Lock consumer data before fd check during destroy
[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.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66
67 #define CONSUMERD_FILE "lttng-consumerd"
68
69 /* Const values */
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71
72 const char *progname;
73 const char *opt_tracing_group;
74 static const char *opt_pidfile;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 int apps_cmd_notify_pipe[2] = { -1, -1 };
152
153 /* Pthread, Mutexes and Semaphores */
154 static pthread_t apps_thread;
155 static pthread_t apps_notify_thread;
156 static pthread_t reg_apps_thread;
157 static pthread_t client_thread;
158 static pthread_t kernel_thread;
159 static pthread_t dispatch_thread;
160 static pthread_t health_thread;
161 static pthread_t ht_cleanup_thread;
162
163 /*
164 * UST registration command queue. This queue is tied with a futex and uses a N
165 * wakers / 1 waiter implemented and detailed in futex.c/.h
166 *
167 * The thread_manage_apps and thread_dispatch_ust_registration interact with
168 * this queue and the wait/wake scheme.
169 */
170 static struct ust_cmd_queue ust_cmd_queue;
171
172 /*
173 * Pointer initialized before thread creation.
174 *
175 * This points to the tracing session list containing the session count and a
176 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
177 * MUST NOT be taken if you call a public function in session.c.
178 *
179 * The lock is nested inside the structure: session_list_ptr->lock. Please use
180 * session_lock_list and session_unlock_list for lock acquisition.
181 */
182 static struct ltt_session_list *session_list_ptr;
183
184 int ust_consumerd64_fd = -1;
185 int ust_consumerd32_fd = -1;
186
187 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
188 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
189 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
190 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
191
192 static const char *module_proc_lttng = "/proc/lttng";
193
194 /*
195 * Consumer daemon state which is changed when spawning it, killing it or in
196 * case of a fatal error.
197 */
198 enum consumerd_state {
199 CONSUMER_STARTED = 1,
200 CONSUMER_STOPPED = 2,
201 CONSUMER_ERROR = 3,
202 };
203
204 /*
205 * This consumer daemon state is used to validate if a client command will be
206 * able to reach the consumer. If not, the client is informed. For instance,
207 * doing a "lttng start" when the consumer state is set to ERROR will return an
208 * error to the client.
209 *
210 * The following example shows a possible race condition of this scheme:
211 *
212 * consumer thread error happens
213 * client cmd arrives
214 * client cmd checks state -> still OK
215 * consumer thread exit, sets error
216 * client cmd try to talk to consumer
217 * ...
218 *
219 * However, since the consumer is a different daemon, we have no way of making
220 * sure the command will reach it safely even with this state flag. This is why
221 * we consider that up to the state validation during command processing, the
222 * command is safe. After that, we can not guarantee the correctness of the
223 * client request vis-a-vis the consumer.
224 */
225 static enum consumerd_state ust_consumerd_state;
226 static enum consumerd_state kernel_consumerd_state;
227
228 /*
229 * Socket timeout for receiving and sending in seconds.
230 */
231 static int app_socket_timeout;
232
233 /* Set in main() with the current page size. */
234 long page_size;
235
236 static
237 void setup_consumerd_path(void)
238 {
239 const char *bin, *libdir;
240
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
245 */
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
256 }
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
259 }
260 #else
261 #error "Unknown bitness"
262 #endif
263
264 /*
265 * runtime env. var. overrides the build default.
266 */
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
270 }
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
274 }
275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
280 if (libdir) {
281 consumerd64_libdir = libdir;
282 }
283 }
284
285 /*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
289 {
290 int ret;
291
292 assert(events);
293
294 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
295 if (ret < 0) {
296 goto error;
297 }
298
299 /* Add quit pipe */
300 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
301 if (ret < 0) {
302 goto error;
303 }
304
305 return 0;
306
307 error:
308 return ret;
309 }
310
311 /*
312 * Check if the thread quit pipe was triggered.
313 *
314 * Return 1 if it was triggered else 0;
315 */
316 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
317 {
318 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
319 return 1;
320 }
321
322 return 0;
323 }
324
325 /*
326 * Return group ID of the tracing group or -1 if not found.
327 */
328 static gid_t allowed_group(void)
329 {
330 struct group *grp;
331
332 if (opt_tracing_group) {
333 grp = getgrnam(opt_tracing_group);
334 } else {
335 grp = getgrnam(default_tracing_group);
336 }
337 if (!grp) {
338 return -1;
339 } else {
340 return grp->gr_gid;
341 }
342 }
343
344 /*
345 * Init thread quit pipe.
346 *
347 * Return -1 on error or 0 if all pipes are created.
348 */
349 static int init_thread_quit_pipe(void)
350 {
351 int ret, i;
352
353 ret = pipe(thread_quit_pipe);
354 if (ret < 0) {
355 PERROR("thread quit pipe");
356 goto error;
357 }
358
359 for (i = 0; i < 2; i++) {
360 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
361 if (ret < 0) {
362 PERROR("fcntl");
363 goto error;
364 }
365 }
366
367 error:
368 return ret;
369 }
370
371 /*
372 * Stop all threads by closing the thread quit pipe.
373 */
374 static void stop_threads(void)
375 {
376 int ret;
377
378 /* Stopping all threads */
379 DBG("Terminating all threads");
380 ret = notify_thread_pipe(thread_quit_pipe[1]);
381 if (ret < 0) {
382 ERR("write error on thread quit pipe");
383 }
384
385 /* Dispatch thread */
386 CMM_STORE_SHARED(dispatch_thread_exit, 1);
387 futex_nto1_wake(&ust_cmd_queue.futex);
388 }
389
390 /*
391 * Close every consumer sockets.
392 */
393 static void close_consumer_sockets(void)
394 {
395 int ret;
396
397 if (kconsumer_data.err_sock >= 0) {
398 ret = close(kconsumer_data.err_sock);
399 if (ret < 0) {
400 PERROR("kernel consumer err_sock close");
401 }
402 }
403 if (ustconsumer32_data.err_sock >= 0) {
404 ret = close(ustconsumer32_data.err_sock);
405 if (ret < 0) {
406 PERROR("UST consumerd32 err_sock close");
407 }
408 }
409 if (ustconsumer64_data.err_sock >= 0) {
410 ret = close(ustconsumer64_data.err_sock);
411 if (ret < 0) {
412 PERROR("UST consumerd64 err_sock close");
413 }
414 }
415 if (kconsumer_data.cmd_sock >= 0) {
416 ret = close(kconsumer_data.cmd_sock);
417 if (ret < 0) {
418 PERROR("kernel consumer cmd_sock close");
419 }
420 }
421 if (ustconsumer32_data.cmd_sock >= 0) {
422 ret = close(ustconsumer32_data.cmd_sock);
423 if (ret < 0) {
424 PERROR("UST consumerd32 cmd_sock close");
425 }
426 }
427 if (ustconsumer64_data.cmd_sock >= 0) {
428 ret = close(ustconsumer64_data.cmd_sock);
429 if (ret < 0) {
430 PERROR("UST consumerd64 cmd_sock close");
431 }
432 }
433 }
434
435 /*
436 * Cleanup the daemon
437 */
438 static void cleanup(void)
439 {
440 int ret;
441 char *cmd = NULL;
442 struct ltt_session *sess, *stmp;
443
444 DBG("Cleaning up");
445
446 /*
447 * Close the thread quit pipe. It has already done its job,
448 * since we are now called.
449 */
450 utils_close_pipe(thread_quit_pipe);
451
452 /*
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
455 */
456 if (opt_pidfile) {
457 ret = remove(opt_pidfile);
458 if (ret < 0) {
459 PERROR("remove pidfile %s", opt_pidfile);
460 }
461 }
462
463 DBG("Removing %s directory", rundir);
464 ret = asprintf(&cmd, "rm -rf %s", rundir);
465 if (ret < 0) {
466 ERR("asprintf failed. Something is really wrong!");
467 }
468
469 /* Remove lttng run directory */
470 ret = system(cmd);
471 if (ret < 0) {
472 ERR("Unable to clean %s", rundir);
473 }
474 free(cmd);
475 free(rundir);
476
477 DBG("Cleaning up all sessions");
478
479 /* Destroy session list mutex */
480 if (session_list_ptr != NULL) {
481 pthread_mutex_destroy(&session_list_ptr->lock);
482
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess, stmp,
485 &session_list_ptr->head, list) {
486 cmd_destroy_session(sess, kernel_poll_pipe[1]);
487 }
488 }
489
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
493
494 if (is_root && !opt_no_kernel) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd >= 0) {
497 ret = close(kernel_tracer_fd);
498 if (ret) {
499 PERROR("close");
500 }
501 }
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
504 }
505
506 close_consumer_sockets();
507
508 /* <fun> */
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
512 /* </fun> */
513 }
514
515 /*
516 * Send data on a unix socket using the liblttsessiondcomm API.
517 *
518 * Return lttcomm error code.
519 */
520 static int send_unix_sock(int sock, void *buf, size_t len)
521 {
522 /* Check valid length */
523 if (len == 0) {
524 return -1;
525 }
526
527 return lttcomm_send_unix_sock(sock, buf, len);
528 }
529
530 /*
531 * Free memory of a command context structure.
532 */
533 static void clean_command_ctx(struct command_ctx **cmd_ctx)
534 {
535 DBG("Clean command context structure");
536 if (*cmd_ctx) {
537 if ((*cmd_ctx)->llm) {
538 free((*cmd_ctx)->llm);
539 }
540 if ((*cmd_ctx)->lsm) {
541 free((*cmd_ctx)->lsm);
542 }
543 free(*cmd_ctx);
544 *cmd_ctx = NULL;
545 }
546 }
547
548 /*
549 * Notify UST applications using the shm mmap futex.
550 */
551 static int notify_ust_apps(int active)
552 {
553 char *wait_shm_mmap;
554
555 DBG("Notifying applications of session daemon state: %d", active);
556
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
559 if (wait_shm_mmap == NULL) {
560 goto error;
561 }
562
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap, active);
565
566 /* Apps notified successfully */
567 return 0;
568
569 error:
570 return -1;
571 }
572
573 /*
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
576 * structure.
577 *
578 * Return total size of the buffer pointed by buf.
579 */
580 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
581 {
582 int ret, buf_size;
583
584 buf_size = size;
585
586 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
587 if (cmd_ctx->llm == NULL) {
588 PERROR("zmalloc");
589 ret = -ENOMEM;
590 goto error;
591 }
592
593 /* Copy common data */
594 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
595 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
596
597 cmd_ctx->llm->data_size = size;
598 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
599
600 return buf_size;
601
602 error:
603 return ret;
604 }
605
606 /*
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
609 */
610 static int update_kernel_poll(struct lttng_poll_event *events)
611 {
612 int ret;
613 struct ltt_session *session;
614 struct ltt_kernel_channel *channel;
615
616 DBG("Updating kernel poll set");
617
618 session_lock_list();
619 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
620 session_lock(session);
621 if (session->kernel_session == NULL) {
622 session_unlock(session);
623 continue;
624 }
625
626 cds_list_for_each_entry(channel,
627 &session->kernel_session->channel_list.head, list) {
628 /* Add channel fd to the kernel poll set */
629 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
630 if (ret < 0) {
631 session_unlock(session);
632 goto error;
633 }
634 DBG("Channel fd %d added to kernel set", channel->fd);
635 }
636 session_unlock(session);
637 }
638 session_unlock_list();
639
640 return 0;
641
642 error:
643 session_unlock_list();
644 return -1;
645 }
646
647 /*
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
650 *
651 * Useful for CPU hotplug feature.
652 */
653 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
654 {
655 int ret = 0;
656 struct ltt_session *session;
657 struct ltt_kernel_session *ksess;
658 struct ltt_kernel_channel *channel;
659
660 DBG("Updating kernel streams for channel fd %d", fd);
661
662 session_lock_list();
663 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
664 session_lock(session);
665 if (session->kernel_session == NULL) {
666 session_unlock(session);
667 continue;
668 }
669 ksess = session->kernel_session;
670
671 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
672 if (channel->fd == fd) {
673 DBG("Channel found, updating kernel streams");
674 ret = kernel_open_channel_stream(channel);
675 if (ret < 0) {
676 goto error;
677 }
678 /* Update the stream global counter */
679 ksess->stream_count_global += ret;
680
681 /*
682 * Have we already sent fds to the consumer? If yes, it means
683 * that tracing is started so it is safe to send our updated
684 * stream fds.
685 */
686 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
687 struct lttng_ht_iter iter;
688 struct consumer_socket *socket;
689
690 rcu_read_lock();
691 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
692 &iter.iter, socket, node.node) {
693 /* Code flow error */
694 assert(socket->fd);
695
696 pthread_mutex_lock(socket->lock);
697 ret = kernel_consumer_send_channel_stream(socket,
698 channel, ksess,
699 session->output_traces ? 1 : 0);
700 pthread_mutex_unlock(socket->lock);
701 if (ret < 0) {
702 rcu_read_unlock();
703 goto error;
704 }
705 }
706 rcu_read_unlock();
707 }
708 goto error;
709 }
710 }
711 session_unlock(session);
712 }
713 session_unlock_list();
714 return ret;
715
716 error:
717 session_unlock(session);
718 session_unlock_list();
719 return ret;
720 }
721
722 /*
723 * For each tracing session, update newly registered apps. The session list
724 * lock MUST be acquired before calling this.
725 */
726 static void update_ust_app(int app_sock)
727 {
728 struct ltt_session *sess, *stmp;
729
730 /* Consumer is in an ERROR state. Stop any application update. */
731 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
732 /* Stop the update process since the consumer is dead. */
733 return;
734 }
735
736 /* For all tracing session(s) */
737 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
738 session_lock(sess);
739 if (sess->ust_session) {
740 ust_app_global_update(sess->ust_session, app_sock);
741 }
742 session_unlock(sess);
743 }
744 }
745
746 /*
747 * This thread manage event coming from the kernel.
748 *
749 * Features supported in this thread:
750 * -) CPU Hotplug
751 */
752 static void *thread_manage_kernel(void *data)
753 {
754 int ret, i, pollfd, update_poll_flag = 1, err = -1;
755 uint32_t revents, nb_fd;
756 char tmp;
757 struct lttng_poll_event events;
758
759 DBG("[thread] Thread manage kernel started");
760
761 health_register(HEALTH_TYPE_KERNEL);
762
763 /*
764 * This first step of the while is to clean this structure which could free
765 * non NULL pointers so initialize it before the loop.
766 */
767 lttng_poll_init(&events);
768
769 if (testpoint(thread_manage_kernel)) {
770 goto error_testpoint;
771 }
772
773 health_code_update();
774
775 if (testpoint(thread_manage_kernel_before_loop)) {
776 goto error_testpoint;
777 }
778
779 while (1) {
780 health_code_update();
781
782 if (update_poll_flag == 1) {
783 /* Clean events object. We are about to populate it again. */
784 lttng_poll_clean(&events);
785
786 ret = sessiond_set_thread_pollset(&events, 2);
787 if (ret < 0) {
788 goto error_poll_create;
789 }
790
791 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
792 if (ret < 0) {
793 goto error;
794 }
795
796 /* This will add the available kernel channel if any. */
797 ret = update_kernel_poll(&events);
798 if (ret < 0) {
799 goto error;
800 }
801 update_poll_flag = 0;
802 }
803
804 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
805
806 /* Poll infinite value of time */
807 restart:
808 health_poll_entry();
809 ret = lttng_poll_wait(&events, -1);
810 health_poll_exit();
811 if (ret < 0) {
812 /*
813 * Restart interrupted system call.
