Fix: Handle empty daemon configuration file lines
[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 #include "syscall.h"
74
75 #define CONSUMERD_FILE "lttng-consumerd"
76
77 const char *progname;
78 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
79 static int tracing_group_name_override;
80 static char *opt_pidfile;
81 static int opt_sig_parent;
82 static int opt_verbose_consumer;
83 static int opt_daemon, opt_background;
84 static int opt_no_kernel;
85 static char *opt_load_session_path;
86 static pid_t ppid; /* Parent PID for --sig-parent option */
87 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
88 static char *rundir;
89 static int lockfile_fd = -1;
90
91 /* Set to 1 when a SIGUSR1 signal is received. */
92 static int recv_child_signal;
93
94 /*
95 * Consumer daemon specific control data. Every value not initialized here is
96 * set to 0 by the static definition.
97 */
98 static struct consumer_data kconsumer_data = {
99 .type = LTTNG_CONSUMER_KERNEL,
100 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer64_data = {
110 .type = LTTNG_CONSUMER64_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120 static struct consumer_data ustconsumer32_data = {
121 .type = LTTNG_CONSUMER32_UST,
122 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
123 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
124 .err_sock = -1,
125 .cmd_sock = -1,
126 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
127 .lock = PTHREAD_MUTEX_INITIALIZER,
128 .cond = PTHREAD_COND_INITIALIZER,
129 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
130 };
131
132 /* Command line options */
133 static const struct option long_options[] = {
134 { "client-sock", 1, 0, 'c' },
135 { "apps-sock", 1, 0, 'a' },
136 { "kconsumerd-cmd-sock", 1, 0, 'C' },
137 { "kconsumerd-err-sock", 1, 0, 'E' },
138 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
139 { "ustconsumerd32-err-sock", 1, 0, 'H' },
140 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
141 { "ustconsumerd64-err-sock", 1, 0, 'F' },
142 { "consumerd32-path", 1, 0, 'u' },
143 { "consumerd32-libdir", 1, 0, 'U' },
144 { "consumerd64-path", 1, 0, 't' },
145 { "consumerd64-libdir", 1, 0, 'T' },
146 { "daemonize", 0, 0, 'd' },
147 { "background", 0, 0, 'b' },
148 { "sig-parent", 0, 0, 'S' },
149 { "help", 0, 0, 'h' },
150 { "group", 1, 0, 'g' },
151 { "version", 0, 0, 'V' },
152 { "quiet", 0, 0, 'q' },
153 { "verbose", 0, 0, 'v' },
154 { "verbose-consumer", 0, 0, 'Z' },
155 { "no-kernel", 0, 0, 'N' },
156 { "pidfile", 1, 0, 'p' },
157 { "agent-tcp-port", 1, 0, 'J' },
158 { "config", 1, 0, 'f' },
159 { "load", 1, 0, 'l' },
160 { "kmod-probes", 1, 0, 'P' },
161 { "extra-kmod-probes", 1, 0, 'e' },
162 { NULL, 0, 0, 0 }
163 };
164
165 /* Command line options to ignore from configuration file */
166 static const char *config_ignore_options[] = { "help", "version", "config" };
167
168 /* Shared between threads */
169 static int dispatch_thread_exit;
170
171 /* Global application Unix socket path */
172 static char apps_unix_sock_path[PATH_MAX];
173 /* Global client Unix socket path */
174 static char client_unix_sock_path[PATH_MAX];
175 /* global wait shm path for UST */
176 static char wait_shm_path[PATH_MAX];
177 /* Global health check unix path */
178 static char health_unix_sock_path[PATH_MAX];
179
180 /* Sockets and FDs */
181 static int client_sock = -1;
182 static int apps_sock = -1;
183 int kernel_tracer_fd = -1;
184 static int kernel_poll_pipe[2] = { -1, -1 };
185
186 /*
187 * Quit pipe for all threads. This permits a single cancellation point
188 * for all threads when receiving an event on the pipe.
189 */
190 static int thread_quit_pipe[2] = { -1, -1 };
191
192 /*
193 * This pipe is used to inform the thread managing application communication
194 * that a command is queued and ready to be processed.
195 */
196 static int apps_cmd_pipe[2] = { -1, -1 };
197
198 int apps_cmd_notify_pipe[2] = { -1, -1 };
199
200 /* Pthread, Mutexes and Semaphores */
201 static pthread_t apps_thread;
202 static pthread_t apps_notify_thread;
203 static pthread_t reg_apps_thread;
204 static pthread_t client_thread;
205 static pthread_t kernel_thread;
206 static pthread_t dispatch_thread;
207 static pthread_t health_thread;
208 static pthread_t ht_cleanup_thread;
209 static pthread_t agent_reg_thread;
210 static pthread_t load_session_thread;
211
212 /*
213 * UST registration command queue. This queue is tied with a futex and uses a N
214 * wakers / 1 waiter implemented and detailed in futex.c/.h
215 *
216 * The thread_registration_apps and thread_dispatch_ust_registration uses this
217 * queue along with the wait/wake scheme. The thread_manage_apps receives down
218 * the line new application socket and monitors it for any I/O error or clean
219 * close that triggers an unregistration of the application.
220 */
221 static struct ust_cmd_queue ust_cmd_queue;
222
223 /*
224 * Pointer initialized before thread creation.
225 *
226 * This points to the tracing session list containing the session count and a
227 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
228 * MUST NOT be taken if you call a public function in session.c.
229 *
230 * The lock is nested inside the structure: session_list_ptr->lock. Please use
231 * session_lock_list and session_unlock_list for lock acquisition.
232 */
233 static struct ltt_session_list *session_list_ptr;
234
235 int ust_consumerd64_fd = -1;
236 int ust_consumerd32_fd = -1;
237
238 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
239 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
240 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
241 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
242 static int consumerd32_bin_override;
243 static int consumerd64_bin_override;
244 static int consumerd32_libdir_override;
245 static int consumerd64_libdir_override;
246
247 static const char *module_proc_lttng = "/proc/lttng";
248
249 /*
250 * Consumer daemon state which is changed when spawning it, killing it or in
251 * case of a fatal error.
252 */
253 enum consumerd_state {
254 CONSUMER_STARTED = 1,
255 CONSUMER_STOPPED = 2,
256 CONSUMER_ERROR = 3,
257 };
258
259 /*
260 * This consumer daemon state is used to validate if a client command will be
261 * able to reach the consumer. If not, the client is informed. For instance,
262 * doing a "lttng start" when the consumer state is set to ERROR will return an
263 * error to the client.
264 *
265 * The following example shows a possible race condition of this scheme:
266 *
267 * consumer thread error happens
268 * client cmd arrives
269 * client cmd checks state -> still OK
270 * consumer thread exit, sets error
271 * client cmd try to talk to consumer
272 * ...
273 *
274 * However, since the consumer is a different daemon, we have no way of making
275 * sure the command will reach it safely even with this state flag. This is why
276 * we consider that up to the state validation during command processing, the
277 * command is safe. After that, we can not guarantee the correctness of the
278 * client request vis-a-vis the consumer.
279 */
280 static enum consumerd_state ust_consumerd_state;
281 static enum consumerd_state kernel_consumerd_state;
282
283 /*
284 * Socket timeout for receiving and sending in seconds.
285 */
286 static int app_socket_timeout;
287
288 /* Set in main() with the current page size. */
289 long page_size;
290
291 /* Application health monitoring */
292 struct health_app *health_sessiond;
293
294 /* Agent TCP port for registration. Used by the agent thread. */
295 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
296
297 /* Am I root or not. */
298 int is_root; /* Set to 1 if the daemon is running as root */
299
300 const char * const config_section_name = "sessiond";
301
302 /* Load session thread information to operate. */
303 struct load_session_thread_data *load_info;
304
305 /*
306 * Whether sessiond is ready for commands/health check requests.
307 * NR_LTTNG_SESSIOND_READY must match the number of calls to
308 * sessiond_notify_ready().
309 */
310 #define NR_LTTNG_SESSIOND_READY 3
311 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
312
313 /* Notify parents that we are ready for cmd and health check */
314 LTTNG_HIDDEN
315 void sessiond_notify_ready(void)
316 {
317 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
318 /*
319 * Notify parent pid that we are ready to accept command
320 * for client side. This ppid is the one from the
321 * external process that spawned us.
322 */
323 if (opt_sig_parent) {
324 kill(ppid, SIGUSR1);
325 }
326
327 /*
328 * Notify the parent of the fork() process that we are
329 * ready.
330 */
331 if (opt_daemon || opt_background) {
332 kill(child_ppid, SIGUSR1);
333 }
334 }
335 }
336
337 static
338 void setup_consumerd_path(void)
339 {
340 const char *bin, *libdir;
341
342 /*
343 * Allow INSTALL_BIN_PATH to be used as a target path for the
344 * native architecture size consumer if CONFIG_CONSUMER*_PATH
345 * has not been defined.
346 */
347 #if (CAA_BITS_PER_LONG == 32)
348 if (!consumerd32_bin[0]) {
349 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
350 }
351 if (!consumerd32_libdir[0]) {
352 consumerd32_libdir = INSTALL_LIB_PATH;
353 }
354 #elif (CAA_BITS_PER_LONG == 64)
355 if (!consumerd64_bin[0]) {
356 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
357 }
358 if (!consumerd64_libdir[0]) {
359 consumerd64_libdir = INSTALL_LIB_PATH;
360 }
361 #else
362 #error "Unknown bitness"
363 #endif
364
365 /*
366 * runtime env. var. overrides the build default.
367 */
368 bin = getenv("LTTNG_CONSUMERD32_BIN");
369 if (bin) {
370 consumerd32_bin = bin;
371 }
372 bin = getenv("LTTNG_CONSUMERD64_BIN");
373 if (bin) {
374 consumerd64_bin = bin;
375 }
376 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
377 if (libdir) {
378 consumerd32_libdir = libdir;
379 }
380 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
381 if (libdir) {
382 consumerd64_libdir = libdir;
383 }
384 }
385
386 /*
387 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
388 */
389 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
390 {
391 int ret;
392
393 assert(events);
394
395 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
396 if (ret < 0) {
397 goto error;
398 }
399
400 /* Add quit pipe */
401 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
402 if (ret < 0) {
403 goto error;
404 }
405
406 return 0;
407
408 error:
409 return ret;
410 }
411
412 /*
413 * Check if the thread quit pipe was triggered.
414 *
415 * Return 1 if it was triggered else 0;
416 */
417 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
418 {
419 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
420 return 1;
421 }
422
423 return 0;
424 }
425
426 /*
427 * Init thread quit pipe.
428 *
429 * Return -1 on error or 0 if all pipes are created.
430 */
431 static int init_thread_quit_pipe(void)
432 {
433 int ret, i;
434
435 ret = pipe(thread_quit_pipe);
436 if (ret < 0) {
437 PERROR("thread quit pipe");
438 goto error;
439 }
440
441 for (i = 0; i < 2; i++) {
442 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
443 if (ret < 0) {
444 PERROR("fcntl");
445 goto error;
446 }
447 }
448
449 error:
450 return ret;
451 }
452
453 /*
454 * Stop all threads by closing the thread quit pipe.
455 */
456 static void stop_threads(void)
457 {
458 int ret;
459
460 /* Stopping all threads */
461 DBG("Terminating all threads");
462 ret = notify_thread_pipe(thread_quit_pipe[1]);
463 if (ret < 0) {
464 ERR("write error on thread quit pipe");
465 }
466
467 /* Dispatch thread */
468 CMM_STORE_SHARED(dispatch_thread_exit, 1);
469 futex_nto1_wake(&ust_cmd_queue.futex);
470 }
471
472 /*
473 * Close every consumer sockets.
474 */
475 static void close_consumer_sockets(void)
476 {
477 int ret;
478
479 if (kconsumer_data.err_sock >= 0) {
480 ret = close(kconsumer_data.err_sock);
481 if (ret < 0) {
482 PERROR("kernel consumer err_sock close");
483 }
484 }
485 if (ustconsumer32_data.err_sock >= 0) {
486 ret = close(ustconsumer32_data.err_sock);
487 if (ret < 0) {
488 PERROR("UST consumerd32 err_sock close");
489 }
490 }
491 if (ustconsumer64_data.err_sock >= 0) {
492 ret = close(ustconsumer64_data.err_sock);
493 if (ret < 0) {
494 PERROR("UST consumerd64 err_sock close");
495 }
496 }
497 if (kconsumer_data.cmd_sock >= 0) {
498 ret = close(kconsumer_data.cmd_sock);
499 if (ret < 0) {
500 PERROR("kernel consumer cmd_sock close");
501 }
502 }
503 if (ustconsumer32_data.cmd_sock >= 0) {
504 ret = close(ustconsumer32_data.cmd_sock);
505 if (ret < 0) {
506 PERROR("UST consumerd32 cmd_sock close");
507 }
508 }
509 if (ustconsumer64_data.cmd_sock >= 0) {
510 ret = close(ustconsumer64_data.cmd_sock);
511 if (ret < 0) {
512 PERROR("UST consumerd64 cmd_sock close");
513 }
514 }
515 }
516
517 /*
518 * Generate the full lock file path using the rundir.
519 *
520 * Return the snprintf() return value thus a negative value is an error.
521 */
522 static int generate_lock_file_path(char *path, size_t len)
523 {
524 int ret;
525
526 assert(path);
527 assert(rundir);
528
529 /* Build lockfile path from rundir. */
530 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
531 if (ret < 0) {
532 PERROR("snprintf lockfile path");
533 }
534
535 return ret;
536 }
537
538 /*
539 * Cleanup the daemon
540 */
541 static void cleanup(void)
542 {
543 int ret;
544 struct ltt_session *sess, *stmp;
545 char path[PATH_MAX];
546
547 DBG("Cleaning up");
548
549 /*
550 * Close the thread quit pipe. It has already done its job,
551 * since we are now called.
552 */
553 utils_close_pipe(thread_quit_pipe);
554
555 /*
556 * If opt_pidfile is undefined, the default file will be wiped when
557 * removing the rundir.
558 */
559 if (opt_pidfile) {
560 ret = remove(opt_pidfile);
561 if (ret < 0) {
562 PERROR("remove pidfile %s", opt_pidfile);
563 }
564 }
565
566 DBG("Removing sessiond and consumerd content of directory %s", rundir);
567
568 /* sessiond */
569 snprintf(path, PATH_MAX,
570 "%s/%s",
571 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
572 DBG("Removing %s", path);
573 (void) unlink(path);
574
575 snprintf(path, PATH_MAX, "%s/%s", rundir,
576 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
577 DBG("Removing %s", path);
578 (void) unlink(path);
579
580 /* kconsumerd */
581 snprintf(path, PATH_MAX,
582 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
583 rundir);
584 DBG("Removing %s", path);
585 (void) unlink(path);
586
587 snprintf(path, PATH_MAX,
588 DEFAULT_KCONSUMERD_PATH,
589 rundir);
590 DBG("Removing directory %s", path);
591 (void) rmdir(path);
592
593 /* ust consumerd 32 */
594 snprintf(path, PATH_MAX,
595 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
596 rundir);
597 DBG("Removing %s", path);
598 (void) unlink(path);
599
600 snprintf(path, PATH_MAX,
601 DEFAULT_USTCONSUMERD32_PATH,
602 rundir);
603 DBG("Removing directory %s", path);
604 (void) rmdir(path);
605
606 /* ust consumerd 64 */
607 snprintf(path, PATH_MAX,
608 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
609 rundir);
610 DBG("Removing %s", path);
611 (void) unlink(path);
612
613 snprintf(path, PATH_MAX,
614 DEFAULT_USTCONSUMERD64_PATH,
615 rundir);
616 DBG("Removing directory %s", path);
617 (void) rmdir(path);
618
619 DBG("Cleaning up all sessions");
620
621 /* Destroy session list mutex */
622 if (session_list_ptr != NULL) {
623 pthread_mutex_destroy(&session_list_ptr->lock);
624
625 /* Cleanup ALL session */
626 cds_list_for_each_entry_safe(sess, stmp,
627 &session_list_ptr->head, list) {
628 cmd_destroy_session(sess, kernel_poll_pipe[1]);
629 }
630 }
631
632 DBG("Closing all UST sockets");
633 ust_app_clean_list();
634 buffer_reg_destroy_registries();
635
636 if (is_root && !opt_no_kernel) {
637 DBG2("Closing kernel fd");
638 if (kernel_tracer_fd >= 0) {
639 ret = close(kernel_tracer_fd);
640 if (ret) {
641 PERROR("close");
642 }
643 }
644 DBG("Unloading kernel modules");
645 modprobe_remove_lttng_all();
646 free(syscall_table);
647 }
648
649 close_consumer_sockets();
650
651 /*
652 * If the override option is set, the pointer points to a *non* const thus
653 * freeing it even though the variable type is set to const.
654 */
655 if (tracing_group_name_override) {
656 free((void *) tracing_group_name);
657 }
658 if (consumerd32_bin_override) {
659 free((void *) consumerd32_bin);
660 }
661 if (consumerd64_bin_override) {
662 free((void *) consumerd64_bin);
663 }
664 if (consumerd32_libdir_override) {
665 free((void *) consumerd32_libdir);
666 }
667 if (consumerd64_libdir_override) {
668 free((void *) consumerd64_libdir);
669 }
670
671 if (opt_pidfile) {
672 free(opt_pidfile);
673 }
674
675 if (opt_load_session_path) {
676 free(opt_load_session_path);
677 }
678
679 if (load_info) {
680 load_session_destroy_data(load_info);
681 free(load_info);
682 }
683
684 /*
685 * Cleanup lock file by deleting it and finaly closing it which will
686 * release the file system lock.
687 */
688 if (lockfile_fd >= 0) {
689 char lockfile_path[PATH_MAX];
690
691 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
692 if (ret > 0) {
693 ret = remove(lockfile_path);
694 if (ret < 0) {
695 PERROR("remove lock file");
696 }
697 ret = close(lockfile_fd);
698 if (ret < 0) {
699 PERROR("close lock file");
700 }
701 }
702 }
703
704 /*
705 * We do NOT rmdir rundir because there are other processes
706 * using it, for instance lttng-relayd, which can start in
707 * parallel with this teardown.
708 */
709
710 free(rundir);
711
712 /* <fun> */
713 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
714 "Matthew, BEET driven development works!%c[%dm",
715 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
716 /* </fun> */
717 }
718
719 /*
720 * Send data on a unix socket using the liblttsessiondcomm API.
721 *
722 * Return lttcomm error code.
723 */
724 static int send_unix_sock(int sock, void *buf, size_t len)
725 {
726 /* Check valid length */
727 if (len == 0) {
728 return -1;
729 }
730
731 return lttcomm_send_unix_sock(sock, buf, len);
732 }
733
734 /*
735 * Free memory of a command context structure.
736 */
737 static void clean_command_ctx(struct command_ctx **cmd_ctx)
738 {
739 DBG("Clean command context structure");
740 if (*cmd_ctx) {
741 if ((*cmd_ctx)->llm) {
742 free((*cmd_ctx)->llm);
743 }
744 if ((*cmd_ctx)->lsm) {
745 free((*cmd_ctx)->lsm);
746 }
747 free(*cmd_ctx);
748 *cmd_ctx = NULL;
749 }
750 }
751
752 /*
753 * Notify UST applications using the shm mmap futex.
