Add rcu_read_ongoing() assertions around process_client_msg
[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 DBG("Cleaning up all agent apps");
716 agent_app_ht_clean();
717
718 DBG("Closing all UST sockets");
719 ust_app_clean_list();
720 buffer_reg_destroy_registries();
721
722 if (is_root && !opt_no_kernel) {
723 DBG2("Closing kernel fd");
724 if (kernel_tracer_fd >= 0) {
725 ret = close(kernel_tracer_fd);
726 if (ret) {
727 PERROR("close");
728 }
729 }
730 DBG("Unloading kernel modules");
731 modprobe_remove_lttng_all();
732 free(syscall_table);
733 }
734
735 close_consumer_sockets();
736
737 if (load_info) {
738 load_session_destroy_data(load_info);
739 free(load_info);
740 }
741
742 /*
743 * Cleanup lock file by deleting it and finaly closing it which will
744 * release the file system lock.
745 */
746 if (lockfile_fd >= 0) {
747 char lockfile_path[PATH_MAX];
748
749 ret = generate_lock_file_path(lockfile_path,
750 sizeof(lockfile_path));
751 if (ret > 0) {
752 ret = remove(lockfile_path);
753 if (ret < 0) {
754 PERROR("remove lock file");
755 }
756 ret = close(lockfile_fd);
757 if (ret < 0) {
758 PERROR("close lock file");
759 }
760 }
761 }
762
763 /*
764 * We do NOT rmdir rundir because there are other processes
765 * using it, for instance lttng-relayd, which can start in
766 * parallel with this teardown.
767 */
768
769 free(rundir);
770 }
771
772 /*
773 * Cleanup the daemon's option data structures.
774 */
775 static void sessiond_cleanup_options(void)
776 {
777 DBG("Cleaning up options");
778
779 /*
780 * If the override option is set, the pointer points to a *non* const
781 * thus freeing it even though the variable type is set to const.
782 */
783 if (tracing_group_name_override) {
784 free((void *) tracing_group_name);
785 }
786 if (consumerd32_bin_override) {
787 free((void *) consumerd32_bin);
788 }
789 if (consumerd64_bin_override) {
790 free((void *) consumerd64_bin);
791 }
792 if (consumerd32_libdir_override) {
793 free((void *) consumerd32_libdir);
794 }
795 if (consumerd64_libdir_override) {
796 free((void *) consumerd64_libdir);
797 }
798
799 free(opt_pidfile);
800 free(opt_load_session_path);
801 free(kmod_probes_list);
802 free(kmod_extra_probes_list);
803
804 /* <fun> */
805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 /* </fun> */
809 }
810
811 /*
812 * Send data on a unix socket using the liblttsessiondcomm API.
813 *
814 * Return lttcomm error code.
815 */
816 static int send_unix_sock(int sock, void *buf, size_t len)
817 {
818 /* Check valid length */
819 if (len == 0) {
820 return -1;
821 }
822
823 return lttcomm_send_unix_sock(sock, buf, len);
824 }
825
826 /*
827 * Free memory of a command context structure.
828 */
829 static void clean_command_ctx(struct command_ctx **cmd_ctx)
830 {
831 DBG("Clean command context structure");
832 if (*cmd_ctx) {
833 if ((*cmd_ctx)->llm) {
834 free((*cmd_ctx)->llm);
835 }
836 if ((*cmd_ctx)->lsm) {
837 free((*cmd_ctx)->lsm);
838 }
839 free(*cmd_ctx);
840 *cmd_ctx = NULL;
841 }
842 }
843
844 /*
845 * Notify UST applications using the shm mmap futex.
846 */
847 static int notify_ust_apps(int active)
848 {
849 char *wait_shm_mmap;
850
851 DBG("Notifying applications of session daemon state: %d", active);
852
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
855 if (wait_shm_mmap == NULL) {
856 goto error;
857 }
858
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap, active);
861
862 /* Apps notified successfully */
863 return 0;
864
865 error:
866 return -1;
867 }
868
869 /*
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
872 * structure.
873 *
874 * Return total size of the buffer pointed by buf.
875 */
876 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
877 {
878 int ret, buf_size;
879
880 buf_size = size;
881
882 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
883 if (cmd_ctx->llm == NULL) {
884 PERROR("zmalloc");
885 ret = -ENOMEM;
886 goto error;
887 }
888
889 /* Copy common data */
890 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
891 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
892
893 cmd_ctx->llm->data_size = size;
894 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
895
896 return buf_size;
897
898 error:
899 return ret;
900 }
901
902 /*
903 * Update the kernel poll set of all channel fd available over all tracing
904 * session. Add the wakeup pipe at the end of the set.
905 */
906 static int update_kernel_poll(struct lttng_poll_event *events)
907 {
908 int ret;
909 struct ltt_session *session;
910 struct ltt_kernel_channel *channel;
911
912 DBG("Updating kernel poll set");
913
914 session_lock_list();
915 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
916 session_lock(session);
917 if (session->kernel_session == NULL) {
918 session_unlock(session);
919 continue;
920 }
921
922 cds_list_for_each_entry(channel,
923 &session->kernel_session->channel_list.head, list) {
924 /* Add channel fd to the kernel poll set */
925 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
926 if (ret < 0) {
927 session_unlock(session);
928 goto error;
929 }
930 DBG("Channel fd %d added to kernel set", channel->fd);
931 }
932 session_unlock(session);
933 }
934 session_unlock_list();
935
936 return 0;
937
938 error:
939 session_unlock_list();
940 return -1;
941 }
942
943 /*
944 * Find the channel fd from 'fd' over all tracing session. When found, check
945 * for new channel stream and send those stream fds to the kernel consumer.
946 *
947 * Useful for CPU hotplug feature.
948 */
949 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
950 {
951 int ret = 0;
952 struct ltt_session *session;
953 struct ltt_kernel_session *ksess;
954 struct ltt_kernel_channel *channel;
955
956 DBG("Updating kernel streams for channel fd %d", fd);
957
958 session_lock_list();
959 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
960 session_lock(session);
961 if (session->kernel_session == NULL) {
962 session_unlock(session);
963 continue;
964 }
965 ksess = session->kernel_session;
966
967 cds_list_for_each_entry(channel,
968 &ksess->channel_list.head, list) {
969 struct lttng_ht_iter iter;
970 struct consumer_socket *socket;
971
972 if (channel->fd != fd) {
973 continue;
974 }
975 DBG("Channel found, updating kernel streams");
976 ret = kernel_open_channel_stream(channel);
977 if (ret < 0) {
978 goto error;
979 }
980 /* Update the stream global counter */
981 ksess->stream_count_global += ret;
982
983 /*
984 * Have we already sent fds to the consumer? If yes, it
985 * means that tracing is started so it is safe to send
986 * our updated stream fds.
987 */
988 if (ksess->consumer_fds_sent != 1
989 || ksess->consumer == NULL) {
990 ret = -1;
991 goto error;
992 }
993
994 rcu_read_lock();
995 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
996 &iter.iter, socket, node.node) {
997 pthread_mutex_lock(socket->lock);
998 ret = kernel_consumer_send_channel_stream(socket,
999 channel, ksess,
1000 session->output_traces ? 1 : 0);
1001 pthread_mutex_unlock(socket->lock);
1002 if (ret < 0) {
1003 rcu_read_unlock();
1004 goto error;
1005 }
1006 }
1007 rcu_read_unlock();
1008 }
1009 session_unlock(session);
1010 }
1011 session_unlock_list();
1012 return ret;
1013
1014 error:
1015 session_unlock(session);
1016 session_unlock_list();
1017 return ret;
1018 }
1019
1020 /*
1021 * For each tracing session, update newly registered apps. The session list
1022 * lock MUST be acquired before calling this.
1023 */
1024 static void update_ust_app(int app_sock)
1025 {
1026 struct ltt_session *sess, *stmp;
1027
1028 /* Consumer is in an ERROR state. Stop any application update. */
1029 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1030 /* Stop the update process since the consumer is dead. */
1031 return;
1032 }
1033
1034 /* For all tracing session(s) */
1035 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1036 struct ust_app *app;
1037
1038 session_lock(sess);
1039 if (!sess->ust_session) {
1040 goto unlock_session;
1041 }
1042
1043 rcu_read_lock();
1044 assert(app_sock >= 0);
1045 app = ust_app_find_by_sock(app_sock);
1046 if (app == NULL) {
1047 /*
1048 * Application can be unregistered before so
1049 * this is possible hence simply stopping the
1050 * update.
1051 */
1052 DBG3("UST app update failed to find app sock %d",
1053 app_sock);
1054 goto unlock_rcu;
1055 }
1056 ust_app_global_update(sess->ust_session, app);
1057 unlock_rcu:
1058 rcu_read_unlock();
1059 unlock_session:
1060 session_unlock(sess);
1061 }
1062 }
1063
1064 /*
1065 * This thread manage event coming from the kernel.
1066 *
1067 * Features supported in this thread:
1068 * -) CPU Hotplug
1069 */
1070 static void *thread_manage_kernel(void *data)
1071 {
1072 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1073 uint32_t revents, nb_fd;
1074 char tmp;
1075 struct lttng_poll_event events;
1076
1077 DBG("[thread] Thread manage kernel started");
1078
1079 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1080
1081 /*
1082 * This first step of the while is to clean this structure which could free
1083 * non NULL pointers so initialize it before the loop.
1084 */
1085 lttng_poll_init(&events);
1086
1087 if (testpoint(sessiond_thread_manage_kernel)) {
1088 goto error_testpoint;
1089 }
1090
1091 health_code_update();
1092
1093 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1094 goto error_testpoint;
1095 }
1096
1097 while (1) {
1098 health_code_update();
1099
1100 if (update_poll_flag == 1) {
1101 /* Clean events object. We are about to populate it again. */
1102 lttng_poll_clean(&events);
1103
1104 ret = sessiond_set_thread_pollset(&events, 2);
1105 if (ret < 0) {
1106 goto error_poll_create;
1107 }
1108
1109 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1110 if (ret < 0) {
1111 goto error;
1112 }
1113
1114 /* This will add the available kernel channel if any. */
1115 ret = update_kernel_poll(&events);
1116 if (ret < 0) {
1117 goto error;
1118 }
1119 update_poll_flag = 0;
1120 }
1121
1122 DBG("Thread kernel polling");
1123
1124 /* Poll infinite value of time */
1125 restart:
1126 health_poll_entry();
1127 ret = lttng_poll_wait(&events, -1);
1128 DBG("Thread kernel return from poll on %d fds",
1129 LTTNG_POLL_GETNB(&events));
1130 health_poll_exit();
1131 if (ret < 0) {
1132 /*
1133 * Restart interrupted system call.
1134 */
1135 if (errno == EINTR) {
1136 goto restart;
1137 }
1138 goto error;
1139 } else if (ret == 0) {
1140 /* Should not happen since timeout is infinite */
1141 ERR("Return value of poll is 0 with an infinite timeout.\n"
1142 "This should not have happened! Continuing...");
1143 continue;
1144 }
1145
1146 nb_fd = ret;
1147
1148 for (i = 0; i < nb_fd; i++) {
1149 /* Fetch once the poll data */
1150 revents = LTTNG_POLL_GETEV(&events, i);
1151 pollfd = LTTNG_POLL_GETFD(&events, i);
1152
1153 health_code_update();
1154
1155 if (!revents) {
1156 /* No activity for this FD (poll implementation). */
1157 continue;
1158 }
1159
1160 /* Thread quit pipe has been closed. Killing thread. */
1161 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1162 if (ret) {
1163 err = 0;
1164 goto exit;
1165 }
1166
1167 /* Check for data on kernel pipe */
1168 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1169 (void) lttng_read(kernel_poll_pipe[0],
1170 &tmp, 1);
1171 /*
1172 * Ret value is useless here, if this pipe gets any actions an
1173 * update is required anyway.
1174 */
1175 update_poll_flag = 1;
1176 continue;
1177 } else {
1178 /*
1179 * New CPU detected by the kernel. Adding kernel stream to
1180 * kernel session and updating the kernel consumer
1181 */
1182 if (revents & LPOLLIN) {
1183 ret = update_kernel_stream(&kconsumer_data, pollfd);
1184 if (ret < 0) {
1185 continue;
1186 }
1187 break;
1188 /*
1189 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1190 * and unregister kernel stream at this point.
1191 */
1192 }
1193 }
1194 }
1195 }
1196
1197 exit:
1198 error:
1199 lttng_poll_clean(&events);
1200 error_poll_create:
1201 error_testpoint:
1202 utils_close_pipe(kernel_poll_pipe);
1203 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1204 if (err) {
1205 health_error();
1206 ERR("Health error occurred in %s", __func__);
1207 WARN("Kernel thread died unexpectedly. "
1208 "Kernel tracing can continue but CPU hotplug is disabled.");
1209 }
1210 health_unregister(health_sessiond);
1211 DBG("Kernel thread dying");
1212 return NULL;
1213 }
1214
1215 /*
1216 * Signal pthread condition of the consumer data that the thread.
1217 */
1218 static void signal_consumer_condition(struct consumer_data *data, int state)
1219 {
1220 pthread_mutex_lock(&data->cond_mutex);
1221
1222 /*
1223 * The state is set before signaling. It can be any value, it's the waiter
1224 * job to correctly interpret this condition variable associated to the
1225 * consumer pthread_cond.
1226 *
1227 * A value of 0 means that the corresponding thread of the consumer data
1228 * was not started. 1 indicates that the thread has started and is ready
1229 * for action. A negative value means that there was an error during the
1230 * thread bootstrap.
1231 */
1232 data->consumer_thread_is_ready = state;
1233 (void) pthread_cond_signal(&data->cond);
1234
1235 pthread_mutex_unlock(&data->cond_mutex);
1236 }
1237
1238 /*
1239 * This thread manage the consumer error sent back to the session daemon.
1240 */
1241 static void *thread_manage_consumer(void *data)
1242 {
1243 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1244 uint32_t revents, nb_fd;
1245 enum lttcomm_return_code code;
1246 struct lttng_poll_event events;
1247 struct consumer_data *consumer_data = data;
1248
1249 DBG("[thread] Manage consumer started");
1250
1251 rcu_register_thread();
1252 rcu_thread_online();
1253
1254 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1255
1256 health_code_update();
1257
1258 /*
1259 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1260 * metadata_sock. Nothing more will be added to this poll set.
1261 */
1262 ret = sessiond_set_thread_pollset(&events, 3);
1263 if (ret < 0) {
1264 goto error_poll;
1265 }
1266
1267 /*
1268 * The error socket here is already in a listening state which was done
1269 * just before spawning this thread to avoid a race between the consumer
1270 * daemon exec trying to connect and the listen() call.
1271 */
1272 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1273 if (ret < 0) {
1274 goto error;
1275 }
1276
1277 health_code_update();
1278
1279 /* Infinite blocking call, waiting for transmission */
1280 restart:
1281 health_poll_entry();
1282
1283 if (testpoint(sessiond_thread_manage_consumer)) {
1284 goto error;
1285 }
1286
1287 ret = lttng_poll_wait(&events, -1);
1288 health_poll_exit();
1289 if (ret < 0) {
1290 /*
1291 * Restart interrupted system call.
1292 */
1293 if (errno == EINTR) {
1294 goto restart;
1295 }
1296 goto error;
1297 }
1298
1299 nb_fd = ret;
1300
1301 for (i = 0; i < nb_fd; i++) {
1302 /* Fetch once the poll data */
1303 revents = LTTNG_POLL_GETEV(&events, i);
1304 pollfd = LTTNG_POLL_GETFD(&events, i);
1305
1306 health_code_update();
1307
1308 if (!revents) {
1309 /* No activity for this FD (poll implementation). */
1310 continue;
1311 }
1312
1313 /* Thread quit pipe has been closed. Killing thread. */
1314 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1315 if (ret) {
1316 err = 0;
1317 goto exit;
1318 }
1319
1320 /* Event on the registration socket */
1321 if (pollfd == consumer_data->err_sock) {
1322 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1323 ERR("consumer err socket poll error");
1324 goto error;
1325 }
1326 }
1327 }
1328
1329 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1330 if (sock < 0) {
1331 goto error;
1332 }
1333
1334 /*
1335 * Set the CLOEXEC flag. Return code is useless because either way, the
1336 * show must go on.
1337 */
1338 (void) utils_set_fd_cloexec(sock);
1339
1340 health_code_update();
1341
1342 DBG2("Receiving code from consumer err_sock");
1343
1344 /* Getting status code from kconsumerd */
1345 ret = lttcomm_recv_unix_sock(sock, &code,
1346 sizeof(enum lttcomm_return_code));
1347 if (ret <= 0) {
1348 goto error;
1349 }
1350
1351 health_code_update();
1352 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1353 /* Connect both socket, command and metadata. */
1354 consumer_data->cmd_sock =
1355 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1356 consumer_data->metadata_fd =
1357 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1358 if (consumer_data->cmd_sock < 0
1359 || consumer_data->metadata_fd < 0) {
1360 PERROR("consumer connect cmd socket");
1361 /* On error, signal condition and quit. */
1362 signal_consumer_condition(consumer_data, -1);
1363 goto error;
1364 }
1365 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1366 /* Create metadata socket lock. */
1367 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1368 if (consumer_data->metadata_sock.lock == NULL) {
1369 PERROR("zmalloc pthread mutex");
1370 ret = -1;
1371 goto error;
1372 }
1373 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1374
1375 signal_consumer_condition(consumer_data, 1);
1376 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1377 DBG("Consumer metadata socket ready (fd: %d)",
1378 consumer_data->metadata_fd);
1379 } else {
1380 ERR("consumer error when waiting for SOCK_READY : %s",
1381 lttcomm_get_readable_code(-code));
1382 goto error;
1383 }
1384
1385 /* Remove the consumerd error sock since we've established a connexion */
1386 ret = lttng_poll_del(&events, consumer_data->err_sock);
1387 if (ret < 0) {
1388 goto error;
1389 }
1390
1391 /* Add new accepted error socket. */
1392 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1393 if (ret < 0) {
1394 goto error;
1395 }
1396
1397 /* Add metadata socket that is successfully connected. */
1398 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1399 LPOLLIN | LPOLLRDHUP);
1400 if (ret < 0) {
1401 goto error;
1402 }
1403
1404 health_code_update();
1405
1406 /* Infinite blocking call, waiting for transmission */
1407 restart_poll:
1408 while (1) {
1409 health_code_update();
1410
1411 /* Exit the thread because the thread quit pipe has been triggered. */
1412 if (should_quit) {
1413 /* Not a health error. */
1414 err = 0;
1415 goto exit;
1416 }
1417
1418 health_poll_entry();
1419 ret = lttng_poll_wait(&events, -1);
1420 health_poll_exit();
1421 if (ret < 0) {
1422 /*
1423 * Restart interrupted system call.
