Fix: leak of reply buffer on data pending check
[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 (void) lttcomm_setsockopt_rcv_timeout(sock,
2215 app_socket_timeout * 1000);
2216 (void) lttcomm_setsockopt_snd_timeout(sock,
2217 app_socket_timeout * 1000);
2218
2219 /*
2220 * Set the CLOEXEC flag. Return code is useless because
2221 * either way, the show must go on.
2222 */
2223 (void) utils_set_fd_cloexec(sock);
2224
2225 /* Create UST registration command for enqueuing */
2226 ust_cmd = zmalloc(sizeof(struct ust_command));
2227 if (ust_cmd == NULL) {
2228 PERROR("ust command zmalloc");
2229 ret = close(sock);
2230 if (ret) {
2231 PERROR("close");
2232 }
2233 goto error;
2234 }
2235
2236 /*
2237 * Using message-based transmissions to ensure we don't
2238 * have to deal with partially received messages.
2239 */
2240 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2241 if (ret < 0) {
2242 ERR("Exhausted file descriptors allowed for applications.");
2243 free(ust_cmd);
2244 ret = close(sock);
2245 if (ret) {
2246 PERROR("close");
2247 }
2248 sock = -1;
2249 continue;
2250 }
2251
2252 health_code_update();
2253 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2254 if (ret < 0) {
2255 free(ust_cmd);
2256 /* Close socket of the application. */
2257 ret = close(sock);
2258 if (ret) {
2259 PERROR("close");
2260 }
2261 lttng_fd_put(LTTNG_FD_APPS, 1);
2262 sock = -1;
2263 continue;
2264 }
2265 health_code_update();
2266
2267 ust_cmd->sock = sock;
2268 sock = -1;
2269
2270 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2271 " gid:%d sock:%d name:%s (version %d.%d)",
2272 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2273 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2274 ust_cmd->sock, ust_cmd->reg_msg.name,
2275 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2276
2277 /*
2278 * Lock free enqueue the registration request. The red pill
2279 * has been taken! This apps will be part of the *system*.
2280 */
2281 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2282
2283 /*
2284 * Wake the registration queue futex. Implicit memory
2285 * barrier with the exchange in cds_wfcq_enqueue.
2286 */
2287 futex_nto1_wake(&ust_cmd_queue.futex);
2288 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2289 ERR("Register apps socket poll error");
2290 goto error;
2291 } else {
2292 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2293 goto error;
2294 }
2295 }
2296 }
2297 }
2298
2299 exit:
2300 error:
2301 /* Notify that the registration thread is gone */
2302 notify_ust_apps(0);
2303
2304 if (apps_sock >= 0) {
2305 ret = close(apps_sock);
2306 if (ret) {
2307 PERROR("close");
2308 }
2309 }
2310 if (sock >= 0) {
2311 ret = close(sock);
2312 if (ret) {
2313 PERROR("close");
2314 }
2315 lttng_fd_put(LTTNG_FD_APPS, 1);
2316 }
2317 unlink(apps_unix_sock_path);
2318
2319 error_poll_add:
2320 lttng_poll_clean(&events);
2321 error_listen:
2322 error_create_poll:
2323 error_testpoint:
2324 DBG("UST Registration thread cleanup complete");
2325 if (err) {
2326 health_error();
2327 ERR("Health error occurred in %s", __func__);
2328 }
2329 health_unregister(health_sessiond);
2330
2331 return NULL;
2332 }
2333
2334 /*
2335 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2336 * exec or it will fails.
2337 */
2338 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2339 {
2340 int ret, clock_ret;
2341 struct timespec timeout;
2342
2343 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2344 consumer_data->consumer_thread_is_ready = 0;
2345
2346 /* Setup pthread condition */
2347 ret = pthread_condattr_init(&consumer_data->condattr);
2348 if (ret) {
2349 errno = ret;
2350 PERROR("pthread_condattr_init consumer data");
2351 goto error;
2352 }
2353
2354 /*
2355 * Set the monotonic clock in order to make sure we DO NOT jump in time
2356 * between the clock_gettime() call and the timedwait call. See bug #324
2357 * for a more details and how we noticed it.
2358 */
2359 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2360 if (ret) {
2361 errno = ret;
2362 PERROR("pthread_condattr_setclock consumer data");
2363 goto error;
2364 }
2365
2366 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2367 if (ret) {
2368 errno = ret;
2369 PERROR("pthread_cond_init consumer data");
2370 goto error;
2371 }
2372
2373 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2374 consumer_data);
2375 if (ret) {
2376 errno = ret;
2377 PERROR("pthread_create consumer");
2378 ret = -1;
2379 goto error;
2380 }
2381
2382 /* We are about to wait on a pthread condition */
2383 pthread_mutex_lock(&consumer_data->cond_mutex);
2384
2385 /* Get time for sem_timedwait absolute timeout */
2386 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2387 /*
2388 * Set the timeout for the condition timed wait even if the clock gettime
2389 * call fails since we might loop on that call and we want to avoid to
2390 * increment the timeout too many times.
2391 */
2392 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2393
2394 /*
2395 * The following loop COULD be skipped in some conditions so this is why we
2396 * set ret to 0 in order to make sure at least one round of the loop is
2397 * done.
2398 */
2399 ret = 0;
2400
2401 /*
2402 * Loop until the condition is reached or when a timeout is reached. Note
2403 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2404 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2405 * possible. This loop does not take any chances and works with both of
2406 * them.
2407 */
2408 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2409 if (clock_ret < 0) {
2410 PERROR("clock_gettime spawn consumer");
2411 /* Infinite wait for the consumerd thread to be ready */
2412 ret = pthread_cond_wait(&consumer_data->cond,
2413 &consumer_data->cond_mutex);
2414 } else {
2415 ret = pthread_cond_timedwait(&consumer_data->cond,
2416 &consumer_data->cond_mutex, &timeout);
2417 }
2418 }
2419
2420 /* Release the pthread condition */
2421 pthread_mutex_unlock(&consumer_data->cond_mutex);
2422
2423 if (ret != 0) {
2424 errno = ret;
2425 if (ret == ETIMEDOUT) {
2426 int pth_ret;
2427
2428 /*
2429 * Call has timed out so we kill the kconsumerd_thread and return
2430 * an error.
2431 */
2432 ERR("Condition timed out. The consumer thread was never ready."
2433 " Killing it");
2434 pth_ret = pthread_cancel(consumer_data->thread);
2435 if (pth_ret < 0) {
2436 PERROR("pthread_cancel consumer thread");
2437 }
2438 } else {
2439 PERROR("pthread_cond_wait failed consumer thread");
2440 }
2441 /* Caller is expecting a negative value on failure. */
2442 ret = -1;
2443 goto error;
2444 }
2445
2446 pthread_mutex_lock(&consumer_data->pid_mutex);
2447 if (consumer_data->pid == 0) {
2448 ERR("Consumerd did not start");
2449 pthread_mutex_unlock(&consumer_data->pid_mutex);
2450 goto error;
2451 }
2452 pthread_mutex_unlock(&consumer_data->pid_mutex);
2453
2454 return 0;
2455
2456 error:
2457 return ret;
2458 }
2459
2460 /*
2461 * Join consumer thread
2462 */
2463 static int join_consumer_thread(struct consumer_data *consumer_data)
2464 {
2465 void *status;
2466
2467 /* Consumer pid must be a real one. */
2468 if (consumer_data->pid > 0) {
2469 int ret;
2470 ret = kill(consumer_data->pid, SIGTERM);
2471 if (ret) {
2472 PERROR("Error killing consumer daemon");
2473 return ret;
2474 }
2475 return pthread_join(consumer_data->thread, &status);
2476 } else {
2477 return 0;
2478 }
2479 }
2480
2481 /*
2482 * Fork and exec a consumer daemon (consumerd).
2483 *
2484 * Return pid if successful else -1.
2485 */
2486 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2487 {
2488 int ret;
2489 pid_t pid;
2490 const char *consumer_to_use;
2491 const char *verbosity;
2492 struct stat st;
2493
2494 DBG("Spawning consumerd");
2495
2496 pid = fork();
2497 if (pid == 0) {
2498 /*
2499 * Exec consumerd.
2500 */
2501 if (opt_verbose_consumer) {
2502 verbosity = "--verbose";
2503 } else if (lttng_opt_quiet) {
2504 verbosity = "--quiet";
2505 } else {
2506 verbosity = "";
2507 }
2508
2509 switch (consumer_data->type) {
2510 case LTTNG_CONSUMER_KERNEL:
2511 /*
2512 * Find out which consumerd to execute. We will first try the
2513 * 64-bit path, then the sessiond's installation directory, and
2514 * fallback on the 32-bit one,
2515 */
2516 DBG3("Looking for a kernel consumer at these locations:");
2517 DBG3(" 1) %s", consumerd64_bin);
2518 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2519 DBG3(" 3) %s", consumerd32_bin);
2520 if (stat(consumerd64_bin, &st) == 0) {
2521 DBG3("Found location #1");
2522 consumer_to_use = consumerd64_bin;
2523 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2524 DBG3("Found location #2");
2525 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2526 } else if (stat(consumerd32_bin, &st) == 0) {
2527 DBG3("Found location #3");
2528 consumer_to_use = consumerd32_bin;
2529 } else {
2530 DBG("Could not find any valid consumerd executable");
2531 ret = -EINVAL;
2532 break;
2533 }
2534 DBG("Using kernel consumer at: %s", consumer_to_use);
2535 ret = execl(consumer_to_use,
2536 "lttng-consumerd", verbosity, "-k",
2537 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2538 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2539 "--group", tracing_group_name,
2540 NULL);
2541 break;
2542 case LTTNG_CONSUMER64_UST:
2543 {
2544 char *tmpnew = NULL;
2545
2546 if (consumerd64_libdir[0] != '\0') {
2547 char *tmp;
2548 size_t tmplen;
2549
2550 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2551 if (!tmp) {
2552 tmp = "";
2553 }
2554 tmplen = strlen("LD_LIBRARY_PATH=")
2555 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2556 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2557 if (!tmpnew) {
2558 ret = -ENOMEM;
2559 goto error;
2560 }
2561 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2562 strcat(tmpnew, consumerd64_libdir);
2563 if (tmp[0] != '\0') {
2564 strcat(tmpnew, ":");
2565 strcat(tmpnew, tmp);
2566 }
2567 ret = putenv(tmpnew);
2568 if (ret) {
2569 ret = -errno;
2570 free(tmpnew);
2571 goto error;
2572 }
2573 }
2574 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2575 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2576 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2577 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2578 "--group", tracing_group_name,
2579 NULL);
2580 if (consumerd64_libdir[0] != '\0') {
2581 free(tmpnew);
2582 }
2583 break;
2584 }
2585 case LTTNG_CONSUMER32_UST:
2586 {
2587 char *tmpnew = NULL;
2588
2589 if (consumerd32_libdir[0] != '\0') {
2590 char *tmp;
2591 size_t tmplen;
2592
2593 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2594 if (!tmp) {
2595 tmp = "";
2596 }
2597 tmplen = strlen("LD_LIBRARY_PATH=")
2598 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2599 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2600 if (!tmpnew) {
2601 ret = -ENOMEM;
2602 goto error;
2603 }
2604 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2605 strcat(tmpnew, consumerd32_libdir);
2606 if (tmp[0] != '\0') {
2607 strcat(tmpnew, ":");
2608 strcat(tmpnew, tmp);
2609 }
2610 ret = putenv(tmpnew);
2611 if (ret) {
2612 ret = -errno;
2613 free(tmpnew);
2614 goto error;
2615 }
2616 }
2617 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2618 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2619 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2620 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2621 "--group", tracing_group_name,
2622 NULL);
2623 if (consumerd32_libdir[0] != '\0') {
2624 free(tmpnew);
2625 }
2626 break;
2627 }
2628 default:
2629 PERROR("unknown consumer type");
2630 exit(EXIT_FAILURE);
2631 }
2632 if (errno != 0) {
2633 PERROR("Consumer execl()");
2634 }
2635 /* Reaching this point, we got a failure on our execl(). */
2636 exit(EXIT_FAILURE);
2637 } else if (pid > 0) {
2638 ret = pid;
2639 } else {
2640 PERROR("start consumer fork");
2641 ret = -errno;
2642 }
2643 error:
2644 return ret;
2645 }
2646
2647 /*
2648 * Spawn the consumerd daemon and session daemon thread.
2649 */
2650 static int start_consumerd(struct consumer_data *consumer_data)
2651 {
2652 int ret;
2653
2654 /*
2655 * Set the listen() state on the socket since there is a possible race
2656 * between the exec() of the consumer daemon and this call if place in the
2657 * consumer thread. See bug #366 for more details.
2658 */
2659 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2660 if (ret < 0) {
2661 goto error;
2662 }
2663
2664 pthread_mutex_lock(&consumer_data->pid_mutex);
2665 if (consumer_data->pid != 0) {
2666 pthread_mutex_unlock(&consumer_data->pid_mutex);
2667 goto end;
2668 }
2669
2670 ret = spawn_consumerd(consumer_data);
2671 if (ret < 0) {
2672 ERR("Spawning consumerd failed");
2673 pthread_mutex_unlock(&consumer_data->pid_mutex);
2674 goto error;
2675 }
2676
2677 /* Setting up the consumer_data pid */
2678 consumer_data->pid = ret;
2679 DBG2("Consumer pid %d", consumer_data->pid);
2680 pthread_mutex_unlock(&consumer_data->pid_mutex);
2681
2682 DBG2("Spawning consumer control thread");
2683 ret = spawn_consumer_thread(consumer_data);
2684 if (ret < 0) {
2685 ERR("Fatal error spawning consumer control thread");
2686 goto error;
2687 }
2688
2689 end:
2690 return 0;
2691
2692 error:
2693 /* Cleanup already created sockets on error. */
2694 if (consumer_data->err_sock >= 0) {
2695 int err;
2696
2697 err = close(consumer_data->err_sock);
2698 if (err < 0) {
2699 PERROR("close consumer data error socket");
2700 }
2701 }
2702 return ret;
2703 }
2704
2705 /*
2706 * Setup necessary data for kernel tracer action.
