Fix: thread_dispatch_ust_registration needs to be a RCU thread
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <ctype.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
52
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
55 #include "channel.h"
56 #include "cmd.h"
57 #include "consumer.h"
58 #include "context.h"
59 #include "event.h"
60 #include "kernel.h"
61 #include "kernel-consumer.h"
62 #include "modprobe.h"
63 #include "shm.h"
64 #include "ust-ctl.h"
65 #include "ust-consumer.h"
66 #include "utils.h"
67 #include "fd-limit.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
72 #include "save.h"
73 #include "load-session-thread.h"
74 #include "syscall.h"
75 #include "agent.h"
76
77 #define CONSUMERD_FILE "lttng-consumerd"
78
79 const char *progname;
80 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
81 static int tracing_group_name_override;
82 static char *opt_pidfile;
83 static int opt_sig_parent;
84 static int opt_verbose_consumer;
85 static int opt_daemon, opt_background;
86 static int opt_no_kernel;
87 static char *opt_load_session_path;
88 static pid_t ppid; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
90 static char *rundir;
91 static int lockfile_fd = -1;
92
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal;
95
96 /*
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
99 */
100 static struct consumer_data kconsumer_data = {
101 .type = LTTNG_CONSUMER_KERNEL,
102 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
103 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
104 .err_sock = -1,
105 .cmd_sock = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
118 .lock = PTHREAD_MUTEX_INITIALIZER,
119 .cond = PTHREAD_COND_INITIALIZER,
120 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
121 };
122 static struct consumer_data ustconsumer32_data = {
123 .type = LTTNG_CONSUMER32_UST,
124 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
125 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
126 .err_sock = -1,
127 .cmd_sock = -1,
128 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
129 .lock = PTHREAD_MUTEX_INITIALIZER,
130 .cond = PTHREAD_COND_INITIALIZER,
131 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
132 };
133
134 /* Command line options */
135 static const struct option long_options[] = {
136 { "client-sock", required_argument, 0, 'c' },
137 { "apps-sock", required_argument, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
144 { "consumerd32-path", required_argument, 0, '\0' },
145 { "consumerd32-libdir", required_argument, 0, '\0' },
146 { "consumerd64-path", required_argument, 0, '\0' },
147 { "consumerd64-libdir", required_argument, 0, '\0' },
148 { "daemonize", no_argument, 0, 'd' },
149 { "background", no_argument, 0, 'b' },
150 { "sig-parent", no_argument, 0, 'S' },
151 { "help", no_argument, 0, 'h' },
152 { "group", required_argument, 0, 'g' },
153 { "version", no_argument, 0, 'V' },
154 { "quiet", no_argument, 0, 'q' },
155 { "verbose", no_argument, 0, 'v' },
156 { "verbose-consumer", no_argument, 0, '\0' },
157 { "no-kernel", no_argument, 0, '\0' },
158 { "pidfile", required_argument, 0, 'p' },
159 { "agent-tcp-port", required_argument, 0, '\0' },
160 { "config", required_argument, 0, 'f' },
161 { "load", required_argument, 0, 'l' },
162 { "kmod-probes", required_argument, 0, '\0' },
163 { "extra-kmod-probes", required_argument, 0, '\0' },
164 { NULL, 0, 0, 0 }
165 };
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path[PATH_MAX];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path[PATH_MAX];
177 /* global wait shm path for UST */
178 static char wait_shm_path[PATH_MAX];
179 /* Global health check unix path */
180 static char health_unix_sock_path[PATH_MAX];
181
182 /* Sockets and FDs */
183 static int client_sock = -1;
184 static int apps_sock = -1;
185 int kernel_tracer_fd = -1;
186 static int kernel_poll_pipe[2] = { -1, -1 };
187
188 /*
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
191 */
192 static int thread_quit_pipe[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
194
195 /*
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
198 */
199 static int apps_cmd_pipe[2] = { -1, -1 };
200
201 int apps_cmd_notify_pipe[2] = { -1, -1 };
202
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread;
205 static pthread_t apps_notify_thread;
206 static pthread_t reg_apps_thread;
207 static pthread_t client_thread;
208 static pthread_t kernel_thread;
209 static pthread_t dispatch_thread;
210 static pthread_t health_thread;
211 static pthread_t ht_cleanup_thread;
212 static pthread_t agent_reg_thread;
213 static pthread_t load_session_thread;
214
215 /*
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 *
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
223 */
224 static struct ust_cmd_queue ust_cmd_queue;
225
226 /*
227 * Pointer initialized before thread creation.
228 *
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
232 *
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
235 */
236 static struct ltt_session_list *session_list_ptr;
237
238 int ust_consumerd64_fd = -1;
239 int ust_consumerd32_fd = -1;
240
241 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
242 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
243 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
244 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
245 static int consumerd32_bin_override;
246 static int consumerd64_bin_override;
247 static int consumerd32_libdir_override;
248 static int consumerd64_libdir_override;
249
250 static const char *module_proc_lttng = "/proc/lttng";
251
252 /*
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
255 */
256 enum consumerd_state {
257 CONSUMER_STARTED = 1,
258 CONSUMER_STOPPED = 2,
259 CONSUMER_ERROR = 3,
260 };
261
262 /*
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
267 *
268 * The following example shows a possible race condition of this scheme:
269 *
270 * consumer thread error happens
271 * client cmd arrives
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
275 * ...
276 *
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
282 */
283 static enum consumerd_state ust_consumerd_state;
284 static enum consumerd_state kernel_consumerd_state;
285
286 /*
287 * Socket timeout for receiving and sending in seconds.
288 */
289 static int app_socket_timeout;
290
291 /* Set in main() with the current page size. */
292 long page_size;
293
294 /* Application health monitoring */
295 struct health_app *health_sessiond;
296
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
299
300 /* Am I root or not. */
301 int is_root; /* Set to 1 if the daemon is running as root */
302
303 const char * const config_section_name = "sessiond";
304
305 /* Load session thread information to operate. */
306 struct load_session_thread_data *load_info;
307
308 /* Global hash tables */
309 struct lttng_ht *agent_apps_ht_by_sock = NULL;
310
311 /*
312 * Whether sessiond is ready for commands/health check requests.
313 * NR_LTTNG_SESSIOND_READY must match the number of calls to
314 * sessiond_notify_ready().
315 */
316 #define NR_LTTNG_SESSIOND_READY 3
317 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
318
319 /* Notify parents that we are ready for cmd and health check */
320 LTTNG_HIDDEN
321 void sessiond_notify_ready(void)
322 {
323 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
324 /*
325 * Notify parent pid that we are ready to accept command
326 * for client side. This ppid is the one from the
327 * external process that spawned us.
328 */
329 if (opt_sig_parent) {
330 kill(ppid, SIGUSR1);
331 }
332
333 /*
334 * Notify the parent of the fork() process that we are
335 * ready.
336 */
337 if (opt_daemon || opt_background) {
338 kill(child_ppid, SIGUSR1);
339 }
340 }
341 }
342
343 static
344 void setup_consumerd_path(void)
345 {
346 const char *bin, *libdir;
347
348 /*
349 * Allow INSTALL_BIN_PATH to be used as a target path for the
350 * native architecture size consumer if CONFIG_CONSUMER*_PATH
351 * has not been defined.
352 */
353 #if (CAA_BITS_PER_LONG == 32)
354 if (!consumerd32_bin[0]) {
355 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
356 }
357 if (!consumerd32_libdir[0]) {
358 consumerd32_libdir = INSTALL_LIB_PATH;
359 }
360 #elif (CAA_BITS_PER_LONG == 64)
361 if (!consumerd64_bin[0]) {
362 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
363 }
364 if (!consumerd64_libdir[0]) {
365 consumerd64_libdir = INSTALL_LIB_PATH;
366 }
367 #else
368 #error "Unknown bitness"
369 #endif
370
371 /*
372 * runtime env. var. overrides the build default.
373 */
374 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
375 if (bin) {
376 consumerd32_bin = bin;
377 }
378 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
379 if (bin) {
380 consumerd64_bin = bin;
381 }
382 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
383 if (libdir) {
384 consumerd32_libdir = libdir;
385 }
386 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
387 if (libdir) {
388 consumerd64_libdir = libdir;
389 }
390 }
391
392 static
393 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
394 int *a_pipe)
395 {
396 int ret;
397
398 assert(events);
399
400 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
401 if (ret < 0) {
402 goto error;
403 }
404
405 /* Add quit pipe */
406 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
407 if (ret < 0) {
408 goto error;
409 }
410
411 return 0;
412
413 error:
414 return ret;
415 }
416
417 /*
418 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
419 */
420 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
421 {
422 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
423 }
424
425 /*
426 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
427 */
428 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
429 size_t size)
430 {
431 return __sessiond_set_thread_pollset(events, size,
432 ht_cleanup_quit_pipe);
433 }
434
435 static
436 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
437 {
438 if (fd == a_pipe && (events & LPOLLIN)) {
439 return 1;
440 }
441 return 0;
442 }
443
444 /*
445 * Check if the thread quit pipe was triggered.
446 *
447 * Return 1 if it was triggered else 0;
448 */
449 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
450 {
451 return __sessiond_check_thread_quit_pipe(fd, events,
452 thread_quit_pipe[0]);
453 }
454
455 /*
456 * Check if the ht_cleanup thread quit pipe was triggered.
457 *
458 * Return 1 if it was triggered else 0;
459 */
460 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
461 {
462 return __sessiond_check_thread_quit_pipe(fd, events,
463 ht_cleanup_quit_pipe[0]);
464 }
465
466 /*
467 * Init thread quit pipe.
468 *
469 * Return -1 on error or 0 if all pipes are created.
470 */
471 static int __init_thread_quit_pipe(int *a_pipe)
472 {
473 int ret, i;
474
475 ret = pipe(a_pipe);
476 if (ret < 0) {
477 PERROR("thread quit pipe");
478 goto error;
479 }
480
481 for (i = 0; i < 2; i++) {
482 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
483 if (ret < 0) {
484 PERROR("fcntl");
485 goto error;
486 }
487 }
488
489 error:
490 return ret;
491 }
492
493 static int init_thread_quit_pipe(void)
494 {
495 return __init_thread_quit_pipe(thread_quit_pipe);
496 }
497
498 static int init_ht_cleanup_quit_pipe(void)
499 {
500 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
501 }
502
503 /*
504 * Stop all threads by closing the thread quit pipe.
505 */
506 static void stop_threads(void)
507 {
508 int ret;
509
510 /* Stopping all threads */
511 DBG("Terminating all threads");
512 ret = notify_thread_pipe(thread_quit_pipe[1]);
513 if (ret < 0) {
514 ERR("write error on thread quit pipe");
515 }
516
517 /* Dispatch thread */
518 CMM_STORE_SHARED(dispatch_thread_exit, 1);
519 futex_nto1_wake(&ust_cmd_queue.futex);
520 }
521
522 /*
523 * Close every consumer sockets.
524 */
525 static void close_consumer_sockets(void)
526 {
527 int ret;
528
529 if (kconsumer_data.err_sock >= 0) {
530 ret = close(kconsumer_data.err_sock);
531 if (ret < 0) {
532 PERROR("kernel consumer err_sock close");
533 }
534 }
535 if (ustconsumer32_data.err_sock >= 0) {
536 ret = close(ustconsumer32_data.err_sock);
537 if (ret < 0) {
538 PERROR("UST consumerd32 err_sock close");
539 }
540 }
541 if (ustconsumer64_data.err_sock >= 0) {
542 ret = close(ustconsumer64_data.err_sock);
543 if (ret < 0) {
544 PERROR("UST consumerd64 err_sock close");
545 }
546 }
547 if (kconsumer_data.cmd_sock >= 0) {
548 ret = close(kconsumer_data.cmd_sock);
549 if (ret < 0) {
550 PERROR("kernel consumer cmd_sock close");
551 }
552 }
553 if (ustconsumer32_data.cmd_sock >= 0) {
554 ret = close(ustconsumer32_data.cmd_sock);
555 if (ret < 0) {
556 PERROR("UST consumerd32 cmd_sock close");
557 }
558 }
559 if (ustconsumer64_data.cmd_sock >= 0) {
560 ret = close(ustconsumer64_data.cmd_sock);
561 if (ret < 0) {
562 PERROR("UST consumerd64 cmd_sock close");
563 }
564 }
565 }
566
567 /*
568 * Generate the full lock file path using the rundir.
569 *
570 * Return the snprintf() return value thus a negative value is an error.
571 */
572 static int generate_lock_file_path(char *path, size_t len)
573 {
574 int ret;
575
576 assert(path);
577 assert(rundir);
578
579 /* Build lockfile path from rundir. */
580 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
581 if (ret < 0) {
582 PERROR("snprintf lockfile path");
583 }
584
585 return ret;
586 }
587
588 /*
589 * Wait on consumer process termination.
590 *
591 * Need to be called with the consumer data lock held or from a context
592 * ensuring no concurrent access to data (e.g: cleanup).
593 */
594 static void wait_consumer(struct consumer_data *consumer_data)
595 {
596 pid_t ret;
597 int status;
598
599 if (consumer_data->pid <= 0) {
600 return;
601 }
602
603 DBG("Waiting for complete teardown of consumerd (PID: %d)",
604 consumer_data->pid);
605 ret = waitpid(consumer_data->pid, &status, 0);
606 if (ret == -1) {
607 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
608 }
609 if (!WIFEXITED(status)) {
610 ERR("consumerd termination with error: %d",
611 WEXITSTATUS(ret));
612 }
613 consumer_data->pid = 0;
614 }
615
616 /*
617 * Cleanup the session daemon's data structures.
618 */
619 static void sessiond_cleanup(void)
620 {
621 int ret;
622 struct ltt_session *sess, *stmp;
623 char path[PATH_MAX];
624
625 DBG("Cleanup sessiond");
626
627 /*
628 * Close the thread quit pipe. It has already done its job,
629 * since we are now called.
630 */
631 utils_close_pipe(thread_quit_pipe);
632
633 /*
634 * If opt_pidfile is undefined, the default file will be wiped when
635 * removing the rundir.
636 */
637 if (opt_pidfile) {
638 ret = remove(opt_pidfile);
639 if (ret < 0) {
640 PERROR("remove pidfile %s", opt_pidfile);
641 }
642 }
643
644 DBG("Removing sessiond and consumerd content of directory %s", rundir);
645
646 /* sessiond */
647 snprintf(path, PATH_MAX,
648 "%s/%s",
649 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
650 DBG("Removing %s", path);
651 (void) unlink(path);
652
653 snprintf(path, PATH_MAX, "%s/%s", rundir,
654 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
655 DBG("Removing %s", path);
656 (void) unlink(path);
657
658 /* kconsumerd */
659 snprintf(path, PATH_MAX,
660 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
661 rundir);
662 DBG("Removing %s", path);
663 (void) unlink(path);
664
665 snprintf(path, PATH_MAX,
666 DEFAULT_KCONSUMERD_PATH,
667 rundir);
668 DBG("Removing directory %s", path);
669 (void) rmdir(path);
670
671 /* ust consumerd 32 */
672 snprintf(path, PATH_MAX,
673 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
674 rundir);
675 DBG("Removing %s", path);
676 (void) unlink(path);
677
678 snprintf(path, PATH_MAX,
679 DEFAULT_USTCONSUMERD32_PATH,
680 rundir);
681 DBG("Removing directory %s", path);
682 (void) rmdir(path);
683
684 /* ust consumerd 64 */
685 snprintf(path, PATH_MAX,
686 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
687 rundir);
688 DBG("Removing %s", path);
689 (void) unlink(path);
690
691 snprintf(path, PATH_MAX,
692 DEFAULT_USTCONSUMERD64_PATH,
693 rundir);
694 DBG("Removing directory %s", path);
695 (void) rmdir(path);
696
697 DBG("Cleaning up all sessions");
698
699 /* Destroy session list mutex */
700 if (session_list_ptr != NULL) {
701 pthread_mutex_destroy(&session_list_ptr->lock);
702
703 /* Cleanup ALL session */
704 cds_list_for_each_entry_safe(sess, stmp,
705 &session_list_ptr->head, list) {
706 cmd_destroy_session(sess, kernel_poll_pipe[1]);
707 }
708 }
709
710 wait_consumer(&kconsumer_data);
711 wait_consumer(&ustconsumer64_data);
712 wait_consumer(&ustconsumer32_data);
713
714 DBG("Cleaning up all agent apps");
715 agent_app_ht_clean();
716
717 DBG("Closing all UST sockets");
718 ust_app_clean_list();
719 buffer_reg_destroy_registries();
720
721 if (is_root && !opt_no_kernel) {
722 DBG2("Closing kernel fd");
723 if (kernel_tracer_fd >= 0) {
724 ret = close(kernel_tracer_fd);
725 if (ret) {
726 PERROR("close");
727 }
728 }
729 DBG("Unloading kernel modules");
730 modprobe_remove_lttng_all();
731 free(syscall_table);
732 }
733
734 close_consumer_sockets();
735
736 if (load_info) {
737 load_session_destroy_data(load_info);
738 free(load_info);
739 }
740
741 /*
742 * Cleanup lock file by deleting it and finaly closing it which will
743 * release the file system lock.
744 */
745 if (lockfile_fd >= 0) {
746 char lockfile_path[PATH_MAX];
747
748 ret = generate_lock_file_path(lockfile_path,
749 sizeof(lockfile_path));
750 if (ret > 0) {
751 ret = remove(lockfile_path);
752 if (ret < 0) {
753 PERROR("remove lock file");
754 }
755 ret = close(lockfile_fd);
756 if (ret < 0) {
757 PERROR("close lock file");
758 }
759 }
760 }
761
762 /*
763 * We do NOT rmdir rundir because there are other processes
764 * using it, for instance lttng-relayd, which can start in
765 * parallel with this teardown.
766 */
767
768 free(rundir);
769 }
770
771 /*
772 * Cleanup the daemon's option data structures.
773 */
774 static void sessiond_cleanup_options(void)
775 {
776 DBG("Cleaning up options");
777
778 /*
779 * If the override option is set, the pointer points to a *non* const
780 * thus freeing it even though the variable type is set to const.
781 */
782 if (tracing_group_name_override) {
783 free((void *) tracing_group_name);
784 }
785 if (consumerd32_bin_override) {
786 free((void *) consumerd32_bin);
787 }
788 if (consumerd64_bin_override) {
789 free((void *) consumerd64_bin);
790 }
791 if (consumerd32_libdir_override) {
792 free((void *) consumerd32_libdir);
793 }
794 if (consumerd64_libdir_override) {
795 free((void *) consumerd64_libdir);
796 }
797
798 free(opt_pidfile);
799 free(opt_load_session_path);
800 free(kmod_probes_list);
801 free(kmod_extra_probes_list);
802
803 run_as_destroy_worker();
804
805 /* <fun> */
806 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
807 "Matthew, BEET driven development works!%c[%dm",
808 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
809 /* </fun> */
810 }
811
812 /*
813 * Send data on a unix socket using the liblttsessiondcomm API.
814 *
815 * Return lttcomm error code.
816 */
817 static int send_unix_sock(int sock, void *buf, size_t len)
818 {
819 /* Check valid length */
820 if (len == 0) {
821 return -1;
822 }
823
824 return lttcomm_send_unix_sock(sock, buf, len);
825 }
826
827 /*
828 * Free memory of a command context structure.
829 */
830 static void clean_command_ctx(struct command_ctx **cmd_ctx)
831 {
832 DBG("Clean command context structure");
833 if (*cmd_ctx) {
834 if ((*cmd_ctx)->llm) {
835 free((*cmd_ctx)->llm);
836 }
837 if ((*cmd_ctx)->lsm) {
838 free((*cmd_ctx)->lsm);
839 }
840 free(*cmd_ctx);
841 *cmd_ctx = NULL;
842 }
843 }
844
845 /*
846 * Notify UST applications using the shm mmap futex.
847 */
848 static int notify_ust_apps(int active)
849 {
850 char *wait_shm_mmap;
851
852 DBG("Notifying applications of session daemon state: %d", active);
853
854 /* See shm.c for this call implying mmap, shm and futex calls */
855 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
856 if (wait_shm_mmap == NULL) {
857 goto error;
858 }
859
860 /* Wake waiting process */
861 futex_wait_update((int32_t *) wait_shm_mmap, active);
862
863 /* Apps notified successfully */
864 return 0;
865
866 error:
867 return -1;
868 }
869
870 /*
871 * Setup the outgoing data buffer for the response (llm) by allocating the
872 * right amount of memory and copying the original information from the lsm
873 * structure.
874 *
875 * Return 0 on success, negative value on error.
876 */
877 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
878 const void *payload_buf, size_t payload_len,
879 const void *cmd_header_buf, size_t cmd_header_len)
880 {
881 int ret = 0;
882 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
883 const size_t cmd_header_offset = header_len;
884 const size_t payload_offset = cmd_header_offset + cmd_header_len;
885 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
886
887 cmd_ctx->llm = zmalloc(total_msg_size);
888
889 if (cmd_ctx->llm == NULL) {
890 PERROR("zmalloc");
891 ret = -ENOMEM;
892 goto end;
893 }
894
895 /* Copy common data */
896 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
897 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
898 cmd_ctx->llm->cmd_header_size = cmd_header_len;
899 cmd_ctx->llm->data_size = payload_len;
900 cmd_ctx->lttng_msg_size = total_msg_size;
901
902 /* Copy command header */
903 if (cmd_header_len) {
904 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
905 cmd_header_len);
906 }
907
908 /* Copy payload */
909 if (payload_len) {
910 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
911 payload_len);
912 }
913
914 end:
915 return ret;
916 }
917
918 /*
919 * Version of setup_lttng_msg() without command header.
920 */
921 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
922 void *payload_buf, size_t payload_len)
923 {
924 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
925 }
926 /*
927 * Update the kernel poll set of all channel fd available over all tracing
928 * session. Add the wakeup pipe at the end of the set.
