2a97c37f6859988360dd283d514429cc8a7f59c8
[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 "agent-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 { "agent-tcp-port", 1, 0, 'J' },
157 { "config", 1, 0, 'f' },
158 { "load", 1, 0, 'l' },
159 { "kmod-probes", 1, 0, 'P' },
160 { "extra-kmod-probes", 1, 0, 'e' },
161 { NULL, 0, 0, 0 }
162 };
163
164 /* Command line options to ignore from configuration file */
165 static const char *config_ignore_options[] = { "help", "version", "config" };
166
167 /* Shared between threads */
168 static int dispatch_thread_exit;
169
170 /* Global application Unix socket path */
171 static char apps_unix_sock_path[PATH_MAX];
172 /* Global client Unix socket path */
173 static char client_unix_sock_path[PATH_MAX];
174 /* global wait shm path for UST */
175 static char wait_shm_path[PATH_MAX];
176 /* Global health check unix path */
177 static char health_unix_sock_path[PATH_MAX];
178
179 /* Sockets and FDs */
180 static int client_sock = -1;
181 static int apps_sock = -1;
182 int kernel_tracer_fd = -1;
183 static int kernel_poll_pipe[2] = { -1, -1 };
184
185 /*
186 * Quit pipe for all threads. This permits a single cancellation point
187 * for all threads when receiving an event on the pipe.
188 */
189 static int thread_quit_pipe[2] = { -1, -1 };
190
191 /*
192 * This pipe is used to inform the thread managing application communication
193 * that a command is queued and ready to be processed.
194 */
195 static int apps_cmd_pipe[2] = { -1, -1 };
196
197 int apps_cmd_notify_pipe[2] = { -1, -1 };
198
199 /* Pthread, Mutexes and Semaphores */
200 static pthread_t apps_thread;
201 static pthread_t apps_notify_thread;
202 static pthread_t reg_apps_thread;
203 static pthread_t client_thread;
204 static pthread_t kernel_thread;
205 static pthread_t dispatch_thread;
206 static pthread_t health_thread;
207 static pthread_t ht_cleanup_thread;
208 static pthread_t agent_reg_thread;
209 static pthread_t load_session_thread;
210
211 /*
212 * UST registration command queue. This queue is tied with a futex and uses a N
213 * wakers / 1 waiter implemented and detailed in futex.c/.h
214 *
215 * The thread_registration_apps and thread_dispatch_ust_registration uses this
216 * queue along with the wait/wake scheme. The thread_manage_apps receives down
217 * the line new application socket and monitors it for any I/O error or clean
218 * close that triggers an unregistration of the application.
219 */
220 static struct ust_cmd_queue ust_cmd_queue;
221
222 /*
223 * Pointer initialized before thread creation.
224 *
225 * This points to the tracing session list containing the session count and a
226 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
227 * MUST NOT be taken if you call a public function in session.c.
228 *
229 * The lock is nested inside the structure: session_list_ptr->lock. Please use
230 * session_lock_list and session_unlock_list for lock acquisition.
231 */
232 static struct ltt_session_list *session_list_ptr;
233
234 int ust_consumerd64_fd = -1;
235 int ust_consumerd32_fd = -1;
236
237 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
238 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
239 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
240 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
241 static int consumerd32_bin_override;
242 static int consumerd64_bin_override;
243 static int consumerd32_libdir_override;
244 static int consumerd64_libdir_override;
245
246 static const char *module_proc_lttng = "/proc/lttng";
247
248 /*
249 * Consumer daemon state which is changed when spawning it, killing it or in
250 * case of a fatal error.
251 */
252 enum consumerd_state {
253 CONSUMER_STARTED = 1,
254 CONSUMER_STOPPED = 2,
255 CONSUMER_ERROR = 3,
256 };
257
258 /*
259 * This consumer daemon state is used to validate if a client command will be
260 * able to reach the consumer. If not, the client is informed. For instance,
261 * doing a "lttng start" when the consumer state is set to ERROR will return an
262 * error to the client.
263 *
264 * The following example shows a possible race condition of this scheme:
265 *
266 * consumer thread error happens
267 * client cmd arrives
268 * client cmd checks state -> still OK
269 * consumer thread exit, sets error
270 * client cmd try to talk to consumer
271 * ...
272 *
273 * However, since the consumer is a different daemon, we have no way of making
274 * sure the command will reach it safely even with this state flag. This is why
275 * we consider that up to the state validation during command processing, the
276 * command is safe. After that, we can not guarantee the correctness of the
277 * client request vis-a-vis the consumer.
278 */
279 static enum consumerd_state ust_consumerd_state;
280 static enum consumerd_state kernel_consumerd_state;
281
282 /*
283 * Socket timeout for receiving and sending in seconds.
284 */
285 static int app_socket_timeout;
286
287 /* Set in main() with the current page size. */
288 long page_size;
289
290 /* Application health monitoring */
291 struct health_app *health_sessiond;
292
293 /* Agent TCP port for registration. Used by the agent thread. */
294 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
295
296 /* Am I root or not. */
297 int is_root; /* Set to 1 if the daemon is running as root */
298
299 const char * const config_section_name = "sessiond";
300
301 /* Load session thread information to operate. */
302 struct load_session_thread_data *load_info;
303
304 /*
305 * Whether sessiond is ready for commands/health check requests.
306 * NR_LTTNG_SESSIOND_READY must match the number of calls to
307 * sessiond_notify_ready().
308 */
309 #define NR_LTTNG_SESSIOND_READY 3
310 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
311
312 /* Notify parents that we are ready for cmd and health check */
313 LTTNG_HIDDEN
314 void sessiond_notify_ready(void)
315 {
316 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
317 /*
318 * Notify parent pid that we are ready to accept command
319 * for client side. This ppid is the one from the
320 * external process that spawned us.
321 */
322 if (opt_sig_parent) {
323 kill(ppid, SIGUSR1);
324 }
325
326 /*
327 * Notify the parent of the fork() process that we are
328 * ready.
329 */
330 if (opt_daemon || opt_background) {
331 kill(child_ppid, SIGUSR1);
332 }
333 }
334 }
335
336 static
337 void setup_consumerd_path(void)
338 {
339 const char *bin, *libdir;
340
341 /*
342 * Allow INSTALL_BIN_PATH to be used as a target path for the
343 * native architecture size consumer if CONFIG_CONSUMER*_PATH
344 * has not been defined.
345 */
346 #if (CAA_BITS_PER_LONG == 32)
347 if (!consumerd32_bin[0]) {
348 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
349 }
350 if (!consumerd32_libdir[0]) {
351 consumerd32_libdir = INSTALL_LIB_PATH;
352 }
353 #elif (CAA_BITS_PER_LONG == 64)
354 if (!consumerd64_bin[0]) {
355 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
356 }
357 if (!consumerd64_libdir[0]) {
358 consumerd64_libdir = INSTALL_LIB_PATH;
359 }
360 #else
361 #error "Unknown bitness"
362 #endif
363
364 /*
365 * runtime env. var. overrides the build default.
366 */
367 bin = getenv("LTTNG_CONSUMERD32_BIN");
368 if (bin) {
369 consumerd32_bin = bin;
370 }
371 bin = getenv("LTTNG_CONSUMERD64_BIN");
372 if (bin) {
373 consumerd64_bin = bin;
374 }
375 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
376 if (libdir) {
377 consumerd32_libdir = libdir;
378 }
379 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
380 if (libdir) {
381 consumerd64_libdir = libdir;
382 }
383 }
384
385 /*
386 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
387 */
388 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
389 {
390 int ret;
391
392 assert(events);
393
394 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
395 if (ret < 0) {
396 goto error;
397 }
398
399 /* Add quit pipe */
400 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
401 if (ret < 0) {
402 goto error;
403 }
404
405 return 0;
406
407 error:
408 return ret;
409 }
410
411 /*
412 * Check if the thread quit pipe was triggered.
413 *
414 * Return 1 if it was triggered else 0;
415 */
416 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
417 {
418 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
419 return 1;
420 }
421
422 return 0;
423 }
424
425 /*
426 * Init thread quit pipe.
427 *
428 * Return -1 on error or 0 if all pipes are created.
429 */
430 static int init_thread_quit_pipe(void)
431 {
432 int ret, i;
433
434 ret = pipe(thread_quit_pipe);
435 if (ret < 0) {
436 PERROR("thread quit pipe");
437 goto error;
438 }
439
440 for (i = 0; i < 2; i++) {
441 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
442 if (ret < 0) {
443 PERROR("fcntl");
444 goto error;
445 }
446 }
447
448 error:
449 return ret;
450 }
451
452 /*
453 * Stop all threads by closing the thread quit pipe.
454 */
455 static void stop_threads(void)
456 {
457 int ret;
458
459 /* Stopping all threads */
460 DBG("Terminating all threads");
461 ret = notify_thread_pipe(thread_quit_pipe[1]);
462 if (ret < 0) {
463 ERR("write error on thread quit pipe");
464 }
465
466 /* Dispatch thread */
467 CMM_STORE_SHARED(dispatch_thread_exit, 1);
468 futex_nto1_wake(&ust_cmd_queue.futex);
469 }
470
471 /*
472 * Close every consumer sockets.
473 */
474 static void close_consumer_sockets(void)
475 {
476 int ret;
477
478 if (kconsumer_data.err_sock >= 0) {
479 ret = close(kconsumer_data.err_sock);
480 if (ret < 0) {
481 PERROR("kernel consumer err_sock close");
482 }
483 }
484 if (ustconsumer32_data.err_sock >= 0) {
485 ret = close(ustconsumer32_data.err_sock);
486 if (ret < 0) {
487 PERROR("UST consumerd32 err_sock close");
488 }
489 }
490 if (ustconsumer64_data.err_sock >= 0) {
491 ret = close(ustconsumer64_data.err_sock);
492 if (ret < 0) {
493 PERROR("UST consumerd64 err_sock close");
494 }
495 }
496 if (kconsumer_data.cmd_sock >= 0) {
497 ret = close(kconsumer_data.cmd_sock);
498 if (ret < 0) {
499 PERROR("kernel consumer cmd_sock close");
500 }
501 }
502 if (ustconsumer32_data.cmd_sock >= 0) {
503 ret = close(ustconsumer32_data.cmd_sock);
504 if (ret < 0) {
505 PERROR("UST consumerd32 cmd_sock close");
506 }
507 }
508 if (ustconsumer64_data.cmd_sock >= 0) {
509 ret = close(ustconsumer64_data.cmd_sock);
510 if (ret < 0) {
511 PERROR("UST consumerd64 cmd_sock close");
512 }
513 }
514 }
515
516 /*
517 * Generate the full lock file path using the rundir.
518 *
519 * Return the snprintf() return value thus a negative value is an error.
520 */
521 static int generate_lock_file_path(char *path, size_t len)
522 {
523 int ret;
524
525 assert(path);
526 assert(rundir);
527
528 /* Build lockfile path from rundir. */
529 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
530 if (ret < 0) {
531 PERROR("snprintf lockfile path");
532 }
533
534 return ret;
535 }
536
537 /*
538 * Cleanup the daemon
539 */
540 static void cleanup(void)
541 {
542 int ret;
543 struct ltt_session *sess, *stmp;
544 char path[PATH_MAX];
545
546 DBG("Cleaning up");
547
548 /*
549 * Close the thread quit pipe. It has already done its job,
550 * since we are now called.
551 */
552 utils_close_pipe(thread_quit_pipe);
553
554 /*
555 * If opt_pidfile is undefined, the default file will be wiped when
556 * removing the rundir.
557 */
558 if (opt_pidfile) {
559 ret = remove(opt_pidfile);
560 if (ret < 0) {
561 PERROR("remove pidfile %s", opt_pidfile);
562 }
563 }
564
565 DBG("Removing sessiond and consumerd content of directory %s", rundir);
566
567 /* sessiond */
568 snprintf(path, PATH_MAX,
569 "%s/%s",
570 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
571 DBG("Removing %s", path);
572 (void) unlink(path);
573
574 snprintf(path, PATH_MAX, "%s/%s", rundir,
575 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
576 DBG("Removing %s", path);
577 (void) unlink(path);
578
579 /* kconsumerd */
580 snprintf(path, PATH_MAX,
581 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
582 rundir);
583 DBG("Removing %s", path);
584 (void) unlink(path);
585
586 snprintf(path, PATH_MAX,
587 DEFAULT_KCONSUMERD_PATH,
588 rundir);
589 DBG("Removing directory %s", path);
590 (void) rmdir(path);
591
592 /* ust consumerd 32 */
593 snprintf(path, PATH_MAX,
594 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
595 rundir);
596 DBG("Removing %s", path);
597 (void) unlink(path);
598
599 snprintf(path, PATH_MAX,
600 DEFAULT_USTCONSUMERD32_PATH,
601 rundir);
602 DBG("Removing directory %s", path);
603 (void) rmdir(path);
604
605 /* ust consumerd 64 */
606 snprintf(path, PATH_MAX,
607 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
608 rundir);
609 DBG("Removing %s", path);
610 (void) unlink(path);
611
612 snprintf(path, PATH_MAX,
613 DEFAULT_USTCONSUMERD64_PATH,
614 rundir);
615 DBG("Removing directory %s", path);
616 (void) rmdir(path);
617
618 DBG("Cleaning up all sessions");
619
620 /* Destroy session list mutex */
621 if (session_list_ptr != NULL) {
622 pthread_mutex_destroy(&session_list_ptr->lock);
623
624 /* Cleanup ALL session */
625 cds_list_for_each_entry_safe(sess, stmp,
626 &session_list_ptr->head, list) {
627 cmd_destroy_session(sess, kernel_poll_pipe[1]);
628 }
629 }
630
631 DBG("Closing all UST sockets");
632 ust_app_clean_list();
633 buffer_reg_destroy_registries();
634
635 if (is_root && !opt_no_kernel) {
636 DBG2("Closing kernel fd");
637 if (kernel_tracer_fd >= 0) {
638 ret = close(kernel_tracer_fd);
639 if (ret) {
640 PERROR("close");
641 }
642 }
643 DBG("Unloading kernel modules");
644 modprobe_remove_lttng_all();
645 }
646
647 close_consumer_sockets();
648
649 /*
650 * If the override option is set, the pointer points to a *non* const thus
651 * freeing it even though the variable type is set to const.
652 */
653 if (tracing_group_name_override) {
654 free((void *) tracing_group_name);
655 }
656 if (consumerd32_bin_override) {
657 free((void *) consumerd32_bin);
658 }
659 if (consumerd64_bin_override) {
660 free((void *) consumerd64_bin);
661 }
662 if (consumerd32_libdir_override) {
663 free((void *) consumerd32_libdir);
664 }
665 if (consumerd64_libdir_override) {
666 free((void *) consumerd64_libdir);
667 }
668
669 if (opt_pidfile) {
670 free(opt_pidfile);
671 }
672
673 if (opt_load_session_path) {
674 free(opt_load_session_path);
675 }
676
677 if (load_info) {
678 load_session_destroy_data(load_info);
679 free(load_info);
680 }
681
682 /*
683 * Cleanup lock file by deleting it and finaly closing it which will
684 * release the file system lock.
685 */
686 if (lockfile_fd >= 0) {
687 char lockfile_path[PATH_MAX];
688
689 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
690 if (ret > 0) {
691 ret = remove(lockfile_path);
692 if (ret < 0) {
693 PERROR("remove lock file");
694 }
695 ret = close(lockfile_fd);
696 if (ret < 0) {
697 PERROR("close lock file");
698 }
699 }
700 }
701
702 /*
703 * We do NOT rmdir rundir because there are other processes
704 * using it, for instance lttng-relayd, which can start in
705 * parallel with this teardown.
706 */
707
708 free(rundir);
709
710 /* <fun> */
711 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
712 "Matthew, BEET driven development works!%c[%dm",
713 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
714 /* </fun> */
715 }
716
717 /*
718 * Send data on a unix socket using the liblttsessiondcomm API.
719 *
720 * Return lttcomm error code.
721 */
722 static int send_unix_sock(int sock, void *buf, size_t len)
723 {
724 /* Check valid length */
725 if (len == 0) {
726 return -1;
727 }
728
729 return lttcomm_send_unix_sock(sock, buf, len);
730 }
731
732 /*
733 * Free memory of a command context structure.
734 */
735 static void clean_command_ctx(struct command_ctx **cmd_ctx)
736 {
737 DBG("Clean command context structure");
738 if (*cmd_ctx) {
739 if ((*cmd_ctx)->llm) {
740 free((*cmd_ctx)->llm);
741 }
742 if ((*cmd_ctx)->lsm) {
743 free((*cmd_ctx)->lsm);
744 }
745 free(*cmd_ctx);
746 *cmd_ctx = NULL;
747 }
748 }
749
750 /*
751 * Notify UST applications using the shm mmap futex.
752 */
753 static int notify_ust_apps(int active)
754 {
755 char *wait_shm_mmap;
756
757 DBG("Notifying applications of session daemon state: %d", active);
758
759 /* See shm.c for this call implying mmap, shm and futex calls */
760 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
761 if (wait_shm_mmap == NULL) {
762 goto error;
763 }
764
765 /* Wake waiting process */
766 futex_wait_update((int32_t *) wait_shm_mmap, active);
767
768 /* Apps notified successfully */
769 return 0;
770
771 error:
772 return -1;
773 }
774
775 /*
776 * Setup the outgoing data buffer for the response (llm) by allocating the
777 * right amount of memory and copying the original information from the lsm
778 * structure.
779 *
780 * Return total size of the buffer pointed by buf.
781 */
782 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
783 {
784 int ret, buf_size;
785
786 buf_size = size;
787
788 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
789 if (cmd_ctx->llm == NULL) {
790 PERROR("zmalloc");
791 ret = -ENOMEM;
792 goto error;
793 }
794
795 /* Copy common data */
796 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
797 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
798
799 cmd_ctx->llm->data_size = size;
800 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
801
802 return buf_size;
803
804 error:
805 return ret;
806 }
807
808 /*
809 * Update the kernel poll set of all channel fd available over all tracing
810 * session. Add the wakeup pipe at the end of the set.
811 */
812 static int update_kernel_poll(struct lttng_poll_event *events)
813 {
814 int ret;
815 struct ltt_session *session;
816 struct ltt_kernel_channel *channel;
817
818 DBG("Updating kernel poll set");
819
820 session_lock_list();
821 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
822 session_lock(session);
823 if (session->kernel_session == NULL) {
824 session_unlock(session);
825 continue;
826 }
827
828 cds_list_for_each_entry(channel,
829 &session->kernel_session->channel_list.head, list) {
830 /* Add channel fd to the kernel poll set */
831 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
832 if (ret < 0) {
833 session_unlock(session);
834 goto error;
835 }
836 DBG("Channel fd %d added to kernel set", channel->fd);
837 }
838 session_unlock(session);
839 }
840 session_unlock_list();
841
842 return 0;
843
844 error:
845 session_unlock_list();
846 return -1;
847 }
848
849 /*
850 * Find the channel fd from 'fd' over all tracing session. When found, check
851 * for new channel stream and send those stream fds to the kernel consumer.
