Cleanup: Remove LTTNG_ENABLE_ALL_EVENT and LTTNG_DISABLE_ALL_EVENT
[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 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <config.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/config.h>
51
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
54 #include "channel.h"
55 #include "cmd.h"
56 #include "consumer.h"
57 #include "context.h"
58 #include "event.h"
59 #include "kernel.h"
60 #include "kernel-consumer.h"
61 #include "modprobe.h"
62 #include "shm.h"
63 #include "ust-ctl.h"
64 #include "ust-consumer.h"
65 #include "utils.h"
66 #include "fd-limit.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "agent-thread.h"
71 #include "save.h"
72 #include "load-session-thread.h"
73
74 #define CONSUMERD_FILE "lttng-consumerd"
75
76 const char *progname;
77 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
78 static int tracing_group_name_override;
79 static char *opt_pidfile;
80 static int opt_sig_parent;
81 static int opt_verbose_consumer;
82 static int opt_daemon, opt_background;
83 static int opt_no_kernel;
84 static char *opt_load_session_path;
85 static pid_t ppid; /* Parent PID for --sig-parent option */
86 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
87 static char *rundir;
88 static int lockfile_fd = -1;
89
90 /* Set to 1 when a SIGUSR1 signal is received. */
91 static int recv_child_signal;
92
93 /*
94 * Consumer daemon specific control data. Every value not initialized here is
95 * set to 0 by the static definition.
96 */
97 static struct consumer_data kconsumer_data = {
98 .type = LTTNG_CONSUMER_KERNEL,
99 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
100 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
101 .err_sock = -1,
102 .cmd_sock = -1,
103 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
104 .lock = PTHREAD_MUTEX_INITIALIZER,
105 .cond = PTHREAD_COND_INITIALIZER,
106 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
107 };
108 static struct consumer_data ustconsumer64_data = {
109 .type = LTTNG_CONSUMER64_UST,
110 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
111 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
112 .err_sock = -1,
113 .cmd_sock = -1,
114 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
115 .lock = PTHREAD_MUTEX_INITIALIZER,
116 .cond = PTHREAD_COND_INITIALIZER,
117 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
118 };
119 static struct consumer_data ustconsumer32_data = {
120 .type = LTTNG_CONSUMER32_UST,
121 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
122 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
123 .err_sock = -1,
124 .cmd_sock = -1,
125 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
126 .lock = PTHREAD_MUTEX_INITIALIZER,
127 .cond = PTHREAD_COND_INITIALIZER,
128 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
129 };
130
131 /* Command line options */
132 static const struct option long_options[] = {
133 { "client-sock", 1, 0, 'c' },
134 { "apps-sock", 1, 0, 'a' },
135 { "kconsumerd-cmd-sock", 1, 0, 'C' },
136 { "kconsumerd-err-sock", 1, 0, 'E' },
137 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
138 { "ustconsumerd32-err-sock", 1, 0, 'H' },
139 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
140 { "ustconsumerd64-err-sock", 1, 0, 'F' },
141 { "consumerd32-path", 1, 0, 'u' },
142 { "consumerd32-libdir", 1, 0, 'U' },
143 { "consumerd64-path", 1, 0, 't' },
144 { "consumerd64-libdir", 1, 0, 'T' },
145 { "daemonize", 0, 0, 'd' },
146 { "background", 0, 0, 'b' },
147 { "sig-parent", 0, 0, 'S' },
148 { "help", 0, 0, 'h' },
149 { "group", 1, 0, 'g' },
150 { "version", 0, 0, 'V' },
151 { "quiet", 0, 0, 'q' },
152 { "verbose", 0, 0, 'v' },
153 { "verbose-consumer", 0, 0, 'Z' },
154 { "no-kernel", 0, 0, 'N' },
155 { "pidfile", 1, 0, 'p' },
156 { "agent-tcp-port", 1, 0, 'J' },
157 { "config", 1, 0, 'f' },
158 { "load", 1, 0, 'l' },
159 { "kmod-probes", 1, 0, 'P' },
160 { "extra-kmod-probes", 1, 0, 'e' },
161 { NULL, 0, 0, 0 }
162 };
163
164 /* Command line options to ignore from configuration file */
165 static const char *config_ignore_options[] = { "help", "version", "config" };
166
167 /* Shared between threads */
168 static int dispatch_thread_exit;
169
170 /* Global application Unix socket path */
171 static char apps_unix_sock_path[PATH_MAX];
172 /* Global client Unix socket path */
173 static char client_unix_sock_path[PATH_MAX];
174 /* global wait shm path for UST */
175 static char wait_shm_path[PATH_MAX];
176 /* Global health check unix path */
177 static char health_unix_sock_path[PATH_MAX];
178
179 /* Sockets and FDs */
180 static int client_sock = -1;
181 static int apps_sock = -1;
182 int kernel_tracer_fd = -1;
183 static int kernel_poll_pipe[2] = { -1, -1 };
184
185 /*
186 * Quit pipe for all threads. This permits a single cancellation point
187 * for all threads when receiving an event on the pipe.
188 */
189 static int thread_quit_pipe[2] = { -1, -1 };
190
191 /*
192 * This pipe is used to inform the thread managing application communication
193 * that a command is queued and ready to be processed.
194 */
195 static int apps_cmd_pipe[2] = { -1, -1 };
196
197 int apps_cmd_notify_pipe[2] = { -1, -1 };
198
199 /* Pthread, Mutexes and Semaphores */
200 static pthread_t apps_thread;
201 static pthread_t apps_notify_thread;
202 static pthread_t reg_apps_thread;
203 static pthread_t client_thread;
204 static pthread_t kernel_thread;
205 static pthread_t dispatch_thread;
206 static pthread_t health_thread;
207 static pthread_t ht_cleanup_thread;
208 static pthread_t agent_reg_thread;
209 static pthread_t load_session_thread;
210
211 /*
212 * UST registration command queue. This queue is tied with a futex and uses a N
213 * wakers / 1 waiter implemented and detailed in futex.c/.h
214 *
215 * The thread_registration_apps and thread_dispatch_ust_registration uses this
216 * queue along with the wait/wake scheme. The thread_manage_apps receives down
217 * the line new application socket and monitors it for any I/O error or clean
218 * close that triggers an unregistration of the application.
219 */
220 static struct ust_cmd_queue ust_cmd_queue;
221
222 /*
223 * Pointer initialized before thread creation.
224 *
225 * This points to the tracing session list containing the session count and a
226 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
227 * MUST NOT be taken if you call a public function in session.c.
228 *
229 * The lock is nested inside the structure: session_list_ptr->lock. Please use
230 * session_lock_list and session_unlock_list for lock acquisition.
231 */
232 static struct ltt_session_list *session_list_ptr;
233
234 int ust_consumerd64_fd = -1;
235 int ust_consumerd32_fd = -1;
236
237 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
238 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
239 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
240 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
241 static int consumerd32_bin_override;
242 static int consumerd64_bin_override;
243 static int consumerd32_libdir_override;
244 static int consumerd64_libdir_override;
245
246 static const char *module_proc_lttng = "/proc/lttng";
247
248 /*
249 * Consumer daemon state which is changed when spawning it, killing it or in
250 * case of a fatal error.
251 */
252 enum consumerd_state {
253 CONSUMER_STARTED = 1,
254 CONSUMER_STOPPED = 2,
255 CONSUMER_ERROR = 3,
256 };
257
258 /*
259 * This consumer daemon state is used to validate if a client command will be
260 * able to reach the consumer. If not, the client is informed. For instance,
261 * doing a "lttng start" when the consumer state is set to ERROR will return an
262 * error to the client.
263 *
264 * The following example shows a possible race condition of this scheme:
265 *
266 * consumer thread error happens
267 * client cmd arrives
268 * client cmd checks state -> still OK
269 * consumer thread exit, sets error
270 * client cmd try to talk to consumer
271 * ...
272 *
273 * However, since the consumer is a different daemon, we have no way of making
274 * sure the command will reach it safely even with this state flag. This is why
275 * we consider that up to the state validation during command processing, the
276 * command is safe. After that, we can not guarantee the correctness of the
277 * client request vis-a-vis the consumer.
278 */
279 static enum consumerd_state ust_consumerd_state;
280 static enum consumerd_state kernel_consumerd_state;
281
282 /*
283 * Socket timeout for receiving and sending in seconds.
284 */
285 static int app_socket_timeout;
286
287 /* Set in main() with the current page size. */
288 long page_size;
289
290 /* Application health monitoring */
291 struct health_app *health_sessiond;
292
293 /* Agent TCP port for registration. Used by the agent thread. */
294 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
295
296 /* Am I root or not. */
297 int is_root; /* Set to 1 if the daemon is running as root */
298
299 const char * const config_section_name = "sessiond";
300
301 /* Load session thread information to operate. */
302 struct load_session_thread_data *load_info;
303
304 /*
305 * Whether sessiond is ready for commands/health check requests.
306 * NR_LTTNG_SESSIOND_READY must match the number of calls to
307 * sessiond_notify_ready().
308 */
309 #define NR_LTTNG_SESSIOND_READY 3
310 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
311
312 /* Notify parents that we are ready for cmd and health check */
313 LTTNG_HIDDEN
314 void sessiond_notify_ready(void)
315 {
316 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
317 /*
318 * Notify parent pid that we are ready to accept command
319 * for client side. This ppid is the one from the
320 * external process that spawned us.
321 */
322 if (opt_sig_parent) {
323 kill(ppid, SIGUSR1);
324 }
325
326 /*
327 * Notify the parent of the fork() process that we are
328 * ready.
329 */
330 if (opt_daemon || opt_background) {
331 kill(child_ppid, SIGUSR1);
332 }
333 }
334 }
335
336 static
337 void setup_consumerd_path(void)
338 {
339 const char *bin, *libdir;
340
341 /*
342 * Allow INSTALL_BIN_PATH to be used as a target path for the
343 * native architecture size consumer if CONFIG_CONSUMER*_PATH
344 * has not been defined.
345 */
346 #if (CAA_BITS_PER_LONG == 32)
347 if (!consumerd32_bin[0]) {
348 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
349 }
350 if (!consumerd32_libdir[0]) {
351 consumerd32_libdir = INSTALL_LIB_PATH;
352 }
353 #elif (CAA_BITS_PER_LONG == 64)
354 if (!consumerd64_bin[0]) {
355 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
356 }
357 if (!consumerd64_libdir[0]) {
358 consumerd64_libdir = INSTALL_LIB_PATH;
359 }
360 #else
361 #error "Unknown bitness"
362 #endif
363
364 /*
365 * runtime env. var. overrides the build default.
366 */
367 bin = getenv("LTTNG_CONSUMERD32_BIN");
368 if (bin) {
369 consumerd32_bin = bin;
370 }
371 bin = getenv("LTTNG_CONSUMERD64_BIN");
372 if (bin) {
373 consumerd64_bin = bin;
374 }
375 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
376 if (libdir) {
377 consumerd32_libdir = libdir;
378 }
379 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
380 if (libdir) {
381 consumerd64_libdir = libdir;
382 }
383 }
384
385 /*
386 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
387 */
388 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
389 {
390 int ret;
391
392 assert(events);
393
394 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
395 if (ret < 0) {
396 goto error;
397 }
398
399 /* Add quit pipe */
400 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
401 if (ret < 0) {
402 goto error;
403 }
404
405 return 0;
406
407 error:
408 return ret;
409 }
410
411 /*
412 * Check if the thread quit pipe was triggered.
413 *
414 * Return 1 if it was triggered else 0;
415 */
416 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
417 {
418 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
419 return 1;
420 }
421
422 return 0;
423 }
424
425 /*
426 * Init thread quit pipe.
427 *
428 * Return -1 on error or 0 if all pipes are created.
429 */
430 static int init_thread_quit_pipe(void)
431 {
432 int ret, i;
433
434 ret = pipe(thread_quit_pipe);
435 if (ret < 0) {
436 PERROR("thread quit pipe");
437 goto error;
438 }
439
440 for (i = 0; i < 2; i++) {
441 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
442 if (ret < 0) {
443 PERROR("fcntl");
444 goto error;
445 }
446 }
447
448 error:
449 return ret;
450 }
451
452 /*
453 * Stop all threads by closing the thread quit pipe.
454 */
455 static void stop_threads(void)
456 {
457 int ret;
458
459 /* Stopping all threads */
460 DBG("Terminating all threads");
461 ret = notify_thread_pipe(thread_quit_pipe[1]);
462 if (ret < 0) {
463 ERR("write error on thread quit pipe");
464 }
465
466 /* Dispatch thread */
467 CMM_STORE_SHARED(dispatch_thread_exit, 1);
468 futex_nto1_wake(&ust_cmd_queue.futex);
469 }
470
471 /*
472 * Close every consumer sockets.
473 */
474 static void close_consumer_sockets(void)
475 {
476 int ret;
477
478 if (kconsumer_data.err_sock >= 0) {
479 ret = close(kconsumer_data.err_sock);
480 if (ret < 0) {
481 PERROR("kernel consumer err_sock close");
482 }
483 }
484 if (ustconsumer32_data.err_sock >= 0) {
485 ret = close(ustconsumer32_data.err_sock);
486 if (ret < 0) {
487 PERROR("UST consumerd32 err_sock close");
488 }
489 }
490 if (ustconsumer64_data.err_sock >= 0) {
491 ret = close(ustconsumer64_data.err_sock);
492 if (ret < 0) {
493 PERROR("UST consumerd64 err_sock close");
494 }
495 }
496 if (kconsumer_data.cmd_sock >= 0) {
497 ret = close(kconsumer_data.cmd_sock);
498 if (ret < 0) {
499 PERROR("kernel consumer cmd_sock close");
500 }
501 }
502 if (ustconsumer32_data.cmd_sock >= 0) {
503 ret = close(ustconsumer32_data.cmd_sock);
504 if (ret < 0) {
505 PERROR("UST consumerd32 cmd_sock close");
506 }
507 }
508 if (ustconsumer64_data.cmd_sock >= 0) {
509 ret = close(ustconsumer64_data.cmd_sock);
510 if (ret < 0) {
511 PERROR("UST consumerd64 cmd_sock close");
512 }
513 }
514 }
515
516 /*
517 * Generate the full lock file path using the rundir.
518 *
519 * Return the snprintf() return value thus a negative value is an error.
520 */
521 static int generate_lock_file_path(char *path, size_t len)
522 {
523 int ret;
524
525 assert(path);
526 assert(rundir);
527
528 /* Build lockfile path from rundir. */
529 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
530 if (ret < 0) {
531 PERROR("snprintf lockfile path");
532 }
533
534 return ret;
535 }
536
537 /*
538 * Cleanup the daemon
539 */
540 static void cleanup(void)
541 {
542 int ret;
543 struct ltt_session *sess, *stmp;
544 char path[PATH_MAX];
545
546 DBG("Cleaning up");
547
548 /*
549 * Close the thread quit pipe. It has already done its job,
550 * since we are now called.
551 */
552 utils_close_pipe(thread_quit_pipe);
553
554 /*
555 * If opt_pidfile is undefined, the default file will be wiped when
556 * removing the rundir.
557 */
558 if (opt_pidfile) {
559 ret = remove(opt_pidfile);
560 if (ret < 0) {
561 PERROR("remove pidfile %s", opt_pidfile);
562 }
563 }
564
565 DBG("Removing sessiond and consumerd content of directory %s", rundir);
566
567 /* sessiond */
568 snprintf(path, PATH_MAX,
569 "%s/%s",
570 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
571 DBG("Removing %s", path);
572 (void) unlink(path);
573
574 snprintf(path, PATH_MAX, "%s/%s", rundir,
575 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
576 DBG("Removing %s", path);
577 (void) unlink(path);
578
579 /* kconsumerd */
580 snprintf(path, PATH_MAX,
581 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
582 rundir);
583 DBG("Removing %s", path);
584 (void) unlink(path);
585
586 snprintf(path, PATH_MAX,
587 DEFAULT_KCONSUMERD_PATH,
588 rundir);
589 DBG("Removing directory %s", path);
590 (void) rmdir(path);
591
592 /* ust consumerd 32 */
593 snprintf(path, PATH_MAX,
594 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
595 rundir);
596 DBG("Removing %s", path);
597 (void) unlink(path);
598
599 snprintf(path, PATH_MAX,
600 DEFAULT_USTCONSUMERD32_PATH,
601 rundir);
602 DBG("Removing directory %s", path);
603 (void) rmdir(path);
604
605 /* ust consumerd 64 */
606 snprintf(path, PATH_MAX,
607 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
608 rundir);
609 DBG("Removing %s", path);
610 (void) unlink(path);
611
612 snprintf(path, PATH_MAX,
613 DEFAULT_USTCONSUMERD64_PATH,
614 rundir);
615 DBG("Removing directory %s", path);
616 (void) rmdir(path);
617
618 DBG("Cleaning up all sessions");
619
620 /* Destroy session list mutex */
621 if (session_list_ptr != NULL) {
622 pthread_mutex_destroy(&session_list_ptr->lock);
623
624 /* Cleanup ALL session */
625 cds_list_for_each_entry_safe(sess, stmp,
626 &session_list_ptr->head, list) {
627 cmd_destroy_session(sess, kernel_poll_pipe[1]);
628 }
629 }
630
631 DBG("Closing all UST sockets");
632 ust_app_clean_list();
633 buffer_reg_destroy_registries();
634
635 if (is_root && !opt_no_kernel) {
636 DBG2("Closing kernel fd");
637 if (kernel_tracer_fd >= 0) {
638 ret = close(kernel_tracer_fd);
639 if (ret) {
640 PERROR("close");
641 }
642 }
643 DBG("Unloading kernel modules");
644 modprobe_remove_lttng_all();
645 }
646
647 close_consumer_sockets();
648
649 /*
650 * If the override option is set, the pointer points to a *non* const thus
651 * freeing it even though the variable type is set to const.
652 */
653 if (tracing_group_name_override) {
654 free((void *) tracing_group_name);
655 }
656 if (consumerd32_bin_override) {
657 free((void *) consumerd32_bin);
658 }
659 if (consumerd64_bin_override) {
660 free((void *) consumerd64_bin);
661 }
662 if (consumerd32_libdir_override) {
663 free((void *) consumerd32_libdir);
664 }
665 if (consumerd64_libdir_override) {
666 free((void *) consumerd64_libdir);
667 }
668
669 if (opt_pidfile) {
670 free(opt_pidfile);
671 }
672
673 if (opt_load_session_path) {
674 free(opt_load_session_path);
675 }
676
677 if (load_info) {
678 load_session_destroy_data(load_info);
679 free(load_info);
680 }
681
682 /*
683 * Cleanup lock file by deleting it and finaly closing it which will
684 * release the file system lock.
685 */
686 if (lockfile_fd >= 0) {
687 char lockfile_path[PATH_MAX];
688
689 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
690 if (ret > 0) {
691 ret = remove(lockfile_path);
692 if (ret < 0) {
693 PERROR("remove lock file");
694 }
695 ret = close(lockfile_fd);
696 if (ret < 0) {
697 PERROR("close lock file");
698 }
699 }
700 }
701
702 /*
703 * We do NOT rmdir rundir because there are other processes
704 * using it, for instance lttng-relayd, which can start in
705 * parallel with this teardown.