814 */
815 if (errno == EINTR) {
816 goto restart;
817 }
818 goto error;
819 } else if (ret == 0) {
820 /* Should not happen since timeout is infinite */
821 ERR("Return value of poll is 0 with an infinite timeout.\n"
822 "This should not have happened! Continuing...");
823 continue;
824 }
825
826 nb_fd = ret;
827
828 for (i = 0; i < nb_fd; i++) {
829 /* Fetch once the poll data */
830 revents = LTTNG_POLL_GETEV(&events, i);
831 pollfd = LTTNG_POLL_GETFD(&events, i);
832
833 health_code_update();
834
835 /* Thread quit pipe has been closed. Killing thread. */
836 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
837 if (ret) {
838 err = 0;
839 goto exit;
840 }
841
842 /* Check for data on kernel pipe */
843 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
844 do {
845 ret = read(kernel_poll_pipe[0], &tmp, 1);
846 } while (ret < 0 && errno == EINTR);
847 /*
848 * Ret value is useless here, if this pipe gets any actions an
849 * update is required anyway.
850 */
851 update_poll_flag = 1;
852 continue;
853 } else {
854 /*
855 * New CPU detected by the kernel. Adding kernel stream to
856 * kernel session and updating the kernel consumer
857 */
858 if (revents & LPOLLIN) {
859 ret = update_kernel_stream(&kconsumer_data, pollfd);
860 if (ret < 0) {
861 continue;
862 }
863 break;
864 /*
865 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
866 * and unregister kernel stream at this point.
867 */
868 }
869 }
870 }
871 }
872
873 exit:
874 error:
875 lttng_poll_clean(&events);
876 error_poll_create:
877 error_testpoint:
878 utils_close_pipe(kernel_poll_pipe);
879 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
880 if (err) {
881 health_error();
882 ERR("Health error occurred in %s", __func__);
883 WARN("Kernel thread died unexpectedly. "
884 "Kernel tracing can continue but CPU hotplug is disabled.");
885 }
886 health_unregister();
887 DBG("Kernel thread dying");
888 return NULL;
889 }
890
891 /*
892 * Signal pthread condition of the consumer data that the thread.
893 */
894 static void signal_consumer_condition(struct consumer_data *data, int state)
895 {
896 pthread_mutex_lock(&data->cond_mutex);
897
898 /*
899 * The state is set before signaling. It can be any value, it's the waiter
900 * job to correctly interpret this condition variable associated to the
901 * consumer pthread_cond.
902 *
903 * A value of 0 means that the corresponding thread of the consumer data
904 * was not started. 1 indicates that the thread has started and is ready
905 * for action. A negative value means that there was an error during the
906 * thread bootstrap.
907 */
908 data->consumer_thread_is_ready = state;
909 (void) pthread_cond_signal(&data->cond);
910
911 pthread_mutex_unlock(&data->cond_mutex);
912 }
913
914 /*
915 * This thread manage the consumer error sent back to the session daemon.
916 */
917 static void *thread_manage_consumer(void *data)
918 {
919 int sock = -1, i, ret, pollfd, err = -1;
920 uint32_t revents, nb_fd;
921 enum lttcomm_return_code code;
922 struct lttng_poll_event events;
923 struct consumer_data *consumer_data = data;
924
925 DBG("[thread] Manage consumer started");
926
927 health_register(HEALTH_TYPE_CONSUMER);
928
929 health_code_update();
930
931 /*
932 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
933 * metadata_sock. Nothing more will be added to this poll set.
934 */
935 ret = sessiond_set_thread_pollset(&events, 3);
936 if (ret < 0) {
937 goto error_poll;
938 }
939
940 /*
941 * The error socket here is already in a listening state which was done
942 * just before spawning this thread to avoid a race between the consumer
943 * daemon exec trying to connect and the listen() call.
944 */
945 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
946 if (ret < 0) {
947 goto error;
948 }
949
950 health_code_update();
951
952 /* Infinite blocking call, waiting for transmission */
953 restart:
954 health_poll_entry();
955
956 if (testpoint(thread_manage_consumer)) {
957 goto error;
958 }
959
960 ret = lttng_poll_wait(&events, -1);
961 health_poll_exit();
962 if (ret < 0) {
963 /*
964 * Restart interrupted system call.
965 */
966 if (errno == EINTR) {
967 goto restart;
968 }
969 goto error;
970 }
971
972 nb_fd = ret;
973
974 for (i = 0; i < nb_fd; i++) {
975 /* Fetch once the poll data */
976 revents = LTTNG_POLL_GETEV(&events, i);
977 pollfd = LTTNG_POLL_GETFD(&events, i);
978
979 health_code_update();
980
981 /* Thread quit pipe has been closed. Killing thread. */
982 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
983 if (ret) {
984 err = 0;
985 goto exit;
986 }
987
988 /* Event on the registration socket */
989 if (pollfd == consumer_data->err_sock) {
990 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
991 ERR("consumer err socket poll error");
992 goto error;
993 }
994 }
995 }
996
997 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
998 if (sock < 0) {
999 goto error;
1000 }
1001
1002 /*
1003 * Set the CLOEXEC flag. Return code is useless because either way, the
1004 * show must go on.
1005 */
1006 (void) utils_set_fd_cloexec(sock);
1007
1008 health_code_update();
1009
1010 DBG2("Receiving code from consumer err_sock");
1011
1012 /* Getting status code from kconsumerd */
1013 ret = lttcomm_recv_unix_sock(sock, &code,
1014 sizeof(enum lttcomm_return_code));
1015 if (ret <= 0) {
1016 goto error;
1017 }
1018
1019 health_code_update();
1020
1021 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1022 /* Connect both socket, command and metadata. */
1023 consumer_data->cmd_sock =
1024 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1025 consumer_data->metadata_fd =
1026 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1027 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1028 PERROR("consumer connect cmd socket");
1029 /* On error, signal condition and quit. */
1030 signal_consumer_condition(consumer_data, -1);
1031 goto error;
1032 }
1033 consumer_data->metadata_sock.fd = &consumer_data->metadata_fd;
1034 /* Create metadata socket lock. */
1035 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1036 if (consumer_data->metadata_sock.lock == NULL) {
1037 PERROR("zmalloc pthread mutex");
1038 ret = -1;
1039 goto error;
1040 }
1041 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1042
1043 signal_consumer_condition(consumer_data, 1);
1044 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1045 DBG("Consumer metadata socket ready (fd: %d)",
1046 consumer_data->metadata_fd);
1047 } else {
1048 ERR("consumer error when waiting for SOCK_READY : %s",
1049 lttcomm_get_readable_code(-code));
1050 goto error;
1051 }
1052
1053 /* Remove the consumerd error sock since we've established a connexion */
1054 ret = lttng_poll_del(&events, consumer_data->err_sock);
1055 if (ret < 0) {
1056 goto error;
1057 }
1058
1059 /* Add new accepted error socket. */
1060 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1061 if (ret < 0) {
1062 goto error;
1063 }
1064
1065 /* Add metadata socket that is successfully connected. */
1066 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1067 LPOLLIN | LPOLLRDHUP);
1068 if (ret < 0) {
1069 goto error;
1070 }
1071
1072 health_code_update();
1073
1074 /* Infinite blocking call, waiting for transmission */
1075 restart_poll:
1076 while (1) {
1077 health_poll_entry();
1078 ret = lttng_poll_wait(&events, -1);
1079 health_poll_exit();
1080 if (ret < 0) {
1081 /*
1082 * Restart interrupted system call.
1083 */
1084 if (errno == EINTR) {
1085 goto restart_poll;
1086 }
1087 goto error;
1088 }
1089
1090 nb_fd = ret;
1091
1092 for (i = 0; i < nb_fd; i++) {
1093 /* Fetch once the poll data */
1094 revents = LTTNG_POLL_GETEV(&events, i);
1095 pollfd = LTTNG_POLL_GETFD(&events, i);
1096
1097 health_code_update();
1098
1099 /* Thread quit pipe has been closed. Killing thread. */
1100 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1101 if (ret) {
1102 err = 0;
1103 goto exit;
1104 }
1105
1106 if (pollfd == sock) {
1107 /* Event on the consumerd socket */
1108 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1109 ERR("consumer err socket second poll error");
1110 goto error;
1111 }
1112 health_code_update();
1113 /* Wait for any kconsumerd error */
1114 ret = lttcomm_recv_unix_sock(sock, &code,
1115 sizeof(enum lttcomm_return_code));
1116 if (ret <= 0) {
1117 ERR("consumer closed the command socket");
1118 goto error;
1119 }
1120
1121 ERR("consumer return code : %s",
1122 lttcomm_get_readable_code(-code));
1123
1124 goto exit;
1125 } else if (pollfd == consumer_data->metadata_fd) {
1126 /* UST metadata requests */
1127 ret = ust_consumer_metadata_request(
1128 &consumer_data->metadata_sock);
1129 if (ret < 0) {
1130 ERR("Handling metadata request");
1131 goto error;
1132 }
1133 break;
1134 } else {
1135 ERR("Unknown pollfd");
1136 goto error;
1137 }
1138 }
1139 health_code_update();
1140 }
1141
1142 exit:
1143 error:
1144 /*
1145 * We lock here because we are about to close the sockets and some other
1146 * thread might be using them so wait before we are exclusive which will
1147 * abort all other consumer command by other threads.
1148 */
1149 pthread_mutex_lock(&consumer_data->lock);
1150
1151 /* Immediately set the consumerd state to stopped */
1152 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1153 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1154 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1155 consumer_data->type == LTTNG_CONSUMER32_UST) {
1156 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1157 } else {
1158 /* Code flow error... */
1159 assert(0);
1160 }
1161
1162 if (consumer_data->err_sock >= 0) {
1163 ret = close(consumer_data->err_sock);
1164 if (ret) {
1165 PERROR("close");
1166 }
1167 consumer_data->err_sock = -1;
1168 }
1169 if (consumer_data->cmd_sock >= 0) {
1170 ret = close(consumer_data->cmd_sock);
1171 if (ret) {
1172 PERROR("close");
1173 }
1174 consumer_data->cmd_sock = -1;
1175 }
1176 if (*consumer_data->metadata_sock.fd >= 0) {
1177 ret = close(*consumer_data->metadata_sock.fd);
1178 if (ret) {
1179 PERROR("close");
1180 }
1181 }
1182
1183 if (sock >= 0) {
1184 ret = close(sock);
1185 if (ret) {
1186 PERROR("close");
1187 }
1188 }
1189
1190 unlink(consumer_data->err_unix_sock_path);
1191 unlink(consumer_data->cmd_unix_sock_path);
1192 consumer_data->pid = 0;
1193 pthread_mutex_unlock(&consumer_data->lock);
1194 /* Cleanup metadata socket mutex. */
1195 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1196 free(consumer_data->metadata_sock.lock);
1197
1198 lttng_poll_clean(&events);
1199 error_poll:
1200 if (err) {
1201 health_error();
1202 ERR("Health error occurred in %s", __func__);
1203 }
1204 health_unregister();
1205 DBG("consumer thread cleanup completed");
1206
1207 return NULL;
1208 }
1209
1210 /*
1211 * This thread manage application communication.
1212 */
1213 static void *thread_manage_apps(void *data)
1214 {
1215 int i, ret, pollfd, err = -1;
1216 uint32_t revents, nb_fd;
1217 struct lttng_poll_event events;
1218
1219 DBG("[thread] Manage application started");
1220
1221 rcu_register_thread();
1222 rcu_thread_online();
1223
1224 health_register(HEALTH_TYPE_APP_MANAGE);
1225
1226 if (testpoint(thread_manage_apps)) {
1227 goto error_testpoint;
1228 }
1229
1230 health_code_update();
1231
1232 ret = sessiond_set_thread_pollset(&events, 2);
1233 if (ret < 0) {
1234 goto error_poll_create;
1235 }
1236
1237 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1238 if (ret < 0) {
1239 goto error;
1240 }
1241
1242 if (testpoint(thread_manage_apps_before_loop)) {
1243 goto error;
1244 }
1245
1246 health_code_update();
1247
1248 while (1) {
1249 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1250
1251 /* Inifinite blocking call, waiting for transmission */
1252 restart:
1253 health_poll_entry();
1254 ret = lttng_poll_wait(&events, -1);
1255 health_poll_exit();
1256 if (ret < 0) {
1257 /*
1258 * Restart interrupted system call.
1259 */
1260 if (errno == EINTR) {
1261 goto restart;
1262 }
1263 goto error;
1264 }
1265
1266 nb_fd = ret;
1267
1268 for (i = 0; i < nb_fd; i++) {
1269 /* Fetch once the poll data */
1270 revents = LTTNG_POLL_GETEV(&events, i);
1271 pollfd = LTTNG_POLL_GETFD(&events, i);
1272
1273 health_code_update();
1274
1275 /* Thread quit pipe has been closed. Killing thread. */
1276 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1277 if (ret) {
1278 err = 0;
1279 goto exit;
1280 }
1281
1282 /* Inspect the apps cmd pipe */
1283 if (pollfd == apps_cmd_pipe[0]) {
1284 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1285 ERR("Apps command pipe error");
1286 goto error;
1287 } else if (revents & LPOLLIN) {
1288 int sock;
1289
1290 /* Empty pipe */
1291 do {
1292 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1293 } while (ret < 0 && errno == EINTR);
1294 if (ret < 0 || ret < sizeof(sock)) {
1295 PERROR("read apps cmd pipe");
1296 goto error;
1297 }
1298
1299 health_code_update();
1300
1301 /*
1302 * We only monitor the error events of the socket. This
1303 * thread does not handle any incoming data from UST
1304 * (POLLIN).
1305 */
1306 ret = lttng_poll_add(&events, sock,
1307 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1308 if (ret < 0) {
1309 goto error;
1310 }
1311
1312 /*
1313 * Set socket timeout for both receiving and ending.
1314 * app_socket_timeout is in seconds, whereas
1315 * lttcomm_setsockopt_rcv_timeout and
1316 * lttcomm_setsockopt_snd_timeout expect msec as
1317 * parameter.
1318 */
1319 (void) lttcomm_setsockopt_rcv_timeout(sock,
1320 app_socket_timeout * 1000);
1321 (void) lttcomm_setsockopt_snd_timeout(sock,
1322 app_socket_timeout * 1000);
1323
1324 DBG("Apps with sock %d added to poll set", sock);
1325
1326 health_code_update();
1327
1328 break;
1329 }
1330 } else {
1331 /*
1332 * At this point, we know that a registered application made
1333 * the event at poll_wait.
1334 */
1335 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1336 /* Removing from the poll set */
1337 ret = lttng_poll_del(&events, pollfd);
1338 if (ret < 0) {
1339 goto error;
1340 }
1341
1342 /* Socket closed on remote end. */
1343 ust_app_unregister(pollfd);
1344 break;
1345 }
1346 }
1347
1348 health_code_update();
1349 }
1350 }
1351
1352 exit:
1353 error:
1354 lttng_poll_clean(&events);
1355 error_poll_create:
1356 error_testpoint:
1357 utils_close_pipe(apps_cmd_pipe);
1358 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1359
1360 /*
1361 * We don't clean the UST app hash table here since already registered
1362 * applications can still be controlled so let them be until the session
1363 * daemon dies or the applications stop.
1364 */
1365
1366 if (err) {
1367 health_error();
1368 ERR("Health error occurred in %s", __func__);
1369 }
1370 health_unregister();
1371 DBG("Application communication apps thread cleanup complete");
1372 rcu_thread_offline();
1373 rcu_unregister_thread();
1374 return NULL;
1375 }
1376
1377 /*
1378 * Send a socket to a thread This is called from the dispatch UST registration
1379 * thread once all sockets are set for the application.
1380 *
1381 * On success, return 0 else a negative value being the errno message of the
1382 * write().