754 */
755 static int notify_ust_apps(int active)
756 {
757 char *wait_shm_mmap;
758
759 DBG("Notifying applications of session daemon state: %d", active);
760
761 /* See shm.c for this call implying mmap, shm and futex calls */
762 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
763 if (wait_shm_mmap == NULL) {
764 goto error;
765 }
766
767 /* Wake waiting process */
768 futex_wait_update((int32_t *) wait_shm_mmap, active);
769
770 /* Apps notified successfully */
771 return 0;
772
773 error:
774 return -1;
775 }
776
777 /*
778 * Setup the outgoing data buffer for the response (llm) by allocating the
779 * right amount of memory and copying the original information from the lsm
780 * structure.
781 *
782 * Return total size of the buffer pointed by buf.
783 */
784 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
785 {
786 int ret, buf_size;
787
788 buf_size = size;
789
790 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
791 if (cmd_ctx->llm == NULL) {
792 PERROR("zmalloc");
793 ret = -ENOMEM;
794 goto error;
795 }
796
797 /* Copy common data */
798 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
799 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
800
801 cmd_ctx->llm->data_size = size;
802 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
803
804 return buf_size;
805
806 error:
807 return ret;
808 }
809
810 /*
811 * Update the kernel poll set of all channel fd available over all tracing
812 * session. Add the wakeup pipe at the end of the set.
813 */
814 static int update_kernel_poll(struct lttng_poll_event *events)
815 {
816 int ret;
817 struct ltt_session *session;
818 struct ltt_kernel_channel *channel;
819
820 DBG("Updating kernel poll set");
821
822 session_lock_list();
823 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
824 session_lock(session);
825 if (session->kernel_session == NULL) {
826 session_unlock(session);
827 continue;
828 }
829
830 cds_list_for_each_entry(channel,
831 &session->kernel_session->channel_list.head, list) {
832 /* Add channel fd to the kernel poll set */
833 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
834 if (ret < 0) {
835 session_unlock(session);
836 goto error;
837 }
838 DBG("Channel fd %d added to kernel set", channel->fd);
839 }
840 session_unlock(session);
841 }
842 session_unlock_list();
843
844 return 0;
845
846 error:
847 session_unlock_list();
848 return -1;
849 }
850
851 /*
852 * Find the channel fd from 'fd' over all tracing session. When found, check
853 * for new channel stream and send those stream fds to the kernel consumer.
854 *
855 * Useful for CPU hotplug feature.
856 */
857 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
858 {
859 int ret = 0;
860 struct ltt_session *session;
861 struct ltt_kernel_session *ksess;
862 struct ltt_kernel_channel *channel;
863
864 DBG("Updating kernel streams for channel fd %d", fd);
865
866 session_lock_list();
867 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
868 session_lock(session);
869 if (session->kernel_session == NULL) {
870 session_unlock(session);
871 continue;
872 }
873 ksess = session->kernel_session;
874
875 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
876 if (channel->fd == fd) {
877 DBG("Channel found, updating kernel streams");
878 ret = kernel_open_channel_stream(channel);
879 if (ret < 0) {
880 goto error;
881 }
882 /* Update the stream global counter */
883 ksess->stream_count_global += ret;
884
885 /*
886 * Have we already sent fds to the consumer? If yes, it means
887 * that tracing is started so it is safe to send our updated
888 * stream fds.
889 */
890 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
891 struct lttng_ht_iter iter;
892 struct consumer_socket *socket;
893
894 rcu_read_lock();
895 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
896 &iter.iter, socket, node.node) {
897 pthread_mutex_lock(socket->lock);
898 ret = kernel_consumer_send_channel_stream(socket,
899 channel, ksess,
900 session->output_traces ? 1 : 0);
901 pthread_mutex_unlock(socket->lock);
902 if (ret < 0) {
903 rcu_read_unlock();
904 goto error;
905 }
906 }
907 rcu_read_unlock();
908 }
909 goto error;
910 }
911 }
912 session_unlock(session);
913 }
914 session_unlock_list();
915 return ret;
916
917 error:
918 session_unlock(session);
919 session_unlock_list();
920 return ret;
921 }
922
923 /*
924 * For each tracing session, update newly registered apps. The session list
925 * lock MUST be acquired before calling this.
926 */
927 static void update_ust_app(int app_sock)
928 {
929 struct ltt_session *sess, *stmp;
930
931 /* Consumer is in an ERROR state. Stop any application update. */
932 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
933 /* Stop the update process since the consumer is dead. */
934 return;
935 }
936
937 /* For all tracing session(s) */
938 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
939 session_lock(sess);
940 if (sess->ust_session) {
941 ust_app_global_update(sess->ust_session, app_sock);
942 }
943 session_unlock(sess);
944 }
945 }
946
947 /*
948 * This thread manage event coming from the kernel.
949 *
950 * Features supported in this thread:
951 * -) CPU Hotplug
952 */
953 static void *thread_manage_kernel(void *data)
954 {
955 int ret, i, pollfd, update_poll_flag = 1, err = -1;
956 uint32_t revents, nb_fd;
957 char tmp;
958 struct lttng_poll_event events;
959
960 DBG("[thread] Thread manage kernel started");
961
962 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
963
964 /*
965 * This first step of the while is to clean this structure which could free
966 * non NULL pointers so initialize it before the loop.
967 */
968 lttng_poll_init(&events);
969
970 if (testpoint(sessiond_thread_manage_kernel)) {
971 goto error_testpoint;
972 }
973
974 health_code_update();
975
976 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
977 goto error_testpoint;
978 }
979
980 while (1) {
981 health_code_update();
982
983 if (update_poll_flag == 1) {
984 /* Clean events object. We are about to populate it again. */
985 lttng_poll_clean(&events);
986
987 ret = sessiond_set_thread_pollset(&events, 2);
988 if (ret < 0) {
989 goto error_poll_create;
990 }
991
992 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
993 if (ret < 0) {
994 goto error;
995 }
996
997 /* This will add the available kernel channel if any. */
998 ret = update_kernel_poll(&events);
999 if (ret < 0) {
1000 goto error;
1001 }
1002 update_poll_flag = 0;
1003 }
1004
1005 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
1006
1007 /* Poll infinite value of time */
1008 restart:
1009 health_poll_entry();
1010 ret = lttng_poll_wait(&events, -1);
1011 health_poll_exit();
1012 if (ret < 0) {
1013 /*
1014 * Restart interrupted system call.
1015 */
1016 if (errno == EINTR) {
1017 goto restart;
1018 }
1019 goto error;
1020 } else if (ret == 0) {
1021 /* Should not happen since timeout is infinite */
1022 ERR("Return value of poll is 0 with an infinite timeout.\n"
1023 "This should not have happened! Continuing...");
1024 continue;
1025 }
1026
1027 nb_fd = ret;
1028
1029 for (i = 0; i < nb_fd; i++) {
1030 /* Fetch once the poll data */
1031 revents = LTTNG_POLL_GETEV(&events, i);
1032 pollfd = LTTNG_POLL_GETFD(&events, i);
1033
1034 health_code_update();
1035
1036 /* Thread quit pipe has been closed. Killing thread. */
1037 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1038 if (ret) {
1039 err = 0;
1040 goto exit;
1041 }
1042
1043 /* Check for data on kernel pipe */
1044 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
1045 (void) lttng_read(kernel_poll_pipe[0],
1046 &tmp, 1);
1047 /*
1048 * Ret value is useless here, if this pipe gets any actions an
1049 * update is required anyway.
1050 */
1051 update_poll_flag = 1;
1052 continue;
1053 } else {
1054 /*
1055 * New CPU detected by the kernel. Adding kernel stream to
1056 * kernel session and updating the kernel consumer
1057 */
1058 if (revents & LPOLLIN) {
1059 ret = update_kernel_stream(&kconsumer_data, pollfd);
1060 if (ret < 0) {
1061 continue;
1062 }
1063 break;
1064 /*
1065 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1066 * and unregister kernel stream at this point.
1067 */
1068 }
1069 }
1070 }
1071 }
1072
1073 exit:
1074 error:
1075 lttng_poll_clean(&events);
1076 error_poll_create:
1077 error_testpoint:
1078 utils_close_pipe(kernel_poll_pipe);
1079 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1080 if (err) {
1081 health_error();
1082 ERR("Health error occurred in %s", __func__);
1083 WARN("Kernel thread died unexpectedly. "
1084 "Kernel tracing can continue but CPU hotplug is disabled.");
1085 }
1086 health_unregister(health_sessiond);
1087 DBG("Kernel thread dying");
1088 return NULL;
1089 }
1090
1091 /*
1092 * Signal pthread condition of the consumer data that the thread.
1093 */
1094 static void signal_consumer_condition(struct consumer_data *data, int state)
1095 {
1096 pthread_mutex_lock(&data->cond_mutex);
1097
1098 /*
1099 * The state is set before signaling. It can be any value, it's the waiter
1100 * job to correctly interpret this condition variable associated to the
1101 * consumer pthread_cond.
1102 *
1103 * A value of 0 means that the corresponding thread of the consumer data
1104 * was not started. 1 indicates that the thread has started and is ready
1105 * for action. A negative value means that there was an error during the
1106 * thread bootstrap.
1107 */
1108 data->consumer_thread_is_ready = state;
1109 (void) pthread_cond_signal(&data->cond);
1110
1111 pthread_mutex_unlock(&data->cond_mutex);
1112 }
1113
1114 /*
1115 * This thread manage the consumer error sent back to the session daemon.
1116 */
1117 static void *thread_manage_consumer(void *data)
1118 {
1119 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1120 uint32_t revents, nb_fd;
1121 enum lttcomm_return_code code;
1122 struct lttng_poll_event events;
1123 struct consumer_data *consumer_data = data;
1124
1125 DBG("[thread] Manage consumer started");
1126
1127 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1128
1129 health_code_update();
1130
1131 /*
1132 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1133 * metadata_sock. Nothing more will be added to this poll set.
1134 */
1135 ret = sessiond_set_thread_pollset(&events, 3);
1136 if (ret < 0) {
1137 goto error_poll;
1138 }
1139
1140 /*
1141 * The error socket here is already in a listening state which was done
1142 * just before spawning this thread to avoid a race between the consumer
1143 * daemon exec trying to connect and the listen() call.
1144 */
1145 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1146 if (ret < 0) {
1147 goto error;
1148 }
1149
1150 health_code_update();
1151
1152 /* Infinite blocking call, waiting for transmission */
1153 restart:
1154 health_poll_entry();
1155
1156 if (testpoint(sessiond_thread_manage_consumer)) {
1157 goto error;
1158 }
1159
1160 ret = lttng_poll_wait(&events, -1);
1161 health_poll_exit();
1162 if (ret < 0) {
1163 /*
1164 * Restart interrupted system call.
1165 */
1166 if (errno == EINTR) {
1167 goto restart;
1168 }
1169 goto error;
1170 }
1171
1172 nb_fd = ret;
1173
1174 for (i = 0; i < nb_fd; i++) {
1175 /* Fetch once the poll data */
1176 revents = LTTNG_POLL_GETEV(&events, i);
1177 pollfd = LTTNG_POLL_GETFD(&events, i);
1178
1179 health_code_update();
1180
1181 /* Thread quit pipe has been closed. Killing thread. */
1182 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1183 if (ret) {
1184 err = 0;
1185 goto exit;
1186 }
1187
1188 /* Event on the registration socket */
1189 if (pollfd == consumer_data->err_sock) {
1190 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1191 ERR("consumer err socket poll error");
1192 goto error;
1193 }
1194 }
1195 }
1196
1197 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1198 if (sock < 0) {
1199 goto error;
1200 }
1201
1202 /*
1203 * Set the CLOEXEC flag. Return code is useless because either way, the
1204 * show must go on.
1205 */
1206 (void) utils_set_fd_cloexec(sock);
1207
1208 health_code_update();
1209
1210 DBG2("Receiving code from consumer err_sock");
1211
1212 /* Getting status code from kconsumerd */
1213 ret = lttcomm_recv_unix_sock(sock, &code,
1214 sizeof(enum lttcomm_return_code));
1215 if (ret <= 0) {
1216 goto error;
1217 }
1218
1219 health_code_update();
1220 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1221 /* Connect both socket, command and metadata. */
1222 consumer_data->cmd_sock =
1223 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1224 consumer_data->metadata_fd =
1225 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1226 if (consumer_data->cmd_sock < 0
1227 || consumer_data->metadata_fd < 0) {
1228 PERROR("consumer connect cmd socket");
1229 /* On error, signal condition and quit. */
1230 signal_consumer_condition(consumer_data, -1);
1231 goto error;
1232 }
1233 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1234 /* Create metadata socket lock. */
1235 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1236 if (consumer_data->metadata_sock.lock == NULL) {
1237 PERROR("zmalloc pthread mutex");
1238 ret = -1;
1239 goto error;
1240 }
1241 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1242
1243 signal_consumer_condition(consumer_data, 1);
1244 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1245 DBG("Consumer metadata socket ready (fd: %d)",
1246 consumer_data->metadata_fd);
1247 } else {
1248 ERR("consumer error when waiting for SOCK_READY : %s",
1249 lttcomm_get_readable_code(-code));
1250 goto error;
1251 }
1252
1253 /* Remove the consumerd error sock since we've established a connexion */
1254 ret = lttng_poll_del(&events, consumer_data->err_sock);
1255 if (ret < 0) {
1256 goto error;
1257 }
1258
1259 /* Add new accepted error socket. */
1260 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1261 if (ret < 0) {
1262 goto error;
1263 }
1264
1265 /* Add metadata socket that is successfully connected. */
1266 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1267 LPOLLIN | LPOLLRDHUP);
1268 if (ret < 0) {
1269 goto error;
1270 }
1271
1272 health_code_update();
1273
1274 /* Infinite blocking call, waiting for transmission */
1275 restart_poll:
1276 while (1) {
1277 health_code_update();
1278
1279 /* Exit the thread because the thread quit pipe has been triggered. */
1280 if (should_quit) {
1281 /* Not a health error. */
1282 err = 0;
1283 goto exit;
1284 }
1285
1286 health_poll_entry();
1287 ret = lttng_poll_wait(&events, -1);
1288 health_poll_exit();
1289 if (ret < 0) {
1290 /*
1291 * Restart interrupted system call.
1292 */
1293 if (errno == EINTR) {
1294 goto restart_poll;
1295 }
1296 goto error;
1297 }
1298
1299 nb_fd = ret;
1300
1301 for (i = 0; i < nb_fd; i++) {
1302 /* Fetch once the poll data */
1303 revents = LTTNG_POLL_GETEV(&events, i);
1304 pollfd = LTTNG_POLL_GETFD(&events, i);
1305
1306 health_code_update();
1307
1308 /*
1309 * Thread quit pipe has been triggered, flag that we should stop
1310 * but continue the current loop to handle potential data from
1311 * consumer.
1312 */
1313 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1314
1315 if (pollfd == sock) {
1316 /* Event on the consumerd socket */
1317 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1318 ERR("consumer err socket second poll error");
1319 goto error;
1320 }
1321 health_code_update();
1322 /* Wait for any kconsumerd error */
1323 ret = lttcomm_recv_unix_sock(sock, &code,
1324 sizeof(enum lttcomm_return_code));
1325 if (ret <= 0) {
1326 ERR("consumer closed the command socket");
1327 goto error;
1328 }
1329
1330 ERR("consumer return code : %s",
1331 lttcomm_get_readable_code(-code));
1332
1333 goto exit;
1334 } else if (pollfd == consumer_data->metadata_fd) {
1335 /* UST metadata requests */
1336 ret = ust_consumer_metadata_request(
1337 &consumer_data->metadata_sock);
1338 if (ret < 0) {
1339 ERR("Handling metadata request");
1340 goto error;
1341 }
1342 }
1343 /* No need for an else branch all FDs are tested prior. */
1344 }
1345 health_code_update();
1346 }
1347
1348 exit:
1349 error:
1350 /*
1351 * We lock here because we are about to close the sockets and some other
1352 * thread might be using them so get exclusive access which will abort all
1353 * other consumer command by other threads.
1354 */
1355 pthread_mutex_lock(&consumer_data->lock);
1356
1357 /* Immediately set the consumerd state to stopped */
1358 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1359 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1360 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1361 consumer_data->type == LTTNG_CONSUMER32_UST) {
1362 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1363 } else {
1364 /* Code flow error... */
1365 assert(0);
1366 }
1367
1368 if (consumer_data->err_sock >= 0) {
1369 ret = close(consumer_data->err_sock);
1370 if (ret) {
1371 PERROR("close");
1372 }
1373 consumer_data->err_sock = -1;
1374 }
1375 if (consumer_data->cmd_sock >= 0) {
1376 ret = close(consumer_data->cmd_sock);
1377 if (ret) {
1378 PERROR("close");
1379 }
1380 consumer_data->cmd_sock = -1;
1381 }
1382 if (consumer_data->metadata_sock.fd_ptr &&
1383 *consumer_data->metadata_sock.fd_ptr >= 0) {
1384 ret = close(*consumer_data->metadata_sock.fd_ptr);
1385 if (ret) {
1386 PERROR("close");
1387 }
1388 }
1389 if (sock >= 0) {
1390 ret = close(sock);
1391 if (ret) {
1392 PERROR("close");
1393 }
1394 }
1395
1396 unlink(consumer_data->err_unix_sock_path);
1397 unlink(consumer_data->cmd_unix_sock_path);
1398 consumer_data->pid = 0;
1399 pthread_mutex_unlock(&consumer_data->lock);
1400
1401 /* Cleanup metadata socket mutex. */
1402 if (consumer_data->metadata_sock.lock) {
1403 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1404 free(consumer_data->metadata_sock.lock);
1405 }
1406 lttng_poll_clean(&events);
1407 error_poll:
1408 if (err) {
1409 health_error();
1410 ERR("Health error occurred in %s", __func__);
1411 }
1412 health_unregister(health_sessiond);
1413 DBG("consumer thread cleanup completed");
1414
1415 return NULL;
1416 }
1417
1418 /*
1419 * This thread manage application communication.
1420 */
1421 static void *thread_manage_apps(void *data)
1422 {
1423 int i, ret, pollfd, err = -1;
1424 ssize_t size_ret;
1425 uint32_t revents, nb_fd;
1426 struct lttng_poll_event events;
1427
1428 DBG("[thread] Manage application started");
1429
1430 rcu_register_thread();
1431 rcu_thread_online();
1432
1433 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1434
1435 if (testpoint(sessiond_thread_manage_apps)) {
1436 goto error_testpoint;
1437 }
1438
1439 health_code_update();
1440
1441 ret = sessiond_set_thread_pollset(&events, 2);
1442 if (ret < 0) {
1443 goto error_poll_create;
1444 }
1445
1446 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1447 if (ret < 0) {
1448 goto error;
1449 }
1450
1451 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1452 goto error;
1453 }
1454
1455 health_code_update();
1456
1457 while (1) {
1458 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1459
1460 /* Inifinite blocking call, waiting for transmission */
1461 restart:
1462 health_poll_entry();
1463 ret = lttng_poll_wait(&events, -1);
1464 health_poll_exit();
1465 if (ret < 0) {
1466 /*
1467 * Restart interrupted system call.
