Add --enable-embedded-help option to embed --help messages in binaries
[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 static const char *help_msg =
83 #ifdef LTTNG_EMBED_HELP
84 #include <lttng-sessiond.8.h>
85 #else
86 NULL
87 #endif
88 ;
89
90 const char *progname;
91 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
92 static int tracing_group_name_override;
93 static char *opt_pidfile;
94 static int opt_sig_parent;
95 static int opt_verbose_consumer;
96 static int opt_daemon, opt_background;
97 static int opt_no_kernel;
98 static char *opt_load_session_path;
99 static pid_t ppid; /* Parent PID for --sig-parent option */
100 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
101 static char *rundir;
102 static int lockfile_fd = -1;
103
104 /* Set to 1 when a SIGUSR1 signal is received. */
105 static int recv_child_signal;
106
107 /*
108 * Consumer daemon specific control data. Every value not initialized here is
109 * set to 0 by the static definition.
110 */
111 static struct consumer_data kconsumer_data = {
112 .type = LTTNG_CONSUMER_KERNEL,
113 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .channel_monitor_pipe = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123 static struct consumer_data ustconsumer64_data = {
124 .type = LTTNG_CONSUMER64_UST,
125 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
126 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
127 .err_sock = -1,
128 .cmd_sock = -1,
129 .channel_monitor_pipe = -1,
130 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
131 .lock = PTHREAD_MUTEX_INITIALIZER,
132 .cond = PTHREAD_COND_INITIALIZER,
133 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
134 };
135 static struct consumer_data ustconsumer32_data = {
136 .type = LTTNG_CONSUMER32_UST,
137 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
138 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
139 .err_sock = -1,
140 .cmd_sock = -1,
141 .channel_monitor_pipe = -1,
142 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
143 .lock = PTHREAD_MUTEX_INITIALIZER,
144 .cond = PTHREAD_COND_INITIALIZER,
145 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
146 };
147
148 /* Command line options */
149 static const struct option long_options[] = {
150 { "client-sock", required_argument, 0, 'c' },
151 { "apps-sock", required_argument, 0, 'a' },
152 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
153 { "kconsumerd-err-sock", required_argument, 0, '\0' },
154 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
155 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
156 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
157 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
158 { "consumerd32-path", required_argument, 0, '\0' },
159 { "consumerd32-libdir", required_argument, 0, '\0' },
160 { "consumerd64-path", required_argument, 0, '\0' },
161 { "consumerd64-libdir", required_argument, 0, '\0' },
162 { "daemonize", no_argument, 0, 'd' },
163 { "background", no_argument, 0, 'b' },
164 { "sig-parent", no_argument, 0, 'S' },
165 { "help", no_argument, 0, 'h' },
166 { "group", required_argument, 0, 'g' },
167 { "version", no_argument, 0, 'V' },
168 { "quiet", no_argument, 0, 'q' },
169 { "verbose", no_argument, 0, 'v' },
170 { "verbose-consumer", no_argument, 0, '\0' },
171 { "no-kernel", no_argument, 0, '\0' },
172 { "pidfile", required_argument, 0, 'p' },
173 { "agent-tcp-port", required_argument, 0, '\0' },
174 { "config", required_argument, 0, 'f' },
175 { "load", required_argument, 0, 'l' },
176 { "kmod-probes", required_argument, 0, '\0' },
177 { "extra-kmod-probes", required_argument, 0, '\0' },
178 { NULL, 0, 0, 0 }
179 };
180
181 /* Command line options to ignore from configuration file */
182 static const char *config_ignore_options[] = { "help", "version", "config" };
183
184 /* Shared between threads */
185 static int dispatch_thread_exit;
186
187 /* Global application Unix socket path */
188 static char apps_unix_sock_path[PATH_MAX];
189 /* Global client Unix socket path */
190 static char client_unix_sock_path[PATH_MAX];
191 /* global wait shm path for UST */
192 static char wait_shm_path[PATH_MAX];
193 /* Global health check unix path */
194 static char health_unix_sock_path[PATH_MAX];
195
196 /* Sockets and FDs */
197 static int client_sock = -1;
198 static int apps_sock = -1;
199 int kernel_tracer_fd = -1;
200 static int kernel_poll_pipe[2] = { -1, -1 };
201
202 /*
203 * Quit pipe for all threads. This permits a single cancellation point
204 * for all threads when receiving an event on the pipe.
205 */
206 static int thread_quit_pipe[2] = { -1, -1 };
207
208 /*
209 * This pipe is used to inform the thread managing application communication
210 * that a command is queued and ready to be processed.
211 */
212 static int apps_cmd_pipe[2] = { -1, -1 };
213
214 int apps_cmd_notify_pipe[2] = { -1, -1 };
215
216 /* Pthread, Mutexes and Semaphores */
217 static pthread_t apps_thread;
218 static pthread_t apps_notify_thread;
219 static pthread_t reg_apps_thread;
220 static pthread_t client_thread;
221 static pthread_t kernel_thread;
222 static pthread_t dispatch_thread;
223 static pthread_t health_thread;
224 static pthread_t ht_cleanup_thread;
225 static pthread_t agent_reg_thread;
226 static pthread_t load_session_thread;
227 static pthread_t notification_thread;
228
229 /*
230 * UST registration command queue. This queue is tied with a futex and uses a N
231 * wakers / 1 waiter implemented and detailed in futex.c/.h
232 *
233 * The thread_registration_apps and thread_dispatch_ust_registration uses this
234 * queue along with the wait/wake scheme. The thread_manage_apps receives down
235 * the line new application socket and monitors it for any I/O error or clean
236 * close that triggers an unregistration of the application.
237 */
238 static struct ust_cmd_queue ust_cmd_queue;
239
240 /*
241 * Pointer initialized before thread creation.
242 *
243 * This points to the tracing session list containing the session count and a
244 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
245 * MUST NOT be taken if you call a public function in session.c.
246 *
247 * The lock is nested inside the structure: session_list_ptr->lock. Please use
248 * session_lock_list and session_unlock_list for lock acquisition.
249 */
250 static struct ltt_session_list *session_list_ptr;
251
252 int ust_consumerd64_fd = -1;
253 int ust_consumerd32_fd = -1;
254
255 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
256 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
257 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
258 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
259 static int consumerd32_bin_override;
260 static int consumerd64_bin_override;
261 static int consumerd32_libdir_override;
262 static int consumerd64_libdir_override;
263
264 static const char *module_proc_lttng = "/proc/lttng";
265
266 /*
267 * Consumer daemon state which is changed when spawning it, killing it or in
268 * case of a fatal error.
269 */
270 enum consumerd_state {
271 CONSUMER_STARTED = 1,
272 CONSUMER_STOPPED = 2,
273 CONSUMER_ERROR = 3,
274 };
275
276 /*
277 * This consumer daemon state is used to validate if a client command will be
278 * able to reach the consumer. If not, the client is informed. For instance,
279 * doing a "lttng start" when the consumer state is set to ERROR will return an
280 * error to the client.
281 *
282 * The following example shows a possible race condition of this scheme:
283 *
284 * consumer thread error happens
285 * client cmd arrives
286 * client cmd checks state -> still OK
287 * consumer thread exit, sets error
288 * client cmd try to talk to consumer
289 * ...
290 *
291 * However, since the consumer is a different daemon, we have no way of making
292 * sure the command will reach it safely even with this state flag. This is why
293 * we consider that up to the state validation during command processing, the
294 * command is safe. After that, we can not guarantee the correctness of the
295 * client request vis-a-vis the consumer.
296 */
297 static enum consumerd_state ust_consumerd_state;
298 static enum consumerd_state kernel_consumerd_state;
299
300 /*
301 * Socket timeout for receiving and sending in seconds.
302 */
303 static int app_socket_timeout;
304
305 /* Set in main() with the current page size. */
306 long page_size;
307
308 /* Application health monitoring */
309 struct health_app *health_sessiond;
310
311 /* Agent TCP port for registration. Used by the agent thread. */
312 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
313
314 /* Am I root or not. */
315 int is_root; /* Set to 1 if the daemon is running as root */
316
317 const char * const config_section_name = "sessiond";
318
319 /* Load session thread information to operate. */
320 struct load_session_thread_data *load_info;
321
322 /* Notification thread handle. */
323 struct notification_thread_handle *notification_thread_handle;
324
325 /* Global hash tables */
326 struct lttng_ht *agent_apps_ht_by_sock = NULL;
327
328 /*
329 * Whether sessiond is ready for commands/notification channel/health check
330 * requests.
331 * NR_LTTNG_SESSIOND_READY must match the number of calls to
332 * sessiond_notify_ready().
333 */
334 #define NR_LTTNG_SESSIOND_READY 4
335 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
336
337 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
338 {
339 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
340 }
341
342 /* Notify parents that we are ready for cmd and health check */
343 LTTNG_HIDDEN
344 void sessiond_notify_ready(void)
345 {
346 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
347 /*
348 * Notify parent pid that we are ready to accept command
349 * for client side. This ppid is the one from the
350 * external process that spawned us.
351 */
352 if (opt_sig_parent) {
353 kill(ppid, SIGUSR1);
354 }
355
356 /*
357 * Notify the parent of the fork() process that we are
358 * ready.
359 */
360 if (opt_daemon || opt_background) {
361 kill(child_ppid, SIGUSR1);
362 }
363 }
364 }
365
366 static
367 void setup_consumerd_path(void)
368 {
369 const char *bin, *libdir;
370
371 /*
372 * Allow INSTALL_BIN_PATH to be used as a target path for the
373 * native architecture size consumer if CONFIG_CONSUMER*_PATH
374 * has not been defined.
375 */
376 #if (CAA_BITS_PER_LONG == 32)
377 if (!consumerd32_bin[0]) {
378 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
379 }
380 if (!consumerd32_libdir[0]) {
381 consumerd32_libdir = INSTALL_LIB_PATH;
382 }
383 #elif (CAA_BITS_PER_LONG == 64)
384 if (!consumerd64_bin[0]) {
385 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
386 }
387 if (!consumerd64_libdir[0]) {
388 consumerd64_libdir = INSTALL_LIB_PATH;
389 }
390 #else
391 #error "Unknown bitness"
392 #endif
393
394 /*
395 * runtime env. var. overrides the build default.
396 */
397 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
398 if (bin) {
399 consumerd32_bin = bin;
400 }
401 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
402 if (bin) {
403 consumerd64_bin = bin;
404 }
405 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
406 if (libdir) {
407 consumerd32_libdir = libdir;
408 }
409 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
410 if (libdir) {
411 consumerd64_libdir = libdir;
412 }
413 }
414
415 static
416 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
417 int *a_pipe)
418 {
419 int ret;
420
421 assert(events);
422
423 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
424 if (ret < 0) {
425 goto error;
426 }
427
428 /* Add quit pipe */
429 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
430 if (ret < 0) {
431 goto error;
432 }
433
434 return 0;
435
436 error:
437 return ret;
438 }
439
440 /*
441 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
442 */
443 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
444 {
445 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
446 }
447
448 /*
449 * Init thread quit pipe.
450 *
451 * Return -1 on error or 0 if all pipes are created.
452 */
453 static int __init_thread_quit_pipe(int *a_pipe)
454 {
455 int ret, i;
456
457 ret = pipe(a_pipe);
458 if (ret < 0) {
459 PERROR("thread quit pipe");
460 goto error;
461 }
462
463 for (i = 0; i < 2; i++) {
464 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
465 if (ret < 0) {
466 PERROR("fcntl");
467 goto error;
468 }
469 }
470
471 error:
472 return ret;
473 }
474
475 static int init_thread_quit_pipe(void)
476 {
477 return __init_thread_quit_pipe(thread_quit_pipe);
478 }
479
480 /*
481 * Stop all threads by closing the thread quit pipe.
482 */
483 static void stop_threads(void)
484 {
485 int ret;
486
487 /* Stopping all threads */
488 DBG("Terminating all threads");
489 ret = notify_thread_pipe(thread_quit_pipe[1]);
490 if (ret < 0) {
491 ERR("write error on thread quit pipe");
492 }
493
494 /* Dispatch thread */
495 CMM_STORE_SHARED(dispatch_thread_exit, 1);
496 futex_nto1_wake(&ust_cmd_queue.futex);
497 }
498
499 /*
500 * Close every consumer sockets.
501 */
502 static void close_consumer_sockets(void)
503 {
504 int ret;
505
506 if (kconsumer_data.err_sock >= 0) {
507 ret = close(kconsumer_data.err_sock);
508 if (ret < 0) {
509 PERROR("kernel consumer err_sock close");
510 }
511 }
512 if (ustconsumer32_data.err_sock >= 0) {
513 ret = close(ustconsumer32_data.err_sock);
514 if (ret < 0) {
515 PERROR("UST consumerd32 err_sock close");
516 }
517 }
518 if (ustconsumer64_data.err_sock >= 0) {
519 ret = close(ustconsumer64_data.err_sock);
520 if (ret < 0) {
521 PERROR("UST consumerd64 err_sock close");
522 }
523 }
524 if (kconsumer_data.cmd_sock >= 0) {
525 ret = close(kconsumer_data.cmd_sock);
526 if (ret < 0) {
527 PERROR("kernel consumer cmd_sock close");
528 }
529 }
530 if (ustconsumer32_data.cmd_sock >= 0) {
531 ret = close(ustconsumer32_data.cmd_sock);
532 if (ret < 0) {
533 PERROR("UST consumerd32 cmd_sock close");
534 }
535 }
536 if (ustconsumer64_data.cmd_sock >= 0) {
537 ret = close(ustconsumer64_data.cmd_sock);
538 if (ret < 0) {
539 PERROR("UST consumerd64 cmd_sock close");
540 }
541 }
542 if (kconsumer_data.channel_monitor_pipe >= 0) {
543 ret = close(kconsumer_data.channel_monitor_pipe);
544 if (ret < 0) {
545 PERROR("kernel consumer channel monitor pipe close");
546 }
547 }
548 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
549 ret = close(ustconsumer32_data.channel_monitor_pipe);
550 if (ret < 0) {
551 PERROR("UST consumerd32 channel monitor pipe close");
552 }
553 }
554 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
555 ret = close(ustconsumer64_data.channel_monitor_pipe);
556 if (ret < 0) {
557 PERROR("UST consumerd64 channel monitor pipe close");
558 }
559 }
560 }
561
562 /*
563 * Generate the full lock file path using the rundir.
564 *
565 * Return the snprintf() return value thus a negative value is an error.
566 */
567 static int generate_lock_file_path(char *path, size_t len)
568 {
569 int ret;
570
571 assert(path);
572 assert(rundir);
573
574 /* Build lockfile path from rundir. */
575 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
576 if (ret < 0) {
577 PERROR("snprintf lockfile path");
578 }
579
580 return ret;
581 }
582
583 /*
584 * Wait on consumer process termination.
585 *
586 * Need to be called with the consumer data lock held or from a context
587 * ensuring no concurrent access to data (e.g: cleanup).
588 */
589 static void wait_consumer(struct consumer_data *consumer_data)
590 {
591 pid_t ret;
592 int status;
593
594 if (consumer_data->pid <= 0) {
595 return;
596 }
597
598 DBG("Waiting for complete teardown of consumerd (PID: %d)",
599 consumer_data->pid);
600 ret = waitpid(consumer_data->pid, &status, 0);
601 if (ret == -1) {
602 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
603 }
604 if (!WIFEXITED(status)) {
605 ERR("consumerd termination with error: %d",
606 WEXITSTATUS(ret));
607 }
608 consumer_data->pid = 0;
609 }
610
611 /*
612 * Cleanup the session daemon's data structures.
613 */
614 static void sessiond_cleanup(void)
615 {
616 int ret;
617 struct ltt_session *sess, *stmp;
618 char path[PATH_MAX];
619
620 DBG("Cleanup sessiond");
621
622 /*
623 * Close the thread quit pipe. It has already done its job,
624 * since we are now called.
625 */
626 utils_close_pipe(thread_quit_pipe);
627
628 /*
629 * If opt_pidfile is undefined, the default file will be wiped when
630 * removing the rundir.
631 */
632 if (opt_pidfile) {
633 ret = remove(opt_pidfile);
634 if (ret < 0) {
635 PERROR("remove pidfile %s", opt_pidfile);
636 }
637 }
638
639 DBG("Removing sessiond and consumerd content of directory %s", rundir);
640
641 /* sessiond */
642 snprintf(path, PATH_MAX,
643 "%s/%s",
644 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
645 DBG("Removing %s", path);
646 (void) unlink(path);
647
648 snprintf(path, PATH_MAX, "%s/%s", rundir,
649 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
650 DBG("Removing %s", path);
651 (void) unlink(path);
652
653 /* kconsumerd */
654 snprintf(path, PATH_MAX,
655 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
656 rundir);
657 DBG("Removing %s", path);
658 (void) unlink(path);
659
660 snprintf(path, PATH_MAX,
661 DEFAULT_KCONSUMERD_PATH,
662 rundir);
663 DBG("Removing directory %s", path);
664 (void) rmdir(path);
665
666 /* ust consumerd 32 */
667 snprintf(path, PATH_MAX,
668 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
669 rundir);
670 DBG("Removing %s", path);
671 (void) unlink(path);
672
673 snprintf(path, PATH_MAX,
674 DEFAULT_USTCONSUMERD32_PATH,
675 rundir);
676 DBG("Removing directory %s", path);
677 (void) rmdir(path);
678
679 /* ust consumerd 64 */
680 snprintf(path, PATH_MAX,
681 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
682 rundir);
683 DBG("Removing %s", path);
684 (void) unlink(path);
685
686 snprintf(path, PATH_MAX,
687 DEFAULT_USTCONSUMERD64_PATH,
688 rundir);
689 DBG("Removing directory %s", path);
690 (void) rmdir(path);
691
692 DBG("Cleaning up all sessions");
693
694 /* Destroy session list mutex */
695 if (session_list_ptr != NULL) {
696 pthread_mutex_destroy(&session_list_ptr->lock);
697
698 /* Cleanup ALL session */
699 cds_list_for_each_entry_safe(sess, stmp,
700 &session_list_ptr->head, list) {
701 cmd_destroy_session(sess, kernel_poll_pipe[1]);
702 }
703 }
704
705 wait_consumer(&kconsumer_data);
706 wait_consumer(&ustconsumer64_data);
707 wait_consumer(&ustconsumer32_data);
708
709 DBG("Cleaning up all agent apps");
710 agent_app_ht_clean();
711
712 DBG("Closing all UST sockets");
713 ust_app_clean_list();
714 buffer_reg_destroy_registries();
715
716 if (is_root && !opt_no_kernel) {
717 DBG2("Closing kernel fd");
718 if (kernel_tracer_fd >= 0) {
719 ret = close(kernel_tracer_fd);
720 if (ret) {
721 PERROR("close");
722 }
723 }
724 DBG("Unloading kernel modules");
725 modprobe_remove_lttng_all();
726 free(syscall_table);
727 }
728
729 close_consumer_sockets();
730
731 if (load_info) {
732 load_session_destroy_data(load_info);
733 free(load_info);
734 }
735
736 /*
737 * Cleanup lock file by deleting it and finaly closing it which will
738 * release the file system lock.
739 */
740 if (lockfile_fd >= 0) {
741 char lockfile_path[PATH_MAX];
742
743 ret = generate_lock_file_path(lockfile_path,
744 sizeof(lockfile_path));
745 if (ret > 0) {
746 ret = remove(lockfile_path);
747 if (ret < 0) {
748 PERROR("remove lock file");
749 }
750 ret = close(lockfile_fd);
751 if (ret < 0) {
752 PERROR("close lock file");
753 }
754 }
755 }
756
757 /*
758 * We do NOT rmdir rundir because there are other processes
759 * using it, for instance lttng-relayd, which can start in
760 * parallel with this teardown.
761 */
762
763 free(rundir);
764 }
765
766 /*
767 * Cleanup the daemon's option data structures.
768 */
769 static void sessiond_cleanup_options(void)
770 {
771 DBG("Cleaning up options");
772
773 /*
774 * If the override option is set, the pointer points to a *non* const
775 * thus freeing it even though the variable type is set to const.
776 */
777 if (tracing_group_name_override) {
778 free((void *) tracing_group_name);
779 }
780 if (consumerd32_bin_override) {
781 free((void *) consumerd32_bin);
782 }
783 if (consumerd64_bin_override) {
784 free((void *) consumerd64_bin);
785 }
786 if (consumerd32_libdir_override) {
787 free((void *) consumerd32_libdir);
788 }
789 if (consumerd64_libdir_override) {
790 free((void *) consumerd64_libdir);
791 }
792
793 free(opt_pidfile);
794 free(opt_load_session_path);
795 free(kmod_probes_list);
796 free(kmod_extra_probes_list);
797
798 run_as_destroy_worker();
799 }
800
801 /*
802 * Send data on a unix socket using the liblttsessiondcomm API.
803 *
804 * Return lttcomm error code.
805 */
806 static int send_unix_sock(int sock, void *buf, size_t len)
807 {
808 /* Check valid length */
809 if (len == 0) {
810 return -1;
811 }
812
813 return lttcomm_send_unix_sock(sock, buf, len);
814 }
815
816 /*
817 * Free memory of a command context structure.
818 */
819 static void clean_command_ctx(struct command_ctx **cmd_ctx)
820 {
821 DBG("Clean command context structure");
822 if (*cmd_ctx) {
823 if ((*cmd_ctx)->llm) {
824 free((*cmd_ctx)->llm);
825 }
826 if ((*cmd_ctx)->lsm) {
827 free((*cmd_ctx)->lsm);
828 }
829 free(*cmd_ctx);
830 *cmd_ctx = NULL;
831 }
832 }
833
834 /*
835 * Notify UST applications using the shm mmap futex.
836 */
837 static int notify_ust_apps(int active)
838 {
839 char *wait_shm_mmap;
840
841 DBG("Notifying applications of session daemon state: %d", active);
842
843 /* See shm.c for this call implying mmap, shm and futex calls */
844 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
845 if (wait_shm_mmap == NULL) {
846 goto error;
847 }
848
849 /* Wake waiting process */
850 futex_wait_update((int32_t *) wait_shm_mmap, active);
851
852 /* Apps notified successfully */
853 return 0;
854
855 error:
856 return -1;
857 }
858
859 /*
860 * Setup the outgoing data buffer for the response (llm) by allocating the
861 * right amount of memory and copying the original information from the lsm
862 * structure.