706 */
707
708 free(rundir);
709
710 /* <fun> */
711 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
712 "Matthew, BEET driven development works!%c[%dm",
713 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
714 /* </fun> */
715 }
716
717 /*
718 * Send data on a unix socket using the liblttsessiondcomm API.
719 *
720 * Return lttcomm error code.
721 */
722 static int send_unix_sock(int sock, void *buf, size_t len)
723 {
724 /* Check valid length */
725 if (len == 0) {
726 return -1;
727 }
728
729 return lttcomm_send_unix_sock(sock, buf, len);
730 }
731
732 /*
733 * Free memory of a command context structure.
734 */
735 static void clean_command_ctx(struct command_ctx **cmd_ctx)
736 {
737 DBG("Clean command context structure");
738 if (*cmd_ctx) {
739 if ((*cmd_ctx)->llm) {
740 free((*cmd_ctx)->llm);
741 }
742 if ((*cmd_ctx)->lsm) {
743 free((*cmd_ctx)->lsm);
744 }
745 free(*cmd_ctx);
746 *cmd_ctx = NULL;
747 }
748 }
749
750 /*
751 * Notify UST applications using the shm mmap futex.
752 */
753 static int notify_ust_apps(int active)
754 {
755 char *wait_shm_mmap;
756
757 DBG("Notifying applications of session daemon state: %d", active);
758
759 /* See shm.c for this call implying mmap, shm and futex calls */
760 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
761 if (wait_shm_mmap == NULL) {
762 goto error;
763 }
764
765 /* Wake waiting process */
766 futex_wait_update((int32_t *) wait_shm_mmap, active);
767
768 /* Apps notified successfully */
769 return 0;
770
771 error:
772 return -1;
773 }
774
775 /*
776 * Setup the outgoing data buffer for the response (llm) by allocating the
777 * right amount of memory and copying the original information from the lsm
778 * structure.
779 *
780 * Return total size of the buffer pointed by buf.
781 */
782 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
783 {
784 int ret, buf_size;
785
786 buf_size = size;
787
788 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
789 if (cmd_ctx->llm == NULL) {
790 PERROR("zmalloc");
791 ret = -ENOMEM;
792 goto error;
793 }
794
795 /* Copy common data */
796 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
797 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
798
799 cmd_ctx->llm->data_size = size;
800 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
801
802 return buf_size;
803
804 error:
805 return ret;
806 }
807
808 /*
809 * Update the kernel poll set of all channel fd available over all tracing
810 * session. Add the wakeup pipe at the end of the set.
811 */
812 static int update_kernel_poll(struct lttng_poll_event *events)
813 {
814 int ret;
815 struct ltt_session *session;
816 struct ltt_kernel_channel *channel;
817
818 DBG("Updating kernel poll set");
819
820 session_lock_list();
821 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
822 session_lock(session);
823 if (session->kernel_session == NULL) {
824 session_unlock(session);
825 continue;
826 }
827
828 cds_list_for_each_entry(channel,
829 &session->kernel_session->channel_list.head, list) {
830 /* Add channel fd to the kernel poll set */
831 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
832 if (ret < 0) {
833 session_unlock(session);
834 goto error;
835 }
836 DBG("Channel fd %d added to kernel set", channel->fd);
837 }
838 session_unlock(session);
839 }
840 session_unlock_list();
841
842 return 0;
843
844 error:
845 session_unlock_list();
846 return -1;
847 }
848
849 /*
850 * Find the channel fd from 'fd' over all tracing session. When found, check
851 * for new channel stream and send those stream fds to the kernel consumer.
852 *
853 * Useful for CPU hotplug feature.
854 */
855 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
856 {
857 int ret = 0;
858 struct ltt_session *session;
859 struct ltt_kernel_session *ksess;
860 struct ltt_kernel_channel *channel;
861
862 DBG("Updating kernel streams for channel fd %d", fd);
863
864 session_lock_list();
865 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
866 session_lock(session);
867 if (session->kernel_session == NULL) {
868 session_unlock(session);
869 continue;
870 }
871 ksess = session->kernel_session;
872
873 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
874 if (channel->fd == fd) {
875 DBG("Channel found, updating kernel streams");
876 ret = kernel_open_channel_stream(channel);
877 if (ret < 0) {
878 goto error;
879 }
880 /* Update the stream global counter */
881 ksess->stream_count_global += ret;
882
883 /*
884 * Have we already sent fds to the consumer? If yes, it means
885 * that tracing is started so it is safe to send our updated
886 * stream fds.
887 */
888 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
889 struct lttng_ht_iter iter;
890 struct consumer_socket *socket;
891
892 rcu_read_lock();
893 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
894 &iter.iter, socket, node.node) {
895 pthread_mutex_lock(socket->lock);
896 ret = kernel_consumer_send_channel_stream(socket,
897 channel, ksess,
898 session->output_traces ? 1 : 0);
899 pthread_mutex_unlock(socket->lock);
900 if (ret < 0) {
901 rcu_read_unlock();
902 goto error;
903 }
904 }
905 rcu_read_unlock();
906 }
907 goto error;
908 }
909 }
910 session_unlock(session);
911 }
912 session_unlock_list();
913 return ret;
914
915 error:
916 session_unlock(session);
917 session_unlock_list();
918 return ret;
919 }
920
921 /*
922 * For each tracing session, update newly registered apps. The session list
923 * lock MUST be acquired before calling this.
924 */
925 static void update_ust_app(int app_sock)
926 {
927 struct ltt_session *sess, *stmp;
928
929 /* Consumer is in an ERROR state. Stop any application update. */
930 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
931 /* Stop the update process since the consumer is dead. */
932 return;
933 }
934
935 /* For all tracing session(s) */
936 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
937 session_lock(sess);
938 if (sess->ust_session) {
939 ust_app_global_update(sess->ust_session, app_sock);
940 }
941 session_unlock(sess);
942 }
943 }
944
945 /*
946 * This thread manage event coming from the kernel.
947 *
948 * Features supported in this thread:
949 * -) CPU Hotplug
950 */
951 static void *thread_manage_kernel(void *data)
952 {
953 int ret, i, pollfd, update_poll_flag = 1, err = -1;
954 uint32_t revents, nb_fd;
955 char tmp;
956 struct lttng_poll_event events;
957
958 DBG("[thread] Thread manage kernel started");
959
960 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
961
962 /*
963 * This first step of the while is to clean this structure which could free
964 * non NULL pointers so initialize it before the loop.
965 */
966 lttng_poll_init(&events);
967
968 if (testpoint(sessiond_thread_manage_kernel)) {
969 goto error_testpoint;
970 }
971
972 health_code_update();
973
974 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
975 goto error_testpoint;
976 }
977
978 while (1) {
979 health_code_update();
980
981 if (update_poll_flag == 1) {
982 /* Clean events object. We are about to populate it again. */
983 lttng_poll_clean(&events);
984
985 ret = sessiond_set_thread_pollset(&events, 2);
986 if (ret < 0) {
987 goto error_poll_create;
988 }
989
990 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
991 if (ret < 0) {
992 goto error;
993 }
994
995 /* This will add the available kernel channel if any. */
996 ret = update_kernel_poll(&events);
997 if (ret < 0) {
998 goto error;
999 }
1000 update_poll_flag = 0;
1001 }
1002
1003 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
1004
1005 /* Poll infinite value of time */
1006 restart:
1007 health_poll_entry();
1008 ret = lttng_poll_wait(&events, -1);
1009 health_poll_exit();
1010 if (ret < 0) {
1011 /*
1012 * Restart interrupted system call.
1013 */
1014 if (errno == EINTR) {
1015 goto restart;
1016 }
1017 goto error;
1018 } else if (ret == 0) {
1019 /* Should not happen since timeout is infinite */
1020 ERR("Return value of poll is 0 with an infinite timeout.\n"
1021 "This should not have happened! Continuing...");
1022 continue;
1023 }
1024
1025 nb_fd = ret;
1026
1027 for (i = 0; i < nb_fd; i++) {
1028 /* Fetch once the poll data */
1029 revents = LTTNG_POLL_GETEV(&events, i);
1030 pollfd = LTTNG_POLL_GETFD(&events, i);
1031
1032 health_code_update();
1033
1034 /* Thread quit pipe has been closed. Killing thread. */
1035 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1036 if (ret) {
1037 err = 0;
1038 goto exit;
1039 }
1040
1041 /* Check for data on kernel pipe */
1042 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1043 (void) lttng_read(kernel_poll_pipe[0],
1044 &tmp, 1);
1045 /*
1046 * Ret value is useless here, if this pipe gets any actions an
1047 * update is required anyway.
1048 */
1049 update_poll_flag = 1;
1050 continue;
1051 } else {
1052 /*
1053 * New CPU detected by the kernel. Adding kernel stream to
1054 * kernel session and updating the kernel consumer
1055 */
1056 if (revents & LPOLLIN) {
1057 ret = update_kernel_stream(&kconsumer_data, pollfd);
1058 if (ret < 0) {
1059 continue;
1060 }
1061 break;
1062 /*
1063 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1064 * and unregister kernel stream at this point.
1065 */
1066 }
1067 }
1068 }
1069 }
1070
1071 exit:
1072 error:
1073 lttng_poll_clean(&events);
1074 error_poll_create:
1075 error_testpoint:
1076 utils_close_pipe(kernel_poll_pipe);
1077 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1078 if (err) {
1079 health_error();
1080 ERR("Health error occurred in %s", __func__);
1081 WARN("Kernel thread died unexpectedly. "
1082 "Kernel tracing can continue but CPU hotplug is disabled.");
1083 }
1084 health_unregister(health_sessiond);
1085 DBG("Kernel thread dying");
1086 return NULL;
1087 }
1088
1089 /*
1090 * Signal pthread condition of the consumer data that the thread.
1091 */
1092 static void signal_consumer_condition(struct consumer_data *data, int state)
1093 {
1094 pthread_mutex_lock(&data->cond_mutex);
1095
1096 /*
1097 * The state is set before signaling. It can be any value, it's the waiter
1098 * job to correctly interpret this condition variable associated to the
1099 * consumer pthread_cond.
1100 *
1101 * A value of 0 means that the corresponding thread of the consumer data
1102 * was not started. 1 indicates that the thread has started and is ready
1103 * for action. A negative value means that there was an error during the
1104 * thread bootstrap.
1105 */
1106 data->consumer_thread_is_ready = state;
1107 (void) pthread_cond_signal(&data->cond);
1108
1109 pthread_mutex_unlock(&data->cond_mutex);
1110 }
1111
1112 /*
1113 * This thread manage the consumer error sent back to the session daemon.
1114 */
1115 static void *thread_manage_consumer(void *data)
1116 {
1117 int sock = -1, i, ret, pollfd, err = -1;
1118 uint32_t revents, nb_fd;
1119 enum lttcomm_return_code code;
1120 struct lttng_poll_event events;
1121 struct consumer_data *consumer_data = data;
1122
1123 DBG("[thread] Manage consumer started");
1124
1125 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1126
1127 health_code_update();
1128
1129 /*
1130 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1131 * metadata_sock. Nothing more will be added to this poll set.
1132 */
1133 ret = sessiond_set_thread_pollset(&events, 3);
1134 if (ret < 0) {
1135 goto error_poll;
1136 }
1137
1138 /*
1139 * The error socket here is already in a listening state which was done
1140 * just before spawning this thread to avoid a race between the consumer
1141 * daemon exec trying to connect and the listen() call.
1142 */
1143 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1144 if (ret < 0) {
1145 goto error;
1146 }
1147
1148 health_code_update();
1149
1150 /* Infinite blocking call, waiting for transmission */
1151 restart:
1152 health_poll_entry();
1153
1154 if (testpoint(sessiond_thread_manage_consumer)) {
1155 goto error;
1156 }
1157
1158 ret = lttng_poll_wait(&events, -1);
1159 health_poll_exit();
1160 if (ret < 0) {
1161 /*
1162 * Restart interrupted system call.
1163 */
1164 if (errno == EINTR) {
1165 goto restart;
1166 }
1167 goto error;
1168 }
1169
1170 nb_fd = ret;
1171
1172 for (i = 0; i < nb_fd; i++) {
1173 /* Fetch once the poll data */
1174 revents = LTTNG_POLL_GETEV(&events, i);
1175 pollfd = LTTNG_POLL_GETFD(&events, i);
1176
1177 health_code_update();
1178
1179 /* Thread quit pipe has been closed. Killing thread. */
1180 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1181 if (ret) {
1182 err = 0;
1183 goto exit;
1184 }
1185
1186 /* Event on the registration socket */
1187 if (pollfd == consumer_data->err_sock) {
1188 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1189 ERR("consumer err socket poll error");
1190 goto error;
1191 }
1192 }
1193 }
1194
1195 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1196 if (sock < 0) {
1197 goto error;
1198 }
1199
1200 /*
1201 * Set the CLOEXEC flag. Return code is useless because either way, the
1202 * show must go on.
1203 */
1204 (void) utils_set_fd_cloexec(sock);
1205
1206 health_code_update();
1207
1208 DBG2("Receiving code from consumer err_sock");
1209
1210 /* Getting status code from kconsumerd */
1211 ret = lttcomm_recv_unix_sock(sock, &code,
1212 sizeof(enum lttcomm_return_code));
1213 if (ret <= 0) {
1214 goto error;
1215 }
1216
1217 health_code_update();
1218 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1219 /* Connect both socket, command and metadata. */
1220 consumer_data->cmd_sock =
1221 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1222 consumer_data->metadata_fd =
1223 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1224 if (consumer_data->cmd_sock < 0
1225 || consumer_data->metadata_fd < 0) {
1226 PERROR("consumer connect cmd socket");
1227 /* On error, signal condition and quit. */
1228 signal_consumer_condition(consumer_data, -1);
1229 goto error;
1230 }
1231 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1232 /* Create metadata socket lock. */
1233 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1234 if (consumer_data->metadata_sock.lock == NULL) {
1235 PERROR("zmalloc pthread mutex");
1236 ret = -1;
1237 goto error;
1238 }
1239 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1240
1241 signal_consumer_condition(consumer_data, 1);
1242 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1243 DBG("Consumer metadata socket ready (fd: %d)",
1244 consumer_data->metadata_fd);
1245 } else {
1246 ERR("consumer error when waiting for SOCK_READY : %s",
1247 lttcomm_get_readable_code(-code));
1248 goto error;
1249 }
1250
1251 /* Remove the consumerd error sock since we've established a connexion */
1252 ret = lttng_poll_del(&events, consumer_data->err_sock);
1253 if (ret < 0) {
1254 goto error;
1255 }
1256
1257 /* Add new accepted error socket. */
1258 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1259 if (ret < 0) {
1260 goto error;
1261 }
1262
1263 /* Add metadata socket that is successfully connected. */
1264 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1265 LPOLLIN | LPOLLRDHUP);
1266 if (ret < 0) {
1267 goto error;
1268 }
1269
1270 health_code_update();
1271
1272 /* Infinite blocking call, waiting for transmission */
1273 restart_poll:
1274 while (1) {
1275 health_poll_entry();
1276 ret = lttng_poll_wait(&events, -1);
1277 health_poll_exit();
1278 if (ret < 0) {
1279 /*
1280 * Restart interrupted system call.
1281 */
1282 if (errno == EINTR) {
1283 goto restart_poll;
1284 }
1285 goto error;
1286 }
1287
1288 nb_fd = ret;
1289
1290 for (i = 0; i < nb_fd; i++) {
1291 /* Fetch once the poll data */
1292 revents = LTTNG_POLL_GETEV(&events, i);
1293 pollfd = LTTNG_POLL_GETFD(&events, i);
1294
1295 health_code_update();
1296
1297 /* Thread quit pipe has been closed. Killing thread. */
1298 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1299 if (ret) {
1300 err = 0;
1301 goto exit;
1302 }
1303
1304 if (pollfd == sock) {
1305 /* Event on the consumerd socket */
1306 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1307 ERR("consumer err socket second poll error");
1308 goto error;
1309 }
1310 health_code_update();
1311 /* Wait for any kconsumerd error */
1312 ret = lttcomm_recv_unix_sock(sock, &code,
1313 sizeof(enum lttcomm_return_code));
1314 if (ret <= 0) {
1315 ERR("consumer closed the command socket");
1316 goto error;
1317 }
1318
1319 ERR("consumer return code : %s",
1320 lttcomm_get_readable_code(-code));
1321
1322 goto exit;
1323 } else if (pollfd == consumer_data->metadata_fd) {
1324 /* UST metadata requests */
1325 ret = ust_consumer_metadata_request(
1326 &consumer_data->metadata_sock);
1327 if (ret < 0) {
1328 ERR("Handling metadata request");
1329 goto error;
1330 }
1331 break;
1332 } else {
1333 ERR("Unknown pollfd");
1334 goto error;
1335 }
1336 }
1337 health_code_update();
1338 }
1339
1340 exit:
1341 error:
1342 /*
1343 * We lock here because we are about to close the sockets and some other
1344 * thread might be using them so get exclusive access which will abort all
1345 * other consumer command by other threads.
1346 */
1347 pthread_mutex_lock(&consumer_data->lock);
1348
1349 /* Immediately set the consumerd state to stopped */
1350 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1351 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1352 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1353 consumer_data->type == LTTNG_CONSUMER32_UST) {
1354 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1355 } else {
1356 /* Code flow error... */
1357 assert(0);
1358 }
1359
1360 if (consumer_data->err_sock >= 0) {
1361 ret = close(consumer_data->err_sock);
1362 if (ret) {
1363 PERROR("close");
1364 }
1365 consumer_data->err_sock = -1;
1366 }
1367 if (consumer_data->cmd_sock >= 0) {
1368 ret = close(consumer_data->cmd_sock);
1369 if (ret) {
1370 PERROR("close");
1371 }
1372 consumer_data->cmd_sock = -1;
1373 }
1374 if (consumer_data->metadata_sock.fd_ptr &&
1375 *consumer_data->metadata_sock.fd_ptr >= 0) {
1376 ret = close(*consumer_data->metadata_sock.fd_ptr);
1377 if (ret) {
1378 PERROR("close");
1379 }
1380 }
1381 if (sock >= 0) {
1382 ret = close(sock);
1383 if (ret) {
1384 PERROR("close");
1385 }
1386 }
1387
1388 unlink(consumer_data->err_unix_sock_path);
1389 unlink(consumer_data->cmd_unix_sock_path);
1390 consumer_data->pid = 0;
1391 pthread_mutex_unlock(&consumer_data->lock);
1392
1393 /* Cleanup metadata socket mutex. */
1394 if (consumer_data->metadata_sock.lock) {
1395 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1396 free(consumer_data->metadata_sock.lock);
1397 }
1398 lttng_poll_clean(&events);
1399 error_poll:
1400 if (err) {
1401 health_error();
1402 ERR("Health error occurred in %s", __func__);
1403 }
1404 health_unregister(health_sessiond);
1405 DBG("consumer thread cleanup completed");
1406
1407 return NULL;
1408 }
1409
1410 /*
1411 * This thread manage application communication.
