Fix: ret is never used on error_open code path
[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 } else if (!WIFEXITED(status)) {
604 ERR("consumerd termination with error: %d",
605 WEXITSTATUS(ret));
606 }
607 consumer_data->pid = 0;
608 }
609
610 /*
611 * Cleanup the session daemon's data structures.
612 */
613 static void sessiond_cleanup(void)
614 {
615 int ret;
616 struct ltt_session *sess, *stmp;
617 char path[PATH_MAX];
618
619 DBG("Cleanup sessiond");
620
621 /*
622 * Close the thread quit pipe. It has already done its job,
623 * since we are now called.
624 */
625 utils_close_pipe(thread_quit_pipe);
626
627 /*
628 * If opt_pidfile is undefined, the default file will be wiped when
629 * removing the rundir.
630 */
631 if (opt_pidfile) {
632 ret = remove(opt_pidfile);
633 if (ret < 0) {
634 PERROR("remove pidfile %s", opt_pidfile);
635 }
636 }
637
638 DBG("Removing sessiond and consumerd content of directory %s", rundir);
639
640 /* sessiond */
641 snprintf(path, PATH_MAX,
642 "%s/%s",
643 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
644 DBG("Removing %s", path);
645 (void) unlink(path);
646
647 snprintf(path, PATH_MAX, "%s/%s", rundir,
648 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
649 DBG("Removing %s", path);
650 (void) unlink(path);
651
652 /* kconsumerd */
653 snprintf(path, PATH_MAX,
654 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
655 rundir);
656 DBG("Removing %s", path);
657 (void) unlink(path);
658
659 snprintf(path, PATH_MAX,
660 DEFAULT_KCONSUMERD_PATH,
661 rundir);
662 DBG("Removing directory %s", path);
663 (void) rmdir(path);
664
665 /* ust consumerd 32 */
666 snprintf(path, PATH_MAX,
667 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
668 rundir);
669 DBG("Removing %s", path);
670 (void) unlink(path);
671
672 snprintf(path, PATH_MAX,
673 DEFAULT_USTCONSUMERD32_PATH,
674 rundir);
675 DBG("Removing directory %s", path);
676 (void) rmdir(path);
677
678 /* ust consumerd 64 */
679 snprintf(path, PATH_MAX,
680 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
681 rundir);
682 DBG("Removing %s", path);
683 (void) unlink(path);
684
685 snprintf(path, PATH_MAX,
686 DEFAULT_USTCONSUMERD64_PATH,
687 rundir);
688 DBG("Removing directory %s", path);
689 (void) rmdir(path);
690
691 DBG("Cleaning up all sessions");
692
693 /* Destroy session list mutex */
694 if (session_list_ptr != NULL) {
695 pthread_mutex_destroy(&session_list_ptr->lock);
696
697 /* Cleanup ALL session */
698 cds_list_for_each_entry_safe(sess, stmp,
699 &session_list_ptr->head, list) {
700 cmd_destroy_session(sess, kernel_poll_pipe[1]);
701 }
702 }
703
704 wait_consumer(&kconsumer_data);
705 wait_consumer(&ustconsumer64_data);
706 wait_consumer(&ustconsumer32_data);
707
708 DBG("Cleaning up all agent apps");
709 agent_app_ht_clean();
710
711 DBG("Closing all UST sockets");
712 ust_app_clean_list();
713 buffer_reg_destroy_registries();
714
715 if (is_root && !opt_no_kernel) {
716 DBG2("Closing kernel fd");
717 if (kernel_tracer_fd >= 0) {
718 ret = close(kernel_tracer_fd);
719 if (ret) {
720 PERROR("close");
721 }
722 }
723 DBG("Unloading kernel modules");
724 modprobe_remove_lttng_all();
725 free(syscall_table);
726 }
727
728 close_consumer_sockets();
729
730 if (load_info) {
731 load_session_destroy_data(load_info);
732 free(load_info);
733 }
734
735 /*
736 * Cleanup lock file by deleting it and finaly closing it which will
737 * release the file system lock.
738 */
739 if (lockfile_fd >= 0) {
740 char lockfile_path[PATH_MAX];
741
742 ret = generate_lock_file_path(lockfile_path,
743 sizeof(lockfile_path));
744 if (ret > 0) {
745 ret = remove(lockfile_path);
746 if (ret < 0) {
747 PERROR("remove lock file");
748 }
749 ret = close(lockfile_fd);
750 if (ret < 0) {
751 PERROR("close lock file");
752 }
753 }
754 }
755
756 /*
757 * We do NOT rmdir rundir because there are other processes
758 * using it, for instance lttng-relayd, which can start in
759 * parallel with this teardown.
760 */
761
762 free(rundir);
763 }
764
765 /*
766 * Cleanup the daemon's option data structures.
767 */
768 static void sessiond_cleanup_options(void)
769 {
770 DBG("Cleaning up options");
771
772 /*
773 * If the override option is set, the pointer points to a *non* const
774 * thus freeing it even though the variable type is set to const.
775 */
776 if (tracing_group_name_override) {
777 free((void *) tracing_group_name);
778 }
779 if (consumerd32_bin_override) {
780 free((void *) consumerd32_bin);
781 }
782 if (consumerd64_bin_override) {
783 free((void *) consumerd64_bin);
784 }
785 if (consumerd32_libdir_override) {
786 free((void *) consumerd32_libdir);
787 }
788 if (consumerd64_libdir_override) {
789 free((void *) consumerd64_libdir);
790 }
791
792 free(opt_pidfile);
793 free(opt_load_session_path);
794 free(kmod_probes_list);
795 free(kmod_extra_probes_list);
796
797 run_as_destroy_worker();
798 }
799
800 /*
801 * Send data on a unix socket using the liblttsessiondcomm API.
802 *
803 * Return lttcomm error code.
804 */
805 static int send_unix_sock(int sock, void *buf, size_t len)
806 {
807 /* Check valid length */
808 if (len == 0) {
809 return -1;
810 }
811
812 return lttcomm_send_unix_sock(sock, buf, len);
813 }
814
815 /*
816 * Free memory of a command context structure.
817 */
818 static void clean_command_ctx(struct command_ctx **cmd_ctx)
819 {
820 DBG("Clean command context structure");
821 if (*cmd_ctx) {
822 if ((*cmd_ctx)->llm) {
823 free((*cmd_ctx)->llm);
824 }
825 if ((*cmd_ctx)->lsm) {
826 free((*cmd_ctx)->lsm);
827 }
828 free(*cmd_ctx);
829 *cmd_ctx = NULL;
830 }
831 }
832
833 /*
834 * Notify UST applications using the shm mmap futex.
835 */
836 static int notify_ust_apps(int active)
837 {
838 char *wait_shm_mmap;
839
840 DBG("Notifying applications of session daemon state: %d", active);
841
842 /* See shm.c for this call implying mmap, shm and futex calls */
843 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
844 if (wait_shm_mmap == NULL) {
845 goto error;
846 }
847
848 /* Wake waiting process */
849 futex_wait_update((int32_t *) wait_shm_mmap, active);
850
851 /* Apps notified successfully */
852 return 0;
853
854 error:
855 return -1;
856 }
857
858 /*
859 * Setup the outgoing data buffer for the response (llm) by allocating the
860 * right amount of memory and copying the original information from the lsm
861 * structure.
862 *
863 * Return 0 on success, negative value on error.
864 */
865 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
866 const void *payload_buf, size_t payload_len,
867 const void *cmd_header_buf, size_t cmd_header_len)
868 {
869 int ret = 0;
870 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
871 const size_t cmd_header_offset = header_len;
872 const size_t payload_offset = cmd_header_offset + cmd_header_len;
873 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
874
875 cmd_ctx->llm = zmalloc(total_msg_size);
876
877 if (cmd_ctx->llm == NULL) {
878 PERROR("zmalloc");
879 ret = -ENOMEM;
880 goto end;
881 }
882
883 /* Copy common data */
884 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
885 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
886 cmd_ctx->llm->cmd_header_size = cmd_header_len;
887 cmd_ctx->llm->data_size = payload_len;
888 cmd_ctx->lttng_msg_size = total_msg_size;
889
890 /* Copy command header */
891 if (cmd_header_len) {
892 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
893 cmd_header_len);
894 }
895
896 /* Copy payload */
897 if (payload_len) {
898 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
899 payload_len);
900 }
901
902 end:
903 return ret;
904 }
905
906 /*
907 * Version of setup_lttng_msg() without command header.
908 */
909 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
910 void *payload_buf, size_t payload_len)
911 {
912 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
913 }
914 /*
915 * Update the kernel poll set of all channel fd available over all tracing
916 * session. Add the wakeup pipe at the end of the set.
917 */
918 static int update_kernel_poll(struct lttng_poll_event *events)
919 {
920 int ret;
921 struct ltt_session *session;
922 struct ltt_kernel_channel *channel;
923
924 DBG("Updating kernel poll set");
925
926 session_lock_list();
927 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
928 session_lock(session);
929 if (session->kernel_session == NULL) {
930 session_unlock(session);
931 continue;
932 }
933
934 cds_list_for_each_entry(channel,
935 &session->kernel_session->channel_list.head, list) {
936 /* Add channel fd to the kernel poll set */
937 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
938 if (ret < 0) {
939 session_unlock(session);
940 goto error;
941 }
942 DBG("Channel fd %d added to kernel set", channel->fd);
943 }
944 session_unlock(session);
945 }
946 session_unlock_list();
947
948 return 0;
949
950 error:
951 session_unlock_list();
952 return -1;
953 }
954
955 /*
956 * Find the channel fd from 'fd' over all tracing session. When found, check
957 * for new channel stream and send those stream fds to the kernel consumer.
958 *
959 * Useful for CPU hotplug feature.
960 */
961 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
962 {
963 int ret = 0;
964 struct ltt_session *session;
965 struct ltt_kernel_session *ksess;
966 struct ltt_kernel_channel *channel;
967
968 DBG("Updating kernel streams for channel fd %d", fd);
969
970 session_lock_list();
971 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
972 session_lock(session);
973 if (session->kernel_session == NULL) {
974 session_unlock(session);
975 continue;
976 }
977 ksess = session->kernel_session;
978
979 cds_list_for_each_entry(channel,
980 &ksess->channel_list.head, list) {
981 struct lttng_ht_iter iter;
982 struct consumer_socket *socket;
983
984 if (channel->fd != fd) {
985 continue;
986 }
987 DBG("Channel found, updating kernel streams");
988 ret = kernel_open_channel_stream(channel);
989 if (ret < 0) {
990 goto error;
991 }
992 /* Update the stream global counter */
993 ksess->stream_count_global += ret;
994
995 /*
996 * Have we already sent fds to the consumer? If yes, it
997 * means that tracing is started so it is safe to send
998 * our updated stream fds.
999 */
1000 if (ksess->consumer_fds_sent != 1
1001 || ksess->consumer == NULL) {
1002 ret = -1;
1003 goto error;
1004 }
1005
1006 rcu_read_lock();
1007 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
1008 &iter.iter, socket, node.node) {
1009 pthread_mutex_lock(socket->lock);
1010 ret = kernel_consumer_send_channel_stream(socket,
1011 channel, ksess,
1012 session->output_traces ? 1 : 0);
1013 pthread_mutex_unlock(socket->lock);
1014 if (ret < 0) {
1015 rcu_read_unlock();
1016 goto error;
1017 }
1018 }
1019 rcu_read_unlock();
1020 }
1021 session_unlock(session);
1022 }
1023 session_unlock_list();
1024 return ret;
1025
1026 error:
1027 session_unlock(session);
1028 session_unlock_list();
1029 return ret;
1030 }
1031
1032 /*
1033 * For each tracing session, update newly registered apps. The session list
1034 * lock MUST be acquired before calling this.
1035 */
1036 static void update_ust_app(int app_sock)
1037 {
1038 struct ltt_session *sess, *stmp;
1039
1040 /* Consumer is in an ERROR state. Stop any application update. */
1041 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1042 /* Stop the update process since the consumer is dead. */
1043 return;
1044 }
1045
1046 /* For all tracing session(s) */
1047 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1048 struct ust_app *app;
1049
1050 session_lock(sess);
1051 if (!sess->ust_session) {
1052 goto unlock_session;
1053 }
1054
1055 rcu_read_lock();
1056 assert(app_sock >= 0);
1057 app = ust_app_find_by_sock(app_sock);
1058 if (app == NULL) {
1059 /*
1060 * Application can be unregistered before so
1061 * this is possible hence simply stopping the
1062 * update.
1063 */
1064 DBG3("UST app update failed to find app sock %d",
1065 app_sock);
1066 goto unlock_rcu;
1067 }
1068 ust_app_global_update(sess->ust_session, app);
1069 unlock_rcu:
1070 rcu_read_unlock();
1071 unlock_session:
1072 session_unlock(sess);
1073 }
1074 }
1075
1076 /*
1077 * This thread manage event coming from the kernel.
1078 *
1079 * Features supported in this thread:
1080 * -) CPU Hotplug
1081 */
1082 static void *thread_manage_kernel(void *data)
1083 {
1084 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1085 uint32_t revents, nb_fd;
1086 char tmp;
1087 struct lttng_poll_event events;
1088
1089 DBG("[thread] Thread manage kernel started");
1090
1091 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1092
1093 /*
1094 * This first step of the while is to clean this structure which could free
1095 * non NULL pointers so initialize it before the loop.
1096 */
1097 lttng_poll_init(&events);
1098
1099 if (testpoint(sessiond_thread_manage_kernel)) {
1100 goto error_testpoint;
1101 }
1102
1103 health_code_update();
1104
1105 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1106 goto error_testpoint;
1107 }
1108
1109 while (1) {
1110 health_code_update();
1111
1112 if (update_poll_flag == 1) {
1113 /* Clean events object. We are about to populate it again. */
1114 lttng_poll_clean(&events);
1115
1116 ret = sessiond_set_thread_pollset(&events, 2);
1117 if (ret < 0) {
1118 goto error_poll_create;
1119 }
1120
1121 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1122 if (ret < 0) {
1123 goto error;
1124 }
1125
1126 /* This will add the available kernel channel if any. */
1127 ret = update_kernel_poll(&events);
1128 if (ret < 0) {
1129 goto error;
1130 }
1131 update_poll_flag = 0;
1132 }
1133
1134 DBG("Thread kernel polling");
1135
1136 /* Poll infinite value of time */
1137 restart:
1138 health_poll_entry();
1139 ret = lttng_poll_wait(&events, -1);
1140 DBG("Thread kernel return from poll on %d fds",
1141 LTTNG_POLL_GETNB(&events));
1142 health_poll_exit();
1143 if (ret < 0) {
1144 /*
1145 * Restart interrupted system call.
1146 */
1147 if (errno == EINTR) {
1148 goto restart;
1149 }
1150 goto error;
1151 } else if (ret == 0) {
1152 /* Should not happen since timeout is infinite */
1153 ERR("Return value of poll is 0 with an infinite timeout.\n"
1154 "This should not have happened! Continuing...");
1155 continue;
1156 }
1157
1158 nb_fd = ret;
1159
1160 for (i = 0; i < nb_fd; i++) {
1161 /* Fetch once the poll data */
1162 revents = LTTNG_POLL_GETEV(&events, i);
1163 pollfd = LTTNG_POLL_GETFD(&events, i);
1164
1165 health_code_update();
1166
1167 if (!revents) {
1168 /* No activity for this FD (poll implementation). */
1169 continue;
1170 }
1171
1172 /* Thread quit pipe has been closed. Killing thread. */
1173 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1174 if (ret) {
1175 err = 0;
1176 goto exit;
1177 }
1178
1179 /* Check for data on kernel pipe */
1180 if (revents & LPOLLIN) {
1181 if (pollfd == kernel_poll_pipe[0]) {
1182 (void) lttng_read(kernel_poll_pipe[0],
1183 &tmp, 1);
1184 /*
1185 * Ret value is useless here, if this pipe gets any actions an
1186 * update is required anyway.
1187 */
1188 update_poll_flag = 1;
1189 continue;
1190 } else {
1191 /*
1192 * New CPU detected by the kernel. Adding kernel stream to
1193 * kernel session and updating the kernel consumer
1194 */
1195 ret = update_kernel_stream(&kconsumer_data, pollfd);
1196 if (ret < 0) {
1197 continue;
1198 }
1199 break;
1200 }
1201 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1202 update_poll_flag = 1;
1203 continue;
1204 } else {
1205 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1206 goto error;
1207 }
1208 }
1209 }
1210
1211 exit:
1212 error:
1213 lttng_poll_clean(&events);
1214 error_poll_create:
1215 error_testpoint:
1216 utils_close_pipe(kernel_poll_pipe);
1217 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1218 if (err) {
1219 health_error();
1220 ERR("Health error occurred in %s", __func__);
1221 WARN("Kernel thread died unexpectedly. "
1222 "Kernel tracing can continue but CPU hotplug is disabled.");
1223 }
1224 health_unregister(health_sessiond);
1225 DBG("Kernel thread dying");
1226 return NULL;
1227 }
1228
1229 /*
1230 * Signal pthread condition of the consumer data that the thread.
1231 */
1232 static void signal_consumer_condition(struct consumer_data *data, int state)
1233 {
1234 pthread_mutex_lock(&data->cond_mutex);
1235
1236 /*
1237 * The state is set before signaling. It can be any value, it's the waiter
1238 * job to correctly interpret this condition variable associated to the
1239 * consumer pthread_cond.
1240 *
1241 * A value of 0 means that the corresponding thread of the consumer data
1242 * was not started. 1 indicates that the thread has started and is ready
1243 * for action. A negative value means that there was an error during the
1244 * thread bootstrap.
1245 */
1246 data->consumer_thread_is_ready = state;
1247 (void) pthread_cond_signal(&data->cond);
1248
1249 pthread_mutex_unlock(&data->cond_mutex);
1250 }
1251
1252 /*
1253 * This thread manage the consumer error sent back to the session daemon.
1254 */
1255 static void *thread_manage_consumer(void *data)
1256 {
1257 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1258 uint32_t revents, nb_fd;
1259 enum lttcomm_return_code code;
1260 struct lttng_poll_event events;
1261 struct consumer_data *consumer_data = data;
1262 struct consumer_socket *cmd_socket_wrapper = NULL;
1263
1264 DBG("[thread] Manage consumer started");
1265
1266 rcu_register_thread();
1267 rcu_thread_online();
1268
1269 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1270
1271 health_code_update();
1272
1273 /*
1274 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1275 * metadata_sock. Nothing more will be added to this poll set.
1276 */
1277 ret = sessiond_set_thread_pollset(&events, 3);
1278 if (ret < 0) {
1279 goto error_poll;
1280 }
1281
1282 /*
1283 * The error socket here is already in a listening state which was done
1284 * just before spawning this thread to avoid a race between the consumer
1285 * daemon exec trying to connect and the listen() call.
1286 */
1287 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1288 if (ret < 0) {
1289 goto error;
1290 }
1291
1292 health_code_update();
1293
1294 /* Infinite blocking call, waiting for transmission */
1295 restart:
1296 health_poll_entry();
1297
1298 if (testpoint(sessiond_thread_manage_consumer)) {
1299 goto error;
1300 }
1301
1302 ret = lttng_poll_wait(&events, -1);
1303 health_poll_exit();
1304 if (ret < 0) {
1305 /*
1306 * Restart interrupted system call.
1307 */
1308 if (errno == EINTR) {
1309 goto restart;
1310 }
1311 goto error;
1312 }
1313
1314 nb_fd = ret;
1315
1316 for (i = 0; i < nb_fd; i++) {
1317 /* Fetch once the poll data */
1318 revents = LTTNG_POLL_GETEV(&events, i);
1319 pollfd = LTTNG_POLL_GETFD(&events, i);
1320
1321 health_code_update();
1322
1323 if (!revents) {
1324 /* No activity for this FD (poll implementation). */
1325 continue;
1326 }
1327
1328 /* Thread quit pipe has been closed. Killing thread. */
1329 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1330 if (ret) {
1331 err = 0;
1332 goto exit;
1333 }
1334
1335 /* Event on the registration socket */
1336 if (pollfd == consumer_data->err_sock) {
1337 if (revents & LPOLLIN) {
1338 continue;
1339 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1340 ERR("consumer err socket poll error");
1341 goto error;
1342 } else {
1343 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1344 goto error;
1345 }
1346 }
1347 }
1348
1349 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1350 if (sock < 0) {
1351 goto error;
1352 }
1353
1354 /*
1355 * Set the CLOEXEC flag. Return code is useless because either way, the
1356 * show must go on.