852 *
853 * Useful for CPU hotplug feature.
854 */
855 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
856 {
857 int ret = 0;
858 struct ltt_session *session;
859 struct ltt_kernel_session *ksess;
860 struct ltt_kernel_channel *channel;
861
862 DBG("Updating kernel streams for channel fd %d", fd);
863
864 session_lock_list();
865 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
866 session_lock(session);
867 if (session->kernel_session == NULL) {
868 session_unlock(session);
869 continue;
870 }
871 ksess = session->kernel_session;
872
873 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
874 if (channel->fd == fd) {
875 DBG("Channel found, updating kernel streams");
876 ret = kernel_open_channel_stream(channel);
877 if (ret < 0) {
878 goto error;
879 }
880 /* Update the stream global counter */
881 ksess->stream_count_global += ret;
882
883 /*
884 * Have we already sent fds to the consumer? If yes, it means
885 * that tracing is started so it is safe to send our updated
886 * stream fds.
887 */
888 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
889 struct lttng_ht_iter iter;
890 struct consumer_socket *socket;
891
892 rcu_read_lock();
893 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
894 &iter.iter, socket, node.node) {
895 pthread_mutex_lock(socket->lock);
896 ret = kernel_consumer_send_channel_stream(socket,
897 channel, ksess,
898 session->output_traces ? 1 : 0);
899 pthread_mutex_unlock(socket->lock);
900 if (ret < 0) {
901 rcu_read_unlock();
902 goto error;
903 }
904 }
905 rcu_read_unlock();
906 }
907 goto error;
908 }
909 }
910 session_unlock(session);
911 }
912 session_unlock_list();
913 return ret;
914
915 error:
916 session_unlock(session);
917 session_unlock_list();
918 return ret;
919 }
920
921 /*
922 * For each tracing session, update newly registered apps. The session list
923 * lock MUST be acquired before calling this.
924 */
925 static void update_ust_app(int app_sock)
926 {
927 struct ltt_session *sess, *stmp;
928
929 /* Consumer is in an ERROR state. Stop any application update. */
930 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
931 /* Stop the update process since the consumer is dead. */
932 return;
933 }
934
935 /* For all tracing session(s) */
936 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
937 session_lock(sess);
938 if (sess->ust_session) {
939 ust_app_global_update(sess->ust_session, app_sock);
940 }
941 session_unlock(sess);
942 }
943 }
944
945 /*
946 * This thread manage event coming from the kernel.
947 *
948 * Features supported in this thread:
949 * -) CPU Hotplug
950 */
951 static void *thread_manage_kernel(void *data)
952 {
953 int ret, i, pollfd, update_poll_flag = 1, err = -1;
954 uint32_t revents, nb_fd;
955 char tmp;
956 struct lttng_poll_event events;
957
958 DBG("[thread] Thread manage kernel started");
959
960 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
961
962 /*
963 * This first step of the while is to clean this structure which could free
964 * non NULL pointers so initialize it before the loop.
965 */
966 lttng_poll_init(&events);
967
968 if (testpoint(sessiond_thread_manage_kernel)) {
969 goto error_testpoint;
970 }
971
972 health_code_update();
973
974 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
975 goto error_testpoint;
976 }
977
978 while (1) {
979 health_code_update();
980
981 if (update_poll_flag == 1) {
982 /* Clean events object. We are about to populate it again. */
983 lttng_poll_clean(&events);
984
985 ret = sessiond_set_thread_pollset(&events, 2);
986 if (ret < 0) {
987 goto error_poll_create;
988 }
989
990 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
991 if (ret < 0) {
992 goto error;
993 }
994
995 /* This will add the available kernel channel if any. */
996 ret = update_kernel_poll(&events);
997 if (ret < 0) {
998 goto error;
999 }
1000 update_poll_flag = 0;
1001 }
1002
1003 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
1004
1005 /* Poll infinite value of time */
1006 restart:
1007 health_poll_entry();
1008 ret = lttng_poll_wait(&events, -1);
1009 health_poll_exit();
1010 if (ret < 0) {
1011 /*
1012 * Restart interrupted system call.
1013 */
1014 if (errno == EINTR) {
1015 goto restart;
1016 }
1017 goto error;
1018 } else if (ret == 0) {
1019 /* Should not happen since timeout is infinite */
1020 ERR("Return value of poll is 0 with an infinite timeout.\n"
1021 "This should not have happened! Continuing...");
1022 continue;
1023 }
1024
1025 nb_fd = ret;
1026
1027 for (i = 0; i < nb_fd; i++) {
1028 /* Fetch once the poll data */
1029 revents = LTTNG_POLL_GETEV(&events, i);
1030 pollfd = LTTNG_POLL_GETFD(&events, i);
1031
1032 health_code_update();
1033
1034 /* Thread quit pipe has been closed. Killing thread. */
1035 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1036 if (ret) {
1037 err = 0;
1038 goto exit;
1039 }
1040
1041 /* Check for data on kernel pipe */
1042 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1043 (void) lttng_read(kernel_poll_pipe[0],
1044 &tmp, 1);
1045 /*
1046 * Ret value is useless here, if this pipe gets any actions an
1047 * update is required anyway.
1048 */
1049 update_poll_flag = 1;
1050 continue;
1051 } else {
1052 /*
1053 * New CPU detected by the kernel. Adding kernel stream to
1054 * kernel session and updating the kernel consumer
1055 */
1056 if (revents & LPOLLIN) {
1057 ret = update_kernel_stream(&kconsumer_data, pollfd);
1058 if (ret < 0) {
1059 continue;
1060 }
1061 break;
1062 /*
1063 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1064 * and unregister kernel stream at this point.
1065 */
1066 }
1067 }
1068 }
1069 }
1070
1071 exit:
1072 error:
1073 lttng_poll_clean(&events);
1074 error_poll_create:
1075 error_testpoint:
1076 utils_close_pipe(kernel_poll_pipe);
1077 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1078 if (err) {
1079 health_error();
1080 ERR("Health error occurred in %s", __func__);
1081 WARN("Kernel thread died unexpectedly. "
1082 "Kernel tracing can continue but CPU hotplug is disabled.");
1083 }
1084 health_unregister(health_sessiond);
1085 DBG("Kernel thread dying");
1086 return NULL;
1087 }
1088
1089 /*
1090 * Signal pthread condition of the consumer data that the thread.
1091 */
1092 static void signal_consumer_condition(struct consumer_data *data, int state)
1093 {
1094 pthread_mutex_lock(&data->cond_mutex);
1095
1096 /*
1097 * The state is set before signaling. It can be any value, it's the waiter
1098 * job to correctly interpret this condition variable associated to the
1099 * consumer pthread_cond.
1100 *
1101 * A value of 0 means that the corresponding thread of the consumer data
1102 * was not started. 1 indicates that the thread has started and is ready
1103 * for action. A negative value means that there was an error during the
1104 * thread bootstrap.
1105 */
1106 data->consumer_thread_is_ready = state;
1107 (void) pthread_cond_signal(&data->cond);
1108
1109 pthread_mutex_unlock(&data->cond_mutex);
1110 }
1111
1112 /*
1113 * This thread manage the consumer error sent back to the session daemon.
1114 */
1115 static void *thread_manage_consumer(void *data)
1116 {
1117 int sock = -1, i, ret, pollfd, err = -1;
1118 uint32_t revents, nb_fd;
1119 enum lttcomm_return_code code;
1120 struct lttng_poll_event events;
1121 struct consumer_data *consumer_data = data;
1122
1123 DBG("[thread] Manage consumer started");
1124
1125 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1126
1127 health_code_update();
1128
1129 /*
1130 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1131 * metadata_sock. Nothing more will be added to this poll set.
1132 */
1133 ret = sessiond_set_thread_pollset(&events, 3);
1134 if (ret < 0) {
1135 goto error_poll;
1136 }
1137
1138 /*
1139 * The error socket here is already in a listening state which was done
1140 * just before spawning this thread to avoid a race between the consumer
1141 * daemon exec trying to connect and the listen() call.
1142 */
1143 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1144 if (ret < 0) {
1145 goto error;
1146 }
1147
1148 health_code_update();
1149
1150 /* Infinite blocking call, waiting for transmission */
1151 restart:
1152 health_poll_entry();
1153
1154 if (testpoint(sessiond_thread_manage_consumer)) {
1155 goto error;
1156 }
1157
1158 ret = lttng_poll_wait(&events, -1);
1159 health_poll_exit();
1160 if (ret < 0) {
1161 /*
1162 * Restart interrupted system call.
1163 */
1164 if (errno == EINTR) {
1165 goto restart;
1166 }
1167 goto error;
1168 }
1169
1170 nb_fd = ret;
1171
1172 for (i = 0; i < nb_fd; i++) {
1173 /* Fetch once the poll data */
1174 revents = LTTNG_POLL_GETEV(&events, i);
1175 pollfd = LTTNG_POLL_GETFD(&events, i);
1176
1177 health_code_update();
1178
1179 /* Thread quit pipe has been closed. Killing thread. */
1180 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1181 if (ret) {
1182 err = 0;
1183 goto exit;
1184 }
1185
1186 /* Event on the registration socket */
1187 if (pollfd == consumer_data->err_sock) {
1188 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1189 ERR("consumer err socket poll error");
1190 goto error;
1191 }
1192 }
1193 }
1194
1195 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1196 if (sock < 0) {
1197 goto error;
1198 }
1199
1200 /*
1201 * Set the CLOEXEC flag. Return code is useless because either way, the
1202 * show must go on.
1203 */
1204 (void) utils_set_fd_cloexec(sock);
1205
1206 health_code_update();
1207
1208 DBG2("Receiving code from consumer err_sock");
1209
1210 /* Getting status code from kconsumerd */
1211 ret = lttcomm_recv_unix_sock(sock, &code,
1212 sizeof(enum lttcomm_return_code));
1213 if (ret <= 0) {
1214 goto error;
1215 }
1216
1217 health_code_update();
1218 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1219 /* Connect both socket, command and metadata. */
1220 consumer_data->cmd_sock =
1221 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1222 consumer_data->metadata_fd =
1223 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1224 if (consumer_data->cmd_sock < 0
1225 || consumer_data->metadata_fd < 0) {
1226 PERROR("consumer connect cmd socket");
1227 /* On error, signal condition and quit. */
1228 signal_consumer_condition(consumer_data, -1);
1229 goto error;
1230 }
1231 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1232 /* Create metadata socket lock. */
1233 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1234 if (consumer_data->metadata_sock.lock == NULL) {
1235 PERROR("zmalloc pthread mutex");
1236 ret = -1;
1237 goto error;
1238 }
1239 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1240
1241 signal_consumer_condition(consumer_data, 1);
1242 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1243 DBG("Consumer metadata socket ready (fd: %d)",
1244 consumer_data->metadata_fd);
1245 } else {
1246 ERR("consumer error when waiting for SOCK_READY : %s",
1247 lttcomm_get_readable_code(-code));
1248 goto error;
1249 }
1250
1251 /* Remove the consumerd error sock since we've established a connexion */
1252 ret = lttng_poll_del(&events, consumer_data->err_sock);
1253 if (ret < 0) {
1254 goto error;
1255 }
1256
1257 /* Add new accepted error socket. */
1258 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1259 if (ret < 0) {
1260 goto error;
1261 }
1262
1263 /* Add metadata socket that is successfully connected. */
1264 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1265 LPOLLIN | LPOLLRDHUP);
1266 if (ret < 0) {
1267 goto error;
1268 }
1269
1270 health_code_update();
1271
1272 /* Infinite blocking call, waiting for transmission */
1273 restart_poll:
1274 while (1) {
1275 health_poll_entry();
1276 ret = lttng_poll_wait(&events, -1);
1277 health_poll_exit();
1278 if (ret < 0) {
1279 /*
1280 * Restart interrupted system call.
1281 */
1282 if (errno == EINTR) {
1283 goto restart_poll;
1284 }
1285 goto error;
1286 }
1287
1288 nb_fd = ret;
1289
1290 for (i = 0; i < nb_fd; i++) {
1291 /* Fetch once the poll data */
1292 revents = LTTNG_POLL_GETEV(&events, i);
1293 pollfd = LTTNG_POLL_GETFD(&events, i);
1294
1295 health_code_update();
1296
1297 /* Thread quit pipe has been closed. Killing thread. */
1298 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1299 if (ret) {
1300 err = 0;
1301 goto exit;
1302 }
1303
1304 if (pollfd == sock) {
1305 /* Event on the consumerd socket */
1306 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1307 ERR("consumer err socket second poll error");
1308 goto error;
1309 }
1310 health_code_update();
1311 /* Wait for any kconsumerd error */
1312 ret = lttcomm_recv_unix_sock(sock, &code,
1313 sizeof(enum lttcomm_return_code));
1314 if (ret <= 0) {
1315 ERR("consumer closed the command socket");
1316 goto error;
1317 }
1318
1319 ERR("consumer return code : %s",
1320 lttcomm_get_readable_code(-code));
1321
1322 goto exit;
1323 } else if (pollfd == consumer_data->metadata_fd) {
1324 /* UST metadata requests */
1325 ret = ust_consumer_metadata_request(
1326 &consumer_data->metadata_sock);
1327 if (ret < 0) {
1328 ERR("Handling metadata request");
1329 goto error;
1330 }
1331 break;
1332 } else {
1333 ERR("Unknown pollfd");
1334 goto error;
1335 }
1336 }
1337 health_code_update();
1338 }
1339
1340 exit:
1341 error:
1342 /*
1343 * We lock here because we are about to close the sockets and some other
1344 * thread might be using them so get exclusive access which will abort all
1345 * other consumer command by other threads.
1346 */
1347 pthread_mutex_lock(&consumer_data->lock);
1348
1349 /* Immediately set the consumerd state to stopped */
1350 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1351 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1352 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1353 consumer_data->type == LTTNG_CONSUMER32_UST) {
1354 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1355 } else {
1356 /* Code flow error... */
1357 assert(0);
1358 }
1359
1360 if (consumer_data->err_sock >= 0) {
1361 ret = close(consumer_data->err_sock);
1362 if (ret) {
1363 PERROR("close");
1364 }
1365 consumer_data->err_sock = -1;
1366 }
1367 if (consumer_data->cmd_sock >= 0) {
1368 ret = close(consumer_data->cmd_sock);
1369 if (ret) {
1370 PERROR("close");
1371 }
1372 consumer_data->cmd_sock = -1;
1373 }
1374 if (consumer_data->metadata_sock.fd_ptr &&
1375 *consumer_data->metadata_sock.fd_ptr >= 0) {
1376 ret = close(*consumer_data->metadata_sock.fd_ptr);
1377 if (ret) {
1378 PERROR("close");
1379 }
1380 }
1381 if (sock >= 0) {
1382 ret = close(sock);
1383 if (ret) {
1384 PERROR("close");
1385 }
1386 }
1387
1388 unlink(consumer_data->err_unix_sock_path);
1389 unlink(consumer_data->cmd_unix_sock_path);
1390 consumer_data->pid = 0;
1391 pthread_mutex_unlock(&consumer_data->lock);
1392
1393 /* Cleanup metadata socket mutex. */
1394 if (consumer_data->metadata_sock.lock) {
1395 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1396 free(consumer_data->metadata_sock.lock);
1397 }
1398 lttng_poll_clean(&events);
1399 error_poll:
1400 if (err) {
1401 health_error();
1402 ERR("Health error occurred in %s", __func__);
1403 }
1404 health_unregister(health_sessiond);
1405 DBG("consumer thread cleanup completed");
1406
1407 return NULL;
1408 }
1409
1410 /*
1411 * This thread manage application communication.
1412 */
1413 static void *thread_manage_apps(void *data)
1414 {
1415 int i, ret, pollfd, err = -1;
1416 ssize_t size_ret;
1417 uint32_t revents, nb_fd;
1418 struct lttng_poll_event events;
1419
1420 DBG("[thread] Manage application started");
1421
1422 rcu_register_thread();
1423 rcu_thread_online();
1424
1425 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1426
1427 if (testpoint(sessiond_thread_manage_apps)) {
1428 goto error_testpoint;
1429 }
1430
1431 health_code_update();
1432
1433 ret = sessiond_set_thread_pollset(&events, 2);
1434 if (ret < 0) {
1435 goto error_poll_create;
1436 }
1437
1438 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1439 if (ret < 0) {
1440 goto error;
1441 }
1442
1443 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1444 goto error;
1445 }
1446
1447 health_code_update();
1448
1449 while (1) {
1450 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1451
1452 /* Inifinite blocking call, waiting for transmission */
1453 restart:
1454 health_poll_entry();
1455 ret = lttng_poll_wait(&events, -1);
1456 health_poll_exit();
1457 if (ret < 0) {
1458 /*
1459 * Restart interrupted system call.
1460 */
1461 if (errno == EINTR) {
1462 goto restart;
1463 }
1464 goto error;
1465 }
1466
1467 nb_fd = ret;
1468
1469 for (i = 0; i < nb_fd; i++) {
1470 /* Fetch once the poll data */
1471 revents = LTTNG_POLL_GETEV(&events, i);
1472 pollfd = LTTNG_POLL_GETFD(&events, i);
1473
1474 health_code_update();
1475
1476 /* Thread quit pipe has been closed. Killing thread. */
1477 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1478 if (ret) {
1479 err = 0;
1480 goto exit;
1481 }
1482
1483 /* Inspect the apps cmd pipe */
1484 if (pollfd == apps_cmd_pipe[0]) {
1485 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1486 ERR("Apps command pipe error");
1487 goto error;
1488 } else if (revents & LPOLLIN) {
1489 int sock;
1490
1491 /* Empty pipe */
1492 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1493 if (size_ret < sizeof(sock)) {
1494 PERROR("read apps cmd pipe");
1495 goto error;
1496 }
1497
1498 health_code_update();
1499
1500 /*
1501 * We only monitor the error events of the socket. This
1502 * thread does not handle any incoming data from UST
1503 * (POLLIN).
1504 */
1505 ret = lttng_poll_add(&events, sock,
1506 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1507 if (ret < 0) {
1508 goto error;
1509 }
1510
1511 DBG("Apps with sock %d added to poll set", sock);
1512 }
1513 } else {
1514 /*
1515 * At this point, we know that a registered application made
1516 * the event at poll_wait.
