGISTER_CONSUMER:
3866 {
3867 struct consumer_data *cdata;
3868
3869 switch (cmd_ctx->lsm->domain.type) {
3870 case LTTNG_DOMAIN_KERNEL:
3871 cdata = &kconsumer_data;
3872 break;
3873 default:
3874 ret = LTTNG_ERR_UND;
3875 goto error;
3876 }
3877
3878 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3879 cmd_ctx->lsm->u.reg.path, cdata);
3880 break;
3881 }
3882 case LTTNG_DATA_PENDING:
3883 {
3884 ret = cmd_data_pending(cmd_ctx->session);
3885 break;
3886 }
3887 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3888 {
3889 struct lttcomm_lttng_output_id reply;
3890
3891 ret = cmd_snapshot_add_output(cmd_ctx->session,
3892 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3893 if (ret != LTTNG_OK) {
3894 goto error;
3895 }
3896
3897 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3898 if (ret < 0) {
3899 goto setup_error;
3900 }
3901
3902 /* Copy output list into message payload */
3903 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3904 ret = LTTNG_OK;
3905 break;
3906 }
3907 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3908 {
3909 ret = cmd_snapshot_del_output(cmd_ctx->session,
3910 &cmd_ctx->lsm->u.snapshot_output.output);
3911 break;
3912 }
3913 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3914 {
3915 ssize_t nb_output;
3916 struct lttng_snapshot_output *outputs = NULL;
3917
3918 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3919 if (nb_output < 0) {
3920 ret = -nb_output;
3921 goto error;
3922 }
3923
3924 ret = setup_lttng_msg(cmd_ctx,
3925 nb_output * sizeof(struct lttng_snapshot_output));
3926 if (ret < 0) {
3927 free(outputs);
3928 goto setup_error;
3929 }
3930
3931 if (outputs) {
3932 /* Copy output list into message payload */
3933 memcpy(cmd_ctx->llm->payload, outputs,
3934 nb_output * sizeof(struct lttng_snapshot_output));
3935 free(outputs);
3936 }
3937
3938 ret = LTTNG_OK;
3939 break;
3940 }
3941 case LTTNG_SNAPSHOT_RECORD:
3942 {
3943 ret = cmd_snapshot_record(cmd_ctx->session,
3944 &cmd_ctx->lsm->u.snapshot_record.output,
3945 cmd_ctx->lsm->u.snapshot_record.wait);
3946 break;
3947 }
3948 case LTTNG_CREATE_SESSION_SNAPSHOT:
3949 {
3950 size_t nb_uri, len;
3951 struct lttng_uri *uris = NULL;
3952
3953 nb_uri = cmd_ctx->lsm->u.uri.size;
3954 len = nb_uri * sizeof(struct lttng_uri);
3955
3956 if (nb_uri > 0) {
3957 uris = zmalloc(len);
3958 if (uris == NULL) {
3959 ret = LTTNG_ERR_FATAL;
3960 goto error;
3961 }
3962
3963 /* Receive variable len data */
3964 DBG("Waiting for %zu URIs from client ...", nb_uri);
3965 ret = lttcomm_recv_unix_sock(sock, uris, len);
3966 if (ret <= 0) {
3967 DBG("No URIs received from client... continuing");
3968 *sock_error = 1;
3969 ret = LTTNG_ERR_SESSION_FAIL;
3970 free(uris);
3971 goto error;
3972 }
3973
3974 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3975 DBG("Creating session with ONE network URI is a bad call");
3976 ret = LTTNG_ERR_SESSION_FAIL;
3977 free(uris);
3978 goto error;
3979 }
3980 }
3981
3982 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3983 nb_uri, &cmd_ctx->creds);
3984 free(uris);
3985 break;
3986 }
3987 case LTTNG_CREATE_SESSION_LIVE:
3988 {
3989 size_t nb_uri, len;
3990 struct lttng_uri *uris = NULL;
3991
3992 nb_uri = cmd_ctx->lsm->u.uri.size;
3993 len = nb_uri * sizeof(struct lttng_uri);
3994
3995 if (nb_uri > 0) {
3996 uris = zmalloc(len);
3997 if (uris == NULL) {
3998 ret = LTTNG_ERR_FATAL;
3999 goto error;
4000 }
4001
4002 /* Receive variable len data */
4003 DBG("Waiting for %zu URIs from client ...", nb_uri);
4004 ret = lttcomm_recv_unix_sock(sock, uris, len);
4005 if (ret <= 0) {
4006 DBG("No URIs received from client... continuing");
4007 *sock_error = 1;
4008 ret = LTTNG_ERR_SESSION_FAIL;
4009 free(uris);
4010 goto error;
4011 }
4012
4013 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4014 DBG("Creating session with ONE network URI is a bad call");
4015 ret = LTTNG_ERR_SESSION_FAIL;
4016 free(uris);
4017 goto error;
4018 }
4019 }
4020
4021 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4022 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4023 free(uris);
4024 break;
4025 }
4026 case LTTNG_SAVE_SESSION:
4027 {
4028 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4029 &cmd_ctx->creds);
4030 break;
4031 }
4032 case LTTNG_SET_SESSION_SHM_PATH:
4033 {
4034 ret = cmd_set_session_shm_path(cmd_ctx->session,
4035 cmd_ctx->lsm->u.set_shm_path.shm_path);
4036 break;
4037 }
4038 default:
4039 ret = LTTNG_ERR_UND;
4040 break;
4041 }
4042
4043 error:
4044 if (cmd_ctx->llm == NULL) {
4045 DBG("Missing llm structure. Allocating one.");
4046 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
4047 goto setup_error;
4048 }
4049 }
4050 /* Set return code */
4051 cmd_ctx->llm->ret_code = ret;
4052 setup_error:
4053 if (cmd_ctx->session) {
4054 session_unlock(cmd_ctx->session);
4055 }
4056 if (need_tracing_session) {
4057 session_unlock_list();
4058 }
4059 init_setup_error:
4060 assert(!rcu_read_ongoing());
4061 return ret;
4062 }
4063
4064 /*
4065 * Thread managing health check socket.
4066 */
4067 static void *thread_manage_health(void *data)
4068 {
4069 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4070 uint32_t revents, nb_fd;
4071 struct lttng_poll_event events;
4072 struct health_comm_msg msg;
4073 struct health_comm_reply reply;
4074
4075 DBG("[thread] Manage health check started");
4076
4077 rcu_register_thread();
4078
4079 /* We might hit an error path before this is created. */
4080 lttng_poll_init(&events);
4081
4082 /* Create unix socket */
4083 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4084 if (sock < 0) {
4085 ERR("Unable to create health check Unix socket");
4086 ret = -1;
4087 goto error;
4088 }
4089
4090 if (is_root) {
4091 /* lttng health client socket path permissions */
4092 ret = chown(health_unix_sock_path, 0,
4093 utils_get_group_id(tracing_group_name));
4094 if (ret < 0) {
4095 ERR("Unable to set group on %s", health_unix_sock_path);
4096 PERROR("chown");
4097 ret = -1;
4098 goto error;
4099 }
4100
4101 ret = chmod(health_unix_sock_path,
4102 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4103 if (ret < 0) {
4104 ERR("Unable to set permissions on %s", health_unix_sock_path);
4105 PERROR("chmod");
4106 ret = -1;
4107 goto error;
4108 }
4109 }
4110
4111 /*
4112 * Set the CLOEXEC flag. Return code is useless because either way, the
4113 * show must go on.
4114 */
4115 (void) utils_set_fd_cloexec(sock);
4116
4117 ret = lttcomm_listen_unix_sock(sock);
4118 if (ret < 0) {
4119 goto error;
4120 }
4121
4122 /*
4123 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4124 * more will be added to this poll set.
4125 */
4126 ret = sessiond_set_thread_pollset(&events, 2);
4127 if (ret < 0) {
4128 goto error;
4129 }
4130
4131 /* Add the application registration socket */
4132 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4133 if (ret < 0) {
4134 goto error;
4135 }
4136
4137 sessiond_notify_ready();
4138
4139 while (1) {
4140 DBG("Health check ready");
4141
4142 /* Inifinite blocking call, waiting for transmission */
4143 restart:
4144 ret = lttng_poll_wait(&events, -1);
4145 if (ret < 0) {
4146 /*
4147 * Restart interrupted system call.
