Mi test: make sure to not auto load session when using lttng-sessiond
[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/daemonize.h>
50 #include <common/config/config.h>
51
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
54 #include "channel.h"
55 #include "cmd.h"
56 #include "consumer.h"
57 #include "context.h"
58 #include "event.h"
59 #include "kernel.h"
60 #include "kernel-consumer.h"
61 #include "modprobe.h"
62 #include "shm.h"
63 #include "ust-ctl.h"
64 #include "ust-consumer.h"
65 #include "utils.h"
66 #include "fd-limit.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "jul-thread.h"
71 #include "save.h"
72 #include "load-session-thread.h"
73
74 #define CONSUMERD_FILE "lttng-consumerd"
75
76 const char *progname;
77 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
78 static int tracing_group_name_override;
79 static char *opt_pidfile;
80 static int opt_sig_parent;
81 static int opt_verbose_consumer;
82 static int opt_daemon, opt_background;
83 static int opt_no_kernel;
84 static char *opt_load_session_path;
85 static pid_t ppid; /* Parent PID for --sig-parent option */
86 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
87 static char *rundir;
88 static int lockfile_fd = -1;
89
90 /* Set to 1 when a SIGUSR1 signal is received. */
91 static int recv_child_signal;
92
93 /*
94 * Consumer daemon specific control data. Every value not initialized here is
95 * set to 0 by the static definition.
96 */
97 static struct consumer_data kconsumer_data = {
98 .type = LTTNG_CONSUMER_KERNEL,
99 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
100 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
101 .err_sock = -1,
102 .cmd_sock = -1,
103 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
104 .lock = PTHREAD_MUTEX_INITIALIZER,
105 .cond = PTHREAD_COND_INITIALIZER,
106 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
107 };
108 static struct consumer_data ustconsumer64_data = {
109 .type = LTTNG_CONSUMER64_UST,
110 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
111 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
112 .err_sock = -1,
113 .cmd_sock = -1,
114 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
115 .lock = PTHREAD_MUTEX_INITIALIZER,
116 .cond = PTHREAD_COND_INITIALIZER,
117 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
118 };
119 static struct consumer_data ustconsumer32_data = {
120 .type = LTTNG_CONSUMER32_UST,
121 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
122 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
123 .err_sock = -1,
124 .cmd_sock = -1,
125 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
126 .lock = PTHREAD_MUTEX_INITIALIZER,
127 .cond = PTHREAD_COND_INITIALIZER,
128 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
129 };
130
131 /* Command line options */
132 static const struct option long_options[] = {
133 { "client-sock", 1, 0, 'c' },
134 { "apps-sock", 1, 0, 'a' },
135 { "kconsumerd-cmd-sock", 1, 0, 'C' },
136 { "kconsumerd-err-sock", 1, 0, 'E' },
137 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
138 { "ustconsumerd32-err-sock", 1, 0, 'H' },
139 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
140 { "ustconsumerd64-err-sock", 1, 0, 'F' },
141 { "consumerd32-path", 1, 0, 'u' },
142 { "consumerd32-libdir", 1, 0, 'U' },
143 { "consumerd64-path", 1, 0, 't' },
144 { "consumerd64-libdir", 1, 0, 'T' },
145 { "daemonize", 0, 0, 'd' },
146 { "background", 0, 0, 'b' },
147 { "sig-parent", 0, 0, 'S' },
148 { "help", 0, 0, 'h' },
149 { "group", 1, 0, 'g' },
150 { "version", 0, 0, 'V' },
151 { "quiet", 0, 0, 'q' },
152 { "verbose", 0, 0, 'v' },
153 { "verbose-consumer", 0, 0, 'Z' },
154 { "no-kernel", 0, 0, 'N' },
155 { "pidfile", 1, 0, 'p' },
156 { "jul-tcp-port", 1, 0, 'J' },
157 { "config", 1, 0, 'f' },
158 { "load", 1, 0, 'l' },
159 { "kmod-probes", 1, 0, 'P' },
160 { NULL, 0, 0, 0 }
161 };
162
163 /* Command line options to ignore from configuration file */
164 static const char *config_ignore_options[] = { "help", "version", "config" };
165
166 /* Shared between threads */
167 static int dispatch_thread_exit;
168
169 /* Global application Unix socket path */
170 static char apps_unix_sock_path[PATH_MAX];
171 /* Global client Unix socket path */
172 static char client_unix_sock_path[PATH_MAX];
173 /* global wait shm path for UST */
174 static char wait_shm_path[PATH_MAX];
175 /* Global health check unix path */
176 static char health_unix_sock_path[PATH_MAX];
177
178 /* Sockets and FDs */
179 static int client_sock = -1;
180 static int apps_sock = -1;
181 int kernel_tracer_fd = -1;
182 static int kernel_poll_pipe[2] = { -1, -1 };
183
184 /*
185 * Quit pipe for all threads. This permits a single cancellation point
186 * for all threads when receiving an event on the pipe.
187 */
188 static int thread_quit_pipe[2] = { -1, -1 };
189
190 /*
191 * This pipe is used to inform the thread managing application communication
192 * that a command is queued and ready to be processed.
193 */
194 static int apps_cmd_pipe[2] = { -1, -1 };
195
196 int apps_cmd_notify_pipe[2] = { -1, -1 };
197
198 /* Pthread, Mutexes and Semaphores */
199 static pthread_t apps_thread;
200 static pthread_t apps_notify_thread;
201 static pthread_t reg_apps_thread;
202 static pthread_t client_thread;
203 static pthread_t kernel_thread;
204 static pthread_t dispatch_thread;
205 static pthread_t health_thread;
206 static pthread_t ht_cleanup_thread;
207 static pthread_t jul_reg_thread;
208 static pthread_t load_session_thread;
209
210 /*
211 * UST registration command queue. This queue is tied with a futex and uses a N
212 * wakers / 1 waiter implemented and detailed in futex.c/.h
213 *
214 * The thread_registration_apps and thread_dispatch_ust_registration uses this
215 * queue along with the wait/wake scheme. The thread_manage_apps receives down
216 * the line new application socket and monitors it for any I/O error or clean
217 * close that triggers an unregistration of the application.
218 */
219 static struct ust_cmd_queue ust_cmd_queue;
220
221 /*
222 * Pointer initialized before thread creation.
223 *
224 * This points to the tracing session list containing the session count and a
225 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
226 * MUST NOT be taken if you call a public function in session.c.
227 *
228 * The lock is nested inside the structure: session_list_ptr->lock. Please use
229 * session_lock_list and session_unlock_list for lock acquisition.
230 */
231 static struct ltt_session_list *session_list_ptr;
232
233 int ust_consumerd64_fd = -1;
234 int ust_consumerd32_fd = -1;
235
236 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
237 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
238 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
239 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
240 static int consumerd32_bin_override;
241 static int consumerd64_bin_override;
242 static int consumerd32_libdir_override;
243 static int consumerd64_libdir_override;
244
245 static const char *module_proc_lttng = "/proc/lttng";
246
247 /*
248 * Consumer daemon state which is changed when spawning it, killing it or in
249 * case of a fatal error.
250 */
251 enum consumerd_state {
252 CONSUMER_STARTED = 1,
253 CONSUMER_STOPPED = 2,
254 CONSUMER_ERROR = 3,
255 };
256
257 /*
258 * This consumer daemon state is used to validate if a client command will be
259 * able to reach the consumer. If not, the client is informed. For instance,
260 * doing a "lttng start" when the consumer state is set to ERROR will return an
261 * error to the client.
262 *
263 * The following example shows a possible race condition of this scheme:
264 *
265 * consumer thread error happens
266 * client cmd arrives
267 * client cmd checks state -> still OK
268 * consumer thread exit, sets error
269 * client cmd try to talk to consumer
270 * ...
271 *
272 * However, since the consumer is a different daemon, we have no way of making
273 * sure the command will reach it safely even with this state flag. This is why
274 * we consider that up to the state validation during command processing, the
275 * command is safe. After that, we can not guarantee the correctness of the
276 * client request vis-a-vis the consumer.
277 */
278 static enum consumerd_state ust_consumerd_state;
279 static enum consumerd_state kernel_consumerd_state;
280
281 /*
282 * Socket timeout for receiving and sending in seconds.
283 */
284 static int app_socket_timeout;
285
286 /* Set in main() with the current page size. */
287 long page_size;
288
289 /* Application health monitoring */
290 struct health_app *health_sessiond;
291
292 /* JUL TCP port for registration. Used by the JUL thread. */
293 unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
294
295 /* Am I root or not. */
296 int is_root; /* Set to 1 if the daemon is running as root */
297
298 const char * const config_section_name = "sessiond";
299
300 /* Load session thread information to operate. */
301 struct load_session_thread_data *load_info;
302
303 /*
304 * Whether sessiond is ready for commands/health check requests.
305 * NR_LTTNG_SESSIOND_READY must match the number of calls to
306 * sessiond_notify_ready().
307 */
308 #define NR_LTTNG_SESSIOND_READY 3
309 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
310
311 /* Notify parents that we are ready for cmd and health check */
312 LTTNG_HIDDEN
313 void sessiond_notify_ready(void)
314 {
315 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
316 /*
317 * Notify parent pid that we are ready to accept command
318 * for client side. This ppid is the one from the
319 * external process that spawned us.
320 */
321 if (opt_sig_parent) {
322 kill(ppid, SIGUSR1);
323 }
324
325 /*
326 * Notify the parent of the fork() process that we are
327 * ready.
328 */
329 if (opt_daemon || opt_background) {
330 kill(child_ppid, SIGUSR1);
331 }
332 }
333 }
334
335 static
336 void setup_consumerd_path(void)
337 {
338 const char *bin, *libdir;
339
340 /*
341 * Allow INSTALL_BIN_PATH to be used as a target path for the
342 * native architecture size consumer if CONFIG_CONSUMER*_PATH
343 * has not been defined.
344 */
345 #if (CAA_BITS_PER_LONG == 32)
346 if (!consumerd32_bin[0]) {
347 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
348 }
349 if (!consumerd32_libdir[0]) {
350 consumerd32_libdir = INSTALL_LIB_PATH;
351 }
352 #elif (CAA_BITS_PER_LONG == 64)
353 if (!consumerd64_bin[0]) {
354 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
355 }
356 if (!consumerd64_libdir[0]) {
357 consumerd64_libdir = INSTALL_LIB_PATH;
358 }
359 #else
360 #error "Unknown bitness"
361 #endif
362
363 /*
364 * runtime env. var. overrides the build default.
365 */
366 bin = getenv("LTTNG_CONSUMERD32_BIN");
367 if (bin) {
368 consumerd32_bin = bin;
369 }
370 bin = getenv("LTTNG_CONSUMERD64_BIN");
371 if (bin) {
372 consumerd64_bin = bin;
373 }
374 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
375 if (libdir) {
376 consumerd32_libdir = libdir;
377 }
378 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
379 if (libdir) {
380 consumerd64_libdir = libdir;
381 }
382 }
383
384 /*
385 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
386 */
387 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
388 {
389 int ret;
390
391 assert(events);
392
393 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
394 if (ret < 0) {
395 goto error;
396 }
397
398 /* Add quit pipe */
399 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
400 if (ret < 0) {
401 goto error;
402 }
403
404 return 0;
405
406 error:
407 return ret;
408 }
409
410 /*
411 * Check if the thread quit pipe was triggered.
412 *
413 * Return 1 if it was triggered else 0;
414 */
415 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
416 {
417 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
418 return 1;
419 }
420
421 return 0;
422 }
423
424 /*
425 * Init thread quit pipe.
426 *
427 * Return -1 on error or 0 if all pipes are created.
428 */
429 static int init_thread_quit_pipe(void)
430 {
431 int ret, i;
432
433 ret = pipe(thread_quit_pipe);
434 if (ret < 0) {
435 PERROR("thread quit pipe");
436 goto error;
437 }
438
439 for (i = 0; i < 2; i++) {
440 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
441 if (ret < 0) {
442 PERROR("fcntl");
443 goto error;
444 }
445 }
446
447 error:
448 return ret;
449 }
450
451 /*
452 * Stop all threads by closing the thread quit pipe.
453 */
454 static void stop_threads(void)
455 {
456 int ret;
457
458 /* Stopping all threads */
459 DBG("Terminating all threads");
460 ret = notify_thread_pipe(thread_quit_pipe[1]);
461 if (ret < 0) {
462 ERR("write error on thread quit pipe");
463 }
464
465 /* Dispatch thread */
466 CMM_STORE_SHARED(dispatch_thread_exit, 1);
467 futex_nto1_wake(&ust_cmd_queue.futex);
468 }
469
470 /*
471 * Close every consumer sockets.
472 */
473 static void close_consumer_sockets(void)
474 {
475 int ret;
476
477 if (kconsumer_data.err_sock >= 0) {
478 ret = close(kconsumer_data.err_sock);
479 if (ret < 0) {
480 PERROR("kernel consumer err_sock close");
481 }
482 }
483 if (ustconsumer32_data.err_sock >= 0) {
484 ret = close(ustconsumer32_data.err_sock);
485 if (ret < 0) {
486 PERROR("UST consumerd32 err_sock close");
487 }
488 }
489 if (ustconsumer64_data.err_sock >= 0) {
490 ret = close(ustconsumer64_data.err_sock);
491 if (ret < 0) {
492 PERROR("UST consumerd64 err_sock close");
493 }
494 }
495 if (kconsumer_data.cmd_sock >= 0) {
496 ret = close(kconsumer_data.cmd_sock);
497 if (ret < 0) {
498 PERROR("kernel consumer cmd_sock close");
499 }
500 }
501 if (ustconsumer32_data.cmd_sock >= 0) {
502 ret = close(ustconsumer32_data.cmd_sock);
503 if (ret < 0) {
504 PERROR("UST consumerd32 cmd_sock close");
505 }
506 }
507 if (ustconsumer64_data.cmd_sock >= 0) {
508 ret = close(ustconsumer64_data.cmd_sock);
509 if (ret < 0) {
510 PERROR("UST consumerd64 cmd_sock close");
511 }
512 }
513 }
514
515 /*
516 * Generate the full lock file path using the rundir.
517 *
518 * Return the snprintf() return value thus a negative value is an error.
519 */
520 static int generate_lock_file_path(char *path, size_t len)
521 {
522 int ret;
523
524 assert(path);
525 assert(rundir);
526
527 /* Build lockfile path from rundir. */
528 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
529 if (ret < 0) {
530 PERROR("snprintf lockfile path");
531 }
532
533 return ret;
534 }
535
536 /*
537 * Cleanup the daemon
538 */
539 static void cleanup(void)
540 {
541 int ret;
542 struct ltt_session *sess, *stmp;
543 char path[PATH_MAX];
544
545 DBG("Cleaning up");
546
547 /*
548 * Close the thread quit pipe. It has already done its job,
549 * since we are now called.
550 */
551 utils_close_pipe(thread_quit_pipe);
552
553 /*
554 * If opt_pidfile is undefined, the default file will be wiped when
555 * removing the rundir.
556 */
557 if (opt_pidfile) {
558 ret = remove(opt_pidfile);
559 if (ret < 0) {
560 PERROR("remove pidfile %s", opt_pidfile);
561 }
562 }
563
564 DBG("Removing sessiond and consumerd content of directory %s", rundir);
565
566 /* sessiond */
567 snprintf(path, PATH_MAX,
568 "%s/%s",
569 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
570 DBG("Removing %s", path);
571 (void) unlink(path);
572
573 snprintf(path, PATH_MAX, "%s/%s", rundir,
574 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
575 DBG("Removing %s", path);
576 (void) unlink(path);
577
578 /* kconsumerd */
579 snprintf(path, PATH_MAX,
580 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
581 rundir);
582 DBG("Removing %s", path);
583 (void) unlink(path);
584
585 snprintf(path, PATH_MAX,
586 DEFAULT_KCONSUMERD_PATH,
587 rundir);
588 DBG("Removing directory %s", path);
589 (void) rmdir(path);
590
591 /* ust consumerd 32 */
592 snprintf(path, PATH_MAX,
593 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
594 rundir);
595 DBG("Removing %s", path);
596 (void) unlink(path);
597
598 snprintf(path, PATH_MAX,
599 DEFAULT_USTCONSUMERD32_PATH,
600 rundir);
601 DBG("Removing directory %s", path);
602 (void) rmdir(path);
603
604 /* ust consumerd 64 */
605 snprintf(path, PATH_MAX,
606 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
607 rundir);
608 DBG("Removing %s", path);
609 (void) unlink(path);
610
611 snprintf(path, PATH_MAX,
612 DEFAULT_USTCONSUMERD64_PATH,
613 rundir);
614 DBG("Removing directory %s", path);
615 (void) rmdir(path);
616
617 DBG("Cleaning up all sessions");
618
619 /* Destroy session list mutex */
620 if (session_list_ptr != NULL) {
621 pthread_mutex_destroy(&session_list_ptr->lock);
622
623 /* Cleanup ALL session */
624 cds_list_for_each_entry_safe(sess, stmp,
625 &session_list_ptr->head, list) {
626 cmd_destroy_session(sess, kernel_poll_pipe[1]);
627 }
628 }
629
630 DBG("Closing all UST sockets");
631 ust_app_clean_list();
632 buffer_reg_destroy_registries();
633
634 if (is_root && !opt_no_kernel) {
635 DBG2("Closing kernel fd");
636 if (kernel_tracer_fd >= 0) {
637 ret = close(kernel_tracer_fd);
638 if (ret) {
639 PERROR("close");
640 }
641 }
642 DBG("Unloading kernel modules");
643 modprobe_remove_lttng_all();
644 }
645
646 close_consumer_sockets();
647
648 /*
649 * If the override option is set, the pointer points to a *non* const thus
650 * freeing it even though the variable type is set to const.
651 */
652 if (tracing_group_name_override) {
653 free((void *) tracing_group_name);
654 }
655 if (consumerd32_bin_override) {
656 free((void *) consumerd32_bin);
657 }
658 if (consumerd64_bin_override) {
659 free((void *) consumerd64_bin);
660 }
661 if (consumerd32_libdir_override) {
662 free((void *) consumerd32_libdir);
663 }
664 if (consumerd64_libdir_override) {
665 free((void *) consumerd64_libdir);
666 }
667
668 if (opt_pidfile) {
669 free(opt_pidfile);
670 }
671
672 if (opt_load_session_path) {
673 free(opt_load_session_path);
674 }
675
676 if (load_info) {
677 load_session_destroy_data(load_info);
678 free(load_info);
679 }
680
681 /*
682 * Cleanup lock file by deleting it and finaly closing it which will
683 * release the file system lock.
684 */
685 if (lockfile_fd >= 0) {
686 char lockfile_path[PATH_MAX];
687
688 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
689 if (ret > 0) {
690 ret = remove(lockfile_path);
691 if (ret < 0) {
692 PERROR("remove lock file");
693 }
694 ret = close(lockfile_fd);
695 if (ret < 0) {
696 PERROR("close lock file");
697 }
698 }
699 }
700
701 /*
702 * We do NOT rmdir rundir because there are other processes
703 * using it, for instance lttng-relayd, which can start in
704 * parallel with this teardown.
