Cleanup: Replace all perror() uses by the PERROR macro
[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/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/config.h>
52
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
55 #include "channel.h"
56 #include "cmd.h"
57 #include "consumer.h"
58 #include "context.h"
59 #include "event.h"
60 #include "kernel.h"
61 #include "kernel-consumer.h"
62 #include "modprobe.h"
63 #include "shm.h"
64 #include "ust-ctl.h"
65 #include "ust-consumer.h"
66 #include "utils.h"
67 #include "fd-limit.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
72 #include "save.h"
73 #include "load-session-thread.h"
74 #include "syscall.h"
75
76 #define CONSUMERD_FILE "lttng-consumerd"
77
78 const char *progname;
79 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
80 static int tracing_group_name_override;
81 static char *opt_pidfile;
82 static int opt_sig_parent;
83 static int opt_verbose_consumer;
84 static int opt_daemon, opt_background;
85 static int opt_no_kernel;
86 static char *opt_load_session_path;
87 static pid_t ppid; /* Parent PID for --sig-parent option */
88 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
89 static char *rundir;
90 static int lockfile_fd = -1;
91
92 /* Set to 1 when a SIGUSR1 signal is received. */
93 static int recv_child_signal;
94
95 /*
96 * Consumer daemon specific control data. Every value not initialized here is
97 * set to 0 by the static definition.
98 */
99 static struct consumer_data kconsumer_data = {
100 .type = LTTNG_CONSUMER_KERNEL,
101 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer64_data = {
111 .type = LTTNG_CONSUMER64_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121 static struct consumer_data ustconsumer32_data = {
122 .type = LTTNG_CONSUMER32_UST,
123 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
124 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
125 .err_sock = -1,
126 .cmd_sock = -1,
127 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
128 .lock = PTHREAD_MUTEX_INITIALIZER,
129 .cond = PTHREAD_COND_INITIALIZER,
130 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
131 };
132
133 /* Command line options */
134 static const struct option long_options[] = {
135 { "client-sock", 1, 0, 'c' },
136 { "apps-sock", 1, 0, 'a' },
137 { "kconsumerd-cmd-sock", 1, 0, 'C' },
138 { "kconsumerd-err-sock", 1, 0, 'E' },
139 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
140 { "ustconsumerd32-err-sock", 1, 0, 'H' },
141 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
142 { "ustconsumerd64-err-sock", 1, 0, 'F' },
143 { "consumerd32-path", 1, 0, 'u' },
144 { "consumerd32-libdir", 1, 0, 'U' },
145 { "consumerd64-path", 1, 0, 't' },
146 { "consumerd64-libdir", 1, 0, 'T' },
147 { "daemonize", 0, 0, 'd' },
148 { "background", 0, 0, 'b' },
149 { "sig-parent", 0, 0, 'S' },
150 { "help", 0, 0, 'h' },
151 { "group", 1, 0, 'g' },
152 { "version", 0, 0, 'V' },
153 { "quiet", 0, 0, 'q' },
154 { "verbose", 0, 0, 'v' },
155 { "verbose-consumer", 0, 0, 'Z' },
156 { "no-kernel", 0, 0, 'N' },
157 { "pidfile", 1, 0, 'p' },
158 { "agent-tcp-port", 1, 0, 'J' },
159 { "config", 1, 0, 'f' },
160 { "load", 1, 0, 'l' },
161 { "kmod-probes", 1, 0, 'P' },
162 { "extra-kmod-probes", 1, 0, 'e' },
163 { NULL, 0, 0, 0 }
164 };
165
166 /* Command line options to ignore from configuration file */
167 static const char *config_ignore_options[] = { "help", "version", "config" };
168
169 /* Shared between threads */
170 static int dispatch_thread_exit;
171
172 /* Global application Unix socket path */
173 static char apps_unix_sock_path[PATH_MAX];
174 /* Global client Unix socket path */
175 static char client_unix_sock_path[PATH_MAX];
176 /* global wait shm path for UST */
177 static char wait_shm_path[PATH_MAX];
178 /* Global health check unix path */
179 static char health_unix_sock_path[PATH_MAX];
180
181 /* Sockets and FDs */
182 static int client_sock = -1;
183 static int apps_sock = -1;
184 int kernel_tracer_fd = -1;
185 static int kernel_poll_pipe[2] = { -1, -1 };
186
187 /*
188 * Quit pipe for all threads. This permits a single cancellation point
189 * for all threads when receiving an event on the pipe.
190 */
191 static int thread_quit_pipe[2] = { -1, -1 };
192 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
193
194 /*
195 * This pipe is used to inform the thread managing application communication
196 * that a command is queued and ready to be processed.
197 */
198 static int apps_cmd_pipe[2] = { -1, -1 };
199
200 int apps_cmd_notify_pipe[2] = { -1, -1 };
201
202 /* Pthread, Mutexes and Semaphores */
203 static pthread_t apps_thread;
204 static pthread_t apps_notify_thread;
205 static pthread_t reg_apps_thread;
206 static pthread_t client_thread;
207 static pthread_t kernel_thread;
208 static pthread_t dispatch_thread;
209 static pthread_t health_thread;
210 static pthread_t ht_cleanup_thread;
211 static pthread_t agent_reg_thread;
212 static pthread_t load_session_thread;
213
214 /*
215 * UST registration command queue. This queue is tied with a futex and uses a N
216 * wakers / 1 waiter implemented and detailed in futex.c/.h
217 *
218 * The thread_registration_apps and thread_dispatch_ust_registration uses this
219 * queue along with the wait/wake scheme. The thread_manage_apps receives down
220 * the line new application socket and monitors it for any I/O error or clean
221 * close that triggers an unregistration of the application.
222 */
223 static struct ust_cmd_queue ust_cmd_queue;
224
225 /*
226 * Pointer initialized before thread creation.
227 *
228 * This points to the tracing session list containing the session count and a
229 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
230 * MUST NOT be taken if you call a public function in session.c.
231 *
232 * The lock is nested inside the structure: session_list_ptr->lock. Please use
233 * session_lock_list and session_unlock_list for lock acquisition.
234 */
235 static struct ltt_session_list *session_list_ptr;
236
237 int ust_consumerd64_fd = -1;
238 int ust_consumerd32_fd = -1;
239
240 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
241 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
242 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
243 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
244 static int consumerd32_bin_override;
245 static int consumerd64_bin_override;
246 static int consumerd32_libdir_override;
247 static int consumerd64_libdir_override;
248
249 static const char *module_proc_lttng = "/proc/lttng";
250
251 /*
252 * Consumer daemon state which is changed when spawning it, killing it or in
253 * case of a fatal error.
254 */
255 enum consumerd_state {
256 CONSUMER_STARTED = 1,
257 CONSUMER_STOPPED = 2,
258 CONSUMER_ERROR = 3,
259 };
260
261 /*
262 * This consumer daemon state is used to validate if a client command will be
263 * able to reach the consumer. If not, the client is informed. For instance,
264 * doing a "lttng start" when the consumer state is set to ERROR will return an
265 * error to the client.
266 *
267 * The following example shows a possible race condition of this scheme:
268 *
269 * consumer thread error happens
270 * client cmd arrives
271 * client cmd checks state -> still OK
272 * consumer thread exit, sets error
273 * client cmd try to talk to consumer
274 * ...
275 *
276 * However, since the consumer is a different daemon, we have no way of making
277 * sure the command will reach it safely even with this state flag. This is why
278 * we consider that up to the state validation during command processing, the
279 * command is safe. After that, we can not guarantee the correctness of the
280 * client request vis-a-vis the consumer.
281 */
282 static enum consumerd_state ust_consumerd_state;
283 static enum consumerd_state kernel_consumerd_state;
284
285 /*
286 * Socket timeout for receiving and sending in seconds.
287 */
288 static int app_socket_timeout;
289
290 /* Set in main() with the current page size. */
291 long page_size;
292
293 /* Application health monitoring */
294 struct health_app *health_sessiond;
295
296 /* Agent TCP port for registration. Used by the agent thread. */
297 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
298
299 /* Am I root or not. */
300 int is_root; /* Set to 1 if the daemon is running as root */
301
302 const char * const config_section_name = "sessiond";
303
304 /* Load session thread information to operate. */
305 struct load_session_thread_data *load_info;
306
307 /*
308 * Whether sessiond is ready for commands/health check requests.
309 * NR_LTTNG_SESSIOND_READY must match the number of calls to
310 * sessiond_notify_ready().
311 */
312 #define NR_LTTNG_SESSIOND_READY 3
313 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
314
315 /* Notify parents that we are ready for cmd and health check */
316 LTTNG_HIDDEN
317 void sessiond_notify_ready(void)
318 {
319 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
320 /*
321 * Notify parent pid that we are ready to accept command
322 * for client side. This ppid is the one from the
323 * external process that spawned us.
324 */
325 if (opt_sig_parent) {
326 kill(ppid, SIGUSR1);
327 }
328
329 /*
330 * Notify the parent of the fork() process that we are
331 * ready.
332 */
333 if (opt_daemon || opt_background) {
334 kill(child_ppid, SIGUSR1);
335 }
336 }
337 }
338
339 static
340 void setup_consumerd_path(void)
341 {
342 const char *bin, *libdir;
343
344 /*
345 * Allow INSTALL_BIN_PATH to be used as a target path for the
346 * native architecture size consumer if CONFIG_CONSUMER*_PATH
347 * has not been defined.
348 */
349 #if (CAA_BITS_PER_LONG == 32)
350 if (!consumerd32_bin[0]) {
351 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
352 }
353 if (!consumerd32_libdir[0]) {
354 consumerd32_libdir = INSTALL_LIB_PATH;
355 }
356 #elif (CAA_BITS_PER_LONG == 64)
357 if (!consumerd64_bin[0]) {
358 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
359 }
360 if (!consumerd64_libdir[0]) {
361 consumerd64_libdir = INSTALL_LIB_PATH;
362 }
363 #else
364 #error "Unknown bitness"
365 #endif
366
367 /*
368 * runtime env. var. overrides the build default.
369 */
370 bin = getenv("LTTNG_CONSUMERD32_BIN");
371 if (bin) {
372 consumerd32_bin = bin;
373 }
374 bin = getenv("LTTNG_CONSUMERD64_BIN");
375 if (bin) {
376 consumerd64_bin = bin;
377 }
378 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
379 if (libdir) {
380 consumerd32_libdir = libdir;
381 }
382 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
383 if (libdir) {
384 consumerd64_libdir = libdir;
385 }
386 }
387
388 static
389 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
390 int *a_pipe)
391 {
392 int ret;
393
394 assert(events);
395
396 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
397 if (ret < 0) {
398 goto error;
399 }
400
401 /* Add quit pipe */
402 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
403 if (ret < 0) {
404 goto error;
405 }
406
407 return 0;
408
409 error:
410 return ret;
411 }
412
413 /*
414 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
415 */
416 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
417 {
418 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
419 }
420
421 /*
422 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
423 */
424 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
425 size_t size)
426 {
427 return __sessiond_set_thread_pollset(events, size,
428 ht_cleanup_quit_pipe);
429 }
430
431 static
432 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
433 {
434 if (fd == a_pipe && (events & LPOLLIN)) {
435 return 1;
436 }
437 return 0;
438 }
439
440 /*
441 * Check if the thread quit pipe was triggered.
442 *
443 * Return 1 if it was triggered else 0;
444 */
445 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
446 {
447 return __sessiond_check_thread_quit_pipe(fd, events,
448 thread_quit_pipe[0]);
449 }
450
451 /*
452 * Check if the ht_cleanup thread quit pipe was triggered.
453 *
454 * Return 1 if it was triggered else 0;
455 */
456 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
457 {
458 return __sessiond_check_thread_quit_pipe(fd, events,
459 ht_cleanup_quit_pipe[0]);
460 }
461
462 /*
463 * Init thread quit pipe.
464 *
465 * Return -1 on error or 0 if all pipes are created.
466 */
467 static int __init_thread_quit_pipe(int *a_pipe)
468 {
469 int ret, i;
470
471 ret = pipe(a_pipe);
472 if (ret < 0) {
473 PERROR("thread quit pipe");
474 goto error;
475 }
476
477 for (i = 0; i < 2; i++) {
478 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
479 if (ret < 0) {
480 PERROR("fcntl");
481 goto error;
482 }
483 }
484
485 error:
486 return ret;
487 }
488
489 static int init_thread_quit_pipe(void)
490 {
491 return __init_thread_quit_pipe(thread_quit_pipe);
492 }
493
494 static int init_ht_cleanup_quit_pipe(void)
495 {
496 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
497 }
498
499 /*
500 * Stop all threads by closing the thread quit pipe.
501 */
502 static void stop_threads(void)
503 {
504 int ret;
505
506 /* Stopping all threads */
507 DBG("Terminating all threads");
508 ret = notify_thread_pipe(thread_quit_pipe[1]);
509 if (ret < 0) {
510 ERR("write error on thread quit pipe");
511 }
512
513 /* Dispatch thread */
514 CMM_STORE_SHARED(dispatch_thread_exit, 1);
515 futex_nto1_wake(&ust_cmd_queue.futex);
516 }
517
518 /*
519 * Close every consumer sockets.
520 */
521 static void close_consumer_sockets(void)
522 {
523 int ret;
524
525 if (kconsumer_data.err_sock >= 0) {
526 ret = close(kconsumer_data.err_sock);
527 if (ret < 0) {
528 PERROR("kernel consumer err_sock close");
529 }
530 }
531 if (ustconsumer32_data.err_sock >= 0) {
532 ret = close(ustconsumer32_data.err_sock);
533 if (ret < 0) {
534 PERROR("UST consumerd32 err_sock close");
535 }
536 }
537 if (ustconsumer64_data.err_sock >= 0) {
538 ret = close(ustconsumer64_data.err_sock);
539 if (ret < 0) {
540 PERROR("UST consumerd64 err_sock close");
541 }
542 }
543 if (kconsumer_data.cmd_sock >= 0) {
544 ret = close(kconsumer_data.cmd_sock);
545 if (ret < 0) {
546 PERROR("kernel consumer cmd_sock close");
547 }
548 }
549 if (ustconsumer32_data.cmd_sock >= 0) {
550 ret = close(ustconsumer32_data.cmd_sock);
551 if (ret < 0) {
552 PERROR("UST consumerd32 cmd_sock close");
553 }
554 }
555 if (ustconsumer64_data.cmd_sock >= 0) {
556 ret = close(ustconsumer64_data.cmd_sock);
557 if (ret < 0) {
558 PERROR("UST consumerd64 cmd_sock close");
559 }
560 }
561 }
562
563 /*
564 * Generate the full lock file path using the rundir.
565 *
566 * Return the snprintf() return value thus a negative value is an error.
567 */
568 static int generate_lock_file_path(char *path, size_t len)
569 {
570 int ret;
571
572 assert(path);
573 assert(rundir);
574
575 /* Build lockfile path from rundir. */
576 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
577 if (ret < 0) {
578 PERROR("snprintf lockfile path");
579 }
580
581 return ret;
582 }
583
584 /*
585 * Cleanup the session daemon's data structures.
586 */
587 static void sessiond_cleanup(void)
588 {
589 int ret;
590 struct ltt_session *sess, *stmp;
591 char path[PATH_MAX];
592
593 DBG("Cleanup sessiond");
594
595 /*
596 * Close the thread quit pipe. It has already done its job,
597 * since we are now called.
598 */
599 utils_close_pipe(thread_quit_pipe);
600
601 /*
602 * If opt_pidfile is undefined, the default file will be wiped when
603 * removing the rundir.
604 */
605 if (opt_pidfile) {
606 ret = remove(opt_pidfile);
607 if (ret < 0) {
608 PERROR("remove pidfile %s", opt_pidfile);
609 }
610 }
611
612 DBG("Removing sessiond and consumerd content of directory %s", rundir);
613
614 /* sessiond */
615 snprintf(path, PATH_MAX,
616 "%s/%s",
617 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
618 DBG("Removing %s", path);
619 (void) unlink(path);
620
621 snprintf(path, PATH_MAX, "%s/%s", rundir,
622 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
623 DBG("Removing %s", path);
624 (void) unlink(path);
625
626 /* kconsumerd */
627 snprintf(path, PATH_MAX,
628 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
629 rundir);
630 DBG("Removing %s", path);
631 (void) unlink(path);
632
633 snprintf(path, PATH_MAX,
634 DEFAULT_KCONSUMERD_PATH,
635 rundir);
636 DBG("Removing directory %s", path);
637 (void) rmdir(path);
638
639 /* ust consumerd 32 */
640 snprintf(path, PATH_MAX,
641 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
642 rundir);
643 DBG("Removing %s", path);
644 (void) unlink(path);
645
646 snprintf(path, PATH_MAX,
647 DEFAULT_USTCONSUMERD32_PATH,
648 rundir);
649 DBG("Removing directory %s", path);
650 (void) rmdir(path);
651
652 /* ust consumerd 64 */
653 snprintf(path, PATH_MAX,
654 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
655 rundir);
656 DBG("Removing %s", path);
657 (void) unlink(path);
658
659 snprintf(path, PATH_MAX,
660 DEFAULT_USTCONSUMERD64_PATH,
661 rundir);
662 DBG("Removing directory %s", path);
663 (void) rmdir(path);
664
665 DBG("Cleaning up all sessions");
666
667 /* Destroy session list mutex */
668 if (session_list_ptr != NULL) {
669 pthread_mutex_destroy(&session_list_ptr->lock);
670
671 /* Cleanup ALL session */
672 cds_list_for_each_entry_safe(sess, stmp,
673 &session_list_ptr->head, list) {
674 cmd_destroy_session(sess, kernel_poll_pipe[1]);
675 }
676 }
677
678 DBG("Closing all UST sockets");
679 ust_app_clean_list();
680 buffer_reg_destroy_registries();
681
682 if (is_root && !opt_no_kernel) {
683 DBG2("Closing kernel fd");
684 if (kernel_tracer_fd >= 0) {
685 ret = close(kernel_tracer_fd);
686 if (ret) {
687 PERROR("close");
688 }
689 }
690 DBG("Unloading kernel modules");
691 modprobe_remove_lttng_all();
692 free(syscall_table);
693 }
694
695 close_consumer_sockets();
696
697 if (load_info) {
698 load_session_destroy_data(load_info);
699 free(load_info);
700 }
701
702 /*
703 * Cleanup lock file by deleting it and finaly closing it which will
704 * release the file system lock.
705 */
706 if (lockfile_fd >= 0) {
707 char lockfile_path[PATH_MAX];
708
709 ret = generate_lock_file_path(lockfile_path,
710 sizeof(lockfile_path));
711 if (ret > 0) {
712 ret = remove(lockfile_path);
713 if (ret < 0) {
714 PERROR("remove lock file");
715 }
716 ret = close(lockfile_fd);
717 if (ret < 0) {
718 PERROR("close lock file");
719 }
720 }
721 }
722
723 /*
724 * We do NOT rmdir rundir because there are other processes
725 * using it, for instance lttng-relayd, which can start in
726 * parallel with this teardown.
727 */
728
729 free(rundir);
730 }
731
732 /*
733 * Cleanup the daemon's option data structures.
734 */
735 static void sessiond_cleanup_options(void)
736 {
737 DBG("Cleaning up options");
738
739 /*
740 * If the override option is set, the pointer points to a *non* const
741 * thus freeing it even though the variable type is set to const.
742 */
743 if (tracing_group_name_override) {
744 free((void *) tracing_group_name);
745 }
746 if (consumerd32_bin_override) {
747 free((void *) consumerd32_bin);
748 }
749 if (consumerd64_bin_override) {
750 free((void *) consumerd64_bin);
751 }
752 if (consumerd32_libdir_override) {
753 free((void *) consumerd32_libdir);
754 }
755 if (consumerd64_libdir_override) {
756 free((void *) consumerd64_libdir);
757 }
758
759 free(opt_pidfile);
760 free(opt_load_session_path);
761 free(kmod_probes_list);
762 free(kmod_extra_probes_list);
763
764 /* <fun> */
765 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
766 "Matthew, BEET driven development works!%c[%dm",
767 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
768 /* </fun> */
769 }
770
771 /*
772 * Send data on a unix socket using the liblttsessiondcomm API.