1383 */
1384 static int send_socket_to_thread(int fd, int sock)
1385 {
1386 int ret;
1387
1388 /* Sockets MUST be set or else this should not have been called. */
1389 assert(fd >= 0);
1390 assert(sock >= 0);
1391
1392 do {
1393 ret = write(fd, &sock, sizeof(sock));
1394 } while (ret < 0 && errno == EINTR);
1395 if (ret < 0 || ret != sizeof(sock)) {
1396 PERROR("write apps pipe %d", fd);
1397 if (ret < 0) {
1398 ret = -errno;
1399 }
1400 goto error;
1401 }
1402
1403 /* All good. Don't send back the write positive ret value. */
1404 ret = 0;
1405 error:
1406 return ret;
1407 }
1408
1409 /*
1410 * Sanitize the wait queue of the dispatch registration thread meaning removing
1411 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1412 * notify socket is never received.
1413 */
1414 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1415 {
1416 int ret, nb_fd = 0, i;
1417 unsigned int fd_added = 0;
1418 struct lttng_poll_event events;
1419 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1420
1421 assert(wait_queue);
1422
1423 lttng_poll_init(&events);
1424
1425 /* Just skip everything for an empty queue. */
1426 if (!wait_queue->count) {
1427 goto end;
1428 }
1429
1430 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1431 if (ret < 0) {
1432 goto error_create;
1433 }
1434
1435 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1436 &wait_queue->head, head) {
1437 assert(wait_node->app);
1438 ret = lttng_poll_add(&events, wait_node->app->sock,
1439 LPOLLHUP | LPOLLERR);
1440 if (ret < 0) {
1441 goto error;
1442 }
1443
1444 fd_added = 1;
1445 }
1446
1447 if (!fd_added) {
1448 goto end;
1449 }
1450
1451 /*
1452 * Poll but don't block so we can quickly identify the faulty events and
1453 * clean them afterwards from the wait queue.
1454 */
1455 ret = lttng_poll_wait(&events, 0);
1456 if (ret < 0) {
1457 goto error;
1458 }
1459 nb_fd = ret;
1460
1461 for (i = 0; i < nb_fd; i++) {
1462 /* Get faulty FD. */
1463 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1464 int pollfd = LTTNG_POLL_GETFD(&events, i);
1465
1466 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1467 &wait_queue->head, head) {
1468 if (pollfd == wait_node->app->sock &&
1469 (revents & (LPOLLHUP | LPOLLERR))) {
1470 cds_list_del(&wait_node->head);
1471 wait_queue->count--;
1472 ust_app_destroy(wait_node->app);
1473 free(wait_node);
1474 break;
1475 }
1476 }
1477 }
1478
1479 if (nb_fd > 0) {
1480 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1481 }
1482
1483 end:
1484 lttng_poll_clean(&events);
1485 return;
1486
1487 error:
1488 lttng_poll_clean(&events);
1489 error_create:
1490 ERR("Unable to sanitize wait queue");
1491 return;
1492 }
1493
1494 /*
1495 * Dispatch request from the registration threads to the application
1496 * communication thread.
1497 */
1498 static void *thread_dispatch_ust_registration(void *data)
1499 {
1500 int ret, err = -1;
1501 struct cds_wfq_node *node;
1502 struct ust_command *ust_cmd = NULL;
1503 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1504 struct ust_reg_wait_queue wait_queue = {
1505 .count = 0,
1506 };
1507
1508 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1509
1510 health_code_update();
1511
1512 CDS_INIT_LIST_HEAD(&wait_queue.head);
1513
1514 DBG("[thread] Dispatch UST command started");
1515
1516 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1517 health_code_update();
1518
1519 /* Atomically prepare the queue futex */
1520 futex_nto1_prepare(&ust_cmd_queue.futex);
1521
1522 do {
1523 struct ust_app *app = NULL;
1524 ust_cmd = NULL;
1525
1526 /*
1527 * Make sure we don't have node(s) that have hung up before receiving
1528 * the notify socket. This is to clean the list in order to avoid
1529 * memory leaks from notify socket that are never seen.
1530 */
1531 sanitize_wait_queue(&wait_queue);
1532
1533 health_code_update();
1534 /* Dequeue command for registration */
1535 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1536 if (node == NULL) {
1537 DBG("Woken up but nothing in the UST command queue");
1538 /* Continue thread execution */
1539 break;
1540 }
1541
1542 ust_cmd = caa_container_of(node, struct ust_command, node);
1543
1544 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1545 " gid:%d sock:%d name:%s (version %d.%d)",
1546 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1547 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1548 ust_cmd->sock, ust_cmd->reg_msg.name,
1549 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1550
1551 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1552 wait_node = zmalloc(sizeof(*wait_node));
1553 if (!wait_node) {
1554 PERROR("zmalloc wait_node dispatch");
1555 ret = close(ust_cmd->sock);
1556 if (ret < 0) {
1557 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1558 }
1559 lttng_fd_put(1, LTTNG_FD_APPS);
1560 free(ust_cmd);
1561 goto error;
1562 }
1563 CDS_INIT_LIST_HEAD(&wait_node->head);
1564
1565 /* Create application object if socket is CMD. */
1566 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1567 ust_cmd->sock);
1568 if (!wait_node->app) {
1569 ret = close(ust_cmd->sock);
1570 if (ret < 0) {
1571 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1572 }
1573 lttng_fd_put(1, LTTNG_FD_APPS);
1574 free(wait_node);
1575 free(ust_cmd);
1576 continue;
1577 }
1578 /*
1579 * Add application to the wait queue so we can set the notify
1580 * socket before putting this object in the global ht.
1581 */
1582 cds_list_add(&wait_node->head, &wait_queue.head);
1583 wait_queue.count++;
1584
1585 free(ust_cmd);
1586 /*
1587 * We have to continue here since we don't have the notify
1588 * socket and the application MUST be added to the hash table
1589 * only at that moment.
1590 */
1591 continue;
1592 } else {
1593 /*
1594 * Look for the application in the local wait queue and set the
1595 * notify socket if found.
1596 */
1597 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1598 &wait_queue.head, head) {
1599 health_code_update();
1600 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1601 wait_node->app->notify_sock = ust_cmd->sock;
1602 cds_list_del(&wait_node->head);
1603 wait_queue.count--;
1604 app = wait_node->app;
1605 free(wait_node);
1606 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1607 break;
1608 }
1609 }
1610
1611 /*
1612 * With no application at this stage the received socket is
1613 * basically useless so close it before we free the cmd data
1614 * structure for good.
1615 */
1616 if (!app) {
1617 ret = close(ust_cmd->sock);
1618 if (ret < 0) {
1619 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1620 }
1621 lttng_fd_put(1, LTTNG_FD_APPS);
1622 }
1623 free(ust_cmd);
1624 }
1625
1626 if (app) {
1627 /*
1628 * @session_lock_list
1629 *
1630 * Lock the global session list so from the register up to the
1631 * registration done message, no thread can see the application
1632 * and change its state.
1633 */
1634 session_lock_list();
1635 rcu_read_lock();
1636
1637 /*
1638 * Add application to the global hash table. This needs to be
1639 * done before the update to the UST registry can locate the
1640 * application.
1641 */
1642 ust_app_add(app);
1643
1644 /* Set app version. This call will print an error if needed. */
1645 (void) ust_app_version(app);
1646
1647 /* Send notify socket through the notify pipe. */
1648 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1649 app->notify_sock);
1650 if (ret < 0) {
1651 rcu_read_unlock();
1652 session_unlock_list();
1653 /* No notify thread, stop the UST tracing. */
1654 goto error;
1655 }
1656
1657 /*
1658 * Update newly registered application with the tracing
1659 * registry info already enabled information.
1660 */
1661 update_ust_app(app->sock);
1662
1663 /*
1664 * Don't care about return value. Let the manage apps threads
1665 * handle app unregistration upon socket close.
1666 */
1667 (void) ust_app_register_done(app->sock);
1668
1669 /*
1670 * Even if the application socket has been closed, send the app
1671 * to the thread and unregistration will take place at that
1672 * place.
1673 */
1674 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1675 if (ret < 0) {
1676 rcu_read_unlock();
1677 session_unlock_list();
1678 /* No apps. thread, stop the UST tracing. */
1679 goto error;
1680 }
1681
1682 rcu_read_unlock();
1683 session_unlock_list();
1684 }
1685 } while (node != NULL);
1686
1687 health_poll_entry();
1688 /* Futex wait on queue. Blocking call on futex() */
1689 futex_nto1_wait(&ust_cmd_queue.futex);
1690 health_poll_exit();
1691 }
1692 /* Normal exit, no error */
1693 err = 0;
1694
1695 error:
1696 /* Clean up wait queue. */
1697 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1698 &wait_queue.head, head) {
1699 cds_list_del(&wait_node->head);
1700 wait_queue.count--;
1701 free(wait_node);
1702 }
1703
1704 DBG("Dispatch thread dying");
1705 if (err) {
1706 health_error();
1707 ERR("Health error occurred in %s", __func__);
1708 }
1709 health_unregister();
1710 return NULL;
1711 }
1712
1713 /*
1714 * This thread manage application registration.
1715 */
1716 static void *thread_registration_apps(void *data)
1717 {
1718 int sock = -1, i, ret, pollfd, err = -1;
1719 uint32_t revents, nb_fd;
1720 struct lttng_poll_event events;
1721 /*
1722 * Get allocated in this thread, enqueued to a global queue, dequeued and
1723 * freed in the manage apps thread.
1724 */
1725 struct ust_command *ust_cmd = NULL;
1726
1727 DBG("[thread] Manage application registration started");
1728
1729 health_register(HEALTH_TYPE_APP_REG);
1730
1731 if (testpoint(thread_registration_apps)) {
1732 goto error_testpoint;
1733 }
1734
1735 ret = lttcomm_listen_unix_sock(apps_sock);
1736 if (ret < 0) {
1737 goto error_listen;
1738 }
1739
1740 /*
1741 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1742 * more will be added to this poll set.
1743 */
1744 ret = sessiond_set_thread_pollset(&events, 2);
1745 if (ret < 0) {
1746 goto error_create_poll;
1747 }
1748
1749 /* Add the application registration socket */
1750 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1751 if (ret < 0) {
1752 goto error_poll_add;
1753 }
1754
1755 /* Notify all applications to register */
1756 ret = notify_ust_apps(1);
1757 if (ret < 0) {
1758 ERR("Failed to notify applications or create the wait shared memory.\n"
1759 "Execution continues but there might be problem for already\n"
1760 "running applications that wishes to register.");
1761 }
1762
1763 while (1) {
1764 DBG("Accepting application registration");
1765
1766 /* Inifinite blocking call, waiting for transmission */
1767 restart:
1768 health_poll_entry();
1769 ret = lttng_poll_wait(&events, -1);
1770 health_poll_exit();
1771 if (ret < 0) {
1772 /*
1773 * Restart interrupted system call.
1774 */
1775 if (errno == EINTR) {
1776 goto restart;
1777 }
1778 goto error;
1779 }
1780
1781 nb_fd = ret;
1782
1783 for (i = 0; i < nb_fd; i++) {
1784 health_code_update();
1785
1786 /* Fetch once the poll data */
1787 revents = LTTNG_POLL_GETEV(&events, i);
1788 pollfd = LTTNG_POLL_GETFD(&events, i);
1789
1790 /* Thread quit pipe has been closed. Killing thread. */
1791 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1792 if (ret) {
1793 err = 0;
1794 goto exit;
1795 }
1796
1797 /* Event on the registration socket */
1798 if (pollfd == apps_sock) {
1799 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1800 ERR("Register apps socket poll error");
1801 goto error;
1802 } else if (revents & LPOLLIN) {
1803 sock = lttcomm_accept_unix_sock(apps_sock);
1804 if (sock < 0) {
1805 goto error;
1806 }
1807
1808 /*
1809 * Set the CLOEXEC flag. Return code is useless because
1810 * either way, the show must go on.
1811 */
1812 (void) utils_set_fd_cloexec(sock);
1813
1814 /* Create UST registration command for enqueuing */
1815 ust_cmd = zmalloc(sizeof(struct ust_command));
1816 if (ust_cmd == NULL) {
1817 PERROR("ust command zmalloc");
1818 goto error;
1819 }
1820
1821 /*
1822 * Using message-based transmissions to ensure we don't
1823 * have to deal with partially received messages.
1824 */
1825 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1826 if (ret < 0) {
1827 ERR("Exhausted file descriptors allowed for applications.");
1828 free(ust_cmd);
1829 ret = close(sock);
1830 if (ret) {
1831 PERROR("close");
1832 }
1833 sock = -1;
1834 continue;
1835 }
1836
1837 health_code_update();
1838 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1839 if (ret < 0) {
1840 free(ust_cmd);
1841 /* Close socket of the application. */
1842 ret = close(sock);
1843 if (ret) {
1844 PERROR("close");
1845 }
1846 lttng_fd_put(LTTNG_FD_APPS, 1);
1847 sock = -1;
1848 continue;
1849 }
1850 health_code_update();
1851
1852 ust_cmd->sock = sock;
1853 sock = -1;
1854
1855 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1856 " gid:%d sock:%d name:%s (version %d.%d)",
1857 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1858 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1859 ust_cmd->sock, ust_cmd->reg_msg.name,
1860 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1861
1862 /*
1863 * Lock free enqueue the registration request. The red pill
1864 * has been taken! This apps will be part of the *system*.
1865 */
1866 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1867
1868 /*
1869 * Wake the registration queue futex. Implicit memory
1870 * barrier with the exchange in cds_wfq_enqueue.
1871 */
1872 futex_nto1_wake(&ust_cmd_queue.futex);
1873 }
1874 }
1875 }
1876 }
1877
1878 exit:
1879 error:
1880 if (err) {
1881 health_error();
1882 ERR("Health error occurred in %s", __func__);
1883 }
1884
1885 /* Notify that the registration thread is gone */
1886 notify_ust_apps(0);
1887
1888 if (apps_sock >= 0) {
1889 ret = close(apps_sock);
1890 if (ret) {
1891 PERROR("close");
1892 }
1893 }
1894 if (sock >= 0) {
1895 ret = close(sock);
1896 if (ret) {
1897 PERROR("close");
1898 }
1899 lttng_fd_put(LTTNG_FD_APPS, 1);
1900 }
1901 unlink(apps_unix_sock_path);
1902
1903 error_poll_add:
1904 lttng_poll_clean(&events);
1905 error_listen:
1906 error_create_poll:
1907 error_testpoint:
1908 DBG("UST Registration thread cleanup complete");
1909 health_unregister();
1910
1911 return NULL;
1912 }
1913
1914 /*
1915 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1916 * exec or it will fails.
1917 */
1918 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1919 {
1920 int ret, clock_ret;
1921 struct timespec timeout;
1922
1923 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1924 consumer_data->consumer_thread_is_ready = 0;
1925
1926 /* Setup pthread condition */
1927 ret = pthread_condattr_init(&consumer_data->condattr);
1928 if (ret != 0) {
1929 errno = ret;
1930 PERROR("pthread_condattr_init consumer data");
1931 goto error;
1932 }
1933
1934 /*
1935 * Set the monotonic clock in order to make sure we DO NOT jump in time
1936 * between the clock_gettime() call and the timedwait call. See bug #324
1937 * for a more details and how we noticed it.
1938 */
1939 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1940 if (ret != 0) {
1941 errno = ret;
1942 PERROR("pthread_condattr_setclock consumer data");
1943 goto error;
1944 }
1945
1946 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1947 if (ret != 0) {
1948 errno = ret;
1949 PERROR("pthread_cond_init consumer data");
1950 goto error;
1951 }
1952
1953 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1954 consumer_data);
1955 if (ret != 0) {
1956 PERROR("pthread_create consumer");
1957 ret = -1;
1958 goto error;
1959 }
1960
1961 /* We are about to wait on a pthread condition */
1962 pthread_mutex_lock(&consumer_data->cond_mutex);
1963
1964 /* Get time for sem_timedwait absolute timeout */
1965 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1966 /*
1967 * Set the timeout for the condition timed wait even if the clock gettime
1968 * call fails since we might loop on that call and we want to avoid to
1969 * increment the timeout too many times.
