545f456e912f6a62bb76cc265350dc9369a7e3e1
[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
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/compat/getenv.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/session-config.h>
51
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
54 #include "channel.h"
55 #include "cmd.h"
56 #include "consumer.h"
57 #include "context.h"
58 #include "event.h"
59 #include "kernel.h"
60 #include "kernel-consumer.h"
61 #include "modprobe.h"
62 #include "shm.h"
63 #include "ust-ctl.h"
64 #include "ust-consumer.h"
65 #include "utils.h"
66 #include "fd-limit.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "agent-thread.h"
71 #include "save.h"
72 #include "load-session-thread.h"
73 #include "syscall.h"
74 #include "agent.h"
75
76 #define CONSUMERD_FILE "lttng-consumerd"
77
78 const char *progname;
79 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
80 static int tracing_group_name_override;
81 static char *opt_pidfile;
82 static int opt_sig_parent;
83 static int opt_verbose_consumer;
84 static int opt_daemon, opt_background;
85 static int opt_no_kernel;
86 static char *opt_load_session_path;
87 static pid_t ppid; /* Parent PID for --sig-parent option */
88 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
89 static char *rundir;
90 static int lockfile_fd = -1;
91
92 /* Set to 1 when a SIGUSR1 signal is received. */
93 static int recv_child_signal;
94
95 /*
96 * Consumer daemon specific control data. Every value not initialized here is
97 * set to 0 by the static definition.
98 */
99 static struct consumer_data kconsumer_data = {
100 .type = LTTNG_CONSUMER_KERNEL,
101 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer64_data = {
111 .type = LTTNG_CONSUMER64_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121 static struct consumer_data ustconsumer32_data = {
122 .type = LTTNG_CONSUMER32_UST,
123 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
124 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
125 .err_sock = -1,
126 .cmd_sock = -1,
127 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
128 .lock = PTHREAD_MUTEX_INITIALIZER,
129 .cond = PTHREAD_COND_INITIALIZER,
130 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
131 };
132
133 /* Command line options */
134 static const struct option long_options[] = {
135 { "client-sock", required_argument, 0, 'c' },
136 { "apps-sock", required_argument, 0, 'a' },
137 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
138 { "kconsumerd-err-sock", required_argument, 0, '\0' },
139 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
143 { "consumerd32-path", required_argument, 0, '\0' },
144 { "consumerd32-libdir", required_argument, 0, '\0' },
145 { "consumerd64-path", required_argument, 0, '\0' },
146 { "consumerd64-libdir", required_argument, 0, '\0' },
147 { "daemonize", no_argument, 0, 'd' },
148 { "background", no_argument, 0, 'b' },
149 { "sig-parent", no_argument, 0, 'S' },
150 { "help", no_argument, 0, 'h' },
151 { "group", required_argument, 0, 'g' },
152 { "version", no_argument, 0, 'V' },
153 { "quiet", no_argument, 0, 'q' },
154 { "verbose", no_argument, 0, 'v' },
155 { "verbose-consumer", no_argument, 0, '\0' },
156 { "no-kernel", no_argument, 0, '\0' },
157 { "pidfile", required_argument, 0, 'p' },
158 { "agent-tcp-port", required_argument, 0, '\0' },
159 { "config", required_argument, 0, 'f' },
160 { "load", required_argument, 0, 'l' },
161 { "kmod-probes", required_argument, 0, '\0' },
162 { "extra-kmod-probes", required_argument, 0, '\0' },
163 { NULL, 0, 0, 0 }
164 };
165
166 /* Command line options to ignore from configuration file */
167 static const char *config_ignore_options[] = { "help", "version", "config" };
168
169 /* Shared between threads */
170 static int dispatch_thread_exit;
171
172 /* Global application Unix socket path */
173 static char apps_unix_sock_path[PATH_MAX];
174 /* Global client Unix socket path */
175 static char client_unix_sock_path[PATH_MAX];
176 /* global wait shm path for UST */
177 static char wait_shm_path[PATH_MAX];
178 /* Global health check unix path */
179 static char health_unix_sock_path[PATH_MAX];
180
181 /* Sockets and FDs */
182 static int client_sock = -1;
183 static int apps_sock = -1;
184 int kernel_tracer_fd = -1;
185 static int kernel_poll_pipe[2] = { -1, -1 };
186
187 /*
188 * Quit pipe for all threads. This permits a single cancellation point
189 * for all threads when receiving an event on the pipe.
190 */
191 static int thread_quit_pipe[2] = { -1, -1 };
192 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
193
194 /*
195 * This pipe is used to inform the thread managing application communication
196 * that a command is queued and ready to be processed.
197 */
198 static int apps_cmd_pipe[2] = { -1, -1 };
199
200 int apps_cmd_notify_pipe[2] = { -1, -1 };
201
202 /* Pthread, Mutexes and Semaphores */
203 static pthread_t apps_thread;
204 static pthread_t apps_notify_thread;
205 static pthread_t reg_apps_thread;
206 static pthread_t client_thread;
207 static pthread_t kernel_thread;
208 static pthread_t dispatch_thread;
209 static pthread_t health_thread;
210 static pthread_t ht_cleanup_thread;
211 static pthread_t agent_reg_thread;
212 static pthread_t load_session_thread;
213
214 /*
215 * UST registration command queue. This queue is tied with a futex and uses a N
216 * wakers / 1 waiter implemented and detailed in futex.c/.h
217 *
218 * The thread_registration_apps and thread_dispatch_ust_registration uses this
219 * queue along with the wait/wake scheme. The thread_manage_apps receives down
220 * the line new application socket and monitors it for any I/O error or clean
221 * close that triggers an unregistration of the application.
222 */
223 static struct ust_cmd_queue ust_cmd_queue;
224
225 /*
226 * Pointer initialized before thread creation.
227 *
228 * This points to the tracing session list containing the session count and a
229 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
230 * MUST NOT be taken if you call a public function in session.c.
231 *
232 * The lock is nested inside the structure: session_list_ptr->lock. Please use
233 * session_lock_list and session_unlock_list for lock acquisition.
234 */
235 static struct ltt_session_list *session_list_ptr;
236
237 int ust_consumerd64_fd = -1;
238 int ust_consumerd32_fd = -1;
239
240 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
241 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
242 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
243 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
244 static int consumerd32_bin_override;
245 static int consumerd64_bin_override;
246 static int consumerd32_libdir_override;
247 static int consumerd64_libdir_override;
248
249 static const char *module_proc_lttng = "/proc/lttng";
250
251 /*
252 * Consumer daemon state which is changed when spawning it, killing it or in
253 * case of a fatal error.
254 */
255 enum consumerd_state {
256 CONSUMER_STARTED = 1,
257 CONSUMER_STOPPED = 2,
258 CONSUMER_ERROR = 3,
259 };
260
261 /*
262 * This consumer daemon state is used to validate if a client command will be
263 * able to reach the consumer. If not, the client is informed. For instance,
264 * doing a "lttng start" when the consumer state is set to ERROR will return an
265 * error to the client.
266 *
267 * The following example shows a possible race condition of this scheme:
268 *
269 * consumer thread error happens
270 * client cmd arrives
271 * client cmd checks state -> still OK
272 * consumer thread exit, sets error
273 * client cmd try to talk to consumer
274 * ...
275 *
276 * However, since the consumer is a different daemon, we have no way of making
277 * sure the command will reach it safely even with this state flag. This is why
278 * we consider that up to the state validation during command processing, the
279 * command is safe. After that, we can not guarantee the correctness of the
280 * client request vis-a-vis the consumer.
281 */
282 static enum consumerd_state ust_consumerd_state;
283 static enum consumerd_state kernel_consumerd_state;
284
285 /*
286 * Socket timeout for receiving and sending in seconds.
287 */
288 static int app_socket_timeout;
289
290 /* Set in main() with the current page size. */
291 long page_size;
292
293 /* Application health monitoring */
294 struct health_app *health_sessiond;
295
296 /* Agent TCP port for registration. Used by the agent thread. */
297 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
298
299 /* Am I root or not. */
300 int is_root; /* Set to 1 if the daemon is running as root */
301
302 const char * const config_section_name = "sessiond";
303
304 /* Load session thread information to operate. */
305 struct load_session_thread_data *load_info;
306
307 /* Global hash tables */
308 struct lttng_ht *agent_apps_ht_by_sock = NULL;
309
310 /*
311 * Whether sessiond is ready for commands/health check requests.
312 * NR_LTTNG_SESSIOND_READY must match the number of calls to
313 * sessiond_notify_ready().
314 */
315 #define NR_LTTNG_SESSIOND_READY 3
316 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
317
318 /* Notify parents that we are ready for cmd and health check */
319 LTTNG_HIDDEN
320 void sessiond_notify_ready(void)
321 {
322 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
323 /*
324 * Notify parent pid that we are ready to accept command
325 * for client side. This ppid is the one from the
326 * external process that spawned us.
327 */
328 if (opt_sig_parent) {
329 kill(ppid, SIGUSR1);
330 }
331
332 /*
333 * Notify the parent of the fork() process that we are
334 * ready.
335 */
336 if (opt_daemon || opt_background) {
337 kill(child_ppid, SIGUSR1);
338 }
339 }
340 }
341
342 static
343 void setup_consumerd_path(void)
344 {
345 const char *bin, *libdir;
346
347 /*
348 * Allow INSTALL_BIN_PATH to be used as a target path for the
349 * native architecture size consumer if CONFIG_CONSUMER*_PATH
350 * has not been defined.
351 */
352 #if (CAA_BITS_PER_LONG == 32)
353 if (!consumerd32_bin[0]) {
354 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
355 }
356 if (!consumerd32_libdir[0]) {
357 consumerd32_libdir = INSTALL_LIB_PATH;
358 }
359 #elif (CAA_BITS_PER_LONG == 64)
360 if (!consumerd64_bin[0]) {
361 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
362 }
363 if (!consumerd64_libdir[0]) {
364 consumerd64_libdir = INSTALL_LIB_PATH;
365 }
366 #else
367 #error "Unknown bitness"
368 #endif
369
370 /*
371 * runtime env. var. overrides the build default.
372 */
373 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
374 if (bin) {
375 consumerd32_bin = bin;
376 }
377 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
378 if (bin) {
379 consumerd64_bin = bin;
380 }
381 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
382 if (libdir) {
383 consumerd32_libdir = libdir;
384 }
385 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
386 if (libdir) {
387 consumerd64_libdir = libdir;
388 }
389 }
390
391 static
392 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
393 int *a_pipe)
394 {
395 int ret;
396
397 assert(events);
398
399 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
400 if (ret < 0) {
401 goto error;
402 }
403
404 /* Add quit pipe */
405 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
406 if (ret < 0) {
407 goto error;
408 }
409
410 return 0;
411
412 error:
413 return ret;
414 }
415
416 /*
417 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
418 */
419 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
420 {
421 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
422 }
423
424 /*
425 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
426 */
427 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
428 size_t size)
429 {
430 return __sessiond_set_thread_pollset(events, size,
431 ht_cleanup_quit_pipe);
432 }
433
434 static
435 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
436 {
437 if (fd == a_pipe && (events & LPOLLIN)) {
438 return 1;
439 }
440 return 0;
441 }
442
443 /*
444 * Check if the thread quit pipe was triggered.
445 *
446 * Return 1 if it was triggered else 0;
447 */
448 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
449 {
450 return __sessiond_check_thread_quit_pipe(fd, events,
451 thread_quit_pipe[0]);
452 }
453
454 /*
455 * Check if the ht_cleanup thread quit pipe was triggered.
456 *
457 * Return 1 if it was triggered else 0;
458 */
459 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
460 {
461 return __sessiond_check_thread_quit_pipe(fd, events,
462 ht_cleanup_quit_pipe[0]);
463 }
464
465 /*
466 * Init thread quit pipe.
467 *
468 * Return -1 on error or 0 if all pipes are created.
469 */
470 static int __init_thread_quit_pipe(int *a_pipe)
471 {
472 int ret, i;
473
474 ret = pipe(a_pipe);
475 if (ret < 0) {
476 PERROR("thread quit pipe");
477 goto error;
478 }
479
480 for (i = 0; i < 2; i++) {
481 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
482 if (ret < 0) {
483 PERROR("fcntl");
484 goto error;
485 }
486 }
487
488 error:
489 return ret;
490 }
491
492 static int init_thread_quit_pipe(void)
493 {
494 return __init_thread_quit_pipe(thread_quit_pipe);
495 }
496
497 static int init_ht_cleanup_quit_pipe(void)
498 {
499 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
500 }
501
502 /*
503 * Stop all threads by closing the thread quit pipe.
504 */
505 static void stop_threads(void)
506 {
507 int ret;
508
509 /* Stopping all threads */
510 DBG("Terminating all threads");
511 ret = notify_thread_pipe(thread_quit_pipe[1]);
512 if (ret < 0) {
513 ERR("write error on thread quit pipe");
514 }
515
516 /* Dispatch thread */
517 CMM_STORE_SHARED(dispatch_thread_exit, 1);
518 futex_nto1_wake(&ust_cmd_queue.futex);
519 }
520
521 /*
522 * Close every consumer sockets.
523 */
524 static void close_consumer_sockets(void)
525 {
526 int ret;
527
528 if (kconsumer_data.err_sock >= 0) {
529 ret = close(kconsumer_data.err_sock);
530 if (ret < 0) {
531 PERROR("kernel consumer err_sock close");
532 }
533 }
534 if (ustconsumer32_data.err_sock >= 0) {
535 ret = close(ustconsumer32_data.err_sock);
536 if (ret < 0) {
537 PERROR("UST consumerd32 err_sock close");
538 }
539 }
540 if (ustconsumer64_data.err_sock >= 0) {
541 ret = close(ustconsumer64_data.err_sock);
542 if (ret < 0) {
543 PERROR("UST consumerd64 err_sock close");
544 }
545 }
546 if (kconsumer_data.cmd_sock >= 0) {
547 ret = close(kconsumer_data.cmd_sock);
548 if (ret < 0) {
549 PERROR("kernel consumer cmd_sock close");
550 }
551 }
552 if (ustconsumer32_data.cmd_sock >= 0) {
553 ret = close(ustconsumer32_data.cmd_sock);
554 if (ret < 0) {
555 PERROR("UST consumerd32 cmd_sock close");
556 }
557 }
558 if (ustconsumer64_data.cmd_sock >= 0) {
559 ret = close(ustconsumer64_data.cmd_sock);
560 if (ret < 0) {
561 PERROR("UST consumerd64 cmd_sock close");
562 }
563 }
564 }
565
566 /*
567 * Generate the full lock file path using the rundir.
568 *
569 * Return the snprintf() return value thus a negative value is an error.
570 */
571 static int generate_lock_file_path(char *path, size_t len)
572 {
573 int ret;
574
575 assert(path);
576 assert(rundir);
577
578 /* Build lockfile path from rundir. */
579 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
580 if (ret < 0) {
581 PERROR("snprintf lockfile path");
582 }
583
584 return ret;
585 }
586
587 /*
588 * Wait on consumer process termination.
589 *
590 * Need to be called with the consumer data lock held or from a context
591 * ensuring no concurrent access to data (e.g: cleanup).
592 */
593 static void wait_consumer(struct consumer_data *consumer_data)
594 {
595 pid_t ret;
596 int status;
597
598 if (consumer_data->pid <= 0) {
599 return;
600 }
601
602 DBG("Waiting for complete teardown of consumerd (PID: %d)",
603 consumer_data->pid);
604 ret = waitpid(consumer_data->pid, &status, 0);
605 if (ret == -1) {
606 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
607 }
608 if (!WIFEXITED(status)) {
609 ERR("consumerd termination with error: %d",
610 WEXITSTATUS(ret));
611 }
612 consumer_data->pid = 0;
613 }
614
615 /*
616 * Cleanup the session daemon's data structures.
617 */
618 static void sessiond_cleanup(void)
619 {
620 int ret;
621 struct ltt_session *sess, *stmp;
622 char path[PATH_MAX];
623
624 DBG("Cleanup sessiond");
625
626 /*
627 * Close the thread quit pipe. It has already done its job,
628 * since we are now called.
629 */
630 utils_close_pipe(thread_quit_pipe);
631
632 /*
633 * If opt_pidfile is undefined, the default file will be wiped when
634 * removing the rundir.
635 */
636 if (opt_pidfile) {
637 ret = remove(opt_pidfile);
638 if (ret < 0) {
639 PERROR("remove pidfile %s", opt_pidfile);
640 }
641 }
642
643 DBG("Removing sessiond and consumerd content of directory %s", rundir);
644
645 /* sessiond */
646 snprintf(path, PATH_MAX,
647 "%s/%s",
648 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
649 DBG("Removing %s", path);
650 (void) unlink(path);
651
652 snprintf(path, PATH_MAX, "%s/%s", rundir,
653 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
654 DBG("Removing %s", path);
655 (void) unlink(path);
656
657 /* kconsumerd */
658 snprintf(path, PATH_MAX,
659 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
660 rundir);
661 DBG("Removing %s", path);
662 (void) unlink(path);
663
664 snprintf(path, PATH_MAX,
665 DEFAULT_KCONSUMERD_PATH,
666 rundir);
667 DBG("Removing directory %s", path);
668 (void) rmdir(path);
669
670 /* ust consumerd 32 */
671 snprintf(path, PATH_MAX,
672 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
673 rundir);
674 DBG("Removing %s", path);
675 (void) unlink(path);
676
677 snprintf(path, PATH_MAX,
678 DEFAULT_USTCONSUMERD32_PATH,
679 rundir);
680 DBG("Removing directory %s", path);
681 (void) rmdir(path);
682
683 /* ust consumerd 64 */
684 snprintf(path, PATH_MAX,
685 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
686 rundir);
687 DBG("Removing %s", path);
688 (void) unlink(path);
689
690 snprintf(path, PATH_MAX,
691 DEFAULT_USTCONSUMERD64_PATH,
692 rundir);
693 DBG("Removing directory %s", path);
694 (void) rmdir(path);
695
696 DBG("Cleaning up all sessions");
697
698 /* Destroy session list mutex */
699 if (session_list_ptr != NULL) {
700 pthread_mutex_destroy(&session_list_ptr->lock);
701
702 /* Cleanup ALL session */
703 cds_list_for_each_entry_safe(sess, stmp,
704 &session_list_ptr->head, list) {
705 cmd_destroy_session(sess, kernel_poll_pipe[1]);
706 }
707 }
708
709 wait_consumer(&kconsumer_data);
710 wait_consumer(&ustconsumer64_data);
711 wait_consumer(&ustconsumer32_data);
712
713 DBG("Cleaning up all agent apps");
714 agent_app_ht_clean();
715
716 DBG("Closing all UST sockets");
717 ust_app_clean_list();
718 buffer_reg_destroy_registries();
719
720 if (is_root && !opt_no_kernel) {
721 DBG2("Closing kernel fd");
722 if (kernel_tracer_fd >= 0) {
723 ret = close(kernel_tracer_fd);
724 if (ret) {
725 PERROR("close");
726 }
727 }
728 DBG("Unloading kernel modules");
729 modprobe_remove_lttng_all();
730 free(syscall_table);
731 }
732
733 close_consumer_sockets();
734
735 if (load_info) {
736 load_session_destroy_data(load_info);
737 free(load_info);
738 }
739
740 /*
741 * Cleanup lock file by deleting it and finaly closing it which will
742 * release the file system lock.
743 */
744 if (lockfile_fd >= 0) {
745 char lockfile_path[PATH_MAX];
746
747 ret = generate_lock_file_path(lockfile_path,
748 sizeof(lockfile_path));
749 if (ret > 0) {
750 ret = remove(lockfile_path);
751 if (ret < 0) {
752 PERROR("remove lock file");
753 }
754 ret = close(lockfile_fd);
755 if (ret < 0) {
756 PERROR("close lock file");
757 }
758 }
759 }
760
761 /*
762 * We do NOT rmdir rundir because there are other processes
763 * using it, for instance lttng-relayd, which can start in
764 * parallel with this teardown.
765 */
766
767 free(rundir);
768 }
769
770 /*
771 * Cleanup the daemon's option data structures.
772 */
773 static void sessiond_cleanup_options(void)
774 {
775 DBG("Cleaning up options");
776
777 /*
778 * If the override option is set, the pointer points to a *non* const
779 * thus freeing it even though the variable type is set to const.
780 */
781 if (tracing_group_name_override) {
782 free((void *) tracing_group_name);
783 }
784 if (consumerd32_bin_override) {
785 free((void *) consumerd32_bin);
786 }
787 if (consumerd64_bin_override) {
788 free((void *) consumerd64_bin);
789 }
790 if (consumerd32_libdir_override) {
791 free((void *) consumerd32_libdir);
792 }
793 if (consumerd64_libdir_override) {
794 free((void *) consumerd64_libdir);
795 }
796
797 free(opt_pidfile);
798 free(opt_load_session_path);
799 free(kmod_probes_list);
800 free(kmod_extra_probes_list);
801
802 run_as_destroy_worker();
803
804 /* <fun> */
805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 /* </fun> */
809 }
810
811 /*
812 * Send data on a unix socket using the liblttsessiondcomm API.
813 *
814 * Return lttcomm error code.
815 */
816 static int send_unix_sock(int sock, void *buf, size_t len)
817 {
818 /* Check valid length */
819 if (len == 0) {
820 return -1;
821 }
822
823 return lttcomm_send_unix_sock(sock, buf, len);
824 }
825
826 /*
827 * Free memory of a command context structure.
