Fix: remove the rundir at the end of the cleanup
[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 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <inttypes.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <config.h>
39
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
50 #include "channel.h"
51 #include "cmd.h"
52 #include "consumer.h"
53 #include "context.h"
54 #include "event.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "modprobe.h"
58 #include "shm.h"
59 #include "ust-ctl.h"
60 #include "ust-consumer.h"
61 #include "utils.h"
62 #include "fd-limit.h"
63 #include "health-sessiond.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66 #include "jul-thread.h"
67
68 #define CONSUMERD_FILE "lttng-consumerd"
69
70 const char *progname;
71 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
72 static const char *opt_pidfile;
73 static int opt_sig_parent;
74 static int opt_verbose_consumer;
75 static int opt_daemon;
76 static int opt_no_kernel;
77 static int is_root; /* Set to 1 if the daemon is running as root */
78 static pid_t ppid; /* Parent PID for --sig-parent option */
79 static char *rundir;
80
81 /*
82 * Consumer daemon specific control data. Every value not initialized here is
83 * set to 0 by the static definition.
84 */
85 static struct consumer_data kconsumer_data = {
86 .type = LTTNG_CONSUMER_KERNEL,
87 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
88 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
89 .err_sock = -1,
90 .cmd_sock = -1,
91 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
92 .lock = PTHREAD_MUTEX_INITIALIZER,
93 .cond = PTHREAD_COND_INITIALIZER,
94 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
95 };
96 static struct consumer_data ustconsumer64_data = {
97 .type = LTTNG_CONSUMER64_UST,
98 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
99 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
100 .err_sock = -1,
101 .cmd_sock = -1,
102 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
103 .lock = PTHREAD_MUTEX_INITIALIZER,
104 .cond = PTHREAD_COND_INITIALIZER,
105 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
106 };
107 static struct consumer_data ustconsumer32_data = {
108 .type = LTTNG_CONSUMER32_UST,
109 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
110 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
111 .err_sock = -1,
112 .cmd_sock = -1,
113 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
114 .lock = PTHREAD_MUTEX_INITIALIZER,
115 .cond = PTHREAD_COND_INITIALIZER,
116 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
117 };
118
119 /* Shared between threads */
120 static int dispatch_thread_exit;
121
122 /* Global application Unix socket path */
123 static char apps_unix_sock_path[PATH_MAX];
124 /* Global client Unix socket path */
125 static char client_unix_sock_path[PATH_MAX];
126 /* global wait shm path for UST */
127 static char wait_shm_path[PATH_MAX];
128 /* Global health check unix path */
129 static char health_unix_sock_path[PATH_MAX];
130
131 /* Sockets and FDs */
132 static int client_sock = -1;
133 static int apps_sock = -1;
134 int kernel_tracer_fd = -1;
135 static int kernel_poll_pipe[2] = { -1, -1 };
136
137 /*
138 * Quit pipe for all threads. This permits a single cancellation point
139 * for all threads when receiving an event on the pipe.
140 */
141 static int thread_quit_pipe[2] = { -1, -1 };
142
143 /*
144 * This pipe is used to inform the thread managing application communication
145 * that a command is queued and ready to be processed.
146 */
147 static int apps_cmd_pipe[2] = { -1, -1 };
148
149 int apps_cmd_notify_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_thread;
153 static pthread_t apps_notify_thread;
154 static pthread_t reg_apps_thread;
155 static pthread_t client_thread;
156 static pthread_t kernel_thread;
157 static pthread_t dispatch_thread;
158 static pthread_t health_thread;
159 static pthread_t ht_cleanup_thread;
160 static pthread_t jul_reg_thread;
161
162 /*
163 * UST registration command queue. This queue is tied with a futex and uses a N
164 * wakers / 1 waiter implemented and detailed in futex.c/.h
165 *
166 * The thread_manage_apps and thread_dispatch_ust_registration interact with
167 * this queue and the wait/wake scheme.
168 */
169 static struct ust_cmd_queue ust_cmd_queue;
170
171 /*
172 * Pointer initialized before thread creation.
173 *
174 * This points to the tracing session list containing the session count and a
175 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
176 * MUST NOT be taken if you call a public function in session.c.
177 *
178 * The lock is nested inside the structure: session_list_ptr->lock. Please use
179 * session_lock_list and session_unlock_list for lock acquisition.
180 */
181 static struct ltt_session_list *session_list_ptr;
182
183 int ust_consumerd64_fd = -1;
184 int ust_consumerd32_fd = -1;
185
186 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
187 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
188 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
189 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
190
191 static const char *module_proc_lttng = "/proc/lttng";
192
193 /*
194 * Consumer daemon state which is changed when spawning it, killing it or in
195 * case of a fatal error.
196 */
197 enum consumerd_state {
198 CONSUMER_STARTED = 1,
199 CONSUMER_STOPPED = 2,
200 CONSUMER_ERROR = 3,
201 };
202
203 /*
204 * This consumer daemon state is used to validate if a client command will be
205 * able to reach the consumer. If not, the client is informed. For instance,
206 * doing a "lttng start" when the consumer state is set to ERROR will return an
207 * error to the client.
208 *
209 * The following example shows a possible race condition of this scheme:
210 *
211 * consumer thread error happens
212 * client cmd arrives
213 * client cmd checks state -> still OK
214 * consumer thread exit, sets error
215 * client cmd try to talk to consumer
216 * ...
217 *
218 * However, since the consumer is a different daemon, we have no way of making
219 * sure the command will reach it safely even with this state flag. This is why
220 * we consider that up to the state validation during command processing, the
221 * command is safe. After that, we can not guarantee the correctness of the
222 * client request vis-a-vis the consumer.
223 */
224 static enum consumerd_state ust_consumerd_state;
225 static enum consumerd_state kernel_consumerd_state;
226
227 /*
228 * Socket timeout for receiving and sending in seconds.
229 */
230 static int app_socket_timeout;
231
232 /* Set in main() with the current page size. */
233 long page_size;
234
235 /* Application health monitoring */
236 struct health_app *health_sessiond;
237
238 /* JUL TCP port for registration. Used by the JUL thread. */
239 unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
240
241 static
242 void setup_consumerd_path(void)
243 {
244 const char *bin, *libdir;
245
246 /*
247 * Allow INSTALL_BIN_PATH to be used as a target path for the
248 * native architecture size consumer if CONFIG_CONSUMER*_PATH
249 * has not been defined.
250 */
251 #if (CAA_BITS_PER_LONG == 32)
252 if (!consumerd32_bin[0]) {
253 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
254 }
255 if (!consumerd32_libdir[0]) {
256 consumerd32_libdir = INSTALL_LIB_PATH;
257 }
258 #elif (CAA_BITS_PER_LONG == 64)
259 if (!consumerd64_bin[0]) {
260 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
261 }
262 if (!consumerd64_libdir[0]) {
263 consumerd64_libdir = INSTALL_LIB_PATH;
264 }
265 #else
266 #error "Unknown bitness"
267 #endif
268
269 /*
270 * runtime env. var. overrides the build default.
271 */
272 bin = getenv("LTTNG_CONSUMERD32_BIN");
273 if (bin) {
274 consumerd32_bin = bin;
275 }
276 bin = getenv("LTTNG_CONSUMERD64_BIN");
277 if (bin) {
278 consumerd64_bin = bin;
279 }
280 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
281 if (libdir) {
282 consumerd32_libdir = libdir;
283 }
284 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
285 if (libdir) {
286 consumerd64_libdir = libdir;
287 }
288 }
289
290 /*
291 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
292 */
293 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
294 {
295 int ret;
296
297 assert(events);
298
299 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
300 if (ret < 0) {
301 goto error;
302 }
303
304 /* Add quit pipe */
305 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
306 if (ret < 0) {
307 goto error;
308 }
309
310 return 0;
311
312 error:
313 return ret;
314 }
315
316 /*
317 * Check if the thread quit pipe was triggered.
318 *
319 * Return 1 if it was triggered else 0;
320 */
321 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
322 {
323 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
324 return 1;
325 }
326
327 return 0;
328 }
329
330 /*
331 * Init thread quit pipe.
332 *
333 * Return -1 on error or 0 if all pipes are created.
334 */
335 static int init_thread_quit_pipe(void)
336 {
337 int ret, i;
338
339 ret = pipe(thread_quit_pipe);
340 if (ret < 0) {
341 PERROR("thread quit pipe");
342 goto error;
343 }
344
345 for (i = 0; i < 2; i++) {
346 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
347 if (ret < 0) {
348 PERROR("fcntl");
349 goto error;
350 }
351 }
352
353 error:
354 return ret;
355 }
356
357 /*
358 * Stop all threads by closing the thread quit pipe.
359 */
360 static void stop_threads(void)
361 {
362 int ret;
363
364 /* Stopping all threads */
365 DBG("Terminating all threads");
366 ret = notify_thread_pipe(thread_quit_pipe[1]);
367 if (ret < 0) {
368 ERR("write error on thread quit pipe");
369 }
370
371 /* Dispatch thread */
372 CMM_STORE_SHARED(dispatch_thread_exit, 1);
373 futex_nto1_wake(&ust_cmd_queue.futex);
374 }
375
376 /*
377 * Close every consumer sockets.
378 */
379 static void close_consumer_sockets(void)
380 {
381 int ret;
382
383 if (kconsumer_data.err_sock >= 0) {
384 ret = close(kconsumer_data.err_sock);
385 if (ret < 0) {
386 PERROR("kernel consumer err_sock close");
387 }
388 }
389 if (ustconsumer32_data.err_sock >= 0) {
390 ret = close(ustconsumer32_data.err_sock);
391 if (ret < 0) {
392 PERROR("UST consumerd32 err_sock close");
393 }
394 }
395 if (ustconsumer64_data.err_sock >= 0) {
396 ret = close(ustconsumer64_data.err_sock);
397 if (ret < 0) {
398 PERROR("UST consumerd64 err_sock close");
399 }
400 }
401 if (kconsumer_data.cmd_sock >= 0) {
402 ret = close(kconsumer_data.cmd_sock);
403 if (ret < 0) {
404 PERROR("kernel consumer cmd_sock close");
405 }
406 }
407 if (ustconsumer32_data.cmd_sock >= 0) {
408 ret = close(ustconsumer32_data.cmd_sock);
409 if (ret < 0) {
410 PERROR("UST consumerd32 cmd_sock close");
411 }
412 }
413 if (ustconsumer64_data.cmd_sock >= 0) {
414 ret = close(ustconsumer64_data.cmd_sock);
415 if (ret < 0) {
416 PERROR("UST consumerd64 cmd_sock close");
417 }
418 }
419 }
420
421 /*
422 * Cleanup the daemon
423 */
424 static void cleanup(void)
425 {
426 int ret;
427 struct ltt_session *sess, *stmp;
428 char path[PATH_MAX];
429
430 DBG("Cleaning up");
431
432 /*
433 * Close the thread quit pipe. It has already done its job,
434 * since we are now called.
435 */
436 utils_close_pipe(thread_quit_pipe);
437
438 /*
439 * If opt_pidfile is undefined, the default file will be wiped when
440 * removing the rundir.
441 */
442 if (opt_pidfile) {
443 ret = remove(opt_pidfile);
444 if (ret < 0) {
445 PERROR("remove pidfile %s", opt_pidfile);
446 }
447 }
448
449 DBG("Removing sessiond and consumerd content of directory %s", rundir);
450
451 /* sessiond */
452 snprintf(path, PATH_MAX,
453 "%s/%s",
454 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
455 DBG("Removing %s", path);
456 (void) unlink(path);
457
458 snprintf(path, PATH_MAX, "%s/%s", rundir,
459 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
460 DBG("Removing %s", path);
461 (void) unlink(path);
462
463 /* kconsumerd */
464 snprintf(path, PATH_MAX,
465 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
466 rundir);
467 DBG("Removing %s", path);
468 (void) unlink(path);
469
470 snprintf(path, PATH_MAX,
471 DEFAULT_KCONSUMERD_PATH,
472 rundir);
473 DBG("Removing directory %s", path);
474 (void) rmdir(path);
475
476 /* ust consumerd 32 */
477 snprintf(path, PATH_MAX,
478 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
479 rundir);
480 DBG("Removing %s", path);
481 (void) unlink(path);
482
483 snprintf(path, PATH_MAX,
484 DEFAULT_USTCONSUMERD32_PATH,
485 rundir);
486 DBG("Removing directory %s", path);
487 (void) rmdir(path);
488
489 /* ust consumerd 64 */
490 snprintf(path, PATH_MAX,
491 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
492 rundir);
493 DBG("Removing %s", path);
494 (void) unlink(path);
495
496 snprintf(path, PATH_MAX,
497 DEFAULT_USTCONSUMERD64_PATH,
498 rundir);
499 DBG("Removing directory %s", path);
500 (void) rmdir(path);
501
502 (void) rmdir(rundir);
503 free(rundir);
504
505 DBG("Cleaning up all sessions");
506
507 /* Destroy session list mutex */
508 if (session_list_ptr != NULL) {
509 pthread_mutex_destroy(&session_list_ptr->lock);
510
511 /* Cleanup ALL session */
512 cds_list_for_each_entry_safe(sess, stmp,
513 &session_list_ptr->head, list) {
514 cmd_destroy_session(sess, kernel_poll_pipe[1]);
515 }
516 }
517
518 DBG("Closing all UST sockets");
519 ust_app_clean_list();
520 buffer_reg_destroy_registries();
521
522 if (is_root && !opt_no_kernel) {
523 DBG2("Closing kernel fd");
524 if (kernel_tracer_fd >= 0) {
525 ret = close(kernel_tracer_fd);
526 if (ret) {
527 PERROR("close");
528 }
529 }
530 DBG("Unloading kernel modules");
531 modprobe_remove_lttng_all();
532 }
533
534 close_consumer_sockets();
535
536 /* <fun> */
537 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
538 "Matthew, BEET driven development works!%c[%dm",
539 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
540 /* </fun> */
541 }
542
543 /*
544 * Send data on a unix socket using the liblttsessiondcomm API.
545 *
546 * Return lttcomm error code.
547 */
548 static int send_unix_sock(int sock, void *buf, size_t len)
549 {
550 /* Check valid length */
551 if (len == 0) {
552 return -1;
553 }
554
555 return lttcomm_send_unix_sock(sock, buf, len);
556 }
557
558 /*
559 * Free memory of a command context structure.
560 */
561 static void clean_command_ctx(struct command_ctx **cmd_ctx)
562 {
563 DBG("Clean command context structure");
564 if (*cmd_ctx) {
565 if ((*cmd_ctx)->llm) {
566 free((*cmd_ctx)->llm);
567 }
568 if ((*cmd_ctx)->lsm) {
569 free((*cmd_ctx)->lsm);
570 }
571 free(*cmd_ctx);
572 *cmd_ctx = NULL;
573 }
574 }
575
576 /*
577 * Notify UST applications using the shm mmap futex.
578 */
579 static int notify_ust_apps(int active)
580 {
581 char *wait_shm_mmap;
582
583 DBG("Notifying applications of session daemon state: %d", active);
584
585 /* See shm.c for this call implying mmap, shm and futex calls */
586 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
587 if (wait_shm_mmap == NULL) {
588 goto error;
589 }
590
591 /* Wake waiting process */
592 futex_wait_update((int32_t *) wait_shm_mmap, active);
593
594 /* Apps notified successfully */
595 return 0;
596
597 error:
598 return -1;
599 }
600
601 /*
602 * Setup the outgoing data buffer for the response (llm) by allocating the
603 * right amount of memory and copying the original information from the lsm
604 * structure.
605 *
606 * Return total size of the buffer pointed by buf.
607 */
608 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
609 {
610 int ret, buf_size;
611
612 buf_size = size;
613
614 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
615 if (cmd_ctx->llm == NULL) {
616 PERROR("zmalloc");
617 ret = -ENOMEM;
618 goto error;
619 }
620
621 /* Copy common data */
622 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
623 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
624
625 cmd_ctx->llm->data_size = size;
626 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
627
628 return buf_size;
629
630 error:
631 return ret;
632 }
633
634 /*
635 * Update the kernel poll set of all channel fd available over all tracing
636 * session. Add the wakeup pipe at the end of the set.
637 */
638 static int update_kernel_poll(struct lttng_poll_event *events)
639 {
640 int ret;
641 struct ltt_session *session;
642 struct ltt_kernel_channel *channel;
643
644 DBG("Updating kernel poll set");
645
646 session_lock_list();
647 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
648 session_lock(session);
649 if (session->kernel_session == NULL) {
650 session_unlock(session);
651 continue;
652 }
653
654 cds_list_for_each_entry(channel,
655 &session->kernel_session->channel_list.head, list) {
656 /* Add channel fd to the kernel poll set */
657 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
658 if (ret < 0) {
659 session_unlock(session);
660 goto error;
661 }
662 DBG("Channel fd %d added to kernel set", channel->fd);
663 }
664 session_unlock(session);
665 }
666 session_unlock_list();
667
668 return 0;
669
670 error:
671 session_unlock_list();
672 return -1;
673 }
674
675 /*
676 * Find the channel fd from 'fd' over all tracing session. When found, check
677 * for new channel stream and send those stream fds to the kernel consumer.
