Fix: race between kconsumerd and sessiond on tear down
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _GNU_SOURCE
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <paths.h>
26 #include <pthread.h>
27 #include <signal.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <string.h>
31 #include <inttypes.h>
32 #include <sys/mman.h>
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
36 #include <sys/stat.h>
37 #include <sys/types.h>
38 #include <sys/wait.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <config.h>
42
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
53
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
56 #include "channel.h"
57 #include "cmd.h"
58 #include "consumer.h"
59 #include "context.h"
60 #include "event.h"
61 #include "kernel.h"
62 #include "kernel-consumer.h"
63 #include "modprobe.h"
64 #include "shm.h"
65 #include "ust-ctl.h"
66 #include "ust-consumer.h"
67 #include "utils.h"
68 #include "fd-limit.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
73 #include "save.h"
74 #include "load-session-thread.h"
75 #include "syscall.h"
76 #include "agent.h"
77
78 #define CONSUMERD_FILE "lttng-consumerd"
79
80 const char *progname;
81 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
82 static int tracing_group_name_override;
83 static char *opt_pidfile;
84 static int opt_sig_parent;
85 static int opt_verbose_consumer;
86 static int opt_daemon, opt_background;
87 static int opt_no_kernel;
88 static char *opt_load_session_path;
89 static pid_t ppid; /* Parent PID for --sig-parent option */
90 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
91 static char *rundir;
92 static int lockfile_fd = -1;
93
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal;
96
97 /*
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
100 */
101 static struct consumer_data kconsumer_data = {
102 .type = LTTNG_CONSUMER_KERNEL,
103 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
104 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
105 .err_sock = -1,
106 .cmd_sock = -1,
107 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
108 .lock = PTHREAD_MUTEX_INITIALIZER,
109 .cond = PTHREAD_COND_INITIALIZER,
110 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
111 };
112 static struct consumer_data ustconsumer64_data = {
113 .type = LTTNG_CONSUMER64_UST,
114 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
115 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
116 .err_sock = -1,
117 .cmd_sock = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123 static struct consumer_data ustconsumer32_data = {
124 .type = LTTNG_CONSUMER32_UST,
125 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
126 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
127 .err_sock = -1,
128 .cmd_sock = -1,
129 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
130 .lock = PTHREAD_MUTEX_INITIALIZER,
131 .cond = PTHREAD_COND_INITIALIZER,
132 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
133 };
134
135 /* Command line options */
136 static const struct option long_options[] = {
137 { "client-sock", required_argument, 0, 'c' },
138 { "apps-sock", required_argument, 0, 'a' },
139 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
140 { "kconsumerd-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
144 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
145 { "consumerd32-path", required_argument, 0, '\0' },
146 { "consumerd32-libdir", required_argument, 0, '\0' },
147 { "consumerd64-path", required_argument, 0, '\0' },
148 { "consumerd64-libdir", required_argument, 0, '\0' },
149 { "daemonize", no_argument, 0, 'd' },
150 { "background", no_argument, 0, 'b' },
151 { "sig-parent", no_argument, 0, 'S' },
152 { "help", no_argument, 0, 'h' },
153 { "group", required_argument, 0, 'g' },
154 { "version", no_argument, 0, 'V' },
155 { "quiet", no_argument, 0, 'q' },
156 { "verbose", no_argument, 0, 'v' },
157 { "verbose-consumer", no_argument, 0, '\0' },
158 { "no-kernel", no_argument, 0, '\0' },
159 { "pidfile", required_argument, 0, 'p' },
160 { "agent-tcp-port", required_argument, 0, '\0' },
161 { "config", required_argument, 0, 'f' },
162 { "load", required_argument, 0, 'l' },
163 { "kmod-probes", required_argument, 0, '\0' },
164 { "extra-kmod-probes", required_argument, 0, '\0' },
165 { NULL, 0, 0, 0 }
166 };
167
168 /* Command line options to ignore from configuration file */
169 static const char *config_ignore_options[] = { "help", "version", "config" };
170
171 /* Shared between threads */
172 static int dispatch_thread_exit;
173
174 /* Global application Unix socket path */
175 static char apps_unix_sock_path[PATH_MAX];
176 /* Global client Unix socket path */
177 static char client_unix_sock_path[PATH_MAX];
178 /* global wait shm path for UST */
179 static char wait_shm_path[PATH_MAX];
180 /* Global health check unix path */
181 static char health_unix_sock_path[PATH_MAX];
182
183 /* Sockets and FDs */
184 static int client_sock = -1;
185 static int apps_sock = -1;
186 int kernel_tracer_fd = -1;
187 static int kernel_poll_pipe[2] = { -1, -1 };
188
189 /*
190 * Quit pipe for all threads. This permits a single cancellation point
191 * for all threads when receiving an event on the pipe.
192 */
193 static int thread_quit_pipe[2] = { -1, -1 };
194 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
195
196 /*
197 * This pipe is used to inform the thread managing application communication
198 * that a command is queued and ready to be processed.
199 */
200 static int apps_cmd_pipe[2] = { -1, -1 };
201
202 int apps_cmd_notify_pipe[2] = { -1, -1 };
203
204 /* Pthread, Mutexes and Semaphores */
205 static pthread_t apps_thread;
206 static pthread_t apps_notify_thread;
207 static pthread_t reg_apps_thread;
208 static pthread_t client_thread;
209 static pthread_t kernel_thread;
210 static pthread_t dispatch_thread;
211 static pthread_t health_thread;
212 static pthread_t ht_cleanup_thread;
213 static pthread_t agent_reg_thread;
214 static pthread_t load_session_thread;
215
216 /*
217 * UST registration command queue. This queue is tied with a futex and uses a N
218 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 *
220 * The thread_registration_apps and thread_dispatch_ust_registration uses this
221 * queue along with the wait/wake scheme. The thread_manage_apps receives down
222 * the line new application socket and monitors it for any I/O error or clean
223 * close that triggers an unregistration of the application.
224 */
225 static struct ust_cmd_queue ust_cmd_queue;
226
227 /*
228 * Pointer initialized before thread creation.
229 *
230 * This points to the tracing session list containing the session count and a
231 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
232 * MUST NOT be taken if you call a public function in session.c.
233 *
234 * The lock is nested inside the structure: session_list_ptr->lock. Please use
235 * session_lock_list and session_unlock_list for lock acquisition.
236 */
237 static struct ltt_session_list *session_list_ptr;
238
239 int ust_consumerd64_fd = -1;
240 int ust_consumerd32_fd = -1;
241
242 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
243 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
244 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
245 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
246 static int consumerd32_bin_override;
247 static int consumerd64_bin_override;
248 static int consumerd32_libdir_override;
249 static int consumerd64_libdir_override;
250
251 static const char *module_proc_lttng = "/proc/lttng";
252
253 /*
254 * Consumer daemon state which is changed when spawning it, killing it or in
255 * case of a fatal error.
256 */
257 enum consumerd_state {
258 CONSUMER_STARTED = 1,
259 CONSUMER_STOPPED = 2,
260 CONSUMER_ERROR = 3,
261 };
262
263 /*
264 * This consumer daemon state is used to validate if a client command will be
265 * able to reach the consumer. If not, the client is informed. For instance,
266 * doing a "lttng start" when the consumer state is set to ERROR will return an
267 * error to the client.
268 *
269 * The following example shows a possible race condition of this scheme:
270 *
271 * consumer thread error happens
272 * client cmd arrives
273 * client cmd checks state -> still OK
274 * consumer thread exit, sets error
275 * client cmd try to talk to consumer
276 * ...
277 *
278 * However, since the consumer is a different daemon, we have no way of making
279 * sure the command will reach it safely even with this state flag. This is why
280 * we consider that up to the state validation during command processing, the
281 * command is safe. After that, we can not guarantee the correctness of the
282 * client request vis-a-vis the consumer.
283 */
284 static enum consumerd_state ust_consumerd_state;
285 static enum consumerd_state kernel_consumerd_state;
286
287 /*
288 * Socket timeout for receiving and sending in seconds.
289 */
290 static int app_socket_timeout;
291
292 /* Set in main() with the current page size. */
293 long page_size;
294
295 /* Application health monitoring */
296 struct health_app *health_sessiond;
297
298 /* Agent TCP port for registration. Used by the agent thread. */
299 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
300
301 /* Am I root or not. */
302 int is_root; /* Set to 1 if the daemon is running as root */
303
304 const char * const config_section_name = "sessiond";
305
306 /* Load session thread information to operate. */
307 struct load_session_thread_data *load_info;
308
309 /* Global hash tables */
310 struct lttng_ht *agent_apps_ht_by_sock = NULL;
311
312 /*
313 * Whether sessiond is ready for commands/health check requests.
314 * NR_LTTNG_SESSIOND_READY must match the number of calls to
315 * sessiond_notify_ready().
316 */
317 #define NR_LTTNG_SESSIOND_READY 3
318 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
319
320 /* Notify parents that we are ready for cmd and health check */
321 LTTNG_HIDDEN
322 void sessiond_notify_ready(void)
323 {
324 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
325 /*
326 * Notify parent pid that we are ready to accept command
327 * for client side. This ppid is the one from the
328 * external process that spawned us.
329 */
330 if (opt_sig_parent) {
331 kill(ppid, SIGUSR1);
332 }
333
334 /*
335 * Notify the parent of the fork() process that we are
336 * ready.
337 */
338 if (opt_daemon || opt_background) {
339 kill(child_ppid, SIGUSR1);
340 }
341 }
342 }
343
344 static
345 void setup_consumerd_path(void)
346 {
347 const char *bin, *libdir;
348
349 /*
350 * Allow INSTALL_BIN_PATH to be used as a target path for the
351 * native architecture size consumer if CONFIG_CONSUMER*_PATH
352 * has not been defined.
353 */
354 #if (CAA_BITS_PER_LONG == 32)
355 if (!consumerd32_bin[0]) {
356 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
357 }
358 if (!consumerd32_libdir[0]) {
359 consumerd32_libdir = INSTALL_LIB_PATH;
360 }
361 #elif (CAA_BITS_PER_LONG == 64)
362 if (!consumerd64_bin[0]) {
363 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
364 }
365 if (!consumerd64_libdir[0]) {
366 consumerd64_libdir = INSTALL_LIB_PATH;
367 }
368 #else
369 #error "Unknown bitness"
370 #endif
371
372 /*
373 * runtime env. var. overrides the build default.
374 */
375 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
376 if (bin) {
377 consumerd32_bin = bin;
378 }
379 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
380 if (bin) {
381 consumerd64_bin = bin;
382 }
383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
384 if (libdir) {
385 consumerd32_libdir = libdir;
386 }
387 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
388 if (libdir) {
389 consumerd64_libdir = libdir;
390 }
391 }
392
393 static
394 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
395 int *a_pipe)
396 {
397 int ret;
398
399 assert(events);
400
401 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
402 if (ret < 0) {
403 goto error;
404 }
405
406 /* Add quit pipe */
407 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
408 if (ret < 0) {
409 goto error;
410 }
411
412 return 0;
413
414 error:
415 return ret;
416 }
417
418 /*
419 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
420 */
421 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
422 {
423 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
424 }
425
426 /*
427 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
428 */
429 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
430 size_t size)
431 {
432 return __sessiond_set_thread_pollset(events, size,
433 ht_cleanup_quit_pipe);
434 }
435
436 static
437 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
438 {
439 if (fd == a_pipe && (events & LPOLLIN)) {
440 return 1;
441 }
442 return 0;
443 }
444
445 /*
446 * Check if the thread quit pipe was triggered.
447 *
448 * Return 1 if it was triggered else 0;
449 */
450 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
451 {
452 return __sessiond_check_thread_quit_pipe(fd, events,
453 thread_quit_pipe[0]);
454 }
455
456 /*
457 * Check if the ht_cleanup thread quit pipe was triggered.
458 *
459 * Return 1 if it was triggered else 0;
460 */
461 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
462 {
463 return __sessiond_check_thread_quit_pipe(fd, events,
464 ht_cleanup_quit_pipe[0]);
465 }
466
467 /*
468 * Init thread quit pipe.
469 *
470 * Return -1 on error or 0 if all pipes are created.
471 */
472 static int __init_thread_quit_pipe(int *a_pipe)
473 {
474 int ret, i;
475
476 ret = pipe(a_pipe);
477 if (ret < 0) {
478 PERROR("thread quit pipe");
479 goto error;
480 }
481
482 for (i = 0; i < 2; i++) {
483 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
484 if (ret < 0) {
485 PERROR("fcntl");
486 goto error;
487 }
488 }
489
490 error:
491 return ret;
492 }
493
494 static int init_thread_quit_pipe(void)
495 {
496 return __init_thread_quit_pipe(thread_quit_pipe);
497 }
498
499 static int init_ht_cleanup_quit_pipe(void)
500 {
501 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
502 }
503
504 /*
505 * Stop all threads by closing the thread quit pipe.
506 */
507 static void stop_threads(void)
508 {
509 int ret;
510
511 /* Stopping all threads */
512 DBG("Terminating all threads");
513 ret = notify_thread_pipe(thread_quit_pipe[1]);
514 if (ret < 0) {
515 ERR("write error on thread quit pipe");
516 }
517
518 /* Dispatch thread */
519 CMM_STORE_SHARED(dispatch_thread_exit, 1);
520 futex_nto1_wake(&ust_cmd_queue.futex);
521 }
522
523 /*
524 * Close every consumer sockets.
525 */
526 static void close_consumer_sockets(void)
527 {
528 int ret;
529
530 if (kconsumer_data.err_sock >= 0) {
531 ret = close(kconsumer_data.err_sock);
532 if (ret < 0) {
533 PERROR("kernel consumer err_sock close");
534 }
535 }
536 if (ustconsumer32_data.err_sock >= 0) {
537 ret = close(ustconsumer32_data.err_sock);
538 if (ret < 0) {
539 PERROR("UST consumerd32 err_sock close");
540 }
541 }
542 if (ustconsumer64_data.err_sock >= 0) {
543 ret = close(ustconsumer64_data.err_sock);
544 if (ret < 0) {
545 PERROR("UST consumerd64 err_sock close");
546 }
547 }
548 if (kconsumer_data.cmd_sock >= 0) {
549 ret = close(kconsumer_data.cmd_sock);
550 if (ret < 0) {
551 PERROR("kernel consumer cmd_sock close");
552 }
553 }
554 if (ustconsumer32_data.cmd_sock >= 0) {
555 ret = close(ustconsumer32_data.cmd_sock);
556 if (ret < 0) {
557 PERROR("UST consumerd32 cmd_sock close");
558 }
559 }
560 if (ustconsumer64_data.cmd_sock >= 0) {
561 ret = close(ustconsumer64_data.cmd_sock);
562 if (ret < 0) {
563 PERROR("UST consumerd64 cmd_sock close");
564 }
565 }
566 }
567
568 /*
569 * Generate the full lock file path using the rundir.
570 *
571 * Return the snprintf() return value thus a negative value is an error.
572 */
573 static int generate_lock_file_path(char *path, size_t len)
574 {
575 int ret;
576
577 assert(path);
578 assert(rundir);
579
580 /* Build lockfile path from rundir. */
581 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
582 if (ret < 0) {
583 PERROR("snprintf lockfile path");
584 }
585
586 return ret;
587 }
588
589 /*
590 * Wait on consumer process termination.
591 *
592 * Need to be called with the consumer data lock held or from a context
593 * ensuring no concurrent access to data (e.g: cleanup).
594 */
595 static void wait_consumer(struct consumer_data *consumer_data)
596 {
597 pid_t ret;
598 int status;
599
600 if (consumer_data->pid <= 0) {
601 return;
602 }
603
604 DBG("Waiting for complete teardown of consumerd (PID: %d)",
605 consumer_data->pid);
606 ret = waitpid(consumer_data->pid, &status, 0);
607 if (ret == -1) {
608 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
609 }
610 if (!WIFEXITED(status)) {
611 ERR("consumerd termination with error: %d",
612 WEXITSTATUS(ret));
613 }
614 consumer_data->pid = 0;
615 }
616
617 /*
618 * Cleanup the session daemon's data structures.
619 */
620 static void sessiond_cleanup(void)
621 {
622 int ret;
623 struct ltt_session *sess, *stmp;
624 char path[PATH_MAX];
625
626 DBG("Cleanup sessiond");
627
628 /*
629 * Close the thread quit pipe. It has already done its job,
630 * since we are now called.
631 */
632 utils_close_pipe(thread_quit_pipe);
633
634 /*
635 * If opt_pidfile is undefined, the default file will be wiped when
636 * removing the rundir.
637 */
638 if (opt_pidfile) {
639 ret = remove(opt_pidfile);
640 if (ret < 0) {
641 PERROR("remove pidfile %s", opt_pidfile);
642 }
643 }
644
645 DBG("Removing sessiond and consumerd content of directory %s", rundir);
646
647 /* sessiond */
648 snprintf(path, PATH_MAX,
649 "%s/%s",
650 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
651 DBG("Removing %s", path);
652 (void) unlink(path);
653
654 snprintf(path, PATH_MAX, "%s/%s", rundir,
655 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
656 DBG("Removing %s", path);
657 (void) unlink(path);
658
659 /* kconsumerd */
660 snprintf(path, PATH_MAX,
661 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
662 rundir);
663 DBG("Removing %s", path);
664 (void) unlink(path);
665
666 snprintf(path, PATH_MAX,
667 DEFAULT_KCONSUMERD_PATH,
668 rundir);
669 DBG("Removing directory %s", path);
670 (void) rmdir(path);
671
672 /* ust consumerd 32 */
673 snprintf(path, PATH_MAX,
674 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
675 rundir);
676 DBG("Removing %s", path);
677 (void) unlink(path);
678
679 snprintf(path, PATH_MAX,
680 DEFAULT_USTCONSUMERD32_PATH,
681 rundir);
682 DBG("Removing directory %s", path);
683 (void) rmdir(path);
684
685 /* ust consumerd 64 */
686 snprintf(path, PATH_MAX,
687 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
688 rundir);
689 DBG("Removing %s", path);
690 (void) unlink(path);
691
692 snprintf(path, PATH_MAX,
693 DEFAULT_USTCONSUMERD64_PATH,
694 rundir);
695 DBG("Removing directory %s", path);
696 (void) rmdir(path);
697
698 DBG("Cleaning up all sessions");
699
700 /* Destroy session list mutex */
701 if (session_list_ptr != NULL) {
702 pthread_mutex_destroy(&session_list_ptr->lock);
703
704 /* Cleanup ALL session */
705 cds_list_for_each_entry_safe(sess, stmp,
706 &session_list_ptr->head, list) {
707 cmd_destroy_session(sess, kernel_poll_pipe[1]);
708 }
709 }
710
711 wait_consumer(&kconsumer_data);
712 wait_consumer(&ustconsumer64_data);
713 wait_consumer(&ustconsumer32_data);
714
715
716 DBG("Cleaning up all agent apps");
717 agent_app_ht_clean();
718
719 DBG("Closing all UST sockets");
720 ust_app_clean_list();
721 buffer_reg_destroy_registries();
722
723 if (is_root && !opt_no_kernel) {
724 DBG2("Closing kernel fd");
725 if (kernel_tracer_fd >= 0) {
726 ret = close(kernel_tracer_fd);
727 if (ret) {
728 PERROR("close");
729 }
730 }
731 DBG("Unloading kernel modules");
732 modprobe_remove_lttng_all();
733 free(syscall_table);
734 }
735
736 close_consumer_sockets();
737
738 if (load_info) {
739 load_session_destroy_data(load_info);
740 free(load_info);
741 }
742
743 /*
744 * Cleanup lock file by deleting it and finaly closing it which will
745 * release the file system lock.
746 */
747 if (lockfile_fd >= 0) {
748 char lockfile_path[PATH_MAX];
749
750 ret = generate_lock_file_path(lockfile_path,
751 sizeof(lockfile_path));
752 if (ret > 0) {
753 ret = remove(lockfile_path);
754 if (ret < 0) {
755 PERROR("remove lock file");
756 }
757 ret = close(lockfile_fd);
758 if (ret < 0) {
759 PERROR("close lock file");
760 }
761 }
762 }
763
764 /*
765 * We do NOT rmdir rundir because there are other processes
766 * using it, for instance lttng-relayd, which can start in
767 * parallel with this teardown.
768 */
769
770 free(rundir);
771 }
772
773 /*
774 * Cleanup the daemon's option data structures.
775 */
776 static void sessiond_cleanup_options(void)
777 {
778 DBG("Cleaning up options");
779
780 /*
781 * If the override option is set, the pointer points to a *non* const
782 * thus freeing it even though the variable type is set to const.
783 */
784 if (tracing_group_name_override) {
785 free((void *) tracing_group_name);
786 }
787 if (consumerd32_bin_override) {
788 free((void *) consumerd32_bin);
789 }
790 if (consumerd64_bin_override) {
791 free((void *) consumerd64_bin);
792 }
793 if (consumerd32_libdir_override) {
794 free((void *) consumerd32_libdir);
795 }
796 if (consumerd64_libdir_override) {
797 free((void *) consumerd64_libdir);
798 }
799
800 free(opt_pidfile);
801 free(opt_load_session_path);
802 free(kmod_probes_list);
803 free(kmod_extra_probes_list);
804
805 /* <fun> */
806 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
807 "Matthew, BEET driven development works!%c[%dm",
808 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
809 /* </fun> */
810 }
811
812 /*
813 * Send data on a unix socket using the liblttsessiondcomm API.
