7a8b5ae42dcfedb4c7af6a159997d3eaa98dd02f
[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/session-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 /*
3347 * An LTTNG_ADD_CONTEXT command might have a supplementary
3348 * payload if the context being added is an application context.
3349 */
3350 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3351 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3352 char *provider_name = NULL, *context_name = NULL;
3353 size_t provider_name_len =
3354 cmd_ctx->lsm->u.context.provider_name_len;
3355 size_t context_name_len =
3356 cmd_ctx->lsm->u.context.context_name_len;
3357
3358 if (provider_name_len == 0 || context_name_len == 0) {
3359 /*
3360 * Application provider and context names MUST
3361 * be provided.
3362 */
3363 ret = -LTTNG_ERR_INVALID;
3364 goto error;
3365 }
3366
3367 provider_name = zmalloc(provider_name_len + 1);
3368 if (!provider_name) {
3369 ret = -LTTNG_ERR_NOMEM;
3370 goto error;
3371 }
3372
3373 context_name = zmalloc(context_name_len + 1);
3374 if (!context_name) {
3375 ret = -LTTNG_ERR_NOMEM;
3376 goto error_add_context;
3377 }
3378
3379 ret = lttcomm_recv_unix_sock(sock, provider_name,
3380 provider_name_len);
3381 if (ret < 0) {
3382 goto error_add_context;
3383 }
3384
3385 ret = lttcomm_recv_unix_sock(sock, context_name,
3386 context_name_len);
3387 if (ret < 0) {
3388 goto error_add_context;
3389 }
3390 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3391 provider_name;
3392 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3393 context_name;
3394 }
3395
3396 /*
3397 * cmd_add_context assumes ownership of the provider and context
3398 * names.
3399 */
3400 ret = cmd_add_context(cmd_ctx->session,
3401 cmd_ctx->lsm->domain.type,
3402 cmd_ctx->lsm->u.context.channel_name,
3403 &cmd_ctx->lsm->u.context.ctx,
3404 kernel_poll_pipe[1]);
3405
3406 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3407 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3408 error_add_context:
3409 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3410 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3411 if (ret < 0) {
3412 goto error;
3413 }
3414 break;
3415 }
3416 case LTTNG_DISABLE_CHANNEL:
3417 {
3418 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3419 cmd_ctx->lsm->u.disable.channel_name);
3420 break;
3421 }
3422 case LTTNG_DISABLE_EVENT:
3423 {
3424
3425 /*
3426 * FIXME: handle filter; for now we just receive the filter's
3427 * bytecode along with the filter expression which are sent by
3428 * liblttng-ctl and discard them.
3429 *
3430 * This fixes an issue where the client may block while sending
3431 * the filter payload and encounter an error because the session
3432 * daemon closes the socket without ever handling this data.
3433 */
3434 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3435 cmd_ctx->lsm->u.disable.bytecode_len;
3436
3437 if (count) {
3438 char data[LTTNG_FILTER_MAX_LEN];
3439
3440 DBG("Discarding disable event command payload of size %zu", count);
3441 while (count) {
3442 ret = lttcomm_recv_unix_sock(sock, data,
3443 count > sizeof(data) ? sizeof(data) : count);
3444 if (ret < 0) {
3445 goto error;
3446 }
3447
3448 count -= (size_t) ret;
3449 }
3450 }
3451 /* FIXME: passing packed structure to non-packed pointer */
3452 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3453 cmd_ctx->lsm->u.disable.channel_name,
3454 &cmd_ctx->lsm->u.disable.event);
3455 break;
3456 }
3457 case LTTNG_ENABLE_CHANNEL:
3458 {
3459 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3460 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3461 break;
3462 }
3463 case LTTNG_TRACK_PID:
3464 {
3465 ret = cmd_track_pid(cmd_ctx->session,
3466 cmd_ctx->lsm->domain.type,
3467 cmd_ctx->lsm->u.pid_tracker.pid);
3468 break;
3469 }
3470 case LTTNG_UNTRACK_PID:
3471 {
3472 ret = cmd_untrack_pid(cmd_ctx->session,
3473 cmd_ctx->lsm->domain.type,
3474 cmd_ctx->lsm->u.pid_tracker.pid);
3475 break;
3476 }
3477 case LTTNG_ENABLE_EVENT:
3478 {
3479 struct lttng_event_exclusion *exclusion = NULL;
3480 struct lttng_filter_bytecode *bytecode = NULL;
3481 char *filter_expression = NULL;
3482
3483 /* Handle exclusion events and receive it from the client. */
3484 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3485 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3486
3487 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3488 (count * LTTNG_SYMBOL_NAME_LEN));
3489 if (!exclusion) {
3490 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3491 goto error;
3492 }
3493
3494 DBG("Receiving var len exclusion event list from client ...");
3495 exclusion->count = count;
3496 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3497 count * LTTNG_SYMBOL_NAME_LEN);
3498 if (ret <= 0) {
3499 DBG("Nothing recv() from client var len data... continuing");
3500 *sock_error = 1;
3501 free(exclusion);
3502 ret = LTTNG_ERR_EXCLUSION_INVAL;
3503 goto error;
3504 }
3505 }
3506
3507 /* Get filter expression from client. */
3508 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3509 size_t expression_len =
3510 cmd_ctx->lsm->u.enable.expression_len;
3511
3512 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3513 ret = LTTNG_ERR_FILTER_INVAL;
3514 free(exclusion);
3515 goto error;
3516 }
3517
3518 filter_expression = zmalloc(expression_len);
3519 if (!filter_expression) {
3520 free(exclusion);
3521 ret = LTTNG_ERR_FILTER_NOMEM;
3522 goto error;
3523 }
3524
3525 /* Receive var. len. data */
3526 DBG("Receiving var len filter's expression from client ...");
3527 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3528 expression_len);
3529 if (ret <= 0) {
3530 DBG("Nothing recv() from client car len data... continuing");
3531 *sock_error = 1;
3532 free(filter_expression);
3533 free(exclusion);
3534 ret = LTTNG_ERR_FILTER_INVAL;
3535 goto error;
3536 }
3537 }
3538
3539 /* Handle filter and get bytecode from client. */
3540 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3541 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3542
3543 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3544 ret = LTTNG_ERR_FILTER_INVAL;
3545 free(filter_expression);
3546 free(exclusion);
3547 goto error;
3548 }
3549
3550 bytecode = zmalloc(bytecode_len);
3551 if (!bytecode) {
3552 free(filter_expression);
3553 free(exclusion);
3554 ret = LTTNG_ERR_FILTER_NOMEM;
3555 goto error;
3556 }
3557
3558 /* Receive var. len. data */
3559 DBG("Receiving var len filter's bytecode from client ...");
3560 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3561 if (ret <= 0) {
3562 DBG("Nothing recv() from client car len data... continuing");
3563 *sock_error = 1;
3564 free(filter_expression);
3565 free(bytecode);
3566 free(exclusion);
3567 ret = LTTNG_ERR_FILTER_INVAL;
3568 goto error;
3569 }
3570
3571 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3572 free(filter_expression);
3573 free(bytecode);
3574 free(exclusion);
3575 ret = LTTNG_ERR_FILTER_INVAL;
3576 goto error;
3577 }
3578 }
3579
3580 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3581 cmd_ctx->lsm->u.enable.channel_name,
3582 &cmd_ctx->lsm->u.enable.event,
3583 filter_expression, bytecode, exclusion,
3584 kernel_poll_pipe[1]);
3585 break;
3586 }
3587 case LTTNG_LIST_TRACEPOINTS:
3588 {
3589 struct lttng_event *events;
3590 ssize_t nb_events;
3591
3592 session_lock_list();
3593 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3594 session_unlock_list();
3595 if (nb_events < 0) {
3596 /* Return value is a negative lttng_error_code. */
3597 ret = -nb_events;
3598 goto error;
3599 }
3600
3601 /*
3602 * Setup lttng message with payload size set to the event list size in
3603 * bytes and then copy list into the llm payload.
3604 */
3605 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3606 if (ret < 0) {
3607 free(events);
3608 goto setup_error;
3609 }
3610
3611 /* Copy event list into message payload */
3612 memcpy(cmd_ctx->llm->payload, events,
3613 sizeof(struct lttng_event) * nb_events);
3614
3615 free(events);
3616
3617 ret = LTTNG_OK;
3618 break;
3619 }
3620 case LTTNG_LIST_TRACEPOINT_FIELDS:
3621 {
3622 struct lttng_event_field *fields;
3623 ssize_t nb_fields;
3624
3625 session_lock_list();
3626 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3627 &fields);
3628 session_unlock_list();
3629 if (nb_fields < 0) {
3630 /* Return value is a negative lttng_error_code. */
3631 ret = -nb_fields;
3632 goto error;
3633 }
3634
3635 /*
3636 * Setup lttng message with payload size set to the event list size in
3637 * bytes and then copy list into the llm payload.
3638 */
3639 ret = setup_lttng_msg(cmd_ctx,
3640 sizeof(struct lttng_event_field) * nb_fields);
3641 if (ret < 0) {
3642 free(fields);
3643 goto setup_error;
3644 }
3645
3646 /* Copy event list into message payload */
3647 memcpy(cmd_ctx->llm->payload, fields,
3648 sizeof(struct lttng_event_field) * nb_fields);
3649
3650 free(fields);
3651
3652 ret = LTTNG_OK;
3653 break;
3654 }
3655 case LTTNG_LIST_SYSCALLS:
3656 {
3657 struct lttng_event *events;
3658 ssize_t nb_events;
3659
3660 nb_events = cmd_list_syscalls(&events);
3661 if (nb_events < 0) {
3662 /* Return value is a negative lttng_error_code. */
3663 ret = -nb_events;
3664 goto error;
3665 }
3666
3667 /*
3668 * Setup lttng message with payload size set to the event list size in
3669 * bytes and then copy list into the llm payload.