2707 */
2708 static int init_kernel_tracer(void)
2709 {
2710 int ret;
2711
2712 /* Modprobe lttng kernel modules */
2713 ret = modprobe_lttng_control();
2714 if (ret < 0) {
2715 goto error;
2716 }
2717
2718 /* Open debugfs lttng */
2719 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2720 if (kernel_tracer_fd < 0) {
2721 DBG("Failed to open %s", module_proc_lttng);
2722 ret = -1;
2723 goto error_open;
2724 }
2725
2726 /* Validate kernel version */
2727 ret = kernel_validate_version(kernel_tracer_fd);
2728 if (ret < 0) {
2729 goto error_version;
2730 }
2731
2732 ret = modprobe_lttng_data();
2733 if (ret < 0) {
2734 goto error_modules;
2735 }
2736
2737 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2738 return 0;
2739
2740 error_version:
2741 modprobe_remove_lttng_control();
2742 ret = close(kernel_tracer_fd);
2743 if (ret) {
2744 PERROR("close");
2745 }
2746 kernel_tracer_fd = -1;
2747 return LTTNG_ERR_KERN_VERSION;
2748
2749 error_modules:
2750 ret = close(kernel_tracer_fd);
2751 if (ret) {
2752 PERROR("close");
2753 }
2754
2755 error_open:
2756 modprobe_remove_lttng_control();
2757
2758 error:
2759 WARN("No kernel tracer available");
2760 kernel_tracer_fd = -1;
2761 if (!is_root) {
2762 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2763 } else {
2764 return LTTNG_ERR_KERN_NA;
2765 }
2766 }
2767
2768
2769 /*
2770 * Copy consumer output from the tracing session to the domain session. The
2771 * function also applies the right modification on a per domain basis for the
2772 * trace files destination directory.
2773 *
2774 * Should *NOT* be called with RCU read-side lock held.
2775 */
2776 static int copy_session_consumer(int domain, struct ltt_session *session)
2777 {
2778 int ret;
2779 const char *dir_name;
2780 struct consumer_output *consumer;
2781
2782 assert(session);
2783 assert(session->consumer);
2784
2785 switch (domain) {
2786 case LTTNG_DOMAIN_KERNEL:
2787 DBG3("Copying tracing session consumer output in kernel session");
2788 /*
2789 * XXX: We should audit the session creation and what this function
2790 * does "extra" in order to avoid a destroy since this function is used
2791 * in the domain session creation (kernel and ust) only. Same for UST
2792 * domain.
2793 */
2794 if (session->kernel_session->consumer) {
2795 consumer_output_put(session->kernel_session->consumer);
2796 }
2797 session->kernel_session->consumer =
2798 consumer_copy_output(session->consumer);
2799 /* Ease our life a bit for the next part */
2800 consumer = session->kernel_session->consumer;
2801 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2802 break;
2803 case LTTNG_DOMAIN_JUL:
2804 case LTTNG_DOMAIN_LOG4J:
2805 case LTTNG_DOMAIN_PYTHON:
2806 case LTTNG_DOMAIN_UST:
2807 DBG3("Copying tracing session consumer output in UST session");
2808 if (session->ust_session->consumer) {
2809 consumer_output_put(session->ust_session->consumer);
2810 }
2811 session->ust_session->consumer =
2812 consumer_copy_output(session->consumer);
2813 /* Ease our life a bit for the next part */
2814 consumer = session->ust_session->consumer;
2815 dir_name = DEFAULT_UST_TRACE_DIR;
2816 break;
2817 default:
2818 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2819 goto error;
2820 }
2821
2822 /* Append correct directory to subdir */
2823 strncat(consumer->subdir, dir_name,
2824 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2825 DBG3("Copy session consumer subdir %s", consumer->subdir);
2826
2827 ret = LTTNG_OK;
2828
2829 error:
2830 return ret;
2831 }
2832
2833 /*
2834 * Create an UST session and add it to the session ust list.
2835 *
2836 * Should *NOT* be called with RCU read-side lock held.
2837 */
2838 static int create_ust_session(struct ltt_session *session,
2839 struct lttng_domain *domain)
2840 {
2841 int ret;
2842 struct ltt_ust_session *lus = NULL;
2843
2844 assert(session);
2845 assert(domain);
2846 assert(session->consumer);
2847
2848 switch (domain->type) {
2849 case LTTNG_DOMAIN_JUL:
2850 case LTTNG_DOMAIN_LOG4J:
2851 case LTTNG_DOMAIN_PYTHON:
2852 case LTTNG_DOMAIN_UST:
2853 break;
2854 default:
2855 ERR("Unknown UST domain on create session %d", domain->type);
2856 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2857 goto error;
2858 }
2859
2860 DBG("Creating UST session");
2861
2862 lus = trace_ust_create_session(session->id);
2863 if (lus == NULL) {
2864 ret = LTTNG_ERR_UST_SESS_FAIL;
2865 goto error;
2866 }
2867
2868 lus->uid = session->uid;
2869 lus->gid = session->gid;
2870 lus->output_traces = session->output_traces;
2871 lus->snapshot_mode = session->snapshot_mode;
2872 lus->live_timer_interval = session->live_timer;
2873 session->ust_session = lus;
2874 if (session->shm_path[0]) {
2875 strncpy(lus->root_shm_path, session->shm_path,
2876 sizeof(lus->root_shm_path));
2877 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2878 strncpy(lus->shm_path, session->shm_path,
2879 sizeof(lus->shm_path));
2880 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2881 strncat(lus->shm_path, "/ust",
2882 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2883 }
2884 /* Copy session output to the newly created UST session */
2885 ret = copy_session_consumer(domain->type, session);
2886 if (ret != LTTNG_OK) {
2887 goto error;
2888 }
2889
2890 return LTTNG_OK;
2891
2892 error:
2893 free(lus);
2894 session->ust_session = NULL;
2895 return ret;
2896 }
2897
2898 /*
2899 * Create a kernel tracer session then create the default channel.
2900 */
2901 static int create_kernel_session(struct ltt_session *session)
2902 {
2903 int ret;
2904
2905 DBG("Creating kernel session");
2906
2907 ret = kernel_create_session(session, kernel_tracer_fd);
2908 if (ret < 0) {
2909 ret = LTTNG_ERR_KERN_SESS_FAIL;
2910 goto error;
2911 }
2912
2913 /* Code flow safety */
2914 assert(session->kernel_session);
2915
2916 /* Copy session output to the newly created Kernel session */
2917 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2918 if (ret != LTTNG_OK) {
2919 goto error;
2920 }
2921
2922 /* Create directory(ies) on local filesystem. */
2923 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2924 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2925 ret = run_as_mkdir_recursive(
2926 session->kernel_session->consumer->dst.trace_path,
2927 S_IRWXU | S_IRWXG, session->uid, session->gid);
2928 if (ret < 0) {
2929 if (errno != EEXIST) {
2930 ERR("Trace directory creation error");
2931 goto error;
2932 }
2933 }
2934 }
2935
2936 session->kernel_session->uid = session->uid;
2937 session->kernel_session->gid = session->gid;
2938 session->kernel_session->output_traces = session->output_traces;
2939 session->kernel_session->snapshot_mode = session->snapshot_mode;
2940
2941 return LTTNG_OK;
2942
2943 error:
2944 trace_kernel_destroy_session(session->kernel_session);
2945 session->kernel_session = NULL;
2946 return ret;
2947 }
2948
2949 /*
2950 * Count number of session permitted by uid/gid.
2951 */
2952 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2953 {
2954 unsigned int i = 0;
2955 struct ltt_session *session;
2956
2957 DBG("Counting number of available session for UID %d GID %d",
2958 uid, gid);
2959 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2960 /*
2961 * Only list the sessions the user can control.
2962 */
2963 if (!session_access_ok(session, uid, gid)) {
2964 continue;
2965 }
2966 i++;
2967 }
2968 return i;
2969 }
2970
2971 /*
2972 * Process the command requested by the lttng client within the command
2973 * context structure. This function make sure that the return structure (llm)
2974 * is set and ready for transmission before returning.
2975 *
2976 * Return any error encountered or 0 for success.
2977 *
2978 * "sock" is only used for special-case var. len data.
2979 *
2980 * Should *NOT* be called with RCU read-side lock held.
2981 */
2982 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2983 int *sock_error)
2984 {
2985 int ret = LTTNG_OK;
2986 int need_tracing_session = 1;
2987 int need_domain;
2988
2989 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2990
2991 assert(!rcu_read_ongoing());
2992
2993 *sock_error = 0;
2994
2995 switch (cmd_ctx->lsm->cmd_type) {
2996 case LTTNG_CREATE_SESSION:
2997 case LTTNG_CREATE_SESSION_SNAPSHOT:
2998 case LTTNG_CREATE_SESSION_LIVE:
2999 case LTTNG_DESTROY_SESSION:
3000 case LTTNG_LIST_SESSIONS:
3001 case LTTNG_LIST_DOMAINS:
3002 case LTTNG_START_TRACE:
3003 case LTTNG_STOP_TRACE:
3004 case LTTNG_DATA_PENDING:
3005 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3006 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3007 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3008 case LTTNG_SNAPSHOT_RECORD:
3009 case LTTNG_SAVE_SESSION:
3010 case LTTNG_SET_SESSION_SHM_PATH:
3011 need_domain = 0;
3012 break;
3013 default:
3014 need_domain = 1;
3015 }
3016
3017 if (opt_no_kernel && need_domain
3018 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3019 if (!is_root) {
3020 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3021 } else {
3022 ret = LTTNG_ERR_KERN_NA;
3023 }
3024 goto error;
3025 }
3026
3027 /* Deny register consumer if we already have a spawned consumer. */
3028 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3029 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3030 if (kconsumer_data.pid > 0) {
3031 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3032 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3033 goto error;
3034 }
3035 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3036 }
3037
3038 /*
3039 * Check for command that don't needs to allocate a returned payload. We do
3040 * this here so we don't have to make the call for no payload at each
3041 * command.
3042 */
3043 switch(cmd_ctx->lsm->cmd_type) {
3044 case LTTNG_LIST_SESSIONS:
3045 case LTTNG_LIST_TRACEPOINTS:
3046 case LTTNG_LIST_TRACEPOINT_FIELDS:
3047 case LTTNG_LIST_DOMAINS:
3048 case LTTNG_LIST_CHANNELS:
3049 case LTTNG_LIST_EVENTS:
3050 case LTTNG_LIST_SYSCALLS:
3051 case LTTNG_LIST_TRACKER_PIDS:
3052 case LTTNG_DATA_PENDING:
3053 break;
3054 default:
3055 /* Setup lttng message with no payload */
3056 ret = setup_lttng_msg(cmd_ctx, 0);
3057 if (ret < 0) {
3058 /* This label does not try to unlock the session */
3059 goto init_setup_error;
3060 }
3061 }
3062
3063 /* Commands that DO NOT need a session. */
3064 switch (cmd_ctx->lsm->cmd_type) {
3065 case LTTNG_CREATE_SESSION:
3066 case LTTNG_CREATE_SESSION_SNAPSHOT:
3067 case LTTNG_CREATE_SESSION_LIVE:
3068 case LTTNG_CALIBRATE:
3069 case LTTNG_LIST_SESSIONS:
3070 case LTTNG_LIST_TRACEPOINTS:
3071 case LTTNG_LIST_SYSCALLS:
3072 case LTTNG_LIST_TRACEPOINT_FIELDS:
3073 case LTTNG_SAVE_SESSION:
3074 need_tracing_session = 0;
3075 break;
3076 default:
3077 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3078 /*
3079 * We keep the session list lock across _all_ commands
3080 * for now, because the per-session lock does not
3081 * handle teardown properly.
3082 */
3083 session_lock_list();
3084 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3085 if (cmd_ctx->session == NULL) {
3086 ret = LTTNG_ERR_SESS_NOT_FOUND;
3087 goto error;
3088 } else {
3089 /* Acquire lock for the session */
3090 session_lock(cmd_ctx->session);
3091 }
3092 break;
3093 }
3094
3095 /*
3096 * Commands that need a valid session but should NOT create one if none
3097 * exists. Instead of creating one and destroying it when the command is
3098 * handled, process that right before so we save some round trip in useless
3099 * code path.
3100 */
3101 switch (cmd_ctx->lsm->cmd_type) {
3102 case LTTNG_DISABLE_CHANNEL:
3103 case LTTNG_DISABLE_EVENT:
3104 switch (cmd_ctx->lsm->domain.type) {
3105 case LTTNG_DOMAIN_KERNEL:
3106 if (!cmd_ctx->session->kernel_session) {
3107 ret = LTTNG_ERR_NO_CHANNEL;
3108 goto error;
3109 }
3110 break;
3111 case LTTNG_DOMAIN_JUL:
3112 case LTTNG_DOMAIN_LOG4J:
3113 case LTTNG_DOMAIN_PYTHON:
3114 case LTTNG_DOMAIN_UST:
3115 if (!cmd_ctx->session->ust_session) {
3116 ret = LTTNG_ERR_NO_CHANNEL;
3117 goto error;
3118 }
3119 break;
3120 default:
3121 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3122 goto error;
3123 }
3124 default:
3125 break;
3126 }
3127
3128 if (!need_domain) {
3129 goto skip_domain;
3130 }
3131
3132 /*
3133 * Check domain type for specific "pre-action".
3134 */
3135 switch (cmd_ctx->lsm->domain.type) {
3136 case LTTNG_DOMAIN_KERNEL:
3137 if (!is_root) {
3138 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3139 goto error;
3140 }
3141
3142 /* Kernel tracer check */
3143 if (kernel_tracer_fd == -1) {
3144 /* Basically, load kernel tracer modules */
3145 ret = init_kernel_tracer();
3146 if (ret != 0) {
3147 goto error;
3148 }
3149 }
3150
3151 /* Consumer is in an ERROR state. Report back to client */
3152 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3153 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3154 goto error;
3155 }
3156
3157 /* Need a session for kernel command */
3158 if (need_tracing_session) {
3159 if (cmd_ctx->session->kernel_session == NULL) {
3160 ret = create_kernel_session(cmd_ctx->session);
3161 if (ret < 0) {
3162 ret = LTTNG_ERR_KERN_SESS_FAIL;
3163 goto error;
3164 }
3165 }
3166
3167 /* Start the kernel consumer daemon */
3168 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3169 if (kconsumer_data.pid == 0 &&
3170 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3171 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3172 ret = start_consumerd(&kconsumer_data);
3173 if (ret < 0) {
3174 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3175 goto error;
3176 }
3177 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3178 } else {
3179 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3180 }
3181
3182 /*
3183 * The consumer was just spawned so we need to add the socket to
3184 * the consumer output of the session if exist.
