Fix: LPOLLHUP and LPOLLERR when there is still data in pipe/socket
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _GNU_SOURCE
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <paths.h>
26 #include <pthread.h>
27 #include <signal.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <string.h>
31 #include <inttypes.h>
32 #include <sys/mman.h>
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
36 #include <sys/stat.h>
37 #include <sys/types.h>
38 #include <sys/wait.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <config.h>
42
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
53
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
56 #include "channel.h"
57 #include "cmd.h"
58 #include "consumer.h"
59 #include "context.h"
60 #include "event.h"
61 #include "kernel.h"
62 #include "kernel-consumer.h"
63 #include "modprobe.h"
64 #include "shm.h"
65 #include "ust-ctl.h"
66 #include "ust-consumer.h"
67 #include "utils.h"
68 #include "fd-limit.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
73 #include "save.h"
74 #include "load-session-thread.h"
75 #include "syscall.h"
76
77 #define CONSUMERD_FILE "lttng-consumerd"
78
79 const char *progname;
80 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
81 static int tracing_group_name_override;
82 static char *opt_pidfile;
83 static int opt_sig_parent;
84 static int opt_verbose_consumer;
85 static int opt_daemon, opt_background;
86 static int opt_no_kernel;
87 static char *opt_load_session_path;
88 static pid_t ppid; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
90 static char *rundir;
91 static int lockfile_fd = -1;
92
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal;
95
96 /*
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
99 */
100 static struct consumer_data kconsumer_data = {
101 .type = LTTNG_CONSUMER_KERNEL,
102 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
103 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
104 .err_sock = -1,
105 .cmd_sock = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
118 .lock = PTHREAD_MUTEX_INITIALIZER,
119 .cond = PTHREAD_COND_INITIALIZER,
120 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
121 };
122 static struct consumer_data ustconsumer32_data = {
123 .type = LTTNG_CONSUMER32_UST,
124 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
125 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
126 .err_sock = -1,
127 .cmd_sock = -1,
128 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
129 .lock = PTHREAD_MUTEX_INITIALIZER,
130 .cond = PTHREAD_COND_INITIALIZER,
131 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
132 };
133
134 /* Command line options */
135 static const struct option long_options[] = {
136 { "client-sock", required_argument, 0, 'c' },
137 { "apps-sock", required_argument, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
144 { "consumerd32-path", required_argument, 0, '\0' },
145 { "consumerd32-libdir", required_argument, 0, '\0' },
146 { "consumerd64-path", required_argument, 0, '\0' },
147 { "consumerd64-libdir", required_argument, 0, '\0' },
148 { "daemonize", no_argument, 0, 'd' },
149 { "background", no_argument, 0, 'b' },
150 { "sig-parent", no_argument, 0, 'S' },
151 { "help", no_argument, 0, 'h' },
152 { "group", required_argument, 0, 'g' },
153 { "version", no_argument, 0, 'V' },
154 { "quiet", no_argument, 0, 'q' },
155 { "verbose", no_argument, 0, 'v' },
156 { "verbose-consumer", no_argument, 0, '\0' },
157 { "no-kernel", no_argument, 0, '\0' },
158 { "pidfile", required_argument, 0, 'p' },
159 { "agent-tcp-port", required_argument, 0, '\0' },
160 { "config", required_argument, 0, 'f' },
161 { "load", required_argument, 0, 'l' },
162 { "kmod-probes", required_argument, 0, '\0' },
163 { "extra-kmod-probes", required_argument, 0, '\0' },
164 { NULL, 0, 0, 0 }
165 };
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path[PATH_MAX];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path[PATH_MAX];
177 /* global wait shm path for UST */
178 static char wait_shm_path[PATH_MAX];
179 /* Global health check unix path */
180 static char health_unix_sock_path[PATH_MAX];
181
182 /* Sockets and FDs */
183 static int client_sock = -1;
184 static int apps_sock = -1;
185 int kernel_tracer_fd = -1;
186 static int kernel_poll_pipe[2] = { -1, -1 };
187
188 /*
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
191 */
192 static int thread_quit_pipe[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
194
195 /*
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
198 */
199 static int apps_cmd_pipe[2] = { -1, -1 };
200
201 int apps_cmd_notify_pipe[2] = { -1, -1 };
202
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread;
205 static pthread_t apps_notify_thread;
206 static pthread_t reg_apps_thread;
207 static pthread_t client_thread;
208 static pthread_t kernel_thread;
209 static pthread_t dispatch_thread;
210 static pthread_t health_thread;
211 static pthread_t ht_cleanup_thread;
212 static pthread_t agent_reg_thread;
213 static pthread_t load_session_thread;
214
215 /*
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 *
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
223 */
224 static struct ust_cmd_queue ust_cmd_queue;
225
226 /*
227 * Pointer initialized before thread creation.
228 *
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
232 *
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
235 */
236 static struct ltt_session_list *session_list_ptr;
237
238 int ust_consumerd64_fd = -1;
239 int ust_consumerd32_fd = -1;
240
241 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
242 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
243 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
244 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
245 static int consumerd32_bin_override;
246 static int consumerd64_bin_override;
247 static int consumerd32_libdir_override;
248 static int consumerd64_libdir_override;
249
250 static const char *module_proc_lttng = "/proc/lttng";
251
252 /*
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
255 */
256 enum consumerd_state {
257 CONSUMER_STARTED = 1,
258 CONSUMER_STOPPED = 2,
259 CONSUMER_ERROR = 3,
260 };
261
262 /*
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
267 *
268 * The following example shows a possible race condition of this scheme:
269 *
270 * consumer thread error happens
271 * client cmd arrives
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
275 * ...
276 *
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
282 */
283 static enum consumerd_state ust_consumerd_state;
284 static enum consumerd_state kernel_consumerd_state;
285
286 /*
287 * Socket timeout for receiving and sending in seconds.
288 */
289 static int app_socket_timeout;
290
291 /* Set in main() with the current page size. */
292 long page_size;
293
294 /* Application health monitoring */
295 struct health_app *health_sessiond;
296
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
299
300 /* Am I root or not. */
301 int is_root; /* Set to 1 if the daemon is running as root */
302
303 const char * const config_section_name = "sessiond";
304
305 /* Load session thread information to operate. */
306 struct load_session_thread_data *load_info;
307
308 /*
309 * Whether sessiond is ready for commands/health check requests.
310 * NR_LTTNG_SESSIOND_READY must match the number of calls to
311 * sessiond_notify_ready().
312 */
313 #define NR_LTTNG_SESSIOND_READY 3
314 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
315
316 /* Notify parents that we are ready for cmd and health check */
317 LTTNG_HIDDEN
318 void sessiond_notify_ready(void)
319 {
320 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
321 /*
322 * Notify parent pid that we are ready to accept command
323 * for client side. This ppid is the one from the
324 * external process that spawned us.
325 */
326 if (opt_sig_parent) {
327 kill(ppid, SIGUSR1);
328 }
329
330 /*
331 * Notify the parent of the fork() process that we are
332 * ready.
333 */
334 if (opt_daemon || opt_background) {
335 kill(child_ppid, SIGUSR1);
336 }
337 }
338 }
339
340 static
341 void setup_consumerd_path(void)
342 {
343 const char *bin, *libdir;
344
345 /*
346 * Allow INSTALL_BIN_PATH to be used as a target path for the
347 * native architecture size consumer if CONFIG_CONSUMER*_PATH
348 * has not been defined.
349 */
350 #if (CAA_BITS_PER_LONG == 32)
351 if (!consumerd32_bin[0]) {
352 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
353 }
354 if (!consumerd32_libdir[0]) {
355 consumerd32_libdir = INSTALL_LIB_PATH;
356 }
357 #elif (CAA_BITS_PER_LONG == 64)
358 if (!consumerd64_bin[0]) {
359 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
360 }
361 if (!consumerd64_libdir[0]) {
362 consumerd64_libdir = INSTALL_LIB_PATH;
363 }
364 #else
365 #error "Unknown bitness"
366 #endif
367
368 /*
369 * runtime env. var. overrides the build default.
370 */
371 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
372 if (bin) {
373 consumerd32_bin = bin;
374 }
375 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
376 if (bin) {
377 consumerd64_bin = bin;
378 }
379 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
380 if (libdir) {
381 consumerd32_libdir = libdir;
382 }
383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
384 if (libdir) {
385 consumerd64_libdir = libdir;
386 }
387 }
388
389 static
390 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
391 int *a_pipe)
392 {
393 int ret;
394
395 assert(events);
396
397 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
398 if (ret < 0) {
399 goto error;
400 }
401
402 /* Add quit pipe */
403 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
404 if (ret < 0) {
405 goto error;
406 }
407
408 return 0;
409
410 error:
411 return ret;
412 }
413
414 /*
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
416 */
417 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
418 {
419 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
420 }
421
422 /*
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 */
425 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
426 size_t size)
427 {
428 return __sessiond_set_thread_pollset(events, size,
429 ht_cleanup_quit_pipe);
430 }
431
432 static
433 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
434 {
435 if (fd == a_pipe && (events & LPOLLIN)) {
436 return 1;
437 }
438 return 0;
439 }
440
441 /*
442 * Check if the thread quit pipe was triggered.
443 *
444 * Return 1 if it was triggered else 0;
445 */
446 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
447 {
448 return __sessiond_check_thread_quit_pipe(fd, events,
449 thread_quit_pipe[0]);
450 }
451
452 /*
453 * Check if the ht_cleanup thread quit pipe was triggered.
454 *
455 * Return 1 if it was triggered else 0;
456 */
457 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
458 {
459 return __sessiond_check_thread_quit_pipe(fd, events,
460 ht_cleanup_quit_pipe[0]);
461 }
462
463 /*
464 * Init thread quit pipe.
465 *
466 * Return -1 on error or 0 if all pipes are created.
467 */
468 static int __init_thread_quit_pipe(int *a_pipe)
469 {
470 int ret, i;
471
472 ret = pipe(a_pipe);
473 if (ret < 0) {
474 PERROR("thread quit pipe");
475 goto error;
476 }
477
478 for (i = 0; i < 2; i++) {
479 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
480 if (ret < 0) {
481 PERROR("fcntl");
482 goto error;
483 }
484 }
485
486 error:
487 return ret;
488 }
489
490 static int init_thread_quit_pipe(void)
491 {
492 return __init_thread_quit_pipe(thread_quit_pipe);
493 }
494
495 static int init_ht_cleanup_quit_pipe(void)
496 {
497 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
498 }
499
500 /*
501 * Stop all threads by closing the thread quit pipe.
502 */
503 static void stop_threads(void)
504 {
505 int ret;
506
507 /* Stopping all threads */
508 DBG("Terminating all threads");
509 ret = notify_thread_pipe(thread_quit_pipe[1]);
510 if (ret < 0) {
511 ERR("write error on thread quit pipe");
512 }
513
514 /* Dispatch thread */
515 CMM_STORE_SHARED(dispatch_thread_exit, 1);
516 futex_nto1_wake(&ust_cmd_queue.futex);
517 }
518
519 /*
520 * Close every consumer sockets.
521 */
522 static void close_consumer_sockets(void)
523 {
524 int ret;
525
526 if (kconsumer_data.err_sock >= 0) {
527 ret = close(kconsumer_data.err_sock);
528 if (ret < 0) {
529 PERROR("kernel consumer err_sock close");
530 }
531 }
532 if (ustconsumer32_data.err_sock >= 0) {
533 ret = close(ustconsumer32_data.err_sock);
534 if (ret < 0) {
535 PERROR("UST consumerd32 err_sock close");
536 }
537 }
538 if (ustconsumer64_data.err_sock >= 0) {
539 ret = close(ustconsumer64_data.err_sock);
540 if (ret < 0) {
541 PERROR("UST consumerd64 err_sock close");
542 }
543 }
544 if (kconsumer_data.cmd_sock >= 0) {
545 ret = close(kconsumer_data.cmd_sock);
546 if (ret < 0) {
547 PERROR("kernel consumer cmd_sock close");
548 }
549 }
550 if (ustconsumer32_data.cmd_sock >= 0) {
551 ret = close(ustconsumer32_data.cmd_sock);
552 if (ret < 0) {
553 PERROR("UST consumerd32 cmd_sock close");
554 }
555 }
556 if (ustconsumer64_data.cmd_sock >= 0) {
557 ret = close(ustconsumer64_data.cmd_sock);
558 if (ret < 0) {
559 PERROR("UST consumerd64 cmd_sock close");
560 }
561 }
562 }
563
564 /*
565 * Generate the full lock file path using the rundir.
566 *
567 * Return the snprintf() return value thus a negative value is an error.
568 */
569 static int generate_lock_file_path(char *path, size_t len)
570 {
571 int ret;
572
573 assert(path);
574 assert(rundir);
575
576 /* Build lockfile path from rundir. */
577 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
578 if (ret < 0) {
579 PERROR("snprintf lockfile path");
580 }
581
582 return ret;
583 }
584
585 /*
586 * Wait on consumer process termination.
587 *
588 * Need to be called with the consumer data lock held or from a context
589 * ensuring no concurrent access to data (e.g: cleanup).
590 */
591 static void wait_consumer(struct consumer_data *consumer_data)
592 {
593 pid_t ret;
594 int status;
595
596 if (consumer_data->pid <= 0) {
597 return;
598 }
599
600 DBG("Waiting for complete teardown of consumerd (PID: %d)",
601 consumer_data->pid);
602 ret = waitpid(consumer_data->pid, &status, 0);
603 if (ret == -1) {
604 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
605 }
606 if (!WIFEXITED(status)) {
607 ERR("consumerd termination with error: %d",
608 WEXITSTATUS(ret));
609 }
610 consumer_data->pid = 0;
611 }
612
613 /*
614 * Cleanup the session daemon's data structures.
615 */
616 static void sessiond_cleanup(void)
617 {
618 int ret;
619 struct ltt_session *sess, *stmp;
620 char path[PATH_MAX];
621
622 DBG("Cleanup sessiond");
623
624 /*
625 * Close the thread quit pipe. It has already done its job,
626 * since we are now called.
627 */
628 utils_close_pipe(thread_quit_pipe);
629
630 /*
631 * If opt_pidfile is undefined, the default file will be wiped when
632 * removing the rundir.
633 */
634 if (opt_pidfile) {
635 ret = remove(opt_pidfile);
636 if (ret < 0) {
637 PERROR("remove pidfile %s", opt_pidfile);
638 }
639 }
640
641 DBG("Removing sessiond and consumerd content of directory %s", rundir);
642
643 /* sessiond */
644 snprintf(path, PATH_MAX,
645 "%s/%s",
646 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
647 DBG("Removing %s", path);
648 (void) unlink(path);
649
650 snprintf(path, PATH_MAX, "%s/%s", rundir,
651 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
652 DBG("Removing %s", path);
653 (void) unlink(path);
654
655 /* kconsumerd */
656 snprintf(path, PATH_MAX,
657 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
658 rundir);
659 DBG("Removing %s", path);
660 (void) unlink(path);
661
662 snprintf(path, PATH_MAX,
663 DEFAULT_KCONSUMERD_PATH,
664 rundir);
665 DBG("Removing directory %s", path);
666 (void) rmdir(path);
667
668 /* ust consumerd 32 */
669 snprintf(path, PATH_MAX,
670 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
671 rundir);
672 DBG("Removing %s", path);
673 (void) unlink(path);
674
675 snprintf(path, PATH_MAX,
676 DEFAULT_USTCONSUMERD32_PATH,
677 rundir);
678 DBG("Removing directory %s", path);
679 (void) rmdir(path);
680
681 /* ust consumerd 64 */
682 snprintf(path, PATH_MAX,
683 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
684 rundir);
685 DBG("Removing %s", path);
686 (void) unlink(path);
687
688 snprintf(path, PATH_MAX,
689 DEFAULT_USTCONSUMERD64_PATH,
690 rundir);
691 DBG("Removing directory %s", path);
692 (void) rmdir(path);
693
694 DBG("Cleaning up all sessions");
695
696 /* Destroy session list mutex */
697 if (session_list_ptr != NULL) {
698 pthread_mutex_destroy(&session_list_ptr->lock);
699
700 /* Cleanup ALL session */
701 cds_list_for_each_entry_safe(sess, stmp,
702 &session_list_ptr->head, list) {
703 cmd_destroy_session(sess, kernel_poll_pipe[1]);
704 }
705 }
706
707 wait_consumer(&kconsumer_data);
708 wait_consumer(&ustconsumer64_data);
709 wait_consumer(&ustconsumer32_data);
710
711 DBG("Cleaning up all agent apps");
712 agent_app_ht_clean();
713
714 DBG("Closing all UST sockets");
715 ust_app_clean_list();
716 buffer_reg_destroy_registries();
717
718 if (is_root && !opt_no_kernel) {
719 DBG2("Closing kernel fd");
720 if (kernel_tracer_fd >= 0) {
721 ret = close(kernel_tracer_fd);
722 if (ret) {
723 PERROR("close");
724 }
725 }
726 DBG("Unloading kernel modules");
727 modprobe_remove_lttng_all();
728 free(syscall_table);
729 }
730
731 close_consumer_sockets();
732
733 if (load_info) {
734 load_session_destroy_data(load_info);
735 free(load_info);
736 }
737
738 /*
739 * Cleanup lock file by deleting it and finaly closing it which will
740 * release the file system lock.
741 */
742 if (lockfile_fd >= 0) {
743 char lockfile_path[PATH_MAX];
744
745 ret = generate_lock_file_path(lockfile_path,
746 sizeof(lockfile_path));
747 if (ret > 0) {
748 ret = remove(lockfile_path);
749 if (ret < 0) {
750 PERROR("remove lock file");
751 }
752 ret = close(lockfile_fd);
753 if (ret < 0) {
754 PERROR("close lock file");
755 }
756 }
757 }
758
759 /*
760 * We do NOT rmdir rundir because there are other processes
761 * using it, for instance lttng-relayd, which can start in
762 * parallel with this teardown.
763 */
764
765 free(rundir);
766 }
767
768 /*
769 * Cleanup the daemon's option data structures.
770 */
771 static void sessiond_cleanup_options(void)
772 {
773 DBG("Cleaning up options");
774
775 /*
776 * If the override option is set, the pointer points to a *non* const
777 * thus freeing it even though the variable type is set to const.
778 */
779 if (tracing_group_name_override) {
780 free((void *) tracing_group_name);
781 }
782 if (consumerd32_bin_override) {
783 free((void *) consumerd32_bin);
784 }
785 if (consumerd64_bin_override) {
786 free((void *) consumerd64_bin);
787 }
788 if (consumerd32_libdir_override) {
789 free((void *) consumerd32_libdir);
790 }
791 if (consumerd64_libdir_override) {
792 free((void *) consumerd64_libdir);
793 }
794
795 free(opt_pidfile);
796 free(opt_load_session_path);
797 free(kmod_probes_list);
798 free(kmod_extra_probes_list);
799
800 /* <fun> */
801 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
802 "Matthew, BEET driven development works!%c[%dm",
803 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
804 /* </fun> */
805 }
806
807 /*
808 * Send data on a unix socket using the liblttsessiondcomm API.
809 *
810 * Return lttcomm error code.
811 */
812 static int send_unix_sock(int sock, void *buf, size_t len)
813 {
814 /* Check valid length */
815 if (len == 0) {
816 return -1;
817 }
818
819 return lttcomm_send_unix_sock(sock, buf, len);
820 }
821
822 /*
823 * Free memory of a command context structure.
824 */
825 static void clean_command_ctx(struct command_ctx **cmd_ctx)
826 {
827 DBG("Clean command context structure");
828 if (*cmd_ctx) {
829 if ((*cmd_ctx)->llm) {
830 free((*cmd_ctx)->llm);
831 }
832 if ((*cmd_ctx)->lsm) {
833 free((*cmd_ctx)->lsm);
834 }
835 free(*cmd_ctx);
836 *cmd_ctx = NULL;
837 }
838 }
839
840 /*
841 * Notify UST applications using the shm mmap futex.
842 */
843 static int notify_ust_apps(int active)
844 {
845 char *wait_shm_mmap;
846
847 DBG("Notifying applications of session daemon state: %d", active);
848
849 /* See shm.c for this call implying mmap, shm and futex calls */
850 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
851 if (wait_shm_mmap == NULL) {
852 goto error;
853 }
854
855 /* Wake waiting process */
856 futex_wait_update((int32_t *) wait_shm_mmap, active);
857
858 /* Apps notified successfully */
859 return 0;
860
861 error:
862 return -1;
863 }
864
865 /*
866 * Setup the outgoing data buffer for the response (llm) by allocating the
867 * right amount of memory and copying the original information from the lsm
868 * structure.
869 *
870 * Return total size of the buffer pointed by buf.