1517 */
1518 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1519 /* Removing from the poll set */
1520 ret = lttng_poll_del(&events, pollfd);
1521 if (ret < 0) {
1522 goto error;
1523 }
1524
1525 /* Socket closed on remote end. */
1526 ust_app_unregister(pollfd);
1527 }
1528 }
1529
1530 health_code_update();
1531 }
1532 }
1533
1534 exit:
1535 error:
1536 lttng_poll_clean(&events);
1537 error_poll_create:
1538 error_testpoint:
1539 utils_close_pipe(apps_cmd_pipe);
1540 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1541
1542 /*
1543 * We don't clean the UST app hash table here since already registered
1544 * applications can still be controlled so let them be until the session
1545 * daemon dies or the applications stop.
1546 */
1547
1548 if (err) {
1549 health_error();
1550 ERR("Health error occurred in %s", __func__);
1551 }
1552 health_unregister(health_sessiond);
1553 DBG("Application communication apps thread cleanup complete");
1554 rcu_thread_offline();
1555 rcu_unregister_thread();
1556 return NULL;
1557 }
1558
1559 /*
1560 * Send a socket to a thread This is called from the dispatch UST registration
1561 * thread once all sockets are set for the application.
1562 *
1563 * The sock value can be invalid, we don't really care, the thread will handle
1564 * it and make the necessary cleanup if so.
1565 *
1566 * On success, return 0 else a negative value being the errno message of the
1567 * write().
1568 */
1569 static int send_socket_to_thread(int fd, int sock)
1570 {
1571 ssize_t ret;
1572
1573 /*
1574 * It's possible that the FD is set as invalid with -1 concurrently just
1575 * before calling this function being a shutdown state of the thread.
1576 */
1577 if (fd < 0) {
1578 ret = -EBADF;
1579 goto error;
1580 }
1581
1582 ret = lttng_write(fd, &sock, sizeof(sock));
1583 if (ret < sizeof(sock)) {
1584 PERROR("write apps pipe %d", fd);
1585 if (ret < 0) {
1586 ret = -errno;
1587 }
1588 goto error;
1589 }
1590
1591 /* All good. Don't send back the write positive ret value. */
1592 ret = 0;
1593 error:
1594 return (int) ret;
1595 }
1596
1597 /*
1598 * Sanitize the wait queue of the dispatch registration thread meaning removing
1599 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1600 * notify socket is never received.
1601 */
1602 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1603 {
1604 int ret, nb_fd = 0, i;
1605 unsigned int fd_added = 0;
1606 struct lttng_poll_event events;
1607 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1608
1609 assert(wait_queue);
1610
1611 lttng_poll_init(&events);
1612
1613 /* Just skip everything for an empty queue. */
1614 if (!wait_queue->count) {
1615 goto end;
1616 }
1617
1618 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1619 if (ret < 0) {
1620 goto error_create;
1621 }
1622
1623 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1624 &wait_queue->head, head) {
1625 assert(wait_node->app);
1626 ret = lttng_poll_add(&events, wait_node->app->sock,
1627 LPOLLHUP | LPOLLERR);
1628 if (ret < 0) {
1629 goto error;
1630 }
1631
1632 fd_added = 1;
1633 }
1634
1635 if (!fd_added) {
1636 goto end;
1637 }
1638
1639 /*
1640 * Poll but don't block so we can quickly identify the faulty events and
1641 * clean them afterwards from the wait queue.
1642 */
1643 ret = lttng_poll_wait(&events, 0);
1644 if (ret < 0) {
1645 goto error;
1646 }
1647 nb_fd = ret;
1648
1649 for (i = 0; i < nb_fd; i++) {
1650 /* Get faulty FD. */
1651 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1652 int pollfd = LTTNG_POLL_GETFD(&events, i);
1653
1654 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1655 &wait_queue->head, head) {
1656 if (pollfd == wait_node->app->sock &&
1657 (revents & (LPOLLHUP | LPOLLERR))) {
1658 cds_list_del(&wait_node->head);
1659 wait_queue->count--;
1660 ust_app_destroy(wait_node->app);
1661 free(wait_node);
1662 break;
1663 }
1664 }
1665 }
1666
1667 if (nb_fd > 0) {
1668 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1669 }
1670
1671 end:
1672 lttng_poll_clean(&events);
1673 return;
1674
1675 error:
1676 lttng_poll_clean(&events);
1677 error_create:
1678 ERR("Unable to sanitize wait queue");
1679 return;
1680 }
1681
1682 /*
1683 * Dispatch request from the registration threads to the application
1684 * communication thread.
1685 */
1686 static void *thread_dispatch_ust_registration(void *data)
1687 {
1688 int ret, err = -1;
1689 struct cds_wfcq_node *node;
1690 struct ust_command *ust_cmd = NULL;
1691 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1692 struct ust_reg_wait_queue wait_queue = {
1693 .count = 0,
1694 };
1695
1696 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1697
1698 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1699 goto error_testpoint;
1700 }
1701
1702 health_code_update();
1703
1704 CDS_INIT_LIST_HEAD(&wait_queue.head);
1705
1706 DBG("[thread] Dispatch UST command started");
1707
1708 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1709 health_code_update();
1710
1711 /* Atomically prepare the queue futex */
1712 futex_nto1_prepare(&ust_cmd_queue.futex);
1713
1714 do {
1715 struct ust_app *app = NULL;
1716 ust_cmd = NULL;
1717
1718 /*
1719 * Make sure we don't have node(s) that have hung up before receiving
1720 * the notify socket. This is to clean the list in order to avoid
1721 * memory leaks from notify socket that are never seen.
1722 */
1723 sanitize_wait_queue(&wait_queue);
1724
1725 health_code_update();
1726 /* Dequeue command for registration */
1727 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1728 if (node == NULL) {
1729 DBG("Woken up but nothing in the UST command queue");
1730 /* Continue thread execution */
1731 break;
1732 }
1733
1734 ust_cmd = caa_container_of(node, struct ust_command, node);
1735
1736 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1737 " gid:%d sock:%d name:%s (version %d.%d)",
1738 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1739 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1740 ust_cmd->sock, ust_cmd->reg_msg.name,
1741 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1742
1743 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1744 wait_node = zmalloc(sizeof(*wait_node));
1745 if (!wait_node) {
1746 PERROR("zmalloc wait_node dispatch");
1747 ret = close(ust_cmd->sock);
1748 if (ret < 0) {
1749 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1750 }
1751 lttng_fd_put(LTTNG_FD_APPS, 1);
1752 free(ust_cmd);
1753 goto error;
1754 }
1755 CDS_INIT_LIST_HEAD(&wait_node->head);
1756
1757 /* Create application object if socket is CMD. */
1758 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1759 ust_cmd->sock);
1760 if (!wait_node->app) {
1761 ret = close(ust_cmd->sock);
1762 if (ret < 0) {
1763 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1764 }
1765 lttng_fd_put(LTTNG_FD_APPS, 1);
1766 free(wait_node);
1767 free(ust_cmd);
1768 continue;
1769 }
1770 /*
1771 * Add application to the wait queue so we can set the notify
1772 * socket before putting this object in the global ht.
1773 */
1774 cds_list_add(&wait_node->head, &wait_queue.head);
1775 wait_queue.count++;
1776
1777 free(ust_cmd);
1778 /*
1779 * We have to continue here since we don't have the notify
1780 * socket and the application MUST be added to the hash table
1781 * only at that moment.
1782 */
1783 continue;
1784 } else {
1785 /*
1786 * Look for the application in the local wait queue and set the
1787 * notify socket if found.
1788 */
1789 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1790 &wait_queue.head, head) {
1791 health_code_update();
1792 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1793 wait_node->app->notify_sock = ust_cmd->sock;
1794 cds_list_del(&wait_node->head);
1795 wait_queue.count--;
1796 app = wait_node->app;
1797 free(wait_node);
1798 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1799 break;
1800 }
1801 }
1802
1803 /*
1804 * With no application at this stage the received socket is
1805 * basically useless so close it before we free the cmd data
1806 * structure for good.
1807 */
1808 if (!app) {
1809 ret = close(ust_cmd->sock);
1810 if (ret < 0) {
1811 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1812 }
1813 lttng_fd_put(LTTNG_FD_APPS, 1);
1814 }
1815 free(ust_cmd);
1816 }
1817
1818 if (app) {
1819 /*
1820 * @session_lock_list
1821 *
1822 * Lock the global session list so from the register up to the
1823 * registration done message, no thread can see the application
1824 * and change its state.
1825 */
1826 session_lock_list();
1827 rcu_read_lock();
1828
1829 /*
1830 * Add application to the global hash table. This needs to be
1831 * done before the update to the UST registry can locate the
1832 * application.
1833 */
1834 ust_app_add(app);
1835
1836 /* Set app version. This call will print an error if needed. */
1837 (void) ust_app_version(app);
1838
1839 /* Send notify socket through the notify pipe. */
1840 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1841 app->notify_sock);
1842 if (ret < 0) {
1843 rcu_read_unlock();
1844 session_unlock_list();
1845 /*
1846 * No notify thread, stop the UST tracing. However, this is
1847 * not an internal error of the this thread thus setting
1848 * the health error code to a normal exit.
1849 */
1850 err = 0;
1851 goto error;
1852 }
1853
1854 /*
1855 * Update newly registered application with the tracing
1856 * registry info already enabled information.
1857 */
1858 update_ust_app(app->sock);
1859
1860 /*
1861 * Don't care about return value. Let the manage apps threads
1862 * handle app unregistration upon socket close.
1863 */
1864 (void) ust_app_register_done(app->sock);
1865
1866 /*
1867 * Even if the application socket has been closed, send the app
1868 * to the thread and unregistration will take place at that
1869 * place.
1870 */
1871 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1872 if (ret < 0) {
1873 rcu_read_unlock();
1874 session_unlock_list();
1875 /*
1876 * No apps. thread, stop the UST tracing. However, this is
1877 * not an internal error of the this thread thus setting
1878 * the health error code to a normal exit.
1879 */
1880 err = 0;
1881 goto error;
1882 }
1883
1884 rcu_read_unlock();
1885 session_unlock_list();
1886 }
1887 } while (node != NULL);
1888
1889 health_poll_entry();
1890 /* Futex wait on queue. Blocking call on futex() */
1891 futex_nto1_wait(&ust_cmd_queue.futex);
1892 health_poll_exit();
1893 }
1894 /* Normal exit, no error */
1895 err = 0;
1896
1897 error:
1898 /* Clean up wait queue. */
1899 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1900 &wait_queue.head, head) {
1901 cds_list_del(&wait_node->head);
1902 wait_queue.count--;
1903 free(wait_node);
1904 }
1905
1906 error_testpoint:
1907 DBG("Dispatch thread dying");
1908 if (err) {
1909 health_error();
1910 ERR("Health error occurred in %s", __func__);
1911 }
1912 health_unregister(health_sessiond);
1913 return NULL;
1914 }
1915
1916 /*
1917 * This thread manage application registration.
1918 */
1919 static void *thread_registration_apps(void *data)
1920 {
1921 int sock = -1, i, ret, pollfd, err = -1;
1922 uint32_t revents, nb_fd;
1923 struct lttng_poll_event events;
1924 /*
1925 * Get allocated in this thread, enqueued to a global queue, dequeued and
1926 * freed in the manage apps thread.
1927 */
1928 struct ust_command *ust_cmd = NULL;
1929
1930 DBG("[thread] Manage application registration started");
1931
1932 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1933
1934 if (testpoint(sessiond_thread_registration_apps)) {
1935 goto error_testpoint;
1936 }
1937
1938 ret = lttcomm_listen_unix_sock(apps_sock);
1939 if (ret < 0) {
1940 goto error_listen;
1941 }
1942
1943 /*
1944 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1945 * more will be added to this poll set.
1946 */
1947 ret = sessiond_set_thread_pollset(&events, 2);
1948 if (ret < 0) {
1949 goto error_create_poll;
1950 }
1951
1952 /* Add the application registration socket */
1953 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1954 if (ret < 0) {
1955 goto error_poll_add;
1956 }
1957
1958 /* Notify all applications to register */
1959 ret = notify_ust_apps(1);
1960 if (ret < 0) {
1961 ERR("Failed to notify applications or create the wait shared memory.\n"
1962 "Execution continues but there might be problem for already\n"
1963 "running applications that wishes to register.");
1964 }
1965
1966 while (1) {
1967 DBG("Accepting application registration");
1968
1969 /* Inifinite blocking call, waiting for transmission */
1970 restart:
1971 health_poll_entry();
1972 ret = lttng_poll_wait(&events, -1);
1973 health_poll_exit();
1974 if (ret < 0) {
1975 /*
1976 * Restart interrupted system call.
1977 */
1978 if (errno == EINTR) {
1979 goto restart;
1980 }
1981 goto error;
1982 }
1983
1984 nb_fd = ret;
1985
1986 for (i = 0; i < nb_fd; i++) {
1987 health_code_update();
1988
1989 /* Fetch once the poll data */
1990 revents = LTTNG_POLL_GETEV(&events, i);
1991 pollfd = LTTNG_POLL_GETFD(&events, i);
1992
1993 /* Thread quit pipe has been closed. Killing thread. */
1994 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1995 if (ret) {
1996 err = 0;
1997 goto exit;
1998 }
1999
2000 /* Event on the registration socket */
2001 if (pollfd == apps_sock) {
2002 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2003 ERR("Register apps socket poll error");
2004 goto error;
2005 } else if (revents & LPOLLIN) {
2006 sock = lttcomm_accept_unix_sock(apps_sock);
2007 if (sock < 0) {
2008 goto error;
2009 }
2010
2011 /*
2012 * Set socket timeout for both receiving and ending.
2013 * app_socket_timeout is in seconds, whereas
2014 * lttcomm_setsockopt_rcv_timeout and
2015 * lttcomm_setsockopt_snd_timeout expect msec as
2016 * parameter.
2017 */
2018 (void) lttcomm_setsockopt_rcv_timeout(sock,
2019 app_socket_timeout * 1000);
2020 (void) lttcomm_setsockopt_snd_timeout(sock,
2021 app_socket_timeout * 1000);
2022
2023 /*
2024 * Set the CLOEXEC flag. Return code is useless because
2025 * either way, the show must go on.
2026 */
2027 (void) utils_set_fd_cloexec(sock);
2028
2029 /* Create UST registration command for enqueuing */
2030 ust_cmd = zmalloc(sizeof(struct ust_command));
2031 if (ust_cmd == NULL) {
2032 PERROR("ust command zmalloc");
2033 goto error;
2034 }
2035
2036 /*
2037 * Using message-based transmissions to ensure we don't
2038 * have to deal with partially received messages.
2039 */
2040 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2041 if (ret < 0) {
2042 ERR("Exhausted file descriptors allowed for applications.");
2043 free(ust_cmd);
2044 ret = close(sock);
2045 if (ret) {
2046 PERROR("close");
2047 }
2048 sock = -1;
2049 continue;
2050 }
2051
2052 health_code_update();
2053 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2054 if (ret < 0) {
2055 free(ust_cmd);
2056 /* Close socket of the application. */
2057 ret = close(sock);
2058 if (ret) {
2059 PERROR("close");
2060 }
2061 lttng_fd_put(LTTNG_FD_APPS, 1);
2062 sock = -1;
2063 continue;
2064 }
2065 health_code_update();
2066
2067 ust_cmd->sock = sock;
2068 sock = -1;
2069
2070 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2071 " gid:%d sock:%d name:%s (version %d.%d)",
2072 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2073 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2074 ust_cmd->sock, ust_cmd->reg_msg.name,
2075 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2076
2077 /*
2078 * Lock free enqueue the registration request. The red pill
2079 * has been taken! This apps will be part of the *system*.
2080 */
2081 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2082
2083 /*
2084 * Wake the registration queue futex. Implicit memory
2085 * barrier with the exchange in cds_wfcq_enqueue.
2086 */
2087 futex_nto1_wake(&ust_cmd_queue.futex);
2088 }
2089 }
2090 }
2091 }
2092
2093 exit:
2094 error:
2095 /* Notify that the registration thread is gone */
2096 notify_ust_apps(0);
2097
2098 if (apps_sock >= 0) {
2099 ret = close(apps_sock);
2100 if (ret) {
2101 PERROR("close");
2102 }
2103 }
2104 if (sock >= 0) {
2105 ret = close(sock);
2106 if (ret) {
2107 PERROR("close");
2108 }
2109 lttng_fd_put(LTTNG_FD_APPS, 1);
2110 }
2111 unlink(apps_unix_sock_path);
2112
2113 error_poll_add:
2114 lttng_poll_clean(&events);
2115 error_listen:
2116 error_create_poll:
2117 error_testpoint:
2118 DBG("UST Registration thread cleanup complete");
2119 if (err) {
2120 health_error();
2121 ERR("Health error occurred in %s", __func__);
2122 }
2123 health_unregister(health_sessiond);
2124
2125 return NULL;
2126 }
2127
2128 /*
2129 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2130 * exec or it will fails.
2131 */
2132 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2133 {
2134 int ret, clock_ret;
2135 struct timespec timeout;
2136
2137 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2138 consumer_data->consumer_thread_is_ready = 0;
2139
2140 /* Setup pthread condition */
2141 ret = pthread_condattr_init(&consumer_data->condattr);
2142 if (ret != 0) {
2143 errno = ret;
2144 PERROR("pthread_condattr_init consumer data");
2145 goto error;
2146 }
2147
2148 /*
2149 * Set the monotonic clock in order to make sure we DO NOT jump in time
2150 * between the clock_gettime() call and the timedwait call. See bug #324
2151 * for a more details and how we noticed it.
2152 */
2153 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2154 if (ret != 0) {
2155 errno = ret;
2156 PERROR("pthread_condattr_setclock consumer data");
2157 goto error;
2158 }
2159
2160 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2161 if (ret != 0) {
2162 errno = ret;
2163 PERROR("pthread_cond_init consumer data");
2164 goto error;
2165 }
2166
2167 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2168 consumer_data);
2169 if (ret != 0) {
2170 PERROR("pthread_create consumer");
2171 ret = -1;
2172 goto error;
2173 }
2174
2175 /* We are about to wait on a pthread condition */
2176 pthread_mutex_lock(&consumer_data->cond_mutex);
2177
2178 /* Get time for sem_timedwait absolute timeout */
2179 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2180 /*
2181 * Set the timeout for the condition timed wait even if the clock gettime
2182 * call fails since we might loop on that call and we want to avoid to
2183 * increment the timeout too many times.
2184 */
2185 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2186
2187 /*
2188 * The following loop COULD be skipped in some conditions so this is why we
2189 * set ret to 0 in order to make sure at least one round of the loop is
2190 * done.