773 *
774 * Return lttcomm error code.
775 */
776 static int send_unix_sock(int sock, void *buf, size_t len)
777 {
778 /* Check valid length */
779 if (len == 0) {
780 return -1;
781 }
782
783 return lttcomm_send_unix_sock(sock, buf, len);
784 }
785
786 /*
787 * Free memory of a command context structure.
788 */
789 static void clean_command_ctx(struct command_ctx **cmd_ctx)
790 {
791 DBG("Clean command context structure");
792 if (*cmd_ctx) {
793 if ((*cmd_ctx)->llm) {
794 free((*cmd_ctx)->llm);
795 }
796 if ((*cmd_ctx)->lsm) {
797 free((*cmd_ctx)->lsm);
798 }
799 free(*cmd_ctx);
800 *cmd_ctx = NULL;
801 }
802 }
803
804 /*
805 * Notify UST applications using the shm mmap futex.
806 */
807 static int notify_ust_apps(int active)
808 {
809 char *wait_shm_mmap;
810
811 DBG("Notifying applications of session daemon state: %d", active);
812
813 /* See shm.c for this call implying mmap, shm and futex calls */
814 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
815 if (wait_shm_mmap == NULL) {
816 goto error;
817 }
818
819 /* Wake waiting process */
820 futex_wait_update((int32_t *) wait_shm_mmap, active);
821
822 /* Apps notified successfully */
823 return 0;
824
825 error:
826 return -1;
827 }
828
829 /*
830 * Setup the outgoing data buffer for the response (llm) by allocating the
831 * right amount of memory and copying the original information from the lsm
832 * structure.
833 *
834 * Return total size of the buffer pointed by buf.
835 */
836 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
837 {
838 int ret, buf_size;
839
840 buf_size = size;
841
842 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
843 if (cmd_ctx->llm == NULL) {
844 PERROR("zmalloc");
845 ret = -ENOMEM;
846 goto error;
847 }
848
849 /* Copy common data */
850 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
851 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
852
853 cmd_ctx->llm->data_size = size;
854 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
855
856 return buf_size;
857
858 error:
859 return ret;
860 }
861
862 /*
863 * Update the kernel poll set of all channel fd available over all tracing
864 * session. Add the wakeup pipe at the end of the set.
865 */
866 static int update_kernel_poll(struct lttng_poll_event *events)
867 {
868 int ret;
869 struct ltt_session *session;
870 struct ltt_kernel_channel *channel;
871
872 DBG("Updating kernel poll set");
873
874 session_lock_list();
875 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
876 session_lock(session);
877 if (session->kernel_session == NULL) {
878 session_unlock(session);
879 continue;
880 }
881
882 cds_list_for_each_entry(channel,
883 &session->kernel_session->channel_list.head, list) {
884 /* Add channel fd to the kernel poll set */
885 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
886 if (ret < 0) {
887 session_unlock(session);
888 goto error;
889 }
890 DBG("Channel fd %d added to kernel set", channel->fd);
891 }
892 session_unlock(session);
893 }
894 session_unlock_list();
895
896 return 0;
897
898 error:
899 session_unlock_list();
900 return -1;
901 }
902
903 /*
904 * Find the channel fd from 'fd' over all tracing session. When found, check
905 * for new channel stream and send those stream fds to the kernel consumer.
906 *
907 * Useful for CPU hotplug feature.
908 */
909 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
910 {
911 int ret = 0;
912 struct ltt_session *session;
913 struct ltt_kernel_session *ksess;
914 struct ltt_kernel_channel *channel;
915
916 DBG("Updating kernel streams for channel fd %d", fd);
917
918 session_lock_list();
919 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
920 session_lock(session);
921 if (session->kernel_session == NULL) {
922 session_unlock(session);
923 continue;
924 }
925 ksess = session->kernel_session;
926
927 cds_list_for_each_entry(channel,
928 &ksess->channel_list.head, list) {
929 struct lttng_ht_iter iter;
930 struct consumer_socket *socket;
931
932 if (channel->fd != fd) {
933 continue;
934 }
935 DBG("Channel found, updating kernel streams");
936 ret = kernel_open_channel_stream(channel);
937 if (ret < 0) {
938 goto error;
939 }
940 /* Update the stream global counter */
941 ksess->stream_count_global += ret;
942
943 /*
944 * Have we already sent fds to the consumer? If yes, it
945 * means that tracing is started so it is safe to send
946 * our updated stream fds.
947 */
948 if (ksess->consumer_fds_sent != 1
949 || ksess->consumer == NULL) {
950 ret = -1;
951 goto error;
952 }
953
954 rcu_read_lock();
955 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
956 &iter.iter, socket, node.node) {
957 pthread_mutex_lock(socket->lock);
958 ret = kernel_consumer_send_channel_stream(socket,
959 channel, ksess,
960 session->output_traces ? 1 : 0);
961 pthread_mutex_unlock(socket->lock);
962 if (ret < 0) {
963 rcu_read_unlock();
964 goto error;
965 }
966 }
967 rcu_read_unlock();
968 }
969 session_unlock(session);
970 }
971 session_unlock_list();
972 return ret;
973
974 error:
975 session_unlock(session);
976 session_unlock_list();
977 return ret;
978 }
979
980 /*
981 * For each tracing session, update newly registered apps. The session list
982 * lock MUST be acquired before calling this.
983 */
984 static void update_ust_app(int app_sock)
985 {
986 struct ltt_session *sess, *stmp;
987
988 /* Consumer is in an ERROR state. Stop any application update. */
989 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
990 /* Stop the update process since the consumer is dead. */
991 return;
992 }
993
994 /* For all tracing session(s) */
995 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
996 session_lock(sess);
997 if (sess->ust_session) {
998 ust_app_global_update(sess->ust_session, app_sock);
999 }
1000 session_unlock(sess);
1001 }
1002 }
1003
1004 /*
1005 * This thread manage event coming from the kernel.
1006 *
1007 * Features supported in this thread:
1008 * -) CPU Hotplug
1009 */
1010 static void *thread_manage_kernel(void *data)
1011 {
1012 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1013 uint32_t revents, nb_fd;
1014 char tmp;
1015 struct lttng_poll_event events;
1016
1017 DBG("[thread] Thread manage kernel started");
1018
1019 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1020
1021 /*
1022 * This first step of the while is to clean this structure which could free
1023 * non NULL pointers so initialize it before the loop.
1024 */
1025 lttng_poll_init(&events);
1026
1027 if (testpoint(sessiond_thread_manage_kernel)) {
1028 goto error_testpoint;
1029 }
1030
1031 health_code_update();
1032
1033 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1034 goto error_testpoint;
1035 }
1036
1037 while (1) {
1038 health_code_update();
1039
1040 if (update_poll_flag == 1) {
1041 /* Clean events object. We are about to populate it again. */
1042 lttng_poll_clean(&events);
1043
1044 ret = sessiond_set_thread_pollset(&events, 2);
1045 if (ret < 0) {
1046 goto error_poll_create;
1047 }
1048
1049 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1050 if (ret < 0) {
1051 goto error;
1052 }
1053
1054 /* This will add the available kernel channel if any. */
1055 ret = update_kernel_poll(&events);
1056 if (ret < 0) {
1057 goto error;
1058 }
1059 update_poll_flag = 0;
1060 }
1061
1062 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
1063
1064 /* Poll infinite value of time */
1065 restart:
1066 health_poll_entry();
1067 ret = lttng_poll_wait(&events, -1);
1068 health_poll_exit();
1069 if (ret < 0) {
1070 /*
1071 * Restart interrupted system call.
1072 */
1073 if (errno == EINTR) {
1074 goto restart;
1075 }
1076 goto error;
1077 } else if (ret == 0) {
1078 /* Should not happen since timeout is infinite */
1079 ERR("Return value of poll is 0 with an infinite timeout.\n"
1080 "This should not have happened! Continuing...");
1081 continue;
1082 }
1083
1084 nb_fd = ret;
1085
1086 for (i = 0; i < nb_fd; i++) {
1087 /* Fetch once the poll data */
1088 revents = LTTNG_POLL_GETEV(&events, i);
1089 pollfd = LTTNG_POLL_GETFD(&events, i);
1090
1091 health_code_update();
1092
1093 /* Thread quit pipe has been closed. Killing thread. */
1094 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1095 if (ret) {
1096 err = 0;
1097 goto exit;
1098 }
1099
1100 /* Check for data on kernel pipe */
1101 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1102 (void) lttng_read(kernel_poll_pipe[0],
1103 &tmp, 1);
1104 /*
1105 * Ret value is useless here, if this pipe gets any actions an
1106 * update is required anyway.
1107 */
1108 update_poll_flag = 1;
1109 continue;
1110 } else {
1111 /*
1112 * New CPU detected by the kernel. Adding kernel stream to
1113 * kernel session and updating the kernel consumer
1114 */
1115 if (revents & LPOLLIN) {
1116 ret = update_kernel_stream(&kconsumer_data, pollfd);
1117 if (ret < 0) {
1118 continue;
1119 }
1120 break;
1121 /*
1122 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1123 * and unregister kernel stream at this point.
1124 */
1125 }
1126 }
1127 }
1128 }
1129
1130 exit:
1131 error:
1132 lttng_poll_clean(&events);
1133 error_poll_create:
1134 error_testpoint:
1135 utils_close_pipe(kernel_poll_pipe);
1136 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1137 if (err) {
1138 health_error();
1139 ERR("Health error occurred in %s", __func__);
1140 WARN("Kernel thread died unexpectedly. "
1141 "Kernel tracing can continue but CPU hotplug is disabled.");
1142 }
1143 health_unregister(health_sessiond);
1144 DBG("Kernel thread dying");
1145 return NULL;
1146 }
1147
1148 /*
1149 * Signal pthread condition of the consumer data that the thread.
1150 */
1151 static void signal_consumer_condition(struct consumer_data *data, int state)
1152 {
1153 pthread_mutex_lock(&data->cond_mutex);
1154
1155 /*
1156 * The state is set before signaling. It can be any value, it's the waiter
1157 * job to correctly interpret this condition variable associated to the
1158 * consumer pthread_cond.
1159 *
1160 * A value of 0 means that the corresponding thread of the consumer data
1161 * was not started. 1 indicates that the thread has started and is ready
1162 * for action. A negative value means that there was an error during the
1163 * thread bootstrap.
1164 */
1165 data->consumer_thread_is_ready = state;
1166 (void) pthread_cond_signal(&data->cond);
1167
1168 pthread_mutex_unlock(&data->cond_mutex);
1169 }
1170
1171 /*
1172 * This thread manage the consumer error sent back to the session daemon.
1173 */
1174 static void *thread_manage_consumer(void *data)
1175 {
1176 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1177 uint32_t revents, nb_fd;
1178 enum lttcomm_return_code code;
1179 struct lttng_poll_event events;
1180 struct consumer_data *consumer_data = data;
1181
1182 DBG("[thread] Manage consumer started");
1183
1184 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1185
1186 health_code_update();
1187
1188 /*
1189 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1190 * metadata_sock. Nothing more will be added to this poll set.
1191 */
1192 ret = sessiond_set_thread_pollset(&events, 3);
1193 if (ret < 0) {
1194 goto error_poll;
1195 }
1196
1197 /*
1198 * The error socket here is already in a listening state which was done
1199 * just before spawning this thread to avoid a race between the consumer
1200 * daemon exec trying to connect and the listen() call.
1201 */
1202 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1203 if (ret < 0) {
1204 goto error;
1205 }
1206
1207 health_code_update();
1208
1209 /* Infinite blocking call, waiting for transmission */
1210 restart:
1211 health_poll_entry();
1212
1213 if (testpoint(sessiond_thread_manage_consumer)) {
1214 goto error;
1215 }
1216
1217 ret = lttng_poll_wait(&events, -1);
1218 health_poll_exit();
1219 if (ret < 0) {
1220 /*
1221 * Restart interrupted system call.
1222 */
1223 if (errno == EINTR) {
1224 goto restart;
1225 }
1226 goto error;
1227 }
1228
1229 nb_fd = ret;
1230
1231 for (i = 0; i < nb_fd; i++) {
1232 /* Fetch once the poll data */
1233 revents = LTTNG_POLL_GETEV(&events, i);
1234 pollfd = LTTNG_POLL_GETFD(&events, i);
1235
1236 health_code_update();
1237
1238 /* Thread quit pipe has been closed. Killing thread. */
1239 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1240 if (ret) {
1241 err = 0;
1242 goto exit;
1243 }
1244
1245 /* Event on the registration socket */
1246 if (pollfd == consumer_data->err_sock) {
1247 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1248 ERR("consumer err socket poll error");
1249 goto error;
1250 }
1251 }
1252 }
1253
1254 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1255 if (sock < 0) {
1256 goto error;
1257 }
1258
1259 /*
1260 * Set the CLOEXEC flag. Return code is useless because either way, the
1261 * show must go on.
1262 */
1263 (void) utils_set_fd_cloexec(sock);
1264
1265 health_code_update();
1266
1267 DBG2("Receiving code from consumer err_sock");
1268
1269 /* Getting status code from kconsumerd */
1270 ret = lttcomm_recv_unix_sock(sock, &code,
1271 sizeof(enum lttcomm_return_code));
1272 if (ret <= 0) {
1273 goto error;
1274 }
1275
1276 health_code_update();
1277 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1278 /* Connect both socket, command and metadata. */
1279 consumer_data->cmd_sock =
1280 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1281 consumer_data->metadata_fd =
1282 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1283 if (consumer_data->cmd_sock < 0
1284 || consumer_data->metadata_fd < 0) {
1285 PERROR("consumer connect cmd socket");
1286 /* On error, signal condition and quit. */
1287 signal_consumer_condition(consumer_data, -1);
1288 goto error;
1289 }
1290 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1291 /* Create metadata socket lock. */
1292 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1293 if (consumer_data->metadata_sock.lock == NULL) {
1294 PERROR("zmalloc pthread mutex");
1295 ret = -1;
1296 goto error;
1297 }
1298 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1299
1300 signal_consumer_condition(consumer_data, 1);
1301 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1302 DBG("Consumer metadata socket ready (fd: %d)",
1303 consumer_data->metadata_fd);
1304 } else {
1305 ERR("consumer error when waiting for SOCK_READY : %s",
1306 lttcomm_get_readable_code(-code));
1307 goto error;
1308 }
1309
1310 /* Remove the consumerd error sock since we've established a connexion */
1311 ret = lttng_poll_del(&events, consumer_data->err_sock);
1312 if (ret < 0) {
1313 goto error;
1314 }
1315
1316 /* Add new accepted error socket. */
1317 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1318 if (ret < 0) {
1319 goto error;
1320 }
1321
1322 /* Add metadata socket that is successfully connected. */
1323 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1324 LPOLLIN | LPOLLRDHUP);
1325 if (ret < 0) {
1326 goto error;
1327 }
1328
1329 health_code_update();
1330
1331 /* Infinite blocking call, waiting for transmission */
1332 restart_poll:
1333 while (1) {
1334 health_code_update();
1335
1336 /* Exit the thread because the thread quit pipe has been triggered. */
1337 if (should_quit) {
1338 /* Not a health error. */
1339 err = 0;
1340 goto exit;
1341 }
1342
1343 health_poll_entry();
1344 ret = lttng_poll_wait(&events, -1);
1345 health_poll_exit();
1346 if (ret < 0) {
1347 /*
1348 * Restart interrupted system call.
1349 */
1350 if (errno == EINTR) {
1351 goto restart_poll;
1352 }
1353 goto error;
1354 }
1355
1356 nb_fd = ret;
1357
1358 for (i = 0; i < nb_fd; i++) {
1359 /* Fetch once the poll data */
1360 revents = LTTNG_POLL_GETEV(&events, i);
1361 pollfd = LTTNG_POLL_GETFD(&events, i);
1362
1363 health_code_update();
1364
1365 /*
1366 * Thread quit pipe has been triggered, flag that we should stop
1367 * but continue the current loop to handle potential data from
1368 * consumer.
1369 */
1370 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1371
1372 if (pollfd == sock) {
1373 /* Event on the consumerd socket */
1374 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1375 ERR("consumer err socket second poll error");
1376 goto error;
1377 }
1378 health_code_update();
1379 /* Wait for any kconsumerd error */
1380 ret = lttcomm_recv_unix_sock(sock, &code,
1381 sizeof(enum lttcomm_return_code));
1382 if (ret <= 0) {
1383 ERR("consumer closed the command socket");
1384 goto error;
1385 }
1386
1387 ERR("consumer return code : %s",
1388 lttcomm_get_readable_code(-code));
1389
1390 goto exit;
1391 } else if (pollfd == consumer_data->metadata_fd) {
1392 /* UST metadata requests */
1393 ret = ust_consumer_metadata_request(
1394 &consumer_data->metadata_sock);
1395 if (ret < 0) {
1396 ERR("Handling metadata request");
1397 goto error;
1398 }
1399 }
1400 /* No need for an else branch all FDs are tested prior. */
1401 }
1402 health_code_update();
1403 }
1404
1405 exit:
1406 error:
1407 /*
1408 * We lock here because we are about to close the sockets and some other
1409 * thread might be using them so get exclusive access which will abort all
1410 * other consumer command by other threads.
1411 */
1412 pthread_mutex_lock(&consumer_data->lock);
1413
1414 /* Immediately set the consumerd state to stopped */
1415 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1416 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1417 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1418 consumer_data->type == LTTNG_CONSUMER32_UST) {
1419 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1420 } else {
1421 /* Code flow error... */
1422 assert(0);
1423 }
1424
1425 if (consumer_data->err_sock >= 0) {
1426 ret = close(consumer_data->err_sock);
1427 if (ret) {
1428 PERROR("close");
1429 }
1430 consumer_data->err_sock = -1;
1431 }
1432 if (consumer_data->cmd_sock >= 0) {
1433 ret = close(consumer_data->cmd_sock);
1434 if (ret) {
1435 PERROR("close");
1436 }
1437 consumer_data->cmd_sock = -1;
1438 }
1439 if (consumer_data->metadata_sock.fd_ptr &&
1440 *consumer_data->metadata_sock.fd_ptr >= 0) {
1441 ret = close(*consumer_data->metadata_sock.fd_ptr);
1442 if (ret) {
1443 PERROR("close");
1444 }
1445 }
1446 if (sock >= 0) {
1447 ret = close(sock);
1448 if (ret) {
1449 PERROR("close");
1450 }
1451 }
1452
1453 unlink(consumer_data->err_unix_sock_path);
1454 unlink(consumer_data->cmd_unix_sock_path);
1455 consumer_data->pid = 0;
1456 pthread_mutex_unlock(&consumer_data->lock);
1457
1458 /* Cleanup metadata socket mutex. */
1459 if (consumer_data->metadata_sock.lock) {
1460 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1461 free(consumer_data->metadata_sock.lock);
1462 }
1463 lttng_poll_clean(&events);
1464 error_poll:
1465 if (err) {
1466 health_error();
1467 ERR("Health error occurred in %s", __func__);
1468 }
1469 health_unregister(health_sessiond);
1470 DBG("consumer thread cleanup completed");
1471
1472 return NULL;
1473 }
1474
1475 /*
1476 * This thread manage application communication.
