Cleanup: Remove fun from code base
[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 #include "ht-cleanup.h"
77
78 #define CONSUMERD_FILE "lttng-consumerd"
79
80 const char *progname;
81 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
82 static int tracing_group_name_override;
83 static char *opt_pidfile;
84 static int opt_sig_parent;
85 static int opt_verbose_consumer;
86 static int opt_daemon, opt_background;
87 static int opt_no_kernel;
88 static char *opt_load_session_path;
89 static pid_t ppid; /* Parent PID for --sig-parent option */
90 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
91 static char *rundir;
92 static int lockfile_fd = -1;
93
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal;
96
97 /*
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
100 */
101 static struct consumer_data kconsumer_data = {
102 .type = LTTNG_CONSUMER_KERNEL,
103 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
104 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
105 .err_sock = -1,
106 .cmd_sock = -1,
107 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
108 .lock = PTHREAD_MUTEX_INITIALIZER,
109 .cond = PTHREAD_COND_INITIALIZER,
110 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
111 };
112 static struct consumer_data ustconsumer64_data = {
113 .type = LTTNG_CONSUMER64_UST,
114 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
115 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
116 .err_sock = -1,
117 .cmd_sock = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123 static struct consumer_data ustconsumer32_data = {
124 .type = LTTNG_CONSUMER32_UST,
125 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
126 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
127 .err_sock = -1,
128 .cmd_sock = -1,
129 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
130 .lock = PTHREAD_MUTEX_INITIALIZER,
131 .cond = PTHREAD_COND_INITIALIZER,
132 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
133 };
134
135 /* Command line options */
136 static const struct option long_options[] = {
137 { "client-sock", required_argument, 0, 'c' },
138 { "apps-sock", required_argument, 0, 'a' },
139 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
140 { "kconsumerd-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
144 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
145 { "consumerd32-path", required_argument, 0, '\0' },
146 { "consumerd32-libdir", required_argument, 0, '\0' },
147 { "consumerd64-path", required_argument, 0, '\0' },
148 { "consumerd64-libdir", required_argument, 0, '\0' },
149 { "daemonize", no_argument, 0, 'd' },
150 { "background", no_argument, 0, 'b' },
151 { "sig-parent", no_argument, 0, 'S' },
152 { "help", no_argument, 0, 'h' },
153 { "group", required_argument, 0, 'g' },
154 { "version", no_argument, 0, 'V' },
155 { "quiet", no_argument, 0, 'q' },
156 { "verbose", no_argument, 0, 'v' },
157 { "verbose-consumer", no_argument, 0, '\0' },
158 { "no-kernel", no_argument, 0, '\0' },
159 { "pidfile", required_argument, 0, 'p' },
160 { "agent-tcp-port", required_argument, 0, '\0' },
161 { "config", required_argument, 0, 'f' },
162 { "load", required_argument, 0, 'l' },
163 { "kmod-probes", required_argument, 0, '\0' },
164 { "extra-kmod-probes", required_argument, 0, '\0' },
165 { NULL, 0, 0, 0 }
166 };
167
168 /* Command line options to ignore from configuration file */
169 static const char *config_ignore_options[] = { "help", "version", "config" };
170
171 /* Shared between threads */
172 static int dispatch_thread_exit;
173
174 /* Global application Unix socket path */
175 static char apps_unix_sock_path[PATH_MAX];
176 /* Global client Unix socket path */
177 static char client_unix_sock_path[PATH_MAX];
178 /* global wait shm path for UST */
179 static char wait_shm_path[PATH_MAX];
180 /* Global health check unix path */
181 static char health_unix_sock_path[PATH_MAX];
182
183 /* Sockets and FDs */
184 static int client_sock = -1;
185 static int apps_sock = -1;
186 int kernel_tracer_fd = -1;
187 static int kernel_poll_pipe[2] = { -1, -1 };
188
189 /*
190 * Quit pipe for all threads. This permits a single cancellation point
191 * for all threads when receiving an event on the pipe.
192 */
193 static int thread_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 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
322 }
323
324 /* Notify parents that we are ready for cmd and health check */
325 LTTNG_HIDDEN
326 void sessiond_notify_ready(void)
327 {
328 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
329 /*
330 * Notify parent pid that we are ready to accept command
331 * for client side. This ppid is the one from the
332 * external process that spawned us.
333 */
334 if (opt_sig_parent) {
335 kill(ppid, SIGUSR1);
336 }
337
338 /*
339 * Notify the parent of the fork() process that we are
340 * ready.
341 */
342 if (opt_daemon || opt_background) {
343 kill(child_ppid, SIGUSR1);
344 }
345 }
346 }
347
348 static
349 void setup_consumerd_path(void)
350 {
351 const char *bin, *libdir;
352
353 /*
354 * Allow INSTALL_BIN_PATH to be used as a target path for the
355 * native architecture size consumer if CONFIG_CONSUMER*_PATH
356 * has not been defined.
357 */
358 #if (CAA_BITS_PER_LONG == 32)
359 if (!consumerd32_bin[0]) {
360 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
361 }
362 if (!consumerd32_libdir[0]) {
363 consumerd32_libdir = INSTALL_LIB_PATH;
364 }
365 #elif (CAA_BITS_PER_LONG == 64)
366 if (!consumerd64_bin[0]) {
367 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
368 }
369 if (!consumerd64_libdir[0]) {
370 consumerd64_libdir = INSTALL_LIB_PATH;
371 }
372 #else
373 #error "Unknown bitness"
374 #endif
375
376 /*
377 * runtime env. var. overrides the build default.
378 */
379 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
380 if (bin) {
381 consumerd32_bin = bin;
382 }
383 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
384 if (bin) {
385 consumerd64_bin = bin;
386 }
387 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
388 if (libdir) {
389 consumerd32_libdir = libdir;
390 }
391 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
392 if (libdir) {
393 consumerd64_libdir = libdir;
394 }
395 }
396
397 static
398 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
399 int *a_pipe)
400 {
401 int ret;
402
403 assert(events);
404
405 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
406 if (ret < 0) {
407 goto error;
408 }
409
410 /* Add quit pipe */
411 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
412 if (ret < 0) {
413 goto error;
414 }
415
416 return 0;
417
418 error:
419 return ret;
420 }
421
422 /*
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 */
425 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
426 {
427 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
428 }
429
430 /*
431 * Init thread quit pipe.
432 *
433 * Return -1 on error or 0 if all pipes are created.
434 */
435 static int __init_thread_quit_pipe(int *a_pipe)
436 {
437 int ret, i;
438
439 ret = pipe(a_pipe);
440 if (ret < 0) {
441 PERROR("thread quit pipe");
442 goto error;
443 }
444
445 for (i = 0; i < 2; i++) {
446 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
447 if (ret < 0) {
448 PERROR("fcntl");
449 goto error;
450 }
451 }
452
453 error:
454 return ret;
455 }
456
457 static int init_thread_quit_pipe(void)
458 {
459 return __init_thread_quit_pipe(thread_quit_pipe);
460 }
461
462 /*
463 * Stop all threads by closing the thread quit pipe.
464 */
465 static void stop_threads(void)
466 {
467 int ret;
468
469 /* Stopping all threads */
470 DBG("Terminating all threads");
471 ret = notify_thread_pipe(thread_quit_pipe[1]);
472 if (ret < 0) {
473 ERR("write error on thread quit pipe");
474 }
475
476 /* Dispatch thread */
477 CMM_STORE_SHARED(dispatch_thread_exit, 1);
478 futex_nto1_wake(&ust_cmd_queue.futex);
479 }
480
481 /*
482 * Close every consumer sockets.
483 */
484 static void close_consumer_sockets(void)
485 {
486 int ret;
487
488 if (kconsumer_data.err_sock >= 0) {
489 ret = close(kconsumer_data.err_sock);
490 if (ret < 0) {
491 PERROR("kernel consumer err_sock close");
492 }
493 }
494 if (ustconsumer32_data.err_sock >= 0) {
495 ret = close(ustconsumer32_data.err_sock);
496 if (ret < 0) {
497 PERROR("UST consumerd32 err_sock close");
498 }
499 }
500 if (ustconsumer64_data.err_sock >= 0) {
501 ret = close(ustconsumer64_data.err_sock);
502 if (ret < 0) {
503 PERROR("UST consumerd64 err_sock close");
504 }
505 }
506 if (kconsumer_data.cmd_sock >= 0) {
507 ret = close(kconsumer_data.cmd_sock);
508 if (ret < 0) {
509 PERROR("kernel consumer cmd_sock close");
510 }
511 }
512 if (ustconsumer32_data.cmd_sock >= 0) {
513 ret = close(ustconsumer32_data.cmd_sock);
514 if (ret < 0) {
515 PERROR("UST consumerd32 cmd_sock close");
516 }
517 }
518 if (ustconsumer64_data.cmd_sock >= 0) {
519 ret = close(ustconsumer64_data.cmd_sock);
520 if (ret < 0) {
521 PERROR("UST consumerd64 cmd_sock close");
522 }
523 }
524 }
525
526 /*
527 * Generate the full lock file path using the rundir.
528 *
529 * Return the snprintf() return value thus a negative value is an error.
530 */
531 static int generate_lock_file_path(char *path, size_t len)
532 {
533 int ret;
534
535 assert(path);
536 assert(rundir);
537
538 /* Build lockfile path from rundir. */
539 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
540 if (ret < 0) {
541 PERROR("snprintf lockfile path");
542 }
543
544 return ret;
545 }
546
547 /*
548 * Wait on consumer process termination.
549 *
550 * Need to be called with the consumer data lock held or from a context
551 * ensuring no concurrent access to data (e.g: cleanup).
552 */
553 static void wait_consumer(struct consumer_data *consumer_data)
554 {
555 pid_t ret;
556 int status;
557
558 if (consumer_data->pid <= 0) {
559 return;
560 }
561
562 DBG("Waiting for complete teardown of consumerd (PID: %d)",
563 consumer_data->pid);
564 ret = waitpid(consumer_data->pid, &status, 0);
565 if (ret == -1) {
566 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
567 }
568 if (!WIFEXITED(status)) {
569 ERR("consumerd termination with error: %d",
570 WEXITSTATUS(ret));
571 }
572 consumer_data->pid = 0;
573 }
574
575 /*
576 * Cleanup the session daemon's data structures.
577 */
578 static void sessiond_cleanup(void)
579 {
580 int ret;
581 struct ltt_session *sess, *stmp;
582 char path[PATH_MAX];
583
584 DBG("Cleanup sessiond");
585
586 /*
587 * Close the thread quit pipe. It has already done its job,
588 * since we are now called.
589 */
590 utils_close_pipe(thread_quit_pipe);
591
592 /*
593 * If opt_pidfile is undefined, the default file will be wiped when
594 * removing the rundir.
