9cde946495436f04590339475e96f95b5ea7421c
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <ctype.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
52
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
55 #include "channel.h"
56 #include "cmd.h"
57 #include "consumer.h"
58 #include "context.h"
59 #include "event.h"
60 #include "kernel.h"
61 #include "kernel-consumer.h"
62 #include "modprobe.h"
63 #include "shm.h"
64 #include "ust-ctl.h"
65 #include "ust-consumer.h"
66 #include "utils.h"
67 #include "fd-limit.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
72 #include "save.h"
73 #include "load-session-thread.h"
74 #include "notification-thread.h"
75 #include "notification-thread-commands.h"
76 #include "syscall.h"
77 #include "agent.h"
78 #include "ht-cleanup.h"
79
80 #define CONSUMERD_FILE "lttng-consumerd"
81
82 const char *progname;
83 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
84 static int tracing_group_name_override;
85 static char *opt_pidfile;
86 static int opt_sig_parent;
87 static int opt_verbose_consumer;
88 static int opt_daemon, opt_background;
89 static int opt_no_kernel;
90 static char *opt_load_session_path;
91 static pid_t ppid; /* Parent PID for --sig-parent option */
92 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
93 static char *rundir;
94 static int lockfile_fd = -1;
95
96 /* Set to 1 when a SIGUSR1 signal is received. */
97 static int recv_child_signal;
98
99 /*
100 * Consumer daemon specific control data. Every value not initialized here is
101 * set to 0 by the static definition.
102 */
103 static struct consumer_data kconsumer_data = {
104 .type = LTTNG_CONSUMER_KERNEL,
105 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
106 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
107 .err_sock = -1,
108 .cmd_sock = -1,
109 .channel_monitor_pipe = -1,
110 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
111 .lock = PTHREAD_MUTEX_INITIALIZER,
112 .cond = PTHREAD_COND_INITIALIZER,
113 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
114 };
115 static struct consumer_data ustconsumer64_data = {
116 .type = LTTNG_CONSUMER64_UST,
117 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
118 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
119 .err_sock = -1,
120 .cmd_sock = -1,
121 .channel_monitor_pipe = -1,
122 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
123 .lock = PTHREAD_MUTEX_INITIALIZER,
124 .cond = PTHREAD_COND_INITIALIZER,
125 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
126 };
127 static struct consumer_data ustconsumer32_data = {
128 .type = LTTNG_CONSUMER32_UST,
129 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
130 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
131 .err_sock = -1,
132 .cmd_sock = -1,
133 .channel_monitor_pipe = -1,
134 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
135 .lock = PTHREAD_MUTEX_INITIALIZER,
136 .cond = PTHREAD_COND_INITIALIZER,
137 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
138 };
139
140 /* Command line options */
141 static const struct option long_options[] = {
142 { "client-sock", required_argument, 0, 'c' },
143 { "apps-sock", required_argument, 0, 'a' },
144 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
145 { "kconsumerd-err-sock", required_argument, 0, '\0' },
146 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
147 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
148 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
149 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
150 { "consumerd32-path", required_argument, 0, '\0' },
151 { "consumerd32-libdir", required_argument, 0, '\0' },
152 { "consumerd64-path", required_argument, 0, '\0' },
153 { "consumerd64-libdir", required_argument, 0, '\0' },
154 { "daemonize", no_argument, 0, 'd' },
155 { "background", no_argument, 0, 'b' },
156 { "sig-parent", no_argument, 0, 'S' },
157 { "help", no_argument, 0, 'h' },
158 { "group", required_argument, 0, 'g' },
159 { "version", no_argument, 0, 'V' },
160 { "quiet", no_argument, 0, 'q' },
161 { "verbose", no_argument, 0, 'v' },
162 { "verbose-consumer", no_argument, 0, '\0' },
163 { "no-kernel", no_argument, 0, '\0' },
164 { "pidfile", required_argument, 0, 'p' },
165 { "agent-tcp-port", required_argument, 0, '\0' },
166 { "config", required_argument, 0, 'f' },
167 { "load", required_argument, 0, 'l' },
168 { "kmod-probes", required_argument, 0, '\0' },
169 { "extra-kmod-probes", required_argument, 0, '\0' },
170 { NULL, 0, 0, 0 }
171 };
172
173 /* Command line options to ignore from configuration file */
174 static const char *config_ignore_options[] = { "help", "version", "config" };
175
176 /* Shared between threads */
177 static int dispatch_thread_exit;
178
179 /* Global application Unix socket path */
180 static char apps_unix_sock_path[PATH_MAX];
181 /* Global client Unix socket path */
182 static char client_unix_sock_path[PATH_MAX];
183 /* global wait shm path for UST */
184 static char wait_shm_path[PATH_MAX];
185 /* Global health check unix path */
186 static char health_unix_sock_path[PATH_MAX];
187
188 /* Sockets and FDs */
189 static int client_sock = -1;
190 static int apps_sock = -1;
191 int kernel_tracer_fd = -1;
192 static int kernel_poll_pipe[2] = { -1, -1 };
193
194 /*
195 * Quit pipe for all threads. This permits a single cancellation point
196 * for all threads when receiving an event on the pipe.
197 */
198 static int thread_quit_pipe[2] = { -1, -1 };
199
200 /*
201 * This pipe is used to inform the thread managing application communication
202 * that a command is queued and ready to be processed.
203 */
204 static int apps_cmd_pipe[2] = { -1, -1 };
205
206 int apps_cmd_notify_pipe[2] = { -1, -1 };
207
208 /* Pthread, Mutexes and Semaphores */
209 static pthread_t apps_thread;
210 static pthread_t apps_notify_thread;
211 static pthread_t reg_apps_thread;
212 static pthread_t client_thread;
213 static pthread_t kernel_thread;
214 static pthread_t dispatch_thread;
215 static pthread_t health_thread;
216 static pthread_t ht_cleanup_thread;
217 static pthread_t agent_reg_thread;
218 static pthread_t load_session_thread;
219 static pthread_t notification_thread;
220
221 /*
222 * UST registration command queue. This queue is tied with a futex and uses a N
223 * wakers / 1 waiter implemented and detailed in futex.c/.h
224 *
225 * The thread_registration_apps and thread_dispatch_ust_registration uses this
226 * queue along with the wait/wake scheme. The thread_manage_apps receives down
227 * the line new application socket and monitors it for any I/O error or clean
228 * close that triggers an unregistration of the application.
229 */
230 static struct ust_cmd_queue ust_cmd_queue;
231
232 /*
233 * Pointer initialized before thread creation.
234 *
235 * This points to the tracing session list containing the session count and a
236 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
237 * MUST NOT be taken if you call a public function in session.c.
238 *
239 * The lock is nested inside the structure: session_list_ptr->lock. Please use
240 * session_lock_list and session_unlock_list for lock acquisition.
241 */
242 static struct ltt_session_list *session_list_ptr;
243
244 int ust_consumerd64_fd = -1;
245 int ust_consumerd32_fd = -1;
246
247 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
248 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
249 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
250 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
251 static int consumerd32_bin_override;
252 static int consumerd64_bin_override;
253 static int consumerd32_libdir_override;
254 static int consumerd64_libdir_override;
255
256 static const char *module_proc_lttng = "/proc/lttng";
257
258 /*
259 * Consumer daemon state which is changed when spawning it, killing it or in
260 * case of a fatal error.
261 */
262 enum consumerd_state {
263 CONSUMER_STARTED = 1,
264 CONSUMER_STOPPED = 2,
265 CONSUMER_ERROR = 3,
266 };
267
268 /*
269 * This consumer daemon state is used to validate if a client command will be
270 * able to reach the consumer. If not, the client is informed. For instance,
271 * doing a "lttng start" when the consumer state is set to ERROR will return an
272 * error to the client.
273 *
274 * The following example shows a possible race condition of this scheme:
275 *
276 * consumer thread error happens
277 * client cmd arrives
278 * client cmd checks state -> still OK
279 * consumer thread exit, sets error
280 * client cmd try to talk to consumer
281 * ...
282 *
283 * However, since the consumer is a different daemon, we have no way of making
284 * sure the command will reach it safely even with this state flag. This is why
285 * we consider that up to the state validation during command processing, the
286 * command is safe. After that, we can not guarantee the correctness of the
287 * client request vis-a-vis the consumer.
288 */
289 static enum consumerd_state ust_consumerd_state;
290 static enum consumerd_state kernel_consumerd_state;
291
292 /*
293 * Socket timeout for receiving and sending in seconds.
294 */
295 static int app_socket_timeout;
296
297 /* Set in main() with the current page size. */
298 long page_size;
299
300 /* Application health monitoring */
301 struct health_app *health_sessiond;
302
303 /* Agent TCP port for registration. Used by the agent thread. */
304 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
305
306 /* Am I root or not. */
307 int is_root; /* Set to 1 if the daemon is running as root */
308
309 const char * const config_section_name = "sessiond";
310
311 /* Load session thread information to operate. */
312 struct load_session_thread_data *load_info;
313
314 /* Notification thread handle. */
315 struct notification_thread_handle *notification_thread_handle;
316
317 /* Global hash tables */
318 struct lttng_ht *agent_apps_ht_by_sock = NULL;
319
320 /*
321 * Whether sessiond is ready for commands/notification channel/health check
322 * requests.
323 * NR_LTTNG_SESSIOND_READY must match the number of calls to
324 * sessiond_notify_ready().
325 */
326 #define NR_LTTNG_SESSIOND_READY 4
327 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
328
329 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
330 {
331 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
332 }
333
334 /* Notify parents that we are ready for cmd and health check */
335 LTTNG_HIDDEN
336 void sessiond_notify_ready(void)
337 {
338 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
339 /*
340 * Notify parent pid that we are ready to accept command
341 * for client side. This ppid is the one from the
342 * external process that spawned us.
343 */
344 if (opt_sig_parent) {
345 kill(ppid, SIGUSR1);
346 }
347
348 /*
349 * Notify the parent of the fork() process that we are
350 * ready.
351 */
352 if (opt_daemon || opt_background) {
353 kill(child_ppid, SIGUSR1);
354 }
355 }
356 }
357
358 static
359 void setup_consumerd_path(void)
360 {
361 const char *bin, *libdir;
362
363 /*
364 * Allow INSTALL_BIN_PATH to be used as a target path for the
365 * native architecture size consumer if CONFIG_CONSUMER*_PATH
366 * has not been defined.
367 */
368 #if (CAA_BITS_PER_LONG == 32)
369 if (!consumerd32_bin[0]) {
370 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
371 }
372 if (!consumerd32_libdir[0]) {
373 consumerd32_libdir = INSTALL_LIB_PATH;
374 }
375 #elif (CAA_BITS_PER_LONG == 64)
376 if (!consumerd64_bin[0]) {
377 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
378 }
379 if (!consumerd64_libdir[0]) {
380 consumerd64_libdir = INSTALL_LIB_PATH;
381 }
382 #else
383 #error "Unknown bitness"
384 #endif
385
386 /*
387 * runtime env. var. overrides the build default.
388 */
389 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
390 if (bin) {
391 consumerd32_bin = bin;
392 }
393 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
394 if (bin) {
395 consumerd64_bin = bin;
396 }
397 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
398 if (libdir) {
399 consumerd32_libdir = libdir;
400 }
401 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
402 if (libdir) {
403 consumerd64_libdir = libdir;
404 }
405 }
406
407 static
408 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
409 int *a_pipe)
410 {
411 int ret;
412
413 assert(events);
414
415 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
416 if (ret < 0) {
417 goto error;
418 }
419
420 /* Add quit pipe */
421 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
422 if (ret < 0) {
423 goto error;
424 }
425
426 return 0;
427
428 error:
429 return ret;
430 }
431
432 /*
433 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
434 */
435 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
436 {
437 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
438 }
439
440 /*
441 * Init thread quit pipe.
442 *
443 * Return -1 on error or 0 if all pipes are created.
444 */
445 static int __init_thread_quit_pipe(int *a_pipe)
446 {
447 int ret, i;
448
449 ret = pipe(a_pipe);
450 if (ret < 0) {
451 PERROR("thread quit pipe");
452 goto error;
453 }
454
455 for (i = 0; i < 2; i++) {
456 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
457 if (ret < 0) {
458 PERROR("fcntl");
459 goto error;
460 }
461 }
462
463 error:
464 return ret;
465 }
466
467 static int init_thread_quit_pipe(void)
468 {
469 return __init_thread_quit_pipe(thread_quit_pipe);
470 }
471
472 /*
473 * Stop all threads by closing the thread quit pipe.
474 */
475 static void stop_threads(void)
476 {
477 int ret;
478
479 /* Stopping all threads */
480 DBG("Terminating all threads");
481 ret = notify_thread_pipe(thread_quit_pipe[1]);
482 if (ret < 0) {
483 ERR("write error on thread quit pipe");
484 }
485
486 /* Dispatch thread */
487 CMM_STORE_SHARED(dispatch_thread_exit, 1);
488 futex_nto1_wake(&ust_cmd_queue.futex);
489 }
490
491 /*
492 * Close every consumer sockets.
493 */
494 static void close_consumer_sockets(void)
495 {
496 int ret;
497
498 if (kconsumer_data.err_sock >= 0) {
499 ret = close(kconsumer_data.err_sock);
500 if (ret < 0) {
501 PERROR("kernel consumer err_sock close");
502 }
503 }
504 if (ustconsumer32_data.err_sock >= 0) {
505 ret = close(ustconsumer32_data.err_sock);
506 if (ret < 0) {
507 PERROR("UST consumerd32 err_sock close");
508 }
509 }
510 if (ustconsumer64_data.err_sock >= 0) {
511 ret = close(ustconsumer64_data.err_sock);
512 if (ret < 0) {
513 PERROR("UST consumerd64 err_sock close");
514 }
515 }
516 if (kconsumer_data.cmd_sock >= 0) {
517 ret = close(kconsumer_data.cmd_sock);
518 if (ret < 0) {
519 PERROR("kernel consumer cmd_sock close");
520 }
521 }
522 if (ustconsumer32_data.cmd_sock >= 0) {
523 ret = close(ustconsumer32_data.cmd_sock);
524 if (ret < 0) {
525 PERROR("UST consumerd32 cmd_sock close");
526 }
527 }
528 if (ustconsumer64_data.cmd_sock >= 0) {
529 ret = close(ustconsumer64_data.cmd_sock);
530 if (ret < 0) {
531 PERROR("UST consumerd64 cmd_sock close");
532 }
533 }
534 if (kconsumer_data.channel_monitor_pipe >= 0) {
535 ret = close(kconsumer_data.channel_monitor_pipe);
536 if (ret < 0) {
537 PERROR("kernel consumer channel monitor pipe close");
538 }
539 }
540 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
541 ret = close(ustconsumer32_data.channel_monitor_pipe);
542 if (ret < 0) {
543 PERROR("UST consumerd32 channel monitor pipe close");
544 }
545 }
546 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
547 ret = close(ustconsumer64_data.channel_monitor_pipe);
548 if (ret < 0) {
549 PERROR("UST consumerd64 channel monitor pipe close");
550 }
551 }
552 }
553
554 /*
555 * Generate the full lock file path using the rundir.
556 *
557 * Return the snprintf() return value thus a negative value is an error.
558 */
559 static int generate_lock_file_path(char *path, size_t len)
560 {
561 int ret;
562
563 assert(path);
564 assert(rundir);
565
566 /* Build lockfile path from rundir. */
567 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
568 if (ret < 0) {
569 PERROR("snprintf lockfile path");
570 }
571
572 return ret;
573 }
574
575 /*
576 * Wait on consumer process termination.
577 *
578 * Need to be called with the consumer data lock held or from a context
579 * ensuring no concurrent access to data (e.g: cleanup).
580 */
581 static void wait_consumer(struct consumer_data *consumer_data)
582 {
583 pid_t ret;
584 int status;
585
586 if (consumer_data->pid <= 0) {
587 return;
588 }
589
590 DBG("Waiting for complete teardown of consumerd (PID: %d)",
591 consumer_data->pid);
592 ret = waitpid(consumer_data->pid, &status, 0);
593 if (ret == -1) {
594 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
595 }
596 if (!WIFEXITED(status)) {
597 ERR("consumerd termination with error: %d",
598 WEXITSTATUS(ret));
599 }
600 consumer_data->pid = 0;
601 }
602
603 /*
604 * Cleanup the session daemon's data structures.
605 */
606 static void sessiond_cleanup(void)
607 {
608 int ret;
609 struct ltt_session *sess, *stmp;
610 char path[PATH_MAX];
611
612 DBG("Cleanup sessiond");
613
614 /*
615 * Close the thread quit pipe. It has already done its job,
616 * since we are now called.
617 */
618 utils_close_pipe(thread_quit_pipe);
619
620 /*
621 * If opt_pidfile is undefined, the default file will be wiped when
622 * removing the rundir.
623 */
624 if (opt_pidfile) {
625 ret = remove(opt_pidfile);
626 if (ret < 0) {
627 PERROR("remove pidfile %s", opt_pidfile);
628 }
629 }
630
631 DBG("Removing sessiond and consumerd content of directory %s", rundir);
632
633 /* sessiond */
634 snprintf(path, PATH_MAX,
635 "%s/%s",
636 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
637 DBG("Removing %s", path);
638 (void) unlink(path);
639
640 snprintf(path, PATH_MAX, "%s/%s", rundir,
641 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
642 DBG("Removing %s", path);
643 (void) unlink(path);
644
645 /* kconsumerd */
646 snprintf(path, PATH_MAX,
647 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
648 rundir);
649 DBG("Removing %s", path);
650 (void) unlink(path);
651
652 snprintf(path, PATH_MAX,
653 DEFAULT_KCONSUMERD_PATH,
654 rundir);
655 DBG("Removing directory %s", path);
656 (void) rmdir(path);
657
658 /* ust consumerd 32 */
659 snprintf(path, PATH_MAX,
660 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
661 rundir);
662 DBG("Removing %s", path);
663 (void) unlink(path);
664
665 snprintf(path, PATH_MAX,
666 DEFAULT_USTCONSUMERD32_PATH,
667 rundir);
668 DBG("Removing directory %s", path);
669 (void) rmdir(path);
670
671 /* ust consumerd 64 */
672 snprintf(path, PATH_MAX,
673 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
674 rundir);
675 DBG("Removing %s", path);
676 (void) unlink(path);
677
678 snprintf(path, PATH_MAX,
679 DEFAULT_USTCONSUMERD64_PATH,
680 rundir);
681 DBG("Removing directory %s", path);
682 (void) rmdir(path);
683
684 DBG("Cleaning up all sessions");
685
686 /* Destroy session list mutex */
687 if (session_list_ptr != NULL) {
688 pthread_mutex_destroy(&session_list_ptr->lock);
689
690 /* Cleanup ALL session */
691 cds_list_for_each_entry_safe(sess, stmp,
692 &session_list_ptr->head, list) {
693 cmd_destroy_session(sess, kernel_poll_pipe[1]);
694 }
695 }
696
697 wait_consumer(&kconsumer_data);
698 wait_consumer(&ustconsumer64_data);
699 wait_consumer(&ustconsumer32_data);
700
701 DBG("Cleaning up all agent apps");
702 agent_app_ht_clean();
703
704 DBG("Closing all UST sockets");
705 ust_app_clean_list();
706 buffer_reg_destroy_registries();
707
708 if (is_root && !opt_no_kernel) {
709 DBG2("Closing kernel fd");
710 if (kernel_tracer_fd >= 0) {
711 ret = close(kernel_tracer_fd);
712 if (ret) {
713 PERROR("close");
714 }
715 }
716 DBG("Unloading kernel modules");
717 modprobe_remove_lttng_all();
718 free(syscall_table);
719 }
720
721 close_consumer_sockets();
722
723 if (load_info) {
724 load_session_destroy_data(load_info);
725 free(load_info);
726 }
727
728 /*
729 * Cleanup lock file by deleting it and finaly closing it which will
730 * release the file system lock.
731 */
732 if (lockfile_fd >= 0) {
733 char lockfile_path[PATH_MAX];
734
735 ret = generate_lock_file_path(lockfile_path,
736 sizeof(lockfile_path));
737 if (ret > 0) {
738 ret = remove(lockfile_path);
739 if (ret < 0) {
740 PERROR("remove lock file");
741 }
742 ret = close(lockfile_fd);
743 if (ret < 0) {
744 PERROR("close lock file");
745 }
746 }
747 }
748
749 /*
750 * We do NOT rmdir rundir because there are other processes
751 * using it, for instance lttng-relayd, which can start in
752 * parallel with this teardown.
753 */
754
755 free(rundir);
756 }
757
758 /*
759 * Cleanup the daemon's option data structures.
760 */
761 static void sessiond_cleanup_options(void)
762 {
763 DBG("Cleaning up options");
764
765 /*
766 * If the override option is set, the pointer points to a *non* const
767 * thus freeing it even though the variable type is set to const.
768 */
769 if (tracing_group_name_override) {
770 free((void *) tracing_group_name);
771 }
772 if (consumerd32_bin_override) {
773 free((void *) consumerd32_bin);
774 }
775 if (consumerd64_bin_override) {
776 free((void *) consumerd64_bin);
777 }
778 if (consumerd32_libdir_override) {
779 free((void *) consumerd32_libdir);
780 }
781 if (consumerd64_libdir_override) {
782 free((void *) consumerd64_libdir);
783 }
784
785 free(opt_pidfile);
786 free(opt_load_session_path);
787 free(kmod_probes_list);
788 free(kmod_extra_probes_list);
789
790 run_as_destroy_worker();
791 }
792
793 /*
794 * Send data on a unix socket using the liblttsessiondcomm API.
795 *
796 * Return lttcomm error code.
797 */
798 static int send_unix_sock(int sock, void *buf, size_t len)
799 {
800 /* Check valid length */
801 if (len == 0) {
802 return -1;
803 }
804
805 return lttcomm_send_unix_sock(sock, buf, len);
806 }
807
808 /*
809 * Free memory of a command context structure.
810 */
811 static void clean_command_ctx(struct command_ctx **cmd_ctx)
812 {
813 DBG("Clean command context structure");
814 if (*cmd_ctx) {
815 if ((*cmd_ctx)->llm) {
816 free((*cmd_ctx)->llm);
817 }
818 if ((*cmd_ctx)->lsm) {
819 free((*cmd_ctx)->lsm);
820 }
821 free(*cmd_ctx);
822 *cmd_ctx = NULL;
823 }
824 }
825
826 /*
827 * Notify UST applications using the shm mmap futex.