871 */
872 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
873 {
874 int ret, buf_size;
875
876 buf_size = size;
877
878 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
879 if (cmd_ctx->llm == NULL) {
880 PERROR("zmalloc");
881 ret = -ENOMEM;
882 goto error;
883 }
884
885 /* Copy common data */
886 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
887 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
888
889 cmd_ctx->llm->data_size = size;
890 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
891
892 return buf_size;
893
894 error:
895 return ret;
896 }
897
898 /*
899 * Update the kernel poll set of all channel fd available over all tracing
900 * session. Add the wakeup pipe at the end of the set.
901 */
902 static int update_kernel_poll(struct lttng_poll_event *events)
903 {
904 int ret;
905 struct ltt_session *session;
906 struct ltt_kernel_channel *channel;
907
908 DBG("Updating kernel poll set");
909
910 session_lock_list();
911 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
912 session_lock(session);
913 if (session->kernel_session == NULL) {
914 session_unlock(session);
915 continue;
916 }
917
918 cds_list_for_each_entry(channel,
919 &session->kernel_session->channel_list.head, list) {
920 /* Add channel fd to the kernel poll set */
921 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
922 if (ret < 0) {
923 session_unlock(session);
924 goto error;
925 }
926 DBG("Channel fd %d added to kernel set", channel->fd);
927 }
928 session_unlock(session);
929 }
930 session_unlock_list();
931
932 return 0;
933
934 error:
935 session_unlock_list();
936 return -1;
937 }
938
939 /*
940 * Find the channel fd from 'fd' over all tracing session. When found, check
941 * for new channel stream and send those stream fds to the kernel consumer.
942 *
943 * Useful for CPU hotplug feature.
944 */
945 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
946 {
947 int ret = 0;
948 struct ltt_session *session;
949 struct ltt_kernel_session *ksess;
950 struct ltt_kernel_channel *channel;
951
952 DBG("Updating kernel streams for channel fd %d", fd);
953
954 session_lock_list();
955 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
956 session_lock(session);
957 if (session->kernel_session == NULL) {
958 session_unlock(session);
959 continue;
960 }
961 ksess = session->kernel_session;
962
963 cds_list_for_each_entry(channel,
964 &ksess->channel_list.head, list) {
965 struct lttng_ht_iter iter;
966 struct consumer_socket *socket;
967
968 if (channel->fd != fd) {
969 continue;
970 }
971 DBG("Channel found, updating kernel streams");
972 ret = kernel_open_channel_stream(channel);
973 if (ret < 0) {
974 goto error;
975 }
976 /* Update the stream global counter */
977 ksess->stream_count_global += ret;
978
979 /*
980 * Have we already sent fds to the consumer? If yes, it
981 * means that tracing is started so it is safe to send
982 * our updated stream fds.
983 */
984 if (ksess->consumer_fds_sent != 1
985 || ksess->consumer == NULL) {
986 ret = -1;
987 goto error;
988 }
989
990 rcu_read_lock();
991 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
992 &iter.iter, socket, node.node) {
993 pthread_mutex_lock(socket->lock);
994 ret = kernel_consumer_send_channel_stream(socket,
995 channel, ksess,
996 session->output_traces ? 1 : 0);
997 pthread_mutex_unlock(socket->lock);
998 if (ret < 0) {
999 rcu_read_unlock();
1000 goto error;
1001 }
1002 }
1003 rcu_read_unlock();
1004 }
1005 session_unlock(session);
1006 }
1007 session_unlock_list();
1008 return ret;
1009
1010 error:
1011 session_unlock(session);
1012 session_unlock_list();
1013 return ret;
1014 }
1015
1016 /*
1017 * For each tracing session, update newly registered apps. The session list
1018 * lock MUST be acquired before calling this.
1019 */
1020 static void update_ust_app(int app_sock)
1021 {
1022 struct ltt_session *sess, *stmp;
1023
1024 /* Consumer is in an ERROR state. Stop any application update. */
1025 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1026 /* Stop the update process since the consumer is dead. */
1027 return;
1028 }
1029
1030 /* For all tracing session(s) */
1031 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1032 struct ust_app *app;
1033
1034 session_lock(sess);
1035 if (!sess->ust_session) {
1036 goto unlock_session;
1037 }
1038
1039 rcu_read_lock();
1040 assert(app_sock >= 0);
1041 app = ust_app_find_by_sock(app_sock);
1042 if (app == NULL) {
1043 /*
1044 * Application can be unregistered before so
1045 * this is possible hence simply stopping the
1046 * update.
1047 */
1048 DBG3("UST app update failed to find app sock %d",
1049 app_sock);
1050 goto unlock_rcu;
1051 }
1052 ust_app_global_update(sess->ust_session, app);
1053 unlock_rcu:
1054 rcu_read_unlock();
1055 unlock_session:
1056 session_unlock(sess);
1057 }
1058 }
1059
1060 /*
1061 * This thread manage event coming from the kernel.
1062 *
1063 * Features supported in this thread:
1064 * -) CPU Hotplug
1065 */
1066 static void *thread_manage_kernel(void *data)
1067 {
1068 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1069 uint32_t revents, nb_fd;
1070 char tmp;
1071 struct lttng_poll_event events;
1072
1073 DBG("[thread] Thread manage kernel started");
1074
1075 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1076
1077 /*
1078 * This first step of the while is to clean this structure which could free
1079 * non NULL pointers so initialize it before the loop.
1080 */
1081 lttng_poll_init(&events);
1082
1083 if (testpoint(sessiond_thread_manage_kernel)) {
1084 goto error_testpoint;
1085 }
1086
1087 health_code_update();
1088
1089 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1090 goto error_testpoint;
1091 }
1092
1093 while (1) {
1094 health_code_update();
1095
1096 if (update_poll_flag == 1) {
1097 /* Clean events object. We are about to populate it again. */
1098 lttng_poll_clean(&events);
1099
1100 ret = sessiond_set_thread_pollset(&events, 2);
1101 if (ret < 0) {
1102 goto error_poll_create;
1103 }
1104
1105 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1106 if (ret < 0) {
1107 goto error;
1108 }
1109
1110 /* This will add the available kernel channel if any. */
1111 ret = update_kernel_poll(&events);
1112 if (ret < 0) {
1113 goto error;
1114 }
1115 update_poll_flag = 0;
1116 }
1117
1118 DBG("Thread kernel polling");
1119
1120 /* Poll infinite value of time */
1121 restart:
1122 health_poll_entry();
1123 ret = lttng_poll_wait(&events, -1);
1124 DBG("Thread kernel return from poll on %d fds",
1125 LTTNG_POLL_GETNB(&events));
1126 health_poll_exit();
1127 if (ret < 0) {
1128 /*
1129 * Restart interrupted system call.
1130 */
1131 if (errno == EINTR) {
1132 goto restart;
1133 }
1134 goto error;
1135 } else if (ret == 0) {
1136 /* Should not happen since timeout is infinite */
1137 ERR("Return value of poll is 0 with an infinite timeout.\n"
1138 "This should not have happened! Continuing...");
1139 continue;
1140 }
1141
1142 nb_fd = ret;
1143
1144 for (i = 0; i < nb_fd; i++) {
1145 /* Fetch once the poll data */
1146 revents = LTTNG_POLL_GETEV(&events, i);
1147 pollfd = LTTNG_POLL_GETFD(&events, i);
1148
1149 health_code_update();
1150
1151 if (!revents) {
1152 /* No activity for this FD (poll implementation). */
1153 continue;
1154 }
1155
1156 /* Thread quit pipe has been closed. Killing thread. */
1157 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1158 if (ret) {
1159 err = 0;
1160 goto exit;
1161 }
1162
1163 /* Check for data on kernel pipe */
1164 if (revents & LPOLLIN) {
1165 if (pollfd == kernel_poll_pipe[0]) {
1166 (void) lttng_read(kernel_poll_pipe[0],
1167 &tmp, 1);
1168 /*
1169 * Ret value is useless here, if this pipe gets any actions an
1170 * update is required anyway.
1171 */
1172 update_poll_flag = 1;
1173 continue;
1174 } else {
1175 /*
1176 * New CPU detected by the kernel. Adding kernel stream to
1177 * kernel session and updating the kernel consumer
1178 */
1179 ret = update_kernel_stream(&kconsumer_data, pollfd);
1180 if (ret < 0) {
1181 continue;
1182 }
1183 break;
1184 }
1185 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1186 update_poll_flag = 1;
1187 continue;
1188 } else {
1189 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1190 goto error;
1191 }
1192 }
1193 }
1194
1195 exit:
1196 error:
1197 lttng_poll_clean(&events);
1198 error_poll_create:
1199 error_testpoint:
1200 utils_close_pipe(kernel_poll_pipe);
1201 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1202 if (err) {
1203 health_error();
1204 ERR("Health error occurred in %s", __func__);
1205 WARN("Kernel thread died unexpectedly. "
1206 "Kernel tracing can continue but CPU hotplug is disabled.");
1207 }
1208 health_unregister(health_sessiond);
1209 DBG("Kernel thread dying");
1210 return NULL;
1211 }
1212
1213 /*
1214 * Signal pthread condition of the consumer data that the thread.
1215 */
1216 static void signal_consumer_condition(struct consumer_data *data, int state)
1217 {
1218 pthread_mutex_lock(&data->cond_mutex);
1219
1220 /*
1221 * The state is set before signaling. It can be any value, it's the waiter
1222 * job to correctly interpret this condition variable associated to the
1223 * consumer pthread_cond.
1224 *
1225 * A value of 0 means that the corresponding thread of the consumer data
1226 * was not started. 1 indicates that the thread has started and is ready
1227 * for action. A negative value means that there was an error during the
1228 * thread bootstrap.
1229 */
1230 data->consumer_thread_is_ready = state;
1231 (void) pthread_cond_signal(&data->cond);
1232
1233 pthread_mutex_unlock(&data->cond_mutex);
1234 }
1235
1236 /*
1237 * This thread manage the consumer error sent back to the session daemon.
1238 */
1239 static void *thread_manage_consumer(void *data)
1240 {
1241 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1242 uint32_t revents, nb_fd;
1243 enum lttcomm_return_code code;
1244 struct lttng_poll_event events;
1245 struct consumer_data *consumer_data = data;
1246
1247 DBG("[thread] Manage consumer started");
1248
1249 rcu_register_thread();
1250 rcu_thread_online();
1251
1252 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1253
1254 health_code_update();
1255
1256 /*
1257 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1258 * metadata_sock. Nothing more will be added to this poll set.
1259 */
1260 ret = sessiond_set_thread_pollset(&events, 3);
1261 if (ret < 0) {
1262 goto error_poll;
1263 }
1264
1265 /*
1266 * The error socket here is already in a listening state which was done
1267 * just before spawning this thread to avoid a race between the consumer
1268 * daemon exec trying to connect and the listen() call.
1269 */
1270 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1271 if (ret < 0) {
1272 goto error;
1273 }
1274
1275 health_code_update();
1276
1277 /* Infinite blocking call, waiting for transmission */
1278 restart:
1279 health_poll_entry();
1280
1281 if (testpoint(sessiond_thread_manage_consumer)) {
1282 goto error;
1283 }
1284
1285 ret = lttng_poll_wait(&events, -1);
1286 health_poll_exit();
1287 if (ret < 0) {
1288 /*
1289 * Restart interrupted system call.
1290 */
1291 if (errno == EINTR) {
1292 goto restart;
1293 }
1294 goto error;
1295 }
1296
1297 nb_fd = ret;
1298
1299 for (i = 0; i < nb_fd; i++) {
1300 /* Fetch once the poll data */
1301 revents = LTTNG_POLL_GETEV(&events, i);
1302 pollfd = LTTNG_POLL_GETFD(&events, i);
1303
1304 health_code_update();
1305
1306 if (!revents) {
1307 /* No activity for this FD (poll implementation). */
1308 continue;
1309 }
1310
1311 /* Thread quit pipe has been closed. Killing thread. */
1312 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1313 if (ret) {
1314 err = 0;
1315 goto exit;
1316 }
1317
1318 /* Event on the registration socket */
1319 if (pollfd == consumer_data->err_sock) {
1320 if (revents & LPOLLIN) {
1321 continue;
1322 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1323 ERR("consumer err socket poll error");
1324 goto error;
1325 } else {
1326 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1327 goto error;
1328 }
1329 }
1330 }
1331
1332 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1333 if (sock < 0) {
1334 goto error;
1335 }
1336
1337 /*
1338 * Set the CLOEXEC flag. Return code is useless because either way, the
1339 * show must go on.
1340 */
1341 (void) utils_set_fd_cloexec(sock);
1342
1343 health_code_update();
1344
1345 DBG2("Receiving code from consumer err_sock");
1346
1347 /* Getting status code from kconsumerd */
1348 ret = lttcomm_recv_unix_sock(sock, &code,
1349 sizeof(enum lttcomm_return_code));
1350 if (ret <= 0) {
1351 goto error;
1352 }
1353
1354 health_code_update();
1355 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1356 /* Connect both socket, command and metadata. */
1357 consumer_data->cmd_sock =
1358 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1359 consumer_data->metadata_fd =
1360 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1361 if (consumer_data->cmd_sock < 0
1362 || consumer_data->metadata_fd < 0) {
1363 PERROR("consumer connect cmd socket");
1364 /* On error, signal condition and quit. */
1365 signal_consumer_condition(consumer_data, -1);
1366 goto error;
1367 }
1368 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1369 /* Create metadata socket lock. */
1370 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1371 if (consumer_data->metadata_sock.lock == NULL) {
1372 PERROR("zmalloc pthread mutex");
1373 ret = -1;
1374 goto error;
1375 }
1376 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1377
1378 signal_consumer_condition(consumer_data, 1);
1379 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1380 DBG("Consumer metadata socket ready (fd: %d)",
1381 consumer_data->metadata_fd);
1382 } else {
1383 ERR("consumer error when waiting for SOCK_READY : %s",
1384 lttcomm_get_readable_code(-code));
1385 goto error;
1386 }
1387
1388 /* Remove the consumerd error sock since we've established a connexion */
1389 ret = lttng_poll_del(&events, consumer_data->err_sock);
1390 if (ret < 0) {
1391 goto error;
1392 }
1393
1394 /* Add new accepted error socket. */
1395 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1396 if (ret < 0) {
1397 goto error;
1398 }
1399
1400 /* Add metadata socket that is successfully connected. */
1401 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1402 LPOLLIN | LPOLLRDHUP);
1403 if (ret < 0) {
1404 goto error;
1405 }
1406
1407 health_code_update();
1408
1409 /* Infinite blocking call, waiting for transmission */
1410 restart_poll:
1411 while (1) {
1412 health_code_update();
1413
1414 /* Exit the thread because the thread quit pipe has been triggered. */
1415 if (should_quit) {
1416 /* Not a health error. */
1417 err = 0;
1418 goto exit;
1419 }
1420
1421 health_poll_entry();
1422 ret = lttng_poll_wait(&events, -1);
1423 health_poll_exit();
1424 if (ret < 0) {
1425 /*
1426 * Restart interrupted system call.
1427 */
1428 if (errno == EINTR) {
1429 goto restart_poll;
1430 }
1431 goto error;
1432 }
1433
1434 nb_fd = ret;
1435
1436 for (i = 0; i < nb_fd; i++) {
1437 /* Fetch once the poll data */
1438 revents = LTTNG_POLL_GETEV(&events, i);
1439 pollfd = LTTNG_POLL_GETFD(&events, i);
1440
1441 health_code_update();
1442
1443 if (!revents) {
1444 /* No activity for this FD (poll implementation). */
1445 continue;
1446 }
1447
1448 /*
1449 * Thread quit pipe has been triggered, flag that we should stop
1450 * but continue the current loop to handle potential data from
1451 * consumer.
1452 */
1453 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1454
1455 if (pollfd == sock) {
1456 /* Event on the consumerd socket */
1457 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1458 && !(revents & LPOLLIN)) {
1459 ERR("consumer err socket second poll error");
1460 goto error;
1461 }
1462 health_code_update();
1463 /* Wait for any kconsumerd error */
1464 ret = lttcomm_recv_unix_sock(sock, &code,
1465 sizeof(enum lttcomm_return_code));
1466 if (ret <= 0) {
1467 ERR("consumer closed the command socket");
1468 goto error;
1469 }
1470
1471 ERR("consumer return code : %s",
1472 lttcomm_get_readable_code(-code));
1473
1474 goto exit;
1475 } else if (pollfd == consumer_data->metadata_fd) {
1476 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1477 && !(revents & LPOLLIN)) {
1478 ERR("consumer err metadata socket second poll error");
1479 goto error;
1480 }
1481 /* UST metadata requests */
1482 ret = ust_consumer_metadata_request(
1483 &consumer_data->metadata_sock);
1484 if (ret < 0) {
1485 ERR("Handling metadata request");
1486 goto error;
1487 }
1488 }
1489 /* No need for an else branch all FDs are tested prior. */
1490 }
1491 health_code_update();
1492 }
1493
1494 exit:
1495 error:
1496 /*
1497 * We lock here because we are about to close the sockets and some other
1498 * thread might be using them so get exclusive access which will abort all
1499 * other consumer command by other threads.
1500 */
1501 pthread_mutex_lock(&consumer_data->lock);
1502
1503 /* Immediately set the consumerd state to stopped */
1504 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1505 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1506 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1507 consumer_data->type == LTTNG_CONSUMER32_UST) {
1508 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1509 } else {
1510 /* Code flow error... */
1511 assert(0);
1512 }
1513
1514 if (consumer_data->err_sock >= 0) {
1515 ret = close(consumer_data->err_sock);
1516 if (ret) {
1517 PERROR("close");
1518 }
1519 consumer_data->err_sock = -1;
1520 }
1521 if (consumer_data->cmd_sock >= 0) {
1522 ret = close(consumer_data->cmd_sock);
1523 if (ret) {
1524 PERROR("close");
1525 }
1526 consumer_data->cmd_sock = -1;
1527 }
1528 if (consumer_data->metadata_sock.fd_ptr &&
1529 *consumer_data->metadata_sock.fd_ptr >= 0) {
1530 ret = close(*consumer_data->metadata_sock.fd_ptr);
1531 if (ret) {
1532 PERROR("close");
1533 }
1534 }
1535 if (sock >= 0) {
1536 ret = close(sock);
1537 if (ret) {
1538 PERROR("close");
1539 }
1540 }
1541
1542 unlink(consumer_data->err_unix_sock_path);
1543 unlink(consumer_data->cmd_unix_sock_path);
1544 pthread_mutex_unlock(&consumer_data->lock);
1545
1546 /* Cleanup metadata socket mutex. */
1547 if (consumer_data->metadata_sock.lock) {
1548 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1549 free(consumer_data->metadata_sock.lock);
1550 }
1551 lttng_poll_clean(&events);
1552 error_poll:
1553 if (err) {
1554 health_error();
1555 ERR("Health error occurred in %s", __func__);
1556 }
1557 health_unregister(health_sessiond);
1558 DBG("consumer thread cleanup completed");
1559
1560 rcu_thread_offline();
1561 rcu_unregister_thread();
1562
1563 return NULL;
1564 }
1565
1566 /*
1567 * This thread manage application communication.
1568 */
1569 static void *thread_manage_apps(void *data)
1570 {
1571 int i, ret, pollfd, err = -1;
1572 ssize_t size_ret;
1573 uint32_t revents, nb_fd;
1574 struct lttng_poll_event events;
1575
1576 DBG("[thread] Manage application started");
1577
1578 rcu_register_thread();
1579 rcu_thread_online();
1580
1581 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1582
1583 if (testpoint(sessiond_thread_manage_apps)) {
1584 goto error_testpoint;
1585 }
1586
1587 health_code_update();
1588
1589 ret = sessiond_set_thread_pollset(&events, 2);
1590 if (ret < 0) {
1591 goto error_poll_create;
1592 }
1593
1594 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1595 if (ret < 0) {
1596 goto error;
1597 }
1598
1599 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1600 goto error;
1601 }
1602
1603 health_code_update();
1604
1605 while (1) {
1606 DBG("Apps thread polling");
1607
1608 /* Inifinite blocking call, waiting for transmission */
1609 restart:
1610 health_poll_entry();
1611 ret = lttng_poll_wait(&events, -1);
1612 DBG("Apps thread return from poll on %d fds",
1613 LTTNG_POLL_GETNB(&events));
1614 health_poll_exit();
1615 if (ret < 0) {
1616 /*
1617 * Restart interrupted system call.
1618 */
1619 if (errno == EINTR) {
1620 goto restart;
1621 }
1622 goto error;
1623 }
1624
1625 nb_fd = ret;
1626
1627 for (i = 0; i < nb_fd; i++) {
1628 /* Fetch once the poll data */
1629 revents = LTTNG_POLL_GETEV(&events, i);
1630 pollfd = LTTNG_POLL_GETFD(&events, i);
1631
1632 health_code_update();
1633
1634 if (!revents) {
1635 /* No activity for this FD (poll implementation). */
1636 continue;
1637 }
1638
1639 /* Thread quit pipe has been closed. Killing thread. */
1640 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1641 if (ret) {
1642 err = 0;
1643 goto exit;
1644 }
1645
1646 /* Inspect the apps cmd pipe */
1647 if (pollfd == apps_cmd_pipe[0]) {
1648 if (revents & LPOLLIN) {
1649 int sock;
1650
1651 /* Empty pipe */
1652 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1653 if (size_ret < sizeof(sock)) {
1654 PERROR("read apps cmd pipe");
1655 goto error;
1656 }
1657
1658 health_code_update();
1659
1660 /*
1661 * Since this is a command socket (write then read),
1662 * we only monitor the error events of the socket.