828 */
829 static void clean_command_ctx(struct command_ctx **cmd_ctx)
830 {
831 DBG("Clean command context structure");
832 if (*cmd_ctx) {
833 if ((*cmd_ctx)->llm) {
834 free((*cmd_ctx)->llm);
835 }
836 if ((*cmd_ctx)->lsm) {
837 free((*cmd_ctx)->lsm);
838 }
839 free(*cmd_ctx);
840 *cmd_ctx = NULL;
841 }
842 }
843
844 /*
845 * Notify UST applications using the shm mmap futex.
846 */
847 static int notify_ust_apps(int active)
848 {
849 char *wait_shm_mmap;
850
851 DBG("Notifying applications of session daemon state: %d", active);
852
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
855 if (wait_shm_mmap == NULL) {
856 goto error;
857 }
858
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap, active);
861
862 /* Apps notified successfully */
863 return 0;
864
865 error:
866 return -1;
867 }
868
869 /*
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
872 * structure.
873 *
874 * Return 0 on success, negative value on error.
875 */
876 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
877 const void *payload_buf, size_t payload_len,
878 const void *cmd_header_buf, size_t cmd_header_len)
879 {
880 int ret = 0;
881 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
882 const size_t cmd_header_offset = header_len;
883 const size_t payload_offset = cmd_header_offset + cmd_header_len;
884 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
885
886 cmd_ctx->llm = zmalloc(total_msg_size);
887
888 if (cmd_ctx->llm == NULL) {
889 PERROR("zmalloc");
890 ret = -ENOMEM;
891 goto end;
892 }
893
894 /* Copy common data */
895 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
896 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
897 cmd_ctx->llm->cmd_header_size = cmd_header_len;
898 cmd_ctx->llm->data_size = payload_len;
899 cmd_ctx->lttng_msg_size = total_msg_size;
900
901 /* Copy command header */
902 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
903 cmd_header_len);
904
905 /* Copy payload */
906 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
907 payload_len);
908
909 end:
910 return ret;
911 }
912
913 /*
914 * Version of setup_lttng_msg() without command header.
915 */
916 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
917 void *payload_buf, size_t payload_len)
918 {
919 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
920 }
921 /*
922 * Update the kernel poll set of all channel fd available over all tracing
923 * session. Add the wakeup pipe at the end of the set.
924 */
925 static int update_kernel_poll(struct lttng_poll_event *events)
926 {
927 int ret;
928 struct ltt_session *session;
929 struct ltt_kernel_channel *channel;
930
931 DBG("Updating kernel poll set");
932
933 session_lock_list();
934 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
935 session_lock(session);
936 if (session->kernel_session == NULL) {
937 session_unlock(session);
938 continue;
939 }
940
941 cds_list_for_each_entry(channel,
942 &session->kernel_session->channel_list.head, list) {
943 /* Add channel fd to the kernel poll set */
944 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
945 if (ret < 0) {
946 session_unlock(session);
947 goto error;
948 }
949 DBG("Channel fd %d added to kernel set", channel->fd);
950 }
951 session_unlock(session);
952 }
953 session_unlock_list();
954
955 return 0;
956
957 error:
958 session_unlock_list();
959 return -1;
960 }
961
962 /*
963 * Find the channel fd from 'fd' over all tracing session. When found, check
964 * for new channel stream and send those stream fds to the kernel consumer.
965 *
966 * Useful for CPU hotplug feature.
967 */
968 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
969 {
970 int ret = 0;
971 struct ltt_session *session;
972 struct ltt_kernel_session *ksess;
973 struct ltt_kernel_channel *channel;
974
975 DBG("Updating kernel streams for channel fd %d", fd);
976
977 session_lock_list();
978 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
979 session_lock(session);
980 if (session->kernel_session == NULL) {
981 session_unlock(session);
982 continue;
983 }
984 ksess = session->kernel_session;
985
986 cds_list_for_each_entry(channel,
987 &ksess->channel_list.head, list) {
988 struct lttng_ht_iter iter;
989 struct consumer_socket *socket;
990
991 if (channel->fd != fd) {
992 continue;
993 }
994 DBG("Channel found, updating kernel streams");
995 ret = kernel_open_channel_stream(channel);
996 if (ret < 0) {
997 goto error;
998 }
999 /* Update the stream global counter */
1000 ksess->stream_count_global += ret;
1001
1002 /*
1003 * Have we already sent fds to the consumer? If yes, it
1004 * means that tracing is started so it is safe to send
1005 * our updated stream fds.
1006 */
1007 if (ksess->consumer_fds_sent != 1
1008 || ksess->consumer == NULL) {
1009 ret = -1;
1010 goto error;
1011 }
1012
1013 rcu_read_lock();
1014 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
1015 &iter.iter, socket, node.node) {
1016 pthread_mutex_lock(socket->lock);
1017 ret = kernel_consumer_send_channel_stream(socket,
1018 channel, ksess,
1019 session->output_traces ? 1 : 0);
1020 pthread_mutex_unlock(socket->lock);
1021 if (ret < 0) {
1022 rcu_read_unlock();
1023 goto error;
1024 }
1025 }
1026 rcu_read_unlock();
1027 }
1028 session_unlock(session);
1029 }
1030 session_unlock_list();
1031 return ret;
1032
1033 error:
1034 session_unlock(session);
1035 session_unlock_list();
1036 return ret;
1037 }
1038
1039 /*
1040 * For each tracing session, update newly registered apps. The session list
1041 * lock MUST be acquired before calling this.
1042 */
1043 static void update_ust_app(int app_sock)
1044 {
1045 struct ltt_session *sess, *stmp;
1046
1047 /* Consumer is in an ERROR state. Stop any application update. */
1048 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1049 /* Stop the update process since the consumer is dead. */
1050 return;
1051 }
1052
1053 /* For all tracing session(s) */
1054 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1055 struct ust_app *app;
1056
1057 session_lock(sess);
1058 if (!sess->ust_session) {
1059 goto unlock_session;
1060 }
1061
1062 rcu_read_lock();
1063 assert(app_sock >= 0);
1064 app = ust_app_find_by_sock(app_sock);
1065 if (app == NULL) {
1066 /*
1067 * Application can be unregistered before so
1068 * this is possible hence simply stopping the
1069 * update.
1070 */
1071 DBG3("UST app update failed to find app sock %d",
1072 app_sock);
1073 goto unlock_rcu;
1074 }
1075 ust_app_global_update(sess->ust_session, app);
1076 unlock_rcu:
1077 rcu_read_unlock();
1078 unlock_session:
1079 session_unlock(sess);
1080 }
1081 }
1082
1083 /*
1084 * This thread manage event coming from the kernel.
1085 *
1086 * Features supported in this thread:
1087 * -) CPU Hotplug
1088 */
1089 static void *thread_manage_kernel(void *data)
1090 {
1091 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1092 uint32_t revents, nb_fd;
1093 char tmp;
1094 struct lttng_poll_event events;
1095
1096 DBG("[thread] Thread manage kernel started");
1097
1098 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1099
1100 /*
1101 * This first step of the while is to clean this structure which could free
1102 * non NULL pointers so initialize it before the loop.
1103 */
1104 lttng_poll_init(&events);
1105
1106 if (testpoint(sessiond_thread_manage_kernel)) {
1107 goto error_testpoint;
1108 }
1109
1110 health_code_update();
1111
1112 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1113 goto error_testpoint;
1114 }
1115
1116 while (1) {
1117 health_code_update();
1118
1119 if (update_poll_flag == 1) {
1120 /* Clean events object. We are about to populate it again. */
1121 lttng_poll_clean(&events);
1122
1123 ret = sessiond_set_thread_pollset(&events, 2);
1124 if (ret < 0) {
1125 goto error_poll_create;
1126 }
1127
1128 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1129 if (ret < 0) {
1130 goto error;
1131 }
1132
1133 /* This will add the available kernel channel if any. */
1134 ret = update_kernel_poll(&events);
1135 if (ret < 0) {
1136 goto error;
1137 }
1138 update_poll_flag = 0;
1139 }
1140
1141 DBG("Thread kernel polling");
1142
1143 /* Poll infinite value of time */
1144 restart:
1145 health_poll_entry();
1146 ret = lttng_poll_wait(&events, -1);
1147 DBG("Thread kernel return from poll on %d fds",
1148 LTTNG_POLL_GETNB(&events));
1149 health_poll_exit();
1150 if (ret < 0) {
1151 /*
1152 * Restart interrupted system call.
1153 */
1154 if (errno == EINTR) {
1155 goto restart;
1156 }
1157 goto error;
1158 } else if (ret == 0) {
1159 /* Should not happen since timeout is infinite */
1160 ERR("Return value of poll is 0 with an infinite timeout.\n"
1161 "This should not have happened! Continuing...");
1162 continue;
1163 }
1164
1165 nb_fd = ret;
1166
1167 for (i = 0; i < nb_fd; i++) {
1168 /* Fetch once the poll data */
1169 revents = LTTNG_POLL_GETEV(&events, i);
1170 pollfd = LTTNG_POLL_GETFD(&events, i);
1171
1172 health_code_update();
1173
1174 if (!revents) {
1175 /* No activity for this FD (poll implementation). */
1176 continue;
1177 }
1178
1179 /* Thread quit pipe has been closed. Killing thread. */
1180 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1181 if (ret) {
1182 err = 0;
1183 goto exit;
1184 }
1185
1186 /* Check for data on kernel pipe */
1187 if (revents & LPOLLIN) {
1188 if (pollfd == kernel_poll_pipe[0]) {
1189 (void) lttng_read(kernel_poll_pipe[0],
1190 &tmp, 1);
1191 /*
1192 * Ret value is useless here, if this pipe gets any actions an
1193 * update is required anyway.
1194 */
1195 update_poll_flag = 1;
1196 continue;
1197 } else {
1198 /*
1199 * New CPU detected by the kernel. Adding kernel stream to
1200 * kernel session and updating the kernel consumer
1201 */
1202 ret = update_kernel_stream(&kconsumer_data, pollfd);
1203 if (ret < 0) {
1204 continue;
1205 }
1206 break;
1207 }
1208 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1209 update_poll_flag = 1;
1210 continue;
1211 } else {
1212 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1213 goto error;
1214 }
1215 }
1216 }
1217
1218 exit:
1219 error:
1220 lttng_poll_clean(&events);
1221 error_poll_create:
1222 error_testpoint:
1223 utils_close_pipe(kernel_poll_pipe);
1224 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1225 if (err) {
1226 health_error();
1227 ERR("Health error occurred in %s", __func__);
1228 WARN("Kernel thread died unexpectedly. "
1229 "Kernel tracing can continue but CPU hotplug is disabled.");
1230 }
1231 health_unregister(health_sessiond);
1232 DBG("Kernel thread dying");
1233 return NULL;
1234 }
1235
1236 /*
1237 * Signal pthread condition of the consumer data that the thread.
1238 */
1239 static void signal_consumer_condition(struct consumer_data *data, int state)
1240 {
1241 pthread_mutex_lock(&data->cond_mutex);
1242
1243 /*
1244 * The state is set before signaling. It can be any value, it's the waiter
1245 * job to correctly interpret this condition variable associated to the
1246 * consumer pthread_cond.
1247 *
1248 * A value of 0 means that the corresponding thread of the consumer data
1249 * was not started. 1 indicates that the thread has started and is ready
1250 * for action. A negative value means that there was an error during the
1251 * thread bootstrap.
1252 */
1253 data->consumer_thread_is_ready = state;
1254 (void) pthread_cond_signal(&data->cond);
1255
1256 pthread_mutex_unlock(&data->cond_mutex);
1257 }
1258
1259 /*
1260 * This thread manage the consumer error sent back to the session daemon.
1261 */
1262 static void *thread_manage_consumer(void *data)
1263 {
1264 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1265 uint32_t revents, nb_fd;
1266 enum lttcomm_return_code code;
1267 struct lttng_poll_event events;
1268 struct consumer_data *consumer_data = data;
1269
1270 DBG("[thread] Manage consumer started");
1271
1272 rcu_register_thread();
1273 rcu_thread_online();
1274
1275 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1276
1277 health_code_update();
1278
1279 /*
1280 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1281 * metadata_sock. Nothing more will be added to this poll set.
1282 */
1283 ret = sessiond_set_thread_pollset(&events, 3);
1284 if (ret < 0) {
1285 goto error_poll;
1286 }
1287
1288 /*
1289 * The error socket here is already in a listening state which was done
1290 * just before spawning this thread to avoid a race between the consumer
1291 * daemon exec trying to connect and the listen() call.
1292 */
1293 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1294 if (ret < 0) {
1295 goto error;
1296 }
1297
1298 health_code_update();
1299
1300 /* Infinite blocking call, waiting for transmission */
1301 restart:
1302 health_poll_entry();
1303
1304 if (testpoint(sessiond_thread_manage_consumer)) {
1305 goto error;
1306 }
1307
1308 ret = lttng_poll_wait(&events, -1);
1309 health_poll_exit();
1310 if (ret < 0) {
1311 /*
1312 * Restart interrupted system call.
1313 */
1314 if (errno == EINTR) {
1315 goto restart;
1316 }
1317 goto error;
1318 }
1319
1320 nb_fd = ret;
1321
1322 for (i = 0; i < nb_fd; i++) {
1323 /* Fetch once the poll data */
1324 revents = LTTNG_POLL_GETEV(&events, i);
1325 pollfd = LTTNG_POLL_GETFD(&events, i);
1326
1327 health_code_update();
1328
1329 if (!revents) {
1330 /* No activity for this FD (poll implementation). */
1331 continue;
1332 }
1333
1334 /* Thread quit pipe has been closed. Killing thread. */
1335 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1336 if (ret) {
1337 err = 0;
1338 goto exit;
1339 }
1340
1341 /* Event on the registration socket */
1342 if (pollfd == consumer_data->err_sock) {
1343 if (revents & LPOLLIN) {
1344 continue;
1345 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1346 ERR("consumer err socket poll error");
1347 goto error;
1348 } else {
1349 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1350 goto error;
1351 }
1352 }
1353 }
1354
1355 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1356 if (sock < 0) {
1357 goto error;
1358 }
1359
1360 /*
1361 * Set the CLOEXEC flag. Return code is useless because either way, the
1362 * show must go on.
1363 */
1364 (void) utils_set_fd_cloexec(sock);
1365
1366 health_code_update();
1367
1368 DBG2("Receiving code from consumer err_sock");
1369
1370 /* Getting status code from kconsumerd */
1371 ret = lttcomm_recv_unix_sock(sock, &code,
1372 sizeof(enum lttcomm_return_code));
1373 if (ret <= 0) {
1374 goto error;
1375 }
1376
1377 health_code_update();
1378 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1379 /* Connect both socket, command and metadata. */
1380 consumer_data->cmd_sock =
1381 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1382 consumer_data->metadata_fd =
1383 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1384 if (consumer_data->cmd_sock < 0
1385 || 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 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1392 /* Create metadata socket lock. */
1393 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1394 if (consumer_data->metadata_sock.lock == NULL) {
1395 PERROR("zmalloc pthread mutex");
1396 ret = -1;
1397 goto error;
1398 }
1399 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1400
1401 signal_consumer_condition(consumer_data, 1);
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 } else {
1406 ERR("consumer error when waiting for SOCK_READY : %s",
1407 lttcomm_get_readable_code(-code));
1408 goto error;
1409 }
1410
1411 /* Remove the consumerd error sock since we've established a connexion */
1412 ret = lttng_poll_del(&events, consumer_data->err_sock);
1413 if (ret < 0) {
1414 goto error;
1415 }
1416
1417 /* Add new accepted error socket. */
1418 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1419 if (ret < 0) {
1420 goto error;
1421 }
1422
1423 /* Add metadata socket that is successfully connected. */
1424 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1425 LPOLLIN | LPOLLRDHUP);
1426 if (ret < 0) {
1427 goto error;
1428 }
1429
1430 health_code_update();
1431
1432 /* Infinite blocking call, waiting for transmission */
1433 restart_poll:
1434 while (1) {
1435 health_code_update();
1436
1437 /* Exit the thread because the thread quit pipe has been triggered. */
1438 if (should_quit) {
1439 /* Not a health error. */
1440 err = 0;
1441 goto exit;
1442 }
1443
1444 health_poll_entry();
1445 ret = lttng_poll_wait(&events, -1);
1446 health_poll_exit();
1447 if (ret < 0) {
1448 /*
1449 * Restart interrupted system call.
1450 */
1451 if (errno == EINTR) {
1452 goto restart_poll;
1453 }
1454 goto error;
1455 }
1456
1457 nb_fd = ret;
1458
1459 for (i = 0; i < nb_fd; i++) {
1460 /* Fetch once the poll data */
1461 revents = LTTNG_POLL_GETEV(&events, i);
1462 pollfd = LTTNG_POLL_GETFD(&events, i);
1463
1464 health_code_update();
1465
1466 if (!revents) {
1467 /* No activity for this FD (poll implementation). */
1468 continue;
1469 }
1470
1471 /*
1472 * Thread quit pipe has been triggered, flag that we should stop
1473 * but continue the current loop to handle potential data from
1474 * consumer.
1475 */
1476 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1477
1478 if (pollfd == sock) {
1479 /* Event on the consumerd socket */
1480 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1481 && !(revents & LPOLLIN)) {
1482 ERR("consumer err socket second poll error");
1483 goto error;
1484 }
1485 health_code_update();
1486 /* Wait for any kconsumerd error */
1487 ret = lttcomm_recv_unix_sock(sock, &code,
1488 sizeof(enum lttcomm_return_code));
1489 if (ret <= 0) {
1490 ERR("consumer closed the command socket");
1491 goto error;
1492 }
1493
1494 ERR("consumer return code : %s",
1495 lttcomm_get_readable_code(-code));
1496
1497 goto exit;
1498 } else if (pollfd == consumer_data->metadata_fd) {
1499 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1500 && !(revents & LPOLLIN)) {
1501 ERR("consumer err metadata socket second poll error");
1502 goto error;
1503 }
1504 /* UST metadata requests */
1505 ret = ust_consumer_metadata_request(
1506 &consumer_data->metadata_sock);
1507 if (ret < 0) {
1508 ERR("Handling metadata request");
1509 goto error;
1510 }
1511 }
1512 /* No need for an else branch all FDs are tested prior. */
1513 }
1514 health_code_update();
1515 }
1516
1517 exit:
1518 error:
1519 /*
1520 * We lock here because we are about to close the sockets and some other
1521 * thread might be using them so get exclusive access which will abort all
1522 * other consumer command by other threads.
1523 */
1524 pthread_mutex_lock(&consumer_data->lock);
1525
1526 /* Immediately set the consumerd state to stopped */
1527 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1528 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1529 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1530 consumer_data->type == LTTNG_CONSUMER32_UST) {
1531 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1532 } else {
1533 /* Code flow error... */
1534 assert(0);
1535 }
1536
1537 if (consumer_data->err_sock >= 0) {
1538 ret = close(consumer_data->err_sock);
1539 if (ret) {
1540 PERROR("close");
1541 }
1542 consumer_data->err_sock = -1;
1543 }
1544 if (consumer_data->cmd_sock >= 0) {
1545 ret = close(consumer_data->cmd_sock);
1546 if (ret) {
1547 PERROR("close");
1548 }
1549 consumer_data->cmd_sock = -1;
1550 }
1551 if (consumer_data->metadata_sock.fd_ptr &&
1552 *consumer_data->metadata_sock.fd_ptr >= 0) {
1553 ret = close(*consumer_data->metadata_sock.fd_ptr);
1554 if (ret) {
1555 PERROR("close");
1556 }
1557 }
1558 if (sock >= 0) {
1559 ret = close(sock);
1560 if (ret) {
1561 PERROR("close");
1562 }
1563 }
1564
1565 unlink(consumer_data->err_unix_sock_path);
1566 unlink(consumer_data->cmd_unix_sock_path);
1567 pthread_mutex_unlock(&consumer_data->lock);
1568
1569 /* Cleanup metadata socket mutex. */
1570 if (consumer_data->metadata_sock.lock) {
1571 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1572 free(consumer_data->metadata_sock.lock);
1573 }
1574 lttng_poll_clean(&events);
1575 error_poll:
1576 if (err) {
1577 health_error();
1578 ERR("Health error occurred in %s", __func__);
1579 }
1580 health_unregister(health_sessiond);
1581 DBG("consumer thread cleanup completed");
1582
1583 rcu_thread_offline();
1584 rcu_unregister_thread();
1585
1586 return NULL;
1587 }
1588
1589 /*
1590 * This thread manage application communication.
1591 */
1592 static void *thread_manage_apps(void *data)
1593 {
1594 int i, ret, pollfd, err = -1;
1595 ssize_t size_ret;
1596 uint32_t revents, nb_fd;
1597 struct lttng_poll_event events;
1598
1599 DBG("[thread] Manage application started");
1600
1601 rcu_register_thread();
1602 rcu_thread_online();
1603
1604 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1605
1606 if (testpoint(sessiond_thread_manage_apps)) {
1607 goto error_testpoint;
1608 }
1609
1610 health_code_update();
1611
1612 ret = sessiond_set_thread_pollset(&events, 2);
1613 if (ret < 0) {
1614 goto error_poll_create;
1615 }
1616
1617 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1618 if (ret < 0) {
1619 goto error;
1620 }
1621
1622 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1623 goto error;
1624 }
1625
1626 health_code_update();
1627
1628 while (1) {
1629 DBG("Apps thread polling");
1630
1631 /* Inifinite blocking call, waiting for transmission */
1632 restart:
1633 health_poll_entry();
1634 ret = lttng_poll_wait(&events, -1);
1635 DBG("Apps thread return from poll on %d fds",
1636 LTTNG_POLL_GETNB(&events));
1637 health_poll_exit();
1638 if (ret < 0) {
1639 /*
1640 * Restart interrupted system call.
