Port: Remove _GNU_SOURCE, defined in config.h
[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 _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/compat/getenv.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/config.h>
51
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
54 #include "channel.h"
55 #include "cmd.h"
56 #include "consumer.h"
57 #include "context.h"
58 #include "event.h"
59 #include "kernel.h"
60 #include "kernel-consumer.h"
61 #include "modprobe.h"
62 #include "shm.h"
63 #include "ust-ctl.h"
64 #include "ust-consumer.h"
65 #include "utils.h"
66 #include "fd-limit.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "agent-thread.h"
71 #include "save.h"
72 #include "load-session-thread.h"
73 #include "syscall.h"
74 #include "agent.h"
75
76 #define CONSUMERD_FILE "lttng-consumerd"
77
78 const char *progname;
79 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
80 static int tracing_group_name_override;
81 static char *opt_pidfile;
82 static int opt_sig_parent;
83 static int opt_verbose_consumer;
84 static int opt_daemon, opt_background;
85 static int opt_no_kernel;
86 static char *opt_load_session_path;
87 static pid_t ppid; /* Parent PID for --sig-parent option */
88 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
89 static char *rundir;
90 static int lockfile_fd = -1;
91
92 /* Set to 1 when a SIGUSR1 signal is received. */
93 static int recv_child_signal;
94
95 /*
96 * Consumer daemon specific control data. Every value not initialized here is
97 * set to 0 by the static definition.
98 */
99 static struct consumer_data kconsumer_data = {
100 .type = LTTNG_CONSUMER_KERNEL,
101 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer64_data = {
111 .type = LTTNG_CONSUMER64_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121 static struct consumer_data ustconsumer32_data = {
122 .type = LTTNG_CONSUMER32_UST,
123 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
124 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
125 .err_sock = -1,
126 .cmd_sock = -1,
127 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
128 .lock = PTHREAD_MUTEX_INITIALIZER,
129 .cond = PTHREAD_COND_INITIALIZER,
130 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
131 };
132
133 /* Command line options */
134 static const struct option long_options[] = {
135 { "client-sock", required_argument, 0, 'c' },
136 { "apps-sock", required_argument, 0, 'a' },
137 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
138 { "kconsumerd-err-sock", required_argument, 0, '\0' },
139 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
143 { "consumerd32-path", required_argument, 0, '\0' },
144 { "consumerd32-libdir", required_argument, 0, '\0' },
145 { "consumerd64-path", required_argument, 0, '\0' },
146 { "consumerd64-libdir", required_argument, 0, '\0' },
147 { "daemonize", no_argument, 0, 'd' },
148 { "background", no_argument, 0, 'b' },
149 { "sig-parent", no_argument, 0, 'S' },
150 { "help", no_argument, 0, 'h' },
151 { "group", required_argument, 0, 'g' },
152 { "version", no_argument, 0, 'V' },
153 { "quiet", no_argument, 0, 'q' },
154 { "verbose", no_argument, 0, 'v' },
155 { "verbose-consumer", no_argument, 0, '\0' },
156 { "no-kernel", no_argument, 0, '\0' },
157 { "pidfile", required_argument, 0, 'p' },
158 { "agent-tcp-port", required_argument, 0, '\0' },
159 { "config", required_argument, 0, 'f' },
160 { "load", required_argument, 0, 'l' },
161 { "kmod-probes", required_argument, 0, '\0' },
162 { "extra-kmod-probes", required_argument, 0, '\0' },
163 { NULL, 0, 0, 0 }
164 };
165
166 /* Command line options to ignore from configuration file */
167 static const char *config_ignore_options[] = { "help", "version", "config" };
168
169 /* Shared between threads */
170 static int dispatch_thread_exit;
171
172 /* Global application Unix socket path */
173 static char apps_unix_sock_path[PATH_MAX];
174 /* Global client Unix socket path */
175 static char client_unix_sock_path[PATH_MAX];
176 /* global wait shm path for UST */
177 static char wait_shm_path[PATH_MAX];
178 /* Global health check unix path */
179 static char health_unix_sock_path[PATH_MAX];
180
181 /* Sockets and FDs */
182 static int client_sock = -1;
183 static int apps_sock = -1;
184 int kernel_tracer_fd = -1;
185 static int kernel_poll_pipe[2] = { -1, -1 };
186
187 /*
188 * Quit pipe for all threads. This permits a single cancellation point
189 * for all threads when receiving an event on the pipe.
190 */
191 static int thread_quit_pipe[2] = { -1, -1 };
192 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
193
194 /*
195 * This pipe is used to inform the thread managing application communication
196 * that a command is queued and ready to be processed.
197 */
198 static int apps_cmd_pipe[2] = { -1, -1 };
199
200 int apps_cmd_notify_pipe[2] = { -1, -1 };
201
202 /* Pthread, Mutexes and Semaphores */
203 static pthread_t apps_thread;
204 static pthread_t apps_notify_thread;
205 static pthread_t reg_apps_thread;
206 static pthread_t client_thread;
207 static pthread_t kernel_thread;
208 static pthread_t dispatch_thread;
209 static pthread_t health_thread;
210 static pthread_t ht_cleanup_thread;
211 static pthread_t agent_reg_thread;
212 static pthread_t load_session_thread;
213
214 /*
215 * UST registration command queue. This queue is tied with a futex and uses a N
216 * wakers / 1 waiter implemented and detailed in futex.c/.h
217 *
218 * The thread_registration_apps and thread_dispatch_ust_registration uses this
219 * queue along with the wait/wake scheme. The thread_manage_apps receives down
220 * the line new application socket and monitors it for any I/O error or clean
221 * close that triggers an unregistration of the application.
222 */
223 static struct ust_cmd_queue ust_cmd_queue;
224
225 /*
226 * Pointer initialized before thread creation.
227 *
228 * This points to the tracing session list containing the session count and a
229 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
230 * MUST NOT be taken if you call a public function in session.c.
231 *
232 * The lock is nested inside the structure: session_list_ptr->lock. Please use
233 * session_lock_list and session_unlock_list for lock acquisition.
234 */
235 static struct ltt_session_list *session_list_ptr;
236
237 int ust_consumerd64_fd = -1;
238 int ust_consumerd32_fd = -1;
239
240 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
241 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
242 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
243 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
244 static int consumerd32_bin_override;
245 static int consumerd64_bin_override;
246 static int consumerd32_libdir_override;
247 static int consumerd64_libdir_override;
248
249 static const char *module_proc_lttng = "/proc/lttng";
250
251 /*
252 * Consumer daemon state which is changed when spawning it, killing it or in
253 * case of a fatal error.
254 */
255 enum consumerd_state {
256 CONSUMER_STARTED = 1,
257 CONSUMER_STOPPED = 2,
258 CONSUMER_ERROR = 3,
259 };
260
261 /*
262 * This consumer daemon state is used to validate if a client command will be
263 * able to reach the consumer. If not, the client is informed. For instance,
264 * doing a "lttng start" when the consumer state is set to ERROR will return an
265 * error to the client.
266 *
267 * The following example shows a possible race condition of this scheme:
268 *
269 * consumer thread error happens
270 * client cmd arrives
271 * client cmd checks state -> still OK
272 * consumer thread exit, sets error
273 * client cmd try to talk to consumer
274 * ...
275 *
276 * However, since the consumer is a different daemon, we have no way of making
277 * sure the command will reach it safely even with this state flag. This is why
278 * we consider that up to the state validation during command processing, the
279 * command is safe. After that, we can not guarantee the correctness of the
280 * client request vis-a-vis the consumer.
281 */
282 static enum consumerd_state ust_consumerd_state;
283 static enum consumerd_state kernel_consumerd_state;
284
285 /*
286 * Socket timeout for receiving and sending in seconds.
287 */
288 static int app_socket_timeout;
289
290 /* Set in main() with the current page size. */
291 long page_size;
292
293 /* Application health monitoring */
294 struct health_app *health_sessiond;
295
296 /* Agent TCP port for registration. Used by the agent thread. */
297 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
298
299 /* Am I root or not. */
300 int is_root; /* Set to 1 if the daemon is running as root */
301
302 const char * const config_section_name = "sessiond";
303
304 /* Load session thread information to operate. */
305 struct load_session_thread_data *load_info;
306
307 /* Global hash tables */
308 struct lttng_ht *agent_apps_ht_by_sock = NULL;
309
310 /*
311 * Whether sessiond is ready for commands/health check requests.
312 * NR_LTTNG_SESSIOND_READY must match the number of calls to
313 * sessiond_notify_ready().
314 */
315 #define NR_LTTNG_SESSIOND_READY 3
316 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
317
318 /* Notify parents that we are ready for cmd and health check */
319 LTTNG_HIDDEN
320 void sessiond_notify_ready(void)
321 {
322 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
323 /*
324 * Notify parent pid that we are ready to accept command
325 * for client side. This ppid is the one from the
326 * external process that spawned us.
327 */
328 if (opt_sig_parent) {
329 kill(ppid, SIGUSR1);
330 }
331
332 /*
333 * Notify the parent of the fork() process that we are
334 * ready.
335 */
336 if (opt_daemon || opt_background) {
337 kill(child_ppid, SIGUSR1);
338 }
339 }
340 }
341
342 static
343 void setup_consumerd_path(void)
344 {
345 const char *bin, *libdir;
346
347 /*
348 * Allow INSTALL_BIN_PATH to be used as a target path for the
349 * native architecture size consumer if CONFIG_CONSUMER*_PATH
350 * has not been defined.
351 */
352 #if (CAA_BITS_PER_LONG == 32)
353 if (!consumerd32_bin[0]) {
354 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
355 }
356 if (!consumerd32_libdir[0]) {
357 consumerd32_libdir = INSTALL_LIB_PATH;
358 }
359 #elif (CAA_BITS_PER_LONG == 64)
360 if (!consumerd64_bin[0]) {
361 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
362 }
363 if (!consumerd64_libdir[0]) {
364 consumerd64_libdir = INSTALL_LIB_PATH;
365 }
366 #else
367 #error "Unknown bitness"
368 #endif
369
370 /*
371 * runtime env. var. overrides the build default.
372 */
373 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
374 if (bin) {
375 consumerd32_bin = bin;
376 }
377 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
378 if (bin) {
379 consumerd64_bin = bin;
380 }
381 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
382 if (libdir) {
383 consumerd32_libdir = libdir;
384 }
385 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
386 if (libdir) {
387 consumerd64_libdir = libdir;
388 }
389 }
390
391 static
392 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
393 int *a_pipe)
394 {
395 int ret;
396
397 assert(events);
398
399 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
400 if (ret < 0) {
401 goto error;
402 }
403
404 /* Add quit pipe */
405 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
406 if (ret < 0) {
407 goto error;
408 }
409
410 return 0;
411
412 error:
413 return ret;
414 }
415
416 /*
417 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
418 */
419 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
420 {
421 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
422 }
423
424 /*
425 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
426 */
427 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
428 size_t size)
429 {
430 return __sessiond_set_thread_pollset(events, size,
431 ht_cleanup_quit_pipe);
432 }
433
434 static
435 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
436 {
437 if (fd == a_pipe && (events & LPOLLIN)) {
438 return 1;
439 }
440 return 0;
441 }
442
443 /*
444 * Check if the thread quit pipe was triggered.
445 *
446 * Return 1 if it was triggered else 0;
447 */
448 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
449 {
450 return __sessiond_check_thread_quit_pipe(fd, events,
451 thread_quit_pipe[0]);
452 }
453
454 /*
455 * Check if the ht_cleanup thread quit pipe was triggered.
456 *
457 * Return 1 if it was triggered else 0;
458 */
459 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
460 {
461 return __sessiond_check_thread_quit_pipe(fd, events,
462 ht_cleanup_quit_pipe[0]);
463 }
464
465 /*
466 * Init thread quit pipe.
467 *
468 * Return -1 on error or 0 if all pipes are created.
469 */
470 static int __init_thread_quit_pipe(int *a_pipe)
471 {
472 int ret, i;
473
474 ret = pipe(a_pipe);
475 if (ret < 0) {
476 PERROR("thread quit pipe");
477 goto error;
478 }
479
480 for (i = 0; i < 2; i++) {
481 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
482 if (ret < 0) {
483 PERROR("fcntl");
484 goto error;
485 }
486 }
487
488 error:
489 return ret;
490 }
491
492 static int init_thread_quit_pipe(void)
493 {
494 return __init_thread_quit_pipe(thread_quit_pipe);
495 }
496
497 static int init_ht_cleanup_quit_pipe(void)
498 {
499 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
500 }
501
502 /*
503 * Stop all threads by closing the thread quit pipe.
504 */
505 static void stop_threads(void)
506 {
507 int ret;
508
509 /* Stopping all threads */
510 DBG("Terminating all threads");
511 ret = notify_thread_pipe(thread_quit_pipe[1]);
512 if (ret < 0) {
513 ERR("write error on thread quit pipe");
514 }
515
516 /* Dispatch thread */
517 CMM_STORE_SHARED(dispatch_thread_exit, 1);
518 futex_nto1_wake(&ust_cmd_queue.futex);
519 }
520
521 /*
522 * Close every consumer sockets.
523 */
524 static void close_consumer_sockets(void)
525 {
526 int ret;
527
528 if (kconsumer_data.err_sock >= 0) {
529 ret = close(kconsumer_data.err_sock);
530 if (ret < 0) {
531 PERROR("kernel consumer err_sock close");
532 }
533 }
534 if (ustconsumer32_data.err_sock >= 0) {
535 ret = close(ustconsumer32_data.err_sock);
536 if (ret < 0) {
537 PERROR("UST consumerd32 err_sock close");
538 }
539 }
540 if (ustconsumer64_data.err_sock >= 0) {
541 ret = close(ustconsumer64_data.err_sock);
542 if (ret < 0) {
543 PERROR("UST consumerd64 err_sock close");
544 }
545 }
546 if (kconsumer_data.cmd_sock >= 0) {
547 ret = close(kconsumer_data.cmd_sock);
548 if (ret < 0) {
549 PERROR("kernel consumer cmd_sock close");
550 }
551 }
552 if (ustconsumer32_data.cmd_sock >= 0) {
553 ret = close(ustconsumer32_data.cmd_sock);
554 if (ret < 0) {
555 PERROR("UST consumerd32 cmd_sock close");
556 }
557 }
558 if (ustconsumer64_data.cmd_sock >= 0) {
559 ret = close(ustconsumer64_data.cmd_sock);
560 if (ret < 0) {
561 PERROR("UST consumerd64 cmd_sock close");
562 }
563 }
564 }
565
566 /*
567 * Generate the full lock file path using the rundir.
568 *
569 * Return the snprintf() return value thus a negative value is an error.
570 */
571 static int generate_lock_file_path(char *path, size_t len)
572 {
573 int ret;
574
575 assert(path);
576 assert(rundir);
577
578 /* Build lockfile path from rundir. */
579 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
580 if (ret < 0) {
581 PERROR("snprintf lockfile path");
582 }
583
584 return ret;
585 }
586
587 /*
588 * Wait on consumer process termination.
589 *
590 * Need to be called with the consumer data lock held or from a context
591 * ensuring no concurrent access to data (e.g: cleanup).
592 */
593 static void wait_consumer(struct consumer_data *consumer_data)
594 {
595 pid_t ret;
596 int status;
597
598 if (consumer_data->pid <= 0) {
599 return;
600 }
601
602 DBG("Waiting for complete teardown of consumerd (PID: %d)",
603 consumer_data->pid);
604 ret = waitpid(consumer_data->pid, &status, 0);
605 if (ret == -1) {
606 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
607 }
608 if (!WIFEXITED(status)) {
609 ERR("consumerd termination with error: %d",
610 WEXITSTATUS(ret));
611 }
612 consumer_data->pid = 0;
613 }
614
615 /*
616 * Cleanup the session daemon's data structures.
617 */
618 static void sessiond_cleanup(void)
619 {
620 int ret;
621 struct ltt_session *sess, *stmp;
622 char path[PATH_MAX];
623
624 DBG("Cleanup sessiond");
625
626 /*
627 * Close the thread quit pipe. It has already done its job,
628 * since we are now called.
629 */
630 utils_close_pipe(thread_quit_pipe);
631
632 /*
633 * If opt_pidfile is undefined, the default file will be wiped when
634 * removing the rundir.
635 */
636 if (opt_pidfile) {
637 ret = remove(opt_pidfile);
638 if (ret < 0) {
639 PERROR("remove pidfile %s", opt_pidfile);
640 }
641 }
642
643 DBG("Removing sessiond and consumerd content of directory %s", rundir);
644
645 /* sessiond */
646 snprintf(path, PATH_MAX,
647 "%s/%s",
648 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
649 DBG("Removing %s", path);
650 (void) unlink(path);
651
652 snprintf(path, PATH_MAX, "%s/%s", rundir,
653 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
654 DBG("Removing %s", path);
655 (void) unlink(path);
656
657 /* kconsumerd */
658 snprintf(path, PATH_MAX,
659 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
660 rundir);
661 DBG("Removing %s", path);
662 (void) unlink(path);
663
664 snprintf(path, PATH_MAX,
665 DEFAULT_KCONSUMERD_PATH,
666 rundir);
667 DBG("Removing directory %s", path);
668 (void) rmdir(path);
669
670 /* ust consumerd 32 */
671 snprintf(path, PATH_MAX,
672 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
673 rundir);
674 DBG("Removing %s", path);
675 (void) unlink(path);
676
677 snprintf(path, PATH_MAX,
678 DEFAULT_USTCONSUMERD32_PATH,
679 rundir);
680 DBG("Removing directory %s", path);
681 (void) rmdir(path);
682
683 /* ust consumerd 64 */
684 snprintf(path, PATH_MAX,
685 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
686 rundir);
687 DBG("Removing %s", path);
688 (void) unlink(path);
689
690 snprintf(path, PATH_MAX,
691 DEFAULT_USTCONSUMERD64_PATH,
692 rundir);
693 DBG("Removing directory %s", path);
694 (void) rmdir(path);
695
696 DBG("Cleaning up all sessions");
697
698 /* Destroy session list mutex */
699 if (session_list_ptr != NULL) {
700 pthread_mutex_destroy(&session_list_ptr->lock);
701
702 /* Cleanup ALL session */
703 cds_list_for_each_entry_safe(sess, stmp,
704 &session_list_ptr->head, list) {
705 cmd_destroy_session(sess, kernel_poll_pipe[1]);
706 }
707 }
708
709 wait_consumer(&kconsumer_data);
710 wait_consumer(&ustconsumer64_data);
711 wait_consumer(&ustconsumer32_data);
712
713 DBG("Cleaning up all agent apps");
714 agent_app_ht_clean();
715
716 DBG("Closing all UST sockets");
717 ust_app_clean_list();
718 buffer_reg_destroy_registries();
719
720 if (is_root && !opt_no_kernel) {
721 DBG2("Closing kernel fd");
722 if (kernel_tracer_fd >= 0) {
723 ret = close(kernel_tracer_fd);
724 if (ret) {
725 PERROR("close");
726 }
727 }
728 DBG("Unloading kernel modules");
729 modprobe_remove_lttng_all();
730 free(syscall_table);
731 }
732
733 close_consumer_sockets();
734
735 if (load_info) {
736 load_session_destroy_data(load_info);
737 free(load_info);
738 }
739
740 /*
741 * Cleanup lock file by deleting it and finaly closing it which will
742 * release the file system lock.
743 */
744 if (lockfile_fd >= 0) {
745 char lockfile_path[PATH_MAX];
746
747 ret = generate_lock_file_path(lockfile_path,
748 sizeof(lockfile_path));
749 if (ret > 0) {
750 ret = remove(lockfile_path);
751 if (ret < 0) {
752 PERROR("remove lock file");
753 }
754 ret = close(lockfile_fd);
755 if (ret < 0) {
756 PERROR("close lock file");
757 }
758 }
759 }
760
761 /*
762 * We do NOT rmdir rundir because there are other processes
763 * using it, for instance lttng-relayd, which can start in
764 * parallel with this teardown.
765 */
766
767 free(rundir);
768 }
769
770 /*
771 * Cleanup the daemon's option data structures.
772 */
773 static void sessiond_cleanup_options(void)
774 {
775 DBG("Cleaning up options");
776
777 /*
778 * If the override option is set, the pointer points to a *non* const
779 * thus freeing it even though the variable type is set to const.
780 */
781 if (tracing_group_name_override) {
782 free((void *) tracing_group_name);
783 }
784 if (consumerd32_bin_override) {
785 free((void *) consumerd32_bin);
786 }
787 if (consumerd64_bin_override) {
788 free((void *) consumerd64_bin);
789 }
790 if (consumerd32_libdir_override) {
791 free((void *) consumerd32_libdir);
792 }
793 if (consumerd64_libdir_override) {
794 free((void *) consumerd64_libdir);
795 }
796
797 free(opt_pidfile);
798 free(opt_load_session_path);
799 free(kmod_probes_list);
800 free(kmod_extra_probes_list);
801
802 run_as_destroy_worker();
803
804 /* <fun> */
805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 /* </fun> */
809 }
810
811 /*
812 * Send data on a unix socket using the liblttsessiondcomm API.
813 *
814 * Return lttcomm error code.
815 */
816 static int send_unix_sock(int sock, void *buf, size_t len)
817 {
818 /* Check valid length */
819 if (len == 0) {
820 return -1;
821 }
822
823 return lttcomm_send_unix_sock(sock, buf, len);
824 }
825
826 /*
827 * Free memory of a command context structure.