3670 */
3671 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3672 if (ret < 0) {
3673 free(events);
3674 goto setup_error;
3675 }
3676
3677 /* Copy event list into message payload */
3678 memcpy(cmd_ctx->llm->payload, events,
3679 sizeof(struct lttng_event) * nb_events);
3680
3681 free(events);
3682
3683 ret = LTTNG_OK;
3684 break;
3685 }
3686 case LTTNG_LIST_TRACKER_PIDS:
3687 {
3688 int32_t *pids = NULL;
3689 ssize_t nr_pids;
3690
3691 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3692 cmd_ctx->lsm->domain.type, &pids);
3693 if (nr_pids < 0) {
3694 /* Return value is a negative lttng_error_code. */
3695 ret = -nr_pids;
3696 goto error;
3697 }
3698
3699 /*
3700 * Setup lttng message with payload size set to the event list size in
3701 * bytes and then copy list into the llm payload.
3702 */
3703 ret = setup_lttng_msg(cmd_ctx, sizeof(int32_t) * nr_pids);
3704 if (ret < 0) {
3705 free(pids);
3706 goto setup_error;
3707 }
3708
3709 /* Copy event list into message payload */
3710 memcpy(cmd_ctx->llm->payload, pids,
3711 sizeof(int) * nr_pids);
3712
3713 free(pids);
3714
3715 ret = LTTNG_OK;
3716 break;
3717 }
3718 case LTTNG_SET_CONSUMER_URI:
3719 {
3720 size_t nb_uri, len;
3721 struct lttng_uri *uris;
3722
3723 nb_uri = cmd_ctx->lsm->u.uri.size;
3724 len = nb_uri * sizeof(struct lttng_uri);
3725
3726 if (nb_uri == 0) {
3727 ret = LTTNG_ERR_INVALID;
3728 goto error;
3729 }
3730
3731 uris = zmalloc(len);
3732 if (uris == NULL) {
3733 ret = LTTNG_ERR_FATAL;
3734 goto error;
3735 }
3736
3737 /* Receive variable len data */
3738 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3739 ret = lttcomm_recv_unix_sock(sock, uris, len);
3740 if (ret <= 0) {
3741 DBG("No URIs received from client... continuing");
3742 *sock_error = 1;
3743 ret = LTTNG_ERR_SESSION_FAIL;
3744 free(uris);
3745 goto error;
3746 }
3747
3748 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3749 free(uris);
3750 if (ret != LTTNG_OK) {
3751 goto error;
3752 }
3753
3754
3755 break;
3756 }
3757 case LTTNG_START_TRACE:
3758 {
3759 ret = cmd_start_trace(cmd_ctx->session);
3760 break;
3761 }
3762 case LTTNG_STOP_TRACE:
3763 {
3764 ret = cmd_stop_trace(cmd_ctx->session);
3765 break;
3766 }
3767 case LTTNG_CREATE_SESSION:
3768 {
3769 size_t nb_uri, len;
3770 struct lttng_uri *uris = NULL;
3771
3772 nb_uri = cmd_ctx->lsm->u.uri.size;
3773 len = nb_uri * sizeof(struct lttng_uri);
3774
3775 if (nb_uri > 0) {
3776 uris = zmalloc(len);
3777 if (uris == NULL) {
3778 ret = LTTNG_ERR_FATAL;
3779 goto error;
3780 }
3781
3782 /* Receive variable len data */
3783 DBG("Waiting for %zu URIs from client ...", nb_uri);
3784 ret = lttcomm_recv_unix_sock(sock, uris, len);
3785 if (ret <= 0) {
3786 DBG("No URIs received from client... continuing");
3787 *sock_error = 1;
3788 ret = LTTNG_ERR_SESSION_FAIL;
3789 free(uris);
3790 goto error;
3791 }
3792
3793 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3794 DBG("Creating session with ONE network URI is a bad call");
3795 ret = LTTNG_ERR_SESSION_FAIL;
3796 free(uris);
3797 goto error;
3798 }
3799 }
3800
3801 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3802 &cmd_ctx->creds, 0);
3803
3804 free(uris);
3805
3806 break;
3807 }
3808 case LTTNG_DESTROY_SESSION:
3809 {
3810 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3811
3812 /* Set session to NULL so we do not unlock it after free. */
3813 cmd_ctx->session = NULL;
3814 break;
3815 }
3816 case LTTNG_LIST_DOMAINS:
3817 {
3818 ssize_t nb_dom;
3819 struct lttng_domain *domains = NULL;
3820
3821 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3822 if (nb_dom < 0) {
3823 /* Return value is a negative lttng_error_code. */
3824 ret = -nb_dom;
3825 goto error;
3826 }
3827
3828 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3829 if (ret < 0) {
3830 free(domains);
3831 goto setup_error;
3832 }
3833
3834 /* Copy event list into message payload */
3835 memcpy(cmd_ctx->llm->payload, domains,
3836 nb_dom * sizeof(struct lttng_domain));
3837
3838 free(domains);
3839
3840 ret = LTTNG_OK;
3841 break;
3842 }
3843 case LTTNG_LIST_CHANNELS:
3844 {
3845 int nb_chan;
3846 struct lttng_channel *channels = NULL;
3847
3848 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3849 cmd_ctx->session, &channels);
3850 if (nb_chan < 0) {
3851 /* Return value is a negative lttng_error_code. */
3852 ret = -nb_chan;
3853 goto error;
3854 }
3855
3856 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3857 if (ret < 0) {
3858 free(channels);
3859 goto setup_error;
3860 }
3861
3862 /* Copy event list into message payload */
3863 memcpy(cmd_ctx->llm->payload, channels,
3864 nb_chan * sizeof(struct lttng_channel));
3865
3866 free(channels);
3867
3868 ret = LTTNG_OK;
3869 break;
3870 }
3871 case LTTNG_LIST_EVENTS:
3872 {
3873 ssize_t nb_event;
3874 struct lttng_event *events = NULL;
3875
3876 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3877 cmd_ctx->lsm->u.list.channel_name, &events);
3878 if (nb_event < 0) {
3879 /* Return value is a negative lttng_error_code. */
3880 ret = -nb_event;
3881 goto error;
3882 }
3883
3884 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3885 if (ret < 0) {
3886 free(events);
3887 goto setup_error;
3888 }
3889
3890 /* Copy event list into message payload */
3891 memcpy(cmd_ctx->llm->payload, events,
3892 nb_event * sizeof(struct lttng_event));
3893
3894 free(events);
3895
3896 ret = LTTNG_OK;
3897 break;
3898 }
3899 case LTTNG_LIST_SESSIONS:
3900 {
3901 unsigned int nr_sessions;
3902
3903 session_lock_list();
3904 nr_sessions = lttng_sessions_count(
3905 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3906 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3907
3908 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3909 if (ret < 0) {
3910 session_unlock_list();
3911 goto setup_error;
3912 }
3913
3914 /* Filled the session array */
3915 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3916 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3917 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3918
3919 session_unlock_list();
3920
3921 ret = LTTNG_OK;
3922 break;
3923 }
3924 case LTTNG_CALIBRATE:
3925 {
3926 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3927 &cmd_ctx->lsm->u.calibrate);
3928 break;
3929 }
3930 case LTTNG_REGISTER_CONSUMER:
3931 {
3932 struct consumer_data *cdata;
3933
3934 switch (cmd_ctx->lsm->domain.type) {
3935 case LTTNG_DOMAIN_KERNEL:
3936 cdata = &kconsumer_data;
3937 break;
3938 default:
3939 ret = LTTNG_ERR_UND;
3940 goto error;
3941 }
3942
3943 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3944 cmd_ctx->lsm->u.reg.path, cdata);
3945 break;
3946 }
3947 case LTTNG_DATA_PENDING:
3948 {
3949 int pending_ret;
3950
3951 /* 1 byte to return whether or not data is pending */
3952 ret = setup_lttng_msg(cmd_ctx, 1);
3953 if (ret < 0) {
3954 goto setup_error;
3955 }
3956
3957 pending_ret = cmd_data_pending(cmd_ctx->session);
3958 /*
3959 * FIXME
3960 *
3961 * This function may returns 0 or 1 to indicate whether or not
3962 * there is data pending. In case of error, it should return an
3963 * LTTNG_ERR code. However, some code paths may still return
3964 * a nondescript error code, which we handle by returning an
3965 * "unknown" error.