1357 */
1358 (void) utils_set_fd_cloexec(sock);
1359
1360 health_code_update();
1361
1362 DBG2("Receiving code from consumer err_sock");
1363
1364 /* Getting status code from kconsumerd */
1365 ret = lttcomm_recv_unix_sock(sock, &code,
1366 sizeof(enum lttcomm_return_code));
1367 if (ret <= 0) {
1368 goto error;
1369 }
1370
1371 health_code_update();
1372 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1373 ERR("consumer error when waiting for SOCK_READY : %s",
1374 lttcomm_get_readable_code(-code));
1375 goto error;
1376 }
1377
1378 /* Connect both command and metadata sockets. */
1379 consumer_data->cmd_sock =
1380 lttcomm_connect_unix_sock(
1381 consumer_data->cmd_unix_sock_path);
1382 consumer_data->metadata_fd =
1383 lttcomm_connect_unix_sock(
1384 consumer_data->cmd_unix_sock_path);
1385 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1386 PERROR("consumer connect cmd socket");
1387 /* On error, signal condition and quit. */
1388 signal_consumer_condition(consumer_data, -1);
1389 goto error;
1390 }
1391
1392 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1393
1394 /* Create metadata socket lock. */
1395 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1396 if (consumer_data->metadata_sock.lock == NULL) {
1397 PERROR("zmalloc pthread mutex");
1398 goto error;
1399 }
1400 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1401
1402 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1403 DBG("Consumer metadata socket ready (fd: %d)",
1404 consumer_data->metadata_fd);
1405
1406 /*
1407 * Remove the consumerd error sock since we've established a connection.
1408 */
1409 ret = lttng_poll_del(&events, consumer_data->err_sock);
1410 if (ret < 0) {
1411 goto error;
1412 }
1413
1414 /* Add new accepted error socket. */
1415 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1416 if (ret < 0) {
1417 goto error;
1418 }
1419
1420 /* Add metadata socket that is successfully connected. */
1421 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1422 LPOLLIN | LPOLLRDHUP);
1423 if (ret < 0) {
1424 goto error;
1425 }
1426
1427 health_code_update();
1428
1429 /*
1430 * Transfer the write-end of the channel monitoring pipe to the
1431 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1432 */
1433 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1434 if (!cmd_socket_wrapper) {
1435 goto error;
1436 }
1437
1438 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1439 consumer_data->channel_monitor_pipe);
1440 if (ret) {
1441 goto error;
1442 }
1443 /* Discard the socket wrapper as it is no longer needed. */
1444 consumer_destroy_socket(cmd_socket_wrapper);
1445 cmd_socket_wrapper = NULL;
1446
1447 /* The thread is completely initialized, signal that it is ready. */
1448 signal_consumer_condition(consumer_data, 1);
1449
1450 /* Infinite blocking call, waiting for transmission */
1451 restart_poll:
1452 while (1) {
1453 health_code_update();
1454
1455 /* Exit the thread because the thread quit pipe has been triggered. */
1456 if (should_quit) {
1457 /* Not a health error. */
1458 err = 0;
1459 goto exit;
1460 }
1461
1462 health_poll_entry();
1463 ret = lttng_poll_wait(&events, -1);
1464 health_poll_exit();
1465 if (ret < 0) {
1466 /*
1467 * Restart interrupted system call.
1468 */
1469 if (errno == EINTR) {
1470 goto restart_poll;
1471 }
1472 goto error;
1473 }
1474
1475 nb_fd = ret;
1476
1477 for (i = 0; i < nb_fd; i++) {
1478 /* Fetch once the poll data */
1479 revents = LTTNG_POLL_GETEV(&events, i);
1480 pollfd = LTTNG_POLL_GETFD(&events, i);
1481
1482 health_code_update();
1483
1484 if (!revents) {
1485 /* No activity for this FD (poll implementation). */
1486 continue;
1487 }
1488
1489 /*
1490 * Thread quit pipe has been triggered, flag that we should stop
1491 * but continue the current loop to handle potential data from
1492 * consumer.
1493 */
1494 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1495
1496 if (pollfd == sock) {
1497 /* Event on the consumerd socket */
1498 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1499 && !(revents & LPOLLIN)) {
1500 ERR("consumer err socket second poll error");
1501 goto error;
1502 }
1503 health_code_update();
1504 /* Wait for any kconsumerd error */
1505 ret = lttcomm_recv_unix_sock(sock, &code,
1506 sizeof(enum lttcomm_return_code));
1507 if (ret <= 0) {
1508 ERR("consumer closed the command socket");
1509 goto error;
1510 }
1511
1512 ERR("consumer return code : %s",
1513 lttcomm_get_readable_code(-code));
1514
1515 goto exit;
1516 } else if (pollfd == consumer_data->metadata_fd) {
1517 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1518 && !(revents & LPOLLIN)) {
1519 ERR("consumer err metadata socket second poll error");
1520 goto error;
1521 }
1522 /* UST metadata requests */
1523 ret = ust_consumer_metadata_request(
1524 &consumer_data->metadata_sock);
1525 if (ret < 0) {
1526 ERR("Handling metadata request");
1527 goto error;
1528 }
1529 }
1530 /* No need for an else branch all FDs are tested prior. */
1531 }
1532 health_code_update();
1533 }
1534
1535 exit:
1536 error:
1537 /*
1538 * We lock here because we are about to close the sockets and some other
1539 * thread might be using them so get exclusive access which will abort all
1540 * other consumer command by other threads.
1541 */
1542 pthread_mutex_lock(&consumer_data->lock);
1543
1544 /* Immediately set the consumerd state to stopped */
1545 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1546 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1547 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1548 consumer_data->type == LTTNG_CONSUMER32_UST) {
1549 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1550 } else {
1551 /* Code flow error... */
1552 assert(0);
1553 }
1554
1555 if (consumer_data->err_sock >= 0) {
1556 ret = close(consumer_data->err_sock);
1557 if (ret) {
1558 PERROR("close");
1559 }
1560 consumer_data->err_sock = -1;
1561 }
1562 if (consumer_data->cmd_sock >= 0) {
1563 ret = close(consumer_data->cmd_sock);
1564 if (ret) {
1565 PERROR("close");
1566 }
1567 consumer_data->cmd_sock = -1;
1568 }
1569 if (consumer_data->metadata_sock.fd_ptr &&
1570 *consumer_data->metadata_sock.fd_ptr >= 0) {
1571 ret = close(*consumer_data->metadata_sock.fd_ptr);
1572 if (ret) {
1573 PERROR("close");
1574 }
1575 }
1576 if (sock >= 0) {
1577 ret = close(sock);
1578 if (ret) {
1579 PERROR("close");
1580 }
1581 }
1582
1583 unlink(consumer_data->err_unix_sock_path);
1584 unlink(consumer_data->cmd_unix_sock_path);
1585 pthread_mutex_unlock(&consumer_data->lock);
1586
1587 /* Cleanup metadata socket mutex. */
1588 if (consumer_data->metadata_sock.lock) {
1589 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1590 free(consumer_data->metadata_sock.lock);
1591 }
1592 lttng_poll_clean(&events);
1593
1594 if (cmd_socket_wrapper) {
1595 consumer_destroy_socket(cmd_socket_wrapper);
1596 }
1597 error_poll:
1598 if (err) {
1599 health_error();
1600 ERR("Health error occurred in %s", __func__);
1601 }
1602 health_unregister(health_sessiond);
1603 DBG("consumer thread cleanup completed");
1604
1605 rcu_thread_offline();
1606 rcu_unregister_thread();
1607
1608 return NULL;
1609 }
1610
1611 /*
1612 * This thread manage application communication.
1613 */
1614 static void *thread_manage_apps(void *data)
1615 {
1616 int i, ret, pollfd, err = -1;
1617 ssize_t size_ret;
1618 uint32_t revents, nb_fd;
1619 struct lttng_poll_event events;
1620
1621 DBG("[thread] Manage application started");
1622
1623 rcu_register_thread();
1624 rcu_thread_online();
1625
1626 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1627
1628 if (testpoint(sessiond_thread_manage_apps)) {
1629 goto error_testpoint;
1630 }
1631
1632 health_code_update();
1633
1634 ret = sessiond_set_thread_pollset(&events, 2);
1635 if (ret < 0) {
1636 goto error_poll_create;
1637 }
1638
1639 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1640 if (ret < 0) {
1641 goto error;
1642 }
1643
1644 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1645 goto error;
1646 }
1647
1648 health_code_update();
1649
1650 while (1) {
1651 DBG("Apps thread polling");
1652
1653 /* Inifinite blocking call, waiting for transmission */
1654 restart:
1655 health_poll_entry();
1656 ret = lttng_poll_wait(&events, -1);
1657 DBG("Apps thread return from poll on %d fds",
1658 LTTNG_POLL_GETNB(&events));
1659 health_poll_exit();
1660 if (ret < 0) {
1661 /*
1662 * Restart interrupted system call.
1663 */
1664 if (errno == EINTR) {
1665 goto restart;
1666 }
1667 goto error;
1668 }
1669
1670 nb_fd = ret;
1671
1672 for (i = 0; i < nb_fd; i++) {
1673 /* Fetch once the poll data */
1674 revents = LTTNG_POLL_GETEV(&events, i);
1675 pollfd = LTTNG_POLL_GETFD(&events, i);
1676
1677 health_code_update();
1678
1679 if (!revents) {
1680 /* No activity for this FD (poll implementation). */
1681 continue;
1682 }
1683
1684 /* Thread quit pipe has been closed. Killing thread. */
1685 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1686 if (ret) {
1687 err = 0;
1688 goto exit;
1689 }
1690
1691 /* Inspect the apps cmd pipe */
1692 if (pollfd == apps_cmd_pipe[0]) {
1693 if (revents & LPOLLIN) {
1694 int sock;
1695
1696 /* Empty pipe */
1697 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1698 if (size_ret < sizeof(sock)) {
1699 PERROR("read apps cmd pipe");
1700 goto error;
1701 }
1702
1703 health_code_update();
1704
1705 /*
1706 * Since this is a command socket (write then read),
1707 * we only monitor the error events of the socket.
1708 */
1709 ret = lttng_poll_add(&events, sock,
1710 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1711 if (ret < 0) {
1712 goto error;
1713 }
1714
1715 DBG("Apps with sock %d added to poll set", sock);
1716 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1717 ERR("Apps command pipe error");
1718 goto error;
1719 } else {
1720 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1721 goto error;
1722 }
1723 } else {
1724 /*
1725 * At this point, we know that a registered application made
1726 * the event at poll_wait.
1727 */
1728 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1729 /* Removing from the poll set */
1730 ret = lttng_poll_del(&events, pollfd);
1731 if (ret < 0) {
1732 goto error;
1733 }
1734
1735 /* Socket closed on remote end. */
1736 ust_app_unregister(pollfd);
1737 } else {
1738 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1739 goto error;
1740 }
1741 }
1742
1743 health_code_update();
1744 }
1745 }
1746
1747 exit:
1748 error:
1749 lttng_poll_clean(&events);
1750 error_poll_create:
1751 error_testpoint:
1752 utils_close_pipe(apps_cmd_pipe);
1753 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1754
1755 /*
1756 * We don't clean the UST app hash table here since already registered
1757 * applications can still be controlled so let them be until the session
1758 * daemon dies or the applications stop.
1759 */
1760
1761 if (err) {
1762 health_error();
1763 ERR("Health error occurred in %s", __func__);
1764 }
1765 health_unregister(health_sessiond);
1766 DBG("Application communication apps thread cleanup complete");
1767 rcu_thread_offline();
1768 rcu_unregister_thread();
1769 return NULL;
1770 }
1771
1772 /*
1773 * Send a socket to a thread This is called from the dispatch UST registration
1774 * thread once all sockets are set for the application.
1775 *
1776 * The sock value can be invalid, we don't really care, the thread will handle
1777 * it and make the necessary cleanup if so.
1778 *
1779 * On success, return 0 else a negative value being the errno message of the
1780 * write().
1781 */
1782 static int send_socket_to_thread(int fd, int sock)
1783 {
1784 ssize_t ret;
1785
1786 /*
1787 * It's possible that the FD is set as invalid with -1 concurrently just
1788 * before calling this function being a shutdown state of the thread.
1789 */
1790 if (fd < 0) {
1791 ret = -EBADF;
1792 goto error;
1793 }
1794
1795 ret = lttng_write(fd, &sock, sizeof(sock));
1796 if (ret < sizeof(sock)) {
1797 PERROR("write apps pipe %d", fd);
1798 if (ret < 0) {
1799 ret = -errno;
1800 }
1801 goto error;
1802 }
1803
1804 /* All good. Don't send back the write positive ret value. */
1805 ret = 0;
1806 error:
1807 return (int) ret;
1808 }
1809
1810 /*
1811 * Sanitize the wait queue of the dispatch registration thread meaning removing
1812 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1813 * notify socket is never received.
1814 */
1815 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1816 {
1817 int ret, nb_fd = 0, i;
1818 unsigned int fd_added = 0;
1819 struct lttng_poll_event events;
1820 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1821
1822 assert(wait_queue);
1823
1824 lttng_poll_init(&events);
1825
1826 /* Just skip everything for an empty queue. */
1827 if (!wait_queue->count) {
1828 goto end;
1829 }
1830
1831 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1832 if (ret < 0) {
1833 goto error_create;
1834 }
1835
1836 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1837 &wait_queue->head, head) {
1838 assert(wait_node->app);
1839 ret = lttng_poll_add(&events, wait_node->app->sock,
1840 LPOLLHUP | LPOLLERR);
1841 if (ret < 0) {
1842 goto error;
1843 }
1844
1845 fd_added = 1;
1846 }
1847
1848 if (!fd_added) {
1849 goto end;
1850 }
1851
1852 /*
1853 * Poll but don't block so we can quickly identify the faulty events and
1854 * clean them afterwards from the wait queue.
1855 */
1856 ret = lttng_poll_wait(&events, 0);
1857 if (ret < 0) {
1858 goto error;
1859 }
1860 nb_fd = ret;
1861
1862 for (i = 0; i < nb_fd; i++) {
1863 /* Get faulty FD. */
1864 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1865 int pollfd = LTTNG_POLL_GETFD(&events, i);
1866
1867 if (!revents) {
1868 /* No activity for this FD (poll implementation). */
1869 continue;
1870 }
1871
1872 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1873 &wait_queue->head, head) {
1874 if (pollfd == wait_node->app->sock &&
1875 (revents & (LPOLLHUP | LPOLLERR))) {
1876 cds_list_del(&wait_node->head);
1877 wait_queue->count--;
1878 ust_app_destroy(wait_node->app);
1879 free(wait_node);
1880 /*
1881 * Silence warning of use-after-free in
1882 * cds_list_for_each_entry_safe which uses
1883 * __typeof__(*wait_node).
1884 */
1885 wait_node = NULL;
1886 break;
1887 } else {
1888 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1889 goto error;
1890 }
1891 }
1892 }
1893
1894 if (nb_fd > 0) {
1895 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1896 }
1897
1898 end:
1899 lttng_poll_clean(&events);
1900 return;
1901
1902 error:
1903 lttng_poll_clean(&events);
1904 error_create:
1905 ERR("Unable to sanitize wait queue");
1906 return;
1907 }
1908
1909 /*
1910 * Dispatch request from the registration threads to the application
1911 * communication thread.
1912 */
1913 static void *thread_dispatch_ust_registration(void *data)
1914 {
1915 int ret, err = -1;
1916 struct cds_wfcq_node *node;
1917 struct ust_command *ust_cmd = NULL;
1918 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1919 struct ust_reg_wait_queue wait_queue = {
1920 .count = 0,
1921 };
1922
1923 rcu_register_thread();
1924
1925 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1926
1927 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1928 goto error_testpoint;
1929 }
1930
1931 health_code_update();
1932
1933 CDS_INIT_LIST_HEAD(&wait_queue.head);
1934
1935 DBG("[thread] Dispatch UST command started");
1936
1937 for (;;) {
1938 health_code_update();
1939
1940 /* Atomically prepare the queue futex */
1941 futex_nto1_prepare(&ust_cmd_queue.futex);
1942
1943 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1944 break;
1945 }
1946
1947 do {
1948 struct ust_app *app = NULL;
1949 ust_cmd = NULL;
1950
1951 /*
1952 * Make sure we don't have node(s) that have hung up before receiving
1953 * the notify socket. This is to clean the list in order to avoid
1954 * memory leaks from notify socket that are never seen.
1955 */
1956 sanitize_wait_queue(&wait_queue);
1957
1958 health_code_update();
1959 /* Dequeue command for registration */
1960 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1961 if (node == NULL) {
1962 DBG("Woken up but nothing in the UST command queue");
1963 /* Continue thread execution */
1964 break;
1965 }
1966
1967 ust_cmd = caa_container_of(node, struct ust_command, node);
1968
1969 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1970 " gid:%d sock:%d name:%s (version %d.%d)",
1971 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1972 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1973 ust_cmd->sock, ust_cmd->reg_msg.name,
1974 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1975
1976 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1977 wait_node = zmalloc(sizeof(*wait_node));
1978 if (!wait_node) {
1979 PERROR("zmalloc wait_node dispatch");
1980 ret = close(ust_cmd->sock);
1981 if (ret < 0) {
1982 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1983 }
1984 lttng_fd_put(LTTNG_FD_APPS, 1);
1985 free(ust_cmd);
1986 goto error;
1987 }
1988 CDS_INIT_LIST_HEAD(&wait_node->head);
1989
1990 /* Create application object if socket is CMD. */
1991 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1992 ust_cmd->sock);
1993 if (!wait_node->app) {
1994 ret = close(ust_cmd->sock);
1995 if (ret < 0) {
1996 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1997 }
1998 lttng_fd_put(LTTNG_FD_APPS, 1);
1999 free(wait_node);
2000 free(ust_cmd);
2001 continue;
2002 }
2003 /*
2004 * Add application to the wait queue so we can set the notify
2005 * socket before putting this object in the global ht.
2006 */
2007 cds_list_add(&wait_node->head, &wait_queue.head);
2008 wait_queue.count++;
2009
2010 free(ust_cmd);
2011 /*
2012 * We have to continue here since we don't have the notify
2013 * socket and the application MUST be added to the hash table
2014 * only at that moment.
2015 */
2016 continue;
2017 } else {
2018 /*
2019 * Look for the application in the local wait queue and set the
2020 * notify socket if found.
2021 */
2022 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2023 &wait_queue.head, head) {
2024 health_code_update();
2025 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
2026 wait_node->app->notify_sock = ust_cmd->sock;
2027 cds_list_del(&wait_node->head);
2028 wait_queue.count--;
2029 app = wait_node->app;
2030 free(wait_node);
2031 DBG3("UST app notify socket %d is set", ust_cmd->sock);
2032 break;
2033 }
2034 }
2035
2036 /*
2037 * With no application at this stage the received socket is
2038 * basically useless so close it before we free the cmd data
2039 * structure for good.
2040 */
2041 if (!app) {
2042 ret = close(ust_cmd->sock);
2043 if (ret < 0) {
2044 PERROR("close ust sock dispatch %d", ust_cmd->sock);
2045 }
2046 lttng_fd_put(LTTNG_FD_APPS, 1);
2047 }
2048 free(ust_cmd);
2049 }
2050
2051 if (app) {
2052 /*
2053 * @session_lock_list
2054 *
2055 * Lock the global session list so from the register up to the
2056 * registration done message, no thread can see the application
2057 * and change its state.
2058 */
2059 session_lock_list();
2060 rcu_read_lock();
2061
2062 /*
2063 * Add application to the global hash table. This needs to be
2064 * done before the update to the UST registry can locate the
2065 * application.
2066 */
2067 ust_app_add(app);
2068
2069 /* Set app version. This call will print an error if needed. */
2070 (void) ust_app_version(app);
2071
2072 /* Send notify socket through the notify pipe. */
2073 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2074 app->notify_sock);
2075 if (ret < 0) {
2076 rcu_read_unlock();
2077 session_unlock_list();
2078 /*
2079 * No notify thread, stop the UST tracing. However, this is
2080 * not an internal error of the this thread thus setting
2081 * the health error code to a normal exit.
2082 */
2083 err = 0;
2084 goto error;
2085 }
2086
2087 /*
2088 * Update newly registered application with the tracing
2089 * registry info already enabled information.