1663 */
1664 ret = lttng_poll_add(&events, sock,
1665 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1666 if (ret < 0) {
1667 goto error;
1668 }
1669
1670 DBG("Apps with sock %d added to poll set", sock);
1671 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1672 ERR("Apps command pipe error");
1673 goto error;
1674 } else {
1675 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1676 goto error;
1677 }
1678 } else {
1679 /*
1680 * At this point, we know that a registered application made
1681 * the event at poll_wait.
1682 */
1683 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1684 /* Removing from the poll set */
1685 ret = lttng_poll_del(&events, pollfd);
1686 if (ret < 0) {
1687 goto error;
1688 }
1689
1690 /* Socket closed on remote end. */
1691 ust_app_unregister(pollfd);
1692 } else {
1693 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1694 goto error;
1695 }
1696 }
1697
1698 health_code_update();
1699 }
1700 }
1701
1702 exit:
1703 error:
1704 lttng_poll_clean(&events);
1705 error_poll_create:
1706 error_testpoint:
1707 utils_close_pipe(apps_cmd_pipe);
1708 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1709
1710 /*
1711 * We don't clean the UST app hash table here since already registered
1712 * applications can still be controlled so let them be until the session
1713 * daemon dies or the applications stop.
1714 */
1715
1716 if (err) {
1717 health_error();
1718 ERR("Health error occurred in %s", __func__);
1719 }
1720 health_unregister(health_sessiond);
1721 DBG("Application communication apps thread cleanup complete");
1722 rcu_thread_offline();
1723 rcu_unregister_thread();
1724 return NULL;
1725 }
1726
1727 /*
1728 * Send a socket to a thread This is called from the dispatch UST registration
1729 * thread once all sockets are set for the application.
1730 *
1731 * The sock value can be invalid, we don't really care, the thread will handle
1732 * it and make the necessary cleanup if so.
1733 *
1734 * On success, return 0 else a negative value being the errno message of the
1735 * write().
1736 */
1737 static int send_socket_to_thread(int fd, int sock)
1738 {
1739 ssize_t ret;
1740
1741 /*
1742 * It's possible that the FD is set as invalid with -1 concurrently just
1743 * before calling this function being a shutdown state of the thread.
1744 */
1745 if (fd < 0) {
1746 ret = -EBADF;
1747 goto error;
1748 }
1749
1750 ret = lttng_write(fd, &sock, sizeof(sock));
1751 if (ret < sizeof(sock)) {
1752 PERROR("write apps pipe %d", fd);
1753 if (ret < 0) {
1754 ret = -errno;
1755 }
1756 goto error;
1757 }
1758
1759 /* All good. Don't send back the write positive ret value. */
1760 ret = 0;
1761 error:
1762 return (int) ret;
1763 }
1764
1765 /*
1766 * Sanitize the wait queue of the dispatch registration thread meaning removing
1767 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1768 * notify socket is never received.
1769 */
1770 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1771 {
1772 int ret, nb_fd = 0, i;
1773 unsigned int fd_added = 0;
1774 struct lttng_poll_event events;
1775 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1776
1777 assert(wait_queue);
1778
1779 lttng_poll_init(&events);
1780
1781 /* Just skip everything for an empty queue. */
1782 if (!wait_queue->count) {
1783 goto end;
1784 }
1785
1786 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1787 if (ret < 0) {
1788 goto error_create;
1789 }
1790
1791 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1792 &wait_queue->head, head) {
1793 assert(wait_node->app);
1794 ret = lttng_poll_add(&events, wait_node->app->sock,
1795 LPOLLHUP | LPOLLERR);
1796 if (ret < 0) {
1797 goto error;
1798 }
1799
1800 fd_added = 1;
1801 }
1802
1803 if (!fd_added) {
1804 goto end;
1805 }
1806
1807 /*
1808 * Poll but don't block so we can quickly identify the faulty events and
1809 * clean them afterwards from the wait queue.
1810 */
1811 ret = lttng_poll_wait(&events, 0);
1812 if (ret < 0) {
1813 goto error;
1814 }
1815 nb_fd = ret;
1816
1817 for (i = 0; i < nb_fd; i++) {
1818 /* Get faulty FD. */
1819 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1820 int pollfd = LTTNG_POLL_GETFD(&events, i);
1821
1822 if (!revents) {
1823 /* No activity for this FD (poll implementation). */
1824 continue;
1825 }
1826
1827 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1828 &wait_queue->head, head) {
1829 if (pollfd == wait_node->app->sock &&
1830 (revents & (LPOLLHUP | LPOLLERR))) {
1831 cds_list_del(&wait_node->head);
1832 wait_queue->count--;
1833 ust_app_destroy(wait_node->app);
1834 free(wait_node);
1835 break;
1836 } else {
1837 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1838 goto error;
1839 }
1840 }
1841 }
1842
1843 if (nb_fd > 0) {
1844 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1845 }
1846
1847 end:
1848 lttng_poll_clean(&events);
1849 return;
1850
1851 error:
1852 lttng_poll_clean(&events);
1853 error_create:
1854 ERR("Unable to sanitize wait queue");
1855 return;
1856 }
1857
1858 /*
1859 * Dispatch request from the registration threads to the application
1860 * communication thread.
1861 */
1862 static void *thread_dispatch_ust_registration(void *data)
1863 {
1864 int ret, err = -1;
1865 struct cds_wfcq_node *node;
1866 struct ust_command *ust_cmd = NULL;
1867 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1868 struct ust_reg_wait_queue wait_queue = {
1869 .count = 0,
1870 };
1871
1872 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1873
1874 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1875 goto error_testpoint;
1876 }
1877
1878 health_code_update();
1879
1880 CDS_INIT_LIST_HEAD(&wait_queue.head);
1881
1882 DBG("[thread] Dispatch UST command started");
1883
1884 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1885 health_code_update();
1886
1887 /* Atomically prepare the queue futex */
1888 futex_nto1_prepare(&ust_cmd_queue.futex);
1889
1890 do {
1891 struct ust_app *app = NULL;
1892 ust_cmd = NULL;
1893
1894 /*
1895 * Make sure we don't have node(s) that have hung up before receiving
1896 * the notify socket. This is to clean the list in order to avoid
1897 * memory leaks from notify socket that are never seen.
1898 */
1899 sanitize_wait_queue(&wait_queue);
1900
1901 health_code_update();
1902 /* Dequeue command for registration */
1903 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1904 if (node == NULL) {
1905 DBG("Woken up but nothing in the UST command queue");
1906 /* Continue thread execution */
1907 break;
1908 }
1909
1910 ust_cmd = caa_container_of(node, struct ust_command, node);
1911
1912 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1913 " gid:%d sock:%d name:%s (version %d.%d)",
1914 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1915 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1916 ust_cmd->sock, ust_cmd->reg_msg.name,
1917 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1918
1919 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1920 wait_node = zmalloc(sizeof(*wait_node));
1921 if (!wait_node) {
1922 PERROR("zmalloc wait_node dispatch");
1923 ret = close(ust_cmd->sock);
1924 if (ret < 0) {
1925 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1926 }
1927 lttng_fd_put(LTTNG_FD_APPS, 1);
1928 free(ust_cmd);
1929 goto error;
1930 }
1931 CDS_INIT_LIST_HEAD(&wait_node->head);
1932
1933 /* Create application object if socket is CMD. */
1934 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1935 ust_cmd->sock);
1936 if (!wait_node->app) {
1937 ret = close(ust_cmd->sock);
1938 if (ret < 0) {
1939 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1940 }
1941 lttng_fd_put(LTTNG_FD_APPS, 1);
1942 free(wait_node);
1943 free(ust_cmd);
1944 continue;
1945 }
1946 /*
1947 * Add application to the wait queue so we can set the notify
1948 * socket before putting this object in the global ht.
1949 */
1950 cds_list_add(&wait_node->head, &wait_queue.head);
1951 wait_queue.count++;
1952
1953 free(ust_cmd);
1954 /*
1955 * We have to continue here since we don't have the notify
1956 * socket and the application MUST be added to the hash table
1957 * only at that moment.
1958 */
1959 continue;
1960 } else {
1961 /*
1962 * Look for the application in the local wait queue and set the
1963 * notify socket if found.
1964 */
1965 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1966 &wait_queue.head, head) {
1967 health_code_update();
1968 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1969 wait_node->app->notify_sock = ust_cmd->sock;
1970 cds_list_del(&wait_node->head);
1971 wait_queue.count--;
1972 app = wait_node->app;
1973 free(wait_node);
1974 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1975 break;
1976 }
1977 }
1978
1979 /*
1980 * With no application at this stage the received socket is
1981 * basically useless so close it before we free the cmd data
1982 * structure for good.
1983 */
1984 if (!app) {
1985 ret = close(ust_cmd->sock);
1986 if (ret < 0) {
1987 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1988 }
1989 lttng_fd_put(LTTNG_FD_APPS, 1);
1990 }
1991 free(ust_cmd);
1992 }
1993
1994 if (app) {
1995 /*
1996 * @session_lock_list
1997 *
1998 * Lock the global session list so from the register up to the
1999 * registration done message, no thread can see the application
2000 * and change its state.
2001 */
2002 session_lock_list();
2003 rcu_read_lock();
2004
2005 /*
2006 * Add application to the global hash table. This needs to be
2007 * done before the update to the UST registry can locate the
2008 * application.
2009 */
2010 ust_app_add(app);
2011
2012 /* Set app version. This call will print an error if needed. */
2013 (void) ust_app_version(app);
2014
2015 /* Send notify socket through the notify pipe. */
2016 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2017 app->notify_sock);
2018 if (ret < 0) {
2019 rcu_read_unlock();
2020 session_unlock_list();
2021 /*
2022 * No notify thread, stop the UST tracing. However, this is
2023 * not an internal error of the this thread thus setting
2024 * the health error code to a normal exit.
2025 */
2026 err = 0;
2027 goto error;
2028 }
2029
2030 /*
2031 * Update newly registered application with the tracing
2032 * registry info already enabled information.
2033 */
2034 update_ust_app(app->sock);
2035
2036 /*
2037 * Don't care about return value. Let the manage apps threads
2038 * handle app unregistration upon socket close.
2039 */
2040 (void) ust_app_register_done(app);
2041
2042 /*
2043 * Even if the application socket has been closed, send the app
2044 * to the thread and unregistration will take place at that
2045 * place.
2046 */
2047 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2048 if (ret < 0) {
2049 rcu_read_unlock();
2050 session_unlock_list();
2051 /*
2052 * No apps. thread, stop the UST tracing. However, this is
2053 * not an internal error of the this thread thus setting
2054 * the health error code to a normal exit.
2055 */
2056 err = 0;
2057 goto error;
2058 }
2059
2060 rcu_read_unlock();
2061 session_unlock_list();
2062 }
2063 } while (node != NULL);
2064
2065 health_poll_entry();
2066 /* Futex wait on queue. Blocking call on futex() */
2067 futex_nto1_wait(&ust_cmd_queue.futex);
2068 health_poll_exit();
2069 }
2070 /* Normal exit, no error */
2071 err = 0;
2072
2073 error:
2074 /* Clean up wait queue. */
2075 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2076 &wait_queue.head, head) {
2077 cds_list_del(&wait_node->head);
2078 wait_queue.count--;
2079 free(wait_node);
2080 }
2081
2082 /* Empty command queue. */
2083 for (;;) {
2084 /* Dequeue command for registration */
2085 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2086 if (node == NULL) {
2087 break;
2088 }
2089 ust_cmd = caa_container_of(node, struct ust_command, node);
2090 ret = close(ust_cmd->sock);
2091 if (ret < 0) {
2092 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2093 }
2094 lttng_fd_put(LTTNG_FD_APPS, 1);
2095 free(ust_cmd);
2096 }
2097
2098 error_testpoint:
2099 DBG("Dispatch thread dying");
2100 if (err) {
2101 health_error();
2102 ERR("Health error occurred in %s", __func__);
2103 }
2104 health_unregister(health_sessiond);
2105 return NULL;
2106 }
2107
2108 /*
2109 * This thread manage application registration.
2110 */
2111 static void *thread_registration_apps(void *data)
2112 {
2113 int sock = -1, i, ret, pollfd, err = -1;
2114 uint32_t revents, nb_fd;
2115 struct lttng_poll_event events;
2116 /*
2117 * Get allocated in this thread, enqueued to a global queue, dequeued and
2118 * freed in the manage apps thread.
2119 */
2120 struct ust_command *ust_cmd = NULL;
2121
2122 DBG("[thread] Manage application registration started");
2123
2124 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2125
2126 if (testpoint(sessiond_thread_registration_apps)) {
2127 goto error_testpoint;
2128 }
2129
2130 ret = lttcomm_listen_unix_sock(apps_sock);
2131 if (ret < 0) {
2132 goto error_listen;
2133 }
2134
2135 /*
2136 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2137 * more will be added to this poll set.
2138 */
2139 ret = sessiond_set_thread_pollset(&events, 2);
2140 if (ret < 0) {
2141 goto error_create_poll;
2142 }
2143
2144 /* Add the application registration socket */
2145 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2146 if (ret < 0) {
2147 goto error_poll_add;
2148 }
2149
2150 /* Notify all applications to register */
2151 ret = notify_ust_apps(1);
2152 if (ret < 0) {
2153 ERR("Failed to notify applications or create the wait shared memory.\n"
2154 "Execution continues but there might be problem for already\n"
2155 "running applications that wishes to register.");
2156 }
2157
2158 while (1) {
2159 DBG("Accepting application registration");
2160
2161 /* Inifinite blocking call, waiting for transmission */
2162 restart:
2163 health_poll_entry();
2164 ret = lttng_poll_wait(&events, -1);
2165 health_poll_exit();
2166 if (ret < 0) {
2167 /*
2168 * Restart interrupted system call.
2169 */
2170 if (errno == EINTR) {
2171 goto restart;
2172 }
2173 goto error;
2174 }
2175
2176 nb_fd = ret;
2177
2178 for (i = 0; i < nb_fd; i++) {
2179 health_code_update();
2180
2181 /* Fetch once the poll data */
2182 revents = LTTNG_POLL_GETEV(&events, i);
2183 pollfd = LTTNG_POLL_GETFD(&events, i);
2184
2185 if (!revents) {
2186 /* No activity for this FD (poll implementation). */
2187 continue;
2188 }
2189
2190 /* Thread quit pipe has been closed. Killing thread. */
2191 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2192 if (ret) {
2193 err = 0;
2194 goto exit;
2195 }
2196
2197 /* Event on the registration socket */
2198 if (pollfd == apps_sock) {
2199 if (revents & LPOLLIN) {
2200 sock = lttcomm_accept_unix_sock(apps_sock);
2201 if (sock < 0) {
2202 goto error;
2203 }
2204
2205 /*
2206 * Set socket timeout for both receiving and ending.
2207 * app_socket_timeout is in seconds, whereas
2208 * lttcomm_setsockopt_rcv_timeout and
2209 * lttcomm_setsockopt_snd_timeout expect msec as
2210 * parameter.
2211 */
2212 (void) lttcomm_setsockopt_rcv_timeout(sock,
2213 app_socket_timeout * 1000);
2214 (void) lttcomm_setsockopt_snd_timeout(sock,
2215 app_socket_timeout * 1000);
2216
2217 /*
2218 * Set the CLOEXEC flag. Return code is useless because
2219 * either way, the show must go on.
2220 */
2221 (void) utils_set_fd_cloexec(sock);
2222
2223 /* Create UST registration command for enqueuing */
2224 ust_cmd = zmalloc(sizeof(struct ust_command));
2225 if (ust_cmd == NULL) {
2226 PERROR("ust command zmalloc");
2227 ret = close(sock);
2228 if (ret) {
2229 PERROR("close");
2230 }
2231 goto error;
2232 }
2233
2234 /*
2235 * Using message-based transmissions to ensure we don't
2236 * have to deal with partially received messages.
2237 */
2238 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2239 if (ret < 0) {
2240 ERR("Exhausted file descriptors allowed for applications.");
2241 free(ust_cmd);
2242 ret = close(sock);
2243 if (ret) {
2244 PERROR("close");
2245 }
2246 sock = -1;
2247 continue;
2248 }
2249
2250 health_code_update();
2251 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2252 if (ret < 0) {
2253 free(ust_cmd);
2254 /* Close socket of the application. */
2255 ret = close(sock);
2256 if (ret) {
2257 PERROR("close");
2258 }
2259 lttng_fd_put(LTTNG_FD_APPS, 1);
2260 sock = -1;
2261 continue;
2262 }
2263 health_code_update();
2264
2265 ust_cmd->sock = sock;
2266 sock = -1;
2267
2268 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2269 " gid:%d sock:%d name:%s (version %d.%d)",
2270 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2271 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2272 ust_cmd->sock, ust_cmd->reg_msg.name,
2273 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2274
2275 /*
2276 * Lock free enqueue the registration request. The red pill
2277 * has been taken! This apps will be part of the *system*.
2278 */
2279 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2280
2281 /*
2282 * Wake the registration queue futex. Implicit memory
2283 * barrier with the exchange in cds_wfcq_enqueue.
2284 */
2285 futex_nto1_wake(&ust_cmd_queue.futex);
2286 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2287 ERR("Register apps socket poll error");
2288 goto error;
2289 } else {
2290 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2291 goto error;
2292 }
2293 }
2294 }
2295 }
2296
2297 exit:
2298 error:
2299 /* Notify that the registration thread is gone */
2300 notify_ust_apps(0);
2301
2302 if (apps_sock >= 0) {
2303 ret = close(apps_sock);
2304 if (ret) {
2305 PERROR("close");
2306 }
2307 }
2308 if (sock >= 0) {
2309 ret = close(sock);
2310 if (ret) {
2311 PERROR("close");
2312 }
2313 lttng_fd_put(LTTNG_FD_APPS, 1);
2314 }
2315 unlink(apps_unix_sock_path);
2316
2317 error_poll_add:
2318 lttng_poll_clean(&events);
2319 error_listen:
2320 error_create_poll:
2321 error_testpoint:
2322 DBG("UST Registration thread cleanup complete");
2323 if (err) {
2324 health_error();
2325 ERR("Health error occurred in %s", __func__);
2326 }
2327 health_unregister(health_sessiond);
2328
2329 return NULL;
2330 }
2331
2332 /*
2333 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2334 * exec or it will fails.
2335 */
2336 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2337 {
2338 int ret, clock_ret;
2339 struct timespec timeout;
2340
2341 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2342 consumer_data->consumer_thread_is_ready = 0;
2343
2344 /* Setup pthread condition */
2345 ret = pthread_condattr_init(&consumer_data->condattr);
2346 if (ret) {
2347 errno = ret;
2348 PERROR("pthread_condattr_init consumer data");
2349 goto error;
2350 }
2351
2352 /*
2353 * Set the monotonic clock in order to make sure we DO NOT jump in time
2354 * between the clock_gettime() call and the timedwait call. See bug #324
2355 * for a more details and how we noticed it.
2356 */
2357 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2358 if (ret) {
2359 errno = ret;
2360 PERROR("pthread_condattr_setclock consumer data");
2361 goto error;
2362 }
2363
2364 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2365 if (ret) {
2366 errno = ret;
2367 PERROR("pthread_cond_init consumer data");
2368 goto error;
2369 }
2370
2371 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2372 consumer_data);
2373 if (ret) {
2374 errno = ret;
2375 PERROR("pthread_create consumer");
2376 ret = -1;
2377 goto error;
2378 }
2379
2380 /* We are about to wait on a pthread condition */
2381 pthread_mutex_lock(&consumer_data->cond_mutex);
2382
2383 /* Get time for sem_timedwait absolute timeout */
2384 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2385 /*
2386 * Set the timeout for the condition timed wait even if the clock gettime
2387 * call fails since we might loop on that call and we want to avoid to
2388 * increment the timeout too many times.
2389 */
2390 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2391
2392 /*
2393 * The following loop COULD be skipped in some conditions so this is why we
2394 * set ret to 0 in order to make sure at least one round of the loop is
2395 * done.
2396 */
2397 ret = 0;
2398
2399 /*
2400 * Loop until the condition is reached or when a timeout is reached. Note
2401 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2402 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2403 * possible. This loop does not take any chances and works with both of
2404 * them.