828 */
829 static void clean_command_ctx(struct command_ctx **cmd_ctx)
830 {
831 DBG("Clean command context structure");
832 if (*cmd_ctx) {
833 if ((*cmd_ctx)->llm) {
834 free((*cmd_ctx)->llm);
835 }
836 if ((*cmd_ctx)->lsm) {
837 free((*cmd_ctx)->lsm);
838 }
839 free(*cmd_ctx);
840 *cmd_ctx = NULL;
841 }
842 }
843
844 /*
845 * Notify UST applications using the shm mmap futex.
846 */
847 static int notify_ust_apps(int active)
848 {
849 char *wait_shm_mmap;
850
851 DBG("Notifying applications of session daemon state: %d", active);
852
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
855 if (wait_shm_mmap == NULL) {
856 goto error;
857 }
858
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap, active);
861
862 /* Apps notified successfully */
863 return 0;
864
865 error:
866 return -1;
867 }
868
869 /*
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
872 * structure.
873 *
874 * Return total size of the buffer pointed by buf.
875 */
876 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
877 {
878 int ret, buf_size;
879
880 buf_size = size;
881
882 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
883 if (cmd_ctx->llm == NULL) {
884 PERROR("zmalloc");
885 ret = -ENOMEM;
886 goto error;
887 }
888
889 /* Copy common data */
890 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
891 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
892
893 cmd_ctx->llm->data_size = size;
894 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
895
896 return buf_size;
897
898 error:
899 return ret;
900 }
901
902 /*
903 * Update the kernel poll set of all channel fd available over all tracing
904 * session. Add the wakeup pipe at the end of the set.
905 */
906 static int update_kernel_poll(struct lttng_poll_event *events)
907 {
908 int ret;
909 struct ltt_session *session;
910 struct ltt_kernel_channel *channel;
911
912 DBG("Updating kernel poll set");
913
914 session_lock_list();
915 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
916 session_lock(session);
917 if (session->kernel_session == NULL) {
918 session_unlock(session);
919 continue;
920 }
921
922 cds_list_for_each_entry(channel,
923 &session->kernel_session->channel_list.head, list) {
924 /* Add channel fd to the kernel poll set */
925 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
926 if (ret < 0) {
927 session_unlock(session);
928 goto error;
929 }
930 DBG("Channel fd %d added to kernel set", channel->fd);
931 }
932 session_unlock(session);
933 }
934 session_unlock_list();
935
936 return 0;
937
938 error:
939 session_unlock_list();
940 return -1;
941 }
942
943 /*
944 * Find the channel fd from 'fd' over all tracing session. When found, check
945 * for new channel stream and send those stream fds to the kernel consumer.
946 *
947 * Useful for CPU hotplug feature.
948 */
949 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
950 {
951 int ret = 0;
952 struct ltt_session *session;
953 struct ltt_kernel_session *ksess;
954 struct ltt_kernel_channel *channel;
955
956 DBG("Updating kernel streams for channel fd %d", fd);
957
958 session_lock_list();
959 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
960 session_lock(session);
961 if (session->kernel_session == NULL) {
962 session_unlock(session);
963 continue;
964 }
965 ksess = session->kernel_session;
966
967 cds_list_for_each_entry(channel,
968 &ksess->channel_list.head, list) {
969 struct lttng_ht_iter iter;
970 struct consumer_socket *socket;
971
972 if (channel->fd != fd) {
973 continue;
974 }
975 DBG("Channel found, updating kernel streams");
976 ret = kernel_open_channel_stream(channel);
977 if (ret < 0) {
978 goto error;
979 }
980 /* Update the stream global counter */
981 ksess->stream_count_global += ret;
982
983 /*
984 * Have we already sent fds to the consumer? If yes, it
985 * means that tracing is started so it is safe to send
986 * our updated stream fds.
987 */
988 if (ksess->consumer_fds_sent != 1
989 || ksess->consumer == NULL) {
990 ret = -1;
991 goto error;
992 }
993
994 rcu_read_lock();
995 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
996 &iter.iter, socket, node.node) {
997 pthread_mutex_lock(socket->lock);
998 ret = kernel_consumer_send_channel_stream(socket,
999 channel, ksess,
1000 session->output_traces ? 1 : 0);
1001 pthread_mutex_unlock(socket->lock);
1002 if (ret < 0) {
1003 rcu_read_unlock();
1004 goto error;
1005 }
1006 }
1007 rcu_read_unlock();
1008 }
1009 session_unlock(session);
1010 }
1011 session_unlock_list();
1012 return ret;
1013
1014 error:
1015 session_unlock(session);
1016 session_unlock_list();
1017 return ret;
1018 }
1019
1020 /*
1021 * For each tracing session, update newly registered apps. The session list
1022 * lock MUST be acquired before calling this.
1023 */
1024 static void update_ust_app(int app_sock)
1025 {
1026 struct ltt_session *sess, *stmp;
1027
1028 /* Consumer is in an ERROR state. Stop any application update. */
1029 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1030 /* Stop the update process since the consumer is dead. */
1031 return;
1032 }
1033
1034 /* For all tracing session(s) */
1035 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1036 struct ust_app *app;
1037
1038 session_lock(sess);
1039 if (!sess->ust_session) {
1040 goto unlock_session;
1041 }
1042
1043 rcu_read_lock();
1044 assert(app_sock >= 0);
1045 app = ust_app_find_by_sock(app_sock);
1046 if (app == NULL) {
1047 /*
1048 * Application can be unregistered before so
1049 * this is possible hence simply stopping the
1050 * update.
1051 */
1052 DBG3("UST app update failed to find app sock %d",
1053 app_sock);
1054 goto unlock_rcu;
1055 }
1056 ust_app_global_update(sess->ust_session, app);
1057 unlock_rcu:
1058 rcu_read_unlock();
1059 unlock_session:
1060 session_unlock(sess);
1061 }
1062 }
1063
1064 /*
1065 * This thread manage event coming from the kernel.
1066 *
1067 * Features supported in this thread:
1068 * -) CPU Hotplug
1069 */
1070 static void *thread_manage_kernel(void *data)
1071 {
1072 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1073 uint32_t revents, nb_fd;
1074 char tmp;
1075 struct lttng_poll_event events;
1076
1077 DBG("[thread] Thread manage kernel started");
1078
1079 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1080
1081 /*
1082 * This first step of the while is to clean this structure which could free
1083 * non NULL pointers so initialize it before the loop.
1084 */
1085 lttng_poll_init(&events);
1086
1087 if (testpoint(sessiond_thread_manage_kernel)) {
1088 goto error_testpoint;
1089 }
1090
1091 health_code_update();
1092
1093 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1094 goto error_testpoint;
1095 }
1096
1097 while (1) {
1098 health_code_update();
1099
1100 if (update_poll_flag == 1) {
1101 /* Clean events object. We are about to populate it again. */
1102 lttng_poll_clean(&events);
1103
1104 ret = sessiond_set_thread_pollset(&events, 2);
1105 if (ret < 0) {
1106 goto error_poll_create;
1107 }
1108
1109 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1110 if (ret < 0) {
1111 goto error;
1112 }
1113
1114 /* This will add the available kernel channel if any. */
1115 ret = update_kernel_poll(&events);
1116 if (ret < 0) {
1117 goto error;
1118 }
1119 update_poll_flag = 0;
1120 }
1121
1122 DBG("Thread kernel polling");
1123
1124 /* Poll infinite value of time */
1125 restart:
1126 health_poll_entry();
1127 ret = lttng_poll_wait(&events, -1);
1128 DBG("Thread kernel return from poll on %d fds",
1129 LTTNG_POLL_GETNB(&events));
1130 health_poll_exit();
1131 if (ret < 0) {
1132 /*
1133 * Restart interrupted system call.
1134 */
1135 if (errno == EINTR) {
1136 goto restart;
1137 }
1138 goto error;
1139 } else if (ret == 0) {
1140 /* Should not happen since timeout is infinite */
1141 ERR("Return value of poll is 0 with an infinite timeout.\n"
1142 "This should not have happened! Continuing...");
1143 continue;
1144 }
1145
1146 nb_fd = ret;
1147
1148 for (i = 0; i < nb_fd; i++) {
1149 /* Fetch once the poll data */
1150 revents = LTTNG_POLL_GETEV(&events, i);
1151 pollfd = LTTNG_POLL_GETFD(&events, i);
1152
1153 health_code_update();
1154
1155 if (!revents) {
1156 /* No activity for this FD (poll implementation). */
1157 continue;
1158 }
1159
1160 /* Thread quit pipe has been closed. Killing thread. */
1161 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1162 if (ret) {
1163 err = 0;
1164 goto exit;
1165 }
1166
1167 /* Check for data on kernel pipe */
1168 if (revents & LPOLLIN) {
1169 if (pollfd == kernel_poll_pipe[0]) {
1170 (void) lttng_read(kernel_poll_pipe[0],
1171 &tmp, 1);
1172 /*
1173 * Ret value is useless here, if this pipe gets any actions an
1174 * update is required anyway.
1175 */
1176 update_poll_flag = 1;
1177 continue;
1178 } else {
1179 /*
1180 * New CPU detected by the kernel. Adding kernel stream to
1181 * kernel session and updating the kernel consumer
1182 */
1183 ret = update_kernel_stream(&kconsumer_data, pollfd);
1184 if (ret < 0) {
1185 continue;
1186 }
1187 break;
1188 }
1189 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1190 update_poll_flag = 1;
1191 continue;
1192 } else {
1193 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1194 goto error;
1195 }
1196 }
1197 }
1198
1199 exit:
1200 error:
1201 lttng_poll_clean(&events);
1202 error_poll_create:
1203 error_testpoint:
1204 utils_close_pipe(kernel_poll_pipe);
1205 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1206 if (err) {
1207 health_error();
1208 ERR("Health error occurred in %s", __func__);
1209 WARN("Kernel thread died unexpectedly. "
1210 "Kernel tracing can continue but CPU hotplug is disabled.");
1211 }
1212 health_unregister(health_sessiond);
1213 DBG("Kernel thread dying");
1214 return NULL;
1215 }
1216
1217 /*
1218 * Signal pthread condition of the consumer data that the thread.
1219 */
1220 static void signal_consumer_condition(struct consumer_data *data, int state)
1221 {
1222 pthread_mutex_lock(&data->cond_mutex);
1223
1224 /*
1225 * The state is set before signaling. It can be any value, it's the waiter
1226 * job to correctly interpret this condition variable associated to the
1227 * consumer pthread_cond.
1228 *
1229 * A value of 0 means that the corresponding thread of the consumer data
1230 * was not started. 1 indicates that the thread has started and is ready
1231 * for action. A negative value means that there was an error during the
1232 * thread bootstrap.
1233 */
1234 data->consumer_thread_is_ready = state;
1235 (void) pthread_cond_signal(&data->cond);
1236
1237 pthread_mutex_unlock(&data->cond_mutex);
1238 }
1239
1240 /*
1241 * This thread manage the consumer error sent back to the session daemon.
1242 */
1243 static void *thread_manage_consumer(void *data)
1244 {
1245 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1246 uint32_t revents, nb_fd;
1247 enum lttcomm_return_code code;
1248 struct lttng_poll_event events;
1249 struct consumer_data *consumer_data = data;
1250
1251 DBG("[thread] Manage consumer started");
1252
1253 rcu_register_thread();
1254 rcu_thread_online();
1255
1256 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1257
1258 health_code_update();
1259
1260 /*
1261 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1262 * metadata_sock. Nothing more will be added to this poll set.
1263 */
1264 ret = sessiond_set_thread_pollset(&events, 3);
1265 if (ret < 0) {
1266 goto error_poll;
1267 }
1268
1269 /*
1270 * The error socket here is already in a listening state which was done
1271 * just before spawning this thread to avoid a race between the consumer
1272 * daemon exec trying to connect and the listen() call.
1273 */
1274 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1275 if (ret < 0) {
1276 goto error;
1277 }
1278
1279 health_code_update();
1280
1281 /* Infinite blocking call, waiting for transmission */
1282 restart:
1283 health_poll_entry();
1284
1285 if (testpoint(sessiond_thread_manage_consumer)) {
1286 goto error;
1287 }
1288
1289 ret = lttng_poll_wait(&events, -1);
1290 health_poll_exit();
1291 if (ret < 0) {
1292 /*
1293 * Restart interrupted system call.
1294 */
1295 if (errno == EINTR) {
1296 goto restart;
1297 }
1298 goto error;
1299 }
1300
1301 nb_fd = ret;
1302
1303 for (i = 0; i < nb_fd; i++) {
1304 /* Fetch once the poll data */
1305 revents = LTTNG_POLL_GETEV(&events, i);
1306 pollfd = LTTNG_POLL_GETFD(&events, i);
1307
1308 health_code_update();
1309
1310 if (!revents) {
1311 /* No activity for this FD (poll implementation). */
1312 continue;
1313 }
1314
1315 /* Thread quit pipe has been closed. Killing thread. */
1316 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1317 if (ret) {
1318 err = 0;
1319 goto exit;
1320 }
1321
1322 /* Event on the registration socket */
1323 if (pollfd == consumer_data->err_sock) {
1324 if (revents & LPOLLIN) {
1325 continue;
1326 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1327 ERR("consumer err socket poll error");
1328 goto error;
1329 } else {
1330 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1331 goto error;
1332 }
1333 }
1334 }
1335
1336 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1337 if (sock < 0) {
1338 goto error;
1339 }
1340
1341 /*
1342 * Set the CLOEXEC flag. Return code is useless because either way, the
1343 * show must go on.
1344 */
1345 (void) utils_set_fd_cloexec(sock);
1346
1347 health_code_update();
1348
1349 DBG2("Receiving code from consumer err_sock");
1350
1351 /* Getting status code from kconsumerd */
1352 ret = lttcomm_recv_unix_sock(sock, &code,
1353 sizeof(enum lttcomm_return_code));
1354 if (ret <= 0) {
1355 goto error;
1356 }
1357
1358 health_code_update();
1359 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1360 /* Connect both socket, command and metadata. */
1361 consumer_data->cmd_sock =
1362 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1363 consumer_data->metadata_fd =
1364 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1365 if (consumer_data->cmd_sock < 0
1366 || consumer_data->metadata_fd < 0) {
1367 PERROR("consumer connect cmd socket");
1368 /* On error, signal condition and quit. */
1369 signal_consumer_condition(consumer_data, -1);
1370 goto error;
1371 }
1372 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1373 /* Create metadata socket lock. */
1374 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1375 if (consumer_data->metadata_sock.lock == NULL) {
1376 PERROR("zmalloc pthread mutex");
1377 ret = -1;
1378 goto error;
1379 }
1380 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1381
1382 signal_consumer_condition(consumer_data, 1);
1383 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1384 DBG("Consumer metadata socket ready (fd: %d)",
1385 consumer_data->metadata_fd);
1386 } else {
1387 ERR("consumer error when waiting for SOCK_READY : %s",
1388 lttcomm_get_readable_code(-code));
1389 goto error;
1390 }
1391
1392 /* Remove the consumerd error sock since we've established a connexion */
1393 ret = lttng_poll_del(&events, consumer_data->err_sock);
1394 if (ret < 0) {
1395 goto error;
1396 }
1397
1398 /* Add new accepted error socket. */
1399 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1400 if (ret < 0) {
1401 goto error;
1402 }
1403
1404 /* Add metadata socket that is successfully connected. */
1405 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1406 LPOLLIN | LPOLLRDHUP);
1407 if (ret < 0) {
1408 goto error;
1409 }
1410
1411 health_code_update();
1412
1413 /* Infinite blocking call, waiting for transmission */
1414 restart_poll:
1415 while (1) {
1416 health_code_update();
1417
1418 /* Exit the thread because the thread quit pipe has been triggered. */
1419 if (should_quit) {
1420 /* Not a health error. */
1421 err = 0;
1422 goto exit;
1423 }
1424
1425 health_poll_entry();
1426 ret = lttng_poll_wait(&events, -1);
1427 health_poll_exit();
1428 if (ret < 0) {
1429 /*
1430 * Restart interrupted system call.
1431 */
1432 if (errno == EINTR) {
1433 goto restart_poll;
1434 }
1435 goto error;
1436 }
1437
1438 nb_fd = ret;
1439
1440 for (i = 0; i < nb_fd; i++) {
1441 /* Fetch once the poll data */
1442 revents = LTTNG_POLL_GETEV(&events, i);
1443 pollfd = LTTNG_POLL_GETFD(&events, i);
1444
1445 health_code_update();
1446
1447 if (!revents) {
1448 /* No activity for this FD (poll implementation). */
1449 continue;
1450 }
1451
1452 /*
1453 * Thread quit pipe has been triggered, flag that we should stop
1454 * but continue the current loop to handle potential data from
1455 * consumer.
1456 */
1457 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1458
1459 if (pollfd == sock) {
1460 /* Event on the consumerd socket */
1461 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1462 && !(revents & LPOLLIN)) {
1463 ERR("consumer err socket second poll error");
1464 goto error;
1465 }
1466 health_code_update();
1467 /* Wait for any kconsumerd error */
1468 ret = lttcomm_recv_unix_sock(sock, &code,
1469 sizeof(enum lttcomm_return_code));
1470 if (ret <= 0) {
1471 ERR("consumer closed the command socket");
1472 goto error;
1473 }
1474
1475 ERR("consumer return code : %s",
1476 lttcomm_get_readable_code(-code));
1477
1478 goto exit;
1479 } else if (pollfd == consumer_data->metadata_fd) {
1480 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1481 && !(revents & LPOLLIN)) {
1482 ERR("consumer err metadata socket second poll error");
1483 goto error;
1484 }
1485 /* UST metadata requests */
1486 ret = ust_consumer_metadata_request(
1487 &consumer_data->metadata_sock);
1488 if (ret < 0) {
1489 ERR("Handling metadata request");
1490 goto error;
1491 }
1492 }
1493 /* No need for an else branch all FDs are tested prior. */
1494 }
1495 health_code_update();
1496 }
1497
1498 exit:
1499 error:
1500 /*
1501 * We lock here because we are about to close the sockets and some other
1502 * thread might be using them so get exclusive access which will abort all
1503 * other consumer command by other threads.
1504 */
1505 pthread_mutex_lock(&consumer_data->lock);
1506
1507 /* Immediately set the consumerd state to stopped */
1508 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1509 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1510 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1511 consumer_data->type == LTTNG_CONSUMER32_UST) {
1512 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1513 } else {
1514 /* Code flow error... */
1515 assert(0);
1516 }
1517
1518 if (consumer_data->err_sock >= 0) {
1519 ret = close(consumer_data->err_sock);
1520 if (ret) {
1521 PERROR("close");
1522 }
1523 consumer_data->err_sock = -1;
1524 }
1525 if (consumer_data->cmd_sock >= 0) {
1526 ret = close(consumer_data->cmd_sock);
1527 if (ret) {
1528 PERROR("close");
1529 }
1530 consumer_data->cmd_sock = -1;
1531 }
1532 if (consumer_data->metadata_sock.fd_ptr &&
1533 *consumer_data->metadata_sock.fd_ptr >= 0) {
1534 ret = close(*consumer_data->metadata_sock.fd_ptr);
1535 if (ret) {
1536 PERROR("close");
1537 }
1538 }
1539 if (sock >= 0) {
1540 ret = close(sock);
1541 if (ret) {
1542 PERROR("close");
1543 }
1544 }
1545
1546 unlink(consumer_data->err_unix_sock_path);
1547 unlink(consumer_data->cmd_unix_sock_path);
1548 pthread_mutex_unlock(&consumer_data->lock);
1549
1550 /* Cleanup metadata socket mutex. */
1551 if (consumer_data->metadata_sock.lock) {
1552 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1553 free(consumer_data->metadata_sock.lock);
1554 }
1555 lttng_poll_clean(&events);
1556 error_poll:
1557 if (err) {
1558 health_error();
1559 ERR("Health error occurred in %s", __func__);
1560 }
1561 health_unregister(health_sessiond);
1562 DBG("consumer thread cleanup completed");
1563
1564 rcu_thread_offline();
1565 rcu_unregister_thread();
1566
1567 return NULL;
1568 }
1569
1570 /*
1571 * This thread manage application communication.
1572 */
1573 static void *thread_manage_apps(void *data)
1574 {
1575 int i, ret, pollfd, err = -1;
1576 ssize_t size_ret;
1577 uint32_t revents, nb_fd;
1578 struct lttng_poll_event events;
1579
1580 DBG("[thread] Manage application started");
1581
1582 rcu_register_thread();
1583 rcu_thread_online();
1584
1585 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1586
1587 if (testpoint(sessiond_thread_manage_apps)) {
1588 goto error_testpoint;
1589 }
1590
1591 health_code_update();
1592
1593 ret = sessiond_set_thread_pollset(&events, 2);
1594 if (ret < 0) {
1595 goto error_poll_create;
1596 }
1597
1598 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1599 if (ret < 0) {
1600 goto error;
1601 }
1602
1603 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1604 goto error;
1605 }
1606
1607 health_code_update();
1608
1609 while (1) {
1610 DBG("Apps thread polling");
1611
1612 /* Inifinite blocking call, waiting for transmission */
1613 restart:
1614 health_poll_entry();
1615 ret = lttng_poll_wait(&events, -1);
1616 DBG("Apps thread return from poll on %d fds",
1617 LTTNG_POLL_GETNB(&events));
1618 health_poll_exit();
1619 if (ret < 0) {
1620 /*
1621 * Restart interrupted system call.