595 */
596 if (opt_pidfile) {
597 ret = remove(opt_pidfile);
598 if (ret < 0) {
599 PERROR("remove pidfile %s", opt_pidfile);
600 }
601 }
602
603 DBG("Removing sessiond and consumerd content of directory %s", rundir);
604
605 /* sessiond */
606 snprintf(path, PATH_MAX,
607 "%s/%s",
608 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
609 DBG("Removing %s", path);
610 (void) unlink(path);
611
612 snprintf(path, PATH_MAX, "%s/%s", rundir,
613 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
614 DBG("Removing %s", path);
615 (void) unlink(path);
616
617 /* kconsumerd */
618 snprintf(path, PATH_MAX,
619 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
620 rundir);
621 DBG("Removing %s", path);
622 (void) unlink(path);
623
624 snprintf(path, PATH_MAX,
625 DEFAULT_KCONSUMERD_PATH,
626 rundir);
627 DBG("Removing directory %s", path);
628 (void) rmdir(path);
629
630 /* ust consumerd 32 */
631 snprintf(path, PATH_MAX,
632 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
633 rundir);
634 DBG("Removing %s", path);
635 (void) unlink(path);
636
637 snprintf(path, PATH_MAX,
638 DEFAULT_USTCONSUMERD32_PATH,
639 rundir);
640 DBG("Removing directory %s", path);
641 (void) rmdir(path);
642
643 /* ust consumerd 64 */
644 snprintf(path, PATH_MAX,
645 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
646 rundir);
647 DBG("Removing %s", path);
648 (void) unlink(path);
649
650 snprintf(path, PATH_MAX,
651 DEFAULT_USTCONSUMERD64_PATH,
652 rundir);
653 DBG("Removing directory %s", path);
654 (void) rmdir(path);
655
656 DBG("Cleaning up all sessions");
657
658 /* Destroy session list mutex */
659 if (session_list_ptr != NULL) {
660 pthread_mutex_destroy(&session_list_ptr->lock);
661
662 /* Cleanup ALL session */
663 cds_list_for_each_entry_safe(sess, stmp,
664 &session_list_ptr->head, list) {
665 cmd_destroy_session(sess, kernel_poll_pipe[1]);
666 }
667 }
668
669 wait_consumer(&kconsumer_data);
670 wait_consumer(&ustconsumer64_data);
671 wait_consumer(&ustconsumer32_data);
672
673 DBG("Cleaning up all agent apps");
674 agent_app_ht_clean();
675
676 DBG("Closing all UST sockets");
677 ust_app_clean_list();
678 buffer_reg_destroy_registries();
679
680 if (is_root && !opt_no_kernel) {
681 DBG2("Closing kernel fd");
682 if (kernel_tracer_fd >= 0) {
683 ret = close(kernel_tracer_fd);
684 if (ret) {
685 PERROR("close");
686 }
687 }
688 DBG("Unloading kernel modules");
689 modprobe_remove_lttng_all();
690 free(syscall_table);
691 }
692
693 close_consumer_sockets();
694
695 if (load_info) {
696 load_session_destroy_data(load_info);
697 free(load_info);
698 }
699
700 /*
701 * Cleanup lock file by deleting it and finaly closing it which will
702 * release the file system lock.
703 */
704 if (lockfile_fd >= 0) {
705 char lockfile_path[PATH_MAX];
706
707 ret = generate_lock_file_path(lockfile_path,
708 sizeof(lockfile_path));
709 if (ret > 0) {
710 ret = remove(lockfile_path);
711 if (ret < 0) {
712 PERROR("remove lock file");
713 }
714 ret = close(lockfile_fd);
715 if (ret < 0) {
716 PERROR("close lock file");
717 }
718 }
719 }
720
721 /*
722 * We do NOT rmdir rundir because there are other processes
723 * using it, for instance lttng-relayd, which can start in
724 * parallel with this teardown.
725 */
726
727 free(rundir);
728 }
729
730 /*
731 * Cleanup the daemon's option data structures.
732 */
733 static void sessiond_cleanup_options(void)
734 {
735 DBG("Cleaning up options");
736
737 /*
738 * If the override option is set, the pointer points to a *non* const
739 * thus freeing it even though the variable type is set to const.
740 */
741 if (tracing_group_name_override) {
742 free((void *) tracing_group_name);
743 }
744 if (consumerd32_bin_override) {
745 free((void *) consumerd32_bin);
746 }
747 if (consumerd64_bin_override) {
748 free((void *) consumerd64_bin);
749 }
750 if (consumerd32_libdir_override) {
751 free((void *) consumerd32_libdir);
752 }
753 if (consumerd64_libdir_override) {
754 free((void *) consumerd64_libdir);
755 }
756
757 free(opt_pidfile);
758 free(opt_load_session_path);
759 free(kmod_probes_list);
760 free(kmod_extra_probes_list);
761
762 run_as_destroy_worker();
763 }
764
765 /*
766 * Send data on a unix socket using the liblttsessiondcomm API.
767 *
768 * Return lttcomm error code.
769 */
770 static int send_unix_sock(int sock, void *buf, size_t len)
771 {
772 /* Check valid length */
773 if (len == 0) {
774 return -1;
775 }
776
777 return lttcomm_send_unix_sock(sock, buf, len);
778 }
779
780 /*
781 * Free memory of a command context structure.
782 */
783 static void clean_command_ctx(struct command_ctx **cmd_ctx)
784 {
785 DBG("Clean command context structure");
786 if (*cmd_ctx) {
787 if ((*cmd_ctx)->llm) {
788 free((*cmd_ctx)->llm);
789 }
790 if ((*cmd_ctx)->lsm) {
791 free((*cmd_ctx)->lsm);
792 }
793 free(*cmd_ctx);
794 *cmd_ctx = NULL;
795 }
796 }
797
798 /*
799 * Notify UST applications using the shm mmap futex.
800 */
801 static int notify_ust_apps(int active)
802 {
803 char *wait_shm_mmap;
804
805 DBG("Notifying applications of session daemon state: %d", active);
806
807 /* See shm.c for this call implying mmap, shm and futex calls */
808 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
809 if (wait_shm_mmap == NULL) {
810 goto error;
811 }
812
813 /* Wake waiting process */
814 futex_wait_update((int32_t *) wait_shm_mmap, active);
815
816 /* Apps notified successfully */
817 return 0;
818
819 error:
820 return -1;
821 }
822
823 /*
824 * Setup the outgoing data buffer for the response (llm) by allocating the
825 * right amount of memory and copying the original information from the lsm
826 * structure.
827 *
828 * Return 0 on success, negative value on error.
829 */
830 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
831 const void *payload_buf, size_t payload_len,
832 const void *cmd_header_buf, size_t cmd_header_len)
833 {
834 int ret = 0;
835 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
836 const size_t cmd_header_offset = header_len;
837 const size_t payload_offset = cmd_header_offset + cmd_header_len;
838 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
839
840 cmd_ctx->llm = zmalloc(total_msg_size);
841
842 if (cmd_ctx->llm == NULL) {
843 PERROR("zmalloc");
844 ret = -ENOMEM;
845 goto end;
846 }
847
848 /* Copy common data */
849 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
850 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
851 cmd_ctx->llm->cmd_header_size = cmd_header_len;
852 cmd_ctx->llm->data_size = payload_len;
853 cmd_ctx->lttng_msg_size = total_msg_size;
854
855 /* Copy command header */
856 if (cmd_header_len) {
857 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
858 cmd_header_len);
859 }
860
861 /* Copy payload */
862 if (payload_len) {
863 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
864 payload_len);
865 }
866
867 end:
868 return ret;
869 }
870
871 /*
872 * Version of setup_lttng_msg() without command header.
873 */
874 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
875 void *payload_buf, size_t payload_len)
876 {
877 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
878 }
879 /*
880 * Update the kernel poll set of all channel fd available over all tracing
881 * session. Add the wakeup pipe at the end of the set.
882 */
883 static int update_kernel_poll(struct lttng_poll_event *events)
884 {
885 int ret;
886 struct ltt_session *session;
887 struct ltt_kernel_channel *channel;
888
889 DBG("Updating kernel poll set");
890
891 session_lock_list();
892 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
893 session_lock(session);
894 if (session->kernel_session == NULL) {
895 session_unlock(session);
896 continue;
897 }
898
899 cds_list_for_each_entry(channel,
900 &session->kernel_session->channel_list.head, list) {
901 /* Add channel fd to the kernel poll set */
902 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
903 if (ret < 0) {
904 session_unlock(session);
905 goto error;
906 }
907 DBG("Channel fd %d added to kernel set", channel->fd);
908 }
909 session_unlock(session);
910 }
911 session_unlock_list();
912
913 return 0;
914
915 error:
916 session_unlock_list();
917 return -1;
918 }
919
920 /*
921 * Find the channel fd from 'fd' over all tracing session. When found, check
922 * for new channel stream and send those stream fds to the kernel consumer.
923 *
924 * Useful for CPU hotplug feature.
925 */
926 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
927 {
928 int ret = 0;
929 struct ltt_session *session;
930 struct ltt_kernel_session *ksess;
931 struct ltt_kernel_channel *channel;
932
933 DBG("Updating kernel streams for channel fd %d", fd);
934
935 session_lock_list();
936 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
937 session_lock(session);
938 if (session->kernel_session == NULL) {
939 session_unlock(session);
940 continue;
941 }
942 ksess = session->kernel_session;
943
944 cds_list_for_each_entry(channel,
945 &ksess->channel_list.head, list) {
946 struct lttng_ht_iter iter;
947 struct consumer_socket *socket;
948
949 if (channel->fd != fd) {
950 continue;
951 }
952 DBG("Channel found, updating kernel streams");
953 ret = kernel_open_channel_stream(channel);
954 if (ret < 0) {
955 goto error;
956 }
957 /* Update the stream global counter */
958 ksess->stream_count_global += ret;
959
960 /*
961 * Have we already sent fds to the consumer? If yes, it
962 * means that tracing is started so it is safe to send
963 * our updated stream fds.
964 */
965 if (ksess->consumer_fds_sent != 1
966 || ksess->consumer == NULL) {
967 ret = -1;
968 goto error;
969 }
970
971 rcu_read_lock();
972 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
973 &iter.iter, socket, node.node) {
974 pthread_mutex_lock(socket->lock);
975 ret = kernel_consumer_send_channel_stream(socket,
976 channel, ksess,
977 session->output_traces ? 1 : 0);
978 pthread_mutex_unlock(socket->lock);
979 if (ret < 0) {
980 rcu_read_unlock();
981 goto error;
982 }
983 }
984 rcu_read_unlock();
985 }
986 session_unlock(session);
987 }
988 session_unlock_list();
989 return ret;
990
991 error:
992 session_unlock(session);
993 session_unlock_list();
994 return ret;
995 }
996
997 /*
998 * For each tracing session, update newly registered apps. The session list
999 * lock MUST be acquired before calling this.