828 */
829 static int notify_ust_apps(int active)
830 {
831 char *wait_shm_mmap;
832
833 DBG("Notifying applications of session daemon state: %d", active);
834
835 /* See shm.c for this call implying mmap, shm and futex calls */
836 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
837 if (wait_shm_mmap == NULL) {
838 goto error;
839 }
840
841 /* Wake waiting process */
842 futex_wait_update((int32_t *) wait_shm_mmap, active);
843
844 /* Apps notified successfully */
845 return 0;
846
847 error:
848 return -1;
849 }
850
851 /*
852 * Setup the outgoing data buffer for the response (llm) by allocating the
853 * right amount of memory and copying the original information from the lsm
854 * structure.
855 *
856 * Return 0 on success, negative value on error.
857 */
858 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
859 const void *payload_buf, size_t payload_len,
860 const void *cmd_header_buf, size_t cmd_header_len)
861 {
862 int ret = 0;
863 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
864 const size_t cmd_header_offset = header_len;
865 const size_t payload_offset = cmd_header_offset + cmd_header_len;
866 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
867
868 cmd_ctx->llm = zmalloc(total_msg_size);
869
870 if (cmd_ctx->llm == NULL) {
871 PERROR("zmalloc");
872 ret = -ENOMEM;
873 goto end;
874 }
875
876 /* Copy common data */
877 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
878 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
879 cmd_ctx->llm->cmd_header_size = cmd_header_len;
880 cmd_ctx->llm->data_size = payload_len;
881 cmd_ctx->lttng_msg_size = total_msg_size;
882
883 /* Copy command header */
884 if (cmd_header_len) {
885 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
886 cmd_header_len);
887 }
888
889 /* Copy payload */
890 if (payload_len) {
891 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
892 payload_len);
893 }
894
895 end:
896 return ret;
897 }
898
899 /*
900 * Version of setup_lttng_msg() without command header.
901 */
902 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
903 void *payload_buf, size_t payload_len)
904 {
905 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
906 }
907 /*
908 * Update the kernel poll set of all channel fd available over all tracing
909 * session. Add the wakeup pipe at the end of the set.
910 */
911 static int update_kernel_poll(struct lttng_poll_event *events)
912 {
913 int ret;
914 struct ltt_session *session;
915 struct ltt_kernel_channel *channel;
916
917 DBG("Updating kernel poll set");
918
919 session_lock_list();
920 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
921 session_lock(session);
922 if (session->kernel_session == NULL) {
923 session_unlock(session);
924 continue;
925 }
926
927 cds_list_for_each_entry(channel,
928 &session->kernel_session->channel_list.head, list) {
929 /* Add channel fd to the kernel poll set */
930 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
931 if (ret < 0) {
932 session_unlock(session);
933 goto error;
934 }
935 DBG("Channel fd %d added to kernel set", channel->fd);
936 }
937 session_unlock(session);
938 }
939 session_unlock_list();
940
941 return 0;
942
943 error:
944 session_unlock_list();
945 return -1;
946 }
947
948 /*
949 * Find the channel fd from 'fd' over all tracing session. When found, check
950 * for new channel stream and send those stream fds to the kernel consumer.
951 *
952 * Useful for CPU hotplug feature.
953 */
954 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
955 {
956 int ret = 0;
957 struct ltt_session *session;
958 struct ltt_kernel_session *ksess;
959 struct ltt_kernel_channel *channel;
960
961 DBG("Updating kernel streams for channel fd %d", fd);
962
963 session_lock_list();
964 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
965 session_lock(session);
966 if (session->kernel_session == NULL) {
967 session_unlock(session);
968 continue;
969 }
970 ksess = session->kernel_session;
971
972 cds_list_for_each_entry(channel,
973 &ksess->channel_list.head, list) {
974 struct lttng_ht_iter iter;
975 struct consumer_socket *socket;
976
977 if (channel->fd != fd) {
978 continue;
979 }
980 DBG("Channel found, updating kernel streams");
981 ret = kernel_open_channel_stream(channel);
982 if (ret < 0) {
983 goto error;
984 }
985 /* Update the stream global counter */
986 ksess->stream_count_global += ret;
987
988 /*
989 * Have we already sent fds to the consumer? If yes, it
990 * means that tracing is started so it is safe to send
991 * our updated stream fds.
992 */
993 if (ksess->consumer_fds_sent != 1
994 || ksess->consumer == NULL) {
995 ret = -1;
996 goto error;
997 }
998
999 rcu_read_lock();
1000 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
1001 &iter.iter, socket, node.node) {
1002 pthread_mutex_lock(socket->lock);
1003 ret = kernel_consumer_send_channel_stream(socket,
1004 channel, ksess,
1005 session->output_traces ? 1 : 0);
1006 pthread_mutex_unlock(socket->lock);
1007 if (ret < 0) {
1008 rcu_read_unlock();
1009 goto error;
1010 }
1011 }
1012 rcu_read_unlock();
1013 }
1014 session_unlock(session);
1015 }
1016 session_unlock_list();
1017 return ret;
1018
1019 error:
1020 session_unlock(session);
1021 session_unlock_list();
1022 return ret;
1023 }
1024
1025 /*
1026 * For each tracing session, update newly registered apps. The session list
1027 * lock MUST be acquired before calling this.
1028 */
1029 static void update_ust_app(int app_sock)
1030 {
1031 struct ltt_session *sess, *stmp;
1032
1033 /* Consumer is in an ERROR state. Stop any application update. */
1034 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1035 /* Stop the update process since the consumer is dead. */
1036 return;
1037 }
1038
1039 /* For all tracing session(s) */
1040 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1041 struct ust_app *app;
1042
1043 session_lock(sess);
1044 if (!sess->ust_session) {
1045 goto unlock_session;
1046 }
1047
1048 rcu_read_lock();
1049 assert(app_sock >= 0);
1050 app = ust_app_find_by_sock(app_sock);
1051 if (app == NULL) {
1052 /*
1053 * Application can be unregistered before so
1054 * this is possible hence simply stopping the
1055 * update.
1056 */
1057 DBG3("UST app update failed to find app sock %d",
1058 app_sock);
1059 goto unlock_rcu;
1060 }
1061 ust_app_global_update(sess->ust_session, app);
1062 unlock_rcu:
1063 rcu_read_unlock();
1064 unlock_session:
1065 session_unlock(sess);
1066 }
1067 }
1068
1069 /*
1070 * This thread manage event coming from the kernel.
1071 *
1072 * Features supported in this thread:
1073 * -) CPU Hotplug
1074 */
1075 static void *thread_manage_kernel(void *data)
1076 {
1077 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1078 uint32_t revents, nb_fd;
1079 char tmp;
1080 struct lttng_poll_event events;
1081
1082 DBG("[thread] Thread manage kernel started");
1083
1084 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1085
1086 /*
1087 * This first step of the while is to clean this structure which could free
1088 * non NULL pointers so initialize it before the loop.
1089 */
1090 lttng_poll_init(&events);
1091
1092 if (testpoint(sessiond_thread_manage_kernel)) {
1093 goto error_testpoint;
1094 }
1095
1096 health_code_update();
1097
1098 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1099 goto error_testpoint;
1100 }
1101
1102 while (1) {
1103 health_code_update();
1104
1105 if (update_poll_flag == 1) {
1106 /* Clean events object. We are about to populate it again. */
1107 lttng_poll_clean(&events);
1108
1109 ret = sessiond_set_thread_pollset(&events, 2);
1110 if (ret < 0) {
1111 goto error_poll_create;
1112 }
1113
1114 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1115 if (ret < 0) {
1116 goto error;
1117 }
1118
1119 /* This will add the available kernel channel if any. */
1120 ret = update_kernel_poll(&events);
1121 if (ret < 0) {
1122 goto error;
1123 }
1124 update_poll_flag = 0;
1125 }
1126
1127 DBG("Thread kernel polling");
1128
1129 /* Poll infinite value of time */
1130 restart:
1131 health_poll_entry();
1132 ret = lttng_poll_wait(&events, -1);
1133 DBG("Thread kernel return from poll on %d fds",
1134 LTTNG_POLL_GETNB(&events));
1135 health_poll_exit();
1136 if (ret < 0) {
1137 /*
1138 * Restart interrupted system call.
1139 */
1140 if (errno == EINTR) {
1141 goto restart;
1142 }
1143 goto error;
1144 } else if (ret == 0) {
1145 /* Should not happen since timeout is infinite */
1146 ERR("Return value of poll is 0 with an infinite timeout.\n"
1147 "This should not have happened! Continuing...");
1148 continue;
1149 }
1150
1151 nb_fd = ret;
1152
1153 for (i = 0; i < nb_fd; i++) {
1154 /* Fetch once the poll data */
1155 revents = LTTNG_POLL_GETEV(&events, i);
1156 pollfd = LTTNG_POLL_GETFD(&events, i);
1157
1158 health_code_update();
1159
1160 if (!revents) {
1161 /* No activity for this FD (poll implementation). */
1162 continue;
1163 }
1164
1165 /* Thread quit pipe has been closed. Killing thread. */
1166 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1167 if (ret) {
1168 err = 0;
1169 goto exit;
1170 }
1171
1172 /* Check for data on kernel pipe */
1173 if (revents & LPOLLIN) {
1174 if (pollfd == kernel_poll_pipe[0]) {
1175 (void) lttng_read(kernel_poll_pipe[0],
1176 &tmp, 1);
1177 /*
1178 * Ret value is useless here, if this pipe gets any actions an
1179 * update is required anyway.
1180 */
1181 update_poll_flag = 1;
1182 continue;
1183 } else {
1184 /*
1185 * New CPU detected by the kernel. Adding kernel stream to
1186 * kernel session and updating the kernel consumer
1187 */
1188 ret = update_kernel_stream(&kconsumer_data, pollfd);
1189 if (ret < 0) {
1190 continue;
1191 }
1192 break;
1193 }
1194 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1195 update_poll_flag = 1;
1196 continue;
1197 } else {
1198 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1199 goto error;
1200 }
1201 }
1202 }
1203
1204 exit:
1205 error:
1206 lttng_poll_clean(&events);
1207 error_poll_create:
1208 error_testpoint:
1209 utils_close_pipe(kernel_poll_pipe);
1210 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1211 if (err) {
1212 health_error();
1213 ERR("Health error occurred in %s", __func__);
1214 WARN("Kernel thread died unexpectedly. "
1215 "Kernel tracing can continue but CPU hotplug is disabled.");
1216 }
1217 health_unregister(health_sessiond);
1218 DBG("Kernel thread dying");
1219 return NULL;
1220 }
1221
1222 /*
1223 * Signal pthread condition of the consumer data that the thread.
1224 */
1225 static void signal_consumer_condition(struct consumer_data *data, int state)
1226 {
1227 pthread_mutex_lock(&data->cond_mutex);
1228
1229 /*
1230 * The state is set before signaling. It can be any value, it's the waiter
1231 * job to correctly interpret this condition variable associated to the
1232 * consumer pthread_cond.
1233 *
1234 * A value of 0 means that the corresponding thread of the consumer data
1235 * was not started. 1 indicates that the thread has started and is ready
1236 * for action. A negative value means that there was an error during the
1237 * thread bootstrap.
1238 */
1239 data->consumer_thread_is_ready = state;
1240 (void) pthread_cond_signal(&data->cond);
1241
1242 pthread_mutex_unlock(&data->cond_mutex);
1243 }
1244
1245 /*
1246 * This thread manage the consumer error sent back to the session daemon.
1247 */
1248 static void *thread_manage_consumer(void *data)
1249 {
1250 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1251 uint32_t revents, nb_fd;
1252 enum lttcomm_return_code code;
1253 struct lttng_poll_event events;
1254 struct consumer_data *consumer_data = data;
1255 struct consumer_socket *cmd_socket_wrapper = NULL;
1256
1257 DBG("[thread] Manage consumer started");
1258
1259 rcu_register_thread();
1260 rcu_thread_online();
1261
1262 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1263
1264 health_code_update();
1265
1266 /*
1267 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1268 * metadata_sock. Nothing more will be added to this poll set.
1269 */
1270 ret = sessiond_set_thread_pollset(&events, 3);
1271 if (ret < 0) {
1272 goto error_poll;
1273 }
1274
1275 /*
1276 * The error socket here is already in a listening state which was done
1277 * just before spawning this thread to avoid a race between the consumer
1278 * daemon exec trying to connect and the listen() call.
1279 */
1280 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1281 if (ret < 0) {
1282 goto error;
1283 }
1284
1285 health_code_update();
1286
1287 /* Infinite blocking call, waiting for transmission */
1288 restart:
1289 health_poll_entry();
1290
1291 if (testpoint(sessiond_thread_manage_consumer)) {
1292 goto error;
1293 }
1294
1295 ret = lttng_poll_wait(&events, -1);
1296 health_poll_exit();
1297 if (ret < 0) {
1298 /*
1299 * Restart interrupted system call.
1300 */
1301 if (errno == EINTR) {
1302 goto restart;
1303 }
1304 goto error;
1305 }
1306
1307 nb_fd = ret;
1308
1309 for (i = 0; i < nb_fd; i++) {
1310 /* Fetch once the poll data */
1311 revents = LTTNG_POLL_GETEV(&events, i);
1312 pollfd = LTTNG_POLL_GETFD(&events, i);
1313
1314 health_code_update();
1315
1316 if (!revents) {
1317 /* No activity for this FD (poll implementation). */
1318 continue;
1319 }
1320
1321 /* Thread quit pipe has been closed. Killing thread. */
1322 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1323 if (ret) {
1324 err = 0;
1325 goto exit;
1326 }
1327
1328 /* Event on the registration socket */
1329 if (pollfd == consumer_data->err_sock) {
1330 if (revents & LPOLLIN) {
1331 continue;
1332 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1333 ERR("consumer err socket poll error");
1334 goto error;
1335 } else {
1336 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1337 goto error;
1338 }
1339 }
1340 }
1341
1342 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1343 if (sock < 0) {
1344 goto error;
1345 }
1346
1347 /*
1348 * Set the CLOEXEC flag. Return code is useless because either way, the
1349 * show must go on.
1350 */
1351 (void) utils_set_fd_cloexec(sock);
1352
1353 health_code_update();
1354
1355 DBG2("Receiving code from consumer err_sock");
1356
1357 /* Getting status code from kconsumerd */
1358 ret = lttcomm_recv_unix_sock(sock, &code,
1359 sizeof(enum lttcomm_return_code));
1360 if (ret <= 0) {
1361 goto error;
1362 }
1363
1364 health_code_update();
1365 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1366 ERR("consumer error when waiting for SOCK_READY : %s",
1367 lttcomm_get_readable_code(-code));
1368 goto error;
1369 }
1370
1371 /* Connect both command and metadata sockets. */
1372 consumer_data->cmd_sock =
1373 lttcomm_connect_unix_sock(
1374 consumer_data->cmd_unix_sock_path);
1375 consumer_data->metadata_fd =
1376 lttcomm_connect_unix_sock(
1377 consumer_data->cmd_unix_sock_path);
1378 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1379 PERROR("consumer connect cmd socket");
1380 /* On error, signal condition and quit. */
1381 signal_consumer_condition(consumer_data, -1);
1382 goto error;
1383 }
1384
1385 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1386
1387 /* Create metadata socket lock. */
1388 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1389 if (consumer_data->metadata_sock.lock == NULL) {
1390 PERROR("zmalloc pthread mutex");
1391 goto error;
1392 }
1393 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1394
1395 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1396 DBG("Consumer metadata socket ready (fd: %d)",
1397 consumer_data->metadata_fd);
1398
1399 /*
1400 * Remove the consumerd error sock since we've established a connection.
1401 */
1402 ret = lttng_poll_del(&events, consumer_data->err_sock);
1403 if (ret < 0) {
1404 goto error;
1405 }
1406
1407 /* Add new accepted error socket. */
1408 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1409 if (ret < 0) {
1410 goto error;
1411 }
1412
1413 /* Add metadata socket that is successfully connected. */
1414 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1415 LPOLLIN | LPOLLRDHUP);
1416 if (ret < 0) {
1417 goto error;
1418 }
1419
1420 health_code_update();
1421
1422 /*
1423 * Transfer the write-end of the channel monitoring pipe to the
1424 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1425 */
1426 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1427 if (!cmd_socket_wrapper) {
1428 goto error;
1429 }
1430
1431 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1432 consumer_data->channel_monitor_pipe);
1433 if (ret) {
1434 goto error;
1435 }
1436 /* Discard the socket wrapper as it is no longer needed. */
1437 consumer_destroy_socket(cmd_socket_wrapper);
1438 cmd_socket_wrapper = NULL;
1439
1440 /* The thread is completely initialized, signal that it is ready. */
1441 signal_consumer_condition(consumer_data, 1);
1442
1443 /* Infinite blocking call, waiting for transmission */
1444 restart_poll:
1445 while (1) {
1446 health_code_update();
1447
1448 /* Exit the thread because the thread quit pipe has been triggered. */
1449 if (should_quit) {
1450 /* Not a health error. */
1451 err = 0;
1452 goto exit;
1453 }
1454
1455 health_poll_entry();
1456 ret = lttng_poll_wait(&events, -1);
1457 health_poll_exit();
1458 if (ret < 0) {
1459 /*
1460 * Restart interrupted system call.
1461 */
1462 if (errno == EINTR) {
1463 goto restart_poll;
1464 }
1465 goto error;
1466 }
1467
1468 nb_fd = ret;
1469
1470 for (i = 0; i < nb_fd; i++) {
1471 /* Fetch once the poll data */
1472 revents = LTTNG_POLL_GETEV(&events, i);
1473 pollfd = LTTNG_POLL_GETFD(&events, i);
1474
1475 health_code_update();
1476
1477 if (!revents) {
1478 /* No activity for this FD (poll implementation). */
1479 continue;
1480 }
1481
1482 /*
1483 * Thread quit pipe has been triggered, flag that we should stop
1484 * but continue the current loop to handle potential data from
1485 * consumer.
1486 */
1487 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1488
1489 if (pollfd == sock) {
1490 /* Event on the consumerd socket */
1491 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1492 && !(revents & LPOLLIN)) {
1493 ERR("consumer err socket second poll error");
1494 goto error;
1495 }
1496 health_code_update();
1497 /* Wait for any kconsumerd error */
1498 ret = lttcomm_recv_unix_sock(sock, &code,
1499 sizeof(enum lttcomm_return_code));
1500 if (ret <= 0) {
1501 ERR("consumer closed the command socket");
1502 goto error;
1503 }
1504
1505 ERR("consumer return code : %s",
1506 lttcomm_get_readable_code(-code));
1507
1508 goto exit;
1509 } else if (pollfd == consumer_data->metadata_fd) {
1510 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1511 && !(revents & LPOLLIN)) {
1512 ERR("consumer err metadata socket second poll error");
1513 goto error;
1514 }
1515 /* UST metadata requests */
1516 ret = ust_consumer_metadata_request(
1517 &consumer_data->metadata_sock);
1518 if (ret < 0) {
1519 ERR("Handling metadata request");
1520 goto error;
1521 }
1522 }
1523 /* No need for an else branch all FDs are tested prior. */
1524 }
1525 health_code_update();
1526 }
1527
1528 exit:
1529 error:
1530 /*
1531 * We lock here because we are about to close the sockets and some other
1532 * thread might be using them so get exclusive access which will abort all
1533 * other consumer command by other threads.
1534 */
1535 pthread_mutex_lock(&consumer_data->lock);
1536
1537 /* Immediately set the consumerd state to stopped */
1538 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1539 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1540 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1541 consumer_data->type == LTTNG_CONSUMER32_UST) {
1542 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1543 } else {
1544 /* Code flow error... */
1545 assert(0);
1546 }
1547
1548 if (consumer_data->err_sock >= 0) {
1549 ret = close(consumer_data->err_sock);
1550 if (ret) {
1551 PERROR("close");
1552 }
1553 consumer_data->err_sock = -1;
1554 }
1555 if (consumer_data->cmd_sock >= 0) {
1556 ret = close(consumer_data->cmd_sock);
1557 if (ret) {
1558 PERROR("close");
1559 }
1560 consumer_data->cmd_sock = -1;
1561 }
1562 if (consumer_data->metadata_sock.fd_ptr &&
1563 *consumer_data->metadata_sock.fd_ptr >= 0) {
1564 ret = close(*consumer_data->metadata_sock.fd_ptr);
1565 if (ret) {
1566 PERROR("close");
1567 }
1568 }
1569 if (sock >= 0) {
1570 ret = close(sock);
1571 if (ret) {
1572 PERROR("close");
1573 }
1574 }
1575
1576 unlink(consumer_data->err_unix_sock_path);
1577 unlink(consumer_data->cmd_unix_sock_path);
1578 pthread_mutex_unlock(&consumer_data->lock);
1579
1580 /* Cleanup metadata socket mutex. */
1581 if (consumer_data->metadata_sock.lock) {
1582 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1583 free(consumer_data->metadata_sock.lock);
1584 }
1585 lttng_poll_clean(&events);
1586
1587 if (cmd_socket_wrapper) {
1588 consumer_destroy_socket(cmd_socket_wrapper);
1589 }
1590 error_poll:
1591 if (err) {
1592 health_error();
1593 ERR("Health error occurred in %s", __func__);
1594 }
1595 health_unregister(health_sessiond);
1596 DBG("consumer thread cleanup completed");
1597
1598 rcu_thread_offline();
1599 rcu_unregister_thread();
1600
1601 return NULL;
1602 }
1603
1604 /*
1605 * This thread manage application communication.
1606 */
1607 static void *thread_manage_apps(void *data)
1608 {
1609 int i, ret, pollfd, err = -1;
1610 ssize_t size_ret;
1611 uint32_t revents, nb_fd;
1612 struct lttng_poll_event events;
1613
1614 DBG("[thread] Manage application started");
1615
1616 rcu_register_thread();
1617 rcu_thread_online();
1618
1619 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1620
1621 if (testpoint(sessiond_thread_manage_apps)) {
1622 goto error_testpoint;
1623 }
1624
1625 health_code_update();
1626
1627 ret = sessiond_set_thread_pollset(&events, 2);
1628 if (ret < 0) {
1629 goto error_poll_create;
1630 }
1631
1632 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1633 if (ret < 0) {
1634 goto error;
1635 }
1636
1637 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1638 goto error;
1639 }
1640
1641 health_code_update();
1642
1643 while (1) {
1644 DBG("Apps thread polling");
1645
1646 /* Inifinite blocking call, waiting for transmission */
1647 restart:
1648 health_poll_entry();
1649 ret = lttng_poll_wait(&events, -1);
1650 DBG("Apps thread return from poll on %d fds",
1651 LTTNG_POLL_GETNB(&events));
1652 health_poll_exit();
1653 if (ret < 0) {
1654 /*
1655 * Restart interrupted system call.