3185 */
3186 ret = consumer_create_socket(&kconsumer_data,
3187 cmd_ctx->session->kernel_session->consumer);
3188 if (ret < 0) {
3189 goto error;
3190 }
3191 }
3192
3193 break;
3194 case LTTNG_DOMAIN_JUL:
3195 case LTTNG_DOMAIN_LOG4J:
3196 case LTTNG_DOMAIN_PYTHON:
3197 case LTTNG_DOMAIN_UST:
3198 {
3199 if (!ust_app_supported()) {
3200 ret = LTTNG_ERR_NO_UST;
3201 goto error;
3202 }
3203 /* Consumer is in an ERROR state. Report back to client */
3204 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3205 ret = LTTNG_ERR_NO_USTCONSUMERD;
3206 goto error;
3207 }
3208
3209 if (need_tracing_session) {
3210 /* Create UST session if none exist. */
3211 if (cmd_ctx->session->ust_session == NULL) {
3212 ret = create_ust_session(cmd_ctx->session,
3213 &cmd_ctx->lsm->domain);
3214 if (ret != LTTNG_OK) {
3215 goto error;
3216 }
3217 }
3218
3219 /* Start the UST consumer daemons */
3220 /* 64-bit */
3221 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3222 if (consumerd64_bin[0] != '\0' &&
3223 ustconsumer64_data.pid == 0 &&
3224 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3225 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3226 ret = start_consumerd(&ustconsumer64_data);
3227 if (ret < 0) {
3228 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3229 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3230 goto error;
3231 }
3232
3233 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3234 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3235 } else {
3236 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3237 }
3238
3239 /*
3240 * Setup socket for consumer 64 bit. No need for atomic access
3241 * since it was set above and can ONLY be set in this thread.
3242 */
3243 ret = consumer_create_socket(&ustconsumer64_data,
3244 cmd_ctx->session->ust_session->consumer);
3245 if (ret < 0) {
3246 goto error;
3247 }
3248
3249 /* 32-bit */
3250 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3251 if (consumerd32_bin[0] != '\0' &&
3252 ustconsumer32_data.pid == 0 &&
3253 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3254 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3255 ret = start_consumerd(&ustconsumer32_data);
3256 if (ret < 0) {
3257 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3258 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3259 goto error;
3260 }
3261
3262 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3263 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3264 } else {
3265 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3266 }
3267
3268 /*
3269 * Setup socket for consumer 64 bit. No need for atomic access
3270 * since it was set above and can ONLY be set in this thread.
3271 */
3272 ret = consumer_create_socket(&ustconsumer32_data,
3273 cmd_ctx->session->ust_session->consumer);
3274 if (ret < 0) {
3275 goto error;
3276 }
3277 }
3278 break;
3279 }
3280 default:
3281 break;
3282 }
3283 skip_domain:
3284
3285 /* Validate consumer daemon state when start/stop trace command */
3286 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3287 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3288 switch (cmd_ctx->lsm->domain.type) {
3289 case LTTNG_DOMAIN_NONE:
3290 break;
3291 case LTTNG_DOMAIN_JUL:
3292 case LTTNG_DOMAIN_LOG4J:
3293 case LTTNG_DOMAIN_PYTHON:
3294 case LTTNG_DOMAIN_UST:
3295 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3296 ret = LTTNG_ERR_NO_USTCONSUMERD;
3297 goto error;
3298 }
3299 break;
3300 case LTTNG_DOMAIN_KERNEL:
3301 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3302 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3303 goto error;
3304 }
3305 break;
3306 default:
3307 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3308 goto error;
3309 }
3310 }
3311
3312 /*
3313 * Check that the UID or GID match that of the tracing session.
3314 * The root user can interact with all sessions.
3315 */
3316 if (need_tracing_session) {
3317 if (!session_access_ok(cmd_ctx->session,
3318 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3319 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3320 ret = LTTNG_ERR_EPERM;
3321 goto error;
3322 }
3323 }
3324
3325 /*
3326 * Send relayd information to consumer as soon as we have a domain and a
3327 * session defined.
3328 */
3329 if (cmd_ctx->session && need_domain) {
3330 /*
3331 * Setup relayd if not done yet. If the relayd information was already
3332 * sent to the consumer, this call will gracefully return.
3333 */
3334 ret = cmd_setup_relayd(cmd_ctx->session);
3335 if (ret != LTTNG_OK) {
3336 goto error;
3337 }
3338 }
3339
3340 /* Process by command type */
3341 switch (cmd_ctx->lsm->cmd_type) {
3342 case LTTNG_ADD_CONTEXT:
3343 {
3344 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3345 cmd_ctx->lsm->u.context.channel_name,
3346 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3347 break;
3348 }
3349 case LTTNG_DISABLE_CHANNEL:
3350 {
3351 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3352 cmd_ctx->lsm->u.disable.channel_name);
3353 break;
3354 }
3355 case LTTNG_DISABLE_EVENT:
3356 {
3357
3358 /*
3359 * FIXME: handle filter; for now we just receive the filter's
3360 * bytecode along with the filter expression which are sent by
3361 * liblttng-ctl and discard them.
3362 *
3363 * This fixes an issue where the client may block while sending
3364 * the filter payload and encounter an error because the session
3365 * daemon closes the socket without ever handling this data.
3366 */
3367 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3368 cmd_ctx->lsm->u.disable.bytecode_len;
3369
3370 if (count) {
3371 char data[LTTNG_FILTER_MAX_LEN];
3372
3373 DBG("Discarding disable event command payload of size %zu", count);
3374 while (count) {
3375 ret = lttcomm_recv_unix_sock(sock, data,
3376 count > sizeof(data) ? sizeof(data) : count);
3377 if (ret < 0) {
3378 goto error;
3379 }
3380
3381 count -= (size_t) ret;
3382 }
3383 }
3384 /* FIXME: passing packed structure to non-packed pointer */
3385 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3386 cmd_ctx->lsm->u.disable.channel_name,
3387 &cmd_ctx->lsm->u.disable.event);
3388 break;
3389 }
3390 case LTTNG_ENABLE_CHANNEL:
3391 {
3392 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3393 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3394 break;
3395 }
3396 case LTTNG_TRACK_PID:
3397 {
3398 ret = cmd_track_pid(cmd_ctx->session,
3399 cmd_ctx->lsm->domain.type,
3400 cmd_ctx->lsm->u.pid_tracker.pid);
3401 break;
3402 }
3403 case LTTNG_UNTRACK_PID:
3404 {
3405 ret = cmd_untrack_pid(cmd_ctx->session,
3406 cmd_ctx->lsm->domain.type,
3407 cmd_ctx->lsm->u.pid_tracker.pid);
3408 break;
3409 }
3410 case LTTNG_ENABLE_EVENT:
3411 {
3412 struct lttng_event_exclusion *exclusion = NULL;
3413 struct lttng_filter_bytecode *bytecode = NULL;
3414 char *filter_expression = NULL;
3415
3416 /* Handle exclusion events and receive it from the client. */
3417 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3418 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3419
3420 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3421 (count * LTTNG_SYMBOL_NAME_LEN));
3422 if (!exclusion) {
3423 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3424 goto error;
3425 }
3426
3427 DBG("Receiving var len exclusion event list from client ...");
3428 exclusion->count = count;
3429 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3430 count * LTTNG_SYMBOL_NAME_LEN);
3431 if (ret <= 0) {
3432 DBG("Nothing recv() from client var len data... continuing");
3433 *sock_error = 1;
3434 free(exclusion);
3435 ret = LTTNG_ERR_EXCLUSION_INVAL;
3436 goto error;
3437 }
3438 }
3439
3440 /* Get filter expression from client. */
3441 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3442 size_t expression_len =
3443 cmd_ctx->lsm->u.enable.expression_len;
3444
3445 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3446 ret = LTTNG_ERR_FILTER_INVAL;
3447 free(exclusion);
3448 goto error;
3449 }
3450
3451 filter_expression = zmalloc(expression_len);
3452 if (!filter_expression) {
3453 free(exclusion);
3454 ret = LTTNG_ERR_FILTER_NOMEM;
3455 goto error;
3456 }
3457
3458 /* Receive var. len. data */
3459 DBG("Receiving var len filter's expression from client ...");
3460 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3461 expression_len);
3462 if (ret <= 0) {
3463 DBG("Nothing recv() from client car len data... continuing");
3464 *sock_error = 1;
3465 free(filter_expression);
3466 free(exclusion);
3467 ret = LTTNG_ERR_FILTER_INVAL;
3468 goto error;
3469 }
3470 }
3471
3472 /* Handle filter and get bytecode from client. */
3473 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3474 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3475
3476 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3477 ret = LTTNG_ERR_FILTER_INVAL;
3478 free(filter_expression);
3479 free(exclusion);
3480 goto error;
3481 }
3482
3483 bytecode = zmalloc(bytecode_len);
3484 if (!bytecode) {
3485 free(filter_expression);
3486 free(exclusion);
3487 ret = LTTNG_ERR_FILTER_NOMEM;
3488 goto error;
3489 }
3490
3491 /* Receive var. len. data */
3492 DBG("Receiving var len filter's bytecode from client ...");
3493 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3494 if (ret <= 0) {
3495 DBG("Nothing recv() from client car len data... continuing");
3496 *sock_error = 1;
3497 free(filter_expression);
3498 free(bytecode);
3499 free(exclusion);
3500 ret = LTTNG_ERR_FILTER_INVAL;
3501 goto error;
3502 }
3503
3504 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3505 free(filter_expression);
3506 free(bytecode);
3507 free(exclusion);
3508 ret = LTTNG_ERR_FILTER_INVAL;
3509 goto error;
3510 }
3511 }
3512
3513 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3514 cmd_ctx->lsm->u.enable.channel_name,
3515 &cmd_ctx->lsm->u.enable.event,
3516 filter_expression, bytecode, exclusion,
3517 kernel_poll_pipe[1]);
3518 break;
3519 }
3520 case LTTNG_LIST_TRACEPOINTS:
3521 {
3522 struct lttng_event *events;
3523 ssize_t nb_events;
3524
3525 session_lock_list();
3526 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3527 session_unlock_list();
3528 if (nb_events < 0) {
3529 /* Return value is a negative lttng_error_code. */
3530 ret = -nb_events;
3531 goto error;
3532 }
3533
3534 /*
3535 * Setup lttng message with payload size set to the event list size in
3536 * bytes and then copy list into the llm payload.
3537 */
3538 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3539 if (ret < 0) {
3540 free(events);
3541 goto setup_error;
3542 }
3543
3544 /* Copy event list into message payload */
3545 memcpy(cmd_ctx->llm->payload, events,
3546 sizeof(struct lttng_event) * nb_events);
3547
3548 free(events);
3549
3550 ret = LTTNG_OK;
3551 break;
3552 }
3553 case LTTNG_LIST_TRACEPOINT_FIELDS:
3554 {
3555 struct lttng_event_field *fields;
3556 ssize_t nb_fields;
3557
3558 session_lock_list();
3559 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3560 &fields);
3561 session_unlock_list();
3562 if (nb_fields < 0) {
3563 /* Return value is a negative lttng_error_code. */
3564 ret = -nb_fields;
3565 goto error;
3566 }
3567
3568 /*
3569 * Setup lttng message with payload size set to the event list size in
3570 * bytes and then copy list into the llm payload.
3571 */
3572 ret = setup_lttng_msg(cmd_ctx,
3573 sizeof(struct lttng_event_field) * nb_fields);
3574 if (ret < 0) {
3575 free(fields);
3576 goto setup_error;
3577 }
3578
3579 /* Copy event list into message payload */
3580 memcpy(cmd_ctx->llm->payload, fields,
3581 sizeof(struct lttng_event_field) * nb_fields);
3582
3583 free(fields);
3584
3585 ret = LTTNG_OK;
3586 break;
3587 }
3588 case LTTNG_LIST_SYSCALLS:
3589 {
3590 struct lttng_event *events;
3591 ssize_t nb_events;
3592
3593 nb_events = cmd_list_syscalls(&events);
3594 if (nb_events < 0) {
3595 /* Return value is a negative lttng_error_code. */
3596 ret = -nb_events;
3597 goto error;
3598 }
3599
3600 /*
3601 * Setup lttng message with payload size set to the event list size in
3602 * bytes and then copy list into the llm payload.
3603 */
3604 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3605 if (ret < 0) {
3606 free(events);
3607 goto setup_error;
3608 }
3609
3610 /* Copy event list into message payload */
3611 memcpy(cmd_ctx->llm->payload, events,
3612 sizeof(struct lttng_event) * nb_events);
3613
3614 free(events);
3615
3616 ret = LTTNG_OK;
3617 break;
3618 }
3619 case LTTNG_LIST_TRACKER_PIDS:
3620 {
3621 int32_t *pids = NULL;
3622 ssize_t nr_pids;
3623
3624 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3625 cmd_ctx->lsm->domain.type, &pids);
3626 if (nr_pids < 0) {
3627 /* Return value is a negative lttng_error_code. */
3628 ret = -nr_pids;
3629 goto error;
3630 }
3631
3632 /*
3633 * Setup lttng message with payload size set to the event list size in
3634 * bytes and then copy list into the llm payload.