3966 */
3967 if (pending_ret == 0 || pending_ret == 1) {
3968 ret = LTTNG_OK;
3969 } else if (pending_ret < 0) {
3970 ret = LTTNG_ERR_UNK;
3971 goto setup_error;
3972 } else {
3973 ret = pending_ret;
3974 goto setup_error;
3975 }
3976
3977 *cmd_ctx->llm->payload = (uint8_t) pending_ret;
3978 break;
3979 }
3980 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3981 {
3982 struct lttcomm_lttng_output_id reply;
3983
3984 ret = cmd_snapshot_add_output(cmd_ctx->session,
3985 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3986 if (ret != LTTNG_OK) {
3987 goto error;
3988 }
3989
3990 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3991 if (ret < 0) {
3992 goto setup_error;
3993 }
3994
3995 /* Copy output list into message payload */
3996 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3997 ret = LTTNG_OK;
3998 break;
3999 }
4000 case LTTNG_SNAPSHOT_DEL_OUTPUT:
4001 {
4002 ret = cmd_snapshot_del_output(cmd_ctx->session,
4003 &cmd_ctx->lsm->u.snapshot_output.output);
4004 break;
4005 }
4006 case LTTNG_SNAPSHOT_LIST_OUTPUT:
4007 {
4008 ssize_t nb_output;
4009 struct lttng_snapshot_output *outputs = NULL;
4010
4011 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
4012 if (nb_output < 0) {
4013 ret = -nb_output;
4014 goto error;
4015 }
4016
4017 ret = setup_lttng_msg(cmd_ctx,
4018 nb_output * sizeof(struct lttng_snapshot_output));
4019 if (ret < 0) {
4020 free(outputs);
4021 goto setup_error;
4022 }
4023
4024 if (outputs) {
4025 /* Copy output list into message payload */
4026 memcpy(cmd_ctx->llm->payload, outputs,
4027 nb_output * sizeof(struct lttng_snapshot_output));
4028 free(outputs);
4029 }
4030
4031 ret = LTTNG_OK;
4032 break;
4033 }
4034 case LTTNG_SNAPSHOT_RECORD:
4035 {
4036 ret = cmd_snapshot_record(cmd_ctx->session,
4037 &cmd_ctx->lsm->u.snapshot_record.output,
4038 cmd_ctx->lsm->u.snapshot_record.wait);
4039 break;
4040 }
4041 case LTTNG_CREATE_SESSION_SNAPSHOT:
4042 {
4043 size_t nb_uri, len;
4044 struct lttng_uri *uris = NULL;
4045
4046 nb_uri = cmd_ctx->lsm->u.uri.size;
4047 len = nb_uri * sizeof(struct lttng_uri);
4048
4049 if (nb_uri > 0) {
4050 uris = zmalloc(len);
4051 if (uris == NULL) {
4052 ret = LTTNG_ERR_FATAL;
4053 goto error;
4054 }
4055
4056 /* Receive variable len data */
4057 DBG("Waiting for %zu URIs from client ...", nb_uri);
4058 ret = lttcomm_recv_unix_sock(sock, uris, len);
4059 if (ret <= 0) {
4060 DBG("No URIs received from client... continuing");
4061 *sock_error = 1;
4062 ret = LTTNG_ERR_SESSION_FAIL;
4063 free(uris);
4064 goto error;
4065 }
4066
4067 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4068 DBG("Creating session with ONE network URI is a bad call");
4069 ret = LTTNG_ERR_SESSION_FAIL;
4070 free(uris);
4071 goto error;
4072 }
4073 }
4074
4075 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4076 nb_uri, &cmd_ctx->creds);
4077 free(uris);
4078 break;
4079 }
4080 case LTTNG_CREATE_SESSION_LIVE:
4081 {
4082 size_t nb_uri, len;
4083 struct lttng_uri *uris = NULL;
4084
4085 nb_uri = cmd_ctx->lsm->u.uri.size;
4086 len = nb_uri * sizeof(struct lttng_uri);
4087
4088 if (nb_uri > 0) {
4089 uris = zmalloc(len);
4090 if (uris == NULL) {
4091 ret = LTTNG_ERR_FATAL;
4092 goto error;
4093 }
4094
4095 /* Receive variable len data */
4096 DBG("Waiting for %zu URIs from client ...", nb_uri);
4097 ret = lttcomm_recv_unix_sock(sock, uris, len);
4098 if (ret <= 0) {
4099 DBG("No URIs received from client... continuing");
4100 *sock_error = 1;
4101 ret = LTTNG_ERR_SESSION_FAIL;
4102 free(uris);
4103 goto error;
4104 }
4105
4106 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4107 DBG("Creating session with ONE network URI is a bad call");
4108 ret = LTTNG_ERR_SESSION_FAIL;
4109 free(uris);
4110 goto error;
4111 }
4112 }
4113
4114 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4115 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4116 free(uris);
4117 break;
4118 }
4119 case LTTNG_SAVE_SESSION:
4120 {
4121 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4122 &cmd_ctx->creds);
4123 break;
4124 }
4125 case LTTNG_SET_SESSION_SHM_PATH:
4126 {
4127 ret = cmd_set_session_shm_path(cmd_ctx->session,
4128 cmd_ctx->lsm->u.set_shm_path.shm_path);
4129 break;
4130 }
4131 default:
4132 ret = LTTNG_ERR_UND;
4133 break;
4134 }
4135
4136 error:
4137 if (cmd_ctx->llm == NULL) {
4138 DBG("Missing llm structure. Allocating one.");
4139 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
4140 goto setup_error;
4141 }
4142 }
4143 /* Set return code */
4144 cmd_ctx->llm->ret_code = ret;
4145 setup_error:
4146 if (cmd_ctx->session) {
4147 session_unlock(cmd_ctx->session);
4148 }
4149 if (need_tracing_session) {
4150 session_unlock_list();
4151 }
4152 init_setup_error:
4153 assert(!rcu_read_ongoing());
4154 return ret;
4155 }
4156
4157 /*
4158 * Thread managing health check socket.
4159 */
4160 static void *thread_manage_health(void *data)
4161 {
4162 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4163 uint32_t revents, nb_fd;
4164 struct lttng_poll_event events;
4165 struct health_comm_msg msg;
4166 struct health_comm_reply reply;
4167
4168 DBG("[thread] Manage health check started");
4169
4170 rcu_register_thread();
4171
4172 /* We might hit an error path before this is created. */
4173 lttng_poll_init(&events);
4174
4175 /* Create unix socket */
4176 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4177 if (sock < 0) {
4178 ERR("Unable to create health check Unix socket");
4179 ret = -1;
4180 goto error;
4181 }
4182
4183 if (is_root) {
4184 /* lttng health client socket path permissions */
4185 ret = chown(health_unix_sock_path, 0,
4186 utils_get_group_id(tracing_group_name));
4187 if (ret < 0) {
4188 ERR("Unable to set group on %s", health_unix_sock_path);
4189 PERROR("chown");
4190 ret = -1;
4191 goto error;
4192 }
4193
4194 ret = chmod(health_unix_sock_path,
4195 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4196 if (ret < 0) {
4197 ERR("Unable to set permissions on %s", health_unix_sock_path);
4198 PERROR("chmod");
4199 ret = -1;
4200 goto error;
4201 }
4202 }
4203
4204 /*
4205 * Set the CLOEXEC flag. Return code is useless because either way, the
4206 * show must go on.
4207 */
4208 (void) utils_set_fd_cloexec(sock);
4209
4210 ret = lttcomm_listen_unix_sock(sock);
4211 if (ret < 0) {
4212 goto error;
4213 }
4214
4215 /*
4216 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4217 * more will be added to this poll set.
4218 */
4219 ret = sessiond_set_thread_pollset(&events, 2);
4220 if (ret < 0) {
4221 goto error;
4222 }
4223
4224 /* Add the application registration socket */
4225 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4226 if (ret < 0) {
4227 goto error;
4228 }
4229
4230 sessiond_notify_ready();
4231
4232 while (1) {
4233 DBG("Health check ready");
4234
4235 /* Inifinite blocking call, waiting for transmission */
4236 restart:
4237 ret = lttng_poll_wait(&events, -1);
4238 if (ret < 0) {
4239 /*
4240 * Restart interrupted system call.
4241 */
4242 if (errno == EINTR) {
4243 goto restart;
4244 }
4245 goto error;
4246 }
4247
4248 nb_fd = ret;
4249
4250 for (i = 0; i < nb_fd; i++) {
4251 /* Fetch once the poll data */
4252 revents = LTTNG_POLL_GETEV(&events, i);
4253 pollfd = LTTNG_POLL_GETFD(&events, i);
4254
4255 if (!revents) {
4256 /* No activity for this FD (poll implementation). */
4257 continue;
4258 }
4259
4260 /* Thread quit pipe has been closed. Killing thread. */
4261 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4262 if (ret) {
4263 err = 0;
4264 goto exit;
4265 }
4266
4267 /* Event on the registration socket */
4268 if (pollfd == sock) {
4269 if (revents & LPOLLIN) {
4270 continue;
4271 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4272 ERR("Health socket poll error");
4273 goto error;
4274 } else {
4275 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4276 goto error;
4277 }
4278 }
4279 }
4280
4281 new_sock = lttcomm_accept_unix_sock(sock);
4282 if (new_sock < 0) {
4283 goto error;
4284 }
4285
4286 /*
4287 * Set the CLOEXEC flag. Return code is useless because either way, the
4288 * show must go on.
4289 */
4290 (void) utils_set_fd_cloexec(new_sock);
4291
4292 DBG("Receiving data from client for health...");
4293 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4294 if (ret <= 0) {
4295 DBG("Nothing recv() from client... continuing");
4296 ret = close(new_sock);
4297 if (ret) {
4298 PERROR("close");
4299 }
4300 new_sock = -1;
4301 continue;
4302 }
4303
4304 rcu_thread_online();
4305
4306 memset(&reply, 0, sizeof(reply));
4307 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4308 /*
4309 * health_check_state returns 0 if health is
4310 * bad.
4311 */
4312 if (!health_check_state(health_sessiond, i)) {
4313 reply.ret_code |= 1ULL << i;
4314 }
4315 }
4316
4317 DBG2("Health check return value %" PRIx64, reply.ret_code);
4318
4319 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4320 if (ret < 0) {
4321 ERR("Failed to send health data back to client");
4322 }
4323
4324 /* End of transmission */
4325 ret = close(new_sock);
4326 if (ret) {
4327 PERROR("close");
4328 }
4329 new_sock = -1;
4330 }
4331
4332 exit:
4333 error:
4334 if (err) {
4335 ERR("Health error occurred in %s", __func__);
4336 }
4337 DBG("Health check thread dying");
4338 unlink(health_unix_sock_path);
4339 if (sock >= 0) {
4340 ret = close(sock);
4341 if (ret) {
4342 PERROR("close");
4343 }
4344 }
4345
4346 lttng_poll_clean(&events);
4347
4348 rcu_unregister_thread();
4349 return NULL;
4350 }
4351
4352 /*
4353 * This thread manage all clients request using the unix client socket for
4354 * communication.
4355 */
4356 static void *thread_manage_clients(void *data)
4357 {
4358 int sock = -1, ret, i, pollfd, err = -1;
4359 int sock_error;
4360 uint32_t revents, nb_fd;
4361 struct command_ctx *cmd_ctx = NULL;
4362 struct lttng_poll_event events;
4363
4364 DBG("[thread] Manage client started");
4365
4366 rcu_register_thread();
4367
4368 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4369
4370 health_code_update();
4371
4372 ret = lttcomm_listen_unix_sock(client_sock);
4373 if (ret < 0) {
4374 goto error_listen;
4375 }
4376
4377 /*
4378 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4379 * more will be added to this poll set.
4380 */
4381 ret = sessiond_set_thread_pollset(&events, 2);
4382 if (ret < 0) {
4383 goto error_create_poll;
4384 }
4385
4386 /* Add the application registration socket */
4387 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4388 if (ret < 0) {
4389 goto error;
4390 }
4391
4392 sessiond_notify_ready();
4393 ret = sem_post(&load_info->message_thread_ready);
4394 if (ret) {
4395 PERROR("sem_post message_thread_ready");
4396 goto error;
4397 }
4398
4399 /* This testpoint is after we signal readiness to the parent. */
4400 if (testpoint(sessiond_thread_manage_clients)) {
4401 goto error;
4402 }
4403
4404 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4405 goto error;
4406 }
4407
4408 health_code_update();
4409
4410 while (1) {
4411 DBG("Accepting client command ...");
4412
4413 /* Inifinite blocking call, waiting for transmission */
4414 restart:
4415 health_poll_entry();
4416 ret = lttng_poll_wait(&events, -1);
4417 health_poll_exit();
4418 if (ret < 0) {
4419 /*
4420 * Restart interrupted system call.