1622 */
1623 if (errno == EINTR) {
1624 goto restart;
1625 }
1626 goto error;
1627 }
1628
1629 nb_fd = ret;
1630
1631 for (i = 0; i < nb_fd; i++) {
1632 /* Fetch once the poll data */
1633 revents = LTTNG_POLL_GETEV(&events, i);
1634 pollfd = LTTNG_POLL_GETFD(&events, i);
1635
1636 health_code_update();
1637
1638 if (!revents) {
1639 /* No activity for this FD (poll implementation). */
1640 continue;
1641 }
1642
1643 /* Thread quit pipe has been closed. Killing thread. */
1644 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1645 if (ret) {
1646 err = 0;
1647 goto exit;
1648 }
1649
1650 /* Inspect the apps cmd pipe */
1651 if (pollfd == apps_cmd_pipe[0]) {
1652 if (revents & LPOLLIN) {
1653 int sock;
1654
1655 /* Empty pipe */
1656 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1657 if (size_ret < sizeof(sock)) {
1658 PERROR("read apps cmd pipe");
1659 goto error;
1660 }
1661
1662 health_code_update();
1663
1664 /*
1665 * Since this is a command socket (write then read),
1666 * we only monitor the error events of the socket.
1667 */
1668 ret = lttng_poll_add(&events, sock,
1669 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1670 if (ret < 0) {
1671 goto error;
1672 }
1673
1674 DBG("Apps with sock %d added to poll set", sock);
1675 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1676 ERR("Apps command pipe error");
1677 goto error;
1678 } else {
1679 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1680 goto error;
1681 }
1682 } else {
1683 /*
1684 * At this point, we know that a registered application made
1685 * the event at poll_wait.
1686 */
1687 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1688 /* Removing from the poll set */
1689 ret = lttng_poll_del(&events, pollfd);
1690 if (ret < 0) {
1691 goto error;
1692 }
1693
1694 /* Socket closed on remote end. */
1695 ust_app_unregister(pollfd);
1696 } else {
1697 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1698 goto error;
1699 }
1700 }
1701
1702 health_code_update();
1703 }
1704 }
1705
1706 exit:
1707 error:
1708 lttng_poll_clean(&events);
1709 error_poll_create:
1710 error_testpoint:
1711 utils_close_pipe(apps_cmd_pipe);
1712 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1713
1714 /*
1715 * We don't clean the UST app hash table here since already registered
1716 * applications can still be controlled so let them be until the session
1717 * daemon dies or the applications stop.
1718 */
1719
1720 if (err) {
1721 health_error();
1722 ERR("Health error occurred in %s", __func__);
1723 }
1724 health_unregister(health_sessiond);
1725 DBG("Application communication apps thread cleanup complete");
1726 rcu_thread_offline();
1727 rcu_unregister_thread();
1728 return NULL;
1729 }
1730
1731 /*
1732 * Send a socket to a thread This is called from the dispatch UST registration
1733 * thread once all sockets are set for the application.
1734 *
1735 * The sock value can be invalid, we don't really care, the thread will handle
1736 * it and make the necessary cleanup if so.
1737 *
1738 * On success, return 0 else a negative value being the errno message of the
1739 * write().
1740 */
1741 static int send_socket_to_thread(int fd, int sock)
1742 {
1743 ssize_t ret;
1744
1745 /*
1746 * It's possible that the FD is set as invalid with -1 concurrently just
1747 * before calling this function being a shutdown state of the thread.
1748 */
1749 if (fd < 0) {
1750 ret = -EBADF;
1751 goto error;
1752 }
1753
1754 ret = lttng_write(fd, &sock, sizeof(sock));
1755 if (ret < sizeof(sock)) {
1756 PERROR("write apps pipe %d", fd);
1757 if (ret < 0) {
1758 ret = -errno;
1759 }
1760 goto error;
1761 }
1762
1763 /* All good. Don't send back the write positive ret value. */
1764 ret = 0;
1765 error:
1766 return (int) ret;
1767 }
1768
1769 /*
1770 * Sanitize the wait queue of the dispatch registration thread meaning removing
1771 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1772 * notify socket is never received.
1773 */
1774 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1775 {
1776 int ret, nb_fd = 0, i;
1777 unsigned int fd_added = 0;
1778 struct lttng_poll_event events;
1779 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1780
1781 assert(wait_queue);
1782
1783 lttng_poll_init(&events);
1784
1785 /* Just skip everything for an empty queue. */
1786 if (!wait_queue->count) {
1787 goto end;
1788 }
1789
1790 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1791 if (ret < 0) {
1792 goto error_create;
1793 }
1794
1795 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1796 &wait_queue->head, head) {
1797 assert(wait_node->app);
1798 ret = lttng_poll_add(&events, wait_node->app->sock,
1799 LPOLLHUP | LPOLLERR);
1800 if (ret < 0) {
1801 goto error;
1802 }
1803
1804 fd_added = 1;
1805 }
1806
1807 if (!fd_added) {
1808 goto end;
1809 }
1810
1811 /*
1812 * Poll but don't block so we can quickly identify the faulty events and
1813 * clean them afterwards from the wait queue.
1814 */
1815 ret = lttng_poll_wait(&events, 0);
1816 if (ret < 0) {
1817 goto error;
1818 }
1819 nb_fd = ret;
1820
1821 for (i = 0; i < nb_fd; i++) {
1822 /* Get faulty FD. */
1823 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1824 int pollfd = LTTNG_POLL_GETFD(&events, i);
1825
1826 if (!revents) {
1827 /* No activity for this FD (poll implementation). */
1828 continue;
1829 }
1830
1831 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1832 &wait_queue->head, head) {
1833 if (pollfd == wait_node->app->sock &&
1834 (revents & (LPOLLHUP | LPOLLERR))) {
1835 cds_list_del(&wait_node->head);
1836 wait_queue->count--;
1837 ust_app_destroy(wait_node->app);
1838 free(wait_node);
1839 /*
1840 * Silence warning of use-after-free in
1841 * cds_list_for_each_entry_safe which uses
1842 * __typeof__(*wait_node).
1843 */
1844 wait_node = NULL;
1845 break;
1846 } else {
1847 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1848 goto error;
1849 }
1850 }
1851 }
1852
1853 if (nb_fd > 0) {
1854 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1855 }
1856
1857 end:
1858 lttng_poll_clean(&events);
1859 return;
1860
1861 error:
1862 lttng_poll_clean(&events);
1863 error_create:
1864 ERR("Unable to sanitize wait queue");
1865 return;
1866 }
1867
1868 /*
1869 * Dispatch request from the registration threads to the application
1870 * communication thread.
1871 */
1872 static void *thread_dispatch_ust_registration(void *data)
1873 {
1874 int ret, err = -1;
1875 struct cds_wfcq_node *node;
1876 struct ust_command *ust_cmd = NULL;
1877 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1878 struct ust_reg_wait_queue wait_queue = {
1879 .count = 0,
1880 };
1881
1882 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1883
1884 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1885 goto error_testpoint;
1886 }
1887
1888 health_code_update();
1889
1890 CDS_INIT_LIST_HEAD(&wait_queue.head);
1891
1892 DBG("[thread] Dispatch UST command started");
1893
1894 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1895 health_code_update();
1896
1897 /* Atomically prepare the queue futex */
1898 futex_nto1_prepare(&ust_cmd_queue.futex);
1899
1900 do {
1901 struct ust_app *app = NULL;
1902 ust_cmd = NULL;
1903
1904 /*
1905 * Make sure we don't have node(s) that have hung up before receiving
1906 * the notify socket. This is to clean the list in order to avoid
1907 * memory leaks from notify socket that are never seen.
1908 */
1909 sanitize_wait_queue(&wait_queue);
1910
1911 health_code_update();
1912 /* Dequeue command for registration */
1913 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1914 if (node == NULL) {
1915 DBG("Woken up but nothing in the UST command queue");
1916 /* Continue thread execution */
1917 break;
1918 }
1919
1920 ust_cmd = caa_container_of(node, struct ust_command, node);
1921
1922 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1923 " gid:%d sock:%d name:%s (version %d.%d)",
1924 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1925 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1926 ust_cmd->sock, ust_cmd->reg_msg.name,
1927 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1928
1929 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1930 wait_node = zmalloc(sizeof(*wait_node));
1931 if (!wait_node) {
1932 PERROR("zmalloc wait_node dispatch");
1933 ret = close(ust_cmd->sock);
1934 if (ret < 0) {
1935 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1936 }
1937 lttng_fd_put(LTTNG_FD_APPS, 1);
1938 free(ust_cmd);
1939 goto error;
1940 }
1941 CDS_INIT_LIST_HEAD(&wait_node->head);
1942
1943 /* Create application object if socket is CMD. */
1944 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1945 ust_cmd->sock);
1946 if (!wait_node->app) {
1947 ret = close(ust_cmd->sock);
1948 if (ret < 0) {
1949 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1950 }
1951 lttng_fd_put(LTTNG_FD_APPS, 1);
1952 free(wait_node);
1953 free(ust_cmd);
1954 continue;
1955 }
1956 /*
1957 * Add application to the wait queue so we can set the notify
1958 * socket before putting this object in the global ht.
1959 */
1960 cds_list_add(&wait_node->head, &wait_queue.head);
1961 wait_queue.count++;
1962
1963 free(ust_cmd);
1964 /*
1965 * We have to continue here since we don't have the notify
1966 * socket and the application MUST be added to the hash table
1967 * only at that moment.
1968 */
1969 continue;
1970 } else {
1971 /*
1972 * Look for the application in the local wait queue and set the
1973 * notify socket if found.
1974 */
1975 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1976 &wait_queue.head, head) {
1977 health_code_update();
1978 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1979 wait_node->app->notify_sock = ust_cmd->sock;
1980 cds_list_del(&wait_node->head);
1981 wait_queue.count--;
1982 app = wait_node->app;
1983 free(wait_node);
1984 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1985 break;
1986 }
1987 }
1988
1989 /*
1990 * With no application at this stage the received socket is
1991 * basically useless so close it before we free the cmd data
1992 * structure for good.
1993 */
1994 if (!app) {
1995 ret = close(ust_cmd->sock);
1996 if (ret < 0) {
1997 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1998 }
1999 lttng_fd_put(LTTNG_FD_APPS, 1);
2000 }
2001 free(ust_cmd);
2002 }
2003
2004 if (app) {
2005 /*
2006 * @session_lock_list
2007 *
2008 * Lock the global session list so from the register up to the
2009 * registration done message, no thread can see the application
2010 * and change its state.
2011 */
2012 session_lock_list();
2013 rcu_read_lock();
2014
2015 /*
2016 * Add application to the global hash table. This needs to be
2017 * done before the update to the UST registry can locate the
2018 * application.
2019 */
2020 ust_app_add(app);
2021
2022 /* Set app version. This call will print an error if needed. */
2023 (void) ust_app_version(app);
2024
2025 /* Send notify socket through the notify pipe. */
2026 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2027 app->notify_sock);
2028 if (ret < 0) {
2029 rcu_read_unlock();
2030 session_unlock_list();
2031 /*
2032 * No notify thread, stop the UST tracing. However, this is
2033 * not an internal error of the this thread thus setting
2034 * the health error code to a normal exit.
2035 */
2036 err = 0;
2037 goto error;
2038 }
2039
2040 /*
2041 * Update newly registered application with the tracing
2042 * registry info already enabled information.
2043 */
2044 update_ust_app(app->sock);
2045
2046 /*
2047 * Don't care about return value. Let the manage apps threads
2048 * handle app unregistration upon socket close.
2049 */
2050 (void) ust_app_register_done(app);
2051
2052 /*
2053 * Even if the application socket has been closed, send the app
2054 * to the thread and unregistration will take place at that
2055 * place.
2056 */
2057 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2058 if (ret < 0) {
2059 rcu_read_unlock();
2060 session_unlock_list();
2061 /*
2062 * No apps. thread, stop the UST tracing. However, this is
2063 * not an internal error of the this thread thus setting
2064 * the health error code to a normal exit.
2065 */
2066 err = 0;
2067 goto error;
2068 }
2069
2070 rcu_read_unlock();
2071 session_unlock_list();
2072 }
2073 } while (node != NULL);
2074
2075 health_poll_entry();
2076 /* Futex wait on queue. Blocking call on futex() */
2077 futex_nto1_wait(&ust_cmd_queue.futex);
2078 health_poll_exit();
2079 }
2080 /* Normal exit, no error */
2081 err = 0;
2082
2083 error:
2084 /* Clean up wait queue. */
2085 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2086 &wait_queue.head, head) {
2087 cds_list_del(&wait_node->head);
2088 wait_queue.count--;
2089 free(wait_node);
2090 }
2091
2092 /* Empty command queue. */
2093 for (;;) {
2094 /* Dequeue command for registration */
2095 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2096 if (node == NULL) {
2097 break;
2098 }
2099 ust_cmd = caa_container_of(node, struct ust_command, node);
2100 ret = close(ust_cmd->sock);
2101 if (ret < 0) {
2102 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2103 }
2104 lttng_fd_put(LTTNG_FD_APPS, 1);
2105 free(ust_cmd);
2106 }
2107
2108 error_testpoint:
2109 DBG("Dispatch thread dying");
2110 if (err) {
2111 health_error();
2112 ERR("Health error occurred in %s", __func__);
2113 }
2114 health_unregister(health_sessiond);
2115 return NULL;
2116 }
2117
2118 /*
2119 * This thread manage application registration.
2120 */
2121 static void *thread_registration_apps(void *data)
2122 {
2123 int sock = -1, i, ret, pollfd, err = -1;
2124 uint32_t revents, nb_fd;
2125 struct lttng_poll_event events;
2126 /*
2127 * Get allocated in this thread, enqueued to a global queue, dequeued and
2128 * freed in the manage apps thread.
2129 */
2130 struct ust_command *ust_cmd = NULL;
2131
2132 DBG("[thread] Manage application registration started");
2133
2134 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2135
2136 if (testpoint(sessiond_thread_registration_apps)) {
2137 goto error_testpoint;
2138 }
2139
2140 ret = lttcomm_listen_unix_sock(apps_sock);
2141 if (ret < 0) {
2142 goto error_listen;
2143 }
2144
2145 /*
2146 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2147 * more will be added to this poll set.
2148 */
2149 ret = sessiond_set_thread_pollset(&events, 2);
2150 if (ret < 0) {
2151 goto error_create_poll;
2152 }
2153
2154 /* Add the application registration socket */
2155 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2156 if (ret < 0) {
2157 goto error_poll_add;
2158 }
2159
2160 /* Notify all applications to register */
2161 ret = notify_ust_apps(1);
2162 if (ret < 0) {
2163 ERR("Failed to notify applications or create the wait shared memory.\n"
2164 "Execution continues but there might be problem for already\n"
2165 "running applications that wishes to register.");
2166 }
2167
2168 while (1) {
2169 DBG("Accepting application registration");
2170
2171 /* Inifinite blocking call, waiting for transmission */
2172 restart:
2173 health_poll_entry();
2174 ret = lttng_poll_wait(&events, -1);
2175 health_poll_exit();
2176 if (ret < 0) {
2177 /*
2178 * Restart interrupted system call.
2179 */
2180 if (errno == EINTR) {
2181 goto restart;
2182 }
2183 goto error;
2184 }
2185
2186 nb_fd = ret;
2187
2188 for (i = 0; i < nb_fd; i++) {
2189 health_code_update();
2190
2191 /* Fetch once the poll data */
2192 revents = LTTNG_POLL_GETEV(&events, i);
2193 pollfd = LTTNG_POLL_GETFD(&events, i);
2194
2195 if (!revents) {
2196 /* No activity for this FD (poll implementation). */
2197 continue;
2198 }
2199
2200 /* Thread quit pipe has been closed. Killing thread. */
2201 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2202 if (ret) {
2203 err = 0;
2204 goto exit;
2205 }
2206
2207 /* Event on the registration socket */
2208 if (pollfd == apps_sock) {
2209 if (revents & LPOLLIN) {
2210 sock = lttcomm_accept_unix_sock(apps_sock);
2211 if (sock < 0) {
2212 goto error;
2213 }
2214
2215 /*
2216 * Set socket timeout for both receiving and ending.
2217 * app_socket_timeout is in seconds, whereas
2218 * lttcomm_setsockopt_rcv_timeout and
2219 * lttcomm_setsockopt_snd_timeout expect msec as
2220 * parameter.
2221 */
2222 (void) lttcomm_setsockopt_rcv_timeout(sock,
2223 app_socket_timeout * 1000);
2224 (void) lttcomm_setsockopt_snd_timeout(sock,
2225 app_socket_timeout * 1000);
2226
2227 /*
2228 * Set the CLOEXEC flag. Return code is useless because
2229 * either way, the show must go on.
2230 */
2231 (void) utils_set_fd_cloexec(sock);
2232
2233 /* Create UST registration command for enqueuing */
2234 ust_cmd = zmalloc(sizeof(struct ust_command));
2235 if (ust_cmd == NULL) {
2236 PERROR("ust command zmalloc");
2237 ret = close(sock);
2238 if (ret) {
2239 PERROR("close");
2240 }
2241 goto error;
2242 }
2243
2244 /*
2245 * Using message-based transmissions to ensure we don't
2246 * have to deal with partially received messages.
2247 */
2248 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2249 if (ret < 0) {
2250 ERR("Exhausted file descriptors allowed for applications.");
2251 free(ust_cmd);
2252 ret = close(sock);
2253 if (ret) {
2254 PERROR("close");
2255 }
2256 sock = -1;
2257 continue;
2258 }
2259
2260 health_code_update();
2261 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2262 if (ret < 0) {
2263 free(ust_cmd);
2264 /* Close socket of the application. */
2265 ret = close(sock);
2266 if (ret) {
2267 PERROR("close");
2268 }
2269 lttng_fd_put(LTTNG_FD_APPS, 1);
2270 sock = -1;
2271 continue;
2272 }
2273 health_code_update();
2274
2275 ust_cmd->sock = sock;
2276 sock = -1;
2277
2278 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2279 " gid:%d sock:%d name:%s (version %d.%d)",
2280 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2281 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2282 ust_cmd->sock, ust_cmd->reg_msg.name,
2283 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2284
2285 /*
2286 * Lock free enqueue the registration request. The red pill
2287 * has been taken! This apps will be part of the *system*.
2288 */
2289 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2290
2291 /*
2292 * Wake the registration queue futex. Implicit memory
2293 * barrier with the exchange in cds_wfcq_enqueue.
2294 */
2295 futex_nto1_wake(&ust_cmd_queue.futex);
2296 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2297 ERR("Register apps socket poll error");
2298 goto error;
2299 } else {
2300 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2301 goto error;
2302 }
2303 }
2304 }
2305 }
2306
2307 exit:
2308 error:
2309 /* Notify that the registration thread is gone */
2310 notify_ust_apps(0);
2311
2312 if (apps_sock >= 0) {
2313 ret = close(apps_sock);
2314 if (ret) {
2315 PERROR("close");
2316 }
2317 }
2318 if (sock >= 0) {
2319 ret = close(sock);
2320 if (ret) {
2321 PERROR("close");
2322 }
2323 lttng_fd_put(LTTNG_FD_APPS, 1);
2324 }
2325 unlink(apps_unix_sock_path);
2326
2327 error_poll_add:
2328 lttng_poll_clean(&events);
2329 error_listen:
2330 error_create_poll:
2331 error_testpoint:
2332 DBG("UST Registration thread cleanup complete");
2333 if (err) {
2334 health_error();
2335 ERR("Health error occurred in %s", __func__);
2336 }
2337 health_unregister(health_sessiond);
2338
2339 return NULL;
2340 }
2341
2342 /*
2343 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2344 * exec or it will fails.
2345 */
2346 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2347 {
2348 int ret, clock_ret;
2349 struct timespec timeout;
2350
2351 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2352 consumer_data->consumer_thread_is_ready = 0;
2353
2354 /* Setup pthread condition */
2355 ret = pthread_condattr_init(&consumer_data->condattr);
2356 if (ret) {
2357 errno = ret;
2358 PERROR("pthread_condattr_init consumer data");
2359 goto error;
2360 }
2361
2362 /*
2363 * Set the monotonic clock in order to make sure we DO NOT jump in time
2364 * between the clock_gettime() call and the timedwait call. See bug #324
2365 * for a more details and how we noticed it.
2366 */
2367 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2368 if (ret) {
2369 errno = ret;
2370 PERROR("pthread_condattr_setclock consumer data");
2371 goto error;
2372 }
2373
2374 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2375 if (ret) {
2376 errno = ret;
2377 PERROR("pthread_cond_init consumer data");
2378 goto error;
2379 }
2380
2381 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2382 consumer_data);
2383 if (ret) {
2384 errno = ret;
2385 PERROR("pthread_create consumer");
2386 ret = -1;
2387 goto error;
2388 }
2389
2390 /* We are about to wait on a pthread condition */
2391 pthread_mutex_lock(&consumer_data->cond_mutex);
2392
2393 /* Get time for sem_timedwait absolute timeout */
2394 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2395 /*
2396 * Set the timeout for the condition timed wait even if the clock gettime
2397 * call fails since we might loop on that call and we want to avoid to
2398 * increment the timeout too many times.