1468 */
1469 if (errno == EINTR) {
1470 goto restart;
1471 }
1472 goto error;
1473 }
1474
1475 nb_fd = ret;
1476
1477 for (i = 0; i < nb_fd; i++) {
1478 /* Fetch once the poll data */
1479 revents = LTTNG_POLL_GETEV(&events, i);
1480 pollfd = LTTNG_POLL_GETFD(&events, i);
1481
1482 health_code_update();
1483
1484 /* Thread quit pipe has been closed. Killing thread. */
1485 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1486 if (ret) {
1487 err = 0;
1488 goto exit;
1489 }
1490
1491 /* Inspect the apps cmd pipe */
1492 if (pollfd == apps_cmd_pipe[0]) {
1493 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1494 ERR("Apps command pipe error");
1495 goto error;
1496 } else if (revents & LPOLLIN) {
1497 int sock;
1498
1499 /* Empty pipe */
1500 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1501 if (size_ret < sizeof(sock)) {
1502 PERROR("read apps cmd pipe");
1503 goto error;
1504 }
1505
1506 health_code_update();
1507
1508 /*
1509 * We only monitor the error events of the socket. This
1510 * thread does not handle any incoming data from UST
1511 * (POLLIN).
1512 */
1513 ret = lttng_poll_add(&events, sock,
1514 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1515 if (ret < 0) {
1516 goto error;
1517 }
1518
1519 DBG("Apps with sock %d added to poll set", sock);
1520 }
1521 } else {
1522 /*
1523 * At this point, we know that a registered application made
1524 * the event at poll_wait.
1525 */
1526 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1527 /* Removing from the poll set */
1528 ret = lttng_poll_del(&events, pollfd);
1529 if (ret < 0) {
1530 goto error;
1531 }
1532
1533 /* Socket closed on remote end. */
1534 ust_app_unregister(pollfd);
1535 }
1536 }
1537
1538 health_code_update();
1539 }
1540 }
1541
1542 exit:
1543 error:
1544 lttng_poll_clean(&events);
1545 error_poll_create:
1546 error_testpoint:
1547 utils_close_pipe(apps_cmd_pipe);
1548 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1549
1550 /*
1551 * We don't clean the UST app hash table here since already registered
1552 * applications can still be controlled so let them be until the session
1553 * daemon dies or the applications stop.
1554 */
1555
1556 if (err) {
1557 health_error();
1558 ERR("Health error occurred in %s", __func__);
1559 }
1560 health_unregister(health_sessiond);
1561 DBG("Application communication apps thread cleanup complete");
1562 rcu_thread_offline();
1563 rcu_unregister_thread();
1564 return NULL;
1565 }
1566
1567 /*
1568 * Send a socket to a thread This is called from the dispatch UST registration
1569 * thread once all sockets are set for the application.
1570 *
1571 * The sock value can be invalid, we don't really care, the thread will handle
1572 * it and make the necessary cleanup if so.
1573 *
1574 * On success, return 0 else a negative value being the errno message of the
1575 * write().
1576 */
1577 static int send_socket_to_thread(int fd, int sock)
1578 {
1579 ssize_t ret;
1580
1581 /*
1582 * It's possible that the FD is set as invalid with -1 concurrently just
1583 * before calling this function being a shutdown state of the thread.
1584 */
1585 if (fd < 0) {
1586 ret = -EBADF;
1587 goto error;
1588 }
1589
1590 ret = lttng_write(fd, &sock, sizeof(sock));
1591 if (ret < sizeof(sock)) {
1592 PERROR("write apps pipe %d", fd);
1593 if (ret < 0) {
1594 ret = -errno;
1595 }
1596 goto error;
1597 }
1598
1599 /* All good. Don't send back the write positive ret value. */
1600 ret = 0;
1601 error:
1602 return (int) ret;
1603 }
1604
1605 /*
1606 * Sanitize the wait queue of the dispatch registration thread meaning removing
1607 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1608 * notify socket is never received.
1609 */
1610 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1611 {
1612 int ret, nb_fd = 0, i;
1613 unsigned int fd_added = 0;
1614 struct lttng_poll_event events;
1615 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1616
1617 assert(wait_queue);
1618
1619 lttng_poll_init(&events);
1620
1621 /* Just skip everything for an empty queue. */
1622 if (!wait_queue->count) {
1623 goto end;
1624 }
1625
1626 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1627 if (ret < 0) {
1628 goto error_create;
1629 }
1630
1631 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1632 &wait_queue->head, head) {
1633 assert(wait_node->app);
1634 ret = lttng_poll_add(&events, wait_node->app->sock,
1635 LPOLLHUP | LPOLLERR);
1636 if (ret < 0) {
1637 goto error;
1638 }
1639
1640 fd_added = 1;
1641 }
1642
1643 if (!fd_added) {
1644 goto end;
1645 }
1646
1647 /*
1648 * Poll but don't block so we can quickly identify the faulty events and
1649 * clean them afterwards from the wait queue.
1650 */
1651 ret = lttng_poll_wait(&events, 0);
1652 if (ret < 0) {
1653 goto error;
1654 }
1655 nb_fd = ret;
1656
1657 for (i = 0; i < nb_fd; i++) {
1658 /* Get faulty FD. */
1659 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1660 int pollfd = LTTNG_POLL_GETFD(&events, i);
1661
1662 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1663 &wait_queue->head, head) {
1664 if (pollfd == wait_node->app->sock &&
1665 (revents & (LPOLLHUP | LPOLLERR))) {
1666 cds_list_del(&wait_node->head);
1667 wait_queue->count--;
1668 ust_app_destroy(wait_node->app);
1669 free(wait_node);
1670 break;
1671 }
1672 }
1673 }
1674
1675 if (nb_fd > 0) {
1676 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1677 }
1678
1679 end:
1680 lttng_poll_clean(&events);
1681 return;
1682
1683 error:
1684 lttng_poll_clean(&events);
1685 error_create:
1686 ERR("Unable to sanitize wait queue");
1687 return;
1688 }
1689
1690 /*
1691 * Dispatch request from the registration threads to the application
1692 * communication thread.
1693 */
1694 static void *thread_dispatch_ust_registration(void *data)
1695 {
1696 int ret, err = -1;
1697 struct cds_wfcq_node *node;
1698 struct ust_command *ust_cmd = NULL;
1699 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1700 struct ust_reg_wait_queue wait_queue = {
1701 .count = 0,
1702 };
1703
1704 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1705
1706 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1707 goto error_testpoint;
1708 }
1709
1710 health_code_update();
1711
1712 CDS_INIT_LIST_HEAD(&wait_queue.head);
1713
1714 DBG("[thread] Dispatch UST command started");
1715
1716 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1717 health_code_update();
1718
1719 /* Atomically prepare the queue futex */
1720 futex_nto1_prepare(&ust_cmd_queue.futex);
1721
1722 do {
1723 struct ust_app *app = NULL;
1724 ust_cmd = NULL;
1725
1726 /*
1727 * Make sure we don't have node(s) that have hung up before receiving
1728 * the notify socket. This is to clean the list in order to avoid
1729 * memory leaks from notify socket that are never seen.
1730 */
1731 sanitize_wait_queue(&wait_queue);
1732
1733 health_code_update();
1734 /* Dequeue command for registration */
1735 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1736 if (node == NULL) {
1737 DBG("Woken up but nothing in the UST command queue");
1738 /* Continue thread execution */
1739 break;
1740 }
1741
1742 ust_cmd = caa_container_of(node, struct ust_command, node);
1743
1744 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1745 " gid:%d sock:%d name:%s (version %d.%d)",
1746 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1747 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1748 ust_cmd->sock, ust_cmd->reg_msg.name,
1749 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1750
1751 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1752 wait_node = zmalloc(sizeof(*wait_node));
1753 if (!wait_node) {
1754 PERROR("zmalloc wait_node dispatch");
1755 ret = close(ust_cmd->sock);
1756 if (ret < 0) {
1757 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1758 }
1759 lttng_fd_put(LTTNG_FD_APPS, 1);
1760 free(ust_cmd);
1761 goto error;
1762 }
1763 CDS_INIT_LIST_HEAD(&wait_node->head);
1764
1765 /* Create application object if socket is CMD. */
1766 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1767 ust_cmd->sock);
1768 if (!wait_node->app) {
1769 ret = close(ust_cmd->sock);
1770 if (ret < 0) {
1771 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1772 }
1773 lttng_fd_put(LTTNG_FD_APPS, 1);
1774 free(wait_node);
1775 free(ust_cmd);
1776 continue;
1777 }
1778 /*
1779 * Add application to the wait queue so we can set the notify
1780 * socket before putting this object in the global ht.
1781 */
1782 cds_list_add(&wait_node->head, &wait_queue.head);
1783 wait_queue.count++;
1784
1785 free(ust_cmd);
1786 /*
1787 * We have to continue here since we don't have the notify
1788 * socket and the application MUST be added to the hash table
1789 * only at that moment.
1790 */
1791 continue;
1792 } else {
1793 /*
1794 * Look for the application in the local wait queue and set the
1795 * notify socket if found.
1796 */
1797 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1798 &wait_queue.head, head) {
1799 health_code_update();
1800 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1801 wait_node->app->notify_sock = ust_cmd->sock;
1802 cds_list_del(&wait_node->head);
1803 wait_queue.count--;
1804 app = wait_node->app;
1805 free(wait_node);
1806 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1807 break;
1808 }
1809 }
1810
1811 /*
1812 * With no application at this stage the received socket is
1813 * basically useless so close it before we free the cmd data
1814 * structure for good.
1815 */
1816 if (!app) {
1817 ret = close(ust_cmd->sock);
1818 if (ret < 0) {
1819 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1820 }
1821 lttng_fd_put(LTTNG_FD_APPS, 1);
1822 }
1823 free(ust_cmd);
1824 }
1825
1826 if (app) {
1827 /*
1828 * @session_lock_list
1829 *
1830 * Lock the global session list so from the register up to the
1831 * registration done message, no thread can see the application
1832 * and change its state.
1833 */
1834 session_lock_list();
1835 rcu_read_lock();
1836
1837 /*
1838 * Add application to the global hash table. This needs to be
1839 * done before the update to the UST registry can locate the
1840 * application.
1841 */
1842 ust_app_add(app);
1843
1844 /* Set app version. This call will print an error if needed. */
1845 (void) ust_app_version(app);
1846
1847 /* Send notify socket through the notify pipe. */
1848 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1849 app->notify_sock);
1850 if (ret < 0) {
1851 rcu_read_unlock();
1852 session_unlock_list();
1853 /*
1854 * No notify thread, stop the UST tracing. However, this is
1855 * not an internal error of the this thread thus setting
1856 * the health error code to a normal exit.
1857 */
1858 err = 0;
1859 goto error;
1860 }
1861
1862 /*
1863 * Update newly registered application with the tracing
1864 * registry info already enabled information.
1865 */
1866 update_ust_app(app->sock);
1867
1868 /*
1869 * Don't care about return value. Let the manage apps threads
1870 * handle app unregistration upon socket close.
1871 */
1872 (void) ust_app_register_done(app->sock);
1873
1874 /*
1875 * Even if the application socket has been closed, send the app
1876 * to the thread and unregistration will take place at that
1877 * place.
1878 */
1879 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1880 if (ret < 0) {
1881 rcu_read_unlock();
1882 session_unlock_list();
1883 /*
1884 * No apps. thread, stop the UST tracing. However, this is
1885 * not an internal error of the this thread thus setting
1886 * the health error code to a normal exit.
1887 */
1888 err = 0;
1889 goto error;
1890 }
1891
1892 rcu_read_unlock();
1893 session_unlock_list();
1894 }
1895 } while (node != NULL);
1896
1897 health_poll_entry();
1898 /* Futex wait on queue. Blocking call on futex() */
1899 futex_nto1_wait(&ust_cmd_queue.futex);
1900 health_poll_exit();
1901 }
1902 /* Normal exit, no error */
1903 err = 0;
1904
1905 error:
1906 /* Clean up wait queue. */
1907 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1908 &wait_queue.head, head) {
1909 cds_list_del(&wait_node->head);
1910 wait_queue.count--;
1911 free(wait_node);
1912 }
1913
1914 error_testpoint:
1915 DBG("Dispatch thread dying");
1916 if (err) {
1917 health_error();
1918 ERR("Health error occurred in %s", __func__);
1919 }
1920 health_unregister(health_sessiond);
1921 return NULL;
1922 }
1923
1924 /*
1925 * This thread manage application registration.
1926 */
1927 static void *thread_registration_apps(void *data)
1928 {
1929 int sock = -1, i, ret, pollfd, err = -1;
1930 uint32_t revents, nb_fd;
1931 struct lttng_poll_event events;
1932 /*
1933 * Get allocated in this thread, enqueued to a global queue, dequeued and
1934 * freed in the manage apps thread.
1935 */
1936 struct ust_command *ust_cmd = NULL;
1937
1938 DBG("[thread] Manage application registration started");
1939
1940 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1941
1942 if (testpoint(sessiond_thread_registration_apps)) {
1943 goto error_testpoint;
1944 }
1945
1946 ret = lttcomm_listen_unix_sock(apps_sock);
1947 if (ret < 0) {
1948 goto error_listen;
1949 }
1950
1951 /*
1952 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1953 * more will be added to this poll set.
1954 */
1955 ret = sessiond_set_thread_pollset(&events, 2);
1956 if (ret < 0) {
1957 goto error_create_poll;
1958 }
1959
1960 /* Add the application registration socket */
1961 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1962 if (ret < 0) {
1963 goto error_poll_add;
1964 }
1965
1966 /* Notify all applications to register */
1967 ret = notify_ust_apps(1);
1968 if (ret < 0) {
1969 ERR("Failed to notify applications or create the wait shared memory.\n"
1970 "Execution continues but there might be problem for already\n"
1971 "running applications that wishes to register.");
1972 }
1973
1974 while (1) {
1975 DBG("Accepting application registration");
1976
1977 /* Inifinite blocking call, waiting for transmission */
1978 restart:
1979 health_poll_entry();
1980 ret = lttng_poll_wait(&events, -1);
1981 health_poll_exit();
1982 if (ret < 0) {
1983 /*
1984 * Restart interrupted system call.
1985 */
1986 if (errno == EINTR) {
1987 goto restart;
1988 }
1989 goto error;
1990 }
1991
1992 nb_fd = ret;
1993
1994 for (i = 0; i < nb_fd; i++) {
1995 health_code_update();
1996
1997 /* Fetch once the poll data */
1998 revents = LTTNG_POLL_GETEV(&events, i);
1999 pollfd = LTTNG_POLL_GETFD(&events, i);
2000
2001 /* Thread quit pipe has been closed. Killing thread. */
2002 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2003 if (ret) {
2004 err = 0;
2005 goto exit;
2006 }
2007
2008 /* Event on the registration socket */
2009 if (pollfd == apps_sock) {
2010 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2011 ERR("Register apps socket poll error");
2012 goto error;
2013 } else if (revents & LPOLLIN) {
2014 sock = lttcomm_accept_unix_sock(apps_sock);
2015 if (sock < 0) {
2016 goto error;
2017 }
2018
2019 /*
2020 * Set socket timeout for both receiving and ending.
2021 * app_socket_timeout is in seconds, whereas
2022 * lttcomm_setsockopt_rcv_timeout and
2023 * lttcomm_setsockopt_snd_timeout expect msec as
2024 * parameter.
2025 */
2026 (void) lttcomm_setsockopt_rcv_timeout(sock,
2027 app_socket_timeout * 1000);
2028 (void) lttcomm_setsockopt_snd_timeout(sock,
2029 app_socket_timeout * 1000);
2030
2031 /*
2032 * Set the CLOEXEC flag. Return code is useless because
2033 * either way, the show must go on.
2034 */
2035 (void) utils_set_fd_cloexec(sock);
2036
2037 /* Create UST registration command for enqueuing */
2038 ust_cmd = zmalloc(sizeof(struct ust_command));
2039 if (ust_cmd == NULL) {
2040 PERROR("ust command zmalloc");
2041 goto error;
2042 }
2043
2044 /*
2045 * Using message-based transmissions to ensure we don't
2046 * have to deal with partially received messages.
2047 */
2048 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2049 if (ret < 0) {
2050 ERR("Exhausted file descriptors allowed for applications.");
2051 free(ust_cmd);
2052 ret = close(sock);
2053 if (ret) {
2054 PERROR("close");
2055 }
2056 sock = -1;
2057 continue;
2058 }
2059
2060 health_code_update();
2061 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2062 if (ret < 0) {
2063 free(ust_cmd);
2064 /* Close socket of the application. */
2065 ret = close(sock);
2066 if (ret) {
2067 PERROR("close");
2068 }
2069 lttng_fd_put(LTTNG_FD_APPS, 1);
2070 sock = -1;
2071 continue;
2072 }
2073 health_code_update();
2074
2075 ust_cmd->sock = sock;
2076 sock = -1;
2077
2078 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2079 " gid:%d sock:%d name:%s (version %d.%d)",
2080 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2081 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2082 ust_cmd->sock, ust_cmd->reg_msg.name,
2083 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2084
2085 /*
2086 * Lock free enqueue the registration request. The red pill
2087 * has been taken! This apps will be part of the *system*.
2088 */
2089 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2090
2091 /*
2092 * Wake the registration queue futex. Implicit memory
2093 * barrier with the exchange in cds_wfcq_enqueue.
2094 */
2095 futex_nto1_wake(&ust_cmd_queue.futex);
2096 }
2097 }
2098 }
2099 }
2100
2101 exit:
2102 error:
2103 /* Notify that the registration thread is gone */
2104 notify_ust_apps(0);
2105
2106 if (apps_sock >= 0) {
2107 ret = close(apps_sock);
2108 if (ret) {
2109 PERROR("close");
2110 }
2111 }
2112 if (sock >= 0) {
2113 ret = close(sock);
2114 if (ret) {
2115 PERROR("close");
2116 }
2117 lttng_fd_put(LTTNG_FD_APPS, 1);
2118 }
2119 unlink(apps_unix_sock_path);
2120
2121 error_poll_add:
2122 lttng_poll_clean(&events);
2123 error_listen:
2124 error_create_poll:
2125 error_testpoint:
2126 DBG("UST Registration thread cleanup complete");
2127 if (err) {
2128 health_error();
2129 ERR("Health error occurred in %s", __func__);
2130 }
2131 health_unregister(health_sessiond);
2132
2133 return NULL;
2134 }
2135
2136 /*
2137 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2138 * exec or it will fails.
2139 */
2140 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2141 {
2142 int ret, clock_ret;
2143 struct timespec timeout;
2144
2145 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2146 consumer_data->consumer_thread_is_ready = 0;
2147
2148 /* Setup pthread condition */
2149 ret = pthread_condattr_init(&consumer_data->condattr);
2150 if (ret != 0) {
2151 errno = ret;
2152 PERROR("pthread_condattr_init consumer data");
2153 goto error;
2154 }
2155
2156 /*
2157 * Set the monotonic clock in order to make sure we DO NOT jump in time
2158 * between the clock_gettime() call and the timedwait call. See bug #324
2159 * for a more details and how we noticed it.