1424 */
1425 if (errno == EINTR) {
1426 goto restart_poll;
1427 }
1428 goto error;
1429 }
1430
1431 nb_fd = ret;
1432
1433 for (i = 0; i < nb_fd; i++) {
1434 /* Fetch once the poll data */
1435 revents = LTTNG_POLL_GETEV(&events, i);
1436 pollfd = LTTNG_POLL_GETFD(&events, i);
1437
1438 health_code_update();
1439
1440 if (!revents) {
1441 /* No activity for this FD (poll implementation). */
1442 continue;
1443 }
1444
1445 /*
1446 * Thread quit pipe has been triggered, flag that we should stop
1447 * but continue the current loop to handle potential data from
1448 * consumer.
1449 */
1450 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1451
1452 if (pollfd == sock) {
1453 /* Event on the consumerd socket */
1454 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1455 ERR("consumer err socket second poll error");
1456 goto error;
1457 }
1458 health_code_update();
1459 /* Wait for any kconsumerd error */
1460 ret = lttcomm_recv_unix_sock(sock, &code,
1461 sizeof(enum lttcomm_return_code));
1462 if (ret <= 0) {
1463 ERR("consumer closed the command socket");
1464 goto error;
1465 }
1466
1467 ERR("consumer return code : %s",
1468 lttcomm_get_readable_code(-code));
1469
1470 goto exit;
1471 } else if (pollfd == consumer_data->metadata_fd) {
1472 /* UST metadata requests */
1473 ret = ust_consumer_metadata_request(
1474 &consumer_data->metadata_sock);
1475 if (ret < 0) {
1476 ERR("Handling metadata request");
1477 goto error;
1478 }
1479 }
1480 /* No need for an else branch all FDs are tested prior. */
1481 }
1482 health_code_update();
1483 }
1484
1485 exit:
1486 error:
1487 /*
1488 * We lock here because we are about to close the sockets and some other
1489 * thread might be using them so get exclusive access which will abort all
1490 * other consumer command by other threads.
1491 */
1492 pthread_mutex_lock(&consumer_data->lock);
1493
1494 /* Immediately set the consumerd state to stopped */
1495 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1496 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1497 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1498 consumer_data->type == LTTNG_CONSUMER32_UST) {
1499 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1500 } else {
1501 /* Code flow error... */
1502 assert(0);
1503 }
1504
1505 if (consumer_data->err_sock >= 0) {
1506 ret = close(consumer_data->err_sock);
1507 if (ret) {
1508 PERROR("close");
1509 }
1510 consumer_data->err_sock = -1;
1511 }
1512 if (consumer_data->cmd_sock >= 0) {
1513 ret = close(consumer_data->cmd_sock);
1514 if (ret) {
1515 PERROR("close");
1516 }
1517 consumer_data->cmd_sock = -1;
1518 }
1519 if (consumer_data->metadata_sock.fd_ptr &&
1520 *consumer_data->metadata_sock.fd_ptr >= 0) {
1521 ret = close(*consumer_data->metadata_sock.fd_ptr);
1522 if (ret) {
1523 PERROR("close");
1524 }
1525 }
1526 if (sock >= 0) {
1527 ret = close(sock);
1528 if (ret) {
1529 PERROR("close");
1530 }
1531 }
1532
1533 unlink(consumer_data->err_unix_sock_path);
1534 unlink(consumer_data->cmd_unix_sock_path);
1535 pthread_mutex_unlock(&consumer_data->lock);
1536
1537 /* Cleanup metadata socket mutex. */
1538 if (consumer_data->metadata_sock.lock) {
1539 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1540 free(consumer_data->metadata_sock.lock);
1541 }
1542 lttng_poll_clean(&events);
1543 error_poll:
1544 if (err) {
1545 health_error();
1546 ERR("Health error occurred in %s", __func__);
1547 }
1548 health_unregister(health_sessiond);
1549 DBG("consumer thread cleanup completed");
1550
1551 rcu_thread_offline();
1552 rcu_unregister_thread();
1553
1554 return NULL;
1555 }
1556
1557 /*
1558 * This thread manage application communication.
1559 */
1560 static void *thread_manage_apps(void *data)
1561 {
1562 int i, ret, pollfd, err = -1;
1563 ssize_t size_ret;
1564 uint32_t revents, nb_fd;
1565 struct lttng_poll_event events;
1566
1567 DBG("[thread] Manage application started");
1568
1569 rcu_register_thread();
1570 rcu_thread_online();
1571
1572 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1573
1574 if (testpoint(sessiond_thread_manage_apps)) {
1575 goto error_testpoint;
1576 }
1577
1578 health_code_update();
1579
1580 ret = sessiond_set_thread_pollset(&events, 2);
1581 if (ret < 0) {
1582 goto error_poll_create;
1583 }
1584
1585 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1586 if (ret < 0) {
1587 goto error;
1588 }
1589
1590 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1591 goto error;
1592 }
1593
1594 health_code_update();
1595
1596 while (1) {
1597 DBG("Apps thread polling");
1598
1599 /* Inifinite blocking call, waiting for transmission */
1600 restart:
1601 health_poll_entry();
1602 ret = lttng_poll_wait(&events, -1);
1603 DBG("Apps thread return from poll on %d fds",
1604 LTTNG_POLL_GETNB(&events));
1605 health_poll_exit();
1606 if (ret < 0) {
1607 /*
1608 * Restart interrupted system call.
1609 */
1610 if (errno == EINTR) {
1611 goto restart;
1612 }
1613 goto error;
1614 }
1615
1616 nb_fd = ret;
1617
1618 for (i = 0; i < nb_fd; i++) {
1619 /* Fetch once the poll data */
1620 revents = LTTNG_POLL_GETEV(&events, i);
1621 pollfd = LTTNG_POLL_GETFD(&events, i);
1622
1623 health_code_update();
1624
1625 if (!revents) {
1626 /* No activity for this FD (poll implementation). */
1627 continue;
1628 }
1629
1630 /* Thread quit pipe has been closed. Killing thread. */
1631 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1632 if (ret) {
1633 err = 0;
1634 goto exit;
1635 }
1636
1637 /* Inspect the apps cmd pipe */
1638 if (pollfd == apps_cmd_pipe[0]) {
1639 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1640 ERR("Apps command pipe error");
1641 goto error;
1642 } else if (revents & LPOLLIN) {
1643 int sock;
1644
1645 /* Empty pipe */
1646 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1647 if (size_ret < sizeof(sock)) {
1648 PERROR("read apps cmd pipe");
1649 goto error;
1650 }
1651
1652 health_code_update();
1653
1654 /*
1655 * We only monitor the error events of the socket. This
1656 * thread does not handle any incoming data from UST
1657 * (POLLIN).
1658 */
1659 ret = lttng_poll_add(&events, sock,
1660 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1661 if (ret < 0) {
1662 goto error;
1663 }
1664
1665 DBG("Apps with sock %d added to poll set", sock);
1666 }
1667 } else {
1668 /*
1669 * At this point, we know that a registered application made
1670 * the event at poll_wait.
1671 */
1672 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1673 /* Removing from the poll set */
1674 ret = lttng_poll_del(&events, pollfd);
1675 if (ret < 0) {
1676 goto error;
1677 }
1678
1679 /* Socket closed on remote end. */
1680 ust_app_unregister(pollfd);
1681 }
1682 }
1683
1684 health_code_update();
1685 }
1686 }
1687
1688 exit:
1689 error:
1690 lttng_poll_clean(&events);
1691 error_poll_create:
1692 error_testpoint:
1693 utils_close_pipe(apps_cmd_pipe);
1694 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1695
1696 /*
1697 * We don't clean the UST app hash table here since already registered
1698 * applications can still be controlled so let them be until the session
1699 * daemon dies or the applications stop.
1700 */
1701
1702 if (err) {
1703 health_error();
1704 ERR("Health error occurred in %s", __func__);
1705 }
1706 health_unregister(health_sessiond);
1707 DBG("Application communication apps thread cleanup complete");
1708 rcu_thread_offline();
1709 rcu_unregister_thread();
1710 return NULL;
1711 }
1712
1713 /*
1714 * Send a socket to a thread This is called from the dispatch UST registration
1715 * thread once all sockets are set for the application.
1716 *
1717 * The sock value can be invalid, we don't really care, the thread will handle
1718 * it and make the necessary cleanup if so.
1719 *
1720 * On success, return 0 else a negative value being the errno message of the
1721 * write().
1722 */
1723 static int send_socket_to_thread(int fd, int sock)
1724 {
1725 ssize_t ret;
1726
1727 /*
1728 * It's possible that the FD is set as invalid with -1 concurrently just
1729 * before calling this function being a shutdown state of the thread.
1730 */
1731 if (fd < 0) {
1732 ret = -EBADF;
1733 goto error;
1734 }
1735
1736 ret = lttng_write(fd, &sock, sizeof(sock));
1737 if (ret < sizeof(sock)) {
1738 PERROR("write apps pipe %d", fd);
1739 if (ret < 0) {
1740 ret = -errno;
1741 }
1742 goto error;
1743 }
1744
1745 /* All good. Don't send back the write positive ret value. */
1746 ret = 0;
1747 error:
1748 return (int) ret;
1749 }
1750
1751 /*
1752 * Sanitize the wait queue of the dispatch registration thread meaning removing
1753 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1754 * notify socket is never received.
1755 */
1756 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1757 {
1758 int ret, nb_fd = 0, i;
1759 unsigned int fd_added = 0;
1760 struct lttng_poll_event events;
1761 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1762
1763 assert(wait_queue);
1764
1765 lttng_poll_init(&events);
1766
1767 /* Just skip everything for an empty queue. */
1768 if (!wait_queue->count) {
1769 goto end;
1770 }
1771
1772 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1773 if (ret < 0) {
1774 goto error_create;
1775 }
1776
1777 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1778 &wait_queue->head, head) {
1779 assert(wait_node->app);
1780 ret = lttng_poll_add(&events, wait_node->app->sock,
1781 LPOLLHUP | LPOLLERR);
1782 if (ret < 0) {
1783 goto error;
1784 }
1785
1786 fd_added = 1;
1787 }
1788
1789 if (!fd_added) {
1790 goto end;
1791 }
1792
1793 /*
1794 * Poll but don't block so we can quickly identify the faulty events and
1795 * clean them afterwards from the wait queue.
1796 */
1797 ret = lttng_poll_wait(&events, 0);
1798 if (ret < 0) {
1799 goto error;
1800 }
1801 nb_fd = ret;
1802
1803 for (i = 0; i < nb_fd; i++) {
1804 /* Get faulty FD. */
1805 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1806 int pollfd = LTTNG_POLL_GETFD(&events, i);
1807
1808 if (!revents) {
1809 /* No activity for this FD (poll implementation). */
1810 continue;
1811 }
1812
1813 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1814 &wait_queue->head, head) {
1815 if (pollfd == wait_node->app->sock &&
1816 (revents & (LPOLLHUP | LPOLLERR))) {
1817 cds_list_del(&wait_node->head);
1818 wait_queue->count--;
1819 ust_app_destroy(wait_node->app);
1820 free(wait_node);
1821 /*
1822 * Silence warning of use-after-free in
1823 * cds_list_for_each_entry_safe which uses
1824 * __typeof__(*wait_node).
1825 */
1826 wait_node = NULL;
1827 break;
1828 }
1829 }
1830 }
1831
1832 if (nb_fd > 0) {
1833 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1834 }
1835
1836 end:
1837 lttng_poll_clean(&events);
1838 return;
1839
1840 error:
1841 lttng_poll_clean(&events);
1842 error_create:
1843 ERR("Unable to sanitize wait queue");
1844 return;
1845 }
1846
1847 /*
1848 * Dispatch request from the registration threads to the application
1849 * communication thread.
1850 */
1851 static void *thread_dispatch_ust_registration(void *data)
1852 {
1853 int ret, err = -1;
1854 struct cds_wfcq_node *node;
1855 struct ust_command *ust_cmd = NULL;
1856 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1857 struct ust_reg_wait_queue wait_queue = {
1858 .count = 0,
1859 };
1860
1861 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1862
1863 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1864 goto error_testpoint;
1865 }
1866
1867 health_code_update();
1868
1869 CDS_INIT_LIST_HEAD(&wait_queue.head);
1870
1871 DBG("[thread] Dispatch UST command started");
1872
1873 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1874 health_code_update();
1875
1876 /* Atomically prepare the queue futex */
1877 futex_nto1_prepare(&ust_cmd_queue.futex);
1878
1879 do {
1880 struct ust_app *app = NULL;
1881 ust_cmd = NULL;
1882
1883 /*
1884 * Make sure we don't have node(s) that have hung up before receiving
1885 * the notify socket. This is to clean the list in order to avoid
1886 * memory leaks from notify socket that are never seen.
1887 */
1888 sanitize_wait_queue(&wait_queue);
1889
1890 health_code_update();
1891 /* Dequeue command for registration */
1892 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1893 if (node == NULL) {
1894 DBG("Woken up but nothing in the UST command queue");
1895 /* Continue thread execution */
1896 break;
1897 }
1898
1899 ust_cmd = caa_container_of(node, struct ust_command, node);
1900
1901 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1902 " gid:%d sock:%d name:%s (version %d.%d)",
1903 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1904 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1905 ust_cmd->sock, ust_cmd->reg_msg.name,
1906 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1907
1908 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1909 wait_node = zmalloc(sizeof(*wait_node));
1910 if (!wait_node) {
1911 PERROR("zmalloc wait_node dispatch");
1912 ret = close(ust_cmd->sock);
1913 if (ret < 0) {
1914 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1915 }
1916 lttng_fd_put(LTTNG_FD_APPS, 1);
1917 free(ust_cmd);
1918 goto error;
1919 }
1920 CDS_INIT_LIST_HEAD(&wait_node->head);
1921
1922 /* Create application object if socket is CMD. */
1923 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1924 ust_cmd->sock);
1925 if (!wait_node->app) {
1926 ret = close(ust_cmd->sock);
1927 if (ret < 0) {
1928 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1929 }
1930 lttng_fd_put(LTTNG_FD_APPS, 1);
1931 free(wait_node);
1932 free(ust_cmd);
1933 continue;
1934 }
1935 /*
1936 * Add application to the wait queue so we can set the notify
1937 * socket before putting this object in the global ht.
1938 */
1939 cds_list_add(&wait_node->head, &wait_queue.head);
1940 wait_queue.count++;
1941
1942 free(ust_cmd);
1943 /*
1944 * We have to continue here since we don't have the notify
1945 * socket and the application MUST be added to the hash table
1946 * only at that moment.
1947 */
1948 continue;
1949 } else {
1950 /*
1951 * Look for the application in the local wait queue and set the
1952 * notify socket if found.
1953 */
1954 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1955 &wait_queue.head, head) {
1956 health_code_update();
1957 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1958 wait_node->app->notify_sock = ust_cmd->sock;
1959 cds_list_del(&wait_node->head);
1960 wait_queue.count--;
1961 app = wait_node->app;
1962 free(wait_node);
1963 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1964 break;
1965 }
1966 }
1967
1968 /*
1969 * With no application at this stage the received socket is
1970 * basically useless so close it before we free the cmd data
1971 * structure for good.
1972 */
1973 if (!app) {
1974 ret = close(ust_cmd->sock);
1975 if (ret < 0) {
1976 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1977 }
1978 lttng_fd_put(LTTNG_FD_APPS, 1);
1979 }
1980 free(ust_cmd);
1981 }
1982
1983 if (app) {
1984 /*
1985 * @session_lock_list
1986 *
1987 * Lock the global session list so from the register up to the
1988 * registration done message, no thread can see the application
1989 * and change its state.
1990 */
1991 session_lock_list();
1992 rcu_read_lock();
1993
1994 /*
1995 * Add application to the global hash table. This needs to be
1996 * done before the update to the UST registry can locate the
1997 * application.
1998 */
1999 ust_app_add(app);
2000
2001 /* Set app version. This call will print an error if needed. */
2002 (void) ust_app_version(app);
2003
2004 /* Send notify socket through the notify pipe. */
2005 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2006 app->notify_sock);
2007 if (ret < 0) {
2008 rcu_read_unlock();
2009 session_unlock_list();
2010 /*
2011 * No notify thread, stop the UST tracing. However, this is
2012 * not an internal error of the this thread thus setting
2013 * the health error code to a normal exit.
2014 */
2015 err = 0;
2016 goto error;
2017 }
2018
2019 /*
2020 * Update newly registered application with the tracing
2021 * registry info already enabled information.
2022 */
2023 update_ust_app(app->sock);
2024
2025 /*
2026 * Don't care about return value. Let the manage apps threads
2027 * handle app unregistration upon socket close.
2028 */
2029 (void) ust_app_register_done(app);
2030
2031 /*
2032 * Even if the application socket has been closed, send the app
2033 * to the thread and unregistration will take place at that
2034 * place.
2035 */
2036 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2037 if (ret < 0) {
2038 rcu_read_unlock();
2039 session_unlock_list();
2040 /*
2041 * No apps. thread, stop the UST tracing. However, this is
2042 * not an internal error of the this thread thus setting
2043 * the health error code to a normal exit.
2044 */
2045 err = 0;
2046 goto error;
2047 }
2048
2049 rcu_read_unlock();
2050 session_unlock_list();
2051 }
2052 } while (node != NULL);
2053
2054 health_poll_entry();
2055 /* Futex wait on queue. Blocking call on futex() */
2056 futex_nto1_wait(&ust_cmd_queue.futex);
2057 health_poll_exit();
2058 }
2059 /* Normal exit, no error */
2060 err = 0;
2061
2062 error:
2063 /* Clean up wait queue. */
2064 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2065 &wait_queue.head, head) {
2066 cds_list_del(&wait_node->head);
2067 wait_queue.count--;
2068 free(wait_node);
2069 }
2070
2071 /* Empty command queue. */
2072 for (;;) {
2073 /* Dequeue command for registration */
2074 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2075 if (node == NULL) {
2076 break;
2077 }
2078 ust_cmd = caa_container_of(node, struct ust_command, node);
2079 ret = close(ust_cmd->sock);
2080 if (ret < 0) {
2081 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2082 }
2083 lttng_fd_put(LTTNG_FD_APPS, 1);
2084 free(ust_cmd);
2085 }
2086
2087 error_testpoint:
2088 DBG("Dispatch thread dying");
2089 if (err) {
2090 health_error();
2091 ERR("Health error occurred in %s", __func__);
2092 }
2093 health_unregister(health_sessiond);
2094 return NULL;
2095 }
2096
2097 /*
2098 * This thread manage application registration.
2099 */
2100 static void *thread_registration_apps(void *data)
2101 {
2102 int sock = -1, i, ret, pollfd, err = -1;
2103 uint32_t revents, nb_fd;
2104 struct lttng_poll_event events;
2105 /*
2106 * Get allocated in this thread, enqueued to a global queue, dequeued and
2107 * freed in the manage apps thread.