1656 */
1657 if (errno == EINTR) {
1658 goto restart;
1659 }
1660 goto error;
1661 }
1662
1663 nb_fd = ret;
1664
1665 for (i = 0; i < nb_fd; i++) {
1666 /* Fetch once the poll data */
1667 revents = LTTNG_POLL_GETEV(&events, i);
1668 pollfd = LTTNG_POLL_GETFD(&events, i);
1669
1670 health_code_update();
1671
1672 if (!revents) {
1673 /* No activity for this FD (poll implementation). */
1674 continue;
1675 }
1676
1677 /* Thread quit pipe has been closed. Killing thread. */
1678 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1679 if (ret) {
1680 err = 0;
1681 goto exit;
1682 }
1683
1684 /* Inspect the apps cmd pipe */
1685 if (pollfd == apps_cmd_pipe[0]) {
1686 if (revents & LPOLLIN) {
1687 int sock;
1688
1689 /* Empty pipe */
1690 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1691 if (size_ret < sizeof(sock)) {
1692 PERROR("read apps cmd pipe");
1693 goto error;
1694 }
1695
1696 health_code_update();
1697
1698 /*
1699 * Since this is a command socket (write then read),
1700 * we only monitor the error events of the socket.
1701 */
1702 ret = lttng_poll_add(&events, sock,
1703 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1704 if (ret < 0) {
1705 goto error;
1706 }
1707
1708 DBG("Apps with sock %d added to poll set", sock);
1709 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1710 ERR("Apps command pipe error");
1711 goto error;
1712 } else {
1713 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1714 goto error;
1715 }
1716 } else {
1717 /*
1718 * At this point, we know that a registered application made
1719 * the event at poll_wait.
1720 */
1721 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1722 /* Removing from the poll set */
1723 ret = lttng_poll_del(&events, pollfd);
1724 if (ret < 0) {
1725 goto error;
1726 }
1727
1728 /* Socket closed on remote end. */
1729 ust_app_unregister(pollfd);
1730 } else {
1731 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1732 goto error;
1733 }
1734 }
1735
1736 health_code_update();
1737 }
1738 }
1739
1740 exit:
1741 error:
1742 lttng_poll_clean(&events);
1743 error_poll_create:
1744 error_testpoint:
1745 utils_close_pipe(apps_cmd_pipe);
1746 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1747
1748 /*
1749 * We don't clean the UST app hash table here since already registered
1750 * applications can still be controlled so let them be until the session
1751 * daemon dies or the applications stop.
1752 */
1753
1754 if (err) {
1755 health_error();
1756 ERR("Health error occurred in %s", __func__);
1757 }
1758 health_unregister(health_sessiond);
1759 DBG("Application communication apps thread cleanup complete");
1760 rcu_thread_offline();
1761 rcu_unregister_thread();
1762 return NULL;
1763 }
1764
1765 /*
1766 * Send a socket to a thread This is called from the dispatch UST registration
1767 * thread once all sockets are set for the application.
1768 *
1769 * The sock value can be invalid, we don't really care, the thread will handle
1770 * it and make the necessary cleanup if so.
1771 *
1772 * On success, return 0 else a negative value being the errno message of the
1773 * write().
1774 */
1775 static int send_socket_to_thread(int fd, int sock)
1776 {
1777 ssize_t ret;
1778
1779 /*
1780 * It's possible that the FD is set as invalid with -1 concurrently just
1781 * before calling this function being a shutdown state of the thread.
1782 */
1783 if (fd < 0) {
1784 ret = -EBADF;
1785 goto error;
1786 }
1787
1788 ret = lttng_write(fd, &sock, sizeof(sock));
1789 if (ret < sizeof(sock)) {
1790 PERROR("write apps pipe %d", fd);
1791 if (ret < 0) {
1792 ret = -errno;
1793 }
1794 goto error;
1795 }
1796
1797 /* All good. Don't send back the write positive ret value. */
1798 ret = 0;
1799 error:
1800 return (int) ret;
1801 }
1802
1803 /*
1804 * Sanitize the wait queue of the dispatch registration thread meaning removing
1805 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1806 * notify socket is never received.
1807 */
1808 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1809 {
1810 int ret, nb_fd = 0, i;
1811 unsigned int fd_added = 0;
1812 struct lttng_poll_event events;
1813 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1814
1815 assert(wait_queue);
1816
1817 lttng_poll_init(&events);
1818
1819 /* Just skip everything for an empty queue. */
1820 if (!wait_queue->count) {
1821 goto end;
1822 }
1823
1824 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1825 if (ret < 0) {
1826 goto error_create;
1827 }
1828
1829 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1830 &wait_queue->head, head) {
1831 assert(wait_node->app);
1832 ret = lttng_poll_add(&events, wait_node->app->sock,
1833 LPOLLHUP | LPOLLERR);
1834 if (ret < 0) {
1835 goto error;
1836 }
1837
1838 fd_added = 1;
1839 }
1840
1841 if (!fd_added) {
1842 goto end;
1843 }
1844
1845 /*
1846 * Poll but don't block so we can quickly identify the faulty events and
1847 * clean them afterwards from the wait queue.
1848 */
1849 ret = lttng_poll_wait(&events, 0);
1850 if (ret < 0) {
1851 goto error;
1852 }
1853 nb_fd = ret;
1854
1855 for (i = 0; i < nb_fd; i++) {
1856 /* Get faulty FD. */
1857 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1858 int pollfd = LTTNG_POLL_GETFD(&events, i);
1859
1860 if (!revents) {
1861 /* No activity for this FD (poll implementation). */
1862 continue;
1863 }
1864
1865 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1866 &wait_queue->head, head) {
1867 if (pollfd == wait_node->app->sock &&
1868 (revents & (LPOLLHUP | LPOLLERR))) {
1869 cds_list_del(&wait_node->head);
1870 wait_queue->count--;
1871 ust_app_destroy(wait_node->app);
1872 free(wait_node);
1873 /*
1874 * Silence warning of use-after-free in
1875 * cds_list_for_each_entry_safe which uses
1876 * __typeof__(*wait_node).
1877 */
1878 wait_node = NULL;
1879 break;
1880 } else {
1881 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1882 goto error;
1883 }
1884 }
1885 }
1886
1887 if (nb_fd > 0) {
1888 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1889 }
1890
1891 end:
1892 lttng_poll_clean(&events);
1893 return;
1894
1895 error:
1896 lttng_poll_clean(&events);
1897 error_create:
1898 ERR("Unable to sanitize wait queue");
1899 return;
1900 }
1901
1902 /*
1903 * Dispatch request from the registration threads to the application
1904 * communication thread.
1905 */
1906 static void *thread_dispatch_ust_registration(void *data)
1907 {
1908 int ret, err = -1;
1909 struct cds_wfcq_node *node;
1910 struct ust_command *ust_cmd = NULL;
1911 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1912 struct ust_reg_wait_queue wait_queue = {
1913 .count = 0,
1914 };
1915
1916 rcu_register_thread();
1917
1918 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1919
1920 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1921 goto error_testpoint;
1922 }
1923
1924 health_code_update();
1925
1926 CDS_INIT_LIST_HEAD(&wait_queue.head);
1927
1928 DBG("[thread] Dispatch UST command started");
1929
1930 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1931 health_code_update();
1932
1933 /* Atomically prepare the queue futex */
1934 futex_nto1_prepare(&ust_cmd_queue.futex);
1935
1936 do {
1937 struct ust_app *app = NULL;
1938 ust_cmd = NULL;
1939
1940 /*
1941 * Make sure we don't have node(s) that have hung up before receiving
1942 * the notify socket. This is to clean the list in order to avoid
1943 * memory leaks from notify socket that are never seen.
1944 */
1945 sanitize_wait_queue(&wait_queue);
1946
1947 health_code_update();
1948 /* Dequeue command for registration */
1949 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1950 if (node == NULL) {
1951 DBG("Woken up but nothing in the UST command queue");
1952 /* Continue thread execution */
1953 break;
1954 }
1955
1956 ust_cmd = caa_container_of(node, struct ust_command, node);
1957
1958 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1959 " gid:%d sock:%d name:%s (version %d.%d)",
1960 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1961 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1962 ust_cmd->sock, ust_cmd->reg_msg.name,
1963 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1964
1965 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1966 wait_node = zmalloc(sizeof(*wait_node));
1967 if (!wait_node) {
1968 PERROR("zmalloc wait_node dispatch");
1969 ret = close(ust_cmd->sock);
1970 if (ret < 0) {
1971 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1972 }
1973 lttng_fd_put(LTTNG_FD_APPS, 1);
1974 free(ust_cmd);
1975 goto error;
1976 }
1977 CDS_INIT_LIST_HEAD(&wait_node->head);
1978
1979 /* Create application object if socket is CMD. */
1980 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1981 ust_cmd->sock);
1982 if (!wait_node->app) {
1983 ret = close(ust_cmd->sock);
1984 if (ret < 0) {
1985 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1986 }
1987 lttng_fd_put(LTTNG_FD_APPS, 1);
1988 free(wait_node);
1989 free(ust_cmd);
1990 continue;
1991 }
1992 /*
1993 * Add application to the wait queue so we can set the notify
1994 * socket before putting this object in the global ht.
1995 */
1996 cds_list_add(&wait_node->head, &wait_queue.head);
1997 wait_queue.count++;
1998
1999 free(ust_cmd);
2000 /*
2001 * We have to continue here since we don't have the notify
2002 * socket and the application MUST be added to the hash table
2003 * only at that moment.
2004 */
2005 continue;
2006 } else {
2007 /*
2008 * Look for the application in the local wait queue and set the
2009 * notify socket if found.
2010 */
2011 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2012 &wait_queue.head, head) {
2013 health_code_update();
2014 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
2015 wait_node->app->notify_sock = ust_cmd->sock;
2016 cds_list_del(&wait_node->head);
2017 wait_queue.count--;
2018 app = wait_node->app;
2019 free(wait_node);
2020 DBG3("UST app notify socket %d is set", ust_cmd->sock);
2021 break;
2022 }
2023 }
2024
2025 /*
2026 * With no application at this stage the received socket is
2027 * basically useless so close it before we free the cmd data
2028 * structure for good.
2029 */
2030 if (!app) {
2031 ret = close(ust_cmd->sock);
2032 if (ret < 0) {
2033 PERROR("close ust sock dispatch %d", ust_cmd->sock);
2034 }
2035 lttng_fd_put(LTTNG_FD_APPS, 1);
2036 }
2037 free(ust_cmd);
2038 }
2039
2040 if (app) {
2041 /*
2042 * @session_lock_list
2043 *
2044 * Lock the global session list so from the register up to the
2045 * registration done message, no thread can see the application
2046 * and change its state.
2047 */
2048 session_lock_list();
2049 rcu_read_lock();
2050
2051 /*
2052 * Add application to the global hash table. This needs to be
2053 * done before the update to the UST registry can locate the
2054 * application.
2055 */
2056 ust_app_add(app);
2057
2058 /* Set app version. This call will print an error if needed. */
2059 (void) ust_app_version(app);
2060
2061 /* Send notify socket through the notify pipe. */
2062 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2063 app->notify_sock);
2064 if (ret < 0) {
2065 rcu_read_unlock();
2066 session_unlock_list();
2067 /*
2068 * No notify thread, stop the UST tracing. However, this is
2069 * not an internal error of the this thread thus setting
2070 * the health error code to a normal exit.
2071 */
2072 err = 0;
2073 goto error;
2074 }
2075
2076 /*
2077 * Update newly registered application with the tracing
2078 * registry info already enabled information.
2079 */
2080 update_ust_app(app->sock);
2081
2082 /*
2083 * Don't care about return value. Let the manage apps threads
2084 * handle app unregistration upon socket close.
2085 */
2086 (void) ust_app_register_done(app);
2087
2088 /*
2089 * Even if the application socket has been closed, send the app
2090 * to the thread and unregistration will take place at that
2091 * place.
2092 */
2093 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2094 if (ret < 0) {
2095 rcu_read_unlock();
2096 session_unlock_list();
2097 /*
2098 * No apps. thread, stop the UST tracing. However, this is
2099 * not an internal error of the this thread thus setting
2100 * the health error code to a normal exit.
2101 */
2102 err = 0;
2103 goto error;
2104 }
2105
2106 rcu_read_unlock();
2107 session_unlock_list();
2108 }
2109 } while (node != NULL);
2110
2111 health_poll_entry();
2112 /* Futex wait on queue. Blocking call on futex() */
2113 futex_nto1_wait(&ust_cmd_queue.futex);
2114 health_poll_exit();
2115 }
2116 /* Normal exit, no error */
2117 err = 0;
2118
2119 error:
2120 /* Clean up wait queue. */
2121 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2122 &wait_queue.head, head) {
2123 cds_list_del(&wait_node->head);
2124 wait_queue.count--;
2125 free(wait_node);
2126 }
2127
2128 /* Empty command queue. */
2129 for (;;) {
2130 /* Dequeue command for registration */
2131 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2132 if (node == NULL) {
2133 break;
2134 }
2135 ust_cmd = caa_container_of(node, struct ust_command, node);
2136 ret = close(ust_cmd->sock);
2137 if (ret < 0) {
2138 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2139 }
2140 lttng_fd_put(LTTNG_FD_APPS, 1);
2141 free(ust_cmd);
2142 }
2143
2144 error_testpoint:
2145 DBG("Dispatch thread dying");
2146 if (err) {
2147 health_error();
2148 ERR("Health error occurred in %s", __func__);
2149 }
2150 health_unregister(health_sessiond);
2151 rcu_unregister_thread();
2152 return NULL;
2153 }
2154
2155 /*
2156 * This thread manage application registration.
2157 */
2158 static void *thread_registration_apps(void *data)
2159 {
2160 int sock = -1, i, ret, pollfd, err = -1;
2161 uint32_t revents, nb_fd;
2162 struct lttng_poll_event events;
2163 /*
2164 * Get allocated in this thread, enqueued to a global queue, dequeued and
2165 * freed in the manage apps thread.
2166 */
2167 struct ust_command *ust_cmd = NULL;
2168
2169 DBG("[thread] Manage application registration started");
2170
2171 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2172
2173 if (testpoint(sessiond_thread_registration_apps)) {
2174 goto error_testpoint;
2175 }
2176
2177 ret = lttcomm_listen_unix_sock(apps_sock);
2178 if (ret < 0) {
2179 goto error_listen;
2180 }
2181
2182 /*
2183 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2184 * more will be added to this poll set.
2185 */
2186 ret = sessiond_set_thread_pollset(&events, 2);
2187 if (ret < 0) {
2188 goto error_create_poll;
2189 }
2190
2191 /* Add the application registration socket */
2192 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2193 if (ret < 0) {
2194 goto error_poll_add;
2195 }
2196
2197 /* Notify all applications to register */
2198 ret = notify_ust_apps(1);
2199 if (ret < 0) {
2200 ERR("Failed to notify applications or create the wait shared memory.\n"
2201 "Execution continues but there might be problem for already\n"
2202 "running applications that wishes to register.");
2203 }
2204
2205 while (1) {
2206 DBG("Accepting application registration");
2207
2208 /* Inifinite blocking call, waiting for transmission */
2209 restart:
2210 health_poll_entry();
2211 ret = lttng_poll_wait(&events, -1);
2212 health_poll_exit();
2213 if (ret < 0) {
2214 /*
2215 * Restart interrupted system call.
2216 */
2217 if (errno == EINTR) {
2218 goto restart;
2219 }
2220 goto error;
2221 }
2222
2223 nb_fd = ret;
2224
2225 for (i = 0; i < nb_fd; i++) {
2226 health_code_update();
2227
2228 /* Fetch once the poll data */
2229 revents = LTTNG_POLL_GETEV(&events, i);
2230 pollfd = LTTNG_POLL_GETFD(&events, i);
2231
2232 if (!revents) {
2233 /* No activity for this FD (poll implementation). */
2234 continue;
2235 }
2236
2237 /* Thread quit pipe has been closed. Killing thread. */
2238 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2239 if (ret) {
2240 err = 0;
2241 goto exit;
2242 }
2243
2244 /* Event on the registration socket */
2245 if (pollfd == apps_sock) {
2246 if (revents & LPOLLIN) {
2247 sock = lttcomm_accept_unix_sock(apps_sock);
2248 if (sock < 0) {
2249 goto error;
2250 }
2251
2252 /*
2253 * Set socket timeout for both receiving and ending.
2254 * app_socket_timeout is in seconds, whereas
2255 * lttcomm_setsockopt_rcv_timeout and
2256 * lttcomm_setsockopt_snd_timeout expect msec as
2257 * parameter.
2258 */
2259 if (app_socket_timeout >= 0) {
2260 (void) lttcomm_setsockopt_rcv_timeout(sock,
2261 app_socket_timeout * 1000);
2262 (void) lttcomm_setsockopt_snd_timeout(sock,
2263 app_socket_timeout * 1000);
2264 }
2265
2266 /*
2267 * Set the CLOEXEC flag. Return code is useless because
2268 * either way, the show must go on.
2269 */
2270 (void) utils_set_fd_cloexec(sock);
2271
2272 /* Create UST registration command for enqueuing */
2273 ust_cmd = zmalloc(sizeof(struct ust_command));
2274 if (ust_cmd == NULL) {
2275 PERROR("ust command zmalloc");
2276 ret = close(sock);
2277 if (ret) {
2278 PERROR("close");
2279 }
2280 goto error;
2281 }
2282
2283 /*
2284 * Using message-based transmissions to ensure we don't
2285 * have to deal with partially received messages.
2286 */
2287 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2288 if (ret < 0) {
2289 ERR("Exhausted file descriptors allowed for applications.");
2290 free(ust_cmd);
2291 ret = close(sock);
2292 if (ret) {
2293 PERROR("close");
2294 }
2295 sock = -1;
2296 continue;
2297 }
2298
2299 health_code_update();
2300 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2301 if (ret < 0) {
2302 free(ust_cmd);
2303 /* Close socket of the application. */
2304 ret = close(sock);
2305 if (ret) {
2306 PERROR("close");
2307 }
2308 lttng_fd_put(LTTNG_FD_APPS, 1);
2309 sock = -1;
2310 continue;
2311 }
2312 health_code_update();
2313
2314 ust_cmd->sock = sock;
2315 sock = -1;
2316
2317 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2318 " gid:%d sock:%d name:%s (version %d.%d)",
2319 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2320 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2321 ust_cmd->sock, ust_cmd->reg_msg.name,
2322 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2323
2324 /*
2325 * Lock free enqueue the registration request. The red pill
2326 * has been taken! This apps will be part of the *system*.
2327 */
2328 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2329
2330 /*
2331 * Wake the registration queue futex. Implicit memory
2332 * barrier with the exchange in cds_wfcq_enqueue.
2333 */
2334 futex_nto1_wake(&ust_cmd_queue.futex);
2335 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2336 ERR("Register apps socket poll error");
2337 goto error;
2338 } else {
2339 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2340 goto error;
2341 }
2342 }
2343 }
2344 }
2345
2346 exit:
2347 error:
2348 /* Notify that the registration thread is gone */
2349 notify_ust_apps(0);
2350
2351 if (apps_sock >= 0) {
2352 ret = close(apps_sock);
2353 if (ret) {
2354 PERROR("close");
2355 }
2356 }
2357 if (sock >= 0) {
2358 ret = close(sock);
2359 if (ret) {
2360 PERROR("close");
2361 }
2362 lttng_fd_put(LTTNG_FD_APPS, 1);
2363 }
2364 unlink(apps_unix_sock_path);
2365
2366 error_poll_add:
2367 lttng_poll_clean(&events);
2368 error_listen:
2369 error_create_poll:
2370 error_testpoint:
2371 DBG("UST Registration thread cleanup complete");
2372 if (err) {
2373 health_error();
2374 ERR("Health error occurred in %s", __func__);
2375 }
2376 health_unregister(health_sessiond);
2377
2378 return NULL;
2379 }
2380
2381 /*
2382 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2383 * exec or it will fails.
2384 */
2385 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2386 {
2387 int ret, clock_ret;
2388 struct timespec timeout;
2389
2390 /*
2391 * Make sure we set the readiness flag to 0 because we are NOT ready.
2392 * This access to consumer_thread_is_ready does not need to be
2393 * protected by consumer_data.cond_mutex (yet) since the consumer
2394 * management thread has not been started at this point.
2395 */
2396 consumer_data->consumer_thread_is_ready = 0;
2397
2398 /* Setup pthread condition */
2399 ret = pthread_condattr_init(&consumer_data->condattr);
2400 if (ret) {
2401 errno = ret;
2402 PERROR("pthread_condattr_init consumer data");
2403 goto error;
2404 }
2405
2406 /*
2407 * Set the monotonic clock in order to make sure we DO NOT jump in time
2408 * between the clock_gettime() call and the timedwait call. See bug #324
2409 * for a more details and how we noticed it.
2410 */
2411 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2412 if (ret) {
2413 errno = ret;
2414 PERROR("pthread_condattr_setclock consumer data");
2415 goto error;
2416 }
2417
2418 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2419 if (ret) {
2420 errno = ret;
2421 PERROR("pthread_cond_init consumer data");
2422 goto error;
2423 }
2424
2425 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2426 thread_manage_consumer, consumer_data);
2427 if (ret) {
2428 errno = ret;
2429 PERROR("pthread_create consumer");
2430 ret = -1;
2431 goto error;
2432 }
2433
2434 /* We are about to wait on a pthread condition */
2435 pthread_mutex_lock(&consumer_data->cond_mutex);
2436
2437 /* Get time for sem_timedwait absolute timeout */
2438 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2439 /*
2440 * Set the timeout for the condition timed wait even if the clock gettime
2441 * call fails since we might loop on that call and we want to avoid to
2442 * increment the timeout too many times.