678 *
679 * Useful for CPU hotplug feature.
680 */
681 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
682 {
683 int ret = 0;
684 struct ltt_session *session;
685 struct ltt_kernel_session *ksess;
686 struct ltt_kernel_channel *channel;
687
688 DBG("Updating kernel streams for channel fd %d", fd);
689
690 session_lock_list();
691 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
692 session_lock(session);
693 if (session->kernel_session == NULL) {
694 session_unlock(session);
695 continue;
696 }
697 ksess = session->kernel_session;
698
699 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
700 if (channel->fd == fd) {
701 DBG("Channel found, updating kernel streams");
702 ret = kernel_open_channel_stream(channel);
703 if (ret < 0) {
704 goto error;
705 }
706 /* Update the stream global counter */
707 ksess->stream_count_global += ret;
708
709 /*
710 * Have we already sent fds to the consumer? If yes, it means
711 * that tracing is started so it is safe to send our updated
712 * stream fds.
713 */
714 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
715 struct lttng_ht_iter iter;
716 struct consumer_socket *socket;
717
718 rcu_read_lock();
719 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
720 &iter.iter, socket, node.node) {
721 pthread_mutex_lock(socket->lock);
722 ret = kernel_consumer_send_channel_stream(socket,
723 channel, ksess,
724 session->output_traces ? 1 : 0);
725 pthread_mutex_unlock(socket->lock);
726 if (ret < 0) {
727 rcu_read_unlock();
728 goto error;
729 }
730 }
731 rcu_read_unlock();
732 }
733 goto error;
734 }
735 }
736 session_unlock(session);
737 }
738 session_unlock_list();
739 return ret;
740
741 error:
742 session_unlock(session);
743 session_unlock_list();
744 return ret;
745 }
746
747 /*
748 * For each tracing session, update newly registered apps. The session list
749 * lock MUST be acquired before calling this.
750 */
751 static void update_ust_app(int app_sock)
752 {
753 struct ltt_session *sess, *stmp;
754
755 /* Consumer is in an ERROR state. Stop any application update. */
756 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
757 /* Stop the update process since the consumer is dead. */
758 return;
759 }
760
761 /* For all tracing session(s) */
762 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
763 session_lock(sess);
764 if (sess->ust_session) {
765 ust_app_global_update(sess->ust_session, app_sock);
766 }
767 session_unlock(sess);
768 }
769 }
770
771 /*
772 * This thread manage event coming from the kernel.
773 *
774 * Features supported in this thread:
775 * -) CPU Hotplug
776 */
777 static void *thread_manage_kernel(void *data)
778 {
779 int ret, i, pollfd, update_poll_flag = 1, err = -1;
780 uint32_t revents, nb_fd;
781 char tmp;
782 struct lttng_poll_event events;
783
784 DBG("[thread] Thread manage kernel started");
785
786 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
787
788 /*
789 * This first step of the while is to clean this structure which could free
790 * non NULL pointers so initialize it before the loop.
791 */
792 lttng_poll_init(&events);
793
794 if (testpoint(thread_manage_kernel)) {
795 goto error_testpoint;
796 }
797
798 health_code_update();
799
800 if (testpoint(thread_manage_kernel_before_loop)) {
801 goto error_testpoint;
802 }
803
804 while (1) {
805 health_code_update();
806
807 if (update_poll_flag == 1) {
808 /* Clean events object. We are about to populate it again. */
809 lttng_poll_clean(&events);
810
811 ret = sessiond_set_thread_pollset(&events, 2);
812 if (ret < 0) {
813 goto error_poll_create;
814 }
815
816 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
817 if (ret < 0) {
818 goto error;
819 }
820
821 /* This will add the available kernel channel if any. */
822 ret = update_kernel_poll(&events);
823 if (ret < 0) {
824 goto error;
825 }
826 update_poll_flag = 0;
827 }
828
829 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
830
831 /* Poll infinite value of time */
832 restart:
833 health_poll_entry();
834 ret = lttng_poll_wait(&events, -1);
835 health_poll_exit();
836 if (ret < 0) {
837 /*
838 * Restart interrupted system call.
839 */
840 if (errno == EINTR) {
841 goto restart;
842 }
843 goto error;
844 } else if (ret == 0) {
845 /* Should not happen since timeout is infinite */
846 ERR("Return value of poll is 0 with an infinite timeout.\n"
847 "This should not have happened! Continuing...");
848 continue;
849 }
850
851 nb_fd = ret;
852
853 for (i = 0; i < nb_fd; i++) {
854 /* Fetch once the poll data */
855 revents = LTTNG_POLL_GETEV(&events, i);
856 pollfd = LTTNG_POLL_GETFD(&events, i);
857
858 health_code_update();
859
860 /* Thread quit pipe has been closed. Killing thread. */
861 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
862 if (ret) {
863 err = 0;
864 goto exit;
865 }
866
867 /* Check for data on kernel pipe */
868 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
869 do {
870 ret = read(kernel_poll_pipe[0], &tmp, 1);
871 } while (ret < 0 && errno == EINTR);
872 /*
873 * Ret value is useless here, if this pipe gets any actions an
874 * update is required anyway.
875 */
876 update_poll_flag = 1;
877 continue;
878 } else {
879 /*
880 * New CPU detected by the kernel. Adding kernel stream to
881 * kernel session and updating the kernel consumer
882 */
883 if (revents & LPOLLIN) {
884 ret = update_kernel_stream(&kconsumer_data, pollfd);
885 if (ret < 0) {
886 continue;
887 }
888 break;
889 /*
890 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
891 * and unregister kernel stream at this point.
892 */
893 }
894 }
895 }
896 }
897
898 exit:
899 error:
900 lttng_poll_clean(&events);
901 error_poll_create:
902 error_testpoint:
903 utils_close_pipe(kernel_poll_pipe);
904 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
905 if (err) {
906 health_error();
907 ERR("Health error occurred in %s", __func__);
908 WARN("Kernel thread died unexpectedly. "
909 "Kernel tracing can continue but CPU hotplug is disabled.");
910 }
911 health_unregister(health_sessiond);
912 DBG("Kernel thread dying");
913 return NULL;
914 }
915
916 /*
917 * Signal pthread condition of the consumer data that the thread.
918 */
919 static void signal_consumer_condition(struct consumer_data *data, int state)
920 {
921 pthread_mutex_lock(&data->cond_mutex);
922
923 /*
924 * The state is set before signaling. It can be any value, it's the waiter
925 * job to correctly interpret this condition variable associated to the
926 * consumer pthread_cond.
927 *
928 * A value of 0 means that the corresponding thread of the consumer data
929 * was not started. 1 indicates that the thread has started and is ready
930 * for action. A negative value means that there was an error during the
931 * thread bootstrap.
932 */
933 data->consumer_thread_is_ready = state;
934 (void) pthread_cond_signal(&data->cond);
935
936 pthread_mutex_unlock(&data->cond_mutex);
937 }
938
939 /*
940 * This thread manage the consumer error sent back to the session daemon.
941 */
942 static void *thread_manage_consumer(void *data)
943 {
944 int sock = -1, i, ret, pollfd, err = -1;
945 uint32_t revents, nb_fd;
946 enum lttcomm_return_code code;
947 struct lttng_poll_event events;
948 struct consumer_data *consumer_data = data;
949
950 DBG("[thread] Manage consumer started");
951
952 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
953
954 health_code_update();
955
956 /*
957 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
958 * metadata_sock. Nothing more will be added to this poll set.
959 */
960 ret = sessiond_set_thread_pollset(&events, 3);
961 if (ret < 0) {
962 goto error_poll;
963 }
964
965 /*
966 * The error socket here is already in a listening state which was done
967 * just before spawning this thread to avoid a race between the consumer
968 * daemon exec trying to connect and the listen() call.
969 */
970 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
971 if (ret < 0) {
972 goto error;
973 }
974
975 health_code_update();
976
977 /* Infinite blocking call, waiting for transmission */
978 restart:
979 health_poll_entry();
980
981 if (testpoint(thread_manage_consumer)) {
982 goto error;
983 }
984
985 ret = lttng_poll_wait(&events, -1);
986 health_poll_exit();
987 if (ret < 0) {
988 /*
989 * Restart interrupted system call.
990 */
991 if (errno == EINTR) {
992 goto restart;
993 }
994 goto error;
995 }
996
997 nb_fd = ret;
998
999 for (i = 0; i < nb_fd; i++) {
1000 /* Fetch once the poll data */
1001 revents = LTTNG_POLL_GETEV(&events, i);
1002 pollfd = LTTNG_POLL_GETFD(&events, i);
1003
1004 health_code_update();
1005
1006 /* Thread quit pipe has been closed. Killing thread. */
1007 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1008 if (ret) {
1009 err = 0;
1010 goto exit;
1011 }
1012
1013 /* Event on the registration socket */
1014 if (pollfd == consumer_data->err_sock) {
1015 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1016 ERR("consumer err socket poll error");
1017 goto error;
1018 }
1019 }
1020 }
1021
1022 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1023 if (sock < 0) {
1024 goto error;
1025 }
1026
1027 /*
1028 * Set the CLOEXEC flag. Return code is useless because either way, the
1029 * show must go on.
1030 */
1031 (void) utils_set_fd_cloexec(sock);
1032
1033 health_code_update();
1034
1035 DBG2("Receiving code from consumer err_sock");
1036
1037 /* Getting status code from kconsumerd */
1038 ret = lttcomm_recv_unix_sock(sock, &code,
1039 sizeof(enum lttcomm_return_code));
1040 if (ret <= 0) {
1041 goto error;
1042 }
1043
1044 health_code_update();
1045
1046 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1047 /* Connect both socket, command and metadata. */
1048 consumer_data->cmd_sock =
1049 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1050 consumer_data->metadata_fd =
1051 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1052 if (consumer_data->cmd_sock < 0
1053 || consumer_data->metadata_fd < 0) {
1054 PERROR("consumer connect cmd socket");
1055 /* On error, signal condition and quit. */
1056 signal_consumer_condition(consumer_data, -1);
1057 goto error;
1058 }
1059 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1060 /* Create metadata socket lock. */
1061 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1062 if (consumer_data->metadata_sock.lock == NULL) {
1063 PERROR("zmalloc pthread mutex");
1064 ret = -1;
1065 goto error;
1066 }
1067 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1068
1069 signal_consumer_condition(consumer_data, 1);
1070 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1071 DBG("Consumer metadata socket ready (fd: %d)",
1072 consumer_data->metadata_fd);
1073 } else {
1074 ERR("consumer error when waiting for SOCK_READY : %s",
1075 lttcomm_get_readable_code(-code));
1076 goto error;
1077 }
1078
1079 /* Remove the consumerd error sock since we've established a connexion */
1080 ret = lttng_poll_del(&events, consumer_data->err_sock);
1081 if (ret < 0) {
1082 goto error;
1083 }
1084
1085 /* Add new accepted error socket. */
1086 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1087 if (ret < 0) {
1088 goto error;
1089 }
1090
1091 /* Add metadata socket that is successfully connected. */
1092 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1093 LPOLLIN | LPOLLRDHUP);
1094 if (ret < 0) {
1095 goto error;
1096 }
1097
1098 health_code_update();
1099
1100 /* Infinite blocking call, waiting for transmission */
1101 restart_poll:
1102 while (1) {
1103 health_poll_entry();
1104 ret = lttng_poll_wait(&events, -1);
1105 health_poll_exit();
1106 if (ret < 0) {
1107 /*
1108 * Restart interrupted system call.
1109 */
1110 if (errno == EINTR) {
1111 goto restart_poll;
1112 }
1113 goto error;
1114 }
1115
1116 nb_fd = ret;
1117
1118 for (i = 0; i < nb_fd; i++) {
1119 /* Fetch once the poll data */
1120 revents = LTTNG_POLL_GETEV(&events, i);
1121 pollfd = LTTNG_POLL_GETFD(&events, i);
1122
1123 health_code_update();
1124
1125 /* Thread quit pipe has been closed. Killing thread. */
1126 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1127 if (ret) {
1128 err = 0;
1129 goto exit;
1130 }
1131
1132 if (pollfd == sock) {
1133 /* Event on the consumerd socket */
1134 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1135 ERR("consumer err socket second poll error");
1136 goto error;
1137 }
1138 health_code_update();
1139 /* Wait for any kconsumerd error */
1140 ret = lttcomm_recv_unix_sock(sock, &code,
1141 sizeof(enum lttcomm_return_code));
1142 if (ret <= 0) {
1143 ERR("consumer closed the command socket");
1144 goto error;
1145 }
1146
1147 ERR("consumer return code : %s",
1148 lttcomm_get_readable_code(-code));
1149
1150 goto exit;
1151 } else if (pollfd == consumer_data->metadata_fd) {
1152 /* UST metadata requests */
1153 ret = ust_consumer_metadata_request(
1154 &consumer_data->metadata_sock);
1155 if (ret < 0) {
1156 ERR("Handling metadata request");
1157 goto error;
1158 }
1159 break;
1160 } else {
1161 ERR("Unknown pollfd");
1162 goto error;
1163 }
1164 }
1165 health_code_update();
1166 }
1167
1168 exit:
1169 error:
1170 /*
1171 * We lock here because we are about to close the sockets and some other
1172 * thread might be using them so get exclusive access which will abort all
1173 * other consumer command by other threads.
1174 */
1175 pthread_mutex_lock(&consumer_data->lock);
1176
1177 /* Immediately set the consumerd state to stopped */
1178 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1179 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1180 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1181 consumer_data->type == LTTNG_CONSUMER32_UST) {
1182 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1183 } else {
1184 /* Code flow error... */
1185 assert(0);
1186 }
1187
1188 if (consumer_data->err_sock >= 0) {
1189 ret = close(consumer_data->err_sock);
1190 if (ret) {
1191 PERROR("close");
1192 }
1193 consumer_data->err_sock = -1;
1194 }
1195 if (consumer_data->cmd_sock >= 0) {
1196 ret = close(consumer_data->cmd_sock);
1197 if (ret) {
1198 PERROR("close");
1199 }
1200 consumer_data->cmd_sock = -1;
1201 }
1202 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1203 ret = close(*consumer_data->metadata_sock.fd_ptr);
1204 if (ret) {
1205 PERROR("close");
1206 }
1207 }
1208
1209 if (sock >= 0) {
1210 ret = close(sock);
1211 if (ret) {
1212 PERROR("close");
1213 }
1214 }
1215
1216 unlink(consumer_data->err_unix_sock_path);
1217 unlink(consumer_data->cmd_unix_sock_path);
1218 consumer_data->pid = 0;
1219 pthread_mutex_unlock(&consumer_data->lock);
1220
1221 /* Cleanup metadata socket mutex. */
1222 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1223 free(consumer_data->metadata_sock.lock);
1224
1225 lttng_poll_clean(&events);
1226 error_poll:
1227 if (err) {
1228 health_error();
1229 ERR("Health error occurred in %s", __func__);
1230 }
1231 health_unregister(health_sessiond);
1232 DBG("consumer thread cleanup completed");
1233
1234 return NULL;
1235 }
1236
1237 /*
1238 * This thread manage application communication.
1239 */
1240 static void *thread_manage_apps(void *data)
1241 {
1242 int i, ret, pollfd, err = -1;
1243 uint32_t revents, nb_fd;
1244 struct lttng_poll_event events;
1245
1246 DBG("[thread] Manage application started");
1247
1248 rcu_register_thread();
1249 rcu_thread_online();
1250
1251 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1252
1253 if (testpoint(thread_manage_apps)) {
1254 goto error_testpoint;
1255 }
1256
1257 health_code_update();
1258
1259 ret = sessiond_set_thread_pollset(&events, 2);
1260 if (ret < 0) {
1261 goto error_poll_create;
1262 }
1263
1264 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1265 if (ret < 0) {
1266 goto error;
1267 }
1268
1269 if (testpoint(thread_manage_apps_before_loop)) {
1270 goto error;
1271 }
1272
1273 health_code_update();
1274
1275 while (1) {
1276 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1277
1278 /* Inifinite blocking call, waiting for transmission */
1279 restart:
1280 health_poll_entry();
1281 ret = lttng_poll_wait(&events, -1);
1282 health_poll_exit();
1283 if (ret < 0) {
1284 /*
1285 * Restart interrupted system call.
1286 */
1287 if (errno == EINTR) {
1288 goto restart;
1289 }
1290 goto error;
1291 }
1292
1293 nb_fd = ret;
1294
1295 for (i = 0; i < nb_fd; i++) {
1296 /* Fetch once the poll data */
1297 revents = LTTNG_POLL_GETEV(&events, i);
1298 pollfd = LTTNG_POLL_GETFD(&events, i);
1299
1300 health_code_update();
1301
1302 /* Thread quit pipe has been closed. Killing thread. */
1303 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1304 if (ret) {
1305 err = 0;
1306 goto exit;
1307 }
1308
1309 /* Inspect the apps cmd pipe */
1310 if (pollfd == apps_cmd_pipe[0]) {
1311 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1312 ERR("Apps command pipe error");
1313 goto error;
1314 } else if (revents & LPOLLIN) {
1315 int sock;
1316
1317 /* Empty pipe */
1318 do {
1319 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1320 } while (ret < 0 && errno == EINTR);
1321 if (ret < 0 || ret < sizeof(sock)) {
1322 PERROR("read apps cmd pipe");
1323 goto error;
1324 }
1325
1326 health_code_update();
1327
1328 /*
1329 * We only monitor the error events of the socket. This
1330 * thread does not handle any incoming data from UST
1331 * (POLLIN).