2405 */
2406 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2407 if (clock_ret < 0) {
2408 PERROR("clock_gettime spawn consumer");
2409 /* Infinite wait for the consumerd thread to be ready */
2410 ret = pthread_cond_wait(&consumer_data->cond,
2411 &consumer_data->cond_mutex);
2412 } else {
2413 ret = pthread_cond_timedwait(&consumer_data->cond,
2414 &consumer_data->cond_mutex, &timeout);
2415 }
2416 }
2417
2418 /* Release the pthread condition */
2419 pthread_mutex_unlock(&consumer_data->cond_mutex);
2420
2421 if (ret != 0) {
2422 errno = ret;
2423 if (ret == ETIMEDOUT) {
2424 int pth_ret;
2425
2426 /*
2427 * Call has timed out so we kill the kconsumerd_thread and return
2428 * an error.
2429 */
2430 ERR("Condition timed out. The consumer thread was never ready."
2431 " Killing it");
2432 pth_ret = pthread_cancel(consumer_data->thread);
2433 if (pth_ret < 0) {
2434 PERROR("pthread_cancel consumer thread");
2435 }
2436 } else {
2437 PERROR("pthread_cond_wait failed consumer thread");
2438 }
2439 /* Caller is expecting a negative value on failure. */
2440 ret = -1;
2441 goto error;
2442 }
2443
2444 pthread_mutex_lock(&consumer_data->pid_mutex);
2445 if (consumer_data->pid == 0) {
2446 ERR("Consumerd did not start");
2447 pthread_mutex_unlock(&consumer_data->pid_mutex);
2448 goto error;
2449 }
2450 pthread_mutex_unlock(&consumer_data->pid_mutex);
2451
2452 return 0;
2453
2454 error:
2455 return ret;
2456 }
2457
2458 /*
2459 * Join consumer thread
2460 */
2461 static int join_consumer_thread(struct consumer_data *consumer_data)
2462 {
2463 void *status;
2464
2465 /* Consumer pid must be a real one. */
2466 if (consumer_data->pid > 0) {
2467 int ret;
2468 ret = kill(consumer_data->pid, SIGTERM);
2469 if (ret) {
2470 PERROR("Error killing consumer daemon");
2471 return ret;
2472 }
2473 return pthread_join(consumer_data->thread, &status);
2474 } else {
2475 return 0;
2476 }
2477 }
2478
2479 /*
2480 * Fork and exec a consumer daemon (consumerd).
2481 *
2482 * Return pid if successful else -1.
2483 */
2484 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2485 {
2486 int ret;
2487 pid_t pid;
2488 const char *consumer_to_use;
2489 const char *verbosity;
2490 struct stat st;
2491
2492 DBG("Spawning consumerd");
2493
2494 pid = fork();
2495 if (pid == 0) {
2496 /*
2497 * Exec consumerd.
2498 */
2499 if (opt_verbose_consumer) {
2500 verbosity = "--verbose";
2501 } else if (lttng_opt_quiet) {
2502 verbosity = "--quiet";
2503 } else {
2504 verbosity = "";
2505 }
2506
2507 switch (consumer_data->type) {
2508 case LTTNG_CONSUMER_KERNEL:
2509 /*
2510 * Find out which consumerd to execute. We will first try the
2511 * 64-bit path, then the sessiond's installation directory, and
2512 * fallback on the 32-bit one,
2513 */
2514 DBG3("Looking for a kernel consumer at these locations:");
2515 DBG3(" 1) %s", consumerd64_bin);
2516 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2517 DBG3(" 3) %s", consumerd32_bin);
2518 if (stat(consumerd64_bin, &st) == 0) {
2519 DBG3("Found location #1");
2520 consumer_to_use = consumerd64_bin;
2521 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2522 DBG3("Found location #2");
2523 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2524 } else if (stat(consumerd32_bin, &st) == 0) {
2525 DBG3("Found location #3");
2526 consumer_to_use = consumerd32_bin;
2527 } else {
2528 DBG("Could not find any valid consumerd executable");
2529 ret = -EINVAL;
2530 break;
2531 }
2532 DBG("Using kernel consumer at: %s", consumer_to_use);
2533 ret = execl(consumer_to_use,
2534 "lttng-consumerd", verbosity, "-k",
2535 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2536 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2537 "--group", tracing_group_name,
2538 NULL);
2539 break;
2540 case LTTNG_CONSUMER64_UST:
2541 {
2542 char *tmpnew = NULL;
2543
2544 if (consumerd64_libdir[0] != '\0') {
2545 char *tmp;
2546 size_t tmplen;
2547
2548 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2549 if (!tmp) {
2550 tmp = "";
2551 }
2552 tmplen = strlen("LD_LIBRARY_PATH=")
2553 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2554 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2555 if (!tmpnew) {
2556 ret = -ENOMEM;
2557 goto error;
2558 }
2559 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2560 strcat(tmpnew, consumerd64_libdir);
2561 if (tmp[0] != '\0') {
2562 strcat(tmpnew, ":");
2563 strcat(tmpnew, tmp);
2564 }
2565 ret = putenv(tmpnew);
2566 if (ret) {
2567 ret = -errno;
2568 free(tmpnew);
2569 goto error;
2570 }
2571 }
2572 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2573 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2574 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2575 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2576 "--group", tracing_group_name,
2577 NULL);
2578 if (consumerd64_libdir[0] != '\0') {
2579 free(tmpnew);
2580 }
2581 break;
2582 }
2583 case LTTNG_CONSUMER32_UST:
2584 {
2585 char *tmpnew = NULL;
2586
2587 if (consumerd32_libdir[0] != '\0') {
2588 char *tmp;
2589 size_t tmplen;
2590
2591 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2592 if (!tmp) {
2593 tmp = "";
2594 }
2595 tmplen = strlen("LD_LIBRARY_PATH=")
2596 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2597 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2598 if (!tmpnew) {
2599 ret = -ENOMEM;
2600 goto error;
2601 }
2602 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2603 strcat(tmpnew, consumerd32_libdir);
2604 if (tmp[0] != '\0') {
2605 strcat(tmpnew, ":");
2606 strcat(tmpnew, tmp);
2607 }
2608 ret = putenv(tmpnew);
2609 if (ret) {
2610 ret = -errno;
2611 free(tmpnew);
2612 goto error;
2613 }
2614 }
2615 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2616 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2617 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2618 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2619 "--group", tracing_group_name,
2620 NULL);
2621 if (consumerd32_libdir[0] != '\0') {
2622 free(tmpnew);
2623 }
2624 break;
2625 }
2626 default:
2627 PERROR("unknown consumer type");
2628 exit(EXIT_FAILURE);
2629 }
2630 if (errno != 0) {
2631 PERROR("Consumer execl()");
2632 }
2633 /* Reaching this point, we got a failure on our execl(). */
2634 exit(EXIT_FAILURE);
2635 } else if (pid > 0) {
2636 ret = pid;
2637 } else {
2638 PERROR("start consumer fork");
2639 ret = -errno;
2640 }
2641 error:
2642 return ret;
2643 }
2644
2645 /*
2646 * Spawn the consumerd daemon and session daemon thread.
2647 */
2648 static int start_consumerd(struct consumer_data *consumer_data)
2649 {
2650 int ret;
2651
2652 /*
2653 * Set the listen() state on the socket since there is a possible race
2654 * between the exec() of the consumer daemon and this call if place in the
2655 * consumer thread. See bug #366 for more details.
2656 */
2657 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2658 if (ret < 0) {
2659 goto error;
2660 }
2661
2662 pthread_mutex_lock(&consumer_data->pid_mutex);
2663 if (consumer_data->pid != 0) {
2664 pthread_mutex_unlock(&consumer_data->pid_mutex);
2665 goto end;
2666 }
2667
2668 ret = spawn_consumerd(consumer_data);
2669 if (ret < 0) {
2670 ERR("Spawning consumerd failed");
2671 pthread_mutex_unlock(&consumer_data->pid_mutex);
2672 goto error;
2673 }
2674
2675 /* Setting up the consumer_data pid */
2676 consumer_data->pid = ret;
2677 DBG2("Consumer pid %d", consumer_data->pid);
2678 pthread_mutex_unlock(&consumer_data->pid_mutex);
2679
2680 DBG2("Spawning consumer control thread");
2681 ret = spawn_consumer_thread(consumer_data);
2682 if (ret < 0) {
2683 ERR("Fatal error spawning consumer control thread");
2684 goto error;
2685 }
2686
2687 end:
2688 return 0;
2689
2690 error:
2691 /* Cleanup already created sockets on error. */
2692 if (consumer_data->err_sock >= 0) {
2693 int err;
2694
2695 err = close(consumer_data->err_sock);
2696 if (err < 0) {
2697 PERROR("close consumer data error socket");
2698 }
2699 }
2700 return ret;
2701 }
2702
2703 /*
2704 * Setup necessary data for kernel tracer action.
2705 */
2706 static int init_kernel_tracer(void)
2707 {
2708 int ret;
2709
2710 /* Modprobe lttng kernel modules */
2711 ret = modprobe_lttng_control();
2712 if (ret < 0) {
2713 goto error;
2714 }
2715
2716 /* Open debugfs lttng */
2717 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2718 if (kernel_tracer_fd < 0) {
2719 DBG("Failed to open %s", module_proc_lttng);
2720 ret = -1;
2721 goto error_open;
2722 }
2723
2724 /* Validate kernel version */
2725 ret = kernel_validate_version(kernel_tracer_fd);
2726 if (ret < 0) {
2727 goto error_version;
2728 }
2729
2730 ret = modprobe_lttng_data();
2731 if (ret < 0) {
2732 goto error_modules;
2733 }
2734
2735 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2736 return 0;
2737
2738 error_version:
2739 modprobe_remove_lttng_control();
2740 ret = close(kernel_tracer_fd);
2741 if (ret) {
2742 PERROR("close");
2743 }
2744 kernel_tracer_fd = -1;
2745 return LTTNG_ERR_KERN_VERSION;
2746
2747 error_modules:
2748 ret = close(kernel_tracer_fd);
2749 if (ret) {
2750 PERROR("close");
2751 }
2752
2753 error_open:
2754 modprobe_remove_lttng_control();
2755
2756 error:
2757 WARN("No kernel tracer available");
2758 kernel_tracer_fd = -1;
2759 if (!is_root) {
2760 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2761 } else {
2762 return LTTNG_ERR_KERN_NA;
2763 }
2764 }
2765
2766
2767 /*
2768 * Copy consumer output from the tracing session to the domain session. The
2769 * function also applies the right modification on a per domain basis for the
2770 * trace files destination directory.
2771 *
2772 * Should *NOT* be called with RCU read-side lock held.
2773 */
2774 static int copy_session_consumer(int domain, struct ltt_session *session)
2775 {
2776 int ret;
2777 const char *dir_name;
2778 struct consumer_output *consumer;
2779
2780 assert(session);
2781 assert(session->consumer);
2782
2783 switch (domain) {
2784 case LTTNG_DOMAIN_KERNEL:
2785 DBG3("Copying tracing session consumer output in kernel session");
2786 /*
2787 * XXX: We should audit the session creation and what this function
2788 * does "extra" in order to avoid a destroy since this function is used
2789 * in the domain session creation (kernel and ust) only. Same for UST
2790 * domain.
2791 */
2792 if (session->kernel_session->consumer) {
2793 consumer_output_put(session->kernel_session->consumer);
2794 }
2795 session->kernel_session->consumer =
2796 consumer_copy_output(session->consumer);
2797 /* Ease our life a bit for the next part */
2798 consumer = session->kernel_session->consumer;
2799 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2800 break;
2801 case LTTNG_DOMAIN_JUL:
2802 case LTTNG_DOMAIN_LOG4J:
2803 case LTTNG_DOMAIN_PYTHON:
2804 case LTTNG_DOMAIN_UST:
2805 DBG3("Copying tracing session consumer output in UST session");
2806 if (session->ust_session->consumer) {
2807 consumer_output_put(session->ust_session->consumer);
2808 }
2809 session->ust_session->consumer =
2810 consumer_copy_output(session->consumer);
2811 /* Ease our life a bit for the next part */
2812 consumer = session->ust_session->consumer;
2813 dir_name = DEFAULT_UST_TRACE_DIR;
2814 break;
2815 default:
2816 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2817 goto error;
2818 }
2819
2820 /* Append correct directory to subdir */
2821 strncat(consumer->subdir, dir_name,
2822 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2823 DBG3("Copy session consumer subdir %s", consumer->subdir);
2824
2825 ret = LTTNG_OK;
2826
2827 error:
2828 return ret;
2829 }
2830
2831 /*
2832 * Create an UST session and add it to the session ust list.
2833 *
2834 * Should *NOT* be called with RCU read-side lock held.
2835 */
2836 static int create_ust_session(struct ltt_session *session,
2837 struct lttng_domain *domain)
2838 {
2839 int ret;
2840 struct ltt_ust_session *lus = NULL;
2841
2842 assert(session);
2843 assert(domain);
2844 assert(session->consumer);
2845
2846 switch (domain->type) {
2847 case LTTNG_DOMAIN_JUL:
2848 case LTTNG_DOMAIN_LOG4J:
2849 case LTTNG_DOMAIN_PYTHON:
2850 case LTTNG_DOMAIN_UST:
2851 break;
2852 default:
2853 ERR("Unknown UST domain on create session %d", domain->type);
2854 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2855 goto error;
2856 }
2857
2858 DBG("Creating UST session");
2859
2860 lus = trace_ust_create_session(session->id);
2861 if (lus == NULL) {
2862 ret = LTTNG_ERR_UST_SESS_FAIL;
2863 goto error;
2864 }
2865
2866 lus->uid = session->uid;
2867 lus->gid = session->gid;
2868 lus->output_traces = session->output_traces;
2869 lus->snapshot_mode = session->snapshot_mode;
2870 lus->live_timer_interval = session->live_timer;
2871 session->ust_session = lus;
2872 if (session->shm_path[0]) {
2873 strncpy(lus->root_shm_path, session->shm_path,
2874 sizeof(lus->root_shm_path));
2875 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2876 strncpy(lus->shm_path, session->shm_path,
2877 sizeof(lus->shm_path));
2878 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2879 strncat(lus->shm_path, "/ust",
2880 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2881 }
2882 /* Copy session output to the newly created UST session */
2883 ret = copy_session_consumer(domain->type, session);
2884 if (ret != LTTNG_OK) {
2885 goto error;
2886 }
2887
2888 return LTTNG_OK;
2889
2890 error:
2891 free(lus);
2892 session->ust_session = NULL;
2893 return ret;
2894 }
2895
2896 /*
2897 * Create a kernel tracer session then create the default channel.
2898 */
2899 static int create_kernel_session(struct ltt_session *session)
2900 {
2901 int ret;
2902
2903 DBG("Creating kernel session");
2904
2905 ret = kernel_create_session(session, kernel_tracer_fd);
2906 if (ret < 0) {
2907 ret = LTTNG_ERR_KERN_SESS_FAIL;
2908 goto error;
2909 }
2910
2911 /* Code flow safety */
2912 assert(session->kernel_session);
2913
2914 /* Copy session output to the newly created Kernel session */
2915 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2916 if (ret != LTTNG_OK) {
2917 goto error;
2918 }
2919
2920 /* Create directory(ies) on local filesystem. */
2921 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2922 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2923 ret = run_as_mkdir_recursive(
2924 session->kernel_session->consumer->dst.trace_path,
2925 S_IRWXU | S_IRWXG, session->uid, session->gid);
2926 if (ret < 0) {
2927 if (errno != EEXIST) {
2928 ERR("Trace directory creation error");
2929 goto error;
2930 }
2931 }
2932 }
2933
2934 session->kernel_session->uid = session->uid;
2935 session->kernel_session->gid = session->gid;
2936 session->kernel_session->output_traces = session->output_traces;
2937 session->kernel_session->snapshot_mode = session->snapshot_mode;
2938
2939 return LTTNG_OK;
2940
2941 error:
2942 trace_kernel_destroy_session(session->kernel_session);
2943 session->kernel_session = NULL;
2944 return ret;
2945 }
2946
2947 /*
2948 * Count number of session permitted by uid/gid.
2949 */
2950 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2951 {
2952 unsigned int i = 0;
2953 struct ltt_session *session;
2954
2955 DBG("Counting number of available session for UID %d GID %d",
2956 uid, gid);
2957 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2958 /*
2959 * Only list the sessions the user can control.
2960 */
2961 if (!session_access_ok(session, uid, gid)) {
2962 continue;
2963 }
2964 i++;
2965 }
2966 return i;
2967 }
2968
2969 /*
2970 * Process the command requested by the lttng client within the command
2971 * context structure. This function make sure that the return structure (llm)
2972 * is set and ready for transmission before returning.
2973 *
2974 * Return any error encountered or 0 for success.
2975 *
2976 * "sock" is only used for special-case var. len data.
2977 *
2978 * Should *NOT* be called with RCU read-side lock held.
2979 */
2980 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2981 int *sock_error)
2982 {
2983 int ret = LTTNG_OK;
2984 int need_tracing_session = 1;
2985 int need_domain;
2986
2987 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2988
2989 assert(!rcu_read_ongoing());
2990
2991 *sock_error = 0;
2992
2993 switch (cmd_ctx->lsm->cmd_type) {
2994 case LTTNG_CREATE_SESSION:
2995 case LTTNG_CREATE_SESSION_SNAPSHOT:
2996 case LTTNG_CREATE_SESSION_LIVE:
2997 case LTTNG_DESTROY_SESSION:
2998 case LTTNG_LIST_SESSIONS:
2999 case LTTNG_LIST_DOMAINS:
3000 case LTTNG_START_TRACE:
3001 case LTTNG_STOP_TRACE:
3002 case LTTNG_DATA_PENDING:
3003 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3004 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3005 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3006 case LTTNG_SNAPSHOT_RECORD:
3007 case LTTNG_SAVE_SESSION:
3008 case LTTNG_SET_SESSION_SHM_PATH:
3009 need_domain = 0;
3010 break;
3011 default:
3012 need_domain = 1;
3013 }
3014
3015 if (opt_no_kernel && need_domain
3016 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3017 if (!is_root) {
3018 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3019 } else {
3020 ret = LTTNG_ERR_KERN_NA;
3021 }
3022 goto error;
3023 }
3024
3025 /* Deny register consumer if we already have a spawned consumer. */
3026 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3027 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3028 if (kconsumer_data.pid > 0) {
3029 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3030 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3031 goto error;
3032 }
3033 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3034 }
3035
3036 /*
3037 * Check for command that don't needs to allocate a returned payload. We do
3038 * this here so we don't have to make the call for no payload at each
3039 * command.
3040 */
3041 switch(cmd_ctx->lsm->cmd_type) {
3042 case LTTNG_LIST_SESSIONS:
3043 case LTTNG_LIST_TRACEPOINTS:
3044 case LTTNG_LIST_TRACEPOINT_FIELDS:
3045 case LTTNG_LIST_DOMAINS:
3046 case LTTNG_LIST_CHANNELS:
3047 case LTTNG_LIST_EVENTS:
3048 case LTTNG_LIST_SYSCALLS:
3049 case LTTNG_LIST_TRACKER_PIDS:
3050 break;
3051 default:
3052 /* Setup lttng message with no payload */
3053 ret = setup_lttng_msg(cmd_ctx, 0);
3054 if (ret < 0) {
3055 /* This label does not try to unlock the session */
3056 goto init_setup_error;
3057 }
3058 }
3059
3060 /* Commands that DO NOT need a session. */
3061 switch (cmd_ctx->lsm->cmd_type) {
3062 case LTTNG_CREATE_SESSION:
3063 case LTTNG_CREATE_SESSION_SNAPSHOT:
3064 case LTTNG_CREATE_SESSION_LIVE:
3065 case LTTNG_CALIBRATE:
3066 case LTTNG_LIST_SESSIONS:
3067 case LTTNG_LIST_TRACEPOINTS:
3068 case LTTNG_LIST_SYSCALLS:
3069 case LTTNG_LIST_TRACEPOINT_FIELDS:
3070 case LTTNG_SAVE_SESSION:
3071 need_tracing_session = 0;
3072 break;
3073 default:
3074 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3075 /*
3076 * We keep the session list lock across _all_ commands
3077 * for now, because the per-session lock does not
3078 * handle teardown properly.
3079 */
3080 session_lock_list();
3081 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3082 if (cmd_ctx->session == NULL) {
3083 ret = LTTNG_ERR_SESS_NOT_FOUND;
3084 goto error;
3085 } else {
3086 /* Acquire lock for the session */
3087 session_lock(cmd_ctx->session);
3088 }
3089 break;
3090 }
3091
3092 /*
3093 * Commands that need a valid session but should NOT create one if none
3094 * exists. Instead of creating one and destroying it when the command is
3095 * handled, process that right before so we save some round trip in useless
3096 * code path.