705 */
706
707 free(rundir);
708
709 /* <fun> */
710 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
711 "Matthew, BEET driven development works!%c[%dm",
712 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
713 /* </fun> */
714 }
715
716 /*
717 * Send data on a unix socket using the liblttsessiondcomm API.
718 *
719 * Return lttcomm error code.
720 */
721 static int send_unix_sock(int sock, void *buf, size_t len)
722 {
723 /* Check valid length */
724 if (len == 0) {
725 return -1;
726 }
727
728 return lttcomm_send_unix_sock(sock, buf, len);
729 }
730
731 /*
732 * Free memory of a command context structure.
733 */
734 static void clean_command_ctx(struct command_ctx **cmd_ctx)
735 {
736 DBG("Clean command context structure");
737 if (*cmd_ctx) {
738 if ((*cmd_ctx)->llm) {
739 free((*cmd_ctx)->llm);
740 }
741 if ((*cmd_ctx)->lsm) {
742 free((*cmd_ctx)->lsm);
743 }
744 free(*cmd_ctx);
745 *cmd_ctx = NULL;
746 }
747 }
748
749 /*
750 * Notify UST applications using the shm mmap futex.
751 */
752 static int notify_ust_apps(int active)
753 {
754 char *wait_shm_mmap;
755
756 DBG("Notifying applications of session daemon state: %d", active);
757
758 /* See shm.c for this call implying mmap, shm and futex calls */
759 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
760 if (wait_shm_mmap == NULL) {
761 goto error;
762 }
763
764 /* Wake waiting process */
765 futex_wait_update((int32_t *) wait_shm_mmap, active);
766
767 /* Apps notified successfully */
768 return 0;
769
770 error:
771 return -1;
772 }
773
774 /*
775 * Setup the outgoing data buffer for the response (llm) by allocating the
776 * right amount of memory and copying the original information from the lsm
777 * structure.
778 *
779 * Return total size of the buffer pointed by buf.
780 */
781 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
782 {
783 int ret, buf_size;
784
785 buf_size = size;
786
787 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
788 if (cmd_ctx->llm == NULL) {
789 PERROR("zmalloc");
790 ret = -ENOMEM;
791 goto error;
792 }
793
794 /* Copy common data */
795 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
796 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
797
798 cmd_ctx->llm->data_size = size;
799 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
800
801 return buf_size;
802
803 error:
804 return ret;
805 }
806
807 /*
808 * Update the kernel poll set of all channel fd available over all tracing
809 * session. Add the wakeup pipe at the end of the set.
810 */
811 static int update_kernel_poll(struct lttng_poll_event *events)
812 {
813 int ret;
814 struct ltt_session *session;
815 struct ltt_kernel_channel *channel;
816
817 DBG("Updating kernel poll set");
818
819 session_lock_list();
820 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
821 session_lock(session);
822 if (session->kernel_session == NULL) {
823 session_unlock(session);
824 continue;
825 }
826
827 cds_list_for_each_entry(channel,
828 &session->kernel_session->channel_list.head, list) {
829 /* Add channel fd to the kernel poll set */
830 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
831 if (ret < 0) {
832 session_unlock(session);
833 goto error;
834 }
835 DBG("Channel fd %d added to kernel set", channel->fd);
836 }
837 session_unlock(session);
838 }
839 session_unlock_list();
840
841 return 0;
842
843 error:
844 session_unlock_list();
845 return -1;
846 }
847
848 /*
849 * Find the channel fd from 'fd' over all tracing session. When found, check
850 * for new channel stream and send those stream fds to the kernel consumer.
851 *
852 * Useful for CPU hotplug feature.
853 */
854 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
855 {
856 int ret = 0;
857 struct ltt_session *session;
858 struct ltt_kernel_session *ksess;
859 struct ltt_kernel_channel *channel;
860
861 DBG("Updating kernel streams for channel fd %d", fd);
862
863 session_lock_list();
864 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
865 session_lock(session);
866 if (session->kernel_session == NULL) {
867 session_unlock(session);
868 continue;
869 }
870 ksess = session->kernel_session;
871
872 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
873 if (channel->fd == fd) {
874 DBG("Channel found, updating kernel streams");
875 ret = kernel_open_channel_stream(channel);
876 if (ret < 0) {
877 goto error;
878 }
879 /* Update the stream global counter */
880 ksess->stream_count_global += ret;
881
882 /*
883 * Have we already sent fds to the consumer? If yes, it means
884 * that tracing is started so it is safe to send our updated
885 * stream fds.
886 */
887 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
888 struct lttng_ht_iter iter;
889 struct consumer_socket *socket;
890
891 rcu_read_lock();
892 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
893 &iter.iter, socket, node.node) {
894 pthread_mutex_lock(socket->lock);
895 ret = kernel_consumer_send_channel_stream(socket,
896 channel, ksess,
897 session->output_traces ? 1 : 0);
898 pthread_mutex_unlock(socket->lock);
899 if (ret < 0) {
900 rcu_read_unlock();
901 goto error;
902 }
903 }
904 rcu_read_unlock();
905 }
906 goto error;
907 }
908 }
909 session_unlock(session);
910 }
911 session_unlock_list();
912 return ret;
913
914 error:
915 session_unlock(session);
916 session_unlock_list();
917 return ret;
918 }
919
920 /*
921 * For each tracing session, update newly registered apps. The session list
922 * lock MUST be acquired before calling this.
923 */
924 static void update_ust_app(int app_sock)
925 {
926 struct ltt_session *sess, *stmp;
927
928 /* Consumer is in an ERROR state. Stop any application update. */
929 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
930 /* Stop the update process since the consumer is dead. */
931 return;
932 }
933
934 /* For all tracing session(s) */
935 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
936 session_lock(sess);
937 if (sess->ust_session) {
938 ust_app_global_update(sess->ust_session, app_sock);
939 }
940 session_unlock(sess);
941 }
942 }
943
944 /*
945 * This thread manage event coming from the kernel.
946 *
947 * Features supported in this thread:
948 * -) CPU Hotplug
949 */
950 static void *thread_manage_kernel(void *data)
951 {
952 int ret, i, pollfd, update_poll_flag = 1, err = -1;
953 uint32_t revents, nb_fd;
954 char tmp;
955 struct lttng_poll_event events;
956
957 DBG("[thread] Thread manage kernel started");
958
959 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
960
961 /*
962 * This first step of the while is to clean this structure which could free
963 * non NULL pointers so initialize it before the loop.
964 */
965 lttng_poll_init(&events);
966
967 if (testpoint(sessiond_thread_manage_kernel)) {
968 goto error_testpoint;
969 }
970
971 health_code_update();
972
973 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
974 goto error_testpoint;
975 }
976
977 while (1) {
978 health_code_update();
979
980 if (update_poll_flag == 1) {
981 /* Clean events object. We are about to populate it again. */
982 lttng_poll_clean(&events);
983
984 ret = sessiond_set_thread_pollset(&events, 2);
985 if (ret < 0) {
986 goto error_poll_create;
987 }
988
989 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
990 if (ret < 0) {
991 goto error;
992 }
993
994 /* This will add the available kernel channel if any. */
995 ret = update_kernel_poll(&events);
996 if (ret < 0) {
997 goto error;
998 }
999 update_poll_flag = 0;
1000 }
1001
1002 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
1003
1004 /* Poll infinite value of time */
1005 restart:
1006 health_poll_entry();
1007 ret = lttng_poll_wait(&events, -1);
1008 health_poll_exit();
1009 if (ret < 0) {
1010 /*
1011 * Restart interrupted system call.
1012 */
1013 if (errno == EINTR) {
1014 goto restart;
1015 }
1016 goto error;
1017 } else if (ret == 0) {
1018 /* Should not happen since timeout is infinite */
1019 ERR("Return value of poll is 0 with an infinite timeout.\n"
1020 "This should not have happened! Continuing...");
1021 continue;
1022 }
1023
1024 nb_fd = ret;
1025
1026 for (i = 0; i < nb_fd; i++) {
1027 /* Fetch once the poll data */
1028 revents = LTTNG_POLL_GETEV(&events, i);
1029 pollfd = LTTNG_POLL_GETFD(&events, i);
1030
1031 health_code_update();
1032
1033 /* Thread quit pipe has been closed. Killing thread. */
1034 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1035 if (ret) {
1036 err = 0;
1037 goto exit;
1038 }
1039
1040 /* Check for data on kernel pipe */
1041 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1042 (void) lttng_read(kernel_poll_pipe[0],
1043 &tmp, 1);
1044 /*
1045 * Ret value is useless here, if this pipe gets any actions an
1046 * update is required anyway.
1047 */
1048 update_poll_flag = 1;
1049 continue;
1050 } else {
1051 /*
1052 * New CPU detected by the kernel. Adding kernel stream to
1053 * kernel session and updating the kernel consumer
1054 */
1055 if (revents & LPOLLIN) {
1056 ret = update_kernel_stream(&kconsumer_data, pollfd);
1057 if (ret < 0) {
1058 continue;
1059 }
1060 break;
1061 /*
1062 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1063 * and unregister kernel stream at this point.
1064 */
1065 }
1066 }
1067 }
1068 }
1069
1070 exit:
1071 error:
1072 lttng_poll_clean(&events);
1073 error_poll_create:
1074 error_testpoint:
1075 utils_close_pipe(kernel_poll_pipe);
1076 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1077 if (err) {
1078 health_error();
1079 ERR("Health error occurred in %s", __func__);
1080 WARN("Kernel thread died unexpectedly. "
1081 "Kernel tracing can continue but CPU hotplug is disabled.");
1082 }
1083 health_unregister(health_sessiond);
1084 DBG("Kernel thread dying");
1085 return NULL;
1086 }
1087
1088 /*
1089 * Signal pthread condition of the consumer data that the thread.
1090 */
1091 static void signal_consumer_condition(struct consumer_data *data, int state)
1092 {
1093 pthread_mutex_lock(&data->cond_mutex);
1094
1095 /*
1096 * The state is set before signaling. It can be any value, it's the waiter
1097 * job to correctly interpret this condition variable associated to the
1098 * consumer pthread_cond.
1099 *
1100 * A value of 0 means that the corresponding thread of the consumer data
1101 * was not started. 1 indicates that the thread has started and is ready
1102 * for action. A negative value means that there was an error during the
1103 * thread bootstrap.
1104 */
1105 data->consumer_thread_is_ready = state;
1106 (void) pthread_cond_signal(&data->cond);
1107
1108 pthread_mutex_unlock(&data->cond_mutex);
1109 }
1110
1111 /*
1112 * This thread manage the consumer error sent back to the session daemon.
1113 */
1114 static void *thread_manage_consumer(void *data)
1115 {
1116 int sock = -1, i, ret, pollfd, err = -1;
1117 uint32_t revents, nb_fd;
1118 enum lttcomm_return_code code;
1119 struct lttng_poll_event events;
1120 struct consumer_data *consumer_data = data;
1121
1122 DBG("[thread] Manage consumer started");
1123
1124 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1125
1126 health_code_update();
1127
1128 /*
1129 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1130 * metadata_sock. Nothing more will be added to this poll set.
1131 */
1132 ret = sessiond_set_thread_pollset(&events, 3);
1133 if (ret < 0) {
1134 goto error_poll;
1135 }
1136
1137 /*
1138 * The error socket here is already in a listening state which was done
1139 * just before spawning this thread to avoid a race between the consumer
1140 * daemon exec trying to connect and the listen() call.
1141 */
1142 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1143 if (ret < 0) {
1144 goto error;
1145 }
1146
1147 health_code_update();
1148
1149 /* Infinite blocking call, waiting for transmission */
1150 restart:
1151 health_poll_entry();
1152
1153 if (testpoint(sessiond_thread_manage_consumer)) {
1154 goto error;
1155 }
1156
1157 ret = lttng_poll_wait(&events, -1);
1158 health_poll_exit();
1159 if (ret < 0) {
1160 /*
1161 * Restart interrupted system call.
1162 */
1163 if (errno == EINTR) {
1164 goto restart;
1165 }
1166 goto error;
1167 }
1168
1169 nb_fd = ret;
1170
1171 for (i = 0; i < nb_fd; i++) {
1172 /* Fetch once the poll data */
1173 revents = LTTNG_POLL_GETEV(&events, i);
1174 pollfd = LTTNG_POLL_GETFD(&events, i);
1175
1176 health_code_update();
1177
1178 /* Thread quit pipe has been closed. Killing thread. */
1179 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1180 if (ret) {
1181 err = 0;
1182 goto exit;
1183 }
1184
1185 /* Event on the registration socket */
1186 if (pollfd == consumer_data->err_sock) {
1187 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1188 ERR("consumer err socket poll error");
1189 goto error;
1190 }
1191 }
1192 }
1193
1194 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1195 if (sock < 0) {
1196 goto error;
1197 }
1198
1199 /*
1200 * Set the CLOEXEC flag. Return code is useless because either way, the
1201 * show must go on.
1202 */
1203 (void) utils_set_fd_cloexec(sock);
1204
1205 health_code_update();
1206
1207 DBG2("Receiving code from consumer err_sock");
1208
1209 /* Getting status code from kconsumerd */
1210 ret = lttcomm_recv_unix_sock(sock, &code,
1211 sizeof(enum lttcomm_return_code));
1212 if (ret <= 0) {
1213 goto error;
1214 }
1215
1216 health_code_update();
1217 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1218 /* Connect both socket, command and metadata. */
1219 consumer_data->cmd_sock =
1220 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1221 consumer_data->metadata_fd =
1222 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1223 if (consumer_data->cmd_sock < 0
1224 || consumer_data->metadata_fd < 0) {
1225 PERROR("consumer connect cmd socket");
1226 /* On error, signal condition and quit. */
1227 signal_consumer_condition(consumer_data, -1);
1228 goto error;
1229 }
1230 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1231 /* Create metadata socket lock. */
1232 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1233 if (consumer_data->metadata_sock.lock == NULL) {
1234 PERROR("zmalloc pthread mutex");
1235 ret = -1;
1236 goto error;
1237 }
1238 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1239
1240 signal_consumer_condition(consumer_data, 1);
1241 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1242 DBG("Consumer metadata socket ready (fd: %d)",
1243 consumer_data->metadata_fd);
1244 } else {
1245 ERR("consumer error when waiting for SOCK_READY : %s",
1246 lttcomm_get_readable_code(-code));
1247 goto error;
1248 }
1249
1250 /* Remove the consumerd error sock since we've established a connexion */
1251 ret = lttng_poll_del(&events, consumer_data->err_sock);
1252 if (ret < 0) {
1253 goto error;
1254 }
1255
1256 /* Add new accepted error socket. */
1257 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1258 if (ret < 0) {
1259 goto error;
1260 }
1261
1262 /* Add metadata socket that is successfully connected. */
1263 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1264 LPOLLIN | LPOLLRDHUP);
1265 if (ret < 0) {
1266 goto error;
1267 }
1268
1269 health_code_update();
1270
1271 /* Infinite blocking call, waiting for transmission */
1272 restart_poll:
1273 while (1) {
1274 health_poll_entry();
1275 ret = lttng_poll_wait(&events, -1);
1276 health_poll_exit();
1277 if (ret < 0) {
1278 /*
1279 * Restart interrupted system call.
1280 */
1281 if (errno == EINTR) {
1282 goto restart_poll;
1283 }
1284 goto error;
1285 }
1286
1287 nb_fd = ret;
1288
1289 for (i = 0; i < nb_fd; i++) {
1290 /* Fetch once the poll data */
1291 revents = LTTNG_POLL_GETEV(&events, i);
1292 pollfd = LTTNG_POLL_GETFD(&events, i);
1293
1294 health_code_update();
1295
1296 /* Thread quit pipe has been closed. Killing thread. */
1297 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1298 if (ret) {
1299 err = 0;
1300 goto exit;
1301 }
1302
1303 if (pollfd == sock) {
1304 /* Event on the consumerd socket */
1305 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1306 ERR("consumer err socket second poll error");
1307 goto error;
1308 }
1309 health_code_update();
1310 /* Wait for any kconsumerd error */
1311 ret = lttcomm_recv_unix_sock(sock, &code,
1312 sizeof(enum lttcomm_return_code));
1313 if (ret <= 0) {
1314 ERR("consumer closed the command socket");
1315 goto error;
1316 }
1317
1318 ERR("consumer return code : %s",
1319 lttcomm_get_readable_code(-code));
1320
1321 goto exit;
1322 } else if (pollfd == consumer_data->metadata_fd) {
1323 /* UST metadata requests */
1324 ret = ust_consumer_metadata_request(
1325 &consumer_data->metadata_sock);
1326 if (ret < 0) {
1327 ERR("Handling metadata request");
1328 goto error;
1329 }
1330 break;
1331 } else {
1332 ERR("Unknown pollfd");
1333 goto error;
1334 }
1335 }
1336 health_code_update();
1337 }
1338
1339 exit:
1340 error:
1341 /*
1342 * We lock here because we are about to close the sockets and some other
1343 * thread might be using them so get exclusive access which will abort all
1344 * other consumer command by other threads.
1345 */
1346 pthread_mutex_lock(&consumer_data->lock);
1347
1348 /* Immediately set the consumerd state to stopped */
1349 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1350 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1351 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1352 consumer_data->type == LTTNG_CONSUMER32_UST) {
1353 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1354 } else {
1355 /* Code flow error... */
1356 assert(0);
1357 }
1358
1359 if (consumer_data->err_sock >= 0) {
1360 ret = close(consumer_data->err_sock);
1361 if (ret) {
1362 PERROR("close");
1363 }
1364 consumer_data->err_sock = -1;
1365 }
1366 if (consumer_data->cmd_sock >= 0) {
1367 ret = close(consumer_data->cmd_sock);
1368 if (ret) {
1369 PERROR("close");
1370 }
1371 consumer_data->cmd_sock = -1;
1372 }
1373 if (consumer_data->metadata_sock.fd_ptr &&
1374 *consumer_data->metadata_sock.fd_ptr >= 0) {
1375 ret = close(*consumer_data->metadata_sock.fd_ptr);
1376 if (ret) {
1377 PERROR("close");
1378 }
1379 }
1380 if (sock >= 0) {
1381 ret = close(sock);
1382 if (ret) {
1383 PERROR("close");
1384 }
1385 }
1386
1387 unlink(consumer_data->err_unix_sock_path);
1388 unlink(consumer_data->cmd_unix_sock_path);
1389 consumer_data->pid = 0;
1390 pthread_mutex_unlock(&consumer_data->lock);
1391
1392 /* Cleanup metadata socket mutex. */
1393 if (consumer_data->metadata_sock.lock) {
1394 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1395 free(consumer_data->metadata_sock.lock);
1396 }
1397 lttng_poll_clean(&events);
1398 error_poll:
1399 if (err) {
1400 health_error();
1401 ERR("Health error occurred in %s", __func__);
1402 }
1403 health_unregister(health_sessiond);
1404 DBG("consumer thread cleanup completed");
1405
1406 return NULL;
1407 }
1408
1409 /*
1410 * This thread manage application communication.
