Load session daemon options from configuration file
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _GNU_SOURCE
21 #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 <config.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/config/config.h>
50
51 #include "lttng-sessiond.h"
52 #include "buffer-registry.h"
53 #include "channel.h"
54 #include "cmd.h"
55 #include "consumer.h"
56 #include "context.h"
57 #include "event.h"
58 #include "kernel.h"
59 #include "kernel-consumer.h"
60 #include "modprobe.h"
61 #include "shm.h"
62 #include "ust-ctl.h"
63 #include "ust-consumer.h"
64 #include "utils.h"
65 #include "fd-limit.h"
66 #include "health-sessiond.h"
67 #include "testpoint.h"
68 #include "ust-thread.h"
69 #include "jul-thread.h"
70
71 #define CONSUMERD_FILE "lttng-consumerd"
72
73 const char *progname;
74 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
75 static int tracing_group_name_override;
76 static char *opt_pidfile;
77 static int opt_sig_parent;
78 static int opt_verbose_consumer;
79 static int opt_daemon;
80 static int opt_no_kernel;
81 static int is_root; /* Set to 1 if the daemon is running as root */
82 static pid_t ppid; /* Parent PID for --sig-parent option */
83 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
84 static char *rundir;
85
86 /* Set to 1 when a SIGUSR1 signal is received. */
87 static int recv_child_signal;
88
89 /*
90 * Consumer daemon specific control data. Every value not initialized here is
91 * set to 0 by the static definition.
92 */
93 static struct consumer_data kconsumer_data = {
94 .type = LTTNG_CONSUMER_KERNEL,
95 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
96 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
97 .err_sock = -1,
98 .cmd_sock = -1,
99 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
100 .lock = PTHREAD_MUTEX_INITIALIZER,
101 .cond = PTHREAD_COND_INITIALIZER,
102 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
103 };
104 static struct consumer_data ustconsumer64_data = {
105 .type = LTTNG_CONSUMER64_UST,
106 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
107 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
108 .err_sock = -1,
109 .cmd_sock = -1,
110 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
111 .lock = PTHREAD_MUTEX_INITIALIZER,
112 .cond = PTHREAD_COND_INITIALIZER,
113 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
114 };
115 static struct consumer_data ustconsumer32_data = {
116 .type = LTTNG_CONSUMER32_UST,
117 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
118 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
119 .err_sock = -1,
120 .cmd_sock = -1,
121 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
122 .lock = PTHREAD_MUTEX_INITIALIZER,
123 .cond = PTHREAD_COND_INITIALIZER,
124 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
125 };
126
127 /* Command line options */
128 static const struct option long_options[] = {
129 { "client-sock", 1, 0, 'c' },
130 { "apps-sock", 1, 0, 'a' },
131 { "kconsumerd-cmd-sock", 1, 0, 'C' },
132 { "kconsumerd-err-sock", 1, 0, 'E' },
133 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
134 { "ustconsumerd32-err-sock", 1, 0, 'H' },
135 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
136 { "ustconsumerd64-err-sock", 1, 0, 'F' },
137 { "consumerd32-path", 1, 0, 'u' },
138 { "consumerd32-libdir", 1, 0, 'U' },
139 { "consumerd64-path", 1, 0, 't' },
140 { "consumerd64-libdir", 1, 0, 'T' },
141 { "daemonize", 0, 0, 'd' },
142 { "sig-parent", 0, 0, 'S' },
143 { "help", 0, 0, 'h' },
144 { "group", 1, 0, 'g' },
145 { "version", 0, 0, 'V' },
146 { "quiet", 0, 0, 'q' },
147 { "verbose", 0, 0, 'v' },
148 { "verbose-consumer", 0, 0, 'Z' },
149 { "no-kernel", 0, 0, 'N' },
150 { "pidfile", 1, 0, 'p' },
151 { "jul-tcp-port", 1, 0, 'J' },
152 { "config", 1, 0, 'f' },
153 { NULL, 0, 0, 0 }
154 };
155
156 /* Command line options to ignore from configuration file */
157 static const char *config_ignore_options[] = { "help", "version", "config" };
158
159 /* Shared between threads */
160 static int dispatch_thread_exit;
161
162 /* Global application Unix socket path */
163 static char apps_unix_sock_path[PATH_MAX];
164 /* Global client Unix socket path */
165 static char client_unix_sock_path[PATH_MAX];
166 /* global wait shm path for UST */
167 static char wait_shm_path[PATH_MAX];
168 /* Global health check unix path */
169 static char health_unix_sock_path[PATH_MAX];
170
171 /* Sockets and FDs */
172 static int client_sock = -1;
173 static int apps_sock = -1;
174 int kernel_tracer_fd = -1;
175 static int kernel_poll_pipe[2] = { -1, -1 };
176
177 /*
178 * Quit pipe for all threads. This permits a single cancellation point
179 * for all threads when receiving an event on the pipe.
180 */
181 static int thread_quit_pipe[2] = { -1, -1 };
182
183 /*
184 * This pipe is used to inform the thread managing application communication
185 * that a command is queued and ready to be processed.
186 */
187 static int apps_cmd_pipe[2] = { -1, -1 };
188
189 int apps_cmd_notify_pipe[2] = { -1, -1 };
190
191 /* Pthread, Mutexes and Semaphores */
192 static pthread_t apps_thread;
193 static pthread_t apps_notify_thread;
194 static pthread_t reg_apps_thread;
195 static pthread_t client_thread;
196 static pthread_t kernel_thread;
197 static pthread_t dispatch_thread;
198 static pthread_t health_thread;
199 static pthread_t ht_cleanup_thread;
200 static pthread_t jul_reg_thread;
201
202 /*
203 * UST registration command queue. This queue is tied with a futex and uses a N
204 * wakers / 1 waiter implemented and detailed in futex.c/.h
205 *
206 * The thread_manage_apps and thread_dispatch_ust_registration interact with
207 * this queue and the wait/wake scheme.
208 */
209 static struct ust_cmd_queue ust_cmd_queue;
210
211 /*
212 * Pointer initialized before thread creation.
213 *
214 * This points to the tracing session list containing the session count and a
215 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
216 * MUST NOT be taken if you call a public function in session.c.
217 *
218 * The lock is nested inside the structure: session_list_ptr->lock. Please use
219 * session_lock_list and session_unlock_list for lock acquisition.
220 */
221 static struct ltt_session_list *session_list_ptr;
222
223 int ust_consumerd64_fd = -1;
224 int ust_consumerd32_fd = -1;
225
226 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
227 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
228 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
229 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
230 static int consumerd32_bin_override;
231 static int consumerd64_bin_override;
232 static int consumerd32_libdir_override;
233 static int consumerd64_libdir_override;
234
235 static const char *module_proc_lttng = "/proc/lttng";
236
237 /*
238 * Consumer daemon state which is changed when spawning it, killing it or in
239 * case of a fatal error.
240 */
241 enum consumerd_state {
242 CONSUMER_STARTED = 1,
243 CONSUMER_STOPPED = 2,
244 CONSUMER_ERROR = 3,
245 };
246
247 /*
248 * This consumer daemon state is used to validate if a client command will be
249 * able to reach the consumer. If not, the client is informed. For instance,
250 * doing a "lttng start" when the consumer state is set to ERROR will return an
251 * error to the client.
252 *
253 * The following example shows a possible race condition of this scheme:
254 *
255 * consumer thread error happens
256 * client cmd arrives
257 * client cmd checks state -> still OK
258 * consumer thread exit, sets error
259 * client cmd try to talk to consumer
260 * ...
261 *
262 * However, since the consumer is a different daemon, we have no way of making
263 * sure the command will reach it safely even with this state flag. This is why
264 * we consider that up to the state validation during command processing, the
265 * command is safe. After that, we can not guarantee the correctness of the
266 * client request vis-a-vis the consumer.
267 */
268 static enum consumerd_state ust_consumerd_state;
269 static enum consumerd_state kernel_consumerd_state;
270
271 /*
272 * Socket timeout for receiving and sending in seconds.
273 */
274 static int app_socket_timeout;
275
276 /* Set in main() with the current page size. */
277 long page_size;
278
279 /* Application health monitoring */
280 struct health_app *health_sessiond;
281
282 /* JUL TCP port for registration. Used by the JUL thread. */
283 unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
284
285 const char * const config_section_name = "sessiond";
286
287 static
288 void setup_consumerd_path(void)
289 {
290 const char *bin, *libdir;
291
292 /*
293 * Allow INSTALL_BIN_PATH to be used as a target path for the
294 * native architecture size consumer if CONFIG_CONSUMER*_PATH
295 * has not been defined.
296 */
297 #if (CAA_BITS_PER_LONG == 32)
298 if (!consumerd32_bin[0]) {
299 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
300 }
301 if (!consumerd32_libdir[0]) {
302 consumerd32_libdir = INSTALL_LIB_PATH;
303 }
304 #elif (CAA_BITS_PER_LONG == 64)
305 if (!consumerd64_bin[0]) {
306 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
307 }
308 if (!consumerd64_libdir[0]) {
309 consumerd64_libdir = INSTALL_LIB_PATH;
310 }
311 #else
312 #error "Unknown bitness"
313 #endif
314
315 /*
316 * runtime env. var. overrides the build default.
317 */
318 bin = getenv("LTTNG_CONSUMERD32_BIN");
319 if (bin) {
320 consumerd32_bin = bin;
321 }
322 bin = getenv("LTTNG_CONSUMERD64_BIN");
323 if (bin) {
324 consumerd64_bin = bin;
325 }
326 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
327 if (libdir) {
328 consumerd32_libdir = libdir;
329 }
330 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
331 if (libdir) {
332 consumerd64_libdir = libdir;
333 }
334 }
335
336 /*
337 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
338 */
339 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
340 {
341 int ret;
342
343 assert(events);
344
345 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
346 if (ret < 0) {
347 goto error;
348 }
349
350 /* Add quit pipe */
351 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
352 if (ret < 0) {
353 goto error;
354 }
355
356 return 0;
357
358 error:
359 return ret;
360 }
361
362 /*
363 * Check if the thread quit pipe was triggered.
364 *
365 * Return 1 if it was triggered else 0;
366 */
367 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
368 {
369 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
370 return 1;
371 }
372
373 return 0;
374 }
375
376 /*
377 * Init thread quit pipe.
378 *
379 * Return -1 on error or 0 if all pipes are created.
380 */
381 static int init_thread_quit_pipe(void)
382 {
383 int ret, i;
384
385 ret = pipe(thread_quit_pipe);
386 if (ret < 0) {
387 PERROR("thread quit pipe");
388 goto error;
389 }
390
391 for (i = 0; i < 2; i++) {
392 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
393 if (ret < 0) {
394 PERROR("fcntl");
395 goto error;
396 }
397 }
398
399 error:
400 return ret;
401 }
402
403 /*
404 * Stop all threads by closing the thread quit pipe.
405 */
406 static void stop_threads(void)
407 {
408 int ret;
409
410 /* Stopping all threads */
411 DBG("Terminating all threads");
412 ret = notify_thread_pipe(thread_quit_pipe[1]);
413 if (ret < 0) {
414 ERR("write error on thread quit pipe");
415 }
416
417 /* Dispatch thread */
418 CMM_STORE_SHARED(dispatch_thread_exit, 1);
419 futex_nto1_wake(&ust_cmd_queue.futex);
420 }
421
422 /*
423 * Close every consumer sockets.
424 */
425 static void close_consumer_sockets(void)
426 {
427 int ret;
428
429 if (kconsumer_data.err_sock >= 0) {
430 ret = close(kconsumer_data.err_sock);
431 if (ret < 0) {
432 PERROR("kernel consumer err_sock close");
433 }
434 }
435 if (ustconsumer32_data.err_sock >= 0) {
436 ret = close(ustconsumer32_data.err_sock);
437 if (ret < 0) {
438 PERROR("UST consumerd32 err_sock close");
439 }
440 }
441 if (ustconsumer64_data.err_sock >= 0) {
442 ret = close(ustconsumer64_data.err_sock);
443 if (ret < 0) {
444 PERROR("UST consumerd64 err_sock close");
445 }
446 }
447 if (kconsumer_data.cmd_sock >= 0) {
448 ret = close(kconsumer_data.cmd_sock);
449 if (ret < 0) {
450 PERROR("kernel consumer cmd_sock close");
451 }
452 }
453 if (ustconsumer32_data.cmd_sock >= 0) {
454 ret = close(ustconsumer32_data.cmd_sock);
455 if (ret < 0) {
456 PERROR("UST consumerd32 cmd_sock close");
457 }
458 }
459 if (ustconsumer64_data.cmd_sock >= 0) {
460 ret = close(ustconsumer64_data.cmd_sock);
461 if (ret < 0) {
462 PERROR("UST consumerd64 cmd_sock close");
463 }
464 }
465 }
466
467 /*
468 * Cleanup the daemon
469 */
470 static void cleanup(void)
471 {
472 int ret;
473 struct ltt_session *sess, *stmp;
474 char path[PATH_MAX];
475
476 DBG("Cleaning up");
477
478 /*
479 * Close the thread quit pipe. It has already done its job,
480 * since we are now called.
481 */
482 utils_close_pipe(thread_quit_pipe);
483
484 /*
485 * If opt_pidfile is undefined, the default file will be wiped when
486 * removing the rundir.
487 */
488 if (opt_pidfile) {
489 ret = remove(opt_pidfile);
490 if (ret < 0) {
491 PERROR("remove pidfile %s", opt_pidfile);
492 }
493 }
494
495 DBG("Removing sessiond and consumerd content of directory %s", rundir);
496
497 /* sessiond */
498 snprintf(path, PATH_MAX,
499 "%s/%s",
500 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
501 DBG("Removing %s", path);
502 (void) unlink(path);
503
504 snprintf(path, PATH_MAX, "%s/%s", rundir,
505 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
506 DBG("Removing %s", path);
507 (void) unlink(path);
508
509 /* kconsumerd */
510 snprintf(path, PATH_MAX,
511 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
512 rundir);
513 DBG("Removing %s", path);
514 (void) unlink(path);
515
516 snprintf(path, PATH_MAX,
517 DEFAULT_KCONSUMERD_PATH,
518 rundir);
519 DBG("Removing directory %s", path);
520 (void) rmdir(path);
521
522 /* ust consumerd 32 */
523 snprintf(path, PATH_MAX,
524 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
525 rundir);
526 DBG("Removing %s", path);
527 (void) unlink(path);
528
529 snprintf(path, PATH_MAX,
530 DEFAULT_USTCONSUMERD32_PATH,
531 rundir);
532 DBG("Removing directory %s", path);
533 (void) rmdir(path);
534
535 /* ust consumerd 64 */
536 snprintf(path, PATH_MAX,
537 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
538 rundir);
539 DBG("Removing %s", path);
540 (void) unlink(path);
541
542 snprintf(path, PATH_MAX,
543 DEFAULT_USTCONSUMERD64_PATH,
544 rundir);
545 DBG("Removing directory %s", path);
546 (void) rmdir(path);
547
548 /*
549 * We do NOT rmdir rundir because there are other processes
550 * using it, for instance lttng-relayd, which can start in
551 * parallel with this teardown.
552 */
553
554 free(rundir);
555
556 DBG("Cleaning up all sessions");
557
558 /* Destroy session list mutex */
559 if (session_list_ptr != NULL) {
560 pthread_mutex_destroy(&session_list_ptr->lock);
561
562 /* Cleanup ALL session */
563 cds_list_for_each_entry_safe(sess, stmp,
564 &session_list_ptr->head, list) {
565 cmd_destroy_session(sess, kernel_poll_pipe[1]);
566 }
567 }
568
569 DBG("Closing all UST sockets");
570 ust_app_clean_list();
571 buffer_reg_destroy_registries();
572
573 if (is_root && !opt_no_kernel) {
574 DBG2("Closing kernel fd");
575 if (kernel_tracer_fd >= 0) {
576 ret = close(kernel_tracer_fd);
577 if (ret) {
578 PERROR("close");
579 }
580 }
581 DBG("Unloading kernel modules");
582 modprobe_remove_lttng_all();
583 }
584
585 close_consumer_sockets();
586
587 /*
588 * If the override option is set, the pointer points to a *non* const thus
589 * freeing it even though the variable type is set to const.
590 */
591 if (tracing_group_name_override) {
592 free((void *) tracing_group_name);
593 }
594 if (consumerd32_bin_override) {
595 free((void *) consumerd32_bin);
596 }
597 if (consumerd64_bin_override) {
598 free((void *) consumerd64_bin);
599 }
600 if (consumerd32_libdir_override) {
601 free((void *) consumerd32_libdir);
602 }
603 if (consumerd64_libdir_override) {
604 free((void *) consumerd64_libdir);
605 }
606
607 if (opt_pidfile) {
608 free(opt_pidfile);
609 }
610
611 /* <fun> */
612 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
613 "Matthew, BEET driven development works!%c[%dm",
614 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
615 /* </fun> */
616 }
617
618 /*
619 * Send data on a unix socket using the liblttsessiondcomm API.
620 *
621 * Return lttcomm error code.
622 */
623 static int send_unix_sock(int sock, void *buf, size_t len)
624 {
625 /* Check valid length */
626 if (len == 0) {
627 return -1;
628 }
629
630 return lttcomm_send_unix_sock(sock, buf, len);
631 }
632
633 /*
634 * Free memory of a command context structure.
635 */
636 static void clean_command_ctx(struct command_ctx **cmd_ctx)
637 {
638 DBG("Clean command context structure");
639 if (*cmd_ctx) {
640 if ((*cmd_ctx)->llm) {
641 free((*cmd_ctx)->llm);
642 }
643 if ((*cmd_ctx)->lsm) {
644 free((*cmd_ctx)->lsm);
645 }
646 free(*cmd_ctx);
647 *cmd_ctx = NULL;
648 }
649 }
650
651 /*
652 * Notify UST applications using the shm mmap futex.
653 */
654 static int notify_ust_apps(int active)
655 {
656 char *wait_shm_mmap;
657
658 DBG("Notifying applications of session daemon state: %d", active);
659
660 /* See shm.c for this call implying mmap, shm and futex calls */
661 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
662 if (wait_shm_mmap == NULL) {
663 goto error;
664 }
665
666 /* Wake waiting process */
667 futex_wait_update((int32_t *) wait_shm_mmap, active);
668
669 /* Apps notified successfully */
670 return 0;
671
672 error:
673 return -1;
674 }
675
676 /*
677 * Setup the outgoing data buffer for the response (llm) by allocating the
678 * right amount of memory and copying the original information from the lsm
679 * structure.