3097 */
3098 switch (cmd_ctx->lsm->cmd_type) {
3099 case LTTNG_DISABLE_CHANNEL:
3100 case LTTNG_DISABLE_EVENT:
3101 switch (cmd_ctx->lsm->domain.type) {
3102 case LTTNG_DOMAIN_KERNEL:
3103 if (!cmd_ctx->session->kernel_session) {
3104 ret = LTTNG_ERR_NO_CHANNEL;
3105 goto error;
3106 }
3107 break;
3108 case LTTNG_DOMAIN_JUL:
3109 case LTTNG_DOMAIN_LOG4J:
3110 case LTTNG_DOMAIN_PYTHON:
3111 case LTTNG_DOMAIN_UST:
3112 if (!cmd_ctx->session->ust_session) {
3113 ret = LTTNG_ERR_NO_CHANNEL;
3114 goto error;
3115 }
3116 break;
3117 default:
3118 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3119 goto error;
3120 }
3121 default:
3122 break;
3123 }
3124
3125 if (!need_domain) {
3126 goto skip_domain;
3127 }
3128
3129 /*
3130 * Check domain type for specific "pre-action".
3131 */
3132 switch (cmd_ctx->lsm->domain.type) {
3133 case LTTNG_DOMAIN_KERNEL:
3134 if (!is_root) {
3135 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3136 goto error;
3137 }
3138
3139 /* Kernel tracer check */
3140 if (kernel_tracer_fd == -1) {
3141 /* Basically, load kernel tracer modules */
3142 ret = init_kernel_tracer();
3143 if (ret != 0) {
3144 goto error;
3145 }
3146 }
3147
3148 /* Consumer is in an ERROR state. Report back to client */
3149 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3150 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3151 goto error;
3152 }
3153
3154 /* Need a session for kernel command */
3155 if (need_tracing_session) {
3156 if (cmd_ctx->session->kernel_session == NULL) {
3157 ret = create_kernel_session(cmd_ctx->session);
3158 if (ret < 0) {
3159 ret = LTTNG_ERR_KERN_SESS_FAIL;
3160 goto error;
3161 }
3162 }
3163
3164 /* Start the kernel consumer daemon */
3165 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3166 if (kconsumer_data.pid == 0 &&
3167 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3168 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3169 ret = start_consumerd(&kconsumer_data);
3170 if (ret < 0) {
3171 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3172 goto error;
3173 }
3174 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3175 } else {
3176 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3177 }
3178
3179 /*
3180 * The consumer was just spawned so we need to add the socket to
3181 * the consumer output of the session if exist.
3182 */
3183 ret = consumer_create_socket(&kconsumer_data,
3184 cmd_ctx->session->kernel_session->consumer);
3185 if (ret < 0) {
3186 goto error;
3187 }
3188 }
3189
3190 break;
3191 case LTTNG_DOMAIN_JUL:
3192 case LTTNG_DOMAIN_LOG4J:
3193 case LTTNG_DOMAIN_PYTHON:
3194 case LTTNG_DOMAIN_UST:
3195 {
3196 if (!ust_app_supported()) {
3197 ret = LTTNG_ERR_NO_UST;
3198 goto error;
3199 }
3200 /* Consumer is in an ERROR state. Report back to client */
3201 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3202 ret = LTTNG_ERR_NO_USTCONSUMERD;
3203 goto error;
3204 }
3205
3206 if (need_tracing_session) {
3207 /* Create UST session if none exist. */
3208 if (cmd_ctx->session->ust_session == NULL) {
3209 ret = create_ust_session(cmd_ctx->session,
3210 &cmd_ctx->lsm->domain);
3211 if (ret != LTTNG_OK) {
3212 goto error;
3213 }
3214 }
3215
3216 /* Start the UST consumer daemons */
3217 /* 64-bit */
3218 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3219 if (consumerd64_bin[0] != '\0' &&
3220 ustconsumer64_data.pid == 0 &&
3221 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3222 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3223 ret = start_consumerd(&ustconsumer64_data);
3224 if (ret < 0) {
3225 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3226 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3227 goto error;
3228 }
3229
3230 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3231 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3232 } else {
3233 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3234 }
3235
3236 /*
3237 * Setup socket for consumer 64 bit. No need for atomic access
3238 * since it was set above and can ONLY be set in this thread.
3239 */
3240 ret = consumer_create_socket(&ustconsumer64_data,
3241 cmd_ctx->session->ust_session->consumer);
3242 if (ret < 0) {
3243 goto error;
3244 }
3245
3246 /* 32-bit */
3247 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3248 if (consumerd32_bin[0] != '\0' &&
3249 ustconsumer32_data.pid == 0 &&
3250 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3251 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3252 ret = start_consumerd(&ustconsumer32_data);
3253 if (ret < 0) {
3254 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3255 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3256 goto error;
3257 }
3258
3259 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3260 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3261 } else {
3262 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3263 }
3264
3265 /*
3266 * Setup socket for consumer 64 bit. No need for atomic access
3267 * since it was set above and can ONLY be set in this thread.
3268 */
3269 ret = consumer_create_socket(&ustconsumer32_data,
3270 cmd_ctx->session->ust_session->consumer);
3271 if (ret < 0) {
3272 goto error;
3273 }
3274 }
3275 break;
3276 }
3277 default:
3278 break;
3279 }
3280 skip_domain:
3281
3282 /* Validate consumer daemon state when start/stop trace command */
3283 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3284 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3285 switch (cmd_ctx->lsm->domain.type) {
3286 case LTTNG_DOMAIN_JUL:
3287 case LTTNG_DOMAIN_LOG4J:
3288 case LTTNG_DOMAIN_PYTHON:
3289 case LTTNG_DOMAIN_UST:
3290 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3291 ret = LTTNG_ERR_NO_USTCONSUMERD;
3292 goto error;
3293 }
3294 break;
3295 case LTTNG_DOMAIN_KERNEL:
3296 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3297 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3298 goto error;
3299 }
3300 break;
3301 }
3302 }
3303
3304 /*
3305 * Check that the UID or GID match that of the tracing session.
3306 * The root user can interact with all sessions.
3307 */
3308 if (need_tracing_session) {
3309 if (!session_access_ok(cmd_ctx->session,
3310 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3311 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3312 ret = LTTNG_ERR_EPERM;
3313 goto error;
3314 }
3315 }
3316
3317 /*
3318 * Send relayd information to consumer as soon as we have a domain and a
3319 * session defined.
3320 */
3321 if (cmd_ctx->session && need_domain) {
3322 /*
3323 * Setup relayd if not done yet. If the relayd information was already
3324 * sent to the consumer, this call will gracefully return.
3325 */
3326 ret = cmd_setup_relayd(cmd_ctx->session);
3327 if (ret != LTTNG_OK) {
3328 goto error;
3329 }
3330 }
3331
3332 /* Process by command type */
3333 switch (cmd_ctx->lsm->cmd_type) {
3334 case LTTNG_ADD_CONTEXT:
3335 {
3336 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3337 cmd_ctx->lsm->u.context.channel_name,
3338 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3339 break;
3340 }
3341 case LTTNG_DISABLE_CHANNEL:
3342 {
3343 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3344 cmd_ctx->lsm->u.disable.channel_name);
3345 break;
3346 }
3347 case LTTNG_DISABLE_EVENT:
3348 {
3349
3350 /*
3351 * FIXME: handle filter; for now we just receive the filter's
3352 * bytecode along with the filter expression which are sent by
3353 * liblttng-ctl and discard them.
3354 *
3355 * This fixes an issue where the client may block while sending
3356 * the filter payload and encounter an error because the session
3357 * daemon closes the socket without ever handling this data.
3358 */
3359 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3360 cmd_ctx->lsm->u.disable.bytecode_len;
3361
3362 if (count) {
3363 char data[LTTNG_FILTER_MAX_LEN];
3364
3365 DBG("Discarding disable event command payload of size %zu", count);
3366 while (count) {
3367 ret = lttcomm_recv_unix_sock(sock, data,
3368 count > sizeof(data) ? sizeof(data) : count);
3369 if (ret < 0) {
3370 goto error;
3371 }
3372
3373 count -= (size_t) ret;
3374 }
3375 }
3376 /* FIXME: passing packed structure to non-packed pointer */
3377 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3378 cmd_ctx->lsm->u.disable.channel_name,
3379 &cmd_ctx->lsm->u.disable.event);
3380 break;
3381 }
3382 case LTTNG_ENABLE_CHANNEL:
3383 {
3384 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3385 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3386 break;
3387 }
3388 case LTTNG_TRACK_PID:
3389 {
3390 ret = cmd_track_pid(cmd_ctx->session,
3391 cmd_ctx->lsm->domain.type,
3392 cmd_ctx->lsm->u.pid_tracker.pid);
3393 break;
3394 }
3395 case LTTNG_UNTRACK_PID:
3396 {
3397 ret = cmd_untrack_pid(cmd_ctx->session,
3398 cmd_ctx->lsm->domain.type,
3399 cmd_ctx->lsm->u.pid_tracker.pid);
3400 break;
3401 }
3402 case LTTNG_ENABLE_EVENT:
3403 {
3404 struct lttng_event_exclusion *exclusion = NULL;
3405 struct lttng_filter_bytecode *bytecode = NULL;
3406 char *filter_expression = NULL;
3407
3408 /* Handle exclusion events and receive it from the client. */
3409 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3410 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3411
3412 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3413 (count * LTTNG_SYMBOL_NAME_LEN));
3414 if (!exclusion) {
3415 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3416 goto error;
3417 }
3418
3419 DBG("Receiving var len exclusion event list from client ...");
3420 exclusion->count = count;
3421 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3422 count * LTTNG_SYMBOL_NAME_LEN);
3423 if (ret <= 0) {
3424 DBG("Nothing recv() from client var len data... continuing");
3425 *sock_error = 1;
3426 free(exclusion);
3427 ret = LTTNG_ERR_EXCLUSION_INVAL;
3428 goto error;
3429 }
3430 }
3431
3432 /* Get filter expression from client. */
3433 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3434 size_t expression_len =
3435 cmd_ctx->lsm->u.enable.expression_len;
3436
3437 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3438 ret = LTTNG_ERR_FILTER_INVAL;
3439 free(exclusion);
3440 goto error;
3441 }
3442
3443 filter_expression = zmalloc(expression_len);
3444 if (!filter_expression) {
3445 free(exclusion);
3446 ret = LTTNG_ERR_FILTER_NOMEM;
3447 goto error;
3448 }
3449
3450 /* Receive var. len. data */
3451 DBG("Receiving var len filter's expression from client ...");
3452 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3453 expression_len);
3454 if (ret <= 0) {
3455 DBG("Nothing recv() from client car len data... continuing");
3456 *sock_error = 1;
3457 free(filter_expression);
3458 free(exclusion);
3459 ret = LTTNG_ERR_FILTER_INVAL;
3460 goto error;
3461 }
3462 }
3463
3464 /* Handle filter and get bytecode from client. */
3465 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3466 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3467
3468 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3469 ret = LTTNG_ERR_FILTER_INVAL;
3470 free(filter_expression);
3471 free(exclusion);
3472 goto error;
3473 }
3474
3475 bytecode = zmalloc(bytecode_len);
3476 if (!bytecode) {
3477 free(filter_expression);
3478 free(exclusion);
3479 ret = LTTNG_ERR_FILTER_NOMEM;
3480 goto error;
3481 }
3482
3483 /* Receive var. len. data */
3484 DBG("Receiving var len filter's bytecode from client ...");
3485 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3486 if (ret <= 0) {
3487 DBG("Nothing recv() from client car len data... continuing");
3488 *sock_error = 1;
3489 free(filter_expression);
3490 free(bytecode);
3491 free(exclusion);
3492 ret = LTTNG_ERR_FILTER_INVAL;
3493 goto error;
3494 }
3495
3496 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3497 free(filter_expression);
3498 free(bytecode);
3499 free(exclusion);
3500 ret = LTTNG_ERR_FILTER_INVAL;
3501 goto error;
3502 }
3503 }
3504
3505 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3506 cmd_ctx->lsm->u.enable.channel_name,
3507 &cmd_ctx->lsm->u.enable.event,
3508 filter_expression, bytecode, exclusion,
3509 kernel_poll_pipe[1]);
3510 break;
3511 }
3512 case LTTNG_LIST_TRACEPOINTS:
3513 {
3514 struct lttng_event *events;
3515 ssize_t nb_events;
3516
3517 session_lock_list();
3518 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3519 session_unlock_list();
3520 if (nb_events < 0) {
3521 /* Return value is a negative lttng_error_code. */
3522 ret = -nb_events;
3523 goto error;
3524 }
3525
3526 /*
3527 * Setup lttng message with payload size set to the event list size in
3528 * bytes and then copy list into the llm payload.
3529 */
3530 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3531 if (ret < 0) {
3532 free(events);
3533 goto setup_error;
3534 }
3535
3536 /* Copy event list into message payload */
3537 memcpy(cmd_ctx->llm->payload, events,
3538 sizeof(struct lttng_event) * nb_events);
3539
3540 free(events);
3541
3542 ret = LTTNG_OK;
3543 break;
3544 }
3545 case LTTNG_LIST_TRACEPOINT_FIELDS:
3546 {
3547 struct lttng_event_field *fields;
3548 ssize_t nb_fields;
3549
3550 session_lock_list();
3551 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3552 &fields);
3553 session_unlock_list();
3554 if (nb_fields < 0) {
3555 /* Return value is a negative lttng_error_code. */
3556 ret = -nb_fields;
3557 goto error;
3558 }
3559
3560 /*
3561 * Setup lttng message with payload size set to the event list size in
3562 * bytes and then copy list into the llm payload.
3563 */
3564 ret = setup_lttng_msg(cmd_ctx,
3565 sizeof(struct lttng_event_field) * nb_fields);
3566 if (ret < 0) {
3567 free(fields);
3568 goto setup_error;
3569 }
3570
3571 /* Copy event list into message payload */
3572 memcpy(cmd_ctx->llm->payload, fields,
3573 sizeof(struct lttng_event_field) * nb_fields);
3574
3575 free(fields);
3576
3577 ret = LTTNG_OK;
3578 break;
3579 }
3580 case LTTNG_LIST_SYSCALLS:
3581 {
3582 struct lttng_event *events;
3583 ssize_t nb_events;
3584
3585 nb_events = cmd_list_syscalls(&events);
3586 if (nb_events < 0) {
3587 /* Return value is a negative lttng_error_code. */
3588 ret = -nb_events;
3589 goto error;
3590 }
3591
3592 /*
3593 * Setup lttng message with payload size set to the event list size in
3594 * bytes and then copy list into the llm payload.
3595 */
3596 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3597 if (ret < 0) {
3598 free(events);
3599 goto setup_error;
3600 }
3601
3602 /* Copy event list into message payload */
3603 memcpy(cmd_ctx->llm->payload, events,
3604 sizeof(struct lttng_event) * nb_events);
3605
3606 free(events);
3607
3608 ret = LTTNG_OK;
3609 break;
3610 }
3611 case LTTNG_LIST_TRACKER_PIDS:
3612 {
3613 int32_t *pids = NULL;
3614 ssize_t nr_pids;
3615
3616 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3617 cmd_ctx->lsm->domain.type, &pids);
3618 if (nr_pids < 0) {
3619 /* Return value is a negative lttng_error_code. */
3620 ret = -nr_pids;
3621 goto error;
3622 }
3623
3624 /*
3625 * Setup lttng message with payload size set to the event list size in
3626 * bytes and then copy list into the llm payload.