1411 */
1412 static void *thread_manage_apps(void *data)
1413 {
1414 int i, ret, pollfd, err = -1;
1415 ssize_t size_ret;
1416 uint32_t revents, nb_fd;
1417 struct lttng_poll_event events;
1418
1419 DBG("[thread] Manage application started");
1420
1421 rcu_register_thread();
1422 rcu_thread_online();
1423
1424 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1425
1426 if (testpoint(sessiond_thread_manage_apps)) {
1427 goto error_testpoint;
1428 }
1429
1430 health_code_update();
1431
1432 ret = sessiond_set_thread_pollset(&events, 2);
1433 if (ret < 0) {
1434 goto error_poll_create;
1435 }
1436
1437 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1438 if (ret < 0) {
1439 goto error;
1440 }
1441
1442 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1443 goto error;
1444 }
1445
1446 health_code_update();
1447
1448 while (1) {
1449 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1450
1451 /* Inifinite blocking call, waiting for transmission */
1452 restart:
1453 health_poll_entry();
1454 ret = lttng_poll_wait(&events, -1);
1455 health_poll_exit();
1456 if (ret < 0) {
1457 /*
1458 * Restart interrupted system call.
1459 */
1460 if (errno == EINTR) {
1461 goto restart;
1462 }
1463 goto error;
1464 }
1465
1466 nb_fd = ret;
1467
1468 for (i = 0; i < nb_fd; i++) {
1469 /* Fetch once the poll data */
1470 revents = LTTNG_POLL_GETEV(&events, i);
1471 pollfd = LTTNG_POLL_GETFD(&events, i);
1472
1473 health_code_update();
1474
1475 /* Thread quit pipe has been closed. Killing thread. */
1476 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1477 if (ret) {
1478 err = 0;
1479 goto exit;
1480 }
1481
1482 /* Inspect the apps cmd pipe */
1483 if (pollfd == apps_cmd_pipe[0]) {
1484 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1485 ERR("Apps command pipe error");
1486 goto error;
1487 } else if (revents & LPOLLIN) {
1488 int sock;
1489
1490 /* Empty pipe */
1491 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1492 if (size_ret < sizeof(sock)) {
1493 PERROR("read apps cmd pipe");
1494 goto error;
1495 }
1496
1497 health_code_update();
1498
1499 /*
1500 * We only monitor the error events of the socket. This
1501 * thread does not handle any incoming data from UST
1502 * (POLLIN).
1503 */
1504 ret = lttng_poll_add(&events, sock,
1505 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1506 if (ret < 0) {
1507 goto error;
1508 }
1509
1510 DBG("Apps with sock %d added to poll set", sock);
1511 }
1512 } else {
1513 /*
1514 * At this point, we know that a registered application made
1515 * the event at poll_wait.
1516 */
1517 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1518 /* Removing from the poll set */
1519 ret = lttng_poll_del(&events, pollfd);
1520 if (ret < 0) {
1521 goto error;
1522 }
1523
1524 /* Socket closed on remote end. */
1525 ust_app_unregister(pollfd);
1526 }
1527 }
1528
1529 health_code_update();
1530 }
1531 }
1532
1533 exit:
1534 error:
1535 lttng_poll_clean(&events);
1536 error_poll_create:
1537 error_testpoint:
1538 utils_close_pipe(apps_cmd_pipe);
1539 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1540
1541 /*
1542 * We don't clean the UST app hash table here since already registered
1543 * applications can still be controlled so let them be until the session
1544 * daemon dies or the applications stop.
1545 */
1546
1547 if (err) {
1548 health_error();
1549 ERR("Health error occurred in %s", __func__);
1550 }
1551 health_unregister(health_sessiond);
1552 DBG("Application communication apps thread cleanup complete");
1553 rcu_thread_offline();
1554 rcu_unregister_thread();
1555 return NULL;
1556 }
1557
1558 /*
1559 * Send a socket to a thread This is called from the dispatch UST registration
1560 * thread once all sockets are set for the application.
1561 *
1562 * The sock value can be invalid, we don't really care, the thread will handle
1563 * it and make the necessary cleanup if so.
1564 *
1565 * On success, return 0 else a negative value being the errno message of the
1566 * write().
1567 */
1568 static int send_socket_to_thread(int fd, int sock)
1569 {
1570 ssize_t ret;
1571
1572 /*
1573 * It's possible that the FD is set as invalid with -1 concurrently just
1574 * before calling this function being a shutdown state of the thread.
1575 */
1576 if (fd < 0) {
1577 ret = -EBADF;
1578 goto error;
1579 }
1580
1581 ret = lttng_write(fd, &sock, sizeof(sock));
1582 if (ret < sizeof(sock)) {
1583 PERROR("write apps pipe %d", fd);
1584 if (ret < 0) {
1585 ret = -errno;
1586 }
1587 goto error;
1588 }
1589
1590 /* All good. Don't send back the write positive ret value. */
1591 ret = 0;
1592 error:
1593 return (int) ret;
1594 }
1595
1596 /*
1597 * Sanitize the wait queue of the dispatch registration thread meaning removing
1598 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1599 * notify socket is never received.
1600 */
1601 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1602 {
1603 int ret, nb_fd = 0, i;
1604 unsigned int fd_added = 0;
1605 struct lttng_poll_event events;
1606 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1607
1608 assert(wait_queue);
1609
1610 lttng_poll_init(&events);
1611
1612 /* Just skip everything for an empty queue. */
1613 if (!wait_queue->count) {
1614 goto end;
1615 }
1616
1617 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1618 if (ret < 0) {
1619 goto error_create;
1620 }
1621
1622 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1623 &wait_queue->head, head) {
1624 assert(wait_node->app);
1625 ret = lttng_poll_add(&events, wait_node->app->sock,
1626 LPOLLHUP | LPOLLERR);
1627 if (ret < 0) {
1628 goto error;
1629 }
1630
1631 fd_added = 1;
1632 }
1633
1634 if (!fd_added) {
1635 goto end;
1636 }
1637
1638 /*
1639 * Poll but don't block so we can quickly identify the faulty events and
1640 * clean them afterwards from the wait queue.
1641 */
1642 ret = lttng_poll_wait(&events, 0);
1643 if (ret < 0) {
1644 goto error;
1645 }
1646 nb_fd = ret;
1647
1648 for (i = 0; i < nb_fd; i++) {
1649 /* Get faulty FD. */
1650 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1651 int pollfd = LTTNG_POLL_GETFD(&events, i);
1652
1653 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1654 &wait_queue->head, head) {
1655 if (pollfd == wait_node->app->sock &&
1656 (revents & (LPOLLHUP | LPOLLERR))) {
1657 cds_list_del(&wait_node->head);
1658 wait_queue->count--;
1659 ust_app_destroy(wait_node->app);
1660 free(wait_node);
1661 break;
1662 }
1663 }
1664 }
1665
1666 if (nb_fd > 0) {
1667 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1668 }
1669
1670 end:
1671 lttng_poll_clean(&events);
1672 return;
1673
1674 error:
1675 lttng_poll_clean(&events);
1676 error_create:
1677 ERR("Unable to sanitize wait queue");
1678 return;
1679 }
1680
1681 /*
1682 * Dispatch request from the registration threads to the application
1683 * communication thread.
1684 */
1685 static void *thread_dispatch_ust_registration(void *data)
1686 {
1687 int ret, err = -1;
1688 struct cds_wfq_node *node;
1689 struct ust_command *ust_cmd = NULL;
1690 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1691 struct ust_reg_wait_queue wait_queue = {
1692 .count = 0,
1693 };
1694
1695 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1696
1697 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1698 goto error_testpoint;
1699 }
1700
1701 health_code_update();
1702
1703 CDS_INIT_LIST_HEAD(&wait_queue.head);
1704
1705 DBG("[thread] Dispatch UST command started");
1706
1707 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1708 health_code_update();
1709
1710 /* Atomically prepare the queue futex */
1711 futex_nto1_prepare(&ust_cmd_queue.futex);
1712
1713 do {
1714 struct ust_app *app = NULL;
1715 ust_cmd = NULL;
1716
1717 /*
1718 * Make sure we don't have node(s) that have hung up before receiving
1719 * the notify socket. This is to clean the list in order to avoid
1720 * memory leaks from notify socket that are never seen.
1721 */
1722 sanitize_wait_queue(&wait_queue);
1723
1724 health_code_update();
1725 /* Dequeue command for registration */
1726 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1727 if (node == NULL) {
1728 DBG("Woken up but nothing in the UST command queue");
1729 /* Continue thread execution */
1730 break;
1731 }
1732
1733 ust_cmd = caa_container_of(node, struct ust_command, node);
1734
1735 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1736 " gid:%d sock:%d name:%s (version %d.%d)",
1737 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1738 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1739 ust_cmd->sock, ust_cmd->reg_msg.name,
1740 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1741
1742 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1743 wait_node = zmalloc(sizeof(*wait_node));
1744 if (!wait_node) {
1745 PERROR("zmalloc wait_node dispatch");
1746 ret = close(ust_cmd->sock);
1747 if (ret < 0) {
1748 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1749 }
1750 lttng_fd_put(LTTNG_FD_APPS, 1);
1751 free(ust_cmd);
1752 goto error;
1753 }
1754 CDS_INIT_LIST_HEAD(&wait_node->head);
1755
1756 /* Create application object if socket is CMD. */
1757 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1758 ust_cmd->sock);
1759 if (!wait_node->app) {
1760 ret = close(ust_cmd->sock);
1761 if (ret < 0) {
1762 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1763 }
1764 lttng_fd_put(LTTNG_FD_APPS, 1);
1765 free(wait_node);
1766 free(ust_cmd);
1767 continue;
1768 }
1769 /*
1770 * Add application to the wait queue so we can set the notify
1771 * socket before putting this object in the global ht.
1772 */
1773 cds_list_add(&wait_node->head, &wait_queue.head);
1774 wait_queue.count++;
1775
1776 free(ust_cmd);
1777 /*
1778 * We have to continue here since we don't have the notify
1779 * socket and the application MUST be added to the hash table
1780 * only at that moment.
1781 */
1782 continue;
1783 } else {
1784 /*
1785 * Look for the application in the local wait queue and set the
1786 * notify socket if found.
1787 */
1788 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1789 &wait_queue.head, head) {
1790 health_code_update();
1791 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1792 wait_node->app->notify_sock = ust_cmd->sock;
1793 cds_list_del(&wait_node->head);
1794 wait_queue.count--;
1795 app = wait_node->app;
1796 free(wait_node);
1797 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1798 break;
1799 }
1800 }
1801
1802 /*
1803 * With no application at this stage the received socket is
1804 * basically useless so close it before we free the cmd data
1805 * structure for good.
1806 */
1807 if (!app) {
1808 ret = close(ust_cmd->sock);
1809 if (ret < 0) {
1810 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1811 }
1812 lttng_fd_put(LTTNG_FD_APPS, 1);
1813 }
1814 free(ust_cmd);
1815 }
1816
1817 if (app) {
1818 /*
1819 * @session_lock_list
1820 *
1821 * Lock the global session list so from the register up to the
1822 * registration done message, no thread can see the application
1823 * and change its state.
1824 */
1825 session_lock_list();
1826 rcu_read_lock();
1827
1828 /*
1829 * Add application to the global hash table. This needs to be
1830 * done before the update to the UST registry can locate the
1831 * application.
1832 */
1833 ust_app_add(app);
1834
1835 /* Set app version. This call will print an error if needed. */
1836 (void) ust_app_version(app);
1837
1838 /* Send notify socket through the notify pipe. */
1839 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1840 app->notify_sock);
1841 if (ret < 0) {
1842 rcu_read_unlock();
1843 session_unlock_list();
1844 /*
1845 * No notify thread, stop the UST tracing. However, this is
1846 * not an internal error of the this thread thus setting
1847 * the health error code to a normal exit.
1848 */
1849 err = 0;
1850 goto error;
1851 }
1852
1853 /*
1854 * Update newly registered application with the tracing
1855 * registry info already enabled information.
1856 */
1857 update_ust_app(app->sock);
1858
1859 /*
1860 * Don't care about return value. Let the manage apps threads
1861 * handle app unregistration upon socket close.
1862 */
1863 (void) ust_app_register_done(app->sock);
1864
1865 /*
1866 * Even if the application socket has been closed, send the app
1867 * to the thread and unregistration will take place at that
1868 * place.
1869 */
1870 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1871 if (ret < 0) {
1872 rcu_read_unlock();
1873 session_unlock_list();
1874 /*
1875 * No apps. thread, stop the UST tracing. However, this is
1876 * not an internal error of the this thread thus setting
1877 * the health error code to a normal exit.
1878 */
1879 err = 0;
1880 goto error;
1881 }
1882
1883 rcu_read_unlock();
1884 session_unlock_list();
1885 }
1886 } while (node != NULL);
1887
1888 health_poll_entry();
1889 /* Futex wait on queue. Blocking call on futex() */
1890 futex_nto1_wait(&ust_cmd_queue.futex);
1891 health_poll_exit();
1892 }
1893 /* Normal exit, no error */
1894 err = 0;
1895
1896 error:
1897 /* Clean up wait queue. */
1898 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1899 &wait_queue.head, head) {
1900 cds_list_del(&wait_node->head);
1901 wait_queue.count--;
1902 free(wait_node);
1903 }
1904
1905 error_testpoint:
1906 DBG("Dispatch thread dying");
1907 if (err) {
1908 health_error();
1909 ERR("Health error occurred in %s", __func__);
1910 }
1911 health_unregister(health_sessiond);
1912 return NULL;
1913 }
1914
1915 /*
1916 * This thread manage application registration.
1917 */
1918 static void *thread_registration_apps(void *data)
1919 {
1920 int sock = -1, i, ret, pollfd, err = -1;
1921 uint32_t revents, nb_fd;
1922 struct lttng_poll_event events;
1923 /*
1924 * Get allocated in this thread, enqueued to a global queue, dequeued and
1925 * freed in the manage apps thread.
1926 */
1927 struct ust_command *ust_cmd = NULL;
1928
1929 DBG("[thread] Manage application registration started");
1930
1931 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1932
1933 if (testpoint(sessiond_thread_registration_apps)) {
1934 goto error_testpoint;
1935 }
1936
1937 ret = lttcomm_listen_unix_sock(apps_sock);
1938 if (ret < 0) {
1939 goto error_listen;
1940 }
1941
1942 /*
1943 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1944 * more will be added to this poll set.
1945 */
1946 ret = sessiond_set_thread_pollset(&events, 2);
1947 if (ret < 0) {
1948 goto error_create_poll;
1949 }
1950
1951 /* Add the application registration socket */
1952 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1953 if (ret < 0) {
1954 goto error_poll_add;
1955 }
1956
1957 /* Notify all applications to register */
1958 ret = notify_ust_apps(1);
1959 if (ret < 0) {
1960 ERR("Failed to notify applications or create the wait shared memory.\n"
1961 "Execution continues but there might be problem for already\n"
1962 "running applications that wishes to register.");
1963 }
1964
1965 while (1) {
1966 DBG("Accepting application registration");
1967
1968 /* Inifinite blocking call, waiting for transmission */
1969 restart:
1970 health_poll_entry();
1971 ret = lttng_poll_wait(&events, -1);
1972 health_poll_exit();
1973 if (ret < 0) {
1974 /*
1975 * Restart interrupted system call.
1976 */
1977 if (errno == EINTR) {
1978 goto restart;
1979 }
1980 goto error;
1981 }
1982
1983 nb_fd = ret;
1984
1985 for (i = 0; i < nb_fd; i++) {
1986 health_code_update();
1987
1988 /* Fetch once the poll data */
1989 revents = LTTNG_POLL_GETEV(&events, i);
1990 pollfd = LTTNG_POLL_GETFD(&events, i);
1991
1992 /* Thread quit pipe has been closed. Killing thread. */
1993 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1994 if (ret) {
1995 err = 0;
1996 goto exit;
1997 }
1998
1999 /* Event on the registration socket */
2000 if (pollfd == apps_sock) {
2001 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2002 ERR("Register apps socket poll error");
2003 goto error;
2004 } else if (revents & LPOLLIN) {
2005 sock = lttcomm_accept_unix_sock(apps_sock);
2006 if (sock < 0) {
2007 goto error;
2008 }
2009
2010 /*
2011 * Set socket timeout for both receiving and ending.
2012 * app_socket_timeout is in seconds, whereas
2013 * lttcomm_setsockopt_rcv_timeout and
2014 * lttcomm_setsockopt_snd_timeout expect msec as
2015 * parameter.
2016 */
2017 (void) lttcomm_setsockopt_rcv_timeout(sock,
2018 app_socket_timeout * 1000);
2019 (void) lttcomm_setsockopt_snd_timeout(sock,
2020 app_socket_timeout * 1000);
2021
2022 /*
2023 * Set the CLOEXEC flag. Return code is useless because
2024 * either way, the show must go on.
2025 */
2026 (void) utils_set_fd_cloexec(sock);
2027
2028 /* Create UST registration command for enqueuing */
2029 ust_cmd = zmalloc(sizeof(struct ust_command));
2030 if (ust_cmd == NULL) {
2031 PERROR("ust command zmalloc");
2032 goto error;
2033 }
2034
2035 /*
2036 * Using message-based transmissions to ensure we don't
2037 * have to deal with partially received messages.
2038 */
2039 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2040 if (ret < 0) {
2041 ERR("Exhausted file descriptors allowed for applications.");
2042 free(ust_cmd);
2043 ret = close(sock);
2044 if (ret) {
2045 PERROR("close");
2046 }
2047 sock = -1;
2048 continue;
2049 }
2050
2051 health_code_update();
2052 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2053 if (ret < 0) {
2054 free(ust_cmd);
2055 /* Close socket of the application. */
2056 ret = close(sock);
2057 if (ret) {
2058 PERROR("close");
2059 }
2060 lttng_fd_put(LTTNG_FD_APPS, 1);
2061 sock = -1;
2062 continue;
2063 }
2064 health_code_update();
2065
2066 ust_cmd->sock = sock;
2067 sock = -1;
2068
2069 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2070 " gid:%d sock:%d name:%s (version %d.%d)",
2071 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2072 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2073 ust_cmd->sock, ust_cmd->reg_msg.name,
2074 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2075
2076 /*
2077 * Lock free enqueue the registration request. The red pill
2078 * has been taken! This apps will be part of the *system*.
2079 */
2080 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
2081
2082 /*
2083 * Wake the registration queue futex. Implicit memory
2084 * barrier with the exchange in cds_wfq_enqueue.