2191 */
2192 ret = 0;
2193
2194 /*
2195 * Loop until the condition is reached or when a timeout is reached. Note
2196 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2197 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2198 * possible. This loop does not take any chances and works with both of
2199 * them.
2200 */
2201 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2202 if (clock_ret < 0) {
2203 PERROR("clock_gettime spawn consumer");
2204 /* Infinite wait for the consumerd thread to be ready */
2205 ret = pthread_cond_wait(&consumer_data->cond,
2206 &consumer_data->cond_mutex);
2207 } else {
2208 ret = pthread_cond_timedwait(&consumer_data->cond,
2209 &consumer_data->cond_mutex, &timeout);
2210 }
2211 }
2212
2213 /* Release the pthread condition */
2214 pthread_mutex_unlock(&consumer_data->cond_mutex);
2215
2216 if (ret != 0) {
2217 errno = ret;
2218 if (ret == ETIMEDOUT) {
2219 int pth_ret;
2220
2221 /*
2222 * Call has timed out so we kill the kconsumerd_thread and return
2223 * an error.
2224 */
2225 ERR("Condition timed out. The consumer thread was never ready."
2226 " Killing it");
2227 pth_ret = pthread_cancel(consumer_data->thread);
2228 if (pth_ret < 0) {
2229 PERROR("pthread_cancel consumer thread");
2230 }
2231 } else {
2232 PERROR("pthread_cond_wait failed consumer thread");
2233 }
2234 /* Caller is expecting a negative value on failure. */
2235 ret = -1;
2236 goto error;
2237 }
2238
2239 pthread_mutex_lock(&consumer_data->pid_mutex);
2240 if (consumer_data->pid == 0) {
2241 ERR("Consumerd did not start");
2242 pthread_mutex_unlock(&consumer_data->pid_mutex);
2243 goto error;
2244 }
2245 pthread_mutex_unlock(&consumer_data->pid_mutex);
2246
2247 return 0;
2248
2249 error:
2250 return ret;
2251 }
2252
2253 /*
2254 * Join consumer thread
2255 */
2256 static int join_consumer_thread(struct consumer_data *consumer_data)
2257 {
2258 void *status;
2259
2260 /* Consumer pid must be a real one. */
2261 if (consumer_data->pid > 0) {
2262 int ret;
2263 ret = kill(consumer_data->pid, SIGTERM);
2264 if (ret) {
2265 ERR("Error killing consumer daemon");
2266 return ret;
2267 }
2268 return pthread_join(consumer_data->thread, &status);
2269 } else {
2270 return 0;
2271 }
2272 }
2273
2274 /*
2275 * Fork and exec a consumer daemon (consumerd).
2276 *
2277 * Return pid if successful else -1.
2278 */
2279 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2280 {
2281 int ret;
2282 pid_t pid;
2283 const char *consumer_to_use;
2284 const char *verbosity;
2285 struct stat st;
2286
2287 DBG("Spawning consumerd");
2288
2289 pid = fork();
2290 if (pid == 0) {
2291 /*
2292 * Exec consumerd.
2293 */
2294 if (opt_verbose_consumer) {
2295 verbosity = "--verbose";
2296 } else if (lttng_opt_quiet) {
2297 verbosity = "--quiet";
2298 } else {
2299 verbosity = "";
2300 }
2301
2302 switch (consumer_data->type) {
2303 case LTTNG_CONSUMER_KERNEL:
2304 /*
2305 * Find out which consumerd to execute. We will first try the
2306 * 64-bit path, then the sessiond's installation directory, and
2307 * fallback on the 32-bit one,
2308 */
2309 DBG3("Looking for a kernel consumer at these locations:");
2310 DBG3(" 1) %s", consumerd64_bin);
2311 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2312 DBG3(" 3) %s", consumerd32_bin);
2313 if (stat(consumerd64_bin, &st) == 0) {
2314 DBG3("Found location #1");
2315 consumer_to_use = consumerd64_bin;
2316 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2317 DBG3("Found location #2");
2318 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2319 } else if (stat(consumerd32_bin, &st) == 0) {
2320 DBG3("Found location #3");
2321 consumer_to_use = consumerd32_bin;
2322 } else {
2323 DBG("Could not find any valid consumerd executable");
2324 ret = -EINVAL;
2325 break;
2326 }
2327 DBG("Using kernel consumer at: %s", consumer_to_use);
2328 ret = execl(consumer_to_use,
2329 "lttng-consumerd", verbosity, "-k",
2330 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2331 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2332 "--group", tracing_group_name,
2333 NULL);
2334 break;
2335 case LTTNG_CONSUMER64_UST:
2336 {
2337 char *tmpnew = NULL;
2338
2339 if (consumerd64_libdir[0] != '\0') {
2340 char *tmp;
2341 size_t tmplen;
2342
2343 tmp = getenv("LD_LIBRARY_PATH");
2344 if (!tmp) {
2345 tmp = "";
2346 }
2347 tmplen = strlen("LD_LIBRARY_PATH=")
2348 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2349 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2350 if (!tmpnew) {
2351 ret = -ENOMEM;
2352 goto error;
2353 }
2354 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2355 strcat(tmpnew, consumerd64_libdir);
2356 if (tmp[0] != '\0') {
2357 strcat(tmpnew, ":");
2358 strcat(tmpnew, tmp);
2359 }
2360 ret = putenv(tmpnew);
2361 if (ret) {
2362 ret = -errno;
2363 free(tmpnew);
2364 goto error;
2365 }
2366 }
2367 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2368 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2369 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2370 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2371 "--group", tracing_group_name,
2372 NULL);
2373 if (consumerd64_libdir[0] != '\0') {
2374 free(tmpnew);
2375 }
2376 break;
2377 }
2378 case LTTNG_CONSUMER32_UST:
2379 {
2380 char *tmpnew = NULL;
2381
2382 if (consumerd32_libdir[0] != '\0') {
2383 char *tmp;
2384 size_t tmplen;
2385
2386 tmp = getenv("LD_LIBRARY_PATH");
2387 if (!tmp) {
2388 tmp = "";
2389 }
2390 tmplen = strlen("LD_LIBRARY_PATH=")
2391 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2392 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2393 if (!tmpnew) {
2394 ret = -ENOMEM;
2395 goto error;
2396 }
2397 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2398 strcat(tmpnew, consumerd32_libdir);
2399 if (tmp[0] != '\0') {
2400 strcat(tmpnew, ":");
2401 strcat(tmpnew, tmp);
2402 }
2403 ret = putenv(tmpnew);
2404 if (ret) {
2405 ret = -errno;
2406 free(tmpnew);
2407 goto error;
2408 }
2409 }
2410 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2411 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2412 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2413 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2414 "--group", tracing_group_name,
2415 NULL);
2416 if (consumerd32_libdir[0] != '\0') {
2417 free(tmpnew);
2418 }
2419 break;
2420 }
2421 default:
2422 PERROR("unknown consumer type");
2423 exit(EXIT_FAILURE);
2424 }
2425 if (errno != 0) {
2426 PERROR("Consumer execl()");
2427 }
2428 /* Reaching this point, we got a failure on our execl(). */
2429 exit(EXIT_FAILURE);
2430 } else if (pid > 0) {
2431 ret = pid;
2432 } else {
2433 PERROR("start consumer fork");
2434 ret = -errno;
2435 }
2436 error:
2437 return ret;
2438 }
2439
2440 /*
2441 * Spawn the consumerd daemon and session daemon thread.
2442 */
2443 static int start_consumerd(struct consumer_data *consumer_data)
2444 {
2445 int ret;
2446
2447 /*
2448 * Set the listen() state on the socket since there is a possible race
2449 * between the exec() of the consumer daemon and this call if place in the
2450 * consumer thread. See bug #366 for more details.
2451 */
2452 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2453 if (ret < 0) {
2454 goto error;
2455 }
2456
2457 pthread_mutex_lock(&consumer_data->pid_mutex);
2458 if (consumer_data->pid != 0) {
2459 pthread_mutex_unlock(&consumer_data->pid_mutex);
2460 goto end;
2461 }
2462
2463 ret = spawn_consumerd(consumer_data);
2464 if (ret < 0) {
2465 ERR("Spawning consumerd failed");
2466 pthread_mutex_unlock(&consumer_data->pid_mutex);
2467 goto error;
2468 }
2469
2470 /* Setting up the consumer_data pid */
2471 consumer_data->pid = ret;
2472 DBG2("Consumer pid %d", consumer_data->pid);
2473 pthread_mutex_unlock(&consumer_data->pid_mutex);
2474
2475 DBG2("Spawning consumer control thread");
2476 ret = spawn_consumer_thread(consumer_data);
2477 if (ret < 0) {
2478 ERR("Fatal error spawning consumer control thread");
2479 goto error;
2480 }
2481
2482 end:
2483 return 0;
2484
2485 error:
2486 /* Cleanup already created sockets on error. */
2487 if (consumer_data->err_sock >= 0) {
2488 int err;
2489
2490 err = close(consumer_data->err_sock);
2491 if (err < 0) {
2492 PERROR("close consumer data error socket");
2493 }
2494 }
2495 return ret;
2496 }
2497
2498 /*
2499 * Setup necessary data for kernel tracer action.
2500 */
2501 static int init_kernel_tracer(void)
2502 {
2503 int ret;
2504
2505 /* Modprobe lttng kernel modules */
2506 ret = modprobe_lttng_control();
2507 if (ret < 0) {
2508 goto error;
2509 }
2510
2511 /* Open debugfs lttng */
2512 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2513 if (kernel_tracer_fd < 0) {
2514 DBG("Failed to open %s", module_proc_lttng);
2515 ret = -1;
2516 goto error_open;
2517 }
2518
2519 /* Validate kernel version */
2520 ret = kernel_validate_version(kernel_tracer_fd);
2521 if (ret < 0) {
2522 goto error_version;
2523 }
2524
2525 ret = modprobe_lttng_data();
2526 if (ret < 0) {
2527 goto error_modules;
2528 }
2529
2530 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2531 return 0;
2532
2533 error_version:
2534 modprobe_remove_lttng_control();
2535 ret = close(kernel_tracer_fd);
2536 if (ret) {
2537 PERROR("close");
2538 }
2539 kernel_tracer_fd = -1;
2540 return LTTNG_ERR_KERN_VERSION;
2541
2542 error_modules:
2543 ret = close(kernel_tracer_fd);
2544 if (ret) {
2545 PERROR("close");
2546 }
2547
2548 error_open:
2549 modprobe_remove_lttng_control();
2550
2551 error:
2552 WARN("No kernel tracer available");
2553 kernel_tracer_fd = -1;
2554 if (!is_root) {
2555 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2556 } else {
2557 return LTTNG_ERR_KERN_NA;
2558 }
2559 }
2560
2561
2562 /*
2563 * Copy consumer output from the tracing session to the domain session. The
2564 * function also applies the right modification on a per domain basis for the
2565 * trace files destination directory.
2566 *
2567 * Should *NOT* be called with RCU read-side lock held.
2568 */
2569 static int copy_session_consumer(int domain, struct ltt_session *session)
2570 {
2571 int ret;
2572 const char *dir_name;
2573 struct consumer_output *consumer;
2574
2575 assert(session);
2576 assert(session->consumer);
2577
2578 switch (domain) {
2579 case LTTNG_DOMAIN_KERNEL:
2580 DBG3("Copying tracing session consumer output in kernel session");
2581 /*
2582 * XXX: We should audit the session creation and what this function
2583 * does "extra" in order to avoid a destroy since this function is used
2584 * in the domain session creation (kernel and ust) only. Same for UST
2585 * domain.
2586 */
2587 if (session->kernel_session->consumer) {
2588 consumer_destroy_output(session->kernel_session->consumer);
2589 }
2590 session->kernel_session->consumer =
2591 consumer_copy_output(session->consumer);
2592 /* Ease our life a bit for the next part */
2593 consumer = session->kernel_session->consumer;
2594 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2595 break;
2596 case LTTNG_DOMAIN_JUL:
2597 case LTTNG_DOMAIN_LOG4J:
2598 case LTTNG_DOMAIN_UST:
2599 DBG3("Copying tracing session consumer output in UST session");
2600 if (session->ust_session->consumer) {
2601 consumer_destroy_output(session->ust_session->consumer);
2602 }
2603 session->ust_session->consumer =
2604 consumer_copy_output(session->consumer);
2605 /* Ease our life a bit for the next part */
2606 consumer = session->ust_session->consumer;
2607 dir_name = DEFAULT_UST_TRACE_DIR;
2608 break;
2609 default:
2610 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2611 goto error;
2612 }
2613
2614 /* Append correct directory to subdir */
2615 strncat(consumer->subdir, dir_name,
2616 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2617 DBG3("Copy session consumer subdir %s", consumer->subdir);
2618
2619 ret = LTTNG_OK;
2620
2621 error:
2622 return ret;
2623 }
2624
2625 /*
2626 * Create an UST session and add it to the session ust list.
2627 *
2628 * Should *NOT* be called with RCU read-side lock held.
2629 */
2630 static int create_ust_session(struct ltt_session *session,
2631 struct lttng_domain *domain)
2632 {
2633 int ret;
2634 struct ltt_ust_session *lus = NULL;
2635
2636 assert(session);
2637 assert(domain);
2638 assert(session->consumer);
2639
2640 switch (domain->type) {
2641 case LTTNG_DOMAIN_JUL:
2642 case LTTNG_DOMAIN_LOG4J:
2643 case LTTNG_DOMAIN_UST:
2644 break;
2645 default:
2646 ERR("Unknown UST domain on create session %d", domain->type);
2647 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2648 goto error;
2649 }
2650
2651 DBG("Creating UST session");
2652
2653 lus = trace_ust_create_session(session->id);
2654 if (lus == NULL) {
2655 ret = LTTNG_ERR_UST_SESS_FAIL;
2656 goto error;
2657 }
2658
2659 lus->uid = session->uid;
2660 lus->gid = session->gid;
2661 lus->output_traces = session->output_traces;
2662 lus->snapshot_mode = session->snapshot_mode;
2663 lus->live_timer_interval = session->live_timer;
2664 session->ust_session = lus;
2665
2666 /* Copy session output to the newly created UST session */
2667 ret = copy_session_consumer(domain->type, session);
2668 if (ret != LTTNG_OK) {
2669 goto error;
2670 }
2671
2672 return LTTNG_OK;
2673
2674 error:
2675 free(lus);
2676 session->ust_session = NULL;
2677 return ret;
2678 }
2679
2680 /*
2681 * Create a kernel tracer session then create the default channel.
2682 */
2683 static int create_kernel_session(struct ltt_session *session)
2684 {
2685 int ret;
2686
2687 DBG("Creating kernel session");
2688
2689 ret = kernel_create_session(session, kernel_tracer_fd);
2690 if (ret < 0) {
2691 ret = LTTNG_ERR_KERN_SESS_FAIL;
2692 goto error;
2693 }
2694
2695 /* Code flow safety */
2696 assert(session->kernel_session);
2697
2698 /* Copy session output to the newly created Kernel session */
2699 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2700 if (ret != LTTNG_OK) {
2701 goto error;
2702 }
2703
2704 /* Create directory(ies) on local filesystem. */
2705 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2706 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2707 ret = run_as_mkdir_recursive(
2708 session->kernel_session->consumer->dst.trace_path,
2709 S_IRWXU | S_IRWXG, session->uid, session->gid);
2710 if (ret < 0) {
2711 if (ret != -EEXIST) {
2712 ERR("Trace directory creation error");
2713 goto error;
2714 }
2715 }
2716 }
2717
2718 session->kernel_session->uid = session->uid;
2719 session->kernel_session->gid = session->gid;
2720 session->kernel_session->output_traces = session->output_traces;
2721 session->kernel_session->snapshot_mode = session->snapshot_mode;
2722
2723 return LTTNG_OK;
2724
2725 error:
2726 trace_kernel_destroy_session(session->kernel_session);
2727 session->kernel_session = NULL;
2728 return ret;
2729 }
2730
2731 /*
2732 * Count number of session permitted by uid/gid.
2733 */
2734 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2735 {
2736 unsigned int i = 0;
2737 struct ltt_session *session;
2738
2739 DBG("Counting number of available session for UID %d GID %d",
2740 uid, gid);
2741 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2742 /*
2743 * Only list the sessions the user can control.
2744 */
2745 if (!session_access_ok(session, uid, gid)) {
2746 continue;
2747 }
2748 i++;
2749 }
2750 return i;
2751 }
2752
2753 /*
2754 * Process the command requested by the lttng client within the command
2755 * context structure. This function make sure that the return structure (llm)
2756 * is set and ready for transmission before returning.
2757 *
2758 * Return any error encountered or 0 for success.
2759 *
2760 * "sock" is only used for special-case var. len data.
2761 *
2762 * Should *NOT* be called with RCU read-side lock held.
2763 */
2764 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2765 int *sock_error)
2766 {
2767 int ret = LTTNG_OK;
2768 int need_tracing_session = 1;
2769 int need_domain;
2770
2771 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2772
2773 *sock_error = 0;
2774
2775 switch (cmd_ctx->lsm->cmd_type) {
2776 case LTTNG_CREATE_SESSION:
2777 case LTTNG_CREATE_SESSION_SNAPSHOT:
2778 case LTTNG_CREATE_SESSION_LIVE:
2779 case LTTNG_DESTROY_SESSION:
2780 case LTTNG_LIST_SESSIONS:
2781 case LTTNG_LIST_DOMAINS:
2782 case LTTNG_START_TRACE:
2783 case LTTNG_STOP_TRACE:
2784 case LTTNG_DATA_PENDING:
2785 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2786 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2787 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2788 case LTTNG_SNAPSHOT_RECORD:
2789 case LTTNG_SAVE_SESSION:
2790 need_domain = 0;
2791 break;
2792 default:
2793 need_domain = 1;
2794 }
2795
2796 if (opt_no_kernel && need_domain
2797 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2798 if (!is_root) {
2799 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2800 } else {
2801 ret = LTTNG_ERR_KERN_NA;
2802 }
2803 goto error;
2804 }
2805
2806 /* Deny register consumer if we already have a spawned consumer. */
2807 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2808 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2809 if (kconsumer_data.pid > 0) {
2810 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2811 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2812 goto error;
2813 }
2814 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2815 }
2816
2817 /*
2818 * Check for command that don't needs to allocate a returned payload. We do
2819 * this here so we don't have to make the call for no payload at each
2820 * command.