2090 */
2091 update_ust_app(app->sock);
2092
2093 /*
2094 * Don't care about return value. Let the manage apps threads
2095 * handle app unregistration upon socket close.
2096 */
2097 (void) ust_app_register_done(app);
2098
2099 /*
2100 * Even if the application socket has been closed, send the app
2101 * to the thread and unregistration will take place at that
2102 * place.
2103 */
2104 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2105 if (ret < 0) {
2106 rcu_read_unlock();
2107 session_unlock_list();
2108 /*
2109 * No apps. thread, stop the UST tracing. However, this is
2110 * not an internal error of the this thread thus setting
2111 * the health error code to a normal exit.
2112 */
2113 err = 0;
2114 goto error;
2115 }
2116
2117 rcu_read_unlock();
2118 session_unlock_list();
2119 }
2120 } while (node != NULL);
2121
2122 health_poll_entry();
2123 /* Futex wait on queue. Blocking call on futex() */
2124 futex_nto1_wait(&ust_cmd_queue.futex);
2125 health_poll_exit();
2126 }
2127 /* Normal exit, no error */
2128 err = 0;
2129
2130 error:
2131 /* Clean up wait queue. */
2132 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2133 &wait_queue.head, head) {
2134 cds_list_del(&wait_node->head);
2135 wait_queue.count--;
2136 free(wait_node);
2137 }
2138
2139 /* Empty command queue. */
2140 for (;;) {
2141 /* Dequeue command for registration */
2142 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2143 if (node == NULL) {
2144 break;
2145 }
2146 ust_cmd = caa_container_of(node, struct ust_command, node);
2147 ret = close(ust_cmd->sock);
2148 if (ret < 0) {
2149 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2150 }
2151 lttng_fd_put(LTTNG_FD_APPS, 1);
2152 free(ust_cmd);
2153 }
2154
2155 error_testpoint:
2156 DBG("Dispatch thread dying");
2157 if (err) {
2158 health_error();
2159 ERR("Health error occurred in %s", __func__);
2160 }
2161 health_unregister(health_sessiond);
2162 rcu_unregister_thread();
2163 return NULL;
2164 }
2165
2166 /*
2167 * This thread manage application registration.
2168 */
2169 static void *thread_registration_apps(void *data)
2170 {
2171 int sock = -1, i, ret, pollfd, err = -1;
2172 uint32_t revents, nb_fd;
2173 struct lttng_poll_event events;
2174 /*
2175 * Get allocated in this thread, enqueued to a global queue, dequeued and
2176 * freed in the manage apps thread.
2177 */
2178 struct ust_command *ust_cmd = NULL;
2179
2180 DBG("[thread] Manage application registration started");
2181
2182 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2183
2184 if (testpoint(sessiond_thread_registration_apps)) {
2185 goto error_testpoint;
2186 }
2187
2188 ret = lttcomm_listen_unix_sock(apps_sock);
2189 if (ret < 0) {
2190 goto error_listen;
2191 }
2192
2193 /*
2194 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2195 * more will be added to this poll set.
2196 */
2197 ret = sessiond_set_thread_pollset(&events, 2);
2198 if (ret < 0) {
2199 goto error_create_poll;
2200 }
2201
2202 /* Add the application registration socket */
2203 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2204 if (ret < 0) {
2205 goto error_poll_add;
2206 }
2207
2208 /* Notify all applications to register */
2209 ret = notify_ust_apps(1);
2210 if (ret < 0) {
2211 ERR("Failed to notify applications or create the wait shared memory.\n"
2212 "Execution continues but there might be problem for already\n"
2213 "running applications that wishes to register.");
2214 }
2215
2216 while (1) {
2217 DBG("Accepting application registration");
2218
2219 /* Inifinite blocking call, waiting for transmission */
2220 restart:
2221 health_poll_entry();
2222 ret = lttng_poll_wait(&events, -1);
2223 health_poll_exit();
2224 if (ret < 0) {
2225 /*
2226 * Restart interrupted system call.
2227 */
2228 if (errno == EINTR) {
2229 goto restart;
2230 }
2231 goto error;
2232 }
2233
2234 nb_fd = ret;
2235
2236 for (i = 0; i < nb_fd; i++) {
2237 health_code_update();
2238
2239 /* Fetch once the poll data */
2240 revents = LTTNG_POLL_GETEV(&events, i);
2241 pollfd = LTTNG_POLL_GETFD(&events, i);
2242
2243 if (!revents) {
2244 /* No activity for this FD (poll implementation). */
2245 continue;
2246 }
2247
2248 /* Thread quit pipe has been closed. Killing thread. */
2249 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2250 if (ret) {
2251 err = 0;
2252 goto exit;
2253 }
2254
2255 /* Event on the registration socket */
2256 if (pollfd == apps_sock) {
2257 if (revents & LPOLLIN) {
2258 sock = lttcomm_accept_unix_sock(apps_sock);
2259 if (sock < 0) {
2260 goto error;
2261 }
2262
2263 /*
2264 * Set socket timeout for both receiving and ending.
2265 * app_socket_timeout is in seconds, whereas
2266 * lttcomm_setsockopt_rcv_timeout and
2267 * lttcomm_setsockopt_snd_timeout expect msec as
2268 * parameter.
2269 */
2270 if (app_socket_timeout >= 0) {
2271 (void) lttcomm_setsockopt_rcv_timeout(sock,
2272 app_socket_timeout * 1000);
2273 (void) lttcomm_setsockopt_snd_timeout(sock,
2274 app_socket_timeout * 1000);
2275 }
2276
2277 /*
2278 * Set the CLOEXEC flag. Return code is useless because
2279 * either way, the show must go on.
2280 */
2281 (void) utils_set_fd_cloexec(sock);
2282
2283 /* Create UST registration command for enqueuing */
2284 ust_cmd = zmalloc(sizeof(struct ust_command));
2285 if (ust_cmd == NULL) {
2286 PERROR("ust command zmalloc");
2287 ret = close(sock);
2288 if (ret) {
2289 PERROR("close");
2290 }
2291 goto error;
2292 }
2293
2294 /*
2295 * Using message-based transmissions to ensure we don't
2296 * have to deal with partially received messages.
2297 */
2298 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2299 if (ret < 0) {
2300 ERR("Exhausted file descriptors allowed for applications.");
2301 free(ust_cmd);
2302 ret = close(sock);
2303 if (ret) {
2304 PERROR("close");
2305 }
2306 sock = -1;
2307 continue;
2308 }
2309
2310 health_code_update();
2311 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2312 if (ret < 0) {
2313 free(ust_cmd);
2314 /* Close socket of the application. */
2315 ret = close(sock);
2316 if (ret) {
2317 PERROR("close");
2318 }
2319 lttng_fd_put(LTTNG_FD_APPS, 1);
2320 sock = -1;
2321 continue;
2322 }
2323 health_code_update();
2324
2325 ust_cmd->sock = sock;
2326 sock = -1;
2327
2328 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2329 " gid:%d sock:%d name:%s (version %d.%d)",
2330 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2331 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2332 ust_cmd->sock, ust_cmd->reg_msg.name,
2333 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2334
2335 /*
2336 * Lock free enqueue the registration request. The red pill
2337 * has been taken! This apps will be part of the *system*.
2338 */
2339 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2340
2341 /*
2342 * Wake the registration queue futex. Implicit memory
2343 * barrier with the exchange in cds_wfcq_enqueue.
2344 */
2345 futex_nto1_wake(&ust_cmd_queue.futex);
2346 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2347 ERR("Register apps socket poll error");
2348 goto error;
2349 } else {
2350 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2351 goto error;
2352 }
2353 }
2354 }
2355 }
2356
2357 exit:
2358 error:
2359 /* Notify that the registration thread is gone */
2360 notify_ust_apps(0);
2361
2362 if (apps_sock >= 0) {
2363 ret = close(apps_sock);
2364 if (ret) {
2365 PERROR("close");
2366 }
2367 }
2368 if (sock >= 0) {
2369 ret = close(sock);
2370 if (ret) {
2371 PERROR("close");
2372 }
2373 lttng_fd_put(LTTNG_FD_APPS, 1);
2374 }
2375 unlink(apps_unix_sock_path);
2376
2377 error_poll_add:
2378 lttng_poll_clean(&events);
2379 error_listen:
2380 error_create_poll:
2381 error_testpoint:
2382 DBG("UST Registration thread cleanup complete");
2383 if (err) {
2384 health_error();
2385 ERR("Health error occurred in %s", __func__);
2386 }
2387 health_unregister(health_sessiond);
2388
2389 return NULL;
2390 }
2391
2392 /*
2393 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2394 * exec or it will fails.
2395 */
2396 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2397 {
2398 int ret, clock_ret;
2399 struct timespec timeout;
2400
2401 /*
2402 * Make sure we set the readiness flag to 0 because we are NOT ready.
2403 * This access to consumer_thread_is_ready does not need to be
2404 * protected by consumer_data.cond_mutex (yet) since the consumer
2405 * management thread has not been started at this point.
2406 */
2407 consumer_data->consumer_thread_is_ready = 0;
2408
2409 /* Setup pthread condition */
2410 ret = pthread_condattr_init(&consumer_data->condattr);
2411 if (ret) {
2412 errno = ret;
2413 PERROR("pthread_condattr_init consumer data");
2414 goto error;
2415 }
2416
2417 /*
2418 * Set the monotonic clock in order to make sure we DO NOT jump in time
2419 * between the clock_gettime() call and the timedwait call. See bug #324
2420 * for a more details and how we noticed it.
2421 */
2422 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2423 if (ret) {
2424 errno = ret;
2425 PERROR("pthread_condattr_setclock consumer data");
2426 goto error;
2427 }
2428
2429 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2430 if (ret) {
2431 errno = ret;
2432 PERROR("pthread_cond_init consumer data");
2433 goto error;
2434 }
2435
2436 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2437 thread_manage_consumer, consumer_data);
2438 if (ret) {
2439 errno = ret;
2440 PERROR("pthread_create consumer");
2441 ret = -1;
2442 goto error;
2443 }
2444
2445 /* We are about to wait on a pthread condition */
2446 pthread_mutex_lock(&consumer_data->cond_mutex);
2447
2448 /* Get time for sem_timedwait absolute timeout */
2449 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2450 /*
2451 * Set the timeout for the condition timed wait even if the clock gettime
2452 * call fails since we might loop on that call and we want to avoid to
2453 * increment the timeout too many times.
2454 */
2455 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2456
2457 /*
2458 * The following loop COULD be skipped in some conditions so this is why we
2459 * set ret to 0 in order to make sure at least one round of the loop is
2460 * done.
2461 */
2462 ret = 0;
2463
2464 /*
2465 * Loop until the condition is reached or when a timeout is reached. Note
2466 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2467 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2468 * possible. This loop does not take any chances and works with both of
2469 * them.
2470 */
2471 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2472 if (clock_ret < 0) {
2473 PERROR("clock_gettime spawn consumer");
2474 /* Infinite wait for the consumerd thread to be ready */
2475 ret = pthread_cond_wait(&consumer_data->cond,
2476 &consumer_data->cond_mutex);
2477 } else {
2478 ret = pthread_cond_timedwait(&consumer_data->cond,
2479 &consumer_data->cond_mutex, &timeout);
2480 }
2481 }
2482
2483 /* Release the pthread condition */
2484 pthread_mutex_unlock(&consumer_data->cond_mutex);
2485
2486 if (ret != 0) {
2487 errno = ret;
2488 if (ret == ETIMEDOUT) {
2489 int pth_ret;
2490
2491 /*
2492 * Call has timed out so we kill the kconsumerd_thread and return
2493 * an error.
2494 */
2495 ERR("Condition timed out. The consumer thread was never ready."
2496 " Killing it");
2497 pth_ret = pthread_cancel(consumer_data->thread);
2498 if (pth_ret < 0) {
2499 PERROR("pthread_cancel consumer thread");
2500 }
2501 } else {
2502 PERROR("pthread_cond_wait failed consumer thread");
2503 }
2504 /* Caller is expecting a negative value on failure. */
2505 ret = -1;
2506 goto error;
2507 }
2508
2509 pthread_mutex_lock(&consumer_data->pid_mutex);
2510 if (consumer_data->pid == 0) {
2511 ERR("Consumerd did not start");
2512 pthread_mutex_unlock(&consumer_data->pid_mutex);
2513 goto error;
2514 }
2515 pthread_mutex_unlock(&consumer_data->pid_mutex);
2516
2517 return 0;
2518
2519 error:
2520 return ret;
2521 }
2522
2523 /*
2524 * Join consumer thread
2525 */
2526 static int join_consumer_thread(struct consumer_data *consumer_data)
2527 {
2528 void *status;
2529
2530 /* Consumer pid must be a real one. */
2531 if (consumer_data->pid > 0) {
2532 int ret;
2533 ret = kill(consumer_data->pid, SIGTERM);
2534 if (ret) {
2535 PERROR("Error killing consumer daemon");
2536 return ret;
2537 }
2538 return pthread_join(consumer_data->thread, &status);
2539 } else {
2540 return 0;
2541 }
2542 }
2543
2544 /*
2545 * Fork and exec a consumer daemon (consumerd).
2546 *
2547 * Return pid if successful else -1.
2548 */
2549 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2550 {
2551 int ret;
2552 pid_t pid;
2553 const char *consumer_to_use;
2554 const char *verbosity;
2555 struct stat st;
2556
2557 DBG("Spawning consumerd");
2558
2559 pid = fork();
2560 if (pid == 0) {
2561 /*
2562 * Exec consumerd.
2563 */
2564 if (opt_verbose_consumer) {
2565 verbosity = "--verbose";
2566 } else if (lttng_opt_quiet) {
2567 verbosity = "--quiet";
2568 } else {
2569 verbosity = "";
2570 }
2571
2572 switch (consumer_data->type) {
2573 case LTTNG_CONSUMER_KERNEL:
2574 /*
2575 * Find out which consumerd to execute. We will first try the
2576 * 64-bit path, then the sessiond's installation directory, and
2577 * fallback on the 32-bit one,
2578 */
2579 DBG3("Looking for a kernel consumer at these locations:");
2580 DBG3(" 1) %s", consumerd64_bin);
2581 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2582 DBG3(" 3) %s", consumerd32_bin);
2583 if (stat(consumerd64_bin, &st) == 0) {
2584 DBG3("Found location #1");
2585 consumer_to_use = consumerd64_bin;
2586 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2587 DBG3("Found location #2");
2588 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2589 } else if (stat(consumerd32_bin, &st) == 0) {
2590 DBG3("Found location #3");
2591 consumer_to_use = consumerd32_bin;
2592 } else {
2593 DBG("Could not find any valid consumerd executable");
2594 ret = -EINVAL;
2595 goto error;
2596 }
2597 DBG("Using kernel consumer at: %s", consumer_to_use);
2598 (void) execl(consumer_to_use,
2599 "lttng-consumerd", verbosity, "-k",
2600 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2601 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2602 "--group", tracing_group_name,
2603 NULL);
2604 break;
2605 case LTTNG_CONSUMER64_UST:
2606 {
2607 char *tmpnew = NULL;
2608
2609 if (consumerd64_libdir[0] != '\0') {
2610 char *tmp;
2611 size_t tmplen;
2612
2613 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2614 if (!tmp) {
2615 tmp = "";
2616 }
2617 tmplen = strlen("LD_LIBRARY_PATH=")
2618 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2619 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2620 if (!tmpnew) {
2621 ret = -ENOMEM;
2622 goto error;
2623 }
2624 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2625 strcat(tmpnew, consumerd64_libdir);
2626 if (tmp[0] != '\0') {
2627 strcat(tmpnew, ":");
2628 strcat(tmpnew, tmp);
2629 }
2630 ret = putenv(tmpnew);
2631 if (ret) {
2632 ret = -errno;
2633 free(tmpnew);
2634 goto error;
2635 }
2636 }
2637 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2638 (void) execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2639 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2640 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2641 "--group", tracing_group_name,
2642 NULL);
2643 if (consumerd64_libdir[0] != '\0') {
2644 free(tmpnew);
2645 }
2646 break;
2647 }
2648 case LTTNG_CONSUMER32_UST:
2649 {
2650 char *tmpnew = NULL;
2651
2652 if (consumerd32_libdir[0] != '\0') {
2653 char *tmp;
2654 size_t tmplen;
2655
2656 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2657 if (!tmp) {
2658 tmp = "";
2659 }
2660 tmplen = strlen("LD_LIBRARY_PATH=")
2661 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2662 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2663 if (!tmpnew) {
2664 ret = -ENOMEM;
2665 goto error;
2666 }
2667 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2668 strcat(tmpnew, consumerd32_libdir);
2669 if (tmp[0] != '\0') {
2670 strcat(tmpnew, ":");
2671 strcat(tmpnew, tmp);
2672 }
2673 ret = putenv(tmpnew);
2674 if (ret) {
2675 ret = -errno;
2676 free(tmpnew);
2677 goto error;
2678 }
2679 }
2680 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2681 (void) execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2682 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2683 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2684 "--group", tracing_group_name,
2685 NULL);
2686 if (consumerd32_libdir[0] != '\0') {
2687 free(tmpnew);
2688 }
2689 break;
2690 }
2691 default:
2692 PERROR("unknown consumer type");
2693 exit(EXIT_FAILURE);
2694 }
2695 if (errno != 0) {
2696 PERROR("Consumer execl()");
2697 }
2698 /* Reaching this point, we got a failure on our execl(). */
2699 exit(EXIT_FAILURE);
2700 } else if (pid > 0) {
2701 ret = pid;
2702 } else {
2703 PERROR("start consumer fork");
2704 ret = -errno;
2705 }
2706 error:
2707 return ret;
2708 }
2709
2710 /*
2711 * Spawn the consumerd daemon and session daemon thread.
2712 */
2713 static int start_consumerd(struct consumer_data *consumer_data)
2714 {
2715 int ret;
2716
2717 /*
2718 * Set the listen() state on the socket since there is a possible race
2719 * between the exec() of the consumer daemon and this call if place in the
2720 * consumer thread. See bug #366 for more details.
2721 */
2722 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2723 if (ret < 0) {
2724 goto error;
2725 }
2726
2727 pthread_mutex_lock(&consumer_data->pid_mutex);
2728 if (consumer_data->pid != 0) {
2729 pthread_mutex_unlock(&consumer_data->pid_mutex);
2730 goto end;
2731 }
2732
2733 ret = spawn_consumerd(consumer_data);
2734 if (ret < 0) {
2735 ERR("Spawning consumerd failed");
2736 pthread_mutex_unlock(&consumer_data->pid_mutex);
2737 goto error;
2738 }
2739
2740 /* Setting up the consumer_data pid */
2741 consumer_data->pid = ret;
2742 DBG2("Consumer pid %d", consumer_data->pid);
2743 pthread_mutex_unlock(&consumer_data->pid_mutex);
2744
2745 DBG2("Spawning consumer control thread");
2746 ret = spawn_consumer_thread(consumer_data);
2747 if (ret < 0) {
2748 ERR("Fatal error spawning consumer control thread");
2749 goto error;
2750 }
2751
2752 end:
2753 return 0;
2754
2755 error:
2756 /* Cleanup already created sockets on error. */
2757 if (consumer_data->err_sock >= 0) {
2758 int err;
2759
2760 err = close(consumer_data->err_sock);
2761 if (err < 0) {
2762 PERROR("close consumer data error socket");
2763 }
2764 }
2765 return ret;
2766 }
2767
2768 /*
2769 * Setup necessary data for kernel tracer action.