863 *
864 * Return 0 on success, negative value on error.
865 */
866 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
867 const void *payload_buf, size_t payload_len,
868 const void *cmd_header_buf, size_t cmd_header_len)
869 {
870 int ret = 0;
871 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
872 const size_t cmd_header_offset = header_len;
873 const size_t payload_offset = cmd_header_offset + cmd_header_len;
874 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
875
876 cmd_ctx->llm = zmalloc(total_msg_size);
877
878 if (cmd_ctx->llm == NULL) {
879 PERROR("zmalloc");
880 ret = -ENOMEM;
881 goto end;
882 }
883
884 /* Copy common data */
885 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
886 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
887 cmd_ctx->llm->cmd_header_size = cmd_header_len;
888 cmd_ctx->llm->data_size = payload_len;
889 cmd_ctx->lttng_msg_size = total_msg_size;
890
891 /* Copy command header */
892 if (cmd_header_len) {
893 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
894 cmd_header_len);
895 }
896
897 /* Copy payload */
898 if (payload_len) {
899 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
900 payload_len);
901 }
902
903 end:
904 return ret;
905 }
906
907 /*
908 * Version of setup_lttng_msg() without command header.
909 */
910 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
911 void *payload_buf, size_t payload_len)
912 {
913 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
914 }
915 /*
916 * Update the kernel poll set of all channel fd available over all tracing
917 * session. Add the wakeup pipe at the end of the set.
918 */
919 static int update_kernel_poll(struct lttng_poll_event *events)
920 {
921 int ret;
922 struct ltt_session *session;
923 struct ltt_kernel_channel *channel;
924
925 DBG("Updating kernel poll set");
926
927 session_lock_list();
928 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
929 session_lock(session);
930 if (session->kernel_session == NULL) {
931 session_unlock(session);
932 continue;
933 }
934
935 cds_list_for_each_entry(channel,
936 &session->kernel_session->channel_list.head, list) {
937 /* Add channel fd to the kernel poll set */
938 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
939 if (ret < 0) {
940 session_unlock(session);
941 goto error;
942 }
943 DBG("Channel fd %d added to kernel set", channel->fd);
944 }
945 session_unlock(session);
946 }
947 session_unlock_list();
948
949 return 0;
950
951 error:
952 session_unlock_list();
953 return -1;
954 }
955
956 /*
957 * Find the channel fd from 'fd' over all tracing session. When found, check
958 * for new channel stream and send those stream fds to the kernel consumer.
959 *
960 * Useful for CPU hotplug feature.
961 */
962 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
963 {
964 int ret = 0;
965 struct ltt_session *session;
966 struct ltt_kernel_session *ksess;
967 struct ltt_kernel_channel *channel;
968
969 DBG("Updating kernel streams for channel fd %d", fd);
970
971 session_lock_list();
972 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
973 session_lock(session);
974 if (session->kernel_session == NULL) {
975 session_unlock(session);
976 continue;
977 }
978 ksess = session->kernel_session;
979
980 cds_list_for_each_entry(channel,
981 &ksess->channel_list.head, list) {
982 struct lttng_ht_iter iter;
983 struct consumer_socket *socket;
984
985 if (channel->fd != fd) {
986 continue;
987 }
988 DBG("Channel found, updating kernel streams");
989 ret = kernel_open_channel_stream(channel);
990 if (ret < 0) {
991 goto error;
992 }
993 /* Update the stream global counter */
994 ksess->stream_count_global += ret;
995
996 /*
997 * Have we already sent fds to the consumer? If yes, it
998 * means that tracing is started so it is safe to send
999 * our updated stream fds.
1000 */
1001 if (ksess->consumer_fds_sent != 1
1002 || ksess->consumer == NULL) {
1003 ret = -1;
1004 goto error;
1005 }
1006
1007 rcu_read_lock();
1008 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
1009 &iter.iter, socket, node.node) {
1010 pthread_mutex_lock(socket->lock);
1011 ret = kernel_consumer_send_channel_stream(socket,
1012 channel, ksess,
1013 session->output_traces ? 1 : 0);
1014 pthread_mutex_unlock(socket->lock);
1015 if (ret < 0) {
1016 rcu_read_unlock();
1017 goto error;
1018 }
1019 }
1020 rcu_read_unlock();
1021 }
1022 session_unlock(session);
1023 }
1024 session_unlock_list();
1025 return ret;
1026
1027 error:
1028 session_unlock(session);
1029 session_unlock_list();
1030 return ret;
1031 }
1032
1033 /*
1034 * For each tracing session, update newly registered apps. The session list
1035 * lock MUST be acquired before calling this.
1036 */
1037 static void update_ust_app(int app_sock)
1038 {
1039 struct ltt_session *sess, *stmp;
1040
1041 /* Consumer is in an ERROR state. Stop any application update. */
1042 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1043 /* Stop the update process since the consumer is dead. */
1044 return;
1045 }
1046
1047 /* For all tracing session(s) */
1048 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1049 struct ust_app *app;
1050
1051 session_lock(sess);
1052 if (!sess->ust_session) {
1053 goto unlock_session;
1054 }
1055
1056 rcu_read_lock();
1057 assert(app_sock >= 0);
1058 app = ust_app_find_by_sock(app_sock);
1059 if (app == NULL) {
1060 /*
1061 * Application can be unregistered before so
1062 * this is possible hence simply stopping the
1063 * update.
1064 */
1065 DBG3("UST app update failed to find app sock %d",
1066 app_sock);
1067 goto unlock_rcu;
1068 }
1069 ust_app_global_update(sess->ust_session, app);
1070 unlock_rcu:
1071 rcu_read_unlock();
1072 unlock_session:
1073 session_unlock(sess);
1074 }
1075 }
1076
1077 /*
1078 * This thread manage event coming from the kernel.
1079 *
1080 * Features supported in this thread:
1081 * -) CPU Hotplug
1082 */
1083 static void *thread_manage_kernel(void *data)
1084 {
1085 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1086 uint32_t revents, nb_fd;
1087 char tmp;
1088 struct lttng_poll_event events;
1089
1090 DBG("[thread] Thread manage kernel started");
1091
1092 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1093
1094 /*
1095 * This first step of the while is to clean this structure which could free
1096 * non NULL pointers so initialize it before the loop.
1097 */
1098 lttng_poll_init(&events);
1099
1100 if (testpoint(sessiond_thread_manage_kernel)) {
1101 goto error_testpoint;
1102 }
1103
1104 health_code_update();
1105
1106 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1107 goto error_testpoint;
1108 }
1109
1110 while (1) {
1111 health_code_update();
1112
1113 if (update_poll_flag == 1) {
1114 /* Clean events object. We are about to populate it again. */
1115 lttng_poll_clean(&events);
1116
1117 ret = sessiond_set_thread_pollset(&events, 2);
1118 if (ret < 0) {
1119 goto error_poll_create;
1120 }
1121
1122 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1123 if (ret < 0) {
1124 goto error;
1125 }
1126
1127 /* This will add the available kernel channel if any. */
1128 ret = update_kernel_poll(&events);
1129 if (ret < 0) {
1130 goto error;
1131 }
1132 update_poll_flag = 0;
1133 }
1134
1135 DBG("Thread kernel polling");
1136
1137 /* Poll infinite value of time */
1138 restart:
1139 health_poll_entry();
1140 ret = lttng_poll_wait(&events, -1);
1141 DBG("Thread kernel return from poll on %d fds",
1142 LTTNG_POLL_GETNB(&events));
1143 health_poll_exit();
1144 if (ret < 0) {
1145 /*
1146 * Restart interrupted system call.
1147 */
1148 if (errno == EINTR) {
1149 goto restart;
1150 }
1151 goto error;
1152 } else if (ret == 0) {
1153 /* Should not happen since timeout is infinite */
1154 ERR("Return value of poll is 0 with an infinite timeout.\n"
1155 "This should not have happened! Continuing...");
1156 continue;
1157 }
1158
1159 nb_fd = ret;
1160
1161 for (i = 0; i < nb_fd; i++) {
1162 /* Fetch once the poll data */
1163 revents = LTTNG_POLL_GETEV(&events, i);
1164 pollfd = LTTNG_POLL_GETFD(&events, i);
1165
1166 health_code_update();
1167
1168 if (!revents) {
1169 /* No activity for this FD (poll implementation). */
1170 continue;
1171 }
1172
1173 /* Thread quit pipe has been closed. Killing thread. */
1174 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1175 if (ret) {
1176 err = 0;
1177 goto exit;
1178 }
1179
1180 /* Check for data on kernel pipe */
1181 if (revents & LPOLLIN) {
1182 if (pollfd == kernel_poll_pipe[0]) {
1183 (void) lttng_read(kernel_poll_pipe[0],
1184 &tmp, 1);
1185 /*
1186 * Ret value is useless here, if this pipe gets any actions an
1187 * update is required anyway.
1188 */
1189 update_poll_flag = 1;
1190 continue;
1191 } else {
1192 /*
1193 * New CPU detected by the kernel. Adding kernel stream to
1194 * kernel session and updating the kernel consumer
1195 */
1196 ret = update_kernel_stream(&kconsumer_data, pollfd);
1197 if (ret < 0) {
1198 continue;
1199 }
1200 break;
1201 }
1202 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1203 update_poll_flag = 1;
1204 continue;
1205 } else {
1206 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1207 goto error;
1208 }
1209 }
1210 }
1211
1212 exit:
1213 error:
1214 lttng_poll_clean(&events);
1215 error_poll_create:
1216 error_testpoint:
1217 utils_close_pipe(kernel_poll_pipe);
1218 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1219 if (err) {
1220 health_error();
1221 ERR("Health error occurred in %s", __func__);
1222 WARN("Kernel thread died unexpectedly. "
1223 "Kernel tracing can continue but CPU hotplug is disabled.");
1224 }
1225 health_unregister(health_sessiond);
1226 DBG("Kernel thread dying");
1227 return NULL;
1228 }
1229
1230 /*
1231 * Signal pthread condition of the consumer data that the thread.
1232 */
1233 static void signal_consumer_condition(struct consumer_data *data, int state)
1234 {
1235 pthread_mutex_lock(&data->cond_mutex);
1236
1237 /*
1238 * The state is set before signaling. It can be any value, it's the waiter
1239 * job to correctly interpret this condition variable associated to the
1240 * consumer pthread_cond.
1241 *
1242 * A value of 0 means that the corresponding thread of the consumer data
1243 * was not started. 1 indicates that the thread has started and is ready
1244 * for action. A negative value means that there was an error during the
1245 * thread bootstrap.
1246 */
1247 data->consumer_thread_is_ready = state;
1248 (void) pthread_cond_signal(&data->cond);
1249
1250 pthread_mutex_unlock(&data->cond_mutex);
1251 }
1252
1253 /*
1254 * This thread manage the consumer error sent back to the session daemon.
1255 */
1256 static void *thread_manage_consumer(void *data)
1257 {
1258 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1259 uint32_t revents, nb_fd;
1260 enum lttcomm_return_code code;
1261 struct lttng_poll_event events;
1262 struct consumer_data *consumer_data = data;
1263 struct consumer_socket *cmd_socket_wrapper = NULL;
1264
1265 DBG("[thread] Manage consumer started");
1266
1267 rcu_register_thread();
1268 rcu_thread_online();
1269
1270 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1271
1272 health_code_update();
1273
1274 /*
1275 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1276 * metadata_sock. Nothing more will be added to this poll set.
1277 */
1278 ret = sessiond_set_thread_pollset(&events, 3);
1279 if (ret < 0) {
1280 goto error_poll;
1281 }
1282
1283 /*
1284 * The error socket here is already in a listening state which was done
1285 * just before spawning this thread to avoid a race between the consumer
1286 * daemon exec trying to connect and the listen() call.
1287 */
1288 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1289 if (ret < 0) {
1290 goto error;
1291 }
1292
1293 health_code_update();
1294
1295 /* Infinite blocking call, waiting for transmission */
1296 restart:
1297 health_poll_entry();
1298
1299 if (testpoint(sessiond_thread_manage_consumer)) {
1300 goto error;
1301 }
1302
1303 ret = lttng_poll_wait(&events, -1);
1304 health_poll_exit();
1305 if (ret < 0) {
1306 /*
1307 * Restart interrupted system call.
1308 */
1309 if (errno == EINTR) {
1310 goto restart;
1311 }
1312 goto error;
1313 }
1314
1315 nb_fd = ret;
1316
1317 for (i = 0; i < nb_fd; i++) {
1318 /* Fetch once the poll data */
1319 revents = LTTNG_POLL_GETEV(&events, i);
1320 pollfd = LTTNG_POLL_GETFD(&events, i);
1321
1322 health_code_update();
1323
1324 if (!revents) {
1325 /* No activity for this FD (poll implementation). */
1326 continue;
1327 }
1328
1329 /* Thread quit pipe has been closed. Killing thread. */
1330 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1331 if (ret) {
1332 err = 0;
1333 goto exit;
1334 }
1335
1336 /* Event on the registration socket */
1337 if (pollfd == consumer_data->err_sock) {
1338 if (revents & LPOLLIN) {
1339 continue;
1340 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1341 ERR("consumer err socket poll error");
1342 goto error;
1343 } else {
1344 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1345 goto error;
1346 }
1347 }
1348 }
1349
1350 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1351 if (sock < 0) {
1352 goto error;
1353 }
1354
1355 /*
1356 * Set the CLOEXEC flag. Return code is useless because either way, the
1357 * show must go on.
1358 */
1359 (void) utils_set_fd_cloexec(sock);
1360
1361 health_code_update();
1362
1363 DBG2("Receiving code from consumer err_sock");
1364
1365 /* Getting status code from kconsumerd */
1366 ret = lttcomm_recv_unix_sock(sock, &code,
1367 sizeof(enum lttcomm_return_code));
1368 if (ret <= 0) {
1369 goto error;
1370 }
1371
1372 health_code_update();
1373 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1374 ERR("consumer error when waiting for SOCK_READY : %s",
1375 lttcomm_get_readable_code(-code));
1376 goto error;
1377 }
1378
1379 /* Connect both command and metadata sockets. */
1380 consumer_data->cmd_sock =
1381 lttcomm_connect_unix_sock(
1382 consumer_data->cmd_unix_sock_path);
1383 consumer_data->metadata_fd =
1384 lttcomm_connect_unix_sock(
1385 consumer_data->cmd_unix_sock_path);
1386 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1387 PERROR("consumer connect cmd socket");
1388 /* On error, signal condition and quit. */
1389 signal_consumer_condition(consumer_data, -1);
1390 goto error;
1391 }
1392
1393 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1394
1395 /* Create metadata socket lock. */
1396 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1397 if (consumer_data->metadata_sock.lock == NULL) {
1398 PERROR("zmalloc pthread mutex");
1399 goto error;
1400 }
1401 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1402
1403 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1404 DBG("Consumer metadata socket ready (fd: %d)",
1405 consumer_data->metadata_fd);
1406
1407 /*
1408 * Remove the consumerd error sock since we've established a connection.
1409 */
1410 ret = lttng_poll_del(&events, consumer_data->err_sock);
1411 if (ret < 0) {
1412 goto error;
1413 }
1414
1415 /* Add new accepted error socket. */
1416 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1417 if (ret < 0) {
1418 goto error;
1419 }
1420
1421 /* Add metadata socket that is successfully connected. */
1422 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1423 LPOLLIN | LPOLLRDHUP);
1424 if (ret < 0) {
1425 goto error;
1426 }
1427
1428 health_code_update();
1429
1430 /*
1431 * Transfer the write-end of the channel monitoring pipe to the
1432 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1433 */
1434 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1435 if (!cmd_socket_wrapper) {
1436 goto error;
1437 }
1438
1439 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1440 consumer_data->channel_monitor_pipe);
1441 if (ret) {
1442 goto error;
1443 }
1444 /* Discard the socket wrapper as it is no longer needed. */
1445 consumer_destroy_socket(cmd_socket_wrapper);
1446 cmd_socket_wrapper = NULL;
1447
1448 /* The thread is completely initialized, signal that it is ready. */
1449 signal_consumer_condition(consumer_data, 1);
1450
1451 /* Infinite blocking call, waiting for transmission */
1452 restart_poll:
1453 while (1) {
1454 health_code_update();
1455
1456 /* Exit the thread because the thread quit pipe has been triggered. */
1457 if (should_quit) {
1458 /* Not a health error. */
1459 err = 0;
1460 goto exit;
1461 }
1462
1463 health_poll_entry();
1464 ret = lttng_poll_wait(&events, -1);
1465 health_poll_exit();
1466 if (ret < 0) {
1467 /*
1468 * Restart interrupted system call.
1469 */
1470 if (errno == EINTR) {
1471 goto restart_poll;
1472 }
1473 goto error;
1474 }
1475
1476 nb_fd = ret;
1477
1478 for (i = 0; i < nb_fd; i++) {
1479 /* Fetch once the poll data */
1480 revents = LTTNG_POLL_GETEV(&events, i);
1481 pollfd = LTTNG_POLL_GETFD(&events, i);
1482
1483 health_code_update();
1484
1485 if (!revents) {
1486 /* No activity for this FD (poll implementation). */
1487 continue;
1488 }
1489
1490 /*
1491 * Thread quit pipe has been triggered, flag that we should stop
1492 * but continue the current loop to handle potential data from
1493 * consumer.
1494 */
1495 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1496
1497 if (pollfd == sock) {
1498 /* Event on the consumerd socket */
1499 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1500 && !(revents & LPOLLIN)) {
1501 ERR("consumer err socket second poll error");
1502 goto error;
1503 }
1504 health_code_update();
1505 /* Wait for any kconsumerd error */
1506 ret = lttcomm_recv_unix_sock(sock, &code,
1507 sizeof(enum lttcomm_return_code));
1508 if (ret <= 0) {
1509 ERR("consumer closed the command socket");
1510 goto error;
1511 }
1512
1513 ERR("consumer return code : %s",
1514 lttcomm_get_readable_code(-code));
1515
1516 goto exit;
1517 } else if (pollfd == consumer_data->metadata_fd) {
1518 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1519 && !(revents & LPOLLIN)) {
1520 ERR("consumer err metadata socket second poll error");
1521 goto error;
1522 }
1523 /* UST metadata requests */
1524 ret = ust_consumer_metadata_request(
1525 &consumer_data->metadata_sock);
1526 if (ret < 0) {
1527 ERR("Handling metadata request");
1528 goto error;
1529 }
1530 }
1531 /* No need for an else branch all FDs are tested prior. */
1532 }
1533 health_code_update();
1534 }
1535
1536 exit:
1537 error:
1538 /*
1539 * We lock here because we are about to close the sockets and some other
1540 * thread might be using them so get exclusive access which will abort all
1541 * other consumer command by other threads.
1542 */
1543 pthread_mutex_lock(&consumer_data->lock);
1544
1545 /* Immediately set the consumerd state to stopped */
1546 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1547 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1548 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1549 consumer_data->type == LTTNG_CONSUMER32_UST) {
1550 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1551 } else {
1552 /* Code flow error... */
1553 assert(0);
1554 }
1555
1556 if (consumer_data->err_sock >= 0) {
1557 ret = close(consumer_data->err_sock);
1558 if (ret) {
1559 PERROR("close");
1560 }
1561 consumer_data->err_sock = -1;
1562 }
1563 if (consumer_data->cmd_sock >= 0) {
1564 ret = close(consumer_data->cmd_sock);
1565 if (ret) {
1566 PERROR("close");
1567 }
1568 consumer_data->cmd_sock = -1;
1569 }
1570 if (consumer_data->metadata_sock.fd_ptr &&
1571 *consumer_data->metadata_sock.fd_ptr >= 0) {
1572 ret = close(*consumer_data->metadata_sock.fd_ptr);
1573 if (ret) {
1574 PERROR("close");
1575 }
1576 }
1577 if (sock >= 0) {
1578 ret = close(sock);
1579 if (ret) {
1580 PERROR("close");
1581 }
1582 }
1583
1584 unlink(consumer_data->err_unix_sock_path);
1585 unlink(consumer_data->cmd_unix_sock_path);
1586 pthread_mutex_unlock(&consumer_data->lock);
1587
1588 /* Cleanup metadata socket mutex. */
1589 if (consumer_data->metadata_sock.lock) {
1590 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1591 free(consumer_data->metadata_sock.lock);
1592 }
1593 lttng_poll_clean(&events);
1594
1595 if (cmd_socket_wrapper) {
1596 consumer_destroy_socket(cmd_socket_wrapper);
1597 }
1598 error_poll:
1599 if (err) {
1600 health_error();
1601 ERR("Health error occurred in %s", __func__);
1602 }
1603 health_unregister(health_sessiond);
1604 DBG("consumer thread cleanup completed");
1605
1606 rcu_thread_offline();
1607 rcu_unregister_thread();
1608
1609 return NULL;
1610 }
1611
1612 /*
1613 * This thread manage application communication.
1614 */
1615 static void *thread_manage_apps(void *data)
1616 {
1617 int i, ret, pollfd, err = -1;
1618 ssize_t size_ret;
1619 uint32_t revents, nb_fd;
1620 struct lttng_poll_event events;
1621
1622 DBG("[thread] Manage application started");
1623
1624 rcu_register_thread();
1625 rcu_thread_online();
1626
1627 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1628
1629 if (testpoint(sessiond_thread_manage_apps)) {
1630 goto error_testpoint;
1631 }
1632
1633 health_code_update();
1634
1635 ret = sessiond_set_thread_pollset(&events, 2);
1636 if (ret < 0) {
1637 goto error_poll_create;
1638 }
1639
1640 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1641 if (ret < 0) {
1642 goto error;
1643 }
1644
1645 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1646 goto error;
1647 }
1648
1649 health_code_update();
1650
1651 while (1) {
1652 DBG("Apps thread polling");
1653
1654 /* Inifinite blocking call, waiting for transmission */
1655 restart:
1656 health_poll_entry();
1657 ret = lttng_poll_wait(&events, -1);
1658 DBG("Apps thread return from poll on %d fds",
1659 LTTNG_POLL_GETNB(&events));
1660 health_poll_exit();
1661 if (ret < 0) {
1662 /*
1663 * Restart interrupted system call.
