Improve comments after review
[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 pthread_mutex_lock(socket->lock);
694 ret = kernel_consumer_send_channel_stream(socket,
695 channel, ksess,
696 session->output_traces ? 1 : 0);
697 pthread_mutex_unlock(socket->lock);
698 if (ret < 0) {
699 rcu_read_unlock();
700 goto error;
701 }
702 }
703 rcu_read_unlock();
704 }
705 goto error;
706 }
707 }
708 session_unlock(session);
709 }
710 session_unlock_list();
711 return ret;
712
713 error:
714 session_unlock(session);
715 session_unlock_list();
716 return ret;
717 }
718
719 /*
720 * For each tracing session, update newly registered apps. The session list
721 * lock MUST be acquired before calling this.
722 */
723 static void update_ust_app(int app_sock)
724 {
725 struct ltt_session *sess, *stmp;
726
727 /* Consumer is in an ERROR state. Stop any application update. */
728 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
729 /* Stop the update process since the consumer is dead. */
730 return;
731 }
732
733 /* For all tracing session(s) */
734 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
735 session_lock(sess);
736 if (sess->ust_session) {
737 ust_app_global_update(sess->ust_session, app_sock);
738 }
739 session_unlock(sess);
740 }
741 }
742
743 /*
744 * This thread manage event coming from the kernel.
745 *
746 * Features supported in this thread:
747 * -) CPU Hotplug
748 */
749 static void *thread_manage_kernel(void *data)
750 {
751 int ret, i, pollfd, update_poll_flag = 1, err = -1;
752 uint32_t revents, nb_fd;
753 char tmp;
754 struct lttng_poll_event events;
755
756 DBG("[thread] Thread manage kernel started");
757
758 health_register(HEALTH_TYPE_KERNEL);
759
760 /*
761 * This first step of the while is to clean this structure which could free
762 * non NULL pointers so initialize it before the loop.
763 */
764 lttng_poll_init(&events);
765
766 if (testpoint(thread_manage_kernel)) {
767 goto error_testpoint;
768 }
769
770 health_code_update();
771
772 if (testpoint(thread_manage_kernel_before_loop)) {
773 goto error_testpoint;
774 }
775
776 while (1) {
777 health_code_update();
778
779 if (update_poll_flag == 1) {
780 /* Clean events object. We are about to populate it again. */
781 lttng_poll_clean(&events);
782
783 ret = sessiond_set_thread_pollset(&events, 2);
784 if (ret < 0) {
785 goto error_poll_create;
786 }
787
788 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
789 if (ret < 0) {
790 goto error;
791 }
792
793 /* This will add the available kernel channel if any. */
794 ret = update_kernel_poll(&events);
795 if (ret < 0) {
796 goto error;
797 }
798 update_poll_flag = 0;
799 }
800
801 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
802
803 /* Poll infinite value of time */
804 restart:
805 health_poll_entry();
806 ret = lttng_poll_wait(&events, -1);
807 health_poll_exit();
808 if (ret < 0) {
809 /*
810 * Restart interrupted system call.
811 */
812 if (errno == EINTR) {
813 goto restart;
814 }
815 goto error;
816 } else if (ret == 0) {
817 /* Should not happen since timeout is infinite */
818 ERR("Return value of poll is 0 with an infinite timeout.\n"
819 "This should not have happened! Continuing...");
820 continue;
821 }
822
823 nb_fd = ret;
824
825 for (i = 0; i < nb_fd; i++) {
826 /* Fetch once the poll data */
827 revents = LTTNG_POLL_GETEV(&events, i);
828 pollfd = LTTNG_POLL_GETFD(&events, i);
829
830 health_code_update();
831
832 /* Thread quit pipe has been closed. Killing thread. */
833 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
834 if (ret) {
835 err = 0;
836 goto exit;
837 }
838
839 /* Check for data on kernel pipe */
840 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
841 do {
842 ret = read(kernel_poll_pipe[0], &tmp, 1);
843 } while (ret < 0 && errno == EINTR);
844 /*
845 * Ret value is useless here, if this pipe gets any actions an
846 * update is required anyway.
847 */
848 update_poll_flag = 1;
849 continue;
850 } else {
851 /*
852 * New CPU detected by the kernel. Adding kernel stream to
853 * kernel session and updating the kernel consumer
854 */
855 if (revents & LPOLLIN) {
856 ret = update_kernel_stream(&kconsumer_data, pollfd);
857 if (ret < 0) {
858 continue;
859 }
860 break;
861 /*
862 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
863 * and unregister kernel stream at this point.
864 */
865 }
866 }
867 }
868 }
869
870 exit:
871 error:
872 lttng_poll_clean(&events);
873 error_poll_create:
874 error_testpoint:
875 utils_close_pipe(kernel_poll_pipe);
876 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
877 if (err) {
878 health_error();
879 ERR("Health error occurred in %s", __func__);
880 WARN("Kernel thread died unexpectedly. "
881 "Kernel tracing can continue but CPU hotplug is disabled.");
882 }
883 health_unregister();
884 DBG("Kernel thread dying");
885 return NULL;
886 }
887
888 /*
889 * Signal pthread condition of the consumer data that the thread.
890 */
891 static void signal_consumer_condition(struct consumer_data *data, int state)
892 {
893 pthread_mutex_lock(&data->cond_mutex);
894
895 /*
896 * The state is set before signaling. It can be any value, it's the waiter
897 * job to correctly interpret this condition variable associated to the
898 * consumer pthread_cond.
899 *
900 * A value of 0 means that the corresponding thread of the consumer data
901 * was not started. 1 indicates that the thread has started and is ready
902 * for action. A negative value means that there was an error during the
903 * thread bootstrap.