680 *
681 * Return total size of the buffer pointed by buf.
682 */
683 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
684 {
685 int ret, buf_size;
686
687 buf_size = size;
688
689 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
690 if (cmd_ctx->llm == NULL) {
691 PERROR("zmalloc");
692 ret = -ENOMEM;
693 goto error;
694 }
695
696 /* Copy common data */
697 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
698 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
699
700 cmd_ctx->llm->data_size = size;
701 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
702
703 return buf_size;
704
705 error:
706 return ret;
707 }
708
709 /*
710 * Update the kernel poll set of all channel fd available over all tracing
711 * session. Add the wakeup pipe at the end of the set.
712 */
713 static int update_kernel_poll(struct lttng_poll_event *events)
714 {
715 int ret;
716 struct ltt_session *session;
717 struct ltt_kernel_channel *channel;
718
719 DBG("Updating kernel poll set");
720
721 session_lock_list();
722 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
723 session_lock(session);
724 if (session->kernel_session == NULL) {
725 session_unlock(session);
726 continue;
727 }
728
729 cds_list_for_each_entry(channel,
730 &session->kernel_session->channel_list.head, list) {
731 /* Add channel fd to the kernel poll set */
732 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
733 if (ret < 0) {
734 session_unlock(session);
735 goto error;
736 }
737 DBG("Channel fd %d added to kernel set", channel->fd);
738 }
739 session_unlock(session);
740 }
741 session_unlock_list();
742
743 return 0;
744
745 error:
746 session_unlock_list();
747 return -1;
748 }
749
750 /*
751 * Find the channel fd from 'fd' over all tracing session. When found, check
752 * for new channel stream and send those stream fds to the kernel consumer.
753 *
754 * Useful for CPU hotplug feature.
755 */
756 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
757 {
758 int ret = 0;
759 struct ltt_session *session;
760 struct ltt_kernel_session *ksess;
761 struct ltt_kernel_channel *channel;
762
763 DBG("Updating kernel streams for channel fd %d", fd);
764
765 session_lock_list();
766 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
767 session_lock(session);
768 if (session->kernel_session == NULL) {
769 session_unlock(session);
770 continue;
771 }
772 ksess = session->kernel_session;
773
774 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
775 if (channel->fd == fd) {
776 DBG("Channel found, updating kernel streams");
777 ret = kernel_open_channel_stream(channel);
778 if (ret < 0) {
779 goto error;
780 }
781 /* Update the stream global counter */
782 ksess->stream_count_global += ret;
783
784 /*
785 * Have we already sent fds to the consumer? If yes, it means
786 * that tracing is started so it is safe to send our updated
787 * stream fds.
788 */
789 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
790 struct lttng_ht_iter iter;
791 struct consumer_socket *socket;
792
793 rcu_read_lock();
794 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
795 &iter.iter, socket, node.node) {
796 pthread_mutex_lock(socket->lock);
797 ret = kernel_consumer_send_channel_stream(socket,
798 channel, ksess,
799 session->output_traces ? 1 : 0);
800 pthread_mutex_unlock(socket->lock);
801 if (ret < 0) {
802 rcu_read_unlock();
803 goto error;
804 }
805 }
806 rcu_read_unlock();
807 }
808 goto error;
809 }
810 }
811 session_unlock(session);
812 }
813 session_unlock_list();
814 return ret;
815
816 error:
817 session_unlock(session);
818 session_unlock_list();
819 return ret;
820 }
821
822 /*
823 * For each tracing session, update newly registered apps. The session list
824 * lock MUST be acquired before calling this.
825 */
826 static void update_ust_app(int app_sock)
827 {
828 struct ltt_session *sess, *stmp;
829
830 /* Consumer is in an ERROR state. Stop any application update. */
831 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
832 /* Stop the update process since the consumer is dead. */
833 return;
834 }
835
836 /* For all tracing session(s) */
837 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
838 session_lock(sess);
839 if (sess->ust_session) {
840 ust_app_global_update(sess->ust_session, app_sock);
841 }
842 session_unlock(sess);
843 }
844 }
845
846 /*
847 * This thread manage event coming from the kernel.
848 *
849 * Features supported in this thread:
850 * -) CPU Hotplug
851 */
852 static void *thread_manage_kernel(void *data)
853 {
854 int ret, i, pollfd, update_poll_flag = 1, err = -1;
855 uint32_t revents, nb_fd;
856 char tmp;
857 struct lttng_poll_event events;
858
859 DBG("[thread] Thread manage kernel started");
860
861 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
862
863 /*
864 * This first step of the while is to clean this structure which could free
865 * non NULL pointers so initialize it before the loop.
866 */
867 lttng_poll_init(&events);
868
869 if (testpoint(thread_manage_kernel)) {
870 goto error_testpoint;
871 }
872
873 health_code_update();
874
875 if (testpoint(thread_manage_kernel_before_loop)) {
876 goto error_testpoint;
877 }
878
879 while (1) {
880 health_code_update();
881
882 if (update_poll_flag == 1) {
883 /* Clean events object. We are about to populate it again. */
884 lttng_poll_clean(&events);
885
886 ret = sessiond_set_thread_pollset(&events, 2);
887 if (ret < 0) {
888 goto error_poll_create;
889 }
890
891 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
892 if (ret < 0) {
893 goto error;
894 }
895
896 /* This will add the available kernel channel if any. */
897 ret = update_kernel_poll(&events);
898 if (ret < 0) {
899 goto error;
900 }
901 update_poll_flag = 0;
902 }
903
904 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
905
906 /* Poll infinite value of time */
907 restart:
908 health_poll_entry();
909 ret = lttng_poll_wait(&events, -1);
910 health_poll_exit();
911 if (ret < 0) {
912 /*
913 * Restart interrupted system call.
914 */
915 if (errno == EINTR) {
916 goto restart;
917 }
918 goto error;
919 } else if (ret == 0) {
920 /* Should not happen since timeout is infinite */
921 ERR("Return value of poll is 0 with an infinite timeout.\n"
922 "This should not have happened! Continuing...");
923 continue;
924 }
925
926 nb_fd = ret;
927
928 for (i = 0; i < nb_fd; i++) {
929 /* Fetch once the poll data */
930 revents = LTTNG_POLL_GETEV(&events, i);
931 pollfd = LTTNG_POLL_GETFD(&events, i);
932
933 health_code_update();
934
935 /* Thread quit pipe has been closed. Killing thread. */
936 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
937 if (ret) {
938 err = 0;
939 goto exit;
940 }
941
942 /* Check for data on kernel pipe */
943 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
944 (void) lttng_read(kernel_poll_pipe[0],
945 &tmp, 1);
946 /*
947 * Ret value is useless here, if this pipe gets any actions an
948 * update is required anyway.
949 */
950 update_poll_flag = 1;
951 continue;
952 } else {
953 /*
954 * New CPU detected by the kernel. Adding kernel stream to
955 * kernel session and updating the kernel consumer
956 */
957 if (revents & LPOLLIN) {
958 ret = update_kernel_stream(&kconsumer_data, pollfd);
959 if (ret < 0) {
960 continue;
961 }
962 break;
963 /*
964 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
965 * and unregister kernel stream at this point.
966 */
967 }
968 }
969 }
970 }
971
972 exit:
973 error:
974 lttng_poll_clean(&events);
975 error_poll_create:
976 error_testpoint:
977 utils_close_pipe(kernel_poll_pipe);
978 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
979 if (err) {
980 health_error();
981 ERR("Health error occurred in %s", __func__);
982 WARN("Kernel thread died unexpectedly. "
983 "Kernel tracing can continue but CPU hotplug is disabled.");
984 }
985 health_unregister(health_sessiond);
986 DBG("Kernel thread dying");
987 return NULL;
988 }
989
990 /*
991 * Signal pthread condition of the consumer data that the thread.
992 */
993 static void signal_consumer_condition(struct consumer_data *data, int state)
994 {
995 pthread_mutex_lock(&data->cond_mutex);
996
997 /*
998 * The state is set before signaling. It can be any value, it's the waiter
999 * job to correctly interpret this condition variable associated to the
1000 * consumer pthread_cond.
1001 *
1002 * A value of 0 means that the corresponding thread of the consumer data
1003 * was not started. 1 indicates that the thread has started and is ready
1004 * for action. A negative value means that there was an error during the
1005 * thread bootstrap.
1006 */
1007 data->consumer_thread_is_ready = state;
1008 (void) pthread_cond_signal(&data->cond);
1009
1010 pthread_mutex_unlock(&data->cond_mutex);
1011 }
1012
1013 /*
1014 * This thread manage the consumer error sent back to the session daemon.
1015 */
1016 static void *thread_manage_consumer(void *data)
1017 {
1018 int sock = -1, i, ret, pollfd, err = -1;
1019 uint32_t revents, nb_fd;
1020 enum lttcomm_return_code code;
1021 struct lttng_poll_event events;
1022 struct consumer_data *consumer_data = data;
1023
1024 DBG("[thread] Manage consumer started");
1025
1026 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1027
1028 health_code_update();
1029
1030 /*
1031 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1032 * metadata_sock. Nothing more will be added to this poll set.
1033 */
1034 ret = sessiond_set_thread_pollset(&events, 3);
1035 if (ret < 0) {
1036 goto error_poll;
1037 }
1038
1039 /*
1040 * The error socket here is already in a listening state which was done
1041 * just before spawning this thread to avoid a race between the consumer
1042 * daemon exec trying to connect and the listen() call.
1043 */
1044 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1045 if (ret < 0) {
1046 goto error;
1047 }
1048
1049 health_code_update();
1050
1051 /* Infinite blocking call, waiting for transmission */
1052 restart:
1053 health_poll_entry();
1054
1055 if (testpoint(thread_manage_consumer)) {
1056 goto error;
1057 }
1058
1059 ret = lttng_poll_wait(&events, -1);
1060 health_poll_exit();
1061 if (ret < 0) {
1062 /*
1063 * Restart interrupted system call.
1064 */
1065 if (errno == EINTR) {
1066 goto restart;
1067 }
1068 goto error;
1069 }
1070
1071 nb_fd = ret;
1072
1073 for (i = 0; i < nb_fd; i++) {
1074 /* Fetch once the poll data */
1075 revents = LTTNG_POLL_GETEV(&events, i);
1076 pollfd = LTTNG_POLL_GETFD(&events, i);
1077
1078 health_code_update();
1079
1080 /* Thread quit pipe has been closed. Killing thread. */
1081 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1082 if (ret) {
1083 err = 0;
1084 goto exit;
1085 }
1086
1087 /* Event on the registration socket */
1088 if (pollfd == consumer_data->err_sock) {
1089 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1090 ERR("consumer err socket poll error");
1091 goto error;
1092 }
1093 }
1094 }
1095
1096 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1097 if (sock < 0) {
1098 goto error;
1099 }
1100
1101 /*
1102 * Set the CLOEXEC flag. Return code is useless because either way, the
1103 * show must go on.
1104 */
1105 (void) utils_set_fd_cloexec(sock);
1106
1107 health_code_update();
1108
1109 DBG2("Receiving code from consumer err_sock");
1110
1111 /* Getting status code from kconsumerd */
1112 ret = lttcomm_recv_unix_sock(sock, &code,
1113 sizeof(enum lttcomm_return_code));
1114 if (ret <= 0) {
1115 goto error;
1116 }
1117
1118 health_code_update();
1119
1120 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1121 /* Connect both socket, command and metadata. */
1122 consumer_data->cmd_sock =
1123 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1124 consumer_data->metadata_fd =
1125 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1126 if (consumer_data->cmd_sock < 0
1127 || consumer_data->metadata_fd < 0) {
1128 PERROR("consumer connect cmd socket");
1129 /* On error, signal condition and quit. */
1130 signal_consumer_condition(consumer_data, -1);
1131 goto error;
1132 }
1133 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1134 /* Create metadata socket lock. */
1135 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1136 if (consumer_data->metadata_sock.lock == NULL) {
1137 PERROR("zmalloc pthread mutex");
1138 ret = -1;
1139 goto error;
1140 }
1141 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1142
1143 signal_consumer_condition(consumer_data, 1);
1144 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1145 DBG("Consumer metadata socket ready (fd: %d)",
1146 consumer_data->metadata_fd);
1147 } else {
1148 ERR("consumer error when waiting for SOCK_READY : %s",
1149 lttcomm_get_readable_code(-code));
1150 goto error;
1151 }
1152
1153 /* Remove the consumerd error sock since we've established a connexion */
1154 ret = lttng_poll_del(&events, consumer_data->err_sock);
1155 if (ret < 0) {
1156 goto error;
1157 }
1158
1159 /* Add new accepted error socket. */
1160 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1161 if (ret < 0) {
1162 goto error;
1163 }
1164
1165 /* Add metadata socket that is successfully connected. */
1166 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1167 LPOLLIN | LPOLLRDHUP);
1168 if (ret < 0) {
1169 goto error;
1170 }
1171
1172 health_code_update();
1173
1174 /* Infinite blocking call, waiting for transmission */
1175 restart_poll:
1176 while (1) {
1177 health_poll_entry();
1178 ret = lttng_poll_wait(&events, -1);
1179 health_poll_exit();
1180 if (ret < 0) {
1181 /*
1182 * Restart interrupted system call.
1183 */
1184 if (errno == EINTR) {
1185 goto restart_poll;
1186 }
1187 goto error;
1188 }
1189
1190 nb_fd = ret;
1191
1192 for (i = 0; i < nb_fd; i++) {
1193 /* Fetch once the poll data */
1194 revents = LTTNG_POLL_GETEV(&events, i);
1195 pollfd = LTTNG_POLL_GETFD(&events, i);
1196
1197 health_code_update();
1198
1199 /* Thread quit pipe has been closed. Killing thread. */
1200 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1201 if (ret) {
1202 err = 0;
1203 goto exit;
1204 }
1205
1206 if (pollfd == sock) {
1207 /* Event on the consumerd socket */
1208 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1209 ERR("consumer err socket second poll error");
1210 goto error;
1211 }
1212 health_code_update();
1213 /* Wait for any kconsumerd error */
1214 ret = lttcomm_recv_unix_sock(sock, &code,
1215 sizeof(enum lttcomm_return_code));
1216 if (ret <= 0) {
1217 ERR("consumer closed the command socket");
1218 goto error;
1219 }
1220
1221 ERR("consumer return code : %s",
1222 lttcomm_get_readable_code(-code));
1223
1224 goto exit;
1225 } else if (pollfd == consumer_data->metadata_fd) {
1226 /* UST metadata requests */
1227 ret = ust_consumer_metadata_request(
1228 &consumer_data->metadata_sock);
1229 if (ret < 0) {
1230 ERR("Handling metadata request");
1231 goto error;
1232 }
1233 break;
1234 } else {
1235 ERR("Unknown pollfd");
1236 goto error;
1237 }
1238 }
1239 health_code_update();
1240 }
1241
1242 exit:
1243 error:
1244 /*
1245 * We lock here because we are about to close the sockets and some other
1246 * thread might be using them so get exclusive access which will abort all
1247 * other consumer command by other threads.
1248 */
1249 pthread_mutex_lock(&consumer_data->lock);
1250
1251 /* Immediately set the consumerd state to stopped */
1252 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1253 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1254 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1255 consumer_data->type == LTTNG_CONSUMER32_UST) {
1256 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1257 } else {
1258 /* Code flow error... */
1259 assert(0);
1260 }
1261
1262 if (consumer_data->err_sock >= 0) {
1263 ret = close(consumer_data->err_sock);
1264 if (ret) {
1265 PERROR("close");
1266 }
1267 consumer_data->err_sock = -1;
1268 }
1269 if (consumer_data->cmd_sock >= 0) {
1270 ret = close(consumer_data->cmd_sock);
1271 if (ret) {
1272 PERROR("close");
1273 }
1274 consumer_data->cmd_sock = -1;
1275 }
1276 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1277 ret = close(*consumer_data->metadata_sock.fd_ptr);
1278 if (ret) {
1279 PERROR("close");
1280 }
1281 }
1282
1283 if (sock >= 0) {
1284 ret = close(sock);
1285 if (ret) {
1286 PERROR("close");
1287 }
1288 }
1289
1290 unlink(consumer_data->err_unix_sock_path);
1291 unlink(consumer_data->cmd_unix_sock_path);
1292 consumer_data->pid = 0;
1293 pthread_mutex_unlock(&consumer_data->lock);
1294
1295 /* Cleanup metadata socket mutex. */
1296 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1297 free(consumer_data->metadata_sock.lock);
1298
1299 lttng_poll_clean(&events);
1300 error_poll:
1301 if (err) {
1302 health_error();
1303 ERR("Health error occurred in %s", __func__);
1304 }
1305 health_unregister(health_sessiond);
1306 DBG("consumer thread cleanup completed");
1307
1308 return NULL;
1309 }
1310
1311 /*
1312 * This thread manage application communication.
1313 */
1314 static void *thread_manage_apps(void *data)
1315 {
1316 int i, ret, pollfd, err = -1;
1317 ssize_t size_ret;
1318 uint32_t revents, nb_fd;
1319 struct lttng_poll_event events;
1320
1321 DBG("[thread] Manage application started");
1322
1323 rcu_register_thread();
1324 rcu_thread_online();
1325
1326 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1327
1328 if (testpoint(thread_manage_apps)) {
1329 goto error_testpoint;
1330 }
1331
1332 health_code_update();
1333
1334 ret = sessiond_set_thread_pollset(&events, 2);
1335 if (ret < 0) {
1336 goto error_poll_create;
1337 }
1338
1339 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1340 if (ret < 0) {
1341 goto error;
1342 }
1343
1344 if (testpoint(thread_manage_apps_before_loop)) {
1345 goto error;
1346 }
1347
1348 health_code_update();
1349
1350 while (1) {
1351 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1352
1353 /* Inifinite blocking call, waiting for transmission */
1354 restart:
1355 health_poll_entry();
1356 ret = lttng_poll_wait(&events, -1);
1357 health_poll_exit();
1358 if (ret < 0) {
1359 /*
1360 * Restart interrupted system call.
