Backport: Use EXTRA_VERSION_NAME and EXTRA_VERSION_DESCRIPTION
[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 #include <ctype.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
52
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
55 #include "channel.h"
56 #include "cmd.h"
57 #include "consumer.h"
58 #include "context.h"
59 #include "event.h"
60 #include "kernel.h"
61 #include "kernel-consumer.h"
62 #include "modprobe.h"
63 #include "shm.h"
64 #include "ust-ctl.h"
65 #include "ust-consumer.h"
66 #include "utils.h"
67 #include "fd-limit.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
72 #include "save.h"
73 #include "load-session-thread.h"
74 #include "syscall.h"
75 #include "agent.h"
76 #include "ht-cleanup.h"
77
78 #define CONSUMERD_FILE "lttng-consumerd"
79
80 const char *progname;
81 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
82 static int tracing_group_name_override;
83 static char *opt_pidfile;
84 static int opt_sig_parent;
85 static int opt_verbose_consumer;
86 static int opt_daemon, opt_background;
87 static int opt_no_kernel;
88 static char *opt_load_session_path;
89 static pid_t ppid; /* Parent PID for --sig-parent option */
90 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
91 static char *rundir;
92 static int lockfile_fd = -1;
93 static int opt_print_version;
94
95 /* Set to 1 when a SIGUSR1 signal is received. */
96 static int recv_child_signal;
97
98 /*
99 * Consumer daemon specific control data. Every value not initialized here is
100 * set to 0 by the static definition.
101 */
102 static struct consumer_data kconsumer_data = {
103 .type = LTTNG_CONSUMER_KERNEL,
104 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
105 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
106 .err_sock = -1,
107 .cmd_sock = -1,
108 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
109 .lock = PTHREAD_MUTEX_INITIALIZER,
110 .cond = PTHREAD_COND_INITIALIZER,
111 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
112 };
113 static struct consumer_data ustconsumer64_data = {
114 .type = LTTNG_CONSUMER64_UST,
115 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
116 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
117 .err_sock = -1,
118 .cmd_sock = -1,
119 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
120 .lock = PTHREAD_MUTEX_INITIALIZER,
121 .cond = PTHREAD_COND_INITIALIZER,
122 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
123 };
124 static struct consumer_data ustconsumer32_data = {
125 .type = LTTNG_CONSUMER32_UST,
126 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
127 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
128 .err_sock = -1,
129 .cmd_sock = -1,
130 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
131 .lock = PTHREAD_MUTEX_INITIALIZER,
132 .cond = PTHREAD_COND_INITIALIZER,
133 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
134 };
135
136 /* Command line options */
137 static const struct option long_options[] = {
138 { "client-sock", required_argument, 0, 'c' },
139 { "apps-sock", required_argument, 0, 'a' },
140 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
141 { "kconsumerd-err-sock", required_argument, 0, '\0' },
142 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
143 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
144 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
145 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
146 { "consumerd32-path", required_argument, 0, '\0' },
147 { "consumerd32-libdir", required_argument, 0, '\0' },
148 { "consumerd64-path", required_argument, 0, '\0' },
149 { "consumerd64-libdir", required_argument, 0, '\0' },
150 { "daemonize", no_argument, 0, 'd' },
151 { "background", no_argument, 0, 'b' },
152 { "sig-parent", no_argument, 0, 'S' },
153 { "help", no_argument, 0, 'h' },
154 { "group", required_argument, 0, 'g' },
155 { "version", no_argument, 0, 'V' },
156 { "quiet", no_argument, 0, 'q' },
157 { "verbose", no_argument, 0, 'v' },
158 { "verbose-consumer", no_argument, 0, '\0' },
159 { "no-kernel", no_argument, 0, '\0' },
160 { "pidfile", required_argument, 0, 'p' },
161 { "agent-tcp-port", required_argument, 0, '\0' },
162 { "config", required_argument, 0, 'f' },
163 { "load", required_argument, 0, 'l' },
164 { "kmod-probes", required_argument, 0, '\0' },
165 { "extra-kmod-probes", required_argument, 0, '\0' },
166 { NULL, 0, 0, 0 }
167 };
168
169 /* Command line options to ignore from configuration file */
170 static const char *config_ignore_options[] = { "help", "version", "config" };
171
172 /* Shared between threads */
173 static int dispatch_thread_exit;
174
175 /* Global application Unix socket path */
176 static char apps_unix_sock_path[PATH_MAX];
177 /* Global client Unix socket path */
178 static char client_unix_sock_path[PATH_MAX];
179 /* global wait shm path for UST */
180 static char wait_shm_path[PATH_MAX];
181 /* Global health check unix path */
182 static char health_unix_sock_path[PATH_MAX];
183
184 /* Sockets and FDs */
185 static int client_sock = -1;
186 static int apps_sock = -1;
187 int kernel_tracer_fd = -1;
188 static int kernel_poll_pipe[2] = { -1, -1 };
189
190 /*
191 * Quit pipe for all threads. This permits a single cancellation point
192 * for all threads when receiving an event on the pipe.
193 */
194 static int thread_quit_pipe[2] = { -1, -1 };
195
196 /*
197 * This pipe is used to inform the thread managing application communication
198 * that a command is queued and ready to be processed.
199 */
200 static int apps_cmd_pipe[2] = { -1, -1 };
201
202 int apps_cmd_notify_pipe[2] = { -1, -1 };
203
204 /* Pthread, Mutexes and Semaphores */
205 static pthread_t apps_thread;
206 static pthread_t apps_notify_thread;
207 static pthread_t reg_apps_thread;
208 static pthread_t client_thread;
209 static pthread_t kernel_thread;
210 static pthread_t dispatch_thread;
211 static pthread_t health_thread;
212 static pthread_t ht_cleanup_thread;
213 static pthread_t agent_reg_thread;
214 static pthread_t load_session_thread;
215
216 /*
217 * UST registration command queue. This queue is tied with a futex and uses a N
218 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 *
220 * The thread_registration_apps and thread_dispatch_ust_registration uses this
221 * queue along with the wait/wake scheme. The thread_manage_apps receives down
222 * the line new application socket and monitors it for any I/O error or clean
223 * close that triggers an unregistration of the application.
224 */
225 static struct ust_cmd_queue ust_cmd_queue;
226
227 /*
228 * Pointer initialized before thread creation.
229 *
230 * This points to the tracing session list containing the session count and a
231 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
232 * MUST NOT be taken if you call a public function in session.c.
233 *
234 * The lock is nested inside the structure: session_list_ptr->lock. Please use
235 * session_lock_list and session_unlock_list for lock acquisition.
236 */
237 static struct ltt_session_list *session_list_ptr;
238
239 int ust_consumerd64_fd = -1;
240 int ust_consumerd32_fd = -1;
241
242 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
243 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
244 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
245 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
246 static int consumerd32_bin_override;
247 static int consumerd64_bin_override;
248 static int consumerd32_libdir_override;
249 static int consumerd64_libdir_override;
250
251 static const char *module_proc_lttng = "/proc/lttng";
252
253 /*
254 * Consumer daemon state which is changed when spawning it, killing it or in
255 * case of a fatal error.
256 */
257 enum consumerd_state {
258 CONSUMER_STARTED = 1,
259 CONSUMER_STOPPED = 2,
260 CONSUMER_ERROR = 3,
261 };
262
263 /*
264 * This consumer daemon state is used to validate if a client command will be
265 * able to reach the consumer. If not, the client is informed. For instance,
266 * doing a "lttng start" when the consumer state is set to ERROR will return an
267 * error to the client.
268 *
269 * The following example shows a possible race condition of this scheme:
270 *
271 * consumer thread error happens
272 * client cmd arrives
273 * client cmd checks state -> still OK
274 * consumer thread exit, sets error
275 * client cmd try to talk to consumer
276 * ...
277 *
278 * However, since the consumer is a different daemon, we have no way of making
279 * sure the command will reach it safely even with this state flag. This is why
280 * we consider that up to the state validation during command processing, the
281 * command is safe. After that, we can not guarantee the correctness of the
282 * client request vis-a-vis the consumer.
283 */
284 static enum consumerd_state ust_consumerd_state;
285 static enum consumerd_state kernel_consumerd_state;
286
287 /*
288 * Socket timeout for receiving and sending in seconds.
289 */
290 static int app_socket_timeout;
291
292 /* Set in main() with the current page size. */
293 long page_size;
294
295 /* Application health monitoring */
296 struct health_app *health_sessiond;
297
298 /* Agent TCP port for registration. Used by the agent thread. */
299 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
300
301 /* Am I root or not. */
302 int is_root; /* Set to 1 if the daemon is running as root */
303
304 const char * const config_section_name = "sessiond";
305
306 /* Load session thread information to operate. */
307 struct load_session_thread_data *load_info;
308
309 /* Global hash tables */
310 struct lttng_ht *agent_apps_ht_by_sock = NULL;
311
312 /*
313 * Whether sessiond is ready for commands/health check requests.
314 * NR_LTTNG_SESSIOND_READY must match the number of calls to
315 * sessiond_notify_ready().
316 */
317 #define NR_LTTNG_SESSIOND_READY 4
318 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
319
320 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
321 {
322 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
323 }
324
325 /* Notify parents that we are ready for cmd and health check */
326 LTTNG_HIDDEN
327 void sessiond_notify_ready(void)
328 {
329 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
330 /*
331 * Notify parent pid that we are ready to accept command
332 * for client side. This ppid is the one from the
333 * external process that spawned us.
334 */
335 if (opt_sig_parent) {
336 kill(ppid, SIGUSR1);
337 }
338
339 /*
340 * Notify the parent of the fork() process that we are
341 * ready.
342 */
343 if (opt_daemon || opt_background) {
344 kill(child_ppid, SIGUSR1);
345 }
346 }
347 }
348
349 static
350 void setup_consumerd_path(void)
351 {
352 const char *bin, *libdir;
353
354 /*
355 * Allow INSTALL_BIN_PATH to be used as a target path for the
356 * native architecture size consumer if CONFIG_CONSUMER*_PATH
357 * has not been defined.
358 */
359 #if (CAA_BITS_PER_LONG == 32)
360 if (!consumerd32_bin[0]) {
361 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
362 }
363 if (!consumerd32_libdir[0]) {
364 consumerd32_libdir = INSTALL_LIB_PATH;
365 }
366 #elif (CAA_BITS_PER_LONG == 64)
367 if (!consumerd64_bin[0]) {
368 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
369 }
370 if (!consumerd64_libdir[0]) {
371 consumerd64_libdir = INSTALL_LIB_PATH;
372 }
373 #else
374 #error "Unknown bitness"
375 #endif
376
377 /*
378 * runtime env. var. overrides the build default.
379 */
380 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
381 if (bin) {
382 consumerd32_bin = bin;
383 }
384 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
385 if (bin) {
386 consumerd64_bin = bin;
387 }
388 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
389 if (libdir) {
390 consumerd32_libdir = libdir;
391 }
392 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
393 if (libdir) {
394 consumerd64_libdir = libdir;
395 }
396 }
397
398 static
399 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
400 int *a_pipe)
401 {
402 int ret;
403
404 assert(events);
405
406 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
407 if (ret < 0) {
408 goto error;
409 }
410
411 /* Add quit pipe */
412 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
413 if (ret < 0) {
414 goto error;
415 }
416
417 return 0;
418
419 error:
420 return ret;
421 }
422
423 /*
424 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
425 */
426 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
427 {
428 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
429 }
430
431 /*
432 * Init thread quit pipe.
433 *
434 * Return -1 on error or 0 if all pipes are created.
435 */
436 static int __init_thread_quit_pipe(int *a_pipe)
437 {
438 int ret, i;
439
440 ret = pipe(a_pipe);
441 if (ret < 0) {
442 PERROR("thread quit pipe");
443 goto error;
444 }
445
446 for (i = 0; i < 2; i++) {
447 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
448 if (ret < 0) {
449 PERROR("fcntl");
450 goto error;
451 }
452 }
453
454 error:
455 return ret;
456 }
457
458 static int init_thread_quit_pipe(void)
459 {
460 return __init_thread_quit_pipe(thread_quit_pipe);
461 }
462
463 /*
464 * Stop all threads by closing the thread quit pipe.
465 */
466 static void stop_threads(void)
467 {
468 int ret;
469
470 /* Stopping all threads */
471 DBG("Terminating all threads");
472 ret = notify_thread_pipe(thread_quit_pipe[1]);
473 if (ret < 0) {
474 ERR("write error on thread quit pipe");
475 }
476
477 /* Dispatch thread */
478 CMM_STORE_SHARED(dispatch_thread_exit, 1);
479 futex_nto1_wake(&ust_cmd_queue.futex);
480 }
481
482 /*
483 * Close every consumer sockets.
484 */
485 static void close_consumer_sockets(void)
486 {
487 int ret;
488
489 if (kconsumer_data.err_sock >= 0) {
490 ret = close(kconsumer_data.err_sock);
491 if (ret < 0) {
492 PERROR("kernel consumer err_sock close");
493 }
494 }
495 if (ustconsumer32_data.err_sock >= 0) {
496 ret = close(ustconsumer32_data.err_sock);
497 if (ret < 0) {
498 PERROR("UST consumerd32 err_sock close");
499 }
500 }
501 if (ustconsumer64_data.err_sock >= 0) {
502 ret = close(ustconsumer64_data.err_sock);
503 if (ret < 0) {
504 PERROR("UST consumerd64 err_sock close");
505 }
506 }
507 if (kconsumer_data.cmd_sock >= 0) {
508 ret = close(kconsumer_data.cmd_sock);
509 if (ret < 0) {
510 PERROR("kernel consumer cmd_sock close");
511 }
512 }
513 if (ustconsumer32_data.cmd_sock >= 0) {
514 ret = close(ustconsumer32_data.cmd_sock);
515 if (ret < 0) {
516 PERROR("UST consumerd32 cmd_sock close");
517 }
518 }
519 if (ustconsumer64_data.cmd_sock >= 0) {
520 ret = close(ustconsumer64_data.cmd_sock);
521 if (ret < 0) {
522 PERROR("UST consumerd64 cmd_sock close");
523 }
524 }
525 }
526
527 /*
528 * Generate the full lock file path using the rundir.
529 *
530 * Return the snprintf() return value thus a negative value is an error.
531 */
532 static int generate_lock_file_path(char *path, size_t len)
533 {
534 int ret;
535
536 assert(path);
537 assert(rundir);
538
539 /* Build lockfile path from rundir. */
540 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
541 if (ret < 0) {
542 PERROR("snprintf lockfile path");
543 }
544
545 return ret;
546 }
547
548 /*
549 * Wait on consumer process termination.
550 *
551 * Need to be called with the consumer data lock held or from a context
552 * ensuring no concurrent access to data (e.g: cleanup).
553 */
554 static void wait_consumer(struct consumer_data *consumer_data)
555 {
556 pid_t ret;
557 int status;
558
559 if (consumer_data->pid <= 0) {
560 return;
561 }
562
563 DBG("Waiting for complete teardown of consumerd (PID: %d)",
564 consumer_data->pid);
565 ret = waitpid(consumer_data->pid, &status, 0);
566 if (ret == -1) {
567 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
568 } else if (!WIFEXITED(status)) {
569 ERR("consumerd termination with error: %d",
570 WEXITSTATUS(ret));
571 }
572 consumer_data->pid = 0;
573 }
574
575 /*
576 * Cleanup the session daemon's data structures.
577 */
578 static void sessiond_cleanup(void)
579 {
580 int ret;
581 struct ltt_session *sess, *stmp;
582 char path[PATH_MAX];
583
584 DBG("Cleanup sessiond");
585
586 /*
587 * Close the thread quit pipe. It has already done its job,
588 * since we are now called.
589 */
590 utils_close_pipe(thread_quit_pipe);
591
592 /*
593 * If opt_pidfile is undefined, the default file will be wiped when
594 * removing the rundir.
595 */
596 if (opt_pidfile) {
597 ret = remove(opt_pidfile);
598 if (ret < 0) {
599 PERROR("remove pidfile %s", opt_pidfile);
600 }
601 }
602
603 DBG("Removing sessiond and consumerd content of directory %s", rundir);
604
605 /* sessiond */
606 snprintf(path, PATH_MAX,
607 "%s/%s",
608 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
609 DBG("Removing %s", path);
610 (void) unlink(path);
611
612 snprintf(path, PATH_MAX, "%s/%s", rundir,
613 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
614 DBG("Removing %s", path);
615 (void) unlink(path);
616
617 /* kconsumerd */
618 snprintf(path, PATH_MAX,
619 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
620 rundir);
621 DBG("Removing %s", path);
622 (void) unlink(path);
623
624 snprintf(path, PATH_MAX,
625 DEFAULT_KCONSUMERD_PATH,
626 rundir);
627 DBG("Removing directory %s", path);
628 (void) rmdir(path);
629
630 /* ust consumerd 32 */
631 snprintf(path, PATH_MAX,
632 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
633 rundir);
634 DBG("Removing %s", path);
635 (void) unlink(path);
636
637 snprintf(path, PATH_MAX,
638 DEFAULT_USTCONSUMERD32_PATH,
639 rundir);
640 DBG("Removing directory %s", path);
641 (void) rmdir(path);
642
643 /* ust consumerd 64 */
644 snprintf(path, PATH_MAX,
645 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
646 rundir);
647 DBG("Removing %s", path);
648 (void) unlink(path);
649
650 snprintf(path, PATH_MAX,
651 DEFAULT_USTCONSUMERD64_PATH,
652 rundir);
653 DBG("Removing directory %s", path);
654 (void) rmdir(path);
655
656 DBG("Cleaning up all sessions");
657
658 /* Destroy session list mutex */
659 if (session_list_ptr != NULL) {
660 pthread_mutex_destroy(&session_list_ptr->lock);
661
662 /* Cleanup ALL session */
663 cds_list_for_each_entry_safe(sess, stmp,
664 &session_list_ptr->head, list) {
665 cmd_destroy_session(sess, kernel_poll_pipe[1]);
666 }
667 }
668
669 wait_consumer(&kconsumer_data);
670 wait_consumer(&ustconsumer64_data);
671 wait_consumer(&ustconsumer32_data);
672
673 DBG("Cleaning up all agent apps");
674 agent_app_ht_clean();
675
676 DBG("Closing all UST sockets");
677 ust_app_clean_list();
678 buffer_reg_destroy_registries();
679
680 if (is_root && !opt_no_kernel) {
681 DBG2("Closing kernel fd");
682 if (kernel_tracer_fd >= 0) {
683 ret = close(kernel_tracer_fd);
684 if (ret) {
685 PERROR("close");
686 }
687 }
688 DBG("Unloading kernel modules");
689 modprobe_remove_lttng_all();
690 free(syscall_table);
691 }
692
693 close_consumer_sockets();
694
695 if (load_info) {
696 load_session_destroy_data(load_info);
697 free(load_info);
698 }
699
700 /*
701 * Cleanup lock file by deleting it and finaly closing it which will
702 * release the file system lock.
703 */
704 if (lockfile_fd >= 0) {
705 char lockfile_path[PATH_MAX];
706
707 ret = generate_lock_file_path(lockfile_path,
708 sizeof(lockfile_path));
709 if (ret > 0) {
710 ret = remove(lockfile_path);
711 if (ret < 0) {
712 PERROR("remove lock file");
713 }
714 ret = close(lockfile_fd);
715 if (ret < 0) {
716 PERROR("close lock file");
717 }
718 }
719 }
720
721 /*
722 * We do NOT rmdir rundir because there are other processes
723 * using it, for instance lttng-relayd, which can start in
724 * parallel with this teardown.
725 */
726
727 free(rundir);
728 }
729
730 /*
731 * Cleanup the daemon's option data structures.
732 */
733 static void sessiond_cleanup_options(void)
734 {
735 DBG("Cleaning up options");
736
737 /*
738 * If the override option is set, the pointer points to a *non* const
739 * thus freeing it even though the variable type is set to const.
740 */
741 if (tracing_group_name_override) {
742 free((void *) tracing_group_name);
743 }
744 if (consumerd32_bin_override) {
745 free((void *) consumerd32_bin);
746 }
747 if (consumerd64_bin_override) {
748 free((void *) consumerd64_bin);
749 }
750 if (consumerd32_libdir_override) {
751 free((void *) consumerd32_libdir);
752 }
753 if (consumerd64_libdir_override) {
754 free((void *) consumerd64_libdir);
755 }
756
757 free(opt_pidfile);
758 free(opt_load_session_path);
759 free(kmod_probes_list);
760 free(kmod_extra_probes_list);
761
762 run_as_destroy_worker();
763 }
764
765 /*
766 * Send data on a unix socket using the liblttsessiondcomm API.
767 *
768 * Return lttcomm error code.
769 */
770 static int send_unix_sock(int sock, void *buf, size_t len)
771 {
772 /* Check valid length */
773 if (len == 0) {
774 return -1;
775 }
776
777 return lttcomm_send_unix_sock(sock, buf, len);
778 }
779
780 /*
781 * Free memory of a command context structure.
782 */
783 static void clean_command_ctx(struct command_ctx **cmd_ctx)
784 {
785 DBG("Clean command context structure");
786 if (*cmd_ctx) {
787 if ((*cmd_ctx)->llm) {
788 free((*cmd_ctx)->llm);
789 }
790 if ((*cmd_ctx)->lsm) {
791 free((*cmd_ctx)->lsm);
792 }
793 free(*cmd_ctx);
794 *cmd_ctx = NULL;
795 }
796 }
797
798 /*
799 * Notify UST applications using the shm mmap futex.
800 */
801 static int notify_ust_apps(int active)
802 {
803 char *wait_shm_mmap;
804
805 DBG("Notifying applications of session daemon state: %d", active);
806
807 /* See shm.c for this call implying mmap, shm and futex calls */
808 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
809 if (wait_shm_mmap == NULL) {
810 goto error;
811 }
812
813 /* Wake waiting process */
814 futex_wait_update((int32_t *) wait_shm_mmap, active);
815
816 /* Apps notified successfully */
817 return 0;
818
819 error:
820 return -1;
821 }
822
823 /*
824 * Setup the outgoing data buffer for the response (llm) by allocating the
825 * right amount of memory and copying the original information from the lsm
826 * structure.