904 */
905 data->consumer_thread_is_ready = state;
906 (void) pthread_cond_signal(&data->cond);
907
908 pthread_mutex_unlock(&data->cond_mutex);
909 }
910
911 /*
912 * This thread manage the consumer error sent back to the session daemon.
913 */
914 static void *thread_manage_consumer(void *data)
915 {
916 int sock = -1, i, ret, pollfd, err = -1;
917 uint32_t revents, nb_fd;
918 enum lttcomm_return_code code;
919 struct lttng_poll_event events;
920 struct consumer_data *consumer_data = data;
921
922 DBG("[thread] Manage consumer started");
923
924 health_register(HEALTH_TYPE_CONSUMER);
925
926 health_code_update();
927
928 /*
929 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
930 * metadata_sock. Nothing more will be added to this poll set.
931 */
932 ret = sessiond_set_thread_pollset(&events, 3);
933 if (ret < 0) {
934 goto error_poll;
935 }
936
937 /*
938 * The error socket here is already in a listening state which was done
939 * just before spawning this thread to avoid a race between the consumer
940 * daemon exec trying to connect and the listen() call.
941 */
942 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
943 if (ret < 0) {
944 goto error;
945 }
946
947 health_code_update();
948
949 /* Infinite blocking call, waiting for transmission */
950 restart:
951 health_poll_entry();
952
953 if (testpoint(thread_manage_consumer)) {
954 goto error;
955 }
956
957 ret = lttng_poll_wait(&events, -1);
958 health_poll_exit();
959 if (ret < 0) {
960 /*
961 * Restart interrupted system call.
962 */
963 if (errno == EINTR) {
964 goto restart;
965 }
966 goto error;
967 }
968
969 nb_fd = ret;
970
971 for (i = 0; i < nb_fd; i++) {
972 /* Fetch once the poll data */
973 revents = LTTNG_POLL_GETEV(&events, i);
974 pollfd = LTTNG_POLL_GETFD(&events, i);
975
976 health_code_update();
977
978 /* Thread quit pipe has been closed. Killing thread. */
979 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
980 if (ret) {
981 err = 0;
982 goto exit;
983 }
984
985 /* Event on the registration socket */
986 if (pollfd == consumer_data->err_sock) {
987 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
988 ERR("consumer err socket poll error");
989 goto error;
990 }
991 }
992 }
993
994 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
995 if (sock < 0) {
996 goto error;
997 }
998
999 /*
1000 * Set the CLOEXEC flag. Return code is useless because either way, the
1001 * show must go on.
1002 */
1003 (void) utils_set_fd_cloexec(sock);
1004
1005 health_code_update();
1006
1007 DBG2("Receiving code from consumer err_sock");
1008
1009 /* Getting status code from kconsumerd */
1010 ret = lttcomm_recv_unix_sock(sock, &code,
1011 sizeof(enum lttcomm_return_code));
1012 if (ret <= 0) {
1013 goto error;
1014 }
1015
1016 health_code_update();
1017
1018 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1019 /* Connect both socket, command and metadata. */
1020 consumer_data->cmd_sock =
1021 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1022 consumer_data->metadata_fd =
1023 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1024 if (consumer_data->cmd_sock < 0
1025 || consumer_data->metadata_fd < 0) {
1026 PERROR("consumer connect cmd socket");
1027 /* On error, signal condition and quit. */
1028 signal_consumer_condition(consumer_data, -1);
1029 goto error;
1030 }
1031 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1032 /* Create metadata socket lock. */
1033 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1034 if (consumer_data->metadata_sock.lock == NULL) {
1035 PERROR("zmalloc pthread mutex");
1036 ret = -1;
1037 goto error;
1038 }
1039 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1040
1041 signal_consumer_condition(consumer_data, 1);
1042 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1043 DBG("Consumer metadata socket ready (fd: %d)",
1044 consumer_data->metadata_fd);
1045 } else {
1046 ERR("consumer error when waiting for SOCK_READY : %s",
1047 lttcomm_get_readable_code(-code));
1048 goto error;
1049 }
1050
1051 /* Remove the consumerd error sock since we've established a connexion */
1052 ret = lttng_poll_del(&events, consumer_data->err_sock);
1053 if (ret < 0) {
1054 goto error;
1055 }
1056
1057 /* Add new accepted error socket. */
1058 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1059 if (ret < 0) {
1060 goto error;
1061 }
1062
1063 /* Add metadata socket that is successfully connected. */
1064 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1065 LPOLLIN | LPOLLRDHUP);
1066 if (ret < 0) {
1067 goto error;
1068 }
1069
1070 health_code_update();
1071
1072 /* Infinite blocking call, waiting for transmission */
1073 restart_poll:
1074 while (1) {
1075 health_poll_entry();
1076 ret = lttng_poll_wait(&events, -1);
1077 health_poll_exit();
1078 if (ret < 0) {
1079 /*
1080 * Restart interrupted system call.