4421 */
4422 if (errno == EINTR) {
4423 goto restart;
4424 }
4425 goto error;
4426 }
4427
4428 nb_fd = ret;
4429
4430 for (i = 0; i < nb_fd; i++) {
4431 /* Fetch once the poll data */
4432 revents = LTTNG_POLL_GETEV(&events, i);
4433 pollfd = LTTNG_POLL_GETFD(&events, i);
4434
4435 health_code_update();
4436
4437 if (!revents) {
4438 /* No activity for this FD (poll implementation). */
4439 continue;
4440 }
4441
4442 /* Thread quit pipe has been closed. Killing thread. */
4443 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4444 if (ret) {
4445 err = 0;
4446 goto exit;
4447 }
4448
4449 /* Event on the registration socket */
4450 if (pollfd == client_sock) {
4451 if (revents & LPOLLIN) {
4452 continue;
4453 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4454 ERR("Client socket poll error");
4455 goto error;
4456 } else {
4457 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4458 goto error;
4459 }
4460 }
4461 }
4462
4463 DBG("Wait for client response");
4464
4465 health_code_update();
4466
4467 sock = lttcomm_accept_unix_sock(client_sock);
4468 if (sock < 0) {
4469 goto error;
4470 }
4471
4472 /*
4473 * Set the CLOEXEC flag. Return code is useless because either way, the
4474 * show must go on.
4475 */
4476 (void) utils_set_fd_cloexec(sock);
4477
4478 /* Set socket option for credentials retrieval */
4479 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4480 if (ret < 0) {
4481 goto error;
4482 }
4483
4484 /* Allocate context command to process the client request */
4485 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4486 if (cmd_ctx == NULL) {
4487 PERROR("zmalloc cmd_ctx");
4488 goto error;
4489 }
4490
4491 /* Allocate data buffer for reception */
4492 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4493 if (cmd_ctx->lsm == NULL) {
4494 PERROR("zmalloc cmd_ctx->lsm");
4495 goto error;
4496 }
4497
4498 cmd_ctx->llm = NULL;
4499 cmd_ctx->session = NULL;
4500
4501 health_code_update();
4502
4503 /*
4504 * Data is received from the lttng client. The struct
4505 * lttcomm_session_msg (lsm) contains the command and data request of
4506 * the client.
4507 */
4508 DBG("Receiving data from client ...");
4509 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4510 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4511 if (ret <= 0) {
4512 DBG("Nothing recv() from client... continuing");
4513 ret = close(sock);
4514 if (ret) {
4515 PERROR("close");
4516 }
4517 sock = -1;
4518 clean_command_ctx(&cmd_ctx);
4519 continue;
4520 }
4521
4522 health_code_update();
4523
4524 // TODO: Validate cmd_ctx including sanity check for
4525 // security purpose.
4526
4527 rcu_thread_online();
4528 /*
4529 * This function dispatch the work to the kernel or userspace tracer
4530 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4531 * informations for the client. The command context struct contains
4532 * everything this function may needs.
4533 */
4534 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4535 rcu_thread_offline();
4536 if (ret < 0) {
4537 ret = close(sock);
4538 if (ret) {
4539 PERROR("close");
4540 }
4541 sock = -1;
4542 /*
4543 * TODO: Inform client somehow of the fatal error. At
4544 * this point, ret < 0 means that a zmalloc failed
4545 * (ENOMEM). Error detected but still accept
4546 * command, unless a socket error has been
4547 * detected.
4548 */
4549 clean_command_ctx(&cmd_ctx);
4550 continue;
4551 }
4552
4553 health_code_update();
4554
4555 DBG("Sending response (size: %d, retcode: %s (%d))",
4556 cmd_ctx->lttng_msg_size,
4557 lttng_strerror(-cmd_ctx->llm->ret_code),
4558 cmd_ctx->llm->ret_code);
4559 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4560 if (ret < 0) {
4561 ERR("Failed to send data back to client");
4562 }
4563
4564 /* End of transmission */
4565 ret = close(sock);
4566 if (ret) {
4567 PERROR("close");
4568 }
4569 sock = -1;
4570
4571 clean_command_ctx(&cmd_ctx);
4572
4573 health_code_update();
4574 }
4575
4576 exit:
4577 error:
4578 if (sock >= 0) {
4579 ret = close(sock);
4580 if (ret) {
4581 PERROR("close");
4582 }
4583 }
4584
4585 lttng_poll_clean(&events);
4586 clean_command_ctx(&cmd_ctx);
4587
4588 error_listen:
4589 error_create_poll:
4590 unlink(client_unix_sock_path);
4591 if (client_sock >= 0) {
4592 ret = close(client_sock);
4593 if (ret) {
4594 PERROR("close");
4595 }
4596 }
4597
4598 if (err) {
4599 health_error();
4600 ERR("Health error occurred in %s", __func__);
4601 }
4602
4603 health_unregister(health_sessiond);
4604
4605 DBG("Client thread dying");
4606
4607 rcu_unregister_thread();
4608
4609 /*
4610 * Since we are creating the consumer threads, we own them, so we need
4611 * to join them before our thread exits.
4612 */
4613 ret = join_consumer_thread(&kconsumer_data);
4614 if (ret) {
4615 errno = ret;
4616 PERROR("join_consumer");
4617 }
4618
4619 ret = join_consumer_thread(&ustconsumer32_data);
4620 if (ret) {
4621 errno = ret;
4622 PERROR("join_consumer ust32");
4623 }
4624
4625 ret = join_consumer_thread(&ustconsumer64_data);
4626 if (ret) {
4627 errno = ret;
4628 PERROR("join_consumer ust64");
4629 }
4630 return NULL;
4631 }
4632
4633
4634 /*
4635 * usage function on stderr
4636 */
4637 static void usage(void)
4638 {
4639 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4640 fprintf(stderr, " -h, --help Display this usage.\n");
4641 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4642 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4643 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4644 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4645 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4646 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4647 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4648 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4649 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4650 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4651 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4652 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4653 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4654 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4655 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4656 fprintf(stderr, " -V, --version Show version number.\n");
4657 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4658 fprintf(stderr, " -q, --quiet No output at all.\n");
4659 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4660 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4661 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4662 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4663 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4664 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4665 fprintf(stderr, " -l --load PATH Load session configuration\n");
4666 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4667 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4668 }
4669
4670 static int string_match(const char *str1, const char *str2)
4671 {
4672 return (str1 && str2) && !strcmp(str1, str2);
4673 }
4674
4675 /*
4676 * Take an option from the getopt output and set it in the right variable to be
4677 * used later.
4678 *
4679 * Return 0 on success else a negative value.
4680 */
4681 static int set_option(int opt, const char *arg, const char *optname)
4682 {
4683 int ret = 0;
4684
4685 if (arg && arg[0] == '\0') {
4686 /*
4687 * This only happens if the value is read from daemon config
4688 * file. This means the option requires an argument and the
4689 * configuration file contains a line such as:
4690 * my_option =
4691 */
4692 ret = -EINVAL;
4693 goto end;
4694 }
4695
4696 if (string_match(optname, "client-sock") || opt == 'c') {
4697 if (lttng_is_setuid_setgid()) {
4698 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4699 "-c, --client-sock");
4700 } else {
4701 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4702 }
4703 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4704 if (lttng_is_setuid_setgid()) {
4705 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4706 "-a, --apps-sock");
4707 } else {
4708 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4709 }
4710 } else if (string_match(optname, "daemonize") || opt == 'd') {
4711 opt_daemon = 1;
4712 } else if (string_match(optname, "background") || opt == 'b') {
4713 opt_background = 1;
4714 } else if (string_match(optname, "group") || opt == 'g') {
4715 if (lttng_is_setuid_setgid()) {
4716 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4717 "-g, --group");
4718 } else {
4719 /*
4720 * If the override option is set, the pointer points to a
4721 * *non* const thus freeing it even though the variable type is
4722 * set to const.