4148 */
4149 if (errno == EINTR) {
4150 goto restart;
4151 }
4152 goto error;
4153 }
4154
4155 nb_fd = ret;
4156
4157 for (i = 0; i < nb_fd; i++) {
4158 /* Fetch once the poll data */
4159 revents = LTTNG_POLL_GETEV(&events, i);
4160 pollfd = LTTNG_POLL_GETFD(&events, i);
4161
4162 if (!revents) {
4163 /* No activity for this FD (poll implementation). */
4164 continue;
4165 }
4166
4167 /* Thread quit pipe has been closed. Killing thread. */
4168 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4169 if (ret) {
4170 err = 0;
4171 goto exit;
4172 }
4173
4174 /* Event on the registration socket */
4175 if (pollfd == sock) {
4176 if (revents & LPOLLIN) {
4177 continue;
4178 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4179 ERR("Health socket poll error");
4180 goto error;
4181 } else {
4182 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4183 goto error;
4184 }
4185 }
4186 }
4187
4188 new_sock = lttcomm_accept_unix_sock(sock);
4189 if (new_sock < 0) {
4190 goto error;
4191 }
4192
4193 /*
4194 * Set the CLOEXEC flag. Return code is useless because either way, the
4195 * show must go on.
4196 */
4197 (void) utils_set_fd_cloexec(new_sock);
4198
4199 DBG("Receiving data from client for health...");
4200 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4201 if (ret <= 0) {
4202 DBG("Nothing recv() from client... continuing");
4203 ret = close(new_sock);
4204 if (ret) {
4205 PERROR("close");
4206 }
4207 new_sock = -1;
4208 continue;
4209 }
4210
4211 rcu_thread_online();
4212
4213 memset(&reply, 0, sizeof(reply));
4214 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4215 /*
4216 * health_check_state returns 0 if health is
4217 * bad.
4218 */
4219 if (!health_check_state(health_sessiond, i)) {
4220 reply.ret_code |= 1ULL << i;
4221 }
4222 }
4223
4224 DBG2("Health check return value %" PRIx64, reply.ret_code);
4225
4226 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4227 if (ret < 0) {
4228 ERR("Failed to send health data back to client");
4229 }
4230
4231 /* End of transmission */
4232 ret = close(new_sock);
4233 if (ret) {
4234 PERROR("close");
4235 }
4236 new_sock = -1;
4237 }
4238
4239 exit:
4240 error:
4241 if (err) {
4242 ERR("Health error occurred in %s", __func__);
4243 }
4244 DBG("Health check thread dying");
4245 unlink(health_unix_sock_path);
4246 if (sock >= 0) {
4247 ret = close(sock);
4248 if (ret) {
4249 PERROR("close");
4250 }
4251 }
4252
4253 lttng_poll_clean(&events);
4254
4255 rcu_unregister_thread();
4256 return NULL;
4257 }
4258
4259 /*
4260 * This thread manage all clients request using the unix client socket for
4261 * communication.
4262 */
4263 static void *thread_manage_clients(void *data)
4264 {
4265 int sock = -1, ret, i, pollfd, err = -1;
4266 int sock_error;
4267 uint32_t revents, nb_fd;
4268 struct command_ctx *cmd_ctx = NULL;
4269 struct lttng_poll_event events;
4270
4271 DBG("[thread] Manage client started");
4272
4273 rcu_register_thread();
4274
4275 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4276
4277 health_code_update();
4278
4279 ret = lttcomm_listen_unix_sock(client_sock);
4280 if (ret < 0) {
4281 goto error_listen;
4282 }
4283
4284 /*
4285 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4286 * more will be added to this poll set.
4287 */
4288 ret = sessiond_set_thread_pollset(&events, 2);
4289 if (ret < 0) {
4290 goto error_create_poll;
4291 }
4292
4293 /* Add the application registration socket */
4294 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4295 if (ret < 0) {
4296 goto error;
4297 }
4298
4299 sessiond_notify_ready();
4300 ret = sem_post(&load_info->message_thread_ready);
4301 if (ret) {
4302 PERROR("sem_post message_thread_ready");
4303 goto error;
4304 }
4305
4306 /* This testpoint is after we signal readiness to the parent. */
4307 if (testpoint(sessiond_thread_manage_clients)) {
4308 goto error;
4309 }
4310
4311 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4312 goto error;
4313 }
4314
4315 health_code_update();
4316
4317 while (1) {
4318 DBG("Accepting client command ...");
4319
4320 /* Inifinite blocking call, waiting for transmission */
4321 restart:
4322 health_poll_entry();
4323 ret = lttng_poll_wait(&events, -1);
4324 health_poll_exit();
4325 if (ret < 0) {
4326 /*
4327 * Restart interrupted system call.
4328 */
4329 if (errno == EINTR) {
4330 goto restart;
4331 }
4332 goto error;
4333 }
4334
4335 nb_fd = ret;
4336
4337 for (i = 0; i < nb_fd; i++) {
4338 /* Fetch once the poll data */
4339 revents = LTTNG_POLL_GETEV(&events, i);
4340 pollfd = LTTNG_POLL_GETFD(&events, i);
4341
4342 health_code_update();
4343
4344 if (!revents) {
4345 /* No activity for this FD (poll implementation). */
4346 continue;
4347 }
4348
4349 /* Thread quit pipe has been closed. Killing thread. */
4350 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4351 if (ret) {
4352 err = 0;
4353 goto exit;
4354 }
4355
4356 /* Event on the registration socket */
4357 if (pollfd == client_sock) {
4358 if (revents & LPOLLIN) {
4359 continue;
4360 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4361 ERR("Client socket poll error");
4362 goto error;
4363 } else {
4364 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4365 goto error;
4366 }
4367 }
4368 }
4369
4370 DBG("Wait for client response");
4371
4372 health_code_update();
4373
4374 sock = lttcomm_accept_unix_sock(client_sock);
4375 if (sock < 0) {
4376 goto error;
4377 }
4378
4379 /*
4380 * Set the CLOEXEC flag. Return code is useless because either way, the
4381 * show must go on.
4382 */
4383 (void) utils_set_fd_cloexec(sock);
4384
4385 /* Set socket option for credentials retrieval */
4386 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4387 if (ret < 0) {
4388 goto error;
4389 }
4390
4391 /* Allocate context command to process the client request */
4392 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4393 if (cmd_ctx == NULL) {
4394 PERROR("zmalloc cmd_ctx");
4395 goto error;
4396 }
4397
4398 /* Allocate data buffer for reception */
4399 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4400 if (cmd_ctx->lsm == NULL) {
4401 PERROR("zmalloc cmd_ctx->lsm");
4402 goto error;
4403 }
4404
4405 cmd_ctx->llm = NULL;
4406 cmd_ctx->session = NULL;
4407
4408 health_code_update();
4409
4410 /*
4411 * Data is received from the lttng client. The struct
4412 * lttcomm_session_msg (lsm) contains the command and data request of
4413 * the client.
4414 */
4415 DBG("Receiving data from client ...");
4416 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4417 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4418 if (ret <= 0) {
4419 DBG("Nothing recv() from client... continuing");
4420 ret = close(sock);
4421 if (ret) {
4422 PERROR("close");
4423 }
4424 sock = -1;
4425 clean_command_ctx(&cmd_ctx);
4426 continue;
4427 }
4428
4429 health_code_update();
4430
4431 // TODO: Validate cmd_ctx including sanity check for
4432 // security purpose.
4433
4434 rcu_thread_online();
4435 /*
4436 * This function dispatch the work to the kernel or userspace tracer
4437 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4438 * informations for the client. The command context struct contains
4439 * everything this function may needs.
4440 */
4441 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4442 rcu_thread_offline();
4443 if (ret < 0) {
4444 ret = close(sock);
4445 if (ret) {
4446 PERROR("close");
4447 }
4448 sock = -1;
4449 /*
4450 * TODO: Inform client somehow of the fatal error. At
4451 * this point, ret < 0 means that a zmalloc failed
4452 * (ENOMEM). Error detected but still accept
4453 * command, unless a socket error has been
4454 * detected.