1641 */
1642 if (errno == EINTR) {
1643 goto restart;
1644 }
1645 goto error;
1646 }
1647
1648 nb_fd = ret;
1649
1650 for (i = 0; i < nb_fd; i++) {
1651 /* Fetch once the poll data */
1652 revents = LTTNG_POLL_GETEV(&events, i);
1653 pollfd = LTTNG_POLL_GETFD(&events, i);
1654
1655 health_code_update();
1656
1657 if (!revents) {
1658 /* No activity for this FD (poll implementation). */
1659 continue;
1660 }
1661
1662 /* Thread quit pipe has been closed. Killing thread. */
1663 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1664 if (ret) {
1665 err = 0;
1666 goto exit;
1667 }
1668
1669 /* Inspect the apps cmd pipe */
1670 if (pollfd == apps_cmd_pipe[0]) {
1671 if (revents & LPOLLIN) {
1672 int sock;
1673
1674 /* Empty pipe */
1675 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1676 if (size_ret < sizeof(sock)) {
1677 PERROR("read apps cmd pipe");
1678 goto error;
1679 }
1680
1681 health_code_update();
1682
1683 /*
1684 * Since this is a command socket (write then read),
1685 * we only monitor the error events of the socket.
1686 */
1687 ret = lttng_poll_add(&events, sock,
1688 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1689 if (ret < 0) {
1690 goto error;
1691 }
1692
1693 DBG("Apps with sock %d added to poll set", sock);
1694 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1695 ERR("Apps command pipe error");
1696 goto error;
1697 } else {
1698 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1699 goto error;
1700 }
1701 } else {
1702 /*
1703 * At this point, we know that a registered application made
1704 * the event at poll_wait.
1705 */
1706 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1707 /* Removing from the poll set */
1708 ret = lttng_poll_del(&events, pollfd);
1709 if (ret < 0) {
1710 goto error;
1711 }
1712
1713 /* Socket closed on remote end. */
1714 ust_app_unregister(pollfd);
1715 } else {
1716 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1717 goto error;
1718 }
1719 }
1720
1721 health_code_update();
1722 }
1723 }
1724
1725 exit:
1726 error:
1727 lttng_poll_clean(&events);
1728 error_poll_create:
1729 error_testpoint:
1730 utils_close_pipe(apps_cmd_pipe);
1731 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1732
1733 /*
1734 * We don't clean the UST app hash table here since already registered
1735 * applications can still be controlled so let them be until the session
1736 * daemon dies or the applications stop.
1737 */
1738
1739 if (err) {
1740 health_error();
1741 ERR("Health error occurred in %s", __func__);
1742 }
1743 health_unregister(health_sessiond);
1744 DBG("Application communication apps thread cleanup complete");
1745 rcu_thread_offline();
1746 rcu_unregister_thread();
1747 return NULL;
1748 }
1749
1750 /*
1751 * Send a socket to a thread This is called from the dispatch UST registration
1752 * thread once all sockets are set for the application.
1753 *
1754 * The sock value can be invalid, we don't really care, the thread will handle
1755 * it and make the necessary cleanup if so.
1756 *
1757 * On success, return 0 else a negative value being the errno message of the
1758 * write().
1759 */
1760 static int send_socket_to_thread(int fd, int sock)
1761 {
1762 ssize_t ret;
1763
1764 /*
1765 * It's possible that the FD is set as invalid with -1 concurrently just
1766 * before calling this function being a shutdown state of the thread.
1767 */
1768 if (fd < 0) {
1769 ret = -EBADF;
1770 goto error;
1771 }
1772
1773 ret = lttng_write(fd, &sock, sizeof(sock));
1774 if (ret < sizeof(sock)) {
1775 PERROR("write apps pipe %d", fd);
1776 if (ret < 0) {
1777 ret = -errno;
1778 }
1779 goto error;
1780 }
1781
1782 /* All good. Don't send back the write positive ret value. */
1783 ret = 0;
1784 error:
1785 return (int) ret;
1786 }
1787
1788 /*
1789 * Sanitize the wait queue of the dispatch registration thread meaning removing
1790 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1791 * notify socket is never received.
1792 */
1793 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1794 {
1795 int ret, nb_fd = 0, i;
1796 unsigned int fd_added = 0;
1797 struct lttng_poll_event events;
1798 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1799
1800 assert(wait_queue);
1801
1802 lttng_poll_init(&events);
1803
1804 /* Just skip everything for an empty queue. */
1805 if (!wait_queue->count) {
1806 goto end;
1807 }
1808
1809 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1810 if (ret < 0) {
1811 goto error_create;
1812 }
1813
1814 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1815 &wait_queue->head, head) {
1816 assert(wait_node->app);
1817 ret = lttng_poll_add(&events, wait_node->app->sock,
1818 LPOLLHUP | LPOLLERR);
1819 if (ret < 0) {
1820 goto error;
1821 }
1822
1823 fd_added = 1;
1824 }
1825
1826 if (!fd_added) {
1827 goto end;
1828 }
1829
1830 /*
1831 * Poll but don't block so we can quickly identify the faulty events and
1832 * clean them afterwards from the wait queue.
1833 */
1834 ret = lttng_poll_wait(&events, 0);
1835 if (ret < 0) {
1836 goto error;
1837 }
1838 nb_fd = ret;
1839
1840 for (i = 0; i < nb_fd; i++) {
1841 /* Get faulty FD. */
1842 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1843 int pollfd = LTTNG_POLL_GETFD(&events, i);
1844
1845 if (!revents) {
1846 /* No activity for this FD (poll implementation). */
1847 continue;
1848 }
1849
1850 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1851 &wait_queue->head, head) {
1852 if (pollfd == wait_node->app->sock &&
1853 (revents & (LPOLLHUP | LPOLLERR))) {
1854 cds_list_del(&wait_node->head);
1855 wait_queue->count--;
1856 ust_app_destroy(wait_node->app);
1857 free(wait_node);
1858 /*
1859 * Silence warning of use-after-free in
1860 * cds_list_for_each_entry_safe which uses
1861 * __typeof__(*wait_node).
1862 */
1863 wait_node = NULL;
1864 break;
1865 } else {
1866 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1867 goto error;
1868 }
1869 }
1870 }
1871
1872 if (nb_fd > 0) {
1873 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1874 }
1875
1876 end:
1877 lttng_poll_clean(&events);
1878 return;
1879
1880 error:
1881 lttng_poll_clean(&events);
1882 error_create:
1883 ERR("Unable to sanitize wait queue");
1884 return;
1885 }
1886
1887 /*
1888 * Dispatch request from the registration threads to the application
1889 * communication thread.
1890 */
1891 static void *thread_dispatch_ust_registration(void *data)
1892 {
1893 int ret, err = -1;
1894 struct cds_wfcq_node *node;
1895 struct ust_command *ust_cmd = NULL;
1896 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1897 struct ust_reg_wait_queue wait_queue = {
1898 .count = 0,
1899 };
1900
1901 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1902
1903 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1904 goto error_testpoint;
1905 }
1906
1907 health_code_update();
1908
1909 CDS_INIT_LIST_HEAD(&wait_queue.head);
1910
1911 DBG("[thread] Dispatch UST command started");
1912
1913 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1914 health_code_update();
1915
1916 /* Atomically prepare the queue futex */
1917 futex_nto1_prepare(&ust_cmd_queue.futex);
1918
1919 do {
1920 struct ust_app *app = NULL;
1921 ust_cmd = NULL;
1922
1923 /*
1924 * Make sure we don't have node(s) that have hung up before receiving
1925 * the notify socket. This is to clean the list in order to avoid
1926 * memory leaks from notify socket that are never seen.
1927 */
1928 sanitize_wait_queue(&wait_queue);
1929
1930 health_code_update();
1931 /* Dequeue command for registration */
1932 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1933 if (node == NULL) {
1934 DBG("Woken up but nothing in the UST command queue");
1935 /* Continue thread execution */
1936 break;
1937 }
1938
1939 ust_cmd = caa_container_of(node, struct ust_command, node);
1940
1941 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1942 " gid:%d sock:%d name:%s (version %d.%d)",
1943 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1944 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1945 ust_cmd->sock, ust_cmd->reg_msg.name,
1946 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1947
1948 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1949 wait_node = zmalloc(sizeof(*wait_node));
1950 if (!wait_node) {
1951 PERROR("zmalloc wait_node dispatch");
1952 ret = close(ust_cmd->sock);
1953 if (ret < 0) {
1954 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1955 }
1956 lttng_fd_put(LTTNG_FD_APPS, 1);
1957 free(ust_cmd);
1958 goto error;
1959 }
1960 CDS_INIT_LIST_HEAD(&wait_node->head);
1961
1962 /* Create application object if socket is CMD. */
1963 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1964 ust_cmd->sock);
1965 if (!wait_node->app) {
1966 ret = close(ust_cmd->sock);
1967 if (ret < 0) {
1968 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1969 }
1970 lttng_fd_put(LTTNG_FD_APPS, 1);
1971 free(wait_node);
1972 free(ust_cmd);
1973 continue;
1974 }
1975 /*
1976 * Add application to the wait queue so we can set the notify
1977 * socket before putting this object in the global ht.
1978 */
1979 cds_list_add(&wait_node->head, &wait_queue.head);
1980 wait_queue.count++;
1981
1982 free(ust_cmd);
1983 /*
1984 * We have to continue here since we don't have the notify
1985 * socket and the application MUST be added to the hash table
1986 * only at that moment.
1987 */
1988 continue;
1989 } else {
1990 /*
1991 * Look for the application in the local wait queue and set the
1992 * notify socket if found.
1993 */
1994 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1995 &wait_queue.head, head) {
1996 health_code_update();
1997 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1998 wait_node->app->notify_sock = ust_cmd->sock;
1999 cds_list_del(&wait_node->head);
2000 wait_queue.count--;
2001 app = wait_node->app;
2002 free(wait_node);
2003 DBG3("UST app notify socket %d is set", ust_cmd->sock);
2004 break;
2005 }
2006 }
2007
2008 /*
2009 * With no application at this stage the received socket is
2010 * basically useless so close it before we free the cmd data
2011 * structure for good.
2012 */
2013 if (!app) {
2014 ret = close(ust_cmd->sock);
2015 if (ret < 0) {
2016 PERROR("close ust sock dispatch %d", ust_cmd->sock);
2017 }
2018 lttng_fd_put(LTTNG_FD_APPS, 1);
2019 }
2020 free(ust_cmd);
2021 }
2022
2023 if (app) {
2024 /*
2025 * @session_lock_list
2026 *
2027 * Lock the global session list so from the register up to the
2028 * registration done message, no thread can see the application
2029 * and change its state.
2030 */
2031 session_lock_list();
2032 rcu_read_lock();
2033
2034 /*
2035 * Add application to the global hash table. This needs to be
2036 * done before the update to the UST registry can locate the
2037 * application.
2038 */
2039 ust_app_add(app);
2040
2041 /* Set app version. This call will print an error if needed. */
2042 (void) ust_app_version(app);
2043
2044 /* Send notify socket through the notify pipe. */
2045 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2046 app->notify_sock);
2047 if (ret < 0) {
2048 rcu_read_unlock();
2049 session_unlock_list();
2050 /*
2051 * No notify thread, stop the UST tracing. However, this is
2052 * not an internal error of the this thread thus setting
2053 * the health error code to a normal exit.
2054 */
2055 err = 0;
2056 goto error;
2057 }
2058
2059 /*
2060 * Update newly registered application with the tracing
2061 * registry info already enabled information.
2062 */
2063 update_ust_app(app->sock);
2064
2065 /*
2066 * Don't care about return value. Let the manage apps threads
2067 * handle app unregistration upon socket close.
2068 */
2069 (void) ust_app_register_done(app);
2070
2071 /*
2072 * Even if the application socket has been closed, send the app
2073 * to the thread and unregistration will take place at that
2074 * place.
2075 */
2076 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2077 if (ret < 0) {
2078 rcu_read_unlock();
2079 session_unlock_list();
2080 /*
2081 * No apps. thread, stop the UST tracing. However, this is
2082 * not an internal error of the this thread thus setting
2083 * the health error code to a normal exit.
2084 */
2085 err = 0;
2086 goto error;
2087 }
2088
2089 rcu_read_unlock();
2090 session_unlock_list();
2091 }
2092 } while (node != NULL);
2093
2094 health_poll_entry();
2095 /* Futex wait on queue. Blocking call on futex() */
2096 futex_nto1_wait(&ust_cmd_queue.futex);
2097 health_poll_exit();
2098 }
2099 /* Normal exit, no error */
2100 err = 0;
2101
2102 error:
2103 /* Clean up wait queue. */
2104 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2105 &wait_queue.head, head) {
2106 cds_list_del(&wait_node->head);
2107 wait_queue.count--;
2108 free(wait_node);
2109 }
2110
2111 /* Empty command queue. */
2112 for (;;) {
2113 /* Dequeue command for registration */
2114 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2115 if (node == NULL) {
2116 break;
2117 }
2118 ust_cmd = caa_container_of(node, struct ust_command, node);
2119 ret = close(ust_cmd->sock);
2120 if (ret < 0) {
2121 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2122 }
2123 lttng_fd_put(LTTNG_FD_APPS, 1);
2124 free(ust_cmd);
2125 }
2126
2127 error_testpoint:
2128 DBG("Dispatch thread dying");
2129 if (err) {
2130 health_error();
2131 ERR("Health error occurred in %s", __func__);
2132 }
2133 health_unregister(health_sessiond);
2134 return NULL;
2135 }
2136
2137 /*
2138 * This thread manage application registration.
2139 */
2140 static void *thread_registration_apps(void *data)
2141 {
2142 int sock = -1, i, ret, pollfd, err = -1;
2143 uint32_t revents, nb_fd;
2144 struct lttng_poll_event events;
2145 /*
2146 * Get allocated in this thread, enqueued to a global queue, dequeued and
2147 * freed in the manage apps thread.
2148 */
2149 struct ust_command *ust_cmd = NULL;
2150
2151 DBG("[thread] Manage application registration started");
2152
2153 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2154
2155 if (testpoint(sessiond_thread_registration_apps)) {
2156 goto error_testpoint;
2157 }
2158
2159 ret = lttcomm_listen_unix_sock(apps_sock);
2160 if (ret < 0) {
2161 goto error_listen;
2162 }
2163
2164 /*
2165 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2166 * more will be added to this poll set.
2167 */
2168 ret = sessiond_set_thread_pollset(&events, 2);
2169 if (ret < 0) {
2170 goto error_create_poll;
2171 }
2172
2173 /* Add the application registration socket */
2174 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2175 if (ret < 0) {
2176 goto error_poll_add;
2177 }
2178
2179 /* Notify all applications to register */
2180 ret = notify_ust_apps(1);
2181 if (ret < 0) {
2182 ERR("Failed to notify applications or create the wait shared memory.\n"
2183 "Execution continues but there might be problem for already\n"
2184 "running applications that wishes to register.");
2185 }
2186
2187 while (1) {
2188 DBG("Accepting application registration");
2189
2190 /* Inifinite blocking call, waiting for transmission */
2191 restart:
2192 health_poll_entry();
2193 ret = lttng_poll_wait(&events, -1);
2194 health_poll_exit();
2195 if (ret < 0) {
2196 /*
2197 * Restart interrupted system call.
2198 */
2199 if (errno == EINTR) {
2200 goto restart;
2201 }
2202 goto error;
2203 }
2204
2205 nb_fd = ret;
2206
2207 for (i = 0; i < nb_fd; i++) {
2208 health_code_update();
2209
2210 /* Fetch once the poll data */
2211 revents = LTTNG_POLL_GETEV(&events, i);
2212 pollfd = LTTNG_POLL_GETFD(&events, i);
2213
2214 if (!revents) {
2215 /* No activity for this FD (poll implementation). */
2216 continue;
2217 }
2218
2219 /* Thread quit pipe has been closed. Killing thread. */
2220 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2221 if (ret) {
2222 err = 0;
2223 goto exit;
2224 }
2225
2226 /* Event on the registration socket */
2227 if (pollfd == apps_sock) {
2228 if (revents & LPOLLIN) {
2229 sock = lttcomm_accept_unix_sock(apps_sock);
2230 if (sock < 0) {
2231 goto error;
2232 }
2233
2234 /*
2235 * Set socket timeout for both receiving and ending.
2236 * app_socket_timeout is in seconds, whereas
2237 * lttcomm_setsockopt_rcv_timeout and
2238 * lttcomm_setsockopt_snd_timeout expect msec as
2239 * parameter.
2240 */
2241 (void) lttcomm_setsockopt_rcv_timeout(sock,
2242 app_socket_timeout * 1000);
2243 (void) lttcomm_setsockopt_snd_timeout(sock,
2244 app_socket_timeout * 1000);
2245
2246 /*
2247 * Set the CLOEXEC flag. Return code is useless because
2248 * either way, the show must go on.
2249 */
2250 (void) utils_set_fd_cloexec(sock);
2251
2252 /* Create UST registration command for enqueuing */
2253 ust_cmd = zmalloc(sizeof(struct ust_command));
2254 if (ust_cmd == NULL) {
2255 PERROR("ust command zmalloc");
2256 ret = close(sock);
2257 if (ret) {
2258 PERROR("close");
2259 }
2260 goto error;
2261 }
2262
2263 /*
2264 * Using message-based transmissions to ensure we don't
2265 * have to deal with partially received messages.
2266 */
2267 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2268 if (ret < 0) {
2269 ERR("Exhausted file descriptors allowed for applications.");
2270 free(ust_cmd);
2271 ret = close(sock);
2272 if (ret) {
2273 PERROR("close");
2274 }
2275 sock = -1;
2276 continue;
2277 }
2278
2279 health_code_update();
2280 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2281 if (ret < 0) {
2282 free(ust_cmd);
2283 /* Close socket of the application. */
2284 ret = close(sock);
2285 if (ret) {
2286 PERROR("close");
2287 }
2288 lttng_fd_put(LTTNG_FD_APPS, 1);
2289 sock = -1;
2290 continue;
2291 }
2292 health_code_update();
2293
2294 ust_cmd->sock = sock;
2295 sock = -1;
2296
2297 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2298 " gid:%d sock:%d name:%s (version %d.%d)",
2299 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2300 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2301 ust_cmd->sock, ust_cmd->reg_msg.name,
2302 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2303
2304 /*
2305 * Lock free enqueue the registration request. The red pill
2306 * has been taken! This apps will be part of the *system*.
2307 */
2308 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2309
2310 /*
2311 * Wake the registration queue futex. Implicit memory
2312 * barrier with the exchange in cds_wfcq_enqueue.
2313 */
2314 futex_nto1_wake(&ust_cmd_queue.futex);
2315 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2316 ERR("Register apps socket poll error");
2317 goto error;
2318 } else {
2319 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2320 goto error;
2321 }
2322 }
2323 }
2324 }
2325
2326 exit:
2327 error:
2328 /* Notify that the registration thread is gone */
2329 notify_ust_apps(0);
2330
2331 if (apps_sock >= 0) {
2332 ret = close(apps_sock);
2333 if (ret) {
2334 PERROR("close");
2335 }
2336 }
2337 if (sock >= 0) {
2338 ret = close(sock);
2339 if (ret) {
2340 PERROR("close");
2341 }
2342 lttng_fd_put(LTTNG_FD_APPS, 1);
2343 }
2344 unlink(apps_unix_sock_path);
2345
2346 error_poll_add:
2347 lttng_poll_clean(&events);
2348 error_listen:
2349 error_create_poll:
2350 error_testpoint:
2351 DBG("UST Registration thread cleanup complete");
2352 if (err) {
2353 health_error();
2354 ERR("Health error occurred in %s", __func__);
2355 }
2356 health_unregister(health_sessiond);
2357
2358 return NULL;
2359 }
2360
2361 /*
2362 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2363 * exec or it will fails.
2364 */
2365 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2366 {
2367 int ret, clock_ret;
2368 struct timespec timeout;
2369
2370 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2371 consumer_data->consumer_thread_is_ready = 0;
2372
2373 /* Setup pthread condition */
2374 ret = pthread_condattr_init(&consumer_data->condattr);
2375 if (ret) {
2376 errno = ret;
2377 PERROR("pthread_condattr_init consumer data");
2378 goto error;
2379 }
2380
2381 /*
2382 * Set the monotonic clock in order to make sure we DO NOT jump in time
2383 * between the clock_gettime() call and the timedwait call. See bug #324
2384 * for a more details and how we noticed it.
2385 */
2386 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2387 if (ret) {
2388 errno = ret;
2389 PERROR("pthread_condattr_setclock consumer data");
2390 goto error;
2391 }
2392
2393 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2394 if (ret) {
2395 errno = ret;
2396 PERROR("pthread_cond_init consumer data");
2397 goto error;
2398 }
2399
2400 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2401 consumer_data);
2402 if (ret) {
2403 errno = ret;
2404 PERROR("pthread_create consumer");
2405 ret = -1;
2406 goto error;
2407 }
2408
2409 /* We are about to wait on a pthread condition */
2410 pthread_mutex_lock(&consumer_data->cond_mutex);
2411
2412 /* Get time for sem_timedwait absolute timeout */
2413 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2414 /*
2415 * Set the timeout for the condition timed wait even if the clock gettime
2416 * call fails since we might loop on that call and we want to avoid to
2417 * increment the timeout too many times.