1664 */
1665 if (errno == EINTR) {
1666 goto restart;
1667 }
1668 goto error;
1669 }
1670
1671 nb_fd = ret;
1672
1673 for (i = 0; i < nb_fd; i++) {
1674 /* Fetch once the poll data */
1675 revents = LTTNG_POLL_GETEV(&events, i);
1676 pollfd = LTTNG_POLL_GETFD(&events, i);
1677
1678 health_code_update();
1679
1680 if (!revents) {
1681 /* No activity for this FD (poll implementation). */
1682 continue;
1683 }
1684
1685 /* Thread quit pipe has been closed. Killing thread. */
1686 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1687 if (ret) {
1688 err = 0;
1689 goto exit;
1690 }
1691
1692 /* Inspect the apps cmd pipe */
1693 if (pollfd == apps_cmd_pipe[0]) {
1694 if (revents & LPOLLIN) {
1695 int sock;
1696
1697 /* Empty pipe */
1698 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1699 if (size_ret < sizeof(sock)) {
1700 PERROR("read apps cmd pipe");
1701 goto error;
1702 }
1703
1704 health_code_update();
1705
1706 /*
1707 * Since this is a command socket (write then read),
1708 * we only monitor the error events of the socket.
1709 */
1710 ret = lttng_poll_add(&events, sock,
1711 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1712 if (ret < 0) {
1713 goto error;
1714 }
1715
1716 DBG("Apps with sock %d added to poll set", sock);
1717 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1718 ERR("Apps command pipe error");
1719 goto error;
1720 } else {
1721 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1722 goto error;
1723 }
1724 } else {
1725 /*
1726 * At this point, we know that a registered application made
1727 * the event at poll_wait.
1728 */
1729 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1730 /* Removing from the poll set */
1731 ret = lttng_poll_del(&events, pollfd);
1732 if (ret < 0) {
1733 goto error;
1734 }
1735
1736 /* Socket closed on remote end. */
1737 ust_app_unregister(pollfd);
1738 } else {
1739 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1740 goto error;
1741 }
1742 }
1743
1744 health_code_update();
1745 }
1746 }
1747
1748 exit:
1749 error:
1750 lttng_poll_clean(&events);
1751 error_poll_create:
1752 error_testpoint:
1753 utils_close_pipe(apps_cmd_pipe);
1754 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1755
1756 /*
1757 * We don't clean the UST app hash table here since already registered
1758 * applications can still be controlled so let them be until the session
1759 * daemon dies or the applications stop.
1760 */
1761
1762 if (err) {
1763 health_error();
1764 ERR("Health error occurred in %s", __func__);
1765 }
1766 health_unregister(health_sessiond);
1767 DBG("Application communication apps thread cleanup complete");
1768 rcu_thread_offline();
1769 rcu_unregister_thread();
1770 return NULL;
1771 }
1772
1773 /*
1774 * Send a socket to a thread This is called from the dispatch UST registration
1775 * thread once all sockets are set for the application.
1776 *
1777 * The sock value can be invalid, we don't really care, the thread will handle
1778 * it and make the necessary cleanup if so.
1779 *
1780 * On success, return 0 else a negative value being the errno message of the
1781 * write().
1782 */
1783 static int send_socket_to_thread(int fd, int sock)
1784 {
1785 ssize_t ret;
1786
1787 /*
1788 * It's possible that the FD is set as invalid with -1 concurrently just
1789 * before calling this function being a shutdown state of the thread.
1790 */
1791 if (fd < 0) {
1792 ret = -EBADF;
1793 goto error;
1794 }
1795
1796 ret = lttng_write(fd, &sock, sizeof(sock));
1797 if (ret < sizeof(sock)) {
1798 PERROR("write apps pipe %d", fd);
1799 if (ret < 0) {
1800 ret = -errno;
1801 }
1802 goto error;
1803 }
1804
1805 /* All good. Don't send back the write positive ret value. */
1806 ret = 0;
1807 error:
1808 return (int) ret;
1809 }
1810
1811 /*
1812 * Sanitize the wait queue of the dispatch registration thread meaning removing
1813 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1814 * notify socket is never received.
1815 */
1816 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1817 {
1818 int ret, nb_fd = 0, i;
1819 unsigned int fd_added = 0;
1820 struct lttng_poll_event events;
1821 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1822
1823 assert(wait_queue);
1824
1825 lttng_poll_init(&events);
1826
1827 /* Just skip everything for an empty queue. */
1828 if (!wait_queue->count) {
1829 goto end;
1830 }
1831
1832 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1833 if (ret < 0) {
1834 goto error_create;
1835 }
1836
1837 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1838 &wait_queue->head, head) {
1839 assert(wait_node->app);
1840 ret = lttng_poll_add(&events, wait_node->app->sock,
1841 LPOLLHUP | LPOLLERR);
1842 if (ret < 0) {
1843 goto error;
1844 }
1845
1846 fd_added = 1;
1847 }
1848
1849 if (!fd_added) {
1850 goto end;
1851 }
1852
1853 /*
1854 * Poll but don't block so we can quickly identify the faulty events and
1855 * clean them afterwards from the wait queue.
1856 */
1857 ret = lttng_poll_wait(&events, 0);
1858 if (ret < 0) {
1859 goto error;
1860 }
1861 nb_fd = ret;
1862
1863 for (i = 0; i < nb_fd; i++) {
1864 /* Get faulty FD. */
1865 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1866 int pollfd = LTTNG_POLL_GETFD(&events, i);
1867
1868 if (!revents) {
1869 /* No activity for this FD (poll implementation). */
1870 continue;
1871 }
1872
1873 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1874 &wait_queue->head, head) {
1875 if (pollfd == wait_node->app->sock &&
1876 (revents & (LPOLLHUP | LPOLLERR))) {
1877 cds_list_del(&wait_node->head);
1878 wait_queue->count--;
1879 ust_app_destroy(wait_node->app);
1880 free(wait_node);
1881 /*
1882 * Silence warning of use-after-free in
1883 * cds_list_for_each_entry_safe which uses
1884 * __typeof__(*wait_node).
1885 */
1886 wait_node = NULL;
1887 break;
1888 } else {
1889 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1890 goto error;
1891 }
1892 }
1893 }
1894
1895 if (nb_fd > 0) {
1896 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1897 }
1898
1899 end:
1900 lttng_poll_clean(&events);
1901 return;
1902
1903 error:
1904 lttng_poll_clean(&events);
1905 error_create:
1906 ERR("Unable to sanitize wait queue");
1907 return;
1908 }
1909
1910 /*
1911 * Dispatch request from the registration threads to the application
1912 * communication thread.
1913 */
1914 static void *thread_dispatch_ust_registration(void *data)
1915 {
1916 int ret, err = -1;
1917 struct cds_wfcq_node *node;
1918 struct ust_command *ust_cmd = NULL;
1919 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1920 struct ust_reg_wait_queue wait_queue = {
1921 .count = 0,
1922 };
1923
1924 rcu_register_thread();
1925
1926 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1927
1928 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1929 goto error_testpoint;
1930 }
1931
1932 health_code_update();
1933
1934 CDS_INIT_LIST_HEAD(&wait_queue.head);
1935
1936 DBG("[thread] Dispatch UST command started");
1937
1938 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1939 health_code_update();
1940
1941 /* Atomically prepare the queue futex */
1942 futex_nto1_prepare(&ust_cmd_queue.futex);
1943
1944 do {
1945 struct ust_app *app = NULL;
1946 ust_cmd = NULL;
1947
1948 /*
1949 * Make sure we don't have node(s) that have hung up before receiving
1950 * the notify socket. This is to clean the list in order to avoid
1951 * memory leaks from notify socket that are never seen.
1952 */
1953 sanitize_wait_queue(&wait_queue);
1954
1955 health_code_update();
1956 /* Dequeue command for registration */
1957 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1958 if (node == NULL) {
1959 DBG("Woken up but nothing in the UST command queue");
1960 /* Continue thread execution */
1961 break;
1962 }
1963
1964 ust_cmd = caa_container_of(node, struct ust_command, node);
1965
1966 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1967 " gid:%d sock:%d name:%s (version %d.%d)",
1968 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1969 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1970 ust_cmd->sock, ust_cmd->reg_msg.name,
1971 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1972
1973 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1974 wait_node = zmalloc(sizeof(*wait_node));
1975 if (!wait_node) {
1976 PERROR("zmalloc wait_node dispatch");
1977 ret = close(ust_cmd->sock);
1978 if (ret < 0) {
1979 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1980 }
1981 lttng_fd_put(LTTNG_FD_APPS, 1);
1982 free(ust_cmd);
1983 goto error;
1984 }
1985 CDS_INIT_LIST_HEAD(&wait_node->head);
1986
1987 /* Create application object if socket is CMD. */
1988 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1989 ust_cmd->sock);
1990 if (!wait_node->app) {
1991 ret = close(ust_cmd->sock);
1992 if (ret < 0) {
1993 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1994 }
1995 lttng_fd_put(LTTNG_FD_APPS, 1);
1996 free(wait_node);
1997 free(ust_cmd);
1998 continue;
1999 }
2000 /*
2001 * Add application to the wait queue so we can set the notify
2002 * socket before putting this object in the global ht.
2003 */
2004 cds_list_add(&wait_node->head, &wait_queue.head);
2005 wait_queue.count++;
2006
2007 free(ust_cmd);
2008 /*
2009 * We have to continue here since we don't have the notify
2010 * socket and the application MUST be added to the hash table
2011 * only at that moment.
2012 */
2013 continue;
2014 } else {
2015 /*
2016 * Look for the application in the local wait queue and set the
2017 * notify socket if found.
2018 */
2019 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2020 &wait_queue.head, head) {
2021 health_code_update();
2022 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
2023 wait_node->app->notify_sock = ust_cmd->sock;
2024 cds_list_del(&wait_node->head);
2025 wait_queue.count--;
2026 app = wait_node->app;
2027 free(wait_node);
2028 DBG3("UST app notify socket %d is set", ust_cmd->sock);
2029 break;
2030 }
2031 }
2032
2033 /*
2034 * With no application at this stage the received socket is
2035 * basically useless so close it before we free the cmd data
2036 * structure for good.
2037 */
2038 if (!app) {
2039 ret = close(ust_cmd->sock);
2040 if (ret < 0) {
2041 PERROR("close ust sock dispatch %d", ust_cmd->sock);
2042 }
2043 lttng_fd_put(LTTNG_FD_APPS, 1);
2044 }
2045 free(ust_cmd);
2046 }
2047
2048 if (app) {
2049 /*
2050 * @session_lock_list
2051 *
2052 * Lock the global session list so from the register up to the
2053 * registration done message, no thread can see the application
2054 * and change its state.
2055 */
2056 session_lock_list();
2057 rcu_read_lock();
2058
2059 /*
2060 * Add application to the global hash table. This needs to be
2061 * done before the update to the UST registry can locate the
2062 * application.
2063 */
2064 ust_app_add(app);
2065
2066 /* Set app version. This call will print an error if needed. */
2067 (void) ust_app_version(app);
2068
2069 /* Send notify socket through the notify pipe. */
2070 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2071 app->notify_sock);
2072 if (ret < 0) {
2073 rcu_read_unlock();
2074 session_unlock_list();
2075 /*
2076 * No notify thread, stop the UST tracing. However, this is
2077 * not an internal error of the this thread thus setting
2078 * the health error code to a normal exit.
2079 */
2080 err = 0;
2081 goto error;
2082 }
2083
2084 /*
2085 * Update newly registered application with the tracing
2086 * registry info already enabled information.
2087 */
2088 update_ust_app(app->sock);
2089
2090 /*
2091 * Don't care about return value. Let the manage apps threads
2092 * handle app unregistration upon socket close.
2093 */
2094 (void) ust_app_register_done(app);
2095
2096 /*
2097 * Even if the application socket has been closed, send the app
2098 * to the thread and unregistration will take place at that
2099 * place.
2100 */
2101 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2102 if (ret < 0) {
2103 rcu_read_unlock();
2104 session_unlock_list();
2105 /*
2106 * No apps. thread, stop the UST tracing. However, this is
2107 * not an internal error of the this thread thus setting
2108 * the health error code to a normal exit.
2109 */
2110 err = 0;
2111 goto error;
2112 }
2113
2114 rcu_read_unlock();
2115 session_unlock_list();
2116 }
2117 } while (node != NULL);
2118
2119 health_poll_entry();
2120 /* Futex wait on queue. Blocking call on futex() */
2121 futex_nto1_wait(&ust_cmd_queue.futex);
2122 health_poll_exit();
2123 }
2124 /* Normal exit, no error */
2125 err = 0;
2126
2127 error:
2128 /* Clean up wait queue. */
2129 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2130 &wait_queue.head, head) {
2131 cds_list_del(&wait_node->head);
2132 wait_queue.count--;
2133 free(wait_node);
2134 }
2135
2136 /* Empty command queue. */
2137 for (;;) {
2138 /* Dequeue command for registration */
2139 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2140 if (node == NULL) {
2141 break;
2142 }
2143 ust_cmd = caa_container_of(node, struct ust_command, node);
2144 ret = close(ust_cmd->sock);
2145 if (ret < 0) {
2146 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2147 }
2148 lttng_fd_put(LTTNG_FD_APPS, 1);
2149 free(ust_cmd);
2150 }
2151
2152 error_testpoint:
2153 DBG("Dispatch thread dying");
2154 if (err) {
2155 health_error();
2156 ERR("Health error occurred in %s", __func__);
2157 }
2158 health_unregister(health_sessiond);
2159 rcu_unregister_thread();
2160 return NULL;
2161 }
2162
2163 /*
2164 * This thread manage application registration.
2165 */
2166 static void *thread_registration_apps(void *data)
2167 {
2168 int sock = -1, i, ret, pollfd, err = -1;
2169 uint32_t revents, nb_fd;
2170 struct lttng_poll_event events;
2171 /*
2172 * Get allocated in this thread, enqueued to a global queue, dequeued and
2173 * freed in the manage apps thread.
2174 */
2175 struct ust_command *ust_cmd = NULL;
2176
2177 DBG("[thread] Manage application registration started");
2178
2179 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2180
2181 if (testpoint(sessiond_thread_registration_apps)) {
2182 goto error_testpoint;
2183 }
2184
2185 ret = lttcomm_listen_unix_sock(apps_sock);
2186 if (ret < 0) {
2187 goto error_listen;
2188 }
2189
2190 /*
2191 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2192 * more will be added to this poll set.
2193 */
2194 ret = sessiond_set_thread_pollset(&events, 2);
2195 if (ret < 0) {
2196 goto error_create_poll;
2197 }
2198
2199 /* Add the application registration socket */
2200 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2201 if (ret < 0) {
2202 goto error_poll_add;
2203 }
2204
2205 /* Notify all applications to register */
2206 ret = notify_ust_apps(1);
2207 if (ret < 0) {
2208 ERR("Failed to notify applications or create the wait shared memory.\n"
2209 "Execution continues but there might be problem for already\n"
2210 "running applications that wishes to register.");
2211 }
2212
2213 while (1) {
2214 DBG("Accepting application registration");
2215
2216 /* Inifinite blocking call, waiting for transmission */
2217 restart:
2218 health_poll_entry();
2219 ret = lttng_poll_wait(&events, -1);
2220 health_poll_exit();
2221 if (ret < 0) {
2222 /*
2223 * Restart interrupted system call.
2224 */
2225 if (errno == EINTR) {
2226 goto restart;
2227 }
2228 goto error;
2229 }
2230
2231 nb_fd = ret;
2232
2233 for (i = 0; i < nb_fd; i++) {
2234 health_code_update();
2235
2236 /* Fetch once the poll data */
2237 revents = LTTNG_POLL_GETEV(&events, i);
2238 pollfd = LTTNG_POLL_GETFD(&events, i);
2239
2240 if (!revents) {
2241 /* No activity for this FD (poll implementation). */
2242 continue;
2243 }
2244
2245 /* Thread quit pipe has been closed. Killing thread. */
2246 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2247 if (ret) {
2248 err = 0;
2249 goto exit;
2250 }
2251
2252 /* Event on the registration socket */
2253 if (pollfd == apps_sock) {
2254 if (revents & LPOLLIN) {
2255 sock = lttcomm_accept_unix_sock(apps_sock);
2256 if (sock < 0) {
2257 goto error;
2258 }
2259
2260 /*
2261 * Set socket timeout for both receiving and ending.
2262 * app_socket_timeout is in seconds, whereas
2263 * lttcomm_setsockopt_rcv_timeout and
2264 * lttcomm_setsockopt_snd_timeout expect msec as
2265 * parameter.
2266 */
2267 if (app_socket_timeout >= 0) {
2268 (void) lttcomm_setsockopt_rcv_timeout(sock,
2269 app_socket_timeout * 1000);
2270 (void) lttcomm_setsockopt_snd_timeout(sock,
2271 app_socket_timeout * 1000);
2272 }
2273
2274 /*
2275 * Set the CLOEXEC flag. Return code is useless because
2276 * either way, the show must go on.
2277 */
2278 (void) utils_set_fd_cloexec(sock);
2279
2280 /* Create UST registration command for enqueuing */
2281 ust_cmd = zmalloc(sizeof(struct ust_command));
2282 if (ust_cmd == NULL) {
2283 PERROR("ust command zmalloc");
2284 ret = close(sock);
2285 if (ret) {
2286 PERROR("close");
2287 }
2288 goto error;
2289 }
2290
2291 /*
2292 * Using message-based transmissions to ensure we don't
2293 * have to deal with partially received messages.
2294 */
2295 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2296 if (ret < 0) {
2297 ERR("Exhausted file descriptors allowed for applications.");
2298 free(ust_cmd);
2299 ret = close(sock);
2300 if (ret) {
2301 PERROR("close");
2302 }
2303 sock = -1;
2304 continue;
2305 }
2306
2307 health_code_update();
2308 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2309 if (ret < 0) {
2310 free(ust_cmd);
2311 /* Close socket of the application. */
2312 ret = close(sock);
2313 if (ret) {
2314 PERROR("close");
2315 }
2316 lttng_fd_put(LTTNG_FD_APPS, 1);
2317 sock = -1;
2318 continue;
2319 }
2320 health_code_update();
2321
2322 ust_cmd->sock = sock;
2323 sock = -1;
2324
2325 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2326 " gid:%d sock:%d name:%s (version %d.%d)",
2327 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2328 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2329 ust_cmd->sock, ust_cmd->reg_msg.name,
2330 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2331
2332 /*
2333 * Lock free enqueue the registration request. The red pill
2334 * has been taken! This apps will be part of the *system*.
2335 */
2336 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2337
2338 /*
2339 * Wake the registration queue futex. Implicit memory
2340 * barrier with the exchange in cds_wfcq_enqueue.
2341 */
2342 futex_nto1_wake(&ust_cmd_queue.futex);
2343 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2344 ERR("Register apps socket poll error");
2345 goto error;
2346 } else {
2347 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2348 goto error;
2349 }
2350 }
2351 }
2352 }
2353
2354 exit:
2355 error:
2356 /* Notify that the registration thread is gone */
2357 notify_ust_apps(0);
2358
2359 if (apps_sock >= 0) {
2360 ret = close(apps_sock);
2361 if (ret) {
2362 PERROR("close");
2363 }
2364 }
2365 if (sock >= 0) {
2366 ret = close(sock);
2367 if (ret) {
2368 PERROR("close");
2369 }
2370 lttng_fd_put(LTTNG_FD_APPS, 1);
2371 }
2372 unlink(apps_unix_sock_path);
2373
2374 error_poll_add:
2375 lttng_poll_clean(&events);
2376 error_listen:
2377 error_create_poll:
2378 error_testpoint:
2379 DBG("UST Registration thread cleanup complete");
2380 if (err) {
2381 health_error();
2382 ERR("Health error occurred in %s", __func__);
2383 }
2384 health_unregister(health_sessiond);
2385
2386 return NULL;
2387 }
2388
2389 /*
2390 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2391 * exec or it will fails.
2392 */
2393 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2394 {
2395 int ret, clock_ret;
2396 struct timespec timeout;
2397
2398 /*
2399 * Make sure we set the readiness flag to 0 because we are NOT ready.
2400 * This access to consumer_thread_is_ready does not need to be
2401 * protected by consumer_data.cond_mutex (yet) since the consumer
2402 * management thread has not been started at this point.
2403 */
2404 consumer_data->consumer_thread_is_ready = 0;
2405
2406 /* Setup pthread condition */
2407 ret = pthread_condattr_init(&consumer_data->condattr);
2408 if (ret) {
2409 errno = ret;
2410 PERROR("pthread_condattr_init consumer data");
2411 goto error;
2412 }
2413
2414 /*
2415 * Set the monotonic clock in order to make sure we DO NOT jump in time
2416 * between the clock_gettime() call and the timedwait call. See bug #324
2417 * for a more details and how we noticed it.
2418 */
2419 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2420 if (ret) {
2421 errno = ret;
2422 PERROR("pthread_condattr_setclock consumer data");
2423 goto error;
2424 }
2425
2426 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2427 if (ret) {
2428 errno = ret;
2429 PERROR("pthread_cond_init consumer data");
2430 goto error;
2431 }
2432
2433 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2434 thread_manage_consumer, consumer_data);
2435 if (ret) {
2436 errno = ret;
2437 PERROR("pthread_create consumer");
2438 ret = -1;
2439 goto error;
2440 }
2441
2442 /* We are about to wait on a pthread condition */
2443 pthread_mutex_lock(&consumer_data->cond_mutex);
2444
2445 /* Get time for sem_timedwait absolute timeout */
2446 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2447 /*
2448 * Set the timeout for the condition timed wait even if the clock gettime
2449 * call fails since we might loop on that call and we want to avoid to
2450 * increment the timeout too many times.
2451 */
2452 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2453
2454 /*
2455 * The following loop COULD be skipped in some conditions so this is why we
2456 * set ret to 0 in order to make sure at least one round of the loop is
2457 * done.
2458 */
2459 ret = 0;
2460
2461 /*
2462 * Loop until the condition is reached or when a timeout is reached. Note
2463 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2464 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2465 * possible. This loop does not take any chances and works with both of
2466 * them.