1361 */
1362 if (errno == EINTR) {
1363 goto restart;
1364 }
1365 goto error;
1366 }
1367
1368 nb_fd = ret;
1369
1370 for (i = 0; i < nb_fd; i++) {
1371 /* Fetch once the poll data */
1372 revents = LTTNG_POLL_GETEV(&events, i);
1373 pollfd = LTTNG_POLL_GETFD(&events, i);
1374
1375 health_code_update();
1376
1377 /* Thread quit pipe has been closed. Killing thread. */
1378 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1379 if (ret) {
1380 err = 0;
1381 goto exit;
1382 }
1383
1384 /* Inspect the apps cmd pipe */
1385 if (pollfd == apps_cmd_pipe[0]) {
1386 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1387 ERR("Apps command pipe error");
1388 goto error;
1389 } else if (revents & LPOLLIN) {
1390 int sock;
1391
1392 /* Empty pipe */
1393 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1394 if (size_ret < sizeof(sock)) {
1395 PERROR("read apps cmd pipe");
1396 goto error;
1397 }
1398
1399 health_code_update();
1400
1401 /*
1402 * We only monitor the error events of the socket. This
1403 * thread does not handle any incoming data from UST
1404 * (POLLIN).
1405 */
1406 ret = lttng_poll_add(&events, sock,
1407 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1408 if (ret < 0) {
1409 goto error;
1410 }
1411
1412 DBG("Apps with sock %d added to poll set", sock);
1413
1414 health_code_update();
1415
1416 break;
1417 }
1418 } else {
1419 /*
1420 * At this point, we know that a registered application made
1421 * the event at poll_wait.
1422 */
1423 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1424 /* Removing from the poll set */
1425 ret = lttng_poll_del(&events, pollfd);
1426 if (ret < 0) {
1427 goto error;
1428 }
1429
1430 /* Socket closed on remote end. */
1431 ust_app_unregister(pollfd);
1432 break;
1433 }
1434 }
1435
1436 health_code_update();
1437 }
1438 }
1439
1440 exit:
1441 error:
1442 lttng_poll_clean(&events);
1443 error_poll_create:
1444 error_testpoint:
1445 utils_close_pipe(apps_cmd_pipe);
1446 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1447
1448 /*
1449 * We don't clean the UST app hash table here since already registered
1450 * applications can still be controlled so let them be until the session
1451 * daemon dies or the applications stop.
1452 */
1453
1454 if (err) {
1455 health_error();
1456 ERR("Health error occurred in %s", __func__);
1457 }
1458 health_unregister(health_sessiond);
1459 DBG("Application communication apps thread cleanup complete");
1460 rcu_thread_offline();
1461 rcu_unregister_thread();
1462 return NULL;
1463 }
1464
1465 /*
1466 * Send a socket to a thread This is called from the dispatch UST registration
1467 * thread once all sockets are set for the application.
1468 *
1469 * The sock value can be invalid, we don't really care, the thread will handle
1470 * it and make the necessary cleanup if so.
1471 *
1472 * On success, return 0 else a negative value being the errno message of the
1473 * write().
1474 */
1475 static int send_socket_to_thread(int fd, int sock)
1476 {
1477 ssize_t ret;
1478
1479 /*
1480 * It's possible that the FD is set as invalid with -1 concurrently just
1481 * before calling this function being a shutdown state of the thread.
1482 */
1483 if (fd < 0) {
1484 ret = -EBADF;
1485 goto error;
1486 }
1487
1488 ret = lttng_write(fd, &sock, sizeof(sock));
1489 if (ret < sizeof(sock)) {
1490 PERROR("write apps pipe %d", fd);
1491 if (ret < 0) {
1492 ret = -errno;
1493 }
1494 goto error;
1495 }
1496
1497 /* All good. Don't send back the write positive ret value. */
1498 ret = 0;
1499 error:
1500 return (int) ret;
1501 }
1502
1503 /*
1504 * Sanitize the wait queue of the dispatch registration thread meaning removing
1505 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1506 * notify socket is never received.
1507 */
1508 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1509 {
1510 int ret, nb_fd = 0, i;
1511 unsigned int fd_added = 0;
1512 struct lttng_poll_event events;
1513 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1514
1515 assert(wait_queue);
1516
1517 lttng_poll_init(&events);
1518
1519 /* Just skip everything for an empty queue. */
1520 if (!wait_queue->count) {
1521 goto end;
1522 }
1523
1524 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1525 if (ret < 0) {
1526 goto error_create;
1527 }
1528
1529 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1530 &wait_queue->head, head) {
1531 assert(wait_node->app);
1532 ret = lttng_poll_add(&events, wait_node->app->sock,
1533 LPOLLHUP | LPOLLERR);
1534 if (ret < 0) {
1535 goto error;
1536 }
1537
1538 fd_added = 1;
1539 }
1540
1541 if (!fd_added) {
1542 goto end;
1543 }
1544
1545 /*
1546 * Poll but don't block so we can quickly identify the faulty events and
1547 * clean them afterwards from the wait queue.
1548 */
1549 ret = lttng_poll_wait(&events, 0);
1550 if (ret < 0) {
1551 goto error;
1552 }
1553 nb_fd = ret;
1554
1555 for (i = 0; i < nb_fd; i++) {
1556 /* Get faulty FD. */
1557 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1558 int pollfd = LTTNG_POLL_GETFD(&events, i);
1559
1560 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1561 &wait_queue->head, head) {
1562 if (pollfd == wait_node->app->sock &&
1563 (revents & (LPOLLHUP | LPOLLERR))) {
1564 cds_list_del(&wait_node->head);
1565 wait_queue->count--;
1566 ust_app_destroy(wait_node->app);
1567 free(wait_node);
1568 break;
1569 }
1570 }
1571 }
1572
1573 if (nb_fd > 0) {
1574 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1575 }
1576
1577 end:
1578 lttng_poll_clean(&events);
1579 return;
1580
1581 error:
1582 lttng_poll_clean(&events);
1583 error_create:
1584 ERR("Unable to sanitize wait queue");
1585 return;
1586 }
1587
1588 /*
1589 * Dispatch request from the registration threads to the application
1590 * communication thread.
1591 */
1592 static void *thread_dispatch_ust_registration(void *data)
1593 {
1594 int ret, err = -1;
1595 struct cds_wfq_node *node;
1596 struct ust_command *ust_cmd = NULL;
1597 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1598 struct ust_reg_wait_queue wait_queue = {
1599 .count = 0,
1600 };
1601
1602 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1603
1604 health_code_update();
1605
1606 CDS_INIT_LIST_HEAD(&wait_queue.head);
1607
1608 DBG("[thread] Dispatch UST command started");
1609
1610 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1611 health_code_update();
1612
1613 /* Atomically prepare the queue futex */
1614 futex_nto1_prepare(&ust_cmd_queue.futex);
1615
1616 do {
1617 struct ust_app *app = NULL;
1618 ust_cmd = NULL;
1619
1620 /*
1621 * Make sure we don't have node(s) that have hung up before receiving
1622 * the notify socket. This is to clean the list in order to avoid
1623 * memory leaks from notify socket that are never seen.
1624 */
1625 sanitize_wait_queue(&wait_queue);
1626
1627 health_code_update();
1628 /* Dequeue command for registration */
1629 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1630 if (node == NULL) {
1631 DBG("Woken up but nothing in the UST command queue");
1632 /* Continue thread execution */
1633 break;
1634 }
1635
1636 ust_cmd = caa_container_of(node, struct ust_command, node);
1637
1638 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1639 " gid:%d sock:%d name:%s (version %d.%d)",
1640 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1641 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1642 ust_cmd->sock, ust_cmd->reg_msg.name,
1643 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1644
1645 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1646 wait_node = zmalloc(sizeof(*wait_node));
1647 if (!wait_node) {
1648 PERROR("zmalloc wait_node dispatch");
1649 ret = close(ust_cmd->sock);
1650 if (ret < 0) {
1651 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1652 }
1653 lttng_fd_put(LTTNG_FD_APPS, 1);
1654 free(ust_cmd);
1655 goto error;
1656 }
1657 CDS_INIT_LIST_HEAD(&wait_node->head);
1658
1659 /* Create application object if socket is CMD. */
1660 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1661 ust_cmd->sock);
1662 if (!wait_node->app) {
1663 ret = close(ust_cmd->sock);
1664 if (ret < 0) {
1665 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1666 }
1667 lttng_fd_put(LTTNG_FD_APPS, 1);
1668 free(wait_node);
1669 free(ust_cmd);
1670 continue;
1671 }
1672 /*
1673 * Add application to the wait queue so we can set the notify
1674 * socket before putting this object in the global ht.
1675 */
1676 cds_list_add(&wait_node->head, &wait_queue.head);
1677 wait_queue.count++;
1678
1679 free(ust_cmd);
1680 /*
1681 * We have to continue here since we don't have the notify
1682 * socket and the application MUST be added to the hash table
1683 * only at that moment.
1684 */
1685 continue;
1686 } else {
1687 /*
1688 * Look for the application in the local wait queue and set the
1689 * notify socket if found.
1690 */
1691 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1692 &wait_queue.head, head) {
1693 health_code_update();
1694 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1695 wait_node->app->notify_sock = ust_cmd->sock;
1696 cds_list_del(&wait_node->head);
1697 wait_queue.count--;
1698 app = wait_node->app;
1699 free(wait_node);
1700 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1701 break;
1702 }
1703 }
1704
1705 /*
1706 * With no application at this stage the received socket is
1707 * basically useless so close it before we free the cmd data
1708 * structure for good.
1709 */
1710 if (!app) {
1711 ret = close(ust_cmd->sock);
1712 if (ret < 0) {
1713 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1714 }
1715 lttng_fd_put(LTTNG_FD_APPS, 1);
1716 }
1717 free(ust_cmd);
1718 }
1719
1720 if (app) {
1721 /*
1722 * @session_lock_list
1723 *
1724 * Lock the global session list so from the register up to the
1725 * registration done message, no thread can see the application
1726 * and change its state.
1727 */
1728 session_lock_list();
1729 rcu_read_lock();
1730
1731 /*
1732 * Add application to the global hash table. This needs to be
1733 * done before the update to the UST registry can locate the
1734 * application.
1735 */
1736 ust_app_add(app);
1737
1738 /* Set app version. This call will print an error if needed. */
1739 (void) ust_app_version(app);
1740
1741 /* Send notify socket through the notify pipe. */
1742 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1743 app->notify_sock);
1744 if (ret < 0) {
1745 rcu_read_unlock();
1746 session_unlock_list();
1747 /*
1748 * No notify thread, stop the UST tracing. However, this is
1749 * not an internal error of the this thread thus setting
1750 * the health error code to a normal exit.
1751 */
1752 err = 0;
1753 goto error;
1754 }
1755
1756 /*
1757 * Update newly registered application with the tracing
1758 * registry info already enabled information.
1759 */
1760 update_ust_app(app->sock);
1761
1762 /*
1763 * Don't care about return value. Let the manage apps threads
1764 * handle app unregistration upon socket close.
1765 */
1766 (void) ust_app_register_done(app->sock);
1767
1768 /*
1769 * Even if the application socket has been closed, send the app
1770 * to the thread and unregistration will take place at that
1771 * place.
1772 */
1773 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1774 if (ret < 0) {
1775 rcu_read_unlock();
1776 session_unlock_list();
1777 /*
1778 * No apps. thread, stop the UST tracing. However, this is
1779 * not an internal error of the this thread thus setting
1780 * the health error code to a normal exit.
1781 */
1782 err = 0;
1783 goto error;
1784 }
1785
1786 rcu_read_unlock();
1787 session_unlock_list();
1788 }
1789 } while (node != NULL);
1790
1791 health_poll_entry();
1792 /* Futex wait on queue. Blocking call on futex() */
1793 futex_nto1_wait(&ust_cmd_queue.futex);
1794 health_poll_exit();
1795 }
1796 /* Normal exit, no error */
1797 err = 0;
1798
1799 error:
1800 /* Clean up wait queue. */
1801 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1802 &wait_queue.head, head) {
1803 cds_list_del(&wait_node->head);
1804 wait_queue.count--;
1805 free(wait_node);
1806 }
1807
1808 DBG("Dispatch thread dying");
1809 if (err) {
1810 health_error();
1811 ERR("Health error occurred in %s", __func__);
1812 }
1813 health_unregister(health_sessiond);
1814 return NULL;
1815 }
1816
1817 /*
1818 * This thread manage application registration.
1819 */
1820 static void *thread_registration_apps(void *data)
1821 {
1822 int sock = -1, i, ret, pollfd, err = -1;
1823 uint32_t revents, nb_fd;
1824 struct lttng_poll_event events;
1825 /*
1826 * Get allocated in this thread, enqueued to a global queue, dequeued and
1827 * freed in the manage apps thread.
1828 */
1829 struct ust_command *ust_cmd = NULL;
1830
1831 DBG("[thread] Manage application registration started");
1832
1833 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1834
1835 if (testpoint(thread_registration_apps)) {
1836 goto error_testpoint;
1837 }
1838
1839 ret = lttcomm_listen_unix_sock(apps_sock);
1840 if (ret < 0) {
1841 goto error_listen;
1842 }
1843
1844 /*
1845 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1846 * more will be added to this poll set.
1847 */
1848 ret = sessiond_set_thread_pollset(&events, 2);
1849 if (ret < 0) {
1850 goto error_create_poll;
1851 }
1852
1853 /* Add the application registration socket */
1854 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1855 if (ret < 0) {
1856 goto error_poll_add;
1857 }
1858
1859 /* Notify all applications to register */
1860 ret = notify_ust_apps(1);
1861 if (ret < 0) {
1862 ERR("Failed to notify applications or create the wait shared memory.\n"
1863 "Execution continues but there might be problem for already\n"
1864 "running applications that wishes to register.");
1865 }
1866
1867 while (1) {
1868 DBG("Accepting application registration");
1869
1870 /* Inifinite blocking call, waiting for transmission */
1871 restart:
1872 health_poll_entry();
1873 ret = lttng_poll_wait(&events, -1);
1874 health_poll_exit();
1875 if (ret < 0) {
1876 /*
1877 * Restart interrupted system call.
1878 */
1879 if (errno == EINTR) {
1880 goto restart;
1881 }
1882 goto error;
1883 }
1884
1885 nb_fd = ret;
1886
1887 for (i = 0; i < nb_fd; i++) {
1888 health_code_update();
1889
1890 /* Fetch once the poll data */
1891 revents = LTTNG_POLL_GETEV(&events, i);
1892 pollfd = LTTNG_POLL_GETFD(&events, i);
1893
1894 /* Thread quit pipe has been closed. Killing thread. */
1895 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1896 if (ret) {
1897 err = 0;
1898 goto exit;
1899 }
1900
1901 /* Event on the registration socket */
1902 if (pollfd == apps_sock) {
1903 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1904 ERR("Register apps socket poll error");
1905 goto error;
1906 } else if (revents & LPOLLIN) {
1907 sock = lttcomm_accept_unix_sock(apps_sock);
1908 if (sock < 0) {
1909 goto error;
1910 }
1911
1912 /*
1913 * Set socket timeout for both receiving and ending.
1914 * app_socket_timeout is in seconds, whereas
1915 * lttcomm_setsockopt_rcv_timeout and
1916 * lttcomm_setsockopt_snd_timeout expect msec as
1917 * parameter.
1918 */
1919 (void) lttcomm_setsockopt_rcv_timeout(sock,
1920 app_socket_timeout * 1000);
1921 (void) lttcomm_setsockopt_snd_timeout(sock,
1922 app_socket_timeout * 1000);
1923
1924 /*
1925 * Set the CLOEXEC flag. Return code is useless because
1926 * either way, the show must go on.
1927 */
1928 (void) utils_set_fd_cloexec(sock);
1929
1930 /* Create UST registration command for enqueuing */
1931 ust_cmd = zmalloc(sizeof(struct ust_command));
1932 if (ust_cmd == NULL) {
1933 PERROR("ust command zmalloc");
1934 goto error;
1935 }
1936
1937 /*
1938 * Using message-based transmissions to ensure we don't
1939 * have to deal with partially received messages.
1940 */
1941 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1942 if (ret < 0) {
1943 ERR("Exhausted file descriptors allowed for applications.");
1944 free(ust_cmd);
1945 ret = close(sock);
1946 if (ret) {
1947 PERROR("close");
1948 }
1949 sock = -1;
1950 continue;
1951 }
1952
1953 health_code_update();
1954 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1955 if (ret < 0) {
1956 free(ust_cmd);
1957 /* Close socket of the application. */
1958 ret = close(sock);
1959 if (ret) {
1960 PERROR("close");
1961 }
1962 lttng_fd_put(LTTNG_FD_APPS, 1);
1963 sock = -1;
1964 continue;
1965 }
1966 health_code_update();
1967
1968 ust_cmd->sock = sock;
1969 sock = -1;
1970
1971 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1972 " gid:%d sock:%d name:%s (version %d.%d)",
1973 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1974 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1975 ust_cmd->sock, ust_cmd->reg_msg.name,
1976 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1977
1978 /*
1979 * Lock free enqueue the registration request. The red pill
1980 * has been taken! This apps will be part of the *system*.
1981 */
1982 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1983
1984 /*
1985 * Wake the registration queue futex. Implicit memory
1986 * barrier with the exchange in cds_wfq_enqueue.
1987 */
1988 futex_nto1_wake(&ust_cmd_queue.futex);
1989 }
1990 }
1991 }
1992 }
1993
1994 exit:
1995 error:
1996 if (err) {
1997 health_error();
1998 ERR("Health error occurred in %s", __func__);
1999 }
2000
2001 /* Notify that the registration thread is gone */
2002 notify_ust_apps(0);
2003
2004 if (apps_sock >= 0) {
2005 ret = close(apps_sock);
2006 if (ret) {
2007 PERROR("close");
2008 }
2009 }
2010 if (sock >= 0) {
2011 ret = close(sock);
2012 if (ret) {
2013 PERROR("close");
2014 }
2015 lttng_fd_put(LTTNG_FD_APPS, 1);
2016 }
2017 unlink(apps_unix_sock_path);
2018
2019 error_poll_add:
2020 lttng_poll_clean(&events);
2021 error_listen:
2022 error_create_poll:
2023 error_testpoint:
2024 DBG("UST Registration thread cleanup complete");
2025 health_unregister(health_sessiond);
2026
2027 return NULL;
2028 }
2029
2030 /*
2031 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2032 * exec or it will fails.