929 */
930 static int update_kernel_poll(struct lttng_poll_event *events)
931 {
932 int ret;
933 struct ltt_session *session;
934 struct ltt_kernel_channel *channel;
935
936 DBG("Updating kernel poll set");
937
938 session_lock_list();
939 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
940 session_lock(session);
941 if (session->kernel_session == NULL) {
942 session_unlock(session);
943 continue;
944 }
945
946 cds_list_for_each_entry(channel,
947 &session->kernel_session->channel_list.head, list) {
948 /* Add channel fd to the kernel poll set */
949 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
950 if (ret < 0) {
951 session_unlock(session);
952 goto error;
953 }
954 DBG("Channel fd %d added to kernel set", channel->fd);
955 }
956 session_unlock(session);
957 }
958 session_unlock_list();
959
960 return 0;
961
962 error:
963 session_unlock_list();
964 return -1;
965 }
966
967 /*
968 * Find the channel fd from 'fd' over all tracing session. When found, check
969 * for new channel stream and send those stream fds to the kernel consumer.
970 *
971 * Useful for CPU hotplug feature.
972 */
973 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
974 {
975 int ret = 0;
976 struct ltt_session *session;
977 struct ltt_kernel_session *ksess;
978 struct ltt_kernel_channel *channel;
979
980 DBG("Updating kernel streams for channel fd %d", fd);
981
982 session_lock_list();
983 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
984 session_lock(session);
985 if (session->kernel_session == NULL) {
986 session_unlock(session);
987 continue;
988 }
989 ksess = session->kernel_session;
990
991 cds_list_for_each_entry(channel,
992 &ksess->channel_list.head, list) {
993 struct lttng_ht_iter iter;
994 struct consumer_socket *socket;
995
996 if (channel->fd != fd) {
997 continue;
998 }
999 DBG("Channel found, updating kernel streams");
1000 ret = kernel_open_channel_stream(channel);
1001 if (ret < 0) {
1002 goto error;
1003 }
1004 /* Update the stream global counter */
1005 ksess->stream_count_global += ret;
1006
1007 /*
1008 * Have we already sent fds to the consumer? If yes, it
1009 * means that tracing is started so it is safe to send
1010 * our updated stream fds.
1011 */
1012 if (ksess->consumer_fds_sent != 1
1013 || ksess->consumer == NULL) {
1014 ret = -1;
1015 goto error;
1016 }
1017
1018 rcu_read_lock();
1019 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
1020 &iter.iter, socket, node.node) {
1021 pthread_mutex_lock(socket->lock);
1022 ret = kernel_consumer_send_channel_stream(socket,
1023 channel, ksess,
1024 session->output_traces ? 1 : 0);
1025 pthread_mutex_unlock(socket->lock);
1026 if (ret < 0) {
1027 rcu_read_unlock();
1028 goto error;
1029 }
1030 }
1031 rcu_read_unlock();
1032 }
1033 session_unlock(session);
1034 }
1035 session_unlock_list();
1036 return ret;
1037
1038 error:
1039 session_unlock(session);
1040 session_unlock_list();
1041 return ret;
1042 }
1043
1044 /*
1045 * For each tracing session, update newly registered apps. The session list
1046 * lock MUST be acquired before calling this.
1047 */
1048 static void update_ust_app(int app_sock)
1049 {
1050 struct ltt_session *sess, *stmp;
1051
1052 /* Consumer is in an ERROR state. Stop any application update. */
1053 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1054 /* Stop the update process since the consumer is dead. */
1055 return;
1056 }
1057
1058 /* For all tracing session(s) */
1059 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1060 struct ust_app *app;
1061
1062 session_lock(sess);
1063 if (!sess->ust_session) {
1064 goto unlock_session;
1065 }
1066
1067 rcu_read_lock();
1068 assert(app_sock >= 0);
1069 app = ust_app_find_by_sock(app_sock);
1070 if (app == NULL) {
1071 /*
1072 * Application can be unregistered before so
1073 * this is possible hence simply stopping the
1074 * update.
1075 */
1076 DBG3("UST app update failed to find app sock %d",
1077 app_sock);
1078 goto unlock_rcu;
1079 }
1080 ust_app_global_update(sess->ust_session, app);
1081 unlock_rcu:
1082 rcu_read_unlock();
1083 unlock_session:
1084 session_unlock(sess);
1085 }
1086 }
1087
1088 /*
1089 * This thread manage event coming from the kernel.
1090 *
1091 * Features supported in this thread:
1092 * -) CPU Hotplug
1093 */
1094 static void *thread_manage_kernel(void *data)
1095 {
1096 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1097 uint32_t revents, nb_fd;
1098 char tmp;
1099 struct lttng_poll_event events;
1100
1101 DBG("[thread] Thread manage kernel started");
1102
1103 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1104
1105 /*
1106 * This first step of the while is to clean this structure which could free
1107 * non NULL pointers so initialize it before the loop.
1108 */
1109 lttng_poll_init(&events);
1110
1111 if (testpoint(sessiond_thread_manage_kernel)) {
1112 goto error_testpoint;
1113 }
1114
1115 health_code_update();
1116
1117 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1118 goto error_testpoint;
1119 }
1120
1121 while (1) {
1122 health_code_update();
1123
1124 if (update_poll_flag == 1) {
1125 /* Clean events object. We are about to populate it again. */
1126 lttng_poll_clean(&events);
1127
1128 ret = sessiond_set_thread_pollset(&events, 2);
1129 if (ret < 0) {
1130 goto error_poll_create;
1131 }
1132
1133 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1134 if (ret < 0) {
1135 goto error;
1136 }
1137
1138 /* This will add the available kernel channel if any. */
1139 ret = update_kernel_poll(&events);
1140 if (ret < 0) {
1141 goto error;
1142 }
1143 update_poll_flag = 0;
1144 }
1145
1146 DBG("Thread kernel polling");
1147
1148 /* Poll infinite value of time */
1149 restart:
1150 health_poll_entry();
1151 ret = lttng_poll_wait(&events, -1);
1152 DBG("Thread kernel return from poll on %d fds",
1153 LTTNG_POLL_GETNB(&events));
1154 health_poll_exit();
1155 if (ret < 0) {
1156 /*
1157 * Restart interrupted system call.
1158 */
1159 if (errno == EINTR) {
1160 goto restart;
1161 }
1162 goto error;
1163 } else if (ret == 0) {
1164 /* Should not happen since timeout is infinite */
1165 ERR("Return value of poll is 0 with an infinite timeout.\n"
1166 "This should not have happened! Continuing...");
1167 continue;
1168 }
1169
1170 nb_fd = ret;
1171
1172 for (i = 0; i < nb_fd; i++) {
1173 /* Fetch once the poll data */
1174 revents = LTTNG_POLL_GETEV(&events, i);
1175 pollfd = LTTNG_POLL_GETFD(&events, i);
1176
1177 health_code_update();
1178
1179 if (!revents) {
1180 /* No activity for this FD (poll implementation). */
1181 continue;
1182 }
1183
1184 /* Thread quit pipe has been closed. Killing thread. */
1185 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1186 if (ret) {
1187 err = 0;
1188 goto exit;
1189 }
1190
1191 /* Check for data on kernel pipe */
1192 if (revents & LPOLLIN) {
1193 if (pollfd == kernel_poll_pipe[0]) {
1194 (void) lttng_read(kernel_poll_pipe[0],
1195 &tmp, 1);
1196 /*
1197 * Ret value is useless here, if this pipe gets any actions an
1198 * update is required anyway.
1199 */
1200 update_poll_flag = 1;
1201 continue;
1202 } else {
1203 /*
1204 * New CPU detected by the kernel. Adding kernel stream to
1205 * kernel session and updating the kernel consumer
1206 */
1207 ret = update_kernel_stream(&kconsumer_data, pollfd);
1208 if (ret < 0) {
1209 continue;
1210 }
1211 break;
1212 }
1213 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1214 update_poll_flag = 1;
1215 continue;
1216 } else {
1217 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1218 goto error;
1219 }
1220 }
1221 }
1222
1223 exit:
1224 error:
1225 lttng_poll_clean(&events);
1226 error_poll_create:
1227 error_testpoint:
1228 utils_close_pipe(kernel_poll_pipe);
1229 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1230 if (err) {
1231 health_error();
1232 ERR("Health error occurred in %s", __func__);
1233 WARN("Kernel thread died unexpectedly. "
1234 "Kernel tracing can continue but CPU hotplug is disabled.");
1235 }
1236 health_unregister(health_sessiond);
1237 DBG("Kernel thread dying");
1238 return NULL;
1239 }
1240
1241 /*
1242 * Signal pthread condition of the consumer data that the thread.
1243 */
1244 static void signal_consumer_condition(struct consumer_data *data, int state)
1245 {
1246 pthread_mutex_lock(&data->cond_mutex);
1247
1248 /*
1249 * The state is set before signaling. It can be any value, it's the waiter
1250 * job to correctly interpret this condition variable associated to the
1251 * consumer pthread_cond.
1252 *
1253 * A value of 0 means that the corresponding thread of the consumer data
1254 * was not started. 1 indicates that the thread has started and is ready
1255 * for action. A negative value means that there was an error during the
1256 * thread bootstrap.
1257 */
1258 data->consumer_thread_is_ready = state;
1259 (void) pthread_cond_signal(&data->cond);
1260
1261 pthread_mutex_unlock(&data->cond_mutex);
1262 }
1263
1264 /*
1265 * This thread manage the consumer error sent back to the session daemon.
1266 */
1267 static void *thread_manage_consumer(void *data)
1268 {
1269 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1270 uint32_t revents, nb_fd;
1271 enum lttcomm_return_code code;
1272 struct lttng_poll_event events;
1273 struct consumer_data *consumer_data = data;
1274
1275 DBG("[thread] Manage consumer started");
1276
1277 rcu_register_thread();
1278 rcu_thread_online();
1279
1280 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1281
1282 health_code_update();
1283
1284 /*
1285 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1286 * metadata_sock. Nothing more will be added to this poll set.
1287 */
1288 ret = sessiond_set_thread_pollset(&events, 3);
1289 if (ret < 0) {
1290 goto error_poll;
1291 }
1292
1293 /*
1294 * The error socket here is already in a listening state which was done
1295 * just before spawning this thread to avoid a race between the consumer
1296 * daemon exec trying to connect and the listen() call.
1297 */
1298 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1299 if (ret < 0) {
1300 goto error;
1301 }
1302
1303 health_code_update();
1304
1305 /* Infinite blocking call, waiting for transmission */
1306 restart:
1307 health_poll_entry();
1308
1309 if (testpoint(sessiond_thread_manage_consumer)) {
1310 goto error;
1311 }
1312
1313 ret = lttng_poll_wait(&events, -1);
1314 health_poll_exit();
1315 if (ret < 0) {
1316 /*
1317 * Restart interrupted system call.
1318 */
1319 if (errno == EINTR) {
1320 goto restart;
1321 }
1322 goto error;
1323 }
1324
1325 nb_fd = ret;
1326
1327 for (i = 0; i < nb_fd; i++) {
1328 /* Fetch once the poll data */
1329 revents = LTTNG_POLL_GETEV(&events, i);
1330 pollfd = LTTNG_POLL_GETFD(&events, i);
1331
1332 health_code_update();
1333
1334 if (!revents) {
1335 /* No activity for this FD (poll implementation). */
1336 continue;
1337 }
1338
1339 /* Thread quit pipe has been closed. Killing thread. */
1340 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1341 if (ret) {
1342 err = 0;
1343 goto exit;
1344 }
1345
1346 /* Event on the registration socket */
1347 if (pollfd == consumer_data->err_sock) {
1348 if (revents & LPOLLIN) {
1349 continue;
1350 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1351 ERR("consumer err socket poll error");
1352 goto error;
1353 } else {
1354 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1355 goto error;
1356 }
1357 }
1358 }
1359
1360 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1361 if (sock < 0) {
1362 goto error;
1363 }
1364
1365 /*
1366 * Set the CLOEXEC flag. Return code is useless because either way, the
1367 * show must go on.
1368 */
1369 (void) utils_set_fd_cloexec(sock);
1370
1371 health_code_update();
1372
1373 DBG2("Receiving code from consumer err_sock");
1374
1375 /* Getting status code from kconsumerd */
1376 ret = lttcomm_recv_unix_sock(sock, &code,
1377 sizeof(enum lttcomm_return_code));
1378 if (ret <= 0) {
1379 goto error;
1380 }
1381
1382 health_code_update();
1383 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1384 /* Connect both socket, command and metadata. */
1385 consumer_data->cmd_sock =
1386 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1387 consumer_data->metadata_fd =
1388 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1389 if (consumer_data->cmd_sock < 0
1390 || consumer_data->metadata_fd < 0) {
1391 PERROR("consumer connect cmd socket");
1392 /* On error, signal condition and quit. */
1393 signal_consumer_condition(consumer_data, -1);
1394 goto error;
1395 }
1396 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1397 /* Create metadata socket lock. */
1398 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1399 if (consumer_data->metadata_sock.lock == NULL) {
1400 PERROR("zmalloc pthread mutex");
1401 ret = -1;
1402 goto error;
1403 }
1404 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1405
1406 signal_consumer_condition(consumer_data, 1);
1407 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1408 DBG("Consumer metadata socket ready (fd: %d)",
1409 consumer_data->metadata_fd);
1410 } else {
1411 ERR("consumer error when waiting for SOCK_READY : %s",
1412 lttcomm_get_readable_code(-code));
1413 goto error;
1414 }
1415
1416 /* Remove the consumerd error sock since we've established a connexion */
1417 ret = lttng_poll_del(&events, consumer_data->err_sock);
1418 if (ret < 0) {
1419 goto error;
1420 }
1421
1422 /* Add new accepted error socket. */
1423 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1424 if (ret < 0) {
1425 goto error;
1426 }
1427
1428 /* Add metadata socket that is successfully connected. */
1429 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1430 LPOLLIN | LPOLLRDHUP);
1431 if (ret < 0) {
1432 goto error;
1433 }
1434
1435 health_code_update();
1436
1437 /* Infinite blocking call, waiting for transmission */
1438 restart_poll:
1439 while (1) {
1440 health_code_update();
1441
1442 /* Exit the thread because the thread quit pipe has been triggered. */
1443 if (should_quit) {
1444 /* Not a health error. */
1445 err = 0;
1446 goto exit;
1447 }
1448
1449 health_poll_entry();
1450 ret = lttng_poll_wait(&events, -1);
1451 health_poll_exit();
1452 if (ret < 0) {
1453 /*
1454 * Restart interrupted system call.
1455 */
1456 if (errno == EINTR) {
1457 goto restart_poll;
1458 }
1459 goto error;
1460 }
1461
1462 nb_fd = ret;
1463
1464 for (i = 0; i < nb_fd; i++) {
1465 /* Fetch once the poll data */
1466 revents = LTTNG_POLL_GETEV(&events, i);
1467 pollfd = LTTNG_POLL_GETFD(&events, i);
1468
1469 health_code_update();
1470
1471 if (!revents) {
1472 /* No activity for this FD (poll implementation). */
1473 continue;
1474 }
1475
1476 /*
1477 * Thread quit pipe has been triggered, flag that we should stop
1478 * but continue the current loop to handle potential data from
1479 * consumer.
1480 */
1481 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1482
1483 if (pollfd == sock) {
1484 /* Event on the consumerd socket */
1485 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1486 && !(revents & LPOLLIN)) {
1487 ERR("consumer err socket second poll error");
1488 goto error;
1489 }
1490 health_code_update();
1491 /* Wait for any kconsumerd error */
1492 ret = lttcomm_recv_unix_sock(sock, &code,
1493 sizeof(enum lttcomm_return_code));
1494 if (ret <= 0) {
1495 ERR("consumer closed the command socket");
1496 goto error;
1497 }
1498
1499 ERR("consumer return code : %s",
1500 lttcomm_get_readable_code(-code));
1501
1502 goto exit;
1503 } else if (pollfd == consumer_data->metadata_fd) {
1504 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1505 && !(revents & LPOLLIN)) {
1506 ERR("consumer err metadata socket second poll error");
1507 goto error;
1508 }
1509 /* UST metadata requests */
1510 ret = ust_consumer_metadata_request(
1511 &consumer_data->metadata_sock);
1512 if (ret < 0) {
1513 ERR("Handling metadata request");
1514 goto error;
1515 }
1516 }
1517 /* No need for an else branch all FDs are tested prior. */
1518 }
1519 health_code_update();
1520 }
1521
1522 exit:
1523 error:
1524 /*
1525 * We lock here because we are about to close the sockets and some other
1526 * thread might be using them so get exclusive access which will abort all
1527 * other consumer command by other threads.
1528 */
1529 pthread_mutex_lock(&consumer_data->lock);
1530
1531 /* Immediately set the consumerd state to stopped */
1532 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1533 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1534 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1535 consumer_data->type == LTTNG_CONSUMER32_UST) {
1536 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1537 } else {
1538 /* Code flow error... */
1539 assert(0);
1540 }
1541
1542 if (consumer_data->err_sock >= 0) {
1543 ret = close(consumer_data->err_sock);
1544 if (ret) {
1545 PERROR("close");
1546 }
1547 consumer_data->err_sock = -1;
1548 }
1549 if (consumer_data->cmd_sock >= 0) {
1550 ret = close(consumer_data->cmd_sock);
1551 if (ret) {
1552 PERROR("close");
1553 }
1554 consumer_data->cmd_sock = -1;
1555 }
1556 if (consumer_data->metadata_sock.fd_ptr &&
1557 *consumer_data->metadata_sock.fd_ptr >= 0) {
1558 ret = close(*consumer_data->metadata_sock.fd_ptr);
1559 if (ret) {
1560 PERROR("close");
1561 }
1562 }
1563 if (sock >= 0) {
1564 ret = close(sock);
1565 if (ret) {
1566 PERROR("close");
1567 }
1568 }
1569
1570 unlink(consumer_data->err_unix_sock_path);
1571 unlink(consumer_data->cmd_unix_sock_path);
1572 pthread_mutex_unlock(&consumer_data->lock);
1573
1574 /* Cleanup metadata socket mutex. */
1575 if (consumer_data->metadata_sock.lock) {
1576 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1577 free(consumer_data->metadata_sock.lock);
1578 }
1579 lttng_poll_clean(&events);
1580 error_poll:
1581 if (err) {
1582 health_error();
1583 ERR("Health error occurred in %s", __func__);
1584 }
1585 health_unregister(health_sessiond);
1586 DBG("consumer thread cleanup completed");
1587
1588 rcu_thread_offline();
1589 rcu_unregister_thread();
1590
1591 return NULL;
1592 }
1593
1594 /*
1595 * This thread manage application communication.
1596 */
1597 static void *thread_manage_apps(void *data)
1598 {
1599 int i, ret, pollfd, err = -1;
1600 ssize_t size_ret;
1601 uint32_t revents, nb_fd;
1602 struct lttng_poll_event events;
1603
1604 DBG("[thread] Manage application started");
1605
1606 rcu_register_thread();
1607 rcu_thread_online();
1608
1609 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1610
1611 if (testpoint(sessiond_thread_manage_apps)) {
1612 goto error_testpoint;
1613 }
1614
1615 health_code_update();
1616
1617 ret = sessiond_set_thread_pollset(&events, 2);
1618 if (ret < 0) {
1619 goto error_poll_create;
1620 }
1621
1622 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1623 if (ret < 0) {
1624 goto error;
1625 }
1626
1627 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1628 goto error;
1629 }
1630
1631 health_code_update();
1632
1633 while (1) {
1634 DBG("Apps thread polling");
1635
1636 /* Inifinite blocking call, waiting for transmission */
1637 restart:
1638 health_poll_entry();
1639 ret = lttng_poll_wait(&events, -1);
1640 DBG("Apps thread return from poll on %d fds",
1641 LTTNG_POLL_GETNB(&events));
1642 health_poll_exit();
1643 if (ret < 0) {
1644 /*
1645 * Restart interrupted system call.
1646 */
1647 if (errno == EINTR) {
1648 goto restart;
1649 }
1650 goto error;
1651 }
1652
1653 nb_fd = ret;
1654
1655 for (i = 0; i < nb_fd; i++) {
1656 /* Fetch once the poll data */
1657 revents = LTTNG_POLL_GETEV(&events, i);
1658 pollfd = LTTNG_POLL_GETFD(&events, i);
1659
1660 health_code_update();
1661
1662 if (!revents) {
1663 /* No activity for this FD (poll implementation). */
1664 continue;
1665 }
1666
1667 /* Thread quit pipe has been closed. Killing thread. */
1668 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1669 if (ret) {
1670 err = 0;
1671 goto exit;
1672 }
1673
1674 /* Inspect the apps cmd pipe */
1675 if (pollfd == apps_cmd_pipe[0]) {
1676 if (revents & LPOLLIN) {
1677 int sock;
1678
1679 /* Empty pipe */
1680 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1681 if (size_ret < sizeof(sock)) {
1682 PERROR("read apps cmd pipe");
1683 goto error;
1684 }
1685
1686 health_code_update();
1687
1688 /*
1689 * Since this is a command socket (write then read),
1690 * we only monitor the error events of the socket.
1691 */
1692 ret = lttng_poll_add(&events, sock,
1693 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1694 if (ret < 0) {
1695 goto error;
1696 }
1697
1698 DBG("Apps with sock %d added to poll set", sock);
1699 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1700 ERR("Apps command pipe error");
1701 goto error;
1702 } else {
1703 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1704 goto error;
1705 }
1706 } else {
1707 /*
1708 * At this point, we know that a registered application made
1709 * the event at poll_wait.