2467 */
2468 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2469 if (clock_ret < 0) {
2470 PERROR("clock_gettime spawn consumer");
2471 /* Infinite wait for the consumerd thread to be ready */
2472 ret = pthread_cond_wait(&consumer_data->cond,
2473 &consumer_data->cond_mutex);
2474 } else {
2475 ret = pthread_cond_timedwait(&consumer_data->cond,
2476 &consumer_data->cond_mutex, &timeout);
2477 }
2478 }
2479
2480 /* Release the pthread condition */
2481 pthread_mutex_unlock(&consumer_data->cond_mutex);
2482
2483 if (ret != 0) {
2484 errno = ret;
2485 if (ret == ETIMEDOUT) {
2486 int pth_ret;
2487
2488 /*
2489 * Call has timed out so we kill the kconsumerd_thread and return
2490 * an error.
2491 */
2492 ERR("Condition timed out. The consumer thread was never ready."
2493 " Killing it");
2494 pth_ret = pthread_cancel(consumer_data->thread);
2495 if (pth_ret < 0) {
2496 PERROR("pthread_cancel consumer thread");
2497 }
2498 } else {
2499 PERROR("pthread_cond_wait failed consumer thread");
2500 }
2501 /* Caller is expecting a negative value on failure. */
2502 ret = -1;
2503 goto error;
2504 }
2505
2506 pthread_mutex_lock(&consumer_data->pid_mutex);
2507 if (consumer_data->pid == 0) {
2508 ERR("Consumerd did not start");
2509 pthread_mutex_unlock(&consumer_data->pid_mutex);
2510 goto error;
2511 }
2512 pthread_mutex_unlock(&consumer_data->pid_mutex);
2513
2514 return 0;
2515
2516 error:
2517 return ret;
2518 }
2519
2520 /*
2521 * Join consumer thread
2522 */
2523 static int join_consumer_thread(struct consumer_data *consumer_data)
2524 {
2525 void *status;
2526
2527 /* Consumer pid must be a real one. */
2528 if (consumer_data->pid > 0) {
2529 int ret;
2530 ret = kill(consumer_data->pid, SIGTERM);
2531 if (ret) {
2532 PERROR("Error killing consumer daemon");
2533 return ret;
2534 }
2535 return pthread_join(consumer_data->thread, &status);
2536 } else {
2537 return 0;
2538 }
2539 }
2540
2541 /*
2542 * Fork and exec a consumer daemon (consumerd).
2543 *
2544 * Return pid if successful else -1.
2545 */
2546 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2547 {
2548 int ret;
2549 pid_t pid;
2550 const char *consumer_to_use;
2551 const char *verbosity;
2552 struct stat st;
2553
2554 DBG("Spawning consumerd");
2555
2556 pid = fork();
2557 if (pid == 0) {
2558 /*
2559 * Exec consumerd.
2560 */
2561 if (opt_verbose_consumer) {
2562 verbosity = "--verbose";
2563 } else if (lttng_opt_quiet) {
2564 verbosity = "--quiet";
2565 } else {
2566 verbosity = "";
2567 }
2568
2569 switch (consumer_data->type) {
2570 case LTTNG_CONSUMER_KERNEL:
2571 /*
2572 * Find out which consumerd to execute. We will first try the
2573 * 64-bit path, then the sessiond's installation directory, and
2574 * fallback on the 32-bit one,
2575 */
2576 DBG3("Looking for a kernel consumer at these locations:");
2577 DBG3(" 1) %s", consumerd64_bin);
2578 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2579 DBG3(" 3) %s", consumerd32_bin);
2580 if (stat(consumerd64_bin, &st) == 0) {
2581 DBG3("Found location #1");
2582 consumer_to_use = consumerd64_bin;
2583 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2584 DBG3("Found location #2");
2585 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2586 } else if (stat(consumerd32_bin, &st) == 0) {
2587 DBG3("Found location #3");
2588 consumer_to_use = consumerd32_bin;
2589 } else {
2590 DBG("Could not find any valid consumerd executable");
2591 ret = -EINVAL;
2592 break;
2593 }
2594 DBG("Using kernel consumer at: %s", consumer_to_use);
2595 ret = execl(consumer_to_use,
2596 "lttng-consumerd", verbosity, "-k",
2597 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2598 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2599 "--group", tracing_group_name,
2600 NULL);
2601 break;
2602 case LTTNG_CONSUMER64_UST:
2603 {
2604 char *tmpnew = NULL;
2605
2606 if (consumerd64_libdir[0] != '\0') {
2607 char *tmp;
2608 size_t tmplen;
2609
2610 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2611 if (!tmp) {
2612 tmp = "";
2613 }
2614 tmplen = strlen("LD_LIBRARY_PATH=")
2615 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2616 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2617 if (!tmpnew) {
2618 ret = -ENOMEM;
2619 goto error;
2620 }
2621 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2622 strcat(tmpnew, consumerd64_libdir);
2623 if (tmp[0] != '\0') {
2624 strcat(tmpnew, ":");
2625 strcat(tmpnew, tmp);
2626 }
2627 ret = putenv(tmpnew);
2628 if (ret) {
2629 ret = -errno;
2630 free(tmpnew);
2631 goto error;
2632 }
2633 }
2634 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2635 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2636 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2637 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2638 "--group", tracing_group_name,
2639 NULL);
2640 if (consumerd64_libdir[0] != '\0') {
2641 free(tmpnew);
2642 }
2643 break;
2644 }
2645 case LTTNG_CONSUMER32_UST:
2646 {
2647 char *tmpnew = NULL;
2648
2649 if (consumerd32_libdir[0] != '\0') {
2650 char *tmp;
2651 size_t tmplen;
2652
2653 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2654 if (!tmp) {
2655 tmp = "";
2656 }
2657 tmplen = strlen("LD_LIBRARY_PATH=")
2658 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2659 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2660 if (!tmpnew) {
2661 ret = -ENOMEM;
2662 goto error;
2663 }
2664 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2665 strcat(tmpnew, consumerd32_libdir);
2666 if (tmp[0] != '\0') {
2667 strcat(tmpnew, ":");
2668 strcat(tmpnew, tmp);
2669 }
2670 ret = putenv(tmpnew);
2671 if (ret) {
2672 ret = -errno;
2673 free(tmpnew);
2674 goto error;
2675 }
2676 }
2677 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2678 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2679 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2680 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2681 "--group", tracing_group_name,
2682 NULL);
2683 if (consumerd32_libdir[0] != '\0') {
2684 free(tmpnew);
2685 }
2686 break;
2687 }
2688 default:
2689 PERROR("unknown consumer type");
2690 exit(EXIT_FAILURE);
2691 }
2692 if (errno != 0) {
2693 PERROR("Consumer execl()");
2694 }
2695 /* Reaching this point, we got a failure on our execl(). */
2696 exit(EXIT_FAILURE);
2697 } else if (pid > 0) {
2698 ret = pid;
2699 } else {
2700 PERROR("start consumer fork");
2701 ret = -errno;
2702 }
2703 error:
2704 return ret;
2705 }
2706
2707 /*
2708 * Spawn the consumerd daemon and session daemon thread.
2709 */
2710 static int start_consumerd(struct consumer_data *consumer_data)
2711 {
2712 int ret;
2713
2714 /*
2715 * Set the listen() state on the socket since there is a possible race
2716 * between the exec() of the consumer daemon and this call if place in the
2717 * consumer thread. See bug #366 for more details.
2718 */
2719 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2720 if (ret < 0) {
2721 goto error;
2722 }
2723
2724 pthread_mutex_lock(&consumer_data->pid_mutex);
2725 if (consumer_data->pid != 0) {
2726 pthread_mutex_unlock(&consumer_data->pid_mutex);
2727 goto end;
2728 }
2729
2730 ret = spawn_consumerd(consumer_data);
2731 if (ret < 0) {
2732 ERR("Spawning consumerd failed");
2733 pthread_mutex_unlock(&consumer_data->pid_mutex);
2734 goto error;
2735 }
2736
2737 /* Setting up the consumer_data pid */
2738 consumer_data->pid = ret;
2739 DBG2("Consumer pid %d", consumer_data->pid);
2740 pthread_mutex_unlock(&consumer_data->pid_mutex);
2741
2742 DBG2("Spawning consumer control thread");
2743 ret = spawn_consumer_thread(consumer_data);
2744 if (ret < 0) {
2745 ERR("Fatal error spawning consumer control thread");
2746 goto error;
2747 }
2748
2749 end:
2750 return 0;
2751
2752 error:
2753 /* Cleanup already created sockets on error. */
2754 if (consumer_data->err_sock >= 0) {
2755 int err;
2756
2757 err = close(consumer_data->err_sock);
2758 if (err < 0) {
2759 PERROR("close consumer data error socket");
2760 }
2761 }
2762 return ret;
2763 }
2764
2765 /*
2766 * Setup necessary data for kernel tracer action.
2767 */
2768 static int init_kernel_tracer(void)
2769 {
2770 int ret;
2771
2772 /* Modprobe lttng kernel modules */
2773 ret = modprobe_lttng_control();
2774 if (ret < 0) {
2775 goto error;
2776 }
2777
2778 /* Open debugfs lttng */
2779 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2780 if (kernel_tracer_fd < 0) {
2781 DBG("Failed to open %s", module_proc_lttng);
2782 ret = -1;
2783 goto error_open;
2784 }
2785
2786 /* Validate kernel version */
2787 ret = kernel_validate_version(kernel_tracer_fd);
2788 if (ret < 0) {
2789 goto error_version;
2790 }
2791
2792 ret = modprobe_lttng_data();
2793 if (ret < 0) {
2794 goto error_modules;
2795 }
2796
2797 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2798 return 0;
2799
2800 error_version:
2801 modprobe_remove_lttng_control();
2802 ret = close(kernel_tracer_fd);
2803 if (ret) {
2804 PERROR("close");
2805 }
2806 kernel_tracer_fd = -1;
2807 return LTTNG_ERR_KERN_VERSION;
2808
2809 error_modules:
2810 ret = close(kernel_tracer_fd);
2811 if (ret) {
2812 PERROR("close");
2813 }
2814
2815 error_open:
2816 modprobe_remove_lttng_control();
2817
2818 error:
2819 WARN("No kernel tracer available");
2820 kernel_tracer_fd = -1;
2821 if (!is_root) {
2822 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2823 } else {
2824 return LTTNG_ERR_KERN_NA;
2825 }
2826 }
2827
2828
2829 /*
2830 * Copy consumer output from the tracing session to the domain session. The
2831 * function also applies the right modification on a per domain basis for the
2832 * trace files destination directory.
2833 *
2834 * Should *NOT* be called with RCU read-side lock held.
2835 */
2836 static int copy_session_consumer(int domain, struct ltt_session *session)
2837 {
2838 int ret;
2839 const char *dir_name;
2840 struct consumer_output *consumer;
2841
2842 assert(session);
2843 assert(session->consumer);
2844
2845 switch (domain) {
2846 case LTTNG_DOMAIN_KERNEL:
2847 DBG3("Copying tracing session consumer output in kernel session");
2848 /*
2849 * XXX: We should audit the session creation and what this function
2850 * does "extra" in order to avoid a destroy since this function is used
2851 * in the domain session creation (kernel and ust) only. Same for UST
2852 * domain.
2853 */
2854 if (session->kernel_session->consumer) {
2855 consumer_output_put(session->kernel_session->consumer);
2856 }
2857 session->kernel_session->consumer =
2858 consumer_copy_output(session->consumer);
2859 /* Ease our life a bit for the next part */
2860 consumer = session->kernel_session->consumer;
2861 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2862 break;
2863 case LTTNG_DOMAIN_JUL:
2864 case LTTNG_DOMAIN_LOG4J:
2865 case LTTNG_DOMAIN_PYTHON:
2866 case LTTNG_DOMAIN_UST:
2867 DBG3("Copying tracing session consumer output in UST session");
2868 if (session->ust_session->consumer) {
2869 consumer_output_put(session->ust_session->consumer);
2870 }
2871 session->ust_session->consumer =
2872 consumer_copy_output(session->consumer);
2873 /* Ease our life a bit for the next part */
2874 consumer = session->ust_session->consumer;
2875 dir_name = DEFAULT_UST_TRACE_DIR;
2876 break;
2877 default:
2878 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2879 goto error;
2880 }
2881
2882 /* Append correct directory to subdir */
2883 strncat(consumer->subdir, dir_name,
2884 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2885 DBG3("Copy session consumer subdir %s", consumer->subdir);
2886
2887 ret = LTTNG_OK;
2888
2889 error:
2890 return ret;
2891 }
2892
2893 /*
2894 * Create an UST session and add it to the session ust list.
2895 *
2896 * Should *NOT* be called with RCU read-side lock held.
2897 */
2898 static int create_ust_session(struct ltt_session *session,
2899 struct lttng_domain *domain)
2900 {
2901 int ret;
2902 struct ltt_ust_session *lus = NULL;
2903
2904 assert(session);
2905 assert(domain);
2906 assert(session->consumer);
2907
2908 switch (domain->type) {
2909 case LTTNG_DOMAIN_JUL:
2910 case LTTNG_DOMAIN_LOG4J:
2911 case LTTNG_DOMAIN_PYTHON:
2912 case LTTNG_DOMAIN_UST:
2913 break;
2914 default:
2915 ERR("Unknown UST domain on create session %d", domain->type);
2916 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2917 goto error;
2918 }
2919
2920 DBG("Creating UST session");
2921
2922 lus = trace_ust_create_session(session->id);
2923 if (lus == NULL) {
2924 ret = LTTNG_ERR_UST_SESS_FAIL;
2925 goto error;
2926 }
2927
2928 lus->uid = session->uid;
2929 lus->gid = session->gid;
2930 lus->output_traces = session->output_traces;
2931 lus->snapshot_mode = session->snapshot_mode;
2932 lus->live_timer_interval = session->live_timer;
2933 session->ust_session = lus;
2934 if (session->shm_path[0]) {
2935 strncpy(lus->root_shm_path, session->shm_path,
2936 sizeof(lus->root_shm_path));
2937 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2938 strncpy(lus->shm_path, session->shm_path,
2939 sizeof(lus->shm_path));
2940 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2941 strncat(lus->shm_path, "/ust",
2942 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2943 }
2944 /* Copy session output to the newly created UST session */
2945 ret = copy_session_consumer(domain->type, session);
2946 if (ret != LTTNG_OK) {
2947 goto error;
2948 }
2949
2950 return LTTNG_OK;
2951
2952 error:
2953 free(lus);
2954 session->ust_session = NULL;
2955 return ret;
2956 }
2957
2958 /*
2959 * Create a kernel tracer session then create the default channel.
2960 */
2961 static int create_kernel_session(struct ltt_session *session)
2962 {
2963 int ret;
2964
2965 DBG("Creating kernel session");
2966
2967 ret = kernel_create_session(session, kernel_tracer_fd);
2968 if (ret < 0) {
2969 ret = LTTNG_ERR_KERN_SESS_FAIL;
2970 goto error;
2971 }
2972
2973 /* Code flow safety */
2974 assert(session->kernel_session);
2975
2976 /* Copy session output to the newly created Kernel session */
2977 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2978 if (ret != LTTNG_OK) {
2979 goto error;
2980 }
2981
2982 /* Create directory(ies) on local filesystem. */
2983 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2984 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2985 ret = run_as_mkdir_recursive(
2986 session->kernel_session->consumer->dst.trace_path,
2987 S_IRWXU | S_IRWXG, session->uid, session->gid);
2988 if (ret < 0) {
2989 if (errno != EEXIST) {
2990 ERR("Trace directory creation error");
2991 goto error;
2992 }
2993 }
2994 }
2995
2996 session->kernel_session->uid = session->uid;
2997 session->kernel_session->gid = session->gid;
2998 session->kernel_session->output_traces = session->output_traces;
2999 session->kernel_session->snapshot_mode = session->snapshot_mode;
3000
3001 return LTTNG_OK;
3002
3003 error:
3004 trace_kernel_destroy_session(session->kernel_session);
3005 session->kernel_session = NULL;
3006 return ret;
3007 }
3008
3009 /*
3010 * Count number of session permitted by uid/gid.
3011 */
3012 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
3013 {
3014 unsigned int i = 0;
3015 struct ltt_session *session;
3016
3017 DBG("Counting number of available session for UID %d GID %d",
3018 uid, gid);
3019 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
3020 /*
3021 * Only list the sessions the user can control.
3022 */
3023 if (!session_access_ok(session, uid, gid)) {
3024 continue;
3025 }
3026 i++;
3027 }
3028 return i;
3029 }
3030
3031 /*
3032 * Process the command requested by the lttng client within the command
3033 * context structure. This function make sure that the return structure (llm)
3034 * is set and ready for transmission before returning.
3035 *
3036 * Return any error encountered or 0 for success.
3037 *
3038 * "sock" is only used for special-case var. len data.
3039 *
3040 * Should *NOT* be called with RCU read-side lock held.
3041 */
3042 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
3043 int *sock_error)
3044 {
3045 int ret = LTTNG_OK;
3046 int need_tracing_session = 1;
3047 int need_domain;
3048
3049 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
3050
3051 assert(!rcu_read_ongoing());
3052
3053 *sock_error = 0;
3054
3055 switch (cmd_ctx->lsm->cmd_type) {
3056 case LTTNG_CREATE_SESSION:
3057 case LTTNG_CREATE_SESSION_SNAPSHOT:
3058 case LTTNG_CREATE_SESSION_LIVE:
3059 case LTTNG_DESTROY_SESSION:
3060 case LTTNG_LIST_SESSIONS:
3061 case LTTNG_LIST_DOMAINS:
3062 case LTTNG_START_TRACE:
3063 case LTTNG_STOP_TRACE:
3064 case LTTNG_DATA_PENDING:
3065 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3066 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3067 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3068 case LTTNG_SNAPSHOT_RECORD:
3069 case LTTNG_SAVE_SESSION:
3070 case LTTNG_SET_SESSION_SHM_PATH:
3071 case LTTNG_REGENERATE_METADATA:
3072 case LTTNG_REGENERATE_STATEDUMP:
3073 case LTTNG_REGISTER_TRIGGER:
3074 case LTTNG_UNREGISTER_TRIGGER:
3075 need_domain = 0;
3076 break;
3077 default:
3078 need_domain = 1;
3079 }
3080
3081 if (opt_no_kernel && need_domain
3082 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3083 if (!is_root) {
3084 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3085 } else {
3086 ret = LTTNG_ERR_KERN_NA;
3087 }
3088 goto error;
3089 }
3090
3091 /* Deny register consumer if we already have a spawned consumer. */
3092 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3093 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3094 if (kconsumer_data.pid > 0) {
3095 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3096 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3097 goto error;
3098 }
3099 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3100 }
3101
3102 /*
3103 * Check for command that don't needs to allocate a returned payload. We do
3104 * this here so we don't have to make the call for no payload at each
3105 * command.
3106 */
3107 switch(cmd_ctx->lsm->cmd_type) {
3108 case LTTNG_LIST_SESSIONS:
3109 case LTTNG_LIST_TRACEPOINTS:
3110 case LTTNG_LIST_TRACEPOINT_FIELDS:
3111 case LTTNG_LIST_DOMAINS:
3112 case LTTNG_LIST_CHANNELS:
3113 case LTTNG_LIST_EVENTS:
3114 case LTTNG_LIST_SYSCALLS:
3115 case LTTNG_LIST_TRACKER_PIDS:
3116 case LTTNG_DATA_PENDING:
3117 break;
3118 default:
3119 /* Setup lttng message with no payload */
3120 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3121 if (ret < 0) {
3122 /* This label does not try to unlock the session */
3123 goto init_setup_error;
3124 }
3125 }
3126
3127 /* Commands that DO NOT need a session. */
3128 switch (cmd_ctx->lsm->cmd_type) {
3129 case LTTNG_CREATE_SESSION:
3130 case LTTNG_CREATE_SESSION_SNAPSHOT:
3131 case LTTNG_CREATE_SESSION_LIVE:
3132 case LTTNG_LIST_SESSIONS:
3133 case LTTNG_LIST_TRACEPOINTS:
3134 case LTTNG_LIST_SYSCALLS:
3135 case LTTNG_LIST_TRACEPOINT_FIELDS:
3136 case LTTNG_SAVE_SESSION:
3137 case LTTNG_REGISTER_TRIGGER:
3138 case LTTNG_UNREGISTER_TRIGGER:
3139 need_tracing_session = 0;
3140 break;
3141 default:
3142 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3143 /*
3144 * We keep the session list lock across _all_ commands
3145 * for now, because the per-session lock does not
3146 * handle teardown properly.
3147 */
3148 session_lock_list();
3149 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3150 if (cmd_ctx->session == NULL) {
3151 ret = LTTNG_ERR_SESS_NOT_FOUND;
3152 goto error;
3153 } else {
3154 /* Acquire lock for the session */
3155 session_lock(cmd_ctx->session);
3156 }
3157 break;
3158 }
3159
3160 /*
3161 * Commands that need a valid session but should NOT create one if none
3162 * exists. Instead of creating one and destroying it when the command is
3163 * handled, process that right before so we save some round trip in useless
3164 * code path.
3165 */
3166 switch (cmd_ctx->lsm->cmd_type) {
3167 case LTTNG_DISABLE_CHANNEL:
3168 case LTTNG_DISABLE_EVENT:
3169 switch (cmd_ctx->lsm->domain.type) {
3170 case LTTNG_DOMAIN_KERNEL:
3171 if (!cmd_ctx->session->kernel_session) {
3172 ret = LTTNG_ERR_NO_CHANNEL;
3173 goto error;
3174 }
3175 break;
3176 case LTTNG_DOMAIN_JUL:
3177 case LTTNG_DOMAIN_LOG4J:
3178 case LTTNG_DOMAIN_PYTHON:
3179 case LTTNG_DOMAIN_UST:
3180 if (!cmd_ctx->session->ust_session) {
3181 ret = LTTNG_ERR_NO_CHANNEL;
3182 goto error;
3183 }
3184 break;
3185 default:
3186 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3187 goto error;
3188 }
3189 default:
3190 break;
3191 }
3192
3193 if (!need_domain) {
3194 goto skip_domain;
3195 }
3196
3197 /*
3198 * Check domain type for specific "pre-action".