2160 */
2161 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2162 if (ret != 0) {
2163 errno = ret;
2164 PERROR("pthread_condattr_setclock consumer data");
2165 goto error;
2166 }
2167
2168 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2169 if (ret != 0) {
2170 errno = ret;
2171 PERROR("pthread_cond_init consumer data");
2172 goto error;
2173 }
2174
2175 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2176 consumer_data);
2177 if (ret != 0) {
2178 PERROR("pthread_create consumer");
2179 ret = -1;
2180 goto error;
2181 }
2182
2183 /* We are about to wait on a pthread condition */
2184 pthread_mutex_lock(&consumer_data->cond_mutex);
2185
2186 /* Get time for sem_timedwait absolute timeout */
2187 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2188 /*
2189 * Set the timeout for the condition timed wait even if the clock gettime
2190 * call fails since we might loop on that call and we want to avoid to
2191 * increment the timeout too many times.
2192 */
2193 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2194
2195 /*
2196 * The following loop COULD be skipped in some conditions so this is why we
2197 * set ret to 0 in order to make sure at least one round of the loop is
2198 * done.
2199 */
2200 ret = 0;
2201
2202 /*
2203 * Loop until the condition is reached or when a timeout is reached. Note
2204 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2205 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2206 * possible. This loop does not take any chances and works with both of
2207 * them.
2208 */
2209 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2210 if (clock_ret < 0) {
2211 PERROR("clock_gettime spawn consumer");
2212 /* Infinite wait for the consumerd thread to be ready */
2213 ret = pthread_cond_wait(&consumer_data->cond,
2214 &consumer_data->cond_mutex);
2215 } else {
2216 ret = pthread_cond_timedwait(&consumer_data->cond,
2217 &consumer_data->cond_mutex, &timeout);
2218 }
2219 }
2220
2221 /* Release the pthread condition */
2222 pthread_mutex_unlock(&consumer_data->cond_mutex);
2223
2224 if (ret != 0) {
2225 errno = ret;
2226 if (ret == ETIMEDOUT) {
2227 int pth_ret;
2228
2229 /*
2230 * Call has timed out so we kill the kconsumerd_thread and return
2231 * an error.
2232 */
2233 ERR("Condition timed out. The consumer thread was never ready."
2234 " Killing it");
2235 pth_ret = pthread_cancel(consumer_data->thread);
2236 if (pth_ret < 0) {
2237 PERROR("pthread_cancel consumer thread");
2238 }
2239 } else {
2240 PERROR("pthread_cond_wait failed consumer thread");
2241 }
2242 /* Caller is expecting a negative value on failure. */
2243 ret = -1;
2244 goto error;
2245 }
2246
2247 pthread_mutex_lock(&consumer_data->pid_mutex);
2248 if (consumer_data->pid == 0) {
2249 ERR("Consumerd did not start");
2250 pthread_mutex_unlock(&consumer_data->pid_mutex);
2251 goto error;
2252 }
2253 pthread_mutex_unlock(&consumer_data->pid_mutex);
2254
2255 return 0;
2256
2257 error:
2258 return ret;
2259 }
2260
2261 /*
2262 * Join consumer thread
2263 */
2264 static int join_consumer_thread(struct consumer_data *consumer_data)
2265 {
2266 void *status;
2267
2268 /* Consumer pid must be a real one. */
2269 if (consumer_data->pid > 0) {
2270 int ret;
2271 ret = kill(consumer_data->pid, SIGTERM);
2272 if (ret) {
2273 ERR("Error killing consumer daemon");
2274 return ret;
2275 }
2276 return pthread_join(consumer_data->thread, &status);
2277 } else {
2278 return 0;
2279 }
2280 }
2281
2282 /*
2283 * Fork and exec a consumer daemon (consumerd).
2284 *
2285 * Return pid if successful else -1.
2286 */
2287 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2288 {
2289 int ret;
2290 pid_t pid;
2291 const char *consumer_to_use;
2292 const char *verbosity;
2293 struct stat st;
2294
2295 DBG("Spawning consumerd");
2296
2297 pid = fork();
2298 if (pid == 0) {
2299 /*
2300 * Exec consumerd.
2301 */
2302 if (opt_verbose_consumer) {
2303 verbosity = "--verbose";
2304 } else if (lttng_opt_quiet) {
2305 verbosity = "--quiet";
2306 } else {
2307 verbosity = "";
2308 }
2309
2310 switch (consumer_data->type) {
2311 case LTTNG_CONSUMER_KERNEL:
2312 /*
2313 * Find out which consumerd to execute. We will first try the
2314 * 64-bit path, then the sessiond's installation directory, and
2315 * fallback on the 32-bit one,
2316 */
2317 DBG3("Looking for a kernel consumer at these locations:");
2318 DBG3(" 1) %s", consumerd64_bin);
2319 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2320 DBG3(" 3) %s", consumerd32_bin);
2321 if (stat(consumerd64_bin, &st) == 0) {
2322 DBG3("Found location #1");
2323 consumer_to_use = consumerd64_bin;
2324 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2325 DBG3("Found location #2");
2326 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2327 } else if (stat(consumerd32_bin, &st) == 0) {
2328 DBG3("Found location #3");
2329 consumer_to_use = consumerd32_bin;
2330 } else {
2331 DBG("Could not find any valid consumerd executable");
2332 ret = -EINVAL;
2333 break;
2334 }
2335 DBG("Using kernel consumer at: %s", consumer_to_use);
2336 ret = execl(consumer_to_use,
2337 "lttng-consumerd", verbosity, "-k",
2338 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2339 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2340 "--group", tracing_group_name,
2341 NULL);
2342 break;
2343 case LTTNG_CONSUMER64_UST:
2344 {
2345 char *tmpnew = NULL;
2346
2347 if (consumerd64_libdir[0] != '\0') {
2348 char *tmp;
2349 size_t tmplen;
2350
2351 tmp = getenv("LD_LIBRARY_PATH");
2352 if (!tmp) {
2353 tmp = "";
2354 }
2355 tmplen = strlen("LD_LIBRARY_PATH=")
2356 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2357 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2358 if (!tmpnew) {
2359 ret = -ENOMEM;
2360 goto error;
2361 }
2362 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2363 strcat(tmpnew, consumerd64_libdir);
2364 if (tmp[0] != '\0') {
2365 strcat(tmpnew, ":");
2366 strcat(tmpnew, tmp);
2367 }
2368 ret = putenv(tmpnew);
2369 if (ret) {
2370 ret = -errno;
2371 free(tmpnew);
2372 goto error;
2373 }
2374 }
2375 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2376 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2377 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2378 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2379 "--group", tracing_group_name,
2380 NULL);
2381 if (consumerd64_libdir[0] != '\0') {
2382 free(tmpnew);
2383 }
2384 break;
2385 }
2386 case LTTNG_CONSUMER32_UST:
2387 {
2388 char *tmpnew = NULL;
2389
2390 if (consumerd32_libdir[0] != '\0') {
2391 char *tmp;
2392 size_t tmplen;
2393
2394 tmp = getenv("LD_LIBRARY_PATH");
2395 if (!tmp) {
2396 tmp = "";
2397 }
2398 tmplen = strlen("LD_LIBRARY_PATH=")
2399 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2400 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2401 if (!tmpnew) {
2402 ret = -ENOMEM;
2403 goto error;
2404 }
2405 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2406 strcat(tmpnew, consumerd32_libdir);
2407 if (tmp[0] != '\0') {
2408 strcat(tmpnew, ":");
2409 strcat(tmpnew, tmp);
2410 }
2411 ret = putenv(tmpnew);
2412 if (ret) {
2413 ret = -errno;
2414 free(tmpnew);
2415 goto error;
2416 }
2417 }
2418 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2419 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2420 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2421 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2422 "--group", tracing_group_name,
2423 NULL);
2424 if (consumerd32_libdir[0] != '\0') {
2425 free(tmpnew);
2426 }
2427 break;
2428 }
2429 default:
2430 PERROR("unknown consumer type");
2431 exit(EXIT_FAILURE);
2432 }
2433 if (errno != 0) {
2434 PERROR("Consumer execl()");
2435 }
2436 /* Reaching this point, we got a failure on our execl(). */
2437 exit(EXIT_FAILURE);
2438 } else if (pid > 0) {
2439 ret = pid;
2440 } else {
2441 PERROR("start consumer fork");
2442 ret = -errno;
2443 }
2444 error:
2445 return ret;
2446 }
2447
2448 /*
2449 * Spawn the consumerd daemon and session daemon thread.
2450 */
2451 static int start_consumerd(struct consumer_data *consumer_data)
2452 {
2453 int ret;
2454
2455 /*
2456 * Set the listen() state on the socket since there is a possible race
2457 * between the exec() of the consumer daemon and this call if place in the
2458 * consumer thread. See bug #366 for more details.
2459 */
2460 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2461 if (ret < 0) {
2462 goto error;
2463 }
2464
2465 pthread_mutex_lock(&consumer_data->pid_mutex);
2466 if (consumer_data->pid != 0) {
2467 pthread_mutex_unlock(&consumer_data->pid_mutex);
2468 goto end;
2469 }
2470
2471 ret = spawn_consumerd(consumer_data);
2472 if (ret < 0) {
2473 ERR("Spawning consumerd failed");
2474 pthread_mutex_unlock(&consumer_data->pid_mutex);
2475 goto error;
2476 }
2477
2478 /* Setting up the consumer_data pid */
2479 consumer_data->pid = ret;
2480 DBG2("Consumer pid %d", consumer_data->pid);
2481 pthread_mutex_unlock(&consumer_data->pid_mutex);
2482
2483 DBG2("Spawning consumer control thread");
2484 ret = spawn_consumer_thread(consumer_data);
2485 if (ret < 0) {
2486 ERR("Fatal error spawning consumer control thread");
2487 goto error;
2488 }
2489
2490 end:
2491 return 0;
2492
2493 error:
2494 /* Cleanup already created sockets on error. */
2495 if (consumer_data->err_sock >= 0) {
2496 int err;
2497
2498 err = close(consumer_data->err_sock);
2499 if (err < 0) {
2500 PERROR("close consumer data error socket");
2501 }
2502 }
2503 return ret;
2504 }
2505
2506 /*
2507 * Setup necessary data for kernel tracer action.
2508 */
2509 static int init_kernel_tracer(void)
2510 {
2511 int ret;
2512
2513 /* Modprobe lttng kernel modules */
2514 ret = modprobe_lttng_control();
2515 if (ret < 0) {
2516 goto error;
2517 }
2518
2519 /* Open debugfs lttng */
2520 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2521 if (kernel_tracer_fd < 0) {
2522 DBG("Failed to open %s", module_proc_lttng);
2523 ret = -1;
2524 goto error_open;
2525 }
2526
2527 /* Validate kernel version */
2528 ret = kernel_validate_version(kernel_tracer_fd);
2529 if (ret < 0) {
2530 goto error_version;
2531 }
2532
2533 ret = modprobe_lttng_data();
2534 if (ret < 0) {
2535 goto error_modules;
2536 }
2537
2538 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2539 return 0;
2540
2541 error_version:
2542 modprobe_remove_lttng_control();
2543 ret = close(kernel_tracer_fd);
2544 if (ret) {
2545 PERROR("close");
2546 }
2547 kernel_tracer_fd = -1;
2548 return LTTNG_ERR_KERN_VERSION;
2549
2550 error_modules:
2551 ret = close(kernel_tracer_fd);
2552 if (ret) {
2553 PERROR("close");
2554 }
2555
2556 error_open:
2557 modprobe_remove_lttng_control();
2558
2559 error:
2560 WARN("No kernel tracer available");
2561 kernel_tracer_fd = -1;
2562 if (!is_root) {
2563 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2564 } else {
2565 return LTTNG_ERR_KERN_NA;
2566 }
2567 }
2568
2569
2570 /*
2571 * Copy consumer output from the tracing session to the domain session. The
2572 * function also applies the right modification on a per domain basis for the
2573 * trace files destination directory.
2574 *
2575 * Should *NOT* be called with RCU read-side lock held.
2576 */
2577 static int copy_session_consumer(int domain, struct ltt_session *session)
2578 {
2579 int ret;
2580 const char *dir_name;
2581 struct consumer_output *consumer;
2582
2583 assert(session);
2584 assert(session->consumer);
2585
2586 switch (domain) {
2587 case LTTNG_DOMAIN_KERNEL:
2588 DBG3("Copying tracing session consumer output in kernel session");
2589 /*
2590 * XXX: We should audit the session creation and what this function
2591 * does "extra" in order to avoid a destroy since this function is used
2592 * in the domain session creation (kernel and ust) only. Same for UST
2593 * domain.
2594 */
2595 if (session->kernel_session->consumer) {
2596 consumer_destroy_output(session->kernel_session->consumer);
2597 }
2598 session->kernel_session->consumer =
2599 consumer_copy_output(session->consumer);
2600 /* Ease our life a bit for the next part */
2601 consumer = session->kernel_session->consumer;
2602 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2603 break;
2604 case LTTNG_DOMAIN_JUL:
2605 case LTTNG_DOMAIN_LOG4J:
2606 case LTTNG_DOMAIN_UST:
2607 DBG3("Copying tracing session consumer output in UST session");
2608 if (session->ust_session->consumer) {
2609 consumer_destroy_output(session->ust_session->consumer);
2610 }
2611 session->ust_session->consumer =
2612 consumer_copy_output(session->consumer);
2613 /* Ease our life a bit for the next part */
2614 consumer = session->ust_session->consumer;
2615 dir_name = DEFAULT_UST_TRACE_DIR;
2616 break;
2617 default:
2618 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2619 goto error;
2620 }
2621
2622 /* Append correct directory to subdir */
2623 strncat(consumer->subdir, dir_name,
2624 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2625 DBG3("Copy session consumer subdir %s", consumer->subdir);
2626
2627 ret = LTTNG_OK;
2628
2629 error:
2630 return ret;
2631 }
2632
2633 /*
2634 * Create an UST session and add it to the session ust list.
2635 *
2636 * Should *NOT* be called with RCU read-side lock held.
2637 */
2638 static int create_ust_session(struct ltt_session *session,
2639 struct lttng_domain *domain)
2640 {
2641 int ret;
2642 struct ltt_ust_session *lus = NULL;
2643
2644 assert(session);
2645 assert(domain);
2646 assert(session->consumer);
2647
2648 switch (domain->type) {
2649 case LTTNG_DOMAIN_JUL:
2650 case LTTNG_DOMAIN_LOG4J:
2651 case LTTNG_DOMAIN_UST:
2652 break;
2653 default:
2654 ERR("Unknown UST domain on create session %d", domain->type);
2655 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2656 goto error;
2657 }
2658
2659 DBG("Creating UST session");
2660
2661 lus = trace_ust_create_session(session->id);
2662 if (lus == NULL) {
2663 ret = LTTNG_ERR_UST_SESS_FAIL;
2664 goto error;
2665 }
2666
2667 lus->uid = session->uid;
2668 lus->gid = session->gid;
2669 lus->output_traces = session->output_traces;
2670 lus->snapshot_mode = session->snapshot_mode;
2671 lus->live_timer_interval = session->live_timer;
2672 session->ust_session = lus;
2673
2674 /* Copy session output to the newly created UST session */
2675 ret = copy_session_consumer(domain->type, session);
2676 if (ret != LTTNG_OK) {
2677 goto error;
2678 }
2679
2680 return LTTNG_OK;
2681
2682 error:
2683 free(lus);
2684 session->ust_session = NULL;
2685 return ret;
2686 }
2687
2688 /*
2689 * Create a kernel tracer session then create the default channel.
2690 */
2691 static int create_kernel_session(struct ltt_session *session)
2692 {
2693 int ret;
2694
2695 DBG("Creating kernel session");
2696
2697 ret = kernel_create_session(session, kernel_tracer_fd);
2698 if (ret < 0) {
2699 ret = LTTNG_ERR_KERN_SESS_FAIL;
2700 goto error;
2701 }
2702
2703 /* Code flow safety */
2704 assert(session->kernel_session);
2705
2706 /* Copy session output to the newly created Kernel session */
2707 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2708 if (ret != LTTNG_OK) {
2709 goto error;
2710 }
2711
2712 /* Create directory(ies) on local filesystem. */
2713 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2714 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2715 ret = run_as_mkdir_recursive(
2716 session->kernel_session->consumer->dst.trace_path,
2717 S_IRWXU | S_IRWXG, session->uid, session->gid);
2718 if (ret < 0) {
2719 if (ret != -EEXIST) {
2720 ERR("Trace directory creation error");
2721 goto error;
2722 }
2723 }
2724 }
2725
2726 session->kernel_session->uid = session->uid;
2727 session->kernel_session->gid = session->gid;
2728 session->kernel_session->output_traces = session->output_traces;
2729 session->kernel_session->snapshot_mode = session->snapshot_mode;
2730
2731 return LTTNG_OK;
2732
2733 error:
2734 trace_kernel_destroy_session(session->kernel_session);
2735 session->kernel_session = NULL;
2736 return ret;
2737 }
2738
2739 /*
2740 * Count number of session permitted by uid/gid.
2741 */
2742 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2743 {
2744 unsigned int i = 0;
2745 struct ltt_session *session;
2746
2747 DBG("Counting number of available session for UID %d GID %d",
2748 uid, gid);
2749 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2750 /*
2751 * Only list the sessions the user can control.
2752 */
2753 if (!session_access_ok(session, uid, gid)) {
2754 continue;
2755 }
2756 i++;
2757 }
2758 return i;
2759 }
2760
2761 /*
2762 * Process the command requested by the lttng client within the command
2763 * context structure. This function make sure that the return structure (llm)
2764 * is set and ready for transmission before returning.
2765 *
2766 * Return any error encountered or 0 for success.
2767 *
2768 * "sock" is only used for special-case var. len data.
2769 *
2770 * Should *NOT* be called with RCU read-side lock held.
2771 */
2772 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2773 int *sock_error)
2774 {
2775 int ret = LTTNG_OK;
2776 int need_tracing_session = 1;
2777 int need_domain;
2778
2779 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2780
2781 *sock_error = 0;
2782
2783 switch (cmd_ctx->lsm->cmd_type) {
2784 case LTTNG_CREATE_SESSION:
2785 case LTTNG_CREATE_SESSION_SNAPSHOT:
2786 case LTTNG_CREATE_SESSION_LIVE:
2787 case LTTNG_DESTROY_SESSION:
2788 case LTTNG_LIST_SESSIONS:
2789 case LTTNG_LIST_DOMAINS:
2790 case LTTNG_START_TRACE:
2791 case LTTNG_STOP_TRACE:
2792 case LTTNG_DATA_PENDING:
2793 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2794 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2795 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2796 case LTTNG_SNAPSHOT_RECORD:
2797 case LTTNG_SAVE_SESSION:
2798 need_domain = 0;
2799 break;
2800 default:
2801 need_domain = 1;
2802 }
2803
2804 if (opt_no_kernel && need_domain
2805 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2806 if (!is_root) {
2807 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2808 } else {
2809 ret = LTTNG_ERR_KERN_NA;
2810 }
2811 goto error;
2812 }
2813
2814 /* Deny register consumer if we already have a spawned consumer. */
2815 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2816 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2817 if (kconsumer_data.pid > 0) {
2818 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2819 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2820 goto error;
2821 }
2822 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2823 }
2824
2825 /*
2826 * Check for command that don't needs to allocate a returned payload. We do
2827 * this here so we don't have to make the call for no payload at each
2828 * command.