1412 */
1413 static void *thread_manage_apps(void *data)
1414 {
1415 int i, ret, pollfd, err = -1;
1416 ssize_t size_ret;
1417 uint32_t revents, nb_fd;
1418 struct lttng_poll_event events;
1419
1420 DBG("[thread] Manage application started");
1421
1422 rcu_register_thread();
1423 rcu_thread_online();
1424
1425 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1426
1427 if (testpoint(sessiond_thread_manage_apps)) {
1428 goto error_testpoint;
1429 }
1430
1431 health_code_update();
1432
1433 ret = sessiond_set_thread_pollset(&events, 2);
1434 if (ret < 0) {
1435 goto error_poll_create;
1436 }
1437
1438 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1439 if (ret < 0) {
1440 goto error;
1441 }
1442
1443 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1444 goto error;
1445 }
1446
1447 health_code_update();
1448
1449 while (1) {
1450 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1451
1452 /* Inifinite blocking call, waiting for transmission */
1453 restart:
1454 health_poll_entry();
1455 ret = lttng_poll_wait(&events, -1);
1456 health_poll_exit();
1457 if (ret < 0) {
1458 /*
1459 * Restart interrupted system call.
1460 */
1461 if (errno == EINTR) {
1462 goto restart;
1463 }
1464 goto error;
1465 }
1466
1467 nb_fd = ret;
1468
1469 for (i = 0; i < nb_fd; i++) {
1470 /* Fetch once the poll data */
1471 revents = LTTNG_POLL_GETEV(&events, i);
1472 pollfd = LTTNG_POLL_GETFD(&events, i);
1473
1474 health_code_update();
1475
1476 /* Thread quit pipe has been closed. Killing thread. */
1477 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1478 if (ret) {
1479 err = 0;
1480 goto exit;
1481 }
1482
1483 /* Inspect the apps cmd pipe */
1484 if (pollfd == apps_cmd_pipe[0]) {
1485 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1486 ERR("Apps command pipe error");
1487 goto error;
1488 } else if (revents & LPOLLIN) {
1489 int sock;
1490
1491 /* Empty pipe */
1492 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1493 if (size_ret < sizeof(sock)) {
1494 PERROR("read apps cmd pipe");
1495 goto error;
1496 }
1497
1498 health_code_update();
1499
1500 /*
1501 * We only monitor the error events of the socket. This
1502 * thread does not handle any incoming data from UST
1503 * (POLLIN).
1504 */
1505 ret = lttng_poll_add(&events, sock,
1506 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1507 if (ret < 0) {
1508 goto error;
1509 }
1510
1511 DBG("Apps with sock %d added to poll set", sock);
1512 }
1513 } else {
1514 /*
1515 * At this point, we know that a registered application made
1516 * the event at poll_wait.
1517 */
1518 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1519 /* Removing from the poll set */
1520 ret = lttng_poll_del(&events, pollfd);
1521 if (ret < 0) {
1522 goto error;
1523 }
1524
1525 /* Socket closed on remote end. */
1526 ust_app_unregister(pollfd);
1527 }
1528 }
1529
1530 health_code_update();
1531 }
1532 }
1533
1534 exit:
1535 error:
1536 lttng_poll_clean(&events);
1537 error_poll_create:
1538 error_testpoint:
1539 utils_close_pipe(apps_cmd_pipe);
1540 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1541
1542 /*
1543 * We don't clean the UST app hash table here since already registered
1544 * applications can still be controlled so let them be until the session
1545 * daemon dies or the applications stop.
1546 */
1547
1548 if (err) {
1549 health_error();
1550 ERR("Health error occurred in %s", __func__);
1551 }
1552 health_unregister(health_sessiond);
1553 DBG("Application communication apps thread cleanup complete");
1554 rcu_thread_offline();
1555 rcu_unregister_thread();
1556 return NULL;
1557 }
1558
1559 /*
1560 * Send a socket to a thread This is called from the dispatch UST registration
1561 * thread once all sockets are set for the application.
1562 *
1563 * The sock value can be invalid, we don't really care, the thread will handle
1564 * it and make the necessary cleanup if so.
1565 *
1566 * On success, return 0 else a negative value being the errno message of the
1567 * write().
1568 */
1569 static int send_socket_to_thread(int fd, int sock)
1570 {
1571 ssize_t ret;
1572
1573 /*
1574 * It's possible that the FD is set as invalid with -1 concurrently just
1575 * before calling this function being a shutdown state of the thread.
1576 */
1577 if (fd < 0) {
1578 ret = -EBADF;
1579 goto error;
1580 }
1581
1582 ret = lttng_write(fd, &sock, sizeof(sock));
1583 if (ret < sizeof(sock)) {
1584 PERROR("write apps pipe %d", fd);
1585 if (ret < 0) {
1586 ret = -errno;
1587 }
1588 goto error;
1589 }
1590
1591 /* All good. Don't send back the write positive ret value. */
1592 ret = 0;
1593 error:
1594 return (int) ret;
1595 }
1596
1597 /*
1598 * Sanitize the wait queue of the dispatch registration thread meaning removing
1599 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1600 * notify socket is never received.
1601 */
1602 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1603 {
1604 int ret, nb_fd = 0, i;
1605 unsigned int fd_added = 0;
1606 struct lttng_poll_event events;
1607 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1608
1609 assert(wait_queue);
1610
1611 lttng_poll_init(&events);
1612
1613 /* Just skip everything for an empty queue. */
1614 if (!wait_queue->count) {
1615 goto end;
1616 }
1617
1618 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1619 if (ret < 0) {
1620 goto error_create;
1621 }
1622
1623 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1624 &wait_queue->head, head) {
1625 assert(wait_node->app);
1626 ret = lttng_poll_add(&events, wait_node->app->sock,
1627 LPOLLHUP | LPOLLERR);
1628 if (ret < 0) {
1629 goto error;
1630 }
1631
1632 fd_added = 1;
1633 }
1634
1635 if (!fd_added) {
1636 goto end;
1637 }
1638
1639 /*
1640 * Poll but don't block so we can quickly identify the faulty events and
1641 * clean them afterwards from the wait queue.
1642 */
1643 ret = lttng_poll_wait(&events, 0);
1644 if (ret < 0) {
1645 goto error;
1646 }
1647 nb_fd = ret;
1648
1649 for (i = 0; i < nb_fd; i++) {
1650 /* Get faulty FD. */
1651 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1652 int pollfd = LTTNG_POLL_GETFD(&events, i);
1653
1654 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1655 &wait_queue->head, head) {
1656 if (pollfd == wait_node->app->sock &&
1657 (revents & (LPOLLHUP | LPOLLERR))) {
1658 cds_list_del(&wait_node->head);
1659 wait_queue->count--;
1660 ust_app_destroy(wait_node->app);
1661 free(wait_node);
1662 break;
1663 }
1664 }
1665 }
1666
1667 if (nb_fd > 0) {
1668 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1669 }
1670
1671 end:
1672 lttng_poll_clean(&events);
1673 return;
1674
1675 error:
1676 lttng_poll_clean(&events);
1677 error_create:
1678 ERR("Unable to sanitize wait queue");
1679 return;
1680 }
1681
1682 /*
1683 * Dispatch request from the registration threads to the application
1684 * communication thread.
1685 */
1686 static void *thread_dispatch_ust_registration(void *data)
1687 {
1688 int ret, err = -1;
1689 struct cds_wfcq_node *node;
1690 struct ust_command *ust_cmd = NULL;
1691 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1692 struct ust_reg_wait_queue wait_queue = {
1693 .count = 0,
1694 };
1695
1696 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1697
1698 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1699 goto error_testpoint;
1700 }
1701
1702 health_code_update();
1703
1704 CDS_INIT_LIST_HEAD(&wait_queue.head);
1705
1706 DBG("[thread] Dispatch UST command started");
1707
1708 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1709 health_code_update();
1710
1711 /* Atomically prepare the queue futex */
1712 futex_nto1_prepare(&ust_cmd_queue.futex);
1713
1714 do {
1715 struct ust_app *app = NULL;
1716 ust_cmd = NULL;
1717
1718 /*
1719 * Make sure we don't have node(s) that have hung up before receiving
1720 * the notify socket. This is to clean the list in order to avoid
1721 * memory leaks from notify socket that are never seen.
1722 */
1723 sanitize_wait_queue(&wait_queue);
1724
1725 health_code_update();
1726 /* Dequeue command for registration */
1727 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1728 if (node == NULL) {
1729 DBG("Woken up but nothing in the UST command queue");
1730 /* Continue thread execution */
1731 break;
1732 }
1733
1734 ust_cmd = caa_container_of(node, struct ust_command, node);
1735
1736 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1737 " gid:%d sock:%d name:%s (version %d.%d)",
1738 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1739 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1740 ust_cmd->sock, ust_cmd->reg_msg.name,
1741 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1742
1743 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1744 wait_node = zmalloc(sizeof(*wait_node));
1745 if (!wait_node) {
1746 PERROR("zmalloc wait_node dispatch");
1747 ret = close(ust_cmd->sock);
1748 if (ret < 0) {
1749 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1750 }
1751 lttng_fd_put(LTTNG_FD_APPS, 1);
1752 free(ust_cmd);
1753 goto error;
1754 }
1755 CDS_INIT_LIST_HEAD(&wait_node->head);
1756
1757 /* Create application object if socket is CMD. */
1758 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1759 ust_cmd->sock);
1760 if (!wait_node->app) {
1761 ret = close(ust_cmd->sock);
1762 if (ret < 0) {
1763 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1764 }
1765 lttng_fd_put(LTTNG_FD_APPS, 1);
1766 free(wait_node);
1767 free(ust_cmd);
1768 continue;
1769 }
1770 /*
1771 * Add application to the wait queue so we can set the notify
1772 * socket before putting this object in the global ht.
1773 */
1774 cds_list_add(&wait_node->head, &wait_queue.head);
1775 wait_queue.count++;
1776
1777 free(ust_cmd);
1778 /*
1779 * We have to continue here since we don't have the notify
1780 * socket and the application MUST be added to the hash table
1781 * only at that moment.
1782 */
1783 continue;
1784 } else {
1785 /*
1786 * Look for the application in the local wait queue and set the
1787 * notify socket if found.
1788 */
1789 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1790 &wait_queue.head, head) {
1791 health_code_update();
1792 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1793 wait_node->app->notify_sock = ust_cmd->sock;
1794 cds_list_del(&wait_node->head);
1795 wait_queue.count--;
1796 app = wait_node->app;
1797 free(wait_node);
1798 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1799 break;
1800 }
1801 }
1802
1803 /*
1804 * With no application at this stage the received socket is
1805 * basically useless so close it before we free the cmd data
1806 * structure for good.
1807 */
1808 if (!app) {
1809 ret = close(ust_cmd->sock);
1810 if (ret < 0) {
1811 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1812 }
1813 lttng_fd_put(LTTNG_FD_APPS, 1);
1814 }
1815 free(ust_cmd);
1816 }
1817
1818 if (app) {
1819 /*
1820 * @session_lock_list
1821 *
1822 * Lock the global session list so from the register up to the
1823 * registration done message, no thread can see the application
1824 * and change its state.
1825 */
1826 session_lock_list();
1827 rcu_read_lock();
1828
1829 /*
1830 * Add application to the global hash table. This needs to be
1831 * done before the update to the UST registry can locate the
1832 * application.
1833 */
1834 ust_app_add(app);
1835
1836 /* Set app version. This call will print an error if needed. */
1837 (void) ust_app_version(app);
1838
1839 /* Send notify socket through the notify pipe. */
1840 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1841 app->notify_sock);
1842 if (ret < 0) {
1843 rcu_read_unlock();
1844 session_unlock_list();
1845 /*
1846 * No notify thread, stop the UST tracing. However, this is
1847 * not an internal error of the this thread thus setting
1848 * the health error code to a normal exit.
1849 */
1850 err = 0;
1851 goto error;
1852 }
1853
1854 /*
1855 * Update newly registered application with the tracing
1856 * registry info already enabled information.
1857 */
1858 update_ust_app(app->sock);
1859
1860 /*
1861 * Don't care about return value. Let the manage apps threads
1862 * handle app unregistration upon socket close.
1863 */
1864 (void) ust_app_register_done(app->sock);
1865
1866 /*
1867 * Even if the application socket has been closed, send the app
1868 * to the thread and unregistration will take place at that
1869 * place.
1870 */
1871 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1872 if (ret < 0) {
1873 rcu_read_unlock();
1874 session_unlock_list();
1875 /*
1876 * No apps. thread, stop the UST tracing. However, this is
1877 * not an internal error of the this thread thus setting
1878 * the health error code to a normal exit.
1879 */
1880 err = 0;
1881 goto error;
1882 }
1883
1884 rcu_read_unlock();
1885 session_unlock_list();
1886 }
1887 } while (node != NULL);
1888
1889 health_poll_entry();
1890 /* Futex wait on queue. Blocking call on futex() */
1891 futex_nto1_wait(&ust_cmd_queue.futex);
1892 health_poll_exit();
1893 }
1894 /* Normal exit, no error */
1895 err = 0;
1896
1897 error:
1898 /* Clean up wait queue. */
1899 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1900 &wait_queue.head, head) {
1901 cds_list_del(&wait_node->head);
1902 wait_queue.count--;
1903 free(wait_node);
1904 }
1905
1906 error_testpoint:
1907 DBG("Dispatch thread dying");
1908 if (err) {
1909 health_error();
1910 ERR("Health error occurred in %s", __func__);
1911 }
1912 health_unregister(health_sessiond);
1913 return NULL;
1914 }
1915
1916 /*
1917 * This thread manage application registration.
1918 */
1919 static void *thread_registration_apps(void *data)
1920 {
1921 int sock = -1, i, ret, pollfd, err = -1;
1922 uint32_t revents, nb_fd;
1923 struct lttng_poll_event events;
1924 /*
1925 * Get allocated in this thread, enqueued to a global queue, dequeued and
1926 * freed in the manage apps thread.
1927 */
1928 struct ust_command *ust_cmd = NULL;
1929
1930 DBG("[thread] Manage application registration started");
1931
1932 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1933
1934 if (testpoint(sessiond_thread_registration_apps)) {
1935 goto error_testpoint;
1936 }
1937
1938 ret = lttcomm_listen_unix_sock(apps_sock);
1939 if (ret < 0) {
1940 goto error_listen;
1941 }
1942
1943 /*
1944 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1945 * more will be added to this poll set.
1946 */
1947 ret = sessiond_set_thread_pollset(&events, 2);
1948 if (ret < 0) {
1949 goto error_create_poll;
1950 }
1951
1952 /* Add the application registration socket */
1953 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1954 if (ret < 0) {
1955 goto error_poll_add;
1956 }
1957
1958 /* Notify all applications to register */
1959 ret = notify_ust_apps(1);
1960 if (ret < 0) {
1961 ERR("Failed to notify applications or create the wait shared memory.\n"
1962 "Execution continues but there might be problem for already\n"
1963 "running applications that wishes to register.");
1964 }
1965
1966 while (1) {
1967 DBG("Accepting application registration");
1968
1969 /* Inifinite blocking call, waiting for transmission */
1970 restart:
1971 health_poll_entry();
1972 ret = lttng_poll_wait(&events, -1);
1973 health_poll_exit();
1974 if (ret < 0) {
1975 /*
1976 * Restart interrupted system call.
1977 */
1978 if (errno == EINTR) {
1979 goto restart;
1980 }
1981 goto error;
1982 }
1983
1984 nb_fd = ret;
1985
1986 for (i = 0; i < nb_fd; i++) {
1987 health_code_update();
1988
1989 /* Fetch once the poll data */
1990 revents = LTTNG_POLL_GETEV(&events, i);
1991 pollfd = LTTNG_POLL_GETFD(&events, i);
1992
1993 /* Thread quit pipe has been closed. Killing thread. */
1994 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1995 if (ret) {
1996 err = 0;
1997 goto exit;
1998 }
1999
2000 /* Event on the registration socket */
2001 if (pollfd == apps_sock) {
2002 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2003 ERR("Register apps socket poll error");
2004 goto error;
2005 } else if (revents & LPOLLIN) {
2006 sock = lttcomm_accept_unix_sock(apps_sock);
2007 if (sock < 0) {
2008 goto error;
2009 }
2010
2011 /*
2012 * Set socket timeout for both receiving and ending.
2013 * app_socket_timeout is in seconds, whereas
2014 * lttcomm_setsockopt_rcv_timeout and
2015 * lttcomm_setsockopt_snd_timeout expect msec as
2016 * parameter.
2017 */
2018 (void) lttcomm_setsockopt_rcv_timeout(sock,
2019 app_socket_timeout * 1000);
2020 (void) lttcomm_setsockopt_snd_timeout(sock,
2021 app_socket_timeout * 1000);
2022
2023 /*
2024 * Set the CLOEXEC flag. Return code is useless because
2025 * either way, the show must go on.
2026 */
2027 (void) utils_set_fd_cloexec(sock);
2028
2029 /* Create UST registration command for enqueuing */
2030 ust_cmd = zmalloc(sizeof(struct ust_command));
2031 if (ust_cmd == NULL) {
2032 PERROR("ust command zmalloc");
2033 goto error;
2034 }
2035
2036 /*
2037 * Using message-based transmissions to ensure we don't
2038 * have to deal with partially received messages.
2039 */
2040 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2041 if (ret < 0) {
2042 ERR("Exhausted file descriptors allowed for applications.");
2043 free(ust_cmd);
2044 ret = close(sock);
2045 if (ret) {
2046 PERROR("close");
2047 }
2048 sock = -1;
2049 continue;
2050 }
2051
2052 health_code_update();
2053 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2054 if (ret < 0) {
2055 free(ust_cmd);
2056 /* Close socket of the application. */
2057 ret = close(sock);
2058 if (ret) {
2059 PERROR("close");
2060 }
2061 lttng_fd_put(LTTNG_FD_APPS, 1);
2062 sock = -1;
2063 continue;
2064 }
2065 health_code_update();
2066
2067 ust_cmd->sock = sock;
2068 sock = -1;
2069
2070 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2071 " gid:%d sock:%d name:%s (version %d.%d)",
2072 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2073 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2074 ust_cmd->sock, ust_cmd->reg_msg.name,
2075 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2076
2077 /*
2078 * Lock free enqueue the registration request. The red pill
2079 * has been taken! This apps will be part of the *system*.
2080 */
2081 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2082
2083 /*
2084 * Wake the registration queue futex. Implicit memory
2085 * barrier with the exchange in cds_wfcq_enqueue.