2085 */
2086 futex_nto1_wake(&ust_cmd_queue.futex);
2087 }
2088 }
2089 }
2090 }
2091
2092 exit:
2093 error:
2094 /* Notify that the registration thread is gone */
2095 notify_ust_apps(0);
2096
2097 if (apps_sock >= 0) {
2098 ret = close(apps_sock);
2099 if (ret) {
2100 PERROR("close");
2101 }
2102 }
2103 if (sock >= 0) {
2104 ret = close(sock);
2105 if (ret) {
2106 PERROR("close");
2107 }
2108 lttng_fd_put(LTTNG_FD_APPS, 1);
2109 }
2110 unlink(apps_unix_sock_path);
2111
2112 error_poll_add:
2113 lttng_poll_clean(&events);
2114 error_listen:
2115 error_create_poll:
2116 error_testpoint:
2117 DBG("UST Registration thread cleanup complete");
2118 if (err) {
2119 health_error();
2120 ERR("Health error occurred in %s", __func__);
2121 }
2122 health_unregister(health_sessiond);
2123
2124 return NULL;
2125 }
2126
2127 /*
2128 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2129 * exec or it will fails.
2130 */
2131 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2132 {
2133 int ret, clock_ret;
2134 struct timespec timeout;
2135
2136 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2137 consumer_data->consumer_thread_is_ready = 0;
2138
2139 /* Setup pthread condition */
2140 ret = pthread_condattr_init(&consumer_data->condattr);
2141 if (ret != 0) {
2142 errno = ret;
2143 PERROR("pthread_condattr_init consumer data");
2144 goto error;
2145 }
2146
2147 /*
2148 * Set the monotonic clock in order to make sure we DO NOT jump in time
2149 * between the clock_gettime() call and the timedwait call. See bug #324
2150 * for a more details and how we noticed it.
2151 */
2152 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2153 if (ret != 0) {
2154 errno = ret;
2155 PERROR("pthread_condattr_setclock consumer data");
2156 goto error;
2157 }
2158
2159 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2160 if (ret != 0) {
2161 errno = ret;
2162 PERROR("pthread_cond_init consumer data");
2163 goto error;
2164 }
2165
2166 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2167 consumer_data);
2168 if (ret != 0) {
2169 PERROR("pthread_create consumer");
2170 ret = -1;
2171 goto error;
2172 }
2173
2174 /* We are about to wait on a pthread condition */
2175 pthread_mutex_lock(&consumer_data->cond_mutex);
2176
2177 /* Get time for sem_timedwait absolute timeout */
2178 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2179 /*
2180 * Set the timeout for the condition timed wait even if the clock gettime
2181 * call fails since we might loop on that call and we want to avoid to
2182 * increment the timeout too many times.
2183 */
2184 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2185
2186 /*
2187 * The following loop COULD be skipped in some conditions so this is why we
2188 * set ret to 0 in order to make sure at least one round of the loop is
2189 * done.
2190 */
2191 ret = 0;
2192
2193 /*
2194 * Loop until the condition is reached or when a timeout is reached. Note
2195 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2196 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2197 * possible. This loop does not take any chances and works with both of
2198 * them.
2199 */
2200 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2201 if (clock_ret < 0) {
2202 PERROR("clock_gettime spawn consumer");
2203 /* Infinite wait for the consumerd thread to be ready */
2204 ret = pthread_cond_wait(&consumer_data->cond,
2205 &consumer_data->cond_mutex);
2206 } else {
2207 ret = pthread_cond_timedwait(&consumer_data->cond,
2208 &consumer_data->cond_mutex, &timeout);
2209 }
2210 }
2211
2212 /* Release the pthread condition */
2213 pthread_mutex_unlock(&consumer_data->cond_mutex);
2214
2215 if (ret != 0) {
2216 errno = ret;
2217 if (ret == ETIMEDOUT) {
2218 int pth_ret;
2219
2220 /*
2221 * Call has timed out so we kill the kconsumerd_thread and return
2222 * an error.
2223 */
2224 ERR("Condition timed out. The consumer thread was never ready."
2225 " Killing it");
2226 pth_ret = pthread_cancel(consumer_data->thread);
2227 if (pth_ret < 0) {
2228 PERROR("pthread_cancel consumer thread");
2229 }
2230 } else {
2231 PERROR("pthread_cond_wait failed consumer thread");
2232 }
2233 /* Caller is expecting a negative value on failure. */
2234 ret = -1;
2235 goto error;
2236 }
2237
2238 pthread_mutex_lock(&consumer_data->pid_mutex);
2239 if (consumer_data->pid == 0) {
2240 ERR("Consumerd did not start");
2241 pthread_mutex_unlock(&consumer_data->pid_mutex);
2242 goto error;
2243 }
2244 pthread_mutex_unlock(&consumer_data->pid_mutex);
2245
2246 return 0;
2247
2248 error:
2249 return ret;
2250 }
2251
2252 /*
2253 * Join consumer thread
2254 */
2255 static int join_consumer_thread(struct consumer_data *consumer_data)
2256 {
2257 void *status;
2258
2259 /* Consumer pid must be a real one. */
2260 if (consumer_data->pid > 0) {
2261 int ret;
2262 ret = kill(consumer_data->pid, SIGTERM);
2263 if (ret) {
2264 ERR("Error killing consumer daemon");
2265 return ret;
2266 }
2267 return pthread_join(consumer_data->thread, &status);
2268 } else {
2269 return 0;
2270 }
2271 }
2272
2273 /*
2274 * Fork and exec a consumer daemon (consumerd).
2275 *
2276 * Return pid if successful else -1.
2277 */
2278 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2279 {
2280 int ret;
2281 pid_t pid;
2282 const char *consumer_to_use;
2283 const char *verbosity;
2284 struct stat st;
2285
2286 DBG("Spawning consumerd");
2287
2288 pid = fork();
2289 if (pid == 0) {
2290 /*
2291 * Exec consumerd.
2292 */
2293 if (opt_verbose_consumer) {
2294 verbosity = "--verbose";
2295 } else if (lttng_opt_quiet) {
2296 verbosity = "--quiet";
2297 } else {
2298 verbosity = "";
2299 }
2300
2301 switch (consumer_data->type) {
2302 case LTTNG_CONSUMER_KERNEL:
2303 /*
2304 * Find out which consumerd to execute. We will first try the
2305 * 64-bit path, then the sessiond's installation directory, and
2306 * fallback on the 32-bit one,
2307 */
2308 DBG3("Looking for a kernel consumer at these locations:");
2309 DBG3(" 1) %s", consumerd64_bin);
2310 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2311 DBG3(" 3) %s", consumerd32_bin);
2312 if (stat(consumerd64_bin, &st) == 0) {
2313 DBG3("Found location #1");
2314 consumer_to_use = consumerd64_bin;
2315 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2316 DBG3("Found location #2");
2317 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2318 } else if (stat(consumerd32_bin, &st) == 0) {
2319 DBG3("Found location #3");
2320 consumer_to_use = consumerd32_bin;
2321 } else {
2322 DBG("Could not find any valid consumerd executable");
2323 ret = -EINVAL;
2324 break;
2325 }
2326 DBG("Using kernel consumer at: %s", consumer_to_use);
2327 ret = execl(consumer_to_use,
2328 "lttng-consumerd", verbosity, "-k",
2329 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2330 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2331 "--group", tracing_group_name,
2332 NULL);
2333 break;
2334 case LTTNG_CONSUMER64_UST:
2335 {
2336 char *tmpnew = NULL;
2337
2338 if (consumerd64_libdir[0] != '\0') {
2339 char *tmp;
2340 size_t tmplen;
2341
2342 tmp = getenv("LD_LIBRARY_PATH");
2343 if (!tmp) {
2344 tmp = "";
2345 }
2346 tmplen = strlen("LD_LIBRARY_PATH=")
2347 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2348 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2349 if (!tmpnew) {
2350 ret = -ENOMEM;
2351 goto error;
2352 }
2353 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2354 strcat(tmpnew, consumerd64_libdir);
2355 if (tmp[0] != '\0') {
2356 strcat(tmpnew, ":");
2357 strcat(tmpnew, tmp);
2358 }
2359 ret = putenv(tmpnew);
2360 if (ret) {
2361 ret = -errno;
2362 free(tmpnew);
2363 goto error;
2364 }
2365 }
2366 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2367 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2368 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2369 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2370 "--group", tracing_group_name,
2371 NULL);
2372 if (consumerd64_libdir[0] != '\0') {
2373 free(tmpnew);
2374 }
2375 break;
2376 }
2377 case LTTNG_CONSUMER32_UST:
2378 {
2379 char *tmpnew = NULL;
2380
2381 if (consumerd32_libdir[0] != '\0') {
2382 char *tmp;
2383 size_t tmplen;
2384
2385 tmp = getenv("LD_LIBRARY_PATH");
2386 if (!tmp) {
2387 tmp = "";
2388 }
2389 tmplen = strlen("LD_LIBRARY_PATH=")
2390 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2391 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2392 if (!tmpnew) {
2393 ret = -ENOMEM;
2394 goto error;
2395 }
2396 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2397 strcat(tmpnew, consumerd32_libdir);
2398 if (tmp[0] != '\0') {
2399 strcat(tmpnew, ":");
2400 strcat(tmpnew, tmp);
2401 }
2402 ret = putenv(tmpnew);
2403 if (ret) {
2404 ret = -errno;
2405 free(tmpnew);
2406 goto error;
2407 }
2408 }
2409 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2410 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2411 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2412 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2413 "--group", tracing_group_name,
2414 NULL);
2415 if (consumerd32_libdir[0] != '\0') {
2416 free(tmpnew);
2417 }
2418 break;
2419 }
2420 default:
2421 PERROR("unknown consumer type");
2422 exit(EXIT_FAILURE);
2423 }
2424 if (errno != 0) {
2425 PERROR("Consumer execl()");
2426 }
2427 /* Reaching this point, we got a failure on our execl(). */
2428 exit(EXIT_FAILURE);
2429 } else if (pid > 0) {
2430 ret = pid;
2431 } else {
2432 PERROR("start consumer fork");
2433 ret = -errno;
2434 }
2435 error:
2436 return ret;
2437 }
2438
2439 /*
2440 * Spawn the consumerd daemon and session daemon thread.
2441 */
2442 static int start_consumerd(struct consumer_data *consumer_data)
2443 {
2444 int ret;
2445
2446 /*
2447 * Set the listen() state on the socket since there is a possible race
2448 * between the exec() of the consumer daemon and this call if place in the
2449 * consumer thread. See bug #366 for more details.
2450 */
2451 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2452 if (ret < 0) {
2453 goto error;
2454 }
2455
2456 pthread_mutex_lock(&consumer_data->pid_mutex);
2457 if (consumer_data->pid != 0) {
2458 pthread_mutex_unlock(&consumer_data->pid_mutex);
2459 goto end;
2460 }
2461
2462 ret = spawn_consumerd(consumer_data);
2463 if (ret < 0) {
2464 ERR("Spawning consumerd failed");
2465 pthread_mutex_unlock(&consumer_data->pid_mutex);
2466 goto error;
2467 }
2468
2469 /* Setting up the consumer_data pid */
2470 consumer_data->pid = ret;
2471 DBG2("Consumer pid %d", consumer_data->pid);
2472 pthread_mutex_unlock(&consumer_data->pid_mutex);
2473
2474 DBG2("Spawning consumer control thread");
2475 ret = spawn_consumer_thread(consumer_data);
2476 if (ret < 0) {
2477 ERR("Fatal error spawning consumer control thread");
2478 goto error;
2479 }
2480
2481 end:
2482 return 0;
2483
2484 error:
2485 /* Cleanup already created sockets on error. */
2486 if (consumer_data->err_sock >= 0) {
2487 int err;
2488
2489 err = close(consumer_data->err_sock);
2490 if (err < 0) {
2491 PERROR("close consumer data error socket");
2492 }
2493 }
2494 return ret;
2495 }
2496
2497 /*
2498 * Setup necessary data for kernel tracer action.
2499 */
2500 static int init_kernel_tracer(void)
2501 {
2502 int ret;
2503
2504 /* Modprobe lttng kernel modules */
2505 ret = modprobe_lttng_control();
2506 if (ret < 0) {
2507 goto error;
2508 }
2509
2510 /* Open debugfs lttng */
2511 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2512 if (kernel_tracer_fd < 0) {
2513 DBG("Failed to open %s", module_proc_lttng);
2514 ret = -1;
2515 goto error_open;
2516 }
2517
2518 /* Validate kernel version */
2519 ret = kernel_validate_version(kernel_tracer_fd);
2520 if (ret < 0) {
2521 goto error_version;
2522 }
2523
2524 ret = modprobe_lttng_data();
2525 if (ret < 0) {
2526 goto error_modules;
2527 }
2528
2529 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2530 return 0;
2531
2532 error_version:
2533 modprobe_remove_lttng_control();
2534 ret = close(kernel_tracer_fd);
2535 if (ret) {
2536 PERROR("close");
2537 }
2538 kernel_tracer_fd = -1;
2539 return LTTNG_ERR_KERN_VERSION;
2540
2541 error_modules:
2542 ret = close(kernel_tracer_fd);
2543 if (ret) {
2544 PERROR("close");
2545 }
2546
2547 error_open:
2548 modprobe_remove_lttng_control();
2549
2550 error:
2551 WARN("No kernel tracer available");
2552 kernel_tracer_fd = -1;
2553 if (!is_root) {
2554 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2555 } else {
2556 return LTTNG_ERR_KERN_NA;
2557 }
2558 }
2559
2560
2561 /*
2562 * Copy consumer output from the tracing session to the domain session. The
2563 * function also applies the right modification on a per domain basis for the
2564 * trace files destination directory.
2565 *
2566 * Should *NOT* be called with RCU read-side lock held.
2567 */
2568 static int copy_session_consumer(int domain, struct ltt_session *session)
2569 {
2570 int ret;
2571 const char *dir_name;
2572 struct consumer_output *consumer;
2573
2574 assert(session);
2575 assert(session->consumer);
2576
2577 switch (domain) {
2578 case LTTNG_DOMAIN_KERNEL:
2579 DBG3("Copying tracing session consumer output in kernel session");
2580 /*
2581 * XXX: We should audit the session creation and what this function
2582 * does "extra" in order to avoid a destroy since this function is used
2583 * in the domain session creation (kernel and ust) only. Same for UST
2584 * domain.
2585 */
2586 if (session->kernel_session->consumer) {
2587 consumer_destroy_output(session->kernel_session->consumer);
2588 }
2589 session->kernel_session->consumer =
2590 consumer_copy_output(session->consumer);
2591 /* Ease our life a bit for the next part */
2592 consumer = session->kernel_session->consumer;
2593 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2594 break;
2595 case LTTNG_DOMAIN_JUL:
2596 case LTTNG_DOMAIN_UST:
2597 DBG3("Copying tracing session consumer output in UST session");
2598 if (session->ust_session->consumer) {
2599 consumer_destroy_output(session->ust_session->consumer);
2600 }
2601 session->ust_session->consumer =
2602 consumer_copy_output(session->consumer);
2603 /* Ease our life a bit for the next part */
2604 consumer = session->ust_session->consumer;
2605 dir_name = DEFAULT_UST_TRACE_DIR;
2606 break;
2607 default:
2608 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2609 goto error;
2610 }
2611
2612 /* Append correct directory to subdir */
2613 strncat(consumer->subdir, dir_name,
2614 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2615 DBG3("Copy session consumer subdir %s", consumer->subdir);
2616
2617 ret = LTTNG_OK;
2618
2619 error:
2620 return ret;
2621 }
2622
2623 /*
2624 * Create an UST session and add it to the session ust list.
2625 *
2626 * Should *NOT* be called with RCU read-side lock held.
2627 */
2628 static int create_ust_session(struct ltt_session *session,
2629 struct lttng_domain *domain)
2630 {
2631 int ret;
2632 struct ltt_ust_session *lus = NULL;
2633
2634 assert(session);
2635 assert(domain);
2636 assert(session->consumer);
2637
2638 switch (domain->type) {
2639 case LTTNG_DOMAIN_JUL:
2640 case LTTNG_DOMAIN_UST:
2641 break;
2642 default:
2643 ERR("Unknown UST domain on create session %d", domain->type);
2644 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2645 goto error;
2646 }
2647
2648 DBG("Creating UST session");
2649
2650 lus = trace_ust_create_session(session->id);
2651 if (lus == NULL) {
2652 ret = LTTNG_ERR_UST_SESS_FAIL;
2653 goto error;
2654 }
2655
2656 lus->uid = session->uid;
2657 lus->gid = session->gid;
2658 lus->output_traces = session->output_traces;
2659 lus->snapshot_mode = session->snapshot_mode;
2660 lus->live_timer_interval = session->live_timer;
2661 session->ust_session = lus;
2662
2663 /* Copy session output to the newly created UST session */
2664 ret = copy_session_consumer(domain->type, session);
2665 if (ret != LTTNG_OK) {
2666 goto error;
2667 }
2668
2669 return LTTNG_OK;
2670
2671 error:
2672 free(lus);
2673 session->ust_session = NULL;
2674 return ret;
2675 }
2676
2677 /*
2678 * Create a kernel tracer session then create the default channel.
2679 */
2680 static int create_kernel_session(struct ltt_session *session)
2681 {
2682 int ret;
2683
2684 DBG("Creating kernel session");
2685
2686 ret = kernel_create_session(session, kernel_tracer_fd);
2687 if (ret < 0) {
2688 ret = LTTNG_ERR_KERN_SESS_FAIL;
2689 goto error;
2690 }
2691
2692 /* Code flow safety */
2693 assert(session->kernel_session);
2694
2695 /* Copy session output to the newly created Kernel session */
2696 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2697 if (ret != LTTNG_OK) {
2698 goto error;
2699 }
2700
2701 /* Create directory(ies) on local filesystem. */
2702 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2703 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2704 ret = run_as_mkdir_recursive(
2705 session->kernel_session->consumer->dst.trace_path,
2706 S_IRWXU | S_IRWXG, session->uid, session->gid);
2707 if (ret < 0) {
2708 if (ret != -EEXIST) {
2709 ERR("Trace directory creation error");
2710 goto error;
2711 }
2712 }
2713 }
2714
2715 session->kernel_session->uid = session->uid;
2716 session->kernel_session->gid = session->gid;
2717 session->kernel_session->output_traces = session->output_traces;
2718 session->kernel_session->snapshot_mode = session->snapshot_mode;
2719
2720 return LTTNG_OK;
2721
2722 error:
2723 trace_kernel_destroy_session(session->kernel_session);
2724 session->kernel_session = NULL;
2725 return ret;
2726 }
2727
2728 /*
2729 * Count number of session permitted by uid/gid.
2730 */
2731 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2732 {
2733 unsigned int i = 0;
2734 struct ltt_session *session;
2735
2736 DBG("Counting number of available session for UID %d GID %d",
2737 uid, gid);
2738 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2739 /*
2740 * Only list the sessions the user can control.
2741 */
2742 if (!session_access_ok(session, uid, gid)) {
2743 continue;
2744 }
2745 i++;
2746 }
2747 return i;
2748 }
2749
2750 /*
2751 * Process the command requested by the lttng client within the command
2752 * context structure. This function make sure that the return structure (llm)
2753 * is set and ready for transmission before returning.