2829 */
2830 switch(cmd_ctx->lsm->cmd_type) {
2831 case LTTNG_LIST_SESSIONS:
2832 case LTTNG_LIST_TRACEPOINTS:
2833 case LTTNG_LIST_TRACEPOINT_FIELDS:
2834 case LTTNG_LIST_DOMAINS:
2835 case LTTNG_LIST_CHANNELS:
2836 case LTTNG_LIST_EVENTS:
2837 case LTTNG_LIST_SYSCALLS:
2838 break;
2839 default:
2840 /* Setup lttng message with no payload */
2841 ret = setup_lttng_msg(cmd_ctx, 0);
2842 if (ret < 0) {
2843 /* This label does not try to unlock the session */
2844 goto init_setup_error;
2845 }
2846 }
2847
2848 /* Commands that DO NOT need a session. */
2849 switch (cmd_ctx->lsm->cmd_type) {
2850 case LTTNG_CREATE_SESSION:
2851 case LTTNG_CREATE_SESSION_SNAPSHOT:
2852 case LTTNG_CREATE_SESSION_LIVE:
2853 case LTTNG_CALIBRATE:
2854 case LTTNG_LIST_SESSIONS:
2855 case LTTNG_LIST_TRACEPOINTS:
2856 case LTTNG_LIST_SYSCALLS:
2857 case LTTNG_LIST_TRACEPOINT_FIELDS:
2858 case LTTNG_SAVE_SESSION:
2859 need_tracing_session = 0;
2860 break;
2861 default:
2862 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2863 /*
2864 * We keep the session list lock across _all_ commands
2865 * for now, because the per-session lock does not
2866 * handle teardown properly.
2867 */
2868 session_lock_list();
2869 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2870 if (cmd_ctx->session == NULL) {
2871 ret = LTTNG_ERR_SESS_NOT_FOUND;
2872 goto error;
2873 } else {
2874 /* Acquire lock for the session */
2875 session_lock(cmd_ctx->session);
2876 }
2877 break;
2878 }
2879
2880 /*
2881 * Commands that need a valid session but should NOT create one if none
2882 * exists. Instead of creating one and destroying it when the command is
2883 * handled, process that right before so we save some round trip in useless
2884 * code path.
2885 */
2886 switch (cmd_ctx->lsm->cmd_type) {
2887 case LTTNG_DISABLE_CHANNEL:
2888 case LTTNG_DISABLE_EVENT:
2889 switch (cmd_ctx->lsm->domain.type) {
2890 case LTTNG_DOMAIN_KERNEL:
2891 if (!cmd_ctx->session->kernel_session) {
2892 ret = LTTNG_ERR_NO_CHANNEL;
2893 goto error;
2894 }
2895 break;
2896 case LTTNG_DOMAIN_JUL:
2897 case LTTNG_DOMAIN_LOG4J:
2898 case LTTNG_DOMAIN_UST:
2899 if (!cmd_ctx->session->ust_session) {
2900 ret = LTTNG_ERR_NO_CHANNEL;
2901 goto error;
2902 }
2903 break;
2904 default:
2905 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2906 goto error;
2907 }
2908 default:
2909 break;
2910 }
2911
2912 if (!need_domain) {
2913 goto skip_domain;
2914 }
2915
2916 /*
2917 * Check domain type for specific "pre-action".
2918 */
2919 switch (cmd_ctx->lsm->domain.type) {
2920 case LTTNG_DOMAIN_KERNEL:
2921 if (!is_root) {
2922 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2923 goto error;
2924 }
2925
2926 /* Kernel tracer check */
2927 if (kernel_tracer_fd == -1) {
2928 /* Basically, load kernel tracer modules */
2929 ret = init_kernel_tracer();
2930 if (ret != 0) {
2931 goto error;
2932 }
2933 }
2934
2935 /* Consumer is in an ERROR state. Report back to client */
2936 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2937 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2938 goto error;
2939 }
2940
2941 /* Need a session for kernel command */
2942 if (need_tracing_session) {
2943 if (cmd_ctx->session->kernel_session == NULL) {
2944 ret = create_kernel_session(cmd_ctx->session);
2945 if (ret < 0) {
2946 ret = LTTNG_ERR_KERN_SESS_FAIL;
2947 goto error;
2948 }
2949 }
2950
2951 /* Start the kernel consumer daemon */
2952 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2953 if (kconsumer_data.pid == 0 &&
2954 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2955 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2956 ret = start_consumerd(&kconsumer_data);
2957 if (ret < 0) {
2958 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2959 goto error;
2960 }
2961 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2962 } else {
2963 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2964 }
2965
2966 /*
2967 * The consumer was just spawned so we need to add the socket to
2968 * the consumer output of the session if exist.
2969 */
2970 ret = consumer_create_socket(&kconsumer_data,
2971 cmd_ctx->session->kernel_session->consumer);
2972 if (ret < 0) {
2973 goto error;
2974 }
2975 }
2976
2977 break;
2978 case LTTNG_DOMAIN_JUL:
2979 case LTTNG_DOMAIN_LOG4J:
2980 case LTTNG_DOMAIN_UST:
2981 {
2982 if (!ust_app_supported()) {
2983 ret = LTTNG_ERR_NO_UST;
2984 goto error;
2985 }
2986 /* Consumer is in an ERROR state. Report back to client */
2987 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2988 ret = LTTNG_ERR_NO_USTCONSUMERD;
2989 goto error;
2990 }
2991
2992 if (need_tracing_session) {
2993 /* Create UST session if none exist. */
2994 if (cmd_ctx->session->ust_session == NULL) {
2995 ret = create_ust_session(cmd_ctx->session,
2996 &cmd_ctx->lsm->domain);
2997 if (ret != LTTNG_OK) {
2998 goto error;
2999 }
3000 }
3001
3002 /* Start the UST consumer daemons */
3003 /* 64-bit */
3004 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3005 if (consumerd64_bin[0] != '\0' &&
3006 ustconsumer64_data.pid == 0 &&
3007 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3008 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3009 ret = start_consumerd(&ustconsumer64_data);
3010 if (ret < 0) {
3011 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3012 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3013 goto error;
3014 }
3015
3016 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3017 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3018 } else {
3019 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3020 }
3021
3022 /*
3023 * Setup socket for consumer 64 bit. No need for atomic access
3024 * since it was set above and can ONLY be set in this thread.
3025 */
3026 ret = consumer_create_socket(&ustconsumer64_data,
3027 cmd_ctx->session->ust_session->consumer);
3028 if (ret < 0) {
3029 goto error;
3030 }
3031
3032 /* 32-bit */
3033 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3034 if (consumerd32_bin[0] != '\0' &&
3035 ustconsumer32_data.pid == 0 &&
3036 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3037 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3038 ret = start_consumerd(&ustconsumer32_data);
3039 if (ret < 0) {
3040 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3041 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3042 goto error;
3043 }
3044
3045 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3046 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3047 } else {
3048 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3049 }
3050
3051 /*
3052 * Setup socket for consumer 64 bit. No need for atomic access
3053 * since it was set above and can ONLY be set in this thread.
3054 */
3055 ret = consumer_create_socket(&ustconsumer32_data,
3056 cmd_ctx->session->ust_session->consumer);
3057 if (ret < 0) {
3058 goto error;
3059 }
3060 }
3061 break;
3062 }
3063 default:
3064 break;
3065 }
3066 skip_domain:
3067
3068 /* Validate consumer daemon state when start/stop trace command */
3069 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3070 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3071 switch (cmd_ctx->lsm->domain.type) {
3072 case LTTNG_DOMAIN_JUL:
3073 case LTTNG_DOMAIN_LOG4J:
3074 case LTTNG_DOMAIN_UST:
3075 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3076 ret = LTTNG_ERR_NO_USTCONSUMERD;
3077 goto error;
3078 }
3079 break;
3080 case LTTNG_DOMAIN_KERNEL:
3081 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3082 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3083 goto error;
3084 }
3085 break;
3086 }
3087 }
3088
3089 /*
3090 * Check that the UID or GID match that of the tracing session.
3091 * The root user can interact with all sessions.
3092 */
3093 if (need_tracing_session) {
3094 if (!session_access_ok(cmd_ctx->session,
3095 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3096 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3097 ret = LTTNG_ERR_EPERM;
3098 goto error;
3099 }
3100 }
3101
3102 /*
3103 * Send relayd information to consumer as soon as we have a domain and a
3104 * session defined.
3105 */
3106 if (cmd_ctx->session && need_domain) {
3107 /*
3108 * Setup relayd if not done yet. If the relayd information was already
3109 * sent to the consumer, this call will gracefully return.
3110 */
3111 ret = cmd_setup_relayd(cmd_ctx->session);
3112 if (ret != LTTNG_OK) {
3113 goto error;
3114 }
3115 }
3116
3117 /* Process by command type */
3118 switch (cmd_ctx->lsm->cmd_type) {
3119 case LTTNG_ADD_CONTEXT:
3120 {
3121 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3122 cmd_ctx->lsm->u.context.channel_name,
3123 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
3124 break;
3125 }
3126 case LTTNG_DISABLE_CHANNEL:
3127 {
3128 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3129 cmd_ctx->lsm->u.disable.channel_name);
3130 break;
3131 }
3132 case LTTNG_DISABLE_EVENT:
3133 {
3134 /* FIXME: passing packed structure to non-packed pointer */
3135 /* TODO: handle filter */
3136 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3137 cmd_ctx->lsm->u.disable.channel_name,
3138 &cmd_ctx->lsm->u.disable.event);
3139 break;
3140 }
3141 case LTTNG_ENABLE_CHANNEL:
3142 {
3143 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3144 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3145 break;
3146 }
3147 case LTTNG_ENABLE_EVENT:
3148 {
3149 struct lttng_event_exclusion *exclusion = NULL;
3150 struct lttng_filter_bytecode *bytecode = NULL;
3151 char *filter_expression = NULL;
3152
3153 /* Handle exclusion events and receive it from the client. */
3154 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3155 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3156
3157 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3158 (count * LTTNG_SYMBOL_NAME_LEN));
3159 if (!exclusion) {
3160 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3161 goto error;
3162 }
3163
3164 DBG("Receiving var len exclusion event list from client ...");
3165 exclusion->count = count;
3166 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3167 count * LTTNG_SYMBOL_NAME_LEN);
3168 if (ret <= 0) {
3169 DBG("Nothing recv() from client var len data... continuing");
3170 *sock_error = 1;
3171 free(exclusion);
3172 ret = LTTNG_ERR_EXCLUSION_INVAL;
3173 goto error;
3174 }
3175 }
3176
3177 /* Get filter expression from client. */
3178 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3179 size_t expression_len =
3180 cmd_ctx->lsm->u.enable.expression_len;
3181
3182 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3183 ret = LTTNG_ERR_FILTER_INVAL;
3184 free(exclusion);
3185 goto error;
3186 }
3187
3188 filter_expression = zmalloc(expression_len);
3189 if (!filter_expression) {
3190 free(exclusion);
3191 ret = LTTNG_ERR_FILTER_NOMEM;
3192 goto error;
3193 }
3194
3195 /* Receive var. len. data */
3196 DBG("Receiving var len filter's expression from client ...");
3197 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3198 expression_len);
3199 if (ret <= 0) {
3200 DBG("Nothing recv() from client car len data... continuing");
3201 *sock_error = 1;
3202 free(filter_expression);
3203 free(exclusion);
3204 ret = LTTNG_ERR_FILTER_INVAL;
3205 goto error;
3206 }
3207 }
3208
3209 /* Handle filter and get bytecode from client. */
3210 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3211 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3212
3213 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3214 ret = LTTNG_ERR_FILTER_INVAL;
3215 free(filter_expression);
3216 free(exclusion);
3217 goto error;
3218 }
3219
3220 bytecode = zmalloc(bytecode_len);
3221 if (!bytecode) {
3222 free(filter_expression);
3223 free(exclusion);
3224 ret = LTTNG_ERR_FILTER_NOMEM;
3225 goto error;
3226 }
3227
3228 /* Receive var. len. data */
3229 DBG("Receiving var len filter's bytecode from client ...");
3230 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3231 if (ret <= 0) {
3232 DBG("Nothing recv() from client car len data... continuing");
3233 *sock_error = 1;
3234 free(filter_expression);
3235 free(bytecode);
3236 free(exclusion);
3237 ret = LTTNG_ERR_FILTER_INVAL;
3238 goto error;
3239 }
3240
3241 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3242 free(filter_expression);
3243 free(bytecode);
3244 free(exclusion);
3245 ret = LTTNG_ERR_FILTER_INVAL;
3246 goto error;
3247 }
3248 }
3249
3250 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3251 cmd_ctx->lsm->u.enable.channel_name,
3252 &cmd_ctx->lsm->u.enable.event,
3253 filter_expression, bytecode, exclusion,
3254 kernel_poll_pipe[1]);
3255 break;
3256 }
3257 case LTTNG_LIST_TRACEPOINTS:
3258 {
3259 struct lttng_event *events;
3260 ssize_t nb_events;
3261
3262 session_lock_list();
3263 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3264 session_unlock_list();
3265 if (nb_events < 0) {
3266 /* Return value is a negative lttng_error_code. */
3267 ret = -nb_events;
3268 goto error;
3269 }
3270
3271 /*
3272 * Setup lttng message with payload size set to the event list size in
3273 * bytes and then copy list into the llm payload.
3274 */
3275 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3276 if (ret < 0) {
3277 free(events);
3278 goto setup_error;
3279 }
3280
3281 /* Copy event list into message payload */
3282 memcpy(cmd_ctx->llm->payload, events,
3283 sizeof(struct lttng_event) * nb_events);
3284
3285 free(events);
3286
3287 ret = LTTNG_OK;
3288 break;
3289 }
3290 case LTTNG_LIST_TRACEPOINT_FIELDS:
3291 {
3292 struct lttng_event_field *fields;
3293 ssize_t nb_fields;
3294
3295 session_lock_list();
3296 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3297 &fields);
3298 session_unlock_list();
3299 if (nb_fields < 0) {
3300 /* Return value is a negative lttng_error_code. */
3301 ret = -nb_fields;
3302 goto error;
3303 }
3304
3305 /*
3306 * Setup lttng message with payload size set to the event list size in
3307 * bytes and then copy list into the llm payload.
3308 */
3309 ret = setup_lttng_msg(cmd_ctx,
3310 sizeof(struct lttng_event_field) * nb_fields);
3311 if (ret < 0) {
3312 free(fields);
3313 goto setup_error;
3314 }
3315
3316 /* Copy event list into message payload */
3317 memcpy(cmd_ctx->llm->payload, fields,
3318 sizeof(struct lttng_event_field) * nb_fields);
3319
3320 free(fields);
3321
3322 ret = LTTNG_OK;
3323 break;
3324 }
3325 case LTTNG_LIST_SYSCALLS:
3326 {
3327 struct lttng_event *events;
3328 ssize_t nb_events;
3329
3330 nb_events = cmd_list_syscalls(&events);
3331 if (nb_events < 0) {
3332 /* Return value is a negative lttng_error_code. */
3333 ret = -nb_events;
3334 goto error;
3335 }
3336
3337 /*
3338 * Setup lttng message with payload size set to the event list size in
3339 * bytes and then copy list into the llm payload.
3340 */
3341 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3342 if (ret < 0) {
3343 free(events);
3344 goto setup_error;
3345 }
3346
3347 /* Copy event list into message payload */
3348 memcpy(cmd_ctx->llm->payload, events,
3349 sizeof(struct lttng_event) * nb_events);
3350
3351 free(events);
3352
3353 ret = LTTNG_OK;
3354 break;
3355 }
3356 case LTTNG_SET_CONSUMER_URI:
3357 {
3358 size_t nb_uri, len;
3359 struct lttng_uri *uris;
3360
3361 nb_uri = cmd_ctx->lsm->u.uri.size;
3362 len = nb_uri * sizeof(struct lttng_uri);
3363
3364 if (nb_uri == 0) {
3365 ret = LTTNG_ERR_INVALID;
3366 goto error;
3367 }
3368
3369 uris = zmalloc(len);
3370 if (uris == NULL) {
3371 ret = LTTNG_ERR_FATAL;
3372 goto error;
3373 }
3374
3375 /* Receive variable len data */
3376 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3377 ret = lttcomm_recv_unix_sock(sock, uris, len);
3378 if (ret <= 0) {
3379 DBG("No URIs received from client... continuing");
3380 *sock_error = 1;
3381 ret = LTTNG_ERR_SESSION_FAIL;
3382 free(uris);
3383 goto error;
3384 }
3385
3386 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3387 nb_uri, uris);
3388 if (ret != LTTNG_OK) {
3389 free(uris);
3390 goto error;
3391 }
3392
3393 /*
3394 * XXX: 0 means that this URI should be applied on the session. Should
3395 * be a DOMAIN enuam.