2086 */
2087 futex_nto1_wake(&ust_cmd_queue.futex);
2088 }
2089 }
2090 }
2091 }
2092
2093 exit:
2094 error:
2095 /* Notify that the registration thread is gone */
2096 notify_ust_apps(0);
2097
2098 if (apps_sock >= 0) {
2099 ret = close(apps_sock);
2100 if (ret) {
2101 PERROR("close");
2102 }
2103 }
2104 if (sock >= 0) {
2105 ret = close(sock);
2106 if (ret) {
2107 PERROR("close");
2108 }
2109 lttng_fd_put(LTTNG_FD_APPS, 1);
2110 }
2111 unlink(apps_unix_sock_path);
2112
2113 error_poll_add:
2114 lttng_poll_clean(&events);
2115 error_listen:
2116 error_create_poll:
2117 error_testpoint:
2118 DBG("UST Registration thread cleanup complete");
2119 if (err) {
2120 health_error();
2121 ERR("Health error occurred in %s", __func__);
2122 }
2123 health_unregister(health_sessiond);
2124
2125 return NULL;
2126 }
2127
2128 /*
2129 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2130 * exec or it will fails.
2131 */
2132 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2133 {
2134 int ret, clock_ret;
2135 struct timespec timeout;
2136
2137 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2138 consumer_data->consumer_thread_is_ready = 0;
2139
2140 /* Setup pthread condition */
2141 ret = pthread_condattr_init(&consumer_data->condattr);
2142 if (ret != 0) {
2143 errno = ret;
2144 PERROR("pthread_condattr_init consumer data");
2145 goto error;
2146 }
2147
2148 /*
2149 * Set the monotonic clock in order to make sure we DO NOT jump in time
2150 * between the clock_gettime() call and the timedwait call. See bug #324
2151 * for a more details and how we noticed it.
2152 */
2153 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2154 if (ret != 0) {
2155 errno = ret;
2156 PERROR("pthread_condattr_setclock consumer data");
2157 goto error;
2158 }
2159
2160 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2161 if (ret != 0) {
2162 errno = ret;
2163 PERROR("pthread_cond_init consumer data");
2164 goto error;
2165 }
2166
2167 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2168 consumer_data);
2169 if (ret != 0) {
2170 PERROR("pthread_create consumer");
2171 ret = -1;
2172 goto error;
2173 }
2174
2175 /* We are about to wait on a pthread condition */
2176 pthread_mutex_lock(&consumer_data->cond_mutex);
2177
2178 /* Get time for sem_timedwait absolute timeout */
2179 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2180 /*
2181 * Set the timeout for the condition timed wait even if the clock gettime
2182 * call fails since we might loop on that call and we want to avoid to
2183 * increment the timeout too many times.
2184 */
2185 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2186
2187 /*
2188 * The following loop COULD be skipped in some conditions so this is why we
2189 * set ret to 0 in order to make sure at least one round of the loop is
2190 * done.
2191 */
2192 ret = 0;
2193
2194 /*
2195 * Loop until the condition is reached or when a timeout is reached. Note
2196 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2197 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2198 * possible. This loop does not take any chances and works with both of
2199 * them.
2200 */
2201 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2202 if (clock_ret < 0) {
2203 PERROR("clock_gettime spawn consumer");
2204 /* Infinite wait for the consumerd thread to be ready */
2205 ret = pthread_cond_wait(&consumer_data->cond,
2206 &consumer_data->cond_mutex);
2207 } else {
2208 ret = pthread_cond_timedwait(&consumer_data->cond,
2209 &consumer_data->cond_mutex, &timeout);
2210 }
2211 }
2212
2213 /* Release the pthread condition */
2214 pthread_mutex_unlock(&consumer_data->cond_mutex);
2215
2216 if (ret != 0) {
2217 errno = ret;
2218 if (ret == ETIMEDOUT) {
2219 int pth_ret;
2220
2221 /*
2222 * Call has timed out so we kill the kconsumerd_thread and return
2223 * an error.
2224 */
2225 ERR("Condition timed out. The consumer thread was never ready."
2226 " Killing it");
2227 pth_ret = pthread_cancel(consumer_data->thread);
2228 if (pth_ret < 0) {
2229 PERROR("pthread_cancel consumer thread");
2230 }
2231 } else {
2232 PERROR("pthread_cond_wait failed consumer thread");
2233 }
2234 /* Caller is expecting a negative value on failure. */
2235 ret = -1;
2236 goto error;
2237 }
2238
2239 pthread_mutex_lock(&consumer_data->pid_mutex);
2240 if (consumer_data->pid == 0) {
2241 ERR("Consumerd did not start");
2242 pthread_mutex_unlock(&consumer_data->pid_mutex);
2243 goto error;
2244 }
2245 pthread_mutex_unlock(&consumer_data->pid_mutex);
2246
2247 return 0;
2248
2249 error:
2250 return ret;
2251 }
2252
2253 /*
2254 * Join consumer thread
2255 */
2256 static int join_consumer_thread(struct consumer_data *consumer_data)
2257 {
2258 void *status;
2259
2260 /* Consumer pid must be a real one. */
2261 if (consumer_data->pid > 0) {
2262 int ret;
2263 ret = kill(consumer_data->pid, SIGTERM);
2264 if (ret) {
2265 ERR("Error killing consumer daemon");
2266 return ret;
2267 }
2268 return pthread_join(consumer_data->thread, &status);
2269 } else {
2270 return 0;
2271 }
2272 }
2273
2274 /*
2275 * Fork and exec a consumer daemon (consumerd).
2276 *
2277 * Return pid if successful else -1.
2278 */
2279 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2280 {
2281 int ret;
2282 pid_t pid;
2283 const char *consumer_to_use;
2284 const char *verbosity;
2285 struct stat st;
2286
2287 DBG("Spawning consumerd");
2288
2289 pid = fork();
2290 if (pid == 0) {
2291 /*
2292 * Exec consumerd.
2293 */
2294 if (opt_verbose_consumer) {
2295 verbosity = "--verbose";
2296 } else if (lttng_opt_quiet) {
2297 verbosity = "--quiet";
2298 } else {
2299 verbosity = "";
2300 }
2301
2302 switch (consumer_data->type) {
2303 case LTTNG_CONSUMER_KERNEL:
2304 /*
2305 * Find out which consumerd to execute. We will first try the
2306 * 64-bit path, then the sessiond's installation directory, and
2307 * fallback on the 32-bit one,
2308 */
2309 DBG3("Looking for a kernel consumer at these locations:");
2310 DBG3(" 1) %s", consumerd64_bin);
2311 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2312 DBG3(" 3) %s", consumerd32_bin);
2313 if (stat(consumerd64_bin, &st) == 0) {
2314 DBG3("Found location #1");
2315 consumer_to_use = consumerd64_bin;
2316 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2317 DBG3("Found location #2");
2318 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2319 } else if (stat(consumerd32_bin, &st) == 0) {
2320 DBG3("Found location #3");
2321 consumer_to_use = consumerd32_bin;
2322 } else {
2323 DBG("Could not find any valid consumerd executable");
2324 ret = -EINVAL;
2325 break;
2326 }
2327 DBG("Using kernel consumer at: %s", consumer_to_use);
2328 ret = execl(consumer_to_use,
2329 "lttng-consumerd", verbosity, "-k",
2330 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2331 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2332 "--group", tracing_group_name,
2333 NULL);
2334 break;
2335 case LTTNG_CONSUMER64_UST:
2336 {
2337 char *tmpnew = NULL;
2338
2339 if (consumerd64_libdir[0] != '\0') {
2340 char *tmp;
2341 size_t tmplen;
2342
2343 tmp = getenv("LD_LIBRARY_PATH");
2344 if (!tmp) {
2345 tmp = "";
2346 }
2347 tmplen = strlen("LD_LIBRARY_PATH=")
2348 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2349 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2350 if (!tmpnew) {
2351 ret = -ENOMEM;
2352 goto error;
2353 }
2354 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2355 strcat(tmpnew, consumerd64_libdir);
2356 if (tmp[0] != '\0') {
2357 strcat(tmpnew, ":");
2358 strcat(tmpnew, tmp);
2359 }
2360 ret = putenv(tmpnew);
2361 if (ret) {
2362 ret = -errno;
2363 free(tmpnew);
2364 goto error;
2365 }
2366 }
2367 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2368 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2369 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2370 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2371 "--group", tracing_group_name,
2372 NULL);
2373 if (consumerd64_libdir[0] != '\0') {
2374 free(tmpnew);
2375 }
2376 break;
2377 }
2378 case LTTNG_CONSUMER32_UST:
2379 {
2380 char *tmpnew = NULL;
2381
2382 if (consumerd32_libdir[0] != '\0') {
2383 char *tmp;
2384 size_t tmplen;
2385
2386 tmp = getenv("LD_LIBRARY_PATH");
2387 if (!tmp) {
2388 tmp = "";
2389 }
2390 tmplen = strlen("LD_LIBRARY_PATH=")
2391 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2392 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2393 if (!tmpnew) {
2394 ret = -ENOMEM;
2395 goto error;
2396 }
2397 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2398 strcat(tmpnew, consumerd32_libdir);
2399 if (tmp[0] != '\0') {
2400 strcat(tmpnew, ":");
2401 strcat(tmpnew, tmp);
2402 }
2403 ret = putenv(tmpnew);
2404 if (ret) {
2405 ret = -errno;
2406 free(tmpnew);
2407 goto error;
2408 }
2409 }
2410 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2411 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2412 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2413 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2414 "--group", tracing_group_name,
2415 NULL);
2416 if (consumerd32_libdir[0] != '\0') {
2417 free(tmpnew);
2418 }
2419 break;
2420 }
2421 default:
2422 PERROR("unknown consumer type");
2423 exit(EXIT_FAILURE);
2424 }
2425 if (errno != 0) {
2426 PERROR("Consumer execl()");
2427 }
2428 /* Reaching this point, we got a failure on our execl(). */
2429 exit(EXIT_FAILURE);
2430 } else if (pid > 0) {
2431 ret = pid;
2432 } else {
2433 PERROR("start consumer fork");
2434 ret = -errno;
2435 }
2436 error:
2437 return ret;
2438 }
2439
2440 /*
2441 * Spawn the consumerd daemon and session daemon thread.
2442 */
2443 static int start_consumerd(struct consumer_data *consumer_data)
2444 {
2445 int ret;
2446
2447 /*
2448 * Set the listen() state on the socket since there is a possible race
2449 * between the exec() of the consumer daemon and this call if place in the
2450 * consumer thread. See bug #366 for more details.
2451 */
2452 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2453 if (ret < 0) {
2454 goto error;
2455 }
2456
2457 pthread_mutex_lock(&consumer_data->pid_mutex);
2458 if (consumer_data->pid != 0) {
2459 pthread_mutex_unlock(&consumer_data->pid_mutex);
2460 goto end;
2461 }
2462
2463 ret = spawn_consumerd(consumer_data);
2464 if (ret < 0) {
2465 ERR("Spawning consumerd failed");
2466 pthread_mutex_unlock(&consumer_data->pid_mutex);
2467 goto error;
2468 }
2469
2470 /* Setting up the consumer_data pid */
2471 consumer_data->pid = ret;
2472 DBG2("Consumer pid %d", consumer_data->pid);
2473 pthread_mutex_unlock(&consumer_data->pid_mutex);
2474
2475 DBG2("Spawning consumer control thread");
2476 ret = spawn_consumer_thread(consumer_data);
2477 if (ret < 0) {
2478 ERR("Fatal error spawning consumer control thread");
2479 goto error;
2480 }
2481
2482 end:
2483 return 0;
2484
2485 error:
2486 /* Cleanup already created sockets on error. */
2487 if (consumer_data->err_sock >= 0) {
2488 int err;
2489
2490 err = close(consumer_data->err_sock);
2491 if (err < 0) {
2492 PERROR("close consumer data error socket");
2493 }
2494 }
2495 return ret;
2496 }
2497
2498 /*
2499 * Setup necessary data for kernel tracer action.
2500 */
2501 static int init_kernel_tracer(void)
2502 {
2503 int ret;
2504
2505 /* Modprobe lttng kernel modules */
2506 ret = modprobe_lttng_control();
2507 if (ret < 0) {
2508 goto error;
2509 }
2510
2511 /* Open debugfs lttng */
2512 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2513 if (kernel_tracer_fd < 0) {
2514 DBG("Failed to open %s", module_proc_lttng);
2515 ret = -1;
2516 goto error_open;
2517 }
2518
2519 /* Validate kernel version */
2520 ret = kernel_validate_version(kernel_tracer_fd);
2521 if (ret < 0) {
2522 goto error_version;
2523 }
2524
2525 ret = modprobe_lttng_data();
2526 if (ret < 0) {
2527 goto error_modules;
2528 }
2529
2530 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2531 return 0;
2532
2533 error_version:
2534 modprobe_remove_lttng_control();
2535 ret = close(kernel_tracer_fd);
2536 if (ret) {
2537 PERROR("close");
2538 }
2539 kernel_tracer_fd = -1;
2540 return LTTNG_ERR_KERN_VERSION;
2541
2542 error_modules:
2543 ret = close(kernel_tracer_fd);
2544 if (ret) {
2545 PERROR("close");
2546 }
2547
2548 error_open:
2549 modprobe_remove_lttng_control();
2550
2551 error:
2552 WARN("No kernel tracer available");
2553 kernel_tracer_fd = -1;
2554 if (!is_root) {
2555 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2556 } else {
2557 return LTTNG_ERR_KERN_NA;
2558 }
2559 }
2560
2561
2562 /*
2563 * Copy consumer output from the tracing session to the domain session. The
2564 * function also applies the right modification on a per domain basis for the
2565 * trace files destination directory.
2566 *
2567 * Should *NOT* be called with RCU read-side lock held.
2568 */
2569 static int copy_session_consumer(int domain, struct ltt_session *session)
2570 {
2571 int ret;
2572 const char *dir_name;
2573 struct consumer_output *consumer;
2574
2575 assert(session);
2576 assert(session->consumer);
2577
2578 switch (domain) {
2579 case LTTNG_DOMAIN_KERNEL:
2580 DBG3("Copying tracing session consumer output in kernel session");
2581 /*
2582 * XXX: We should audit the session creation and what this function
2583 * does "extra" in order to avoid a destroy since this function is used
2584 * in the domain session creation (kernel and ust) only. Same for UST
2585 * domain.
2586 */
2587 if (session->kernel_session->consumer) {
2588 consumer_destroy_output(session->kernel_session->consumer);
2589 }
2590 session->kernel_session->consumer =
2591 consumer_copy_output(session->consumer);
2592 /* Ease our life a bit for the next part */
2593 consumer = session->kernel_session->consumer;
2594 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2595 break;
2596 case LTTNG_DOMAIN_JUL:
2597 case LTTNG_DOMAIN_LOG4J:
2598 case LTTNG_DOMAIN_UST:
2599 DBG3("Copying tracing session consumer output in UST session");
2600 if (session->ust_session->consumer) {
2601 consumer_destroy_output(session->ust_session->consumer);
2602 }
2603 session->ust_session->consumer =
2604 consumer_copy_output(session->consumer);
2605 /* Ease our life a bit for the next part */
2606 consumer = session->ust_session->consumer;
2607 dir_name = DEFAULT_UST_TRACE_DIR;
2608 break;
2609 default:
2610 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2611 goto error;
2612 }
2613
2614 /* Append correct directory to subdir */
2615 strncat(consumer->subdir, dir_name,
2616 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2617 DBG3("Copy session consumer subdir %s", consumer->subdir);
2618
2619 ret = LTTNG_OK;
2620
2621 error:
2622 return ret;
2623 }
2624
2625 /*
2626 * Create an UST session and add it to the session ust list.
2627 *
2628 * Should *NOT* be called with RCU read-side lock held.
2629 */
2630 static int create_ust_session(struct ltt_session *session,
2631 struct lttng_domain *domain)
2632 {
2633 int ret;
2634 struct ltt_ust_session *lus = NULL;
2635
2636 assert(session);
2637 assert(domain);
2638 assert(session->consumer);
2639
2640 switch (domain->type) {
2641 case LTTNG_DOMAIN_JUL:
2642 case LTTNG_DOMAIN_LOG4J:
2643 case LTTNG_DOMAIN_UST:
2644 break;
2645 default:
2646 ERR("Unknown UST domain on create session %d", domain->type);
2647 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2648 goto error;
2649 }
2650
2651 DBG("Creating UST session");
2652
2653 lus = trace_ust_create_session(session->id);
2654 if (lus == NULL) {
2655 ret = LTTNG_ERR_UST_SESS_FAIL;
2656 goto error;
2657 }
2658
2659 lus->uid = session->uid;
2660 lus->gid = session->gid;
2661 lus->output_traces = session->output_traces;
2662 lus->snapshot_mode = session->snapshot_mode;
2663 lus->live_timer_interval = session->live_timer;
2664 session->ust_session = lus;
2665
2666 /* Copy session output to the newly created UST session */
2667 ret = copy_session_consumer(domain->type, session);
2668 if (ret != LTTNG_OK) {
2669 goto error;
2670 }
2671
2672 return LTTNG_OK;
2673
2674 error:
2675 free(lus);
2676 session->ust_session = NULL;
2677 return ret;
2678 }
2679
2680 /*
2681 * Create a kernel tracer session then create the default channel.
2682 */
2683 static int create_kernel_session(struct ltt_session *session)
2684 {
2685 int ret;
2686
2687 DBG("Creating kernel session");
2688
2689 ret = kernel_create_session(session, kernel_tracer_fd);
2690 if (ret < 0) {
2691 ret = LTTNG_ERR_KERN_SESS_FAIL;
2692 goto error;
2693 }
2694
2695 /* Code flow safety */
2696 assert(session->kernel_session);
2697
2698 /* Copy session output to the newly created Kernel session */
2699 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2700 if (ret != LTTNG_OK) {
2701 goto error;
2702 }
2703
2704 /* Create directory(ies) on local filesystem. */
2705 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2706 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2707 ret = run_as_mkdir_recursive(
2708 session->kernel_session->consumer->dst.trace_path,
2709 S_IRWXU | S_IRWXG, session->uid, session->gid);
2710 if (ret < 0) {
2711 if (ret != -EEXIST) {
2712 ERR("Trace directory creation error");
2713 goto error;
2714 }
2715 }
2716 }
2717
2718 session->kernel_session->uid = session->uid;
2719 session->kernel_session->gid = session->gid;
2720 session->kernel_session->output_traces = session->output_traces;
2721 session->kernel_session->snapshot_mode = session->snapshot_mode;
2722
2723 return LTTNG_OK;
2724
2725 error:
2726 trace_kernel_destroy_session(session->kernel_session);
2727 session->kernel_session = NULL;
2728 return ret;
2729 }
2730
2731 /*
2732 * Count number of session permitted by uid/gid.
2733 */
2734 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2735 {
2736 unsigned int i = 0;
2737 struct ltt_session *session;
2738
2739 DBG("Counting number of available session for UID %d GID %d",
2740 uid, gid);
2741 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2742 /*
2743 * Only list the sessions the user can control.
2744 */
2745 if (!session_access_ok(session, uid, gid)) {
2746 continue;
2747 }
2748 i++;
2749 }
2750 return i;
2751 }
2752
2753 /*
2754 * Process the command requested by the lttng client within the command
2755 * context structure. This function make sure that the return structure (llm)
2756 * is set and ready for transmission before returning.