1081 */
1082 if (errno == EINTR) {
1083 goto restart_poll;
1084 }
1085 goto error;
1086 }
1087
1088 nb_fd = ret;
1089
1090 for (i = 0; i < nb_fd; i++) {
1091 /* Fetch once the poll data */
1092 revents = LTTNG_POLL_GETEV(&events, i);
1093 pollfd = LTTNG_POLL_GETFD(&events, i);
1094
1095 health_code_update();
1096
1097 /* Thread quit pipe has been closed. Killing thread. */
1098 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1099 if (ret) {
1100 err = 0;
1101 goto exit;
1102 }
1103
1104 if (pollfd == sock) {
1105 /* Event on the consumerd socket */
1106 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1107 ERR("consumer err socket second poll error");
1108 goto error;
1109 }
1110 health_code_update();
1111 /* Wait for any kconsumerd error */
1112 ret = lttcomm_recv_unix_sock(sock, &code,
1113 sizeof(enum lttcomm_return_code));
1114 if (ret <= 0) {
1115 ERR("consumer closed the command socket");
1116 goto error;
1117 }
1118
1119 ERR("consumer return code : %s",
1120 lttcomm_get_readable_code(-code));
1121
1122 goto exit;
1123 } else if (pollfd == consumer_data->metadata_fd) {
1124 /* UST metadata requests */
1125 ret = ust_consumer_metadata_request(
1126 &consumer_data->metadata_sock);
1127 if (ret < 0) {
1128 ERR("Handling metadata request");
1129 goto error;
1130 }
1131 break;
1132 } else {
1133 ERR("Unknown pollfd");
1134 goto error;
1135 }
1136 }
1137 health_code_update();
1138 }
1139
1140 exit:
1141 error:
1142 /*
1143 * We lock here because we are about to close the sockets and some other
1144 * thread might be using them so get exclusive access which will abort all
1145 * other consumer command by other threads.
1146 */
1147 pthread_mutex_lock(&consumer_data->lock);
1148
1149 /* Immediately set the consumerd state to stopped */
1150 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1151 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1152 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1153 consumer_data->type == LTTNG_CONSUMER32_UST) {
1154 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1155 } else {
1156 /* Code flow error... */
1157 assert(0);
1158 }
1159
1160 if (consumer_data->err_sock >= 0) {
1161 ret = close(consumer_data->err_sock);
1162 if (ret) {
1163 PERROR("close");
1164 }
1165 consumer_data->err_sock = -1;
1166 }
1167 if (consumer_data->cmd_sock >= 0) {
1168 ret = close(consumer_data->cmd_sock);
1169 if (ret) {
1170 PERROR("close");
1171 }
1172 consumer_data->cmd_sock = -1;
1173 }
1174 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1175 ret = close(*consumer_data->metadata_sock.fd_ptr);
1176 if (ret) {
1177 PERROR("close");
1178 }
1179 }
1180
1181 if (sock >= 0) {
1182 ret = close(sock);
1183 if (ret) {
1184 PERROR("close");
1185 }
1186 }
1187
1188 unlink(consumer_data->err_unix_sock_path);
1189 unlink(consumer_data->cmd_unix_sock_path);
1190 consumer_data->pid = 0;
1191 pthread_mutex_unlock(&consumer_data->lock);
1192
1193 /* Cleanup metadata socket mutex. */
1194 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1195 free(consumer_data->metadata_sock.lock);
1196
1197 lttng_poll_clean(&events);
1198 error_poll:
1199 if (err) {
1200 health_error();
1201 ERR("Health error occurred in %s", __func__);
1202 }
1203 health_unregister();
1204 DBG("consumer thread cleanup completed");
1205
1206 return NULL;
1207 }
1208
1209 /*
1210 * This thread manage application communication.
1211 */
1212 static void *thread_manage_apps(void *data)
1213 {
1214 int i, ret, pollfd, err = -1;
1215 uint32_t revents, nb_fd;
1216 struct lttng_poll_event events;
1217
1218 DBG("[thread] Manage application started");
1219
1220 rcu_register_thread();
1221 rcu_thread_online();
1222
1223 health_register(HEALTH_TYPE_APP_MANAGE);
1224
1225 if (testpoint(thread_manage_apps)) {
1226 goto error_testpoint;
1227 }
1228
1229 health_code_update();
1230
1231 ret = sessiond_set_thread_pollset(&events, 2);
1232 if (ret < 0) {
1233 goto error_poll_create;
1234 }
1235
1236 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1237 if (ret < 0) {
1238 goto error;
1239 }
1240
1241 if (testpoint(thread_manage_apps_before_loop)) {
1242 goto error;
1243 }
1244
1245 health_code_update();
1246
1247 while (1) {
1248 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1249
1250 /* Inifinite blocking call, waiting for transmission */
1251 restart:
1252 health_poll_entry();
1253 ret = lttng_poll_wait(&events, -1);
1254 health_poll_exit();
1255 if (ret < 0) {
1256 /*
1257 * Restart interrupted system call.
1258 */
1259 if (errno == EINTR) {
1260 goto restart;
1261 }
1262 goto error;
1263 }
1264
1265 nb_fd = ret;
1266
1267 for (i = 0; i < nb_fd; i++) {
1268 /* Fetch once the poll data */
1269 revents = LTTNG_POLL_GETEV(&events, i);
1270 pollfd = LTTNG_POLL_GETFD(&events, i);
1271
1272 health_code_update();
1273
1274 /* Thread quit pipe has been closed. Killing thread. */
1275 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1276 if (ret) {
1277 err = 0;
1278 goto exit;
1279 }
1280
1281 /* Inspect the apps cmd pipe */
1282 if (pollfd == apps_cmd_pipe[0]) {
1283 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1284 ERR("Apps command pipe error");
1285 goto error;
1286 } else if (revents & LPOLLIN) {
1287 int sock;
1288
1289 /* Empty pipe */
1290 do {
1291 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1292 } while (ret < 0 && errno == EINTR);
1293 if (ret < 0 || ret < sizeof(sock)) {
1294 PERROR("read apps cmd pipe");
1295 goto error;
1296 }
1297
1298 health_code_update();
1299
1300 /*
1301 * We only monitor the error events of the socket. This
1302 * thread does not handle any incoming data from UST
1303 * (POLLIN).