2399 */
2400 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2401
2402 /*
2403 * The following loop COULD be skipped in some conditions so this is why we
2404 * set ret to 0 in order to make sure at least one round of the loop is
2405 * done.
2406 */
2407 ret = 0;
2408
2409 /*
2410 * Loop until the condition is reached or when a timeout is reached. Note
2411 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2412 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2413 * possible. This loop does not take any chances and works with both of
2414 * them.
2415 */
2416 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2417 if (clock_ret < 0) {
2418 PERROR("clock_gettime spawn consumer");
2419 /* Infinite wait for the consumerd thread to be ready */
2420 ret = pthread_cond_wait(&consumer_data->cond,
2421 &consumer_data->cond_mutex);
2422 } else {
2423 ret = pthread_cond_timedwait(&consumer_data->cond,
2424 &consumer_data->cond_mutex, &timeout);
2425 }
2426 }
2427
2428 /* Release the pthread condition */
2429 pthread_mutex_unlock(&consumer_data->cond_mutex);
2430
2431 if (ret != 0) {
2432 errno = ret;
2433 if (ret == ETIMEDOUT) {
2434 int pth_ret;
2435
2436 /*
2437 * Call has timed out so we kill the kconsumerd_thread and return
2438 * an error.
2439 */
2440 ERR("Condition timed out. The consumer thread was never ready."
2441 " Killing it");
2442 pth_ret = pthread_cancel(consumer_data->thread);
2443 if (pth_ret < 0) {
2444 PERROR("pthread_cancel consumer thread");
2445 }
2446 } else {
2447 PERROR("pthread_cond_wait failed consumer thread");
2448 }
2449 /* Caller is expecting a negative value on failure. */
2450 ret = -1;
2451 goto error;
2452 }
2453
2454 pthread_mutex_lock(&consumer_data->pid_mutex);
2455 if (consumer_data->pid == 0) {
2456 ERR("Consumerd did not start");
2457 pthread_mutex_unlock(&consumer_data->pid_mutex);
2458 goto error;
2459 }
2460 pthread_mutex_unlock(&consumer_data->pid_mutex);
2461
2462 return 0;
2463
2464 error:
2465 return ret;
2466 }
2467
2468 /*
2469 * Join consumer thread
2470 */
2471 static int join_consumer_thread(struct consumer_data *consumer_data)
2472 {
2473 void *status;
2474
2475 /* Consumer pid must be a real one. */
2476 if (consumer_data->pid > 0) {
2477 int ret;
2478 ret = kill(consumer_data->pid, SIGTERM);
2479 if (ret) {
2480 PERROR("Error killing consumer daemon");
2481 return ret;
2482 }
2483 return pthread_join(consumer_data->thread, &status);
2484 } else {
2485 return 0;
2486 }
2487 }
2488
2489 /*
2490 * Fork and exec a consumer daemon (consumerd).
2491 *
2492 * Return pid if successful else -1.
2493 */
2494 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2495 {
2496 int ret;
2497 pid_t pid;
2498 const char *consumer_to_use;
2499 const char *verbosity;
2500 struct stat st;
2501
2502 DBG("Spawning consumerd");
2503
2504 pid = fork();
2505 if (pid == 0) {
2506 /*
2507 * Exec consumerd.
2508 */
2509 if (opt_verbose_consumer) {
2510 verbosity = "--verbose";
2511 } else if (lttng_opt_quiet) {
2512 verbosity = "--quiet";
2513 } else {
2514 verbosity = "";
2515 }
2516
2517 switch (consumer_data->type) {
2518 case LTTNG_CONSUMER_KERNEL:
2519 /*
2520 * Find out which consumerd to execute. We will first try the
2521 * 64-bit path, then the sessiond's installation directory, and
2522 * fallback on the 32-bit one,
2523 */
2524 DBG3("Looking for a kernel consumer at these locations:");
2525 DBG3(" 1) %s", consumerd64_bin);
2526 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2527 DBG3(" 3) %s", consumerd32_bin);
2528 if (stat(consumerd64_bin, &st) == 0) {
2529 DBG3("Found location #1");
2530 consumer_to_use = consumerd64_bin;
2531 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2532 DBG3("Found location #2");
2533 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2534 } else if (stat(consumerd32_bin, &st) == 0) {
2535 DBG3("Found location #3");
2536 consumer_to_use = consumerd32_bin;
2537 } else {
2538 DBG("Could not find any valid consumerd executable");
2539 ret = -EINVAL;
2540 break;
2541 }
2542 DBG("Using kernel consumer at: %s", consumer_to_use);
2543 ret = execl(consumer_to_use,
2544 "lttng-consumerd", verbosity, "-k",
2545 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2546 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2547 "--group", tracing_group_name,
2548 NULL);
2549 break;
2550 case LTTNG_CONSUMER64_UST:
2551 {
2552 char *tmpnew = NULL;
2553
2554 if (consumerd64_libdir[0] != '\0') {
2555 char *tmp;
2556 size_t tmplen;
2557
2558 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2559 if (!tmp) {
2560 tmp = "";
2561 }
2562 tmplen = strlen("LD_LIBRARY_PATH=")
2563 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2564 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2565 if (!tmpnew) {
2566 ret = -ENOMEM;
2567 goto error;
2568 }
2569 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2570 strcat(tmpnew, consumerd64_libdir);
2571 if (tmp[0] != '\0') {
2572 strcat(tmpnew, ":");
2573 strcat(tmpnew, tmp);
2574 }
2575 ret = putenv(tmpnew);
2576 if (ret) {
2577 ret = -errno;
2578 free(tmpnew);
2579 goto error;
2580 }
2581 }
2582 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2583 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2584 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2585 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2586 "--group", tracing_group_name,
2587 NULL);
2588 if (consumerd64_libdir[0] != '\0') {
2589 free(tmpnew);
2590 }
2591 break;
2592 }
2593 case LTTNG_CONSUMER32_UST:
2594 {
2595 char *tmpnew = NULL;
2596
2597 if (consumerd32_libdir[0] != '\0') {
2598 char *tmp;
2599 size_t tmplen;
2600
2601 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2602 if (!tmp) {
2603 tmp = "";
2604 }
2605 tmplen = strlen("LD_LIBRARY_PATH=")
2606 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2607 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2608 if (!tmpnew) {
2609 ret = -ENOMEM;
2610 goto error;
2611 }
2612 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2613 strcat(tmpnew, consumerd32_libdir);
2614 if (tmp[0] != '\0') {
2615 strcat(tmpnew, ":");
2616 strcat(tmpnew, tmp);
2617 }
2618 ret = putenv(tmpnew);
2619 if (ret) {
2620 ret = -errno;
2621 free(tmpnew);
2622 goto error;
2623 }
2624 }
2625 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2626 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2627 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2628 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2629 "--group", tracing_group_name,
2630 NULL);
2631 if (consumerd32_libdir[0] != '\0') {
2632 free(tmpnew);
2633 }
2634 break;
2635 }
2636 default:
2637 PERROR("unknown consumer type");
2638 exit(EXIT_FAILURE);
2639 }
2640 if (errno != 0) {
2641 PERROR("Consumer execl()");
2642 }
2643 /* Reaching this point, we got a failure on our execl(). */
2644 exit(EXIT_FAILURE);
2645 } else if (pid > 0) {
2646 ret = pid;
2647 } else {
2648 PERROR("start consumer fork");
2649 ret = -errno;
2650 }
2651 error:
2652 return ret;
2653 }
2654
2655 /*
2656 * Spawn the consumerd daemon and session daemon thread.
2657 */
2658 static int start_consumerd(struct consumer_data *consumer_data)
2659 {
2660 int ret;
2661
2662 /*
2663 * Set the listen() state on the socket since there is a possible race
2664 * between the exec() of the consumer daemon and this call if place in the
2665 * consumer thread. See bug #366 for more details.
2666 */
2667 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2668 if (ret < 0) {
2669 goto error;
2670 }
2671
2672 pthread_mutex_lock(&consumer_data->pid_mutex);
2673 if (consumer_data->pid != 0) {
2674 pthread_mutex_unlock(&consumer_data->pid_mutex);
2675 goto end;
2676 }
2677
2678 ret = spawn_consumerd(consumer_data);
2679 if (ret < 0) {
2680 ERR("Spawning consumerd failed");
2681 pthread_mutex_unlock(&consumer_data->pid_mutex);
2682 goto error;
2683 }
2684
2685 /* Setting up the consumer_data pid */
2686 consumer_data->pid = ret;
2687 DBG2("Consumer pid %d", consumer_data->pid);
2688 pthread_mutex_unlock(&consumer_data->pid_mutex);
2689
2690 DBG2("Spawning consumer control thread");
2691 ret = spawn_consumer_thread(consumer_data);
2692 if (ret < 0) {
2693 ERR("Fatal error spawning consumer control thread");
2694 goto error;
2695 }
2696
2697 end:
2698 return 0;
2699
2700 error:
2701 /* Cleanup already created sockets on error. */
2702 if (consumer_data->err_sock >= 0) {
2703 int err;
2704
2705 err = close(consumer_data->err_sock);
2706 if (err < 0) {
2707 PERROR("close consumer data error socket");
2708 }
2709 }
2710 return ret;
2711 }
2712
2713 /*
2714 * Setup necessary data for kernel tracer action.
2715 */
2716 static int init_kernel_tracer(void)
2717 {
2718 int ret;
2719
2720 /* Modprobe lttng kernel modules */
2721 ret = modprobe_lttng_control();
2722 if (ret < 0) {
2723 goto error;
2724 }
2725
2726 /* Open debugfs lttng */
2727 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2728 if (kernel_tracer_fd < 0) {
2729 DBG("Failed to open %s", module_proc_lttng);
2730 ret = -1;
2731 goto error_open;
2732 }
2733
2734 /* Validate kernel version */
2735 ret = kernel_validate_version(kernel_tracer_fd);
2736 if (ret < 0) {
2737 goto error_version;
2738 }
2739
2740 ret = modprobe_lttng_data();
2741 if (ret < 0) {
2742 goto error_modules;
2743 }
2744
2745 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2746 return 0;
2747
2748 error_version:
2749 modprobe_remove_lttng_control();
2750 ret = close(kernel_tracer_fd);
2751 if (ret) {
2752 PERROR("close");
2753 }
2754 kernel_tracer_fd = -1;
2755 return LTTNG_ERR_KERN_VERSION;
2756
2757 error_modules:
2758 ret = close(kernel_tracer_fd);
2759 if (ret) {
2760 PERROR("close");
2761 }
2762
2763 error_open:
2764 modprobe_remove_lttng_control();
2765
2766 error:
2767 WARN("No kernel tracer available");
2768 kernel_tracer_fd = -1;
2769 if (!is_root) {
2770 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2771 } else {
2772 return LTTNG_ERR_KERN_NA;
2773 }
2774 }
2775
2776
2777 /*
2778 * Copy consumer output from the tracing session to the domain session. The
2779 * function also applies the right modification on a per domain basis for the
2780 * trace files destination directory.
2781 *
2782 * Should *NOT* be called with RCU read-side lock held.
2783 */
2784 static int copy_session_consumer(int domain, struct ltt_session *session)
2785 {
2786 int ret;
2787 const char *dir_name;
2788 struct consumer_output *consumer;
2789
2790 assert(session);
2791 assert(session->consumer);
2792
2793 switch (domain) {
2794 case LTTNG_DOMAIN_KERNEL:
2795 DBG3("Copying tracing session consumer output in kernel session");
2796 /*
2797 * XXX: We should audit the session creation and what this function
2798 * does "extra" in order to avoid a destroy since this function is used
2799 * in the domain session creation (kernel and ust) only. Same for UST
2800 * domain.
2801 */
2802 if (session->kernel_session->consumer) {
2803 consumer_output_put(session->kernel_session->consumer);
2804 }
2805 session->kernel_session->consumer =
2806 consumer_copy_output(session->consumer);
2807 /* Ease our life a bit for the next part */
2808 consumer = session->kernel_session->consumer;
2809 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2810 break;
2811 case LTTNG_DOMAIN_JUL:
2812 case LTTNG_DOMAIN_LOG4J:
2813 case LTTNG_DOMAIN_PYTHON:
2814 case LTTNG_DOMAIN_UST:
2815 DBG3("Copying tracing session consumer output in UST session");
2816 if (session->ust_session->consumer) {
2817 consumer_output_put(session->ust_session->consumer);
2818 }
2819 session->ust_session->consumer =
2820 consumer_copy_output(session->consumer);
2821 /* Ease our life a bit for the next part */
2822 consumer = session->ust_session->consumer;
2823 dir_name = DEFAULT_UST_TRACE_DIR;
2824 break;
2825 default:
2826 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2827 goto error;
2828 }
2829
2830 /* Append correct directory to subdir */
2831 strncat(consumer->subdir, dir_name,
2832 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2833 DBG3("Copy session consumer subdir %s", consumer->subdir);
2834
2835 ret = LTTNG_OK;
2836
2837 error:
2838 return ret;
2839 }
2840
2841 /*
2842 * Create an UST session and add it to the session ust list.
2843 *
2844 * Should *NOT* be called with RCU read-side lock held.
2845 */
2846 static int create_ust_session(struct ltt_session *session,
2847 struct lttng_domain *domain)
2848 {
2849 int ret;
2850 struct ltt_ust_session *lus = NULL;
2851
2852 assert(session);
2853 assert(domain);
2854 assert(session->consumer);
2855
2856 switch (domain->type) {
2857 case LTTNG_DOMAIN_JUL:
2858 case LTTNG_DOMAIN_LOG4J:
2859 case LTTNG_DOMAIN_PYTHON:
2860 case LTTNG_DOMAIN_UST:
2861 break;
2862 default:
2863 ERR("Unknown UST domain on create session %d", domain->type);
2864 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2865 goto error;
2866 }
2867
2868 DBG("Creating UST session");
2869
2870 lus = trace_ust_create_session(session->id);
2871 if (lus == NULL) {
2872 ret = LTTNG_ERR_UST_SESS_FAIL;
2873 goto error;
2874 }
2875
2876 lus->uid = session->uid;
2877 lus->gid = session->gid;
2878 lus->output_traces = session->output_traces;
2879 lus->snapshot_mode = session->snapshot_mode;
2880 lus->live_timer_interval = session->live_timer;
2881 session->ust_session = lus;
2882 if (session->shm_path[0]) {
2883 strncpy(lus->root_shm_path, session->shm_path,
2884 sizeof(lus->root_shm_path));
2885 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2886 strncpy(lus->shm_path, session->shm_path,
2887 sizeof(lus->shm_path));
2888 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2889 strncat(lus->shm_path, "/ust",
2890 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2891 }
2892 /* Copy session output to the newly created UST session */
2893 ret = copy_session_consumer(domain->type, session);
2894 if (ret != LTTNG_OK) {
2895 goto error;
2896 }
2897
2898 return LTTNG_OK;
2899
2900 error:
2901 free(lus);
2902 session->ust_session = NULL;
2903 return ret;
2904 }
2905
2906 /*
2907 * Create a kernel tracer session then create the default channel.
2908 */
2909 static int create_kernel_session(struct ltt_session *session)
2910 {
2911 int ret;
2912
2913 DBG("Creating kernel session");
2914
2915 ret = kernel_create_session(session, kernel_tracer_fd);
2916 if (ret < 0) {
2917 ret = LTTNG_ERR_KERN_SESS_FAIL;
2918 goto error;
2919 }
2920
2921 /* Code flow safety */
2922 assert(session->kernel_session);
2923
2924 /* Copy session output to the newly created Kernel session */
2925 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2926 if (ret != LTTNG_OK) {
2927 goto error;
2928 }
2929
2930 /* Create directory(ies) on local filesystem. */
2931 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2932 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2933 ret = run_as_mkdir_recursive(
2934 session->kernel_session->consumer->dst.trace_path,
2935 S_IRWXU | S_IRWXG, session->uid, session->gid);
2936 if (ret < 0) {
2937 if (errno != EEXIST) {
2938 ERR("Trace directory creation error");
2939 goto error;
2940 }
2941 }
2942 }
2943
2944 session->kernel_session->uid = session->uid;
2945 session->kernel_session->gid = session->gid;
2946 session->kernel_session->output_traces = session->output_traces;
2947 session->kernel_session->snapshot_mode = session->snapshot_mode;
2948
2949 return LTTNG_OK;
2950
2951 error:
2952 trace_kernel_destroy_session(session->kernel_session);
2953 session->kernel_session = NULL;
2954 return ret;
2955 }
2956
2957 /*
2958 * Count number of session permitted by uid/gid.
2959 */
2960 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2961 {
2962 unsigned int i = 0;
2963 struct ltt_session *session;
2964
2965 DBG("Counting number of available session for UID %d GID %d",
2966 uid, gid);
2967 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2968 /*
2969 * Only list the sessions the user can control.
2970 */
2971 if (!session_access_ok(session, uid, gid)) {
2972 continue;
2973 }
2974 i++;
2975 }
2976 return i;
2977 }
2978
2979 /*
2980 * Process the command requested by the lttng client within the command
2981 * context structure. This function make sure that the return structure (llm)
2982 * is set and ready for transmission before returning.
2983 *
2984 * Return any error encountered or 0 for success.
2985 *
2986 * "sock" is only used for special-case var. len data.
2987 *
2988 * Should *NOT* be called with RCU read-side lock held.
2989 */
2990 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2991 int *sock_error)
2992 {
2993 int ret = LTTNG_OK;
2994 int need_tracing_session = 1;
2995 int need_domain;
2996
2997 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2998
2999 assert(!rcu_read_ongoing());
3000
3001 *sock_error = 0;
3002
3003 switch (cmd_ctx->lsm->cmd_type) {
3004 case LTTNG_CREATE_SESSION:
3005 case LTTNG_CREATE_SESSION_SNAPSHOT:
3006 case LTTNG_CREATE_SESSION_LIVE:
3007 case LTTNG_DESTROY_SESSION:
3008 case LTTNG_LIST_SESSIONS:
3009 case LTTNG_LIST_DOMAINS:
3010 case LTTNG_START_TRACE:
3011 case LTTNG_STOP_TRACE:
3012 case LTTNG_DATA_PENDING:
3013 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3014 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3015 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3016 case LTTNG_SNAPSHOT_RECORD:
3017 case LTTNG_SAVE_SESSION:
3018 case LTTNG_SET_SESSION_SHM_PATH:
3019 need_domain = 0;
3020 break;
3021 default:
3022 need_domain = 1;
3023 }
3024
3025 if (opt_no_kernel && need_domain
3026 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3027 if (!is_root) {
3028 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3029 } else {
3030 ret = LTTNG_ERR_KERN_NA;
3031 }
3032 goto error;
3033 }
3034
3035 /* Deny register consumer if we already have a spawned consumer. */
3036 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3037 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3038 if (kconsumer_data.pid > 0) {
3039 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3040 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3041 goto error;
3042 }
3043 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3044 }
3045
3046 /*
3047 * Check for command that don't needs to allocate a returned payload. We do
3048 * this here so we don't have to make the call for no payload at each
3049 * command.
3050 */
3051 switch(cmd_ctx->lsm->cmd_type) {
3052 case LTTNG_LIST_SESSIONS:
3053 case LTTNG_LIST_TRACEPOINTS:
3054 case LTTNG_LIST_TRACEPOINT_FIELDS:
3055 case LTTNG_LIST_DOMAINS:
3056 case LTTNG_LIST_CHANNELS:
3057 case LTTNG_LIST_EVENTS:
3058 case LTTNG_LIST_SYSCALLS:
3059 case LTTNG_LIST_TRACKER_PIDS:
3060 break;
3061 default:
3062 /* Setup lttng message with no payload */
3063 ret = setup_lttng_msg(cmd_ctx, 0);
3064 if (ret < 0) {
3065 /* This label does not try to unlock the session */
3066 goto init_setup_error;
3067 }
3068 }
3069
3070 /* Commands that DO NOT need a session. */
3071 switch (cmd_ctx->lsm->cmd_type) {
3072 case LTTNG_CREATE_SESSION:
3073 case LTTNG_CREATE_SESSION_SNAPSHOT:
3074 case LTTNG_CREATE_SESSION_LIVE:
3075 case LTTNG_CALIBRATE:
3076 case LTTNG_LIST_SESSIONS:
3077 case LTTNG_LIST_TRACEPOINTS:
3078 case LTTNG_LIST_SYSCALLS:
3079 case LTTNG_LIST_TRACEPOINT_FIELDS:
3080 case LTTNG_SAVE_SESSION:
3081 need_tracing_session = 0;
3082 break;
3083 default:
3084 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3085 /*
3086 * We keep the session list lock across _all_ commands
3087 * for now, because the per-session lock does not
3088 * handle teardown properly.
3089 */
3090 session_lock_list();
3091 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3092 if (cmd_ctx->session == NULL) {
3093 ret = LTTNG_ERR_SESS_NOT_FOUND;
3094 goto error;
3095 } else {
3096 /* Acquire lock for the session */
3097 session_lock(cmd_ctx->session);
3098 }
3099 break;
3100 }
3101
3102 /*
3103 * Commands that need a valid session but should NOT create one if none
3104 * exists. Instead of creating one and destroying it when the command is
3105 * handled, process that right before so we save some round trip in useless
3106 * code path.