4455 */
4456 clean_command_ctx(&cmd_ctx);
4457 continue;
4458 }
4459
4460 health_code_update();
4461
4462 DBG("Sending response (size: %d, retcode: %s)",
4463 cmd_ctx->lttng_msg_size,
4464 lttng_strerror(-cmd_ctx->llm->ret_code));
4465 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4466 if (ret < 0) {
4467 ERR("Failed to send data back to client");
4468 }
4469
4470 /* End of transmission */
4471 ret = close(sock);
4472 if (ret) {
4473 PERROR("close");
4474 }
4475 sock = -1;
4476
4477 clean_command_ctx(&cmd_ctx);
4478
4479 health_code_update();
4480 }
4481
4482 exit:
4483 error:
4484 if (sock >= 0) {
4485 ret = close(sock);
4486 if (ret) {
4487 PERROR("close");
4488 }
4489 }
4490
4491 lttng_poll_clean(&events);
4492 clean_command_ctx(&cmd_ctx);
4493
4494 error_listen:
4495 error_create_poll:
4496 unlink(client_unix_sock_path);
4497 if (client_sock >= 0) {
4498 ret = close(client_sock);
4499 if (ret) {
4500 PERROR("close");
4501 }
4502 }
4503
4504 if (err) {
4505 health_error();
4506 ERR("Health error occurred in %s", __func__);
4507 }
4508
4509 health_unregister(health_sessiond);
4510
4511 DBG("Client thread dying");
4512
4513 rcu_unregister_thread();
4514
4515 /*
4516 * Since we are creating the consumer threads, we own them, so we need
4517 * to join them before our thread exits.
4518 */
4519 ret = join_consumer_thread(&kconsumer_data);
4520 if (ret) {
4521 errno = ret;
4522 PERROR("join_consumer");
4523 }
4524
4525 ret = join_consumer_thread(&ustconsumer32_data);
4526 if (ret) {
4527 errno = ret;
4528 PERROR("join_consumer ust32");
4529 }
4530
4531 ret = join_consumer_thread(&ustconsumer64_data);
4532 if (ret) {
4533 errno = ret;
4534 PERROR("join_consumer ust64");
4535 }
4536 return NULL;
4537 }
4538
4539
4540 /*
4541 * usage function on stderr
4542 */
4543 static void usage(void)
4544 {
4545 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4546 fprintf(stderr, " -h, --help Display this usage.\n");
4547 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4548 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4549 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4550 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4551 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4552 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4553 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4554 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4555 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4556 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4557 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4558 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4559 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4560 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4561 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4562 fprintf(stderr, " -V, --version Show version number.\n");
4563 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4564 fprintf(stderr, " -q, --quiet No output at all.\n");
4565 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4566 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4567 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4568 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4569 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4570 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4571 fprintf(stderr, " -l --load PATH Load session configuration\n");
4572 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4573 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4574 }
4575
4576 static int string_match(const char *str1, const char *str2)
4577 {
4578 return (str1 && str2) && !strcmp(str1, str2);
4579 }
4580
4581 /*
4582 * Take an option from the getopt output and set it in the right variable to be
4583 * used later.
4584 *
4585 * Return 0 on success else a negative value.
4586 */
4587 static int set_option(int opt, const char *arg, const char *optname)
4588 {
4589 int ret = 0;
4590
4591 if (string_match(optname, "client-sock") || opt == 'c') {
4592 if (!arg || *arg == '\0') {
4593 ret = -EINVAL;
4594 goto end;
4595 }
4596 if (lttng_is_setuid_setgid()) {
4597 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4598 "-c, --client-sock");
4599 } else {
4600 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4601 }
4602 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4603 if (!arg || *arg == '\0') {
4604 ret = -EINVAL;
4605 goto end;
4606 }
4607 if (lttng_is_setuid_setgid()) {
4608 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4609 "-a, --apps-sock");
4610 } else {
4611 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4612 }
4613 } else if (string_match(optname, "daemonize") || opt == 'd') {
4614 opt_daemon = 1;
4615 } else if (string_match(optname, "background") || opt == 'b') {
4616 opt_background = 1;
4617 } else if (string_match(optname, "group") || opt == 'g') {
4618 if (!arg || *arg == '\0') {
4619 ret = -EINVAL;
4620 goto end;
4621 }
4622 if (lttng_is_setuid_setgid()) {
4623 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4624 "-g, --group");
4625 } else {
4626 /*
4627 * If the override option is set, the pointer points to a
4628 * *non* const thus freeing it even though the variable type is
4629 * set to const.
4630 */
4631 if (tracing_group_name_override) {
4632 free((void *) tracing_group_name);
4633 }
4634 tracing_group_name = strdup(arg);
4635 if (!tracing_group_name) {
4636 PERROR("strdup");
4637 ret = -ENOMEM;
4638 }
4639 tracing_group_name_override = 1;
4640 }
4641 } else if (string_match(optname, "help") || opt == 'h') {
4642 usage();
4643 exit(EXIT_SUCCESS);
4644 } else if (string_match(optname, "version") || opt == 'V') {
4645 fprintf(stdout, "%s\n", VERSION);
4646 exit(EXIT_SUCCESS);
4647 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4648 opt_sig_parent = 1;
4649 } else if (string_match(optname, "kconsumerd-err-sock")) {
4650 if (!arg || *arg == '\0') {
4651 ret = -EINVAL;
4652 goto end;
4653 }
4654 if (lttng_is_setuid_setgid()) {
4655 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4656 "--kconsumerd-err-sock");
4657 } else {
4658 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4659 }
4660 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4661 if (!arg || *arg == '\0') {
4662 ret = -EINVAL;
4663 goto end;
4664 }
4665 if (lttng_is_setuid_setgid()) {
4666 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4667 "--kconsumerd-cmd-sock");
4668 } else {
4669 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4670 }
4671 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4672 if (!arg || *arg == '\0') {
4673 ret = -EINVAL;
4674 goto end;
4675 }
4676 if (lttng_is_setuid_setgid()) {
4677 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4678 "--ustconsumerd64-err-sock");
4679 } else {
4680 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4681 }
4682 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4683 if (!arg || *arg == '\0') {
4684 ret = -EINVAL;
4685 goto end;
4686 }
4687 if (lttng_is_setuid_setgid()) {
4688 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4689 "--ustconsumerd64-cmd-sock");
4690 } else {
4691 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4692 }
4693 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4694 if (!arg || *arg == '\0') {
4695 ret = -EINVAL;
4696 goto end;
4697 }
4698 if (lttng_is_setuid_setgid()) {
4699 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4700 "--ustconsumerd32-err-sock");
4701 } else {
4702 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4703 }
4704 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4705 if (!arg || *arg == '\0') {
4706 ret = -EINVAL;
4707 goto end;
4708 }
4709 if (lttng_is_setuid_setgid()) {
4710 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4711 "--ustconsumerd32-cmd-sock");
4712 } else {
4713 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4714 }
4715 } else if (string_match(optname, "no-kernel")) {
4716 opt_no_kernel = 1;
4717 } else if (string_match(optname, "quiet") || opt == 'q') {
4718 lttng_opt_quiet = 1;
4719 } else if (string_match(optname, "verbose") || opt == 'v') {
4720 /* Verbose level can increase using multiple -v */
4721 if (arg) {
4722 /* Value obtained from config file */
4723 lttng_opt_verbose = config_parse_value(arg);
4724 } else {
4725 /* -v used on command line */
4726 lttng_opt_verbose++;
4727 }
4728 /* Clamp value to [0, 3] */
