Fix: LPOLLHUP and LPOLLERR when there is still data in pipe/socket
[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 _GNU_SOURCE
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <paths.h>
26 #include <pthread.h>
27 #include <signal.h>
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <string.h>
31 #include <inttypes.h>
32 #include <sys/mman.h>
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
36 #include <sys/stat.h>
37 #include <sys/types.h>
38 #include <sys/wait.h>
39 #include <urcu/uatomic.h>
40 #include <unistd.h>
41 #include <config.h>
42
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
53
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
56 #include "channel.h"
57 #include "cmd.h"
58 #include "consumer.h"
59 #include "context.h"
60 #include "event.h"
61 #include "kernel.h"
62 #include "kernel-consumer.h"
63 #include "modprobe.h"
64 #include "shm.h"
65 #include "ust-ctl.h"
66 #include "ust-consumer.h"
67 #include "utils.h"
68 #include "fd-limit.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
73 #include "save.h"
74 #include "load-session-thread.h"
75 #include "syscall.h"
76 #include "agent.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 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
195
196 /*
197 * This pipe is used to inform the thread managing application communication
198 * that a command is queued and ready to be processed.
199 */
200 static int apps_cmd_pipe[2] = { -1, -1 };
201
202 int apps_cmd_notify_pipe[2] = { -1, -1 };
203
204 /* Pthread, Mutexes and Semaphores */
205 static pthread_t apps_thread;
206 static pthread_t apps_notify_thread;
207 static pthread_t reg_apps_thread;
208 static pthread_t client_thread;
209 static pthread_t kernel_thread;
210 static pthread_t dispatch_thread;
211 static pthread_t health_thread;
212 static pthread_t ht_cleanup_thread;
213 static pthread_t agent_reg_thread;
214 static pthread_t load_session_thread;
215
216 /*
217 * UST registration command queue. This queue is tied with a futex and uses a N
218 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 *
220 * The thread_registration_apps and thread_dispatch_ust_registration uses this
221 * queue along with the wait/wake scheme. The thread_manage_apps receives down
222 * the line new application socket and monitors it for any I/O error or clean
223 * close that triggers an unregistration of the application.
224 */
225 static struct ust_cmd_queue ust_cmd_queue;
226
227 /*
228 * Pointer initialized before thread creation.
229 *
230 * This points to the tracing session list containing the session count and a
231 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
232 * MUST NOT be taken if you call a public function in session.c.
233 *
234 * The lock is nested inside the structure: session_list_ptr->lock. Please use
235 * session_lock_list and session_unlock_list for lock acquisition.
236 */
237 static struct ltt_session_list *session_list_ptr;
238
239 int ust_consumerd64_fd = -1;
240 int ust_consumerd32_fd = -1;
241
242 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
243 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
244 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
245 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
246 static int consumerd32_bin_override;
247 static int consumerd64_bin_override;
248 static int consumerd32_libdir_override;
249 static int consumerd64_libdir_override;
250
251 static const char *module_proc_lttng = "/proc/lttng";
252
253 /*
254 * Consumer daemon state which is changed when spawning it, killing it or in
255 * case of a fatal error.
256 */
257 enum consumerd_state {
258 CONSUMER_STARTED = 1,
259 CONSUMER_STOPPED = 2,
260 CONSUMER_ERROR = 3,
261 };
262
263 /*
264 * This consumer daemon state is used to validate if a client command will be
265 * able to reach the consumer. If not, the client is informed. For instance,
266 * doing a "lttng start" when the consumer state is set to ERROR will return an
267 * error to the client.
268 *
269 * The following example shows a possible race condition of this scheme:
270 *
271 * consumer thread error happens
272 * client cmd arrives
273 * client cmd checks state -> still OK
274 * consumer thread exit, sets error
275 * client cmd try to talk to consumer
276 * ...
277 *
278 * However, since the consumer is a different daemon, we have no way of making
279 * sure the command will reach it safely even with this state flag. This is why
280 * we consider that up to the state validation during command processing, the
281 * command is safe. After that, we can not guarantee the correctness of the
282 * client request vis-a-vis the consumer.
283 */
284 static enum consumerd_state ust_consumerd_state;
285 static enum consumerd_state kernel_consumerd_state;
286
287 /*
288 * Socket timeout for receiving and sending in seconds.
289 */
290 static int app_socket_timeout;
291
292 /* Set in main() with the current page size. */
293 long page_size;
294
295 /* Application health monitoring */
296 struct health_app *health_sessiond;
297
298 /* Agent TCP port for registration. Used by the agent thread. */
299 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
300
301 /* Am I root or not. */
302 int is_root; /* Set to 1 if the daemon is running as root */
303
304 const char * const config_section_name = "sessiond";
305
306 /* Load session thread information to operate. */
307 struct load_session_thread_data *load_info;
308
309 /* Global hash tables */
310 struct lttng_ht *agent_apps_ht_by_sock = NULL;
311
312 /*
313 * Whether sessiond is ready for commands/health check requests.
314 * NR_LTTNG_SESSIOND_READY must match the number of calls to
315 * sessiond_notify_ready().
316 */
317 #define NR_LTTNG_SESSIOND_READY 3
318 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
319
320 /* Notify parents that we are ready for cmd and health check */
321 LTTNG_HIDDEN
322 void sessiond_notify_ready(void)
323 {
324 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
325 /*
326 * Notify parent pid that we are ready to accept command
327 * for client side. This ppid is the one from the
328 * external process that spawned us.
329 */
330 if (opt_sig_parent) {
331 kill(ppid, SIGUSR1);
332 }
333
334 /*
335 * Notify the parent of the fork() process that we are
336 * ready.
337 */
338 if (opt_daemon || opt_background) {
339 kill(child_ppid, SIGUSR1);
340 }
341 }
342 }
343
344 static
345 void setup_consumerd_path(void)
346 {
347 const char *bin, *libdir;
348
349 /*
350 * Allow INSTALL_BIN_PATH to be used as a target path for the
351 * native architecture size consumer if CONFIG_CONSUMER*_PATH
352 * has not been defined.
353 */
354 #if (CAA_BITS_PER_LONG == 32)
355 if (!consumerd32_bin[0]) {
356 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
357 }
358 if (!consumerd32_libdir[0]) {
359 consumerd32_libdir = INSTALL_LIB_PATH;
360 }
361 #elif (CAA_BITS_PER_LONG == 64)
362 if (!consumerd64_bin[0]) {
363 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
364 }
365 if (!consumerd64_libdir[0]) {
366 consumerd64_libdir = INSTALL_LIB_PATH;
367 }
368 #else
369 #error "Unknown bitness"
370 #endif
371
372 /*
373 * runtime env. var. overrides the build default.
374 */
375 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
376 if (bin) {
377 consumerd32_bin = bin;
378 }
379 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
380 if (bin) {
381 consumerd64_bin = bin;
382 }
383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
384 if (libdir) {
385 consumerd32_libdir = libdir;
386 }
387 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
388 if (libdir) {
389 consumerd64_libdir = libdir;
390 }
391 }
392
393 static
394 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
395 int *a_pipe)
396 {
397 int ret;
398
399 assert(events);
400
401 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
402 if (ret < 0) {
403 goto error;
404 }
405
406 /* Add quit pipe */
407 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
408 if (ret < 0) {
409 goto error;
410 }
411
412 return 0;
413
414 error:
415 return ret;
416 }
417
418 /*
419 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
420 */
421 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
422 {
423 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
424 }
425
426 /*
427 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
428 */
429 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
430 size_t size)
431 {
432 return __sessiond_set_thread_pollset(events, size,
433 ht_cleanup_quit_pipe);
434 }
435
436 static
437 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
438 {
439 if (fd == a_pipe && (events & LPOLLIN)) {
440 return 1;
441 }
442 return 0;
443 }
444
445 /*
446 * Check if the thread quit pipe was triggered.
447 *
448 * Return 1 if it was triggered else 0;
449 */
450 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
451 {
452 return __sessiond_check_thread_quit_pipe(fd, events,
453 thread_quit_pipe[0]);
454 }
455
456 /*
457 * Check if the ht_cleanup thread quit pipe was triggered.
458 *
459 * Return 1 if it was triggered else 0;
460 */
461 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
462 {
463 return __sessiond_check_thread_quit_pipe(fd, events,
464 ht_cleanup_quit_pipe[0]);
465 }
466
467 /*
468 * Init thread quit pipe.
469 *
470 * Return -1 on error or 0 if all pipes are created.
471 */
472 static int __init_thread_quit_pipe(int *a_pipe)
473 {
474 int ret, i;
475
476 ret = pipe(a_pipe);
477 if (ret < 0) {
478 PERROR("thread quit pipe");
479 goto error;
480 }
481
482 for (i = 0; i < 2; i++) {
483 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
484 if (ret < 0) {
485 PERROR("fcntl");
486 goto error;
487 }
488 }
489
490 error:
491 return ret;
492 }
493
494 static int init_thread_quit_pipe(void)
495 {
496 return __init_thread_quit_pipe(thread_quit_pipe);
497 }
498
499 static int init_ht_cleanup_quit_pipe(void)
500 {
501 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
502 }
503
504 /*
505 * Stop all threads by closing the thread quit pipe.
506 */
507 static void stop_threads(void)
508 {
509 int ret;
510
511 /* Stopping all threads */
512 DBG("Terminating all threads");
513 ret = notify_thread_pipe(thread_quit_pipe[1]);
514 if (ret < 0) {
515 ERR("write error on thread quit pipe");
516 }
517
518 /* Dispatch thread */
519 CMM_STORE_SHARED(dispatch_thread_exit, 1);
520 futex_nto1_wake(&ust_cmd_queue.futex);
521 }
522
523 /*
524 * Close every consumer sockets.
525 */
526 static void close_consumer_sockets(void)
527 {
528 int ret;
529
530 if (kconsumer_data.err_sock >= 0) {
531 ret = close(kconsumer_data.err_sock);
532 if (ret < 0) {
533 PERROR("kernel consumer err_sock close");
534 }
535 }
536 if (ustconsumer32_data.err_sock >= 0) {
537 ret = close(ustconsumer32_data.err_sock);
538 if (ret < 0) {
539 PERROR("UST consumerd32 err_sock close");
540 }
541 }
542 if (ustconsumer64_data.err_sock >= 0) {
543 ret = close(ustconsumer64_data.err_sock);
544 if (ret < 0) {
545 PERROR("UST consumerd64 err_sock close");
546 }
547 }
548 if (kconsumer_data.cmd_sock >= 0) {
549 ret = close(kconsumer_data.cmd_sock);
550 if (ret < 0) {
551 PERROR("kernel consumer cmd_sock close");
552 }
553 }
554 if (ustconsumer32_data.cmd_sock >= 0) {
555 ret = close(ustconsumer32_data.cmd_sock);
556 if (ret < 0) {
557 PERROR("UST consumerd32 cmd_sock close");
558 }
559 }
560 if (ustconsumer64_data.cmd_sock >= 0) {
561 ret = close(ustconsumer64_data.cmd_sock);
562 if (ret < 0) {
563 PERROR("UST consumerd64 cmd_sock close");
564 }
565 }
566 }
567
568 /*
569 * Generate the full lock file path using the rundir.