3627 */
3628 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3629 if (ret < 0) {
3630 free(pids);
3631 goto setup_error;
3632 }
3633
3634 /* Copy event list into message payload */
3635 memcpy(cmd_ctx->llm->payload, pids,
3636 sizeof(int) * nr_pids);
3637
3638 free(pids);
3639
3640 ret = LTTNG_OK;
3641 break;
3642 }
3643 case LTTNG_SET_CONSUMER_URI:
3644 {
3645 size_t nb_uri, len;
3646 struct lttng_uri *uris;
3647
3648 nb_uri = cmd_ctx->lsm->u.uri.size;
3649 len = nb_uri * sizeof(struct lttng_uri);
3650
3651 if (nb_uri == 0) {
3652 ret = LTTNG_ERR_INVALID;
3653 goto error;
3654 }
3655
3656 uris = zmalloc(len);
3657 if (uris == NULL) {
3658 ret = LTTNG_ERR_FATAL;
3659 goto error;
3660 }
3661
3662 /* Receive variable len data */
3663 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3664 ret = lttcomm_recv_unix_sock(sock, uris, len);
3665 if (ret <= 0) {
3666 DBG("No URIs received from client... continuing");
3667 *sock_error = 1;
3668 ret = LTTNG_ERR_SESSION_FAIL;
3669 free(uris);
3670 goto error;
3671 }
3672
3673 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3674 free(uris);
3675 if (ret != LTTNG_OK) {
3676 goto error;
3677 }
3678
3679
3680 break;
3681 }
3682 case LTTNG_START_TRACE:
3683 {
3684 ret = cmd_start_trace(cmd_ctx->session);
3685 break;
3686 }
3687 case LTTNG_STOP_TRACE:
3688 {
3689 ret = cmd_stop_trace(cmd_ctx->session);
3690 break;
3691 }
3692 case LTTNG_CREATE_SESSION:
3693 {
3694 size_t nb_uri, len;
3695 struct lttng_uri *uris = NULL;
3696
3697 nb_uri = cmd_ctx->lsm->u.uri.size;
3698 len = nb_uri * sizeof(struct lttng_uri);
3699
3700 if (nb_uri > 0) {
3701 uris = zmalloc(len);
3702 if (uris == NULL) {
3703 ret = LTTNG_ERR_FATAL;
3704 goto error;
3705 }
3706
3707 /* Receive variable len data */
3708 DBG("Waiting for %zu URIs from client ...", nb_uri);
3709 ret = lttcomm_recv_unix_sock(sock, uris, len);
3710 if (ret <= 0) {
3711 DBG("No URIs received from client... continuing");
3712 *sock_error = 1;
3713 ret = LTTNG_ERR_SESSION_FAIL;
3714 free(uris);
3715 goto error;
3716 }
3717
3718 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3719 DBG("Creating session with ONE network URI is a bad call");
3720 ret = LTTNG_ERR_SESSION_FAIL;
3721 free(uris);
3722 goto error;
3723 }
3724 }
3725
3726 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3727 &cmd_ctx->creds, 0);
3728
3729 free(uris);
3730
3731 break;
3732 }
3733 case LTTNG_DESTROY_SESSION:
3734 {
3735 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3736
3737 /* Set session to NULL so we do not unlock it after free. */
3738 cmd_ctx->session = NULL;
3739 break;
3740 }
3741 case LTTNG_LIST_DOMAINS:
3742 {
3743 ssize_t nb_dom;
3744 struct lttng_domain *domains = NULL;
3745
3746 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3747 if (nb_dom < 0) {
3748 /* Return value is a negative lttng_error_code. */
3749 ret = -nb_dom;
3750 goto error;
3751 }
3752
3753 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3754 if (ret < 0) {
3755 free(domains);
3756 goto setup_error;
3757 }
3758
3759 /* Copy event list into message payload */
3760 memcpy(cmd_ctx->llm->payload, domains,
3761 nb_dom * sizeof(struct lttng_domain));
3762
3763 free(domains);
3764
3765 ret = LTTNG_OK;
3766 break;
3767 }
3768 case LTTNG_LIST_CHANNELS:
3769 {
3770 int nb_chan;
3771 struct lttng_channel *channels = NULL;
3772
3773 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3774 cmd_ctx->session, &channels);
3775 if (nb_chan < 0) {
3776 /* Return value is a negative lttng_error_code. */
3777 ret = -nb_chan;
3778 goto error;
3779 }
3780
3781 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3782 if (ret < 0) {
3783 free(channels);
3784 goto setup_error;
3785 }
3786
3787 /* Copy event list into message payload */
3788 memcpy(cmd_ctx->llm->payload, channels,
3789 nb_chan * sizeof(struct lttng_channel));
3790
3791 free(channels);
3792
3793 ret = LTTNG_OK;
3794 break;
3795 }
3796 case LTTNG_LIST_EVENTS:
3797 {
3798 ssize_t nb_event;
3799 struct lttng_event *events = NULL;
3800
3801 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3802 cmd_ctx->lsm->u.list.channel_name, &events);
3803 if (nb_event < 0) {
3804 /* Return value is a negative lttng_error_code. */
3805 ret = -nb_event;
3806 goto error;
3807 }
3808
3809 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3810 if (ret < 0) {
3811 free(events);
3812 goto setup_error;
3813 }
3814
3815 /* Copy event list into message payload */
3816 memcpy(cmd_ctx->llm->payload, events,
3817 nb_event * sizeof(struct lttng_event));
3818
3819 free(events);
3820
3821 ret = LTTNG_OK;
3822 break;
3823 }
3824 case LTTNG_LIST_SESSIONS:
3825 {
3826 unsigned int nr_sessions;
3827
3828 session_lock_list();
3829 nr_sessions = lttng_sessions_count(
3830 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3831 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3832
3833 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3834 if (ret < 0) {
3835 session_unlock_list();
3836 goto setup_error;
3837 }
3838
3839 /* Filled the session array */
3840 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3841 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3842 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3843
3844 session_unlock_list();
3845
3846 ret = LTTNG_OK;
3847 break;
3848 }
3849 case LTTNG_CALIBRATE:
3850 {
3851 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3852 &cmd_ctx->lsm->u.calibrate);
3853 break;
3854 }
3855 case LTTNG_REGISTER_CONSUMER:
3856 {
3857 struct consumer_data *cdata;
3858
3859 switch (cmd_ctx->lsm->domain.type) {
3860 case LTTNG_DOMAIN_KERNEL:
3861 cdata = &kconsumer_data;
3862 break;
3863 default:
3864 ret = LTTNG_ERR_UND;
3865 goto error;
3866 }
3867
3868 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3869 cmd_ctx->lsm->u.reg.path, cdata);
3870 break;
3871 }
3872 case LTTNG_DATA_PENDING:
3873 {
3874 ret = cmd_data_pending(cmd_ctx->session);
3875 break;
3876 }
3877 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3878 {
3879 struct lttcomm_lttng_output_id reply;
3880
3881 ret = cmd_snapshot_add_output(cmd_ctx->session,
3882 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3883 if (ret != LTTNG_OK) {
3884 goto error;
3885 }
3886
3887 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3888 if (ret < 0) {
3889 goto setup_error;
3890 }
3891
3892 /* Copy output list into message payload */
3893 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3894 ret = LTTNG_OK;
3895 break;
3896 }
3897 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3898 {
3899 ret = cmd_snapshot_del_output(cmd_ctx->session,
3900 &cmd_ctx->lsm->u.snapshot_output.output);
3901 break;
3902 }
3903 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3904 {
3905 ssize_t nb_output;
3906 struct lttng_snapshot_output *outputs = NULL;
3907
3908 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3909 if (nb_output < 0) {
3910 ret = -nb_output;
3911 goto error;
3912 }
3913
3914 ret = setup_lttng_msg(cmd_ctx,
3915 nb_output * sizeof(struct lttng_snapshot_output));
3916 if (ret < 0) {
3917 free(outputs);
3918 goto setup_error;
3919 }
3920
3921 if (outputs) {
3922 /* Copy output list into message payload */
3923 memcpy(cmd_ctx->llm->payload, outputs,
3924 nb_output * sizeof(struct lttng_snapshot_output));
3925 free(outputs);
3926 }
3927
3928 ret = LTTNG_OK;
3929 break;
3930 }
3931 case LTTNG_SNAPSHOT_RECORD:
3932 {
3933 ret = cmd_snapshot_record(cmd_ctx->session,
3934 &cmd_ctx->lsm->u.snapshot_record.output,
3935 cmd_ctx->lsm->u.snapshot_record.wait);
3936 break;
3937 }
3938 case LTTNG_CREATE_SESSION_SNAPSHOT:
3939 {
3940 size_t nb_uri, len;
3941 struct lttng_uri *uris = NULL;
3942
3943 nb_uri = cmd_ctx->lsm->u.uri.size;
3944 len = nb_uri * sizeof(struct lttng_uri);
3945
3946 if (nb_uri > 0) {
3947 uris = zmalloc(len);
3948 if (uris == NULL) {
3949 ret = LTTNG_ERR_FATAL;
3950 goto error;
3951 }
3952
3953 /* Receive variable len data */
3954 DBG("Waiting for %zu URIs from client ...", nb_uri);
3955 ret = lttcomm_recv_unix_sock(sock, uris, len);
3956 if (ret <= 0) {
3957 DBG("No URIs received from client... continuing");
3958 *sock_error = 1;
3959 ret = LTTNG_ERR_SESSION_FAIL;
3960 free(uris);
3961 goto error;
3962 }
3963
3964 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3965 DBG("Creating session with ONE network URI is a bad call");
3966 ret = LTTNG_ERR_SESSION_FAIL;
3967 free(uris);
3968 goto error;
3969 }
3970 }
3971
3972 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3973 nb_uri, &cmd_ctx->creds);
3974 free(uris);
3975 break;
3976 }
3977 case LTTNG_CREATE_SESSION_LIVE:
3978 {
3979 size_t nb_uri, len;
3980 struct lttng_uri *uris = NULL;
3981
3982 nb_uri = cmd_ctx->lsm->u.uri.size;
3983 len = nb_uri * sizeof(struct lttng_uri);
3984
3985 if (nb_uri > 0) {
3986 uris = zmalloc(len);
3987 if (uris == NULL) {
3988 ret = LTTNG_ERR_FATAL;
3989 goto error;
3990 }
3991
3992 /* Receive variable len data */
3993 DBG("Waiting for %zu URIs from client ...", nb_uri);
3994 ret = lttcomm_recv_unix_sock(sock, uris, len);
3995 if (ret <= 0) {
3996 DBG("No URIs received from client... continuing");
3997 *sock_error = 1;
3998 ret = LTTNG_ERR_SESSION_FAIL;
3999 free(uris);
4000 goto error;
4001 }
4002
4003 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4004 DBG("Creating session with ONE network URI is a bad call");
4005 ret = LTTNG_ERR_SESSION_FAIL;
4006 free(uris);
4007 goto error;
4008 }
4009 }
4010
4011 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4012 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4013 free(uris);
4014 break;
4015 }
4016 case LTTNG_SAVE_SESSION:
4017 {
4018 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4019 &cmd_ctx->creds);
4020 break;
4021 }
4022 case LTTNG_SET_SESSION_SHM_PATH:
4023 {
4024 ret = cmd_set_session_shm_path(cmd_ctx->session,
4025 cmd_ctx->lsm->u.set_shm_path.shm_path);
4026 break;
4027 }
4028 default:
4029 ret = LTTNG_ERR_UND;
4030 break;
4031 }
4032
4033 error:
4034 if (cmd_ctx->llm == NULL) {
4035 DBG("Missing llm structure. Allocating one.");
4036 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
4037 goto setup_error;
4038 }
4039 }
4040 /* Set return code */
4041 cmd_ctx->llm->ret_code = ret;
4042 setup_error:
4043 if (cmd_ctx->session) {
4044 session_unlock(cmd_ctx->session);
4045 }
4046 if (need_tracing_session) {
4047 session_unlock_list();
4048 }
4049 init_setup_error:
4050 assert(!rcu_read_ongoing());
4051 return ret;
4052 }
4053
4054 /*
4055 * Thread managing health check socket.
4056 */
4057 static void *thread_manage_health(void *data)
4058 {
4059 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4060 uint32_t revents, nb_fd;
4061 struct lttng_poll_event events;
4062 struct health_comm_msg msg;
4063 struct health_comm_reply reply;
4064
4065 DBG("[thread] Manage health check started");
4066
4067 rcu_register_thread();
4068
4069 /* We might hit an error path before this is created. */
4070 lttng_poll_init(&events);
4071
4072 /* Create unix socket */
4073 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4074 if (sock < 0) {
4075 ERR("Unable to create health check Unix socket");
4076 ret = -1;
4077 goto error;
4078 }
4079
4080 if (is_root) {
4081 /* lttng health client socket path permissions */
4082 ret = chown(health_unix_sock_path, 0,
4083 utils_get_group_id(tracing_group_name));
4084 if (ret < 0) {
4085 ERR("Unable to set group on %s", health_unix_sock_path);
4086 PERROR("chown");
4087 ret = -1;
4088 goto error;
4089 }
4090
4091 ret = chmod(health_unix_sock_path,
4092 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4093 if (ret < 0) {
4094 ERR("Unable to set permissions on %s", health_unix_sock_path);
4095 PERROR("chmod");
4096 ret = -1;
4097 goto error;
4098 }
4099 }
4100
4101 /*
4102 * Set the CLOEXEC flag. Return code is useless because either way, the
4103 * show must go on.
4104 */
4105 (void) utils_set_fd_cloexec(sock);
4106
4107 ret = lttcomm_listen_unix_sock(sock);
4108 if (ret < 0) {
4109 goto error;
4110 }
4111
4112 /*
4113 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4114 * more will be added to this poll set.
4115 */
4116 ret = sessiond_set_thread_pollset(&events, 2);
4117 if (ret < 0) {
4118 goto error;
4119 }
4120
4121 /* Add the application registration socket */
4122 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4123 if (ret < 0) {
4124 goto error;
4125 }
4126
4127 sessiond_notify_ready();
4128
4129 while (1) {
4130 DBG("Health check ready");
4131
4132 /* Inifinite blocking call, waiting for transmission */
4133 restart:
4134 ret = lttng_poll_wait(&events, -1);
4135 if (ret < 0) {
4136 /*
4137 * Restart interrupted system call.
4138 */
4139 if (errno == EINTR) {
4140 goto restart;
4141 }
4142 goto error;
4143 }
4144
4145 nb_fd = ret;
4146
4147 for (i = 0; i < nb_fd; i++) {
4148 /* Fetch once the poll data */
4149 revents = LTTNG_POLL_GETEV(&events, i);
4150 pollfd = LTTNG_POLL_GETFD(&events, i);
4151
4152 if (!revents) {
4153 /* No activity for this FD (poll implementation). */
4154 continue;
4155 }
4156
4157 /* Thread quit pipe has been closed. Killing thread. */
4158 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4159 if (ret) {
4160 err = 0;
4161 goto exit;
4162 }
4163
4164 /* Event on the registration socket */
4165 if (pollfd == sock) {
4166 if (revents & LPOLLIN) {
4167 continue;
4168 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4169 ERR("Health socket poll error");
4170 goto error;
4171 } else {
4172 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4173 goto error;
4174 }
4175 }
4176 }
4177
4178 new_sock = lttcomm_accept_unix_sock(sock);
4179 if (new_sock < 0) {
4180 goto error;
4181 }
4182
4183 /*
4184 * Set the CLOEXEC flag. Return code is useless because either way, the
4185 * show must go on.
4186 */
4187 (void) utils_set_fd_cloexec(new_sock);
4188
4189 DBG("Receiving data from client for health...");
4190 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4191 if (ret <= 0) {
4192 DBG("Nothing recv() from client... continuing");
4193 ret = close(new_sock);
4194 if (ret) {
4195 PERROR("close");
4196 }
4197 new_sock = -1;
4198 continue;
4199 }
4200
4201 rcu_thread_online();
4202
4203 memset(&reply, 0, sizeof(reply));
4204 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4205 /*
4206 * health_check_state returns 0 if health is
4207 * bad.
4208 */
4209 if (!health_check_state(health_sessiond, i)) {
4210 reply.ret_code |= 1ULL << i;
4211 }
4212 }
4213
4214 DBG2("Health check return value %" PRIx64, reply.ret_code);
4215
4216 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4217 if (ret < 0) {
4218 ERR("Failed to send health data back to client");
4219 }
4220
4221 /* End of transmission */
4222 ret = close(new_sock);
4223 if (ret) {
4224 PERROR("close");
4225 }
4226 new_sock = -1;
4227 }
4228
4229 exit:
4230 error:
4231 if (err) {
4232 ERR("Health error occurred in %s", __func__);
4233 }
4234 DBG("Health check thread dying");
4235 unlink(health_unix_sock_path);
4236 if (sock >= 0) {
4237 ret = close(sock);
4238 if (ret) {
4239 PERROR("close");
4240 }
4241 }
4242
4243 lttng_poll_clean(&events);
4244
4245 rcu_unregister_thread();
4246 return NULL;
4247 }
4248
4249 /*
4250 * This thread manage all clients request using the unix client socket for
4251 * communication.
4252 */
4253 static void *thread_manage_clients(void *data)
4254 {
4255 int sock = -1, ret, i, pollfd, err = -1;
4256 int sock_error;
4257 uint32_t revents, nb_fd;
4258 struct command_ctx *cmd_ctx = NULL;
4259 struct lttng_poll_event events;
4260
4261 DBG("[thread] Manage client started");
4262
4263 rcu_register_thread();
4264
4265 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4266
4267 health_code_update();
4268
4269 ret = lttcomm_listen_unix_sock(client_sock);
4270 if (ret < 0) {
4271 goto error_listen;
4272 }
4273
4274 /*
4275 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4276 * more will be added to this poll set.
4277 */
4278 ret = sessiond_set_thread_pollset(&events, 2);
4279 if (ret < 0) {
4280 goto error_create_poll;
4281 }
4282
4283 /* Add the application registration socket */
4284 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4285 if (ret < 0) {
4286 goto error;
4287 }
4288
4289 sessiond_notify_ready();
4290 ret = sem_post(&load_info->message_thread_ready);
4291 if (ret) {
4292 PERROR("sem_post message_thread_ready");
4293 goto error;
4294 }
4295
4296 /* This testpoint is after we signal readiness to the parent. */
4297 if (testpoint(sessiond_thread_manage_clients)) {
4298 goto error;
4299 }
4300
4301 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4302 goto error;
4303 }
4304
4305 health_code_update();
4306
4307 while (1) {
4308 DBG("Accepting client command ...");
4309
4310 /* Inifinite blocking call, waiting for transmission */
4311 restart:
4312 health_poll_entry();
4313 ret = lttng_poll_wait(&events, -1);
4314 health_poll_exit();
4315 if (ret < 0) {
4316 /*
4317 * Restart interrupted system call.
4318 */
4319 if (errno == EINTR) {
4320 goto restart;
4321 }
4322 goto error;
4323 }
4324
4325 nb_fd = ret;
4326
4327 for (i = 0; i < nb_fd; i++) {
4328 /* Fetch once the poll data */
4329 revents = LTTNG_POLL_GETEV(&events, i);
4330 pollfd = LTTNG_POLL_GETFD(&events, i);
4331
4332 health_code_update();
4333
4334 if (!revents) {
4335 /* No activity for this FD (poll implementation). */
4336 continue;
4337 }
4338
4339 /* Thread quit pipe has been closed. Killing thread. */
4340 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4341 if (ret) {
4342 err = 0;
4343 goto exit;
4344 }
4345
4346 /* Event on the registration socket */
4347 if (pollfd == client_sock) {
4348 if (revents & LPOLLIN) {
4349 continue;
4350 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4351 ERR("Client socket poll error");
4352 goto error;
4353 } else {
4354 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4355 goto error;
4356 }
4357 }
4358 }
4359
4360 DBG("Wait for client response");
4361
4362 health_code_update();
4363
4364 sock = lttcomm_accept_unix_sock(client_sock);
4365 if (sock < 0) {
4366 goto error;
4367 }
4368
4369 /*
4370 * Set the CLOEXEC flag. Return code is useless because either way, the
4371 * show must go on.
4372 */
4373 (void) utils_set_fd_cloexec(sock);
4374
4375 /* Set socket option for credentials retrieval */
4376 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4377 if (ret < 0) {
4378 goto error;
4379 }
4380
4381 /* Allocate context command to process the client request */
4382 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4383 if (cmd_ctx == NULL) {
4384 PERROR("zmalloc cmd_ctx");
4385 goto error;
4386 }
4387
4388 /* Allocate data buffer for reception */
4389 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4390 if (cmd_ctx->lsm == NULL) {
4391 PERROR("zmalloc cmd_ctx->lsm");
4392 goto error;
4393 }
4394
4395 cmd_ctx->llm = NULL;
4396 cmd_ctx->session = NULL;
4397
4398 health_code_update();
4399
4400 /*
4401 * Data is received from the lttng client. The struct
4402 * lttcomm_session_msg (lsm) contains the command and data request of
4403 * the client.
4404 */
4405 DBG("Receiving data from client ...");
4406 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4407 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4408 if (ret <= 0) {
4409 DBG("Nothing recv() from client... continuing");
4410 ret = close(sock);
4411 if (ret) {
4412 PERROR("close");
4413 }
4414 sock = -1;
4415 clean_command_ctx(&cmd_ctx);
4416 continue;
4417 }
4418
4419 health_code_update();
4420
4421 // TODO: Validate cmd_ctx including sanity check for
4422 // security purpose.
4423
4424 rcu_thread_online();
4425 /*
4426 * This function dispatch the work to the kernel or userspace tracer
4427 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4428 * informations for the client. The command context struct contains
4429 * everything this function may needs.
4430 */
4431 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4432 rcu_thread_offline();
4433 if (ret < 0) {
4434 ret = close(sock);
4435 if (ret) {
4436 PERROR("close");
4437 }
4438 sock = -1;
4439 /*
4440 * TODO: Inform client somehow of the fatal error. At
4441 * this point, ret < 0 means that a zmalloc failed
4442 * (ENOMEM). Error detected but still accept
4443 * command, unless a socket error has been
4444 * detected.
4445 */
4446 clean_command_ctx(&cmd_ctx);
4447 continue;
4448 }
4449
4450 health_code_update();
4451
4452 DBG("Sending response (size: %d, retcode: %s)",
4453 cmd_ctx->lttng_msg_size,
4454 lttng_strerror(-cmd_ctx->llm->ret_code));
4455 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4456 if (ret < 0) {
4457 ERR("Failed to send data back to client");
4458 }
4459
4460 /* End of transmission */
4461 ret = close(sock);
4462 if (ret) {
4463 PERROR("close");
4464 }
4465 sock = -1;
4466
4467 clean_command_ctx(&cmd_ctx);
4468
4469 health_code_update();
4470 }
4471
4472 exit:
4473 error:
4474 if (sock >= 0) {
4475 ret = close(sock);
4476 if (ret) {
4477 PERROR("close");
4478 }
4479 }
4480
4481 lttng_poll_clean(&events);
4482 clean_command_ctx(&cmd_ctx);
4483
4484 error_listen:
4485 error_create_poll:
4486 unlink(client_unix_sock_path);
4487 if (client_sock >= 0) {
4488 ret = close(client_sock);
4489 if (ret) {
4490 PERROR("close");
4491 }
4492 }
4493
4494 if (err) {
4495 health_error();
4496 ERR("Health error occurred in %s", __func__);
4497 }
4498
4499 health_unregister(health_sessiond);
4500
4501 DBG("Client thread dying");
4502
4503 rcu_unregister_thread();
4504
4505 /*
4506 * Since we are creating the consumer threads, we own them, so we need
4507 * to join them before our thread exits.
4508 */
4509 ret = join_consumer_thread(&kconsumer_data);
4510 if (ret) {
4511 errno = ret;
4512 PERROR("join_consumer");
4513 }
4514
4515 ret = join_consumer_thread(&ustconsumer32_data);
4516 if (ret) {
4517 errno = ret;
4518 PERROR("join_consumer ust32");
4519 }
4520
4521 ret = join_consumer_thread(&ustconsumer64_data);
4522 if (ret) {
4523 errno = ret;
4524 PERROR("join_consumer ust64");
4525 }
4526 return NULL;
4527 }
4528
4529
4530 /*
4531 * usage function on stderr
4532 */
4533 static void usage(void)
4534 {
4535 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4536 fprintf(stderr, " -h, --help Display this usage.\n");
4537 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4538 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4539 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4540 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4541 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4542 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4543 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4544 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4545 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4546 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4547 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4548 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4549 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4550 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4551 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4552 fprintf(stderr, " -V, --version Show version number.\n");
4553 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4554 fprintf(stderr, " -q, --quiet No output at all.\n");
4555 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4556 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4557 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4558 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4559 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4560 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4561 fprintf(stderr, " -l --load PATH Load session configuration\n");
4562 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4563 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4564 }
4565
4566 static int string_match(const char *str1, const char *str2)
4567 {
4568 return (str1 && str2) && !strcmp(str1, str2);
4569 }
4570
4571 /*
4572 * Take an option from the getopt output and set it in the right variable to be
4573 * used later.
4574 *
4575 * Return 0 on success else a negative value.
4576 */
4577 static int set_option(int opt, const char *arg, const char *optname)
4578 {
4579 int ret = 0;
4580
4581 if (arg && arg[0] == '\0') {
4582 /*
4583 * This only happens if the value is read from daemon config
4584 * file. This means the option requires an argument and the
4585 * configuration file contains a line such as:
4586 * my_option =
4587 */
4588 ret = -EINVAL;
4589 goto end;
4590 }
4591
4592 if (string_match(optname, "client-sock") || opt == 'c') {
4593 if (lttng_is_setuid_setgid()) {
4594 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4595 "-c, --client-sock");
4596 } else {
4597 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4598 }
4599 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4600 if (lttng_is_setuid_setgid()) {
4601 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4602 "-a, --apps-sock");
4603 } else {
4604 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4605 }
4606 } else if (string_match(optname, "daemonize") || opt == 'd') {
4607 opt_daemon = 1;
4608 } else if (string_match(optname, "background") || opt == 'b') {
4609 opt_background = 1;
4610 } else if (string_match(optname, "group") || opt == 'g') {
4611 if (lttng_is_setuid_setgid()) {
4612 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4613 "-g, --group");
4614 } else {
4615 /*
4616 * If the override option is set, the pointer points to a
4617 * *non* const thus freeing it even though the variable type is
4618 * set to const.