3635 */
3636 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3637 if (ret < 0) {
3638 free(pids);
3639 goto setup_error;
3640 }
3641
3642 /* Copy event list into message payload */
3643 memcpy(cmd_ctx->llm->payload, pids,
3644 sizeof(int) * nr_pids);
3645
3646 free(pids);
3647
3648 ret = LTTNG_OK;
3649 break;
3650 }
3651 case LTTNG_SET_CONSUMER_URI:
3652 {
3653 size_t nb_uri, len;
3654 struct lttng_uri *uris;
3655
3656 nb_uri = cmd_ctx->lsm->u.uri.size;
3657 len = nb_uri * sizeof(struct lttng_uri);
3658
3659 if (nb_uri == 0) {
3660 ret = LTTNG_ERR_INVALID;
3661 goto error;
3662 }
3663
3664 uris = zmalloc(len);
3665 if (uris == NULL) {
3666 ret = LTTNG_ERR_FATAL;
3667 goto error;
3668 }
3669
3670 /* Receive variable len data */
3671 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3672 ret = lttcomm_recv_unix_sock(sock, uris, len);
3673 if (ret <= 0) {
3674 DBG("No URIs received from client... continuing");
3675 *sock_error = 1;
3676 ret = LTTNG_ERR_SESSION_FAIL;
3677 free(uris);
3678 goto error;
3679 }
3680
3681 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3682 free(uris);
3683 if (ret != LTTNG_OK) {
3684 goto error;
3685 }
3686
3687
3688 break;
3689 }
3690 case LTTNG_START_TRACE:
3691 {
3692 ret = cmd_start_trace(cmd_ctx->session);
3693 break;
3694 }
3695 case LTTNG_STOP_TRACE:
3696 {
3697 ret = cmd_stop_trace(cmd_ctx->session);
3698 break;
3699 }
3700 case LTTNG_CREATE_SESSION:
3701 {
3702 size_t nb_uri, len;
3703 struct lttng_uri *uris = NULL;
3704
3705 nb_uri = cmd_ctx->lsm->u.uri.size;
3706 len = nb_uri * sizeof(struct lttng_uri);
3707
3708 if (nb_uri > 0) {
3709 uris = zmalloc(len);
3710 if (uris == NULL) {
3711 ret = LTTNG_ERR_FATAL;
3712 goto error;
3713 }
3714
3715 /* Receive variable len data */
3716 DBG("Waiting for %zu URIs from client ...", nb_uri);
3717 ret = lttcomm_recv_unix_sock(sock, uris, len);
3718 if (ret <= 0) {
3719 DBG("No URIs received from client... continuing");
3720 *sock_error = 1;
3721 ret = LTTNG_ERR_SESSION_FAIL;
3722 free(uris);
3723 goto error;
3724 }
3725
3726 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3727 DBG("Creating session with ONE network URI is a bad call");
3728 ret = LTTNG_ERR_SESSION_FAIL;
3729 free(uris);
3730 goto error;
3731 }
3732 }
3733
3734 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3735 &cmd_ctx->creds, 0);
3736
3737 free(uris);
3738
3739 break;
3740 }
3741 case LTTNG_DESTROY_SESSION:
3742 {
3743 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3744
3745 /* Set session to NULL so we do not unlock it after free. */
3746 cmd_ctx->session = NULL;
3747 break;
3748 }
3749 case LTTNG_LIST_DOMAINS:
3750 {
3751 ssize_t nb_dom;
3752 struct lttng_domain *domains = NULL;
3753
3754 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3755 if (nb_dom < 0) {
3756 /* Return value is a negative lttng_error_code. */
3757 ret = -nb_dom;
3758 goto error;
3759 }
3760
3761 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3762 if (ret < 0) {
3763 free(domains);
3764 goto setup_error;
3765 }
3766
3767 /* Copy event list into message payload */
3768 memcpy(cmd_ctx->llm->payload, domains,
3769 nb_dom * sizeof(struct lttng_domain));
3770
3771 free(domains);
3772
3773 ret = LTTNG_OK;
3774 break;
3775 }
3776 case LTTNG_LIST_CHANNELS:
3777 {
3778 int nb_chan;
3779 struct lttng_channel *channels = NULL;
3780
3781 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3782 cmd_ctx->session, &channels);
3783 if (nb_chan < 0) {
3784 /* Return value is a negative lttng_error_code. */
3785 ret = -nb_chan;
3786 goto error;
3787 }
3788
3789 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3790 if (ret < 0) {
3791 free(channels);
3792 goto setup_error;
3793 }
3794
3795 /* Copy event list into message payload */
3796 memcpy(cmd_ctx->llm->payload, channels,
3797 nb_chan * sizeof(struct lttng_channel));
3798
3799 free(channels);
3800
3801 ret = LTTNG_OK;
3802 break;
3803 }
3804 case LTTNG_LIST_EVENTS:
3805 {
3806 ssize_t nb_event;
3807 struct lttng_event *events = NULL;
3808
3809 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3810 cmd_ctx->lsm->u.list.channel_name, &events);
3811 if (nb_event < 0) {
3812 /* Return value is a negative lttng_error_code. */
3813 ret = -nb_event;
3814 goto error;
3815 }
3816
3817 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3818 if (ret < 0) {
3819 free(events);
3820 goto setup_error;
3821 }
3822
3823 /* Copy event list into message payload */
3824 memcpy(cmd_ctx->llm->payload, events,
3825 nb_event * sizeof(struct lttng_event));
3826
3827 free(events);
3828
3829 ret = LTTNG_OK;
3830 break;
3831 }
3832 case LTTNG_LIST_SESSIONS:
3833 {
3834 unsigned int nr_sessions;
3835
3836 session_lock_list();
3837 nr_sessions = lttng_sessions_count(
3838 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3839 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3840
3841 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3842 if (ret < 0) {
3843 session_unlock_list();
3844 goto setup_error;
3845 }
3846
3847 /* Filled the session array */
3848 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3849 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3850 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3851
3852 session_unlock_list();
3853
3854 ret = LTTNG_OK;
3855 break;
3856 }
3857 case LTTNG_CALIBRATE:
3858 {
3859 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3860 &cmd_ctx->lsm->u.calibrate);
3861 break;
3862 }
3863 case LTTNG_REGISTER_CONSUMER:
3864 {
3865 struct consumer_data *cdata;
3866
3867 switch (cmd_ctx->lsm->domain.type) {
3868 case LTTNG_DOMAIN_KERNEL:
3869 cdata = &kconsumer_data;
3870 break;
3871 default:
3872 ret = LTTNG_ERR_UND;
3873 goto error;
3874 }
3875
3876 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3877 cmd_ctx->lsm->u.reg.path, cdata);
3878 break;
3879 }
3880 case LTTNG_DATA_PENDING:
3881 {
3882 ret = cmd_data_pending(cmd_ctx->session);
3883 break;
3884 }
3885 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3886 {
3887 struct lttcomm_lttng_output_id reply;
3888
3889 ret = cmd_snapshot_add_output(cmd_ctx->session,
3890 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3891 if (ret != LTTNG_OK) {
3892 goto error;
3893 }
3894
3895 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3896 if (ret < 0) {
3897 goto setup_error;
3898 }
3899
3900 /* Copy output list into message payload */
3901 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3902 ret = LTTNG_OK;
3903 break;
3904 }
3905 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3906 {
3907 ret = cmd_snapshot_del_output(cmd_ctx->session,
3908 &cmd_ctx->lsm->u.snapshot_output.output);
3909 break;
3910 }
3911 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3912 {
3913 ssize_t nb_output;
3914 struct lttng_snapshot_output *outputs = NULL;
3915
3916 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3917 if (nb_output < 0) {
3918 ret = -nb_output;
3919 goto error;
3920 }
3921
3922 ret = setup_lttng_msg(cmd_ctx,
3923 nb_output * sizeof(struct lttng_snapshot_output));
3924 if (ret < 0) {
3925 free(outputs);
3926 goto setup_error;
3927 }
3928
3929 if (outputs) {
3930 /* Copy output list into message payload */
3931 memcpy(cmd_ctx->llm->payload, outputs,
3932 nb_output * sizeof(struct lttng_snapshot_output));
3933 free(outputs);
3934 }
3935
3936 ret = LTTNG_OK;
3937 break;
3938 }
3939 case LTTNG_SNAPSHOT_RECORD:
3940 {
3941 ret = cmd_snapshot_record(cmd_ctx->session,
3942 &cmd_ctx->lsm->u.snapshot_record.output,
3943 cmd_ctx->lsm->u.snapshot_record.wait);
3944 break;
3945 }
3946 case LTTNG_CREATE_SESSION_SNAPSHOT:
3947 {
3948 size_t nb_uri, len;
3949 struct lttng_uri *uris = NULL;
3950
3951 nb_uri = cmd_ctx->lsm->u.uri.size;
3952 len = nb_uri * sizeof(struct lttng_uri);
3953
3954 if (nb_uri > 0) {
3955 uris = zmalloc(len);
3956 if (uris == NULL) {
3957 ret = LTTNG_ERR_FATAL;
3958 goto error;
3959 }
3960
3961 /* Receive variable len data */
3962 DBG("Waiting for %zu URIs from client ...", nb_uri);
3963 ret = lttcomm_recv_unix_sock(sock, uris, len);
3964 if (ret <= 0) {
3965 DBG("No URIs received from client... continuing");
3966 *sock_error = 1;
3967 ret = LTTNG_ERR_SESSION_FAIL;
3968 free(uris);
3969 goto error;
3970 }
3971
3972 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3973 DBG("Creating session with ONE network URI is a bad call");
3974 ret = LTTNG_ERR_SESSION_FAIL;
3975 free(uris);
3976 goto error;
3977 }
3978 }
3979
3980 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3981 nb_uri, &cmd_ctx->creds);
3982 free(uris);
3983 break;
3984 }
3985 case LTTNG_CREATE_SESSION_LIVE:
3986 {
3987 size_t nb_uri, len;
3988 struct lttng_uri *uris = NULL;
3989
3990 nb_uri = cmd_ctx->lsm->u.uri.size;
3991 len = nb_uri * sizeof(struct lttng_uri);
3992
3993 if (nb_uri > 0) {
3994 uris = zmalloc(len);
3995 if (uris == NULL) {
3996 ret = LTTNG_ERR_FATAL;
3997 goto error;
3998 }
3999
4000 /* Receive variable len data */
4001 DBG("Waiting for %zu URIs from client ...", nb_uri);
4002 ret = lttcomm_recv_unix_sock(sock, uris, len);
4003 if (ret <= 0) {
4004 DBG("No URIs received from client... continuing");
4005 *sock_error = 1;
4006 ret = LTTNG_ERR_SESSION_FAIL;
4007 free(uris);
4008 goto error;
4009 }
4010
4011 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4012 DBG("Creating session with ONE network URI is a bad call");
4013 ret = LTTNG_ERR_SESSION_FAIL;
4014 free(uris);
4015 goto error;
4016 }
4017 }
4018
4019 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4020 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4021 free(uris);
4022 break;
4023 }
4024 case LTTNG_SAVE_SESSION:
4025 {
4026 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4027 &cmd_ctx->creds);
4028 break;
4029 }
4030 case LTTNG_SET_SESSION_SHM_PATH:
4031 {
4032 ret = cmd_set_session_shm_path(cmd_ctx->session,
4033 cmd_ctx->lsm->u.set_shm_path.shm_path);
4034 break;
4035 }
4036 default:
4037 ret = LTTNG_ERR_UND;
4038 break;
4039 }
4040
4041 error:
4042 if (cmd_ctx->llm == NULL) {
4043 DBG("Missing llm structure. Allocating one.");
4044 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
4045 goto setup_error;
4046 }
4047 }
4048 /* Set return code */
4049 cmd_ctx->llm->ret_code = ret;
4050 setup_error:
4051 if (cmd_ctx->session) {
4052 session_unlock(cmd_ctx->session);
4053 }
4054 if (need_tracing_session) {
4055 session_unlock_list();
4056 }
4057 init_setup_error:
4058 assert(!rcu_read_ongoing());
4059 return ret;
4060 }
4061
4062 /*
4063 * Thread managing health check socket.
4064 */
4065 static void *thread_manage_health(void *data)
4066 {
4067 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4068 uint32_t revents, nb_fd;
4069 struct lttng_poll_event events;
4070 struct health_comm_msg msg;
4071 struct health_comm_reply reply;
4072
4073 DBG("[thread] Manage health check started");
4074
4075 rcu_register_thread();
4076
4077 /* We might hit an error path before this is created. */
4078 lttng_poll_init(&events);
4079
4080 /* Create unix socket */
4081 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4082 if (sock < 0) {
4083 ERR("Unable to create health check Unix socket");
4084 ret = -1;
4085 goto error;
4086 }
4087
4088 if (is_root) {
4089 /* lttng health client socket path permissions */
4090 ret = chown(health_unix_sock_path, 0,
4091 utils_get_group_id(tracing_group_name));
4092 if (ret < 0) {
4093 ERR("Unable to set group on %s", health_unix_sock_path);
4094 PERROR("chown");
4095 ret = -1;
4096 goto error;
4097 }
4098
4099 ret = chmod(health_unix_sock_path,
4100 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4101 if (ret < 0) {
4102 ERR("Unable to set permissions on %s", health_unix_sock_path);
4103 PERROR("chmod");
4104 ret = -1;
4105 goto error;
4106 }
4107 }
4108
4109 /*
4110 * Set the CLOEXEC flag. Return code is useless because either way, the
4111 * show must go on.
4112 */
4113 (void) utils_set_fd_cloexec(sock);
4114
4115 ret = lttcomm_listen_unix_sock(sock);
4116 if (ret < 0) {
4117 goto error;
4118 }
4119
4120 /*
4121 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4122 * more will be added to this poll set.
4123 */
4124 ret = sessiond_set_thread_pollset(&events, 2);
4125 if (ret < 0) {
4126 goto error;
4127 }
4128
4129 /* Add the application registration socket */
4130 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4131 if (ret < 0) {
4132 goto error;
4133 }
4134
4135 sessiond_notify_ready();
4136
4137 while (1) {
4138 DBG("Health check ready");
4139
4140 /* Inifinite blocking call, waiting for transmission */
4141 restart:
4142 ret = lttng_poll_wait(&events, -1);
4143 if (ret < 0) {
4144 /*
4145 * Restart interrupted system call.