4723 */
4724 if (tracing_group_name_override) {
4725 free((void *) tracing_group_name);
4726 }
4727 tracing_group_name = strdup(arg);
4728 if (!tracing_group_name) {
4729 PERROR("strdup");
4730 ret = -ENOMEM;
4731 }
4732 tracing_group_name_override = 1;
4733 }
4734 } else if (string_match(optname, "help") || opt == 'h') {
4735 usage();
4736 exit(EXIT_SUCCESS);
4737 } else if (string_match(optname, "version") || opt == 'V') {
4738 fprintf(stdout, "%s\n", VERSION);
4739 exit(EXIT_SUCCESS);
4740 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4741 opt_sig_parent = 1;
4742 } else if (string_match(optname, "kconsumerd-err-sock")) {
4743 if (lttng_is_setuid_setgid()) {
4744 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4745 "--kconsumerd-err-sock");
4746 } else {
4747 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4748 }
4749 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4750 if (lttng_is_setuid_setgid()) {
4751 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4752 "--kconsumerd-cmd-sock");
4753 } else {
4754 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4755 }
4756 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4757 if (lttng_is_setuid_setgid()) {
4758 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4759 "--ustconsumerd64-err-sock");
4760 } else {
4761 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4762 }
4763 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4764 if (lttng_is_setuid_setgid()) {
4765 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4766 "--ustconsumerd64-cmd-sock");
4767 } else {
4768 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4769 }
4770 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4771 if (lttng_is_setuid_setgid()) {
4772 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4773 "--ustconsumerd32-err-sock");
4774 } else {
4775 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4776 }
4777 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4778 if (lttng_is_setuid_setgid()) {
4779 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4780 "--ustconsumerd32-cmd-sock");
4781 } else {
4782 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4783 }
4784 } else if (string_match(optname, "no-kernel")) {
4785 opt_no_kernel = 1;
4786 } else if (string_match(optname, "quiet") || opt == 'q') {
4787 lttng_opt_quiet = 1;
4788 } else if (string_match(optname, "verbose") || opt == 'v') {
4789 /* Verbose level can increase using multiple -v */
4790 if (arg) {
4791 /* Value obtained from config file */
4792 lttng_opt_verbose = config_parse_value(arg);
4793 } else {
4794 /* -v used on command line */
4795 lttng_opt_verbose++;
4796 }
4797 /* Clamp value to [0, 3] */
4798 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4799 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4800 } else if (string_match(optname, "verbose-consumer")) {
4801 if (arg) {
4802 opt_verbose_consumer = config_parse_value(arg);
4803 } else {
4804 opt_verbose_consumer += 1;
4805 }
4806 } else if (string_match(optname, "consumerd32-path")) {
4807 if (lttng_is_setuid_setgid()) {
4808 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4809 "--consumerd32-path");
4810 } else {
4811 if (consumerd32_bin_override) {
4812 free((void *) consumerd32_bin);
4813 }
4814 consumerd32_bin = strdup(arg);
4815 if (!consumerd32_bin) {
4816 PERROR("strdup");
4817 ret = -ENOMEM;
4818 }
4819 consumerd32_bin_override = 1;
4820 }
4821 } else if (string_match(optname, "consumerd32-libdir")) {
4822 if (lttng_is_setuid_setgid()) {
4823 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4824 "--consumerd32-libdir");
4825 } else {
4826 if (consumerd32_libdir_override) {
4827 free((void *) consumerd32_libdir);
4828 }
4829 consumerd32_libdir = strdup(arg);
4830 if (!consumerd32_libdir) {
4831 PERROR("strdup");
4832 ret = -ENOMEM;
4833 }
4834 consumerd32_libdir_override = 1;
4835 }
4836 } else if (string_match(optname, "consumerd64-path")) {
4837 if (lttng_is_setuid_setgid()) {
4838 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4839 "--consumerd64-path");
4840 } else {
4841 if (consumerd64_bin_override) {
4842 free((void *) consumerd64_bin);
4843 }
4844 consumerd64_bin = strdup(arg);
4845 if (!consumerd64_bin) {
4846 PERROR("strdup");
4847 ret = -ENOMEM;
4848 }
4849 consumerd64_bin_override = 1;
4850 }
4851 } else if (string_match(optname, "consumerd64-libdir")) {
4852 if (lttng_is_setuid_setgid()) {
4853 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4854 "--consumerd64-libdir");
4855 } else {
4856 if (consumerd64_libdir_override) {
4857 free((void *) consumerd64_libdir);
4858 }
4859 consumerd64_libdir = strdup(arg);
4860 if (!consumerd64_libdir) {
4861 PERROR("strdup");
4862 ret = -ENOMEM;
4863 }
4864 consumerd64_libdir_override = 1;
4865 }
4866 } else if (string_match(optname, "pidfile") || opt == 'p') {
4867 if (lttng_is_setuid_setgid()) {
4868 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4869 "-p, --pidfile");
4870 } else {
4871 free(opt_pidfile);
4872 opt_pidfile = strdup(arg);
4873 if (!opt_pidfile) {
4874 PERROR("strdup");
4875 ret = -ENOMEM;
4876 }
4877 }
4878 } else if (string_match(optname, "agent-tcp-port")) {
4879 if (lttng_is_setuid_setgid()) {
4880 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4881 "--agent-tcp-port");
4882 } else {
4883 unsigned long v;
4884
4885 if (!arg) {
4886 ret = -EINVAL;
4887 goto end;
4888 }
4889 errno = 0;
4890 v = strtoul(arg, NULL, 0);
4891 if (errno != 0 || !isdigit(arg[0])) {
4892 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4893 return -1;
4894 }
4895 if (v == 0 || v >= 65535) {
4896 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4897 return -1;
4898 }
4899 agent_tcp_port = (uint32_t) v;
4900 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4901 }
4902 } else if (string_match(optname, "load") || opt == 'l') {
4903 if (lttng_is_setuid_setgid()) {
4904 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4905 "-l, --load");
4906 } else {
4907 free(opt_load_session_path);
4908 opt_load_session_path = strdup(arg);
4909 if (!opt_load_session_path) {
4910 PERROR("strdup");
4911 ret = -ENOMEM;
4912 }
4913 }
4914 } else if (string_match(optname, "kmod-probes")) {
4915 if (lttng_is_setuid_setgid()) {
4916 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4917 "--kmod-probes");
4918 } else {
4919 free(kmod_probes_list);
4920 kmod_probes_list = strdup(arg);
4921 if (!kmod_probes_list) {
4922 PERROR("strdup");
4923 ret = -ENOMEM;
4924 }
4925 }
4926 } else if (string_match(optname, "extra-kmod-probes")) {
4927 if (lttng_is_setuid_setgid()) {
4928 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4929 "--extra-kmod-probes");
4930 } else {
4931 free(kmod_extra_probes_list);
4932 kmod_extra_probes_list = strdup(arg);
4933 if (!kmod_extra_probes_list) {
4934 PERROR("strdup");
4935 ret = -ENOMEM;
4936 }
4937 }
4938 } else if (string_match(optname, "config") || opt == 'f') {
4939 /* This is handled in set_options() thus silent skip. */
4940 goto end;
4941 } else {
4942 /* Unknown option or other error.
4943 * Error is printed by getopt, just return */
4944 ret = -1;
4945 }
4946
4947 end:
4948 if (ret == -EINVAL) {
4949 const char *opt_name = "unknown";
4950 int i;
4951
4952 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4953 i++) {
4954 if (opt == long_options[i].val) {
4955 opt_name = long_options[i].name;
4956 break;
4957 }
4958 }
4959
4960 WARN("Invalid argument provided for option \"%s\", using default value.",
4961 opt_name);
4962 }
4963
4964 return ret;
4965 }
4966
4967 /*
4968 * config_entry_handler_cb used to handle options read from a config file.
4969 * See config_entry_handler_cb comment in common/config/session-config.h for the
4970 * return value conventions.
4971 */
4972 static int config_entry_handler(const struct config_entry *entry, void *unused)
4973 {
4974 int ret = 0, i;
4975
4976 if (!entry || !entry->name || !entry->value) {
4977 ret = -EINVAL;
4978 goto end;
4979 }
4980
4981 /* Check if the option is to be ignored */
4982 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4983 if (!strcmp(entry->name, config_ignore_options[i])) {
4984 goto end;
4985 }
4986 }
4987
4988 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4989 i++) {
4990
4991 /* Ignore if not fully matched. */
4992 if (strcmp(entry->name, long_options[i].name)) {
4993 continue;
4994 }
4995
4996 /*
4997 * If the option takes no argument on the command line, we have to
4998 * check if the value is "true". We support non-zero numeric values,
4999 * true, on and yes.
5000 */
5001 if (!long_options[i].has_arg) {
5002 ret = config_parse_value(entry->value);
5003 if (ret <= 0) {
5004 if (ret) {
5005 WARN("Invalid configuration value \"%s\" for option %s",
5006 entry->value, entry->name);
5007 }
5008 /* False, skip boolean config option. */
5009 goto end;
5010 }
5011 }
5012
5013 ret = set_option(long_options[i].val, entry->value, entry->name);
5014 goto end;
5015 }
5016
5017 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5018
5019 end:
5020 return ret;
5021 }
5022
5023 /*
5024 * daemon configuration loading and argument parsing
5025 */
5026 static int set_options(int argc, char **argv)
5027 {
5028 int ret = 0, c = 0, option_index = 0;
5029 int orig_optopt = optopt, orig_optind = optind;
5030 char *optstring;
5031 const char *config_path = NULL;
5032
5033 optstring = utils_generate_optstring(long_options,
5034 sizeof(long_options) / sizeof(struct option));
5035 if (!optstring) {
5036 ret = -ENOMEM;
5037 goto end;
5038 }
5039
5040 /* Check for the --config option */
5041 while ((c = getopt_long(argc, argv, optstring, long_options,
5042 &option_index)) != -1) {
5043 if (c == '?') {
5044 ret = -EINVAL;
5045 goto end;
5046 } else if (c != 'f') {
5047 /* if not equal to --config option. */
5048 continue;
5049 }
5050
5051 if (lttng_is_setuid_setgid()) {
5052 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5053 "-f, --config");
5054 } else {
5055 config_path = utils_expand_path(optarg);
5056 if (!config_path) {
5057 ERR("Failed to resolve path: %s", optarg);
5058 }
5059 }
5060 }
5061
5062 ret = config_get_section_entries(config_path, config_section_name,
5063 config_entry_handler, NULL);
5064 if (ret) {
5065 if (ret > 0) {
5066 ERR("Invalid configuration option at line %i", ret);
5067 ret = -1;
5068 }
5069 goto end;
5070 }
5071
5072 /* Reset getopt's global state */
5073 optopt = orig_optopt;
5074 optind = orig_optind;
5075 while (1) {
5076 option_index = -1;
5077 /*
5078 * getopt_long() will not set option_index if it encounters a
5079 * short option.
5080 */
5081 c = getopt_long(argc, argv, optstring, long_options,
5082 &option_index);
5083 if (c == -1) {
5084 break;
5085 }
5086
5087 /*
5088 * Pass NULL as the long option name if popt left the index
5089 * unset.
5090 */
5091 ret = set_option(c, optarg,
5092 option_index < 0 ? NULL :
5093 long_options[option_index].name);
5094 if (ret < 0) {
5095 break;
5096 }
5097 }
5098
5099 end:
5100 free(optstring);
5101 return ret;
5102 }
5103
5104 /*
5105 * Creates the two needed socket by the daemon.
5106 * apps_sock - The communication socket for all UST apps.
5107 * client_sock - The communication of the cli tool (lttng).