1970 */
1971 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1972
1973 /*
1974 * The following loop COULD be skipped in some conditions so this is why we
1975 * set ret to 0 in order to make sure at least one round of the loop is
1976 * done.
1977 */
1978 ret = 0;
1979
1980 /*
1981 * Loop until the condition is reached or when a timeout is reached. Note
1982 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1983 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1984 * possible. This loop does not take any chances and works with both of
1985 * them.
1986 */
1987 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1988 if (clock_ret < 0) {
1989 PERROR("clock_gettime spawn consumer");
1990 /* Infinite wait for the consumerd thread to be ready */
1991 ret = pthread_cond_wait(&consumer_data->cond,
1992 &consumer_data->cond_mutex);
1993 } else {
1994 ret = pthread_cond_timedwait(&consumer_data->cond,
1995 &consumer_data->cond_mutex, &timeout);
1996 }
1997 }
1998
1999 /* Release the pthread condition */
2000 pthread_mutex_unlock(&consumer_data->cond_mutex);
2001
2002 if (ret != 0) {
2003 errno = ret;
2004 if (ret == ETIMEDOUT) {
2005 /*
2006 * Call has timed out so we kill the kconsumerd_thread and return
2007 * an error.
2008 */
2009 ERR("Condition timed out. The consumer thread was never ready."
2010 " Killing it");
2011 ret = pthread_cancel(consumer_data->thread);
2012 if (ret < 0) {
2013 PERROR("pthread_cancel consumer thread");
2014 }
2015 } else {
2016 PERROR("pthread_cond_wait failed consumer thread");
2017 }
2018 goto error;
2019 }
2020
2021 pthread_mutex_lock(&consumer_data->pid_mutex);
2022 if (consumer_data->pid == 0) {
2023 ERR("Consumerd did not start");
2024 pthread_mutex_unlock(&consumer_data->pid_mutex);
2025 goto error;
2026 }
2027 pthread_mutex_unlock(&consumer_data->pid_mutex);
2028
2029 return 0;
2030
2031 error:
2032 return ret;
2033 }
2034
2035 /*
2036 * Join consumer thread
2037 */
2038 static int join_consumer_thread(struct consumer_data *consumer_data)
2039 {
2040 void *status;
2041
2042 /* Consumer pid must be a real one. */
2043 if (consumer_data->pid > 0) {
2044 int ret;
2045 ret = kill(consumer_data->pid, SIGTERM);
2046 if (ret) {
2047 ERR("Error killing consumer daemon");
2048 return ret;
2049 }
2050 return pthread_join(consumer_data->thread, &status);
2051 } else {
2052 return 0;
2053 }
2054 }
2055
2056 /*
2057 * Fork and exec a consumer daemon (consumerd).
2058 *
2059 * Return pid if successful else -1.
2060 */
2061 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2062 {
2063 int ret;
2064 pid_t pid;
2065 const char *consumer_to_use;
2066 const char *verbosity;
2067 struct stat st;
2068
2069 DBG("Spawning consumerd");
2070
2071 pid = fork();
2072 if (pid == 0) {
2073 /*
2074 * Exec consumerd.
2075 */
2076 if (opt_verbose_consumer) {
2077 verbosity = "--verbose";
2078 } else {
2079 verbosity = "--quiet";
2080 }
2081 switch (consumer_data->type) {
2082 case LTTNG_CONSUMER_KERNEL:
2083 /*
2084 * Find out which consumerd to execute. We will first try the
2085 * 64-bit path, then the sessiond's installation directory, and
2086 * fallback on the 32-bit one,
2087 */
2088 DBG3("Looking for a kernel consumer at these locations:");
2089 DBG3(" 1) %s", consumerd64_bin);
2090 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2091 DBG3(" 3) %s", consumerd32_bin);
2092 if (stat(consumerd64_bin, &st) == 0) {
2093 DBG3("Found location #1");
2094 consumer_to_use = consumerd64_bin;
2095 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2096 DBG3("Found location #2");
2097 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2098 } else if (stat(consumerd32_bin, &st) == 0) {
2099 DBG3("Found location #3");
2100 consumer_to_use = consumerd32_bin;
2101 } else {
2102 DBG("Could not find any valid consumerd executable");
2103 break;
2104 }
2105 DBG("Using kernel consumer at: %s", consumer_to_use);
2106 execl(consumer_to_use,
2107 "lttng-consumerd", verbosity, "-k",
2108 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2109 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2110 NULL);
2111 break;
2112 case LTTNG_CONSUMER64_UST:
2113 {
2114 char *tmpnew = NULL;
2115
2116 if (consumerd64_libdir[0] != '\0') {
2117 char *tmp;
2118 size_t tmplen;
2119
2120 tmp = getenv("LD_LIBRARY_PATH");
2121 if (!tmp) {
2122 tmp = "";
2123 }
2124 tmplen = strlen("LD_LIBRARY_PATH=")
2125 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2126 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2127 if (!tmpnew) {
2128 ret = -ENOMEM;
2129 goto error;
2130 }
2131 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2132 strcat(tmpnew, consumerd64_libdir);
2133 if (tmp[0] != '\0') {
2134 strcat(tmpnew, ":");
2135 strcat(tmpnew, tmp);
2136 }
2137 ret = putenv(tmpnew);
2138 if (ret) {
2139 ret = -errno;
2140 free(tmpnew);
2141 goto error;
2142 }
2143 }
2144 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2145 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2146 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2147 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2148 NULL);
2149 if (consumerd64_libdir[0] != '\0') {
2150 free(tmpnew);
2151 }
2152 if (ret) {
2153 goto error;
2154 }
2155 break;
2156 }
2157 case LTTNG_CONSUMER32_UST:
2158 {
2159 char *tmpnew = NULL;
2160
2161 if (consumerd32_libdir[0] != '\0') {
2162 char *tmp;
2163 size_t tmplen;
2164
2165 tmp = getenv("LD_LIBRARY_PATH");
2166 if (!tmp) {
2167 tmp = "";
2168 }
2169 tmplen = strlen("LD_LIBRARY_PATH=")
2170 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2171 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2172 if (!tmpnew) {
2173 ret = -ENOMEM;
2174 goto error;
2175 }
2176 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2177 strcat(tmpnew, consumerd32_libdir);
2178 if (tmp[0] != '\0') {
2179 strcat(tmpnew, ":");
2180 strcat(tmpnew, tmp);
2181 }
2182 ret = putenv(tmpnew);
2183 if (ret) {
2184 ret = -errno;
2185 free(tmpnew);
2186 goto error;
2187 }
2188 }
2189 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2190 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2191 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2192 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2193 NULL);
2194 if (consumerd32_libdir[0] != '\0') {
2195 free(tmpnew);
2196 }
2197 if (ret) {
2198 goto error;
2199 }
2200 break;
2201 }
2202 default:
2203 PERROR("unknown consumer type");
2204 exit(EXIT_FAILURE);
2205 }
2206 if (errno != 0) {
2207 PERROR("kernel start consumer exec");
2208 }
2209 exit(EXIT_FAILURE);
2210 } else if (pid > 0) {
2211 ret = pid;
2212 } else {
2213 PERROR("start consumer fork");
2214 ret = -errno;
2215 }
2216 error:
2217 return ret;
2218 }
2219
2220 /*
2221 * Spawn the consumerd daemon and session daemon thread.
2222 */
2223 static int start_consumerd(struct consumer_data *consumer_data)
2224 {
2225 int ret;
2226
2227 /*
2228 * Set the listen() state on the socket since there is a possible race
2229 * between the exec() of the consumer daemon and this call if place in the
2230 * consumer thread. See bug #366 for more details.
2231 */
2232 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2233 if (ret < 0) {
2234 goto error;
2235 }
2236
2237 pthread_mutex_lock(&consumer_data->pid_mutex);
2238 if (consumer_data->pid != 0) {
2239 pthread_mutex_unlock(&consumer_data->pid_mutex);
2240 goto end;
2241 }
2242
2243 ret = spawn_consumerd(consumer_data);
2244 if (ret < 0) {
2245 ERR("Spawning consumerd failed");
2246 pthread_mutex_unlock(&consumer_data->pid_mutex);
2247 goto error;
2248 }
2249
2250 /* Setting up the consumer_data pid */
2251 consumer_data->pid = ret;
2252 DBG2("Consumer pid %d", consumer_data->pid);
2253 pthread_mutex_unlock(&consumer_data->pid_mutex);
2254
2255 DBG2("Spawning consumer control thread");
2256 ret = spawn_consumer_thread(consumer_data);
2257 if (ret < 0) {
2258 ERR("Fatal error spawning consumer control thread");
2259 goto error;
2260 }
2261
2262 end:
2263 return 0;
2264
2265 error:
2266 /* Cleanup already created sockets on error. */
2267 if (consumer_data->err_sock >= 0) {
2268 int err;
2269
2270 err = close(consumer_data->err_sock);
2271 if (err < 0) {
2272 PERROR("close consumer data error socket");
2273 }
2274 }
2275 return ret;
2276 }
2277
2278 /*
2279 * Compute health status of each consumer. If one of them is zero (bad
2280 * state), we return 0.
2281 */
2282 static int check_consumer_health(void)
2283 {
2284 int ret;
2285
2286 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2287
2288 DBG3("Health consumer check %d", ret);
2289
2290 return ret;
2291 }
2292
2293 /*
2294 * Setup necessary data for kernel tracer action.
2295 */
2296 static int init_kernel_tracer(void)
2297 {
2298 int ret;
2299
2300 /* Modprobe lttng kernel modules */
2301 ret = modprobe_lttng_control();
2302 if (ret < 0) {
2303 goto error;
2304 }
2305
2306 /* Open debugfs lttng */
2307 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2308 if (kernel_tracer_fd < 0) {
2309 DBG("Failed to open %s", module_proc_lttng);
2310 ret = -1;
2311 goto error_open;
2312 }
2313
2314 /* Validate kernel version */
2315 ret = kernel_validate_version(kernel_tracer_fd);
2316 if (ret < 0) {
2317 goto error_version;
2318 }
2319
2320 ret = modprobe_lttng_data();
2321 if (ret < 0) {
2322 goto error_modules;
2323 }
2324
2325 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2326 return 0;
2327
2328 error_version:
2329 modprobe_remove_lttng_control();
2330 ret = close(kernel_tracer_fd);
2331 if (ret) {
2332 PERROR("close");
2333 }
2334 kernel_tracer_fd = -1;
2335 return LTTNG_ERR_KERN_VERSION;
2336
2337 error_modules:
2338 ret = close(kernel_tracer_fd);
2339 if (ret) {
2340 PERROR("close");
2341 }
2342
2343 error_open:
2344 modprobe_remove_lttng_control();
2345
2346 error:
2347 WARN("No kernel tracer available");
2348 kernel_tracer_fd = -1;
2349 if (!is_root) {
2350 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2351 } else {
2352 return LTTNG_ERR_KERN_NA;
2353 }
2354 }
2355
2356
2357 /*
2358 * Copy consumer output from the tracing session to the domain session. The
2359 * function also applies the right modification on a per domain basis for the
2360 * trace files destination directory.
2361 *
2362 * Should *NOT* be called with RCU read-side lock held.
2363 */
2364 static int copy_session_consumer(int domain, struct ltt_session *session)
2365 {
2366 int ret;
2367 const char *dir_name;
2368 struct consumer_output *consumer;
2369
2370 assert(session);
2371 assert(session->consumer);
2372
2373 switch (domain) {
2374 case LTTNG_DOMAIN_KERNEL:
2375 DBG3("Copying tracing session consumer output in kernel session");
2376 /*
2377 * XXX: We should audit the session creation and what this function
2378 * does "extra" in order to avoid a destroy since this function is used
2379 * in the domain session creation (kernel and ust) only. Same for UST
2380 * domain.
2381 */
2382 if (session->kernel_session->consumer) {
2383 consumer_destroy_output(session->kernel_session->consumer);
2384 }
2385 session->kernel_session->consumer =
2386 consumer_copy_output(session->consumer);
2387 /* Ease our life a bit for the next part */
2388 consumer = session->kernel_session->consumer;
2389 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2390 break;
2391 case LTTNG_DOMAIN_UST:
2392 DBG3("Copying tracing session consumer output in UST session");
2393 if (session->ust_session->consumer) {
2394 consumer_destroy_output(session->ust_session->consumer);
2395 }
2396 session->ust_session->consumer =
2397 consumer_copy_output(session->consumer);
2398 /* Ease our life a bit for the next part */
2399 consumer = session->ust_session->consumer;
2400 dir_name = DEFAULT_UST_TRACE_DIR;
2401 break;
2402 default:
2403 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2404 goto error;
2405 }
2406
2407 /* Append correct directory to subdir */
2408 strncat(consumer->subdir, dir_name,
2409 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2410 DBG3("Copy session consumer subdir %s", consumer->subdir);
2411
2412 ret = LTTNG_OK;
2413
2414 error:
2415 return ret;
2416 }
2417
2418 /*
2419 * Create an UST session and add it to the session ust list.
2420 *
2421 * Should *NOT* be called with RCU read-side lock held.
2422 */
2423 static int create_ust_session(struct ltt_session *session,
2424 struct lttng_domain *domain)
2425 {
2426 int ret;
2427 struct ltt_ust_session *lus = NULL;
2428
2429 assert(session);
2430 assert(domain);
2431 assert(session->consumer);
2432
2433 switch (domain->type) {
2434 case LTTNG_DOMAIN_UST:
2435 break;
2436 default:
2437 ERR("Unknown UST domain on create session %d", domain->type);
2438 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2439 goto error;
2440 }
2441
2442 DBG("Creating UST session");
2443
2444 lus = trace_ust_create_session(session->id);
2445 if (lus == NULL) {
2446 ret = LTTNG_ERR_UST_SESS_FAIL;
2447 goto error;
2448 }
2449
2450 lus->uid = session->uid;
2451 lus->gid = session->gid;
2452 lus->output_traces = session->output_traces;
2453 lus->snapshot_mode = session->snapshot_mode;
2454 session->ust_session = lus;
2455
2456 /* Copy session output to the newly created UST session */
2457 ret = copy_session_consumer(domain->type, session);
2458 if (ret != LTTNG_OK) {
2459 goto error;
2460 }
2461
2462 return LTTNG_OK;
2463
2464 error:
2465 free(lus);
2466 session->ust_session = NULL;
2467 return ret;
2468 }
2469
2470 /*
2471 * Create a kernel tracer session then create the default channel.
2472 */
2473 static int create_kernel_session(struct ltt_session *session)
2474 {
2475 int ret;
2476
2477 DBG("Creating kernel session");
2478
2479 ret = kernel_create_session(session, kernel_tracer_fd);
2480 if (ret < 0) {
2481 ret = LTTNG_ERR_KERN_SESS_FAIL;
2482 goto error;
2483 }
2484
2485 /* Code flow safety */
2486 assert(session->kernel_session);
2487
2488 /* Copy session output to the newly created Kernel session */
2489 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2490 if (ret != LTTNG_OK) {
2491 goto error;
2492 }
2493
2494 /* Create directory(ies) on local filesystem. */
2495 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2496 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2497 ret = run_as_mkdir_recursive(
2498 session->kernel_session->consumer->dst.trace_path,
2499 S_IRWXU | S_IRWXG, session->uid, session->gid);
2500 if (ret < 0) {
2501 if (ret != -EEXIST) {
2502 ERR("Trace directory creation error");
2503 goto error;
2504 }
2505 }
2506 }
2507
2508 session->kernel_session->uid = session->uid;
2509 session->kernel_session->gid = session->gid;
2510 session->kernel_session->output_traces = session->output_traces;
2511 session->kernel_session->snapshot_mode = session->snapshot_mode;
2512
2513 return LTTNG_OK;
2514
2515 error:
2516 trace_kernel_destroy_session(session->kernel_session);
2517 session->kernel_session = NULL;
2518 return ret;
2519 }
2520
2521 /*
2522 * Count number of session permitted by uid/gid.