1304 */
1305 ret = lttng_poll_add(&events, sock,
1306 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1307 if (ret < 0) {
1308 goto error;
1309 }
1310
1311 /*
1312 * Set socket timeout for both receiving and ending.
1313 * app_socket_timeout is in seconds, whereas
1314 * lttcomm_setsockopt_rcv_timeout and
1315 * lttcomm_setsockopt_snd_timeout expect msec as
1316 * parameter.
1317 */
1318 (void) lttcomm_setsockopt_rcv_timeout(sock,
1319 app_socket_timeout * 1000);
1320 (void) lttcomm_setsockopt_snd_timeout(sock,
1321 app_socket_timeout * 1000);
1322
1323 DBG("Apps with sock %d added to poll set", sock);
1324
1325 health_code_update();
1326
1327 break;
1328 }
1329 } else {
1330 /*
1331 * At this point, we know that a registered application made
1332 * the event at poll_wait.
1333 */
1334 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1335 /* Removing from the poll set */
1336 ret = lttng_poll_del(&events, pollfd);
1337 if (ret < 0) {
1338 goto error;
1339 }
1340
1341 /* Socket closed on remote end. */
1342 ust_app_unregister(pollfd);
1343 break;
1344 }
1345 }
1346
1347 health_code_update();
1348 }
1349 }
1350
1351 exit:
1352 error:
1353 lttng_poll_clean(&events);
1354 error_poll_create:
1355 error_testpoint:
1356 utils_close_pipe(apps_cmd_pipe);
1357 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1358
1359 /*
1360 * We don't clean the UST app hash table here since already registered
1361 * applications can still be controlled so let them be until the session
1362 * daemon dies or the applications stop.
1363 */
1364
1365 if (err) {
1366 health_error();
1367 ERR("Health error occurred in %s", __func__);
1368 }
1369 health_unregister();
1370 DBG("Application communication apps thread cleanup complete");
1371 rcu_thread_offline();
1372 rcu_unregister_thread();
1373 return NULL;
1374 }
1375
1376 /*
1377 * Send a socket to a thread This is called from the dispatch UST registration
1378 * thread once all sockets are set for the application.
1379 *
1380 * On success, return 0 else a negative value being the errno message of the
1381 * write().
1382 */
1383 static int send_socket_to_thread(int fd, int sock)
1384 {
1385 int ret;
1386
1387 /* Sockets MUST be set or else this should not have been called. */
1388 assert(fd >= 0);
1389 assert(sock >= 0);
1390
1391 do {
1392 ret = write(fd, &sock, sizeof(sock));
1393 } while (ret < 0 && errno == EINTR);
1394 if (ret < 0 || ret != sizeof(sock)) {
1395 PERROR("write apps pipe %d", fd);
1396 if (ret < 0) {
1397 ret = -errno;
1398 }
1399 goto error;
1400 }
1401
1402 /* All good. Don't send back the write positive ret value. */
1403 ret = 0;
1404 error:
1405 return ret;
1406 }
1407
1408 /*
1409 * Sanitize the wait queue of the dispatch registration thread meaning removing
1410 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1411 * notify socket is never received.
1412 */
1413 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1414 {
1415 int ret, nb_fd = 0, i;
1416 unsigned int fd_added = 0;
1417 struct lttng_poll_event events;
1418 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1419
1420 assert(wait_queue);
1421
1422 lttng_poll_init(&events);
1423
1424 /* Just skip everything for an empty queue. */
1425 if (!wait_queue->count) {
1426 goto end;
1427 }
1428
1429 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1430 if (ret < 0) {
1431 goto error_create;
1432 }
1433
1434 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1435 &wait_queue->head, head) {
1436 assert(wait_node->app);
1437 ret = lttng_poll_add(&events, wait_node->app->sock,
1438 LPOLLHUP | LPOLLERR);
1439 if (ret < 0) {
1440 goto error;
1441 }
1442
1443 fd_added = 1;
1444 }
1445
1446 if (!fd_added) {
1447 goto end;
1448 }
1449
1450 /*
1451 * Poll but don't block so we can quickly identify the faulty events and
1452 * clean them afterwards from the wait queue.
1453 */
1454 ret = lttng_poll_wait(&events, 0);
1455 if (ret < 0) {
1456 goto error;
1457 }
1458 nb_fd = ret;
1459
1460 for (i = 0; i < nb_fd; i++) {
1461 /* Get faulty FD. */
1462 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1463 int pollfd = LTTNG_POLL_GETFD(&events, i);
1464
1465 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1466 &wait_queue->head, head) {
1467 if (pollfd == wait_node->app->sock &&
1468 (revents & (LPOLLHUP | LPOLLERR))) {
1469 cds_list_del(&wait_node->head);
1470 wait_queue->count--;
1471 ust_app_destroy(wait_node->app);
1472 free(wait_node);
1473 break;
1474 }
1475 }
1476 }
1477
1478 if (nb_fd > 0) {
1479 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1480 }
1481
1482 end:
1483 lttng_poll_clean(&events);
1484 return;
1485
1486 error:
1487 lttng_poll_clean(&events);
1488 error_create:
1489 ERR("Unable to sanitize wait queue");
1490 return;
1491 }
1492
1493 /*
1494 * Dispatch request from the registration threads to the application
1495 * communication thread.