570 *
571 * Return the snprintf() return value thus a negative value is an error.
572 */
573 static int generate_lock_file_path(char *path, size_t len)
574 {
575 int ret;
576
577 assert(path);
578 assert(rundir);
579
580 /* Build lockfile path from rundir. */
581 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
582 if (ret < 0) {
583 PERROR("snprintf lockfile path");
584 }
585
586 return ret;
587 }
588
589 /*
590 * Wait on consumer process termination.
591 *
592 * Need to be called with the consumer data lock held or from a context
593 * ensuring no concurrent access to data (e.g: cleanup).
594 */
595 static void wait_consumer(struct consumer_data *consumer_data)
596 {
597 pid_t ret;
598 int status;
599
600 if (consumer_data->pid <= 0) {
601 return;
602 }
603
604 DBG("Waiting for complete teardown of consumerd (PID: %d)",
605 consumer_data->pid);
606 ret = waitpid(consumer_data->pid, &status, 0);
607 if (ret == -1) {
608 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
609 }
610 if (!WIFEXITED(status)) {
611 ERR("consumerd termination with error: %d",
612 WEXITSTATUS(ret));
613 }
614 consumer_data->pid = 0;
615 }
616
617 /*
618 * Cleanup the session daemon's data structures.
619 */
620 static void sessiond_cleanup(void)
621 {
622 int ret;
623 struct ltt_session *sess, *stmp;
624 char path[PATH_MAX];
625
626 DBG("Cleanup sessiond");
627
628 /*
629 * Close the thread quit pipe. It has already done its job,
630 * since we are now called.
631 */
632 utils_close_pipe(thread_quit_pipe);
633
634 /*
635 * If opt_pidfile is undefined, the default file will be wiped when
636 * removing the rundir.
637 */
638 if (opt_pidfile) {
639 ret = remove(opt_pidfile);
640 if (ret < 0) {
641 PERROR("remove pidfile %s", opt_pidfile);
642 }
643 }
644
645 DBG("Removing sessiond and consumerd content of directory %s", rundir);
646
647 /* sessiond */
648 snprintf(path, PATH_MAX,
649 "%s/%s",
650 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
651 DBG("Removing %s", path);
652 (void) unlink(path);
653
654 snprintf(path, PATH_MAX, "%s/%s", rundir,
655 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
656 DBG("Removing %s", path);
657 (void) unlink(path);
658
659 /* kconsumerd */
660 snprintf(path, PATH_MAX,
661 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
662 rundir);
663 DBG("Removing %s", path);
664 (void) unlink(path);
665
666 snprintf(path, PATH_MAX,
667 DEFAULT_KCONSUMERD_PATH,
668 rundir);
669 DBG("Removing directory %s", path);
670 (void) rmdir(path);
671
672 /* ust consumerd 32 */
673 snprintf(path, PATH_MAX,
674 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
675 rundir);
676 DBG("Removing %s", path);
677 (void) unlink(path);
678
679 snprintf(path, PATH_MAX,
680 DEFAULT_USTCONSUMERD32_PATH,
681 rundir);
682 DBG("Removing directory %s", path);
683 (void) rmdir(path);
684
685 /* ust consumerd 64 */
686 snprintf(path, PATH_MAX,
687 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
688 rundir);
689 DBG("Removing %s", path);
690 (void) unlink(path);
691
692 snprintf(path, PATH_MAX,
693 DEFAULT_USTCONSUMERD64_PATH,
694 rundir);
695 DBG("Removing directory %s", path);
696 (void) rmdir(path);
697
698 DBG("Cleaning up all sessions");
699
700 /* Destroy session list mutex */
701 if (session_list_ptr != NULL) {
702 pthread_mutex_destroy(&session_list_ptr->lock);
703
704 /* Cleanup ALL session */
705 cds_list_for_each_entry_safe(sess, stmp,
706 &session_list_ptr->head, list) {
707 cmd_destroy_session(sess, kernel_poll_pipe[1]);
708 }
709 }
710
711 wait_consumer(&kconsumer_data);
712 wait_consumer(&ustconsumer64_data);
713 wait_consumer(&ustconsumer32_data);
714
715 DBG("Cleaning up all agent apps");
716 agent_app_ht_clean();
717
718 DBG("Closing all UST sockets");
719 ust_app_clean_list();
720 buffer_reg_destroy_registries();
721
722 if (is_root && !opt_no_kernel) {
723 DBG2("Closing kernel fd");
724 if (kernel_tracer_fd >= 0) {
725 ret = close(kernel_tracer_fd);
726 if (ret) {
727 PERROR("close");
728 }
729 }
730 DBG("Unloading kernel modules");
731 modprobe_remove_lttng_all();
732 free(syscall_table);
733 }
734
735 close_consumer_sockets();
736
737 if (load_info) {
738 load_session_destroy_data(load_info);
739 free(load_info);
740 }
741
742 /*
743 * Cleanup lock file by deleting it and finaly closing it which will
744 * release the file system lock.
745 */
746 if (lockfile_fd >= 0) {
747 char lockfile_path[PATH_MAX];
748
749 ret = generate_lock_file_path(lockfile_path,
750 sizeof(lockfile_path));
751 if (ret > 0) {
752 ret = remove(lockfile_path);
753 if (ret < 0) {
754 PERROR("remove lock file");
755 }
756 ret = close(lockfile_fd);
757 if (ret < 0) {
758 PERROR("close lock file");
759 }
760 }
761 }
762
763 /*
764 * We do NOT rmdir rundir because there are other processes
765 * using it, for instance lttng-relayd, which can start in
766 * parallel with this teardown.
767 */
768
769 free(rundir);
770 }
771
772 /*
773 * Cleanup the daemon's option data structures.
774 */
775 static void sessiond_cleanup_options(void)
776 {
777 DBG("Cleaning up options");
778
779 /*
780 * If the override option is set, the pointer points to a *non* const
781 * thus freeing it even though the variable type is set to const.
782 */
783 if (tracing_group_name_override) {
784 free((void *) tracing_group_name);
785 }
786 if (consumerd32_bin_override) {
787 free((void *) consumerd32_bin);
788 }
789 if (consumerd64_bin_override) {
790 free((void *) consumerd64_bin);
791 }
792 if (consumerd32_libdir_override) {
793 free((void *) consumerd32_libdir);
794 }
795 if (consumerd64_libdir_override) {
796 free((void *) consumerd64_libdir);
797 }
798
799 free(opt_pidfile);
800 free(opt_load_session_path);
801 free(kmod_probes_list);
802 free(kmod_extra_probes_list);
803
804 /* <fun> */
805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 /* </fun> */
809 }
810
811 /*
812 * Send data on a unix socket using the liblttsessiondcomm API.
813 *
814 * Return lttcomm error code.
815 */
816 static int send_unix_sock(int sock, void *buf, size_t len)
817 {
818 /* Check valid length */
819 if (len == 0) {
820 return -1;
821 }
822
823 return lttcomm_send_unix_sock(sock, buf, len);
824 }
825
826 /*
827 * Free memory of a command context structure.
828 */
829 static void clean_command_ctx(struct command_ctx **cmd_ctx)
830 {
831 DBG("Clean command context structure");
832 if (*cmd_ctx) {
833 if ((*cmd_ctx)->llm) {
834 free((*cmd_ctx)->llm);
835 }
836 if ((*cmd_ctx)->lsm) {
837 free((*cmd_ctx)->lsm);
838 }
839 free(*cmd_ctx);
840 *cmd_ctx = NULL;
841 }
842 }
843
844 /*
845 * Notify UST applications using the shm mmap futex.
846 */
847 static int notify_ust_apps(int active)
848 {
849 char *wait_shm_mmap;
850
851 DBG("Notifying applications of session daemon state: %d", active);
852
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
855 if (wait_shm_mmap == NULL) {
856 goto error;
857 }
858
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap, active);
861
862 /* Apps notified successfully */
863 return 0;
864
865 error:
866 return -1;
867 }
868
869 /*
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
872 * structure.
873 *
874 * Return total size of the buffer pointed by buf.
875 */
876 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
877 {
878 int ret, buf_size;
879
880 buf_size = size;
881
882 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
883 if (cmd_ctx->llm == NULL) {
884 PERROR("zmalloc");
885 ret = -ENOMEM;
886 goto error;
887 }
888
889 /* Copy common data */
890 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
891 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
892
893 cmd_ctx->llm->data_size = size;
894 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
895
896 return buf_size;
897
898 error:
899 return ret;
900 }
901
902 /*
903 * Update the kernel poll set of all channel fd available over all tracing
904 * session. Add the wakeup pipe at the end of the set.
905 */
906 static int update_kernel_poll(struct lttng_poll_event *events)
907 {
908 int ret;
909 struct ltt_session *session;
910 struct ltt_kernel_channel *channel;
911
912 DBG("Updating kernel poll set");
913
914 session_lock_list();
915 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
916 session_lock(session);
917 if (session->kernel_session == NULL) {
918 session_unlock(session);
919 continue;
920 }
921
922 cds_list_for_each_entry(channel,
923 &session->kernel_session->channel_list.head, list) {
924 /* Add channel fd to the kernel poll set */
925 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
926 if (ret < 0) {
927 session_unlock(session);
928 goto error;
929 }
930 DBG("Channel fd %d added to kernel set", channel->fd);
931 }
932 session_unlock(session);
933 }
934 session_unlock_list();
935
936 return 0;
937
938 error:
939 session_unlock_list();
940 return -1;
941 }
942
943 /*
944 * Find the channel fd from 'fd' over all tracing session. When found, check
945 * for new channel stream and send those stream fds to the kernel consumer.
946 *
947 * Useful for CPU hotplug feature.
948 */
949 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
950 {
951 int ret = 0;
952 struct ltt_session *session;
953 struct ltt_kernel_session *ksess;
954 struct ltt_kernel_channel *channel;
955
956 DBG("Updating kernel streams for channel fd %d", fd);
957
958 session_lock_list();
959 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
960 session_lock(session);
961 if (session->kernel_session == NULL) {
962 session_unlock(session);
963 continue;
964 }
965 ksess = session->kernel_session;
966
967 cds_list_for_each_entry(channel,
968 &ksess->channel_list.head, list) {
969 struct lttng_ht_iter iter;
970 struct consumer_socket *socket;
971
972 if (channel->fd != fd) {
973 continue;
974 }
975 DBG("Channel found, updating kernel streams");
976 ret = kernel_open_channel_stream(channel);
977 if (ret < 0) {
978 goto error;
979 }
980 /* Update the stream global counter */
981 ksess->stream_count_global += ret;
982
983 /*
984 * Have we already sent fds to the consumer? If yes, it
985 * means that tracing is started so it is safe to send
986 * our updated stream fds.