2821 */
2822 switch(cmd_ctx->lsm->cmd_type) {
2823 case LTTNG_LIST_SESSIONS:
2824 case LTTNG_LIST_TRACEPOINTS:
2825 case LTTNG_LIST_TRACEPOINT_FIELDS:
2826 case LTTNG_LIST_DOMAINS:
2827 case LTTNG_LIST_CHANNELS:
2828 case LTTNG_LIST_EVENTS:
2829 break;
2830 default:
2831 /* Setup lttng message with no payload */
2832 ret = setup_lttng_msg(cmd_ctx, 0);
2833 if (ret < 0) {
2834 /* This label does not try to unlock the session */
2835 goto init_setup_error;
2836 }
2837 }
2838
2839 /* Commands that DO NOT need a session. */
2840 switch (cmd_ctx->lsm->cmd_type) {
2841 case LTTNG_CREATE_SESSION:
2842 case LTTNG_CREATE_SESSION_SNAPSHOT:
2843 case LTTNG_CREATE_SESSION_LIVE:
2844 case LTTNG_CALIBRATE:
2845 case LTTNG_LIST_SESSIONS:
2846 case LTTNG_LIST_TRACEPOINTS:
2847 case LTTNG_LIST_TRACEPOINT_FIELDS:
2848 case LTTNG_SAVE_SESSION:
2849 need_tracing_session = 0;
2850 break;
2851 default:
2852 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2853 /*
2854 * We keep the session list lock across _all_ commands
2855 * for now, because the per-session lock does not
2856 * handle teardown properly.
2857 */
2858 session_lock_list();
2859 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2860 if (cmd_ctx->session == NULL) {
2861 ret = LTTNG_ERR_SESS_NOT_FOUND;
2862 goto error;
2863 } else {
2864 /* Acquire lock for the session */
2865 session_lock(cmd_ctx->session);
2866 }
2867 break;
2868 }
2869
2870 /*
2871 * Commands that need a valid session but should NOT create one if none
2872 * exists. Instead of creating one and destroying it when the command is
2873 * handled, process that right before so we save some round trip in useless
2874 * code path.
2875 */
2876 switch (cmd_ctx->lsm->cmd_type) {
2877 case LTTNG_DISABLE_CHANNEL:
2878 case LTTNG_DISABLE_EVENT:
2879 case LTTNG_DISABLE_ALL_EVENT:
2880 switch (cmd_ctx->lsm->domain.type) {
2881 case LTTNG_DOMAIN_KERNEL:
2882 if (!cmd_ctx->session->kernel_session) {
2883 ret = LTTNG_ERR_NO_CHANNEL;
2884 goto error;
2885 }
2886 break;
2887 case LTTNG_DOMAIN_JUL:
2888 case LTTNG_DOMAIN_LOG4J:
2889 case LTTNG_DOMAIN_UST:
2890 if (!cmd_ctx->session->ust_session) {
2891 ret = LTTNG_ERR_NO_CHANNEL;
2892 goto error;
2893 }
2894 break;
2895 default:
2896 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2897 goto error;
2898 }
2899 default:
2900 break;
2901 }
2902
2903 if (!need_domain) {
2904 goto skip_domain;
2905 }
2906
2907 /*
2908 * Check domain type for specific "pre-action".
2909 */
2910 switch (cmd_ctx->lsm->domain.type) {
2911 case LTTNG_DOMAIN_KERNEL:
2912 if (!is_root) {
2913 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2914 goto error;
2915 }
2916
2917 /* Kernel tracer check */
2918 if (kernel_tracer_fd == -1) {
2919 /* Basically, load kernel tracer modules */
2920 ret = init_kernel_tracer();
2921 if (ret != 0) {
2922 goto error;
2923 }
2924 }
2925
2926 /* Consumer is in an ERROR state. Report back to client */
2927 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2928 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2929 goto error;
2930 }
2931
2932 /* Need a session for kernel command */
2933 if (need_tracing_session) {
2934 if (cmd_ctx->session->kernel_session == NULL) {
2935 ret = create_kernel_session(cmd_ctx->session);
2936 if (ret < 0) {
2937 ret = LTTNG_ERR_KERN_SESS_FAIL;
2938 goto error;
2939 }
2940 }
2941
2942 /* Start the kernel consumer daemon */
2943 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2944 if (kconsumer_data.pid == 0 &&
2945 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2946 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2947 ret = start_consumerd(&kconsumer_data);
2948 if (ret < 0) {
2949 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2950 goto error;
2951 }
2952 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2953 } else {
2954 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2955 }
2956
2957 /*
2958 * The consumer was just spawned so we need to add the socket to
2959 * the consumer output of the session if exist.
2960 */
2961 ret = consumer_create_socket(&kconsumer_data,
2962 cmd_ctx->session->kernel_session->consumer);
2963 if (ret < 0) {
2964 goto error;
2965 }
2966 }
2967
2968 break;
2969 case LTTNG_DOMAIN_JUL:
2970 case LTTNG_DOMAIN_LOG4J:
2971 case LTTNG_DOMAIN_UST:
2972 {
2973 if (!ust_app_supported()) {
2974 ret = LTTNG_ERR_NO_UST;
2975 goto error;
2976 }
2977 /* Consumer is in an ERROR state. Report back to client */
2978 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2979 ret = LTTNG_ERR_NO_USTCONSUMERD;
2980 goto error;
2981 }
2982
2983 if (need_tracing_session) {
2984 /* Create UST session if none exist. */
2985 if (cmd_ctx->session->ust_session == NULL) {
2986 ret = create_ust_session(cmd_ctx->session,
2987 &cmd_ctx->lsm->domain);
2988 if (ret != LTTNG_OK) {
2989 goto error;
2990 }
2991 }
2992
2993 /* Start the UST consumer daemons */
2994 /* 64-bit */
2995 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2996 if (consumerd64_bin[0] != '\0' &&
2997 ustconsumer64_data.pid == 0 &&
2998 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2999 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3000 ret = start_consumerd(&ustconsumer64_data);
3001 if (ret < 0) {
3002 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3003 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3004 goto error;
3005 }
3006
3007 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3008 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3009 } else {
3010 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3011 }
3012
3013 /*
3014 * Setup socket for consumer 64 bit. No need for atomic access
3015 * since it was set above and can ONLY be set in this thread.
3016 */
3017 ret = consumer_create_socket(&ustconsumer64_data,
3018 cmd_ctx->session->ust_session->consumer);
3019 if (ret < 0) {
3020 goto error;
3021 }
3022
3023 /* 32-bit */
3024 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3025 if (consumerd32_bin[0] != '\0' &&
3026 ustconsumer32_data.pid == 0 &&
3027 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3028 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3029 ret = start_consumerd(&ustconsumer32_data);
3030 if (ret < 0) {
3031 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3032 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3033 goto error;
3034 }
3035
3036 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3037 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3038 } else {
3039 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3040 }
3041
3042 /*
3043 * Setup socket for consumer 64 bit. No need for atomic access
3044 * since it was set above and can ONLY be set in this thread.
3045 */
3046 ret = consumer_create_socket(&ustconsumer32_data,
3047 cmd_ctx->session->ust_session->consumer);
3048 if (ret < 0) {
3049 goto error;
3050 }
3051 }
3052 break;
3053 }
3054 default:
3055 break;
3056 }
3057 skip_domain:
3058
3059 /* Validate consumer daemon state when start/stop trace command */
3060 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3061 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3062 switch (cmd_ctx->lsm->domain.type) {
3063 case LTTNG_DOMAIN_JUL:
3064 case LTTNG_DOMAIN_LOG4J:
3065 case LTTNG_DOMAIN_UST:
3066 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3067 ret = LTTNG_ERR_NO_USTCONSUMERD;
3068 goto error;
3069 }
3070 break;
3071 case LTTNG_DOMAIN_KERNEL:
3072 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3073 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3074 goto error;
3075 }
3076 break;
3077 }
3078 }
3079
3080 /*
3081 * Check that the UID or GID match that of the tracing session.
3082 * The root user can interact with all sessions.
3083 */
3084 if (need_tracing_session) {
3085 if (!session_access_ok(cmd_ctx->session,
3086 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3087 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3088 ret = LTTNG_ERR_EPERM;
3089 goto error;
3090 }
3091 }
3092
3093 /*
3094 * Send relayd information to consumer as soon as we have a domain and a
3095 * session defined.
3096 */
3097 if (cmd_ctx->session && need_domain) {
3098 /*
3099 * Setup relayd if not done yet. If the relayd information was already
3100 * sent to the consumer, this call will gracefully return.
3101 */
3102 ret = cmd_setup_relayd(cmd_ctx->session);
3103 if (ret != LTTNG_OK) {
3104 goto error;
3105 }
3106 }
3107
3108 /* Process by command type */
3109 switch (cmd_ctx->lsm->cmd_type) {
3110 case LTTNG_ADD_CONTEXT:
3111 {
3112 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3113 cmd_ctx->lsm->u.context.channel_name,
3114 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3115 break;
3116 }
3117 case LTTNG_DISABLE_CHANNEL:
3118 {
3119 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3120 cmd_ctx->lsm->u.disable.channel_name);
3121 break;
3122 }
3123 case LTTNG_DISABLE_EVENT:
3124 {
3125 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3126 cmd_ctx->lsm->u.disable.channel_name,
3127 cmd_ctx->lsm->u.disable.name);
3128 break;
3129 }
3130 case LTTNG_DISABLE_ALL_EVENT:
3131 {
3132 DBG("Disabling all events");
3133
3134 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3135 cmd_ctx->lsm->u.disable.channel_name);
3136 break;
3137 }
3138 case LTTNG_ENABLE_CHANNEL:
3139 {
3140 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3141 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3142 break;
3143 }
3144 case LTTNG_ENABLE_EVENT:
3145 {
3146 struct lttng_event_exclusion *exclusion = NULL;
3147 struct lttng_filter_bytecode *bytecode = NULL;
3148 char *filter_expression = NULL;
3149
3150 /* Handle exclusion events and receive it from the client. */
3151 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3152 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3153
3154 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3155 (count * LTTNG_SYMBOL_NAME_LEN));
3156 if (!exclusion) {
3157 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3158 goto error;
3159 }
3160
3161 DBG("Receiving var len exclusion event list from client ...");
3162 exclusion->count = count;
3163 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3164 count * LTTNG_SYMBOL_NAME_LEN);
3165 if (ret <= 0) {
3166 DBG("Nothing recv() from client var len data... continuing");
3167 *sock_error = 1;
3168 free(exclusion);
3169 ret = LTTNG_ERR_EXCLUSION_INVAL;
3170 goto error;
3171 }
3172 }
3173
3174 /* Get filter expression from client. */
3175 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3176 size_t expression_len =
3177 cmd_ctx->lsm->u.enable.expression_len;
3178
3179 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3180 ret = LTTNG_ERR_FILTER_INVAL;
3181 free(exclusion);
3182 goto error;
3183 }
3184
3185 filter_expression = zmalloc(expression_len);
3186 if (!filter_expression) {
3187 free(exclusion);
3188 ret = LTTNG_ERR_FILTER_NOMEM;
3189 goto error;
3190 }
3191
3192 /* Receive var. len. data */
3193 DBG("Receiving var len filter's expression from client ...");
3194 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3195 expression_len);
3196 if (ret <= 0) {
3197 DBG("Nothing recv() from client car len data... continuing");
3198 *sock_error = 1;
3199 free(filter_expression);
3200 free(exclusion);
3201 ret = LTTNG_ERR_FILTER_INVAL;
3202 goto error;
3203 }
3204 }
3205
3206 /* Handle filter and get bytecode from client. */
3207 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3208 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3209
3210 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3211 ret = LTTNG_ERR_FILTER_INVAL;
3212 free(exclusion);
3213 goto error;
3214 }
3215
3216 bytecode = zmalloc(bytecode_len);
3217 if (!bytecode) {
3218 free(exclusion);
3219 ret = LTTNG_ERR_FILTER_NOMEM;
3220 goto error;
3221 }
3222
3223 /* Receive var. len. data */
3224 DBG("Receiving var len filter's bytecode from client ...");
3225 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3226 if (ret <= 0) {
3227 DBG("Nothing recv() from client car len data... continuing");
3228 *sock_error = 1;
3229 free(bytecode);
3230 free(exclusion);
3231 ret = LTTNG_ERR_FILTER_INVAL;
3232 goto error;
3233 }
3234
3235 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3236 free(bytecode);
3237 free(exclusion);
3238 ret = LTTNG_ERR_FILTER_INVAL;
3239 goto error;
3240 }
3241 }
3242
3243 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3244 cmd_ctx->lsm->u.enable.channel_name,
3245 &cmd_ctx->lsm->u.enable.event,
3246 filter_expression, bytecode, exclusion,
3247 kernel_poll_pipe[1]);
3248 break;
3249 }
3250 case LTTNG_ENABLE_ALL_EVENT:
3251 {
3252 DBG("Enabling all events");
3253
3254 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
3255 cmd_ctx->lsm->u.enable.channel_name,
3256 cmd_ctx->lsm->u.enable.event.type, NULL, NULL,
3257 kernel_poll_pipe[1]);
3258 break;
3259 }
3260 case LTTNG_LIST_TRACEPOINTS:
3261 {
3262 struct lttng_event *events;
3263 ssize_t nb_events;
3264
3265 session_lock_list();
3266 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3267 session_unlock_list();
3268 if (nb_events < 0) {
3269 /* Return value is a negative lttng_error_code. */
3270 ret = -nb_events;
3271 goto error;
3272 }
3273
3274 /*
3275 * Setup lttng message with payload size set to the event list size in
3276 * bytes and then copy list into the llm payload.
3277 */
3278 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3279 if (ret < 0) {
3280 free(events);
3281 goto setup_error;
3282 }
3283
3284 /* Copy event list into message payload */
3285 memcpy(cmd_ctx->llm->payload, events,
3286 sizeof(struct lttng_event) * nb_events);
3287
3288 free(events);
3289
3290 ret = LTTNG_OK;
3291 break;
3292 }
3293 case LTTNG_LIST_TRACEPOINT_FIELDS:
3294 {
3295 struct lttng_event_field *fields;
3296 ssize_t nb_fields;
3297
3298 session_lock_list();
3299 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3300 &fields);
3301 session_unlock_list();
3302 if (nb_fields < 0) {
3303 /* Return value is a negative lttng_error_code. */
3304 ret = -nb_fields;
3305 goto error;
3306 }
3307
3308 /*
3309 * Setup lttng message with payload size set to the event list size in
3310 * bytes and then copy list into the llm payload.
3311 */
3312 ret = setup_lttng_msg(cmd_ctx,
3313 sizeof(struct lttng_event_field) * nb_fields);
3314 if (ret < 0) {
3315 free(fields);
3316 goto setup_error;
3317 }
3318
3319 /* Copy event list into message payload */
3320 memcpy(cmd_ctx->llm->payload, fields,
3321 sizeof(struct lttng_event_field) * nb_fields);
3322
3323 free(fields);
3324
3325 ret = LTTNG_OK;
3326 break;
3327 }
3328 case LTTNG_SET_CONSUMER_URI:
3329 {
3330 size_t nb_uri, len;
3331 struct lttng_uri *uris;
3332
3333 nb_uri = cmd_ctx->lsm->u.uri.size;
3334 len = nb_uri * sizeof(struct lttng_uri);
3335
3336 if (nb_uri == 0) {
3337 ret = LTTNG_ERR_INVALID;
3338 goto error;
3339 }
3340
3341 uris = zmalloc(len);
3342 if (uris == NULL) {
3343 ret = LTTNG_ERR_FATAL;
3344 goto error;
3345 }
3346
3347 /* Receive variable len data */
3348 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3349 ret = lttcomm_recv_unix_sock(sock, uris, len);
3350 if (ret <= 0) {
3351 DBG("No URIs received from client... continuing");
3352 *sock_error = 1;
3353 ret = LTTNG_ERR_SESSION_FAIL;
3354 free(uris);
3355 goto error;
3356 }
3357
3358 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3359 nb_uri, uris);
3360 if (ret != LTTNG_OK) {
3361 free(uris);
3362 goto error;
3363 }
3364
3365 /*
3366 * XXX: 0 means that this URI should be applied on the session. Should
3367 * be a DOMAIN enuam.