2523 */
2524 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2525 {
2526 unsigned int i = 0;
2527 struct ltt_session *session;
2528
2529 DBG("Counting number of available session for UID %d GID %d",
2530 uid, gid);
2531 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2532 /*
2533 * Only list the sessions the user can control.
2534 */
2535 if (!session_access_ok(session, uid, gid)) {
2536 continue;
2537 }
2538 i++;
2539 }
2540 return i;
2541 }
2542
2543 /*
2544 * Process the command requested by the lttng client within the command
2545 * context structure. This function make sure that the return structure (llm)
2546 * is set and ready for transmission before returning.
2547 *
2548 * Return any error encountered or 0 for success.
2549 *
2550 * "sock" is only used for special-case var. len data.
2551 *
2552 * Should *NOT* be called with RCU read-side lock held.
2553 */
2554 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2555 int *sock_error)
2556 {
2557 int ret = LTTNG_OK;
2558 int need_tracing_session = 1;
2559 int need_domain;
2560
2561 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2562
2563 *sock_error = 0;
2564
2565 switch (cmd_ctx->lsm->cmd_type) {
2566 case LTTNG_CREATE_SESSION:
2567 case LTTNG_CREATE_SESSION_SNAPSHOT:
2568 case LTTNG_DESTROY_SESSION:
2569 case LTTNG_LIST_SESSIONS:
2570 case LTTNG_LIST_DOMAINS:
2571 case LTTNG_START_TRACE:
2572 case LTTNG_STOP_TRACE:
2573 case LTTNG_DATA_PENDING:
2574 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2575 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2576 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2577 case LTTNG_SNAPSHOT_RECORD:
2578 need_domain = 0;
2579 break;
2580 default:
2581 need_domain = 1;
2582 }
2583
2584 if (opt_no_kernel && need_domain
2585 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2586 if (!is_root) {
2587 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2588 } else {
2589 ret = LTTNG_ERR_KERN_NA;
2590 }
2591 goto error;
2592 }
2593
2594 /* Deny register consumer if we already have a spawned consumer. */
2595 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2596 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2597 if (kconsumer_data.pid > 0) {
2598 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2599 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2600 goto error;
2601 }
2602 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2603 }
2604
2605 /*
2606 * Check for command that don't needs to allocate a returned payload. We do
2607 * this here so we don't have to make the call for no payload at each
2608 * command.
2609 */
2610 switch(cmd_ctx->lsm->cmd_type) {
2611 case LTTNG_LIST_SESSIONS:
2612 case LTTNG_LIST_TRACEPOINTS:
2613 case LTTNG_LIST_TRACEPOINT_FIELDS:
2614 case LTTNG_LIST_DOMAINS:
2615 case LTTNG_LIST_CHANNELS:
2616 case LTTNG_LIST_EVENTS:
2617 break;
2618 default:
2619 /* Setup lttng message with no payload */
2620 ret = setup_lttng_msg(cmd_ctx, 0);
2621 if (ret < 0) {
2622 /* This label does not try to unlock the session */
2623 goto init_setup_error;
2624 }
2625 }
2626
2627 /* Commands that DO NOT need a session. */
2628 switch (cmd_ctx->lsm->cmd_type) {
2629 case LTTNG_CREATE_SESSION:
2630 case LTTNG_CREATE_SESSION_SNAPSHOT:
2631 case LTTNG_CALIBRATE:
2632 case LTTNG_LIST_SESSIONS:
2633 case LTTNG_LIST_TRACEPOINTS:
2634 case LTTNG_LIST_TRACEPOINT_FIELDS:
2635 need_tracing_session = 0;
2636 break;
2637 default:
2638 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2639 /*
2640 * We keep the session list lock across _all_ commands
2641 * for now, because the per-session lock does not
2642 * handle teardown properly.
2643 */
2644 session_lock_list();
2645 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2646 if (cmd_ctx->session == NULL) {
2647 ret = LTTNG_ERR_SESS_NOT_FOUND;
2648 goto error;
2649 } else {
2650 /* Acquire lock for the session */
2651 session_lock(cmd_ctx->session);
2652 }
2653 break;
2654 }
2655
2656 if (!need_domain) {
2657 goto skip_domain;
2658 }
2659
2660 /*
2661 * Check domain type for specific "pre-action".
2662 */
2663 switch (cmd_ctx->lsm->domain.type) {
2664 case LTTNG_DOMAIN_KERNEL:
2665 if (!is_root) {
2666 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2667 goto error;
2668 }
2669
2670 /* Kernel tracer check */
2671 if (kernel_tracer_fd == -1) {
2672 /* Basically, load kernel tracer modules */
2673 ret = init_kernel_tracer();
2674 if (ret != 0) {
2675 goto error;
2676 }
2677 }
2678
2679 /* Consumer is in an ERROR state. Report back to client */
2680 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2681 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2682 goto error;
2683 }
2684
2685 /* Need a session for kernel command */
2686 if (need_tracing_session) {
2687 if (cmd_ctx->session->kernel_session == NULL) {
2688 ret = create_kernel_session(cmd_ctx->session);
2689 if (ret < 0) {
2690 ret = LTTNG_ERR_KERN_SESS_FAIL;
2691 goto error;
2692 }
2693 }
2694
2695 /* Start the kernel consumer daemon */
2696 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2697 if (kconsumer_data.pid == 0 &&
2698 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2699 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2700 ret = start_consumerd(&kconsumer_data);
2701 if (ret < 0) {
2702 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2703 goto error;
2704 }
2705 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2706 } else {
2707 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2708 }
2709
2710 /*
2711 * The consumer was just spawned so we need to add the socket to
2712 * the consumer output of the session if exist.
2713 */
2714 ret = consumer_create_socket(&kconsumer_data,
2715 cmd_ctx->session->kernel_session->consumer);
2716 if (ret < 0) {
2717 goto error;
2718 }
2719 }
2720
2721 break;
2722 case LTTNG_DOMAIN_UST:
2723 {
2724 if (!ust_app_supported()) {
2725 ret = LTTNG_ERR_NO_UST;
2726 goto error;
2727 }
2728 /* Consumer is in an ERROR state. Report back to client */
2729 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2730 ret = LTTNG_ERR_NO_USTCONSUMERD;
2731 goto error;
2732 }
2733
2734 if (need_tracing_session) {
2735 /* Create UST session if none exist. */
2736 if (cmd_ctx->session->ust_session == NULL) {
2737 ret = create_ust_session(cmd_ctx->session,
2738 &cmd_ctx->lsm->domain);
2739 if (ret != LTTNG_OK) {
2740 goto error;
2741 }
2742 }
2743
2744 /* Start the UST consumer daemons */
2745 /* 64-bit */
2746 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2747 if (consumerd64_bin[0] != '\0' &&
2748 ustconsumer64_data.pid == 0 &&
2749 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2750 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2751 ret = start_consumerd(&ustconsumer64_data);
2752 if (ret < 0) {
2753 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2754 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2755 goto error;
2756 }
2757
2758 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2759 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2760 } else {
2761 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2762 }
2763
2764 /*
2765 * Setup socket for consumer 64 bit. No need for atomic access
2766 * since it was set above and can ONLY be set in this thread.
2767 */
2768 ret = consumer_create_socket(&ustconsumer64_data,
2769 cmd_ctx->session->ust_session->consumer);
2770 if (ret < 0) {
2771 goto error;
2772 }
2773
2774 /* 32-bit */
2775 if (consumerd32_bin[0] != '\0' &&
2776 ustconsumer32_data.pid == 0 &&
2777 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2778 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2779 ret = start_consumerd(&ustconsumer32_data);
2780 if (ret < 0) {
2781 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2782 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2783 goto error;
2784 }
2785
2786 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2787 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2788 } else {
2789 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2790 }
2791
2792 /*
2793 * Setup socket for consumer 64 bit. No need for atomic access
2794 * since it was set above and can ONLY be set in this thread.
2795 */
2796 ret = consumer_create_socket(&ustconsumer32_data,
2797 cmd_ctx->session->ust_session->consumer);
2798 if (ret < 0) {
2799 goto error;
2800 }
2801 }
2802 break;
2803 }
2804 default:
2805 break;
2806 }
2807 skip_domain:
2808
2809 /* Validate consumer daemon state when start/stop trace command */
2810 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2811 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2812 switch (cmd_ctx->lsm->domain.type) {
2813 case LTTNG_DOMAIN_UST:
2814 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2815 ret = LTTNG_ERR_NO_USTCONSUMERD;
2816 goto error;
2817 }
2818 break;
2819 case LTTNG_DOMAIN_KERNEL:
2820 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2821 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2822 goto error;
2823 }
2824 break;
2825 }
2826 }
2827
2828 /*
2829 * Check that the UID or GID match that of the tracing session.
2830 * The root user can interact with all sessions.
2831 */
2832 if (need_tracing_session) {
2833 if (!session_access_ok(cmd_ctx->session,
2834 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2835 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2836 ret = LTTNG_ERR_EPERM;
2837 goto error;
2838 }
2839 }
2840
2841 /*
2842 * Send relayd information to consumer as soon as we have a domain and a
2843 * session defined.
2844 */
2845 if (cmd_ctx->session && need_domain) {
2846 /*
2847 * Setup relayd if not done yet. If the relayd information was already
2848 * sent to the consumer, this call will gracefully return.
2849 */
2850 ret = cmd_setup_relayd(cmd_ctx->session);
2851 if (ret != LTTNG_OK) {
2852 goto error;
2853 }
2854 }
2855
2856 /* Process by command type */
2857 switch (cmd_ctx->lsm->cmd_type) {
2858 case LTTNG_ADD_CONTEXT:
2859 {
2860 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2861 cmd_ctx->lsm->u.context.channel_name,
2862 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2863 break;
2864 }
2865 case LTTNG_DISABLE_CHANNEL:
2866 {
2867 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2868 cmd_ctx->lsm->u.disable.channel_name);
2869 break;
2870 }
2871 case LTTNG_DISABLE_EVENT:
2872 {
2873 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2874 cmd_ctx->lsm->u.disable.channel_name,
2875 cmd_ctx->lsm->u.disable.name);
2876 break;
2877 }
2878 case LTTNG_DISABLE_ALL_EVENT:
2879 {
2880 DBG("Disabling all events");
2881
2882 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2883 cmd_ctx->lsm->u.disable.channel_name);
2884 break;
2885 }
2886 case LTTNG_ENABLE_CHANNEL:
2887 {
2888 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2889 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2890 break;
2891 }
2892 case LTTNG_ENABLE_EVENT:
2893 {
2894 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2895 cmd_ctx->lsm->u.enable.channel_name,
2896 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2897 break;
2898 }
2899 case LTTNG_ENABLE_ALL_EVENT:
2900 {
2901 DBG("Enabling all events");
2902
2903 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2904 cmd_ctx->lsm->u.enable.channel_name,
2905 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2906 break;
2907 }
2908 case LTTNG_LIST_TRACEPOINTS:
2909 {
2910 struct lttng_event *events;
2911 ssize_t nb_events;
2912
2913 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2914 if (nb_events < 0) {
2915 /* Return value is a negative lttng_error_code. */
2916 ret = -nb_events;
2917 goto error;
2918 }
2919
2920 /*
2921 * Setup lttng message with payload size set to the event list size in
2922 * bytes and then copy list into the llm payload.
2923 */
2924 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2925 if (ret < 0) {
2926 free(events);
2927 goto setup_error;
2928 }
2929
2930 /* Copy event list into message payload */
2931 memcpy(cmd_ctx->llm->payload, events,
2932 sizeof(struct lttng_event) * nb_events);
2933
2934 free(events);
2935
2936 ret = LTTNG_OK;
2937 break;
2938 }
2939 case LTTNG_LIST_TRACEPOINT_FIELDS:
2940 {
2941 struct lttng_event_field *fields;
2942 ssize_t nb_fields;
2943
2944 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2945 &fields);
2946 if (nb_fields < 0) {
2947 /* Return value is a negative lttng_error_code. */
2948 ret = -nb_fields;
2949 goto error;
2950 }
2951
2952 /*
2953 * Setup lttng message with payload size set to the event list size in
2954 * bytes and then copy list into the llm payload.
2955 */
2956 ret = setup_lttng_msg(cmd_ctx,
2957 sizeof(struct lttng_event_field) * nb_fields);
2958 if (ret < 0) {
2959 free(fields);
2960 goto setup_error;
2961 }
2962
2963 /* Copy event list into message payload */
2964 memcpy(cmd_ctx->llm->payload, fields,
2965 sizeof(struct lttng_event_field) * nb_fields);
2966
2967 free(fields);
2968
2969 ret = LTTNG_OK;
2970 break;
2971 }
2972 case LTTNG_SET_CONSUMER_URI:
2973 {
2974 size_t nb_uri, len;
2975 struct lttng_uri *uris;
2976
2977 nb_uri = cmd_ctx->lsm->u.uri.size;
2978 len = nb_uri * sizeof(struct lttng_uri);
2979
2980 if (nb_uri == 0) {
2981 ret = LTTNG_ERR_INVALID;
2982 goto error;
2983 }
2984
2985 uris = zmalloc(len);
2986 if (uris == NULL) {
2987 ret = LTTNG_ERR_FATAL;
2988 goto error;
2989 }
2990
2991 /* Receive variable len data */
2992 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2993 ret = lttcomm_recv_unix_sock(sock, uris, len);
2994 if (ret <= 0) {
2995 DBG("No URIs received from client... continuing");
2996 *sock_error = 1;
2997 ret = LTTNG_ERR_SESSION_FAIL;
2998 free(uris);
2999 goto error;
3000 }
3001
3002 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3003 nb_uri, uris);
3004 if (ret != LTTNG_OK) {
3005 free(uris);
3006 goto error;
3007 }
3008
3009 /*
3010 * XXX: 0 means that this URI should be applied on the session. Should
3011 * be a DOMAIN enuam.