2757 *
2758 * Return any error encountered or 0 for success.
2759 *
2760 * "sock" is only used for special-case var. len data.
2761 *
2762 * Should *NOT* be called with RCU read-side lock held.
2763 */
2764 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2765 int *sock_error)
2766 {
2767 int ret = LTTNG_OK;
2768 int need_tracing_session = 1;
2769 int need_domain;
2770
2771 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2772
2773 *sock_error = 0;
2774
2775 switch (cmd_ctx->lsm->cmd_type) {
2776 case LTTNG_CREATE_SESSION:
2777 case LTTNG_CREATE_SESSION_SNAPSHOT:
2778 case LTTNG_CREATE_SESSION_LIVE:
2779 case LTTNG_DESTROY_SESSION:
2780 case LTTNG_LIST_SESSIONS:
2781 case LTTNG_LIST_DOMAINS:
2782 case LTTNG_START_TRACE:
2783 case LTTNG_STOP_TRACE:
2784 case LTTNG_DATA_PENDING:
2785 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2786 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2787 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2788 case LTTNG_SNAPSHOT_RECORD:
2789 case LTTNG_SAVE_SESSION:
2790 need_domain = 0;
2791 break;
2792 default:
2793 need_domain = 1;
2794 }
2795
2796 if (opt_no_kernel && need_domain
2797 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2798 if (!is_root) {
2799 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2800 } else {
2801 ret = LTTNG_ERR_KERN_NA;
2802 }
2803 goto error;
2804 }
2805
2806 /* Deny register consumer if we already have a spawned consumer. */
2807 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2808 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2809 if (kconsumer_data.pid > 0) {
2810 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2811 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2812 goto error;
2813 }
2814 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2815 }
2816
2817 /*
2818 * Check for command that don't needs to allocate a returned payload. We do
2819 * this here so we don't have to make the call for no payload at each
2820 * command.
2821 */
2822 switch(cmd_ctx->lsm->cmd_type) {
2823 case LTTNG_LIST_SESSIONS:
2824 case LTTNG_LIST_TRACEPOINTS:
2825 case LTTNG_LIST_TRACEPOINT_FIELDS:
2826 case LTTNG_LIST_DOMAINS:
2827 case LTTNG_LIST_CHANNELS:
2828 case LTTNG_LIST_EVENTS:
2829 break;
2830 default:
2831 /* Setup lttng message with no payload */
2832 ret = setup_lttng_msg(cmd_ctx, 0);
2833 if (ret < 0) {
2834 /* This label does not try to unlock the session */
2835 goto init_setup_error;
2836 }
2837 }
2838
2839 /* Commands that DO NOT need a session. */
2840 switch (cmd_ctx->lsm->cmd_type) {
2841 case LTTNG_CREATE_SESSION:
2842 case LTTNG_CREATE_SESSION_SNAPSHOT:
2843 case LTTNG_CREATE_SESSION_LIVE:
2844 case LTTNG_CALIBRATE:
2845 case LTTNG_LIST_SESSIONS:
2846 case LTTNG_LIST_TRACEPOINTS:
2847 case LTTNG_LIST_TRACEPOINT_FIELDS:
2848 case LTTNG_SAVE_SESSION:
2849 need_tracing_session = 0;
2850 break;
2851 default:
2852 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2853 /*
2854 * We keep the session list lock across _all_ commands
2855 * for now, because the per-session lock does not
2856 * handle teardown properly.
2857 */
2858 session_lock_list();
2859 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2860 if (cmd_ctx->session == NULL) {
2861 ret = LTTNG_ERR_SESS_NOT_FOUND;
2862 goto error;
2863 } else {
2864 /* Acquire lock for the session */
2865 session_lock(cmd_ctx->session);
2866 }
2867 break;
2868 }
2869
2870 /*
2871 * Commands that need a valid session but should NOT create one if none
2872 * exists. Instead of creating one and destroying it when the command is
2873 * handled, process that right before so we save some round trip in useless
2874 * code path.
2875 */
2876 switch (cmd_ctx->lsm->cmd_type) {
2877 case LTTNG_DISABLE_CHANNEL:
2878 case LTTNG_DISABLE_EVENT:
2879 switch (cmd_ctx->lsm->domain.type) {
2880 case LTTNG_DOMAIN_KERNEL:
2881 if (!cmd_ctx->session->kernel_session) {
2882 ret = LTTNG_ERR_NO_CHANNEL;
2883 goto error;
2884 }
2885 break;
2886 case LTTNG_DOMAIN_JUL:
2887 case LTTNG_DOMAIN_LOG4J:
2888 case LTTNG_DOMAIN_UST:
2889 if (!cmd_ctx->session->ust_session) {
2890 ret = LTTNG_ERR_NO_CHANNEL;
2891 goto error;
2892 }
2893 break;
2894 default:
2895 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2896 goto error;
2897 }
2898 default:
2899 break;
2900 }
2901
2902 if (!need_domain) {
2903 goto skip_domain;
2904 }
2905
2906 /*
2907 * Check domain type for specific "pre-action".
2908 */
2909 switch (cmd_ctx->lsm->domain.type) {
2910 case LTTNG_DOMAIN_KERNEL:
2911 if (!is_root) {
2912 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2913 goto error;
2914 }
2915
2916 /* Kernel tracer check */
2917 if (kernel_tracer_fd == -1) {
2918 /* Basically, load kernel tracer modules */
2919 ret = init_kernel_tracer();
2920 if (ret != 0) {
2921 goto error;
2922 }
2923 }
2924
2925 /* Consumer is in an ERROR state. Report back to client */
2926 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2927 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2928 goto error;
2929 }
2930
2931 /* Need a session for kernel command */
2932 if (need_tracing_session) {
2933 if (cmd_ctx->session->kernel_session == NULL) {
2934 ret = create_kernel_session(cmd_ctx->session);
2935 if (ret < 0) {
2936 ret = LTTNG_ERR_KERN_SESS_FAIL;
2937 goto error;
2938 }
2939 }
2940
2941 /* Start the kernel consumer daemon */
2942 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2943 if (kconsumer_data.pid == 0 &&
2944 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2945 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2946 ret = start_consumerd(&kconsumer_data);
2947 if (ret < 0) {
2948 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2949 goto error;
2950 }
2951 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2952 } else {
2953 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2954 }
2955
2956 /*
2957 * The consumer was just spawned so we need to add the socket to
2958 * the consumer output of the session if exist.
2959 */
2960 ret = consumer_create_socket(&kconsumer_data,
2961 cmd_ctx->session->kernel_session->consumer);
2962 if (ret < 0) {
2963 goto error;
2964 }
2965 }
2966
2967 break;
2968 case LTTNG_DOMAIN_JUL:
2969 case LTTNG_DOMAIN_LOG4J:
2970 case LTTNG_DOMAIN_UST:
2971 {
2972 if (!ust_app_supported()) {
2973 ret = LTTNG_ERR_NO_UST;
2974 goto error;
2975 }
2976 /* Consumer is in an ERROR state. Report back to client */
2977 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2978 ret = LTTNG_ERR_NO_USTCONSUMERD;
2979 goto error;
2980 }
2981
2982 if (need_tracing_session) {
2983 /* Create UST session if none exist. */
2984 if (cmd_ctx->session->ust_session == NULL) {
2985 ret = create_ust_session(cmd_ctx->session,
2986 &cmd_ctx->lsm->domain);
2987 if (ret != LTTNG_OK) {
2988 goto error;
2989 }
2990 }
2991
2992 /* Start the UST consumer daemons */
2993 /* 64-bit */
2994 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2995 if (consumerd64_bin[0] != '\0' &&
2996 ustconsumer64_data.pid == 0 &&
2997 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2998 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2999 ret = start_consumerd(&ustconsumer64_data);
3000 if (ret < 0) {
3001 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3002 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3003 goto error;
3004 }
3005
3006 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3007 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3008 } else {
3009 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3010 }
3011
3012 /*
3013 * Setup socket for consumer 64 bit. No need for atomic access
3014 * since it was set above and can ONLY be set in this thread.
3015 */
3016 ret = consumer_create_socket(&ustconsumer64_data,
3017 cmd_ctx->session->ust_session->consumer);
3018 if (ret < 0) {
3019 goto error;
3020 }
3021
3022 /* 32-bit */
3023 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3024 if (consumerd32_bin[0] != '\0' &&
3025 ustconsumer32_data.pid == 0 &&
3026 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3027 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3028 ret = start_consumerd(&ustconsumer32_data);
3029 if (ret < 0) {
3030 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3031 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3032 goto error;
3033 }
3034
3035 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3036 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3037 } else {
3038 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3039 }
3040
3041 /*
3042 * Setup socket for consumer 64 bit. No need for atomic access
3043 * since it was set above and can ONLY be set in this thread.
3044 */
3045 ret = consumer_create_socket(&ustconsumer32_data,
3046 cmd_ctx->session->ust_session->consumer);
3047 if (ret < 0) {
3048 goto error;
3049 }
3050 }
3051 break;
3052 }
3053 default:
3054 break;
3055 }
3056 skip_domain:
3057
3058 /* Validate consumer daemon state when start/stop trace command */
3059 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3060 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3061 switch (cmd_ctx->lsm->domain.type) {
3062 case LTTNG_DOMAIN_JUL:
3063 case LTTNG_DOMAIN_LOG4J:
3064 case LTTNG_DOMAIN_UST:
3065 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3066 ret = LTTNG_ERR_NO_USTCONSUMERD;
3067 goto error;
3068 }
3069 break;
3070 case LTTNG_DOMAIN_KERNEL:
3071 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3072 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3073 goto error;
3074 }
3075 break;
3076 }
3077 }
3078
3079 /*
3080 * Check that the UID or GID match that of the tracing session.
3081 * The root user can interact with all sessions.
3082 */
3083 if (need_tracing_session) {
3084 if (!session_access_ok(cmd_ctx->session,
3085 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3086 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3087 ret = LTTNG_ERR_EPERM;
3088 goto error;
3089 }
3090 }
3091
3092 /*
3093 * Send relayd information to consumer as soon as we have a domain and a
3094 * session defined.
3095 */
3096 if (cmd_ctx->session && need_domain) {
3097 /*
3098 * Setup relayd if not done yet. If the relayd information was already
3099 * sent to the consumer, this call will gracefully return.
3100 */
3101 ret = cmd_setup_relayd(cmd_ctx->session);
3102 if (ret != LTTNG_OK) {
3103 goto error;
3104 }
3105 }
3106
3107 /* Process by command type */
3108 switch (cmd_ctx->lsm->cmd_type) {
3109 case LTTNG_ADD_CONTEXT:
3110 {
3111 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3112 cmd_ctx->lsm->u.context.channel_name,
3113 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3114 break;
3115 }
3116 case LTTNG_DISABLE_CHANNEL:
3117 {
3118 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3119 cmd_ctx->lsm->u.disable.channel_name);
3120 break;
3121 }
3122 case LTTNG_DISABLE_EVENT:
3123 {
3124 /* FIXME: passing packed structure to non-packed pointer */
3125 /* TODO: handle filter */
3126 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3127 cmd_ctx->lsm->u.disable.channel_name,
3128 &cmd_ctx->lsm->u.disable.event);
3129 break;
3130 }
3131 case LTTNG_ENABLE_CHANNEL:
3132 {
3133 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3134 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3135 break;
3136 }
3137 case LTTNG_ENABLE_EVENT:
3138 {
3139 struct lttng_event_exclusion *exclusion = NULL;
3140 struct lttng_filter_bytecode *bytecode = NULL;
3141 char *filter_expression = NULL;
3142
3143 /* Handle exclusion events and receive it from the client. */
3144 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3145 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3146
3147 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3148 (count * LTTNG_SYMBOL_NAME_LEN));
3149 if (!exclusion) {
3150 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3151 goto error;
3152 }
3153
3154 DBG("Receiving var len exclusion event list from client ...");
3155 exclusion->count = count;
3156 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3157 count * LTTNG_SYMBOL_NAME_LEN);
3158 if (ret <= 0) {
3159 DBG("Nothing recv() from client var len data... continuing");
3160 *sock_error = 1;
3161 free(exclusion);
3162 ret = LTTNG_ERR_EXCLUSION_INVAL;
3163 goto error;
3164 }
3165 }
3166
3167 /* Get filter expression from client. */
3168 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3169 size_t expression_len =
3170 cmd_ctx->lsm->u.enable.expression_len;
3171
3172 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3173 ret = LTTNG_ERR_FILTER_INVAL;
3174 free(exclusion);
3175 goto error;
3176 }
3177
3178 filter_expression = zmalloc(expression_len);
3179 if (!filter_expression) {
3180 free(exclusion);
3181 ret = LTTNG_ERR_FILTER_NOMEM;
3182 goto error;
3183 }
3184
3185 /* Receive var. len. data */
3186 DBG("Receiving var len filter's expression from client ...");
3187 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3188 expression_len);
3189 if (ret <= 0) {
3190 DBG("Nothing recv() from client car len data... continuing");
3191 *sock_error = 1;
3192 free(filter_expression);
3193 free(exclusion);
3194 ret = LTTNG_ERR_FILTER_INVAL;
3195 goto error;
3196 }
3197 }
3198
3199 /* Handle filter and get bytecode from client. */
3200 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3201 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3202
3203 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3204 ret = LTTNG_ERR_FILTER_INVAL;
3205 free(exclusion);
3206 goto error;
3207 }
3208
3209 bytecode = zmalloc(bytecode_len);
3210 if (!bytecode) {
3211 free(exclusion);
3212 ret = LTTNG_ERR_FILTER_NOMEM;
3213 goto error;
3214 }
3215
3216 /* Receive var. len. data */
3217 DBG("Receiving var len filter's bytecode from client ...");
3218 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3219 if (ret <= 0) {
3220 DBG("Nothing recv() from client car len data... continuing");
3221 *sock_error = 1;
3222 free(bytecode);
3223 free(exclusion);
3224 ret = LTTNG_ERR_FILTER_INVAL;
3225 goto error;
3226 }
3227
3228 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3229 free(bytecode);
3230 free(exclusion);
3231 ret = LTTNG_ERR_FILTER_INVAL;
3232 goto error;
3233 }
3234 }
3235
3236 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3237 cmd_ctx->lsm->u.enable.channel_name,
3238 &cmd_ctx->lsm->u.enable.event,
3239 filter_expression, bytecode, exclusion,
3240 kernel_poll_pipe[1]);
3241 break;
3242 }
3243 case LTTNG_LIST_TRACEPOINTS:
3244 {
3245 struct lttng_event *events;
3246 ssize_t nb_events;
3247
3248 session_lock_list();
3249 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3250 session_unlock_list();
3251 if (nb_events < 0) {
3252 /* Return value is a negative lttng_error_code. */
3253 ret = -nb_events;
3254 goto error;
3255 }
3256
3257 /*
3258 * Setup lttng message with payload size set to the event list size in
3259 * bytes and then copy list into the llm payload.
3260 */
3261 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3262 if (ret < 0) {
3263 free(events);
3264 goto setup_error;
3265 }
3266
3267 /* Copy event list into message payload */
3268 memcpy(cmd_ctx->llm->payload, events,
3269 sizeof(struct lttng_event) * nb_events);
3270
3271 free(events);
3272
3273 ret = LTTNG_OK;
3274 break;
3275 }
3276 case LTTNG_LIST_TRACEPOINT_FIELDS:
3277 {
3278 struct lttng_event_field *fields;
3279 ssize_t nb_fields;
3280
3281 session_lock_list();
3282 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3283 &fields);
3284 session_unlock_list();
3285 if (nb_fields < 0) {
3286 /* Return value is a negative lttng_error_code. */
3287 ret = -nb_fields;
3288 goto error;
3289 }
3290
3291 /*
3292 * Setup lttng message with payload size set to the event list size in
3293 * bytes and then copy list into the llm payload.
3294 */
3295 ret = setup_lttng_msg(cmd_ctx,
3296 sizeof(struct lttng_event_field) * nb_fields);
3297 if (ret < 0) {
3298 free(fields);
3299 goto setup_error;
3300 }
3301
3302 /* Copy event list into message payload */
3303 memcpy(cmd_ctx->llm->payload, fields,
3304 sizeof(struct lttng_event_field) * nb_fields);
3305
3306 free(fields);
3307
3308 ret = LTTNG_OK;
3309 break;
3310 }
3311 case LTTNG_SET_CONSUMER_URI:
3312 {
3313 size_t nb_uri, len;
3314 struct lttng_uri *uris;
3315
3316 nb_uri = cmd_ctx->lsm->u.uri.size;
3317 len = nb_uri * sizeof(struct lttng_uri);
3318
3319 if (nb_uri == 0) {
3320 ret = LTTNG_ERR_INVALID;
3321 goto error;
3322 }
3323
3324 uris = zmalloc(len);
3325 if (uris == NULL) {
3326 ret = LTTNG_ERR_FATAL;
3327 goto error;
3328 }
3329
3330 /* Receive variable len data */
3331 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3332 ret = lttcomm_recv_unix_sock(sock, uris, len);
3333 if (ret <= 0) {
3334 DBG("No URIs received from client... continuing");
3335 *sock_error = 1;
3336 ret = LTTNG_ERR_SESSION_FAIL;
3337 free(uris);
3338 goto error;
3339 }
3340
3341 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3342 nb_uri, uris);
3343 if (ret != LTTNG_OK) {
3344 free(uris);
3345 goto error;
3346 }
3347
3348 /*
3349 * XXX: 0 means that this URI should be applied on the session. Should
3350 * be a DOMAIN enuam.