2033 */
2034 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2035 {
2036 int ret, clock_ret;
2037 struct timespec timeout;
2038
2039 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2040 consumer_data->consumer_thread_is_ready = 0;
2041
2042 /* Setup pthread condition */
2043 ret = pthread_condattr_init(&consumer_data->condattr);
2044 if (ret != 0) {
2045 errno = ret;
2046 PERROR("pthread_condattr_init consumer data");
2047 goto error;
2048 }
2049
2050 /*
2051 * Set the monotonic clock in order to make sure we DO NOT jump in time
2052 * between the clock_gettime() call and the timedwait call. See bug #324
2053 * for a more details and how we noticed it.
2054 */
2055 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2056 if (ret != 0) {
2057 errno = ret;
2058 PERROR("pthread_condattr_setclock consumer data");
2059 goto error;
2060 }
2061
2062 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2063 if (ret != 0) {
2064 errno = ret;
2065 PERROR("pthread_cond_init consumer data");
2066 goto error;
2067 }
2068
2069 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2070 consumer_data);
2071 if (ret != 0) {
2072 PERROR("pthread_create consumer");
2073 ret = -1;
2074 goto error;
2075 }
2076
2077 /* We are about to wait on a pthread condition */
2078 pthread_mutex_lock(&consumer_data->cond_mutex);
2079
2080 /* Get time for sem_timedwait absolute timeout */
2081 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2082 /*
2083 * Set the timeout for the condition timed wait even if the clock gettime
2084 * call fails since we might loop on that call and we want to avoid to
2085 * increment the timeout too many times.
2086 */
2087 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2088
2089 /*
2090 * The following loop COULD be skipped in some conditions so this is why we
2091 * set ret to 0 in order to make sure at least one round of the loop is
2092 * done.
2093 */
2094 ret = 0;
2095
2096 /*
2097 * Loop until the condition is reached or when a timeout is reached. Note
2098 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2099 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2100 * possible. This loop does not take any chances and works with both of
2101 * them.
2102 */
2103 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2104 if (clock_ret < 0) {
2105 PERROR("clock_gettime spawn consumer");
2106 /* Infinite wait for the consumerd thread to be ready */
2107 ret = pthread_cond_wait(&consumer_data->cond,
2108 &consumer_data->cond_mutex);
2109 } else {
2110 ret = pthread_cond_timedwait(&consumer_data->cond,
2111 &consumer_data->cond_mutex, &timeout);
2112 }
2113 }
2114
2115 /* Release the pthread condition */
2116 pthread_mutex_unlock(&consumer_data->cond_mutex);
2117
2118 if (ret != 0) {
2119 errno = ret;
2120 if (ret == ETIMEDOUT) {
2121 /*
2122 * Call has timed out so we kill the kconsumerd_thread and return
2123 * an error.
2124 */
2125 ERR("Condition timed out. The consumer thread was never ready."
2126 " Killing it");
2127 ret = pthread_cancel(consumer_data->thread);
2128 if (ret < 0) {
2129 PERROR("pthread_cancel consumer thread");
2130 }
2131 } else {
2132 PERROR("pthread_cond_wait failed consumer thread");
2133 }
2134 goto error;
2135 }
2136
2137 pthread_mutex_lock(&consumer_data->pid_mutex);
2138 if (consumer_data->pid == 0) {
2139 ERR("Consumerd did not start");
2140 pthread_mutex_unlock(&consumer_data->pid_mutex);
2141 goto error;
2142 }
2143 pthread_mutex_unlock(&consumer_data->pid_mutex);
2144
2145 return 0;
2146
2147 error:
2148 return ret;
2149 }
2150
2151 /*
2152 * Join consumer thread
2153 */
2154 static int join_consumer_thread(struct consumer_data *consumer_data)
2155 {
2156 void *status;
2157
2158 /* Consumer pid must be a real one. */
2159 if (consumer_data->pid > 0) {
2160 int ret;
2161 ret = kill(consumer_data->pid, SIGTERM);
2162 if (ret) {
2163 ERR("Error killing consumer daemon");
2164 return ret;
2165 }
2166 return pthread_join(consumer_data->thread, &status);
2167 } else {
2168 return 0;
2169 }
2170 }
2171
2172 /*
2173 * Fork and exec a consumer daemon (consumerd).
2174 *
2175 * Return pid if successful else -1.
2176 */
2177 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2178 {
2179 int ret;
2180 pid_t pid;
2181 const char *consumer_to_use;
2182 const char *verbosity;
2183 struct stat st;
2184
2185 DBG("Spawning consumerd");
2186
2187 pid = fork();
2188 if (pid == 0) {
2189 /*
2190 * Exec consumerd.
2191 */
2192 if (opt_verbose_consumer) {
2193 verbosity = "--verbose";
2194 } else {
2195 verbosity = "--quiet";
2196 }
2197 switch (consumer_data->type) {
2198 case LTTNG_CONSUMER_KERNEL:
2199 /*
2200 * Find out which consumerd to execute. We will first try the
2201 * 64-bit path, then the sessiond's installation directory, and
2202 * fallback on the 32-bit one,
2203 */
2204 DBG3("Looking for a kernel consumer at these locations:");
2205 DBG3(" 1) %s", consumerd64_bin);
2206 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2207 DBG3(" 3) %s", consumerd32_bin);
2208 if (stat(consumerd64_bin, &st) == 0) {
2209 DBG3("Found location #1");
2210 consumer_to_use = consumerd64_bin;
2211 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2212 DBG3("Found location #2");
2213 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2214 } else if (stat(consumerd32_bin, &st) == 0) {
2215 DBG3("Found location #3");
2216 consumer_to_use = consumerd32_bin;
2217 } else {
2218 DBG("Could not find any valid consumerd executable");
2219 break;
2220 }
2221 DBG("Using kernel consumer at: %s", consumer_to_use);
2222 execl(consumer_to_use,
2223 "lttng-consumerd", verbosity, "-k",
2224 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2225 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2226 "--group", tracing_group_name,
2227 NULL);
2228 break;
2229 case LTTNG_CONSUMER64_UST:
2230 {
2231 char *tmpnew = NULL;
2232
2233 if (consumerd64_libdir[0] != '\0') {
2234 char *tmp;
2235 size_t tmplen;
2236
2237 tmp = getenv("LD_LIBRARY_PATH");
2238 if (!tmp) {
2239 tmp = "";
2240 }
2241 tmplen = strlen("LD_LIBRARY_PATH=")
2242 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2243 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2244 if (!tmpnew) {
2245 ret = -ENOMEM;
2246 goto error;
2247 }
2248 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2249 strcat(tmpnew, consumerd64_libdir);
2250 if (tmp[0] != '\0') {
2251 strcat(tmpnew, ":");
2252 strcat(tmpnew, tmp);
2253 }
2254 ret = putenv(tmpnew);
2255 if (ret) {
2256 ret = -errno;
2257 free(tmpnew);
2258 goto error;
2259 }
2260 }
2261 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2262 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2263 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2264 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2265 "--group", tracing_group_name,
2266 NULL);
2267 if (consumerd64_libdir[0] != '\0') {
2268 free(tmpnew);
2269 }
2270 if (ret) {
2271 goto error;
2272 }
2273 break;
2274 }
2275 case LTTNG_CONSUMER32_UST:
2276 {
2277 char *tmpnew = NULL;
2278
2279 if (consumerd32_libdir[0] != '\0') {
2280 char *tmp;
2281 size_t tmplen;
2282
2283 tmp = getenv("LD_LIBRARY_PATH");
2284 if (!tmp) {
2285 tmp = "";
2286 }
2287 tmplen = strlen("LD_LIBRARY_PATH=")
2288 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2289 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2290 if (!tmpnew) {
2291 ret = -ENOMEM;
2292 goto error;
2293 }
2294 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2295 strcat(tmpnew, consumerd32_libdir);
2296 if (tmp[0] != '\0') {
2297 strcat(tmpnew, ":");
2298 strcat(tmpnew, tmp);
2299 }
2300 ret = putenv(tmpnew);
2301 if (ret) {
2302 ret = -errno;
2303 free(tmpnew);
2304 goto error;
2305 }
2306 }
2307 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2308 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2309 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2310 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2311 "--group", tracing_group_name,
2312 NULL);
2313 if (consumerd32_libdir[0] != '\0') {
2314 free(tmpnew);
2315 }
2316 if (ret) {
2317 goto error;
2318 }
2319 break;
2320 }
2321 default:
2322 PERROR("unknown consumer type");
2323 exit(EXIT_FAILURE);
2324 }
2325 if (errno != 0) {
2326 PERROR("kernel start consumer exec");
2327 }
2328 exit(EXIT_FAILURE);
2329 } else if (pid > 0) {
2330 ret = pid;
2331 } else {
2332 PERROR("start consumer fork");
2333 ret = -errno;
2334 }
2335 error:
2336 return ret;
2337 }
2338
2339 /*
2340 * Spawn the consumerd daemon and session daemon thread.
2341 */
2342 static int start_consumerd(struct consumer_data *consumer_data)
2343 {
2344 int ret;
2345
2346 /*
2347 * Set the listen() state on the socket since there is a possible race
2348 * between the exec() of the consumer daemon and this call if place in the
2349 * consumer thread. See bug #366 for more details.
2350 */
2351 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2352 if (ret < 0) {
2353 goto error;
2354 }
2355
2356 pthread_mutex_lock(&consumer_data->pid_mutex);
2357 if (consumer_data->pid != 0) {
2358 pthread_mutex_unlock(&consumer_data->pid_mutex);
2359 goto end;
2360 }
2361
2362 ret = spawn_consumerd(consumer_data);
2363 if (ret < 0) {
2364 ERR("Spawning consumerd failed");
2365 pthread_mutex_unlock(&consumer_data->pid_mutex);
2366 goto error;
2367 }
2368
2369 /* Setting up the consumer_data pid */
2370 consumer_data->pid = ret;
2371 DBG2("Consumer pid %d", consumer_data->pid);
2372 pthread_mutex_unlock(&consumer_data->pid_mutex);
2373
2374 DBG2("Spawning consumer control thread");
2375 ret = spawn_consumer_thread(consumer_data);
2376 if (ret < 0) {
2377 ERR("Fatal error spawning consumer control thread");
2378 goto error;
2379 }
2380
2381 end:
2382 return 0;
2383
2384 error:
2385 /* Cleanup already created sockets on error. */
2386 if (consumer_data->err_sock >= 0) {
2387 int err;
2388
2389 err = close(consumer_data->err_sock);
2390 if (err < 0) {
2391 PERROR("close consumer data error socket");
2392 }
2393 }
2394 return ret;
2395 }
2396
2397 /*
2398 * Setup necessary data for kernel tracer action.
2399 */
2400 static int init_kernel_tracer(void)
2401 {
2402 int ret;
2403
2404 /* Modprobe lttng kernel modules */
2405 ret = modprobe_lttng_control();
2406 if (ret < 0) {
2407 goto error;
2408 }
2409
2410 /* Open debugfs lttng */
2411 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2412 if (kernel_tracer_fd < 0) {
2413 DBG("Failed to open %s", module_proc_lttng);
2414 ret = -1;
2415 goto error_open;
2416 }
2417
2418 /* Validate kernel version */
2419 ret = kernel_validate_version(kernel_tracer_fd);
2420 if (ret < 0) {
2421 goto error_version;
2422 }
2423
2424 ret = modprobe_lttng_data();
2425 if (ret < 0) {
2426 goto error_modules;
2427 }
2428
2429 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2430 return 0;
2431
2432 error_version:
2433 modprobe_remove_lttng_control();
2434 ret = close(kernel_tracer_fd);
2435 if (ret) {
2436 PERROR("close");
2437 }
2438 kernel_tracer_fd = -1;
2439 return LTTNG_ERR_KERN_VERSION;
2440
2441 error_modules:
2442 ret = close(kernel_tracer_fd);
2443 if (ret) {
2444 PERROR("close");
2445 }
2446
2447 error_open:
2448 modprobe_remove_lttng_control();
2449
2450 error:
2451 WARN("No kernel tracer available");
2452 kernel_tracer_fd = -1;
2453 if (!is_root) {
2454 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2455 } else {
2456 return LTTNG_ERR_KERN_NA;
2457 }
2458 }
2459
2460
2461 /*
2462 * Copy consumer output from the tracing session to the domain session. The
2463 * function also applies the right modification on a per domain basis for the
2464 * trace files destination directory.
2465 *
2466 * Should *NOT* be called with RCU read-side lock held.
2467 */
2468 static int copy_session_consumer(int domain, struct ltt_session *session)
2469 {
2470 int ret;
2471 const char *dir_name;
2472 struct consumer_output *consumer;
2473
2474 assert(session);
2475 assert(session->consumer);
2476
2477 switch (domain) {
2478 case LTTNG_DOMAIN_KERNEL:
2479 DBG3("Copying tracing session consumer output in kernel session");
2480 /*
2481 * XXX: We should audit the session creation and what this function
2482 * does "extra" in order to avoid a destroy since this function is used
2483 * in the domain session creation (kernel and ust) only. Same for UST
2484 * domain.
2485 */
2486 if (session->kernel_session->consumer) {
2487 consumer_destroy_output(session->kernel_session->consumer);
2488 }
2489 session->kernel_session->consumer =
2490 consumer_copy_output(session->consumer);
2491 /* Ease our life a bit for the next part */
2492 consumer = session->kernel_session->consumer;
2493 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2494 break;
2495 case LTTNG_DOMAIN_JUL:
2496 case LTTNG_DOMAIN_UST:
2497 DBG3("Copying tracing session consumer output in UST session");
2498 if (session->ust_session->consumer) {
2499 consumer_destroy_output(session->ust_session->consumer);
2500 }
2501 session->ust_session->consumer =
2502 consumer_copy_output(session->consumer);
2503 /* Ease our life a bit for the next part */
2504 consumer = session->ust_session->consumer;
2505 dir_name = DEFAULT_UST_TRACE_DIR;
2506 break;
2507 default:
2508 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2509 goto error;
2510 }
2511
2512 /* Append correct directory to subdir */
2513 strncat(consumer->subdir, dir_name,
2514 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2515 DBG3("Copy session consumer subdir %s", consumer->subdir);
2516
2517 ret = LTTNG_OK;
2518
2519 error:
2520 return ret;
2521 }
2522
2523 /*
2524 * Create an UST session and add it to the session ust list.
2525 *
2526 * Should *NOT* be called with RCU read-side lock held.
2527 */
2528 static int create_ust_session(struct ltt_session *session,
2529 struct lttng_domain *domain)
2530 {
2531 int ret;
2532 struct ltt_ust_session *lus = NULL;
2533
2534 assert(session);
2535 assert(domain);
2536 assert(session->consumer);
2537
2538 switch (domain->type) {
2539 case LTTNG_DOMAIN_JUL:
2540 case LTTNG_DOMAIN_UST:
2541 break;
2542 default:
2543 ERR("Unknown UST domain on create session %d", domain->type);
2544 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2545 goto error;
2546 }
2547
2548 DBG("Creating UST session");
2549
2550 lus = trace_ust_create_session(session->id);
2551 if (lus == NULL) {
2552 ret = LTTNG_ERR_UST_SESS_FAIL;
2553 goto error;
2554 }
2555
2556 lus->uid = session->uid;
2557 lus->gid = session->gid;
2558 lus->output_traces = session->output_traces;
2559 lus->snapshot_mode = session->snapshot_mode;
2560 lus->live_timer_interval = session->live_timer;
2561 session->ust_session = lus;
2562
2563 /* Copy session output to the newly created UST session */
2564 ret = copy_session_consumer(domain->type, session);
2565 if (ret != LTTNG_OK) {
2566 goto error;
2567 }
2568
2569 return LTTNG_OK;
2570
2571 error:
2572 free(lus);
2573 session->ust_session = NULL;
2574 return ret;
2575 }
2576
2577 /*
2578 * Create a kernel tracer session then create the default channel.
2579 */
2580 static int create_kernel_session(struct ltt_session *session)
2581 {
2582 int ret;
2583
2584 DBG("Creating kernel session");
2585
2586 ret = kernel_create_session(session, kernel_tracer_fd);
2587 if (ret < 0) {
2588 ret = LTTNG_ERR_KERN_SESS_FAIL;
2589 goto error;
2590 }
2591
2592 /* Code flow safety */
2593 assert(session->kernel_session);
2594
2595 /* Copy session output to the newly created Kernel session */
2596 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2597 if (ret != LTTNG_OK) {
2598 goto error;
2599 }
2600
2601 /* Create directory(ies) on local filesystem. */
2602 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2603 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2604 ret = run_as_mkdir_recursive(
2605 session->kernel_session->consumer->dst.trace_path,
2606 S_IRWXU | S_IRWXG, session->uid, session->gid);
2607 if (ret < 0) {
2608 if (ret != -EEXIST) {
2609 ERR("Trace directory creation error");
2610 goto error;
2611 }
2612 }
2613 }
2614
2615 session->kernel_session->uid = session->uid;
2616 session->kernel_session->gid = session->gid;
2617 session->kernel_session->output_traces = session->output_traces;
2618 session->kernel_session->snapshot_mode = session->snapshot_mode;
2619
2620 return LTTNG_OK;
2621
2622 error:
2623 trace_kernel_destroy_session(session->kernel_session);
2624 session->kernel_session = NULL;
2625 return ret;
2626 }
2627
2628 /*
2629 * Count number of session permitted by uid/gid.
2630 */
2631 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2632 {
2633 unsigned int i = 0;
2634 struct ltt_session *session;
2635
2636 DBG("Counting number of available session for UID %d GID %d",
2637 uid, gid);
2638 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2639 /*
2640 * Only list the sessions the user can control.
2641 */
2642 if (!session_access_ok(session, uid, gid)) {
2643 continue;
2644 }
2645 i++;
2646 }
2647 return i;
2648 }
2649
2650 /*
2651 * Process the command requested by the lttng client within the command
2652 * context structure. This function make sure that the return structure (llm)
2653 * is set and ready for transmission before returning.