3368 */
3369 if (cmd_ctx->lsm->domain.type == 0) {
3370 /* Add the URI for the UST session if a consumer is present. */
3371 if (cmd_ctx->session->ust_session &&
3372 cmd_ctx->session->ust_session->consumer) {
3373 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3374 nb_uri, uris);
3375 } else if (cmd_ctx->session->kernel_session &&
3376 cmd_ctx->session->kernel_session->consumer) {
3377 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3378 cmd_ctx->session, nb_uri, uris);
3379 }
3380 }
3381
3382 free(uris);
3383
3384 break;
3385 }
3386 case LTTNG_START_TRACE:
3387 {
3388 ret = cmd_start_trace(cmd_ctx->session);
3389 break;
3390 }
3391 case LTTNG_STOP_TRACE:
3392 {
3393 ret = cmd_stop_trace(cmd_ctx->session);
3394 break;
3395 }
3396 case LTTNG_CREATE_SESSION:
3397 {
3398 size_t nb_uri, len;
3399 struct lttng_uri *uris = NULL;
3400
3401 nb_uri = cmd_ctx->lsm->u.uri.size;
3402 len = nb_uri * sizeof(struct lttng_uri);
3403
3404 if (nb_uri > 0) {
3405 uris = zmalloc(len);
3406 if (uris == NULL) {
3407 ret = LTTNG_ERR_FATAL;
3408 goto error;
3409 }
3410
3411 /* Receive variable len data */
3412 DBG("Waiting for %zu URIs from client ...", nb_uri);
3413 ret = lttcomm_recv_unix_sock(sock, uris, len);
3414 if (ret <= 0) {
3415 DBG("No URIs received from client... continuing");
3416 *sock_error = 1;
3417 ret = LTTNG_ERR_SESSION_FAIL;
3418 free(uris);
3419 goto error;
3420 }
3421
3422 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3423 DBG("Creating session with ONE network URI is a bad call");
3424 ret = LTTNG_ERR_SESSION_FAIL;
3425 free(uris);
3426 goto error;
3427 }
3428 }
3429
3430 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3431 &cmd_ctx->creds, 0);
3432
3433 free(uris);
3434
3435 break;
3436 }
3437 case LTTNG_DESTROY_SESSION:
3438 {
3439 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3440
3441 /* Set session to NULL so we do not unlock it after free. */
3442 cmd_ctx->session = NULL;
3443 break;
3444 }
3445 case LTTNG_LIST_DOMAINS:
3446 {
3447 ssize_t nb_dom;
3448 struct lttng_domain *domains;
3449
3450 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3451 if (nb_dom < 0) {
3452 /* Return value is a negative lttng_error_code. */
3453 ret = -nb_dom;
3454 goto error;
3455 }
3456
3457 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3458 if (ret < 0) {
3459 free(domains);
3460 goto setup_error;
3461 }
3462
3463 /* Copy event list into message payload */
3464 memcpy(cmd_ctx->llm->payload, domains,
3465 nb_dom * sizeof(struct lttng_domain));
3466
3467 free(domains);
3468
3469 ret = LTTNG_OK;
3470 break;
3471 }
3472 case LTTNG_LIST_CHANNELS:
3473 {
3474 int nb_chan;
3475 struct lttng_channel *channels;
3476
3477 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3478 cmd_ctx->session, &channels);
3479 if (nb_chan < 0) {
3480 /* Return value is a negative lttng_error_code. */
3481 ret = -nb_chan;
3482 goto error;
3483 }
3484
3485 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3486 if (ret < 0) {
3487 free(channels);
3488 goto setup_error;
3489 }
3490
3491 /* Copy event list into message payload */
3492 memcpy(cmd_ctx->llm->payload, channels,
3493 nb_chan * sizeof(struct lttng_channel));
3494
3495 free(channels);
3496
3497 ret = LTTNG_OK;
3498 break;
3499 }
3500 case LTTNG_LIST_EVENTS:
3501 {
3502 ssize_t nb_event;
3503 struct lttng_event *events = NULL;
3504
3505 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3506 cmd_ctx->lsm->u.list.channel_name, &events);
3507 if (nb_event < 0) {
3508 /* Return value is a negative lttng_error_code. */
3509 ret = -nb_event;
3510 goto error;
3511 }
3512
3513 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3514 if (ret < 0) {
3515 free(events);
3516 goto setup_error;
3517 }
3518
3519 /* Copy event list into message payload */
3520 memcpy(cmd_ctx->llm->payload, events,
3521 nb_event * sizeof(struct lttng_event));
3522
3523 free(events);
3524
3525 ret = LTTNG_OK;
3526 break;
3527 }
3528 case LTTNG_LIST_SESSIONS:
3529 {
3530 unsigned int nr_sessions;
3531
3532 session_lock_list();
3533 nr_sessions = lttng_sessions_count(
3534 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3535 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3536
3537 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3538 if (ret < 0) {
3539 session_unlock_list();
3540 goto setup_error;
3541 }
3542
3543 /* Filled the session array */
3544 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3545 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3546 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3547
3548 session_unlock_list();
3549
3550 ret = LTTNG_OK;
3551 break;
3552 }
3553 case LTTNG_CALIBRATE:
3554 {
3555 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3556 &cmd_ctx->lsm->u.calibrate);
3557 break;
3558 }
3559 case LTTNG_REGISTER_CONSUMER:
3560 {
3561 struct consumer_data *cdata;
3562
3563 switch (cmd_ctx->lsm->domain.type) {
3564 case LTTNG_DOMAIN_KERNEL:
3565 cdata = &kconsumer_data;
3566 break;
3567 default:
3568 ret = LTTNG_ERR_UND;
3569 goto error;
3570 }
3571
3572 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3573 cmd_ctx->lsm->u.reg.path, cdata);
3574 break;
3575 }
3576 case LTTNG_DATA_PENDING:
3577 {
3578 ret = cmd_data_pending(cmd_ctx->session);
3579 break;
3580 }
3581 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3582 {
3583 struct lttcomm_lttng_output_id reply;
3584
3585 ret = cmd_snapshot_add_output(cmd_ctx->session,
3586 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3587 if (ret != LTTNG_OK) {
3588 goto error;
3589 }
3590
3591 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3592 if (ret < 0) {
3593 goto setup_error;
3594 }
3595
3596 /* Copy output list into message payload */
3597 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3598 ret = LTTNG_OK;
3599 break;
3600 }
3601 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3602 {
3603 ret = cmd_snapshot_del_output(cmd_ctx->session,
3604 &cmd_ctx->lsm->u.snapshot_output.output);
3605 break;
3606 }
3607 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3608 {
3609 ssize_t nb_output;
3610 struct lttng_snapshot_output *outputs = NULL;
3611
3612 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3613 if (nb_output < 0) {
3614 ret = -nb_output;
3615 goto error;
3616 }
3617
3618 ret = setup_lttng_msg(cmd_ctx,
3619 nb_output * sizeof(struct lttng_snapshot_output));
3620 if (ret < 0) {
3621 free(outputs);
3622 goto setup_error;
3623 }
3624
3625 if (outputs) {
3626 /* Copy output list into message payload */
3627 memcpy(cmd_ctx->llm->payload, outputs,
3628 nb_output * sizeof(struct lttng_snapshot_output));
3629 free(outputs);
3630 }
3631
3632 ret = LTTNG_OK;
3633 break;
3634 }
3635 case LTTNG_SNAPSHOT_RECORD:
3636 {
3637 ret = cmd_snapshot_record(cmd_ctx->session,
3638 &cmd_ctx->lsm->u.snapshot_record.output,
3639 cmd_ctx->lsm->u.snapshot_record.wait);
3640 break;
3641 }
3642 case LTTNG_CREATE_SESSION_SNAPSHOT:
3643 {
3644 size_t nb_uri, len;
3645 struct lttng_uri *uris = NULL;
3646
3647 nb_uri = cmd_ctx->lsm->u.uri.size;
3648 len = nb_uri * sizeof(struct lttng_uri);
3649
3650 if (nb_uri > 0) {
3651 uris = zmalloc(len);
3652 if (uris == NULL) {
3653 ret = LTTNG_ERR_FATAL;
3654 goto error;
3655 }
3656
3657 /* Receive variable len data */
3658 DBG("Waiting for %zu URIs from client ...", nb_uri);
3659 ret = lttcomm_recv_unix_sock(sock, uris, len);
3660 if (ret <= 0) {
3661 DBG("No URIs received from client... continuing");
3662 *sock_error = 1;
3663 ret = LTTNG_ERR_SESSION_FAIL;
3664 free(uris);
3665 goto error;
3666 }
3667
3668 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3669 DBG("Creating session with ONE network URI is a bad call");
3670 ret = LTTNG_ERR_SESSION_FAIL;
3671 free(uris);
3672 goto error;
3673 }
3674 }
3675
3676 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3677 nb_uri, &cmd_ctx->creds);
3678 free(uris);
3679 break;
3680 }
3681 case LTTNG_CREATE_SESSION_LIVE:
3682 {
3683 size_t nb_uri, len;
3684 struct lttng_uri *uris = NULL;
3685
3686 nb_uri = cmd_ctx->lsm->u.uri.size;
3687 len = nb_uri * sizeof(struct lttng_uri);
3688
3689 if (nb_uri > 0) {
3690 uris = zmalloc(len);
3691 if (uris == NULL) {
3692 ret = LTTNG_ERR_FATAL;
3693 goto error;
3694 }
3695
3696 /* Receive variable len data */
3697 DBG("Waiting for %zu URIs from client ...", nb_uri);
3698 ret = lttcomm_recv_unix_sock(sock, uris, len);
3699 if (ret <= 0) {
3700 DBG("No URIs received from client... continuing");
3701 *sock_error = 1;
3702 ret = LTTNG_ERR_SESSION_FAIL;
3703 free(uris);
3704 goto error;
3705 }
3706
3707 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3708 DBG("Creating session with ONE network URI is a bad call");
3709 ret = LTTNG_ERR_SESSION_FAIL;
3710 free(uris);
3711 goto error;
3712 }
3713 }
3714
3715 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3716 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3717 free(uris);
3718 break;
3719 }
3720 case LTTNG_SAVE_SESSION:
3721 {
3722 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3723 &cmd_ctx->creds);
3724 break;
3725 }
3726 default:
3727 ret = LTTNG_ERR_UND;
3728 break;
3729 }
3730
3731 error:
3732 if (cmd_ctx->llm == NULL) {
3733 DBG("Missing llm structure. Allocating one.");
3734 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3735 goto setup_error;
3736 }
3737 }
3738 /* Set return code */
3739 cmd_ctx->llm->ret_code = ret;
3740 setup_error:
3741 if (cmd_ctx->session) {
3742 session_unlock(cmd_ctx->session);
3743 }
3744 if (need_tracing_session) {
3745 session_unlock_list();
3746 }
3747 init_setup_error:
3748 return ret;
3749 }
3750
3751 /*
3752 * Thread managing health check socket.
3753 */
3754 static void *thread_manage_health(void *data)
3755 {
3756 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3757 uint32_t revents, nb_fd;
3758 struct lttng_poll_event events;
3759 struct health_comm_msg msg;
3760 struct health_comm_reply reply;
3761
3762 DBG("[thread] Manage health check started");
3763
3764 rcu_register_thread();
3765
3766 /* We might hit an error path before this is created. */
3767 lttng_poll_init(&events);
3768
3769 /* Create unix socket */
3770 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3771 if (sock < 0) {
3772 ERR("Unable to create health check Unix socket");
3773 ret = -1;
3774 goto error;
3775 }
3776
3777 if (is_root) {
3778 /* lttng health client socket path permissions */
3779 ret = chown(health_unix_sock_path, 0,
3780 utils_get_group_id(tracing_group_name));
3781 if (ret < 0) {
3782 ERR("Unable to set group on %s", health_unix_sock_path);
3783 PERROR("chown");
3784 ret = -1;
3785 goto error;
3786 }
3787
3788 ret = chmod(health_unix_sock_path,
3789 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3790 if (ret < 0) {
3791 ERR("Unable to set permissions on %s", health_unix_sock_path);
3792 PERROR("chmod");
3793 ret = -1;
3794 goto error;
3795 }
3796 }
3797
3798 /*
3799 * Set the CLOEXEC flag. Return code is useless because either way, the
3800 * show must go on.
3801 */
3802 (void) utils_set_fd_cloexec(sock);
3803
3804 ret = lttcomm_listen_unix_sock(sock);
3805 if (ret < 0) {
3806 goto error;
3807 }
3808
3809 /*
3810 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3811 * more will be added to this poll set.
3812 */
3813 ret = sessiond_set_thread_pollset(&events, 2);
3814 if (ret < 0) {
3815 goto error;
3816 }
3817
3818 /* Add the application registration socket */
3819 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3820 if (ret < 0) {
3821 goto error;
3822 }
3823
3824 sessiond_notify_ready();
3825
3826 while (1) {
3827 DBG("Health check ready");
3828
3829 /* Inifinite blocking call, waiting for transmission */
3830 restart:
3831 ret = lttng_poll_wait(&events, -1);
3832 if (ret < 0) {
3833 /*
3834 * Restart interrupted system call.
3835 */
3836 if (errno == EINTR) {
3837 goto restart;
3838 }
3839 goto error;
3840 }
3841
3842 nb_fd = ret;
3843
3844 for (i = 0; i < nb_fd; i++) {
3845 /* Fetch once the poll data */
3846 revents = LTTNG_POLL_GETEV(&events, i);
3847 pollfd = LTTNG_POLL_GETFD(&events, i);
3848
3849 /* Thread quit pipe has been closed. Killing thread. */
3850 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3851 if (ret) {
3852 err = 0;
3853 goto exit;
3854 }
3855
3856 /* Event on the registration socket */
3857 if (pollfd == sock) {
3858 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3859 ERR("Health socket poll error");
3860 goto error;
3861 }
3862 }
3863 }
3864
3865 new_sock = lttcomm_accept_unix_sock(sock);
3866 if (new_sock < 0) {
3867 goto error;
3868 }
3869
3870 /*
3871 * Set the CLOEXEC flag. Return code is useless because either way, the
3872 * show must go on.
3873 */
3874 (void) utils_set_fd_cloexec(new_sock);
3875
3876 DBG("Receiving data from client for health...");
3877 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3878 if (ret <= 0) {
3879 DBG("Nothing recv() from client... continuing");
3880 ret = close(new_sock);
3881 if (ret) {
3882 PERROR("close");
3883 }
3884 new_sock = -1;
3885 continue;
3886 }
3887
3888 rcu_thread_online();
3889
3890 memset(&reply, 0, sizeof(reply));
3891 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3892 /*
3893 * health_check_state returns 0 if health is
3894 * bad.
3895 */
3896 if (!health_check_state(health_sessiond, i)) {
3897 reply.ret_code |= 1ULL << i;
3898 }
3899 }
3900
3901 DBG2("Health check return value %" PRIx64, reply.ret_code);
3902
3903 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3904 if (ret < 0) {
3905 ERR("Failed to send health data back to client");
3906 }
3907
3908 /* End of transmission */
3909 ret = close(new_sock);
3910 if (ret) {
3911 PERROR("close");
3912 }
3913 new_sock = -1;
3914 }
3915
3916 exit:
3917 error:
3918 if (err) {
3919 ERR("Health error occurred in %s", __func__);
3920 }
3921 DBG("Health check thread dying");
3922 unlink(health_unix_sock_path);
3923 if (sock >= 0) {
3924 ret = close(sock);
3925 if (ret) {
3926 PERROR("close");
3927 }
3928 }
3929
3930 lttng_poll_clean(&events);
3931
3932 rcu_unregister_thread();
3933 return NULL;
3934 }
3935
3936 /*
3937 * This thread manage all clients request using the unix client socket for
3938 * communication.
3939 */
3940 static void *thread_manage_clients(void *data)
3941 {
3942 int sock = -1, ret, i, pollfd, err = -1;
3943 int sock_error;
3944 uint32_t revents, nb_fd;
3945 struct command_ctx *cmd_ctx = NULL;
3946 struct lttng_poll_event events;
3947
3948 DBG("[thread] Manage client started");
3949
3950 rcu_register_thread();
3951
3952 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3953
3954 health_code_update();
3955
3956 ret = lttcomm_listen_unix_sock(client_sock);
3957 if (ret < 0) {
3958 goto error_listen;
3959 }
3960
3961 /*
3962 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3963 * more will be added to this poll set.
3964 */
3965 ret = sessiond_set_thread_pollset(&events, 2);
3966 if (ret < 0) {
3967 goto error_create_poll;
3968 }
3969
3970 /* Add the application registration socket */
3971 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3972 if (ret < 0) {
3973 goto error;
3974 }
3975
3976 sessiond_notify_ready();
3977 ret = sem_post(&load_info->message_thread_ready);
3978 if (ret) {
3979 PERROR("sem_post message_thread_ready");
3980 goto error;
3981 }
3982
3983 /* This testpoint is after we signal readiness to the parent. */
3984 if (testpoint(sessiond_thread_manage_clients)) {
3985 goto error;
3986 }
3987
3988 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
3989 goto error;
3990 }
3991
3992 health_code_update();
3993
3994 while (1) {
3995 DBG("Accepting client command ...");
3996
3997 /* Inifinite blocking call, waiting for transmission */
3998 restart:
3999 health_poll_entry();
4000 ret = lttng_poll_wait(&events, -1);
4001 health_poll_exit();
4002 if (ret < 0) {
4003 /*
4004 * Restart interrupted system call.
4005 */
4006 if (errno == EINTR) {
4007 goto restart;
4008 }
4009 goto error;
4010 }
4011
4012 nb_fd = ret;
4013
4014 for (i = 0; i < nb_fd; i++) {
4015 /* Fetch once the poll data */
4016 revents = LTTNG_POLL_GETEV(&events, i);
4017 pollfd = LTTNG_POLL_GETFD(&events, i);
4018
4019 health_code_update();
4020
4021 /* Thread quit pipe has been closed. Killing thread. */
4022 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4023 if (ret) {
4024 err = 0;
4025 goto exit;
4026 }
4027
4028 /* Event on the registration socket */
4029 if (pollfd == client_sock) {
4030 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4031 ERR("Client socket poll error");
4032 goto error;
4033 }
4034 }
4035 }
4036
4037 DBG("Wait for client response");
4038
4039 health_code_update();
4040
4041 sock = lttcomm_accept_unix_sock(client_sock);
4042 if (sock < 0) {
4043 goto error;
4044 }
4045
4046 /*
4047 * Set the CLOEXEC flag. Return code is useless because either way, the
4048 * show must go on.
4049 */
4050 (void) utils_set_fd_cloexec(sock);
4051
4052 /* Set socket option for credentials retrieval */
4053 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4054 if (ret < 0) {
4055 goto error;
4056 }
4057
4058 /* Allocate context command to process the client request */
4059 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4060 if (cmd_ctx == NULL) {
4061 PERROR("zmalloc cmd_ctx");
4062 goto error;
4063 }
4064
4065 /* Allocate data buffer for reception */
4066 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4067 if (cmd_ctx->lsm == NULL) {
4068 PERROR("zmalloc cmd_ctx->lsm");
4069 goto error;
4070 }
4071
4072 cmd_ctx->llm = NULL;
4073 cmd_ctx->session = NULL;
4074
4075 health_code_update();
4076
4077 /*
4078 * Data is received from the lttng client. The struct
4079 * lttcomm_session_msg (lsm) contains the command and data request of
4080 * the client.
4081 */
4082 DBG("Receiving data from client ...");
4083 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4084 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4085 if (ret <= 0) {
4086 DBG("Nothing recv() from client... continuing");
4087 ret = close(sock);
4088 if (ret) {
4089 PERROR("close");
4090 }
4091 sock = -1;
4092 clean_command_ctx(&cmd_ctx);
4093 continue;
4094 }
4095
4096 health_code_update();
4097
4098 // TODO: Validate cmd_ctx including sanity check for
4099 // security purpose.
4100
4101 rcu_thread_online();
4102 /*
4103 * This function dispatch the work to the kernel or userspace tracer
4104 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4105 * informations for the client. The command context struct contains
4106 * everything this function may needs.
4107 */
4108 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4109 rcu_thread_offline();
4110 if (ret < 0) {
4111 ret = close(sock);
4112 if (ret) {
4113 PERROR("close");
4114 }
4115 sock = -1;
4116 /*
4117 * TODO: Inform client somehow of the fatal error. At
4118 * this point, ret < 0 means that a zmalloc failed
4119 * (ENOMEM). Error detected but still accept
4120 * command, unless a socket error has been
4121 * detected.