827 *
828 * Return 0 on success, negative value on error.
829 */
830 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
831 const void *payload_buf, size_t payload_len,
832 const void *cmd_header_buf, size_t cmd_header_len)
833 {
834 int ret = 0;
835 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
836 const size_t cmd_header_offset = header_len;
837 const size_t payload_offset = cmd_header_offset + cmd_header_len;
838 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
839
840 cmd_ctx->llm = zmalloc(total_msg_size);
841
842 if (cmd_ctx->llm == NULL) {
843 PERROR("zmalloc");
844 ret = -ENOMEM;
845 goto end;
846 }
847
848 /* Copy common data */
849 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
850 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
851 cmd_ctx->llm->cmd_header_size = cmd_header_len;
852 cmd_ctx->llm->data_size = payload_len;
853 cmd_ctx->lttng_msg_size = total_msg_size;
854
855 /* Copy command header */
856 if (cmd_header_len) {
857 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
858 cmd_header_len);
859 }
860
861 /* Copy payload */
862 if (payload_len) {
863 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
864 payload_len);
865 }
866
867 end:
868 return ret;
869 }
870
871 /*
872 * Version of setup_lttng_msg() without command header.
873 */
874 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
875 void *payload_buf, size_t payload_len)
876 {
877 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
878 }
879 /*
880 * Update the kernel poll set of all channel fd available over all tracing
881 * session. Add the wakeup pipe at the end of the set.
882 */
883 static int update_kernel_poll(struct lttng_poll_event *events)
884 {
885 int ret;
886 struct ltt_session *session;
887 struct ltt_kernel_channel *channel;
888
889 DBG("Updating kernel poll set");
890
891 session_lock_list();
892 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
893 session_lock(session);
894 if (session->kernel_session == NULL) {
895 session_unlock(session);
896 continue;
897 }
898
899 cds_list_for_each_entry(channel,
900 &session->kernel_session->channel_list.head, list) {
901 /* Add channel fd to the kernel poll set */
902 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
903 if (ret < 0) {
904 session_unlock(session);
905 goto error;
906 }
907 DBG("Channel fd %d added to kernel set", channel->fd);
908 }
909 session_unlock(session);
910 }
911 session_unlock_list();
912
913 return 0;
914
915 error:
916 session_unlock_list();
917 return -1;
918 }
919
920 /*
921 * Find the channel fd from 'fd' over all tracing session. When found, check
922 * for new channel stream and send those stream fds to the kernel consumer.
923 *
924 * Useful for CPU hotplug feature.
925 */
926 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
927 {
928 int ret = 0;
929 struct ltt_session *session;
930 struct ltt_kernel_session *ksess;
931 struct ltt_kernel_channel *channel;
932
933 DBG("Updating kernel streams for channel fd %d", fd);
934
935 session_lock_list();
936 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
937 session_lock(session);
938 if (session->kernel_session == NULL) {
939 session_unlock(session);
940 continue;
941 }
942 ksess = session->kernel_session;
943
944 cds_list_for_each_entry(channel,
945 &ksess->channel_list.head, list) {
946 struct lttng_ht_iter iter;
947 struct consumer_socket *socket;
948
949 if (channel->fd != fd) {
950 continue;
951 }
952 DBG("Channel found, updating kernel streams");
953 ret = kernel_open_channel_stream(channel);
954 if (ret < 0) {
955 goto error;
956 }
957 /* Update the stream global counter */
958 ksess->stream_count_global += ret;
959
960 /*
961 * Have we already sent fds to the consumer? If yes, it
962 * means that tracing is started so it is safe to send
963 * our updated stream fds.
964 */
965 if (ksess->consumer_fds_sent != 1
966 || ksess->consumer == NULL) {
967 ret = -1;
968 goto error;
969 }
970
971 rcu_read_lock();
972 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
973 &iter.iter, socket, node.node) {
974 pthread_mutex_lock(socket->lock);
975 ret = kernel_consumer_send_channel_stream(socket,
976 channel, ksess,
977 session->output_traces ? 1 : 0);
978 pthread_mutex_unlock(socket->lock);
979 if (ret < 0) {
980 rcu_read_unlock();
981 goto error;
982 }
983 }
984 rcu_read_unlock();
985 }
986 session_unlock(session);
987 }
988 session_unlock_list();
989 return ret;
990
991 error:
992 session_unlock(session);
993 session_unlock_list();
994 return ret;
995 }
996
997 /*
998 * For each tracing session, update newly registered apps. The session list
999 * lock MUST be acquired before calling this.
1000 */
1001 static void update_ust_app(int app_sock)
1002 {
1003 struct ltt_session *sess, *stmp;
1004
1005 /* Consumer is in an ERROR state. Stop any application update. */
1006 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1007 /* Stop the update process since the consumer is dead. */
1008 return;
1009 }
1010
1011 /* For all tracing session(s) */
1012 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1013 struct ust_app *app;
1014
1015 session_lock(sess);
1016 if (!sess->ust_session) {
1017 goto unlock_session;
1018 }
1019
1020 rcu_read_lock();
1021 assert(app_sock >= 0);
1022 app = ust_app_find_by_sock(app_sock);
1023 if (app == NULL) {
1024 /*
1025 * Application can be unregistered before so
1026 * this is possible hence simply stopping the
1027 * update.
1028 */
1029 DBG3("UST app update failed to find app sock %d",
1030 app_sock);
1031 goto unlock_rcu;
1032 }
1033 ust_app_global_update(sess->ust_session, app);
1034 unlock_rcu:
1035 rcu_read_unlock();
1036 unlock_session:
1037 session_unlock(sess);
1038 }
1039 }
1040
1041 /*
1042 * This thread manage event coming from the kernel.
1043 *
1044 * Features supported in this thread:
1045 * -) CPU Hotplug
1046 */
1047 static void *thread_manage_kernel(void *data)
1048 {
1049 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1050 uint32_t revents, nb_fd;
1051 char tmp;
1052 struct lttng_poll_event events;
1053
1054 DBG("[thread] Thread manage kernel started");
1055
1056 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1057
1058 /*
1059 * This first step of the while is to clean this structure which could free
1060 * non NULL pointers so initialize it before the loop.
1061 */
1062 lttng_poll_init(&events);
1063
1064 if (testpoint(sessiond_thread_manage_kernel)) {
1065 goto error_testpoint;
1066 }
1067
1068 health_code_update();
1069
1070 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1071 goto error_testpoint;
1072 }
1073
1074 while (1) {
1075 health_code_update();
1076
1077 if (update_poll_flag == 1) {
1078 /* Clean events object. We are about to populate it again. */
1079 lttng_poll_clean(&events);
1080
1081 ret = sessiond_set_thread_pollset(&events, 2);
1082 if (ret < 0) {
1083 goto error_poll_create;
1084 }
1085
1086 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1087 if (ret < 0) {
1088 goto error;
1089 }
1090
1091 /* This will add the available kernel channel if any. */
1092 ret = update_kernel_poll(&events);
1093 if (ret < 0) {
1094 goto error;
1095 }
1096 update_poll_flag = 0;
1097 }
1098
1099 DBG("Thread kernel polling");
1100
1101 /* Poll infinite value of time */
1102 restart:
1103 health_poll_entry();
1104 ret = lttng_poll_wait(&events, -1);
1105 DBG("Thread kernel return from poll on %d fds",
1106 LTTNG_POLL_GETNB(&events));
1107 health_poll_exit();
1108 if (ret < 0) {
1109 /*
1110 * Restart interrupted system call.
1111 */
1112 if (errno == EINTR) {
1113 goto restart;
1114 }
1115 goto error;
1116 } else if (ret == 0) {
1117 /* Should not happen since timeout is infinite */
1118 ERR("Return value of poll is 0 with an infinite timeout.\n"
1119 "This should not have happened! Continuing...");
1120 continue;
1121 }
1122
1123 nb_fd = ret;
1124
1125 for (i = 0; i < nb_fd; i++) {
1126 /* Fetch once the poll data */
1127 revents = LTTNG_POLL_GETEV(&events, i);
1128 pollfd = LTTNG_POLL_GETFD(&events, i);
1129
1130 health_code_update();
1131
1132 if (!revents) {
1133 /* No activity for this FD (poll implementation). */
1134 continue;
1135 }
1136
1137 /* Thread quit pipe has been closed. Killing thread. */
1138 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1139 if (ret) {
1140 err = 0;
1141 goto exit;
1142 }
1143
1144 /* Check for data on kernel pipe */
1145 if (revents & LPOLLIN) {
1146 if (pollfd == kernel_poll_pipe[0]) {
1147 (void) lttng_read(kernel_poll_pipe[0],
1148 &tmp, 1);
1149 /*
1150 * Ret value is useless here, if this pipe gets any actions an
1151 * update is required anyway.
1152 */
1153 update_poll_flag = 1;
1154 continue;
1155 } else {
1156 /*
1157 * New CPU detected by the kernel. Adding kernel stream to
1158 * kernel session and updating the kernel consumer
1159 */
1160 ret = update_kernel_stream(&kconsumer_data, pollfd);
1161 if (ret < 0) {
1162 continue;
1163 }
1164 break;
1165 }
1166 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1167 update_poll_flag = 1;
1168 continue;
1169 } else {
1170 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1171 goto error;
1172 }
1173 }
1174 }
1175
1176 exit:
1177 error:
1178 lttng_poll_clean(&events);
1179 error_poll_create:
1180 error_testpoint:
1181 utils_close_pipe(kernel_poll_pipe);
1182 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1183 if (err) {
1184 health_error();
1185 ERR("Health error occurred in %s", __func__);
1186 WARN("Kernel thread died unexpectedly. "
1187 "Kernel tracing can continue but CPU hotplug is disabled.");
1188 }
1189 health_unregister(health_sessiond);
1190 DBG("Kernel thread dying");
1191 return NULL;
1192 }
1193
1194 /*
1195 * Signal pthread condition of the consumer data that the thread.
1196 */
1197 static void signal_consumer_condition(struct consumer_data *data, int state)
1198 {
1199 pthread_mutex_lock(&data->cond_mutex);
1200
1201 /*
1202 * The state is set before signaling. It can be any value, it's the waiter
1203 * job to correctly interpret this condition variable associated to the
1204 * consumer pthread_cond.
1205 *
1206 * A value of 0 means that the corresponding thread of the consumer data
1207 * was not started. 1 indicates that the thread has started and is ready
1208 * for action. A negative value means that there was an error during the
1209 * thread bootstrap.
1210 */
1211 data->consumer_thread_is_ready = state;
1212 (void) pthread_cond_signal(&data->cond);
1213
1214 pthread_mutex_unlock(&data->cond_mutex);
1215 }
1216
1217 /*
1218 * This thread manage the consumer error sent back to the session daemon.
1219 */
1220 static void *thread_manage_consumer(void *data)
1221 {
1222 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1223 uint32_t revents, nb_fd;
1224 enum lttcomm_return_code code;
1225 struct lttng_poll_event events;
1226 struct consumer_data *consumer_data = data;
1227
1228 DBG("[thread] Manage consumer started");
1229
1230 rcu_register_thread();
1231 rcu_thread_online();
1232
1233 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1234
1235 health_code_update();
1236
1237 /*
1238 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1239 * metadata_sock. Nothing more will be added to this poll set.
1240 */
1241 ret = sessiond_set_thread_pollset(&events, 3);
1242 if (ret < 0) {
1243 goto error_poll;
1244 }
1245
1246 /*
1247 * The error socket here is already in a listening state which was done
1248 * just before spawning this thread to avoid a race between the consumer
1249 * daemon exec trying to connect and the listen() call.
1250 */
1251 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1252 if (ret < 0) {
1253 goto error;
1254 }
1255
1256 health_code_update();
1257
1258 /* Infinite blocking call, waiting for transmission */
1259 restart:
1260 health_poll_entry();
1261
1262 if (testpoint(sessiond_thread_manage_consumer)) {
1263 goto error;
1264 }
1265
1266 ret = lttng_poll_wait(&events, -1);
1267 health_poll_exit();
1268 if (ret < 0) {
1269 /*
1270 * Restart interrupted system call.
1271 */
1272 if (errno == EINTR) {
1273 goto restart;
1274 }
1275 goto error;
1276 }
1277
1278 nb_fd = ret;
1279
1280 for (i = 0; i < nb_fd; i++) {
1281 /* Fetch once the poll data */
1282 revents = LTTNG_POLL_GETEV(&events, i);
1283 pollfd = LTTNG_POLL_GETFD(&events, i);
1284
1285 health_code_update();
1286
1287 if (!revents) {
1288 /* No activity for this FD (poll implementation). */
1289 continue;
1290 }
1291
1292 /* Thread quit pipe has been closed. Killing thread. */
1293 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1294 if (ret) {
1295 err = 0;
1296 goto exit;
1297 }
1298
1299 /* Event on the registration socket */
1300 if (pollfd == consumer_data->err_sock) {
1301 if (revents & LPOLLIN) {
1302 continue;
1303 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1304 ERR("consumer err socket poll error");
1305 goto error;
1306 } else {
1307 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1308 goto error;
1309 }
1310 }
1311 }
1312
1313 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1314 if (sock < 0) {
1315 goto error;
1316 }
1317
1318 /*
1319 * Set the CLOEXEC flag. Return code is useless because either way, the
1320 * show must go on.
1321 */
1322 (void) utils_set_fd_cloexec(sock);
1323
1324 health_code_update();
1325
1326 DBG2("Receiving code from consumer err_sock");
1327
1328 /* Getting status code from kconsumerd */
1329 ret = lttcomm_recv_unix_sock(sock, &code,
1330 sizeof(enum lttcomm_return_code));
1331 if (ret <= 0) {
1332 goto error;
1333 }
1334
1335 health_code_update();
1336 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1337 /* Connect both socket, command and metadata. */
1338 consumer_data->cmd_sock =
1339 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1340 consumer_data->metadata_fd =
1341 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1342 if (consumer_data->cmd_sock < 0
1343 || consumer_data->metadata_fd < 0) {
1344 PERROR("consumer connect cmd socket");
1345 /* On error, signal condition and quit. */
1346 signal_consumer_condition(consumer_data, -1);
1347 goto error;
1348 }
1349 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1350 /* Create metadata socket lock. */
1351 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1352 if (consumer_data->metadata_sock.lock == NULL) {
1353 PERROR("zmalloc pthread mutex");
1354 goto error;
1355 }
1356 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1357
1358 signal_consumer_condition(consumer_data, 1);
1359 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1360 DBG("Consumer metadata socket ready (fd: %d)",
1361 consumer_data->metadata_fd);
1362 } else {
1363 ERR("consumer error when waiting for SOCK_READY : %s",
1364 lttcomm_get_readable_code(-code));
1365 goto error;
1366 }
1367
1368 /* Remove the consumerd error sock since we've established a connexion */
1369 ret = lttng_poll_del(&events, consumer_data->err_sock);
1370 if (ret < 0) {
1371 goto error;
1372 }
1373
1374 /* Add new accepted error socket. */
1375 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1376 if (ret < 0) {
1377 goto error;
1378 }
1379
1380 /* Add metadata socket that is successfully connected. */
1381 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1382 LPOLLIN | LPOLLRDHUP);
1383 if (ret < 0) {
1384 goto error;
1385 }
1386
1387 health_code_update();
1388
1389 /* Infinite blocking call, waiting for transmission */
1390 restart_poll:
1391 while (1) {
1392 health_code_update();
1393
1394 /* Exit the thread because the thread quit pipe has been triggered. */
1395 if (should_quit) {
1396 /* Not a health error. */
1397 err = 0;
1398 goto exit;
1399 }
1400
1401 health_poll_entry();
1402 ret = lttng_poll_wait(&events, -1);
1403 health_poll_exit();
1404 if (ret < 0) {
1405 /*
1406 * Restart interrupted system call.
1407 */
1408 if (errno == EINTR) {
1409 goto restart_poll;
1410 }
1411 goto error;
1412 }
1413
1414 nb_fd = ret;
1415
1416 for (i = 0; i < nb_fd; i++) {
1417 /* Fetch once the poll data */
1418 revents = LTTNG_POLL_GETEV(&events, i);
1419 pollfd = LTTNG_POLL_GETFD(&events, i);
1420
1421 health_code_update();
1422
1423 if (!revents) {
1424 /* No activity for this FD (poll implementation). */
1425 continue;
1426 }
1427
1428 /*
1429 * Thread quit pipe has been triggered, flag that we should stop
1430 * but continue the current loop to handle potential data from
1431 * consumer.
1432 */
1433 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1434
1435 if (pollfd == sock) {
1436 /* Event on the consumerd socket */
1437 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1438 && !(revents & LPOLLIN)) {
1439 ERR("consumer err socket second poll error");
1440 goto error;
1441 }
1442 health_code_update();
1443 /* Wait for any kconsumerd error */
1444 ret = lttcomm_recv_unix_sock(sock, &code,
1445 sizeof(enum lttcomm_return_code));
1446 if (ret <= 0) {
1447 ERR("consumer closed the command socket");
1448 goto error;
1449 }
1450
1451 ERR("consumer return code : %s",
1452 lttcomm_get_readable_code(-code));
1453
1454 goto exit;
1455 } else if (pollfd == consumer_data->metadata_fd) {
1456 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1457 && !(revents & LPOLLIN)) {
1458 ERR("consumer err metadata socket second poll error");
1459 goto error;
1460 }
1461 /* UST metadata requests */
1462 ret = ust_consumer_metadata_request(
1463 &consumer_data->metadata_sock);
1464 if (ret < 0) {
1465 ERR("Handling metadata request");
1466 goto error;
1467 }
1468 }
1469 /* No need for an else branch all FDs are tested prior. */
1470 }
1471 health_code_update();
1472 }
1473
1474 exit:
1475 error:
1476 /*
1477 * We lock here because we are about to close the sockets and some other
1478 * thread might be using them so get exclusive access which will abort all
1479 * other consumer command by other threads.
1480 */
1481 pthread_mutex_lock(&consumer_data->lock);
1482
1483 /* Immediately set the consumerd state to stopped */
1484 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1485 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1486 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1487 consumer_data->type == LTTNG_CONSUMER32_UST) {
1488 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1489 } else {
1490 /* Code flow error... */
1491 assert(0);
1492 }
1493
1494 if (consumer_data->err_sock >= 0) {
1495 ret = close(consumer_data->err_sock);
1496 if (ret) {
1497 PERROR("close");
1498 }
1499 consumer_data->err_sock = -1;
1500 }
1501 if (consumer_data->cmd_sock >= 0) {
1502 ret = close(consumer_data->cmd_sock);
1503 if (ret) {
1504 PERROR("close");
1505 }
1506 consumer_data->cmd_sock = -1;
1507 }
1508 if (consumer_data->metadata_sock.fd_ptr &&
1509 *consumer_data->metadata_sock.fd_ptr >= 0) {
1510 ret = close(*consumer_data->metadata_sock.fd_ptr);
1511 if (ret) {
1512 PERROR("close");
1513 }
1514 }
1515 if (sock >= 0) {
1516 ret = close(sock);
1517 if (ret) {
1518 PERROR("close");
1519 }
1520 }
1521
1522 unlink(consumer_data->err_unix_sock_path);
1523 unlink(consumer_data->cmd_unix_sock_path);
1524 pthread_mutex_unlock(&consumer_data->lock);
1525
1526 /* Cleanup metadata socket mutex. */
1527 if (consumer_data->metadata_sock.lock) {
1528 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1529 free(consumer_data->metadata_sock.lock);
1530 }
1531 lttng_poll_clean(&events);
1532 error_poll:
1533 if (err) {
1534 health_error();
1535 ERR("Health error occurred in %s", __func__);
1536 }
1537 health_unregister(health_sessiond);
1538 DBG("consumer thread cleanup completed");
1539
1540 rcu_thread_offline();
1541 rcu_unregister_thread();
1542
1543 return NULL;
1544 }
1545
1546 /*
1547 * This thread manage application communication.
1548 */
1549 static void *thread_manage_apps(void *data)
1550 {
1551 int i, ret, pollfd, err = -1;
1552 ssize_t size_ret;
1553 uint32_t revents, nb_fd;
1554 struct lttng_poll_event events;
1555
1556 DBG("[thread] Manage application started");
1557
1558 rcu_register_thread();
1559 rcu_thread_online();
1560
1561 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1562
1563 if (testpoint(sessiond_thread_manage_apps)) {
1564 goto error_testpoint;
1565 }
1566
1567 health_code_update();
1568
1569 ret = sessiond_set_thread_pollset(&events, 2);
1570 if (ret < 0) {
1571 goto error_poll_create;
1572 }
1573
1574 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1575 if (ret < 0) {
1576 goto error;
1577 }
1578
1579 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1580 goto error;
1581 }
1582
1583 health_code_update();
1584
1585 while (1) {
1586 DBG("Apps thread polling");
1587
1588 /* Inifinite blocking call, waiting for transmission */
1589 restart:
1590 health_poll_entry();
1591 ret = lttng_poll_wait(&events, -1);
1592 DBG("Apps thread return from poll on %d fds",
1593 LTTNG_POLL_GETNB(&events));
1594 health_poll_exit();
1595 if (ret < 0) {
1596 /*
1597 * Restart interrupted system call.