3012 */
3013 if (cmd_ctx->lsm->domain.type == 0) {
3014 /* Add the URI for the UST session if a consumer is present. */
3015 if (cmd_ctx->session->ust_session &&
3016 cmd_ctx->session->ust_session->consumer) {
3017 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3018 nb_uri, uris);
3019 } else if (cmd_ctx->session->kernel_session &&
3020 cmd_ctx->session->kernel_session->consumer) {
3021 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3022 cmd_ctx->session, nb_uri, uris);
3023 }
3024 }
3025
3026 free(uris);
3027
3028 break;
3029 }
3030 case LTTNG_START_TRACE:
3031 {
3032 ret = cmd_start_trace(cmd_ctx->session);
3033 break;
3034 }
3035 case LTTNG_STOP_TRACE:
3036 {
3037 ret = cmd_stop_trace(cmd_ctx->session);
3038 break;
3039 }
3040 case LTTNG_CREATE_SESSION:
3041 {
3042 size_t nb_uri, len;
3043 struct lttng_uri *uris = NULL;
3044
3045 nb_uri = cmd_ctx->lsm->u.uri.size;
3046 len = nb_uri * sizeof(struct lttng_uri);
3047
3048 if (nb_uri > 0) {
3049 uris = zmalloc(len);
3050 if (uris == NULL) {
3051 ret = LTTNG_ERR_FATAL;
3052 goto error;
3053 }
3054
3055 /* Receive variable len data */
3056 DBG("Waiting for %zu URIs from client ...", nb_uri);
3057 ret = lttcomm_recv_unix_sock(sock, uris, len);
3058 if (ret <= 0) {
3059 DBG("No URIs received from client... continuing");
3060 *sock_error = 1;
3061 ret = LTTNG_ERR_SESSION_FAIL;
3062 free(uris);
3063 goto error;
3064 }
3065
3066 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3067 DBG("Creating session with ONE network URI is a bad call");
3068 ret = LTTNG_ERR_SESSION_FAIL;
3069 free(uris);
3070 goto error;
3071 }
3072 }
3073
3074 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3075 &cmd_ctx->creds);
3076
3077 free(uris);
3078
3079 break;
3080 }
3081 case LTTNG_DESTROY_SESSION:
3082 {
3083 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3084
3085 /* Set session to NULL so we do not unlock it after free. */
3086 cmd_ctx->session = NULL;
3087 break;
3088 }
3089 case LTTNG_LIST_DOMAINS:
3090 {
3091 ssize_t nb_dom;
3092 struct lttng_domain *domains;
3093
3094 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3095 if (nb_dom < 0) {
3096 /* Return value is a negative lttng_error_code. */
3097 ret = -nb_dom;
3098 goto error;
3099 }
3100
3101 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3102 if (ret < 0) {
3103 free(domains);
3104 goto setup_error;
3105 }
3106
3107 /* Copy event list into message payload */
3108 memcpy(cmd_ctx->llm->payload, domains,
3109 nb_dom * sizeof(struct lttng_domain));
3110
3111 free(domains);
3112
3113 ret = LTTNG_OK;
3114 break;
3115 }
3116 case LTTNG_LIST_CHANNELS:
3117 {
3118 int nb_chan;
3119 struct lttng_channel *channels;
3120
3121 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3122 cmd_ctx->session, &channels);
3123 if (nb_chan < 0) {
3124 /* Return value is a negative lttng_error_code. */
3125 ret = -nb_chan;
3126 goto error;
3127 }
3128
3129 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3130 if (ret < 0) {
3131 free(channels);
3132 goto setup_error;
3133 }
3134
3135 /* Copy event list into message payload */
3136 memcpy(cmd_ctx->llm->payload, channels,
3137 nb_chan * sizeof(struct lttng_channel));
3138
3139 free(channels);
3140
3141 ret = LTTNG_OK;
3142 break;
3143 }
3144 case LTTNG_LIST_EVENTS:
3145 {
3146 ssize_t nb_event;
3147 struct lttng_event *events = NULL;
3148
3149 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3150 cmd_ctx->lsm->u.list.channel_name, &events);
3151 if (nb_event < 0) {
3152 /* Return value is a negative lttng_error_code. */
3153 ret = -nb_event;
3154 goto error;
3155 }
3156
3157 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3158 if (ret < 0) {
3159 free(events);
3160 goto setup_error;
3161 }
3162
3163 /* Copy event list into message payload */
3164 memcpy(cmd_ctx->llm->payload, events,
3165 nb_event * sizeof(struct lttng_event));
3166
3167 free(events);
3168
3169 ret = LTTNG_OK;
3170 break;
3171 }
3172 case LTTNG_LIST_SESSIONS:
3173 {
3174 unsigned int nr_sessions;
3175
3176 session_lock_list();
3177 nr_sessions = lttng_sessions_count(
3178 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3179 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3180
3181 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3182 if (ret < 0) {
3183 session_unlock_list();
3184 goto setup_error;
3185 }
3186
3187 /* Filled the session array */
3188 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3189 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3190 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3191
3192 session_unlock_list();
3193
3194 ret = LTTNG_OK;
3195 break;
3196 }
3197 case LTTNG_CALIBRATE:
3198 {
3199 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3200 &cmd_ctx->lsm->u.calibrate);
3201 break;
3202 }
3203 case LTTNG_REGISTER_CONSUMER:
3204 {
3205 struct consumer_data *cdata;
3206
3207 switch (cmd_ctx->lsm->domain.type) {
3208 case LTTNG_DOMAIN_KERNEL:
3209 cdata = &kconsumer_data;
3210 break;
3211 default:
3212 ret = LTTNG_ERR_UND;
3213 goto error;
3214 }
3215
3216 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3217 cmd_ctx->lsm->u.reg.path, cdata);
3218 break;
3219 }
3220 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3221 {
3222 struct lttng_filter_bytecode *bytecode;
3223
3224 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3225 ret = LTTNG_ERR_FILTER_INVAL;
3226 goto error;
3227 }
3228 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3229 ret = LTTNG_ERR_FILTER_INVAL;
3230 goto error;
3231 }
3232 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3233 if (!bytecode) {
3234 ret = LTTNG_ERR_FILTER_NOMEM;
3235 goto error;
3236 }
3237 /* Receive var. len. data */
3238 DBG("Receiving var len data from client ...");
3239 ret = lttcomm_recv_unix_sock(sock, bytecode,
3240 cmd_ctx->lsm->u.enable.bytecode_len);
3241 if (ret <= 0) {
3242 DBG("Nothing recv() from client var len data... continuing");
3243 *sock_error = 1;
3244 ret = LTTNG_ERR_FILTER_INVAL;
3245 goto error;
3246 }
3247
3248 if (bytecode->len + sizeof(*bytecode)
3249 != cmd_ctx->lsm->u.enable.bytecode_len) {
3250 free(bytecode);
3251 ret = LTTNG_ERR_FILTER_INVAL;
3252 goto error;
3253 }
3254
3255 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3256 cmd_ctx->lsm->u.enable.channel_name,
3257 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3258 break;
3259 }
3260 case LTTNG_DATA_PENDING:
3261 {
3262 ret = cmd_data_pending(cmd_ctx->session);
3263 break;
3264 }
3265 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3266 {
3267 struct lttcomm_lttng_output_id reply;
3268
3269 ret = cmd_snapshot_add_output(cmd_ctx->session,
3270 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3271 if (ret != LTTNG_OK) {
3272 goto error;
3273 }
3274
3275 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3276 if (ret < 0) {
3277 goto setup_error;
3278 }
3279
3280 /* Copy output list into message payload */
3281 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3282 ret = LTTNG_OK;
3283 break;
3284 }
3285 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3286 {
3287 ret = cmd_snapshot_del_output(cmd_ctx->session,
3288 &cmd_ctx->lsm->u.snapshot_output.output);
3289 break;
3290 }
3291 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3292 {
3293 ssize_t nb_output;
3294 struct lttng_snapshot_output *outputs = NULL;
3295
3296 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3297 if (nb_output < 0) {
3298 ret = -nb_output;
3299 goto error;
3300 }
3301
3302 ret = setup_lttng_msg(cmd_ctx,
3303 nb_output * sizeof(struct lttng_snapshot_output));
3304 if (ret < 0) {
3305 free(outputs);
3306 goto setup_error;
3307 }
3308
3309 if (outputs) {
3310 /* Copy output list into message payload */
3311 memcpy(cmd_ctx->llm->payload, outputs,
3312 nb_output * sizeof(struct lttng_snapshot_output));
3313 free(outputs);
3314 }
3315
3316 ret = LTTNG_OK;
3317 break;
3318 }
3319 case LTTNG_SNAPSHOT_RECORD:
3320 {
3321 ret = cmd_snapshot_record(cmd_ctx->session,
3322 &cmd_ctx->lsm->u.snapshot_record.output,
3323 cmd_ctx->lsm->u.snapshot_record.wait);
3324 break;
3325 }
3326 case LTTNG_CREATE_SESSION_SNAPSHOT:
3327 {
3328 size_t nb_uri, len;
3329 struct lttng_uri *uris = NULL;
3330
3331 nb_uri = cmd_ctx->lsm->u.uri.size;
3332 len = nb_uri * sizeof(struct lttng_uri);
3333
3334 if (nb_uri > 0) {
3335 uris = zmalloc(len);
3336 if (uris == NULL) {
3337 ret = LTTNG_ERR_FATAL;
3338 goto error;
3339 }
3340
3341 /* Receive variable len data */
3342 DBG("Waiting for %zu URIs from client ...", nb_uri);
3343 ret = lttcomm_recv_unix_sock(sock, uris, len);
3344 if (ret <= 0) {
3345 DBG("No URIs received from client... continuing");
3346 *sock_error = 1;
3347 ret = LTTNG_ERR_SESSION_FAIL;
3348 free(uris);
3349 goto error;
3350 }
3351
3352 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3353 DBG("Creating session with ONE network URI is a bad call");
3354 ret = LTTNG_ERR_SESSION_FAIL;
3355 free(uris);
3356 goto error;
3357 }
3358 }
3359
3360 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3361 nb_uri, &cmd_ctx->creds);
3362 free(uris);
3363 break;
3364 }
3365 default:
3366 ret = LTTNG_ERR_UND;
3367 break;
3368 }
3369
3370 error:
3371 if (cmd_ctx->llm == NULL) {
3372 DBG("Missing llm structure. Allocating one.");
3373 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3374 goto setup_error;
3375 }
3376 }
3377 /* Set return code */
3378 cmd_ctx->llm->ret_code = ret;
3379 setup_error:
3380 if (cmd_ctx->session) {
3381 session_unlock(cmd_ctx->session);
3382 }
3383 if (need_tracing_session) {
3384 session_unlock_list();
3385 }
3386 init_setup_error:
3387 return ret;
3388 }
3389
3390 /*
3391 * Thread managing health check socket.
3392 */
3393 static void *thread_manage_health(void *data)
3394 {
3395 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3396 uint32_t revents, nb_fd;
3397 struct lttng_poll_event events;
3398 struct lttcomm_health_msg msg;
3399 struct lttcomm_health_data reply;
3400
3401 DBG("[thread] Manage health check started");
3402
3403 rcu_register_thread();
3404
3405 /* We might hit an error path before this is created. */
3406 lttng_poll_init(&events);
3407
3408 /* Create unix socket */
3409 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3410 if (sock < 0) {
3411 ERR("Unable to create health check Unix socket");
3412 ret = -1;
3413 goto error;
3414 }
3415
3416 /*
3417 * Set the CLOEXEC flag. Return code is useless because either way, the
3418 * show must go on.
3419 */
3420 (void) utils_set_fd_cloexec(sock);
3421
3422 ret = lttcomm_listen_unix_sock(sock);
3423 if (ret < 0) {
3424 goto error;
3425 }
3426
3427 /*
3428 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3429 * more will be added to this poll set.
3430 */
3431 ret = sessiond_set_thread_pollset(&events, 2);
3432 if (ret < 0) {
3433 goto error;
3434 }
3435
3436 /* Add the application registration socket */
3437 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3438 if (ret < 0) {
3439 goto error;
3440 }
3441
3442 while (1) {
3443 DBG("Health check ready");
3444
3445 /* Inifinite blocking call, waiting for transmission */
3446 restart:
3447 ret = lttng_poll_wait(&events, -1);
3448 if (ret < 0) {
3449 /*
3450 * Restart interrupted system call.
3451 */
3452 if (errno == EINTR) {
3453 goto restart;
3454 }
3455 goto error;
3456 }
3457
3458 nb_fd = ret;
3459
3460 for (i = 0; i < nb_fd; i++) {
3461 /* Fetch once the poll data */
3462 revents = LTTNG_POLL_GETEV(&events, i);
3463 pollfd = LTTNG_POLL_GETFD(&events, i);
3464
3465 /* Thread quit pipe has been closed. Killing thread. */
3466 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3467 if (ret) {
3468 err = 0;
3469 goto exit;
3470 }
3471
3472 /* Event on the registration socket */
3473 if (pollfd == sock) {
3474 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3475 ERR("Health socket poll error");
3476 goto error;
3477 }
3478 }
3479 }
3480
3481 new_sock = lttcomm_accept_unix_sock(sock);
3482 if (new_sock < 0) {
3483 goto error;
3484 }
3485
3486 /*
3487 * Set the CLOEXEC flag. Return code is useless because either way, the
3488 * show must go on.
3489 */
3490 (void) utils_set_fd_cloexec(new_sock);
3491
3492 DBG("Receiving data from client for health...");
3493 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3494 if (ret <= 0) {
3495 DBG("Nothing recv() from client... continuing");
3496 ret = close(new_sock);
3497 if (ret) {
3498 PERROR("close");
3499 }
3500 new_sock = -1;
3501 continue;
3502 }
3503
3504 rcu_thread_online();
3505
3506 switch (msg.component) {
3507 case LTTNG_HEALTH_CMD:
3508 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3509 break;
3510 case LTTNG_HEALTH_APP_MANAGE:
3511 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3512 break;
3513 case LTTNG_HEALTH_APP_REG:
3514 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3515 break;
3516 case LTTNG_HEALTH_KERNEL:
3517 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3518 break;
3519 case LTTNG_HEALTH_CONSUMER:
3520 reply.ret_code = check_consumer_health();
3521 break;
3522 case LTTNG_HEALTH_HT_CLEANUP:
3523 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3524 break;
3525 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3526 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3527 break;
3528 case LTTNG_HEALTH_APP_REG_DISPATCH:
3529 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3530 break;
3531 case LTTNG_HEALTH_ALL:
3532 reply.ret_code =
3533 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3534 health_check_state(HEALTH_TYPE_APP_REG) &&
3535 health_check_state(HEALTH_TYPE_CMD) &&
3536 health_check_state(HEALTH_TYPE_KERNEL) &&
3537 check_consumer_health() &&
3538 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3539 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3540 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3541 break;
3542 default:
3543 reply.ret_code = LTTNG_ERR_UND;
3544 break;
3545 }
3546
3547 /*
3548 * Flip ret value since 0 is a success and 1 indicates a bad health for
3549 * the client where in the sessiond it is the opposite. Again, this is
3550 * just to make things easier for us poor developer which enjoy a lot
3551 * lazyness.
3552 */
3553 if (reply.ret_code == 0 || reply.ret_code == 1) {
3554 reply.ret_code = !reply.ret_code;
3555 }
3556
3557 DBG2("Health check return value %d", reply.ret_code);
3558
3559 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3560 if (ret < 0) {
3561 ERR("Failed to send health data back to client");
3562 }
3563
3564 /* End of transmission */
3565 ret = close(new_sock);
3566 if (ret) {
3567 PERROR("close");
3568 }
3569 new_sock = -1;
3570 }
3571
3572 exit:
3573 error:
3574 if (err) {
3575 ERR("Health error occurred in %s", __func__);
3576 }
3577 DBG("Health check thread dying");
3578 unlink(health_unix_sock_path);
3579 if (sock >= 0) {
3580 ret = close(sock);
3581 if (ret) {
3582 PERROR("close");
3583 }
3584 }
3585
3586 lttng_poll_clean(&events);
3587
3588 rcu_unregister_thread();
3589 return NULL;
3590 }
3591
3592 /*
3593 * This thread manage all clients request using the unix client socket for
3594 * communication.
3595 */
3596 static void *thread_manage_clients(void *data)
3597 {
3598 int sock = -1, ret, i, pollfd, err = -1;
3599 int sock_error;
3600 uint32_t revents, nb_fd;
3601 struct command_ctx *cmd_ctx = NULL;
3602 struct lttng_poll_event events;
3603
3604 DBG("[thread] Manage client started");
3605
3606 rcu_register_thread();
3607
3608 health_register(HEALTH_TYPE_CMD);
3609
3610 if (testpoint(thread_manage_clients)) {
3611 goto error_testpoint;
3612 }
3613
3614 health_code_update();
3615
3616 ret = lttcomm_listen_unix_sock(client_sock);
3617 if (ret < 0) {
3618 goto error_listen;
3619 }
3620
3621 /*
3622 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3623 * more will be added to this poll set.
3624 */
3625 ret = sessiond_set_thread_pollset(&events, 2);
3626 if (ret < 0) {
3627 goto error_create_poll;
3628 }
3629
3630 /* Add the application registration socket */
3631 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3632 if (ret < 0) {
3633 goto error;
3634 }
3635
3636 /*
3637 * Notify parent pid that we are ready to accept command for client side.
3638 */
3639 if (opt_sig_parent) {
3640 kill(ppid, SIGUSR1);
3641 }
3642
3643 if (testpoint(thread_manage_clients_before_loop)) {
3644 goto error;
3645 }
3646
3647 health_code_update();
3648
3649 while (1) {
3650 DBG("Accepting client command ...");
3651
3652 /* Inifinite blocking call, waiting for transmission */
3653 restart:
3654 health_poll_entry();
3655 ret = lttng_poll_wait(&events, -1);
3656 health_poll_exit();
3657 if (ret < 0) {
3658 /*
3659 * Restart interrupted system call.