1000 */
1001 static void update_ust_app(int app_sock)
1002 {
1003 struct ltt_session *sess, *stmp;
1004
1005 /* Consumer is in an ERROR state. Stop any application update. */
1006 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1007 /* Stop the update process since the consumer is dead. */
1008 return;
1009 }
1010
1011 /* For all tracing session(s) */
1012 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1013 struct ust_app *app;
1014
1015 session_lock(sess);
1016 if (!sess->ust_session) {
1017 goto unlock_session;
1018 }
1019
1020 rcu_read_lock();
1021 assert(app_sock >= 0);
1022 app = ust_app_find_by_sock(app_sock);
1023 if (app == NULL) {
1024 /*
1025 * Application can be unregistered before so
1026 * this is possible hence simply stopping the
1027 * update.
1028 */
1029 DBG3("UST app update failed to find app sock %d",
1030 app_sock);
1031 goto unlock_rcu;
1032 }
1033 ust_app_global_update(sess->ust_session, app);
1034 unlock_rcu:
1035 rcu_read_unlock();
1036 unlock_session:
1037 session_unlock(sess);
1038 }
1039 }
1040
1041 /*
1042 * This thread manage event coming from the kernel.
1043 *
1044 * Features supported in this thread:
1045 * -) CPU Hotplug
1046 */
1047 static void *thread_manage_kernel(void *data)
1048 {
1049 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1050 uint32_t revents, nb_fd;
1051 char tmp;
1052 struct lttng_poll_event events;
1053
1054 DBG("[thread] Thread manage kernel started");
1055
1056 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1057
1058 /*
1059 * This first step of the while is to clean this structure which could free
1060 * non NULL pointers so initialize it before the loop.
1061 */
1062 lttng_poll_init(&events);
1063
1064 if (testpoint(sessiond_thread_manage_kernel)) {
1065 goto error_testpoint;
1066 }
1067
1068 health_code_update();
1069
1070 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1071 goto error_testpoint;
1072 }
1073
1074 while (1) {
1075 health_code_update();
1076
1077 if (update_poll_flag == 1) {
1078 /* Clean events object. We are about to populate it again. */
1079 lttng_poll_clean(&events);
1080
1081 ret = sessiond_set_thread_pollset(&events, 2);
1082 if (ret < 0) {
1083 goto error_poll_create;
1084 }
1085
1086 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1087 if (ret < 0) {
1088 goto error;
1089 }
1090
1091 /* This will add the available kernel channel if any. */
1092 ret = update_kernel_poll(&events);
1093 if (ret < 0) {
1094 goto error;
1095 }
1096 update_poll_flag = 0;
1097 }
1098
1099 DBG("Thread kernel polling");
1100
1101 /* Poll infinite value of time */
1102 restart:
1103 health_poll_entry();
1104 ret = lttng_poll_wait(&events, -1);
1105 DBG("Thread kernel return from poll on %d fds",
1106 LTTNG_POLL_GETNB(&events));
1107 health_poll_exit();
1108 if (ret < 0) {
1109 /*
1110 * Restart interrupted system call.
1111 */
1112 if (errno == EINTR) {
1113 goto restart;
1114 }
1115 goto error;
1116 } else if (ret == 0) {
1117 /* Should not happen since timeout is infinite */
1118 ERR("Return value of poll is 0 with an infinite timeout.\n"
1119 "This should not have happened! Continuing...");
1120 continue;
1121 }
1122
1123 nb_fd = ret;
1124
1125 for (i = 0; i < nb_fd; i++) {
1126 /* Fetch once the poll data */
1127 revents = LTTNG_POLL_GETEV(&events, i);
1128 pollfd = LTTNG_POLL_GETFD(&events, i);
1129
1130 health_code_update();
1131
1132 if (!revents) {
1133 /* No activity for this FD (poll implementation). */
1134 continue;
1135 }
1136
1137 /* Thread quit pipe has been closed. Killing thread. */
1138 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1139 if (ret) {
1140 err = 0;
1141 goto exit;
1142 }
1143
1144 /* Check for data on kernel pipe */
1145 if (revents & LPOLLIN) {
1146 if (pollfd == kernel_poll_pipe[0]) {
1147 (void) lttng_read(kernel_poll_pipe[0],
1148 &tmp, 1);
1149 /*
1150 * Ret value is useless here, if this pipe gets any actions an
1151 * update is required anyway.
1152 */
1153 update_poll_flag = 1;
1154 continue;
1155 } else {
1156 /*
1157 * New CPU detected by the kernel. Adding kernel stream to
1158 * kernel session and updating the kernel consumer
1159 */
1160 ret = update_kernel_stream(&kconsumer_data, pollfd);
1161 if (ret < 0) {
1162 continue;
1163 }
1164 break;
1165 }
1166 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1167 update_poll_flag = 1;
1168 continue;
1169 } else {
1170 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1171 goto error;
1172 }
1173 }
1174 }
1175
1176 exit:
1177 error:
1178 lttng_poll_clean(&events);
1179 error_poll_create:
1180 error_testpoint:
1181 utils_close_pipe(kernel_poll_pipe);
1182 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1183 if (err) {
1184 health_error();
1185 ERR("Health error occurred in %s", __func__);
1186 WARN("Kernel thread died unexpectedly. "
1187 "Kernel tracing can continue but CPU hotplug is disabled.");
1188 }
1189 health_unregister(health_sessiond);
1190 DBG("Kernel thread dying");
1191 return NULL;
1192 }
1193
1194 /*
1195 * Signal pthread condition of the consumer data that the thread.
1196 */
1197 static void signal_consumer_condition(struct consumer_data *data, int state)
1198 {
1199 pthread_mutex_lock(&data->cond_mutex);
1200
1201 /*
1202 * The state is set before signaling. It can be any value, it's the waiter
1203 * job to correctly interpret this condition variable associated to the
1204 * consumer pthread_cond.
1205 *
1206 * A value of 0 means that the corresponding thread of the consumer data
1207 * was not started. 1 indicates that the thread has started and is ready
1208 * for action. A negative value means that there was an error during the
1209 * thread bootstrap.
1210 */
1211 data->consumer_thread_is_ready = state;
1212 (void) pthread_cond_signal(&data->cond);
1213
1214 pthread_mutex_unlock(&data->cond_mutex);
1215 }
1216
1217 /*
1218 * This thread manage the consumer error sent back to the session daemon.
1219 */
1220 static void *thread_manage_consumer(void *data)
1221 {
1222 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1223 uint32_t revents, nb_fd;
1224 enum lttcomm_return_code code;
1225 struct lttng_poll_event events;
1226 struct consumer_data *consumer_data = data;
1227
1228 DBG("[thread] Manage consumer started");
1229
1230 rcu_register_thread();
1231 rcu_thread_online();
1232
1233 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1234
1235 health_code_update();
1236
1237 /*
1238 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1239 * metadata_sock. Nothing more will be added to this poll set.
1240 */
1241 ret = sessiond_set_thread_pollset(&events, 3);
1242 if (ret < 0) {
1243 goto error_poll;
1244 }
1245
1246 /*
1247 * The error socket here is already in a listening state which was done
1248 * just before spawning this thread to avoid a race between the consumer
1249 * daemon exec trying to connect and the listen() call.
1250 */
1251 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1252 if (ret < 0) {
1253 goto error;
1254 }
1255
1256 health_code_update();
1257
1258 /* Infinite blocking call, waiting for transmission */
1259 restart:
1260 health_poll_entry();
1261
1262 if (testpoint(sessiond_thread_manage_consumer)) {
1263 goto error;
1264 }
1265
1266 ret = lttng_poll_wait(&events, -1);
1267 health_poll_exit();
1268 if (ret < 0) {
1269 /*
1270 * Restart interrupted system call.
1271 */
1272 if (errno == EINTR) {
1273 goto restart;
1274 }
1275 goto error;
1276 }
1277
1278 nb_fd = ret;
1279
1280 for (i = 0; i < nb_fd; i++) {
1281 /* Fetch once the poll data */
1282 revents = LTTNG_POLL_GETEV(&events, i);
1283 pollfd = LTTNG_POLL_GETFD(&events, i);
1284
1285 health_code_update();
1286
1287 if (!revents) {
1288 /* No activity for this FD (poll implementation). */
1289 continue;
1290 }
1291
1292 /* Thread quit pipe has been closed. Killing thread. */
1293 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1294 if (ret) {
1295 err = 0;
1296 goto exit;
1297 }
1298
1299 /* Event on the registration socket */
1300 if (pollfd == consumer_data->err_sock) {
1301 if (revents & LPOLLIN) {
1302 continue;
1303 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1304 ERR("consumer err socket poll error");
1305 goto error;
1306 } else {
1307 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1308 goto error;
1309 }
1310 }
1311 }
1312
1313 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1314 if (sock < 0) {
1315 goto error;
1316 }
1317
1318 /*
1319 * Set the CLOEXEC flag. Return code is useless because either way, the
1320 * show must go on.
1321 */
1322 (void) utils_set_fd_cloexec(sock);
1323
1324 health_code_update();
1325
1326 DBG2("Receiving code from consumer err_sock");
1327
1328 /* Getting status code from kconsumerd */
1329 ret = lttcomm_recv_unix_sock(sock, &code,
1330 sizeof(enum lttcomm_return_code));
1331 if (ret <= 0) {
1332 goto error;
1333 }
1334
1335 health_code_update();
1336 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1337 /* Connect both socket, command and metadata. */
1338 consumer_data->cmd_sock =
1339 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1340 consumer_data->metadata_fd =
1341 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1342 if (consumer_data->cmd_sock < 0
1343 || consumer_data->metadata_fd < 0) {
1344 PERROR("consumer connect cmd socket");
1345 /* On error, signal condition and quit. */
1346 signal_consumer_condition(consumer_data, -1);
1347 goto error;
1348 }
1349 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1350 /* Create metadata socket lock. */
1351 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1352 if (consumer_data->metadata_sock.lock == NULL) {
1353 PERROR("zmalloc pthread mutex");
1354 goto error;
1355 }
1356 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1357
1358 signal_consumer_condition(consumer_data, 1);
1359 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1360 DBG("Consumer metadata socket ready (fd: %d)",
1361 consumer_data->metadata_fd);
1362 } else {
1363 ERR("consumer error when waiting for SOCK_READY : %s",
1364 lttcomm_get_readable_code(-code));
1365 goto error;
1366 }
1367
1368 /* Remove the consumerd error sock since we've established a connexion */
1369 ret = lttng_poll_del(&events, consumer_data->err_sock);
1370 if (ret < 0) {
1371 goto error;
1372 }
1373
1374 /* Add new accepted error socket. */
1375 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1376 if (ret < 0) {
1377 goto error;
1378 }
1379
1380 /* Add metadata socket that is successfully connected. */
1381 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1382 LPOLLIN | LPOLLRDHUP);
1383 if (ret < 0) {
1384 goto error;
1385 }
1386
1387 health_code_update();
1388
1389 /* Infinite blocking call, waiting for transmission */
1390 restart_poll:
1391 while (1) {
1392 health_code_update();
1393
1394 /* Exit the thread because the thread quit pipe has been triggered. */
1395 if (should_quit) {
1396 /* Not a health error. */
1397 err = 0;
1398 goto exit;
1399 }
1400
1401 health_poll_entry();
1402 ret = lttng_poll_wait(&events, -1);
1403 health_poll_exit();
1404 if (ret < 0) {
1405 /*
1406 * Restart interrupted system call.