3351 */
3352 if (cmd_ctx->lsm->domain.type == 0) {
3353 /* Add the URI for the UST session if a consumer is present. */
3354 if (cmd_ctx->session->ust_session &&
3355 cmd_ctx->session->ust_session->consumer) {
3356 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3357 nb_uri, uris);
3358 } else if (cmd_ctx->session->kernel_session &&
3359 cmd_ctx->session->kernel_session->consumer) {
3360 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3361 cmd_ctx->session, nb_uri, uris);
3362 }
3363 }
3364
3365 free(uris);
3366
3367 break;
3368 }
3369 case LTTNG_START_TRACE:
3370 {
3371 ret = cmd_start_trace(cmd_ctx->session);
3372 break;
3373 }
3374 case LTTNG_STOP_TRACE:
3375 {
3376 ret = cmd_stop_trace(cmd_ctx->session);
3377 break;
3378 }
3379 case LTTNG_CREATE_SESSION:
3380 {
3381 size_t nb_uri, len;
3382 struct lttng_uri *uris = NULL;
3383
3384 nb_uri = cmd_ctx->lsm->u.uri.size;
3385 len = nb_uri * sizeof(struct lttng_uri);
3386
3387 if (nb_uri > 0) {
3388 uris = zmalloc(len);
3389 if (uris == NULL) {
3390 ret = LTTNG_ERR_FATAL;
3391 goto error;
3392 }
3393
3394 /* Receive variable len data */
3395 DBG("Waiting for %zu URIs from client ...", nb_uri);
3396 ret = lttcomm_recv_unix_sock(sock, uris, len);
3397 if (ret <= 0) {
3398 DBG("No URIs received from client... continuing");
3399 *sock_error = 1;
3400 ret = LTTNG_ERR_SESSION_FAIL;
3401 free(uris);
3402 goto error;
3403 }
3404
3405 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3406 DBG("Creating session with ONE network URI is a bad call");
3407 ret = LTTNG_ERR_SESSION_FAIL;
3408 free(uris);
3409 goto error;
3410 }
3411 }
3412
3413 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3414 &cmd_ctx->creds, 0);
3415
3416 free(uris);
3417
3418 break;
3419 }
3420 case LTTNG_DESTROY_SESSION:
3421 {
3422 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3423
3424 /* Set session to NULL so we do not unlock it after free. */
3425 cmd_ctx->session = NULL;
3426 break;
3427 }
3428 case LTTNG_LIST_DOMAINS:
3429 {
3430 ssize_t nb_dom;
3431 struct lttng_domain *domains;
3432
3433 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3434 if (nb_dom < 0) {
3435 /* Return value is a negative lttng_error_code. */
3436 ret = -nb_dom;
3437 goto error;
3438 }
3439
3440 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3441 if (ret < 0) {
3442 free(domains);
3443 goto setup_error;
3444 }
3445
3446 /* Copy event list into message payload */
3447 memcpy(cmd_ctx->llm->payload, domains,
3448 nb_dom * sizeof(struct lttng_domain));
3449
3450 free(domains);
3451
3452 ret = LTTNG_OK;
3453 break;
3454 }
3455 case LTTNG_LIST_CHANNELS:
3456 {
3457 int nb_chan;
3458 struct lttng_channel *channels;
3459
3460 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3461 cmd_ctx->session, &channels);
3462 if (nb_chan < 0) {
3463 /* Return value is a negative lttng_error_code. */
3464 ret = -nb_chan;
3465 goto error;
3466 }
3467
3468 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3469 if (ret < 0) {
3470 free(channels);
3471 goto setup_error;
3472 }
3473
3474 /* Copy event list into message payload */
3475 memcpy(cmd_ctx->llm->payload, channels,
3476 nb_chan * sizeof(struct lttng_channel));
3477
3478 free(channels);
3479
3480 ret = LTTNG_OK;
3481 break;
3482 }
3483 case LTTNG_LIST_EVENTS:
3484 {
3485 ssize_t nb_event;
3486 struct lttng_event *events = NULL;
3487
3488 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3489 cmd_ctx->lsm->u.list.channel_name, &events);
3490 if (nb_event < 0) {
3491 /* Return value is a negative lttng_error_code. */
3492 ret = -nb_event;
3493 goto error;
3494 }
3495
3496 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3497 if (ret < 0) {
3498 free(events);
3499 goto setup_error;
3500 }
3501
3502 /* Copy event list into message payload */
3503 memcpy(cmd_ctx->llm->payload, events,
3504 nb_event * sizeof(struct lttng_event));
3505
3506 free(events);
3507
3508 ret = LTTNG_OK;
3509 break;
3510 }
3511 case LTTNG_LIST_SESSIONS:
3512 {
3513 unsigned int nr_sessions;
3514
3515 session_lock_list();
3516 nr_sessions = lttng_sessions_count(
3517 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3518 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3519
3520 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3521 if (ret < 0) {
3522 session_unlock_list();
3523 goto setup_error;
3524 }
3525
3526 /* Filled the session array */
3527 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3528 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3529 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3530
3531 session_unlock_list();
3532
3533 ret = LTTNG_OK;
3534 break;
3535 }
3536 case LTTNG_CALIBRATE:
3537 {
3538 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3539 &cmd_ctx->lsm->u.calibrate);
3540 break;
3541 }
3542 case LTTNG_REGISTER_CONSUMER:
3543 {
3544 struct consumer_data *cdata;
3545
3546 switch (cmd_ctx->lsm->domain.type) {
3547 case LTTNG_DOMAIN_KERNEL:
3548 cdata = &kconsumer_data;
3549 break;
3550 default:
3551 ret = LTTNG_ERR_UND;
3552 goto error;
3553 }
3554
3555 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3556 cmd_ctx->lsm->u.reg.path, cdata);
3557 break;
3558 }
3559 case LTTNG_DATA_PENDING:
3560 {
3561 ret = cmd_data_pending(cmd_ctx->session);
3562 break;
3563 }
3564 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3565 {
3566 struct lttcomm_lttng_output_id reply;
3567
3568 ret = cmd_snapshot_add_output(cmd_ctx->session,
3569 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3570 if (ret != LTTNG_OK) {
3571 goto error;
3572 }
3573
3574 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3575 if (ret < 0) {
3576 goto setup_error;
3577 }
3578
3579 /* Copy output list into message payload */
3580 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3581 ret = LTTNG_OK;
3582 break;
3583 }
3584 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3585 {
3586 ret = cmd_snapshot_del_output(cmd_ctx->session,
3587 &cmd_ctx->lsm->u.snapshot_output.output);
3588 break;
3589 }
3590 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3591 {
3592 ssize_t nb_output;
3593 struct lttng_snapshot_output *outputs = NULL;
3594
3595 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3596 if (nb_output < 0) {
3597 ret = -nb_output;
3598 goto error;
3599 }
3600
3601 ret = setup_lttng_msg(cmd_ctx,
3602 nb_output * sizeof(struct lttng_snapshot_output));
3603 if (ret < 0) {
3604 free(outputs);
3605 goto setup_error;
3606 }
3607
3608 if (outputs) {
3609 /* Copy output list into message payload */
3610 memcpy(cmd_ctx->llm->payload, outputs,
3611 nb_output * sizeof(struct lttng_snapshot_output));
3612 free(outputs);
3613 }
3614
3615 ret = LTTNG_OK;
3616 break;
3617 }
3618 case LTTNG_SNAPSHOT_RECORD:
3619 {
3620 ret = cmd_snapshot_record(cmd_ctx->session,
3621 &cmd_ctx->lsm->u.snapshot_record.output,
3622 cmd_ctx->lsm->u.snapshot_record.wait);
3623 break;
3624 }
3625 case LTTNG_CREATE_SESSION_SNAPSHOT:
3626 {
3627 size_t nb_uri, len;
3628 struct lttng_uri *uris = NULL;
3629
3630 nb_uri = cmd_ctx->lsm->u.uri.size;
3631 len = nb_uri * sizeof(struct lttng_uri);
3632
3633 if (nb_uri > 0) {
3634 uris = zmalloc(len);
3635 if (uris == NULL) {
3636 ret = LTTNG_ERR_FATAL;
3637 goto error;
3638 }
3639
3640 /* Receive variable len data */
3641 DBG("Waiting for %zu URIs from client ...", nb_uri);
3642 ret = lttcomm_recv_unix_sock(sock, uris, len);
3643 if (ret <= 0) {
3644 DBG("No URIs received from client... continuing");
3645 *sock_error = 1;
3646 ret = LTTNG_ERR_SESSION_FAIL;
3647 free(uris);
3648 goto error;
3649 }
3650
3651 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3652 DBG("Creating session with ONE network URI is a bad call");
3653 ret = LTTNG_ERR_SESSION_FAIL;
3654 free(uris);
3655 goto error;
3656 }
3657 }
3658
3659 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3660 nb_uri, &cmd_ctx->creds);
3661 free(uris);
3662 break;
3663 }
3664 case LTTNG_CREATE_SESSION_LIVE:
3665 {
3666 size_t nb_uri, len;
3667 struct lttng_uri *uris = NULL;
3668
3669 nb_uri = cmd_ctx->lsm->u.uri.size;
3670 len = nb_uri * sizeof(struct lttng_uri);
3671
3672 if (nb_uri > 0) {
3673 uris = zmalloc(len);
3674 if (uris == NULL) {
3675 ret = LTTNG_ERR_FATAL;
3676 goto error;
3677 }
3678
3679 /* Receive variable len data */
3680 DBG("Waiting for %zu URIs from client ...", nb_uri);
3681 ret = lttcomm_recv_unix_sock(sock, uris, len);
3682 if (ret <= 0) {
3683 DBG("No URIs received from client... continuing");
3684 *sock_error = 1;
3685 ret = LTTNG_ERR_SESSION_FAIL;
3686 free(uris);
3687 goto error;
3688 }
3689
3690 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3691 DBG("Creating session with ONE network URI is a bad call");
3692 ret = LTTNG_ERR_SESSION_FAIL;
3693 free(uris);
3694 goto error;
3695 }
3696 }
3697
3698 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3699 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3700 free(uris);
3701 break;
3702 }
3703 case LTTNG_SAVE_SESSION:
3704 {
3705 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3706 &cmd_ctx->creds);
3707 break;
3708 }
3709 default:
3710 ret = LTTNG_ERR_UND;
3711 break;
3712 }
3713
3714 error:
3715 if (cmd_ctx->llm == NULL) {
3716 DBG("Missing llm structure. Allocating one.");
3717 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3718 goto setup_error;
3719 }
3720 }
3721 /* Set return code */
3722 cmd_ctx->llm->ret_code = ret;
3723 setup_error:
3724 if (cmd_ctx->session) {
3725 session_unlock(cmd_ctx->session);
3726 }
3727 if (need_tracing_session) {
3728 session_unlock_list();
3729 }
3730 init_setup_error:
3731 return ret;
3732 }
3733
3734 /*
3735 * Thread managing health check socket.
3736 */
3737 static void *thread_manage_health(void *data)
3738 {
3739 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3740 uint32_t revents, nb_fd;
3741 struct lttng_poll_event events;
3742 struct health_comm_msg msg;
3743 struct health_comm_reply reply;
3744
3745 DBG("[thread] Manage health check started");
3746
3747 rcu_register_thread();
3748
3749 /* We might hit an error path before this is created. */
3750 lttng_poll_init(&events);
3751
3752 /* Create unix socket */
3753 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3754 if (sock < 0) {
3755 ERR("Unable to create health check Unix socket");
3756 ret = -1;
3757 goto error;
3758 }
3759
3760 if (is_root) {
3761 /* lttng health client socket path permissions */
3762 ret = chown(health_unix_sock_path, 0,
3763 utils_get_group_id(tracing_group_name));
3764 if (ret < 0) {
3765 ERR("Unable to set group on %s", health_unix_sock_path);
3766 PERROR("chown");
3767 ret = -1;
3768 goto error;
3769 }
3770
3771 ret = chmod(health_unix_sock_path,
3772 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3773 if (ret < 0) {
3774 ERR("Unable to set permissions on %s", health_unix_sock_path);
3775 PERROR("chmod");
3776 ret = -1;
3777 goto error;
3778 }
3779 }
3780
3781 /*
3782 * Set the CLOEXEC flag. Return code is useless because either way, the
3783 * show must go on.
3784 */
3785 (void) utils_set_fd_cloexec(sock);
3786
3787 ret = lttcomm_listen_unix_sock(sock);
3788 if (ret < 0) {
3789 goto error;
3790 }
3791
3792 /*
3793 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3794 * more will be added to this poll set.
3795 */
3796 ret = sessiond_set_thread_pollset(&events, 2);
3797 if (ret < 0) {
3798 goto error;
3799 }
3800
3801 /* Add the application registration socket */
3802 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3803 if (ret < 0) {
3804 goto error;
3805 }
3806
3807 sessiond_notify_ready();
3808
3809 while (1) {
3810 DBG("Health check ready");
3811
3812 /* Inifinite blocking call, waiting for transmission */
3813 restart:
3814 ret = lttng_poll_wait(&events, -1);
3815 if (ret < 0) {
3816 /*
3817 * Restart interrupted system call.
3818 */
3819 if (errno == EINTR) {
3820 goto restart;
3821 }
3822 goto error;
3823 }
3824
3825 nb_fd = ret;
3826
3827 for (i = 0; i < nb_fd; i++) {
3828 /* Fetch once the poll data */
3829 revents = LTTNG_POLL_GETEV(&events, i);
3830 pollfd = LTTNG_POLL_GETFD(&events, i);
3831
3832 /* Thread quit pipe has been closed. Killing thread. */
3833 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3834 if (ret) {
3835 err = 0;
3836 goto exit;
3837 }
3838
3839 /* Event on the registration socket */
3840 if (pollfd == sock) {
3841 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3842 ERR("Health socket poll error");
3843 goto error;
3844 }
3845 }
3846 }
3847
3848 new_sock = lttcomm_accept_unix_sock(sock);
3849 if (new_sock < 0) {
3850 goto error;
3851 }
3852
3853 /*
3854 * Set the CLOEXEC flag. Return code is useless because either way, the
3855 * show must go on.
3856 */
3857 (void) utils_set_fd_cloexec(new_sock);
3858
3859 DBG("Receiving data from client for health...");
3860 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3861 if (ret <= 0) {
3862 DBG("Nothing recv() from client... continuing");
3863 ret = close(new_sock);
3864 if (ret) {
3865 PERROR("close");
3866 }
3867 new_sock = -1;
3868 continue;
3869 }
3870
3871 rcu_thread_online();
3872
3873 memset(&reply, 0, sizeof(reply));
3874 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3875 /*
3876 * health_check_state returns 0 if health is
3877 * bad.
3878 */
3879 if (!health_check_state(health_sessiond, i)) {
3880 reply.ret_code |= 1ULL << i;
3881 }
3882 }
3883
3884 DBG2("Health check return value %" PRIx64, reply.ret_code);
3885
3886 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3887 if (ret < 0) {
3888 ERR("Failed to send health data back to client");
3889 }
3890
3891 /* End of transmission */
3892 ret = close(new_sock);
3893 if (ret) {
3894 PERROR("close");
3895 }
3896 new_sock = -1;
3897 }
3898
3899 exit:
3900 error:
3901 if (err) {
3902 ERR("Health error occurred in %s", __func__);
3903 }
3904 DBG("Health check thread dying");
3905 unlink(health_unix_sock_path);
3906 if (sock >= 0) {
3907 ret = close(sock);
3908 if (ret) {
3909 PERROR("close");
3910 }
3911 }
3912
3913 lttng_poll_clean(&events);
3914
3915 rcu_unregister_thread();
3916 return NULL;
3917 }
3918
3919 /*
3920 * This thread manage all clients request using the unix client socket for
3921 * communication.
3922 */
3923 static void *thread_manage_clients(void *data)
3924 {
3925 int sock = -1, ret, i, pollfd, err = -1;
3926 int sock_error;
3927 uint32_t revents, nb_fd;
3928 struct command_ctx *cmd_ctx = NULL;
3929 struct lttng_poll_event events;
3930
3931 DBG("[thread] Manage client started");
3932
3933 rcu_register_thread();
3934
3935 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3936
3937 health_code_update();
3938
3939 ret = lttcomm_listen_unix_sock(client_sock);
3940 if (ret < 0) {
3941 goto error_listen;
3942 }
3943
3944 /*
3945 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3946 * more will be added to this poll set.
3947 */
3948 ret = sessiond_set_thread_pollset(&events, 2);
3949 if (ret < 0) {
3950 goto error_create_poll;
3951 }
3952
3953 /* Add the application registration socket */
3954 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3955 if (ret < 0) {
3956 goto error;
3957 }
3958
3959 sessiond_notify_ready();
3960 ret = sem_post(&load_info->message_thread_ready);
3961 if (ret) {
3962 PERROR("sem_post message_thread_ready");
3963 goto error;
3964 }
3965
3966 /* This testpoint is after we signal readiness to the parent. */
3967 if (testpoint(sessiond_thread_manage_clients)) {
3968 goto error;
3969 }
3970
3971 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
3972 goto error;
3973 }
3974
3975 health_code_update();
3976
3977 while (1) {
3978 DBG("Accepting client command ...");
3979
3980 /* Inifinite blocking call, waiting for transmission */
3981 restart:
3982 health_poll_entry();
3983 ret = lttng_poll_wait(&events, -1);
3984 health_poll_exit();
3985 if (ret < 0) {
3986 /*
3987 * Restart interrupted system call.
3988 */
3989 if (errno == EINTR) {
3990 goto restart;
3991 }
3992 goto error;
3993 }
3994
3995 nb_fd = ret;
3996
3997 for (i = 0; i < nb_fd; i++) {
3998 /* Fetch once the poll data */
3999 revents = LTTNG_POLL_GETEV(&events, i);
4000 pollfd = LTTNG_POLL_GETFD(&events, i);
4001
4002 health_code_update();
4003
4004 /* Thread quit pipe has been closed. Killing thread. */
4005 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4006 if (ret) {
4007 err = 0;
4008 goto exit;
4009 }
4010
4011 /* Event on the registration socket */
4012 if (pollfd == client_sock) {
4013 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4014 ERR("Client socket poll error");
4015 goto error;
4016 }
4017 }
4018 }
4019
4020 DBG("Wait for client response");
4021
4022 health_code_update();
4023
4024 sock = lttcomm_accept_unix_sock(client_sock);
4025 if (sock < 0) {
4026 goto error;
4027 }
4028
4029 /*
4030 * Set the CLOEXEC flag. Return code is useless because either way, the
4031 * show must go on.
4032 */
4033 (void) utils_set_fd_cloexec(sock);
4034
4035 /* Set socket option for credentials retrieval */
4036 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4037 if (ret < 0) {
4038 goto error;
4039 }
4040
4041 /* Allocate context command to process the client request */
4042 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4043 if (cmd_ctx == NULL) {
4044 PERROR("zmalloc cmd_ctx");
4045 goto error;
4046 }
4047
4048 /* Allocate data buffer for reception */
4049 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4050 if (cmd_ctx->lsm == NULL) {
4051 PERROR("zmalloc cmd_ctx->lsm");
4052 goto error;
4053 }
4054
4055 cmd_ctx->llm = NULL;
4056 cmd_ctx->session = NULL;
4057
4058 health_code_update();
4059
4060 /*
4061 * Data is received from the lttng client. The struct
4062 * lttcomm_session_msg (lsm) contains the command and data request of
4063 * the client.
4064 */
4065 DBG("Receiving data from client ...");
4066 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4067 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4068 if (ret <= 0) {
4069 DBG("Nothing recv() from client... continuing");
4070 ret = close(sock);
4071 if (ret) {
4072 PERROR("close");
4073 }
4074 sock = -1;
4075 clean_command_ctx(&cmd_ctx);
4076 continue;
4077 }
4078
4079 health_code_update();
4080
4081 // TODO: Validate cmd_ctx including sanity check for
4082 // security purpose.
4083
4084 rcu_thread_online();
4085 /*
4086 * This function dispatch the work to the kernel or userspace tracer
4087 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4088 * informations for the client. The command context struct contains
4089 * everything this function may needs.
4090 */
4091 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4092 rcu_thread_offline();
4093 if (ret < 0) {
4094 ret = close(sock);
4095 if (ret) {
4096 PERROR("close");
4097 }
4098 sock = -1;
4099 /*
4100 * TODO: Inform client somehow of the fatal error. At
4101 * this point, ret < 0 means that a zmalloc failed
4102 * (ENOMEM). Error detected but still accept
4103 * command, unless a socket error has been
4104 * detected.