3199 */
3200 switch (cmd_ctx->lsm->domain.type) {
3201 case LTTNG_DOMAIN_KERNEL:
3202 if (!is_root) {
3203 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3204 goto error;
3205 }
3206
3207 /* Kernel tracer check */
3208 if (kernel_tracer_fd == -1) {
3209 /* Basically, load kernel tracer modules */
3210 ret = init_kernel_tracer();
3211 if (ret != 0) {
3212 goto error;
3213 }
3214 }
3215
3216 /* Consumer is in an ERROR state. Report back to client */
3217 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3218 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3219 goto error;
3220 }
3221
3222 /* Need a session for kernel command */
3223 if (need_tracing_session) {
3224 if (cmd_ctx->session->kernel_session == NULL) {
3225 ret = create_kernel_session(cmd_ctx->session);
3226 if (ret < 0) {
3227 ret = LTTNG_ERR_KERN_SESS_FAIL;
3228 goto error;
3229 }
3230 }
3231
3232 /* Start the kernel consumer daemon */
3233 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3234 if (kconsumer_data.pid == 0 &&
3235 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3236 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3237 ret = start_consumerd(&kconsumer_data);
3238 if (ret < 0) {
3239 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3240 goto error;
3241 }
3242 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3243 } else {
3244 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3245 }
3246
3247 /*
3248 * The consumer was just spawned so we need to add the socket to
3249 * the consumer output of the session if exist.
3250 */
3251 ret = consumer_create_socket(&kconsumer_data,
3252 cmd_ctx->session->kernel_session->consumer);
3253 if (ret < 0) {
3254 goto error;
3255 }
3256 }
3257
3258 break;
3259 case LTTNG_DOMAIN_JUL:
3260 case LTTNG_DOMAIN_LOG4J:
3261 case LTTNG_DOMAIN_PYTHON:
3262 case LTTNG_DOMAIN_UST:
3263 {
3264 if (!ust_app_supported()) {
3265 ret = LTTNG_ERR_NO_UST;
3266 goto error;
3267 }
3268 /* Consumer is in an ERROR state. Report back to client */
3269 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3270 ret = LTTNG_ERR_NO_USTCONSUMERD;
3271 goto error;
3272 }
3273
3274 if (need_tracing_session) {
3275 /* Create UST session if none exist. */
3276 if (cmd_ctx->session->ust_session == NULL) {
3277 ret = create_ust_session(cmd_ctx->session,
3278 &cmd_ctx->lsm->domain);
3279 if (ret != LTTNG_OK) {
3280 goto error;
3281 }
3282 }
3283
3284 /* Start the UST consumer daemons */
3285 /* 64-bit */
3286 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3287 if (consumerd64_bin[0] != '\0' &&
3288 ustconsumer64_data.pid == 0 &&
3289 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3290 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3291 ret = start_consumerd(&ustconsumer64_data);
3292 if (ret < 0) {
3293 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3294 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3295 goto error;
3296 }
3297
3298 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3299 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3300 } else {
3301 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3302 }
3303
3304 /*
3305 * Setup socket for consumer 64 bit. No need for atomic access
3306 * since it was set above and can ONLY be set in this thread.
3307 */
3308 ret = consumer_create_socket(&ustconsumer64_data,
3309 cmd_ctx->session->ust_session->consumer);
3310 if (ret < 0) {
3311 goto error;
3312 }
3313
3314 /* 32-bit */
3315 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3316 if (consumerd32_bin[0] != '\0' &&
3317 ustconsumer32_data.pid == 0 &&
3318 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3319 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3320 ret = start_consumerd(&ustconsumer32_data);
3321 if (ret < 0) {
3322 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3323 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3324 goto error;
3325 }
3326
3327 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3328 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3329 } else {
3330 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3331 }
3332
3333 /*
3334 * Setup socket for consumer 64 bit. No need for atomic access
3335 * since it was set above and can ONLY be set in this thread.
3336 */
3337 ret = consumer_create_socket(&ustconsumer32_data,
3338 cmd_ctx->session->ust_session->consumer);
3339 if (ret < 0) {
3340 goto error;
3341 }
3342 }
3343 break;
3344 }
3345 default:
3346 break;
3347 }
3348 skip_domain:
3349
3350 /* Validate consumer daemon state when start/stop trace command */
3351 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3352 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3353 switch (cmd_ctx->lsm->domain.type) {
3354 case LTTNG_DOMAIN_NONE:
3355 break;
3356 case LTTNG_DOMAIN_JUL:
3357 case LTTNG_DOMAIN_LOG4J:
3358 case LTTNG_DOMAIN_PYTHON:
3359 case LTTNG_DOMAIN_UST:
3360 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3361 ret = LTTNG_ERR_NO_USTCONSUMERD;
3362 goto error;
3363 }
3364 break;
3365 case LTTNG_DOMAIN_KERNEL:
3366 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3367 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3368 goto error;
3369 }
3370 break;
3371 default:
3372 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3373 goto error;
3374 }
3375 }
3376
3377 /*
3378 * Check that the UID or GID match that of the tracing session.
3379 * The root user can interact with all sessions.
3380 */
3381 if (need_tracing_session) {
3382 if (!session_access_ok(cmd_ctx->session,
3383 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3384 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3385 ret = LTTNG_ERR_EPERM;
3386 goto error;
3387 }
3388 }
3389
3390 /*
3391 * Send relayd information to consumer as soon as we have a domain and a
3392 * session defined.
3393 */
3394 if (cmd_ctx->session && need_domain) {
3395 /*
3396 * Setup relayd if not done yet. If the relayd information was already
3397 * sent to the consumer, this call will gracefully return.
3398 */
3399 ret = cmd_setup_relayd(cmd_ctx->session);
3400 if (ret != LTTNG_OK) {
3401 goto error;
3402 }
3403 }
3404
3405 /* Process by command type */
3406 switch (cmd_ctx->lsm->cmd_type) {
3407 case LTTNG_ADD_CONTEXT:
3408 {
3409 /*
3410 * An LTTNG_ADD_CONTEXT command might have a supplementary
3411 * payload if the context being added is an application context.
3412 */
3413 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3414 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3415 char *provider_name = NULL, *context_name = NULL;
3416 size_t provider_name_len =
3417 cmd_ctx->lsm->u.context.provider_name_len;
3418 size_t context_name_len =
3419 cmd_ctx->lsm->u.context.context_name_len;
3420
3421 if (provider_name_len == 0 || context_name_len == 0) {
3422 /*
3423 * Application provider and context names MUST
3424 * be provided.
3425 */
3426 ret = -LTTNG_ERR_INVALID;
3427 goto error;
3428 }
3429
3430 provider_name = zmalloc(provider_name_len + 1);
3431 if (!provider_name) {
3432 ret = -LTTNG_ERR_NOMEM;
3433 goto error;
3434 }
3435 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3436 provider_name;
3437
3438 context_name = zmalloc(context_name_len + 1);
3439 if (!context_name) {
3440 ret = -LTTNG_ERR_NOMEM;
3441 goto error_add_context;
3442 }
3443 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3444 context_name;
3445
3446 ret = lttcomm_recv_unix_sock(sock, provider_name,
3447 provider_name_len);
3448 if (ret < 0) {
3449 goto error_add_context;
3450 }
3451
3452 ret = lttcomm_recv_unix_sock(sock, context_name,
3453 context_name_len);
3454 if (ret < 0) {
3455 goto error_add_context;
3456 }
3457 }
3458
3459 /*
3460 * cmd_add_context assumes ownership of the provider and context
3461 * names.
3462 */
3463 ret = cmd_add_context(cmd_ctx->session,
3464 cmd_ctx->lsm->domain.type,
3465 cmd_ctx->lsm->u.context.channel_name,
3466 &cmd_ctx->lsm->u.context.ctx,
3467 kernel_poll_pipe[1]);
3468
3469 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3470 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3471 error_add_context:
3472 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3473 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3474 if (ret < 0) {
3475 goto error;
3476 }
3477 break;
3478 }
3479 case LTTNG_DISABLE_CHANNEL:
3480 {
3481 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3482 cmd_ctx->lsm->u.disable.channel_name);
3483 break;
3484 }
3485 case LTTNG_DISABLE_EVENT:
3486 {
3487
3488 /*
3489 * FIXME: handle filter; for now we just receive the filter's
3490 * bytecode along with the filter expression which are sent by
3491 * liblttng-ctl and discard them.
3492 *
3493 * This fixes an issue where the client may block while sending
3494 * the filter payload and encounter an error because the session
3495 * daemon closes the socket without ever handling this data.
3496 */
3497 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3498 cmd_ctx->lsm->u.disable.bytecode_len;
3499
3500 if (count) {
3501 char data[LTTNG_FILTER_MAX_LEN];
3502
3503 DBG("Discarding disable event command payload of size %zu", count);
3504 while (count) {
3505 ret = lttcomm_recv_unix_sock(sock, data,
3506 count > sizeof(data) ? sizeof(data) : count);
3507 if (ret < 0) {
3508 goto error;
3509 }
3510
3511 count -= (size_t) ret;
3512 }
3513 }
3514 /* FIXME: passing packed structure to non-packed pointer */
3515 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3516 cmd_ctx->lsm->u.disable.channel_name,
3517 &cmd_ctx->lsm->u.disable.event);
3518 break;
3519 }
3520 case LTTNG_ENABLE_CHANNEL:
3521 {
3522 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3523 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3524 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3525 &cmd_ctx->lsm->u.channel.chan,
3526 kernel_poll_pipe[1]);
3527 break;
3528 }
3529 case LTTNG_TRACK_PID:
3530 {
3531 ret = cmd_track_pid(cmd_ctx->session,
3532 cmd_ctx->lsm->domain.type,
3533 cmd_ctx->lsm->u.pid_tracker.pid);
3534 break;
3535 }
3536 case LTTNG_UNTRACK_PID:
3537 {
3538 ret = cmd_untrack_pid(cmd_ctx->session,
3539 cmd_ctx->lsm->domain.type,
3540 cmd_ctx->lsm->u.pid_tracker.pid);
3541 break;
3542 }
3543 case LTTNG_ENABLE_EVENT:
3544 {
3545 struct lttng_event_exclusion *exclusion = NULL;
3546 struct lttng_filter_bytecode *bytecode = NULL;
3547 char *filter_expression = NULL;
3548
3549 /* Handle exclusion events and receive it from the client. */
3550 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3551 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3552
3553 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3554 (count * LTTNG_SYMBOL_NAME_LEN));
3555 if (!exclusion) {
3556 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3557 goto error;
3558 }
3559
3560 DBG("Receiving var len exclusion event list from client ...");
3561 exclusion->count = count;
3562 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3563 count * LTTNG_SYMBOL_NAME_LEN);
3564 if (ret <= 0) {
3565 DBG("Nothing recv() from client var len data... continuing");
3566 *sock_error = 1;
3567 free(exclusion);
3568 ret = LTTNG_ERR_EXCLUSION_INVAL;
3569 goto error;
3570 }
3571 }
3572
3573 /* Get filter expression from client. */
3574 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3575 size_t expression_len =
3576 cmd_ctx->lsm->u.enable.expression_len;
3577
3578 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3579 ret = LTTNG_ERR_FILTER_INVAL;
3580 free(exclusion);
3581 goto error;
3582 }
3583
3584 filter_expression = zmalloc(expression_len);
3585 if (!filter_expression) {
3586 free(exclusion);
3587 ret = LTTNG_ERR_FILTER_NOMEM;
3588 goto error;
3589 }
3590
3591 /* Receive var. len. data */
3592 DBG("Receiving var len filter's expression from client ...");
3593 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3594 expression_len);
3595 if (ret <= 0) {
3596 DBG("Nothing recv() from client car len data... continuing");
3597 *sock_error = 1;
3598 free(filter_expression);
3599 free(exclusion);
3600 ret = LTTNG_ERR_FILTER_INVAL;
3601 goto error;
3602 }
3603 }
3604
3605 /* Handle filter and get bytecode from client. */
3606 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3607 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3608
3609 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3610 ret = LTTNG_ERR_FILTER_INVAL;
3611 free(filter_expression);
3612 free(exclusion);
3613 goto error;
3614 }
3615
3616 bytecode = zmalloc(bytecode_len);
3617 if (!bytecode) {
3618 free(filter_expression);
3619 free(exclusion);
3620 ret = LTTNG_ERR_FILTER_NOMEM;
3621 goto error;
3622 }
3623
3624 /* Receive var. len. data */
3625 DBG("Receiving var len filter's bytecode from client ...");
3626 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3627 if (ret <= 0) {
3628 DBG("Nothing recv() from client car len data... continuing");
3629 *sock_error = 1;
3630 free(filter_expression);
3631 free(bytecode);
3632 free(exclusion);
3633 ret = LTTNG_ERR_FILTER_INVAL;
3634 goto error;
3635 }
3636
3637 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3638 free(filter_expression);
3639 free(bytecode);
3640 free(exclusion);
3641 ret = LTTNG_ERR_FILTER_INVAL;
3642 goto error;
3643 }
3644 }
3645
3646 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3647 cmd_ctx->lsm->u.enable.channel_name,
3648 &cmd_ctx->lsm->u.enable.event,
3649 filter_expression, bytecode, exclusion,
3650 kernel_poll_pipe[1]);
3651 break;
3652 }
3653 case LTTNG_LIST_TRACEPOINTS:
3654 {
3655 struct lttng_event *events;
3656 ssize_t nb_events;
3657
3658 session_lock_list();
3659 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3660 session_unlock_list();
3661 if (nb_events < 0) {
3662 /* Return value is a negative lttng_error_code. */
3663 ret = -nb_events;
3664 goto error;
3665 }
3666
3667 /*
3668 * Setup lttng message with payload size set to the event list size in
3669 * bytes and then copy list into the llm payload.
3670 */
3671 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3672 sizeof(struct lttng_event) * nb_events);
3673 free(events);
3674
3675 if (ret < 0) {
3676 goto setup_error;
3677 }
3678
3679 ret = LTTNG_OK;
3680 break;
3681 }
3682 case LTTNG_LIST_TRACEPOINT_FIELDS:
3683 {
3684 struct lttng_event_field *fields;
3685 ssize_t nb_fields;
3686
3687 session_lock_list();
3688 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3689 &fields);
3690 session_unlock_list();
3691 if (nb_fields < 0) {
3692 /* Return value is a negative lttng_error_code. */
3693 ret = -nb_fields;
3694 goto error;
3695 }
3696
3697 /*
3698 * Setup lttng message with payload size set to the event list size in
3699 * bytes and then copy list into the llm payload.
3700 */
3701 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3702 sizeof(struct lttng_event_field) * nb_fields);
3703 free(fields);
3704
3705 if (ret < 0) {
3706 goto setup_error;
3707 }
3708
3709 ret = LTTNG_OK;
3710 break;
3711 }
3712 case LTTNG_LIST_SYSCALLS:
3713 {
3714 struct lttng_event *events;
3715 ssize_t nb_events;
3716
3717 nb_events = cmd_list_syscalls(&events);
3718 if (nb_events < 0) {
3719 /* Return value is a negative lttng_error_code. */
3720 ret = -nb_events;
3721 goto error;
3722 }
3723
3724 /*
3725 * Setup lttng message with payload size set to the event list size in
3726 * bytes and then copy list into the llm payload.
3727 */
3728 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3729 sizeof(struct lttng_event) * nb_events);
3730 free(events);
3731
3732 if (ret < 0) {
3733 goto setup_error;
3734 }
3735
3736 ret = LTTNG_OK;
3737 break;
3738 }
3739 case LTTNG_LIST_TRACKER_PIDS:
3740 {
3741 int32_t *pids = NULL;
3742 ssize_t nr_pids;
3743
3744 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3745 cmd_ctx->lsm->domain.type, &pids);
3746 if (nr_pids < 0) {
3747 /* Return value is a negative lttng_error_code. */
3748 ret = -nr_pids;
3749 goto error;
3750 }
3751
3752 /*
3753 * Setup lttng message with payload size set to the event list size in
3754 * bytes and then copy list into the llm payload.
3755 */
3756 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3757 sizeof(int32_t) * nr_pids);
3758 free(pids);
3759
3760 if (ret < 0) {
3761 goto setup_error;
3762 }
3763
3764 ret = LTTNG_OK;
3765 break;
3766 }
3767 case LTTNG_SET_CONSUMER_URI:
3768 {
3769 size_t nb_uri, len;
3770 struct lttng_uri *uris;
3771
3772 nb_uri = cmd_ctx->lsm->u.uri.size;
3773 len = nb_uri * sizeof(struct lttng_uri);
3774
3775 if (nb_uri == 0) {
3776 ret = LTTNG_ERR_INVALID;
3777 goto error;
3778 }
3779
3780 uris = zmalloc(len);
3781 if (uris == NULL) {
3782 ret = LTTNG_ERR_FATAL;
3783 goto error;
3784 }
3785
3786 /* Receive variable len data */
3787 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3788 ret = lttcomm_recv_unix_sock(sock, uris, len);
3789 if (ret <= 0) {
3790 DBG("No URIs received from client... continuing");
3791 *sock_error = 1;
3792 ret = LTTNG_ERR_SESSION_FAIL;
3793 free(uris);
3794 goto error;
3795 }
3796
3797 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3798 free(uris);
3799 if (ret != LTTNG_OK) {
3800 goto error;
3801 }
3802
3803
3804 break;
3805 }
3806 case LTTNG_START_TRACE:
3807 {
3808 ret = cmd_start_trace(cmd_ctx->session);
3809 break;
3810 }
3811 case LTTNG_STOP_TRACE:
3812 {
3813 ret = cmd_stop_trace(cmd_ctx->session);
3814 break;
3815 }
3816 case LTTNG_CREATE_SESSION:
3817 {
3818 size_t nb_uri, len;
3819 struct lttng_uri *uris = NULL;
3820
3821 nb_uri = cmd_ctx->lsm->u.uri.size;
3822 len = nb_uri * sizeof(struct lttng_uri);
3823
3824 if (nb_uri > 0) {
3825 uris = zmalloc(len);
3826 if (uris == NULL) {
3827 ret = LTTNG_ERR_FATAL;
3828 goto error;
3829 }
3830
3831 /* Receive variable len data */
3832 DBG("Waiting for %zu URIs from client ...", nb_uri);
3833 ret = lttcomm_recv_unix_sock(sock, uris, len);
3834 if (ret <= 0) {
3835 DBG("No URIs received from client... continuing");
3836 *sock_error = 1;
3837 ret = LTTNG_ERR_SESSION_FAIL;
3838 free(uris);
3839 goto error;
3840 }
3841
3842 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3843 DBG("Creating session with ONE network URI is a bad call");
3844 ret = LTTNG_ERR_SESSION_FAIL;
3845 free(uris);
3846 goto error;
3847 }
3848 }
3849
3850 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3851 &cmd_ctx->creds, 0);
3852
3853 free(uris);
3854
3855 break;
3856 }
3857 case LTTNG_DESTROY_SESSION:
3858 {
3859 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3860
3861 /* Set session to NULL so we do not unlock it after free. */
3862 cmd_ctx->session = NULL;
3863 break;
3864 }
3865 case LTTNG_LIST_DOMAINS:
3866 {
3867 ssize_t nb_dom;
3868 struct lttng_domain *domains = NULL;
3869
3870 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3871 if (nb_dom < 0) {
3872 /* Return value is a negative lttng_error_code. */
3873 ret = -nb_dom;
3874 goto error;
3875 }
3876
3877 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3878 nb_dom * sizeof(struct lttng_domain));
3879 free(domains);
3880
3881 if (ret < 0) {
3882 goto setup_error;
3883 }
3884
3885 ret = LTTNG_OK;
3886 break;
3887 }
3888 case LTTNG_LIST_CHANNELS:
3889 {
3890 ssize_t payload_size;
3891 struct lttng_channel *channels = NULL;
3892
3893 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3894 cmd_ctx->session, &channels);
3895 if (payload_size < 0) {
3896 /* Return value is a negative lttng_error_code. */
3897 ret = -payload_size;
3898 goto error;
3899 }
3900
3901 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3902 payload_size);
3903 free(channels);
3904
3905 if (ret < 0) {
3906 goto setup_error;
3907 }
3908
3909 ret = LTTNG_OK;
3910 break;
3911 }
3912 case LTTNG_LIST_EVENTS:
3913 {
3914 ssize_t nb_event;
3915 struct lttng_event *events = NULL;
3916 struct lttcomm_event_command_header cmd_header;
3917 size_t total_size;
3918
3919 memset(&cmd_header, 0, sizeof(cmd_header));
3920 /* Extended infos are included at the end of events */
3921 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3922 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3923 &events, &total_size);
3924
3925 if (nb_event < 0) {
3926 /* Return value is a negative lttng_error_code. */
3927 ret = -nb_event;
3928 goto error;
3929 }
3930
3931 cmd_header.nb_events = nb_event;
3932 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3933 &cmd_header, sizeof(cmd_header));
3934 free(events);
3935
3936 if (ret < 0) {
3937 goto setup_error;
3938 }
3939
3940 ret = LTTNG_OK;
3941 break;
3942 }
3943 case LTTNG_LIST_SESSIONS:
3944 {
3945 unsigned int nr_sessions;
3946 void *sessions_payload;
3947 size_t payload_len;
3948
3949 session_lock_list();
3950 nr_sessions = lttng_sessions_count(
3951 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3952 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3953 payload_len = sizeof(struct lttng_session) * nr_sessions;
3954 sessions_payload = zmalloc(payload_len);
3955
3956 if (!sessions_payload) {
3957 session_unlock_list();
3958 ret = -ENOMEM;
3959 goto setup_error;
3960 }
3961
3962 cmd_list_lttng_sessions(sessions_payload,
3963 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3964 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3965 session_unlock_list();
3966
3967 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3968 payload_len);
3969 free(sessions_payload);
3970
3971 if (ret < 0) {
3972 goto setup_error;
3973 }
3974
3975 ret = LTTNG_OK;
3976 break;
3977 }
3978 case LTTNG_REGISTER_CONSUMER:
3979 {
3980 struct consumer_data *cdata;
3981
3982 switch (cmd_ctx->lsm->domain.type) {
3983 case LTTNG_DOMAIN_KERNEL:
3984 cdata = &kconsumer_data;
3985 break;
3986 default:
3987 ret = LTTNG_ERR_UND;
3988 goto error;
3989 }
3990
3991 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3992 cmd_ctx->lsm->u.reg.path, cdata);
3993 break;
3994 }
3995 case LTTNG_DATA_PENDING:
3996 {
3997 int pending_ret;
3998 uint8_t pending_ret_byte;
3999
4000 pending_ret = cmd_data_pending(cmd_ctx->session);
4001
4002 /*
4003 * FIXME
4004 *
4005 * This function may returns 0 or 1 to indicate whether or not
4006 * there is data pending. In case of error, it should return an
4007 * LTTNG_ERR code. However, some code paths may still return
4008 * a nondescript error code, which we handle by returning an
4009 * "unknown" error.