1332 */
1333 ret = lttng_poll_add(&events, sock,
1334 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1335 if (ret < 0) {
1336 goto error;
1337 }
1338
1339 DBG("Apps with sock %d added to poll set", sock);
1340
1341 health_code_update();
1342
1343 break;
1344 }
1345 } else {
1346 /*
1347 * At this point, we know that a registered application made
1348 * the event at poll_wait.
1349 */
1350 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1351 /* Removing from the poll set */
1352 ret = lttng_poll_del(&events, pollfd);
1353 if (ret < 0) {
1354 goto error;
1355 }
1356
1357 /* Socket closed on remote end. */
1358 ust_app_unregister(pollfd);
1359 break;
1360 }
1361 }
1362
1363 health_code_update();
1364 }
1365 }
1366
1367 exit:
1368 error:
1369 lttng_poll_clean(&events);
1370 error_poll_create:
1371 error_testpoint:
1372 utils_close_pipe(apps_cmd_pipe);
1373 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1374
1375 /*
1376 * We don't clean the UST app hash table here since already registered
1377 * applications can still be controlled so let them be until the session
1378 * daemon dies or the applications stop.
1379 */
1380
1381 if (err) {
1382 health_error();
1383 ERR("Health error occurred in %s", __func__);
1384 }
1385 health_unregister(health_sessiond);
1386 DBG("Application communication apps thread cleanup complete");
1387 rcu_thread_offline();
1388 rcu_unregister_thread();
1389 return NULL;
1390 }
1391
1392 /*
1393 * Send a socket to a thread This is called from the dispatch UST registration
1394 * thread once all sockets are set for the application.
1395 *
1396 * The sock value can be invalid, we don't really care, the thread will handle
1397 * it and make the necessary cleanup if so.
1398 *
1399 * On success, return 0 else a negative value being the errno message of the
1400 * write().
1401 */
1402 static int send_socket_to_thread(int fd, int sock)
1403 {
1404 int ret;
1405
1406 /*
1407 * It's possible that the FD is set as invalid with -1 concurrently just
1408 * before calling this function being a shutdown state of the thread.
1409 */
1410 if (fd < 0) {
1411 ret = -EBADF;
1412 goto error;
1413 }
1414
1415 do {
1416 ret = write(fd, &sock, sizeof(sock));
1417 } while (ret < 0 && errno == EINTR);
1418 if (ret < 0 || ret != sizeof(sock)) {
1419 PERROR("write apps pipe %d", fd);
1420 if (ret < 0) {
1421 ret = -errno;
1422 }
1423 goto error;
1424 }
1425
1426 /* All good. Don't send back the write positive ret value. */
1427 ret = 0;
1428 error:
1429 return ret;
1430 }
1431
1432 /*
1433 * Sanitize the wait queue of the dispatch registration thread meaning removing
1434 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1435 * notify socket is never received.
1436 */
1437 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1438 {
1439 int ret, nb_fd = 0, i;
1440 unsigned int fd_added = 0;
1441 struct lttng_poll_event events;
1442 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1443
1444 assert(wait_queue);
1445
1446 lttng_poll_init(&events);
1447
1448 /* Just skip everything for an empty queue. */
1449 if (!wait_queue->count) {
1450 goto end;
1451 }
1452
1453 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1454 if (ret < 0) {
1455 goto error_create;
1456 }
1457
1458 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1459 &wait_queue->head, head) {
1460 assert(wait_node->app);
1461 ret = lttng_poll_add(&events, wait_node->app->sock,
1462 LPOLLHUP | LPOLLERR);
1463 if (ret < 0) {
1464 goto error;
1465 }
1466
1467 fd_added = 1;
1468 }
1469
1470 if (!fd_added) {
1471 goto end;
1472 }
1473
1474 /*
1475 * Poll but don't block so we can quickly identify the faulty events and
1476 * clean them afterwards from the wait queue.
1477 */
1478 ret = lttng_poll_wait(&events, 0);
1479 if (ret < 0) {
1480 goto error;
1481 }
1482 nb_fd = ret;
1483
1484 for (i = 0; i < nb_fd; i++) {
1485 /* Get faulty FD. */
1486 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1487 int pollfd = LTTNG_POLL_GETFD(&events, i);
1488
1489 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1490 &wait_queue->head, head) {
1491 if (pollfd == wait_node->app->sock &&
1492 (revents & (LPOLLHUP | LPOLLERR))) {
1493 cds_list_del(&wait_node->head);
1494 wait_queue->count--;
1495 ust_app_destroy(wait_node->app);
1496 free(wait_node);
1497 break;
1498 }
1499 }
1500 }
1501
1502 if (nb_fd > 0) {
1503 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1504 }
1505
1506 end:
1507 lttng_poll_clean(&events);
1508 return;
1509
1510 error:
1511 lttng_poll_clean(&events);
1512 error_create:
1513 ERR("Unable to sanitize wait queue");
1514 return;
1515 }
1516
1517 /*
1518 * Dispatch request from the registration threads to the application
1519 * communication thread.
1520 */
1521 static void *thread_dispatch_ust_registration(void *data)
1522 {
1523 int ret, err = -1;
1524 struct cds_wfq_node *node;
1525 struct ust_command *ust_cmd = NULL;
1526 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1527 struct ust_reg_wait_queue wait_queue = {
1528 .count = 0,
1529 };
1530
1531 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1532
1533 health_code_update();
1534
1535 CDS_INIT_LIST_HEAD(&wait_queue.head);
1536
1537 DBG("[thread] Dispatch UST command started");
1538
1539 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1540 health_code_update();
1541
1542 /* Atomically prepare the queue futex */
1543 futex_nto1_prepare(&ust_cmd_queue.futex);
1544
1545 do {
1546 struct ust_app *app = NULL;
1547 ust_cmd = NULL;
1548
1549 /*
1550 * Make sure we don't have node(s) that have hung up before receiving
1551 * the notify socket. This is to clean the list in order to avoid
1552 * memory leaks from notify socket that are never seen.
1553 */
1554 sanitize_wait_queue(&wait_queue);
1555
1556 health_code_update();
1557 /* Dequeue command for registration */
1558 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1559 if (node == NULL) {
1560 DBG("Woken up but nothing in the UST command queue");
1561 /* Continue thread execution */
1562 break;
1563 }
1564
1565 ust_cmd = caa_container_of(node, struct ust_command, node);
1566
1567 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1568 " gid:%d sock:%d name:%s (version %d.%d)",
1569 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1570 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1571 ust_cmd->sock, ust_cmd->reg_msg.name,
1572 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1573
1574 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1575 wait_node = zmalloc(sizeof(*wait_node));
1576 if (!wait_node) {
1577 PERROR("zmalloc wait_node dispatch");
1578 ret = close(ust_cmd->sock);
1579 if (ret < 0) {
1580 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1581 }
1582 lttng_fd_put(LTTNG_FD_APPS, 1);
1583 free(ust_cmd);
1584 goto error;
1585 }
1586 CDS_INIT_LIST_HEAD(&wait_node->head);
1587
1588 /* Create application object if socket is CMD. */
1589 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1590 ust_cmd->sock);
1591 if (!wait_node->app) {
1592 ret = close(ust_cmd->sock);
1593 if (ret < 0) {
1594 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1595 }
1596 lttng_fd_put(LTTNG_FD_APPS, 1);
1597 free(wait_node);
1598 free(ust_cmd);
1599 continue;
1600 }
1601 /*
1602 * Add application to the wait queue so we can set the notify
1603 * socket before putting this object in the global ht.
1604 */
1605 cds_list_add(&wait_node->head, &wait_queue.head);
1606 wait_queue.count++;
1607
1608 free(ust_cmd);
1609 /*
1610 * We have to continue here since we don't have the notify
1611 * socket and the application MUST be added to the hash table
1612 * only at that moment.
1613 */
1614 continue;
1615 } else {
1616 /*
1617 * Look for the application in the local wait queue and set the
1618 * notify socket if found.
1619 */
1620 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1621 &wait_queue.head, head) {
1622 health_code_update();
1623 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1624 wait_node->app->notify_sock = ust_cmd->sock;
1625 cds_list_del(&wait_node->head);
1626 wait_queue.count--;
1627 app = wait_node->app;
1628 free(wait_node);
1629 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1630 break;
1631 }
1632 }
1633
1634 /*
1635 * With no application at this stage the received socket is
1636 * basically useless so close it before we free the cmd data
1637 * structure for good.
1638 */
1639 if (!app) {
1640 ret = close(ust_cmd->sock);
1641 if (ret < 0) {
1642 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1643 }
1644 lttng_fd_put(LTTNG_FD_APPS, 1);
1645 }
1646 free(ust_cmd);
1647 }
1648
1649 if (app) {
1650 /*
1651 * @session_lock_list
1652 *
1653 * Lock the global session list so from the register up to the
1654 * registration done message, no thread can see the application
1655 * and change its state.
1656 */
1657 session_lock_list();
1658 rcu_read_lock();
1659
1660 /*
1661 * Add application to the global hash table. This needs to be
1662 * done before the update to the UST registry can locate the
1663 * application.
1664 */
1665 ust_app_add(app);
1666
1667 /* Set app version. This call will print an error if needed. */
1668 (void) ust_app_version(app);
1669
1670 /* Send notify socket through the notify pipe. */
1671 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1672 app->notify_sock);
1673 if (ret < 0) {
1674 rcu_read_unlock();
1675 session_unlock_list();
1676 /*
1677 * No notify thread, stop the UST tracing. However, this is
1678 * not an internal error of the this thread thus setting
1679 * the health error code to a normal exit.
1680 */
1681 err = 0;
1682 goto error;
1683 }
1684
1685 /*
1686 * Update newly registered application with the tracing
1687 * registry info already enabled information.
1688 */
1689 update_ust_app(app->sock);
1690
1691 /*
1692 * Don't care about return value. Let the manage apps threads
1693 * handle app unregistration upon socket close.
1694 */
1695 (void) ust_app_register_done(app->sock);
1696
1697 /*
1698 * Even if the application socket has been closed, send the app
1699 * to the thread and unregistration will take place at that
1700 * place.
1701 */
1702 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1703 if (ret < 0) {
1704 rcu_read_unlock();
1705 session_unlock_list();
1706 /*
1707 * No apps. thread, stop the UST tracing. However, this is
1708 * not an internal error of the this thread thus setting
1709 * the health error code to a normal exit.
1710 */
1711 err = 0;
1712 goto error;
1713 }
1714
1715 rcu_read_unlock();
1716 session_unlock_list();
1717 }
1718 } while (node != NULL);
1719
1720 health_poll_entry();
1721 /* Futex wait on queue. Blocking call on futex() */
1722 futex_nto1_wait(&ust_cmd_queue.futex);
1723 health_poll_exit();
1724 }
1725 /* Normal exit, no error */
1726 err = 0;
1727
1728 error:
1729 /* Clean up wait queue. */
1730 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1731 &wait_queue.head, head) {
1732 cds_list_del(&wait_node->head);
1733 wait_queue.count--;
1734 free(wait_node);
1735 }
1736
1737 DBG("Dispatch thread dying");
1738 if (err) {
1739 health_error();
1740 ERR("Health error occurred in %s", __func__);
1741 }
1742 health_unregister(health_sessiond);
1743 return NULL;
1744 }
1745
1746 /*
1747 * This thread manage application registration.
1748 */
1749 static void *thread_registration_apps(void *data)
1750 {
1751 int sock = -1, i, ret, pollfd, err = -1;
1752 uint32_t revents, nb_fd;
1753 struct lttng_poll_event events;
1754 /*
1755 * Get allocated in this thread, enqueued to a global queue, dequeued and
1756 * freed in the manage apps thread.
1757 */
1758 struct ust_command *ust_cmd = NULL;
1759
1760 DBG("[thread] Manage application registration started");
1761
1762 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1763
1764 if (testpoint(thread_registration_apps)) {
1765 goto error_testpoint;
1766 }
1767
1768 ret = lttcomm_listen_unix_sock(apps_sock);
1769 if (ret < 0) {
1770 goto error_listen;
1771 }
1772
1773 /*
1774 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1775 * more will be added to this poll set.
1776 */
1777 ret = sessiond_set_thread_pollset(&events, 2);
1778 if (ret < 0) {
1779 goto error_create_poll;
1780 }
1781
1782 /* Add the application registration socket */
1783 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1784 if (ret < 0) {
1785 goto error_poll_add;
1786 }
1787
1788 /* Notify all applications to register */
1789 ret = notify_ust_apps(1);
1790 if (ret < 0) {
1791 ERR("Failed to notify applications or create the wait shared memory.\n"
1792 "Execution continues but there might be problem for already\n"
1793 "running applications that wishes to register.");
1794 }
1795
1796 while (1) {
1797 DBG("Accepting application registration");
1798
1799 /* Inifinite blocking call, waiting for transmission */
1800 restart:
1801 health_poll_entry();
1802 ret = lttng_poll_wait(&events, -1);
1803 health_poll_exit();
1804 if (ret < 0) {
1805 /*
1806 * Restart interrupted system call.
1807 */
1808 if (errno == EINTR) {
1809 goto restart;
1810 }
1811 goto error;
1812 }
1813
1814 nb_fd = ret;
1815
1816 for (i = 0; i < nb_fd; i++) {
1817 health_code_update();
1818
1819 /* Fetch once the poll data */
1820 revents = LTTNG_POLL_GETEV(&events, i);
1821 pollfd = LTTNG_POLL_GETFD(&events, i);
1822
1823 /* Thread quit pipe has been closed. Killing thread. */
1824 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1825 if (ret) {
1826 err = 0;
1827 goto exit;
1828 }
1829
1830 /* Event on the registration socket */
1831 if (pollfd == apps_sock) {
1832 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1833 ERR("Register apps socket poll error");
1834 goto error;
1835 } else if (revents & LPOLLIN) {
1836 sock = lttcomm_accept_unix_sock(apps_sock);
1837 if (sock < 0) {
1838 goto error;
1839 }
1840
1841 /*
1842 * Set socket timeout for both receiving and ending.
1843 * app_socket_timeout is in seconds, whereas
1844 * lttcomm_setsockopt_rcv_timeout and
1845 * lttcomm_setsockopt_snd_timeout expect msec as
1846 * parameter.
1847 */
1848 (void) lttcomm_setsockopt_rcv_timeout(sock,
1849 app_socket_timeout * 1000);
1850 (void) lttcomm_setsockopt_snd_timeout(sock,
1851 app_socket_timeout * 1000);
1852
1853 /*
1854 * Set the CLOEXEC flag. Return code is useless because
1855 * either way, the show must go on.
1856 */
1857 (void) utils_set_fd_cloexec(sock);
1858
1859 /* Create UST registration command for enqueuing */
1860 ust_cmd = zmalloc(sizeof(struct ust_command));
1861 if (ust_cmd == NULL) {
1862 PERROR("ust command zmalloc");
1863 goto error;
1864 }
1865
1866 /*
1867 * Using message-based transmissions to ensure we don't
1868 * have to deal with partially received messages.
1869 */
1870 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1871 if (ret < 0) {
1872 ERR("Exhausted file descriptors allowed for applications.");
1873 free(ust_cmd);
1874 ret = close(sock);
1875 if (ret) {
1876 PERROR("close");
1877 }
1878 sock = -1;
1879 continue;
1880 }
1881
1882 health_code_update();
1883 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1884 if (ret < 0) {
1885 free(ust_cmd);
1886 /* Close socket of the application. */
1887 ret = close(sock);
1888 if (ret) {
1889 PERROR("close");
1890 }
1891 lttng_fd_put(LTTNG_FD_APPS, 1);
1892 sock = -1;
1893 continue;
1894 }
1895 health_code_update();
1896
1897 ust_cmd->sock = sock;
1898 sock = -1;
1899
1900 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1901 " gid:%d sock:%d name:%s (version %d.%d)",
1902 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1903 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1904 ust_cmd->sock, ust_cmd->reg_msg.name,
1905 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1906
1907 /*
1908 * Lock free enqueue the registration request. The red pill
1909 * has been taken! This apps will be part of the *system*.
1910 */
1911 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1912
1913 /*
1914 * Wake the registration queue futex. Implicit memory
1915 * barrier with the exchange in cds_wfq_enqueue.