1496 */
1497 static void *thread_dispatch_ust_registration(void *data)
1498 {
1499 int ret, err = -1;
1500 struct cds_wfq_node *node;
1501 struct ust_command *ust_cmd = NULL;
1502 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1503 struct ust_reg_wait_queue wait_queue = {
1504 .count = 0,
1505 };
1506
1507 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1508
1509 health_code_update();
1510
1511 CDS_INIT_LIST_HEAD(&wait_queue.head);
1512
1513 DBG("[thread] Dispatch UST command started");
1514
1515 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1516 health_code_update();
1517
1518 /* Atomically prepare the queue futex */
1519 futex_nto1_prepare(&ust_cmd_queue.futex);
1520
1521 do {
1522 struct ust_app *app = NULL;
1523 ust_cmd = NULL;
1524
1525 /*
1526 * Make sure we don't have node(s) that have hung up before receiving
1527 * the notify socket. This is to clean the list in order to avoid
1528 * memory leaks from notify socket that are never seen.
1529 */
1530 sanitize_wait_queue(&wait_queue);
1531
1532 health_code_update();
1533 /* Dequeue command for registration */
1534 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1535 if (node == NULL) {
1536 DBG("Woken up but nothing in the UST command queue");
1537 /* Continue thread execution */
1538 break;
1539 }
1540
1541 ust_cmd = caa_container_of(node, struct ust_command, node);
1542
1543 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1544 " gid:%d sock:%d name:%s (version %d.%d)",
1545 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1546 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1547 ust_cmd->sock, ust_cmd->reg_msg.name,
1548 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1549
1550 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1551 wait_node = zmalloc(sizeof(*wait_node));
1552 if (!wait_node) {
1553 PERROR("zmalloc wait_node dispatch");
1554 ret = close(ust_cmd->sock);
1555 if (ret < 0) {
1556 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1557 }
1558 lttng_fd_put(1, LTTNG_FD_APPS);
1559 free(ust_cmd);
1560 goto error;
1561 }
1562 CDS_INIT_LIST_HEAD(&wait_node->head);
1563
1564 /* Create application object if socket is CMD. */
1565 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1566 ust_cmd->sock);
1567 if (!wait_node->app) {
1568 ret = close(ust_cmd->sock);
1569 if (ret < 0) {
1570 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1571 }
1572 lttng_fd_put(1, LTTNG_FD_APPS);
1573 free(wait_node);
1574 free(ust_cmd);
1575 continue;
1576 }
1577 /*
1578 * Add application to the wait queue so we can set the notify
1579 * socket before putting this object in the global ht.
1580 */
1581 cds_list_add(&wait_node->head, &wait_queue.head);
1582 wait_queue.count++;
1583
1584 free(ust_cmd);
1585 /*
1586 * We have to continue here since we don't have the notify
1587 * socket and the application MUST be added to the hash table
1588 * only at that moment.
1589 */
1590 continue;
1591 } else {
1592 /*
1593 * Look for the application in the local wait queue and set the
1594 * notify socket if found.
1595 */
1596 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1597 &wait_queue.head, head) {
1598 health_code_update();
1599 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1600 wait_node->app->notify_sock = ust_cmd->sock;
1601 cds_list_del(&wait_node->head);
1602 wait_queue.count--;
1603 app = wait_node->app;
1604 free(wait_node);
1605 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1606 break;
1607 }
1608 }
1609
1610 /*
1611 * With no application at this stage the received socket is
1612 * basically useless so close it before we free the cmd data
1613 * structure for good.
1614 */
1615 if (!app) {
1616 ret = close(ust_cmd->sock);
1617 if (ret < 0) {
1618 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1619 }
1620 lttng_fd_put(1, LTTNG_FD_APPS);
1621 }
1622 free(ust_cmd);
1623 }
1624
1625 if (app) {
1626 /*
1627 * @session_lock_list
1628 *
1629 * Lock the global session list so from the register up to the
1630 * registration done message, no thread can see the application
1631 * and change its state.
1632 */
1633 session_lock_list();
1634 rcu_read_lock();
1635
1636 /*
1637 * Add application to the global hash table. This needs to be
1638 * done before the update to the UST registry can locate the
1639 * application.
1640 */
1641 ust_app_add(app);
1642
1643 /* Set app version. This call will print an error if needed. */
1644 (void) ust_app_version(app);
1645
1646 /* Send notify socket through the notify pipe. */
1647 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1648 app->notify_sock);
1649 if (ret < 0) {
1650 rcu_read_unlock();
1651 session_unlock_list();
1652 /* No notify thread, stop the UST tracing. */
1653 goto error;
1654 }
1655
1656 /*
1657 * Update newly registered application with the tracing
1658 * registry info already enabled information.
1659 */
1660 update_ust_app(app->sock);
1661
1662 /*
1663 * Don't care about return value. Let the manage apps threads
1664 * handle app unregistration upon socket close.
1665 */
1666 (void) ust_app_register_done(app->sock);
1667
1668 /*
1669 * Even if the application socket has been closed, send the app
1670 * to the thread and unregistration will take place at that
1671 * place.
1672 */
1673 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1674 if (ret < 0) {
1675 rcu_read_unlock();
1676 session_unlock_list();
1677 /* No apps. thread, stop the UST tracing. */
1678 goto error;
1679 }
1680
1681 rcu_read_unlock();
1682 session_unlock_list();
1683 }
1684 } while (node != NULL);
1685
1686 health_poll_entry();
1687 /* Futex wait on queue. Blocking call on futex() */
1688 futex_nto1_wait(&ust_cmd_queue.futex);
1689 health_poll_exit();
1690 }
1691 /* Normal exit, no error */
1692 err = 0;
1693
1694 error:
1695 /* Clean up wait queue. */
1696 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1697 &wait_queue.head, head) {
1698 cds_list_del(&wait_node->head);
1699 wait_queue.count--;
1700 free(wait_node);
1701 }
1702
1703 DBG("Dispatch thread dying");
1704 if (err) {
1705 health_error();
1706 ERR("Health error occurred in %s", __func__);
1707 }
1708 health_unregister();
1709 return NULL;
1710 }
1711
1712 /*
1713 * This thread manage application registration.