1407 */
1408 if (errno == EINTR) {
1409 goto restart_poll;
1410 }
1411 goto error;
1412 }
1413
1414 nb_fd = ret;
1415
1416 for (i = 0; i < nb_fd; i++) {
1417 /* Fetch once the poll data */
1418 revents = LTTNG_POLL_GETEV(&events, i);
1419 pollfd = LTTNG_POLL_GETFD(&events, i);
1420
1421 health_code_update();
1422
1423 if (!revents) {
1424 /* No activity for this FD (poll implementation). */
1425 continue;
1426 }
1427
1428 /*
1429 * Thread quit pipe has been triggered, flag that we should stop
1430 * but continue the current loop to handle potential data from
1431 * consumer.
1432 */
1433 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1434
1435 if (pollfd == sock) {
1436 /* Event on the consumerd socket */
1437 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1438 && !(revents & LPOLLIN)) {
1439 ERR("consumer err socket second poll error");
1440 goto error;
1441 }
1442 health_code_update();
1443 /* Wait for any kconsumerd error */
1444 ret = lttcomm_recv_unix_sock(sock, &code,
1445 sizeof(enum lttcomm_return_code));
1446 if (ret <= 0) {
1447 ERR("consumer closed the command socket");
1448 goto error;
1449 }
1450
1451 ERR("consumer return code : %s",
1452 lttcomm_get_readable_code(-code));
1453
1454 goto exit;
1455 } else if (pollfd == consumer_data->metadata_fd) {
1456 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1457 && !(revents & LPOLLIN)) {
1458 ERR("consumer err metadata socket second poll error");
1459 goto error;
1460 }
1461 /* UST metadata requests */
1462 ret = ust_consumer_metadata_request(
1463 &consumer_data->metadata_sock);
1464 if (ret < 0) {
1465 ERR("Handling metadata request");
1466 goto error;
1467 }
1468 }
1469 /* No need for an else branch all FDs are tested prior. */
1470 }
1471 health_code_update();
1472 }
1473
1474 exit:
1475 error:
1476 /*
1477 * We lock here because we are about to close the sockets and some other
1478 * thread might be using them so get exclusive access which will abort all
1479 * other consumer command by other threads.
1480 */
1481 pthread_mutex_lock(&consumer_data->lock);
1482
1483 /* Immediately set the consumerd state to stopped */
1484 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1485 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1486 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1487 consumer_data->type == LTTNG_CONSUMER32_UST) {
1488 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1489 } else {
1490 /* Code flow error... */
1491 assert(0);
1492 }
1493
1494 if (consumer_data->err_sock >= 0) {
1495 ret = close(consumer_data->err_sock);
1496 if (ret) {
1497 PERROR("close");
1498 }
1499 consumer_data->err_sock = -1;
1500 }
1501 if (consumer_data->cmd_sock >= 0) {
1502 ret = close(consumer_data->cmd_sock);
1503 if (ret) {
1504 PERROR("close");
1505 }
1506 consumer_data->cmd_sock = -1;
1507 }
1508 if (consumer_data->metadata_sock.fd_ptr &&
1509 *consumer_data->metadata_sock.fd_ptr >= 0) {
1510 ret = close(*consumer_data->metadata_sock.fd_ptr);
1511 if (ret) {
1512 PERROR("close");
1513 }
1514 }
1515 if (sock >= 0) {
1516 ret = close(sock);
1517 if (ret) {
1518 PERROR("close");
1519 }
1520 }
1521
1522 unlink(consumer_data->err_unix_sock_path);
1523 unlink(consumer_data->cmd_unix_sock_path);
1524 pthread_mutex_unlock(&consumer_data->lock);
1525
1526 /* Cleanup metadata socket mutex. */
1527 if (consumer_data->metadata_sock.lock) {
1528 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1529 free(consumer_data->metadata_sock.lock);
1530 }
1531 lttng_poll_clean(&events);
1532 error_poll:
1533 if (err) {
1534 health_error();
1535 ERR("Health error occurred in %s", __func__);
1536 }
1537 health_unregister(health_sessiond);
1538 DBG("consumer thread cleanup completed");
1539
1540 rcu_thread_offline();
1541 rcu_unregister_thread();
1542
1543 return NULL;
1544 }
1545
1546 /*
1547 * This thread manage application communication.
1548 */
1549 static void *thread_manage_apps(void *data)
1550 {
1551 int i, ret, pollfd, err = -1;
1552 ssize_t size_ret;
1553 uint32_t revents, nb_fd;
1554 struct lttng_poll_event events;
1555
1556 DBG("[thread] Manage application started");
1557
1558 rcu_register_thread();
1559 rcu_thread_online();
1560
1561 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1562
1563 if (testpoint(sessiond_thread_manage_apps)) {
1564 goto error_testpoint;
1565 }
1566
1567 health_code_update();
1568
1569 ret = sessiond_set_thread_pollset(&events, 2);
1570 if (ret < 0) {
1571 goto error_poll_create;
1572 }
1573
1574 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1575 if (ret < 0) {
1576 goto error;
1577 }
1578
1579 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1580 goto error;
1581 }
1582
1583 health_code_update();
1584
1585 while (1) {
1586 DBG("Apps thread polling");
1587
1588 /* Inifinite blocking call, waiting for transmission */
1589 restart:
1590 health_poll_entry();
1591 ret = lttng_poll_wait(&events, -1);
1592 DBG("Apps thread return from poll on %d fds",
1593 LTTNG_POLL_GETNB(&events));
1594 health_poll_exit();
1595 if (ret < 0) {
1596 /*
1597 * Restart interrupted system call.
1598 */
1599 if (errno == EINTR) {
1600 goto restart;
1601 }
1602 goto error;
1603 }
1604
1605 nb_fd = ret;
1606
1607 for (i = 0; i < nb_fd; i++) {
1608 /* Fetch once the poll data */
1609 revents = LTTNG_POLL_GETEV(&events, i);
1610 pollfd = LTTNG_POLL_GETFD(&events, i);
1611
1612 health_code_update();
1613
1614 if (!revents) {
1615 /* No activity for this FD (poll implementation). */
1616 continue;
1617 }
1618
1619 /* Thread quit pipe has been closed. Killing thread. */
1620 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1621 if (ret) {
1622 err = 0;
1623 goto exit;
1624 }
1625
1626 /* Inspect the apps cmd pipe */
1627 if (pollfd == apps_cmd_pipe[0]) {
1628 if (revents & LPOLLIN) {
1629 int sock;
1630
1631 /* Empty pipe */
1632 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1633 if (size_ret < sizeof(sock)) {
1634 PERROR("read apps cmd pipe");
1635 goto error;
1636 }
1637
1638 health_code_update();
1639
1640 /*
1641 * Since this is a command socket (write then read),
1642 * we only monitor the error events of the socket.
1643 */
1644 ret = lttng_poll_add(&events, sock,
1645 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1646 if (ret < 0) {
1647 goto error;
1648 }
1649
1650 DBG("Apps with sock %d added to poll set", sock);
1651 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1652 ERR("Apps command pipe error");
1653 goto error;
1654 } else {
1655 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1656 goto error;
1657 }
1658 } else {
1659 /*
1660 * At this point, we know that a registered application made
1661 * the event at poll_wait.
1662 */
1663 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1664 /* Removing from the poll set */
1665 ret = lttng_poll_del(&events, pollfd);
1666 if (ret < 0) {
1667 goto error;
1668 }
1669
1670 /* Socket closed on remote end. */
1671 ust_app_unregister(pollfd);
1672 } else {
1673 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1674 goto error;
1675 }
1676 }
1677
1678 health_code_update();
1679 }
1680 }
1681
1682 exit:
1683 error:
1684 lttng_poll_clean(&events);
1685 error_poll_create:
1686 error_testpoint:
1687 utils_close_pipe(apps_cmd_pipe);
1688 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1689
1690 /*
1691 * We don't clean the UST app hash table here since already registered
1692 * applications can still be controlled so let them be until the session
1693 * daemon dies or the applications stop.
1694 */
1695
1696 if (err) {
1697 health_error();
1698 ERR("Health error occurred in %s", __func__);
1699 }
1700 health_unregister(health_sessiond);
1701 DBG("Application communication apps thread cleanup complete");
1702 rcu_thread_offline();
1703 rcu_unregister_thread();
1704 return NULL;
1705 }
1706
1707 /*
1708 * Send a socket to a thread This is called from the dispatch UST registration
1709 * thread once all sockets are set for the application.
1710 *
1711 * The sock value can be invalid, we don't really care, the thread will handle
1712 * it and make the necessary cleanup if so.
1713 *
1714 * On success, return 0 else a negative value being the errno message of the
1715 * write().
1716 */
1717 static int send_socket_to_thread(int fd, int sock)
1718 {
1719 ssize_t ret;
1720
1721 /*
1722 * It's possible that the FD is set as invalid with -1 concurrently just
1723 * before calling this function being a shutdown state of the thread.
1724 */
1725 if (fd < 0) {
1726 ret = -EBADF;
1727 goto error;
1728 }
1729
1730 ret = lttng_write(fd, &sock, sizeof(sock));
1731 if (ret < sizeof(sock)) {
1732 PERROR("write apps pipe %d", fd);
1733 if (ret < 0) {
1734 ret = -errno;
1735 }
1736 goto error;
1737 }
1738
1739 /* All good. Don't send back the write positive ret value. */
1740 ret = 0;
1741 error:
1742 return (int) ret;
1743 }
1744
1745 /*
1746 * Sanitize the wait queue of the dispatch registration thread meaning removing
1747 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1748 * notify socket is never received.