4729 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4730 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4731 } else if (string_match(optname, "verbose-consumer")) {
4732 if (arg) {
4733 opt_verbose_consumer = config_parse_value(arg);
4734 } else {
4735 opt_verbose_consumer += 1;
4736 }
4737 } else if (string_match(optname, "consumerd32-path")) {
4738 if (!arg || *arg == '\0') {
4739 ret = -EINVAL;
4740 goto end;
4741 }
4742 if (lttng_is_setuid_setgid()) {
4743 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4744 "--consumerd32-path");
4745 } else {
4746 if (consumerd32_bin_override) {
4747 free((void *) consumerd32_bin);
4748 }
4749 consumerd32_bin = strdup(arg);
4750 if (!consumerd32_bin) {
4751 PERROR("strdup");
4752 ret = -ENOMEM;
4753 }
4754 consumerd32_bin_override = 1;
4755 }
4756 } else if (string_match(optname, "consumerd32-libdir")) {
4757 if (!arg || *arg == '\0') {
4758 ret = -EINVAL;
4759 goto end;
4760 }
4761 if (lttng_is_setuid_setgid()) {
4762 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4763 "--consumerd32-libdir");
4764 } else {
4765 if (consumerd32_libdir_override) {
4766 free((void *) consumerd32_libdir);
4767 }
4768 consumerd32_libdir = strdup(arg);
4769 if (!consumerd32_libdir) {
4770 PERROR("strdup");
4771 ret = -ENOMEM;
4772 }
4773 consumerd32_libdir_override = 1;
4774 }
4775 } else if (string_match(optname, "consumerd64-path")) {
4776 if (!arg || *arg == '\0') {
4777 ret = -EINVAL;
4778 goto end;
4779 }
4780 if (lttng_is_setuid_setgid()) {
4781 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4782 "--consumerd64-path");
4783 } else {
4784 if (consumerd64_bin_override) {
4785 free((void *) consumerd64_bin);
4786 }
4787 consumerd64_bin = strdup(arg);
4788 if (!consumerd64_bin) {
4789 PERROR("strdup");
4790 ret = -ENOMEM;
4791 }
4792 consumerd64_bin_override = 1;
4793 }
4794 } else if (string_match(optname, "consumerd64-libdir")) {
4795 if (!arg || *arg == '\0') {
4796 ret = -EINVAL;
4797 goto end;
4798 }
4799 if (lttng_is_setuid_setgid()) {
4800 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4801 "--consumerd64-libdir");
4802 } else {
4803 if (consumerd64_libdir_override) {
4804 free((void *) consumerd64_libdir);
4805 }
4806 consumerd64_libdir = strdup(arg);
4807 if (!consumerd64_libdir) {
4808 PERROR("strdup");
4809 ret = -ENOMEM;
4810 }
4811 consumerd64_libdir_override = 1;
4812 }
4813 } else if (string_match(optname, "pidfile") || opt == 'p') {
4814 if (!arg || *arg == '\0') {
4815 ret = -EINVAL;
4816 goto end;
4817 }
4818 if (lttng_is_setuid_setgid()) {
4819 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4820 "-p, --pidfile");
4821 } else {
4822 free(opt_pidfile);
4823 opt_pidfile = strdup(arg);
4824 if (!opt_pidfile) {
4825 PERROR("strdup");
4826 ret = -ENOMEM;
4827 }
4828 }
4829 } else if (string_match(optname, "agent-tcp-port")) {
4830 if (!arg || *arg == '\0') {
4831 ret = -EINVAL;
4832 goto end;
4833 }
4834 if (lttng_is_setuid_setgid()) {
4835 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4836 "--agent-tcp-port");
4837 } else {
4838 unsigned long v;
4839
4840 if (!arg) {
4841 ret = -EINVAL;
4842 goto end;
4843 }
4844 errno = 0;
4845 v = strtoul(arg, NULL, 0);
4846 if (errno != 0 || !isdigit(arg[0])) {
4847 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4848 return -1;
4849 }
4850 if (v == 0 || v >= 65535) {
4851 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4852 return -1;
4853 }
4854 agent_tcp_port = (uint32_t) v;
4855 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4856 }
4857 } else if (string_match(optname, "load") || opt == 'l') {
4858 if (!arg || *arg == '\0') {
4859 ret = -EINVAL;
4860 goto end;
4861 }
4862 if (lttng_is_setuid_setgid()) {
4863 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4864 "-l, --load");
4865 } else {
4866 free(opt_load_session_path);
4867 opt_load_session_path = strdup(arg);
4868 if (!opt_load_session_path) {
4869 PERROR("strdup");
4870 ret = -ENOMEM;
4871 }
4872 }
4873 } else if (string_match(optname, "kmod-probes")) {
4874 if (!arg || *arg == '\0') {
4875 ret = -EINVAL;
4876 goto end;
4877 }
4878 if (lttng_is_setuid_setgid()) {
4879 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4880 "--kmod-probes");
4881 } else {
4882 free(kmod_probes_list);
4883 kmod_probes_list = strdup(arg);
4884 if (!kmod_probes_list) {
4885 PERROR("strdup");
4886 ret = -ENOMEM;
4887 }
4888 }
4889 } else if (string_match(optname, "extra-kmod-probes")) {
4890 if (!arg || *arg == '\0') {
4891 ret = -EINVAL;
4892 goto end;
4893 }
4894 if (lttng_is_setuid_setgid()) {
4895 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4896 "--extra-kmod-probes");
4897 } else {
4898 free(kmod_extra_probes_list);
4899 kmod_extra_probes_list = strdup(arg);
4900 if (!kmod_extra_probes_list) {
4901 PERROR("strdup");
4902 ret = -ENOMEM;
4903 }
4904 }
4905 } else if (string_match(optname, "config") || opt == 'f') {
4906 /* This is handled in set_options() thus silent skip. */
4907 goto end;
4908 } else {
4909 /* Unknown option or other error.
4910 * Error is printed by getopt, just return */
4911 ret = -1;
4912 }
4913
4914 end:
4915 if (ret == -EINVAL) {
4916 const char *opt_name = "unknown";
4917 int i;
4918
4919 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4920 i++) {
4921 if (opt == long_options[i].val) {
4922 opt_name = long_options[i].name;
4923 break;
4924 }
4925 }
4926
4927 WARN("Invalid argument provided for option \"%s\", using default value.",
4928 opt_name);
4929 }
4930
4931 return ret;
4932 }
4933
4934 /*
4935 * config_entry_handler_cb used to handle options read from a config file.
4936 * See config_entry_handler_cb comment in common/config/config.h for the
4937 * return value conventions.
4938 */
4939 static int config_entry_handler(const struct config_entry *entry, void *unused)
4940 {
4941 int ret = 0, i;
4942
4943 if (!entry || !entry->name || !entry->value) {
4944 ret = -EINVAL;
4945 goto end;
4946 }
4947
4948 /* Check if the option is to be ignored */
4949 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4950 if (!strcmp(entry->name, config_ignore_options[i])) {
4951 goto end;
4952 }
4953 }
4954
4955 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4956 i++) {
4957
4958 /* Ignore if not fully matched. */
4959 if (strcmp(entry->name, long_options[i].name)) {
4960 continue;
4961 }
4962
4963 /*
4964 * If the option takes no argument on the command line, we have to
4965 * check if the value is "true". We support non-zero numeric values,
4966 * true, on and yes.
4967 */
4968 if (!long_options[i].has_arg) {
4969 ret = config_parse_value(entry->value);
4970 if (ret <= 0) {
4971 if (ret) {
4972 WARN("Invalid configuration value \"%s\" for option %s",
4973 entry->value, entry->name);
4974 }
4975 /* False, skip boolean config option. */
4976 goto end;
4977 }
4978 }
4979
4980 ret = set_option(long_options[i].val, entry->value, entry->name);
4981 goto end;
4982 }
4983
4984 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4985
4986 end:
4987 return ret;
4988 }
4989
4990 /*
4991 * daemon configuration loading and argument parsing
4992 */
4993 static int set_options(int argc, char **argv)
4994 {
4995 int ret = 0, c = 0, option_index = 0;
4996 int orig_optopt = optopt, orig_optind = optind;
4997 char *optstring;
4998 const char *config_path = NULL;
4999
5000 optstring = utils_generate_optstring(long_options,
5001 sizeof(long_options) / sizeof(struct option));
5002 if (!optstring) {
5003 ret = -ENOMEM;
5004 goto end;
5005 }
5006
5007 /* Check for the --config option */
5008 while ((c = getopt_long(argc, argv, optstring, long_options,
5009 &option_index)) != -1) {
5010 if (c == '?') {
5011 ret = -EINVAL;
5012 goto end;
5013 } else if (c != 'f') {
5014 /* if not equal to --config option. */
5015 continue;
5016 }
5017
5018 if (lttng_is_setuid_setgid()) {
5019 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5020 "-f, --config");
5021 } else {
5022 config_path = utils_expand_path(optarg);
5023 if (!config_path) {
5024 ERR("Failed to resolve path: %s", optarg);
5025 }
5026 }
5027 }
5028
5029 ret = config_get_section_entries(config_path, config_section_name,
5030 config_entry_handler, NULL);
5031 if (ret) {
5032 if (ret > 0) {
5033 ERR("Invalid configuration option at line %i", ret);
5034 ret = -1;
5035 }
5036 goto end;
5037 }
5038
5039 /* Reset getopt's global state */
5040 optopt = orig_optopt;
5041 optind = orig_optind;
5042 while (1) {
5043 option_index = -1;
5044 /*
5045 * getopt_long() will not set option_index if it encounters a
5046 * short option.