3107 */
3108 switch (cmd_ctx->lsm->cmd_type) {
3109 case LTTNG_DISABLE_CHANNEL:
3110 case LTTNG_DISABLE_EVENT:
3111 switch (cmd_ctx->lsm->domain.type) {
3112 case LTTNG_DOMAIN_KERNEL:
3113 if (!cmd_ctx->session->kernel_session) {
3114 ret = LTTNG_ERR_NO_CHANNEL;
3115 goto error;
3116 }
3117 break;
3118 case LTTNG_DOMAIN_JUL:
3119 case LTTNG_DOMAIN_LOG4J:
3120 case LTTNG_DOMAIN_PYTHON:
3121 case LTTNG_DOMAIN_UST:
3122 if (!cmd_ctx->session->ust_session) {
3123 ret = LTTNG_ERR_NO_CHANNEL;
3124 goto error;
3125 }
3126 break;
3127 default:
3128 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3129 goto error;
3130 }
3131 default:
3132 break;
3133 }
3134
3135 if (!need_domain) {
3136 goto skip_domain;
3137 }
3138
3139 /*
3140 * Check domain type for specific "pre-action".
3141 */
3142 switch (cmd_ctx->lsm->domain.type) {
3143 case LTTNG_DOMAIN_KERNEL:
3144 if (!is_root) {
3145 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3146 goto error;
3147 }
3148
3149 /* Kernel tracer check */
3150 if (kernel_tracer_fd == -1) {
3151 /* Basically, load kernel tracer modules */
3152 ret = init_kernel_tracer();
3153 if (ret != 0) {
3154 goto error;
3155 }
3156 }
3157
3158 /* Consumer is in an ERROR state. Report back to client */
3159 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3160 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3161 goto error;
3162 }
3163
3164 /* Need a session for kernel command */
3165 if (need_tracing_session) {
3166 if (cmd_ctx->session->kernel_session == NULL) {
3167 ret = create_kernel_session(cmd_ctx->session);
3168 if (ret < 0) {
3169 ret = LTTNG_ERR_KERN_SESS_FAIL;
3170 goto error;
3171 }
3172 }
3173
3174 /* Start the kernel consumer daemon */
3175 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3176 if (kconsumer_data.pid == 0 &&
3177 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3178 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3179 ret = start_consumerd(&kconsumer_data);
3180 if (ret < 0) {
3181 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3182 goto error;
3183 }
3184 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3185 } else {
3186 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3187 }
3188
3189 /*
3190 * The consumer was just spawned so we need to add the socket to
3191 * the consumer output of the session if exist.
3192 */
3193 ret = consumer_create_socket(&kconsumer_data,
3194 cmd_ctx->session->kernel_session->consumer);
3195 if (ret < 0) {
3196 goto error;
3197 }
3198 }
3199
3200 break;
3201 case LTTNG_DOMAIN_JUL:
3202 case LTTNG_DOMAIN_LOG4J:
3203 case LTTNG_DOMAIN_PYTHON:
3204 case LTTNG_DOMAIN_UST:
3205 {
3206 if (!ust_app_supported()) {
3207 ret = LTTNG_ERR_NO_UST;
3208 goto error;
3209 }
3210 /* Consumer is in an ERROR state. Report back to client */
3211 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3212 ret = LTTNG_ERR_NO_USTCONSUMERD;
3213 goto error;
3214 }
3215
3216 if (need_tracing_session) {
3217 /* Create UST session if none exist. */
3218 if (cmd_ctx->session->ust_session == NULL) {
3219 ret = create_ust_session(cmd_ctx->session,
3220 &cmd_ctx->lsm->domain);
3221 if (ret != LTTNG_OK) {
3222 goto error;
3223 }
3224 }
3225
3226 /* Start the UST consumer daemons */
3227 /* 64-bit */
3228 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3229 if (consumerd64_bin[0] != '\0' &&
3230 ustconsumer64_data.pid == 0 &&
3231 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3232 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3233 ret = start_consumerd(&ustconsumer64_data);
3234 if (ret < 0) {
3235 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3236 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3237 goto error;
3238 }
3239
3240 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3241 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3242 } else {
3243 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3244 }
3245
3246 /*
3247 * Setup socket for consumer 64 bit. No need for atomic access
3248 * since it was set above and can ONLY be set in this thread.
3249 */
3250 ret = consumer_create_socket(&ustconsumer64_data,
3251 cmd_ctx->session->ust_session->consumer);
3252 if (ret < 0) {
3253 goto error;
3254 }
3255
3256 /* 32-bit */
3257 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3258 if (consumerd32_bin[0] != '\0' &&
3259 ustconsumer32_data.pid == 0 &&
3260 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3261 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3262 ret = start_consumerd(&ustconsumer32_data);
3263 if (ret < 0) {
3264 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3265 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3266 goto error;
3267 }
3268
3269 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3270 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3271 } else {
3272 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3273 }
3274
3275 /*
3276 * Setup socket for consumer 64 bit. No need for atomic access
3277 * since it was set above and can ONLY be set in this thread.
3278 */
3279 ret = consumer_create_socket(&ustconsumer32_data,
3280 cmd_ctx->session->ust_session->consumer);
3281 if (ret < 0) {
3282 goto error;
3283 }
3284 }
3285 break;
3286 }
3287 default:
3288 break;
3289 }
3290 skip_domain:
3291
3292 /* Validate consumer daemon state when start/stop trace command */
3293 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3294 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3295 switch (cmd_ctx->lsm->domain.type) {
3296 case LTTNG_DOMAIN_JUL:
3297 case LTTNG_DOMAIN_LOG4J:
3298 case LTTNG_DOMAIN_PYTHON:
3299 case LTTNG_DOMAIN_UST:
3300 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3301 ret = LTTNG_ERR_NO_USTCONSUMERD;
3302 goto error;
3303 }
3304 break;
3305 case LTTNG_DOMAIN_KERNEL:
3306 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3307 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3308 goto error;
3309 }
3310 break;
3311 }
3312 }
3313
3314 /*
3315 * Check that the UID or GID match that of the tracing session.
3316 * The root user can interact with all sessions.
3317 */
3318 if (need_tracing_session) {
3319 if (!session_access_ok(cmd_ctx->session,
3320 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3321 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3322 ret = LTTNG_ERR_EPERM;
3323 goto error;
3324 }
3325 }
3326
3327 /*
3328 * Send relayd information to consumer as soon as we have a domain and a
3329 * session defined.
3330 */
3331 if (cmd_ctx->session && need_domain) {
3332 /*
3333 * Setup relayd if not done yet. If the relayd information was already
3334 * sent to the consumer, this call will gracefully return.
3335 */
3336 ret = cmd_setup_relayd(cmd_ctx->session);
3337 if (ret != LTTNG_OK) {
3338 goto error;
3339 }
3340 }
3341
3342 /* Process by command type */
3343 switch (cmd_ctx->lsm->cmd_type) {
3344 case LTTNG_ADD_CONTEXT:
3345 {
3346 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3347 cmd_ctx->lsm->u.context.channel_name,
3348 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3349 break;
3350 }
3351 case LTTNG_DISABLE_CHANNEL:
3352 {
3353 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3354 cmd_ctx->lsm->u.disable.channel_name);
3355 break;
3356 }
3357 case LTTNG_DISABLE_EVENT:
3358 {
3359
3360 /*
3361 * FIXME: handle filter; for now we just receive the filter's
3362 * bytecode along with the filter expression which are sent by
3363 * liblttng-ctl and discard them.
3364 *
3365 * This fixes an issue where the client may block while sending
3366 * the filter payload and encounter an error because the session
3367 * daemon closes the socket without ever handling this data.
3368 */
3369 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3370 cmd_ctx->lsm->u.disable.bytecode_len;
3371
3372 if (count) {
3373 char data[LTTNG_FILTER_MAX_LEN];
3374
3375 DBG("Discarding disable event command payload of size %zu", count);
3376 while (count) {
3377 ret = lttcomm_recv_unix_sock(sock, data,
3378 count > sizeof(data) ? sizeof(data) : count);
3379 if (ret < 0) {
3380 goto error;
3381 }
3382
3383 count -= (size_t) ret;
3384 }
3385 }
3386 /* FIXME: passing packed structure to non-packed pointer */
3387 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3388 cmd_ctx->lsm->u.disable.channel_name,
3389 &cmd_ctx->lsm->u.disable.event);
3390 break;
3391 }
3392 case LTTNG_ENABLE_CHANNEL:
3393 {
3394 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3395 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3396 break;
3397 }
3398 case LTTNG_TRACK_PID:
3399 {
3400 ret = cmd_track_pid(cmd_ctx->session,
3401 cmd_ctx->lsm->domain.type,
3402 cmd_ctx->lsm->u.pid_tracker.pid);
3403 break;
3404 }
3405 case LTTNG_UNTRACK_PID:
3406 {
3407 ret = cmd_untrack_pid(cmd_ctx->session,
3408 cmd_ctx->lsm->domain.type,
3409 cmd_ctx->lsm->u.pid_tracker.pid);
3410 break;
3411 }
3412 case LTTNG_ENABLE_EVENT:
3413 {
3414 struct lttng_event_exclusion *exclusion = NULL;
3415 struct lttng_filter_bytecode *bytecode = NULL;
3416 char *filter_expression = NULL;
3417
3418 /* Handle exclusion events and receive it from the client. */
3419 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3420 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3421
3422 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3423 (count * LTTNG_SYMBOL_NAME_LEN));
3424 if (!exclusion) {
3425 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3426 goto error;
3427 }
3428
3429 DBG("Receiving var len exclusion event list from client ...");
3430 exclusion->count = count;
3431 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3432 count * LTTNG_SYMBOL_NAME_LEN);
3433 if (ret <= 0) {
3434 DBG("Nothing recv() from client var len data... continuing");
3435 *sock_error = 1;
3436 free(exclusion);
3437 ret = LTTNG_ERR_EXCLUSION_INVAL;
3438 goto error;
3439 }
3440 }
3441
3442 /* Get filter expression from client. */
3443 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3444 size_t expression_len =
3445 cmd_ctx->lsm->u.enable.expression_len;
3446
3447 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3448 ret = LTTNG_ERR_FILTER_INVAL;
3449 free(exclusion);
3450 goto error;
3451 }
3452
3453 filter_expression = zmalloc(expression_len);
3454 if (!filter_expression) {
3455 free(exclusion);
3456 ret = LTTNG_ERR_FILTER_NOMEM;
3457 goto error;
3458 }
3459
3460 /* Receive var. len. data */
3461 DBG("Receiving var len filter's expression from client ...");
3462 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3463 expression_len);
3464 if (ret <= 0) {
3465 DBG("Nothing recv() from client car len data... continuing");
3466 *sock_error = 1;
3467 free(filter_expression);
3468 free(exclusion);
3469 ret = LTTNG_ERR_FILTER_INVAL;
3470 goto error;
3471 }
3472 }
3473
3474 /* Handle filter and get bytecode from client. */
3475 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3476 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3477
3478 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3479 ret = LTTNG_ERR_FILTER_INVAL;
3480 free(filter_expression);
3481 free(exclusion);
3482 goto error;
3483 }
3484
3485 bytecode = zmalloc(bytecode_len);
3486 if (!bytecode) {
3487 free(filter_expression);
3488 free(exclusion);
3489 ret = LTTNG_ERR_FILTER_NOMEM;
3490 goto error;
3491 }
3492
3493 /* Receive var. len. data */
3494 DBG("Receiving var len filter's bytecode from client ...");
3495 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3496 if (ret <= 0) {
3497 DBG("Nothing recv() from client car len data... continuing");
3498 *sock_error = 1;
3499 free(filter_expression);
3500 free(bytecode);
3501 free(exclusion);
3502 ret = LTTNG_ERR_FILTER_INVAL;
3503 goto error;
3504 }
3505
3506 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3507 free(filter_expression);
3508 free(bytecode);
3509 free(exclusion);
3510 ret = LTTNG_ERR_FILTER_INVAL;
3511 goto error;
3512 }
3513 }
3514
3515 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3516 cmd_ctx->lsm->u.enable.channel_name,
3517 &cmd_ctx->lsm->u.enable.event,
3518 filter_expression, bytecode, exclusion,
3519 kernel_poll_pipe[1]);
3520 break;
3521 }
3522 case LTTNG_LIST_TRACEPOINTS:
3523 {
3524 struct lttng_event *events;
3525 ssize_t nb_events;
3526
3527 session_lock_list();
3528 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3529 session_unlock_list();
3530 if (nb_events < 0) {
3531 /* Return value is a negative lttng_error_code. */
3532 ret = -nb_events;
3533 goto error;
3534 }
3535
3536 /*
3537 * Setup lttng message with payload size set to the event list size in
3538 * bytes and then copy list into the llm payload.
3539 */
3540 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3541 if (ret < 0) {
3542 free(events);
3543 goto setup_error;
3544 }
3545
3546 /* Copy event list into message payload */
3547 memcpy(cmd_ctx->llm->payload, events,
3548 sizeof(struct lttng_event) * nb_events);
3549
3550 free(events);
3551
3552 ret = LTTNG_OK;
3553 break;
3554 }
3555 case LTTNG_LIST_TRACEPOINT_FIELDS:
3556 {
3557 struct lttng_event_field *fields;
3558 ssize_t nb_fields;
3559
3560 session_lock_list();
3561 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3562 &fields);
3563 session_unlock_list();
3564 if (nb_fields < 0) {
3565 /* Return value is a negative lttng_error_code. */
3566 ret = -nb_fields;
3567 goto error;
3568 }
3569
3570 /*
3571 * Setup lttng message with payload size set to the event list size in
3572 * bytes and then copy list into the llm payload.
3573 */
3574 ret = setup_lttng_msg(cmd_ctx,
3575 sizeof(struct lttng_event_field) * nb_fields);
3576 if (ret < 0) {
3577 free(fields);
3578 goto setup_error;
3579 }
3580
3581 /* Copy event list into message payload */
3582 memcpy(cmd_ctx->llm->payload, fields,
3583 sizeof(struct lttng_event_field) * nb_fields);
3584
3585 free(fields);
3586
3587 ret = LTTNG_OK;
3588 break;
3589 }
3590 case LTTNG_LIST_SYSCALLS:
3591 {
3592 struct lttng_event *events;
3593 ssize_t nb_events;
3594
3595 nb_events = cmd_list_syscalls(&events);
3596 if (nb_events < 0) {
3597 /* Return value is a negative lttng_error_code. */
3598 ret = -nb_events;
3599 goto error;
3600 }
3601
3602 /*
3603 * Setup lttng message with payload size set to the event list size in
3604 * bytes and then copy list into the llm payload.
3605 */
3606 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3607 if (ret < 0) {
3608 free(events);
3609 goto setup_error;
3610 }
3611
3612 /* Copy event list into message payload */
3613 memcpy(cmd_ctx->llm->payload, events,
3614 sizeof(struct lttng_event) * nb_events);
3615
3616 free(events);
3617
3618 ret = LTTNG_OK;
3619 break;
3620 }
3621 case LTTNG_LIST_TRACKER_PIDS:
3622 {
3623 int32_t *pids = NULL;
3624 ssize_t nr_pids;
3625
3626 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3627 cmd_ctx->lsm->domain.type, &pids);
3628 if (nr_pids < 0) {
3629 /* Return value is a negative lttng_error_code. */
3630 ret = -nr_pids;
3631 goto error;
3632 }
3633
3634 /*
3635 * Setup lttng message with payload size set to the event list size in
3636 * bytes and then copy list into the llm payload.
3637 */
3638 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3639 if (ret < 0) {
3640 free(pids);
3641 goto setup_error;
3642 }
3643
3644 /* Copy event list into message payload */
3645 memcpy(cmd_ctx->llm->payload, pids,
3646 sizeof(int) * nr_pids);
3647
3648 free(pids);
3649
3650 ret = LTTNG_OK;
3651 break;
3652 }
3653 case LTTNG_SET_CONSUMER_URI:
3654 {
3655 size_t nb_uri, len;
3656 struct lttng_uri *uris;
3657
3658 nb_uri = cmd_ctx->lsm->u.uri.size;
3659 len = nb_uri * sizeof(struct lttng_uri);
3660
3661 if (nb_uri == 0) {
3662 ret = LTTNG_ERR_INVALID;
3663 goto error;
3664 }
3665
3666 uris = zmalloc(len);
3667 if (uris == NULL) {
3668 ret = LTTNG_ERR_FATAL;
3669 goto error;
3670 }
3671
3672 /* Receive variable len data */
3673 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3674 ret = lttcomm_recv_unix_sock(sock, uris, len);
3675 if (ret <= 0) {
3676 DBG("No URIs received from client... continuing");
3677 *sock_error = 1;
3678 ret = LTTNG_ERR_SESSION_FAIL;
3679 free(uris);
3680 goto error;
3681 }
3682
3683 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3684 free(uris);
3685 if (ret != LTTNG_OK) {
3686 goto error;
3687 }
3688
3689
3690 break;
3691 }
3692 case LTTNG_START_TRACE:
3693 {
3694 ret = cmd_start_trace(cmd_ctx->session);
3695 break;
3696 }
3697 case LTTNG_STOP_TRACE:
3698 {
3699 ret = cmd_stop_trace(cmd_ctx->session);
3700 break;
3701 }
3702 case LTTNG_CREATE_SESSION:
3703 {
3704 size_t nb_uri, len;
3705 struct lttng_uri *uris = NULL;
3706
3707 nb_uri = cmd_ctx->lsm->u.uri.size;
3708 len = nb_uri * sizeof(struct lttng_uri);
3709
3710 if (nb_uri > 0) {
3711 uris = zmalloc(len);
3712 if (uris == NULL) {
3713 ret = LTTNG_ERR_FATAL;
3714 goto error;
3715 }
3716
3717 /* Receive variable len data */
3718 DBG("Waiting for %zu URIs from client ...", nb_uri);
3719 ret = lttcomm_recv_unix_sock(sock, uris, len);
3720 if (ret <= 0) {
3721 DBG("No URIs received from client... continuing");
3722 *sock_error = 1;
3723 ret = LTTNG_ERR_SESSION_FAIL;
3724 free(uris);
3725 goto error;
3726 }
3727
3728 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3729 DBG("Creating session with ONE network URI is a bad call");
3730 ret = LTTNG_ERR_SESSION_FAIL;
3731 free(uris);
3732 goto error;
3733 }
3734 }
3735
3736 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3737 &cmd_ctx->creds, 0);
3738
3739 free(uris);
3740
3741 break;
3742 }
3743 case LTTNG_DESTROY_SESSION:
3744 {
3745 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3746
3747 /* Set session to NULL so we do not unlock it after free. */
3748 cmd_ctx->session = NULL;
3749 break;
3750 }
3751 case LTTNG_LIST_DOMAINS:
3752 {
3753 ssize_t nb_dom;
3754 struct lttng_domain *domains = NULL;
3755
3756 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3757 if (nb_dom < 0) {
3758 /* Return value is a negative lttng_error_code. */
3759 ret = -nb_dom;
3760 goto error;
3761 }
3762
3763 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3764 if (ret < 0) {
3765 free(domains);
3766 goto setup_error;
3767 }
3768
3769 /* Copy event list into message payload */
3770 memcpy(cmd_ctx->llm->payload, domains,
3771 nb_dom * sizeof(struct lttng_domain));
3772
3773 free(domains);
3774
3775 ret = LTTNG_OK;
3776 break;
3777 }
3778 case LTTNG_LIST_CHANNELS:
3779 {
3780 int nb_chan;
3781 struct lttng_channel *channels = NULL;
3782
3783 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3784 cmd_ctx->session, &channels);
3785 if (nb_chan < 0) {
3786 /* Return value is a negative lttng_error_code. */
3787 ret = -nb_chan;
3788 goto error;
3789 }
3790
3791 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3792 if (ret < 0) {
3793 free(channels);
3794 goto setup_error;
3795 }
3796
3797 /* Copy event list into message payload */
3798 memcpy(cmd_ctx->llm->payload, channels,
3799 nb_chan * sizeof(struct lttng_channel));
3800
3801 free(channels);
3802
3803 ret = LTTNG_OK;
3804 break;
3805 }
3806 case LTTNG_LIST_EVENTS:
3807 {
3808 ssize_t nb_event;
3809 struct lttng_event *events = NULL;
3810
3811 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3812 cmd_ctx->lsm->u.list.channel_name, &events);
3813 if (nb_event < 0) {
3814 /* Return value is a negative lttng_error_code. */
3815 ret = -nb_event;
3816 goto error;
3817 }
3818
3819 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3820 if (ret < 0) {
3821 free(events);
3822 goto setup_error;
3823 }
3824
3825 /* Copy event list into message payload */
3826 memcpy(cmd_ctx->llm->payload, events,
3827 nb_event * sizeof(struct lttng_event));
3828
3829 free(events);
3830
3831 ret = LTTNG_OK;
3832 break;
3833 }
3834 case LTTNG_LIST_SESSIONS:
3835 {
3836 unsigned int nr_sessions;
3837
3838 session_lock_list();
3839 nr_sessions = lttng_sessions_count(
3840 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3841 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3842
3843 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3844 if (ret < 0) {
3845 session_unlock_list();
3846 goto setup_error;
3847 }
3848
3849 /* Filled the session array */
3850 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3851 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3852 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3853
3854 session_unlock_list();
3855
3856 ret = LTTNG_OK;
3857 break;
3858 }
3859 case LTTNG_CALIBRATE:
3860 {
3861 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3862 &cmd_ctx->lsm->u.calibrate);
3863 break;
3864 }
3865 case LTTNG_REGISTER_CONSUMER:
3866 {
3867 struct consumer_data *cdata;
3868
3869 switch (cmd_ctx->lsm->domain.type) {
3870 case LTTNG_DOMAIN_KERNEL:
3871 cdata = &kconsumer_data;
3872 break;
3873 default:
3874 ret = LTTNG_ERR_UND;
3875 goto error;
3876 }
3877
3878 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3879 cmd_ctx->lsm->u.reg.path, cdata);
3880 break;
3881 }
3882 case LTTNG_DATA_PENDING:
3883 {
3884 int pending_ret;
3885
3886 /* 1 byte to return whether or not data is pending */
3887 ret = setup_lttng_msg(cmd_ctx, 1);
3888 if (ret < 0) {
3889 goto setup_error;
3890 }
3891
3892 pending_ret = cmd_data_pending(cmd_ctx->session);
3893 /*
3894 * FIXME
3895 *
3896 * This function may returns 0 or 1 to indicate whether or not
3897 * there is data pending. In case of error, it should return an
3898 * LTTNG_ERR code. However, some code paths may still return
3899 * a nondescript error code, which we handle by returning an
3900 * "unknown" error.