1598 */
1599 if (errno == EINTR) {
1600 goto restart;
1601 }
1602 goto error;
1603 }
1604
1605 nb_fd = ret;
1606
1607 for (i = 0; i < nb_fd; i++) {
1608 /* Fetch once the poll data */
1609 revents = LTTNG_POLL_GETEV(&events, i);
1610 pollfd = LTTNG_POLL_GETFD(&events, i);
1611
1612 health_code_update();
1613
1614 if (!revents) {
1615 /* No activity for this FD (poll implementation). */
1616 continue;
1617 }
1618
1619 /* Thread quit pipe has been closed. Killing thread. */
1620 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1621 if (ret) {
1622 err = 0;
1623 goto exit;
1624 }
1625
1626 /* Inspect the apps cmd pipe */
1627 if (pollfd == apps_cmd_pipe[0]) {
1628 if (revents & LPOLLIN) {
1629 int sock;
1630
1631 /* Empty pipe */
1632 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1633 if (size_ret < sizeof(sock)) {
1634 PERROR("read apps cmd pipe");
1635 goto error;
1636 }
1637
1638 health_code_update();
1639
1640 /*
1641 * Since this is a command socket (write then read),
1642 * we only monitor the error events of the socket.
1643 */
1644 ret = lttng_poll_add(&events, sock,
1645 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1646 if (ret < 0) {
1647 goto error;
1648 }
1649
1650 DBG("Apps with sock %d added to poll set", sock);
1651 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1652 ERR("Apps command pipe error");
1653 goto error;
1654 } else {
1655 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1656 goto error;
1657 }
1658 } else {
1659 /*
1660 * At this point, we know that a registered application made
1661 * the event at poll_wait.
1662 */
1663 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1664 /* Removing from the poll set */
1665 ret = lttng_poll_del(&events, pollfd);
1666 if (ret < 0) {
1667 goto error;
1668 }
1669
1670 /* Socket closed on remote end. */
1671 ust_app_unregister(pollfd);
1672 } else {
1673 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1674 goto error;
1675 }
1676 }
1677
1678 health_code_update();
1679 }
1680 }
1681
1682 exit:
1683 error:
1684 lttng_poll_clean(&events);
1685 error_poll_create:
1686 error_testpoint:
1687 utils_close_pipe(apps_cmd_pipe);
1688 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1689
1690 /*
1691 * We don't clean the UST app hash table here since already registered
1692 * applications can still be controlled so let them be until the session
1693 * daemon dies or the applications stop.
1694 */
1695
1696 if (err) {
1697 health_error();
1698 ERR("Health error occurred in %s", __func__);
1699 }
1700 health_unregister(health_sessiond);
1701 DBG("Application communication apps thread cleanup complete");
1702 rcu_thread_offline();
1703 rcu_unregister_thread();
1704 return NULL;
1705 }
1706
1707 /*
1708 * Send a socket to a thread This is called from the dispatch UST registration
1709 * thread once all sockets are set for the application.
1710 *
1711 * The sock value can be invalid, we don't really care, the thread will handle
1712 * it and make the necessary cleanup if so.
1713 *
1714 * On success, return 0 else a negative value being the errno message of the
1715 * write().
1716 */
1717 static int send_socket_to_thread(int fd, int sock)
1718 {
1719 ssize_t ret;
1720
1721 /*
1722 * It's possible that the FD is set as invalid with -1 concurrently just
1723 * before calling this function being a shutdown state of the thread.
1724 */
1725 if (fd < 0) {
1726 ret = -EBADF;
1727 goto error;
1728 }
1729
1730 ret = lttng_write(fd, &sock, sizeof(sock));
1731 if (ret < sizeof(sock)) {
1732 PERROR("write apps pipe %d", fd);
1733 if (ret < 0) {
1734 ret = -errno;
1735 }
1736 goto error;
1737 }
1738
1739 /* All good. Don't send back the write positive ret value. */
1740 ret = 0;
1741 error:
1742 return (int) ret;
1743 }
1744
1745 /*
1746 * Sanitize the wait queue of the dispatch registration thread meaning removing
1747 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1748 * notify socket is never received.
1749 */
1750 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1751 {
1752 int ret, nb_fd = 0, i;
1753 unsigned int fd_added = 0;
1754 struct lttng_poll_event events;
1755 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1756
1757 assert(wait_queue);
1758
1759 lttng_poll_init(&events);
1760
1761 /* Just skip everything for an empty queue. */
1762 if (!wait_queue->count) {
1763 goto end;
1764 }
1765
1766 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1767 if (ret < 0) {
1768 goto error_create;
1769 }
1770
1771 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1772 &wait_queue->head, head) {
1773 assert(wait_node->app);
1774 ret = lttng_poll_add(&events, wait_node->app->sock,
1775 LPOLLHUP | LPOLLERR);
1776 if (ret < 0) {
1777 goto error;
1778 }
1779
1780 fd_added = 1;
1781 }
1782
1783 if (!fd_added) {
1784 goto end;
1785 }
1786
1787 /*
1788 * Poll but don't block so we can quickly identify the faulty events and
1789 * clean them afterwards from the wait queue.
1790 */
1791 ret = lttng_poll_wait(&events, 0);
1792 if (ret < 0) {
1793 goto error;
1794 }
1795 nb_fd = ret;
1796
1797 for (i = 0; i < nb_fd; i++) {
1798 /* Get faulty FD. */
1799 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1800 int pollfd = LTTNG_POLL_GETFD(&events, i);
1801
1802 if (!revents) {
1803 /* No activity for this FD (poll implementation). */
1804 continue;
1805 }
1806
1807 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1808 &wait_queue->head, head) {
1809 if (pollfd == wait_node->app->sock &&
1810 (revents & (LPOLLHUP | LPOLLERR))) {
1811 cds_list_del(&wait_node->head);
1812 wait_queue->count--;
1813 ust_app_destroy(wait_node->app);
1814 free(wait_node);
1815 /*
1816 * Silence warning of use-after-free in
1817 * cds_list_for_each_entry_safe which uses
1818 * __typeof__(*wait_node).
1819 */
1820 wait_node = NULL;
1821 break;
1822 } else {
1823 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1824 goto error;
1825 }
1826 }
1827 }
1828
1829 if (nb_fd > 0) {
1830 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1831 }
1832
1833 end:
1834 lttng_poll_clean(&events);
1835 return;
1836
1837 error:
1838 lttng_poll_clean(&events);
1839 error_create:
1840 ERR("Unable to sanitize wait queue");
1841 return;
1842 }
1843
1844 /*
1845 * Dispatch request from the registration threads to the application
1846 * communication thread.
1847 */
1848 static void *thread_dispatch_ust_registration(void *data)
1849 {
1850 int ret, err = -1;
1851 struct cds_wfcq_node *node;
1852 struct ust_command *ust_cmd = NULL;
1853 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1854 struct ust_reg_wait_queue wait_queue = {
1855 .count = 0,
1856 };
1857
1858 rcu_register_thread();
1859
1860 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1861
1862 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1863 goto error_testpoint;
1864 }
1865
1866 health_code_update();
1867
1868 CDS_INIT_LIST_HEAD(&wait_queue.head);
1869
1870 DBG("[thread] Dispatch UST command started");
1871
1872 for (;;) {
1873 health_code_update();
1874
1875 /* Atomically prepare the queue futex */
1876 futex_nto1_prepare(&ust_cmd_queue.futex);
1877
1878 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1879 break;
1880 }
1881
1882 do {
1883 struct ust_app *app = NULL;
1884 ust_cmd = NULL;
1885
1886 /*
1887 * Make sure we don't have node(s) that have hung up before receiving
1888 * the notify socket. This is to clean the list in order to avoid
1889 * memory leaks from notify socket that are never seen.
1890 */
1891 sanitize_wait_queue(&wait_queue);
1892
1893 health_code_update();
1894 /* Dequeue command for registration */
1895 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1896 if (node == NULL) {
1897 DBG("Woken up but nothing in the UST command queue");
1898 /* Continue thread execution */
1899 break;
1900 }
1901
1902 ust_cmd = caa_container_of(node, struct ust_command, node);
1903
1904 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1905 " gid:%d sock:%d name:%s (version %d.%d)",
1906 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1907 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1908 ust_cmd->sock, ust_cmd->reg_msg.name,
1909 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1910
1911 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1912 wait_node = zmalloc(sizeof(*wait_node));
1913 if (!wait_node) {
1914 PERROR("zmalloc wait_node dispatch");
1915 ret = close(ust_cmd->sock);
1916 if (ret < 0) {
1917 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1918 }
1919 lttng_fd_put(LTTNG_FD_APPS, 1);
1920 free(ust_cmd);
1921 goto error;
1922 }
1923 CDS_INIT_LIST_HEAD(&wait_node->head);
1924
1925 /* Create application object if socket is CMD. */
1926 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1927 ust_cmd->sock);
1928 if (!wait_node->app) {
1929 ret = close(ust_cmd->sock);
1930 if (ret < 0) {
1931 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1932 }
1933 lttng_fd_put(LTTNG_FD_APPS, 1);
1934 free(wait_node);
1935 free(ust_cmd);
1936 continue;
1937 }
1938 /*
1939 * Add application to the wait queue so we can set the notify
1940 * socket before putting this object in the global ht.
1941 */
1942 cds_list_add(&wait_node->head, &wait_queue.head);
1943 wait_queue.count++;
1944
1945 free(ust_cmd);
1946 /*
1947 * We have to continue here since we don't have the notify
1948 * socket and the application MUST be added to the hash table
1949 * only at that moment.
1950 */
1951 continue;
1952 } else {
1953 /*
1954 * Look for the application in the local wait queue and set the
1955 * notify socket if found.
1956 */
1957 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1958 &wait_queue.head, head) {
1959 health_code_update();
1960 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1961 wait_node->app->notify_sock = ust_cmd->sock;
1962 cds_list_del(&wait_node->head);
1963 wait_queue.count--;
1964 app = wait_node->app;
1965 free(wait_node);
1966 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1967 break;
1968 }
1969 }
1970
1971 /*
1972 * With no application at this stage the received socket is
1973 * basically useless so close it before we free the cmd data
1974 * structure for good.
1975 */
1976 if (!app) {
1977 ret = close(ust_cmd->sock);
1978 if (ret < 0) {
1979 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1980 }
1981 lttng_fd_put(LTTNG_FD_APPS, 1);
1982 }
1983 free(ust_cmd);
1984 }
1985
1986 if (app) {
1987 /*
1988 * @session_lock_list
1989 *
1990 * Lock the global session list so from the register up to the
1991 * registration done message, no thread can see the application
1992 * and change its state.
1993 */
1994 session_lock_list();
1995 rcu_read_lock();
1996
1997 /*
1998 * Add application to the global hash table. This needs to be
1999 * done before the update to the UST registry can locate the
2000 * application.
2001 */
2002 ust_app_add(app);
2003
2004 /* Set app version. This call will print an error if needed. */
2005 (void) ust_app_version(app);
2006
2007 /* Send notify socket through the notify pipe. */
2008 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2009 app->notify_sock);
2010 if (ret < 0) {
2011 rcu_read_unlock();
2012 session_unlock_list();
2013 /*
2014 * No notify thread, stop the UST tracing. However, this is
2015 * not an internal error of the this thread thus setting
2016 * the health error code to a normal exit.
2017 */
2018 err = 0;
2019 goto error;
2020 }
2021
2022 /*
2023 * Update newly registered application with the tracing
2024 * registry info already enabled information.
2025 */
2026 update_ust_app(app->sock);
2027
2028 /*
2029 * Don't care about return value. Let the manage apps threads
2030 * handle app unregistration upon socket close.
2031 */
2032 (void) ust_app_register_done(app);
2033
2034 /*
2035 * Even if the application socket has been closed, send the app
2036 * to the thread and unregistration will take place at that
2037 * place.
2038 */
2039 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2040 if (ret < 0) {
2041 rcu_read_unlock();
2042 session_unlock_list();
2043 /*
2044 * No apps. thread, stop the UST tracing. However, this is
2045 * not an internal error of the this thread thus setting
2046 * the health error code to a normal exit.
2047 */
2048 err = 0;
2049 goto error;
2050 }
2051
2052 rcu_read_unlock();
2053 session_unlock_list();
2054 }
2055 } while (node != NULL);
2056
2057 health_poll_entry();
2058 /* Futex wait on queue. Blocking call on futex() */
2059 futex_nto1_wait(&ust_cmd_queue.futex);
2060 health_poll_exit();
2061 }
2062 /* Normal exit, no error */
2063 err = 0;
2064
2065 error:
2066 /* Clean up wait queue. */
2067 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2068 &wait_queue.head, head) {
2069 cds_list_del(&wait_node->head);
2070 wait_queue.count--;
2071 free(wait_node);
2072 }
2073
2074 /* Empty command queue. */
2075 for (;;) {
2076 /* Dequeue command for registration */
2077 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2078 if (node == NULL) {
2079 break;
2080 }
2081 ust_cmd = caa_container_of(node, struct ust_command, node);
2082 ret = close(ust_cmd->sock);
2083 if (ret < 0) {
2084 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2085 }
2086 lttng_fd_put(LTTNG_FD_APPS, 1);
2087 free(ust_cmd);
2088 }
2089
2090 error_testpoint:
2091 DBG("Dispatch thread dying");
2092 if (err) {
2093 health_error();
2094 ERR("Health error occurred in %s", __func__);
2095 }
2096 health_unregister(health_sessiond);
2097 rcu_unregister_thread();
2098 return NULL;
2099 }
2100
2101 /*
2102 * This thread manage application registration.
2103 */
2104 static void *thread_registration_apps(void *data)
2105 {
2106 int sock = -1, i, ret, pollfd, err = -1;
2107 uint32_t revents, nb_fd;
2108 struct lttng_poll_event events;
2109 /*
2110 * Get allocated in this thread, enqueued to a global queue, dequeued and
2111 * freed in the manage apps thread.
2112 */
2113 struct ust_command *ust_cmd = NULL;
2114
2115 DBG("[thread] Manage application registration started");
2116
2117 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2118
2119 if (testpoint(sessiond_thread_registration_apps)) {
2120 goto error_testpoint;
2121 }
2122
2123 ret = lttcomm_listen_unix_sock(apps_sock);
2124 if (ret < 0) {
2125 goto error_listen;
2126 }
2127
2128 /*
2129 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2130 * more will be added to this poll set.
2131 */
2132 ret = sessiond_set_thread_pollset(&events, 2);
2133 if (ret < 0) {
2134 goto error_create_poll;
2135 }
2136
2137 /* Add the application registration socket */
2138 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2139 if (ret < 0) {
2140 goto error_poll_add;
2141 }
2142
2143 /* Notify all applications to register */
2144 ret = notify_ust_apps(1);
2145 if (ret < 0) {
2146 ERR("Failed to notify applications or create the wait shared memory.\n"
2147 "Execution continues but there might be problem for already\n"
2148 "running applications that wishes to register.");
2149 }
2150
2151 while (1) {
2152 DBG("Accepting application registration");
2153
2154 /* Inifinite blocking call, waiting for transmission */
2155 restart:
2156 health_poll_entry();
2157 ret = lttng_poll_wait(&events, -1);
2158 health_poll_exit();
2159 if (ret < 0) {
2160 /*
2161 * Restart interrupted system call.
2162 */
2163 if (errno == EINTR) {
2164 goto restart;
2165 }
2166 goto error;
2167 }
2168
2169 nb_fd = ret;
2170
2171 for (i = 0; i < nb_fd; i++) {
2172 health_code_update();
2173
2174 /* Fetch once the poll data */
2175 revents = LTTNG_POLL_GETEV(&events, i);
2176 pollfd = LTTNG_POLL_GETFD(&events, i);
2177
2178 if (!revents) {
2179 /* No activity for this FD (poll implementation). */
2180 continue;
2181 }
2182
2183 /* Thread quit pipe has been closed. Killing thread. */
2184 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2185 if (ret) {
2186 err = 0;
2187 goto exit;
2188 }
2189
2190 /* Event on the registration socket */
2191 if (pollfd == apps_sock) {
2192 if (revents & LPOLLIN) {
2193 sock = lttcomm_accept_unix_sock(apps_sock);
2194 if (sock < 0) {
2195 goto error;
2196 }
2197
2198 /*
2199 * Set socket timeout for both receiving and ending.
2200 * app_socket_timeout is in seconds, whereas
2201 * lttcomm_setsockopt_rcv_timeout and
2202 * lttcomm_setsockopt_snd_timeout expect msec as
2203 * parameter.
2204 */
2205 if (app_socket_timeout >= 0) {
2206 (void) lttcomm_setsockopt_rcv_timeout(sock,
2207 app_socket_timeout * 1000);
2208 (void) lttcomm_setsockopt_snd_timeout(sock,
2209 app_socket_timeout * 1000);
2210 }
2211
2212 /*
2213 * Set the CLOEXEC flag. Return code is useless because
2214 * either way, the show must go on.
2215 */
2216 (void) utils_set_fd_cloexec(sock);
2217
2218 /* Create UST registration command for enqueuing */
2219 ust_cmd = zmalloc(sizeof(struct ust_command));
2220 if (ust_cmd == NULL) {
2221 PERROR("ust command zmalloc");
2222 ret = close(sock);
2223 if (ret) {
2224 PERROR("close");
2225 }
2226 goto error;
2227 }
2228
2229 /*
2230 * Using message-based transmissions to ensure we don't
2231 * have to deal with partially received messages.
2232 */
2233 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2234 if (ret < 0) {
2235 ERR("Exhausted file descriptors allowed for applications.");
2236 free(ust_cmd);
2237 ret = close(sock);
2238 if (ret) {
2239 PERROR("close");
2240 }
2241 sock = -1;
2242 continue;
2243 }
2244
2245 health_code_update();
2246 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2247 if (ret < 0) {
2248 free(ust_cmd);
2249 /* Close socket of the application. */
2250 ret = close(sock);
2251 if (ret) {
2252 PERROR("close");
2253 }
2254 lttng_fd_put(LTTNG_FD_APPS, 1);
2255 sock = -1;
2256 continue;
2257 }
2258 health_code_update();
2259
2260 ust_cmd->sock = sock;
2261 sock = -1;
2262
2263 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2264 " gid:%d sock:%d name:%s (version %d.%d)",
2265 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2266 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2267 ust_cmd->sock, ust_cmd->reg_msg.name,
2268 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2269
2270 /*
2271 * Lock free enqueue the registration request. The red pill
2272 * has been taken! This apps will be part of the *system*.
2273 */
2274 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2275
2276 /*
2277 * Wake the registration queue futex. Implicit memory
2278 * barrier with the exchange in cds_wfcq_enqueue.
2279 */
2280 futex_nto1_wake(&ust_cmd_queue.futex);
2281 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2282 ERR("Register apps socket poll error");
2283 goto error;
2284 } else {
2285 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2286 goto error;
2287 }
2288 }
2289 }
2290 }
2291
2292 exit:
2293 error:
2294 /* Notify that the registration thread is gone */
2295 notify_ust_apps(0);
2296
2297 if (apps_sock >= 0) {
2298 ret = close(apps_sock);
2299 if (ret) {
2300 PERROR("close");
2301 }
2302 }
2303 if (sock >= 0) {
2304 ret = close(sock);
2305 if (ret) {
2306 PERROR("close");
2307 }
2308 lttng_fd_put(LTTNG_FD_APPS, 1);
2309 }
2310 unlink(apps_unix_sock_path);
2311
2312 error_poll_add:
2313 lttng_poll_clean(&events);
2314 error_listen:
2315 error_create_poll:
2316 error_testpoint:
2317 DBG("UST Registration thread cleanup complete");
2318 if (err) {
2319 health_error();
2320 ERR("Health error occurred in %s", __func__);
2321 }
2322 health_unregister(health_sessiond);
2323
2324 return NULL;
2325 }
2326
2327 /*
2328 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2329 * exec or it will fails.
2330 */
2331 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2332 {
2333 int ret, clock_ret;
2334 struct timespec timeout;
2335
2336 /*
2337 * Make sure we set the readiness flag to 0 because we are NOT ready.
2338 * This access to consumer_thread_is_ready does not need to be
2339 * protected by consumer_data.cond_mutex (yet) since the consumer
2340 * management thread has not been started at this point.
2341 */
2342 consumer_data->consumer_thread_is_ready = 0;
2343
2344 /* Setup pthread condition */
2345 ret = pthread_condattr_init(&consumer_data->condattr);
2346 if (ret) {
2347 errno = ret;
2348 PERROR("pthread_condattr_init consumer data");
2349 goto error;
2350 }
2351
2352 /*
2353 * Set the monotonic clock in order to make sure we DO NOT jump in time
2354 * between the clock_gettime() call and the timedwait call. See bug #324
2355 * for a more details and how we noticed it.
2356 */
2357 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2358 if (ret) {
2359 errno = ret;
2360 PERROR("pthread_condattr_setclock consumer data");
2361 goto error;
2362 }
2363
2364 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2365 if (ret) {
2366 errno = ret;
2367 PERROR("pthread_cond_init consumer data");
2368 goto error;
2369 }
2370
2371 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2372 thread_manage_consumer, consumer_data);
2373 if (ret) {
2374 errno = ret;
2375 PERROR("pthread_create consumer");
2376 ret = -1;
2377 goto error;
2378 }
2379
2380 /* We are about to wait on a pthread condition */
2381 pthread_mutex_lock(&consumer_data->cond_mutex);
2382
2383 /* Get time for sem_timedwait absolute timeout */
2384 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2385 /*
2386 * Set the timeout for the condition timed wait even if the clock gettime
2387 * call fails since we might loop on that call and we want to avoid to
2388 * increment the timeout too many times.
2389 */
2390 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2391
2392 /*
2393 * The following loop COULD be skipped in some conditions so this is why we
2394 * set ret to 0 in order to make sure at least one round of the loop is
2395 * done.
2396 */
2397 ret = 0;
2398
2399 /*
2400 * Loop until the condition is reached or when a timeout is reached. Note
2401 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2402 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2403 * possible. This loop does not take any chances and works with both of
2404 * them.