5108 */
5109 static int init_daemon_socket(void)
5110 {
5111 int ret = 0;
5112 mode_t old_umask;
5113
5114 old_umask = umask(0);
5115
5116 /* Create client tool unix socket */
5117 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5118 if (client_sock < 0) {
5119 ERR("Create unix sock failed: %s", client_unix_sock_path);
5120 ret = -1;
5121 goto end;
5122 }
5123
5124 /* Set the cloexec flag */
5125 ret = utils_set_fd_cloexec(client_sock);
5126 if (ret < 0) {
5127 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5128 "Continuing but note that the consumer daemon will have a "
5129 "reference to this socket on exec()", client_sock);
5130 }
5131
5132 /* File permission MUST be 660 */
5133 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5134 if (ret < 0) {
5135 ERR("Set file permissions failed: %s", client_unix_sock_path);
5136 PERROR("chmod");
5137 goto end;
5138 }
5139
5140 /* Create the application unix socket */
5141 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5142 if (apps_sock < 0) {
5143 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5144 ret = -1;
5145 goto end;
5146 }
5147
5148 /* Set the cloexec flag */
5149 ret = utils_set_fd_cloexec(apps_sock);
5150 if (ret < 0) {
5151 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5152 "Continuing but note that the consumer daemon will have a "
5153 "reference to this socket on exec()", apps_sock);
5154 }
5155
5156 /* File permission MUST be 666 */
5157 ret = chmod(apps_unix_sock_path,
5158 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5159 if (ret < 0) {
5160 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5161 PERROR("chmod");
5162 goto end;
5163 }
5164
5165 DBG3("Session daemon client socket %d and application socket %d created",
5166 client_sock, apps_sock);
5167
5168 end:
5169 umask(old_umask);
5170 return ret;
5171 }
5172
5173 /*
5174 * Check if the global socket is available, and if a daemon is answering at the
5175 * other side. If yes, error is returned.
5176 */
5177 static int check_existing_daemon(void)
5178 {
5179 /* Is there anybody out there ? */
5180 if (lttng_session_daemon_alive()) {
5181 return -EEXIST;
5182 }
5183
5184 return 0;
5185 }
5186
5187 /*
5188 * Set the tracing group gid onto the client socket.
5189 *
5190 * Race window between mkdir and chown is OK because we are going from more
5191 * permissive (root.root) to less permissive (root.tracing).
5192 */
5193 static int set_permissions(char *rundir)
5194 {
5195 int ret;
5196 gid_t gid;
5197
5198 gid = utils_get_group_id(tracing_group_name);
5199
5200 /* Set lttng run dir */
5201 ret = chown(rundir, 0, gid);
5202 if (ret < 0) {
5203 ERR("Unable to set group on %s", rundir);
5204 PERROR("chown");
5205 }
5206
5207 /*
5208 * Ensure all applications and tracing group can search the run
5209 * dir. Allow everyone to read the directory, since it does not
5210 * buy us anything to hide its content.
5211 */
5212 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5213 if (ret < 0) {
5214 ERR("Unable to set permissions on %s", rundir);
5215 PERROR("chmod");
5216 }
5217
5218 /* lttng client socket path */
5219 ret = chown(client_unix_sock_path, 0, gid);
5220 if (ret < 0) {
5221 ERR("Unable to set group on %s", client_unix_sock_path);
5222 PERROR("chown");
5223 }
5224
5225 /* kconsumer error socket path */
5226 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5227 if (ret < 0) {
5228 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5229 PERROR("chown");
5230 }
5231
5232 /* 64-bit ustconsumer error socket path */
5233 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5234 if (ret < 0) {
5235 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5236 PERROR("chown");
5237 }
5238
5239 /* 32-bit ustconsumer compat32 error socket path */
5240 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5241 if (ret < 0) {
5242 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5243 PERROR("chown");
5244 }
5245
5246 DBG("All permissions are set");
5247
5248 return ret;
5249 }
5250
5251 /*
5252 * Create the lttng run directory needed for all global sockets and pipe.
5253 */
5254 static int create_lttng_rundir(const char *rundir)
5255 {
5256 int ret;
5257
5258 DBG3("Creating LTTng run directory: %s", rundir);
5259
5260 ret = mkdir(rundir, S_IRWXU);
5261 if (ret < 0) {
5262 if (errno != EEXIST) {
5263 ERR("Unable to create %s", rundir);
5264 goto error;
5265 } else {
5266 ret = 0;
5267 }
5268 }
5269
5270 error:
5271 return ret;
5272 }
5273
5274 /*
5275 * Setup sockets and directory needed by the kconsumerd communication with the
5276 * session daemon.
5277 */
5278 static int set_consumer_sockets(struct consumer_data *consumer_data,
5279 const char *rundir)
5280 {
5281 int ret;
5282 char path[PATH_MAX];
5283
5284 switch (consumer_data->type) {
5285 case LTTNG_CONSUMER_KERNEL:
5286 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5287 break;
5288 case LTTNG_CONSUMER64_UST:
5289 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5290 break;
5291 case LTTNG_CONSUMER32_UST:
5292 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5293 break;
5294 default:
5295 ERR("Consumer type unknown");
5296 ret = -EINVAL;
5297 goto error;
5298 }
5299
5300 DBG2("Creating consumer directory: %s", path);
5301
5302 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5303 if (ret < 0) {
5304 if (errno != EEXIST) {
5305 PERROR("mkdir");
5306 ERR("Failed to create %s", path);
5307 goto error;
5308 }
5309 ret = -1;
5310 }
5311 if (is_root) {
5312 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5313 if (ret < 0) {
5314 ERR("Unable to set group on %s", path);
5315 PERROR("chown");
5316 goto error;
5317 }
5318 }
5319
5320 /* Create the kconsumerd error unix socket */
5321 consumer_data->err_sock =
5322 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5323 if (consumer_data->err_sock < 0) {
5324 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5325 ret = -1;
5326 goto error;
5327 }
5328
5329 /*
5330 * Set the CLOEXEC flag. Return code is useless because either way, the
5331 * show must go on.
5332 */
5333 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5334 if (ret < 0) {
5335 PERROR("utils_set_fd_cloexec");
5336 /* continue anyway */
5337 }
5338
5339 /* File permission MUST be 660 */
5340 ret = chmod(consumer_data->err_unix_sock_path,
5341 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5342 if (ret < 0) {
5343 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5344 PERROR("chmod");
5345 goto error;
5346 }
5347
5348 error:
5349 return ret;
5350 }
5351
5352 /*
5353 * Signal handler for the daemon
5354 *
5355 * Simply stop all worker threads, leaving main() return gracefully after
5356 * joining all threads and calling cleanup().
5357 */
5358 static void sighandler(int sig)
5359 {
5360 switch (sig) {
5361 case SIGPIPE:
5362 DBG("SIGPIPE caught");
5363 return;
5364 case SIGINT:
5365 DBG("SIGINT caught");
5366 stop_threads();
5367 break;
5368 case SIGTERM:
5369 DBG("SIGTERM caught");
5370 stop_threads();
5371 break;
5372 case SIGUSR1:
5373 CMM_STORE_SHARED(recv_child_signal, 1);
5374 break;
5375 default:
5376 break;
5377 }
5378 }
5379
5380 /*
5381 * Setup signal handler for :
5382 * SIGINT, SIGTERM, SIGPIPE
5383 */
5384 static int set_signal_handler(void)
5385 {
5386 int ret = 0;
5387 struct sigaction sa;
5388 sigset_t sigset;
5389
5390 if ((ret = sigemptyset(&sigset)) < 0) {
5391 PERROR("sigemptyset");
5392 return ret;
5393 }
5394
5395 sa.sa_handler = sighandler;
5396 sa.sa_mask = sigset;
5397 sa.sa_flags = 0;
5398 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5399 PERROR("sigaction");
5400 return ret;
5401 }
5402
5403 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5404 PERROR("sigaction");
5405 return ret;
5406 }
5407
5408 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5409 PERROR("sigaction");
5410 return ret;
5411 }
5412
5413 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5414 PERROR("sigaction");
5415 return ret;
5416 }
5417
5418 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5419
5420 return ret;
5421 }
5422
5423 /*
5424 * Set open files limit to unlimited. This daemon can open a large number of
5425 * file descriptors in order to consumer multiple kernel traces.
5426 */
5427 static void set_ulimit(void)
5428 {
5429 int ret;
5430 struct rlimit lim;
5431
5432 /* The kernel does not allowed an infinite limit for open files */
5433 lim.rlim_cur = 65535;
5434 lim.rlim_max = 65535;
5435
5436 ret = setrlimit(RLIMIT_NOFILE, &lim);
5437 if (ret < 0) {
5438 PERROR("failed to set open files limit");
5439 }
5440 }
5441
5442 /*
5443 * Write pidfile using the rundir and opt_pidfile.
5444 */
5445 static int write_pidfile(void)
5446 {
5447 int ret;
5448 char pidfile_path[PATH_MAX];
5449
5450 assert(rundir);
5451
5452 if (opt_pidfile) {
5453 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5454 } else {
5455 /* Build pidfile path from rundir and opt_pidfile. */
5456 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5457 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5458 if (ret < 0) {
5459 PERROR("snprintf pidfile path");
5460 goto error;
5461 }
5462 }
5463
5464 /*
5465 * Create pid file in rundir.
5466 */
5467 ret = utils_create_pid_file(getpid(), pidfile_path);
5468 error:
5469 return ret;
5470 }
5471
5472 /*
5473 * Create lockfile using the rundir and return its fd.
5474 */
5475 static int create_lockfile(void)
5476 {
5477 int ret;
5478 char lockfile_path[PATH_MAX];
5479
5480 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5481 if (ret < 0) {
5482 goto error;
5483 }
5484
5485 ret = utils_create_lock_file(lockfile_path);
5486 error:
5487 return ret;
5488 }
5489
5490 /*
5491 * Write agent TCP port using the rundir.
5492 */
5493 static int write_agent_port(void)
5494 {
5495 int ret;
5496 char path[PATH_MAX];
5497
5498 assert(rundir);
5499
5500 ret = snprintf(path, sizeof(path), "%s/"
5501 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5502 if (ret < 0) {
5503 PERROR("snprintf agent port path");
5504 goto error;
5505 }
5506
5507 /*
5508 * Create TCP agent port file in rundir.