3901 */
3902 if (pending_ret == 0 || pending_ret == 1) {
3903 ret = LTTNG_OK;
3904 } else if (pending_ret < 0) {
3905 ret = LTTNG_ERR_UNK;
3906 goto setup_error;
3907 } else {
3908 ret = pending_ret;
3909 goto setup_error;
3910 }
3911
3912 *cmd_ctx->llm->payload = (uint8_t) pending_ret;
3913 break;
3914 }
3915 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3916 {
3917 struct lttcomm_lttng_output_id reply;
3918
3919 ret = cmd_snapshot_add_output(cmd_ctx->session,
3920 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3921 if (ret != LTTNG_OK) {
3922 goto error;
3923 }
3924
3925 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3926 if (ret < 0) {
3927 goto setup_error;
3928 }
3929
3930 /* Copy output list into message payload */
3931 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3932 ret = LTTNG_OK;
3933 break;
3934 }
3935 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3936 {
3937 ret = cmd_snapshot_del_output(cmd_ctx->session,
3938 &cmd_ctx->lsm->u.snapshot_output.output);
3939 break;
3940 }
3941 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3942 {
3943 ssize_t nb_output;
3944 struct lttng_snapshot_output *outputs = NULL;
3945
3946 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3947 if (nb_output < 0) {
3948 ret = -nb_output;
3949 goto error;
3950 }
3951
3952 ret = setup_lttng_msg(cmd_ctx,
3953 nb_output * sizeof(struct lttng_snapshot_output));
3954 if (ret < 0) {
3955 free(outputs);
3956 goto setup_error;
3957 }
3958
3959 if (outputs) {
3960 /* Copy output list into message payload */
3961 memcpy(cmd_ctx->llm->payload, outputs,
3962 nb_output * sizeof(struct lttng_snapshot_output));
3963 free(outputs);
3964 }
3965
3966 ret = LTTNG_OK;
3967 break;
3968 }
3969 case LTTNG_SNAPSHOT_RECORD:
3970 {
3971 ret = cmd_snapshot_record(cmd_ctx->session,
3972 &cmd_ctx->lsm->u.snapshot_record.output,
3973 cmd_ctx->lsm->u.snapshot_record.wait);
3974 break;
3975 }
3976 case LTTNG_CREATE_SESSION_SNAPSHOT:
3977 {
3978 size_t nb_uri, len;
3979 struct lttng_uri *uris = NULL;
3980
3981 nb_uri = cmd_ctx->lsm->u.uri.size;
3982 len = nb_uri * sizeof(struct lttng_uri);
3983
3984 if (nb_uri > 0) {
3985 uris = zmalloc(len);
3986 if (uris == NULL) {
3987 ret = LTTNG_ERR_FATAL;
3988 goto error;
3989 }
3990
3991 /* Receive variable len data */
3992 DBG("Waiting for %zu URIs from client ...", nb_uri);
3993 ret = lttcomm_recv_unix_sock(sock, uris, len);
3994 if (ret <= 0) {
3995 DBG("No URIs received from client... continuing");
3996 *sock_error = 1;
3997 ret = LTTNG_ERR_SESSION_FAIL;
3998 free(uris);
3999 goto error;
4000 }
4001
4002 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4003 DBG("Creating session with ONE network URI is a bad call");
4004 ret = LTTNG_ERR_SESSION_FAIL;
4005 free(uris);
4006 goto error;
4007 }
4008 }
4009
4010 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4011 nb_uri, &cmd_ctx->creds);
4012 free(uris);
4013 break;
4014 }
4015 case LTTNG_CREATE_SESSION_LIVE:
4016 {
4017 size_t nb_uri, len;
4018 struct lttng_uri *uris = NULL;
4019
4020 nb_uri = cmd_ctx->lsm->u.uri.size;
4021 len = nb_uri * sizeof(struct lttng_uri);
4022
4023 if (nb_uri > 0) {
4024 uris = zmalloc(len);
4025 if (uris == NULL) {
4026 ret = LTTNG_ERR_FATAL;
4027 goto error;
4028 }
4029
4030 /* Receive variable len data */
4031 DBG("Waiting for %zu URIs from client ...", nb_uri);
4032 ret = lttcomm_recv_unix_sock(sock, uris, len);
4033 if (ret <= 0) {
4034 DBG("No URIs received from client... continuing");
4035 *sock_error = 1;
4036 ret = LTTNG_ERR_SESSION_FAIL;
4037 free(uris);
4038 goto error;
4039 }
4040
4041 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4042 DBG("Creating session with ONE network URI is a bad call");
4043 ret = LTTNG_ERR_SESSION_FAIL;
4044 free(uris);
4045 goto error;
4046 }
4047 }
4048
4049 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4050 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4051 free(uris);
4052 break;
4053 }
4054 case LTTNG_SAVE_SESSION:
4055 {
4056 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4057 &cmd_ctx->creds);
4058 break;
4059 }
4060 case LTTNG_SET_SESSION_SHM_PATH:
4061 {
4062 ret = cmd_set_session_shm_path(cmd_ctx->session,
4063 cmd_ctx->lsm->u.set_shm_path.shm_path);
4064 break;
4065 }
4066 default:
4067 ret = LTTNG_ERR_UND;
4068 break;
4069 }
4070
4071 error:
4072 if (cmd_ctx->llm == NULL) {
4073 DBG("Missing llm structure. Allocating one.");
4074 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
4075 goto setup_error;
4076 }
4077 }
4078 /* Set return code */
4079 cmd_ctx->llm->ret_code = ret;
4080 setup_error:
4081 if (cmd_ctx->session) {
4082 session_unlock(cmd_ctx->session);
4083 }
4084 if (need_tracing_session) {
4085 session_unlock_list();
4086 }
4087 init_setup_error:
4088 assert(!rcu_read_ongoing());
4089 return ret;
4090 }
4091
4092 /*
4093 * Thread managing health check socket.
4094 */
4095 static void *thread_manage_health(void *data)
4096 {
4097 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4098 uint32_t revents, nb_fd;
4099 struct lttng_poll_event events;
4100 struct health_comm_msg msg;
4101 struct health_comm_reply reply;
4102
4103 DBG("[thread] Manage health check started");
4104
4105 rcu_register_thread();
4106
4107 /* We might hit an error path before this is created. */
4108 lttng_poll_init(&events);
4109
4110 /* Create unix socket */
4111 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4112 if (sock < 0) {
4113 ERR("Unable to create health check Unix socket");
4114 ret = -1;
4115 goto error;
4116 }
4117
4118 if (is_root) {
4119 /* lttng health client socket path permissions */
4120 ret = chown(health_unix_sock_path, 0,
4121 utils_get_group_id(tracing_group_name));
4122 if (ret < 0) {
4123 ERR("Unable to set group on %s", health_unix_sock_path);
4124 PERROR("chown");
4125 ret = -1;
4126 goto error;
4127 }
4128
4129 ret = chmod(health_unix_sock_path,
4130 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4131 if (ret < 0) {
4132 ERR("Unable to set permissions on %s", health_unix_sock_path);
4133 PERROR("chmod");
4134 ret = -1;
4135 goto error;
4136 }
4137 }
4138
4139 /*
4140 * Set the CLOEXEC flag. Return code is useless because either way, the
4141 * show must go on.
4142 */
4143 (void) utils_set_fd_cloexec(sock);
4144
4145 ret = lttcomm_listen_unix_sock(sock);
4146 if (ret < 0) {
4147 goto error;
4148 }
4149
4150 /*
4151 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4152 * more will be added to this poll set.
4153 */
4154 ret = sessiond_set_thread_pollset(&events, 2);
4155 if (ret < 0) {
4156 goto error;
4157 }
4158
4159 /* Add the application registration socket */
4160 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4161 if (ret < 0) {
4162 goto error;
4163 }
4164
4165 sessiond_notify_ready();
4166
4167 while (1) {
4168 DBG("Health check ready");
4169
4170 /* Inifinite blocking call, waiting for transmission */
4171 restart:
4172 ret = lttng_poll_wait(&events, -1);
4173 if (ret < 0) {
4174 /*
4175 * Restart interrupted system call.
4176 */
4177 if (errno == EINTR) {
4178 goto restart;
4179 }
4180 goto error;
4181 }
4182
4183 nb_fd = ret;
4184
4185 for (i = 0; i < nb_fd; i++) {
4186 /* Fetch once the poll data */
4187 revents = LTTNG_POLL_GETEV(&events, i);
4188 pollfd = LTTNG_POLL_GETFD(&events, i);
4189
4190 if (!revents) {
4191 /* No activity for this FD (poll implementation). */
4192 continue;
4193 }
4194
4195 /* Thread quit pipe has been closed. Killing thread. */
4196 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4197 if (ret) {
4198 err = 0;
4199 goto exit;
4200 }
4201
4202 /* Event on the registration socket */
4203 if (pollfd == sock) {
4204 if (revents & LPOLLIN) {
4205 continue;
4206 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4207 ERR("Health socket poll error");
4208 goto error;
4209 } else {
4210 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4211 goto error;
4212 }
4213 }
4214 }
4215
4216 new_sock = lttcomm_accept_unix_sock(sock);
4217 if (new_sock < 0) {
4218 goto error;
4219 }
4220
4221 /*
4222 * Set the CLOEXEC flag. Return code is useless because either way, the
4223 * show must go on.
4224 */
4225 (void) utils_set_fd_cloexec(new_sock);
4226
4227 DBG("Receiving data from client for health...");
4228 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4229 if (ret <= 0) {
4230 DBG("Nothing recv() from client... continuing");
4231 ret = close(new_sock);
4232 if (ret) {
4233 PERROR("close");
4234 }
4235 new_sock = -1;
4236 continue;
4237 }
4238
4239 rcu_thread_online();
4240
4241 memset(&reply, 0, sizeof(reply));
4242 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4243 /*
4244 * health_check_state returns 0 if health is
4245 * bad.
4246 */
4247 if (!health_check_state(health_sessiond, i)) {
4248 reply.ret_code |= 1ULL << i;
4249 }
4250 }
4251
4252 DBG2("Health check return value %" PRIx64, reply.ret_code);
4253
4254 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4255 if (ret < 0) {
4256 ERR("Failed to send health data back to client");
4257 }
4258
4259 /* End of transmission */
4260 ret = close(new_sock);
4261 if (ret) {
4262 PERROR("close");
4263 }
4264 new_sock = -1;
4265 }
4266
4267 exit:
4268 error:
4269 if (err) {
4270 ERR("Health error occurred in %s", __func__);
4271 }
4272 DBG("Health check thread dying");
4273 unlink(health_unix_sock_path);
4274 if (sock >= 0) {
4275 ret = close(sock);
4276 if (ret) {
4277 PERROR("close");
4278 }
4279 }
4280
4281 lttng_poll_clean(&events);
4282
4283 rcu_unregister_thread();
4284 return NULL;
4285 }
4286
4287 /*
4288 * This thread manage all clients request using the unix client socket for
4289 * communication.
4290 */
4291 static void *thread_manage_clients(void *data)
4292 {
4293 int sock = -1, ret, i, pollfd, err = -1;
4294 int sock_error;
4295 uint32_t revents, nb_fd;
4296 struct command_ctx *cmd_ctx = NULL;
4297 struct lttng_poll_event events;
4298
4299 DBG("[thread] Manage client started");
4300
4301 rcu_register_thread();
4302
4303 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4304
4305 health_code_update();
4306
4307 ret = lttcomm_listen_unix_sock(client_sock);
4308 if (ret < 0) {
4309 goto error_listen;
4310 }
4311
4312 /*
4313 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4314 * more will be added to this poll set.
4315 */
4316 ret = sessiond_set_thread_pollset(&events, 2);
4317 if (ret < 0) {
4318 goto error_create_poll;
4319 }
4320
4321 /* Add the application registration socket */
4322 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4323 if (ret < 0) {
4324 goto error;
4325 }
4326
4327 sessiond_notify_ready();
4328 ret = sem_post(&load_info->message_thread_ready);
4329 if (ret) {
4330 PERROR("sem_post message_thread_ready");
4331 goto error;
4332 }
4333
4334 /* This testpoint is after we signal readiness to the parent. */
4335 if (testpoint(sessiond_thread_manage_clients)) {
4336 goto error;
4337 }
4338
4339 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4340 goto error;
4341 }
4342
4343 health_code_update();
4344
4345 while (1) {
4346 DBG("Accepting client command ...");
4347
4348 /* Inifinite blocking call, waiting for transmission */
4349 restart:
4350 health_poll_entry();
4351 ret = lttng_poll_wait(&events, -1);
4352 health_poll_exit();
4353 if (ret < 0) {
4354 /*
4355 * Restart interrupted system call.
4356 */
4357 if (errno == EINTR) {
4358 goto restart;
4359 }
4360 goto error;
4361 }
4362
4363 nb_fd = ret;
4364
4365 for (i = 0; i < nb_fd; i++) {
4366 /* Fetch once the poll data */
4367 revents = LTTNG_POLL_GETEV(&events, i);
4368 pollfd = LTTNG_POLL_GETFD(&events, i);
4369
4370 health_code_update();
4371
4372 if (!revents) {
4373 /* No activity for this FD (poll implementation). */
4374 continue;
4375 }
4376
4377 /* Thread quit pipe has been closed. Killing thread. */
4378 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4379 if (ret) {
4380 err = 0;
4381 goto exit;
4382 }
4383
4384 /* Event on the registration socket */
4385 if (pollfd == client_sock) {
4386 if (revents & LPOLLIN) {
4387 continue;
4388 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4389 ERR("Client socket poll error");
4390 goto error;
4391 } else {
4392 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4393 goto error;
4394 }
4395 }
4396 }
4397
4398 DBG("Wait for client response");
4399
4400 health_code_update();
4401
4402 sock = lttcomm_accept_unix_sock(client_sock);
4403 if (sock < 0) {
4404 goto error;
4405 }
4406
4407 /*
4408 * Set the CLOEXEC flag. Return code is useless because either way, the
4409 * show must go on.
4410 */
4411 (void) utils_set_fd_cloexec(sock);
4412
4413 /* Set socket option for credentials retrieval */
4414 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4415 if (ret < 0) {
4416 goto error;
4417 }
4418
4419 /* Allocate context command to process the client request */
4420 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4421 if (cmd_ctx == NULL) {
4422 PERROR("zmalloc cmd_ctx");
4423 goto error;
4424 }
4425
4426 /* Allocate data buffer for reception */
4427 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4428 if (cmd_ctx->lsm == NULL) {
4429 PERROR("zmalloc cmd_ctx->lsm");
4430 goto error;
4431 }
4432
4433 cmd_ctx->llm = NULL;
4434 cmd_ctx->session = NULL;
4435
4436 health_code_update();
4437
4438 /*
4439 * Data is received from the lttng client. The struct
4440 * lttcomm_session_msg (lsm) contains the command and data request of
4441 * the client.
4442 */
4443 DBG("Receiving data from client ...");
4444 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4445 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4446 if (ret <= 0) {
4447 DBG("Nothing recv() from client... continuing");
4448 ret = close(sock);
4449 if (ret) {
4450 PERROR("close");
4451 }
4452 sock = -1;
4453 clean_command_ctx(&cmd_ctx);
4454 continue;
4455 }
4456
4457 health_code_update();
4458
4459 // TODO: Validate cmd_ctx including sanity check for
4460 // security purpose.
4461
4462 rcu_thread_online();
4463 /*
4464 * This function dispatch the work to the kernel or userspace tracer
4465 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4466 * informations for the client. The command context struct contains
4467 * everything this function may needs.
4468 */
4469 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4470 rcu_thread_offline();
4471 if (ret < 0) {
4472 ret = close(sock);
4473 if (ret) {
4474 PERROR("close");
4475 }
4476 sock = -1;
4477 /*
4478 * TODO: Inform client somehow of the fatal error. At
4479 * this point, ret < 0 means that a zmalloc failed
4480 * (ENOMEM). Error detected but still accept
4481 * command, unless a socket error has been
4482 * detected.
4483 */
4484 clean_command_ctx(&cmd_ctx);
4485 continue;
4486 }
4487
4488 health_code_update();
4489
4490 DBG("Sending response (size: %d, retcode: %s (%d))",
4491 cmd_ctx->lttng_msg_size,
4492 lttng_strerror(-cmd_ctx->llm->ret_code),
4493 cmd_ctx->llm->ret_code);
4494 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4495 if (ret < 0) {
4496 ERR("Failed to send data back to client");
4497 }
4498
4499 /* End of transmission */
4500 ret = close(sock);
4501 if (ret) {
4502 PERROR("close");
4503 }
4504 sock = -1;
4505
4506 clean_command_ctx(&cmd_ctx);
4507
4508 health_code_update();
4509 }
4510
4511 exit:
4512 error:
4513 if (sock >= 0) {
4514 ret = close(sock);
4515 if (ret) {
4516 PERROR("close");
4517 }
4518 }
4519
4520 lttng_poll_clean(&events);
4521 clean_command_ctx(&cmd_ctx);
4522
4523 error_listen:
4524 error_create_poll:
4525 unlink(client_unix_sock_path);
4526 if (client_sock >= 0) {
4527 ret = close(client_sock);
4528 if (ret) {
4529 PERROR("close");
4530 }
4531 }
4532
4533 if (err) {
4534 health_error();
4535 ERR("Health error occurred in %s", __func__);
4536 }
4537
4538 health_unregister(health_sessiond);
4539
4540 DBG("Client thread dying");
4541
4542 rcu_unregister_thread();
4543
4544 /*
4545 * Since we are creating the consumer threads, we own them, so we need
4546 * to join them before our thread exits.
4547 */
4548 ret = join_consumer_thread(&kconsumer_data);
4549 if (ret) {
4550 errno = ret;
4551 PERROR("join_consumer");
4552 }
4553
4554 ret = join_consumer_thread(&ustconsumer32_data);
4555 if (ret) {
4556 errno = ret;
4557 PERROR("join_consumer ust32");
4558 }
4559
4560 ret = join_consumer_thread(&ustconsumer64_data);
4561 if (ret) {
4562 errno = ret;
4563 PERROR("join_consumer ust64");
4564 }
4565 return NULL;
4566 }
4567
4568
4569 /*
4570 * usage function on stderr
4571 */
4572 static void usage(void)
4573 {
4574 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4575 fprintf(stderr, " -h, --help Display this usage.\n");
4576 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4577 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4578 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4579 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4580 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4581 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4582 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4583 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4584 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4585 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4586 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4587 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4588 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4589 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4590 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4591 fprintf(stderr, " -V, --version Show version number.\n");
4592 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4593 fprintf(stderr, " -q, --quiet No output at all.\n");
4594 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4595 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4596 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4597 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4598 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4599 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4600 fprintf(stderr, " -l --load PATH Load session configuration\n");
4601 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4602 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4603 }
4604
4605 static int string_match(const char *str1, const char *str2)
4606 {
4607 return (str1 && str2) && !strcmp(str1, str2);
4608 }
4609
4610 /*
4611 * Take an option from the getopt output and set it in the right variable to be
4612 * used later.
4613 *
4614 * Return 0 on success else a negative value.
4615 */
4616 static int set_option(int opt, const char *arg, const char *optname)
4617 {
4618 int ret = 0;
4619
4620 if (arg && arg[0] == '\0') {
4621 /*
4622 * This only happens if the value is read from daemon config
4623 * file. This means the option requires an argument and the
4624 * configuration file contains a line such as:
4625 * my_option =
4626 */
4627 ret = -EINVAL;
4628 goto end;
4629 }
4630
4631 if (string_match(optname, "client-sock") || opt == 'c') {
4632 if (lttng_is_setuid_setgid()) {
4633 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4634 "-c, --client-sock");
4635 } else {
4636 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4637 }
4638 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4639 if (lttng_is_setuid_setgid()) {
4640 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4641 "-a, --apps-sock");
4642 } else {
4643 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4644 }
4645 } else if (string_match(optname, "daemonize") || opt == 'd') {
4646 opt_daemon = 1;
4647 } else if (string_match(optname, "background") || opt == 'b') {
4648 opt_background = 1;
4649 } else if (string_match(optname, "group") || opt == 'g') {
4650 if (lttng_is_setuid_setgid()) {
4651 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4652 "-g, --group");
4653 } else {
4654 /*
4655 * If the override option is set, the pointer points to a
4656 * *non* const thus freeing it even though the variable type is
4657 * set to const.