1710 */
1711 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1712 /* Removing from the poll set */
1713 ret = lttng_poll_del(&events, pollfd);
1714 if (ret < 0) {
1715 goto error;
1716 }
1717
1718 /* Socket closed on remote end. */
1719 ust_app_unregister(pollfd);
1720 } else {
1721 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1722 goto error;
1723 }
1724 }
1725
1726 health_code_update();
1727 }
1728 }
1729
1730 exit:
1731 error:
1732 lttng_poll_clean(&events);
1733 error_poll_create:
1734 error_testpoint:
1735 utils_close_pipe(apps_cmd_pipe);
1736 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1737
1738 /*
1739 * We don't clean the UST app hash table here since already registered
1740 * applications can still be controlled so let them be until the session
1741 * daemon dies or the applications stop.
1742 */
1743
1744 if (err) {
1745 health_error();
1746 ERR("Health error occurred in %s", __func__);
1747 }
1748 health_unregister(health_sessiond);
1749 DBG("Application communication apps thread cleanup complete");
1750 rcu_thread_offline();
1751 rcu_unregister_thread();
1752 return NULL;
1753 }
1754
1755 /*
1756 * Send a socket to a thread This is called from the dispatch UST registration
1757 * thread once all sockets are set for the application.
1758 *
1759 * The sock value can be invalid, we don't really care, the thread will handle
1760 * it and make the necessary cleanup if so.
1761 *
1762 * On success, return 0 else a negative value being the errno message of the
1763 * write().
1764 */
1765 static int send_socket_to_thread(int fd, int sock)
1766 {
1767 ssize_t ret;
1768
1769 /*
1770 * It's possible that the FD is set as invalid with -1 concurrently just
1771 * before calling this function being a shutdown state of the thread.
1772 */
1773 if (fd < 0) {
1774 ret = -EBADF;
1775 goto error;
1776 }
1777
1778 ret = lttng_write(fd, &sock, sizeof(sock));
1779 if (ret < sizeof(sock)) {
1780 PERROR("write apps pipe %d", fd);
1781 if (ret < 0) {
1782 ret = -errno;
1783 }
1784 goto error;
1785 }
1786
1787 /* All good. Don't send back the write positive ret value. */
1788 ret = 0;
1789 error:
1790 return (int) ret;
1791 }
1792
1793 /*
1794 * Sanitize the wait queue of the dispatch registration thread meaning removing
1795 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1796 * notify socket is never received.
1797 */
1798 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1799 {
1800 int ret, nb_fd = 0, i;
1801 unsigned int fd_added = 0;
1802 struct lttng_poll_event events;
1803 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1804
1805 assert(wait_queue);
1806
1807 lttng_poll_init(&events);
1808
1809 /* Just skip everything for an empty queue. */
1810 if (!wait_queue->count) {
1811 goto end;
1812 }
1813
1814 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1815 if (ret < 0) {
1816 goto error_create;
1817 }
1818
1819 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1820 &wait_queue->head, head) {
1821 assert(wait_node->app);
1822 ret = lttng_poll_add(&events, wait_node->app->sock,
1823 LPOLLHUP | LPOLLERR);
1824 if (ret < 0) {
1825 goto error;
1826 }
1827
1828 fd_added = 1;
1829 }
1830
1831 if (!fd_added) {
1832 goto end;
1833 }
1834
1835 /*
1836 * Poll but don't block so we can quickly identify the faulty events and
1837 * clean them afterwards from the wait queue.
1838 */
1839 ret = lttng_poll_wait(&events, 0);
1840 if (ret < 0) {
1841 goto error;
1842 }
1843 nb_fd = ret;
1844
1845 for (i = 0; i < nb_fd; i++) {
1846 /* Get faulty FD. */
1847 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1848 int pollfd = LTTNG_POLL_GETFD(&events, i);
1849
1850 if (!revents) {
1851 /* No activity for this FD (poll implementation). */
1852 continue;
1853 }
1854
1855 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1856 &wait_queue->head, head) {
1857 if (pollfd == wait_node->app->sock &&
1858 (revents & (LPOLLHUP | LPOLLERR))) {
1859 cds_list_del(&wait_node->head);
1860 wait_queue->count--;
1861 ust_app_destroy(wait_node->app);
1862 free(wait_node);
1863 /*
1864 * Silence warning of use-after-free in
1865 * cds_list_for_each_entry_safe which uses
1866 * __typeof__(*wait_node).
1867 */
1868 wait_node = NULL;
1869 break;
1870 } else {
1871 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1872 goto error;
1873 }
1874 }
1875 }
1876
1877 if (nb_fd > 0) {
1878 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1879 }
1880
1881 end:
1882 lttng_poll_clean(&events);
1883 return;
1884
1885 error:
1886 lttng_poll_clean(&events);
1887 error_create:
1888 ERR("Unable to sanitize wait queue");
1889 return;
1890 }
1891
1892 /*
1893 * Dispatch request from the registration threads to the application
1894 * communication thread.
1895 */
1896 static void *thread_dispatch_ust_registration(void *data)
1897 {
1898 int ret, err = -1;
1899 struct cds_wfcq_node *node;
1900 struct ust_command *ust_cmd = NULL;
1901 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1902 struct ust_reg_wait_queue wait_queue = {
1903 .count = 0,
1904 };
1905
1906 rcu_register_thread();
1907
1908 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1909
1910 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1911 goto error_testpoint;
1912 }
1913
1914 health_code_update();
1915
1916 CDS_INIT_LIST_HEAD(&wait_queue.head);
1917
1918 DBG("[thread] Dispatch UST command started");
1919
1920 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1921 health_code_update();
1922
1923 /* Atomically prepare the queue futex */
1924 futex_nto1_prepare(&ust_cmd_queue.futex);
1925
1926 do {
1927 struct ust_app *app = NULL;
1928 ust_cmd = NULL;
1929
1930 /*
1931 * Make sure we don't have node(s) that have hung up before receiving
1932 * the notify socket. This is to clean the list in order to avoid
1933 * memory leaks from notify socket that are never seen.
1934 */
1935 sanitize_wait_queue(&wait_queue);
1936
1937 health_code_update();
1938 /* Dequeue command for registration */
1939 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1940 if (node == NULL) {
1941 DBG("Woken up but nothing in the UST command queue");
1942 /* Continue thread execution */
1943 break;
1944 }
1945
1946 ust_cmd = caa_container_of(node, struct ust_command, node);
1947
1948 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1949 " gid:%d sock:%d name:%s (version %d.%d)",
1950 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1951 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1952 ust_cmd->sock, ust_cmd->reg_msg.name,
1953 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1954
1955 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1956 wait_node = zmalloc(sizeof(*wait_node));
1957 if (!wait_node) {
1958 PERROR("zmalloc wait_node dispatch");
1959 ret = close(ust_cmd->sock);
1960 if (ret < 0) {
1961 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1962 }
1963 lttng_fd_put(LTTNG_FD_APPS, 1);
1964 free(ust_cmd);
1965 goto error;
1966 }
1967 CDS_INIT_LIST_HEAD(&wait_node->head);
1968
1969 /* Create application object if socket is CMD. */
1970 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1971 ust_cmd->sock);
1972 if (!wait_node->app) {
1973 ret = close(ust_cmd->sock);
1974 if (ret < 0) {
1975 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1976 }
1977 lttng_fd_put(LTTNG_FD_APPS, 1);
1978 free(wait_node);
1979 free(ust_cmd);
1980 continue;
1981 }
1982 /*
1983 * Add application to the wait queue so we can set the notify
1984 * socket before putting this object in the global ht.
1985 */
1986 cds_list_add(&wait_node->head, &wait_queue.head);
1987 wait_queue.count++;
1988
1989 free(ust_cmd);
1990 /*
1991 * We have to continue here since we don't have the notify
1992 * socket and the application MUST be added to the hash table
1993 * only at that moment.
1994 */
1995 continue;
1996 } else {
1997 /*
1998 * Look for the application in the local wait queue and set the
1999 * notify socket if found.
2000 */
2001 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2002 &wait_queue.head, head) {
2003 health_code_update();
2004 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
2005 wait_node->app->notify_sock = ust_cmd->sock;
2006 cds_list_del(&wait_node->head);
2007 wait_queue.count--;
2008 app = wait_node->app;
2009 free(wait_node);
2010 DBG3("UST app notify socket %d is set", ust_cmd->sock);
2011 break;
2012 }
2013 }
2014
2015 /*
2016 * With no application at this stage the received socket is
2017 * basically useless so close it before we free the cmd data
2018 * structure for good.
2019 */
2020 if (!app) {
2021 ret = close(ust_cmd->sock);
2022 if (ret < 0) {
2023 PERROR("close ust sock dispatch %d", ust_cmd->sock);
2024 }
2025 lttng_fd_put(LTTNG_FD_APPS, 1);
2026 }
2027 free(ust_cmd);
2028 }
2029
2030 if (app) {
2031 /*
2032 * @session_lock_list
2033 *
2034 * Lock the global session list so from the register up to the
2035 * registration done message, no thread can see the application
2036 * and change its state.
2037 */
2038 session_lock_list();
2039 rcu_read_lock();
2040
2041 /*
2042 * Add application to the global hash table. This needs to be
2043 * done before the update to the UST registry can locate the
2044 * application.
2045 */
2046 ust_app_add(app);
2047
2048 /* Set app version. This call will print an error if needed. */
2049 (void) ust_app_version(app);
2050
2051 /* Send notify socket through the notify pipe. */
2052 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2053 app->notify_sock);
2054 if (ret < 0) {
2055 rcu_read_unlock();
2056 session_unlock_list();
2057 /*
2058 * No notify thread, stop the UST tracing. However, this is
2059 * not an internal error of the this thread thus setting
2060 * the health error code to a normal exit.
2061 */
2062 err = 0;
2063 goto error;
2064 }
2065
2066 /*
2067 * Update newly registered application with the tracing
2068 * registry info already enabled information.
2069 */
2070 update_ust_app(app->sock);
2071
2072 /*
2073 * Don't care about return value. Let the manage apps threads
2074 * handle app unregistration upon socket close.
2075 */
2076 (void) ust_app_register_done(app);
2077
2078 /*
2079 * Even if the application socket has been closed, send the app
2080 * to the thread and unregistration will take place at that
2081 * place.
2082 */
2083 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2084 if (ret < 0) {
2085 rcu_read_unlock();
2086 session_unlock_list();
2087 /*
2088 * No apps. thread, stop the UST tracing. However, this is
2089 * not an internal error of the this thread thus setting
2090 * the health error code to a normal exit.
2091 */
2092 err = 0;
2093 goto error;
2094 }
2095
2096 rcu_read_unlock();
2097 session_unlock_list();
2098 }
2099 } while (node != NULL);
2100
2101 health_poll_entry();
2102 /* Futex wait on queue. Blocking call on futex() */
2103 futex_nto1_wait(&ust_cmd_queue.futex);
2104 health_poll_exit();
2105 }
2106 /* Normal exit, no error */
2107 err = 0;
2108
2109 error:
2110 /* Clean up wait queue. */
2111 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2112 &wait_queue.head, head) {
2113 cds_list_del(&wait_node->head);
2114 wait_queue.count--;
2115 free(wait_node);
2116 }
2117
2118 /* Empty command queue. */
2119 for (;;) {
2120 /* Dequeue command for registration */
2121 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2122 if (node == NULL) {
2123 break;
2124 }
2125 ust_cmd = caa_container_of(node, struct ust_command, node);
2126 ret = close(ust_cmd->sock);
2127 if (ret < 0) {
2128 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2129 }
2130 lttng_fd_put(LTTNG_FD_APPS, 1);
2131 free(ust_cmd);
2132 }
2133
2134 error_testpoint:
2135 DBG("Dispatch thread dying");
2136 if (err) {
2137 health_error();
2138 ERR("Health error occurred in %s", __func__);
2139 }
2140 health_unregister(health_sessiond);
2141 rcu_unregister_thread();
2142 return NULL;
2143 }
2144
2145 /*
2146 * This thread manage application registration.
2147 */
2148 static void *thread_registration_apps(void *data)
2149 {
2150 int sock = -1, i, ret, pollfd, err = -1;
2151 uint32_t revents, nb_fd;
2152 struct lttng_poll_event events;
2153 /*
2154 * Get allocated in this thread, enqueued to a global queue, dequeued and
2155 * freed in the manage apps thread.
2156 */
2157 struct ust_command *ust_cmd = NULL;
2158
2159 DBG("[thread] Manage application registration started");
2160
2161 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2162
2163 if (testpoint(sessiond_thread_registration_apps)) {
2164 goto error_testpoint;
2165 }
2166
2167 ret = lttcomm_listen_unix_sock(apps_sock);
2168 if (ret < 0) {
2169 goto error_listen;
2170 }
2171
2172 /*
2173 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2174 * more will be added to this poll set.
2175 */
2176 ret = sessiond_set_thread_pollset(&events, 2);
2177 if (ret < 0) {
2178 goto error_create_poll;
2179 }
2180
2181 /* Add the application registration socket */
2182 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2183 if (ret < 0) {
2184 goto error_poll_add;
2185 }
2186
2187 /* Notify all applications to register */
2188 ret = notify_ust_apps(1);
2189 if (ret < 0) {
2190 ERR("Failed to notify applications or create the wait shared memory.\n"
2191 "Execution continues but there might be problem for already\n"
2192 "running applications that wishes to register.");
2193 }
2194
2195 while (1) {
2196 DBG("Accepting application registration");
2197
2198 /* Inifinite blocking call, waiting for transmission */
2199 restart:
2200 health_poll_entry();
2201 ret = lttng_poll_wait(&events, -1);
2202 health_poll_exit();
2203 if (ret < 0) {
2204 /*
2205 * Restart interrupted system call.
2206 */
2207 if (errno == EINTR) {
2208 goto restart;
2209 }
2210 goto error;
2211 }
2212
2213 nb_fd = ret;
2214
2215 for (i = 0; i < nb_fd; i++) {
2216 health_code_update();
2217
2218 /* Fetch once the poll data */
2219 revents = LTTNG_POLL_GETEV(&events, i);
2220 pollfd = LTTNG_POLL_GETFD(&events, i);
2221
2222 if (!revents) {
2223 /* No activity for this FD (poll implementation). */
2224 continue;
2225 }
2226
2227 /* Thread quit pipe has been closed. Killing thread. */
2228 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2229 if (ret) {
2230 err = 0;
2231 goto exit;
2232 }
2233
2234 /* Event on the registration socket */
2235 if (pollfd == apps_sock) {
2236 if (revents & LPOLLIN) {
2237 sock = lttcomm_accept_unix_sock(apps_sock);
2238 if (sock < 0) {
2239 goto error;
2240 }
2241
2242 /*
2243 * Set socket timeout for both receiving and ending.
2244 * app_socket_timeout is in seconds, whereas
2245 * lttcomm_setsockopt_rcv_timeout and
2246 * lttcomm_setsockopt_snd_timeout expect msec as
2247 * parameter.
2248 */
2249 if (app_socket_timeout >= 0) {
2250 (void) lttcomm_setsockopt_rcv_timeout(sock,
2251 app_socket_timeout * 1000);
2252 (void) lttcomm_setsockopt_snd_timeout(sock,
2253 app_socket_timeout * 1000);
2254 }
2255
2256 /*
2257 * Set the CLOEXEC flag. Return code is useless because
2258 * either way, the show must go on.
2259 */
2260 (void) utils_set_fd_cloexec(sock);
2261
2262 /* Create UST registration command for enqueuing */
2263 ust_cmd = zmalloc(sizeof(struct ust_command));
2264 if (ust_cmd == NULL) {
2265 PERROR("ust command zmalloc");
2266 ret = close(sock);
2267 if (ret) {
2268 PERROR("close");
2269 }
2270 goto error;
2271 }
2272
2273 /*
2274 * Using message-based transmissions to ensure we don't
2275 * have to deal with partially received messages.
2276 */
2277 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2278 if (ret < 0) {
2279 ERR("Exhausted file descriptors allowed for applications.");
2280 free(ust_cmd);
2281 ret = close(sock);
2282 if (ret) {
2283 PERROR("close");
2284 }
2285 sock = -1;
2286 continue;
2287 }
2288
2289 health_code_update();
2290 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2291 if (ret < 0) {
2292 free(ust_cmd);
2293 /* Close socket of the application. */
2294 ret = close(sock);
2295 if (ret) {
2296 PERROR("close");
2297 }
2298 lttng_fd_put(LTTNG_FD_APPS, 1);
2299 sock = -1;
2300 continue;
2301 }
2302 health_code_update();
2303
2304 ust_cmd->sock = sock;
2305 sock = -1;
2306
2307 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2308 " gid:%d sock:%d name:%s (version %d.%d)",
2309 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2310 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2311 ust_cmd->sock, ust_cmd->reg_msg.name,
2312 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2313
2314 /*
2315 * Lock free enqueue the registration request. The red pill
2316 * has been taken! This apps will be part of the *system*.
2317 */
2318 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2319
2320 /*
2321 * Wake the registration queue futex. Implicit memory
2322 * barrier with the exchange in cds_wfcq_enqueue.
2323 */
2324 futex_nto1_wake(&ust_cmd_queue.futex);
2325 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2326 ERR("Register apps socket poll error");
2327 goto error;
2328 } else {
2329 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2330 goto error;
2331 }
2332 }
2333 }
2334 }
2335
2336 exit:
2337 error:
2338 /* Notify that the registration thread is gone */
2339 notify_ust_apps(0);
2340
2341 if (apps_sock >= 0) {
2342 ret = close(apps_sock);
2343 if (ret) {
2344 PERROR("close");
2345 }
2346 }
2347 if (sock >= 0) {
2348 ret = close(sock);
2349 if (ret) {
2350 PERROR("close");
2351 }
2352 lttng_fd_put(LTTNG_FD_APPS, 1);
2353 }
2354 unlink(apps_unix_sock_path);
2355
2356 error_poll_add:
2357 lttng_poll_clean(&events);
2358 error_listen:
2359 error_create_poll:
2360 error_testpoint:
2361 DBG("UST Registration thread cleanup complete");
2362 if (err) {
2363 health_error();
2364 ERR("Health error occurred in %s", __func__);
2365 }
2366 health_unregister(health_sessiond);
2367
2368 return NULL;
2369 }
2370
2371 /*
2372 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2373 * exec or it will fails.
2374 */
2375 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2376 {
2377 int ret, clock_ret;
2378 struct timespec timeout;
2379
2380 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2381 consumer_data->consumer_thread_is_ready = 0;
2382
2383 /* Setup pthread condition */
2384 ret = pthread_condattr_init(&consumer_data->condattr);
2385 if (ret) {
2386 errno = ret;
2387 PERROR("pthread_condattr_init consumer data");
2388 goto error;
2389 }
2390
2391 /*
2392 * Set the monotonic clock in order to make sure we DO NOT jump in time
2393 * between the clock_gettime() call and the timedwait call. See bug #324
2394 * for a more details and how we noticed it.
2395 */
2396 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2397 if (ret) {
2398 errno = ret;
2399 PERROR("pthread_condattr_setclock consumer data");
2400 goto error;
2401 }
2402
2403 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2404 if (ret) {
2405 errno = ret;
2406 PERROR("pthread_cond_init consumer data");
2407 goto error;
2408 }
2409
2410 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2411 consumer_data);
2412 if (ret) {
2413 errno = ret;
2414 PERROR("pthread_create consumer");
2415 ret = -1;
2416 goto error;
2417 }
2418
2419 /* We are about to wait on a pthread condition */
2420 pthread_mutex_lock(&consumer_data->cond_mutex);
2421
2422 /* Get time for sem_timedwait absolute timeout */
2423 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2424 /*
2425 * Set the timeout for the condition timed wait even if the clock gettime
2426 * call fails since we might loop on that call and we want to avoid to
2427 * increment the timeout too many times.
2428 */
2429 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2430
2431 /*
2432 * The following loop COULD be skipped in some conditions so this is why we
2433 * set ret to 0 in order to make sure at least one round of the loop is
2434 * done.
2435 */
2436 ret = 0;
2437
2438 /*
2439 * Loop until the condition is reached or when a timeout is reached. Note
2440 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2441 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2442 * possible. This loop does not take any chances and works with both of
2443 * them.
2444 */
2445 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2446 if (clock_ret < 0) {
2447 PERROR("clock_gettime spawn consumer");
2448 /* Infinite wait for the consumerd thread to be ready */
2449 ret = pthread_cond_wait(&consumer_data->cond,
2450 &consumer_data->cond_mutex);
2451 } else {
2452 ret = pthread_cond_timedwait(&consumer_data->cond,
2453 &consumer_data->cond_mutex, &timeout);
2454 }
2455 }
2456
2457 /* Release the pthread condition */
2458 pthread_mutex_unlock(&consumer_data->cond_mutex);
2459
2460 if (ret != 0) {
2461 errno = ret;
2462 if (ret == ETIMEDOUT) {
2463 int pth_ret;
2464
2465 /*
2466 * Call has timed out so we kill the kconsumerd_thread and return
2467 * an error.
2468 */
2469 ERR("Condition timed out. The consumer thread was never ready."
2470 " Killing it");
2471 pth_ret = pthread_cancel(consumer_data->thread);
2472 if (pth_ret < 0) {
2473 PERROR("pthread_cancel consumer thread");
2474 }
2475 } else {
2476 PERROR("pthread_cond_wait failed consumer thread");
2477 }
2478 /* Caller is expecting a negative value on failure. */
2479 ret = -1;
2480 goto error;
2481 }
2482
2483 pthread_mutex_lock(&consumer_data->pid_mutex);
2484 if (consumer_data->pid == 0) {
2485 ERR("Consumerd did not start");
2486 pthread_mutex_unlock(&consumer_data->pid_mutex);
2487 goto error;
2488 }
2489 pthread_mutex_unlock(&consumer_data->pid_mutex);
2490
2491 return 0;
2492
2493 error:
2494 return ret;
2495 }
2496
2497 /*
2498 * Join consumer thread
2499 */
2500 static int join_consumer_thread(struct consumer_data *consumer_data)
2501 {
2502 void *status;
2503
2504 /* Consumer pid must be a real one. */
2505 if (consumer_data->pid > 0) {
2506 int ret;
2507 ret = kill(consumer_data->pid, SIGTERM);
2508 if (ret) {
2509 PERROR("Error killing consumer daemon");
2510 return ret;
2511 }
2512 return pthread_join(consumer_data->thread, &status);
2513 } else {
2514 return 0;
2515 }
2516 }
2517
2518 /*
2519 * Fork and exec a consumer daemon (consumerd).