2754 *
2755 * Return any error encountered or 0 for success.
2756 *
2757 * "sock" is only used for special-case var. len data.
2758 *
2759 * Should *NOT* be called with RCU read-side lock held.
2760 */
2761 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2762 int *sock_error)
2763 {
2764 int ret = LTTNG_OK;
2765 int need_tracing_session = 1;
2766 int need_domain;
2767
2768 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2769
2770 *sock_error = 0;
2771
2772 switch (cmd_ctx->lsm->cmd_type) {
2773 case LTTNG_CREATE_SESSION:
2774 case LTTNG_CREATE_SESSION_SNAPSHOT:
2775 case LTTNG_CREATE_SESSION_LIVE:
2776 case LTTNG_DESTROY_SESSION:
2777 case LTTNG_LIST_SESSIONS:
2778 case LTTNG_LIST_DOMAINS:
2779 case LTTNG_START_TRACE:
2780 case LTTNG_STOP_TRACE:
2781 case LTTNG_DATA_PENDING:
2782 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2783 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2784 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2785 case LTTNG_SNAPSHOT_RECORD:
2786 case LTTNG_SAVE_SESSION:
2787 need_domain = 0;
2788 break;
2789 default:
2790 need_domain = 1;
2791 }
2792
2793 if (opt_no_kernel && need_domain
2794 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2795 if (!is_root) {
2796 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2797 } else {
2798 ret = LTTNG_ERR_KERN_NA;
2799 }
2800 goto error;
2801 }
2802
2803 /* Deny register consumer if we already have a spawned consumer. */
2804 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2805 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2806 if (kconsumer_data.pid > 0) {
2807 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2808 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2809 goto error;
2810 }
2811 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2812 }
2813
2814 /*
2815 * Check for command that don't needs to allocate a returned payload. We do
2816 * this here so we don't have to make the call for no payload at each
2817 * command.
2818 */
2819 switch(cmd_ctx->lsm->cmd_type) {
2820 case LTTNG_LIST_SESSIONS:
2821 case LTTNG_LIST_TRACEPOINTS:
2822 case LTTNG_LIST_TRACEPOINT_FIELDS:
2823 case LTTNG_LIST_DOMAINS:
2824 case LTTNG_LIST_CHANNELS:
2825 case LTTNG_LIST_EVENTS:
2826 break;
2827 default:
2828 /* Setup lttng message with no payload */
2829 ret = setup_lttng_msg(cmd_ctx, 0);
2830 if (ret < 0) {
2831 /* This label does not try to unlock the session */
2832 goto init_setup_error;
2833 }
2834 }
2835
2836 /* Commands that DO NOT need a session. */
2837 switch (cmd_ctx->lsm->cmd_type) {
2838 case LTTNG_CREATE_SESSION:
2839 case LTTNG_CREATE_SESSION_SNAPSHOT:
2840 case LTTNG_CREATE_SESSION_LIVE:
2841 case LTTNG_CALIBRATE:
2842 case LTTNG_LIST_SESSIONS:
2843 case LTTNG_LIST_TRACEPOINTS:
2844 case LTTNG_LIST_TRACEPOINT_FIELDS:
2845 case LTTNG_SAVE_SESSION:
2846 need_tracing_session = 0;
2847 break;
2848 default:
2849 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2850 /*
2851 * We keep the session list lock across _all_ commands
2852 * for now, because the per-session lock does not
2853 * handle teardown properly.
2854 */
2855 session_lock_list();
2856 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2857 if (cmd_ctx->session == NULL) {
2858 ret = LTTNG_ERR_SESS_NOT_FOUND;
2859 goto error;
2860 } else {
2861 /* Acquire lock for the session */
2862 session_lock(cmd_ctx->session);
2863 }
2864 break;
2865 }
2866
2867 /*
2868 * Commands that need a valid session but should NOT create one if none
2869 * exists. Instead of creating one and destroying it when the command is
2870 * handled, process that right before so we save some round trip in useless
2871 * code path.
2872 */
2873 switch (cmd_ctx->lsm->cmd_type) {
2874 case LTTNG_DISABLE_CHANNEL:
2875 case LTTNG_DISABLE_EVENT:
2876 case LTTNG_DISABLE_ALL_EVENT:
2877 switch (cmd_ctx->lsm->domain.type) {
2878 case LTTNG_DOMAIN_KERNEL:
2879 if (!cmd_ctx->session->kernel_session) {
2880 ret = LTTNG_ERR_NO_CHANNEL;
2881 goto error;
2882 }
2883 break;
2884 case LTTNG_DOMAIN_JUL:
2885 case LTTNG_DOMAIN_UST:
2886 if (!cmd_ctx->session->ust_session) {
2887 ret = LTTNG_ERR_NO_CHANNEL;
2888 goto error;
2889 }
2890 break;
2891 default:
2892 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2893 goto error;
2894 }
2895 default:
2896 break;
2897 }
2898
2899 if (!need_domain) {
2900 goto skip_domain;
2901 }
2902
2903 /*
2904 * Check domain type for specific "pre-action".
2905 */
2906 switch (cmd_ctx->lsm->domain.type) {
2907 case LTTNG_DOMAIN_KERNEL:
2908 if (!is_root) {
2909 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2910 goto error;
2911 }
2912
2913 /* Kernel tracer check */
2914 if (kernel_tracer_fd == -1) {
2915 /* Basically, load kernel tracer modules */
2916 ret = init_kernel_tracer();
2917 if (ret != 0) {
2918 goto error;
2919 }
2920 }
2921
2922 /* Consumer is in an ERROR state. Report back to client */
2923 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2924 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2925 goto error;
2926 }
2927
2928 /* Need a session for kernel command */
2929 if (need_tracing_session) {
2930 if (cmd_ctx->session->kernel_session == NULL) {
2931 ret = create_kernel_session(cmd_ctx->session);
2932 if (ret < 0) {
2933 ret = LTTNG_ERR_KERN_SESS_FAIL;
2934 goto error;
2935 }
2936 }
2937
2938 /* Start the kernel consumer daemon */
2939 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2940 if (kconsumer_data.pid == 0 &&
2941 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2942 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2943 ret = start_consumerd(&kconsumer_data);
2944 if (ret < 0) {
2945 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2946 goto error;
2947 }
2948 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2949 } else {
2950 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2951 }
2952
2953 /*
2954 * The consumer was just spawned so we need to add the socket to
2955 * the consumer output of the session if exist.
2956 */
2957 ret = consumer_create_socket(&kconsumer_data,
2958 cmd_ctx->session->kernel_session->consumer);
2959 if (ret < 0) {
2960 goto error;
2961 }
2962 }
2963
2964 break;
2965 case LTTNG_DOMAIN_JUL:
2966 case LTTNG_DOMAIN_UST:
2967 {
2968 if (!ust_app_supported()) {
2969 ret = LTTNG_ERR_NO_UST;
2970 goto error;
2971 }
2972 /* Consumer is in an ERROR state. Report back to client */
2973 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2974 ret = LTTNG_ERR_NO_USTCONSUMERD;
2975 goto error;
2976 }
2977
2978 if (need_tracing_session) {
2979 /* Create UST session if none exist. */
2980 if (cmd_ctx->session->ust_session == NULL) {
2981 ret = create_ust_session(cmd_ctx->session,
2982 &cmd_ctx->lsm->domain);
2983 if (ret != LTTNG_OK) {
2984 goto error;
2985 }
2986 }
2987
2988 /* Start the UST consumer daemons */
2989 /* 64-bit */
2990 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2991 if (consumerd64_bin[0] != '\0' &&
2992 ustconsumer64_data.pid == 0 &&
2993 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2994 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2995 ret = start_consumerd(&ustconsumer64_data);
2996 if (ret < 0) {
2997 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2998 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2999 goto error;
3000 }
3001
3002 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3003 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3004 } else {
3005 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3006 }
3007
3008 /*
3009 * Setup socket for consumer 64 bit. No need for atomic access
3010 * since it was set above and can ONLY be set in this thread.
3011 */
3012 ret = consumer_create_socket(&ustconsumer64_data,
3013 cmd_ctx->session->ust_session->consumer);
3014 if (ret < 0) {
3015 goto error;
3016 }
3017
3018 /* 32-bit */
3019 if (consumerd32_bin[0] != '\0' &&
3020 ustconsumer32_data.pid == 0 &&
3021 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3022 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3023 ret = start_consumerd(&ustconsumer32_data);
3024 if (ret < 0) {
3025 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3026 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3027 goto error;
3028 }
3029
3030 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3031 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3032 } else {
3033 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3034 }
3035
3036 /*
3037 * Setup socket for consumer 64 bit. No need for atomic access
3038 * since it was set above and can ONLY be set in this thread.
3039 */
3040 ret = consumer_create_socket(&ustconsumer32_data,
3041 cmd_ctx->session->ust_session->consumer);
3042 if (ret < 0) {
3043 goto error;
3044 }
3045 }
3046 break;
3047 }
3048 default:
3049 break;
3050 }
3051 skip_domain:
3052
3053 /* Validate consumer daemon state when start/stop trace command */
3054 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3055 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3056 switch (cmd_ctx->lsm->domain.type) {
3057 case LTTNG_DOMAIN_JUL:
3058 case LTTNG_DOMAIN_UST:
3059 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3060 ret = LTTNG_ERR_NO_USTCONSUMERD;
3061 goto error;
3062 }
3063 break;
3064 case LTTNG_DOMAIN_KERNEL:
3065 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3066 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3067 goto error;
3068 }
3069 break;
3070 }
3071 }
3072
3073 /*
3074 * Check that the UID or GID match that of the tracing session.
3075 * The root user can interact with all sessions.
3076 */
3077 if (need_tracing_session) {
3078 if (!session_access_ok(cmd_ctx->session,
3079 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3080 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3081 ret = LTTNG_ERR_EPERM;
3082 goto error;
3083 }
3084 }
3085
3086 /*
3087 * Send relayd information to consumer as soon as we have a domain and a
3088 * session defined.
3089 */
3090 if (cmd_ctx->session && need_domain) {
3091 /*
3092 * Setup relayd if not done yet. If the relayd information was already
3093 * sent to the consumer, this call will gracefully return.
3094 */
3095 ret = cmd_setup_relayd(cmd_ctx->session);
3096 if (ret != LTTNG_OK) {
3097 goto error;
3098 }
3099 }
3100
3101 /* Process by command type */
3102 switch (cmd_ctx->lsm->cmd_type) {
3103 case LTTNG_ADD_CONTEXT:
3104 {
3105 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3106 cmd_ctx->lsm->u.context.channel_name,
3107 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3108 break;
3109 }
3110 case LTTNG_DISABLE_CHANNEL:
3111 {
3112 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3113 cmd_ctx->lsm->u.disable.channel_name);
3114 break;
3115 }
3116 case LTTNG_DISABLE_EVENT:
3117 {
3118 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3119 cmd_ctx->lsm->u.disable.channel_name,
3120 cmd_ctx->lsm->u.disable.name);
3121 break;
3122 }
3123 case LTTNG_DISABLE_ALL_EVENT:
3124 {
3125 DBG("Disabling all events");
3126
3127 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3128 cmd_ctx->lsm->u.disable.channel_name);
3129 break;
3130 }
3131 case LTTNG_ENABLE_CHANNEL:
3132 {
3133 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3134 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3135 break;
3136 }
3137 case LTTNG_ENABLE_EVENT:
3138 {
3139 struct lttng_event_exclusion *exclusion = NULL;
3140 struct lttng_filter_bytecode *bytecode = NULL;
3141 char *filter_expression = NULL;
3142
3143 /* Handle exclusion events and receive it from the client. */
3144 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3145 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3146
3147 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3148 (count * LTTNG_SYMBOL_NAME_LEN));
3149 if (!exclusion) {
3150 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3151 goto error;
3152 }
3153
3154 DBG("Receiving var len exclusion event list from client ...");
3155 exclusion->count = count;
3156 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3157 count * LTTNG_SYMBOL_NAME_LEN);
3158 if (ret <= 0) {
3159 DBG("Nothing recv() from client var len data... continuing");
3160 *sock_error = 1;
3161 free(exclusion);
3162 ret = LTTNG_ERR_EXCLUSION_INVAL;
3163 goto error;
3164 }
3165 }
3166
3167 /* Get filter expression from client. */
3168 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3169 size_t expression_len =
3170 cmd_ctx->lsm->u.enable.expression_len;
3171
3172 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3173 ret = LTTNG_ERR_FILTER_INVAL;
3174 free(exclusion);
3175 goto error;
3176 }
3177
3178 filter_expression = zmalloc(expression_len);
3179 if (!filter_expression) {
3180 free(exclusion);
3181 ret = LTTNG_ERR_FILTER_NOMEM;
3182 goto error;
3183 }
3184
3185 /* Receive var. len. data */
3186 DBG("Receiving var len filter's expression from client ...");
3187 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3188 expression_len);
3189 if (ret <= 0) {
3190 DBG("Nothing recv() from client car len data... continuing");
3191 *sock_error = 1;
3192 free(filter_expression);
3193 free(exclusion);
3194 ret = LTTNG_ERR_FILTER_INVAL;
3195 goto error;
3196 }
3197 }
3198
3199 /* Handle filter and get bytecode from client. */
3200 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3201 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3202
3203 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3204 ret = LTTNG_ERR_FILTER_INVAL;
3205 free(exclusion);
3206 goto error;
3207 }
3208
3209 bytecode = zmalloc(bytecode_len);
3210 if (!bytecode) {
3211 free(exclusion);
3212 ret = LTTNG_ERR_FILTER_NOMEM;
3213 goto error;
3214 }
3215
3216 /* Receive var. len. data */
3217 DBG("Receiving var len filter's bytecode from client ...");
3218 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3219 if (ret <= 0) {
3220 DBG("Nothing recv() from client car len data... continuing");
3221 *sock_error = 1;
3222 free(bytecode);
3223 free(exclusion);
3224 ret = LTTNG_ERR_FILTER_INVAL;
3225 goto error;
3226 }
3227
3228 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3229 free(bytecode);
3230 free(exclusion);
3231 ret = LTTNG_ERR_FILTER_INVAL;
3232 goto error;
3233 }
3234 }
3235
3236 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3237 cmd_ctx->lsm->u.enable.channel_name,
3238 &cmd_ctx->lsm->u.enable.event,
3239 filter_expression, bytecode, exclusion,
3240 kernel_poll_pipe[1]);
3241 break;
3242 }
3243 case LTTNG_ENABLE_ALL_EVENT:
3244 {
3245 DBG("Enabling all events");
3246
3247 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
3248 cmd_ctx->lsm->u.enable.channel_name,
3249 cmd_ctx->lsm->u.enable.event.type, NULL, NULL,
3250 kernel_poll_pipe[1]);
3251 break;
3252 }
3253 case LTTNG_LIST_TRACEPOINTS:
3254 {
3255 struct lttng_event *events;
3256 ssize_t nb_events;
3257
3258 session_lock_list();
3259 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3260 session_unlock_list();
3261 if (nb_events < 0) {
3262 /* Return value is a negative lttng_error_code. */
3263 ret = -nb_events;
3264 goto error;
3265 }
3266
3267 /*
3268 * Setup lttng message with payload size set to the event list size in
3269 * bytes and then copy list into the llm payload.
3270 */
3271 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3272 if (ret < 0) {
3273 free(events);
3274 goto setup_error;
3275 }
3276
3277 /* Copy event list into message payload */
3278 memcpy(cmd_ctx->llm->payload, events,
3279 sizeof(struct lttng_event) * nb_events);
3280
3281 free(events);
3282
3283 ret = LTTNG_OK;
3284 break;
3285 }
3286 case LTTNG_LIST_TRACEPOINT_FIELDS:
3287 {
3288 struct lttng_event_field *fields;
3289 ssize_t nb_fields;
3290
3291 session_lock_list();
3292 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3293 &fields);
3294 session_unlock_list();
3295 if (nb_fields < 0) {
3296 /* Return value is a negative lttng_error_code. */
3297 ret = -nb_fields;
3298 goto error;
3299 }
3300
3301 /*
3302 * Setup lttng message with payload size set to the event list size in
3303 * bytes and then copy list into the llm payload.
3304 */
3305 ret = setup_lttng_msg(cmd_ctx,
3306 sizeof(struct lttng_event_field) * nb_fields);
3307 if (ret < 0) {
3308 free(fields);
3309 goto setup_error;
3310 }
3311
3312 /* Copy event list into message payload */
3313 memcpy(cmd_ctx->llm->payload, fields,
3314 sizeof(struct lttng_event_field) * nb_fields);
3315
3316 free(fields);
3317
3318 ret = LTTNG_OK;
3319 break;
3320 }
3321 case LTTNG_SET_CONSUMER_URI:
3322 {
3323 size_t nb_uri, len;
3324 struct lttng_uri *uris;
3325
3326 nb_uri = cmd_ctx->lsm->u.uri.size;
3327 len = nb_uri * sizeof(struct lttng_uri);
3328
3329 if (nb_uri == 0) {
3330 ret = LTTNG_ERR_INVALID;
3331 goto error;
3332 }
3333
3334 uris = zmalloc(len);
3335 if (uris == NULL) {
3336 ret = LTTNG_ERR_FATAL;
3337 goto error;
3338 }
3339
3340 /* Receive variable len data */
3341 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3342 ret = lttcomm_recv_unix_sock(sock, uris, len);
3343 if (ret <= 0) {
3344 DBG("No URIs received from client... continuing");
3345 *sock_error = 1;
3346 ret = LTTNG_ERR_SESSION_FAIL;
3347 free(uris);
3348 goto error;
3349 }
3350
3351 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3352 nb_uri, uris);
3353 if (ret != LTTNG_OK) {
3354 free(uris);
3355 goto error;
3356 }
3357
3358 /*
3359 * XXX: 0 means that this URI should be applied on the session. Should
3360 * be a DOMAIN enuam.