1749 */
1750 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1751 {
1752 int ret, nb_fd = 0, i;
1753 unsigned int fd_added = 0;
1754 struct lttng_poll_event events;
1755 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1756
1757 assert(wait_queue);
1758
1759 lttng_poll_init(&events);
1760
1761 /* Just skip everything for an empty queue. */
1762 if (!wait_queue->count) {
1763 goto end;
1764 }
1765
1766 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1767 if (ret < 0) {
1768 goto error_create;
1769 }
1770
1771 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1772 &wait_queue->head, head) {
1773 assert(wait_node->app);
1774 ret = lttng_poll_add(&events, wait_node->app->sock,
1775 LPOLLHUP | LPOLLERR);
1776 if (ret < 0) {
1777 goto error;
1778 }
1779
1780 fd_added = 1;
1781 }
1782
1783 if (!fd_added) {
1784 goto end;
1785 }
1786
1787 /*
1788 * Poll but don't block so we can quickly identify the faulty events and
1789 * clean them afterwards from the wait queue.
1790 */
1791 ret = lttng_poll_wait(&events, 0);
1792 if (ret < 0) {
1793 goto error;
1794 }
1795 nb_fd = ret;
1796
1797 for (i = 0; i < nb_fd; i++) {
1798 /* Get faulty FD. */
1799 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1800 int pollfd = LTTNG_POLL_GETFD(&events, i);
1801
1802 if (!revents) {
1803 /* No activity for this FD (poll implementation). */
1804 continue;
1805 }
1806
1807 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1808 &wait_queue->head, head) {
1809 if (pollfd == wait_node->app->sock &&
1810 (revents & (LPOLLHUP | LPOLLERR))) {
1811 cds_list_del(&wait_node->head);
1812 wait_queue->count--;
1813 ust_app_destroy(wait_node->app);
1814 free(wait_node);
1815 /*
1816 * Silence warning of use-after-free in
1817 * cds_list_for_each_entry_safe which uses
1818 * __typeof__(*wait_node).
1819 */
1820 wait_node = NULL;
1821 break;
1822 } else {
1823 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1824 goto error;
1825 }
1826 }
1827 }
1828
1829 if (nb_fd > 0) {
1830 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1831 }
1832
1833 end:
1834 lttng_poll_clean(&events);
1835 return;
1836
1837 error:
1838 lttng_poll_clean(&events);
1839 error_create:
1840 ERR("Unable to sanitize wait queue");
1841 return;
1842 }
1843
1844 /*
1845 * Dispatch request from the registration threads to the application
1846 * communication thread.
1847 */
1848 static void *thread_dispatch_ust_registration(void *data)
1849 {
1850 int ret, err = -1;
1851 struct cds_wfcq_node *node;
1852 struct ust_command *ust_cmd = NULL;
1853 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1854 struct ust_reg_wait_queue wait_queue = {
1855 .count = 0,
1856 };
1857
1858 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1859
1860 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1861 goto error_testpoint;
1862 }
1863
1864 health_code_update();
1865
1866 CDS_INIT_LIST_HEAD(&wait_queue.head);
1867
1868 DBG("[thread] Dispatch UST command started");
1869
1870 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1871 health_code_update();
1872
1873 /* Atomically prepare the queue futex */
1874 futex_nto1_prepare(&ust_cmd_queue.futex);
1875
1876 do {
1877 struct ust_app *app = NULL;
1878 ust_cmd = NULL;
1879
1880 /*
1881 * Make sure we don't have node(s) that have hung up before receiving
1882 * the notify socket. This is to clean the list in order to avoid
1883 * memory leaks from notify socket that are never seen.
1884 */
1885 sanitize_wait_queue(&wait_queue);
1886
1887 health_code_update();
1888 /* Dequeue command for registration */
1889 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1890 if (node == NULL) {
1891 DBG("Woken up but nothing in the UST command queue");
1892 /* Continue thread execution */
1893 break;
1894 }
1895
1896 ust_cmd = caa_container_of(node, struct ust_command, node);
1897
1898 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1899 " gid:%d sock:%d name:%s (version %d.%d)",
1900 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1901 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1902 ust_cmd->sock, ust_cmd->reg_msg.name,
1903 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1904
1905 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1906 wait_node = zmalloc(sizeof(*wait_node));
1907 if (!wait_node) {
1908 PERROR("zmalloc wait_node dispatch");
1909 ret = close(ust_cmd->sock);
1910 if (ret < 0) {
1911 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1912 }
1913 lttng_fd_put(LTTNG_FD_APPS, 1);
1914 free(ust_cmd);
1915 goto error;
1916 }
1917 CDS_INIT_LIST_HEAD(&wait_node->head);
1918
1919 /* Create application object if socket is CMD. */
1920 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1921 ust_cmd->sock);
1922 if (!wait_node->app) {
1923 ret = close(ust_cmd->sock);
1924 if (ret < 0) {
1925 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1926 }
1927 lttng_fd_put(LTTNG_FD_APPS, 1);
1928 free(wait_node);
1929 free(ust_cmd);
1930 continue;
1931 }
1932 /*
1933 * Add application to the wait queue so we can set the notify
1934 * socket before putting this object in the global ht.
1935 */
1936 cds_list_add(&wait_node->head, &wait_queue.head);
1937 wait_queue.count++;
1938
1939 free(ust_cmd);
1940 /*
1941 * We have to continue here since we don't have the notify
1942 * socket and the application MUST be added to the hash table
1943 * only at that moment.
1944 */
1945 continue;
1946 } else {
1947 /*
1948 * Look for the application in the local wait queue and set the
1949 * notify socket if found.
1950 */
1951 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1952 &wait_queue.head, head) {
1953 health_code_update();
1954 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1955 wait_node->app->notify_sock = ust_cmd->sock;
1956 cds_list_del(&wait_node->head);
1957 wait_queue.count--;
1958 app = wait_node->app;
1959 free(wait_node);
1960 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1961 break;
1962 }
1963 }
1964
1965 /*
1966 * With no application at this stage the received socket is
1967 * basically useless so close it before we free the cmd data
1968 * structure for good.
1969 */
1970 if (!app) {
1971 ret = close(ust_cmd->sock);
1972 if (ret < 0) {
1973 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1974 }
1975 lttng_fd_put(LTTNG_FD_APPS, 1);
1976 }
1977 free(ust_cmd);
1978 }
1979
1980 if (app) {
1981 /*
1982 * @session_lock_list
1983 *
1984 * Lock the global session list so from the register up to the
1985 * registration done message, no thread can see the application
1986 * and change its state.
1987 */
1988 session_lock_list();
1989 rcu_read_lock();
1990
1991 /*
1992 * Add application to the global hash table. This needs to be
1993 * done before the update to the UST registry can locate the
1994 * application.
1995 */
1996 ust_app_add(app);
1997
1998 /* Set app version. This call will print an error if needed. */
1999 (void) ust_app_version(app);
2000
2001 /* Send notify socket through the notify pipe. */
2002 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2003 app->notify_sock);
2004 if (ret < 0) {
2005 rcu_read_unlock();
2006 session_unlock_list();
2007 /*
2008 * No notify thread, stop the UST tracing. However, this is
2009 * not an internal error of the this thread thus setting
2010 * the health error code to a normal exit.
2011 */
2012 err = 0;
2013 goto error;
2014 }
2015
2016 /*
2017 * Update newly registered application with the tracing
2018 * registry info already enabled information.
2019 */
2020 update_ust_app(app->sock);
2021
2022 /*
2023 * Don't care about return value. Let the manage apps threads
2024 * handle app unregistration upon socket close.
2025 */
2026 (void) ust_app_register_done(app);
2027
2028 /*
2029 * Even if the application socket has been closed, send the app
2030 * to the thread and unregistration will take place at that
2031 * place.
2032 */
2033 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2034 if (ret < 0) {
2035 rcu_read_unlock();
2036 session_unlock_list();
2037 /*
2038 * No apps. thread, stop the UST tracing. However, this is
2039 * not an internal error of the this thread thus setting
2040 * the health error code to a normal exit.
2041 */
2042 err = 0;
2043 goto error;
2044 }
2045
2046 rcu_read_unlock();
2047 session_unlock_list();
2048 }
2049 } while (node != NULL);
2050
2051 health_poll_entry();
2052 /* Futex wait on queue. Blocking call on futex() */
2053 futex_nto1_wait(&ust_cmd_queue.futex);
2054 health_poll_exit();
2055 }
2056 /* Normal exit, no error */
2057 err = 0;
2058
2059 error:
2060 /* Clean up wait queue. */
2061 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2062 &wait_queue.head, head) {
2063 cds_list_del(&wait_node->head);
2064 wait_queue.count--;
2065 free(wait_node);
2066 }
2067
2068 /* Empty command queue. */
2069 for (;;) {
2070 /* Dequeue command for registration */
2071 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2072 if (node == NULL) {
2073 break;
2074 }
2075 ust_cmd = caa_container_of(node, struct ust_command, node);
2076 ret = close(ust_cmd->sock);
2077 if (ret < 0) {
2078 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2079 }
2080 lttng_fd_put(LTTNG_FD_APPS, 1);
2081 free(ust_cmd);
2082 }
2083
2084 error_testpoint:
2085 DBG("Dispatch thread dying");
2086 if (err) {
2087 health_error();
2088 ERR("Health error occurred in %s", __func__);
2089 }
2090 health_unregister(health_sessiond);
2091 return NULL;
2092 }
2093
2094 /*
2095 * This thread manage application registration.
2096 */
2097 static void *thread_registration_apps(void *data)
2098 {
2099 int sock = -1, i, ret, pollfd, err = -1;
2100 uint32_t revents, nb_fd;
2101 struct lttng_poll_event events;
2102 /*
2103 * Get allocated in this thread, enqueued to a global queue, dequeued and
2104 * freed in the manage apps thread.
2105 */
2106 struct ust_command *ust_cmd = NULL;
2107
2108 DBG("[thread] Manage application registration started");
2109
2110 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2111
2112 if (testpoint(sessiond_thread_registration_apps)) {
2113 goto error_testpoint;
2114 }
2115
2116 ret = lttcomm_listen_unix_sock(apps_sock);
2117 if (ret < 0) {
2118 goto error_listen;
2119 }
2120
2121 /*
2122 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2123 * more will be added to this poll set.
2124 */
2125 ret = sessiond_set_thread_pollset(&events, 2);
2126 if (ret < 0) {
2127 goto error_create_poll;
2128 }
2129
2130 /* Add the application registration socket */
2131 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2132 if (ret < 0) {
2133 goto error_poll_add;
2134 }
2135
2136 /* Notify all applications to register */
2137 ret = notify_ust_apps(1);
2138 if (ret < 0) {
2139 ERR("Failed to notify applications or create the wait shared memory.\n"
2140 "Execution continues but there might be problem for already\n"
2141 "running applications that wishes to register.");
2142 }
2143
2144 while (1) {
2145 DBG("Accepting application registration");
2146
2147 /* Inifinite blocking call, waiting for transmission */
2148 restart:
2149 health_poll_entry();
2150 ret = lttng_poll_wait(&events, -1);
2151 health_poll_exit();
2152 if (ret < 0) {
2153 /*
2154 * Restart interrupted system call.