2418 */
2419 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2420
2421 /*
2422 * The following loop COULD be skipped in some conditions so this is why we
2423 * set ret to 0 in order to make sure at least one round of the loop is
2424 * done.
2425 */
2426 ret = 0;
2427
2428 /*
2429 * Loop until the condition is reached or when a timeout is reached. Note
2430 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2431 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2432 * possible. This loop does not take any chances and works with both of
2433 * them.
2434 */
2435 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2436 if (clock_ret < 0) {
2437 PERROR("clock_gettime spawn consumer");
2438 /* Infinite wait for the consumerd thread to be ready */
2439 ret = pthread_cond_wait(&consumer_data->cond,
2440 &consumer_data->cond_mutex);
2441 } else {
2442 ret = pthread_cond_timedwait(&consumer_data->cond,
2443 &consumer_data->cond_mutex, &timeout);
2444 }
2445 }
2446
2447 /* Release the pthread condition */
2448 pthread_mutex_unlock(&consumer_data->cond_mutex);
2449
2450 if (ret != 0) {
2451 errno = ret;
2452 if (ret == ETIMEDOUT) {
2453 int pth_ret;
2454
2455 /*
2456 * Call has timed out so we kill the kconsumerd_thread and return
2457 * an error.
2458 */
2459 ERR("Condition timed out. The consumer thread was never ready."
2460 " Killing it");
2461 pth_ret = pthread_cancel(consumer_data->thread);
2462 if (pth_ret < 0) {
2463 PERROR("pthread_cancel consumer thread");
2464 }
2465 } else {
2466 PERROR("pthread_cond_wait failed consumer thread");
2467 }
2468 /* Caller is expecting a negative value on failure. */
2469 ret = -1;
2470 goto error;
2471 }
2472
2473 pthread_mutex_lock(&consumer_data->pid_mutex);
2474 if (consumer_data->pid == 0) {
2475 ERR("Consumerd did not start");
2476 pthread_mutex_unlock(&consumer_data->pid_mutex);
2477 goto error;
2478 }
2479 pthread_mutex_unlock(&consumer_data->pid_mutex);
2480
2481 return 0;
2482
2483 error:
2484 return ret;
2485 }
2486
2487 /*
2488 * Join consumer thread
2489 */
2490 static int join_consumer_thread(struct consumer_data *consumer_data)
2491 {
2492 void *status;
2493
2494 /* Consumer pid must be a real one. */
2495 if (consumer_data->pid > 0) {
2496 int ret;
2497 ret = kill(consumer_data->pid, SIGTERM);
2498 if (ret) {
2499 PERROR("Error killing consumer daemon");
2500 return ret;
2501 }
2502 return pthread_join(consumer_data->thread, &status);
2503 } else {
2504 return 0;
2505 }
2506 }
2507
2508 /*
2509 * Fork and exec a consumer daemon (consumerd).
2510 *
2511 * Return pid if successful else -1.
2512 */
2513 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2514 {
2515 int ret;
2516 pid_t pid;
2517 const char *consumer_to_use;
2518 const char *verbosity;
2519 struct stat st;
2520
2521 DBG("Spawning consumerd");
2522
2523 pid = fork();
2524 if (pid == 0) {
2525 /*
2526 * Exec consumerd.
2527 */
2528 if (opt_verbose_consumer) {
2529 verbosity = "--verbose";
2530 } else if (lttng_opt_quiet) {
2531 verbosity = "--quiet";
2532 } else {
2533 verbosity = "";
2534 }
2535
2536 switch (consumer_data->type) {
2537 case LTTNG_CONSUMER_KERNEL:
2538 /*
2539 * Find out which consumerd to execute. We will first try the
2540 * 64-bit path, then the sessiond's installation directory, and
2541 * fallback on the 32-bit one,
2542 */
2543 DBG3("Looking for a kernel consumer at these locations:");
2544 DBG3(" 1) %s", consumerd64_bin);
2545 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2546 DBG3(" 3) %s", consumerd32_bin);
2547 if (stat(consumerd64_bin, &st) == 0) {
2548 DBG3("Found location #1");
2549 consumer_to_use = consumerd64_bin;
2550 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2551 DBG3("Found location #2");
2552 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2553 } else if (stat(consumerd32_bin, &st) == 0) {
2554 DBG3("Found location #3");
2555 consumer_to_use = consumerd32_bin;
2556 } else {
2557 DBG("Could not find any valid consumerd executable");
2558 ret = -EINVAL;
2559 break;
2560 }
2561 DBG("Using kernel consumer at: %s", consumer_to_use);
2562 ret = execl(consumer_to_use,
2563 "lttng-consumerd", verbosity, "-k",
2564 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2565 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2566 "--group", tracing_group_name,
2567 NULL);
2568 break;
2569 case LTTNG_CONSUMER64_UST:
2570 {
2571 char *tmpnew = NULL;
2572
2573 if (consumerd64_libdir[0] != '\0') {
2574 char *tmp;
2575 size_t tmplen;
2576
2577 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2578 if (!tmp) {
2579 tmp = "";
2580 }
2581 tmplen = strlen("LD_LIBRARY_PATH=")
2582 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2583 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2584 if (!tmpnew) {
2585 ret = -ENOMEM;
2586 goto error;
2587 }
2588 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2589 strcat(tmpnew, consumerd64_libdir);
2590 if (tmp[0] != '\0') {
2591 strcat(tmpnew, ":");
2592 strcat(tmpnew, tmp);
2593 }
2594 ret = putenv(tmpnew);
2595 if (ret) {
2596 ret = -errno;
2597 free(tmpnew);
2598 goto error;
2599 }
2600 }
2601 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2602 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2603 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2604 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2605 "--group", tracing_group_name,
2606 NULL);
2607 if (consumerd64_libdir[0] != '\0') {
2608 free(tmpnew);
2609 }
2610 break;
2611 }
2612 case LTTNG_CONSUMER32_UST:
2613 {
2614 char *tmpnew = NULL;
2615
2616 if (consumerd32_libdir[0] != '\0') {
2617 char *tmp;
2618 size_t tmplen;
2619
2620 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2621 if (!tmp) {
2622 tmp = "";
2623 }
2624 tmplen = strlen("LD_LIBRARY_PATH=")
2625 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2626 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2627 if (!tmpnew) {
2628 ret = -ENOMEM;
2629 goto error;
2630 }
2631 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2632 strcat(tmpnew, consumerd32_libdir);
2633 if (tmp[0] != '\0') {
2634 strcat(tmpnew, ":");
2635 strcat(tmpnew, tmp);
2636 }
2637 ret = putenv(tmpnew);
2638 if (ret) {
2639 ret = -errno;
2640 free(tmpnew);
2641 goto error;
2642 }
2643 }
2644 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2645 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2646 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2647 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2648 "--group", tracing_group_name,
2649 NULL);
2650 if (consumerd32_libdir[0] != '\0') {
2651 free(tmpnew);
2652 }
2653 break;
2654 }
2655 default:
2656 PERROR("unknown consumer type");
2657 exit(EXIT_FAILURE);
2658 }
2659 if (errno != 0) {
2660 PERROR("Consumer execl()");
2661 }
2662 /* Reaching this point, we got a failure on our execl(). */
2663 exit(EXIT_FAILURE);
2664 } else if (pid > 0) {
2665 ret = pid;
2666 } else {
2667 PERROR("start consumer fork");
2668 ret = -errno;
2669 }
2670 error:
2671 return ret;
2672 }
2673
2674 /*
2675 * Spawn the consumerd daemon and session daemon thread.
2676 */
2677 static int start_consumerd(struct consumer_data *consumer_data)
2678 {
2679 int ret;
2680
2681 /*
2682 * Set the listen() state on the socket since there is a possible race
2683 * between the exec() of the consumer daemon and this call if place in the
2684 * consumer thread. See bug #366 for more details.
2685 */
2686 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2687 if (ret < 0) {
2688 goto error;
2689 }
2690
2691 pthread_mutex_lock(&consumer_data->pid_mutex);
2692 if (consumer_data->pid != 0) {
2693 pthread_mutex_unlock(&consumer_data->pid_mutex);
2694 goto end;
2695 }
2696
2697 ret = spawn_consumerd(consumer_data);
2698 if (ret < 0) {
2699 ERR("Spawning consumerd failed");
2700 pthread_mutex_unlock(&consumer_data->pid_mutex);
2701 goto error;
2702 }
2703
2704 /* Setting up the consumer_data pid */
2705 consumer_data->pid = ret;
2706 DBG2("Consumer pid %d", consumer_data->pid);
2707 pthread_mutex_unlock(&consumer_data->pid_mutex);
2708
2709 DBG2("Spawning consumer control thread");
2710 ret = spawn_consumer_thread(consumer_data);
2711 if (ret < 0) {
2712 ERR("Fatal error spawning consumer control thread");
2713 goto error;
2714 }
2715
2716 end:
2717 return 0;
2718
2719 error:
2720 /* Cleanup already created sockets on error. */
2721 if (consumer_data->err_sock >= 0) {
2722 int err;
2723
2724 err = close(consumer_data->err_sock);
2725 if (err < 0) {
2726 PERROR("close consumer data error socket");
2727 }
2728 }
2729 return ret;
2730 }
2731
2732 /*
2733 * Setup necessary data for kernel tracer action.
2734 */
2735 static int init_kernel_tracer(void)
2736 {
2737 int ret;
2738
2739 /* Modprobe lttng kernel modules */
2740 ret = modprobe_lttng_control();
2741 if (ret < 0) {
2742 goto error;
2743 }
2744
2745 /* Open debugfs lttng */
2746 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2747 if (kernel_tracer_fd < 0) {
2748 DBG("Failed to open %s", module_proc_lttng);
2749 ret = -1;
2750 goto error_open;
2751 }
2752
2753 /* Validate kernel version */
2754 ret = kernel_validate_version(kernel_tracer_fd);
2755 if (ret < 0) {
2756 goto error_version;
2757 }
2758
2759 ret = modprobe_lttng_data();
2760 if (ret < 0) {
2761 goto error_modules;
2762 }
2763
2764 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2765 return 0;
2766
2767 error_version:
2768 modprobe_remove_lttng_control();
2769 ret = close(kernel_tracer_fd);
2770 if (ret) {
2771 PERROR("close");
2772 }
2773 kernel_tracer_fd = -1;
2774 return LTTNG_ERR_KERN_VERSION;
2775
2776 error_modules:
2777 ret = close(kernel_tracer_fd);
2778 if (ret) {
2779 PERROR("close");
2780 }
2781
2782 error_open:
2783 modprobe_remove_lttng_control();
2784
2785 error:
2786 WARN("No kernel tracer available");
2787 kernel_tracer_fd = -1;
2788 if (!is_root) {
2789 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2790 } else {
2791 return LTTNG_ERR_KERN_NA;
2792 }
2793 }
2794
2795
2796 /*
2797 * Copy consumer output from the tracing session to the domain session. The
2798 * function also applies the right modification on a per domain basis for the
2799 * trace files destination directory.
2800 *
2801 * Should *NOT* be called with RCU read-side lock held.
2802 */
2803 static int copy_session_consumer(int domain, struct ltt_session *session)
2804 {
2805 int ret;
2806 const char *dir_name;
2807 struct consumer_output *consumer;
2808
2809 assert(session);
2810 assert(session->consumer);
2811
2812 switch (domain) {
2813 case LTTNG_DOMAIN_KERNEL:
2814 DBG3("Copying tracing session consumer output in kernel session");
2815 /*
2816 * XXX: We should audit the session creation and what this function
2817 * does "extra" in order to avoid a destroy since this function is used
2818 * in the domain session creation (kernel and ust) only. Same for UST
2819 * domain.
2820 */
2821 if (session->kernel_session->consumer) {
2822 consumer_output_put(session->kernel_session->consumer);
2823 }
2824 session->kernel_session->consumer =
2825 consumer_copy_output(session->consumer);
2826 /* Ease our life a bit for the next part */
2827 consumer = session->kernel_session->consumer;
2828 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2829 break;
2830 case LTTNG_DOMAIN_JUL:
2831 case LTTNG_DOMAIN_LOG4J:
2832 case LTTNG_DOMAIN_PYTHON:
2833 case LTTNG_DOMAIN_UST:
2834 DBG3("Copying tracing session consumer output in UST session");
2835 if (session->ust_session->consumer) {
2836 consumer_output_put(session->ust_session->consumer);
2837 }
2838 session->ust_session->consumer =
2839 consumer_copy_output(session->consumer);
2840 /* Ease our life a bit for the next part */
2841 consumer = session->ust_session->consumer;
2842 dir_name = DEFAULT_UST_TRACE_DIR;
2843 break;
2844 default:
2845 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2846 goto error;
2847 }
2848
2849 /* Append correct directory to subdir */
2850 strncat(consumer->subdir, dir_name,
2851 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2852 DBG3("Copy session consumer subdir %s", consumer->subdir);
2853
2854 ret = LTTNG_OK;
2855
2856 error:
2857 return ret;
2858 }
2859
2860 /*
2861 * Create an UST session and add it to the session ust list.
2862 *
2863 * Should *NOT* be called with RCU read-side lock held.
2864 */
2865 static int create_ust_session(struct ltt_session *session,
2866 struct lttng_domain *domain)
2867 {
2868 int ret;
2869 struct ltt_ust_session *lus = NULL;
2870
2871 assert(session);
2872 assert(domain);
2873 assert(session->consumer);
2874
2875 switch (domain->type) {
2876 case LTTNG_DOMAIN_JUL:
2877 case LTTNG_DOMAIN_LOG4J:
2878 case LTTNG_DOMAIN_PYTHON:
2879 case LTTNG_DOMAIN_UST:
2880 break;
2881 default:
2882 ERR("Unknown UST domain on create session %d", domain->type);
2883 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2884 goto error;
2885 }
2886
2887 DBG("Creating UST session");
2888
2889 lus = trace_ust_create_session(session->id);
2890 if (lus == NULL) {
2891 ret = LTTNG_ERR_UST_SESS_FAIL;
2892 goto error;
2893 }
2894
2895 lus->uid = session->uid;
2896 lus->gid = session->gid;
2897 lus->output_traces = session->output_traces;
2898 lus->snapshot_mode = session->snapshot_mode;
2899 lus->live_timer_interval = session->live_timer;
2900 session->ust_session = lus;
2901 if (session->shm_path[0]) {
2902 strncpy(lus->root_shm_path, session->shm_path,
2903 sizeof(lus->root_shm_path));
2904 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2905 strncpy(lus->shm_path, session->shm_path,
2906 sizeof(lus->shm_path));
2907 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2908 strncat(lus->shm_path, "/ust",
2909 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2910 }
2911 /* Copy session output to the newly created UST session */
2912 ret = copy_session_consumer(domain->type, session);
2913 if (ret != LTTNG_OK) {
2914 goto error;
2915 }
2916
2917 return LTTNG_OK;
2918
2919 error:
2920 free(lus);
2921 session->ust_session = NULL;
2922 return ret;
2923 }
2924
2925 /*
2926 * Create a kernel tracer session then create the default channel.
2927 */
2928 static int create_kernel_session(struct ltt_session *session)
2929 {
2930 int ret;
2931
2932 DBG("Creating kernel session");
2933
2934 ret = kernel_create_session(session, kernel_tracer_fd);
2935 if (ret < 0) {
2936 ret = LTTNG_ERR_KERN_SESS_FAIL;
2937 goto error;
2938 }
2939
2940 /* Code flow safety */
2941 assert(session->kernel_session);
2942
2943 /* Copy session output to the newly created Kernel session */
2944 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2945 if (ret != LTTNG_OK) {
2946 goto error;
2947 }
2948
2949 /* Create directory(ies) on local filesystem. */
2950 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2951 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2952 ret = run_as_mkdir_recursive(
2953 session->kernel_session->consumer->dst.trace_path,
2954 S_IRWXU | S_IRWXG, session->uid, session->gid);
2955 if (ret < 0) {
2956 if (errno != EEXIST) {
2957 ERR("Trace directory creation error");
2958 goto error;
2959 }
2960 }
2961 }
2962
2963 session->kernel_session->uid = session->uid;
2964 session->kernel_session->gid = session->gid;
2965 session->kernel_session->output_traces = session->output_traces;
2966 session->kernel_session->snapshot_mode = session->snapshot_mode;
2967
2968 return LTTNG_OK;
2969
2970 error:
2971 trace_kernel_destroy_session(session->kernel_session);
2972 session->kernel_session = NULL;
2973 return ret;
2974 }
2975
2976 /*
2977 * Count number of session permitted by uid/gid.
2978 */
2979 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2980 {
2981 unsigned int i = 0;
2982 struct ltt_session *session;
2983
2984 DBG("Counting number of available session for UID %d GID %d",
2985 uid, gid);
2986 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2987 /*
2988 * Only list the sessions the user can control.
2989 */
2990 if (!session_access_ok(session, uid, gid)) {
2991 continue;
2992 }
2993 i++;
2994 }
2995 return i;
2996 }
2997
2998 /*
2999 * Process the command requested by the lttng client within the command
3000 * context structure. This function make sure that the return structure (llm)
3001 * is set and ready for transmission before returning.
3002 *
3003 * Return any error encountered or 0 for success.
3004 *
3005 * "sock" is only used for special-case var. len data.
3006 *
3007 * Should *NOT* be called with RCU read-side lock held.
3008 */
3009 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
3010 int *sock_error)
3011 {
3012 int ret = LTTNG_OK;
3013 int need_tracing_session = 1;
3014 int need_domain;
3015
3016 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
3017
3018 assert(!rcu_read_ongoing());
3019
3020 *sock_error = 0;
3021
3022 switch (cmd_ctx->lsm->cmd_type) {
3023 case LTTNG_CREATE_SESSION:
3024 case LTTNG_CREATE_SESSION_SNAPSHOT:
3025 case LTTNG_CREATE_SESSION_LIVE:
3026 case LTTNG_DESTROY_SESSION:
3027 case LTTNG_LIST_SESSIONS:
3028 case LTTNG_LIST_DOMAINS:
3029 case LTTNG_START_TRACE:
3030 case LTTNG_STOP_TRACE:
3031 case LTTNG_DATA_PENDING:
3032 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3033 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3034 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3035 case LTTNG_SNAPSHOT_RECORD:
3036 case LTTNG_SAVE_SESSION:
3037 case LTTNG_SET_SESSION_SHM_PATH:
3038 need_domain = 0;
3039 break;
3040 default:
3041 need_domain = 1;
3042 }
3043
3044 if (opt_no_kernel && need_domain
3045 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3046 if (!is_root) {
3047 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3048 } else {
3049 ret = LTTNG_ERR_KERN_NA;
3050 }
3051 goto error;
3052 }
3053
3054 /* Deny register consumer if we already have a spawned consumer. */
3055 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3056 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3057 if (kconsumer_data.pid > 0) {
3058 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3059 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3060 goto error;
3061 }
3062 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3063 }
3064
3065 /*
3066 * Check for command that don't needs to allocate a returned payload. We do
3067 * this here so we don't have to make the call for no payload at each
3068 * command.
3069 */
3070 switch(cmd_ctx->lsm->cmd_type) {
3071 case LTTNG_LIST_SESSIONS:
3072 case LTTNG_LIST_TRACEPOINTS:
3073 case LTTNG_LIST_TRACEPOINT_FIELDS:
3074 case LTTNG_LIST_DOMAINS:
3075 case LTTNG_LIST_CHANNELS:
3076 case LTTNG_LIST_EVENTS:
3077 case LTTNG_LIST_SYSCALLS:
3078 case LTTNG_LIST_TRACKER_PIDS:
3079 break;
3080 default:
3081 /* Setup lttng message with no payload */
3082 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3083 if (ret < 0) {
3084 /* This label does not try to unlock the session */
3085 goto init_setup_error;
3086 }
3087 }
3088
3089 /* Commands that DO NOT need a session. */
3090 switch (cmd_ctx->lsm->cmd_type) {
3091 case LTTNG_CREATE_SESSION:
3092 case LTTNG_CREATE_SESSION_SNAPSHOT:
3093 case LTTNG_CREATE_SESSION_LIVE:
3094 case LTTNG_CALIBRATE:
3095 case LTTNG_LIST_SESSIONS:
3096 case LTTNG_LIST_TRACEPOINTS:
3097 case LTTNG_LIST_SYSCALLS:
3098 case LTTNG_LIST_TRACEPOINT_FIELDS:
3099 case LTTNG_SAVE_SESSION:
3100 need_tracing_session = 0;
3101 break;
3102 default:
3103 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3104 /*
3105 * We keep the session list lock across _all_ commands
3106 * for now, because the per-session lock does not
3107 * handle teardown properly.
3108 */
3109 session_lock_list();
3110 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3111 if (cmd_ctx->session == NULL) {
3112 ret = LTTNG_ERR_SESS_NOT_FOUND;
3113 goto error;
3114 } else {
3115 /* Acquire lock for the session */
3116 session_lock(cmd_ctx->session);
3117 }
3118 break;
3119 }
3120
3121 /*
3122 * Commands that need a valid session but should NOT create one if none
3123 * exists. Instead of creating one and destroying it when the command is
3124 * handled, process that right before so we save some round trip in useless
3125 * code path.