5509 */
5510 ret = utils_create_pid_file(agent_tcp_port, path);
5511
5512 error:
5513 return ret;
5514 }
5515
5516 /*
5517 * main
5518 */
5519 int main(int argc, char **argv)
5520 {
5521 int ret = 0, retval = 0;
5522 void *status;
5523 const char *home_path, *env_app_timeout;
5524
5525 init_kernel_workarounds();
5526
5527 rcu_register_thread();
5528
5529 if (set_signal_handler()) {
5530 retval = -1;
5531 goto exit_set_signal_handler;
5532 }
5533
5534 setup_consumerd_path();
5535
5536 page_size = sysconf(_SC_PAGESIZE);
5537 if (page_size < 0) {
5538 PERROR("sysconf _SC_PAGESIZE");
5539 page_size = LONG_MAX;
5540 WARN("Fallback page size to %ld", page_size);
5541 }
5542
5543 /*
5544 * Parse arguments and load the daemon configuration file.
5545 *
5546 * We have an exit_options exit path to free memory reserved by
5547 * set_options. This is needed because the rest of sessiond_cleanup()
5548 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5549 * depends on set_options.
5550 */
5551 progname = argv[0];
5552 if (set_options(argc, argv)) {
5553 retval = -1;
5554 goto exit_options;
5555 }
5556
5557 /* Daemonize */
5558 if (opt_daemon || opt_background) {
5559 int i;
5560
5561 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5562 !opt_background);
5563 if (ret < 0) {
5564 retval = -1;
5565 goto exit_options;
5566 }
5567
5568 /*
5569 * We are in the child. Make sure all other file descriptors are
5570 * closed, in case we are called with more opened file
5571 * descriptors than the standard ones.
5572 */
5573 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5574 (void) close(i);
5575 }
5576 }
5577
5578 if (run_as_create_worker(argv[0]) < 0) {
5579 goto exit_create_run_as_worker_cleanup;
5580 }
5581
5582 /*
5583 * Starting from here, we can create threads. This needs to be after
5584 * lttng_daemonize due to RCU.
5585 */
5586
5587 /*
5588 * Initialize the health check subsystem. This call should set the
5589 * appropriate time values.
5590 */
5591 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5592 if (!health_sessiond) {
5593 PERROR("health_app_create error");
5594 retval = -1;
5595 goto exit_health_sessiond_cleanup;
5596 }
5597
5598 if (init_ht_cleanup_quit_pipe()) {
5599 retval = -1;
5600 goto exit_ht_cleanup_quit_pipe;
5601 }
5602
5603 /* Setup the thread ht_cleanup communication pipe. */
5604 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5605 retval = -1;
5606 goto exit_ht_cleanup_pipe;
5607 }
5608
5609 /* Set up max poll set size */
5610 if (lttng_poll_set_max_size()) {
5611 retval = -1;
5612 goto exit_set_max_size;
5613 }
5614
5615 /* Create thread to clean up RCU hash tables */
5616 ret = pthread_create(&ht_cleanup_thread, NULL,
5617 thread_ht_cleanup, (void *) NULL);
5618 if (ret) {
5619 errno = ret;
5620 PERROR("pthread_create ht_cleanup");
5621 retval = -1;
5622 goto exit_ht_cleanup;
5623 }
5624
5625 /* Create thread quit pipe */
5626 if (init_thread_quit_pipe()) {
5627 retval = -1;
5628 goto exit_init_data;
5629 }
5630
5631 /* Check if daemon is UID = 0 */
5632 is_root = !getuid();
5633
5634 if (is_root) {
5635 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5636 if (!rundir) {
5637 retval = -1;
5638 goto exit_init_data;
5639 }
5640
5641 /* Create global run dir with root access */
5642 if (create_lttng_rundir(rundir)) {
5643 retval = -1;
5644 goto exit_init_data;
5645 }
5646
5647 if (strlen(apps_unix_sock_path) == 0) {
5648 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5649 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5650 if (ret < 0) {
5651 retval = -1;
5652 goto exit_init_data;
5653 }
5654 }
5655
5656 if (strlen(client_unix_sock_path) == 0) {
5657 ret = snprintf(client_unix_sock_path, PATH_MAX,
5658 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5659 if (ret < 0) {
5660 retval = -1;
5661 goto exit_init_data;
5662 }
5663 }
5664
5665 /* Set global SHM for ust */
5666 if (strlen(wait_shm_path) == 0) {
5667 ret = snprintf(wait_shm_path, PATH_MAX,
5668 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5669 if (ret < 0) {
5670 retval = -1;
5671 goto exit_init_data;
5672 }
5673 }
5674
5675 if (strlen(health_unix_sock_path) == 0) {
5676 ret = snprintf(health_unix_sock_path,
5677 sizeof(health_unix_sock_path),
5678 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5679 if (ret < 0) {
5680 retval = -1;
5681 goto exit_init_data;
5682 }
5683 }
5684
5685 /* Setup kernel consumerd path */
5686 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5687 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5688 if (ret < 0) {
5689 retval = -1;
5690 goto exit_init_data;
5691 }
5692 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5693 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5694 if (ret < 0) {
5695 retval = -1;
5696 goto exit_init_data;
5697 }
5698
5699 DBG2("Kernel consumer err path: %s",
5700 kconsumer_data.err_unix_sock_path);
5701 DBG2("Kernel consumer cmd path: %s",
5702 kconsumer_data.cmd_unix_sock_path);
5703 } else {
5704 home_path = utils_get_home_dir();
5705 if (home_path == NULL) {
5706 /* TODO: Add --socket PATH option */
5707 ERR("Can't get HOME directory for sockets creation.");
5708 retval = -1;
5709 goto exit_init_data;
5710 }
5711
5712 /*
5713 * Create rundir from home path. This will create something like
5714 * $HOME/.lttng
5715 */
5716 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5717 if (ret < 0) {
5718 retval = -1;
5719 goto exit_init_data;
5720 }
5721
5722 if (create_lttng_rundir(rundir)) {
5723 retval = -1;
5724 goto exit_init_data;
5725 }
5726
5727 if (strlen(apps_unix_sock_path) == 0) {
5728 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5729 DEFAULT_HOME_APPS_UNIX_SOCK,
5730 home_path);
5731 if (ret < 0) {
5732 retval = -1;
5733 goto exit_init_data;
5734 }
5735 }
5736
5737 /* Set the cli tool unix socket path */
5738 if (strlen(client_unix_sock_path) == 0) {
5739 ret = snprintf(client_unix_sock_path, PATH_MAX,
5740 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5741 home_path);
5742 if (ret < 0) {
5743 retval = -1;
5744 goto exit_init_data;
5745 }
5746 }
5747
5748 /* Set global SHM for ust */
5749 if (strlen(wait_shm_path) == 0) {
5750 ret = snprintf(wait_shm_path, PATH_MAX,
5751 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5752 getuid());
5753 if (ret < 0) {
5754 retval = -1;
5755 goto exit_init_data;
5756 }
5757 }
5758
5759 /* Set health check Unix path */
5760 if (strlen(health_unix_sock_path) == 0) {
5761 ret = snprintf(health_unix_sock_path,
5762 sizeof(health_unix_sock_path),
5763 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5764 home_path);
5765 if (ret < 0) {
5766 retval = -1;
5767 goto exit_init_data;
5768 }
5769 }
5770 }
5771
5772 lockfile_fd = create_lockfile();
5773 if (lockfile_fd < 0) {
5774 retval = -1;
5775 goto exit_init_data;
5776 }
5777
5778 /* Set consumer initial state */
5779 kernel_consumerd_state = CONSUMER_STOPPED;
5780 ust_consumerd_state = CONSUMER_STOPPED;
5781
5782 DBG("Client socket path %s", client_unix_sock_path);
5783 DBG("Application socket path %s", apps_unix_sock_path);
5784 DBG("Application wait path %s", wait_shm_path);
5785 DBG("LTTng run directory path: %s", rundir);
5786
5787 /* 32 bits consumerd path setup */
5788 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5789 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5790 if (ret < 0) {
5791 PERROR("snprintf 32-bit consumer error socket path");
5792 retval = -1;
5793 goto exit_init_data;
5794 }
5795 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5796 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5797 if (ret < 0) {
5798 PERROR("snprintf 32-bit consumer command socket path");
5799 retval = -1;
5800 goto exit_init_data;
5801 }
5802
5803 DBG2("UST consumer 32 bits err path: %s",
5804 ustconsumer32_data.err_unix_sock_path);
5805 DBG2("UST consumer 32 bits cmd path: %s",
5806 ustconsumer32_data.cmd_unix_sock_path);
5807
5808 /* 64 bits consumerd path setup */
5809 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5810 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5811 if (ret < 0) {
5812 PERROR("snprintf 64-bit consumer error socket path");
5813 retval = -1;
5814 goto exit_init_data;
5815 }
5816 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5817 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5818 if (ret < 0) {
5819 PERROR("snprintf 64-bit consumer command socket path");
5820 retval = -1;
5821 goto exit_init_data;
5822 }
5823
5824 DBG2("UST consumer 64 bits err path: %s",
5825 ustconsumer64_data.err_unix_sock_path);
5826 DBG2("UST consumer 64 bits cmd path: %s",
5827 ustconsumer64_data.cmd_unix_sock_path);
5828
5829 /*
5830 * See if daemon already exist.
5831 */
5832 if (check_existing_daemon()) {
5833 ERR("Already running daemon.\n");
5834 /*
5835 * We do not goto exit because we must not cleanup()
5836 * because a daemon is already running.
5837 */
5838 retval = -1;
5839 goto exit_init_data;
5840 }
5841
5842 /*
5843 * Init UST app hash table. Alloc hash table before this point since
5844 * cleanup() can get called after that point.
5845 */
5846 if (ust_app_ht_alloc()) {
5847 ERR("Failed to allocate UST app hash table");
5848 retval = -1;
5849 goto exit_init_data;
5850 }
5851
5852 /*
5853 * Initialize agent app hash table. We allocate the hash table here
5854 * since cleanup() can get called after this point.
5855 */
5856 if (agent_app_ht_alloc()) {
5857 ERR("Failed to allocate Agent app hash table");
5858 retval = -1;
5859 goto exit_init_data;
5860 }
5861
5862 /*
5863 * These actions must be executed as root. We do that *after* setting up
5864 * the sockets path because we MUST make the check for another daemon using
5865 * those paths *before* trying to set the kernel consumer sockets and init
5866 * kernel tracer.