2108 */
2109 struct ust_command *ust_cmd = NULL;
2110
2111 DBG("[thread] Manage application registration started");
2112
2113 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2114
2115 if (testpoint(sessiond_thread_registration_apps)) {
2116 goto error_testpoint;
2117 }
2118
2119 ret = lttcomm_listen_unix_sock(apps_sock);
2120 if (ret < 0) {
2121 goto error_listen;
2122 }
2123
2124 /*
2125 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2126 * more will be added to this poll set.
2127 */
2128 ret = sessiond_set_thread_pollset(&events, 2);
2129 if (ret < 0) {
2130 goto error_create_poll;
2131 }
2132
2133 /* Add the application registration socket */
2134 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2135 if (ret < 0) {
2136 goto error_poll_add;
2137 }
2138
2139 /* Notify all applications to register */
2140 ret = notify_ust_apps(1);
2141 if (ret < 0) {
2142 ERR("Failed to notify applications or create the wait shared memory.\n"
2143 "Execution continues but there might be problem for already\n"
2144 "running applications that wishes to register.");
2145 }
2146
2147 while (1) {
2148 DBG("Accepting application registration");
2149
2150 /* Inifinite blocking call, waiting for transmission */
2151 restart:
2152 health_poll_entry();
2153 ret = lttng_poll_wait(&events, -1);
2154 health_poll_exit();
2155 if (ret < 0) {
2156 /*
2157 * Restart interrupted system call.
2158 */
2159 if (errno == EINTR) {
2160 goto restart;
2161 }
2162 goto error;
2163 }
2164
2165 nb_fd = ret;
2166
2167 for (i = 0; i < nb_fd; i++) {
2168 health_code_update();
2169
2170 /* Fetch once the poll data */
2171 revents = LTTNG_POLL_GETEV(&events, i);
2172 pollfd = LTTNG_POLL_GETFD(&events, i);
2173
2174 if (!revents) {
2175 /* No activity for this FD (poll implementation). */
2176 continue;
2177 }
2178
2179 /* Thread quit pipe has been closed. Killing thread. */
2180 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2181 if (ret) {
2182 err = 0;
2183 goto exit;
2184 }
2185
2186 /* Event on the registration socket */
2187 if (pollfd == apps_sock) {
2188 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2189 ERR("Register apps socket poll error");
2190 goto error;
2191 } else if (revents & LPOLLIN) {
2192 sock = lttcomm_accept_unix_sock(apps_sock);
2193 if (sock < 0) {
2194 goto error;
2195 }
2196
2197 /*
2198 * Set socket timeout for both receiving and ending.
2199 * app_socket_timeout is in seconds, whereas
2200 * lttcomm_setsockopt_rcv_timeout and
2201 * lttcomm_setsockopt_snd_timeout expect msec as
2202 * parameter.
2203 */
2204 (void) lttcomm_setsockopt_rcv_timeout(sock,
2205 app_socket_timeout * 1000);
2206 (void) lttcomm_setsockopt_snd_timeout(sock,
2207 app_socket_timeout * 1000);
2208
2209 /*
2210 * Set the CLOEXEC flag. Return code is useless because
2211 * either way, the show must go on.
2212 */
2213 (void) utils_set_fd_cloexec(sock);
2214
2215 /* Create UST registration command for enqueuing */
2216 ust_cmd = zmalloc(sizeof(struct ust_command));
2217 if (ust_cmd == NULL) {
2218 PERROR("ust command zmalloc");
2219 ret = close(sock);
2220 if (ret) {
2221 PERROR("close");
2222 }
2223 goto error;
2224 }
2225
2226 /*
2227 * Using message-based transmissions to ensure we don't
2228 * have to deal with partially received messages.
2229 */
2230 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2231 if (ret < 0) {
2232 ERR("Exhausted file descriptors allowed for applications.");
2233 free(ust_cmd);
2234 ret = close(sock);
2235 if (ret) {
2236 PERROR("close");
2237 }
2238 sock = -1;
2239 continue;
2240 }
2241
2242 health_code_update();
2243 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2244 if (ret < 0) {
2245 free(ust_cmd);
2246 /* Close socket of the application. */
2247 ret = close(sock);
2248 if (ret) {
2249 PERROR("close");
2250 }
2251 lttng_fd_put(LTTNG_FD_APPS, 1);
2252 sock = -1;
2253 continue;
2254 }
2255 health_code_update();
2256
2257 ust_cmd->sock = sock;
2258 sock = -1;
2259
2260 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2261 " gid:%d sock:%d name:%s (version %d.%d)",
2262 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2263 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2264 ust_cmd->sock, ust_cmd->reg_msg.name,
2265 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2266
2267 /*
2268 * Lock free enqueue the registration request. The red pill
2269 * has been taken! This apps will be part of the *system*.
2270 */
2271 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2272
2273 /*
2274 * Wake the registration queue futex. Implicit memory
2275 * barrier with the exchange in cds_wfcq_enqueue.
2276 */
2277 futex_nto1_wake(&ust_cmd_queue.futex);
2278 }
2279 }
2280 }
2281 }
2282
2283 exit:
2284 error:
2285 /* Notify that the registration thread is gone */
2286 notify_ust_apps(0);
2287
2288 if (apps_sock >= 0) {
2289 ret = close(apps_sock);
2290 if (ret) {
2291 PERROR("close");
2292 }
2293 }
2294 if (sock >= 0) {
2295 ret = close(sock);
2296 if (ret) {
2297 PERROR("close");
2298 }
2299 lttng_fd_put(LTTNG_FD_APPS, 1);
2300 }
2301 unlink(apps_unix_sock_path);
2302
2303 error_poll_add:
2304 lttng_poll_clean(&events);
2305 error_listen:
2306 error_create_poll:
2307 error_testpoint:
2308 DBG("UST Registration thread cleanup complete");
2309 if (err) {
2310 health_error();
2311 ERR("Health error occurred in %s", __func__);
2312 }
2313 health_unregister(health_sessiond);
2314
2315 return NULL;
2316 }
2317
2318 /*
2319 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2320 * exec or it will fails.
2321 */
2322 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2323 {
2324 int ret, clock_ret;
2325 struct timespec timeout;
2326
2327 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2328 consumer_data->consumer_thread_is_ready = 0;
2329
2330 /* Setup pthread condition */
2331 ret = pthread_condattr_init(&consumer_data->condattr);
2332 if (ret) {
2333 errno = ret;
2334 PERROR("pthread_condattr_init consumer data");
2335 goto error;
2336 }
2337
2338 /*
2339 * Set the monotonic clock in order to make sure we DO NOT jump in time
2340 * between the clock_gettime() call and the timedwait call. See bug #324
2341 * for a more details and how we noticed it.
2342 */
2343 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2344 if (ret) {
2345 errno = ret;
2346 PERROR("pthread_condattr_setclock consumer data");
2347 goto error;
2348 }
2349
2350 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2351 if (ret) {
2352 errno = ret;
2353 PERROR("pthread_cond_init consumer data");
2354 goto error;
2355 }
2356
2357 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2358 consumer_data);
2359 if (ret) {
2360 errno = ret;
2361 PERROR("pthread_create consumer");
2362 ret = -1;
2363 goto error;
2364 }
2365
2366 /* We are about to wait on a pthread condition */
2367 pthread_mutex_lock(&consumer_data->cond_mutex);
2368
2369 /* Get time for sem_timedwait absolute timeout */
2370 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2371 /*
2372 * Set the timeout for the condition timed wait even if the clock gettime
2373 * call fails since we might loop on that call and we want to avoid to
2374 * increment the timeout too many times.
2375 */
2376 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2377
2378 /*
2379 * The following loop COULD be skipped in some conditions so this is why we
2380 * set ret to 0 in order to make sure at least one round of the loop is
2381 * done.
2382 */
2383 ret = 0;
2384
2385 /*
2386 * Loop until the condition is reached or when a timeout is reached. Note
2387 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2388 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2389 * possible. This loop does not take any chances and works with both of
2390 * them.
2391 */
2392 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2393 if (clock_ret < 0) {
2394 PERROR("clock_gettime spawn consumer");
2395 /* Infinite wait for the consumerd thread to be ready */
2396 ret = pthread_cond_wait(&consumer_data->cond,
2397 &consumer_data->cond_mutex);
2398 } else {
2399 ret = pthread_cond_timedwait(&consumer_data->cond,
2400 &consumer_data->cond_mutex, &timeout);
2401 }
2402 }
2403
2404 /* Release the pthread condition */
2405 pthread_mutex_unlock(&consumer_data->cond_mutex);
2406
2407 if (ret != 0) {
2408 errno = ret;
2409 if (ret == ETIMEDOUT) {
2410 int pth_ret;
2411
2412 /*
2413 * Call has timed out so we kill the kconsumerd_thread and return
2414 * an error.
2415 */
2416 ERR("Condition timed out. The consumer thread was never ready."
2417 " Killing it");
2418 pth_ret = pthread_cancel(consumer_data->thread);
2419 if (pth_ret < 0) {
2420 PERROR("pthread_cancel consumer thread");
2421 }
2422 } else {
2423 PERROR("pthread_cond_wait failed consumer thread");
2424 }
2425 /* Caller is expecting a negative value on failure. */
2426 ret = -1;
2427 goto error;
2428 }
2429
2430 pthread_mutex_lock(&consumer_data->pid_mutex);
2431 if (consumer_data->pid == 0) {
2432 ERR("Consumerd did not start");
2433 pthread_mutex_unlock(&consumer_data->pid_mutex);
2434 goto error;
2435 }
2436 pthread_mutex_unlock(&consumer_data->pid_mutex);
2437
2438 return 0;
2439
2440 error:
2441 return ret;
2442 }
2443
2444 /*
2445 * Join consumer thread
2446 */
2447 static int join_consumer_thread(struct consumer_data *consumer_data)
2448 {
2449 void *status;
2450
2451 /* Consumer pid must be a real one. */
2452 if (consumer_data->pid > 0) {
2453 int ret;
2454 ret = kill(consumer_data->pid, SIGTERM);
2455 if (ret) {
2456 PERROR("Error killing consumer daemon");
2457 return ret;
2458 }
2459 return pthread_join(consumer_data->thread, &status);
2460 } else {
2461 return 0;
2462 }
2463 }
2464
2465 /*
2466 * Fork and exec a consumer daemon (consumerd).
2467 *
2468 * Return pid if successful else -1.
2469 */
2470 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2471 {
2472 int ret;
2473 pid_t pid;
2474 const char *consumer_to_use;
2475 const char *verbosity;
2476 struct stat st;
2477
2478 DBG("Spawning consumerd");
2479
2480 pid = fork();
2481 if (pid == 0) {
2482 /*
2483 * Exec consumerd.
2484 */
2485 if (opt_verbose_consumer) {
2486 verbosity = "--verbose";
2487 } else if (lttng_opt_quiet) {
2488 verbosity = "--quiet";
2489 } else {
2490 verbosity = "";
2491 }
2492
2493 switch (consumer_data->type) {
2494 case LTTNG_CONSUMER_KERNEL:
2495 /*
2496 * Find out which consumerd to execute. We will first try the
2497 * 64-bit path, then the sessiond's installation directory, and
2498 * fallback on the 32-bit one,
2499 */
2500 DBG3("Looking for a kernel consumer at these locations:");
2501 DBG3(" 1) %s", consumerd64_bin);
2502 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2503 DBG3(" 3) %s", consumerd32_bin);
2504 if (stat(consumerd64_bin, &st) == 0) {
2505 DBG3("Found location #1");
2506 consumer_to_use = consumerd64_bin;
2507 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2508 DBG3("Found location #2");
2509 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2510 } else if (stat(consumerd32_bin, &st) == 0) {
2511 DBG3("Found location #3");
2512 consumer_to_use = consumerd32_bin;
2513 } else {
2514 DBG("Could not find any valid consumerd executable");
2515 ret = -EINVAL;
2516 break;
2517 }
2518 DBG("Using kernel consumer at: %s", consumer_to_use);
2519 ret = execl(consumer_to_use,
2520 "lttng-consumerd", verbosity, "-k",
2521 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2522 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2523 "--group", tracing_group_name,
2524 NULL);
2525 break;
2526 case LTTNG_CONSUMER64_UST:
2527 {
2528 char *tmpnew = NULL;
2529
2530 if (consumerd64_libdir[0] != '\0') {
2531 char *tmp;
2532 size_t tmplen;
2533
2534 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2535 if (!tmp) {
2536 tmp = "";
2537 }
2538 tmplen = strlen("LD_LIBRARY_PATH=")
2539 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2540 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2541 if (!tmpnew) {
2542 ret = -ENOMEM;
2543 goto error;
2544 }
2545 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2546 strcat(tmpnew, consumerd64_libdir);
2547 if (tmp[0] != '\0') {
2548 strcat(tmpnew, ":");
2549 strcat(tmpnew, tmp);
2550 }
2551 ret = putenv(tmpnew);
2552 if (ret) {
2553 ret = -errno;
2554 free(tmpnew);
2555 goto error;
2556 }
2557 }
2558 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2559 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2560 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2561 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2562 "--group", tracing_group_name,
2563 NULL);
2564 if (consumerd64_libdir[0] != '\0') {
2565 free(tmpnew);
2566 }
2567 break;
2568 }
2569 case LTTNG_CONSUMER32_UST:
2570 {
2571 char *tmpnew = NULL;
2572
2573 if (consumerd32_libdir[0] != '\0') {
2574 char *tmp;
2575 size_t tmplen;
2576
2577 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2578 if (!tmp) {
2579 tmp = "";
2580 }
2581 tmplen = strlen("LD_LIBRARY_PATH=")
2582 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2583 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2584 if (!tmpnew) {
2585 ret = -ENOMEM;
2586 goto error;
2587 }
2588 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2589 strcat(tmpnew, consumerd32_libdir);
2590 if (tmp[0] != '\0') {
2591 strcat(tmpnew, ":");
2592 strcat(tmpnew, tmp);
2593 }
2594 ret = putenv(tmpnew);
2595 if (ret) {
2596 ret = -errno;
2597 free(tmpnew);
2598 goto error;
2599 }
2600 }
2601 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2602 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2603 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2604 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2605 "--group", tracing_group_name,
2606 NULL);
2607 if (consumerd32_libdir[0] != '\0') {
2608 free(tmpnew);
2609 }
2610 break;
2611 }
2612 default:
2613 PERROR("unknown consumer type");
2614 exit(EXIT_FAILURE);
2615 }
2616 if (errno != 0) {
2617 PERROR("Consumer execl()");
2618 }
2619 /* Reaching this point, we got a failure on our execl(). */
2620 exit(EXIT_FAILURE);
2621 } else if (pid > 0) {
2622 ret = pid;
2623 } else {
2624 PERROR("start consumer fork");
2625 ret = -errno;
2626 }
2627 error:
2628 return ret;
2629 }
2630
2631 /*
2632 * Spawn the consumerd daemon and session daemon thread.
2633 */
2634 static int start_consumerd(struct consumer_data *consumer_data)
2635 {
2636 int ret;
2637
2638 /*
2639 * Set the listen() state on the socket since there is a possible race
2640 * between the exec() of the consumer daemon and this call if place in the
2641 * consumer thread. See bug #366 for more details.
2642 */
2643 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2644 if (ret < 0) {
2645 goto error;
2646 }
2647
2648 pthread_mutex_lock(&consumer_data->pid_mutex);
2649 if (consumer_data->pid != 0) {
2650 pthread_mutex_unlock(&consumer_data->pid_mutex);
2651 goto end;
2652 }
2653
2654 ret = spawn_consumerd(consumer_data);
2655 if (ret < 0) {
2656 ERR("Spawning consumerd failed");
2657 pthread_mutex_unlock(&consumer_data->pid_mutex);
2658 goto error;
2659 }
2660
2661 /* Setting up the consumer_data pid */
2662 consumer_data->pid = ret;
2663 DBG2("Consumer pid %d", consumer_data->pid);
2664 pthread_mutex_unlock(&consumer_data->pid_mutex);
2665
2666 DBG2("Spawning consumer control thread");
2667 ret = spawn_consumer_thread(consumer_data);
2668 if (ret < 0) {
2669 ERR("Fatal error spawning consumer control thread");
2670 goto error;
2671 }
2672
2673 end:
2674 return 0;
2675
2676 error:
2677 /* Cleanup already created sockets on error. */
2678 if (consumer_data->err_sock >= 0) {
2679 int err;
2680
2681 err = close(consumer_data->err_sock);
2682 if (err < 0) {
2683 PERROR("close consumer data error socket");
2684 }
2685 }
2686 return ret;
2687 }
2688
2689 /*
2690 * Setup necessary data for kernel tracer action.
2691 */
2692 static int init_kernel_tracer(void)
2693 {
2694 int ret;
2695
2696 /* Modprobe lttng kernel modules */
2697 ret = modprobe_lttng_control();
2698 if (ret < 0) {
2699 goto error;
2700 }
2701
2702 /* Open debugfs lttng */
2703 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2704 if (kernel_tracer_fd < 0) {
2705 DBG("Failed to open %s", module_proc_lttng);
2706 ret = -1;
2707 goto error_open;
2708 }
2709
2710 /* Validate kernel version */
2711 ret = kernel_validate_version(kernel_tracer_fd);
2712 if (ret < 0) {
2713 goto error_version;
2714 }
2715
2716 ret = modprobe_lttng_data();
2717 if (ret < 0) {
2718 goto error_modules;
2719 }
2720
2721 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2722 return 0;
2723
2724 error_version:
2725 modprobe_remove_lttng_control();
2726 ret = close(kernel_tracer_fd);
2727 if (ret) {
2728 PERROR("close");
2729 }
2730 kernel_tracer_fd = -1;
2731 return LTTNG_ERR_KERN_VERSION;
2732
2733 error_modules:
2734 ret = close(kernel_tracer_fd);
2735 if (ret) {
2736 PERROR("close");
2737 }
2738
2739 error_open:
2740 modprobe_remove_lttng_control();
2741
2742 error:
2743 WARN("No kernel tracer available");
2744 kernel_tracer_fd = -1;
2745 if (!is_root) {
2746 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2747 } else {
2748 return LTTNG_ERR_KERN_NA;
2749 }
2750 }
2751
2752
2753 /*
2754 * Copy consumer output from the tracing session to the domain session. The
2755 * function also applies the right modification on a per domain basis for the
2756 * trace files destination directory.
2757 *
2758 * Should *NOT* be called with RCU read-side lock held.
2759 */
2760 static int copy_session_consumer(int domain, struct ltt_session *session)
2761 {
2762 int ret;
2763 const char *dir_name;
2764 struct consumer_output *consumer;
2765
2766 assert(session);
2767 assert(session->consumer);
2768
2769 switch (domain) {
2770 case LTTNG_DOMAIN_KERNEL:
2771 DBG3("Copying tracing session consumer output in kernel session");
2772 /*
2773 * XXX: We should audit the session creation and what this function
2774 * does "extra" in order to avoid a destroy since this function is used
2775 * in the domain session creation (kernel and ust) only. Same for UST
2776 * domain.