2654 *
2655 * Return any error encountered or 0 for success.
2656 *
2657 * "sock" is only used for special-case var. len data.
2658 *
2659 * Should *NOT* be called with RCU read-side lock held.
2660 */
2661 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2662 int *sock_error)
2663 {
2664 int ret = LTTNG_OK;
2665 int need_tracing_session = 1;
2666 int need_domain;
2667
2668 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2669
2670 *sock_error = 0;
2671
2672 switch (cmd_ctx->lsm->cmd_type) {
2673 case LTTNG_CREATE_SESSION:
2674 case LTTNG_CREATE_SESSION_SNAPSHOT:
2675 case LTTNG_CREATE_SESSION_LIVE:
2676 case LTTNG_DESTROY_SESSION:
2677 case LTTNG_LIST_SESSIONS:
2678 case LTTNG_LIST_DOMAINS:
2679 case LTTNG_START_TRACE:
2680 case LTTNG_STOP_TRACE:
2681 case LTTNG_DATA_PENDING:
2682 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2683 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2684 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2685 case LTTNG_SNAPSHOT_RECORD:
2686 need_domain = 0;
2687 break;
2688 default:
2689 need_domain = 1;
2690 }
2691
2692 if (opt_no_kernel && need_domain
2693 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2694 if (!is_root) {
2695 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2696 } else {
2697 ret = LTTNG_ERR_KERN_NA;
2698 }
2699 goto error;
2700 }
2701
2702 /* Deny register consumer if we already have a spawned consumer. */
2703 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2704 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2705 if (kconsumer_data.pid > 0) {
2706 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2707 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2708 goto error;
2709 }
2710 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2711 }
2712
2713 /*
2714 * Check for command that don't needs to allocate a returned payload. We do
2715 * this here so we don't have to make the call for no payload at each
2716 * command.
2717 */
2718 switch(cmd_ctx->lsm->cmd_type) {
2719 case LTTNG_LIST_SESSIONS:
2720 case LTTNG_LIST_TRACEPOINTS:
2721 case LTTNG_LIST_TRACEPOINT_FIELDS:
2722 case LTTNG_LIST_DOMAINS:
2723 case LTTNG_LIST_CHANNELS:
2724 case LTTNG_LIST_EVENTS:
2725 break;
2726 default:
2727 /* Setup lttng message with no payload */
2728 ret = setup_lttng_msg(cmd_ctx, 0);
2729 if (ret < 0) {
2730 /* This label does not try to unlock the session */
2731 goto init_setup_error;
2732 }
2733 }
2734
2735 /* Commands that DO NOT need a session. */
2736 switch (cmd_ctx->lsm->cmd_type) {
2737 case LTTNG_CREATE_SESSION:
2738 case LTTNG_CREATE_SESSION_SNAPSHOT:
2739 case LTTNG_CREATE_SESSION_LIVE:
2740 case LTTNG_CALIBRATE:
2741 case LTTNG_LIST_SESSIONS:
2742 case LTTNG_LIST_TRACEPOINTS:
2743 case LTTNG_LIST_TRACEPOINT_FIELDS:
2744 need_tracing_session = 0;
2745 break;
2746 default:
2747 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2748 /*
2749 * We keep the session list lock across _all_ commands
2750 * for now, because the per-session lock does not
2751 * handle teardown properly.
2752 */
2753 session_lock_list();
2754 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2755 if (cmd_ctx->session == NULL) {
2756 ret = LTTNG_ERR_SESS_NOT_FOUND;
2757 goto error;
2758 } else {
2759 /* Acquire lock for the session */
2760 session_lock(cmd_ctx->session);
2761 }
2762 break;
2763 }
2764
2765 if (!need_domain) {
2766 goto skip_domain;
2767 }
2768
2769 /*
2770 * Check domain type for specific "pre-action".
2771 */
2772 switch (cmd_ctx->lsm->domain.type) {
2773 case LTTNG_DOMAIN_KERNEL:
2774 if (!is_root) {
2775 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2776 goto error;
2777 }
2778
2779 /* Kernel tracer check */
2780 if (kernel_tracer_fd == -1) {
2781 /* Basically, load kernel tracer modules */
2782 ret = init_kernel_tracer();
2783 if (ret != 0) {
2784 goto error;
2785 }
2786 }
2787
2788 /* Consumer is in an ERROR state. Report back to client */
2789 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2790 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2791 goto error;
2792 }
2793
2794 /* Need a session for kernel command */
2795 if (need_tracing_session) {
2796 if (cmd_ctx->session->kernel_session == NULL) {
2797 ret = create_kernel_session(cmd_ctx->session);
2798 if (ret < 0) {
2799 ret = LTTNG_ERR_KERN_SESS_FAIL;
2800 goto error;
2801 }
2802 }
2803
2804 /* Start the kernel consumer daemon */
2805 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2806 if (kconsumer_data.pid == 0 &&
2807 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2808 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2809 ret = start_consumerd(&kconsumer_data);
2810 if (ret < 0) {
2811 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2812 goto error;
2813 }
2814 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2815 } else {
2816 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2817 }
2818
2819 /*
2820 * The consumer was just spawned so we need to add the socket to
2821 * the consumer output of the session if exist.
2822 */
2823 ret = consumer_create_socket(&kconsumer_data,
2824 cmd_ctx->session->kernel_session->consumer);
2825 if (ret < 0) {
2826 goto error;
2827 }
2828 }
2829
2830 break;
2831 case LTTNG_DOMAIN_JUL:
2832 case LTTNG_DOMAIN_UST:
2833 {
2834 if (!ust_app_supported()) {
2835 ret = LTTNG_ERR_NO_UST;
2836 goto error;
2837 }
2838 /* Consumer is in an ERROR state. Report back to client */
2839 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2840 ret = LTTNG_ERR_NO_USTCONSUMERD;
2841 goto error;
2842 }
2843
2844 if (need_tracing_session) {
2845 /* Create UST session if none exist. */
2846 if (cmd_ctx->session->ust_session == NULL) {
2847 ret = create_ust_session(cmd_ctx->session,
2848 &cmd_ctx->lsm->domain);
2849 if (ret != LTTNG_OK) {
2850 goto error;
2851 }
2852 }
2853
2854 /* Start the UST consumer daemons */
2855 /* 64-bit */
2856 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2857 if (consumerd64_bin[0] != '\0' &&
2858 ustconsumer64_data.pid == 0 &&
2859 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2860 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2861 ret = start_consumerd(&ustconsumer64_data);
2862 if (ret < 0) {
2863 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2864 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2865 goto error;
2866 }
2867
2868 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2869 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2870 } else {
2871 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2872 }
2873
2874 /*
2875 * Setup socket for consumer 64 bit. No need for atomic access
2876 * since it was set above and can ONLY be set in this thread.
2877 */
2878 ret = consumer_create_socket(&ustconsumer64_data,
2879 cmd_ctx->session->ust_session->consumer);
2880 if (ret < 0) {
2881 goto error;
2882 }
2883
2884 /* 32-bit */
2885 if (consumerd32_bin[0] != '\0' &&
2886 ustconsumer32_data.pid == 0 &&
2887 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2888 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2889 ret = start_consumerd(&ustconsumer32_data);
2890 if (ret < 0) {
2891 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2892 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2893 goto error;
2894 }
2895
2896 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2897 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2898 } else {
2899 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2900 }
2901
2902 /*
2903 * Setup socket for consumer 64 bit. No need for atomic access
2904 * since it was set above and can ONLY be set in this thread.
2905 */
2906 ret = consumer_create_socket(&ustconsumer32_data,
2907 cmd_ctx->session->ust_session->consumer);
2908 if (ret < 0) {
2909 goto error;
2910 }
2911 }
2912 break;
2913 }
2914 default:
2915 break;
2916 }
2917 skip_domain:
2918
2919 /* Validate consumer daemon state when start/stop trace command */
2920 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2921 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2922 switch (cmd_ctx->lsm->domain.type) {
2923 case LTTNG_DOMAIN_JUL:
2924 case LTTNG_DOMAIN_UST:
2925 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2926 ret = LTTNG_ERR_NO_USTCONSUMERD;
2927 goto error;
2928 }
2929 break;
2930 case LTTNG_DOMAIN_KERNEL:
2931 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2932 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2933 goto error;
2934 }
2935 break;
2936 }
2937 }
2938
2939 /*
2940 * Check that the UID or GID match that of the tracing session.
2941 * The root user can interact with all sessions.
2942 */
2943 if (need_tracing_session) {
2944 if (!session_access_ok(cmd_ctx->session,
2945 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2946 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2947 ret = LTTNG_ERR_EPERM;
2948 goto error;
2949 }
2950 }
2951
2952 /*
2953 * Send relayd information to consumer as soon as we have a domain and a
2954 * session defined.
2955 */
2956 if (cmd_ctx->session && need_domain) {
2957 /*
2958 * Setup relayd if not done yet. If the relayd information was already
2959 * sent to the consumer, this call will gracefully return.
2960 */
2961 ret = cmd_setup_relayd(cmd_ctx->session);
2962 if (ret != LTTNG_OK) {
2963 goto error;
2964 }
2965 }
2966
2967 /* Process by command type */
2968 switch (cmd_ctx->lsm->cmd_type) {
2969 case LTTNG_ADD_CONTEXT:
2970 {
2971 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2972 cmd_ctx->lsm->u.context.channel_name,
2973 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2974 break;
2975 }
2976 case LTTNG_DISABLE_CHANNEL:
2977 {
2978 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2979 cmd_ctx->lsm->u.disable.channel_name);
2980 break;
2981 }
2982 case LTTNG_DISABLE_EVENT:
2983 {
2984 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2985 cmd_ctx->lsm->u.disable.channel_name,
2986 cmd_ctx->lsm->u.disable.name);
2987 break;
2988 }
2989 case LTTNG_DISABLE_ALL_EVENT:
2990 {
2991 DBG("Disabling all events");
2992
2993 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2994 cmd_ctx->lsm->u.disable.channel_name);
2995 break;
2996 }
2997 case LTTNG_ENABLE_CHANNEL:
2998 {
2999 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3000 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3001 break;
3002 }
3003 case LTTNG_ENABLE_EVENT:
3004 {
3005 struct lttng_event_exclusion *exclusion = NULL;
3006 struct lttng_filter_bytecode *bytecode = NULL;
3007
3008 /* Handle exclusion events and receive it from the client. */
3009 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3010 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3011
3012 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3013 (count * LTTNG_SYMBOL_NAME_LEN));
3014 if (!exclusion) {
3015 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3016 goto error;
3017 }
3018
3019 DBG("Receiving var len exclusion event list from client ...");
3020 exclusion->count = count;
3021 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3022 count * LTTNG_SYMBOL_NAME_LEN);
3023 if (ret <= 0) {
3024 DBG("Nothing recv() from client var len data... continuing");
3025 *sock_error = 1;
3026 free(exclusion);
3027 ret = LTTNG_ERR_EXCLUSION_INVAL;
3028 goto error;
3029 }
3030 }
3031
3032 /* Handle filter and get bytecode from client. */
3033 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3034 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3035
3036 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3037 ret = LTTNG_ERR_FILTER_INVAL;
3038 free(exclusion);
3039 goto error;
3040 }
3041
3042 bytecode = zmalloc(bytecode_len);
3043 if (!bytecode) {
3044 free(exclusion);
3045 ret = LTTNG_ERR_FILTER_NOMEM;
3046 goto error;
3047 }
3048
3049 /* Receive var. len. data */
3050 DBG("Receiving var len filter's bytecode from client ...");
3051 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3052 if (ret <= 0) {
3053 DBG("Nothing recv() from client car len data... continuing");
3054 *sock_error = 1;
3055 free(bytecode);
3056 free(exclusion);
3057 ret = LTTNG_ERR_FILTER_INVAL;
3058 goto error;
3059 }
3060
3061 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3062 free(bytecode);
3063 free(exclusion);
3064 ret = LTTNG_ERR_FILTER_INVAL;
3065 goto error;
3066 }
3067 }
3068
3069 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3070 cmd_ctx->lsm->u.enable.channel_name,
3071 &cmd_ctx->lsm->u.enable.event, bytecode, exclusion,
3072 kernel_poll_pipe[1]);
3073 break;
3074 }
3075 case LTTNG_ENABLE_ALL_EVENT:
3076 {
3077 DBG("Enabling all events");
3078
3079 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
3080 cmd_ctx->lsm->u.enable.channel_name,
3081 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
3082 break;
3083 }
3084 case LTTNG_LIST_TRACEPOINTS:
3085 {
3086 struct lttng_event *events;
3087 ssize_t nb_events;
3088
3089 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3090 if (nb_events < 0) {
3091 /* Return value is a negative lttng_error_code. */
3092 ret = -nb_events;
3093 goto error;
3094 }
3095
3096 /*
3097 * Setup lttng message with payload size set to the event list size in
3098 * bytes and then copy list into the llm payload.
3099 */
3100 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3101 if (ret < 0) {
3102 free(events);
3103 goto setup_error;
3104 }
3105
3106 /* Copy event list into message payload */
3107 memcpy(cmd_ctx->llm->payload, events,
3108 sizeof(struct lttng_event) * nb_events);
3109
3110 free(events);
3111
3112 ret = LTTNG_OK;
3113 break;
3114 }
3115 case LTTNG_LIST_TRACEPOINT_FIELDS:
3116 {
3117 struct lttng_event_field *fields;
3118 ssize_t nb_fields;
3119
3120 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3121 &fields);
3122 if (nb_fields < 0) {
3123 /* Return value is a negative lttng_error_code. */
3124 ret = -nb_fields;
3125 goto error;
3126 }
3127
3128 /*
3129 * Setup lttng message with payload size set to the event list size in
3130 * bytes and then copy list into the llm payload.
3131 */
3132 ret = setup_lttng_msg(cmd_ctx,
3133 sizeof(struct lttng_event_field) * nb_fields);
3134 if (ret < 0) {
3135 free(fields);
3136 goto setup_error;
3137 }
3138
3139 /* Copy event list into message payload */
3140 memcpy(cmd_ctx->llm->payload, fields,
3141 sizeof(struct lttng_event_field) * nb_fields);
3142
3143 free(fields);
3144
3145 ret = LTTNG_OK;
3146 break;
3147 }
3148 case LTTNG_SET_CONSUMER_URI:
3149 {
3150 size_t nb_uri, len;
3151 struct lttng_uri *uris;
3152
3153 nb_uri = cmd_ctx->lsm->u.uri.size;
3154 len = nb_uri * sizeof(struct lttng_uri);
3155
3156 if (nb_uri == 0) {
3157 ret = LTTNG_ERR_INVALID;
3158 goto error;
3159 }
3160
3161 uris = zmalloc(len);
3162 if (uris == NULL) {
3163 ret = LTTNG_ERR_FATAL;
3164 goto error;
3165 }
3166
3167 /* Receive variable len data */
3168 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3169 ret = lttcomm_recv_unix_sock(sock, uris, len);
3170 if (ret <= 0) {
3171 DBG("No URIs received from client... continuing");
3172 *sock_error = 1;
3173 ret = LTTNG_ERR_SESSION_FAIL;
3174 free(uris);
3175 goto error;
3176 }
3177
3178 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3179 nb_uri, uris);
3180 if (ret != LTTNG_OK) {
3181 free(uris);
3182 goto error;
3183 }
3184
3185 /*
3186 * XXX: 0 means that this URI should be applied on the session. Should
3187 * be a DOMAIN enuam.