3126 */
3127 switch (cmd_ctx->lsm->cmd_type) {
3128 case LTTNG_DISABLE_CHANNEL:
3129 case LTTNG_DISABLE_EVENT:
3130 switch (cmd_ctx->lsm->domain.type) {
3131 case LTTNG_DOMAIN_KERNEL:
3132 if (!cmd_ctx->session->kernel_session) {
3133 ret = LTTNG_ERR_NO_CHANNEL;
3134 goto error;
3135 }
3136 break;
3137 case LTTNG_DOMAIN_JUL:
3138 case LTTNG_DOMAIN_LOG4J:
3139 case LTTNG_DOMAIN_PYTHON:
3140 case LTTNG_DOMAIN_UST:
3141 if (!cmd_ctx->session->ust_session) {
3142 ret = LTTNG_ERR_NO_CHANNEL;
3143 goto error;
3144 }
3145 break;
3146 default:
3147 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3148 goto error;
3149 }
3150 default:
3151 break;
3152 }
3153
3154 if (!need_domain) {
3155 goto skip_domain;
3156 }
3157
3158 /*
3159 * Check domain type for specific "pre-action".
3160 */
3161 switch (cmd_ctx->lsm->domain.type) {
3162 case LTTNG_DOMAIN_KERNEL:
3163 if (!is_root) {
3164 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3165 goto error;
3166 }
3167
3168 /* Kernel tracer check */
3169 if (kernel_tracer_fd == -1) {
3170 /* Basically, load kernel tracer modules */
3171 ret = init_kernel_tracer();
3172 if (ret != 0) {
3173 goto error;
3174 }
3175 }
3176
3177 /* Consumer is in an ERROR state. Report back to client */
3178 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3179 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3180 goto error;
3181 }
3182
3183 /* Need a session for kernel command */
3184 if (need_tracing_session) {
3185 if (cmd_ctx->session->kernel_session == NULL) {
3186 ret = create_kernel_session(cmd_ctx->session);
3187 if (ret < 0) {
3188 ret = LTTNG_ERR_KERN_SESS_FAIL;
3189 goto error;
3190 }
3191 }
3192
3193 /* Start the kernel consumer daemon */
3194 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3195 if (kconsumer_data.pid == 0 &&
3196 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3197 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3198 ret = start_consumerd(&kconsumer_data);
3199 if (ret < 0) {
3200 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3201 goto error;
3202 }
3203 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3204 } else {
3205 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3206 }
3207
3208 /*
3209 * The consumer was just spawned so we need to add the socket to
3210 * the consumer output of the session if exist.
3211 */
3212 ret = consumer_create_socket(&kconsumer_data,
3213 cmd_ctx->session->kernel_session->consumer);
3214 if (ret < 0) {
3215 goto error;
3216 }
3217 }
3218
3219 break;
3220 case LTTNG_DOMAIN_JUL:
3221 case LTTNG_DOMAIN_LOG4J:
3222 case LTTNG_DOMAIN_PYTHON:
3223 case LTTNG_DOMAIN_UST:
3224 {
3225 if (!ust_app_supported()) {
3226 ret = LTTNG_ERR_NO_UST;
3227 goto error;
3228 }
3229 /* Consumer is in an ERROR state. Report back to client */
3230 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3231 ret = LTTNG_ERR_NO_USTCONSUMERD;
3232 goto error;
3233 }
3234
3235 if (need_tracing_session) {
3236 /* Create UST session if none exist. */
3237 if (cmd_ctx->session->ust_session == NULL) {
3238 ret = create_ust_session(cmd_ctx->session,
3239 &cmd_ctx->lsm->domain);
3240 if (ret != LTTNG_OK) {
3241 goto error;
3242 }
3243 }
3244
3245 /* Start the UST consumer daemons */
3246 /* 64-bit */
3247 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3248 if (consumerd64_bin[0] != '\0' &&
3249 ustconsumer64_data.pid == 0 &&
3250 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3251 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3252 ret = start_consumerd(&ustconsumer64_data);
3253 if (ret < 0) {
3254 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3255 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3256 goto error;
3257 }
3258
3259 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3260 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3261 } else {
3262 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3263 }
3264
3265 /*
3266 * Setup socket for consumer 64 bit. No need for atomic access
3267 * since it was set above and can ONLY be set in this thread.
3268 */
3269 ret = consumer_create_socket(&ustconsumer64_data,
3270 cmd_ctx->session->ust_session->consumer);
3271 if (ret < 0) {
3272 goto error;
3273 }
3274
3275 /* 32-bit */
3276 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3277 if (consumerd32_bin[0] != '\0' &&
3278 ustconsumer32_data.pid == 0 &&
3279 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3280 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3281 ret = start_consumerd(&ustconsumer32_data);
3282 if (ret < 0) {
3283 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3284 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3285 goto error;
3286 }
3287
3288 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3289 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3290 } else {
3291 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3292 }
3293
3294 /*
3295 * Setup socket for consumer 64 bit. No need for atomic access
3296 * since it was set above and can ONLY be set in this thread.
3297 */
3298 ret = consumer_create_socket(&ustconsumer32_data,
3299 cmd_ctx->session->ust_session->consumer);
3300 if (ret < 0) {
3301 goto error;
3302 }
3303 }
3304 break;
3305 }
3306 default:
3307 break;
3308 }
3309 skip_domain:
3310
3311 /* Validate consumer daemon state when start/stop trace command */
3312 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3313 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3314 switch (cmd_ctx->lsm->domain.type) {
3315 case LTTNG_DOMAIN_JUL:
3316 case LTTNG_DOMAIN_LOG4J:
3317 case LTTNG_DOMAIN_PYTHON:
3318 case LTTNG_DOMAIN_UST:
3319 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3320 ret = LTTNG_ERR_NO_USTCONSUMERD;
3321 goto error;
3322 }
3323 break;
3324 case LTTNG_DOMAIN_KERNEL:
3325 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3326 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3327 goto error;
3328 }
3329 break;
3330 }
3331 }
3332
3333 /*
3334 * Check that the UID or GID match that of the tracing session.
3335 * The root user can interact with all sessions.
3336 */
3337 if (need_tracing_session) {
3338 if (!session_access_ok(cmd_ctx->session,
3339 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3340 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3341 ret = LTTNG_ERR_EPERM;
3342 goto error;
3343 }
3344 }
3345
3346 /*
3347 * Send relayd information to consumer as soon as we have a domain and a
3348 * session defined.
3349 */
3350 if (cmd_ctx->session && need_domain) {
3351 /*
3352 * Setup relayd if not done yet. If the relayd information was already
3353 * sent to the consumer, this call will gracefully return.
3354 */
3355 ret = cmd_setup_relayd(cmd_ctx->session);
3356 if (ret != LTTNG_OK) {
3357 goto error;
3358 }
3359 }
3360
3361 /* Process by command type */
3362 switch (cmd_ctx->lsm->cmd_type) {
3363 case LTTNG_ADD_CONTEXT:
3364 {
3365 /*
3366 * An LTTNG_ADD_CONTEXT command might have a supplementary
3367 * payload if the context being added is an application context.
3368 */
3369 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3370 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3371 char *provider_name = NULL, *context_name = NULL;
3372 size_t provider_name_len =
3373 cmd_ctx->lsm->u.context.provider_name_len;
3374 size_t context_name_len =
3375 cmd_ctx->lsm->u.context.context_name_len;
3376
3377 if (provider_name_len == 0 || context_name_len == 0) {
3378 /*
3379 * Application provider and context names MUST
3380 * be provided.
3381 */
3382 ret = -LTTNG_ERR_INVALID;
3383 goto error;
3384 }
3385
3386 provider_name = zmalloc(provider_name_len + 1);
3387 if (!provider_name) {
3388 ret = -LTTNG_ERR_NOMEM;
3389 goto error;
3390 }
3391 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3392 provider_name;
3393
3394 context_name = zmalloc(context_name_len + 1);
3395 if (!context_name) {
3396 ret = -LTTNG_ERR_NOMEM;
3397 goto error_add_context;
3398 }
3399 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3400 context_name;
3401
3402 ret = lttcomm_recv_unix_sock(sock, provider_name,
3403 provider_name_len);
3404 if (ret < 0) {
3405 goto error_add_context;
3406 }
3407
3408 ret = lttcomm_recv_unix_sock(sock, context_name,
3409 context_name_len);
3410 if (ret < 0) {
3411 goto error_add_context;
3412 }
3413 }
3414
3415 /*
3416 * cmd_add_context assumes ownership of the provider and context
3417 * names.
3418 */
3419 ret = cmd_add_context(cmd_ctx->session,
3420 cmd_ctx->lsm->domain.type,
3421 cmd_ctx->lsm->u.context.channel_name,
3422 &cmd_ctx->lsm->u.context.ctx,
3423 kernel_poll_pipe[1]);
3424
3425 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3426 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3427 error_add_context:
3428 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3429 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3430 if (ret < 0) {
3431 goto error;
3432 }
3433 break;
3434 }
3435 case LTTNG_DISABLE_CHANNEL:
3436 {
3437 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3438 cmd_ctx->lsm->u.disable.channel_name);
3439 break;
3440 }
3441 case LTTNG_DISABLE_EVENT:
3442 {
3443
3444 /*
3445 * FIXME: handle filter; for now we just receive the filter's
3446 * bytecode along with the filter expression which are sent by
3447 * liblttng-ctl and discard them.
3448 *
3449 * This fixes an issue where the client may block while sending
3450 * the filter payload and encounter an error because the session
3451 * daemon closes the socket without ever handling this data.
3452 */
3453 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3454 cmd_ctx->lsm->u.disable.bytecode_len;
3455
3456 if (count) {
3457 char data[LTTNG_FILTER_MAX_LEN];
3458
3459 DBG("Discarding disable event command payload of size %zu", count);
3460 while (count) {
3461 ret = lttcomm_recv_unix_sock(sock, data,
3462 count > sizeof(data) ? sizeof(data) : count);
3463 if (ret < 0) {
3464 goto error;
3465 }
3466
3467 count -= (size_t) ret;
3468 }
3469 }
3470 /* FIXME: passing packed structure to non-packed pointer */
3471 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3472 cmd_ctx->lsm->u.disable.channel_name,
3473 &cmd_ctx->lsm->u.disable.event);
3474 break;
3475 }
3476 case LTTNG_ENABLE_CHANNEL:
3477 {
3478 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3479 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3480 break;
3481 }
3482 case LTTNG_TRACK_PID:
3483 {
3484 ret = cmd_track_pid(cmd_ctx->session,
3485 cmd_ctx->lsm->domain.type,
3486 cmd_ctx->lsm->u.pid_tracker.pid);
3487 break;
3488 }
3489 case LTTNG_UNTRACK_PID:
3490 {
3491 ret = cmd_untrack_pid(cmd_ctx->session,
3492 cmd_ctx->lsm->domain.type,
3493 cmd_ctx->lsm->u.pid_tracker.pid);
3494 break;
3495 }
3496 case LTTNG_ENABLE_EVENT:
3497 {
3498 struct lttng_event_exclusion *exclusion = NULL;
3499 struct lttng_filter_bytecode *bytecode = NULL;
3500 char *filter_expression = NULL;
3501
3502 /* Handle exclusion events and receive it from the client. */
3503 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3504 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3505
3506 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3507 (count * LTTNG_SYMBOL_NAME_LEN));
3508 if (!exclusion) {
3509 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3510 goto error;
3511 }
3512
3513 DBG("Receiving var len exclusion event list from client ...");
3514 exclusion->count = count;
3515 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3516 count * LTTNG_SYMBOL_NAME_LEN);
3517 if (ret <= 0) {
3518 DBG("Nothing recv() from client var len data... continuing");
3519 *sock_error = 1;
3520 free(exclusion);
3521 ret = LTTNG_ERR_EXCLUSION_INVAL;
3522 goto error;
3523 }
3524 }
3525
3526 /* Get filter expression from client. */
3527 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3528 size_t expression_len =
3529 cmd_ctx->lsm->u.enable.expression_len;
3530
3531 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3532 ret = LTTNG_ERR_FILTER_INVAL;
3533 free(exclusion);
3534 goto error;
3535 }
3536
3537 filter_expression = zmalloc(expression_len);
3538 if (!filter_expression) {
3539 free(exclusion);
3540 ret = LTTNG_ERR_FILTER_NOMEM;
3541 goto error;
3542 }
3543
3544 /* Receive var. len. data */
3545 DBG("Receiving var len filter's expression from client ...");
3546 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3547 expression_len);
3548 if (ret <= 0) {
3549 DBG("Nothing recv() from client car len data... continuing");
3550 *sock_error = 1;
3551 free(filter_expression);
3552 free(exclusion);
3553 ret = LTTNG_ERR_FILTER_INVAL;
3554 goto error;
3555 }
3556 }
3557
3558 /* Handle filter and get bytecode from client. */
3559 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3560 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3561
3562 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3563 ret = LTTNG_ERR_FILTER_INVAL;
3564 free(filter_expression);
3565 free(exclusion);
3566 goto error;
3567 }
3568
3569 bytecode = zmalloc(bytecode_len);
3570 if (!bytecode) {
3571 free(filter_expression);
3572 free(exclusion);
3573 ret = LTTNG_ERR_FILTER_NOMEM;
3574 goto error;
3575 }
3576
3577 /* Receive var. len. data */
3578 DBG("Receiving var len filter's bytecode from client ...");
3579 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3580 if (ret <= 0) {
3581 DBG("Nothing recv() from client car len data... continuing");
3582 *sock_error = 1;
3583 free(filter_expression);
3584 free(bytecode);
3585 free(exclusion);
3586 ret = LTTNG_ERR_FILTER_INVAL;
3587 goto error;
3588 }
3589
3590 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3591 free(filter_expression);
3592 free(bytecode);
3593 free(exclusion);
3594 ret = LTTNG_ERR_FILTER_INVAL;
3595 goto error;
3596 }
3597 }
3598
3599 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3600 cmd_ctx->lsm->u.enable.channel_name,
3601 &cmd_ctx->lsm->u.enable.event,
3602 filter_expression, bytecode, exclusion,
3603 kernel_poll_pipe[1]);
3604 break;
3605 }
3606 case LTTNG_LIST_TRACEPOINTS:
3607 {
3608 struct lttng_event *events;
3609 ssize_t nb_events;
3610
3611 session_lock_list();
3612 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3613 session_unlock_list();
3614 if (nb_events < 0) {
3615 /* Return value is a negative lttng_error_code. */
3616 ret = -nb_events;
3617 goto error;
3618 }
3619
3620 /*
3621 * Setup lttng message with payload size set to the event list size in
3622 * bytes and then copy list into the llm payload.
3623 */
3624 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3625 sizeof(struct lttng_event) * nb_events);
3626 free(events);
3627
3628 if (ret < 0) {
3629 goto setup_error;
3630 }
3631
3632 ret = LTTNG_OK;
3633 break;
3634 }
3635 case LTTNG_LIST_TRACEPOINT_FIELDS:
3636 {
3637 struct lttng_event_field *fields;
3638 ssize_t nb_fields;
3639
3640 session_lock_list();
3641 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3642 &fields);
3643 session_unlock_list();
3644 if (nb_fields < 0) {
3645 /* Return value is a negative lttng_error_code. */
3646 ret = -nb_fields;
3647 goto error;
3648 }
3649
3650 /*
3651 * Setup lttng message with payload size set to the event list size in
3652 * bytes and then copy list into the llm payload.
3653 */
3654 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3655 sizeof(struct lttng_event_field) * nb_fields);
3656 free(fields);
3657
3658 if (ret < 0) {
3659 goto setup_error;
3660 }
3661
3662 ret = LTTNG_OK;
3663 break;
3664 }
3665 case LTTNG_LIST_SYSCALLS:
3666 {
3667 struct lttng_event *events;
3668 ssize_t nb_events;
3669
3670 nb_events = cmd_list_syscalls(&events);
3671 if (nb_events < 0) {
3672 /* Return value is a negative lttng_error_code. */
3673 ret = -nb_events;
3674 goto error;
3675 }
3676
3677 /*
3678 * Setup lttng message with payload size set to the event list size in
3679 * bytes and then copy list into the llm payload.
3680 */
3681 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3682 sizeof(struct lttng_event) * nb_events);
3683 free(events);
3684
3685 if (ret < 0) {
3686 goto setup_error;
3687 }
3688
3689 ret = LTTNG_OK;
3690 break;
3691 }
3692 case LTTNG_LIST_TRACKER_PIDS:
3693 {
3694 int32_t *pids = NULL;
3695 ssize_t nr_pids;
3696
3697 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3698 cmd_ctx->lsm->domain.type, &pids);
3699 if (nr_pids < 0) {
3700 /* Return value is a negative lttng_error_code. */
3701 ret = -nr_pids;
3702 goto error;
3703 }
3704
3705 /*
3706 * Setup lttng message with payload size set to the event list size in
3707 * bytes and then copy list into the llm payload.
3708 */
3709 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3710 sizeof(int32_t) * nr_pids);
3711 free(pids);
3712
3713 if (ret < 0) {
3714 goto setup_error;
3715 }
3716
3717 ret = LTTNG_OK;
3718 break;
3719 }
3720 case LTTNG_SET_CONSUMER_URI:
3721 {
3722 size_t nb_uri, len;
3723 struct lttng_uri *uris;
3724
3725 nb_uri = cmd_ctx->lsm->u.uri.size;
3726 len = nb_uri * sizeof(struct lttng_uri);
3727
3728 if (nb_uri == 0) {
3729 ret = LTTNG_ERR_INVALID;
3730 goto error;
3731 }
3732
3733 uris = zmalloc(len);
3734 if (uris == NULL) {
3735 ret = LTTNG_ERR_FATAL;
3736 goto error;
3737 }
3738
3739 /* Receive variable len data */
3740 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3741 ret = lttcomm_recv_unix_sock(sock, uris, len);
3742 if (ret <= 0) {
3743 DBG("No URIs received from client... continuing");
3744 *sock_error = 1;
3745 ret = LTTNG_ERR_SESSION_FAIL;
3746 free(uris);
3747 goto error;
3748 }
3749
3750 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3751 free(uris);
3752 if (ret != LTTNG_OK) {
3753 goto error;
3754 }
3755
3756
3757 break;
3758 }
3759 case LTTNG_START_TRACE:
3760 {
3761 ret = cmd_start_trace(cmd_ctx->session);
3762 break;
3763 }
3764 case LTTNG_STOP_TRACE:
3765 {
3766 ret = cmd_stop_trace(cmd_ctx->session);
3767 break;
3768 }
3769 case LTTNG_CREATE_SESSION:
3770 {
3771 size_t nb_uri, len;
3772 struct lttng_uri *uris = NULL;
3773
3774 nb_uri = cmd_ctx->lsm->u.uri.size;
3775 len = nb_uri * sizeof(struct lttng_uri);
3776
3777 if (nb_uri > 0) {
3778 uris = zmalloc(len);
3779 if (uris == NULL) {
3780 ret = LTTNG_ERR_FATAL;
3781 goto error;
3782 }
3783
3784 /* Receive variable len data */
3785 DBG("Waiting for %zu URIs from client ...", nb_uri);
3786 ret = lttcomm_recv_unix_sock(sock, uris, len);
3787 if (ret <= 0) {
3788 DBG("No URIs received from client... continuing");
3789 *sock_error = 1;
3790 ret = LTTNG_ERR_SESSION_FAIL;
3791 free(uris);
3792 goto error;
3793 }
3794
3795 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3796 DBG("Creating session with ONE network URI is a bad call");
3797 ret = LTTNG_ERR_SESSION_FAIL;
3798 free(uris);
3799 goto error;
3800 }
3801 }
3802
3803 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3804 &cmd_ctx->creds, 0);
3805
3806 free(uris);
3807
3808 break;
3809 }
3810 case LTTNG_DESTROY_SESSION:
3811 {
3812 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3813
3814 /* Set session to NULL so we do not unlock it after free. */
3815 cmd_ctx->session = NULL;
3816 break;
3817 }
3818 case LTTNG_LIST_DOMAINS:
3819 {
3820 ssize_t nb_dom;
3821 struct lttng_domain *domains = NULL;
3822
3823 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3824 if (nb_dom < 0) {
3825 /* Return value is a negative lttng_error_code. */
3826 ret = -nb_dom;
3827 goto error;
3828 }
3829
3830 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3831 nb_dom * sizeof(struct lttng_domain));
3832 free(domains);
3833
3834 if (ret < 0) {
3835 goto setup_error;
3836 }
3837
3838 ret = LTTNG_OK;
3839 break;
3840 }
3841 case LTTNG_LIST_CHANNELS:
3842 {
3843 int nb_chan;
3844 struct lttng_channel *channels = NULL;
3845
3846 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3847 cmd_ctx->session, &channels);
3848 if (nb_chan < 0) {
3849 /* Return value is a negative lttng_error_code. */
3850 ret = -nb_chan;
3851 goto error;
3852 }
3853
3854 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3855 nb_chan * sizeof(struct lttng_channel));
3856 free(channels);
3857
3858 if (ret < 0) {
3859 goto setup_error;
3860 }
3861
3862 ret = LTTNG_OK;
3863 break;
3864 }
3865 case LTTNG_LIST_EVENTS:
3866 {
3867 ssize_t nb_event;
3868 struct lttng_event *events = NULL;
3869
3870 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3871 cmd_ctx->lsm->u.list.channel_name, &events);
3872 if (nb_event < 0) {
3873 /* Return value is a negative lttng_error_code. */
3874 ret = -nb_event;
3875 goto error;
3876 }
3877
3878 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3879 nb_event * sizeof(struct lttng_event));
3880 free(events);
3881
3882 if (ret < 0) {
3883 goto setup_error;
3884 }
3885
3886 ret = LTTNG_OK;
3887 break;
3888 }
3889 case LTTNG_LIST_SESSIONS:
3890 {
3891 unsigned int nr_sessions;
3892 void *sessions_payload;
3893 size_t payload_len;
3894
3895 session_lock_list();
3896 nr_sessions = lttng_sessions_count(
3897 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3898 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3899 payload_len = sizeof(struct lttng_session) * nr_sessions;
3900 sessions_payload = zmalloc(payload_len);
3901
3902 if (!sessions_payload) {
3903 session_unlock_list();
3904 ret = -ENOMEM;
3905 goto setup_error;
3906 }
3907
3908 cmd_list_lttng_sessions(sessions_payload,
3909 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3910 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3911 session_unlock_list();
3912
3913 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3914 payload_len);
3915 free(sessions_payload);
3916
3917 if (ret < 0) {
3918 goto setup_error;
3919 }
3920
3921 ret = LTTNG_OK;
3922 break;
3923 }
3924 case LTTNG_CALIBRATE:
3925 {
3926 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3927 &cmd_ctx->lsm->u.calibrate);
3928 break;
3929 }
3930 case LTTNG_REGISTER_CONSUMER:
3931 {
3932 struct consumer_data *cdata;
3933
3934 switch (cmd_ctx->lsm->domain.type) {
3935 case LTTNG_DOMAIN_KERNEL:
3936 cdata = &kconsumer_data;
3937 break;
3938 default:
3939 ret = LTTNG_ERR_UND;
3940 goto error;
3941 }
3942
3943 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3944 cmd_ctx->lsm->u.reg.path, cdata);
3945 break;
3946 }
3947 case LTTNG_DATA_PENDING:
3948 {
3949 int pending_ret;
3950 uint8_t pending_ret_byte;
3951
3952 pending_ret = cmd_data_pending(cmd_ctx->session);
3953
3954 /*
3955 * FIXME
3956 *
3957 * This function may returns 0 or 1 to indicate whether or not
3958 * there is data pending. In case of error, it should return an
3959 * LTTNG_ERR code. However, some code paths may still return
3960 * a nondescript error code, which we handle by returning an
3961 * "unknown" error.