987 */
988 if (ksess->consumer_fds_sent != 1
989 || ksess->consumer == NULL) {
990 ret = -1;
991 goto error;
992 }
993
994 rcu_read_lock();
995 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
996 &iter.iter, socket, node.node) {
997 pthread_mutex_lock(socket->lock);
998 ret = kernel_consumer_send_channel_stream(socket,
999 channel, ksess,
1000 session->output_traces ? 1 : 0);
1001 pthread_mutex_unlock(socket->lock);
1002 if (ret < 0) {
1003 rcu_read_unlock();
1004 goto error;
1005 }
1006 }
1007 rcu_read_unlock();
1008 }
1009 session_unlock(session);
1010 }
1011 session_unlock_list();
1012 return ret;
1013
1014 error:
1015 session_unlock(session);
1016 session_unlock_list();
1017 return ret;
1018 }
1019
1020 /*
1021 * For each tracing session, update newly registered apps. The session list
1022 * lock MUST be acquired before calling this.
1023 */
1024 static void update_ust_app(int app_sock)
1025 {
1026 struct ltt_session *sess, *stmp;
1027
1028 /* Consumer is in an ERROR state. Stop any application update. */
1029 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1030 /* Stop the update process since the consumer is dead. */
1031 return;
1032 }
1033
1034 /* For all tracing session(s) */
1035 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1036 struct ust_app *app;
1037
1038 session_lock(sess);
1039 if (!sess->ust_session) {
1040 goto unlock_session;
1041 }
1042
1043 rcu_read_lock();
1044 assert(app_sock >= 0);
1045 app = ust_app_find_by_sock(app_sock);
1046 if (app == NULL) {
1047 /*
1048 * Application can be unregistered before so
1049 * this is possible hence simply stopping the
1050 * update.
1051 */
1052 DBG3("UST app update failed to find app sock %d",
1053 app_sock);
1054 goto unlock_rcu;
1055 }
1056 ust_app_global_update(sess->ust_session, app);
1057 unlock_rcu:
1058 rcu_read_unlock();
1059 unlock_session:
1060 session_unlock(sess);
1061 }
1062 }
1063
1064 /*
1065 * This thread manage event coming from the kernel.
1066 *
1067 * Features supported in this thread:
1068 * -) CPU Hotplug
1069 */
1070 static void *thread_manage_kernel(void *data)
1071 {
1072 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1073 uint32_t revents, nb_fd;
1074 char tmp;
1075 struct lttng_poll_event events;
1076
1077 DBG("[thread] Thread manage kernel started");
1078
1079 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1080
1081 /*
1082 * This first step of the while is to clean this structure which could free
1083 * non NULL pointers so initialize it before the loop.
1084 */
1085 lttng_poll_init(&events);
1086
1087 if (testpoint(sessiond_thread_manage_kernel)) {
1088 goto error_testpoint;
1089 }
1090
1091 health_code_update();
1092
1093 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1094 goto error_testpoint;
1095 }
1096
1097 while (1) {
1098 health_code_update();
1099
1100 if (update_poll_flag == 1) {
1101 /* Clean events object. We are about to populate it again. */
1102 lttng_poll_clean(&events);
1103
1104 ret = sessiond_set_thread_pollset(&events, 2);
1105 if (ret < 0) {
1106 goto error_poll_create;
1107 }
1108
1109 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1110 if (ret < 0) {
1111 goto error;
1112 }
1113
1114 /* This will add the available kernel channel if any. */
1115 ret = update_kernel_poll(&events);
1116 if (ret < 0) {
1117 goto error;
1118 }
1119 update_poll_flag = 0;
1120 }
1121
1122 DBG("Thread kernel polling");
1123
1124 /* Poll infinite value of time */
1125 restart:
1126 health_poll_entry();
1127 ret = lttng_poll_wait(&events, -1);
1128 DBG("Thread kernel return from poll on %d fds",
1129 LTTNG_POLL_GETNB(&events));
1130 health_poll_exit();
1131 if (ret < 0) {
1132 /*
1133 * Restart interrupted system call.
1134 */
1135 if (errno == EINTR) {
1136 goto restart;
1137 }
1138 goto error;
1139 } else if (ret == 0) {
1140 /* Should not happen since timeout is infinite */
1141 ERR("Return value of poll is 0 with an infinite timeout.\n"
1142 "This should not have happened! Continuing...");
1143 continue;
1144 }
1145
1146 nb_fd = ret;
1147
1148 for (i = 0; i < nb_fd; i++) {
1149 /* Fetch once the poll data */
1150 revents = LTTNG_POLL_GETEV(&events, i);
1151 pollfd = LTTNG_POLL_GETFD(&events, i);
1152
1153 health_code_update();
1154
1155 if (!revents) {
1156 /* No activity for this FD (poll implementation). */
1157 continue;
1158 }
1159
1160 /* Thread quit pipe has been closed. Killing thread. */
1161 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1162 if (ret) {
1163 err = 0;
1164 goto exit;
1165 }
1166
1167 /* Check for data on kernel pipe */
1168 if (revents & LPOLLIN) {
1169 if (pollfd == kernel_poll_pipe[0]) {
1170 (void) lttng_read(kernel_poll_pipe[0],
1171 &tmp, 1);
1172 /*
1173 * Ret value is useless here, if this pipe gets any actions an
1174 * update is required anyway.
1175 */
1176 update_poll_flag = 1;
1177 continue;
1178 } else {
1179 /*
1180 * New CPU detected by the kernel. Adding kernel stream to
1181 * kernel session and updating the kernel consumer
1182 */
1183 ret = update_kernel_stream(&kconsumer_data, pollfd);
1184 if (ret < 0) {
1185 continue;
1186 }
1187 break;
1188 }
1189 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1190 update_poll_flag = 1;
1191 continue;
1192 } else {
1193 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1194 goto error;
1195 }
1196 }
1197 }
1198
1199 exit:
1200 error:
1201 lttng_poll_clean(&events);
1202 error_poll_create:
1203 error_testpoint:
1204 utils_close_pipe(kernel_poll_pipe);
1205 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1206 if (err) {
1207 health_error();
1208 ERR("Health error occurred in %s", __func__);
1209 WARN("Kernel thread died unexpectedly. "
1210 "Kernel tracing can continue but CPU hotplug is disabled.");
1211 }
1212 health_unregister(health_sessiond);
1213 DBG("Kernel thread dying");
1214 return NULL;
1215 }
1216
1217 /*
1218 * Signal pthread condition of the consumer data that the thread.
1219 */
1220 static void signal_consumer_condition(struct consumer_data *data, int state)
1221 {
1222 pthread_mutex_lock(&data->cond_mutex);
1223
1224 /*
1225 * The state is set before signaling. It can be any value, it's the waiter
1226 * job to correctly interpret this condition variable associated to the
1227 * consumer pthread_cond.
1228 *
1229 * A value of 0 means that the corresponding thread of the consumer data
1230 * was not started. 1 indicates that the thread has started and is ready
1231 * for action. A negative value means that there was an error during the
1232 * thread bootstrap.
1233 */
1234 data->consumer_thread_is_ready = state;
1235 (void) pthread_cond_signal(&data->cond);
1236
1237 pthread_mutex_unlock(&data->cond_mutex);
1238 }
1239
1240 /*
1241 * This thread manage the consumer error sent back to the session daemon.
1242 */
1243 static void *thread_manage_consumer(void *data)
1244 {
1245 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1246 uint32_t revents, nb_fd;
1247 enum lttcomm_return_code code;
1248 struct lttng_poll_event events;
1249 struct consumer_data *consumer_data = data;
1250
1251 DBG("[thread] Manage consumer started");
1252
1253 rcu_register_thread();
1254 rcu_thread_online();
1255
1256 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1257
1258 health_code_update();
1259
1260 /*
1261 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1262 * metadata_sock. Nothing more will be added to this poll set.
1263 */
1264 ret = sessiond_set_thread_pollset(&events, 3);
1265 if (ret < 0) {
1266 goto error_poll;
1267 }
1268
1269 /*
1270 * The error socket here is already in a listening state which was done
1271 * just before spawning this thread to avoid a race between the consumer
1272 * daemon exec trying to connect and the listen() call.
1273 */
1274 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1275 if (ret < 0) {
1276 goto error;
1277 }
1278
1279 health_code_update();
1280
1281 /* Infinite blocking call, waiting for transmission */
1282 restart:
1283 health_poll_entry();
1284
1285 if (testpoint(sessiond_thread_manage_consumer)) {
1286 goto error;
1287 }
1288
1289 ret = lttng_poll_wait(&events, -1);
1290 health_poll_exit();
1291 if (ret < 0) {
1292 /*
1293 * Restart interrupted system call.
1294 */
1295 if (errno == EINTR) {
1296 goto restart;
1297 }
1298 goto error;
1299 }
1300
1301 nb_fd = ret;
1302
1303 for (i = 0; i < nb_fd; i++) {
1304 /* Fetch once the poll data */
1305 revents = LTTNG_POLL_GETEV(&events, i);
1306 pollfd = LTTNG_POLL_GETFD(&events, i);
1307
1308 health_code_update();
1309
1310 if (!revents) {
1311 /* No activity for this FD (poll implementation). */
1312 continue;
1313 }
1314
1315 /* Thread quit pipe has been closed. Killing thread. */
1316 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1317 if (ret) {
1318 err = 0;
1319 goto exit;
1320 }
1321
1322 /* Event on the registration socket */
1323 if (pollfd == consumer_data->err_sock) {
1324 if (revents & LPOLLIN) {
1325 continue;
1326 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1327 ERR("consumer err socket poll error");
1328 goto error;
1329 } else {
1330 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1331 goto error;
1332 }
1333 }
1334 }
1335
1336 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1337 if (sock < 0) {
1338 goto error;
1339 }
1340
1341 /*
1342 * Set the CLOEXEC flag. Return code is useless because either way, the
1343 * show must go on.