3396 */
3397 if (cmd_ctx->lsm->domain.type == 0) {
3398 /* Add the URI for the UST session if a consumer is present. */
3399 if (cmd_ctx->session->ust_session &&
3400 cmd_ctx->session->ust_session->consumer) {
3401 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3402 nb_uri, uris);
3403 } else if (cmd_ctx->session->kernel_session &&
3404 cmd_ctx->session->kernel_session->consumer) {
3405 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3406 cmd_ctx->session, nb_uri, uris);
3407 }
3408 }
3409
3410 free(uris);
3411
3412 break;
3413 }
3414 case LTTNG_START_TRACE:
3415 {
3416 ret = cmd_start_trace(cmd_ctx->session);
3417 break;
3418 }
3419 case LTTNG_STOP_TRACE:
3420 {
3421 ret = cmd_stop_trace(cmd_ctx->session);
3422 break;
3423 }
3424 case LTTNG_CREATE_SESSION:
3425 {
3426 size_t nb_uri, len;
3427 struct lttng_uri *uris = NULL;
3428
3429 nb_uri = cmd_ctx->lsm->u.uri.size;
3430 len = nb_uri * sizeof(struct lttng_uri);
3431
3432 if (nb_uri > 0) {
3433 uris = zmalloc(len);
3434 if (uris == NULL) {
3435 ret = LTTNG_ERR_FATAL;
3436 goto error;
3437 }
3438
3439 /* Receive variable len data */
3440 DBG("Waiting for %zu URIs from client ...", nb_uri);
3441 ret = lttcomm_recv_unix_sock(sock, uris, len);
3442 if (ret <= 0) {
3443 DBG("No URIs received from client... continuing");
3444 *sock_error = 1;
3445 ret = LTTNG_ERR_SESSION_FAIL;
3446 free(uris);
3447 goto error;
3448 }
3449
3450 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3451 DBG("Creating session with ONE network URI is a bad call");
3452 ret = LTTNG_ERR_SESSION_FAIL;
3453 free(uris);
3454 goto error;
3455 }
3456 }
3457
3458 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3459 &cmd_ctx->creds, 0);
3460
3461 free(uris);
3462
3463 break;
3464 }
3465 case LTTNG_DESTROY_SESSION:
3466 {
3467 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3468
3469 /* Set session to NULL so we do not unlock it after free. */
3470 cmd_ctx->session = NULL;
3471 break;
3472 }
3473 case LTTNG_LIST_DOMAINS:
3474 {
3475 ssize_t nb_dom;
3476 struct lttng_domain *domains;
3477
3478 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3479 if (nb_dom < 0) {
3480 /* Return value is a negative lttng_error_code. */
3481 ret = -nb_dom;
3482 goto error;
3483 }
3484
3485 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3486 if (ret < 0) {
3487 free(domains);
3488 goto setup_error;
3489 }
3490
3491 /* Copy event list into message payload */
3492 memcpy(cmd_ctx->llm->payload, domains,
3493 nb_dom * sizeof(struct lttng_domain));
3494
3495 free(domains);
3496
3497 ret = LTTNG_OK;
3498 break;
3499 }
3500 case LTTNG_LIST_CHANNELS:
3501 {
3502 int nb_chan;
3503 struct lttng_channel *channels = NULL;
3504
3505 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3506 cmd_ctx->session, &channels);
3507 if (nb_chan < 0) {
3508 /* Return value is a negative lttng_error_code. */
3509 ret = -nb_chan;
3510 goto error;
3511 }
3512
3513 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3514 if (ret < 0) {
3515 free(channels);
3516 goto setup_error;
3517 }
3518
3519 /* Copy event list into message payload */
3520 memcpy(cmd_ctx->llm->payload, channels,
3521 nb_chan * sizeof(struct lttng_channel));
3522
3523 free(channels);
3524
3525 ret = LTTNG_OK;
3526 break;
3527 }
3528 case LTTNG_LIST_EVENTS:
3529 {
3530 ssize_t nb_event;
3531 struct lttng_event *events = NULL;
3532
3533 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3534 cmd_ctx->lsm->u.list.channel_name, &events);
3535 if (nb_event < 0) {
3536 /* Return value is a negative lttng_error_code. */
3537 ret = -nb_event;
3538 goto error;
3539 }
3540
3541 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3542 if (ret < 0) {
3543 free(events);
3544 goto setup_error;
3545 }
3546
3547 /* Copy event list into message payload */
3548 memcpy(cmd_ctx->llm->payload, events,
3549 nb_event * sizeof(struct lttng_event));
3550
3551 free(events);
3552
3553 ret = LTTNG_OK;
3554 break;
3555 }
3556 case LTTNG_LIST_SESSIONS:
3557 {
3558 unsigned int nr_sessions;
3559
3560 session_lock_list();
3561 nr_sessions = lttng_sessions_count(
3562 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3563 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3564
3565 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3566 if (ret < 0) {
3567 session_unlock_list();
3568 goto setup_error;
3569 }
3570
3571 /* Filled the session array */
3572 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3573 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3574 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3575
3576 session_unlock_list();
3577
3578 ret = LTTNG_OK;
3579 break;
3580 }
3581 case LTTNG_CALIBRATE:
3582 {
3583 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3584 &cmd_ctx->lsm->u.calibrate);
3585 break;
3586 }
3587 case LTTNG_REGISTER_CONSUMER:
3588 {
3589 struct consumer_data *cdata;
3590
3591 switch (cmd_ctx->lsm->domain.type) {
3592 case LTTNG_DOMAIN_KERNEL:
3593 cdata = &kconsumer_data;
3594 break;
3595 default:
3596 ret = LTTNG_ERR_UND;
3597 goto error;
3598 }
3599
3600 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3601 cmd_ctx->lsm->u.reg.path, cdata);
3602 break;
3603 }
3604 case LTTNG_DATA_PENDING:
3605 {
3606 ret = cmd_data_pending(cmd_ctx->session);
3607 break;
3608 }
3609 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3610 {
3611 struct lttcomm_lttng_output_id reply;
3612
3613 ret = cmd_snapshot_add_output(cmd_ctx->session,
3614 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3615 if (ret != LTTNG_OK) {
3616 goto error;
3617 }
3618
3619 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3620 if (ret < 0) {
3621 goto setup_error;
3622 }
3623
3624 /* Copy output list into message payload */
3625 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3626 ret = LTTNG_OK;
3627 break;
3628 }
3629 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3630 {
3631 ret = cmd_snapshot_del_output(cmd_ctx->session,
3632 &cmd_ctx->lsm->u.snapshot_output.output);
3633 break;
3634 }
3635 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3636 {
3637 ssize_t nb_output;
3638 struct lttng_snapshot_output *outputs = NULL;
3639
3640 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3641 if (nb_output < 0) {
3642 ret = -nb_output;
3643 goto error;
3644 }
3645
3646 ret = setup_lttng_msg(cmd_ctx,
3647 nb_output * sizeof(struct lttng_snapshot_output));
3648 if (ret < 0) {
3649 free(outputs);
3650 goto setup_error;
3651 }
3652
3653 if (outputs) {
3654 /* Copy output list into message payload */
3655 memcpy(cmd_ctx->llm->payload, outputs,
3656 nb_output * sizeof(struct lttng_snapshot_output));
3657 free(outputs);
3658 }
3659
3660 ret = LTTNG_OK;
3661 break;
3662 }
3663 case LTTNG_SNAPSHOT_RECORD:
3664 {
3665 ret = cmd_snapshot_record(cmd_ctx->session,
3666 &cmd_ctx->lsm->u.snapshot_record.output,
3667 cmd_ctx->lsm->u.snapshot_record.wait);
3668 break;
3669 }
3670 case LTTNG_CREATE_SESSION_SNAPSHOT:
3671 {
3672 size_t nb_uri, len;
3673 struct lttng_uri *uris = NULL;
3674
3675 nb_uri = cmd_ctx->lsm->u.uri.size;
3676 len = nb_uri * sizeof(struct lttng_uri);
3677
3678 if (nb_uri > 0) {
3679 uris = zmalloc(len);
3680 if (uris == NULL) {
3681 ret = LTTNG_ERR_FATAL;
3682 goto error;
3683 }
3684
3685 /* Receive variable len data */
3686 DBG("Waiting for %zu URIs from client ...", nb_uri);
3687 ret = lttcomm_recv_unix_sock(sock, uris, len);
3688 if (ret <= 0) {
3689 DBG("No URIs received from client... continuing");
3690 *sock_error = 1;
3691 ret = LTTNG_ERR_SESSION_FAIL;
3692 free(uris);
3693 goto error;
3694 }
3695
3696 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3697 DBG("Creating session with ONE network URI is a bad call");
3698 ret = LTTNG_ERR_SESSION_FAIL;
3699 free(uris);
3700 goto error;
3701 }
3702 }
3703
3704 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3705 nb_uri, &cmd_ctx->creds);
3706 free(uris);
3707 break;
3708 }
3709 case LTTNG_CREATE_SESSION_LIVE:
3710 {
3711 size_t nb_uri, len;
3712 struct lttng_uri *uris = NULL;
3713
3714 nb_uri = cmd_ctx->lsm->u.uri.size;
3715 len = nb_uri * sizeof(struct lttng_uri);
3716
3717 if (nb_uri > 0) {
3718 uris = zmalloc(len);
3719 if (uris == NULL) {
3720 ret = LTTNG_ERR_FATAL;
3721 goto error;
3722 }
3723
3724 /* Receive variable len data */
3725 DBG("Waiting for %zu URIs from client ...", nb_uri);
3726 ret = lttcomm_recv_unix_sock(sock, uris, len);
3727 if (ret <= 0) {
3728 DBG("No URIs received from client... continuing");
3729 *sock_error = 1;
3730 ret = LTTNG_ERR_SESSION_FAIL;
3731 free(uris);
3732 goto error;
3733 }
3734
3735 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3736 DBG("Creating session with ONE network URI is a bad call");
3737 ret = LTTNG_ERR_SESSION_FAIL;
3738 free(uris);
3739 goto error;
3740 }
3741 }
3742
3743 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3744 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3745 free(uris);
3746 break;
3747 }
3748 case LTTNG_SAVE_SESSION:
3749 {
3750 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
3751 &cmd_ctx->creds);
3752 break;
3753 }
3754 default:
3755 ret = LTTNG_ERR_UND;
3756 break;
3757 }
3758
3759 error:
3760 if (cmd_ctx->llm == NULL) {
3761 DBG("Missing llm structure. Allocating one.");
3762 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3763 goto setup_error;
3764 }
3765 }
3766 /* Set return code */
3767 cmd_ctx->llm->ret_code = ret;
3768 setup_error:
3769 if (cmd_ctx->session) {
3770 session_unlock(cmd_ctx->session);
3771 }
3772 if (need_tracing_session) {
3773 session_unlock_list();
3774 }
3775 init_setup_error:
3776 return ret;
3777 }
3778
3779 /*
3780 * Thread managing health check socket.
3781 */
3782 static void *thread_manage_health(void *data)
3783 {
3784 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3785 uint32_t revents, nb_fd;
3786 struct lttng_poll_event events;
3787 struct health_comm_msg msg;
3788 struct health_comm_reply reply;
3789
3790 DBG("[thread] Manage health check started");
3791
3792 rcu_register_thread();
3793
3794 /* We might hit an error path before this is created. */
3795 lttng_poll_init(&events);
3796
3797 /* Create unix socket */
3798 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3799 if (sock < 0) {
3800 ERR("Unable to create health check Unix socket");
3801 ret = -1;
3802 goto error;
3803 }
3804
3805 if (is_root) {
3806 /* lttng health client socket path permissions */
3807 ret = chown(health_unix_sock_path, 0,
3808 utils_get_group_id(tracing_group_name));
3809 if (ret < 0) {
3810 ERR("Unable to set group on %s", health_unix_sock_path);
3811 PERROR("chown");
3812 ret = -1;
3813 goto error;
3814 }
3815
3816 ret = chmod(health_unix_sock_path,
3817 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3818 if (ret < 0) {
3819 ERR("Unable to set permissions on %s", health_unix_sock_path);
3820 PERROR("chmod");
3821 ret = -1;
3822 goto error;
3823 }
3824 }
3825
3826 /*
3827 * Set the CLOEXEC flag. Return code is useless because either way, the
3828 * show must go on.
3829 */
3830 (void) utils_set_fd_cloexec(sock);
3831
3832 ret = lttcomm_listen_unix_sock(sock);
3833 if (ret < 0) {
3834 goto error;
3835 }
3836
3837 /*
3838 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3839 * more will be added to this poll set.
3840 */
3841 ret = sessiond_set_thread_pollset(&events, 2);
3842 if (ret < 0) {
3843 goto error;
3844 }
3845
3846 /* Add the application registration socket */
3847 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3848 if (ret < 0) {
3849 goto error;
3850 }
3851
3852 sessiond_notify_ready();
3853
3854 while (1) {
3855 DBG("Health check ready");
3856
3857 /* Inifinite blocking call, waiting for transmission */
3858 restart:
3859 ret = lttng_poll_wait(&events, -1);
3860 if (ret < 0) {
3861 /*
3862 * Restart interrupted system call.
3863 */
3864 if (errno == EINTR) {
3865 goto restart;
3866 }
3867 goto error;
3868 }
3869
3870 nb_fd = ret;
3871
3872 for (i = 0; i < nb_fd; i++) {
3873 /* Fetch once the poll data */
3874 revents = LTTNG_POLL_GETEV(&events, i);
3875 pollfd = LTTNG_POLL_GETFD(&events, i);
3876
3877 /* Thread quit pipe has been closed. Killing thread. */
3878 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3879 if (ret) {
3880 err = 0;
3881 goto exit;
3882 }
3883
3884 /* Event on the registration socket */
3885 if (pollfd == sock) {
3886 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3887 ERR("Health socket poll error");
3888 goto error;
3889 }
3890 }
3891 }
3892
3893 new_sock = lttcomm_accept_unix_sock(sock);
3894 if (new_sock < 0) {
3895 goto error;
3896 }
3897
3898 /*
3899 * Set the CLOEXEC flag. Return code is useless because either way, the
3900 * show must go on.
3901 */
3902 (void) utils_set_fd_cloexec(new_sock);
3903
3904 DBG("Receiving data from client for health...");
3905 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3906 if (ret <= 0) {
3907 DBG("Nothing recv() from client... continuing");
3908 ret = close(new_sock);
3909 if (ret) {
3910 PERROR("close");
3911 }
3912 new_sock = -1;
3913 continue;
3914 }
3915
3916 rcu_thread_online();
3917
3918 memset(&reply, 0, sizeof(reply));
3919 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3920 /*
3921 * health_check_state returns 0 if health is
3922 * bad.
3923 */
3924 if (!health_check_state(health_sessiond, i)) {
3925 reply.ret_code |= 1ULL << i;
3926 }
3927 }
3928
3929 DBG2("Health check return value %" PRIx64, reply.ret_code);
3930
3931 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3932 if (ret < 0) {
3933 ERR("Failed to send health data back to client");
3934 }
3935
3936 /* End of transmission */
3937 ret = close(new_sock);
3938 if (ret) {
3939 PERROR("close");
3940 }
3941 new_sock = -1;
3942 }
3943
3944 exit:
3945 error:
3946 if (err) {
3947 ERR("Health error occurred in %s", __func__);
3948 }
3949 DBG("Health check thread dying");
3950 unlink(health_unix_sock_path);
3951 if (sock >= 0) {
3952 ret = close(sock);
3953 if (ret) {
3954 PERROR("close");
3955 }
3956 }
3957
3958 lttng_poll_clean(&events);
3959
3960 rcu_unregister_thread();
3961 return NULL;
3962 }
3963
3964 /*
3965 * This thread manage all clients request using the unix client socket for
3966 * communication.
3967 */
3968 static void *thread_manage_clients(void *data)
3969 {
3970 int sock = -1, ret, i, pollfd, err = -1;
3971 int sock_error;
3972 uint32_t revents, nb_fd;
3973 struct command_ctx *cmd_ctx = NULL;
3974 struct lttng_poll_event events;
3975
3976 DBG("[thread] Manage client started");
3977
3978 rcu_register_thread();
3979
3980 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3981
3982 health_code_update();
3983
3984 ret = lttcomm_listen_unix_sock(client_sock);
3985 if (ret < 0) {
3986 goto error_listen;
3987 }
3988
3989 /*
3990 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3991 * more will be added to this poll set.
3992 */
3993 ret = sessiond_set_thread_pollset(&events, 2);
3994 if (ret < 0) {
3995 goto error_create_poll;
3996 }
3997
3998 /* Add the application registration socket */
3999 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4000 if (ret < 0) {
4001 goto error;
4002 }
4003
4004 sessiond_notify_ready();
4005 ret = sem_post(&load_info->message_thread_ready);
4006 if (ret) {
4007 PERROR("sem_post message_thread_ready");
4008 goto error;
4009 }
4010
4011 /* This testpoint is after we signal readiness to the parent. */
4012 if (testpoint(sessiond_thread_manage_clients)) {
4013 goto error;
4014 }
4015
4016 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4017 goto error;
4018 }
4019
4020 health_code_update();
4021
4022 while (1) {
4023 DBG("Accepting client command ...");
4024
4025 /* Inifinite blocking call, waiting for transmission */
4026 restart:
4027 health_poll_entry();
4028 ret = lttng_poll_wait(&events, -1);
4029 health_poll_exit();
4030 if (ret < 0) {
4031 /*
4032 * Restart interrupted system call.
4033 */
4034 if (errno == EINTR) {
4035 goto restart;
4036 }
4037 goto error;
4038 }
4039
4040 nb_fd = ret;
4041
4042 for (i = 0; i < nb_fd; i++) {
4043 /* Fetch once the poll data */
4044 revents = LTTNG_POLL_GETEV(&events, i);
4045 pollfd = LTTNG_POLL_GETFD(&events, i);
4046
4047 health_code_update();
4048
4049 /* Thread quit pipe has been closed. Killing thread. */
4050 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4051 if (ret) {
4052 err = 0;
4053 goto exit;
4054 }
4055
4056 /* Event on the registration socket */
4057 if (pollfd == client_sock) {
4058 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4059 ERR("Client socket poll error");
4060 goto error;
4061 }
4062 }
4063 }
4064
4065 DBG("Wait for client response");
4066
4067 health_code_update();
4068
4069 sock = lttcomm_accept_unix_sock(client_sock);
4070 if (sock < 0) {
4071 goto error;
4072 }
4073
4074 /*
4075 * Set the CLOEXEC flag. Return code is useless because either way, the
4076 * show must go on.
4077 */
4078 (void) utils_set_fd_cloexec(sock);
4079
4080 /* Set socket option for credentials retrieval */
4081 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4082 if (ret < 0) {
4083 goto error;
4084 }
4085
4086 /* Allocate context command to process the client request */
4087 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4088 if (cmd_ctx == NULL) {
4089 PERROR("zmalloc cmd_ctx");
4090 goto error;
4091 }
4092
4093 /* Allocate data buffer for reception */
4094 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4095 if (cmd_ctx->lsm == NULL) {
4096 PERROR("zmalloc cmd_ctx->lsm");
4097 goto error;
4098 }
4099
4100 cmd_ctx->llm = NULL;
4101 cmd_ctx->session = NULL;
4102
4103 health_code_update();
4104
4105 /*
4106 * Data is received from the lttng client. The struct
4107 * lttcomm_session_msg (lsm) contains the command and data request of
4108 * the client.
4109 */
4110 DBG("Receiving data from client ...");
4111 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4112 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4113 if (ret <= 0) {
4114 DBG("Nothing recv() from client... continuing");
4115 ret = close(sock);
4116 if (ret) {
4117 PERROR("close");
4118 }
4119 sock = -1;
4120 clean_command_ctx(&cmd_ctx);
4121 continue;
4122 }
4123
4124 health_code_update();
4125
4126 // TODO: Validate cmd_ctx including sanity check for
4127 // security purpose.
4128
4129 rcu_thread_online();
4130 /*
4131 * This function dispatch the work to the kernel or userspace tracer
4132 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4133 * informations for the client. The command context struct contains
4134 * everything this function may needs.
4135 */
4136 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4137 rcu_thread_offline();
4138 if (ret < 0) {
4139 ret = close(sock);
4140 if (ret) {
4141 PERROR("close");
4142 }
4143 sock = -1;
4144 /*
4145 * TODO: Inform client somehow of the fatal error. At
4146 * this point, ret < 0 means that a zmalloc failed
4147 * (ENOMEM). Error detected but still accept
4148 * command, unless a socket error has been
4149 * detected.