1916 */
1917 futex_nto1_wake(&ust_cmd_queue.futex);
1918 }
1919 }
1920 }
1921 }
1922
1923 exit:
1924 error:
1925 if (err) {
1926 health_error();
1927 ERR("Health error occurred in %s", __func__);
1928 }
1929
1930 /* Notify that the registration thread is gone */
1931 notify_ust_apps(0);
1932
1933 if (apps_sock >= 0) {
1934 ret = close(apps_sock);
1935 if (ret) {
1936 PERROR("close");
1937 }
1938 }
1939 if (sock >= 0) {
1940 ret = close(sock);
1941 if (ret) {
1942 PERROR("close");
1943 }
1944 lttng_fd_put(LTTNG_FD_APPS, 1);
1945 }
1946 unlink(apps_unix_sock_path);
1947
1948 error_poll_add:
1949 lttng_poll_clean(&events);
1950 error_listen:
1951 error_create_poll:
1952 error_testpoint:
1953 DBG("UST Registration thread cleanup complete");
1954 health_unregister(health_sessiond);
1955
1956 return NULL;
1957 }
1958
1959 /*
1960 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1961 * exec or it will fails.
1962 */
1963 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1964 {
1965 int ret, clock_ret;
1966 struct timespec timeout;
1967
1968 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1969 consumer_data->consumer_thread_is_ready = 0;
1970
1971 /* Setup pthread condition */
1972 ret = pthread_condattr_init(&consumer_data->condattr);
1973 if (ret != 0) {
1974 errno = ret;
1975 PERROR("pthread_condattr_init consumer data");
1976 goto error;
1977 }
1978
1979 /*
1980 * Set the monotonic clock in order to make sure we DO NOT jump in time
1981 * between the clock_gettime() call and the timedwait call. See bug #324
1982 * for a more details and how we noticed it.
1983 */
1984 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1985 if (ret != 0) {
1986 errno = ret;
1987 PERROR("pthread_condattr_setclock consumer data");
1988 goto error;
1989 }
1990
1991 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1992 if (ret != 0) {
1993 errno = ret;
1994 PERROR("pthread_cond_init consumer data");
1995 goto error;
1996 }
1997
1998 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1999 consumer_data);
2000 if (ret != 0) {
2001 PERROR("pthread_create consumer");
2002 ret = -1;
2003 goto error;
2004 }
2005
2006 /* We are about to wait on a pthread condition */
2007 pthread_mutex_lock(&consumer_data->cond_mutex);
2008
2009 /* Get time for sem_timedwait absolute timeout */
2010 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2011 /*
2012 * Set the timeout for the condition timed wait even if the clock gettime
2013 * call fails since we might loop on that call and we want to avoid to
2014 * increment the timeout too many times.
2015 */
2016 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2017
2018 /*
2019 * The following loop COULD be skipped in some conditions so this is why we
2020 * set ret to 0 in order to make sure at least one round of the loop is
2021 * done.
2022 */
2023 ret = 0;
2024
2025 /*
2026 * Loop until the condition is reached or when a timeout is reached. Note
2027 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2028 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2029 * possible. This loop does not take any chances and works with both of
2030 * them.
2031 */
2032 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2033 if (clock_ret < 0) {
2034 PERROR("clock_gettime spawn consumer");
2035 /* Infinite wait for the consumerd thread to be ready */
2036 ret = pthread_cond_wait(&consumer_data->cond,
2037 &consumer_data->cond_mutex);
2038 } else {
2039 ret = pthread_cond_timedwait(&consumer_data->cond,
2040 &consumer_data->cond_mutex, &timeout);
2041 }
2042 }
2043
2044 /* Release the pthread condition */
2045 pthread_mutex_unlock(&consumer_data->cond_mutex);
2046
2047 if (ret != 0) {
2048 errno = ret;
2049 if (ret == ETIMEDOUT) {
2050 /*
2051 * Call has timed out so we kill the kconsumerd_thread and return
2052 * an error.
2053 */
2054 ERR("Condition timed out. The consumer thread was never ready."
2055 " Killing it");
2056 ret = pthread_cancel(consumer_data->thread);
2057 if (ret < 0) {
2058 PERROR("pthread_cancel consumer thread");
2059 }
2060 } else {
2061 PERROR("pthread_cond_wait failed consumer thread");
2062 }
2063 goto error;
2064 }
2065
2066 pthread_mutex_lock(&consumer_data->pid_mutex);
2067 if (consumer_data->pid == 0) {
2068 ERR("Consumerd did not start");
2069 pthread_mutex_unlock(&consumer_data->pid_mutex);
2070 goto error;
2071 }
2072 pthread_mutex_unlock(&consumer_data->pid_mutex);
2073
2074 return 0;
2075
2076 error:
2077 return ret;
2078 }
2079
2080 /*
2081 * Join consumer thread
2082 */
2083 static int join_consumer_thread(struct consumer_data *consumer_data)
2084 {
2085 void *status;
2086
2087 /* Consumer pid must be a real one. */
2088 if (consumer_data->pid > 0) {
2089 int ret;
2090 ret = kill(consumer_data->pid, SIGTERM);
2091 if (ret) {
2092 ERR("Error killing consumer daemon");
2093 return ret;
2094 }
2095 return pthread_join(consumer_data->thread, &status);
2096 } else {
2097 return 0;
2098 }
2099 }
2100
2101 /*
2102 * Fork and exec a consumer daemon (consumerd).
2103 *
2104 * Return pid if successful else -1.
2105 */
2106 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2107 {
2108 int ret;
2109 pid_t pid;
2110 const char *consumer_to_use;
2111 const char *verbosity;
2112 struct stat st;
2113
2114 DBG("Spawning consumerd");
2115
2116 pid = fork();
2117 if (pid == 0) {
2118 /*
2119 * Exec consumerd.
2120 */
2121 if (opt_verbose_consumer) {
2122 verbosity = "--verbose";
2123 } else {
2124 verbosity = "--quiet";
2125 }
2126 switch (consumer_data->type) {
2127 case LTTNG_CONSUMER_KERNEL:
2128 /*
2129 * Find out which consumerd to execute. We will first try the
2130 * 64-bit path, then the sessiond's installation directory, and
2131 * fallback on the 32-bit one,
2132 */
2133 DBG3("Looking for a kernel consumer at these locations:");
2134 DBG3(" 1) %s", consumerd64_bin);
2135 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2136 DBG3(" 3) %s", consumerd32_bin);
2137 if (stat(consumerd64_bin, &st) == 0) {
2138 DBG3("Found location #1");
2139 consumer_to_use = consumerd64_bin;
2140 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2141 DBG3("Found location #2");
2142 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2143 } else if (stat(consumerd32_bin, &st) == 0) {
2144 DBG3("Found location #3");
2145 consumer_to_use = consumerd32_bin;
2146 } else {
2147 DBG("Could not find any valid consumerd executable");
2148 break;
2149 }
2150 DBG("Using kernel consumer at: %s", consumer_to_use);
2151 execl(consumer_to_use,
2152 "lttng-consumerd", verbosity, "-k",
2153 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2154 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2155 "--group", tracing_group_name,
2156 NULL);
2157 break;
2158 case LTTNG_CONSUMER64_UST:
2159 {
2160 char *tmpnew = NULL;
2161
2162 if (consumerd64_libdir[0] != '\0') {
2163 char *tmp;
2164 size_t tmplen;
2165
2166 tmp = getenv("LD_LIBRARY_PATH");
2167 if (!tmp) {
2168 tmp = "";
2169 }
2170 tmplen = strlen("LD_LIBRARY_PATH=")
2171 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2172 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2173 if (!tmpnew) {
2174 ret = -ENOMEM;
2175 goto error;
2176 }
2177 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2178 strcat(tmpnew, consumerd64_libdir);
2179 if (tmp[0] != '\0') {
2180 strcat(tmpnew, ":");
2181 strcat(tmpnew, tmp);
2182 }
2183 ret = putenv(tmpnew);
2184 if (ret) {
2185 ret = -errno;
2186 free(tmpnew);
2187 goto error;
2188 }
2189 }
2190 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2191 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2192 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2193 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2194 "--group", tracing_group_name,
2195 NULL);
2196 if (consumerd64_libdir[0] != '\0') {
2197 free(tmpnew);
2198 }
2199 if (ret) {
2200 goto error;
2201 }
2202 break;
2203 }
2204 case LTTNG_CONSUMER32_UST:
2205 {
2206 char *tmpnew = NULL;
2207
2208 if (consumerd32_libdir[0] != '\0') {
2209 char *tmp;
2210 size_t tmplen;
2211
2212 tmp = getenv("LD_LIBRARY_PATH");
2213 if (!tmp) {
2214 tmp = "";
2215 }
2216 tmplen = strlen("LD_LIBRARY_PATH=")
2217 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2218 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2219 if (!tmpnew) {
2220 ret = -ENOMEM;
2221 goto error;
2222 }
2223 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2224 strcat(tmpnew, consumerd32_libdir);
2225 if (tmp[0] != '\0') {
2226 strcat(tmpnew, ":");
2227 strcat(tmpnew, tmp);
2228 }
2229 ret = putenv(tmpnew);
2230 if (ret) {
2231 ret = -errno;
2232 free(tmpnew);
2233 goto error;
2234 }
2235 }
2236 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2237 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2238 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2239 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2240 "--group", tracing_group_name,
2241 NULL);
2242 if (consumerd32_libdir[0] != '\0') {
2243 free(tmpnew);
2244 }
2245 if (ret) {
2246 goto error;
2247 }
2248 break;
2249 }
2250 default:
2251 PERROR("unknown consumer type");
2252 exit(EXIT_FAILURE);
2253 }
2254 if (errno != 0) {
2255 PERROR("kernel start consumer exec");
2256 }
2257 exit(EXIT_FAILURE);
2258 } else if (pid > 0) {
2259 ret = pid;
2260 } else {
2261 PERROR("start consumer fork");
2262 ret = -errno;
2263 }
2264 error:
2265 return ret;
2266 }
2267
2268 /*
2269 * Spawn the consumerd daemon and session daemon thread.
2270 */
2271 static int start_consumerd(struct consumer_data *consumer_data)
2272 {
2273 int ret;
2274
2275 /*
2276 * Set the listen() state on the socket since there is a possible race
2277 * between the exec() of the consumer daemon and this call if place in the
2278 * consumer thread. See bug #366 for more details.
2279 */
2280 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2281 if (ret < 0) {
2282 goto error;
2283 }
2284
2285 pthread_mutex_lock(&consumer_data->pid_mutex);
2286 if (consumer_data->pid != 0) {
2287 pthread_mutex_unlock(&consumer_data->pid_mutex);
2288 goto end;
2289 }
2290
2291 ret = spawn_consumerd(consumer_data);
2292 if (ret < 0) {
2293 ERR("Spawning consumerd failed");
2294 pthread_mutex_unlock(&consumer_data->pid_mutex);
2295 goto error;
2296 }
2297
2298 /* Setting up the consumer_data pid */
2299 consumer_data->pid = ret;
2300 DBG2("Consumer pid %d", consumer_data->pid);
2301 pthread_mutex_unlock(&consumer_data->pid_mutex);
2302
2303 DBG2("Spawning consumer control thread");
2304 ret = spawn_consumer_thread(consumer_data);
2305 if (ret < 0) {
2306 ERR("Fatal error spawning consumer control thread");
2307 goto error;
2308 }
2309
2310 end:
2311 return 0;
2312
2313 error:
2314 /* Cleanup already created sockets on error. */
2315 if (consumer_data->err_sock >= 0) {
2316 int err;
2317
2318 err = close(consumer_data->err_sock);
2319 if (err < 0) {
2320 PERROR("close consumer data error socket");
2321 }
2322 }
2323 return ret;
2324 }
2325
2326 /*
2327 * Setup necessary data for kernel tracer action.
2328 */
2329 static int init_kernel_tracer(void)
2330 {
2331 int ret;
2332
2333 /* Modprobe lttng kernel modules */
2334 ret = modprobe_lttng_control();
2335 if (ret < 0) {
2336 goto error;
2337 }
2338
2339 /* Open debugfs lttng */
2340 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2341 if (kernel_tracer_fd < 0) {
2342 DBG("Failed to open %s", module_proc_lttng);
2343 ret = -1;
2344 goto error_open;
2345 }
2346
2347 /* Validate kernel version */
2348 ret = kernel_validate_version(kernel_tracer_fd);
2349 if (ret < 0) {
2350 goto error_version;
2351 }
2352
2353 ret = modprobe_lttng_data();
2354 if (ret < 0) {
2355 goto error_modules;
2356 }
2357
2358 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2359 return 0;
2360
2361 error_version:
2362 modprobe_remove_lttng_control();
2363 ret = close(kernel_tracer_fd);
2364 if (ret) {
2365 PERROR("close");
2366 }
2367 kernel_tracer_fd = -1;
2368 return LTTNG_ERR_KERN_VERSION;
2369
2370 error_modules:
2371 ret = close(kernel_tracer_fd);
2372 if (ret) {
2373 PERROR("close");
2374 }
2375
2376 error_open:
2377 modprobe_remove_lttng_control();
2378
2379 error:
2380 WARN("No kernel tracer available");
2381 kernel_tracer_fd = -1;
2382 if (!is_root) {
2383 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2384 } else {
2385 return LTTNG_ERR_KERN_NA;
2386 }
2387 }
2388
2389
2390 /*
2391 * Copy consumer output from the tracing session to the domain session. The
2392 * function also applies the right modification on a per domain basis for the
2393 * trace files destination directory.
2394 *
2395 * Should *NOT* be called with RCU read-side lock held.
2396 */
2397 static int copy_session_consumer(int domain, struct ltt_session *session)
2398 {
2399 int ret;
2400 const char *dir_name;
2401 struct consumer_output *consumer;
2402
2403 assert(session);
2404 assert(session->consumer);
2405
2406 switch (domain) {
2407 case LTTNG_DOMAIN_KERNEL:
2408 DBG3("Copying tracing session consumer output in kernel session");
2409 /*
2410 * XXX: We should audit the session creation and what this function
2411 * does "extra" in order to avoid a destroy since this function is used
2412 * in the domain session creation (kernel and ust) only. Same for UST
2413 * domain.
2414 */
2415 if (session->kernel_session->consumer) {
2416 consumer_destroy_output(session->kernel_session->consumer);
2417 }
2418 session->kernel_session->consumer =
2419 consumer_copy_output(session->consumer);
2420 /* Ease our life a bit for the next part */
2421 consumer = session->kernel_session->consumer;
2422 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2423 break;
2424 case LTTNG_DOMAIN_JUL:
2425 case LTTNG_DOMAIN_UST:
2426 DBG3("Copying tracing session consumer output in UST session");
2427 if (session->ust_session->consumer) {
2428 consumer_destroy_output(session->ust_session->consumer);
2429 }
2430 session->ust_session->consumer =
2431 consumer_copy_output(session->consumer);
2432 /* Ease our life a bit for the next part */
2433 consumer = session->ust_session->consumer;
2434 dir_name = DEFAULT_UST_TRACE_DIR;
2435 break;
2436 default:
2437 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2438 goto error;
2439 }
2440
2441 /* Append correct directory to subdir */
2442 strncat(consumer->subdir, dir_name,
2443 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2444 DBG3("Copy session consumer subdir %s", consumer->subdir);
2445
2446 ret = LTTNG_OK;
2447
2448 error:
2449 return ret;
2450 }
2451
2452 /*
2453 * Create an UST session and add it to the session ust list.
2454 *
2455 * Should *NOT* be called with RCU read-side lock held.
2456 */
2457 static int create_ust_session(struct ltt_session *session,
2458 struct lttng_domain *domain)
2459 {
2460 int ret;
2461 struct ltt_ust_session *lus = NULL;
2462
2463 assert(session);
2464 assert(domain);
2465 assert(session->consumer);
2466
2467 switch (domain->type) {
2468 case LTTNG_DOMAIN_JUL:
2469 case LTTNG_DOMAIN_UST:
2470 break;
2471 default:
2472 ERR("Unknown UST domain on create session %d", domain->type);
2473 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2474 goto error;
2475 }
2476
2477 DBG("Creating UST session");
2478
2479 lus = trace_ust_create_session(session->id);
2480 if (lus == NULL) {
2481 ret = LTTNG_ERR_UST_SESS_FAIL;
2482 goto error;
2483 }
2484
2485 lus->uid = session->uid;
2486 lus->gid = session->gid;
2487 lus->output_traces = session->output_traces;
2488 lus->snapshot_mode = session->snapshot_mode;
2489 lus->live_timer_interval = session->live_timer;
2490 session->ust_session = lus;
2491
2492 /* Copy session output to the newly created UST session */
2493 ret = copy_session_consumer(domain->type, session);
2494 if (ret != LTTNG_OK) {
2495 goto error;
2496 }
2497
2498 return LTTNG_OK;
2499
2500 error:
2501 free(lus);
2502 session->ust_session = NULL;
2503 return ret;
2504 }
2505
2506 /*
2507 * Create a kernel tracer session then create the default channel.