814 *
815 * Return lttcomm error code.
816 */
817 static int send_unix_sock(int sock, void *buf, size_t len)
818 {
819 /* Check valid length */
820 if (len == 0) {
821 return -1;
822 }
823
824 return lttcomm_send_unix_sock(sock, buf, len);
825 }
826
827 /*
828 * Free memory of a command context structure.
829 */
830 static void clean_command_ctx(struct command_ctx **cmd_ctx)
831 {
832 DBG("Clean command context structure");
833 if (*cmd_ctx) {
834 if ((*cmd_ctx)->llm) {
835 free((*cmd_ctx)->llm);
836 }
837 if ((*cmd_ctx)->lsm) {
838 free((*cmd_ctx)->lsm);
839 }
840 free(*cmd_ctx);
841 *cmd_ctx = NULL;
842 }
843 }
844
845 /*
846 * Notify UST applications using the shm mmap futex.
847 */
848 static int notify_ust_apps(int active)
849 {
850 char *wait_shm_mmap;
851
852 DBG("Notifying applications of session daemon state: %d", active);
853
854 /* See shm.c for this call implying mmap, shm and futex calls */
855 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
856 if (wait_shm_mmap == NULL) {
857 goto error;
858 }
859
860 /* Wake waiting process */
861 futex_wait_update((int32_t *) wait_shm_mmap, active);
862
863 /* Apps notified successfully */
864 return 0;
865
866 error:
867 return -1;
868 }
869
870 /*
871 * Setup the outgoing data buffer for the response (llm) by allocating the
872 * right amount of memory and copying the original information from the lsm
873 * structure.
874 *
875 * Return total size of the buffer pointed by buf.
876 */
877 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
878 {
879 int ret, buf_size;
880
881 buf_size = size;
882
883 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
884 if (cmd_ctx->llm == NULL) {
885 PERROR("zmalloc");
886 ret = -ENOMEM;
887 goto error;
888 }
889
890 /* Copy common data */
891 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
892 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
893
894 cmd_ctx->llm->data_size = size;
895 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
896
897 return buf_size;
898
899 error:
900 return ret;
901 }
902
903 /*
904 * Update the kernel poll set of all channel fd available over all tracing
905 * session. Add the wakeup pipe at the end of the set.
906 */
907 static int update_kernel_poll(struct lttng_poll_event *events)
908 {
909 int ret;
910 struct ltt_session *session;
911 struct ltt_kernel_channel *channel;
912
913 DBG("Updating kernel poll set");
914
915 session_lock_list();
916 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
917 session_lock(session);
918 if (session->kernel_session == NULL) {
919 session_unlock(session);
920 continue;
921 }
922
923 cds_list_for_each_entry(channel,
924 &session->kernel_session->channel_list.head, list) {
925 /* Add channel fd to the kernel poll set */
926 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
927 if (ret < 0) {
928 session_unlock(session);
929 goto error;
930 }
931 DBG("Channel fd %d added to kernel set", channel->fd);
932 }
933 session_unlock(session);
934 }
935 session_unlock_list();
936
937 return 0;
938
939 error:
940 session_unlock_list();
941 return -1;
942 }
943
944 /*
945 * Find the channel fd from 'fd' over all tracing session. When found, check
946 * for new channel stream and send those stream fds to the kernel consumer.
947 *
948 * Useful for CPU hotplug feature.
949 */
950 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
951 {
952 int ret = 0;
953 struct ltt_session *session;
954 struct ltt_kernel_session *ksess;
955 struct ltt_kernel_channel *channel;
956
957 DBG("Updating kernel streams for channel fd %d", fd);
958
959 session_lock_list();
960 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
961 session_lock(session);
962 if (session->kernel_session == NULL) {
963 session_unlock(session);
964 continue;
965 }
966 ksess = session->kernel_session;
967
968 cds_list_for_each_entry(channel,
969 &ksess->channel_list.head, list) {
970 struct lttng_ht_iter iter;
971 struct consumer_socket *socket;
972
973 if (channel->fd != fd) {
974 continue;
975 }
976 DBG("Channel found, updating kernel streams");
977 ret = kernel_open_channel_stream(channel);
978 if (ret < 0) {
979 goto error;
980 }
981 /* Update the stream global counter */
982 ksess->stream_count_global += ret;
983
984 /*
985 * Have we already sent fds to the consumer? If yes, it
986 * means that tracing is started so it is safe to send
987 * our updated stream fds.
988 */
989 if (ksess->consumer_fds_sent != 1
990 || ksess->consumer == NULL) {
991 ret = -1;
992 goto error;
993 }
994
995 rcu_read_lock();
996 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
997 &iter.iter, socket, node.node) {
998 pthread_mutex_lock(socket->lock);
999 ret = kernel_consumer_send_channel_stream(socket,
1000 channel, ksess,
1001 session->output_traces ? 1 : 0);
1002 pthread_mutex_unlock(socket->lock);
1003 if (ret < 0) {
1004 rcu_read_unlock();
1005 goto error;
1006 }
1007 }
1008 rcu_read_unlock();
1009 }
1010 session_unlock(session);
1011 }
1012 session_unlock_list();
1013 return ret;
1014
1015 error:
1016 session_unlock(session);
1017 session_unlock_list();
1018 return ret;
1019 }
1020
1021 /*
1022 * For each tracing session, update newly registered apps. The session list
1023 * lock MUST be acquired before calling this.
1024 */
1025 static void update_ust_app(int app_sock)
1026 {
1027 struct ltt_session *sess, *stmp;
1028
1029 /* Consumer is in an ERROR state. Stop any application update. */
1030 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1031 /* Stop the update process since the consumer is dead. */
1032 return;
1033 }
1034
1035 /* For all tracing session(s) */
1036 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1037 struct ust_app *app;
1038
1039 session_lock(sess);
1040 if (!sess->ust_session) {
1041 goto unlock_session;
1042 }
1043
1044 rcu_read_lock();
1045 assert(app_sock >= 0);
1046 app = ust_app_find_by_sock(app_sock);
1047 if (app == NULL) {
1048 /*
1049 * Application can be unregistered before so
1050 * this is possible hence simply stopping the
1051 * update.
1052 */
1053 DBG3("UST app update failed to find app sock %d",
1054 app_sock);
1055 goto unlock_rcu;
1056 }
1057 ust_app_global_update(sess->ust_session, app);
1058 unlock_rcu:
1059 rcu_read_unlock();
1060 unlock_session:
1061 session_unlock(sess);
1062 }
1063 }
1064
1065 /*
1066 * This thread manage event coming from the kernel.
1067 *
1068 * Features supported in this thread:
1069 * -) CPU Hotplug
1070 */
1071 static void *thread_manage_kernel(void *data)
1072 {
1073 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1074 uint32_t revents, nb_fd;
1075 char tmp;
1076 struct lttng_poll_event events;
1077
1078 DBG("[thread] Thread manage kernel started");
1079
1080 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1081
1082 /*
1083 * This first step of the while is to clean this structure which could free
1084 * non NULL pointers so initialize it before the loop.
1085 */
1086 lttng_poll_init(&events);
1087
1088 if (testpoint(sessiond_thread_manage_kernel)) {
1089 goto error_testpoint;
1090 }
1091
1092 health_code_update();
1093
1094 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1095 goto error_testpoint;
1096 }
1097
1098 while (1) {
1099 health_code_update();
1100
1101 if (update_poll_flag == 1) {
1102 /* Clean events object. We are about to populate it again. */
1103 lttng_poll_clean(&events);
1104
1105 ret = sessiond_set_thread_pollset(&events, 2);
1106 if (ret < 0) {
1107 goto error_poll_create;
1108 }
1109
1110 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1111 if (ret < 0) {
1112 goto error;
1113 }
1114
1115 /* This will add the available kernel channel if any. */
1116 ret = update_kernel_poll(&events);
1117 if (ret < 0) {
1118 goto error;
1119 }
1120 update_poll_flag = 0;
1121 }
1122
1123 DBG("Thread kernel polling");
1124
1125 /* Poll infinite value of time */
1126 restart:
1127 health_poll_entry();
1128 ret = lttng_poll_wait(&events, -1);
1129 DBG("Thread kernel return from poll on %d fds",
1130 LTTNG_POLL_GETNB(&events));
1131 health_poll_exit();
1132 if (ret < 0) {
1133 /*
1134 * Restart interrupted system call.
1135 */
1136 if (errno == EINTR) {
1137 goto restart;
1138 }
1139 goto error;
1140 } else if (ret == 0) {
1141 /* Should not happen since timeout is infinite */
1142 ERR("Return value of poll is 0 with an infinite timeout.\n"
1143 "This should not have happened! Continuing...");
1144 continue;
1145 }
1146
1147 nb_fd = ret;
1148
1149 for (i = 0; i < nb_fd; i++) {
1150 /* Fetch once the poll data */
1151 revents = LTTNG_POLL_GETEV(&events, i);
1152 pollfd = LTTNG_POLL_GETFD(&events, i);
1153
1154 health_code_update();
1155
1156 if (!revents) {
1157 /* No activity for this FD (poll implementation). */
1158 continue;
1159 }
1160
1161 /* Thread quit pipe has been closed. Killing thread. */
1162 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1163 if (ret) {
1164 err = 0;
1165 goto exit;
1166 }
1167
1168 /* Check for data on kernel pipe */
1169 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1170 (void) lttng_read(kernel_poll_pipe[0],
1171 &tmp, 1);
1172 /*
1173 * Ret value is useless here, if this pipe gets any actions an
1174 * update is required anyway.
1175 */
1176 update_poll_flag = 1;
1177 continue;
1178 } else {
1179 /*
1180 * New CPU detected by the kernel. Adding kernel stream to
1181 * kernel session and updating the kernel consumer
1182 */
1183 if (revents & LPOLLIN) {
1184 ret = update_kernel_stream(&kconsumer_data, pollfd);
1185 if (ret < 0) {
1186 continue;
1187 }
1188 break;
1189 /*
1190 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1191 * and unregister kernel stream at this point.
1192 */
1193 }
1194 }
1195 }
1196 }
1197
1198 exit:
1199 error:
1200 lttng_poll_clean(&events);
1201 error_poll_create:
1202 error_testpoint:
1203 utils_close_pipe(kernel_poll_pipe);
1204 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1205 if (err) {
1206 health_error();
1207 ERR("Health error occurred in %s", __func__);
1208 WARN("Kernel thread died unexpectedly. "
1209 "Kernel tracing can continue but CPU hotplug is disabled.");
1210 }
1211 health_unregister(health_sessiond);
1212 DBG("Kernel thread dying");
1213 return NULL;
1214 }
1215
1216 /*
1217 * Signal pthread condition of the consumer data that the thread.
1218 */
1219 static void signal_consumer_condition(struct consumer_data *data, int state)
1220 {
1221 pthread_mutex_lock(&data->cond_mutex);
1222
1223 /*
1224 * The state is set before signaling. It can be any value, it's the waiter
1225 * job to correctly interpret this condition variable associated to the
1226 * consumer pthread_cond.
1227 *
1228 * A value of 0 means that the corresponding thread of the consumer data
1229 * was not started. 1 indicates that the thread has started and is ready
1230 * for action. A negative value means that there was an error during the
1231 * thread bootstrap.
1232 */
1233 data->consumer_thread_is_ready = state;
1234 (void) pthread_cond_signal(&data->cond);
1235
1236 pthread_mutex_unlock(&data->cond_mutex);
1237 }
1238
1239 /*
1240 * This thread manage the consumer error sent back to the session daemon.
1241 */
1242 static void *thread_manage_consumer(void *data)
1243 {
1244 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1245 uint32_t revents, nb_fd;
1246 enum lttcomm_return_code code;
1247 struct lttng_poll_event events;
1248 struct consumer_data *consumer_data = data;
1249
1250 DBG("[thread] Manage consumer started");
1251
1252 rcu_register_thread();
1253 rcu_thread_online();
1254
1255 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1256
1257 health_code_update();
1258
1259 /*
1260 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1261 * metadata_sock. Nothing more will be added to this poll set.
1262 */
1263 ret = sessiond_set_thread_pollset(&events, 3);
1264 if (ret < 0) {
1265 goto error_poll;
1266 }
1267
1268 /*
1269 * The error socket here is already in a listening state which was done
1270 * just before spawning this thread to avoid a race between the consumer
1271 * daemon exec trying to connect and the listen() call.
1272 */
1273 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1274 if (ret < 0) {
1275 goto error;
1276 }
1277
1278 health_code_update();
1279
1280 /* Infinite blocking call, waiting for transmission */
1281 restart:
1282 health_poll_entry();
1283
1284 if (testpoint(sessiond_thread_manage_consumer)) {
1285 goto error;
1286 }
1287
1288 ret = lttng_poll_wait(&events, -1);
1289 health_poll_exit();
1290 if (ret < 0) {
1291 /*
1292 * Restart interrupted system call.
1293 */
1294 if (errno == EINTR) {
1295 goto restart;
1296 }
1297 goto error;
1298 }
1299
1300 nb_fd = ret;
1301
1302 for (i = 0; i < nb_fd; i++) {
1303 /* Fetch once the poll data */
1304 revents = LTTNG_POLL_GETEV(&events, i);
1305 pollfd = LTTNG_POLL_GETFD(&events, i);
1306
1307 health_code_update();
1308
1309 if (!revents) {
1310 /* No activity for this FD (poll implementation). */
1311 continue;
1312 }
1313
1314 /* Thread quit pipe has been closed. Killing thread. */
1315 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1316 if (ret) {
1317 err = 0;
1318 goto exit;
1319 }
1320
1321 /* Event on the registration socket */
1322 if (pollfd == consumer_data->err_sock) {
1323 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1324 ERR("consumer err socket poll error");
1325 goto error;
1326 }
1327 }
1328 }
1329
1330 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1331 if (sock < 0) {
1332 goto error;
1333 }
1334
1335 /*
1336 * Set the CLOEXEC flag. Return code is useless because either way, the
1337 * show must go on.
1338 */
1339 (void) utils_set_fd_cloexec(sock);
1340
1341 health_code_update();
1342
1343 DBG2("Receiving code from consumer err_sock");
1344
1345 /* Getting status code from kconsumerd */
1346 ret = lttcomm_recv_unix_sock(sock, &code,
1347 sizeof(enum lttcomm_return_code));
1348 if (ret <= 0) {
1349 goto error;
1350 }
1351
1352 health_code_update();
1353 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1354 /* Connect both socket, command and metadata. */
1355 consumer_data->cmd_sock =
1356 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1357 consumer_data->metadata_fd =
1358 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1359 if (consumer_data->cmd_sock < 0
1360 || consumer_data->metadata_fd < 0) {
1361 PERROR("consumer connect cmd socket");
1362 /* On error, signal condition and quit. */
1363 signal_consumer_condition(consumer_data, -1);
1364 goto error;
1365 }
1366 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1367 /* Create metadata socket lock. */
1368 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1369 if (consumer_data->metadata_sock.lock == NULL) {
1370 PERROR("zmalloc pthread mutex");
1371 ret = -1;
1372 goto error;
1373 }
1374 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1375
1376 signal_consumer_condition(consumer_data, 1);
1377 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1378 DBG("Consumer metadata socket ready (fd: %d)",
1379 consumer_data->metadata_fd);
1380 } else {
1381 ERR("consumer error when waiting for SOCK_READY : %s",
1382 lttcomm_get_readable_code(-code));
1383 goto error;
1384 }
1385
1386 /* Remove the consumerd error sock since we've established a connexion */
1387 ret = lttng_poll_del(&events, consumer_data->err_sock);
1388 if (ret < 0) {
1389 goto error;
1390 }
1391
1392 /* Add new accepted error socket. */
1393 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1394 if (ret < 0) {
1395 goto error;
1396 }
1397
1398 /* Add metadata socket that is successfully connected. */
1399 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1400 LPOLLIN | LPOLLRDHUP);
1401 if (ret < 0) {
1402 goto error;
1403 }
1404
1405 health_code_update();
1406
1407 /* Infinite blocking call, waiting for transmission */
1408 restart_poll:
1409 while (1) {
1410 health_code_update();
1411
1412 /* Exit the thread because the thread quit pipe has been triggered. */
1413 if (should_quit) {
1414 /* Not a health error. */
1415 err = 0;
1416 goto exit;
1417 }
1418
1419 health_poll_entry();
1420 ret = lttng_poll_wait(&events, -1);
1421 health_poll_exit();
1422 if (ret < 0) {
1423 /*
1424 * Restart interrupted system call.
1425 */
1426 if (errno == EINTR) {
1427 goto restart_poll;
1428 }
1429 goto error;
1430 }
1431
1432 nb_fd = ret;
1433
1434 for (i = 0; i < nb_fd; i++) {
1435 /* Fetch once the poll data */
1436 revents = LTTNG_POLL_GETEV(&events, i);
1437 pollfd = LTTNG_POLL_GETFD(&events, i);
1438
1439 health_code_update();
1440
1441 if (!revents) {
1442 /* No activity for this FD (poll implementation). */
1443 continue;
1444 }
1445
1446 /*
1447 * Thread quit pipe has been triggered, flag that we should stop
1448 * but continue the current loop to handle potential data from
1449 * consumer.
1450 */
1451 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1452
1453 if (pollfd == sock) {
1454 /* Event on the consumerd socket */
1455 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1456 ERR("consumer err socket second poll error");
1457 goto error;
1458 }
1459 health_code_update();
1460 /* Wait for any kconsumerd error */
1461 ret = lttcomm_recv_unix_sock(sock, &code,
1462 sizeof(enum lttcomm_return_code));
1463 if (ret <= 0) {
1464 ERR("consumer closed the command socket");
1465 goto error;
1466 }
1467
1468 ERR("consumer return code : %s",
1469 lttcomm_get_readable_code(-code));
1470
1471 goto exit;
1472 } else if (pollfd == consumer_data->metadata_fd) {
1473 /* UST metadata requests */
1474 ret = ust_consumer_metadata_request(
1475 &consumer_data->metadata_sock);
1476 if (ret < 0) {
1477 ERR("Handling metadata request");
1478 goto error;
1479 }
1480 }
1481 /* No need for an else branch all FDs are tested prior. */
1482 }
1483 health_code_update();
1484 }
1485
1486 exit:
1487 error:
1488 /*
1489 * We lock here because we are about to close the sockets and some other
1490 * thread might be using them so get exclusive access which will abort all
1491 * other consumer command by other threads.
1492 */
1493 pthread_mutex_lock(&consumer_data->lock);
1494
1495 /* Immediately set the consumerd state to stopped */
1496 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1497 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1498 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1499 consumer_data->type == LTTNG_CONSUMER32_UST) {
1500 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1501 } else {
1502 /* Code flow error... */
1503 assert(0);
1504 }
1505
1506 if (consumer_data->err_sock >= 0) {
1507 ret = close(consumer_data->err_sock);
1508 if (ret) {
1509 PERROR("close");
1510 }
1511 consumer_data->err_sock = -1;
1512 }
1513 if (consumer_data->cmd_sock >= 0) {
1514 ret = close(consumer_data->cmd_sock);
1515 if (ret) {
1516 PERROR("close");
1517 }
1518 consumer_data->cmd_sock = -1;
1519 }
1520 if (consumer_data->metadata_sock.fd_ptr &&
1521 *consumer_data->metadata_sock.fd_ptr >= 0) {
1522 ret = close(*consumer_data->metadata_sock.fd_ptr);
1523 if (ret) {
1524 PERROR("close");
1525 }
1526 }
1527 if (sock >= 0) {
1528 ret = close(sock);
1529 if (ret) {
1530 PERROR("close");
1531 }
1532 }
1533
1534 unlink(consumer_data->err_unix_sock_path);
1535 unlink(consumer_data->cmd_unix_sock_path);
1536 pthread_mutex_unlock(&consumer_data->lock);
1537
1538 /* Cleanup metadata socket mutex. */
1539 if (consumer_data->metadata_sock.lock) {
1540 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1541 free(consumer_data->metadata_sock.lock);
1542 }
1543 lttng_poll_clean(&events);
1544 error_poll:
1545 if (err) {
1546 health_error();
1547 ERR("Health error occurred in %s", __func__);
1548 }
1549 health_unregister(health_sessiond);
1550 DBG("consumer thread cleanup completed");
1551
1552 rcu_thread_offline();
1553 rcu_unregister_thread();
1554
1555 return NULL;
1556 }
1557
1558 /*
1559 * This thread manage application communication.
1560 */
1561 static void *thread_manage_apps(void *data)
1562 {
1563 int i, ret, pollfd, err = -1;
1564 ssize_t size_ret;
1565 uint32_t revents, nb_fd;
1566 struct lttng_poll_event events;
1567
1568 DBG("[thread] Manage application started");
1569
1570 rcu_register_thread();
1571 rcu_thread_online();
1572
1573 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1574
1575 if (testpoint(sessiond_thread_manage_apps)) {
1576 goto error_testpoint;
1577 }
1578
1579 health_code_update();
1580
1581 ret = sessiond_set_thread_pollset(&events, 2);
1582 if (ret < 0) {
1583 goto error_poll_create;
1584 }
1585
1586 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1587 if (ret < 0) {
1588 goto error;
1589 }
1590
1591 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1592 goto error;
1593 }
1594
1595 health_code_update();
1596
1597 while (1) {
1598 DBG("Apps thread polling");
1599
1600 /* Inifinite blocking call, waiting for transmission */
1601 restart:
1602 health_poll_entry();
1603 ret = lttng_poll_wait(&events, -1);
1604 DBG("Apps thread return from poll on %d fds",
1605 LTTNG_POLL_GETNB(&events));
1606 health_poll_exit();
1607 if (ret < 0) {
1608 /*
1609 * Restart interrupted system call.