2520 *
2521 * Return pid if successful else -1.
2522 */
2523 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2524 {
2525 int ret;
2526 pid_t pid;
2527 const char *consumer_to_use;
2528 const char *verbosity;
2529 struct stat st;
2530
2531 DBG("Spawning consumerd");
2532
2533 pid = fork();
2534 if (pid == 0) {
2535 /*
2536 * Exec consumerd.
2537 */
2538 if (opt_verbose_consumer) {
2539 verbosity = "--verbose";
2540 } else if (lttng_opt_quiet) {
2541 verbosity = "--quiet";
2542 } else {
2543 verbosity = "";
2544 }
2545
2546 switch (consumer_data->type) {
2547 case LTTNG_CONSUMER_KERNEL:
2548 /*
2549 * Find out which consumerd to execute. We will first try the
2550 * 64-bit path, then the sessiond's installation directory, and
2551 * fallback on the 32-bit one,
2552 */
2553 DBG3("Looking for a kernel consumer at these locations:");
2554 DBG3(" 1) %s", consumerd64_bin);
2555 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2556 DBG3(" 3) %s", consumerd32_bin);
2557 if (stat(consumerd64_bin, &st) == 0) {
2558 DBG3("Found location #1");
2559 consumer_to_use = consumerd64_bin;
2560 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2561 DBG3("Found location #2");
2562 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2563 } else if (stat(consumerd32_bin, &st) == 0) {
2564 DBG3("Found location #3");
2565 consumer_to_use = consumerd32_bin;
2566 } else {
2567 DBG("Could not find any valid consumerd executable");
2568 ret = -EINVAL;
2569 break;
2570 }
2571 DBG("Using kernel consumer at: %s", consumer_to_use);
2572 ret = execl(consumer_to_use,
2573 "lttng-consumerd", verbosity, "-k",
2574 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2575 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2576 "--group", tracing_group_name,
2577 NULL);
2578 break;
2579 case LTTNG_CONSUMER64_UST:
2580 {
2581 char *tmpnew = NULL;
2582
2583 if (consumerd64_libdir[0] != '\0') {
2584 char *tmp;
2585 size_t tmplen;
2586
2587 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2588 if (!tmp) {
2589 tmp = "";
2590 }
2591 tmplen = strlen("LD_LIBRARY_PATH=")
2592 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2593 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2594 if (!tmpnew) {
2595 ret = -ENOMEM;
2596 goto error;
2597 }
2598 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2599 strcat(tmpnew, consumerd64_libdir);
2600 if (tmp[0] != '\0') {
2601 strcat(tmpnew, ":");
2602 strcat(tmpnew, tmp);
2603 }
2604 ret = putenv(tmpnew);
2605 if (ret) {
2606 ret = -errno;
2607 free(tmpnew);
2608 goto error;
2609 }
2610 }
2611 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2612 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2613 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2614 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2615 "--group", tracing_group_name,
2616 NULL);
2617 if (consumerd64_libdir[0] != '\0') {
2618 free(tmpnew);
2619 }
2620 break;
2621 }
2622 case LTTNG_CONSUMER32_UST:
2623 {
2624 char *tmpnew = NULL;
2625
2626 if (consumerd32_libdir[0] != '\0') {
2627 char *tmp;
2628 size_t tmplen;
2629
2630 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2631 if (!tmp) {
2632 tmp = "";
2633 }
2634 tmplen = strlen("LD_LIBRARY_PATH=")
2635 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2636 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2637 if (!tmpnew) {
2638 ret = -ENOMEM;
2639 goto error;
2640 }
2641 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2642 strcat(tmpnew, consumerd32_libdir);
2643 if (tmp[0] != '\0') {
2644 strcat(tmpnew, ":");
2645 strcat(tmpnew, tmp);
2646 }
2647 ret = putenv(tmpnew);
2648 if (ret) {
2649 ret = -errno;
2650 free(tmpnew);
2651 goto error;
2652 }
2653 }
2654 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2655 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2656 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2657 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2658 "--group", tracing_group_name,
2659 NULL);
2660 if (consumerd32_libdir[0] != '\0') {
2661 free(tmpnew);
2662 }
2663 break;
2664 }
2665 default:
2666 PERROR("unknown consumer type");
2667 exit(EXIT_FAILURE);
2668 }
2669 if (errno != 0) {
2670 PERROR("Consumer execl()");
2671 }
2672 /* Reaching this point, we got a failure on our execl(). */
2673 exit(EXIT_FAILURE);
2674 } else if (pid > 0) {
2675 ret = pid;
2676 } else {
2677 PERROR("start consumer fork");
2678 ret = -errno;
2679 }
2680 error:
2681 return ret;
2682 }
2683
2684 /*
2685 * Spawn the consumerd daemon and session daemon thread.
2686 */
2687 static int start_consumerd(struct consumer_data *consumer_data)
2688 {
2689 int ret;
2690
2691 /*
2692 * Set the listen() state on the socket since there is a possible race
2693 * between the exec() of the consumer daemon and this call if place in the
2694 * consumer thread. See bug #366 for more details.
2695 */
2696 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2697 if (ret < 0) {
2698 goto error;
2699 }
2700
2701 pthread_mutex_lock(&consumer_data->pid_mutex);
2702 if (consumer_data->pid != 0) {
2703 pthread_mutex_unlock(&consumer_data->pid_mutex);
2704 goto end;
2705 }
2706
2707 ret = spawn_consumerd(consumer_data);
2708 if (ret < 0) {
2709 ERR("Spawning consumerd failed");
2710 pthread_mutex_unlock(&consumer_data->pid_mutex);
2711 goto error;
2712 }
2713
2714 /* Setting up the consumer_data pid */
2715 consumer_data->pid = ret;
2716 DBG2("Consumer pid %d", consumer_data->pid);
2717 pthread_mutex_unlock(&consumer_data->pid_mutex);
2718
2719 DBG2("Spawning consumer control thread");
2720 ret = spawn_consumer_thread(consumer_data);
2721 if (ret < 0) {
2722 ERR("Fatal error spawning consumer control thread");
2723 goto error;
2724 }
2725
2726 end:
2727 return 0;
2728
2729 error:
2730 /* Cleanup already created sockets on error. */
2731 if (consumer_data->err_sock >= 0) {
2732 int err;
2733
2734 err = close(consumer_data->err_sock);
2735 if (err < 0) {
2736 PERROR("close consumer data error socket");
2737 }
2738 }
2739 return ret;
2740 }
2741
2742 /*
2743 * Setup necessary data for kernel tracer action.
2744 */
2745 static int init_kernel_tracer(void)
2746 {
2747 int ret;
2748
2749 /* Modprobe lttng kernel modules */
2750 ret = modprobe_lttng_control();
2751 if (ret < 0) {
2752 goto error;
2753 }
2754
2755 /* Open debugfs lttng */
2756 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2757 if (kernel_tracer_fd < 0) {
2758 DBG("Failed to open %s", module_proc_lttng);
2759 ret = -1;
2760 goto error_open;
2761 }
2762
2763 /* Validate kernel version */
2764 ret = kernel_validate_version(kernel_tracer_fd);
2765 if (ret < 0) {
2766 goto error_version;
2767 }
2768
2769 ret = modprobe_lttng_data();
2770 if (ret < 0) {
2771 goto error_modules;
2772 }
2773
2774 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2775 return 0;
2776
2777 error_version:
2778 modprobe_remove_lttng_control();
2779 ret = close(kernel_tracer_fd);
2780 if (ret) {
2781 PERROR("close");
2782 }
2783 kernel_tracer_fd = -1;
2784 return LTTNG_ERR_KERN_VERSION;
2785
2786 error_modules:
2787 ret = close(kernel_tracer_fd);
2788 if (ret) {
2789 PERROR("close");
2790 }
2791
2792 error_open:
2793 modprobe_remove_lttng_control();
2794
2795 error:
2796 WARN("No kernel tracer available");
2797 kernel_tracer_fd = -1;
2798 if (!is_root) {
2799 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2800 } else {
2801 return LTTNG_ERR_KERN_NA;
2802 }
2803 }
2804
2805
2806 /*
2807 * Copy consumer output from the tracing session to the domain session. The
2808 * function also applies the right modification on a per domain basis for the
2809 * trace files destination directory.
2810 *
2811 * Should *NOT* be called with RCU read-side lock held.
2812 */
2813 static int copy_session_consumer(int domain, struct ltt_session *session)
2814 {
2815 int ret;
2816 const char *dir_name;
2817 struct consumer_output *consumer;
2818
2819 assert(session);
2820 assert(session->consumer);
2821
2822 switch (domain) {
2823 case LTTNG_DOMAIN_KERNEL:
2824 DBG3("Copying tracing session consumer output in kernel session");
2825 /*
2826 * XXX: We should audit the session creation and what this function
2827 * does "extra" in order to avoid a destroy since this function is used
2828 * in the domain session creation (kernel and ust) only. Same for UST
2829 * domain.
2830 */
2831 if (session->kernel_session->consumer) {
2832 consumer_output_put(session->kernel_session->consumer);
2833 }
2834 session->kernel_session->consumer =
2835 consumer_copy_output(session->consumer);
2836 /* Ease our life a bit for the next part */
2837 consumer = session->kernel_session->consumer;
2838 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2839 break;
2840 case LTTNG_DOMAIN_JUL:
2841 case LTTNG_DOMAIN_LOG4J:
2842 case LTTNG_DOMAIN_PYTHON:
2843 case LTTNG_DOMAIN_UST:
2844 DBG3("Copying tracing session consumer output in UST session");
2845 if (session->ust_session->consumer) {
2846 consumer_output_put(session->ust_session->consumer);
2847 }
2848 session->ust_session->consumer =
2849 consumer_copy_output(session->consumer);
2850 /* Ease our life a bit for the next part */
2851 consumer = session->ust_session->consumer;
2852 dir_name = DEFAULT_UST_TRACE_DIR;
2853 break;
2854 default:
2855 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2856 goto error;
2857 }
2858
2859 /* Append correct directory to subdir */
2860 strncat(consumer->subdir, dir_name,
2861 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2862 DBG3("Copy session consumer subdir %s", consumer->subdir);
2863
2864 ret = LTTNG_OK;
2865
2866 error:
2867 return ret;
2868 }
2869
2870 /*
2871 * Create an UST session and add it to the session ust list.
2872 *
2873 * Should *NOT* be called with RCU read-side lock held.
2874 */
2875 static int create_ust_session(struct ltt_session *session,
2876 struct lttng_domain *domain)
2877 {
2878 int ret;
2879 struct ltt_ust_session *lus = NULL;
2880
2881 assert(session);
2882 assert(domain);
2883 assert(session->consumer);
2884
2885 switch (domain->type) {
2886 case LTTNG_DOMAIN_JUL:
2887 case LTTNG_DOMAIN_LOG4J:
2888 case LTTNG_DOMAIN_PYTHON:
2889 case LTTNG_DOMAIN_UST:
2890 break;
2891 default:
2892 ERR("Unknown UST domain on create session %d", domain->type);
2893 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2894 goto error;
2895 }
2896
2897 DBG("Creating UST session");
2898
2899 lus = trace_ust_create_session(session->id);
2900 if (lus == NULL) {
2901 ret = LTTNG_ERR_UST_SESS_FAIL;
2902 goto error;
2903 }
2904
2905 lus->uid = session->uid;
2906 lus->gid = session->gid;
2907 lus->output_traces = session->output_traces;
2908 lus->snapshot_mode = session->snapshot_mode;
2909 lus->live_timer_interval = session->live_timer;
2910 session->ust_session = lus;
2911 if (session->shm_path[0]) {
2912 strncpy(lus->root_shm_path, session->shm_path,
2913 sizeof(lus->root_shm_path));
2914 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2915 strncpy(lus->shm_path, session->shm_path,
2916 sizeof(lus->shm_path));
2917 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2918 strncat(lus->shm_path, "/ust",
2919 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2920 }
2921 /* Copy session output to the newly created UST session */
2922 ret = copy_session_consumer(domain->type, session);
2923 if (ret != LTTNG_OK) {
2924 goto error;
2925 }
2926
2927 return LTTNG_OK;
2928
2929 error:
2930 free(lus);
2931 session->ust_session = NULL;
2932 return ret;
2933 }
2934
2935 /*
2936 * Create a kernel tracer session then create the default channel.
2937 */
2938 static int create_kernel_session(struct ltt_session *session)
2939 {
2940 int ret;
2941
2942 DBG("Creating kernel session");
2943
2944 ret = kernel_create_session(session, kernel_tracer_fd);
2945 if (ret < 0) {
2946 ret = LTTNG_ERR_KERN_SESS_FAIL;
2947 goto error;
2948 }
2949
2950 /* Code flow safety */
2951 assert(session->kernel_session);
2952
2953 /* Copy session output to the newly created Kernel session */
2954 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2955 if (ret != LTTNG_OK) {
2956 goto error;
2957 }
2958
2959 /* Create directory(ies) on local filesystem. */
2960 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2961 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2962 ret = run_as_mkdir_recursive(
2963 session->kernel_session->consumer->dst.trace_path,
2964 S_IRWXU | S_IRWXG, session->uid, session->gid);
2965 if (ret < 0) {
2966 if (errno != EEXIST) {
2967 ERR("Trace directory creation error");
2968 goto error;
2969 }
2970 }
2971 }
2972
2973 session->kernel_session->uid = session->uid;
2974 session->kernel_session->gid = session->gid;
2975 session->kernel_session->output_traces = session->output_traces;
2976 session->kernel_session->snapshot_mode = session->snapshot_mode;
2977
2978 return LTTNG_OK;
2979
2980 error:
2981 trace_kernel_destroy_session(session->kernel_session);
2982 session->kernel_session = NULL;
2983 return ret;
2984 }
2985
2986 /*
2987 * Count number of session permitted by uid/gid.
2988 */
2989 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2990 {
2991 unsigned int i = 0;
2992 struct ltt_session *session;
2993
2994 DBG("Counting number of available session for UID %d GID %d",
2995 uid, gid);
2996 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2997 /*
2998 * Only list the sessions the user can control.
2999 */
3000 if (!session_access_ok(session, uid, gid)) {
3001 continue;
3002 }
3003 i++;
3004 }
3005 return i;
3006 }
3007
3008 /*
3009 * Process the command requested by the lttng client within the command
3010 * context structure. This function make sure that the return structure (llm)
3011 * is set and ready for transmission before returning.
3012 *
3013 * Return any error encountered or 0 for success.
3014 *
3015 * "sock" is only used for special-case var. len data.
3016 *
3017 * Should *NOT* be called with RCU read-side lock held.
3018 */
3019 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
3020 int *sock_error)
3021 {
3022 int ret = LTTNG_OK;
3023 int need_tracing_session = 1;
3024 int need_domain;
3025
3026 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
3027
3028 assert(!rcu_read_ongoing());
3029
3030 *sock_error = 0;
3031
3032 switch (cmd_ctx->lsm->cmd_type) {
3033 case LTTNG_CREATE_SESSION:
3034 case LTTNG_CREATE_SESSION_SNAPSHOT:
3035 case LTTNG_CREATE_SESSION_LIVE:
3036 case LTTNG_DESTROY_SESSION:
3037 case LTTNG_LIST_SESSIONS:
3038 case LTTNG_LIST_DOMAINS:
3039 case LTTNG_START_TRACE:
3040 case LTTNG_STOP_TRACE:
3041 case LTTNG_DATA_PENDING:
3042 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3043 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3044 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3045 case LTTNG_SNAPSHOT_RECORD:
3046 case LTTNG_SAVE_SESSION:
3047 case LTTNG_SET_SESSION_SHM_PATH:
3048 case LTTNG_METADATA_REGENERATE:
3049 need_domain = 0;
3050 break;
3051 default:
3052 need_domain = 1;
3053 }
3054
3055 if (opt_no_kernel && need_domain
3056 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3057 if (!is_root) {
3058 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3059 } else {
3060 ret = LTTNG_ERR_KERN_NA;
3061 }
3062 goto error;
3063 }
3064
3065 /* Deny register consumer if we already have a spawned consumer. */
3066 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3067 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3068 if (kconsumer_data.pid > 0) {
3069 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3070 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3071 goto error;
3072 }
3073 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3074 }
3075
3076 /*
3077 * Check for command that don't needs to allocate a returned payload. We do
3078 * this here so we don't have to make the call for no payload at each
3079 * command.
3080 */
3081 switch(cmd_ctx->lsm->cmd_type) {
3082 case LTTNG_LIST_SESSIONS:
3083 case LTTNG_LIST_TRACEPOINTS:
3084 case LTTNG_LIST_TRACEPOINT_FIELDS:
3085 case LTTNG_LIST_DOMAINS:
3086 case LTTNG_LIST_CHANNELS:
3087 case LTTNG_LIST_EVENTS:
3088 case LTTNG_LIST_SYSCALLS:
3089 case LTTNG_LIST_TRACKER_PIDS:
3090 case LTTNG_DATA_PENDING:
3091 break;
3092 default:
3093 /* Setup lttng message with no payload */
3094 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3095 if (ret < 0) {
3096 /* This label does not try to unlock the session */
3097 goto init_setup_error;
3098 }
3099 }
3100
3101 /* Commands that DO NOT need a session. */
3102 switch (cmd_ctx->lsm->cmd_type) {
3103 case LTTNG_CREATE_SESSION:
3104 case LTTNG_CREATE_SESSION_SNAPSHOT:
3105 case LTTNG_CREATE_SESSION_LIVE:
3106 case LTTNG_CALIBRATE:
3107 case LTTNG_LIST_SESSIONS:
3108 case LTTNG_LIST_TRACEPOINTS:
3109 case LTTNG_LIST_SYSCALLS:
3110 case LTTNG_LIST_TRACEPOINT_FIELDS:
3111 case LTTNG_SAVE_SESSION:
3112 need_tracing_session = 0;
3113 break;
3114 default:
3115 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3116 /*
3117 * We keep the session list lock across _all_ commands
3118 * for now, because the per-session lock does not
3119 * handle teardown properly.
3120 */
3121 session_lock_list();
3122 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3123 if (cmd_ctx->session == NULL) {
3124 ret = LTTNG_ERR_SESS_NOT_FOUND;
3125 goto error;
3126 } else {
3127 /* Acquire lock for the session */
3128 session_lock(cmd_ctx->session);
3129 }
3130 break;
3131 }
3132
3133 /*
3134 * Commands that need a valid session but should NOT create one if none
3135 * exists. Instead of creating one and destroying it when the command is
3136 * handled, process that right before so we save some round trip in useless
3137 * code path.
3138 */
3139 switch (cmd_ctx->lsm->cmd_type) {
3140 case LTTNG_DISABLE_CHANNEL:
3141 case LTTNG_DISABLE_EVENT:
3142 switch (cmd_ctx->lsm->domain.type) {
3143 case LTTNG_DOMAIN_KERNEL:
3144 if (!cmd_ctx->session->kernel_session) {
3145 ret = LTTNG_ERR_NO_CHANNEL;
3146 goto error;
3147 }
3148 break;
3149 case LTTNG_DOMAIN_JUL:
3150 case LTTNG_DOMAIN_LOG4J:
3151 case LTTNG_DOMAIN_PYTHON:
3152 case LTTNG_DOMAIN_UST:
3153 if (!cmd_ctx->session->ust_session) {
3154 ret = LTTNG_ERR_NO_CHANNEL;
3155 goto error;
3156 }
3157 break;
3158 default:
3159 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3160 goto error;
3161 }
3162 default:
3163 break;
3164 }
3165
3166 if (!need_domain) {
3167 goto skip_domain;
3168 }
3169
3170 /*
3171 * Check domain type for specific "pre-action".
3172 */
3173 switch (cmd_ctx->lsm->domain.type) {
3174 case LTTNG_DOMAIN_KERNEL:
3175 if (!is_root) {
3176 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3177 goto error;
3178 }
3179
3180 /* Kernel tracer check */
3181 if (kernel_tracer_fd == -1) {
3182 /* Basically, load kernel tracer modules */
3183 ret = init_kernel_tracer();
3184 if (ret != 0) {
3185 goto error;
3186 }
3187 }
3188
3189 /* Consumer is in an ERROR state. Report back to client */
3190 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3191 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3192 goto error;
3193 }
3194
3195 /* Need a session for kernel command */
3196 if (need_tracing_session) {
3197 if (cmd_ctx->session->kernel_session == NULL) {
3198 ret = create_kernel_session(cmd_ctx->session);
3199 if (ret < 0) {
3200 ret = LTTNG_ERR_KERN_SESS_FAIL;
3201 goto error;
3202 }
3203 }
3204
3205 /* Start the kernel consumer daemon */
3206 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3207 if (kconsumer_data.pid == 0 &&
3208 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3209 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3210 ret = start_consumerd(&kconsumer_data);
3211 if (ret < 0) {
3212 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3213 goto error;
3214 }
3215 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3216 } else {
3217 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3218 }
3219
3220 /*
3221 * The consumer was just spawned so we need to add the socket to
3222 * the consumer output of the session if exist.