4010 */
4011 if (pending_ret == 0 || pending_ret == 1) {
4012 /*
4013 * ret will be set to LTTNG_OK at the end of
4014 * this function.
4015 */
4016 } else if (pending_ret < 0) {
4017 ret = LTTNG_ERR_UNK;
4018 goto setup_error;
4019 } else {
4020 ret = pending_ret;
4021 goto setup_error;
4022 }
4023
4024 pending_ret_byte = (uint8_t) pending_ret;
4025
4026 /* 1 byte to return whether or not data is pending */
4027 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
4028 &pending_ret_byte, 1);
4029
4030 if (ret < 0) {
4031 goto setup_error;
4032 }
4033
4034 ret = LTTNG_OK;
4035 break;
4036 }
4037 case LTTNG_SNAPSHOT_ADD_OUTPUT:
4038 {
4039 struct lttcomm_lttng_output_id reply;
4040
4041 ret = cmd_snapshot_add_output(cmd_ctx->session,
4042 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
4043 if (ret != LTTNG_OK) {
4044 goto error;
4045 }
4046
4047 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
4048 sizeof(reply));
4049 if (ret < 0) {
4050 goto setup_error;
4051 }
4052
4053 /* Copy output list into message payload */
4054 ret = LTTNG_OK;
4055 break;
4056 }
4057 case LTTNG_SNAPSHOT_DEL_OUTPUT:
4058 {
4059 ret = cmd_snapshot_del_output(cmd_ctx->session,
4060 &cmd_ctx->lsm->u.snapshot_output.output);
4061 break;
4062 }
4063 case LTTNG_SNAPSHOT_LIST_OUTPUT:
4064 {
4065 ssize_t nb_output;
4066 struct lttng_snapshot_output *outputs = NULL;
4067
4068 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
4069 if (nb_output < 0) {
4070 ret = -nb_output;
4071 goto error;
4072 }
4073
4074 assert((nb_output > 0 && outputs) || nb_output == 0);
4075 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
4076 nb_output * sizeof(struct lttng_snapshot_output));
4077 free(outputs);
4078
4079 if (ret < 0) {
4080 goto setup_error;
4081 }
4082
4083 ret = LTTNG_OK;
4084 break;
4085 }
4086 case LTTNG_SNAPSHOT_RECORD:
4087 {
4088 ret = cmd_snapshot_record(cmd_ctx->session,
4089 &cmd_ctx->lsm->u.snapshot_record.output,
4090 cmd_ctx->lsm->u.snapshot_record.wait);
4091 break;
4092 }
4093 case LTTNG_CREATE_SESSION_SNAPSHOT:
4094 {
4095 size_t nb_uri, len;
4096 struct lttng_uri *uris = NULL;
4097
4098 nb_uri = cmd_ctx->lsm->u.uri.size;
4099 len = nb_uri * sizeof(struct lttng_uri);
4100
4101 if (nb_uri > 0) {
4102 uris = zmalloc(len);
4103 if (uris == NULL) {
4104 ret = LTTNG_ERR_FATAL;
4105 goto error;
4106 }
4107
4108 /* Receive variable len data */
4109 DBG("Waiting for %zu URIs from client ...", nb_uri);
4110 ret = lttcomm_recv_unix_sock(sock, uris, len);
4111 if (ret <= 0) {
4112 DBG("No URIs received from client... continuing");
4113 *sock_error = 1;
4114 ret = LTTNG_ERR_SESSION_FAIL;
4115 free(uris);
4116 goto error;
4117 }
4118
4119 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4120 DBG("Creating session with ONE network URI is a bad call");
4121 ret = LTTNG_ERR_SESSION_FAIL;
4122 free(uris);
4123 goto error;
4124 }
4125 }
4126
4127 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4128 nb_uri, &cmd_ctx->creds);
4129 free(uris);
4130 break;
4131 }
4132 case LTTNG_CREATE_SESSION_LIVE:
4133 {
4134 size_t nb_uri, len;
4135 struct lttng_uri *uris = NULL;
4136
4137 nb_uri = cmd_ctx->lsm->u.uri.size;
4138 len = nb_uri * sizeof(struct lttng_uri);
4139
4140 if (nb_uri > 0) {
4141 uris = zmalloc(len);
4142 if (uris == NULL) {
4143 ret = LTTNG_ERR_FATAL;
4144 goto error;
4145 }
4146
4147 /* Receive variable len data */
4148 DBG("Waiting for %zu URIs from client ...", nb_uri);
4149 ret = lttcomm_recv_unix_sock(sock, uris, len);
4150 if (ret <= 0) {
4151 DBG("No URIs received from client... continuing");
4152 *sock_error = 1;
4153 ret = LTTNG_ERR_SESSION_FAIL;
4154 free(uris);
4155 goto error;
4156 }
4157
4158 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4159 DBG("Creating session with ONE network URI is a bad call");
4160 ret = LTTNG_ERR_SESSION_FAIL;
4161 free(uris);
4162 goto error;
4163 }
4164 }
4165
4166 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4167 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4168 free(uris);
4169 break;
4170 }
4171 case LTTNG_SAVE_SESSION:
4172 {
4173 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4174 &cmd_ctx->creds);
4175 break;
4176 }
4177 case LTTNG_SET_SESSION_SHM_PATH:
4178 {
4179 ret = cmd_set_session_shm_path(cmd_ctx->session,
4180 cmd_ctx->lsm->u.set_shm_path.shm_path);
4181 break;
4182 }
4183 case LTTNG_REGENERATE_METADATA:
4184 {
4185 ret = cmd_regenerate_metadata(cmd_ctx->session);
4186 break;
4187 }
4188 case LTTNG_REGENERATE_STATEDUMP:
4189 {
4190 ret = cmd_regenerate_statedump(cmd_ctx->session);
4191 break;
4192 }
4193 case LTTNG_REGISTER_TRIGGER:
4194 {
4195 ret = cmd_register_trigger(cmd_ctx, sock,
4196 notification_thread_handle);
4197 break;
4198 }
4199 case LTTNG_UNREGISTER_TRIGGER:
4200 {
4201 ret = cmd_unregister_trigger(cmd_ctx, sock,
4202 notification_thread_handle);
4203 break;
4204 }
4205 default:
4206 ret = LTTNG_ERR_UND;
4207 break;
4208 }
4209
4210 error:
4211 if (cmd_ctx->llm == NULL) {
4212 DBG("Missing llm structure. Allocating one.");
4213 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4214 goto setup_error;
4215 }
4216 }
4217 /* Set return code */
4218 cmd_ctx->llm->ret_code = ret;
4219 setup_error:
4220 if (cmd_ctx->session) {
4221 session_unlock(cmd_ctx->session);
4222 }
4223 if (need_tracing_session) {
4224 session_unlock_list();
4225 }
4226 init_setup_error:
4227 assert(!rcu_read_ongoing());
4228 return ret;
4229 }
4230
4231 /*
4232 * Thread managing health check socket.
4233 */
4234 static void *thread_manage_health(void *data)
4235 {
4236 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4237 uint32_t revents, nb_fd;
4238 struct lttng_poll_event events;
4239 struct health_comm_msg msg;
4240 struct health_comm_reply reply;
4241
4242 DBG("[thread] Manage health check started");
4243
4244 rcu_register_thread();
4245
4246 /* We might hit an error path before this is created. */
4247 lttng_poll_init(&events);
4248
4249 /* Create unix socket */
4250 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4251 if (sock < 0) {
4252 ERR("Unable to create health check Unix socket");
4253 goto error;
4254 }
4255
4256 if (is_root) {
4257 /* lttng health client socket path permissions */
4258 ret = chown(health_unix_sock_path, 0,
4259 utils_get_group_id(tracing_group_name));
4260 if (ret < 0) {
4261 ERR("Unable to set group on %s", health_unix_sock_path);
4262 PERROR("chown");
4263 goto error;
4264 }
4265
4266 ret = chmod(health_unix_sock_path,
4267 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4268 if (ret < 0) {
4269 ERR("Unable to set permissions on %s", health_unix_sock_path);
4270 PERROR("chmod");
4271 goto error;
4272 }
4273 }
4274
4275 /*
4276 * Set the CLOEXEC flag. Return code is useless because either way, the
4277 * show must go on.
4278 */
4279 (void) utils_set_fd_cloexec(sock);
4280
4281 ret = lttcomm_listen_unix_sock(sock);
4282 if (ret < 0) {
4283 goto error;
4284 }
4285
4286 /*
4287 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4288 * more will be added to this poll set.
4289 */
4290 ret = sessiond_set_thread_pollset(&events, 2);
4291 if (ret < 0) {
4292 goto error;
4293 }
4294
4295 /* Add the application registration socket */
4296 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4297 if (ret < 0) {
4298 goto error;
4299 }
4300
4301 sessiond_notify_ready();
4302
4303 while (1) {
4304 DBG("Health check ready");
4305
4306 /* Inifinite blocking call, waiting for transmission */
4307 restart:
4308 ret = lttng_poll_wait(&events, -1);
4309 if (ret < 0) {
4310 /*
4311 * Restart interrupted system call.
4312 */
4313 if (errno == EINTR) {
4314 goto restart;
4315 }
4316 goto error;
4317 }
4318
4319 nb_fd = ret;
4320
4321 for (i = 0; i < nb_fd; i++) {
4322 /* Fetch once the poll data */
4323 revents = LTTNG_POLL_GETEV(&events, i);
4324 pollfd = LTTNG_POLL_GETFD(&events, i);
4325
4326 if (!revents) {
4327 /* No activity for this FD (poll implementation). */
4328 continue;
4329 }
4330
4331 /* Thread quit pipe has been closed. Killing thread. */
4332 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4333 if (ret) {
4334 err = 0;
4335 goto exit;
4336 }
4337
4338 /* Event on the registration socket */
4339 if (pollfd == sock) {
4340 if (revents & LPOLLIN) {
4341 continue;
4342 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4343 ERR("Health socket poll error");
4344 goto error;
4345 } else {
4346 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4347 goto error;
4348 }
4349 }
4350 }
4351
4352 new_sock = lttcomm_accept_unix_sock(sock);
4353 if (new_sock < 0) {
4354 goto error;
4355 }
4356
4357 /*
4358 * Set the CLOEXEC flag. Return code is useless because either way, the
4359 * show must go on.
4360 */
4361 (void) utils_set_fd_cloexec(new_sock);
4362
4363 DBG("Receiving data from client for health...");
4364 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4365 if (ret <= 0) {
4366 DBG("Nothing recv() from client... continuing");
4367 ret = close(new_sock);
4368 if (ret) {
4369 PERROR("close");
4370 }
4371 continue;
4372 }
4373
4374 rcu_thread_online();
4375
4376 memset(&reply, 0, sizeof(reply));
4377 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4378 /*
4379 * health_check_state returns 0 if health is
4380 * bad.
4381 */
4382 if (!health_check_state(health_sessiond, i)) {
4383 reply.ret_code |= 1ULL << i;
4384 }
4385 }
4386
4387 DBG2("Health check return value %" PRIx64, reply.ret_code);
4388
4389 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4390 if (ret < 0) {
4391 ERR("Failed to send health data back to client");
4392 }
4393
4394 /* End of transmission */
4395 ret = close(new_sock);
4396 if (ret) {
4397 PERROR("close");
4398 }
4399 }
4400
4401 exit:
4402 error:
4403 if (err) {
4404 ERR("Health error occurred in %s", __func__);
4405 }
4406 DBG("Health check thread dying");
4407 unlink(health_unix_sock_path);
4408 if (sock >= 0) {
4409 ret = close(sock);
4410 if (ret) {
4411 PERROR("close");
4412 }
4413 }
4414
4415 lttng_poll_clean(&events);
4416 stop_threads();
4417 rcu_unregister_thread();
4418 return NULL;
4419 }
4420
4421 /*
4422 * This thread manage all clients request using the unix client socket for
4423 * communication.
4424 */
4425 static void *thread_manage_clients(void *data)
4426 {
4427 int sock = -1, ret, i, pollfd, err = -1;
4428 int sock_error;
4429 uint32_t revents, nb_fd;
4430 struct command_ctx *cmd_ctx = NULL;
4431 struct lttng_poll_event events;
4432
4433 DBG("[thread] Manage client started");
4434
4435 rcu_register_thread();
4436
4437 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4438
4439 health_code_update();
4440
4441 ret = lttcomm_listen_unix_sock(client_sock);
4442 if (ret < 0) {
4443 goto error_listen;
4444 }
4445
4446 /*
4447 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4448 * more will be added to this poll set.
4449 */
4450 ret = sessiond_set_thread_pollset(&events, 2);
4451 if (ret < 0) {
4452 goto error_create_poll;
4453 }
4454
4455 /* Add the application registration socket */
4456 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4457 if (ret < 0) {
4458 goto error;
4459 }
4460
4461 sessiond_notify_ready();
4462 ret = sem_post(&load_info->message_thread_ready);
4463 if (ret) {
4464 PERROR("sem_post message_thread_ready");
4465 goto error;
4466 }
4467
4468 /* This testpoint is after we signal readiness to the parent. */
4469 if (testpoint(sessiond_thread_manage_clients)) {
4470 goto error;
4471 }
4472
4473 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4474 goto error;
4475 }
4476
4477 health_code_update();
4478
4479 while (1) {
4480 DBG("Accepting client command ...");
4481
4482 /* Inifinite blocking call, waiting for transmission */
4483 restart:
4484 health_poll_entry();
4485 ret = lttng_poll_wait(&events, -1);
4486 health_poll_exit();
4487 if (ret < 0) {
4488 /*
4489 * Restart interrupted system call.
4490 */
4491 if (errno == EINTR) {
4492 goto restart;
4493 }
4494 goto error;
4495 }
4496
4497 nb_fd = ret;
4498
4499 for (i = 0; i < nb_fd; i++) {
4500 /* Fetch once the poll data */
4501 revents = LTTNG_POLL_GETEV(&events, i);
4502 pollfd = LTTNG_POLL_GETFD(&events, i);
4503
4504 health_code_update();
4505
4506 if (!revents) {
4507 /* No activity for this FD (poll implementation). */
4508 continue;
4509 }
4510
4511 /* Thread quit pipe has been closed. Killing thread. */
4512 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4513 if (ret) {
4514 err = 0;
4515 goto exit;
4516 }
4517
4518 /* Event on the registration socket */
4519 if (pollfd == client_sock) {
4520 if (revents & LPOLLIN) {
4521 continue;
4522 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4523 ERR("Client socket poll error");
4524 goto error;
4525 } else {
4526 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4527 goto error;
4528 }
4529 }
4530 }
4531
4532 DBG("Wait for client response");
4533
4534 health_code_update();
4535
4536 sock = lttcomm_accept_unix_sock(client_sock);
4537 if (sock < 0) {
4538 goto error;
4539 }
4540
4541 /*
4542 * Set the CLOEXEC flag. Return code is useless because either way, the
4543 * show must go on.
4544 */
4545 (void) utils_set_fd_cloexec(sock);
4546
4547 /* Set socket option for credentials retrieval */
4548 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4549 if (ret < 0) {
4550 goto error;
4551 }
4552
4553 /* Allocate context command to process the client request */
4554 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4555 if (cmd_ctx == NULL) {
4556 PERROR("zmalloc cmd_ctx");
4557 goto error;
4558 }
4559
4560 /* Allocate data buffer for reception */
4561 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4562 if (cmd_ctx->lsm == NULL) {
4563 PERROR("zmalloc cmd_ctx->lsm");
4564 goto error;
4565 }
4566
4567 cmd_ctx->llm = NULL;
4568 cmd_ctx->session = NULL;
4569
4570 health_code_update();
4571
4572 /*
4573 * Data is received from the lttng client. The struct
4574 * lttcomm_session_msg (lsm) contains the command and data request of
4575 * the client.
4576 */
4577 DBG("Receiving data from client ...");
4578 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4579 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4580 if (ret <= 0) {
4581 DBG("Nothing recv() from client... continuing");
4582 ret = close(sock);
4583 if (ret) {
4584 PERROR("close");
4585 }
4586 sock = -1;
4587 clean_command_ctx(&cmd_ctx);
4588 continue;
4589 }
4590
4591 health_code_update();
4592
4593 // TODO: Validate cmd_ctx including sanity check for
4594 // security purpose.
4595
4596 rcu_thread_online();
4597 /*
4598 * This function dispatch the work to the kernel or userspace tracer
4599 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4600 * informations for the client. The command context struct contains
4601 * everything this function may needs.
4602 */
4603 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4604 rcu_thread_offline();
4605 if (ret < 0) {
4606 ret = close(sock);
4607 if (ret) {
4608 PERROR("close");
4609 }
4610 sock = -1;
4611 /*
4612 * TODO: Inform client somehow of the fatal error. At
4613 * this point, ret < 0 means that a zmalloc failed
4614 * (ENOMEM). Error detected but still accept
4615 * command, unless a socket error has been
4616 * detected.
4617 */
4618 clean_command_ctx(&cmd_ctx);
4619 continue;
4620 }
4621
4622 health_code_update();
4623
4624 DBG("Sending response (size: %d, retcode: %s (%d))",
4625 cmd_ctx->lttng_msg_size,
4626 lttng_strerror(-cmd_ctx->llm->ret_code),
4627 cmd_ctx->llm->ret_code);
4628 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4629 if (ret < 0) {
4630 ERR("Failed to send data back to client");
4631 }
4632
4633 /* End of transmission */
4634 ret = close(sock);
4635 if (ret) {
4636 PERROR("close");
4637 }
4638 sock = -1;
4639
4640 clean_command_ctx(&cmd_ctx);
4641
4642 health_code_update();
4643 }
4644
4645 exit:
4646 error:
4647 if (sock >= 0) {
4648 ret = close(sock);
4649 if (ret) {
4650 PERROR("close");
4651 }
4652 }
4653
4654 lttng_poll_clean(&events);
4655 clean_command_ctx(&cmd_ctx);
4656
4657 error_listen:
4658 error_create_poll:
4659 unlink(client_unix_sock_path);
4660 if (client_sock >= 0) {
4661 ret = close(client_sock);
4662 if (ret) {
4663 PERROR("close");
4664 }
4665 }
4666
4667 if (err) {
4668 health_error();
4669 ERR("Health error occurred in %s", __func__);
4670 }
4671
4672 health_unregister(health_sessiond);
4673
4674 DBG("Client thread dying");
4675
4676 rcu_unregister_thread();
4677
4678 /*
4679 * Since we are creating the consumer threads, we own them, so we need
4680 * to join them before our thread exits.
4681 */
4682 ret = join_consumer_thread(&kconsumer_data);
4683 if (ret) {
4684 errno = ret;
4685 PERROR("join_consumer");
4686 }
4687
4688 ret = join_consumer_thread(&ustconsumer32_data);
4689 if (ret) {
4690 errno = ret;
4691 PERROR("join_consumer ust32");
4692 }
4693
4694 ret = join_consumer_thread(&ustconsumer64_data);
4695 if (ret) {
4696 errno = ret;
4697 PERROR("join_consumer ust64");
4698 }
4699 return NULL;
4700 }
4701
4702 static int string_match(const char *str1, const char *str2)
4703 {
4704 return (str1 && str2) && !strcmp(str1, str2);
4705 }
4706
4707 /*
4708 * Take an option from the getopt output and set it in the right variable to be
4709 * used later.
4710 *
4711 * Return 0 on success else a negative value.
4712 */
4713 static int set_option(int opt, const char *arg, const char *optname)
4714 {
4715 int ret = 0;
4716
4717 if (string_match(optname, "client-sock") || opt == 'c') {
4718 if (!arg || *arg == '\0') {
4719 ret = -EINVAL;
4720 goto end;
4721 }
4722 if (lttng_is_setuid_setgid()) {
4723 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4724 "-c, --client-sock");
4725 } else {
4726 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4727 }
4728 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4729 if (!arg || *arg == '\0') {
4730 ret = -EINVAL;
4731 goto end;
4732 }
4733 if (lttng_is_setuid_setgid()) {
4734 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4735 "-a, --apps-sock");
4736 } else {
4737 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4738 }
4739 } else if (string_match(optname, "daemonize") || opt == 'd') {
4740 opt_daemon = 1;
4741 } else if (string_match(optname, "background") || opt == 'b') {
4742 opt_background = 1;
4743 } else if (string_match(optname, "group") || opt == 'g') {
4744 if (!arg || *arg == '\0') {
4745 ret = -EINVAL;
4746 goto end;
4747 }
4748 if (lttng_is_setuid_setgid()) {
4749 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4750 "-g, --group");
4751 } else {
4752 /*
4753 * If the override option is set, the pointer points to a
4754 * *non* const thus freeing it even though the variable type is
4755 * set to const.