3188 */
3189 if (cmd_ctx->lsm->domain.type == 0) {
3190 /* Add the URI for the UST session if a consumer is present. */
3191 if (cmd_ctx->session->ust_session &&
3192 cmd_ctx->session->ust_session->consumer) {
3193 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3194 nb_uri, uris);
3195 } else if (cmd_ctx->session->kernel_session &&
3196 cmd_ctx->session->kernel_session->consumer) {
3197 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3198 cmd_ctx->session, nb_uri, uris);
3199 }
3200 }
3201
3202 free(uris);
3203
3204 break;
3205 }
3206 case LTTNG_START_TRACE:
3207 {
3208 ret = cmd_start_trace(cmd_ctx->session);
3209 break;
3210 }
3211 case LTTNG_STOP_TRACE:
3212 {
3213 ret = cmd_stop_trace(cmd_ctx->session);
3214 break;
3215 }
3216 case LTTNG_CREATE_SESSION:
3217 {
3218 size_t nb_uri, len;
3219 struct lttng_uri *uris = NULL;
3220
3221 nb_uri = cmd_ctx->lsm->u.uri.size;
3222 len = nb_uri * sizeof(struct lttng_uri);
3223
3224 if (nb_uri > 0) {
3225 uris = zmalloc(len);
3226 if (uris == NULL) {
3227 ret = LTTNG_ERR_FATAL;
3228 goto error;
3229 }
3230
3231 /* Receive variable len data */
3232 DBG("Waiting for %zu URIs from client ...", nb_uri);
3233 ret = lttcomm_recv_unix_sock(sock, uris, len);
3234 if (ret <= 0) {
3235 DBG("No URIs received from client... continuing");
3236 *sock_error = 1;
3237 ret = LTTNG_ERR_SESSION_FAIL;
3238 free(uris);
3239 goto error;
3240 }
3241
3242 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3243 DBG("Creating session with ONE network URI is a bad call");
3244 ret = LTTNG_ERR_SESSION_FAIL;
3245 free(uris);
3246 goto error;
3247 }
3248 }
3249
3250 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3251 &cmd_ctx->creds, 0);
3252
3253 free(uris);
3254
3255 break;
3256 }
3257 case LTTNG_DESTROY_SESSION:
3258 {
3259 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3260
3261 /* Set session to NULL so we do not unlock it after free. */
3262 cmd_ctx->session = NULL;
3263 break;
3264 }
3265 case LTTNG_LIST_DOMAINS:
3266 {
3267 ssize_t nb_dom;
3268 struct lttng_domain *domains;
3269
3270 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3271 if (nb_dom < 0) {
3272 /* Return value is a negative lttng_error_code. */
3273 ret = -nb_dom;
3274 goto error;
3275 }
3276
3277 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3278 if (ret < 0) {
3279 free(domains);
3280 goto setup_error;
3281 }
3282
3283 /* Copy event list into message payload */
3284 memcpy(cmd_ctx->llm->payload, domains,
3285 nb_dom * sizeof(struct lttng_domain));
3286
3287 free(domains);
3288
3289 ret = LTTNG_OK;
3290 break;
3291 }
3292 case LTTNG_LIST_CHANNELS:
3293 {
3294 int nb_chan;
3295 struct lttng_channel *channels;
3296
3297 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3298 cmd_ctx->session, &channels);
3299 if (nb_chan < 0) {
3300 /* Return value is a negative lttng_error_code. */
3301 ret = -nb_chan;
3302 goto error;
3303 }
3304
3305 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3306 if (ret < 0) {
3307 free(channels);
3308 goto setup_error;
3309 }
3310
3311 /* Copy event list into message payload */
3312 memcpy(cmd_ctx->llm->payload, channels,
3313 nb_chan * sizeof(struct lttng_channel));
3314
3315 free(channels);
3316
3317 ret = LTTNG_OK;
3318 break;
3319 }
3320 case LTTNG_LIST_EVENTS:
3321 {
3322 ssize_t nb_event;
3323 struct lttng_event *events = NULL;
3324
3325 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3326 cmd_ctx->lsm->u.list.channel_name, &events);
3327 if (nb_event < 0) {
3328 /* Return value is a negative lttng_error_code. */
3329 ret = -nb_event;
3330 goto error;
3331 }
3332
3333 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3334 if (ret < 0) {
3335 free(events);
3336 goto setup_error;
3337 }
3338
3339 /* Copy event list into message payload */
3340 memcpy(cmd_ctx->llm->payload, events,
3341 nb_event * sizeof(struct lttng_event));
3342
3343 free(events);
3344
3345 ret = LTTNG_OK;
3346 break;
3347 }
3348 case LTTNG_LIST_SESSIONS:
3349 {
3350 unsigned int nr_sessions;
3351
3352 session_lock_list();
3353 nr_sessions = lttng_sessions_count(
3354 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3355 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3356
3357 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3358 if (ret < 0) {
3359 session_unlock_list();
3360 goto setup_error;
3361 }
3362
3363 /* Filled the session array */
3364 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3365 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3366 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3367
3368 session_unlock_list();
3369
3370 ret = LTTNG_OK;
3371 break;
3372 }
3373 case LTTNG_CALIBRATE:
3374 {
3375 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3376 &cmd_ctx->lsm->u.calibrate);
3377 break;
3378 }
3379 case LTTNG_REGISTER_CONSUMER:
3380 {
3381 struct consumer_data *cdata;
3382
3383 switch (cmd_ctx->lsm->domain.type) {
3384 case LTTNG_DOMAIN_KERNEL:
3385 cdata = &kconsumer_data;
3386 break;
3387 default:
3388 ret = LTTNG_ERR_UND;
3389 goto error;
3390 }
3391
3392 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3393 cmd_ctx->lsm->u.reg.path, cdata);
3394 break;
3395 }
3396 case LTTNG_DATA_PENDING:
3397 {
3398 ret = cmd_data_pending(cmd_ctx->session);
3399 break;
3400 }
3401 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3402 {
3403 struct lttcomm_lttng_output_id reply;
3404
3405 ret = cmd_snapshot_add_output(cmd_ctx->session,
3406 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3407 if (ret != LTTNG_OK) {
3408 goto error;
3409 }
3410
3411 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3412 if (ret < 0) {
3413 goto setup_error;
3414 }
3415
3416 /* Copy output list into message payload */
3417 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3418 ret = LTTNG_OK;
3419 break;
3420 }
3421 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3422 {
3423 ret = cmd_snapshot_del_output(cmd_ctx->session,
3424 &cmd_ctx->lsm->u.snapshot_output.output);
3425 break;
3426 }
3427 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3428 {
3429 ssize_t nb_output;
3430 struct lttng_snapshot_output *outputs = NULL;
3431
3432 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3433 if (nb_output < 0) {
3434 ret = -nb_output;
3435 goto error;
3436 }
3437
3438 ret = setup_lttng_msg(cmd_ctx,
3439 nb_output * sizeof(struct lttng_snapshot_output));
3440 if (ret < 0) {
3441 free(outputs);
3442 goto setup_error;
3443 }
3444
3445 if (outputs) {
3446 /* Copy output list into message payload */
3447 memcpy(cmd_ctx->llm->payload, outputs,
3448 nb_output * sizeof(struct lttng_snapshot_output));
3449 free(outputs);
3450 }
3451
3452 ret = LTTNG_OK;
3453 break;
3454 }
3455 case LTTNG_SNAPSHOT_RECORD:
3456 {
3457 ret = cmd_snapshot_record(cmd_ctx->session,
3458 &cmd_ctx->lsm->u.snapshot_record.output,
3459 cmd_ctx->lsm->u.snapshot_record.wait);
3460 break;
3461 }
3462 case LTTNG_CREATE_SESSION_SNAPSHOT:
3463 {
3464 size_t nb_uri, len;
3465 struct lttng_uri *uris = NULL;
3466
3467 nb_uri = cmd_ctx->lsm->u.uri.size;
3468 len = nb_uri * sizeof(struct lttng_uri);
3469
3470 if (nb_uri > 0) {
3471 uris = zmalloc(len);
3472 if (uris == NULL) {
3473 ret = LTTNG_ERR_FATAL;
3474 goto error;
3475 }
3476
3477 /* Receive variable len data */
3478 DBG("Waiting for %zu URIs from client ...", nb_uri);
3479 ret = lttcomm_recv_unix_sock(sock, uris, len);
3480 if (ret <= 0) {
3481 DBG("No URIs received from client... continuing");
3482 *sock_error = 1;
3483 ret = LTTNG_ERR_SESSION_FAIL;
3484 free(uris);
3485 goto error;
3486 }
3487
3488 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3489 DBG("Creating session with ONE network URI is a bad call");
3490 ret = LTTNG_ERR_SESSION_FAIL;
3491 free(uris);
3492 goto error;
3493 }
3494 }
3495
3496 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3497 nb_uri, &cmd_ctx->creds);
3498 free(uris);
3499 break;
3500 }
3501 case LTTNG_CREATE_SESSION_LIVE:
3502 {
3503 size_t nb_uri, len;
3504 struct lttng_uri *uris = NULL;
3505
3506 nb_uri = cmd_ctx->lsm->u.uri.size;
3507 len = nb_uri * sizeof(struct lttng_uri);
3508
3509 if (nb_uri > 0) {
3510 uris = zmalloc(len);
3511 if (uris == NULL) {
3512 ret = LTTNG_ERR_FATAL;
3513 goto error;
3514 }
3515
3516 /* Receive variable len data */
3517 DBG("Waiting for %zu URIs from client ...", nb_uri);
3518 ret = lttcomm_recv_unix_sock(sock, uris, len);
3519 if (ret <= 0) {
3520 DBG("No URIs received from client... continuing");
3521 *sock_error = 1;
3522 ret = LTTNG_ERR_SESSION_FAIL;
3523 free(uris);
3524 goto error;
3525 }
3526
3527 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3528 DBG("Creating session with ONE network URI is a bad call");
3529 ret = LTTNG_ERR_SESSION_FAIL;
3530 free(uris);
3531 goto error;
3532 }
3533 }
3534
3535 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3536 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3537 free(uris);
3538 break;
3539 }
3540 default:
3541 ret = LTTNG_ERR_UND;
3542 break;
3543 }
3544
3545 error:
3546 if (cmd_ctx->llm == NULL) {
3547 DBG("Missing llm structure. Allocating one.");
3548 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3549 goto setup_error;
3550 }
3551 }
3552 /* Set return code */
3553 cmd_ctx->llm->ret_code = ret;
3554 setup_error:
3555 if (cmd_ctx->session) {
3556 session_unlock(cmd_ctx->session);
3557 }
3558 if (need_tracing_session) {
3559 session_unlock_list();
3560 }
3561 init_setup_error:
3562 return ret;
3563 }
3564
3565 /*
3566 * Thread managing health check socket.
3567 */
3568 static void *thread_manage_health(void *data)
3569 {
3570 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3571 uint32_t revents, nb_fd;
3572 struct lttng_poll_event events;
3573 struct health_comm_msg msg;
3574 struct health_comm_reply reply;
3575
3576 DBG("[thread] Manage health check started");
3577
3578 rcu_register_thread();
3579
3580 /* We might hit an error path before this is created. */
3581 lttng_poll_init(&events);
3582
3583 /* Create unix socket */
3584 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3585 if (sock < 0) {
3586 ERR("Unable to create health check Unix socket");
3587 ret = -1;
3588 goto error;
3589 }
3590
3591 if (is_root) {
3592 /* lttng health client socket path permissions */
3593 ret = chown(health_unix_sock_path, 0,
3594 utils_get_group_id(tracing_group_name));
3595 if (ret < 0) {
3596 ERR("Unable to set group on %s", health_unix_sock_path);
3597 PERROR("chown");
3598 ret = -1;
3599 goto error;
3600 }
3601
3602 ret = chmod(health_unix_sock_path,
3603 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3604 if (ret < 0) {
3605 ERR("Unable to set permissions on %s", health_unix_sock_path);
3606 PERROR("chmod");
3607 ret = -1;
3608 goto error;
3609 }
3610 }
3611
3612 /*
3613 * Set the CLOEXEC flag. Return code is useless because either way, the
3614 * show must go on.
3615 */
3616 (void) utils_set_fd_cloexec(sock);
3617
3618 ret = lttcomm_listen_unix_sock(sock);
3619 if (ret < 0) {
3620 goto error;
3621 }
3622
3623 /*
3624 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3625 * more will be added to this poll set.
3626 */
3627 ret = sessiond_set_thread_pollset(&events, 2);
3628 if (ret < 0) {
3629 goto error;
3630 }
3631
3632 /* Add the application registration socket */
3633 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3634 if (ret < 0) {
3635 goto error;
3636 }
3637
3638 while (1) {
3639 DBG("Health check ready");
3640
3641 /* Inifinite blocking call, waiting for transmission */
3642 restart:
3643 ret = lttng_poll_wait(&events, -1);
3644 if (ret < 0) {
3645 /*
3646 * Restart interrupted system call.
3647 */
3648 if (errno == EINTR) {
3649 goto restart;
3650 }
3651 goto error;
3652 }
3653
3654 nb_fd = ret;
3655
3656 for (i = 0; i < nb_fd; i++) {
3657 /* Fetch once the poll data */
3658 revents = LTTNG_POLL_GETEV(&events, i);
3659 pollfd = LTTNG_POLL_GETFD(&events, i);
3660
3661 /* Thread quit pipe has been closed. Killing thread. */
3662 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3663 if (ret) {
3664 err = 0;
3665 goto exit;
3666 }
3667
3668 /* Event on the registration socket */
3669 if (pollfd == sock) {
3670 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3671 ERR("Health socket poll error");
3672 goto error;
3673 }
3674 }
3675 }
3676
3677 new_sock = lttcomm_accept_unix_sock(sock);
3678 if (new_sock < 0) {
3679 goto error;
3680 }
3681
3682 /*
3683 * Set the CLOEXEC flag. Return code is useless because either way, the
3684 * show must go on.
3685 */
3686 (void) utils_set_fd_cloexec(new_sock);
3687
3688 DBG("Receiving data from client for health...");
3689 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3690 if (ret <= 0) {
3691 DBG("Nothing recv() from client... continuing");
3692 ret = close(new_sock);
3693 if (ret) {
3694 PERROR("close");
3695 }
3696 new_sock = -1;
3697 continue;
3698 }
3699
3700 rcu_thread_online();
3701
3702 reply.ret_code = 0;
3703 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3704 /*
3705 * health_check_state returns 0 if health is
3706 * bad.
3707 */
3708 if (!health_check_state(health_sessiond, i)) {
3709 reply.ret_code |= 1ULL << i;
3710 }
3711 }
3712
3713 DBG2("Health check return value %" PRIx64, reply.ret_code);
3714
3715 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3716 if (ret < 0) {
3717 ERR("Failed to send health data back to client");
3718 }
3719
3720 /* End of transmission */
3721 ret = close(new_sock);
3722 if (ret) {
3723 PERROR("close");
3724 }
3725 new_sock = -1;
3726 }
3727
3728 exit:
3729 error:
3730 if (err) {
3731 ERR("Health error occurred in %s", __func__);
3732 }
3733 DBG("Health check thread dying");
3734 unlink(health_unix_sock_path);
3735 if (sock >= 0) {
3736 ret = close(sock);
3737 if (ret) {
3738 PERROR("close");
3739 }
3740 }
3741
3742 lttng_poll_clean(&events);
3743
3744 rcu_unregister_thread();
3745 return NULL;
3746 }
3747
3748 /*
3749 * This thread manage all clients request using the unix client socket for
3750 * communication.
3751 */
3752 static void *thread_manage_clients(void *data)
3753 {
3754 int sock = -1, ret, i, pollfd, err = -1;
3755 int sock_error;
3756 uint32_t revents, nb_fd;
3757 struct command_ctx *cmd_ctx = NULL;
3758 struct lttng_poll_event events;
3759
3760 DBG("[thread] Manage client started");
3761
3762 rcu_register_thread();
3763
3764 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3765
3766 if (testpoint(thread_manage_clients)) {
3767 goto error_testpoint;
3768 }
3769
3770 health_code_update();
3771
3772 ret = lttcomm_listen_unix_sock(client_sock);
3773 if (ret < 0) {
3774 goto error_listen;
3775 }
3776
3777 /*
3778 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3779 * more will be added to this poll set.
3780 */
3781 ret = sessiond_set_thread_pollset(&events, 2);
3782 if (ret < 0) {
3783 goto error_create_poll;
3784 }
3785
3786 /* Add the application registration socket */
3787 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3788 if (ret < 0) {
3789 goto error;
3790 }
3791
3792 /*
3793 * Notify parent pid that we are ready to accept command for client side.
3794 * This ppid is the one from the external process that spawned us.
3795 */
3796 if (opt_sig_parent) {
3797 kill(ppid, SIGUSR1);
3798 }
3799
3800 /* Notify the parent of the fork() process that we are ready. */
3801 if (opt_daemon) {
3802 kill(child_ppid, SIGUSR1);
3803 }
3804
3805 if (testpoint(thread_manage_clients_before_loop)) {
3806 goto error;
3807 }
3808
3809 health_code_update();
3810
3811 while (1) {
3812 DBG("Accepting client command ...");
3813
3814 /* Inifinite blocking call, waiting for transmission */
3815 restart:
3816 health_poll_entry();
3817 ret = lttng_poll_wait(&events, -1);
3818 health_poll_exit();
3819 if (ret < 0) {
3820 /*
3821 * Restart interrupted system call.
3822 */
3823 if (errno == EINTR) {
3824 goto restart;
3825 }
3826 goto error;
3827 }
3828
3829 nb_fd = ret;
3830
3831 for (i = 0; i < nb_fd; i++) {
3832 /* Fetch once the poll data */
3833 revents = LTTNG_POLL_GETEV(&events, i);
3834 pollfd = LTTNG_POLL_GETFD(&events, i);
3835
3836 health_code_update();
3837
3838 /* Thread quit pipe has been closed. Killing thread. */
3839 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3840 if (ret) {
3841 err = 0;
3842 goto exit;
3843 }
3844
3845 /* Event on the registration socket */
3846 if (pollfd == client_sock) {
3847 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3848 ERR("Client socket poll error");
3849 goto error;
3850 }
3851 }
3852 }
3853
3854 DBG("Wait for client response");
3855
3856 health_code_update();
3857
3858 sock = lttcomm_accept_unix_sock(client_sock);
3859 if (sock < 0) {
3860 goto error;
3861 }
3862
3863 /*
3864 * Set the CLOEXEC flag. Return code is useless because either way, the
3865 * show must go on.
3866 */
3867 (void) utils_set_fd_cloexec(sock);
3868
3869 /* Set socket option for credentials retrieval */
3870 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3871 if (ret < 0) {
3872 goto error;
3873 }
3874
3875 /* Allocate context command to process the client request */
3876 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3877 if (cmd_ctx == NULL) {
3878 PERROR("zmalloc cmd_ctx");
3879 goto error;
3880 }
3881
3882 /* Allocate data buffer for reception */
3883 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3884 if (cmd_ctx->lsm == NULL) {
3885 PERROR("zmalloc cmd_ctx->lsm");
3886 goto error;
3887 }
3888
3889 cmd_ctx->llm = NULL;
3890 cmd_ctx->session = NULL;
3891
3892 health_code_update();
3893
3894 /*
3895 * Data is received from the lttng client. The struct
3896 * lttcomm_session_msg (lsm) contains the command and data request of
3897 * the client.
3898 */
3899 DBG("Receiving data from client ...");
3900 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3901 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3902 if (ret <= 0) {
3903 DBG("Nothing recv() from client... continuing");
3904 ret = close(sock);
3905 if (ret) {
3906 PERROR("close");
3907 }
3908 sock = -1;
3909 clean_command_ctx(&cmd_ctx);
3910 continue;
3911 }
3912
3913 health_code_update();
3914
3915 // TODO: Validate cmd_ctx including sanity check for
3916 // security purpose.
3917
3918 rcu_thread_online();
3919 /*
3920 * This function dispatch the work to the kernel or userspace tracer
3921 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3922 * informations for the client. The command context struct contains
3923 * everything this function may needs.
3924 */
3925 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3926 rcu_thread_offline();
3927 if (ret < 0) {
3928 ret = close(sock);
3929 if (ret) {
3930 PERROR("close");
3931 }
3932 sock = -1;
3933 /*
3934 * TODO: Inform client somehow of the fatal error. At
3935 * this point, ret < 0 means that a zmalloc failed
3936 * (ENOMEM). Error detected but still accept
3937 * command, unless a socket error has been
3938 * detected.