1344 */
1345 (void) utils_set_fd_cloexec(sock);
1346
1347 health_code_update();
1348
1349 DBG2("Receiving code from consumer err_sock");
1350
1351 /* Getting status code from kconsumerd */
1352 ret = lttcomm_recv_unix_sock(sock, &code,
1353 sizeof(enum lttcomm_return_code));
1354 if (ret <= 0) {
1355 goto error;
1356 }
1357
1358 health_code_update();
1359 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1360 /* Connect both socket, command and metadata. */
1361 consumer_data->cmd_sock =
1362 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1363 consumer_data->metadata_fd =
1364 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1365 if (consumer_data->cmd_sock < 0
1366 || consumer_data->metadata_fd < 0) {
1367 PERROR("consumer connect cmd socket");
1368 /* On error, signal condition and quit. */
1369 signal_consumer_condition(consumer_data, -1);
1370 goto error;
1371 }
1372 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1373 /* Create metadata socket lock. */
1374 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1375 if (consumer_data->metadata_sock.lock == NULL) {
1376 PERROR("zmalloc pthread mutex");
1377 ret = -1;
1378 goto error;
1379 }
1380 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1381
1382 signal_consumer_condition(consumer_data, 1);
1383 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1384 DBG("Consumer metadata socket ready (fd: %d)",
1385 consumer_data->metadata_fd);
1386 } else {
1387 ERR("consumer error when waiting for SOCK_READY : %s",
1388 lttcomm_get_readable_code(-code));
1389 goto error;
1390 }
1391
1392 /* Remove the consumerd error sock since we've established a connexion */
1393 ret = lttng_poll_del(&events, consumer_data->err_sock);
1394 if (ret < 0) {
1395 goto error;
1396 }
1397
1398 /* Add new accepted error socket. */
1399 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1400 if (ret < 0) {
1401 goto error;
1402 }
1403
1404 /* Add metadata socket that is successfully connected. */
1405 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1406 LPOLLIN | LPOLLRDHUP);
1407 if (ret < 0) {
1408 goto error;
1409 }
1410
1411 health_code_update();
1412
1413 /* Infinite blocking call, waiting for transmission */
1414 restart_poll:
1415 while (1) {
1416 health_code_update();
1417
1418 /* Exit the thread because the thread quit pipe has been triggered. */
1419 if (should_quit) {
1420 /* Not a health error. */
1421 err = 0;
1422 goto exit;
1423 }
1424
1425 health_poll_entry();
1426 ret = lttng_poll_wait(&events, -1);
1427 health_poll_exit();
1428 if (ret < 0) {
1429 /*
1430 * Restart interrupted system call.
1431 */
1432 if (errno == EINTR) {
1433 goto restart_poll;
1434 }
1435 goto error;
1436 }
1437
1438 nb_fd = ret;
1439
1440 for (i = 0; i < nb_fd; i++) {
1441 /* Fetch once the poll data */
1442 revents = LTTNG_POLL_GETEV(&events, i);
1443 pollfd = LTTNG_POLL_GETFD(&events, i);
1444
1445 health_code_update();
1446
1447 if (!revents) {
1448 /* No activity for this FD (poll implementation). */
1449 continue;
1450 }
1451
1452 /*
1453 * Thread quit pipe has been triggered, flag that we should stop
1454 * but continue the current loop to handle potential data from
1455 * consumer.
1456 */
1457 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1458
1459 if (pollfd == sock) {
1460 /* Event on the consumerd socket */
1461 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1462 && !(revents & LPOLLIN)) {
1463 ERR("consumer err socket second poll error");
1464 goto error;
1465 }
1466 health_code_update();
1467 /* Wait for any kconsumerd error */
1468 ret = lttcomm_recv_unix_sock(sock, &code,
1469 sizeof(enum lttcomm_return_code));
1470 if (ret <= 0) {
1471 ERR("consumer closed the command socket");
1472 goto error;
1473 }
1474
1475 ERR("consumer return code : %s",
1476 lttcomm_get_readable_code(-code));
1477
1478 goto exit;
1479 } else if (pollfd == consumer_data->metadata_fd) {
1480 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1481 && !(revents & LPOLLIN)) {
1482 ERR("consumer err metadata socket second poll error");
1483 goto error;
1484 }
1485 /* UST metadata requests */
1486 ret = ust_consumer_metadata_request(
1487 &consumer_data->metadata_sock);
1488 if (ret < 0) {
1489 ERR("Handling metadata request");
1490 goto error;
1491 }
1492 }
1493 /* No need for an else branch all FDs are tested prior. */
1494 }
1495 health_code_update();
1496 }
1497
1498 exit:
1499 error:
1500 /*
1501 * We lock here because we are about to close the sockets and some other
1502 * thread might be using them so get exclusive access which will abort all
1503 * other consumer command by other threads.
1504 */
1505 pthread_mutex_lock(&consumer_data->lock);
1506
1507 /* Immediately set the consumerd state to stopped */
1508 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1509 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1510 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1511 consumer_data->type == LTTNG_CONSUMER32_UST) {
1512 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1513 } else {
1514 /* Code flow error... */
1515 assert(0);
1516 }
1517
1518 if (consumer_data->err_sock >= 0) {
1519 ret = close(consumer_data->err_sock);
1520 if (ret) {
1521 PERROR("close");
1522 }
1523 consumer_data->err_sock = -1;
1524 }
1525 if (consumer_data->cmd_sock >= 0) {
1526 ret = close(consumer_data->cmd_sock);
1527 if (ret) {
1528 PERROR("close");
1529 }
1530 consumer_data->cmd_sock = -1;
1531 }
1532 if (consumer_data->metadata_sock.fd_ptr &&
1533 *consumer_data->metadata_sock.fd_ptr >= 0) {
1534 ret = close(*consumer_data->metadata_sock.fd_ptr);
1535 if (ret) {
1536 PERROR("close");
1537 }
1538 }
1539 if (sock >= 0) {
1540 ret = close(sock);
1541 if (ret) {
1542 PERROR("close");
1543 }
1544 }
1545
1546 unlink(consumer_data->err_unix_sock_path);
1547 unlink(consumer_data->cmd_unix_sock_path);
1548 pthread_mutex_unlock(&consumer_data->lock);
1549
1550 /* Cleanup metadata socket mutex. */
1551 if (consumer_data->metadata_sock.lock) {
1552 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1553 free(consumer_data->metadata_sock.lock);
1554 }
1555 lttng_poll_clean(&events);
1556 error_poll:
1557 if (err) {
1558 health_error();
1559 ERR("Health error occurred in %s", __func__);
1560 }
1561 health_unregister(health_sessiond);
1562 DBG("consumer thread cleanup completed");
1563
1564 rcu_thread_offline();
1565 rcu_unregister_thread();
1566
1567 return NULL;
1568 }
1569
1570 /*
1571 * This thread manage application communication.
1572 */
1573 static void *thread_manage_apps(void *data)
1574 {
1575 int i, ret, pollfd, err = -1;
1576 ssize_t size_ret;
1577 uint32_t revents, nb_fd;
1578 struct lttng_poll_event events;
1579
1580 DBG("[thread] Manage application started");
1581
1582 rcu_register_thread();
1583 rcu_thread_online();
1584
1585 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1586
1587 if (testpoint(sessiond_thread_manage_apps)) {
1588 goto error_testpoint;
1589 }
1590
1591 health_code_update();
1592
1593 ret = sessiond_set_thread_pollset(&events, 2);
1594 if (ret < 0) {
1595 goto error_poll_create;
1596 }
1597
1598 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1599 if (ret < 0) {
1600 goto error;
1601 }
1602
1603 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1604 goto error;
1605 }
1606
1607 health_code_update();
1608
1609 while (1) {
1610 DBG("Apps thread polling");
1611
1612 /* Inifinite blocking call, waiting for transmission */
1613 restart:
1614 health_poll_entry();
1615 ret = lttng_poll_wait(&events, -1);
1616 DBG("Apps thread return from poll on %d fds",
1617 LTTNG_POLL_GETNB(&events));
1618 health_poll_exit();
1619 if (ret < 0) {
1620 /*
1621 * Restart interrupted system call.
1622 */
1623 if (errno == EINTR) {
1624 goto restart;
1625 }
1626 goto error;
1627 }
1628
1629 nb_fd = ret;
1630
1631 for (i = 0; i < nb_fd; i++) {
1632 /* Fetch once the poll data */
1633 revents = LTTNG_POLL_GETEV(&events, i);
1634 pollfd = LTTNG_POLL_GETFD(&events, i);
1635
1636 health_code_update();
1637
1638 if (!revents) {
1639 /* No activity for this FD (poll implementation). */
1640 continue;
1641 }
1642
1643 /* Thread quit pipe has been closed. Killing thread. */
1644 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1645 if (ret) {
1646 err = 0;
1647 goto exit;
1648 }
1649
1650 /* Inspect the apps cmd pipe */
1651 if (pollfd == apps_cmd_pipe[0]) {
1652 if (revents & LPOLLIN) {
1653 int sock;
1654
1655 /* Empty pipe */
1656 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1657 if (size_ret < sizeof(sock)) {
1658 PERROR("read apps cmd pipe");
1659 goto error;
1660 }
1661
1662 health_code_update();
1663
1664 /*
1665 * Since this is a command socket (write then read),
1666 * we only monitor the error events of the socket.
1667 */
1668 ret = lttng_poll_add(&events, sock,
1669 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1670 if (ret < 0) {
1671 goto error;
1672 }
1673
1674 DBG("Apps with sock %d added to poll set", sock);
1675 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1676 ERR("Apps command pipe error");
1677 goto error;
1678 } else {
1679 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1680 goto error;
1681 }
1682 } else {
1683 /*
1684 * At this point, we know that a registered application made
1685 * the event at poll_wait.
1686 */
1687 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1688 /* Removing from the poll set */
1689 ret = lttng_poll_del(&events, pollfd);
1690 if (ret < 0) {
1691 goto error;
1692 }
1693
1694 /* Socket closed on remote end. */
1695 ust_app_unregister(pollfd);
1696 } else {
1697 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1698 goto error;
1699 }
1700 }
1701
1702 health_code_update();
1703 }
1704 }
1705
1706 exit:
1707 error:
1708 lttng_poll_clean(&events);
1709 error_poll_create:
1710 error_testpoint:
1711 utils_close_pipe(apps_cmd_pipe);
1712 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1713
1714 /*
1715 * We don't clean the UST app hash table here since already registered
1716 * applications can still be controlled so let them be until the session
1717 * daemon dies or the applications stop.
1718 */
1719
1720 if (err) {
1721 health_error();
1722 ERR("Health error occurred in %s", __func__);
1723 }
1724 health_unregister(health_sessiond);
1725 DBG("Application communication apps thread cleanup complete");
1726 rcu_thread_offline();
1727 rcu_unregister_thread();
1728 return NULL;
1729 }
1730
1731 /*
1732 * Send a socket to a thread This is called from the dispatch UST registration
1733 * thread once all sockets are set for the application.
1734 *
1735 * The sock value can be invalid, we don't really care, the thread will handle
1736 * it and make the necessary cleanup if so.
1737 *
1738 * On success, return 0 else a negative value being the errno message of the
1739 * write().
1740 */
1741 static int send_socket_to_thread(int fd, int sock)
1742 {
1743 ssize_t ret;
1744
1745 /*
1746 * It's possible that the FD is set as invalid with -1 concurrently just
1747 * before calling this function being a shutdown state of the thread.