1477 */
1478 static void *thread_manage_apps(void *data)
1479 {
1480 int i, ret, pollfd, err = -1;
1481 ssize_t size_ret;
1482 uint32_t revents, nb_fd;
1483 struct lttng_poll_event events;
1484
1485 DBG("[thread] Manage application started");
1486
1487 rcu_register_thread();
1488 rcu_thread_online();
1489
1490 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1491
1492 if (testpoint(sessiond_thread_manage_apps)) {
1493 goto error_testpoint;
1494 }
1495
1496 health_code_update();
1497
1498 ret = sessiond_set_thread_pollset(&events, 2);
1499 if (ret < 0) {
1500 goto error_poll_create;
1501 }
1502
1503 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1504 if (ret < 0) {
1505 goto error;
1506 }
1507
1508 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1509 goto error;
1510 }
1511
1512 health_code_update();
1513
1514 while (1) {
1515 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1516
1517 /* Inifinite blocking call, waiting for transmission */
1518 restart:
1519 health_poll_entry();
1520 ret = lttng_poll_wait(&events, -1);
1521 health_poll_exit();
1522 if (ret < 0) {
1523 /*
1524 * Restart interrupted system call.
1525 */
1526 if (errno == EINTR) {
1527 goto restart;
1528 }
1529 goto error;
1530 }
1531
1532 nb_fd = ret;
1533
1534 for (i = 0; i < nb_fd; i++) {
1535 /* Fetch once the poll data */
1536 revents = LTTNG_POLL_GETEV(&events, i);
1537 pollfd = LTTNG_POLL_GETFD(&events, i);
1538
1539 health_code_update();
1540
1541 /* Thread quit pipe has been closed. Killing thread. */
1542 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1543 if (ret) {
1544 err = 0;
1545 goto exit;
1546 }
1547
1548 /* Inspect the apps cmd pipe */
1549 if (pollfd == apps_cmd_pipe[0]) {
1550 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1551 ERR("Apps command pipe error");
1552 goto error;
1553 } else if (revents & LPOLLIN) {
1554 int sock;
1555
1556 /* Empty pipe */
1557 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1558 if (size_ret < sizeof(sock)) {
1559 PERROR("read apps cmd pipe");
1560 goto error;
1561 }
1562
1563 health_code_update();
1564
1565 /*
1566 * We only monitor the error events of the socket. This
1567 * thread does not handle any incoming data from UST
1568 * (POLLIN).
1569 */
1570 ret = lttng_poll_add(&events, sock,
1571 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1572 if (ret < 0) {
1573 goto error;
1574 }
1575
1576 DBG("Apps with sock %d added to poll set", sock);
1577 }
1578 } else {
1579 /*
1580 * At this point, we know that a registered application made
1581 * the event at poll_wait.
1582 */
1583 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1584 /* Removing from the poll set */
1585 ret = lttng_poll_del(&events, pollfd);
1586 if (ret < 0) {
1587 goto error;
1588 }
1589
1590 /* Socket closed on remote end. */
1591 ust_app_unregister(pollfd);
1592 }
1593 }
1594
1595 health_code_update();
1596 }
1597 }
1598
1599 exit:
1600 error:
1601 lttng_poll_clean(&events);
1602 error_poll_create:
1603 error_testpoint:
1604 utils_close_pipe(apps_cmd_pipe);
1605 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1606
1607 /*
1608 * We don't clean the UST app hash table here since already registered
1609 * applications can still be controlled so let them be until the session
1610 * daemon dies or the applications stop.
1611 */
1612
1613 if (err) {
1614 health_error();
1615 ERR("Health error occurred in %s", __func__);
1616 }
1617 health_unregister(health_sessiond);
1618 DBG("Application communication apps thread cleanup complete");
1619 rcu_thread_offline();
1620 rcu_unregister_thread();
1621 return NULL;
1622 }
1623
1624 /*
1625 * Send a socket to a thread This is called from the dispatch UST registration
1626 * thread once all sockets are set for the application.
1627 *
1628 * The sock value can be invalid, we don't really care, the thread will handle
1629 * it and make the necessary cleanup if so.
1630 *
1631 * On success, return 0 else a negative value being the errno message of the
1632 * write().
1633 */
1634 static int send_socket_to_thread(int fd, int sock)
1635 {
1636 ssize_t ret;
1637
1638 /*
1639 * It's possible that the FD is set as invalid with -1 concurrently just
1640 * before calling this function being a shutdown state of the thread.
1641 */
1642 if (fd < 0) {
1643 ret = -EBADF;
1644 goto error;
1645 }
1646
1647 ret = lttng_write(fd, &sock, sizeof(sock));
1648 if (ret < sizeof(sock)) {
1649 PERROR("write apps pipe %d", fd);
1650 if (ret < 0) {
1651 ret = -errno;
1652 }
1653 goto error;
1654 }
1655
1656 /* All good. Don't send back the write positive ret value. */
1657 ret = 0;
1658 error:
1659 return (int) ret;
1660 }
1661
1662 /*
1663 * Sanitize the wait queue of the dispatch registration thread meaning removing
1664 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1665 * notify socket is never received.
1666 */
1667 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1668 {
1669 int ret, nb_fd = 0, i;
1670 unsigned int fd_added = 0;
1671 struct lttng_poll_event events;
1672 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1673
1674 assert(wait_queue);
1675
1676 lttng_poll_init(&events);
1677
1678 /* Just skip everything for an empty queue. */
1679 if (!wait_queue->count) {
1680 goto end;
1681 }
1682
1683 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1684 if (ret < 0) {
1685 goto error_create;
1686 }
1687
1688 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1689 &wait_queue->head, head) {
1690 assert(wait_node->app);
1691 ret = lttng_poll_add(&events, wait_node->app->sock,
1692 LPOLLHUP | LPOLLERR);
1693 if (ret < 0) {
1694 goto error;
1695 }
1696
1697 fd_added = 1;
1698 }
1699
1700 if (!fd_added) {
1701 goto end;
1702 }
1703
1704 /*
1705 * Poll but don't block so we can quickly identify the faulty events and
1706 * clean them afterwards from the wait queue.
1707 */
1708 ret = lttng_poll_wait(&events, 0);
1709 if (ret < 0) {
1710 goto error;
1711 }
1712 nb_fd = ret;
1713
1714 for (i = 0; i < nb_fd; i++) {
1715 /* Get faulty FD. */
1716 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1717 int pollfd = LTTNG_POLL_GETFD(&events, i);
1718
1719 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1720 &wait_queue->head, head) {
1721 if (pollfd == wait_node->app->sock &&
1722 (revents & (LPOLLHUP | LPOLLERR))) {
1723 cds_list_del(&wait_node->head);
1724 wait_queue->count--;
1725 ust_app_destroy(wait_node->app);
1726 free(wait_node);
1727 break;
1728 }
1729 }
1730 }
1731
1732 if (nb_fd > 0) {
1733 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1734 }
1735
1736 end:
1737 lttng_poll_clean(&events);
1738 return;
1739
1740 error:
1741 lttng_poll_clean(&events);
1742 error_create:
1743 ERR("Unable to sanitize wait queue");
1744 return;
1745 }
1746
1747 /*
1748 * Dispatch request from the registration threads to the application
1749 * communication thread.
1750 */
1751 static void *thread_dispatch_ust_registration(void *data)
1752 {
1753 int ret, err = -1;
1754 struct cds_wfcq_node *node;
1755 struct ust_command *ust_cmd = NULL;
1756 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1757 struct ust_reg_wait_queue wait_queue = {
1758 .count = 0,
1759 };
1760
1761 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1762
1763 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1764 goto error_testpoint;
1765 }
1766
1767 health_code_update();
1768
1769 CDS_INIT_LIST_HEAD(&wait_queue.head);
1770
1771 DBG("[thread] Dispatch UST command started");
1772
1773 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1774 health_code_update();
1775
1776 /* Atomically prepare the queue futex */
1777 futex_nto1_prepare(&ust_cmd_queue.futex);
1778
1779 do {
1780 struct ust_app *app = NULL;
1781 ust_cmd = NULL;
1782
1783 /*
1784 * Make sure we don't have node(s) that have hung up before receiving
1785 * the notify socket. This is to clean the list in order to avoid
1786 * memory leaks from notify socket that are never seen.
1787 */
1788 sanitize_wait_queue(&wait_queue);
1789
1790 health_code_update();
1791 /* Dequeue command for registration */
1792 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1793 if (node == NULL) {
1794 DBG("Woken up but nothing in the UST command queue");
1795 /* Continue thread execution */
1796 break;
1797 }
1798
1799 ust_cmd = caa_container_of(node, struct ust_command, node);
1800
1801 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1802 " gid:%d sock:%d name:%s (version %d.%d)",
1803 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1804 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1805 ust_cmd->sock, ust_cmd->reg_msg.name,
1806 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1807
1808 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1809 wait_node = zmalloc(sizeof(*wait_node));
1810 if (!wait_node) {
1811 PERROR("zmalloc wait_node dispatch");
1812 ret = close(ust_cmd->sock);
1813 if (ret < 0) {
1814 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1815 }
1816 lttng_fd_put(LTTNG_FD_APPS, 1);
1817 free(ust_cmd);
1818 goto error;
1819 }
1820 CDS_INIT_LIST_HEAD(&wait_node->head);
1821
1822 /* Create application object if socket is CMD. */
1823 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1824 ust_cmd->sock);
1825 if (!wait_node->app) {
1826 ret = close(ust_cmd->sock);
1827 if (ret < 0) {
1828 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1829 }
1830 lttng_fd_put(LTTNG_FD_APPS, 1);
1831 free(wait_node);
1832 free(ust_cmd);
1833 continue;
1834 }
1835 /*
1836 * Add application to the wait queue so we can set the notify
1837 * socket before putting this object in the global ht.
1838 */
1839 cds_list_add(&wait_node->head, &wait_queue.head);
1840 wait_queue.count++;
1841
1842 free(ust_cmd);
1843 /*
1844 * We have to continue here since we don't have the notify
1845 * socket and the application MUST be added to the hash table
1846 * only at that moment.
1847 */
1848 continue;
1849 } else {
1850 /*
1851 * Look for the application in the local wait queue and set the
1852 * notify socket if found.
1853 */
1854 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1855 &wait_queue.head, head) {
1856 health_code_update();
1857 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1858 wait_node->app->notify_sock = ust_cmd->sock;
1859 cds_list_del(&wait_node->head);
1860 wait_queue.count--;
1861 app = wait_node->app;
1862 free(wait_node);
1863 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1864 break;
1865 }
1866 }
1867
1868 /*
1869 * With no application at this stage the received socket is
1870 * basically useless so close it before we free the cmd data
1871 * structure for good.
1872 */
1873 if (!app) {
1874 ret = close(ust_cmd->sock);
1875 if (ret < 0) {
1876 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1877 }
1878 lttng_fd_put(LTTNG_FD_APPS, 1);
1879 }
1880 free(ust_cmd);
1881 }
1882
1883 if (app) {
1884 /*
1885 * @session_lock_list
1886 *
1887 * Lock the global session list so from the register up to the
1888 * registration done message, no thread can see the application
1889 * and change its state.
1890 */
1891 session_lock_list();
1892 rcu_read_lock();
1893
1894 /*
1895 * Add application to the global hash table. This needs to be
1896 * done before the update to the UST registry can locate the
1897 * application.
1898 */
1899 ust_app_add(app);
1900
1901 /* Set app version. This call will print an error if needed. */
1902 (void) ust_app_version(app);
1903
1904 /* Send notify socket through the notify pipe. */
1905 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1906 app->notify_sock);
1907 if (ret < 0) {
1908 rcu_read_unlock();
1909 session_unlock_list();
1910 /*
1911 * No notify thread, stop the UST tracing. However, this is
1912 * not an internal error of the this thread thus setting
1913 * the health error code to a normal exit.
1914 */
1915 err = 0;
1916 goto error;
1917 }
1918
1919 /*
1920 * Update newly registered application with the tracing
1921 * registry info already enabled information.
1922 */
1923 update_ust_app(app->sock);
1924
1925 /*
1926 * Don't care about return value. Let the manage apps threads
1927 * handle app unregistration upon socket close.
1928 */
1929 (void) ust_app_register_done(app->sock);
1930
1931 /*
1932 * Even if the application socket has been closed, send the app
1933 * to the thread and unregistration will take place at that
1934 * place.
1935 */
1936 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1937 if (ret < 0) {
1938 rcu_read_unlock();
1939 session_unlock_list();
1940 /*
1941 * No apps. thread, stop the UST tracing. However, this is
1942 * not an internal error of the this thread thus setting
1943 * the health error code to a normal exit.
1944 */
1945 err = 0;
1946 goto error;
1947 }
1948
1949 rcu_read_unlock();
1950 session_unlock_list();
1951 }
1952 } while (node != NULL);
1953
1954 health_poll_entry();
1955 /* Futex wait on queue. Blocking call on futex() */
1956 futex_nto1_wait(&ust_cmd_queue.futex);
1957 health_poll_exit();
1958 }
1959 /* Normal exit, no error */
1960 err = 0;
1961
1962 error:
1963 /* Clean up wait queue. */
1964 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1965 &wait_queue.head, head) {
1966 cds_list_del(&wait_node->head);
1967 wait_queue.count--;
1968 free(wait_node);
1969 }
1970
1971 error_testpoint:
1972 DBG("Dispatch thread dying");
1973 if (err) {
1974 health_error();
1975 ERR("Health error occurred in %s", __func__);
1976 }
1977 health_unregister(health_sessiond);
1978 return NULL;
1979 }
1980
1981 /*
1982 * This thread manage application registration.
1983 */
1984 static void *thread_registration_apps(void *data)
1985 {
1986 int sock = -1, i, ret, pollfd, err = -1;
1987 uint32_t revents, nb_fd;
1988 struct lttng_poll_event events;
1989 /*
1990 * Get allocated in this thread, enqueued to a global queue, dequeued and
1991 * freed in the manage apps thread.
1992 */
1993 struct ust_command *ust_cmd = NULL;
1994
1995 DBG("[thread] Manage application registration started");
1996
1997 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1998
1999 if (testpoint(sessiond_thread_registration_apps)) {
2000 goto error_testpoint;
2001 }
2002
2003 ret = lttcomm_listen_unix_sock(apps_sock);
2004 if (ret < 0) {
2005 goto error_listen;
2006 }
2007
2008 /*
2009 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2010 * more will be added to this poll set.
2011 */
2012 ret = sessiond_set_thread_pollset(&events, 2);
2013 if (ret < 0) {
2014 goto error_create_poll;
2015 }
2016
2017 /* Add the application registration socket */
2018 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2019 if (ret < 0) {
2020 goto error_poll_add;
2021 }
2022
2023 /* Notify all applications to register */
2024 ret = notify_ust_apps(1);
2025 if (ret < 0) {
2026 ERR("Failed to notify applications or create the wait shared memory.\n"
2027 "Execution continues but there might be problem for already\n"
2028 "running applications that wishes to register.");
2029 }
2030
2031 while (1) {
2032 DBG("Accepting application registration");
2033
2034 /* Inifinite blocking call, waiting for transmission */
2035 restart:
2036 health_poll_entry();
2037 ret = lttng_poll_wait(&events, -1);
2038 health_poll_exit();
2039 if (ret < 0) {
2040 /*
2041 * Restart interrupted system call.
2042 */
2043 if (errno == EINTR) {
2044 goto restart;
2045 }
2046 goto error;
2047 }
2048
2049 nb_fd = ret;
2050
2051 for (i = 0; i < nb_fd; i++) {
2052 health_code_update();
2053
2054 /* Fetch once the poll data */
2055 revents = LTTNG_POLL_GETEV(&events, i);
2056 pollfd = LTTNG_POLL_GETFD(&events, i);
2057
2058 /* Thread quit pipe has been closed. Killing thread. */
2059 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2060 if (ret) {
2061 err = 0;
2062 goto exit;
2063 }
2064
2065 /* Event on the registration socket */
2066 if (pollfd == apps_sock) {
2067 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2068 ERR("Register apps socket poll error");
2069 goto error;
2070 } else if (revents & LPOLLIN) {
2071 sock = lttcomm_accept_unix_sock(apps_sock);
2072 if (sock < 0) {
2073 goto error;
2074 }
2075
2076 /*
2077 * Set socket timeout for both receiving and ending.
2078 * app_socket_timeout is in seconds, whereas
2079 * lttcomm_setsockopt_rcv_timeout and
2080 * lttcomm_setsockopt_snd_timeout expect msec as
2081 * parameter.
2082 */
2083 (void) lttcomm_setsockopt_rcv_timeout(sock,
2084 app_socket_timeout * 1000);
2085 (void) lttcomm_setsockopt_snd_timeout(sock,
2086 app_socket_timeout * 1000);
2087
2088 /*
2089 * Set the CLOEXEC flag. Return code is useless because
2090 * either way, the show must go on.
2091 */
2092 (void) utils_set_fd_cloexec(sock);
2093
2094 /* Create UST registration command for enqueuing */
2095 ust_cmd = zmalloc(sizeof(struct ust_command));
2096 if (ust_cmd == NULL) {
2097 PERROR("ust command zmalloc");
2098 goto error;
2099 }
2100
2101 /*
2102 * Using message-based transmissions to ensure we don't
2103 * have to deal with partially received messages.
2104 */
2105 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2106 if (ret < 0) {
2107 ERR("Exhausted file descriptors allowed for applications.");
2108 free(ust_cmd);
2109 ret = close(sock);
2110 if (ret) {
2111 PERROR("close");
2112 }
2113 sock = -1;
2114 continue;
2115 }
2116
2117 health_code_update();
2118 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2119 if (ret < 0) {
2120 free(ust_cmd);
2121 /* Close socket of the application. */
2122 ret = close(sock);
2123 if (ret) {
2124 PERROR("close");
2125 }
2126 lttng_fd_put(LTTNG_FD_APPS, 1);
2127 sock = -1;
2128 continue;
2129 }
2130 health_code_update();
2131
2132 ust_cmd->sock = sock;
2133 sock = -1;
2134
2135 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2136 " gid:%d sock:%d name:%s (version %d.%d)",
2137 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2138 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2139 ust_cmd->sock, ust_cmd->reg_msg.name,
2140 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2141
2142 /*
2143 * Lock free enqueue the registration request. The red pill
2144 * has been taken! This apps will be part of the *system*.
2145 */
2146 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2147
2148 /*
2149 * Wake the registration queue futex. Implicit memory
2150 * barrier with the exchange in cds_wfcq_enqueue.
2151 */
2152 futex_nto1_wake(&ust_cmd_queue.futex);
2153 }
2154 }
2155 }
2156 }
2157
2158 exit:
2159 error:
2160 /* Notify that the registration thread is gone */
2161 notify_ust_apps(0);
2162
2163 if (apps_sock >= 0) {
2164 ret = close(apps_sock);
2165 if (ret) {
2166 PERROR("close");
2167 }
2168 }
2169 if (sock >= 0) {
2170 ret = close(sock);
2171 if (ret) {
2172 PERROR("close");
2173 }
2174 lttng_fd_put(LTTNG_FD_APPS, 1);
2175 }
2176 unlink(apps_unix_sock_path);
2177
2178 error_poll_add:
2179 lttng_poll_clean(&events);
2180 error_listen:
2181 error_create_poll:
2182 error_testpoint:
2183 DBG("UST Registration thread cleanup complete");
2184 if (err) {
2185 health_error();
2186 ERR("Health error occurred in %s", __func__);
2187 }
2188 health_unregister(health_sessiond);
2189
2190 return NULL;
2191 }
2192
2193 /*
2194 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2195 * exec or it will fails.
2196 */
2197 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2198 {
2199 int ret, clock_ret;
2200 struct timespec timeout;
2201
2202 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2203 consumer_data->consumer_thread_is_ready = 0;
2204
2205 /* Setup pthread condition */
2206 ret = pthread_condattr_init(&consumer_data->condattr);
2207 if (ret) {
2208 errno = ret;
2209 PERROR("pthread_condattr_init consumer data");
2210 goto error;
2211 }
2212
2213 /*
2214 * Set the monotonic clock in order to make sure we DO NOT jump in time
2215 * between the clock_gettime() call and the timedwait call. See bug #324
2216 * for a more details and how we noticed it.