2770 */
2771 static int init_kernel_tracer(void)
2772 {
2773 int ret;
2774
2775 /* Modprobe lttng kernel modules */
2776 ret = modprobe_lttng_control();
2777 if (ret < 0) {
2778 goto error;
2779 }
2780
2781 /* Open debugfs lttng */
2782 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2783 if (kernel_tracer_fd < 0) {
2784 DBG("Failed to open %s", module_proc_lttng);
2785 goto error_open;
2786 }
2787
2788 /* Validate kernel version */
2789 ret = kernel_validate_version(kernel_tracer_fd);
2790 if (ret < 0) {
2791 goto error_version;
2792 }
2793
2794 ret = modprobe_lttng_data();
2795 if (ret < 0) {
2796 goto error_modules;
2797 }
2798
2799 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2800 kernel_tracer_fd);
2801 if (ret < 0) {
2802 goto error_modules;
2803 }
2804
2805 if (ret < 1) {
2806 WARN("Kernel tracer does not support buffer monitoring. "
2807 "The monitoring timer of channels in the kernel domain "
2808 "will be set to 0 (disabled).");
2809 }
2810
2811 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2812 return 0;
2813
2814 error_version:
2815 modprobe_remove_lttng_control();
2816 ret = close(kernel_tracer_fd);
2817 if (ret) {
2818 PERROR("close");
2819 }
2820 kernel_tracer_fd = -1;
2821 return LTTNG_ERR_KERN_VERSION;
2822
2823 error_modules:
2824 ret = close(kernel_tracer_fd);
2825 if (ret) {
2826 PERROR("close");
2827 }
2828
2829 error_open:
2830 modprobe_remove_lttng_control();
2831
2832 error:
2833 WARN("No kernel tracer available");
2834 kernel_tracer_fd = -1;
2835 if (!is_root) {
2836 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2837 } else {
2838 return LTTNG_ERR_KERN_NA;
2839 }
2840 }
2841
2842
2843 /*
2844 * Copy consumer output from the tracing session to the domain session. The
2845 * function also applies the right modification on a per domain basis for the
2846 * trace files destination directory.
2847 *
2848 * Should *NOT* be called with RCU read-side lock held.
2849 */
2850 static int copy_session_consumer(int domain, struct ltt_session *session)
2851 {
2852 int ret;
2853 const char *dir_name;
2854 struct consumer_output *consumer;
2855
2856 assert(session);
2857 assert(session->consumer);
2858
2859 switch (domain) {
2860 case LTTNG_DOMAIN_KERNEL:
2861 DBG3("Copying tracing session consumer output in kernel session");
2862 /*
2863 * XXX: We should audit the session creation and what this function
2864 * does "extra" in order to avoid a destroy since this function is used
2865 * in the domain session creation (kernel and ust) only. Same for UST
2866 * domain.
2867 */
2868 if (session->kernel_session->consumer) {
2869 consumer_output_put(session->kernel_session->consumer);
2870 }
2871 session->kernel_session->consumer =
2872 consumer_copy_output(session->consumer);
2873 /* Ease our life a bit for the next part */
2874 consumer = session->kernel_session->consumer;
2875 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2876 break;
2877 case LTTNG_DOMAIN_JUL:
2878 case LTTNG_DOMAIN_LOG4J:
2879 case LTTNG_DOMAIN_PYTHON:
2880 case LTTNG_DOMAIN_UST:
2881 DBG3("Copying tracing session consumer output in UST session");
2882 if (session->ust_session->consumer) {
2883 consumer_output_put(session->ust_session->consumer);
2884 }
2885 session->ust_session->consumer =
2886 consumer_copy_output(session->consumer);
2887 /* Ease our life a bit for the next part */
2888 consumer = session->ust_session->consumer;
2889 dir_name = DEFAULT_UST_TRACE_DIR;
2890 break;
2891 default:
2892 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2893 goto error;
2894 }
2895
2896 /* Append correct directory to subdir */
2897 strncat(consumer->subdir, dir_name,
2898 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2899 DBG3("Copy session consumer subdir %s", consumer->subdir);
2900
2901 ret = LTTNG_OK;
2902
2903 error:
2904 return ret;
2905 }
2906
2907 /*
2908 * Create an UST session and add it to the session ust list.
2909 *
2910 * Should *NOT* be called with RCU read-side lock held.
2911 */
2912 static int create_ust_session(struct ltt_session *session,
2913 struct lttng_domain *domain)
2914 {
2915 int ret;
2916 struct ltt_ust_session *lus = NULL;
2917
2918 assert(session);
2919 assert(domain);
2920 assert(session->consumer);
2921
2922 switch (domain->type) {
2923 case LTTNG_DOMAIN_JUL:
2924 case LTTNG_DOMAIN_LOG4J:
2925 case LTTNG_DOMAIN_PYTHON:
2926 case LTTNG_DOMAIN_UST:
2927 break;
2928 default:
2929 ERR("Unknown UST domain on create session %d", domain->type);
2930 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2931 goto error;
2932 }
2933
2934 DBG("Creating UST session");
2935
2936 lus = trace_ust_create_session(session->id);
2937 if (lus == NULL) {
2938 ret = LTTNG_ERR_UST_SESS_FAIL;
2939 goto error;
2940 }
2941
2942 lus->uid = session->uid;
2943 lus->gid = session->gid;
2944 lus->output_traces = session->output_traces;
2945 lus->snapshot_mode = session->snapshot_mode;
2946 lus->live_timer_interval = session->live_timer;
2947 session->ust_session = lus;
2948 if (session->shm_path[0]) {
2949 strncpy(lus->root_shm_path, session->shm_path,
2950 sizeof(lus->root_shm_path));
2951 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2952 strncpy(lus->shm_path, session->shm_path,
2953 sizeof(lus->shm_path));
2954 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2955 strncat(lus->shm_path, "/ust",
2956 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2957 }
2958 /* Copy session output to the newly created UST session */
2959 ret = copy_session_consumer(domain->type, session);
2960 if (ret != LTTNG_OK) {
2961 goto error;
2962 }
2963
2964 return LTTNG_OK;
2965
2966 error:
2967 free(lus);
2968 session->ust_session = NULL;
2969 return ret;
2970 }
2971
2972 /*
2973 * Create a kernel tracer session then create the default channel.
2974 */
2975 static int create_kernel_session(struct ltt_session *session)
2976 {
2977 int ret;
2978
2979 DBG("Creating kernel session");
2980
2981 ret = kernel_create_session(session, kernel_tracer_fd);
2982 if (ret < 0) {
2983 ret = LTTNG_ERR_KERN_SESS_FAIL;
2984 goto error;
2985 }
2986
2987 /* Code flow safety */
2988 assert(session->kernel_session);
2989
2990 /* Copy session output to the newly created Kernel session */
2991 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2992 if (ret != LTTNG_OK) {
2993 goto error;
2994 }
2995
2996 /* Create directory(ies) on local filesystem. */
2997 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2998 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2999 ret = run_as_mkdir_recursive(
3000 session->kernel_session->consumer->dst.trace_path,
3001 S_IRWXU | S_IRWXG, session->uid, session->gid);
3002 if (ret < 0) {
3003 if (errno != EEXIST) {
3004 ERR("Trace directory creation error");
3005 goto error;
3006 }
3007 }
3008 }
3009
3010 session->kernel_session->uid = session->uid;
3011 session->kernel_session->gid = session->gid;
3012 session->kernel_session->output_traces = session->output_traces;
3013 session->kernel_session->snapshot_mode = session->snapshot_mode;
3014
3015 return LTTNG_OK;
3016
3017 error:
3018 trace_kernel_destroy_session(session->kernel_session);
3019 session->kernel_session = NULL;
3020 return ret;
3021 }
3022
3023 /*
3024 * Count number of session permitted by uid/gid.
3025 */
3026 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
3027 {
3028 unsigned int i = 0;
3029 struct ltt_session *session;
3030
3031 DBG("Counting number of available session for UID %d GID %d",
3032 uid, gid);
3033 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
3034 /*
3035 * Only list the sessions the user can control.
3036 */
3037 if (!session_access_ok(session, uid, gid)) {
3038 continue;
3039 }
3040 i++;
3041 }
3042 return i;
3043 }
3044
3045 /*
3046 * Process the command requested by the lttng client within the command
3047 * context structure. This function make sure that the return structure (llm)
3048 * is set and ready for transmission before returning.
3049 *
3050 * Return any error encountered or 0 for success.
3051 *
3052 * "sock" is only used for special-case var. len data.
3053 *
3054 * Should *NOT* be called with RCU read-side lock held.
3055 */
3056 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
3057 int *sock_error)
3058 {
3059 int ret = LTTNG_OK;
3060 int need_tracing_session = 1;
3061 int need_domain;
3062
3063 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
3064
3065 assert(!rcu_read_ongoing());
3066
3067 *sock_error = 0;
3068
3069 switch (cmd_ctx->lsm->cmd_type) {
3070 case LTTNG_CREATE_SESSION:
3071 case LTTNG_CREATE_SESSION_SNAPSHOT:
3072 case LTTNG_CREATE_SESSION_LIVE:
3073 case LTTNG_DESTROY_SESSION:
3074 case LTTNG_LIST_SESSIONS:
3075 case LTTNG_LIST_DOMAINS:
3076 case LTTNG_START_TRACE:
3077 case LTTNG_STOP_TRACE:
3078 case LTTNG_DATA_PENDING:
3079 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3080 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3081 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3082 case LTTNG_SNAPSHOT_RECORD:
3083 case LTTNG_SAVE_SESSION:
3084 case LTTNG_SET_SESSION_SHM_PATH:
3085 case LTTNG_REGENERATE_METADATA:
3086 case LTTNG_REGENERATE_STATEDUMP:
3087 case LTTNG_REGISTER_TRIGGER:
3088 case LTTNG_UNREGISTER_TRIGGER:
3089 need_domain = 0;
3090 break;
3091 default:
3092 need_domain = 1;
3093 }
3094
3095 if (opt_no_kernel && need_domain
3096 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3097 if (!is_root) {
3098 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3099 } else {
3100 ret = LTTNG_ERR_KERN_NA;
3101 }
3102 goto error;
3103 }
3104
3105 /* Deny register consumer if we already have a spawned consumer. */
3106 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3107 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3108 if (kconsumer_data.pid > 0) {
3109 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3110 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3111 goto error;
3112 }
3113 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3114 }
3115
3116 /*
3117 * Check for command that don't needs to allocate a returned payload. We do
3118 * this here so we don't have to make the call for no payload at each
3119 * command.
3120 */
3121 switch(cmd_ctx->lsm->cmd_type) {
3122 case LTTNG_LIST_SESSIONS:
3123 case LTTNG_LIST_TRACEPOINTS:
3124 case LTTNG_LIST_TRACEPOINT_FIELDS:
3125 case LTTNG_LIST_DOMAINS:
3126 case LTTNG_LIST_CHANNELS:
3127 case LTTNG_LIST_EVENTS:
3128 case LTTNG_LIST_SYSCALLS:
3129 case LTTNG_LIST_TRACKER_PIDS:
3130 case LTTNG_DATA_PENDING:
3131 break;
3132 default:
3133 /* Setup lttng message with no payload */
3134 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3135 if (ret < 0) {
3136 /* This label does not try to unlock the session */
3137 goto init_setup_error;
3138 }
3139 }
3140
3141 /* Commands that DO NOT need a session. */
3142 switch (cmd_ctx->lsm->cmd_type) {
3143 case LTTNG_CREATE_SESSION:
3144 case LTTNG_CREATE_SESSION_SNAPSHOT:
3145 case LTTNG_CREATE_SESSION_LIVE:
3146 case LTTNG_LIST_SESSIONS:
3147 case LTTNG_LIST_TRACEPOINTS:
3148 case LTTNG_LIST_SYSCALLS:
3149 case LTTNG_LIST_TRACEPOINT_FIELDS:
3150 case LTTNG_SAVE_SESSION:
3151 case LTTNG_REGISTER_TRIGGER:
3152 case LTTNG_UNREGISTER_TRIGGER:
3153 need_tracing_session = 0;
3154 break;
3155 default:
3156 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3157 /*
3158 * We keep the session list lock across _all_ commands
3159 * for now, because the per-session lock does not
3160 * handle teardown properly.
3161 */
3162 session_lock_list();
3163 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3164 if (cmd_ctx->session == NULL) {
3165 ret = LTTNG_ERR_SESS_NOT_FOUND;
3166 goto error;
3167 } else {
3168 /* Acquire lock for the session */
3169 session_lock(cmd_ctx->session);
3170 }
3171 break;
3172 }
3173
3174 /*
3175 * Commands that need a valid session but should NOT create one if none
3176 * exists. Instead of creating one and destroying it when the command is
3177 * handled, process that right before so we save some round trip in useless
3178 * code path.
3179 */
3180 switch (cmd_ctx->lsm->cmd_type) {
3181 case LTTNG_DISABLE_CHANNEL:
3182 case LTTNG_DISABLE_EVENT:
3183 switch (cmd_ctx->lsm->domain.type) {
3184 case LTTNG_DOMAIN_KERNEL:
3185 if (!cmd_ctx->session->kernel_session) {
3186 ret = LTTNG_ERR_NO_CHANNEL;
3187 goto error;
3188 }
3189 break;
3190 case LTTNG_DOMAIN_JUL:
3191 case LTTNG_DOMAIN_LOG4J:
3192 case LTTNG_DOMAIN_PYTHON:
3193 case LTTNG_DOMAIN_UST:
3194 if (!cmd_ctx->session->ust_session) {
3195 ret = LTTNG_ERR_NO_CHANNEL;
3196 goto error;
3197 }
3198 break;
3199 default:
3200 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3201 goto error;
3202 }
3203 default:
3204 break;
3205 }
3206
3207 if (!need_domain) {
3208 goto skip_domain;
3209 }
3210
3211 /*
3212 * Check domain type for specific "pre-action".
3213 */
3214 switch (cmd_ctx->lsm->domain.type) {
3215 case LTTNG_DOMAIN_KERNEL:
3216 if (!is_root) {
3217 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3218 goto error;
3219 }
3220
3221 /* Kernel tracer check */
3222 if (kernel_tracer_fd == -1) {
3223 /* Basically, load kernel tracer modules */
3224 ret = init_kernel_tracer();
3225 if (ret != 0) {
3226 goto error;
3227 }
3228 }
3229
3230 /* Consumer is in an ERROR state. Report back to client */
3231 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3232 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3233 goto error;
3234 }
3235
3236 /* Need a session for kernel command */
3237 if (need_tracing_session) {
3238 if (cmd_ctx->session->kernel_session == NULL) {
3239 ret = create_kernel_session(cmd_ctx->session);
3240 if (ret < 0) {
3241 ret = LTTNG_ERR_KERN_SESS_FAIL;
3242 goto error;
3243 }
3244 }
3245
3246 /* Start the kernel consumer daemon */
3247 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3248 if (kconsumer_data.pid == 0 &&
3249 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3250 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3251 ret = start_consumerd(&kconsumer_data);
3252 if (ret < 0) {
3253 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3254 goto error;
3255 }
3256 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3257 } else {
3258 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3259 }
3260
3261 /*
3262 * The consumer was just spawned so we need to add the socket to
3263 * the consumer output of the session if exist.
3264 */
3265 ret = consumer_create_socket(&kconsumer_data,
3266 cmd_ctx->session->kernel_session->consumer);
3267 if (ret < 0) {
3268 goto error;
3269 }
3270 }
3271
3272 break;
3273 case LTTNG_DOMAIN_JUL:
3274 case LTTNG_DOMAIN_LOG4J:
3275 case LTTNG_DOMAIN_PYTHON:
3276 case LTTNG_DOMAIN_UST:
3277 {
3278 if (!ust_app_supported()) {
3279 ret = LTTNG_ERR_NO_UST;
3280 goto error;
3281 }
3282 /* Consumer is in an ERROR state. Report back to client */
3283 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3284 ret = LTTNG_ERR_NO_USTCONSUMERD;
3285 goto error;
3286 }
3287
3288 if (need_tracing_session) {
3289 /* Create UST session if none exist. */
3290 if (cmd_ctx->session->ust_session == NULL) {
3291 ret = create_ust_session(cmd_ctx->session,
3292 &cmd_ctx->lsm->domain);
3293 if (ret != LTTNG_OK) {
3294 goto error;
3295 }
3296 }
3297
3298 /* Start the UST consumer daemons */
3299 /* 64-bit */
3300 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3301 if (consumerd64_bin[0] != '\0' &&
3302 ustconsumer64_data.pid == 0 &&
3303 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3304 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3305 ret = start_consumerd(&ustconsumer64_data);
3306 if (ret < 0) {
3307 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3308 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3309 goto error;
3310 }
3311
3312 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3313 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3314 } else {
3315 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3316 }
3317
3318 /*
3319 * Setup socket for consumer 64 bit. No need for atomic access
3320 * since it was set above and can ONLY be set in this thread.
3321 */
3322 ret = consumer_create_socket(&ustconsumer64_data,
3323 cmd_ctx->session->ust_session->consumer);
3324 if (ret < 0) {
3325 goto error;
3326 }
3327
3328 /* 32-bit */
3329 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3330 if (consumerd32_bin[0] != '\0' &&
3331 ustconsumer32_data.pid == 0 &&
3332 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3333 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3334 ret = start_consumerd(&ustconsumer32_data);
3335 if (ret < 0) {
3336 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3337 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3338 goto error;
3339 }
3340
3341 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3342 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3343 } else {
3344 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3345 }
3346
3347 /*
3348 * Setup socket for consumer 64 bit. No need for atomic access
3349 * since it was set above and can ONLY be set in this thread.
3350 */
3351 ret = consumer_create_socket(&ustconsumer32_data,
3352 cmd_ctx->session->ust_session->consumer);
3353 if (ret < 0) {
3354 goto error;
3355 }
3356 }
3357 break;
3358 }
3359 default:
3360 break;
3361 }
3362 skip_domain:
3363
3364 /* Validate consumer daemon state when start/stop trace command */
3365 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3366 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3367 switch (cmd_ctx->lsm->domain.type) {
3368 case LTTNG_DOMAIN_NONE:
3369 break;
3370 case LTTNG_DOMAIN_JUL:
3371 case LTTNG_DOMAIN_LOG4J:
3372 case LTTNG_DOMAIN_PYTHON:
3373 case LTTNG_DOMAIN_UST:
3374 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3375 ret = LTTNG_ERR_NO_USTCONSUMERD;
3376 goto error;
3377 }
3378 break;
3379 case LTTNG_DOMAIN_KERNEL:
3380 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3381 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3382 goto error;
3383 }
3384 break;
3385 default:
3386 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3387 goto error;
3388 }
3389 }
3390
3391 /*
3392 * Check that the UID or GID match that of the tracing session.
3393 * The root user can interact with all sessions.
3394 */
3395 if (need_tracing_session) {
3396 if (!session_access_ok(cmd_ctx->session,
3397 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3398 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3399 ret = LTTNG_ERR_EPERM;
3400 goto error;
3401 }
3402 }
3403
3404 /*
3405 * Send relayd information to consumer as soon as we have a domain and a
3406 * session defined.
3407 */
3408 if (cmd_ctx->session && need_domain) {
3409 /*
3410 * Setup relayd if not done yet. If the relayd information was already
3411 * sent to the consumer, this call will gracefully return.
3412 */
3413 ret = cmd_setup_relayd(cmd_ctx->session);
3414 if (ret != LTTNG_OK) {
3415 goto error;
3416 }
3417 }
3418
3419 /* Process by command type */
3420 switch (cmd_ctx->lsm->cmd_type) {
3421 case LTTNG_ADD_CONTEXT:
3422 {
3423 /*
3424 * An LTTNG_ADD_CONTEXT command might have a supplementary
3425 * payload if the context being added is an application context.
3426 */
3427 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3428 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3429 char *provider_name = NULL, *context_name = NULL;
3430 size_t provider_name_len =
3431 cmd_ctx->lsm->u.context.provider_name_len;
3432 size_t context_name_len =
3433 cmd_ctx->lsm->u.context.context_name_len;
3434
3435 if (provider_name_len == 0 || context_name_len == 0) {
3436 /*
3437 * Application provider and context names MUST
3438 * be provided.
3439 */
3440 ret = -LTTNG_ERR_INVALID;
3441 goto error;
3442 }
3443
3444 provider_name = zmalloc(provider_name_len + 1);
3445 if (!provider_name) {
3446 ret = -LTTNG_ERR_NOMEM;
3447 goto error;
3448 }
3449 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3450 provider_name;
3451
3452 context_name = zmalloc(context_name_len + 1);
3453 if (!context_name) {
3454 ret = -LTTNG_ERR_NOMEM;
3455 goto error_add_context;
3456 }
3457 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3458 context_name;
3459
3460 ret = lttcomm_recv_unix_sock(sock, provider_name,
3461 provider_name_len);
3462 if (ret < 0) {
3463 goto error_add_context;
3464 }
3465
3466 ret = lttcomm_recv_unix_sock(sock, context_name,
3467 context_name_len);
3468 if (ret < 0) {
3469 goto error_add_context;
3470 }
3471 }
3472
3473 /*
3474 * cmd_add_context assumes ownership of the provider and context
3475 * names.