1610 */
1611 if (errno == EINTR) {
1612 goto restart;
1613 }
1614 goto error;
1615 }
1616
1617 nb_fd = ret;
1618
1619 for (i = 0; i < nb_fd; i++) {
1620 /* Fetch once the poll data */
1621 revents = LTTNG_POLL_GETEV(&events, i);
1622 pollfd = LTTNG_POLL_GETFD(&events, i);
1623
1624 health_code_update();
1625
1626 if (!revents) {
1627 /* No activity for this FD (poll implementation). */
1628 continue;
1629 }
1630
1631 /* Thread quit pipe has been closed. Killing thread. */
1632 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1633 if (ret) {
1634 err = 0;
1635 goto exit;
1636 }
1637
1638 /* Inspect the apps cmd pipe */
1639 if (pollfd == apps_cmd_pipe[0]) {
1640 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1641 ERR("Apps command pipe error");
1642 goto error;
1643 } else if (revents & LPOLLIN) {
1644 int sock;
1645
1646 /* Empty pipe */
1647 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1648 if (size_ret < sizeof(sock)) {
1649 PERROR("read apps cmd pipe");
1650 goto error;
1651 }
1652
1653 health_code_update();
1654
1655 /*
1656 * We only monitor the error events of the socket. This
1657 * thread does not handle any incoming data from UST
1658 * (POLLIN).
1659 */
1660 ret = lttng_poll_add(&events, sock,
1661 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1662 if (ret < 0) {
1663 goto error;
1664 }
1665
1666 DBG("Apps with sock %d added to poll set", sock);
1667 }
1668 } else {
1669 /*
1670 * At this point, we know that a registered application made
1671 * the event at poll_wait.
1672 */
1673 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1674 /* Removing from the poll set */
1675 ret = lttng_poll_del(&events, pollfd);
1676 if (ret < 0) {
1677 goto error;
1678 }
1679
1680 /* Socket closed on remote end. */
1681 ust_app_unregister(pollfd);
1682 }
1683 }
1684
1685 health_code_update();
1686 }
1687 }
1688
1689 exit:
1690 error:
1691 lttng_poll_clean(&events);
1692 error_poll_create:
1693 error_testpoint:
1694 utils_close_pipe(apps_cmd_pipe);
1695 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1696
1697 /*
1698 * We don't clean the UST app hash table here since already registered
1699 * applications can still be controlled so let them be until the session
1700 * daemon dies or the applications stop.
1701 */
1702
1703 if (err) {
1704 health_error();
1705 ERR("Health error occurred in %s", __func__);
1706 }
1707 health_unregister(health_sessiond);
1708 DBG("Application communication apps thread cleanup complete");
1709 rcu_thread_offline();
1710 rcu_unregister_thread();
1711 return NULL;
1712 }
1713
1714 /*
1715 * Send a socket to a thread This is called from the dispatch UST registration
1716 * thread once all sockets are set for the application.
1717 *
1718 * The sock value can be invalid, we don't really care, the thread will handle
1719 * it and make the necessary cleanup if so.
1720 *
1721 * On success, return 0 else a negative value being the errno message of the
1722 * write().
1723 */
1724 static int send_socket_to_thread(int fd, int sock)
1725 {
1726 ssize_t ret;
1727
1728 /*
1729 * It's possible that the FD is set as invalid with -1 concurrently just
1730 * before calling this function being a shutdown state of the thread.
1731 */
1732 if (fd < 0) {
1733 ret = -EBADF;
1734 goto error;
1735 }
1736
1737 ret = lttng_write(fd, &sock, sizeof(sock));
1738 if (ret < sizeof(sock)) {
1739 PERROR("write apps pipe %d", fd);
1740 if (ret < 0) {
1741 ret = -errno;
1742 }
1743 goto error;
1744 }
1745
1746 /* All good. Don't send back the write positive ret value. */
1747 ret = 0;
1748 error:
1749 return (int) ret;
1750 }
1751
1752 /*
1753 * Sanitize the wait queue of the dispatch registration thread meaning removing
1754 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1755 * notify socket is never received.
1756 */
1757 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1758 {
1759 int ret, nb_fd = 0, i;
1760 unsigned int fd_added = 0;
1761 struct lttng_poll_event events;
1762 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1763
1764 assert(wait_queue);
1765
1766 lttng_poll_init(&events);
1767
1768 /* Just skip everything for an empty queue. */
1769 if (!wait_queue->count) {
1770 goto end;
1771 }
1772
1773 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1774 if (ret < 0) {
1775 goto error_create;
1776 }
1777
1778 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1779 &wait_queue->head, head) {
1780 assert(wait_node->app);
1781 ret = lttng_poll_add(&events, wait_node->app->sock,
1782 LPOLLHUP | LPOLLERR);
1783 if (ret < 0) {
1784 goto error;
1785 }
1786
1787 fd_added = 1;
1788 }
1789
1790 if (!fd_added) {
1791 goto end;
1792 }
1793
1794 /*
1795 * Poll but don't block so we can quickly identify the faulty events and
1796 * clean them afterwards from the wait queue.
1797 */
1798 ret = lttng_poll_wait(&events, 0);
1799 if (ret < 0) {
1800 goto error;
1801 }
1802 nb_fd = ret;
1803
1804 for (i = 0; i < nb_fd; i++) {
1805 /* Get faulty FD. */
1806 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1807 int pollfd = LTTNG_POLL_GETFD(&events, i);
1808
1809 if (!revents) {
1810 /* No activity for this FD (poll implementation). */
1811 continue;
1812 }
1813
1814 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1815 &wait_queue->head, head) {
1816 if (pollfd == wait_node->app->sock &&
1817 (revents & (LPOLLHUP | LPOLLERR))) {
1818 cds_list_del(&wait_node->head);
1819 wait_queue->count--;
1820 ust_app_destroy(wait_node->app);
1821 free(wait_node);
1822 /*
1823 * Silence warning of use-after-free in
1824 * cds_list_for_each_entry_safe which uses
1825 * __typeof__(*wait_node).
1826 */
1827 wait_node = NULL;
1828 break;
1829 }
1830 }
1831 }
1832
1833 if (nb_fd > 0) {
1834 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1835 }
1836
1837 end:
1838 lttng_poll_clean(&events);
1839 return;
1840
1841 error:
1842 lttng_poll_clean(&events);
1843 error_create:
1844 ERR("Unable to sanitize wait queue");
1845 return;
1846 }
1847
1848 /*
1849 * Dispatch request from the registration threads to the application
1850 * communication thread.
1851 */
1852 static void *thread_dispatch_ust_registration(void *data)
1853 {
1854 int ret, err = -1;
1855 struct cds_wfcq_node *node;
1856 struct ust_command *ust_cmd = NULL;
1857 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1858 struct ust_reg_wait_queue wait_queue = {
1859 .count = 0,
1860 };
1861
1862 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1863
1864 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1865 goto error_testpoint;
1866 }
1867
1868 health_code_update();
1869
1870 CDS_INIT_LIST_HEAD(&wait_queue.head);
1871
1872 DBG("[thread] Dispatch UST command started");
1873
1874 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1875 health_code_update();
1876
1877 /* Atomically prepare the queue futex */
1878 futex_nto1_prepare(&ust_cmd_queue.futex);
1879
1880 do {
1881 struct ust_app *app = NULL;
1882 ust_cmd = NULL;
1883
1884 /*
1885 * Make sure we don't have node(s) that have hung up before receiving
1886 * the notify socket. This is to clean the list in order to avoid
1887 * memory leaks from notify socket that are never seen.
1888 */
1889 sanitize_wait_queue(&wait_queue);
1890
1891 health_code_update();
1892 /* Dequeue command for registration */
1893 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1894 if (node == NULL) {
1895 DBG("Woken up but nothing in the UST command queue");
1896 /* Continue thread execution */
1897 break;
1898 }
1899
1900 ust_cmd = caa_container_of(node, struct ust_command, node);
1901
1902 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1903 " gid:%d sock:%d name:%s (version %d.%d)",
1904 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1905 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1906 ust_cmd->sock, ust_cmd->reg_msg.name,
1907 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1908
1909 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1910 wait_node = zmalloc(sizeof(*wait_node));
1911 if (!wait_node) {
1912 PERROR("zmalloc wait_node dispatch");
1913 ret = close(ust_cmd->sock);
1914 if (ret < 0) {
1915 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1916 }
1917 lttng_fd_put(LTTNG_FD_APPS, 1);
1918 free(ust_cmd);
1919 goto error;
1920 }
1921 CDS_INIT_LIST_HEAD(&wait_node->head);
1922
1923 /* Create application object if socket is CMD. */
1924 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1925 ust_cmd->sock);
1926 if (!wait_node->app) {
1927 ret = close(ust_cmd->sock);
1928 if (ret < 0) {
1929 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1930 }
1931 lttng_fd_put(LTTNG_FD_APPS, 1);
1932 free(wait_node);
1933 free(ust_cmd);
1934 continue;
1935 }
1936 /*
1937 * Add application to the wait queue so we can set the notify
1938 * socket before putting this object in the global ht.
1939 */
1940 cds_list_add(&wait_node->head, &wait_queue.head);
1941 wait_queue.count++;
1942
1943 free(ust_cmd);
1944 /*
1945 * We have to continue here since we don't have the notify
1946 * socket and the application MUST be added to the hash table
1947 * only at that moment.
1948 */
1949 continue;
1950 } else {
1951 /*
1952 * Look for the application in the local wait queue and set the
1953 * notify socket if found.
1954 */
1955 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1956 &wait_queue.head, head) {
1957 health_code_update();
1958 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1959 wait_node->app->notify_sock = ust_cmd->sock;
1960 cds_list_del(&wait_node->head);
1961 wait_queue.count--;
1962 app = wait_node->app;
1963 free(wait_node);
1964 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1965 break;
1966 }
1967 }
1968
1969 /*
1970 * With no application at this stage the received socket is
1971 * basically useless so close it before we free the cmd data
1972 * structure for good.
1973 */
1974 if (!app) {
1975 ret = close(ust_cmd->sock);
1976 if (ret < 0) {
1977 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1978 }
1979 lttng_fd_put(LTTNG_FD_APPS, 1);
1980 }
1981 free(ust_cmd);
1982 }
1983
1984 if (app) {
1985 /*
1986 * @session_lock_list
1987 *
1988 * Lock the global session list so from the register up to the
1989 * registration done message, no thread can see the application
1990 * and change its state.
1991 */
1992 session_lock_list();
1993 rcu_read_lock();
1994
1995 /*
1996 * Add application to the global hash table. This needs to be
1997 * done before the update to the UST registry can locate the
1998 * application.
1999 */
2000 ust_app_add(app);
2001
2002 /* Set app version. This call will print an error if needed. */
2003 (void) ust_app_version(app);
2004
2005 /* Send notify socket through the notify pipe. */
2006 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2007 app->notify_sock);
2008 if (ret < 0) {
2009 rcu_read_unlock();
2010 session_unlock_list();
2011 /*
2012 * No notify thread, stop the UST tracing. However, this is
2013 * not an internal error of the this thread thus setting
2014 * the health error code to a normal exit.
2015 */
2016 err = 0;
2017 goto error;
2018 }
2019
2020 /*
2021 * Update newly registered application with the tracing
2022 * registry info already enabled information.
2023 */
2024 update_ust_app(app->sock);
2025
2026 /*
2027 * Don't care about return value. Let the manage apps threads
2028 * handle app unregistration upon socket close.
2029 */
2030 (void) ust_app_register_done(app);
2031
2032 /*
2033 * Even if the application socket has been closed, send the app
2034 * to the thread and unregistration will take place at that
2035 * place.
2036 */
2037 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2038 if (ret < 0) {
2039 rcu_read_unlock();
2040 session_unlock_list();
2041 /*
2042 * No apps. thread, stop the UST tracing. However, this is
2043 * not an internal error of the this thread thus setting
2044 * the health error code to a normal exit.
2045 */
2046 err = 0;
2047 goto error;
2048 }
2049
2050 rcu_read_unlock();
2051 session_unlock_list();
2052 }
2053 } while (node != NULL);
2054
2055 health_poll_entry();
2056 /* Futex wait on queue. Blocking call on futex() */
2057 futex_nto1_wait(&ust_cmd_queue.futex);
2058 health_poll_exit();
2059 }
2060 /* Normal exit, no error */
2061 err = 0;
2062
2063 error:
2064 /* Clean up wait queue. */
2065 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2066 &wait_queue.head, head) {
2067 cds_list_del(&wait_node->head);
2068 wait_queue.count--;
2069 free(wait_node);
2070 }
2071
2072 /* Empty command queue. */
2073 for (;;) {
2074 /* Dequeue command for registration */
2075 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2076 if (node == NULL) {
2077 break;
2078 }
2079 ust_cmd = caa_container_of(node, struct ust_command, node);
2080 ret = close(ust_cmd->sock);
2081 if (ret < 0) {
2082 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2083 }
2084 lttng_fd_put(LTTNG_FD_APPS, 1);
2085 free(ust_cmd);
2086 }
2087
2088 error_testpoint:
2089 DBG("Dispatch thread dying");
2090 if (err) {
2091 health_error();
2092 ERR("Health error occurred in %s", __func__);
2093 }
2094 health_unregister(health_sessiond);
2095 return NULL;
2096 }
2097
2098 /*
2099 * This thread manage application registration.
2100 */
2101 static void *thread_registration_apps(void *data)
2102 {
2103 int sock = -1, i, ret, pollfd, err = -1;
2104 uint32_t revents, nb_fd;
2105 struct lttng_poll_event events;
2106 /*
2107 * Get allocated in this thread, enqueued to a global queue, dequeued and
2108 * freed in the manage apps thread.
2109 */
2110 struct ust_command *ust_cmd = NULL;
2111
2112 DBG("[thread] Manage application registration started");
2113
2114 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2115
2116 if (testpoint(sessiond_thread_registration_apps)) {
2117 goto error_testpoint;
2118 }
2119
2120 ret = lttcomm_listen_unix_sock(apps_sock);
2121 if (ret < 0) {
2122 goto error_listen;
2123 }
2124
2125 /*
2126 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2127 * more will be added to this poll set.
2128 */
2129 ret = sessiond_set_thread_pollset(&events, 2);
2130 if (ret < 0) {
2131 goto error_create_poll;
2132 }
2133
2134 /* Add the application registration socket */
2135 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2136 if (ret < 0) {
2137 goto error_poll_add;
2138 }
2139
2140 /* Notify all applications to register */
2141 ret = notify_ust_apps(1);
2142 if (ret < 0) {
2143 ERR("Failed to notify applications or create the wait shared memory.\n"
2144 "Execution continues but there might be problem for already\n"
2145 "running applications that wishes to register.");
2146 }
2147
2148 while (1) {
2149 DBG("Accepting application registration");
2150
2151 /* Inifinite blocking call, waiting for transmission */
2152 restart:
2153 health_poll_entry();
2154 ret = lttng_poll_wait(&events, -1);
2155 health_poll_exit();
2156 if (ret < 0) {
2157 /*
2158 * Restart interrupted system call.
2159 */
2160 if (errno == EINTR) {
2161 goto restart;
2162 }
2163 goto error;
2164 }
2165
2166 nb_fd = ret;
2167
2168 for (i = 0; i < nb_fd; i++) {
2169 health_code_update();
2170
2171 /* Fetch once the poll data */
2172 revents = LTTNG_POLL_GETEV(&events, i);
2173 pollfd = LTTNG_POLL_GETFD(&events, i);
2174
2175 if (!revents) {
2176 /* No activity for this FD (poll implementation). */
2177 continue;
2178 }
2179
2180 /* Thread quit pipe has been closed. Killing thread. */
2181 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2182 if (ret) {
2183 err = 0;
2184 goto exit;
2185 }
2186
2187 /* Event on the registration socket */
2188 if (pollfd == apps_sock) {
2189 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2190 ERR("Register apps socket poll error");
2191 goto error;
2192 } else if (revents & LPOLLIN) {
2193 sock = lttcomm_accept_unix_sock(apps_sock);
2194 if (sock < 0) {
2195 goto error;
2196 }
2197
2198 /*
2199 * Set socket timeout for both receiving and ending.
2200 * app_socket_timeout is in seconds, whereas
2201 * lttcomm_setsockopt_rcv_timeout and
2202 * lttcomm_setsockopt_snd_timeout expect msec as
2203 * parameter.
2204 */
2205 (void) lttcomm_setsockopt_rcv_timeout(sock,
2206 app_socket_timeout * 1000);
2207 (void) lttcomm_setsockopt_snd_timeout(sock,
2208 app_socket_timeout * 1000);
2209
2210 /*
2211 * Set the CLOEXEC flag. Return code is useless because
2212 * either way, the show must go on.
2213 */
2214 (void) utils_set_fd_cloexec(sock);
2215
2216 /* Create UST registration command for enqueuing */
2217 ust_cmd = zmalloc(sizeof(struct ust_command));
2218 if (ust_cmd == NULL) {
2219 PERROR("ust command zmalloc");
2220 ret = close(sock);
2221 if (ret) {
2222 PERROR("close");
2223 }
2224 goto error;
2225 }
2226
2227 /*
2228 * Using message-based transmissions to ensure we don't
2229 * have to deal with partially received messages.
2230 */
2231 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2232 if (ret < 0) {
2233 ERR("Exhausted file descriptors allowed for applications.");
2234 free(ust_cmd);
2235 ret = close(sock);
2236 if (ret) {
2237 PERROR("close");
2238 }
2239 sock = -1;
2240 continue;
2241 }
2242
2243 health_code_update();
2244 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2245 if (ret < 0) {
2246 free(ust_cmd);
2247 /* Close socket of the application. */
2248 ret = close(sock);
2249 if (ret) {
2250 PERROR("close");
2251 }
2252 lttng_fd_put(LTTNG_FD_APPS, 1);
2253 sock = -1;
2254 continue;
2255 }
2256 health_code_update();
2257
2258 ust_cmd->sock = sock;
2259 sock = -1;
2260
2261 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2262 " gid:%d sock:%d name:%s (version %d.%d)",
2263 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2264 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2265 ust_cmd->sock, ust_cmd->reg_msg.name,
2266 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2267
2268 /*
2269 * Lock free enqueue the registration request. The red pill
2270 * has been taken! This apps will be part of the *system*.
2271 */
2272 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2273
2274 /*
2275 * Wake the registration queue futex. Implicit memory
2276 * barrier with the exchange in cds_wfcq_enqueue.
2277 */
2278 futex_nto1_wake(&ust_cmd_queue.futex);
2279 }
2280 }
2281 }
2282 }
2283
2284 exit:
2285 error:
2286 /* Notify that the registration thread is gone */
2287 notify_ust_apps(0);
2288
2289 if (apps_sock >= 0) {
2290 ret = close(apps_sock);
2291 if (ret) {
2292 PERROR("close");
2293 }
2294 }
2295 if (sock >= 0) {
2296 ret = close(sock);
2297 if (ret) {
2298 PERROR("close");
2299 }
2300 lttng_fd_put(LTTNG_FD_APPS, 1);
2301 }
2302 unlink(apps_unix_sock_path);
2303
2304 error_poll_add:
2305 lttng_poll_clean(&events);
2306 error_listen:
2307 error_create_poll:
2308 error_testpoint:
2309 DBG("UST Registration thread cleanup complete");
2310 if (err) {
2311 health_error();
2312 ERR("Health error occurred in %s", __func__);
2313 }
2314 health_unregister(health_sessiond);
2315
2316 return NULL;
2317 }
2318
2319 /*
2320 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2321 * exec or it will fails.
2322 */
2323 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2324 {
2325 int ret, clock_ret;
2326 struct timespec timeout;
2327
2328 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2329 consumer_data->consumer_thread_is_ready = 0;
2330
2331 /* Setup pthread condition */
2332 ret = pthread_condattr_init(&consumer_data->condattr);
2333 if (ret) {
2334 errno = ret;
2335 PERROR("pthread_condattr_init consumer data");
2336 goto error;
2337 }
2338
2339 /*
2340 * Set the monotonic clock in order to make sure we DO NOT jump in time
2341 * between the clock_gettime() call and the timedwait call. See bug #324
2342 * for a more details and how we noticed it.
2343 */
2344 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2345 if (ret) {
2346 errno = ret;
2347 PERROR("pthread_condattr_setclock consumer data");
2348 goto error;
2349 }
2350
2351 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2352 if (ret) {
2353 errno = ret;
2354 PERROR("pthread_cond_init consumer data");
2355 goto error;
2356 }
2357
2358 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2359 consumer_data);
2360 if (ret) {
2361 errno = ret;
2362 PERROR("pthread_create consumer");
2363 ret = -1;
2364 goto error;
2365 }
2366
2367 /* We are about to wait on a pthread condition */
2368 pthread_mutex_lock(&consumer_data->cond_mutex);
2369
2370 /* Get time for sem_timedwait absolute timeout */
2371 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2372 /*
2373 * Set the timeout for the condition timed wait even if the clock gettime
2374 * call fails since we might loop on that call and we want to avoid to
2375 * increment the timeout too many times.