4658 */
4659 if (tracing_group_name_override) {
4660 free((void *) tracing_group_name);
4661 }
4662 tracing_group_name = strdup(arg);
4663 if (!tracing_group_name) {
4664 PERROR("strdup");
4665 ret = -ENOMEM;
4666 }
4667 tracing_group_name_override = 1;
4668 }
4669 } else if (string_match(optname, "help") || opt == 'h') {
4670 usage();
4671 exit(EXIT_SUCCESS);
4672 } else if (string_match(optname, "version") || opt == 'V') {
4673 fprintf(stdout, "%s\n", VERSION);
4674 exit(EXIT_SUCCESS);
4675 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4676 opt_sig_parent = 1;
4677 } else if (string_match(optname, "kconsumerd-err-sock")) {
4678 if (lttng_is_setuid_setgid()) {
4679 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4680 "--kconsumerd-err-sock");
4681 } else {
4682 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4683 }
4684 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4685 if (lttng_is_setuid_setgid()) {
4686 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4687 "--kconsumerd-cmd-sock");
4688 } else {
4689 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4690 }
4691 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4692 if (lttng_is_setuid_setgid()) {
4693 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4694 "--ustconsumerd64-err-sock");
4695 } else {
4696 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4697 }
4698 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4699 if (lttng_is_setuid_setgid()) {
4700 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4701 "--ustconsumerd64-cmd-sock");
4702 } else {
4703 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4704 }
4705 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4706 if (lttng_is_setuid_setgid()) {
4707 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4708 "--ustconsumerd32-err-sock");
4709 } else {
4710 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4711 }
4712 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4713 if (lttng_is_setuid_setgid()) {
4714 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4715 "--ustconsumerd32-cmd-sock");
4716 } else {
4717 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4718 }
4719 } else if (string_match(optname, "no-kernel")) {
4720 opt_no_kernel = 1;
4721 } else if (string_match(optname, "quiet") || opt == 'q') {
4722 lttng_opt_quiet = 1;
4723 } else if (string_match(optname, "verbose") || opt == 'v') {
4724 /* Verbose level can increase using multiple -v */
4725 if (arg) {
4726 /* Value obtained from config file */
4727 lttng_opt_verbose = config_parse_value(arg);
4728 } else {
4729 /* -v used on command line */
4730 lttng_opt_verbose++;
4731 }
4732 /* Clamp value to [0, 3] */
4733 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4734 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4735 } else if (string_match(optname, "verbose-consumer")) {
4736 if (arg) {
4737 opt_verbose_consumer = config_parse_value(arg);
4738 } else {
4739 opt_verbose_consumer += 1;
4740 }
4741 } else if (string_match(optname, "consumerd32-path")) {
4742 if (lttng_is_setuid_setgid()) {
4743 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4744 "--consumerd32-path");
4745 } else {
4746 if (consumerd32_bin_override) {
4747 free((void *) consumerd32_bin);
4748 }
4749 consumerd32_bin = strdup(arg);
4750 if (!consumerd32_bin) {
4751 PERROR("strdup");
4752 ret = -ENOMEM;
4753 }
4754 consumerd32_bin_override = 1;
4755 }
4756 } else if (string_match(optname, "consumerd32-libdir")) {
4757 if (lttng_is_setuid_setgid()) {
4758 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4759 "--consumerd32-libdir");
4760 } else {
4761 if (consumerd32_libdir_override) {
4762 free((void *) consumerd32_libdir);
4763 }
4764 consumerd32_libdir = strdup(arg);
4765 if (!consumerd32_libdir) {
4766 PERROR("strdup");
4767 ret = -ENOMEM;
4768 }
4769 consumerd32_libdir_override = 1;
4770 }
4771 } else if (string_match(optname, "consumerd64-path")) {
4772 if (lttng_is_setuid_setgid()) {
4773 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4774 "--consumerd64-path");
4775 } else {
4776 if (consumerd64_bin_override) {
4777 free((void *) consumerd64_bin);
4778 }
4779 consumerd64_bin = strdup(arg);
4780 if (!consumerd64_bin) {
4781 PERROR("strdup");
4782 ret = -ENOMEM;
4783 }
4784 consumerd64_bin_override = 1;
4785 }
4786 } else if (string_match(optname, "consumerd64-libdir")) {
4787 if (lttng_is_setuid_setgid()) {
4788 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4789 "--consumerd64-libdir");
4790 } else {
4791 if (consumerd64_libdir_override) {
4792 free((void *) consumerd64_libdir);
4793 }
4794 consumerd64_libdir = strdup(arg);
4795 if (!consumerd64_libdir) {
4796 PERROR("strdup");
4797 ret = -ENOMEM;
4798 }
4799 consumerd64_libdir_override = 1;
4800 }
4801 } else if (string_match(optname, "pidfile") || opt == 'p') {
4802 if (lttng_is_setuid_setgid()) {
4803 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4804 "-p, --pidfile");
4805 } else {
4806 free(opt_pidfile);
4807 opt_pidfile = strdup(arg);
4808 if (!opt_pidfile) {
4809 PERROR("strdup");
4810 ret = -ENOMEM;
4811 }
4812 }
4813 } else if (string_match(optname, "agent-tcp-port")) {
4814 if (lttng_is_setuid_setgid()) {
4815 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4816 "--agent-tcp-port");
4817 } else {
4818 unsigned long v;
4819
4820 if (!arg) {
4821 ret = -EINVAL;
4822 goto end;
4823 }
4824 errno = 0;
4825 v = strtoul(arg, NULL, 0);
4826 if (errno != 0 || !isdigit(arg[0])) {
4827 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4828 return -1;
4829 }
4830 if (v == 0 || v >= 65535) {
4831 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4832 return -1;
4833 }
4834 agent_tcp_port = (uint32_t) v;
4835 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4836 }
4837 } else if (string_match(optname, "load") || opt == 'l') {
4838 if (lttng_is_setuid_setgid()) {
4839 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4840 "-l, --load");
4841 } else {
4842 free(opt_load_session_path);
4843 opt_load_session_path = strdup(arg);
4844 if (!opt_load_session_path) {
4845 PERROR("strdup");
4846 ret = -ENOMEM;
4847 }
4848 }
4849 } else if (string_match(optname, "kmod-probes")) {
4850 if (lttng_is_setuid_setgid()) {
4851 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4852 "--kmod-probes");
4853 } else {
4854 free(kmod_probes_list);
4855 kmod_probes_list = strdup(arg);
4856 if (!kmod_probes_list) {
4857 PERROR("strdup");
4858 ret = -ENOMEM;
4859 }
4860 }
4861 } else if (string_match(optname, "extra-kmod-probes")) {
4862 if (lttng_is_setuid_setgid()) {
4863 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4864 "--extra-kmod-probes");
4865 } else {
4866 free(kmod_extra_probes_list);
4867 kmod_extra_probes_list = strdup(arg);
4868 if (!kmod_extra_probes_list) {
4869 PERROR("strdup");
4870 ret = -ENOMEM;
4871 }
4872 }
4873 } else if (string_match(optname, "config") || opt == 'f') {
4874 /* This is handled in set_options() thus silent skip. */
4875 goto end;
4876 } else {
4877 /* Unknown option or other error.
4878 * Error is printed by getopt, just return */
4879 ret = -1;
4880 }
4881
4882 end:
4883 if (ret == -EINVAL) {
4884 const char *opt_name = "unknown";
4885 int i;
4886
4887 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4888 i++) {
4889 if (opt == long_options[i].val) {
4890 opt_name = long_options[i].name;
4891 break;
4892 }
4893 }
4894
4895 WARN("Invalid argument provided for option \"%s\", using default value.",
4896 opt_name);
4897 }
4898
4899 return ret;
4900 }
4901
4902 /*
4903 * config_entry_handler_cb used to handle options read from a config file.
4904 * See config_entry_handler_cb comment in common/config/config.h for the
4905 * return value conventions.
4906 */
4907 static int config_entry_handler(const struct config_entry *entry, void *unused)
4908 {
4909 int ret = 0, i;
4910
4911 if (!entry || !entry->name || !entry->value) {
4912 ret = -EINVAL;
4913 goto end;
4914 }
4915
4916 /* Check if the option is to be ignored */
4917 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4918 if (!strcmp(entry->name, config_ignore_options[i])) {
4919 goto end;
4920 }
4921 }
4922
4923 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4924 i++) {
4925
4926 /* Ignore if not fully matched. */
4927 if (strcmp(entry->name, long_options[i].name)) {
4928 continue;
4929 }
4930
4931 /*
4932 * If the option takes no argument on the command line, we have to
4933 * check if the value is "true". We support non-zero numeric values,
4934 * true, on and yes.
4935 */
4936 if (!long_options[i].has_arg) {
4937 ret = config_parse_value(entry->value);
4938 if (ret <= 0) {
4939 if (ret) {
4940 WARN("Invalid configuration value \"%s\" for option %s",
4941 entry->value, entry->name);
4942 }
4943 /* False, skip boolean config option. */
4944 goto end;
4945 }
4946 }
4947
4948 ret = set_option(long_options[i].val, entry->value, entry->name);
4949 goto end;
4950 }
4951
4952 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4953
4954 end:
4955 return ret;
4956 }
4957
4958 /*
4959 * daemon configuration loading and argument parsing
4960 */
4961 static int set_options(int argc, char **argv)
4962 {
4963 int ret = 0, c = 0, option_index = 0;
4964 int orig_optopt = optopt, orig_optind = optind;
4965 char *optstring;
4966 const char *config_path = NULL;
4967
4968 optstring = utils_generate_optstring(long_options,
4969 sizeof(long_options) / sizeof(struct option));
4970 if (!optstring) {
4971 ret = -ENOMEM;
4972 goto end;
4973 }
4974
4975 /* Check for the --config option */
4976 while ((c = getopt_long(argc, argv, optstring, long_options,
4977 &option_index)) != -1) {
4978 if (c == '?') {
4979 ret = -EINVAL;
4980 goto end;
4981 } else if (c != 'f') {
4982 /* if not equal to --config option. */
4983 continue;
4984 }
4985
4986 if (lttng_is_setuid_setgid()) {
4987 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4988 "-f, --config");
4989 } else {
4990 config_path = utils_expand_path(optarg);
4991 if (!config_path) {
4992 ERR("Failed to resolve path: %s", optarg);
4993 }
4994 }
4995 }
4996
4997 ret = config_get_section_entries(config_path, config_section_name,
4998 config_entry_handler, NULL);
4999 if (ret) {
5000 if (ret > 0) {
5001 ERR("Invalid configuration option at line %i", ret);
5002 ret = -1;
5003 }
5004 goto end;
5005 }
5006
5007 /* Reset getopt's global state */
5008 optopt = orig_optopt;
5009 optind = orig_optind;
5010 while (1) {
5011 option_index = -1;
5012 /*
5013 * getopt_long() will not set option_index if it encounters a
5014 * short option.
5015 */
5016 c = getopt_long(argc, argv, optstring, long_options,
5017 &option_index);
5018 if (c == -1) {
5019 break;
5020 }
5021
5022 /*
5023 * Pass NULL as the long option name if popt left the index
5024 * unset.
5025 */
5026 ret = set_option(c, optarg,
5027 option_index < 0 ? NULL :
5028 long_options[option_index].name);
5029 if (ret < 0) {
5030 break;
5031 }
5032 }
5033
5034 end:
5035 free(optstring);
5036 return ret;
5037 }
5038
5039 /*
5040 * Creates the two needed socket by the daemon.
5041 * apps_sock - The communication socket for all UST apps.
5042 * client_sock - The communication of the cli tool (lttng).
5043 */
5044 static int init_daemon_socket(void)
5045 {
5046 int ret = 0;
5047 mode_t old_umask;
5048
5049 old_umask = umask(0);
5050
5051 /* Create client tool unix socket */
5052 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5053 if (client_sock < 0) {
5054 ERR("Create unix sock failed: %s", client_unix_sock_path);
5055 ret = -1;
5056 goto end;
5057 }
5058
5059 /* Set the cloexec flag */
5060 ret = utils_set_fd_cloexec(client_sock);
5061 if (ret < 0) {
5062 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5063 "Continuing but note that the consumer daemon will have a "
5064 "reference to this socket on exec()", client_sock);
5065 }
5066
5067 /* File permission MUST be 660 */
5068 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5069 if (ret < 0) {
5070 ERR("Set file permissions failed: %s", client_unix_sock_path);
5071 PERROR("chmod");
5072 goto end;
5073 }
5074
5075 /* Create the application unix socket */
5076 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5077 if (apps_sock < 0) {
5078 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5079 ret = -1;
5080 goto end;
5081 }
5082
5083 /* Set the cloexec flag */
5084 ret = utils_set_fd_cloexec(apps_sock);
5085 if (ret < 0) {
5086 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5087 "Continuing but note that the consumer daemon will have a "
5088 "reference to this socket on exec()", apps_sock);
5089 }
5090
5091 /* File permission MUST be 666 */
5092 ret = chmod(apps_unix_sock_path,
5093 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5094 if (ret < 0) {
5095 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5096 PERROR("chmod");
5097 goto end;
5098 }
5099
5100 DBG3("Session daemon client socket %d and application socket %d created",
5101 client_sock, apps_sock);
5102
5103 end:
5104 umask(old_umask);
5105 return ret;
5106 }
5107
5108 /*
5109 * Check if the global socket is available, and if a daemon is answering at the
5110 * other side. If yes, error is returned.
5111 */
5112 static int check_existing_daemon(void)
5113 {
5114 /* Is there anybody out there ? */
5115 if (lttng_session_daemon_alive()) {
5116 return -EEXIST;
5117 }
5118
5119 return 0;
5120 }
5121
5122 /*
5123 * Set the tracing group gid onto the client socket.
5124 *
5125 * Race window between mkdir and chown is OK because we are going from more
5126 * permissive (root.root) to less permissive (root.tracing).
5127 */
5128 static int set_permissions(char *rundir)
5129 {
5130 int ret;
5131 gid_t gid;
5132
5133 gid = utils_get_group_id(tracing_group_name);
5134
5135 /* Set lttng run dir */
5136 ret = chown(rundir, 0, gid);
5137 if (ret < 0) {
5138 ERR("Unable to set group on %s", rundir);
5139 PERROR("chown");
5140 }
5141
5142 /*
5143 * Ensure all applications and tracing group can search the run
5144 * dir. Allow everyone to read the directory, since it does not
5145 * buy us anything to hide its content.
5146 */
5147 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5148 if (ret < 0) {
5149 ERR("Unable to set permissions on %s", rundir);
5150 PERROR("chmod");
5151 }
5152
5153 /* lttng client socket path */
5154 ret = chown(client_unix_sock_path, 0, gid);
5155 if (ret < 0) {
5156 ERR("Unable to set group on %s", client_unix_sock_path);
5157 PERROR("chown");
5158 }
5159
5160 /* kconsumer error socket path */
5161 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5162 if (ret < 0) {
5163 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5164 PERROR("chown");
5165 }
5166
5167 /* 64-bit ustconsumer error socket path */
5168 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5169 if (ret < 0) {
5170 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5171 PERROR("chown");
5172 }
5173
5174 /* 32-bit ustconsumer compat32 error socket path */
5175 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5176 if (ret < 0) {
5177 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5178 PERROR("chown");
5179 }
5180
5181 DBG("All permissions are set");
5182
5183 return ret;
5184 }
5185
5186 /*
5187 * Create the lttng run directory needed for all global sockets and pipe.
5188 */
5189 static int create_lttng_rundir(const char *rundir)
5190 {
5191 int ret;
5192
5193 DBG3("Creating LTTng run directory: %s", rundir);
5194
5195 ret = mkdir(rundir, S_IRWXU);
5196 if (ret < 0) {
5197 if (errno != EEXIST) {
5198 ERR("Unable to create %s", rundir);
5199 goto error;
5200 } else {
5201 ret = 0;
5202 }
5203 }
5204
5205 error:
5206 return ret;
5207 }
5208
5209 /*
5210 * Setup sockets and directory needed by the kconsumerd communication with the
5211 * session daemon.
5212 */
5213 static int set_consumer_sockets(struct consumer_data *consumer_data,
5214 const char *rundir)
5215 {
5216 int ret;
5217 char path[PATH_MAX];
5218
5219 switch (consumer_data->type) {
5220 case LTTNG_CONSUMER_KERNEL:
5221 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5222 break;
5223 case LTTNG_CONSUMER64_UST:
5224 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5225 break;
5226 case LTTNG_CONSUMER32_UST:
5227 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5228 break;
5229 default:
5230 ERR("Consumer type unknown");
5231 ret = -EINVAL;
5232 goto error;
5233 }
5234
5235 DBG2("Creating consumer directory: %s", path);
5236
5237 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5238 if (ret < 0) {
5239 if (errno != EEXIST) {
5240 PERROR("mkdir");
5241 ERR("Failed to create %s", path);
5242 goto error;
5243 }
5244 ret = -1;
5245 }
5246 if (is_root) {
5247 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5248 if (ret < 0) {
5249 ERR("Unable to set group on %s", path);
5250 PERROR("chown");
5251 goto error;
5252 }
5253 }
5254
5255 /* Create the kconsumerd error unix socket */
5256 consumer_data->err_sock =
5257 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5258 if (consumer_data->err_sock < 0) {
5259 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5260 ret = -1;
5261 goto error;
5262 }
5263
5264 /*
5265 * Set the CLOEXEC flag. Return code is useless because either way, the
5266 * show must go on.
5267 */
5268 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5269 if (ret < 0) {
5270 PERROR("utils_set_fd_cloexec");
5271 /* continue anyway */
5272 }
5273
5274 /* File permission MUST be 660 */
5275 ret = chmod(consumer_data->err_unix_sock_path,
5276 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5277 if (ret < 0) {
5278 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5279 PERROR("chmod");
5280 goto error;
5281 }
5282
5283 error:
5284 return ret;
5285 }
5286
5287 /*
5288 * Signal handler for the daemon
5289 *
5290 * Simply stop all worker threads, leaving main() return gracefully after
5291 * joining all threads and calling cleanup().
5292 */
5293 static void sighandler(int sig)
5294 {
5295 switch (sig) {
5296 case SIGPIPE:
5297 DBG("SIGPIPE caught");
5298 return;
5299 case SIGINT:
5300 DBG("SIGINT caught");
5301 stop_threads();
5302 break;
5303 case SIGTERM:
5304 DBG("SIGTERM caught");
5305 stop_threads();
5306 break;
5307 case SIGUSR1:
5308 CMM_STORE_SHARED(recv_child_signal, 1);
5309 break;
5310 default:
5311 break;
5312 }
5313 }
5314
5315 /*
5316 * Setup signal handler for :
5317 * SIGINT, SIGTERM, SIGPIPE
5318 */
5319 static int set_signal_handler(void)
5320 {
5321 int ret = 0;
5322 struct sigaction sa;
5323 sigset_t sigset;
5324
5325 if ((ret = sigemptyset(&sigset)) < 0) {
5326 PERROR("sigemptyset");
5327 return ret;
5328 }
5329
5330 sa.sa_handler = sighandler;
5331 sa.sa_mask = sigset;
5332 sa.sa_flags = 0;
5333 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5334 PERROR("sigaction");
5335 return ret;
5336 }
5337
5338 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5339 PERROR("sigaction");
5340 return ret;
5341 }
5342
5343 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5344 PERROR("sigaction");
5345 return ret;
5346 }
5347
5348 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5349 PERROR("sigaction");
5350 return ret;
5351 }
5352
5353 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5354
5355 return ret;
5356 }
5357
5358 /*
5359 * Set open files limit to unlimited. This daemon can open a large number of
5360 * file descriptors in order to consumer multiple kernel traces.
5361 */
5362 static void set_ulimit(void)
5363 {
5364 int ret;
5365 struct rlimit lim;
5366
5367 /* The kernel does not allowed an infinite limit for open files */
5368 lim.rlim_cur = 65535;
5369 lim.rlim_max = 65535;
5370
5371 ret = setrlimit(RLIMIT_NOFILE, &lim);
5372 if (ret < 0) {
5373 PERROR("failed to set open files limit");
5374 }
5375 }
5376
5377 /*
5378 * Write pidfile using the rundir and opt_pidfile.
5379 */
5380 static int write_pidfile(void)
5381 {
5382 int ret;
5383 char pidfile_path[PATH_MAX];
5384
5385 assert(rundir);
5386
5387 if (opt_pidfile) {
5388 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5389 } else {
5390 /* Build pidfile path from rundir and opt_pidfile. */
5391 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5392 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5393 if (ret < 0) {
5394 PERROR("snprintf pidfile path");
5395 goto error;
5396 }
5397 }
5398
5399 /*
5400 * Create pid file in rundir.