3361 */
3362 if (cmd_ctx->lsm->domain.type == 0) {
3363 /* Add the URI for the UST session if a consumer is present. */
3364 if (cmd_ctx->session->ust_session &&
3365 cmd_ctx->session->ust_session->consumer) {
3366 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3367 nb_uri, uris);
3368 } else if (cmd_ctx->session->kernel_session &&
3369 cmd_ctx->session->kernel_session->consumer) {
3370 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3371 cmd_ctx->session, nb_uri, uris);
3372 }
3373 }
3374
3375 free(uris);
3376
3377 break;
3378 }
3379 case LTTNG_START_TRACE:
3380 {
3381 ret = cmd_start_trace(cmd_ctx->session);
3382 break;
3383 }
3384 case LTTNG_STOP_TRACE:
3385 {
3386 ret = cmd_stop_trace(cmd_ctx->session);
3387 break;
3388 }
3389 case LTTNG_CREATE_SESSION:
3390 {
3391 size_t nb_uri, len;
3392 struct lttng_uri *uris = NULL;
3393
3394 nb_uri = cmd_ctx->lsm->u.uri.size;
3395 len = nb_uri * sizeof(struct lttng_uri);
3396
3397 if (nb_uri > 0) {
3398 uris = zmalloc(len);
3399 if (uris == NULL) {
3400 ret = LTTNG_ERR_FATAL;
3401 goto error;
3402 }
3403
3404 /* Receive variable len data */
3405 DBG("Waiting for %zu URIs from client ...", nb_uri);
3406 ret = lttcomm_recv_unix_sock(sock, uris, len);
3407 if (ret <= 0) {
3408 DBG("No URIs received from client... continuing");
3409 *sock_error = 1;
3410 ret = LTTNG_ERR_SESSION_FAIL;
3411 free(uris);
3412 goto error;
3413 }
3414
3415 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3416 DBG("Creating session with ONE network URI is a bad call");
3417 ret = LTTNG_ERR_SESSION_FAIL;
3418 free(uris);
3419 goto error;
3420 }
3421 }
3422
3423 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3424 &cmd_ctx->creds, 0);
3425
3426 free(uris);
3427
3428 break;
3429 }
3430 case LTTNG_DESTROY_SESSION:
3431 {
3432 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3433
3434 /* Set session to NULL so we do not unlock it after free. */
3435 cmd_ctx->session = NULL;
3436 break;
3437 }
3438 case LTTNG_LIST_DOMAINS:
3439 {
3440 ssize_t nb_dom;
3441 struct lttng_domain *domains;
3442
3443 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3444 if (nb_dom < 0) {
3445 /* Return value is a negative lttng_error_code. */
3446 ret = -nb_dom;
3447 goto error;
3448 }
3449
3450 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3451 if (ret < 0) {
3452 free(domains);
3453 goto setup_error;
3454 }
3455
3456 /* Copy event list into message payload */
3457 memcpy(cmd_ctx->llm->payload, domains,
3458 nb_dom * sizeof(struct lttng_domain));
3459
3460 free(domains);
3461
3462 ret = LTTNG_OK;
3463 break;
3464 }
3465 case LTTNG_LIST_CHANNELS:
3466 {
3467 int nb_chan;
3468 struct lttng_channel *channels;
3469
3470 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3471 cmd_ctx->session, &channels);
3472 if (nb_chan < 0) {
3473 /* Return value is a negative lttng_error_code. */
3474 ret = -nb_chan;
3475 goto error;
3476 }
3477
3478 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3479 if (ret < 0) {
3480 free(channels);
3481 goto setup_error;
3482 }
3483
3484 /* Copy event list into message payload */
3485 memcpy(cmd_ctx->llm->payload, channels,
3486 nb_chan * sizeof(struct lttng_channel));
3487
3488 free(channels);
3489
3490 ret = LTTNG_OK;
3491 break;
3492 }
3493 case LTTNG_LIST_EVENTS:
3494 {
3495 ssize_t nb_event;
3496 struct lttng_event *events = NULL;
3497
3498 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3499 cmd_ctx->lsm->u.list.channel_name, &events);
3500 if (nb_event < 0) {
3501 /* Return value is a negative lttng_error_code. */
3502 ret = -nb_event;
3503 goto error;
3504 }
3505
3506 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3507 if (ret < 0) {
3508 free(events);
3509 goto setup_error;
3510 }
3511
3512 /* Copy event list into message payload */
3513 memcpy(cmd_ctx->llm->payload, events,
3514 nb_event * sizeof(struct lttng_event));
3515
3516 free(events);
3517
3518 ret = LTTNG_OK;
3519 break;
3520 }
3521 case LTTNG_LIST_SESSIONS:
3522 {
3523 unsigned int nr_sessions;
3524
3525 session_lock_list();
3526 nr_sessions = lttng_sessions_count(
3527 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3528 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3529
3530 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3531 if (ret < 0) {
3532 session_unlock_list();
3533 goto setup_error;
3534 }
3535
3536 /* Filled the session array */
3537 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3538 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3539 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3540
3541 session_unlock_list();
3542
3543 ret = LTTNG_OK;
3544 break;
3545 }
3546 case LTTNG_CALIBRATE:
3547 {
3548 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3549 &cmd_ctx->lsm->u.calibrate);
3550 break;
3551 }
3552 case LTTNG_REGISTER_CONSUMER:
3553 {
3554 struct consumer_data *cdata;
3555
3556 switch (cmd_ctx->lsm->domain.type) {
3557 case LTTNG_DOMAIN_KERNEL:
3558 cdata = &kconsumer_data;
3559 break;
3560 default:
3561 ret = LTTNG_ERR_UND;
3562 goto error;
3563 }
3564
3565 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3566 cmd_ctx->lsm->u.reg.path, cdata);
3567 break;
3568 }
3569 case LTTNG_DATA_PENDING:
3570 {
3571 ret = cmd_data_pending(cmd_ctx->session);
3572 break;
3573 }
3574 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3575 {
3576 struct lttcomm_lttng_output_id reply;
3577
3578 ret = cmd_snapshot_add_output(cmd_ctx->session,
3579 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3580 if (ret != LTTNG_OK) {
3581 goto error;
3582 }
3583
3584 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3585 if (ret < 0) {
3586 goto setup_error;
3587 }
3588
3589 /* Copy output list into message payload */
3590 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3591 ret = LTTNG_OK;
3592 break;
3593 }
3594 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3595 {
3596 ret = cmd_snapshot_del_output(cmd_ctx->session,
3597 &cmd_ctx->lsm->u.snapshot_output.output);
3598 break;
3599 }
3600 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3601 {
3602 ssize_t nb_output;
3603 struct lttng_snapshot_output *outputs = NULL;
3604
3605 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3606 if (nb_output < 0) {
3607 ret = -nb_output;
3608 goto error;
3609 }
3610
3611 ret = setup_lttng_msg(cmd_ctx,
3612 nb_output * sizeof(struct lttng_snapshot_output));
3613 if (ret < 0) {
3614 free(outputs);
3615 goto setup_error;
3616 }
3617
3618 if (outputs) {
3619 /* Copy output list into message payload */
3620 memcpy(cmd_ctx->llm->payload, outputs,
3621 nb_output * sizeof(struct lttng_snapshot_output));
3622 free(outputs);
3623 }
3624
3625 ret = LTTNG_OK;
3626 break;
3627 }
3628 case LTTNG_SNAPSHOT_RECORD:
3629 {
3630 ret = cmd_snapshot_record(cmd_ctx->session,
3631 &cmd_ctx->lsm->u.snapshot_record.output,
3632 cmd_ctx->lsm->u.snapshot_record.wait);
3633 break;
3634 }
3635 case LTTNG_CREATE_SESSION_SNAPSHOT:
3636 {
3637 size_t nb_uri, len;
3638 struct lttng_uri *uris = NULL;
3639
3640 nb_uri = cmd_ctx->lsm->u.uri.size;
3641 len = nb_uri * sizeof(struct lttng_uri);
3642
3643 if (nb_uri > 0) {
3644 uris = zmalloc(len);
3645 if (uris == NULL) {
3646 ret = LTTNG_ERR_FATAL;
3647 goto error;
3648 }
3649
3650 /* Receive variable len data */
3651 DBG("Waiting for %zu URIs from client ...", nb_uri);
3652 ret = lttcomm_recv_unix_sock(sock, uris, len);
3653 if (ret <= 0) {
3654 DBG("No URIs received from client... continuing");
3655 *sock_error = 1;
3656 ret = LTTNG_ERR_SESSION_FAIL;
3657 free(uris);
3658 goto error;
3659 }
3660
3661 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3662 DBG("Creating session with ONE network URI is a bad call");
3663 ret = LTTNG_ERR_SESSION_FAIL;
3664 free(uris);
3665 goto error;
3666 }
3667 }
3668
3669 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3670 nb_uri, &cmd_ctx->creds);
3671 free(uris);
3672 break;
3673 }
3674 case LTTNG_CREATE_SESSION_LIVE:
3675 {
3676 size_t nb_uri, len;
3677 struct lttng_uri *uris = NULL;
3678
3679 nb_uri = cmd_ctx->lsm->u.uri.size;
3680 len = nb_uri * sizeof(struct lttng_uri);
3681
3682 if (nb_uri > 0) {
3683 uris = zmalloc(len);
3684 if (uris == NULL) {
3685 ret = LTTNG_ERR_FATAL;
3686 goto error;
3687 }
3688
3689 /* Receive variable len data */
3690 DBG("Waiting for %zu URIs from client ...", nb_uri);
3691 ret = lttcomm_recv_unix_sock(sock, uris, len);
3692 if (ret <= 0) {
3693 DBG("No URIs received from client... continuing");
3694 *sock_error = 1;
3695 ret = LTTNG_ERR_SESSION_FAIL;
3696 free(uris);
3697 goto error;
3698 }
3699
3700 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3701 DBG("Creating session with ONE network URI is a bad call");
3702 ret = LTTNG_ERR_SESSION_FAIL;
3703 free(uris);
3704 goto error;
3705 }
3706 }
3707
3708 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3709 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3710 free(uris);
3711 break;
3712 }
3713 case LTTNG_SAVE_SESSION:
3714 {
3715 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3716 &cmd_ctx->creds);
3717 break;
3718 }
3719 default:
3720 ret = LTTNG_ERR_UND;
3721 break;
3722 }
3723
3724 error:
3725 if (cmd_ctx->llm == NULL) {
3726 DBG("Missing llm structure. Allocating one.");
3727 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3728 goto setup_error;
3729 }
3730 }
3731 /* Set return code */
3732 cmd_ctx->llm->ret_code = ret;
3733 setup_error:
3734 if (cmd_ctx->session) {
3735 session_unlock(cmd_ctx->session);
3736 }
3737 if (need_tracing_session) {
3738 session_unlock_list();
3739 }
3740 init_setup_error:
3741 return ret;
3742 }
3743
3744 /*
3745 * Thread managing health check socket.
3746 */
3747 static void *thread_manage_health(void *data)
3748 {
3749 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3750 uint32_t revents, nb_fd;
3751 struct lttng_poll_event events;
3752 struct health_comm_msg msg;
3753 struct health_comm_reply reply;
3754
3755 DBG("[thread] Manage health check started");
3756
3757 rcu_register_thread();
3758
3759 /* We might hit an error path before this is created. */
3760 lttng_poll_init(&events);
3761
3762 /* Create unix socket */
3763 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3764 if (sock < 0) {
3765 ERR("Unable to create health check Unix socket");
3766 ret = -1;
3767 goto error;
3768 }
3769
3770 if (is_root) {
3771 /* lttng health client socket path permissions */
3772 ret = chown(health_unix_sock_path, 0,
3773 utils_get_group_id(tracing_group_name));
3774 if (ret < 0) {
3775 ERR("Unable to set group on %s", health_unix_sock_path);
3776 PERROR("chown");
3777 ret = -1;
3778 goto error;
3779 }
3780
3781 ret = chmod(health_unix_sock_path,
3782 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3783 if (ret < 0) {
3784 ERR("Unable to set permissions on %s", health_unix_sock_path);
3785 PERROR("chmod");
3786 ret = -1;
3787 goto error;
3788 }
3789 }
3790
3791 /*
3792 * Set the CLOEXEC flag. Return code is useless because either way, the
3793 * show must go on.
3794 */
3795 (void) utils_set_fd_cloexec(sock);
3796
3797 ret = lttcomm_listen_unix_sock(sock);
3798 if (ret < 0) {
3799 goto error;
3800 }
3801
3802 /*
3803 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3804 * more will be added to this poll set.
3805 */
3806 ret = sessiond_set_thread_pollset(&events, 2);
3807 if (ret < 0) {
3808 goto error;
3809 }
3810
3811 /* Add the application registration socket */
3812 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3813 if (ret < 0) {
3814 goto error;
3815 }
3816
3817 sessiond_notify_ready();
3818
3819 while (1) {
3820 DBG("Health check ready");
3821
3822 /* Inifinite blocking call, waiting for transmission */
3823 restart:
3824 ret = lttng_poll_wait(&events, -1);
3825 if (ret < 0) {
3826 /*
3827 * Restart interrupted system call.
3828 */
3829 if (errno == EINTR) {
3830 goto restart;
3831 }
3832 goto error;
3833 }
3834
3835 nb_fd = ret;
3836
3837 for (i = 0; i < nb_fd; i++) {
3838 /* Fetch once the poll data */
3839 revents = LTTNG_POLL_GETEV(&events, i);
3840 pollfd = LTTNG_POLL_GETFD(&events, i);
3841
3842 /* Thread quit pipe has been closed. Killing thread. */
3843 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3844 if (ret) {
3845 err = 0;
3846 goto exit;
3847 }
3848
3849 /* Event on the registration socket */
3850 if (pollfd == sock) {
3851 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3852 ERR("Health socket poll error");
3853 goto error;
3854 }
3855 }
3856 }
3857
3858 new_sock = lttcomm_accept_unix_sock(sock);
3859 if (new_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(new_sock);
3868
3869 DBG("Receiving data from client for health...");
3870 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3871 if (ret <= 0) {
3872 DBG("Nothing recv() from client... continuing");
3873 ret = close(new_sock);
3874 if (ret) {
3875 PERROR("close");
3876 }
3877 new_sock = -1;
3878 continue;
3879 }
3880
3881 rcu_thread_online();
3882
3883 memset(&reply, 0, sizeof(reply));
3884 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3885 /*
3886 * health_check_state returns 0 if health is
3887 * bad.
3888 */
3889 if (!health_check_state(health_sessiond, i)) {
3890 reply.ret_code |= 1ULL << i;
3891 }
3892 }
3893
3894 DBG2("Health check return value %" PRIx64, reply.ret_code);
3895
3896 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3897 if (ret < 0) {
3898 ERR("Failed to send health data back to client");
3899 }
3900
3901 /* End of transmission */
3902 ret = close(new_sock);
3903 if (ret) {
3904 PERROR("close");
3905 }
3906 new_sock = -1;
3907 }
3908
3909 exit:
3910 error:
3911 if (err) {
3912 ERR("Health error occurred in %s", __func__);
3913 }
3914 DBG("Health check thread dying");
3915 unlink(health_unix_sock_path);
3916 if (sock >= 0) {
3917 ret = close(sock);
3918 if (ret) {
3919 PERROR("close");
3920 }
3921 }
3922
3923 lttng_poll_clean(&events);
3924
3925 rcu_unregister_thread();
3926 return NULL;
3927 }
3928
3929 /*
3930 * This thread manage all clients request using the unix client socket for
3931 * communication.
3932 */
3933 static void *thread_manage_clients(void *data)
3934 {
3935 int sock = -1, ret, i, pollfd, err = -1;
3936 int sock_error;
3937 uint32_t revents, nb_fd;
3938 struct command_ctx *cmd_ctx = NULL;
3939 struct lttng_poll_event events;
3940
3941 DBG("[thread] Manage client started");
3942
3943 rcu_register_thread();
3944
3945 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3946
3947 health_code_update();
3948
3949 ret = lttcomm_listen_unix_sock(client_sock);
3950 if (ret < 0) {
3951 goto error_listen;
3952 }
3953
3954 /*
3955 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3956 * more will be added to this poll set.
3957 */
3958 ret = sessiond_set_thread_pollset(&events, 2);
3959 if (ret < 0) {
3960 goto error_create_poll;
3961 }
3962
3963 /* Add the application registration socket */
3964 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3965 if (ret < 0) {
3966 goto error;
3967 }
3968
3969 sessiond_notify_ready();
3970 ret = sem_post(&load_info->message_thread_ready);
3971 if (ret) {
3972 PERROR("sem_post message_thread_ready");
3973 goto error;
3974 }
3975
3976 /* This testpoint is after we signal readiness to the parent. */
3977 if (testpoint(sessiond_thread_manage_clients)) {
3978 goto error;
3979 }
3980
3981 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
3982 goto error;
3983 }
3984
3985 health_code_update();
3986
3987 while (1) {
3988 DBG("Accepting client command ...");
3989
3990 /* Inifinite blocking call, waiting for transmission */
3991 restart:
3992 health_poll_entry();
3993 ret = lttng_poll_wait(&events, -1);
3994 health_poll_exit();
3995 if (ret < 0) {
3996 /*
3997 * Restart interrupted system call.
3998 */
3999 if (errno == EINTR) {
4000 goto restart;
4001 }
4002 goto error;
4003 }
4004
4005 nb_fd = ret;
4006
4007 for (i = 0; i < nb_fd; i++) {
4008 /* Fetch once the poll data */
4009 revents = LTTNG_POLL_GETEV(&events, i);
4010 pollfd = LTTNG_POLL_GETFD(&events, i);
4011
4012 health_code_update();
4013
4014 /* Thread quit pipe has been closed. Killing thread. */
4015 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4016 if (ret) {
4017 err = 0;
4018 goto exit;
4019 }
4020
4021 /* Event on the registration socket */
4022 if (pollfd == client_sock) {
4023 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4024 ERR("Client socket poll error");
4025 goto error;
4026 }
4027 }
4028 }
4029
4030 DBG("Wait for client response");
4031
4032 health_code_update();
4033
4034 sock = lttcomm_accept_unix_sock(client_sock);
4035 if (sock < 0) {
4036 goto error;
4037 }
4038
4039 /*
4040 * Set the CLOEXEC flag. Return code is useless because either way, the
4041 * show must go on.
4042 */
4043 (void) utils_set_fd_cloexec(sock);
4044
4045 /* Set socket option for credentials retrieval */
4046 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4047 if (ret < 0) {
4048 goto error;
4049 }
4050
4051 /* Allocate context command to process the client request */
4052 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4053 if (cmd_ctx == NULL) {
4054 PERROR("zmalloc cmd_ctx");
4055 goto error;
4056 }
4057
4058 /* Allocate data buffer for reception */
4059 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4060 if (cmd_ctx->lsm == NULL) {
4061 PERROR("zmalloc cmd_ctx->lsm");
4062 goto error;
4063 }
4064
4065 cmd_ctx->llm = NULL;
4066 cmd_ctx->session = NULL;
4067
4068 health_code_update();
4069
4070 /*
4071 * Data is received from the lttng client. The struct
4072 * lttcomm_session_msg (lsm) contains the command and data request of
4073 * the client.
4074 */
4075 DBG("Receiving data from client ...");
4076 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4077 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4078 if (ret <= 0) {
4079 DBG("Nothing recv() from client... continuing");
4080 ret = close(sock);
4081 if (ret) {
4082 PERROR("close");
4083 }
4084 sock = -1;
4085 clean_command_ctx(&cmd_ctx);
4086 continue;
4087 }
4088
4089 health_code_update();
4090
4091 // TODO: Validate cmd_ctx including sanity check for
4092 // security purpose.
4093
4094 rcu_thread_online();
4095 /*
4096 * This function dispatch the work to the kernel or userspace tracer
4097 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4098 * informations for the client. The command context struct contains
4099 * everything this function may needs.
4100 */
4101 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4102 rcu_thread_offline();
4103 if (ret < 0) {
4104 ret = close(sock);
4105 if (ret) {
4106 PERROR("close");
4107 }
4108 sock = -1;
4109 /*
4110 * TODO: Inform client somehow of the fatal error. At
4111 * this point, ret < 0 means that a zmalloc failed
4112 * (ENOMEM). Error detected but still accept
4113 * command, unless a socket error has been
4114 * detected.