2155 */
2156 if (errno == EINTR) {
2157 goto restart;
2158 }
2159 goto error;
2160 }
2161
2162 nb_fd = ret;
2163
2164 for (i = 0; i < nb_fd; i++) {
2165 health_code_update();
2166
2167 /* Fetch once the poll data */
2168 revents = LTTNG_POLL_GETEV(&events, i);
2169 pollfd = LTTNG_POLL_GETFD(&events, i);
2170
2171 if (!revents) {
2172 /* No activity for this FD (poll implementation). */
2173 continue;
2174 }
2175
2176 /* Thread quit pipe has been closed. Killing thread. */
2177 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2178 if (ret) {
2179 err = 0;
2180 goto exit;
2181 }
2182
2183 /* Event on the registration socket */
2184 if (pollfd == apps_sock) {
2185 if (revents & LPOLLIN) {
2186 sock = lttcomm_accept_unix_sock(apps_sock);
2187 if (sock < 0) {
2188 goto error;
2189 }
2190
2191 /*
2192 * Set socket timeout for both receiving and ending.
2193 * app_socket_timeout is in seconds, whereas
2194 * lttcomm_setsockopt_rcv_timeout and
2195 * lttcomm_setsockopt_snd_timeout expect msec as
2196 * parameter.
2197 */
2198 (void) lttcomm_setsockopt_rcv_timeout(sock,
2199 app_socket_timeout * 1000);
2200 (void) lttcomm_setsockopt_snd_timeout(sock,
2201 app_socket_timeout * 1000);
2202
2203 /*
2204 * Set the CLOEXEC flag. Return code is useless because
2205 * either way, the show must go on.
2206 */
2207 (void) utils_set_fd_cloexec(sock);
2208
2209 /* Create UST registration command for enqueuing */
2210 ust_cmd = zmalloc(sizeof(struct ust_command));
2211 if (ust_cmd == NULL) {
2212 PERROR("ust command zmalloc");
2213 ret = close(sock);
2214 if (ret) {
2215 PERROR("close");
2216 }
2217 goto error;
2218 }
2219
2220 /*
2221 * Using message-based transmissions to ensure we don't
2222 * have to deal with partially received messages.
2223 */
2224 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2225 if (ret < 0) {
2226 ERR("Exhausted file descriptors allowed for applications.");
2227 free(ust_cmd);
2228 ret = close(sock);
2229 if (ret) {
2230 PERROR("close");
2231 }
2232 sock = -1;
2233 continue;
2234 }
2235
2236 health_code_update();
2237 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2238 if (ret < 0) {
2239 free(ust_cmd);
2240 /* Close socket of the application. */
2241 ret = close(sock);
2242 if (ret) {
2243 PERROR("close");
2244 }
2245 lttng_fd_put(LTTNG_FD_APPS, 1);
2246 sock = -1;
2247 continue;
2248 }
2249 health_code_update();
2250
2251 ust_cmd->sock = sock;
2252 sock = -1;
2253
2254 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2255 " gid:%d sock:%d name:%s (version %d.%d)",
2256 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2257 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2258 ust_cmd->sock, ust_cmd->reg_msg.name,
2259 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2260
2261 /*
2262 * Lock free enqueue the registration request. The red pill
2263 * has been taken! This apps will be part of the *system*.
2264 */
2265 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2266
2267 /*
2268 * Wake the registration queue futex. Implicit memory
2269 * barrier with the exchange in cds_wfcq_enqueue.
2270 */
2271 futex_nto1_wake(&ust_cmd_queue.futex);
2272 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2273 ERR("Register apps socket poll error");
2274 goto error;
2275 } else {
2276 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2277 goto error;
2278 }
2279 }
2280 }
2281 }
2282
2283 exit:
2284 error:
2285 /* Notify that the registration thread is gone */
2286 notify_ust_apps(0);
2287
2288 if (apps_sock >= 0) {
2289 ret = close(apps_sock);
2290 if (ret) {
2291 PERROR("close");
2292 }
2293 }
2294 if (sock >= 0) {
2295 ret = close(sock);
2296 if (ret) {
2297 PERROR("close");
2298 }
2299 lttng_fd_put(LTTNG_FD_APPS, 1);
2300 }
2301 unlink(apps_unix_sock_path);
2302
2303 error_poll_add:
2304 lttng_poll_clean(&events);
2305 error_listen:
2306 error_create_poll:
2307 error_testpoint:
2308 DBG("UST Registration thread cleanup complete");
2309 if (err) {
2310 health_error();
2311 ERR("Health error occurred in %s", __func__);
2312 }
2313 health_unregister(health_sessiond);
2314
2315 return NULL;
2316 }
2317
2318 /*
2319 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2320 * exec or it will fails.
2321 */
2322 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2323 {
2324 int ret, clock_ret;
2325 struct timespec timeout;
2326
2327 /*
2328 * Make sure we set the readiness flag to 0 because we are NOT ready.
2329 * This access to consumer_thread_is_ready does not need to be
2330 * protected by consumer_data.cond_mutex (yet) since the consumer
2331 * management thread has not been started at this point.
2332 */
2333 consumer_data->consumer_thread_is_ready = 0;
2334
2335 /* Setup pthread condition */
2336 ret = pthread_condattr_init(&consumer_data->condattr);
2337 if (ret) {
2338 errno = ret;
2339 PERROR("pthread_condattr_init consumer data");
2340 goto error;
2341 }
2342
2343 /*
2344 * Set the monotonic clock in order to make sure we DO NOT jump in time
2345 * between the clock_gettime() call and the timedwait call. See bug #324
2346 * for a more details and how we noticed it.
2347 */
2348 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2349 if (ret) {
2350 errno = ret;
2351 PERROR("pthread_condattr_setclock consumer data");
2352 goto error;
2353 }
2354
2355 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2356 if (ret) {
2357 errno = ret;
2358 PERROR("pthread_cond_init consumer data");
2359 goto error;
2360 }
2361
2362 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2363 thread_manage_consumer, consumer_data);
2364 if (ret) {
2365 errno = ret;
2366 PERROR("pthread_create consumer");
2367 ret = -1;
2368 goto error;
2369 }
2370
2371 /* We are about to wait on a pthread condition */
2372 pthread_mutex_lock(&consumer_data->cond_mutex);
2373
2374 /* Get time for sem_timedwait absolute timeout */
2375 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2376 /*
2377 * Set the timeout for the condition timed wait even if the clock gettime
2378 * call fails since we might loop on that call and we want to avoid to
2379 * increment the timeout too many times.
2380 */
2381 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2382
2383 /*
2384 * The following loop COULD be skipped in some conditions so this is why we
2385 * set ret to 0 in order to make sure at least one round of the loop is
2386 * done.
2387 */
2388 ret = 0;
2389
2390 /*
2391 * Loop until the condition is reached or when a timeout is reached. Note
2392 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2393 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2394 * possible. This loop does not take any chances and works with both of
2395 * them.
2396 */
2397 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2398 if (clock_ret < 0) {
2399 PERROR("clock_gettime spawn consumer");
2400 /* Infinite wait for the consumerd thread to be ready */
2401 ret = pthread_cond_wait(&consumer_data->cond,
2402 &consumer_data->cond_mutex);
2403 } else {
2404 ret = pthread_cond_timedwait(&consumer_data->cond,
2405 &consumer_data->cond_mutex, &timeout);
2406 }
2407 }
2408
2409 /* Release the pthread condition */
2410 pthread_mutex_unlock(&consumer_data->cond_mutex);
2411
2412 if (ret != 0) {
2413 errno = ret;
2414 if (ret == ETIMEDOUT) {
2415 int pth_ret;
2416
2417 /*
2418 * Call has timed out so we kill the kconsumerd_thread and return
2419 * an error.
2420 */
2421 ERR("Condition timed out. The consumer thread was never ready."
2422 " Killing it");
2423 pth_ret = pthread_cancel(consumer_data->thread);
2424 if (pth_ret < 0) {
2425 PERROR("pthread_cancel consumer thread");
2426 }
2427 } else {
2428 PERROR("pthread_cond_wait failed consumer thread");
2429 }
2430 /* Caller is expecting a negative value on failure. */
2431 ret = -1;
2432 goto error;
2433 }
2434
2435 pthread_mutex_lock(&consumer_data->pid_mutex);
2436 if (consumer_data->pid == 0) {
2437 ERR("Consumerd did not start");
2438 pthread_mutex_unlock(&consumer_data->pid_mutex);
2439 goto error;
2440 }
2441 pthread_mutex_unlock(&consumer_data->pid_mutex);
2442
2443 return 0;
2444
2445 error:
2446 return ret;
2447 }
2448
2449 /*
2450 * Join consumer thread
2451 */
2452 static int join_consumer_thread(struct consumer_data *consumer_data)
2453 {
2454 void *status;
2455
2456 /* Consumer pid must be a real one. */
2457 if (consumer_data->pid > 0) {
2458 int ret;
2459 ret = kill(consumer_data->pid, SIGTERM);
2460 if (ret) {
2461 PERROR("Error killing consumer daemon");
2462 return ret;
2463 }
2464 return pthread_join(consumer_data->thread, &status);
2465 } else {
2466 return 0;
2467 }
2468 }
2469
2470 /*
2471 * Fork and exec a consumer daemon (consumerd).
2472 *
2473 * Return pid if successful else -1.
2474 */
2475 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2476 {
2477 int ret;
2478 pid_t pid;
2479 const char *consumer_to_use;
2480 const char *verbosity;
2481 struct stat st;
2482
2483 DBG("Spawning consumerd");
2484
2485 pid = fork();
2486 if (pid == 0) {
2487 /*
2488 * Exec consumerd.