4105 */
4106 clean_command_ctx(&cmd_ctx);
4107 continue;
4108 }
4109
4110 health_code_update();
4111
4112 DBG("Sending response (size: %d, retcode: %s)",
4113 cmd_ctx->lttng_msg_size,
4114 lttng_strerror(-cmd_ctx->llm->ret_code));
4115 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4116 if (ret < 0) {
4117 ERR("Failed to send data back to client");
4118 }
4119
4120 /* End of transmission */
4121 ret = close(sock);
4122 if (ret) {
4123 PERROR("close");
4124 }
4125 sock = -1;
4126
4127 clean_command_ctx(&cmd_ctx);
4128
4129 health_code_update();
4130 }
4131
4132 exit:
4133 error:
4134 if (sock >= 0) {
4135 ret = close(sock);
4136 if (ret) {
4137 PERROR("close");
4138 }
4139 }
4140
4141 lttng_poll_clean(&events);
4142 clean_command_ctx(&cmd_ctx);
4143
4144 error_listen:
4145 error_create_poll:
4146 unlink(client_unix_sock_path);
4147 if (client_sock >= 0) {
4148 ret = close(client_sock);
4149 if (ret) {
4150 PERROR("close");
4151 }
4152 }
4153
4154 if (err) {
4155 health_error();
4156 ERR("Health error occurred in %s", __func__);
4157 }
4158
4159 health_unregister(health_sessiond);
4160
4161 DBG("Client thread dying");
4162
4163 rcu_unregister_thread();
4164 return NULL;
4165 }
4166
4167
4168 /*
4169 * usage function on stderr
4170 */
4171 static void usage(void)
4172 {
4173 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4174 fprintf(stderr, " -h, --help Display this usage.\n");
4175 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4176 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4177 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4178 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4179 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4180 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4181 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4182 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4183 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4184 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4185 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4186 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4187 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4188 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4189 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4190 fprintf(stderr, " -V, --version Show version number.\n");
4191 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4192 fprintf(stderr, " -q, --quiet No output at all.\n");
4193 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4194 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4195 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4196 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4197 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4198 fprintf(stderr, " -f --config Load daemon configuration file\n");
4199 fprintf(stderr, " -l --load PATH Load session configuration\n");
4200 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4201 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4202 }
4203
4204 /*
4205 * Take an option from the getopt output and set it in the right variable to be
4206 * used later.
4207 *
4208 * Return 0 on success else a negative value.
4209 */
4210 static int set_option(int opt, const char *arg, const char *optname)
4211 {
4212 int ret = 0;
4213
4214 switch (opt) {
4215 case 0:
4216 fprintf(stderr, "option %s", optname);
4217 if (arg) {
4218 fprintf(stderr, " with arg %s\n", arg);
4219 }
4220 break;
4221 case 'c':
4222 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4223 break;
4224 case 'a':
4225 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4226 break;
4227 case 'd':
4228 opt_daemon = 1;
4229 break;
4230 case 'b':
4231 opt_background = 1;
4232 break;
4233 case 'g':
4234 /*
4235 * If the override option is set, the pointer points to a
4236 * *non* const thus freeing it even though the variable type is
4237 * set to const.
4238 */
4239 if (tracing_group_name_override) {
4240 free((void *) tracing_group_name);
4241 }
4242 tracing_group_name = strdup(arg);
4243 if (!tracing_group_name) {
4244 perror("strdup");
4245 ret = -ENOMEM;
4246 }
4247 tracing_group_name_override = 1;
4248 break;
4249 case 'h':
4250 usage();
4251 exit(EXIT_FAILURE);
4252 case 'V':
4253 fprintf(stdout, "%s\n", VERSION);
4254 exit(EXIT_SUCCESS);
4255 case 'S':
4256 opt_sig_parent = 1;
4257 break;
4258 case 'E':
4259 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4260 break;
4261 case 'C':
4262 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4263 break;
4264 case 'F':
4265 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4266 break;
4267 case 'D':
4268 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4269 break;
4270 case 'H':
4271 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4272 break;
4273 case 'G':
4274 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4275 break;
4276 case 'N':
4277 opt_no_kernel = 1;
4278 break;
4279 case 'q':
4280 lttng_opt_quiet = 1;
4281 break;
4282 case 'v':
4283 /* Verbose level can increase using multiple -v */
4284 if (arg) {
4285 /* Value obtained from config file */
4286 lttng_opt_verbose = config_parse_value(arg);
4287 } else {
4288 /* -v used on command line */
4289 lttng_opt_verbose++;
4290 }
4291 /* Clamp value to [0, 3] */
4292 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4293 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4294 break;
4295 case 'Z':
4296 if (arg) {
4297 opt_verbose_consumer = config_parse_value(arg);
4298 } else {
4299 opt_verbose_consumer += 1;
4300 }
4301 break;
4302 case 'u':
4303 if (consumerd32_bin_override) {
4304 free((void *) consumerd32_bin);
4305 }
4306 consumerd32_bin = strdup(arg);
4307 if (!consumerd32_bin) {
4308 perror("strdup");
4309 ret = -ENOMEM;
4310 }
4311 consumerd32_bin_override = 1;
4312 break;
4313 case 'U':
4314 if (consumerd32_libdir_override) {
4315 free((void *) consumerd32_libdir);
4316 }
4317 consumerd32_libdir = strdup(arg);
4318 if (!consumerd32_libdir) {
4319 perror("strdup");
4320 ret = -ENOMEM;
4321 }
4322 consumerd32_libdir_override = 1;
4323 break;
4324 case 't':
4325 if (consumerd64_bin_override) {
4326 free((void *) consumerd64_bin);
4327 }
4328 consumerd64_bin = strdup(arg);
4329 if (!consumerd64_bin) {
4330 perror("strdup");
4331 ret = -ENOMEM;
4332 }
4333 consumerd64_bin_override = 1;
4334 break;
4335 case 'T':
4336 if (consumerd64_libdir_override) {
4337 free((void *) consumerd64_libdir);
4338 }
4339 consumerd64_libdir = strdup(arg);
4340 if (!consumerd64_libdir) {
4341 perror("strdup");
4342 ret = -ENOMEM;
4343 }
4344 consumerd64_libdir_override = 1;
4345 break;
4346 case 'p':
4347 free(opt_pidfile);
4348 opt_pidfile = strdup(arg);
4349 if (!opt_pidfile) {
4350 perror("strdup");
4351 ret = -ENOMEM;
4352 }
4353 break;
4354 case 'J': /* Agent TCP port. */
4355 {
4356 unsigned long v;
4357
4358 errno = 0;
4359 v = strtoul(arg, NULL, 0);
4360 if (errno != 0 || !isdigit(arg[0])) {
4361 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4362 return -1;
4363 }
4364 if (v == 0 || v >= 65535) {
4365 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4366 return -1;
4367 }
4368 agent_tcp_port = (uint32_t) v;
4369 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4370 break;
4371 }
4372 case 'l':
4373 free(opt_load_session_path);
4374 opt_load_session_path = strdup(arg);
4375 if (!opt_load_session_path) {
4376 perror("strdup");
4377 ret = -ENOMEM;
4378 }
4379 break;
4380 case 'P': /* probe modules list */
4381 free(kmod_probes_list);
4382 kmod_probes_list = strdup(arg);
4383 if (!kmod_probes_list) {
4384 perror("strdup");
4385 ret = -ENOMEM;
4386 }
4387 break;
4388 case 'e':
4389 free(kmod_extra_probes_list);
4390 kmod_extra_probes_list = strdup(arg);
4391 if (!kmod_extra_probes_list) {
4392 perror("strdup");
4393 ret = -ENOMEM;
4394 }
4395 break;
4396 case 'f':
4397 /* This is handled in set_options() thus silent break. */
4398 break;
4399 default:
4400 /* Unknown option or other error.
4401 * Error is printed by getopt, just return */
4402 ret = -1;
4403 }
4404
4405 return ret;
4406 }
4407
4408 /*
4409 * config_entry_handler_cb used to handle options read from a config file.
4410 * See config_entry_handler_cb comment in common/config/config.h for the
4411 * return value conventions.
4412 */
4413 static int config_entry_handler(const struct config_entry *entry, void *unused)
4414 {
4415 int ret = 0, i;
4416
4417 if (!entry || !entry->name || !entry->value) {
4418 ret = -EINVAL;
4419 goto end;
4420 }
4421
4422 /* Check if the option is to be ignored */
4423 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4424 if (!strcmp(entry->name, config_ignore_options[i])) {
4425 goto end;
4426 }
4427 }
4428
4429 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4430 i++) {
4431
4432 /* Ignore if not fully matched. */
4433 if (strcmp(entry->name, long_options[i].name)) {
4434 continue;
4435 }
4436
4437 /*
4438 * If the option takes no argument on the command line, we have to
4439 * check if the value is "true". We support non-zero numeric values,
4440 * true, on and yes.
4441 */
4442 if (!long_options[i].has_arg) {
4443 ret = config_parse_value(entry->value);
4444 if (ret <= 0) {
4445 if (ret) {
4446 WARN("Invalid configuration value \"%s\" for option %s",
4447 entry->value, entry->name);
4448 }
4449 /* False, skip boolean config option. */
4450 goto end;
4451 }
4452 }
4453
4454 ret = set_option(long_options[i].val, entry->value, entry->name);
4455 goto end;
4456 }
4457
4458 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4459
4460 end:
4461 return ret;
4462 }
4463
4464 /*
4465 * daemon configuration loading and argument parsing
4466 */
4467 static int set_options(int argc, char **argv)
4468 {
4469 int ret = 0, c = 0, option_index = 0;
4470 int orig_optopt = optopt, orig_optind = optind;
4471 char *optstring;
4472 const char *config_path = NULL;
4473
4474 optstring = utils_generate_optstring(long_options,
4475 sizeof(long_options) / sizeof(struct option));
4476 if (!optstring) {
4477 ret = -ENOMEM;
4478 goto end;
4479 }
4480
4481 /* Check for the --config option */
4482 while ((c = getopt_long(argc, argv, optstring, long_options,
4483 &option_index)) != -1) {
4484 if (c == '?') {
4485 ret = -EINVAL;
4486 goto end;
4487 } else if (c != 'f') {
4488 /* if not equal to --config option. */
4489 continue;
4490 }
4491
4492 config_path = utils_expand_path(optarg);
4493 if (!config_path) {
4494 ERR("Failed to resolve path: %s", optarg);
4495 }
4496 }
4497
4498 ret = config_get_section_entries(config_path, config_section_name,
4499 config_entry_handler, NULL);
4500 if (ret) {
4501 if (ret > 0) {
4502 ERR("Invalid configuration option at line %i", ret);
4503 ret = -1;
4504 }
4505 goto end;
4506 }
4507
4508 /* Reset getopt's global state */
4509 optopt = orig_optopt;
4510 optind = orig_optind;
4511 while (1) {
4512 c = getopt_long(argc, argv, optstring, long_options, &option_index);
4513 if (c == -1) {
4514 break;
4515 }
4516
4517 ret = set_option(c, optarg, long_options[option_index].name);
4518 if (ret < 0) {
4519 break;
4520 }
4521 }
4522
4523 end:
4524 free(optstring);
4525 return ret;
4526 }
4527
4528 /*
4529 * Creates the two needed socket by the daemon.
4530 * apps_sock - The communication socket for all UST apps.
4531 * client_sock - The communication of the cli tool (lttng).
4532 */
4533 static int init_daemon_socket(void)
4534 {
4535 int ret = 0;
4536 mode_t old_umask;
4537
4538 old_umask = umask(0);
4539
4540 /* Create client tool unix socket */
4541 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4542 if (client_sock < 0) {
4543 ERR("Create unix sock failed: %s", client_unix_sock_path);
4544 ret = -1;
4545 goto end;
4546 }
4547
4548 /* Set the cloexec flag */
4549 ret = utils_set_fd_cloexec(client_sock);
4550 if (ret < 0) {
4551 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4552 "Continuing but note that the consumer daemon will have a "
4553 "reference to this socket on exec()", client_sock);
4554 }
4555
4556 /* File permission MUST be 660 */
4557 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4558 if (ret < 0) {
4559 ERR("Set file permissions failed: %s", client_unix_sock_path);
4560 PERROR("chmod");
4561 goto end;
4562 }
4563
4564 /* Create the application unix socket */
4565 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4566 if (apps_sock < 0) {
4567 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4568 ret = -1;
4569 goto end;
4570 }
4571
4572 /* Set the cloexec flag */
4573 ret = utils_set_fd_cloexec(apps_sock);
4574 if (ret < 0) {
4575 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4576 "Continuing but note that the consumer daemon will have a "
4577 "reference to this socket on exec()", apps_sock);
4578 }
4579
4580 /* File permission MUST be 666 */
4581 ret = chmod(apps_unix_sock_path,
4582 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4583 if (ret < 0) {
4584 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4585 PERROR("chmod");
4586 goto end;
4587 }
4588
4589 DBG3("Session daemon client socket %d and application socket %d created",
4590 client_sock, apps_sock);
4591
4592 end:
4593 umask(old_umask);
4594 return ret;
4595 }
4596
4597 /*
4598 * Check if the global socket is available, and if a daemon is answering at the
4599 * other side. If yes, error is returned.
4600 */
4601 static int check_existing_daemon(void)
4602 {
4603 /* Is there anybody out there ? */
4604 if (lttng_session_daemon_alive()) {
4605 return -EEXIST;
4606 }
4607
4608 return 0;
4609 }
4610
4611 /*
4612 * Set the tracing group gid onto the client socket.
4613 *
4614 * Race window between mkdir and chown is OK because we are going from more
4615 * permissive (root.root) to less permissive (root.tracing).
4616 */
4617 static int set_permissions(char *rundir)
4618 {
4619 int ret;
4620 gid_t gid;
4621
4622 gid = utils_get_group_id(tracing_group_name);
4623
4624 /* Set lttng run dir */
4625 ret = chown(rundir, 0, gid);
4626 if (ret < 0) {
4627 ERR("Unable to set group on %s", rundir);
4628 PERROR("chown");
4629 }
4630
4631 /*
4632 * Ensure all applications and tracing group can search the run
4633 * dir. Allow everyone to read the directory, since it does not
4634 * buy us anything to hide its content.
4635 */
4636 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4637 if (ret < 0) {
4638 ERR("Unable to set permissions on %s", rundir);
4639 PERROR("chmod");
4640 }
4641
4642 /* lttng client socket path */
4643 ret = chown(client_unix_sock_path, 0, gid);
4644 if (ret < 0) {
4645 ERR("Unable to set group on %s", client_unix_sock_path);
4646 PERROR("chown");
4647 }
4648
4649 /* kconsumer error socket path */
4650 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4651 if (ret < 0) {
4652 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4653 PERROR("chown");
4654 }
4655
4656 /* 64-bit ustconsumer error socket path */
4657 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4658 if (ret < 0) {
4659 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4660 PERROR("chown");
4661 }
4662
4663 /* 32-bit ustconsumer compat32 error socket path */
4664 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4665 if (ret < 0) {
4666 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4667 PERROR("chown");
4668 }
4669
4670 DBG("All permissions are set");
4671
4672 return ret;
4673 }
4674
4675 /*
4676 * Create the lttng run directory needed for all global sockets and pipe.
4677 */
4678 static int create_lttng_rundir(const char *rundir)
4679 {
4680 int ret;
4681
4682 DBG3("Creating LTTng run directory: %s", rundir);
4683
4684 ret = mkdir(rundir, S_IRWXU);
4685 if (ret < 0) {
4686 if (errno != EEXIST) {
4687 ERR("Unable to create %s", rundir);
4688 goto error;
4689 } else {
4690 ret = 0;
4691 }
4692 }
4693
4694 error:
4695 return ret;
4696 }
4697
4698 /*
4699 * Setup sockets and directory needed by the kconsumerd communication with the
4700 * session daemon.
4701 */
4702 static int set_consumer_sockets(struct consumer_data *consumer_data,
4703 const char *rundir)
4704 {
4705 int ret;
4706 char path[PATH_MAX];
4707
4708 switch (consumer_data->type) {
4709 case LTTNG_CONSUMER_KERNEL:
4710 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4711 break;
4712 case LTTNG_CONSUMER64_UST:
4713 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4714 break;
4715 case LTTNG_CONSUMER32_UST:
4716 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4717 break;
4718 default:
4719 ERR("Consumer type unknown");
4720 ret = -EINVAL;
4721 goto error;
4722 }
4723
4724 DBG2("Creating consumer directory: %s", path);
4725
4726 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4727 if (ret < 0) {
4728 if (errno != EEXIST) {
4729 PERROR("mkdir");
4730 ERR("Failed to create %s", path);
4731 goto error;
4732 }
4733 ret = -1;
4734 }
4735 if (is_root) {
4736 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4737 if (ret < 0) {
4738 ERR("Unable to set group on %s", path);
4739 PERROR("chown");
4740 goto error;
4741 }
4742 }
4743
4744 /* Create the kconsumerd error unix socket */
4745 consumer_data->err_sock =
4746 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4747 if (consumer_data->err_sock < 0) {
4748 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4749 ret = -1;
4750 goto error;
4751 }
4752
4753 /*
4754 * Set the CLOEXEC flag. Return code is useless because either way, the
4755 * show must go on.
4756 */
4757 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4758 if (ret < 0) {
4759 PERROR("utils_set_fd_cloexec");
4760 /* continue anyway */
4761 }
4762
4763 /* File permission MUST be 660 */
4764 ret = chmod(consumer_data->err_unix_sock_path,
4765 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4766 if (ret < 0) {
4767 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4768 PERROR("chmod");
4769 goto error;
4770 }
4771
4772 error:
4773 return ret;
4774 }
4775
4776 /*
4777 * Signal handler for the daemon
4778 *
4779 * Simply stop all worker threads, leaving main() return gracefully after
4780 * joining all threads and calling cleanup().
4781 */
4782 static void sighandler(int sig)
4783 {
4784 switch (sig) {
4785 case SIGPIPE:
4786 DBG("SIGPIPE caught");
4787 return;
4788 case SIGINT:
4789 DBG("SIGINT caught");
4790 stop_threads();
4791 break;
4792 case SIGTERM:
4793 DBG("SIGTERM caught");
4794 stop_threads();
4795 break;
4796 case SIGUSR1:
4797 CMM_STORE_SHARED(recv_child_signal, 1);
4798 break;
4799 default:
4800 break;
4801 }
4802 }
4803
4804 /*
4805 * Setup signal handler for :
4806 * SIGINT, SIGTERM, SIGPIPE
4807 */
4808 static int set_signal_handler(void)
4809 {
4810 int ret = 0;
4811 struct sigaction sa;
4812 sigset_t sigset;
4813
4814 if ((ret = sigemptyset(&sigset)) < 0) {
4815 PERROR("sigemptyset");
4816 return ret;
4817 }
4818
4819 sa.sa_handler = sighandler;
4820 sa.sa_mask = sigset;
4821 sa.sa_flags = 0;
4822 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4823 PERROR("sigaction");
4824 return ret;
4825 }
4826
4827 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4828 PERROR("sigaction");
4829 return ret;
4830 }
4831
4832 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4833 PERROR("sigaction");
4834 return ret;
4835 }
4836
4837 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4838 PERROR("sigaction");
4839 return ret;
4840 }
4841
4842 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4843
4844 return ret;
4845 }
4846
4847 /*
4848 * Set open files limit to unlimited. This daemon can open a large number of
4849 * file descriptors in order to consumer multiple kernel traces.