2217 */
2218 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2219 if (ret) {
2220 errno = ret;
2221 PERROR("pthread_condattr_setclock consumer data");
2222 goto error;
2223 }
2224
2225 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2226 if (ret) {
2227 errno = ret;
2228 PERROR("pthread_cond_init consumer data");
2229 goto error;
2230 }
2231
2232 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2233 consumer_data);
2234 if (ret) {
2235 errno = ret;
2236 PERROR("pthread_create consumer");
2237 ret = -1;
2238 goto error;
2239 }
2240
2241 /* We are about to wait on a pthread condition */
2242 pthread_mutex_lock(&consumer_data->cond_mutex);
2243
2244 /* Get time for sem_timedwait absolute timeout */
2245 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2246 /*
2247 * Set the timeout for the condition timed wait even if the clock gettime
2248 * call fails since we might loop on that call and we want to avoid to
2249 * increment the timeout too many times.
2250 */
2251 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2252
2253 /*
2254 * The following loop COULD be skipped in some conditions so this is why we
2255 * set ret to 0 in order to make sure at least one round of the loop is
2256 * done.
2257 */
2258 ret = 0;
2259
2260 /*
2261 * Loop until the condition is reached or when a timeout is reached. Note
2262 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2263 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2264 * possible. This loop does not take any chances and works with both of
2265 * them.
2266 */
2267 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2268 if (clock_ret < 0) {
2269 PERROR("clock_gettime spawn consumer");
2270 /* Infinite wait for the consumerd thread to be ready */
2271 ret = pthread_cond_wait(&consumer_data->cond,
2272 &consumer_data->cond_mutex);
2273 } else {
2274 ret = pthread_cond_timedwait(&consumer_data->cond,
2275 &consumer_data->cond_mutex, &timeout);
2276 }
2277 }
2278
2279 /* Release the pthread condition */
2280 pthread_mutex_unlock(&consumer_data->cond_mutex);
2281
2282 if (ret != 0) {
2283 errno = ret;
2284 if (ret == ETIMEDOUT) {
2285 int pth_ret;
2286
2287 /*
2288 * Call has timed out so we kill the kconsumerd_thread and return
2289 * an error.
2290 */
2291 ERR("Condition timed out. The consumer thread was never ready."
2292 " Killing it");
2293 pth_ret = pthread_cancel(consumer_data->thread);
2294 if (pth_ret < 0) {
2295 PERROR("pthread_cancel consumer thread");
2296 }
2297 } else {
2298 PERROR("pthread_cond_wait failed consumer thread");
2299 }
2300 /* Caller is expecting a negative value on failure. */
2301 ret = -1;
2302 goto error;
2303 }
2304
2305 pthread_mutex_lock(&consumer_data->pid_mutex);
2306 if (consumer_data->pid == 0) {
2307 ERR("Consumerd did not start");
2308 pthread_mutex_unlock(&consumer_data->pid_mutex);
2309 goto error;
2310 }
2311 pthread_mutex_unlock(&consumer_data->pid_mutex);
2312
2313 return 0;
2314
2315 error:
2316 return ret;
2317 }
2318
2319 /*
2320 * Join consumer thread
2321 */
2322 static int join_consumer_thread(struct consumer_data *consumer_data)
2323 {
2324 void *status;
2325
2326 /* Consumer pid must be a real one. */
2327 if (consumer_data->pid > 0) {
2328 int ret;
2329 ret = kill(consumer_data->pid, SIGTERM);
2330 if (ret) {
2331 PERROR("Error killing consumer daemon");
2332 return ret;
2333 }
2334 return pthread_join(consumer_data->thread, &status);
2335 } else {
2336 return 0;
2337 }
2338 }
2339
2340 /*
2341 * Fork and exec a consumer daemon (consumerd).
2342 *
2343 * Return pid if successful else -1.
2344 */
2345 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2346 {
2347 int ret;
2348 pid_t pid;
2349 const char *consumer_to_use;
2350 const char *verbosity;
2351 struct stat st;
2352
2353 DBG("Spawning consumerd");
2354
2355 pid = fork();
2356 if (pid == 0) {
2357 /*
2358 * Exec consumerd.
2359 */
2360 if (opt_verbose_consumer) {
2361 verbosity = "--verbose";
2362 } else if (lttng_opt_quiet) {
2363 verbosity = "--quiet";
2364 } else {
2365 verbosity = "";
2366 }
2367
2368 switch (consumer_data->type) {
2369 case LTTNG_CONSUMER_KERNEL:
2370 /*
2371 * Find out which consumerd to execute. We will first try the
2372 * 64-bit path, then the sessiond's installation directory, and
2373 * fallback on the 32-bit one,
2374 */
2375 DBG3("Looking for a kernel consumer at these locations:");
2376 DBG3(" 1) %s", consumerd64_bin);
2377 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2378 DBG3(" 3) %s", consumerd32_bin);
2379 if (stat(consumerd64_bin, &st) == 0) {
2380 DBG3("Found location #1");
2381 consumer_to_use = consumerd64_bin;
2382 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2383 DBG3("Found location #2");
2384 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2385 } else if (stat(consumerd32_bin, &st) == 0) {
2386 DBG3("Found location #3");
2387 consumer_to_use = consumerd32_bin;
2388 } else {
2389 DBG("Could not find any valid consumerd executable");
2390 ret = -EINVAL;
2391 break;
2392 }
2393 DBG("Using kernel consumer at: %s", consumer_to_use);
2394 ret = execl(consumer_to_use,
2395 "lttng-consumerd", verbosity, "-k",
2396 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2397 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2398 "--group", tracing_group_name,
2399 NULL);
2400 break;
2401 case LTTNG_CONSUMER64_UST:
2402 {
2403 char *tmpnew = NULL;
2404
2405 if (consumerd64_libdir[0] != '\0') {
2406 char *tmp;
2407 size_t tmplen;
2408
2409 tmp = getenv("LD_LIBRARY_PATH");
2410 if (!tmp) {
2411 tmp = "";
2412 }
2413 tmplen = strlen("LD_LIBRARY_PATH=")
2414 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2415 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2416 if (!tmpnew) {
2417 ret = -ENOMEM;
2418 goto error;
2419 }
2420 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2421 strcat(tmpnew, consumerd64_libdir);
2422 if (tmp[0] != '\0') {
2423 strcat(tmpnew, ":");
2424 strcat(tmpnew, tmp);
2425 }
2426 ret = putenv(tmpnew);
2427 if (ret) {
2428 ret = -errno;
2429 free(tmpnew);
2430 goto error;
2431 }
2432 }
2433 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2434 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2435 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2436 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2437 "--group", tracing_group_name,
2438 NULL);
2439 if (consumerd64_libdir[0] != '\0') {
2440 free(tmpnew);
2441 }
2442 break;
2443 }
2444 case LTTNG_CONSUMER32_UST:
2445 {
2446 char *tmpnew = NULL;
2447
2448 if (consumerd32_libdir[0] != '\0') {
2449 char *tmp;
2450 size_t tmplen;
2451
2452 tmp = getenv("LD_LIBRARY_PATH");
2453 if (!tmp) {
2454 tmp = "";
2455 }
2456 tmplen = strlen("LD_LIBRARY_PATH=")
2457 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2458 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2459 if (!tmpnew) {
2460 ret = -ENOMEM;
2461 goto error;
2462 }
2463 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2464 strcat(tmpnew, consumerd32_libdir);
2465 if (tmp[0] != '\0') {
2466 strcat(tmpnew, ":");
2467 strcat(tmpnew, tmp);
2468 }
2469 ret = putenv(tmpnew);
2470 if (ret) {
2471 ret = -errno;
2472 free(tmpnew);
2473 goto error;
2474 }
2475 }
2476 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2477 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2478 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2479 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2480 "--group", tracing_group_name,
2481 NULL);
2482 if (consumerd32_libdir[0] != '\0') {
2483 free(tmpnew);
2484 }
2485 break;
2486 }
2487 default:
2488 PERROR("unknown consumer type");
2489 exit(EXIT_FAILURE);
2490 }
2491 if (errno != 0) {
2492 PERROR("Consumer execl()");
2493 }
2494 /* Reaching this point, we got a failure on our execl(). */
2495 exit(EXIT_FAILURE);
2496 } else if (pid > 0) {
2497 ret = pid;
2498 } else {
2499 PERROR("start consumer fork");
2500 ret = -errno;
2501 }
2502 error:
2503 return ret;
2504 }
2505
2506 /*
2507 * Spawn the consumerd daemon and session daemon thread.
2508 */
2509 static int start_consumerd(struct consumer_data *consumer_data)
2510 {
2511 int ret;
2512
2513 /*
2514 * Set the listen() state on the socket since there is a possible race
2515 * between the exec() of the consumer daemon and this call if place in the
2516 * consumer thread. See bug #366 for more details.
2517 */
2518 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2519 if (ret < 0) {
2520 goto error;
2521 }
2522
2523 pthread_mutex_lock(&consumer_data->pid_mutex);
2524 if (consumer_data->pid != 0) {
2525 pthread_mutex_unlock(&consumer_data->pid_mutex);
2526 goto end;
2527 }
2528
2529 ret = spawn_consumerd(consumer_data);
2530 if (ret < 0) {
2531 ERR("Spawning consumerd failed");
2532 pthread_mutex_unlock(&consumer_data->pid_mutex);
2533 goto error;
2534 }
2535
2536 /* Setting up the consumer_data pid */
2537 consumer_data->pid = ret;
2538 DBG2("Consumer pid %d", consumer_data->pid);
2539 pthread_mutex_unlock(&consumer_data->pid_mutex);
2540
2541 DBG2("Spawning consumer control thread");
2542 ret = spawn_consumer_thread(consumer_data);
2543 if (ret < 0) {
2544 ERR("Fatal error spawning consumer control thread");
2545 goto error;
2546 }
2547
2548 end:
2549 return 0;
2550
2551 error:
2552 /* Cleanup already created sockets on error. */
2553 if (consumer_data->err_sock >= 0) {
2554 int err;
2555
2556 err = close(consumer_data->err_sock);
2557 if (err < 0) {
2558 PERROR("close consumer data error socket");
2559 }
2560 }
2561 return ret;
2562 }
2563
2564 /*
2565 * Setup necessary data for kernel tracer action.
2566 */
2567 static int init_kernel_tracer(void)
2568 {
2569 int ret;
2570
2571 /* Modprobe lttng kernel modules */
2572 ret = modprobe_lttng_control();
2573 if (ret < 0) {
2574 goto error;
2575 }
2576
2577 /* Open debugfs lttng */
2578 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2579 if (kernel_tracer_fd < 0) {
2580 DBG("Failed to open %s", module_proc_lttng);
2581 ret = -1;
2582 goto error_open;
2583 }
2584
2585 /* Validate kernel version */
2586 ret = kernel_validate_version(kernel_tracer_fd);
2587 if (ret < 0) {
2588 goto error_version;
2589 }
2590
2591 ret = modprobe_lttng_data();
2592 if (ret < 0) {
2593 goto error_modules;
2594 }
2595
2596 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2597 return 0;
2598
2599 error_version:
2600 modprobe_remove_lttng_control();
2601 ret = close(kernel_tracer_fd);
2602 if (ret) {
2603 PERROR("close");
2604 }
2605 kernel_tracer_fd = -1;
2606 return LTTNG_ERR_KERN_VERSION;
2607
2608 error_modules:
2609 ret = close(kernel_tracer_fd);
2610 if (ret) {
2611 PERROR("close");
2612 }
2613
2614 error_open:
2615 modprobe_remove_lttng_control();
2616
2617 error:
2618 WARN("No kernel tracer available");
2619 kernel_tracer_fd = -1;
2620 if (!is_root) {
2621 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2622 } else {
2623 return LTTNG_ERR_KERN_NA;
2624 }
2625 }
2626
2627
2628 /*
2629 * Copy consumer output from the tracing session to the domain session. The
2630 * function also applies the right modification on a per domain basis for the
2631 * trace files destination directory.
2632 *
2633 * Should *NOT* be called with RCU read-side lock held.
2634 */
2635 static int copy_session_consumer(int domain, struct ltt_session *session)
2636 {
2637 int ret;
2638 const char *dir_name;
2639 struct consumer_output *consumer;
2640
2641 assert(session);
2642 assert(session->consumer);
2643
2644 switch (domain) {
2645 case LTTNG_DOMAIN_KERNEL:
2646 DBG3("Copying tracing session consumer output in kernel session");
2647 /*
2648 * XXX: We should audit the session creation and what this function
2649 * does "extra" in order to avoid a destroy since this function is used
2650 * in the domain session creation (kernel and ust) only. Same for UST
2651 * domain.
2652 */
2653 if (session->kernel_session->consumer) {
2654 consumer_destroy_output(session->kernel_session->consumer);
2655 }
2656 session->kernel_session->consumer =
2657 consumer_copy_output(session->consumer);
2658 /* Ease our life a bit for the next part */
2659 consumer = session->kernel_session->consumer;
2660 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2661 break;
2662 case LTTNG_DOMAIN_JUL:
2663 case LTTNG_DOMAIN_LOG4J:
2664 case LTTNG_DOMAIN_PYTHON:
2665 case LTTNG_DOMAIN_UST:
2666 DBG3("Copying tracing session consumer output in UST session");
2667 if (session->ust_session->consumer) {
2668 consumer_destroy_output(session->ust_session->consumer);
2669 }
2670 session->ust_session->consumer =
2671 consumer_copy_output(session->consumer);
2672 /* Ease our life a bit for the next part */
2673 consumer = session->ust_session->consumer;
2674 dir_name = DEFAULT_UST_TRACE_DIR;
2675 break;
2676 default:
2677 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2678 goto error;
2679 }
2680
2681 /* Append correct directory to subdir */
2682 strncat(consumer->subdir, dir_name,
2683 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2684 DBG3("Copy session consumer subdir %s", consumer->subdir);
2685
2686 ret = LTTNG_OK;
2687
2688 error:
2689 return ret;
2690 }
2691
2692 /*
2693 * Create an UST session and add it to the session ust list.
2694 *
2695 * Should *NOT* be called with RCU read-side lock held.
2696 */
2697 static int create_ust_session(struct ltt_session *session,
2698 struct lttng_domain *domain)
2699 {
2700 int ret;
2701 struct ltt_ust_session *lus = NULL;
2702
2703 assert(session);
2704 assert(domain);
2705 assert(session->consumer);
2706
2707 switch (domain->type) {
2708 case LTTNG_DOMAIN_JUL:
2709 case LTTNG_DOMAIN_LOG4J:
2710 case LTTNG_DOMAIN_PYTHON:
2711 case LTTNG_DOMAIN_UST:
2712 break;
2713 default:
2714 ERR("Unknown UST domain on create session %d", domain->type);
2715 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2716 goto error;
2717 }
2718
2719 DBG("Creating UST session");
2720
2721 lus = trace_ust_create_session(session->id);
2722 if (lus == NULL) {
2723 ret = LTTNG_ERR_UST_SESS_FAIL;
2724 goto error;
2725 }
2726
2727 lus->uid = session->uid;
2728 lus->gid = session->gid;
2729 lus->output_traces = session->output_traces;
2730 lus->snapshot_mode = session->snapshot_mode;
2731 lus->live_timer_interval = session->live_timer;
2732 session->ust_session = lus;
2733
2734 /* Copy session output to the newly created UST session */
2735 ret = copy_session_consumer(domain->type, session);
2736 if (ret != LTTNG_OK) {
2737 goto error;
2738 }
2739
2740 return LTTNG_OK;
2741
2742 error:
2743 free(lus);
2744 session->ust_session = NULL;
2745 return ret;
2746 }
2747
2748 /*
2749 * Create a kernel tracer session then create the default channel.
2750 */
2751 static int create_kernel_session(struct ltt_session *session)
2752 {
2753 int ret;
2754
2755 DBG("Creating kernel session");
2756
2757 ret = kernel_create_session(session, kernel_tracer_fd);
2758 if (ret < 0) {
2759 ret = LTTNG_ERR_KERN_SESS_FAIL;
2760 goto error;
2761 }
2762
2763 /* Code flow safety */
2764 assert(session->kernel_session);
2765
2766 /* Copy session output to the newly created Kernel session */
2767 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2768 if (ret != LTTNG_OK) {
2769 goto error;
2770 }
2771
2772 /* Create directory(ies) on local filesystem. */
2773 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2774 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2775 ret = run_as_mkdir_recursive(
2776 session->kernel_session->consumer->dst.trace_path,
2777 S_IRWXU | S_IRWXG, session->uid, session->gid);
2778 if (ret < 0) {
2779 if (ret != -EEXIST) {
2780 ERR("Trace directory creation error");
2781 goto error;
2782 }
2783 }
2784 }
2785
2786 session->kernel_session->uid = session->uid;
2787 session->kernel_session->gid = session->gid;
2788 session->kernel_session->output_traces = session->output_traces;
2789 session->kernel_session->snapshot_mode = session->snapshot_mode;
2790
2791 return LTTNG_OK;
2792
2793 error:
2794 trace_kernel_destroy_session(session->kernel_session);
2795 session->kernel_session = NULL;
2796 return ret;
2797 }
2798
2799 /*
2800 * Count number of session permitted by uid/gid.
2801 */
2802 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2803 {
2804 unsigned int i = 0;
2805 struct ltt_session *session;
2806
2807 DBG("Counting number of available session for UID %d GID %d",
2808 uid, gid);
2809 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2810 /*
2811 * Only list the sessions the user can control.
2812 */
2813 if (!session_access_ok(session, uid, gid)) {
2814 continue;
2815 }
2816 i++;
2817 }
2818 return i;
2819 }
2820
2821 /*
2822 * Process the command requested by the lttng client within the command
2823 * context structure. This function make sure that the return structure (llm)
2824 * is set and ready for transmission before returning.
2825 *
2826 * Return any error encountered or 0 for success.
2827 *
2828 * "sock" is only used for special-case var. len data.
2829 *
2830 * Should *NOT* be called with RCU read-side lock held.
2831 */
2832 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2833 int *sock_error)
2834 {
2835 int ret = LTTNG_OK;
2836 int need_tracing_session = 1;
2837 int need_domain;
2838
2839 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2840
2841 *sock_error = 0;
2842
2843 switch (cmd_ctx->lsm->cmd_type) {
2844 case LTTNG_CREATE_SESSION:
2845 case LTTNG_CREATE_SESSION_SNAPSHOT:
2846 case LTTNG_CREATE_SESSION_LIVE:
2847 case LTTNG_DESTROY_SESSION:
2848 case LTTNG_LIST_SESSIONS:
2849 case LTTNG_LIST_DOMAINS:
2850 case LTTNG_START_TRACE:
2851 case LTTNG_STOP_TRACE:
2852 case LTTNG_DATA_PENDING:
2853 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2854 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2855 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2856 case LTTNG_SNAPSHOT_RECORD:
2857 case LTTNG_SAVE_SESSION:
2858 need_domain = 0;
2859 break;
2860 default:
2861 need_domain = 1;
2862 }
2863
2864 if (opt_no_kernel && need_domain
2865 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2866 if (!is_root) {
2867 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2868 } else {
2869 ret = LTTNG_ERR_KERN_NA;
2870 }
2871 goto error;
2872 }
2873
2874 /* Deny register consumer if we already have a spawned consumer. */
2875 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2876 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2877 if (kconsumer_data.pid > 0) {
2878 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2879 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2880 goto error;
2881 }
2882 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2883 }
2884
2885 /*
2886 * Check for command that don't needs to allocate a returned payload. We do
2887 * this here so we don't have to make the call for no payload at each
2888 * command.