4122 */
4123 clean_command_ctx(&cmd_ctx);
4124 continue;
4125 }
4126
4127 health_code_update();
4128
4129 DBG("Sending response (size: %d, retcode: %s)",
4130 cmd_ctx->lttng_msg_size,
4131 lttng_strerror(-cmd_ctx->llm->ret_code));
4132 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4133 if (ret < 0) {
4134 ERR("Failed to send data back to client");
4135 }
4136
4137 /* End of transmission */
4138 ret = close(sock);
4139 if (ret) {
4140 PERROR("close");
4141 }
4142 sock = -1;
4143
4144 clean_command_ctx(&cmd_ctx);
4145
4146 health_code_update();
4147 }
4148
4149 exit:
4150 error:
4151 if (sock >= 0) {
4152 ret = close(sock);
4153 if (ret) {
4154 PERROR("close");
4155 }
4156 }
4157
4158 lttng_poll_clean(&events);
4159 clean_command_ctx(&cmd_ctx);
4160
4161 error_listen:
4162 error_create_poll:
4163 unlink(client_unix_sock_path);
4164 if (client_sock >= 0) {
4165 ret = close(client_sock);
4166 if (ret) {
4167 PERROR("close");
4168 }
4169 }
4170
4171 if (err) {
4172 health_error();
4173 ERR("Health error occurred in %s", __func__);
4174 }
4175
4176 health_unregister(health_sessiond);
4177
4178 DBG("Client thread dying");
4179
4180 rcu_unregister_thread();
4181 return NULL;
4182 }
4183
4184
4185 /*
4186 * usage function on stderr
4187 */
4188 static void usage(void)
4189 {
4190 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4191 fprintf(stderr, " -h, --help Display this usage.\n");
4192 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4193 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4194 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4195 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4196 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4197 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4198 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4199 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4200 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4201 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4202 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4203 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4204 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4205 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4206 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4207 fprintf(stderr, " -V, --version Show version number.\n");
4208 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4209 fprintf(stderr, " -q, --quiet No output at all.\n");
4210 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4211 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4212 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4213 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4214 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4215 fprintf(stderr, " -f --config Load daemon configuration file\n");
4216 fprintf(stderr, " -l --load PATH Load session configuration\n");
4217 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4218 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4219 }
4220
4221 /*
4222 * Take an option from the getopt output and set it in the right variable to be
4223 * used later.
4224 *
4225 * Return 0 on success else a negative value.
4226 */
4227 static int set_option(int opt, const char *arg, const char *optname)
4228 {
4229 int ret = 0;
4230
4231 switch (opt) {
4232 case 0:
4233 fprintf(stderr, "option %s", optname);
4234 if (arg) {
4235 fprintf(stderr, " with arg %s\n", arg);
4236 }
4237 break;
4238 case 'c':
4239 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4240 break;
4241 case 'a':
4242 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4243 break;
4244 case 'd':
4245 opt_daemon = 1;
4246 break;
4247 case 'b':
4248 opt_background = 1;
4249 break;
4250 case 'g':
4251 /*
4252 * If the override option is set, the pointer points to a
4253 * *non* const thus freeing it even though the variable type is
4254 * set to const.
4255 */
4256 if (tracing_group_name_override) {
4257 free((void *) tracing_group_name);
4258 }
4259 tracing_group_name = strdup(arg);
4260 if (!tracing_group_name) {
4261 perror("strdup");
4262 ret = -ENOMEM;
4263 }
4264 tracing_group_name_override = 1;
4265 break;
4266 case 'h':
4267 usage();
4268 exit(EXIT_FAILURE);
4269 case 'V':
4270 fprintf(stdout, "%s\n", VERSION);
4271 exit(EXIT_SUCCESS);
4272 case 'S':
4273 opt_sig_parent = 1;
4274 break;
4275 case 'E':
4276 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4277 break;
4278 case 'C':
4279 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4280 break;
4281 case 'F':
4282 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4283 break;
4284 case 'D':
4285 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4286 break;
4287 case 'H':
4288 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4289 break;
4290 case 'G':
4291 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4292 break;
4293 case 'N':
4294 opt_no_kernel = 1;
4295 break;
4296 case 'q':
4297 lttng_opt_quiet = 1;
4298 break;
4299 case 'v':
4300 /* Verbose level can increase using multiple -v */
4301 if (arg) {
4302 /* Value obtained from config file */
4303 lttng_opt_verbose = config_parse_value(arg);
4304 } else {
4305 /* -v used on command line */
4306 lttng_opt_verbose++;
4307 }
4308 /* Clamp value to [0, 3] */
4309 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4310 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4311 break;
4312 case 'Z':
4313 if (arg) {
4314 opt_verbose_consumer = config_parse_value(arg);
4315 } else {
4316 opt_verbose_consumer += 1;
4317 }
4318 break;
4319 case 'u':
4320 if (consumerd32_bin_override) {
4321 free((void *) consumerd32_bin);
4322 }
4323 consumerd32_bin = strdup(arg);
4324 if (!consumerd32_bin) {
4325 perror("strdup");
4326 ret = -ENOMEM;
4327 }
4328 consumerd32_bin_override = 1;
4329 break;
4330 case 'U':
4331 if (consumerd32_libdir_override) {
4332 free((void *) consumerd32_libdir);
4333 }
4334 consumerd32_libdir = strdup(arg);
4335 if (!consumerd32_libdir) {
4336 perror("strdup");
4337 ret = -ENOMEM;
4338 }
4339 consumerd32_libdir_override = 1;
4340 break;
4341 case 't':
4342 if (consumerd64_bin_override) {
4343 free((void *) consumerd64_bin);
4344 }
4345 consumerd64_bin = strdup(arg);
4346 if (!consumerd64_bin) {
4347 perror("strdup");
4348 ret = -ENOMEM;
4349 }
4350 consumerd64_bin_override = 1;
4351 break;
4352 case 'T':
4353 if (consumerd64_libdir_override) {
4354 free((void *) consumerd64_libdir);
4355 }
4356 consumerd64_libdir = strdup(arg);
4357 if (!consumerd64_libdir) {
4358 perror("strdup");
4359 ret = -ENOMEM;
4360 }
4361 consumerd64_libdir_override = 1;
4362 break;
4363 case 'p':
4364 free(opt_pidfile);
4365 opt_pidfile = strdup(arg);
4366 if (!opt_pidfile) {
4367 perror("strdup");
4368 ret = -ENOMEM;
4369 }
4370 break;
4371 case 'J': /* Agent TCP port. */
4372 {
4373 unsigned long v;
4374
4375 errno = 0;
4376 v = strtoul(arg, NULL, 0);
4377 if (errno != 0 || !isdigit(arg[0])) {
4378 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4379 return -1;
4380 }
4381 if (v == 0 || v >= 65535) {
4382 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4383 return -1;
4384 }
4385 agent_tcp_port = (uint32_t) v;
4386 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4387 break;
4388 }
4389 case 'l':
4390 free(opt_load_session_path);
4391 opt_load_session_path = strdup(arg);
4392 if (!opt_load_session_path) {
4393 perror("strdup");
4394 ret = -ENOMEM;
4395 }
4396 break;
4397 case 'P': /* probe modules list */
4398 free(kmod_probes_list);
4399 kmod_probes_list = strdup(arg);
4400 if (!kmod_probes_list) {
4401 perror("strdup");
4402 ret = -ENOMEM;
4403 }
4404 break;
4405 case 'e':
4406 free(kmod_extra_probes_list);
4407 kmod_extra_probes_list = strdup(arg);
4408 if (!kmod_extra_probes_list) {
4409 perror("strdup");
4410 ret = -ENOMEM;
4411 }
4412 break;
4413 case 'f':
4414 /* This is handled in set_options() thus silent break. */
4415 break;
4416 default:
4417 /* Unknown option or other error.
4418 * Error is printed by getopt, just return */
4419 ret = -1;
4420 }
4421
4422 return ret;
4423 }
4424
4425 /*
4426 * config_entry_handler_cb used to handle options read from a config file.
4427 * See config_entry_handler_cb comment in common/config/config.h for the
4428 * return value conventions.
4429 */
4430 static int config_entry_handler(const struct config_entry *entry, void *unused)
4431 {
4432 int ret = 0, i;
4433
4434 if (!entry || !entry->name || !entry->value) {
4435 ret = -EINVAL;
4436 goto end;
4437 }
4438
4439 /* Check if the option is to be ignored */
4440 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4441 if (!strcmp(entry->name, config_ignore_options[i])) {
4442 goto end;
4443 }
4444 }
4445
4446 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4447 i++) {
4448
4449 /* Ignore if not fully matched. */
4450 if (strcmp(entry->name, long_options[i].name)) {
4451 continue;
4452 }
4453
4454 /*
4455 * If the option takes no argument on the command line, we have to
4456 * check if the value is "true". We support non-zero numeric values,
4457 * true, on and yes.
4458 */
4459 if (!long_options[i].has_arg) {
4460 ret = config_parse_value(entry->value);
4461 if (ret <= 0) {
4462 if (ret) {
4463 WARN("Invalid configuration value \"%s\" for option %s",
4464 entry->value, entry->name);
4465 }
4466 /* False, skip boolean config option. */
4467 goto end;
4468 }
4469 }
4470
4471 ret = set_option(long_options[i].val, entry->value, entry->name);
4472 goto end;
4473 }
4474
4475 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4476
4477 end:
4478 return ret;
4479 }
4480
4481 /*
4482 * daemon configuration loading and argument parsing
4483 */
4484 static int set_options(int argc, char **argv)
4485 {
4486 int ret = 0, c = 0, option_index = 0;
4487 int orig_optopt = optopt, orig_optind = optind;
4488 char *optstring;
4489 const char *config_path = NULL;
4490
4491 optstring = utils_generate_optstring(long_options,
4492 sizeof(long_options) / sizeof(struct option));
4493 if (!optstring) {
4494 ret = -ENOMEM;
4495 goto end;
4496 }
4497
4498 /* Check for the --config option */
4499 while ((c = getopt_long(argc, argv, optstring, long_options,
4500 &option_index)) != -1) {
4501 if (c == '?') {
4502 ret = -EINVAL;
4503 goto end;
4504 } else if (c != 'f') {
4505 /* if not equal to --config option. */
4506 continue;
4507 }
4508
4509 config_path = utils_expand_path(optarg);
4510 if (!config_path) {
4511 ERR("Failed to resolve path: %s", optarg);
4512 }
4513 }
4514
4515 ret = config_get_section_entries(config_path, config_section_name,
4516 config_entry_handler, NULL);
4517 if (ret) {
4518 if (ret > 0) {
4519 ERR("Invalid configuration option at line %i", ret);
4520 ret = -1;
4521 }
4522 goto end;
4523 }
4524
4525 /* Reset getopt's global state */
4526 optopt = orig_optopt;
4527 optind = orig_optind;
4528 while (1) {
4529 c = getopt_long(argc, argv, optstring, long_options, &option_index);
4530 if (c == -1) {
4531 break;
4532 }
4533
4534 ret = set_option(c, optarg, long_options[option_index].name);
4535 if (ret < 0) {
4536 break;
4537 }
4538 }
4539
4540 end:
4541 free(optstring);
4542 return ret;
4543 }
4544
4545 /*
4546 * Creates the two needed socket by the daemon.
4547 * apps_sock - The communication socket for all UST apps.
4548 * client_sock - The communication of the cli tool (lttng).
4549 */
4550 static int init_daemon_socket(void)
4551 {
4552 int ret = 0;
4553 mode_t old_umask;
4554
4555 old_umask = umask(0);
4556
4557 /* Create client tool unix socket */
4558 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4559 if (client_sock < 0) {
4560 ERR("Create unix sock failed: %s", client_unix_sock_path);
4561 ret = -1;
4562 goto end;
4563 }
4564
4565 /* Set the cloexec flag */
4566 ret = utils_set_fd_cloexec(client_sock);
4567 if (ret < 0) {
4568 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4569 "Continuing but note that the consumer daemon will have a "
4570 "reference to this socket on exec()", client_sock);
4571 }
4572
4573 /* File permission MUST be 660 */
4574 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4575 if (ret < 0) {
4576 ERR("Set file permissions failed: %s", client_unix_sock_path);
4577 PERROR("chmod");
4578 goto end;
4579 }
4580
4581 /* Create the application unix socket */
4582 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4583 if (apps_sock < 0) {
4584 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4585 ret = -1;
4586 goto end;
4587 }
4588
4589 /* Set the cloexec flag */
4590 ret = utils_set_fd_cloexec(apps_sock);
4591 if (ret < 0) {
4592 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4593 "Continuing but note that the consumer daemon will have a "
4594 "reference to this socket on exec()", apps_sock);
4595 }
4596
4597 /* File permission MUST be 666 */
4598 ret = chmod(apps_unix_sock_path,
4599 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4600 if (ret < 0) {
4601 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4602 PERROR("chmod");
4603 goto end;
4604 }
4605
4606 DBG3("Session daemon client socket %d and application socket %d created",
4607 client_sock, apps_sock);
4608
4609 end:
4610 umask(old_umask);
4611 return ret;
4612 }
4613
4614 /*
4615 * Check if the global socket is available, and if a daemon is answering at the
4616 * other side. If yes, error is returned.
4617 */
4618 static int check_existing_daemon(void)
4619 {
4620 /* Is there anybody out there ? */
4621 if (lttng_session_daemon_alive()) {
4622 return -EEXIST;
4623 }
4624
4625 return 0;
4626 }
4627
4628 /*
4629 * Set the tracing group gid onto the client socket.
4630 *
4631 * Race window between mkdir and chown is OK because we are going from more
4632 * permissive (root.root) to less permissive (root.tracing).
4633 */
4634 static int set_permissions(char *rundir)
4635 {
4636 int ret;
4637 gid_t gid;
4638
4639 gid = utils_get_group_id(tracing_group_name);
4640
4641 /* Set lttng run dir */
4642 ret = chown(rundir, 0, gid);
4643 if (ret < 0) {
4644 ERR("Unable to set group on %s", rundir);
4645 PERROR("chown");
4646 }
4647
4648 /*
4649 * Ensure all applications and tracing group can search the run
4650 * dir. Allow everyone to read the directory, since it does not
4651 * buy us anything to hide its content.
4652 */
4653 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4654 if (ret < 0) {
4655 ERR("Unable to set permissions on %s", rundir);
4656 PERROR("chmod");
4657 }
4658
4659 /* lttng client socket path */
4660 ret = chown(client_unix_sock_path, 0, gid);
4661 if (ret < 0) {
4662 ERR("Unable to set group on %s", client_unix_sock_path);
4663 PERROR("chown");
4664 }
4665
4666 /* kconsumer error socket path */
4667 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4668 if (ret < 0) {
4669 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4670 PERROR("chown");
4671 }
4672
4673 /* 64-bit ustconsumer error socket path */
4674 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4675 if (ret < 0) {
4676 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4677 PERROR("chown");
4678 }
4679
4680 /* 32-bit ustconsumer compat32 error socket path */
4681 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4682 if (ret < 0) {
4683 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4684 PERROR("chown");
4685 }
4686
4687 DBG("All permissions are set");
4688
4689 return ret;
4690 }
4691
4692 /*
4693 * Create the lttng run directory needed for all global sockets and pipe.
4694 */
4695 static int create_lttng_rundir(const char *rundir)
4696 {
4697 int ret;
4698
4699 DBG3("Creating LTTng run directory: %s", rundir);
4700
4701 ret = mkdir(rundir, S_IRWXU);
4702 if (ret < 0) {
4703 if (errno != EEXIST) {
4704 ERR("Unable to create %s", rundir);
4705 goto error;
4706 } else {
4707 ret = 0;
4708 }
4709 }
4710
4711 error:
4712 return ret;
4713 }
4714
4715 /*
4716 * Setup sockets and directory needed by the kconsumerd communication with the
4717 * session daemon.
4718 */
4719 static int set_consumer_sockets(struct consumer_data *consumer_data,
4720 const char *rundir)
4721 {
4722 int ret;
4723 char path[PATH_MAX];
4724
4725 switch (consumer_data->type) {
4726 case LTTNG_CONSUMER_KERNEL:
4727 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4728 break;
4729 case LTTNG_CONSUMER64_UST:
4730 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4731 break;
4732 case LTTNG_CONSUMER32_UST:
4733 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4734 break;
4735 default:
4736 ERR("Consumer type unknown");
4737 ret = -EINVAL;
4738 goto error;
4739 }
4740
4741 DBG2("Creating consumer directory: %s", path);
4742
4743 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4744 if (ret < 0) {
4745 if (errno != EEXIST) {
4746 PERROR("mkdir");
4747 ERR("Failed to create %s", path);
4748 goto error;
4749 }
4750 ret = -1;
4751 }
4752 if (is_root) {
4753 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4754 if (ret < 0) {
4755 ERR("Unable to set group on %s", path);
4756 PERROR("chown");
4757 goto error;
4758 }
4759 }
4760
4761 /* Create the kconsumerd error unix socket */
4762 consumer_data->err_sock =
4763 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4764 if (consumer_data->err_sock < 0) {
4765 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4766 ret = -1;
4767 goto error;
4768 }
4769
4770 /*
4771 * Set the CLOEXEC flag. Return code is useless because either way, the
4772 * show must go on.
4773 */
4774 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4775 if (ret < 0) {
4776 PERROR("utils_set_fd_cloexec");
4777 /* continue anyway */
4778 }
4779
4780 /* File permission MUST be 660 */
4781 ret = chmod(consumer_data->err_unix_sock_path,
4782 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4783 if (ret < 0) {
4784 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4785 PERROR("chmod");
4786 goto error;
4787 }
4788
4789 error:
4790 return ret;
4791 }
4792
4793 /*
4794 * Signal handler for the daemon
4795 *
4796 * Simply stop all worker threads, leaving main() return gracefully after
4797 * joining all threads and calling cleanup().
4798 */
4799 static void sighandler(int sig)
4800 {
4801 switch (sig) {
4802 case SIGPIPE:
4803 DBG("SIGPIPE caught");
4804 return;
4805 case SIGINT:
4806 DBG("SIGINT caught");
4807 stop_threads();
4808 break;
4809 case SIGTERM:
4810 DBG("SIGTERM caught");
4811 stop_threads();
4812 break;
4813 case SIGUSR1:
4814 CMM_STORE_SHARED(recv_child_signal, 1);
4815 break;
4816 default:
4817 break;
4818 }
4819 }
4820
4821 /*
4822 * Setup signal handler for :
4823 * SIGINT, SIGTERM, SIGPIPE
4824 */
4825 static int set_signal_handler(void)
4826 {
4827 int ret = 0;
4828 struct sigaction sa;
4829 sigset_t sigset;
4830
4831 if ((ret = sigemptyset(&sigset)) < 0) {
4832 PERROR("sigemptyset");
4833 return ret;
4834 }
4835
4836 sa.sa_handler = sighandler;
4837 sa.sa_mask = sigset;
4838 sa.sa_flags = 0;
4839 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4840 PERROR("sigaction");
4841 return ret;
4842 }
4843
4844 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4845 PERROR("sigaction");
4846 return ret;
4847 }
4848
4849 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4850 PERROR("sigaction");
4851 return ret;
4852 }
4853
4854 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4855 PERROR("sigaction");
4856 return ret;
4857 }
4858
4859 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4860
4861 return ret;
4862 }
4863
4864 /*
4865 * Set open files limit to unlimited. This daemon can open a large number of
4866 * file descriptors in order to consumer multiple kernel traces.