1748 */
1749 if (fd < 0) {
1750 ret = -EBADF;
1751 goto error;
1752 }
1753
1754 ret = lttng_write(fd, &sock, sizeof(sock));
1755 if (ret < sizeof(sock)) {
1756 PERROR("write apps pipe %d", fd);
1757 if (ret < 0) {
1758 ret = -errno;
1759 }
1760 goto error;
1761 }
1762
1763 /* All good. Don't send back the write positive ret value. */
1764 ret = 0;
1765 error:
1766 return (int) ret;
1767 }
1768
1769 /*
1770 * Sanitize the wait queue of the dispatch registration thread meaning removing
1771 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1772 * notify socket is never received.
1773 */
1774 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1775 {
1776 int ret, nb_fd = 0, i;
1777 unsigned int fd_added = 0;
1778 struct lttng_poll_event events;
1779 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1780
1781 assert(wait_queue);
1782
1783 lttng_poll_init(&events);
1784
1785 /* Just skip everything for an empty queue. */
1786 if (!wait_queue->count) {
1787 goto end;
1788 }
1789
1790 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1791 if (ret < 0) {
1792 goto error_create;
1793 }
1794
1795 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1796 &wait_queue->head, head) {
1797 assert(wait_node->app);
1798 ret = lttng_poll_add(&events, wait_node->app->sock,
1799 LPOLLHUP | LPOLLERR);
1800 if (ret < 0) {
1801 goto error;
1802 }
1803
1804 fd_added = 1;
1805 }
1806
1807 if (!fd_added) {
1808 goto end;
1809 }
1810
1811 /*
1812 * Poll but don't block so we can quickly identify the faulty events and
1813 * clean them afterwards from the wait queue.
1814 */
1815 ret = lttng_poll_wait(&events, 0);
1816 if (ret < 0) {
1817 goto error;
1818 }
1819 nb_fd = ret;
1820
1821 for (i = 0; i < nb_fd; i++) {
1822 /* Get faulty FD. */
1823 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1824 int pollfd = LTTNG_POLL_GETFD(&events, i);
1825
1826 if (!revents) {
1827 /* No activity for this FD (poll implementation). */
1828 continue;
1829 }
1830
1831 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1832 &wait_queue->head, head) {
1833 if (pollfd == wait_node->app->sock &&
1834 (revents & (LPOLLHUP | LPOLLERR))) {
1835 cds_list_del(&wait_node->head);
1836 wait_queue->count--;
1837 ust_app_destroy(wait_node->app);
1838 free(wait_node);
1839 /*
1840 * Silence warning of use-after-free in
1841 * cds_list_for_each_entry_safe which uses
1842 * __typeof__(*wait_node).
1843 */
1844 wait_node = NULL;
1845 break;
1846 } else {
1847 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1848 goto error;
1849 }
1850 }
1851 }
1852
1853 if (nb_fd > 0) {
1854 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1855 }
1856
1857 end:
1858 lttng_poll_clean(&events);
1859 return;
1860
1861 error:
1862 lttng_poll_clean(&events);
1863 error_create:
1864 ERR("Unable to sanitize wait queue");
1865 return;
1866 }
1867
1868 /*
1869 * Dispatch request from the registration threads to the application
1870 * communication thread.
1871 */
1872 static void *thread_dispatch_ust_registration(void *data)
1873 {
1874 int ret, err = -1;
1875 struct cds_wfcq_node *node;
1876 struct ust_command *ust_cmd = NULL;
1877 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1878 struct ust_reg_wait_queue wait_queue = {
1879 .count = 0,
1880 };
1881
1882 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1883
1884 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1885 goto error_testpoint;
1886 }
1887
1888 health_code_update();
1889
1890 CDS_INIT_LIST_HEAD(&wait_queue.head);
1891
1892 DBG("[thread] Dispatch UST command started");
1893
1894 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1895 health_code_update();
1896
1897 /* Atomically prepare the queue futex */
1898 futex_nto1_prepare(&ust_cmd_queue.futex);
1899
1900 do {
1901 struct ust_app *app = NULL;
1902 ust_cmd = NULL;
1903
1904 /*
1905 * Make sure we don't have node(s) that have hung up before receiving
1906 * the notify socket. This is to clean the list in order to avoid
1907 * memory leaks from notify socket that are never seen.
1908 */
1909 sanitize_wait_queue(&wait_queue);
1910
1911 health_code_update();
1912 /* Dequeue command for registration */
1913 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1914 if (node == NULL) {
1915 DBG("Woken up but nothing in the UST command queue");
1916 /* Continue thread execution */
1917 break;
1918 }
1919
1920 ust_cmd = caa_container_of(node, struct ust_command, node);
1921
1922 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1923 " gid:%d sock:%d name:%s (version %d.%d)",
1924 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1925 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1926 ust_cmd->sock, ust_cmd->reg_msg.name,
1927 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1928
1929 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1930 wait_node = zmalloc(sizeof(*wait_node));
1931 if (!wait_node) {
1932 PERROR("zmalloc wait_node dispatch");
1933 ret = close(ust_cmd->sock);
1934 if (ret < 0) {
1935 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1936 }
1937 lttng_fd_put(LTTNG_FD_APPS, 1);
1938 free(ust_cmd);
1939 goto error;
1940 }
1941 CDS_INIT_LIST_HEAD(&wait_node->head);
1942
1943 /* Create application object if socket is CMD. */
1944 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1945 ust_cmd->sock);
1946 if (!wait_node->app) {
1947 ret = close(ust_cmd->sock);
1948 if (ret < 0) {
1949 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1950 }
1951 lttng_fd_put(LTTNG_FD_APPS, 1);
1952 free(wait_node);
1953 free(ust_cmd);
1954 continue;
1955 }
1956 /*
1957 * Add application to the wait queue so we can set the notify
1958 * socket before putting this object in the global ht.
1959 */
1960 cds_list_add(&wait_node->head, &wait_queue.head);
1961 wait_queue.count++;
1962
1963 free(ust_cmd);
1964 /*
1965 * We have to continue here since we don't have the notify
1966 * socket and the application MUST be added to the hash table
1967 * only at that moment.
1968 */
1969 continue;
1970 } else {
1971 /*
1972 * Look for the application in the local wait queue and set the
1973 * notify socket if found.
1974 */
1975 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1976 &wait_queue.head, head) {
1977 health_code_update();
1978 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1979 wait_node->app->notify_sock = ust_cmd->sock;
1980 cds_list_del(&wait_node->head);
1981 wait_queue.count--;
1982 app = wait_node->app;
1983 free(wait_node);
1984 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1985 break;
1986 }
1987 }
1988
1989 /*
1990 * With no application at this stage the received socket is
1991 * basically useless so close it before we free the cmd data
1992 * structure for good.
1993 */
1994 if (!app) {
1995 ret = close(ust_cmd->sock);
1996 if (ret < 0) {
1997 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1998 }
1999 lttng_fd_put(LTTNG_FD_APPS, 1);
2000 }
2001 free(ust_cmd);
2002 }
2003
2004 if (app) {
2005 /*
2006 * @session_lock_list
2007 *
2008 * Lock the global session list so from the register up to the
2009 * registration done message, no thread can see the application
2010 * and change its state.
2011 */
2012 session_lock_list();
2013 rcu_read_lock();
2014
2015 /*
2016 * Add application to the global hash table. This needs to be
2017 * done before the update to the UST registry can locate the
2018 * application.
2019 */
2020 ust_app_add(app);
2021
2022 /* Set app version. This call will print an error if needed. */
2023 (void) ust_app_version(app);
2024
2025 /* Send notify socket through the notify pipe. */
2026 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2027 app->notify_sock);
2028 if (ret < 0) {
2029 rcu_read_unlock();
2030 session_unlock_list();
2031 /*
2032 * No notify thread, stop the UST tracing. However, this is
2033 * not an internal error of the this thread thus setting
2034 * the health error code to a normal exit.
2035 */
2036 err = 0;
2037 goto error;
2038 }
2039
2040 /*
2041 * Update newly registered application with the tracing
2042 * registry info already enabled information.
2043 */
2044 update_ust_app(app->sock);
2045
2046 /*
2047 * Don't care about return value. Let the manage apps threads
2048 * handle app unregistration upon socket close.
2049 */
2050 (void) ust_app_register_done(app);
2051
2052 /*
2053 * Even if the application socket has been closed, send the app
2054 * to the thread and unregistration will take place at that
2055 * place.
2056 */
2057 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2058 if (ret < 0) {
2059 rcu_read_unlock();
2060 session_unlock_list();
2061 /*
2062 * No apps. thread, stop the UST tracing. However, this is
2063 * not an internal error of the this thread thus setting
2064 * the health error code to a normal exit.
2065 */
2066 err = 0;
2067 goto error;
2068 }
2069
2070 rcu_read_unlock();
2071 session_unlock_list();
2072 }
2073 } while (node != NULL);
2074
2075 health_poll_entry();
2076 /* Futex wait on queue. Blocking call on futex() */
2077 futex_nto1_wait(&ust_cmd_queue.futex);
2078 health_poll_exit();
2079 }
2080 /* Normal exit, no error */
2081 err = 0;
2082
2083 error:
2084 /* Clean up wait queue. */
2085 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2086 &wait_queue.head, head) {
2087 cds_list_del(&wait_node->head);
2088 wait_queue.count--;
2089 free(wait_node);
2090 }
2091
2092 /* Empty command queue. */
2093 for (;;) {
2094 /* Dequeue command for registration */
2095 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2096 if (node == NULL) {
2097 break;
2098 }
2099 ust_cmd = caa_container_of(node, struct ust_command, node);
2100 ret = close(ust_cmd->sock);
2101 if (ret < 0) {
2102 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2103 }
2104 lttng_fd_put(LTTNG_FD_APPS, 1);
2105 free(ust_cmd);
2106 }
2107
2108 error_testpoint:
2109 DBG("Dispatch thread dying");
2110 if (err) {
2111 health_error();
2112 ERR("Health error occurred in %s", __func__);
2113 }
2114 health_unregister(health_sessiond);
2115 return NULL;
2116 }
2117
2118 /*
2119 * This thread manage application registration.
2120 */
2121 static void *thread_registration_apps(void *data)
2122 {
2123 int sock = -1, i, ret, pollfd, err = -1;
2124 uint32_t revents, nb_fd;
2125 struct lttng_poll_event events;
2126 /*
2127 * Get allocated in this thread, enqueued to a global queue, dequeued and
2128 * freed in the manage apps thread.
2129 */
2130 struct ust_command *ust_cmd = NULL;
2131
2132 DBG("[thread] Manage application registration started");
2133
2134 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2135
2136 if (testpoint(sessiond_thread_registration_apps)) {
2137 goto error_testpoint;
2138 }
2139
2140 ret = lttcomm_listen_unix_sock(apps_sock);
2141 if (ret < 0) {
2142 goto error_listen;
2143 }
2144
2145 /*
2146 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2147 * more will be added to this poll set.