4146 */
4147 if (errno == EINTR) {
4148 goto restart;
4149 }
4150 goto error;
4151 }
4152
4153 nb_fd = ret;
4154
4155 for (i = 0; i < nb_fd; i++) {
4156 /* Fetch once the poll data */
4157 revents = LTTNG_POLL_GETEV(&events, i);
4158 pollfd = LTTNG_POLL_GETFD(&events, i);
4159
4160 if (!revents) {
4161 /* No activity for this FD (poll implementation). */
4162 continue;
4163 }
4164
4165 /* Thread quit pipe has been closed. Killing thread. */
4166 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4167 if (ret) {
4168 err = 0;
4169 goto exit;
4170 }
4171
4172 /* Event on the registration socket */
4173 if (pollfd == sock) {
4174 if (revents & LPOLLIN) {
4175 continue;
4176 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4177 ERR("Health socket poll error");
4178 goto error;
4179 } else {
4180 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4181 goto error;
4182 }
4183 }
4184 }
4185
4186 new_sock = lttcomm_accept_unix_sock(sock);
4187 if (new_sock < 0) {
4188 goto error;
4189 }
4190
4191 /*
4192 * Set the CLOEXEC flag. Return code is useless because either way, the
4193 * show must go on.
4194 */
4195 (void) utils_set_fd_cloexec(new_sock);
4196
4197 DBG("Receiving data from client for health...");
4198 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4199 if (ret <= 0) {
4200 DBG("Nothing recv() from client... continuing");
4201 ret = close(new_sock);
4202 if (ret) {
4203 PERROR("close");
4204 }
4205 new_sock = -1;
4206 continue;
4207 }
4208
4209 rcu_thread_online();
4210
4211 memset(&reply, 0, sizeof(reply));
4212 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4213 /*
4214 * health_check_state returns 0 if health is
4215 * bad.
4216 */
4217 if (!health_check_state(health_sessiond, i)) {
4218 reply.ret_code |= 1ULL << i;
4219 }
4220 }
4221
4222 DBG2("Health check return value %" PRIx64, reply.ret_code);
4223
4224 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4225 if (ret < 0) {
4226 ERR("Failed to send health data back to client");
4227 }
4228
4229 /* End of transmission */
4230 ret = close(new_sock);
4231 if (ret) {
4232 PERROR("close");
4233 }
4234 new_sock = -1;
4235 }
4236
4237 exit:
4238 error:
4239 if (err) {
4240 ERR("Health error occurred in %s", __func__);
4241 }
4242 DBG("Health check thread dying");
4243 unlink(health_unix_sock_path);
4244 if (sock >= 0) {
4245 ret = close(sock);
4246 if (ret) {
4247 PERROR("close");
4248 }
4249 }
4250
4251 lttng_poll_clean(&events);
4252
4253 rcu_unregister_thread();
4254 return NULL;
4255 }
4256
4257 /*
4258 * This thread manage all clients request using the unix client socket for
4259 * communication.
4260 */
4261 static void *thread_manage_clients(void *data)
4262 {
4263 int sock = -1, ret, i, pollfd, err = -1;
4264 int sock_error;
4265 uint32_t revents, nb_fd;
4266 struct command_ctx *cmd_ctx = NULL;
4267 struct lttng_poll_event events;
4268
4269 DBG("[thread] Manage client started");
4270
4271 rcu_register_thread();
4272
4273 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4274
4275 health_code_update();
4276
4277 ret = lttcomm_listen_unix_sock(client_sock);
4278 if (ret < 0) {
4279 goto error_listen;
4280 }
4281
4282 /*
4283 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4284 * more will be added to this poll set.
4285 */
4286 ret = sessiond_set_thread_pollset(&events, 2);
4287 if (ret < 0) {
4288 goto error_create_poll;
4289 }
4290
4291 /* Add the application registration socket */
4292 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4293 if (ret < 0) {
4294 goto error;
4295 }
4296
4297 sessiond_notify_ready();
4298 ret = sem_post(&load_info->message_thread_ready);
4299 if (ret) {
4300 PERROR("sem_post message_thread_ready");
4301 goto error;
4302 }
4303
4304 /* This testpoint is after we signal readiness to the parent. */
4305 if (testpoint(sessiond_thread_manage_clients)) {
4306 goto error;
4307 }
4308
4309 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4310 goto error;
4311 }
4312
4313 health_code_update();
4314
4315 while (1) {
4316 DBG("Accepting client command ...");
4317
4318 /* Inifinite blocking call, waiting for transmission */
4319 restart:
4320 health_poll_entry();
4321 ret = lttng_poll_wait(&events, -1);
4322 health_poll_exit();
4323 if (ret < 0) {
4324 /*
4325 * Restart interrupted system call.
4326 */
4327 if (errno == EINTR) {
4328 goto restart;
4329 }
4330 goto error;
4331 }
4332
4333 nb_fd = ret;
4334
4335 for (i = 0; i < nb_fd; i++) {
4336 /* Fetch once the poll data */
4337 revents = LTTNG_POLL_GETEV(&events, i);
4338 pollfd = LTTNG_POLL_GETFD(&events, i);
4339
4340 health_code_update();
4341
4342 if (!revents) {
4343 /* No activity for this FD (poll implementation). */
4344 continue;
4345 }
4346
4347 /* Thread quit pipe has been closed. Killing thread. */
4348 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4349 if (ret) {
4350 err = 0;
4351 goto exit;
4352 }
4353
4354 /* Event on the registration socket */
4355 if (pollfd == client_sock) {
4356 if (revents & LPOLLIN) {
4357 continue;
4358 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4359 ERR("Client socket poll error");
4360 goto error;
4361 } else {
4362 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4363 goto error;
4364 }
4365 }
4366 }
4367
4368 DBG("Wait for client response");
4369
4370 health_code_update();
4371
4372 sock = lttcomm_accept_unix_sock(client_sock);
4373 if (sock < 0) {
4374 goto error;
4375 }
4376
4377 /*
4378 * Set the CLOEXEC flag. Return code is useless because either way, the
4379 * show must go on.
4380 */
4381 (void) utils_set_fd_cloexec(sock);
4382
4383 /* Set socket option for credentials retrieval */
4384 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4385 if (ret < 0) {
4386 goto error;
4387 }
4388
4389 /* Allocate context command to process the client request */
4390 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4391 if (cmd_ctx == NULL) {
4392 PERROR("zmalloc cmd_ctx");
4393 goto error;
4394 }
4395
4396 /* Allocate data buffer for reception */
4397 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4398 if (cmd_ctx->lsm == NULL) {
4399 PERROR("zmalloc cmd_ctx->lsm");
4400 goto error;
4401 }
4402
4403 cmd_ctx->llm = NULL;
4404 cmd_ctx->session = NULL;
4405
4406 health_code_update();
4407
4408 /*
4409 * Data is received from the lttng client. The struct
4410 * lttcomm_session_msg (lsm) contains the command and data request of
4411 * the client.
4412 */
4413 DBG("Receiving data from client ...");
4414 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4415 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4416 if (ret <= 0) {
4417 DBG("Nothing recv() from client... continuing");
4418 ret = close(sock);
4419 if (ret) {
4420 PERROR("close");
4421 }
4422 sock = -1;
4423 clean_command_ctx(&cmd_ctx);
4424 continue;
4425 }
4426
4427 health_code_update();
4428
4429 // TODO: Validate cmd_ctx including sanity check for
4430 // security purpose.
4431
4432 rcu_thread_online();
4433 /*
4434 * This function dispatch the work to the kernel or userspace tracer
4435 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4436 * informations for the client. The command context struct contains
4437 * everything this function may needs.
4438 */
4439 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4440 rcu_thread_offline();
4441 if (ret < 0) {
4442 ret = close(sock);
4443 if (ret) {
4444 PERROR("close");
4445 }
4446 sock = -1;
4447 /*
4448 * TODO: Inform client somehow of the fatal error. At
4449 * this point, ret < 0 means that a zmalloc failed
4450 * (ENOMEM). Error detected but still accept
4451 * command, unless a socket error has been
4452 * detected.
4453 */
4454 clean_command_ctx(&cmd_ctx);
4455 continue;
4456 }
4457
4458 health_code_update();
4459
4460 DBG("Sending response (size: %d, retcode: %s)",
4461 cmd_ctx->lttng_msg_size,
4462 lttng_strerror(-cmd_ctx->llm->ret_code));
4463 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4464 if (ret < 0) {
4465 ERR("Failed to send data back to client");
4466 }
4467
4468 /* End of transmission */
4469 ret = close(sock);
4470 if (ret) {
4471 PERROR("close");
4472 }
4473 sock = -1;
4474
4475 clean_command_ctx(&cmd_ctx);
4476
4477 health_code_update();
4478 }
4479
4480 exit:
4481 error:
4482 if (sock >= 0) {
4483 ret = close(sock);
4484 if (ret) {
4485 PERROR("close");
4486 }
4487 }
4488
4489 lttng_poll_clean(&events);
4490 clean_command_ctx(&cmd_ctx);
4491
4492 error_listen:
4493 error_create_poll:
4494 unlink(client_unix_sock_path);
4495 if (client_sock >= 0) {
4496 ret = close(client_sock);
4497 if (ret) {
4498 PERROR("close");
4499 }
4500 }
4501
4502 if (err) {
4503 health_error();
4504 ERR("Health error occurred in %s", __func__);
4505 }
4506
4507 health_unregister(health_sessiond);
4508
4509 DBG("Client thread dying");
4510
4511 rcu_unregister_thread();
4512
4513 /*
4514 * Since we are creating the consumer threads, we own them, so we need
4515 * to join them before our thread exits.
4516 */
4517 ret = join_consumer_thread(&kconsumer_data);
4518 if (ret) {
4519 errno = ret;
4520 PERROR("join_consumer");
4521 }
4522
4523 ret = join_consumer_thread(&ustconsumer32_data);
4524 if (ret) {
4525 errno = ret;
4526 PERROR("join_consumer ust32");
4527 }
4528
4529 ret = join_consumer_thread(&ustconsumer64_data);
4530 if (ret) {
4531 errno = ret;
4532 PERROR("join_consumer ust64");
4533 }
4534 return NULL;
4535 }
4536
4537
4538 /*
4539 * usage function on stderr
4540 */
4541 static void usage(void)
4542 {
4543 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4544 fprintf(stderr, " -h, --help Display this usage.\n");
4545 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4546 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4547 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4548 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4549 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4550 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4551 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4552 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4553 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4554 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4555 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4556 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4557 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4558 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4559 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4560 fprintf(stderr, " -V, --version Show version number.\n");
4561 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4562 fprintf(stderr, " -q, --quiet No output at all.\n");
4563 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4564 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4565 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4566 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4567 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4568 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4569 fprintf(stderr, " -l --load PATH Load session configuration\n");
4570 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4571 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4572 }
4573
4574 static int string_match(const char *str1, const char *str2)
4575 {
4576 return (str1 && str2) && !strcmp(str1, str2);
4577 }
4578
4579 /*
4580 * Take an option from the getopt output and set it in the right variable to be
4581 * used later.
4582 *
4583 * Return 0 on success else a negative value.
4584 */
4585 static int set_option(int opt, const char *arg, const char *optname)
4586 {
4587 int ret = 0;
4588
4589 if (string_match(optname, "client-sock") || opt == 'c') {
4590 if (!arg || *arg == '\0') {
4591 ret = -EINVAL;
4592 goto end;
4593 }
4594 if (lttng_is_setuid_setgid()) {
4595 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4596 "-c, --client-sock");
4597 } else {
4598 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4599 }
4600 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4601 if (!arg || *arg == '\0') {
4602 ret = -EINVAL;
4603 goto end;
4604 }
4605 if (lttng_is_setuid_setgid()) {
4606 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4607 "-a, --apps-sock");
4608 } else {
4609 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4610 }
4611 } else if (string_match(optname, "daemonize") || opt == 'd') {
4612 opt_daemon = 1;
4613 } else if (string_match(optname, "background") || opt == 'b') {
4614 opt_background = 1;
4615 } else if (string_match(optname, "group") || opt == 'g') {
4616 if (!arg || *arg == '\0') {
4617 ret = -EINVAL;
4618 goto end;
4619 }
4620 if (lttng_is_setuid_setgid()) {
4621 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4622 "-g, --group");
4623 } else {
4624 /*
4625 * If the override option is set, the pointer points to a
4626 * *non* const thus freeing it even though the variable type is
4627 * set to const.