2376 */
2377 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2378
2379 /*
2380 * The following loop COULD be skipped in some conditions so this is why we
2381 * set ret to 0 in order to make sure at least one round of the loop is
2382 * done.
2383 */
2384 ret = 0;
2385
2386 /*
2387 * Loop until the condition is reached or when a timeout is reached. Note
2388 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2389 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2390 * possible. This loop does not take any chances and works with both of
2391 * them.
2392 */
2393 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2394 if (clock_ret < 0) {
2395 PERROR("clock_gettime spawn consumer");
2396 /* Infinite wait for the consumerd thread to be ready */
2397 ret = pthread_cond_wait(&consumer_data->cond,
2398 &consumer_data->cond_mutex);
2399 } else {
2400 ret = pthread_cond_timedwait(&consumer_data->cond,
2401 &consumer_data->cond_mutex, &timeout);
2402 }
2403 }
2404
2405 /* Release the pthread condition */
2406 pthread_mutex_unlock(&consumer_data->cond_mutex);
2407
2408 if (ret != 0) {
2409 errno = ret;
2410 if (ret == ETIMEDOUT) {
2411 int pth_ret;
2412
2413 /*
2414 * Call has timed out so we kill the kconsumerd_thread and return
2415 * an error.
2416 */
2417 ERR("Condition timed out. The consumer thread was never ready."
2418 " Killing it");
2419 pth_ret = pthread_cancel(consumer_data->thread);
2420 if (pth_ret < 0) {
2421 PERROR("pthread_cancel consumer thread");
2422 }
2423 } else {
2424 PERROR("pthread_cond_wait failed consumer thread");
2425 }
2426 /* Caller is expecting a negative value on failure. */
2427 ret = -1;
2428 goto error;
2429 }
2430
2431 pthread_mutex_lock(&consumer_data->pid_mutex);
2432 if (consumer_data->pid == 0) {
2433 ERR("Consumerd did not start");
2434 pthread_mutex_unlock(&consumer_data->pid_mutex);
2435 goto error;
2436 }
2437 pthread_mutex_unlock(&consumer_data->pid_mutex);
2438
2439 return 0;
2440
2441 error:
2442 return ret;
2443 }
2444
2445 /*
2446 * Join consumer thread
2447 */
2448 static int join_consumer_thread(struct consumer_data *consumer_data)
2449 {
2450 void *status;
2451
2452 /* Consumer pid must be a real one. */
2453 if (consumer_data->pid > 0) {
2454 int ret;
2455 ret = kill(consumer_data->pid, SIGTERM);
2456 if (ret) {
2457 PERROR("Error killing consumer daemon");
2458 return ret;
2459 }
2460 return pthread_join(consumer_data->thread, &status);
2461 } else {
2462 return 0;
2463 }
2464 }
2465
2466 /*
2467 * Fork and exec a consumer daemon (consumerd).
2468 *
2469 * Return pid if successful else -1.
2470 */
2471 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2472 {
2473 int ret;
2474 pid_t pid;
2475 const char *consumer_to_use;
2476 const char *verbosity;
2477 struct stat st;
2478
2479 DBG("Spawning consumerd");
2480
2481 pid = fork();
2482 if (pid == 0) {
2483 /*
2484 * Exec consumerd.
2485 */
2486 if (opt_verbose_consumer) {
2487 verbosity = "--verbose";
2488 } else if (lttng_opt_quiet) {
2489 verbosity = "--quiet";
2490 } else {
2491 verbosity = "";
2492 }
2493
2494 switch (consumer_data->type) {
2495 case LTTNG_CONSUMER_KERNEL:
2496 /*
2497 * Find out which consumerd to execute. We will first try the
2498 * 64-bit path, then the sessiond's installation directory, and
2499 * fallback on the 32-bit one,
2500 */
2501 DBG3("Looking for a kernel consumer at these locations:");
2502 DBG3(" 1) %s", consumerd64_bin);
2503 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2504 DBG3(" 3) %s", consumerd32_bin);
2505 if (stat(consumerd64_bin, &st) == 0) {
2506 DBG3("Found location #1");
2507 consumer_to_use = consumerd64_bin;
2508 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2509 DBG3("Found location #2");
2510 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2511 } else if (stat(consumerd32_bin, &st) == 0) {
2512 DBG3("Found location #3");
2513 consumer_to_use = consumerd32_bin;
2514 } else {
2515 DBG("Could not find any valid consumerd executable");
2516 ret = -EINVAL;
2517 break;
2518 }
2519 DBG("Using kernel consumer at: %s", consumer_to_use);
2520 ret = execl(consumer_to_use,
2521 "lttng-consumerd", verbosity, "-k",
2522 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2523 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2524 "--group", tracing_group_name,
2525 NULL);
2526 break;
2527 case LTTNG_CONSUMER64_UST:
2528 {
2529 char *tmpnew = NULL;
2530
2531 if (consumerd64_libdir[0] != '\0') {
2532 char *tmp;
2533 size_t tmplen;
2534
2535 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2536 if (!tmp) {
2537 tmp = "";
2538 }
2539 tmplen = strlen("LD_LIBRARY_PATH=")
2540 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2541 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2542 if (!tmpnew) {
2543 ret = -ENOMEM;
2544 goto error;
2545 }
2546 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2547 strcat(tmpnew, consumerd64_libdir);
2548 if (tmp[0] != '\0') {
2549 strcat(tmpnew, ":");
2550 strcat(tmpnew, tmp);
2551 }
2552 ret = putenv(tmpnew);
2553 if (ret) {
2554 ret = -errno;
2555 free(tmpnew);
2556 goto error;
2557 }
2558 }
2559 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2560 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2561 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2562 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2563 "--group", tracing_group_name,
2564 NULL);
2565 if (consumerd64_libdir[0] != '\0') {
2566 free(tmpnew);
2567 }
2568 break;
2569 }
2570 case LTTNG_CONSUMER32_UST:
2571 {
2572 char *tmpnew = NULL;
2573
2574 if (consumerd32_libdir[0] != '\0') {
2575 char *tmp;
2576 size_t tmplen;
2577
2578 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2579 if (!tmp) {
2580 tmp = "";
2581 }
2582 tmplen = strlen("LD_LIBRARY_PATH=")
2583 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2584 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2585 if (!tmpnew) {
2586 ret = -ENOMEM;
2587 goto error;
2588 }
2589 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2590 strcat(tmpnew, consumerd32_libdir);
2591 if (tmp[0] != '\0') {
2592 strcat(tmpnew, ":");
2593 strcat(tmpnew, tmp);
2594 }
2595 ret = putenv(tmpnew);
2596 if (ret) {
2597 ret = -errno;
2598 free(tmpnew);
2599 goto error;
2600 }
2601 }
2602 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2603 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2604 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2605 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2606 "--group", tracing_group_name,
2607 NULL);
2608 if (consumerd32_libdir[0] != '\0') {
2609 free(tmpnew);
2610 }
2611 break;
2612 }
2613 default:
2614 PERROR("unknown consumer type");
2615 exit(EXIT_FAILURE);
2616 }
2617 if (errno != 0) {
2618 PERROR("Consumer execl()");
2619 }
2620 /* Reaching this point, we got a failure on our execl(). */
2621 exit(EXIT_FAILURE);
2622 } else if (pid > 0) {
2623 ret = pid;
2624 } else {
2625 PERROR("start consumer fork");
2626 ret = -errno;
2627 }
2628 error:
2629 return ret;
2630 }
2631
2632 /*
2633 * Spawn the consumerd daemon and session daemon thread.
2634 */
2635 static int start_consumerd(struct consumer_data *consumer_data)
2636 {
2637 int ret;
2638
2639 /*
2640 * Set the listen() state on the socket since there is a possible race
2641 * between the exec() of the consumer daemon and this call if place in the
2642 * consumer thread. See bug #366 for more details.
2643 */
2644 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2645 if (ret < 0) {
2646 goto error;
2647 }
2648
2649 pthread_mutex_lock(&consumer_data->pid_mutex);
2650 if (consumer_data->pid != 0) {
2651 pthread_mutex_unlock(&consumer_data->pid_mutex);
2652 goto end;
2653 }
2654
2655 ret = spawn_consumerd(consumer_data);
2656 if (ret < 0) {
2657 ERR("Spawning consumerd failed");
2658 pthread_mutex_unlock(&consumer_data->pid_mutex);
2659 goto error;
2660 }
2661
2662 /* Setting up the consumer_data pid */
2663 consumer_data->pid = ret;
2664 DBG2("Consumer pid %d", consumer_data->pid);
2665 pthread_mutex_unlock(&consumer_data->pid_mutex);
2666
2667 DBG2("Spawning consumer control thread");
2668 ret = spawn_consumer_thread(consumer_data);
2669 if (ret < 0) {
2670 ERR("Fatal error spawning consumer control thread");
2671 goto error;
2672 }
2673
2674 end:
2675 return 0;
2676
2677 error:
2678 /* Cleanup already created sockets on error. */
2679 if (consumer_data->err_sock >= 0) {
2680 int err;
2681
2682 err = close(consumer_data->err_sock);
2683 if (err < 0) {
2684 PERROR("close consumer data error socket");
2685 }
2686 }
2687 return ret;
2688 }
2689
2690 /*
2691 * Setup necessary data for kernel tracer action.
2692 */
2693 static int init_kernel_tracer(void)
2694 {
2695 int ret;
2696
2697 /* Modprobe lttng kernel modules */
2698 ret = modprobe_lttng_control();
2699 if (ret < 0) {
2700 goto error;
2701 }
2702
2703 /* Open debugfs lttng */
2704 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2705 if (kernel_tracer_fd < 0) {
2706 DBG("Failed to open %s", module_proc_lttng);
2707 ret = -1;
2708 goto error_open;
2709 }
2710
2711 /* Validate kernel version */
2712 ret = kernel_validate_version(kernel_tracer_fd);
2713 if (ret < 0) {
2714 goto error_version;
2715 }
2716
2717 ret = modprobe_lttng_data();
2718 if (ret < 0) {
2719 goto error_modules;
2720 }
2721
2722 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2723 return 0;
2724
2725 error_version:
2726 modprobe_remove_lttng_control();
2727 ret = close(kernel_tracer_fd);
2728 if (ret) {
2729 PERROR("close");
2730 }
2731 kernel_tracer_fd = -1;
2732 return LTTNG_ERR_KERN_VERSION;
2733
2734 error_modules:
2735 ret = close(kernel_tracer_fd);
2736 if (ret) {
2737 PERROR("close");
2738 }
2739
2740 error_open:
2741 modprobe_remove_lttng_control();
2742
2743 error:
2744 WARN("No kernel tracer available");
2745 kernel_tracer_fd = -1;
2746 if (!is_root) {
2747 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2748 } else {
2749 return LTTNG_ERR_KERN_NA;
2750 }
2751 }
2752
2753
2754 /*
2755 * Copy consumer output from the tracing session to the domain session. The
2756 * function also applies the right modification on a per domain basis for the
2757 * trace files destination directory.
2758 *
2759 * Should *NOT* be called with RCU read-side lock held.
2760 */
2761 static int copy_session_consumer(int domain, struct ltt_session *session)
2762 {
2763 int ret;
2764 const char *dir_name;
2765 struct consumer_output *consumer;
2766
2767 assert(session);
2768 assert(session->consumer);
2769
2770 switch (domain) {
2771 case LTTNG_DOMAIN_KERNEL:
2772 DBG3("Copying tracing session consumer output in kernel session");
2773 /*
2774 * XXX: We should audit the session creation and what this function
2775 * does "extra" in order to avoid a destroy since this function is used
2776 * in the domain session creation (kernel and ust) only. Same for UST
2777 * domain.
2778 */
2779 if (session->kernel_session->consumer) {
2780 consumer_output_put(session->kernel_session->consumer);
2781 }
2782 session->kernel_session->consumer =
2783 consumer_copy_output(session->consumer);
2784 /* Ease our life a bit for the next part */
2785 consumer = session->kernel_session->consumer;
2786 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2787 break;
2788 case LTTNG_DOMAIN_JUL:
2789 case LTTNG_DOMAIN_LOG4J:
2790 case LTTNG_DOMAIN_PYTHON:
2791 case LTTNG_DOMAIN_UST:
2792 DBG3("Copying tracing session consumer output in UST session");
2793 if (session->ust_session->consumer) {
2794 consumer_output_put(session->ust_session->consumer);
2795 }
2796 session->ust_session->consumer =
2797 consumer_copy_output(session->consumer);
2798 /* Ease our life a bit for the next part */
2799 consumer = session->ust_session->consumer;
2800 dir_name = DEFAULT_UST_TRACE_DIR;
2801 break;
2802 default:
2803 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2804 goto error;
2805 }
2806
2807 /* Append correct directory to subdir */
2808 strncat(consumer->subdir, dir_name,
2809 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2810 DBG3("Copy session consumer subdir %s", consumer->subdir);
2811
2812 ret = LTTNG_OK;
2813
2814 error:
2815 return ret;
2816 }
2817
2818 /*
2819 * Create an UST session and add it to the session ust list.
2820 *
2821 * Should *NOT* be called with RCU read-side lock held.
2822 */
2823 static int create_ust_session(struct ltt_session *session,
2824 struct lttng_domain *domain)
2825 {
2826 int ret;
2827 struct ltt_ust_session *lus = NULL;
2828
2829 assert(session);
2830 assert(domain);
2831 assert(session->consumer);
2832
2833 switch (domain->type) {
2834 case LTTNG_DOMAIN_JUL:
2835 case LTTNG_DOMAIN_LOG4J:
2836 case LTTNG_DOMAIN_PYTHON:
2837 case LTTNG_DOMAIN_UST:
2838 break;
2839 default:
2840 ERR("Unknown UST domain on create session %d", domain->type);
2841 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2842 goto error;
2843 }
2844
2845 DBG("Creating UST session");
2846
2847 lus = trace_ust_create_session(session->id);
2848 if (lus == NULL) {
2849 ret = LTTNG_ERR_UST_SESS_FAIL;
2850 goto error;
2851 }
2852
2853 lus->uid = session->uid;
2854 lus->gid = session->gid;
2855 lus->output_traces = session->output_traces;
2856 lus->snapshot_mode = session->snapshot_mode;
2857 lus->live_timer_interval = session->live_timer;
2858 session->ust_session = lus;
2859 if (session->shm_path[0]) {
2860 strncpy(lus->root_shm_path, session->shm_path,
2861 sizeof(lus->root_shm_path));
2862 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2863 strncpy(lus->shm_path, session->shm_path,
2864 sizeof(lus->shm_path));
2865 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2866 strncat(lus->shm_path, "/ust",
2867 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2868 }
2869 /* Copy session output to the newly created UST session */
2870 ret = copy_session_consumer(domain->type, session);
2871 if (ret != LTTNG_OK) {
2872 goto error;
2873 }
2874
2875 return LTTNG_OK;
2876
2877 error:
2878 free(lus);
2879 session->ust_session = NULL;
2880 return ret;
2881 }
2882
2883 /*
2884 * Create a kernel tracer session then create the default channel.
2885 */
2886 static int create_kernel_session(struct ltt_session *session)
2887 {
2888 int ret;
2889
2890 DBG("Creating kernel session");
2891
2892 ret = kernel_create_session(session, kernel_tracer_fd);
2893 if (ret < 0) {
2894 ret = LTTNG_ERR_KERN_SESS_FAIL;
2895 goto error;
2896 }
2897
2898 /* Code flow safety */
2899 assert(session->kernel_session);
2900
2901 /* Copy session output to the newly created Kernel session */
2902 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2903 if (ret != LTTNG_OK) {
2904 goto error;
2905 }
2906
2907 /* Create directory(ies) on local filesystem. */
2908 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2909 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2910 ret = run_as_mkdir_recursive(
2911 session->kernel_session->consumer->dst.trace_path,
2912 S_IRWXU | S_IRWXG, session->uid, session->gid);
2913 if (ret < 0) {
2914 if (errno != EEXIST) {
2915 ERR("Trace directory creation error");
2916 goto error;
2917 }
2918 }
2919 }
2920
2921 session->kernel_session->uid = session->uid;
2922 session->kernel_session->gid = session->gid;
2923 session->kernel_session->output_traces = session->output_traces;
2924 session->kernel_session->snapshot_mode = session->snapshot_mode;
2925
2926 return LTTNG_OK;
2927
2928 error:
2929 trace_kernel_destroy_session(session->kernel_session);
2930 session->kernel_session = NULL;
2931 return ret;
2932 }
2933
2934 /*
2935 * Count number of session permitted by uid/gid.
2936 */
2937 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2938 {
2939 unsigned int i = 0;
2940 struct ltt_session *session;
2941
2942 DBG("Counting number of available session for UID %d GID %d",
2943 uid, gid);
2944 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2945 /*
2946 * Only list the sessions the user can control.
2947 */
2948 if (!session_access_ok(session, uid, gid)) {
2949 continue;
2950 }
2951 i++;
2952 }
2953 return i;
2954 }
2955
2956 /*
2957 * Process the command requested by the lttng client within the command
2958 * context structure. This function make sure that the return structure (llm)
2959 * is set and ready for transmission before returning.
2960 *
2961 * Return any error encountered or 0 for success.
2962 *
2963 * "sock" is only used for special-case var. len data.
2964 *
2965 * Should *NOT* be called with RCU read-side lock held.
2966 */
2967 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2968 int *sock_error)
2969 {
2970 int ret = LTTNG_OK;
2971 int need_tracing_session = 1;
2972 int need_domain;
2973
2974 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2975
2976 *sock_error = 0;
2977
2978 switch (cmd_ctx->lsm->cmd_type) {
2979 case LTTNG_CREATE_SESSION:
2980 case LTTNG_CREATE_SESSION_SNAPSHOT:
2981 case LTTNG_CREATE_SESSION_LIVE:
2982 case LTTNG_DESTROY_SESSION:
2983 case LTTNG_LIST_SESSIONS:
2984 case LTTNG_LIST_DOMAINS:
2985 case LTTNG_START_TRACE:
2986 case LTTNG_STOP_TRACE:
2987 case LTTNG_DATA_PENDING:
2988 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2989 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2990 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2991 case LTTNG_SNAPSHOT_RECORD:
2992 case LTTNG_SAVE_SESSION:
2993 case LTTNG_SET_SESSION_SHM_PATH:
2994 need_domain = 0;
2995 break;
2996 default:
2997 need_domain = 1;
2998 }
2999
3000 if (opt_no_kernel && need_domain
3001 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3002 if (!is_root) {
3003 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3004 } else {
3005 ret = LTTNG_ERR_KERN_NA;
3006 }
3007 goto error;
3008 }
3009
3010 /* Deny register consumer if we already have a spawned consumer. */
3011 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3012 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3013 if (kconsumer_data.pid > 0) {
3014 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3015 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3016 goto error;
3017 }
3018 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3019 }
3020
3021 /*
3022 * Check for command that don't needs to allocate a returned payload. We do
3023 * this here so we don't have to make the call for no payload at each
3024 * command.
3025 */
3026 switch(cmd_ctx->lsm->cmd_type) {
3027 case LTTNG_LIST_SESSIONS:
3028 case LTTNG_LIST_TRACEPOINTS:
3029 case LTTNG_LIST_TRACEPOINT_FIELDS:
3030 case LTTNG_LIST_DOMAINS:
3031 case LTTNG_LIST_CHANNELS:
3032 case LTTNG_LIST_EVENTS:
3033 case LTTNG_LIST_SYSCALLS:
3034 case LTTNG_LIST_TRACKER_PIDS:
3035 break;
3036 default:
3037 /* Setup lttng message with no payload */
3038 ret = setup_lttng_msg(cmd_ctx, 0);
3039 if (ret < 0) {
3040 /* This label does not try to unlock the session */
3041 goto init_setup_error;
3042 }
3043 }
3044
3045 /* Commands that DO NOT need a session. */
3046 switch (cmd_ctx->lsm->cmd_type) {
3047 case LTTNG_CREATE_SESSION:
3048 case LTTNG_CREATE_SESSION_SNAPSHOT:
3049 case LTTNG_CREATE_SESSION_LIVE:
3050 case LTTNG_CALIBRATE:
3051 case LTTNG_LIST_SESSIONS:
3052 case LTTNG_LIST_TRACEPOINTS:
3053 case LTTNG_LIST_SYSCALLS:
3054 case LTTNG_LIST_TRACEPOINT_FIELDS:
3055 case LTTNG_SAVE_SESSION:
3056 need_tracing_session = 0;
3057 break;
3058 default:
3059 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3060 /*
3061 * We keep the session list lock across _all_ commands
3062 * for now, because the per-session lock does not
3063 * handle teardown properly.