4619 */
4620 if (tracing_group_name_override) {
4621 free((void *) tracing_group_name);
4622 }
4623 tracing_group_name = strdup(arg);
4624 if (!tracing_group_name) {
4625 PERROR("strdup");
4626 ret = -ENOMEM;
4627 }
4628 tracing_group_name_override = 1;
4629 }
4630 } else if (string_match(optname, "help") || opt == 'h') {
4631 usage();
4632 exit(EXIT_SUCCESS);
4633 } else if (string_match(optname, "version") || opt == 'V') {
4634 fprintf(stdout, "%s\n", VERSION);
4635 exit(EXIT_SUCCESS);
4636 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4637 opt_sig_parent = 1;
4638 } else if (string_match(optname, "kconsumerd-err-sock")) {
4639 if (lttng_is_setuid_setgid()) {
4640 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4641 "--kconsumerd-err-sock");
4642 } else {
4643 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4644 }
4645 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4646 if (lttng_is_setuid_setgid()) {
4647 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4648 "--kconsumerd-cmd-sock");
4649 } else {
4650 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4651 }
4652 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4653 if (lttng_is_setuid_setgid()) {
4654 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4655 "--ustconsumerd64-err-sock");
4656 } else {
4657 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4658 }
4659 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4660 if (lttng_is_setuid_setgid()) {
4661 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4662 "--ustconsumerd64-cmd-sock");
4663 } else {
4664 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4665 }
4666 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4667 if (lttng_is_setuid_setgid()) {
4668 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4669 "--ustconsumerd32-err-sock");
4670 } else {
4671 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4672 }
4673 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4674 if (lttng_is_setuid_setgid()) {
4675 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4676 "--ustconsumerd32-cmd-sock");
4677 } else {
4678 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4679 }
4680 } else if (string_match(optname, "no-kernel")) {
4681 opt_no_kernel = 1;
4682 } else if (string_match(optname, "quiet") || opt == 'q') {
4683 lttng_opt_quiet = 1;
4684 } else if (string_match(optname, "verbose") || opt == 'v') {
4685 /* Verbose level can increase using multiple -v */
4686 if (arg) {
4687 /* Value obtained from config file */
4688 lttng_opt_verbose = config_parse_value(arg);
4689 } else {
4690 /* -v used on command line */
4691 lttng_opt_verbose++;
4692 }
4693 /* Clamp value to [0, 3] */
4694 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4695 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4696 } else if (string_match(optname, "verbose-consumer")) {
4697 if (arg) {
4698 opt_verbose_consumer = config_parse_value(arg);
4699 } else {
4700 opt_verbose_consumer += 1;
4701 }
4702 } else if (string_match(optname, "consumerd32-path")) {
4703 if (lttng_is_setuid_setgid()) {
4704 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4705 "--consumerd32-path");
4706 } else {
4707 if (consumerd32_bin_override) {
4708 free((void *) consumerd32_bin);
4709 }
4710 consumerd32_bin = strdup(arg);
4711 if (!consumerd32_bin) {
4712 PERROR("strdup");
4713 ret = -ENOMEM;
4714 }
4715 consumerd32_bin_override = 1;
4716 }
4717 } else if (string_match(optname, "consumerd32-libdir")) {
4718 if (lttng_is_setuid_setgid()) {
4719 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4720 "--consumerd32-libdir");
4721 } else {
4722 if (consumerd32_libdir_override) {
4723 free((void *) consumerd32_libdir);
4724 }
4725 consumerd32_libdir = strdup(arg);
4726 if (!consumerd32_libdir) {
4727 PERROR("strdup");
4728 ret = -ENOMEM;
4729 }
4730 consumerd32_libdir_override = 1;
4731 }
4732 } else if (string_match(optname, "consumerd64-path")) {
4733 if (lttng_is_setuid_setgid()) {
4734 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4735 "--consumerd64-path");
4736 } else {
4737 if (consumerd64_bin_override) {
4738 free((void *) consumerd64_bin);
4739 }
4740 consumerd64_bin = strdup(arg);
4741 if (!consumerd64_bin) {
4742 PERROR("strdup");
4743 ret = -ENOMEM;
4744 }
4745 consumerd64_bin_override = 1;
4746 }
4747 } else if (string_match(optname, "consumerd64-libdir")) {
4748 if (lttng_is_setuid_setgid()) {
4749 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4750 "--consumerd64-libdir");
4751 } else {
4752 if (consumerd64_libdir_override) {
4753 free((void *) consumerd64_libdir);
4754 }
4755 consumerd64_libdir = strdup(arg);
4756 if (!consumerd64_libdir) {
4757 PERROR("strdup");
4758 ret = -ENOMEM;
4759 }
4760 consumerd64_libdir_override = 1;
4761 }
4762 } else if (string_match(optname, "pidfile") || opt == 'p') {
4763 if (lttng_is_setuid_setgid()) {
4764 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4765 "-p, --pidfile");
4766 } else {
4767 free(opt_pidfile);
4768 opt_pidfile = strdup(arg);
4769 if (!opt_pidfile) {
4770 PERROR("strdup");
4771 ret = -ENOMEM;
4772 }
4773 }
4774 } else if (string_match(optname, "agent-tcp-port")) {
4775 if (lttng_is_setuid_setgid()) {
4776 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4777 "--agent-tcp-port");
4778 } else {
4779 unsigned long v;
4780
4781 if (!arg) {
4782 ret = -EINVAL;
4783 goto end;
4784 }
4785 errno = 0;
4786 v = strtoul(arg, NULL, 0);
4787 if (errno != 0 || !isdigit(arg[0])) {
4788 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4789 return -1;
4790 }
4791 if (v == 0 || v >= 65535) {
4792 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4793 return -1;
4794 }
4795 agent_tcp_port = (uint32_t) v;
4796 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4797 }
4798 } else if (string_match(optname, "load") || opt == 'l') {
4799 if (lttng_is_setuid_setgid()) {
4800 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4801 "-l, --load");
4802 } else {
4803 free(opt_load_session_path);
4804 opt_load_session_path = strdup(arg);
4805 if (!opt_load_session_path) {
4806 PERROR("strdup");
4807 ret = -ENOMEM;
4808 }
4809 }
4810 } else if (string_match(optname, "kmod-probes")) {
4811 if (lttng_is_setuid_setgid()) {
4812 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4813 "--kmod-probes");
4814 } else {
4815 free(kmod_probes_list);
4816 kmod_probes_list = strdup(arg);
4817 if (!kmod_probes_list) {
4818 PERROR("strdup");
4819 ret = -ENOMEM;
4820 }
4821 }
4822 } else if (string_match(optname, "extra-kmod-probes")) {
4823 if (lttng_is_setuid_setgid()) {
4824 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4825 "--extra-kmod-probes");
4826 } else {
4827 free(kmod_extra_probes_list);
4828 kmod_extra_probes_list = strdup(arg);
4829 if (!kmod_extra_probes_list) {
4830 PERROR("strdup");
4831 ret = -ENOMEM;
4832 }
4833 }
4834 } else if (string_match(optname, "config") || opt == 'f') {
4835 /* This is handled in set_options() thus silent skip. */
4836 goto end;
4837 } else {
4838 /* Unknown option or other error.
4839 * Error is printed by getopt, just return */
4840 ret = -1;
4841 }
4842
4843 end:
4844 if (ret == -EINVAL) {
4845 const char *opt_name = "unknown";
4846 int i;
4847
4848 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4849 i++) {
4850 if (opt == long_options[i].val) {
4851 opt_name = long_options[i].name;
4852 break;
4853 }
4854 }
4855
4856 WARN("Invalid argument provided for option \"%s\", using default value.",
4857 opt_name);
4858 }
4859
4860 return ret;
4861 }
4862
4863 /*
4864 * config_entry_handler_cb used to handle options read from a config file.
4865 * See config_entry_handler_cb comment in common/config/config.h for the
4866 * return value conventions.
4867 */
4868 static int config_entry_handler(const struct config_entry *entry, void *unused)
4869 {
4870 int ret = 0, i;
4871
4872 if (!entry || !entry->name || !entry->value) {
4873 ret = -EINVAL;
4874 goto end;
4875 }
4876
4877 /* Check if the option is to be ignored */
4878 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4879 if (!strcmp(entry->name, config_ignore_options[i])) {
4880 goto end;
4881 }
4882 }
4883
4884 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4885 i++) {
4886
4887 /* Ignore if not fully matched. */
4888 if (strcmp(entry->name, long_options[i].name)) {
4889 continue;
4890 }
4891
4892 /*
4893 * If the option takes no argument on the command line, we have to
4894 * check if the value is "true". We support non-zero numeric values,
4895 * true, on and yes.
4896 */
4897 if (!long_options[i].has_arg) {
4898 ret = config_parse_value(entry->value);
4899 if (ret <= 0) {
4900 if (ret) {
4901 WARN("Invalid configuration value \"%s\" for option %s",
4902 entry->value, entry->name);
4903 }
4904 /* False, skip boolean config option. */
4905 goto end;
4906 }
4907 }
4908
4909 ret = set_option(long_options[i].val, entry->value, entry->name);
4910 goto end;
4911 }
4912
4913 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4914
4915 end:
4916 return ret;
4917 }
4918
4919 /*
4920 * daemon configuration loading and argument parsing
4921 */
4922 static int set_options(int argc, char **argv)
4923 {
4924 int ret = 0, c = 0, option_index = 0;
4925 int orig_optopt = optopt, orig_optind = optind;
4926 char *optstring;
4927 const char *config_path = NULL;
4928
4929 optstring = utils_generate_optstring(long_options,
4930 sizeof(long_options) / sizeof(struct option));
4931 if (!optstring) {
4932 ret = -ENOMEM;
4933 goto end;
4934 }
4935
4936 /* Check for the --config option */
4937 while ((c = getopt_long(argc, argv, optstring, long_options,
4938 &option_index)) != -1) {
4939 if (c == '?') {
4940 ret = -EINVAL;
4941 goto end;
4942 } else if (c != 'f') {
4943 /* if not equal to --config option. */
4944 continue;
4945 }
4946
4947 if (lttng_is_setuid_setgid()) {
4948 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4949 "-f, --config");
4950 } else {
4951 config_path = utils_expand_path(optarg);
4952 if (!config_path) {
4953 ERR("Failed to resolve path: %s", optarg);
4954 }
4955 }
4956 }
4957
4958 ret = config_get_section_entries(config_path, config_section_name,
4959 config_entry_handler, NULL);
4960 if (ret) {
4961 if (ret > 0) {
4962 ERR("Invalid configuration option at line %i", ret);
4963 ret = -1;
4964 }
4965 goto end;
4966 }
4967
4968 /* Reset getopt's global state */
4969 optopt = orig_optopt;
4970 optind = orig_optind;
4971 while (1) {
4972 option_index = -1;
4973 /*
4974 * getopt_long() will not set option_index if it encounters a
4975 * short option.
4976 */
4977 c = getopt_long(argc, argv, optstring, long_options,
4978 &option_index);
4979 if (c == -1) {
4980 break;
4981 }
4982
4983 /*
4984 * Pass NULL as the long option name if popt left the index
4985 * unset.
4986 */
4987 ret = set_option(c, optarg,
4988 option_index < 0 ? NULL :
4989 long_options[option_index].name);
4990 if (ret < 0) {
4991 break;
4992 }
4993 }
4994
4995 end:
4996 free(optstring);
4997 return ret;
4998 }
4999
5000 /*
5001 * Creates the two needed socket by the daemon.
5002 * apps_sock - The communication socket for all UST apps.
5003 * client_sock - The communication of the cli tool (lttng).
5004 */
5005 static int init_daemon_socket(void)
5006 {
5007 int ret = 0;
5008 mode_t old_umask;
5009
5010 old_umask = umask(0);
5011
5012 /* Create client tool unix socket */
5013 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5014 if (client_sock < 0) {
5015 ERR("Create unix sock failed: %s", client_unix_sock_path);
5016 ret = -1;
5017 goto end;
5018 }
5019
5020 /* Set the cloexec flag */
5021 ret = utils_set_fd_cloexec(client_sock);
5022 if (ret < 0) {
5023 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5024 "Continuing but note that the consumer daemon will have a "
5025 "reference to this socket on exec()", client_sock);
5026 }
5027
5028 /* File permission MUST be 660 */
5029 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5030 if (ret < 0) {
5031 ERR("Set file permissions failed: %s", client_unix_sock_path);
5032 PERROR("chmod");
5033 goto end;
5034 }
5035
5036 /* Create the application unix socket */
5037 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5038 if (apps_sock < 0) {
5039 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5040 ret = -1;
5041 goto end;
5042 }
5043
5044 /* Set the cloexec flag */
5045 ret = utils_set_fd_cloexec(apps_sock);
5046 if (ret < 0) {
5047 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5048 "Continuing but note that the consumer daemon will have a "
5049 "reference to this socket on exec()", apps_sock);
5050 }
5051
5052 /* File permission MUST be 666 */
5053 ret = chmod(apps_unix_sock_path,
5054 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5055 if (ret < 0) {
5056 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5057 PERROR("chmod");
5058 goto end;
5059 }
5060
5061 DBG3("Session daemon client socket %d and application socket %d created",
5062 client_sock, apps_sock);
5063
5064 end:
5065 umask(old_umask);
5066 return ret;
5067 }
5068
5069 /*
5070 * Check if the global socket is available, and if a daemon is answering at the
5071 * other side. If yes, error is returned.
5072 */
5073 static int check_existing_daemon(void)
5074 {
5075 /* Is there anybody out there ? */
5076 if (lttng_session_daemon_alive()) {
5077 return -EEXIST;
5078 }
5079
5080 return 0;
5081 }
5082
5083 /*
5084 * Set the tracing group gid onto the client socket.
5085 *
5086 * Race window between mkdir and chown is OK because we are going from more
5087 * permissive (root.root) to less permissive (root.tracing).
5088 */
5089 static int set_permissions(char *rundir)
5090 {
5091 int ret;
5092 gid_t gid;
5093
5094 gid = utils_get_group_id(tracing_group_name);
5095
5096 /* Set lttng run dir */
5097 ret = chown(rundir, 0, gid);
5098 if (ret < 0) {
5099 ERR("Unable to set group on %s", rundir);
5100 PERROR("chown");
5101 }
5102
5103 /*
5104 * Ensure all applications and tracing group can search the run
5105 * dir. Allow everyone to read the directory, since it does not
5106 * buy us anything to hide its content.
5107 */
5108 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5109 if (ret < 0) {
5110 ERR("Unable to set permissions on %s", rundir);
5111 PERROR("chmod");
5112 }
5113
5114 /* lttng client socket path */
5115 ret = chown(client_unix_sock_path, 0, gid);
5116 if (ret < 0) {
5117 ERR("Unable to set group on %s", client_unix_sock_path);
5118 PERROR("chown");
5119 }
5120
5121 /* kconsumer error socket path */
5122 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5123 if (ret < 0) {
5124 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5125 PERROR("chown");
5126 }
5127
5128 /* 64-bit ustconsumer error socket path */
5129 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5130 if (ret < 0) {
5131 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5132 PERROR("chown");
5133 }
5134
5135 /* 32-bit ustconsumer compat32 error socket path */
5136 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5137 if (ret < 0) {
5138 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5139 PERROR("chown");
5140 }
5141
5142 DBG("All permissions are set");
5143
5144 return ret;
5145 }
5146
5147 /*
5148 * Create the lttng run directory needed for all global sockets and pipe.
5149 */
5150 static int create_lttng_rundir(const char *rundir)
5151 {
5152 int ret;
5153
5154 DBG3("Creating LTTng run directory: %s", rundir);
5155
5156 ret = mkdir(rundir, S_IRWXU);
5157 if (ret < 0) {
5158 if (errno != EEXIST) {
5159 ERR("Unable to create %s", rundir);
5160 goto error;
5161 } else {
5162 ret = 0;
5163 }
5164 }
5165
5166 error:
5167 return ret;
5168 }
5169
5170 /*
5171 * Setup sockets and directory needed by the kconsumerd communication with the
5172 * session daemon.
5173 */
5174 static int set_consumer_sockets(struct consumer_data *consumer_data,
5175 const char *rundir)
5176 {
5177 int ret;
5178 char path[PATH_MAX];
5179
5180 switch (consumer_data->type) {
5181 case LTTNG_CONSUMER_KERNEL:
5182 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5183 break;
5184 case LTTNG_CONSUMER64_UST:
5185 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5186 break;
5187 case LTTNG_CONSUMER32_UST:
5188 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5189 break;
5190 default:
5191 ERR("Consumer type unknown");
5192 ret = -EINVAL;
5193 goto error;
5194 }
5195
5196 DBG2("Creating consumer directory: %s", path);
5197
5198 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5199 if (ret < 0) {
5200 if (errno != EEXIST) {
5201 PERROR("mkdir");
5202 ERR("Failed to create %s", path);
5203 goto error;
5204 }
5205 ret = -1;
5206 }
5207 if (is_root) {
5208 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5209 if (ret < 0) {
5210 ERR("Unable to set group on %s", path);
5211 PERROR("chown");
5212 goto error;
5213 }
5214 }
5215
5216 /* Create the kconsumerd error unix socket */
5217 consumer_data->err_sock =
5218 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5219 if (consumer_data->err_sock < 0) {
5220 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5221 ret = -1;
5222 goto error;
5223 }
5224
5225 /*
5226 * Set the CLOEXEC flag. Return code is useless because either way, the
5227 * show must go on.
5228 */
5229 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5230 if (ret < 0) {
5231 PERROR("utils_set_fd_cloexec");
5232 /* continue anyway */
5233 }
5234
5235 /* File permission MUST be 660 */
5236 ret = chmod(consumer_data->err_unix_sock_path,
5237 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5238 if (ret < 0) {
5239 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5240 PERROR("chmod");
5241 goto error;
5242 }
5243
5244 error:
5245 return ret;
5246 }
5247
5248 /*
5249 * Signal handler for the daemon
5250 *
5251 * Simply stop all worker threads, leaving main() return gracefully after
5252 * joining all threads and calling cleanup().
5253 */
5254 static void sighandler(int sig)
5255 {
5256 switch (sig) {
5257 case SIGPIPE:
5258 DBG("SIGPIPE caught");
5259 return;
5260 case SIGINT:
5261 DBG("SIGINT caught");
5262 stop_threads();
5263 break;
5264 case SIGTERM:
5265 DBG("SIGTERM caught");
5266 stop_threads();
5267 break;
5268 case SIGUSR1:
5269 CMM_STORE_SHARED(recv_child_signal, 1);
5270 break;
5271 default:
5272 break;
5273 }
5274 }
5275
5276 /*
5277 * Setup signal handler for :
5278 * SIGINT, SIGTERM, SIGPIPE
5279 */
5280 static int set_signal_handler(void)
5281 {
5282 int ret = 0;
5283 struct sigaction sa;
5284 sigset_t sigset;
5285
5286 if ((ret = sigemptyset(&sigset)) < 0) {
5287 PERROR("sigemptyset");
5288 return ret;
5289 }
5290
5291 sa.sa_handler = sighandler;
5292 sa.sa_mask = sigset;
5293 sa.sa_flags = 0;
5294 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5295 PERROR("sigaction");
5296 return ret;
5297 }
5298
5299 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5300 PERROR("sigaction");
5301 return ret;
5302 }
5303
5304 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5305 PERROR("sigaction");
5306 return ret;
5307 }
5308
5309 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5310 PERROR("sigaction");
5311 return ret;
5312 }
5313
5314 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5315
5316 return ret;
5317 }
5318
5319 /*
5320 * Set open files limit to unlimited. This daemon can open a large number of
5321 * file descriptors in order to consumer multiple kernel traces.
5322 */
5323 static void set_ulimit(void)
5324 {
5325 int ret;
5326 struct rlimit lim;
5327
5328 /* The kernel does not allowed an infinite limit for open files */
5329 lim.rlim_cur = 65535;
5330 lim.rlim_max = 65535;
5331
5332 ret = setrlimit(RLIMIT_NOFILE, &lim);
5333 if (ret < 0) {
5334 PERROR("failed to set open files limit");
5335 }
5336 }
5337
5338 /*
5339 * Write pidfile using the rundir and opt_pidfile.
5340 */
5341 static int write_pidfile(void)
5342 {
5343 int ret;
5344 char pidfile_path[PATH_MAX];
5345
5346 assert(rundir);
5347
5348 if (opt_pidfile) {
5349 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5350 } else {
5351 /* Build pidfile path from rundir and opt_pidfile. */
5352 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5353 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5354 if (ret < 0) {
5355 PERROR("snprintf pidfile path");
5356 goto error;
5357 }
5358 }
5359
5360 /*
5361 * Create pid file in rundir.
5362 */
5363 ret = utils_create_pid_file(getpid(), pidfile_path);
5364 error:
5365 return ret;
5366 }
5367
5368 /*
5369 * Create lockfile using the rundir and return its fd.