3223 */
3224 ret = consumer_create_socket(&kconsumer_data,
3225 cmd_ctx->session->kernel_session->consumer);
3226 if (ret < 0) {
3227 goto error;
3228 }
3229 }
3230
3231 break;
3232 case LTTNG_DOMAIN_JUL:
3233 case LTTNG_DOMAIN_LOG4J:
3234 case LTTNG_DOMAIN_PYTHON:
3235 case LTTNG_DOMAIN_UST:
3236 {
3237 if (!ust_app_supported()) {
3238 ret = LTTNG_ERR_NO_UST;
3239 goto error;
3240 }
3241 /* Consumer is in an ERROR state. Report back to client */
3242 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3243 ret = LTTNG_ERR_NO_USTCONSUMERD;
3244 goto error;
3245 }
3246
3247 if (need_tracing_session) {
3248 /* Create UST session if none exist. */
3249 if (cmd_ctx->session->ust_session == NULL) {
3250 ret = create_ust_session(cmd_ctx->session,
3251 &cmd_ctx->lsm->domain);
3252 if (ret != LTTNG_OK) {
3253 goto error;
3254 }
3255 }
3256
3257 /* Start the UST consumer daemons */
3258 /* 64-bit */
3259 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3260 if (consumerd64_bin[0] != '\0' &&
3261 ustconsumer64_data.pid == 0 &&
3262 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3263 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3264 ret = start_consumerd(&ustconsumer64_data);
3265 if (ret < 0) {
3266 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3267 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3268 goto error;
3269 }
3270
3271 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3272 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3273 } else {
3274 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3275 }
3276
3277 /*
3278 * Setup socket for consumer 64 bit. No need for atomic access
3279 * since it was set above and can ONLY be set in this thread.
3280 */
3281 ret = consumer_create_socket(&ustconsumer64_data,
3282 cmd_ctx->session->ust_session->consumer);
3283 if (ret < 0) {
3284 goto error;
3285 }
3286
3287 /* 32-bit */
3288 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3289 if (consumerd32_bin[0] != '\0' &&
3290 ustconsumer32_data.pid == 0 &&
3291 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3292 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3293 ret = start_consumerd(&ustconsumer32_data);
3294 if (ret < 0) {
3295 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3296 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3297 goto error;
3298 }
3299
3300 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3301 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3302 } else {
3303 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3304 }
3305
3306 /*
3307 * Setup socket for consumer 64 bit. No need for atomic access
3308 * since it was set above and can ONLY be set in this thread.
3309 */
3310 ret = consumer_create_socket(&ustconsumer32_data,
3311 cmd_ctx->session->ust_session->consumer);
3312 if (ret < 0) {
3313 goto error;
3314 }
3315 }
3316 break;
3317 }
3318 default:
3319 break;
3320 }
3321 skip_domain:
3322
3323 /* Validate consumer daemon state when start/stop trace command */
3324 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3325 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3326 switch (cmd_ctx->lsm->domain.type) {
3327 case LTTNG_DOMAIN_NONE:
3328 break;
3329 case LTTNG_DOMAIN_JUL:
3330 case LTTNG_DOMAIN_LOG4J:
3331 case LTTNG_DOMAIN_PYTHON:
3332 case LTTNG_DOMAIN_UST:
3333 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3334 ret = LTTNG_ERR_NO_USTCONSUMERD;
3335 goto error;
3336 }
3337 break;
3338 case LTTNG_DOMAIN_KERNEL:
3339 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3340 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3341 goto error;
3342 }
3343 break;
3344 default:
3345 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3346 goto error;
3347 }
3348 }
3349
3350 /*
3351 * Check that the UID or GID match that of the tracing session.
3352 * The root user can interact with all sessions.
3353 */
3354 if (need_tracing_session) {
3355 if (!session_access_ok(cmd_ctx->session,
3356 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3357 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
3358 ret = LTTNG_ERR_EPERM;
3359 goto error;
3360 }
3361 }
3362
3363 /*
3364 * Send relayd information to consumer as soon as we have a domain and a
3365 * session defined.
3366 */
3367 if (cmd_ctx->session && need_domain) {
3368 /*
3369 * Setup relayd if not done yet. If the relayd information was already
3370 * sent to the consumer, this call will gracefully return.
3371 */
3372 ret = cmd_setup_relayd(cmd_ctx->session);
3373 if (ret != LTTNG_OK) {
3374 goto error;
3375 }
3376 }
3377
3378 /* Process by command type */
3379 switch (cmd_ctx->lsm->cmd_type) {
3380 case LTTNG_ADD_CONTEXT:
3381 {
3382 /*
3383 * An LTTNG_ADD_CONTEXT command might have a supplementary
3384 * payload if the context being added is an application context.
3385 */
3386 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3387 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3388 char *provider_name = NULL, *context_name = NULL;
3389 size_t provider_name_len =
3390 cmd_ctx->lsm->u.context.provider_name_len;
3391 size_t context_name_len =
3392 cmd_ctx->lsm->u.context.context_name_len;
3393
3394 if (provider_name_len == 0 || context_name_len == 0) {
3395 /*
3396 * Application provider and context names MUST
3397 * be provided.
3398 */
3399 ret = -LTTNG_ERR_INVALID;
3400 goto error;
3401 }
3402
3403 provider_name = zmalloc(provider_name_len + 1);
3404 if (!provider_name) {
3405 ret = -LTTNG_ERR_NOMEM;
3406 goto error;
3407 }
3408 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3409 provider_name;
3410
3411 context_name = zmalloc(context_name_len + 1);
3412 if (!context_name) {
3413 ret = -LTTNG_ERR_NOMEM;
3414 goto error_add_context;
3415 }
3416 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3417 context_name;
3418
3419 ret = lttcomm_recv_unix_sock(sock, provider_name,
3420 provider_name_len);
3421 if (ret < 0) {
3422 goto error_add_context;
3423 }
3424
3425 ret = lttcomm_recv_unix_sock(sock, context_name,
3426 context_name_len);
3427 if (ret < 0) {
3428 goto error_add_context;
3429 }
3430 }
3431
3432 /*
3433 * cmd_add_context assumes ownership of the provider and context
3434 * names.
3435 */
3436 ret = cmd_add_context(cmd_ctx->session,
3437 cmd_ctx->lsm->domain.type,
3438 cmd_ctx->lsm->u.context.channel_name,
3439 &cmd_ctx->lsm->u.context.ctx,
3440 kernel_poll_pipe[1]);
3441
3442 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3443 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3444 error_add_context:
3445 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3446 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3447 if (ret < 0) {
3448 goto error;
3449 }
3450 break;
3451 }
3452 case LTTNG_DISABLE_CHANNEL:
3453 {
3454 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3455 cmd_ctx->lsm->u.disable.channel_name);
3456 break;
3457 }
3458 case LTTNG_DISABLE_EVENT:
3459 {
3460
3461 /*
3462 * FIXME: handle filter; for now we just receive the filter's
3463 * bytecode along with the filter expression which are sent by
3464 * liblttng-ctl and discard them.
3465 *
3466 * This fixes an issue where the client may block while sending
3467 * the filter payload and encounter an error because the session
3468 * daemon closes the socket without ever handling this data.
3469 */
3470 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3471 cmd_ctx->lsm->u.disable.bytecode_len;
3472
3473 if (count) {
3474 char data[LTTNG_FILTER_MAX_LEN];
3475
3476 DBG("Discarding disable event command payload of size %zu", count);
3477 while (count) {
3478 ret = lttcomm_recv_unix_sock(sock, data,
3479 count > sizeof(data) ? sizeof(data) : count);
3480 if (ret < 0) {
3481 goto error;
3482 }
3483
3484 count -= (size_t) ret;
3485 }
3486 }
3487 /* FIXME: passing packed structure to non-packed pointer */
3488 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3489 cmd_ctx->lsm->u.disable.channel_name,
3490 &cmd_ctx->lsm->u.disable.event);
3491 break;
3492 }
3493 case LTTNG_ENABLE_CHANNEL:
3494 {
3495 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3496 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
3497 break;
3498 }
3499 case LTTNG_TRACK_PID:
3500 {
3501 ret = cmd_track_pid(cmd_ctx->session,
3502 cmd_ctx->lsm->domain.type,
3503 cmd_ctx->lsm->u.pid_tracker.pid);
3504 break;
3505 }
3506 case LTTNG_UNTRACK_PID:
3507 {
3508 ret = cmd_untrack_pid(cmd_ctx->session,
3509 cmd_ctx->lsm->domain.type,
3510 cmd_ctx->lsm->u.pid_tracker.pid);
3511 break;
3512 }
3513 case LTTNG_ENABLE_EVENT:
3514 {
3515 struct lttng_event_exclusion *exclusion = NULL;
3516 struct lttng_filter_bytecode *bytecode = NULL;
3517 char *filter_expression = NULL;
3518
3519 /* Handle exclusion events and receive it from the client. */
3520 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3521 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3522
3523 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3524 (count * LTTNG_SYMBOL_NAME_LEN));
3525 if (!exclusion) {
3526 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3527 goto error;
3528 }
3529
3530 DBG("Receiving var len exclusion event list from client ...");
3531 exclusion->count = count;
3532 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3533 count * LTTNG_SYMBOL_NAME_LEN);
3534 if (ret <= 0) {
3535 DBG("Nothing recv() from client var len data... continuing");
3536 *sock_error = 1;
3537 free(exclusion);
3538 ret = LTTNG_ERR_EXCLUSION_INVAL;
3539 goto error;
3540 }
3541 }
3542
3543 /* Get filter expression from client. */
3544 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3545 size_t expression_len =
3546 cmd_ctx->lsm->u.enable.expression_len;
3547
3548 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3549 ret = LTTNG_ERR_FILTER_INVAL;
3550 free(exclusion);
3551 goto error;
3552 }
3553
3554 filter_expression = zmalloc(expression_len);
3555 if (!filter_expression) {
3556 free(exclusion);
3557 ret = LTTNG_ERR_FILTER_NOMEM;
3558 goto error;
3559 }
3560
3561 /* Receive var. len. data */
3562 DBG("Receiving var len filter's expression from client ...");
3563 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3564 expression_len);
3565 if (ret <= 0) {
3566 DBG("Nothing recv() from client car len data... continuing");
3567 *sock_error = 1;
3568 free(filter_expression);
3569 free(exclusion);
3570 ret = LTTNG_ERR_FILTER_INVAL;
3571 goto error;
3572 }
3573 }
3574
3575 /* Handle filter and get bytecode from client. */
3576 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3577 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3578
3579 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3580 ret = LTTNG_ERR_FILTER_INVAL;
3581 free(filter_expression);
3582 free(exclusion);
3583 goto error;
3584 }
3585
3586 bytecode = zmalloc(bytecode_len);
3587 if (!bytecode) {
3588 free(filter_expression);
3589 free(exclusion);
3590 ret = LTTNG_ERR_FILTER_NOMEM;
3591 goto error;
3592 }
3593
3594 /* Receive var. len. data */
3595 DBG("Receiving var len filter's bytecode from client ...");
3596 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3597 if (ret <= 0) {
3598 DBG("Nothing recv() from client car len data... continuing");
3599 *sock_error = 1;
3600 free(filter_expression);
3601 free(bytecode);
3602 free(exclusion);
3603 ret = LTTNG_ERR_FILTER_INVAL;
3604 goto error;
3605 }
3606
3607 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3608 free(filter_expression);
3609 free(bytecode);
3610 free(exclusion);
3611 ret = LTTNG_ERR_FILTER_INVAL;
3612 goto error;
3613 }
3614 }
3615
3616 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3617 cmd_ctx->lsm->u.enable.channel_name,
3618 &cmd_ctx->lsm->u.enable.event,
3619 filter_expression, bytecode, exclusion,
3620 kernel_poll_pipe[1]);
3621 break;
3622 }
3623 case LTTNG_LIST_TRACEPOINTS:
3624 {
3625 struct lttng_event *events;
3626 ssize_t nb_events;
3627
3628 session_lock_list();
3629 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3630 session_unlock_list();
3631 if (nb_events < 0) {
3632 /* Return value is a negative lttng_error_code. */
3633 ret = -nb_events;
3634 goto error;
3635 }
3636
3637 /*
3638 * Setup lttng message with payload size set to the event list size in
3639 * bytes and then copy list into the llm payload.
3640 */
3641 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3642 sizeof(struct lttng_event) * nb_events);
3643 free(events);
3644
3645 if (ret < 0) {
3646 goto setup_error;
3647 }
3648
3649 ret = LTTNG_OK;
3650 break;
3651 }
3652 case LTTNG_LIST_TRACEPOINT_FIELDS:
3653 {
3654 struct lttng_event_field *fields;
3655 ssize_t nb_fields;
3656
3657 session_lock_list();
3658 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3659 &fields);
3660 session_unlock_list();
3661 if (nb_fields < 0) {
3662 /* Return value is a negative lttng_error_code. */
3663 ret = -nb_fields;
3664 goto error;
3665 }
3666
3667 /*
3668 * Setup lttng message with payload size set to the event list size in
3669 * bytes and then copy list into the llm payload.
3670 */
3671 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3672 sizeof(struct lttng_event_field) * nb_fields);
3673 free(fields);
3674
3675 if (ret < 0) {
3676 goto setup_error;
3677 }
3678
3679 ret = LTTNG_OK;
3680 break;
3681 }
3682 case LTTNG_LIST_SYSCALLS:
3683 {
3684 struct lttng_event *events;
3685 ssize_t nb_events;
3686
3687 nb_events = cmd_list_syscalls(&events);
3688 if (nb_events < 0) {
3689 /* Return value is a negative lttng_error_code. */
3690 ret = -nb_events;
3691 goto error;
3692 }
3693
3694 /*
3695 * Setup lttng message with payload size set to the event list size in
3696 * bytes and then copy list into the llm payload.
3697 */
3698 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3699 sizeof(struct lttng_event) * nb_events);
3700 free(events);
3701
3702 if (ret < 0) {
3703 goto setup_error;
3704 }
3705
3706 ret = LTTNG_OK;
3707 break;
3708 }
3709 case LTTNG_LIST_TRACKER_PIDS:
3710 {
3711 int32_t *pids = NULL;
3712 ssize_t nr_pids;
3713
3714 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3715 cmd_ctx->lsm->domain.type, &pids);
3716 if (nr_pids < 0) {
3717 /* Return value is a negative lttng_error_code. */
3718 ret = -nr_pids;
3719 goto error;
3720 }
3721
3722 /*
3723 * Setup lttng message with payload size set to the event list size in
3724 * bytes and then copy list into the llm payload.
3725 */
3726 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3727 sizeof(int32_t) * nr_pids);
3728 free(pids);
3729
3730 if (ret < 0) {
3731 goto setup_error;
3732 }
3733
3734 ret = LTTNG_OK;
3735 break;
3736 }
3737 case LTTNG_SET_CONSUMER_URI:
3738 {
3739 size_t nb_uri, len;
3740 struct lttng_uri *uris;
3741
3742 nb_uri = cmd_ctx->lsm->u.uri.size;
3743 len = nb_uri * sizeof(struct lttng_uri);
3744
3745 if (nb_uri == 0) {
3746 ret = LTTNG_ERR_INVALID;
3747 goto error;
3748 }
3749
3750 uris = zmalloc(len);
3751 if (uris == NULL) {
3752 ret = LTTNG_ERR_FATAL;
3753 goto error;
3754 }
3755
3756 /* Receive variable len data */
3757 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3758 ret = lttcomm_recv_unix_sock(sock, uris, len);
3759 if (ret <= 0) {
3760 DBG("No URIs received from client... continuing");
3761 *sock_error = 1;
3762 ret = LTTNG_ERR_SESSION_FAIL;
3763 free(uris);
3764 goto error;
3765 }
3766
3767 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3768 free(uris);
3769 if (ret != LTTNG_OK) {
3770 goto error;
3771 }
3772
3773
3774 break;
3775 }
3776 case LTTNG_START_TRACE:
3777 {
3778 ret = cmd_start_trace(cmd_ctx->session);
3779 break;
3780 }
3781 case LTTNG_STOP_TRACE:
3782 {
3783 ret = cmd_stop_trace(cmd_ctx->session);
3784 break;
3785 }
3786 case LTTNG_CREATE_SESSION:
3787 {
3788 size_t nb_uri, len;
3789 struct lttng_uri *uris = NULL;
3790
3791 nb_uri = cmd_ctx->lsm->u.uri.size;
3792 len = nb_uri * sizeof(struct lttng_uri);
3793
3794 if (nb_uri > 0) {
3795 uris = zmalloc(len);
3796 if (uris == NULL) {
3797 ret = LTTNG_ERR_FATAL;
3798 goto error;
3799 }
3800
3801 /* Receive variable len data */
3802 DBG("Waiting for %zu URIs from client ...", nb_uri);
3803 ret = lttcomm_recv_unix_sock(sock, uris, len);
3804 if (ret <= 0) {
3805 DBG("No URIs received from client... continuing");
3806 *sock_error = 1;
3807 ret = LTTNG_ERR_SESSION_FAIL;
3808 free(uris);
3809 goto error;
3810 }
3811
3812 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3813 DBG("Creating session with ONE network URI is a bad call");
3814 ret = LTTNG_ERR_SESSION_FAIL;
3815 free(uris);
3816 goto error;
3817 }
3818 }
3819
3820 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3821 &cmd_ctx->creds, 0);
3822
3823 free(uris);
3824
3825 break;
3826 }
3827 case LTTNG_DESTROY_SESSION:
3828 {
3829 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3830
3831 /* Set session to NULL so we do not unlock it after free. */
3832 cmd_ctx->session = NULL;
3833 break;
3834 }
3835 case LTTNG_LIST_DOMAINS:
3836 {
3837 ssize_t nb_dom;
3838 struct lttng_domain *domains = NULL;
3839
3840 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3841 if (nb_dom < 0) {
3842 /* Return value is a negative lttng_error_code. */
3843 ret = -nb_dom;
3844 goto error;
3845 }
3846
3847 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3848 nb_dom * sizeof(struct lttng_domain));
3849 free(domains);
3850
3851 if (ret < 0) {
3852 goto setup_error;
3853 }
3854
3855 ret = LTTNG_OK;
3856 break;
3857 }
3858 case LTTNG_LIST_CHANNELS:
3859 {
3860 ssize_t payload_size;
3861 struct lttng_channel *channels = NULL;
3862
3863 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3864 cmd_ctx->session, &channels);
3865 if (payload_size < 0) {
3866 /* Return value is a negative lttng_error_code. */
3867 ret = -payload_size;
3868 goto error;
3869 }
3870
3871 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3872 payload_size);
3873 free(channels);
3874
3875 if (ret < 0) {
3876 goto setup_error;
3877 }
3878
3879 ret = LTTNG_OK;
3880 break;
3881 }
3882 case LTTNG_LIST_EVENTS:
3883 {
3884 ssize_t nb_event;
3885 struct lttng_event *events = NULL;
3886 struct lttcomm_event_command_header cmd_header;
3887 size_t total_size;
3888
3889 memset(&cmd_header, 0, sizeof(cmd_header));
3890 /* Extended infos are included at the end of events */
3891 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3892 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3893 &events, &total_size);
3894
3895 if (nb_event < 0) {
3896 /* Return value is a negative lttng_error_code. */
3897 ret = -nb_event;
3898 goto error;
3899 }
3900
3901 cmd_header.nb_events = nb_event;
3902 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3903 &cmd_header, sizeof(cmd_header));
3904 free(events);
3905
3906 if (ret < 0) {
3907 goto setup_error;
3908 }
3909
3910 ret = LTTNG_OK;
3911 break;
3912 }
3913 case LTTNG_LIST_SESSIONS:
3914 {
3915 unsigned int nr_sessions;
3916 void *sessions_payload;
3917 size_t payload_len;
3918
3919 session_lock_list();
3920 nr_sessions = lttng_sessions_count(
3921 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3922 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3923 payload_len = sizeof(struct lttng_session) * nr_sessions;
3924 sessions_payload = zmalloc(payload_len);
3925
3926 if (!sessions_payload) {
3927 session_unlock_list();
3928 ret = -ENOMEM;
3929 goto setup_error;
3930 }
3931
3932 cmd_list_lttng_sessions(sessions_payload,
3933 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3934 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3935 session_unlock_list();
3936
3937 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3938 payload_len);
3939 free(sessions_payload);
3940
3941 if (ret < 0) {
3942 goto setup_error;
3943 }
3944
3945 ret = LTTNG_OK;
3946 break;
3947 }
3948 case LTTNG_CALIBRATE:
3949 {
3950 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3951 &cmd_ctx->lsm->u.calibrate);
3952 break;
3953 }
3954 case LTTNG_REGISTER_CONSUMER:
3955 {
3956 struct consumer_data *cdata;
3957
3958 switch (cmd_ctx->lsm->domain.type) {
3959 case LTTNG_DOMAIN_KERNEL:
3960 cdata = &kconsumer_data;
3961 break;
3962 default:
3963 ret = LTTNG_ERR_UND;
3964 goto error;
3965 }
3966
3967 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3968 cmd_ctx->lsm->u.reg.path, cdata);
3969 break;
3970 }
3971 case LTTNG_DATA_PENDING:
3972 {
3973 int pending_ret;
3974 uint8_t pending_ret_byte;
3975
3976 pending_ret = cmd_data_pending(cmd_ctx->session);
3977
3978 /*
3979 * FIXME
3980 *
3981 * This function may returns 0 or 1 to indicate whether or not
3982 * there is data pending. In case of error, it should return an
3983 * LTTNG_ERR code. However, some code paths may still return
3984 * a nondescript error code, which we handle by returning an
3985 * "unknown" error.
3986 */
3987 if (pending_ret == 0 || pending_ret == 1) {
3988 /*
3989 * ret will be set to LTTNG_OK at the end of
3990 * this function.