2405 */
2406 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2407 if (clock_ret < 0) {
2408 PERROR("clock_gettime spawn consumer");
2409 /* Infinite wait for the consumerd thread to be ready */
2410 ret = pthread_cond_wait(&consumer_data->cond,
2411 &consumer_data->cond_mutex);
2412 } else {
2413 ret = pthread_cond_timedwait(&consumer_data->cond,
2414 &consumer_data->cond_mutex, &timeout);
2415 }
2416 }
2417
2418 /* Release the pthread condition */
2419 pthread_mutex_unlock(&consumer_data->cond_mutex);
2420
2421 if (ret != 0) {
2422 errno = ret;
2423 if (ret == ETIMEDOUT) {
2424 int pth_ret;
2425
2426 /*
2427 * Call has timed out so we kill the kconsumerd_thread and return
2428 * an error.
2429 */
2430 ERR("Condition timed out. The consumer thread was never ready."
2431 " Killing it");
2432 pth_ret = pthread_cancel(consumer_data->thread);
2433 if (pth_ret < 0) {
2434 PERROR("pthread_cancel consumer thread");
2435 }
2436 } else {
2437 PERROR("pthread_cond_wait failed consumer thread");
2438 }
2439 /* Caller is expecting a negative value on failure. */
2440 ret = -1;
2441 goto error;
2442 }
2443
2444 pthread_mutex_lock(&consumer_data->pid_mutex);
2445 if (consumer_data->pid == 0) {
2446 ERR("Consumerd did not start");
2447 pthread_mutex_unlock(&consumer_data->pid_mutex);
2448 goto error;
2449 }
2450 pthread_mutex_unlock(&consumer_data->pid_mutex);
2451
2452 return 0;
2453
2454 error:
2455 return ret;
2456 }
2457
2458 /*
2459 * Join consumer thread
2460 */
2461 static int join_consumer_thread(struct consumer_data *consumer_data)
2462 {
2463 void *status;
2464
2465 /* Consumer pid must be a real one. */
2466 if (consumer_data->pid > 0) {
2467 int ret;
2468 ret = kill(consumer_data->pid, SIGTERM);
2469 if (ret) {
2470 PERROR("Error killing consumer daemon");
2471 return ret;
2472 }
2473 return pthread_join(consumer_data->thread, &status);
2474 } else {
2475 return 0;
2476 }
2477 }
2478
2479 /*
2480 * Fork and exec a consumer daemon (consumerd).
2481 *
2482 * Return pid if successful else -1.
2483 */
2484 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2485 {
2486 int ret;
2487 pid_t pid;
2488 const char *consumer_to_use;
2489 const char *verbosity;
2490 struct stat st;
2491
2492 DBG("Spawning consumerd");
2493
2494 pid = fork();
2495 if (pid == 0) {
2496 /*
2497 * Exec consumerd.
2498 */
2499 if (opt_verbose_consumer) {
2500 verbosity = "--verbose";
2501 } else if (lttng_opt_quiet) {
2502 verbosity = "--quiet";
2503 } else {
2504 verbosity = "";
2505 }
2506
2507 switch (consumer_data->type) {
2508 case LTTNG_CONSUMER_KERNEL:
2509 /*
2510 * Find out which consumerd to execute. We will first try the
2511 * 64-bit path, then the sessiond's installation directory, and
2512 * fallback on the 32-bit one,
2513 */
2514 DBG3("Looking for a kernel consumer at these locations:");
2515 DBG3(" 1) %s", consumerd64_bin);
2516 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2517 DBG3(" 3) %s", consumerd32_bin);
2518 if (stat(consumerd64_bin, &st) == 0) {
2519 DBG3("Found location #1");
2520 consumer_to_use = consumerd64_bin;
2521 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2522 DBG3("Found location #2");
2523 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2524 } else if (stat(consumerd32_bin, &st) == 0) {
2525 DBG3("Found location #3");
2526 consumer_to_use = consumerd32_bin;
2527 } else {
2528 DBG("Could not find any valid consumerd executable");
2529 ret = -EINVAL;
2530 goto error;
2531 }
2532 DBG("Using kernel consumer at: %s", consumer_to_use);
2533 ret = execl(consumer_to_use,
2534 "lttng-consumerd", verbosity, "-k",
2535 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2536 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2537 "--group", tracing_group_name,
2538 NULL);
2539 break;
2540 case LTTNG_CONSUMER64_UST:
2541 {
2542 char *tmpnew = NULL;
2543
2544 if (consumerd64_libdir[0] != '\0') {
2545 char *tmp;
2546 size_t tmplen;
2547
2548 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2549 if (!tmp) {
2550 tmp = "";
2551 }
2552 tmplen = strlen("LD_LIBRARY_PATH=")
2553 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2554 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2555 if (!tmpnew) {
2556 ret = -ENOMEM;
2557 goto error;
2558 }
2559 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2560 strcat(tmpnew, consumerd64_libdir);
2561 if (tmp[0] != '\0') {
2562 strcat(tmpnew, ":");
2563 strcat(tmpnew, tmp);
2564 }
2565 ret = putenv(tmpnew);
2566 if (ret) {
2567 ret = -errno;
2568 free(tmpnew);
2569 goto error;
2570 }
2571 }
2572 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2573 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2574 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2575 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2576 "--group", tracing_group_name,
2577 NULL);
2578 if (consumerd64_libdir[0] != '\0') {
2579 free(tmpnew);
2580 }
2581 break;
2582 }
2583 case LTTNG_CONSUMER32_UST:
2584 {
2585 char *tmpnew = NULL;
2586
2587 if (consumerd32_libdir[0] != '\0') {
2588 char *tmp;
2589 size_t tmplen;
2590
2591 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2592 if (!tmp) {
2593 tmp = "";
2594 }
2595 tmplen = strlen("LD_LIBRARY_PATH=")
2596 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2597 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2598 if (!tmpnew) {
2599 ret = -ENOMEM;
2600 goto error;
2601 }
2602 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2603 strcat(tmpnew, consumerd32_libdir);
2604 if (tmp[0] != '\0') {
2605 strcat(tmpnew, ":");
2606 strcat(tmpnew, tmp);
2607 }
2608 ret = putenv(tmpnew);
2609 if (ret) {
2610 ret = -errno;
2611 free(tmpnew);
2612 goto error;
2613 }
2614 }
2615 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2616 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2617 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2618 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2619 "--group", tracing_group_name,
2620 NULL);
2621 if (consumerd32_libdir[0] != '\0') {
2622 free(tmpnew);
2623 }
2624 break;
2625 }
2626 default:
2627 ERR("unknown consumer type");
2628 exit(EXIT_FAILURE);
2629 }
2630 if (errno != 0) {
2631 PERROR("Consumer execl()");
2632 }
2633 /* Reaching this point, we got a failure on our execl(). */
2634 exit(EXIT_FAILURE);
2635 } else if (pid > 0) {
2636 ret = pid;
2637 } else {
2638 PERROR("start consumer fork");
2639 ret = -errno;
2640 }
2641 error:
2642 return ret;
2643 }
2644
2645 /*
2646 * Spawn the consumerd daemon and session daemon thread.
2647 */
2648 static int start_consumerd(struct consumer_data *consumer_data)
2649 {
2650 int ret;
2651
2652 /*
2653 * Set the listen() state on the socket since there is a possible race
2654 * between the exec() of the consumer daemon and this call if place in the
2655 * consumer thread. See bug #366 for more details.
2656 */
2657 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2658 if (ret < 0) {
2659 goto error;
2660 }
2661
2662 pthread_mutex_lock(&consumer_data->pid_mutex);
2663 if (consumer_data->pid != 0) {
2664 pthread_mutex_unlock(&consumer_data->pid_mutex);
2665 goto end;
2666 }
2667
2668 ret = spawn_consumerd(consumer_data);
2669 if (ret < 0) {
2670 ERR("Spawning consumerd failed");
2671 pthread_mutex_unlock(&consumer_data->pid_mutex);
2672 goto error;
2673 }
2674
2675 /* Setting up the consumer_data pid */
2676 consumer_data->pid = ret;
2677 DBG2("Consumer pid %d", consumer_data->pid);
2678 pthread_mutex_unlock(&consumer_data->pid_mutex);
2679
2680 DBG2("Spawning consumer control thread");
2681 ret = spawn_consumer_thread(consumer_data);
2682 if (ret < 0) {
2683 ERR("Fatal error spawning consumer control thread");
2684 goto error;
2685 }
2686
2687 end:
2688 return 0;
2689
2690 error:
2691 /* Cleanup already created sockets on error. */
2692 if (consumer_data->err_sock >= 0) {
2693 int err;
2694
2695 err = close(consumer_data->err_sock);
2696 if (err < 0) {
2697 PERROR("close consumer data error socket");
2698 }
2699 }
2700 return ret;
2701 }
2702
2703 /*
2704 * Setup necessary data for kernel tracer action.
2705 */
2706 static int init_kernel_tracer(void)
2707 {
2708 int ret;
2709
2710 /* Modprobe lttng kernel modules */
2711 ret = modprobe_lttng_control();
2712 if (ret < 0) {
2713 goto error;
2714 }
2715
2716 /* Open debugfs lttng */
2717 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2718 if (kernel_tracer_fd < 0) {
2719 DBG("Failed to open %s", module_proc_lttng);
2720 goto error_open;
2721 }
2722
2723 /* Validate kernel version */
2724 ret = kernel_validate_version(kernel_tracer_fd);
2725 if (ret < 0) {
2726 goto error_version;
2727 }
2728
2729 ret = modprobe_lttng_data();
2730 if (ret < 0) {
2731 goto error_modules;
2732 }
2733
2734 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2735 return 0;
2736
2737 error_version:
2738 modprobe_remove_lttng_control();
2739 ret = close(kernel_tracer_fd);
2740 if (ret) {
2741 PERROR("close");
2742 }
2743 kernel_tracer_fd = -1;
2744 return LTTNG_ERR_KERN_VERSION;
2745
2746 error_modules:
2747 ret = close(kernel_tracer_fd);
2748 if (ret) {
2749 PERROR("close");
2750 }
2751
2752 error_open:
2753 modprobe_remove_lttng_control();
2754
2755 error:
2756 WARN("No kernel tracer available");
2757 kernel_tracer_fd = -1;
2758 if (!is_root) {
2759 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2760 } else {
2761 return LTTNG_ERR_KERN_NA;
2762 }
2763 }
2764
2765
2766 /*
2767 * Copy consumer output from the tracing session to the domain session. The
2768 * function also applies the right modification on a per domain basis for the
2769 * trace files destination directory.
2770 *
2771 * Should *NOT* be called with RCU read-side lock held.
2772 */
2773 static int copy_session_consumer(int domain, struct ltt_session *session)
2774 {
2775 int ret;
2776 const char *dir_name;
2777 struct consumer_output *consumer;
2778
2779 assert(session);
2780 assert(session->consumer);
2781
2782 switch (domain) {
2783 case LTTNG_DOMAIN_KERNEL:
2784 DBG3("Copying tracing session consumer output in kernel session");
2785 /*
2786 * XXX: We should audit the session creation and what this function
2787 * does "extra" in order to avoid a destroy since this function is used
2788 * in the domain session creation (kernel and ust) only. Same for UST
2789 * domain.
2790 */
2791 if (session->kernel_session->consumer) {
2792 consumer_output_put(session->kernel_session->consumer);
2793 }
2794 session->kernel_session->consumer =
2795 consumer_copy_output(session->consumer);
2796 /* Ease our life a bit for the next part */
2797 consumer = session->kernel_session->consumer;
2798 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2799 break;
2800 case LTTNG_DOMAIN_JUL:
2801 case LTTNG_DOMAIN_LOG4J:
2802 case LTTNG_DOMAIN_PYTHON:
2803 case LTTNG_DOMAIN_UST:
2804 DBG3("Copying tracing session consumer output in UST session");
2805 if (session->ust_session->consumer) {
2806 consumer_output_put(session->ust_session->consumer);
2807 }
2808 session->ust_session->consumer =
2809 consumer_copy_output(session->consumer);
2810 /* Ease our life a bit for the next part */
2811 consumer = session->ust_session->consumer;
2812 dir_name = DEFAULT_UST_TRACE_DIR;
2813 break;
2814 default:
2815 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2816 goto error;
2817 }
2818
2819 /* Append correct directory to subdir */
2820 strncat(consumer->subdir, dir_name,
2821 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2822 DBG3("Copy session consumer subdir %s", consumer->subdir);
2823
2824 ret = LTTNG_OK;
2825
2826 error:
2827 return ret;
2828 }
2829
2830 /*
2831 * Create an UST session and add it to the session ust list.
2832 *
2833 * Should *NOT* be called with RCU read-side lock held.
2834 */
2835 static int create_ust_session(struct ltt_session *session,
2836 struct lttng_domain *domain)
2837 {
2838 int ret;
2839 struct ltt_ust_session *lus = NULL;
2840
2841 assert(session);
2842 assert(domain);
2843 assert(session->consumer);
2844
2845 switch (domain->type) {
2846 case LTTNG_DOMAIN_JUL:
2847 case LTTNG_DOMAIN_LOG4J:
2848 case LTTNG_DOMAIN_PYTHON:
2849 case LTTNG_DOMAIN_UST:
2850 break;
2851 default:
2852 ERR("Unknown UST domain on create session %d", domain->type);
2853 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2854 goto error;
2855 }
2856
2857 DBG("Creating UST session");
2858
2859 lus = trace_ust_create_session(session->id);
2860 if (lus == NULL) {
2861 ret = LTTNG_ERR_UST_SESS_FAIL;
2862 goto error;
2863 }
2864
2865 lus->uid = session->uid;
2866 lus->gid = session->gid;
2867 lus->output_traces = session->output_traces;
2868 lus->snapshot_mode = session->snapshot_mode;
2869 lus->live_timer_interval = session->live_timer;
2870 session->ust_session = lus;
2871 if (session->shm_path[0]) {
2872 strncpy(lus->root_shm_path, session->shm_path,
2873 sizeof(lus->root_shm_path));
2874 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2875 strncpy(lus->shm_path, session->shm_path,
2876 sizeof(lus->shm_path));
2877 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2878 strncat(lus->shm_path, "/ust",
2879 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2880 }
2881 /* Copy session output to the newly created UST session */
2882 ret = copy_session_consumer(domain->type, session);
2883 if (ret != LTTNG_OK) {
2884 goto error;
2885 }
2886
2887 return LTTNG_OK;
2888
2889 error:
2890 free(lus);
2891 session->ust_session = NULL;
2892 return ret;
2893 }
2894
2895 /*
2896 * Create a kernel tracer session then create the default channel.
2897 */
2898 static int create_kernel_session(struct ltt_session *session)
2899 {
2900 int ret;
2901
2902 DBG("Creating kernel session");
2903
2904 ret = kernel_create_session(session, kernel_tracer_fd);
2905 if (ret < 0) {
2906 ret = LTTNG_ERR_KERN_SESS_FAIL;
2907 goto error;
2908 }
2909
2910 /* Code flow safety */
2911 assert(session->kernel_session);
2912
2913 /* Copy session output to the newly created Kernel session */
2914 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2915 if (ret != LTTNG_OK) {
2916 goto error;
2917 }
2918
2919 /* Create directory(ies) on local filesystem. */
2920 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2921 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2922 ret = run_as_mkdir_recursive(
2923 session->kernel_session->consumer->dst.trace_path,
2924 S_IRWXU | S_IRWXG, session->uid, session->gid);
2925 if (ret < 0) {
2926 if (errno != EEXIST) {
2927 ERR("Trace directory creation error");
2928 goto error;
2929 }
2930 }
2931 }
2932
2933 session->kernel_session->uid = session->uid;
2934 session->kernel_session->gid = session->gid;
2935 session->kernel_session->output_traces = session->output_traces;
2936 session->kernel_session->snapshot_mode = session->snapshot_mode;
2937
2938 return LTTNG_OK;
2939
2940 error:
2941 trace_kernel_destroy_session(session->kernel_session);
2942 session->kernel_session = NULL;
2943 return ret;
2944 }
2945
2946 /*
2947 * Count number of session permitted by uid/gid.
2948 */
2949 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2950 {
2951 unsigned int i = 0;
2952 struct ltt_session *session;
2953
2954 DBG("Counting number of available session for UID %d GID %d",
2955 uid, gid);
2956 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2957 /*
2958 * Only list the sessions the user can control.
2959 */
2960 if (!session_access_ok(session, uid, gid)) {
2961 continue;
2962 }
2963 i++;
2964 }
2965 return i;
2966 }
2967
2968 /*
2969 * Process the command requested by the lttng client within the command
2970 * context structure. This function make sure that the return structure (llm)
2971 * is set and ready for transmission before returning.
2972 *
2973 * Return any error encountered or 0 for success.
2974 *
2975 * "sock" is only used for special-case var. len data.
2976 *
2977 * Should *NOT* be called with RCU read-side lock held.
2978 */
2979 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2980 int *sock_error)
2981 {
2982 int ret = LTTNG_OK;
2983 int need_tracing_session = 1;
2984 int need_domain;
2985
2986 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2987
2988 assert(!rcu_read_ongoing());
2989
2990 *sock_error = 0;
2991
2992 switch (cmd_ctx->lsm->cmd_type) {
2993 case LTTNG_CREATE_SESSION:
2994 case LTTNG_CREATE_SESSION_SNAPSHOT:
2995 case LTTNG_CREATE_SESSION_LIVE:
2996 case LTTNG_DESTROY_SESSION:
2997 case LTTNG_LIST_SESSIONS:
2998 case LTTNG_LIST_DOMAINS:
2999 case LTTNG_START_TRACE:
3000 case LTTNG_STOP_TRACE:
3001 case LTTNG_DATA_PENDING:
3002 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3003 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3004 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3005 case LTTNG_SNAPSHOT_RECORD:
3006 case LTTNG_SAVE_SESSION:
3007 case LTTNG_SET_SESSION_SHM_PATH:
3008 case LTTNG_REGENERATE_METADATA:
3009 case LTTNG_REGENERATE_STATEDUMP:
3010 need_domain = 0;
3011 break;
3012 default:
3013 need_domain = 1;
3014 }
3015
3016 if (opt_no_kernel && need_domain
3017 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3018 if (!is_root) {
3019 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3020 } else {
3021 ret = LTTNG_ERR_KERN_NA;
3022 }
3023 goto error;
3024 }
3025
3026 /* Deny register consumer if we already have a spawned consumer. */
3027 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3028 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3029 if (kconsumer_data.pid > 0) {
3030 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3031 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3032 goto error;
3033 }
3034 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3035 }
3036
3037 /*
3038 * Check for command that don't needs to allocate a returned payload. We do
3039 * this here so we don't have to make the call for no payload at each
3040 * command.
3041 */
3042 switch(cmd_ctx->lsm->cmd_type) {
3043 case LTTNG_LIST_SESSIONS:
3044 case LTTNG_LIST_TRACEPOINTS:
3045 case LTTNG_LIST_TRACEPOINT_FIELDS:
3046 case LTTNG_LIST_DOMAINS:
3047 case LTTNG_LIST_CHANNELS:
3048 case LTTNG_LIST_EVENTS:
3049 case LTTNG_LIST_SYSCALLS:
3050 case LTTNG_LIST_TRACKER_PIDS:
3051 case LTTNG_DATA_PENDING:
3052 break;
3053 default:
3054 /* Setup lttng message with no payload */
3055 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3056 if (ret < 0) {
3057 /* This label does not try to unlock the session */
3058 goto init_setup_error;
3059 }
3060 }
3061
3062 /* Commands that DO NOT need a session. */
3063 switch (cmd_ctx->lsm->cmd_type) {
3064 case LTTNG_CREATE_SESSION:
3065 case LTTNG_CREATE_SESSION_SNAPSHOT:
3066 case LTTNG_CREATE_SESSION_LIVE:
3067 case LTTNG_LIST_SESSIONS:
3068 case LTTNG_LIST_TRACEPOINTS:
3069 case LTTNG_LIST_SYSCALLS:
3070 case LTTNG_LIST_TRACEPOINT_FIELDS:
3071 case LTTNG_SAVE_SESSION:
3072 need_tracing_session = 0;
3073 break;
3074 default:
3075 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3076 /*
3077 * We keep the session list lock across _all_ commands
3078 * for now, because the per-session lock does not
3079 * handle teardown properly.
3080 */
3081 session_lock_list();
3082 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3083 if (cmd_ctx->session == NULL) {
3084 ret = LTTNG_ERR_SESS_NOT_FOUND;
3085 goto error;
3086 } else {
3087 /* Acquire lock for the session */
3088 session_lock(cmd_ctx->session);
3089 }
3090 break;
3091 }
3092
3093 /*
3094 * Commands that need a valid session but should NOT create one if none
3095 * exists. Instead of creating one and destroying it when the command is
3096 * handled, process that right before so we save some round trip in useless
3097 * code path.
3098 */
3099 switch (cmd_ctx->lsm->cmd_type) {
3100 case LTTNG_DISABLE_CHANNEL:
3101 case LTTNG_DISABLE_EVENT:
3102 switch (cmd_ctx->lsm->domain.type) {
3103 case LTTNG_DOMAIN_KERNEL:
3104 if (!cmd_ctx->session->kernel_session) {
3105 ret = LTTNG_ERR_NO_CHANNEL;
3106 goto error;
3107 }
3108 break;
3109 case LTTNG_DOMAIN_JUL:
3110 case LTTNG_DOMAIN_LOG4J:
3111 case LTTNG_DOMAIN_PYTHON:
3112 case LTTNG_DOMAIN_UST:
3113 if (!cmd_ctx->session->ust_session) {
3114 ret = LTTNG_ERR_NO_CHANNEL;
3115 goto error;
3116 }
3117 break;
3118 default:
3119 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3120 goto error;
3121 }
3122 default:
3123 break;
3124 }
3125
3126 if (!need_domain) {
3127 goto skip_domain;
3128 }
3129
3130 /*
3131 * Check domain type for specific "pre-action".