2489 */
2490 if (opt_verbose_consumer) {
2491 verbosity = "--verbose";
2492 } else if (lttng_opt_quiet) {
2493 verbosity = "--quiet";
2494 } else {
2495 verbosity = "";
2496 }
2497
2498 switch (consumer_data->type) {
2499 case LTTNG_CONSUMER_KERNEL:
2500 /*
2501 * Find out which consumerd to execute. We will first try the
2502 * 64-bit path, then the sessiond's installation directory, and
2503 * fallback on the 32-bit one,
2504 */
2505 DBG3("Looking for a kernel consumer at these locations:");
2506 DBG3(" 1) %s", consumerd64_bin);
2507 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2508 DBG3(" 3) %s", consumerd32_bin);
2509 if (stat(consumerd64_bin, &st) == 0) {
2510 DBG3("Found location #1");
2511 consumer_to_use = consumerd64_bin;
2512 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2513 DBG3("Found location #2");
2514 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2515 } else if (stat(consumerd32_bin, &st) == 0) {
2516 DBG3("Found location #3");
2517 consumer_to_use = consumerd32_bin;
2518 } else {
2519 DBG("Could not find any valid consumerd executable");
2520 ret = -EINVAL;
2521 break;
2522 }
2523 DBG("Using kernel consumer at: %s", consumer_to_use);
2524 ret = execl(consumer_to_use,
2525 "lttng-consumerd", verbosity, "-k",
2526 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2527 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2528 "--group", tracing_group_name,
2529 NULL);
2530 break;
2531 case LTTNG_CONSUMER64_UST:
2532 {
2533 char *tmpnew = NULL;
2534
2535 if (consumerd64_libdir[0] != '\0') {
2536 char *tmp;
2537 size_t tmplen;
2538
2539 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2540 if (!tmp) {
2541 tmp = "";
2542 }
2543 tmplen = strlen("LD_LIBRARY_PATH=")
2544 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2545 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2546 if (!tmpnew) {
2547 ret = -ENOMEM;
2548 goto error;
2549 }
2550 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2551 strcat(tmpnew, consumerd64_libdir);
2552 if (tmp[0] != '\0') {
2553 strcat(tmpnew, ":");
2554 strcat(tmpnew, tmp);
2555 }
2556 ret = putenv(tmpnew);
2557 if (ret) {
2558 ret = -errno;
2559 free(tmpnew);
2560 goto error;
2561 }
2562 }
2563 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2564 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2565 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2566 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2567 "--group", tracing_group_name,
2568 NULL);
2569 if (consumerd64_libdir[0] != '\0') {
2570 free(tmpnew);
2571 }
2572 break;
2573 }
2574 case LTTNG_CONSUMER32_UST:
2575 {
2576 char *tmpnew = NULL;
2577
2578 if (consumerd32_libdir[0] != '\0') {
2579 char *tmp;
2580 size_t tmplen;
2581
2582 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2583 if (!tmp) {
2584 tmp = "";
2585 }
2586 tmplen = strlen("LD_LIBRARY_PATH=")
2587 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2588 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2589 if (!tmpnew) {
2590 ret = -ENOMEM;
2591 goto error;
2592 }
2593 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2594 strcat(tmpnew, consumerd32_libdir);
2595 if (tmp[0] != '\0') {
2596 strcat(tmpnew, ":");
2597 strcat(tmpnew, tmp);
2598 }
2599 ret = putenv(tmpnew);
2600 if (ret) {
2601 ret = -errno;
2602 free(tmpnew);
2603 goto error;
2604 }
2605 }
2606 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2607 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2608 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2609 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2610 "--group", tracing_group_name,
2611 NULL);
2612 if (consumerd32_libdir[0] != '\0') {
2613 free(tmpnew);
2614 }
2615 break;
2616 }
2617 default:
2618 PERROR("unknown consumer type");
2619 exit(EXIT_FAILURE);
2620 }
2621 if (errno != 0) {
2622 PERROR("Consumer execl()");
2623 }
2624 /* Reaching this point, we got a failure on our execl(). */
2625 exit(EXIT_FAILURE);
2626 } else if (pid > 0) {
2627 ret = pid;
2628 } else {
2629 PERROR("start consumer fork");
2630 ret = -errno;
2631 }
2632 error:
2633 return ret;
2634 }
2635
2636 /*
2637 * Spawn the consumerd daemon and session daemon thread.
2638 */
2639 static int start_consumerd(struct consumer_data *consumer_data)
2640 {
2641 int ret;
2642
2643 /*
2644 * Set the listen() state on the socket since there is a possible race
2645 * between the exec() of the consumer daemon and this call if place in the
2646 * consumer thread. See bug #366 for more details.
2647 */
2648 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2649 if (ret < 0) {
2650 goto error;
2651 }
2652
2653 pthread_mutex_lock(&consumer_data->pid_mutex);
2654 if (consumer_data->pid != 0) {
2655 pthread_mutex_unlock(&consumer_data->pid_mutex);
2656 goto end;
2657 }
2658
2659 ret = spawn_consumerd(consumer_data);
2660 if (ret < 0) {
2661 ERR("Spawning consumerd failed");
2662 pthread_mutex_unlock(&consumer_data->pid_mutex);
2663 goto error;
2664 }
2665
2666 /* Setting up the consumer_data pid */
2667 consumer_data->pid = ret;
2668 DBG2("Consumer pid %d", consumer_data->pid);
2669 pthread_mutex_unlock(&consumer_data->pid_mutex);
2670
2671 DBG2("Spawning consumer control thread");
2672 ret = spawn_consumer_thread(consumer_data);
2673 if (ret < 0) {
2674 ERR("Fatal error spawning consumer control thread");
2675 goto error;
2676 }
2677
2678 end:
2679 return 0;
2680
2681 error:
2682 /* Cleanup already created sockets on error. */
2683 if (consumer_data->err_sock >= 0) {
2684 int err;
2685
2686 err = close(consumer_data->err_sock);
2687 if (err < 0) {
2688 PERROR("close consumer data error socket");
2689 }
2690 }
2691 return ret;
2692 }
2693
2694 /*
2695 * Setup necessary data for kernel tracer action.
2696 */
2697 static int init_kernel_tracer(void)
2698 {
2699 int ret;
2700
2701 /* Modprobe lttng kernel modules */
2702 ret = modprobe_lttng_control();
2703 if (ret < 0) {
2704 goto error;
2705 }
2706
2707 /* Open debugfs lttng */
2708 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2709 if (kernel_tracer_fd < 0) {
2710 DBG("Failed to open %s", module_proc_lttng);
2711 ret = -1;
2712 goto error_open;
2713 }
2714
2715 /* Validate kernel version */
2716 ret = kernel_validate_version(kernel_tracer_fd);
2717 if (ret < 0) {
2718 goto error_version;
2719 }
2720
2721 ret = modprobe_lttng_data();
2722 if (ret < 0) {
2723 goto error_modules;
2724 }
2725
2726 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2727 return 0;
2728
2729 error_version:
2730 modprobe_remove_lttng_control();
2731 ret = close(kernel_tracer_fd);
2732 if (ret) {
2733 PERROR("close");
2734 }
2735 kernel_tracer_fd = -1;
2736 return LTTNG_ERR_KERN_VERSION;
2737
2738 error_modules:
2739 ret = close(kernel_tracer_fd);
2740 if (ret) {
2741 PERROR("close");
2742 }
2743
2744 error_open:
2745 modprobe_remove_lttng_control();
2746
2747 error:
2748 WARN("No kernel tracer available");
2749 kernel_tracer_fd = -1;
2750 if (!is_root) {
2751 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2752 } else {
2753 return LTTNG_ERR_KERN_NA;
2754 }
2755 }
2756
2757
2758 /*
2759 * Copy consumer output from the tracing session to the domain session. The
2760 * function also applies the right modification on a per domain basis for the
2761 * trace files destination directory.
2762 *
2763 * Should *NOT* be called with RCU read-side lock held.
2764 */
2765 static int copy_session_consumer(int domain, struct ltt_session *session)
2766 {
2767 int ret;
2768 const char *dir_name;
2769 struct consumer_output *consumer;
2770
2771 assert(session);
2772 assert(session->consumer);
2773
2774 switch (domain) {
2775 case LTTNG_DOMAIN_KERNEL:
2776 DBG3("Copying tracing session consumer output in kernel session");
2777 /*
2778 * XXX: We should audit the session creation and what this function
2779 * does "extra" in order to avoid a destroy since this function is used
2780 * in the domain session creation (kernel and ust) only. Same for UST
2781 * domain.
2782 */
2783 if (session->kernel_session->consumer) {
2784 consumer_output_put(session->kernel_session->consumer);
2785 }
2786 session->kernel_session->consumer =
2787 consumer_copy_output(session->consumer);
2788 /* Ease our life a bit for the next part */
2789 consumer = session->kernel_session->consumer;
2790 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2791 break;
2792 case LTTNG_DOMAIN_JUL:
2793 case LTTNG_DOMAIN_LOG4J:
2794 case LTTNG_DOMAIN_PYTHON:
2795 case LTTNG_DOMAIN_UST:
2796 DBG3("Copying tracing session consumer output in UST session");
2797 if (session->ust_session->consumer) {
2798 consumer_output_put(session->ust_session->consumer);
2799 }
2800 session->ust_session->consumer =
2801 consumer_copy_output(session->consumer);
2802 /* Ease our life a bit for the next part */
2803 consumer = session->ust_session->consumer;
2804 dir_name = DEFAULT_UST_TRACE_DIR;
2805 break;
2806 default:
2807 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2808 goto error;
2809 }
2810
2811 /* Append correct directory to subdir */
2812 strncat(consumer->subdir, dir_name,
2813 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2814 DBG3("Copy session consumer subdir %s", consumer->subdir);
2815
2816 ret = LTTNG_OK;
2817
2818 error:
2819 return ret;
2820 }
2821
2822 /*
2823 * Create an UST session and add it to the session ust list.
2824 *
2825 * Should *NOT* be called with RCU read-side lock held.
2826 */
2827 static int create_ust_session(struct ltt_session *session,
2828 struct lttng_domain *domain)
2829 {
2830 int ret;
2831 struct ltt_ust_session *lus = NULL;
2832
2833 assert(session);
2834 assert(domain);
2835 assert(session->consumer);
2836
2837 switch (domain->type) {
2838 case LTTNG_DOMAIN_JUL:
2839 case LTTNG_DOMAIN_LOG4J:
2840 case LTTNG_DOMAIN_PYTHON:
2841 case LTTNG_DOMAIN_UST:
2842 break;
2843 default:
2844 ERR("Unknown UST domain on create session %d", domain->type);
2845 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2846 goto error;
2847 }
2848
2849 DBG("Creating UST session");
2850
2851 lus = trace_ust_create_session(session->id);
2852 if (lus == NULL) {
2853 ret = LTTNG_ERR_UST_SESS_FAIL;
2854 goto error;
2855 }
2856
2857 lus->uid = session->uid;
2858 lus->gid = session->gid;
2859 lus->output_traces = session->output_traces;
2860 lus->snapshot_mode = session->snapshot_mode;
2861 lus->live_timer_interval = session->live_timer;
2862 session->ust_session = lus;
2863 if (session->shm_path[0]) {
2864 strncpy(lus->root_shm_path, session->shm_path,
2865 sizeof(lus->root_shm_path));
2866 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2867 strncpy(lus->shm_path, session->shm_path,
2868 sizeof(lus->shm_path));
2869 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2870 strncat(lus->shm_path, "/ust",
2871 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2872 }
2873 /* Copy session output to the newly created UST session */
2874 ret = copy_session_consumer(domain->type, session);
2875 if (ret != LTTNG_OK) {
2876 goto error;
2877 }
2878
2879 return LTTNG_OK;
2880
2881 error:
2882 free(lus);
2883 session->ust_session = NULL;
2884 return ret;
2885 }
2886
2887 /*
2888 * Create a kernel tracer session then create the default channel.