3476 */
3477 ret = cmd_add_context(cmd_ctx->session,
3478 cmd_ctx->lsm->domain.type,
3479 cmd_ctx->lsm->u.context.channel_name,
3480 &cmd_ctx->lsm->u.context.ctx,
3481 kernel_poll_pipe[1]);
3482
3483 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3484 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3485 error_add_context:
3486 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3487 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3488 if (ret < 0) {
3489 goto error;
3490 }
3491 break;
3492 }
3493 case LTTNG_DISABLE_CHANNEL:
3494 {
3495 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3496 cmd_ctx->lsm->u.disable.channel_name);
3497 break;
3498 }
3499 case LTTNG_DISABLE_EVENT:
3500 {
3501
3502 /*
3503 * FIXME: handle filter; for now we just receive the filter's
3504 * bytecode along with the filter expression which are sent by
3505 * liblttng-ctl and discard them.
3506 *
3507 * This fixes an issue where the client may block while sending
3508 * the filter payload and encounter an error because the session
3509 * daemon closes the socket without ever handling this data.
3510 */
3511 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3512 cmd_ctx->lsm->u.disable.bytecode_len;
3513
3514 if (count) {
3515 char data[LTTNG_FILTER_MAX_LEN];
3516
3517 DBG("Discarding disable event command payload of size %zu", count);
3518 while (count) {
3519 ret = lttcomm_recv_unix_sock(sock, data,
3520 count > sizeof(data) ? sizeof(data) : count);
3521 if (ret < 0) {
3522 goto error;
3523 }
3524
3525 count -= (size_t) ret;
3526 }
3527 }
3528 /* FIXME: passing packed structure to non-packed pointer */
3529 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3530 cmd_ctx->lsm->u.disable.channel_name,
3531 &cmd_ctx->lsm->u.disable.event);
3532 break;
3533 }
3534 case LTTNG_ENABLE_CHANNEL:
3535 {
3536 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3537 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3538 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3539 &cmd_ctx->lsm->u.channel.chan,
3540 kernel_poll_pipe[1]);
3541 break;
3542 }
3543 case LTTNG_TRACK_PID:
3544 {
3545 ret = cmd_track_pid(cmd_ctx->session,
3546 cmd_ctx->lsm->domain.type,
3547 cmd_ctx->lsm->u.pid_tracker.pid);
3548 break;
3549 }
3550 case LTTNG_UNTRACK_PID:
3551 {
3552 ret = cmd_untrack_pid(cmd_ctx->session,
3553 cmd_ctx->lsm->domain.type,
3554 cmd_ctx->lsm->u.pid_tracker.pid);
3555 break;
3556 }
3557 case LTTNG_ENABLE_EVENT:
3558 {
3559 struct lttng_event_exclusion *exclusion = NULL;
3560 struct lttng_filter_bytecode *bytecode = NULL;
3561 char *filter_expression = NULL;
3562
3563 /* Handle exclusion events and receive it from the client. */
3564 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3565 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3566
3567 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3568 (count * LTTNG_SYMBOL_NAME_LEN));
3569 if (!exclusion) {
3570 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3571 goto error;
3572 }
3573
3574 DBG("Receiving var len exclusion event list from client ...");
3575 exclusion->count = count;
3576 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3577 count * LTTNG_SYMBOL_NAME_LEN);
3578 if (ret <= 0) {
3579 DBG("Nothing recv() from client var len data... continuing");
3580 *sock_error = 1;
3581 free(exclusion);
3582 ret = LTTNG_ERR_EXCLUSION_INVAL;
3583 goto error;
3584 }
3585 }
3586
3587 /* Get filter expression from client. */
3588 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3589 size_t expression_len =
3590 cmd_ctx->lsm->u.enable.expression_len;
3591
3592 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3593 ret = LTTNG_ERR_FILTER_INVAL;
3594 free(exclusion);
3595 goto error;
3596 }
3597
3598 filter_expression = zmalloc(expression_len);
3599 if (!filter_expression) {
3600 free(exclusion);
3601 ret = LTTNG_ERR_FILTER_NOMEM;
3602 goto error;
3603 }
3604
3605 /* Receive var. len. data */
3606 DBG("Receiving var len filter's expression from client ...");
3607 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3608 expression_len);
3609 if (ret <= 0) {
3610 DBG("Nothing recv() from client car len data... continuing");
3611 *sock_error = 1;
3612 free(filter_expression);
3613 free(exclusion);
3614 ret = LTTNG_ERR_FILTER_INVAL;
3615 goto error;
3616 }
3617 }
3618
3619 /* Handle filter and get bytecode from client. */
3620 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3621 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3622
3623 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3624 ret = LTTNG_ERR_FILTER_INVAL;
3625 free(filter_expression);
3626 free(exclusion);
3627 goto error;
3628 }
3629
3630 bytecode = zmalloc(bytecode_len);
3631 if (!bytecode) {
3632 free(filter_expression);
3633 free(exclusion);
3634 ret = LTTNG_ERR_FILTER_NOMEM;
3635 goto error;
3636 }
3637
3638 /* Receive var. len. data */
3639 DBG("Receiving var len filter's bytecode from client ...");
3640 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3641 if (ret <= 0) {
3642 DBG("Nothing recv() from client car len data... continuing");
3643 *sock_error = 1;
3644 free(filter_expression);
3645 free(bytecode);
3646 free(exclusion);
3647 ret = LTTNG_ERR_FILTER_INVAL;
3648 goto error;
3649 }
3650
3651 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3652 free(filter_expression);
3653 free(bytecode);
3654 free(exclusion);
3655 ret = LTTNG_ERR_FILTER_INVAL;
3656 goto error;
3657 }
3658 }
3659
3660 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3661 cmd_ctx->lsm->u.enable.channel_name,
3662 &cmd_ctx->lsm->u.enable.event,
3663 filter_expression, bytecode, exclusion,
3664 kernel_poll_pipe[1]);
3665 break;
3666 }
3667 case LTTNG_LIST_TRACEPOINTS:
3668 {
3669 struct lttng_event *events;
3670 ssize_t nb_events;
3671
3672 session_lock_list();
3673 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3674 session_unlock_list();
3675 if (nb_events < 0) {
3676 /* Return value is a negative lttng_error_code. */
3677 ret = -nb_events;
3678 goto error;
3679 }
3680
3681 /*
3682 * Setup lttng message with payload size set to the event list size in
3683 * bytes and then copy list into the llm payload.
3684 */
3685 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3686 sizeof(struct lttng_event) * nb_events);
3687 free(events);
3688
3689 if (ret < 0) {
3690 goto setup_error;
3691 }
3692
3693 ret = LTTNG_OK;
3694 break;
3695 }
3696 case LTTNG_LIST_TRACEPOINT_FIELDS:
3697 {
3698 struct lttng_event_field *fields;
3699 ssize_t nb_fields;
3700
3701 session_lock_list();
3702 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3703 &fields);
3704 session_unlock_list();
3705 if (nb_fields < 0) {
3706 /* Return value is a negative lttng_error_code. */
3707 ret = -nb_fields;
3708 goto error;
3709 }
3710
3711 /*
3712 * Setup lttng message with payload size set to the event list size in
3713 * bytes and then copy list into the llm payload.
3714 */
3715 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3716 sizeof(struct lttng_event_field) * nb_fields);
3717 free(fields);
3718
3719 if (ret < 0) {
3720 goto setup_error;
3721 }
3722
3723 ret = LTTNG_OK;
3724 break;
3725 }
3726 case LTTNG_LIST_SYSCALLS:
3727 {
3728 struct lttng_event *events;
3729 ssize_t nb_events;
3730
3731 nb_events = cmd_list_syscalls(&events);
3732 if (nb_events < 0) {
3733 /* Return value is a negative lttng_error_code. */
3734 ret = -nb_events;
3735 goto error;
3736 }
3737
3738 /*
3739 * Setup lttng message with payload size set to the event list size in
3740 * bytes and then copy list into the llm payload.
3741 */
3742 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3743 sizeof(struct lttng_event) * nb_events);
3744 free(events);
3745
3746 if (ret < 0) {
3747 goto setup_error;
3748 }
3749
3750 ret = LTTNG_OK;
3751 break;
3752 }
3753 case LTTNG_LIST_TRACKER_PIDS:
3754 {
3755 int32_t *pids = NULL;
3756 ssize_t nr_pids;
3757
3758 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3759 cmd_ctx->lsm->domain.type, &pids);
3760 if (nr_pids < 0) {
3761 /* Return value is a negative lttng_error_code. */
3762 ret = -nr_pids;
3763 goto error;
3764 }
3765
3766 /*
3767 * Setup lttng message with payload size set to the event list size in
3768 * bytes and then copy list into the llm payload.
3769 */
3770 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3771 sizeof(int32_t) * nr_pids);
3772 free(pids);
3773
3774 if (ret < 0) {
3775 goto setup_error;
3776 }
3777
3778 ret = LTTNG_OK;
3779 break;
3780 }
3781 case LTTNG_SET_CONSUMER_URI:
3782 {
3783 size_t nb_uri, len;
3784 struct lttng_uri *uris;
3785
3786 nb_uri = cmd_ctx->lsm->u.uri.size;
3787 len = nb_uri * sizeof(struct lttng_uri);
3788
3789 if (nb_uri == 0) {
3790 ret = LTTNG_ERR_INVALID;
3791 goto error;
3792 }
3793
3794 uris = zmalloc(len);
3795 if (uris == NULL) {
3796 ret = LTTNG_ERR_FATAL;
3797 goto error;
3798 }
3799
3800 /* Receive variable len data */
3801 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3802 ret = lttcomm_recv_unix_sock(sock, uris, len);
3803 if (ret <= 0) {
3804 DBG("No URIs received from client... continuing");
3805 *sock_error = 1;
3806 ret = LTTNG_ERR_SESSION_FAIL;
3807 free(uris);
3808 goto error;
3809 }
3810
3811 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3812 free(uris);
3813 if (ret != LTTNG_OK) {
3814 goto error;
3815 }
3816
3817
3818 break;
3819 }
3820 case LTTNG_START_TRACE:
3821 {
3822 ret = cmd_start_trace(cmd_ctx->session);
3823 break;
3824 }
3825 case LTTNG_STOP_TRACE:
3826 {
3827 ret = cmd_stop_trace(cmd_ctx->session);
3828 break;
3829 }
3830 case LTTNG_CREATE_SESSION:
3831 {
3832 size_t nb_uri, len;
3833 struct lttng_uri *uris = NULL;
3834
3835 nb_uri = cmd_ctx->lsm->u.uri.size;
3836 len = nb_uri * sizeof(struct lttng_uri);
3837
3838 if (nb_uri > 0) {
3839 uris = zmalloc(len);
3840 if (uris == NULL) {
3841 ret = LTTNG_ERR_FATAL;
3842 goto error;
3843 }
3844
3845 /* Receive variable len data */
3846 DBG("Waiting for %zu URIs from client ...", nb_uri);
3847 ret = lttcomm_recv_unix_sock(sock, uris, len);
3848 if (ret <= 0) {
3849 DBG("No URIs received from client... continuing");
3850 *sock_error = 1;
3851 ret = LTTNG_ERR_SESSION_FAIL;
3852 free(uris);
3853 goto error;
3854 }
3855
3856 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3857 DBG("Creating session with ONE network URI is a bad call");
3858 ret = LTTNG_ERR_SESSION_FAIL;
3859 free(uris);
3860 goto error;
3861 }
3862 }
3863
3864 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3865 &cmd_ctx->creds, 0);
3866
3867 free(uris);
3868
3869 break;
3870 }
3871 case LTTNG_DESTROY_SESSION:
3872 {
3873 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3874
3875 /* Set session to NULL so we do not unlock it after free. */
3876 cmd_ctx->session = NULL;
3877 break;
3878 }
3879 case LTTNG_LIST_DOMAINS:
3880 {
3881 ssize_t nb_dom;
3882 struct lttng_domain *domains = NULL;
3883
3884 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3885 if (nb_dom < 0) {
3886 /* Return value is a negative lttng_error_code. */
3887 ret = -nb_dom;
3888 goto error;
3889 }
3890
3891 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3892 nb_dom * sizeof(struct lttng_domain));
3893 free(domains);
3894
3895 if (ret < 0) {
3896 goto setup_error;
3897 }
3898
3899 ret = LTTNG_OK;
3900 break;
3901 }
3902 case LTTNG_LIST_CHANNELS:
3903 {
3904 ssize_t payload_size;
3905 struct lttng_channel *channels = NULL;
3906
3907 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3908 cmd_ctx->session, &channels);
3909 if (payload_size < 0) {
3910 /* Return value is a negative lttng_error_code. */
3911 ret = -payload_size;
3912 goto error;
3913 }
3914
3915 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3916 payload_size);
3917 free(channels);
3918
3919 if (ret < 0) {
3920 goto setup_error;
3921 }
3922
3923 ret = LTTNG_OK;
3924 break;
3925 }
3926 case LTTNG_LIST_EVENTS:
3927 {
3928 ssize_t nb_event;
3929 struct lttng_event *events = NULL;
3930 struct lttcomm_event_command_header cmd_header;
3931 size_t total_size;
3932
3933 memset(&cmd_header, 0, sizeof(cmd_header));
3934 /* Extended infos are included at the end of events */
3935 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3936 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3937 &events, &total_size);
3938
3939 if (nb_event < 0) {
3940 /* Return value is a negative lttng_error_code. */
3941 ret = -nb_event;
3942 goto error;
3943 }
3944
3945 cmd_header.nb_events = nb_event;
3946 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3947 &cmd_header, sizeof(cmd_header));
3948 free(events);
3949
3950 if (ret < 0) {
3951 goto setup_error;
3952 }
3953
3954 ret = LTTNG_OK;
3955 break;
3956 }
3957 case LTTNG_LIST_SESSIONS:
3958 {
3959 unsigned int nr_sessions;
3960 void *sessions_payload;
3961 size_t payload_len;
3962
3963 session_lock_list();
3964 nr_sessions = lttng_sessions_count(
3965 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3966 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3967 payload_len = sizeof(struct lttng_session) * nr_sessions;
3968 sessions_payload = zmalloc(payload_len);
3969
3970 if (!sessions_payload) {
3971 session_unlock_list();
3972 ret = -ENOMEM;
3973 goto setup_error;
3974 }
3975
3976 cmd_list_lttng_sessions(sessions_payload,
3977 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3978 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3979 session_unlock_list();
3980
3981 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3982 payload_len);
3983 free(sessions_payload);
3984
3985 if (ret < 0) {
3986 goto setup_error;
3987 }
3988
3989 ret = LTTNG_OK;
3990 break;
3991 }
3992 case LTTNG_REGISTER_CONSUMER:
3993 {
3994 struct consumer_data *cdata;
3995
3996 switch (cmd_ctx->lsm->domain.type) {
3997 case LTTNG_DOMAIN_KERNEL:
3998 cdata = &kconsumer_data;
3999 break;
4000 default:
4001 ret = LTTNG_ERR_UND;
4002 goto error;
4003 }
4004
4005 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
4006 cmd_ctx->lsm->u.reg.path, cdata);
4007 break;
4008 }
4009 case LTTNG_DATA_PENDING:
4010 {
4011 int pending_ret;
4012 uint8_t pending_ret_byte;
4013
4014 pending_ret = cmd_data_pending(cmd_ctx->session);
4015
4016 /*
4017 * FIXME
4018 *
4019 * This function may returns 0 or 1 to indicate whether or not
4020 * there is data pending. In case of error, it should return an
4021 * LTTNG_ERR code. However, some code paths may still return
4022 * a nondescript error code, which we handle by returning an
4023 * "unknown" error.
4024 */
4025 if (pending_ret == 0 || pending_ret == 1) {
4026 /*
4027 * ret will be set to LTTNG_OK at the end of
4028 * this function.
4029 */
4030 } else if (pending_ret < 0) {
4031 ret = LTTNG_ERR_UNK;
4032 goto setup_error;
4033 } else {
4034 ret = pending_ret;
4035 goto setup_error;
4036 }
4037
4038 pending_ret_byte = (uint8_t) pending_ret;
4039
4040 /* 1 byte to return whether or not data is pending */
4041 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
4042 &pending_ret_byte, 1);
4043
4044 if (ret < 0) {
4045 goto setup_error;
4046 }
4047
4048 ret = LTTNG_OK;
4049 break;
4050 }
4051 case LTTNG_SNAPSHOT_ADD_OUTPUT:
4052 {
4053 struct lttcomm_lttng_output_id reply;
4054
4055 ret = cmd_snapshot_add_output(cmd_ctx->session,
4056 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
4057 if (ret != LTTNG_OK) {
4058 goto error;
4059 }
4060
4061 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
4062 sizeof(reply));
4063 if (ret < 0) {
4064 goto setup_error;
4065 }
4066
4067 /* Copy output list into message payload */
4068 ret = LTTNG_OK;
4069 break;
4070 }
4071 case LTTNG_SNAPSHOT_DEL_OUTPUT:
4072 {
4073 ret = cmd_snapshot_del_output(cmd_ctx->session,
4074 &cmd_ctx->lsm->u.snapshot_output.output);
4075 break;
4076 }
4077 case LTTNG_SNAPSHOT_LIST_OUTPUT:
4078 {
4079 ssize_t nb_output;
4080 struct lttng_snapshot_output *outputs = NULL;
4081
4082 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
4083 if (nb_output < 0) {
4084 ret = -nb_output;
4085 goto error;
4086 }
4087
4088 assert((nb_output > 0 && outputs) || nb_output == 0);
4089 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
4090 nb_output * sizeof(struct lttng_snapshot_output));
4091 free(outputs);
4092
4093 if (ret < 0) {
4094 goto setup_error;
4095 }
4096
4097 ret = LTTNG_OK;
4098 break;
4099 }
4100 case LTTNG_SNAPSHOT_RECORD:
4101 {
4102 ret = cmd_snapshot_record(cmd_ctx->session,
4103 &cmd_ctx->lsm->u.snapshot_record.output,
4104 cmd_ctx->lsm->u.snapshot_record.wait);
4105 break;
4106 }
4107 case LTTNG_CREATE_SESSION_SNAPSHOT:
4108 {
4109 size_t nb_uri, len;
4110 struct lttng_uri *uris = NULL;
4111
4112 nb_uri = cmd_ctx->lsm->u.uri.size;
4113 len = nb_uri * sizeof(struct lttng_uri);
4114
4115 if (nb_uri > 0) {
4116 uris = zmalloc(len);
4117 if (uris == NULL) {
4118 ret = LTTNG_ERR_FATAL;
4119 goto error;
4120 }
4121
4122 /* Receive variable len data */
4123 DBG("Waiting for %zu URIs from client ...", nb_uri);
4124 ret = lttcomm_recv_unix_sock(sock, uris, len);
4125 if (ret <= 0) {
4126 DBG("No URIs received from client... continuing");
4127 *sock_error = 1;
4128 ret = LTTNG_ERR_SESSION_FAIL;
4129 free(uris);
4130 goto error;
4131 }
4132
4133 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4134 DBG("Creating session with ONE network URI is a bad call");
4135 ret = LTTNG_ERR_SESSION_FAIL;
4136 free(uris);
4137 goto error;
4138 }
4139 }
4140
4141 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4142 nb_uri, &cmd_ctx->creds);
4143 free(uris);
4144 break;
4145 }
4146 case LTTNG_CREATE_SESSION_LIVE:
4147 {
4148 size_t nb_uri, len;
4149 struct lttng_uri *uris = NULL;
4150
4151 nb_uri = cmd_ctx->lsm->u.uri.size;
4152 len = nb_uri * sizeof(struct lttng_uri);
4153
4154 if (nb_uri > 0) {
4155 uris = zmalloc(len);
4156 if (uris == NULL) {
4157 ret = LTTNG_ERR_FATAL;
4158 goto error;
4159 }
4160
4161 /* Receive variable len data */
4162 DBG("Waiting for %zu URIs from client ...", nb_uri);
4163 ret = lttcomm_recv_unix_sock(sock, uris, len);
4164 if (ret <= 0) {
4165 DBG("No URIs received from client... continuing");
4166 *sock_error = 1;
4167 ret = LTTNG_ERR_SESSION_FAIL;
4168 free(uris);
4169 goto error;
4170 }
4171
4172 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4173 DBG("Creating session with ONE network URI is a bad call");
4174 ret = LTTNG_ERR_SESSION_FAIL;
4175 free(uris);
4176 goto error;
4177 }
4178 }
4179
4180 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4181 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4182 free(uris);
4183 break;
4184 }
4185 case LTTNG_SAVE_SESSION:
4186 {
4187 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4188 &cmd_ctx->creds);
4189 break;
4190 }
4191 case LTTNG_SET_SESSION_SHM_PATH:
4192 {
4193 ret = cmd_set_session_shm_path(cmd_ctx->session,
4194 cmd_ctx->lsm->u.set_shm_path.shm_path);
4195 break;
4196 }
4197 case LTTNG_REGENERATE_METADATA:
4198 {
4199 ret = cmd_regenerate_metadata(cmd_ctx->session);
4200 break;
4201 }
4202 case LTTNG_REGENERATE_STATEDUMP:
4203 {
4204 ret = cmd_regenerate_statedump(cmd_ctx->session);
4205 break;
4206 }
4207 case LTTNG_REGISTER_TRIGGER:
4208 {
4209 ret = cmd_register_trigger(cmd_ctx, sock,
4210 notification_thread_handle);
4211 break;
4212 }
4213 case LTTNG_UNREGISTER_TRIGGER:
4214 {
4215 ret = cmd_unregister_trigger(cmd_ctx, sock,
4216 notification_thread_handle);
4217 break;
4218 }
4219 default:
4220 ret = LTTNG_ERR_UND;
4221 break;
4222 }
4223
4224 error:
4225 if (cmd_ctx->llm == NULL) {
4226 DBG("Missing llm structure. Allocating one.");
4227 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4228 goto setup_error;
4229 }
4230 }
4231 /* Set return code */
4232 cmd_ctx->llm->ret_code = ret;
4233 setup_error:
4234 if (cmd_ctx->session) {
4235 session_unlock(cmd_ctx->session);
4236 }
4237 if (need_tracing_session) {
4238 session_unlock_list();
4239 }
4240 init_setup_error:
4241 assert(!rcu_read_ongoing());
4242 return ret;
4243 }
4244
4245 /*
4246 * Thread managing health check socket.