5047 */
5048 c = getopt_long(argc, argv, optstring, long_options,
5049 &option_index);
5050 if (c == -1) {
5051 break;
5052 }
5053
5054 /*
5055 * Pass NULL as the long option name if popt left the index
5056 * unset.
5057 */
5058 ret = set_option(c, optarg,
5059 option_index < 0 ? NULL :
5060 long_options[option_index].name);
5061 if (ret < 0) {
5062 break;
5063 }
5064 }
5065
5066 end:
5067 free(optstring);
5068 return ret;
5069 }
5070
5071 /*
5072 * Creates the two needed socket by the daemon.
5073 * apps_sock - The communication socket for all UST apps.
5074 * client_sock - The communication of the cli tool (lttng).
5075 */
5076 static int init_daemon_socket(void)
5077 {
5078 int ret = 0;
5079 mode_t old_umask;
5080
5081 old_umask = umask(0);
5082
5083 /* Create client tool unix socket */
5084 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5085 if (client_sock < 0) {
5086 ERR("Create unix sock failed: %s", client_unix_sock_path);
5087 ret = -1;
5088 goto end;
5089 }
5090
5091 /* Set the cloexec flag */
5092 ret = utils_set_fd_cloexec(client_sock);
5093 if (ret < 0) {
5094 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5095 "Continuing but note that the consumer daemon will have a "
5096 "reference to this socket on exec()", client_sock);
5097 }
5098
5099 /* File permission MUST be 660 */
5100 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5101 if (ret < 0) {
5102 ERR("Set file permissions failed: %s", client_unix_sock_path);
5103 PERROR("chmod");
5104 goto end;
5105 }
5106
5107 /* Create the application unix socket */
5108 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5109 if (apps_sock < 0) {
5110 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5111 ret = -1;
5112 goto end;
5113 }
5114
5115 /* Set the cloexec flag */
5116 ret = utils_set_fd_cloexec(apps_sock);
5117 if (ret < 0) {
5118 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5119 "Continuing but note that the consumer daemon will have a "
5120 "reference to this socket on exec()", apps_sock);
5121 }
5122
5123 /* File permission MUST be 666 */
5124 ret = chmod(apps_unix_sock_path,
5125 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5126 if (ret < 0) {
5127 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5128 PERROR("chmod");
5129 goto end;
5130 }
5131
5132 DBG3("Session daemon client socket %d and application socket %d created",
5133 client_sock, apps_sock);
5134
5135 end:
5136 umask(old_umask);
5137 return ret;
5138 }
5139
5140 /*
5141 * Check if the global socket is available, and if a daemon is answering at the
5142 * other side. If yes, error is returned.
5143 */
5144 static int check_existing_daemon(void)
5145 {
5146 /* Is there anybody out there ? */
5147 if (lttng_session_daemon_alive()) {
5148 return -EEXIST;
5149 }
5150
5151 return 0;
5152 }
5153
5154 /*
5155 * Set the tracing group gid onto the client socket.
5156 *
5157 * Race window between mkdir and chown is OK because we are going from more
5158 * permissive (root.root) to less permissive (root.tracing).
5159 */
5160 static int set_permissions(char *rundir)
5161 {
5162 int ret;
5163 gid_t gid;
5164
5165 gid = utils_get_group_id(tracing_group_name);
5166
5167 /* Set lttng run dir */
5168 ret = chown(rundir, 0, gid);
5169 if (ret < 0) {
5170 ERR("Unable to set group on %s", rundir);
5171 PERROR("chown");
5172 }
5173
5174 /*
5175 * Ensure all applications and tracing group can search the run
5176 * dir. Allow everyone to read the directory, since it does not
5177 * buy us anything to hide its content.
5178 */
5179 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5180 if (ret < 0) {
5181 ERR("Unable to set permissions on %s", rundir);
5182 PERROR("chmod");
5183 }
5184
5185 /* lttng client socket path */
5186 ret = chown(client_unix_sock_path, 0, gid);
5187 if (ret < 0) {
5188 ERR("Unable to set group on %s", client_unix_sock_path);
5189 PERROR("chown");
5190 }
5191
5192 /* kconsumer error socket path */
5193 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5194 if (ret < 0) {
5195 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5196 PERROR("chown");
5197 }
5198
5199 /* 64-bit ustconsumer error socket path */
5200 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5201 if (ret < 0) {
5202 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5203 PERROR("chown");
5204 }
5205
5206 /* 32-bit ustconsumer compat32 error socket path */
5207 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5208 if (ret < 0) {
5209 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5210 PERROR("chown");
5211 }
5212
5213 DBG("All permissions are set");
5214
5215 return ret;
5216 }
5217
5218 /*
5219 * Create the lttng run directory needed for all global sockets and pipe.
5220 */
5221 static int create_lttng_rundir(const char *rundir)
5222 {
5223 int ret;
5224
5225 DBG3("Creating LTTng run directory: %s", rundir);
5226
5227 ret = mkdir(rundir, S_IRWXU);
5228 if (ret < 0) {
5229 if (errno != EEXIST) {
5230 ERR("Unable to create %s", rundir);
5231 goto error;
5232 } else {
5233 ret = 0;
5234 }
5235 }
5236
5237 error:
5238 return ret;
5239 }
5240
5241 /*
5242 * Setup sockets and directory needed by the kconsumerd communication with the
5243 * session daemon.
5244 */
5245 static int set_consumer_sockets(struct consumer_data *consumer_data,
5246 const char *rundir)
5247 {
5248 int ret;
5249 char path[PATH_MAX];
5250
5251 switch (consumer_data->type) {
5252 case LTTNG_CONSUMER_KERNEL:
5253 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5254 break;
5255 case LTTNG_CONSUMER64_UST:
5256 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5257 break;
5258 case LTTNG_CONSUMER32_UST:
5259 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5260 break;
5261 default:
5262 ERR("Consumer type unknown");
5263 ret = -EINVAL;
5264 goto error;
5265 }
5266
5267 DBG2("Creating consumer directory: %s", path);
5268
5269 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5270 if (ret < 0) {
5271 if (errno != EEXIST) {
5272 PERROR("mkdir");
5273 ERR("Failed to create %s", path);
5274 goto error;
5275 }
5276 ret = -1;
5277 }
5278 if (is_root) {
5279 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5280 if (ret < 0) {
5281 ERR("Unable to set group on %s", path);
5282 PERROR("chown");
5283 goto error;
5284 }
5285 }
5286
5287 /* Create the kconsumerd error unix socket */
5288 consumer_data->err_sock =
5289 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5290 if (consumer_data->err_sock < 0) {
5291 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5292 ret = -1;
5293 goto error;
5294 }
5295
5296 /*
5297 * Set the CLOEXEC flag. Return code is useless because either way, the
5298 * show must go on.
5299 */
5300 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5301 if (ret < 0) {
5302 PERROR("utils_set_fd_cloexec");
5303 /* continue anyway */
5304 }
5305
5306 /* File permission MUST be 660 */
5307 ret = chmod(consumer_data->err_unix_sock_path,
5308 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5309 if (ret < 0) {
5310 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5311 PERROR("chmod");
5312 goto error;
5313 }
5314
5315 error:
5316 return ret;
5317 }
5318
5319 /*
5320 * Signal handler for the daemon
5321 *
5322 * Simply stop all worker threads, leaving main() return gracefully after
5323 * joining all threads and calling cleanup().