4628 */
4629 if (tracing_group_name_override) {
4630 free((void *) tracing_group_name);
4631 }
4632 tracing_group_name = strdup(arg);
4633 if (!tracing_group_name) {
4634 PERROR("strdup");
4635 ret = -ENOMEM;
4636 }
4637 tracing_group_name_override = 1;
4638 }
4639 } else if (string_match(optname, "help") || opt == 'h') {
4640 usage();
4641 exit(EXIT_SUCCESS);
4642 } else if (string_match(optname, "version") || opt == 'V') {
4643 fprintf(stdout, "%s\n", VERSION);
4644 exit(EXIT_SUCCESS);
4645 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4646 opt_sig_parent = 1;
4647 } else if (string_match(optname, "kconsumerd-err-sock")) {
4648 if (!arg || *arg == '\0') {
4649 ret = -EINVAL;
4650 goto end;
4651 }
4652 if (lttng_is_setuid_setgid()) {
4653 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4654 "--kconsumerd-err-sock");
4655 } else {
4656 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4657 }
4658 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4659 if (!arg || *arg == '\0') {
4660 ret = -EINVAL;
4661 goto end;
4662 }
4663 if (lttng_is_setuid_setgid()) {
4664 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4665 "--kconsumerd-cmd-sock");
4666 } else {
4667 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4668 }
4669 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4670 if (!arg || *arg == '\0') {
4671 ret = -EINVAL;
4672 goto end;
4673 }
4674 if (lttng_is_setuid_setgid()) {
4675 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4676 "--ustconsumerd64-err-sock");
4677 } else {
4678 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4679 }
4680 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4681 if (!arg || *arg == '\0') {
4682 ret = -EINVAL;
4683 goto end;
4684 }
4685 if (lttng_is_setuid_setgid()) {
4686 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4687 "--ustconsumerd64-cmd-sock");
4688 } else {
4689 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4690 }
4691 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4692 if (!arg || *arg == '\0') {
4693 ret = -EINVAL;
4694 goto end;
4695 }
4696 if (lttng_is_setuid_setgid()) {
4697 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4698 "--ustconsumerd32-err-sock");
4699 } else {
4700 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4701 }
4702 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4703 if (!arg || *arg == '\0') {
4704 ret = -EINVAL;
4705 goto end;
4706 }
4707 if (lttng_is_setuid_setgid()) {
4708 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4709 "--ustconsumerd32-cmd-sock");
4710 } else {
4711 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4712 }
4713 } else if (string_match(optname, "no-kernel")) {
4714 opt_no_kernel = 1;
4715 } else if (string_match(optname, "quiet") || opt == 'q') {
4716 lttng_opt_quiet = 1;
4717 } else if (string_match(optname, "verbose") || opt == 'v') {
4718 /* Verbose level can increase using multiple -v */
4719 if (arg) {
4720 /* Value obtained from config file */
4721 lttng_opt_verbose = config_parse_value(arg);
4722 } else {
4723 /* -v used on command line */
4724 lttng_opt_verbose++;
4725 }
4726 /* Clamp value to [0, 3] */
4727 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4728 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4729 } else if (string_match(optname, "verbose-consumer")) {
4730 if (arg) {
4731 opt_verbose_consumer = config_parse_value(arg);
4732 } else {
4733 opt_verbose_consumer += 1;
4734 }
4735 } else if (string_match(optname, "consumerd32-path")) {
4736 if (!arg || *arg == '\0') {
4737 ret = -EINVAL;
4738 goto end;
4739 }
4740 if (lttng_is_setuid_setgid()) {
4741 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4742 "--consumerd32-path");
4743 } else {
4744 if (consumerd32_bin_override) {
4745 free((void *) consumerd32_bin);
4746 }
4747 consumerd32_bin = strdup(arg);
4748 if (!consumerd32_bin) {
4749 PERROR("strdup");
4750 ret = -ENOMEM;
4751 }
4752 consumerd32_bin_override = 1;
4753 }
4754 } else if (string_match(optname, "consumerd32-libdir")) {
4755 if (!arg || *arg == '\0') {
4756 ret = -EINVAL;
4757 goto end;
4758 }
4759 if (lttng_is_setuid_setgid()) {
4760 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4761 "--consumerd32-libdir");
4762 } else {
4763 if (consumerd32_libdir_override) {
4764 free((void *) consumerd32_libdir);
4765 }
4766 consumerd32_libdir = strdup(arg);
4767 if (!consumerd32_libdir) {
4768 PERROR("strdup");
4769 ret = -ENOMEM;
4770 }
4771 consumerd32_libdir_override = 1;
4772 }
4773 } else if (string_match(optname, "consumerd64-path")) {
4774 if (!arg || *arg == '\0') {
4775 ret = -EINVAL;
4776 goto end;
4777 }
4778 if (lttng_is_setuid_setgid()) {
4779 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4780 "--consumerd64-path");
4781 } else {
4782 if (consumerd64_bin_override) {
4783 free((void *) consumerd64_bin);
4784 }
4785 consumerd64_bin = strdup(arg);
4786 if (!consumerd64_bin) {
4787 PERROR("strdup");
4788 ret = -ENOMEM;
4789 }
4790 consumerd64_bin_override = 1;
4791 }
4792 } else if (string_match(optname, "consumerd64-libdir")) {
4793 if (!arg || *arg == '\0') {
4794 ret = -EINVAL;
4795 goto end;
4796 }
4797 if (lttng_is_setuid_setgid()) {
4798 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4799 "--consumerd64-libdir");
4800 } else {
4801 if (consumerd64_libdir_override) {
4802 free((void *) consumerd64_libdir);
4803 }
4804 consumerd64_libdir = strdup(arg);
4805 if (!consumerd64_libdir) {
4806 PERROR("strdup");
4807 ret = -ENOMEM;
4808 }
4809 consumerd64_libdir_override = 1;
4810 }
4811 } else if (string_match(optname, "pidfile") || opt == 'p') {
4812 if (!arg || *arg == '\0') {
4813 ret = -EINVAL;
4814 goto end;
4815 }
4816 if (lttng_is_setuid_setgid()) {
4817 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4818 "-p, --pidfile");
4819 } else {
4820 free(opt_pidfile);
4821 opt_pidfile = strdup(arg);
4822 if (!opt_pidfile) {
4823 PERROR("strdup");
4824 ret = -ENOMEM;
4825 }
4826 }
4827 } else if (string_match(optname, "agent-tcp-port")) {
4828 if (!arg || *arg == '\0') {
4829 ret = -EINVAL;
4830 goto end;
4831 }
4832 if (lttng_is_setuid_setgid()) {
4833 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4834 "--agent-tcp-port");
4835 } else {
4836 unsigned long v;
4837
4838 if (!arg) {
4839 ret = -EINVAL;
4840 goto end;
4841 }
4842 errno = 0;
4843 v = strtoul(arg, NULL, 0);
4844 if (errno != 0 || !isdigit(arg[0])) {
4845 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4846 return -1;
4847 }
4848 if (v == 0 || v >= 65535) {
4849 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4850 return -1;
4851 }
4852 agent_tcp_port = (uint32_t) v;
4853 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4854 }
4855 } else if (string_match(optname, "load") || opt == 'l') {
4856 if (!arg || *arg == '\0') {
4857 ret = -EINVAL;
4858 goto end;
4859 }
4860 if (lttng_is_setuid_setgid()) {
4861 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4862 "-l, --load");
4863 } else {
4864 free(opt_load_session_path);
4865 opt_load_session_path = strdup(arg);
4866 if (!opt_load_session_path) {
4867 PERROR("strdup");
4868 ret = -ENOMEM;
4869 }
4870 }
4871 } else if (string_match(optname, "kmod-probes")) {
4872 if (!arg || *arg == '\0') {
4873 ret = -EINVAL;
4874 goto end;
4875 }
4876 if (lttng_is_setuid_setgid()) {
4877 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4878 "--kmod-probes");
4879 } else {
4880 free(kmod_probes_list);
4881 kmod_probes_list = strdup(arg);
4882 if (!kmod_probes_list) {
4883 PERROR("strdup");
4884 ret = -ENOMEM;
4885 }
4886 }
4887 } else if (string_match(optname, "extra-kmod-probes")) {
4888 if (!arg || *arg == '\0') {
4889 ret = -EINVAL;
4890 goto end;
4891 }
4892 if (lttng_is_setuid_setgid()) {
4893 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4894 "--extra-kmod-probes");
4895 } else {
4896 free(kmod_extra_probes_list);
4897 kmod_extra_probes_list = strdup(arg);
4898 if (!kmod_extra_probes_list) {
4899 PERROR("strdup");
4900 ret = -ENOMEM;
4901 }
4902 }
4903 } else if (string_match(optname, "config") || opt == 'f') {
4904 /* This is handled in set_options() thus silent skip. */
4905 goto end;
4906 } else {
4907 /* Unknown option or other error.
4908 * Error is printed by getopt, just return */
4909 ret = -1;
4910 }
4911
4912 end:
4913 if (ret == -EINVAL) {
4914 const char *opt_name = "unknown";
4915 int i;
4916
4917 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4918 i++) {
4919 if (opt == long_options[i].val) {
4920 opt_name = long_options[i].name;
4921 break;
4922 }
4923 }
4924
4925 WARN("Invalid argument provided for option \"%s\", using default value.",
4926 opt_name);
4927 }
4928
4929 return ret;
4930 }
4931
4932 /*
4933 * config_entry_handler_cb used to handle options read from a config file.
4934 * See config_entry_handler_cb comment in common/config/config.h for the
4935 * return value conventions.
4936 */
4937 static int config_entry_handler(const struct config_entry *entry, void *unused)
4938 {
4939 int ret = 0, i;
4940
4941 if (!entry || !entry->name || !entry->value) {
4942 ret = -EINVAL;
4943 goto end;
4944 }
4945
4946 /* Check if the option is to be ignored */
4947 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4948 if (!strcmp(entry->name, config_ignore_options[i])) {
4949 goto end;
4950 }
4951 }
4952
4953 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4954 i++) {
4955
4956 /* Ignore if not fully matched. */
4957 if (strcmp(entry->name, long_options[i].name)) {
4958 continue;
4959 }
4960
4961 /*
4962 * If the option takes no argument on the command line, we have to
4963 * check if the value is "true". We support non-zero numeric values,
4964 * true, on and yes.
4965 */
4966 if (!long_options[i].has_arg) {
4967 ret = config_parse_value(entry->value);
4968 if (ret <= 0) {
4969 if (ret) {
4970 WARN("Invalid configuration value \"%s\" for option %s",
4971 entry->value, entry->name);
4972 }
4973 /* False, skip boolean config option. */
4974 goto end;
4975 }
4976 }
4977
4978 ret = set_option(long_options[i].val, entry->value, entry->name);
4979 goto end;
4980 }
4981
4982 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4983
4984 end:
4985 return ret;
4986 }
4987
4988 /*
4989 * daemon configuration loading and argument parsing
4990 */
4991 static int set_options(int argc, char **argv)
4992 {
4993 int ret = 0, c = 0, option_index = 0;
4994 int orig_optopt = optopt, orig_optind = optind;
4995 char *optstring;
4996 const char *config_path = NULL;
4997
4998 optstring = utils_generate_optstring(long_options,
4999 sizeof(long_options) / sizeof(struct option));
5000 if (!optstring) {
5001 ret = -ENOMEM;
5002 goto end;
5003 }
5004
5005 /* Check for the --config option */
5006 while ((c = getopt_long(argc, argv, optstring, long_options,
5007 &option_index)) != -1) {
5008 if (c == '?') {
5009 ret = -EINVAL;
5010 goto end;
5011 } else if (c != 'f') {
5012 /* if not equal to --config option. */
5013 continue;
5014 }
5015
5016 if (lttng_is_setuid_setgid()) {
5017 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5018 "-f, --config");
5019 } else {
5020 config_path = utils_expand_path(optarg);
5021 if (!config_path) {
5022 ERR("Failed to resolve path: %s", optarg);
5023 }
5024 }
5025 }
5026
5027 ret = config_get_section_entries(config_path, config_section_name,
5028 config_entry_handler, NULL);
5029 if (ret) {
5030 if (ret > 0) {
5031 ERR("Invalid configuration option at line %i", ret);
5032 ret = -1;
5033 }
5034 goto end;
5035 }
5036
5037 /* Reset getopt's global state */
5038 optopt = orig_optopt;
5039 optind = orig_optind;
5040 while (1) {
5041 option_index = -1;
5042 /*
5043 * getopt_long() will not set option_index if it encounters a
5044 * short option.
5045 */
5046 c = getopt_long(argc, argv, optstring, long_options,
5047 &option_index);
5048 if (c == -1) {
5049 break;
5050 }
5051
5052 /*
5053 * Pass NULL as the long option name if popt left the index
5054 * unset.
5055 */
5056 ret = set_option(c, optarg,
5057 option_index < 0 ? NULL :
5058 long_options[option_index].name);
5059 if (ret < 0) {
5060 break;
5061 }
5062 }
5063
5064 end:
5065 free(optstring);
5066 return ret;
5067 }
5068
5069 /*
5070 * Creates the two needed socket by the daemon.
5071 * apps_sock - The communication socket for all UST apps.
5072 * client_sock - The communication of the cli tool (lttng).
5073 */
5074 static int init_daemon_socket(void)
5075 {
5076 int ret = 0;
5077 mode_t old_umask;
5078
5079 old_umask = umask(0);
5080
5081 /* Create client tool unix socket */
5082 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5083 if (client_sock < 0) {
5084 ERR("Create unix sock failed: %s", client_unix_sock_path);
5085 ret = -1;
5086 goto end;
5087 }
5088
5089 /* Set the cloexec flag */
5090 ret = utils_set_fd_cloexec(client_sock);
5091 if (ret < 0) {
5092 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5093 "Continuing but note that the consumer daemon will have a "
5094 "reference to this socket on exec()", client_sock);
5095 }
5096
5097 /* File permission MUST be 660 */
5098 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5099 if (ret < 0) {
5100 ERR("Set file permissions failed: %s", client_unix_sock_path);
5101 PERROR("chmod");
5102 goto end;
5103 }
5104
5105 /* Create the application unix socket */
5106 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5107 if (apps_sock < 0) {
5108 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5109 ret = -1;
5110 goto end;
5111 }
5112
5113 /* Set the cloexec flag */
5114 ret = utils_set_fd_cloexec(apps_sock);
5115 if (ret < 0) {
5116 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5117 "Continuing but note that the consumer daemon will have a "
5118 "reference to this socket on exec()", apps_sock);
5119 }
5120
5121 /* File permission MUST be 666 */
5122 ret = chmod(apps_unix_sock_path,
5123 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5124 if (ret < 0) {
5125 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5126 PERROR("chmod");
5127 goto end;
5128 }
5129
5130 DBG3("Session daemon client socket %d and application socket %d created",
5131 client_sock, apps_sock);
5132
5133 end:
5134 umask(old_umask);
5135 return ret;
5136 }
5137
5138 /*
5139 * Check if the global socket is available, and if a daemon is answering at the
5140 * other side. If yes, error is returned.
5141 */
5142 static int check_existing_daemon(void)
5143 {
5144 /* Is there anybody out there ? */
5145 if (lttng_session_daemon_alive()) {
5146 return -EEXIST;
5147 }
5148
5149 return 0;
5150 }
5151
5152 /*
5153 * Set the tracing group gid onto the client socket.
5154 *
5155 * Race window between mkdir and chown is OK because we are going from more
5156 * permissive (root.root) to less permissive (root.tracing).
5157 */
5158 static int set_permissions(char *rundir)
5159 {
5160 int ret;
5161 gid_t gid;
5162
5163 gid = utils_get_group_id(tracing_group_name);
5164
5165 /* Set lttng run dir */
5166 ret = chown(rundir, 0, gid);
5167 if (ret < 0) {
5168 ERR("Unable to set group on %s", rundir);
5169 PERROR("chown");
5170 }
5171
5172 /*
5173 * Ensure all applications and tracing group can search the run
5174 * dir. Allow everyone to read the directory, since it does not
5175 * buy us anything to hide its content.