4867 */
4868 static void set_ulimit(void)
4869 {
4870 int ret;
4871 struct rlimit lim;
4872
4873 /* The kernel does not allowed an infinite limit for open files */
4874 lim.rlim_cur = 65535;
4875 lim.rlim_max = 65535;
4876
4877 ret = setrlimit(RLIMIT_NOFILE, &lim);
4878 if (ret < 0) {
4879 PERROR("failed to set open files limit");
4880 }
4881 }
4882
4883 /*
4884 * Write pidfile using the rundir and opt_pidfile.
4885 */
4886 static void write_pidfile(void)
4887 {
4888 int ret;
4889 char pidfile_path[PATH_MAX];
4890
4891 assert(rundir);
4892
4893 if (opt_pidfile) {
4894 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4895 } else {
4896 /* Build pidfile path from rundir and opt_pidfile. */
4897 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4898 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4899 if (ret < 0) {
4900 PERROR("snprintf pidfile path");
4901 goto error;
4902 }
4903 }
4904
4905 /*
4906 * Create pid file in rundir. Return value is of no importance. The
4907 * execution will continue even though we are not able to write the file.
4908 */
4909 (void) utils_create_pid_file(getpid(), pidfile_path);
4910
4911 error:
4912 return;
4913 }
4914
4915 /*
4916 * Create lockfile using the rundir and return its fd.
4917 */
4918 static int create_lockfile(void)
4919 {
4920 int ret;
4921 char lockfile_path[PATH_MAX];
4922
4923 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
4924 if (ret < 0) {
4925 goto error;
4926 }
4927
4928 ret = utils_create_lock_file(lockfile_path);
4929 error:
4930 return ret;
4931 }
4932
4933 /*
4934 * Write agent TCP port using the rundir.
4935 */
4936 static void write_agent_port(void)
4937 {
4938 int ret;
4939 char path[PATH_MAX];
4940
4941 assert(rundir);
4942
4943 ret = snprintf(path, sizeof(path), "%s/"
4944 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
4945 if (ret < 0) {
4946 PERROR("snprintf agent port path");
4947 goto error;
4948 }
4949
4950 /*
4951 * Create TCP agent port file in rundir. Return value is of no importance.
4952 * The execution will continue even though we are not able to write the
4953 * file.
4954 */
4955 (void) utils_create_pid_file(agent_tcp_port, path);
4956
4957 error:
4958 return;
4959 }
4960
4961 /*
4962 * Start the load session thread and dettach from it so the main thread can
4963 * continue. This does not return a value since whatever the outcome, the main
4964 * thread will continue.
4965 */
4966 static void start_load_session_thread(void)
4967 {
4968 int ret;
4969
4970 /* Create session loading thread. */
4971 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
4972 load_info);
4973 if (ret != 0) {
4974 PERROR("pthread_create load_session_thread");
4975 goto error_create;
4976 }
4977
4978 ret = pthread_detach(load_session_thread);
4979 if (ret != 0) {
4980 PERROR("pthread_detach load_session_thread");
4981 }
4982
4983 /* Everything went well so don't cleanup anything. */
4984
4985 error_create:
4986 /* The cleanup() function will destroy the load_info data. */
4987 return;
4988 }
4989
4990 /*
4991 * main
4992 */
4993 int main(int argc, char **argv)
4994 {
4995 int ret = 0;
4996 void *status;
4997 const char *home_path, *env_app_timeout;
4998
4999 init_kernel_workarounds();
5000
5001 rcu_register_thread();
5002
5003 if ((ret = set_signal_handler()) < 0) {
5004 goto error;
5005 }
5006
5007 setup_consumerd_path();
5008
5009 page_size = sysconf(_SC_PAGESIZE);
5010 if (page_size < 0) {
5011 PERROR("sysconf _SC_PAGESIZE");
5012 page_size = LONG_MAX;
5013 WARN("Fallback page size to %ld", page_size);
5014 }
5015
5016 /* Parse arguments and load the daemon configuration file */
5017 progname = argv[0];
5018 if ((ret = set_options(argc, argv)) < 0) {
5019 goto error;
5020 }
5021
5022 /* Daemonize */
5023 if (opt_daemon || opt_background) {
5024 int i;
5025
5026 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5027 !opt_background);
5028 if (ret < 0) {
5029 goto error;
5030 }
5031
5032 /*
5033 * We are in the child. Make sure all other file descriptors are
5034 * closed, in case we are called with more opened file descriptors than
5035 * the standard ones.
5036 */
5037 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5038 (void) close(i);
5039 }
5040 }
5041
5042 /* Create thread quit pipe */
5043 if ((ret = init_thread_quit_pipe()) < 0) {
5044 goto error;
5045 }
5046
5047 /* Check if daemon is UID = 0 */
5048 is_root = !getuid();
5049
5050 if (is_root) {
5051 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5052
5053 /* Create global run dir with root access */
5054 ret = create_lttng_rundir(rundir);
5055 if (ret < 0) {
5056 goto error;
5057 }
5058
5059 if (strlen(apps_unix_sock_path) == 0) {
5060 snprintf(apps_unix_sock_path, PATH_MAX,
5061 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5062 }
5063
5064 if (strlen(client_unix_sock_path) == 0) {
5065 snprintf(client_unix_sock_path, PATH_MAX,
5066 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5067 }
5068
5069 /* Set global SHM for ust */
5070 if (strlen(wait_shm_path) == 0) {
5071 snprintf(wait_shm_path, PATH_MAX,
5072 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5073 }
5074
5075 if (strlen(health_unix_sock_path) == 0) {
5076 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
5077 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5078 }
5079
5080 /* Setup kernel consumerd path */
5081 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5082 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5083 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5084 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5085
5086 DBG2("Kernel consumer err path: %s",
5087 kconsumer_data.err_unix_sock_path);
5088 DBG2("Kernel consumer cmd path: %s",
5089 kconsumer_data.cmd_unix_sock_path);
5090 } else {
5091 home_path = utils_get_home_dir();
5092 if (home_path == NULL) {
5093 /* TODO: Add --socket PATH option */
5094 ERR("Can't get HOME directory for sockets creation.");
5095 ret = -EPERM;
5096 goto error;
5097 }
5098
5099 /*
5100 * Create rundir from home path. This will create something like
5101 * $HOME/.lttng
5102 */
5103 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5104 if (ret < 0) {
5105 ret = -ENOMEM;
5106 goto error;
5107 }
5108
5109 ret = create_lttng_rundir(rundir);
5110 if (ret < 0) {
5111 goto error;
5112 }
5113
5114 if (strlen(apps_unix_sock_path) == 0) {
5115 snprintf(apps_unix_sock_path, PATH_MAX,
5116 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
5117 }
5118
5119 /* Set the cli tool unix socket path */
5120 if (strlen(client_unix_sock_path) == 0) {
5121 snprintf(client_unix_sock_path, PATH_MAX,
5122 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
5123 }
5124
5125 /* Set global SHM for ust */
5126 if (strlen(wait_shm_path) == 0) {
5127 snprintf(wait_shm_path, PATH_MAX,
5128 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
5129 }
5130
5131 /* Set health check Unix path */
5132 if (strlen(health_unix_sock_path) == 0) {
5133 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
5134 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
5135 }
5136 }
5137
5138 lockfile_fd = create_lockfile();
5139 if (lockfile_fd < 0) {
5140 goto error;
5141 }
5142
5143 /* Set consumer initial state */
5144 kernel_consumerd_state = CONSUMER_STOPPED;
5145 ust_consumerd_state = CONSUMER_STOPPED;
5146
5147 DBG("Client socket path %s", client_unix_sock_path);
5148 DBG("Application socket path %s", apps_unix_sock_path);
5149 DBG("Application wait path %s", wait_shm_path);
5150 DBG("LTTng run directory path: %s", rundir);
5151
5152 /* 32 bits consumerd path setup */
5153 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5154 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5155 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5156 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5157
5158 DBG2("UST consumer 32 bits err path: %s",
5159 ustconsumer32_data.err_unix_sock_path);
5160 DBG2("UST consumer 32 bits cmd path: %s",
5161 ustconsumer32_data.cmd_unix_sock_path);
5162
5163 /* 64 bits consumerd path setup */
5164 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5165 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5166 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5167 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5168
5169 DBG2("UST consumer 64 bits err path: %s",
5170 ustconsumer64_data.err_unix_sock_path);
5171 DBG2("UST consumer 64 bits cmd path: %s",
5172 ustconsumer64_data.cmd_unix_sock_path);
5173
5174 /*
5175 * See if daemon already exist.
5176 */
5177 if ((ret = check_existing_daemon()) < 0) {
5178 ERR("Already running daemon.\n");
5179 /*
5180 * We do not goto exit because we must not cleanup()
5181 * because a daemon is already running.
5182 */
5183 goto error;
5184 }
5185
5186 /*
5187 * Init UST app hash table. Alloc hash table before this point since
5188 * cleanup() can get called after that point.
5189 */
5190 ust_app_ht_alloc();
5191
5192 /* Initialize agent domain subsystem. */
5193 if ((ret = agent_setup()) < 0) {
5194 /* ENOMEM at this point. */
5195 goto error;
5196 }
5197
5198 /* After this point, we can safely call cleanup() with "goto exit" */
5199
5200 /*
5201 * These actions must be executed as root. We do that *after* setting up
5202 * the sockets path because we MUST make the check for another daemon using
5203 * those paths *before* trying to set the kernel consumer sockets and init
5204 * kernel tracer.
5205 */
5206 if (is_root) {
5207 ret = set_consumer_sockets(&kconsumer_data, rundir);
5208 if (ret < 0) {
5209 goto exit;
5210 }
5211
5212 /* Setup kernel tracer */
5213 if (!opt_no_kernel) {
5214 init_kernel_tracer();
5215 }
5216
5217 /* Set ulimit for open files */
5218 set_ulimit();
5219 }
5220 /* init lttng_fd tracking must be done after set_ulimit. */
5221 lttng_fd_init();
5222
5223 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
5224 if (ret < 0) {
5225 goto exit;
5226 }
5227
5228 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
5229 if (ret < 0) {
5230 goto exit;
5231 }
5232
5233 /* Setup the needed unix socket */
5234 if ((ret = init_daemon_socket()) < 0) {
5235 goto exit;
5236 }
5237
5238 /* Set credentials to socket */
5239 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
5240 goto exit;
5241 }
5242
5243 /* Get parent pid if -S, --sig-parent is specified. */
5244 if (opt_sig_parent) {
5245 ppid = getppid();
5246 }
5247
5248 /* Setup the kernel pipe for waking up the kernel thread */
5249 if (is_root && !opt_no_kernel) {
5250 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
5251 goto exit;
5252 }
5253 }
5254
5255 /* Setup the thread ht_cleanup communication pipe. */
5256 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
5257 goto exit;
5258 }
5259
5260 /* Setup the thread apps communication pipe. */
5261 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
5262 goto exit;
5263 }
5264
5265 /* Setup the thread apps notify communication pipe. */
5266 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
5267 goto exit;
5268 }
5269
5270 /* Initialize global buffer per UID and PID registry. */
5271 buffer_reg_init_uid_registry();
5272 buffer_reg_init_pid_registry();
5273
5274 /* Init UST command queue. */
5275 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5276
5277 /*
5278 * Get session list pointer. This pointer MUST NOT be free(). This list is
5279 * statically declared in session.c
5280 */
5281 session_list_ptr = session_get_list();
5282
5283 /* Set up max poll set size */
5284 lttng_poll_set_max_size();
5285
5286 cmd_init();
5287
5288 /* Check for the application socket timeout env variable. */
5289 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5290 if (env_app_timeout) {
5291 app_socket_timeout = atoi(env_app_timeout);
5292 } else {
5293 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5294 }
5295
5296 write_pidfile();
5297 write_agent_port();
5298
5299 /* Initialize communication library */
5300 lttcomm_init();
5301 /* This is to get the TCP timeout value. */
5302 lttcomm_inet_init();
5303
5304 if (load_session_init_data(&load_info) < 0) {
5305 goto exit;
5306 }
5307 load_info->path = opt_load_session_path;
5308
5309 /*
5310 * Initialize the health check subsystem. This call should set the
5311 * appropriate time values.
5312 */
5313 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5314 if (!health_sessiond) {
5315 PERROR("health_app_create error");
5316 goto exit_health_sessiond_cleanup;
5317 }
5318
5319 /* Create thread to clean up RCU hash tables */
5320 ret = pthread_create(&ht_cleanup_thread, NULL,
5321 thread_ht_cleanup, (void *) NULL);
5322 if (ret != 0) {
5323 PERROR("pthread_create ht_cleanup");
5324 goto exit_ht_cleanup;
5325 }
5326
5327 /* Create health-check thread */
5328 ret = pthread_create(&health_thread, NULL,
5329 thread_manage_health, (void *) NULL);
5330 if (ret != 0) {
5331 PERROR("pthread_create health");
5332 goto exit_health;
5333 }
5334
5335 /* Create thread to manage the client socket */
5336 ret = pthread_create(&client_thread, NULL,
5337 thread_manage_clients, (void *) NULL);
5338 if (ret != 0) {
5339 PERROR("pthread_create clients");
5340 goto exit_client;
5341 }
5342
5343 /* Create thread to dispatch registration */
5344 ret = pthread_create(&dispatch_thread, NULL,
5345 thread_dispatch_ust_registration, (void *) NULL);
5346 if (ret != 0) {
5347 PERROR("pthread_create dispatch");
5348 goto exit_dispatch;
5349 }
5350
5351 /* Create thread to manage application registration. */
5352 ret = pthread_create(&reg_apps_thread, NULL,
5353 thread_registration_apps, (void *) NULL);
5354 if (ret != 0) {
5355 PERROR("pthread_create registration");
5356 goto exit_reg_apps;
5357 }
5358
5359 /* Create thread to manage application socket */
5360 ret = pthread_create(&apps_thread, NULL,
5361 thread_manage_apps, (void *) NULL);
5362 if (ret != 0) {
5363 PERROR("pthread_create apps");
5364 goto exit_apps;
5365 }
5366
5367 /* Create thread to manage application notify socket */
5368 ret = pthread_create(&apps_notify_thread, NULL,
5369 ust_thread_manage_notify, (void *) NULL);
5370 if (ret != 0) {
5371 PERROR("pthread_create notify");
5372 goto exit_apps_notify;
5373 }
5374
5375 /* Create agent registration thread. */
5376 ret = pthread_create(&agent_reg_thread, NULL,
5377 agent_thread_manage_registration, (void *) NULL);
5378 if (ret != 0) {
5379 PERROR("pthread_create agent");
5380 goto exit_agent_reg;
5381 }
5382
5383 /* Don't start this thread if kernel tracing is not requested nor root */
5384 if (is_root && !opt_no_kernel) {
5385 /* Create kernel thread to manage kernel event */
5386 ret = pthread_create(&kernel_thread, NULL,
5387 thread_manage_kernel, (void *) NULL);
5388 if (ret != 0) {
5389 PERROR("pthread_create kernel");
5390 goto exit_kernel;
5391 }
5392 }
5393
5394 /* Load possible session(s). */
5395 start_load_session_thread();
5396
5397 if (is_root && !opt_no_kernel) {
5398 ret = pthread_join(kernel_thread, &status);
5399 if (ret != 0) {
5400 PERROR("pthread_join");
5401 goto error; /* join error, exit without cleanup */
5402 }
5403 }
5404
5405 exit_kernel:
5406 ret = pthread_join(agent_reg_thread, &status);
5407 if (ret != 0) {
5408 PERROR("pthread_join agent");
5409 goto error; /* join error, exit without cleanup */
5410 }
5411
5412 exit_agent_reg:
5413 ret = pthread_join(apps_notify_thread, &status);
5414 if (ret != 0) {
5415 PERROR("pthread_join apps notify");
5416 goto error; /* join error, exit without cleanup */
5417 }
5418
5419 exit_apps_notify:
5420 ret = pthread_join(apps_thread, &status);
5421 if (ret != 0) {
5422 PERROR("pthread_join apps");
5423 goto error; /* join error, exit without cleanup */
5424 }
5425
5426
5427 exit_apps:
5428 ret = pthread_join(reg_apps_thread, &status);
5429 if (ret != 0) {
5430 PERROR("pthread_join");
5431 goto error; /* join error, exit without cleanup */
5432 }
5433
5434 exit_reg_apps:
5435 ret = pthread_join(dispatch_thread, &status);
5436 if (ret != 0) {
5437 PERROR("pthread_join");
5438 goto error; /* join error, exit without cleanup */
5439 }
5440
5441 exit_dispatch:
5442 ret = pthread_join(client_thread, &status);
5443 if (ret != 0) {
5444 PERROR("pthread_join");
5445 goto error; /* join error, exit without cleanup */
5446 }
5447
5448 ret = join_consumer_thread(&kconsumer_data);
5449 if (ret != 0) {
5450 PERROR("join_consumer");
5451 goto error; /* join error, exit without cleanup */
5452 }
5453
5454 ret = join_consumer_thread(&ustconsumer32_data);
5455 if (ret != 0) {
5456 PERROR("join_consumer ust32");
5457 goto error; /* join error, exit without cleanup */
5458 }
5459
5460 ret = join_consumer_thread(&ustconsumer64_data);
5461 if (ret != 0) {
5462 PERROR("join_consumer ust64");
5463 goto error; /* join error, exit without cleanup */
5464 }
5465
5466 exit_client:
5467 ret = pthread_join(health_thread, &status);
5468 if (ret != 0) {
5469 PERROR("pthread_join health thread");
5470 goto error; /* join error, exit without cleanup */
5471 }
5472
5473 exit_health:
5474 ret = pthread_join(ht_cleanup_thread, &status);
5475 if (ret != 0) {
5476 PERROR("pthread_join ht cleanup thread");
5477 goto error; /* join error, exit without cleanup */
5478 }
5479 exit_ht_cleanup:
5480 health_app_destroy(health_sessiond);
5481 exit_health_sessiond_cleanup:
5482 exit:
5483 /*
5484 * cleanup() is called when no other thread is running.
5485 */
5486 rcu_thread_online();
5487 cleanup();
5488 rcu_thread_offline();
5489 rcu_unregister_thread();
5490 if (!ret) {
5491 exit(EXIT_SUCCESS);
5492 }
5493 error:
5494 exit(EXIT_FAILURE);
5495 }
This page took 0.198778 seconds and 4 git commands to generate.