2148 */
2149 ret = sessiond_set_thread_pollset(&events, 2);
2150 if (ret < 0) {
2151 goto error_create_poll;
2152 }
2153
2154 /* Add the application registration socket */
2155 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2156 if (ret < 0) {
2157 goto error_poll_add;
2158 }
2159
2160 /* Notify all applications to register */
2161 ret = notify_ust_apps(1);
2162 if (ret < 0) {
2163 ERR("Failed to notify applications or create the wait shared memory.\n"
2164 "Execution continues but there might be problem for already\n"
2165 "running applications that wishes to register.");
2166 }
2167
2168 while (1) {
2169 DBG("Accepting application registration");
2170
2171 /* Inifinite blocking call, waiting for transmission */
2172 restart:
2173 health_poll_entry();
2174 ret = lttng_poll_wait(&events, -1);
2175 health_poll_exit();
2176 if (ret < 0) {
2177 /*
2178 * Restart interrupted system call.
2179 */
2180 if (errno == EINTR) {
2181 goto restart;
2182 }
2183 goto error;
2184 }
2185
2186 nb_fd = ret;
2187
2188 for (i = 0; i < nb_fd; i++) {
2189 health_code_update();
2190
2191 /* Fetch once the poll data */
2192 revents = LTTNG_POLL_GETEV(&events, i);
2193 pollfd = LTTNG_POLL_GETFD(&events, i);
2194
2195 if (!revents) {
2196 /* No activity for this FD (poll implementation). */
2197 continue;
2198 }
2199
2200 /* Thread quit pipe has been closed. Killing thread. */
2201 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2202 if (ret) {
2203 err = 0;
2204 goto exit;
2205 }
2206
2207 /* Event on the registration socket */
2208 if (pollfd == apps_sock) {
2209 if (revents & LPOLLIN) {
2210 sock = lttcomm_accept_unix_sock(apps_sock);
2211 if (sock < 0) {
2212 goto error;
2213 }
2214
2215 /*
2216 * Set socket timeout for both receiving and ending.
2217 * app_socket_timeout is in seconds, whereas
2218 * lttcomm_setsockopt_rcv_timeout and
2219 * lttcomm_setsockopt_snd_timeout expect msec as
2220 * parameter.
2221 */
2222 (void) lttcomm_setsockopt_rcv_timeout(sock,
2223 app_socket_timeout * 1000);
2224 (void) lttcomm_setsockopt_snd_timeout(sock,
2225 app_socket_timeout * 1000);
2226
2227 /*
2228 * Set the CLOEXEC flag. Return code is useless because
2229 * either way, the show must go on.
2230 */
2231 (void) utils_set_fd_cloexec(sock);
2232
2233 /* Create UST registration command for enqueuing */
2234 ust_cmd = zmalloc(sizeof(struct ust_command));
2235 if (ust_cmd == NULL) {
2236 PERROR("ust command zmalloc");
2237 ret = close(sock);
2238 if (ret) {
2239 PERROR("close");
2240 }
2241 goto error;
2242 }
2243
2244 /*
2245 * Using message-based transmissions to ensure we don't
2246 * have to deal with partially received messages.
2247 */
2248 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2249 if (ret < 0) {
2250 ERR("Exhausted file descriptors allowed for applications.");
2251 free(ust_cmd);
2252 ret = close(sock);
2253 if (ret) {
2254 PERROR("close");
2255 }
2256 sock = -1;
2257 continue;
2258 }
2259
2260 health_code_update();
2261 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2262 if (ret < 0) {
2263 free(ust_cmd);
2264 /* Close socket of the application. */
2265 ret = close(sock);
2266 if (ret) {
2267 PERROR("close");
2268 }
2269 lttng_fd_put(LTTNG_FD_APPS, 1);
2270 sock = -1;
2271 continue;
2272 }
2273 health_code_update();
2274
2275 ust_cmd->sock = sock;
2276 sock = -1;
2277
2278 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2279 " gid:%d sock:%d name:%s (version %d.%d)",
2280 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2281 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2282 ust_cmd->sock, ust_cmd->reg_msg.name,
2283 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2284
2285 /*
2286 * Lock free enqueue the registration request. The red pill
2287 * has been taken! This apps will be part of the *system*.
2288 */
2289 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2290
2291 /*
2292 * Wake the registration queue futex. Implicit memory
2293 * barrier with the exchange in cds_wfcq_enqueue.
2294 */
2295 futex_nto1_wake(&ust_cmd_queue.futex);
2296 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2297 ERR("Register apps socket poll error");
2298 goto error;
2299 } else {
2300 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2301 goto error;
2302 }
2303 }
2304 }
2305 }
2306
2307 exit:
2308 error:
2309 /* Notify that the registration thread is gone */
2310 notify_ust_apps(0);
2311
2312 if (apps_sock >= 0) {
2313 ret = close(apps_sock);
2314 if (ret) {
2315 PERROR("close");
2316 }
2317 }
2318 if (sock >= 0) {
2319 ret = close(sock);
2320 if (ret) {
2321 PERROR("close");
2322 }
2323 lttng_fd_put(LTTNG_FD_APPS, 1);
2324 }
2325 unlink(apps_unix_sock_path);
2326
2327 error_poll_add:
2328 lttng_poll_clean(&events);
2329 error_listen:
2330 error_create_poll:
2331 error_testpoint:
2332 DBG("UST Registration thread cleanup complete");
2333 if (err) {
2334 health_error();
2335 ERR("Health error occurred in %s", __func__);
2336 }
2337 health_unregister(health_sessiond);
2338
2339 return NULL;
2340 }
2341
2342 /*
2343 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2344 * exec or it will fails.
2345 */
2346 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2347 {
2348 int ret, clock_ret;
2349 struct timespec timeout;
2350
2351 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2352 consumer_data->consumer_thread_is_ready = 0;
2353
2354 /* Setup pthread condition */
2355 ret = pthread_condattr_init(&consumer_data->condattr);
2356 if (ret) {
2357 errno = ret;
2358 PERROR("pthread_condattr_init consumer data");
2359 goto error;
2360 }
2361
2362 /*
2363 * Set the monotonic clock in order to make sure we DO NOT jump in time
2364 * between the clock_gettime() call and the timedwait call. See bug #324
2365 * for a more details and how we noticed it.
2366 */
2367 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2368 if (ret) {
2369 errno = ret;
2370 PERROR("pthread_condattr_setclock consumer data");
2371 goto error;
2372 }
2373
2374 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2375 if (ret) {
2376 errno = ret;
2377 PERROR("pthread_cond_init consumer data");
2378 goto error;
2379 }
2380
2381 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2382 consumer_data);
2383 if (ret) {
2384 errno = ret;
2385 PERROR("pthread_create consumer");
2386 ret = -1;
2387 goto error;
2388 }
2389
2390 /* We are about to wait on a pthread condition */
2391 pthread_mutex_lock(&consumer_data->cond_mutex);
2392
2393 /* Get time for sem_timedwait absolute timeout */
2394 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2395 /*
2396 * Set the timeout for the condition timed wait even if the clock gettime
2397 * call fails since we might loop on that call and we want to avoid to
2398 * increment the timeout too many times.
2399 */
2400 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2401
2402 /*
2403 * The following loop COULD be skipped in some conditions so this is why we
2404 * set ret to 0 in order to make sure at least one round of the loop is
2405 * done.
2406 */
2407 ret = 0;
2408
2409 /*
2410 * Loop until the condition is reached or when a timeout is reached. Note
2411 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2412 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2413 * possible. This loop does not take any chances and works with both of
2414 * them.
2415 */
2416 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2417 if (clock_ret < 0) {
2418 PERROR("clock_gettime spawn consumer");
2419 /* Infinite wait for the consumerd thread to be ready */
2420 ret = pthread_cond_wait(&consumer_data->cond,
2421 &consumer_data->cond_mutex);
2422 } else {
2423 ret = pthread_cond_timedwait(&consumer_data->cond,
2424 &consumer_data->cond_mutex, &timeout);
2425 }
2426 }
2427
2428 /* Release the pthread condition */
2429 pthread_mutex_unlock(&consumer_data->cond_mutex);
2430
2431 if (ret != 0) {
2432 errno = ret;
2433 if (ret == ETIMEDOUT) {
2434 int pth_ret;
2435
2436 /*
2437 * Call has timed out so we kill the kconsumerd_thread and return
2438 * an error.
2439 */
2440 ERR("Condition timed out. The consumer thread was never ready."
2441 " Killing it");
2442 pth_ret = pthread_cancel(consumer_data->thread);
2443 if (pth_ret < 0) {
2444 PERROR("pthread_cancel consumer thread");
2445 }
2446 } else {
2447 PERROR("pthread_cond_wait failed consumer thread");
2448 }
2449 /* Caller is expecting a negative value on failure. */
2450 ret = -1;
2451 goto error;
2452 }
2453
2454 pthread_mutex_lock(&consumer_data->pid_mutex);
2455 if (consumer_data->pid == 0) {
2456 ERR("Consumerd did not start");
2457 pthread_mutex_unlock(&consumer_data->pid_mutex);
2458 goto error;
2459 }
2460 pthread_mutex_unlock(&consumer_data->pid_mutex);
2461
2462 return 0;
2463
2464 error:
2465 return ret;
2466 }
2467
2468 /*
2469 * Join consumer thread
2470 */
2471 static int join_consumer_thread(struct consumer_data *consumer_data)
2472 {
2473 void *status;
2474
2475 /* Consumer pid must be a real one. */
2476 if (consumer_data->pid > 0) {
2477 int ret;
2478 ret = kill(consumer_data->pid, SIGTERM);
2479 if (ret) {
2480 PERROR("Error killing consumer daemon");
2481 return ret;
2482 }
2483 return pthread_join(consumer_data->thread, &status);
2484 } else {
2485 return 0;
2486 }
2487 }
2488
2489 /*
2490 * Fork and exec a consumer daemon (consumerd).
2491 *
2492 * Return pid if successful else -1.
2493 */
2494 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2495 {
2496 int ret;
2497 pid_t pid;
2498 const char *consumer_to_use;
2499 const char *verbosity;
2500 struct stat st;
2501
2502 DBG("Spawning consumerd");
2503
2504 pid = fork();
2505 if (pid == 0) {
2506 /*
2507 * Exec consumerd.