3939 */
3940 clean_command_ctx(&cmd_ctx);
3941 continue;
3942 }
3943
3944 health_code_update();
3945
3946 DBG("Sending response (size: %d, retcode: %s)",
3947 cmd_ctx->lttng_msg_size,
3948 lttng_strerror(-cmd_ctx->llm->ret_code));
3949 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3950 if (ret < 0) {
3951 ERR("Failed to send data back to client");
3952 }
3953
3954 /* End of transmission */
3955 ret = close(sock);
3956 if (ret) {
3957 PERROR("close");
3958 }
3959 sock = -1;
3960
3961 clean_command_ctx(&cmd_ctx);
3962
3963 health_code_update();
3964 }
3965
3966 exit:
3967 error:
3968 if (sock >= 0) {
3969 ret = close(sock);
3970 if (ret) {
3971 PERROR("close");
3972 }
3973 }
3974
3975 lttng_poll_clean(&events);
3976 clean_command_ctx(&cmd_ctx);
3977
3978 error_listen:
3979 error_create_poll:
3980 error_testpoint:
3981 unlink(client_unix_sock_path);
3982 if (client_sock >= 0) {
3983 ret = close(client_sock);
3984 if (ret) {
3985 PERROR("close");
3986 }
3987 }
3988
3989 if (err) {
3990 health_error();
3991 ERR("Health error occurred in %s", __func__);
3992 }
3993
3994 health_unregister(health_sessiond);
3995
3996 DBG("Client thread dying");
3997
3998 rcu_unregister_thread();
3999 return NULL;
4000 }
4001
4002
4003 /*
4004 * usage function on stderr
4005 */
4006 static void usage(void)
4007 {
4008 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4009 fprintf(stderr, " -h, --help Display this usage.\n");
4010 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4011 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4012 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4013 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4014 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4015 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4016 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4017 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4018 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4019 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4020 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4021 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4022 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4023 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4024 fprintf(stderr, " -V, --version Show version number.\n");
4025 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4026 fprintf(stderr, " -q, --quiet No output at all.\n");
4027 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4028 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4029 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4030 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4031 fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
4032 fprintf(stderr, " -f --config Load daemon configuration file\n");
4033 }
4034
4035 /*
4036 * Take an option from the getopt output and set it in the right variable to be
4037 * used later.
4038 *
4039 * Return 0 on success else a negative value.
4040 */
4041 static int set_option(int opt, const char *arg, const char *optname)
4042 {
4043 int ret = 0;
4044
4045 switch (opt) {
4046 case 0:
4047 fprintf(stderr, "option %s", optname);
4048 if (arg) {
4049 fprintf(stderr, " with arg %s\n", arg);
4050 }
4051 break;
4052 case 'c':
4053 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4054 break;
4055 case 'a':
4056 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4057 break;
4058 case 'd':
4059 opt_daemon = 1;
4060 break;
4061 case 'g':
4062 tracing_group_name = strdup(arg);
4063 break;
4064 case 'h':
4065 usage();
4066 exit(EXIT_FAILURE);
4067 case 'V':
4068 fprintf(stdout, "%s\n", VERSION);
4069 exit(EXIT_SUCCESS);
4070 case 'S':
4071 opt_sig_parent = 1;
4072 break;
4073 case 'E':
4074 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4075 break;
4076 case 'C':
4077 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4078 break;
4079 case 'F':
4080 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4081 break;
4082 case 'D':
4083 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4084 break;
4085 case 'H':
4086 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4087 break;
4088 case 'G':
4089 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4090 break;
4091 case 'N':
4092 opt_no_kernel = 1;
4093 break;
4094 case 'q':
4095 lttng_opt_quiet = 1;
4096 break;
4097 case 'v':
4098 /* Verbose level can increase using multiple -v */
4099 if (arg) {
4100 lttng_opt_verbose = config_parse_value(arg);
4101 } else {
4102 lttng_opt_verbose += 1;
4103 }
4104 break;
4105 case 'Z':
4106 if (arg) {
4107 opt_verbose_consumer = config_parse_value(arg);
4108 } else {
4109 opt_verbose_consumer += 1;
4110 }
4111 break;
4112 case 'u':
4113 consumerd32_bin = strdup(arg);
4114 consumerd32_bin_override = 1;
4115 break;
4116 case 'U':
4117 consumerd32_libdir = strdup(arg);
4118 consumerd32_libdir_override = 1;
4119 break;
4120 case 't':
4121 consumerd64_bin = strdup(arg);
4122 consumerd64_bin_override = 1;
4123 break;
4124 case 'T':
4125 consumerd64_libdir = strdup(arg);
4126 consumerd64_libdir_override = 1;
4127 break;
4128 case 'p':
4129 opt_pidfile = strdup(arg);
4130 break;
4131 case 'J': /* JUL TCP port. */
4132 {
4133 unsigned long v;
4134
4135 errno = 0;
4136 v = strtoul(arg, NULL, 0);
4137 if (errno != 0 || !isdigit(arg[0])) {
4138 ERR("Wrong value in --jul-tcp-port parameter: %s", arg);
4139 return -1;
4140 }
4141 if (v == 0 || v >= 65535) {
4142 ERR("Port overflow in --jul-tcp-port parameter: %s", arg);
4143 return -1;
4144 }
4145 jul_tcp_port = (uint32_t) v;
4146 DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
4147 break;
4148 }
4149 default:
4150 /* Unknown option or other error.
4151 * Error is printed by getopt, just return */
4152 ret = -1;
4153 }
4154
4155 return ret;
4156 }
4157
4158 /*
4159 * config_entry_handler_cb used to handle options read from a config file.
4160 * See config_entry_handler_cb comment in common/config/config.h for the
4161 * return value conventions.
4162 */
4163 static int config_entry_handler(const struct config_entry *entry, void *unused)
4164 {
4165 int ret = 0, i;
4166
4167 if (!entry || !entry->name || !entry->value) {
4168 ret = -EINVAL;
4169 goto end;
4170 }
4171
4172 /* Check if the option is to be ignored */
4173 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4174 if (!strcmp(entry->name, config_ignore_options[i])) {
4175 goto end;
4176 }
4177 }
4178
4179 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4180 i++) {
4181
4182 /* Ignore if not fully matched. */
4183 if (strcmp(entry->name, long_options[i].name)) {
4184 continue;
4185 }
4186
4187 /*
4188 * If the option takes no argument on the command line, we have to
4189 * check if the value is "true". We support non-zero numeric values,
4190 * true, on and yes.
4191 */
4192 if (!long_options[i].has_arg) {
4193 ret = config_parse_value(entry->value);
4194 if (ret <= 0) {
4195 if (ret) {
4196 WARN("Invalid configuration value \"%s\" for option %s",
4197 entry->value, entry->name);
4198 }
4199 /* False, skip boolean config option. */
4200 goto end;
4201 }
4202 }
4203
4204 ret = set_option(long_options[i].val, entry->value, entry->name);
4205 goto end;
4206 }
4207
4208 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4209
4210 end:
4211 return ret;
4212 }
4213
4214 /*
4215 * daemon configuration loading and argument parsing
4216 */
4217 static int set_options(int argc, char **argv)
4218 {
4219 int ret = 0, c = 0, option_index = 0;
4220 int orig_optopt = optopt, orig_optind = optind;
4221 char *optstring;
4222 const char *config_path = NULL;
4223
4224 optstring = utils_generate_optstring(long_options,
4225 sizeof(long_options) / sizeof(struct option));
4226 if (!optstring) {
4227 ret = -ENOMEM;
4228 goto end;
4229 }
4230
4231 /* Check for the --config option */
4232 while ((c = getopt_long(argc, argv, optstring, long_options,
4233 &option_index)) != -1) {
4234 if (c == '?') {
4235 ret = -EINVAL;
4236 goto end;
4237 } else if (c != 'f') {
4238 /* if not equal to --config option. */
4239 continue;
4240 }
4241
4242 config_path = utils_expand_path(optarg);
4243 if (!config_path) {
4244 ERR("Failed to resolve path: %s", optarg);
4245 }
4246 }
4247
4248 ret = config_get_section_entries(config_path, config_section_name,
4249 config_entry_handler, NULL);
4250 if (ret) {
4251 if (ret > 0) {
4252 ERR("Invalid configuration option at line %i", ret);
4253 ret = -1;
4254 }
4255 goto end;
4256 }
4257
4258 /* Reset getopt's global state */
4259 optopt = orig_optopt;
4260 optind = orig_optind;
4261 while (1) {
4262 c = getopt_long(argc, argv, optstring, long_options, &option_index);
4263 if (c == -1) {
4264 break;
4265 }
4266
4267 ret = set_option(c, optarg, long_options[option_index].name);
4268 if (ret < 0) {
4269 break;
4270 }
4271 }
4272
4273 end:
4274 free(optstring);
4275 return ret;
4276 }
4277
4278 /*
4279 * Creates the two needed socket by the daemon.
4280 * apps_sock - The communication socket for all UST apps.
4281 * client_sock - The communication of the cli tool (lttng).
4282 */
4283 static int init_daemon_socket(void)
4284 {
4285 int ret = 0;
4286 mode_t old_umask;
4287
4288 old_umask = umask(0);
4289
4290 /* Create client tool unix socket */
4291 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4292 if (client_sock < 0) {
4293 ERR("Create unix sock failed: %s", client_unix_sock_path);
4294 ret = -1;
4295 goto end;
4296 }
4297
4298 /* Set the cloexec flag */
4299 ret = utils_set_fd_cloexec(client_sock);
4300 if (ret < 0) {
4301 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4302 "Continuing but note that the consumer daemon will have a "
4303 "reference to this socket on exec()", client_sock);
4304 }
4305
4306 /* File permission MUST be 660 */
4307 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4308 if (ret < 0) {
4309 ERR("Set file permissions failed: %s", client_unix_sock_path);
4310 PERROR("chmod");
4311 goto end;
4312 }
4313
4314 /* Create the application unix socket */
4315 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4316 if (apps_sock < 0) {
4317 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4318 ret = -1;
4319 goto end;
4320 }
4321
4322 /* Set the cloexec flag */
4323 ret = utils_set_fd_cloexec(apps_sock);
4324 if (ret < 0) {
4325 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4326 "Continuing but note that the consumer daemon will have a "
4327 "reference to this socket on exec()", apps_sock);
4328 }
4329
4330 /* File permission MUST be 666 */
4331 ret = chmod(apps_unix_sock_path,
4332 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4333 if (ret < 0) {
4334 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4335 PERROR("chmod");
4336 goto end;
4337 }
4338
4339 DBG3("Session daemon client socket %d and application socket %d created",
4340 client_sock, apps_sock);
4341
4342 end:
4343 umask(old_umask);
4344 return ret;
4345 }
4346
4347 /*
4348 * Check if the global socket is available, and if a daemon is answering at the
4349 * other side. If yes, error is returned.
4350 */
4351 static int check_existing_daemon(void)
4352 {
4353 /* Is there anybody out there ? */
4354 if (lttng_session_daemon_alive()) {
4355 return -EEXIST;
4356 }
4357
4358 return 0;
4359 }
4360
4361 /*
4362 * Set the tracing group gid onto the client socket.
4363 *
4364 * Race window between mkdir and chown is OK because we are going from more
4365 * permissive (root.root) to less permissive (root.tracing).
4366 */
4367 static int set_permissions(char *rundir)
4368 {
4369 int ret;
4370 gid_t gid;
4371
4372 gid = utils_get_group_id(tracing_group_name);
4373
4374 /* Set lttng run dir */
4375 ret = chown(rundir, 0, gid);
4376 if (ret < 0) {
4377 ERR("Unable to set group on %s", rundir);
4378 PERROR("chown");
4379 }
4380
4381 /*
4382 * Ensure all applications and tracing group can search the run
4383 * dir. Allow everyone to read the directory, since it does not
4384 * buy us anything to hide its content.
4385 */
4386 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4387 if (ret < 0) {
4388 ERR("Unable to set permissions on %s", rundir);
4389 PERROR("chmod");
4390 }
4391
4392 /* lttng client socket path */
4393 ret = chown(client_unix_sock_path, 0, gid);
4394 if (ret < 0) {
4395 ERR("Unable to set group on %s", client_unix_sock_path);
4396 PERROR("chown");
4397 }
4398
4399 /* kconsumer error socket path */
4400 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4401 if (ret < 0) {
4402 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4403 PERROR("chown");
4404 }
4405
4406 /* 64-bit ustconsumer error socket path */
4407 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4408 if (ret < 0) {
4409 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4410 PERROR("chown");
4411 }
4412
4413 /* 32-bit ustconsumer compat32 error socket path */
4414 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4415 if (ret < 0) {
4416 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4417 PERROR("chown");
4418 }
4419
4420 DBG("All permissions are set");
4421
4422 return ret;
4423 }
4424
4425 /*
4426 * Create the lttng run directory needed for all global sockets and pipe.
4427 */
4428 static int create_lttng_rundir(const char *rundir)
4429 {
4430 int ret;
4431
4432 DBG3("Creating LTTng run directory: %s", rundir);
4433
4434 ret = mkdir(rundir, S_IRWXU);
4435 if (ret < 0) {
4436 if (errno != EEXIST) {
4437 ERR("Unable to create %s", rundir);
4438 goto error;
4439 } else {
4440 ret = 0;
4441 }
4442 }
4443
4444 error:
4445 return ret;
4446 }
4447
4448 /*
4449 * Setup sockets and directory needed by the kconsumerd communication with the
4450 * session daemon.
4451 */
4452 static int set_consumer_sockets(struct consumer_data *consumer_data,
4453 const char *rundir)
4454 {
4455 int ret;
4456 char path[PATH_MAX];
4457
4458 switch (consumer_data->type) {
4459 case LTTNG_CONSUMER_KERNEL:
4460 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4461 break;
4462 case LTTNG_CONSUMER64_UST:
4463 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4464 break;
4465 case LTTNG_CONSUMER32_UST:
4466 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4467 break;
4468 default:
4469 ERR("Consumer type unknown");
4470 ret = -EINVAL;
4471 goto error;
4472 }
4473
4474 DBG2("Creating consumer directory: %s", path);
4475
4476 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4477 if (ret < 0) {
4478 if (errno != EEXIST) {
4479 PERROR("mkdir");
4480 ERR("Failed to create %s", path);
4481 goto error;
4482 }
4483 ret = -1;
4484 }
4485 if (is_root) {
4486 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4487 if (ret < 0) {
4488 ERR("Unable to set group on %s", path);
4489 PERROR("chown");
4490 goto error;
4491 }
4492 }
4493
4494 /* Create the kconsumerd error unix socket */
4495 consumer_data->err_sock =
4496 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4497 if (consumer_data->err_sock < 0) {
4498 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4499 ret = -1;
4500 goto error;
4501 }
4502
4503 /*
4504 * Set the CLOEXEC flag. Return code is useless because either way, the
4505 * show must go on.
4506 */
4507 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4508 if (ret < 0) {
4509 PERROR("utils_set_fd_cloexec");
4510 /* continue anyway */
4511 }
4512
4513 /* File permission MUST be 660 */
4514 ret = chmod(consumer_data->err_unix_sock_path,
4515 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4516 if (ret < 0) {
4517 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4518 PERROR("chmod");
4519 goto error;
4520 }
4521
4522 error:
4523 return ret;
4524 }
4525
4526 /*
4527 * Signal handler for the daemon
4528 *
4529 * Simply stop all worker threads, leaving main() return gracefully after
4530 * joining all threads and calling cleanup().
4531 */
4532 static void sighandler(int sig)
4533 {
4534 switch (sig) {
4535 case SIGPIPE:
4536 DBG("SIGPIPE caught");
4537 return;
4538 case SIGINT:
4539 DBG("SIGINT caught");
4540 stop_threads();
4541 break;
4542 case SIGTERM:
4543 DBG("SIGTERM caught");
4544 stop_threads();
4545 break;
4546 case SIGUSR1:
4547 CMM_STORE_SHARED(recv_child_signal, 1);
4548 break;
4549 default:
4550 break;
4551 }
4552 }
4553
4554 /*
4555 * Setup signal handler for :
4556 * SIGINT, SIGTERM, SIGPIPE
4557 */
4558 static int set_signal_handler(void)
4559 {
4560 int ret = 0;
4561 struct sigaction sa;
4562 sigset_t sigset;
4563
4564 if ((ret = sigemptyset(&sigset)) < 0) {
4565 PERROR("sigemptyset");
4566 return ret;
4567 }
4568
4569 sa.sa_handler = sighandler;
4570 sa.sa_mask = sigset;
4571 sa.sa_flags = 0;
4572 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4573 PERROR("sigaction");
4574 return ret;
4575 }
4576
4577 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4578 PERROR("sigaction");
4579 return ret;
4580 }
4581
4582 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4583 PERROR("sigaction");
4584 return ret;
4585 }
4586
4587 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4588 PERROR("sigaction");
4589 return ret;
4590 }
4591
4592 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4593
4594 return ret;
4595 }
4596
4597 /*
4598 * Set open files limit to unlimited. This daemon can open a large number of
4599 * file descriptors in order to consumer multiple kernel traces.
4600 */
4601 static void set_ulimit(void)
4602 {
4603 int ret;
4604 struct rlimit lim;
4605
4606 /* The kernel does not allowed an infinite limit for open files */
4607 lim.rlim_cur = 65535;
4608 lim.rlim_max = 65535;
4609
4610 ret = setrlimit(RLIMIT_NOFILE, &lim);
4611 if (ret < 0) {
4612 PERROR("failed to set open files limit");
4613 }
4614 }
4615
4616 /*
4617 * Write pidfile using the rundir and opt_pidfile.
4618 */
4619 static void write_pidfile(void)
4620 {
4621 int ret;
4622 char pidfile_path[PATH_MAX];
4623
4624 assert(rundir);
4625
4626 if (opt_pidfile) {
4627 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4628 } else {
4629 /* Build pidfile path from rundir and opt_pidfile. */
4630 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4631 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4632 if (ret < 0) {
4633 PERROR("snprintf pidfile path");
4634 goto error;
4635 }
4636 }
4637
4638 /*
4639 * Create pid file in rundir. Return value is of no importance. The
4640 * execution will continue even though we are not able to write the file.
4641 */
4642 (void) utils_create_pid_file(getpid(), pidfile_path);
4643
4644 error:
4645 return;
4646 }
4647
4648 /*
4649 * Write JUL TCP port using the rundir.