3962 */
3963 if (pending_ret == 0 || pending_ret == 1) {
3964 /*
3965 * ret will be set to LTTNG_OK at the end of
3966 * this function.
3967 */
3968 } else if (pending_ret < 0) {
3969 ret = LTTNG_ERR_UNK;
3970 goto setup_error;
3971 } else {
3972 ret = pending_ret;
3973 goto setup_error;
3974 }
3975
3976 pending_ret_byte = (uint8_t) pending_ret;
3977
3978 /* 1 byte to return whether or not data is pending */
3979 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3980 &pending_ret_byte, 1);
3981
3982 if (ret < 0) {
3983 goto setup_error;
3984 }
3985
3986 ret = LTTNG_OK;
3987 break;
3988 }
3989 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3990 {
3991 struct lttcomm_lttng_output_id reply;
3992
3993 ret = cmd_snapshot_add_output(cmd_ctx->session,
3994 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3995 if (ret != LTTNG_OK) {
3996 goto error;
3997 }
3998
3999 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
4000 sizeof(reply));
4001 if (ret < 0) {
4002 goto setup_error;
4003 }
4004
4005 /* Copy output list into message payload */
4006 ret = LTTNG_OK;
4007 break;
4008 }
4009 case LTTNG_SNAPSHOT_DEL_OUTPUT:
4010 {
4011 ret = cmd_snapshot_del_output(cmd_ctx->session,
4012 &cmd_ctx->lsm->u.snapshot_output.output);
4013 break;
4014 }
4015 case LTTNG_SNAPSHOT_LIST_OUTPUT:
4016 {
4017 ssize_t nb_output;
4018 struct lttng_snapshot_output *outputs = NULL;
4019
4020 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
4021 if (nb_output < 0) {
4022 ret = -nb_output;
4023 goto error;
4024 }
4025
4026 assert((nb_output > 0 && outputs) || nb_output == 0);
4027 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
4028 nb_output * sizeof(struct lttng_snapshot_output));
4029 free(outputs);
4030
4031 if (ret < 0) {
4032 goto setup_error;
4033 }
4034
4035 ret = LTTNG_OK;
4036 break;
4037 }
4038 case LTTNG_SNAPSHOT_RECORD:
4039 {
4040 ret = cmd_snapshot_record(cmd_ctx->session,
4041 &cmd_ctx->lsm->u.snapshot_record.output,
4042 cmd_ctx->lsm->u.snapshot_record.wait);
4043 break;
4044 }
4045 case LTTNG_CREATE_SESSION_SNAPSHOT:
4046 {
4047 size_t nb_uri, len;
4048 struct lttng_uri *uris = NULL;
4049
4050 nb_uri = cmd_ctx->lsm->u.uri.size;
4051 len = nb_uri * sizeof(struct lttng_uri);
4052
4053 if (nb_uri > 0) {
4054 uris = zmalloc(len);
4055 if (uris == NULL) {
4056 ret = LTTNG_ERR_FATAL;
4057 goto error;
4058 }
4059
4060 /* Receive variable len data */
4061 DBG("Waiting for %zu URIs from client ...", nb_uri);
4062 ret = lttcomm_recv_unix_sock(sock, uris, len);
4063 if (ret <= 0) {
4064 DBG("No URIs received from client... continuing");
4065 *sock_error = 1;
4066 ret = LTTNG_ERR_SESSION_FAIL;
4067 free(uris);
4068 goto error;
4069 }
4070
4071 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4072 DBG("Creating session with ONE network URI is a bad call");
4073 ret = LTTNG_ERR_SESSION_FAIL;
4074 free(uris);
4075 goto error;
4076 }
4077 }
4078
4079 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4080 nb_uri, &cmd_ctx->creds);
4081 free(uris);
4082 break;
4083 }
4084 case LTTNG_CREATE_SESSION_LIVE:
4085 {
4086 size_t nb_uri, len;
4087 struct lttng_uri *uris = NULL;
4088
4089 nb_uri = cmd_ctx->lsm->u.uri.size;
4090 len = nb_uri * sizeof(struct lttng_uri);
4091
4092 if (nb_uri > 0) {
4093 uris = zmalloc(len);
4094 if (uris == NULL) {
4095 ret = LTTNG_ERR_FATAL;
4096 goto error;
4097 }
4098
4099 /* Receive variable len data */
4100 DBG("Waiting for %zu URIs from client ...", nb_uri);
4101 ret = lttcomm_recv_unix_sock(sock, uris, len);
4102 if (ret <= 0) {
4103 DBG("No URIs received from client... continuing");
4104 *sock_error = 1;
4105 ret = LTTNG_ERR_SESSION_FAIL;
4106 free(uris);
4107 goto error;
4108 }
4109
4110 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4111 DBG("Creating session with ONE network URI is a bad call");
4112 ret = LTTNG_ERR_SESSION_FAIL;
4113 free(uris);
4114 goto error;
4115 }
4116 }
4117
4118 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4119 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4120 free(uris);
4121 break;
4122 }
4123 case LTTNG_SAVE_SESSION:
4124 {
4125 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4126 &cmd_ctx->creds);
4127 break;
4128 }
4129 case LTTNG_SET_SESSION_SHM_PATH:
4130 {
4131 ret = cmd_set_session_shm_path(cmd_ctx->session,
4132 cmd_ctx->lsm->u.set_shm_path.shm_path);
4133 break;
4134 }
4135 default:
4136 ret = LTTNG_ERR_UND;
4137 break;
4138 }
4139
4140 error:
4141 if (cmd_ctx->llm == NULL) {
4142 DBG("Missing llm structure. Allocating one.");
4143 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4144 goto setup_error;
4145 }
4146 }
4147 /* Set return code */
4148 cmd_ctx->llm->ret_code = ret;
4149 setup_error:
4150 if (cmd_ctx->session) {
4151 session_unlock(cmd_ctx->session);
4152 }
4153 if (need_tracing_session) {
4154 session_unlock_list();
4155 }
4156 init_setup_error:
4157 assert(!rcu_read_ongoing());
4158 return ret;
4159 }
4160
4161 /*
4162 * Thread managing health check socket.
4163 */
4164 static void *thread_manage_health(void *data)
4165 {
4166 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4167 uint32_t revents, nb_fd;
4168 struct lttng_poll_event events;
4169 struct health_comm_msg msg;
4170 struct health_comm_reply reply;
4171
4172 DBG("[thread] Manage health check started");
4173
4174 rcu_register_thread();
4175
4176 /* We might hit an error path before this is created. */
4177 lttng_poll_init(&events);
4178
4179 /* Create unix socket */
4180 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4181 if (sock < 0) {
4182 ERR("Unable to create health check Unix socket");
4183 ret = -1;
4184 goto error;
4185 }
4186
4187 if (is_root) {
4188 /* lttng health client socket path permissions */
4189 ret = chown(health_unix_sock_path, 0,
4190 utils_get_group_id(tracing_group_name));
4191 if (ret < 0) {
4192 ERR("Unable to set group on %s", health_unix_sock_path);
4193 PERROR("chown");
4194 ret = -1;
4195 goto error;
4196 }
4197
4198 ret = chmod(health_unix_sock_path,
4199 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4200 if (ret < 0) {
4201 ERR("Unable to set permissions on %s", health_unix_sock_path);
4202 PERROR("chmod");
4203 ret = -1;
4204 goto error;
4205 }
4206 }
4207
4208 /*
4209 * Set the CLOEXEC flag. Return code is useless because either way, the
4210 * show must go on.
4211 */
4212 (void) utils_set_fd_cloexec(sock);
4213
4214 ret = lttcomm_listen_unix_sock(sock);
4215 if (ret < 0) {
4216 goto error;
4217 }
4218
4219 /*
4220 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4221 * more will be added to this poll set.
4222 */
4223 ret = sessiond_set_thread_pollset(&events, 2);
4224 if (ret < 0) {
4225 goto error;
4226 }
4227
4228 /* Add the application registration socket */
4229 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4230 if (ret < 0) {
4231 goto error;
4232 }
4233
4234 sessiond_notify_ready();
4235
4236 while (1) {
4237 DBG("Health check ready");
4238
4239 /* Inifinite blocking call, waiting for transmission */
4240 restart:
4241 ret = lttng_poll_wait(&events, -1);
4242 if (ret < 0) {
4243 /*
4244 * Restart interrupted system call.
4245 */
4246 if (errno == EINTR) {
4247 goto restart;
4248 }
4249 goto error;
4250 }
4251
4252 nb_fd = ret;
4253
4254 for (i = 0; i < nb_fd; i++) {
4255 /* Fetch once the poll data */
4256 revents = LTTNG_POLL_GETEV(&events, i);
4257 pollfd = LTTNG_POLL_GETFD(&events, i);
4258
4259 if (!revents) {
4260 /* No activity for this FD (poll implementation). */
4261 continue;
4262 }
4263
4264 /* Thread quit pipe has been closed. Killing thread. */
4265 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4266 if (ret) {
4267 err = 0;
4268 goto exit;
4269 }
4270
4271 /* Event on the registration socket */
4272 if (pollfd == sock) {
4273 if (revents & LPOLLIN) {
4274 continue;
4275 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4276 ERR("Health socket poll error");
4277 goto error;
4278 } else {
4279 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4280 goto error;
4281 }
4282 }
4283 }
4284
4285 new_sock = lttcomm_accept_unix_sock(sock);
4286 if (new_sock < 0) {
4287 goto error;
4288 }
4289
4290 /*
4291 * Set the CLOEXEC flag. Return code is useless because either way, the
4292 * show must go on.
4293 */
4294 (void) utils_set_fd_cloexec(new_sock);
4295
4296 DBG("Receiving data from client for health...");
4297 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4298 if (ret <= 0) {
4299 DBG("Nothing recv() from client... continuing");
4300 ret = close(new_sock);
4301 if (ret) {
4302 PERROR("close");
4303 }
4304 new_sock = -1;
4305 continue;
4306 }
4307
4308 rcu_thread_online();
4309
4310 memset(&reply, 0, sizeof(reply));
4311 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4312 /*
4313 * health_check_state returns 0 if health is
4314 * bad.
4315 */
4316 if (!health_check_state(health_sessiond, i)) {
4317 reply.ret_code |= 1ULL << i;
4318 }
4319 }
4320
4321 DBG2("Health check return value %" PRIx64, reply.ret_code);
4322
4323 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4324 if (ret < 0) {
4325 ERR("Failed to send health data back to client");
4326 }
4327
4328 /* End of transmission */
4329 ret = close(new_sock);
4330 if (ret) {
4331 PERROR("close");
4332 }
4333 new_sock = -1;
4334 }
4335
4336 exit:
4337 error:
4338 if (err) {
4339 ERR("Health error occurred in %s", __func__);
4340 }
4341 DBG("Health check thread dying");
4342 unlink(health_unix_sock_path);
4343 if (sock >= 0) {
4344 ret = close(sock);
4345 if (ret) {
4346 PERROR("close");
4347 }
4348 }
4349
4350 lttng_poll_clean(&events);
4351
4352 rcu_unregister_thread();
4353 return NULL;
4354 }
4355
4356 /*
4357 * This thread manage all clients request using the unix client socket for
4358 * communication.
4359 */
4360 static void *thread_manage_clients(void *data)
4361 {
4362 int sock = -1, ret, i, pollfd, err = -1;
4363 int sock_error;
4364 uint32_t revents, nb_fd;
4365 struct command_ctx *cmd_ctx = NULL;
4366 struct lttng_poll_event events;
4367
4368 DBG("[thread] Manage client started");
4369
4370 rcu_register_thread();
4371
4372 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4373
4374 health_code_update();
4375
4376 ret = lttcomm_listen_unix_sock(client_sock);
4377 if (ret < 0) {
4378 goto error_listen;
4379 }
4380
4381 /*
4382 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4383 * more will be added to this poll set.
4384 */
4385 ret = sessiond_set_thread_pollset(&events, 2);
4386 if (ret < 0) {
4387 goto error_create_poll;
4388 }
4389
4390 /* Add the application registration socket */
4391 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4392 if (ret < 0) {
4393 goto error;
4394 }
4395
4396 sessiond_notify_ready();
4397 ret = sem_post(&load_info->message_thread_ready);
4398 if (ret) {
4399 PERROR("sem_post message_thread_ready");
4400 goto error;
4401 }
4402
4403 /* This testpoint is after we signal readiness to the parent. */
4404 if (testpoint(sessiond_thread_manage_clients)) {
4405 goto error;
4406 }
4407
4408 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4409 goto error;
4410 }
4411
4412 health_code_update();
4413
4414 while (1) {
4415 DBG("Accepting client command ...");
4416
4417 /* Inifinite blocking call, waiting for transmission */
4418 restart:
4419 health_poll_entry();
4420 ret = lttng_poll_wait(&events, -1);
4421 health_poll_exit();
4422 if (ret < 0) {
4423 /*
4424 * Restart interrupted system call.
4425 */
4426 if (errno == EINTR) {
4427 goto restart;
4428 }
4429 goto error;
4430 }
4431
4432 nb_fd = ret;
4433
4434 for (i = 0; i < nb_fd; i++) {
4435 /* Fetch once the poll data */
4436 revents = LTTNG_POLL_GETEV(&events, i);
4437 pollfd = LTTNG_POLL_GETFD(&events, i);
4438
4439 health_code_update();
4440
4441 if (!revents) {
4442 /* No activity for this FD (poll implementation). */
4443 continue;
4444 }
4445
4446 /* Thread quit pipe has been closed. Killing thread. */
4447 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4448 if (ret) {
4449 err = 0;
4450 goto exit;
4451 }
4452
4453 /* Event on the registration socket */
4454 if (pollfd == client_sock) {
4455 if (revents & LPOLLIN) {
4456 continue;
4457 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4458 ERR("Client socket poll error");
4459 goto error;
4460 } else {
4461 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4462 goto error;
4463 }
4464 }
4465 }
4466
4467 DBG("Wait for client response");
4468
4469 health_code_update();
4470
4471 sock = lttcomm_accept_unix_sock(client_sock);
4472 if (sock < 0) {
4473 goto error;
4474 }
4475
4476 /*
4477 * Set the CLOEXEC flag. Return code is useless because either way, the
4478 * show must go on.
4479 */
4480 (void) utils_set_fd_cloexec(sock);
4481
4482 /* Set socket option for credentials retrieval */
4483 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4484 if (ret < 0) {
4485 goto error;
4486 }
4487
4488 /* Allocate context command to process the client request */
4489 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4490 if (cmd_ctx == NULL) {
4491 PERROR("zmalloc cmd_ctx");
4492 goto error;
4493 }
4494
4495 /* Allocate data buffer for reception */
4496 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4497 if (cmd_ctx->lsm == NULL) {
4498 PERROR("zmalloc cmd_ctx->lsm");
4499 goto error;
4500 }
4501
4502 cmd_ctx->llm = NULL;
4503 cmd_ctx->session = NULL;
4504
4505 health_code_update();
4506
4507 /*
4508 * Data is received from the lttng client. The struct
4509 * lttcomm_session_msg (lsm) contains the command and data request of
4510 * the client.
4511 */
4512 DBG("Receiving data from client ...");
4513 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4514 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4515 if (ret <= 0) {
4516 DBG("Nothing recv() from client... continuing");
4517 ret = close(sock);
4518 if (ret) {
4519 PERROR("close");
4520 }
4521 sock = -1;
4522 clean_command_ctx(&cmd_ctx);
4523 continue;
4524 }
4525
4526 health_code_update();
4527
4528 // TODO: Validate cmd_ctx including sanity check for
4529 // security purpose.
4530
4531 rcu_thread_online();
4532 /*
4533 * This function dispatch the work to the kernel or userspace tracer
4534 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4535 * informations for the client. The command context struct contains
4536 * everything this function may needs.
4537 */
4538 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4539 rcu_thread_offline();
4540 if (ret < 0) {
4541 ret = close(sock);
4542 if (ret) {
4543 PERROR("close");
4544 }
4545 sock = -1;
4546 /*
4547 * TODO: Inform client somehow of the fatal error. At
4548 * this point, ret < 0 means that a zmalloc failed
4549 * (ENOMEM). Error detected but still accept
4550 * command, unless a socket error has been
4551 * detected.
4552 */
4553 clean_command_ctx(&cmd_ctx);
4554 continue;
4555 }
4556
4557 health_code_update();
4558
4559 DBG("Sending response (size: %d, retcode: %s (%d))",
4560 cmd_ctx->lttng_msg_size,
4561 lttng_strerror(-cmd_ctx->llm->ret_code),
4562 cmd_ctx->llm->ret_code);
4563 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4564 if (ret < 0) {
4565 ERR("Failed to send data back to client");
4566 }
4567
4568 /* End of transmission */
4569 ret = close(sock);
4570 if (ret) {
4571 PERROR("close");
4572 }
4573 sock = -1;
4574
4575 clean_command_ctx(&cmd_ctx);
4576
4577 health_code_update();
4578 }
4579
4580 exit:
4581 error:
4582 if (sock >= 0) {
4583 ret = close(sock);
4584 if (ret) {
4585 PERROR("close");
4586 }
4587 }
4588
4589 lttng_poll_clean(&events);
4590 clean_command_ctx(&cmd_ctx);
4591
4592 error_listen:
4593 error_create_poll:
4594 unlink(client_unix_sock_path);
4595 if (client_sock >= 0) {
4596 ret = close(client_sock);
4597 if (ret) {
4598 PERROR("close");
4599 }
4600 }
4601
4602 if (err) {
4603 health_error();
4604 ERR("Health error occurred in %s", __func__);
4605 }
4606
4607 health_unregister(health_sessiond);
4608
4609 DBG("Client thread dying");
4610
4611 rcu_unregister_thread();
4612
4613 /*
4614 * Since we are creating the consumer threads, we own them, so we need
4615 * to join them before our thread exits.
4616 */
4617 ret = join_consumer_thread(&kconsumer_data);
4618 if (ret) {
4619 errno = ret;
4620 PERROR("join_consumer");
4621 }
4622
4623 ret = join_consumer_thread(&ustconsumer32_data);
4624 if (ret) {
4625 errno = ret;
4626 PERROR("join_consumer ust32");
4627 }
4628
4629 ret = join_consumer_thread(&ustconsumer64_data);
4630 if (ret) {
4631 errno = ret;
4632 PERROR("join_consumer ust64");
4633 }
4634 return NULL;
4635 }
4636
4637
4638 /*
4639 * usage function on stderr
4640 */
4641 static void usage(void)
4642 {
4643 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
4644 fprintf(stderr, " -h, --help Display this usage.\n");
4645 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4646 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4647 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4648 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4649 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4650 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4651 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4652 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4653 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4654 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4655 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4656 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4657 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
4658 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
4659 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4660 fprintf(stderr, " -V, --version Show version number.\n");
4661 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4662 fprintf(stderr, " -q, --quiet No output at all.\n");
4663 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4664 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4665 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4666 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4667 fprintf(stderr, " --agent-tcp-port Agent registration TCP port\n");
4668 fprintf(stderr, " -f --config PATH Load daemon configuration file\n");
4669 fprintf(stderr, " -l --load PATH Load session configuration\n");
4670 fprintf(stderr, " --kmod-probes Specify kernel module probes to load\n");
4671 fprintf(stderr, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4672 }
4673
4674 static int string_match(const char *str1, const char *str2)
4675 {
4676 return (str1 && str2) && !strcmp(str1, str2);
4677 }
4678
4679 /*
4680 * Take an option from the getopt output and set it in the right variable to be
4681 * used later.
4682 *
4683 * Return 0 on success else a negative value.