2777 */
2778 if (session->kernel_session->consumer) {
2779 consumer_output_put(session->kernel_session->consumer);
2780 }
2781 session->kernel_session->consumer =
2782 consumer_copy_output(session->consumer);
2783 /* Ease our life a bit for the next part */
2784 consumer = session->kernel_session->consumer;
2785 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2786 break;
2787 case LTTNG_DOMAIN_JUL:
2788 case LTTNG_DOMAIN_LOG4J:
2789 case LTTNG_DOMAIN_PYTHON:
2790 case LTTNG_DOMAIN_UST:
2791 DBG3("Copying tracing session consumer output in UST session");
2792 if (session->ust_session->consumer) {
2793 consumer_output_put(session->ust_session->consumer);
2794 }
2795 session->ust_session->consumer =
2796 consumer_copy_output(session->consumer);
2797 /* Ease our life a bit for the next part */
2798 consumer = session->ust_session->consumer;
2799 dir_name = DEFAULT_UST_TRACE_DIR;
2800 break;
2801 default:
2802 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2803 goto error;
2804 }
2805
2806 /* Append correct directory to subdir */
2807 strncat(consumer->subdir, dir_name,
2808 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2809 DBG3("Copy session consumer subdir %s", consumer->subdir);
2810
2811 ret = LTTNG_OK;
2812
2813 error:
2814 return ret;
2815 }
2816
2817 /*
2818 * Create an UST session and add it to the session ust list.
2819 *
2820 * Should *NOT* be called with RCU read-side lock held.
2821 */
2822 static int create_ust_session(struct ltt_session *session,
2823 struct lttng_domain *domain)
2824 {
2825 int ret;
2826 struct ltt_ust_session *lus = NULL;
2827
2828 assert(session);
2829 assert(domain);
2830 assert(session->consumer);
2831
2832 switch (domain->type) {
2833 case LTTNG_DOMAIN_JUL:
2834 case LTTNG_DOMAIN_LOG4J:
2835 case LTTNG_DOMAIN_PYTHON:
2836 case LTTNG_DOMAIN_UST:
2837 break;
2838 default:
2839 ERR("Unknown UST domain on create session %d", domain->type);
2840 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2841 goto error;
2842 }
2843
2844 DBG("Creating UST session");
2845
2846 lus = trace_ust_create_session(session->id);
2847 if (lus == NULL) {
2848 ret = LTTNG_ERR_UST_SESS_FAIL;
2849 goto error;
2850 }
2851
2852 lus->uid = session->uid;
2853 lus->gid = session->gid;
2854 lus->output_traces = session->output_traces;
2855 lus->snapshot_mode = session->snapshot_mode;
2856 lus->live_timer_interval = session->live_timer;
2857 session->ust_session = lus;
2858 if (session->shm_path[0]) {
2859 strncpy(lus->root_shm_path, session->shm_path,
2860 sizeof(lus->root_shm_path));
2861 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2862 strncpy(lus->shm_path, session->shm_path,
2863 sizeof(lus->shm_path));
2864 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2865 strncat(lus->shm_path, "/ust",
2866 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2867 }
2868 /* Copy session output to the newly created UST session */
2869 ret = copy_session_consumer(domain->type, session);
2870 if (ret != LTTNG_OK) {
2871 goto error;
2872 }
2873
2874 return LTTNG_OK;
2875
2876 error:
2877 free(lus);
2878 session->ust_session = NULL;
2879 return ret;
2880 }
2881
2882 /*
2883 * Create a kernel tracer session then create the default channel.
2884 */
2885 static int create_kernel_session(struct ltt_session *session)
2886 {
2887 int ret;
2888
2889 DBG("Creating kernel session");
2890
2891 ret = kernel_create_session(session, kernel_tracer_fd);
2892 if (ret < 0) {
2893 ret = LTTNG_ERR_KERN_SESS_FAIL;
2894 goto error;
2895 }
2896
2897 /* Code flow safety */
2898 assert(session->kernel_session);
2899
2900 /* Copy session output to the newly created Kernel session */
2901 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2902 if (ret != LTTNG_OK) {
2903 goto error;
2904 }
2905
2906 /* Create directory(ies) on local filesystem. */
2907 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2908 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2909 ret = run_as_mkdir_recursive(
2910 session->kernel_session->consumer->dst.trace_path,
2911 S_IRWXU | S_IRWXG, session->uid, session->gid);
2912 if (ret < 0) {
2913 if (errno != EEXIST) {
2914 ERR("Trace directory creation error");
2915 goto error;
2916 }
2917 }
2918 }
2919
2920 session->kernel_session->uid = session->uid;
2921 session->kernel_session->gid = session->gid;
2922 session->kernel_session->output_traces = session->output_traces;
2923 session->kernel_session->snapshot_mode = session->snapshot_mode;
2924
2925 return LTTNG_OK;
2926
2927 error:
2928 trace_kernel_destroy_session(session->kernel_session);
2929 session->kernel_session = NULL;
2930 return ret;
2931 }
2932
2933 /*
2934 * Count number of session permitted by uid/gid.
2935 */
2936 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2937 {
2938 unsigned int i = 0;
2939 struct ltt_session *session;
2940
2941 DBG("Counting number of available session for UID %d GID %d",
2942 uid, gid);
2943 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2944 /*
2945 * Only list the sessions the user can control.
2946 */
2947 if (!session_access_ok(session, uid, gid)) {
2948 continue;
2949 }
2950 i++;
2951 }
2952 return i;
2953 }
2954
2955 /*
2956 * Process the command requested by the lttng client within the command
2957 * context structure. This function make sure that the return structure (llm)
2958 * is set and ready for transmission before returning.
2959 *
2960 * Return any error encountered or 0 for success.
2961 *
2962 * "sock" is only used for special-case var. len data.
2963 *
2964 * Should *NOT* be called with RCU read-side lock held.
2965 */
2966 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2967 int *sock_error)
2968 {
2969 int ret = LTTNG_OK;
2970 int need_tracing_session = 1;
2971 int need_domain;
2972
2973 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2974
2975 assert(!rcu_read_ongoing());
2976
2977 *sock_error = 0;
2978
2979 switch (cmd_ctx->lsm->cmd_type) {
2980 case LTTNG_CREATE_SESSION:
2981 case LTTNG_CREATE_SESSION_SNAPSHOT:
2982 case LTTNG_CREATE_SESSION_LIVE:
2983 case LTTNG_DESTROY_SESSION:
2984 case LTTNG_LIST_SESSIONS:
2985 case LTTNG_LIST_DOMAINS:
2986 case LTTNG_START_TRACE:
2987 case LTTNG_STOP_TRACE:
2988 case LTTNG_DATA_PENDING:
2989 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2990 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2991 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2992 case LTTNG_SNAPSHOT_RECORD:
2993 case LTTNG_SAVE_SESSION:
2994 case LTTNG_SET_SESSION_SHM_PATH:
2995 need_domain = 0;
2996 break;
2997 default:
2998 need_domain = 1;
2999 }
3000
3001 if (opt_no_kernel && need_domain
3002 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3003 if (!is_root) {
3004 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3005 } else {
3006 ret = LTTNG_ERR_KERN_NA;
3007 }
3008 goto error;
3009 }
3010
3011 /* Deny register consumer if we already have a spawned consumer. */
3012 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3013 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3014 if (kconsumer_data.pid > 0) {
3015 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3016 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3017 goto error;
3018 }
3019 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3020 }
3021
3022 /*
3023 * Check for command that don't needs to allocate a returned payload. We do
3024 * this here so we don't have to make the call for no payload at each
3025 * command.
3026 */
3027 switch(cmd_ctx->lsm->cmd_type) {
3028 case LTTNG_LIST_SESSIONS:
3029 case LTTNG_LIST_TRACEPOINTS:
3030 case LTTNG_LIST_TRACEPOINT_FIELDS:
3031 case LTTNG_LIST_DOMAINS:
3032 case LTTNG_LIST_CHANNELS:
3033 case LTTNG_LIST_EVENTS:
3034 case LTTNG_LIST_SYSCALLS:
3035 case LTTNG_LIST_TRACKER_PIDS:
3036 break;
3037 default:
3038 /* Setup lttng message with no payload */
3039 ret = setup_lttng_msg(cmd_ctx, 0);
3040 if (ret < 0) {
3041 /* This label does not try to unlock the session */
3042 goto init_setup_error;
3043 }
3044 }
3045
3046 /* Commands that DO NOT need a session. */
3047 switch (cmd_ctx->lsm->cmd_type) {
3048 case LTTNG_CREATE_SESSION:
3049 case LTTNG_CREATE_SESSION_SNAPSHOT:
3050 case LTTNG_CREATE_SESSION_LIVE:
3051 case LTTNG_CALIBRATE:
3052 case LTTNG_LIST_SESSIONS:
3053 case LTTNG_LIST_TRACEPOINTS:
3054 case LTTNG_LIST_SYSCALLS:
3055 case LTTNG_LIST_TRACEPOINT_FIELDS:
3056 case LTTNG_SAVE_SESSION:
3057 need_tracing_session = 0;
3058 break;
3059 default:
3060 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3061 /*
3062 * We keep the session list lock across _all_ commands
3063 * for now, because the per-session lock does not
3064 * handle teardown properly.
3065 */
3066 session_lock_list();
3067 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3068 if (cmd_ctx->session == NULL) {
3069 ret = LTTNG_ERR_SESS_NOT_FOUND;
3070 goto error;
3071 } else {
3072 /* Acquire lock for the session */
3073 session_lock(cmd_ctx->session);
3074 }
3075 break;
3076 }
3077
3078 /*
3079 * Commands that need a valid session but should NOT create one if none
3080 * exists. Instead of creating one and destroying it when the command is
3081 * handled, process that right before so we save some round trip in useless
3082 * code path.
3083 */
3084 switch (cmd_ctx->lsm->cmd_type) {
3085 case LTTNG_DISABLE_CHANNEL:
3086 case LTTNG_DISABLE_EVENT:
3087 switch (cmd_ctx->lsm->domain.type) {
3088 case LTTNG_DOMAIN_KERNEL:
3089 if (!cmd_ctx->session->kernel_session) {
3090 ret = LTTNG_ERR_NO_CHANNEL;
3091 goto error;
3092 }
3093 break;
3094 case LTTNG_DOMAIN_JUL:
3095 case LTTNG_DOMAIN_LOG4J:
3096 case LTTNG_DOMAIN_PYTHON:
3097 case LTTNG_DOMAIN_UST:
3098 if (!cmd_ctx->session->ust_session) {
3099 ret = LTTNG_ERR_NO_CHANNEL;
3100 goto error;
3101 }
3102 break;
3103 default:
3104 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3105 goto error;
3106 }
3107 default:
3108 break;
3109 }
3110
3111 if (!need_domain) {
3112 goto skip_domain;
3113 }
3114
3115 /*
3116 * Check domain type for specific "pre-action".
3117 */
3118 switch (cmd_ctx->lsm->domain.type) {
3119 case LTTNG_DOMAIN_KERNEL:
3120 if (!is_root) {
3121 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3122 goto error;
3123 }
3124
3125 /* Kernel tracer check */
3126 if (kernel_tracer_fd == -1) {
3127 /* Basically, load kernel tracer modules */
3128 ret = init_kernel_tracer();
3129 if (ret != 0) {
3130 goto error;
3131 }
3132 }
3133
3134 /* Consumer is in an ERROR state. Report back to client */
3135 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3136 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3137 goto error;
3138 }
3139
3140 /* Need a session for kernel command */
3141 if (need_tracing_session) {
3142 if (cmd_ctx->session->kernel_session == NULL) {
3143 ret = create_kernel_session(cmd_ctx->session);
3144 if (ret < 0) {
3145 ret = LTTNG_ERR_KERN_SESS_FAIL;
3146 goto error;
3147 }
3148 }
3149
3150 /* Start the kernel consumer daemon */
3151 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3152 if (kconsumer_data.pid == 0 &&
3153 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3154 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3155 ret = start_consumerd(&kconsumer_data);
3156 if (ret < 0) {
3157 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3158 goto error;
3159 }
3160 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3161 } else {
3162 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3163 }
3164
3165 /*
3166 * The consumer was just spawned so we need to add the socket to
3167 * the consumer output of the session if exist.
3168 */
3169 ret = consumer_create_socket(&kconsumer_data,
3170 cmd_ctx->session->kernel_session->consumer);
3171 if (ret < 0) {
3172 goto error;
3173 }
3174 }
3175
3176 break;
3177 case LTTNG_DOMAIN_JUL:
3178 case LTTNG_DOMAIN_LOG4J:
3179 case LTTNG_DOMAIN_PYTHON:
3180 case LTTNG_DOMAIN_UST:
3181 {
3182 if (!ust_app_supported()) {
3183 ret = LTTNG_ERR_NO_UST;
3184 goto error;
3185 }
3186 /* Consumer is in an ERROR state. Report back to client */
3187 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3188 ret = LTTNG_ERR_NO_USTCONSUMERD;
3189 goto error;
3190 }
3191
3192 if (need_tracing_session) {
3193 /* Create UST session if none exist. */
3194 if (cmd_ctx->session->ust_session == NULL) {
3195 ret = create_ust_session(cmd_ctx->session,
3196 &cmd_ctx->lsm->domain);
3197 if (ret != LTTNG_OK) {
3198 goto error;
3199 }
3200 }
3201
3202 /* Start the UST consumer daemons */
3203 /* 64-bit */
3204 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3205 if (consumerd64_bin[0] != '\0' &&
3206 ustconsumer64_data.pid == 0 &&
3207 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3208 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3209 ret = start_consumerd(&ustconsumer64_data);
3210 if (ret < 0) {
3211 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3212 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3213 goto error;
3214 }
3215
3216 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3217 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3218 } else {
3219 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3220 }
3221
3222 /*
3223 * Setup socket for consumer 64 bit. No need for atomic access
3224 * since it was set above and can ONLY be set in this thread.
3225 */
3226 ret = consumer_create_socket(&ustconsumer64_data,
3227 cmd_ctx->session->ust_session->consumer);
3228 if (ret < 0) {
3229 goto error;
3230 }
3231
3232 /* 32-bit */
3233 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3234 if (consumerd32_bin[0] != '\0' &&
3235 ustconsumer32_data.pid == 0 &&
3236 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3237 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3238 ret = start_consumerd(&ustconsumer32_data);
3239 if (ret < 0) {
3240 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3241 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3242 goto error;
3243 }
3244
3245 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3246 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3247 } else {
3248 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3249 }
3250
3251 /*
3252 * Setup socket for consumer 64 bit. No need for atomic access
3253 * since it was set above and can ONLY be set in this thread.
3254 */
3255 ret = consumer_create_socket(&ustconsumer32_data,
3256 cmd_ctx->session->ust_session->consumer);
3257 if (ret < 0) {
3258 goto error;
3259 }
3260 }
3261 break;
3262 }
3263 default:
3264 break;
3265 }
3266 skip_domain:
3267
3268 /* Validate consumer daemon state when start/stop trace command */
3269 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3270 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3271 switch (cmd_ctx->lsm->domain.type) {
3272 case LTTNG_DOMAIN_JUL:
3273 case LTTNG_DOMAIN_LOG4J:
3274 case LTTNG_DOMAIN_PYTHON:
3275 case LTTNG_DOMAIN_UST:
3276 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3277 ret = LTTNG_ERR_NO_USTCONSUMERD;
3278 goto error;
3279 }
3280 break;
3281 case LTTNG_DOMAIN_KERNEL:
3282 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3283 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3284 goto error;
3285 }
3286 break;
3287 }
3288 }
3289
3290 /*
3291 * Check that the UID or GID match that of the tracing session.
3292 * The root user can interact with all sessions.
3293 */
3294 if (need_tracing_session) {
3295 if (!session_access_ok(cmd_ctx->session,
3296 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3297 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3298 ret = LTTNG_ERR_EPERM;
3299 goto error;
3300 }
3301 }
3302
3303 /*
3304 * Send relayd information to consumer as soon as we have a domain and a
3305 * session defined.
3306 */
3307 if (cmd_ctx->session && need_domain) {
3308 /*
3309 * Setup relayd if not done yet. If the relayd information was already
3310 * sent to the consumer, this call will gracefully return.
3311 */
3312 ret = cmd_setup_relayd(cmd_ctx->session);
3313 if (ret != LTTNG_OK) {
3314 goto error;
3315 }
3316 }
3317
3318 /* Process by command type */
3319 switch (cmd_ctx->lsm->cmd_type) {
3320 case LTTNG_ADD_CONTEXT:
3321 {
3322 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3323 cmd_ctx->lsm->u.context.channel_name,
3324 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3325 break;
3326 }
3327 case LTTNG_DISABLE_CHANNEL:
3328 {
3329 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3330 cmd_ctx->lsm->u.disable.channel_name);
3331 break;
3332 }
3333 case LTTNG_DISABLE_EVENT:
3334 {
3335
3336 /*
3337 * FIXME: handle filter; for now we just receive the filter's
3338 * bytecode along with the filter expression which are sent by
3339 * liblttng-ctl and discard them.
3340 *
3341 * This fixes an issue where the client may block while sending
3342 * the filter payload and encounter an error because the session
3343 * daemon closes the socket without ever handling this data.