4650 */
4651 static void write_julport(void)
4652 {
4653 int ret;
4654 char path[PATH_MAX];
4655
4656 assert(rundir);
4657
4658 ret = snprintf(path, sizeof(path), "%s/"
4659 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
4660 if (ret < 0) {
4661 PERROR("snprintf julport path");
4662 goto error;
4663 }
4664
4665 /*
4666 * Create TCP JUL port file in rundir. Return value is of no importance.
4667 * The execution will continue even though we are not able to write the
4668 * file.
4669 */
4670 (void) utils_create_pid_file(jul_tcp_port, path);
4671
4672 error:
4673 return;
4674 }
4675
4676 /*
4677 * Daemonize this process by forking and making the parent wait for the child
4678 * to signal it indicating readiness. Once received, the parent successfully
4679 * quits.
4680 *
4681 * The child process undergoes the same action that daemon(3) does meaning
4682 * setsid, chdir, and dup /dev/null into 0, 1 and 2.
4683 *
4684 * Return 0 on success else -1 on error.
4685 */
4686 static int daemonize(void)
4687 {
4688 int ret;
4689 pid_t pid;
4690
4691 /* Get parent pid of this process. */
4692 child_ppid = getppid();
4693
4694 pid = fork();
4695 if (pid < 0) {
4696 PERROR("fork");
4697 goto error;
4698 } else if (pid == 0) {
4699 int fd;
4700 pid_t sid;
4701
4702 /* Child */
4703
4704 /*
4705 * Get the newly created parent pid so we can signal that process when
4706 * we are ready to operate.
4707 */
4708 child_ppid = getppid();
4709
4710 sid = setsid();
4711 if (sid < 0) {
4712 PERROR("setsid");
4713 goto error;
4714 }
4715
4716 /* Try to change directory to /. If we can't well at least notify. */
4717 ret = chdir("/");
4718 if (ret < 0) {
4719 PERROR("chdir");
4720 }
4721
4722 fd = open(_PATH_DEVNULL, O_RDWR, 0);
4723 if (fd < 0) {
4724 PERROR("open %s", _PATH_DEVNULL);
4725 /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */
4726 } else {
4727 (void) dup2(fd, STDIN_FILENO);
4728 (void) dup2(fd, STDOUT_FILENO);
4729 (void) dup2(fd, STDERR_FILENO);
4730 if (fd > 2) {
4731 ret = close(fd);
4732 if (ret < 0) {
4733 PERROR("close");
4734 }
4735 }
4736 }
4737 goto end;
4738 } else {
4739 /* Parent */
4740
4741 /*
4742 * Waiting for child to notify this parent that it can exit. Note that
4743 * sleep() is interrupted before the 1 second delay as soon as the
4744 * signal is received, so it will not cause visible delay for the
4745 * user.
4746 */
4747 while (!CMM_LOAD_SHARED(recv_child_signal)) {
4748 int status;
4749 pid_t ret;
4750
4751 /*
4752 * Check if child exists without blocking. If so, we have to stop
4753 * this parent process and return an error.
4754 */
4755 ret = waitpid(pid, &status, WNOHANG);
4756 if (ret < 0 || (ret != 0 && WIFEXITED(status))) {
4757 /* The child exited somehow or was not valid. */
4758 goto error;
4759 }
4760 sleep(1);
4761 }
4762
4763 /*
4764 * From this point on, the parent can exit and the child is now an
4765 * operationnal session daemon ready to serve clients and applications.
4766 */
4767 exit(EXIT_SUCCESS);
4768 }
4769
4770 end:
4771 return 0;
4772
4773 error:
4774 return -1;
4775 }
4776
4777 /*
4778 * main
4779 */
4780 int main(int argc, char **argv)
4781 {
4782 int ret = 0;
4783 void *status;
4784 const char *home_path, *env_app_timeout;
4785
4786 init_kernel_workarounds();
4787
4788 rcu_register_thread();
4789
4790 if ((ret = set_signal_handler()) < 0) {
4791 goto error;
4792 }
4793
4794 setup_consumerd_path();
4795
4796 page_size = sysconf(_SC_PAGESIZE);
4797 if (page_size < 0) {
4798 PERROR("sysconf _SC_PAGESIZE");
4799 page_size = LONG_MAX;
4800 WARN("Fallback page size to %ld", page_size);
4801 }
4802
4803 /* Parse arguments and load the daemon configuration file */
4804 progname = argv[0];
4805 if ((ret = set_options(argc, argv)) < 0) {
4806 goto error;
4807 }
4808
4809 /* Daemonize */
4810 if (opt_daemon) {
4811 int i;
4812
4813 ret = daemonize();
4814 if (ret < 0) {
4815 goto error;
4816 }
4817
4818 /*
4819 * We are in the child. Make sure all other file descriptors are
4820 * closed, in case we are called with more opened file descriptors than
4821 * the standard ones.
4822 */
4823 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4824 (void) close(i);
4825 }
4826 }
4827
4828 /* Create thread quit pipe */
4829 if ((ret = init_thread_quit_pipe()) < 0) {
4830 goto error;
4831 }
4832
4833 /* Check if daemon is UID = 0 */
4834 is_root = !getuid();
4835
4836 if (is_root) {
4837 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4838
4839 /* Create global run dir with root access */
4840 ret = create_lttng_rundir(rundir);
4841 if (ret < 0) {
4842 goto error;
4843 }
4844
4845 if (strlen(apps_unix_sock_path) == 0) {
4846 snprintf(apps_unix_sock_path, PATH_MAX,
4847 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4848 }
4849
4850 if (strlen(client_unix_sock_path) == 0) {
4851 snprintf(client_unix_sock_path, PATH_MAX,
4852 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4853 }
4854
4855 /* Set global SHM for ust */
4856 if (strlen(wait_shm_path) == 0) {
4857 snprintf(wait_shm_path, PATH_MAX,
4858 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4859 }
4860
4861 if (strlen(health_unix_sock_path) == 0) {
4862 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4863 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4864 }
4865
4866 /* Setup kernel consumerd path */
4867 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4868 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4869 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4870 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4871
4872 DBG2("Kernel consumer err path: %s",
4873 kconsumer_data.err_unix_sock_path);
4874 DBG2("Kernel consumer cmd path: %s",
4875 kconsumer_data.cmd_unix_sock_path);
4876 } else {
4877 home_path = utils_get_home_dir();
4878 if (home_path == NULL) {
4879 /* TODO: Add --socket PATH option */
4880 ERR("Can't get HOME directory for sockets creation.");
4881 ret = -EPERM;
4882 goto error;
4883 }
4884
4885 /*
4886 * Create rundir from home path. This will create something like
4887 * $HOME/.lttng
4888 */
4889 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4890 if (ret < 0) {
4891 ret = -ENOMEM;
4892 goto error;
4893 }
4894
4895 ret = create_lttng_rundir(rundir);
4896 if (ret < 0) {
4897 goto error;
4898 }
4899
4900 if (strlen(apps_unix_sock_path) == 0) {
4901 snprintf(apps_unix_sock_path, PATH_MAX,
4902 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4903 }
4904
4905 /* Set the cli tool unix socket path */
4906 if (strlen(client_unix_sock_path) == 0) {
4907 snprintf(client_unix_sock_path, PATH_MAX,
4908 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4909 }
4910
4911 /* Set global SHM for ust */
4912 if (strlen(wait_shm_path) == 0) {
4913 snprintf(wait_shm_path, PATH_MAX,
4914 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4915 }
4916
4917 /* Set health check Unix path */
4918 if (strlen(health_unix_sock_path) == 0) {
4919 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4920 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4921 }
4922 }
4923
4924 /* Set consumer initial state */
4925 kernel_consumerd_state = CONSUMER_STOPPED;
4926 ust_consumerd_state = CONSUMER_STOPPED;
4927
4928 DBG("Client socket path %s", client_unix_sock_path);
4929 DBG("Application socket path %s", apps_unix_sock_path);
4930 DBG("Application wait path %s", wait_shm_path);
4931 DBG("LTTng run directory path: %s", rundir);
4932
4933 /* 32 bits consumerd path setup */
4934 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4935 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4936 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4937 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4938
4939 DBG2("UST consumer 32 bits err path: %s",
4940 ustconsumer32_data.err_unix_sock_path);
4941 DBG2("UST consumer 32 bits cmd path: %s",
4942 ustconsumer32_data.cmd_unix_sock_path);
4943
4944 /* 64 bits consumerd path setup */
4945 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4946 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4947 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4948 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4949
4950 DBG2("UST consumer 64 bits err path: %s",
4951 ustconsumer64_data.err_unix_sock_path);
4952 DBG2("UST consumer 64 bits cmd path: %s",
4953 ustconsumer64_data.cmd_unix_sock_path);
4954
4955 /*
4956 * See if daemon already exist.
4957 */
4958 if ((ret = check_existing_daemon()) < 0) {
4959 ERR("Already running daemon.\n");
4960 /*
4961 * We do not goto exit because we must not cleanup()
4962 * because a daemon is already running.
4963 */
4964 goto error;
4965 }
4966
4967 /*
4968 * Init UST app hash table. Alloc hash table before this point since
4969 * cleanup() can get called after that point.
4970 */
4971 ust_app_ht_alloc();
4972
4973 /* Initialize JUL domain subsystem. */
4974 if ((ret = jul_init()) < 0) {
4975 /* ENOMEM at this point. */
4976 goto error;
4977 }
4978
4979 /* After this point, we can safely call cleanup() with "goto exit" */
4980
4981 /*
4982 * These actions must be executed as root. We do that *after* setting up
4983 * the sockets path because we MUST make the check for another daemon using
4984 * those paths *before* trying to set the kernel consumer sockets and init
4985 * kernel tracer.
4986 */
4987 if (is_root) {
4988 ret = set_consumer_sockets(&kconsumer_data, rundir);
4989 if (ret < 0) {
4990 goto exit;
4991 }
4992
4993 /* Setup kernel tracer */
4994 if (!opt_no_kernel) {
4995 init_kernel_tracer();
4996 }
4997
4998 /* Set ulimit for open files */
4999 set_ulimit();
5000 }
5001 /* init lttng_fd tracking must be done after set_ulimit. */
5002 lttng_fd_init();
5003
5004 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
5005 if (ret < 0) {
5006 goto exit;
5007 }
5008
5009 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
5010 if (ret < 0) {
5011 goto exit;
5012 }
5013
5014 /* Setup the needed unix socket */
5015 if ((ret = init_daemon_socket()) < 0) {
5016 goto exit;
5017 }
5018
5019 /* Set credentials to socket */
5020 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
5021 goto exit;
5022 }
5023
5024 /* Get parent pid if -S, --sig-parent is specified. */
5025 if (opt_sig_parent) {
5026 ppid = getppid();
5027 }
5028
5029 /* Setup the kernel pipe for waking up the kernel thread */
5030 if (is_root && !opt_no_kernel) {
5031 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
5032 goto exit;
5033 }
5034 }
5035
5036 /* Setup the thread ht_cleanup communication pipe. */
5037 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
5038 goto exit;
5039 }
5040
5041 /* Setup the thread apps communication pipe. */
5042 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
5043 goto exit;
5044 }
5045
5046 /* Setup the thread apps notify communication pipe. */
5047 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
5048 goto exit;
5049 }
5050
5051 /* Initialize global buffer per UID and PID registry. */
5052 buffer_reg_init_uid_registry();
5053 buffer_reg_init_pid_registry();
5054
5055 /* Init UST command queue. */
5056 cds_wfq_init(&ust_cmd_queue.queue);
5057
5058 /*
5059 * Get session list pointer. This pointer MUST NOT be free(). This list is
5060 * statically declared in session.c
5061 */
5062 session_list_ptr = session_get_list();
5063
5064 /* Set up max poll set size */
5065 lttng_poll_set_max_size();
5066
5067 cmd_init();
5068
5069 /* Check for the application socket timeout env variable. */
5070 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5071 if (env_app_timeout) {
5072 app_socket_timeout = atoi(env_app_timeout);
5073 } else {
5074 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5075 }
5076
5077 write_pidfile();
5078 write_julport();
5079
5080 /* Initialize communication library */
5081 lttcomm_init();
5082 /* This is to get the TCP timeout value. */
5083 lttcomm_inet_init();
5084
5085 /*
5086 * Initialize the health check subsystem. This call should set the
5087 * appropriate time values.
5088 */
5089 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5090 if (!health_sessiond) {
5091 PERROR("health_app_create error");
5092 goto exit_health_sessiond_cleanup;
5093 }
5094
5095 /* Create thread to clean up RCU hash tables */
5096 ret = pthread_create(&ht_cleanup_thread, NULL,
5097 thread_ht_cleanup, (void *) NULL);
5098 if (ret != 0) {
5099 PERROR("pthread_create ht_cleanup");
5100 goto exit_ht_cleanup;
5101 }
5102
5103 /* Create health-check thread */
5104 ret = pthread_create(&health_thread, NULL,
5105 thread_manage_health, (void *) NULL);
5106 if (ret != 0) {
5107 PERROR("pthread_create health");
5108 goto exit_health;
5109 }
5110
5111 /* Create thread to manage the client socket */
5112 ret = pthread_create(&client_thread, NULL,
5113 thread_manage_clients, (void *) NULL);
5114 if (ret != 0) {
5115 PERROR("pthread_create clients");
5116 goto exit_client;
5117 }
5118
5119 /* Create thread to dispatch registration */
5120 ret = pthread_create(&dispatch_thread, NULL,
5121 thread_dispatch_ust_registration, (void *) NULL);
5122 if (ret != 0) {
5123 PERROR("pthread_create dispatch");
5124 goto exit_dispatch;
5125 }
5126
5127 /* Create thread to manage application registration. */
5128 ret = pthread_create(&reg_apps_thread, NULL,
5129 thread_registration_apps, (void *) NULL);
5130 if (ret != 0) {
5131 PERROR("pthread_create registration");
5132 goto exit_reg_apps;
5133 }
5134
5135 /* Create thread to manage application socket */
5136 ret = pthread_create(&apps_thread, NULL,
5137 thread_manage_apps, (void *) NULL);
5138 if (ret != 0) {
5139 PERROR("pthread_create apps");
5140 goto exit_apps;
5141 }
5142
5143 /* Create thread to manage application notify socket */
5144 ret = pthread_create(&apps_notify_thread, NULL,
5145 ust_thread_manage_notify, (void *) NULL);
5146 if (ret != 0) {
5147 PERROR("pthread_create apps");
5148 goto exit_apps_notify;
5149 }
5150
5151 /* Create JUL registration thread. */
5152 ret = pthread_create(&jul_reg_thread, NULL,
5153 jul_thread_manage_registration, (void *) NULL);
5154 if (ret != 0) {
5155 PERROR("pthread_create apps");
5156 goto exit_jul_reg;
5157 }
5158
5159 /* Don't start this thread if kernel tracing is not requested nor root */
5160 if (is_root && !opt_no_kernel) {
5161 /* Create kernel thread to manage kernel event */
5162 ret = pthread_create(&kernel_thread, NULL,
5163 thread_manage_kernel, (void *) NULL);
5164 if (ret != 0) {
5165 PERROR("pthread_create kernel");
5166 goto exit_kernel;
5167 }
5168
5169 ret = pthread_join(kernel_thread, &status);
5170 if (ret != 0) {
5171 PERROR("pthread_join");
5172 goto error; /* join error, exit without cleanup */
5173 }
5174 }
5175
5176 exit_kernel:
5177 ret = pthread_join(jul_reg_thread, &status);
5178 if (ret != 0) {
5179 PERROR("pthread_join JUL");
5180 goto error; /* join error, exit without cleanup */
5181 }
5182
5183 exit_jul_reg:
5184 ret = pthread_join(apps_notify_thread, &status);
5185 if (ret != 0) {
5186 PERROR("pthread_join apps notify");
5187 goto error; /* join error, exit without cleanup */
5188 }
5189
5190 exit_apps_notify:
5191 ret = pthread_join(apps_thread, &status);
5192 if (ret != 0) {
5193 PERROR("pthread_join apps");
5194 goto error; /* join error, exit without cleanup */
5195 }
5196
5197
5198 exit_apps:
5199 ret = pthread_join(reg_apps_thread, &status);
5200 if (ret != 0) {
5201 PERROR("pthread_join");
5202 goto error; /* join error, exit without cleanup */
5203 }
5204
5205 exit_reg_apps:
5206 ret = pthread_join(dispatch_thread, &status);
5207 if (ret != 0) {
5208 PERROR("pthread_join");
5209 goto error; /* join error, exit without cleanup */
5210 }
5211
5212 exit_dispatch:
5213 ret = pthread_join(client_thread, &status);
5214 if (ret != 0) {
5215 PERROR("pthread_join");
5216 goto error; /* join error, exit without cleanup */
5217 }
5218
5219 ret = join_consumer_thread(&kconsumer_data);
5220 if (ret != 0) {
5221 PERROR("join_consumer");
5222 goto error; /* join error, exit without cleanup */
5223 }
5224
5225 ret = join_consumer_thread(&ustconsumer32_data);
5226 if (ret != 0) {
5227 PERROR("join_consumer ust32");
5228 goto error; /* join error, exit without cleanup */
5229 }
5230
5231 ret = join_consumer_thread(&ustconsumer64_data);
5232 if (ret != 0) {
5233 PERROR("join_consumer ust64");
5234 goto error; /* join error, exit without cleanup */
5235 }
5236
5237 exit_client:
5238 ret = pthread_join(health_thread, &status);
5239 if (ret != 0) {
5240 PERROR("pthread_join health thread");
5241 goto error; /* join error, exit without cleanup */
5242 }
5243
5244 exit_health:
5245 ret = pthread_join(ht_cleanup_thread, &status);
5246 if (ret != 0) {
5247 PERROR("pthread_join ht cleanup thread");
5248 goto error; /* join error, exit without cleanup */
5249 }
5250 exit_ht_cleanup:
5251 health_app_destroy(health_sessiond);
5252 exit_health_sessiond_cleanup:
5253 exit:
5254 /*
5255 * cleanup() is called when no other thread is running.
5256 */
5257 rcu_thread_online();
5258 cleanup();
5259 rcu_thread_offline();
5260 rcu_unregister_thread();
5261 if (!ret) {
5262 exit(EXIT_SUCCESS);
5263 }
5264 error:
5265 exit(EXIT_FAILURE);
5266 }
This page took 0.173517 seconds and 5 git commands to generate.