3132 */
3133 switch (cmd_ctx->lsm->domain.type) {
3134 case LTTNG_DOMAIN_KERNEL:
3135 if (!is_root) {
3136 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3137 goto error;
3138 }
3139
3140 /* Kernel tracer check */
3141 if (kernel_tracer_fd == -1) {
3142 /* Basically, load kernel tracer modules */
3143 ret = init_kernel_tracer();
3144 if (ret != 0) {
3145 goto error;
3146 }
3147 }
3148
3149 /* Consumer is in an ERROR state. Report back to client */
3150 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3151 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3152 goto error;
3153 }
3154
3155 /* Need a session for kernel command */
3156 if (need_tracing_session) {
3157 if (cmd_ctx->session->kernel_session == NULL) {
3158 ret = create_kernel_session(cmd_ctx->session);
3159 if (ret < 0) {
3160 ret = LTTNG_ERR_KERN_SESS_FAIL;
3161 goto error;
3162 }
3163 }
3164
3165 /* Start the kernel consumer daemon */
3166 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3167 if (kconsumer_data.pid == 0 &&
3168 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3169 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3170 ret = start_consumerd(&kconsumer_data);
3171 if (ret < 0) {
3172 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3173 goto error;
3174 }
3175 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3176 } else {
3177 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3178 }
3179
3180 /*
3181 * The consumer was just spawned so we need to add the socket to
3182 * the consumer output of the session if exist.
3183 */
3184 ret = consumer_create_socket(&kconsumer_data,
3185 cmd_ctx->session->kernel_session->consumer);
3186 if (ret < 0) {
3187 goto error;
3188 }
3189 }
3190
3191 break;
3192 case LTTNG_DOMAIN_JUL:
3193 case LTTNG_DOMAIN_LOG4J:
3194 case LTTNG_DOMAIN_PYTHON:
3195 case LTTNG_DOMAIN_UST:
3196 {
3197 if (!ust_app_supported()) {
3198 ret = LTTNG_ERR_NO_UST;
3199 goto error;
3200 }
3201 /* Consumer is in an ERROR state. Report back to client */
3202 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3203 ret = LTTNG_ERR_NO_USTCONSUMERD;
3204 goto error;
3205 }
3206
3207 if (need_tracing_session) {
3208 /* Create UST session if none exist. */
3209 if (cmd_ctx->session->ust_session == NULL) {
3210 ret = create_ust_session(cmd_ctx->session,
3211 &cmd_ctx->lsm->domain);
3212 if (ret != LTTNG_OK) {
3213 goto error;
3214 }
3215 }
3216
3217 /* Start the UST consumer daemons */
3218 /* 64-bit */
3219 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3220 if (consumerd64_bin[0] != '\0' &&
3221 ustconsumer64_data.pid == 0 &&
3222 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3223 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3224 ret = start_consumerd(&ustconsumer64_data);
3225 if (ret < 0) {
3226 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3227 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3228 goto error;
3229 }
3230
3231 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3232 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3233 } else {
3234 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3235 }
3236
3237 /*
3238 * Setup socket for consumer 64 bit. No need for atomic access
3239 * since it was set above and can ONLY be set in this thread.
3240 */
3241 ret = consumer_create_socket(&ustconsumer64_data,
3242 cmd_ctx->session->ust_session->consumer);
3243 if (ret < 0) {
3244 goto error;
3245 }
3246
3247 /* 32-bit */
3248 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3249 if (consumerd32_bin[0] != '\0' &&
3250 ustconsumer32_data.pid == 0 &&
3251 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3252 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3253 ret = start_consumerd(&ustconsumer32_data);
3254 if (ret < 0) {
3255 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3256 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3257 goto error;
3258 }
3259
3260 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3261 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3262 } else {
3263 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3264 }
3265
3266 /*
3267 * Setup socket for consumer 64 bit. No need for atomic access
3268 * since it was set above and can ONLY be set in this thread.
3269 */
3270 ret = consumer_create_socket(&ustconsumer32_data,
3271 cmd_ctx->session->ust_session->consumer);
3272 if (ret < 0) {
3273 goto error;
3274 }
3275 }
3276 break;
3277 }
3278 default:
3279 break;
3280 }
3281 skip_domain:
3282
3283 /* Validate consumer daemon state when start/stop trace command */
3284 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3285 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3286 switch (cmd_ctx->lsm->domain.type) {
3287 case LTTNG_DOMAIN_NONE:
3288 break;
3289 case LTTNG_DOMAIN_JUL:
3290 case LTTNG_DOMAIN_LOG4J:
3291 case LTTNG_DOMAIN_PYTHON:
3292 case LTTNG_DOMAIN_UST:
3293 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3294 ret = LTTNG_ERR_NO_USTCONSUMERD;
3295 goto error;
3296 }
3297 break;
3298 case LTTNG_DOMAIN_KERNEL:
3299 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3300 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3301 goto error;
3302 }
3303 break;
3304 default:
3305 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3306 goto error;
3307 }
3308 }
3309
3310 /*
3311 * Check that the UID or GID match that of the tracing session.
3312 * The root user can interact with all sessions.
3313 */
3314 if (need_tracing_session) {
3315 if (!session_access_ok(cmd_ctx->session,
3316 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3317 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3318 ret = LTTNG_ERR_EPERM;
3319 goto error;
3320 }
3321 }
3322
3323 /*
3324 * Send relayd information to consumer as soon as we have a domain and a
3325 * session defined.
3326 */
3327 if (cmd_ctx->session && need_domain) {
3328 /*
3329 * Setup relayd if not done yet. If the relayd information was already
3330 * sent to the consumer, this call will gracefully return.
3331 */
3332 ret = cmd_setup_relayd(cmd_ctx->session);
3333 if (ret != LTTNG_OK) {
3334 goto error;
3335 }
3336 }
3337
3338 /* Process by command type */
3339 switch (cmd_ctx->lsm->cmd_type) {
3340 case LTTNG_ADD_CONTEXT:
3341 {
3342 /*
3343 * An LTTNG_ADD_CONTEXT command might have a supplementary
3344 * payload if the context being added is an application context.
3345 */
3346 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3347 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3348 char *provider_name = NULL, *context_name = NULL;
3349 size_t provider_name_len =
3350 cmd_ctx->lsm->u.context.provider_name_len;
3351 size_t context_name_len =
3352 cmd_ctx->lsm->u.context.context_name_len;
3353
3354 if (provider_name_len == 0 || context_name_len == 0) {
3355 /*
3356 * Application provider and context names MUST
3357 * be provided.
3358 */
3359 ret = -LTTNG_ERR_INVALID;
3360 goto error;
3361 }
3362
3363 provider_name = zmalloc(provider_name_len + 1);
3364 if (!provider_name) {
3365 ret = -LTTNG_ERR_NOMEM;
3366 goto error;
3367 }
3368 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3369 provider_name;
3370
3371 context_name = zmalloc(context_name_len + 1);
3372 if (!context_name) {
3373 ret = -LTTNG_ERR_NOMEM;
3374 goto error_add_context;
3375 }
3376 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3377 context_name;
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 }
3391
3392 /*
3393 * cmd_add_context assumes ownership of the provider and context
3394 * names.
3395 */
3396 ret = cmd_add_context(cmd_ctx->session,
3397 cmd_ctx->lsm->domain.type,
3398 cmd_ctx->lsm->u.context.channel_name,
3399 &cmd_ctx->lsm->u.context.ctx,
3400 kernel_poll_pipe[1]);
3401
3402 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3403 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3404 error_add_context:
3405 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3406 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3407 if (ret < 0) {
3408 goto error;
3409 }
3410 break;
3411 }
3412 case LTTNG_DISABLE_CHANNEL:
3413 {
3414 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3415 cmd_ctx->lsm->u.disable.channel_name);
3416 break;
3417 }
3418 case LTTNG_DISABLE_EVENT:
3419 {
3420
3421 /*
3422 * FIXME: handle filter; for now we just receive the filter's
3423 * bytecode along with the filter expression which are sent by
3424 * liblttng-ctl and discard them.
3425 *
3426 * This fixes an issue where the client may block while sending
3427 * the filter payload and encounter an error because the session
3428 * daemon closes the socket without ever handling this data.
3429 */
3430 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3431 cmd_ctx->lsm->u.disable.bytecode_len;
3432
3433 if (count) {
3434 char data[LTTNG_FILTER_MAX_LEN];
3435
3436 DBG("Discarding disable event command payload of size %zu", count);
3437 while (count) {
3438 ret = lttcomm_recv_unix_sock(sock, data,
3439 count > sizeof(data) ? sizeof(data) : count);
3440 if (ret < 0) {
3441 goto error;
3442 }
3443
3444 count -= (size_t) ret;
3445 }
3446 }
3447 /* FIXME: passing packed structure to non-packed pointer */
3448 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3449 cmd_ctx->lsm->u.disable.channel_name,
3450 &cmd_ctx->lsm->u.disable.event);
3451 break;
3452 }
3453 case LTTNG_ENABLE_CHANNEL:
3454 {
3455 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3456 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3457 break;
3458 }
3459 case LTTNG_TRACK_PID:
3460 {
3461 ret = cmd_track_pid(cmd_ctx->session,
3462 cmd_ctx->lsm->domain.type,
3463 cmd_ctx->lsm->u.pid_tracker.pid);
3464 break;
3465 }
3466 case LTTNG_UNTRACK_PID:
3467 {
3468 ret = cmd_untrack_pid(cmd_ctx->session,
3469 cmd_ctx->lsm->domain.type,
3470 cmd_ctx->lsm->u.pid_tracker.pid);
3471 break;
3472 }
3473 case LTTNG_ENABLE_EVENT:
3474 {
3475 struct lttng_event_exclusion *exclusion = NULL;
3476 struct lttng_filter_bytecode *bytecode = NULL;
3477 char *filter_expression = NULL;
3478
3479 /* Handle exclusion events and receive it from the client. */
3480 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3481 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3482
3483 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3484 (count * LTTNG_SYMBOL_NAME_LEN));
3485 if (!exclusion) {
3486 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3487 goto error;
3488 }
3489
3490 DBG("Receiving var len exclusion event list from client ...");
3491 exclusion->count = count;
3492 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3493 count * LTTNG_SYMBOL_NAME_LEN);
3494 if (ret <= 0) {
3495 DBG("Nothing recv() from client var len data... continuing");
3496 *sock_error = 1;
3497 free(exclusion);
3498 ret = LTTNG_ERR_EXCLUSION_INVAL;
3499 goto error;
3500 }
3501 }
3502
3503 /* Get filter expression from client. */
3504 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3505 size_t expression_len =
3506 cmd_ctx->lsm->u.enable.expression_len;
3507
3508 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3509 ret = LTTNG_ERR_FILTER_INVAL;
3510 free(exclusion);
3511 goto error;
3512 }
3513
3514 filter_expression = zmalloc(expression_len);
3515 if (!filter_expression) {
3516 free(exclusion);
3517 ret = LTTNG_ERR_FILTER_NOMEM;
3518 goto error;
3519 }
3520
3521 /* Receive var. len. data */
3522 DBG("Receiving var len filter's expression from client ...");
3523 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3524 expression_len);
3525 if (ret <= 0) {
3526 DBG("Nothing recv() from client car len data... continuing");
3527 *sock_error = 1;
3528 free(filter_expression);
3529 free(exclusion);
3530 ret = LTTNG_ERR_FILTER_INVAL;
3531 goto error;
3532 }
3533 }
3534
3535 /* Handle filter and get bytecode from client. */
3536 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3537 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3538
3539 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3540 ret = LTTNG_ERR_FILTER_INVAL;
3541 free(filter_expression);
3542 free(exclusion);
3543 goto error;
3544 }
3545
3546 bytecode = zmalloc(bytecode_len);
3547 if (!bytecode) {
3548 free(filter_expression);
3549 free(exclusion);
3550 ret = LTTNG_ERR_FILTER_NOMEM;
3551 goto error;
3552 }
3553
3554 /* Receive var. len. data */
3555 DBG("Receiving var len filter's bytecode from client ...");
3556 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3557 if (ret <= 0) {
3558 DBG("Nothing recv() from client car len data... continuing");
3559 *sock_error = 1;
3560 free(filter_expression);
3561 free(bytecode);
3562 free(exclusion);
3563 ret = LTTNG_ERR_FILTER_INVAL;
3564 goto error;
3565 }
3566
3567 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3568 free(filter_expression);
3569 free(bytecode);
3570 free(exclusion);
3571 ret = LTTNG_ERR_FILTER_INVAL;
3572 goto error;
3573 }
3574 }
3575
3576 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3577 cmd_ctx->lsm->u.enable.channel_name,
3578 &cmd_ctx->lsm->u.enable.event,
3579 filter_expression, bytecode, exclusion,
3580 kernel_poll_pipe[1]);
3581 break;
3582 }
3583 case LTTNG_LIST_TRACEPOINTS:
3584 {
3585 struct lttng_event *events;
3586 ssize_t nb_events;
3587
3588 session_lock_list();
3589 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3590 session_unlock_list();
3591 if (nb_events < 0) {
3592 /* Return value is a negative lttng_error_code. */
3593 ret = -nb_events;
3594 goto error;
3595 }
3596
3597 /*
3598 * Setup lttng message with payload size set to the event list size in
3599 * bytes and then copy list into the llm payload.
3600 */
3601 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3602 sizeof(struct lttng_event) * nb_events);
3603 free(events);
3604
3605 if (ret < 0) {
3606 goto setup_error;
3607 }
3608
3609 ret = LTTNG_OK;
3610 break;
3611 }
3612 case LTTNG_LIST_TRACEPOINT_FIELDS:
3613 {
3614 struct lttng_event_field *fields;
3615 ssize_t nb_fields;
3616
3617 session_lock_list();
3618 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3619 &fields);
3620 session_unlock_list();
3621 if (nb_fields < 0) {
3622 /* Return value is a negative lttng_error_code. */
3623 ret = -nb_fields;
3624 goto error;
3625 }
3626
3627 /*
3628 * Setup lttng message with payload size set to the event list size in
3629 * bytes and then copy list into the llm payload.
3630 */
3631 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3632 sizeof(struct lttng_event_field) * nb_fields);
3633 free(fields);
3634
3635 if (ret < 0) {
3636 goto setup_error;
3637 }
3638
3639 ret = LTTNG_OK;
3640 break;
3641 }
3642 case LTTNG_LIST_SYSCALLS:
3643 {
3644 struct lttng_event *events;
3645 ssize_t nb_events;
3646
3647 nb_events = cmd_list_syscalls(&events);
3648 if (nb_events < 0) {
3649 /* Return value is a negative lttng_error_code. */
3650 ret = -nb_events;
3651 goto error;
3652 }
3653
3654 /*
3655 * Setup lttng message with payload size set to the event list size in
3656 * bytes and then copy list into the llm payload.
3657 */
3658 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3659 sizeof(struct lttng_event) * nb_events);
3660 free(events);
3661
3662 if (ret < 0) {
3663 goto setup_error;
3664 }
3665
3666 ret = LTTNG_OK;
3667 break;
3668 }
3669 case LTTNG_LIST_TRACKER_PIDS:
3670 {
3671 int32_t *pids = NULL;
3672 ssize_t nr_pids;
3673
3674 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3675 cmd_ctx->lsm->domain.type, &pids);
3676 if (nr_pids < 0) {
3677 /* Return value is a negative lttng_error_code. */
3678 ret = -nr_pids;
3679 goto error;
3680 }
3681
3682 /*
3683 * Setup lttng message with payload size set to the event list size in
3684 * bytes and then copy list into the llm payload.
3685 */
3686 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3687 sizeof(int32_t) * nr_pids);
3688 free(pids);
3689
3690 if (ret < 0) {
3691 goto setup_error;
3692 }
3693
3694 ret = LTTNG_OK;
3695 break;
3696 }
3697 case LTTNG_SET_CONSUMER_URI:
3698 {
3699 size_t nb_uri, len;
3700 struct lttng_uri *uris;
3701
3702 nb_uri = cmd_ctx->lsm->u.uri.size;
3703 len = nb_uri * sizeof(struct lttng_uri);
3704
3705 if (nb_uri == 0) {
3706 ret = LTTNG_ERR_INVALID;
3707 goto error;
3708 }
3709
3710 uris = zmalloc(len);
3711 if (uris == NULL) {
3712 ret = LTTNG_ERR_FATAL;
3713 goto error;
3714 }
3715
3716 /* Receive variable len data */
3717 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3718 ret = lttcomm_recv_unix_sock(sock, uris, len);
3719 if (ret <= 0) {
3720 DBG("No URIs received from client... continuing");
3721 *sock_error = 1;
3722 ret = LTTNG_ERR_SESSION_FAIL;
3723 free(uris);
3724 goto error;
3725 }
3726
3727 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3728 free(uris);
3729 if (ret != LTTNG_OK) {
3730 goto error;
3731 }
3732
3733
3734 break;
3735 }
3736 case LTTNG_START_TRACE:
3737 {
3738 ret = cmd_start_trace(cmd_ctx->session);
3739 break;
3740 }
3741 case LTTNG_STOP_TRACE:
3742 {
3743 ret = cmd_stop_trace(cmd_ctx->session);
3744 break;
3745 }
3746 case LTTNG_CREATE_SESSION:
3747 {
3748 size_t nb_uri, len;
3749 struct lttng_uri *uris = NULL;
3750
3751 nb_uri = cmd_ctx->lsm->u.uri.size;
3752 len = nb_uri * sizeof(struct lttng_uri);
3753
3754 if (nb_uri > 0) {
3755 uris = zmalloc(len);
3756 if (uris == NULL) {
3757 ret = LTTNG_ERR_FATAL;
3758 goto error;
3759 }
3760
3761 /* Receive variable len data */
3762 DBG("Waiting for %zu URIs from client ...", nb_uri);
3763 ret = lttcomm_recv_unix_sock(sock, uris, len);
3764 if (ret <= 0) {
3765 DBG("No URIs received from client... continuing");
3766 *sock_error = 1;
3767 ret = LTTNG_ERR_SESSION_FAIL;
3768 free(uris);
3769 goto error;
3770 }
3771
3772 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3773 DBG("Creating session with ONE network URI is a bad call");
3774 ret = LTTNG_ERR_SESSION_FAIL;
3775 free(uris);
3776 goto error;
3777 }
3778 }
3779
3780 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3781 &cmd_ctx->creds, 0);
3782
3783 free(uris);
3784
3785 break;
3786 }
3787 case LTTNG_DESTROY_SESSION:
3788 {
3789 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3790
3791 /* Set session to NULL so we do not unlock it after free. */
3792 cmd_ctx->session = NULL;
3793 break;
3794 }
3795 case LTTNG_LIST_DOMAINS:
3796 {
3797 ssize_t nb_dom;
3798 struct lttng_domain *domains = NULL;
3799
3800 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3801 if (nb_dom < 0) {
3802 /* Return value is a negative lttng_error_code. */
3803 ret = -nb_dom;
3804 goto error;
3805 }
3806
3807 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3808 nb_dom * sizeof(struct lttng_domain));
3809 free(domains);
3810
3811 if (ret < 0) {
3812 goto setup_error;
3813 }
3814
3815 ret = LTTNG_OK;
3816 break;
3817 }
3818 case LTTNG_LIST_CHANNELS:
3819 {
3820 ssize_t payload_size;
3821 struct lttng_channel *channels = NULL;
3822
3823 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3824 cmd_ctx->session, &channels);
3825 if (payload_size < 0) {
3826 /* Return value is a negative lttng_error_code. */
3827 ret = -payload_size;
3828 goto error;
3829 }
3830
3831 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3832 payload_size);
3833 free(channels);
3834
3835 if (ret < 0) {
3836 goto setup_error;
3837 }
3838
3839 ret = LTTNG_OK;
3840 break;
3841 }
3842 case LTTNG_LIST_EVENTS:
3843 {
3844 ssize_t nb_event;
3845 struct lttng_event *events = NULL;
3846 struct lttcomm_event_command_header cmd_header;
3847 size_t total_size;
3848
3849 memset(&cmd_header, 0, sizeof(cmd_header));
3850 /* Extended infos are included at the end of events */
3851 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3852 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3853 &events, &total_size);
3854
3855 if (nb_event < 0) {
3856 /* Return value is a negative lttng_error_code. */
3857 ret = -nb_event;
3858 goto error;
3859 }
3860
3861 cmd_header.nb_events = nb_event;
3862 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3863 &cmd_header, sizeof(cmd_header));
3864 free(events);
3865
3866 if (ret < 0) {
3867 goto setup_error;
3868 }
3869
3870 ret = LTTNG_OK;
3871 break;
3872 }
3873 case LTTNG_LIST_SESSIONS:
3874 {
3875 unsigned int nr_sessions;
3876 void *sessions_payload;
3877 size_t payload_len;
3878
3879 session_lock_list();
3880 nr_sessions = lttng_sessions_count(
3881 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3882 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3883 payload_len = sizeof(struct lttng_session) * nr_sessions;
3884 sessions_payload = zmalloc(payload_len);
3885
3886 if (!sessions_payload) {
3887 session_unlock_list();
3888 ret = -ENOMEM;
3889 goto setup_error;
3890 }
3891
3892 cmd_list_lttng_sessions(sessions_payload,
3893 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3894 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3895 session_unlock_list();
3896
3897 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3898 payload_len);
3899 free(sessions_payload);
3900
3901 if (ret < 0) {
3902 goto setup_error;
3903 }
3904
3905 ret = LTTNG_OK;
3906 break;
3907 }
3908 case LTTNG_REGISTER_CONSUMER:
3909 {
3910 struct consumer_data *cdata;
3911
3912 switch (cmd_ctx->lsm->domain.type) {
3913 case LTTNG_DOMAIN_KERNEL:
3914 cdata = &kconsumer_data;
3915 break;
3916 default:
3917 ret = LTTNG_ERR_UND;
3918 goto error;
3919 }
3920
3921 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3922 cmd_ctx->lsm->u.reg.path, cdata);
3923 break;
3924 }
3925 case LTTNG_DATA_PENDING:
3926 {
3927 int pending_ret;
3928 uint8_t pending_ret_byte;
3929
3930 pending_ret = cmd_data_pending(cmd_ctx->session);
3931
3932 /*
3933 * FIXME
3934 *
3935 * This function may returns 0 or 1 to indicate whether or not
3936 * there is data pending. In case of error, it should return an
3937 * LTTNG_ERR code. However, some code paths may still return
3938 * a nondescript error code, which we handle by returning an
3939 * "unknown" error.