4850 */
4851 static void set_ulimit(void)
4852 {
4853 int ret;
4854 struct rlimit lim;
4855
4856 /* The kernel does not allowed an infinite limit for open files */
4857 lim.rlim_cur = 65535;
4858 lim.rlim_max = 65535;
4859
4860 ret = setrlimit(RLIMIT_NOFILE, &lim);
4861 if (ret < 0) {
4862 PERROR("failed to set open files limit");
4863 }
4864 }
4865
4866 /*
4867 * Write pidfile using the rundir and opt_pidfile.
4868 */
4869 static void write_pidfile(void)
4870 {
4871 int ret;
4872 char pidfile_path[PATH_MAX];
4873
4874 assert(rundir);
4875
4876 if (opt_pidfile) {
4877 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4878 } else {
4879 /* Build pidfile path from rundir and opt_pidfile. */
4880 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4881 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4882 if (ret < 0) {
4883 PERROR("snprintf pidfile path");
4884 goto error;
4885 }
4886 }
4887
4888 /*
4889 * Create pid file in rundir. Return value is of no importance. The
4890 * execution will continue even though we are not able to write the file.
4891 */
4892 (void) utils_create_pid_file(getpid(), pidfile_path);
4893
4894 error:
4895 return;
4896 }
4897
4898 /*
4899 * Create lockfile using the rundir and return its fd.
4900 */
4901 static int create_lockfile(void)
4902 {
4903 int ret;
4904 char lockfile_path[PATH_MAX];
4905
4906 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
4907 if (ret < 0) {
4908 goto error;
4909 }
4910
4911 ret = utils_create_lock_file(lockfile_path);
4912 error:
4913 return ret;
4914 }
4915
4916 /*
4917 * Write agent TCP port using the rundir.
4918 */
4919 static void write_agent_port(void)
4920 {
4921 int ret;
4922 char path[PATH_MAX];
4923
4924 assert(rundir);
4925
4926 ret = snprintf(path, sizeof(path), "%s/"
4927 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
4928 if (ret < 0) {
4929 PERROR("snprintf agent port path");
4930 goto error;
4931 }
4932
4933 /*
4934 * Create TCP agent port file in rundir. Return value is of no importance.
4935 * The execution will continue even though we are not able to write the
4936 * file.
4937 */
4938 (void) utils_create_pid_file(agent_tcp_port, path);
4939
4940 error:
4941 return;
4942 }
4943
4944 /*
4945 * Start the load session thread and dettach from it so the main thread can
4946 * continue. This does not return a value since whatever the outcome, the main
4947 * thread will continue.
4948 */
4949 static void start_load_session_thread(void)
4950 {
4951 int ret;
4952
4953 /* Create session loading thread. */
4954 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
4955 load_info);
4956 if (ret != 0) {
4957 PERROR("pthread_create load_session_thread");
4958 goto error_create;
4959 }
4960
4961 ret = pthread_detach(load_session_thread);
4962 if (ret != 0) {
4963 PERROR("pthread_detach load_session_thread");
4964 }
4965
4966 /* Everything went well so don't cleanup anything. */
4967
4968 error_create:
4969 /* The cleanup() function will destroy the load_info data. */
4970 return;
4971 }
4972
4973 /*
4974 * main
4975 */
4976 int main(int argc, char **argv)
4977 {
4978 int ret = 0;
4979 void *status;
4980 const char *home_path, *env_app_timeout;
4981
4982 init_kernel_workarounds();
4983
4984 rcu_register_thread();
4985
4986 if ((ret = set_signal_handler()) < 0) {
4987 goto error;
4988 }
4989
4990 setup_consumerd_path();
4991
4992 page_size = sysconf(_SC_PAGESIZE);
4993 if (page_size < 0) {
4994 PERROR("sysconf _SC_PAGESIZE");
4995 page_size = LONG_MAX;
4996 WARN("Fallback page size to %ld", page_size);
4997 }
4998
4999 /* Parse arguments and load the daemon configuration file */
5000 progname = argv[0];
5001 if ((ret = set_options(argc, argv)) < 0) {
5002 goto error;
5003 }
5004
5005 /* Daemonize */
5006 if (opt_daemon || opt_background) {
5007 int i;
5008
5009 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5010 !opt_background);
5011 if (ret < 0) {
5012 goto error;
5013 }
5014
5015 /*
5016 * We are in the child. Make sure all other file descriptors are
5017 * closed, in case we are called with more opened file descriptors than
5018 * the standard ones.
5019 */
5020 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5021 (void) close(i);
5022 }
5023 }
5024
5025 /* Create thread quit pipe */
5026 if ((ret = init_thread_quit_pipe()) < 0) {
5027 goto error;
5028 }
5029
5030 /* Check if daemon is UID = 0 */
5031 is_root = !getuid();
5032
5033 if (is_root) {
5034 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5035
5036 /* Create global run dir with root access */
5037 ret = create_lttng_rundir(rundir);
5038 if (ret < 0) {
5039 goto error;
5040 }
5041
5042 if (strlen(apps_unix_sock_path) == 0) {
5043 snprintf(apps_unix_sock_path, PATH_MAX,
5044 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5045 }
5046
5047 if (strlen(client_unix_sock_path) == 0) {
5048 snprintf(client_unix_sock_path, PATH_MAX,
5049 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5050 }
5051
5052 /* Set global SHM for ust */
5053 if (strlen(wait_shm_path) == 0) {
5054 snprintf(wait_shm_path, PATH_MAX,
5055 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5056 }
5057
5058 if (strlen(health_unix_sock_path) == 0) {
5059 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
5060 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5061 }
5062
5063 /* Setup kernel consumerd path */
5064 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5065 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5066 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5067 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5068
5069 DBG2("Kernel consumer err path: %s",
5070 kconsumer_data.err_unix_sock_path);
5071 DBG2("Kernel consumer cmd path: %s",
5072 kconsumer_data.cmd_unix_sock_path);
5073 } else {
5074 home_path = utils_get_home_dir();
5075 if (home_path == NULL) {
5076 /* TODO: Add --socket PATH option */
5077 ERR("Can't get HOME directory for sockets creation.");
5078 ret = -EPERM;
5079 goto error;
5080 }
5081
5082 /*
5083 * Create rundir from home path. This will create something like
5084 * $HOME/.lttng
5085 */
5086 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5087 if (ret < 0) {
5088 ret = -ENOMEM;
5089 goto error;
5090 }
5091
5092 ret = create_lttng_rundir(rundir);
5093 if (ret < 0) {
5094 goto error;
5095 }
5096
5097 if (strlen(apps_unix_sock_path) == 0) {
5098 snprintf(apps_unix_sock_path, PATH_MAX,
5099 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
5100 }
5101
5102 /* Set the cli tool unix socket path */
5103 if (strlen(client_unix_sock_path) == 0) {
5104 snprintf(client_unix_sock_path, PATH_MAX,
5105 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
5106 }
5107
5108 /* Set global SHM for ust */
5109 if (strlen(wait_shm_path) == 0) {
5110 snprintf(wait_shm_path, PATH_MAX,
5111 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
5112 }
5113
5114 /* Set health check Unix path */
5115 if (strlen(health_unix_sock_path) == 0) {
5116 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
5117 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
5118 }
5119 }
5120
5121 lockfile_fd = create_lockfile();
5122 if (lockfile_fd < 0) {
5123 goto error;
5124 }
5125
5126 /* Set consumer initial state */
5127 kernel_consumerd_state = CONSUMER_STOPPED;
5128 ust_consumerd_state = CONSUMER_STOPPED;
5129
5130 DBG("Client socket path %s", client_unix_sock_path);
5131 DBG("Application socket path %s", apps_unix_sock_path);
5132 DBG("Application wait path %s", wait_shm_path);
5133 DBG("LTTng run directory path: %s", rundir);
5134
5135 /* 32 bits consumerd path setup */
5136 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5137 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5138 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5139 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5140
5141 DBG2("UST consumer 32 bits err path: %s",
5142 ustconsumer32_data.err_unix_sock_path);
5143 DBG2("UST consumer 32 bits cmd path: %s",
5144 ustconsumer32_data.cmd_unix_sock_path);
5145
5146 /* 64 bits consumerd path setup */
5147 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5148 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5149 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5150 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5151
5152 DBG2("UST consumer 64 bits err path: %s",
5153 ustconsumer64_data.err_unix_sock_path);
5154 DBG2("UST consumer 64 bits cmd path: %s",
5155 ustconsumer64_data.cmd_unix_sock_path);
5156
5157 /*
5158 * See if daemon already exist.
5159 */
5160 if ((ret = check_existing_daemon()) < 0) {
5161 ERR("Already running daemon.\n");
5162 /*
5163 * We do not goto exit because we must not cleanup()
5164 * because a daemon is already running.
5165 */
5166 goto error;
5167 }
5168
5169 /*
5170 * Init UST app hash table. Alloc hash table before this point since
5171 * cleanup() can get called after that point.
5172 */
5173 ust_app_ht_alloc();
5174
5175 /* Initialize agent domain subsystem. */
5176 if ((ret = agent_setup()) < 0) {
5177 /* ENOMEM at this point. */
5178 goto error;
5179 }
5180
5181 /* After this point, we can safely call cleanup() with "goto exit" */
5182
5183 /*
5184 * These actions must be executed as root. We do that *after* setting up
5185 * the sockets path because we MUST make the check for another daemon using
5186 * those paths *before* trying to set the kernel consumer sockets and init
5187 * kernel tracer.
5188 */
5189 if (is_root) {
5190 ret = set_consumer_sockets(&kconsumer_data, rundir);
5191 if (ret < 0) {
5192 goto exit;
5193 }
5194
5195 /* Setup kernel tracer */
5196 if (!opt_no_kernel) {
5197 init_kernel_tracer();
5198 }
5199
5200 /* Set ulimit for open files */
5201 set_ulimit();
5202 }
5203 /* init lttng_fd tracking must be done after set_ulimit. */
5204 lttng_fd_init();
5205
5206 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
5207 if (ret < 0) {
5208 goto exit;
5209 }
5210
5211 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
5212 if (ret < 0) {
5213 goto exit;
5214 }
5215
5216 /* Setup the needed unix socket */
5217 if ((ret = init_daemon_socket()) < 0) {
5218 goto exit;
5219 }
5220
5221 /* Set credentials to socket */
5222 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
5223 goto exit;
5224 }
5225
5226 /* Get parent pid if -S, --sig-parent is specified. */
5227 if (opt_sig_parent) {
5228 ppid = getppid();
5229 }
5230
5231 /* Setup the kernel pipe for waking up the kernel thread */
5232 if (is_root && !opt_no_kernel) {
5233 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
5234 goto exit;
5235 }
5236 }
5237
5238 /* Setup the thread ht_cleanup communication pipe. */
5239 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
5240 goto exit;
5241 }
5242
5243 /* Setup the thread apps communication pipe. */
5244 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
5245 goto exit;
5246 }
5247
5248 /* Setup the thread apps notify communication pipe. */
5249 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
5250 goto exit;
5251 }
5252
5253 /* Initialize global buffer per UID and PID registry. */
5254 buffer_reg_init_uid_registry();
5255 buffer_reg_init_pid_registry();
5256
5257 /* Init UST command queue. */
5258 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5259
5260 /*
5261 * Get session list pointer. This pointer MUST NOT be free(). This list is
5262 * statically declared in session.c
5263 */
5264 session_list_ptr = session_get_list();
5265
5266 /* Set up max poll set size */
5267 lttng_poll_set_max_size();
5268
5269 cmd_init();
5270
5271 /* Check for the application socket timeout env variable. */
5272 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5273 if (env_app_timeout) {
5274 app_socket_timeout = atoi(env_app_timeout);
5275 } else {
5276 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5277 }
5278
5279 write_pidfile();
5280 write_agent_port();
5281
5282 /* Initialize communication library */
5283 lttcomm_init();
5284 /* This is to get the TCP timeout value. */
5285 lttcomm_inet_init();
5286
5287 if (load_session_init_data(&load_info) < 0) {
5288 goto exit;
5289 }
5290 load_info->path = opt_load_session_path;
5291
5292 /*
5293 * Initialize the health check subsystem. This call should set the
5294 * appropriate time values.
5295 */
5296 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5297 if (!health_sessiond) {
5298 PERROR("health_app_create error");
5299 goto exit_health_sessiond_cleanup;
5300 }
5301
5302 /* Create thread to clean up RCU hash tables */
5303 ret = pthread_create(&ht_cleanup_thread, NULL,
5304 thread_ht_cleanup, (void *) NULL);
5305 if (ret != 0) {
5306 PERROR("pthread_create ht_cleanup");
5307 goto exit_ht_cleanup;
5308 }
5309
5310 /* Create health-check thread */
5311 ret = pthread_create(&health_thread, NULL,
5312 thread_manage_health, (void *) NULL);
5313 if (ret != 0) {
5314 PERROR("pthread_create health");
5315 goto exit_health;
5316 }
5317
5318 /* Create thread to manage the client socket */
5319 ret = pthread_create(&client_thread, NULL,
5320 thread_manage_clients, (void *) NULL);
5321 if (ret != 0) {
5322 PERROR("pthread_create clients");
5323 goto exit_client;
5324 }
5325
5326 /* Create thread to dispatch registration */
5327 ret = pthread_create(&dispatch_thread, NULL,
5328 thread_dispatch_ust_registration, (void *) NULL);
5329 if (ret != 0) {
5330 PERROR("pthread_create dispatch");
5331 goto exit_dispatch;
5332 }
5333
5334 /* Create thread to manage application registration. */
5335 ret = pthread_create(&reg_apps_thread, NULL,
5336 thread_registration_apps, (void *) NULL);
5337 if (ret != 0) {
5338 PERROR("pthread_create registration");
5339 goto exit_reg_apps;
5340 }
5341
5342 /* Create thread to manage application socket */
5343 ret = pthread_create(&apps_thread, NULL,
5344 thread_manage_apps, (void *) NULL);
5345 if (ret != 0) {
5346 PERROR("pthread_create apps");
5347 goto exit_apps;
5348 }
5349
5350 /* Create thread to manage application notify socket */
5351 ret = pthread_create(&apps_notify_thread, NULL,
5352 ust_thread_manage_notify, (void *) NULL);
5353 if (ret != 0) {
5354 PERROR("pthread_create notify");
5355 goto exit_apps_notify;
5356 }
5357
5358 /* Create agent registration thread. */
5359 ret = pthread_create(&agent_reg_thread, NULL,
5360 agent_thread_manage_registration, (void *) NULL);
5361 if (ret != 0) {
5362 PERROR("pthread_create agent");
5363 goto exit_agent_reg;
5364 }
5365
5366 /* Don't start this thread if kernel tracing is not requested nor root */
5367 if (is_root && !opt_no_kernel) {
5368 /* Create kernel thread to manage kernel event */
5369 ret = pthread_create(&kernel_thread, NULL,
5370 thread_manage_kernel, (void *) NULL);
5371 if (ret != 0) {
5372 PERROR("pthread_create kernel");
5373 goto exit_kernel;
5374 }
5375 }
5376
5377 /* Load possible session(s). */
5378 start_load_session_thread();
5379
5380 if (is_root && !opt_no_kernel) {
5381 ret = pthread_join(kernel_thread, &status);
5382 if (ret != 0) {
5383 PERROR("pthread_join");
5384 goto error; /* join error, exit without cleanup */
5385 }
5386 }
5387
5388 exit_kernel:
5389 ret = pthread_join(agent_reg_thread, &status);
5390 if (ret != 0) {
5391 PERROR("pthread_join agent");
5392 goto error; /* join error, exit without cleanup */
5393 }
5394
5395 exit_agent_reg:
5396 ret = pthread_join(apps_notify_thread, &status);
5397 if (ret != 0) {
5398 PERROR("pthread_join apps notify");
5399 goto error; /* join error, exit without cleanup */
5400 }
5401
5402 exit_apps_notify:
5403 ret = pthread_join(apps_thread, &status);
5404 if (ret != 0) {
5405 PERROR("pthread_join apps");
5406 goto error; /* join error, exit without cleanup */
5407 }
5408
5409
5410 exit_apps:
5411 ret = pthread_join(reg_apps_thread, &status);
5412 if (ret != 0) {
5413 PERROR("pthread_join");
5414 goto error; /* join error, exit without cleanup */
5415 }
5416
5417 exit_reg_apps:
5418 ret = pthread_join(dispatch_thread, &status);
5419 if (ret != 0) {
5420 PERROR("pthread_join");
5421 goto error; /* join error, exit without cleanup */
5422 }
5423
5424 exit_dispatch:
5425 ret = pthread_join(client_thread, &status);
5426 if (ret != 0) {
5427 PERROR("pthread_join");
5428 goto error; /* join error, exit without cleanup */
5429 }
5430
5431 ret = join_consumer_thread(&kconsumer_data);
5432 if (ret != 0) {
5433 PERROR("join_consumer");
5434 goto error; /* join error, exit without cleanup */
5435 }
5436
5437 ret = join_consumer_thread(&ustconsumer32_data);
5438 if (ret != 0) {
5439 PERROR("join_consumer ust32");
5440 goto error; /* join error, exit without cleanup */
5441 }
5442
5443 ret = join_consumer_thread(&ustconsumer64_data);
5444 if (ret != 0) {
5445 PERROR("join_consumer ust64");
5446 goto error; /* join error, exit without cleanup */
5447 }
5448
5449 exit_client:
5450 ret = pthread_join(health_thread, &status);
5451 if (ret != 0) {
5452 PERROR("pthread_join health thread");
5453 goto error; /* join error, exit without cleanup */
5454 }
5455
5456 exit_health:
5457 ret = pthread_join(ht_cleanup_thread, &status);
5458 if (ret != 0) {
5459 PERROR("pthread_join ht cleanup thread");
5460 goto error; /* join error, exit without cleanup */
5461 }
5462 exit_ht_cleanup:
5463 health_app_destroy(health_sessiond);
5464 exit_health_sessiond_cleanup:
5465 exit:
5466 /*
5467 * cleanup() is called when no other thread is running.
5468 */
5469 rcu_thread_online();
5470 cleanup();
5471 rcu_thread_offline();
5472 rcu_unregister_thread();
5473 if (!ret) {
5474 exit(EXIT_SUCCESS);
5475 }
5476 error:
5477 exit(EXIT_FAILURE);
5478 }
This page took 0.178856 seconds and 5 git commands to generate.