2508 */
2509 if (opt_verbose_consumer) {
2510 verbosity = "--verbose";
2511 } else if (lttng_opt_quiet) {
2512 verbosity = "--quiet";
2513 } else {
2514 verbosity = "";
2515 }
2516
2517 switch (consumer_data->type) {
2518 case LTTNG_CONSUMER_KERNEL:
2519 /*
2520 * Find out which consumerd to execute. We will first try the
2521 * 64-bit path, then the sessiond's installation directory, and
2522 * fallback on the 32-bit one,
2523 */
2524 DBG3("Looking for a kernel consumer at these locations:");
2525 DBG3(" 1) %s", consumerd64_bin);
2526 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2527 DBG3(" 3) %s", consumerd32_bin);
2528 if (stat(consumerd64_bin, &st) == 0) {
2529 DBG3("Found location #1");
2530 consumer_to_use = consumerd64_bin;
2531 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2532 DBG3("Found location #2");
2533 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2534 } else if (stat(consumerd32_bin, &st) == 0) {
2535 DBG3("Found location #3");
2536 consumer_to_use = consumerd32_bin;
2537 } else {
2538 DBG("Could not find any valid consumerd executable");
2539 ret = -EINVAL;
2540 break;
2541 }
2542 DBG("Using kernel consumer at: %s", consumer_to_use);
2543 ret = execl(consumer_to_use,
2544 "lttng-consumerd", verbosity, "-k",
2545 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2546 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2547 "--group", tracing_group_name,
2548 NULL);
2549 break;
2550 case LTTNG_CONSUMER64_UST:
2551 {
2552 char *tmpnew = NULL;
2553
2554 if (consumerd64_libdir[0] != '\0') {
2555 char *tmp;
2556 size_t tmplen;
2557
2558 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2559 if (!tmp) {
2560 tmp = "";
2561 }
2562 tmplen = strlen("LD_LIBRARY_PATH=")
2563 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2564 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2565 if (!tmpnew) {
2566 ret = -ENOMEM;
2567 goto error;
2568 }
2569 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2570 strcat(tmpnew, consumerd64_libdir);
2571 if (tmp[0] != '\0') {
2572 strcat(tmpnew, ":");
2573 strcat(tmpnew, tmp);
2574 }
2575 ret = putenv(tmpnew);
2576 if (ret) {
2577 ret = -errno;
2578 free(tmpnew);
2579 goto error;
2580 }
2581 }
2582 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2583 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2584 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2585 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2586 "--group", tracing_group_name,
2587 NULL);
2588 if (consumerd64_libdir[0] != '\0') {
2589 free(tmpnew);
2590 }
2591 break;
2592 }
2593 case LTTNG_CONSUMER32_UST:
2594 {
2595 char *tmpnew = NULL;
2596
2597 if (consumerd32_libdir[0] != '\0') {
2598 char *tmp;
2599 size_t tmplen;
2600
2601 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2602 if (!tmp) {
2603 tmp = "";
2604 }
2605 tmplen = strlen("LD_LIBRARY_PATH=")
2606 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2607 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2608 if (!tmpnew) {
2609 ret = -ENOMEM;
2610 goto error;
2611 }
2612 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2613 strcat(tmpnew, consumerd32_libdir);
2614 if (tmp[0] != '\0') {
2615 strcat(tmpnew, ":");
2616 strcat(tmpnew, tmp);
2617 }
2618 ret = putenv(tmpnew);
2619 if (ret) {
2620 ret = -errno;
2621 free(tmpnew);
2622 goto error;
2623 }
2624 }
2625 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2626 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2627 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2628 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2629 "--group", tracing_group_name,
2630 NULL);
2631 if (consumerd32_libdir[0] != '\0') {
2632 free(tmpnew);
2633 }
2634 break;
2635 }
2636 default:
2637 PERROR("unknown consumer type");
2638 exit(EXIT_FAILURE);
2639 }
2640 if (errno != 0) {
2641 PERROR("Consumer execl()");
2642 }
2643 /* Reaching this point, we got a failure on our execl(). */
2644 exit(EXIT_FAILURE);
2645 } else if (pid > 0) {
2646 ret = pid;
2647 } else {
2648 PERROR("start consumer fork");
2649 ret = -errno;
2650 }
2651 error:
2652 return ret;
2653 }
2654
2655 /*
2656 * Spawn the consumerd daemon and session daemon thread.
2657 */
2658 static int start_consumerd(struct consumer_data *consumer_data)
2659 {
2660 int ret;
2661
2662 /*
2663 * Set the listen() state on the socket since there is a possible race
2664 * between the exec() of the consumer daemon and this call if place in the
2665 * consumer thread. See bug #366 for more details.
2666 */
2667 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2668 if (ret < 0) {
2669 goto error;
2670 }
2671
2672 pthread_mutex_lock(&consumer_data->pid_mutex);
2673 if (consumer_data->pid != 0) {
2674 pthread_mutex_unlock(&consumer_data->pid_mutex);
2675 goto end;
2676 }
2677
2678 ret = spawn_consumerd(consumer_data);
2679 if (ret < 0) {
2680 ERR("Spawning consumerd failed");
2681 pthread_mutex_unlock(&consumer_data->pid_mutex);
2682 goto error;
2683 }
2684
2685 /* Setting up the consumer_data pid */
2686 consumer_data->pid = ret;
2687 DBG2("Consumer pid %d", consumer_data->pid);
2688 pthread_mutex_unlock(&consumer_data->pid_mutex);
2689
2690 DBG2("Spawning consumer control thread");
2691 ret = spawn_consumer_thread(consumer_data);
2692 if (ret < 0) {
2693 ERR("Fatal error spawning consumer control thread");
2694 goto error;
2695 }
2696
2697 end:
2698 return 0;
2699
2700 error:
2701 /* Cleanup already created sockets on error. */
2702 if (consumer_data->err_sock >= 0) {
2703 int err;
2704
2705 err = close(consumer_data->err_sock);
2706 if (err < 0) {
2707 PERROR("close consumer data error socket");
2708 }
2709 }
2710 return ret;
2711 }
2712
2713 /*
2714 * Setup necessary data for kernel tracer action.
2715 */
2716 static int init_kernel_tracer(void)
2717 {
2718 int ret;
2719
2720 /* Modprobe lttng kernel modules */
2721 ret = modprobe_lttng_control();
2722 if (ret < 0) {
2723 goto error;
2724 }
2725
2726 /* Open debugfs lttng */
2727 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2728 if (kernel_tracer_fd < 0) {
2729 DBG("Failed to open %s", module_proc_lttng);
2730 ret = -1;
2731 goto error_open;
2732 }
2733
2734 /* Validate kernel version */
2735 ret = kernel_validate_version(kernel_tracer_fd);
2736 if (ret < 0) {
2737 goto error_version;
2738 }
2739
2740 ret = modprobe_lttng_data();
2741 if (ret < 0) {
2742 goto error_modules;
2743 }
2744
2745 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2746 return 0;
2747
2748 error_version:
2749 modprobe_remove_lttng_control();
2750 ret = close(kernel_tracer_fd);
2751 if (ret) {
2752 PERROR("close");
2753 }
2754 kernel_tracer_fd = -1;
2755 return LTTNG_ERR_KERN_VERSION;
2756
2757 error_modules:
2758 ret = close(kernel_tracer_fd);
2759 if (ret) {
2760 PERROR("close");
2761 }
2762
2763 error_open:
2764 modprobe_remove_lttng_control();
2765
2766 error:
2767 WARN("No kernel tracer available");
2768 kernel_tracer_fd = -1;
2769 if (!is_root) {
2770 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2771 } else {
2772 return LTTNG_ERR_KERN_NA;
2773 }
2774 }
2775
2776
2777 /*
2778 * Copy consumer output from the tracing session to the domain session. The
2779 * function also applies the right modification on a per domain basis for the
2780 * trace files destination directory.
2781 *
2782 * Should *NOT* be called with RCU read-side lock held.
2783 */
2784 static int copy_session_consumer(int domain, struct ltt_session *session)
2785 {
2786 int ret;
2787 const char *dir_name;
2788 struct consumer_output *consumer;
2789
2790 assert(session);
2791 assert(session->consumer);
2792
2793 switch (domain) {
2794 case LTTNG_DOMAIN_KERNEL:
2795 DBG3("Copying tracing session consumer output in kernel session");
2796 /*
2797 * XXX: We should audit the session creation and what this function
2798 * does "extra" in order to avoid a destroy since this function is used
2799 * in the domain session creation (kernel and ust) only. Same for UST
2800 * domain.
2801 */
2802 if (session->kernel_session->consumer) {
2803 consumer_output_put(session->kernel_session->consumer);
2804 }
2805 session->kernel_session->consumer =
2806 consumer_copy_output(session->consumer);
2807 /* Ease our life a bit for the next part */
2808 consumer = session->kernel_session->consumer;
2809 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2810 break;
2811 case LTTNG_DOMAIN_JUL:
2812 case LTTNG_DOMAIN_LOG4J:
2813 case LTTNG_DOMAIN_PYTHON:
2814 case LTTNG_DOMAIN_UST:
2815 DBG3("Copying tracing session consumer output in UST session");
2816 if (session->ust_session->consumer) {
2817 consumer_output_put(session->ust_session->consumer);
2818 }
2819 session->ust_session->consumer =
2820 consumer_copy_output(session->consumer);
2821 /* Ease our life a bit for the next part */
2822 consumer = session->ust_session->consumer;
2823 dir_name = DEFAULT_UST_TRACE_DIR;
2824 break;
2825 default:
2826 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2827 goto error;
2828 }
2829
2830 /* Append correct directory to subdir */
2831 strncat(consumer->subdir, dir_name,
2832 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2833 DBG3("Copy session consumer subdir %s", consumer->subdir);
2834
2835 ret = LTTNG_OK;
2836
2837 error:
2838 return ret;
2839 }
2840
2841 /*
2842 * Create an UST session and add it to the session ust list.
2843 *
2844 * Should *NOT* be called with RCU read-side lock held.
2845 */
2846 static int create_ust_session(struct ltt_session *session,
2847 struct lttng_domain *domain)
2848 {
2849 int ret;
2850 struct ltt_ust_session *lus = NULL;
2851
2852 assert(session);
2853 assert(domain);
2854 assert(session->consumer);
2855
2856 switch (domain->type) {
2857 case LTTNG_DOMAIN_JUL:
2858 case LTTNG_DOMAIN_LOG4J:
2859 case LTTNG_DOMAIN_PYTHON:
2860 case LTTNG_DOMAIN_UST:
2861 break;
2862 default:
2863 ERR("Unknown UST domain on create session %d", domain->type);
2864 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2865 goto error;
2866 }
2867
2868 DBG("Creating UST session");
2869
2870 lus = trace_ust_create_session(session->id);
2871 if (lus == NULL) {
2872 ret = LTTNG_ERR_UST_SESS_FAIL;
2873 goto error;
2874 }
2875
2876 lus->uid = session->uid;
2877 lus->gid = session->gid;
2878 lus->output_traces = session->output_traces;
2879 lus->snapshot_mode = session->snapshot_mode;
2880 lus->live_timer_interval = session->live_timer;
2881 session->ust_session = lus;
2882 if (session->shm_path[0]) {
2883 strncpy(lus->root_shm_path, session->shm_path,
2884 sizeof(lus->root_shm_path));
2885 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2886 strncpy(lus->shm_path, session->shm_path,
2887 sizeof(lus->shm_path));
2888 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2889 strncat(lus->shm_path, "/ust",
2890 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2891 }
2892 /* Copy session output to the newly created UST session */
2893 ret = copy_session_consumer(domain->type, session);
2894 if (ret != LTTNG_OK) {
2895 goto error;
2896 }
2897
2898 return LTTNG_OK;
2899
2900 error:
2901 free(lus);
2902 session->ust_session = NULL;
2903 return ret;
2904 }
2905
2906 /*
2907 * Create a kernel tracer session then create the default channel.