5324 */
5325 static void sighandler(int sig)
5326 {
5327 switch (sig) {
5328 case SIGPIPE:
5329 DBG("SIGPIPE caught");
5330 return;
5331 case SIGINT:
5332 DBG("SIGINT caught");
5333 stop_threads();
5334 break;
5335 case SIGTERM:
5336 DBG("SIGTERM caught");
5337 stop_threads();
5338 break;
5339 case SIGUSR1:
5340 CMM_STORE_SHARED(recv_child_signal, 1);
5341 break;
5342 default:
5343 break;
5344 }
5345 }
5346
5347 /*
5348 * Setup signal handler for :
5349 * SIGINT, SIGTERM, SIGPIPE
5350 */
5351 static int set_signal_handler(void)
5352 {
5353 int ret = 0;
5354 struct sigaction sa;
5355 sigset_t sigset;
5356
5357 if ((ret = sigemptyset(&sigset)) < 0) {
5358 PERROR("sigemptyset");
5359 return ret;
5360 }
5361
5362 sa.sa_handler = sighandler;
5363 sa.sa_mask = sigset;
5364 sa.sa_flags = 0;
5365 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5366 PERROR("sigaction");
5367 return ret;
5368 }
5369
5370 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5371 PERROR("sigaction");
5372 return ret;
5373 }
5374
5375 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5376 PERROR("sigaction");
5377 return ret;
5378 }
5379
5380 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5381 PERROR("sigaction");
5382 return ret;
5383 }
5384
5385 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5386
5387 return ret;
5388 }
5389
5390 /*
5391 * Set open files limit to unlimited. This daemon can open a large number of
5392 * file descriptors in order to consumer multiple kernel traces.
5393 */
5394 static void set_ulimit(void)
5395 {
5396 int ret;
5397 struct rlimit lim;
5398
5399 /* The kernel does not allowed an infinite limit for open files */
5400 lim.rlim_cur = 65535;
5401 lim.rlim_max = 65535;
5402
5403 ret = setrlimit(RLIMIT_NOFILE, &lim);
5404 if (ret < 0) {
5405 PERROR("failed to set open files limit");
5406 }
5407 }
5408
5409 /*
5410 * Write pidfile using the rundir and opt_pidfile.
5411 */
5412 static int write_pidfile(void)
5413 {
5414 int ret;
5415 char pidfile_path[PATH_MAX];
5416
5417 assert(rundir);
5418
5419 if (opt_pidfile) {
5420 if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) {
5421 ret = -1;
5422 goto error;
5423 }
5424 } else {
5425 /* Build pidfile path from rundir and opt_pidfile. */
5426 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5427 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5428 if (ret < 0) {
5429 PERROR("snprintf pidfile path");
5430 goto error;
5431 }
5432 }
5433
5434 /*
5435 * Create pid file in rundir.
5436 */
5437 ret = utils_create_pid_file(getpid(), pidfile_path);
5438 error:
5439 return ret;
5440 }
5441
5442 /*
5443 * Create lockfile using the rundir and return its fd.
5444 */
5445 static int create_lockfile(void)
5446 {
5447 int ret;
5448 char lockfile_path[PATH_MAX];
5449
5450 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5451 if (ret < 0) {
5452 goto error;
5453 }
5454
5455 ret = utils_create_lock_file(lockfile_path);
5456 error:
5457 return ret;
5458 }
5459
5460 /*
5461 * Write agent TCP port using the rundir.
5462 */
5463 static int write_agent_port(void)
5464 {
5465 int ret;
5466 char path[PATH_MAX];
5467
5468 assert(rundir);
5469
5470 ret = snprintf(path, sizeof(path), "%s/"
5471 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5472 if (ret < 0) {
5473 PERROR("snprintf agent port path");
5474 goto error;
5475 }
5476
5477 /*
5478 * Create TCP agent port file in rundir.
5479 */
5480 ret = utils_create_pid_file(agent_tcp_port, path);
5481
5482 error:
5483 return ret;
5484 }
5485
5486 /*
5487 * main
5488 */
5489 int main(int argc, char **argv)
5490 {
5491 int ret = 0, retval = 0;
5492 void *status;
5493 const char *home_path, *env_app_timeout;
5494
5495 /* Initialize agent apps ht global variable */
5496 agent_apps_ht_by_sock = NULL;
5497
5498 init_kernel_workarounds();
5499
5500 rcu_register_thread();
5501
5502 if (set_signal_handler()) {
5503 retval = -1;
5504 goto exit_set_signal_handler;
5505 }
5506
5507 setup_consumerd_path();
5508
5509 page_size = sysconf(_SC_PAGESIZE);
5510 if (page_size < 0) {
5511 PERROR("sysconf _SC_PAGESIZE");
5512 page_size = LONG_MAX;
5513 WARN("Fallback page size to %ld", page_size);
5514 }
5515
5516 /*
5517 * Parse arguments and load the daemon configuration file.
5518 *
5519 * We have an exit_options exit path to free memory reserved by
5520 * set_options. This is needed because the rest of sessiond_cleanup()
5521 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5522 * depends on set_options.
5523 */
5524 progname = argv[0];
5525 if (set_options(argc, argv)) {
5526 retval = -1;
5527 goto exit_options;
5528 }
5529
5530 /* Daemonize */
5531 if (opt_daemon || opt_background) {
5532 int i;
5533
5534 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5535 !opt_background);
5536 if (ret < 0) {
5537 retval = -1;
5538 goto exit_options;
5539 }
5540
5541 /*
5542 * We are in the child. Make sure all other file descriptors are
5543 * closed, in case we are called with more opened file
5544 * descriptors than the standard ones.
5545 */
5546 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5547 (void) close(i);
5548 }
5549 }
5550
5551 if (run_as_create_worker(argv[0]) < 0) {
5552 goto exit_create_run_as_worker_cleanup;
5553 }
5554
5555 /*
5556 * Starting from here, we can create threads. This needs to be after
5557 * lttng_daemonize due to RCU.
5558 */
5559
5560 /*
5561 * Initialize the health check subsystem. This call should set the
5562 * appropriate time values.