5867 */
5868 if (is_root) {
5869 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5870 retval = -1;
5871 goto exit_init_data;
5872 }
5873
5874 /* Setup kernel tracer */
5875 if (!opt_no_kernel) {
5876 init_kernel_tracer();
5877 if (kernel_tracer_fd >= 0) {
5878 ret = syscall_init_table();
5879 if (ret < 0) {
5880 ERR("Unable to populate syscall table. "
5881 "Syscall tracing won't work "
5882 "for this session daemon.");
5883 }
5884 }
5885 }
5886
5887 /* Set ulimit for open files */
5888 set_ulimit();
5889 }
5890 /* init lttng_fd tracking must be done after set_ulimit. */
5891 lttng_fd_init();
5892
5893 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5894 retval = -1;
5895 goto exit_init_data;
5896 }
5897
5898 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5899 retval = -1;
5900 goto exit_init_data;
5901 }
5902
5903 /* Setup the needed unix socket */
5904 if (init_daemon_socket()) {
5905 retval = -1;
5906 goto exit_init_data;
5907 }
5908
5909 /* Set credentials to socket */
5910 if (is_root && set_permissions(rundir)) {
5911 retval = -1;
5912 goto exit_init_data;
5913 }
5914
5915 /* Get parent pid if -S, --sig-parent is specified. */
5916 if (opt_sig_parent) {
5917 ppid = getppid();
5918 }
5919
5920 /* Setup the kernel pipe for waking up the kernel thread */
5921 if (is_root && !opt_no_kernel) {
5922 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5923 retval = -1;
5924 goto exit_init_data;
5925 }
5926 }
5927
5928 /* Setup the thread apps communication pipe. */
5929 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5930 retval = -1;
5931 goto exit_init_data;
5932 }
5933
5934 /* Setup the thread apps notify communication pipe. */
5935 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5936 retval = -1;
5937 goto exit_init_data;
5938 }
5939
5940 /* Initialize global buffer per UID and PID registry. */
5941 buffer_reg_init_uid_registry();
5942 buffer_reg_init_pid_registry();
5943
5944 /* Init UST command queue. */
5945 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5946
5947 /*
5948 * Get session list pointer. This pointer MUST NOT be free'd. This list
5949 * is statically declared in session.c
5950 */
5951 session_list_ptr = session_get_list();
5952
5953 cmd_init();
5954
5955 /* Check for the application socket timeout env variable. */
5956 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5957 if (env_app_timeout) {
5958 app_socket_timeout = atoi(env_app_timeout);
5959 } else {
5960 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5961 }
5962
5963 ret = write_pidfile();
5964 if (ret) {
5965 ERR("Error in write_pidfile");
5966 retval = -1;
5967 goto exit_init_data;
5968 }
5969 ret = write_agent_port();
5970 if (ret) {
5971 ERR("Error in write_agent_port");
5972 retval = -1;
5973 goto exit_init_data;
5974 }
5975
5976 /* Initialize communication library */
5977 lttcomm_init();
5978 /* Initialize TCP timeout values */
5979 lttcomm_inet_init();
5980
5981 if (load_session_init_data(&load_info) < 0) {
5982 retval = -1;
5983 goto exit_init_data;
5984 }
5985 load_info->path = opt_load_session_path;
5986
5987 /* Create health-check thread */
5988 ret = pthread_create(&health_thread, NULL,
5989 thread_manage_health, (void *) NULL);
5990 if (ret) {
5991 errno = ret;
5992 PERROR("pthread_create health");
5993 retval = -1;
5994 goto exit_health;
5995 }
5996
5997 /* Create thread to manage the client socket */
5998 ret = pthread_create(&client_thread, NULL,
5999 thread_manage_clients, (void *) NULL);
6000 if (ret) {
6001 errno = ret;
6002 PERROR("pthread_create clients");
6003 retval = -1;
6004 goto exit_client;
6005 }
6006
6007 /* Create thread to dispatch registration */
6008 ret = pthread_create(&dispatch_thread, NULL,
6009 thread_dispatch_ust_registration, (void *) NULL);
6010 if (ret) {
6011 errno = ret;
6012 PERROR("pthread_create dispatch");
6013 retval = -1;
6014 goto exit_dispatch;
6015 }
6016
6017 /* Create thread to manage application registration. */
6018 ret = pthread_create(&reg_apps_thread, NULL,
6019 thread_registration_apps, (void *) NULL);
6020 if (ret) {
6021 errno = ret;
6022 PERROR("pthread_create registration");
6023 retval = -1;
6024 goto exit_reg_apps;
6025 }
6026
6027 /* Create thread to manage application socket */
6028 ret = pthread_create(&apps_thread, NULL,
6029 thread_manage_apps, (void *) NULL);
6030 if (ret) {
6031 errno = ret;
6032 PERROR("pthread_create apps");
6033 retval = -1;
6034 goto exit_apps;
6035 }
6036
6037 /* Create thread to manage application notify socket */
6038 ret = pthread_create(&apps_notify_thread, NULL,
6039 ust_thread_manage_notify, (void *) NULL);
6040 if (ret) {
6041 errno = ret;
6042 PERROR("pthread_create notify");
6043 retval = -1;
6044 goto exit_apps_notify;
6045 }
6046
6047 /* Create agent registration thread. */
6048 ret = pthread_create(&agent_reg_thread, NULL,
6049 agent_thread_manage_registration, (void *) NULL);
6050 if (ret) {
6051 errno = ret;
6052 PERROR("pthread_create agent");
6053 retval = -1;
6054 goto exit_agent_reg;
6055 }
6056
6057 /* Don't start this thread if kernel tracing is not requested nor root */
6058 if (is_root && !opt_no_kernel) {
6059 /* Create kernel thread to manage kernel event */
6060 ret = pthread_create(&kernel_thread, NULL,
6061 thread_manage_kernel, (void *) NULL);
6062 if (ret) {
6063 errno = ret;
6064 PERROR("pthread_create kernel");
6065 retval = -1;
6066 goto exit_kernel;
6067 }
6068 }
6069
6070 /* Create session loading thread. */
6071 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
6072 load_info);
6073 if (ret) {
6074 errno = ret;
6075 PERROR("pthread_create load_session_thread");
6076 retval = -1;
6077 goto exit_load_session;
6078 }
6079
6080 /*
6081 * This is where we start awaiting program completion (e.g. through
6082 * signal that asks threads to teardown).
6083 */
6084
6085 ret = pthread_join(load_session_thread, &status);
6086 if (ret) {
6087 errno = ret;
6088 PERROR("pthread_join load_session_thread");
6089 retval = -1;
6090 }
6091 exit_load_session:
6092
6093 if (is_root && !opt_no_kernel) {
6094 ret = pthread_join(kernel_thread, &status);
6095 if (ret) {
6096 errno = ret;
6097 PERROR("pthread_join");
6098 retval = -1;
6099 }
6100 }
6101 exit_kernel:
6102
6103 ret = pthread_join(agent_reg_thread, &status);
6104 if (ret) {
6105 errno = ret;
6106 PERROR("pthread_join agent");
6107 retval = -1;
6108 }
6109 exit_agent_reg:
6110
6111 ret = pthread_join(apps_notify_thread, &status);
6112 if (ret) {
6113 errno = ret;
6114 PERROR("pthread_join apps notify");
6115 retval = -1;
6116 }
6117 exit_apps_notify:
6118
6119 ret = pthread_join(apps_thread, &status);
6120 if (ret) {
6121 errno = ret;
6122 PERROR("pthread_join apps");
6123 retval = -1;
6124 }
6125 exit_apps:
6126
6127 ret = pthread_join(reg_apps_thread, &status);
6128 if (ret) {
6129 errno = ret;
6130 PERROR("pthread_join");
6131 retval = -1;
6132 }
6133 exit_reg_apps:
6134
6135 /*
6136 * Join dispatch thread after joining reg_apps_thread to ensure
6137 * we don't leak applications in the queue.
6138 */
6139 ret = pthread_join(dispatch_thread, &status);
6140 if (ret) {
6141 errno = ret;
6142 PERROR("pthread_join");
6143 retval = -1;
6144 }
6145 exit_dispatch:
6146
6147 ret = pthread_join(client_thread, &status);
6148 if (ret) {
6149 errno = ret;
6150 PERROR("pthread_join");
6151 retval = -1;
6152 }
6153 exit_client:
6154
6155 ret = pthread_join(health_thread, &status);
6156 if (ret) {
6157 errno = ret;
6158 PERROR("pthread_join health thread");
6159 retval = -1;
6160 }
6161 exit_health:
6162
6163 exit_init_data:
6164 /*
6165 * sessiond_cleanup() is called when no other thread is running, except
6166 * the ht_cleanup thread, which is needed to destroy the hash tables.
6167 */
6168 rcu_thread_online();
6169 sessiond_cleanup();
6170 rcu_thread_offline();
6171 rcu_unregister_thread();
6172
6173 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6174 if (ret < 0) {
6175 ERR("write error on ht_cleanup quit pipe");
6176 retval = -1;
6177 }
6178
6179 ret = pthread_join(ht_cleanup_thread, &status);
6180 if (ret) {
6181 errno = ret;
6182 PERROR("pthread_join ht cleanup thread");
6183 retval = -1;
6184 }
6185 exit_ht_cleanup:
6186 exit_set_max_size:
6187
6188 utils_close_pipe(ht_cleanup_pipe);
6189 exit_ht_cleanup_pipe:
6190
6191 /*
6192 * Close the ht_cleanup quit pipe.
6193 */
6194 utils_close_pipe(ht_cleanup_quit_pipe);
6195 exit_ht_cleanup_quit_pipe:
6196
6197 health_app_destroy(health_sessiond);
6198 exit_health_sessiond_cleanup:
6199 exit_create_run_as_worker_cleanup:
6200
6201 exit_options:
6202 /* Ensure all prior call_rcu are done. */
6203 rcu_barrier();
6204
6205 sessiond_cleanup_options();
6206
6207 exit_set_signal_handler:
6208
6209 if (!retval) {
6210 exit(EXIT_SUCCESS);
6211 } else {
6212 exit(EXIT_FAILURE);
6213 }
6214 }
This page took 0.237559 seconds and 3 git commands to generate.