2443 */
2444 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2445
2446 /*
2447 * The following loop COULD be skipped in some conditions so this is why we
2448 * set ret to 0 in order to make sure at least one round of the loop is
2449 * done.
2450 */
2451 ret = 0;
2452
2453 /*
2454 * Loop until the condition is reached or when a timeout is reached. Note
2455 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2456 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2457 * possible. This loop does not take any chances and works with both of
2458 * them.
2459 */
2460 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2461 if (clock_ret < 0) {
2462 PERROR("clock_gettime spawn consumer");
2463 /* Infinite wait for the consumerd thread to be ready */
2464 ret = pthread_cond_wait(&consumer_data->cond,
2465 &consumer_data->cond_mutex);
2466 } else {
2467 ret = pthread_cond_timedwait(&consumer_data->cond,
2468 &consumer_data->cond_mutex, &timeout);
2469 }
2470 }
2471
2472 /* Release the pthread condition */
2473 pthread_mutex_unlock(&consumer_data->cond_mutex);
2474
2475 if (ret != 0) {
2476 errno = ret;
2477 if (ret == ETIMEDOUT) {
2478 int pth_ret;
2479
2480 /*
2481 * Call has timed out so we kill the kconsumerd_thread and return
2482 * an error.
2483 */
2484 ERR("Condition timed out. The consumer thread was never ready."
2485 " Killing it");
2486 pth_ret = pthread_cancel(consumer_data->thread);
2487 if (pth_ret < 0) {
2488 PERROR("pthread_cancel consumer thread");
2489 }
2490 } else {
2491 PERROR("pthread_cond_wait failed consumer thread");
2492 }
2493 /* Caller is expecting a negative value on failure. */
2494 ret = -1;
2495 goto error;
2496 }
2497
2498 pthread_mutex_lock(&consumer_data->pid_mutex);
2499 if (consumer_data->pid == 0) {
2500 ERR("Consumerd did not start");
2501 pthread_mutex_unlock(&consumer_data->pid_mutex);
2502 goto error;
2503 }
2504 pthread_mutex_unlock(&consumer_data->pid_mutex);
2505
2506 return 0;
2507
2508 error:
2509 return ret;
2510 }
2511
2512 /*
2513 * Join consumer thread
2514 */
2515 static int join_consumer_thread(struct consumer_data *consumer_data)
2516 {
2517 void *status;
2518
2519 /* Consumer pid must be a real one. */
2520 if (consumer_data->pid > 0) {
2521 int ret;
2522 ret = kill(consumer_data->pid, SIGTERM);
2523 if (ret) {
2524 PERROR("Error killing consumer daemon");
2525 return ret;
2526 }
2527 return pthread_join(consumer_data->thread, &status);
2528 } else {
2529 return 0;
2530 }
2531 }
2532
2533 /*
2534 * Fork and exec a consumer daemon (consumerd).
2535 *
2536 * Return pid if successful else -1.
2537 */
2538 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2539 {
2540 int ret;
2541 pid_t pid;
2542 const char *consumer_to_use;
2543 const char *verbosity;
2544 struct stat st;
2545
2546 DBG("Spawning consumerd");
2547
2548 pid = fork();
2549 if (pid == 0) {
2550 /*
2551 * Exec consumerd.
2552 */
2553 if (opt_verbose_consumer) {
2554 verbosity = "--verbose";
2555 } else if (lttng_opt_quiet) {
2556 verbosity = "--quiet";
2557 } else {
2558 verbosity = "";
2559 }
2560
2561 switch (consumer_data->type) {
2562 case LTTNG_CONSUMER_KERNEL:
2563 /*
2564 * Find out which consumerd to execute. We will first try the
2565 * 64-bit path, then the sessiond's installation directory, and
2566 * fallback on the 32-bit one,
2567 */
2568 DBG3("Looking for a kernel consumer at these locations:");
2569 DBG3(" 1) %s", consumerd64_bin);
2570 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2571 DBG3(" 3) %s", consumerd32_bin);
2572 if (stat(consumerd64_bin, &st) == 0) {
2573 DBG3("Found location #1");
2574 consumer_to_use = consumerd64_bin;
2575 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2576 DBG3("Found location #2");
2577 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2578 } else if (stat(consumerd32_bin, &st) == 0) {
2579 DBG3("Found location #3");
2580 consumer_to_use = consumerd32_bin;
2581 } else {
2582 DBG("Could not find any valid consumerd executable");
2583 ret = -EINVAL;
2584 break;
2585 }
2586 DBG("Using kernel consumer at: %s", consumer_to_use);
2587 ret = execl(consumer_to_use,
2588 "lttng-consumerd", verbosity, "-k",
2589 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2590 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2591 "--group", tracing_group_name,
2592 NULL);
2593 break;
2594 case LTTNG_CONSUMER64_UST:
2595 {
2596 char *tmpnew = NULL;
2597
2598 if (consumerd64_libdir[0] != '\0') {
2599 char *tmp;
2600 size_t tmplen;
2601
2602 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2603 if (!tmp) {
2604 tmp = "";
2605 }
2606 tmplen = strlen("LD_LIBRARY_PATH=")
2607 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2608 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2609 if (!tmpnew) {
2610 ret = -ENOMEM;
2611 goto error;
2612 }
2613 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2614 strcat(tmpnew, consumerd64_libdir);
2615 if (tmp[0] != '\0') {
2616 strcat(tmpnew, ":");
2617 strcat(tmpnew, tmp);
2618 }
2619 ret = putenv(tmpnew);
2620 if (ret) {
2621 ret = -errno;
2622 free(tmpnew);
2623 goto error;
2624 }
2625 }
2626 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2627 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2628 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2629 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2630 "--group", tracing_group_name,
2631 NULL);
2632 if (consumerd64_libdir[0] != '\0') {
2633 free(tmpnew);
2634 }
2635 break;
2636 }
2637 case LTTNG_CONSUMER32_UST:
2638 {
2639 char *tmpnew = NULL;
2640
2641 if (consumerd32_libdir[0] != '\0') {
2642 char *tmp;
2643 size_t tmplen;
2644
2645 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2646 if (!tmp) {
2647 tmp = "";
2648 }
2649 tmplen = strlen("LD_LIBRARY_PATH=")
2650 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2651 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2652 if (!tmpnew) {
2653 ret = -ENOMEM;
2654 goto error;
2655 }
2656 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2657 strcat(tmpnew, consumerd32_libdir);
2658 if (tmp[0] != '\0') {
2659 strcat(tmpnew, ":");
2660 strcat(tmpnew, tmp);
2661 }
2662 ret = putenv(tmpnew);
2663 if (ret) {
2664 ret = -errno;
2665 free(tmpnew);
2666 goto error;
2667 }
2668 }
2669 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2670 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2671 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2672 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2673 "--group", tracing_group_name,
2674 NULL);
2675 if (consumerd32_libdir[0] != '\0') {
2676 free(tmpnew);
2677 }
2678 break;
2679 }
2680 default:
2681 PERROR("unknown consumer type");
2682 exit(EXIT_FAILURE);
2683 }
2684 if (errno != 0) {
2685 PERROR("Consumer execl()");
2686 }
2687 /* Reaching this point, we got a failure on our execl(). */
2688 exit(EXIT_FAILURE);
2689 } else if (pid > 0) {
2690 ret = pid;
2691 } else {
2692 PERROR("start consumer fork");
2693 ret = -errno;
2694 }
2695 error:
2696 return ret;
2697 }
2698
2699 /*
2700 * Spawn the consumerd daemon and session daemon thread.
2701 */
2702 static int start_consumerd(struct consumer_data *consumer_data)
2703 {
2704 int ret;
2705
2706 /*
2707 * Set the listen() state on the socket since there is a possible race
2708 * between the exec() of the consumer daemon and this call if place in the
2709 * consumer thread. See bug #366 for more details.
2710 */
2711 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2712 if (ret < 0) {
2713 goto error;
2714 }
2715
2716 pthread_mutex_lock(&consumer_data->pid_mutex);
2717 if (consumer_data->pid != 0) {
2718 pthread_mutex_unlock(&consumer_data->pid_mutex);
2719 goto end;
2720 }
2721
2722 ret = spawn_consumerd(consumer_data);
2723 if (ret < 0) {
2724 ERR("Spawning consumerd failed");
2725 pthread_mutex_unlock(&consumer_data->pid_mutex);
2726 goto error;
2727 }
2728
2729 /* Setting up the consumer_data pid */
2730 consumer_data->pid = ret;
2731 DBG2("Consumer pid %d", consumer_data->pid);
2732 pthread_mutex_unlock(&consumer_data->pid_mutex);
2733
2734 DBG2("Spawning consumer control thread");
2735 ret = spawn_consumer_thread(consumer_data);
2736 if (ret < 0) {
2737 ERR("Fatal error spawning consumer control thread");
2738 goto error;
2739 }
2740
2741 end:
2742 return 0;
2743
2744 error:
2745 /* Cleanup already created sockets on error. */
2746 if (consumer_data->err_sock >= 0) {
2747 int err;
2748
2749 err = close(consumer_data->err_sock);
2750 if (err < 0) {
2751 PERROR("close consumer data error socket");
2752 }
2753 }
2754 return ret;
2755 }
2756
2757 /*
2758 * Setup necessary data for kernel tracer action.
2759 */
2760 static int init_kernel_tracer(void)
2761 {
2762 int ret;
2763
2764 /* Modprobe lttng kernel modules */
2765 ret = modprobe_lttng_control();
2766 if (ret < 0) {
2767 goto error;
2768 }
2769
2770 /* Open debugfs lttng */
2771 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2772 if (kernel_tracer_fd < 0) {
2773 DBG("Failed to open %s", module_proc_lttng);
2774 ret = -1;
2775 goto error_open;
2776 }
2777
2778 /* Validate kernel version */
2779 ret = kernel_validate_version(kernel_tracer_fd);
2780 if (ret < 0) {
2781 goto error_version;
2782 }
2783
2784 ret = modprobe_lttng_data();
2785 if (ret < 0) {
2786 goto error_modules;
2787 }
2788
2789 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2790 return 0;
2791
2792 error_version:
2793 modprobe_remove_lttng_control();
2794 ret = close(kernel_tracer_fd);
2795 if (ret) {
2796 PERROR("close");
2797 }
2798 kernel_tracer_fd = -1;
2799 return LTTNG_ERR_KERN_VERSION;
2800
2801 error_modules:
2802 ret = close(kernel_tracer_fd);
2803 if (ret) {
2804 PERROR("close");
2805 }
2806
2807 error_open:
2808 modprobe_remove_lttng_control();
2809
2810 error:
2811 WARN("No kernel tracer available");
2812 kernel_tracer_fd = -1;
2813 if (!is_root) {
2814 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2815 } else {
2816 return LTTNG_ERR_KERN_NA;
2817 }
2818 }
2819
2820
2821 /*
2822 * Copy consumer output from the tracing session to the domain session. The
2823 * function also applies the right modification on a per domain basis for the
2824 * trace files destination directory.
2825 *
2826 * Should *NOT* be called with RCU read-side lock held.
2827 */
2828 static int copy_session_consumer(int domain, struct ltt_session *session)
2829 {
2830 int ret;
2831 const char *dir_name;
2832 struct consumer_output *consumer;
2833
2834 assert(session);
2835 assert(session->consumer);
2836
2837 switch (domain) {
2838 case LTTNG_DOMAIN_KERNEL:
2839 DBG3("Copying tracing session consumer output in kernel session");
2840 /*
2841 * XXX: We should audit the session creation and what this function
2842 * does "extra" in order to avoid a destroy since this function is used
2843 * in the domain session creation (kernel and ust) only. Same for UST
2844 * domain.
2845 */
2846 if (session->kernel_session->consumer) {
2847 consumer_output_put(session->kernel_session->consumer);
2848 }
2849 session->kernel_session->consumer =
2850 consumer_copy_output(session->consumer);
2851 /* Ease our life a bit for the next part */
2852 consumer = session->kernel_session->consumer;
2853 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2854 break;
2855 case LTTNG_DOMAIN_JUL:
2856 case LTTNG_DOMAIN_LOG4J:
2857 case LTTNG_DOMAIN_PYTHON:
2858 case LTTNG_DOMAIN_UST:
2859 DBG3("Copying tracing session consumer output in UST session");
2860 if (session->ust_session->consumer) {
2861 consumer_output_put(session->ust_session->consumer);
2862 }
2863 session->ust_session->consumer =
2864 consumer_copy_output(session->consumer);
2865 /* Ease our life a bit for the next part */
2866 consumer = session->ust_session->consumer;
2867 dir_name = DEFAULT_UST_TRACE_DIR;
2868 break;
2869 default:
2870 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2871 goto error;
2872 }
2873
2874 /* Append correct directory to subdir */
2875 strncat(consumer->subdir, dir_name,
2876 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2877 DBG3("Copy session consumer subdir %s", consumer->subdir);
2878
2879 ret = LTTNG_OK;
2880
2881 error:
2882 return ret;
2883 }
2884
2885 /*
2886 * Create an UST session and add it to the session ust list.
2887 *
2888 * Should *NOT* be called with RCU read-side lock held.
2889 */
2890 static int create_ust_session(struct ltt_session *session,
2891 struct lttng_domain *domain)
2892 {
2893 int ret;
2894 struct ltt_ust_session *lus = NULL;
2895
2896 assert(session);
2897 assert(domain);
2898 assert(session->consumer);
2899
2900 switch (domain->type) {
2901 case LTTNG_DOMAIN_JUL:
2902 case LTTNG_DOMAIN_LOG4J:
2903 case LTTNG_DOMAIN_PYTHON:
2904 case LTTNG_DOMAIN_UST:
2905 break;
2906 default:
2907 ERR("Unknown UST domain on create session %d", domain->type);
2908 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2909 goto error;
2910 }
2911
2912 DBG("Creating UST session");
2913
2914 lus = trace_ust_create_session(session->id);
2915 if (lus == NULL) {
2916 ret = LTTNG_ERR_UST_SESS_FAIL;
2917 goto error;
2918 }
2919
2920 lus->uid = session->uid;
2921 lus->gid = session->gid;
2922 lus->output_traces = session->output_traces;
2923 lus->snapshot_mode = session->snapshot_mode;
2924 lus->live_timer_interval = session->live_timer;
2925 session->ust_session = lus;
2926 if (session->shm_path[0]) {
2927 strncpy(lus->root_shm_path, session->shm_path,
2928 sizeof(lus->root_shm_path));
2929 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2930 strncpy(lus->shm_path, session->shm_path,
2931 sizeof(lus->shm_path));
2932 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2933 strncat(lus->shm_path, "/ust",
2934 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2935 }
2936 /* Copy session output to the newly created UST session */
2937 ret = copy_session_consumer(domain->type, session);
2938 if (ret != LTTNG_OK) {
2939 goto error;
2940 }
2941
2942 return LTTNG_OK;
2943
2944 error:
2945 free(lus);
2946 session->ust_session = NULL;
2947 return ret;
2948 }
2949
2950 /*
2951 * Create a kernel tracer session then create the default channel.
2952 */
2953 static int create_kernel_session(struct ltt_session *session)
2954 {
2955 int ret;
2956
2957 DBG("Creating kernel session");
2958
2959 ret = kernel_create_session(session, kernel_tracer_fd);
2960 if (ret < 0) {
2961 ret = LTTNG_ERR_KERN_SESS_FAIL;
2962 goto error;
2963 }
2964
2965 /* Code flow safety */
2966 assert(session->kernel_session);
2967
2968 /* Copy session output to the newly created Kernel session */
2969 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2970 if (ret != LTTNG_OK) {
2971 goto error;
2972 }
2973
2974 /* Create directory(ies) on local filesystem. */
2975 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2976 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2977 ret = run_as_mkdir_recursive(
2978 session->kernel_session->consumer->dst.trace_path,
2979 S_IRWXU | S_IRWXG, session->uid, session->gid);
2980 if (ret < 0) {
2981 if (errno != EEXIST) {
2982 ERR("Trace directory creation error");
2983 goto error;
2984 }
2985 }
2986 }
2987
2988 session->kernel_session->uid = session->uid;
2989 session->kernel_session->gid = session->gid;
2990 session->kernel_session->output_traces = session->output_traces;
2991 session->kernel_session->snapshot_mode = session->snapshot_mode;
2992
2993 return LTTNG_OK;
2994
2995 error:
2996 trace_kernel_destroy_session(session->kernel_session);
2997 session->kernel_session = NULL;
2998 return ret;
2999 }
3000
3001 /*
3002 * Count number of session permitted by uid/gid.
3003 */
3004 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
3005 {
3006 unsigned int i = 0;
3007 struct ltt_session *session;
3008
3009 DBG("Counting number of available session for UID %d GID %d",
3010 uid, gid);
3011 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
3012 /*
3013 * Only list the sessions the user can control.
3014 */
3015 if (!session_access_ok(session, uid, gid)) {
3016 continue;
3017 }
3018 i++;
3019 }
3020 return i;
3021 }
3022
3023 /*
3024 * Process the command requested by the lttng client within the command
3025 * context structure. This function make sure that the return structure (llm)
3026 * is set and ready for transmission before returning.
3027 *
3028 * Return any error encountered or 0 for success.
3029 *
3030 * "sock" is only used for special-case var. len data.
3031 *
3032 * Should *NOT* be called with RCU read-side lock held.
3033 */
3034 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
3035 int *sock_error)
3036 {
3037 int ret = LTTNG_OK;
3038 int need_tracing_session = 1;
3039 int need_domain;
3040
3041 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
3042
3043 assert(!rcu_read_ongoing());
3044
3045 *sock_error = 0;
3046
3047 switch (cmd_ctx->lsm->cmd_type) {
3048 case LTTNG_CREATE_SESSION:
3049 case LTTNG_CREATE_SESSION_SNAPSHOT:
3050 case LTTNG_CREATE_SESSION_LIVE:
3051 case LTTNG_DESTROY_SESSION:
3052 case LTTNG_LIST_SESSIONS:
3053 case LTTNG_LIST_DOMAINS:
3054 case LTTNG_START_TRACE:
3055 case LTTNG_STOP_TRACE:
3056 case LTTNG_DATA_PENDING:
3057 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3058 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3059 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3060 case LTTNG_SNAPSHOT_RECORD:
3061 case LTTNG_SAVE_SESSION:
3062 case LTTNG_SET_SESSION_SHM_PATH:
3063 case LTTNG_REGENERATE_METADATA:
3064 case LTTNG_REGENERATE_STATEDUMP:
3065 case LTTNG_REGISTER_TRIGGER:
3066 case LTTNG_UNREGISTER_TRIGGER:
3067 need_domain = 0;
3068 break;
3069 default:
3070 need_domain = 1;
3071 }
3072
3073 if (opt_no_kernel && need_domain
3074 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3075 if (!is_root) {
3076 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3077 } else {
3078 ret = LTTNG_ERR_KERN_NA;
3079 }
3080 goto error;
3081 }
3082
3083 /* Deny register consumer if we already have a spawned consumer. */
3084 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3085 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3086 if (kconsumer_data.pid > 0) {
3087 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3088 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3089 goto error;
3090 }
3091 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3092 }
3093
3094 /*
3095 * Check for command that don't needs to allocate a returned payload. We do
3096 * this here so we don't have to make the call for no payload at each
3097 * command.
3098 */
3099 switch(cmd_ctx->lsm->cmd_type) {
3100 case LTTNG_LIST_SESSIONS:
3101 case LTTNG_LIST_TRACEPOINTS:
3102 case LTTNG_LIST_TRACEPOINT_FIELDS:
3103 case LTTNG_LIST_DOMAINS:
3104 case LTTNG_LIST_CHANNELS:
3105 case LTTNG_LIST_EVENTS:
3106 case LTTNG_LIST_SYSCALLS:
3107 case LTTNG_LIST_TRACKER_PIDS:
3108 case LTTNG_DATA_PENDING:
3109 break;
3110 default:
3111 /* Setup lttng message with no payload */
3112 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3113 if (ret < 0) {
3114 /* This label does not try to unlock the session */
3115 goto init_setup_error;
3116 }
3117 }
3118
3119 /* Commands that DO NOT need a session. */
3120 switch (cmd_ctx->lsm->cmd_type) {
3121 case LTTNG_CREATE_SESSION:
3122 case LTTNG_CREATE_SESSION_SNAPSHOT:
3123 case LTTNG_CREATE_SESSION_LIVE:
3124 case LTTNG_LIST_SESSIONS:
3125 case LTTNG_LIST_TRACEPOINTS:
3126 case LTTNG_LIST_SYSCALLS:
3127 case LTTNG_LIST_TRACEPOINT_FIELDS:
3128 case LTTNG_SAVE_SESSION:
3129 case LTTNG_REGISTER_TRIGGER:
3130 case LTTNG_UNREGISTER_TRIGGER:
3131 need_tracing_session = 0;
3132 break;
3133 default:
3134 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3135 /*
3136 * We keep the session list lock across _all_ commands
3137 * for now, because the per-session lock does not
3138 * handle teardown properly.
3139 */
3140 session_lock_list();
3141 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3142 if (cmd_ctx->session == NULL) {
3143 ret = LTTNG_ERR_SESS_NOT_FOUND;
3144 goto error;
3145 } else {
3146 /* Acquire lock for the session */
3147 session_lock(cmd_ctx->session);
3148 }
3149 break;
3150 }
3151
3152 /*
3153 * Commands that need a valid session but should NOT create one if none
3154 * exists. Instead of creating one and destroying it when the command is
3155 * handled, process that right before so we save some round trip in useless
3156 * code path.
3157 */
3158 switch (cmd_ctx->lsm->cmd_type) {
3159 case LTTNG_DISABLE_CHANNEL:
3160 case LTTNG_DISABLE_EVENT:
3161 switch (cmd_ctx->lsm->domain.type) {
3162 case LTTNG_DOMAIN_KERNEL:
3163 if (!cmd_ctx->session->kernel_session) {
3164 ret = LTTNG_ERR_NO_CHANNEL;
3165 goto error;
3166 }
3167 break;
3168 case LTTNG_DOMAIN_JUL:
3169 case LTTNG_DOMAIN_LOG4J:
3170 case LTTNG_DOMAIN_PYTHON:
3171 case LTTNG_DOMAIN_UST:
3172 if (!cmd_ctx->session->ust_session) {
3173 ret = LTTNG_ERR_NO_CHANNEL;
3174 goto error;
3175 }
3176 break;
3177 default:
3178 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3179 goto error;
3180 }
3181 default:
3182 break;
3183 }
3184
3185 if (!need_domain) {
3186 goto skip_domain;
3187 }
3188
3189 /*
3190 * Check domain type for specific "pre-action".