4247 */
4248 static void *thread_manage_health(void *data)
4249 {
4250 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4251 uint32_t revents, nb_fd;
4252 struct lttng_poll_event events;
4253 struct health_comm_msg msg;
4254 struct health_comm_reply reply;
4255
4256 DBG("[thread] Manage health check started");
4257
4258 rcu_register_thread();
4259
4260 /* We might hit an error path before this is created. */
4261 lttng_poll_init(&events);
4262
4263 /* Create unix socket */
4264 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4265 if (sock < 0) {
4266 ERR("Unable to create health check Unix socket");
4267 goto error;
4268 }
4269
4270 if (is_root) {
4271 /* lttng health client socket path permissions */
4272 ret = chown(health_unix_sock_path, 0,
4273 utils_get_group_id(tracing_group_name));
4274 if (ret < 0) {
4275 ERR("Unable to set group on %s", health_unix_sock_path);
4276 PERROR("chown");
4277 goto error;
4278 }
4279
4280 ret = chmod(health_unix_sock_path,
4281 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4282 if (ret < 0) {
4283 ERR("Unable to set permissions on %s", health_unix_sock_path);
4284 PERROR("chmod");
4285 goto error;
4286 }
4287 }
4288
4289 /*
4290 * Set the CLOEXEC flag. Return code is useless because either way, the
4291 * show must go on.
4292 */
4293 (void) utils_set_fd_cloexec(sock);
4294
4295 ret = lttcomm_listen_unix_sock(sock);
4296 if (ret < 0) {
4297 goto error;
4298 }
4299
4300 /*
4301 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4302 * more will be added to this poll set.
4303 */
4304 ret = sessiond_set_thread_pollset(&events, 2);
4305 if (ret < 0) {
4306 goto error;
4307 }
4308
4309 /* Add the application registration socket */
4310 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4311 if (ret < 0) {
4312 goto error;
4313 }
4314
4315 sessiond_notify_ready();
4316
4317 while (1) {
4318 DBG("Health check ready");
4319
4320 /* Inifinite blocking call, waiting for transmission */
4321 restart:
4322 ret = lttng_poll_wait(&events, -1);
4323 if (ret < 0) {
4324 /*
4325 * Restart interrupted system call.
4326 */
4327 if (errno == EINTR) {
4328 goto restart;
4329 }
4330 goto error;
4331 }
4332
4333 nb_fd = ret;
4334
4335 for (i = 0; i < nb_fd; i++) {
4336 /* Fetch once the poll data */
4337 revents = LTTNG_POLL_GETEV(&events, i);
4338 pollfd = LTTNG_POLL_GETFD(&events, i);
4339
4340 if (!revents) {
4341 /* No activity for this FD (poll implementation). */
4342 continue;
4343 }
4344
4345 /* Thread quit pipe has been closed. Killing thread. */
4346 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4347 if (ret) {
4348 err = 0;
4349 goto exit;
4350 }
4351
4352 /* Event on the registration socket */
4353 if (pollfd == sock) {
4354 if (revents & LPOLLIN) {
4355 continue;
4356 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4357 ERR("Health socket poll error");
4358 goto error;
4359 } else {
4360 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4361 goto error;
4362 }
4363 }
4364 }
4365
4366 new_sock = lttcomm_accept_unix_sock(sock);
4367 if (new_sock < 0) {
4368 goto error;
4369 }
4370
4371 /*
4372 * Set the CLOEXEC flag. Return code is useless because either way, the
4373 * show must go on.
4374 */
4375 (void) utils_set_fd_cloexec(new_sock);
4376
4377 DBG("Receiving data from client for health...");
4378 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4379 if (ret <= 0) {
4380 DBG("Nothing recv() from client... continuing");
4381 ret = close(new_sock);
4382 if (ret) {
4383 PERROR("close");
4384 }
4385 continue;
4386 }
4387
4388 rcu_thread_online();
4389
4390 memset(&reply, 0, sizeof(reply));
4391 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4392 /*
4393 * health_check_state returns 0 if health is
4394 * bad.
4395 */
4396 if (!health_check_state(health_sessiond, i)) {
4397 reply.ret_code |= 1ULL << i;
4398 }
4399 }
4400
4401 DBG2("Health check return value %" PRIx64, reply.ret_code);
4402
4403 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4404 if (ret < 0) {
4405 ERR("Failed to send health data back to client");
4406 }
4407
4408 /* End of transmission */
4409 ret = close(new_sock);
4410 if (ret) {
4411 PERROR("close");
4412 }
4413 }
4414
4415 exit:
4416 error:
4417 if (err) {
4418 ERR("Health error occurred in %s", __func__);
4419 }
4420 DBG("Health check thread dying");
4421 unlink(health_unix_sock_path);
4422 if (sock >= 0) {
4423 ret = close(sock);
4424 if (ret) {
4425 PERROR("close");
4426 }
4427 }
4428
4429 lttng_poll_clean(&events);
4430 stop_threads();
4431 rcu_unregister_thread();
4432 return NULL;
4433 }
4434
4435 /*
4436 * This thread manage all clients request using the unix client socket for
4437 * communication.
4438 */
4439 static void *thread_manage_clients(void *data)
4440 {
4441 int sock = -1, ret, i, pollfd, err = -1;
4442 int sock_error;
4443 uint32_t revents, nb_fd;
4444 struct command_ctx *cmd_ctx = NULL;
4445 struct lttng_poll_event events;
4446
4447 DBG("[thread] Manage client started");
4448
4449 rcu_register_thread();
4450
4451 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4452
4453 health_code_update();
4454
4455 ret = lttcomm_listen_unix_sock(client_sock);
4456 if (ret < 0) {
4457 goto error_listen;
4458 }
4459
4460 /*
4461 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4462 * more will be added to this poll set.
4463 */
4464 ret = sessiond_set_thread_pollset(&events, 2);
4465 if (ret < 0) {
4466 goto error_create_poll;
4467 }
4468
4469 /* Add the application registration socket */
4470 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4471 if (ret < 0) {
4472 goto error;
4473 }
4474
4475 sessiond_notify_ready();
4476 ret = sem_post(&load_info->message_thread_ready);
4477 if (ret) {
4478 PERROR("sem_post message_thread_ready");
4479 goto error;
4480 }
4481
4482 /* This testpoint is after we signal readiness to the parent. */
4483 if (testpoint(sessiond_thread_manage_clients)) {
4484 goto error;
4485 }
4486
4487 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4488 goto error;
4489 }
4490
4491 health_code_update();
4492
4493 while (1) {
4494 DBG("Accepting client command ...");
4495
4496 /* Inifinite blocking call, waiting for transmission */
4497 restart:
4498 health_poll_entry();
4499 ret = lttng_poll_wait(&events, -1);
4500 health_poll_exit();
4501 if (ret < 0) {
4502 /*
4503 * Restart interrupted system call.
4504 */
4505 if (errno == EINTR) {
4506 goto restart;
4507 }
4508 goto error;
4509 }
4510
4511 nb_fd = ret;
4512
4513 for (i = 0; i < nb_fd; i++) {
4514 /* Fetch once the poll data */
4515 revents = LTTNG_POLL_GETEV(&events, i);
4516 pollfd = LTTNG_POLL_GETFD(&events, i);
4517
4518 health_code_update();
4519
4520 if (!revents) {
4521 /* No activity for this FD (poll implementation). */
4522 continue;
4523 }
4524
4525 /* Thread quit pipe has been closed. Killing thread. */
4526 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4527 if (ret) {
4528 err = 0;
4529 goto exit;
4530 }
4531
4532 /* Event on the registration socket */
4533 if (pollfd == client_sock) {
4534 if (revents & LPOLLIN) {
4535 continue;
4536 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4537 ERR("Client socket poll error");
4538 goto error;
4539 } else {
4540 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4541 goto error;
4542 }
4543 }
4544 }
4545
4546 DBG("Wait for client response");
4547
4548 health_code_update();
4549
4550 sock = lttcomm_accept_unix_sock(client_sock);
4551 if (sock < 0) {
4552 goto error;
4553 }
4554
4555 /*
4556 * Set the CLOEXEC flag. Return code is useless because either way, the
4557 * show must go on.
4558 */
4559 (void) utils_set_fd_cloexec(sock);
4560
4561 /* Set socket option for credentials retrieval */
4562 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4563 if (ret < 0) {
4564 goto error;
4565 }
4566
4567 /* Allocate context command to process the client request */
4568 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4569 if (cmd_ctx == NULL) {
4570 PERROR("zmalloc cmd_ctx");
4571 goto error;
4572 }
4573
4574 /* Allocate data buffer for reception */
4575 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4576 if (cmd_ctx->lsm == NULL) {
4577 PERROR("zmalloc cmd_ctx->lsm");
4578 goto error;
4579 }
4580
4581 cmd_ctx->llm = NULL;
4582 cmd_ctx->session = NULL;
4583
4584 health_code_update();
4585
4586 /*
4587 * Data is received from the lttng client. The struct
4588 * lttcomm_session_msg (lsm) contains the command and data request of
4589 * the client.
4590 */
4591 DBG("Receiving data from client ...");
4592 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4593 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4594 if (ret <= 0) {
4595 DBG("Nothing recv() from client... continuing");
4596 ret = close(sock);
4597 if (ret) {
4598 PERROR("close");
4599 }
4600 sock = -1;
4601 clean_command_ctx(&cmd_ctx);
4602 continue;
4603 }
4604
4605 health_code_update();
4606
4607 // TODO: Validate cmd_ctx including sanity check for
4608 // security purpose.
4609
4610 rcu_thread_online();
4611 /*
4612 * This function dispatch the work to the kernel or userspace tracer
4613 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4614 * informations for the client. The command context struct contains
4615 * everything this function may needs.
4616 */
4617 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4618 rcu_thread_offline();
4619 if (ret < 0) {
4620 ret = close(sock);
4621 if (ret) {
4622 PERROR("close");
4623 }
4624 sock = -1;
4625 /*
4626 * TODO: Inform client somehow of the fatal error. At
4627 * this point, ret < 0 means that a zmalloc failed
4628 * (ENOMEM). Error detected but still accept
4629 * command, unless a socket error has been
4630 * detected.
4631 */
4632 clean_command_ctx(&cmd_ctx);
4633 continue;
4634 }
4635
4636 health_code_update();
4637
4638 DBG("Sending response (size: %d, retcode: %s (%d))",
4639 cmd_ctx->lttng_msg_size,
4640 lttng_strerror(-cmd_ctx->llm->ret_code),
4641 cmd_ctx->llm->ret_code);
4642 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4643 if (ret < 0) {
4644 ERR("Failed to send data back to client");
4645 }
4646
4647 /* End of transmission */
4648 ret = close(sock);
4649 if (ret) {
4650 PERROR("close");
4651 }
4652 sock = -1;
4653
4654 clean_command_ctx(&cmd_ctx);
4655
4656 health_code_update();
4657 }
4658
4659 exit:
4660 error:
4661 if (sock >= 0) {
4662 ret = close(sock);
4663 if (ret) {
4664 PERROR("close");
4665 }
4666 }
4667
4668 lttng_poll_clean(&events);
4669 clean_command_ctx(&cmd_ctx);
4670
4671 error_listen:
4672 error_create_poll:
4673 unlink(client_unix_sock_path);
4674 if (client_sock >= 0) {
4675 ret = close(client_sock);
4676 if (ret) {
4677 PERROR("close");
4678 }
4679 }
4680
4681 if (err) {
4682 health_error();
4683 ERR("Health error occurred in %s", __func__);
4684 }
4685
4686 health_unregister(health_sessiond);
4687
4688 DBG("Client thread dying");
4689
4690 rcu_unregister_thread();
4691
4692 /*
4693 * Since we are creating the consumer threads, we own them, so we need
4694 * to join them before our thread exits.
4695 */
4696 ret = join_consumer_thread(&kconsumer_data);
4697 if (ret) {
4698 errno = ret;
4699 PERROR("join_consumer");
4700 }
4701
4702 ret = join_consumer_thread(&ustconsumer32_data);
4703 if (ret) {
4704 errno = ret;
4705 PERROR("join_consumer ust32");
4706 }
4707
4708 ret = join_consumer_thread(&ustconsumer64_data);
4709 if (ret) {
4710 errno = ret;
4711 PERROR("join_consumer ust64");
4712 }
4713 return NULL;
4714 }
4715
4716 static int string_match(const char *str1, const char *str2)
4717 {
4718 return (str1 && str2) && !strcmp(str1, str2);
4719 }
4720
4721 /*
4722 * Take an option from the getopt output and set it in the right variable to be
4723 * used later.
4724 *
4725 * Return 0 on success else a negative value.
4726 */
4727 static int set_option(int opt, const char *arg, const char *optname)
4728 {
4729 int ret = 0;
4730
4731 if (string_match(optname, "client-sock") || opt == 'c') {
4732 if (!arg || *arg == '\0') {
4733 ret = -EINVAL;
4734 goto end;
4735 }
4736 if (lttng_is_setuid_setgid()) {
4737 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4738 "-c, --client-sock");
4739 } else {
4740 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4741 }
4742 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4743 if (!arg || *arg == '\0') {
4744 ret = -EINVAL;
4745 goto end;
4746 }
4747 if (lttng_is_setuid_setgid()) {
4748 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4749 "-a, --apps-sock");
4750 } else {
4751 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4752 }
4753 } else if (string_match(optname, "daemonize") || opt == 'd') {
4754 opt_daemon = 1;
4755 } else if (string_match(optname, "background") || opt == 'b') {
4756 opt_background = 1;
4757 } else if (string_match(optname, "group") || opt == 'g') {
4758 if (!arg || *arg == '\0') {
4759 ret = -EINVAL;
4760 goto end;
4761 }
4762 if (lttng_is_setuid_setgid()) {
4763 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4764 "-g, --group");
4765 } else {
4766 /*
4767 * If the override option is set, the pointer points to a
4768 * *non* const thus freeing it even though the variable type is
4769 * set to const.
4770 */
4771 if (tracing_group_name_override) {
4772 free((void *) tracing_group_name);
4773 }
4774 tracing_group_name = strdup(arg);
4775 if (!tracing_group_name) {
4776 PERROR("strdup");
4777 ret = -ENOMEM;
4778 }
4779 tracing_group_name_override = 1;
4780 }
4781 } else if (string_match(optname, "help") || opt == 'h') {
4782 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4783 if (ret) {
4784 ERR("Cannot show --help for `lttng-sessiond`");
4785 perror("exec");
4786 }
4787 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4788 } else if (string_match(optname, "version") || opt == 'V') {
4789 fprintf(stdout, "%s\n", VERSION);
4790 exit(EXIT_SUCCESS);
4791 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4792 opt_sig_parent = 1;
4793 } else if (string_match(optname, "kconsumerd-err-sock")) {
4794 if (!arg || *arg == '\0') {
4795 ret = -EINVAL;
4796 goto end;
4797 }
4798 if (lttng_is_setuid_setgid()) {
4799 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4800 "--kconsumerd-err-sock");
4801 } else {
4802 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4803 }
4804 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4805 if (!arg || *arg == '\0') {
4806 ret = -EINVAL;
4807 goto end;
4808 }
4809 if (lttng_is_setuid_setgid()) {
4810 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4811 "--kconsumerd-cmd-sock");
4812 } else {
4813 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4814 }
4815 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4816 if (!arg || *arg == '\0') {
4817 ret = -EINVAL;
4818 goto end;
4819 }
4820 if (lttng_is_setuid_setgid()) {
4821 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4822 "--ustconsumerd64-err-sock");
4823 } else {
4824 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4825 }
4826 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4827 if (!arg || *arg == '\0') {
4828 ret = -EINVAL;
4829 goto end;
4830 }
4831 if (lttng_is_setuid_setgid()) {
4832 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4833 "--ustconsumerd64-cmd-sock");
4834 } else {
4835 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4836 }
4837 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4838 if (!arg || *arg == '\0') {
4839 ret = -EINVAL;
4840 goto end;
4841 }
4842 if (lttng_is_setuid_setgid()) {
4843 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4844 "--ustconsumerd32-err-sock");
4845 } else {
4846 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4847 }
4848 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4849 if (!arg || *arg == '\0') {
4850 ret = -EINVAL;
4851 goto end;
4852 }
4853 if (lttng_is_setuid_setgid()) {
4854 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4855 "--ustconsumerd32-cmd-sock");
4856 } else {
4857 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4858 }
4859 } else if (string_match(optname, "no-kernel")) {
4860 opt_no_kernel = 1;
4861 } else if (string_match(optname, "quiet") || opt == 'q') {
4862 lttng_opt_quiet = 1;
4863 } else if (string_match(optname, "verbose") || opt == 'v') {
4864 /* Verbose level can increase using multiple -v */
4865 if (arg) {
4866 /* Value obtained from config file */
4867 lttng_opt_verbose = config_parse_value(arg);
4868 } else {
4869 /* -v used on command line */
4870 lttng_opt_verbose++;
4871 }
4872 /* Clamp value to [0, 3] */
4873 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4874 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4875 } else if (string_match(optname, "verbose-consumer")) {
4876 if (arg) {
4877 opt_verbose_consumer = config_parse_value(arg);
4878 } else {
4879 opt_verbose_consumer++;
4880 }
4881 } else if (string_match(optname, "consumerd32-path")) {
4882 if (!arg || *arg == '\0') {
4883 ret = -EINVAL;
4884 goto end;
4885 }
4886 if (lttng_is_setuid_setgid()) {
4887 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4888 "--consumerd32-path");
4889 } else {
4890 if (consumerd32_bin_override) {
4891 free((void *) consumerd32_bin);
4892 }
4893 consumerd32_bin = strdup(arg);
4894 if (!consumerd32_bin) {
4895 PERROR("strdup");
4896 ret = -ENOMEM;
4897 }
4898 consumerd32_bin_override = 1;
4899 }
4900 } else if (string_match(optname, "consumerd32-libdir")) {
4901 if (!arg || *arg == '\0') {
4902 ret = -EINVAL;
4903 goto end;
4904 }
4905 if (lttng_is_setuid_setgid()) {
4906 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4907 "--consumerd32-libdir");
4908 } else {
4909 if (consumerd32_libdir_override) {
4910 free((void *) consumerd32_libdir);
4911 }
4912 consumerd32_libdir = strdup(arg);
4913 if (!consumerd32_libdir) {
4914 PERROR("strdup");
4915 ret = -ENOMEM;
4916 }
4917 consumerd32_libdir_override = 1;
4918 }
4919 } else if (string_match(optname, "consumerd64-path")) {
4920 if (!arg || *arg == '\0') {
4921 ret = -EINVAL;
4922 goto end;
4923 }
4924 if (lttng_is_setuid_setgid()) {
4925 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4926 "--consumerd64-path");
4927 } else {
4928 if (consumerd64_bin_override) {
4929 free((void *) consumerd64_bin);
4930 }
4931 consumerd64_bin = strdup(arg);
4932 if (!consumerd64_bin) {
4933 PERROR("strdup");
4934 ret = -ENOMEM;
4935 }
4936 consumerd64_bin_override = 1;
4937 }
4938 } else if (string_match(optname, "consumerd64-libdir")) {
4939 if (!arg || *arg == '\0') {
4940 ret = -EINVAL;
4941 goto end;
4942 }
4943 if (lttng_is_setuid_setgid()) {
4944 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4945 "--consumerd64-libdir");
4946 } else {
4947 if (consumerd64_libdir_override) {
4948 free((void *) consumerd64_libdir);
4949 }
4950 consumerd64_libdir = strdup(arg);
4951 if (!consumerd64_libdir) {
4952 PERROR("strdup");
4953 ret = -ENOMEM;
4954 }
4955 consumerd64_libdir_override = 1;
4956 }
4957 } else if (string_match(optname, "pidfile") || opt == 'p') {
4958 if (!arg || *arg == '\0') {
4959 ret = -EINVAL;
4960 goto end;
4961 }
4962 if (lttng_is_setuid_setgid()) {
4963 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4964 "-p, --pidfile");
4965 } else {
4966 free(opt_pidfile);
4967 opt_pidfile = strdup(arg);
4968 if (!opt_pidfile) {
4969 PERROR("strdup");
4970 ret = -ENOMEM;
4971 }
4972 }
4973 } else if (string_match(optname, "agent-tcp-port")) {
4974 if (!arg || *arg == '\0') {
4975 ret = -EINVAL;
4976 goto end;
4977 }
4978 if (lttng_is_setuid_setgid()) {
4979 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4980 "--agent-tcp-port");
4981 } else {
4982 unsigned long v;
4983
4984 errno = 0;
4985 v = strtoul(arg, NULL, 0);
4986 if (errno != 0 || !isdigit(arg[0])) {
4987 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4988 return -1;
4989 }
4990 if (v == 0 || v >= 65535) {
4991 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4992 return -1;
4993 }
4994 agent_tcp_port = (uint32_t) v;
4995 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4996 }
4997 } else if (string_match(optname, "load") || opt == 'l') {
4998 if (!arg || *arg == '\0') {
4999 ret = -EINVAL;
5000 goto end;
5001 }
5002 if (lttng_is_setuid_setgid()) {
5003 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5004 "-l, --load");
5005 } else {
5006 free(opt_load_session_path);
5007 opt_load_session_path = strdup(arg);
5008 if (!opt_load_session_path) {
5009 PERROR("strdup");
5010 ret = -ENOMEM;
5011 }
5012 }
5013 } else if (string_match(optname, "kmod-probes")) {
5014 if (!arg || *arg == '\0') {
5015 ret = -EINVAL;
5016 goto end;
5017 }
5018 if (lttng_is_setuid_setgid()) {
5019 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5020 "--kmod-probes");
5021 } else {
5022 free(kmod_probes_list);
5023 kmod_probes_list = strdup(arg);
5024 if (!kmod_probes_list) {
5025 PERROR("strdup");
5026 ret = -ENOMEM;
5027 }
5028 }
5029 } else if (string_match(optname, "extra-kmod-probes")) {
5030 if (!arg || *arg == '\0') {
5031 ret = -EINVAL;
5032 goto end;
5033 }
5034 if (lttng_is_setuid_setgid()) {
5035 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5036 "--extra-kmod-probes");
5037 } else {
5038 free(kmod_extra_probes_list);
5039 kmod_extra_probes_list = strdup(arg);
5040 if (!kmod_extra_probes_list) {
5041 PERROR("strdup");
5042 ret = -ENOMEM;
5043 }
5044 }
5045 } else if (string_match(optname, "config") || opt == 'f') {
5046 /* This is handled in set_options() thus silent skip. */
5047 goto end;
5048 } else {
5049 /* Unknown option or other error.