3064 */
3065 session_lock_list();
3066 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3067 if (cmd_ctx->session == NULL) {
3068 ret = LTTNG_ERR_SESS_NOT_FOUND;
3069 goto error;
3070 } else {
3071 /* Acquire lock for the session */
3072 session_lock(cmd_ctx->session);
3073 }
3074 break;
3075 }
3076
3077 /*
3078 * Commands that need a valid session but should NOT create one if none
3079 * exists. Instead of creating one and destroying it when the command is
3080 * handled, process that right before so we save some round trip in useless
3081 * code path.
3082 */
3083 switch (cmd_ctx->lsm->cmd_type) {
3084 case LTTNG_DISABLE_CHANNEL:
3085 case LTTNG_DISABLE_EVENT:
3086 switch (cmd_ctx->lsm->domain.type) {
3087 case LTTNG_DOMAIN_KERNEL:
3088 if (!cmd_ctx->session->kernel_session) {
3089 ret = LTTNG_ERR_NO_CHANNEL;
3090 goto error;
3091 }
3092 break;
3093 case LTTNG_DOMAIN_JUL:
3094 case LTTNG_DOMAIN_LOG4J:
3095 case LTTNG_DOMAIN_PYTHON:
3096 case LTTNG_DOMAIN_UST:
3097 if (!cmd_ctx->session->ust_session) {
3098 ret = LTTNG_ERR_NO_CHANNEL;
3099 goto error;
3100 }
3101 break;
3102 default:
3103 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3104 goto error;
3105 }
3106 default:
3107 break;
3108 }
3109
3110 if (!need_domain) {
3111 goto skip_domain;
3112 }
3113
3114 /*
3115 * Check domain type for specific "pre-action".
3116 */
3117 switch (cmd_ctx->lsm->domain.type) {
3118 case LTTNG_DOMAIN_KERNEL:
3119 if (!is_root) {
3120 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3121 goto error;
3122 }
3123
3124 /* Kernel tracer check */
3125 if (kernel_tracer_fd == -1) {
3126 /* Basically, load kernel tracer modules */
3127 ret = init_kernel_tracer();
3128 if (ret != 0) {
3129 goto error;
3130 }
3131 }
3132
3133 /* Consumer is in an ERROR state. Report back to client */
3134 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3135 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3136 goto error;
3137 }
3138
3139 /* Need a session for kernel command */
3140 if (need_tracing_session) {
3141 if (cmd_ctx->session->kernel_session == NULL) {
3142 ret = create_kernel_session(cmd_ctx->session);
3143 if (ret < 0) {
3144 ret = LTTNG_ERR_KERN_SESS_FAIL;
3145 goto error;
3146 }
3147 }
3148
3149 /* Start the kernel consumer daemon */
3150 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3151 if (kconsumer_data.pid == 0 &&
3152 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3153 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3154 ret = start_consumerd(&kconsumer_data);
3155 if (ret < 0) {
3156 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3157 goto error;
3158 }
3159 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3160 } else {
3161 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3162 }
3163
3164 /*
3165 * The consumer was just spawned so we need to add the socket to
3166 * the consumer output of the session if exist.
3167 */
3168 ret = consumer_create_socket(&kconsumer_data,
3169 cmd_ctx->session->kernel_session->consumer);
3170 if (ret < 0) {
3171 goto error;
3172 }
3173 }
3174
3175 break;
3176 case LTTNG_DOMAIN_JUL:
3177 case LTTNG_DOMAIN_LOG4J:
3178 case LTTNG_DOMAIN_PYTHON:
3179 case LTTNG_DOMAIN_UST:
3180 {
3181 if (!ust_app_supported()) {
3182 ret = LTTNG_ERR_NO_UST;
3183 goto error;
3184 }
3185 /* Consumer is in an ERROR state. Report back to client */
3186 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3187 ret = LTTNG_ERR_NO_USTCONSUMERD;
3188 goto error;
3189 }
3190
3191 if (need_tracing_session) {
3192 /* Create UST session if none exist. */
3193 if (cmd_ctx->session->ust_session == NULL) {
3194 ret = create_ust_session(cmd_ctx->session,
3195 &cmd_ctx->lsm->domain);
3196 if (ret != LTTNG_OK) {
3197 goto error;
3198 }
3199 }
3200
3201 /* Start the UST consumer daemons */
3202 /* 64-bit */
3203 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3204 if (consumerd64_bin[0] != '\0' &&
3205 ustconsumer64_data.pid == 0 &&
3206 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3207 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3208 ret = start_consumerd(&ustconsumer64_data);
3209 if (ret < 0) {
3210 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3211 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3212 goto error;
3213 }
3214
3215 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3216 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3217 } else {
3218 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3219 }
3220
3221 /*
3222 * Setup socket for consumer 64 bit. No need for atomic access
3223 * since it was set above and can ONLY be set in this thread.
3224 */
3225 ret = consumer_create_socket(&ustconsumer64_data,
3226 cmd_ctx->session->ust_session->consumer);
3227 if (ret < 0) {
3228 goto error;
3229 }
3230
3231 /* 32-bit */
3232 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3233 if (consumerd32_bin[0] != '\0' &&
3234 ustconsumer32_data.pid == 0 &&
3235 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3236 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3237 ret = start_consumerd(&ustconsumer32_data);
3238 if (ret < 0) {
3239 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3240 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3241 goto error;
3242 }
3243
3244 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3245 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3246 } else {
3247 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3248 }
3249
3250 /*
3251 * Setup socket for consumer 64 bit. No need for atomic access
3252 * since it was set above and can ONLY be set in this thread.
3253 */
3254 ret = consumer_create_socket(&ustconsumer32_data,
3255 cmd_ctx->session->ust_session->consumer);
3256 if (ret < 0) {
3257 goto error;
3258 }
3259 }
3260 break;
3261 }
3262 default:
3263 break;
3264 }
3265 skip_domain:
3266
3267 /* Validate consumer daemon state when start/stop trace command */
3268 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3269 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3270 switch (cmd_ctx->lsm->domain.type) {
3271 case LTTNG_DOMAIN_JUL:
3272 case LTTNG_DOMAIN_LOG4J:
3273 case LTTNG_DOMAIN_PYTHON:
3274 case LTTNG_DOMAIN_UST:
3275 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3276 ret = LTTNG_ERR_NO_USTCONSUMERD;
3277 goto error;
3278 }
3279 break;
3280 case LTTNG_DOMAIN_KERNEL:
3281 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3282 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3283 goto error;
3284 }
3285 break;
3286 }
3287 }
3288
3289 /*
3290 * Check that the UID or GID match that of the tracing session.
3291 * The root user can interact with all sessions.
3292 */
3293 if (need_tracing_session) {
3294 if (!session_access_ok(cmd_ctx->session,
3295 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3296 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3297 ret = LTTNG_ERR_EPERM;
3298 goto error;
3299 }
3300 }
3301
3302 /*
3303 * Send relayd information to consumer as soon as we have a domain and a
3304 * session defined.
3305 */
3306 if (cmd_ctx->session && need_domain) {
3307 /*
3308 * Setup relayd if not done yet. If the relayd information was already
3309 * sent to the consumer, this call will gracefully return.
3310 */
3311 ret = cmd_setup_relayd(cmd_ctx->session);
3312 if (ret != LTTNG_OK) {
3313 goto error;
3314 }
3315 }
3316
3317 /* Process by command type */
3318 switch (cmd_ctx->lsm->cmd_type) {
3319 case LTTNG_ADD_CONTEXT:
3320 {
3321 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3322 cmd_ctx->lsm->u.context.channel_name,
3323 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3324 break;
3325 }
3326 case LTTNG_DISABLE_CHANNEL:
3327 {
3328 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3329 cmd_ctx->lsm->u.disable.channel_name);
3330 break;
3331 }
3332 case LTTNG_DISABLE_EVENT:
3333 {
3334
3335 /*
3336 * FIXME: handle filter; for now we just receive the filter's
3337 * bytecode along with the filter expression which are sent by
3338 * liblttng-ctl and discard them.
3339 *
3340 * This fixes an issue where the client may block while sending
3341 * the filter payload and encounter an error because the session
3342 * daemon closes the socket without ever handling this data.
3343 */
3344 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3345 cmd_ctx->lsm->u.disable.bytecode_len;
3346
3347 if (count) {
3348 char data[LTTNG_FILTER_MAX_LEN];
3349
3350 DBG("Discarding disable event command payload of size %zu", count);
3351 while (count) {
3352 ret = lttcomm_recv_unix_sock(sock, data,
3353 count > sizeof(data) ? sizeof(data) : count);
3354 if (ret < 0) {
3355 goto error;
3356 }
3357
3358 count -= (size_t) ret;
3359 }
3360 }
3361 /* FIXME: passing packed structure to non-packed pointer */
3362 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3363 cmd_ctx->lsm->u.disable.channel_name,
3364 &cmd_ctx->lsm->u.disable.event);
3365 break;
3366 }
3367 case LTTNG_ENABLE_CHANNEL:
3368 {
3369 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3370 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3371 break;
3372 }
3373 case LTTNG_TRACK_PID:
3374 {
3375 ret = cmd_track_pid(cmd_ctx->session,
3376 cmd_ctx->lsm->domain.type,
3377 cmd_ctx->lsm->u.pid_tracker.pid);
3378 break;
3379 }
3380 case LTTNG_UNTRACK_PID:
3381 {
3382 ret = cmd_untrack_pid(cmd_ctx->session,
3383 cmd_ctx->lsm->domain.type,
3384 cmd_ctx->lsm->u.pid_tracker.pid);
3385 break;
3386 }
3387 case LTTNG_ENABLE_EVENT:
3388 {
3389 struct lttng_event_exclusion *exclusion = NULL;
3390 struct lttng_filter_bytecode *bytecode = NULL;
3391 char *filter_expression = NULL;
3392
3393 /* Handle exclusion events and receive it from the client. */
3394 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3395 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3396
3397 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3398 (count * LTTNG_SYMBOL_NAME_LEN));
3399 if (!exclusion) {
3400 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3401 goto error;
3402 }
3403
3404 DBG("Receiving var len exclusion event list from client ...");
3405 exclusion->count = count;
3406 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3407 count * LTTNG_SYMBOL_NAME_LEN);
3408 if (ret <= 0) {
3409 DBG("Nothing recv() from client var len data... continuing");
3410 *sock_error = 1;
3411 free(exclusion);
3412 ret = LTTNG_ERR_EXCLUSION_INVAL;
3413 goto error;
3414 }
3415 }
3416
3417 /* Get filter expression from client. */
3418 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3419 size_t expression_len =
3420 cmd_ctx->lsm->u.enable.expression_len;
3421
3422 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3423 ret = LTTNG_ERR_FILTER_INVAL;
3424 free(exclusion);
3425 goto error;
3426 }
3427
3428 filter_expression = zmalloc(expression_len);
3429 if (!filter_expression) {
3430 free(exclusion);
3431 ret = LTTNG_ERR_FILTER_NOMEM;
3432 goto error;
3433 }
3434
3435 /* Receive var. len. data */
3436 DBG("Receiving var len filter's expression from client ...");
3437 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3438 expression_len);
3439 if (ret <= 0) {
3440 DBG("Nothing recv() from client car len data... continuing");
3441 *sock_error = 1;
3442 free(filter_expression);
3443 free(exclusion);
3444 ret = LTTNG_ERR_FILTER_INVAL;
3445 goto error;
3446 }
3447 }
3448
3449 /* Handle filter and get bytecode from client. */
3450 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3451 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3452
3453 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3454 ret = LTTNG_ERR_FILTER_INVAL;
3455 free(filter_expression);
3456 free(exclusion);
3457 goto error;
3458 }
3459
3460 bytecode = zmalloc(bytecode_len);
3461 if (!bytecode) {
3462 free(filter_expression);
3463 free(exclusion);
3464 ret = LTTNG_ERR_FILTER_NOMEM;
3465 goto error;
3466 }
3467
3468 /* Receive var. len. data */
3469 DBG("Receiving var len filter's bytecode from client ...");
3470 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3471 if (ret <= 0) {
3472 DBG("Nothing recv() from client car len data... continuing");
3473 *sock_error = 1;
3474 free(filter_expression);
3475 free(bytecode);
3476 free(exclusion);
3477 ret = LTTNG_ERR_FILTER_INVAL;
3478 goto error;
3479 }
3480
3481 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3482 free(filter_expression);
3483 free(bytecode);
3484 free(exclusion);
3485 ret = LTTNG_ERR_FILTER_INVAL;
3486 goto error;
3487 }
3488 }
3489
3490 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3491 cmd_ctx->lsm->u.enable.channel_name,
3492 &cmd_ctx->lsm->u.enable.event,
3493 filter_expression, bytecode, exclusion,
3494 kernel_poll_pipe[1]);
3495 break;
3496 }
3497 case LTTNG_LIST_TRACEPOINTS:
3498 {
3499 struct lttng_event *events;
3500 ssize_t nb_events;
3501
3502 session_lock_list();
3503 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3504 session_unlock_list();
3505 if (nb_events < 0) {
3506 /* Return value is a negative lttng_error_code. */
3507 ret = -nb_events;
3508 goto error;
3509 }
3510
3511 /*
3512 * Setup lttng message with payload size set to the event list size in
3513 * bytes and then copy list into the llm payload.
3514 */
3515 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3516 if (ret < 0) {
3517 free(events);
3518 goto setup_error;
3519 }
3520
3521 /* Copy event list into message payload */
3522 memcpy(cmd_ctx->llm->payload, events,
3523 sizeof(struct lttng_event) * nb_events);
3524
3525 free(events);
3526
3527 ret = LTTNG_OK;
3528 break;
3529 }
3530 case LTTNG_LIST_TRACEPOINT_FIELDS:
3531 {
3532 struct lttng_event_field *fields;
3533 ssize_t nb_fields;
3534
3535 session_lock_list();
3536 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3537 &fields);
3538 session_unlock_list();
3539 if (nb_fields < 0) {
3540 /* Return value is a negative lttng_error_code. */
3541 ret = -nb_fields;
3542 goto error;
3543 }
3544
3545 /*
3546 * Setup lttng message with payload size set to the event list size in
3547 * bytes and then copy list into the llm payload.
3548 */
3549 ret = setup_lttng_msg(cmd_ctx,
3550 sizeof(struct lttng_event_field) * nb_fields);
3551 if (ret < 0) {
3552 free(fields);
3553 goto setup_error;
3554 }
3555
3556 /* Copy event list into message payload */
3557 memcpy(cmd_ctx->llm->payload, fields,
3558 sizeof(struct lttng_event_field) * nb_fields);
3559
3560 free(fields);
3561
3562 ret = LTTNG_OK;
3563 break;
3564 }
3565 case LTTNG_LIST_SYSCALLS:
3566 {
3567 struct lttng_event *events;
3568 ssize_t nb_events;
3569
3570 nb_events = cmd_list_syscalls(&events);
3571 if (nb_events < 0) {
3572 /* Return value is a negative lttng_error_code. */
3573 ret = -nb_events;
3574 goto error;
3575 }
3576
3577 /*
3578 * Setup lttng message with payload size set to the event list size in
3579 * bytes and then copy list into the llm payload.
3580 */
3581 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3582 if (ret < 0) {
3583 free(events);
3584 goto setup_error;
3585 }
3586
3587 /* Copy event list into message payload */
3588 memcpy(cmd_ctx->llm->payload, events,
3589 sizeof(struct lttng_event) * nb_events);
3590
3591 free(events);
3592
3593 ret = LTTNG_OK;
3594 break;
3595 }
3596 case LTTNG_LIST_TRACKER_PIDS:
3597 {
3598 int32_t *pids = NULL;
3599 ssize_t nr_pids;
3600
3601 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3602 cmd_ctx->lsm->domain.type, &pids);
3603 if (nr_pids < 0) {
3604 /* Return value is a negative lttng_error_code. */
3605 ret = -nr_pids;
3606 goto error;
3607 }
3608
3609 /*
3610 * Setup lttng message with payload size set to the event list size in
3611 * bytes and then copy list into the llm payload.
3612 */
3613 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3614 if (ret < 0) {
3615 free(pids);
3616 goto setup_error;
3617 }
3618
3619 /* Copy event list into message payload */
3620 memcpy(cmd_ctx->llm->payload, pids,
3621 sizeof(int) * nr_pids);
3622
3623 free(pids);
3624
3625 ret = LTTNG_OK;
3626 break;
3627 }
3628 case LTTNG_SET_CONSUMER_URI:
3629 {
3630 size_t nb_uri, len;
3631 struct lttng_uri *uris;
3632
3633 nb_uri = cmd_ctx->lsm->u.uri.size;
3634 len = nb_uri * sizeof(struct lttng_uri);
3635
3636 if (nb_uri == 0) {
3637 ret = LTTNG_ERR_INVALID;
3638 goto error;
3639 }
3640
3641 uris = zmalloc(len);
3642 if (uris == NULL) {
3643 ret = LTTNG_ERR_FATAL;
3644 goto error;
3645 }
3646
3647 /* Receive variable len data */
3648 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3649 ret = lttcomm_recv_unix_sock(sock, uris, len);
3650 if (ret <= 0) {
3651 DBG("No URIs received from client... continuing");
3652 *sock_error = 1;
3653 ret = LTTNG_ERR_SESSION_FAIL;
3654 free(uris);
3655 goto error;
3656 }
3657
3658 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3659 free(uris);
3660 if (ret != LTTNG_OK) {
3661 goto error;
3662 }
3663
3664
3665 break;
3666 }
3667 case LTTNG_START_TRACE:
3668 {
3669 ret = cmd_start_trace(cmd_ctx->session);
3670 break;
3671 }
3672 case LTTNG_STOP_TRACE:
3673 {
3674 ret = cmd_stop_trace(cmd_ctx->session);
3675 break;
3676 }
3677 case LTTNG_CREATE_SESSION:
3678 {
3679 size_t nb_uri, len;
3680 struct lttng_uri *uris = NULL;
3681
3682 nb_uri = cmd_ctx->lsm->u.uri.size;
3683 len = nb_uri * sizeof(struct lttng_uri);
3684
3685 if (nb_uri > 0) {
3686 uris = zmalloc(len);
3687 if (uris == NULL) {
3688 ret = LTTNG_ERR_FATAL;
3689 goto error;
3690 }
3691
3692 /* Receive variable len data */
3693 DBG("Waiting for %zu URIs from client ...", nb_uri);
3694 ret = lttcomm_recv_unix_sock(sock, uris, len);
3695 if (ret <= 0) {
3696 DBG("No URIs received from client... continuing");
3697 *sock_error = 1;
3698 ret = LTTNG_ERR_SESSION_FAIL;
3699 free(uris);
3700 goto error;
3701 }
3702
3703 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3704 DBG("Creating session with ONE network URI is a bad call");
3705 ret = LTTNG_ERR_SESSION_FAIL;
3706 free(uris);
3707 goto error;
3708 }
3709 }
3710
3711 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3712 &cmd_ctx->creds, 0);
3713
3714 free(uris);
3715
3716 break;
3717 }
3718 case LTTNG_DESTROY_SESSION:
3719 {
3720 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3721
3722 /* Set session to NULL so we do not unlock it after free. */
3723 cmd_ctx->session = NULL;
3724 break;
3725 }
3726 case LTTNG_LIST_DOMAINS:
3727 {
3728 ssize_t nb_dom;
3729 struct lttng_domain *domains = NULL;
3730
3731 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3732 if (nb_dom < 0) {
3733 /* Return value is a negative lttng_error_code. */
3734 ret = -nb_dom;
3735 goto error;
3736 }
3737
3738 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3739 if (ret < 0) {
3740 free(domains);
3741 goto setup_error;
3742 }
3743
3744 /* Copy event list into message payload */
3745 memcpy(cmd_ctx->llm->payload, domains,
3746 nb_dom * sizeof(struct lttng_domain));
3747
3748 free(domains);
3749
3750 ret = LTTNG_OK;
3751 break;
3752 }
3753 case LTTNG_LIST_CHANNELS:
3754 {
3755 int nb_chan;
3756 struct lttng_channel *channels = NULL;
3757
3758 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3759 cmd_ctx->session, &channels);
3760 if (nb_chan < 0) {
3761 /* Return value is a negative lttng_error_code. */
3762 ret = -nb_chan;
3763 goto error;
3764 }
3765
3766 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3767 if (ret < 0) {
3768 free(channels);
3769 goto setup_error;
3770 }
3771
3772 /* Copy event list into message payload */
3773 memcpy(cmd_ctx->llm->payload, channels,
3774 nb_chan * sizeof(struct lttng_channel));
3775
3776 free(channels);
3777
3778 ret = LTTNG_OK;
3779 break;
3780 }
3781 case LTTNG_LIST_EVENTS:
3782 {
3783 ssize_t nb_event;
3784 struct lttng_event *events = NULL;
3785
3786 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3787 cmd_ctx->lsm->u.list.channel_name, &events);
3788 if (nb_event < 0) {
3789 /* Return value is a negative lttng_error_code. */
3790 ret = -nb_event;
3791 goto error;
3792 }
3793
3794 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3795 if (ret < 0) {
3796 free(events);
3797 goto setup_error;
3798 }
3799
3800 /* Copy event list into message payload */
3801 memcpy(cmd_ctx->llm->payload, events,
3802 nb_event * sizeof(struct lttng_event));
3803
3804 free(events);
3805
3806 ret = LTTNG_OK;
3807 break;
3808 }
3809 case LTTNG_LIST_SESSIONS:
3810 {
3811 unsigned int nr_sessions;
3812
3813 session_lock_list();
3814 nr_sessions = lttng_sessions_count(
3815 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3816 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3817
3818 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3819 if (ret < 0) {
3820 session_unlock_list();
3821 goto setup_error;
3822 }
3823
3824 /* Filled the session array */
3825 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3826 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3827 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3828
3829 session_unlock_list();
3830
3831 ret = LTTNG_OK;
3832 break;
3833 }
3834 case LTTNG_CALIBRATE:
3835 {
3836 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3837 &cmd_ctx->lsm->u.calibrate);
3838 break;
3839 }
3840 case LTTNG_REGISTER_CONSUMER:
3841 {
3842 struct consumer_data *cdata;
3843
3844 switch (cmd_ctx->lsm->domain.type) {
3845 case LTTNG_DOMAIN_KERNEL:
3846 cdata = &kconsumer_data;
3847 break;
3848 default:
3849 ret = LTTNG_ERR_UND;
3850 goto error;
3851 }
3852
3853 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3854 cmd_ctx->lsm->u.reg.path, cdata);
3855 break;
3856 }
3857 case LTTNG_DATA_PENDING:
3858 {
3859 int pending_ret;
3860
3861 /* 1 byte to return whether or not data is pending */
3862 ret = setup_lttng_msg(cmd_ctx, 1);
3863 if (ret < 0) {
3864 goto setup_error;
3865 }
3866
3867 pending_ret = cmd_data_pending(cmd_ctx->session);
3868 /*
3869 * FIXME
3870 *
3871 * This function may returns 0 or 1 to indicate whether or not
3872 * there is data pending. In case of error, it should return an
3873 * LTTNG_ERR code. However, some code paths may still return
3874 * a nondescript error code, which we handle by returning an
3875 * "unknown" error.