4684 */
4685 static int set_option(int opt, const char *arg, const char *optname)
4686 {
4687 int ret = 0;
4688
4689 if (arg && arg[0] == '\0') {
4690 /*
4691 * This only happens if the value is read from daemon config
4692 * file. This means the option requires an argument and the
4693 * configuration file contains a line such as:
4694 * my_option =
4695 */
4696 ret = -EINVAL;
4697 goto end;
4698 }
4699
4700 if (string_match(optname, "client-sock") || opt == 'c') {
4701 if (lttng_is_setuid_setgid()) {
4702 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4703 "-c, --client-sock");
4704 } else {
4705 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4706 }
4707 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4708 if (lttng_is_setuid_setgid()) {
4709 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4710 "-a, --apps-sock");
4711 } else {
4712 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4713 }
4714 } else if (string_match(optname, "daemonize") || opt == 'd') {
4715 opt_daemon = 1;
4716 } else if (string_match(optname, "background") || opt == 'b') {
4717 opt_background = 1;
4718 } else if (string_match(optname, "group") || opt == 'g') {
4719 if (lttng_is_setuid_setgid()) {
4720 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4721 "-g, --group");
4722 } else {
4723 /*
4724 * If the override option is set, the pointer points to a
4725 * *non* const thus freeing it even though the variable type is
4726 * set to const.
4727 */
4728 if (tracing_group_name_override) {
4729 free((void *) tracing_group_name);
4730 }
4731 tracing_group_name = strdup(arg);
4732 if (!tracing_group_name) {
4733 PERROR("strdup");
4734 ret = -ENOMEM;
4735 }
4736 tracing_group_name_override = 1;
4737 }
4738 } else if (string_match(optname, "help") || opt == 'h') {
4739 usage();
4740 exit(EXIT_SUCCESS);
4741 } else if (string_match(optname, "version") || opt == 'V') {
4742 fprintf(stdout, "%s\n", VERSION);
4743 exit(EXIT_SUCCESS);
4744 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4745 opt_sig_parent = 1;
4746 } else if (string_match(optname, "kconsumerd-err-sock")) {
4747 if (lttng_is_setuid_setgid()) {
4748 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4749 "--kconsumerd-err-sock");
4750 } else {
4751 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4752 }
4753 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4754 if (lttng_is_setuid_setgid()) {
4755 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4756 "--kconsumerd-cmd-sock");
4757 } else {
4758 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4759 }
4760 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4761 if (lttng_is_setuid_setgid()) {
4762 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4763 "--ustconsumerd64-err-sock");
4764 } else {
4765 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4766 }
4767 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4768 if (lttng_is_setuid_setgid()) {
4769 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4770 "--ustconsumerd64-cmd-sock");
4771 } else {
4772 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4773 }
4774 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4775 if (lttng_is_setuid_setgid()) {
4776 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4777 "--ustconsumerd32-err-sock");
4778 } else {
4779 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4780 }
4781 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4782 if (lttng_is_setuid_setgid()) {
4783 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4784 "--ustconsumerd32-cmd-sock");
4785 } else {
4786 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4787 }
4788 } else if (string_match(optname, "no-kernel")) {
4789 opt_no_kernel = 1;
4790 } else if (string_match(optname, "quiet") || opt == 'q') {
4791 lttng_opt_quiet = 1;
4792 } else if (string_match(optname, "verbose") || opt == 'v') {
4793 /* Verbose level can increase using multiple -v */
4794 if (arg) {
4795 /* Value obtained from config file */
4796 lttng_opt_verbose = config_parse_value(arg);
4797 } else {
4798 /* -v used on command line */
4799 lttng_opt_verbose++;
4800 }
4801 /* Clamp value to [0, 3] */
4802 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4803 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4804 } else if (string_match(optname, "verbose-consumer")) {
4805 if (arg) {
4806 opt_verbose_consumer = config_parse_value(arg);
4807 } else {
4808 opt_verbose_consumer += 1;
4809 }
4810 } else if (string_match(optname, "consumerd32-path")) {
4811 if (lttng_is_setuid_setgid()) {
4812 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4813 "--consumerd32-path");
4814 } else {
4815 if (consumerd32_bin_override) {
4816 free((void *) consumerd32_bin);
4817 }
4818 consumerd32_bin = strdup(arg);
4819 if (!consumerd32_bin) {
4820 PERROR("strdup");
4821 ret = -ENOMEM;
4822 }
4823 consumerd32_bin_override = 1;
4824 }
4825 } else if (string_match(optname, "consumerd32-libdir")) {
4826 if (lttng_is_setuid_setgid()) {
4827 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4828 "--consumerd32-libdir");
4829 } else {
4830 if (consumerd32_libdir_override) {
4831 free((void *) consumerd32_libdir);
4832 }
4833 consumerd32_libdir = strdup(arg);
4834 if (!consumerd32_libdir) {
4835 PERROR("strdup");
4836 ret = -ENOMEM;
4837 }
4838 consumerd32_libdir_override = 1;
4839 }
4840 } else if (string_match(optname, "consumerd64-path")) {
4841 if (lttng_is_setuid_setgid()) {
4842 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4843 "--consumerd64-path");
4844 } else {
4845 if (consumerd64_bin_override) {
4846 free((void *) consumerd64_bin);
4847 }
4848 consumerd64_bin = strdup(arg);
4849 if (!consumerd64_bin) {
4850 PERROR("strdup");
4851 ret = -ENOMEM;
4852 }
4853 consumerd64_bin_override = 1;
4854 }
4855 } else if (string_match(optname, "consumerd64-libdir")) {
4856 if (lttng_is_setuid_setgid()) {
4857 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4858 "--consumerd64-libdir");
4859 } else {
4860 if (consumerd64_libdir_override) {
4861 free((void *) consumerd64_libdir);
4862 }
4863 consumerd64_libdir = strdup(arg);
4864 if (!consumerd64_libdir) {
4865 PERROR("strdup");
4866 ret = -ENOMEM;
4867 }
4868 consumerd64_libdir_override = 1;
4869 }
4870 } else if (string_match(optname, "pidfile") || opt == 'p') {
4871 if (lttng_is_setuid_setgid()) {
4872 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4873 "-p, --pidfile");
4874 } else {
4875 free(opt_pidfile);
4876 opt_pidfile = strdup(arg);
4877 if (!opt_pidfile) {
4878 PERROR("strdup");
4879 ret = -ENOMEM;
4880 }
4881 }
4882 } else if (string_match(optname, "agent-tcp-port")) {
4883 if (lttng_is_setuid_setgid()) {
4884 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4885 "--agent-tcp-port");
4886 } else {
4887 unsigned long v;
4888
4889 if (!arg) {
4890 ret = -EINVAL;
4891 goto end;
4892 }
4893 errno = 0;
4894 v = strtoul(arg, NULL, 0);
4895 if (errno != 0 || !isdigit(arg[0])) {
4896 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4897 return -1;
4898 }
4899 if (v == 0 || v >= 65535) {
4900 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4901 return -1;
4902 }
4903 agent_tcp_port = (uint32_t) v;
4904 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4905 }
4906 } else if (string_match(optname, "load") || opt == 'l') {
4907 if (lttng_is_setuid_setgid()) {
4908 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4909 "-l, --load");
4910 } else {
4911 free(opt_load_session_path);
4912 opt_load_session_path = strdup(arg);
4913 if (!opt_load_session_path) {
4914 PERROR("strdup");
4915 ret = -ENOMEM;
4916 }
4917 }
4918 } else if (string_match(optname, "kmod-probes")) {
4919 if (lttng_is_setuid_setgid()) {
4920 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4921 "--kmod-probes");
4922 } else {
4923 free(kmod_probes_list);
4924 kmod_probes_list = strdup(arg);
4925 if (!kmod_probes_list) {
4926 PERROR("strdup");
4927 ret = -ENOMEM;
4928 }
4929 }
4930 } else if (string_match(optname, "extra-kmod-probes")) {
4931 if (lttng_is_setuid_setgid()) {
4932 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4933 "--extra-kmod-probes");
4934 } else {
4935 free(kmod_extra_probes_list);
4936 kmod_extra_probes_list = strdup(arg);
4937 if (!kmod_extra_probes_list) {
4938 PERROR("strdup");
4939 ret = -ENOMEM;
4940 }
4941 }
4942 } else if (string_match(optname, "config") || opt == 'f') {
4943 /* This is handled in set_options() thus silent skip. */
4944 goto end;
4945 } else {
4946 /* Unknown option or other error.
4947 * Error is printed by getopt, just return */
4948 ret = -1;
4949 }
4950
4951 end:
4952 if (ret == -EINVAL) {
4953 const char *opt_name = "unknown";
4954 int i;
4955
4956 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
4957 i++) {
4958 if (opt == long_options[i].val) {
4959 opt_name = long_options[i].name;
4960 break;
4961 }
4962 }
4963
4964 WARN("Invalid argument provided for option \"%s\", using default value.",
4965 opt_name);
4966 }
4967
4968 return ret;
4969 }
4970
4971 /*
4972 * config_entry_handler_cb used to handle options read from a config file.
4973 * See config_entry_handler_cb comment in common/config/session-config.h for the
4974 * return value conventions.
4975 */
4976 static int config_entry_handler(const struct config_entry *entry, void *unused)
4977 {
4978 int ret = 0, i;
4979
4980 if (!entry || !entry->name || !entry->value) {
4981 ret = -EINVAL;
4982 goto end;
4983 }
4984
4985 /* Check if the option is to be ignored */
4986 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
4987 if (!strcmp(entry->name, config_ignore_options[i])) {
4988 goto end;
4989 }
4990 }
4991
4992 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
4993 i++) {
4994
4995 /* Ignore if not fully matched. */
4996 if (strcmp(entry->name, long_options[i].name)) {
4997 continue;
4998 }
4999
5000 /*
5001 * If the option takes no argument on the command line, we have to
5002 * check if the value is "true". We support non-zero numeric values,
5003 * true, on and yes.
5004 */
5005 if (!long_options[i].has_arg) {
5006 ret = config_parse_value(entry->value);
5007 if (ret <= 0) {
5008 if (ret) {
5009 WARN("Invalid configuration value \"%s\" for option %s",
5010 entry->value, entry->name);
5011 }
5012 /* False, skip boolean config option. */
5013 goto end;
5014 }
5015 }
5016
5017 ret = set_option(long_options[i].val, entry->value, entry->name);
5018 goto end;
5019 }
5020
5021 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5022
5023 end:
5024 return ret;
5025 }
5026
5027 /*
5028 * daemon configuration loading and argument parsing
5029 */
5030 static int set_options(int argc, char **argv)
5031 {
5032 int ret = 0, c = 0, option_index = 0;
5033 int orig_optopt = optopt, orig_optind = optind;
5034 char *optstring;
5035 const char *config_path = NULL;
5036
5037 optstring = utils_generate_optstring(long_options,
5038 sizeof(long_options) / sizeof(struct option));
5039 if (!optstring) {
5040 ret = -ENOMEM;
5041 goto end;
5042 }
5043
5044 /* Check for the --config option */
5045 while ((c = getopt_long(argc, argv, optstring, long_options,
5046 &option_index)) != -1) {
5047 if (c == '?') {
5048 ret = -EINVAL;
5049 goto end;
5050 } else if (c != 'f') {
5051 /* if not equal to --config option. */
5052 continue;
5053 }
5054
5055 if (lttng_is_setuid_setgid()) {
5056 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5057 "-f, --config");
5058 } else {
5059 config_path = utils_expand_path(optarg);
5060 if (!config_path) {
5061 ERR("Failed to resolve path: %s", optarg);
5062 }
5063 }
5064 }
5065
5066 ret = config_get_section_entries(config_path, config_section_name,
5067 config_entry_handler, NULL);
5068 if (ret) {
5069 if (ret > 0) {
5070 ERR("Invalid configuration option at line %i", ret);
5071 ret = -1;
5072 }
5073 goto end;
5074 }
5075
5076 /* Reset getopt's global state */
5077 optopt = orig_optopt;
5078 optind = orig_optind;
5079 while (1) {
5080 option_index = -1;
5081 /*
5082 * getopt_long() will not set option_index if it encounters a
5083 * short option.
5084 */
5085 c = getopt_long(argc, argv, optstring, long_options,
5086 &option_index);
5087 if (c == -1) {
5088 break;
5089 }
5090
5091 /*
5092 * Pass NULL as the long option name if popt left the index
5093 * unset.
5094 */
5095 ret = set_option(c, optarg,
5096 option_index < 0 ? NULL :
5097 long_options[option_index].name);
5098 if (ret < 0) {
5099 break;
5100 }
5101 }
5102
5103 end:
5104 free(optstring);
5105 return ret;
5106 }
5107
5108 /*
5109 * Creates the two needed socket by the daemon.
5110 * apps_sock - The communication socket for all UST apps.
5111 * client_sock - The communication of the cli tool (lttng).
5112 */
5113 static int init_daemon_socket(void)
5114 {
5115 int ret = 0;
5116 mode_t old_umask;
5117
5118 old_umask = umask(0);
5119
5120 /* Create client tool unix socket */
5121 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5122 if (client_sock < 0) {
5123 ERR("Create unix sock failed: %s", client_unix_sock_path);
5124 ret = -1;
5125 goto end;
5126 }
5127
5128 /* Set the cloexec flag */
5129 ret = utils_set_fd_cloexec(client_sock);
5130 if (ret < 0) {
5131 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5132 "Continuing but note that the consumer daemon will have a "
5133 "reference to this socket on exec()", client_sock);
5134 }
5135
5136 /* File permission MUST be 660 */
5137 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5138 if (ret < 0) {
5139 ERR("Set file permissions failed: %s", client_unix_sock_path);
5140 PERROR("chmod");
5141 goto end;
5142 }
5143
5144 /* Create the application unix socket */
5145 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5146 if (apps_sock < 0) {
5147 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5148 ret = -1;
5149 goto end;
5150 }
5151
5152 /* Set the cloexec flag */
5153 ret = utils_set_fd_cloexec(apps_sock);
5154 if (ret < 0) {
5155 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5156 "Continuing but note that the consumer daemon will have a "
5157 "reference to this socket on exec()", apps_sock);
5158 }
5159
5160 /* File permission MUST be 666 */
5161 ret = chmod(apps_unix_sock_path,
5162 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5163 if (ret < 0) {
5164 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5165 PERROR("chmod");
5166 goto end;
5167 }
5168
5169 DBG3("Session daemon client socket %d and application socket %d created",
5170 client_sock, apps_sock);
5171
5172 end:
5173 umask(old_umask);
5174 return ret;
5175 }
5176
5177 /*
5178 * Check if the global socket is available, and if a daemon is answering at the
5179 * other side. If yes, error is returned.
5180 */
5181 static int check_existing_daemon(void)
5182 {
5183 /* Is there anybody out there ? */
5184 if (lttng_session_daemon_alive()) {
5185 return -EEXIST;
5186 }
5187
5188 return 0;
5189 }
5190
5191 /*
5192 * Set the tracing group gid onto the client socket.
5193 *
5194 * Race window between mkdir and chown is OK because we are going from more
5195 * permissive (root.root) to less permissive (root.tracing).
5196 */
5197 static int set_permissions(char *rundir)
5198 {
5199 int ret;
5200 gid_t gid;
5201
5202 gid = utils_get_group_id(tracing_group_name);
5203
5204 /* Set lttng run dir */
5205 ret = chown(rundir, 0, gid);
5206 if (ret < 0) {
5207 ERR("Unable to set group on %s", rundir);
5208 PERROR("chown");
5209 }
5210
5211 /*
5212 * Ensure all applications and tracing group can search the run
5213 * dir. Allow everyone to read the directory, since it does not
5214 * buy us anything to hide its content.
5215 */
5216 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5217 if (ret < 0) {
5218 ERR("Unable to set permissions on %s", rundir);
5219 PERROR("chmod");
5220 }
5221
5222 /* lttng client socket path */
5223 ret = chown(client_unix_sock_path, 0, gid);
5224 if (ret < 0) {
5225 ERR("Unable to set group on %s", client_unix_sock_path);
5226 PERROR("chown");
5227 }
5228
5229 /* kconsumer error socket path */
5230 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5231 if (ret < 0) {
5232 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5233 PERROR("chown");
5234 }
5235
5236 /* 64-bit ustconsumer error socket path */
5237 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5238 if (ret < 0) {
5239 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5240 PERROR("chown");
5241 }
5242
5243 /* 32-bit ustconsumer compat32 error socket path */
5244 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5245 if (ret < 0) {
5246 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5247 PERROR("chown");
5248 }
5249
5250 DBG("All permissions are set");
5251
5252 return ret;
5253 }
5254
5255 /*
5256 * Create the lttng run directory needed for all global sockets and pipe.
5257 */
5258 static int create_lttng_rundir(const char *rundir)
5259 {
5260 int ret;
5261
5262 DBG3("Creating LTTng run directory: %s", rundir);
5263
5264 ret = mkdir(rundir, S_IRWXU);
5265 if (ret < 0) {
5266 if (errno != EEXIST) {
5267 ERR("Unable to create %s", rundir);
5268 goto error;
5269 } else {
5270 ret = 0;
5271 }
5272 }
5273
5274 error:
5275 return ret;
5276 }
5277
5278 /*
5279 * Setup sockets and directory needed by the kconsumerd communication with the
5280 * session daemon.
5281 */
5282 static int set_consumer_sockets(struct consumer_data *consumer_data,
5283 const char *rundir)
5284 {
5285 int ret;
5286 char path[PATH_MAX];
5287
5288 switch (consumer_data->type) {
5289 case LTTNG_CONSUMER_KERNEL:
5290 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5291 break;
5292 case LTTNG_CONSUMER64_UST:
5293 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5294 break;
5295 case LTTNG_CONSUMER32_UST:
5296 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5297 break;
5298 default:
5299 ERR("Consumer type unknown");
5300 ret = -EINVAL;
5301 goto error;
5302 }
5303
5304 DBG2("Creating consumer directory: %s", path);
5305
5306 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5307 if (ret < 0) {
5308 if (errno != EEXIST) {
5309 PERROR("mkdir");
5310 ERR("Failed to create %s", path);
5311 goto error;
5312 }
5313 ret = -1;
5314 }
5315 if (is_root) {
5316 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5317 if (ret < 0) {
5318 ERR("Unable to set group on %s", path);
5319 PERROR("chown");
5320 goto error;
5321 }
5322 }
5323
5324 /* Create the kconsumerd error unix socket */
5325 consumer_data->err_sock =
5326 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5327 if (consumer_data->err_sock < 0) {
5328 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5329 ret = -1;
5330 goto error;
5331 }
5332
5333 /*
5334 * Set the CLOEXEC flag. Return code is useless because either way, the
5335 * show must go on.
5336 */
5337 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5338 if (ret < 0) {
5339 PERROR("utils_set_fd_cloexec");
5340 /* continue anyway */
5341 }
5342
5343 /* File permission MUST be 660 */
5344 ret = chmod(consumer_data->err_unix_sock_path,
5345 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5346 if (ret < 0) {
5347 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5348 PERROR("chmod");
5349 goto error;
5350 }
5351
5352 error:
5353 return ret;
5354 }
5355
5356 /*
5357 * Signal handler for the daemon
5358 *
5359 * Simply stop all worker threads, leaving main() return gracefully after
5360 * joining all threads and calling cleanup().
5361 */
5362 static void sighandler(int sig)
5363 {
5364 switch (sig) {
5365 case SIGPIPE:
5366 DBG("SIGPIPE caught");
5367 return;
5368 case SIGINT:
5369 DBG("SIGINT caught");
5370 stop_threads();
5371 break;
5372 case SIGTERM:
5373 DBG("SIGTERM caught");
5374 stop_threads();
5375 break;
5376 case SIGUSR1:
5377 CMM_STORE_SHARED(recv_child_signal, 1);
5378 break;
5379 default:
5380 break;
5381 }
5382 }
5383
5384 /*
5385 * Setup signal handler for :
5386 * SIGINT, SIGTERM, SIGPIPE
5387 */
5388 static int set_signal_handler(void)
5389 {
5390 int ret = 0;
5391 struct sigaction sa;
5392 sigset_t sigset;
5393
5394 if ((ret = sigemptyset(&sigset)) < 0) {
5395 PERROR("sigemptyset");
5396 return ret;
5397 }
5398
5399 sa.sa_handler = sighandler;
5400 sa.sa_mask = sigset;
5401 sa.sa_flags = 0;
5402 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5403 PERROR("sigaction");
5404 return ret;
5405 }
5406
5407 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5408 PERROR("sigaction");
5409 return ret;
5410 }
5411
5412 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5413 PERROR("sigaction");
5414 return ret;
5415 }
5416
5417 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5418 PERROR("sigaction");
5419 return ret;
5420 }
5421
5422 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5423
5424 return ret;
5425 }
5426
5427 /*
5428 * Set open files limit to unlimited. This daemon can open a large number of
5429 * file descriptors in order to consumer multiple kernel traces.
5430 */
5431 static void set_ulimit(void)
5432 {
5433 int ret;
5434 struct rlimit lim;
5435
5436 /* The kernel does not allowed an infinite limit for open files */
5437 lim.rlim_cur = 65535;
5438 lim.rlim_max = 65535;
5439
5440 ret = setrlimit(RLIMIT_NOFILE, &lim);
5441 if (ret < 0) {
5442 PERROR("failed to set open files limit");
5443 }
5444 }
5445
5446 /*
5447 * Write pidfile using the rundir and opt_pidfile.
5448 */
5449 static int write_pidfile(void)
5450 {
5451 int ret;
5452 char pidfile_path[PATH_MAX];
5453
5454 assert(rundir);
5455
5456 if (opt_pidfile) {
5457 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
5458 } else {
5459 /* Build pidfile path from rundir and opt_pidfile. */
5460 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5461 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5462 if (ret < 0) {
5463 PERROR("snprintf pidfile path");
5464 goto error;
5465 }
5466 }
5467
5468 /*
5469 * Create pid file in rundir.