4756 */
4757 if (tracing_group_name_override) {
4758 free((void *) tracing_group_name);
4759 }
4760 tracing_group_name = strdup(arg);
4761 if (!tracing_group_name) {
4762 PERROR("strdup");
4763 ret = -ENOMEM;
4764 }
4765 tracing_group_name_override = 1;
4766 }
4767 } else if (string_match(optname, "help") || opt == 'h') {
4768 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4769 if (ret) {
4770 ERR("Cannot show --help for `lttng-sessiond`");
4771 perror("exec");
4772 }
4773 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4774 } else if (string_match(optname, "version") || opt == 'V') {
4775 fprintf(stdout, "%s\n", VERSION);
4776 exit(EXIT_SUCCESS);
4777 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4778 opt_sig_parent = 1;
4779 } else if (string_match(optname, "kconsumerd-err-sock")) {
4780 if (!arg || *arg == '\0') {
4781 ret = -EINVAL;
4782 goto end;
4783 }
4784 if (lttng_is_setuid_setgid()) {
4785 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4786 "--kconsumerd-err-sock");
4787 } else {
4788 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4789 }
4790 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4791 if (!arg || *arg == '\0') {
4792 ret = -EINVAL;
4793 goto end;
4794 }
4795 if (lttng_is_setuid_setgid()) {
4796 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4797 "--kconsumerd-cmd-sock");
4798 } else {
4799 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4800 }
4801 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4802 if (!arg || *arg == '\0') {
4803 ret = -EINVAL;
4804 goto end;
4805 }
4806 if (lttng_is_setuid_setgid()) {
4807 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4808 "--ustconsumerd64-err-sock");
4809 } else {
4810 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4811 }
4812 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4813 if (!arg || *arg == '\0') {
4814 ret = -EINVAL;
4815 goto end;
4816 }
4817 if (lttng_is_setuid_setgid()) {
4818 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4819 "--ustconsumerd64-cmd-sock");
4820 } else {
4821 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4822 }
4823 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4824 if (!arg || *arg == '\0') {
4825 ret = -EINVAL;
4826 goto end;
4827 }
4828 if (lttng_is_setuid_setgid()) {
4829 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4830 "--ustconsumerd32-err-sock");
4831 } else {
4832 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4833 }
4834 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4835 if (!arg || *arg == '\0') {
4836 ret = -EINVAL;
4837 goto end;
4838 }
4839 if (lttng_is_setuid_setgid()) {
4840 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4841 "--ustconsumerd32-cmd-sock");
4842 } else {
4843 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4844 }
4845 } else if (string_match(optname, "no-kernel")) {
4846 opt_no_kernel = 1;
4847 } else if (string_match(optname, "quiet") || opt == 'q') {
4848 lttng_opt_quiet = 1;
4849 } else if (string_match(optname, "verbose") || opt == 'v') {
4850 /* Verbose level can increase using multiple -v */
4851 if (arg) {
4852 /* Value obtained from config file */
4853 lttng_opt_verbose = config_parse_value(arg);
4854 } else {
4855 /* -v used on command line */
4856 lttng_opt_verbose++;
4857 }
4858 /* Clamp value to [0, 3] */
4859 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4860 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4861 } else if (string_match(optname, "verbose-consumer")) {
4862 if (arg) {
4863 opt_verbose_consumer = config_parse_value(arg);
4864 } else {
4865 opt_verbose_consumer++;
4866 }
4867 } else if (string_match(optname, "consumerd32-path")) {
4868 if (!arg || *arg == '\0') {
4869 ret = -EINVAL;
4870 goto end;
4871 }
4872 if (lttng_is_setuid_setgid()) {
4873 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4874 "--consumerd32-path");
4875 } else {
4876 if (consumerd32_bin_override) {
4877 free((void *) consumerd32_bin);
4878 }
4879 consumerd32_bin = strdup(arg);
4880 if (!consumerd32_bin) {
4881 PERROR("strdup");
4882 ret = -ENOMEM;
4883 }
4884 consumerd32_bin_override = 1;
4885 }
4886 } else if (string_match(optname, "consumerd32-libdir")) {
4887 if (!arg || *arg == '\0') {
4888 ret = -EINVAL;
4889 goto end;
4890 }
4891 if (lttng_is_setuid_setgid()) {
4892 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4893 "--consumerd32-libdir");
4894 } else {
4895 if (consumerd32_libdir_override) {
4896 free((void *) consumerd32_libdir);
4897 }
4898 consumerd32_libdir = strdup(arg);
4899 if (!consumerd32_libdir) {
4900 PERROR("strdup");
4901 ret = -ENOMEM;
4902 }
4903 consumerd32_libdir_override = 1;
4904 }
4905 } else if (string_match(optname, "consumerd64-path")) {
4906 if (!arg || *arg == '\0') {
4907 ret = -EINVAL;
4908 goto end;
4909 }
4910 if (lttng_is_setuid_setgid()) {
4911 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4912 "--consumerd64-path");
4913 } else {
4914 if (consumerd64_bin_override) {
4915 free((void *) consumerd64_bin);
4916 }
4917 consumerd64_bin = strdup(arg);
4918 if (!consumerd64_bin) {
4919 PERROR("strdup");
4920 ret = -ENOMEM;
4921 }
4922 consumerd64_bin_override = 1;
4923 }
4924 } else if (string_match(optname, "consumerd64-libdir")) {
4925 if (!arg || *arg == '\0') {
4926 ret = -EINVAL;
4927 goto end;
4928 }
4929 if (lttng_is_setuid_setgid()) {
4930 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4931 "--consumerd64-libdir");
4932 } else {
4933 if (consumerd64_libdir_override) {
4934 free((void *) consumerd64_libdir);
4935 }
4936 consumerd64_libdir = strdup(arg);
4937 if (!consumerd64_libdir) {
4938 PERROR("strdup");
4939 ret = -ENOMEM;
4940 }
4941 consumerd64_libdir_override = 1;
4942 }
4943 } else if (string_match(optname, "pidfile") || opt == 'p') {
4944 if (!arg || *arg == '\0') {
4945 ret = -EINVAL;
4946 goto end;
4947 }
4948 if (lttng_is_setuid_setgid()) {
4949 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4950 "-p, --pidfile");
4951 } else {
4952 free(opt_pidfile);
4953 opt_pidfile = strdup(arg);
4954 if (!opt_pidfile) {
4955 PERROR("strdup");
4956 ret = -ENOMEM;
4957 }
4958 }
4959 } else if (string_match(optname, "agent-tcp-port")) {
4960 if (!arg || *arg == '\0') {
4961 ret = -EINVAL;
4962 goto end;
4963 }
4964 if (lttng_is_setuid_setgid()) {
4965 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4966 "--agent-tcp-port");
4967 } else {
4968 unsigned long v;
4969
4970 errno = 0;
4971 v = strtoul(arg, NULL, 0);
4972 if (errno != 0 || !isdigit(arg[0])) {
4973 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4974 return -1;
4975 }
4976 if (v == 0 || v >= 65535) {
4977 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4978 return -1;
4979 }
4980 agent_tcp_port = (uint32_t) v;
4981 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4982 }
4983 } else if (string_match(optname, "load") || opt == 'l') {
4984 if (!arg || *arg == '\0') {
4985 ret = -EINVAL;
4986 goto end;
4987 }
4988 if (lttng_is_setuid_setgid()) {
4989 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4990 "-l, --load");
4991 } else {
4992 free(opt_load_session_path);
4993 opt_load_session_path = strdup(arg);
4994 if (!opt_load_session_path) {
4995 PERROR("strdup");
4996 ret = -ENOMEM;
4997 }
4998 }
4999 } else if (string_match(optname, "kmod-probes")) {
5000 if (!arg || *arg == '\0') {
5001 ret = -EINVAL;
5002 goto end;
5003 }
5004 if (lttng_is_setuid_setgid()) {
5005 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5006 "--kmod-probes");
5007 } else {
5008 free(kmod_probes_list);
5009 kmod_probes_list = strdup(arg);
5010 if (!kmod_probes_list) {
5011 PERROR("strdup");
5012 ret = -ENOMEM;
5013 }
5014 }
5015 } else if (string_match(optname, "extra-kmod-probes")) {
5016 if (!arg || *arg == '\0') {
5017 ret = -EINVAL;
5018 goto end;
5019 }
5020 if (lttng_is_setuid_setgid()) {
5021 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5022 "--extra-kmod-probes");
5023 } else {
5024 free(kmod_extra_probes_list);
5025 kmod_extra_probes_list = strdup(arg);
5026 if (!kmod_extra_probes_list) {
5027 PERROR("strdup");
5028 ret = -ENOMEM;
5029 }
5030 }
5031 } else if (string_match(optname, "config") || opt == 'f') {
5032 /* This is handled in set_options() thus silent skip. */
5033 goto end;
5034 } else {
5035 /* Unknown option or other error.
5036 * Error is printed by getopt, just return */
5037 ret = -1;
5038 }
5039
5040 end:
5041 if (ret == -EINVAL) {
5042 const char *opt_name = "unknown";
5043 int i;
5044
5045 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
5046 i++) {
5047 if (opt == long_options[i].val) {
5048 opt_name = long_options[i].name;
5049 break;
5050 }
5051 }
5052
5053 WARN("Invalid argument provided for option \"%s\", using default value.",
5054 opt_name);
5055 }
5056
5057 return ret;
5058 }
5059
5060 /*
5061 * config_entry_handler_cb used to handle options read from a config file.
5062 * See config_entry_handler_cb comment in common/config/session-config.h for the
5063 * return value conventions.
5064 */
5065 static int config_entry_handler(const struct config_entry *entry, void *unused)
5066 {
5067 int ret = 0, i;
5068
5069 if (!entry || !entry->name || !entry->value) {
5070 ret = -EINVAL;
5071 goto end;
5072 }
5073
5074 /* Check if the option is to be ignored */
5075 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5076 if (!strcmp(entry->name, config_ignore_options[i])) {
5077 goto end;
5078 }
5079 }
5080
5081 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5082 i++) {
5083
5084 /* Ignore if not fully matched. */
5085 if (strcmp(entry->name, long_options[i].name)) {
5086 continue;
5087 }
5088
5089 /*
5090 * If the option takes no argument on the command line, we have to
5091 * check if the value is "true". We support non-zero numeric values,
5092 * true, on and yes.
5093 */
5094 if (!long_options[i].has_arg) {
5095 ret = config_parse_value(entry->value);
5096 if (ret <= 0) {
5097 if (ret) {
5098 WARN("Invalid configuration value \"%s\" for option %s",
5099 entry->value, entry->name);
5100 }
5101 /* False, skip boolean config option. */
5102 goto end;
5103 }
5104 }
5105
5106 ret = set_option(long_options[i].val, entry->value, entry->name);
5107 goto end;
5108 }
5109
5110 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5111
5112 end:
5113 return ret;
5114 }
5115
5116 /*
5117 * daemon configuration loading and argument parsing
5118 */
5119 static int set_options(int argc, char **argv)
5120 {
5121 int ret = 0, c = 0, option_index = 0;
5122 int orig_optopt = optopt, orig_optind = optind;
5123 char *optstring;
5124 const char *config_path = NULL;
5125
5126 optstring = utils_generate_optstring(long_options,
5127 sizeof(long_options) / sizeof(struct option));
5128 if (!optstring) {
5129 ret = -ENOMEM;
5130 goto end;
5131 }
5132
5133 /* Check for the --config option */
5134 while ((c = getopt_long(argc, argv, optstring, long_options,
5135 &option_index)) != -1) {
5136 if (c == '?') {
5137 ret = -EINVAL;
5138 goto end;
5139 } else if (c != 'f') {
5140 /* if not equal to --config option. */
5141 continue;
5142 }
5143
5144 if (lttng_is_setuid_setgid()) {
5145 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5146 "-f, --config");
5147 } else {
5148 config_path = utils_expand_path(optarg);
5149 if (!config_path) {
5150 ERR("Failed to resolve path: %s", optarg);
5151 }
5152 }
5153 }
5154
5155 ret = config_get_section_entries(config_path, config_section_name,
5156 config_entry_handler, NULL);
5157 if (ret) {
5158 if (ret > 0) {
5159 ERR("Invalid configuration option at line %i", ret);
5160 ret = -1;
5161 }
5162 goto end;
5163 }
5164
5165 /* Reset getopt's global state */
5166 optopt = orig_optopt;
5167 optind = orig_optind;
5168 while (1) {
5169 option_index = -1;
5170 /*
5171 * getopt_long() will not set option_index if it encounters a
5172 * short option.
5173 */
5174 c = getopt_long(argc, argv, optstring, long_options,
5175 &option_index);
5176 if (c == -1) {
5177 break;
5178 }
5179
5180 /*
5181 * Pass NULL as the long option name if popt left the index
5182 * unset.
5183 */
5184 ret = set_option(c, optarg,
5185 option_index < 0 ? NULL :
5186 long_options[option_index].name);
5187 if (ret < 0) {
5188 break;
5189 }
5190 }
5191
5192 end:
5193 free(optstring);
5194 return ret;
5195 }
5196
5197 /*
5198 * Creates the two needed socket by the daemon.
5199 * apps_sock - The communication socket for all UST apps.
5200 * client_sock - The communication of the cli tool (lttng).
5201 */
5202 static int init_daemon_socket(void)
5203 {
5204 int ret = 0;
5205 mode_t old_umask;
5206
5207 old_umask = umask(0);
5208
5209 /* Create client tool unix socket */
5210 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5211 if (client_sock < 0) {
5212 ERR("Create unix sock failed: %s", client_unix_sock_path);
5213 ret = -1;
5214 goto end;
5215 }
5216
5217 /* Set the cloexec flag */
5218 ret = utils_set_fd_cloexec(client_sock);
5219 if (ret < 0) {
5220 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5221 "Continuing but note that the consumer daemon will have a "
5222 "reference to this socket on exec()", client_sock);
5223 }
5224
5225 /* File permission MUST be 660 */
5226 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5227 if (ret < 0) {
5228 ERR("Set file permissions failed: %s", client_unix_sock_path);
5229 PERROR("chmod");
5230 goto end;
5231 }
5232
5233 /* Create the application unix socket */
5234 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5235 if (apps_sock < 0) {
5236 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5237 ret = -1;
5238 goto end;
5239 }
5240
5241 /* Set the cloexec flag */
5242 ret = utils_set_fd_cloexec(apps_sock);
5243 if (ret < 0) {
5244 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5245 "Continuing but note that the consumer daemon will have a "
5246 "reference to this socket on exec()", apps_sock);
5247 }
5248
5249 /* File permission MUST be 666 */
5250 ret = chmod(apps_unix_sock_path,
5251 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5252 if (ret < 0) {
5253 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5254 PERROR("chmod");
5255 goto end;
5256 }
5257
5258 DBG3("Session daemon client socket %d and application socket %d created",
5259 client_sock, apps_sock);
5260
5261 end:
5262 umask(old_umask);
5263 return ret;
5264 }
5265
5266 /*
5267 * Check if the global socket is available, and if a daemon is answering at the
5268 * other side. If yes, error is returned.
5269 */
5270 static int check_existing_daemon(void)
5271 {
5272 /* Is there anybody out there ? */
5273 if (lttng_session_daemon_alive()) {
5274 return -EEXIST;
5275 }
5276
5277 return 0;
5278 }
5279
5280 /*
5281 * Set the tracing group gid onto the client socket.
5282 *
5283 * Race window between mkdir and chown is OK because we are going from more
5284 * permissive (root.root) to less permissive (root.tracing).
5285 */
5286 static int set_permissions(char *rundir)
5287 {
5288 int ret;
5289 gid_t gid;
5290
5291 gid = utils_get_group_id(tracing_group_name);
5292
5293 /* Set lttng run dir */
5294 ret = chown(rundir, 0, gid);
5295 if (ret < 0) {
5296 ERR("Unable to set group on %s", rundir);
5297 PERROR("chown");
5298 }
5299
5300 /*
5301 * Ensure all applications and tracing group can search the run
5302 * dir. Allow everyone to read the directory, since it does not
5303 * buy us anything to hide its content.
5304 */
5305 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5306 if (ret < 0) {
5307 ERR("Unable to set permissions on %s", rundir);
5308 PERROR("chmod");
5309 }
5310
5311 /* lttng client socket path */
5312 ret = chown(client_unix_sock_path, 0, gid);
5313 if (ret < 0) {
5314 ERR("Unable to set group on %s", client_unix_sock_path);
5315 PERROR("chown");
5316 }
5317
5318 /* kconsumer error socket path */
5319 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5320 if (ret < 0) {
5321 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5322 PERROR("chown");
5323 }
5324
5325 /* 64-bit ustconsumer error socket path */
5326 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5327 if (ret < 0) {
5328 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5329 PERROR("chown");
5330 }
5331
5332 /* 32-bit ustconsumer compat32 error socket path */
5333 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5334 if (ret < 0) {
5335 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5336 PERROR("chown");
5337 }
5338
5339 DBG("All permissions are set");
5340
5341 return ret;
5342 }
5343
5344 /*
5345 * Create the lttng run directory needed for all global sockets and pipe.
5346 */
5347 static int create_lttng_rundir(const char *rundir)
5348 {
5349 int ret;
5350
5351 DBG3("Creating LTTng run directory: %s", rundir);
5352
5353 ret = mkdir(rundir, S_IRWXU);
5354 if (ret < 0) {
5355 if (errno != EEXIST) {
5356 ERR("Unable to create %s", rundir);
5357 goto error;
5358 } else {
5359 ret = 0;
5360 }
5361 }
5362
5363 error:
5364 return ret;
5365 }
5366
5367 /*
5368 * Setup sockets and directory needed by the kconsumerd communication with the
5369 * session daemon.
5370 */
5371 static int set_consumer_sockets(struct consumer_data *consumer_data,
5372 const char *rundir)
5373 {
5374 int ret;
5375 char path[PATH_MAX];
5376
5377 switch (consumer_data->type) {
5378 case LTTNG_CONSUMER_KERNEL:
5379 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5380 break;
5381 case LTTNG_CONSUMER64_UST:
5382 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5383 break;
5384 case LTTNG_CONSUMER32_UST:
5385 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5386 break;
5387 default:
5388 ERR("Consumer type unknown");
5389 ret = -EINVAL;
5390 goto error;
5391 }
5392
5393 DBG2("Creating consumer directory: %s", path);
5394
5395 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5396 if (ret < 0) {
5397 if (errno != EEXIST) {
5398 PERROR("mkdir");
5399 ERR("Failed to create %s", path);
5400 goto error;
5401 }
5402 ret = -1;
5403 }
5404 if (is_root) {
5405 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5406 if (ret < 0) {
5407 ERR("Unable to set group on %s", path);
5408 PERROR("chown");
5409 goto error;
5410 }
5411 }
5412
5413 /* Create the kconsumerd error unix socket */
5414 consumer_data->err_sock =
5415 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5416 if (consumer_data->err_sock < 0) {
5417 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5418 ret = -1;
5419 goto error;
5420 }
5421
5422 /*
5423 * Set the CLOEXEC flag. Return code is useless because either way, the
5424 * show must go on.
5425 */
5426 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5427 if (ret < 0) {
5428 PERROR("utils_set_fd_cloexec");
5429 /* continue anyway */
5430 }
5431
5432 /* File permission MUST be 660 */
5433 ret = chmod(consumer_data->err_unix_sock_path,
5434 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5435 if (ret < 0) {
5436 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5437 PERROR("chmod");
5438 goto error;
5439 }
5440
5441 error:
5442 return ret;
5443 }
5444
5445 /*
5446 * Signal handler for the daemon
5447 *
5448 * Simply stop all worker threads, leaving main() return gracefully after
5449 * joining all threads and calling cleanup().
5450 */
5451 static void sighandler(int sig)
5452 {
5453 switch (sig) {
5454 case SIGINT:
5455 DBG("SIGINT caught");
5456 stop_threads();
5457 break;
5458 case SIGTERM:
5459 DBG("SIGTERM caught");
5460 stop_threads();
5461 break;
5462 case SIGUSR1:
5463 CMM_STORE_SHARED(recv_child_signal, 1);
5464 break;
5465 default:
5466 break;
5467 }
5468 }
5469
5470 /*
5471 * Setup signal handler for :
5472 * SIGINT, SIGTERM, SIGPIPE
5473 */
5474 static int set_signal_handler(void)
5475 {
5476 int ret = 0;
5477 struct sigaction sa;
5478 sigset_t sigset;
5479
5480 if ((ret = sigemptyset(&sigset)) < 0) {
5481 PERROR("sigemptyset");
5482 return ret;
5483 }
5484
5485 sa.sa_mask = sigset;
5486 sa.sa_flags = 0;
5487
5488 sa.sa_handler = sighandler;
5489 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5490 PERROR("sigaction");
5491 return ret;
5492 }
5493
5494 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5495 PERROR("sigaction");
5496 return ret;
5497 }
5498
5499 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5500 PERROR("sigaction");
5501 return ret;
5502 }
5503
5504 sa.sa_handler = SIG_IGN;
5505 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5506 PERROR("sigaction");
5507 return ret;
5508 }
5509
5510 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5511
5512 return ret;
5513 }
5514
5515 /*
5516 * Set open files limit to unlimited. This daemon can open a large number of
5517 * file descriptors in order to consume multiple kernel traces.
5518 */
5519 static void set_ulimit(void)
5520 {
5521 int ret;
5522 struct rlimit lim;
5523
5524 /* The kernel does not allow an infinite limit for open files */
5525 lim.rlim_cur = 65535;
5526 lim.rlim_max = 65535;
5527
5528 ret = setrlimit(RLIMIT_NOFILE, &lim);
5529 if (ret < 0) {
5530 PERROR("failed to set open files limit");
5531 }
5532 }
5533
5534 /*
5535 * Write pidfile using the rundir and opt_pidfile.
5536 */
5537 static int write_pidfile(void)
5538 {
5539 int ret;
5540 char pidfile_path[PATH_MAX];
5541
5542 assert(rundir);
5543
5544 if (opt_pidfile) {
5545 if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) {
5546 ret = -1;
5547 goto error;
5548 }
5549 } else {
5550 /* Build pidfile path from rundir and opt_pidfile. */
5551 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5552 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5553 if (ret < 0) {
5554 PERROR("snprintf pidfile path");
5555 goto error;
5556 }
5557 }
5558
5559 /*
5560 * Create pid file in rundir.
5561 */
5562 ret = utils_create_pid_file(getpid(), pidfile_path);
5563 error:
5564 return ret;
5565 }
5566
5567 /*
5568 * Create lockfile using the rundir and return its fd.
5569 */
5570 static int create_lockfile(void)
5571 {
5572 int ret;
5573 char lockfile_path[PATH_MAX];
5574
5575 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5576 if (ret < 0) {
5577 goto error;
5578 }
5579
5580 ret = utils_create_lock_file(lockfile_path);
5581 error:
5582 return ret;
5583 }
5584
5585 /*
5586 * Write agent TCP port using the rundir.