3876 */
3877 if (pending_ret == 0 || pending_ret == 1) {
3878 ret = LTTNG_OK;
3879 } else if (pending_ret < 0) {
3880 ret = LTTNG_ERR_UNK;
3881 goto setup_error;
3882 } else {
3883 ret = pending_ret;
3884 goto setup_error;
3885 }
3886
3887 *cmd_ctx->llm->payload = (uint8_t) pending_ret;
3888 break;
3889 }
3890 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3891 {
3892 struct lttcomm_lttng_output_id reply;
3893
3894 ret = cmd_snapshot_add_output(cmd_ctx->session,
3895 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3896 if (ret != LTTNG_OK) {
3897 goto error;
3898 }
3899
3900 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3901 if (ret < 0) {
3902 goto setup_error;
3903 }
3904
3905 /* Copy output list into message payload */
3906 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3907 ret = LTTNG_OK;
3908 break;
3909 }
3910 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3911 {
3912 ret = cmd_snapshot_del_output(cmd_ctx->session,
3913 &cmd_ctx->lsm->u.snapshot_output.output);
3914 break;
3915 }
3916 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3917 {
3918 ssize_t nb_output;
3919 struct lttng_snapshot_output *outputs = NULL;
3920
3921 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3922 if (nb_output < 0) {
3923 ret = -nb_output;
3924 goto error;
3925 }
3926
3927 ret = setup_lttng_msg(cmd_ctx,
3928 nb_output * sizeof(struct lttng_snapshot_output));
3929 if (ret < 0) {
3930 free(outputs);
3931 goto setup_error;
3932 }
3933
3934 if (outputs) {
3935 /* Copy output list into message payload */
3936 memcpy(cmd_ctx->llm->payload, outputs,
3937 nb_output * sizeof(struct lttng_snapshot_output));
3938 free(outputs);
3939 }
3940
3941 ret = LTTNG_OK;
3942 break;
3943 }
3944 case LTTNG_SNAPSHOT_RECORD:
3945 {
3946 ret = cmd_snapshot_record(cmd_ctx->session,
3947 &cmd_ctx->lsm->u.snapshot_record.output,
3948 cmd_ctx->lsm->u.snapshot_record.wait);
3949 break;
3950 }
3951 case LTTNG_CREATE_SESSION_SNAPSHOT:
3952 {
3953 size_t nb_uri, len;
3954 struct lttng_uri *uris = NULL;
3955
3956 nb_uri = cmd_ctx->lsm->u.uri.size;
3957 len = nb_uri * sizeof(struct lttng_uri);
3958
3959 if (nb_uri > 0) {
3960 uris = zmalloc(len);
3961 if (uris == NULL) {
3962 ret = LTTNG_ERR_FATAL;
3963 goto error;
3964 }
3965
3966 /* Receive variable len data */
3967 DBG("Waiting for %zu URIs from client ...", nb_uri);
3968 ret = lttcomm_recv_unix_sock(sock, uris, len);
3969 if (ret <= 0) {
3970 DBG("No URIs received from client... continuing");
3971 *sock_error = 1;
3972 ret = LTTNG_ERR_SESSION_FAIL;
3973 free(uris);
3974 goto error;
3975 }
3976
3977 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3978 DBG("Creating session with ONE network URI is a bad call");
3979 ret = LTTNG_ERR_SESSION_FAIL;
3980 free(uris);
3981 goto error;
3982 }
3983 }
3984
3985 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3986 nb_uri, &cmd_ctx->creds);
3987 free(uris);
3988 break;
3989 }
3990 case LTTNG_CREATE_SESSION_LIVE:
3991 {
3992 size_t nb_uri, len;
3993 struct lttng_uri *uris = NULL;
3994
3995 nb_uri = cmd_ctx->lsm->u.uri.size;
3996 len = nb_uri * sizeof(struct lttng_uri);
3997
3998 if (nb_uri > 0) {
3999 uris = zmalloc(len);
4000 if (uris == NULL) {
4001 ret = LTTNG_ERR_FATAL;
4002 goto error;
4003 }
4004
4005 /* Receive variable len data */
4006 DBG("Waiting for %zu URIs from client ...", nb_uri);
4007 ret = lttcomm_recv_unix_sock(sock, uris, len);
4008 if (ret <= 0) {
4009 DBG("No URIs received from client... continuing");
4010 *sock_error = 1;
4011 ret = LTTNG_ERR_SESSION_FAIL;
4012 free(uris);
4013 goto error;
4014 }
4015
4016 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4017 DBG("Creating session with ONE network URI is a bad call");
4018 ret = LTTNG_ERR_SESSION_FAIL;
4019 free(uris);
4020 goto error;
4021 }
4022 }
4023
4024 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4025 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4026 free(uris);
4027 break;
4028 }
4029 case LTTNG_SAVE_SESSION:
4030 {
4031 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4032 &cmd_ctx->creds);
4033 break;
4034 }
4035 case LTTNG_SET_SESSION_SHM_PATH:
4036 {
4037 ret = cmd_set_session_shm_path(cmd_ctx->session,
4038 cmd_ctx->lsm->u.set_shm_path.shm_path);
4039 break;
4040 }
4041 default:
4042 ret = LTTNG_ERR_UND;
4043 break;
4044 }
4045
4046 error:
4047 if (cmd_ctx->llm == NULL) {
4048 DBG("Missing llm structure. Allocating one.");
4049 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
4050 goto setup_error;
4051 }
4052 }
4053 /* Set return code */
4054 cmd_ctx->llm->ret_code = ret;
4055 setup_error:
4056 if (cmd_ctx->session) {
4057 session_unlock(cmd_ctx->session);
4058 }
4059 if (need_tracing_session) {
4060 session_unlock_list();
4061 }
4062 init_setup_error:
4063 return ret;
4064 }
4065
4066 /*
4067 * Thread managing health check socket.
4068 */
4069 static void *thread_manage_health(void *data)
4070 {
4071 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4072 uint32_t revents, nb_fd;
4073 struct lttng_poll_event events;
4074 struct health_comm_msg msg;
4075 struct health_comm_reply reply;
4076
4077 DBG("[thread] Manage health check started");
4078
4079 rcu_register_thread();
4080
4081 /* We might hit an error path before this is created. */
4082 lttng_poll_init(&events);
4083
4084 /* Create unix socket */
4085 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4086 if (sock < 0) {
4087 ERR("Unable to create health check Unix socket");
4088 ret = -1;
4089 goto error;
4090 }
4091
4092 if (is_root) {
4093 /* lttng health client socket path permissions */
4094 ret = chown(health_unix_sock_path, 0,
4095 utils_get_group_id(tracing_group_name));
4096 if (ret < 0) {
4097 ERR("Unable to set group on %s", health_unix_sock_path);
4098 PERROR("chown");
4099 ret = -1;
4100 goto error;
4101 }
4102
4103 ret = chmod(health_unix_sock_path,
4104 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4105 if (ret < 0) {
4106 ERR("Unable to set permissions on %s", health_unix_sock_path);
4107 PERROR("chmod");
4108 ret = -1;
4109 goto error;
4110 }
4111 }
4112
4113 /*
4114 * Set the CLOEXEC flag. Return code is useless because either way, the
4115 * show must go on.
4116 */
4117 (void) utils_set_fd_cloexec(sock);
4118
4119 ret = lttcomm_listen_unix_sock(sock);
4120 if (ret < 0) {
4121 goto error;
4122 }
4123
4124 /*
4125 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4126 * more will be added to this poll set.
4127 */
4128 ret = sessiond_set_thread_pollset(&events, 2);
4129 if (ret < 0) {
4130 goto error;
4131 }
4132
4133 /* Add the application registration socket */
4134 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4135 if (ret < 0) {
4136 goto error;
4137 }
4138
4139 sessiond_notify_ready();
4140
4141 while (1) {
4142 DBG("Health check ready");
4143
4144 /* Inifinite blocking call, waiting for transmission */
4145 restart:
4146 ret = lttng_poll_wait(&events, -1);
4147 if (ret < 0) {
4148 /*
4149 * Restart interrupted system call.
4150 */
4151 if (errno == EINTR) {
4152 goto restart;
4153 }
4154 goto error;
4155 }
4156
4157 nb_fd = ret;
4158
4159 for (i = 0; i < nb_fd; i++) {
4160 /* Fetch once the poll data */
4161 revents = LTTNG_POLL_GETEV(&events, i);
4162 pollfd = LTTNG_POLL_GETFD(&events, i);
4163
4164 if (!revents) {
4165 /* No activity for this FD (poll implementation). */
4166 continue;
4167 }
4168
4169 /* Thread quit pipe has been closed. Killing thread. */
4170 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4171 if (ret) {
4172 err = 0;
4173 goto exit;
4174 }
4175
4176 /* Event on the registration socket */
4177 if (pollfd == sock) {
4178 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4179 ERR("Health socket poll error");
4180 goto error;
4181 }
4182 }
4183 }
4184
4185 new_sock = lttcomm_accept_unix_sock(sock);
4186 if (new_sock < 0) {
4187 goto error;
4188 }
4189
4190 /*
4191 * Set the CLOEXEC flag. Return code is useless because either way, the
4192 * show must go on.
4193 */
4194 (void) utils_set_fd_cloexec(new_sock);
4195
4196 DBG("Receiving data from client for health...");
4197 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4198 if (ret <= 0) {
4199 DBG("Nothing recv() from client... continuing");
4200 ret = close(new_sock);
4201 if (ret) {
4202 PERROR("close");
4203 }
4204 new_sock = -1;
4205 continue;
4206 }
4207
4208 rcu_thread_online();
4209
4210 memset(&reply, 0, sizeof(reply));
4211 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4212 /*
4213 * health_check_state returns 0 if health is
4214 * bad.
4215 */
4216 if (!health_check_state(health_sessiond, i)) {
4217 reply.ret_code |= 1ULL << i;
4218 }
4219 }
4220
4221 DBG2("Health check return value %" PRIx64, reply.ret_code);
4222
4223 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4224 if (ret < 0) {
4225 ERR("Failed to send health data back to client");
4226 }
4227
4228 /* End of transmission */
4229 ret = close(new_sock);
4230 if (ret) {
4231 PERROR("close");
4232 }
4233 new_sock = -1;
4234 }
4235
4236 exit:
4237 error:
4238 if (err) {
4239 ERR("Health error occurred in %s", __func__);
4240 }
4241 DBG("Health check thread dying");
4242 unlink(health_unix_sock_path);
4243 if (sock >= 0) {
4244 ret = close(sock);
4245 if (ret) {
4246 PERROR("close");
4247 }
4248 }
4249
4250 lttng_poll_clean(&events);
4251
4252 rcu_unregister_thread();
4253 return NULL;
4254 }
4255
4256 /*
4257 * This thread manage all clients request using the unix client socket for
4258 * communication.
4259 */
4260 static void *thread_manage_clients(void *data)
4261 {
4262 int sock = -1, ret, i, pollfd, err = -1;
4263 int sock_error;
4264 uint32_t revents, nb_fd;
4265 struct command_ctx *cmd_ctx = NULL;
4266 struct lttng_poll_event events;
4267
4268 DBG("[thread] Manage client started");
4269
4270 rcu_register_thread();
4271
4272 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4273
4274 health_code_update();
4275
4276 ret = lttcomm_listen_unix_sock(client_sock);
4277 if (ret < 0) {
4278 goto error_listen;
4279 }
4280
4281 /*
4282 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4283 * more will be added to this poll set.
4284 */
4285 ret = sessiond_set_thread_pollset(&events, 2);
4286 if (ret < 0) {
4287 goto error_create_poll;
4288 }
4289
4290 /* Add the application registration socket */
4291 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4292 if (ret < 0) {
4293 goto error;
4294 }
4295
4296 sessiond_notify_ready();
4297 ret = sem_post(&load_info->message_thread_ready);
4298 if (ret) {
4299 PERROR("sem_post message_thread_ready");
4300 goto error;
4301 }
4302
4303 /* This testpoint is after we signal readiness to the parent. */
4304 if (testpoint(sessiond_thread_manage_clients)) {
4305 goto error;
4306 }
4307
4308 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4309 goto error;
4310 }
4311
4312 health_code_update();
4313
4314 while (1) {
4315 DBG("Accepting client command ...");
4316
4317 /* Inifinite blocking call, waiting for transmission */
4318 restart:
4319 health_poll_entry();
4320 ret = lttng_poll_wait(&events, -1);
4321 health_poll_exit();
4322 if (ret < 0) {
4323 /*
4324 * Restart interrupted system call.
4325 */
4326 if (errno == EINTR) {
4327 goto restart;
4328 }
4329 goto error;
4330 }
4331
4332 nb_fd = ret;
4333
4334 for (i = 0; i < nb_fd; i++) {
4335 /* Fetch once the poll data */
4336 revents = LTTNG_POLL_GETEV(&events, i);
4337 pollfd = LTTNG_POLL_GETFD(&events, i);
4338
4339 health_code_update();
4340
4341 if (!revents) {
4342 /* No activity for this FD (poll implementation). */
4343 continue;
4344 }
4345
4346 /* Thread quit pipe has been closed. Killing thread. */
4347 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4348 if (ret) {
4349 err = 0;
4350 goto exit;
4351 }
4352
4353 /* Event on the registration socket */
4354 if (pollfd == client_sock) {
4355 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4356 ERR("Client socket poll error");
4357 goto error;
4358 }
4359 }
4360 }
4361
4362 DBG("Wait for client response");
4363
4364 health_code_update();
4365
4366 sock = lttcomm_accept_unix_sock(client_sock);
4367 if (sock < 0) {
4368 goto error;
4369 }
4370
4371 /*
4372 * Set the CLOEXEC flag. Return code is useless because either way, the
4373 * show must go on.
4374 */
4375 (void) utils_set_fd_cloexec(sock);
4376
4377 /* Set socket option for credentials retrieval */
4378 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4379 if (ret < 0) {
4380 goto error;
4381 }
4382
4383 /* Allocate context command to process the client request */
4384 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4385 if (cmd_ctx == NULL) {
4386 PERROR("zmalloc cmd_ctx");
4387 goto error;
4388 }
4389
4390 /* Allocate data buffer for reception */
4391 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4392 if (cmd_ctx->lsm == NULL) {
4393 PERROR("zmalloc cmd_ctx->lsm");
4394 goto error;
4395 }
4396
4397 cmd_ctx->llm = NULL;
4398 cmd_ctx->session = NULL;
4399
4400 health_code_update();
4401
4402 /*
4403 * Data is received from the lttng client. The struct
4404 * lttcomm_session_msg (lsm) contains the command and data request of
4405 * the client.
4406 */
4407 DBG("Receiving data from client ...");
4408 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4409 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4410 if (ret <= 0) {
4411 DBG("Nothing recv() from client... continuing");
4412 ret = close(sock);
4413 if (ret) {
4414 PERROR("close");
4415 }
4416 sock = -1;
4417 clean_command_ctx(&cmd_ctx);
4418 continue;
4419 }
4420
4421 health_code_update();
4422
4423 // TODO: Validate cmd_ctx including sanity check for
4424 // security purpose.
4425
4426 rcu_thread_online();
4427 /*
4428 * This function dispatch the work to the kernel or userspace tracer
4429 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4430 * informations for the client. The command context struct contains
4431 * everything this function may needs.
4432 */
4433 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4434 rcu_thread_offline();
4435 if (ret < 0) {
4436 ret = close(sock);
4437 if (ret) {
4438 PERROR("close");
4439 }
4440 sock = -1;
4441 /*
4442 * TODO: Inform client somehow of the fatal error. At
4443 * this point, ret < 0 means that a zmalloc failed
4444 * (ENOMEM). Error detected but still accept
4445 * command, unless a socket error has been
4446 * detected.
4447 */
4448 clean_command_ctx(&cmd_ctx);
4449 continue;
4450 }
4451
4452 health_code_update();
4453
4454 DBG("Sending response (size: %d, retcode: %s (%d))",
4455 cmd_ctx->lttng_msg_size,
4456 lttng_strerror(-cmd_ctx->llm->ret_code),
4457 cmd_ctx->llm->ret_code);
4458 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4459 if (ret < 0) {
4460 ERR("Failed to send data back to client");
4461 }
4462
4463 /* End of transmission */
4464 ret = close(sock);
4465 if (ret) {
4466 PERROR("close");
4467 }
4468 sock = -1;
4469
4470 clean_command_ctx(&cmd_ctx);
4471
4472 health_code_update();
4473 }
4474
4475 exit:
4476 error:
4477 if (sock >= 0) {
4478 ret = close(sock);
4479 if (ret) {
4480 PERROR("close");
4481 }
4482 }
4483
4484 lttng_poll_clean(&events);
4485 clean_command_ctx(&cmd_ctx);
4486
4487 error_listen:
4488 error_create_poll:
4489 unlink(client_unix_sock_path);
4490 if (client_sock >= 0) {
4491 ret = close(client_sock);
4492 if (ret) {
4493 PERROR("close");
4494 }
4495 }
4496
4497 if (err) {
4498 health_error();
4499 ERR("Health error occurred in %s", __func__);
4500 }
4501
4502 health_unregister(health_sessiond);
4503
4504 DBG("Client thread dying");
4505
4506 rcu_unregister_thread();
4507
4508 /*
4509 * Since we are creating the consumer threads, we own them, so we need
4510 * to join them before our thread exits.
4511 */
4512 ret = join_consumer_thread(&kconsumer_data);
4513 if (ret) {
4514 errno = ret;
4515 PERROR("join_consumer");
4516 }
4517
4518 ret = join_consumer_thread(&ustconsumer32_data);
4519 if (ret) {
4520 errno = ret;
4521 PERROR("join_consumer ust32");
4522 }
4523
4524 ret = join_consumer_thread(&ustconsumer64_data);
4525 if (ret) {
4526 errno = ret;
4527 PERROR("join_consumer ust64");
4528 }
4529 return NULL;
4530 }
4531
4532
4533 /*
4534 * usage function on stderr
4535 */
4536 static void usage(void)
4537 {
4538 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4539 fprintf(stderr, " -h, --help Display this usage.\n");
4540 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4541 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4542 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4543 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4544 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4545 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4546 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4547 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4548 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4549 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4550 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4551 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4552 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4553 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4554 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4555 fprintf(stderr, " -V, --version Show version number.\n");
4556 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4557 fprintf(stderr, " -q, --quiet No output at all.\n");
4558 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4559 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4560 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4561 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4562 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4563 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4564 fprintf(stderr, " -l --load PATH Load session configuration\n");
4565 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4566 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4567 }
4568
4569 static int string_match(const char *str1, const char *str2)
4570 {
4571 return (str1 && str2) && !strcmp(str1, str2);
4572 }
4573
4574 /*
4575 * Take an option from the getopt output and set it in the right variable to be
4576 * used later.
4577 *
4578 * Return 0 on success else a negative value.
4579 */
4580 static int set_option(int opt, const char *arg, const char *optname)
4581 {
4582 int ret = 0;
4583
4584 if (arg && arg[0] == '\0') {
4585 /*
4586 * This only happens if the value is read from daemon config
4587 * file. This means the option requires an argument and the
4588 * configuration file contains a line such as:
4589 * my_option =
4590 */
4591 ret = -EINVAL;
4592 goto end;
4593 }
4594
4595 if (string_match(optname, "client-sock") || opt == 'c') {
4596 if (lttng_is_setuid_setgid()) {
4597 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4598 "-c, --client-sock");
4599 } else {
4600 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4601 }
4602 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4603 if (lttng_is_setuid_setgid()) {
4604 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4605 "-a, --apps-sock");
4606 } else {
4607 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4608 }
4609 } else if (string_match(optname, "daemonize") || opt == 'd') {
4610 opt_daemon = 1;
4611 } else if (string_match(optname, "background") || opt == 'b') {
4612 opt_background = 1;
4613 } else if (string_match(optname, "group") || opt == 'g') {
4614 if (lttng_is_setuid_setgid()) {
4615 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4616 "-g, --group");
4617 } else {
4618 /*
4619 * If the override option is set, the pointer points to a
4620 * *non* const thus freeing it even though the variable type is
4621 * set to const.