3940 */
3941 if (pending_ret == 0 || pending_ret == 1) {
3942 /*
3943 * ret will be set to LTTNG_OK at the end of
3944 * this function.
3945 */
3946 } else if (pending_ret < 0) {
3947 ret = LTTNG_ERR_UNK;
3948 goto setup_error;
3949 } else {
3950 ret = pending_ret;
3951 goto setup_error;
3952 }
3953
3954 pending_ret_byte = (uint8_t) pending_ret;
3955
3956 /* 1 byte to return whether or not data is pending */
3957 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3958 &pending_ret_byte, 1);
3959
3960 if (ret < 0) {
3961 goto setup_error;
3962 }
3963
3964 ret = LTTNG_OK;
3965 break;
3966 }
3967 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3968 {
3969 struct lttcomm_lttng_output_id reply;
3970
3971 ret = cmd_snapshot_add_output(cmd_ctx->session,
3972 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3973 if (ret != LTTNG_OK) {
3974 goto error;
3975 }
3976
3977 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
3978 sizeof(reply));
3979 if (ret < 0) {
3980 goto setup_error;
3981 }
3982
3983 /* Copy output list into message payload */
3984 ret = LTTNG_OK;
3985 break;
3986 }
3987 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3988 {
3989 ret = cmd_snapshot_del_output(cmd_ctx->session,
3990 &cmd_ctx->lsm->u.snapshot_output.output);
3991 break;
3992 }
3993 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3994 {
3995 ssize_t nb_output;
3996 struct lttng_snapshot_output *outputs = NULL;
3997
3998 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3999 if (nb_output < 0) {
4000 ret = -nb_output;
4001 goto error;
4002 }
4003
4004 assert((nb_output > 0 && outputs) || nb_output == 0);
4005 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
4006 nb_output * sizeof(struct lttng_snapshot_output));
4007 free(outputs);
4008
4009 if (ret < 0) {
4010 goto setup_error;
4011 }
4012
4013 ret = LTTNG_OK;
4014 break;
4015 }
4016 case LTTNG_SNAPSHOT_RECORD:
4017 {
4018 ret = cmd_snapshot_record(cmd_ctx->session,
4019 &cmd_ctx->lsm->u.snapshot_record.output,
4020 cmd_ctx->lsm->u.snapshot_record.wait);
4021 break;
4022 }
4023 case LTTNG_CREATE_SESSION_SNAPSHOT:
4024 {
4025 size_t nb_uri, len;
4026 struct lttng_uri *uris = NULL;
4027
4028 nb_uri = cmd_ctx->lsm->u.uri.size;
4029 len = nb_uri * sizeof(struct lttng_uri);
4030
4031 if (nb_uri > 0) {
4032 uris = zmalloc(len);
4033 if (uris == NULL) {
4034 ret = LTTNG_ERR_FATAL;
4035 goto error;
4036 }
4037
4038 /* Receive variable len data */
4039 DBG("Waiting for %zu URIs from client ...", nb_uri);
4040 ret = lttcomm_recv_unix_sock(sock, uris, len);
4041 if (ret <= 0) {
4042 DBG("No URIs received from client... continuing");
4043 *sock_error = 1;
4044 ret = LTTNG_ERR_SESSION_FAIL;
4045 free(uris);
4046 goto error;
4047 }
4048
4049 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4050 DBG("Creating session with ONE network URI is a bad call");
4051 ret = LTTNG_ERR_SESSION_FAIL;
4052 free(uris);
4053 goto error;
4054 }
4055 }
4056
4057 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4058 nb_uri, &cmd_ctx->creds);
4059 free(uris);
4060 break;
4061 }
4062 case LTTNG_CREATE_SESSION_LIVE:
4063 {
4064 size_t nb_uri, len;
4065 struct lttng_uri *uris = NULL;
4066
4067 nb_uri = cmd_ctx->lsm->u.uri.size;
4068 len = nb_uri * sizeof(struct lttng_uri);
4069
4070 if (nb_uri > 0) {
4071 uris = zmalloc(len);
4072 if (uris == NULL) {
4073 ret = LTTNG_ERR_FATAL;
4074 goto error;
4075 }
4076
4077 /* Receive variable len data */
4078 DBG("Waiting for %zu URIs from client ...", nb_uri);
4079 ret = lttcomm_recv_unix_sock(sock, uris, len);
4080 if (ret <= 0) {
4081 DBG("No URIs received from client... continuing");
4082 *sock_error = 1;
4083 ret = LTTNG_ERR_SESSION_FAIL;
4084 free(uris);
4085 goto error;
4086 }
4087
4088 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4089 DBG("Creating session with ONE network URI is a bad call");
4090 ret = LTTNG_ERR_SESSION_FAIL;
4091 free(uris);
4092 goto error;
4093 }
4094 }
4095
4096 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4097 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4098 free(uris);
4099 break;
4100 }
4101 case LTTNG_SAVE_SESSION:
4102 {
4103 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4104 &cmd_ctx->creds);
4105 break;
4106 }
4107 case LTTNG_SET_SESSION_SHM_PATH:
4108 {
4109 ret = cmd_set_session_shm_path(cmd_ctx->session,
4110 cmd_ctx->lsm->u.set_shm_path.shm_path);
4111 break;
4112 }
4113 case LTTNG_REGENERATE_METADATA:
4114 {
4115 ret = cmd_regenerate_metadata(cmd_ctx->session);
4116 break;
4117 }
4118 case LTTNG_REGENERATE_STATEDUMP:
4119 {
4120 ret = cmd_regenerate_statedump(cmd_ctx->session);
4121 break;
4122 }
4123 default:
4124 ret = LTTNG_ERR_UND;
4125 break;
4126 }
4127
4128 error:
4129 if (cmd_ctx->llm == NULL) {
4130 DBG("Missing llm structure. Allocating one.");
4131 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4132 goto setup_error;
4133 }
4134 }
4135 /* Set return code */
4136 cmd_ctx->llm->ret_code = ret;
4137 setup_error:
4138 if (cmd_ctx->session) {
4139 session_unlock(cmd_ctx->session);
4140 }
4141 if (need_tracing_session) {
4142 session_unlock_list();
4143 }
4144 init_setup_error:
4145 assert(!rcu_read_ongoing());
4146 return ret;
4147 }
4148
4149 /*
4150 * Thread managing health check socket.
4151 */
4152 static void *thread_manage_health(void *data)
4153 {
4154 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4155 uint32_t revents, nb_fd;
4156 struct lttng_poll_event events;
4157 struct health_comm_msg msg;
4158 struct health_comm_reply reply;
4159
4160 DBG("[thread] Manage health check started");
4161
4162 rcu_register_thread();
4163
4164 /* We might hit an error path before this is created. */
4165 lttng_poll_init(&events);
4166
4167 /* Create unix socket */
4168 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4169 if (sock < 0) {
4170 ERR("Unable to create health check Unix socket");
4171 goto error;
4172 }
4173
4174 if (is_root) {
4175 /* lttng health client socket path permissions */
4176 ret = chown(health_unix_sock_path, 0,
4177 utils_get_group_id(tracing_group_name));
4178 if (ret < 0) {
4179 ERR("Unable to set group on %s", health_unix_sock_path);
4180 PERROR("chown");
4181 goto error;
4182 }
4183
4184 ret = chmod(health_unix_sock_path,
4185 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4186 if (ret < 0) {
4187 ERR("Unable to set permissions on %s", health_unix_sock_path);
4188 PERROR("chmod");
4189 goto error;
4190 }
4191 }
4192
4193 /*
4194 * Set the CLOEXEC flag. Return code is useless because either way, the
4195 * show must go on.
4196 */
4197 (void) utils_set_fd_cloexec(sock);
4198
4199 ret = lttcomm_listen_unix_sock(sock);
4200 if (ret < 0) {
4201 goto error;
4202 }
4203
4204 /*
4205 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4206 * more will be added to this poll set.
4207 */
4208 ret = sessiond_set_thread_pollset(&events, 2);
4209 if (ret < 0) {
4210 goto error;
4211 }
4212
4213 /* Add the application registration socket */
4214 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4215 if (ret < 0) {
4216 goto error;
4217 }
4218
4219 sessiond_notify_ready();
4220
4221 while (1) {
4222 DBG("Health check ready");
4223
4224 /* Inifinite blocking call, waiting for transmission */
4225 restart:
4226 ret = lttng_poll_wait(&events, -1);
4227 if (ret < 0) {
4228 /*
4229 * Restart interrupted system call.
4230 */
4231 if (errno == EINTR) {
4232 goto restart;
4233 }
4234 goto error;
4235 }
4236
4237 nb_fd = ret;
4238
4239 for (i = 0; i < nb_fd; i++) {
4240 /* Fetch once the poll data */
4241 revents = LTTNG_POLL_GETEV(&events, i);
4242 pollfd = LTTNG_POLL_GETFD(&events, i);
4243
4244 if (!revents) {
4245 /* No activity for this FD (poll implementation). */
4246 continue;
4247 }
4248
4249 /* Thread quit pipe has been closed. Killing thread. */
4250 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4251 if (ret) {
4252 err = 0;
4253 goto exit;
4254 }
4255
4256 /* Event on the registration socket */
4257 if (pollfd == sock) {
4258 if (revents & LPOLLIN) {
4259 continue;
4260 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4261 ERR("Health socket poll error");
4262 goto error;
4263 } else {
4264 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4265 goto error;
4266 }
4267 }
4268 }
4269
4270 new_sock = lttcomm_accept_unix_sock(sock);
4271 if (new_sock < 0) {
4272 goto error;
4273 }
4274
4275 /*
4276 * Set the CLOEXEC flag. Return code is useless because either way, the
4277 * show must go on.
4278 */
4279 (void) utils_set_fd_cloexec(new_sock);
4280
4281 DBG("Receiving data from client for health...");
4282 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4283 if (ret <= 0) {
4284 DBG("Nothing recv() from client... continuing");
4285 ret = close(new_sock);
4286 if (ret) {
4287 PERROR("close");
4288 }
4289 continue;
4290 }
4291
4292 rcu_thread_online();
4293
4294 memset(&reply, 0, sizeof(reply));
4295 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4296 /*
4297 * health_check_state returns 0 if health is
4298 * bad.
4299 */
4300 if (!health_check_state(health_sessiond, i)) {
4301 reply.ret_code |= 1ULL << i;
4302 }
4303 }
4304
4305 DBG2("Health check return value %" PRIx64, reply.ret_code);
4306
4307 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4308 if (ret < 0) {
4309 ERR("Failed to send health data back to client");
4310 }
4311
4312 /* End of transmission */
4313 ret = close(new_sock);
4314 if (ret) {
4315 PERROR("close");
4316 }
4317 }
4318
4319 exit:
4320 error:
4321 if (err) {
4322 ERR("Health error occurred in %s", __func__);
4323 }
4324 DBG("Health check thread dying");
4325 unlink(health_unix_sock_path);
4326 if (sock >= 0) {
4327 ret = close(sock);
4328 if (ret) {
4329 PERROR("close");
4330 }
4331 }
4332
4333 lttng_poll_clean(&events);
4334 stop_threads();
4335 rcu_unregister_thread();
4336 return NULL;
4337 }
4338
4339 /*
4340 * This thread manage all clients request using the unix client socket for
4341 * communication.
4342 */
4343 static void *thread_manage_clients(void *data)
4344 {
4345 int sock = -1, ret, i, pollfd, err = -1;
4346 int sock_error;
4347 uint32_t revents, nb_fd;
4348 struct command_ctx *cmd_ctx = NULL;
4349 struct lttng_poll_event events;
4350
4351 DBG("[thread] Manage client started");
4352
4353 rcu_register_thread();
4354
4355 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4356
4357 health_code_update();
4358
4359 ret = lttcomm_listen_unix_sock(client_sock);
4360 if (ret < 0) {
4361 goto error_listen;
4362 }
4363
4364 /*
4365 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4366 * more will be added to this poll set.
4367 */
4368 ret = sessiond_set_thread_pollset(&events, 2);
4369 if (ret < 0) {
4370 goto error_create_poll;
4371 }
4372
4373 /* Add the application registration socket */
4374 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4375 if (ret < 0) {
4376 goto error;
4377 }
4378
4379 sessiond_notify_ready();
4380
4381 ret = sem_post(&load_info->message_thread_ready);
4382 if (ret) {
4383 PERROR("sem_post message_thread_ready");
4384 goto error;
4385 }
4386
4387 /*
4388 * Wait until all support threads are initialized before accepting
4389 * commands.
4390 */
4391 while (uatomic_read(&lttng_sessiond_ready) != 0) {
4392 fd_set read_fds;
4393 struct timeval timeout;
4394
4395 FD_ZERO(&read_fds);
4396 FD_SET(thread_quit_pipe[0], &read_fds);
4397 memset(&timeout, 0, sizeof(timeout));
4398 timeout.tv_usec = 1000;
4399
4400 /*
4401 * If a support thread failed to launch, it may signal that
4402 * we must exit and the sessiond would never be marked as
4403 * "ready".
4404 *
4405 * The timeout is set to 1ms, which serves as a way to
4406 * pace down this check.
4407 */
4408 ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
4409 &timeout);
4410 if (ret > 0 || (ret < 0 && errno != EINTR)) {
4411 goto exit;
4412 }
4413 }
4414
4415 /* This testpoint is after we signal readiness to the parent. */
4416 if (testpoint(sessiond_thread_manage_clients)) {
4417 goto error;
4418 }
4419
4420 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4421 goto error;
4422 }
4423
4424 health_code_update();
4425
4426 while (1) {
4427 DBG("Accepting client command ...");
4428
4429 /* Inifinite blocking call, waiting for transmission */
4430 restart:
4431 health_poll_entry();
4432 ret = lttng_poll_wait(&events, -1);
4433 health_poll_exit();
4434 if (ret < 0) {
4435 /*
4436 * Restart interrupted system call.
4437 */
4438 if (errno == EINTR) {
4439 goto restart;
4440 }
4441 goto error;
4442 }
4443
4444 nb_fd = ret;
4445
4446 for (i = 0; i < nb_fd; i++) {
4447 /* Fetch once the poll data */
4448 revents = LTTNG_POLL_GETEV(&events, i);
4449 pollfd = LTTNG_POLL_GETFD(&events, i);
4450
4451 health_code_update();
4452
4453 if (!revents) {
4454 /* No activity for this FD (poll implementation). */
4455 continue;
4456 }
4457
4458 /* Thread quit pipe has been closed. Killing thread. */
4459 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4460 if (ret) {
4461 err = 0;
4462 goto exit;
4463 }
4464
4465 /* Event on the registration socket */
4466 if (pollfd == client_sock) {
4467 if (revents & LPOLLIN) {
4468 continue;
4469 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4470 ERR("Client socket poll error");
4471 goto error;
4472 } else {
4473 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4474 goto error;
4475 }
4476 }
4477 }
4478
4479 DBG("Wait for client response");
4480
4481 health_code_update();
4482
4483 sock = lttcomm_accept_unix_sock(client_sock);
4484 if (sock < 0) {
4485 goto error;
4486 }
4487
4488 /*
4489 * Set the CLOEXEC flag. Return code is useless because either way, the
4490 * show must go on.
4491 */
4492 (void) utils_set_fd_cloexec(sock);
4493
4494 /* Set socket option for credentials retrieval */
4495 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4496 if (ret < 0) {
4497 goto error;
4498 }
4499
4500 /* Allocate context command to process the client request */
4501 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4502 if (cmd_ctx == NULL) {
4503 PERROR("zmalloc cmd_ctx");
4504 goto error;
4505 }
4506
4507 /* Allocate data buffer for reception */
4508 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4509 if (cmd_ctx->lsm == NULL) {
4510 PERROR("zmalloc cmd_ctx->lsm");
4511 goto error;
4512 }
4513
4514 cmd_ctx->llm = NULL;
4515 cmd_ctx->session = NULL;
4516
4517 health_code_update();
4518
4519 /*
4520 * Data is received from the lttng client. The struct
4521 * lttcomm_session_msg (lsm) contains the command and data request of
4522 * the client.
4523 */
4524 DBG("Receiving data from client ...");
4525 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4526 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4527 if (ret <= 0) {
4528 DBG("Nothing recv() from client... continuing");
4529 ret = close(sock);
4530 if (ret) {
4531 PERROR("close");
4532 }
4533 sock = -1;
4534 clean_command_ctx(&cmd_ctx);
4535 continue;
4536 }
4537
4538 health_code_update();
4539
4540 // TODO: Validate cmd_ctx including sanity check for
4541 // security purpose.
4542
4543 rcu_thread_online();
4544 /*
4545 * This function dispatch the work to the kernel or userspace tracer
4546 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4547 * informations for the client. The command context struct contains
4548 * everything this function may needs.
4549 */
4550 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4551 rcu_thread_offline();
4552 if (ret < 0) {
4553 ret = close(sock);
4554 if (ret) {
4555 PERROR("close");
4556 }
4557 sock = -1;
4558 /*
4559 * TODO: Inform client somehow of the fatal error. At
4560 * this point, ret < 0 means that a zmalloc failed
4561 * (ENOMEM). Error detected but still accept
4562 * command, unless a socket error has been
4563 * detected.
4564 */
4565 clean_command_ctx(&cmd_ctx);
4566 continue;
4567 }
4568
4569 health_code_update();
4570
4571 DBG("Sending response (size: %d, retcode: %s (%d))",
4572 cmd_ctx->lttng_msg_size,
4573 lttng_strerror(-cmd_ctx->llm->ret_code),
4574 cmd_ctx->llm->ret_code);
4575 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4576 if (ret < 0) {
4577 ERR("Failed to send data back to client");
4578 }
4579
4580 /* End of transmission */
4581 ret = close(sock);
4582 if (ret) {
4583 PERROR("close");
4584 }
4585 sock = -1;
4586
4587 clean_command_ctx(&cmd_ctx);
4588
4589 health_code_update();
4590 }
4591
4592 exit:
4593 error:
4594 if (sock >= 0) {
4595 ret = close(sock);
4596 if (ret) {
4597 PERROR("close");
4598 }
4599 }
4600
4601 lttng_poll_clean(&events);
4602 clean_command_ctx(&cmd_ctx);
4603
4604 error_listen:
4605 error_create_poll:
4606 unlink(client_unix_sock_path);
4607 if (client_sock >= 0) {
4608 ret = close(client_sock);
4609 if (ret) {
4610 PERROR("close");
4611 }
4612 }
4613
4614 if (err) {
4615 health_error();
4616 ERR("Health error occurred in %s", __func__);
4617 }
4618
4619 health_unregister(health_sessiond);
4620
4621 DBG("Client thread dying");
4622
4623 rcu_unregister_thread();
4624
4625 /*
4626 * Since we are creating the consumer threads, we own them, so we need
4627 * to join them before our thread exits.
4628 */
4629 ret = join_consumer_thread(&kconsumer_data);
4630 if (ret) {
4631 errno = ret;
4632 PERROR("join_consumer");
4633 }
4634
4635 ret = join_consumer_thread(&ustconsumer32_data);
4636 if (ret) {
4637 errno = ret;
4638 PERROR("join_consumer ust32");
4639 }
4640
4641 ret = join_consumer_thread(&ustconsumer64_data);
4642 if (ret) {
4643 errno = ret;
4644 PERROR("join_consumer ust64");
4645 }
4646 return NULL;
4647 }
4648
4649 static int string_match(const char *str1, const char *str2)
4650 {
4651 return (str1 && str2) && !strcmp(str1, str2);
4652 }
4653
4654 /*
4655 * Take an option from the getopt output and set it in the right variable to be
4656 * used later.
4657 *
4658 * Return 0 on success else a negative value.
4659 */
4660 static int set_option(int opt, const char *arg, const char *optname)
4661 {
4662 int ret = 0;
4663
4664 if (string_match(optname, "client-sock") || opt == 'c') {
4665 if (!arg || *arg == '\0') {
4666 ret = -EINVAL;
4667 goto end;
4668 }
4669 if (lttng_is_setuid_setgid()) {
4670 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4671 "-c, --client-sock");
4672 } else {
4673 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4674 }
4675 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4676 if (!arg || *arg == '\0') {
4677 ret = -EINVAL;
4678 goto end;
4679 }
4680 if (lttng_is_setuid_setgid()) {
4681 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4682 "-a, --apps-sock");
4683 } else {
4684 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4685 }
4686 } else if (string_match(optname, "daemonize") || opt == 'd') {
4687 opt_daemon = 1;
4688 } else if (string_match(optname, "background") || opt == 'b') {
4689 opt_background = 1;
4690 } else if (string_match(optname, "group") || opt == 'g') {
4691 if (!arg || *arg == '\0') {
4692 ret = -EINVAL;
4693 goto end;
4694 }
4695 if (lttng_is_setuid_setgid()) {
4696 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4697 "-g, --group");
4698 } else {
4699 /*
4700 * If the override option is set, the pointer points to a
4701 * *non* const thus freeing it even though the variable type is
4702 * set to const.