3344 */
3345 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3346 cmd_ctx->lsm->u.disable.bytecode_len;
3347
3348 if (count) {
3349 char data[LTTNG_FILTER_MAX_LEN];
3350
3351 DBG("Discarding disable event command payload of size %zu", count);
3352 while (count) {
3353 ret = lttcomm_recv_unix_sock(sock, data,
3354 count > sizeof(data) ? sizeof(data) : count);
3355 if (ret < 0) {
3356 goto error;
3357 }
3358
3359 count -= (size_t) ret;
3360 }
3361 }
3362 /* FIXME: passing packed structure to non-packed pointer */
3363 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3364 cmd_ctx->lsm->u.disable.channel_name,
3365 &cmd_ctx->lsm->u.disable.event);
3366 break;
3367 }
3368 case LTTNG_ENABLE_CHANNEL:
3369 {
3370 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3371 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3372 break;
3373 }
3374 case LTTNG_TRACK_PID:
3375 {
3376 ret = cmd_track_pid(cmd_ctx->session,
3377 cmd_ctx->lsm->domain.type,
3378 cmd_ctx->lsm->u.pid_tracker.pid);
3379 break;
3380 }
3381 case LTTNG_UNTRACK_PID:
3382 {
3383 ret = cmd_untrack_pid(cmd_ctx->session,
3384 cmd_ctx->lsm->domain.type,
3385 cmd_ctx->lsm->u.pid_tracker.pid);
3386 break;
3387 }
3388 case LTTNG_ENABLE_EVENT:
3389 {
3390 struct lttng_event_exclusion *exclusion = NULL;
3391 struct lttng_filter_bytecode *bytecode = NULL;
3392 char *filter_expression = NULL;
3393
3394 /* Handle exclusion events and receive it from the client. */
3395 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3396 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3397
3398 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3399 (count * LTTNG_SYMBOL_NAME_LEN));
3400 if (!exclusion) {
3401 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3402 goto error;
3403 }
3404
3405 DBG("Receiving var len exclusion event list from client ...");
3406 exclusion->count = count;
3407 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3408 count * LTTNG_SYMBOL_NAME_LEN);
3409 if (ret <= 0) {
3410 DBG("Nothing recv() from client var len data... continuing");
3411 *sock_error = 1;
3412 free(exclusion);
3413 ret = LTTNG_ERR_EXCLUSION_INVAL;
3414 goto error;
3415 }
3416 }
3417
3418 /* Get filter expression from client. */
3419 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3420 size_t expression_len =
3421 cmd_ctx->lsm->u.enable.expression_len;
3422
3423 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3424 ret = LTTNG_ERR_FILTER_INVAL;
3425 free(exclusion);
3426 goto error;
3427 }
3428
3429 filter_expression = zmalloc(expression_len);
3430 if (!filter_expression) {
3431 free(exclusion);
3432 ret = LTTNG_ERR_FILTER_NOMEM;
3433 goto error;
3434 }
3435
3436 /* Receive var. len. data */
3437 DBG("Receiving var len filter's expression from client ...");
3438 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3439 expression_len);
3440 if (ret <= 0) {
3441 DBG("Nothing recv() from client car len data... continuing");
3442 *sock_error = 1;
3443 free(filter_expression);
3444 free(exclusion);
3445 ret = LTTNG_ERR_FILTER_INVAL;
3446 goto error;
3447 }
3448 }
3449
3450 /* Handle filter and get bytecode from client. */
3451 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3452 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3453
3454 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3455 ret = LTTNG_ERR_FILTER_INVAL;
3456 free(filter_expression);
3457 free(exclusion);
3458 goto error;
3459 }
3460
3461 bytecode = zmalloc(bytecode_len);
3462 if (!bytecode) {
3463 free(filter_expression);
3464 free(exclusion);
3465 ret = LTTNG_ERR_FILTER_NOMEM;
3466 goto error;
3467 }
3468
3469 /* Receive var. len. data */
3470 DBG("Receiving var len filter's bytecode from client ...");
3471 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3472 if (ret <= 0) {
3473 DBG("Nothing recv() from client car len data... continuing");
3474 *sock_error = 1;
3475 free(filter_expression);
3476 free(bytecode);
3477 free(exclusion);
3478 ret = LTTNG_ERR_FILTER_INVAL;
3479 goto error;
3480 }
3481
3482 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3483 free(filter_expression);
3484 free(bytecode);
3485 free(exclusion);
3486 ret = LTTNG_ERR_FILTER_INVAL;
3487 goto error;
3488 }
3489 }
3490
3491 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3492 cmd_ctx->lsm->u.enable.channel_name,
3493 &cmd_ctx->lsm->u.enable.event,
3494 filter_expression, bytecode, exclusion,
3495 kernel_poll_pipe[1]);
3496 break;
3497 }
3498 case LTTNG_LIST_TRACEPOINTS:
3499 {
3500 struct lttng_event *events;
3501 ssize_t nb_events;
3502
3503 session_lock_list();
3504 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3505 session_unlock_list();
3506 if (nb_events < 0) {
3507 /* Return value is a negative lttng_error_code. */
3508 ret = -nb_events;
3509 goto error;
3510 }
3511
3512 /*
3513 * Setup lttng message with payload size set to the event list size in
3514 * bytes and then copy list into the llm payload.
3515 */
3516 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3517 if (ret < 0) {
3518 free(events);
3519 goto setup_error;
3520 }
3521
3522 /* Copy event list into message payload */
3523 memcpy(cmd_ctx->llm->payload, events,
3524 sizeof(struct lttng_event) * nb_events);
3525
3526 free(events);
3527
3528 ret = LTTNG_OK;
3529 break;
3530 }
3531 case LTTNG_LIST_TRACEPOINT_FIELDS:
3532 {
3533 struct lttng_event_field *fields;
3534 ssize_t nb_fields;
3535
3536 session_lock_list();
3537 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3538 &fields);
3539 session_unlock_list();
3540 if (nb_fields < 0) {
3541 /* Return value is a negative lttng_error_code. */
3542 ret = -nb_fields;
3543 goto error;
3544 }
3545
3546 /*
3547 * Setup lttng message with payload size set to the event list size in
3548 * bytes and then copy list into the llm payload.
3549 */
3550 ret = setup_lttng_msg(cmd_ctx,
3551 sizeof(struct lttng_event_field) * nb_fields);
3552 if (ret < 0) {
3553 free(fields);
3554 goto setup_error;
3555 }
3556
3557 /* Copy event list into message payload */
3558 memcpy(cmd_ctx->llm->payload, fields,
3559 sizeof(struct lttng_event_field) * nb_fields);
3560
3561 free(fields);
3562
3563 ret = LTTNG_OK;
3564 break;
3565 }
3566 case LTTNG_LIST_SYSCALLS:
3567 {
3568 struct lttng_event *events;
3569 ssize_t nb_events;
3570
3571 nb_events = cmd_list_syscalls(&events);
3572 if (nb_events < 0) {
3573 /* Return value is a negative lttng_error_code. */
3574 ret = -nb_events;
3575 goto error;
3576 }
3577
3578 /*
3579 * Setup lttng message with payload size set to the event list size in
3580 * bytes and then copy list into the llm payload.
3581 */
3582 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3583 if (ret < 0) {
3584 free(events);
3585 goto setup_error;
3586 }
3587
3588 /* Copy event list into message payload */
3589 memcpy(cmd_ctx->llm->payload, events,
3590 sizeof(struct lttng_event) * nb_events);
3591
3592 free(events);
3593
3594 ret = LTTNG_OK;
3595 break;
3596 }
3597 case LTTNG_LIST_TRACKER_PIDS:
3598 {
3599 int32_t *pids = NULL;
3600 ssize_t nr_pids;
3601
3602 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3603 cmd_ctx->lsm->domain.type, &pids);
3604 if (nr_pids < 0) {
3605 /* Return value is a negative lttng_error_code. */
3606 ret = -nr_pids;
3607 goto error;
3608 }
3609
3610 /*
3611 * Setup lttng message with payload size set to the event list size in
3612 * bytes and then copy list into the llm payload.
3613 */
3614 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3615 if (ret < 0) {
3616 free(pids);
3617 goto setup_error;
3618 }
3619
3620 /* Copy event list into message payload */
3621 memcpy(cmd_ctx->llm->payload, pids,
3622 sizeof(int) * nr_pids);
3623
3624 free(pids);
3625
3626 ret = LTTNG_OK;
3627 break;
3628 }
3629 case LTTNG_SET_CONSUMER_URI:
3630 {
3631 size_t nb_uri, len;
3632 struct lttng_uri *uris;
3633
3634 nb_uri = cmd_ctx->lsm->u.uri.size;
3635 len = nb_uri * sizeof(struct lttng_uri);
3636
3637 if (nb_uri == 0) {
3638 ret = LTTNG_ERR_INVALID;
3639 goto error;
3640 }
3641
3642 uris = zmalloc(len);
3643 if (uris == NULL) {
3644 ret = LTTNG_ERR_FATAL;
3645 goto error;
3646 }
3647
3648 /* Receive variable len data */
3649 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3650 ret = lttcomm_recv_unix_sock(sock, uris, len);
3651 if (ret <= 0) {
3652 DBG("No URIs received from client... continuing");
3653 *sock_error = 1;
3654 ret = LTTNG_ERR_SESSION_FAIL;
3655 free(uris);
3656 goto error;
3657 }
3658
3659 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3660 free(uris);
3661 if (ret != LTTNG_OK) {
3662 goto error;
3663 }
3664
3665
3666 break;
3667 }
3668 case LTTNG_START_TRACE:
3669 {
3670 ret = cmd_start_trace(cmd_ctx->session);
3671 break;
3672 }
3673 case LTTNG_STOP_TRACE:
3674 {
3675 ret = cmd_stop_trace(cmd_ctx->session);
3676 break;
3677 }
3678 case LTTNG_CREATE_SESSION:
3679 {
3680 size_t nb_uri, len;
3681 struct lttng_uri *uris = NULL;
3682
3683 nb_uri = cmd_ctx->lsm->u.uri.size;
3684 len = nb_uri * sizeof(struct lttng_uri);
3685
3686 if (nb_uri > 0) {
3687 uris = zmalloc(len);
3688 if (uris == NULL) {
3689 ret = LTTNG_ERR_FATAL;
3690 goto error;
3691 }
3692
3693 /* Receive variable len data */
3694 DBG("Waiting for %zu URIs from client ...", nb_uri);
3695 ret = lttcomm_recv_unix_sock(sock, uris, len);
3696 if (ret <= 0) {
3697 DBG("No URIs received from client... continuing");
3698 *sock_error = 1;
3699 ret = LTTNG_ERR_SESSION_FAIL;
3700 free(uris);
3701 goto error;
3702 }
3703
3704 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3705 DBG("Creating session with ONE network URI is a bad call");
3706 ret = LTTNG_ERR_SESSION_FAIL;
3707 free(uris);
3708 goto error;
3709 }
3710 }
3711
3712 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3713 &cmd_ctx->creds, 0);
3714
3715 free(uris);
3716
3717 break;
3718 }
3719 case LTTNG_DESTROY_SESSION:
3720 {
3721 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3722
3723 /* Set session to NULL so we do not unlock it after free. */
3724 cmd_ctx->session = NULL;
3725 break;
3726 }
3727 case LTTNG_LIST_DOMAINS:
3728 {
3729 ssize_t nb_dom;
3730 struct lttng_domain *domains = NULL;
3731
3732 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3733 if (nb_dom < 0) {
3734 /* Return value is a negative lttng_error_code. */
3735 ret = -nb_dom;
3736 goto error;
3737 }
3738
3739 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3740 if (ret < 0) {
3741 free(domains);
3742 goto setup_error;
3743 }
3744
3745 /* Copy event list into message payload */
3746 memcpy(cmd_ctx->llm->payload, domains,
3747 nb_dom * sizeof(struct lttng_domain));
3748
3749 free(domains);
3750
3751 ret = LTTNG_OK;
3752 break;
3753 }
3754 case LTTNG_LIST_CHANNELS:
3755 {
3756 int nb_chan;
3757 struct lttng_channel *channels = NULL;
3758
3759 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3760 cmd_ctx->session, &channels);
3761 if (nb_chan < 0) {
3762 /* Return value is a negative lttng_error_code. */
3763 ret = -nb_chan;
3764 goto error;
3765 }
3766
3767 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3768 if (ret < 0) {
3769 free(channels);
3770 goto setup_error;
3771 }
3772
3773 /* Copy event list into message payload */
3774 memcpy(cmd_ctx->llm->payload, channels,
3775 nb_chan * sizeof(struct lttng_channel));
3776
3777 free(channels);
3778
3779 ret = LTTNG_OK;
3780 break;
3781 }
3782 case LTTNG_LIST_EVENTS:
3783 {
3784 ssize_t nb_event;
3785 struct lttng_event *events = NULL;
3786
3787 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3788 cmd_ctx->lsm->u.list.channel_name, &events);
3789 if (nb_event < 0) {
3790 /* Return value is a negative lttng_error_code. */
3791 ret = -nb_event;
3792 goto error;
3793 }
3794
3795 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3796 if (ret < 0) {
3797 free(events);
3798 goto setup_error;
3799 }
3800
3801 /* Copy event list into message payload */
3802 memcpy(cmd_ctx->llm->payload, events,
3803 nb_event * sizeof(struct lttng_event));
3804
3805 free(events);
3806
3807 ret = LTTNG_OK;
3808 break;
3809 }
3810 case LTTNG_LIST_SESSIONS:
3811 {
3812 unsigned int nr_sessions;
3813
3814 session_lock_list();
3815 nr_sessions = lttng_sessions_count(
3816 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3817 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3818
3819 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3820 if (ret < 0) {
3821 session_unlock_list();
3822 goto setup_error;
3823 }
3824
3825 /* Filled the session array */
3826 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3827 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3828 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3829
3830 session_unlock_list();
3831
3832 ret = LTTNG_OK;
3833 break;
3834 }
3835 case LTTNG_CALIBRATE:
3836 {
3837 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3838 &cmd_ctx->lsm->u.calibrate);
3839 break;
3840 }
3841 case LTTNG_REGISTER_CONSUMER:
3842 {
3843 struct consumer_data *cdata;
3844
3845 switch (cmd_ctx->lsm->domain.type) {
3846 case LTTNG_DOMAIN_KERNEL:
3847 cdata = &kconsumer_data;
3848 break;
3849 default:
3850 ret = LTTNG_ERR_UND;
3851 goto error;
3852 }
3853
3854 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3855 cmd_ctx->lsm->u.reg.path, cdata);
3856 break;
3857 }
3858 case LTTNG_DATA_PENDING:
3859 {
3860 int pending_ret;
3861
3862 /* 1 byte to return whether or not data is pending */
3863 ret = setup_lttng_msg(cmd_ctx, 1);
3864 if (ret < 0) {
3865 goto setup_error;
3866 }
3867
3868 pending_ret = cmd_data_pending(cmd_ctx->session);
3869 /*
3870 * FIXME
3871 *
3872 * This function may returns 0 or 1 to indicate whether or not
3873 * there is data pending. In case of error, it should return an
3874 * LTTNG_ERR code. However, some code paths may still return
3875 * a nondescript error code, which we handle by returning an
3876 * "unknown" error.
3877 */
3878 if (pending_ret == 0 || pending_ret == 1) {
3879 ret = LTTNG_OK;
3880 } else if (pending_ret < 0) {
3881 ret = LTTNG_ERR_UNK;
3882 goto setup_error;
3883 } else {
3884 ret = pending_ret;
3885 goto setup_error;
3886 }
3887
3888 *cmd_ctx->llm->payload = (uint8_t) pending_ret;
3889 break;
3890 }
3891 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3892 {
3893 struct lttcomm_lttng_output_id reply;
3894
3895 ret = cmd_snapshot_add_output(cmd_ctx->session,
3896 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3897 if (ret != LTTNG_OK) {
3898 goto error;
3899 }
3900
3901 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3902 if (ret < 0) {
3903 goto setup_error;
3904 }
3905
3906 /* Copy output list into message payload */
3907 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3908 ret = LTTNG_OK;
3909 break;
3910 }
3911 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3912 {
3913 ret = cmd_snapshot_del_output(cmd_ctx->session,
3914 &cmd_ctx->lsm->u.snapshot_output.output);
3915 break;
3916 }
3917 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3918 {
3919 ssize_t nb_output;
3920 struct lttng_snapshot_output *outputs = NULL;
3921
3922 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3923 if (nb_output < 0) {
3924 ret = -nb_output;
3925 goto error;
3926 }
3927
3928 ret = setup_lttng_msg(cmd_ctx,
3929 nb_output * sizeof(struct lttng_snapshot_output));
3930 if (ret < 0) {
3931 free(outputs);
3932 goto setup_error;
3933 }
3934
3935 if (outputs) {
3936 /* Copy output list into message payload */
3937 memcpy(cmd_ctx->llm->payload, outputs,
3938 nb_output * sizeof(struct lttng_snapshot_output));
3939 free(outputs);
3940 }
3941
3942 ret = LTTNG_OK;
3943 break;
3944 }
3945 case LTTNG_SNAPSHOT_RECORD:
3946 {
3947 ret = cmd_snapshot_record(cmd_ctx->session,
3948 &cmd_ctx->lsm->u.snapshot_record.output,
3949 cmd_ctx->lsm->u.snapshot_record.wait);
3950 break;
3951 }
3952 case LTTNG_CREATE_SESSION_SNAPSHOT:
3953 {
3954 size_t nb_uri, len;
3955 struct lttng_uri *uris = NULL;
3956
3957 nb_uri = cmd_ctx->lsm->u.uri.size;
3958 len = nb_uri * sizeof(struct lttng_uri);
3959
3960 if (nb_uri > 0) {
3961 uris = zmalloc(len);
3962 if (uris == NULL) {
3963 ret = LTTNG_ERR_FATAL;
3964 goto error;
3965 }
3966
3967 /* Receive variable len data */
3968 DBG("Waiting for %zu URIs from client ...", nb_uri);
3969 ret = lttcomm_recv_unix_sock(sock, uris, len);
3970 if (ret <= 0) {
3971 DBG("No URIs received from client... continuing");
3972 *sock_error = 1;
3973 ret = LTTNG_ERR_SESSION_FAIL;
3974 free(uris);
3975 goto error;
3976 }
3977
3978 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3979 DBG("Creating session with ONE network URI is a bad call");
3980 ret = LTTNG_ERR_SESSION_FAIL;
3981 free(uris);
3982 goto error;
3983 }
3984 }
3985
3986 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3987 nb_uri, &cmd_ctx->creds);
3988 free(uris);
3989 break;
3990 }
3991 case LTTNG_CREATE_SESSION_LIVE:
3992 {
3993 size_t nb_uri, len;
3994 struct lttng_uri *uris = NULL;
3995
3996 nb_uri = cmd_ctx->lsm->u.uri.size;
3997 len = nb_uri * sizeof(struct lttng_uri);
3998
3999 if (nb_uri > 0) {
4000 uris = zmalloc(len);
4001 if (uris == NULL) {
4002 ret = LTTNG_ERR_FATAL;
4003 goto error;
4004 }
4005
4006 /* Receive variable len data */
4007 DBG("Waiting for %zu URIs from client ...", nb_uri);
4008 ret = lttcomm_recv_unix_sock(sock, uris, len);
4009 if (ret <= 0) {
4010 DBG("No URIs received from client... continuing");
4011 *sock_error = 1;
4012 ret = LTTNG_ERR_SESSION_FAIL;
4013 free(uris);
4014 goto error;
4015 }
4016
4017 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4018 DBG("Creating session with ONE network URI is a bad call");
4019 ret = LTTNG_ERR_SESSION_FAIL;
4020 free(uris);
4021 goto error;
4022 }
4023 }
4024
4025 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4026 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4027 free(uris);
4028 break;
4029 }
4030 case LTTNG_SAVE_SESSION:
4031 {
4032 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4033 &cmd_ctx->creds);
4034 break;
4035 }
4036 case LTTNG_SET_SESSION_SHM_PATH:
4037 {
4038 ret = cmd_set_session_shm_path(cmd_ctx->session,
4039 cmd_ctx->lsm->u.set_shm_path.shm_path);
4040 break;
4041 }
4042 default:
4043 ret = LTTNG_ERR_UND;
4044 break;
4045 }
4046
4047 error:
4048 if (cmd_ctx->llm == NULL) {
4049 DBG("Missing llm structure. Allocating one.");
4050 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
4051 goto setup_error;
4052 }
4053 }
4054 /* Set return code */
4055 cmd_ctx->llm->ret_code = ret;
4056 setup_error:
4057 if (cmd_ctx->session) {
4058 session_unlock(cmd_ctx->session);
4059 }
4060 if (need_tracing_session) {
4061 session_unlock_list();
4062 }
4063 init_setup_error:
4064 assert(!rcu_read_ongoing());
4065 return ret;
4066 }
4067
4068 /*
4069 * Thread managing health check socket.