4115 */
4116 clean_command_ctx(&cmd_ctx);
4117 continue;
4118 }
4119
4120 health_code_update();
4121
4122 DBG("Sending response (size: %d, retcode: %s)",
4123 cmd_ctx->lttng_msg_size,
4124 lttng_strerror(-cmd_ctx->llm->ret_code));
4125 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4126 if (ret < 0) {
4127 ERR("Failed to send data back to client");
4128 }
4129
4130 /* End of transmission */
4131 ret = close(sock);
4132 if (ret) {
4133 PERROR("close");
4134 }
4135 sock = -1;
4136
4137 clean_command_ctx(&cmd_ctx);
4138
4139 health_code_update();
4140 }
4141
4142 exit:
4143 error:
4144 if (sock >= 0) {
4145 ret = close(sock);
4146 if (ret) {
4147 PERROR("close");
4148 }
4149 }
4150
4151 lttng_poll_clean(&events);
4152 clean_command_ctx(&cmd_ctx);
4153
4154 error_listen:
4155 error_create_poll:
4156 unlink(client_unix_sock_path);
4157 if (client_sock >= 0) {
4158 ret = close(client_sock);
4159 if (ret) {
4160 PERROR("close");
4161 }
4162 }
4163
4164 if (err) {
4165 health_error();
4166 ERR("Health error occurred in %s", __func__);
4167 }
4168
4169 health_unregister(health_sessiond);
4170
4171 DBG("Client thread dying");
4172
4173 rcu_unregister_thread();
4174 return NULL;
4175 }
4176
4177
4178 /*
4179 * usage function on stderr
4180 */
4181 static void usage(void)
4182 {
4183 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4184 fprintf(stderr, " -h, --help Display this usage.\n");
4185 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4186 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4187 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4188 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4189 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4190 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4191 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4192 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4193 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4194 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4195 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4196 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4197 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4198 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4199 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4200 fprintf(stderr, " -V, --version Show version number.\n");
4201 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4202 fprintf(stderr, " -q, --quiet No output at all.\n");
4203 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4204 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4205 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4206 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4207 fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
4208 fprintf(stderr, " -f --config Load daemon configuration file\n");
4209 fprintf(stderr, " -l --load PATH Load session configuration\n");
4210 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4211 }
4212
4213 /*
4214 * Take an option from the getopt output and set it in the right variable to be
4215 * used later.
4216 *
4217 * Return 0 on success else a negative value.
4218 */
4219 static int set_option(int opt, const char *arg, const char *optname)
4220 {
4221 int ret = 0;
4222
4223 switch (opt) {
4224 case 0:
4225 fprintf(stderr, "option %s", optname);
4226 if (arg) {
4227 fprintf(stderr, " with arg %s\n", arg);
4228 }
4229 break;
4230 case 'c':
4231 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4232 break;
4233 case 'a':
4234 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4235 break;
4236 case 'd':
4237 opt_daemon = 1;
4238 break;
4239 case 'b':
4240 opt_background = 1;
4241 break;
4242 case 'g':
4243 /*
4244 * If the override option is set, the pointer points to a
4245 * *non* const thus freeing it even though the variable type is
4246 * set to const.
4247 */
4248 if (tracing_group_name_override) {
4249 free((void *) tracing_group_name);
4250 }
4251 tracing_group_name = strdup(arg);
4252 if (!tracing_group_name) {
4253 perror("strdup");
4254 ret = -ENOMEM;
4255 }
4256 tracing_group_name_override = 1;
4257 break;
4258 case 'h':
4259 usage();
4260 exit(EXIT_FAILURE);
4261 case 'V':
4262 fprintf(stdout, "%s\n", VERSION);
4263 exit(EXIT_SUCCESS);
4264 case 'S':
4265 opt_sig_parent = 1;
4266 break;
4267 case 'E':
4268 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4269 break;
4270 case 'C':
4271 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4272 break;
4273 case 'F':
4274 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4275 break;
4276 case 'D':
4277 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4278 break;
4279 case 'H':
4280 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4281 break;
4282 case 'G':
4283 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4284 break;
4285 case 'N':
4286 opt_no_kernel = 1;
4287 break;
4288 case 'q':
4289 lttng_opt_quiet = 1;
4290 break;
4291 case 'v':
4292 /* Verbose level can increase using multiple -v */
4293 if (arg) {
4294 /* Value obtained from config file */
4295 lttng_opt_verbose = config_parse_value(arg);
4296 } else {
4297 /* -v used on command line */
4298 lttng_opt_verbose++;
4299 }
4300 /* Clamp value to [0, 3] */
4301 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4302 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4303 break;
4304 case 'Z':
4305 if (arg) {
4306 opt_verbose_consumer = config_parse_value(arg);
4307 } else {
4308 opt_verbose_consumer += 1;
4309 }
4310 break;
4311 case 'u':
4312 if (consumerd32_bin_override) {
4313 free((void *) consumerd32_bin);
4314 }
4315 consumerd32_bin = strdup(arg);
4316 if (!consumerd32_bin) {
4317 perror("strdup");
4318 ret = -ENOMEM;
4319 }
4320 consumerd32_bin_override = 1;
4321 break;
4322 case 'U':
4323 if (consumerd32_libdir_override) {
4324 free((void *) consumerd32_libdir);
4325 }
4326 consumerd32_libdir = strdup(arg);
4327 if (!consumerd32_libdir) {
4328 perror("strdup");
4329 ret = -ENOMEM;
4330 }
4331 consumerd32_libdir_override = 1;
4332 break;
4333 case 't':
4334 if (consumerd64_bin_override) {
4335 free((void *) consumerd64_bin);
4336 }
4337 consumerd64_bin = strdup(arg);
4338 if (!consumerd64_bin) {
4339 perror("strdup");
4340 ret = -ENOMEM;
4341 }
4342 consumerd64_bin_override = 1;
4343 break;
4344 case 'T':
4345 if (consumerd64_libdir_override) {
4346 free((void *) consumerd64_libdir);
4347 }
4348 consumerd64_libdir = strdup(arg);
4349 if (!consumerd64_libdir) {
4350 perror("strdup");
4351 ret = -ENOMEM;
4352 }
4353 consumerd64_libdir_override = 1;
4354 break;
4355 case 'p':
4356 free(opt_pidfile);
4357 opt_pidfile = strdup(arg);
4358 if (!opt_pidfile) {
4359 perror("strdup");
4360 ret = -ENOMEM;
4361 }
4362 break;
4363 case 'J': /* JUL TCP port. */
4364 {
4365 unsigned long v;
4366
4367 errno = 0;
4368 v = strtoul(arg, NULL, 0);
4369 if (errno != 0 || !isdigit(arg[0])) {
4370 ERR("Wrong value in --jul-tcp-port parameter: %s", arg);
4371 return -1;
4372 }
4373 if (v == 0 || v >= 65535) {
4374 ERR("Port overflow in --jul-tcp-port parameter: %s", arg);
4375 return -1;
4376 }
4377 jul_tcp_port = (uint32_t) v;
4378 DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
4379 break;
4380 }
4381 case 'l':
4382 free(opt_load_session_path);
4383 opt_load_session_path = strdup(arg);
4384 if (!opt_load_session_path) {
4385 perror("strdup");
4386 ret = -ENOMEM;
4387 }
4388 break;
4389 case 'P': /* probe modules list */
4390 free(kmod_probes_list);
4391 kmod_probes_list = strdup(arg);
4392 if (!kmod_probes_list) {
4393 perror("strdup");
4394 ret = -ENOMEM;
4395 }
4396 break;
4397 case 'f':
4398 /* This is handled in set_options() thus silent break. */
4399 break;
4400 default:
4401 /* Unknown option or other error.
4402 * Error is printed by getopt, just return */
4403 ret = -1;
4404 }
4405
4406 return ret;
4407 }
4408
4409 /*
4410 * config_entry_handler_cb used to handle options read from a config file.
4411 * See config_entry_handler_cb comment in common/config/config.h for the
4412 * return value conventions.
4413 */
4414 static int config_entry_handler(const struct config_entry *entry, void *unused)
4415 {
4416 int ret = 0, i;
4417
4418 if (!entry || !entry->name || !entry->value) {
4419 ret = -EINVAL;
4420 goto end;
4421 }
4422
4423 /* Check if the option is to be ignored */
4424 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4425 if (!strcmp(entry->name, config_ignore_options[i])) {
4426 goto end;
4427 }
4428 }
4429
4430 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4431 i++) {
4432
4433 /* Ignore if not fully matched. */
4434 if (strcmp(entry->name, long_options[i].name)) {
4435 continue;
4436 }
4437
4438 /*
4439 * If the option takes no argument on the command line, we have to
4440 * check if the value is "true". We support non-zero numeric values,
4441 * true, on and yes.
4442 */
4443 if (!long_options[i].has_arg) {
4444 ret = config_parse_value(entry->value);
4445 if (ret <= 0) {
4446 if (ret) {
4447 WARN("Invalid configuration value \"%s\" for option %s",
4448 entry->value, entry->name);
4449 }
4450 /* False, skip boolean config option. */
4451 goto end;
4452 }
4453 }
4454
4455 ret = set_option(long_options[i].val, entry->value, entry->name);
4456 goto end;
4457 }
4458
4459 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4460
4461 end:
4462 return ret;
4463 }
4464
4465 /*
4466 * daemon configuration loading and argument parsing
4467 */
4468 static int set_options(int argc, char **argv)
4469 {
4470 int ret = 0, c = 0, option_index = 0;
4471 int orig_optopt = optopt, orig_optind = optind;
4472 char *optstring;
4473 const char *config_path = NULL;
4474
4475 optstring = utils_generate_optstring(long_options,
4476 sizeof(long_options) / sizeof(struct option));
4477 if (!optstring) {
4478 ret = -ENOMEM;
4479 goto end;
4480 }
4481
4482 /* Check for the --config option */
4483 while ((c = getopt_long(argc, argv, optstring, long_options,
4484 &option_index)) != -1) {
4485 if (c == '?') {
4486 ret = -EINVAL;
4487 goto end;
4488 } else if (c != 'f') {
4489 /* if not equal to --config option. */
4490 continue;
4491 }
4492
4493 config_path = utils_expand_path(optarg);
4494 if (!config_path) {
4495 ERR("Failed to resolve path: %s", optarg);
4496 }
4497 }
4498
4499 ret = config_get_section_entries(config_path, config_section_name,
4500 config_entry_handler, NULL);
4501 if (ret) {
4502 if (ret > 0) {
4503 ERR("Invalid configuration option at line %i", ret);
4504 ret = -1;
4505 }
4506 goto end;
4507 }
4508
4509 /* Reset getopt's global state */
4510 optopt = orig_optopt;
4511 optind = orig_optind;
4512 while (1) {
4513 c = getopt_long(argc, argv, optstring, long_options, &option_index);
4514 if (c == -1) {
4515 break;
4516 }
4517
4518 ret = set_option(c, optarg, long_options[option_index].name);
4519 if (ret < 0) {
4520 break;
4521 }
4522 }
4523
4524 end:
4525 free(optstring);
4526 return ret;
4527 }
4528
4529 /*
4530 * Creates the two needed socket by the daemon.
4531 * apps_sock - The communication socket for all UST apps.
4532 * client_sock - The communication of the cli tool (lttng).
4533 */
4534 static int init_daemon_socket(void)
4535 {
4536 int ret = 0;
4537 mode_t old_umask;
4538
4539 old_umask = umask(0);
4540
4541 /* Create client tool unix socket */
4542 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4543 if (client_sock < 0) {
4544 ERR("Create unix sock failed: %s", client_unix_sock_path);
4545 ret = -1;
4546 goto end;
4547 }
4548
4549 /* Set the cloexec flag */
4550 ret = utils_set_fd_cloexec(client_sock);
4551 if (ret < 0) {
4552 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4553 "Continuing but note that the consumer daemon will have a "
4554 "reference to this socket on exec()", client_sock);
4555 }
4556
4557 /* File permission MUST be 660 */
4558 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4559 if (ret < 0) {
4560 ERR("Set file permissions failed: %s", client_unix_sock_path);
4561 PERROR("chmod");
4562 goto end;
4563 }
4564
4565 /* Create the application unix socket */
4566 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4567 if (apps_sock < 0) {
4568 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4569 ret = -1;
4570 goto end;
4571 }
4572
4573 /* Set the cloexec flag */
4574 ret = utils_set_fd_cloexec(apps_sock);
4575 if (ret < 0) {
4576 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4577 "Continuing but note that the consumer daemon will have a "
4578 "reference to this socket on exec()", apps_sock);
4579 }
4580
4581 /* File permission MUST be 666 */
4582 ret = chmod(apps_unix_sock_path,
4583 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4584 if (ret < 0) {
4585 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4586 PERROR("chmod");
4587 goto end;
4588 }
4589
4590 DBG3("Session daemon client socket %d and application socket %d created",
4591 client_sock, apps_sock);
4592
4593 end:
4594 umask(old_umask);
4595 return ret;
4596 }
4597
4598 /*
4599 * Check if the global socket is available, and if a daemon is answering at the
4600 * other side. If yes, error is returned.
4601 */
4602 static int check_existing_daemon(void)
4603 {
4604 /* Is there anybody out there ? */
4605 if (lttng_session_daemon_alive()) {
4606 return -EEXIST;
4607 }
4608
4609 return 0;
4610 }
4611
4612 /*
4613 * Set the tracing group gid onto the client socket.
4614 *
4615 * Race window between mkdir and chown is OK because we are going from more
4616 * permissive (root.root) to less permissive (root.tracing).
4617 */
4618 static int set_permissions(char *rundir)
4619 {
4620 int ret;
4621 gid_t gid;
4622
4623 gid = utils_get_group_id(tracing_group_name);
4624
4625 /* Set lttng run dir */
4626 ret = chown(rundir, 0, gid);
4627 if (ret < 0) {
4628 ERR("Unable to set group on %s", rundir);
4629 PERROR("chown");
4630 }
4631
4632 /*
4633 * Ensure all applications and tracing group can search the run
4634 * dir. Allow everyone to read the directory, since it does not
4635 * buy us anything to hide its content.
4636 */
4637 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4638 if (ret < 0) {
4639 ERR("Unable to set permissions on %s", rundir);
4640 PERROR("chmod");
4641 }
4642
4643 /* lttng client socket path */
4644 ret = chown(client_unix_sock_path, 0, gid);
4645 if (ret < 0) {
4646 ERR("Unable to set group on %s", client_unix_sock_path);
4647 PERROR("chown");
4648 }
4649
4650 /* kconsumer error socket path */
4651 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4652 if (ret < 0) {
4653 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4654 PERROR("chown");
4655 }
4656
4657 /* 64-bit ustconsumer error socket path */
4658 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4659 if (ret < 0) {
4660 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4661 PERROR("chown");
4662 }
4663
4664 /* 32-bit ustconsumer compat32 error socket path */
4665 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4666 if (ret < 0) {
4667 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4668 PERROR("chown");
4669 }
4670
4671 DBG("All permissions are set");
4672
4673 return ret;
4674 }
4675
4676 /*
4677 * Create the lttng run directory needed for all global sockets and pipe.
4678 */
4679 static int create_lttng_rundir(const char *rundir)
4680 {
4681 int ret;
4682
4683 DBG3("Creating LTTng run directory: %s", rundir);
4684
4685 ret = mkdir(rundir, S_IRWXU);
4686 if (ret < 0) {
4687 if (errno != EEXIST) {
4688 ERR("Unable to create %s", rundir);
4689 goto error;
4690 } else {
4691 ret = 0;
4692 }
4693 }
4694
4695 error:
4696 return ret;
4697 }
4698
4699 /*
4700 * Setup sockets and directory needed by the kconsumerd communication with the
4701 * session daemon.
4702 */
4703 static int set_consumer_sockets(struct consumer_data *consumer_data,
4704 const char *rundir)
4705 {
4706 int ret;
4707 char path[PATH_MAX];
4708
4709 switch (consumer_data->type) {
4710 case LTTNG_CONSUMER_KERNEL:
4711 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4712 break;
4713 case LTTNG_CONSUMER64_UST:
4714 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4715 break;
4716 case LTTNG_CONSUMER32_UST:
4717 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4718 break;
4719 default:
4720 ERR("Consumer type unknown");
4721 ret = -EINVAL;
4722 goto error;
4723 }
4724
4725 DBG2("Creating consumer directory: %s", path);
4726
4727 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4728 if (ret < 0) {
4729 if (errno != EEXIST) {
4730 PERROR("mkdir");
4731 ERR("Failed to create %s", path);
4732 goto error;
4733 }
4734 ret = -1;
4735 }
4736 if (is_root) {
4737 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4738 if (ret < 0) {
4739 ERR("Unable to set group on %s", path);
4740 PERROR("chown");
4741 goto error;
4742 }
4743 }
4744
4745 /* Create the kconsumerd error unix socket */
4746 consumer_data->err_sock =
4747 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4748 if (consumer_data->err_sock < 0) {
4749 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4750 ret = -1;
4751 goto error;
4752 }
4753
4754 /*
4755 * Set the CLOEXEC flag. Return code is useless because either way, the
4756 * show must go on.
4757 */
4758 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4759 if (ret < 0) {
4760 PERROR("utils_set_fd_cloexec");
4761 /* continue anyway */
4762 }
4763
4764 /* File permission MUST be 660 */
4765 ret = chmod(consumer_data->err_unix_sock_path,
4766 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4767 if (ret < 0) {
4768 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4769 PERROR("chmod");
4770 goto error;
4771 }
4772
4773 error:
4774 return ret;
4775 }
4776
4777 /*
4778 * Signal handler for the daemon
4779 *
4780 * Simply stop all worker threads, leaving main() return gracefully after
4781 * joining all threads and calling cleanup().
4782 */
4783 static void sighandler(int sig)
4784 {
4785 switch (sig) {
4786 case SIGPIPE:
4787 DBG("SIGPIPE caught");
4788 return;
4789 case SIGINT:
4790 DBG("SIGINT caught");
4791 stop_threads();
4792 break;
4793 case SIGTERM:
4794 DBG("SIGTERM caught");
4795 stop_threads();
4796 break;
4797 case SIGUSR1:
4798 CMM_STORE_SHARED(recv_child_signal, 1);
4799 break;
4800 default:
4801 break;
4802 }
4803 }
4804
4805 /*
4806 * Setup signal handler for :
4807 * SIGINT, SIGTERM, SIGPIPE
4808 */
4809 static int set_signal_handler(void)
4810 {
4811 int ret = 0;
4812 struct sigaction sa;
4813 sigset_t sigset;
4814
4815 if ((ret = sigemptyset(&sigset)) < 0) {
4816 PERROR("sigemptyset");
4817 return ret;
4818 }
4819
4820 sa.sa_handler = sighandler;
4821 sa.sa_mask = sigset;
4822 sa.sa_flags = 0;
4823 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4824 PERROR("sigaction");
4825 return ret;
4826 }
4827
4828 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4829 PERROR("sigaction");
4830 return ret;
4831 }
4832
4833 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4834 PERROR("sigaction");
4835 return ret;
4836 }
4837
4838 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4839 PERROR("sigaction");
4840 return ret;
4841 }
4842
4843 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4844
4845 return ret;
4846 }
4847
4848 /*
4849 * Set open files limit to unlimited. This daemon can open a large number of
4850 * file descriptors in order to consumer multiple kernel traces.