5563 */
5564 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5565 if (!health_sessiond) {
5566 PERROR("health_app_create error");
5567 retval = -1;
5568 goto exit_health_sessiond_cleanup;
5569 }
5570
5571 if (init_ht_cleanup_quit_pipe()) {
5572 retval = -1;
5573 goto exit_ht_cleanup_quit_pipe;
5574 }
5575
5576 /* Setup the thread ht_cleanup communication pipe. */
5577 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5578 retval = -1;
5579 goto exit_ht_cleanup_pipe;
5580 }
5581
5582 /* Set up max poll set size */
5583 if (lttng_poll_set_max_size()) {
5584 retval = -1;
5585 goto exit_set_max_size;
5586 }
5587
5588 /* Create thread to clean up RCU hash tables */
5589 ret = pthread_create(&ht_cleanup_thread, NULL,
5590 thread_ht_cleanup, (void *) NULL);
5591 if (ret) {
5592 errno = ret;
5593 PERROR("pthread_create ht_cleanup");
5594 retval = -1;
5595 goto exit_ht_cleanup;
5596 }
5597
5598 /* Create thread quit pipe */
5599 if (init_thread_quit_pipe()) {
5600 retval = -1;
5601 goto exit_init_data;
5602 }
5603
5604 /* Check if daemon is UID = 0 */
5605 is_root = !getuid();
5606
5607 if (is_root) {
5608 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5609 if (!rundir) {
5610 retval = -1;
5611 goto exit_init_data;
5612 }
5613
5614 /* Create global run dir with root access */
5615 if (create_lttng_rundir(rundir)) {
5616 retval = -1;
5617 goto exit_init_data;
5618 }
5619
5620 if (strlen(apps_unix_sock_path) == 0) {
5621 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5622 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5623 if (ret < 0) {
5624 retval = -1;
5625 goto exit_init_data;
5626 }
5627 }
5628
5629 if (strlen(client_unix_sock_path) == 0) {
5630 ret = snprintf(client_unix_sock_path, PATH_MAX,
5631 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5632 if (ret < 0) {
5633 retval = -1;
5634 goto exit_init_data;
5635 }
5636 }
5637
5638 /* Set global SHM for ust */
5639 if (strlen(wait_shm_path) == 0) {
5640 ret = snprintf(wait_shm_path, PATH_MAX,
5641 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5642 if (ret < 0) {
5643 retval = -1;
5644 goto exit_init_data;
5645 }
5646 }
5647
5648 if (strlen(health_unix_sock_path) == 0) {
5649 ret = snprintf(health_unix_sock_path,
5650 sizeof(health_unix_sock_path),
5651 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5652 if (ret < 0) {
5653 retval = -1;
5654 goto exit_init_data;
5655 }
5656 }
5657
5658 /* Setup kernel consumerd path */
5659 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5660 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5661 if (ret < 0) {
5662 retval = -1;
5663 goto exit_init_data;
5664 }
5665 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5666 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5667 if (ret < 0) {
5668 retval = -1;
5669 goto exit_init_data;
5670 }
5671
5672 DBG2("Kernel consumer err path: %s",
5673 kconsumer_data.err_unix_sock_path);
5674 DBG2("Kernel consumer cmd path: %s",
5675 kconsumer_data.cmd_unix_sock_path);
5676 } else {
5677 home_path = utils_get_home_dir();
5678 if (home_path == NULL) {
5679 /* TODO: Add --socket PATH option */
5680 ERR("Can't get HOME directory for sockets creation.");
5681 retval = -1;
5682 goto exit_init_data;
5683 }
5684
5685 /*
5686 * Create rundir from home path. This will create something like
5687 * $HOME/.lttng
5688 */
5689 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5690 if (ret < 0) {
5691 retval = -1;
5692 goto exit_init_data;
5693 }
5694
5695 if (create_lttng_rundir(rundir)) {
5696 retval = -1;
5697 goto exit_init_data;
5698 }
5699
5700 if (strlen(apps_unix_sock_path) == 0) {
5701 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5702 DEFAULT_HOME_APPS_UNIX_SOCK,
5703 home_path);
5704 if (ret < 0) {
5705 retval = -1;
5706 goto exit_init_data;
5707 }
5708 }
5709
5710 /* Set the cli tool unix socket path */
5711 if (strlen(client_unix_sock_path) == 0) {
5712 ret = snprintf(client_unix_sock_path, PATH_MAX,
5713 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5714 home_path);
5715 if (ret < 0) {
5716 retval = -1;
5717 goto exit_init_data;
5718 }
5719 }
5720
5721 /* Set global SHM for ust */
5722 if (strlen(wait_shm_path) == 0) {
5723 ret = snprintf(wait_shm_path, PATH_MAX,
5724 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5725 getuid());
5726 if (ret < 0) {
5727 retval = -1;
5728 goto exit_init_data;
5729 }
5730 }
5731
5732 /* Set health check Unix path */
5733 if (strlen(health_unix_sock_path) == 0) {
5734 ret = snprintf(health_unix_sock_path,
5735 sizeof(health_unix_sock_path),
5736 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5737 home_path);
5738 if (ret < 0) {
5739 retval = -1;
5740 goto exit_init_data;
5741 }
5742 }
5743 }
5744
5745 lockfile_fd = create_lockfile();
5746 if (lockfile_fd < 0) {
5747 retval = -1;
5748 goto exit_init_data;
5749 }
5750
5751 /* Set consumer initial state */
5752 kernel_consumerd_state = CONSUMER_STOPPED;
5753 ust_consumerd_state = CONSUMER_STOPPED;
5754
5755 DBG("Client socket path %s", client_unix_sock_path);
5756 DBG("Application socket path %s", apps_unix_sock_path);
5757 DBG("Application wait path %s", wait_shm_path);
5758 DBG("LTTng run directory path: %s", rundir);
5759
5760 /* 32 bits consumerd path setup */
5761 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5762 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5763 if (ret < 0) {
5764 PERROR("snprintf 32-bit consumer error socket path");
5765 retval = -1;
5766 goto exit_init_data;
5767 }
5768 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5769 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5770 if (ret < 0) {
5771 PERROR("snprintf 32-bit consumer command socket path");
5772 retval = -1;
5773 goto exit_init_data;
5774 }
5775
5776 DBG2("UST consumer 32 bits err path: %s",
5777 ustconsumer32_data.err_unix_sock_path);
5778 DBG2("UST consumer 32 bits cmd path: %s",
5779 ustconsumer32_data.cmd_unix_sock_path);
5780
5781 /* 64 bits consumerd path setup */
5782 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5783 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5784 if (ret < 0) {
5785 PERROR("snprintf 64-bit consumer error socket path");
5786 retval = -1;
5787 goto exit_init_data;
5788 }
5789 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5790 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5791 if (ret < 0) {
5792 PERROR("snprintf 64-bit consumer command socket path");
5793 retval = -1;
5794 goto exit_init_data;
5795 }
5796
5797 DBG2("UST consumer 64 bits err path: %s",
5798 ustconsumer64_data.err_unix_sock_path);
5799 DBG2("UST consumer 64 bits cmd path: %s",
5800 ustconsumer64_data.cmd_unix_sock_path);
5801
5802 /*
5803 * See if daemon already exist.
5804 */
5805 if (check_existing_daemon()) {
5806 ERR("Already running daemon.\n");
5807 /*
5808 * We do not goto exit because we must not cleanup()
5809 * because a daemon is already running.
5810 */
5811 retval = -1;
5812 goto exit_init_data;
5813 }
5814
5815 /*
5816 * Init UST app hash table. Alloc hash table before this point since
5817 * cleanup() can get called after that point.
5818 */
5819 if (ust_app_ht_alloc()) {
5820 ERR("Failed to allocate UST app hash table");
5821 retval = -1;
5822 goto exit_init_data;
5823 }
5824
5825 /*
5826 * Initialize agent app hash table. We allocate the hash table here
5827 * since cleanup() can get called after this point.
5828 */
5829 if (agent_app_ht_alloc()) {
5830 ERR("Failed to allocate Agent app hash table");
5831 retval = -1;
5832 goto exit_init_data;
5833 }
5834
5835 /*
5836 * These actions must be executed as root. We do that *after* setting up
5837 * the sockets path because we MUST make the check for another daemon using
5838 * those paths *before* trying to set the kernel consumer sockets and init
5839 * kernel tracer.