3660 */
3661 if (errno == EINTR) {
3662 goto restart;
3663 }
3664 goto error;
3665 }
3666
3667 nb_fd = ret;
3668
3669 for (i = 0; i < nb_fd; i++) {
3670 /* Fetch once the poll data */
3671 revents = LTTNG_POLL_GETEV(&events, i);
3672 pollfd = LTTNG_POLL_GETFD(&events, i);
3673
3674 health_code_update();
3675
3676 /* Thread quit pipe has been closed. Killing thread. */
3677 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3678 if (ret) {
3679 err = 0;
3680 goto exit;
3681 }
3682
3683 /* Event on the registration socket */
3684 if (pollfd == client_sock) {
3685 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3686 ERR("Client socket poll error");
3687 goto error;
3688 }
3689 }
3690 }
3691
3692 DBG("Wait for client response");
3693
3694 health_code_update();
3695
3696 sock = lttcomm_accept_unix_sock(client_sock);
3697 if (sock < 0) {
3698 goto error;
3699 }
3700
3701 /*
3702 * Set the CLOEXEC flag. Return code is useless because either way, the
3703 * show must go on.
3704 */
3705 (void) utils_set_fd_cloexec(sock);
3706
3707 /* Set socket option for credentials retrieval */
3708 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3709 if (ret < 0) {
3710 goto error;
3711 }
3712
3713 /* Allocate context command to process the client request */
3714 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3715 if (cmd_ctx == NULL) {
3716 PERROR("zmalloc cmd_ctx");
3717 goto error;
3718 }
3719
3720 /* Allocate data buffer for reception */
3721 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3722 if (cmd_ctx->lsm == NULL) {
3723 PERROR("zmalloc cmd_ctx->lsm");
3724 goto error;
3725 }
3726
3727 cmd_ctx->llm = NULL;
3728 cmd_ctx->session = NULL;
3729
3730 health_code_update();
3731
3732 /*
3733 * Data is received from the lttng client. The struct
3734 * lttcomm_session_msg (lsm) contains the command and data request of
3735 * the client.
3736 */
3737 DBG("Receiving data from client ...");
3738 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3739 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3740 if (ret <= 0) {
3741 DBG("Nothing recv() from client... continuing");
3742 ret = close(sock);
3743 if (ret) {
3744 PERROR("close");
3745 }
3746 sock = -1;
3747 clean_command_ctx(&cmd_ctx);
3748 continue;
3749 }
3750
3751 health_code_update();
3752
3753 // TODO: Validate cmd_ctx including sanity check for
3754 // security purpose.
3755
3756 rcu_thread_online();
3757 /*
3758 * This function dispatch the work to the kernel or userspace tracer
3759 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3760 * informations for the client. The command context struct contains
3761 * everything this function may needs.
3762 */
3763 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3764 rcu_thread_offline();
3765 if (ret < 0) {
3766 ret = close(sock);
3767 if (ret) {
3768 PERROR("close");
3769 }
3770 sock = -1;
3771 /*
3772 * TODO: Inform client somehow of the fatal error. At
3773 * this point, ret < 0 means that a zmalloc failed
3774 * (ENOMEM). Error detected but still accept
3775 * command, unless a socket error has been
3776 * detected.
3777 */
3778 clean_command_ctx(&cmd_ctx);
3779 continue;
3780 }
3781
3782 health_code_update();
3783
3784 DBG("Sending response (size: %d, retcode: %s)",
3785 cmd_ctx->lttng_msg_size,
3786 lttng_strerror(-cmd_ctx->llm->ret_code));
3787 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3788 if (ret < 0) {
3789 ERR("Failed to send data back to client");
3790 }
3791
3792 /* End of transmission */
3793 ret = close(sock);
3794 if (ret) {
3795 PERROR("close");
3796 }
3797 sock = -1;
3798
3799 clean_command_ctx(&cmd_ctx);
3800
3801 health_code_update();
3802 }
3803
3804 exit:
3805 error:
3806 if (sock >= 0) {
3807 ret = close(sock);
3808 if (ret) {
3809 PERROR("close");
3810 }
3811 }
3812
3813 lttng_poll_clean(&events);
3814 clean_command_ctx(&cmd_ctx);
3815
3816 error_listen:
3817 error_create_poll:
3818 error_testpoint:
3819 unlink(client_unix_sock_path);
3820 if (client_sock >= 0) {
3821 ret = close(client_sock);
3822 if (ret) {
3823 PERROR("close");
3824 }
3825 }
3826
3827 if (err) {
3828 health_error();
3829 ERR("Health error occurred in %s", __func__);
3830 }
3831
3832 health_unregister();
3833
3834 DBG("Client thread dying");
3835
3836 rcu_unregister_thread();
3837 return NULL;
3838 }
3839
3840
3841 /*
3842 * usage function on stderr
3843 */
3844 static void usage(void)
3845 {
3846 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3847 fprintf(stderr, " -h, --help Display this usage.\n");
3848 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3849 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3850 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3851 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3852 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3853 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3854 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3855 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3856 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3857 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3858 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3859 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3860 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3861 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3862 fprintf(stderr, " -V, --version Show version number.\n");
3863 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3864 fprintf(stderr, " -q, --quiet No output at all.\n");
3865 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3866 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3867 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3868 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3869 }
3870
3871 /*
3872 * daemon argument parsing
3873 */
3874 static int parse_args(int argc, char **argv)
3875 {
3876 int c;
3877
3878 static struct option long_options[] = {
3879 { "client-sock", 1, 0, 'c' },
3880 { "apps-sock", 1, 0, 'a' },
3881 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3882 { "kconsumerd-err-sock", 1, 0, 'E' },
3883 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3884 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3885 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3886 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3887 { "consumerd32-path", 1, 0, 'u' },
3888 { "consumerd32-libdir", 1, 0, 'U' },
3889 { "consumerd64-path", 1, 0, 't' },
3890 { "consumerd64-libdir", 1, 0, 'T' },
3891 { "daemonize", 0, 0, 'd' },
3892 { "sig-parent", 0, 0, 'S' },
3893 { "help", 0, 0, 'h' },
3894 { "group", 1, 0, 'g' },
3895 { "version", 0, 0, 'V' },
3896 { "quiet", 0, 0, 'q' },
3897 { "verbose", 0, 0, 'v' },
3898 { "verbose-consumer", 0, 0, 'Z' },
3899 { "no-kernel", 0, 0, 'N' },
3900 { "pidfile", 1, 0, 'p' },
3901 { NULL, 0, 0, 0 }
3902 };
3903
3904 while (1) {
3905 int option_index = 0;
3906 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3907 long_options, &option_index);
3908 if (c == -1) {
3909 break;
3910 }
3911
3912 switch (c) {
3913 case 0:
3914 fprintf(stderr, "option %s", long_options[option_index].name);
3915 if (optarg) {
3916 fprintf(stderr, " with arg %s\n", optarg);
3917 }
3918 break;
3919 case 'c':
3920 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3921 break;
3922 case 'a':
3923 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3924 break;
3925 case 'd':
3926 opt_daemon = 1;
3927 break;
3928 case 'g':
3929 opt_tracing_group = optarg;
3930 break;
3931 case 'h':
3932 usage();
3933 exit(EXIT_FAILURE);
3934 case 'V':
3935 fprintf(stdout, "%s\n", VERSION);
3936 exit(EXIT_SUCCESS);
3937 case 'S':
3938 opt_sig_parent = 1;
3939 break;
3940 case 'E':
3941 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3942 break;
3943 case 'C':
3944 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3945 break;
3946 case 'F':
3947 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3948 break;
3949 case 'D':
3950 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3951 break;
3952 case 'H':
3953 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3954 break;
3955 case 'G':
3956 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3957 break;
3958 case 'N':
3959 opt_no_kernel = 1;
3960 break;
3961 case 'q':
3962 lttng_opt_quiet = 1;
3963 break;
3964 case 'v':
3965 /* Verbose level can increase using multiple -v */
3966 lttng_opt_verbose += 1;
3967 break;
3968 case 'Z':
3969 opt_verbose_consumer += 1;
3970 break;
3971 case 'u':
3972 consumerd32_bin= optarg;
3973 break;
3974 case 'U':
3975 consumerd32_libdir = optarg;
3976 break;
3977 case 't':
3978 consumerd64_bin = optarg;
3979 break;
3980 case 'T':
3981 consumerd64_libdir = optarg;
3982 break;
3983 case 'p':
3984 opt_pidfile = optarg;
3985 break;
3986 default:
3987 /* Unknown option or other error.
3988 * Error is printed by getopt, just return */
3989 return -1;
3990 }
3991 }
3992
3993 return 0;
3994 }
3995
3996 /*
3997 * Creates the two needed socket by the daemon.
3998 * apps_sock - The communication socket for all UST apps.
3999 * client_sock - The communication of the cli tool (lttng).
4000 */
4001 static int init_daemon_socket(void)
4002 {
4003 int ret = 0;
4004 mode_t old_umask;
4005
4006 old_umask = umask(0);
4007
4008 /* Create client tool unix socket */
4009 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4010 if (client_sock < 0) {
4011 ERR("Create unix sock failed: %s", client_unix_sock_path);
4012 ret = -1;
4013 goto end;
4014 }
4015
4016 /* Set the cloexec flag */
4017 ret = utils_set_fd_cloexec(client_sock);
4018 if (ret < 0) {
4019 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4020 "Continuing but note that the consumer daemon will have a "
4021 "reference to this socket on exec()", client_sock);
4022 }
4023
4024 /* File permission MUST be 660 */
4025 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4026 if (ret < 0) {
4027 ERR("Set file permissions failed: %s", client_unix_sock_path);
4028 PERROR("chmod");
4029 goto end;
4030 }
4031
4032 /* Create the application unix socket */
4033 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4034 if (apps_sock < 0) {
4035 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4036 ret = -1;
4037 goto end;
4038 }
4039
4040 /* Set the cloexec flag */
4041 ret = utils_set_fd_cloexec(apps_sock);
4042 if (ret < 0) {
4043 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4044 "Continuing but note that the consumer daemon will have a "
4045 "reference to this socket on exec()", apps_sock);
4046 }
4047
4048 /* File permission MUST be 666 */
4049 ret = chmod(apps_unix_sock_path,
4050 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4051 if (ret < 0) {
4052 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4053 PERROR("chmod");
4054 goto end;
4055 }
4056
4057 DBG3("Session daemon client socket %d and application socket %d created",
4058 client_sock, apps_sock);
4059
4060 end:
4061 umask(old_umask);
4062 return ret;
4063 }
4064
4065 /*
4066 * Check if the global socket is available, and if a daemon is answering at the
4067 * other side. If yes, error is returned.
4068 */
4069 static int check_existing_daemon(void)
4070 {
4071 /* Is there anybody out there ? */
4072 if (lttng_session_daemon_alive()) {
4073 return -EEXIST;
4074 }
4075
4076 return 0;
4077 }
4078
4079 /*
4080 * Set the tracing group gid onto the client socket.
4081 *
4082 * Race window between mkdir and chown is OK because we are going from more
4083 * permissive (root.root) to less permissive (root.tracing).
4084 */
4085 static int set_permissions(char *rundir)
4086 {
4087 int ret;
4088 gid_t gid;
4089
4090 ret = allowed_group();
4091 if (ret < 0) {
4092 WARN("No tracing group detected");
4093 /* Setting gid to 0 if no tracing group is found */
4094 gid = 0;
4095 } else {
4096 gid = ret;
4097 }
4098
4099 /* Set lttng run dir */
4100 ret = chown(rundir, 0, gid);
4101 if (ret < 0) {
4102 ERR("Unable to set group on %s", rundir);
4103 PERROR("chown");
4104 }
4105
4106 /* Ensure all applications and tracing group can search the run dir */
4107 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
4108 if (ret < 0) {
4109 ERR("Unable to set permissions on %s", rundir);
4110 PERROR("chmod");
4111 }
4112
4113 /* lttng client socket path */
4114 ret = chown(client_unix_sock_path, 0, gid);
4115 if (ret < 0) {
4116 ERR("Unable to set group on %s", client_unix_sock_path);
4117 PERROR("chown");
4118 }
4119
4120 /* kconsumer error socket path */
4121 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4122 if (ret < 0) {
4123 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4124 PERROR("chown");
4125 }
4126
4127 /* 64-bit ustconsumer error socket path */
4128 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4129 if (ret < 0) {
4130 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4131 PERROR("chown");
4132 }
4133
4134 /* 32-bit ustconsumer compat32 error socket path */
4135 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4136 if (ret < 0) {
4137 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4138 PERROR("chown");
4139 }
4140
4141 DBG("All permissions are set");
4142
4143 return ret;
4144 }
4145
4146 /*
4147 * Create the lttng run directory needed for all global sockets and pipe.
4148 */
4149 static int create_lttng_rundir(const char *rundir)
4150 {
4151 int ret;
4152
4153 DBG3("Creating LTTng run directory: %s", rundir);
4154
4155 ret = mkdir(rundir, S_IRWXU);
4156 if (ret < 0) {
4157 if (errno != EEXIST) {
4158 ERR("Unable to create %s", rundir);
4159 goto error;
4160 } else {
4161 ret = 0;
4162 }
4163 }
4164
4165 error:
4166 return ret;
4167 }
4168
4169 /*
4170 * Setup sockets and directory needed by the kconsumerd communication with the
4171 * session daemon.
4172 */
4173 static int set_consumer_sockets(struct consumer_data *consumer_data,
4174 const char *rundir)
4175 {
4176 int ret;
4177 char path[PATH_MAX];
4178
4179 switch (consumer_data->type) {
4180 case LTTNG_CONSUMER_KERNEL:
4181 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4182 break;
4183 case LTTNG_CONSUMER64_UST:
4184 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4185 break;
4186 case LTTNG_CONSUMER32_UST:
4187 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4188 break;
4189 default:
4190 ERR("Consumer type unknown");
4191 ret = -EINVAL;
4192 goto error;
4193 }
4194
4195 DBG2("Creating consumer directory: %s", path);
4196
4197 ret = mkdir(path, S_IRWXU);
4198 if (ret < 0) {
4199 if (errno != EEXIST) {
4200 PERROR("mkdir");
4201 ERR("Failed to create %s", path);
4202 goto error;
4203 }
4204 ret = -1;
4205 }
4206
4207 /* Create the kconsumerd error unix socket */
4208 consumer_data->err_sock =
4209 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4210 if (consumer_data->err_sock < 0) {
4211 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4212 ret = -1;
4213 goto error;
4214 }
4215
4216 /*
4217 * Set the CLOEXEC flag. Return code is useless because either way, the
4218 * show must go on.
4219 */
4220 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4221 if (ret < 0) {
4222 PERROR("utils_set_fd_cloexec");
4223 /* continue anyway */
4224 }
4225
4226 /* File permission MUST be 660 */
4227 ret = chmod(consumer_data->err_unix_sock_path,
4228 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4229 if (ret < 0) {
4230 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4231 PERROR("chmod");
4232 goto error;
4233 }
4234
4235 error:
4236 return ret;
4237 }
4238
4239 /*
4240 * Signal handler for the daemon
4241 *
4242 * Simply stop all worker threads, leaving main() return gracefully after
4243 * joining all threads and calling cleanup().