2889 */
2890 switch(cmd_ctx->lsm->cmd_type) {
2891 case LTTNG_LIST_SESSIONS:
2892 case LTTNG_LIST_TRACEPOINTS:
2893 case LTTNG_LIST_TRACEPOINT_FIELDS:
2894 case LTTNG_LIST_DOMAINS:
2895 case LTTNG_LIST_CHANNELS:
2896 case LTTNG_LIST_EVENTS:
2897 case LTTNG_LIST_SYSCALLS:
2898 break;
2899 default:
2900 /* Setup lttng message with no payload */
2901 ret = setup_lttng_msg(cmd_ctx, 0);
2902 if (ret < 0) {
2903 /* This label does not try to unlock the session */
2904 goto init_setup_error;
2905 }
2906 }
2907
2908 /* Commands that DO NOT need a session. */
2909 switch (cmd_ctx->lsm->cmd_type) {
2910 case LTTNG_CREATE_SESSION:
2911 case LTTNG_CREATE_SESSION_SNAPSHOT:
2912 case LTTNG_CREATE_SESSION_LIVE:
2913 case LTTNG_CALIBRATE:
2914 case LTTNG_LIST_SESSIONS:
2915 case LTTNG_LIST_TRACEPOINTS:
2916 case LTTNG_LIST_SYSCALLS:
2917 case LTTNG_LIST_TRACEPOINT_FIELDS:
2918 case LTTNG_SAVE_SESSION:
2919 need_tracing_session = 0;
2920 break;
2921 default:
2922 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2923 /*
2924 * We keep the session list lock across _all_ commands
2925 * for now, because the per-session lock does not
2926 * handle teardown properly.
2927 */
2928 session_lock_list();
2929 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2930 if (cmd_ctx->session == NULL) {
2931 ret = LTTNG_ERR_SESS_NOT_FOUND;
2932 goto error;
2933 } else {
2934 /* Acquire lock for the session */
2935 session_lock(cmd_ctx->session);
2936 }
2937 break;
2938 }
2939
2940 /*
2941 * Commands that need a valid session but should NOT create one if none
2942 * exists. Instead of creating one and destroying it when the command is
2943 * handled, process that right before so we save some round trip in useless
2944 * code path.
2945 */
2946 switch (cmd_ctx->lsm->cmd_type) {
2947 case LTTNG_DISABLE_CHANNEL:
2948 case LTTNG_DISABLE_EVENT:
2949 switch (cmd_ctx->lsm->domain.type) {
2950 case LTTNG_DOMAIN_KERNEL:
2951 if (!cmd_ctx->session->kernel_session) {
2952 ret = LTTNG_ERR_NO_CHANNEL;
2953 goto error;
2954 }
2955 break;
2956 case LTTNG_DOMAIN_JUL:
2957 case LTTNG_DOMAIN_LOG4J:
2958 case LTTNG_DOMAIN_PYTHON:
2959 case LTTNG_DOMAIN_UST:
2960 if (!cmd_ctx->session->ust_session) {
2961 ret = LTTNG_ERR_NO_CHANNEL;
2962 goto error;
2963 }
2964 break;
2965 default:
2966 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2967 goto error;
2968 }
2969 default:
2970 break;
2971 }
2972
2973 if (!need_domain) {
2974 goto skip_domain;
2975 }
2976
2977 /*
2978 * Check domain type for specific "pre-action".
2979 */
2980 switch (cmd_ctx->lsm->domain.type) {
2981 case LTTNG_DOMAIN_KERNEL:
2982 if (!is_root) {
2983 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2984 goto error;
2985 }
2986
2987 /* Kernel tracer check */
2988 if (kernel_tracer_fd == -1) {
2989 /* Basically, load kernel tracer modules */
2990 ret = init_kernel_tracer();
2991 if (ret != 0) {
2992 goto error;
2993 }
2994 }
2995
2996 /* Consumer is in an ERROR state. Report back to client */
2997 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2998 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2999 goto error;
3000 }
3001
3002 /* Need a session for kernel command */
3003 if (need_tracing_session) {
3004 if (cmd_ctx->session->kernel_session == NULL) {
3005 ret = create_kernel_session(cmd_ctx->session);
3006 if (ret < 0) {
3007 ret = LTTNG_ERR_KERN_SESS_FAIL;
3008 goto error;
3009 }
3010 }
3011
3012 /* Start the kernel consumer daemon */
3013 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3014 if (kconsumer_data.pid == 0 &&
3015 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3016 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3017 ret = start_consumerd(&kconsumer_data);
3018 if (ret < 0) {
3019 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3020 goto error;
3021 }
3022 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3023 } else {
3024 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3025 }
3026
3027 /*
3028 * The consumer was just spawned so we need to add the socket to
3029 * the consumer output of the session if exist.
3030 */
3031 ret = consumer_create_socket(&kconsumer_data,
3032 cmd_ctx->session->kernel_session->consumer);
3033 if (ret < 0) {
3034 goto error;
3035 }
3036 }
3037
3038 break;
3039 case LTTNG_DOMAIN_JUL:
3040 case LTTNG_DOMAIN_LOG4J:
3041 case LTTNG_DOMAIN_PYTHON:
3042 case LTTNG_DOMAIN_UST:
3043 {
3044 if (!ust_app_supported()) {
3045 ret = LTTNG_ERR_NO_UST;
3046 goto error;
3047 }
3048 /* Consumer is in an ERROR state. Report back to client */
3049 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3050 ret = LTTNG_ERR_NO_USTCONSUMERD;
3051 goto error;
3052 }
3053
3054 if (need_tracing_session) {
3055 /* Create UST session if none exist. */
3056 if (cmd_ctx->session->ust_session == NULL) {
3057 ret = create_ust_session(cmd_ctx->session,
3058 &cmd_ctx->lsm->domain);
3059 if (ret != LTTNG_OK) {
3060 goto error;
3061 }
3062 }
3063
3064 /* Start the UST consumer daemons */
3065 /* 64-bit */
3066 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3067 if (consumerd64_bin[0] != '\0' &&
3068 ustconsumer64_data.pid == 0 &&
3069 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3070 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3071 ret = start_consumerd(&ustconsumer64_data);
3072 if (ret < 0) {
3073 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3074 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3075 goto error;
3076 }
3077
3078 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3079 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3080 } else {
3081 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3082 }
3083
3084 /*
3085 * Setup socket for consumer 64 bit. No need for atomic access
3086 * since it was set above and can ONLY be set in this thread.
3087 */
3088 ret = consumer_create_socket(&ustconsumer64_data,
3089 cmd_ctx->session->ust_session->consumer);
3090 if (ret < 0) {
3091 goto error;
3092 }
3093
3094 /* 32-bit */
3095 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3096 if (consumerd32_bin[0] != '\0' &&
3097 ustconsumer32_data.pid == 0 &&
3098 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3099 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3100 ret = start_consumerd(&ustconsumer32_data);
3101 if (ret < 0) {
3102 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3103 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3104 goto error;
3105 }
3106
3107 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3108 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3109 } else {
3110 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3111 }
3112
3113 /*
3114 * Setup socket for consumer 64 bit. No need for atomic access
3115 * since it was set above and can ONLY be set in this thread.
3116 */
3117 ret = consumer_create_socket(&ustconsumer32_data,
3118 cmd_ctx->session->ust_session->consumer);
3119 if (ret < 0) {
3120 goto error;
3121 }
3122 }
3123 break;
3124 }
3125 default:
3126 break;
3127 }
3128 skip_domain:
3129
3130 /* Validate consumer daemon state when start/stop trace command */
3131 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3132 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3133 switch (cmd_ctx->lsm->domain.type) {
3134 case LTTNG_DOMAIN_JUL:
3135 case LTTNG_DOMAIN_LOG4J:
3136 case LTTNG_DOMAIN_PYTHON:
3137 case LTTNG_DOMAIN_UST:
3138 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3139 ret = LTTNG_ERR_NO_USTCONSUMERD;
3140 goto error;
3141 }
3142 break;
3143 case LTTNG_DOMAIN_KERNEL:
3144 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3145 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3146 goto error;
3147 }
3148 break;
3149 }
3150 }
3151
3152 /*
3153 * Check that the UID or GID match that of the tracing session.
3154 * The root user can interact with all sessions.
3155 */
3156 if (need_tracing_session) {
3157 if (!session_access_ok(cmd_ctx->session,
3158 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3159 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3160 ret = LTTNG_ERR_EPERM;
3161 goto error;
3162 }
3163 }
3164
3165 /*
3166 * Send relayd information to consumer as soon as we have a domain and a
3167 * session defined.
3168 */
3169 if (cmd_ctx->session && need_domain) {
3170 /*
3171 * Setup relayd if not done yet. If the relayd information was already
3172 * sent to the consumer, this call will gracefully return.
3173 */
3174 ret = cmd_setup_relayd(cmd_ctx->session);
3175 if (ret != LTTNG_OK) {
3176 goto error;
3177 }
3178 }
3179
3180 /* Process by command type */
3181 switch (cmd_ctx->lsm->cmd_type) {
3182 case LTTNG_ADD_CONTEXT:
3183 {
3184 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3185 cmd_ctx->lsm->u.context.channel_name,
3186 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3187 break;
3188 }
3189 case LTTNG_DISABLE_CHANNEL:
3190 {
3191 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3192 cmd_ctx->lsm->u.disable.channel_name);
3193 break;
3194 }
3195 case LTTNG_DISABLE_EVENT:
3196 {
3197 /* FIXME: passing packed structure to non-packed pointer */
3198 /* TODO: handle filter */
3199 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3200 cmd_ctx->lsm->u.disable.channel_name,
3201 &cmd_ctx->lsm->u.disable.event);
3202 break;
3203 }
3204 case LTTNG_ENABLE_CHANNEL:
3205 {
3206 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3207 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3208 break;
3209 }
3210 case LTTNG_ENABLE_EVENT:
3211 {
3212 struct lttng_event_exclusion *exclusion = NULL;
3213 struct lttng_filter_bytecode *bytecode = NULL;
3214 char *filter_expression = NULL;
3215
3216 /* Handle exclusion events and receive it from the client. */
3217 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3218 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3219
3220 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3221 (count * LTTNG_SYMBOL_NAME_LEN));
3222 if (!exclusion) {
3223 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3224 goto error;
3225 }
3226
3227 DBG("Receiving var len exclusion event list from client ...");
3228 exclusion->count = count;
3229 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3230 count * LTTNG_SYMBOL_NAME_LEN);
3231 if (ret <= 0) {
3232 DBG("Nothing recv() from client var len data... continuing");
3233 *sock_error = 1;
3234 free(exclusion);
3235 ret = LTTNG_ERR_EXCLUSION_INVAL;
3236 goto error;
3237 }
3238 }
3239
3240 /* Get filter expression from client. */
3241 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3242 size_t expression_len =
3243 cmd_ctx->lsm->u.enable.expression_len;
3244
3245 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3246 ret = LTTNG_ERR_FILTER_INVAL;
3247 free(exclusion);
3248 goto error;
3249 }
3250
3251 filter_expression = zmalloc(expression_len);
3252 if (!filter_expression) {
3253 free(exclusion);
3254 ret = LTTNG_ERR_FILTER_NOMEM;
3255 goto error;
3256 }
3257
3258 /* Receive var. len. data */
3259 DBG("Receiving var len filter's expression from client ...");
3260 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3261 expression_len);
3262 if (ret <= 0) {
3263 DBG("Nothing recv() from client car len data... continuing");
3264 *sock_error = 1;
3265 free(filter_expression);
3266 free(exclusion);
3267 ret = LTTNG_ERR_FILTER_INVAL;
3268 goto error;
3269 }
3270 }
3271
3272 /* Handle filter and get bytecode from client. */
3273 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3274 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3275
3276 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3277 ret = LTTNG_ERR_FILTER_INVAL;
3278 free(filter_expression);
3279 free(exclusion);
3280 goto error;
3281 }
3282
3283 bytecode = zmalloc(bytecode_len);
3284 if (!bytecode) {
3285 free(filter_expression);
3286 free(exclusion);
3287 ret = LTTNG_ERR_FILTER_NOMEM;
3288 goto error;
3289 }
3290
3291 /* Receive var. len. data */
3292 DBG("Receiving var len filter's bytecode from client ...");
3293 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3294 if (ret <= 0) {
3295 DBG("Nothing recv() from client car len data... continuing");
3296 *sock_error = 1;
3297 free(filter_expression);
3298 free(bytecode);
3299 free(exclusion);
3300 ret = LTTNG_ERR_FILTER_INVAL;
3301 goto error;
3302 }
3303
3304 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3305 free(filter_expression);
3306 free(bytecode);
3307 free(exclusion);
3308 ret = LTTNG_ERR_FILTER_INVAL;
3309 goto error;
3310 }
3311 }
3312
3313 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3314 cmd_ctx->lsm->u.enable.channel_name,
3315 &cmd_ctx->lsm->u.enable.event,
3316 filter_expression, bytecode, exclusion,
3317 kernel_poll_pipe[1]);
3318 break;
3319 }
3320 case LTTNG_LIST_TRACEPOINTS:
3321 {
3322 struct lttng_event *events;
3323 ssize_t nb_events;
3324
3325 session_lock_list();
3326 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3327 session_unlock_list();
3328 if (nb_events < 0) {
3329 /* Return value is a negative lttng_error_code. */
3330 ret = -nb_events;
3331 goto error;
3332 }
3333
3334 /*
3335 * Setup lttng message with payload size set to the event list size in
3336 * bytes and then copy list into the llm payload.
3337 */
3338 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3339 if (ret < 0) {
3340 free(events);
3341 goto setup_error;
3342 }
3343
3344 /* Copy event list into message payload */
3345 memcpy(cmd_ctx->llm->payload, events,
3346 sizeof(struct lttng_event) * nb_events);
3347
3348 free(events);
3349
3350 ret = LTTNG_OK;
3351 break;
3352 }
3353 case LTTNG_LIST_TRACEPOINT_FIELDS:
3354 {
3355 struct lttng_event_field *fields;
3356 ssize_t nb_fields;
3357
3358 session_lock_list();
3359 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3360 &fields);
3361 session_unlock_list();
3362 if (nb_fields < 0) {
3363 /* Return value is a negative lttng_error_code. */
3364 ret = -nb_fields;
3365 goto error;
3366 }
3367
3368 /*
3369 * Setup lttng message with payload size set to the event list size in
3370 * bytes and then copy list into the llm payload.
3371 */
3372 ret = setup_lttng_msg(cmd_ctx,
3373 sizeof(struct lttng_event_field) * nb_fields);
3374 if (ret < 0) {
3375 free(fields);
3376 goto setup_error;
3377 }
3378
3379 /* Copy event list into message payload */
3380 memcpy(cmd_ctx->llm->payload, fields,
3381 sizeof(struct lttng_event_field) * nb_fields);
3382
3383 free(fields);
3384
3385 ret = LTTNG_OK;
3386 break;
3387 }
3388 case LTTNG_LIST_SYSCALLS:
3389 {
3390 struct lttng_event *events;
3391 ssize_t nb_events;
3392
3393 nb_events = cmd_list_syscalls(&events);
3394 if (nb_events < 0) {
3395 /* Return value is a negative lttng_error_code. */
3396 ret = -nb_events;
3397 goto error;
3398 }
3399
3400 /*
3401 * Setup lttng message with payload size set to the event list size in
3402 * bytes and then copy list into the llm payload.