4622 */
4623 if (tracing_group_name_override) {
4624 free((void *) tracing_group_name);
4625 }
4626 tracing_group_name = strdup(arg);
4627 if (!tracing_group_name) {
4628 PERROR("strdup");
4629 ret = -ENOMEM;
4630 }
4631 tracing_group_name_override = 1;
4632 }
4633 } else if (string_match(optname, "help") || opt == 'h') {
4634 usage();
4635 exit(EXIT_SUCCESS);
4636 } else if (string_match(optname, "version") || opt == 'V') {
4637 fprintf(stdout, "%s\n", VERSION);
4638 exit(EXIT_SUCCESS);
4639 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4640 opt_sig_parent = 1;
4641 } else if (string_match(optname, "kconsumerd-err-sock")) {
4642 if (lttng_is_setuid_setgid()) {
4643 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4644 "--kconsumerd-err-sock");
4645 } else {
4646 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4647 }
4648 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4649 if (lttng_is_setuid_setgid()) {
4650 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4651 "--kconsumerd-cmd-sock");
4652 } else {
4653 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4654 }
4655 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4656 if (lttng_is_setuid_setgid()) {
4657 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4658 "--ustconsumerd64-err-sock");
4659 } else {
4660 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4661 }
4662 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4663 if (lttng_is_setuid_setgid()) {
4664 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4665 "--ustconsumerd64-cmd-sock");
4666 } else {
4667 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4668 }
4669 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4670 if (lttng_is_setuid_setgid()) {
4671 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4672 "--ustconsumerd32-err-sock");
4673 } else {
4674 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4675 }
4676 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4677 if (lttng_is_setuid_setgid()) {
4678 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4679 "--ustconsumerd32-cmd-sock");
4680 } else {
4681 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4682 }
4683 } else if (string_match(optname, "no-kernel")) {
4684 opt_no_kernel = 1;
4685 } else if (string_match(optname, "quiet") || opt == 'q') {
4686 lttng_opt_quiet = 1;
4687 } else if (string_match(optname, "verbose") || opt == 'v') {
4688 /* Verbose level can increase using multiple -v */
4689 if (arg) {
4690 /* Value obtained from config file */
4691 lttng_opt_verbose = config_parse_value(arg);
4692 } else {
4693 /* -v used on command line */
4694 lttng_opt_verbose++;
4695 }
4696 /* Clamp value to [0, 3] */
4697 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4698 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4699 } else if (string_match(optname, "verbose-consumer")) {
4700 if (arg) {
4701 opt_verbose_consumer = config_parse_value(arg);
4702 } else {
4703 opt_verbose_consumer += 1;
4704 }
4705 } else if (string_match(optname, "consumerd32-path")) {
4706 if (lttng_is_setuid_setgid()) {
4707 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4708 "--consumerd32-path");
4709 } else {
4710 if (consumerd32_bin_override) {
4711 free((void *) consumerd32_bin);
4712 }
4713 consumerd32_bin = strdup(arg);
4714 if (!consumerd32_bin) {
4715 PERROR("strdup");
4716 ret = -ENOMEM;
4717 }
4718 consumerd32_bin_override = 1;
4719 }
4720 } else if (string_match(optname, "consumerd32-libdir")) {
4721 if (lttng_is_setuid_setgid()) {
4722 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4723 "--consumerd32-libdir");
4724 } else {
4725 if (consumerd32_libdir_override) {
4726 free((void *) consumerd32_libdir);
4727 }
4728 consumerd32_libdir = strdup(arg);
4729 if (!consumerd32_libdir) {
4730 PERROR("strdup");
4731 ret = -ENOMEM;
4732 }
4733 consumerd32_libdir_override = 1;
4734 }
4735 } else if (string_match(optname, "consumerd64-path")) {
4736 if (lttng_is_setuid_setgid()) {
4737 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4738 "--consumerd64-path");
4739 } else {
4740 if (consumerd64_bin_override) {
4741 free((void *) consumerd64_bin);
4742 }
4743 consumerd64_bin = strdup(arg);
4744 if (!consumerd64_bin) {
4745 PERROR("strdup");
4746 ret = -ENOMEM;
4747 }
4748 consumerd64_bin_override = 1;
4749 }
4750 } else if (string_match(optname, "consumerd64-libdir")) {
4751 if (lttng_is_setuid_setgid()) {
4752 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4753 "--consumerd64-libdir");
4754 } else {
4755 if (consumerd64_libdir_override) {
4756 free((void *) consumerd64_libdir);
4757 }
4758 consumerd64_libdir = strdup(arg);
4759 if (!consumerd64_libdir) {
4760 PERROR("strdup");
4761 ret = -ENOMEM;
4762 }
4763 consumerd64_libdir_override = 1;
4764 }
4765 } else if (string_match(optname, "pidfile") || opt == 'p') {
4766 if (lttng_is_setuid_setgid()) {
4767 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4768 "-p, --pidfile");
4769 } else {
4770 free(opt_pidfile);
4771 opt_pidfile = strdup(arg);
4772 if (!opt_pidfile) {
4773 PERROR("strdup");
4774 ret = -ENOMEM;
4775 }
4776 }
4777 } else if (string_match(optname, "agent-tcp-port")) {
4778 if (lttng_is_setuid_setgid()) {
4779 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4780 "--agent-tcp-port");
4781 } else {
4782 unsigned long v;
4783
4784 if (!arg) {
4785 ret = -EINVAL;
4786 goto end;
4787 }
4788 errno = 0;
4789 v = strtoul(arg, NULL, 0);
4790 if (errno != 0 || !isdigit(arg[0])) {
4791 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4792 return -1;
4793 }
4794 if (v == 0 || v >= 65535) {
4795 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4796 return -1;
4797 }
4798 agent_tcp_port = (uint32_t) v;
4799 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4800 }
4801 } else if (string_match(optname, "load") || opt == 'l') {
4802 if (lttng_is_setuid_setgid()) {
4803 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4804 "-l, --load");
4805 } else {
4806 free(opt_load_session_path);
4807 opt_load_session_path = strdup(arg);
4808 if (!opt_load_session_path) {
4809 PERROR("strdup");
4810 ret = -ENOMEM;
4811 }
4812 }
4813 } else if (string_match(optname, "kmod-probes")) {
4814 if (lttng_is_setuid_setgid()) {
4815 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4816 "--kmod-probes");
4817 } else {
4818 free(kmod_probes_list);
4819 kmod_probes_list = strdup(arg);
4820 if (!kmod_probes_list) {
4821 PERROR("strdup");
4822 ret = -ENOMEM;
4823 }
4824 }
4825 } else if (string_match(optname, "extra-kmod-probes")) {
4826 if (lttng_is_setuid_setgid()) {
4827 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4828 "--extra-kmod-probes");
4829 } else {
4830 free(kmod_extra_probes_list);
4831 kmod_extra_probes_list = strdup(arg);
4832 if (!kmod_extra_probes_list) {
4833 PERROR("strdup");
4834 ret = -ENOMEM;
4835 }
4836 }
4837 } else if (string_match(optname, "config") || opt == 'f') {
4838 /* This is handled in set_options() thus silent skip. */
4839 goto end;
4840 } else {
4841 /* Unknown option or other error.
4842 * Error is printed by getopt, just return */
4843 ret = -1;
4844 }
4845
4846 end:
4847 if (ret == -EINVAL) {
4848 const char *opt_name = "unknown";
4849 int i;
4850
4851 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4852 i++) {
4853 if (opt == long_options[i].val) {
4854 opt_name = long_options[i].name;
4855 break;
4856 }
4857 }
4858
4859 WARN("Invalid argument provided for option \"%s\", using default value.",
4860 opt_name);
4861 }
4862
4863 return ret;
4864 }
4865
4866 /*
4867 * config_entry_handler_cb used to handle options read from a config file.
4868 * See config_entry_handler_cb comment in common/config/config.h for the
4869 * return value conventions.
4870 */
4871 static int config_entry_handler(const struct config_entry *entry, void *unused)
4872 {
4873 int ret = 0, i;
4874
4875 if (!entry || !entry->name || !entry->value) {
4876 ret = -EINVAL;
4877 goto end;
4878 }
4879
4880 /* Check if the option is to be ignored */
4881 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4882 if (!strcmp(entry->name, config_ignore_options[i])) {
4883 goto end;
4884 }
4885 }
4886
4887 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4888 i++) {
4889
4890 /* Ignore if not fully matched. */
4891 if (strcmp(entry->name, long_options[i].name)) {
4892 continue;
4893 }
4894
4895 /*
4896 * If the option takes no argument on the command line, we have to
4897 * check if the value is "true". We support non-zero numeric values,
4898 * true, on and yes.
4899 */
4900 if (!long_options[i].has_arg) {
4901 ret = config_parse_value(entry->value);
4902 if (ret <= 0) {
4903 if (ret) {
4904 WARN("Invalid configuration value \"%s\" for option %s",
4905 entry->value, entry->name);
4906 }
4907 /* False, skip boolean config option. */
4908 goto end;
4909 }
4910 }
4911
4912 ret = set_option(long_options[i].val, entry->value, entry->name);
4913 goto end;
4914 }
4915
4916 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4917
4918 end:
4919 return ret;
4920 }
4921
4922 /*
4923 * daemon configuration loading and argument parsing
4924 */
4925 static int set_options(int argc, char **argv)
4926 {
4927 int ret = 0, c = 0, option_index = 0;
4928 int orig_optopt = optopt, orig_optind = optind;
4929 char *optstring;
4930 const char *config_path = NULL;
4931
4932 optstring = utils_generate_optstring(long_options,
4933 sizeof(long_options) / sizeof(struct option));
4934 if (!optstring) {
4935 ret = -ENOMEM;
4936 goto end;
4937 }
4938
4939 /* Check for the --config option */
4940 while ((c = getopt_long(argc, argv, optstring, long_options,
4941 &option_index)) != -1) {
4942 if (c == '?') {
4943 ret = -EINVAL;
4944 goto end;
4945 } else if (c != 'f') {
4946 /* if not equal to --config option. */
4947 continue;
4948 }
4949
4950 if (lttng_is_setuid_setgid()) {
4951 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4952 "-f, --config");
4953 } else {
4954 config_path = utils_expand_path(optarg);
4955 if (!config_path) {
4956 ERR("Failed to resolve path: %s", optarg);
4957 }
4958 }
4959 }
4960
4961 ret = config_get_section_entries(config_path, config_section_name,
4962 config_entry_handler, NULL);
4963 if (ret) {
4964 if (ret > 0) {
4965 ERR("Invalid configuration option at line %i", ret);
4966 ret = -1;
4967 }
4968 goto end;
4969 }
4970
4971 /* Reset getopt's global state */
4972 optopt = orig_optopt;
4973 optind = orig_optind;
4974 while (1) {
4975 option_index = -1;
4976 /*
4977 * getopt_long() will not set option_index if it encounters a
4978 * short option.
4979 */
4980 c = getopt_long(argc, argv, optstring, long_options,
4981 &option_index);
4982 if (c == -1) {
4983 break;
4984 }
4985
4986 /*
4987 * Pass NULL as the long option name if popt left the index
4988 * unset.
4989 */
4990 ret = set_option(c, optarg,
4991 option_index < 0 ? NULL :
4992 long_options[option_index].name);
4993 if (ret < 0) {
4994 break;
4995 }
4996 }
4997
4998 end:
4999 free(optstring);
5000 return ret;
5001 }
5002
5003 /*
5004 * Creates the two needed socket by the daemon.
5005 * apps_sock - The communication socket for all UST apps.
5006 * client_sock - The communication of the cli tool (lttng).
5007 */
5008 static int init_daemon_socket(void)
5009 {
5010 int ret = 0;
5011 mode_t old_umask;
5012
5013 old_umask = umask(0);
5014
5015 /* Create client tool unix socket */
5016 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5017 if (client_sock < 0) {
5018 ERR("Create unix sock failed: %s", client_unix_sock_path);
5019 ret = -1;
5020 goto end;
5021 }
5022
5023 /* Set the cloexec flag */
5024 ret = utils_set_fd_cloexec(client_sock);
5025 if (ret < 0) {
5026 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5027 "Continuing but note that the consumer daemon will have a "
5028 "reference to this socket on exec()", client_sock);
5029 }
5030
5031 /* File permission MUST be 660 */
5032 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5033 if (ret < 0) {
5034 ERR("Set file permissions failed: %s", client_unix_sock_path);
5035 PERROR("chmod");
5036 goto end;
5037 }
5038
5039 /* Create the application unix socket */
5040 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5041 if (apps_sock < 0) {
5042 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5043 ret = -1;
5044 goto end;
5045 }
5046
5047 /* Set the cloexec flag */
5048 ret = utils_set_fd_cloexec(apps_sock);
5049 if (ret < 0) {
5050 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5051 "Continuing but note that the consumer daemon will have a "
5052 "reference to this socket on exec()", apps_sock);
5053 }
5054
5055 /* File permission MUST be 666 */
5056 ret = chmod(apps_unix_sock_path,
5057 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5058 if (ret < 0) {
5059 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5060 PERROR("chmod");
5061 goto end;
5062 }
5063
5064 DBG3("Session daemon client socket %d and application socket %d created",
5065 client_sock, apps_sock);
5066
5067 end:
5068 umask(old_umask);
5069 return ret;
5070 }
5071
5072 /*
5073 * Check if the global socket is available, and if a daemon is answering at the
5074 * other side. If yes, error is returned.
5075 */
5076 static int check_existing_daemon(void)
5077 {
5078 /* Is there anybody out there ? */
5079 if (lttng_session_daemon_alive()) {
5080 return -EEXIST;
5081 }
5082
5083 return 0;
5084 }
5085
5086 /*
5087 * Set the tracing group gid onto the client socket.
5088 *
5089 * Race window between mkdir and chown is OK because we are going from more
5090 * permissive (root.root) to less permissive (root.tracing).
5091 */
5092 static int set_permissions(char *rundir)
5093 {
5094 int ret;
5095 gid_t gid;
5096
5097 gid = utils_get_group_id(tracing_group_name);
5098
5099 /* Set lttng run dir */
5100 ret = chown(rundir, 0, gid);
5101 if (ret < 0) {
5102 ERR("Unable to set group on %s", rundir);
5103 PERROR("chown");
5104 }
5105
5106 /*
5107 * Ensure all applications and tracing group can search the run
5108 * dir. Allow everyone to read the directory, since it does not
5109 * buy us anything to hide its content.
5110 */
5111 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5112 if (ret < 0) {
5113 ERR("Unable to set permissions on %s", rundir);
5114 PERROR("chmod");
5115 }
5116
5117 /* lttng client socket path */
5118 ret = chown(client_unix_sock_path, 0, gid);
5119 if (ret < 0) {
5120 ERR("Unable to set group on %s", client_unix_sock_path);
5121 PERROR("chown");
5122 }
5123
5124 /* kconsumer error socket path */
5125 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5126 if (ret < 0) {
5127 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5128 PERROR("chown");
5129 }
5130
5131 /* 64-bit ustconsumer error socket path */
5132 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5133 if (ret < 0) {
5134 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5135 PERROR("chown");
5136 }
5137
5138 /* 32-bit ustconsumer compat32 error socket path */
5139 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5140 if (ret < 0) {
5141 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5142 PERROR("chown");
5143 }
5144
5145 DBG("All permissions are set");
5146
5147 return ret;
5148 }
5149
5150 /*
5151 * Create the lttng run directory needed for all global sockets and pipe.
5152 */
5153 static int create_lttng_rundir(const char *rundir)
5154 {
5155 int ret;
5156
5157 DBG3("Creating LTTng run directory: %s", rundir);
5158
5159 ret = mkdir(rundir, S_IRWXU);
5160 if (ret < 0) {
5161 if (errno != EEXIST) {
5162 ERR("Unable to create %s", rundir);
5163 goto error;
5164 } else {
5165 ret = 0;
5166 }
5167 }
5168
5169 error:
5170 return ret;
5171 }
5172
5173 /*
5174 * Setup sockets and directory needed by the kconsumerd communication with the
5175 * session daemon.
5176 */
5177 static int set_consumer_sockets(struct consumer_data *consumer_data,
5178 const char *rundir)
5179 {
5180 int ret;
5181 char path[PATH_MAX];
5182
5183 switch (consumer_data->type) {
5184 case LTTNG_CONSUMER_KERNEL:
5185 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5186 break;
5187 case LTTNG_CONSUMER64_UST:
5188 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5189 break;
5190 case LTTNG_CONSUMER32_UST:
5191 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5192 break;
5193 default:
5194 ERR("Consumer type unknown");
5195 ret = -EINVAL;
5196 goto error;
5197 }
5198
5199 DBG2("Creating consumer directory: %s", path);
5200
5201 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5202 if (ret < 0) {
5203 if (errno != EEXIST) {
5204 PERROR("mkdir");
5205 ERR("Failed to create %s", path);
5206 goto error;
5207 }
5208 ret = -1;
5209 }
5210 if (is_root) {
5211 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5212 if (ret < 0) {
5213 ERR("Unable to set group on %s", path);
5214 PERROR("chown");
5215 goto error;
5216 }
5217 }
5218
5219 /* Create the kconsumerd error unix socket */
5220 consumer_data->err_sock =
5221 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5222 if (consumer_data->err_sock < 0) {
5223 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5224 ret = -1;
5225 goto error;
5226 }
5227
5228 /*
5229 * Set the CLOEXEC flag. Return code is useless because either way, the
5230 * show must go on.
5231 */
5232 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5233 if (ret < 0) {
5234 PERROR("utils_set_fd_cloexec");
5235 /* continue anyway */
5236 }
5237
5238 /* File permission MUST be 660 */
5239 ret = chmod(consumer_data->err_unix_sock_path,
5240 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5241 if (ret < 0) {
5242 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5243 PERROR("chmod");
5244 goto error;
5245 }
5246
5247 error:
5248 return ret;
5249 }
5250
5251 /*
5252 * Signal handler for the daemon
5253 *
5254 * Simply stop all worker threads, leaving main() return gracefully after
5255 * joining all threads and calling cleanup().
5256 */
5257 static void sighandler(int sig)
5258 {
5259 switch (sig) {
5260 case SIGPIPE:
5261 DBG("SIGPIPE caught");
5262 return;
5263 case SIGINT:
5264 DBG("SIGINT caught");
5265 stop_threads();
5266 break;
5267 case SIGTERM:
5268 DBG("SIGTERM caught");
5269 stop_threads();
5270 break;
5271 case SIGUSR1:
5272 CMM_STORE_SHARED(recv_child_signal, 1);
5273 break;
5274 default:
5275 break;
5276 }
5277 }
5278
5279 /*
5280 * Setup signal handler for :
5281 * SIGINT, SIGTERM, SIGPIPE
5282 */
5283 static int set_signal_handler(void)
5284 {
5285 int ret = 0;
5286 struct sigaction sa;
5287 sigset_t sigset;
5288
5289 if ((ret = sigemptyset(&sigset)) < 0) {
5290 PERROR("sigemptyset");
5291 return ret;
5292 }
5293
5294 sa.sa_handler = sighandler;
5295 sa.sa_mask = sigset;
5296 sa.sa_flags = 0;
5297 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5298 PERROR("sigaction");
5299 return ret;
5300 }
5301
5302 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5303 PERROR("sigaction");
5304 return ret;
5305 }
5306
5307 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5308 PERROR("sigaction");
5309 return ret;
5310 }
5311
5312 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5313 PERROR("sigaction");
5314 return ret;
5315 }
5316
5317 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5318
5319 return ret;
5320 }
5321
5322 /*
5323 * Set open files limit to unlimited. This daemon can open a large number of
5324 * file descriptors in order to consumer multiple kernel traces.
5325 */
5326 static void set_ulimit(void)
5327 {
5328 int ret;
5329 struct rlimit lim;
5330
5331 /* The kernel does not allowed an infinite limit for open files */
5332 lim.rlim_cur = 65535;
5333 lim.rlim_max = 65535;
5334
5335 ret = setrlimit(RLIMIT_NOFILE, &lim);
5336 if (ret < 0) {
5337 PERROR("failed to set open files limit");
5338 }
5339 }
5340
5341 /*
5342 * Write pidfile using the rundir and opt_pidfile.
5343 */
5344 static int write_pidfile(void)
5345 {
5346 int ret;
5347 char pidfile_path[PATH_MAX];
5348
5349 assert(rundir);
5350
5351 if (opt_pidfile) {
5352 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5353 } else {
5354 /* Build pidfile path from rundir and opt_pidfile. */
5355 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5356 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5357 if (ret < 0) {
5358 PERROR("snprintf pidfile path");
5359 goto error;
5360 }
5361 }
5362
5363 /*
5364 * Create pid file in rundir.