5840 */
5841 if (is_root) {
5842 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5843 retval = -1;
5844 goto exit_init_data;
5845 }
5846
5847 /* Setup kernel tracer */
5848 if (!opt_no_kernel) {
5849 init_kernel_tracer();
5850 if (kernel_tracer_fd >= 0) {
5851 ret = syscall_init_table();
5852 if (ret < 0) {
5853 ERR("Unable to populate syscall table. "
5854 "Syscall tracing won't work "
5855 "for this session daemon.");
5856 }
5857 }
5858 }
5859
5860 /* Set ulimit for open files */
5861 set_ulimit();
5862 }
5863 /* init lttng_fd tracking must be done after set_ulimit. */
5864 lttng_fd_init();
5865
5866 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5867 retval = -1;
5868 goto exit_init_data;
5869 }
5870
5871 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5872 retval = -1;
5873 goto exit_init_data;
5874 }
5875
5876 /* Setup the needed unix socket */
5877 if (init_daemon_socket()) {
5878 retval = -1;
5879 goto exit_init_data;
5880 }
5881
5882 /* Set credentials to socket */
5883 if (is_root && set_permissions(rundir)) {
5884 retval = -1;
5885 goto exit_init_data;
5886 }
5887
5888 /* Get parent pid if -S, --sig-parent is specified. */
5889 if (opt_sig_parent) {
5890 ppid = getppid();
5891 }
5892
5893 /* Setup the kernel pipe for waking up the kernel thread */
5894 if (is_root && !opt_no_kernel) {
5895 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5896 retval = -1;
5897 goto exit_init_data;
5898 }
5899 }
5900
5901 /* Setup the thread apps communication pipe. */
5902 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5903 retval = -1;
5904 goto exit_init_data;
5905 }
5906
5907 /* Setup the thread apps notify communication pipe. */
5908 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5909 retval = -1;
5910 goto exit_init_data;
5911 }
5912
5913 /* Initialize global buffer per UID and PID registry. */
5914 buffer_reg_init_uid_registry();
5915 buffer_reg_init_pid_registry();
5916
5917 /* Init UST command queue. */
5918 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5919
5920 /*
5921 * Get session list pointer. This pointer MUST NOT be free'd. This list
5922 * is statically declared in session.c
5923 */
5924 session_list_ptr = session_get_list();
5925
5926 cmd_init();
5927
5928 /* Check for the application socket timeout env variable. */
5929 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5930 if (env_app_timeout) {
5931 app_socket_timeout = atoi(env_app_timeout);
5932 } else {
5933 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5934 }
5935
5936 ret = write_pidfile();
5937 if (ret) {
5938 ERR("Error in write_pidfile");
5939 retval = -1;
5940 goto exit_init_data;
5941 }
5942 ret = write_agent_port();
5943 if (ret) {
5944 ERR("Error in write_agent_port");
5945 retval = -1;
5946 goto exit_init_data;
5947 }
5948
5949 /* Initialize communication library */
5950 lttcomm_init();
5951 /* Initialize TCP timeout values */
5952 lttcomm_inet_init();
5953
5954 if (load_session_init_data(&load_info) < 0) {
5955 retval = -1;
5956 goto exit_init_data;
5957 }
5958 load_info->path = opt_load_session_path;
5959
5960 /* Create health-check thread */
5961 ret = pthread_create(&health_thread, NULL,
5962 thread_manage_health, (void *) NULL);
5963 if (ret) {
5964 errno = ret;
5965 PERROR("pthread_create health");
5966 retval = -1;
5967 goto exit_health;
5968 }
5969
5970 /* Create thread to manage the client socket */
5971 ret = pthread_create(&client_thread, NULL,
5972 thread_manage_clients, (void *) NULL);
5973 if (ret) {
5974 errno = ret;
5975 PERROR("pthread_create clients");
5976 retval = -1;
5977 goto exit_client;
5978 }
5979
5980 /* Create thread to dispatch registration */
5981 ret = pthread_create(&dispatch_thread, NULL,
5982 thread_dispatch_ust_registration, (void *) NULL);
5983 if (ret) {
5984 errno = ret;
5985 PERROR("pthread_create dispatch");
5986 retval = -1;
5987 goto exit_dispatch;
5988 }
5989
5990 /* Create thread to manage application registration. */
5991 ret = pthread_create(&reg_apps_thread, NULL,
5992 thread_registration_apps, (void *) NULL);
5993 if (ret) {
5994 errno = ret;
5995 PERROR("pthread_create registration");
5996 retval = -1;
5997 goto exit_reg_apps;
5998 }
5999
6000 /* Create thread to manage application socket */
6001 ret = pthread_create(&apps_thread, NULL,
6002 thread_manage_apps, (void *) NULL);
6003 if (ret) {
6004 errno = ret;
6005 PERROR("pthread_create apps");
6006 retval = -1;
6007 goto exit_apps;
6008 }
6009
6010 /* Create thread to manage application notify socket */
6011 ret = pthread_create(&apps_notify_thread, NULL,
6012 ust_thread_manage_notify, (void *) NULL);
6013 if (ret) {
6014 errno = ret;
6015 PERROR("pthread_create notify");
6016 retval = -1;
6017 goto exit_apps_notify;
6018 }
6019
6020 /* Create agent registration thread. */
6021 ret = pthread_create(&agent_reg_thread, NULL,
6022 agent_thread_manage_registration, (void *) NULL);
6023 if (ret) {
6024 errno = ret;
6025 PERROR("pthread_create agent");
6026 retval = -1;
6027 goto exit_agent_reg;
6028 }
6029
6030 /* Don't start this thread if kernel tracing is not requested nor root */
6031 if (is_root && !opt_no_kernel) {
6032 /* Create kernel thread to manage kernel event */
6033 ret = pthread_create(&kernel_thread, NULL,
6034 thread_manage_kernel, (void *) NULL);
6035 if (ret) {
6036 errno = ret;
6037 PERROR("pthread_create kernel");
6038 retval = -1;
6039 goto exit_kernel;
6040 }
6041 }
6042
6043 /* Create session loading thread. */
6044 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
6045 load_info);
6046 if (ret) {
6047 errno = ret;
6048 PERROR("pthread_create load_session_thread");
6049 retval = -1;
6050 goto exit_load_session;
6051 }
6052
6053 /*
6054 * This is where we start awaiting program completion (e.g. through
6055 * signal that asks threads to teardown).
6056 */
6057
6058 ret = pthread_join(load_session_thread, &status);
6059 if (ret) {
6060 errno = ret;
6061 PERROR("pthread_join load_session_thread");
6062 retval = -1;
6063 }
6064 exit_load_session:
6065
6066 if (is_root && !opt_no_kernel) {
6067 ret = pthread_join(kernel_thread, &status);
6068 if (ret) {
6069 errno = ret;
6070 PERROR("pthread_join");
6071 retval = -1;
6072 }
6073 }
6074 exit_kernel:
6075
6076 ret = pthread_join(agent_reg_thread, &status);
6077 if (ret) {
6078 errno = ret;
6079 PERROR("pthread_join agent");
6080 retval = -1;
6081 }
6082 exit_agent_reg:
6083
6084 ret = pthread_join(apps_notify_thread, &status);
6085 if (ret) {
6086 errno = ret;
6087 PERROR("pthread_join apps notify");
6088 retval = -1;
6089 }
6090 exit_apps_notify:
6091
6092 ret = pthread_join(apps_thread, &status);
6093 if (ret) {
6094 errno = ret;
6095 PERROR("pthread_join apps");
6096 retval = -1;
6097 }
6098 exit_apps:
6099
6100 ret = pthread_join(reg_apps_thread, &status);
6101 if (ret) {
6102 errno = ret;
6103 PERROR("pthread_join");
6104 retval = -1;
6105 }
6106 exit_reg_apps:
6107
6108 /*
6109 * Join dispatch thread after joining reg_apps_thread to ensure
6110 * we don't leak applications in the queue.
6111 */
6112 ret = pthread_join(dispatch_thread, &status);
6113 if (ret) {
6114 errno = ret;
6115 PERROR("pthread_join");
6116 retval = -1;
6117 }
6118 exit_dispatch:
6119
6120 ret = pthread_join(client_thread, &status);
6121 if (ret) {
6122 errno = ret;
6123 PERROR("pthread_join");
6124 retval = -1;
6125 }
6126 exit_client:
6127
6128 ret = pthread_join(health_thread, &status);
6129 if (ret) {
6130 errno = ret;
6131 PERROR("pthread_join health thread");
6132 retval = -1;
6133 }
6134 exit_health:
6135
6136 exit_init_data:
6137 /*
6138 * sessiond_cleanup() is called when no other thread is running, except
6139 * the ht_cleanup thread, which is needed to destroy the hash tables.
6140 */
6141 rcu_thread_online();
6142 sessiond_cleanup();
6143 rcu_thread_offline();
6144 rcu_unregister_thread();
6145
6146 /*
6147 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6148 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6149 * the queue is empty before shutting down the clean-up thread.
6150 */
6151 rcu_barrier();
6152
6153 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6154 if (ret < 0) {
6155 ERR("write error on ht_cleanup quit pipe");
6156 retval = -1;
6157 }
6158
6159 ret = pthread_join(ht_cleanup_thread, &status);
6160 if (ret) {
6161 errno = ret;
6162 PERROR("pthread_join ht cleanup thread");
6163 retval = -1;
6164 }
6165 exit_ht_cleanup:
6166 exit_set_max_size:
6167
6168 utils_close_pipe(ht_cleanup_pipe);
6169 exit_ht_cleanup_pipe:
6170
6171 /*
6172 * Close the ht_cleanup quit pipe.
6173 */
6174 utils_close_pipe(ht_cleanup_quit_pipe);
6175 exit_ht_cleanup_quit_pipe:
6176
6177 health_app_destroy(health_sessiond);
6178 exit_health_sessiond_cleanup:
6179 exit_create_run_as_worker_cleanup:
6180
6181 exit_options:
6182 sessiond_cleanup_options();
6183
6184 exit_set_signal_handler:
6185
6186 if (!retval) {
6187 exit(EXIT_SUCCESS);
6188 } else {
6189 exit(EXIT_FAILURE);
6190 }
6191 }
This page took 0.156571 seconds and 4 git commands to generate.