3403 */
3404 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3405 if (ret < 0) {
3406 free(events);
3407 goto setup_error;
3408 }
3409
3410 /* Copy event list into message payload */
3411 memcpy(cmd_ctx->llm->payload, events,
3412 sizeof(struct lttng_event) * nb_events);
3413
3414 free(events);
3415
3416 ret = LTTNG_OK;
3417 break;
3418 }
3419 case LTTNG_SET_CONSUMER_URI:
3420 {
3421 size_t nb_uri, len;
3422 struct lttng_uri *uris;
3423
3424 nb_uri = cmd_ctx->lsm->u.uri.size;
3425 len = nb_uri * sizeof(struct lttng_uri);
3426
3427 if (nb_uri == 0) {
3428 ret = LTTNG_ERR_INVALID;
3429 goto error;
3430 }
3431
3432 uris = zmalloc(len);
3433 if (uris == NULL) {
3434 ret = LTTNG_ERR_FATAL;
3435 goto error;
3436 }
3437
3438 /* Receive variable len data */
3439 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3440 ret = lttcomm_recv_unix_sock(sock, uris, len);
3441 if (ret <= 0) {
3442 DBG("No URIs received from client... continuing");
3443 *sock_error = 1;
3444 ret = LTTNG_ERR_SESSION_FAIL;
3445 free(uris);
3446 goto error;
3447 }
3448
3449 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3450 free(uris);
3451 if (ret != LTTNG_OK) {
3452 goto error;
3453 }
3454
3455
3456 break;
3457 }
3458 case LTTNG_START_TRACE:
3459 {
3460 ret = cmd_start_trace(cmd_ctx->session);
3461 break;
3462 }
3463 case LTTNG_STOP_TRACE:
3464 {
3465 ret = cmd_stop_trace(cmd_ctx->session);
3466 break;
3467 }
3468 case LTTNG_CREATE_SESSION:
3469 {
3470 size_t nb_uri, len;
3471 struct lttng_uri *uris = NULL;
3472
3473 nb_uri = cmd_ctx->lsm->u.uri.size;
3474 len = nb_uri * sizeof(struct lttng_uri);
3475
3476 if (nb_uri > 0) {
3477 uris = zmalloc(len);
3478 if (uris == NULL) {
3479 ret = LTTNG_ERR_FATAL;
3480 goto error;
3481 }
3482
3483 /* Receive variable len data */
3484 DBG("Waiting for %zu URIs from client ...", nb_uri);
3485 ret = lttcomm_recv_unix_sock(sock, uris, len);
3486 if (ret <= 0) {
3487 DBG("No URIs received from client... continuing");
3488 *sock_error = 1;
3489 ret = LTTNG_ERR_SESSION_FAIL;
3490 free(uris);
3491 goto error;
3492 }
3493
3494 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3495 DBG("Creating session with ONE network URI is a bad call");
3496 ret = LTTNG_ERR_SESSION_FAIL;
3497 free(uris);
3498 goto error;
3499 }
3500 }
3501
3502 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3503 &cmd_ctx->creds, 0);
3504
3505 free(uris);
3506
3507 break;
3508 }
3509 case LTTNG_DESTROY_SESSION:
3510 {
3511 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3512
3513 /* Set session to NULL so we do not unlock it after free. */
3514 cmd_ctx->session = NULL;
3515 break;
3516 }
3517 case LTTNG_LIST_DOMAINS:
3518 {
3519 ssize_t nb_dom;
3520 struct lttng_domain *domains = NULL;
3521
3522 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3523 if (nb_dom < 0) {
3524 /* Return value is a negative lttng_error_code. */
3525 ret = -nb_dom;
3526 goto error;
3527 }
3528
3529 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3530 if (ret < 0) {
3531 free(domains);
3532 goto setup_error;
3533 }
3534
3535 /* Copy event list into message payload */
3536 memcpy(cmd_ctx->llm->payload, domains,
3537 nb_dom * sizeof(struct lttng_domain));
3538
3539 free(domains);
3540
3541 ret = LTTNG_OK;
3542 break;
3543 }
3544 case LTTNG_LIST_CHANNELS:
3545 {
3546 int nb_chan;
3547 struct lttng_channel *channels = NULL;
3548
3549 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3550 cmd_ctx->session, &channels);
3551 if (nb_chan < 0) {
3552 /* Return value is a negative lttng_error_code. */
3553 ret = -nb_chan;
3554 goto error;
3555 }
3556
3557 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3558 if (ret < 0) {
3559 free(channels);
3560 goto setup_error;
3561 }
3562
3563 /* Copy event list into message payload */
3564 memcpy(cmd_ctx->llm->payload, channels,
3565 nb_chan * sizeof(struct lttng_channel));
3566
3567 free(channels);
3568
3569 ret = LTTNG_OK;
3570 break;
3571 }
3572 case LTTNG_LIST_EVENTS:
3573 {
3574 ssize_t nb_event;
3575 struct lttng_event *events = NULL;
3576
3577 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3578 cmd_ctx->lsm->u.list.channel_name, &events);
3579 if (nb_event < 0) {
3580 /* Return value is a negative lttng_error_code. */
3581 ret = -nb_event;
3582 goto error;
3583 }
3584
3585 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3586 if (ret < 0) {
3587 free(events);
3588 goto setup_error;
3589 }
3590
3591 /* Copy event list into message payload */
3592 memcpy(cmd_ctx->llm->payload, events,
3593 nb_event * sizeof(struct lttng_event));
3594
3595 free(events);
3596
3597 ret = LTTNG_OK;
3598 break;
3599 }
3600 case LTTNG_LIST_SESSIONS:
3601 {
3602 unsigned int nr_sessions;
3603
3604 session_lock_list();
3605 nr_sessions = lttng_sessions_count(
3606 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3607 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3608
3609 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3610 if (ret < 0) {
3611 session_unlock_list();
3612 goto setup_error;
3613 }
3614
3615 /* Filled the session array */
3616 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3617 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3618 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3619
3620 session_unlock_list();
3621
3622 ret = LTTNG_OK;
3623 break;
3624 }
3625 case LTTNG_CALIBRATE:
3626 {
3627 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3628 &cmd_ctx->lsm->u.calibrate);
3629 break;
3630 }
3631 case LTTNG_REGISTER_CONSUMER:
3632 {
3633 struct consumer_data *cdata;
3634
3635 switch (cmd_ctx->lsm->domain.type) {
3636 case LTTNG_DOMAIN_KERNEL:
3637 cdata = &kconsumer_data;
3638 break;
3639 default:
3640 ret = LTTNG_ERR_UND;
3641 goto error;
3642 }
3643
3644 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3645 cmd_ctx->lsm->u.reg.path, cdata);
3646 break;
3647 }
3648 case LTTNG_DATA_PENDING:
3649 {
3650 ret = cmd_data_pending(cmd_ctx->session);
3651 break;
3652 }
3653 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3654 {
3655 struct lttcomm_lttng_output_id reply;
3656
3657 ret = cmd_snapshot_add_output(cmd_ctx->session,
3658 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3659 if (ret != LTTNG_OK) {
3660 goto error;
3661 }
3662
3663 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3664 if (ret < 0) {
3665 goto setup_error;
3666 }
3667
3668 /* Copy output list into message payload */
3669 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3670 ret = LTTNG_OK;
3671 break;
3672 }
3673 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3674 {
3675 ret = cmd_snapshot_del_output(cmd_ctx->session,
3676 &cmd_ctx->lsm->u.snapshot_output.output);
3677 break;
3678 }
3679 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3680 {
3681 ssize_t nb_output;
3682 struct lttng_snapshot_output *outputs = NULL;
3683
3684 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3685 if (nb_output < 0) {
3686 ret = -nb_output;
3687 goto error;
3688 }
3689
3690 ret = setup_lttng_msg(cmd_ctx,
3691 nb_output * sizeof(struct lttng_snapshot_output));
3692 if (ret < 0) {
3693 free(outputs);
3694 goto setup_error;
3695 }
3696
3697 if (outputs) {
3698 /* Copy output list into message payload */
3699 memcpy(cmd_ctx->llm->payload, outputs,
3700 nb_output * sizeof(struct lttng_snapshot_output));
3701 free(outputs);
3702 }
3703
3704 ret = LTTNG_OK;
3705 break;
3706 }
3707 case LTTNG_SNAPSHOT_RECORD:
3708 {
3709 ret = cmd_snapshot_record(cmd_ctx->session,
3710 &cmd_ctx->lsm->u.snapshot_record.output,
3711 cmd_ctx->lsm->u.snapshot_record.wait);
3712 break;
3713 }
3714 case LTTNG_CREATE_SESSION_SNAPSHOT:
3715 {
3716 size_t nb_uri, len;
3717 struct lttng_uri *uris = NULL;
3718
3719 nb_uri = cmd_ctx->lsm->u.uri.size;
3720 len = nb_uri * sizeof(struct lttng_uri);
3721
3722 if (nb_uri > 0) {
3723 uris = zmalloc(len);
3724 if (uris == NULL) {
3725 ret = LTTNG_ERR_FATAL;
3726 goto error;
3727 }
3728
3729 /* Receive variable len data */
3730 DBG("Waiting for %zu URIs from client ...", nb_uri);
3731 ret = lttcomm_recv_unix_sock(sock, uris, len);
3732 if (ret <= 0) {
3733 DBG("No URIs received from client... continuing");
3734 *sock_error = 1;
3735 ret = LTTNG_ERR_SESSION_FAIL;
3736 free(uris);
3737 goto error;
3738 }
3739
3740 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3741 DBG("Creating session with ONE network URI is a bad call");
3742 ret = LTTNG_ERR_SESSION_FAIL;
3743 free(uris);
3744 goto error;
3745 }
3746 }
3747
3748 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3749 nb_uri, &cmd_ctx->creds);
3750 free(uris);
3751 break;
3752 }
3753 case LTTNG_CREATE_SESSION_LIVE:
3754 {
3755 size_t nb_uri, len;
3756 struct lttng_uri *uris = NULL;
3757
3758 nb_uri = cmd_ctx->lsm->u.uri.size;
3759 len = nb_uri * sizeof(struct lttng_uri);
3760
3761 if (nb_uri > 0) {
3762 uris = zmalloc(len);
3763 if (uris == NULL) {
3764 ret = LTTNG_ERR_FATAL;
3765 goto error;
3766 }
3767
3768 /* Receive variable len data */
3769 DBG("Waiting for %zu URIs from client ...", nb_uri);
3770 ret = lttcomm_recv_unix_sock(sock, uris, len);
3771 if (ret <= 0) {
3772 DBG("No URIs received from client... continuing");
3773 *sock_error = 1;
3774 ret = LTTNG_ERR_SESSION_FAIL;
3775 free(uris);
3776 goto error;
3777 }
3778
3779 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3780 DBG("Creating session with ONE network URI is a bad call");
3781 ret = LTTNG_ERR_SESSION_FAIL;
3782 free(uris);
3783 goto error;
3784 }
3785 }
3786
3787 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3788 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3789 free(uris);
3790 break;
3791 }
3792 case LTTNG_SAVE_SESSION:
3793 {
3794 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3795 &cmd_ctx->creds);
3796 break;
3797 }
3798 default:
3799 ret = LTTNG_ERR_UND;
3800 break;
3801 }
3802
3803 error:
3804 if (cmd_ctx->llm == NULL) {
3805 DBG("Missing llm structure. Allocating one.");
3806 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3807 goto setup_error;
3808 }
3809 }
3810 /* Set return code */
3811 cmd_ctx->llm->ret_code = ret;
3812 setup_error:
3813 if (cmd_ctx->session) {
3814 session_unlock(cmd_ctx->session);
3815 }
3816 if (need_tracing_session) {
3817 session_unlock_list();
3818 }
3819 init_setup_error:
3820 return ret;
3821 }
3822
3823 /*
3824 * Thread managing health check socket.
3825 */
3826 static void *thread_manage_health(void *data)
3827 {
3828 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3829 uint32_t revents, nb_fd;
3830 struct lttng_poll_event events;
3831 struct health_comm_msg msg;
3832 struct health_comm_reply reply;
3833
3834 DBG("[thread] Manage health check started");
3835
3836 rcu_register_thread();
3837
3838 /* We might hit an error path before this is created. */
3839 lttng_poll_init(&events);
3840
3841 /* Create unix socket */
3842 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3843 if (sock < 0) {
3844 ERR("Unable to create health check Unix socket");
3845 ret = -1;
3846 goto error;
3847 }
3848
3849 if (is_root) {
3850 /* lttng health client socket path permissions */
3851 ret = chown(health_unix_sock_path, 0,
3852 utils_get_group_id(tracing_group_name));
3853 if (ret < 0) {
3854 ERR("Unable to set group on %s", health_unix_sock_path);
3855 PERROR("chown");
3856 ret = -1;
3857 goto error;
3858 }
3859
3860 ret = chmod(health_unix_sock_path,
3861 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3862 if (ret < 0) {
3863 ERR("Unable to set permissions on %s", health_unix_sock_path);
3864 PERROR("chmod");
3865 ret = -1;
3866 goto error;
3867 }
3868 }
3869
3870 /*
3871 * Set the CLOEXEC flag. Return code is useless because either way, the
3872 * show must go on.
3873 */
3874 (void) utils_set_fd_cloexec(sock);
3875
3876 ret = lttcomm_listen_unix_sock(sock);
3877 if (ret < 0) {
3878 goto error;
3879 }
3880
3881 /*
3882 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3883 * more will be added to this poll set.
3884 */
3885 ret = sessiond_set_thread_pollset(&events, 2);
3886 if (ret < 0) {
3887 goto error;
3888 }
3889
3890 /* Add the application registration socket */
3891 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3892 if (ret < 0) {
3893 goto error;
3894 }
3895
3896 sessiond_notify_ready();
3897
3898 while (1) {
3899 DBG("Health check ready");
3900
3901 /* Inifinite blocking call, waiting for transmission */
3902 restart:
3903 ret = lttng_poll_wait(&events, -1);
3904 if (ret < 0) {
3905 /*
3906 * Restart interrupted system call.
3907 */
3908 if (errno == EINTR) {
3909 goto restart;
3910 }
3911 goto error;
3912 }
3913
3914 nb_fd = ret;
3915
3916 for (i = 0; i < nb_fd; i++) {
3917 /* Fetch once the poll data */
3918 revents = LTTNG_POLL_GETEV(&events, i);
3919 pollfd = LTTNG_POLL_GETFD(&events, i);
3920
3921 /* Thread quit pipe has been closed. Killing thread. */
3922 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3923 if (ret) {
3924 err = 0;
3925 goto exit;
3926 }
3927
3928 /* Event on the registration socket */
3929 if (pollfd == sock) {
3930 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3931 ERR("Health socket poll error");
3932 goto error;
3933 }
3934 }
3935 }
3936
3937 new_sock = lttcomm_accept_unix_sock(sock);
3938 if (new_sock < 0) {
3939 goto error;
3940 }
3941
3942 /*
3943 * Set the CLOEXEC flag. Return code is useless because either way, the
3944 * show must go on.
3945 */
3946 (void) utils_set_fd_cloexec(new_sock);
3947
3948 DBG("Receiving data from client for health...");
3949 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3950 if (ret <= 0) {
3951 DBG("Nothing recv() from client... continuing");
3952 ret = close(new_sock);
3953 if (ret) {
3954 PERROR("close");
3955 }
3956 new_sock = -1;
3957 continue;
3958 }
3959
3960 rcu_thread_online();
3961
3962 memset(&reply, 0, sizeof(reply));
3963 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3964 /*
3965 * health_check_state returns 0 if health is
3966 * bad.
3967 */
3968 if (!health_check_state(health_sessiond, i)) {
3969 reply.ret_code |= 1ULL << i;
3970 }
3971 }
3972
3973 DBG2("Health check return value %" PRIx64, reply.ret_code);
3974
3975 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3976 if (ret < 0) {
3977 ERR("Failed to send health data back to client");
3978 }
3979
3980 /* End of transmission */
3981 ret = close(new_sock);
3982 if (ret) {
3983 PERROR("close");
3984 }
3985 new_sock = -1;
3986 }
3987
3988 exit:
3989 error:
3990 if (err) {
3991 ERR("Health error occurred in %s", __func__);
3992 }
3993 DBG("Health check thread dying");
3994 unlink(health_unix_sock_path);
3995 if (sock >= 0) {
3996 ret = close(sock);
3997 if (ret) {
3998 PERROR("close");
3999 }
4000 }
4001
4002 lttng_poll_clean(&events);
4003
4004 rcu_unregister_thread();
4005 return NULL;
4006 }
4007
4008 /*
4009 * This thread manage all clients request using the unix client socket for
4010 * communication.
4011 */
4012 static void *thread_manage_clients(void *data)
4013 {
4014 int sock = -1, ret, i, pollfd, err = -1;
4015 int sock_error;
4016 uint32_t revents, nb_fd;
4017 struct command_ctx *cmd_ctx = NULL;
4018 struct lttng_poll_event events;
4019
4020 DBG("[thread] Manage client started");
4021
4022 rcu_register_thread();
4023
4024 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4025
4026 health_code_update();
4027
4028 ret = lttcomm_listen_unix_sock(client_sock);
4029 if (ret < 0) {
4030 goto error_listen;
4031 }
4032
4033 /*
4034 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4035 * more will be added to this poll set.
4036 */
4037 ret = sessiond_set_thread_pollset(&events, 2);
4038 if (ret < 0) {
4039 goto error_create_poll;
4040 }
4041
4042 /* Add the application registration socket */
4043 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4044 if (ret < 0) {
4045 goto error;
4046 }
4047
4048 sessiond_notify_ready();
4049 ret = sem_post(&load_info->message_thread_ready);
4050 if (ret) {
4051 PERROR("sem_post message_thread_ready");
4052 goto error;
4053 }
4054
4055 /* This testpoint is after we signal readiness to the parent. */
4056 if (testpoint(sessiond_thread_manage_clients)) {
4057 goto error;
4058 }
4059
4060 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4061 goto error;
4062 }
4063
4064 health_code_update();
4065
4066 while (1) {
4067 DBG("Accepting client command ...");
4068
4069 /* Inifinite blocking call, waiting for transmission */
4070 restart:
4071 health_poll_entry();
4072 ret = lttng_poll_wait(&events, -1);
4073 health_poll_exit();
4074 if (ret < 0) {
4075 /*
4076 * Restart interrupted system call.
4077 */
4078 if (errno == EINTR) {
4079 goto restart;
4080 }
4081 goto error;
4082 }
4083
4084 nb_fd = ret;
4085
4086 for (i = 0; i < nb_fd; i++) {
4087 /* Fetch once the poll data */
4088 revents = LTTNG_POLL_GETEV(&events, i);
4089 pollfd = LTTNG_POLL_GETFD(&events, i);
4090
4091 health_code_update();
4092
4093 /* Thread quit pipe has been closed. Killing thread. */
4094 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4095 if (ret) {
4096 err = 0;
4097 goto exit;
4098 }
4099
4100 /* Event on the registration socket */
4101 if (pollfd == client_sock) {
4102 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4103 ERR("Client socket poll error");
4104 goto error;
4105 }
4106 }
4107 }
4108
4109 DBG("Wait for client response");
4110
4111 health_code_update();
4112
4113 sock = lttcomm_accept_unix_sock(client_sock);
4114 if (sock < 0) {
4115 goto error;
4116 }
4117
4118 /*
4119 * Set the CLOEXEC flag. Return code is useless because either way, the
4120 * show must go on.
4121 */
4122 (void) utils_set_fd_cloexec(sock);
4123
4124 /* Set socket option for credentials retrieval */
4125 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4126 if (ret < 0) {
4127 goto error;
4128 }
4129
4130 /* Allocate context command to process the client request */
4131 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4132 if (cmd_ctx == NULL) {
4133 PERROR("zmalloc cmd_ctx");
4134 goto error;
4135 }
4136
4137 /* Allocate data buffer for reception */
4138 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4139 if (cmd_ctx->lsm == NULL) {
4140 PERROR("zmalloc cmd_ctx->lsm");
4141 goto error;
4142 }
4143
4144 cmd_ctx->llm = NULL;
4145 cmd_ctx->session = NULL;
4146
4147 health_code_update();
4148
4149 /*
4150 * Data is received from the lttng client. The struct
4151 * lttcomm_session_msg (lsm) contains the command and data request of
4152 * the client.
4153 */
4154 DBG("Receiving data from client ...");
4155 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4156 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4157 if (ret <= 0) {
4158 DBG("Nothing recv() from client... continuing");
4159 ret = close(sock);
4160 if (ret) {
4161 PERROR("close");
4162 }
4163 sock = -1;
4164 clean_command_ctx(&cmd_ctx);
4165 continue;
4166 }
4167
4168 health_code_update();
4169
4170 // TODO: Validate cmd_ctx including sanity check for
4171 // security purpose.
4172
4173 rcu_thread_online();
4174 /*
4175 * This function dispatch the work to the kernel or userspace tracer
4176 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4177 * informations for the client. The command context struct contains
4178 * everything this function may needs.
4179 */
4180 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4181 rcu_thread_offline();
4182 if (ret < 0) {
4183 ret = close(sock);
4184 if (ret) {
4185 PERROR("close");
4186 }
4187 sock = -1;
4188 /*
4189 * TODO: Inform client somehow of the fatal error. At
4190 * this point, ret < 0 means that a zmalloc failed
4191 * (ENOMEM). Error detected but still accept
4192 * command, unless a socket error has been
4193 * detected.
4194 */
4195 clean_command_ctx(&cmd_ctx);
4196 continue;
4197 }
4198
4199 health_code_update();
4200
4201 DBG("Sending response (size: %d, retcode: %s)",
4202 cmd_ctx->lttng_msg_size,
4203 lttng_strerror(-cmd_ctx->llm->ret_code));
4204 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4205 if (ret < 0) {
4206 ERR("Failed to send data back to client");
4207 }
4208
4209 /* End of transmission */
4210 ret = close(sock);
4211 if (ret) {
4212 PERROR("close");
4213 }
4214 sock = -1;
4215
4216 clean_command_ctx(&cmd_ctx);
4217
4218 health_code_update();
4219 }
4220
4221 exit:
4222 error:
4223 if (sock >= 0) {
4224 ret = close(sock);
4225 if (ret) {
4226 PERROR("close");
4227 }
4228 }
4229
4230 lttng_poll_clean(&events);
4231 clean_command_ctx(&cmd_ctx);
4232
4233 error_listen:
4234 error_create_poll:
4235 unlink(client_unix_sock_path);
4236 if (client_sock >= 0) {
4237 ret = close(client_sock);
4238 if (ret) {
4239 PERROR("close");
4240 }
4241 }
4242
4243 if (err) {
4244 health_error();
4245 ERR("Health error occurred in %s", __func__);
4246 }
4247
4248 health_unregister(health_sessiond);
4249
4250 DBG("Client thread dying");
4251
4252 rcu_unregister_thread();
4253
4254 /*
4255 * Since we are creating the consumer threads, we own them, so we need
4256 * to join them before our thread exits.
4257 */
4258 ret = join_consumer_thread(&kconsumer_data);
4259 if (ret) {
4260 errno = ret;
4261 PERROR("join_consumer");
4262 }
4263
4264 ret = join_consumer_thread(&ustconsumer32_data);
4265 if (ret) {
4266 errno = ret;
4267 PERROR("join_consumer ust32");
4268 }
4269
4270 ret = join_consumer_thread(&ustconsumer64_data);
4271 if (ret) {
4272 errno = ret;
4273 PERROR("join_consumer ust64");
4274 }
4275 return NULL;
4276 }
4277
4278
4279 /*
4280 * usage function on stderr
4281 */
4282 static void usage(void)
4283 {
4284 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4285 fprintf(stderr, " -h, --help Display this usage.\n");
4286 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4287 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4288 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4289 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4290 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4291 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4292 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4293 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4294 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4295 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4296 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4297 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4298 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4299 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4300 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4301 fprintf(stderr, " -V, --version Show version number.\n");
4302 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4303 fprintf(stderr, " -q, --quiet No output at all.\n");
4304 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4305 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4306 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4307 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4308 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4309 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4310 fprintf(stderr, " -l --load PATH Load session configuration\n");
4311 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4312 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4313 }
4314
4315 /*
4316 * Take an option from the getopt output and set it in the right variable to be
4317 * used later.