4070 */
4071 static void *thread_manage_health(void *data)
4072 {
4073 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4074 uint32_t revents, nb_fd;
4075 struct lttng_poll_event events;
4076 struct health_comm_msg msg;
4077 struct health_comm_reply reply;
4078
4079 DBG("[thread] Manage health check started");
4080
4081 rcu_register_thread();
4082
4083 /* We might hit an error path before this is created. */
4084 lttng_poll_init(&events);
4085
4086 /* Create unix socket */
4087 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4088 if (sock < 0) {
4089 ERR("Unable to create health check Unix socket");
4090 ret = -1;
4091 goto error;
4092 }
4093
4094 if (is_root) {
4095 /* lttng health client socket path permissions */
4096 ret = chown(health_unix_sock_path, 0,
4097 utils_get_group_id(tracing_group_name));
4098 if (ret < 0) {
4099 ERR("Unable to set group on %s", health_unix_sock_path);
4100 PERROR("chown");
4101 ret = -1;
4102 goto error;
4103 }
4104
4105 ret = chmod(health_unix_sock_path,
4106 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4107 if (ret < 0) {
4108 ERR("Unable to set permissions on %s", health_unix_sock_path);
4109 PERROR("chmod");
4110 ret = -1;
4111 goto error;
4112 }
4113 }
4114
4115 /*
4116 * Set the CLOEXEC flag. Return code is useless because either way, the
4117 * show must go on.
4118 */
4119 (void) utils_set_fd_cloexec(sock);
4120
4121 ret = lttcomm_listen_unix_sock(sock);
4122 if (ret < 0) {
4123 goto error;
4124 }
4125
4126 /*
4127 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4128 * more will be added to this poll set.
4129 */
4130 ret = sessiond_set_thread_pollset(&events, 2);
4131 if (ret < 0) {
4132 goto error;
4133 }
4134
4135 /* Add the application registration socket */
4136 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4137 if (ret < 0) {
4138 goto error;
4139 }
4140
4141 sessiond_notify_ready();
4142
4143 while (1) {
4144 DBG("Health check ready");
4145
4146 /* Inifinite blocking call, waiting for transmission */
4147 restart:
4148 ret = lttng_poll_wait(&events, -1);
4149 if (ret < 0) {
4150 /*
4151 * Restart interrupted system call.
4152 */
4153 if (errno == EINTR) {
4154 goto restart;
4155 }
4156 goto error;
4157 }
4158
4159 nb_fd = ret;
4160
4161 for (i = 0; i < nb_fd; i++) {
4162 /* Fetch once the poll data */
4163 revents = LTTNG_POLL_GETEV(&events, i);
4164 pollfd = LTTNG_POLL_GETFD(&events, i);
4165
4166 if (!revents) {
4167 /* No activity for this FD (poll implementation). */
4168 continue;
4169 }
4170
4171 /* Thread quit pipe has been closed. Killing thread. */
4172 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4173 if (ret) {
4174 err = 0;
4175 goto exit;
4176 }
4177
4178 /* Event on the registration socket */
4179 if (pollfd == sock) {
4180 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4181 ERR("Health socket poll error");
4182 goto error;
4183 }
4184 }
4185 }
4186
4187 new_sock = lttcomm_accept_unix_sock(sock);
4188 if (new_sock < 0) {
4189 goto error;
4190 }
4191
4192 /*
4193 * Set the CLOEXEC flag. Return code is useless because either way, the
4194 * show must go on.
4195 */
4196 (void) utils_set_fd_cloexec(new_sock);
4197
4198 DBG("Receiving data from client for health...");
4199 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4200 if (ret <= 0) {
4201 DBG("Nothing recv() from client... continuing");
4202 ret = close(new_sock);
4203 if (ret) {
4204 PERROR("close");
4205 }
4206 new_sock = -1;
4207 continue;
4208 }
4209
4210 rcu_thread_online();
4211
4212 memset(&reply, 0, sizeof(reply));
4213 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4214 /*
4215 * health_check_state returns 0 if health is
4216 * bad.
4217 */
4218 if (!health_check_state(health_sessiond, i)) {
4219 reply.ret_code |= 1ULL << i;
4220 }
4221 }
4222
4223 DBG2("Health check return value %" PRIx64, reply.ret_code);
4224
4225 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4226 if (ret < 0) {
4227 ERR("Failed to send health data back to client");
4228 }
4229
4230 /* End of transmission */
4231 ret = close(new_sock);
4232 if (ret) {
4233 PERROR("close");
4234 }
4235 new_sock = -1;
4236 }
4237
4238 exit:
4239 error:
4240 if (err) {
4241 ERR("Health error occurred in %s", __func__);
4242 }
4243 DBG("Health check thread dying");
4244 unlink(health_unix_sock_path);
4245 if (sock >= 0) {
4246 ret = close(sock);
4247 if (ret) {
4248 PERROR("close");
4249 }
4250 }
4251
4252 lttng_poll_clean(&events);
4253
4254 rcu_unregister_thread();
4255 return NULL;
4256 }
4257
4258 /*
4259 * This thread manage all clients request using the unix client socket for
4260 * communication.
4261 */
4262 static void *thread_manage_clients(void *data)
4263 {
4264 int sock = -1, ret, i, pollfd, err = -1;
4265 int sock_error;
4266 uint32_t revents, nb_fd;
4267 struct command_ctx *cmd_ctx = NULL;
4268 struct lttng_poll_event events;
4269
4270 DBG("[thread] Manage client started");
4271
4272 rcu_register_thread();
4273
4274 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4275
4276 health_code_update();
4277
4278 ret = lttcomm_listen_unix_sock(client_sock);
4279 if (ret < 0) {
4280 goto error_listen;
4281 }
4282
4283 /*
4284 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4285 * more will be added to this poll set.
4286 */
4287 ret = sessiond_set_thread_pollset(&events, 2);
4288 if (ret < 0) {
4289 goto error_create_poll;
4290 }
4291
4292 /* Add the application registration socket */
4293 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4294 if (ret < 0) {
4295 goto error;
4296 }
4297
4298 sessiond_notify_ready();
4299 ret = sem_post(&load_info->message_thread_ready);
4300 if (ret) {
4301 PERROR("sem_post message_thread_ready");
4302 goto error;
4303 }
4304
4305 /* This testpoint is after we signal readiness to the parent. */
4306 if (testpoint(sessiond_thread_manage_clients)) {
4307 goto error;
4308 }
4309
4310 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4311 goto error;
4312 }
4313
4314 health_code_update();
4315
4316 while (1) {
4317 DBG("Accepting client command ...");
4318
4319 /* Inifinite blocking call, waiting for transmission */
4320 restart:
4321 health_poll_entry();
4322 ret = lttng_poll_wait(&events, -1);
4323 health_poll_exit();
4324 if (ret < 0) {
4325 /*
4326 * Restart interrupted system call.
4327 */
4328 if (errno == EINTR) {
4329 goto restart;
4330 }
4331 goto error;
4332 }
4333
4334 nb_fd = ret;
4335
4336 for (i = 0; i < nb_fd; i++) {
4337 /* Fetch once the poll data */
4338 revents = LTTNG_POLL_GETEV(&events, i);
4339 pollfd = LTTNG_POLL_GETFD(&events, i);
4340
4341 health_code_update();
4342
4343 if (!revents) {
4344 /* No activity for this FD (poll implementation). */
4345 continue;
4346 }
4347
4348 /* Thread quit pipe has been closed. Killing thread. */
4349 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4350 if (ret) {
4351 err = 0;
4352 goto exit;
4353 }
4354
4355 /* Event on the registration socket */
4356 if (pollfd == client_sock) {
4357 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4358 ERR("Client socket poll error");
4359 goto error;
4360 }
4361 }
4362 }
4363
4364 DBG("Wait for client response");
4365
4366 health_code_update();
4367
4368 sock = lttcomm_accept_unix_sock(client_sock);
4369 if (sock < 0) {
4370 goto error;
4371 }
4372
4373 /*
4374 * Set the CLOEXEC flag. Return code is useless because either way, the
4375 * show must go on.
4376 */
4377 (void) utils_set_fd_cloexec(sock);
4378
4379 /* Set socket option for credentials retrieval */
4380 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4381 if (ret < 0) {
4382 goto error;
4383 }
4384
4385 /* Allocate context command to process the client request */
4386 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4387 if (cmd_ctx == NULL) {
4388 PERROR("zmalloc cmd_ctx");
4389 goto error;
4390 }
4391
4392 /* Allocate data buffer for reception */
4393 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4394 if (cmd_ctx->lsm == NULL) {
4395 PERROR("zmalloc cmd_ctx->lsm");
4396 goto error;
4397 }
4398
4399 cmd_ctx->llm = NULL;
4400 cmd_ctx->session = NULL;
4401
4402 health_code_update();
4403
4404 /*
4405 * Data is received from the lttng client. The struct
4406 * lttcomm_session_msg (lsm) contains the command and data request of
4407 * the client.
4408 */
4409 DBG("Receiving data from client ...");
4410 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4411 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4412 if (ret <= 0) {
4413 DBG("Nothing recv() from client... continuing");
4414 ret = close(sock);
4415 if (ret) {
4416 PERROR("close");
4417 }
4418 sock = -1;
4419 clean_command_ctx(&cmd_ctx);
4420 continue;
4421 }
4422
4423 health_code_update();
4424
4425 // TODO: Validate cmd_ctx including sanity check for
4426 // security purpose.
4427
4428 rcu_thread_online();
4429 /*
4430 * This function dispatch the work to the kernel or userspace tracer
4431 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4432 * informations for the client. The command context struct contains
4433 * everything this function may needs.
4434 */
4435 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4436 rcu_thread_offline();
4437 if (ret < 0) {
4438 ret = close(sock);
4439 if (ret) {
4440 PERROR("close");
4441 }
4442 sock = -1;
4443 /*
4444 * TODO: Inform client somehow of the fatal error. At
4445 * this point, ret < 0 means that a zmalloc failed
4446 * (ENOMEM). Error detected but still accept
4447 * command, unless a socket error has been
4448 * detected.
4449 */
4450 clean_command_ctx(&cmd_ctx);
4451 continue;
4452 }
4453
4454 health_code_update();
4455
4456 DBG("Sending response (size: %d, retcode: %s (%d))",
4457 cmd_ctx->lttng_msg_size,
4458 lttng_strerror(-cmd_ctx->llm->ret_code),
4459 cmd_ctx->llm->ret_code);
4460 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4461 if (ret < 0) {
4462 ERR("Failed to send data back to client");
4463 }
4464
4465 /* End of transmission */
4466 ret = close(sock);
4467 if (ret) {
4468 PERROR("close");
4469 }
4470 sock = -1;
4471
4472 clean_command_ctx(&cmd_ctx);
4473
4474 health_code_update();
4475 }
4476
4477 exit:
4478 error:
4479 if (sock >= 0) {
4480 ret = close(sock);
4481 if (ret) {
4482 PERROR("close");
4483 }
4484 }
4485
4486 lttng_poll_clean(&events);
4487 clean_command_ctx(&cmd_ctx);
4488
4489 error_listen:
4490 error_create_poll:
4491 unlink(client_unix_sock_path);
4492 if (client_sock >= 0) {
4493 ret = close(client_sock);
4494 if (ret) {
4495 PERROR("close");
4496 }
4497 }
4498
4499 if (err) {
4500 health_error();
4501 ERR("Health error occurred in %s", __func__);
4502 }
4503
4504 health_unregister(health_sessiond);
4505
4506 DBG("Client thread dying");
4507
4508 rcu_unregister_thread();
4509
4510 /*
4511 * Since we are creating the consumer threads, we own them, so we need
4512 * to join them before our thread exits.
4513 */
4514 ret = join_consumer_thread(&kconsumer_data);
4515 if (ret) {
4516 errno = ret;
4517 PERROR("join_consumer");
4518 }
4519
4520 ret = join_consumer_thread(&ustconsumer32_data);
4521 if (ret) {
4522 errno = ret;
4523 PERROR("join_consumer ust32");
4524 }
4525
4526 ret = join_consumer_thread(&ustconsumer64_data);
4527 if (ret) {
4528 errno = ret;
4529 PERROR("join_consumer ust64");
4530 }
4531 return NULL;
4532 }
4533
4534
4535 /*
4536 * usage function on stderr
4537 */
4538 static void usage(void)
4539 {
4540 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4541 fprintf(stderr, " -h, --help Display this usage.\n");
4542 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4543 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4544 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4545 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4546 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4547 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4548 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4549 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4550 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4551 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4552 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4553 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4554 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4555 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4556 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4557 fprintf(stderr, " -V, --version Show version number.\n");
4558 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4559 fprintf(stderr, " -q, --quiet No output at all.\n");
4560 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4561 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4562 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4563 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4564 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4565 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4566 fprintf(stderr, " -l --load PATH Load session configuration\n");
4567 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4568 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4569 }
4570
4571 static int string_match(const char *str1, const char *str2)
4572 {
4573 return (str1 && str2) && !strcmp(str1, str2);
4574 }
4575
4576 /*
4577 * Take an option from the getopt output and set it in the right variable to be
4578 * used later.
4579 *
4580 * Return 0 on success else a negative value.
4581 */
4582 static int set_option(int opt, const char *arg, const char *optname)
4583 {
4584 int ret = 0;
4585
4586 if (arg && arg[0] == '\0') {
4587 /*
4588 * This only happens if the value is read from daemon config
4589 * file. This means the option requires an argument and the
4590 * configuration file contains a line such as:
4591 * my_option =
4592 */
4593 ret = -EINVAL;
4594 goto end;
4595 }
4596
4597 if (string_match(optname, "client-sock") || opt == 'c') {
4598 if (lttng_is_setuid_setgid()) {
4599 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4600 "-c, --client-sock");
4601 } else {
4602 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4603 }
4604 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4605 if (lttng_is_setuid_setgid()) {
4606 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4607 "-a, --apps-sock");
4608 } else {
4609 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4610 }
4611 } else if (string_match(optname, "daemonize") || opt == 'd') {
4612 opt_daemon = 1;
4613 } else if (string_match(optname, "background") || opt == 'b') {
4614 opt_background = 1;
4615 } else if (string_match(optname, "group") || opt == 'g') {
4616 if (lttng_is_setuid_setgid()) {
4617 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4618 "-g, --group");
4619 } else {
4620 /*
4621 * If the override option is set, the pointer points to a
4622 * *non* const thus freeing it even though the variable type is
4623 * set to const.
4624 */
4625 if (tracing_group_name_override) {
4626 free((void *) tracing_group_name);
4627 }
4628 tracing_group_name = strdup(arg);
4629 if (!tracing_group_name) {
4630 PERROR("strdup");
4631 ret = -ENOMEM;
4632 }
4633 tracing_group_name_override = 1;
4634 }
4635 } else if (string_match(optname, "help") || opt == 'h') {
4636 usage();
4637 exit(EXIT_SUCCESS);
4638 } else if (string_match(optname, "version") || opt == 'V') {
4639 fprintf(stdout, "%s\n", VERSION);
4640 exit(EXIT_SUCCESS);
4641 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4642 opt_sig_parent = 1;
4643 } else if (string_match(optname, "kconsumerd-err-sock")) {
4644 if (lttng_is_setuid_setgid()) {
4645 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4646 "--kconsumerd-err-sock");
4647 } else {
4648 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4649 }
4650 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4651 if (lttng_is_setuid_setgid()) {
4652 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4653 "--kconsumerd-cmd-sock");
4654 } else {
4655 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4656 }
4657 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4658 if (lttng_is_setuid_setgid()) {
4659 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4660 "--ustconsumerd64-err-sock");
4661 } else {
4662 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4663 }
4664 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4665 if (lttng_is_setuid_setgid()) {
4666 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4667 "--ustconsumerd64-cmd-sock");
4668 } else {
4669 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4670 }
4671 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4672 if (lttng_is_setuid_setgid()) {
4673 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4674 "--ustconsumerd32-err-sock");
4675 } else {
4676 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4677 }
4678 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4679 if (lttng_is_setuid_setgid()) {
4680 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4681 "--ustconsumerd32-cmd-sock");
4682 } else {
4683 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4684 }
4685 } else if (string_match(optname, "no-kernel")) {
4686 opt_no_kernel = 1;
4687 } else if (string_match(optname, "quiet") || opt == 'q') {
4688 lttng_opt_quiet = 1;
4689 } else if (string_match(optname, "verbose") || opt == 'v') {
4690 /* Verbose level can increase using multiple -v */
4691 if (arg) {
4692 /* Value obtained from config file */
4693 lttng_opt_verbose = config_parse_value(arg);
4694 } else {
4695 /* -v used on command line */
4696 lttng_opt_verbose++;
4697 }
4698 /* Clamp value to [0, 3] */
4699 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4700 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4701 } else if (string_match(optname, "verbose-consumer")) {
4702 if (arg) {
4703 opt_verbose_consumer = config_parse_value(arg);
4704 } else {
4705 opt_verbose_consumer += 1;
4706 }
4707 } else if (string_match(optname, "consumerd32-path")) {
4708 if (lttng_is_setuid_setgid()) {
4709 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4710 "--consumerd32-path");
4711 } else {
4712 if (consumerd32_bin_override) {
4713 free((void *) consumerd32_bin);
4714 }
4715 consumerd32_bin = strdup(arg);
4716 if (!consumerd32_bin) {
4717 PERROR("strdup");
4718 ret = -ENOMEM;
4719 }
4720 consumerd32_bin_override = 1;
4721 }
4722 } else if (string_match(optname, "consumerd32-libdir")) {
4723 if (lttng_is_setuid_setgid()) {
4724 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4725 "--consumerd32-libdir");
4726 } else {
4727 if (consumerd32_libdir_override) {
4728 free((void *) consumerd32_libdir);
4729 }
4730 consumerd32_libdir = strdup(arg);
4731 if (!consumerd32_libdir) {
4732 PERROR("strdup");
4733 ret = -ENOMEM;
4734 }
4735 consumerd32_libdir_override = 1;
4736 }
4737 } else if (string_match(optname, "consumerd64-path")) {
4738 if (lttng_is_setuid_setgid()) {
4739 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4740 "--consumerd64-path");
4741 } else {
4742 if (consumerd64_bin_override) {
4743 free((void *) consumerd64_bin);
4744 }
4745 consumerd64_bin = strdup(arg);
4746 if (!consumerd64_bin) {
4747 PERROR("strdup");
4748 ret = -ENOMEM;
4749 }
4750 consumerd64_bin_override = 1;
4751 }
4752 } else if (string_match(optname, "consumerd64-libdir")) {
4753 if (lttng_is_setuid_setgid()) {
4754 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4755 "--consumerd64-libdir");
4756 } else {
4757 if (consumerd64_libdir_override) {
4758 free((void *) consumerd64_libdir);
4759 }
4760 consumerd64_libdir = strdup(arg);
4761 if (!consumerd64_libdir) {
4762 PERROR("strdup");
4763 ret = -ENOMEM;
4764 }
4765 consumerd64_libdir_override = 1;
4766 }
4767 } else if (string_match(optname, "pidfile") || opt == 'p') {
4768 if (lttng_is_setuid_setgid()) {
4769 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4770 "-p, --pidfile");
4771 } else {
4772 free(opt_pidfile);
4773 opt_pidfile = strdup(arg);
4774 if (!opt_pidfile) {
4775 PERROR("strdup");
4776 ret = -ENOMEM;
4777 }
4778 }
4779 } else if (string_match(optname, "agent-tcp-port")) {
4780 if (lttng_is_setuid_setgid()) {
4781 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4782 "--agent-tcp-port");
4783 } else {
4784 unsigned long v;
4785
4786 if (!arg) {
4787 ret = -EINVAL;
4788 goto end;
4789 }
4790 errno = 0;
4791 v = strtoul(arg, NULL, 0);
4792 if (errno != 0 || !isdigit(arg[0])) {
4793 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4794 return -1;
4795 }
4796 if (v == 0 || v >= 65535) {
4797 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4798 return -1;
4799 }
4800 agent_tcp_port = (uint32_t) v;
4801 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4802 }
4803 } else if (string_match(optname, "load") || opt == 'l') {
4804 if (lttng_is_setuid_setgid()) {
4805 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4806 "-l, --load");
4807 } else {
4808 free(opt_load_session_path);
4809 opt_load_session_path = strdup(arg);
4810 if (!opt_load_session_path) {
4811 PERROR("strdup");
4812 ret = -ENOMEM;
4813 }
4814 }
4815 } else if (string_match(optname, "kmod-probes")) {
4816 if (lttng_is_setuid_setgid()) {
4817 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4818 "--kmod-probes");
4819 } else {
4820 free(kmod_probes_list);
4821 kmod_probes_list = strdup(arg);
4822 if (!kmod_probes_list) {
4823 PERROR("strdup");
4824 ret = -ENOMEM;
4825 }
4826 }
4827 } else if (string_match(optname, "extra-kmod-probes")) {
4828 if (lttng_is_setuid_setgid()) {
4829 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4830 "--extra-kmod-probes");
4831 } else {
4832 free(kmod_extra_probes_list);
4833 kmod_extra_probes_list = strdup(arg);
4834 if (!kmod_extra_probes_list) {
4835 PERROR("strdup");
4836 ret = -ENOMEM;
4837 }
4838 }
4839 } else if (string_match(optname, "config") || opt == 'f') {
4840 /* This is handled in set_options() thus silent skip. */
4841 goto end;
4842 } else {
4843 /* Unknown option or other error.