3191 */
3192 switch (cmd_ctx->lsm->domain.type) {
3193 case LTTNG_DOMAIN_KERNEL:
3194 if (!is_root) {
3195 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3196 goto error;
3197 }
3198
3199 /* Kernel tracer check */
3200 if (kernel_tracer_fd == -1) {
3201 /* Basically, load kernel tracer modules */
3202 ret = init_kernel_tracer();
3203 if (ret != 0) {
3204 goto error;
3205 }
3206 }
3207
3208 /* Consumer is in an ERROR state. Report back to client */
3209 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3210 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3211 goto error;
3212 }
3213
3214 /* Need a session for kernel command */
3215 if (need_tracing_session) {
3216 if (cmd_ctx->session->kernel_session == NULL) {
3217 ret = create_kernel_session(cmd_ctx->session);
3218 if (ret < 0) {
3219 ret = LTTNG_ERR_KERN_SESS_FAIL;
3220 goto error;
3221 }
3222 }
3223
3224 /* Start the kernel consumer daemon */
3225 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3226 if (kconsumer_data.pid == 0 &&
3227 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3228 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3229 ret = start_consumerd(&kconsumer_data);
3230 if (ret < 0) {
3231 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3232 goto error;
3233 }
3234 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3235 } else {
3236 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3237 }
3238
3239 /*
3240 * The consumer was just spawned so we need to add the socket to
3241 * the consumer output of the session if exist.
3242 */
3243 ret = consumer_create_socket(&kconsumer_data,
3244 cmd_ctx->session->kernel_session->consumer);
3245 if (ret < 0) {
3246 goto error;
3247 }
3248 }
3249
3250 break;
3251 case LTTNG_DOMAIN_JUL:
3252 case LTTNG_DOMAIN_LOG4J:
3253 case LTTNG_DOMAIN_PYTHON:
3254 case LTTNG_DOMAIN_UST:
3255 {
3256 if (!ust_app_supported()) {
3257 ret = LTTNG_ERR_NO_UST;
3258 goto error;
3259 }
3260 /* Consumer is in an ERROR state. Report back to client */
3261 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3262 ret = LTTNG_ERR_NO_USTCONSUMERD;
3263 goto error;
3264 }
3265
3266 if (need_tracing_session) {
3267 /* Create UST session if none exist. */
3268 if (cmd_ctx->session->ust_session == NULL) {
3269 ret = create_ust_session(cmd_ctx->session,
3270 &cmd_ctx->lsm->domain);
3271 if (ret != LTTNG_OK) {
3272 goto error;
3273 }
3274 }
3275
3276 /* Start the UST consumer daemons */
3277 /* 64-bit */
3278 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3279 if (consumerd64_bin[0] != '\0' &&
3280 ustconsumer64_data.pid == 0 &&
3281 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3282 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3283 ret = start_consumerd(&ustconsumer64_data);
3284 if (ret < 0) {
3285 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3286 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3287 goto error;
3288 }
3289
3290 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3291 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3292 } else {
3293 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3294 }
3295
3296 /*
3297 * Setup socket for consumer 64 bit. No need for atomic access
3298 * since it was set above and can ONLY be set in this thread.
3299 */
3300 ret = consumer_create_socket(&ustconsumer64_data,
3301 cmd_ctx->session->ust_session->consumer);
3302 if (ret < 0) {
3303 goto error;
3304 }
3305
3306 /* 32-bit */
3307 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3308 if (consumerd32_bin[0] != '\0' &&
3309 ustconsumer32_data.pid == 0 &&
3310 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3311 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3312 ret = start_consumerd(&ustconsumer32_data);
3313 if (ret < 0) {
3314 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3315 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3316 goto error;
3317 }
3318
3319 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3320 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3321 } else {
3322 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3323 }
3324
3325 /*
3326 * Setup socket for consumer 64 bit. No need for atomic access
3327 * since it was set above and can ONLY be set in this thread.
3328 */
3329 ret = consumer_create_socket(&ustconsumer32_data,
3330 cmd_ctx->session->ust_session->consumer);
3331 if (ret < 0) {
3332 goto error;
3333 }
3334 }
3335 break;
3336 }
3337 default:
3338 break;
3339 }
3340 skip_domain:
3341
3342 /* Validate consumer daemon state when start/stop trace command */
3343 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3344 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3345 switch (cmd_ctx->lsm->domain.type) {
3346 case LTTNG_DOMAIN_NONE:
3347 break;
3348 case LTTNG_DOMAIN_JUL:
3349 case LTTNG_DOMAIN_LOG4J:
3350 case LTTNG_DOMAIN_PYTHON:
3351 case LTTNG_DOMAIN_UST:
3352 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3353 ret = LTTNG_ERR_NO_USTCONSUMERD;
3354 goto error;
3355 }
3356 break;
3357 case LTTNG_DOMAIN_KERNEL:
3358 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3359 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3360 goto error;
3361 }
3362 break;
3363 default:
3364 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3365 goto error;
3366 }
3367 }
3368
3369 /*
3370 * Check that the UID or GID match that of the tracing session.
3371 * The root user can interact with all sessions.
3372 */
3373 if (need_tracing_session) {
3374 if (!session_access_ok(cmd_ctx->session,
3375 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3376 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3377 ret = LTTNG_ERR_EPERM;
3378 goto error;
3379 }
3380 }
3381
3382 /*
3383 * Send relayd information to consumer as soon as we have a domain and a
3384 * session defined.
3385 */
3386 if (cmd_ctx->session && need_domain) {
3387 /*
3388 * Setup relayd if not done yet. If the relayd information was already
3389 * sent to the consumer, this call will gracefully return.
3390 */
3391 ret = cmd_setup_relayd(cmd_ctx->session);
3392 if (ret != LTTNG_OK) {
3393 goto error;
3394 }
3395 }
3396
3397 /* Process by command type */
3398 switch (cmd_ctx->lsm->cmd_type) {
3399 case LTTNG_ADD_CONTEXT:
3400 {
3401 /*
3402 * An LTTNG_ADD_CONTEXT command might have a supplementary
3403 * payload if the context being added is an application context.
3404 */
3405 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3406 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3407 char *provider_name = NULL, *context_name = NULL;
3408 size_t provider_name_len =
3409 cmd_ctx->lsm->u.context.provider_name_len;
3410 size_t context_name_len =
3411 cmd_ctx->lsm->u.context.context_name_len;
3412
3413 if (provider_name_len == 0 || context_name_len == 0) {
3414 /*
3415 * Application provider and context names MUST
3416 * be provided.
3417 */
3418 ret = -LTTNG_ERR_INVALID;
3419 goto error;
3420 }
3421
3422 provider_name = zmalloc(provider_name_len + 1);
3423 if (!provider_name) {
3424 ret = -LTTNG_ERR_NOMEM;
3425 goto error;
3426 }
3427 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3428 provider_name;
3429
3430 context_name = zmalloc(context_name_len + 1);
3431 if (!context_name) {
3432 ret = -LTTNG_ERR_NOMEM;
3433 goto error_add_context;
3434 }
3435 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3436 context_name;
3437
3438 ret = lttcomm_recv_unix_sock(sock, provider_name,
3439 provider_name_len);
3440 if (ret < 0) {
3441 goto error_add_context;
3442 }
3443
3444 ret = lttcomm_recv_unix_sock(sock, context_name,
3445 context_name_len);
3446 if (ret < 0) {
3447 goto error_add_context;
3448 }
3449 }
3450
3451 /*
3452 * cmd_add_context assumes ownership of the provider and context
3453 * names.
3454 */
3455 ret = cmd_add_context(cmd_ctx->session,
3456 cmd_ctx->lsm->domain.type,
3457 cmd_ctx->lsm->u.context.channel_name,
3458 &cmd_ctx->lsm->u.context.ctx,
3459 kernel_poll_pipe[1]);
3460
3461 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3462 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3463 error_add_context:
3464 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3465 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3466 if (ret < 0) {
3467 goto error;
3468 }
3469 break;
3470 }
3471 case LTTNG_DISABLE_CHANNEL:
3472 {
3473 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3474 cmd_ctx->lsm->u.disable.channel_name);
3475 break;
3476 }
3477 case LTTNG_DISABLE_EVENT:
3478 {
3479
3480 /*
3481 * FIXME: handle filter; for now we just receive the filter's
3482 * bytecode along with the filter expression which are sent by
3483 * liblttng-ctl and discard them.
3484 *
3485 * This fixes an issue where the client may block while sending
3486 * the filter payload and encounter an error because the session
3487 * daemon closes the socket without ever handling this data.
3488 */
3489 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3490 cmd_ctx->lsm->u.disable.bytecode_len;
3491
3492 if (count) {
3493 char data[LTTNG_FILTER_MAX_LEN];
3494
3495 DBG("Discarding disable event command payload of size %zu", count);
3496 while (count) {
3497 ret = lttcomm_recv_unix_sock(sock, data,
3498 count > sizeof(data) ? sizeof(data) : count);
3499 if (ret < 0) {
3500 goto error;
3501 }
3502
3503 count -= (size_t) ret;
3504 }
3505 }
3506 /* FIXME: passing packed structure to non-packed pointer */
3507 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3508 cmd_ctx->lsm->u.disable.channel_name,
3509 &cmd_ctx->lsm->u.disable.event);
3510 break;
3511 }
3512 case LTTNG_ENABLE_CHANNEL:
3513 {
3514 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3515 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3516 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3517 &cmd_ctx->lsm->u.channel.chan,
3518 kernel_poll_pipe[1]);
3519 break;
3520 }
3521 case LTTNG_TRACK_PID:
3522 {
3523 ret = cmd_track_pid(cmd_ctx->session,
3524 cmd_ctx->lsm->domain.type,
3525 cmd_ctx->lsm->u.pid_tracker.pid);
3526 break;
3527 }
3528 case LTTNG_UNTRACK_PID:
3529 {
3530 ret = cmd_untrack_pid(cmd_ctx->session,
3531 cmd_ctx->lsm->domain.type,
3532 cmd_ctx->lsm->u.pid_tracker.pid);
3533 break;
3534 }
3535 case LTTNG_ENABLE_EVENT:
3536 {
3537 struct lttng_event_exclusion *exclusion = NULL;
3538 struct lttng_filter_bytecode *bytecode = NULL;
3539 char *filter_expression = NULL;
3540
3541 /* Handle exclusion events and receive it from the client. */
3542 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3543 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3544
3545 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3546 (count * LTTNG_SYMBOL_NAME_LEN));
3547 if (!exclusion) {
3548 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3549 goto error;
3550 }
3551
3552 DBG("Receiving var len exclusion event list from client ...");
3553 exclusion->count = count;
3554 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3555 count * LTTNG_SYMBOL_NAME_LEN);
3556 if (ret <= 0) {
3557 DBG("Nothing recv() from client var len data... continuing");
3558 *sock_error = 1;
3559 free(exclusion);
3560 ret = LTTNG_ERR_EXCLUSION_INVAL;
3561 goto error;
3562 }
3563 }
3564
3565 /* Get filter expression from client. */
3566 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3567 size_t expression_len =
3568 cmd_ctx->lsm->u.enable.expression_len;
3569
3570 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3571 ret = LTTNG_ERR_FILTER_INVAL;
3572 free(exclusion);
3573 goto error;
3574 }
3575
3576 filter_expression = zmalloc(expression_len);
3577 if (!filter_expression) {
3578 free(exclusion);
3579 ret = LTTNG_ERR_FILTER_NOMEM;
3580 goto error;
3581 }
3582
3583 /* Receive var. len. data */
3584 DBG("Receiving var len filter's expression from client ...");
3585 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3586 expression_len);
3587 if (ret <= 0) {
3588 DBG("Nothing recv() from client car len data... continuing");
3589 *sock_error = 1;
3590 free(filter_expression);
3591 free(exclusion);
3592 ret = LTTNG_ERR_FILTER_INVAL;
3593 goto error;
3594 }
3595 }
3596
3597 /* Handle filter and get bytecode from client. */
3598 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3599 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3600
3601 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3602 ret = LTTNG_ERR_FILTER_INVAL;
3603 free(filter_expression);
3604 free(exclusion);
3605 goto error;
3606 }
3607
3608 bytecode = zmalloc(bytecode_len);
3609 if (!bytecode) {
3610 free(filter_expression);
3611 free(exclusion);
3612 ret = LTTNG_ERR_FILTER_NOMEM;
3613 goto error;
3614 }
3615
3616 /* Receive var. len. data */
3617 DBG("Receiving var len filter's bytecode from client ...");
3618 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3619 if (ret <= 0) {
3620 DBG("Nothing recv() from client car len data... continuing");
3621 *sock_error = 1;
3622 free(filter_expression);
3623 free(bytecode);
3624 free(exclusion);
3625 ret = LTTNG_ERR_FILTER_INVAL;
3626 goto error;
3627 }
3628
3629 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3630 free(filter_expression);
3631 free(bytecode);
3632 free(exclusion);
3633 ret = LTTNG_ERR_FILTER_INVAL;
3634 goto error;
3635 }
3636 }
3637
3638 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3639 cmd_ctx->lsm->u.enable.channel_name,
3640 &cmd_ctx->lsm->u.enable.event,
3641 filter_expression, bytecode, exclusion,
3642 kernel_poll_pipe[1]);
3643 break;
3644 }
3645 case LTTNG_LIST_TRACEPOINTS:
3646 {
3647 struct lttng_event *events;
3648 ssize_t nb_events;
3649
3650 session_lock_list();
3651 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3652 session_unlock_list();
3653 if (nb_events < 0) {
3654 /* Return value is a negative lttng_error_code. */
3655 ret = -nb_events;
3656 goto error;
3657 }
3658
3659 /*
3660 * Setup lttng message with payload size set to the event list size in
3661 * bytes and then copy list into the llm payload.
3662 */
3663 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3664 sizeof(struct lttng_event) * nb_events);
3665 free(events);
3666
3667 if (ret < 0) {
3668 goto setup_error;
3669 }
3670
3671 ret = LTTNG_OK;
3672 break;
3673 }
3674 case LTTNG_LIST_TRACEPOINT_FIELDS:
3675 {
3676 struct lttng_event_field *fields;
3677 ssize_t nb_fields;
3678
3679 session_lock_list();
3680 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3681 &fields);
3682 session_unlock_list();
3683 if (nb_fields < 0) {
3684 /* Return value is a negative lttng_error_code. */
3685 ret = -nb_fields;
3686 goto error;
3687 }
3688
3689 /*
3690 * Setup lttng message with payload size set to the event list size in
3691 * bytes and then copy list into the llm payload.
3692 */
3693 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3694 sizeof(struct lttng_event_field) * nb_fields);
3695 free(fields);
3696
3697 if (ret < 0) {
3698 goto setup_error;
3699 }
3700
3701 ret = LTTNG_OK;
3702 break;
3703 }
3704 case LTTNG_LIST_SYSCALLS:
3705 {
3706 struct lttng_event *events;
3707 ssize_t nb_events;
3708
3709 nb_events = cmd_list_syscalls(&events);
3710 if (nb_events < 0) {
3711 /* Return value is a negative lttng_error_code. */
3712 ret = -nb_events;
3713 goto error;
3714 }
3715
3716 /*
3717 * Setup lttng message with payload size set to the event list size in
3718 * bytes and then copy list into the llm payload.
3719 */
3720 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3721 sizeof(struct lttng_event) * nb_events);
3722 free(events);
3723
3724 if (ret < 0) {
3725 goto setup_error;
3726 }
3727
3728 ret = LTTNG_OK;
3729 break;
3730 }
3731 case LTTNG_LIST_TRACKER_PIDS:
3732 {
3733 int32_t *pids = NULL;
3734 ssize_t nr_pids;
3735
3736 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3737 cmd_ctx->lsm->domain.type, &pids);
3738 if (nr_pids < 0) {
3739 /* Return value is a negative lttng_error_code. */
3740 ret = -nr_pids;
3741 goto error;
3742 }
3743
3744 /*
3745 * Setup lttng message with payload size set to the event list size in
3746 * bytes and then copy list into the llm payload.
3747 */
3748 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3749 sizeof(int32_t) * nr_pids);
3750 free(pids);
3751
3752 if (ret < 0) {
3753 goto setup_error;
3754 }
3755
3756 ret = LTTNG_OK;
3757 break;
3758 }
3759 case LTTNG_SET_CONSUMER_URI:
3760 {
3761 size_t nb_uri, len;
3762 struct lttng_uri *uris;
3763
3764 nb_uri = cmd_ctx->lsm->u.uri.size;
3765 len = nb_uri * sizeof(struct lttng_uri);
3766
3767 if (nb_uri == 0) {
3768 ret = LTTNG_ERR_INVALID;
3769 goto error;
3770 }
3771
3772 uris = zmalloc(len);
3773 if (uris == NULL) {
3774 ret = LTTNG_ERR_FATAL;
3775 goto error;
3776 }
3777
3778 /* Receive variable len data */
3779 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3780 ret = lttcomm_recv_unix_sock(sock, uris, len);
3781 if (ret <= 0) {
3782 DBG("No URIs received from client... continuing");
3783 *sock_error = 1;
3784 ret = LTTNG_ERR_SESSION_FAIL;
3785 free(uris);
3786 goto error;
3787 }
3788
3789 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3790 free(uris);
3791 if (ret != LTTNG_OK) {
3792 goto error;
3793 }
3794
3795
3796 break;
3797 }
3798 case LTTNG_START_TRACE:
3799 {
3800 ret = cmd_start_trace(cmd_ctx->session);
3801 break;
3802 }
3803 case LTTNG_STOP_TRACE:
3804 {
3805 ret = cmd_stop_trace(cmd_ctx->session);
3806 break;
3807 }
3808 case LTTNG_CREATE_SESSION:
3809 {
3810 size_t nb_uri, len;
3811 struct lttng_uri *uris = NULL;
3812
3813 nb_uri = cmd_ctx->lsm->u.uri.size;
3814 len = nb_uri * sizeof(struct lttng_uri);
3815
3816 if (nb_uri > 0) {
3817 uris = zmalloc(len);
3818 if (uris == NULL) {
3819 ret = LTTNG_ERR_FATAL;
3820 goto error;
3821 }
3822
3823 /* Receive variable len data */
3824 DBG("Waiting for %zu URIs from client ...", nb_uri);
3825 ret = lttcomm_recv_unix_sock(sock, uris, len);
3826 if (ret <= 0) {
3827 DBG("No URIs received from client... continuing");
3828 *sock_error = 1;
3829 ret = LTTNG_ERR_SESSION_FAIL;
3830 free(uris);
3831 goto error;
3832 }
3833
3834 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3835 DBG("Creating session with ONE network URI is a bad call");
3836 ret = LTTNG_ERR_SESSION_FAIL;
3837 free(uris);
3838 goto error;
3839 }
3840 }
3841
3842 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3843 &cmd_ctx->creds, 0);
3844
3845 free(uris);
3846
3847 break;
3848 }
3849 case LTTNG_DESTROY_SESSION:
3850 {
3851 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3852
3853 /* Set session to NULL so we do not unlock it after free. */
3854 cmd_ctx->session = NULL;
3855 break;
3856 }
3857 case LTTNG_LIST_DOMAINS:
3858 {
3859 ssize_t nb_dom;
3860 struct lttng_domain *domains = NULL;
3861
3862 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3863 if (nb_dom < 0) {
3864 /* Return value is a negative lttng_error_code. */
3865 ret = -nb_dom;
3866 goto error;
3867 }
3868
3869 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3870 nb_dom * sizeof(struct lttng_domain));
3871 free(domains);
3872
3873 if (ret < 0) {
3874 goto setup_error;
3875 }
3876
3877 ret = LTTNG_OK;
3878 break;
3879 }
3880 case LTTNG_LIST_CHANNELS:
3881 {
3882 ssize_t payload_size;
3883 struct lttng_channel *channels = NULL;
3884
3885 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3886 cmd_ctx->session, &channels);
3887 if (payload_size < 0) {
3888 /* Return value is a negative lttng_error_code. */
3889 ret = -payload_size;
3890 goto error;
3891 }
3892
3893 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3894 payload_size);
3895 free(channels);
3896
3897 if (ret < 0) {
3898 goto setup_error;
3899 }
3900
3901 ret = LTTNG_OK;
3902 break;
3903 }
3904 case LTTNG_LIST_EVENTS:
3905 {
3906 ssize_t nb_event;
3907 struct lttng_event *events = NULL;
3908 struct lttcomm_event_command_header cmd_header;
3909 size_t total_size;
3910
3911 memset(&cmd_header, 0, sizeof(cmd_header));
3912 /* Extended infos are included at the end of events */
3913 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3914 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3915 &events, &total_size);
3916
3917 if (nb_event < 0) {
3918 /* Return value is a negative lttng_error_code. */
3919 ret = -nb_event;
3920 goto error;
3921 }
3922
3923 cmd_header.nb_events = nb_event;
3924 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3925 &cmd_header, sizeof(cmd_header));
3926 free(events);
3927
3928 if (ret < 0) {
3929 goto setup_error;
3930 }
3931
3932 ret = LTTNG_OK;
3933 break;
3934 }
3935 case LTTNG_LIST_SESSIONS:
3936 {
3937 unsigned int nr_sessions;
3938 void *sessions_payload;
3939 size_t payload_len;
3940
3941 session_lock_list();
3942 nr_sessions = lttng_sessions_count(
3943 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3944 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3945 payload_len = sizeof(struct lttng_session) * nr_sessions;
3946 sessions_payload = zmalloc(payload_len);
3947
3948 if (!sessions_payload) {
3949 session_unlock_list();
3950 ret = -ENOMEM;
3951 goto setup_error;
3952 }
3953
3954 cmd_list_lttng_sessions(sessions_payload,
3955 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3956 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3957 session_unlock_list();
3958
3959 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3960 payload_len);
3961 free(sessions_payload);
3962
3963 if (ret < 0) {
3964 goto setup_error;
3965 }
3966
3967 ret = LTTNG_OK;
3968 break;
3969 }
3970 case LTTNG_REGISTER_CONSUMER:
3971 {
3972 struct consumer_data *cdata;
3973
3974 switch (cmd_ctx->lsm->domain.type) {
3975 case LTTNG_DOMAIN_KERNEL:
3976 cdata = &kconsumer_data;
3977 break;
3978 default:
3979 ret = LTTNG_ERR_UND;
3980 goto error;
3981 }
3982
3983 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3984 cmd_ctx->lsm->u.reg.path, cdata);
3985 break;
3986 }
3987 case LTTNG_DATA_PENDING:
3988 {
3989 int pending_ret;
3990 uint8_t pending_ret_byte;
3991
3992 pending_ret = cmd_data_pending(cmd_ctx->session);
3993
3994 /*
3995 * FIXME
3996 *
3997 * This function may returns 0 or 1 to indicate whether or not
3998 * there is data pending. In case of error, it should return an
3999 * LTTNG_ERR code. However, some code paths may still return
4000 * a nondescript error code, which we handle by returning an
4001 * "unknown" error.