2908 */
2909 static int create_kernel_session(struct ltt_session *session)
2910 {
2911 int ret;
2912
2913 DBG("Creating kernel session");
2914
2915 ret = kernel_create_session(session, kernel_tracer_fd);
2916 if (ret < 0) {
2917 ret = LTTNG_ERR_KERN_SESS_FAIL;
2918 goto error;
2919 }
2920
2921 /* Code flow safety */
2922 assert(session->kernel_session);
2923
2924 /* Copy session output to the newly created Kernel session */
2925 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2926 if (ret != LTTNG_OK) {
2927 goto error;
2928 }
2929
2930 /* Create directory(ies) on local filesystem. */
2931 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2932 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2933 ret = run_as_mkdir_recursive(
2934 session->kernel_session->consumer->dst.trace_path,
2935 S_IRWXU | S_IRWXG, session->uid, session->gid);
2936 if (ret < 0) {
2937 if (errno != EEXIST) {
2938 ERR("Trace directory creation error");
2939 goto error;
2940 }
2941 }
2942 }
2943
2944 session->kernel_session->uid = session->uid;
2945 session->kernel_session->gid = session->gid;
2946 session->kernel_session->output_traces = session->output_traces;
2947 session->kernel_session->snapshot_mode = session->snapshot_mode;
2948
2949 return LTTNG_OK;
2950
2951 error:
2952 trace_kernel_destroy_session(session->kernel_session);
2953 session->kernel_session = NULL;
2954 return ret;
2955 }
2956
2957 /*
2958 * Count number of session permitted by uid/gid.
2959 */
2960 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2961 {
2962 unsigned int i = 0;
2963 struct ltt_session *session;
2964
2965 DBG("Counting number of available session for UID %d GID %d",
2966 uid, gid);
2967 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2968 /*
2969 * Only list the sessions the user can control.
2970 */
2971 if (!session_access_ok(session, uid, gid)) {
2972 continue;
2973 }
2974 i++;
2975 }
2976 return i;
2977 }
2978
2979 /*
2980 * Process the command requested by the lttng client within the command
2981 * context structure. This function make sure that the return structure (llm)
2982 * is set and ready for transmission before returning.
2983 *
2984 * Return any error encountered or 0 for success.
2985 *
2986 * "sock" is only used for special-case var. len data.
2987 *
2988 * Should *NOT* be called with RCU read-side lock held.
2989 */
2990 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2991 int *sock_error)
2992 {
2993 int ret = LTTNG_OK;
2994 int need_tracing_session = 1;
2995 int need_domain;
2996
2997 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2998
2999 assert(!rcu_read_ongoing());
3000
3001 *sock_error = 0;
3002
3003 switch (cmd_ctx->lsm->cmd_type) {
3004 case LTTNG_CREATE_SESSION:
3005 case LTTNG_CREATE_SESSION_SNAPSHOT:
3006 case LTTNG_CREATE_SESSION_LIVE:
3007 case LTTNG_DESTROY_SESSION:
3008 case LTTNG_LIST_SESSIONS:
3009 case LTTNG_LIST_DOMAINS:
3010 case LTTNG_START_TRACE:
3011 case LTTNG_STOP_TRACE:
3012 case LTTNG_DATA_PENDING:
3013 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3014 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3015 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3016 case LTTNG_SNAPSHOT_RECORD:
3017 case LTTNG_SAVE_SESSION:
3018 case LTTNG_SET_SESSION_SHM_PATH:
3019 need_domain = 0;
3020 break;
3021 default:
3022 need_domain = 1;
3023 }
3024
3025 if (opt_no_kernel && need_domain
3026 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3027 if (!is_root) {
3028 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3029 } else {
3030 ret = LTTNG_ERR_KERN_NA;
3031 }
3032 goto error;
3033 }
3034
3035 /* Deny register consumer if we already have a spawned consumer. */
3036 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3037 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3038 if (kconsumer_data.pid > 0) {
3039 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3040 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3041 goto error;
3042 }
3043 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3044 }
3045
3046 /*
3047 * Check for command that don't needs to allocate a returned payload. We do
3048 * this here so we don't have to make the call for no payload at each
3049 * command.
3050 */
3051 switch(cmd_ctx->lsm->cmd_type) {
3052 case LTTNG_LIST_SESSIONS:
3053 case LTTNG_LIST_TRACEPOINTS:
3054 case LTTNG_LIST_TRACEPOINT_FIELDS:
3055 case LTTNG_LIST_DOMAINS:
3056 case LTTNG_LIST_CHANNELS:
3057 case LTTNG_LIST_EVENTS:
3058 case LTTNG_LIST_SYSCALLS:
3059 case LTTNG_LIST_TRACKER_PIDS:
3060 break;
3061 default:
3062 /* Setup lttng message with no payload */
3063 ret = setup_lttng_msg(cmd_ctx, 0);
3064 if (ret < 0) {
3065 /* This label does not try to unlock the session */
3066 goto init_setup_error;
3067 }
3068 }
3069
3070 /* Commands that DO NOT need a session. */
3071 switch (cmd_ctx->lsm->cmd_type) {
3072 case LTTNG_CREATE_SESSION:
3073 case LTTNG_CREATE_SESSION_SNAPSHOT:
3074 case LTTNG_CREATE_SESSION_LIVE:
3075 case LTTNG_CALIBRATE:
3076 case LTTNG_LIST_SESSIONS:
3077 case LTTNG_LIST_TRACEPOINTS:
3078 case LTTNG_LIST_SYSCALLS:
3079 case LTTNG_LIST_TRACEPOINT_FIELDS:
3080 case LTTNG_SAVE_SESSION:
3081 need_tracing_session = 0;
3082 break;
3083 default:
3084 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3085 /*
3086 * We keep the session list lock across _all_ commands
3087 * for now, because the per-session lock does not
3088 * handle teardown properly.
3089 */
3090 session_lock_list();
3091 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3092 if (cmd_ctx->session == NULL) {
3093 ret = LTTNG_ERR_SESS_NOT_FOUND;
3094 goto error;
3095 } else {
3096 /* Acquire lock for the session */
3097 session_lock(cmd_ctx->session);
3098 }
3099 break;
3100 }
3101
3102 /*
3103 * Commands that need a valid session but should NOT create one if none
3104 * exists. Instead of creating one and destroying it when the command is
3105 * handled, process that right before so we save some round trip in useless
3106 * code path.
3107 */
3108 switch (cmd_ctx->lsm->cmd_type) {
3109 case LTTNG_DISABLE_CHANNEL:
3110 case LTTNG_DISABLE_EVENT:
3111 switch (cmd_ctx->lsm->domain.type) {
3112 case LTTNG_DOMAIN_KERNEL:
3113 if (!cmd_ctx->session->kernel_session) {
3114 ret = LTTNG_ERR_NO_CHANNEL;
3115 goto error;
3116 }
3117 break;
3118 case LTTNG_DOMAIN_JUL:
3119 case LTTNG_DOMAIN_LOG4J:
3120 case LTTNG_DOMAIN_PYTHON:
3121 case LTTNG_DOMAIN_UST:
3122 if (!cmd_ctx->session->ust_session) {
3123 ret = LTTNG_ERR_NO_CHANNEL;
3124 goto error;
3125 }
3126 break;
3127 default:
3128 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3129 goto error;
3130 }
3131 default:
3132 break;
3133 }
3134
3135 if (!need_domain) {
3136 goto skip_domain;
3137 }
3138
3139 /*
3140 * Check domain type for specific "pre-action".
3141 */
3142 switch (cmd_ctx->lsm->domain.type) {
3143 case LTTNG_DOMAIN_KERNEL:
3144 if (!is_root) {
3145 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3146 goto error;
3147 }
3148
3149 /* Kernel tracer check */
3150 if (kernel_tracer_fd == -1) {
3151 /* Basically, load kernel tracer modules */
3152 ret = init_kernel_tracer();
3153 if (ret != 0) {
3154 goto error;
3155 }
3156 }
3157
3158 /* Consumer is in an ERROR state. Report back to client */
3159 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3160 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3161 goto error;
3162 }
3163
3164 /* Need a session for kernel command */
3165 if (need_tracing_session) {
3166 if (cmd_ctx->session->kernel_session == NULL) {
3167 ret = create_kernel_session(cmd_ctx->session);
3168 if (ret < 0) {
3169 ret = LTTNG_ERR_KERN_SESS_FAIL;
3170 goto error;
3171 }
3172 }
3173
3174 /* Start the kernel consumer daemon */
3175 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3176 if (kconsumer_data.pid == 0 &&
3177 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3178 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3179 ret = start_consumerd(&kconsumer_data);
3180 if (ret < 0) {
3181 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3182 goto error;
3183 }
3184 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3185 } else {
3186 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3187 }
3188
3189 /*
3190 * The consumer was just spawned so we need to add the socket to
3191 * the consumer output of the session if exist.
3192 */
3193 ret = consumer_create_socket(&kconsumer_data,
3194 cmd_ctx->session->kernel_session->consumer);
3195 if (ret < 0) {
3196 goto error;
3197 }
3198 }
3199
3200 break;
3201 case LTTNG_DOMAIN_JUL:
3202 case LTTNG_DOMAIN_LOG4J:
3203 case LTTNG_DOMAIN_PYTHON:
3204 case LTTNG_DOMAIN_UST:
3205 {
3206 if (!ust_app_supported()) {
3207 ret = LTTNG_ERR_NO_UST;
3208 goto error;
3209 }
3210 /* Consumer is in an ERROR state. Report back to client */
3211 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3212 ret = LTTNG_ERR_NO_USTCONSUMERD;
3213 goto error;
3214 }
3215
3216 if (need_tracing_session) {
3217 /* Create UST session if none exist. */
3218 if (cmd_ctx->session->ust_session == NULL) {
3219 ret = create_ust_session(cmd_ctx->session,
3220 &cmd_ctx->lsm->domain);
3221 if (ret != LTTNG_OK) {
3222 goto error;
3223 }
3224 }
3225
3226 /* Start the UST consumer daemons */
3227 /* 64-bit */
3228 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3229 if (consumerd64_bin[0] != '\0' &&
3230 ustconsumer64_data.pid == 0 &&
3231 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3232 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3233 ret = start_consumerd(&ustconsumer64_data);
3234 if (ret < 0) {