2508 */
2509 static int create_kernel_session(struct ltt_session *session)
2510 {
2511 int ret;
2512
2513 DBG("Creating kernel session");
2514
2515 ret = kernel_create_session(session, kernel_tracer_fd);
2516 if (ret < 0) {
2517 ret = LTTNG_ERR_KERN_SESS_FAIL;
2518 goto error;
2519 }
2520
2521 /* Code flow safety */
2522 assert(session->kernel_session);
2523
2524 /* Copy session output to the newly created Kernel session */
2525 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2526 if (ret != LTTNG_OK) {
2527 goto error;
2528 }
2529
2530 /* Create directory(ies) on local filesystem. */
2531 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2532 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2533 ret = run_as_mkdir_recursive(
2534 session->kernel_session->consumer->dst.trace_path,
2535 S_IRWXU | S_IRWXG, session->uid, session->gid);
2536 if (ret < 0) {
2537 if (ret != -EEXIST) {
2538 ERR("Trace directory creation error");
2539 goto error;
2540 }
2541 }
2542 }
2543
2544 session->kernel_session->uid = session->uid;
2545 session->kernel_session->gid = session->gid;
2546 session->kernel_session->output_traces = session->output_traces;
2547 session->kernel_session->snapshot_mode = session->snapshot_mode;
2548
2549 return LTTNG_OK;
2550
2551 error:
2552 trace_kernel_destroy_session(session->kernel_session);
2553 session->kernel_session = NULL;
2554 return ret;
2555 }
2556
2557 /*
2558 * Count number of session permitted by uid/gid.
2559 */
2560 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2561 {
2562 unsigned int i = 0;
2563 struct ltt_session *session;
2564
2565 DBG("Counting number of available session for UID %d GID %d",
2566 uid, gid);
2567 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2568 /*
2569 * Only list the sessions the user can control.
2570 */
2571 if (!session_access_ok(session, uid, gid)) {
2572 continue;
2573 }
2574 i++;
2575 }
2576 return i;
2577 }
2578
2579 /*
2580 * Process the command requested by the lttng client within the command
2581 * context structure. This function make sure that the return structure (llm)
2582 * is set and ready for transmission before returning.
2583 *
2584 * Return any error encountered or 0 for success.
2585 *
2586 * "sock" is only used for special-case var. len data.
2587 *
2588 * Should *NOT* be called with RCU read-side lock held.
2589 */
2590 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2591 int *sock_error)
2592 {
2593 int ret = LTTNG_OK;
2594 int need_tracing_session = 1;
2595 int need_domain;
2596
2597 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2598
2599 *sock_error = 0;
2600
2601 switch (cmd_ctx->lsm->cmd_type) {
2602 case LTTNG_CREATE_SESSION:
2603 case LTTNG_CREATE_SESSION_SNAPSHOT:
2604 case LTTNG_CREATE_SESSION_LIVE:
2605 case LTTNG_DESTROY_SESSION:
2606 case LTTNG_LIST_SESSIONS:
2607 case LTTNG_LIST_DOMAINS:
2608 case LTTNG_START_TRACE:
2609 case LTTNG_STOP_TRACE:
2610 case LTTNG_DATA_PENDING:
2611 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2612 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2613 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2614 case LTTNG_SNAPSHOT_RECORD:
2615 need_domain = 0;
2616 break;
2617 default:
2618 need_domain = 1;
2619 }
2620
2621 if (opt_no_kernel && need_domain
2622 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2623 if (!is_root) {
2624 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2625 } else {
2626 ret = LTTNG_ERR_KERN_NA;
2627 }
2628 goto error;
2629 }
2630
2631 /* Deny register consumer if we already have a spawned consumer. */
2632 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2633 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2634 if (kconsumer_data.pid > 0) {
2635 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2636 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2637 goto error;
2638 }
2639 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2640 }
2641
2642 /*
2643 * Check for command that don't needs to allocate a returned payload. We do
2644 * this here so we don't have to make the call for no payload at each
2645 * command.
2646 */
2647 switch(cmd_ctx->lsm->cmd_type) {
2648 case LTTNG_LIST_SESSIONS:
2649 case LTTNG_LIST_TRACEPOINTS:
2650 case LTTNG_LIST_TRACEPOINT_FIELDS:
2651 case LTTNG_LIST_DOMAINS:
2652 case LTTNG_LIST_CHANNELS:
2653 case LTTNG_LIST_EVENTS:
2654 break;
2655 default:
2656 /* Setup lttng message with no payload */
2657 ret = setup_lttng_msg(cmd_ctx, 0);
2658 if (ret < 0) {
2659 /* This label does not try to unlock the session */
2660 goto init_setup_error;
2661 }
2662 }
2663
2664 /* Commands that DO NOT need a session. */
2665 switch (cmd_ctx->lsm->cmd_type) {
2666 case LTTNG_CREATE_SESSION:
2667 case LTTNG_CREATE_SESSION_SNAPSHOT:
2668 case LTTNG_CREATE_SESSION_LIVE:
2669 case LTTNG_CALIBRATE:
2670 case LTTNG_LIST_SESSIONS:
2671 case LTTNG_LIST_TRACEPOINTS:
2672 case LTTNG_LIST_TRACEPOINT_FIELDS:
2673 need_tracing_session = 0;
2674 break;
2675 default:
2676 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2677 /*
2678 * We keep the session list lock across _all_ commands
2679 * for now, because the per-session lock does not
2680 * handle teardown properly.
2681 */
2682 session_lock_list();
2683 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2684 if (cmd_ctx->session == NULL) {
2685 ret = LTTNG_ERR_SESS_NOT_FOUND;
2686 goto error;
2687 } else {
2688 /* Acquire lock for the session */
2689 session_lock(cmd_ctx->session);
2690 }
2691 break;
2692 }
2693
2694 if (!need_domain) {
2695 goto skip_domain;
2696 }
2697
2698 /*
2699 * Check domain type for specific "pre-action".
2700 */
2701 switch (cmd_ctx->lsm->domain.type) {
2702 case LTTNG_DOMAIN_KERNEL:
2703 if (!is_root) {
2704 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2705 goto error;
2706 }
2707
2708 /* Kernel tracer check */
2709 if (kernel_tracer_fd == -1) {
2710 /* Basically, load kernel tracer modules */
2711 ret = init_kernel_tracer();
2712 if (ret != 0) {
2713 goto error;
2714 }
2715 }
2716
2717 /* Consumer is in an ERROR state. Report back to client */
2718 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2719 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2720 goto error;
2721 }
2722
2723 /* Need a session for kernel command */
2724 if (need_tracing_session) {
2725 if (cmd_ctx->session->kernel_session == NULL) {
2726 ret = create_kernel_session(cmd_ctx->session);
2727 if (ret < 0) {
2728 ret = LTTNG_ERR_KERN_SESS_FAIL;
2729 goto error;
2730 }
2731 }
2732
2733 /* Start the kernel consumer daemon */
2734 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2735 if (kconsumer_data.pid == 0 &&
2736 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2737 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2738 ret = start_consumerd(&kconsumer_data);
2739 if (ret < 0) {
2740 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2741 goto error;
2742 }
2743 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2744 } else {
2745 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2746 }
2747
2748 /*
2749 * The consumer was just spawned so we need to add the socket to
2750 * the consumer output of the session if exist.
2751 */
2752 ret = consumer_create_socket(&kconsumer_data,
2753 cmd_ctx->session->kernel_session->consumer);
2754 if (ret < 0) {
2755 goto error;
2756 }
2757 }
2758
2759 break;
2760 case LTTNG_DOMAIN_JUL:
2761 case LTTNG_DOMAIN_UST:
2762 {
2763 if (!ust_app_supported()) {
2764 ret = LTTNG_ERR_NO_UST;
2765 goto error;
2766 }
2767 /* Consumer is in an ERROR state. Report back to client */
2768 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2769 ret = LTTNG_ERR_NO_USTCONSUMERD;
2770 goto error;
2771 }
2772
2773 if (need_tracing_session) {
2774 /* Create UST session if none exist. */
2775 if (cmd_ctx->session->ust_session == NULL) {
2776 ret = create_ust_session(cmd_ctx->session,
2777 &cmd_ctx->lsm->domain);
2778 if (ret != LTTNG_OK) {
2779 goto error;
2780 }
2781 }
2782
2783 /* Start the UST consumer daemons */
2784 /* 64-bit */
2785 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2786 if (consumerd64_bin[0] != '\0' &&
2787 ustconsumer64_data.pid == 0 &&
2788 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2789 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2790 ret = start_consumerd(&ustconsumer64_data);
2791 if (ret < 0) {
2792 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2793 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2794 goto error;
2795 }
2796
2797 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2798 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2799 } else {
2800 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2801 }
2802
2803 /*
2804 * Setup socket for consumer 64 bit. No need for atomic access
2805 * since it was set above and can ONLY be set in this thread.
2806 */
2807 ret = consumer_create_socket(&ustconsumer64_data,
2808 cmd_ctx->session->ust_session->consumer);
2809 if (ret < 0) {
2810 goto error;
2811 }
2812
2813 /* 32-bit */
2814 if (consumerd32_bin[0] != '\0' &&
2815 ustconsumer32_data.pid == 0 &&
2816 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2817 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2818 ret = start_consumerd(&ustconsumer32_data);
2819 if (ret < 0) {
2820 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2821 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2822 goto error;
2823 }
2824
2825 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2826 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2827 } else {
2828 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2829 }
2830
2831 /*
2832 * Setup socket for consumer 64 bit. No need for atomic access
2833 * since it was set above and can ONLY be set in this thread.
2834 */
2835 ret = consumer_create_socket(&ustconsumer32_data,
2836 cmd_ctx->session->ust_session->consumer);
2837 if (ret < 0) {
2838 goto error;
2839 }
2840 }
2841 break;
2842 }
2843 default:
2844 break;
2845 }
2846 skip_domain:
2847
2848 /* Validate consumer daemon state when start/stop trace command */
2849 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2850 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2851 switch (cmd_ctx->lsm->domain.type) {
2852 case LTTNG_DOMAIN_JUL:
2853 case LTTNG_DOMAIN_UST:
2854 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2855 ret = LTTNG_ERR_NO_USTCONSUMERD;
2856 goto error;
2857 }
2858 break;
2859 case LTTNG_DOMAIN_KERNEL:
2860 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2861 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2862 goto error;
2863 }
2864 break;
2865 }
2866 }
2867
2868 /*
2869 * Check that the UID or GID match that of the tracing session.
2870 * The root user can interact with all sessions.
2871 */
2872 if (need_tracing_session) {
2873 if (!session_access_ok(cmd_ctx->session,
2874 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2875 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2876 ret = LTTNG_ERR_EPERM;
2877 goto error;
2878 }
2879 }
2880
2881 /*
2882 * Send relayd information to consumer as soon as we have a domain and a
2883 * session defined.
2884 */
2885 if (cmd_ctx->session && need_domain) {
2886 /*
2887 * Setup relayd if not done yet. If the relayd information was already
2888 * sent to the consumer, this call will gracefully return.
2889 */
2890 ret = cmd_setup_relayd(cmd_ctx->session);
2891 if (ret != LTTNG_OK) {
2892 goto error;
2893 }
2894 }
2895
2896 /* Process by command type */
2897 switch (cmd_ctx->lsm->cmd_type) {
2898 case LTTNG_ADD_CONTEXT:
2899 {
2900 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2901 cmd_ctx->lsm->u.context.channel_name,
2902 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2903 break;
2904 }
2905 case LTTNG_DISABLE_CHANNEL:
2906 {
2907 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2908 cmd_ctx->lsm->u.disable.channel_name);
2909 break;
2910 }
2911 case LTTNG_DISABLE_EVENT:
2912 {
2913 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2914 cmd_ctx->lsm->u.disable.channel_name,
2915 cmd_ctx->lsm->u.disable.name);
2916 break;
2917 }
2918 case LTTNG_DISABLE_ALL_EVENT:
2919 {
2920 DBG("Disabling all events");
2921
2922 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2923 cmd_ctx->lsm->u.disable.channel_name);
2924 break;
2925 }
2926 case LTTNG_ENABLE_CHANNEL:
2927 {
2928 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2929 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2930 break;
2931 }
2932 case LTTNG_ENABLE_EVENT:
2933 {
2934 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2935 cmd_ctx->lsm->u.enable.channel_name,
2936 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2937 break;
2938 }
2939 case LTTNG_ENABLE_ALL_EVENT:
2940 {
2941 DBG("Enabling all events");
2942
2943 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2944 cmd_ctx->lsm->u.enable.channel_name,
2945 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2946 break;
2947 }
2948 case LTTNG_LIST_TRACEPOINTS:
2949 {
2950 struct lttng_event *events;
2951 ssize_t nb_events;
2952
2953 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2954 if (nb_events < 0) {
2955 /* Return value is a negative lttng_error_code. */
2956 ret = -nb_events;
2957 goto error;
2958 }
2959
2960 /*
2961 * Setup lttng message with payload size set to the event list size in
2962 * bytes and then copy list into the llm payload.
2963 */
2964 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2965 if (ret < 0) {
2966 free(events);
2967 goto setup_error;
2968 }
2969
2970 /* Copy event list into message payload */
2971 memcpy(cmd_ctx->llm->payload, events,
2972 sizeof(struct lttng_event) * nb_events);
2973
2974 free(events);
2975
2976 ret = LTTNG_OK;
2977 break;
2978 }
2979 case LTTNG_LIST_TRACEPOINT_FIELDS:
2980 {
2981 struct lttng_event_field *fields;
2982 ssize_t nb_fields;
2983
2984 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2985 &fields);
2986 if (nb_fields < 0) {
2987 /* Return value is a negative lttng_error_code. */
2988 ret = -nb_fields;
2989 goto error;
2990 }
2991
2992 /*
2993 * Setup lttng message with payload size set to the event list size in
2994 * bytes and then copy list into the llm payload.
2995 */
2996 ret = setup_lttng_msg(cmd_ctx,
2997 sizeof(struct lttng_event_field) * nb_fields);
2998 if (ret < 0) {
2999 free(fields);
3000 goto setup_error;
3001 }
3002
3003 /* Copy event list into message payload */
3004 memcpy(cmd_ctx->llm->payload, fields,
3005 sizeof(struct lttng_event_field) * nb_fields);
3006
3007 free(fields);
3008
3009 ret = LTTNG_OK;
3010 break;
3011 }
3012 case LTTNG_SET_CONSUMER_URI:
3013 {
3014 size_t nb_uri, len;
3015 struct lttng_uri *uris;
3016
3017 nb_uri = cmd_ctx->lsm->u.uri.size;
3018 len = nb_uri * sizeof(struct lttng_uri);
3019
3020 if (nb_uri == 0) {
3021 ret = LTTNG_ERR_INVALID;
3022 goto error;
3023 }
3024
3025 uris = zmalloc(len);
3026 if (uris == NULL) {
3027 ret = LTTNG_ERR_FATAL;
3028 goto error;
3029 }
3030
3031 /* Receive variable len data */
3032 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3033 ret = lttcomm_recv_unix_sock(sock, uris, len);
3034 if (ret <= 0) {
3035 DBG("No URIs received from client... continuing");
3036 *sock_error = 1;
3037 ret = LTTNG_ERR_SESSION_FAIL;
3038 free(uris);
3039 goto error;
3040 }
3041
3042 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3043 nb_uri, uris);
3044 if (ret != LTTNG_OK) {
3045 free(uris);
3046 goto error;
3047 }
3048
3049 /*
3050 * XXX: 0 means that this URI should be applied on the session. Should
3051 * be a DOMAIN enuam.