4002 */
4003 if (pending_ret == 0 || pending_ret == 1) {
4004 /*
4005 * ret will be set to LTTNG_OK at the end of
4006 * this function.
4007 */
4008 } else if (pending_ret < 0) {
4009 ret = LTTNG_ERR_UNK;
4010 goto setup_error;
4011 } else {
4012 ret = pending_ret;
4013 goto setup_error;
4014 }
4015
4016 pending_ret_byte = (uint8_t) pending_ret;
4017
4018 /* 1 byte to return whether or not data is pending */
4019 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
4020 &pending_ret_byte, 1);
4021
4022 if (ret < 0) {
4023 goto setup_error;
4024 }
4025
4026 ret = LTTNG_OK;
4027 break;
4028 }
4029 case LTTNG_SNAPSHOT_ADD_OUTPUT:
4030 {
4031 struct lttcomm_lttng_output_id reply;
4032
4033 ret = cmd_snapshot_add_output(cmd_ctx->session,
4034 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
4035 if (ret != LTTNG_OK) {
4036 goto error;
4037 }
4038
4039 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
4040 sizeof(reply));
4041 if (ret < 0) {
4042 goto setup_error;
4043 }
4044
4045 /* Copy output list into message payload */
4046 ret = LTTNG_OK;
4047 break;
4048 }
4049 case LTTNG_SNAPSHOT_DEL_OUTPUT:
4050 {
4051 ret = cmd_snapshot_del_output(cmd_ctx->session,
4052 &cmd_ctx->lsm->u.snapshot_output.output);
4053 break;
4054 }
4055 case LTTNG_SNAPSHOT_LIST_OUTPUT:
4056 {
4057 ssize_t nb_output;
4058 struct lttng_snapshot_output *outputs = NULL;
4059
4060 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
4061 if (nb_output < 0) {
4062 ret = -nb_output;
4063 goto error;
4064 }
4065
4066 assert((nb_output > 0 && outputs) || nb_output == 0);
4067 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
4068 nb_output * sizeof(struct lttng_snapshot_output));
4069 free(outputs);
4070
4071 if (ret < 0) {
4072 goto setup_error;
4073 }
4074
4075 ret = LTTNG_OK;
4076 break;
4077 }
4078 case LTTNG_SNAPSHOT_RECORD:
4079 {
4080 ret = cmd_snapshot_record(cmd_ctx->session,
4081 &cmd_ctx->lsm->u.snapshot_record.output,
4082 cmd_ctx->lsm->u.snapshot_record.wait);
4083 break;
4084 }
4085 case LTTNG_CREATE_SESSION_SNAPSHOT:
4086 {
4087 size_t nb_uri, len;
4088 struct lttng_uri *uris = NULL;
4089
4090 nb_uri = cmd_ctx->lsm->u.uri.size;
4091 len = nb_uri * sizeof(struct lttng_uri);
4092
4093 if (nb_uri > 0) {
4094 uris = zmalloc(len);
4095 if (uris == NULL) {
4096 ret = LTTNG_ERR_FATAL;
4097 goto error;
4098 }
4099
4100 /* Receive variable len data */
4101 DBG("Waiting for %zu URIs from client ...", nb_uri);
4102 ret = lttcomm_recv_unix_sock(sock, uris, len);
4103 if (ret <= 0) {
4104 DBG("No URIs received from client... continuing");
4105 *sock_error = 1;
4106 ret = LTTNG_ERR_SESSION_FAIL;
4107 free(uris);
4108 goto error;
4109 }
4110
4111 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4112 DBG("Creating session with ONE network URI is a bad call");
4113 ret = LTTNG_ERR_SESSION_FAIL;
4114 free(uris);
4115 goto error;
4116 }
4117 }
4118
4119 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4120 nb_uri, &cmd_ctx->creds);
4121 free(uris);
4122 break;
4123 }
4124 case LTTNG_CREATE_SESSION_LIVE:
4125 {
4126 size_t nb_uri, len;
4127 struct lttng_uri *uris = NULL;
4128
4129 nb_uri = cmd_ctx->lsm->u.uri.size;
4130 len = nb_uri * sizeof(struct lttng_uri);
4131
4132 if (nb_uri > 0) {
4133 uris = zmalloc(len);
4134 if (uris == NULL) {
4135 ret = LTTNG_ERR_FATAL;
4136 goto error;
4137 }
4138
4139 /* Receive variable len data */
4140 DBG("Waiting for %zu URIs from client ...", nb_uri);
4141 ret = lttcomm_recv_unix_sock(sock, uris, len);
4142 if (ret <= 0) {
4143 DBG("No URIs received from client... continuing");
4144 *sock_error = 1;
4145 ret = LTTNG_ERR_SESSION_FAIL;
4146 free(uris);
4147 goto error;
4148 }
4149
4150 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4151 DBG("Creating session with ONE network URI is a bad call");
4152 ret = LTTNG_ERR_SESSION_FAIL;
4153 free(uris);
4154 goto error;
4155 }
4156 }
4157
4158 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4159 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4160 free(uris);
4161 break;
4162 }
4163 case LTTNG_SAVE_SESSION:
4164 {
4165 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4166 &cmd_ctx->creds);
4167 break;
4168 }
4169 case LTTNG_SET_SESSION_SHM_PATH:
4170 {
4171 ret = cmd_set_session_shm_path(cmd_ctx->session,
4172 cmd_ctx->lsm->u.set_shm_path.shm_path);
4173 break;
4174 }
4175 case LTTNG_REGENERATE_METADATA:
4176 {
4177 ret = cmd_regenerate_metadata(cmd_ctx->session);
4178 break;
4179 }
4180 case LTTNG_REGENERATE_STATEDUMP:
4181 {
4182 ret = cmd_regenerate_statedump(cmd_ctx->session);
4183 break;
4184 }
4185 case LTTNG_REGISTER_TRIGGER:
4186 {
4187 ret = cmd_register_trigger(cmd_ctx, sock,
4188 notification_thread_handle);
4189 break;
4190 }
4191 case LTTNG_UNREGISTER_TRIGGER:
4192 {
4193 ret = cmd_unregister_trigger(cmd_ctx, sock,
4194 notification_thread_handle);
4195 break;
4196 }
4197 default:
4198 ret = LTTNG_ERR_UND;
4199 break;
4200 }
4201
4202 error:
4203 if (cmd_ctx->llm == NULL) {
4204 DBG("Missing llm structure. Allocating one.");
4205 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4206 goto setup_error;
4207 }
4208 }
4209 /* Set return code */
4210 cmd_ctx->llm->ret_code = ret;
4211 setup_error:
4212 if (cmd_ctx->session) {
4213 session_unlock(cmd_ctx->session);
4214 }
4215 if (need_tracing_session) {
4216 session_unlock_list();
4217 }
4218 init_setup_error:
4219 assert(!rcu_read_ongoing());
4220 return ret;
4221 }
4222
4223 /*
4224 * Thread managing health check socket.
4225 */
4226 static void *thread_manage_health(void *data)
4227 {
4228 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4229 uint32_t revents, nb_fd;
4230 struct lttng_poll_event events;
4231 struct health_comm_msg msg;
4232 struct health_comm_reply reply;
4233
4234 DBG("[thread] Manage health check started");
4235
4236 rcu_register_thread();
4237
4238 /* We might hit an error path before this is created. */
4239 lttng_poll_init(&events);
4240
4241 /* Create unix socket */
4242 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4243 if (sock < 0) {
4244 ERR("Unable to create health check Unix socket");
4245 goto error;
4246 }
4247
4248 if (is_root) {
4249 /* lttng health client socket path permissions */
4250 ret = chown(health_unix_sock_path, 0,
4251 utils_get_group_id(tracing_group_name));
4252 if (ret < 0) {
4253 ERR("Unable to set group on %s", health_unix_sock_path);
4254 PERROR("chown");
4255 goto error;
4256 }
4257
4258 ret = chmod(health_unix_sock_path,
4259 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4260 if (ret < 0) {
4261 ERR("Unable to set permissions on %s", health_unix_sock_path);
4262 PERROR("chmod");
4263 goto error;
4264 }
4265 }
4266
4267 /*
4268 * Set the CLOEXEC flag. Return code is useless because either way, the
4269 * show must go on.
4270 */
4271 (void) utils_set_fd_cloexec(sock);
4272
4273 ret = lttcomm_listen_unix_sock(sock);
4274 if (ret < 0) {
4275 goto error;
4276 }
4277
4278 /*
4279 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4280 * more will be added to this poll set.
4281 */
4282 ret = sessiond_set_thread_pollset(&events, 2);
4283 if (ret < 0) {
4284 goto error;
4285 }
4286
4287 /* Add the application registration socket */
4288 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4289 if (ret < 0) {
4290 goto error;
4291 }
4292
4293 sessiond_notify_ready();
4294
4295 while (1) {
4296 DBG("Health check ready");
4297
4298 /* Inifinite blocking call, waiting for transmission */
4299 restart:
4300 ret = lttng_poll_wait(&events, -1);
4301 if (ret < 0) {
4302 /*
4303 * Restart interrupted system call.
4304 */
4305 if (errno == EINTR) {
4306 goto restart;
4307 }
4308 goto error;
4309 }
4310
4311 nb_fd = ret;
4312
4313 for (i = 0; i < nb_fd; i++) {
4314 /* Fetch once the poll data */
4315 revents = LTTNG_POLL_GETEV(&events, i);
4316 pollfd = LTTNG_POLL_GETFD(&events, i);
4317
4318 if (!revents) {
4319 /* No activity for this FD (poll implementation). */
4320 continue;
4321 }
4322
4323 /* Thread quit pipe has been closed. Killing thread. */
4324 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4325 if (ret) {
4326 err = 0;
4327 goto exit;
4328 }
4329
4330 /* Event on the registration socket */
4331 if (pollfd == sock) {
4332 if (revents & LPOLLIN) {
4333 continue;
4334 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4335 ERR("Health socket poll error");
4336 goto error;
4337 } else {
4338 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4339 goto error;
4340 }
4341 }
4342 }
4343
4344 new_sock = lttcomm_accept_unix_sock(sock);
4345 if (new_sock < 0) {
4346 goto error;
4347 }
4348
4349 /*
4350 * Set the CLOEXEC flag. Return code is useless because either way, the
4351 * show must go on.
4352 */
4353 (void) utils_set_fd_cloexec(new_sock);
4354
4355 DBG("Receiving data from client for health...");
4356 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4357 if (ret <= 0) {
4358 DBG("Nothing recv() from client... continuing");
4359 ret = close(new_sock);
4360 if (ret) {
4361 PERROR("close");
4362 }
4363 continue;
4364 }
4365
4366 rcu_thread_online();
4367
4368 memset(&reply, 0, sizeof(reply));
4369 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4370 /*
4371 * health_check_state returns 0 if health is
4372 * bad.
4373 */
4374 if (!health_check_state(health_sessiond, i)) {
4375 reply.ret_code |= 1ULL << i;
4376 }
4377 }
4378
4379 DBG2("Health check return value %" PRIx64, reply.ret_code);
4380
4381 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4382 if (ret < 0) {
4383 ERR("Failed to send health data back to client");
4384 }
4385
4386 /* End of transmission */
4387 ret = close(new_sock);
4388 if (ret) {
4389 PERROR("close");
4390 }
4391 }
4392
4393 exit:
4394 error:
4395 if (err) {
4396 ERR("Health error occurred in %s", __func__);
4397 }
4398 DBG("Health check thread dying");
4399 unlink(health_unix_sock_path);
4400 if (sock >= 0) {
4401 ret = close(sock);
4402 if (ret) {
4403 PERROR("close");
4404 }
4405 }
4406
4407 lttng_poll_clean(&events);
4408 stop_threads();
4409 rcu_unregister_thread();
4410 return NULL;
4411 }
4412
4413 /*
4414 * This thread manage all clients request using the unix client socket for
4415 * communication.
4416 */
4417 static void *thread_manage_clients(void *data)
4418 {
4419 int sock = -1, ret, i, pollfd, err = -1;
4420 int sock_error;
4421 uint32_t revents, nb_fd;
4422 struct command_ctx *cmd_ctx = NULL;
4423 struct lttng_poll_event events;
4424
4425 DBG("[thread] Manage client started");
4426
4427 rcu_register_thread();
4428
4429 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4430
4431 health_code_update();
4432
4433 ret = lttcomm_listen_unix_sock(client_sock);
4434 if (ret < 0) {
4435 goto error_listen;
4436 }
4437
4438 /*
4439 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4440 * more will be added to this poll set.
4441 */
4442 ret = sessiond_set_thread_pollset(&events, 2);
4443 if (ret < 0) {
4444 goto error_create_poll;
4445 }
4446
4447 /* Add the application registration socket */
4448 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4449 if (ret < 0) {
4450 goto error;
4451 }
4452
4453 sessiond_notify_ready();
4454 ret = sem_post(&load_info->message_thread_ready);
4455 if (ret) {
4456 PERROR("sem_post message_thread_ready");
4457 goto error;
4458 }
4459
4460 /* This testpoint is after we signal readiness to the parent. */
4461 if (testpoint(sessiond_thread_manage_clients)) {
4462 goto error;
4463 }
4464
4465 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4466 goto error;
4467 }
4468
4469 health_code_update();
4470
4471 while (1) {
4472 DBG("Accepting client command ...");
4473
4474 /* Inifinite blocking call, waiting for transmission */
4475 restart:
4476 health_poll_entry();
4477 ret = lttng_poll_wait(&events, -1);
4478 health_poll_exit();
4479 if (ret < 0) {
4480 /*
4481 * Restart interrupted system call.
4482 */
4483 if (errno == EINTR) {
4484 goto restart;
4485 }
4486 goto error;
4487 }
4488
4489 nb_fd = ret;
4490
4491 for (i = 0; i < nb_fd; i++) {
4492 /* Fetch once the poll data */
4493 revents = LTTNG_POLL_GETEV(&events, i);
4494 pollfd = LTTNG_POLL_GETFD(&events, i);
4495
4496 health_code_update();
4497
4498 if (!revents) {
4499 /* No activity for this FD (poll implementation). */
4500 continue;
4501 }
4502
4503 /* Thread quit pipe has been closed. Killing thread. */
4504 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4505 if (ret) {
4506 err = 0;
4507 goto exit;
4508 }
4509
4510 /* Event on the registration socket */
4511 if (pollfd == client_sock) {
4512 if (revents & LPOLLIN) {
4513 continue;
4514 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4515 ERR("Client socket poll error");
4516 goto error;
4517 } else {
4518 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4519 goto error;
4520 }
4521 }
4522 }
4523
4524 DBG("Wait for client response");
4525
4526 health_code_update();
4527
4528 sock = lttcomm_accept_unix_sock(client_sock);
4529 if (sock < 0) {
4530 goto error;
4531 }
4532
4533 /*
4534 * Set the CLOEXEC flag. Return code is useless because either way, the
4535 * show must go on.
4536 */
4537 (void) utils_set_fd_cloexec(sock);
4538
4539 /* Set socket option for credentials retrieval */
4540 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4541 if (ret < 0) {
4542 goto error;
4543 }
4544
4545 /* Allocate context command to process the client request */
4546 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4547 if (cmd_ctx == NULL) {
4548 PERROR("zmalloc cmd_ctx");
4549 goto error;
4550 }
4551
4552 /* Allocate data buffer for reception */
4553 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4554 if (cmd_ctx->lsm == NULL) {
4555 PERROR("zmalloc cmd_ctx->lsm");
4556 goto error;
4557 }
4558
4559 cmd_ctx->llm = NULL;
4560 cmd_ctx->session = NULL;
4561
4562 health_code_update();
4563
4564 /*
4565 * Data is received from the lttng client. The struct
4566 * lttcomm_session_msg (lsm) contains the command and data request of
4567 * the client.
4568 */
4569 DBG("Receiving data from client ...");
4570 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4571 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4572 if (ret <= 0) {
4573 DBG("Nothing recv() from client... continuing");
4574 ret = close(sock);
4575 if (ret) {
4576 PERROR("close");
4577 }
4578 sock = -1;
4579 clean_command_ctx(&cmd_ctx);
4580 continue;
4581 }
4582
4583 health_code_update();
4584
4585 // TODO: Validate cmd_ctx including sanity check for
4586 // security purpose.
4587
4588 rcu_thread_online();
4589 /*
4590 * This function dispatch the work to the kernel or userspace tracer
4591 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4592 * informations for the client. The command context struct contains
4593 * everything this function may needs.
4594 */
4595 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4596 rcu_thread_offline();
4597 if (ret < 0) {
4598 ret = close(sock);
4599 if (ret) {
4600 PERROR("close");
4601 }
4602 sock = -1;
4603 /*
4604 * TODO: Inform client somehow of the fatal error. At
4605 * this point, ret < 0 means that a zmalloc failed
4606 * (ENOMEM). Error detected but still accept
4607 * command, unless a socket error has been
4608 * detected.
4609 */
4610 clean_command_ctx(&cmd_ctx);
4611 continue;
4612 }
4613
4614 health_code_update();
4615
4616 DBG("Sending response (size: %d, retcode: %s (%d))",
4617 cmd_ctx->lttng_msg_size,
4618 lttng_strerror(-cmd_ctx->llm->ret_code),
4619 cmd_ctx->llm->ret_code);
4620 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4621 if (ret < 0) {
4622 ERR("Failed to send data back to client");
4623 }
4624
4625 /* End of transmission */
4626 ret = close(sock);
4627 if (ret) {
4628 PERROR("close");
4629 }
4630 sock = -1;
4631
4632 clean_command_ctx(&cmd_ctx);
4633
4634 health_code_update();
4635 }
4636
4637 exit:
4638 error:
4639 if (sock >= 0) {
4640 ret = close(sock);
4641 if (ret) {
4642 PERROR("close");
4643 }
4644 }
4645
4646 lttng_poll_clean(&events);
4647 clean_command_ctx(&cmd_ctx);
4648
4649 error_listen:
4650 error_create_poll:
4651 unlink(client_unix_sock_path);
4652 if (client_sock >= 0) {
4653 ret = close(client_sock);
4654 if (ret) {
4655 PERROR("close");
4656 }
4657 }
4658
4659 if (err) {
4660 health_error();
4661 ERR("Health error occurred in %s", __func__);
4662 }
4663
4664 health_unregister(health_sessiond);
4665
4666 DBG("Client thread dying");
4667
4668 rcu_unregister_thread();
4669
4670 /*
4671 * Since we are creating the consumer threads, we own them, so we need
4672 * to join them before our thread exits.
4673 */
4674 ret = join_consumer_thread(&kconsumer_data);
4675 if (ret) {
4676 errno = ret;
4677 PERROR("join_consumer");
4678 }
4679
4680 ret = join_consumer_thread(&ustconsumer32_data);
4681 if (ret) {
4682 errno = ret;
4683 PERROR("join_consumer ust32");
4684 }
4685
4686 ret = join_consumer_thread(&ustconsumer64_data);
4687 if (ret) {
4688 errno = ret;
4689 PERROR("join_consumer ust64");
4690 }
4691 return NULL;
4692 }
4693
4694 static int string_match(const char *str1, const char *str2)
4695 {
4696 return (str1 && str2) && !strcmp(str1, str2);
4697 }
4698
4699 /*
4700 * Take an option from the getopt output and set it in the right variable to be
4701 * used later.
4702 *
4703 * Return 0 on success else a negative value.
4704 */
4705 static int set_option(int opt, const char *arg, const char *optname)
4706 {
4707 int ret = 0;
4708
4709 if (string_match(optname, "client-sock") || opt == 'c') {
4710 if (!arg || *arg == '\0') {
4711 ret = -EINVAL;
4712 goto end;
4713 }
4714 if (lttng_is_setuid_setgid()) {
4715 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4716 "-c, --client-sock");
4717 } else {
4718 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4719 }
4720 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4721 if (!arg || *arg == '\0') {
4722 ret = -EINVAL;
4723 goto end;
4724 }
4725 if (lttng_is_setuid_setgid()) {
4726 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4727 "-a, --apps-sock");
4728 } else {
4729 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4730 }
4731 } else if (string_match(optname, "daemonize") || opt == 'd') {
4732 opt_daemon = 1;
4733 } else if (string_match(optname, "background") || opt == 'b') {
4734 opt_background = 1;
4735 } else if (string_match(optname, "group") || opt == 'g') {
4736 if (!arg || *arg == '\0') {
4737 ret = -EINVAL;
4738 goto end;
4739 }
4740 if (lttng_is_setuid_setgid()) {
4741 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4742 "-g, --group");
4743 } else {
4744 /*
4745 * If the override option is set, the pointer points to a
4746 * *non* const thus freeing it even though the variable type is
4747 * set to const.