4318 *
4319 * Return 0 on success else a negative value.
4320 */
4321 static int set_option(int opt, const char *arg, const char *optname)
4322 {
4323 int ret = 0;
4324
4325 if (arg && arg[0] == '\0') {
4326 /*
4327 * This only happens if the value is read from daemon config
4328 * file. This means the option requires an argument and the
4329 * configuration file contains a line such as:
4330 * my_option =
4331 */
4332 ret = -EINVAL;
4333 goto end;
4334 }
4335
4336 switch (opt) {
4337 case 0:
4338 fprintf(stderr, "option %s", optname);
4339 if (arg) {
4340 fprintf(stderr, " with arg %s\n", arg);
4341 }
4342 break;
4343 case 'c':
4344 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4345 break;
4346 case 'a':
4347 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4348 break;
4349 case 'd':
4350 opt_daemon = 1;
4351 break;
4352 case 'b':
4353 opt_background = 1;
4354 break;
4355 case 'g':
4356 /*
4357 * If the override option is set, the pointer points to a
4358 * *non* const thus freeing it even though the variable type is
4359 * set to const.
4360 */
4361 if (tracing_group_name_override) {
4362 free((void *) tracing_group_name);
4363 }
4364 tracing_group_name = strdup(arg);
4365 if (!tracing_group_name) {
4366 PERROR("strdup");
4367 ret = -ENOMEM;
4368 }
4369 tracing_group_name_override = 1;
4370 break;
4371 case 'h':
4372 usage();
4373 exit(EXIT_SUCCESS);
4374 case 'V':
4375 fprintf(stdout, "%s\n", VERSION);
4376 exit(EXIT_SUCCESS);
4377 case 'S':
4378 opt_sig_parent = 1;
4379 break;
4380 case 'E':
4381 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4382 break;
4383 case 'C':
4384 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4385 break;
4386 case 'F':
4387 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4388 break;
4389 case 'D':
4390 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4391 break;
4392 case 'H':
4393 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4394 break;
4395 case 'G':
4396 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4397 break;
4398 case 'N':
4399 opt_no_kernel = 1;
4400 break;
4401 case 'q':
4402 lttng_opt_quiet = 1;
4403 break;
4404 case 'v':
4405 /* Verbose level can increase using multiple -v */
4406 if (arg) {
4407 /* Value obtained from config file */
4408 lttng_opt_verbose = config_parse_value(arg);
4409 } else {
4410 /* -v used on command line */
4411 lttng_opt_verbose++;
4412 }
4413 /* Clamp value to [0, 3] */
4414 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4415 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4416 break;
4417 case 'Z':
4418 if (arg) {
4419 opt_verbose_consumer = config_parse_value(arg);
4420 } else {
4421 opt_verbose_consumer += 1;
4422 }
4423 break;
4424 case 'u':
4425 if (consumerd32_bin_override) {
4426 free((void *) consumerd32_bin);
4427 }
4428 consumerd32_bin = strdup(arg);
4429 if (!consumerd32_bin) {
4430 PERROR("strdup");
4431 ret = -ENOMEM;
4432 }
4433 consumerd32_bin_override = 1;
4434 break;
4435 case 'U':
4436 if (consumerd32_libdir_override) {
4437 free((void *) consumerd32_libdir);
4438 }
4439 consumerd32_libdir = strdup(arg);
4440 if (!consumerd32_libdir) {
4441 PERROR("strdup");
4442 ret = -ENOMEM;
4443 }
4444 consumerd32_libdir_override = 1;
4445 break;
4446 case 't':
4447 if (consumerd64_bin_override) {
4448 free((void *) consumerd64_bin);
4449 }
4450 consumerd64_bin = strdup(arg);
4451 if (!consumerd64_bin) {
4452 PERROR("strdup");
4453 ret = -ENOMEM;
4454 }
4455 consumerd64_bin_override = 1;
4456 break;
4457 case 'T':
4458 if (consumerd64_libdir_override) {
4459 free((void *) consumerd64_libdir);
4460 }
4461 consumerd64_libdir = strdup(arg);
4462 if (!consumerd64_libdir) {
4463 PERROR("strdup");
4464 ret = -ENOMEM;
4465 }
4466 consumerd64_libdir_override = 1;
4467 break;
4468 case 'p':
4469 free(opt_pidfile);
4470 opt_pidfile = strdup(arg);
4471 if (!opt_pidfile) {
4472 PERROR("strdup");
4473 ret = -ENOMEM;
4474 }
4475 break;
4476 case 'J': /* Agent TCP port. */
4477 {
4478 unsigned long v;
4479
4480 errno = 0;
4481 v = strtoul(arg, NULL, 0);
4482 if (errno != 0 || !isdigit(arg[0])) {
4483 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4484 return -1;
4485 }
4486 if (v == 0 || v >= 65535) {
4487 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4488 return -1;
4489 }
4490 agent_tcp_port = (uint32_t) v;
4491 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4492 break;
4493 }
4494 case 'l':
4495 free(opt_load_session_path);
4496 opt_load_session_path = strdup(arg);
4497 if (!opt_load_session_path) {
4498 PERROR("strdup");
4499 ret = -ENOMEM;
4500 }
4501 break;
4502 case 'P': /* probe modules list */
4503 free(kmod_probes_list);
4504 kmod_probes_list = strdup(arg);
4505 if (!kmod_probes_list) {
4506 PERROR("strdup");
4507 ret = -ENOMEM;
4508 }
4509 break;
4510 case 'e':
4511 free(kmod_extra_probes_list);
4512 kmod_extra_probes_list = strdup(arg);
4513 if (!kmod_extra_probes_list) {
4514 PERROR("strdup");
4515 ret = -ENOMEM;
4516 }
4517 break;
4518 case 'f':
4519 /* This is handled in set_options() thus silent break. */
4520 break;
4521 default:
4522 /* Unknown option or other error.
4523 * Error is printed by getopt, just return */
4524 ret = -1;
4525 }
4526
4527 end:
4528 if (ret == -EINVAL) {
4529 const char *opt_name = "unknown";
4530 int i;
4531
4532 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4533 i++) {
4534 if (opt == long_options[i].val) {
4535 opt_name = long_options[i].name;
4536 break;
4537 }
4538 }
4539
4540 WARN("Invalid argument provided for option \"%s\", using default value.",
4541 opt_name);
4542 }
4543
4544 return ret;
4545 }
4546
4547 /*
4548 * config_entry_handler_cb used to handle options read from a config file.
4549 * See config_entry_handler_cb comment in common/config/config.h for the
4550 * return value conventions.
4551 */
4552 static int config_entry_handler(const struct config_entry *entry, void *unused)
4553 {
4554 int ret = 0, i;
4555
4556 if (!entry || !entry->name || !entry->value) {
4557 ret = -EINVAL;
4558 goto end;
4559 }
4560
4561 /* Check if the option is to be ignored */
4562 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4563 if (!strcmp(entry->name, config_ignore_options[i])) {
4564 goto end;
4565 }
4566 }
4567
4568 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4569 i++) {
4570
4571 /* Ignore if not fully matched. */
4572 if (strcmp(entry->name, long_options[i].name)) {
4573 continue;
4574 }
4575
4576 /*
4577 * If the option takes no argument on the command line, we have to
4578 * check if the value is "true". We support non-zero numeric values,
4579 * true, on and yes.
4580 */
4581 if (!long_options[i].has_arg) {
4582 ret = config_parse_value(entry->value);
4583 if (ret <= 0) {
4584 if (ret) {
4585 WARN("Invalid configuration value \"%s\" for option %s",
4586 entry->value, entry->name);
4587 }
4588 /* False, skip boolean config option. */
4589 goto end;
4590 }
4591 }
4592
4593 ret = set_option(long_options[i].val, entry->value, entry->name);
4594 goto end;
4595 }
4596
4597 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4598
4599 end:
4600 return ret;
4601 }
4602
4603 /*
4604 * daemon configuration loading and argument parsing
4605 */
4606 static int set_options(int argc, char **argv)
4607 {
4608 int ret = 0, c = 0, option_index = 0;
4609 int orig_optopt = optopt, orig_optind = optind;
4610 char *optstring;
4611 const char *config_path = NULL;
4612
4613 optstring = utils_generate_optstring(long_options,
4614 sizeof(long_options) / sizeof(struct option));
4615 if (!optstring) {
4616 ret = -ENOMEM;
4617 goto end;
4618 }
4619
4620 /* Check for the --config option */
4621 while ((c = getopt_long(argc, argv, optstring, long_options,
4622 &option_index)) != -1) {
4623 if (c == '?') {
4624 ret = -EINVAL;
4625 goto end;
4626 } else if (c != 'f') {
4627 /* if not equal to --config option. */
4628 continue;
4629 }
4630
4631 config_path = utils_expand_path(optarg);
4632 if (!config_path) {
4633 ERR("Failed to resolve path: %s", optarg);
4634 }
4635 }
4636
4637 ret = config_get_section_entries(config_path, config_section_name,
4638 config_entry_handler, NULL);
4639 if (ret) {
4640 if (ret > 0) {
4641 ERR("Invalid configuration option at line %i", ret);
4642 ret = -1;
4643 }
4644 goto end;
4645 }
4646
4647 /* Reset getopt's global state */
4648 optopt = orig_optopt;
4649 optind = orig_optind;
4650 while (1) {
4651 c = getopt_long(argc, argv, optstring, long_options, &option_index);
4652 if (c == -1) {
4653 break;
4654 }
4655
4656 ret = set_option(c, optarg, long_options[option_index].name);
4657 if (ret < 0) {
4658 break;
4659 }
4660 }
4661
4662 end:
4663 free(optstring);
4664 return ret;
4665 }
4666
4667 /*
4668 * Creates the two needed socket by the daemon.
4669 * apps_sock - The communication socket for all UST apps.
4670 * client_sock - The communication of the cli tool (lttng).
4671 */
4672 static int init_daemon_socket(void)
4673 {
4674 int ret = 0;
4675 mode_t old_umask;
4676
4677 old_umask = umask(0);
4678
4679 /* Create client tool unix socket */
4680 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4681 if (client_sock < 0) {
4682 ERR("Create unix sock failed: %s", client_unix_sock_path);
4683 ret = -1;
4684 goto end;
4685 }
4686
4687 /* Set the cloexec flag */
4688 ret = utils_set_fd_cloexec(client_sock);
4689 if (ret < 0) {
4690 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4691 "Continuing but note that the consumer daemon will have a "
4692 "reference to this socket on exec()", client_sock);
4693 }
4694
4695 /* File permission MUST be 660 */
4696 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4697 if (ret < 0) {
4698 ERR("Set file permissions failed: %s", client_unix_sock_path);
4699 PERROR("chmod");
4700 goto end;
4701 }
4702
4703 /* Create the application unix socket */
4704 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4705 if (apps_sock < 0) {
4706 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4707 ret = -1;
4708 goto end;
4709 }
4710
4711 /* Set the cloexec flag */
4712 ret = utils_set_fd_cloexec(apps_sock);
4713 if (ret < 0) {
4714 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4715 "Continuing but note that the consumer daemon will have a "
4716 "reference to this socket on exec()", apps_sock);
4717 }
4718
4719 /* File permission MUST be 666 */
4720 ret = chmod(apps_unix_sock_path,
4721 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4722 if (ret < 0) {
4723 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4724 PERROR("chmod");
4725 goto end;
4726 }
4727
4728 DBG3("Session daemon client socket %d and application socket %d created",
4729 client_sock, apps_sock);
4730
4731 end:
4732 umask(old_umask);
4733 return ret;
4734 }
4735
4736 /*
4737 * Check if the global socket is available, and if a daemon is answering at the
4738 * other side. If yes, error is returned.
4739 */
4740 static int check_existing_daemon(void)
4741 {
4742 /* Is there anybody out there ? */
4743 if (lttng_session_daemon_alive()) {
4744 return -EEXIST;
4745 }
4746
4747 return 0;
4748 }
4749
4750 /*
4751 * Set the tracing group gid onto the client socket.
4752 *
4753 * Race window between mkdir and chown is OK because we are going from more
4754 * permissive (root.root) to less permissive (root.tracing).
4755 */
4756 static int set_permissions(char *rundir)
4757 {
4758 int ret;
4759 gid_t gid;
4760
4761 gid = utils_get_group_id(tracing_group_name);
4762
4763 /* Set lttng run dir */
4764 ret = chown(rundir, 0, gid);
4765 if (ret < 0) {
4766 ERR("Unable to set group on %s", rundir);
4767 PERROR("chown");
4768 }
4769
4770 /*
4771 * Ensure all applications and tracing group can search the run
4772 * dir. Allow everyone to read the directory, since it does not
4773 * buy us anything to hide its content.
4774 */
4775 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4776 if (ret < 0) {
4777 ERR("Unable to set permissions on %s", rundir);
4778 PERROR("chmod");
4779 }
4780
4781 /* lttng client socket path */
4782 ret = chown(client_unix_sock_path, 0, gid);
4783 if (ret < 0) {
4784 ERR("Unable to set group on %s", client_unix_sock_path);
4785 PERROR("chown");
4786 }
4787
4788 /* kconsumer error socket path */
4789 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4790 if (ret < 0) {
4791 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4792 PERROR("chown");
4793 }
4794
4795 /* 64-bit ustconsumer error socket path */
4796 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4797 if (ret < 0) {
4798 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4799 PERROR("chown");
4800 }
4801
4802 /* 32-bit ustconsumer compat32 error socket path */
4803 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4804 if (ret < 0) {
4805 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4806 PERROR("chown");
4807 }
4808
4809 DBG("All permissions are set");
4810
4811 return ret;
4812 }
4813
4814 /*
4815 * Create the lttng run directory needed for all global sockets and pipe.
4816 */
4817 static int create_lttng_rundir(const char *rundir)
4818 {
4819 int ret;
4820
4821 DBG3("Creating LTTng run directory: %s", rundir);
4822
4823 ret = mkdir(rundir, S_IRWXU);
4824 if (ret < 0) {
4825 if (errno != EEXIST) {
4826 ERR("Unable to create %s", rundir);
4827 goto error;
4828 } else {
4829 ret = 0;
4830 }
4831 }
4832
4833 error:
4834 return ret;
4835 }
4836
4837 /*
4838 * Setup sockets and directory needed by the kconsumerd communication with the
4839 * session daemon.
4840 */
4841 static int set_consumer_sockets(struct consumer_data *consumer_data,
4842 const char *rundir)
4843 {
4844 int ret;
4845 char path[PATH_MAX];
4846
4847 switch (consumer_data->type) {
4848 case LTTNG_CONSUMER_KERNEL:
4849 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4850 break;
4851 case LTTNG_CONSUMER64_UST:
4852 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4853 break;
4854 case LTTNG_CONSUMER32_UST:
4855 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4856 break;
4857 default:
4858 ERR("Consumer type unknown");
4859 ret = -EINVAL;
4860 goto error;
4861 }
4862
4863 DBG2("Creating consumer directory: %s", path);
4864
4865 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4866 if (ret < 0) {
4867 if (errno != EEXIST) {
4868 PERROR("mkdir");
4869 ERR("Failed to create %s", path);
4870 goto error;
4871 }
4872 ret = -1;
4873 }
4874 if (is_root) {
4875 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4876 if (ret < 0) {
4877 ERR("Unable to set group on %s", path);
4878 PERROR("chown");
4879 goto error;
4880 }
4881 }
4882
4883 /* Create the kconsumerd error unix socket */
4884 consumer_data->err_sock =
4885 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4886 if (consumer_data->err_sock < 0) {
4887 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4888 ret = -1;
4889 goto error;
4890 }
4891
4892 /*
4893 * Set the CLOEXEC flag. Return code is useless because either way, the
4894 * show must go on.
4895 */
4896 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4897 if (ret < 0) {
4898 PERROR("utils_set_fd_cloexec");
4899 /* continue anyway */
4900 }
4901
4902 /* File permission MUST be 660 */
4903 ret = chmod(consumer_data->err_unix_sock_path,
4904 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4905 if (ret < 0) {
4906 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4907 PERROR("chmod");
4908 goto error;
4909 }
4910
4911 error:
4912 return ret;
4913 }
4914
4915 /*
4916 * Signal handler for the daemon
4917 *
4918 * Simply stop all worker threads, leaving main() return gracefully after
4919 * joining all threads and calling cleanup().
4920 */
4921 static void sighandler(int sig)
4922 {
4923 switch (sig) {
4924 case SIGPIPE:
4925 DBG("SIGPIPE caught");
4926 return;
4927 case SIGINT:
4928 DBG("SIGINT caught");
4929 stop_threads();
4930 break;
4931 case SIGTERM:
4932 DBG("SIGTERM caught");
4933 stop_threads();
4934 break;
4935 case SIGUSR1:
4936 CMM_STORE_SHARED(recv_child_signal, 1);
4937 break;
4938 default:
4939 break;
4940 }
4941 }
4942
4943 /*
4944 * Setup signal handler for :
4945 * SIGINT, SIGTERM, SIGPIPE
4946 */
4947 static int set_signal_handler(void)
4948 {
4949 int ret = 0;
4950 struct sigaction sa;
4951 sigset_t sigset;
4952
4953 if ((ret = sigemptyset(&sigset)) < 0) {
4954 PERROR("sigemptyset");
4955 return ret;
4956 }
4957
4958 sa.sa_handler = sighandler;
4959 sa.sa_mask = sigset;
4960 sa.sa_flags = 0;
4961 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4962 PERROR("sigaction");
4963 return ret;
4964 }
4965
4966 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4967 PERROR("sigaction");
4968 return ret;
4969 }
4970
4971 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4972 PERROR("sigaction");
4973 return ret;
4974 }
4975
4976 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4977 PERROR("sigaction");
4978 return ret;
4979 }
4980
4981 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4982
4983 return ret;
4984 }
4985
4986 /*
4987 * Set open files limit to unlimited. This daemon can open a large number of
4988 * file descriptors in order to consumer multiple kernel traces.
4989 */
4990 static void set_ulimit(void)
4991 {
4992 int ret;
4993 struct rlimit lim;
4994
4995 /* The kernel does not allowed an infinite limit for open files */
4996 lim.rlim_cur = 65535;
4997 lim.rlim_max = 65535;
4998
4999 ret = setrlimit(RLIMIT_NOFILE, &lim);
5000 if (ret < 0) {
5001 PERROR("failed to set open files limit");
5002 }
5003 }
5004
5005 /*
5006 * Write pidfile using the rundir and opt_pidfile.
5007 */
5008 static int write_pidfile(void)
5009 {
5010 int ret;
5011 char pidfile_path[PATH_MAX];
5012
5013 assert(rundir);
5014
5015 if (opt_pidfile) {
5016 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5017 } else {
5018 /* Build pidfile path from rundir and opt_pidfile. */
5019 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5020 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5021 if (ret < 0) {
5022 PERROR("snprintf pidfile path");
5023 goto error;
5024 }
5025 }
5026
5027 /*
5028 * Create pid file in rundir.
5029 */
5030 ret = utils_create_pid_file(getpid(), pidfile_path);
5031 error:
5032 return ret;
5033 }
5034
5035 /*
5036 * Create lockfile using the rundir and return its fd.
5037 */
5038 static int create_lockfile(void)
5039 {
5040 int ret;
5041 char lockfile_path[PATH_MAX];
5042
5043 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5044 if (ret < 0) {
5045 goto error;
5046 }
5047
5048 ret = utils_create_lock_file(lockfile_path);
5049 error:
5050 return ret;
5051 }
5052
5053 /*
5054 * Write agent TCP port using the rundir.
5055 */
5056 static int write_agent_port(void)
5057 {
5058 int ret;
5059 char path[PATH_MAX];
5060
5061 assert(rundir);
5062
5063 ret = snprintf(path, sizeof(path), "%s/"
5064 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5065 if (ret < 0) {
5066 PERROR("snprintf agent port path");
5067 goto error;
5068 }
5069
5070 /*
5071 * Create TCP agent port file in rundir.