1714 */
1715 static void *thread_registration_apps(void *data)
1716 {
1717 int sock = -1, i, ret, pollfd, err = -1;
1718 uint32_t revents, nb_fd;
1719 struct lttng_poll_event events;
1720 /*
1721 * Get allocated in this thread, enqueued to a global queue, dequeued and
1722 * freed in the manage apps thread.
1723 */
1724 struct ust_command *ust_cmd = NULL;
1725
1726 DBG("[thread] Manage application registration started");
1727
1728 health_register(HEALTH_TYPE_APP_REG);
1729
1730 if (testpoint(thread_registration_apps)) {
1731 goto error_testpoint;
1732 }
1733
1734 ret = lttcomm_listen_unix_sock(apps_sock);
1735 if (ret < 0) {
1736 goto error_listen;
1737 }
1738
1739 /*
1740 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1741 * more will be added to this poll set.
1742 */
1743 ret = sessiond_set_thread_pollset(&events, 2);
1744 if (ret < 0) {
1745 goto error_create_poll;
1746 }
1747
1748 /* Add the application registration socket */
1749 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1750 if (ret < 0) {
1751 goto error_poll_add;
1752 }
1753
1754 /* Notify all applications to register */
1755 ret = notify_ust_apps(1);
1756 if (ret < 0) {
1757 ERR("Failed to notify applications or create the wait shared memory.\n"
1758 "Execution continues but there might be problem for already\n"
1759 "running applications that wishes to register.");
1760 }
1761
1762 while (1) {
1763 DBG("Accepting application registration");
1764
1765 /* Inifinite blocking call, waiting for transmission */
1766 restart:
1767 health_poll_entry();
1768 ret = lttng_poll_wait(&events, -1);
1769 health_poll_exit();
1770 if (ret < 0) {
1771 /*
1772 * Restart interrupted system call.
1773 */
1774 if (errno == EINTR) {
1775 goto restart;
1776 }
1777 goto error;
1778 }
1779
1780 nb_fd = ret;
1781
1782 for (i = 0; i < nb_fd; i++) {
1783 health_code_update();
1784
1785 /* Fetch once the poll data */
1786 revents = LTTNG_POLL_GETEV(&events, i);
1787 pollfd = LTTNG_POLL_GETFD(&events, i);
1788
1789 /* Thread quit pipe has been closed. Killing thread. */
1790 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1791 if (ret) {
1792 err = 0;
1793 goto exit;
1794 }
1795
1796 /* Event on the registration socket */
1797 if (pollfd == apps_sock) {
1798 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1799 ERR("Register apps socket poll error");
1800 goto error;
1801 } else if (revents & LPOLLIN) {
1802 sock = lttcomm_accept_unix_sock(apps_sock);
1803 if (sock < 0) {
1804 goto error;
1805 }
1806
1807 /*
1808 * Set the CLOEXEC flag. Return code is useless because
1809 * either way, the show must go on.
1810 */
1811 (void) utils_set_fd_cloexec(sock);
1812
1813 /* Create UST registration command for enqueuing */
1814 ust_cmd = zmalloc(sizeof(struct ust_command));
1815 if (ust_cmd == NULL) {
1816 PERROR("ust command zmalloc");
1817 goto error;
1818 }
1819
1820 /*
1821 * Using message-based transmissions to ensure we don't
1822 * have to deal with partially received messages.
1823 */
1824 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1825 if (ret < 0) {
1826 ERR("Exhausted file descriptors allowed for applications.");
1827 free(ust_cmd);
1828 ret = close(sock);
1829 if (ret) {
1830 PERROR("close");
1831 }
1832 sock = -1;
1833 continue;
1834 }
1835
1836 health_code_update();
1837 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1838 if (ret < 0) {
1839 free(ust_cmd);
1840 /* Close socket of the application. */
1841 ret = close(sock);
1842 if (ret) {
1843 PERROR("close");
1844 }
1845 lttng_fd_put(LTTNG_FD_APPS, 1);
1846 sock = -1;
1847 continue;
1848 }
1849 health_code_update();
1850
1851 ust_cmd->sock = sock;
1852 sock = -1;
1853
1854 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1855 " gid:%d sock:%d name:%s (version %d.%d)",
1856 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1857 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1858 ust_cmd->sock, ust_cmd->reg_msg.name,
1859 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1860
1861 /*
1862 * Lock free enqueue the registration request. The red pill
1863 * has been taken! This apps will be part of the *system*.
1864 */
1865 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1866
1867 /*
1868 * Wake the registration queue futex. Implicit memory
1869 * barrier with the exchange in cds_wfq_enqueue.
1870 */
1871 futex_nto1_wake(&ust_cmd_queue.futex);
1872 }
1873 }
1874 }
1875 }
1876
1877 exit:
1878 error:
1879 if (err) {
1880 health_error();
1881 ERR("Health error occurred in %s", __func__);
1882 }
1883
1884 /* Notify that the registration thread is gone */
1885 notify_ust_apps(0);
1886
1887 if (apps_sock >= 0) {
1888 ret = close(apps_sock);
1889 if (ret) {
1890 PERROR("close");
1891 }
1892 }
1893 if (sock >= 0) {
1894 ret = close(sock);
1895 if (ret) {
1896 PERROR("close");
1897 }
1898 lttng_fd_put(LTTNG_FD_APPS, 1);
1899 }
1900 unlink(apps_unix_sock_path);
1901
1902 error_poll_add:
1903 lttng_poll_clean(&events);
1904 error_listen:
1905 error_create_poll:
1906 error_testpoint:
1907 DBG("UST Registration thread cleanup complete");
1908 health_unregister();
1909
1910 return NULL;
1911 }
1912
1913 /*
1914 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1915 * exec or it will fails.