4748 */
4749 if (tracing_group_name_override) {
4750 free((void *) tracing_group_name);
4751 }
4752 tracing_group_name = strdup(arg);
4753 if (!tracing_group_name) {
4754 PERROR("strdup");
4755 ret = -ENOMEM;
4756 }
4757 tracing_group_name_override = 1;
4758 }
4759 } else if (string_match(optname, "help") || opt == 'h') {
4760 ret = utils_show_man_page(8, "lttng-sessiond");
4761 if (ret) {
4762 ERR("Cannot view man page lttng-sessiond(8)");
4763 perror("exec");
4764 }
4765 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4766 } else if (string_match(optname, "version") || opt == 'V') {
4767 fprintf(stdout, "%s\n", VERSION);
4768 exit(EXIT_SUCCESS);
4769 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4770 opt_sig_parent = 1;
4771 } else if (string_match(optname, "kconsumerd-err-sock")) {
4772 if (!arg || *arg == '\0') {
4773 ret = -EINVAL;
4774 goto end;
4775 }
4776 if (lttng_is_setuid_setgid()) {
4777 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4778 "--kconsumerd-err-sock");
4779 } else {
4780 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4781 }
4782 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4783 if (!arg || *arg == '\0') {
4784 ret = -EINVAL;
4785 goto end;
4786 }
4787 if (lttng_is_setuid_setgid()) {
4788 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4789 "--kconsumerd-cmd-sock");
4790 } else {
4791 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4792 }
4793 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4794 if (!arg || *arg == '\0') {
4795 ret = -EINVAL;
4796 goto end;
4797 }
4798 if (lttng_is_setuid_setgid()) {
4799 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4800 "--ustconsumerd64-err-sock");
4801 } else {
4802 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4803 }
4804 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4805 if (!arg || *arg == '\0') {
4806 ret = -EINVAL;
4807 goto end;
4808 }
4809 if (lttng_is_setuid_setgid()) {
4810 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4811 "--ustconsumerd64-cmd-sock");
4812 } else {
4813 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4814 }
4815 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4816 if (!arg || *arg == '\0') {
4817 ret = -EINVAL;
4818 goto end;
4819 }
4820 if (lttng_is_setuid_setgid()) {
4821 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4822 "--ustconsumerd32-err-sock");
4823 } else {
4824 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4825 }
4826 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4827 if (!arg || *arg == '\0') {
4828 ret = -EINVAL;
4829 goto end;
4830 }
4831 if (lttng_is_setuid_setgid()) {
4832 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4833 "--ustconsumerd32-cmd-sock");
4834 } else {
4835 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4836 }
4837 } else if (string_match(optname, "no-kernel")) {
4838 opt_no_kernel = 1;
4839 } else if (string_match(optname, "quiet") || opt == 'q') {
4840 lttng_opt_quiet = 1;
4841 } else if (string_match(optname, "verbose") || opt == 'v') {
4842 /* Verbose level can increase using multiple -v */
4843 if (arg) {
4844 /* Value obtained from config file */
4845 lttng_opt_verbose = config_parse_value(arg);
4846 } else {
4847 /* -v used on command line */
4848 lttng_opt_verbose++;
4849 }
4850 /* Clamp value to [0, 3] */
4851 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4852 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4853 } else if (string_match(optname, "verbose-consumer")) {
4854 if (arg) {
4855 opt_verbose_consumer = config_parse_value(arg);
4856 } else {
4857 opt_verbose_consumer++;
4858 }
4859 } else if (string_match(optname, "consumerd32-path")) {
4860 if (!arg || *arg == '\0') {
4861 ret = -EINVAL;
4862 goto end;
4863 }
4864 if (lttng_is_setuid_setgid()) {
4865 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4866 "--consumerd32-path");
4867 } else {
4868 if (consumerd32_bin_override) {
4869 free((void *) consumerd32_bin);
4870 }
4871 consumerd32_bin = strdup(arg);
4872 if (!consumerd32_bin) {
4873 PERROR("strdup");
4874 ret = -ENOMEM;
4875 }
4876 consumerd32_bin_override = 1;
4877 }
4878 } else if (string_match(optname, "consumerd32-libdir")) {
4879 if (!arg || *arg == '\0') {
4880 ret = -EINVAL;
4881 goto end;
4882 }
4883 if (lttng_is_setuid_setgid()) {
4884 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4885 "--consumerd32-libdir");
4886 } else {
4887 if (consumerd32_libdir_override) {
4888 free((void *) consumerd32_libdir);
4889 }
4890 consumerd32_libdir = strdup(arg);
4891 if (!consumerd32_libdir) {
4892 PERROR("strdup");
4893 ret = -ENOMEM;
4894 }
4895 consumerd32_libdir_override = 1;
4896 }
4897 } else if (string_match(optname, "consumerd64-path")) {
4898 if (!arg || *arg == '\0') {
4899 ret = -EINVAL;
4900 goto end;
4901 }
4902 if (lttng_is_setuid_setgid()) {
4903 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4904 "--consumerd64-path");
4905 } else {
4906 if (consumerd64_bin_override) {
4907 free((void *) consumerd64_bin);
4908 }
4909 consumerd64_bin = strdup(arg);
4910 if (!consumerd64_bin) {
4911 PERROR("strdup");
4912 ret = -ENOMEM;
4913 }
4914 consumerd64_bin_override = 1;
4915 }
4916 } else if (string_match(optname, "consumerd64-libdir")) {
4917 if (!arg || *arg == '\0') {
4918 ret = -EINVAL;
4919 goto end;
4920 }
4921 if (lttng_is_setuid_setgid()) {
4922 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4923 "--consumerd64-libdir");
4924 } else {
4925 if (consumerd64_libdir_override) {
4926 free((void *) consumerd64_libdir);
4927 }
4928 consumerd64_libdir = strdup(arg);
4929 if (!consumerd64_libdir) {
4930 PERROR("strdup");
4931 ret = -ENOMEM;
4932 }
4933 consumerd64_libdir_override = 1;
4934 }
4935 } else if (string_match(optname, "pidfile") || opt == 'p') {
4936 if (!arg || *arg == '\0') {
4937 ret = -EINVAL;
4938 goto end;
4939 }
4940 if (lttng_is_setuid_setgid()) {
4941 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4942 "-p, --pidfile");
4943 } else {
4944 free(opt_pidfile);
4945 opt_pidfile = strdup(arg);
4946 if (!opt_pidfile) {
4947 PERROR("strdup");
4948 ret = -ENOMEM;
4949 }
4950 }
4951 } else if (string_match(optname, "agent-tcp-port")) {
4952 if (!arg || *arg == '\0') {
4953 ret = -EINVAL;
4954 goto end;
4955 }
4956 if (lttng_is_setuid_setgid()) {
4957 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4958 "--agent-tcp-port");
4959 } else {
4960 unsigned long v;
4961
4962 errno = 0;
4963 v = strtoul(arg, NULL, 0);
4964 if (errno != 0 || !isdigit(arg[0])) {
4965 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4966 return -1;
4967 }
4968 if (v == 0 || v >= 65535) {
4969 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4970 return -1;
4971 }
4972 agent_tcp_port = (uint32_t) v;
4973 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4974 }
4975 } else if (string_match(optname, "load") || opt == 'l') {
4976 if (!arg || *arg == '\0') {
4977 ret = -EINVAL;
4978 goto end;
4979 }
4980 if (lttng_is_setuid_setgid()) {
4981 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4982 "-l, --load");
4983 } else {
4984 free(opt_load_session_path);
4985 opt_load_session_path = strdup(arg);
4986 if (!opt_load_session_path) {
4987 PERROR("strdup");
4988 ret = -ENOMEM;
4989 }
4990 }
4991 } else if (string_match(optname, "kmod-probes")) {
4992 if (!arg || *arg == '\0') {
4993 ret = -EINVAL;
4994 goto end;
4995 }
4996 if (lttng_is_setuid_setgid()) {
4997 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4998 "--kmod-probes");
4999 } else {
5000 free(kmod_probes_list);
5001 kmod_probes_list = strdup(arg);
5002 if (!kmod_probes_list) {
5003 PERROR("strdup");
5004 ret = -ENOMEM;
5005 }
5006 }
5007 } else if (string_match(optname, "extra-kmod-probes")) {
5008 if (!arg || *arg == '\0') {
5009 ret = -EINVAL;
5010 goto end;
5011 }
5012 if (lttng_is_setuid_setgid()) {
5013 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5014 "--extra-kmod-probes");
5015 } else {
5016 free(kmod_extra_probes_list);
5017 kmod_extra_probes_list = strdup(arg);
5018 if (!kmod_extra_probes_list) {
5019 PERROR("strdup");
5020 ret = -ENOMEM;
5021 }
5022 }
5023 } else if (string_match(optname, "config") || opt == 'f') {
5024 /* This is handled in set_options() thus silent skip. */
5025 goto end;
5026 } else {
5027 /* Unknown option or other error.
5028 * Error is printed by getopt, just return */
5029 ret = -1;
5030 }
5031
5032 end:
5033 if (ret == -EINVAL) {
5034 const char *opt_name = "unknown";
5035 int i;
5036
5037 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
5038 i++) {
5039 if (opt == long_options[i].val) {
5040 opt_name = long_options[i].name;
5041 break;
5042 }
5043 }
5044
5045 WARN("Invalid argument provided for option \"%s\", using default value.",
5046 opt_name);
5047 }
5048
5049 return ret;
5050 }
5051
5052 /*
5053 * config_entry_handler_cb used to handle options read from a config file.
5054 * See config_entry_handler_cb comment in common/config/session-config.h for the
5055 * return value conventions.
5056 */
5057 static int config_entry_handler(const struct config_entry *entry, void *unused)
5058 {
5059 int ret = 0, i;
5060
5061 if (!entry || !entry->name || !entry->value) {
5062 ret = -EINVAL;
5063 goto end;
5064 }
5065
5066 /* Check if the option is to be ignored */
5067 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5068 if (!strcmp(entry->name, config_ignore_options[i])) {
5069 goto end;
5070 }
5071 }
5072
5073 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5074 i++) {
5075
5076 /* Ignore if not fully matched. */
5077 if (strcmp(entry->name, long_options[i].name)) {
5078 continue;
5079 }
5080
5081 /*
5082 * If the option takes no argument on the command line, we have to
5083 * check if the value is "true". We support non-zero numeric values,
5084 * true, on and yes.
5085 */
5086 if (!long_options[i].has_arg) {
5087 ret = config_parse_value(entry->value);
5088 if (ret <= 0) {
5089 if (ret) {
5090 WARN("Invalid configuration value \"%s\" for option %s",
5091 entry->value, entry->name);
5092 }
5093 /* False, skip boolean config option. */
5094 goto end;
5095 }
5096 }
5097
5098 ret = set_option(long_options[i].val, entry->value, entry->name);
5099 goto end;
5100 }
5101
5102 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5103
5104 end:
5105 return ret;
5106 }
5107
5108 /*
5109 * daemon configuration loading and argument parsing
5110 */
5111 static int set_options(int argc, char **argv)
5112 {
5113 int ret = 0, c = 0, option_index = 0;
5114 int orig_optopt = optopt, orig_optind = optind;
5115 char *optstring;
5116 const char *config_path = NULL;
5117
5118 optstring = utils_generate_optstring(long_options,
5119 sizeof(long_options) / sizeof(struct option));
5120 if (!optstring) {
5121 ret = -ENOMEM;
5122 goto end;
5123 }
5124
5125 /* Check for the --config option */
5126 while ((c = getopt_long(argc, argv, optstring, long_options,
5127 &option_index)) != -1) {
5128 if (c == '?') {
5129 ret = -EINVAL;
5130 goto end;
5131 } else if (c != 'f') {
5132 /* if not equal to --config option. */
5133 continue;
5134 }
5135
5136 if (lttng_is_setuid_setgid()) {
5137 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5138 "-f, --config");
5139 } else {
5140 config_path = utils_expand_path(optarg);
5141 if (!config_path) {
5142 ERR("Failed to resolve path: %s", optarg);
5143 }
5144 }
5145 }
5146
5147 ret = config_get_section_entries(config_path, config_section_name,
5148 config_entry_handler, NULL);
5149 if (ret) {
5150 if (ret > 0) {
5151 ERR("Invalid configuration option at line %i", ret);
5152 ret = -1;
5153 }
5154 goto end;
5155 }
5156
5157 /* Reset getopt's global state */
5158 optopt = orig_optopt;
5159 optind = orig_optind;
5160 while (1) {
5161 option_index = -1;
5162 /*
5163 * getopt_long() will not set option_index if it encounters a
5164 * short option.
5165 */
5166 c = getopt_long(argc, argv, optstring, long_options,
5167 &option_index);
5168 if (c == -1) {
5169 break;
5170 }
5171
5172 /*
5173 * Pass NULL as the long option name if popt left the index
5174 * unset.
5175 */
5176 ret = set_option(c, optarg,
5177 option_index < 0 ? NULL :
5178 long_options[option_index].name);
5179 if (ret < 0) {
5180 break;
5181 }
5182 }
5183
5184 end:
5185 free(optstring);
5186 return ret;
5187 }
5188
5189 /*
5190 * Creates the two needed socket by the daemon.
5191 * apps_sock - The communication socket for all UST apps.
5192 * client_sock - The communication of the cli tool (lttng).
5193 */
5194 static int init_daemon_socket(void)
5195 {
5196 int ret = 0;
5197 mode_t old_umask;
5198
5199 old_umask = umask(0);
5200
5201 /* Create client tool unix socket */
5202 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5203 if (client_sock < 0) {
5204 ERR("Create unix sock failed: %s", client_unix_sock_path);
5205 ret = -1;
5206 goto end;
5207 }
5208
5209 /* Set the cloexec flag */
5210 ret = utils_set_fd_cloexec(client_sock);
5211 if (ret < 0) {
5212 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5213 "Continuing but note that the consumer daemon will have a "
5214 "reference to this socket on exec()", client_sock);
5215 }
5216
5217 /* File permission MUST be 660 */
5218 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5219 if (ret < 0) {
5220 ERR("Set file permissions failed: %s", client_unix_sock_path);
5221 PERROR("chmod");
5222 goto end;
5223 }
5224
5225 /* Create the application unix socket */
5226 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5227 if (apps_sock < 0) {
5228 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5229 ret = -1;
5230 goto end;
5231 }
5232
5233 /* Set the cloexec flag */
5234 ret = utils_set_fd_cloexec(apps_sock);
5235 if (ret < 0) {
5236 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5237 "Continuing but note that the consumer daemon will have a "
5238 "reference to this socket on exec()", apps_sock);
5239 }
5240
5241 /* File permission MUST be 666 */
5242 ret = chmod(apps_unix_sock_path,
5243 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5244 if (ret < 0) {
5245 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5246 PERROR("chmod");
5247 goto end;
5248 }
5249
5250 DBG3("Session daemon client socket %d and application socket %d created",
5251 client_sock, apps_sock);
5252
5253 end:
5254 umask(old_umask);
5255 return ret;
5256 }
5257
5258 /*
5259 * Check if the global socket is available, and if a daemon is answering at the
5260 * other side. If yes, error is returned.
5261 */
5262 static int check_existing_daemon(void)
5263 {
5264 /* Is there anybody out there ? */
5265 if (lttng_session_daemon_alive()) {
5266 return -EEXIST;
5267 }
5268
5269 return 0;
5270 }
5271
5272 /*
5273 * Set the tracing group gid onto the client socket.
5274 *
5275 * Race window between mkdir and chown is OK because we are going from more
5276 * permissive (root.root) to less permissive (root.tracing).
5277 */
5278 static int set_permissions(char *rundir)
5279 {
5280 int ret;
5281 gid_t gid;
5282
5283 gid = utils_get_group_id(tracing_group_name);
5284
5285 /* Set lttng run dir */
5286 ret = chown(rundir, 0, gid);
5287 if (ret < 0) {
5288 ERR("Unable to set group on %s", rundir);
5289 PERROR("chown");
5290 }
5291
5292 /*
5293 * Ensure all applications and tracing group can search the run
5294 * dir. Allow everyone to read the directory, since it does not
5295 * buy us anything to hide its content.
5296 */
5297 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5298 if (ret < 0) {
5299 ERR("Unable to set permissions on %s", rundir);
5300 PERROR("chmod");
5301 }
5302
5303 /* lttng client socket path */
5304 ret = chown(client_unix_sock_path, 0, gid);
5305 if (ret < 0) {
5306 ERR("Unable to set group on %s", client_unix_sock_path);
5307 PERROR("chown");
5308 }
5309
5310 /* kconsumer error socket path */
5311 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5312 if (ret < 0) {
5313 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5314 PERROR("chown");
5315 }
5316
5317 /* 64-bit ustconsumer error socket path */
5318 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5319 if (ret < 0) {
5320 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5321 PERROR("chown");
5322 }
5323
5324 /* 32-bit ustconsumer compat32 error socket path */
5325 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5326 if (ret < 0) {
5327 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5328 PERROR("chown");
5329 }
5330
5331 DBG("All permissions are set");
5332
5333 return ret;
5334 }
5335
5336 /*
5337 * Create the lttng run directory needed for all global sockets and pipe.
5338 */
5339 static int create_lttng_rundir(const char *rundir)
5340 {
5341 int ret;
5342
5343 DBG3("Creating LTTng run directory: %s", rundir);
5344
5345 ret = mkdir(rundir, S_IRWXU);
5346 if (ret < 0) {
5347 if (errno != EEXIST) {
5348 ERR("Unable to create %s", rundir);
5349 goto error;
5350 } else {
5351 ret = 0;
5352 }
5353 }
5354
5355 error:
5356 return ret;
5357 }
5358
5359 /*
5360 * Setup sockets and directory needed by the kconsumerd communication with the
5361 * session daemon.
5362 */
5363 static int set_consumer_sockets(struct consumer_data *consumer_data,
5364 const char *rundir)
5365 {
5366 int ret;
5367 char path[PATH_MAX];
5368
5369 switch (consumer_data->type) {
5370 case LTTNG_CONSUMER_KERNEL:
5371 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5372 break;
5373 case LTTNG_CONSUMER64_UST:
5374 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5375 break;
5376 case LTTNG_CONSUMER32_UST:
5377 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5378 break;
5379 default:
5380 ERR("Consumer type unknown");
5381 ret = -EINVAL;
5382 goto error;
5383 }
5384
5385 DBG2("Creating consumer directory: %s", path);
5386
5387 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5388 if (ret < 0) {
5389 if (errno != EEXIST) {
5390 PERROR("mkdir");
5391 ERR("Failed to create %s", path);
5392 goto error;
5393 }
5394 ret = -1;
5395 }
5396 if (is_root) {
5397 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5398 if (ret < 0) {
5399 ERR("Unable to set group on %s", path);
5400 PERROR("chown");
5401 goto error;
5402 }
5403 }
5404
5405 /* Create the kconsumerd error unix socket */
5406 consumer_data->err_sock =
5407 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5408 if (consumer_data->err_sock < 0) {
5409 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5410 ret = -1;
5411 goto error;
5412 }
5413
5414 /*
5415 * Set the CLOEXEC flag. Return code is useless because either way, the
5416 * show must go on.
5417 */
5418 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5419 if (ret < 0) {
5420 PERROR("utils_set_fd_cloexec");
5421 /* continue anyway */
5422 }
5423
5424 /* File permission MUST be 660 */
5425 ret = chmod(consumer_data->err_unix_sock_path,
5426 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5427 if (ret < 0) {
5428 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5429 PERROR("chmod");
5430 goto error;
5431 }
5432
5433 error:
5434 return ret;
5435 }
5436
5437 /*
5438 * Signal handler for the daemon
5439 *
5440 * Simply stop all worker threads, leaving main() return gracefully after
5441 * joining all threads and calling cleanup().
5442 */
5443 static void sighandler(int sig)
5444 {
5445 switch (sig) {
5446 case SIGINT:
5447 DBG("SIGINT caught");
5448 stop_threads();
5449 break;
5450 case SIGTERM:
5451 DBG("SIGTERM caught");
5452 stop_threads();
5453 break;
5454 case SIGUSR1:
5455 CMM_STORE_SHARED(recv_child_signal, 1);
5456 break;
5457 default:
5458 break;
5459 }
5460 }
5461
5462 /*
5463 * Setup signal handler for :
5464 * SIGINT, SIGTERM, SIGPIPE
5465 */
5466 static int set_signal_handler(void)
5467 {
5468 int ret = 0;
5469 struct sigaction sa;
5470 sigset_t sigset;
5471
5472 if ((ret = sigemptyset(&sigset)) < 0) {
5473 PERROR("sigemptyset");
5474 return ret;
5475 }
5476
5477 sa.sa_mask = sigset;
5478 sa.sa_flags = 0;
5479
5480 sa.sa_handler = sighandler;
5481 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5482 PERROR("sigaction");
5483 return ret;
5484 }
5485
5486 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5487 PERROR("sigaction");
5488 return ret;
5489 }
5490
5491 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5492 PERROR("sigaction");
5493 return ret;
5494 }
5495
5496 sa.sa_handler = SIG_IGN;
5497 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5498 PERROR("sigaction");
5499 return ret;
5500 }
5501
5502 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5503
5504 return ret;
5505 }
5506
5507 /*
5508 * Set open files limit to unlimited. This daemon can open a large number of
5509 * file descriptors in order to consume multiple kernel traces.
5510 */
5511 static void set_ulimit(void)
5512 {
5513 int ret;
5514 struct rlimit lim;
5515
5516 /* The kernel does not allow an infinite limit for open files */
5517 lim.rlim_cur = 65535;
5518 lim.rlim_max = 65535;
5519
5520 ret = setrlimit(RLIMIT_NOFILE, &lim);
5521 if (ret < 0) {
5522 PERROR("failed to set open files limit");
5523 }
5524 }
5525
5526 /*
5527 * Write pidfile using the rundir and opt_pidfile.
5528 */
5529 static int write_pidfile(void)
5530 {
5531 int ret;
5532 char pidfile_path[PATH_MAX];
5533
5534 assert(rundir);
5535
5536 if (opt_pidfile) {
5537 if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) {
5538 ret = -1;
5539 goto error;
5540 }
5541 } else {
5542 /* Build pidfile path from rundir and opt_pidfile. */
5543 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5544 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5545 if (ret < 0) {
5546 PERROR("snprintf pidfile path");
5547 goto error;
5548 }
5549 }
5550
5551 /*
5552 * Create pid file in rundir.
5553 */
5554 ret = utils_create_pid_file(getpid(), pidfile_path);
5555 error:
5556 return ret;
5557 }
5558
5559 /*
5560 * Create lockfile using the rundir and return its fd.
5561 */
5562 static int create_lockfile(void)
5563 {
5564 int ret;
5565 char lockfile_path[PATH_MAX];
5566
5567 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5568 if (ret < 0) {
5569 goto error;
5570 }
5571
5572 ret = utils_create_lock_file(lockfile_path);
5573 error:
5574 return ret;
5575 }
5576
5577 /*
5578 * Write agent TCP port using the rundir.