4150 */
4151 clean_command_ctx(&cmd_ctx);
4152 continue;
4153 }
4154
4155 health_code_update();
4156
4157 DBG("Sending response (size: %d, retcode: %s)",
4158 cmd_ctx->lttng_msg_size,
4159 lttng_strerror(-cmd_ctx->llm->ret_code));
4160 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4161 if (ret < 0) {
4162 ERR("Failed to send data back to client");
4163 }
4164
4165 /* End of transmission */
4166 ret = close(sock);
4167 if (ret) {
4168 PERROR("close");
4169 }
4170 sock = -1;
4171
4172 clean_command_ctx(&cmd_ctx);
4173
4174 health_code_update();
4175 }
4176
4177 exit:
4178 error:
4179 if (sock >= 0) {
4180 ret = close(sock);
4181 if (ret) {
4182 PERROR("close");
4183 }
4184 }
4185
4186 lttng_poll_clean(&events);
4187 clean_command_ctx(&cmd_ctx);
4188
4189 error_listen:
4190 error_create_poll:
4191 unlink(client_unix_sock_path);
4192 if (client_sock >= 0) {
4193 ret = close(client_sock);
4194 if (ret) {
4195 PERROR("close");
4196 }
4197 }
4198
4199 if (err) {
4200 health_error();
4201 ERR("Health error occurred in %s", __func__);
4202 }
4203
4204 health_unregister(health_sessiond);
4205
4206 DBG("Client thread dying");
4207
4208 rcu_unregister_thread();
4209 return NULL;
4210 }
4211
4212
4213 /*
4214 * usage function on stderr
4215 */
4216 static void usage(void)
4217 {
4218 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4219 fprintf(stderr, " -h, --help Display this usage.\n");
4220 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4221 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4222 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4223 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4224 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4225 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4226 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4227 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4228 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4229 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4230 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4231 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4232 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4233 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4234 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4235 fprintf(stderr, " -V, --version Show version number.\n");
4236 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4237 fprintf(stderr, " -q, --quiet No output at all.\n");
4238 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4239 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4240 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4241 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4242 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4243 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4244 fprintf(stderr, " -l --load PATH Load session configuration\n");
4245 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4246 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4247 }
4248
4249 /*
4250 * Take an option from the getopt output and set it in the right variable to be
4251 * used later.
4252 *
4253 * Return 0 on success else a negative value.
4254 */
4255 static int set_option(int opt, const char *arg, const char *optname)
4256 {
4257 int ret = 0;
4258
4259 if (arg && arg[0] == '\0') {
4260 /*
4261 * This only happens if the value is read from daemon config
4262 * file. This means the option requires an argument and the
4263 * configuration file contains a line such as:
4264 * my_option =
4265 */
4266 ret = -EINVAL;
4267 goto end;
4268 }
4269
4270 switch (opt) {
4271 case 0:
4272 fprintf(stderr, "option %s", optname);
4273 if (arg) {
4274 fprintf(stderr, " with arg %s\n", arg);
4275 }
4276 break;
4277 case 'c':
4278 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4279 break;
4280 case 'a':
4281 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4282 break;
4283 case 'd':
4284 opt_daemon = 1;
4285 break;
4286 case 'b':
4287 opt_background = 1;
4288 break;
4289 case 'g':
4290 /*
4291 * If the override option is set, the pointer points to a
4292 * *non* const thus freeing it even though the variable type is
4293 * set to const.
4294 */
4295 if (tracing_group_name_override) {
4296 free((void *) tracing_group_name);
4297 }
4298 tracing_group_name = strdup(arg);
4299 if (!tracing_group_name) {
4300 perror("strdup");
4301 ret = -ENOMEM;
4302 }
4303 tracing_group_name_override = 1;
4304 break;
4305 case 'h':
4306 usage();
4307 exit(EXIT_FAILURE);
4308 case 'V':
4309 fprintf(stdout, "%s\n", VERSION);
4310 exit(EXIT_SUCCESS);
4311 case 'S':
4312 opt_sig_parent = 1;
4313 break;
4314 case 'E':
4315 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4316 break;
4317 case 'C':
4318 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4319 break;
4320 case 'F':
4321 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4322 break;
4323 case 'D':
4324 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4325 break;
4326 case 'H':
4327 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4328 break;
4329 case 'G':
4330 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4331 break;
4332 case 'N':
4333 opt_no_kernel = 1;
4334 break;
4335 case 'q':
4336 lttng_opt_quiet = 1;
4337 break;
4338 case 'v':
4339 /* Verbose level can increase using multiple -v */
4340 if (arg) {
4341 /* Value obtained from config file */
4342 lttng_opt_verbose = config_parse_value(arg);
4343 } else {
4344 /* -v used on command line */
4345 lttng_opt_verbose++;
4346 }
4347 /* Clamp value to [0, 3] */
4348 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4349 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4350 break;
4351 case 'Z':
4352 if (arg) {
4353 opt_verbose_consumer = config_parse_value(arg);
4354 } else {
4355 opt_verbose_consumer += 1;
4356 }
4357 break;
4358 case 'u':
4359 if (consumerd32_bin_override) {
4360 free((void *) consumerd32_bin);
4361 }
4362 consumerd32_bin = strdup(arg);
4363 if (!consumerd32_bin) {
4364 perror("strdup");
4365 ret = -ENOMEM;
4366 }
4367 consumerd32_bin_override = 1;
4368 break;
4369 case 'U':
4370 if (consumerd32_libdir_override) {
4371 free((void *) consumerd32_libdir);
4372 }
4373 consumerd32_libdir = strdup(arg);
4374 if (!consumerd32_libdir) {
4375 perror("strdup");
4376 ret = -ENOMEM;
4377 }
4378 consumerd32_libdir_override = 1;
4379 break;
4380 case 't':
4381 if (consumerd64_bin_override) {
4382 free((void *) consumerd64_bin);
4383 }
4384 consumerd64_bin = strdup(arg);
4385 if (!consumerd64_bin) {
4386 perror("strdup");
4387 ret = -ENOMEM;
4388 }
4389 consumerd64_bin_override = 1;
4390 break;
4391 case 'T':
4392 if (consumerd64_libdir_override) {
4393 free((void *) consumerd64_libdir);
4394 }
4395 consumerd64_libdir = strdup(arg);
4396 if (!consumerd64_libdir) {
4397 perror("strdup");
4398 ret = -ENOMEM;
4399 }
4400 consumerd64_libdir_override = 1;
4401 break;
4402 case 'p':
4403 free(opt_pidfile);
4404 opt_pidfile = strdup(arg);
4405 if (!opt_pidfile) {
4406 perror("strdup");
4407 ret = -ENOMEM;
4408 }
4409 break;
4410 case 'J': /* Agent TCP port. */
4411 {
4412 unsigned long v;
4413
4414 errno = 0;
4415 v = strtoul(arg, NULL, 0);
4416 if (errno != 0 || !isdigit(arg[0])) {
4417 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4418 return -1;
4419 }
4420 if (v == 0 || v >= 65535) {
4421 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4422 return -1;
4423 }
4424 agent_tcp_port = (uint32_t) v;
4425 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4426 break;
4427 }
4428 case 'l':
4429 free(opt_load_session_path);
4430 opt_load_session_path = strdup(arg);
4431 if (!opt_load_session_path) {
4432 perror("strdup");
4433 ret = -ENOMEM;
4434 }
4435 break;
4436 case 'P': /* probe modules list */
4437 free(kmod_probes_list);
4438 kmod_probes_list = strdup(arg);
4439 if (!kmod_probes_list) {
4440 perror("strdup");
4441 ret = -ENOMEM;
4442 }
4443 break;
4444 case 'e':
4445 free(kmod_extra_probes_list);
4446 kmod_extra_probes_list = strdup(arg);
4447 if (!kmod_extra_probes_list) {
4448 perror("strdup");
4449 ret = -ENOMEM;
4450 }
4451 break;
4452 case 'f':
4453 /* This is handled in set_options() thus silent break. */
4454 break;
4455 default:
4456 /* Unknown option or other error.
4457 * Error is printed by getopt, just return */
4458 ret = -1;
4459 }
4460
4461 if (ret == -EINVAL) {
4462 const char *opt_name = "unknown";
4463 int i;
4464
4465 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4466 i++) {
4467 if (opt == long_options[i].val) {
4468 opt_name = long_options[i].name;
4469 break;
4470 }
4471 }
4472
4473 WARN("Invalid argument provided for option \"%s\", using default value.",
4474 opt_name);
4475 }
4476 end:
4477 return ret;
4478 }
4479
4480 /*
4481 * config_entry_handler_cb used to handle options read from a config file.
4482 * See config_entry_handler_cb comment in common/config/config.h for the
4483 * return value conventions.
4484 */
4485 static int config_entry_handler(const struct config_entry *entry, void *unused)
4486 {
4487 int ret = 0, i;
4488
4489 if (!entry || !entry->name || !entry->value) {
4490 ret = -EINVAL;
4491 goto end;
4492 }
4493
4494 /* Check if the option is to be ignored */
4495 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4496 if (!strcmp(entry->name, config_ignore_options[i])) {
4497 goto end;
4498 }
4499 }
4500
4501 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4502 i++) {
4503
4504 /* Ignore if not fully matched. */
4505 if (strcmp(entry->name, long_options[i].name)) {
4506 continue;
4507 }
4508
4509 /*
4510 * If the option takes no argument on the command line, we have to
4511 * check if the value is "true". We support non-zero numeric values,
4512 * true, on and yes.
4513 */
4514 if (!long_options[i].has_arg) {
4515 ret = config_parse_value(entry->value);
4516 if (ret <= 0) {
4517 if (ret) {
4518 WARN("Invalid configuration value \"%s\" for option %s",
4519 entry->value, entry->name);
4520 }
4521 /* False, skip boolean config option. */
4522 goto end;
4523 }
4524 }
4525
4526 ret = set_option(long_options[i].val, entry->value, entry->name);
4527 goto end;
4528 }
4529
4530 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
4531
4532 end:
4533 return ret;
4534 }
4535
4536 /*
4537 * daemon configuration loading and argument parsing
4538 */
4539 static int set_options(int argc, char **argv)
4540 {
4541 int ret = 0, c = 0, option_index = 0;
4542 int orig_optopt = optopt, orig_optind = optind;
4543 char *optstring;
4544 const char *config_path = NULL;
4545
4546 optstring = utils_generate_optstring(long_options,
4547 sizeof(long_options) / sizeof(struct option));
4548 if (!optstring) {
4549 ret = -ENOMEM;
4550 goto end;
4551 }
4552
4553 /* Check for the --config option */
4554 while ((c = getopt_long(argc, argv, optstring, long_options,
4555 &option_index)) != -1) {
4556 if (c == '?') {
4557 ret = -EINVAL;
4558 goto end;
4559 } else if (c != 'f') {
4560 /* if not equal to --config option. */
4561 continue;
4562 }
4563
4564 config_path = utils_expand_path(optarg);
4565 if (!config_path) {
4566 ERR("Failed to resolve path: %s", optarg);
4567 }
4568 }
4569
4570 ret = config_get_section_entries(config_path, config_section_name,
4571 config_entry_handler, NULL);
4572 if (ret) {
4573 if (ret > 0) {
4574 ERR("Invalid configuration option at line %i", ret);
4575 ret = -1;
4576 }
4577 goto end;
4578 }
4579
4580 /* Reset getopt's global state */
4581 optopt = orig_optopt;
4582 optind = orig_optind;
4583 while (1) {
4584 c = getopt_long(argc, argv, optstring, long_options, &option_index);
4585 if (c == -1) {
4586 break;
4587 }
4588
4589 ret = set_option(c, optarg, long_options[option_index].name);
4590 if (ret < 0) {
4591 break;
4592 }
4593 }
4594
4595 end:
4596 free(optstring);
4597 return ret;
4598 }
4599
4600 /*
4601 * Creates the two needed socket by the daemon.
4602 * apps_sock - The communication socket for all UST apps.
4603 * client_sock - The communication of the cli tool (lttng).
4604 */
4605 static int init_daemon_socket(void)
4606 {
4607 int ret = 0;
4608 mode_t old_umask;
4609
4610 old_umask = umask(0);
4611
4612 /* Create client tool unix socket */
4613 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4614 if (client_sock < 0) {
4615 ERR("Create unix sock failed: %s", client_unix_sock_path);
4616 ret = -1;
4617 goto end;
4618 }
4619
4620 /* Set the cloexec flag */
4621 ret = utils_set_fd_cloexec(client_sock);
4622 if (ret < 0) {
4623 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4624 "Continuing but note that the consumer daemon will have a "
4625 "reference to this socket on exec()", client_sock);
4626 }
4627
4628 /* File permission MUST be 660 */
4629 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4630 if (ret < 0) {
4631 ERR("Set file permissions failed: %s", client_unix_sock_path);
4632 PERROR("chmod");
4633 goto end;
4634 }
4635
4636 /* Create the application unix socket */
4637 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4638 if (apps_sock < 0) {
4639 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4640 ret = -1;
4641 goto end;
4642 }
4643
4644 /* Set the cloexec flag */
4645 ret = utils_set_fd_cloexec(apps_sock);
4646 if (ret < 0) {
4647 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4648 "Continuing but note that the consumer daemon will have a "
4649 "reference to this socket on exec()", apps_sock);
4650 }
4651
4652 /* File permission MUST be 666 */
4653 ret = chmod(apps_unix_sock_path,
4654 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4655 if (ret < 0) {
4656 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4657 PERROR("chmod");
4658 goto end;
4659 }
4660
4661 DBG3("Session daemon client socket %d and application socket %d created",
4662 client_sock, apps_sock);
4663
4664 end:
4665 umask(old_umask);
4666 return ret;
4667 }
4668
4669 /*
4670 * Check if the global socket is available, and if a daemon is answering at the
4671 * other side. If yes, error is returned.
4672 */
4673 static int check_existing_daemon(void)
4674 {
4675 /* Is there anybody out there ? */
4676 if (lttng_session_daemon_alive()) {
4677 return -EEXIST;
4678 }
4679
4680 return 0;
4681 }
4682
4683 /*
4684 * Set the tracing group gid onto the client socket.
4685 *
4686 * Race window between mkdir and chown is OK because we are going from more
4687 * permissive (root.root) to less permissive (root.tracing).
4688 */
4689 static int set_permissions(char *rundir)
4690 {
4691 int ret;
4692 gid_t gid;
4693
4694 gid = utils_get_group_id(tracing_group_name);
4695
4696 /* Set lttng run dir */
4697 ret = chown(rundir, 0, gid);
4698 if (ret < 0) {
4699 ERR("Unable to set group on %s", rundir);
4700 PERROR("chown");
4701 }
4702
4703 /*
4704 * Ensure all applications and tracing group can search the run
4705 * dir. Allow everyone to read the directory, since it does not
4706 * buy us anything to hide its content.
4707 */
4708 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4709 if (ret < 0) {
4710 ERR("Unable to set permissions on %s", rundir);
4711 PERROR("chmod");
4712 }
4713
4714 /* lttng client socket path */
4715 ret = chown(client_unix_sock_path, 0, gid);
4716 if (ret < 0) {
4717 ERR("Unable to set group on %s", client_unix_sock_path);
4718 PERROR("chown");
4719 }
4720
4721 /* kconsumer error socket path */
4722 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4723 if (ret < 0) {
4724 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4725 PERROR("chown");
4726 }
4727
4728 /* 64-bit ustconsumer error socket path */
4729 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4730 if (ret < 0) {
4731 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4732 PERROR("chown");
4733 }
4734
4735 /* 32-bit ustconsumer compat32 error socket path */
4736 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4737 if (ret < 0) {
4738 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4739 PERROR("chown");
4740 }
4741
4742 DBG("All permissions are set");
4743
4744 return ret;
4745 }
4746
4747 /*
4748 * Create the lttng run directory needed for all global sockets and pipe.
4749 */
4750 static int create_lttng_rundir(const char *rundir)
4751 {
4752 int ret;
4753
4754 DBG3("Creating LTTng run directory: %s", rundir);
4755
4756 ret = mkdir(rundir, S_IRWXU);
4757 if (ret < 0) {
4758 if (errno != EEXIST) {
4759 ERR("Unable to create %s", rundir);
4760 goto error;
4761 } else {
4762 ret = 0;
4763 }
4764 }
4765
4766 error:
4767 return ret;
4768 }
4769
4770 /*
4771 * Setup sockets and directory needed by the kconsumerd communication with the
4772 * session daemon.
4773 */
4774 static int set_consumer_sockets(struct consumer_data *consumer_data,
4775 const char *rundir)
4776 {
4777 int ret;
4778 char path[PATH_MAX];
4779
4780 switch (consumer_data->type) {
4781 case LTTNG_CONSUMER_KERNEL:
4782 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4783 break;
4784 case LTTNG_CONSUMER64_UST:
4785 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4786 break;
4787 case LTTNG_CONSUMER32_UST:
4788 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4789 break;
4790 default:
4791 ERR("Consumer type unknown");
4792 ret = -EINVAL;
4793 goto error;
4794 }
4795
4796 DBG2("Creating consumer directory: %s", path);
4797
4798 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4799 if (ret < 0) {
4800 if (errno != EEXIST) {
4801 PERROR("mkdir");
4802 ERR("Failed to create %s", path);
4803 goto error;
4804 }
4805 ret = -1;
4806 }
4807 if (is_root) {
4808 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4809 if (ret < 0) {
4810 ERR("Unable to set group on %s", path);
4811 PERROR("chown");
4812 goto error;
4813 }
4814 }
4815
4816 /* Create the kconsumerd error unix socket */
4817 consumer_data->err_sock =
4818 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4819 if (consumer_data->err_sock < 0) {
4820 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4821 ret = -1;
4822 goto error;
4823 }
4824
4825 /*
4826 * Set the CLOEXEC flag. Return code is useless because either way, the
4827 * show must go on.
4828 */
4829 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4830 if (ret < 0) {
4831 PERROR("utils_set_fd_cloexec");
4832 /* continue anyway */
4833 }
4834
4835 /* File permission MUST be 660 */
4836 ret = chmod(consumer_data->err_unix_sock_path,
4837 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4838 if (ret < 0) {
4839 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4840 PERROR("chmod");
4841 goto error;
4842 }
4843
4844 error:
4845 return ret;
4846 }
4847
4848 /*
4849 * Signal handler for the daemon
4850 *
4851 * Simply stop all worker threads, leaving main() return gracefully after
4852 * joining all threads and calling cleanup().
4853 */
4854 static void sighandler(int sig)
4855 {
4856 switch (sig) {
4857 case SIGPIPE:
4858 DBG("SIGPIPE caught");
4859 return;
4860 case SIGINT:
4861 DBG("SIGINT caught");
4862 stop_threads();
4863 break;
4864 case SIGTERM:
4865 DBG("SIGTERM caught");
4866 stop_threads();
4867 break;
4868 case SIGUSR1:
4869 CMM_STORE_SHARED(recv_child_signal, 1);
4870 break;
4871 default:
4872 break;
4873 }
4874 }
4875
4876 /*
4877 * Setup signal handler for :
4878 * SIGINT, SIGTERM, SIGPIPE
4879 */
4880 static int set_signal_handler(void)
4881 {
4882 int ret = 0;
4883 struct sigaction sa;
4884 sigset_t sigset;
4885
4886 if ((ret = sigemptyset(&sigset)) < 0) {
4887 PERROR("sigemptyset");
4888 return ret;
4889 }
4890
4891 sa.sa_handler = sighandler;
4892 sa.sa_mask = sigset;
4893 sa.sa_flags = 0;
4894 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4895 PERROR("sigaction");
4896 return ret;
4897 }
4898
4899 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4900 PERROR("sigaction");
4901 return ret;
4902 }
4903
4904 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4905 PERROR("sigaction");
4906 return ret;
4907 }
4908
4909 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4910 PERROR("sigaction");
4911 return ret;
4912 }
4913
4914 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4915
4916 return ret;
4917 }
4918
4919 /*
4920 * Set open files limit to unlimited. This daemon can open a large number of
4921 * file descriptors in order to consumer multiple kernel traces.