3052 */
3053 if (cmd_ctx->lsm->domain.type == 0) {
3054 /* Add the URI for the UST session if a consumer is present. */
3055 if (cmd_ctx->session->ust_session &&
3056 cmd_ctx->session->ust_session->consumer) {
3057 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3058 nb_uri, uris);
3059 } else if (cmd_ctx->session->kernel_session &&
3060 cmd_ctx->session->kernel_session->consumer) {
3061 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3062 cmd_ctx->session, nb_uri, uris);
3063 }
3064 }
3065
3066 free(uris);
3067
3068 break;
3069 }
3070 case LTTNG_START_TRACE:
3071 {
3072 ret = cmd_start_trace(cmd_ctx->session);
3073 break;
3074 }
3075 case LTTNG_STOP_TRACE:
3076 {
3077 ret = cmd_stop_trace(cmd_ctx->session);
3078 break;
3079 }
3080 case LTTNG_CREATE_SESSION:
3081 {
3082 size_t nb_uri, len;
3083 struct lttng_uri *uris = NULL;
3084
3085 nb_uri = cmd_ctx->lsm->u.uri.size;
3086 len = nb_uri * sizeof(struct lttng_uri);
3087
3088 if (nb_uri > 0) {
3089 uris = zmalloc(len);
3090 if (uris == NULL) {
3091 ret = LTTNG_ERR_FATAL;
3092 goto error;
3093 }
3094
3095 /* Receive variable len data */
3096 DBG("Waiting for %zu URIs from client ...", nb_uri);
3097 ret = lttcomm_recv_unix_sock(sock, uris, len);
3098 if (ret <= 0) {
3099 DBG("No URIs received from client... continuing");
3100 *sock_error = 1;
3101 ret = LTTNG_ERR_SESSION_FAIL;
3102 free(uris);
3103 goto error;
3104 }
3105
3106 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3107 DBG("Creating session with ONE network URI is a bad call");
3108 ret = LTTNG_ERR_SESSION_FAIL;
3109 free(uris);
3110 goto error;
3111 }
3112 }
3113
3114 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3115 &cmd_ctx->creds, 0);
3116
3117 free(uris);
3118
3119 break;
3120 }
3121 case LTTNG_DESTROY_SESSION:
3122 {
3123 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3124
3125 /* Set session to NULL so we do not unlock it after free. */
3126 cmd_ctx->session = NULL;
3127 break;
3128 }
3129 case LTTNG_LIST_DOMAINS:
3130 {
3131 ssize_t nb_dom;
3132 struct lttng_domain *domains;
3133
3134 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3135 if (nb_dom < 0) {
3136 /* Return value is a negative lttng_error_code. */
3137 ret = -nb_dom;
3138 goto error;
3139 }
3140
3141 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3142 if (ret < 0) {
3143 free(domains);
3144 goto setup_error;
3145 }
3146
3147 /* Copy event list into message payload */
3148 memcpy(cmd_ctx->llm->payload, domains,
3149 nb_dom * sizeof(struct lttng_domain));
3150
3151 free(domains);
3152
3153 ret = LTTNG_OK;
3154 break;
3155 }
3156 case LTTNG_LIST_CHANNELS:
3157 {
3158 int nb_chan;
3159 struct lttng_channel *channels;
3160
3161 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3162 cmd_ctx->session, &channels);
3163 if (nb_chan < 0) {
3164 /* Return value is a negative lttng_error_code. */
3165 ret = -nb_chan;
3166 goto error;
3167 }
3168
3169 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3170 if (ret < 0) {
3171 free(channels);
3172 goto setup_error;
3173 }
3174
3175 /* Copy event list into message payload */
3176 memcpy(cmd_ctx->llm->payload, channels,
3177 nb_chan * sizeof(struct lttng_channel));
3178
3179 free(channels);
3180
3181 ret = LTTNG_OK;
3182 break;
3183 }
3184 case LTTNG_LIST_EVENTS:
3185 {
3186 ssize_t nb_event;
3187 struct lttng_event *events = NULL;
3188
3189 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3190 cmd_ctx->lsm->u.list.channel_name, &events);
3191 if (nb_event < 0) {
3192 /* Return value is a negative lttng_error_code. */
3193 ret = -nb_event;
3194 goto error;
3195 }
3196
3197 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3198 if (ret < 0) {
3199 free(events);
3200 goto setup_error;
3201 }
3202
3203 /* Copy event list into message payload */
3204 memcpy(cmd_ctx->llm->payload, events,
3205 nb_event * sizeof(struct lttng_event));
3206
3207 free(events);
3208
3209 ret = LTTNG_OK;
3210 break;
3211 }
3212 case LTTNG_LIST_SESSIONS:
3213 {
3214 unsigned int nr_sessions;
3215
3216 session_lock_list();
3217 nr_sessions = lttng_sessions_count(
3218 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3219 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3220
3221 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3222 if (ret < 0) {
3223 session_unlock_list();
3224 goto setup_error;
3225 }
3226
3227 /* Filled the session array */
3228 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3229 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3230 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3231
3232 session_unlock_list();
3233
3234 ret = LTTNG_OK;
3235 break;
3236 }
3237 case LTTNG_CALIBRATE:
3238 {
3239 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3240 &cmd_ctx->lsm->u.calibrate);
3241 break;
3242 }
3243 case LTTNG_REGISTER_CONSUMER:
3244 {
3245 struct consumer_data *cdata;
3246
3247 switch (cmd_ctx->lsm->domain.type) {
3248 case LTTNG_DOMAIN_KERNEL:
3249 cdata = &kconsumer_data;
3250 break;
3251 default:
3252 ret = LTTNG_ERR_UND;
3253 goto error;
3254 }
3255
3256 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3257 cmd_ctx->lsm->u.reg.path, cdata);
3258 break;
3259 }
3260 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3261 {
3262 struct lttng_filter_bytecode *bytecode;
3263
3264 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3265 ret = LTTNG_ERR_FILTER_INVAL;
3266 goto error;
3267 }
3268 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3269 ret = LTTNG_ERR_FILTER_INVAL;
3270 goto error;
3271 }
3272 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3273 if (!bytecode) {
3274 ret = LTTNG_ERR_FILTER_NOMEM;
3275 goto error;
3276 }
3277 /* Receive var. len. data */
3278 DBG("Receiving var len data from client ...");
3279 ret = lttcomm_recv_unix_sock(sock, bytecode,
3280 cmd_ctx->lsm->u.enable.bytecode_len);
3281 if (ret <= 0) {
3282 DBG("Nothing recv() from client var len data... continuing");
3283 *sock_error = 1;
3284 ret = LTTNG_ERR_FILTER_INVAL;
3285 goto error;
3286 }
3287
3288 if (bytecode->len + sizeof(*bytecode)
3289 != cmd_ctx->lsm->u.enable.bytecode_len) {
3290 free(bytecode);
3291 ret = LTTNG_ERR_FILTER_INVAL;
3292 goto error;
3293 }
3294
3295 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3296 cmd_ctx->lsm->u.enable.channel_name,
3297 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3298 break;
3299 }
3300 case LTTNG_DATA_PENDING:
3301 {
3302 ret = cmd_data_pending(cmd_ctx->session);
3303 break;
3304 }
3305 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3306 {
3307 struct lttcomm_lttng_output_id reply;
3308
3309 ret = cmd_snapshot_add_output(cmd_ctx->session,
3310 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3311 if (ret != LTTNG_OK) {
3312 goto error;
3313 }
3314
3315 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3316 if (ret < 0) {
3317 goto setup_error;
3318 }
3319
3320 /* Copy output list into message payload */
3321 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3322 ret = LTTNG_OK;
3323 break;
3324 }
3325 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3326 {
3327 ret = cmd_snapshot_del_output(cmd_ctx->session,
3328 &cmd_ctx->lsm->u.snapshot_output.output);
3329 break;
3330 }
3331 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3332 {
3333 ssize_t nb_output;
3334 struct lttng_snapshot_output *outputs = NULL;
3335
3336 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3337 if (nb_output < 0) {
3338 ret = -nb_output;
3339 goto error;
3340 }
3341
3342 ret = setup_lttng_msg(cmd_ctx,
3343 nb_output * sizeof(struct lttng_snapshot_output));
3344 if (ret < 0) {
3345 free(outputs);
3346 goto setup_error;
3347 }
3348
3349 if (outputs) {
3350 /* Copy output list into message payload */
3351 memcpy(cmd_ctx->llm->payload, outputs,
3352 nb_output * sizeof(struct lttng_snapshot_output));
3353 free(outputs);
3354 }
3355
3356 ret = LTTNG_OK;
3357 break;
3358 }
3359 case LTTNG_SNAPSHOT_RECORD:
3360 {
3361 ret = cmd_snapshot_record(cmd_ctx->session,
3362 &cmd_ctx->lsm->u.snapshot_record.output,
3363 cmd_ctx->lsm->u.snapshot_record.wait);
3364 break;
3365 }
3366 case LTTNG_CREATE_SESSION_SNAPSHOT:
3367 {
3368 size_t nb_uri, len;
3369 struct lttng_uri *uris = NULL;
3370
3371 nb_uri = cmd_ctx->lsm->u.uri.size;
3372 len = nb_uri * sizeof(struct lttng_uri);
3373
3374 if (nb_uri > 0) {
3375 uris = zmalloc(len);
3376 if (uris == NULL) {
3377 ret = LTTNG_ERR_FATAL;
3378 goto error;
3379 }
3380
3381 /* Receive variable len data */
3382 DBG("Waiting for %zu URIs from client ...", nb_uri);
3383 ret = lttcomm_recv_unix_sock(sock, uris, len);
3384 if (ret <= 0) {
3385 DBG("No URIs received from client... continuing");
3386 *sock_error = 1;
3387 ret = LTTNG_ERR_SESSION_FAIL;
3388 free(uris);
3389 goto error;
3390 }
3391
3392 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3393 DBG("Creating session with ONE network URI is a bad call");
3394 ret = LTTNG_ERR_SESSION_FAIL;
3395 free(uris);
3396 goto error;
3397 }
3398 }
3399
3400 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3401 nb_uri, &cmd_ctx->creds);
3402 free(uris);
3403 break;
3404 }
3405 case LTTNG_CREATE_SESSION_LIVE:
3406 {
3407 size_t nb_uri, len;
3408 struct lttng_uri *uris = NULL;
3409
3410 nb_uri = cmd_ctx->lsm->u.uri.size;
3411 len = nb_uri * sizeof(struct lttng_uri);
3412
3413 if (nb_uri > 0) {
3414 uris = zmalloc(len);
3415 if (uris == NULL) {
3416 ret = LTTNG_ERR_FATAL;
3417 goto error;
3418 }
3419
3420 /* Receive variable len data */
3421 DBG("Waiting for %zu URIs from client ...", nb_uri);
3422 ret = lttcomm_recv_unix_sock(sock, uris, len);
3423 if (ret <= 0) {
3424 DBG("No URIs received from client... continuing");
3425 *sock_error = 1;
3426 ret = LTTNG_ERR_SESSION_FAIL;
3427 free(uris);
3428 goto error;
3429 }
3430
3431 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3432 DBG("Creating session with ONE network URI is a bad call");
3433 ret = LTTNG_ERR_SESSION_FAIL;
3434 free(uris);
3435 goto error;
3436 }
3437 }
3438
3439 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3440 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3441 free(uris);
3442 break;
3443 }
3444 default:
3445 ret = LTTNG_ERR_UND;
3446 break;
3447 }
3448
3449 error:
3450 if (cmd_ctx->llm == NULL) {
3451 DBG("Missing llm structure. Allocating one.");
3452 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3453 goto setup_error;
3454 }
3455 }
3456 /* Set return code */
3457 cmd_ctx->llm->ret_code = ret;
3458 setup_error:
3459 if (cmd_ctx->session) {
3460 session_unlock(cmd_ctx->session);
3461 }
3462 if (need_tracing_session) {
3463 session_unlock_list();
3464 }
3465 init_setup_error:
3466 return ret;
3467 }
3468
3469 /*
3470 * Thread managing health check socket.
3471 */
3472 static void *thread_manage_health(void *data)
3473 {
3474 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3475 uint32_t revents, nb_fd;
3476 struct lttng_poll_event events;
3477 struct health_comm_msg msg;
3478 struct health_comm_reply reply;
3479
3480 DBG("[thread] Manage health check started");
3481
3482 rcu_register_thread();
3483
3484 /* We might hit an error path before this is created. */
3485 lttng_poll_init(&events);
3486
3487 /* Create unix socket */
3488 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3489 if (sock < 0) {
3490 ERR("Unable to create health check Unix socket");
3491 ret = -1;
3492 goto error;
3493 }
3494
3495 if (is_root) {
3496 /* lttng health client socket path permissions */
3497 ret = chown(health_unix_sock_path, 0,
3498 utils_get_group_id(tracing_group_name));
3499 if (ret < 0) {
3500 ERR("Unable to set group on %s", health_unix_sock_path);
3501 PERROR("chown");
3502 ret = -1;
3503 goto error;
3504 }
3505
3506 ret = chmod(health_unix_sock_path,
3507 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3508 if (ret < 0) {
3509 ERR("Unable to set permissions on %s", health_unix_sock_path);
3510 PERROR("chmod");
3511 ret = -1;
3512 goto error;
3513 }
3514 }
3515
3516 /*
3517 * Set the CLOEXEC flag. Return code is useless because either way, the
3518 * show must go on.
3519 */
3520 (void) utils_set_fd_cloexec(sock);
3521
3522 ret = lttcomm_listen_unix_sock(sock);
3523 if (ret < 0) {
3524 goto error;
3525 }
3526
3527 /*
3528 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3529 * more will be added to this poll set.
3530 */
3531 ret = sessiond_set_thread_pollset(&events, 2);
3532 if (ret < 0) {
3533 goto error;
3534 }
3535
3536 /* Add the application registration socket */
3537 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3538 if (ret < 0) {
3539 goto error;
3540 }
3541
3542 while (1) {
3543 DBG("Health check ready");
3544
3545 /* Inifinite blocking call, waiting for transmission */
3546 restart:
3547 ret = lttng_poll_wait(&events, -1);
3548 if (ret < 0) {
3549 /*
3550 * Restart interrupted system call.
3551 */
3552 if (errno == EINTR) {
3553 goto restart;
3554 }
3555 goto error;
3556 }
3557
3558 nb_fd = ret;
3559
3560 for (i = 0; i < nb_fd; i++) {
3561 /* Fetch once the poll data */
3562 revents = LTTNG_POLL_GETEV(&events, i);
3563 pollfd = LTTNG_POLL_GETFD(&events, i);
3564
3565 /* Thread quit pipe has been closed. Killing thread. */
3566 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3567 if (ret) {
3568 err = 0;
3569 goto exit;
3570 }
3571
3572 /* Event on the registration socket */
3573 if (pollfd == sock) {
3574 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3575 ERR("Health socket poll error");
3576 goto error;
3577 }
3578 }
3579 }
3580
3581 new_sock = lttcomm_accept_unix_sock(sock);
3582 if (new_sock < 0) {
3583 goto error;
3584 }
3585
3586 /*
3587 * Set the CLOEXEC flag. Return code is useless because either way, the
3588 * show must go on.
3589 */
3590 (void) utils_set_fd_cloexec(new_sock);
3591
3592 DBG("Receiving data from client for health...");
3593 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3594 if (ret <= 0) {
3595 DBG("Nothing recv() from client... continuing");
3596 ret = close(new_sock);
3597 if (ret) {
3598 PERROR("close");
3599 }
3600 new_sock = -1;
3601 continue;
3602 }
3603
3604 rcu_thread_online();
3605
3606 reply.ret_code = 0;
3607 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3608 /*
3609 * health_check_state returns 0 if health is
3610 * bad.
3611 */
3612 if (!health_check_state(health_sessiond, i)) {
3613 reply.ret_code |= 1ULL << i;
3614 }
3615 }
3616
3617 DBG2("Health check return value %" PRIx64, reply.ret_code);
3618
3619 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3620 if (ret < 0) {
3621 ERR("Failed to send health data back to client");
3622 }
3623
3624 /* End of transmission */
3625 ret = close(new_sock);
3626 if (ret) {
3627 PERROR("close");
3628 }
3629 new_sock = -1;
3630 }
3631
3632 exit:
3633 error:
3634 if (err) {
3635 ERR("Health error occurred in %s", __func__);
3636 }
3637 DBG("Health check thread dying");
3638 unlink(health_unix_sock_path);
3639 if (sock >= 0) {
3640 ret = close(sock);
3641 if (ret) {
3642 PERROR("close");
3643 }
3644 }
3645
3646 lttng_poll_clean(&events);
3647
3648 rcu_unregister_thread();
3649 return NULL;
3650 }
3651
3652 /*
3653 * This thread manage all clients request using the unix client socket for
3654 * communication.
3655 */
3656 static void *thread_manage_clients(void *data)
3657 {
3658 int sock = -1, ret, i, pollfd, err = -1;
3659 int sock_error;
3660 uint32_t revents, nb_fd;
3661 struct command_ctx *cmd_ctx = NULL;
3662 struct lttng_poll_event events;
3663
3664 DBG("[thread] Manage client started");
3665
3666 rcu_register_thread();
3667
3668 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3669
3670 if (testpoint(thread_manage_clients)) {
3671 goto error_testpoint;
3672 }
3673
3674 health_code_update();
3675
3676 ret = lttcomm_listen_unix_sock(client_sock);
3677 if (ret < 0) {
3678 goto error_listen;
3679 }
3680
3681 /*
3682 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3683 * more will be added to this poll set.
3684 */
3685 ret = sessiond_set_thread_pollset(&events, 2);
3686 if (ret < 0) {
3687 goto error_create_poll;
3688 }
3689
3690 /* Add the application registration socket */
3691 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3692 if (ret < 0) {
3693 goto error;
3694 }
3695
3696 /*
3697 * Notify parent pid that we are ready to accept command for client side.
3698 */
3699 if (opt_sig_parent) {
3700 kill(ppid, SIGUSR1);
3701 }
3702
3703 if (testpoint(thread_manage_clients_before_loop)) {
3704 goto error;
3705 }
3706
3707 health_code_update();
3708
3709 while (1) {
3710 DBG("Accepting client command ...");
3711
3712 /* Inifinite blocking call, waiting for transmission */
3713 restart:
3714 health_poll_entry();
3715 ret = lttng_poll_wait(&events, -1);
3716 health_poll_exit();
3717 if (ret < 0) {
3718 /*
3719 * Restart interrupted system call.