5470 */
5471 ret = utils_create_pid_file(getpid(), pidfile_path);
5472 error:
5473 return ret;
5474 }
5475
5476 /*
5477 * Create lockfile using the rundir and return its fd.
5478 */
5479 static int create_lockfile(void)
5480 {
5481 int ret;
5482 char lockfile_path[PATH_MAX];
5483
5484 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5485 if (ret < 0) {
5486 goto error;
5487 }
5488
5489 ret = utils_create_lock_file(lockfile_path);
5490 error:
5491 return ret;
5492 }
5493
5494 /*
5495 * Write agent TCP port using the rundir.
5496 */
5497 static int write_agent_port(void)
5498 {
5499 int ret;
5500 char path[PATH_MAX];
5501
5502 assert(rundir);
5503
5504 ret = snprintf(path, sizeof(path), "%s/"
5505 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5506 if (ret < 0) {
5507 PERROR("snprintf agent port path");
5508 goto error;
5509 }
5510
5511 /*
5512 * Create TCP agent port file in rundir.
5513 */
5514 ret = utils_create_pid_file(agent_tcp_port, path);
5515
5516 error:
5517 return ret;
5518 }
5519
5520 /*
5521 * main
5522 */
5523 int main(int argc, char **argv)
5524 {
5525 int ret = 0, retval = 0;
5526 void *status;
5527 const char *home_path, *env_app_timeout;
5528
5529 init_kernel_workarounds();
5530
5531 rcu_register_thread();
5532
5533 if (set_signal_handler()) {
5534 retval = -1;
5535 goto exit_set_signal_handler;
5536 }
5537
5538 setup_consumerd_path();
5539
5540 page_size = sysconf(_SC_PAGESIZE);
5541 if (page_size < 0) {
5542 PERROR("sysconf _SC_PAGESIZE");
5543 page_size = LONG_MAX;
5544 WARN("Fallback page size to %ld", page_size);
5545 }
5546
5547 /*
5548 * Parse arguments and load the daemon configuration file.
5549 *
5550 * We have an exit_options exit path to free memory reserved by
5551 * set_options. This is needed because the rest of sessiond_cleanup()
5552 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5553 * depends on set_options.
5554 */
5555 progname = argv[0];
5556 if (set_options(argc, argv)) {
5557 retval = -1;
5558 goto exit_options;
5559 }
5560
5561 /* Daemonize */
5562 if (opt_daemon || opt_background) {
5563 int i;
5564
5565 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5566 !opt_background);
5567 if (ret < 0) {
5568 retval = -1;
5569 goto exit_options;
5570 }
5571
5572 /*
5573 * We are in the child. Make sure all other file descriptors are
5574 * closed, in case we are called with more opened file
5575 * descriptors than the standard ones.
5576 */
5577 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5578 (void) close(i);
5579 }
5580 }
5581
5582 if (run_as_create_worker(argv[0]) < 0) {
5583 goto exit_create_run_as_worker_cleanup;
5584 }
5585
5586 /*
5587 * Starting from here, we can create threads. This needs to be after
5588 * lttng_daemonize due to RCU.
5589 */
5590
5591 /*
5592 * Initialize the health check subsystem. This call should set the
5593 * appropriate time values.
5594 */
5595 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5596 if (!health_sessiond) {
5597 PERROR("health_app_create error");
5598 retval = -1;
5599 goto exit_health_sessiond_cleanup;
5600 }
5601
5602 if (init_ht_cleanup_quit_pipe()) {
5603 retval = -1;
5604 goto exit_ht_cleanup_quit_pipe;
5605 }
5606
5607 /* Setup the thread ht_cleanup communication pipe. */
5608 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5609 retval = -1;
5610 goto exit_ht_cleanup_pipe;
5611 }
5612
5613 /* Set up max poll set size */
5614 if (lttng_poll_set_max_size()) {
5615 retval = -1;
5616 goto exit_set_max_size;
5617 }
5618
5619 /* Create thread to clean up RCU hash tables */
5620 ret = pthread_create(&ht_cleanup_thread, NULL,
5621 thread_ht_cleanup, (void *) NULL);
5622 if (ret) {
5623 errno = ret;
5624 PERROR("pthread_create ht_cleanup");
5625 retval = -1;
5626 goto exit_ht_cleanup;
5627 }
5628
5629 /* Create thread quit pipe */
5630 if (init_thread_quit_pipe()) {
5631 retval = -1;
5632 goto exit_init_data;
5633 }
5634
5635 /* Check if daemon is UID = 0 */
5636 is_root = !getuid();
5637
5638 if (is_root) {
5639 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5640 if (!rundir) {
5641 retval = -1;
5642 goto exit_init_data;
5643 }
5644
5645 /* Create global run dir with root access */
5646 if (create_lttng_rundir(rundir)) {
5647 retval = -1;
5648 goto exit_init_data;
5649 }
5650
5651 if (strlen(apps_unix_sock_path) == 0) {
5652 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5653 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5654 if (ret < 0) {
5655 retval = -1;
5656 goto exit_init_data;
5657 }
5658 }
5659
5660 if (strlen(client_unix_sock_path) == 0) {
5661 ret = snprintf(client_unix_sock_path, PATH_MAX,
5662 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5663 if (ret < 0) {
5664 retval = -1;
5665 goto exit_init_data;
5666 }
5667 }
5668
5669 /* Set global SHM for ust */
5670 if (strlen(wait_shm_path) == 0) {
5671 ret = snprintf(wait_shm_path, PATH_MAX,
5672 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5673 if (ret < 0) {
5674 retval = -1;
5675 goto exit_init_data;
5676 }
5677 }
5678
5679 if (strlen(health_unix_sock_path) == 0) {
5680 ret = snprintf(health_unix_sock_path,
5681 sizeof(health_unix_sock_path),
5682 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5683 if (ret < 0) {
5684 retval = -1;
5685 goto exit_init_data;
5686 }
5687 }
5688
5689 /* Setup kernel consumerd path */
5690 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5691 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5692 if (ret < 0) {
5693 retval = -1;
5694 goto exit_init_data;
5695 }
5696 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5697 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5698 if (ret < 0) {
5699 retval = -1;
5700 goto exit_init_data;
5701 }
5702
5703 DBG2("Kernel consumer err path: %s",
5704 kconsumer_data.err_unix_sock_path);
5705 DBG2("Kernel consumer cmd path: %s",
5706 kconsumer_data.cmd_unix_sock_path);
5707 } else {
5708 home_path = utils_get_home_dir();
5709 if (home_path == NULL) {
5710 /* TODO: Add --socket PATH option */
5711 ERR("Can't get HOME directory for sockets creation.");
5712 retval = -1;
5713 goto exit_init_data;
5714 }
5715
5716 /*
5717 * Create rundir from home path. This will create something like
5718 * $HOME/.lttng
5719 */
5720 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5721 if (ret < 0) {
5722 retval = -1;
5723 goto exit_init_data;
5724 }
5725
5726 if (create_lttng_rundir(rundir)) {
5727 retval = -1;
5728 goto exit_init_data;
5729 }
5730
5731 if (strlen(apps_unix_sock_path) == 0) {
5732 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5733 DEFAULT_HOME_APPS_UNIX_SOCK,
5734 home_path);
5735 if (ret < 0) {
5736 retval = -1;
5737 goto exit_init_data;
5738 }
5739 }
5740
5741 /* Set the cli tool unix socket path */
5742 if (strlen(client_unix_sock_path) == 0) {
5743 ret = snprintf(client_unix_sock_path, PATH_MAX,
5744 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5745 home_path);
5746 if (ret < 0) {
5747 retval = -1;
5748 goto exit_init_data;
5749 }
5750 }
5751
5752 /* Set global SHM for ust */
5753 if (strlen(wait_shm_path) == 0) {
5754 ret = snprintf(wait_shm_path, PATH_MAX,
5755 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5756 getuid());
5757 if (ret < 0) {
5758 retval = -1;
5759 goto exit_init_data;
5760 }
5761 }
5762
5763 /* Set health check Unix path */
5764 if (strlen(health_unix_sock_path) == 0) {
5765 ret = snprintf(health_unix_sock_path,
5766 sizeof(health_unix_sock_path),
5767 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5768 home_path);
5769 if (ret < 0) {
5770 retval = -1;
5771 goto exit_init_data;
5772 }
5773 }
5774 }
5775
5776 lockfile_fd = create_lockfile();
5777 if (lockfile_fd < 0) {
5778 retval = -1;
5779 goto exit_init_data;
5780 }
5781
5782 /* Set consumer initial state */
5783 kernel_consumerd_state = CONSUMER_STOPPED;
5784 ust_consumerd_state = CONSUMER_STOPPED;
5785
5786 DBG("Client socket path %s", client_unix_sock_path);
5787 DBG("Application socket path %s", apps_unix_sock_path);
5788 DBG("Application wait path %s", wait_shm_path);
5789 DBG("LTTng run directory path: %s", rundir);
5790
5791 /* 32 bits consumerd path setup */
5792 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5793 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5794 if (ret < 0) {
5795 PERROR("snprintf 32-bit consumer error socket path");
5796 retval = -1;
5797 goto exit_init_data;
5798 }
5799 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5800 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5801 if (ret < 0) {
5802 PERROR("snprintf 32-bit consumer command socket path");
5803 retval = -1;
5804 goto exit_init_data;
5805 }
5806
5807 DBG2("UST consumer 32 bits err path: %s",
5808 ustconsumer32_data.err_unix_sock_path);
5809 DBG2("UST consumer 32 bits cmd path: %s",
5810 ustconsumer32_data.cmd_unix_sock_path);
5811
5812 /* 64 bits consumerd path setup */
5813 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5814 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5815 if (ret < 0) {
5816 PERROR("snprintf 64-bit consumer error socket path");
5817 retval = -1;
5818 goto exit_init_data;
5819 }
5820 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5821 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5822 if (ret < 0) {
5823 PERROR("snprintf 64-bit consumer command socket path");
5824 retval = -1;
5825 goto exit_init_data;
5826 }
5827
5828 DBG2("UST consumer 64 bits err path: %s",
5829 ustconsumer64_data.err_unix_sock_path);
5830 DBG2("UST consumer 64 bits cmd path: %s",
5831 ustconsumer64_data.cmd_unix_sock_path);
5832
5833 /*
5834 * See if daemon already exist.
5835 */
5836 if (check_existing_daemon()) {
5837 ERR("Already running daemon.\n");
5838 /*
5839 * We do not goto exit because we must not cleanup()
5840 * because a daemon is already running.
5841 */
5842 retval = -1;
5843 goto exit_init_data;
5844 }
5845
5846 /*
5847 * Init UST app hash table. Alloc hash table before this point since
5848 * cleanup() can get called after that point.
5849 */
5850 if (ust_app_ht_alloc()) {
5851 ERR("Failed to allocate UST app hash table");
5852 retval = -1;
5853 goto exit_init_data;
5854 }
5855
5856 /*
5857 * Initialize agent app hash table. We allocate the hash table here
5858 * since cleanup() can get called after this point.
5859 */
5860 if (agent_app_ht_alloc()) {
5861 ERR("Failed to allocate Agent app hash table");
5862 retval = -1;
5863 goto exit_init_data;
5864 }
5865
5866 /*
5867 * These actions must be executed as root. We do that *after* setting up
5868 * the sockets path because we MUST make the check for another daemon using
5869 * those paths *before* trying to set the kernel consumer sockets and init
5870 * kernel tracer.
5871 */
5872 if (is_root) {
5873 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5874 retval = -1;
5875 goto exit_init_data;
5876 }
5877
5878 /* Setup kernel tracer */
5879 if (!opt_no_kernel) {
5880 init_kernel_tracer();
5881 if (kernel_tracer_fd >= 0) {
5882 ret = syscall_init_table();
5883 if (ret < 0) {
5884 ERR("Unable to populate syscall table. "
5885 "Syscall tracing won't work "
5886 "for this session daemon.");
5887 }
5888 }
5889 }
5890
5891 /* Set ulimit for open files */
5892 set_ulimit();
5893 }
5894 /* init lttng_fd tracking must be done after set_ulimit. */
5895 lttng_fd_init();
5896
5897 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5898 retval = -1;
5899 goto exit_init_data;
5900 }
5901
5902 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5903 retval = -1;
5904 goto exit_init_data;
5905 }
5906
5907 /* Setup the needed unix socket */
5908 if (init_daemon_socket()) {
5909 retval = -1;
5910 goto exit_init_data;
5911 }
5912
5913 /* Set credentials to socket */
5914 if (is_root && set_permissions(rundir)) {
5915 retval = -1;
5916 goto exit_init_data;
5917 }
5918
5919 /* Get parent pid if -S, --sig-parent is specified. */
5920 if (opt_sig_parent) {
5921 ppid = getppid();
5922 }
5923
5924 /* Setup the kernel pipe for waking up the kernel thread */
5925 if (is_root && !opt_no_kernel) {
5926 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5927 retval = -1;
5928 goto exit_init_data;
5929 }
5930 }
5931
5932 /* Setup the thread apps communication pipe. */
5933 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5934 retval = -1;
5935 goto exit_init_data;
5936 }
5937
5938 /* Setup the thread apps notify communication pipe. */
5939 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5940 retval = -1;
5941 goto exit_init_data;
5942 }
5943
5944 /* Initialize global buffer per UID and PID registry. */
5945 buffer_reg_init_uid_registry();
5946 buffer_reg_init_pid_registry();
5947
5948 /* Init UST command queue. */
5949 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
5950
5951 /*
5952 * Get session list pointer. This pointer MUST NOT be free'd. This list
5953 * is statically declared in session.c
5954 */
5955 session_list_ptr = session_get_list();
5956
5957 cmd_init();
5958
5959 /* Check for the application socket timeout env variable. */
5960 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
5961 if (env_app_timeout) {
5962 app_socket_timeout = atoi(env_app_timeout);
5963 } else {
5964 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
5965 }
5966
5967 ret = write_pidfile();
5968 if (ret) {
5969 ERR("Error in write_pidfile");
5970 retval = -1;
5971 goto exit_init_data;
5972 }
5973 ret = write_agent_port();
5974 if (ret) {
5975 ERR("Error in write_agent_port");
5976 retval = -1;
5977 goto exit_init_data;
5978 }
5979
5980 /* Initialize communication library */
5981 lttcomm_init();
5982 /* Initialize TCP timeout values */
5983 lttcomm_inet_init();
5984
5985 if (load_session_init_data(&load_info) < 0) {
5986 retval = -1;
5987 goto exit_init_data;
5988 }
5989 load_info->path = opt_load_session_path;
5990
5991 /* Create health-check thread */
5992 ret = pthread_create(&health_thread, NULL,
5993 thread_manage_health, (void *) NULL);
5994 if (ret) {
5995 errno = ret;
5996 PERROR("pthread_create health");
5997 retval = -1;
5998 goto exit_health;
5999 }
6000
6001 /* Create thread to manage the client socket */
6002 ret = pthread_create(&client_thread, NULL,
6003 thread_manage_clients, (void *) NULL);
6004 if (ret) {
6005 errno = ret;
6006 PERROR("pthread_create clients");
6007 retval = -1;
6008 goto exit_client;
6009 }
6010
6011 /* Create thread to dispatch registration */
6012 ret = pthread_create(&dispatch_thread, NULL,
6013 thread_dispatch_ust_registration, (void *) NULL);
6014 if (ret) {
6015 errno = ret;
6016 PERROR("pthread_create dispatch");
6017 retval = -1;
6018 goto exit_dispatch;
6019 }
6020
6021 /* Create thread to manage application registration. */
6022 ret = pthread_create(&reg_apps_thread, NULL,
6023 thread_registration_apps, (void *) NULL);
6024 if (ret) {
6025 errno = ret;
6026 PERROR("pthread_create registration");
6027 retval = -1;
6028 goto exit_reg_apps;
6029 }
6030
6031 /* Create thread to manage application socket */
6032 ret = pthread_create(&apps_thread, NULL,
6033 thread_manage_apps, (void *) NULL);
6034 if (ret) {
6035 errno = ret;
6036 PERROR("pthread_create apps");
6037 retval = -1;
6038 goto exit_apps;
6039 }
6040
6041 /* Create thread to manage application notify socket */
6042 ret = pthread_create(&apps_notify_thread, NULL,
6043 ust_thread_manage_notify, (void *) NULL);
6044 if (ret) {
6045 errno = ret;
6046 PERROR("pthread_create notify");
6047 retval = -1;
6048 goto exit_apps_notify;
6049 }
6050
6051 /* Create agent registration thread. */
6052 ret = pthread_create(&agent_reg_thread, NULL,
6053 agent_thread_manage_registration, (void *) NULL);
6054 if (ret) {
6055 errno = ret;
6056 PERROR("pthread_create agent");
6057 retval = -1;
6058 goto exit_agent_reg;
6059 }
6060
6061 /* Don't start this thread if kernel tracing is not requested nor root */
6062 if (is_root && !opt_no_kernel) {
6063 /* Create kernel thread to manage kernel event */
6064 ret = pthread_create(&kernel_thread, NULL,
6065 thread_manage_kernel, (void *) NULL);
6066 if (ret) {
6067 errno = ret;
6068 PERROR("pthread_create kernel");
6069 retval = -1;
6070 goto exit_kernel;
6071 }
6072 }
6073
6074 /* Create session loading thread. */
6075 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
6076 load_info);
6077 if (ret) {
6078 errno = ret;
6079 PERROR("pthread_create load_session_thread");
6080 retval = -1;
6081 goto exit_load_session;
6082 }
6083
6084 /*
6085 * This is where we start awaiting program completion (e.g. through
6086 * signal that asks threads to teardown).
6087 */
6088
6089 ret = pthread_join(load_session_thread, &status);
6090 if (ret) {
6091 errno = ret;
6092 PERROR("pthread_join load_session_thread");
6093 retval = -1;
6094 }
6095 exit_load_session:
6096
6097 if (is_root && !opt_no_kernel) {
6098 ret = pthread_join(kernel_thread, &status);
6099 if (ret) {
6100 errno = ret;
6101 PERROR("pthread_join");
6102 retval = -1;
6103 }
6104 }
6105 exit_kernel:
6106
6107 ret = pthread_join(agent_reg_thread, &status);
6108 if (ret) {
6109 errno = ret;
6110 PERROR("pthread_join agent");
6111 retval = -1;
6112 }
6113 exit_agent_reg:
6114
6115 ret = pthread_join(apps_notify_thread, &status);
6116 if (ret) {
6117 errno = ret;
6118 PERROR("pthread_join apps notify");
6119 retval = -1;
6120 }
6121 exit_apps_notify:
6122
6123 ret = pthread_join(apps_thread, &status);
6124 if (ret) {
6125 errno = ret;
6126 PERROR("pthread_join apps");
6127 retval = -1;
6128 }
6129 exit_apps:
6130
6131 ret = pthread_join(reg_apps_thread, &status);
6132 if (ret) {
6133 errno = ret;
6134 PERROR("pthread_join");
6135 retval = -1;
6136 }
6137 exit_reg_apps:
6138
6139 /*
6140 * Join dispatch thread after joining reg_apps_thread to ensure
6141 * we don't leak applications in the queue.
6142 */
6143 ret = pthread_join(dispatch_thread, &status);
6144 if (ret) {
6145 errno = ret;
6146 PERROR("pthread_join");
6147 retval = -1;
6148 }
6149 exit_dispatch:
6150
6151 ret = pthread_join(client_thread, &status);
6152 if (ret) {
6153 errno = ret;
6154 PERROR("pthread_join");
6155 retval = -1;
6156 }
6157 exit_client:
6158
6159 ret = pthread_join(health_thread, &status);
6160 if (ret) {
6161 errno = ret;
6162 PERROR("pthread_join health thread");
6163 retval = -1;
6164 }
6165 exit_health:
6166
6167 exit_init_data:
6168 /*
6169 * sessiond_cleanup() is called when no other thread is running, except
6170 * the ht_cleanup thread, which is needed to destroy the hash tables.
6171 */
6172 rcu_thread_online();
6173 sessiond_cleanup();
6174 rcu_thread_offline();
6175 rcu_unregister_thread();
6176
6177 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6178 if (ret < 0) {
6179 ERR("write error on ht_cleanup quit pipe");
6180 retval = -1;
6181 }
6182
6183 ret = pthread_join(ht_cleanup_thread, &status);
6184 if (ret) {
6185 errno = ret;
6186 PERROR("pthread_join ht cleanup thread");
6187 retval = -1;
6188 }
6189 exit_ht_cleanup:
6190 exit_set_max_size:
6191
6192 utils_close_pipe(ht_cleanup_pipe);
6193 exit_ht_cleanup_pipe:
6194
6195 /*
6196 * Close the ht_cleanup quit pipe.
6197 */
6198 utils_close_pipe(ht_cleanup_quit_pipe);
6199 exit_ht_cleanup_quit_pipe:
6200
6201 health_app_destroy(health_sessiond);
6202 exit_health_sessiond_cleanup:
6203 exit_create_run_as_worker_cleanup:
6204
6205 exit_options:
6206 /* Ensure all prior call_rcu are done. */
6207 rcu_barrier();
6208
6209 sessiond_cleanup_options();
6210
6211 exit_set_signal_handler:
6212
6213 if (!retval) {
6214 exit(EXIT_SUCCESS);
6215 } else {
6216 exit(EXIT_FAILURE);
6217 }
6218 }
This page took 0.204005 seconds and 3 git commands to generate.