5176 */
5177 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5178 if (ret < 0) {
5179 ERR("Unable to set permissions on %s", rundir);
5180 PERROR("chmod");
5181 }
5182
5183 /* lttng client socket path */
5184 ret = chown(client_unix_sock_path, 0, gid);
5185 if (ret < 0) {
5186 ERR("Unable to set group on %s", client_unix_sock_path);
5187 PERROR("chown");
5188 }
5189
5190 /* kconsumer error socket path */
5191 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5192 if (ret < 0) {
5193 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5194 PERROR("chown");
5195 }
5196
5197 /* 64-bit ustconsumer error socket path */
5198 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5199 if (ret < 0) {
5200 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5201 PERROR("chown");
5202 }
5203
5204 /* 32-bit ustconsumer compat32 error socket path */
5205 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5206 if (ret < 0) {
5207 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5208 PERROR("chown");
5209 }
5210
5211 DBG("All permissions are set");
5212
5213 return ret;
5214 }
5215
5216 /*
5217 * Create the lttng run directory needed for all global sockets and pipe.
5218 */
5219 static int create_lttng_rundir(const char *rundir)
5220 {
5221 int ret;
5222
5223 DBG3("Creating LTTng run directory: %s", rundir);
5224
5225 ret = mkdir(rundir, S_IRWXU);
5226 if (ret < 0) {
5227 if (errno != EEXIST) {
5228 ERR("Unable to create %s", rundir);
5229 goto error;
5230 } else {
5231 ret = 0;
5232 }
5233 }
5234
5235 error:
5236 return ret;
5237 }
5238
5239 /*
5240 * Setup sockets and directory needed by the kconsumerd communication with the
5241 * session daemon.
5242 */
5243 static int set_consumer_sockets(struct consumer_data *consumer_data,
5244 const char *rundir)
5245 {
5246 int ret;
5247 char path[PATH_MAX];
5248
5249 switch (consumer_data->type) {
5250 case LTTNG_CONSUMER_KERNEL:
5251 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5252 break;
5253 case LTTNG_CONSUMER64_UST:
5254 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5255 break;
5256 case LTTNG_CONSUMER32_UST:
5257 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5258 break;
5259 default:
5260 ERR("Consumer type unknown");
5261 ret = -EINVAL;
5262 goto error;
5263 }
5264
5265 DBG2("Creating consumer directory: %s", path);
5266
5267 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5268 if (ret < 0) {
5269 if (errno != EEXIST) {
5270 PERROR("mkdir");
5271 ERR("Failed to create %s", path);
5272 goto error;
5273 }
5274 ret = -1;
5275 }
5276 if (is_root) {
5277 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5278 if (ret < 0) {
5279 ERR("Unable to set group on %s", path);
5280 PERROR("chown");
5281 goto error;
5282 }
5283 }
5284
5285 /* Create the kconsumerd error unix socket */
5286 consumer_data->err_sock =
5287 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5288 if (consumer_data->err_sock < 0) {
5289 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5290 ret = -1;
5291 goto error;
5292 }
5293
5294 /*
5295 * Set the CLOEXEC flag. Return code is useless because either way, the
5296 * show must go on.
5297 */
5298 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5299 if (ret < 0) {
5300 PERROR("utils_set_fd_cloexec");
5301 /* continue anyway */
5302 }
5303
5304 /* File permission MUST be 660 */
5305 ret = chmod(consumer_data->err_unix_sock_path,
5306 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5307 if (ret < 0) {
5308 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5309 PERROR("chmod");
5310 goto error;
5311 }
5312
5313 error:
5314 return ret;
5315 }
5316
5317 /*
5318 * Signal handler for the daemon
5319 *
5320 * Simply stop all worker threads, leaving main() return gracefully after
5321 * joining all threads and calling cleanup().
5322 */
5323 static void sighandler(int sig)
5324 {
5325 switch (sig) {
5326 case SIGPIPE:
5327 DBG("SIGPIPE caught");
5328 return;
5329 case SIGINT:
5330 DBG("SIGINT caught");
5331 stop_threads();
5332 break;
5333 case SIGTERM:
5334 DBG("SIGTERM caught");
5335 stop_threads();
5336 break;
5337 case SIGUSR1:
5338 CMM_STORE_SHARED(recv_child_signal, 1);
5339 break;
5340 default:
5341 break;
5342 }
5343 }
5344
5345 /*
5346 * Setup signal handler for :
5347 * SIGINT, SIGTERM, SIGPIPE
5348 */
5349 static int set_signal_handler(void)
5350 {
5351 int ret = 0;
5352 struct sigaction sa;
5353 sigset_t sigset;
5354
5355 if ((ret = sigemptyset(&sigset)) < 0) {
5356 PERROR("sigemptyset");
5357 return ret;
5358 }
5359
5360 sa.sa_handler = sighandler;
5361 sa.sa_mask = sigset;
5362 sa.sa_flags = 0;
5363 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5364 PERROR("sigaction");
5365 return ret;
5366 }
5367
5368 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5369 PERROR("sigaction");
5370 return ret;
5371 }
5372
5373 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5374 PERROR("sigaction");
5375 return ret;
5376 }
5377
5378 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5379 PERROR("sigaction");
5380 return ret;
5381 }
5382
5383 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5384
5385 return ret;
5386 }
5387
5388 /*
5389 * Set open files limit to unlimited. This daemon can open a large number of
5390 * file descriptors in order to consumer multiple kernel traces.
5391 */
5392 static void set_ulimit(void)
5393 {
5394 int ret;
5395 struct rlimit lim;
5396
5397 /* The kernel does not allowed an infinite limit for open files */
5398 lim.rlim_cur = 65535;
5399 lim.rlim_max = 65535;
5400
5401 ret = setrlimit(RLIMIT_NOFILE, &lim);
5402 if (ret < 0) {
5403 PERROR("failed to set open files limit");
5404 }
5405 }
5406
5407 /*
5408 * Write pidfile using the rundir and opt_pidfile.
5409 */
5410 static int write_pidfile(void)
5411 {
5412 int ret;
5413 char pidfile_path[PATH_MAX];
5414
5415 assert(rundir);
5416
5417 if (opt_pidfile) {
5418 if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) {
5419 ret = -1;
5420 goto error;
5421 }
5422 } else {
5423 /* Build pidfile path from rundir and opt_pidfile. */
5424 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5425 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5426 if (ret < 0) {
5427 PERROR("snprintf pidfile path");
5428 goto error;
5429 }
5430 }
5431
5432 /*
5433 * Create pid file in rundir.
5434 */
5435 ret = utils_create_pid_file(getpid(), pidfile_path);
5436 error:
5437 return ret;
5438 }
5439
5440 /*
5441 * Create lockfile using the rundir and return its fd.
5442 */
5443 static int create_lockfile(void)
5444 {
5445 int ret;
5446 char lockfile_path[PATH_MAX];
5447
5448 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5449 if (ret < 0) {
5450 goto error;
5451 }
5452
5453 ret = utils_create_lock_file(lockfile_path);
5454 error:
5455 return ret;
5456 }
5457
5458 /*
5459 * Write agent TCP port using the rundir.
5460 */
5461 static int write_agent_port(void)
5462 {
5463 int ret;
5464 char path[PATH_MAX];
5465
5466 assert(rundir);
5467
5468 ret = snprintf(path, sizeof(path), "%s/"
5469 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5470 if (ret < 0) {
5471 PERROR("snprintf agent port path");
5472 goto error;
5473 }
5474
5475 /*
5476 * Create TCP agent port file in rundir.
5477 */
5478 ret = utils_create_pid_file(agent_tcp_port, path);
5479
5480 error:
5481 return ret;
5482 }
5483
5484 /*
5485 * main
5486 */
5487 int main(int argc, char **argv)
5488 {
5489 int ret = 0, retval = 0;
5490 void *status;
5491 const char *home_path, *env_app_timeout;
5492
5493 /* Initialize agent apps ht global variable */
5494 agent_apps_ht_by_sock = NULL;
5495
5496 init_kernel_workarounds();
5497
5498 rcu_register_thread();
5499
5500 if (set_signal_handler()) {
5501 retval = -1;
5502 goto exit_set_signal_handler;
5503 }
5504
5505 setup_consumerd_path();
5506
5507 page_size = sysconf(_SC_PAGESIZE);
5508 if (page_size < 0) {
5509 PERROR("sysconf _SC_PAGESIZE");
5510 page_size = LONG_MAX;
5511 WARN("Fallback page size to %ld", page_size);
5512 }
5513
5514 /*
5515 * Parse arguments and load the daemon configuration file.
5516 *
5517 * We have an exit_options exit path to free memory reserved by
5518 * set_options. This is needed because the rest of sessiond_cleanup()
5519 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5520 * depends on set_options.
5521 */
5522 progname = argv[0];
5523 if (set_options(argc, argv)) {
5524 retval = -1;
5525 goto exit_options;
5526 }
5527
5528 /* Daemonize */
5529 if (opt_daemon || opt_background) {
5530 int i;
5531
5532 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5533 !opt_background);
5534 if (ret < 0) {
5535 retval = -1;
5536 goto exit_options;
5537 }
5538
5539 /*
5540 * We are in the child. Make sure all other file descriptors are
5541 * closed, in case we are called with more opened file
5542 * descriptors than the standard ones.
5543 */
5544 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5545 (void) close(i);
5546 }
5547 }
5548
5549 if (run_as_create_worker(argv[0]) < 0) {
5550 goto exit_create_run_as_worker_cleanup;
5551 }
5552
5553 /*
5554 * Starting from here, we can create threads. This needs to be after
5555 * lttng_daemonize due to RCU.
5556 */
5557
5558 /*
5559 * Initialize the health check subsystem. This call should set the
5560 * appropriate time values.
5561 */
5562 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5563 if (!health_sessiond) {
5564 PERROR("health_app_create error");
5565 retval = -1;
5566 goto exit_health_sessiond_cleanup;
5567 }
5568
5569 if (init_ht_cleanup_quit_pipe()) {
5570 retval = -1;
5571 goto exit_ht_cleanup_quit_pipe;
5572 }
5573
5574 /* Setup the thread ht_cleanup communication pipe. */
5575 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5576 retval = -1;
5577 goto exit_ht_cleanup_pipe;
5578 }
5579
5580 /* Set up max poll set size */
5581 if (lttng_poll_set_max_size()) {
5582 retval = -1;
5583 goto exit_set_max_size;
5584 }
5585
5586 /* Create thread to clean up RCU hash tables */
5587 ret = pthread_create(&ht_cleanup_thread, NULL,
5588 thread_ht_cleanup, (void *) NULL);
5589 if (ret) {
5590 errno = ret;
5591 PERROR("pthread_create ht_cleanup");
5592 retval = -1;
5593 goto exit_ht_cleanup;
5594 }
5595
5596 /* Create thread quit pipe */
5597 if (init_thread_quit_pipe()) {
5598 retval = -1;
5599 goto exit_init_data;
5600 }
5601
5602 /* Check if daemon is UID = 0 */
5603 is_root = !getuid();
5604
5605 if (is_root) {
5606 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5607 if (!rundir) {
5608 retval = -1;
5609 goto exit_init_data;
5610 }
5611
5612 /* Create global run dir with root access */
5613 if (create_lttng_rundir(rundir)) {
5614 retval = -1;
5615 goto exit_init_data;
5616 }
5617
5618 if (strlen(apps_unix_sock_path) == 0) {
5619 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5620 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5621 if (ret < 0) {
5622 retval = -1;
5623 goto exit_init_data;
5624 }
5625 }
5626
5627 if (strlen(client_unix_sock_path) == 0) {
5628 ret = snprintf(client_unix_sock_path, PATH_MAX,
5629 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5630 if (ret < 0) {
5631 retval = -1;
5632 goto exit_init_data;
5633 }
5634 }
5635
5636 /* Set global SHM for ust */
5637 if (strlen(wait_shm_path) == 0) {
5638 ret = snprintf(wait_shm_path, PATH_MAX,
5639 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5640 if (ret < 0) {
5641 retval = -1;
5642 goto exit_init_data;
5643 }
5644 }
5645
5646 if (strlen(health_unix_sock_path) == 0) {
5647 ret = snprintf(health_unix_sock_path,
5648 sizeof(health_unix_sock_path),
5649 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5650 if (ret < 0) {
5651 retval = -1;
5652 goto exit_init_data;
5653 }
5654 }
5655
5656 /* Setup kernel consumerd path */
5657 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5658 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5659 if (ret < 0) {
5660 retval = -1;
5661 goto exit_init_data;
5662 }
5663 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5664 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5665 if (ret < 0) {
5666 retval = -1;
5667 goto exit_init_data;
5668 }
5669
5670 DBG2("Kernel consumer err path: %s",
5671 kconsumer_data.err_unix_sock_path);
5672 DBG2("Kernel consumer cmd path: %s",
5673 kconsumer_data.cmd_unix_sock_path);
5674 } else {
5675 home_path = utils_get_home_dir();
5676 if (home_path == NULL) {
5677 /* TODO: Add --socket PATH option */
5678 ERR("Can't get HOME directory for sockets creation.");
5679 retval = -1;
5680 goto exit_init_data;
5681 }
5682
5683 /*
5684 * Create rundir from home path. This will create something like
5685 * $HOME/.lttng
5686 */
5687 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5688 if (ret < 0) {
5689 retval = -1;
5690 goto exit_init_data;
5691 }
5692
5693 if (create_lttng_rundir(rundir)) {
5694 retval = -1;
5695 goto exit_init_data;
5696 }
5697
5698 if (strlen(apps_unix_sock_path) == 0) {
5699 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5700 DEFAULT_HOME_APPS_UNIX_SOCK,
5701 home_path);
5702 if (ret < 0) {
5703 retval = -1;
5704 goto exit_init_data;
5705 }
5706 }
5707
5708 /* Set the cli tool unix socket path */
5709 if (strlen(client_unix_sock_path) == 0) {
5710 ret = snprintf(client_unix_sock_path, PATH_MAX,
5711 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5712 home_path);
5713 if (ret < 0) {
5714 retval = -1;
5715 goto exit_init_data;
5716 }
5717 }
5718
5719 /* Set global SHM for ust */
5720 if (strlen(wait_shm_path) == 0) {
5721 ret = snprintf(wait_shm_path, PATH_MAX,
5722 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5723 getuid());
5724 if (ret < 0) {
5725 retval = -1;
5726 goto exit_init_data;
5727 }
5728 }
5729
5730 /* Set health check Unix path */
5731 if (strlen(health_unix_sock_path) == 0) {
5732 ret = snprintf(health_unix_sock_path,
5733 sizeof(health_unix_sock_path),
5734 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5735 home_path);
5736 if (ret < 0) {
5737 retval = -1;
5738 goto exit_init_data;
5739 }
5740 }
5741 }
5742
5743 lockfile_fd = create_lockfile();
5744 if (lockfile_fd < 0) {
5745 retval = -1;
5746 goto exit_init_data;
5747 }
5748
5749 /* Set consumer initial state */
5750 kernel_consumerd_state = CONSUMER_STOPPED;
5751 ust_consumerd_state = CONSUMER_STOPPED;
5752
5753 DBG("Client socket path %s", client_unix_sock_path);
5754 DBG("Application socket path %s", apps_unix_sock_path);
5755 DBG("Application wait path %s", wait_shm_path);
5756 DBG("LTTng run directory path: %s", rundir);
5757
5758 /* 32 bits consumerd path setup */
5759 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5760 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5761 if (ret < 0) {
5762 PERROR("snprintf 32-bit consumer error socket path");
5763 retval = -1;
5764 goto exit_init_data;
5765 }
5766 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5767 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5768 if (ret < 0) {
5769 PERROR("snprintf 32-bit consumer command socket path");
5770 retval = -1;
5771 goto exit_init_data;
5772 }
5773
5774 DBG2("UST consumer 32 bits err path: %s",
5775 ustconsumer32_data.err_unix_sock_path);
5776 DBG2("UST consumer 32 bits cmd path: %s",
5777 ustconsumer32_data.cmd_unix_sock_path);
5778
5779 /* 64 bits consumerd path setup */
5780 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5781 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5782 if (ret < 0) {
5783 PERROR("snprintf 64-bit consumer error socket path");
5784 retval = -1;
5785 goto exit_init_data;
5786 }
5787 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5788 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5789 if (ret < 0) {
5790 PERROR("snprintf 64-bit consumer command socket path");
5791 retval = -1;
5792 goto exit_init_data;
5793 }
5794
5795 DBG2("UST consumer 64 bits err path: %s",
5796 ustconsumer64_data.err_unix_sock_path);
5797 DBG2("UST consumer 64 bits cmd path: %s",
5798 ustconsumer64_data.cmd_unix_sock_path);
5799
5800 /*
5801 * See if daemon already exist.