4844 * Error is printed by getopt, just return */
4845 ret = -1;
4846 }
4847
4848 end:
4849 if (ret == -EINVAL) {
4850 const char *opt_name = "unknown";
4851 int i;
4852
4853 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4854 i++) {
4855 if (opt == long_options[i].val) {
4856 opt_name = long_options[i].name;
4857 break;
4858 }
4859 }
4860
4861 WARN("Invalid argument provided for option \"%s\", using default value.",
4862 opt_name);
4863 }
4864
4865 return ret;
4866 }
4867
4868 /*
4869 * config_entry_handler_cb used to handle options read from a config file.
4870 * See config_entry_handler_cb comment in common/config/config.h for the
4871 * return value conventions.
4872 */
4873 static int config_entry_handler(const struct config_entry *entry, void *unused)
4874 {
4875 int ret = 0, i;
4876
4877 if (!entry || !entry->name || !entry->value) {
4878 ret = -EINVAL;
4879 goto end;
4880 }
4881
4882 /* Check if the option is to be ignored */
4883 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4884 if (!strcmp(entry->name, config_ignore_options[i])) {
4885 goto end;
4886 }
4887 }
4888
4889 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4890 i++) {
4891
4892 /* Ignore if not fully matched. */
4893 if (strcmp(entry->name, long_options[i].name)) {
4894 continue;
4895 }
4896
4897 /*
4898 * If the option takes no argument on the command line, we have to
4899 * check if the value is "true". We support non-zero numeric values,
4900 * true, on and yes.
4901 */
4902 if (!long_options[i].has_arg) {
4903 ret = config_parse_value(entry->value);
4904 if (ret <= 0) {
4905 if (ret) {
4906 WARN("Invalid configuration value \"%s\" for option %s",
4907 entry->value, entry->name);
4908 }
4909 /* False, skip boolean config option. */
4910 goto end;
4911 }
4912 }
4913
4914 ret = set_option(long_options[i].val, entry->value, entry->name);
4915 goto end;
4916 }
4917
4918 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4919
4920 end:
4921 return ret;
4922 }
4923
4924 /*
4925 * daemon configuration loading and argument parsing
4926 */
4927 static int set_options(int argc, char **argv)
4928 {
4929 int ret = 0, c = 0, option_index = 0;
4930 int orig_optopt = optopt, orig_optind = optind;
4931 char *optstring;
4932 const char *config_path = NULL;
4933
4934 optstring = utils_generate_optstring(long_options,
4935 sizeof(long_options) / sizeof(struct option));
4936 if (!optstring) {
4937 ret = -ENOMEM;
4938 goto end;
4939 }
4940
4941 /* Check for the --config option */
4942 while ((c = getopt_long(argc, argv, optstring, long_options,
4943 &option_index)) != -1) {
4944 if (c == '?') {
4945 ret = -EINVAL;
4946 goto end;
4947 } else if (c != 'f') {
4948 /* if not equal to --config option. */
4949 continue;
4950 }
4951
4952 if (lttng_is_setuid_setgid()) {
4953 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4954 "-f, --config");
4955 } else {
4956 config_path = utils_expand_path(optarg);
4957 if (!config_path) {
4958 ERR("Failed to resolve path: %s", optarg);
4959 }
4960 }
4961 }
4962
4963 ret = config_get_section_entries(config_path, config_section_name,
4964 config_entry_handler, NULL);
4965 if (ret) {
4966 if (ret > 0) {
4967 ERR("Invalid configuration option at line %i", ret);
4968 ret = -1;
4969 }
4970 goto end;
4971 }
4972
4973 /* Reset getopt's global state */
4974 optopt = orig_optopt;
4975 optind = orig_optind;
4976 while (1) {
4977 option_index = -1;
4978 /*
4979 * getopt_long() will not set option_index if it encounters a
4980 * short option.
4981 */
4982 c = getopt_long(argc, argv, optstring, long_options,
4983 &option_index);
4984 if (c == -1) {
4985 break;
4986 }
4987
4988 /*
4989 * Pass NULL as the long option name if popt left the index
4990 * unset.
4991 */
4992 ret = set_option(c, optarg,
4993 option_index < 0 ? NULL :
4994 long_options[option_index].name);
4995 if (ret < 0) {
4996 break;
4997 }
4998 }
4999
5000 end:
5001 free(optstring);
5002 return ret;
5003 }
5004
5005 /*
5006 * Creates the two needed socket by the daemon.
5007 * apps_sock - The communication socket for all UST apps.
5008 * client_sock - The communication of the cli tool (lttng).
5009 */
5010 static int init_daemon_socket(void)
5011 {
5012 int ret = 0;
5013 mode_t old_umask;
5014
5015 old_umask = umask(0);
5016
5017 /* Create client tool unix socket */
5018 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5019 if (client_sock < 0) {
5020 ERR("Create unix sock failed: %s", client_unix_sock_path);
5021 ret = -1;
5022 goto end;
5023 }
5024
5025 /* Set the cloexec flag */
5026 ret = utils_set_fd_cloexec(client_sock);
5027 if (ret < 0) {
5028 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5029 "Continuing but note that the consumer daemon will have a "
5030 "reference to this socket on exec()", client_sock);
5031 }
5032
5033 /* File permission MUST be 660 */
5034 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5035 if (ret < 0) {
5036 ERR("Set file permissions failed: %s", client_unix_sock_path);
5037 PERROR("chmod");
5038 goto end;
5039 }
5040
5041 /* Create the application unix socket */
5042 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5043 if (apps_sock < 0) {
5044 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5045 ret = -1;
5046 goto end;
5047 }
5048
5049 /* Set the cloexec flag */
5050 ret = utils_set_fd_cloexec(apps_sock);
5051 if (ret < 0) {
5052 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5053 "Continuing but note that the consumer daemon will have a "
5054 "reference to this socket on exec()", apps_sock);
5055 }
5056
5057 /* File permission MUST be 666 */
5058 ret = chmod(apps_unix_sock_path,
5059 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5060 if (ret < 0) {
5061 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5062 PERROR("chmod");
5063 goto end;
5064 }
5065
5066 DBG3("Session daemon client socket %d and application socket %d created",
5067 client_sock, apps_sock);
5068
5069 end:
5070 umask(old_umask);
5071 return ret;
5072 }
5073
5074 /*
5075 * Check if the global socket is available, and if a daemon is answering at the
5076 * other side. If yes, error is returned.
5077 */
5078 static int check_existing_daemon(void)
5079 {
5080 /* Is there anybody out there ? */
5081 if (lttng_session_daemon_alive()) {
5082 return -EEXIST;
5083 }
5084
5085 return 0;
5086 }
5087
5088 /*
5089 * Set the tracing group gid onto the client socket.
5090 *
5091 * Race window between mkdir and chown is OK because we are going from more
5092 * permissive (root.root) to less permissive (root.tracing).
5093 */
5094 static int set_permissions(char *rundir)
5095 {
5096 int ret;
5097 gid_t gid;
5098
5099 gid = utils_get_group_id(tracing_group_name);
5100
5101 /* Set lttng run dir */
5102 ret = chown(rundir, 0, gid);
5103 if (ret < 0) {
5104 ERR("Unable to set group on %s", rundir);
5105 PERROR("chown");
5106 }
5107
5108 /*
5109 * Ensure all applications and tracing group can search the run
5110 * dir. Allow everyone to read the directory, since it does not
5111 * buy us anything to hide its content.
5112 */
5113 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5114 if (ret < 0) {
5115 ERR("Unable to set permissions on %s", rundir);
5116 PERROR("chmod");
5117 }
5118
5119 /* lttng client socket path */
5120 ret = chown(client_unix_sock_path, 0, gid);
5121 if (ret < 0) {
5122 ERR("Unable to set group on %s", client_unix_sock_path);
5123 PERROR("chown");
5124 }
5125
5126 /* kconsumer error socket path */
5127 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5128 if (ret < 0) {
5129 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5130 PERROR("chown");
5131 }
5132
5133 /* 64-bit ustconsumer error socket path */
5134 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5135 if (ret < 0) {
5136 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5137 PERROR("chown");
5138 }
5139
5140 /* 32-bit ustconsumer compat32 error socket path */
5141 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5142 if (ret < 0) {
5143 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5144 PERROR("chown");
5145 }
5146
5147 DBG("All permissions are set");
5148
5149 return ret;
5150 }
5151
5152 /*
5153 * Create the lttng run directory needed for all global sockets and pipe.
5154 */
5155 static int create_lttng_rundir(const char *rundir)
5156 {
5157 int ret;
5158
5159 DBG3("Creating LTTng run directory: %s", rundir);
5160
5161 ret = mkdir(rundir, S_IRWXU);
5162 if (ret < 0) {
5163 if (errno != EEXIST) {
5164 ERR("Unable to create %s", rundir);
5165 goto error;
5166 } else {
5167 ret = 0;
5168 }
5169 }
5170
5171 error:
5172 return ret;
5173 }
5174
5175 /*
5176 * Setup sockets and directory needed by the kconsumerd communication with the
5177 * session daemon.
5178 */
5179 static int set_consumer_sockets(struct consumer_data *consumer_data,
5180 const char *rundir)
5181 {
5182 int ret;
5183 char path[PATH_MAX];
5184
5185 switch (consumer_data->type) {
5186 case LTTNG_CONSUMER_KERNEL:
5187 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5188 break;
5189 case LTTNG_CONSUMER64_UST:
5190 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5191 break;
5192 case LTTNG_CONSUMER32_UST:
5193 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5194 break;
5195 default:
5196 ERR("Consumer type unknown");
5197 ret = -EINVAL;
5198 goto error;
5199 }
5200
5201 DBG2("Creating consumer directory: %s", path);
5202
5203 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5204 if (ret < 0) {
5205 if (errno != EEXIST) {
5206 PERROR("mkdir");
5207 ERR("Failed to create %s", path);
5208 goto error;
5209 }
5210 ret = -1;
5211 }
5212 if (is_root) {
5213 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5214 if (ret < 0) {
5215 ERR("Unable to set group on %s", path);
5216 PERROR("chown");
5217 goto error;
5218 }
5219 }
5220
5221 /* Create the kconsumerd error unix socket */
5222 consumer_data->err_sock =
5223 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5224 if (consumer_data->err_sock < 0) {
5225 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5226 ret = -1;
5227 goto error;
5228 }
5229
5230 /*
5231 * Set the CLOEXEC flag. Return code is useless because either way, the
5232 * show must go on.
5233 */
5234 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5235 if (ret < 0) {
5236 PERROR("utils_set_fd_cloexec");
5237 /* continue anyway */
5238 }
5239
5240 /* File permission MUST be 660 */
5241 ret = chmod(consumer_data->err_unix_sock_path,
5242 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5243 if (ret < 0) {
5244 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5245 PERROR("chmod");
5246 goto error;
5247 }
5248
5249 error:
5250 return ret;
5251 }
5252
5253 /*
5254 * Signal handler for the daemon
5255 *
5256 * Simply stop all worker threads, leaving main() return gracefully after
5257 * joining all threads and calling cleanup().
5258 */
5259 static void sighandler(int sig)
5260 {
5261 switch (sig) {
5262 case SIGPIPE:
5263 DBG("SIGPIPE caught");
5264 return;
5265 case SIGINT:
5266 DBG("SIGINT caught");
5267 stop_threads();
5268 break;
5269 case SIGTERM:
5270 DBG("SIGTERM caught");
5271 stop_threads();
5272 break;
5273 case SIGUSR1:
5274 CMM_STORE_SHARED(recv_child_signal, 1);
5275 break;
5276 default:
5277 break;
5278 }
5279 }
5280
5281 /*
5282 * Setup signal handler for :
5283 * SIGINT, SIGTERM, SIGPIPE
5284 */
5285 static int set_signal_handler(void)
5286 {
5287 int ret = 0;
5288 struct sigaction sa;
5289 sigset_t sigset;
5290
5291 if ((ret = sigemptyset(&sigset)) < 0) {
5292 PERROR("sigemptyset");
5293 return ret;
5294 }
5295
5296 sa.sa_handler = sighandler;
5297 sa.sa_mask = sigset;
5298 sa.sa_flags = 0;
5299 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5300 PERROR("sigaction");
5301 return ret;
5302 }
5303
5304 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5305 PERROR("sigaction");
5306 return ret;
5307 }
5308
5309 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5310 PERROR("sigaction");
5311 return ret;
5312 }
5313
5314 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5315 PERROR("sigaction");
5316 return ret;
5317 }
5318
5319 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5320
5321 return ret;
5322 }
5323
5324 /*
5325 * Set open files limit to unlimited. This daemon can open a large number of
5326 * file descriptors in order to consumer multiple kernel traces.
5327 */
5328 static void set_ulimit(void)
5329 {
5330 int ret;
5331 struct rlimit lim;
5332
5333 /* The kernel does not allowed an infinite limit for open files */
5334 lim.rlim_cur = 65535;
5335 lim.rlim_max = 65535;
5336
5337 ret = setrlimit(RLIMIT_NOFILE, &lim);
5338 if (ret < 0) {
5339 PERROR("failed to set open files limit");
5340 }
5341 }
5342
5343 /*
5344 * Write pidfile using the rundir and opt_pidfile.
5345 */
5346 static int write_pidfile(void)
5347 {
5348 int ret;
5349 char pidfile_path[PATH_MAX];
5350
5351 assert(rundir);
5352
5353 if (opt_pidfile) {
5354 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5355 } else {
5356 /* Build pidfile path from rundir and opt_pidfile. */
5357 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5358 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5359 if (ret < 0) {
5360 PERROR("snprintf pidfile path");
5361 goto error;
5362 }
5363 }
5364
5365 /*
5366 * Create pid file in rundir.
5367 */
5368 ret = utils_create_pid_file(getpid(), pidfile_path);
5369 error:
5370 return ret;
5371 }
5372
5373 /*
5374 * Create lockfile using the rundir and return its fd.
5375 */
5376 static int create_lockfile(void)
5377 {
5378 int ret;
5379 char lockfile_path[PATH_MAX];
5380
5381 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5382 if (ret < 0) {
5383 goto error;
5384 }
5385
5386 ret = utils_create_lock_file(lockfile_path);
5387 error:
5388 return ret;
5389 }
5390
5391 /*
5392 * Write agent TCP port using the rundir.
5393 */
5394 static int write_agent_port(void)
5395 {
5396 int ret;
5397 char path[PATH_MAX];
5398
5399 assert(rundir);
5400
5401 ret = snprintf(path, sizeof(path), "%s/"
5402 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5403 if (ret < 0) {
5404 PERROR("snprintf agent port path");
5405 goto error;
5406 }
5407
5408 /*
5409 * Create TCP agent port file in rundir.
5410 */
5411 ret = utils_create_pid_file(agent_tcp_port, path);
5412
5413 error:
5414 return ret;
5415 }
5416
5417 /*
5418 * main
5419 */
5420 int main(int argc, char **argv)
5421 {
5422 int ret = 0, retval = 0;
5423 void *status;
5424 const char *home_path, *env_app_timeout;
5425
5426 init_kernel_workarounds();
5427
5428 rcu_register_thread();
5429
5430 if (set_signal_handler()) {
5431 retval = -1;
5432 goto exit_set_signal_handler;
5433 }
5434
5435 setup_consumerd_path();
5436
5437 page_size = sysconf(_SC_PAGESIZE);
5438 if (page_size < 0) {
5439 PERROR("sysconf _SC_PAGESIZE");
5440 page_size = LONG_MAX;
5441 WARN("Fallback page size to %ld", page_size);
5442 }
5443
5444 /*
5445 * Parse arguments and load the daemon configuration file.
5446 *
5447 * We have an exit_options exit path to free memory reserved by
5448 * set_options. This is needed because the rest of sessiond_cleanup()
5449 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5450 * depends on set_options.