5587 */
5588 static int write_agent_port(void)
5589 {
5590 int ret;
5591 char path[PATH_MAX];
5592
5593 assert(rundir);
5594
5595 ret = snprintf(path, sizeof(path), "%s/"
5596 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5597 if (ret < 0) {
5598 PERROR("snprintf agent port path");
5599 goto error;
5600 }
5601
5602 /*
5603 * Create TCP agent port file in rundir.
5604 */
5605 ret = utils_create_pid_file(agent_tcp_port, path);
5606
5607 error:
5608 return ret;
5609 }
5610
5611 /*
5612 * main
5613 */
5614 int main(int argc, char **argv)
5615 {
5616 int ret = 0, retval = 0;
5617 void *status;
5618 const char *home_path, *env_app_timeout;
5619 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5620 *ust64_channel_monitor_pipe = NULL,
5621 *kernel_channel_monitor_pipe = NULL;
5622
5623 init_kernel_workarounds();
5624
5625 rcu_register_thread();
5626
5627 if (set_signal_handler()) {
5628 retval = -1;
5629 goto exit_set_signal_handler;
5630 }
5631
5632 setup_consumerd_path();
5633
5634 page_size = sysconf(_SC_PAGESIZE);
5635 if (page_size < 0) {
5636 PERROR("sysconf _SC_PAGESIZE");
5637 page_size = LONG_MAX;
5638 WARN("Fallback page size to %ld", page_size);
5639 }
5640
5641 /*
5642 * Parse arguments and load the daemon configuration file.
5643 *
5644 * We have an exit_options exit path to free memory reserved by
5645 * set_options. This is needed because the rest of sessiond_cleanup()
5646 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5647 * depends on set_options.
5648 */
5649 progname = argv[0];
5650 if (set_options(argc, argv)) {
5651 retval = -1;
5652 goto exit_options;
5653 }
5654
5655 /* Daemonize */
5656 if (opt_daemon || opt_background) {
5657 int i;
5658
5659 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5660 !opt_background);
5661 if (ret < 0) {
5662 retval = -1;
5663 goto exit_options;
5664 }
5665
5666 /*
5667 * We are in the child. Make sure all other file descriptors are
5668 * closed, in case we are called with more opened file
5669 * descriptors than the standard ones.
5670 */
5671 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5672 (void) close(i);
5673 }
5674 }
5675
5676 if (run_as_create_worker(argv[0]) < 0) {
5677 goto exit_create_run_as_worker_cleanup;
5678 }
5679
5680 /*
5681 * Starting from here, we can create threads. This needs to be after
5682 * lttng_daemonize due to RCU.
5683 */
5684
5685 /*
5686 * Initialize the health check subsystem. This call should set the
5687 * appropriate time values.
5688 */
5689 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5690 if (!health_sessiond) {
5691 PERROR("health_app_create error");
5692 retval = -1;
5693 goto exit_health_sessiond_cleanup;
5694 }
5695
5696 /* Create thread to clean up RCU hash tables */
5697 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5698 retval = -1;
5699 goto exit_ht_cleanup;
5700 }
5701
5702 /* Create thread quit pipe */
5703 if (init_thread_quit_pipe()) {
5704 retval = -1;
5705 goto exit_init_data;
5706 }
5707
5708 /* Check if daemon is UID = 0 */
5709 is_root = !getuid();
5710
5711 if (is_root) {
5712 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5713 if (!rundir) {
5714 retval = -1;
5715 goto exit_init_data;
5716 }
5717
5718 /* Create global run dir with root access */
5719 if (create_lttng_rundir(rundir)) {
5720 retval = -1;
5721 goto exit_init_data;
5722 }
5723
5724 if (strlen(apps_unix_sock_path) == 0) {
5725 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5726 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5727 if (ret < 0) {
5728 retval = -1;
5729 goto exit_init_data;
5730 }
5731 }
5732
5733 if (strlen(client_unix_sock_path) == 0) {
5734 ret = snprintf(client_unix_sock_path, PATH_MAX,
5735 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5736 if (ret < 0) {
5737 retval = -1;
5738 goto exit_init_data;
5739 }
5740 }
5741
5742 /* Set global SHM for ust */
5743 if (strlen(wait_shm_path) == 0) {
5744 ret = snprintf(wait_shm_path, PATH_MAX,
5745 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5746 if (ret < 0) {
5747 retval = -1;
5748 goto exit_init_data;
5749 }
5750 }
5751
5752 if (strlen(health_unix_sock_path) == 0) {
5753 ret = snprintf(health_unix_sock_path,
5754 sizeof(health_unix_sock_path),
5755 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5756 if (ret < 0) {
5757 retval = -1;
5758 goto exit_init_data;
5759 }
5760 }
5761
5762 /* Setup kernel consumerd path */
5763 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5764 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5765 if (ret < 0) {
5766 retval = -1;
5767 goto exit_init_data;
5768 }
5769 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5770 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5771 if (ret < 0) {
5772 retval = -1;
5773 goto exit_init_data;
5774 }
5775
5776 DBG2("Kernel consumer err path: %s",
5777 kconsumer_data.err_unix_sock_path);
5778 DBG2("Kernel consumer cmd path: %s",
5779 kconsumer_data.cmd_unix_sock_path);
5780 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5781 if (!kernel_channel_monitor_pipe) {
5782 ERR("Failed to create kernel consumer channel monitor pipe");
5783 retval = -1;
5784 goto exit_init_data;
5785 }
5786 kconsumer_data.channel_monitor_pipe =
5787 lttng_pipe_release_writefd(
5788 kernel_channel_monitor_pipe);
5789 if (kconsumer_data.channel_monitor_pipe < 0) {
5790 retval = -1;
5791 goto exit_init_data;
5792 }
5793 } else {
5794 home_path = utils_get_home_dir();
5795 if (home_path == NULL) {
5796 /* TODO: Add --socket PATH option */
5797 ERR("Can't get HOME directory for sockets creation.");
5798 retval = -1;
5799 goto exit_init_data;
5800 }
5801
5802 /*
5803 * Create rundir from home path. This will create something like
5804 * $HOME/.lttng
5805 */
5806 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5807 if (ret < 0) {
5808 retval = -1;
5809 goto exit_init_data;
5810 }
5811
5812 if (create_lttng_rundir(rundir)) {
5813 retval = -1;
5814 goto exit_init_data;
5815 }
5816
5817 if (strlen(apps_unix_sock_path) == 0) {
5818 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5819 DEFAULT_HOME_APPS_UNIX_SOCK,
5820 home_path);
5821 if (ret < 0) {
5822 retval = -1;
5823 goto exit_init_data;
5824 }
5825 }
5826
5827 /* Set the cli tool unix socket path */
5828 if (strlen(client_unix_sock_path) == 0) {
5829 ret = snprintf(client_unix_sock_path, PATH_MAX,
5830 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5831 home_path);
5832 if (ret < 0) {
5833 retval = -1;
5834 goto exit_init_data;
5835 }
5836 }
5837
5838 /* Set global SHM for ust */
5839 if (strlen(wait_shm_path) == 0) {
5840 ret = snprintf(wait_shm_path, PATH_MAX,
5841 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5842 getuid());
5843 if (ret < 0) {
5844 retval = -1;
5845 goto exit_init_data;
5846 }
5847 }
5848
5849 /* Set health check Unix path */
5850 if (strlen(health_unix_sock_path) == 0) {
5851 ret = snprintf(health_unix_sock_path,
5852 sizeof(health_unix_sock_path),
5853 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5854 home_path);
5855 if (ret < 0) {
5856 retval = -1;
5857 goto exit_init_data;
5858 }
5859 }
5860 }
5861
5862 lockfile_fd = create_lockfile();
5863 if (lockfile_fd < 0) {
5864 retval = -1;
5865 goto exit_init_data;
5866 }
5867
5868 /* Set consumer initial state */
5869 kernel_consumerd_state = CONSUMER_STOPPED;
5870 ust_consumerd_state = CONSUMER_STOPPED;
5871
5872 DBG("Client socket path %s", client_unix_sock_path);
5873 DBG("Application socket path %s", apps_unix_sock_path);
5874 DBG("Application wait path %s", wait_shm_path);
5875 DBG("LTTng run directory path: %s", rundir);
5876
5877 /* 32 bits consumerd path setup */
5878 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5879 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5880 if (ret < 0) {
5881 PERROR("snprintf 32-bit consumer error socket path");
5882 retval = -1;
5883 goto exit_init_data;
5884 }
5885 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5886 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5887 if (ret < 0) {
5888 PERROR("snprintf 32-bit consumer command socket path");
5889 retval = -1;
5890 goto exit_init_data;
5891 }
5892
5893 DBG2("UST consumer 32 bits err path: %s",
5894 ustconsumer32_data.err_unix_sock_path);
5895 DBG2("UST consumer 32 bits cmd path: %s",
5896 ustconsumer32_data.cmd_unix_sock_path);
5897 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5898 if (!ust32_channel_monitor_pipe) {
5899 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5900 retval = -1;
5901 goto exit_init_data;
5902 }
5903 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5904 ust32_channel_monitor_pipe);
5905 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5906 retval = -1;
5907 goto exit_init_data;
5908 }
5909
5910 /* 64 bits consumerd path setup */
5911 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5912 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5913 if (ret < 0) {
5914 PERROR("snprintf 64-bit consumer error socket path");
5915 retval = -1;
5916 goto exit_init_data;
5917 }
5918 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5919 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5920 if (ret < 0) {
5921 PERROR("snprintf 64-bit consumer command socket path");
5922 retval = -1;
5923 goto exit_init_data;
5924 }
5925
5926 DBG2("UST consumer 64 bits err path: %s",
5927 ustconsumer64_data.err_unix_sock_path);
5928 DBG2("UST consumer 64 bits cmd path: %s",
5929 ustconsumer64_data.cmd_unix_sock_path);
5930 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5931 if (!ust64_channel_monitor_pipe) {
5932 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5933 retval = -1;
5934 goto exit_init_data;
5935 }
5936 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5937 ust64_channel_monitor_pipe);
5938 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5939 retval = -1;
5940 goto exit_init_data;
5941 }
5942
5943 /*
5944 * See if daemon already exist.
5945 */
5946 if (check_existing_daemon()) {
5947 ERR("Already running daemon.\n");
5948 /*
5949 * We do not goto exit because we must not cleanup()
5950 * because a daemon is already running.
5951 */
5952 retval = -1;
5953 goto exit_init_data;
5954 }
5955
5956 /*
5957 * Init UST app hash table. Alloc hash table before this point since
5958 * cleanup() can get called after that point.
5959 */
5960 if (ust_app_ht_alloc()) {
5961 ERR("Failed to allocate UST app hash table");
5962 retval = -1;
5963 goto exit_init_data;
5964 }
5965
5966 /*
5967 * Initialize agent app hash table. We allocate the hash table here
5968 * since cleanup() can get called after this point.
5969 */
5970 if (agent_app_ht_alloc()) {
5971 ERR("Failed to allocate Agent app hash table");
5972 retval = -1;
5973 goto exit_init_data;
5974 }
5975
5976 /*
5977 * These actions must be executed as root. We do that *after* setting up
5978 * the sockets path because we MUST make the check for another daemon using
5979 * those paths *before* trying to set the kernel consumer sockets and init
5980 * kernel tracer.
5981 */
5982 if (is_root) {
5983 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5984 retval = -1;
5985 goto exit_init_data;
5986 }
5987
5988 /* Setup kernel tracer */
5989 if (!opt_no_kernel) {
5990 init_kernel_tracer();
5991 if (kernel_tracer_fd >= 0) {
5992 ret = syscall_init_table();
5993 if (ret < 0) {
5994 ERR("Unable to populate syscall table. "
5995 "Syscall tracing won't work "
5996 "for this session daemon.");
5997 }
5998 }
5999 }
6000
6001 /* Set ulimit for open files */
6002 set_ulimit();
6003 }
6004 /* init lttng_fd tracking must be done after set_ulimit. */
6005 lttng_fd_init();
6006
6007 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
6008 retval = -1;
6009 goto exit_init_data;
6010 }
6011
6012 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
6013 retval = -1;
6014 goto exit_init_data;
6015 }
6016
6017 /* Setup the needed unix socket */
6018 if (init_daemon_socket()) {
6019 retval = -1;
6020 goto exit_init_data;
6021 }
6022
6023 /* Set credentials to socket */
6024 if (is_root && set_permissions(rundir)) {
6025 retval = -1;
6026 goto exit_init_data;
6027 }
6028
6029 /* Get parent pid if -S, --sig-parent is specified. */
6030 if (opt_sig_parent) {
6031 ppid = getppid();
6032 }
6033
6034 /* Setup the kernel pipe for waking up the kernel thread */
6035 if (is_root && !opt_no_kernel) {
6036 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
6037 retval = -1;
6038 goto exit_init_data;
6039 }
6040 }
6041
6042 /* Setup the thread apps communication pipe. */
6043 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
6044 retval = -1;
6045 goto exit_init_data;
6046 }
6047
6048 /* Setup the thread apps notify communication pipe. */
6049 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
6050 retval = -1;
6051 goto exit_init_data;
6052 }
6053
6054 /* Initialize global buffer per UID and PID registry. */
6055 buffer_reg_init_uid_registry();
6056 buffer_reg_init_pid_registry();
6057
6058 /* Init UST command queue. */
6059 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
6060
6061 /*
6062 * Get session list pointer. This pointer MUST NOT be free'd. This list
6063 * is statically declared in session.c
6064 */
6065 session_list_ptr = session_get_list();
6066
6067 cmd_init();
6068
6069 /* Check for the application socket timeout env variable. */
6070 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
6071 if (env_app_timeout) {
6072 app_socket_timeout = atoi(env_app_timeout);
6073 } else {
6074 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
6075 }
6076
6077 ret = write_pidfile();
6078 if (ret) {
6079 ERR("Error in write_pidfile");
6080 retval = -1;
6081 goto exit_init_data;
6082 }
6083 ret = write_agent_port();
6084 if (ret) {
6085 ERR("Error in write_agent_port");
6086 retval = -1;
6087 goto exit_init_data;
6088 }
6089
6090 /* Initialize communication library */
6091 lttcomm_init();
6092 /* Initialize TCP timeout values */
6093 lttcomm_inet_init();
6094
6095 if (load_session_init_data(&load_info) < 0) {
6096 retval = -1;
6097 goto exit_init_data;
6098 }
6099 load_info->path = opt_load_session_path;
6100
6101 /* Create health-check thread. */
6102 ret = pthread_create(&health_thread, default_pthread_attr(),
6103 thread_manage_health, (void *) NULL);
6104 if (ret) {
6105 errno = ret;
6106 PERROR("pthread_create health");
6107 retval = -1;
6108 goto exit_health;
6109 }
6110
6111 /* notification_thread_data acquires the pipes' read side. */
6112 notification_thread_handle = notification_thread_handle_create(
6113 ust32_channel_monitor_pipe,
6114 ust64_channel_monitor_pipe,
6115 kernel_channel_monitor_pipe);
6116 if (!notification_thread_handle) {
6117 retval = -1;
6118 ERR("Failed to create notification thread shared data");
6119 stop_threads();
6120 goto exit_notification;
6121 }
6122
6123 /* Create notification thread. */
6124 ret = pthread_create(&notification_thread, default_pthread_attr(),
6125 thread_notification, notification_thread_handle);
6126 if (ret) {
6127 errno = ret;
6128 PERROR("pthread_create notification");
6129 retval = -1;
6130 stop_threads();
6131 goto exit_notification;
6132 }
6133
6134 /* Create thread to manage the client socket */
6135 ret = pthread_create(&client_thread, default_pthread_attr(),
6136 thread_manage_clients, (void *) NULL);
6137 if (ret) {
6138 errno = ret;
6139 PERROR("pthread_create clients");
6140 retval = -1;
6141 stop_threads();
6142 goto exit_client;
6143 }
6144
6145 /* Create thread to dispatch registration */
6146 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
6147 thread_dispatch_ust_registration, (void *) NULL);
6148 if (ret) {
6149 errno = ret;
6150 PERROR("pthread_create dispatch");
6151 retval = -1;
6152 stop_threads();
6153 goto exit_dispatch;
6154 }
6155
6156 /* Create thread to manage application registration. */
6157 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
6158 thread_registration_apps, (void *) NULL);
6159 if (ret) {
6160 errno = ret;
6161 PERROR("pthread_create registration");
6162 retval = -1;
6163 stop_threads();
6164 goto exit_reg_apps;
6165 }
6166
6167 /* Create thread to manage application socket */
6168 ret = pthread_create(&apps_thread, default_pthread_attr(),
6169 thread_manage_apps, (void *) NULL);
6170 if (ret) {
6171 errno = ret;
6172 PERROR("pthread_create apps");
6173 retval = -1;
6174 stop_threads();
6175 goto exit_apps;
6176 }
6177
6178 /* Create thread to manage application notify socket */
6179 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
6180 ust_thread_manage_notify, (void *) NULL);
6181 if (ret) {
6182 errno = ret;
6183 PERROR("pthread_create notify");
6184 retval = -1;
6185 stop_threads();
6186 goto exit_apps_notify;
6187 }
6188
6189 /* Create agent registration thread. */
6190 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
6191 agent_thread_manage_registration, (void *) NULL);
6192 if (ret) {
6193 errno = ret;
6194 PERROR("pthread_create agent");
6195 retval = -1;
6196 stop_threads();
6197 goto exit_agent_reg;
6198 }
6199
6200 /* Don't start this thread if kernel tracing is not requested nor root */
6201 if (is_root && !opt_no_kernel) {
6202 /* Create kernel thread to manage kernel event */
6203 ret = pthread_create(&kernel_thread, default_pthread_attr(),
6204 thread_manage_kernel, (void *) NULL);
6205 if (ret) {
6206 errno = ret;
6207 PERROR("pthread_create kernel");
6208 retval = -1;
6209 stop_threads();
6210 goto exit_kernel;
6211 }
6212 }
6213
6214 /* Create session loading thread. */
6215 ret = pthread_create(&load_session_thread, default_pthread_attr(),
6216 thread_load_session, load_info);
6217 if (ret) {
6218 errno = ret;
6219 PERROR("pthread_create load_session_thread");
6220 retval = -1;
6221 stop_threads();
6222 goto exit_load_session;
6223 }
6224
6225 /*
6226 * This is where we start awaiting program completion (e.g. through
6227 * signal that asks threads to teardown).
6228 */
6229
6230 ret = pthread_join(load_session_thread, &status);
6231 if (ret) {
6232 errno = ret;
6233 PERROR("pthread_join load_session_thread");
6234 retval = -1;
6235 }
6236 exit_load_session:
6237
6238 if (is_root && !opt_no_kernel) {
6239 ret = pthread_join(kernel_thread, &status);
6240 if (ret) {
6241 errno = ret;
6242 PERROR("pthread_join");
6243 retval = -1;
6244 }
6245 }
6246 exit_kernel:
6247
6248 ret = pthread_join(agent_reg_thread, &status);
6249 if (ret) {
6250 errno = ret;
6251 PERROR("pthread_join agent");
6252 retval = -1;
6253 }
6254 exit_agent_reg:
6255
6256 ret = pthread_join(apps_notify_thread, &status);
6257 if (ret) {
6258 errno = ret;
6259 PERROR("pthread_join apps notify");
6260 retval = -1;
6261 }
6262 exit_apps_notify:
6263
6264 ret = pthread_join(apps_thread, &status);
6265 if (ret) {
6266 errno = ret;
6267 PERROR("pthread_join apps");
6268 retval = -1;
6269 }
6270 exit_apps:
6271
6272 ret = pthread_join(reg_apps_thread, &status);
6273 if (ret) {
6274 errno = ret;
6275 PERROR("pthread_join");
6276 retval = -1;
6277 }
6278 exit_reg_apps:
6279
6280 /*
6281 * Join dispatch thread after joining reg_apps_thread to ensure
6282 * we don't leak applications in the queue.
6283 */
6284 ret = pthread_join(dispatch_thread, &status);
6285 if (ret) {
6286 errno = ret;
6287 PERROR("pthread_join");
6288 retval = -1;
6289 }
6290 exit_dispatch:
6291
6292 ret = pthread_join(client_thread, &status);
6293 if (ret) {
6294 errno = ret;
6295 PERROR("pthread_join");
6296 retval = -1;
6297 }
6298
6299 exit_client:
6300 exit_notification:
6301 ret = pthread_join(health_thread, &status);
6302 if (ret) {
6303 errno = ret;
6304 PERROR("pthread_join health thread");
6305 retval = -1;
6306 }
6307
6308 exit_health:
6309 exit_init_data:
6310 /*
6311 * Wait for all pending call_rcu work to complete before tearing
6312 * down data structures. call_rcu worker may be trying to
6313 * perform lookups in those structures.
6314 */
6315 rcu_barrier();
6316 /*
6317 * sessiond_cleanup() is called when no other thread is running, except
6318 * the ht_cleanup thread, which is needed to destroy the hash tables.
6319 */
6320 rcu_thread_online();
6321 sessiond_cleanup();
6322
6323 /*
6324 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6325 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6326 * the queue is empty before shutting down the clean-up thread.
6327 */
6328 rcu_barrier();
6329
6330 /*
6331 * The teardown of the notification system is performed after the
6332 * session daemon's teardown in order to allow it to be notified
6333 * of the active session and channels at the moment of the teardown.
6334 */
6335 if (notification_thread_handle) {
6336 notification_thread_command_quit(notification_thread_handle);
6337 notification_thread_handle_destroy(notification_thread_handle);
6338 }
6339
6340 ret = pthread_join(notification_thread, &status);
6341 if (ret) {
6342 errno = ret;
6343 PERROR("pthread_join notification thread");
6344 retval = -1;
6345 }
6346
6347 rcu_thread_offline();
6348 rcu_unregister_thread();
6349
6350 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6351 if (ret) {
6352 retval = -1;
6353 }
6354 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6355 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6356 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6357 exit_ht_cleanup:
6358
6359 health_app_destroy(health_sessiond);
6360 exit_health_sessiond_cleanup:
6361 exit_create_run_as_worker_cleanup:
6362
6363 exit_options:
6364 sessiond_cleanup_options();
6365
6366 exit_set_signal_handler:
6367 if (!retval) {
6368 exit(EXIT_SUCCESS);
6369 } else {
6370 exit(EXIT_FAILURE);
6371 }
6372 }
This page took 0.190677 seconds and 5 git commands to generate.