5050 * Error is printed by getopt, just return */
5051 ret = -1;
5052 }
5053
5054 end:
5055 if (ret == -EINVAL) {
5056 const char *opt_name = "unknown";
5057 int i;
5058
5059 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
5060 i++) {
5061 if (opt == long_options[i].val) {
5062 opt_name = long_options[i].name;
5063 break;
5064 }
5065 }
5066
5067 WARN("Invalid argument provided for option \"%s\", using default value.",
5068 opt_name);
5069 }
5070
5071 return ret;
5072 }
5073
5074 /*
5075 * config_entry_handler_cb used to handle options read from a config file.
5076 * See config_entry_handler_cb comment in common/config/session-config.h for the
5077 * return value conventions.
5078 */
5079 static int config_entry_handler(const struct config_entry *entry, void *unused)
5080 {
5081 int ret = 0, i;
5082
5083 if (!entry || !entry->name || !entry->value) {
5084 ret = -EINVAL;
5085 goto end;
5086 }
5087
5088 /* Check if the option is to be ignored */
5089 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5090 if (!strcmp(entry->name, config_ignore_options[i])) {
5091 goto end;
5092 }
5093 }
5094
5095 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5096 i++) {
5097
5098 /* Ignore if not fully matched. */
5099 if (strcmp(entry->name, long_options[i].name)) {
5100 continue;
5101 }
5102
5103 /*
5104 * If the option takes no argument on the command line, we have to
5105 * check if the value is "true". We support non-zero numeric values,
5106 * true, on and yes.
5107 */
5108 if (!long_options[i].has_arg) {
5109 ret = config_parse_value(entry->value);
5110 if (ret <= 0) {
5111 if (ret) {
5112 WARN("Invalid configuration value \"%s\" for option %s",
5113 entry->value, entry->name);
5114 }
5115 /* False, skip boolean config option. */
5116 goto end;
5117 }
5118 }
5119
5120 ret = set_option(long_options[i].val, entry->value, entry->name);
5121 goto end;
5122 }
5123
5124 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5125
5126 end:
5127 return ret;
5128 }
5129
5130 /*
5131 * daemon configuration loading and argument parsing
5132 */
5133 static int set_options(int argc, char **argv)
5134 {
5135 int ret = 0, c = 0, option_index = 0;
5136 int orig_optopt = optopt, orig_optind = optind;
5137 char *optstring;
5138 const char *config_path = NULL;
5139
5140 optstring = utils_generate_optstring(long_options,
5141 sizeof(long_options) / sizeof(struct option));
5142 if (!optstring) {
5143 ret = -ENOMEM;
5144 goto end;
5145 }
5146
5147 /* Check for the --config option */
5148 while ((c = getopt_long(argc, argv, optstring, long_options,
5149 &option_index)) != -1) {
5150 if (c == '?') {
5151 ret = -EINVAL;
5152 goto end;
5153 } else if (c != 'f') {
5154 /* if not equal to --config option. */
5155 continue;
5156 }
5157
5158 if (lttng_is_setuid_setgid()) {
5159 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5160 "-f, --config");
5161 } else {
5162 config_path = utils_expand_path(optarg);
5163 if (!config_path) {
5164 ERR("Failed to resolve path: %s", optarg);
5165 }
5166 }
5167 }
5168
5169 ret = config_get_section_entries(config_path, config_section_name,
5170 config_entry_handler, NULL);
5171 if (ret) {
5172 if (ret > 0) {
5173 ERR("Invalid configuration option at line %i", ret);
5174 ret = -1;
5175 }
5176 goto end;
5177 }
5178
5179 /* Reset getopt's global state */
5180 optopt = orig_optopt;
5181 optind = orig_optind;
5182 while (1) {
5183 option_index = -1;
5184 /*
5185 * getopt_long() will not set option_index if it encounters a
5186 * short option.
5187 */
5188 c = getopt_long(argc, argv, optstring, long_options,
5189 &option_index);
5190 if (c == -1) {
5191 break;
5192 }
5193
5194 /*
5195 * Pass NULL as the long option name if popt left the index
5196 * unset.
5197 */
5198 ret = set_option(c, optarg,
5199 option_index < 0 ? NULL :
5200 long_options[option_index].name);
5201 if (ret < 0) {
5202 break;
5203 }
5204 }
5205
5206 end:
5207 free(optstring);
5208 return ret;
5209 }
5210
5211 /*
5212 * Creates the two needed socket by the daemon.
5213 * apps_sock - The communication socket for all UST apps.
5214 * client_sock - The communication of the cli tool (lttng).
5215 */
5216 static int init_daemon_socket(void)
5217 {
5218 int ret = 0;
5219 mode_t old_umask;
5220
5221 old_umask = umask(0);
5222
5223 /* Create client tool unix socket */
5224 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5225 if (client_sock < 0) {
5226 ERR("Create unix sock failed: %s", client_unix_sock_path);
5227 ret = -1;
5228 goto end;
5229 }
5230
5231 /* Set the cloexec flag */
5232 ret = utils_set_fd_cloexec(client_sock);
5233 if (ret < 0) {
5234 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5235 "Continuing but note that the consumer daemon will have a "
5236 "reference to this socket on exec()", client_sock);
5237 }
5238
5239 /* File permission MUST be 660 */
5240 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5241 if (ret < 0) {
5242 ERR("Set file permissions failed: %s", client_unix_sock_path);
5243 PERROR("chmod");
5244 goto end;
5245 }
5246
5247 /* Create the application unix socket */
5248 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5249 if (apps_sock < 0) {
5250 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5251 ret = -1;
5252 goto end;
5253 }
5254
5255 /* Set the cloexec flag */
5256 ret = utils_set_fd_cloexec(apps_sock);
5257 if (ret < 0) {
5258 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5259 "Continuing but note that the consumer daemon will have a "
5260 "reference to this socket on exec()", apps_sock);
5261 }
5262
5263 /* File permission MUST be 666 */
5264 ret = chmod(apps_unix_sock_path,
5265 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5266 if (ret < 0) {
5267 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5268 PERROR("chmod");
5269 goto end;
5270 }
5271
5272 DBG3("Session daemon client socket %d and application socket %d created",
5273 client_sock, apps_sock);
5274
5275 end:
5276 umask(old_umask);
5277 return ret;
5278 }
5279
5280 /*
5281 * Check if the global socket is available, and if a daemon is answering at the
5282 * other side. If yes, error is returned.
5283 */
5284 static int check_existing_daemon(void)
5285 {
5286 /* Is there anybody out there ? */
5287 if (lttng_session_daemon_alive()) {
5288 return -EEXIST;
5289 }
5290
5291 return 0;
5292 }
5293
5294 /*
5295 * Set the tracing group gid onto the client socket.
5296 *
5297 * Race window between mkdir and chown is OK because we are going from more
5298 * permissive (root.root) to less permissive (root.tracing).
5299 */
5300 static int set_permissions(char *rundir)
5301 {
5302 int ret;
5303 gid_t gid;
5304
5305 gid = utils_get_group_id(tracing_group_name);
5306
5307 /* Set lttng run dir */
5308 ret = chown(rundir, 0, gid);
5309 if (ret < 0) {
5310 ERR("Unable to set group on %s", rundir);
5311 PERROR("chown");
5312 }
5313
5314 /*
5315 * Ensure all applications and tracing group can search the run
5316 * dir. Allow everyone to read the directory, since it does not
5317 * buy us anything to hide its content.
5318 */
5319 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5320 if (ret < 0) {
5321 ERR("Unable to set permissions on %s", rundir);
5322 PERROR("chmod");
5323 }
5324
5325 /* lttng client socket path */
5326 ret = chown(client_unix_sock_path, 0, gid);
5327 if (ret < 0) {
5328 ERR("Unable to set group on %s", client_unix_sock_path);
5329 PERROR("chown");
5330 }
5331
5332 /* kconsumer error socket path */
5333 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5334 if (ret < 0) {
5335 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5336 PERROR("chown");
5337 }
5338
5339 /* 64-bit ustconsumer error socket path */
5340 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5341 if (ret < 0) {
5342 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5343 PERROR("chown");
5344 }
5345
5346 /* 32-bit ustconsumer compat32 error socket path */
5347 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5348 if (ret < 0) {
5349 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5350 PERROR("chown");
5351 }
5352
5353 DBG("All permissions are set");
5354
5355 return ret;
5356 }
5357
5358 /*
5359 * Create the lttng run directory needed for all global sockets and pipe.
5360 */
5361 static int create_lttng_rundir(const char *rundir)
5362 {
5363 int ret;
5364
5365 DBG3("Creating LTTng run directory: %s", rundir);
5366
5367 ret = mkdir(rundir, S_IRWXU);
5368 if (ret < 0) {
5369 if (errno != EEXIST) {
5370 ERR("Unable to create %s", rundir);
5371 goto error;
5372 } else {
5373 ret = 0;
5374 }
5375 }
5376
5377 error:
5378 return ret;
5379 }
5380
5381 /*
5382 * Setup sockets and directory needed by the kconsumerd communication with the
5383 * session daemon.
5384 */
5385 static int set_consumer_sockets(struct consumer_data *consumer_data,
5386 const char *rundir)
5387 {
5388 int ret;
5389 char path[PATH_MAX];
5390
5391 switch (consumer_data->type) {
5392 case LTTNG_CONSUMER_KERNEL:
5393 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5394 break;
5395 case LTTNG_CONSUMER64_UST:
5396 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5397 break;
5398 case LTTNG_CONSUMER32_UST:
5399 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5400 break;
5401 default:
5402 ERR("Consumer type unknown");
5403 ret = -EINVAL;
5404 goto error;
5405 }
5406
5407 DBG2("Creating consumer directory: %s", path);
5408
5409 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5410 if (ret < 0 && errno != EEXIST) {
5411 PERROR("mkdir");
5412 ERR("Failed to create %s", path);
5413 goto error;
5414 }
5415 if (is_root) {
5416 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5417 if (ret < 0) {
5418 ERR("Unable to set group on %s", path);
5419 PERROR("chown");
5420 goto error;
5421 }
5422 }
5423
5424 /* Create the kconsumerd error unix socket */
5425 consumer_data->err_sock =
5426 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5427 if (consumer_data->err_sock < 0) {
5428 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5429 ret = -1;
5430 goto error;
5431 }
5432
5433 /*
5434 * Set the CLOEXEC flag. Return code is useless because either way, the
5435 * show must go on.
5436 */
5437 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5438 if (ret < 0) {
5439 PERROR("utils_set_fd_cloexec");
5440 /* continue anyway */
5441 }
5442
5443 /* File permission MUST be 660 */
5444 ret = chmod(consumer_data->err_unix_sock_path,
5445 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5446 if (ret < 0) {
5447 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5448 PERROR("chmod");
5449 goto error;
5450 }
5451
5452 error:
5453 return ret;
5454 }
5455
5456 /*
5457 * Signal handler for the daemon
5458 *
5459 * Simply stop all worker threads, leaving main() return gracefully after
5460 * joining all threads and calling cleanup().
5461 */
5462 static void sighandler(int sig)
5463 {
5464 switch (sig) {
5465 case SIGINT:
5466 DBG("SIGINT caught");
5467 stop_threads();
5468 break;
5469 case SIGTERM:
5470 DBG("SIGTERM caught");
5471 stop_threads();
5472 break;
5473 case SIGUSR1:
5474 CMM_STORE_SHARED(recv_child_signal, 1);
5475 break;
5476 default:
5477 break;
5478 }
5479 }
5480
5481 /*
5482 * Setup signal handler for :
5483 * SIGINT, SIGTERM, SIGPIPE
5484 */
5485 static int set_signal_handler(void)
5486 {
5487 int ret = 0;
5488 struct sigaction sa;
5489 sigset_t sigset;
5490
5491 if ((ret = sigemptyset(&sigset)) < 0) {
5492 PERROR("sigemptyset");
5493 return ret;
5494 }
5495
5496 sa.sa_mask = sigset;
5497 sa.sa_flags = 0;
5498
5499 sa.sa_handler = sighandler;
5500 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5501 PERROR("sigaction");
5502 return ret;
5503 }
5504
5505 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5506 PERROR("sigaction");
5507 return ret;
5508 }
5509
5510 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5511 PERROR("sigaction");
5512 return ret;
5513 }
5514
5515 sa.sa_handler = SIG_IGN;
5516 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5517 PERROR("sigaction");
5518 return ret;
5519 }
5520
5521 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5522
5523 return ret;
5524 }
5525
5526 /*
5527 * Set open files limit to unlimited. This daemon can open a large number of
5528 * file descriptors in order to consume multiple kernel traces.
5529 */
5530 static void set_ulimit(void)
5531 {
5532 int ret;
5533 struct rlimit lim;
5534
5535 /* The kernel does not allow an infinite limit for open files */
5536 lim.rlim_cur = 65535;
5537 lim.rlim_max = 65535;
5538
5539 ret = setrlimit(RLIMIT_NOFILE, &lim);
5540 if (ret < 0) {
5541 PERROR("failed to set open files limit");
5542 }
5543 }
5544
5545 /*
5546 * Write pidfile using the rundir and opt_pidfile.
5547 */
5548 static int write_pidfile(void)
5549 {
5550 int ret;
5551 char pidfile_path[PATH_MAX];
5552
5553 assert(rundir);
5554
5555 if (opt_pidfile) {
5556 if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) {
5557 ret = -1;
5558 goto error;
5559 }
5560 } else {
5561 /* Build pidfile path from rundir and opt_pidfile. */
5562 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5563 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5564 if (ret < 0) {
5565 PERROR("snprintf pidfile path");
5566 goto error;
5567 }
5568 }
5569
5570 /*
5571 * Create pid file in rundir.
5572 */
5573 ret = utils_create_pid_file(getpid(), pidfile_path);
5574 error:
5575 return ret;
5576 }
5577
5578 /*
5579 * Create lockfile using the rundir and return its fd.
5580 */
5581 static int create_lockfile(void)
5582 {
5583 int ret;
5584 char lockfile_path[PATH_MAX];
5585
5586 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5587 if (ret < 0) {
5588 goto error;
5589 }
5590
5591 ret = utils_create_lock_file(lockfile_path);
5592 error:
5593 return ret;
5594 }
5595
5596 /*
5597 * Write agent TCP port using the rundir.
5598 */
5599 static int write_agent_port(void)
5600 {
5601 int ret;
5602 char path[PATH_MAX];
5603
5604 assert(rundir);
5605
5606 ret = snprintf(path, sizeof(path), "%s/"
5607 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5608 if (ret < 0) {
5609 PERROR("snprintf agent port path");
5610 goto error;
5611 }
5612
5613 /*
5614 * Create TCP agent port file in rundir.
5615 */
5616 ret = utils_create_pid_file(agent_tcp_port, path);
5617
5618 error:
5619 return ret;
5620 }
5621
5622 /*
5623 * main
5624 */
5625 int main(int argc, char **argv)
5626 {
5627 int ret = 0, retval = 0;
5628 void *status;
5629 const char *home_path, *env_app_timeout;
5630 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5631 *ust64_channel_monitor_pipe = NULL,
5632 *kernel_channel_monitor_pipe = NULL;
5633 bool notification_thread_running = false;
5634
5635 init_kernel_workarounds();
5636
5637 rcu_register_thread();
5638
5639 if (set_signal_handler()) {
5640 retval = -1;
5641 goto exit_set_signal_handler;
5642 }
5643
5644 setup_consumerd_path();
5645
5646 page_size = sysconf(_SC_PAGESIZE);
5647 if (page_size < 0) {
5648 PERROR("sysconf _SC_PAGESIZE");
5649 page_size = LONG_MAX;
5650 WARN("Fallback page size to %ld", page_size);
5651 }
5652
5653 /*
5654 * Parse arguments and load the daemon configuration file.
5655 *
5656 * We have an exit_options exit path to free memory reserved by
5657 * set_options. This is needed because the rest of sessiond_cleanup()
5658 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5659 * depends on set_options.
5660 */
5661 progname = argv[0];
5662 if (set_options(argc, argv)) {
5663 retval = -1;
5664 goto exit_options;
5665 }
5666
5667 /* Daemonize */
5668 if (opt_daemon || opt_background) {
5669 int i;
5670
5671 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5672 !opt_background);
5673 if (ret < 0) {
5674 retval = -1;
5675 goto exit_options;
5676 }
5677
5678 /*
5679 * We are in the child. Make sure all other file descriptors are
5680 * closed, in case we are called with more opened file
5681 * descriptors than the standard ones.