5072 */
5073 ret = utils_create_pid_file(agent_tcp_port, path);
5074
5075 error:
5076 return ret;
5077 }
5078
5079 /*
5080 * main
5081 */
5082 int main(int argc, char **argv)
5083 {
5084 int ret = 0, retval = 0;
5085 void *status;
5086 const char *home_path, *env_app_timeout;
5087
5088 init_kernel_workarounds();
5089
5090 rcu_register_thread();
5091
5092 if (set_signal_handler()) {
5093 retval = -1;
5094 goto exit_set_signal_handler;
5095 }
5096
5097 setup_consumerd_path();
5098
5099 page_size = sysconf(_SC_PAGESIZE);
5100 if (page_size < 0) {
5101 PERROR("sysconf _SC_PAGESIZE");
5102 page_size = LONG_MAX;
5103 WARN("Fallback page size to %ld", page_size);
5104 }
5105
5106 /*
5107 * Parse arguments and load the daemon configuration file.
5108 *
5109 * We have an exit_options exit path to free memory reserved by
5110 * set_options. This is needed because the rest of sessiond_cleanup()
5111 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5112 * depends on set_options.
5113 */
5114 progname = argv[0];
5115 if (set_options(argc, argv)) {
5116 retval = -1;
5117 goto exit_options;
5118 }
5119
5120 /* Daemonize */
5121 if (opt_daemon || opt_background) {
5122 int i;
5123
5124 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5125 !opt_background);
5126 if (ret < 0) {
5127 retval = -1;
5128 goto exit_options;
5129 }
5130
5131 /*
5132 * We are in the child. Make sure all other file descriptors are
5133 * closed, in case we are called with more opened file
5134 * descriptors than the standard ones.
5135 */
5136 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5137 (void) close(i);
5138 }
5139 }
5140
5141 /*
5142 * Starting from here, we can create threads. This needs to be after
5143 * lttng_daemonize due to RCU.
5144 */
5145
5146 /*
5147 * Initialize the health check subsystem. This call should set the
5148 * appropriate time values.
5149 */
5150 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5151 if (!health_sessiond) {
5152 PERROR("health_app_create error");
5153 retval = -1;
5154 goto exit_health_sessiond_cleanup;
5155 }
5156
5157 if (init_ht_cleanup_quit_pipe()) {
5158 retval = -1;
5159 goto exit_ht_cleanup_quit_pipe;
5160 }
5161
5162 /* Setup the thread ht_cleanup communication pipe. */
5163 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5164 retval = -1;
5165 goto exit_ht_cleanup_pipe;
5166 }
5167
5168 /* Create thread to clean up RCU hash tables */
5169 ret = pthread_create(&ht_cleanup_thread, NULL,
5170 thread_ht_cleanup, (void *) NULL);
5171 if (ret) {
5172 errno = ret;
5173 PERROR("pthread_create ht_cleanup");
5174 retval = -1;
5175 goto exit_ht_cleanup;
5176 }
5177
5178 /* Create thread quit pipe */
5179 if (init_thread_quit_pipe()) {
5180 retval = -1;
5181 goto exit_init_data;
5182 }
5183
5184 /* Check if daemon is UID = 0 */
5185 is_root = !getuid();
5186
5187 if (is_root) {
5188 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5189 if (!rundir) {
5190 retval = -1;
5191 goto exit_init_data;
5192 }
5193
5194 /* Create global run dir with root access */
5195 if (create_lttng_rundir(rundir)) {
5196 retval = -1;
5197 goto exit_init_data;
5198 }
5199
5200 if (strlen(apps_unix_sock_path) == 0) {
5201 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5202 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5203 if (ret < 0) {
5204 retval = -1;
5205 goto exit_init_data;
5206 }
5207 }
5208
5209 if (strlen(client_unix_sock_path) == 0) {
5210 ret = snprintf(client_unix_sock_path, PATH_MAX,
5211 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5212 if (ret < 0) {
5213 retval = -1;
5214 goto exit_init_data;
5215 }
5216 }
5217
5218 /* Set global SHM for ust */
5219 if (strlen(wait_shm_path) == 0) {
5220 ret = snprintf(wait_shm_path, PATH_MAX,
5221 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5222 if (ret < 0) {
5223 retval = -1;
5224 goto exit_init_data;
5225 }
5226 }
5227
5228 if (strlen(health_unix_sock_path) == 0) {
5229 ret = snprintf(health_unix_sock_path,
5230 sizeof(health_unix_sock_path),
5231 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5232 if (ret < 0) {
5233 retval = -1;
5234 goto exit_init_data;
5235 }
5236 }
5237
5238 /* Setup kernel consumerd path */
5239 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5240 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5241 if (ret < 0) {
5242 retval = -1;
5243 goto exit_init_data;
5244 }
5245 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5246 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5247 if (ret < 0) {
5248 retval = -1;
5249 goto exit_init_data;
5250 }
5251
5252 DBG2("Kernel consumer err path: %s",
5253 kconsumer_data.err_unix_sock_path);
5254 DBG2("Kernel consumer cmd path: %s",
5255 kconsumer_data.cmd_unix_sock_path);
5256 } else {
5257 home_path = utils_get_home_dir();
5258 if (home_path == NULL) {
5259 /* TODO: Add --socket PATH option */
5260 ERR("Can't get HOME directory for sockets creation.");
5261 retval = -1;
5262 goto exit_init_data;
5263 }
5264
5265 /*
5266 * Create rundir from home path. This will create something like
5267 * $HOME/.lttng
5268 */
5269 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5270 if (ret < 0) {
5271 retval = -1;
5272 goto exit_init_data;
5273 }
5274
5275 if (create_lttng_rundir(rundir)) {
5276 retval = -1;
5277 goto exit_init_data;
5278 }
5279
5280 if (strlen(apps_unix_sock_path) == 0) {
5281 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5282 DEFAULT_HOME_APPS_UNIX_SOCK,
5283 home_path);
5284 if (ret < 0) {
5285 retval = -1;
5286 goto exit_init_data;
5287 }
5288 }
5289
5290 /* Set the cli tool unix socket path */
5291 if (strlen(client_unix_sock_path) == 0) {
5292 ret = snprintf(client_unix_sock_path, PATH_MAX,
5293 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5294 home_path);
5295 if (ret < 0) {
5296 retval = -1;
5297 goto exit_init_data;
5298 }
5299 }
5300
5301 /* Set global SHM for ust */
5302 if (strlen(wait_shm_path) == 0) {
5303 ret = snprintf(wait_shm_path, PATH_MAX,
5304 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5305 getuid());
5306 if (ret < 0) {
5307 retval = -1;
5308 goto exit_init_data;
5309 }
5310 }
5311
5312 /* Set health check Unix path */
5313 if (strlen(health_unix_sock_path) == 0) {
5314 ret = snprintf(health_unix_sock_path,
5315 sizeof(health_unix_sock_path),
5316 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5317 home_path);
5318 if (ret < 0) {
5319 retval = -1;
5320 goto exit_init_data;
5321 }
5322 }
5323 }
5324
5325 lockfile_fd = create_lockfile();
5326 if (lockfile_fd < 0) {
5327 retval = -1;
5328 goto exit_init_data;
5329 }
5330
5331 /* Set consumer initial state */
5332 kernel_consumerd_state = CONSUMER_STOPPED;
5333 ust_consumerd_state = CONSUMER_STOPPED;
5334
5335 DBG("Client socket path %s", client_unix_sock_path);
5336 DBG("Application socket path %s", apps_unix_sock_path);
5337 DBG("Application wait path %s", wait_shm_path);
5338 DBG("LTTng run directory path: %s", rundir);
5339
5340 /* 32 bits consumerd path setup */
5341 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5342 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5343 if (ret < 0) {
5344 PERROR("snprintf 32-bit consumer error socket path");
5345 retval = -1;
5346 goto exit_init_data;
5347 }
5348 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5349 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5350 if (ret < 0) {
5351 PERROR("snprintf 32-bit consumer command socket path");
5352 retval = -1;
5353 goto exit_init_data;
5354 }
5355
5356 DBG2("UST consumer 32 bits err path: %s",
5357 ustconsumer32_data.err_unix_sock_path);
5358 DBG2("UST consumer 32 bits cmd path: %s",
5359 ustconsumer32_data.cmd_unix_sock_path);
5360
5361 /* 64 bits consumerd path setup */
5362 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5363 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5364 if (ret < 0) {
5365 PERROR("snprintf 64-bit consumer error socket path");
5366 retval = -1;
5367 goto exit_init_data;
5368 }
5369 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5370 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5371 if (ret < 0) {
5372 PERROR("snprintf 64-bit consumer command socket path");
5373 retval = -1;
5374 goto exit_init_data;
5375 }
5376
5377 DBG2("UST consumer 64 bits err path: %s",
5378 ustconsumer64_data.err_unix_sock_path);
5379 DBG2("UST consumer 64 bits cmd path: %s",
5380 ustconsumer64_data.cmd_unix_sock_path);
5381
5382 /*
5383 * See if daemon already exist.
5384 */
5385 if (check_existing_daemon()) {
5386 ERR("Already running daemon.\n");
5387 /*
5388 * We do not goto exit because we must not cleanup()
5389 * because a daemon is already running.
5390 */
5391 retval = -1;
5392 goto exit_init_data;
5393 }
5394
5395 /*
5396 * Init UST app hash table. Alloc hash table before this point since
5397 * cleanup() can get called after that point.
5398 */
5399 if (ust_app_ht_alloc()) {
5400 retval = -1;
5401 goto exit_init_data;
5402 }
5403
5404 /* Initialize agent domain subsystem. */
5405 if (agent_setup()) {
5406 /* ENOMEM at this point. */
5407 retval = -1;
5408 goto exit_init_data;
5409 }
5410
5411 /*
5412 * These actions must be executed as root. We do that *after* setting up
5413 * the sockets path because we MUST make the check for another daemon using
5414 * those paths *before* trying to set the kernel consumer sockets and init
5415 * kernel tracer.
5416 */
5417 if (is_root) {
5418 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5419 retval = -1;
5420 goto exit_init_data;
5421 }
5422
5423 /* Setup kernel tracer */
5424 if (!opt_no_kernel) {
5425 init_kernel_tracer();
5426 if (kernel_tracer_fd >= 0) {
5427 ret = syscall_init_table();
5428 if (ret < 0) {
5429 ERR("Unable to populate syscall table. "
5430 "Syscall tracing won't work "
5431 "for this session daemon.");
5432 }
5433 }
5434 }
5435
5436 /* Set ulimit for open files */
5437 set_ulimit();
5438 }
5439 /* init lttng_fd tracking must be done after set_ulimit. */
5440 lttng_fd_init();
5441
5442 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5443 retval = -1;
5444 goto exit_init_data;
5445 }
5446
5447 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5448 retval = -1;
5449 goto exit_init_data;
5450 }
5451
5452 /* Setup the needed unix socket */
5453 if (init_daemon_socket()) {
5454 retval = -1;
5455 goto exit_init_data;
5456 }
5457
5458 /* Set credentials to socket */
5459 if (is_root && set_permissions(rundir)) {
5460 retval = -1;
5461 goto exit_init_data;
5462 }
5463
5464 /* Get parent pid if -S, --sig-parent is specified. */
5465 if (opt_sig_parent) {
5466 ppid = getppid();
5467 }
5468
5469 /* Setup the kernel pipe for waking up the kernel thread */
5470 if (is_root && !opt_no_kernel) {
5471 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5472 retval = -1;
5473 goto exit_init_data;
5474 }
5475 }
5476
5477 /* Setup the thread apps communication pipe. */
5478 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5479 retval = -1;
5480 goto exit_init_data;
5481 }
5482
5483 /* Setup the thread apps notify communication pipe. */
5484 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5485 retval = -1;
5486 goto exit_init_data;
5487 }
5488
5489 /* Initialize global buffer per UID and PID registry. */
5490 buffer_reg_init_uid_registry();
5491 buffer_reg_init_pid_registry();
5492
5493 /* Init UST command queue. */
5494 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5495
5496 /*
5497 * Get session list pointer. This pointer MUST NOT be free'd. This list
5498 * is statically declared in session.c
5499 */
5500 session_list_ptr = session_get_list();
5501
5502 /* Set up max poll set size */
5503 lttng_poll_set_max_size();
5504
5505 cmd_init();
5506
5507 /* Check for the application socket timeout env variable. */
5508 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5509 if (env_app_timeout) {
5510 app_socket_timeout = atoi(env_app_timeout);
5511 } else {
5512 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5513 }
5514
5515 ret = write_pidfile();
5516 if (ret) {
5517 ERR("Error in write_pidfile");
5518 retval = -1;
5519 goto exit_init_data;
5520 }
5521 ret = write_agent_port();
5522 if (ret) {
5523 ERR("Error in write_agent_port");
5524 retval = -1;
5525 goto exit_init_data;
5526 }
5527
5528 /* Initialize communication library */
5529 lttcomm_init();
5530 /* Initialize TCP timeout values */
5531 lttcomm_inet_init();
5532
5533 if (load_session_init_data(&load_info) < 0) {
5534 retval = -1;
5535 goto exit_init_data;
5536 }
5537 load_info->path = opt_load_session_path;
5538
5539 /* Create health-check thread */
5540 ret = pthread_create(&health_thread, NULL,
5541 thread_manage_health, (void *) NULL);
5542 if (ret) {
5543 errno = ret;
5544 PERROR("pthread_create health");
5545 retval = -1;
5546 goto exit_health;
5547 }
5548
5549 /* Create thread to manage the client socket */
5550 ret = pthread_create(&client_thread, NULL,
5551 thread_manage_clients, (void *) NULL);
5552 if (ret) {
5553 errno = ret;
5554 PERROR("pthread_create clients");
5555 retval = -1;
5556 goto exit_client;
5557 }
5558
5559 /* Create thread to dispatch registration */
5560 ret = pthread_create(&dispatch_thread, NULL,
5561 thread_dispatch_ust_registration, (void *) NULL);
5562 if (ret) {
5563 errno = ret;
5564 PERROR("pthread_create dispatch");
5565 retval = -1;
5566 goto exit_dispatch;
5567 }
5568
5569 /* Create thread to manage application registration. */
5570 ret = pthread_create(&reg_apps_thread, NULL,
5571 thread_registration_apps, (void *) NULL);
5572 if (ret) {
5573 errno = ret;
5574 PERROR("pthread_create registration");
5575 retval = -1;
5576 goto exit_reg_apps;
5577 }
5578
5579 /* Create thread to manage application socket */
5580 ret = pthread_create(&apps_thread, NULL,
5581 thread_manage_apps, (void *) NULL);
5582 if (ret) {
5583 errno = ret;
5584 PERROR("pthread_create apps");
5585 retval = -1;
5586 goto exit_apps;
5587 }
5588
5589 /* Create thread to manage application notify socket */
5590 ret = pthread_create(&apps_notify_thread, NULL,
5591 ust_thread_manage_notify, (void *) NULL);
5592 if (ret) {
5593 errno = ret;
5594 PERROR("pthread_create notify");
5595 retval = -1;
5596 goto exit_apps_notify;
5597 }
5598
5599 /* Create agent registration thread. */
5600 ret = pthread_create(&agent_reg_thread, NULL,
5601 agent_thread_manage_registration, (void *) NULL);
5602 if (ret) {
5603 errno = ret;
5604 PERROR("pthread_create agent");
5605 retval = -1;
5606 goto exit_agent_reg;
5607 }
5608
5609 /* Don't start this thread if kernel tracing is not requested nor root */
5610 if (is_root && !opt_no_kernel) {
5611 /* Create kernel thread to manage kernel event */
5612 ret = pthread_create(&kernel_thread, NULL,
5613 thread_manage_kernel, (void *) NULL);
5614 if (ret) {
5615 errno = ret;
5616 PERROR("pthread_create kernel");
5617 retval = -1;
5618 goto exit_kernel;
5619 }
5620 }
5621
5622 /* Create session loading thread. */
5623 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5624 load_info);
5625 if (ret) {
5626 errno = ret;
5627 PERROR("pthread_create load_session_thread");
5628 retval = -1;
5629 goto exit_load_session;
5630 }
5631
5632 /*
5633 * This is where we start awaiting program completion (e.g. through
5634 * signal that asks threads to teardown).
5635 */
5636
5637 ret = pthread_join(load_session_thread, &status);
5638 if (ret) {
5639 errno = ret;
5640 PERROR("pthread_join load_session_thread");
5641 retval = -1;
5642 }
5643 exit_load_session:
5644
5645 if (is_root && !opt_no_kernel) {
5646 ret = pthread_join(kernel_thread, &status);
5647 if (ret) {
5648 errno = ret;
5649 PERROR("pthread_join");
5650 retval = -1;
5651 }
5652 }
5653 exit_kernel:
5654
5655 ret = pthread_join(agent_reg_thread, &status);
5656 if (ret) {
5657 errno = ret;
5658 PERROR("pthread_join agent");
5659 retval = -1;
5660 }
5661 exit_agent_reg:
5662
5663 ret = pthread_join(apps_notify_thread, &status);
5664 if (ret) {
5665 errno = ret;
5666 PERROR("pthread_join apps notify");
5667 retval = -1;
5668 }
5669 exit_apps_notify:
5670
5671 ret = pthread_join(apps_thread, &status);
5672 if (ret) {
5673 errno = ret;
5674 PERROR("pthread_join apps");
5675 retval = -1;
5676 }
5677 exit_apps:
5678
5679 ret = pthread_join(reg_apps_thread, &status);
5680 if (ret) {
5681 errno = ret;
5682 PERROR("pthread_join");
5683 retval = -1;
5684 }
5685 exit_reg_apps:
5686
5687 ret = pthread_join(dispatch_thread, &status);
5688 if (ret) {
5689 errno = ret;
5690 PERROR("pthread_join");
5691 retval = -1;
5692 }
5693 exit_dispatch:
5694
5695 ret = pthread_join(client_thread, &status);
5696 if (ret) {
5697 errno = ret;
5698 PERROR("pthread_join");
5699 retval = -1;
5700 }
5701 exit_client:
5702
5703 ret = pthread_join(health_thread, &status);
5704 if (ret) {
5705 errno = ret;
5706 PERROR("pthread_join health thread");
5707 retval = -1;
5708 }
5709 exit_health:
5710
5711 exit_init_data:
5712 /*
5713 * sessiond_cleanup() is called when no other thread is running, except
5714 * the ht_cleanup thread, which is needed to destroy the hash tables.
5715 */
5716 rcu_thread_online();
5717 sessiond_cleanup();
5718 rcu_thread_offline();
5719 rcu_unregister_thread();
5720
5721 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
5722 if (ret < 0) {
5723 ERR("write error on ht_cleanup quit pipe");
5724 retval = -1;
5725 }
5726
5727 ret = pthread_join(ht_cleanup_thread, &status);
5728 if (ret) {
5729 errno = ret;
5730 PERROR("pthread_join ht cleanup thread");
5731 retval = -1;
5732 }
5733 exit_ht_cleanup:
5734
5735 utils_close_pipe(ht_cleanup_pipe);
5736 exit_ht_cleanup_pipe:
5737
5738 /*
5739 * Close the ht_cleanup quit pipe.
5740 */
5741 utils_close_pipe(ht_cleanup_quit_pipe);
5742 exit_ht_cleanup_quit_pipe:
5743
5744 health_app_destroy(health_sessiond);
5745 exit_health_sessiond_cleanup:
5746
5747 exit_options:
5748 sessiond_cleanup_options();
5749
5750 exit_set_signal_handler:
5751 if (!retval) {
5752 exit(EXIT_SUCCESS);
5753 } else {
5754 exit(EXIT_FAILURE);
5755 }
5756 }
This page took 0.186633 seconds and 6 git commands to generate.