4922 */
4923 static void set_ulimit(void)
4924 {
4925 int ret;
4926 struct rlimit lim;
4927
4928 /* The kernel does not allowed an infinite limit for open files */
4929 lim.rlim_cur = 65535;
4930 lim.rlim_max = 65535;
4931
4932 ret = setrlimit(RLIMIT_NOFILE, &lim);
4933 if (ret < 0) {
4934 PERROR("failed to set open files limit");
4935 }
4936 }
4937
4938 /*
4939 * Write pidfile using the rundir and opt_pidfile.
4940 */
4941 static void write_pidfile(void)
4942 {
4943 int ret;
4944 char pidfile_path[PATH_MAX];
4945
4946 assert(rundir);
4947
4948 if (opt_pidfile) {
4949 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4950 } else {
4951 /* Build pidfile path from rundir and opt_pidfile. */
4952 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4953 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4954 if (ret < 0) {
4955 PERROR("snprintf pidfile path");
4956 goto error;
4957 }
4958 }
4959
4960 /*
4961 * Create pid file in rundir. Return value is of no importance. The
4962 * execution will continue even though we are not able to write the file.
4963 */
4964 (void) utils_create_pid_file(getpid(), pidfile_path);
4965
4966 error:
4967 return;
4968 }
4969
4970 /*
4971 * Create lockfile using the rundir and return its fd.
4972 */
4973 static int create_lockfile(void)
4974 {
4975 int ret;
4976 char lockfile_path[PATH_MAX];
4977
4978 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
4979 if (ret < 0) {
4980 goto error;
4981 }
4982
4983 ret = utils_create_lock_file(lockfile_path);
4984 error:
4985 return ret;
4986 }
4987
4988 /*
4989 * Write agent TCP port using the rundir.
4990 */
4991 static void write_agent_port(void)
4992 {
4993 int ret;
4994 char path[PATH_MAX];
4995
4996 assert(rundir);
4997
4998 ret = snprintf(path, sizeof(path), "%s/"
4999 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5000 if (ret < 0) {
5001 PERROR("snprintf agent port path");
5002 goto error;
5003 }
5004
5005 /*
5006 * Create TCP agent port file in rundir. Return value is of no importance.
5007 * The execution will continue even though we are not able to write the
5008 * file.
5009 */
5010 (void) utils_create_pid_file(agent_tcp_port, path);
5011
5012 error:
5013 return;
5014 }
5015
5016 /*
5017 * Start the load session thread and dettach from it so the main thread can
5018 * continue. This does not return a value since whatever the outcome, the main
5019 * thread will continue.
5020 */
5021 static void start_load_session_thread(void)
5022 {
5023 int ret;
5024
5025 /* Create session loading thread. */
5026 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
5027 load_info);
5028 if (ret != 0) {
5029 PERROR("pthread_create load_session_thread");
5030 goto error_create;
5031 }
5032
5033 ret = pthread_detach(load_session_thread);
5034 if (ret != 0) {
5035 PERROR("pthread_detach load_session_thread");
5036 }
5037
5038 /* Everything went well so don't cleanup anything. */
5039
5040 error_create:
5041 /* The cleanup() function will destroy the load_info data. */
5042 return;
5043 }
5044
5045 /*
5046 * main
5047 */
5048 int main(int argc, char **argv)
5049 {
5050 int ret = 0;
5051 void *status;
5052 const char *home_path, *env_app_timeout;
5053
5054 init_kernel_workarounds();
5055
5056 rcu_register_thread();
5057
5058 if ((ret = set_signal_handler()) < 0) {
5059 goto error;
5060 }
5061
5062 setup_consumerd_path();
5063
5064 page_size = sysconf(_SC_PAGESIZE);
5065 if (page_size < 0) {
5066 PERROR("sysconf _SC_PAGESIZE");
5067 page_size = LONG_MAX;
5068 WARN("Fallback page size to %ld", page_size);
5069 }
5070
5071 /* Parse arguments and load the daemon configuration file */
5072 progname = argv[0];
5073 if ((ret = set_options(argc, argv)) < 0) {
5074 goto error;
5075 }
5076
5077 /* Daemonize */
5078 if (opt_daemon || opt_background) {
5079 int i;
5080
5081 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5082 !opt_background);
5083 if (ret < 0) {
5084 goto error;
5085 }
5086
5087 /*
5088 * We are in the child. Make sure all other file descriptors are
5089 * closed, in case we are called with more opened file descriptors than
5090 * the standard ones.
5091 */
5092 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5093 (void) close(i);
5094 }
5095 }
5096
5097 /* Create thread quit pipe */
5098 if ((ret = init_thread_quit_pipe()) < 0) {
5099 goto error;
5100 }
5101
5102 /* Check if daemon is UID = 0 */
5103 is_root = !getuid();
5104
5105 if (is_root) {
5106 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5107
5108 /* Create global run dir with root access */
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_GLOBAL_APPS_UNIX_SOCK);
5117 }
5118
5119 if (strlen(client_unix_sock_path) == 0) {
5120 snprintf(client_unix_sock_path, PATH_MAX,
5121 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5122 }
5123
5124 /* Set global SHM for ust */
5125 if (strlen(wait_shm_path) == 0) {
5126 snprintf(wait_shm_path, PATH_MAX,
5127 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5128 }
5129
5130 if (strlen(health_unix_sock_path) == 0) {
5131 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
5132 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5133 }
5134
5135 /* Setup kernel consumerd path */
5136 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5137 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5138 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5139 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5140
5141 DBG2("Kernel consumer err path: %s",
5142 kconsumer_data.err_unix_sock_path);
5143 DBG2("Kernel consumer cmd path: %s",
5144 kconsumer_data.cmd_unix_sock_path);
5145 } else {
5146 home_path = utils_get_home_dir();
5147 if (home_path == NULL) {
5148 /* TODO: Add --socket PATH option */
5149 ERR("Can't get HOME directory for sockets creation.");
5150 ret = -EPERM;
5151 goto error;
5152 }
5153
5154 /*
5155 * Create rundir from home path. This will create something like
5156 * $HOME/.lttng
5157 */
5158 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5159 if (ret < 0) {
5160 ret = -ENOMEM;
5161 goto error;
5162 }
5163
5164 ret = create_lttng_rundir(rundir);
5165 if (ret < 0) {
5166 goto error;
5167 }
5168
5169 if (strlen(apps_unix_sock_path) == 0) {
5170 snprintf(apps_unix_sock_path, PATH_MAX,
5171 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
5172 }
5173
5174 /* Set the cli tool unix socket path */
5175 if (strlen(client_unix_sock_path) == 0) {
5176 snprintf(client_unix_sock_path, PATH_MAX,
5177 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
5178 }
5179
5180 /* Set global SHM for ust */
5181 if (strlen(wait_shm_path) == 0) {
5182 snprintf(wait_shm_path, PATH_MAX,
5183 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
5184 }
5185
5186 /* Set health check Unix path */
5187 if (strlen(health_unix_sock_path) == 0) {
5188 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
5189 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
5190 }
5191 }
5192
5193 lockfile_fd = create_lockfile();
5194 if (lockfile_fd < 0) {
5195 goto error;
5196 }
5197
5198 /* Set consumer initial state */
5199 kernel_consumerd_state = CONSUMER_STOPPED;
5200 ust_consumerd_state = CONSUMER_STOPPED;
5201
5202 DBG("Client socket path %s", client_unix_sock_path);
5203 DBG("Application socket path %s", apps_unix_sock_path);
5204 DBG("Application wait path %s", wait_shm_path);
5205 DBG("LTTng run directory path: %s", rundir);
5206
5207 /* 32 bits consumerd path setup */
5208 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5209 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5210 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5211 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5212
5213 DBG2("UST consumer 32 bits err path: %s",
5214 ustconsumer32_data.err_unix_sock_path);
5215 DBG2("UST consumer 32 bits cmd path: %s",
5216 ustconsumer32_data.cmd_unix_sock_path);
5217
5218 /* 64 bits consumerd path setup */
5219 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5220 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5221 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5222 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5223
5224 DBG2("UST consumer 64 bits err path: %s",
5225 ustconsumer64_data.err_unix_sock_path);
5226 DBG2("UST consumer 64 bits cmd path: %s",
5227 ustconsumer64_data.cmd_unix_sock_path);
5228
5229 /*
5230 * See if daemon already exist.
5231 */
5232 if ((ret = check_existing_daemon()) < 0) {
5233 ERR("Already running daemon.\n");
5234 /*
5235 * We do not goto exit because we must not cleanup()
5236 * because a daemon is already running.
5237 */
5238 goto error;
5239 }
5240
5241 /*
5242 * Init UST app hash table. Alloc hash table before this point since
5243 * cleanup() can get called after that point.
5244 */
5245 ust_app_ht_alloc();
5246
5247 /* Initialize agent domain subsystem. */
5248 if ((ret = agent_setup()) < 0) {
5249 /* ENOMEM at this point. */
5250 goto error;
5251 }
5252
5253 /* After this point, we can safely call cleanup() with "goto exit" */
5254
5255 /*
5256 * These actions must be executed as root. We do that *after* setting up
5257 * the sockets path because we MUST make the check for another daemon using
5258 * those paths *before* trying to set the kernel consumer sockets and init
5259 * kernel tracer.
5260 */
5261 if (is_root) {
5262 ret = set_consumer_sockets(&kconsumer_data, rundir);
5263 if (ret < 0) {
5264 goto exit;
5265 }
5266
5267 /* Setup kernel tracer */
5268 if (!opt_no_kernel) {
5269 init_kernel_tracer();
5270 if (kernel_tracer_fd >= 0) {
5271 ret = syscall_init_table();
5272 if (ret < 0) {
5273 ERR("Unable to populate syscall table. Syscall tracing"
5274 " won't work for this session daemon.");
5275 }
5276 }
5277 }
5278
5279 /* Set ulimit for open files */
5280 set_ulimit();
5281 }
5282 /* init lttng_fd tracking must be done after set_ulimit. */
5283 lttng_fd_init();
5284
5285 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
5286 if (ret < 0) {
5287 goto exit;
5288 }
5289
5290 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
5291 if (ret < 0) {
5292 goto exit;
5293 }
5294
5295 /* Setup the needed unix socket */
5296 if ((ret = init_daemon_socket()) < 0) {
5297 goto exit;
5298 }
5299
5300 /* Set credentials to socket */
5301 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
5302 goto exit;
5303 }
5304
5305 /* Get parent pid if -S, --sig-parent is specified. */
5306 if (opt_sig_parent) {
5307 ppid = getppid();
5308 }
5309
5310 /* Setup the kernel pipe for waking up the kernel thread */
5311 if (is_root && !opt_no_kernel) {
5312 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
5313 goto exit;
5314 }
5315 }
5316
5317 /* Setup the thread ht_cleanup communication pipe. */
5318 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
5319 goto exit;
5320 }
5321
5322 /* Setup the thread apps communication pipe. */
5323 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
5324 goto exit;
5325 }
5326
5327 /* Setup the thread apps notify communication pipe. */
5328 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
5329 goto exit;
5330 }
5331
5332 /* Initialize global buffer per UID and PID registry. */
5333 buffer_reg_init_uid_registry();
5334 buffer_reg_init_pid_registry();
5335
5336 /* Init UST command queue. */
5337 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5338
5339 /*
5340 * Get session list pointer. This pointer MUST NOT be free(). This list is
5341 * statically declared in session.c
5342 */
5343 session_list_ptr = session_get_list();
5344
5345 /* Set up max poll set size */
5346 lttng_poll_set_max_size();
5347
5348 cmd_init();
5349
5350 /* Check for the application socket timeout env variable. */
5351 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5352 if (env_app_timeout) {
5353 app_socket_timeout = atoi(env_app_timeout);
5354 } else {
5355 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5356 }
5357
5358 write_pidfile();
5359 write_agent_port();
5360
5361 /* Initialize communication library */
5362 lttcomm_init();
5363 /* This is to get the TCP timeout value. */
5364 lttcomm_inet_init();
5365
5366 if (load_session_init_data(&load_info) < 0) {
5367 goto exit;
5368 }
5369 load_info->path = opt_load_session_path;
5370
5371 /*
5372 * Initialize the health check subsystem. This call should set the
5373 * appropriate time values.
5374 */
5375 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5376 if (!health_sessiond) {
5377 PERROR("health_app_create error");
5378 goto exit_health_sessiond_cleanup;
5379 }
5380
5381 /* Create thread to clean up RCU hash tables */
5382 ret = pthread_create(&ht_cleanup_thread, NULL,
5383 thread_ht_cleanup, (void *) NULL);
5384 if (ret != 0) {
5385 PERROR("pthread_create ht_cleanup");
5386 goto exit_ht_cleanup;
5387 }
5388
5389 /* Create health-check thread */
5390 ret = pthread_create(&health_thread, NULL,
5391 thread_manage_health, (void *) NULL);
5392 if (ret != 0) {
5393 PERROR("pthread_create health");
5394 goto exit_health;
5395 }
5396
5397 /* Create thread to manage the client socket */
5398 ret = pthread_create(&client_thread, NULL,
5399 thread_manage_clients, (void *) NULL);
5400 if (ret != 0) {
5401 PERROR("pthread_create clients");
5402 goto exit_client;
5403 }
5404
5405 /* Create thread to dispatch registration */
5406 ret = pthread_create(&dispatch_thread, NULL,
5407 thread_dispatch_ust_registration, (void *) NULL);
5408 if (ret != 0) {
5409 PERROR("pthread_create dispatch");
5410 goto exit_dispatch;
5411 }
5412
5413 /* Create thread to manage application registration. */
5414 ret = pthread_create(&reg_apps_thread, NULL,
5415 thread_registration_apps, (void *) NULL);
5416 if (ret != 0) {
5417 PERROR("pthread_create registration");
5418 goto exit_reg_apps;
5419 }
5420
5421 /* Create thread to manage application socket */
5422 ret = pthread_create(&apps_thread, NULL,
5423 thread_manage_apps, (void *) NULL);
5424 if (ret != 0) {
5425 PERROR("pthread_create apps");
5426 goto exit_apps;
5427 }
5428
5429 /* Create thread to manage application notify socket */
5430 ret = pthread_create(&apps_notify_thread, NULL,
5431 ust_thread_manage_notify, (void *) NULL);
5432 if (ret != 0) {
5433 PERROR("pthread_create notify");
5434 goto exit_apps_notify;
5435 }
5436
5437 /* Create agent registration thread. */
5438 ret = pthread_create(&agent_reg_thread, NULL,
5439 agent_thread_manage_registration, (void *) NULL);
5440 if (ret != 0) {
5441 PERROR("pthread_create agent");
5442 goto exit_agent_reg;
5443 }
5444
5445 /* Don't start this thread if kernel tracing is not requested nor root */
5446 if (is_root && !opt_no_kernel) {
5447 /* Create kernel thread to manage kernel event */
5448 ret = pthread_create(&kernel_thread, NULL,
5449 thread_manage_kernel, (void *) NULL);
5450 if (ret != 0) {
5451 PERROR("pthread_create kernel");
5452 goto exit_kernel;
5453 }
5454 }
5455
5456 /* Load possible session(s). */
5457 start_load_session_thread();
5458
5459 if (is_root && !opt_no_kernel) {
5460 ret = pthread_join(kernel_thread, &status);
5461 if (ret != 0) {
5462 PERROR("pthread_join");
5463 goto error; /* join error, exit without cleanup */
5464 }
5465 }
5466
5467 exit_kernel:
5468 ret = pthread_join(agent_reg_thread, &status);
5469 if (ret != 0) {
5470 PERROR("pthread_join agent");
5471 goto error; /* join error, exit without cleanup */
5472 }
5473
5474 exit_agent_reg:
5475 ret = pthread_join(apps_notify_thread, &status);
5476 if (ret != 0) {
5477 PERROR("pthread_join apps notify");
5478 goto error; /* join error, exit without cleanup */
5479 }
5480
5481 exit_apps_notify:
5482 ret = pthread_join(apps_thread, &status);
5483 if (ret != 0) {
5484 PERROR("pthread_join apps");
5485 goto error; /* join error, exit without cleanup */
5486 }
5487
5488
5489 exit_apps:
5490 ret = pthread_join(reg_apps_thread, &status);
5491 if (ret != 0) {
5492 PERROR("pthread_join");
5493 goto error; /* join error, exit without cleanup */
5494 }
5495
5496 exit_reg_apps:
5497 ret = pthread_join(dispatch_thread, &status);
5498 if (ret != 0) {
5499 PERROR("pthread_join");
5500 goto error; /* join error, exit without cleanup */
5501 }
5502
5503 exit_dispatch:
5504 ret = pthread_join(client_thread, &status);
5505 if (ret != 0) {
5506 PERROR("pthread_join");
5507 goto error; /* join error, exit without cleanup */
5508 }
5509
5510 ret = join_consumer_thread(&kconsumer_data);
5511 if (ret != 0) {
5512 PERROR("join_consumer");
5513 goto error; /* join error, exit without cleanup */
5514 }
5515
5516 ret = join_consumer_thread(&ustconsumer32_data);
5517 if (ret != 0) {
5518 PERROR("join_consumer ust32");
5519 goto error; /* join error, exit without cleanup */
5520 }
5521
5522 ret = join_consumer_thread(&ustconsumer64_data);
5523 if (ret != 0) {
5524 PERROR("join_consumer ust64");
5525 goto error; /* join error, exit without cleanup */
5526 }
5527
5528 exit_client:
5529 ret = pthread_join(health_thread, &status);
5530 if (ret != 0) {
5531 PERROR("pthread_join health thread");
5532 goto error; /* join error, exit without cleanup */
5533 }
5534
5535 exit_health:
5536 ret = pthread_join(ht_cleanup_thread, &status);
5537 if (ret != 0) {
5538 PERROR("pthread_join ht cleanup thread");
5539 goto error; /* join error, exit without cleanup */
5540 }
5541 exit_ht_cleanup:
5542 health_app_destroy(health_sessiond);
5543 exit_health_sessiond_cleanup:
5544 exit:
5545 /*
5546 * cleanup() is called when no other thread is running.
5547 */
5548 rcu_thread_online();
5549 cleanup();
5550 rcu_thread_offline();
5551 rcu_unregister_thread();
5552 if (!ret) {
5553 exit(EXIT_SUCCESS);
5554 }
5555 error:
5556 exit(EXIT_FAILURE);
5557 }
This page took 0.179664 seconds and 4 git commands to generate.