5401 */
5402 ret = utils_create_pid_file(getpid(), pidfile_path);
5403 error:
5404 return ret;
5405 }
5406
5407 /*
5408 * Create lockfile using the rundir and return its fd.
5409 */
5410 static int create_lockfile(void)
5411 {
5412 int ret;
5413 char lockfile_path[PATH_MAX];
5414
5415 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5416 if (ret < 0) {
5417 goto error;
5418 }
5419
5420 ret = utils_create_lock_file(lockfile_path);
5421 error:
5422 return ret;
5423 }
5424
5425 /*
5426 * Write agent TCP port using the rundir.
5427 */
5428 static int write_agent_port(void)
5429 {
5430 int ret;
5431 char path[PATH_MAX];
5432
5433 assert(rundir);
5434
5435 ret = snprintf(path, sizeof(path), "%s/"
5436 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5437 if (ret < 0) {
5438 PERROR("snprintf agent port path");
5439 goto error;
5440 }
5441
5442 /*
5443 * Create TCP agent port file in rundir.
5444 */
5445 ret = utils_create_pid_file(agent_tcp_port, path);
5446
5447 error:
5448 return ret;
5449 }
5450
5451 /*
5452 * main
5453 */
5454 int main(int argc, char **argv)
5455 {
5456 int ret = 0, retval = 0;
5457 void *status;
5458 const char *home_path, *env_app_timeout;
5459
5460 init_kernel_workarounds();
5461
5462 rcu_register_thread();
5463
5464 if (set_signal_handler()) {
5465 retval = -1;
5466 goto exit_set_signal_handler;
5467 }
5468
5469 setup_consumerd_path();
5470
5471 page_size = sysconf(_SC_PAGESIZE);
5472 if (page_size < 0) {
5473 PERROR("sysconf _SC_PAGESIZE");
5474 page_size = LONG_MAX;
5475 WARN("Fallback page size to %ld", page_size);
5476 }
5477
5478 /*
5479 * Parse arguments and load the daemon configuration file.
5480 *
5481 * We have an exit_options exit path to free memory reserved by
5482 * set_options. This is needed because the rest of sessiond_cleanup()
5483 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5484 * depends on set_options.
5485 */
5486 progname = argv[0];
5487 if (set_options(argc, argv)) {
5488 retval = -1;
5489 goto exit_options;
5490 }
5491
5492 /* Daemonize */
5493 if (opt_daemon || opt_background) {
5494 int i;
5495
5496 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5497 !opt_background);
5498 if (ret < 0) {
5499 retval = -1;
5500 goto exit_options;
5501 }
5502
5503 /*
5504 * We are in the child. Make sure all other file descriptors are
5505 * closed, in case we are called with more opened file
5506 * descriptors than the standard ones.
5507 */
5508 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5509 (void) close(i);
5510 }
5511 }
5512
5513 if (run_as_create_worker(argv[0]) < 0) {
5514 goto exit_create_run_as_worker_cleanup;
5515 }
5516
5517 /*
5518 * Starting from here, we can create threads. This needs to be after
5519 * lttng_daemonize due to RCU.
5520 */
5521
5522 /*
5523 * Initialize the health check subsystem. This call should set the
5524 * appropriate time values.
5525 */
5526 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5527 if (!health_sessiond) {
5528 PERROR("health_app_create error");
5529 retval = -1;
5530 goto exit_health_sessiond_cleanup;
5531 }
5532
5533 if (init_ht_cleanup_quit_pipe()) {
5534 retval = -1;
5535 goto exit_ht_cleanup_quit_pipe;
5536 }
5537
5538 /* Setup the thread ht_cleanup communication pipe. */
5539 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5540 retval = -1;
5541 goto exit_ht_cleanup_pipe;
5542 }
5543
5544 /* Set up max poll set size */
5545 if (lttng_poll_set_max_size()) {
5546 retval = -1;
5547 goto exit_set_max_size;
5548 }
5549
5550 /* Create thread to clean up RCU hash tables */
5551 ret = pthread_create(&ht_cleanup_thread, NULL,
5552 thread_ht_cleanup, (void *) NULL);
5553 if (ret) {
5554 errno = ret;
5555 PERROR("pthread_create ht_cleanup");
5556 retval = -1;
5557 goto exit_ht_cleanup;
5558 }
5559
5560 /* Create thread quit pipe */
5561 if (init_thread_quit_pipe()) {
5562 retval = -1;
5563 goto exit_init_data;
5564 }
5565
5566 /* Check if daemon is UID = 0 */
5567 is_root = !getuid();
5568
5569 if (is_root) {
5570 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5571 if (!rundir) {
5572 retval = -1;
5573 goto exit_init_data;
5574 }
5575
5576 /* Create global run dir with root access */
5577 if (create_lttng_rundir(rundir)) {
5578 retval = -1;
5579 goto exit_init_data;
5580 }
5581
5582 if (strlen(apps_unix_sock_path) == 0) {
5583 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5584 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5585 if (ret < 0) {
5586 retval = -1;
5587 goto exit_init_data;
5588 }
5589 }
5590
5591 if (strlen(client_unix_sock_path) == 0) {
5592 ret = snprintf(client_unix_sock_path, PATH_MAX,
5593 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5594 if (ret < 0) {
5595 retval = -1;
5596 goto exit_init_data;
5597 }
5598 }
5599
5600 /* Set global SHM for ust */
5601 if (strlen(wait_shm_path) == 0) {
5602 ret = snprintf(wait_shm_path, PATH_MAX,
5603 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5604 if (ret < 0) {
5605 retval = -1;
5606 goto exit_init_data;
5607 }
5608 }
5609
5610 if (strlen(health_unix_sock_path) == 0) {
5611 ret = snprintf(health_unix_sock_path,
5612 sizeof(health_unix_sock_path),
5613 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5614 if (ret < 0) {
5615 retval = -1;
5616 goto exit_init_data;
5617 }
5618 }
5619
5620 /* Setup kernel consumerd path */
5621 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5622 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5623 if (ret < 0) {
5624 retval = -1;
5625 goto exit_init_data;
5626 }
5627 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5628 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5629 if (ret < 0) {
5630 retval = -1;
5631 goto exit_init_data;
5632 }
5633
5634 DBG2("Kernel consumer err path: %s",
5635 kconsumer_data.err_unix_sock_path);
5636 DBG2("Kernel consumer cmd path: %s",
5637 kconsumer_data.cmd_unix_sock_path);
5638 } else {
5639 home_path = utils_get_home_dir();
5640 if (home_path == NULL) {
5641 /* TODO: Add --socket PATH option */
5642 ERR("Can't get HOME directory for sockets creation.");
5643 retval = -1;
5644 goto exit_init_data;
5645 }
5646
5647 /*
5648 * Create rundir from home path. This will create something like
5649 * $HOME/.lttng
5650 */
5651 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5652 if (ret < 0) {
5653 retval = -1;
5654 goto exit_init_data;
5655 }
5656
5657 if (create_lttng_rundir(rundir)) {
5658 retval = -1;
5659 goto exit_init_data;
5660 }
5661
5662 if (strlen(apps_unix_sock_path) == 0) {
5663 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5664 DEFAULT_HOME_APPS_UNIX_SOCK,
5665 home_path);
5666 if (ret < 0) {
5667 retval = -1;
5668 goto exit_init_data;
5669 }
5670 }
5671
5672 /* Set the cli tool unix socket path */
5673 if (strlen(client_unix_sock_path) == 0) {
5674 ret = snprintf(client_unix_sock_path, PATH_MAX,
5675 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5676 home_path);
5677 if (ret < 0) {
5678 retval = -1;
5679 goto exit_init_data;
5680 }
5681 }
5682
5683 /* Set global SHM for ust */
5684 if (strlen(wait_shm_path) == 0) {
5685 ret = snprintf(wait_shm_path, PATH_MAX,
5686 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5687 getuid());
5688 if (ret < 0) {
5689 retval = -1;
5690 goto exit_init_data;
5691 }
5692 }
5693
5694 /* Set health check Unix path */
5695 if (strlen(health_unix_sock_path) == 0) {
5696 ret = snprintf(health_unix_sock_path,
5697 sizeof(health_unix_sock_path),
5698 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5699 home_path);
5700 if (ret < 0) {
5701 retval = -1;
5702 goto exit_init_data;
5703 }
5704 }
5705 }
5706
5707 lockfile_fd = create_lockfile();
5708 if (lockfile_fd < 0) {
5709 retval = -1;
5710 goto exit_init_data;
5711 }
5712
5713 /* Set consumer initial state */
5714 kernel_consumerd_state = CONSUMER_STOPPED;
5715 ust_consumerd_state = CONSUMER_STOPPED;
5716
5717 DBG("Client socket path %s", client_unix_sock_path);
5718 DBG("Application socket path %s", apps_unix_sock_path);
5719 DBG("Application wait path %s", wait_shm_path);
5720 DBG("LTTng run directory path: %s", rundir);
5721
5722 /* 32 bits consumerd path setup */
5723 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5724 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5725 if (ret < 0) {
5726 PERROR("snprintf 32-bit consumer error socket path");
5727 retval = -1;
5728 goto exit_init_data;
5729 }
5730 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5731 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5732 if (ret < 0) {
5733 PERROR("snprintf 32-bit consumer command socket path");
5734 retval = -1;
5735 goto exit_init_data;
5736 }
5737
5738 DBG2("UST consumer 32 bits err path: %s",
5739 ustconsumer32_data.err_unix_sock_path);
5740 DBG2("UST consumer 32 bits cmd path: %s",
5741 ustconsumer32_data.cmd_unix_sock_path);
5742
5743 /* 64 bits consumerd path setup */
5744 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5745 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5746 if (ret < 0) {
5747 PERROR("snprintf 64-bit consumer error socket path");
5748 retval = -1;
5749 goto exit_init_data;
5750 }
5751 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5752 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5753 if (ret < 0) {
5754 PERROR("snprintf 64-bit consumer command socket path");
5755 retval = -1;
5756 goto exit_init_data;
5757 }
5758
5759 DBG2("UST consumer 64 bits err path: %s",
5760 ustconsumer64_data.err_unix_sock_path);
5761 DBG2("UST consumer 64 bits cmd path: %s",
5762 ustconsumer64_data.cmd_unix_sock_path);
5763
5764 /*
5765 * See if daemon already exist.
5766 */
5767 if (check_existing_daemon()) {
5768 ERR("Already running daemon.\n");
5769 /*
5770 * We do not goto exit because we must not cleanup()
5771 * because a daemon is already running.
5772 */
5773 retval = -1;
5774 goto exit_init_data;
5775 }
5776
5777 /*
5778 * Init UST app hash table. Alloc hash table before this point since
5779 * cleanup() can get called after that point.
5780 */
5781 if (ust_app_ht_alloc()) {
5782 ERR("Failed to allocate UST app hash table");
5783 retval = -1;
5784 goto exit_init_data;
5785 }
5786
5787 /*
5788 * Initialize agent app hash table. We allocate the hash table here
5789 * since cleanup() can get called after this point.
5790 */
5791 if (agent_app_ht_alloc()) {
5792 ERR("Failed to allocate Agent app hash table");
5793 retval = -1;
5794 goto exit_init_data;
5795 }
5796
5797 /*
5798 * These actions must be executed as root. We do that *after* setting up
5799 * the sockets path because we MUST make the check for another daemon using
5800 * those paths *before* trying to set the kernel consumer sockets and init
5801 * kernel tracer.
5802 */
5803 if (is_root) {
5804 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5805 retval = -1;
5806 goto exit_init_data;
5807 }
5808
5809 /* Setup kernel tracer */
5810 if (!opt_no_kernel) {
5811 init_kernel_tracer();
5812 if (kernel_tracer_fd >= 0) {
5813 ret = syscall_init_table();
5814 if (ret < 0) {
5815 ERR("Unable to populate syscall table. "
5816 "Syscall tracing won't work "
5817 "for this session daemon.");
5818 }
5819 }
5820 }
5821
5822 /* Set ulimit for open files */
5823 set_ulimit();
5824 }
5825 /* init lttng_fd tracking must be done after set_ulimit. */
5826 lttng_fd_init();
5827
5828 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5829 retval = -1;
5830 goto exit_init_data;
5831 }
5832
5833 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5834 retval = -1;
5835 goto exit_init_data;
5836 }
5837
5838 /* Setup the needed unix socket */
5839 if (init_daemon_socket()) {
5840 retval = -1;
5841 goto exit_init_data;
5842 }
5843
5844 /* Set credentials to socket */
5845 if (is_root && set_permissions(rundir)) {
5846 retval = -1;
5847 goto exit_init_data;
5848 }
5849
5850 /* Get parent pid if -S, --sig-parent is specified. */
5851 if (opt_sig_parent) {
5852 ppid = getppid();
5853 }
5854
5855 /* Setup the kernel pipe for waking up the kernel thread */
5856 if (is_root && !opt_no_kernel) {
5857 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5858 retval = -1;
5859 goto exit_init_data;
5860 }
5861 }
5862
5863 /* Setup the thread apps communication pipe. */
5864 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5865 retval = -1;
5866 goto exit_init_data;
5867 }
5868
5869 /* Setup the thread apps notify communication pipe. */
5870 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5871 retval = -1;
5872 goto exit_init_data;
5873 }
5874
5875 /* Initialize global buffer per UID and PID registry. */
5876 buffer_reg_init_uid_registry();
5877 buffer_reg_init_pid_registry();
5878
5879 /* Init UST command queue. */
5880 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5881
5882 /*
5883 * Get session list pointer. This pointer MUST NOT be free'd. This list
5884 * is statically declared in session.c
5885 */
5886 session_list_ptr = session_get_list();
5887
5888 cmd_init();
5889
5890 /* Check for the application socket timeout env variable. */
5891 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5892 if (env_app_timeout) {
5893 app_socket_timeout = atoi(env_app_timeout);
5894 } else {
5895 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5896 }
5897
5898 ret = write_pidfile();
5899 if (ret) {
5900 ERR("Error in write_pidfile");
5901 retval = -1;
5902 goto exit_init_data;
5903 }
5904 ret = write_agent_port();
5905 if (ret) {
5906 ERR("Error in write_agent_port");
5907 retval = -1;
5908 goto exit_init_data;
5909 }
5910
5911 /* Initialize communication library */
5912 lttcomm_init();
5913 /* Initialize TCP timeout values */
5914 lttcomm_inet_init();
5915
5916 if (load_session_init_data(&load_info) < 0) {
5917 retval = -1;
5918 goto exit_init_data;
5919 }
5920 load_info->path = opt_load_session_path;
5921
5922 /* Create health-check thread */
5923 ret = pthread_create(&health_thread, NULL,
5924 thread_manage_health, (void *) NULL);
5925 if (ret) {
5926 errno = ret;
5927 PERROR("pthread_create health");
5928 retval = -1;
5929 goto exit_health;
5930 }
5931
5932 /* Create thread to manage the client socket */
5933 ret = pthread_create(&client_thread, NULL,
5934 thread_manage_clients, (void *) NULL);
5935 if (ret) {
5936 errno = ret;
5937 PERROR("pthread_create clients");
5938 retval = -1;
5939 goto exit_client;
5940 }
5941
5942 /* Create thread to dispatch registration */
5943 ret = pthread_create(&dispatch_thread, NULL,
5944 thread_dispatch_ust_registration, (void *) NULL);
5945 if (ret) {
5946 errno = ret;
5947 PERROR("pthread_create dispatch");
5948 retval = -1;
5949 goto exit_dispatch;
5950 }
5951
5952 /* Create thread to manage application registration. */
5953 ret = pthread_create(&reg_apps_thread, NULL,
5954 thread_registration_apps, (void *) NULL);
5955 if (ret) {
5956 errno = ret;
5957 PERROR("pthread_create registration");
5958 retval = -1;
5959 goto exit_reg_apps;
5960 }
5961
5962 /* Create thread to manage application socket */
5963 ret = pthread_create(&apps_thread, NULL,
5964 thread_manage_apps, (void *) NULL);
5965 if (ret) {
5966 errno = ret;
5967 PERROR("pthread_create apps");
5968 retval = -1;
5969 goto exit_apps;
5970 }
5971
5972 /* Create thread to manage application notify socket */
5973 ret = pthread_create(&apps_notify_thread, NULL,
5974 ust_thread_manage_notify, (void *) NULL);
5975 if (ret) {
5976 errno = ret;
5977 PERROR("pthread_create notify");
5978 retval = -1;
5979 goto exit_apps_notify;
5980 }
5981
5982 /* Create agent registration thread. */
5983 ret = pthread_create(&agent_reg_thread, NULL,
5984 agent_thread_manage_registration, (void *) NULL);
5985 if (ret) {
5986 errno = ret;
5987 PERROR("pthread_create agent");
5988 retval = -1;
5989 goto exit_agent_reg;
5990 }
5991
5992 /* Don't start this thread if kernel tracing is not requested nor root */
5993 if (is_root && !opt_no_kernel) {
5994 /* Create kernel thread to manage kernel event */
5995 ret = pthread_create(&kernel_thread, NULL,
5996 thread_manage_kernel, (void *) NULL);
5997 if (ret) {
5998 errno = ret;
5999 PERROR("pthread_create kernel");
6000 retval = -1;
6001 goto exit_kernel;
6002 }
6003 }
6004
6005 /* Create session loading thread. */
6006 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
6007 load_info);
6008 if (ret) {
6009 errno = ret;
6010 PERROR("pthread_create load_session_thread");
6011 retval = -1;
6012 goto exit_load_session;
6013 }
6014
6015 /*
6016 * This is where we start awaiting program completion (e.g. through
6017 * signal that asks threads to teardown).
6018 */
6019
6020 ret = pthread_join(load_session_thread, &status);
6021 if (ret) {
6022 errno = ret;
6023 PERROR("pthread_join load_session_thread");
6024 retval = -1;
6025 }
6026 exit_load_session:
6027
6028 if (is_root && !opt_no_kernel) {
6029 ret = pthread_join(kernel_thread, &status);
6030 if (ret) {
6031 errno = ret;
6032 PERROR("pthread_join");
6033 retval = -1;
6034 }
6035 }
6036 exit_kernel:
6037
6038 ret = pthread_join(agent_reg_thread, &status);
6039 if (ret) {
6040 errno = ret;
6041 PERROR("pthread_join agent");
6042 retval = -1;
6043 }
6044 exit_agent_reg:
6045
6046 ret = pthread_join(apps_notify_thread, &status);
6047 if (ret) {
6048 errno = ret;
6049 PERROR("pthread_join apps notify");
6050 retval = -1;
6051 }
6052 exit_apps_notify:
6053
6054 ret = pthread_join(apps_thread, &status);
6055 if (ret) {
6056 errno = ret;
6057 PERROR("pthread_join apps");
6058 retval = -1;
6059 }
6060 exit_apps:
6061
6062 ret = pthread_join(reg_apps_thread, &status);
6063 if (ret) {
6064 errno = ret;
6065 PERROR("pthread_join");
6066 retval = -1;
6067 }
6068 exit_reg_apps:
6069
6070 /*
6071 * Join dispatch thread after joining reg_apps_thread to ensure
6072 * we don't leak applications in the queue.
6073 */
6074 ret = pthread_join(dispatch_thread, &status);
6075 if (ret) {
6076 errno = ret;
6077 PERROR("pthread_join");
6078 retval = -1;
6079 }
6080 exit_dispatch:
6081
6082 ret = pthread_join(client_thread, &status);
6083 if (ret) {
6084 errno = ret;
6085 PERROR("pthread_join");
6086 retval = -1;
6087 }
6088 exit_client:
6089
6090 ret = pthread_join(health_thread, &status);
6091 if (ret) {
6092 errno = ret;
6093 PERROR("pthread_join health thread");
6094 retval = -1;
6095 }
6096 exit_health:
6097
6098 exit_init_data:
6099 /*
6100 * sessiond_cleanup() is called when no other thread is running, except
6101 * the ht_cleanup thread, which is needed to destroy the hash tables.
6102 */
6103 rcu_thread_online();
6104 sessiond_cleanup();
6105 rcu_thread_offline();
6106 rcu_unregister_thread();
6107
6108 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6109 if (ret < 0) {
6110 ERR("write error on ht_cleanup quit pipe");
6111 retval = -1;
6112 }
6113
6114 ret = pthread_join(ht_cleanup_thread, &status);
6115 if (ret) {
6116 errno = ret;
6117 PERROR("pthread_join ht cleanup thread");
6118 retval = -1;
6119 }
6120 exit_ht_cleanup:
6121 exit_set_max_size:
6122
6123 utils_close_pipe(ht_cleanup_pipe);
6124 exit_ht_cleanup_pipe:
6125
6126 /*
6127 * Close the ht_cleanup quit pipe.
6128 */
6129 utils_close_pipe(ht_cleanup_quit_pipe);
6130 exit_ht_cleanup_quit_pipe:
6131
6132 health_app_destroy(health_sessiond);
6133 exit_health_sessiond_cleanup:
6134 exit_create_run_as_worker_cleanup:
6135
6136 exit_options:
6137 /* Ensure all prior call_rcu are done. */
6138 rcu_barrier();
6139
6140 sessiond_cleanup_options();
6141
6142 exit_set_signal_handler:
6143
6144 if (!retval) {
6145 exit(EXIT_SUCCESS);
6146 } else {
6147 exit(EXIT_FAILURE);
6148 }
6149 }
This page took 0.14765 seconds and 5 git commands to generate.