5579 */
5580 static int write_agent_port(void)
5581 {
5582 int ret;
5583 char path[PATH_MAX];
5584
5585 assert(rundir);
5586
5587 ret = snprintf(path, sizeof(path), "%s/"
5588 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5589 if (ret < 0) {
5590 PERROR("snprintf agent port path");
5591 goto error;
5592 }
5593
5594 /*
5595 * Create TCP agent port file in rundir.
5596 */
5597 ret = utils_create_pid_file(agent_tcp_port, path);
5598
5599 error:
5600 return ret;
5601 }
5602
5603 /*
5604 * main
5605 */
5606 int main(int argc, char **argv)
5607 {
5608 int ret = 0, retval = 0;
5609 void *status;
5610 const char *home_path, *env_app_timeout;
5611 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5612 *ust64_channel_monitor_pipe = NULL,
5613 *kernel_channel_monitor_pipe = NULL;
5614
5615 init_kernel_workarounds();
5616
5617 rcu_register_thread();
5618
5619 if (set_signal_handler()) {
5620 retval = -1;
5621 goto exit_set_signal_handler;
5622 }
5623
5624 setup_consumerd_path();
5625
5626 page_size = sysconf(_SC_PAGESIZE);
5627 if (page_size < 0) {
5628 PERROR("sysconf _SC_PAGESIZE");
5629 page_size = LONG_MAX;
5630 WARN("Fallback page size to %ld", page_size);
5631 }
5632
5633 /*
5634 * Parse arguments and load the daemon configuration file.
5635 *
5636 * We have an exit_options exit path to free memory reserved by
5637 * set_options. This is needed because the rest of sessiond_cleanup()
5638 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5639 * depends on set_options.
5640 */
5641 progname = argv[0];
5642 if (set_options(argc, argv)) {
5643 retval = -1;
5644 goto exit_options;
5645 }
5646
5647 /* Daemonize */
5648 if (opt_daemon || opt_background) {
5649 int i;
5650
5651 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5652 !opt_background);
5653 if (ret < 0) {
5654 retval = -1;
5655 goto exit_options;
5656 }
5657
5658 /*
5659 * We are in the child. Make sure all other file descriptors are
5660 * closed, in case we are called with more opened file
5661 * descriptors than the standard ones.
5662 */
5663 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5664 (void) close(i);
5665 }
5666 }
5667
5668 if (run_as_create_worker(argv[0]) < 0) {
5669 goto exit_create_run_as_worker_cleanup;
5670 }
5671
5672 /*
5673 * Starting from here, we can create threads. This needs to be after
5674 * lttng_daemonize due to RCU.
5675 */
5676
5677 /*
5678 * Initialize the health check subsystem. This call should set the
5679 * appropriate time values.
5680 */
5681 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5682 if (!health_sessiond) {
5683 PERROR("health_app_create error");
5684 retval = -1;
5685 goto exit_health_sessiond_cleanup;
5686 }
5687
5688 /* Create thread to clean up RCU hash tables */
5689 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5690 retval = -1;
5691 goto exit_ht_cleanup;
5692 }
5693
5694 /* Create thread quit pipe */
5695 if (init_thread_quit_pipe()) {
5696 retval = -1;
5697 goto exit_init_data;
5698 }
5699
5700 /* Check if daemon is UID = 0 */
5701 is_root = !getuid();
5702
5703 if (is_root) {
5704 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5705 if (!rundir) {
5706 retval = -1;
5707 goto exit_init_data;
5708 }
5709
5710 /* Create global run dir with root access */
5711 if (create_lttng_rundir(rundir)) {
5712 retval = -1;
5713 goto exit_init_data;
5714 }
5715
5716 if (strlen(apps_unix_sock_path) == 0) {
5717 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5718 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5719 if (ret < 0) {
5720 retval = -1;
5721 goto exit_init_data;
5722 }
5723 }
5724
5725 if (strlen(client_unix_sock_path) == 0) {
5726 ret = snprintf(client_unix_sock_path, PATH_MAX,
5727 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5728 if (ret < 0) {
5729 retval = -1;
5730 goto exit_init_data;
5731 }
5732 }
5733
5734 /* Set global SHM for ust */
5735 if (strlen(wait_shm_path) == 0) {
5736 ret = snprintf(wait_shm_path, PATH_MAX,
5737 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5738 if (ret < 0) {
5739 retval = -1;
5740 goto exit_init_data;
5741 }
5742 }
5743
5744 if (strlen(health_unix_sock_path) == 0) {
5745 ret = snprintf(health_unix_sock_path,
5746 sizeof(health_unix_sock_path),
5747 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5748 if (ret < 0) {
5749 retval = -1;
5750 goto exit_init_data;
5751 }
5752 }
5753
5754 /* Setup kernel consumerd path */
5755 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5756 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5757 if (ret < 0) {
5758 retval = -1;
5759 goto exit_init_data;
5760 }
5761 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5762 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5763 if (ret < 0) {
5764 retval = -1;
5765 goto exit_init_data;
5766 }
5767
5768 DBG2("Kernel consumer err path: %s",
5769 kconsumer_data.err_unix_sock_path);
5770 DBG2("Kernel consumer cmd path: %s",
5771 kconsumer_data.cmd_unix_sock_path);
5772 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5773 if (!kernel_channel_monitor_pipe) {
5774 ERR("Failed to create kernel consumer channel monitor pipe");
5775 retval = -1;
5776 goto exit_init_data;
5777 }
5778 kconsumer_data.channel_monitor_pipe =
5779 lttng_pipe_release_writefd(
5780 kernel_channel_monitor_pipe);
5781 if (kconsumer_data.channel_monitor_pipe < 0) {
5782 retval = -1;
5783 goto exit_init_data;
5784 }
5785 } else {
5786 home_path = utils_get_home_dir();
5787 if (home_path == NULL) {
5788 /* TODO: Add --socket PATH option */
5789 ERR("Can't get HOME directory for sockets creation.");
5790 retval = -1;
5791 goto exit_init_data;
5792 }
5793
5794 /*
5795 * Create rundir from home path. This will create something like
5796 * $HOME/.lttng
5797 */
5798 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5799 if (ret < 0) {
5800 retval = -1;
5801 goto exit_init_data;
5802 }
5803
5804 if (create_lttng_rundir(rundir)) {
5805 retval = -1;
5806 goto exit_init_data;
5807 }
5808
5809 if (strlen(apps_unix_sock_path) == 0) {
5810 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5811 DEFAULT_HOME_APPS_UNIX_SOCK,
5812 home_path);
5813 if (ret < 0) {
5814 retval = -1;
5815 goto exit_init_data;
5816 }
5817 }
5818
5819 /* Set the cli tool unix socket path */
5820 if (strlen(client_unix_sock_path) == 0) {
5821 ret = snprintf(client_unix_sock_path, PATH_MAX,
5822 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5823 home_path);
5824 if (ret < 0) {
5825 retval = -1;
5826 goto exit_init_data;
5827 }
5828 }
5829
5830 /* Set global SHM for ust */
5831 if (strlen(wait_shm_path) == 0) {
5832 ret = snprintf(wait_shm_path, PATH_MAX,
5833 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5834 getuid());
5835 if (ret < 0) {
5836 retval = -1;
5837 goto exit_init_data;
5838 }
5839 }
5840
5841 /* Set health check Unix path */
5842 if (strlen(health_unix_sock_path) == 0) {
5843 ret = snprintf(health_unix_sock_path,
5844 sizeof(health_unix_sock_path),
5845 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5846 home_path);
5847 if (ret < 0) {
5848 retval = -1;
5849 goto exit_init_data;
5850 }
5851 }
5852 }
5853
5854 lockfile_fd = create_lockfile();
5855 if (lockfile_fd < 0) {
5856 retval = -1;
5857 goto exit_init_data;
5858 }
5859
5860 /* Set consumer initial state */
5861 kernel_consumerd_state = CONSUMER_STOPPED;
5862 ust_consumerd_state = CONSUMER_STOPPED;
5863
5864 DBG("Client socket path %s", client_unix_sock_path);
5865 DBG("Application socket path %s", apps_unix_sock_path);
5866 DBG("Application wait path %s", wait_shm_path);
5867 DBG("LTTng run directory path: %s", rundir);
5868
5869 /* 32 bits consumerd path setup */
5870 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5871 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5872 if (ret < 0) {
5873 PERROR("snprintf 32-bit consumer error socket path");
5874 retval = -1;
5875 goto exit_init_data;
5876 }
5877 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5878 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5879 if (ret < 0) {
5880 PERROR("snprintf 32-bit consumer command socket path");
5881 retval = -1;
5882 goto exit_init_data;
5883 }
5884
5885 DBG2("UST consumer 32 bits err path: %s",
5886 ustconsumer32_data.err_unix_sock_path);
5887 DBG2("UST consumer 32 bits cmd path: %s",
5888 ustconsumer32_data.cmd_unix_sock_path);
5889 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5890 if (!ust32_channel_monitor_pipe) {
5891 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5892 retval = -1;
5893 goto exit_init_data;
5894 }
5895 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5896 ust32_channel_monitor_pipe);
5897 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5898 retval = -1;
5899 goto exit_init_data;
5900 }
5901
5902 /* 64 bits consumerd path setup */
5903 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5904 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5905 if (ret < 0) {
5906 PERROR("snprintf 64-bit consumer error socket path");
5907 retval = -1;
5908 goto exit_init_data;
5909 }
5910 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5911 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5912 if (ret < 0) {
5913 PERROR("snprintf 64-bit consumer command socket path");
5914 retval = -1;
5915 goto exit_init_data;
5916 }
5917
5918 DBG2("UST consumer 64 bits err path: %s",
5919 ustconsumer64_data.err_unix_sock_path);
5920 DBG2("UST consumer 64 bits cmd path: %s",
5921 ustconsumer64_data.cmd_unix_sock_path);
5922 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5923 if (!ust64_channel_monitor_pipe) {
5924 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5925 retval = -1;
5926 goto exit_init_data;
5927 }
5928 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5929 ust64_channel_monitor_pipe);
5930 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5931 retval = -1;
5932 goto exit_init_data;
5933 }
5934
5935 /*
5936 * See if daemon already exist.
5937 */
5938 if (check_existing_daemon()) {
5939 ERR("Already running daemon.\n");
5940 /*
5941 * We do not goto exit because we must not cleanup()
5942 * because a daemon is already running.
5943 */
5944 retval = -1;
5945 goto exit_init_data;
5946 }
5947
5948 /*
5949 * Init UST app hash table. Alloc hash table before this point since
5950 * cleanup() can get called after that point.
5951 */
5952 if (ust_app_ht_alloc()) {
5953 ERR("Failed to allocate UST app hash table");
5954 retval = -1;
5955 goto exit_init_data;
5956 }
5957
5958 /*
5959 * Initialize agent app hash table. We allocate the hash table here
5960 * since cleanup() can get called after this point.
5961 */
5962 if (agent_app_ht_alloc()) {
5963 ERR("Failed to allocate Agent app hash table");
5964 retval = -1;
5965 goto exit_init_data;
5966 }
5967
5968 /*
5969 * These actions must be executed as root. We do that *after* setting up
5970 * the sockets path because we MUST make the check for another daemon using
5971 * those paths *before* trying to set the kernel consumer sockets and init
5972 * kernel tracer.
5973 */
5974 if (is_root) {
5975 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5976 retval = -1;
5977 goto exit_init_data;
5978 }
5979
5980 /* Setup kernel tracer */
5981 if (!opt_no_kernel) {
5982 init_kernel_tracer();
5983 if (kernel_tracer_fd >= 0) {
5984 ret = syscall_init_table();
5985 if (ret < 0) {
5986 ERR("Unable to populate syscall table. "
5987 "Syscall tracing won't work "
5988 "for this session daemon.");
5989 }
5990 }
5991 }
5992
5993 /* Set ulimit for open files */
5994 set_ulimit();
5995 }
5996 /* init lttng_fd tracking must be done after set_ulimit. */
5997 lttng_fd_init();
5998
5999 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
6000 retval = -1;
6001 goto exit_init_data;
6002 }
6003
6004 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
6005 retval = -1;
6006 goto exit_init_data;
6007 }
6008
6009 /* Setup the needed unix socket */
6010 if (init_daemon_socket()) {
6011 retval = -1;
6012 goto exit_init_data;
6013 }
6014
6015 /* Set credentials to socket */
6016 if (is_root && set_permissions(rundir)) {
6017 retval = -1;
6018 goto exit_init_data;
6019 }
6020
6021 /* Get parent pid if -S, --sig-parent is specified. */
6022 if (opt_sig_parent) {
6023 ppid = getppid();
6024 }
6025
6026 /* Setup the kernel pipe for waking up the kernel thread */
6027 if (is_root && !opt_no_kernel) {
6028 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
6029 retval = -1;
6030 goto exit_init_data;
6031 }
6032 }
6033
6034 /* Setup the thread apps communication pipe. */
6035 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
6036 retval = -1;
6037 goto exit_init_data;
6038 }
6039
6040 /* Setup the thread apps notify communication pipe. */
6041 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
6042 retval = -1;
6043 goto exit_init_data;
6044 }
6045
6046 /* Initialize global buffer per UID and PID registry. */
6047 buffer_reg_init_uid_registry();
6048 buffer_reg_init_pid_registry();
6049
6050 /* Init UST command queue. */
6051 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
6052
6053 /*
6054 * Get session list pointer. This pointer MUST NOT be free'd. This list
6055 * is statically declared in session.c
6056 */
6057 session_list_ptr = session_get_list();
6058
6059 cmd_init();
6060
6061 /* Check for the application socket timeout env variable. */
6062 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
6063 if (env_app_timeout) {
6064 app_socket_timeout = atoi(env_app_timeout);
6065 } else {
6066 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
6067 }
6068
6069 ret = write_pidfile();
6070 if (ret) {
6071 ERR("Error in write_pidfile");
6072 retval = -1;
6073 goto exit_init_data;
6074 }
6075 ret = write_agent_port();
6076 if (ret) {
6077 ERR("Error in write_agent_port");
6078 retval = -1;
6079 goto exit_init_data;
6080 }
6081
6082 /* Initialize communication library */
6083 lttcomm_init();
6084 /* Initialize TCP timeout values */
6085 lttcomm_inet_init();
6086
6087 if (load_session_init_data(&load_info) < 0) {
6088 retval = -1;
6089 goto exit_init_data;
6090 }
6091 load_info->path = opt_load_session_path;
6092
6093 /* Create health-check thread. */
6094 ret = pthread_create(&health_thread, default_pthread_attr(),
6095 thread_manage_health, (void *) NULL);
6096 if (ret) {
6097 errno = ret;
6098 PERROR("pthread_create health");
6099 retval = -1;
6100 goto exit_health;
6101 }
6102
6103 /* notification_thread_data acquires the pipes' read side. */
6104 notification_thread_handle = notification_thread_handle_create(
6105 ust32_channel_monitor_pipe,
6106 ust64_channel_monitor_pipe,
6107 kernel_channel_monitor_pipe);
6108 if (!notification_thread_handle) {
6109 retval = -1;
6110 ERR("Failed to create notification thread shared data");
6111 stop_threads();
6112 goto exit_notification;
6113 }
6114
6115 /* Create notification thread. */
6116 ret = pthread_create(&notification_thread, default_pthread_attr(),
6117 thread_notification, notification_thread_handle);
6118 if (ret) {
6119 errno = ret;
6120 PERROR("pthread_create notification");
6121 retval = -1;
6122 stop_threads();
6123 goto exit_notification;
6124 }
6125
6126 /* Create thread to manage the client socket */
6127 ret = pthread_create(&client_thread, default_pthread_attr(),
6128 thread_manage_clients, (void *) NULL);
6129 if (ret) {
6130 errno = ret;
6131 PERROR("pthread_create clients");
6132 retval = -1;
6133 stop_threads();
6134 goto exit_client;
6135 }
6136
6137 /* Create thread to dispatch registration */
6138 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
6139 thread_dispatch_ust_registration, (void *) NULL);
6140 if (ret) {
6141 errno = ret;
6142 PERROR("pthread_create dispatch");
6143 retval = -1;
6144 stop_threads();
6145 goto exit_dispatch;
6146 }
6147
6148 /* Create thread to manage application registration. */
6149 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
6150 thread_registration_apps, (void *) NULL);
6151 if (ret) {
6152 errno = ret;
6153 PERROR("pthread_create registration");
6154 retval = -1;
6155 stop_threads();
6156 goto exit_reg_apps;
6157 }
6158
6159 /* Create thread to manage application socket */
6160 ret = pthread_create(&apps_thread, default_pthread_attr(),
6161 thread_manage_apps, (void *) NULL);
6162 if (ret) {
6163 errno = ret;
6164 PERROR("pthread_create apps");
6165 retval = -1;
6166 stop_threads();
6167 goto exit_apps;
6168 }
6169
6170 /* Create thread to manage application notify socket */
6171 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
6172 ust_thread_manage_notify, (void *) NULL);
6173 if (ret) {
6174 errno = ret;
6175 PERROR("pthread_create notify");
6176 retval = -1;
6177 stop_threads();
6178 goto exit_apps_notify;
6179 }
6180
6181 /* Create agent registration thread. */
6182 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
6183 agent_thread_manage_registration, (void *) NULL);
6184 if (ret) {
6185 errno = ret;
6186 PERROR("pthread_create agent");
6187 retval = -1;
6188 stop_threads();
6189 goto exit_agent_reg;
6190 }
6191
6192 /* Don't start this thread if kernel tracing is not requested nor root */
6193 if (is_root && !opt_no_kernel) {
6194 /* Create kernel thread to manage kernel event */
6195 ret = pthread_create(&kernel_thread, default_pthread_attr(),
6196 thread_manage_kernel, (void *) NULL);
6197 if (ret) {
6198 errno = ret;
6199 PERROR("pthread_create kernel");
6200 retval = -1;
6201 stop_threads();
6202 goto exit_kernel;
6203 }
6204 }
6205
6206 /* Create session loading thread. */
6207 ret = pthread_create(&load_session_thread, default_pthread_attr(),
6208 thread_load_session, load_info);
6209 if (ret) {
6210 errno = ret;
6211 PERROR("pthread_create load_session_thread");
6212 retval = -1;
6213 stop_threads();
6214 goto exit_load_session;
6215 }
6216
6217 /*
6218 * This is where we start awaiting program completion (e.g. through
6219 * signal that asks threads to teardown).
6220 */
6221
6222 ret = pthread_join(load_session_thread, &status);
6223 if (ret) {
6224 errno = ret;
6225 PERROR("pthread_join load_session_thread");
6226 retval = -1;
6227 }
6228 exit_load_session:
6229
6230 if (is_root && !opt_no_kernel) {
6231 ret = pthread_join(kernel_thread, &status);
6232 if (ret) {
6233 errno = ret;
6234 PERROR("pthread_join");
6235 retval = -1;
6236 }
6237 }
6238 exit_kernel:
6239
6240 ret = pthread_join(agent_reg_thread, &status);
6241 if (ret) {
6242 errno = ret;
6243 PERROR("pthread_join agent");
6244 retval = -1;
6245 }
6246 exit_agent_reg:
6247
6248 ret = pthread_join(apps_notify_thread, &status);
6249 if (ret) {
6250 errno = ret;
6251 PERROR("pthread_join apps notify");
6252 retval = -1;
6253 }
6254 exit_apps_notify:
6255
6256 ret = pthread_join(apps_thread, &status);
6257 if (ret) {
6258 errno = ret;
6259 PERROR("pthread_join apps");
6260 retval = -1;
6261 }
6262 exit_apps:
6263
6264 ret = pthread_join(reg_apps_thread, &status);
6265 if (ret) {
6266 errno = ret;
6267 PERROR("pthread_join");
6268 retval = -1;
6269 }
6270 exit_reg_apps:
6271
6272 /*
6273 * Join dispatch thread after joining reg_apps_thread to ensure
6274 * we don't leak applications in the queue.
6275 */
6276 ret = pthread_join(dispatch_thread, &status);
6277 if (ret) {
6278 errno = ret;
6279 PERROR("pthread_join");
6280 retval = -1;
6281 }
6282 exit_dispatch:
6283
6284 ret = pthread_join(client_thread, &status);
6285 if (ret) {
6286 errno = ret;
6287 PERROR("pthread_join");
6288 retval = -1;
6289 }
6290
6291 exit_client:
6292 exit_notification:
6293 ret = pthread_join(health_thread, &status);
6294 if (ret) {
6295 errno = ret;
6296 PERROR("pthread_join health thread");
6297 retval = -1;
6298 }
6299
6300 exit_health:
6301 exit_init_data:
6302 /*
6303 * Wait for all pending call_rcu work to complete before tearing
6304 * down data structures. call_rcu worker may be trying to
6305 * perform lookups in those structures.
6306 */
6307 rcu_barrier();
6308 /*
6309 * sessiond_cleanup() is called when no other thread is running, except
6310 * the ht_cleanup thread, which is needed to destroy the hash tables.
6311 */
6312 rcu_thread_online();
6313 sessiond_cleanup();
6314
6315 /*
6316 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6317 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6318 * the queue is empty before shutting down the clean-up thread.
6319 */
6320 rcu_barrier();
6321
6322 /*
6323 * The teardown of the notification system is performed after the
6324 * session daemon's teardown in order to allow it to be notified
6325 * of the active session and channels at the moment of the teardown.
6326 */
6327 if (notification_thread_handle) {
6328 notification_thread_command_quit(notification_thread_handle);
6329 notification_thread_handle_destroy(notification_thread_handle);
6330 }
6331
6332 ret = pthread_join(notification_thread, &status);
6333 if (ret) {
6334 errno = ret;
6335 PERROR("pthread_join notification thread");
6336 retval = -1;
6337 }
6338
6339 rcu_thread_offline();
6340 rcu_unregister_thread();
6341
6342 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6343 if (ret) {
6344 retval = -1;
6345 }
6346 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6347 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6348 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6349 exit_ht_cleanup:
6350
6351 health_app_destroy(health_sessiond);
6352 exit_health_sessiond_cleanup:
6353 exit_create_run_as_worker_cleanup:
6354
6355 exit_options:
6356 sessiond_cleanup_options();
6357
6358 exit_set_signal_handler:
6359 if (!retval) {
6360 exit(EXIT_SUCCESS);
6361 } else {
6362 exit(EXIT_FAILURE);
6363 }
6364 }
This page took 0.151652 seconds and 4 git commands to generate.