5365 */
5366 ret = utils_create_pid_file(getpid(), pidfile_path);
5367 error:
5368 return ret;
5369 }
5370
5371 /*
5372 * Create lockfile using the rundir and return its fd.
5373 */
5374 static int create_lockfile(void)
5375 {
5376 int ret;
5377 char lockfile_path[PATH_MAX];
5378
5379 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5380 if (ret < 0) {
5381 goto error;
5382 }
5383
5384 ret = utils_create_lock_file(lockfile_path);
5385 error:
5386 return ret;
5387 }
5388
5389 /*
5390 * Write agent TCP port using the rundir.
5391 */
5392 static int write_agent_port(void)
5393 {
5394 int ret;
5395 char path[PATH_MAX];
5396
5397 assert(rundir);
5398
5399 ret = snprintf(path, sizeof(path), "%s/"
5400 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5401 if (ret < 0) {
5402 PERROR("snprintf agent port path");
5403 goto error;
5404 }
5405
5406 /*
5407 * Create TCP agent port file in rundir.
5408 */
5409 ret = utils_create_pid_file(agent_tcp_port, path);
5410
5411 error:
5412 return ret;
5413 }
5414
5415 /*
5416 * main
5417 */
5418 int main(int argc, char **argv)
5419 {
5420 int ret = 0, retval = 0;
5421 void *status;
5422 const char *home_path, *env_app_timeout;
5423
5424 init_kernel_workarounds();
5425
5426 rcu_register_thread();
5427
5428 if (set_signal_handler()) {
5429 retval = -1;
5430 goto exit_set_signal_handler;
5431 }
5432
5433 setup_consumerd_path();
5434
5435 page_size = sysconf(_SC_PAGESIZE);
5436 if (page_size < 0) {
5437 PERROR("sysconf _SC_PAGESIZE");
5438 page_size = LONG_MAX;
5439 WARN("Fallback page size to %ld", page_size);
5440 }
5441
5442 /*
5443 * Parse arguments and load the daemon configuration file.
5444 *
5445 * We have an exit_options exit path to free memory reserved by
5446 * set_options. This is needed because the rest of sessiond_cleanup()
5447 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5448 * depends on set_options.
5449 */
5450 progname = argv[0];
5451 if (set_options(argc, argv)) {
5452 retval = -1;
5453 goto exit_options;
5454 }
5455
5456 /* Daemonize */
5457 if (opt_daemon || opt_background) {
5458 int i;
5459
5460 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5461 !opt_background);
5462 if (ret < 0) {
5463 retval = -1;
5464 goto exit_options;
5465 }
5466
5467 /*
5468 * We are in the child. Make sure all other file descriptors are
5469 * closed, in case we are called with more opened file
5470 * descriptors than the standard ones.
5471 */
5472 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5473 (void) close(i);
5474 }
5475 }
5476
5477 /*
5478 * Starting from here, we can create threads. This needs to be after
5479 * lttng_daemonize due to RCU.
5480 */
5481
5482 /*
5483 * Initialize the health check subsystem. This call should set the
5484 * appropriate time values.
5485 */
5486 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5487 if (!health_sessiond) {
5488 PERROR("health_app_create error");
5489 retval = -1;
5490 goto exit_health_sessiond_cleanup;
5491 }
5492
5493 if (init_ht_cleanup_quit_pipe()) {
5494 retval = -1;
5495 goto exit_ht_cleanup_quit_pipe;
5496 }
5497
5498 /* Setup the thread ht_cleanup communication pipe. */
5499 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5500 retval = -1;
5501 goto exit_ht_cleanup_pipe;
5502 }
5503
5504 /* Set up max poll set size */
5505 if (lttng_poll_set_max_size()) {
5506 retval = -1;
5507 goto exit_set_max_size;
5508 }
5509
5510 /* Create thread to clean up RCU hash tables */
5511 ret = pthread_create(&ht_cleanup_thread, NULL,
5512 thread_ht_cleanup, (void *) NULL);
5513 if (ret) {
5514 errno = ret;
5515 PERROR("pthread_create ht_cleanup");
5516 retval = -1;
5517 goto exit_ht_cleanup;
5518 }
5519
5520 /* Create thread quit pipe */
5521 if (init_thread_quit_pipe()) {
5522 retval = -1;
5523 goto exit_init_data;
5524 }
5525
5526 /* Check if daemon is UID = 0 */
5527 is_root = !getuid();
5528
5529 if (is_root) {
5530 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5531 if (!rundir) {
5532 retval = -1;
5533 goto exit_init_data;
5534 }
5535
5536 /* Create global run dir with root access */
5537 if (create_lttng_rundir(rundir)) {
5538 retval = -1;
5539 goto exit_init_data;
5540 }
5541
5542 if (strlen(apps_unix_sock_path) == 0) {
5543 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5544 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5545 if (ret < 0) {
5546 retval = -1;
5547 goto exit_init_data;
5548 }
5549 }
5550
5551 if (strlen(client_unix_sock_path) == 0) {
5552 ret = snprintf(client_unix_sock_path, PATH_MAX,
5553 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5554 if (ret < 0) {
5555 retval = -1;
5556 goto exit_init_data;
5557 }
5558 }
5559
5560 /* Set global SHM for ust */
5561 if (strlen(wait_shm_path) == 0) {
5562 ret = snprintf(wait_shm_path, PATH_MAX,
5563 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5564 if (ret < 0) {
5565 retval = -1;
5566 goto exit_init_data;
5567 }
5568 }
5569
5570 if (strlen(health_unix_sock_path) == 0) {
5571 ret = snprintf(health_unix_sock_path,
5572 sizeof(health_unix_sock_path),
5573 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5574 if (ret < 0) {
5575 retval = -1;
5576 goto exit_init_data;
5577 }
5578 }
5579
5580 /* Setup kernel consumerd path */
5581 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5582 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5583 if (ret < 0) {
5584 retval = -1;
5585 goto exit_init_data;
5586 }
5587 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5588 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5589 if (ret < 0) {
5590 retval = -1;
5591 goto exit_init_data;
5592 }
5593
5594 DBG2("Kernel consumer err path: %s",
5595 kconsumer_data.err_unix_sock_path);
5596 DBG2("Kernel consumer cmd path: %s",
5597 kconsumer_data.cmd_unix_sock_path);
5598 } else {
5599 home_path = utils_get_home_dir();
5600 if (home_path == NULL) {
5601 /* TODO: Add --socket PATH option */
5602 ERR("Can't get HOME directory for sockets creation.");
5603 retval = -1;
5604 goto exit_init_data;
5605 }
5606
5607 /*
5608 * Create rundir from home path. This will create something like
5609 * $HOME/.lttng
5610 */
5611 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5612 if (ret < 0) {
5613 retval = -1;
5614 goto exit_init_data;
5615 }
5616
5617 if (create_lttng_rundir(rundir)) {
5618 retval = -1;
5619 goto exit_init_data;
5620 }
5621
5622 if (strlen(apps_unix_sock_path) == 0) {
5623 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5624 DEFAULT_HOME_APPS_UNIX_SOCK,
5625 home_path);
5626 if (ret < 0) {
5627 retval = -1;
5628 goto exit_init_data;
5629 }
5630 }
5631
5632 /* Set the cli tool unix socket path */
5633 if (strlen(client_unix_sock_path) == 0) {
5634 ret = snprintf(client_unix_sock_path, PATH_MAX,
5635 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5636 home_path);
5637 if (ret < 0) {
5638 retval = -1;
5639 goto exit_init_data;
5640 }
5641 }
5642
5643 /* Set global SHM for ust */
5644 if (strlen(wait_shm_path) == 0) {
5645 ret = snprintf(wait_shm_path, PATH_MAX,
5646 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5647 getuid());
5648 if (ret < 0) {
5649 retval = -1;
5650 goto exit_init_data;
5651 }
5652 }
5653
5654 /* Set health check Unix path */
5655 if (strlen(health_unix_sock_path) == 0) {
5656 ret = snprintf(health_unix_sock_path,
5657 sizeof(health_unix_sock_path),
5658 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5659 home_path);
5660 if (ret < 0) {
5661 retval = -1;
5662 goto exit_init_data;
5663 }
5664 }
5665 }
5666
5667 lockfile_fd = create_lockfile();
5668 if (lockfile_fd < 0) {
5669 retval = -1;
5670 goto exit_init_data;
5671 }
5672
5673 /* Set consumer initial state */
5674 kernel_consumerd_state = CONSUMER_STOPPED;
5675 ust_consumerd_state = CONSUMER_STOPPED;
5676
5677 DBG("Client socket path %s", client_unix_sock_path);
5678 DBG("Application socket path %s", apps_unix_sock_path);
5679 DBG("Application wait path %s", wait_shm_path);
5680 DBG("LTTng run directory path: %s", rundir);
5681
5682 /* 32 bits consumerd path setup */
5683 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5684 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5685 if (ret < 0) {
5686 PERROR("snprintf 32-bit consumer error socket path");
5687 retval = -1;
5688 goto exit_init_data;
5689 }
5690 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5691 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5692 if (ret < 0) {
5693 PERROR("snprintf 32-bit consumer command socket path");
5694 retval = -1;
5695 goto exit_init_data;
5696 }
5697
5698 DBG2("UST consumer 32 bits err path: %s",
5699 ustconsumer32_data.err_unix_sock_path);
5700 DBG2("UST consumer 32 bits cmd path: %s",
5701 ustconsumer32_data.cmd_unix_sock_path);
5702
5703 /* 64 bits consumerd path setup */
5704 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5705 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5706 if (ret < 0) {
5707 PERROR("snprintf 64-bit consumer error socket path");
5708 retval = -1;
5709 goto exit_init_data;
5710 }
5711 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5712 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5713 if (ret < 0) {
5714 PERROR("snprintf 64-bit consumer command socket path");
5715 retval = -1;
5716 goto exit_init_data;
5717 }
5718
5719 DBG2("UST consumer 64 bits err path: %s",
5720 ustconsumer64_data.err_unix_sock_path);
5721 DBG2("UST consumer 64 bits cmd path: %s",
5722 ustconsumer64_data.cmd_unix_sock_path);
5723
5724 /*
5725 * See if daemon already exist.
5726 */
5727 if (check_existing_daemon()) {
5728 ERR("Already running daemon.\n");
5729 /*
5730 * We do not goto exit because we must not cleanup()
5731 * because a daemon is already running.
5732 */
5733 retval = -1;
5734 goto exit_init_data;
5735 }
5736
5737 /*
5738 * Init UST app hash table. Alloc hash table before this point since
5739 * cleanup() can get called after that point.
5740 */
5741 if (ust_app_ht_alloc()) {
5742 ERR("Failed to allocate UST app hash table");
5743 retval = -1;
5744 goto exit_init_data;
5745 }
5746
5747 /*
5748 * Initialize agent app hash table. We allocate the hash table here
5749 * since cleanup() can get called after this point.
5750 */
5751 if (agent_app_ht_alloc()) {
5752 ERR("Failed to allocate Agent app hash table");
5753 retval = -1;
5754 goto exit_init_data;
5755 }
5756
5757 /*
5758 * These actions must be executed as root. We do that *after* setting up
5759 * the sockets path because we MUST make the check for another daemon using
5760 * those paths *before* trying to set the kernel consumer sockets and init
5761 * kernel tracer.
5762 */
5763 if (is_root) {
5764 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5765 retval = -1;
5766 goto exit_init_data;
5767 }
5768
5769 /* Setup kernel tracer */
5770 if (!opt_no_kernel) {
5771 init_kernel_tracer();
5772 if (kernel_tracer_fd >= 0) {
5773 ret = syscall_init_table();
5774 if (ret < 0) {
5775 ERR("Unable to populate syscall table. "
5776 "Syscall tracing won't work "
5777 "for this session daemon.");
5778 }
5779 }
5780 }
5781
5782 /* Set ulimit for open files */
5783 set_ulimit();
5784 }
5785 /* init lttng_fd tracking must be done after set_ulimit. */
5786 lttng_fd_init();
5787
5788 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5789 retval = -1;
5790 goto exit_init_data;
5791 }
5792
5793 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5794 retval = -1;
5795 goto exit_init_data;
5796 }
5797
5798 /* Setup the needed unix socket */
5799 if (init_daemon_socket()) {
5800 retval = -1;
5801 goto exit_init_data;
5802 }
5803
5804 /* Set credentials to socket */
5805 if (is_root && set_permissions(rundir)) {
5806 retval = -1;
5807 goto exit_init_data;
5808 }
5809
5810 /* Get parent pid if -S, --sig-parent is specified. */
5811 if (opt_sig_parent) {
5812 ppid = getppid();
5813 }
5814
5815 /* Setup the kernel pipe for waking up the kernel thread */
5816 if (is_root && !opt_no_kernel) {
5817 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5818 retval = -1;
5819 goto exit_init_data;
5820 }
5821 }
5822
5823 /* Setup the thread apps communication pipe. */
5824 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5825 retval = -1;
5826 goto exit_init_data;
5827 }
5828
5829 /* Setup the thread apps notify communication pipe. */
5830 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5831 retval = -1;
5832 goto exit_init_data;
5833 }
5834
5835 /* Initialize global buffer per UID and PID registry. */
5836 buffer_reg_init_uid_registry();
5837 buffer_reg_init_pid_registry();
5838
5839 /* Init UST command queue. */
5840 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5841
5842 /*
5843 * Get session list pointer. This pointer MUST NOT be free'd. This list
5844 * is statically declared in session.c
5845 */
5846 session_list_ptr = session_get_list();
5847
5848 cmd_init();
5849
5850 /* Check for the application socket timeout env variable. */
5851 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5852 if (env_app_timeout) {
5853 app_socket_timeout = atoi(env_app_timeout);
5854 } else {
5855 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5856 }
5857
5858 ret = write_pidfile();
5859 if (ret) {
5860 ERR("Error in write_pidfile");
5861 retval = -1;
5862 goto exit_init_data;
5863 }
5864 ret = write_agent_port();
5865 if (ret) {
5866 ERR("Error in write_agent_port");
5867 retval = -1;
5868 goto exit_init_data;
5869 }
5870
5871 /* Initialize communication library */
5872 lttcomm_init();
5873 /* Initialize TCP timeout values */
5874 lttcomm_inet_init();
5875
5876 if (load_session_init_data(&load_info) < 0) {
5877 retval = -1;
5878 goto exit_init_data;
5879 }
5880 load_info->path = opt_load_session_path;
5881
5882 /* Create health-check thread */
5883 ret = pthread_create(&health_thread, NULL,
5884 thread_manage_health, (void *) NULL);
5885 if (ret) {
5886 errno = ret;
5887 PERROR("pthread_create health");
5888 retval = -1;
5889 goto exit_health;
5890 }
5891
5892 /* Create thread to manage the client socket */
5893 ret = pthread_create(&client_thread, NULL,
5894 thread_manage_clients, (void *) NULL);
5895 if (ret) {
5896 errno = ret;
5897 PERROR("pthread_create clients");
5898 retval = -1;
5899 goto exit_client;
5900 }
5901
5902 /* Create thread to dispatch registration */
5903 ret = pthread_create(&dispatch_thread, NULL,
5904 thread_dispatch_ust_registration, (void *) NULL);
5905 if (ret) {
5906 errno = ret;
5907 PERROR("pthread_create dispatch");
5908 retval = -1;
5909 goto exit_dispatch;
5910 }
5911
5912 /* Create thread to manage application registration. */
5913 ret = pthread_create(&reg_apps_thread, NULL,
5914 thread_registration_apps, (void *) NULL);
5915 if (ret) {
5916 errno = ret;
5917 PERROR("pthread_create registration");
5918 retval = -1;
5919 goto exit_reg_apps;
5920 }
5921
5922 /* Create thread to manage application socket */
5923 ret = pthread_create(&apps_thread, NULL,
5924 thread_manage_apps, (void *) NULL);
5925 if (ret) {
5926 errno = ret;
5927 PERROR("pthread_create apps");
5928 retval = -1;
5929 goto exit_apps;
5930 }
5931
5932 /* Create thread to manage application notify socket */
5933 ret = pthread_create(&apps_notify_thread, NULL,
5934 ust_thread_manage_notify, (void *) NULL);
5935 if (ret) {
5936 errno = ret;
5937 PERROR("pthread_create notify");
5938 retval = -1;
5939 goto exit_apps_notify;
5940 }
5941
5942 /* Create agent registration thread. */
5943 ret = pthread_create(&agent_reg_thread, NULL,
5944 agent_thread_manage_registration, (void *) NULL);
5945 if (ret) {
5946 errno = ret;
5947 PERROR("pthread_create agent");
5948 retval = -1;
5949 goto exit_agent_reg;
5950 }
5951
5952 /* Don't start this thread if kernel tracing is not requested nor root */
5953 if (is_root && !opt_no_kernel) {
5954 /* Create kernel thread to manage kernel event */
5955 ret = pthread_create(&kernel_thread, NULL,
5956 thread_manage_kernel, (void *) NULL);
5957 if (ret) {
5958 errno = ret;
5959 PERROR("pthread_create kernel");
5960 retval = -1;
5961 goto exit_kernel;
5962 }
5963 }
5964
5965 /* Create session loading thread. */
5966 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5967 load_info);
5968 if (ret) {
5969 errno = ret;
5970 PERROR("pthread_create load_session_thread");
5971 retval = -1;
5972 goto exit_load_session;
5973 }
5974
5975 /*
5976 * This is where we start awaiting program completion (e.g. through
5977 * signal that asks threads to teardown).
5978 */
5979
5980 ret = pthread_join(load_session_thread, &status);
5981 if (ret) {
5982 errno = ret;
5983 PERROR("pthread_join load_session_thread");
5984 retval = -1;
5985 }
5986 exit_load_session:
5987
5988 if (is_root && !opt_no_kernel) {
5989 ret = pthread_join(kernel_thread, &status);
5990 if (ret) {
5991 errno = ret;
5992 PERROR("pthread_join");
5993 retval = -1;
5994 }
5995 }
5996 exit_kernel:
5997
5998 ret = pthread_join(agent_reg_thread, &status);
5999 if (ret) {
6000 errno = ret;
6001 PERROR("pthread_join agent");
6002 retval = -1;
6003 }
6004 exit_agent_reg:
6005
6006 ret = pthread_join(apps_notify_thread, &status);
6007 if (ret) {
6008 errno = ret;
6009 PERROR("pthread_join apps notify");
6010 retval = -1;
6011 }
6012 exit_apps_notify:
6013
6014 ret = pthread_join(apps_thread, &status);
6015 if (ret) {
6016 errno = ret;
6017 PERROR("pthread_join apps");
6018 retval = -1;
6019 }
6020 exit_apps:
6021
6022 ret = pthread_join(reg_apps_thread, &status);
6023 if (ret) {
6024 errno = ret;
6025 PERROR("pthread_join");
6026 retval = -1;
6027 }
6028 exit_reg_apps:
6029
6030 /*
6031 * Join dispatch thread after joining reg_apps_thread to ensure
6032 * we don't leak applications in the queue.
6033 */
6034 ret = pthread_join(dispatch_thread, &status);
6035 if (ret) {
6036 errno = ret;
6037 PERROR("pthread_join");
6038 retval = -1;
6039 }
6040 exit_dispatch:
6041
6042 ret = pthread_join(client_thread, &status);
6043 if (ret) {
6044 errno = ret;
6045 PERROR("pthread_join");
6046 retval = -1;
6047 }
6048 exit_client:
6049
6050 ret = pthread_join(health_thread, &status);
6051 if (ret) {
6052 errno = ret;
6053 PERROR("pthread_join health thread");
6054 retval = -1;
6055 }
6056 exit_health:
6057
6058 exit_init_data:
6059 /*
6060 * sessiond_cleanup() is called when no other thread is running, except
6061 * the ht_cleanup thread, which is needed to destroy the hash tables.
6062 */
6063 rcu_thread_online();
6064 sessiond_cleanup();
6065 rcu_thread_offline();
6066 rcu_unregister_thread();
6067
6068 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6069 if (ret < 0) {
6070 ERR("write error on ht_cleanup quit pipe");
6071 retval = -1;
6072 }
6073
6074 ret = pthread_join(ht_cleanup_thread, &status);
6075 if (ret) {
6076 errno = ret;
6077 PERROR("pthread_join ht cleanup thread");
6078 retval = -1;
6079 }
6080 exit_ht_cleanup:
6081 exit_set_max_size:
6082
6083 utils_close_pipe(ht_cleanup_pipe);
6084 exit_ht_cleanup_pipe:
6085
6086 /*
6087 * Close the ht_cleanup quit pipe.
6088 */
6089 utils_close_pipe(ht_cleanup_quit_pipe);
6090 exit_ht_cleanup_quit_pipe:
6091
6092 health_app_destroy(health_sessiond);
6093 exit_health_sessiond_cleanup:
6094
6095 exit_options:
6096 sessiond_cleanup_options();
6097
6098 exit_set_signal_handler:
6099 /* Ensure all prior call_rcu are done. */
6100 rcu_barrier();
6101
6102 if (!retval) {
6103 exit(EXIT_SUCCESS);
6104 } else {
6105 exit(EXIT_FAILURE);
6106 }
6107 }
This page took 0.2043 seconds and 4 git commands to generate.