3991 */
3992 } else if (pending_ret < 0) {
3993 ret = LTTNG_ERR_UNK;
3994 goto setup_error;
3995 } else {
3996 ret = pending_ret;
3997 goto setup_error;
3998 }
3999
4000 pending_ret_byte = (uint8_t) pending_ret;
4001
4002 /* 1 byte to return whether or not data is pending */
4003 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
4004 &pending_ret_byte, 1);
4005
4006 if (ret < 0) {
4007 goto setup_error;
4008 }
4009
4010 ret = LTTNG_OK;
4011 break;
4012 }
4013 case LTTNG_SNAPSHOT_ADD_OUTPUT:
4014 {
4015 struct lttcomm_lttng_output_id reply;
4016
4017 ret = cmd_snapshot_add_output(cmd_ctx->session,
4018 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
4019 if (ret != LTTNG_OK) {
4020 goto error;
4021 }
4022
4023 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
4024 sizeof(reply));
4025 if (ret < 0) {
4026 goto setup_error;
4027 }
4028
4029 /* Copy output list into message payload */
4030 ret = LTTNG_OK;
4031 break;
4032 }
4033 case LTTNG_SNAPSHOT_DEL_OUTPUT:
4034 {
4035 ret = cmd_snapshot_del_output(cmd_ctx->session,
4036 &cmd_ctx->lsm->u.snapshot_output.output);
4037 break;
4038 }
4039 case LTTNG_SNAPSHOT_LIST_OUTPUT:
4040 {
4041 ssize_t nb_output;
4042 struct lttng_snapshot_output *outputs = NULL;
4043
4044 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
4045 if (nb_output < 0) {
4046 ret = -nb_output;
4047 goto error;
4048 }
4049
4050 assert((nb_output > 0 && outputs) || nb_output == 0);
4051 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
4052 nb_output * sizeof(struct lttng_snapshot_output));
4053 free(outputs);
4054
4055 if (ret < 0) {
4056 goto setup_error;
4057 }
4058
4059 ret = LTTNG_OK;
4060 break;
4061 }
4062 case LTTNG_SNAPSHOT_RECORD:
4063 {
4064 ret = cmd_snapshot_record(cmd_ctx->session,
4065 &cmd_ctx->lsm->u.snapshot_record.output,
4066 cmd_ctx->lsm->u.snapshot_record.wait);
4067 break;
4068 }
4069 case LTTNG_CREATE_SESSION_SNAPSHOT:
4070 {
4071 size_t nb_uri, len;
4072 struct lttng_uri *uris = NULL;
4073
4074 nb_uri = cmd_ctx->lsm->u.uri.size;
4075 len = nb_uri * sizeof(struct lttng_uri);
4076
4077 if (nb_uri > 0) {
4078 uris = zmalloc(len);
4079 if (uris == NULL) {
4080 ret = LTTNG_ERR_FATAL;
4081 goto error;
4082 }
4083
4084 /* Receive variable len data */
4085 DBG("Waiting for %zu URIs from client ...", nb_uri);
4086 ret = lttcomm_recv_unix_sock(sock, uris, len);
4087 if (ret <= 0) {
4088 DBG("No URIs received from client... continuing");
4089 *sock_error = 1;
4090 ret = LTTNG_ERR_SESSION_FAIL;
4091 free(uris);
4092 goto error;
4093 }
4094
4095 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4096 DBG("Creating session with ONE network URI is a bad call");
4097 ret = LTTNG_ERR_SESSION_FAIL;
4098 free(uris);
4099 goto error;
4100 }
4101 }
4102
4103 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
4104 nb_uri, &cmd_ctx->creds);
4105 free(uris);
4106 break;
4107 }
4108 case LTTNG_CREATE_SESSION_LIVE:
4109 {
4110 size_t nb_uri, len;
4111 struct lttng_uri *uris = NULL;
4112
4113 nb_uri = cmd_ctx->lsm->u.uri.size;
4114 len = nb_uri * sizeof(struct lttng_uri);
4115
4116 if (nb_uri > 0) {
4117 uris = zmalloc(len);
4118 if (uris == NULL) {
4119 ret = LTTNG_ERR_FATAL;
4120 goto error;
4121 }
4122
4123 /* Receive variable len data */
4124 DBG("Waiting for %zu URIs from client ...", nb_uri);
4125 ret = lttcomm_recv_unix_sock(sock, uris, len);
4126 if (ret <= 0) {
4127 DBG("No URIs received from client... continuing");
4128 *sock_error = 1;
4129 ret = LTTNG_ERR_SESSION_FAIL;
4130 free(uris);
4131 goto error;
4132 }
4133
4134 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4135 DBG("Creating session with ONE network URI is a bad call");
4136 ret = LTTNG_ERR_SESSION_FAIL;
4137 free(uris);
4138 goto error;
4139 }
4140 }
4141
4142 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4143 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4144 free(uris);
4145 break;
4146 }
4147 case LTTNG_SAVE_SESSION:
4148 {
4149 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4150 &cmd_ctx->creds);
4151 break;
4152 }
4153 case LTTNG_SET_SESSION_SHM_PATH:
4154 {
4155 ret = cmd_set_session_shm_path(cmd_ctx->session,
4156 cmd_ctx->lsm->u.set_shm_path.shm_path);
4157 break;
4158 }
4159 case LTTNG_METADATA_REGENERATE:
4160 {
4161 ret = cmd_metadata_regenerate(cmd_ctx->session);
4162 break;
4163 }
4164 default:
4165 ret = LTTNG_ERR_UND;
4166 break;
4167 }
4168
4169 error:
4170 if (cmd_ctx->llm == NULL) {
4171 DBG("Missing llm structure. Allocating one.");
4172 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4173 goto setup_error;
4174 }
4175 }
4176 /* Set return code */
4177 cmd_ctx->llm->ret_code = ret;
4178 setup_error:
4179 if (cmd_ctx->session) {
4180 session_unlock(cmd_ctx->session);
4181 }
4182 if (need_tracing_session) {
4183 session_unlock_list();
4184 }
4185 init_setup_error:
4186 assert(!rcu_read_ongoing());
4187 return ret;
4188 }
4189
4190 /*
4191 * Thread managing health check socket.
4192 */
4193 static void *thread_manage_health(void *data)
4194 {
4195 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4196 uint32_t revents, nb_fd;
4197 struct lttng_poll_event events;
4198 struct health_comm_msg msg;
4199 struct health_comm_reply reply;
4200
4201 DBG("[thread] Manage health check started");
4202
4203 rcu_register_thread();
4204
4205 /* We might hit an error path before this is created. */
4206 lttng_poll_init(&events);
4207
4208 /* Create unix socket */
4209 sock = lttcomm_create_unix_sock(health_unix_sock_path);
4210 if (sock < 0) {
4211 ERR("Unable to create health check Unix socket");
4212 ret = -1;
4213 goto error;
4214 }
4215
4216 if (is_root) {
4217 /* lttng health client socket path permissions */
4218 ret = chown(health_unix_sock_path, 0,
4219 utils_get_group_id(tracing_group_name));
4220 if (ret < 0) {
4221 ERR("Unable to set group on %s", health_unix_sock_path);
4222 PERROR("chown");
4223 ret = -1;
4224 goto error;
4225 }
4226
4227 ret = chmod(health_unix_sock_path,
4228 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4229 if (ret < 0) {
4230 ERR("Unable to set permissions on %s", health_unix_sock_path);
4231 PERROR("chmod");
4232 ret = -1;
4233 goto error;
4234 }
4235 }
4236
4237 /*
4238 * Set the CLOEXEC flag. Return code is useless because either way, the
4239 * show must go on.
4240 */
4241 (void) utils_set_fd_cloexec(sock);
4242
4243 ret = lttcomm_listen_unix_sock(sock);
4244 if (ret < 0) {
4245 goto error;
4246 }
4247
4248 /*
4249 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4250 * more will be added to this poll set.
4251 */
4252 ret = sessiond_set_thread_pollset(&events, 2);
4253 if (ret < 0) {
4254 goto error;
4255 }
4256
4257 /* Add the application registration socket */
4258 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4259 if (ret < 0) {
4260 goto error;
4261 }
4262
4263 sessiond_notify_ready();
4264
4265 while (1) {
4266 DBG("Health check ready");
4267
4268 /* Inifinite blocking call, waiting for transmission */
4269 restart:
4270 ret = lttng_poll_wait(&events, -1);
4271 if (ret < 0) {
4272 /*
4273 * Restart interrupted system call.
4274 */
4275 if (errno == EINTR) {
4276 goto restart;
4277 }
4278 goto error;
4279 }
4280
4281 nb_fd = ret;
4282
4283 for (i = 0; i < nb_fd; i++) {
4284 /* Fetch once the poll data */
4285 revents = LTTNG_POLL_GETEV(&events, i);
4286 pollfd = LTTNG_POLL_GETFD(&events, i);
4287
4288 if (!revents) {
4289 /* No activity for this FD (poll implementation). */
4290 continue;
4291 }
4292
4293 /* Thread quit pipe has been closed. Killing thread. */
4294 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4295 if (ret) {
4296 err = 0;
4297 goto exit;
4298 }
4299
4300 /* Event on the registration socket */
4301 if (pollfd == sock) {
4302 if (revents & LPOLLIN) {
4303 continue;
4304 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4305 ERR("Health socket poll error");
4306 goto error;
4307 } else {
4308 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4309 goto error;
4310 }
4311 }
4312 }
4313
4314 new_sock = lttcomm_accept_unix_sock(sock);
4315 if (new_sock < 0) {
4316 goto error;
4317 }
4318
4319 /*
4320 * Set the CLOEXEC flag. Return code is useless because either way, the
4321 * show must go on.
4322 */
4323 (void) utils_set_fd_cloexec(new_sock);
4324
4325 DBG("Receiving data from client for health...");
4326 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4327 if (ret <= 0) {
4328 DBG("Nothing recv() from client... continuing");
4329 ret = close(new_sock);
4330 if (ret) {
4331 PERROR("close");
4332 }
4333 new_sock = -1;
4334 continue;
4335 }
4336
4337 rcu_thread_online();
4338
4339 memset(&reply, 0, sizeof(reply));
4340 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4341 /*
4342 * health_check_state returns 0 if health is
4343 * bad.
4344 */
4345 if (!health_check_state(health_sessiond, i)) {
4346 reply.ret_code |= 1ULL << i;
4347 }
4348 }
4349
4350 DBG2("Health check return value %" PRIx64, reply.ret_code);
4351
4352 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4353 if (ret < 0) {
4354 ERR("Failed to send health data back to client");
4355 }
4356
4357 /* End of transmission */
4358 ret = close(new_sock);
4359 if (ret) {
4360 PERROR("close");
4361 }
4362 new_sock = -1;
4363 }
4364
4365 exit:
4366 error:
4367 if (err) {
4368 ERR("Health error occurred in %s", __func__);
4369 }
4370 DBG("Health check thread dying");
4371 unlink(health_unix_sock_path);
4372 if (sock >= 0) {
4373 ret = close(sock);
4374 if (ret) {
4375 PERROR("close");
4376 }
4377 }
4378
4379 lttng_poll_clean(&events);
4380 stop_threads();
4381 rcu_unregister_thread();
4382 return NULL;
4383 }
4384
4385 /*
4386 * This thread manage all clients request using the unix client socket for
4387 * communication.
4388 */
4389 static void *thread_manage_clients(void *data)
4390 {
4391 int sock = -1, ret, i, pollfd, err = -1;
4392 int sock_error;
4393 uint32_t revents, nb_fd;
4394 struct command_ctx *cmd_ctx = NULL;
4395 struct lttng_poll_event events;
4396
4397 DBG("[thread] Manage client started");
4398
4399 rcu_register_thread();
4400
4401 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4402
4403 health_code_update();
4404
4405 ret = lttcomm_listen_unix_sock(client_sock);
4406 if (ret < 0) {
4407 goto error_listen;
4408 }
4409
4410 /*
4411 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4412 * more will be added to this poll set.
4413 */
4414 ret = sessiond_set_thread_pollset(&events, 2);
4415 if (ret < 0) {
4416 goto error_create_poll;
4417 }
4418
4419 /* Add the application registration socket */
4420 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4421 if (ret < 0) {
4422 goto error;
4423 }
4424
4425 sessiond_notify_ready();
4426 ret = sem_post(&load_info->message_thread_ready);
4427 if (ret) {
4428 PERROR("sem_post message_thread_ready");
4429 goto error;
4430 }
4431
4432 /* This testpoint is after we signal readiness to the parent. */
4433 if (testpoint(sessiond_thread_manage_clients)) {
4434 goto error;
4435 }
4436
4437 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4438 goto error;
4439 }
4440
4441 health_code_update();
4442
4443 while (1) {
4444 DBG("Accepting client command ...");
4445
4446 /* Inifinite blocking call, waiting for transmission */
4447 restart:
4448 health_poll_entry();
4449 ret = lttng_poll_wait(&events, -1);
4450 health_poll_exit();
4451 if (ret < 0) {
4452 /*
4453 * Restart interrupted system call.
4454 */
4455 if (errno == EINTR) {
4456 goto restart;
4457 }
4458 goto error;
4459 }
4460
4461 nb_fd = ret;
4462
4463 for (i = 0; i < nb_fd; i++) {
4464 /* Fetch once the poll data */
4465 revents = LTTNG_POLL_GETEV(&events, i);
4466 pollfd = LTTNG_POLL_GETFD(&events, i);
4467
4468 health_code_update();
4469
4470 if (!revents) {
4471 /* No activity for this FD (poll implementation). */
4472 continue;
4473 }
4474
4475 /* Thread quit pipe has been closed. Killing thread. */
4476 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4477 if (ret) {
4478 err = 0;
4479 goto exit;
4480 }
4481
4482 /* Event on the registration socket */
4483 if (pollfd == client_sock) {
4484 if (revents & LPOLLIN) {
4485 continue;
4486 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4487 ERR("Client socket poll error");
4488 goto error;
4489 } else {
4490 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4491 goto error;
4492 }
4493 }
4494 }
4495
4496 DBG("Wait for client response");
4497
4498 health_code_update();
4499
4500 sock = lttcomm_accept_unix_sock(client_sock);
4501 if (sock < 0) {
4502 goto error;
4503 }
4504
4505 /*
4506 * Set the CLOEXEC flag. Return code is useless because either way, the
4507 * show must go on.
4508 */
4509 (void) utils_set_fd_cloexec(sock);
4510
4511 /* Set socket option for credentials retrieval */
4512 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4513 if (ret < 0) {
4514 goto error;
4515 }
4516
4517 /* Allocate context command to process the client request */
4518 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4519 if (cmd_ctx == NULL) {
4520 PERROR("zmalloc cmd_ctx");
4521 goto error;
4522 }
4523
4524 /* Allocate data buffer for reception */
4525 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4526 if (cmd_ctx->lsm == NULL) {
4527 PERROR("zmalloc cmd_ctx->lsm");
4528 goto error;
4529 }
4530
4531 cmd_ctx->llm = NULL;
4532 cmd_ctx->session = NULL;
4533
4534 health_code_update();
4535
4536 /*
4537 * Data is received from the lttng client. The struct
4538 * lttcomm_session_msg (lsm) contains the command and data request of
4539 * the client.
4540 */
4541 DBG("Receiving data from client ...");
4542 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4543 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4544 if (ret <= 0) {
4545 DBG("Nothing recv() from client... continuing");
4546 ret = close(sock);
4547 if (ret) {
4548 PERROR("close");
4549 }
4550 sock = -1;
4551 clean_command_ctx(&cmd_ctx);
4552 continue;
4553 }
4554
4555 health_code_update();
4556
4557 // TODO: Validate cmd_ctx including sanity check for
4558 // security purpose.
4559
4560 rcu_thread_online();
4561 /*
4562 * This function dispatch the work to the kernel or userspace tracer
4563 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4564 * informations for the client. The command context struct contains
4565 * everything this function may needs.
4566 */
4567 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4568 rcu_thread_offline();
4569 if (ret < 0) {
4570 ret = close(sock);
4571 if (ret) {
4572 PERROR("close");
4573 }
4574 sock = -1;
4575 /*
4576 * TODO: Inform client somehow of the fatal error. At
4577 * this point, ret < 0 means that a zmalloc failed
4578 * (ENOMEM). Error detected but still accept
4579 * command, unless a socket error has been
4580 * detected.
4581 */
4582 clean_command_ctx(&cmd_ctx);
4583 continue;
4584 }
4585
4586 health_code_update();
4587
4588 DBG("Sending response (size: %d, retcode: %s (%d))",
4589 cmd_ctx->lttng_msg_size,
4590 lttng_strerror(-cmd_ctx->llm->ret_code),
4591 cmd_ctx->llm->ret_code);
4592 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4593 if (ret < 0) {
4594 ERR("Failed to send data back to client");
4595 }
4596
4597 /* End of transmission */
4598 ret = close(sock);
4599 if (ret) {
4600 PERROR("close");
4601 }
4602 sock = -1;
4603
4604 clean_command_ctx(&cmd_ctx);
4605
4606 health_code_update();
4607 }
4608
4609 exit:
4610 error:
4611 if (sock >= 0) {
4612 ret = close(sock);
4613 if (ret) {
4614 PERROR("close");
4615 }
4616 }
4617
4618 lttng_poll_clean(&events);
4619 clean_command_ctx(&cmd_ctx);
4620
4621 error_listen:
4622 error_create_poll:
4623 unlink(client_unix_sock_path);
4624 if (client_sock >= 0) {
4625 ret = close(client_sock);
4626 if (ret) {
4627 PERROR("close");
4628 }
4629 }
4630
4631 if (err) {
4632 health_error();
4633 ERR("Health error occurred in %s", __func__);
4634 }
4635
4636 health_unregister(health_sessiond);
4637
4638 DBG("Client thread dying");
4639
4640 rcu_unregister_thread();
4641
4642 /*
4643 * Since we are creating the consumer threads, we own them, so we need
4644 * to join them before our thread exits.
4645 */
4646 ret = join_consumer_thread(&kconsumer_data);
4647 if (ret) {
4648 errno = ret;
4649 PERROR("join_consumer");
4650 }
4651
4652 ret = join_consumer_thread(&ustconsumer32_data);
4653 if (ret) {
4654 errno = ret;
4655 PERROR("join_consumer ust32");
4656 }
4657
4658 ret = join_consumer_thread(&ustconsumer64_data);
4659 if (ret) {
4660 errno = ret;
4661 PERROR("join_consumer ust64");
4662 }
4663 return NULL;
4664 }
4665
4666 static int string_match(const char *str1, const char *str2)
4667 {
4668 return (str1 && str2) && !strcmp(str1, str2);
4669 }
4670
4671 /*
4672 * Take an option from the getopt output and set it in the right variable to be
4673 * used later.
4674 *
4675 * Return 0 on success else a negative value.
4676 */
4677 static int set_option(int opt, const char *arg, const char *optname)
4678 {
4679 int ret = 0;
4680
4681 if (string_match(optname, "client-sock") || opt == 'c') {
4682 if (!arg || *arg == '\0') {
4683 ret = -EINVAL;
4684 goto end;
4685 }
4686 if (lttng_is_setuid_setgid()) {
4687 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4688 "-c, --client-sock");
4689 } else {
4690 snprintf(client_unix_sock_path, PATH_MAX, "%s", arg);
4691 }
4692 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4693 if (!arg || *arg == '\0') {
4694 ret = -EINVAL;
4695 goto end;
4696 }
4697 if (lttng_is_setuid_setgid()) {
4698 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4699 "-a, --apps-sock");
4700 } else {
4701 snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg);
4702 }
4703 } else if (string_match(optname, "daemonize") || opt == 'd') {
4704 opt_daemon = 1;
4705 } else if (string_match(optname, "background") || opt == 'b') {
4706 opt_background = 1;
4707 } else if (string_match(optname, "group") || opt == 'g') {
4708 if (!arg || *arg == '\0') {
4709 ret = -EINVAL;
4710 goto end;
4711 }
4712 if (lttng_is_setuid_setgid()) {
4713 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4714 "-g, --group");
4715 } else {
4716 /*
4717 * If the override option is set, the pointer points to a
4718 * *non* const thus freeing it even though the variable type is
4719 * set to const.