3720 */
3721 if (errno == EINTR) {
3722 goto restart;
3723 }
3724 goto error;
3725 }
3726
3727 nb_fd = ret;
3728
3729 for (i = 0; i < nb_fd; i++) {
3730 /* Fetch once the poll data */
3731 revents = LTTNG_POLL_GETEV(&events, i);
3732 pollfd = LTTNG_POLL_GETFD(&events, i);
3733
3734 health_code_update();
3735
3736 /* Thread quit pipe has been closed. Killing thread. */
3737 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3738 if (ret) {
3739 err = 0;
3740 goto exit;
3741 }
3742
3743 /* Event on the registration socket */
3744 if (pollfd == client_sock) {
3745 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3746 ERR("Client socket poll error");
3747 goto error;
3748 }
3749 }
3750 }
3751
3752 DBG("Wait for client response");
3753
3754 health_code_update();
3755
3756 sock = lttcomm_accept_unix_sock(client_sock);
3757 if (sock < 0) {
3758 goto error;
3759 }
3760
3761 /*
3762 * Set the CLOEXEC flag. Return code is useless because either way, the
3763 * show must go on.
3764 */
3765 (void) utils_set_fd_cloexec(sock);
3766
3767 /* Set socket option for credentials retrieval */
3768 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3769 if (ret < 0) {
3770 goto error;
3771 }
3772
3773 /* Allocate context command to process the client request */
3774 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3775 if (cmd_ctx == NULL) {
3776 PERROR("zmalloc cmd_ctx");
3777 goto error;
3778 }
3779
3780 /* Allocate data buffer for reception */
3781 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3782 if (cmd_ctx->lsm == NULL) {
3783 PERROR("zmalloc cmd_ctx->lsm");
3784 goto error;
3785 }
3786
3787 cmd_ctx->llm = NULL;
3788 cmd_ctx->session = NULL;
3789
3790 health_code_update();
3791
3792 /*
3793 * Data is received from the lttng client. The struct
3794 * lttcomm_session_msg (lsm) contains the command and data request of
3795 * the client.
3796 */
3797 DBG("Receiving data from client ...");
3798 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3799 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3800 if (ret <= 0) {
3801 DBG("Nothing recv() from client... continuing");
3802 ret = close(sock);
3803 if (ret) {
3804 PERROR("close");
3805 }
3806 sock = -1;
3807 clean_command_ctx(&cmd_ctx);
3808 continue;
3809 }
3810
3811 health_code_update();
3812
3813 // TODO: Validate cmd_ctx including sanity check for
3814 // security purpose.
3815
3816 rcu_thread_online();
3817 /*
3818 * This function dispatch the work to the kernel or userspace tracer
3819 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3820 * informations for the client. The command context struct contains
3821 * everything this function may needs.
3822 */
3823 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3824 rcu_thread_offline();
3825 if (ret < 0) {
3826 ret = close(sock);
3827 if (ret) {
3828 PERROR("close");
3829 }
3830 sock = -1;
3831 /*
3832 * TODO: Inform client somehow of the fatal error. At
3833 * this point, ret < 0 means that a zmalloc failed
3834 * (ENOMEM). Error detected but still accept
3835 * command, unless a socket error has been
3836 * detected.
3837 */
3838 clean_command_ctx(&cmd_ctx);
3839 continue;
3840 }
3841
3842 health_code_update();
3843
3844 DBG("Sending response (size: %d, retcode: %s)",
3845 cmd_ctx->lttng_msg_size,
3846 lttng_strerror(-cmd_ctx->llm->ret_code));
3847 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3848 if (ret < 0) {
3849 ERR("Failed to send data back to client");
3850 }
3851
3852 /* End of transmission */
3853 ret = close(sock);
3854 if (ret) {
3855 PERROR("close");
3856 }
3857 sock = -1;
3858
3859 clean_command_ctx(&cmd_ctx);
3860
3861 health_code_update();
3862 }
3863
3864 exit:
3865 error:
3866 if (sock >= 0) {
3867 ret = close(sock);
3868 if (ret) {
3869 PERROR("close");
3870 }
3871 }
3872
3873 lttng_poll_clean(&events);
3874 clean_command_ctx(&cmd_ctx);
3875
3876 error_listen:
3877 error_create_poll:
3878 error_testpoint:
3879 unlink(client_unix_sock_path);
3880 if (client_sock >= 0) {
3881 ret = close(client_sock);
3882 if (ret) {
3883 PERROR("close");
3884 }
3885 }
3886
3887 if (err) {
3888 health_error();
3889 ERR("Health error occurred in %s", __func__);
3890 }
3891
3892 health_unregister(health_sessiond);
3893
3894 DBG("Client thread dying");
3895
3896 rcu_unregister_thread();
3897 return NULL;
3898 }
3899
3900
3901 /*
3902 * usage function on stderr
3903 */
3904 static void usage(void)
3905 {
3906 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3907 fprintf(stderr, " -h, --help Display this usage.\n");
3908 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3909 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3910 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3911 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3912 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3913 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3914 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3915 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3916 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3917 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3918 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3919 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3920 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3921 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3922 fprintf(stderr, " -V, --version Show version number.\n");
3923 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3924 fprintf(stderr, " -q, --quiet No output at all.\n");
3925 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3926 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3927 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3928 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3929 fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
3930 }
3931
3932 /*
3933 * daemon argument parsing
3934 */
3935 static int parse_args(int argc, char **argv)
3936 {
3937 int c;
3938
3939 static struct option long_options[] = {
3940 { "client-sock", 1, 0, 'c' },
3941 { "apps-sock", 1, 0, 'a' },
3942 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3943 { "kconsumerd-err-sock", 1, 0, 'E' },
3944 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3945 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3946 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3947 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3948 { "consumerd32-path", 1, 0, 'u' },
3949 { "consumerd32-libdir", 1, 0, 'U' },
3950 { "consumerd64-path", 1, 0, 't' },
3951 { "consumerd64-libdir", 1, 0, 'T' },
3952 { "daemonize", 0, 0, 'd' },
3953 { "sig-parent", 0, 0, 'S' },
3954 { "help", 0, 0, 'h' },
3955 { "group", 1, 0, 'g' },
3956 { "version", 0, 0, 'V' },
3957 { "quiet", 0, 0, 'q' },
3958 { "verbose", 0, 0, 'v' },
3959 { "verbose-consumer", 0, 0, 'Z' },
3960 { "no-kernel", 0, 0, 'N' },
3961 { "pidfile", 1, 0, 'p' },
3962 { "jul-tcp-port", 1, 0, 'J' },
3963 { NULL, 0, 0, 0 }
3964 };
3965
3966 while (1) {
3967 int option_index = 0;
3968 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
3969 long_options, &option_index);
3970 if (c == -1) {
3971 break;
3972 }
3973
3974 switch (c) {
3975 case 0:
3976 fprintf(stderr, "option %s", long_options[option_index].name);
3977 if (optarg) {
3978 fprintf(stderr, " with arg %s\n", optarg);
3979 }
3980 break;
3981 case 'c':
3982 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3983 break;
3984 case 'a':
3985 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3986 break;
3987 case 'd':
3988 opt_daemon = 1;
3989 break;
3990 case 'g':
3991 tracing_group_name = optarg;
3992 break;
3993 case 'h':
3994 usage();
3995 exit(EXIT_FAILURE);
3996 case 'V':
3997 fprintf(stdout, "%s\n", VERSION);
3998 exit(EXIT_SUCCESS);
3999 case 'S':
4000 opt_sig_parent = 1;
4001 break;
4002 case 'E':
4003 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4004 break;
4005 case 'C':
4006 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4007 break;
4008 case 'F':
4009 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4010 break;
4011 case 'D':
4012 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4013 break;
4014 case 'H':
4015 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4016 break;
4017 case 'G':
4018 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4019 break;
4020 case 'N':
4021 opt_no_kernel = 1;
4022 break;
4023 case 'q':
4024 lttng_opt_quiet = 1;
4025 break;
4026 case 'v':
4027 /* Verbose level can increase using multiple -v */
4028 lttng_opt_verbose += 1;
4029 break;
4030 case 'Z':
4031 opt_verbose_consumer += 1;
4032 break;
4033 case 'u':
4034 consumerd32_bin= optarg;
4035 break;
4036 case 'U':
4037 consumerd32_libdir = optarg;
4038 break;
4039 case 't':
4040 consumerd64_bin = optarg;
4041 break;
4042 case 'T':
4043 consumerd64_libdir = optarg;
4044 break;
4045 case 'p':
4046 opt_pidfile = optarg;
4047 break;
4048 case 'J': /* JUL TCP port. */
4049 {
4050 unsigned long v;
4051
4052 errno = 0;
4053 v = strtoul(optarg, NULL, 0);
4054 if (errno != 0 || !isdigit(optarg[0])) {
4055 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg);
4056 return -1;
4057 }
4058 if (v == 0 || v >= 65535) {
4059 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg);
4060 return -1;
4061 }
4062 jul_tcp_port = (uint32_t) v;
4063 DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
4064 break;
4065 }
4066 default:
4067 /* Unknown option or other error.
4068 * Error is printed by getopt, just return */
4069 return -1;
4070 }
4071 }
4072
4073 return 0;
4074 }
4075
4076 /*
4077 * Creates the two needed socket by the daemon.
4078 * apps_sock - The communication socket for all UST apps.
4079 * client_sock - The communication of the cli tool (lttng).
4080 */
4081 static int init_daemon_socket(void)
4082 {
4083 int ret = 0;
4084 mode_t old_umask;
4085
4086 old_umask = umask(0);
4087
4088 /* Create client tool unix socket */
4089 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4090 if (client_sock < 0) {
4091 ERR("Create unix sock failed: %s", client_unix_sock_path);
4092 ret = -1;
4093 goto end;
4094 }
4095
4096 /* Set the cloexec flag */
4097 ret = utils_set_fd_cloexec(client_sock);
4098 if (ret < 0) {
4099 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4100 "Continuing but note that the consumer daemon will have a "
4101 "reference to this socket on exec()", client_sock);
4102 }
4103
4104 /* File permission MUST be 660 */
4105 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4106 if (ret < 0) {
4107 ERR("Set file permissions failed: %s", client_unix_sock_path);
4108 PERROR("chmod");
4109 goto end;
4110 }
4111
4112 /* Create the application unix socket */
4113 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4114 if (apps_sock < 0) {
4115 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4116 ret = -1;
4117 goto end;
4118 }
4119
4120 /* Set the cloexec flag */
4121 ret = utils_set_fd_cloexec(apps_sock);
4122 if (ret < 0) {
4123 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4124 "Continuing but note that the consumer daemon will have a "
4125 "reference to this socket on exec()", apps_sock);
4126 }
4127
4128 /* File permission MUST be 666 */
4129 ret = chmod(apps_unix_sock_path,
4130 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4131 if (ret < 0) {
4132 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4133 PERROR("chmod");
4134 goto end;
4135 }
4136
4137 DBG3("Session daemon client socket %d and application socket %d created",
4138 client_sock, apps_sock);
4139
4140 end:
4141 umask(old_umask);
4142 return ret;
4143 }
4144
4145 /*
4146 * Check if the global socket is available, and if a daemon is answering at the
4147 * other side. If yes, error is returned.
4148 */
4149 static int check_existing_daemon(void)
4150 {
4151 /* Is there anybody out there ? */
4152 if (lttng_session_daemon_alive()) {
4153 return -EEXIST;
4154 }
4155
4156 return 0;
4157 }
4158
4159 /*
4160 * Set the tracing group gid onto the client socket.
4161 *
4162 * Race window between mkdir and chown is OK because we are going from more
4163 * permissive (root.root) to less permissive (root.tracing).
4164 */
4165 static int set_permissions(char *rundir)
4166 {
4167 int ret;
4168 gid_t gid;
4169
4170 gid = utils_get_group_id(tracing_group_name);
4171
4172 /* Set lttng run dir */
4173 ret = chown(rundir, 0, gid);
4174 if (ret < 0) {
4175 ERR("Unable to set group on %s", rundir);
4176 PERROR("chown");
4177 }
4178
4179 /*
4180 * Ensure all applications and tracing group can search the run
4181 * dir. Allow everyone to read the directory, since it does not
4182 * buy us anything to hide its content.
4183 */
4184 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4185 if (ret < 0) {
4186 ERR("Unable to set permissions on %s", rundir);
4187 PERROR("chmod");
4188 }
4189
4190 /* lttng client socket path */
4191 ret = chown(client_unix_sock_path, 0, gid);
4192 if (ret < 0) {
4193 ERR("Unable to set group on %s", client_unix_sock_path);
4194 PERROR("chown");
4195 }
4196
4197 /* kconsumer error socket path */
4198 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4199 if (ret < 0) {
4200 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4201 PERROR("chown");
4202 }
4203
4204 /* 64-bit ustconsumer error socket path */
4205 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4206 if (ret < 0) {
4207 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4208 PERROR("chown");
4209 }
4210
4211 /* 32-bit ustconsumer compat32 error socket path */
4212 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4213 if (ret < 0) {
4214 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4215 PERROR("chown");
4216 }
4217
4218 DBG("All permissions are set");
4219
4220 return ret;
4221 }
4222
4223 /*
4224 * Create the lttng run directory needed for all global sockets and pipe.
4225 */
4226 static int create_lttng_rundir(const char *rundir)
4227 {
4228 int ret;
4229
4230 DBG3("Creating LTTng run directory: %s", rundir);
4231
4232 ret = mkdir(rundir, S_IRWXU);
4233 if (ret < 0) {
4234 if (errno != EEXIST) {
4235 ERR("Unable to create %s", rundir);
4236 goto error;
4237 } else {
4238 ret = 0;
4239 }
4240 }
4241
4242 error:
4243 return ret;
4244 }
4245
4246 /*
4247 * Setup sockets and directory needed by the kconsumerd communication with the
4248 * session daemon.
4249 */
4250 static int set_consumer_sockets(struct consumer_data *consumer_data,
4251 const char *rundir)
4252 {
4253 int ret;
4254 char path[PATH_MAX];
4255
4256 switch (consumer_data->type) {
4257 case LTTNG_CONSUMER_KERNEL:
4258 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4259 break;
4260 case LTTNG_CONSUMER64_UST:
4261 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4262 break;
4263 case LTTNG_CONSUMER32_UST:
4264 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4265 break;
4266 default:
4267 ERR("Consumer type unknown");
4268 ret = -EINVAL;
4269 goto error;
4270 }
4271
4272 DBG2("Creating consumer directory: %s", path);
4273
4274 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4275 if (ret < 0) {
4276 if (errno != EEXIST) {
4277 PERROR("mkdir");
4278 ERR("Failed to create %s", path);
4279 goto error;
4280 }
4281 ret = -1;
4282 }
4283 if (is_root) {
4284 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4285 if (ret < 0) {
4286 ERR("Unable to set group on %s", path);
4287 PERROR("chown");
4288 goto error;
4289 }
4290 }
4291
4292 /* Create the kconsumerd error unix socket */
4293 consumer_data->err_sock =
4294 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4295 if (consumer_data->err_sock < 0) {
4296 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4297 ret = -1;
4298 goto error;
4299 }
4300
4301 /*
4302 * Set the CLOEXEC flag. Return code is useless because either way, the
4303 * show must go on.
4304 */
4305 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4306 if (ret < 0) {
4307 PERROR("utils_set_fd_cloexec");
4308 /* continue anyway */
4309 }
4310
4311 /* File permission MUST be 660 */
4312 ret = chmod(consumer_data->err_unix_sock_path,
4313 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4314 if (ret < 0) {
4315 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4316 PERROR("chmod");
4317 goto error;
4318 }
4319
4320 error:
4321 return ret;
4322 }
4323
4324 /*
4325 * Signal handler for the daemon
4326 *
4327 * Simply stop all worker threads, leaving main() return gracefully after
4328 * joining all threads and calling cleanup().
4329 */
4330 static void sighandler(int sig)
4331 {
4332 switch (sig) {
4333 case SIGPIPE:
4334 DBG("SIGPIPE caught");
4335 return;
4336 case SIGINT:
4337 DBG("SIGINT caught");
4338 stop_threads();
4339 break;
4340 case SIGTERM:
4341 DBG("SIGTERM caught");
4342 stop_threads();
4343 break;
4344 default:
4345 break;
4346 }
4347 }
4348
4349 /*
4350 * Setup signal handler for :
4351 * SIGINT, SIGTERM, SIGPIPE
4352 */
4353 static int set_signal_handler(void)
4354 {
4355 int ret = 0;
4356 struct sigaction sa;
4357 sigset_t sigset;
4358
4359 if ((ret = sigemptyset(&sigset)) < 0) {
4360 PERROR("sigemptyset");
4361 return ret;
4362 }
4363
4364 sa.sa_handler = sighandler;
4365 sa.sa_mask = sigset;
4366 sa.sa_flags = 0;
4367 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4368 PERROR("sigaction");
4369 return ret;
4370 }
4371
4372 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4373 PERROR("sigaction");
4374 return ret;
4375 }
4376
4377 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4378 PERROR("sigaction");
4379 return ret;
4380 }
4381
4382 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4383
4384 return ret;
4385 }
4386
4387 /*
4388 * Set open files limit to unlimited. This daemon can open a large number of
4389 * file descriptors in order to consumer multiple kernel traces.