5370 */
5371 static int create_lockfile(void)
5372 {
5373 int ret;
5374 char lockfile_path[PATH_MAX];
5375
5376 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5377 if (ret < 0) {
5378 goto error;
5379 }
5380
5381 ret = utils_create_lock_file(lockfile_path);
5382 error:
5383 return ret;
5384 }
5385
5386 /*
5387 * Write agent TCP port using the rundir.
5388 */
5389 static int write_agent_port(void)
5390 {
5391 int ret;
5392 char path[PATH_MAX];
5393
5394 assert(rundir);
5395
5396 ret = snprintf(path, sizeof(path), "%s/"
5397 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5398 if (ret < 0) {
5399 PERROR("snprintf agent port path");
5400 goto error;
5401 }
5402
5403 /*
5404 * Create TCP agent port file in rundir.
5405 */
5406 ret = utils_create_pid_file(agent_tcp_port, path);
5407
5408 error:
5409 return ret;
5410 }
5411
5412 /*
5413 * main
5414 */
5415 int main(int argc, char **argv)
5416 {
5417 int ret = 0, retval = 0;
5418 void *status;
5419 const char *home_path, *env_app_timeout;
5420
5421 /* Initialize agent apps ht global variable */
5422 agent_apps_ht_by_sock = NULL;
5423
5424 init_kernel_workarounds();
5425
5426 rcu_register_thread();
5427
5428 if (set_signal_handler()) {
5429 retval = -1;
5430 goto exit_set_signal_handler;
5431 }
5432
5433 setup_consumerd_path();
5434
5435 page_size = sysconf(_SC_PAGESIZE);
5436 if (page_size < 0) {
5437 PERROR("sysconf _SC_PAGESIZE");
5438 page_size = LONG_MAX;
5439 WARN("Fallback page size to %ld", page_size);
5440 }
5441
5442 /*
5443 * Parse arguments and load the daemon configuration file.
5444 *
5445 * We have an exit_options exit path to free memory reserved by
5446 * set_options. This is needed because the rest of sessiond_cleanup()
5447 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5448 * depends on set_options.
5449 */
5450 progname = argv[0];
5451 if (set_options(argc, argv)) {
5452 retval = -1;
5453 goto exit_options;
5454 }
5455
5456 /* Daemonize */
5457 if (opt_daemon || opt_background) {
5458 int i;
5459
5460 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5461 !opt_background);
5462 if (ret < 0) {
5463 retval = -1;
5464 goto exit_options;
5465 }
5466
5467 /*
5468 * We are in the child. Make sure all other file descriptors are
5469 * closed, in case we are called with more opened file
5470 * descriptors than the standard ones.
5471 */
5472 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5473 (void) close(i);
5474 }
5475 }
5476
5477 /*
5478 * Starting from here, we can create threads. This needs to be after
5479 * lttng_daemonize due to RCU.
5480 */
5481
5482 /*
5483 * Initialize the health check subsystem. This call should set the
5484 * appropriate time values.
5485 */
5486 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5487 if (!health_sessiond) {
5488 PERROR("health_app_create error");
5489 retval = -1;
5490 goto exit_health_sessiond_cleanup;
5491 }
5492
5493 if (init_ht_cleanup_quit_pipe()) {
5494 retval = -1;
5495 goto exit_ht_cleanup_quit_pipe;
5496 }
5497
5498 /* Setup the thread ht_cleanup communication pipe. */
5499 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5500 retval = -1;
5501 goto exit_ht_cleanup_pipe;
5502 }
5503
5504 /* Set up max poll set size */
5505 if (lttng_poll_set_max_size()) {
5506 retval = -1;
5507 goto exit_set_max_size;
5508 }
5509
5510 /* Create thread to clean up RCU hash tables */
5511 ret = pthread_create(&ht_cleanup_thread, NULL,
5512 thread_ht_cleanup, (void *) NULL);
5513 if (ret) {
5514 errno = ret;
5515 PERROR("pthread_create ht_cleanup");
5516 retval = -1;
5517 goto exit_ht_cleanup;
5518 }
5519
5520 /* Create thread quit pipe */
5521 if (init_thread_quit_pipe()) {
5522 retval = -1;
5523 goto exit_init_data;
5524 }
5525
5526 /* Check if daemon is UID = 0 */
5527 is_root = !getuid();
5528
5529 if (is_root) {
5530 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5531 if (!rundir) {
5532 retval = -1;
5533 goto exit_init_data;
5534 }
5535
5536 /* Create global run dir with root access */
5537 if (create_lttng_rundir(rundir)) {
5538 retval = -1;
5539 goto exit_init_data;
5540 }
5541
5542 if (strlen(apps_unix_sock_path) == 0) {
5543 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5544 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5545 if (ret < 0) {
5546 retval = -1;
5547 goto exit_init_data;
5548 }
5549 }
5550
5551 if (strlen(client_unix_sock_path) == 0) {
5552 ret = snprintf(client_unix_sock_path, PATH_MAX,
5553 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5554 if (ret < 0) {
5555 retval = -1;
5556 goto exit_init_data;
5557 }
5558 }
5559
5560 /* Set global SHM for ust */
5561 if (strlen(wait_shm_path) == 0) {
5562 ret = snprintf(wait_shm_path, PATH_MAX,
5563 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5564 if (ret < 0) {
5565 retval = -1;
5566 goto exit_init_data;
5567 }
5568 }
5569
5570 if (strlen(health_unix_sock_path) == 0) {
5571 ret = snprintf(health_unix_sock_path,
5572 sizeof(health_unix_sock_path),
5573 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5574 if (ret < 0) {
5575 retval = -1;
5576 goto exit_init_data;
5577 }
5578 }
5579
5580 /* Setup kernel consumerd path */
5581 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5582 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5583 if (ret < 0) {
5584 retval = -1;
5585 goto exit_init_data;
5586 }
5587 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5588 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5589 if (ret < 0) {
5590 retval = -1;
5591 goto exit_init_data;
5592 }
5593
5594 DBG2("Kernel consumer err path: %s",
5595 kconsumer_data.err_unix_sock_path);
5596 DBG2("Kernel consumer cmd path: %s",
5597 kconsumer_data.cmd_unix_sock_path);
5598 } else {
5599 home_path = utils_get_home_dir();
5600 if (home_path == NULL) {
5601 /* TODO: Add --socket PATH option */
5602 ERR("Can't get HOME directory for sockets creation.");
5603 retval = -1;
5604 goto exit_init_data;
5605 }
5606
5607 /*
5608 * Create rundir from home path. This will create something like
5609 * $HOME/.lttng
5610 */
5611 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5612 if (ret < 0) {
5613 retval = -1;
5614 goto exit_init_data;
5615 }
5616
5617 if (create_lttng_rundir(rundir)) {
5618 retval = -1;
5619 goto exit_init_data;
5620 }
5621
5622 if (strlen(apps_unix_sock_path) == 0) {
5623 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5624 DEFAULT_HOME_APPS_UNIX_SOCK,
5625 home_path);
5626 if (ret < 0) {
5627 retval = -1;
5628 goto exit_init_data;
5629 }
5630 }
5631
5632 /* Set the cli tool unix socket path */
5633 if (strlen(client_unix_sock_path) == 0) {
5634 ret = snprintf(client_unix_sock_path, PATH_MAX,
5635 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5636 home_path);
5637 if (ret < 0) {
5638 retval = -1;
5639 goto exit_init_data;
5640 }
5641 }
5642
5643 /* Set global SHM for ust */
5644 if (strlen(wait_shm_path) == 0) {
5645 ret = snprintf(wait_shm_path, PATH_MAX,
5646 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5647 getuid());
5648 if (ret < 0) {
5649 retval = -1;
5650 goto exit_init_data;
5651 }
5652 }
5653
5654 /* Set health check Unix path */
5655 if (strlen(health_unix_sock_path) == 0) {
5656 ret = snprintf(health_unix_sock_path,
5657 sizeof(health_unix_sock_path),
5658 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5659 home_path);
5660 if (ret < 0) {
5661 retval = -1;
5662 goto exit_init_data;
5663 }
5664 }
5665 }
5666
5667 lockfile_fd = create_lockfile();
5668 if (lockfile_fd < 0) {
5669 retval = -1;
5670 goto exit_init_data;
5671 }
5672
5673 /* Set consumer initial state */
5674 kernel_consumerd_state = CONSUMER_STOPPED;
5675 ust_consumerd_state = CONSUMER_STOPPED;
5676
5677 DBG("Client socket path %s", client_unix_sock_path);
5678 DBG("Application socket path %s", apps_unix_sock_path);
5679 DBG("Application wait path %s", wait_shm_path);
5680 DBG("LTTng run directory path: %s", rundir);
5681
5682 /* 32 bits consumerd path setup */
5683 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5684 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5685 if (ret < 0) {
5686 PERROR("snprintf 32-bit consumer error socket path");
5687 retval = -1;
5688 goto exit_init_data;
5689 }
5690 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5691 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5692 if (ret < 0) {
5693 PERROR("snprintf 32-bit consumer command socket path");
5694 retval = -1;
5695 goto exit_init_data;
5696 }
5697
5698 DBG2("UST consumer 32 bits err path: %s",
5699 ustconsumer32_data.err_unix_sock_path);
5700 DBG2("UST consumer 32 bits cmd path: %s",
5701 ustconsumer32_data.cmd_unix_sock_path);
5702
5703 /* 64 bits consumerd path setup */
5704 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5705 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5706 if (ret < 0) {
5707 PERROR("snprintf 64-bit consumer error socket path");
5708 retval = -1;
5709 goto exit_init_data;
5710 }
5711 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5712 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5713 if (ret < 0) {
5714 PERROR("snprintf 64-bit consumer command socket path");
5715 retval = -1;
5716 goto exit_init_data;
5717 }
5718
5719 DBG2("UST consumer 64 bits err path: %s",
5720 ustconsumer64_data.err_unix_sock_path);
5721 DBG2("UST consumer 64 bits cmd path: %s",
5722 ustconsumer64_data.cmd_unix_sock_path);
5723
5724 /*
5725 * See if daemon already exist.
5726 */
5727 if (check_existing_daemon()) {
5728 ERR("Already running daemon.\n");
5729 /*
5730 * We do not goto exit because we must not cleanup()
5731 * because a daemon is already running.
5732 */
5733 retval = -1;
5734 goto exit_init_data;
5735 }
5736
5737 /*
5738 * Init UST app hash table. Alloc hash table before this point since
5739 * cleanup() can get called after that point.
5740 */
5741 if (ust_app_ht_alloc()) {
5742 ERR("Failed to allocate UST app hash table");
5743 retval = -1;
5744 goto exit_init_data;
5745 }
5746
5747 /*
5748 * Initialize agent app hash table. We allocate the hash table here
5749 * since cleanup() can get called after this point.
5750 */
5751 if (agent_app_ht_alloc()) {
5752 ERR("Failed to allocate Agent app hash table");
5753 retval = -1;
5754 goto exit_init_data;
5755 }
5756
5757 /*
5758 * These actions must be executed as root. We do that *after* setting up
5759 * the sockets path because we MUST make the check for another daemon using
5760 * those paths *before* trying to set the kernel consumer sockets and init
5761 * kernel tracer.
5762 */
5763 if (is_root) {
5764 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5765 retval = -1;
5766 goto exit_init_data;
5767 }
5768
5769 /* Setup kernel tracer */
5770 if (!opt_no_kernel) {
5771 init_kernel_tracer();
5772 if (kernel_tracer_fd >= 0) {
5773 ret = syscall_init_table();
5774 if (ret < 0) {
5775 ERR("Unable to populate syscall table. "
5776 "Syscall tracing won't work "
5777 "for this session daemon.");
5778 }
5779 }
5780 }
5781
5782 /* Set ulimit for open files */
5783 set_ulimit();
5784 }
5785 /* init lttng_fd tracking must be done after set_ulimit. */
5786 lttng_fd_init();
5787
5788 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5789 retval = -1;
5790 goto exit_init_data;
5791 }
5792
5793 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5794 retval = -1;
5795 goto exit_init_data;
5796 }
5797
5798 /* Setup the needed unix socket */
5799 if (init_daemon_socket()) {
5800 retval = -1;
5801 goto exit_init_data;
5802 }
5803
5804 /* Set credentials to socket */
5805 if (is_root && set_permissions(rundir)) {
5806 retval = -1;
5807 goto exit_init_data;
5808 }
5809
5810 /* Get parent pid if -S, --sig-parent is specified. */
5811 if (opt_sig_parent) {
5812 ppid = getppid();
5813 }
5814
5815 /* Setup the kernel pipe for waking up the kernel thread */
5816 if (is_root && !opt_no_kernel) {
5817 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5818 retval = -1;
5819 goto exit_init_data;
5820 }
5821 }
5822
5823 /* Setup the thread apps communication pipe. */
5824 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5825 retval = -1;
5826 goto exit_init_data;
5827 }
5828
5829 /* Setup the thread apps notify communication pipe. */
5830 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5831 retval = -1;
5832 goto exit_init_data;
5833 }
5834
5835 /* Initialize global buffer per UID and PID registry. */
5836 buffer_reg_init_uid_registry();
5837 buffer_reg_init_pid_registry();
5838
5839 /* Init UST command queue. */
5840 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5841
5842 /*
5843 * Get session list pointer. This pointer MUST NOT be free'd. This list
5844 * is statically declared in session.c
5845 */
5846 session_list_ptr = session_get_list();
5847
5848 cmd_init();
5849
5850 /* Check for the application socket timeout env variable. */
5851 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5852 if (env_app_timeout) {
5853 app_socket_timeout = atoi(env_app_timeout);
5854 } else {
5855 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5856 }
5857
5858 ret = write_pidfile();
5859 if (ret) {
5860 ERR("Error in write_pidfile");
5861 retval = -1;
5862 goto exit_init_data;
5863 }
5864 ret = write_agent_port();
5865 if (ret) {
5866 ERR("Error in write_agent_port");
5867 retval = -1;
5868 goto exit_init_data;
5869 }
5870
5871 /* Initialize communication library */
5872 lttcomm_init();
5873 /* Initialize TCP timeout values */
5874 lttcomm_inet_init();
5875
5876 if (load_session_init_data(&load_info) < 0) {
5877 retval = -1;
5878 goto exit_init_data;
5879 }
5880 load_info->path = opt_load_session_path;
5881
5882 /* Create health-check thread */
5883 ret = pthread_create(&health_thread, NULL,
5884 thread_manage_health, (void *) NULL);
5885 if (ret) {
5886 errno = ret;
5887 PERROR("pthread_create health");
5888 retval = -1;
5889 goto exit_health;
5890 }
5891
5892 /* Create thread to manage the client socket */
5893 ret = pthread_create(&client_thread, NULL,
5894 thread_manage_clients, (void *) NULL);
5895 if (ret) {
5896 errno = ret;
5897 PERROR("pthread_create clients");
5898 retval = -1;
5899 goto exit_client;
5900 }
5901
5902 /* Create thread to dispatch registration */
5903 ret = pthread_create(&dispatch_thread, NULL,
5904 thread_dispatch_ust_registration, (void *) NULL);
5905 if (ret) {
5906 errno = ret;
5907 PERROR("pthread_create dispatch");
5908 retval = -1;
5909 goto exit_dispatch;
5910 }
5911
5912 /* Create thread to manage application registration. */
5913 ret = pthread_create(&reg_apps_thread, NULL,
5914 thread_registration_apps, (void *) NULL);
5915 if (ret) {
5916 errno = ret;
5917 PERROR("pthread_create registration");
5918 retval = -1;
5919 goto exit_reg_apps;
5920 }
5921
5922 /* Create thread to manage application socket */
5923 ret = pthread_create(&apps_thread, NULL,
5924 thread_manage_apps, (void *) NULL);
5925 if (ret) {
5926 errno = ret;
5927 PERROR("pthread_create apps");
5928 retval = -1;
5929 goto exit_apps;
5930 }
5931
5932 /* Create thread to manage application notify socket */
5933 ret = pthread_create(&apps_notify_thread, NULL,
5934 ust_thread_manage_notify, (void *) NULL);
5935 if (ret) {
5936 errno = ret;
5937 PERROR("pthread_create notify");
5938 retval = -1;
5939 goto exit_apps_notify;
5940 }
5941
5942 /* Create agent registration thread. */
5943 ret = pthread_create(&agent_reg_thread, NULL,
5944 agent_thread_manage_registration, (void *) NULL);
5945 if (ret) {
5946 errno = ret;
5947 PERROR("pthread_create agent");
5948 retval = -1;
5949 goto exit_agent_reg;
5950 }
5951
5952 /* Don't start this thread if kernel tracing is not requested nor root */
5953 if (is_root && !opt_no_kernel) {
5954 /* Create kernel thread to manage kernel event */
5955 ret = pthread_create(&kernel_thread, NULL,
5956 thread_manage_kernel, (void *) NULL);
5957 if (ret) {
5958 errno = ret;
5959 PERROR("pthread_create kernel");
5960 retval = -1;
5961 goto exit_kernel;
5962 }
5963 }
5964
5965 /* Create session loading thread. */
5966 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5967 load_info);
5968 if (ret) {
5969 errno = ret;
5970 PERROR("pthread_create load_session_thread");
5971 retval = -1;
5972 goto exit_load_session;
5973 }
5974
5975 /*
5976 * This is where we start awaiting program completion (e.g. through
5977 * signal that asks threads to teardown).
5978 */
5979
5980 ret = pthread_join(load_session_thread, &status);
5981 if (ret) {
5982 errno = ret;
5983 PERROR("pthread_join load_session_thread");
5984 retval = -1;
5985 }
5986 exit_load_session:
5987
5988 if (is_root && !opt_no_kernel) {
5989 ret = pthread_join(kernel_thread, &status);
5990 if (ret) {
5991 errno = ret;
5992 PERROR("pthread_join");
5993 retval = -1;
5994 }
5995 }
5996 exit_kernel:
5997
5998 ret = pthread_join(agent_reg_thread, &status);
5999 if (ret) {
6000 errno = ret;
6001 PERROR("pthread_join agent");
6002 retval = -1;
6003 }
6004 exit_agent_reg:
6005
6006 ret = pthread_join(apps_notify_thread, &status);
6007 if (ret) {
6008 errno = ret;
6009 PERROR("pthread_join apps notify");
6010 retval = -1;
6011 }
6012 exit_apps_notify:
6013
6014 ret = pthread_join(apps_thread, &status);
6015 if (ret) {
6016 errno = ret;
6017 PERROR("pthread_join apps");
6018 retval = -1;
6019 }
6020 exit_apps:
6021
6022 ret = pthread_join(reg_apps_thread, &status);
6023 if (ret) {
6024 errno = ret;
6025 PERROR("pthread_join");
6026 retval = -1;
6027 }
6028 exit_reg_apps:
6029
6030 /*
6031 * Join dispatch thread after joining reg_apps_thread to ensure
6032 * we don't leak applications in the queue.
6033 */
6034 ret = pthread_join(dispatch_thread, &status);
6035 if (ret) {
6036 errno = ret;
6037 PERROR("pthread_join");
6038 retval = -1;
6039 }
6040 exit_dispatch:
6041
6042 ret = pthread_join(client_thread, &status);
6043 if (ret) {
6044 errno = ret;
6045 PERROR("pthread_join");
6046 retval = -1;
6047 }
6048 exit_client:
6049
6050 ret = pthread_join(health_thread, &status);
6051 if (ret) {
6052 errno = ret;
6053 PERROR("pthread_join health thread");
6054 retval = -1;
6055 }
6056 exit_health:
6057
6058 exit_init_data:
6059 /*
6060 * sessiond_cleanup() is called when no other thread is running, except
6061 * the ht_cleanup thread, which is needed to destroy the hash tables.
6062 */
6063 rcu_thread_online();
6064 sessiond_cleanup();
6065 rcu_thread_offline();
6066 rcu_unregister_thread();
6067
6068 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6069 if (ret < 0) {
6070 ERR("write error on ht_cleanup quit pipe");
6071 retval = -1;
6072 }
6073
6074 ret = pthread_join(ht_cleanup_thread, &status);
6075 if (ret) {
6076 errno = ret;
6077 PERROR("pthread_join ht cleanup thread");
6078 retval = -1;
6079 }
6080 exit_ht_cleanup:
6081 exit_set_max_size:
6082
6083 utils_close_pipe(ht_cleanup_pipe);
6084 exit_ht_cleanup_pipe:
6085
6086 /*
6087 * Close the ht_cleanup quit pipe.
6088 */
6089 utils_close_pipe(ht_cleanup_quit_pipe);
6090 exit_ht_cleanup_quit_pipe:
6091
6092 health_app_destroy(health_sessiond);
6093 exit_health_sessiond_cleanup:
6094
6095 exit_options:
6096 sessiond_cleanup_options();
6097
6098 exit_set_signal_handler:
6099 /* Ensure all prior call_rcu are done. */
6100 rcu_barrier();
6101
6102 if (!retval) {
6103 exit(EXIT_SUCCESS);
6104 } else {
6105 exit(EXIT_FAILURE);
6106 }
6107 }
This page took 0.149901 seconds and 5 git commands to generate.