4720 */
4721 if (tracing_group_name_override) {
4722 free((void *) tracing_group_name);
4723 }
4724 tracing_group_name = strdup(arg);
4725 if (!tracing_group_name) {
4726 PERROR("strdup");
4727 ret = -ENOMEM;
4728 }
4729 tracing_group_name_override = 1;
4730 }
4731 } else if (string_match(optname, "help") || opt == 'h') {
4732 ret = utils_show_man_page(8, "lttng-sessiond");
4733 if (ret) {
4734 ERR("Cannot view man page lttng-sessiond(8)");
4735 perror("exec");
4736 }
4737 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4738 } else if (string_match(optname, "version") || opt == 'V') {
4739 fprintf(stdout, "%s\n", VERSION);
4740 exit(EXIT_SUCCESS);
4741 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4742 opt_sig_parent = 1;
4743 } else if (string_match(optname, "kconsumerd-err-sock")) {
4744 if (!arg || *arg == '\0') {
4745 ret = -EINVAL;
4746 goto end;
4747 }
4748 if (lttng_is_setuid_setgid()) {
4749 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4750 "--kconsumerd-err-sock");
4751 } else {
4752 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4753 }
4754 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4755 if (!arg || *arg == '\0') {
4756 ret = -EINVAL;
4757 goto end;
4758 }
4759 if (lttng_is_setuid_setgid()) {
4760 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4761 "--kconsumerd-cmd-sock");
4762 } else {
4763 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4764 }
4765 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4766 if (!arg || *arg == '\0') {
4767 ret = -EINVAL;
4768 goto end;
4769 }
4770 if (lttng_is_setuid_setgid()) {
4771 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4772 "--ustconsumerd64-err-sock");
4773 } else {
4774 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4775 }
4776 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4777 if (!arg || *arg == '\0') {
4778 ret = -EINVAL;
4779 goto end;
4780 }
4781 if (lttng_is_setuid_setgid()) {
4782 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4783 "--ustconsumerd64-cmd-sock");
4784 } else {
4785 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4786 }
4787 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4788 if (!arg || *arg == '\0') {
4789 ret = -EINVAL;
4790 goto end;
4791 }
4792 if (lttng_is_setuid_setgid()) {
4793 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4794 "--ustconsumerd32-err-sock");
4795 } else {
4796 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg);
4797 }
4798 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4799 if (!arg || *arg == '\0') {
4800 ret = -EINVAL;
4801 goto end;
4802 }
4803 if (lttng_is_setuid_setgid()) {
4804 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4805 "--ustconsumerd32-cmd-sock");
4806 } else {
4807 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg);
4808 }
4809 } else if (string_match(optname, "no-kernel")) {
4810 opt_no_kernel = 1;
4811 } else if (string_match(optname, "quiet") || opt == 'q') {
4812 lttng_opt_quiet = 1;
4813 } else if (string_match(optname, "verbose") || opt == 'v') {
4814 /* Verbose level can increase using multiple -v */
4815 if (arg) {
4816 /* Value obtained from config file */
4817 lttng_opt_verbose = config_parse_value(arg);
4818 } else {
4819 /* -v used on command line */
4820 lttng_opt_verbose++;
4821 }
4822 /* Clamp value to [0, 3] */
4823 lttng_opt_verbose = lttng_opt_verbose < 0 ? 0 :
4824 (lttng_opt_verbose <= 3 ? lttng_opt_verbose : 3);
4825 } else if (string_match(optname, "verbose-consumer")) {
4826 if (arg) {
4827 opt_verbose_consumer = config_parse_value(arg);
4828 } else {
4829 opt_verbose_consumer++;
4830 }
4831 } else if (string_match(optname, "consumerd32-path")) {
4832 if (!arg || *arg == '\0') {
4833 ret = -EINVAL;
4834 goto end;
4835 }
4836 if (lttng_is_setuid_setgid()) {
4837 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4838 "--consumerd32-path");
4839 } else {
4840 if (consumerd32_bin_override) {
4841 free((void *) consumerd32_bin);
4842 }
4843 consumerd32_bin = strdup(arg);
4844 if (!consumerd32_bin) {
4845 PERROR("strdup");
4846 ret = -ENOMEM;
4847 }
4848 consumerd32_bin_override = 1;
4849 }
4850 } else if (string_match(optname, "consumerd32-libdir")) {
4851 if (!arg || *arg == '\0') {
4852 ret = -EINVAL;
4853 goto end;
4854 }
4855 if (lttng_is_setuid_setgid()) {
4856 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4857 "--consumerd32-libdir");
4858 } else {
4859 if (consumerd32_libdir_override) {
4860 free((void *) consumerd32_libdir);
4861 }
4862 consumerd32_libdir = strdup(arg);
4863 if (!consumerd32_libdir) {
4864 PERROR("strdup");
4865 ret = -ENOMEM;
4866 }
4867 consumerd32_libdir_override = 1;
4868 }
4869 } else if (string_match(optname, "consumerd64-path")) {
4870 if (!arg || *arg == '\0') {
4871 ret = -EINVAL;
4872 goto end;
4873 }
4874 if (lttng_is_setuid_setgid()) {
4875 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4876 "--consumerd64-path");
4877 } else {
4878 if (consumerd64_bin_override) {
4879 free((void *) consumerd64_bin);
4880 }
4881 consumerd64_bin = strdup(arg);
4882 if (!consumerd64_bin) {
4883 PERROR("strdup");
4884 ret = -ENOMEM;
4885 }
4886 consumerd64_bin_override = 1;
4887 }
4888 } else if (string_match(optname, "consumerd64-libdir")) {
4889 if (!arg || *arg == '\0') {
4890 ret = -EINVAL;
4891 goto end;
4892 }
4893 if (lttng_is_setuid_setgid()) {
4894 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4895 "--consumerd64-libdir");
4896 } else {
4897 if (consumerd64_libdir_override) {
4898 free((void *) consumerd64_libdir);
4899 }
4900 consumerd64_libdir = strdup(arg);
4901 if (!consumerd64_libdir) {
4902 PERROR("strdup");
4903 ret = -ENOMEM;
4904 }
4905 consumerd64_libdir_override = 1;
4906 }
4907 } else if (string_match(optname, "pidfile") || opt == 'p') {
4908 if (!arg || *arg == '\0') {
4909 ret = -EINVAL;
4910 goto end;
4911 }
4912 if (lttng_is_setuid_setgid()) {
4913 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4914 "-p, --pidfile");
4915 } else {
4916 free(opt_pidfile);
4917 opt_pidfile = strdup(arg);
4918 if (!opt_pidfile) {
4919 PERROR("strdup");
4920 ret = -ENOMEM;
4921 }
4922 }
4923 } else if (string_match(optname, "agent-tcp-port")) {
4924 if (!arg || *arg == '\0') {
4925 ret = -EINVAL;
4926 goto end;
4927 }
4928 if (lttng_is_setuid_setgid()) {
4929 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4930 "--agent-tcp-port");
4931 } else {
4932 unsigned long v;
4933
4934 if (!arg) {
4935 ret = -EINVAL;
4936 goto end;
4937 }
4938 errno = 0;
4939 v = strtoul(arg, NULL, 0);
4940 if (errno != 0 || !isdigit(arg[0])) {
4941 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
4942 return -1;
4943 }
4944 if (v == 0 || v >= 65535) {
4945 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
4946 return -1;
4947 }
4948 agent_tcp_port = (uint32_t) v;
4949 DBG3("Agent TCP port set to non default: %u", agent_tcp_port);
4950 }
4951 } else if (string_match(optname, "load") || opt == 'l') {
4952 if (!arg || *arg == '\0') {
4953 ret = -EINVAL;
4954 goto end;
4955 }
4956 if (lttng_is_setuid_setgid()) {
4957 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4958 "-l, --load");
4959 } else {
4960 free(opt_load_session_path);
4961 opt_load_session_path = strdup(arg);
4962 if (!opt_load_session_path) {
4963 PERROR("strdup");
4964 ret = -ENOMEM;
4965 }
4966 }
4967 } else if (string_match(optname, "kmod-probes")) {
4968 if (!arg || *arg == '\0') {
4969 ret = -EINVAL;
4970 goto end;
4971 }
4972 if (lttng_is_setuid_setgid()) {
4973 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4974 "--kmod-probes");
4975 } else {
4976 free(kmod_probes_list);
4977 kmod_probes_list = strdup(arg);
4978 if (!kmod_probes_list) {
4979 PERROR("strdup");
4980 ret = -ENOMEM;
4981 }
4982 }
4983 } else if (string_match(optname, "extra-kmod-probes")) {
4984 if (!arg || *arg == '\0') {
4985 ret = -EINVAL;
4986 goto end;
4987 }
4988 if (lttng_is_setuid_setgid()) {
4989 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4990 "--extra-kmod-probes");
4991 } else {
4992 free(kmod_extra_probes_list);
4993 kmod_extra_probes_list = strdup(arg);
4994 if (!kmod_extra_probes_list) {
4995 PERROR("strdup");
4996 ret = -ENOMEM;
4997 }
4998 }
4999 } else if (string_match(optname, "config") || opt == 'f') {
5000 /* This is handled in set_options() thus silent skip. */
5001 goto end;
5002 } else {
5003 /* Unknown option or other error.
5004 * Error is printed by getopt, just return */
5005 ret = -1;
5006 }
5007
5008 end:
5009 if (ret == -EINVAL) {
5010 const char *opt_name = "unknown";
5011 int i;
5012
5013 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
5014 i++) {
5015 if (opt == long_options[i].val) {
5016 opt_name = long_options[i].name;
5017 break;
5018 }
5019 }
5020
5021 WARN("Invalid argument provided for option \"%s\", using default value.",
5022 opt_name);
5023 }
5024
5025 return ret;
5026 }
5027
5028 /*
5029 * config_entry_handler_cb used to handle options read from a config file.
5030 * See config_entry_handler_cb comment in common/config/session-config.h for the
5031 * return value conventions.
5032 */
5033 static int config_entry_handler(const struct config_entry *entry, void *unused)
5034 {
5035 int ret = 0, i;
5036
5037 if (!entry || !entry->name || !entry->value) {
5038 ret = -EINVAL;
5039 goto end;
5040 }
5041
5042 /* Check if the option is to be ignored */
5043 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5044 if (!strcmp(entry->name, config_ignore_options[i])) {
5045 goto end;
5046 }
5047 }
5048
5049 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5050 i++) {
5051
5052 /* Ignore if not fully matched. */
5053 if (strcmp(entry->name, long_options[i].name)) {
5054 continue;
5055 }
5056
5057 /*
5058 * If the option takes no argument on the command line, we have to
5059 * check if the value is "true". We support non-zero numeric values,
5060 * true, on and yes.
5061 */
5062 if (!long_options[i].has_arg) {
5063 ret = config_parse_value(entry->value);
5064 if (ret <= 0) {
5065 if (ret) {
5066 WARN("Invalid configuration value \"%s\" for option %s",
5067 entry->value, entry->name);
5068 }
5069 /* False, skip boolean config option. */
5070 goto end;
5071 }
5072 }
5073
5074 ret = set_option(long_options[i].val, entry->value, entry->name);
5075 goto end;
5076 }
5077
5078 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5079
5080 end:
5081 return ret;
5082 }
5083
5084 /*
5085 * daemon configuration loading and argument parsing
5086 */
5087 static int set_options(int argc, char **argv)
5088 {
5089 int ret = 0, c = 0, option_index = 0;
5090 int orig_optopt = optopt, orig_optind = optind;
5091 char *optstring;
5092 const char *config_path = NULL;
5093
5094 optstring = utils_generate_optstring(long_options,
5095 sizeof(long_options) / sizeof(struct option));
5096 if (!optstring) {
5097 ret = -ENOMEM;
5098 goto end;
5099 }
5100
5101 /* Check for the --config option */
5102 while ((c = getopt_long(argc, argv, optstring, long_options,
5103 &option_index)) != -1) {
5104 if (c == '?') {
5105 ret = -EINVAL;
5106 goto end;
5107 } else if (c != 'f') {
5108 /* if not equal to --config option. */
5109 continue;
5110 }
5111
5112 if (lttng_is_setuid_setgid()) {
5113 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5114 "-f, --config");
5115 } else {
5116 config_path = utils_expand_path(optarg);
5117 if (!config_path) {
5118 ERR("Failed to resolve path: %s", optarg);
5119 }
5120 }
5121 }
5122
5123 ret = config_get_section_entries(config_path, config_section_name,
5124 config_entry_handler, NULL);
5125 if (ret) {
5126 if (ret > 0) {
5127 ERR("Invalid configuration option at line %i", ret);
5128 ret = -1;
5129 }
5130 goto end;
5131 }
5132
5133 /* Reset getopt's global state */
5134 optopt = orig_optopt;
5135 optind = orig_optind;
5136 while (1) {
5137 option_index = -1;
5138 /*
5139 * getopt_long() will not set option_index if it encounters a
5140 * short option.
5141 */
5142 c = getopt_long(argc, argv, optstring, long_options,
5143 &option_index);
5144 if (c == -1) {
5145 break;
5146 }
5147
5148 /*
5149 * Pass NULL as the long option name if popt left the index
5150 * unset.
5151 */
5152 ret = set_option(c, optarg,
5153 option_index < 0 ? NULL :
5154 long_options[option_index].name);
5155 if (ret < 0) {
5156 break;
5157 }
5158 }
5159
5160 end:
5161 free(optstring);
5162 return ret;
5163 }
5164
5165 /*
5166 * Creates the two needed socket by the daemon.
5167 * apps_sock - The communication socket for all UST apps.
5168 * client_sock - The communication of the cli tool (lttng).
5169 */
5170 static int init_daemon_socket(void)
5171 {
5172 int ret = 0;
5173 mode_t old_umask;
5174
5175 old_umask = umask(0);
5176
5177 /* Create client tool unix socket */
5178 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
5179 if (client_sock < 0) {
5180 ERR("Create unix sock failed: %s", client_unix_sock_path);
5181 ret = -1;
5182 goto end;
5183 }
5184
5185 /* Set the cloexec flag */
5186 ret = utils_set_fd_cloexec(client_sock);
5187 if (ret < 0) {
5188 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5189 "Continuing but note that the consumer daemon will have a "
5190 "reference to this socket on exec()", client_sock);
5191 }
5192
5193 /* File permission MUST be 660 */
5194 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5195 if (ret < 0) {
5196 ERR("Set file permissions failed: %s", client_unix_sock_path);
5197 PERROR("chmod");
5198 goto end;
5199 }
5200
5201 /* Create the application unix socket */
5202 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
5203 if (apps_sock < 0) {
5204 ERR("Create unix sock failed: %s", apps_unix_sock_path);
5205 ret = -1;
5206 goto end;
5207 }
5208
5209 /* Set the cloexec flag */
5210 ret = utils_set_fd_cloexec(apps_sock);
5211 if (ret < 0) {
5212 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5213 "Continuing but note that the consumer daemon will have a "
5214 "reference to this socket on exec()", apps_sock);
5215 }
5216
5217 /* File permission MUST be 666 */
5218 ret = chmod(apps_unix_sock_path,
5219 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5220 if (ret < 0) {
5221 ERR("Set file permissions failed: %s", apps_unix_sock_path);
5222 PERROR("chmod");
5223 goto end;
5224 }
5225
5226 DBG3("Session daemon client socket %d and application socket %d created",
5227 client_sock, apps_sock);
5228
5229 end:
5230 umask(old_umask);
5231 return ret;
5232 }
5233
5234 /*
5235 * Check if the global socket is available, and if a daemon is answering at the
5236 * other side. If yes, error is returned.
5237 */
5238 static int check_existing_daemon(void)
5239 {
5240 /* Is there anybody out there ? */
5241 if (lttng_session_daemon_alive()) {
5242 return -EEXIST;
5243 }
5244
5245 return 0;
5246 }
5247
5248 /*
5249 * Set the tracing group gid onto the client socket.
5250 *
5251 * Race window between mkdir and chown is OK because we are going from more
5252 * permissive (root.root) to less permissive (root.tracing).
5253 */
5254 static int set_permissions(char *rundir)
5255 {
5256 int ret;
5257 gid_t gid;
5258
5259 gid = utils_get_group_id(tracing_group_name);
5260
5261 /* Set lttng run dir */
5262 ret = chown(rundir, 0, gid);
5263 if (ret < 0) {
5264 ERR("Unable to set group on %s", rundir);
5265 PERROR("chown");
5266 }
5267
5268 /*
5269 * Ensure all applications and tracing group can search the run
5270 * dir. Allow everyone to read the directory, since it does not
5271 * buy us anything to hide its content.
5272 */
5273 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5274 if (ret < 0) {
5275 ERR("Unable to set permissions on %s", rundir);
5276 PERROR("chmod");
5277 }
5278
5279 /* lttng client socket path */
5280 ret = chown(client_unix_sock_path, 0, gid);
5281 if (ret < 0) {
5282 ERR("Unable to set group on %s", client_unix_sock_path);
5283 PERROR("chown");
5284 }
5285
5286 /* kconsumer error socket path */
5287 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5288 if (ret < 0) {
5289 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5290 PERROR("chown");
5291 }
5292
5293 /* 64-bit ustconsumer error socket path */
5294 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5295 if (ret < 0) {
5296 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5297 PERROR("chown");
5298 }
5299
5300 /* 32-bit ustconsumer compat32 error socket path */
5301 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5302 if (ret < 0) {
5303 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5304 PERROR("chown");
5305 }
5306
5307 DBG("All permissions are set");
5308
5309 return ret;
5310 }
5311
5312 /*
5313 * Create the lttng run directory needed for all global sockets and pipe.
5314 */
5315 static int create_lttng_rundir(const char *rundir)
5316 {
5317 int ret;
5318
5319 DBG3("Creating LTTng run directory: %s", rundir);
5320
5321 ret = mkdir(rundir, S_IRWXU);
5322 if (ret < 0) {
5323 if (errno != EEXIST) {
5324 ERR("Unable to create %s", rundir);
5325 goto error;
5326 } else {
5327 ret = 0;
5328 }
5329 }
5330
5331 error:
5332 return ret;
5333 }
5334
5335 /*
5336 * Setup sockets and directory needed by the kconsumerd communication with the
5337 * session daemon.
5338 */
5339 static int set_consumer_sockets(struct consumer_data *consumer_data,
5340 const char *rundir)
5341 {
5342 int ret;
5343 char path[PATH_MAX];
5344
5345 switch (consumer_data->type) {
5346 case LTTNG_CONSUMER_KERNEL:
5347 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
5348 break;
5349 case LTTNG_CONSUMER64_UST:
5350 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
5351 break;
5352 case LTTNG_CONSUMER32_UST:
5353 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
5354 break;
5355 default:
5356 ERR("Consumer type unknown");
5357 ret = -EINVAL;
5358 goto error;
5359 }
5360
5361 DBG2("Creating consumer directory: %s", path);
5362
5363 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5364 if (ret < 0) {
5365 if (errno != EEXIST) {
5366 PERROR("mkdir");
5367 ERR("Failed to create %s", path);
5368 goto error;
5369 }
5370 ret = -1;
5371 }
5372 if (is_root) {
5373 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
5374 if (ret < 0) {
5375 ERR("Unable to set group on %s", path);
5376 PERROR("chown");
5377 goto error;
5378 }
5379 }
5380
5381 /* Create the kconsumerd error unix socket */
5382 consumer_data->err_sock =
5383 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5384 if (consumer_data->err_sock < 0) {
5385 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5386 ret = -1;
5387 goto error;
5388 }
5389
5390 /*
5391 * Set the CLOEXEC flag. Return code is useless because either way, the
5392 * show must go on.
5393 */
5394 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5395 if (ret < 0) {
5396 PERROR("utils_set_fd_cloexec");
5397 /* continue anyway */
5398 }
5399
5400 /* File permission MUST be 660 */
5401 ret = chmod(consumer_data->err_unix_sock_path,
5402 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5403 if (ret < 0) {
5404 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5405 PERROR("chmod");
5406 goto error;
5407 }
5408
5409 error:
5410 return ret;
5411 }
5412
5413 /*
5414 * Signal handler for the daemon
5415 *
5416 * Simply stop all worker threads, leaving main() return gracefully after
5417 * joining all threads and calling cleanup().
5418 */
5419 static void sighandler(int sig)
5420 {
5421 switch (sig) {
5422 case SIGINT:
5423 DBG("SIGINT caught");
5424 stop_threads();
5425 break;
5426 case SIGTERM:
5427 DBG("SIGTERM caught");
5428 stop_threads();
5429 break;
5430 case SIGUSR1:
5431 CMM_STORE_SHARED(recv_child_signal, 1);
5432 break;
5433 default:
5434 break;
5435 }
5436 }
5437
5438 /*
5439 * Setup signal handler for :
5440 * SIGINT, SIGTERM, SIGPIPE
5441 */
5442 static int set_signal_handler(void)
5443 {
5444 int ret = 0;
5445 struct sigaction sa;
5446 sigset_t sigset;
5447
5448 if ((ret = sigemptyset(&sigset)) < 0) {
5449 PERROR("sigemptyset");
5450 return ret;
5451 }
5452
5453 sa.sa_mask = sigset;
5454 sa.sa_flags = 0;
5455
5456 sa.sa_handler = sighandler;
5457 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5458 PERROR("sigaction");
5459 return ret;
5460 }
5461
5462 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5463 PERROR("sigaction");
5464 return ret;
5465 }
5466
5467 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5468 PERROR("sigaction");
5469 return ret;
5470 }
5471
5472 sa.sa_handler = SIG_IGN;
5473 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5474 PERROR("sigaction");
5475 return ret;
5476 }
5477
5478 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5479
5480 return ret;
5481 }
5482
5483 /*
5484 * Set open files limit to unlimited. This daemon can open a large number of
5485 * file descriptors in order to consumer multiple kernel traces.
5486 */
5487 static void set_ulimit(void)
5488 {
5489 int ret;
5490 struct rlimit lim;
5491
5492 /* The kernel does not allowed an infinite limit for open files */
5493 lim.rlim_cur = 65535;
5494 lim.rlim_max = 65535;
5495
5496 ret = setrlimit(RLIMIT_NOFILE, &lim);
5497 if (ret < 0) {
5498 PERROR("failed to set open files limit");
5499 }
5500 }
5501
5502 /*
5503 * Write pidfile using the rundir and opt_pidfile.
5504 */
5505 static int write_pidfile(void)
5506 {
5507 int ret;
5508 char pidfile_path[PATH_MAX];
5509
5510 assert(rundir);
5511
5512 if (opt_pidfile) {
5513 if (lttng_strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path))) {
5514 ret = -1;
5515 goto error;
5516 }
5517 } else {
5518 /* Build pidfile path from rundir and opt_pidfile. */
5519 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
5520 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
5521 if (ret < 0) {
5522 PERROR("snprintf pidfile path");
5523 goto error;
5524 }
5525 }
5526
5527 /*
5528 * Create pid file in rundir.
5529 */
5530 ret = utils_create_pid_file(getpid(), pidfile_path);
5531 error:
5532 return ret;
5533 }
5534
5535 /*
5536 * Create lockfile using the rundir and return its fd.
5537 */
5538 static int create_lockfile(void)
5539 {
5540 int ret;
5541 char lockfile_path[PATH_MAX];
5542
5543 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
5544 if (ret < 0) {
5545 goto error;
5546 }
5547
5548 ret = utils_create_lock_file(lockfile_path);
5549 error:
5550 return ret;
5551 }
5552
5553 /*
5554 * Write agent TCP port using the rundir.