2889 */
2890 static int create_kernel_session(struct ltt_session *session)
2891 {
2892 int ret;
2893
2894 DBG("Creating kernel session");
2895
2896 ret = kernel_create_session(session, kernel_tracer_fd);
2897 if (ret < 0) {
2898 ret = LTTNG_ERR_KERN_SESS_FAIL;
2899 goto error;
2900 }
2901
2902 /* Code flow safety */
2903 assert(session->kernel_session);
2904
2905 /* Copy session output to the newly created Kernel session */
2906 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2907 if (ret != LTTNG_OK) {
2908 goto error;
2909 }
2910
2911 /* Create directory(ies) on local filesystem. */
2912 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2913 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2914 ret = run_as_mkdir_recursive(
2915 session->kernel_session->consumer->dst.trace_path,
2916 S_IRWXU | S_IRWXG, session->uid, session->gid);
2917 if (ret < 0) {
2918 if (errno != EEXIST) {
2919 ERR("Trace directory creation error");
2920 goto error;
2921 }
2922 }
2923 }
2924
2925 session->kernel_session->uid = session->uid;
2926 session->kernel_session->gid = session->gid;
2927 session->kernel_session->output_traces = session->output_traces;
2928 session->kernel_session->snapshot_mode = session->snapshot_mode;
2929
2930 return LTTNG_OK;
2931
2932 error:
2933 trace_kernel_destroy_session(session->kernel_session);
2934 session->kernel_session = NULL;
2935 return ret;
2936 }
2937
2938 /*
2939 * Count number of session permitted by uid/gid.
2940 */
2941 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2942 {
2943 unsigned int i = 0;
2944 struct ltt_session *session;
2945
2946 DBG("Counting number of available session for UID %d GID %d",
2947 uid, gid);
2948 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2949 /*
2950 * Only list the sessions the user can control.
2951 */
2952 if (!session_access_ok(session, uid, gid)) {
2953 continue;
2954 }
2955 i++;
2956 }
2957 return i;
2958 }
2959
2960 /*
2961 * Process the command requested by the lttng client within the command
2962 * context structure. This function make sure that the return structure (llm)
2963 * is set and ready for transmission before returning.
2964 *
2965 * Return any error encountered or 0 for success.
2966 *
2967 * "sock" is only used for special-case var. len data.
2968 *
2969 * Should *NOT* be called with RCU read-side lock held.
2970 */
2971 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2972 int *sock_error)
2973 {
2974 int ret = LTTNG_OK;
2975 int need_tracing_session = 1;
2976 int need_domain;
2977
2978 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2979
2980 assert(!rcu_read_ongoing());
2981
2982 *sock_error = 0;
2983
2984 switch (cmd_ctx->lsm->cmd_type) {
2985 case LTTNG_CREATE_SESSION:
2986 case LTTNG_CREATE_SESSION_SNAPSHOT:
2987 case LTTNG_CREATE_SESSION_LIVE:
2988 case LTTNG_DESTROY_SESSION:
2989 case LTTNG_LIST_SESSIONS:
2990 case LTTNG_LIST_DOMAINS:
2991 case LTTNG_START_TRACE:
2992 case LTTNG_STOP_TRACE:
2993 case LTTNG_DATA_PENDING:
2994 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2995 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2996 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2997 case LTTNG_SNAPSHOT_RECORD:
2998 case LTTNG_SAVE_SESSION:
2999 case LTTNG_SET_SESSION_SHM_PATH:
3000 case LTTNG_REGENERATE_METADATA:
3001 case LTTNG_REGENERATE_STATEDUMP:
3002 need_domain = 0;
3003 break;
3004 default:
3005 need_domain = 1;
3006 }
3007
3008 if (opt_no_kernel && need_domain
3009 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3010 if (!is_root) {
3011 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3012 } else {
3013 ret = LTTNG_ERR_KERN_NA;
3014 }
3015 goto error;
3016 }
3017
3018 /* Deny register consumer if we already have a spawned consumer. */
3019 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3020 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3021 if (kconsumer_data.pid > 0) {
3022 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3023 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3024 goto error;
3025 }
3026 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3027 }
3028
3029 /*
3030 * Check for command that don't needs to allocate a returned payload. We do
3031 * this here so we don't have to make the call for no payload at each
3032 * command.
3033 */
3034 switch(cmd_ctx->lsm->cmd_type) {
3035 case LTTNG_LIST_SESSIONS:
3036 case LTTNG_LIST_TRACEPOINTS:
3037 case LTTNG_LIST_TRACEPOINT_FIELDS:
3038 case LTTNG_LIST_DOMAINS:
3039 case LTTNG_LIST_CHANNELS:
3040 case LTTNG_LIST_EVENTS:
3041 case LTTNG_LIST_SYSCALLS:
3042 case LTTNG_LIST_TRACKER_PIDS:
3043 case LTTNG_DATA_PENDING:
3044 break;
3045 default:
3046 /* Setup lttng message with no payload */
3047 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3048 if (ret < 0) {
3049 /* This label does not try to unlock the session */
3050 goto init_setup_error;
3051 }
3052 }
3053
3054 /* Commands that DO NOT need a session. */
3055 switch (cmd_ctx->lsm->cmd_type) {
3056 case LTTNG_CREATE_SESSION:
3057 case LTTNG_CREATE_SESSION_SNAPSHOT:
3058 case LTTNG_CREATE_SESSION_LIVE:
3059 case LTTNG_CALIBRATE:
3060 case LTTNG_LIST_SESSIONS:
3061 case LTTNG_LIST_TRACEPOINTS:
3062 case LTTNG_LIST_SYSCALLS:
3063 case LTTNG_LIST_TRACEPOINT_FIELDS:
3064 case LTTNG_SAVE_SESSION:
3065 need_tracing_session = 0;
3066 break;
3067 default:
3068 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3069 /*
3070 * We keep the session list lock across _all_ commands
3071 * for now, because the per-session lock does not
3072 * handle teardown properly.
3073 */
3074 session_lock_list();
3075 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3076 if (cmd_ctx->session == NULL) {
3077 ret = LTTNG_ERR_SESS_NOT_FOUND;
3078 goto error;
3079 } else {
3080 /* Acquire lock for the session */
3081 session_lock(cmd_ctx->session);
3082 }
3083 break;
3084 }
3085
3086 /*
3087 * Commands that need a valid session but should NOT create one if none
3088 * exists. Instead of creating one and destroying it when the command is
3089 * handled, process that right before so we save some round trip in useless
3090 * code path.
3091 */
3092 switch (cmd_ctx->lsm->cmd_type) {
3093 case LTTNG_DISABLE_CHANNEL:
3094 case LTTNG_DISABLE_EVENT:
3095 switch (cmd_ctx->lsm->domain.type) {
3096 case LTTNG_DOMAIN_KERNEL:
3097 if (!cmd_ctx->session->kernel_session) {
3098 ret = LTTNG_ERR_NO_CHANNEL;
3099 goto error;
3100 }
3101 break;
3102 case LTTNG_DOMAIN_JUL:
3103 case LTTNG_DOMAIN_LOG4J:
3104 case LTTNG_DOMAIN_PYTHON:
3105 case LTTNG_DOMAIN_UST:
3106 if (!cmd_ctx->session->ust_session) {
3107 ret = LTTNG_ERR_NO_CHANNEL;
3108 goto error;
3109 }
3110 break;
3111 default:
3112 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3113 goto error;
3114 }
3115 default:
3116 break;
3117 }
3118
3119 if (!need_domain) {
3120 goto skip_domain;
3121 }
3122
3123 /*
3124 * Check domain type for specific "pre-action".
3125 */
3126 switch (cmd_ctx->lsm->domain.type) {
3127 case LTTNG_DOMAIN_KERNEL:
3128 if (!is_root) {
3129 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3130 goto error;
3131 }
3132
3133 /* Kernel tracer check */
3134 if (kernel_tracer_fd == -1) {
3135 /* Basically, load kernel tracer modules */
3136 ret = init_kernel_tracer();
3137 if (ret != 0) {
3138 goto error;
3139 }
3140 }
3141
3142 /* Consumer is in an ERROR state. Report back to client */
3143 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3144 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3145 goto error;
3146 }
3147
3148 /* Need a session for kernel command */
3149 if (need_tracing_session) {
3150 if (cmd_ctx->session->kernel_session == NULL) {
3151 ret = create_kernel_session(cmd_ctx->session);
3152 if (ret < 0) {
3153 ret = LTTNG_ERR_KERN_SESS_FAIL;
3154 goto error;
3155 }
3156 }
3157
3158 /* Start the kernel consumer daemon */
3159 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3160 if (kconsumer_data.pid == 0 &&
3161 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3162 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3163 ret = start_consumerd(&kconsumer_data);
3164 if (ret < 0) {
3165 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3166 goto error;
3167 }
3168 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3169 } else {
3170 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3171 }
3172
3173 /*
3174 * The consumer was just spawned so we need to add the socket to
3175 * the consumer output of the session if exist.
3176 */
3177 ret = consumer_create_socket(&kconsumer_data,
3178 cmd_ctx->session->kernel_session->consumer);
3179 if (ret < 0) {
3180 goto error;
3181 }
3182 }
3183
3184 break;
3185 case LTTNG_DOMAIN_JUL:
3186 case LTTNG_DOMAIN_LOG4J:
3187 case LTTNG_DOMAIN_PYTHON:
3188 case LTTNG_DOMAIN_UST:
3189 {
3190 if (!ust_app_supported()) {
3191 ret = LTTNG_ERR_NO_UST;
3192 goto error;
3193 }
3194 /* Consumer is in an ERROR state. Report back to client */
3195 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3196 ret = LTTNG_ERR_NO_USTCONSUMERD;
3197 goto error;
3198 }
3199
3200 if (need_tracing_session) {
3201 /* Create UST session if none exist. */
3202 if (cmd_ctx->session->ust_session == NULL) {
3203 ret = create_ust_session(cmd_ctx->session,
3204 &cmd_ctx->lsm->domain);
3205 if (ret != LTTNG_OK) {
3206 goto error;
3207 }
3208 }
3209
3210 /* Start the UST consumer daemons */
3211 /* 64-bit */
3212 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3213 if (consumerd64_bin[0] != '\0' &&
3214 ustconsumer64_data.pid == 0 &&
3215 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3216 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3217 ret = start_consumerd(&ustconsumer64_data);
3218 if (ret < 0) {
3219 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3220 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3221 goto error;
3222 }
3223
3224 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3225 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3226 } else {
3227 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3228 }
3229
3230 /*
3231 * Setup socket for consumer 64 bit. No need for atomic