4703 */
4704 if (tracing_group_name_override) {
4705 free((void *) tracing_group_name);
4706 }
4707 tracing_group_name = strdup(arg);
4708 if (!tracing_group_name) {
4709 PERROR("strdup");
4710 ret = -ENOMEM;
4711 }
4712 tracing_group_name_override = 1;
4713 }
4714 } else if (string_match(optname, "help") || opt == 'h') {
4715 ret = utils_show_man_page(8, "lttng-sessiond");
4716 if (ret) {
4717 ERR("Cannot view man page lttng-sessiond(8)");
4718 perror("exec");
4719 }
4720 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4721 } else if (string_match(optname, "version") || opt == 'V') {
4722 opt_print_version = 1;
4723 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4724 opt_sig_parent = 1;
4725 } else if (string_match(optname, "kconsumerd-err-sock")) {
4726 if (!arg || *arg == '\0') {
4727 ret = -EINVAL;
4728 goto end;
4729 }
4730 if (lttng_is_setuid_setgid()) {
4731 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4732 "--kconsumerd-err-sock");
4733 } else {
4734 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4735 }
4736 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4737 if (!arg || *arg == '\0') {
4738 ret = -EINVAL;
4739 goto end;
4740 }
4741 if (lttng_is_setuid_setgid()) {
4742 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4743 "--kconsumerd-cmd-sock");
4744 } else {
4745 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4746 }
4747 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4748 if (!arg || *arg == '\0') {
4749 ret = -EINVAL;
4750 goto end;
4751 }
4752 if (lttng_is_setuid_setgid()) {
4753 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4754 "--ustconsumerd64-err-sock");
4755 } else {
4756 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4757 }
4758 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4759 if (!arg || *arg == '\0') {
4760 ret = -EINVAL;
4761 goto end;
4762 }
4763 if (lttng_is_setuid_setgid()) {
4764 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4765 "--ustconsumerd64-cmd-sock");
4766 } else {
4767 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4768 }
4769 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4770 if (!arg || *arg == '\0') {
4771 ret = -EINVAL;
4772 goto end;
4773 }
4774 if (lttng_is_setuid_setgid()) {
4775 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4776 "--ustconsumerd32-err-sock");
4777 } else {
4778 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4779 }
4780 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4781 if (!arg || *arg == '\0') {
4782 ret = -EINVAL;
4783 goto end;
4784 }
4785 if (lttng_is_setuid_setgid()) {
4786 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4787 "--ustconsumerd32-cmd-sock");
4788 } else {
4789 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4790 }
4791 } else if (string_match(optname, "no-kernel")) {
4792 opt_no_kernel = 1;
4793 } else if (string_match(optname, "quiet") || opt == 'q') {
4794 lttng_opt_quiet = 1;
4795 } else if (string_match(optname, "verbose") || opt == 'v') {
4796 /* Verbose level can increase using multiple -v */
4797 if (arg) {
4798 /* Value obtained from config file */
4799 lttng_opt_verbose = config_parse_value(arg);
4800 } else {
4801 /* -v used on command line */
4802 lttng_opt_verbose++;
4803 }
4804 /* Clamp value to [0, 3] */
4805 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4806 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4807 } else if (string_match(optname, "verbose-consumer")) {
4808 if (arg) {
4809 opt_verbose_consumer = config_parse_value(arg);
4810 } else {
4811 opt_verbose_consumer++;
4812 }
4813 } else if (string_match(optname, "consumerd32-path")) {
4814 if (!arg || *arg == '\0') {
4815 ret = -EINVAL;
4816 goto end;
4817 }
4818 if (lttng_is_setuid_setgid()) {
4819 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4820 "--consumerd32-path");
4821 } else {
4822 if (consumerd32_bin_override) {
4823 free((void *) consumerd32_bin);
4824 }
4825 consumerd32_bin = strdup(arg);
4826 if (!consumerd32_bin) {
4827 PERROR("strdup");
4828 ret = -ENOMEM;
4829 }
4830 consumerd32_bin_override = 1;
4831 }
4832 } else if (string_match(optname, "consumerd32-libdir")) {
4833 if (!arg || *arg == '\0') {
4834 ret = -EINVAL;
4835 goto end;
4836 }
4837 if (lttng_is_setuid_setgid()) {
4838 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4839 "--consumerd32-libdir");
4840 } else {
4841 if (consumerd32_libdir_override) {
4842 free((void *) consumerd32_libdir);
4843 }
4844 consumerd32_libdir = strdup(arg);
4845 if (!consumerd32_libdir) {
4846 PERROR("strdup");
4847 ret = -ENOMEM;
4848 }
4849 consumerd32_libdir_override = 1;
4850 }
4851 } else if (string_match(optname, "consumerd64-path")) {
4852 if (!arg || *arg == '\0') {
4853 ret = -EINVAL;
4854 goto end;
4855 }
4856 if (lttng_is_setuid_setgid()) {
4857 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4858 "--consumerd64-path");
4859 } else {
4860 if (consumerd64_bin_override) {
4861 free((void *) consumerd64_bin);
4862 }
4863 consumerd64_bin = strdup(arg);
4864 if (!consumerd64_bin) {
4865 PERROR("strdup");
4866 ret = -ENOMEM;
4867 }
4868 consumerd64_bin_override = 1;
4869 }
4870 } else if (string_match(optname, "consumerd64-libdir")) {
4871 if (!arg || *arg == '\0') {
4872 ret = -EINVAL;
4873 goto end;
4874 }
4875 if (lttng_is_setuid_setgid()) {
4876 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4877 "--consumerd64-libdir");
4878 } else {
4879 if (consumerd64_libdir_override) {
4880 free((void *) consumerd64_libdir);
4881 }
4882 consumerd64_libdir = strdup(arg);
4883 if (!consumerd64_libdir) {
4884 PERROR("strdup");
4885 ret = -ENOMEM;
4886 }
4887 consumerd64_libdir_override = 1;
4888 }
4889 } else if (string_match(optname, "pidfile") || opt == 'p') {
4890 if (!arg || *arg == '\0') {
4891 ret = -EINVAL;
4892 goto end;
4893 }
4894 if (lttng_is_setuid_setgid()) {
4895 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4896 "-p, --pidfile");
4897 } else {
4898 free(opt_pidfile);
4899 opt_pidfile = strdup(arg);
4900 if (!opt_pidfile) {
4901 PERROR("strdup");
4902 ret = -ENOMEM;
4903 }
4904 }
4905 } else if (string_match(optname, "agent-tcp-port")) {
4906 if (!arg || *arg == '\0') {
4907 ret = -EINVAL;
4908 goto end;
4909 }
4910 if (lttng_is_setuid_setgid()) {
4911 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4912 "--agent-tcp-port");
4913 } else {
4914 unsigned long v;
4915
4916 errno = 0;
4917 v = strtoul(arg, NULL, 0);
4918 if (errno != 0 || !isdigit(arg[0])) {
4919 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4920 return -1;
4921 }
4922 if (v == 0 || v >= 65535) {
4923 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4924 return -1;
4925 }
4926 agent_tcp_port = (uint32_t) v;
4927 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4928 }
4929 } else if (string_match(optname, "load") || opt == 'l') {
4930 if (!arg || *arg == '\0') {
4931 ret = -EINVAL;
4932 goto end;
4933 }
4934 if (lttng_is_setuid_setgid()) {
4935 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4936 "-l, --load");
4937 } else {
4938 free(opt_load_session_path);
4939 opt_load_session_path = strdup(arg);
4940 if (!opt_load_session_path) {
4941 PERROR("strdup");
4942 ret = -ENOMEM;
4943 }
4944 }
4945 } else if (string_match(optname, "kmod-probes")) {
4946 if (!arg || *arg == '\0') {
4947 ret = -EINVAL;
4948 goto end;
4949 }
4950 if (lttng_is_setuid_setgid()) {
4951 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4952 "--kmod-probes");
4953 } else {
4954 free(kmod_probes_list);
4955 kmod_probes_list = strdup(arg);
4956 if (!kmod_probes_list) {
4957 PERROR("strdup");
4958 ret = -ENOMEM;
4959 }
4960 }
4961 } else if (string_match(optname, "extra-kmod-probes")) {
4962 if (!arg || *arg == '\0') {
4963 ret = -EINVAL;
4964 goto end;
4965 }
4966 if (lttng_is_setuid_setgid()) {
4967 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4968 "--extra-kmod-probes");
4969 } else {
4970 free(kmod_extra_probes_list);
4971 kmod_extra_probes_list = strdup(arg);
4972 if (!kmod_extra_probes_list) {
4973 PERROR("strdup");
4974 ret = -ENOMEM;
4975 }
4976 }
4977 } else if (string_match(optname, "config") || opt == 'f') {
4978 /* This is handled in set_options() thus silent skip. */
4979 goto end;
4980 } else {
4981 /* Unknown option or other error.
4982 * Error is printed by getopt, just return */
4983 ret = -1;
4984 }
4985
4986 end:
4987 if (ret == -EINVAL) {
4988 const char *opt_name = "unknown";
4989 int i;
4990
4991 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4992 i++) {
4993 if (opt == long_options[i].val) {
4994 opt_name = long_options[i].name;
4995 break;
4996 }
4997 }
4998
4999 WARN("Invalid argument provided for option \"%s\", using default value.",
5000 opt_name);
5001 }
5002
5003 return ret;
5004 }
5005
5006 /*
5007 * config_entry_handler_cb used to handle options read from a config file.
5008 * See config_entry_handler_cb comment in common/config/session-config.h for the
5009 * return value conventions.
5010 */
5011 static int config_entry_handler(const struct config_entry *entry, void *unused)
5012 {
5013 int ret = 0, i;
5014
5015 if (!entry || !entry->name || !entry->value) {
5016 ret = -EINVAL;
5017 goto end;
5018 }
5019
5020 /* Check if the option is to be ignored */
5021 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5022 if (!strcmp(entry->name, config_ignore_options[i])) {
5023 goto end;
5024 }
5025 }
5026
5027 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5028 i++) {
5029
5030 /* Ignore if not fully matched. */
5031 if (strcmp(entry->name, long_options[i].name)) {
5032 continue;
5033 }
5034
5035 /*
5036 * If the option takes no argument on the command line, we have to
5037 * check if the value is "true". We support non-zero numeric values,
5038 * true, on and yes.
5039 */
5040 if (!long_options[i].has_arg) {
5041 ret = config_parse_value(entry->value);
5042 if (ret <= 0) {
5043 if (ret) {
5044 WARN("Invalid configuration value \"%s\" for option %s",
5045 entry->value, entry->name);
5046 }
5047 /* False, skip boolean config option. */
5048 goto end;
5049 }
5050 }
5051
5052 ret = set_option(long_options[i].val, entry->value, entry->name);
5053 goto end;
5054 }
5055
5056 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5057
5058 end:
5059 return ret;
5060 }
5061
5062 static void sessiond_config_log(void)
5063 {
5064 DBG("LTTng-sessiond " VERSION " - " VERSION_NAME "%s%s",
5065 GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
5066 EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
5067 if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
5068 DBG("LTTng-sessiond extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
5069 }
5070 }
5071
5072 static void print_version(void) {
5073 fprintf(stdout, "%s\n", VERSION);
5074 }
5075
5076 /*
5077 * daemon configuration loading and argument parsing
5078 */
5079 static int set_options(int argc, char **argv)
5080 {
5081 int ret = 0, c = 0, option_index = 0;
5082 int orig_optopt = optopt, orig_optind = optind;
5083 char *optstring;
5084 const char *config_path = NULL;
5085
5086 optstring = utils_generate_optstring(long_options,
5087 sizeof(long_options) / sizeof(struct option));
5088 if (!optstring) {
5089 ret = -ENOMEM;
5090 goto end;
5091 }
5092
5093 /* Check for the --config option */
5094 while ((c = getopt_long(argc, argv, optstring, long_options,
5095 &option_index)) != -1) {
5096 if (c == '?') {
5097 ret = -EINVAL;
5098 goto end;
5099 } else if (c != 'f') {
5100 /* if not equal to --config option. */
5101 continue;
5102 }
5103
5104 if (lttng_is_setuid_setgid()) {
5105 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5106 "-f, --config");
5107 } else {
5108 config_path = utils_expand_path(optarg);
5109 if (!config_path) {
5110 ERR("Failed to resolve path: %s", optarg);
5111 }
5112 }
5113 }
5114
5115 ret = config_get_section_entries(config_path, config_section_name,
5116 config_entry_handler, NULL);
5117 if (ret) {
5118 if (ret > 0) {
5119 ERR("Invalid configuration option at line %i", ret);
5120 ret = -1;
5121 }
5122 goto end;
5123 }
5124
5125 /* Reset getopt's global state */
5126 optopt = orig_optopt;
5127 optind = orig_optind;
5128 while (1) {
5129 option_index = -1;
5130 /*
5131 * getopt_long() will not set option_index if it encounters a
5132 * short option.
5133 */
5134 c = getopt_long(argc, argv, optstring, long_options,
5135 &option_index);
5136 if (c == -1) {
5137 break;
5138 }
5139
5140 /*
5141 * Pass NULL as the long option name if popt left the index
5142 * unset.
5143 */
5144 ret = set_option(c, optarg,
5145 option_index < 0 ? NULL :
5146 long_options[option_index].name);
5147 if (ret < 0) {
5148 break;
5149 }
5150 }
5151
5152 end:
5153 free(optstring);
5154 return ret;
5155 }
5156
5157 /*
5158 * Creates the two needed socket by the daemon.
5159 * apps_sock - The communication socket for all UST apps.
5160 * client_sock - The communication of the cli tool (lttng).
5161 */
5162 static int init_daemon_socket(void)
5163 {
5164 int ret = 0;
5165 mode_t old_umask;
5166
5167 old_umask = umask(0);
5168
5169 /* Create client tool unix socket */
5170 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5171 if (client_sock < 0) {
5172 ERR("Create unix sock failed: %s", client_unix_sock_path);
5173 ret = -1;
5174 goto end;
5175 }
5176
5177 /* Set the cloexec flag */
5178 ret = utils_set_fd_cloexec(client_sock);
5179 if (ret < 0) {
5180 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5181 "Continuing but note that the consumer daemon will have a "
5182 "reference to this socket on exec()", client_sock);
5183 }
5184
5185 /* File permission MUST be 660 */
5186 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5187 if (ret < 0) {
5188 ERR("Set file permissions failed: %s", client_unix_sock_path);
5189 PERROR("chmod");
5190 goto end;
5191 }
5192
5193 /* Create the application unix socket */
5194 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5195 if (apps_sock < 0) {
5196 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5197 ret = -1;
5198 goto end;
5199 }
5200
5201 /* Set the cloexec flag */
5202 ret = utils_set_fd_cloexec(apps_sock);
5203 if (ret < 0) {
5204 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5205 "Continuing but note that the consumer daemon will have a "
5206 "reference to this socket on exec()", apps_sock);
5207 }
5208
5209 /* File permission MUST be 666 */
5210 ret = chmod(apps_unix_sock_path,
5211 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5212 if (ret < 0) {
5213 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5214 PERROR("chmod");
5215 goto end;
5216 }
5217
5218 DBG3("Session daemon client socket %d and application socket %d created",
5219 client_sock, apps_sock);
5220
5221 end:
5222 umask(old_umask);
5223 return ret;
5224 }
5225
5226 /*
5227 * Check if the global socket is available, and if a daemon is answering at the
5228 * other side. If yes, error is returned.
5229 */
5230 static int check_existing_daemon(void)
5231 {
5232 /* Is there anybody out there ? */
5233 if (lttng_session_daemon_alive()) {
5234 return -EEXIST;
5235 }
5236
5237 return 0;
5238 }
5239
5240 /*
5241 * Set the tracing group gid onto the client socket.
5242 *
5243 * Race window between mkdir and chown is OK because we are going from more
5244 * permissive (root.root) to less permissive (root.tracing).
5245 */
5246 static int set_permissions(char *rundir)
5247 {
5248 int ret;
5249 gid_t gid;
5250
5251 gid = utils_get_group_id(tracing_group_name);
5252
5253 /* Set lttng run dir */
5254 ret = chown(rundir, 0, gid);
5255 if (ret < 0) {
5256 ERR("Unable to set group on %s", rundir);
5257 PERROR("chown");
5258 }
5259
5260 /*
5261 * Ensure all applications and tracing group can search the run
5262 * dir. Allow everyone to read the directory, since it does not
5263 * buy us anything to hide its content.
5264 */
5265 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5266 if (ret < 0) {
5267 ERR("Unable to set permissions on %s", rundir);
5268 PERROR("chmod");
5269 }
5270
5271 /* lttng client socket path */
5272 ret = chown(client_unix_sock_path, 0, gid);
5273 if (ret < 0) {
5274 ERR("Unable to set group on %s", client_unix_sock_path);
5275 PERROR("chown");
5276 }
5277
5278 /* kconsumer error socket path */
5279 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5280 if (ret < 0) {
5281 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5282 PERROR("chown");
5283 }
5284
5285 /* 64-bit ustconsumer error socket path */
5286 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5287 if (ret < 0) {
5288 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5289 PERROR("chown");
5290 }
5291
5292 /* 32-bit ustconsumer compat32 error socket path */
5293 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5294 if (ret < 0) {
5295 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5296 PERROR("chown");
5297 }
5298
5299 DBG("All permissions are set");
5300
5301 return ret;
5302 }
5303
5304 /*
5305 * Create the lttng run directory needed for all global sockets and pipe.
5306 */
5307 static int create_lttng_rundir(const char *rundir)
5308 {
5309 int ret;
5310
5311 DBG3("Creating LTTng run directory: %s", rundir);
5312
5313 ret = mkdir(rundir, S_IRWXU);
5314 if (ret < 0) {
5315 if (errno != EEXIST) {
5316 ERR("Unable to create %s", rundir);
5317 goto error;
5318 } else {
5319 ret = 0;
5320 }
5321 }
5322
5323 error:
5324 return ret;
5325 }
5326
5327 /*
5328 * Setup sockets and directory needed by the kconsumerd communication with the
5329 * session daemon.
5330 */
5331 static int set_consumer_sockets(struct consumer_data *consumer_data,
5332 const char *rundir)
5333 {
5334 int ret;
5335 char path[PATH_MAX];
5336
5337 switch (consumer_data->type) {
5338 case LTTNG_CONSUMER_KERNEL:
5339 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5340 break;
5341 case LTTNG_CONSUMER64_UST:
5342 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5343 break;
5344 case LTTNG_CONSUMER32_UST:
5345 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5346 break;
5347 default:
5348 ERR("Consumer type unknown");
5349 ret = -EINVAL;
5350 goto error;
5351 }
5352
5353 DBG2("Creating consumer directory: %s", path);
5354
5355 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5356 if (ret < 0) {
5357 if (errno != EEXIST) {
5358 PERROR("mkdir");
5359 ERR("Failed to create %s", path);
5360 goto error;
5361 }
5362 ret = -1;
5363 }
5364 if (is_root) {
5365 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5366 if (ret < 0) {
5367 ERR("Unable to set group on %s", path);
5368 PERROR("chown");
5369 goto error;
5370 }
5371 }
5372
5373 /* Create the kconsumerd error unix socket */
5374 consumer_data->err_sock =
5375 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5376 if (consumer_data->err_sock < 0) {
5377 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5378 ret = -1;
5379 goto error;
5380 }
5381
5382 /*
5383 * Set the CLOEXEC flag. Return code is useless because either way, the
5384 * show must go on.
5385 */
5386 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5387 if (ret < 0) {
5388 PERROR("utils_set_fd_cloexec");
5389 /* continue anyway */
5390 }
5391
5392 /* File permission MUST be 660 */
5393 ret = chmod(consumer_data->err_unix_sock_path,
5394 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5395 if (ret < 0) {
5396 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5397 PERROR("chmod");
5398 goto error;
5399 }
5400
5401 error:
5402 return ret;
5403 }
5404
5405 /*
5406 * Signal handler for the daemon
5407 *
5408 * Simply stop all worker threads, leaving main() return gracefully after
5409 * joining all threads and calling cleanup().
5410 */
5411 static void sighandler(int sig)
5412 {
5413 switch (sig) {
5414 case SIGINT:
5415 DBG("SIGINT caught");
5416 stop_threads();
5417 break;
5418 case SIGTERM:
5419 DBG("SIGTERM caught");
5420 stop_threads();
5421 break;
5422 case SIGUSR1:
5423 CMM_STORE_SHARED(recv_child_signal, 1);
5424 break;
5425 default:
5426 break;
5427 }
5428 }
5429
5430 /*
5431 * Setup signal handler for :
5432 * SIGINT, SIGTERM, SIGPIPE
5433 */
5434 static int set_signal_handler(void)
5435 {
5436 int ret = 0;
5437 struct sigaction sa;
5438 sigset_t sigset;
5439
5440 if ((ret = sigemptyset(&sigset)) < 0) {
5441 PERROR("sigemptyset");
5442 return ret;
5443 }
5444
5445 sa.sa_mask = sigset;
5446 sa.sa_flags = 0;
5447
5448 sa.sa_handler = sighandler;
5449 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5450 PERROR("sigaction");
5451 return ret;
5452 }
5453
5454 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5455 PERROR("sigaction");
5456 return ret;
5457 }
5458
5459 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5460 PERROR("sigaction");
5461 return ret;
5462 }
5463
5464 sa.sa_handler = SIG_IGN;
5465 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5466 PERROR("sigaction");
5467 return ret;
5468 }
5469
5470 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5471
5472 return ret;
5473 }
5474
5475 /*
5476 * Set open files limit to unlimited. This daemon can open a large number of
5477 * file descriptors in order to consume multiple kernel traces.
5478 */
5479 static void set_ulimit(void)
5480 {
5481 int ret;
5482 struct rlimit lim;
5483
5484 /* The kernel does not allow an infinite limit for open files */
5485 lim.rlim_cur = 65535;
5486 lim.rlim_max = 65535;
5487
5488 ret = setrlimit(RLIMIT_NOFILE, &lim);
5489 if (ret < 0) {
5490 PERROR("failed to set open files limit");
5491 }
5492 }
5493
5494 /*
5495 * Write pidfile using the rundir and opt_pidfile.