1916 */
1917 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1918 {
1919 int ret, clock_ret;
1920 struct timespec timeout;
1921
1922 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1923 consumer_data->consumer_thread_is_ready = 0;
1924
1925 /* Setup pthread condition */
1926 ret = pthread_condattr_init(&consumer_data->condattr);
1927 if (ret != 0) {
1928 errno = ret;
1929 PERROR("pthread_condattr_init consumer data");
1930 goto error;
1931 }
1932
1933 /*
1934 * Set the monotonic clock in order to make sure we DO NOT jump in time
1935 * between the clock_gettime() call and the timedwait call. See bug #324
1936 * for a more details and how we noticed it.
1937 */
1938 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1939 if (ret != 0) {
1940 errno = ret;
1941 PERROR("pthread_condattr_setclock consumer data");
1942 goto error;
1943 }
1944
1945 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1946 if (ret != 0) {
1947 errno = ret;
1948 PERROR("pthread_cond_init consumer data");
1949 goto error;
1950 }
1951
1952 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1953 consumer_data);
1954 if (ret != 0) {
1955 PERROR("pthread_create consumer");
1956 ret = -1;
1957 goto error;
1958 }
1959
1960 /* We are about to wait on a pthread condition */
1961 pthread_mutex_lock(&consumer_data->cond_mutex);
1962
1963 /* Get time for sem_timedwait absolute timeout */
1964 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1965 /*
1966 * Set the timeout for the condition timed wait even if the clock gettime
1967 * call fails since we might loop on that call and we want to avoid to
1968 * increment the timeout too many times.
1969 */
1970 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1971
1972 /*
1973 * The following loop COULD be skipped in some conditions so this is why we
1974 * set ret to 0 in order to make sure at least one round of the loop is
1975 * done.
1976 */
1977 ret = 0;
1978
1979 /*
1980 * Loop until the condition is reached or when a timeout is reached. Note
1981 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1982 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1983 * possible. This loop does not take any chances and works with both of
1984 * them.
1985 */
1986 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1987 if (clock_ret < 0) {
1988 PERROR("clock_gettime spawn consumer");
1989 /* Infinite wait for the consumerd thread to be ready */
1990 ret = pthread_cond_wait(&consumer_data->cond,
1991 &consumer_data->cond_mutex);
1992 } else {
1993 ret = pthread_cond_timedwait(&consumer_data->cond,
1994 &consumer_data->cond_mutex, &timeout);
1995 }
1996 }
1997
1998 /* Release the pthread condition */
1999 pthread_mutex_unlock(&consumer_data->cond_mutex);
2000
2001 if (ret != 0) {
2002 errno = ret;
2003 if (ret == ETIMEDOUT) {
2004 /*
2005 * Call has timed out so we kill the kconsumerd_thread and return
2006 * an error.
2007 */
2008 ERR("Condition timed out. The consumer thread was never ready."
2009 " Killing it");
2010 ret = pthread_cancel(consumer_data->thread);
2011 if (ret < 0) {
2012 PERROR("pthread_cancel consumer thread");
2013 }
2014 } else {
2015 PERROR("pthread_cond_wait failed consumer thread");
2016 }
2017 goto error;
2018 }
2019
2020 pthread_mutex_lock(&consumer_data->pid_mutex);
2021 if (consumer_data->pid == 0) {
2022 ERR("Consumerd did not start");
2023 pthread_mutex_unlock(&consumer_data->pid_mutex);
2024 goto error;
2025 }
2026 pthread_mutex_unlock(&consumer_data->pid_mutex);
2027
2028 return 0;
2029
2030 error:
2031 return ret;
2032 }
2033
2034 /*
2035 * Join consumer thread
2036 */
2037 static int join_consumer_thread(struct consumer_data *consumer_data)
2038 {
2039 void *status;
2040
2041 /* Consumer pid must be a real one. */
2042 if (consumer_data->pid > 0) {
2043 int ret;
2044 ret = kill(consumer_data->pid, SIGTERM);
2045 if (ret) {
2046 ERR("Error killing consumer daemon");
2047 return ret;
2048 }
2049 return pthread_join(consumer_data->thread, &status);
2050 } else {
2051 return 0;
2052 }
2053 }
2054
2055 /*
2056 * Fork and exec a consumer daemon (consumerd).
2057 *
2058 * Return pid if successful else -1.
2059 */
2060 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2061 {
2062 int ret;
2063 pid_t pid;
2064 const char *consumer_to_use;
2065 const char *verbosity;
2066 struct stat st;
2067
2068 DBG("Spawning consumerd");
2069
2070 pid = fork();
2071 if (pid == 0) {
2072 /*
2073 * Exec consumerd.