5555 */
5556 static int write_agent_port(void)
5557 {
5558 int ret;
5559 char path[PATH_MAX];
5560
5561 assert(rundir);
5562
5563 ret = snprintf(path, sizeof(path), "%s/"
5564 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE, rundir);
5565 if (ret < 0) {
5566 PERROR("snprintf agent port path");
5567 goto error;
5568 }
5569
5570 /*
5571 * Create TCP agent port file in rundir.
5572 */
5573 ret = utils_create_pid_file(agent_tcp_port, path);
5574
5575 error:
5576 return ret;
5577 }
5578
5579 /*
5580 * main
5581 */
5582 int main(int argc, char **argv)
5583 {
5584 int ret = 0, retval = 0;
5585 void *status;
5586 const char *home_path, *env_app_timeout;
5587
5588 init_kernel_workarounds();
5589
5590 rcu_register_thread();
5591
5592 if (set_signal_handler()) {
5593 retval = -1;
5594 goto exit_set_signal_handler;
5595 }
5596
5597 setup_consumerd_path();
5598
5599 page_size = sysconf(_SC_PAGESIZE);
5600 if (page_size < 0) {
5601 PERROR("sysconf _SC_PAGESIZE");
5602 page_size = LONG_MAX;
5603 WARN("Fallback page size to %ld", page_size);
5604 }
5605
5606 /*
5607 * Parse arguments and load the daemon configuration file.
5608 *
5609 * We have an exit_options exit path to free memory reserved by
5610 * set_options. This is needed because the rest of sessiond_cleanup()
5611 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5612 * depends on set_options.
5613 */
5614 progname = argv[0];
5615 if (set_options(argc, argv)) {
5616 retval = -1;
5617 goto exit_options;
5618 }
5619
5620 /* Daemonize */
5621 if (opt_daemon || opt_background) {
5622 int i;
5623
5624 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5625 !opt_background);
5626 if (ret < 0) {
5627 retval = -1;
5628 goto exit_options;
5629 }
5630
5631 /*
5632 * We are in the child. Make sure all other file descriptors are
5633 * closed, in case we are called with more opened file
5634 * descriptors than the standard ones.
5635 */
5636 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5637 (void) close(i);
5638 }
5639 }
5640
5641 if (run_as_create_worker(argv[0]) < 0) {
5642 goto exit_create_run_as_worker_cleanup;
5643 }
5644
5645 /*
5646 * Starting from here, we can create threads. This needs to be after
5647 * lttng_daemonize due to RCU.
5648 */
5649
5650 /*
5651 * Initialize the health check subsystem. This call should set the
5652 * appropriate time values.
5653 */
5654 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5655 if (!health_sessiond) {
5656 PERROR("health_app_create error");
5657 retval = -1;
5658 goto exit_health_sessiond_cleanup;
5659 }
5660
5661 if (init_ht_cleanup_quit_pipe()) {
5662 retval = -1;
5663 goto exit_ht_cleanup_quit_pipe;
5664 }
5665
5666 /* Setup the thread ht_cleanup communication pipe. */
5667 if (utils_create_pipe_cloexec(ht_cleanup_pipe)) {
5668 retval = -1;
5669 goto exit_ht_cleanup_pipe;
5670 }
5671
5672 /* Set up max poll set size */
5673 if (lttng_poll_set_max_size()) {
5674 retval = -1;
5675 goto exit_set_max_size;
5676 }
5677
5678 /* Create thread to clean up RCU hash tables */
5679 ret = pthread_create(&ht_cleanup_thread, NULL,
5680 thread_ht_cleanup, (void *) NULL);
5681 if (ret) {
5682 errno = ret;
5683 PERROR("pthread_create ht_cleanup");
5684 retval = -1;
5685 goto exit_ht_cleanup;
5686 }
5687
5688 /* Create thread quit pipe */
5689 if (init_thread_quit_pipe()) {
5690 retval = -1;
5691 goto exit_init_data;
5692 }
5693
5694 /* Check if daemon is UID = 0 */
5695 is_root = !getuid();
5696
5697 if (is_root) {
5698 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
5699 if (!rundir) {
5700 retval = -1;
5701 goto exit_init_data;
5702 }
5703
5704 /* Create global run dir with root access */
5705 if (create_lttng_rundir(rundir)) {
5706 retval = -1;
5707 goto exit_init_data;
5708 }
5709
5710 if (strlen(apps_unix_sock_path) == 0) {
5711 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5712 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
5713 if (ret < 0) {
5714 retval = -1;
5715 goto exit_init_data;
5716 }
5717 }
5718
5719 if (strlen(client_unix_sock_path) == 0) {
5720 ret = snprintf(client_unix_sock_path, PATH_MAX,
5721 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
5722 if (ret < 0) {
5723 retval = -1;
5724 goto exit_init_data;
5725 }
5726 }
5727
5728 /* Set global SHM for ust */
5729 if (strlen(wait_shm_path) == 0) {
5730 ret = snprintf(wait_shm_path, PATH_MAX,
5731 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
5732 if (ret < 0) {
5733 retval = -1;
5734 goto exit_init_data;
5735 }
5736 }
5737
5738 if (strlen(health_unix_sock_path) == 0) {
5739 ret = snprintf(health_unix_sock_path,
5740 sizeof(health_unix_sock_path),
5741 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
5742 if (ret < 0) {
5743 retval = -1;
5744 goto exit_init_data;
5745 }
5746 }
5747
5748 /* Setup kernel consumerd path */
5749 ret = snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
5750 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
5751 if (ret < 0) {
5752 retval = -1;
5753 goto exit_init_data;
5754 }
5755 ret = snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
5756 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
5757 if (ret < 0) {
5758 retval = -1;
5759 goto exit_init_data;
5760 }
5761
5762 DBG2("Kernel consumer err path: %s",
5763 kconsumer_data.err_unix_sock_path);
5764 DBG2("Kernel consumer cmd path: %s",
5765 kconsumer_data.cmd_unix_sock_path);
5766 } else {
5767 home_path = utils_get_home_dir();
5768 if (home_path == NULL) {
5769 /* TODO: Add --socket PATH option */
5770 ERR("Can't get HOME directory for sockets creation.");
5771 retval = -1;
5772 goto exit_init_data;
5773 }
5774
5775 /*
5776 * Create rundir from home path. This will create something like
5777 * $HOME/.lttng
5778 */
5779 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
5780 if (ret < 0) {
5781 retval = -1;
5782 goto exit_init_data;
5783 }
5784
5785 if (create_lttng_rundir(rundir)) {
5786 retval = -1;
5787 goto exit_init_data;
5788 }
5789
5790 if (strlen(apps_unix_sock_path) == 0) {
5791 ret = snprintf(apps_unix_sock_path, PATH_MAX,
5792 DEFAULT_HOME_APPS_UNIX_SOCK,
5793 home_path);
5794 if (ret < 0) {
5795 retval = -1;
5796 goto exit_init_data;
5797 }
5798 }
5799
5800 /* Set the cli tool unix socket path */
5801 if (strlen(client_unix_sock_path) == 0) {
5802 ret = snprintf(client_unix_sock_path, PATH_MAX,
5803 DEFAULT_HOME_CLIENT_UNIX_SOCK,
5804 home_path);
5805 if (ret < 0) {
5806 retval = -1;
5807 goto exit_init_data;
5808 }
5809 }
5810
5811 /* Set global SHM for ust */
5812 if (strlen(wait_shm_path) == 0) {
5813 ret = snprintf(wait_shm_path, PATH_MAX,
5814 DEFAULT_HOME_APPS_WAIT_SHM_PATH,
5815 getuid());
5816 if (ret < 0) {
5817 retval = -1;
5818 goto exit_init_data;
5819 }
5820 }
5821
5822 /* Set health check Unix path */
5823 if (strlen(health_unix_sock_path) == 0) {
5824 ret = snprintf(health_unix_sock_path,
5825 sizeof(health_unix_sock_path),
5826 DEFAULT_HOME_HEALTH_UNIX_SOCK,
5827 home_path);
5828 if (ret < 0) {
5829 retval = -1;
5830 goto exit_init_data;
5831 }
5832 }
5833 }
5834
5835 lockfile_fd = create_lockfile();
5836 if (lockfile_fd < 0) {
5837 retval = -1;
5838 goto exit_init_data;
5839 }
5840
5841 /* Set consumer initial state */
5842 kernel_consumerd_state = CONSUMER_STOPPED;
5843 ust_consumerd_state = CONSUMER_STOPPED;
5844
5845 DBG("Client socket path %s", client_unix_sock_path);
5846 DBG("Application socket path %s", apps_unix_sock_path);
5847 DBG("Application wait path %s", wait_shm_path);
5848 DBG("LTTng run directory path: %s", rundir);
5849
5850 /* 32 bits consumerd path setup */
5851 ret = snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
5852 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
5853 if (ret < 0) {
5854 PERROR("snprintf 32-bit consumer error socket path");
5855 retval = -1;
5856 goto exit_init_data;
5857 }
5858 ret = snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
5859 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
5860 if (ret < 0) {
5861 PERROR("snprintf 32-bit consumer command socket path");
5862 retval = -1;
5863 goto exit_init_data;
5864 }
5865
5866 DBG2("UST consumer 32 bits err path: %s",
5867 ustconsumer32_data.err_unix_sock_path);
5868 DBG2("UST consumer 32 bits cmd path: %s",
5869 ustconsumer32_data.cmd_unix_sock_path);
5870
5871 /* 64 bits consumerd path setup */
5872 ret = snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
5873 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
5874 if (ret < 0) {
5875 PERROR("snprintf 64-bit consumer error socket path");
5876 retval = -1;
5877 goto exit_init_data;
5878 }
5879 ret = snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
5880 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
5881 if (ret < 0) {
5882 PERROR("snprintf 64-bit consumer command socket path");
5883 retval = -1;
5884 goto exit_init_data;
5885 }
5886
5887 DBG2("UST consumer 64 bits err path: %s",
5888 ustconsumer64_data.err_unix_sock_path);
5889 DBG2("UST consumer 64 bits cmd path: %s",
5890 ustconsumer64_data.cmd_unix_sock_path);
5891
5892 /*
5893 * See if daemon already exist.
5894 */
5895 if (check_existing_daemon()) {
5896 ERR("Already running daemon.\n");
5897 /*
5898 * We do not goto exit because we must not cleanup()
5899 * because a daemon is already running.
5900 */
5901 retval = -1;
5902 goto exit_init_data;
5903 }
5904
5905 /*
5906 * Init UST app hash table. Alloc hash table before this point since
5907 * cleanup() can get called after that point.
5908 */
5909 if (ust_app_ht_alloc()) {
5910 ERR("Failed to allocate UST app hash table");
5911 retval = -1;
5912 goto exit_init_data;
5913 }
5914
5915 /*
5916 * Initialize agent app hash table. We allocate the hash table here
5917 * since cleanup() can get called after this point.
5918 */
5919 if (agent_app_ht_alloc()) {
5920 ERR("Failed to allocate Agent app hash table");
5921 retval = -1;
5922 goto exit_init_data;
5923 }
5924
5925 /*
5926 * These actions must be executed as root. We do that *after* setting up
5927 * the sockets path because we MUST make the check for another daemon using
5928 * those paths *before* trying to set the kernel consumer sockets and init
5929 * kernel tracer.
5930 */
5931 if (is_root) {
5932 if (set_consumer_sockets(&kconsumer_data, rundir)) {
5933 retval = -1;
5934 goto exit_init_data;
5935 }
5936
5937 /* Setup kernel tracer */
5938 if (!opt_no_kernel) {
5939 init_kernel_tracer();
5940 if (kernel_tracer_fd >= 0) {
5941 ret = syscall_init_table();
5942 if (ret < 0) {
5943 ERR("Unable to populate syscall table. "
5944 "Syscall tracing won't work "
5945 "for this session daemon.");
5946 }
5947 }
5948 }
5949
5950 /* Set ulimit for open files */
5951 set_ulimit();
5952 }
5953 /* init lttng_fd tracking must be done after set_ulimit. */
5954 lttng_fd_init();
5955
5956 if (set_consumer_sockets(&ustconsumer64_data, rundir)) {
5957 retval = -1;
5958 goto exit_init_data;
5959 }
5960
5961 if (set_consumer_sockets(&ustconsumer32_data, rundir)) {
5962 retval = -1;
5963 goto exit_init_data;
5964 }
5965
5966 /* Setup the needed unix socket */
5967 if (init_daemon_socket()) {
5968 retval = -1;
5969 goto exit_init_data;
5970 }
5971
5972 /* Set credentials to socket */
5973 if (is_root && set_permissions(rundir)) {
5974 retval = -1;
5975 goto exit_init_data;
5976 }
5977
5978 /* Get parent pid if -S, --sig-parent is specified. */
5979 if (opt_sig_parent) {
5980 ppid = getppid();
5981 }
5982
5983 /* Setup the kernel pipe for waking up the kernel thread */
5984 if (is_root && !opt_no_kernel) {
5985 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5986 retval = -1;
5987 goto exit_init_data;
5988 }
5989 }
5990
5991 /* Setup the thread apps communication pipe. */
5992 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5993 retval = -1;
5994 goto exit_init_data;
5995 }
5996
5997 /* Setup the thread apps notify communication pipe. */
5998 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
5999 retval = -1;
6000 goto exit_init_data;
6001 }
6002
6003 /* Initialize global buffer per UID and PID registry. */
6004 buffer_reg_init_uid_registry();
6005 buffer_reg_init_pid_registry();
6006
6007 /* Init UST command queue. */
6008 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
6009
6010 /*
6011 * Get session list pointer. This pointer MUST NOT be free'd. This list
6012 * is statically declared in session.c
6013 */
6014 session_list_ptr = session_get_list();
6015
6016 cmd_init();
6017
6018 /* Check for the application socket timeout env variable. */
6019 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
6020 if (env_app_timeout) {
6021 app_socket_timeout = atoi(env_app_timeout);
6022 } else {
6023 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
6024 }
6025
6026 ret = write_pidfile();
6027 if (ret) {
6028 ERR("Error in write_pidfile");
6029 retval = -1;
6030 goto exit_init_data;
6031 }
6032 ret = write_agent_port();
6033 if (ret) {
6034 ERR("Error in write_agent_port");
6035 retval = -1;
6036 goto exit_init_data;
6037 }
6038
6039 /* Initialize communication library */
6040 lttcomm_init();
6041 /* Initialize TCP timeout values */
6042 lttcomm_inet_init();
6043
6044 if (load_session_init_data(&load_info) < 0) {
6045 retval = -1;
6046 goto exit_init_data;
6047 }
6048 load_info->path = opt_load_session_path;
6049
6050 /* Create health-check thread */
6051 ret = pthread_create(&health_thread, NULL,
6052 thread_manage_health, (void *) NULL);
6053 if (ret) {
6054 errno = ret;
6055 PERROR("pthread_create health");
6056 retval = -1;
6057 goto exit_health;
6058 }
6059
6060 /* Create thread to manage the client socket */
6061 ret = pthread_create(&client_thread, NULL,
6062 thread_manage_clients, (void *) NULL);
6063 if (ret) {
6064 errno = ret;
6065 PERROR("pthread_create clients");
6066 retval = -1;
6067 goto exit_client;
6068 }
6069
6070 /* Create thread to dispatch registration */
6071 ret = pthread_create(&dispatch_thread, NULL,
6072 thread_dispatch_ust_registration, (void *) NULL);
6073 if (ret) {
6074 errno = ret;
6075 PERROR("pthread_create dispatch");
6076 retval = -1;
6077 goto exit_dispatch;
6078 }
6079
6080 /* Create thread to manage application registration. */
6081 ret = pthread_create(&reg_apps_thread, NULL,
6082 thread_registration_apps, (void *) NULL);
6083 if (ret) {
6084 errno = ret;
6085 PERROR("pthread_create registration");
6086 retval = -1;
6087 goto exit_reg_apps;
6088 }
6089
6090 /* Create thread to manage application socket */
6091 ret = pthread_create(&apps_thread, NULL,
6092 thread_manage_apps, (void *) NULL);
6093 if (ret) {
6094 errno = ret;
6095 PERROR("pthread_create apps");
6096 retval = -1;
6097 goto exit_apps;
6098 }
6099
6100 /* Create thread to manage application notify socket */
6101 ret = pthread_create(&apps_notify_thread, NULL,
6102 ust_thread_manage_notify, (void *) NULL);
6103 if (ret) {
6104 errno = ret;
6105 PERROR("pthread_create notify");
6106 retval = -1;
6107 goto exit_apps_notify;
6108 }
6109
6110 /* Create agent registration thread. */
6111 ret = pthread_create(&agent_reg_thread, NULL,
6112 agent_thread_manage_registration, (void *) NULL);
6113 if (ret) {
6114 errno = ret;
6115 PERROR("pthread_create agent");
6116 retval = -1;
6117 goto exit_agent_reg;
6118 }
6119
6120 /* Don't start this thread if kernel tracing is not requested nor root */
6121 if (is_root && !opt_no_kernel) {
6122 /* Create kernel thread to manage kernel event */
6123 ret = pthread_create(&kernel_thread, NULL,
6124 thread_manage_kernel, (void *) NULL);
6125 if (ret) {
6126 errno = ret;
6127 PERROR("pthread_create kernel");
6128 retval = -1;
6129 goto exit_kernel;
6130 }
6131 }
6132
6133 /* Create session loading thread. */
6134 ret = pthread_create(&load_session_thread, NULL, thread_load_session,
6135 load_info);
6136 if (ret) {
6137 errno = ret;
6138 PERROR("pthread_create load_session_thread");
6139 retval = -1;
6140 goto exit_load_session;
6141 }
6142
6143 /*
6144 * This is where we start awaiting program completion (e.g. through
6145 * signal that asks threads to teardown).
6146 */
6147
6148 ret = pthread_join(load_session_thread, &status);
6149 if (ret) {
6150 errno = ret;
6151 PERROR("pthread_join load_session_thread");
6152 retval = -1;
6153 }
6154 exit_load_session:
6155
6156 if (is_root && !opt_no_kernel) {
6157 ret = pthread_join(kernel_thread, &status);
6158 if (ret) {
6159 errno = ret;
6160 PERROR("pthread_join");
6161 retval = -1;
6162 }
6163 }
6164 exit_kernel:
6165
6166 ret = pthread_join(agent_reg_thread, &status);
6167 if (ret) {
6168 errno = ret;
6169 PERROR("pthread_join agent");
6170 retval = -1;
6171 }
6172 exit_agent_reg:
6173
6174 ret = pthread_join(apps_notify_thread, &status);
6175 if (ret) {
6176 errno = ret;
6177 PERROR("pthread_join apps notify");
6178 retval = -1;
6179 }
6180 exit_apps_notify:
6181
6182 ret = pthread_join(apps_thread, &status);
6183 if (ret) {
6184 errno = ret;
6185 PERROR("pthread_join apps");
6186 retval = -1;
6187 }
6188 exit_apps:
6189
6190 ret = pthread_join(reg_apps_thread, &status);
6191 if (ret) {
6192 errno = ret;
6193 PERROR("pthread_join");
6194 retval = -1;
6195 }
6196 exit_reg_apps:
6197
6198 /*
6199 * Join dispatch thread after joining reg_apps_thread to ensure
6200 * we don't leak applications in the queue.
6201 */
6202 ret = pthread_join(dispatch_thread, &status);
6203 if (ret) {
6204 errno = ret;
6205 PERROR("pthread_join");
6206 retval = -1;
6207 }
6208 exit_dispatch:
6209
6210 ret = pthread_join(client_thread, &status);
6211 if (ret) {
6212 errno = ret;
6213 PERROR("pthread_join");
6214 retval = -1;
6215 }
6216 exit_client:
6217
6218 ret = pthread_join(health_thread, &status);
6219 if (ret) {
6220 errno = ret;
6221 PERROR("pthread_join health thread");
6222 retval = -1;
6223 }
6224 exit_health:
6225
6226 exit_init_data:
6227 /*
6228 * Wait for all pending call_rcu work to complete before tearing
6229 * down data structures. call_rcu worker may be trying to
6230 * perform lookups in those structures.
6231 */
6232 rcu_barrier();
6233 /*
6234 * sessiond_cleanup() is called when no other thread is running, except
6235 * the ht_cleanup thread, which is needed to destroy the hash tables.
6236 */
6237 rcu_thread_online();
6238 sessiond_cleanup();
6239 rcu_thread_offline();
6240 rcu_unregister_thread();
6241
6242 /*
6243 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6244 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6245 * the queue is empty before shutting down the clean-up thread.
6246 */
6247 rcu_barrier();
6248
6249 ret = notify_thread_pipe(ht_cleanup_quit_pipe[1]);
6250 if (ret < 0) {
6251 ERR("write error on ht_cleanup quit pipe");
6252 retval = -1;
6253 }
6254
6255 ret = pthread_join(ht_cleanup_thread, &status);
6256 if (ret) {
6257 errno = ret;
6258 PERROR("pthread_join ht cleanup thread");
6259 retval = -1;
6260 }
6261 exit_ht_cleanup:
6262 exit_set_max_size:
6263
6264 utils_close_pipe(ht_cleanup_pipe);
6265 exit_ht_cleanup_pipe:
6266
6267 /*
6268 * Close the ht_cleanup quit pipe.
6269 */
6270 utils_close_pipe(ht_cleanup_quit_pipe);
6271 exit_ht_cleanup_quit_pipe:
6272
6273 health_app_destroy(health_sessiond);
6274 exit_health_sessiond_cleanup:
6275 exit_create_run_as_worker_cleanup:
6276
6277 exit_options:
6278 sessiond_cleanup_options();
6279
6280 exit_set_signal_handler:
6281
6282 if (!retval) {
6283 exit(EXIT_SUCCESS);
6284 } else {
6285 exit(EXIT_FAILURE);
6286 }
6287 }
This page took 0.219441 seconds and 4 git commands to generate.