5496 */
5497 static int write_pidfile(void)
5498 {
5499 int ret;
5500 char pidfile_path[PATH_MAX];
5501
5502 assert(rundir);
5503
5504 if (opt_pidfile) {
5505 if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) {
5506 ret = -1;
5507 goto error;
5508 }
5509 } else {
5510 /* Build pidfile path from rundir and opt_pidfile. */
5511 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5512 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5513 if (ret < 0) {
5514 PERROR("snprintf pidfile path");
5515 goto error;
5516 }
5517 }
5518
5519 /*
5520 * Create pid file in rundir.
5521 */
5522 ret = utils_create_pid_file(getpid(), pidfile_path);
5523 error:
5524 return ret;
5525 }
5526
5527 /*
5528 * Create lockfile using the rundir and return its fd.
5529 */
5530 static int create_lockfile(void)
5531 {
5532 int ret;
5533 char lockfile_path[PATH_MAX];
5534
5535 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5536 if (ret < 0) {
5537 goto error;
5538 }
5539
5540 ret = utils_create_lock_file(lockfile_path);
5541 error:
5542 return ret;
5543 }
5544
5545 /*
5546 * Write agent TCP port using the rundir.
5547 */
5548 static int write_agent_port(void)
5549 {
5550 int ret;
5551 char path[PATH_MAX];
5552
5553 assert(rundir);
5554
5555 ret = snprintf(path, sizeof(path), "%s/"
5556 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5557 if (ret < 0) {
5558 PERROR("snprintf agent port path");
5559 goto error;
5560 }
5561
5562 /*
5563 * Create TCP agent port file in rundir.
5564 */
5565 ret = utils_create_pid_file(agent_tcp_port, path);
5566
5567 error:
5568 return ret;
5569 }
5570
5571 static int set_clock_plugin_env(void)
5572 {
5573 int ret = 0;
5574 const char *original_env_value;
5575 char *full_path = NULL;
5576 char *new_env_value = NULL;
5577
5578 original_env_value = getenv("LTTNG_UST_CLOCK_PLUGIN");
5579 if (!original_env_value) {
5580 goto end;
5581 }
5582
5583 full_path = utils_expand_path(original_env_value);
5584 if (!full_path) {
5585 ERR("Failed to expand LTTNG_UST_CLOCK_PLUGIN path \"%s\"",
5586 original_env_value);
5587 ret = -1;
5588 goto end;
5589 }
5590 ret = asprintf(&new_env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5591 full_path);
5592 free(full_path);
5593 if (ret < 0) {
5594 PERROR("asprintf");
5595 goto end;
5596 }
5597
5598 DBG("Updating environment: %s", new_env_value);
5599 ret = putenv(new_env_value);
5600 if (ret) {
5601 free(new_env_value);
5602 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5603 goto end;
5604 }
5605 end:
5606 return ret;
5607 }
5608
5609 /*
5610 * main
5611 */
5612 int main(int argc, char **argv)
5613 {
5614 int ret = 0, retval = 0;
5615 void *status;
5616 const char *home_path, *env_app_timeout;
5617
5618 init_kernel_workarounds();
5619
5620 rcu_register_thread();
5621
5622 if (set_signal_handler()) {
5623 retval = -1;
5624 goto exit_set_signal_handler;
5625 }
5626
5627 setup_consumerd_path();
5628
5629 page_size = sysconf(_SC_PAGESIZE);
5630 if (page_size < 0) {
5631 PERROR("sysconf _SC_PAGESIZE");
5632 page_size = LONG_MAX;
5633 WARN("Fallback page size to %ld", page_size);
5634 }
5635
5636 /*
5637 * Parse arguments and load the daemon configuration file.
5638 *
5639 * We have an exit_options exit path to free memory reserved by
5640 * set_options. This is needed because the rest of sessiond_cleanup()
5641 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5642 * depends on set_options.
5643 */
5644 progname = argv[0];
5645 if (set_options(argc, argv)) {
5646 retval = -1;
5647 goto exit_options;
5648 }
5649
5650 sessiond_config_log();
5651
5652 if (opt_print_version) {
5653 print_version();
5654 retval = 0;
5655 goto exit_options;
5656 }
5657
5658 ret = set_clock_plugin_env();
5659 if (ret) {
5660 retval = -1;
5661 goto exit_options;
5662 }
5663
5664 /* Daemonize */
5665 if (opt_daemon || opt_background) {
5666 int i;
5667
5668 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5669 !opt_background);
5670 if (ret < 0) {
5671 retval = -1;
5672 goto exit_options;
5673 }
5674
5675 /*
5676 * We are in the child. Make sure all other file descriptors are
5677 * closed, in case we are called with more opened file
5678 * descriptors than the standard ones.
5679 */
5680 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5681 (void) close(i);
5682 }
5683 }
5684
5685 if (run_as_create_worker(argv[0]) < 0) {
5686 goto exit_create_run_as_worker_cleanup;
5687 }
5688
5689 /*
5690 * Starting from here, we can create threads. This needs to be after
5691 * lttng_daemonize due to RCU.
5692 */
5693
5694 /*
5695 * Initialize the health check subsystem. This call should set the
5696 * appropriate time values.
5697 */
5698 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5699 if (!health_sessiond) {
5700 PERROR("health_app_create error");
5701 retval = -1;
5702 goto exit_health_sessiond_cleanup;
5703 }
5704
5705 /* Create thread to clean up RCU hash tables */
5706 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5707 retval = -1;
5708 goto exit_ht_cleanup;
5709 }
5710
5711 /* Create thread quit pipe */
5712 if (init_thread_quit_pipe()) {
5713 retval = -1;
5714 goto exit_init_data;
5715 }
5716
5717 /* Check if daemon is UID = 0 */
5718 is_root = !getuid();
5719
5720 if (is_root) {
5721 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5722 if (!rundir) {
5723 retval = -1;
5724 goto exit_init_data;
5725 }
5726
5727 /* Create global run dir with root access */
5728 if (create_lttng_rundir(rundir)) {
5729 retval = -1;
5730 goto exit_init_data;
5731 }
5732
5733 if (strlen(apps_unix_sock_path) == 0) {
5734 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5735 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5736 if (ret < 0) {
5737 retval = -1;
5738 goto exit_init_data;
5739 }
5740 }
5741
5742 if (strlen(client_unix_sock_path) == 0) {
5743 ret = snprintf(client_unix_sock_path, PATH_MAX,
5744 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5745 if (ret < 0) {
5746 retval = -1;
5747 goto exit_init_data;
5748 }
5749 }
5750
5751 /* Set global SHM for ust */
5752 if (strlen(wait_shm_path) == 0) {
5753 ret = snprintf(wait_shm_path, PATH_MAX,
5754 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5755 if (ret < 0) {
5756 retval = -1;
5757 goto exit_init_data;
5758 }
5759 }
5760
5761 if (strlen(health_unix_sock_path) == 0) {
5762 ret = snprintf(health_unix_sock_path,
5763 sizeof(health_unix_sock_path),
5764 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5765 if (ret < 0) {
5766 retval = -1;
5767 goto exit_init_data;
5768 }
5769 }
5770
5771 /* Setup kernel consumerd path */
5772 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5773 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5774 if (ret < 0) {
5775 retval = -1;
5776 goto exit_init_data;
5777 }
5778 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5779 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5780 if (ret < 0) {
5781 retval = -1;
5782 goto exit_init_data;
5783 }
5784
5785 DBG2("Kernel consumer err path: %s",
5786 kconsumer_data.err_unix_sock_path);
5787 DBG2("Kernel consumer cmd path: %s",
5788 kconsumer_data.cmd_unix_sock_path);
5789 } else {
5790 home_path = utils_get_home_dir();
5791 if (home_path == NULL) {
5792 /* TODO: Add --socket PATH option */
5793 ERR("Can't get HOME directory for sockets creation.");
5794 retval = -1;
5795 goto exit_init_data;
5796 }
5797
5798 /*
5799 * Create rundir from home path. This will create something like
5800 * $HOME/.lttng
5801 */
5802 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5803 if (ret < 0) {
5804 retval = -1;
5805 goto exit_init_data;
5806 }
5807
5808 if (create_lttng_rundir(rundir)) {
5809 retval = -1;
5810 goto exit_init_data;
5811 }
5812
5813 if (strlen(apps_unix_sock_path) == 0) {
5814 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5815 DEFAULT_HOME_APPS_UNIX_SOCK,
5816 home_path);
5817 if (ret < 0) {
5818 retval = -1;
5819 goto exit_init_data;
5820 }
5821 }
5822
5823 /* Set the cli tool unix socket path */
5824 if (strlen(client_unix_sock_path) == 0) {
5825 ret = snprintf(client_unix_sock_path, PATH_MAX,
5826 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5827 home_path);
5828 if (ret < 0) {
5829 retval = -1;
5830 goto exit_init_data;
5831 }
5832 }
5833
5834 /* Set global SHM for ust */
5835 if (strlen(wait_shm_path) == 0) {
5836 ret = snprintf(wait_shm_path, PATH_MAX,
5837 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5838 getuid());
5839 if (ret < 0) {
5840 retval = -1;
5841 goto exit_init_data;
5842 }
5843 }
5844
5845 /* Set health check Unix path */
5846 if (strlen(health_unix_sock_path) == 0) {
5847 ret = snprintf(health_unix_sock_path,
5848 sizeof(health_unix_sock_path),
5849 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5850 home_path);
5851 if (ret < 0) {
5852 retval = -1;
5853 goto exit_init_data;
5854 }
5855 }
5856 }
5857
5858 lockfile_fd = create_lockfile();
5859 if (lockfile_fd < 0) {
5860 retval = -1;
5861 goto exit_init_data;
5862 }
5863
5864 /* Set consumer initial state */
5865 kernel_consumerd_state = CONSUMER_STOPPED;
5866 ust_consumerd_state = CONSUMER_STOPPED;
5867
5868 DBG("Client socket path %s", client_unix_sock_path);
5869 DBG("Application socket path %s", apps_unix_sock_path);
5870 DBG("Application wait path %s", wait_shm_path);
5871 DBG("LTTng run directory path: %s", rundir);
5872
5873 /* 32 bits consumerd path setup */
5874 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5875 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5876 if (ret < 0) {
5877 PERROR("snprintf 32-bit consumer error socket path");
5878 retval = -1;
5879 goto exit_init_data;
5880 }
5881 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5882 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5883 if (ret < 0) {
5884 PERROR("snprintf 32-bit consumer command socket path");
5885 retval = -1;
5886 goto exit_init_data;
5887 }
5888
5889 DBG2("UST consumer 32 bits err path: %s",
5890 ustconsumer32_data.err_unix_sock_path);
5891 DBG2("UST consumer 32 bits cmd path: %s",
5892 ustconsumer32_data.cmd_unix_sock_path);
5893
5894 /* 64 bits consumerd path setup */
5895 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5896 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5897 if (ret < 0) {
5898 PERROR("snprintf 64-bit consumer error socket path");
5899 retval = -1;
5900 goto exit_init_data;
5901 }
5902 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5903 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5904 if (ret < 0) {
5905 PERROR("snprintf 64-bit consumer command socket path");
5906 retval = -1;
5907 goto exit_init_data;
5908 }
5909
5910 DBG2("UST consumer 64 bits err path: %s",
5911 ustconsumer64_data.err_unix_sock_path);
5912 DBG2("UST consumer 64 bits cmd path: %s",
5913 ustconsumer64_data.cmd_unix_sock_path);
5914
5915 /*
5916 * See if daemon already exist.
5917 */
5918 if (check_existing_daemon()) {
5919 ERR("Already running daemon.\n");
5920 /*
5921 * We do not goto exit because we must not cleanup()
5922 * because a daemon is already running.
5923 */
5924 retval = -1;
5925 goto exit_init_data;
5926 }
5927
5928 /*
5929 * Init UST app hash table. Alloc hash table before this point since
5930 * cleanup() can get called after that point.
5931 */
5932 if (ust_app_ht_alloc()) {
5933 ERR("Failed to allocate UST app hash table");
5934 retval = -1;
5935 goto exit_init_data;
5936 }
5937
5938 /*
5939 * Initialize agent app hash table. We allocate the hash table here
5940 * since cleanup() can get called after this point.
5941 */
5942 if (agent_app_ht_alloc()) {
5943 ERR("Failed to allocate Agent app hash table");
5944 retval = -1;
5945 goto exit_init_data;
5946 }
5947
5948 /*
5949 * These actions must be executed as root. We do that *after* setting up
5950 * the sockets path because we MUST make the check for another daemon using
5951 * those paths *before* trying to set the kernel consumer sockets and init
5952 * kernel tracer.
5953 */
5954 if (is_root) {
5955 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5956 retval = -1;
5957 goto exit_init_data;
5958 }
5959
5960 /* Setup kernel tracer */
5961 if (!opt_no_kernel) {
5962 init_kernel_tracer();
5963 if (kernel_tracer_fd >= 0) {
5964 ret = syscall_init_table();
5965 if (ret < 0) {
5966 ERR("Unable to populate syscall table. "
5967 "Syscall tracing won't work "
5968 "for this session daemon.");
5969 }
5970 }
5971 }
5972
5973 /* Set ulimit for open files */
5974 set_ulimit();
5975 }
5976 /* init lttng_fd tracking must be done after set_ulimit. */
5977 lttng_fd_init();
5978
5979 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5980 retval = -1;
5981 goto exit_init_data;
5982 }
5983
5984 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5985 retval = -1;
5986 goto exit_init_data;
5987 }
5988
5989 /* Setup the needed unix socket */
5990 if (init_daemon_socket()) {
5991 retval = -1;
5992 goto exit_init_data;
5993 }
5994
5995 /* Set credentials to socket */
5996 if (is_root && set_permissions(rundir)) {
5997 retval = -1;
5998 goto exit_init_data;
5999 }
6000
6001 /* Get parent pid if -S, --sig-parent is specified. */
6002 if (opt_sig_parent) {
6003 ppid = getppid();
6004 }
6005
6006 /* Setup the kernel pipe for waking up the kernel thread */
6007 if (is_root && !opt_no_kernel) {
6008 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
6009 retval = -1;
6010 goto exit_init_data;
6011 }
6012 }
6013
6014 /* Setup the thread apps communication pipe. */
6015 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
6016 retval = -1;
6017 goto exit_init_data;
6018 }
6019
6020 /* Setup the thread apps notify communication pipe. */
6021 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
6022 retval = -1;
6023 goto exit_init_data;
6024 }
6025
6026 /* Initialize global buffer per UID and PID registry. */
6027 buffer_reg_init_uid_registry();
6028 buffer_reg_init_pid_registry();
6029
6030 /* Init UST command queue. */
6031 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
6032
6033 /*
6034 * Get session list pointer. This pointer MUST NOT be free'd. This list
6035 * is statically declared in session.c
6036 */
6037 session_list_ptr = session_get_list();
6038
6039 cmd_init();
6040
6041 /* Check for the application socket timeout env variable. */
6042 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
6043 if (env_app_timeout) {
6044 app_socket_timeout = atoi(env_app_timeout);
6045 } else {
6046 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
6047 }
6048
6049 ret = write_pidfile();
6050 if (ret) {
6051 ERR("Error in write_pidfile");
6052 retval = -1;
6053 goto exit_init_data;
6054 }
6055 ret = write_agent_port();
6056 if (ret) {
6057 ERR("Error in write_agent_port");
6058 retval = -1;
6059 goto exit_init_data;
6060 }
6061
6062 /* Initialize communication library */
6063 lttcomm_init();
6064 /* Initialize TCP timeout values */
6065 lttcomm_inet_init();
6066
6067 if (load_session_init_data(&load_info) < 0) {
6068 retval = -1;
6069 goto exit_init_data;
6070 }
6071 load_info->path = opt_load_session_path;
6072
6073 /* Create health-check thread */
6074 ret = pthread_create(&health_thread, default_pthread_attr(),
6075 thread_manage_health, (void *) NULL);
6076 if (ret) {
6077 errno = ret;
6078 PERROR("pthread_create health");
6079 retval = -1;
6080 goto exit_health;
6081 }
6082
6083 /* Create thread to manage the client socket */
6084 ret = pthread_create(&client_thread, default_pthread_attr(),
6085 thread_manage_clients, (void *) NULL);
6086 if (ret) {
6087 errno = ret;
6088 PERROR("pthread_create clients");
6089 retval = -1;
6090 goto exit_client;
6091 }
6092
6093 /* Create thread to dispatch registration */
6094 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
6095 thread_dispatch_ust_registration, (void *) NULL);
6096 if (ret) {
6097 errno = ret;
6098 PERROR("pthread_create dispatch");
6099 retval = -1;
6100 goto exit_dispatch;
6101 }
6102
6103 /* Create thread to manage application registration. */
6104 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
6105 thread_registration_apps, (void *) NULL);
6106 if (ret) {
6107 errno = ret;
6108 PERROR("pthread_create registration");
6109 retval = -1;
6110 goto exit_reg_apps;
6111 }
6112
6113 /* Create thread to manage application socket */
6114 ret = pthread_create(&apps_thread, default_pthread_attr(),
6115 thread_manage_apps, (void *) NULL);
6116 if (ret) {
6117 errno = ret;
6118 PERROR("pthread_create apps");
6119 retval = -1;
6120 goto exit_apps;
6121 }
6122
6123 /* Create thread to manage application notify socket */
6124 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
6125 ust_thread_manage_notify, (void *) NULL);
6126 if (ret) {
6127 errno = ret;
6128 PERROR("pthread_create notify");
6129 retval = -1;
6130 goto exit_apps_notify;
6131 }
6132
6133 /* Create agent registration thread. */
6134 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
6135 agent_thread_manage_registration, (void *) NULL);
6136 if (ret) {
6137 errno = ret;
6138 PERROR("pthread_create agent");
6139 retval = -1;
6140 goto exit_agent_reg;
6141 }
6142
6143 /* Don't start this thread if kernel tracing is not requested nor root */
6144 if (is_root && !opt_no_kernel) {
6145 /* Create kernel thread to manage kernel event */
6146 ret = pthread_create(&kernel_thread, default_pthread_attr(),
6147 thread_manage_kernel, (void *) NULL);
6148 if (ret) {
6149 errno = ret;
6150 PERROR("pthread_create kernel");
6151 retval = -1;
6152 goto exit_kernel;
6153 }
6154 }
6155
6156 /* Create session loading thread. */
6157 ret = pthread_create(&load_session_thread, default_pthread_attr(),
6158 thread_load_session, load_info);
6159 if (ret) {
6160 errno = ret;
6161 PERROR("pthread_create load_session_thread");
6162 retval = -1;
6163 goto exit_load_session;
6164 }
6165
6166 /*
6167 * This is where we start awaiting program completion (e.g. through
6168 * signal that asks threads to teardown).
6169 */
6170
6171 ret = pthread_join(load_session_thread, &status);
6172 if (ret) {
6173 errno = ret;
6174 PERROR("pthread_join load_session_thread");
6175 retval = -1;
6176 }
6177 exit_load_session:
6178
6179 if (is_root && !opt_no_kernel) {
6180 ret = pthread_join(kernel_thread, &status);
6181 if (ret) {
6182 errno = ret;
6183 PERROR("pthread_join");
6184 retval = -1;
6185 }
6186 }
6187 exit_kernel:
6188
6189 ret = pthread_join(agent_reg_thread, &status);
6190 if (ret) {
6191 errno = ret;
6192 PERROR("pthread_join agent");
6193 retval = -1;
6194 }
6195 exit_agent_reg:
6196
6197 ret = pthread_join(apps_notify_thread, &status);
6198 if (ret) {
6199 errno = ret;
6200 PERROR("pthread_join apps notify");
6201 retval = -1;
6202 }
6203 exit_apps_notify:
6204
6205 ret = pthread_join(apps_thread, &status);
6206 if (ret) {
6207 errno = ret;
6208 PERROR("pthread_join apps");
6209 retval = -1;
6210 }
6211 exit_apps:
6212
6213 ret = pthread_join(reg_apps_thread, &status);
6214 if (ret) {
6215 errno = ret;
6216 PERROR("pthread_join");
6217 retval = -1;
6218 }
6219 exit_reg_apps:
6220
6221 /*
6222 * Join dispatch thread after joining reg_apps_thread to ensure
6223 * we don't leak applications in the queue.
6224 */
6225 ret = pthread_join(dispatch_thread, &status);
6226 if (ret) {
6227 errno = ret;
6228 PERROR("pthread_join");
6229 retval = -1;
6230 }
6231 exit_dispatch:
6232
6233 ret = pthread_join(client_thread, &status);
6234 if (ret) {
6235 errno = ret;
6236 PERROR("pthread_join");
6237 retval = -1;
6238 }
6239 exit_client:
6240
6241 ret = pthread_join(health_thread, &status);
6242 if (ret) {
6243 errno = ret;
6244 PERROR("pthread_join health thread");
6245 retval = -1;
6246 }
6247 exit_health:
6248
6249 exit_init_data:
6250 /*
6251 * Wait for all pending call_rcu work to complete before tearing
6252 * down data structures. call_rcu worker may be trying to
6253 * perform lookups in those structures.
6254 */
6255 rcu_barrier();
6256 /*
6257 * sessiond_cleanup() is called when no other thread is running, except
6258 * the ht_cleanup thread, which is needed to destroy the hash tables.
6259 */
6260 rcu_thread_online();
6261 sessiond_cleanup();
6262 rcu_thread_offline();
6263 rcu_unregister_thread();
6264
6265 /*
6266 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6267 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6268 * the queue is empty before shutting down the clean-up thread.
6269 */
6270 rcu_barrier();
6271
6272 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6273 if (ret) {
6274 retval = -1;
6275 }
6276 exit_ht_cleanup:
6277
6278 health_app_destroy(health_sessiond);
6279 exit_health_sessiond_cleanup:
6280 exit_create_run_as_worker_cleanup:
6281
6282 exit_options:
6283 sessiond_cleanup_options();
6284
6285 exit_set_signal_handler:
6286
6287 if (!retval) {
6288 exit(EXIT_SUCCESS);
6289 } else {
6290 exit(EXIT_FAILURE);
6291 }
6292 }
This page took 0.236979 seconds and 5 git commands to generate.