5802 */
5803 if (check_existing_daemon()) {
5804 ERR("Already running daemon.\n");
5805 /*
5806 * We do not goto exit because we must not cleanup()
5807 * because a daemon is already running.
5808 */
5809 retval = -1;
5810 goto exit_init_data;
5811 }
5812
5813 /*
5814 * Init UST app hash table. Alloc hash table before this point since
5815 * cleanup() can get called after that point.
5816 */
5817 if (ust_app_ht_alloc()) {
5818 ERR("Failed to allocate UST app hash table");
5819 retval = -1;
5820 goto exit_init_data;
5821 }
5822
5823 /*
5824 * Initialize agent app hash table. We allocate the hash table here
5825 * since cleanup() can get called after this point.
5826 */
5827 if (agent_app_ht_alloc()) {
5828 ERR("Failed to allocate Agent app hash table");
5829 retval = -1;
5830 goto exit_init_data;
5831 }
5832
5833 /*
5834 * These actions must be executed as root. We do that *after* setting up
5835 * the sockets path because we MUST make the check for another daemon using
5836 * those paths *before* trying to set the kernel consumer sockets and init
5837 * kernel tracer.
5838 */
5839 if (is_root) {
5840 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5841 retval = -1;
5842 goto exit_init_data;
5843 }
5844
5845 /* Setup kernel tracer */
5846 if (!opt_no_kernel) {
5847 init_kernel_tracer();
5848 if (kernel_tracer_fd >= 0) {
5849 ret = syscall_init_table();
5850 if (ret < 0) {
5851 ERR("Unable to populate syscall table. "
5852 "Syscall tracing won't work "
5853 "for this session daemon.");
5854 }
5855 }
5856 }
5857
5858 /* Set ulimit for open files */
5859 set_ulimit();
5860 }
5861 /* init lttng_fd tracking must be done after set_ulimit. */
5862 lttng_fd_init();
5863
5864 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5865 retval = -1;
5866 goto exit_init_data;
5867 }
5868
5869 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5870 retval = -1;
5871 goto exit_init_data;
5872 }
5873
5874 /* Setup the needed unix socket */
5875 if (init_daemon_socket()) {
5876 retval = -1;
5877 goto exit_init_data;
5878 }
5879
5880 /* Set credentials to socket */
5881 if (is_root && set_permissions(rundir)) {
5882 retval = -1;
5883 goto exit_init_data;
5884 }
5885
5886 /* Get parent pid if -S, --sig-parent is specified. */
5887 if (opt_sig_parent) {
5888 ppid = getppid();
5889 }
5890
5891 /* Setup the kernel pipe for waking up the kernel thread */
5892 if (is_root && !opt_no_kernel) {
5893 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5894 retval = -1;
5895 goto exit_init_data;
5896 }
5897 }
5898
5899 /* Setup the thread apps communication pipe. */
5900 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5901 retval = -1;
5902 goto exit_init_data;
5903 }
5904
5905 /* Setup the thread apps notify communication pipe. */
5906 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5907 retval = -1;
5908 goto exit_init_data;
5909 }
5910
5911 /* Initialize global buffer per UID and PID registry. */
5912 buffer_reg_init_uid_registry();
5913 buffer_reg_init_pid_registry();
5914
5915 /* Init UST command queue. */
5916 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5917
5918 /*
5919 * Get session list pointer. This pointer MUST NOT be free'd. This list
5920 * is statically declared in session.c
5921 */
5922 session_list_ptr = session_get_list();
5923
5924 cmd_init();
5925
5926 /* Check for the application socket timeout env variable. */
5927 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5928 if (env_app_timeout) {
5929 app_socket_timeout = atoi(env_app_timeout);
5930 } else {
5931 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5932 }
5933
5934 ret = write_pidfile();
5935 if (ret) {
5936 ERR("Error in write_pidfile");
5937 retval = -1;
5938 goto exit_init_data;
5939 }
5940 ret = write_agent_port();
5941 if (ret) {
5942 ERR("Error in write_agent_port");
5943 retval = -1;
5944 goto exit_init_data;
5945 }
5946
5947 /* Initialize communication library */
5948 lttcomm_init();
5949 /* Initialize TCP timeout values */
5950 lttcomm_inet_init();
5951
5952 if (load_session_init_data(&load_info) < 0) {
5953 retval = -1;
5954 goto exit_init_data;
5955 }
5956 load_info->path = opt_load_session_path;
5957
5958 /* Create health-check thread */
5959 ret = pthread_create(&health_thread, NULL,
5960 thread_manage_health, (void *) NULL);
5961 if (ret) {
5962 errno = ret;
5963 PERROR("pthread_create health");
5964 retval = -1;
5965 goto exit_health;
5966 }
5967
5968 /* Create thread to manage the client socket */
5969 ret = pthread_create(&client_thread, NULL,
5970 thread_manage_clients, (void *) NULL);
5971 if (ret) {
5972 errno = ret;
5973 PERROR("pthread_create clients");
5974 retval = -1;
5975 goto exit_client;
5976 }
5977
5978 /* Create thread to dispatch registration */
5979 ret = pthread_create(&dispatch_thread, NULL,
5980 thread_dispatch_ust_registration, (void *) NULL);
5981 if (ret) {
5982 errno = ret;
5983 PERROR("pthread_create dispatch");
5984 retval = -1;
5985 goto exit_dispatch;
5986 }
5987
5988 /* Create thread to manage application registration. */
5989 ret = pthread_create(&reg_apps_thread, NULL,
5990 thread_registration_apps, (void *) NULL);
5991 if (ret) {
5992 errno = ret;
5993 PERROR("pthread_create registration");
5994 retval = -1;
5995 goto exit_reg_apps;
5996 }
5997
5998 /* Create thread to manage application socket */
5999 ret = pthread_create(&apps_thread, NULL,
6000 thread_manage_apps, (void *) NULL);
6001 if (ret) {
6002 errno = ret;
6003 PERROR("pthread_create apps");
6004 retval = -1;
6005 goto exit_apps;
6006 }
6007
6008 /* Create thread to manage application notify socket */
6009 ret = pthread_create(&apps_notify_thread, NULL,
6010 ust_thread_manage_notify, (void *) NULL);
6011 if (ret) {
6012 errno = ret;
6013 PERROR("pthread_create notify");
6014 retval = -1;
6015 goto exit_apps_notify;
6016 }
6017
6018 /* Create agent registration thread. */
6019 ret = pthread_create(&agent_reg_thread, NULL,
6020 agent_thread_manage_registration, (void *) NULL);
6021 if (ret) {
6022 errno = ret;
6023 PERROR("pthread_create agent");
6024 retval = -1;
6025 goto exit_agent_reg;
6026 }
6027
6028 /* Don't start this thread if kernel tracing is not requested nor root */
6029 if (is_root && !opt_no_kernel) {
6030 /* Create kernel thread to manage kernel event */
6031 ret = pthread_create(&kernel_thread, NULL,
6032 thread_manage_kernel, (void *) NULL);
6033 if (ret) {
6034 errno = ret;
6035 PERROR("pthread_create kernel");
6036 retval = -1;
6037 goto exit_kernel;
6038 }
6039 }
6040
6041 /* Create session loading thread. */
6042 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
6043 load_info);
6044 if (ret) {
6045 errno = ret;
6046 PERROR("pthread_create load_session_thread");
6047 retval = -1;
6048 goto exit_load_session;
6049 }
6050
6051 /*
6052 * This is where we start awaiting program completion (e.g. through
6053 * signal that asks threads to teardown).
6054 */
6055
6056 ret = pthread_join(load_session_thread, &status);
6057 if (ret) {
6058 errno = ret;
6059 PERROR("pthread_join load_session_thread");
6060 retval = -1;
6061 }
6062 exit_load_session:
6063
6064 if (is_root && !opt_no_kernel) {
6065 ret = pthread_join(kernel_thread, &status);
6066 if (ret) {
6067 errno = ret;
6068 PERROR("pthread_join");
6069 retval = -1;
6070 }
6071 }
6072 exit_kernel:
6073
6074 ret = pthread_join(agent_reg_thread, &status);
6075 if (ret) {
6076 errno = ret;
6077 PERROR("pthread_join agent");
6078 retval = -1;
6079 }
6080 exit_agent_reg:
6081
6082 ret = pthread_join(apps_notify_thread, &status);
6083 if (ret) {
6084 errno = ret;
6085 PERROR("pthread_join apps notify");
6086 retval = -1;
6087 }
6088 exit_apps_notify:
6089
6090 ret = pthread_join(apps_thread, &status);
6091 if (ret) {
6092 errno = ret;
6093 PERROR("pthread_join apps");
6094 retval = -1;
6095 }
6096 exit_apps:
6097
6098 ret = pthread_join(reg_apps_thread, &status);
6099 if (ret) {
6100 errno = ret;
6101 PERROR("pthread_join");
6102 retval = -1;
6103 }
6104 exit_reg_apps:
6105
6106 /*
6107 * Join dispatch thread after joining reg_apps_thread to ensure
6108 * we don't leak applications in the queue.
6109 */
6110 ret = pthread_join(dispatch_thread, &status);
6111 if (ret) {
6112 errno = ret;
6113 PERROR("pthread_join");
6114 retval = -1;
6115 }
6116 exit_dispatch:
6117
6118 ret = pthread_join(client_thread, &status);
6119 if (ret) {
6120 errno = ret;
6121 PERROR("pthread_join");
6122 retval = -1;
6123 }
6124 exit_client:
6125
6126 ret = pthread_join(health_thread, &status);
6127 if (ret) {
6128 errno = ret;
6129 PERROR("pthread_join health thread");
6130 retval = -1;
6131 }
6132 exit_health:
6133
6134 exit_init_data:
6135 /*
6136 * sessiond_cleanup() is called when no other thread is running, except
6137 * the ht_cleanup thread, which is needed to destroy the hash tables.
6138 */
6139 rcu_thread_online();
6140 sessiond_cleanup();
6141 rcu_thread_offline();
6142 rcu_unregister_thread();
6143
6144 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6145 if (ret < 0) {
6146 ERR("write error on ht_cleanup quit pipe");
6147 retval = -1;
6148 }
6149
6150 ret = pthread_join(ht_cleanup_thread, &status);
6151 if (ret) {
6152 errno = ret;
6153 PERROR("pthread_join ht cleanup thread");
6154 retval = -1;
6155 }
6156 exit_ht_cleanup:
6157 exit_set_max_size:
6158
6159 utils_close_pipe(ht_cleanup_pipe);
6160 exit_ht_cleanup_pipe:
6161
6162 /*
6163 * Close the ht_cleanup quit pipe.
6164 */
6165 utils_close_pipe(ht_cleanup_quit_pipe);
6166 exit_ht_cleanup_quit_pipe:
6167
6168 health_app_destroy(health_sessiond);
6169 exit_health_sessiond_cleanup:
6170 exit_create_run_as_worker_cleanup:
6171
6172 exit_options:
6173 /* Ensure all prior call_rcu are done. */
6174 rcu_barrier();
6175
6176 sessiond_cleanup_options();
6177
6178 exit_set_signal_handler:
6179
6180 if (!retval) {
6181 exit(EXIT_SUCCESS);
6182 } else {
6183 exit(EXIT_FAILURE);
6184 }
6185 }
This page took 0.157181 seconds and 4 git commands to generate.