4390 */
4391 static void set_ulimit(void)
4392 {
4393 int ret;
4394 struct rlimit lim;
4395
4396 /* The kernel does not allowed an infinite limit for open files */
4397 lim.rlim_cur = 65535;
4398 lim.rlim_max = 65535;
4399
4400 ret = setrlimit(RLIMIT_NOFILE, &lim);
4401 if (ret < 0) {
4402 PERROR("failed to set open files limit");
4403 }
4404 }
4405
4406 /*
4407 * Write pidfile using the rundir and opt_pidfile.
4408 */
4409 static void write_pidfile(void)
4410 {
4411 int ret;
4412 char pidfile_path[PATH_MAX];
4413
4414 assert(rundir);
4415
4416 if (opt_pidfile) {
4417 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4418 } else {
4419 /* Build pidfile path from rundir and opt_pidfile. */
4420 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4421 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4422 if (ret < 0) {
4423 PERROR("snprintf pidfile path");
4424 goto error;
4425 }
4426 }
4427
4428 /*
4429 * Create pid file in rundir. Return value is of no importance. The
4430 * execution will continue even though we are not able to write the file.
4431 */
4432 (void) utils_create_pid_file(getpid(), pidfile_path);
4433
4434 error:
4435 return;
4436 }
4437
4438 /*
4439 * Write JUL TCP port using the rundir.
4440 */
4441 static void write_julport(void)
4442 {
4443 int ret;
4444 char path[PATH_MAX];
4445
4446 assert(rundir);
4447
4448 ret = snprintf(path, sizeof(path), "%s/"
4449 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
4450 if (ret < 0) {
4451 PERROR("snprintf julport path");
4452 goto error;
4453 }
4454
4455 /*
4456 * Create TCP JUL port file in rundir. Return value is of no importance.
4457 * The execution will continue even though we are not able to write the
4458 * file.
4459 */
4460 (void) utils_create_pid_file(jul_tcp_port, path);
4461
4462 error:
4463 return;
4464 }
4465
4466 /*
4467 * main
4468 */
4469 int main(int argc, char **argv)
4470 {
4471 int ret = 0;
4472 void *status;
4473 const char *home_path, *env_app_timeout;
4474
4475 init_kernel_workarounds();
4476
4477 rcu_register_thread();
4478
4479 setup_consumerd_path();
4480
4481 page_size = sysconf(_SC_PAGESIZE);
4482 if (page_size < 0) {
4483 PERROR("sysconf _SC_PAGESIZE");
4484 page_size = LONG_MAX;
4485 WARN("Fallback page size to %ld", page_size);
4486 }
4487
4488 /* Parse arguments */
4489 progname = argv[0];
4490 if ((ret = parse_args(argc, argv)) < 0) {
4491 goto error;
4492 }
4493
4494 /* Daemonize */
4495 if (opt_daemon) {
4496 int i;
4497
4498 /*
4499 * fork
4500 * child: setsid, close FD 0, 1, 2, chdir /
4501 * parent: exit (if fork is successful)
4502 */
4503 ret = daemon(0, 0);
4504 if (ret < 0) {
4505 PERROR("daemon");
4506 goto error;
4507 }
4508 /*
4509 * We are in the child. Make sure all other file
4510 * descriptors are closed, in case we are called with
4511 * more opened file descriptors than the standard ones.
4512 */
4513 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4514 (void) close(i);
4515 }
4516 }
4517
4518 /* Create thread quit pipe */
4519 if ((ret = init_thread_quit_pipe()) < 0) {
4520 goto error;
4521 }
4522
4523 /* Check if daemon is UID = 0 */
4524 is_root = !getuid();
4525
4526 if (is_root) {
4527 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4528
4529 /* Create global run dir with root access */
4530 ret = create_lttng_rundir(rundir);
4531 if (ret < 0) {
4532 goto error;
4533 }
4534
4535 if (strlen(apps_unix_sock_path) == 0) {
4536 snprintf(apps_unix_sock_path, PATH_MAX,
4537 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4538 }
4539
4540 if (strlen(client_unix_sock_path) == 0) {
4541 snprintf(client_unix_sock_path, PATH_MAX,
4542 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4543 }
4544
4545 /* Set global SHM for ust */
4546 if (strlen(wait_shm_path) == 0) {
4547 snprintf(wait_shm_path, PATH_MAX,
4548 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4549 }
4550
4551 if (strlen(health_unix_sock_path) == 0) {
4552 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4553 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4554 }
4555
4556 /* Setup kernel consumerd path */
4557 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4558 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4559 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4560 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4561
4562 DBG2("Kernel consumer err path: %s",
4563 kconsumer_data.err_unix_sock_path);
4564 DBG2("Kernel consumer cmd path: %s",
4565 kconsumer_data.cmd_unix_sock_path);
4566 } else {
4567 home_path = utils_get_home_dir();
4568 if (home_path == NULL) {
4569 /* TODO: Add --socket PATH option */
4570 ERR("Can't get HOME directory for sockets creation.");
4571 ret = -EPERM;
4572 goto error;
4573 }
4574
4575 /*
4576 * Create rundir from home path. This will create something like
4577 * $HOME/.lttng
4578 */
4579 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4580 if (ret < 0) {
4581 ret = -ENOMEM;
4582 goto error;
4583 }
4584
4585 ret = create_lttng_rundir(rundir);
4586 if (ret < 0) {
4587 goto error;
4588 }
4589
4590 if (strlen(apps_unix_sock_path) == 0) {
4591 snprintf(apps_unix_sock_path, PATH_MAX,
4592 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4593 }
4594
4595 /* Set the cli tool unix socket path */
4596 if (strlen(client_unix_sock_path) == 0) {
4597 snprintf(client_unix_sock_path, PATH_MAX,
4598 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4599 }
4600
4601 /* Set global SHM for ust */
4602 if (strlen(wait_shm_path) == 0) {
4603 snprintf(wait_shm_path, PATH_MAX,
4604 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4605 }
4606
4607 /* Set health check Unix path */
4608 if (strlen(health_unix_sock_path) == 0) {
4609 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4610 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4611 }
4612 }
4613
4614 /* Set consumer initial state */
4615 kernel_consumerd_state = CONSUMER_STOPPED;
4616 ust_consumerd_state = CONSUMER_STOPPED;
4617
4618 DBG("Client socket path %s", client_unix_sock_path);
4619 DBG("Application socket path %s", apps_unix_sock_path);
4620 DBG("Application wait path %s", wait_shm_path);
4621 DBG("LTTng run directory path: %s", rundir);
4622
4623 /* 32 bits consumerd path setup */
4624 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4625 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4626 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4627 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4628
4629 DBG2("UST consumer 32 bits err path: %s",
4630 ustconsumer32_data.err_unix_sock_path);
4631 DBG2("UST consumer 32 bits cmd path: %s",
4632 ustconsumer32_data.cmd_unix_sock_path);
4633
4634 /* 64 bits consumerd path setup */
4635 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4636 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4637 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4638 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4639
4640 DBG2("UST consumer 64 bits err path: %s",
4641 ustconsumer64_data.err_unix_sock_path);
4642 DBG2("UST consumer 64 bits cmd path: %s",
4643 ustconsumer64_data.cmd_unix_sock_path);
4644
4645 /*
4646 * See if daemon already exist.
4647 */
4648 if ((ret = check_existing_daemon()) < 0) {
4649 ERR("Already running daemon.\n");
4650 /*
4651 * We do not goto exit because we must not cleanup()
4652 * because a daemon is already running.
4653 */
4654 goto error;
4655 }
4656
4657 /*
4658 * Init UST app hash table. Alloc hash table before this point since
4659 * cleanup() can get called after that point.
4660 */
4661 ust_app_ht_alloc();
4662
4663 /* Initialize JUL domain subsystem. */
4664 if ((ret = jul_init()) < 0) {
4665 /* ENOMEM at this point. */
4666 goto error;
4667 }
4668
4669 /* After this point, we can safely call cleanup() with "goto exit" */
4670
4671 /*
4672 * These actions must be executed as root. We do that *after* setting up
4673 * the sockets path because we MUST make the check for another daemon using
4674 * those paths *before* trying to set the kernel consumer sockets and init
4675 * kernel tracer.
4676 */
4677 if (is_root) {
4678 ret = set_consumer_sockets(&kconsumer_data, rundir);
4679 if (ret < 0) {
4680 goto exit;
4681 }
4682
4683 /* Setup kernel tracer */
4684 if (!opt_no_kernel) {
4685 init_kernel_tracer();
4686 }
4687
4688 /* Set ulimit for open files */
4689 set_ulimit();
4690 }
4691 /* init lttng_fd tracking must be done after set_ulimit. */
4692 lttng_fd_init();
4693
4694 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4695 if (ret < 0) {
4696 goto exit;
4697 }
4698
4699 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4700 if (ret < 0) {
4701 goto exit;
4702 }
4703
4704 if ((ret = set_signal_handler()) < 0) {
4705 goto exit;
4706 }
4707
4708 /* Setup the needed unix socket */
4709 if ((ret = init_daemon_socket()) < 0) {
4710 goto exit;
4711 }
4712
4713 /* Set credentials to socket */
4714 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4715 goto exit;
4716 }
4717
4718 /* Get parent pid if -S, --sig-parent is specified. */
4719 if (opt_sig_parent) {
4720 ppid = getppid();
4721 }
4722
4723 /* Setup the kernel pipe for waking up the kernel thread */
4724 if (is_root && !opt_no_kernel) {
4725 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4726 goto exit;
4727 }
4728 }
4729
4730 /* Setup the thread ht_cleanup communication pipe. */
4731 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4732 goto exit;
4733 }
4734
4735 /* Setup the thread apps communication pipe. */
4736 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4737 goto exit;
4738 }
4739
4740 /* Setup the thread apps notify communication pipe. */
4741 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4742 goto exit;
4743 }
4744
4745 /* Initialize global buffer per UID and PID registry. */
4746 buffer_reg_init_uid_registry();
4747 buffer_reg_init_pid_registry();
4748
4749 /* Init UST command queue. */
4750 cds_wfq_init(&ust_cmd_queue.queue);
4751
4752 /*
4753 * Get session list pointer. This pointer MUST NOT be free(). This list is
4754 * statically declared in session.c
4755 */
4756 session_list_ptr = session_get_list();
4757
4758 /* Set up max poll set size */
4759 lttng_poll_set_max_size();
4760
4761 cmd_init();
4762
4763 /* Check for the application socket timeout env variable. */
4764 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4765 if (env_app_timeout) {
4766 app_socket_timeout = atoi(env_app_timeout);
4767 } else {
4768 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4769 }
4770
4771 write_pidfile();
4772 write_julport();
4773
4774 /* Initialize communication library */
4775 lttcomm_init();
4776 /* This is to get the TCP timeout value. */
4777 lttcomm_inet_init();
4778
4779 /*
4780 * Initialize the health check subsystem. This call should set the
4781 * appropriate time values.
4782 */
4783 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
4784 if (!health_sessiond) {
4785 PERROR("health_app_create error");
4786 goto exit_health_sessiond_cleanup;
4787 }
4788
4789 /* Create thread to manage the client socket */
4790 ret = pthread_create(&ht_cleanup_thread, NULL,
4791 thread_ht_cleanup, (void *) NULL);
4792 if (ret != 0) {
4793 PERROR("pthread_create ht_cleanup");
4794 goto exit_ht_cleanup;
4795 }
4796
4797 /* Create thread to manage the client socket */
4798 ret = pthread_create(&health_thread, NULL,
4799 thread_manage_health, (void *) NULL);
4800 if (ret != 0) {
4801 PERROR("pthread_create health");
4802 goto exit_health;
4803 }
4804
4805 /* Create thread to manage the client socket */
4806 ret = pthread_create(&client_thread, NULL,
4807 thread_manage_clients, (void *) NULL);
4808 if (ret != 0) {
4809 PERROR("pthread_create clients");
4810 goto exit_client;
4811 }
4812
4813 /* Create thread to dispatch registration */
4814 ret = pthread_create(&dispatch_thread, NULL,
4815 thread_dispatch_ust_registration, (void *) NULL);
4816 if (ret != 0) {
4817 PERROR("pthread_create dispatch");
4818 goto exit_dispatch;
4819 }
4820
4821 /* Create thread to manage application registration. */
4822 ret = pthread_create(&reg_apps_thread, NULL,
4823 thread_registration_apps, (void *) NULL);
4824 if (ret != 0) {
4825 PERROR("pthread_create registration");
4826 goto exit_reg_apps;
4827 }
4828
4829 /* Create thread to manage application socket */
4830 ret = pthread_create(&apps_thread, NULL,
4831 thread_manage_apps, (void *) NULL);
4832 if (ret != 0) {
4833 PERROR("pthread_create apps");
4834 goto exit_apps;
4835 }
4836
4837 /* Create thread to manage application notify socket */
4838 ret = pthread_create(&apps_notify_thread, NULL,
4839 ust_thread_manage_notify, (void *) NULL);
4840 if (ret != 0) {
4841 PERROR("pthread_create apps");
4842 goto exit_apps_notify;
4843 }
4844
4845 /* Create JUL registration thread. */
4846 ret = pthread_create(&jul_reg_thread, NULL,
4847 jul_thread_manage_registration, (void *) NULL);
4848 if (ret != 0) {
4849 PERROR("pthread_create apps");
4850 goto exit_jul_reg;
4851 }
4852
4853 /* Don't start this thread if kernel tracing is not requested nor root */
4854 if (is_root && !opt_no_kernel) {
4855 /* Create kernel thread to manage kernel event */
4856 ret = pthread_create(&kernel_thread, NULL,
4857 thread_manage_kernel, (void *) NULL);
4858 if (ret != 0) {
4859 PERROR("pthread_create kernel");
4860 goto exit_kernel;
4861 }
4862
4863 ret = pthread_join(kernel_thread, &status);
4864 if (ret != 0) {
4865 PERROR("pthread_join");
4866 goto error; /* join error, exit without cleanup */
4867 }
4868 }
4869
4870 exit_kernel:
4871 ret = pthread_join(jul_reg_thread, &status);
4872 if (ret != 0) {
4873 PERROR("pthread_join JUL");
4874 goto error; /* join error, exit without cleanup */
4875 }
4876
4877 exit_jul_reg:
4878 ret = pthread_join(apps_notify_thread, &status);
4879 if (ret != 0) {
4880 PERROR("pthread_join apps notify");
4881 goto error; /* join error, exit without cleanup */
4882 }
4883
4884 exit_apps_notify:
4885 ret = pthread_join(apps_thread, &status);
4886 if (ret != 0) {
4887 PERROR("pthread_join apps");
4888 goto error; /* join error, exit without cleanup */
4889 }
4890
4891
4892 exit_apps:
4893 ret = pthread_join(reg_apps_thread, &status);
4894 if (ret != 0) {
4895 PERROR("pthread_join");
4896 goto error; /* join error, exit without cleanup */
4897 }
4898
4899 exit_reg_apps:
4900 ret = pthread_join(dispatch_thread, &status);
4901 if (ret != 0) {
4902 PERROR("pthread_join");
4903 goto error; /* join error, exit without cleanup */
4904 }
4905
4906 exit_dispatch:
4907 ret = pthread_join(client_thread, &status);
4908 if (ret != 0) {
4909 PERROR("pthread_join");
4910 goto error; /* join error, exit without cleanup */
4911 }
4912
4913 ret = join_consumer_thread(&kconsumer_data);
4914 if (ret != 0) {
4915 PERROR("join_consumer");
4916 goto error; /* join error, exit without cleanup */
4917 }
4918
4919 ret = join_consumer_thread(&ustconsumer32_data);
4920 if (ret != 0) {
4921 PERROR("join_consumer ust32");
4922 goto error; /* join error, exit without cleanup */
4923 }
4924
4925 ret = join_consumer_thread(&ustconsumer64_data);
4926 if (ret != 0) {
4927 PERROR("join_consumer ust64");
4928 goto error; /* join error, exit without cleanup */
4929 }
4930
4931 exit_client:
4932 ret = pthread_join(health_thread, &status);
4933 if (ret != 0) {
4934 PERROR("pthread_join health thread");
4935 goto error; /* join error, exit without cleanup */
4936 }
4937
4938 exit_health:
4939 ret = pthread_join(ht_cleanup_thread, &status);
4940 if (ret != 0) {
4941 PERROR("pthread_join ht cleanup thread");
4942 goto error; /* join error, exit without cleanup */
4943 }
4944 exit_ht_cleanup:
4945 health_app_destroy(health_sessiond);
4946 exit_health_sessiond_cleanup:
4947 exit:
4948 /*
4949 * cleanup() is called when no other thread is running.
4950 */
4951 rcu_thread_online();
4952 cleanup();
4953 rcu_thread_offline();
4954 rcu_unregister_thread();
4955 if (!ret) {
4956 exit(EXIT_SUCCESS);
4957 }
4958 error:
4959 exit(EXIT_FAILURE);
4960 }
This page took 0.172909 seconds and 5 git commands to generate.