4851 */
4852 static void set_ulimit(void)
4853 {
4854 int ret;
4855 struct rlimit lim;
4856
4857 /* The kernel does not allowed an infinite limit for open files */
4858 lim.rlim_cur = 65535;
4859 lim.rlim_max = 65535;
4860
4861 ret = setrlimit(RLIMIT_NOFILE, &lim);
4862 if (ret < 0) {
4863 PERROR("failed to set open files limit");
4864 }
4865 }
4866
4867 /*
4868 * Write pidfile using the rundir and opt_pidfile.
4869 */
4870 static void write_pidfile(void)
4871 {
4872 int ret;
4873 char pidfile_path[PATH_MAX];
4874
4875 assert(rundir);
4876
4877 if (opt_pidfile) {
4878 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4879 } else {
4880 /* Build pidfile path from rundir and opt_pidfile. */
4881 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4882 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4883 if (ret < 0) {
4884 PERROR("snprintf pidfile path");
4885 goto error;
4886 }
4887 }
4888
4889 /*
4890 * Create pid file in rundir. Return value is of no importance. The
4891 * execution will continue even though we are not able to write the file.
4892 */
4893 (void) utils_create_pid_file(getpid(), pidfile_path);
4894
4895 error:
4896 return;
4897 }
4898
4899 /*
4900 * Create lockfile using the rundir and return its fd.
4901 */
4902 static int create_lockfile(void)
4903 {
4904 int ret;
4905 char lockfile_path[PATH_MAX];
4906
4907 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
4908 if (ret < 0) {
4909 goto error;
4910 }
4911
4912 ret = utils_create_lock_file(lockfile_path);
4913 error:
4914 return ret;
4915 }
4916
4917 /*
4918 * Write JUL TCP port using the rundir.
4919 */
4920 static void write_julport(void)
4921 {
4922 int ret;
4923 char path[PATH_MAX];
4924
4925 assert(rundir);
4926
4927 ret = snprintf(path, sizeof(path), "%s/"
4928 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
4929 if (ret < 0) {
4930 PERROR("snprintf julport path");
4931 goto error;
4932 }
4933
4934 /*
4935 * Create TCP JUL port file in rundir. Return value is of no importance.
4936 * The execution will continue even though we are not able to write the
4937 * file.
4938 */
4939 (void) utils_create_pid_file(jul_tcp_port, path);
4940
4941 error:
4942 return;
4943 }
4944
4945 /*
4946 * Start the load session thread and dettach from it so the main thread can
4947 * continue. This does not return a value since whatever the outcome, the main
4948 * thread will continue.
4949 */
4950 static void start_load_session_thread(void)
4951 {
4952 int ret;
4953
4954 /* Create session loading thread. */
4955 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
4956 load_info);
4957 if (ret != 0) {
4958 PERROR("pthread_create load_session_thread");
4959 goto error_create;
4960 }
4961
4962 ret = pthread_detach(load_session_thread);
4963 if (ret != 0) {
4964 PERROR("pthread_detach load_session_thread");
4965 }
4966
4967 /* Everything went well so don't cleanup anything. */
4968
4969 error_create:
4970 /* The cleanup() function will destroy the load_info data. */
4971 return;
4972 }
4973
4974 /*
4975 * main
4976 */
4977 int main(int argc, char **argv)
4978 {
4979 int ret = 0;
4980 void *status;
4981 const char *home_path, *env_app_timeout;
4982
4983 init_kernel_workarounds();
4984
4985 rcu_register_thread();
4986
4987 if ((ret = set_signal_handler()) < 0) {
4988 goto error;
4989 }
4990
4991 setup_consumerd_path();
4992
4993 page_size = sysconf(_SC_PAGESIZE);
4994 if (page_size < 0) {
4995 PERROR("sysconf _SC_PAGESIZE");
4996 page_size = LONG_MAX;
4997 WARN("Fallback page size to %ld", page_size);
4998 }
4999
5000 /* Parse arguments and load the daemon configuration file */
5001 progname = argv[0];
5002 if ((ret = set_options(argc, argv)) < 0) {
5003 goto error;
5004 }
5005
5006 /* Daemonize */
5007 if (opt_daemon || opt_background) {
5008 int i;
5009
5010 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5011 !opt_background);
5012 if (ret < 0) {
5013 goto error;
5014 }
5015
5016 /*
5017 * We are in the child. Make sure all other file descriptors are
5018 * closed, in case we are called with more opened file descriptors than
5019 * the standard ones.
5020 */
5021 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5022 (void) close(i);
5023 }
5024 }
5025
5026 /* Create thread quit pipe */
5027 if ((ret = init_thread_quit_pipe()) < 0) {
5028 goto error;
5029 }
5030
5031 /* Check if daemon is UID = 0 */
5032 is_root = !getuid();
5033
5034 if (is_root) {
5035 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5036
5037 /* Create global run dir with root access */
5038 ret = create_lttng_rundir(rundir);
5039 if (ret < 0) {
5040 goto error;
5041 }
5042
5043 if (strlen(apps_unix_sock_path) == 0) {
5044 snprintf(apps_unix_sock_path, PATH_MAX,
5045 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5046 }
5047
5048 if (strlen(client_unix_sock_path) == 0) {
5049 snprintf(client_unix_sock_path, PATH_MAX,
5050 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5051 }
5052
5053 /* Set global SHM for ust */
5054 if (strlen(wait_shm_path) == 0) {
5055 snprintf(wait_shm_path, PATH_MAX,
5056 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5057 }
5058
5059 if (strlen(health_unix_sock_path) == 0) {
5060 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
5061 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5062 }
5063
5064 /* Setup kernel consumerd path */
5065 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5066 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5067 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5068 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5069
5070 DBG2("Kernel consumer err path: %s",
5071 kconsumer_data.err_unix_sock_path);
5072 DBG2("Kernel consumer cmd path: %s",
5073 kconsumer_data.cmd_unix_sock_path);
5074 } else {
5075 home_path = utils_get_home_dir();
5076 if (home_path == NULL) {
5077 /* TODO: Add --socket PATH option */
5078 ERR("Can't get HOME directory for sockets creation.");
5079 ret = -EPERM;
5080 goto error;
5081 }
5082
5083 /*
5084 * Create rundir from home path. This will create something like
5085 * $HOME/.lttng
5086 */
5087 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5088 if (ret < 0) {
5089 ret = -ENOMEM;
5090 goto error;
5091 }
5092
5093 ret = create_lttng_rundir(rundir);
5094 if (ret < 0) {
5095 goto error;
5096 }
5097
5098 if (strlen(apps_unix_sock_path) == 0) {
5099 snprintf(apps_unix_sock_path, PATH_MAX,
5100 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
5101 }
5102
5103 /* Set the cli tool unix socket path */
5104 if (strlen(client_unix_sock_path) == 0) {
5105 snprintf(client_unix_sock_path, PATH_MAX,
5106 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
5107 }
5108
5109 /* Set global SHM for ust */
5110 if (strlen(wait_shm_path) == 0) {
5111 snprintf(wait_shm_path, PATH_MAX,
5112 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
5113 }
5114
5115 /* Set health check Unix path */
5116 if (strlen(health_unix_sock_path) == 0) {
5117 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
5118 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
5119 }
5120 }
5121
5122 lockfile_fd = create_lockfile();
5123 if (lockfile_fd < 0) {
5124 goto error;
5125 }
5126
5127 /* Set consumer initial state */
5128 kernel_consumerd_state = CONSUMER_STOPPED;
5129 ust_consumerd_state = CONSUMER_STOPPED;
5130
5131 DBG("Client socket path %s", client_unix_sock_path);
5132 DBG("Application socket path %s", apps_unix_sock_path);
5133 DBG("Application wait path %s", wait_shm_path);
5134 DBG("LTTng run directory path: %s", rundir);
5135
5136 /* 32 bits consumerd path setup */
5137 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5138 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5139 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5140 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5141
5142 DBG2("UST consumer 32 bits err path: %s",
5143 ustconsumer32_data.err_unix_sock_path);
5144 DBG2("UST consumer 32 bits cmd path: %s",
5145 ustconsumer32_data.cmd_unix_sock_path);
5146
5147 /* 64 bits consumerd path setup */
5148 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5149 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5150 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5151 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5152
5153 DBG2("UST consumer 64 bits err path: %s",
5154 ustconsumer64_data.err_unix_sock_path);
5155 DBG2("UST consumer 64 bits cmd path: %s",
5156 ustconsumer64_data.cmd_unix_sock_path);
5157
5158 /*
5159 * See if daemon already exist.
5160 */
5161 if ((ret = check_existing_daemon()) < 0) {
5162 ERR("Already running daemon.\n");
5163 /*
5164 * We do not goto exit because we must not cleanup()
5165 * because a daemon is already running.
5166 */
5167 goto error;
5168 }
5169
5170 /*
5171 * Init UST app hash table. Alloc hash table before this point since
5172 * cleanup() can get called after that point.
5173 */
5174 ust_app_ht_alloc();
5175
5176 /* Initialize JUL domain subsystem. */
5177 if ((ret = jul_init()) < 0) {
5178 /* ENOMEM at this point. */
5179 goto error;
5180 }
5181
5182 /* After this point, we can safely call cleanup() with "goto exit" */
5183
5184 /*
5185 * These actions must be executed as root. We do that *after* setting up
5186 * the sockets path because we MUST make the check for another daemon using
5187 * those paths *before* trying to set the kernel consumer sockets and init
5188 * kernel tracer.
5189 */
5190 if (is_root) {
5191 ret = set_consumer_sockets(&kconsumer_data, rundir);
5192 if (ret < 0) {
5193 goto exit;
5194 }
5195
5196 /* Setup kernel tracer */
5197 if (!opt_no_kernel) {
5198 init_kernel_tracer();
5199 }
5200
5201 /* Set ulimit for open files */
5202 set_ulimit();
5203 }
5204 /* init lttng_fd tracking must be done after set_ulimit. */
5205 lttng_fd_init();
5206
5207 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
5208 if (ret < 0) {
5209 goto exit;
5210 }
5211
5212 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
5213 if (ret < 0) {
5214 goto exit;
5215 }
5216
5217 /* Setup the needed unix socket */
5218 if ((ret = init_daemon_socket()) < 0) {
5219 goto exit;
5220 }
5221
5222 /* Set credentials to socket */
5223 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
5224 goto exit;
5225 }
5226
5227 /* Get parent pid if -S, --sig-parent is specified. */
5228 if (opt_sig_parent) {
5229 ppid = getppid();
5230 }
5231
5232 /* Setup the kernel pipe for waking up the kernel thread */
5233 if (is_root && !opt_no_kernel) {
5234 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
5235 goto exit;
5236 }
5237 }
5238
5239 /* Setup the thread ht_cleanup communication pipe. */
5240 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
5241 goto exit;
5242 }
5243
5244 /* Setup the thread apps communication pipe. */
5245 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
5246 goto exit;
5247 }
5248
5249 /* Setup the thread apps notify communication pipe. */
5250 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
5251 goto exit;
5252 }
5253
5254 /* Initialize global buffer per UID and PID registry. */
5255 buffer_reg_init_uid_registry();
5256 buffer_reg_init_pid_registry();
5257
5258 /* Init UST command queue. */
5259 cds_wfq_init(&ust_cmd_queue.queue);
5260
5261 /*
5262 * Get session list pointer. This pointer MUST NOT be free(). This list is
5263 * statically declared in session.c
5264 */
5265 session_list_ptr = session_get_list();
5266
5267 /* Set up max poll set size */
5268 lttng_poll_set_max_size();
5269
5270 cmd_init();
5271
5272 /* Check for the application socket timeout env variable. */
5273 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5274 if (env_app_timeout) {
5275 app_socket_timeout = atoi(env_app_timeout);
5276 } else {
5277 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5278 }
5279
5280 write_pidfile();
5281 write_julport();
5282
5283 /* Initialize communication library */
5284 lttcomm_init();
5285 /* This is to get the TCP timeout value. */
5286 lttcomm_inet_init();
5287
5288 if (load_session_init_data(&load_info) < 0) {
5289 goto exit;
5290 }
5291 load_info->path = opt_load_session_path;
5292
5293 /*
5294 * Initialize the health check subsystem. This call should set the
5295 * appropriate time values.
5296 */
5297 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5298 if (!health_sessiond) {
5299 PERROR("health_app_create error");
5300 goto exit_health_sessiond_cleanup;
5301 }
5302
5303 /* Create thread to clean up RCU hash tables */
5304 ret = pthread_create(&ht_cleanup_thread, NULL,
5305 thread_ht_cleanup, (void *) NULL);
5306 if (ret != 0) {
5307 PERROR("pthread_create ht_cleanup");
5308 goto exit_ht_cleanup;
5309 }
5310
5311 /* Create health-check thread */
5312 ret = pthread_create(&health_thread, NULL,
5313 thread_manage_health, (void *) NULL);
5314 if (ret != 0) {
5315 PERROR("pthread_create health");
5316 goto exit_health;
5317 }
5318
5319 /* Create thread to manage the client socket */
5320 ret = pthread_create(&client_thread, NULL,
5321 thread_manage_clients, (void *) NULL);
5322 if (ret != 0) {
5323 PERROR("pthread_create clients");
5324 goto exit_client;
5325 }
5326
5327 /* Create thread to dispatch registration */
5328 ret = pthread_create(&dispatch_thread, NULL,
5329 thread_dispatch_ust_registration, (void *) NULL);
5330 if (ret != 0) {
5331 PERROR("pthread_create dispatch");
5332 goto exit_dispatch;
5333 }
5334
5335 /* Create thread to manage application registration. */
5336 ret = pthread_create(&reg_apps_thread, NULL,
5337 thread_registration_apps, (void *) NULL);
5338 if (ret != 0) {
5339 PERROR("pthread_create registration");
5340 goto exit_reg_apps;
5341 }
5342
5343 /* Create thread to manage application socket */
5344 ret = pthread_create(&apps_thread, NULL,
5345 thread_manage_apps, (void *) NULL);
5346 if (ret != 0) {
5347 PERROR("pthread_create apps");
5348 goto exit_apps;
5349 }
5350
5351 /* Create thread to manage application notify socket */
5352 ret = pthread_create(&apps_notify_thread, NULL,
5353 ust_thread_manage_notify, (void *) NULL);
5354 if (ret != 0) {
5355 PERROR("pthread_create notify");
5356 goto exit_apps_notify;
5357 }
5358
5359 /* Create JUL registration thread. */
5360 ret = pthread_create(&jul_reg_thread, NULL,
5361 jul_thread_manage_registration, (void *) NULL);
5362 if (ret != 0) {
5363 PERROR("pthread_create JUL");
5364 goto exit_jul_reg;
5365 }
5366
5367 /* Don't start this thread if kernel tracing is not requested nor root */
5368 if (is_root && !opt_no_kernel) {
5369 /* Create kernel thread to manage kernel event */
5370 ret = pthread_create(&kernel_thread, NULL,
5371 thread_manage_kernel, (void *) NULL);
5372 if (ret != 0) {
5373 PERROR("pthread_create kernel");
5374 goto exit_kernel;
5375 }
5376 }
5377
5378 /* Load possible session(s). */
5379 start_load_session_thread();
5380
5381 if (is_root && !opt_no_kernel) {
5382 ret = pthread_join(kernel_thread, &status);
5383 if (ret != 0) {
5384 PERROR("pthread_join");
5385 goto error; /* join error, exit without cleanup */
5386 }
5387 }
5388
5389 exit_kernel:
5390 ret = pthread_join(jul_reg_thread, &status);
5391 if (ret != 0) {
5392 PERROR("pthread_join JUL");
5393 goto error; /* join error, exit without cleanup */
5394 }
5395
5396 exit_jul_reg:
5397 ret = pthread_join(apps_notify_thread, &status);
5398 if (ret != 0) {
5399 PERROR("pthread_join apps notify");
5400 goto error; /* join error, exit without cleanup */
5401 }
5402
5403 exit_apps_notify:
5404 ret = pthread_join(apps_thread, &status);
5405 if (ret != 0) {
5406 PERROR("pthread_join apps");
5407 goto error; /* join error, exit without cleanup */
5408 }
5409
5410
5411 exit_apps:
5412 ret = pthread_join(reg_apps_thread, &status);
5413 if (ret != 0) {
5414 PERROR("pthread_join");
5415 goto error; /* join error, exit without cleanup */
5416 }
5417
5418 exit_reg_apps:
5419 ret = pthread_join(dispatch_thread, &status);
5420 if (ret != 0) {
5421 PERROR("pthread_join");
5422 goto error; /* join error, exit without cleanup */
5423 }
5424
5425 exit_dispatch:
5426 ret = pthread_join(client_thread, &status);
5427 if (ret != 0) {
5428 PERROR("pthread_join");
5429 goto error; /* join error, exit without cleanup */
5430 }
5431
5432 ret = join_consumer_thread(&kconsumer_data);
5433 if (ret != 0) {
5434 PERROR("join_consumer");
5435 goto error; /* join error, exit without cleanup */
5436 }
5437
5438 ret = join_consumer_thread(&ustconsumer32_data);
5439 if (ret != 0) {
5440 PERROR("join_consumer ust32");
5441 goto error; /* join error, exit without cleanup */
5442 }
5443
5444 ret = join_consumer_thread(&ustconsumer64_data);
5445 if (ret != 0) {
5446 PERROR("join_consumer ust64");
5447 goto error; /* join error, exit without cleanup */
5448 }
5449
5450 exit_client:
5451 ret = pthread_join(health_thread, &status);
5452 if (ret != 0) {
5453 PERROR("pthread_join health thread");
5454 goto error; /* join error, exit without cleanup */
5455 }
5456
5457 exit_health:
5458 ret = pthread_join(ht_cleanup_thread, &status);
5459 if (ret != 0) {
5460 PERROR("pthread_join ht cleanup thread");
5461 goto error; /* join error, exit without cleanup */
5462 }
5463 exit_ht_cleanup:
5464 health_app_destroy(health_sessiond);
5465 exit_health_sessiond_cleanup:
5466 exit:
5467 /*
5468 * cleanup() is called when no other thread is running.
5469 */
5470 rcu_thread_online();
5471 cleanup();
5472 rcu_thread_offline();
5473 rcu_unregister_thread();
5474 if (!ret) {
5475 exit(EXIT_SUCCESS);
5476 }
5477 error:
5478 exit(EXIT_FAILURE);
5479 }
This page took 0.136468 seconds and 5 git commands to generate.