4244 */
4245 static void sighandler(int sig)
4246 {
4247 switch (sig) {
4248 case SIGPIPE:
4249 DBG("SIGPIPE caught");
4250 return;
4251 case SIGINT:
4252 DBG("SIGINT caught");
4253 stop_threads();
4254 break;
4255 case SIGTERM:
4256 DBG("SIGTERM caught");
4257 stop_threads();
4258 break;
4259 default:
4260 break;
4261 }
4262 }
4263
4264 /*
4265 * Setup signal handler for :
4266 * SIGINT, SIGTERM, SIGPIPE
4267 */
4268 static int set_signal_handler(void)
4269 {
4270 int ret = 0;
4271 struct sigaction sa;
4272 sigset_t sigset;
4273
4274 if ((ret = sigemptyset(&sigset)) < 0) {
4275 PERROR("sigemptyset");
4276 return ret;
4277 }
4278
4279 sa.sa_handler = sighandler;
4280 sa.sa_mask = sigset;
4281 sa.sa_flags = 0;
4282 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4283 PERROR("sigaction");
4284 return ret;
4285 }
4286
4287 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4288 PERROR("sigaction");
4289 return ret;
4290 }
4291
4292 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4293 PERROR("sigaction");
4294 return ret;
4295 }
4296
4297 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4298
4299 return ret;
4300 }
4301
4302 /*
4303 * Set open files limit to unlimited. This daemon can open a large number of
4304 * file descriptors in order to consumer multiple kernel traces.
4305 */
4306 static void set_ulimit(void)
4307 {
4308 int ret;
4309 struct rlimit lim;
4310
4311 /* The kernel does not allowed an infinite limit for open files */
4312 lim.rlim_cur = 65535;
4313 lim.rlim_max = 65535;
4314
4315 ret = setrlimit(RLIMIT_NOFILE, &lim);
4316 if (ret < 0) {
4317 PERROR("failed to set open files limit");
4318 }
4319 }
4320
4321 /*
4322 * Write pidfile using the rundir and opt_pidfile.
4323 */
4324 static void write_pidfile(void)
4325 {
4326 int ret;
4327 char pidfile_path[PATH_MAX];
4328
4329 assert(rundir);
4330
4331 if (opt_pidfile) {
4332 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4333 } else {
4334 /* Build pidfile path from rundir and opt_pidfile. */
4335 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4336 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4337 if (ret < 0) {
4338 PERROR("snprintf pidfile path");
4339 goto error;
4340 }
4341 }
4342
4343 /*
4344 * Create pid file in rundir. Return value is of no importance. The
4345 * execution will continue even though we are not able to write the file.
4346 */
4347 (void) utils_create_pid_file(getpid(), pidfile_path);
4348
4349 error:
4350 return;
4351 }
4352
4353 /*
4354 * main
4355 */
4356 int main(int argc, char **argv)
4357 {
4358 int ret = 0;
4359 void *status;
4360 const char *home_path, *env_app_timeout;
4361
4362 init_kernel_workarounds();
4363
4364 rcu_register_thread();
4365
4366 setup_consumerd_path();
4367
4368 page_size = sysconf(_SC_PAGESIZE);
4369 if (page_size < 0) {
4370 PERROR("sysconf _SC_PAGESIZE");
4371 page_size = LONG_MAX;
4372 WARN("Fallback page size to %ld", page_size);
4373 }
4374
4375 /* Parse arguments */
4376 progname = argv[0];
4377 if ((ret = parse_args(argc, argv)) < 0) {
4378 goto error;
4379 }
4380
4381 /* Daemonize */
4382 if (opt_daemon) {
4383 int i;
4384
4385 /*
4386 * fork
4387 * child: setsid, close FD 0, 1, 2, chdir /
4388 * parent: exit (if fork is successful)
4389 */
4390 ret = daemon(0, 0);
4391 if (ret < 0) {
4392 PERROR("daemon");
4393 goto error;
4394 }
4395 /*
4396 * We are in the child. Make sure all other file
4397 * descriptors are closed, in case we are called with
4398 * more opened file descriptors than the standard ones.
4399 */
4400 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4401 (void) close(i);
4402 }
4403 }
4404
4405 /* Create thread quit pipe */
4406 if ((ret = init_thread_quit_pipe()) < 0) {
4407 goto error;
4408 }
4409
4410 /* Check if daemon is UID = 0 */
4411 is_root = !getuid();
4412
4413 if (is_root) {
4414 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4415
4416 /* Create global run dir with root access */
4417 ret = create_lttng_rundir(rundir);
4418 if (ret < 0) {
4419 goto error;
4420 }
4421
4422 if (strlen(apps_unix_sock_path) == 0) {
4423 snprintf(apps_unix_sock_path, PATH_MAX,
4424 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4425 }
4426
4427 if (strlen(client_unix_sock_path) == 0) {
4428 snprintf(client_unix_sock_path, PATH_MAX,
4429 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4430 }
4431
4432 /* Set global SHM for ust */
4433 if (strlen(wait_shm_path) == 0) {
4434 snprintf(wait_shm_path, PATH_MAX,
4435 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4436 }
4437
4438 if (strlen(health_unix_sock_path) == 0) {
4439 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4440 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4441 }
4442
4443 /* Setup kernel consumerd path */
4444 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4445 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4446 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4447 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4448
4449 DBG2("Kernel consumer err path: %s",
4450 kconsumer_data.err_unix_sock_path);
4451 DBG2("Kernel consumer cmd path: %s",
4452 kconsumer_data.cmd_unix_sock_path);
4453 } else {
4454 home_path = utils_get_home_dir();
4455 if (home_path == NULL) {
4456 /* TODO: Add --socket PATH option */
4457 ERR("Can't get HOME directory for sockets creation.");
4458 ret = -EPERM;
4459 goto error;
4460 }
4461
4462 /*
4463 * Create rundir from home path. This will create something like
4464 * $HOME/.lttng
4465 */
4466 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4467 if (ret < 0) {
4468 ret = -ENOMEM;
4469 goto error;
4470 }
4471
4472 ret = create_lttng_rundir(rundir);
4473 if (ret < 0) {
4474 goto error;
4475 }
4476
4477 if (strlen(apps_unix_sock_path) == 0) {
4478 snprintf(apps_unix_sock_path, PATH_MAX,
4479 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4480 }
4481
4482 /* Set the cli tool unix socket path */
4483 if (strlen(client_unix_sock_path) == 0) {
4484 snprintf(client_unix_sock_path, PATH_MAX,
4485 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4486 }
4487
4488 /* Set global SHM for ust */
4489 if (strlen(wait_shm_path) == 0) {
4490 snprintf(wait_shm_path, PATH_MAX,
4491 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4492 }
4493
4494 /* Set health check Unix path */
4495 if (strlen(health_unix_sock_path) == 0) {
4496 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4497 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4498 }
4499 }
4500
4501 /* Set consumer initial state */
4502 kernel_consumerd_state = CONSUMER_STOPPED;
4503 ust_consumerd_state = CONSUMER_STOPPED;
4504
4505 DBG("Client socket path %s", client_unix_sock_path);
4506 DBG("Application socket path %s", apps_unix_sock_path);
4507 DBG("Application wait path %s", wait_shm_path);
4508 DBG("LTTng run directory path: %s", rundir);
4509
4510 /* 32 bits consumerd path setup */
4511 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4512 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4513 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4514 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4515
4516 DBG2("UST consumer 32 bits err path: %s",
4517 ustconsumer32_data.err_unix_sock_path);
4518 DBG2("UST consumer 32 bits cmd path: %s",
4519 ustconsumer32_data.cmd_unix_sock_path);
4520
4521 /* 64 bits consumerd path setup */
4522 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4523 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4524 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4525 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4526
4527 DBG2("UST consumer 64 bits err path: %s",
4528 ustconsumer64_data.err_unix_sock_path);
4529 DBG2("UST consumer 64 bits cmd path: %s",
4530 ustconsumer64_data.cmd_unix_sock_path);
4531
4532 /*
4533 * See if daemon already exist.
4534 */
4535 if ((ret = check_existing_daemon()) < 0) {
4536 ERR("Already running daemon.\n");
4537 /*
4538 * We do not goto exit because we must not cleanup()
4539 * because a daemon is already running.
4540 */
4541 goto error;
4542 }
4543
4544 /*
4545 * Init UST app hash table. Alloc hash table before this point since
4546 * cleanup() can get called after that point.
4547 */
4548 ust_app_ht_alloc();
4549
4550 /* After this point, we can safely call cleanup() with "goto exit" */
4551
4552 /*
4553 * These actions must be executed as root. We do that *after* setting up
4554 * the sockets path because we MUST make the check for another daemon using
4555 * those paths *before* trying to set the kernel consumer sockets and init
4556 * kernel tracer.
4557 */
4558 if (is_root) {
4559 ret = set_consumer_sockets(&kconsumer_data, rundir);
4560 if (ret < 0) {
4561 goto exit;
4562 }
4563
4564 /* Setup kernel tracer */
4565 if (!opt_no_kernel) {
4566 init_kernel_tracer();
4567 }
4568
4569 /* Set ulimit for open files */
4570 set_ulimit();
4571 }
4572 /* init lttng_fd tracking must be done after set_ulimit. */
4573 lttng_fd_init();
4574
4575 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4576 if (ret < 0) {
4577 goto exit;
4578 }
4579
4580 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4581 if (ret < 0) {
4582 goto exit;
4583 }
4584
4585 if ((ret = set_signal_handler()) < 0) {
4586 goto exit;
4587 }
4588
4589 /* Setup the needed unix socket */
4590 if ((ret = init_daemon_socket()) < 0) {
4591 goto exit;
4592 }
4593
4594 /* Set credentials to socket */
4595 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4596 goto exit;
4597 }
4598
4599 /* Get parent pid if -S, --sig-parent is specified. */
4600 if (opt_sig_parent) {
4601 ppid = getppid();
4602 }
4603
4604 /* Setup the kernel pipe for waking up the kernel thread */
4605 if (is_root && !opt_no_kernel) {
4606 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4607 goto exit;
4608 }
4609 }
4610
4611 /* Setup the thread ht_cleanup communication pipe. */
4612 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4613 goto exit;
4614 }
4615
4616 /* Setup the thread apps communication pipe. */
4617 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4618 goto exit;
4619 }
4620
4621 /* Setup the thread apps notify communication pipe. */
4622 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4623 goto exit;
4624 }
4625
4626 /* Initialize global buffer per UID and PID registry. */
4627 buffer_reg_init_uid_registry();
4628 buffer_reg_init_pid_registry();
4629
4630 /* Init UST command queue. */
4631 cds_wfq_init(&ust_cmd_queue.queue);
4632
4633 /*
4634 * Get session list pointer. This pointer MUST NOT be free(). This list is
4635 * statically declared in session.c
4636 */
4637 session_list_ptr = session_get_list();
4638
4639 /* Set up max poll set size */
4640 lttng_poll_set_max_size();
4641
4642 cmd_init();
4643
4644 /* Check for the application socket timeout env variable. */
4645 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4646 if (env_app_timeout) {
4647 app_socket_timeout = atoi(env_app_timeout);
4648 } else {
4649 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4650 }
4651
4652 write_pidfile();
4653
4654 /* Initialize communication library */
4655 lttcomm_init();
4656
4657 /* Create thread to manage the client socket */
4658 ret = pthread_create(&ht_cleanup_thread, NULL,
4659 thread_ht_cleanup, (void *) NULL);
4660 if (ret != 0) {
4661 PERROR("pthread_create ht_cleanup");
4662 goto exit_ht_cleanup;
4663 }
4664
4665 /* Create thread to manage the client socket */
4666 ret = pthread_create(&health_thread, NULL,
4667 thread_manage_health, (void *) NULL);
4668 if (ret != 0) {
4669 PERROR("pthread_create health");
4670 goto exit_health;
4671 }
4672
4673 /* Create thread to manage the client socket */
4674 ret = pthread_create(&client_thread, NULL,
4675 thread_manage_clients, (void *) NULL);
4676 if (ret != 0) {
4677 PERROR("pthread_create clients");
4678 goto exit_client;
4679 }
4680
4681 /* Create thread to dispatch registration */
4682 ret = pthread_create(&dispatch_thread, NULL,
4683 thread_dispatch_ust_registration, (void *) NULL);
4684 if (ret != 0) {
4685 PERROR("pthread_create dispatch");
4686 goto exit_dispatch;
4687 }
4688
4689 /* Create thread to manage application registration. */
4690 ret = pthread_create(&reg_apps_thread, NULL,
4691 thread_registration_apps, (void *) NULL);
4692 if (ret != 0) {
4693 PERROR("pthread_create registration");
4694 goto exit_reg_apps;
4695 }
4696
4697 /* Create thread to manage application socket */
4698 ret = pthread_create(&apps_thread, NULL,
4699 thread_manage_apps, (void *) NULL);
4700 if (ret != 0) {
4701 PERROR("pthread_create apps");
4702 goto exit_apps;
4703 }
4704
4705 /* Create thread to manage application notify socket */
4706 ret = pthread_create(&apps_notify_thread, NULL,
4707 ust_thread_manage_notify, (void *) NULL);
4708 if (ret != 0) {
4709 PERROR("pthread_create apps");
4710 goto exit_apps;
4711 }
4712
4713 /* Don't start this thread if kernel tracing is not requested nor root */
4714 if (is_root && !opt_no_kernel) {
4715 /* Create kernel thread to manage kernel event */
4716 ret = pthread_create(&kernel_thread, NULL,
4717 thread_manage_kernel, (void *) NULL);
4718 if (ret != 0) {
4719 PERROR("pthread_create kernel");
4720 goto exit_kernel;
4721 }
4722
4723 ret = pthread_join(kernel_thread, &status);
4724 if (ret != 0) {
4725 PERROR("pthread_join");
4726 goto error; /* join error, exit without cleanup */
4727 }
4728 }
4729
4730 exit_kernel:
4731 ret = pthread_join(apps_thread, &status);
4732 if (ret != 0) {
4733 PERROR("pthread_join");
4734 goto error; /* join error, exit without cleanup */
4735 }
4736
4737 exit_apps:
4738 ret = pthread_join(reg_apps_thread, &status);
4739 if (ret != 0) {
4740 PERROR("pthread_join");
4741 goto error; /* join error, exit without cleanup */
4742 }
4743
4744 exit_reg_apps:
4745 ret = pthread_join(dispatch_thread, &status);
4746 if (ret != 0) {
4747 PERROR("pthread_join");
4748 goto error; /* join error, exit without cleanup */
4749 }
4750
4751 exit_dispatch:
4752 ret = pthread_join(client_thread, &status);
4753 if (ret != 0) {
4754 PERROR("pthread_join");
4755 goto error; /* join error, exit without cleanup */
4756 }
4757
4758 ret = join_consumer_thread(&kconsumer_data);
4759 if (ret != 0) {
4760 PERROR("join_consumer");
4761 goto error; /* join error, exit without cleanup */
4762 }
4763
4764 ret = join_consumer_thread(&ustconsumer32_data);
4765 if (ret != 0) {
4766 PERROR("join_consumer ust32");
4767 goto error; /* join error, exit without cleanup */
4768 }
4769
4770 ret = join_consumer_thread(&ustconsumer64_data);
4771 if (ret != 0) {
4772 PERROR("join_consumer ust64");
4773 goto error; /* join error, exit without cleanup */
4774 }
4775
4776 exit_client:
4777 ret = pthread_join(health_thread, &status);
4778 if (ret != 0) {
4779 PERROR("pthread_join health thread");
4780 goto error; /* join error, exit without cleanup */
4781 }
4782
4783 exit_health:
4784 ret = pthread_join(ht_cleanup_thread, &status);
4785 if (ret != 0) {
4786 PERROR("pthread_join ht cleanup thread");
4787 goto error; /* join error, exit without cleanup */
4788 }
4789 exit_ht_cleanup:
4790 exit:
4791 /*
4792 * cleanup() is called when no other thread is running.
4793 */
4794 rcu_thread_online();
4795 cleanup();
4796 rcu_thread_offline();
4797 rcu_unregister_thread();
4798 if (!ret) {
4799 exit(EXIT_SUCCESS);
4800 }
4801 error:
4802 exit(EXIT_FAILURE);
4803 }
This page took 0.128527 seconds and 4 git commands to generate.