5682 */
5683 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5684 (void) close(i);
5685 }
5686 }
5687
5688 if (run_as_create_worker(argv[0]) < 0) {
5689 goto exit_create_run_as_worker_cleanup;
5690 }
5691
5692 /*
5693 * Starting from here, we can create threads. This needs to be after
5694 * lttng_daemonize due to RCU.
5695 */
5696
5697 /*
5698 * Initialize the health check subsystem. This call should set the
5699 * appropriate time values.
5700 */
5701 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5702 if (!health_sessiond) {
5703 PERROR("health_app_create error");
5704 retval = -1;
5705 goto exit_health_sessiond_cleanup;
5706 }
5707
5708 /* Create thread to clean up RCU hash tables */
5709 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5710 retval = -1;
5711 goto exit_ht_cleanup;
5712 }
5713
5714 /* Create thread quit pipe */
5715 if (init_thread_quit_pipe()) {
5716 retval = -1;
5717 goto exit_init_data;
5718 }
5719
5720 /* Check if daemon is UID = 0 */
5721 is_root = !getuid();
5722
5723 if (is_root) {
5724 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5725 if (!rundir) {
5726 retval = -1;
5727 goto exit_init_data;
5728 }
5729
5730 /* Create global run dir with root access */
5731 if (create_lttng_rundir(rundir)) {
5732 retval = -1;
5733 goto exit_init_data;
5734 }
5735
5736 if (strlen(apps_unix_sock_path) == 0) {
5737 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5738 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5739 if (ret < 0) {
5740 retval = -1;
5741 goto exit_init_data;
5742 }
5743 }
5744
5745 if (strlen(client_unix_sock_path) == 0) {
5746 ret = snprintf(client_unix_sock_path, PATH_MAX,
5747 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5748 if (ret < 0) {
5749 retval = -1;
5750 goto exit_init_data;
5751 }
5752 }
5753
5754 /* Set global SHM for ust */
5755 if (strlen(wait_shm_path) == 0) {
5756 ret = snprintf(wait_shm_path, PATH_MAX,
5757 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5758 if (ret < 0) {
5759 retval = -1;
5760 goto exit_init_data;
5761 }
5762 }
5763
5764 if (strlen(health_unix_sock_path) == 0) {
5765 ret = snprintf(health_unix_sock_path,
5766 sizeof(health_unix_sock_path),
5767 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5768 if (ret < 0) {
5769 retval = -1;
5770 goto exit_init_data;
5771 }
5772 }
5773
5774 /* Setup kernel consumerd path */
5775 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5776 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5777 if (ret < 0) {
5778 retval = -1;
5779 goto exit_init_data;
5780 }
5781 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5782 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5783 if (ret < 0) {
5784 retval = -1;
5785 goto exit_init_data;
5786 }
5787
5788 DBG2("Kernel consumer err path: %s",
5789 kconsumer_data.err_unix_sock_path);
5790 DBG2("Kernel consumer cmd path: %s",
5791 kconsumer_data.cmd_unix_sock_path);
5792 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5793 if (!kernel_channel_monitor_pipe) {
5794 ERR("Failed to create kernel consumer channel monitor pipe");
5795 retval = -1;
5796 goto exit_init_data;
5797 }
5798 kconsumer_data.channel_monitor_pipe =
5799 lttng_pipe_release_writefd(
5800 kernel_channel_monitor_pipe);
5801 if (kconsumer_data.channel_monitor_pipe < 0) {
5802 retval = -1;
5803 goto exit_init_data;
5804 }
5805 } else {
5806 home_path = utils_get_home_dir();
5807 if (home_path == NULL) {
5808 /* TODO: Add --socket PATH option */
5809 ERR("Can't get HOME directory for sockets creation.");
5810 retval = -1;
5811 goto exit_init_data;
5812 }
5813
5814 /*
5815 * Create rundir from home path. This will create something like
5816 * $HOME/.lttng
5817 */
5818 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5819 if (ret < 0) {
5820 retval = -1;
5821 goto exit_init_data;
5822 }
5823
5824 if (create_lttng_rundir(rundir)) {
5825 retval = -1;
5826 goto exit_init_data;
5827 }
5828
5829 if (strlen(apps_unix_sock_path) == 0) {
5830 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5831 DEFAULT_HOME_APPS_UNIX_SOCK,
5832 home_path);
5833 if (ret < 0) {
5834 retval = -1;
5835 goto exit_init_data;
5836 }
5837 }
5838
5839 /* Set the cli tool unix socket path */
5840 if (strlen(client_unix_sock_path) == 0) {
5841 ret = snprintf(client_unix_sock_path, PATH_MAX,
5842 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5843 home_path);
5844 if (ret < 0) {
5845 retval = -1;
5846 goto exit_init_data;
5847 }
5848 }
5849
5850 /* Set global SHM for ust */
5851 if (strlen(wait_shm_path) == 0) {
5852 ret = snprintf(wait_shm_path, PATH_MAX,
5853 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5854 getuid());
5855 if (ret < 0) {
5856 retval = -1;
5857 goto exit_init_data;
5858 }
5859 }
5860
5861 /* Set health check Unix path */
5862 if (strlen(health_unix_sock_path) == 0) {
5863 ret = snprintf(health_unix_sock_path,
5864 sizeof(health_unix_sock_path),
5865 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5866 home_path);
5867 if (ret < 0) {
5868 retval = -1;
5869 goto exit_init_data;
5870 }
5871 }
5872 }
5873
5874 lockfile_fd = create_lockfile();
5875 if (lockfile_fd < 0) {
5876 retval = -1;
5877 goto exit_init_data;
5878 }
5879
5880 /* Set consumer initial state */
5881 kernel_consumerd_state = CONSUMER_STOPPED;
5882 ust_consumerd_state = CONSUMER_STOPPED;
5883
5884 DBG("Client socket path %s", client_unix_sock_path);
5885 DBG("Application socket path %s", apps_unix_sock_path);
5886 DBG("Application wait path %s", wait_shm_path);
5887 DBG("LTTng run directory path: %s", rundir);
5888
5889 /* 32 bits consumerd path setup */
5890 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5891 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5892 if (ret < 0) {
5893 PERROR("snprintf 32-bit consumer error socket path");
5894 retval = -1;
5895 goto exit_init_data;
5896 }
5897 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5898 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5899 if (ret < 0) {
5900 PERROR("snprintf 32-bit consumer command socket path");
5901 retval = -1;
5902 goto exit_init_data;
5903 }
5904
5905 DBG2("UST consumer 32 bits err path: %s",
5906 ustconsumer32_data.err_unix_sock_path);
5907 DBG2("UST consumer 32 bits cmd path: %s",
5908 ustconsumer32_data.cmd_unix_sock_path);
5909 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5910 if (!ust32_channel_monitor_pipe) {
5911 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5912 retval = -1;
5913 goto exit_init_data;
5914 }
5915 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5916 ust32_channel_monitor_pipe);
5917 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5918 retval = -1;
5919 goto exit_init_data;
5920 }
5921
5922 /* 64 bits consumerd path setup */
5923 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5924 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5925 if (ret < 0) {
5926 PERROR("snprintf 64-bit consumer error socket path");
5927 retval = -1;
5928 goto exit_init_data;
5929 }
5930 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5931 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5932 if (ret < 0) {
5933 PERROR("snprintf 64-bit consumer command socket path");
5934 retval = -1;
5935 goto exit_init_data;
5936 }
5937
5938 DBG2("UST consumer 64 bits err path: %s",
5939 ustconsumer64_data.err_unix_sock_path);
5940 DBG2("UST consumer 64 bits cmd path: %s",
5941 ustconsumer64_data.cmd_unix_sock_path);
5942 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5943 if (!ust64_channel_monitor_pipe) {
5944 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5945 retval = -1;
5946 goto exit_init_data;
5947 }
5948 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5949 ust64_channel_monitor_pipe);
5950 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5951 retval = -1;
5952 goto exit_init_data;
5953 }
5954
5955 /*
5956 * See if daemon already exist.
5957 */
5958 if (check_existing_daemon()) {
5959 ERR("Already running daemon.\n");
5960 /*
5961 * We do not goto exit because we must not cleanup()
5962 * because a daemon is already running.
5963 */
5964 retval = -1;
5965 goto exit_init_data;
5966 }
5967
5968 /*
5969 * Init UST app hash table. Alloc hash table before this point since
5970 * cleanup() can get called after that point.
5971 */
5972 if (ust_app_ht_alloc()) {
5973 ERR("Failed to allocate UST app hash table");
5974 retval = -1;
5975 goto exit_init_data;
5976 }
5977
5978 /*
5979 * Initialize agent app hash table. We allocate the hash table here
5980 * since cleanup() can get called after this point.
5981 */
5982 if (agent_app_ht_alloc()) {
5983 ERR("Failed to allocate Agent app hash table");
5984 retval = -1;
5985 goto exit_init_data;
5986 }
5987
5988 /*
5989 * These actions must be executed as root. We do that *after* setting up
5990 * the sockets path because we MUST make the check for another daemon using
5991 * those paths *before* trying to set the kernel consumer sockets and init
5992 * kernel tracer.
5993 */
5994 if (is_root) {
5995 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5996 retval = -1;
5997 goto exit_init_data;
5998 }
5999
6000 /* Setup kernel tracer */
6001 if (!opt_no_kernel) {
6002 init_kernel_tracer();
6003 if (kernel_tracer_fd >= 0) {
6004 ret = syscall_init_table();
6005 if (ret < 0) {
6006 ERR("Unable to populate syscall table. "
6007 "Syscall tracing won't work "
6008 "for this session daemon.");
6009 }
6010 }
6011 }
6012
6013 /* Set ulimit for open files */
6014 set_ulimit();
6015 }
6016 /* init lttng_fd tracking must be done after set_ulimit. */
6017 lttng_fd_init();
6018
6019 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
6020 retval = -1;
6021 goto exit_init_data;
6022 }
6023
6024 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
6025 retval = -1;
6026 goto exit_init_data;
6027 }
6028
6029 /* Setup the needed unix socket */
6030 if (init_daemon_socket()) {
6031 retval = -1;
6032 goto exit_init_data;
6033 }
6034
6035 /* Set credentials to socket */
6036 if (is_root && set_permissions(rundir)) {
6037 retval = -1;
6038 goto exit_init_data;
6039 }
6040
6041 /* Get parent pid if -S, --sig-parent is specified. */
6042 if (opt_sig_parent) {
6043 ppid = getppid();
6044 }
6045
6046 /* Setup the kernel pipe for waking up the kernel thread */
6047 if (is_root && !opt_no_kernel) {
6048 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
6049 retval = -1;
6050 goto exit_init_data;
6051 }
6052 }
6053
6054 /* Setup the thread apps communication pipe. */
6055 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
6056 retval = -1;
6057 goto exit_init_data;
6058 }
6059
6060 /* Setup the thread apps notify communication pipe. */
6061 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
6062 retval = -1;
6063 goto exit_init_data;
6064 }
6065
6066 /* Initialize global buffer per UID and PID registry. */
6067 buffer_reg_init_uid_registry();
6068 buffer_reg_init_pid_registry();
6069
6070 /* Init UST command queue. */
6071 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
6072
6073 /*
6074 * Get session list pointer. This pointer MUST NOT be free'd. This list
6075 * is statically declared in session.c
6076 */
6077 session_list_ptr = session_get_list();
6078
6079 cmd_init();
6080
6081 /* Check for the application socket timeout env variable. */
6082 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
6083 if (env_app_timeout) {
6084 app_socket_timeout = atoi(env_app_timeout);
6085 } else {
6086 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
6087 }
6088
6089 ret = write_pidfile();
6090 if (ret) {
6091 ERR("Error in write_pidfile");
6092 retval = -1;
6093 goto exit_init_data;
6094 }
6095 ret = write_agent_port();
6096 if (ret) {
6097 ERR("Error in write_agent_port");
6098 retval = -1;
6099 goto exit_init_data;
6100 }
6101
6102 /* Initialize communication library */
6103 lttcomm_init();
6104 /* Initialize TCP timeout values */
6105 lttcomm_inet_init();
6106
6107 if (load_session_init_data(&load_info) < 0) {
6108 retval = -1;
6109 goto exit_init_data;
6110 }
6111 load_info->path = opt_load_session_path;
6112
6113 /* Create health-check thread. */
6114 ret = pthread_create(&health_thread, default_pthread_attr(),
6115 thread_manage_health, (void *) NULL);
6116 if (ret) {
6117 errno = ret;
6118 PERROR("pthread_create health");
6119 retval = -1;
6120 goto exit_health;
6121 }
6122
6123 /* notification_thread_data acquires the pipes' read side. */
6124 notification_thread_handle = notification_thread_handle_create(
6125 ust32_channel_monitor_pipe,
6126 ust64_channel_monitor_pipe,
6127 kernel_channel_monitor_pipe);
6128 if (!notification_thread_handle) {
6129 retval = -1;
6130 ERR("Failed to create notification thread shared data");
6131 stop_threads();
6132 goto exit_notification;
6133 }
6134
6135 /* Create notification thread. */
6136 ret = pthread_create(&notification_thread, default_pthread_attr(),
6137 thread_notification, notification_thread_handle);
6138 if (ret) {
6139 errno = ret;
6140 PERROR("pthread_create notification");
6141 retval = -1;
6142 stop_threads();
6143 goto exit_notification;
6144 }
6145 notification_thread_running = true;
6146
6147 /* Create thread to manage the client socket */
6148 ret = pthread_create(&client_thread, default_pthread_attr(),
6149 thread_manage_clients, (void *) NULL);
6150 if (ret) {
6151 errno = ret;
6152 PERROR("pthread_create clients");
6153 retval = -1;
6154 stop_threads();
6155 goto exit_client;
6156 }
6157
6158 /* Create thread to dispatch registration */
6159 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
6160 thread_dispatch_ust_registration, (void *) NULL);
6161 if (ret) {
6162 errno = ret;
6163 PERROR("pthread_create dispatch");
6164 retval = -1;
6165 stop_threads();
6166 goto exit_dispatch;
6167 }
6168
6169 /* Create thread to manage application registration. */
6170 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
6171 thread_registration_apps, (void *) NULL);
6172 if (ret) {
6173 errno = ret;
6174 PERROR("pthread_create registration");
6175 retval = -1;
6176 stop_threads();
6177 goto exit_reg_apps;
6178 }
6179
6180 /* Create thread to manage application socket */
6181 ret = pthread_create(&apps_thread, default_pthread_attr(),
6182 thread_manage_apps, (void *) NULL);
6183 if (ret) {
6184 errno = ret;
6185 PERROR("pthread_create apps");
6186 retval = -1;
6187 stop_threads();
6188 goto exit_apps;
6189 }
6190
6191 /* Create thread to manage application notify socket */
6192 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
6193 ust_thread_manage_notify, (void *) NULL);
6194 if (ret) {
6195 errno = ret;
6196 PERROR("pthread_create notify");
6197 retval = -1;
6198 stop_threads();
6199 goto exit_apps_notify;
6200 }
6201
6202 /* Create agent registration thread. */
6203 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
6204 agent_thread_manage_registration, (void *) NULL);
6205 if (ret) {
6206 errno = ret;
6207 PERROR("pthread_create agent");
6208 retval = -1;
6209 stop_threads();
6210 goto exit_agent_reg;
6211 }
6212
6213 /* Don't start this thread if kernel tracing is not requested nor root */
6214 if (is_root && !opt_no_kernel) {
6215 /* Create kernel thread to manage kernel event */
6216 ret = pthread_create(&kernel_thread, default_pthread_attr(),
6217 thread_manage_kernel, (void *) NULL);
6218 if (ret) {
6219 errno = ret;
6220 PERROR("pthread_create kernel");
6221 retval = -1;
6222 stop_threads();
6223 goto exit_kernel;
6224 }
6225 }
6226
6227 /* Create session loading thread. */
6228 ret = pthread_create(&load_session_thread, default_pthread_attr(),
6229 thread_load_session, load_info);
6230 if (ret) {
6231 errno = ret;
6232 PERROR("pthread_create load_session_thread");
6233 retval = -1;
6234 stop_threads();
6235 goto exit_load_session;
6236 }
6237
6238 /*
6239 * This is where we start awaiting program completion (e.g. through
6240 * signal that asks threads to teardown).
6241 */
6242
6243 ret = pthread_join(load_session_thread, &status);
6244 if (ret) {
6245 errno = ret;
6246 PERROR("pthread_join load_session_thread");
6247 retval = -1;
6248 }
6249 exit_load_session:
6250
6251 if (is_root && !opt_no_kernel) {
6252 ret = pthread_join(kernel_thread, &status);
6253 if (ret) {
6254 errno = ret;
6255 PERROR("pthread_join");
6256 retval = -1;
6257 }
6258 }
6259 exit_kernel:
6260
6261 ret = pthread_join(agent_reg_thread, &status);
6262 if (ret) {
6263 errno = ret;
6264 PERROR("pthread_join agent");
6265 retval = -1;
6266 }
6267 exit_agent_reg:
6268
6269 ret = pthread_join(apps_notify_thread, &status);
6270 if (ret) {
6271 errno = ret;
6272 PERROR("pthread_join apps notify");
6273 retval = -1;
6274 }
6275 exit_apps_notify:
6276
6277 ret = pthread_join(apps_thread, &status);
6278 if (ret) {
6279 errno = ret;
6280 PERROR("pthread_join apps");
6281 retval = -1;
6282 }
6283 exit_apps:
6284
6285 ret = pthread_join(reg_apps_thread, &status);
6286 if (ret) {
6287 errno = ret;
6288 PERROR("pthread_join");
6289 retval = -1;
6290 }
6291 exit_reg_apps:
6292
6293 /*
6294 * Join dispatch thread after joining reg_apps_thread to ensure
6295 * we don't leak applications in the queue.
6296 */
6297 ret = pthread_join(dispatch_thread, &status);
6298 if (ret) {
6299 errno = ret;
6300 PERROR("pthread_join");
6301 retval = -1;
6302 }
6303 exit_dispatch:
6304
6305 ret = pthread_join(client_thread, &status);
6306 if (ret) {
6307 errno = ret;
6308 PERROR("pthread_join");
6309 retval = -1;
6310 }
6311
6312 exit_client:
6313 exit_notification:
6314 ret = pthread_join(health_thread, &status);
6315 if (ret) {
6316 errno = ret;
6317 PERROR("pthread_join health thread");
6318 retval = -1;
6319 }
6320
6321 exit_health:
6322 exit_init_data:
6323 /*
6324 * Wait for all pending call_rcu work to complete before tearing
6325 * down data structures. call_rcu worker may be trying to
6326 * perform lookups in those structures.
6327 */
6328 rcu_barrier();
6329 /*
6330 * sessiond_cleanup() is called when no other thread is running, except
6331 * the ht_cleanup thread, which is needed to destroy the hash tables.
6332 */
6333 rcu_thread_online();
6334 sessiond_cleanup();
6335
6336 /*
6337 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6338 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6339 * the queue is empty before shutting down the clean-up thread.
6340 */
6341 rcu_barrier();
6342
6343 /*
6344 * The teardown of the notification system is performed after the
6345 * session daemon's teardown in order to allow it to be notified
6346 * of the active session and channels at the moment of the teardown.
6347 */
6348 if (notification_thread_handle) {
6349 if (notification_thread_running) {
6350 notification_thread_command_quit(
6351 notification_thread_handle);
6352 ret = pthread_join(notification_thread, &status);
6353 if (ret) {
6354 errno = ret;
6355 PERROR("pthread_join notification thread");
6356 retval = -1;
6357 }
6358 }
6359 notification_thread_handle_destroy(notification_thread_handle);
6360 }
6361
6362 rcu_thread_offline();
6363 rcu_unregister_thread();
6364
6365 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6366 if (ret) {
6367 retval = -1;
6368 }
6369 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6370 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6371 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6372 exit_ht_cleanup:
6373
6374 health_app_destroy(health_sessiond);
6375 exit_health_sessiond_cleanup:
6376 exit_create_run_as_worker_cleanup:
6377
6378 exit_options:
6379 sessiond_cleanup_options();
6380
6381 exit_set_signal_handler:
6382 if (!retval) {
6383 exit(EXIT_SUCCESS);
6384 } else {
6385 exit(EXIT_FAILURE);
6386 }
6387 }
This page took 0.189515 seconds and 6 git commands to generate.