5451 */
5452 progname = argv[0];
5453 if (set_options(argc, argv)) {
5454 retval = -1;
5455 goto exit_options;
5456 }
5457
5458 /* Daemonize */
5459 if (opt_daemon || opt_background) {
5460 int i;
5461
5462 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5463 !opt_background);
5464 if (ret < 0) {
5465 retval = -1;
5466 goto exit_options;
5467 }
5468
5469 /*
5470 * We are in the child. Make sure all other file descriptors are
5471 * closed, in case we are called with more opened file
5472 * descriptors than the standard ones.
5473 */
5474 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5475 (void) close(i);
5476 }
5477 }
5478
5479 /*
5480 * Starting from here, we can create threads. This needs to be after
5481 * lttng_daemonize due to RCU.
5482 */
5483
5484 /*
5485 * Initialize the health check subsystem. This call should set the
5486 * appropriate time values.
5487 */
5488 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5489 if (!health_sessiond) {
5490 PERROR("health_app_create error");
5491 retval = -1;
5492 goto exit_health_sessiond_cleanup;
5493 }
5494
5495 if (init_ht_cleanup_quit_pipe()) {
5496 retval = -1;
5497 goto exit_ht_cleanup_quit_pipe;
5498 }
5499
5500 /* Setup the thread ht_cleanup communication pipe. */
5501 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5502 retval = -1;
5503 goto exit_ht_cleanup_pipe;
5504 }
5505
5506 /* Set up max poll set size */
5507 if (lttng_poll_set_max_size()) {
5508 retval = -1;
5509 goto exit_set_max_size;
5510 }
5511
5512 /* Create thread to clean up RCU hash tables */
5513 ret = pthread_create(&ht_cleanup_thread, NULL,
5514 thread_ht_cleanup, (void *) NULL);
5515 if (ret) {
5516 errno = ret;
5517 PERROR("pthread_create ht_cleanup");
5518 retval = -1;
5519 goto exit_ht_cleanup;
5520 }
5521
5522 /* Create thread quit pipe */
5523 if (init_thread_quit_pipe()) {
5524 retval = -1;
5525 goto exit_init_data;
5526 }
5527
5528 /* Check if daemon is UID = 0 */
5529 is_root = !getuid();
5530
5531 if (is_root) {
5532 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5533 if (!rundir) {
5534 retval = -1;
5535 goto exit_init_data;
5536 }
5537
5538 /* Create global run dir with root access */
5539 if (create_lttng_rundir(rundir)) {
5540 retval = -1;
5541 goto exit_init_data;
5542 }
5543
5544 if (strlen(apps_unix_sock_path) == 0) {
5545 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5546 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5547 if (ret < 0) {
5548 retval = -1;
5549 goto exit_init_data;
5550 }
5551 }
5552
5553 if (strlen(client_unix_sock_path) == 0) {
5554 ret = snprintf(client_unix_sock_path, PATH_MAX,
5555 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5556 if (ret < 0) {
5557 retval = -1;
5558 goto exit_init_data;
5559 }
5560 }
5561
5562 /* Set global SHM for ust */
5563 if (strlen(wait_shm_path) == 0) {
5564 ret = snprintf(wait_shm_path, PATH_MAX,
5565 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5566 if (ret < 0) {
5567 retval = -1;
5568 goto exit_init_data;
5569 }
5570 }
5571
5572 if (strlen(health_unix_sock_path) == 0) {
5573 ret = snprintf(health_unix_sock_path,
5574 sizeof(health_unix_sock_path),
5575 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5576 if (ret < 0) {
5577 retval = -1;
5578 goto exit_init_data;
5579 }
5580 }
5581
5582 /* Setup kernel consumerd path */
5583 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5584 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5585 if (ret < 0) {
5586 retval = -1;
5587 goto exit_init_data;
5588 }
5589 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5590 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5591 if (ret < 0) {
5592 retval = -1;
5593 goto exit_init_data;
5594 }
5595
5596 DBG2("Kernel consumer err path: %s",
5597 kconsumer_data.err_unix_sock_path);
5598 DBG2("Kernel consumer cmd path: %s",
5599 kconsumer_data.cmd_unix_sock_path);
5600 } else {
5601 home_path = utils_get_home_dir();
5602 if (home_path == NULL) {
5603 /* TODO: Add --socket PATH option */
5604 ERR("Can't get HOME directory for sockets creation.");
5605 retval = -1;
5606 goto exit_init_data;
5607 }
5608
5609 /*
5610 * Create rundir from home path. This will create something like
5611 * $HOME/.lttng
5612 */
5613 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5614 if (ret < 0) {
5615 retval = -1;
5616 goto exit_init_data;
5617 }
5618
5619 if (create_lttng_rundir(rundir)) {
5620 retval = -1;
5621 goto exit_init_data;
5622 }
5623
5624 if (strlen(apps_unix_sock_path) == 0) {
5625 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5626 DEFAULT_HOME_APPS_UNIX_SOCK,
5627 home_path);
5628 if (ret < 0) {
5629 retval = -1;
5630 goto exit_init_data;
5631 }
5632 }
5633
5634 /* Set the cli tool unix socket path */
5635 if (strlen(client_unix_sock_path) == 0) {
5636 ret = snprintf(client_unix_sock_path, PATH_MAX,
5637 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5638 home_path);
5639 if (ret < 0) {
5640 retval = -1;
5641 goto exit_init_data;
5642 }
5643 }
5644
5645 /* Set global SHM for ust */
5646 if (strlen(wait_shm_path) == 0) {
5647 ret = snprintf(wait_shm_path, PATH_MAX,
5648 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5649 getuid());
5650 if (ret < 0) {
5651 retval = -1;
5652 goto exit_init_data;
5653 }
5654 }
5655
5656 /* Set health check Unix path */
5657 if (strlen(health_unix_sock_path) == 0) {
5658 ret = snprintf(health_unix_sock_path,
5659 sizeof(health_unix_sock_path),
5660 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5661 home_path);
5662 if (ret < 0) {
5663 retval = -1;
5664 goto exit_init_data;
5665 }
5666 }
5667 }
5668
5669 lockfile_fd = create_lockfile();
5670 if (lockfile_fd < 0) {
5671 retval = -1;
5672 goto exit_init_data;
5673 }
5674
5675 /* Set consumer initial state */
5676 kernel_consumerd_state = CONSUMER_STOPPED;
5677 ust_consumerd_state = CONSUMER_STOPPED;
5678
5679 DBG("Client socket path %s", client_unix_sock_path);
5680 DBG("Application socket path %s", apps_unix_sock_path);
5681 DBG("Application wait path %s", wait_shm_path);
5682 DBG("LTTng run directory path: %s", rundir);
5683
5684 /* 32 bits consumerd path setup */
5685 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5686 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5687 if (ret < 0) {
5688 PERROR("snprintf 32-bit consumer error socket path");
5689 retval = -1;
5690 goto exit_init_data;
5691 }
5692 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5693 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5694 if (ret < 0) {
5695 PERROR("snprintf 32-bit consumer command socket path");
5696 retval = -1;
5697 goto exit_init_data;
5698 }
5699
5700 DBG2("UST consumer 32 bits err path: %s",
5701 ustconsumer32_data.err_unix_sock_path);
5702 DBG2("UST consumer 32 bits cmd path: %s",
5703 ustconsumer32_data.cmd_unix_sock_path);
5704
5705 /* 64 bits consumerd path setup */
5706 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5707 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5708 if (ret < 0) {
5709 PERROR("snprintf 64-bit consumer error socket path");
5710 retval = -1;
5711 goto exit_init_data;
5712 }
5713 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5714 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5715 if (ret < 0) {
5716 PERROR("snprintf 64-bit consumer command socket path");
5717 retval = -1;
5718 goto exit_init_data;
5719 }
5720
5721 DBG2("UST consumer 64 bits err path: %s",
5722 ustconsumer64_data.err_unix_sock_path);
5723 DBG2("UST consumer 64 bits cmd path: %s",
5724 ustconsumer64_data.cmd_unix_sock_path);
5725
5726 /*
5727 * See if daemon already exist.
5728 */
5729 if (check_existing_daemon()) {
5730 ERR("Already running daemon.\n");
5731 /*
5732 * We do not goto exit because we must not cleanup()
5733 * because a daemon is already running.
5734 */
5735 retval = -1;
5736 goto exit_init_data;
5737 }
5738
5739 /*
5740 * Init UST app hash table. Alloc hash table before this point since
5741 * cleanup() can get called after that point.
5742 */
5743 if (ust_app_ht_alloc()) {
5744 ERR("Failed to allocate UST app hash table");
5745 retval = -1;
5746 goto exit_init_data;
5747 }
5748
5749 /*
5750 * Initialize agent app hash table. We allocate the hash table here
5751 * since cleanup() can get called after this point.
5752 */
5753 if (agent_app_ht_alloc()) {
5754 ERR("Failed to allocate Agent app hash table");
5755 retval = -1;
5756 goto exit_init_data;
5757 }
5758
5759 /*
5760 * These actions must be executed as root. We do that *after* setting up
5761 * the sockets path because we MUST make the check for another daemon using
5762 * those paths *before* trying to set the kernel consumer sockets and init
5763 * kernel tracer.
5764 */
5765 if (is_root) {
5766 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5767 retval = -1;
5768 goto exit_init_data;
5769 }
5770
5771 /* Setup kernel tracer */
5772 if (!opt_no_kernel) {
5773 init_kernel_tracer();
5774 if (kernel_tracer_fd >= 0) {
5775 ret = syscall_init_table();
5776 if (ret < 0) {
5777 ERR("Unable to populate syscall table. "
5778 "Syscall tracing won't work "
5779 "for this session daemon.");
5780 }
5781 }
5782 }
5783
5784 /* Set ulimit for open files */
5785 set_ulimit();
5786 }
5787 /* init lttng_fd tracking must be done after set_ulimit. */
5788 lttng_fd_init();
5789
5790 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5791 retval = -1;
5792 goto exit_init_data;
5793 }
5794
5795 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5796 retval = -1;
5797 goto exit_init_data;
5798 }
5799
5800 /* Setup the needed unix socket */
5801 if (init_daemon_socket()) {
5802 retval = -1;
5803 goto exit_init_data;
5804 }
5805
5806 /* Set credentials to socket */
5807 if (is_root && set_permissions(rundir)) {
5808 retval = -1;
5809 goto exit_init_data;
5810 }
5811
5812 /* Get parent pid if -S, --sig-parent is specified. */
5813 if (opt_sig_parent) {
5814 ppid = getppid();
5815 }
5816
5817 /* Setup the kernel pipe for waking up the kernel thread */
5818 if (is_root && !opt_no_kernel) {
5819 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5820 retval = -1;
5821 goto exit_init_data;
5822 }
5823 }
5824
5825 /* Setup the thread apps communication pipe. */
5826 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5827 retval = -1;
5828 goto exit_init_data;
5829 }
5830
5831 /* Setup the thread apps notify communication pipe. */
5832 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5833 retval = -1;
5834 goto exit_init_data;
5835 }
5836
5837 /* Initialize global buffer per UID and PID registry. */
5838 buffer_reg_init_uid_registry();
5839 buffer_reg_init_pid_registry();
5840
5841 /* Init UST command queue. */
5842 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5843
5844 /*
5845 * Get session list pointer. This pointer MUST NOT be free'd. This list
5846 * is statically declared in session.c
5847 */
5848 session_list_ptr = session_get_list();
5849
5850 cmd_init();
5851
5852 /* Check for the application socket timeout env variable. */
5853 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5854 if (env_app_timeout) {
5855 app_socket_timeout = atoi(env_app_timeout);
5856 } else {
5857 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5858 }
5859
5860 ret = write_pidfile();
5861 if (ret) {
5862 ERR("Error in write_pidfile");
5863 retval = -1;
5864 goto exit_init_data;
5865 }
5866 ret = write_agent_port();
5867 if (ret) {
5868 ERR("Error in write_agent_port");
5869 retval = -1;
5870 goto exit_init_data;
5871 }
5872
5873 /* Initialize communication library */
5874 lttcomm_init();
5875 /* Initialize TCP timeout values */
5876 lttcomm_inet_init();
5877
5878 if (load_session_init_data(&load_info) < 0) {
5879 retval = -1;
5880 goto exit_init_data;
5881 }
5882 load_info->path = opt_load_session_path;
5883
5884 /* Create health-check thread */
5885 ret = pthread_create(&health_thread, NULL,
5886 thread_manage_health, (void *) NULL);
5887 if (ret) {
5888 errno = ret;
5889 PERROR("pthread_create health");
5890 retval = -1;
5891 goto exit_health;
5892 }
5893
5894 /* Create thread to manage the client socket */
5895 ret = pthread_create(&client_thread, NULL,
5896 thread_manage_clients, (void *) NULL);
5897 if (ret) {
5898 errno = ret;
5899 PERROR("pthread_create clients");
5900 retval = -1;
5901 goto exit_client;
5902 }
5903
5904 /* Create thread to dispatch registration */
5905 ret = pthread_create(&dispatch_thread, NULL,
5906 thread_dispatch_ust_registration, (void *) NULL);
5907 if (ret) {
5908 errno = ret;
5909 PERROR("pthread_create dispatch");
5910 retval = -1;
5911 goto exit_dispatch;
5912 }
5913
5914 /* Create thread to manage application registration. */
5915 ret = pthread_create(&reg_apps_thread, NULL,
5916 thread_registration_apps, (void *) NULL);
5917 if (ret) {
5918 errno = ret;
5919 PERROR("pthread_create registration");
5920 retval = -1;
5921 goto exit_reg_apps;
5922 }
5923
5924 /* Create thread to manage application socket */
5925 ret = pthread_create(&apps_thread, NULL,
5926 thread_manage_apps, (void *) NULL);
5927 if (ret) {
5928 errno = ret;
5929 PERROR("pthread_create apps");
5930 retval = -1;
5931 goto exit_apps;
5932 }
5933
5934 /* Create thread to manage application notify socket */
5935 ret = pthread_create(&apps_notify_thread, NULL,
5936 ust_thread_manage_notify, (void *) NULL);
5937 if (ret) {
5938 errno = ret;
5939 PERROR("pthread_create notify");
5940 retval = -1;
5941 goto exit_apps_notify;
5942 }
5943
5944 /* Create agent registration thread. */
5945 ret = pthread_create(&agent_reg_thread, NULL,
5946 agent_thread_manage_registration, (void *) NULL);
5947 if (ret) {
5948 errno = ret;
5949 PERROR("pthread_create agent");
5950 retval = -1;
5951 goto exit_agent_reg;
5952 }
5953
5954 /* Don't start this thread if kernel tracing is not requested nor root */
5955 if (is_root && !opt_no_kernel) {
5956 /* Create kernel thread to manage kernel event */
5957 ret = pthread_create(&kernel_thread, NULL,
5958 thread_manage_kernel, (void *) NULL);
5959 if (ret) {
5960 errno = ret;
5961 PERROR("pthread_create kernel");
5962 retval = -1;
5963 goto exit_kernel;
5964 }
5965 }
5966
5967 /* Create session loading thread. */
5968 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5969 load_info);
5970 if (ret) {
5971 errno = ret;
5972 PERROR("pthread_create load_session_thread");
5973 retval = -1;
5974 goto exit_load_session;
5975 }
5976
5977 /*
5978 * This is where we start awaiting program completion (e.g. through
5979 * signal that asks threads to teardown).
5980 */
5981
5982 ret = pthread_join(load_session_thread, &status);
5983 if (ret) {
5984 errno = ret;
5985 PERROR("pthread_join load_session_thread");
5986 retval = -1;
5987 }
5988 exit_load_session:
5989
5990 if (is_root && !opt_no_kernel) {
5991 ret = pthread_join(kernel_thread, &status);
5992 if (ret) {
5993 errno = ret;
5994 PERROR("pthread_join");
5995 retval = -1;
5996 }
5997 }
5998 exit_kernel:
5999
6000 ret = pthread_join(agent_reg_thread, &status);
6001 if (ret) {
6002 errno = ret;
6003 PERROR("pthread_join agent");
6004 retval = -1;
6005 }
6006 exit_agent_reg:
6007
6008 ret = pthread_join(apps_notify_thread, &status);
6009 if (ret) {
6010 errno = ret;
6011 PERROR("pthread_join apps notify");
6012 retval = -1;
6013 }
6014 exit_apps_notify:
6015
6016 ret = pthread_join(apps_thread, &status);
6017 if (ret) {
6018 errno = ret;
6019 PERROR("pthread_join apps");
6020 retval = -1;
6021 }
6022 exit_apps:
6023
6024 ret = pthread_join(reg_apps_thread, &status);
6025 if (ret) {
6026 errno = ret;
6027 PERROR("pthread_join");
6028 retval = -1;
6029 }
6030 exit_reg_apps:
6031
6032 /*
6033 * Join dispatch thread after joining reg_apps_thread to ensure
6034 * we don't leak applications in the queue.
6035 */
6036 ret = pthread_join(dispatch_thread, &status);
6037 if (ret) {
6038 errno = ret;
6039 PERROR("pthread_join");
6040 retval = -1;
6041 }
6042 exit_dispatch:
6043
6044 ret = pthread_join(client_thread, &status);
6045 if (ret) {
6046 errno = ret;
6047 PERROR("pthread_join");
6048 retval = -1;
6049 }
6050 exit_client:
6051
6052 ret = pthread_join(health_thread, &status);
6053 if (ret) {
6054 errno = ret;
6055 PERROR("pthread_join health thread");
6056 retval = -1;
6057 }
6058 exit_health:
6059
6060 exit_init_data:
6061 /*
6062 * sessiond_cleanup() is called when no other thread is running, except
6063 * the ht_cleanup thread, which is needed to destroy the hash tables.
6064 */
6065 rcu_thread_online();
6066 sessiond_cleanup();
6067 rcu_thread_offline();
6068 rcu_unregister_thread();
6069
6070 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6071 if (ret < 0) {
6072 ERR("write error on ht_cleanup quit pipe");
6073 retval = -1;
6074 }
6075
6076 ret = pthread_join(ht_cleanup_thread, &status);
6077 if (ret) {
6078 errno = ret;
6079 PERROR("pthread_join ht cleanup thread");
6080 retval = -1;
6081 }
6082 exit_ht_cleanup:
6083 exit_set_max_size:
6084
6085 utils_close_pipe(ht_cleanup_pipe);
6086 exit_ht_cleanup_pipe:
6087
6088 /*
6089 * Close the ht_cleanup quit pipe.
6090 */
6091 utils_close_pipe(ht_cleanup_quit_pipe);
6092 exit_ht_cleanup_quit_pipe:
6093
6094 health_app_destroy(health_sessiond);
6095 exit_health_sessiond_cleanup:
6096
6097 exit_options:
6098 sessiond_cleanup_options();
6099
6100 exit_set_signal_handler:
6101 /* Ensure all prior call_rcu are done. */
6102 rcu_barrier();
6103
6104 if (!retval) {
6105 exit(EXIT_SUCCESS);
6106 } else {
6107 exit(EXIT_FAILURE);
6108 }
6109 }
This page took 0.236693 seconds and 4 git commands to generate.