2074 */
2075 if (opt_verbose_consumer) {
2076 verbosity = "--verbose";
2077 } else {
2078 verbosity = "--quiet";
2079 }
2080 switch (consumer_data->type) {
2081 case LTTNG_CONSUMER_KERNEL:
2082 /*
2083 * Find out which consumerd to execute. We will first try the
2084 * 64-bit path, then the sessiond's installation directory, and
2085 * fallback on the 32-bit one,
2086 */
2087 DBG3("Looking for a kernel consumer at these locations:");
2088 DBG3(" 1) %s", consumerd64_bin);
2089 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2090 DBG3(" 3) %s", consumerd32_bin);
2091 if (stat(consumerd64_bin, &st) == 0) {
2092 DBG3("Found location #1");
2093 consumer_to_use = consumerd64_bin;
2094 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2095 DBG3("Found location #2");
2096 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2097 } else if (stat(consumerd32_bin, &st) == 0) {
2098 DBG3("Found location #3");
2099 consumer_to_use = consumerd32_bin;
2100 } else {
2101 DBG("Could not find any valid consumerd executable");
2102 break;
2103 }
2104 DBG("Using kernel consumer at: %s", consumer_to_use);
2105 execl(consumer_to_use,
2106 "lttng-consumerd", verbosity, "-k",
2107 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2108 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2109 NULL);
2110 break;
2111 case LTTNG_CONSUMER64_UST:
2112 {
2113 char *tmpnew = NULL;
2114
2115 if (consumerd64_libdir[0] != '\0') {
2116 char *tmp;
2117 size_t tmplen;
2118
2119 tmp = getenv("LD_LIBRARY_PATH");
2120 if (!tmp) {
2121 tmp = "";
2122 }
2123 tmplen = strlen("LD_LIBRARY_PATH=")
2124 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2125 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2126 if (!tmpnew) {
2127 ret = -ENOMEM;
2128 goto error;
2129 }
2130 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2131 strcat(tmpnew, consumerd64_libdir);
2132 if (tmp[0] != '\0') {
2133 strcat(tmpnew, ":");
2134 strcat(tmpnew, tmp);
2135 }
2136 ret = putenv(tmpnew);
2137 if (ret) {
2138 ret = -errno;
2139 free(tmpnew);
2140 goto error;
2141 }
2142 }
2143 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2144 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2145 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2146 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2147 NULL);
2148 if (consumerd64_libdir[0] != '\0') {
2149 free(tmpnew);
2150 }
2151 if (ret) {
2152 goto error;
2153 }
2154 break;
2155 }
2156 case LTTNG_CONSUMER32_UST:
2157 {
2158 char *tmpnew = NULL;
2159
2160 if (consumerd32_libdir[0] != '\0') {
2161 char *tmp;
2162 size_t tmplen;
2163
2164 tmp = getenv("LD_LIBRARY_PATH");
2165 if (!tmp) {
2166 tmp = "";
2167 }
2168 tmplen = strlen("LD_LIBRARY_PATH=")
2169 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2170 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2171 if (!tmpnew) {
2172 ret = -ENOMEM;
2173 goto error;
2174 }
2175 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2176 strcat(tmpnew, consumerd32_libdir);
2177 if (tmp[0] != '\0') {
2178 strcat(tmpnew, ":");
2179 strcat(tmpnew, tmp);
2180 }
2181 ret = putenv(tmpnew);
2182 if (ret) {
2183 ret = -errno;
2184 free(tmpnew);
2185 goto error;
2186 }
2187 }
2188 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2189 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2190 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2191 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2192 NULL);
2193 if (consumerd32_libdir[0] != '\0') {
2194 free(tmpnew);
2195 }
2196 if (ret) {
2197 goto error;
2198 }
2199 break;
2200 }
2201 default:
2202 PERROR("unknown consumer type");
2203 exit(EXIT_FAILURE);
2204 }
2205 if (errno != 0) {
2206 PERROR("kernel start consumer exec");
2207 }
2208 exit(EXIT_FAILURE);
2209 } else if (pid > 0) {
2210 ret = pid;
2211 } else {
2212 PERROR("start consumer fork");
2213 ret = -errno;
2214 }
2215 error:
2216 return ret;
2217 }
2218
2219 /*
2220 * Spawn the consumerd daemon and session daemon thread.
2221 */
2222 static int start_consumerd(struct consumer_data *consumer_data)
2223 {
2224 int ret;
2225
2226 /*
2227 * Set the listen() state on the socket since there is a possible race
2228 * between the exec() of the consumer daemon and this call if place in the
2229 * consumer thread. See bug #366 for more details.
2230 */
2231 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2232 if (ret < 0) {
2233 goto error;
2234 }
2235
2236 pthread_mutex_lock(&consumer_data->pid_mutex);
2237 if (consumer_data->pid != 0) {
2238 pthread_mutex_unlock(&consumer_data->pid_mutex);
2239 goto end;
2240 }
2241
2242 ret = spawn_consumerd(consumer_data);
2243 if (ret < 0) {
2244 ERR("Spawning consumerd failed");
2245 pthread_mutex_unlock(&consumer_data->pid_mutex);
2246 goto error;
2247 }
2248
2249 /* Setting up the consumer_data pid */
2250 consumer_data->pid = ret;
2251 DBG2("Consumer pid %d", consumer_data->pid);
2252 pthread_mutex_unlock(&consumer_data->pid_mutex);
2253
2254 DBG2("Spawning consumer control thread");
2255 ret = spawn_consumer_thread(consumer_data);
2256 if (ret < 0) {
2257 ERR("Fatal error spawning consumer control thread");
2258 goto error;
2259 }
2260
2261 end:
2262 return 0;
2263
2264 error:
2265 /* Cleanup already created sockets on error. */
2266 if (consumer_data->err_sock >= 0) {
2267 int err;
2268
2269 err = close(consumer_data->err_sock);
2270 if (err < 0) {
2271 PERROR("close consumer data error socket");
2272 }
2273 }
2274 return ret;
2275 }
2276
2277 /*
2278 * Compute health status of each consumer. If one of them is zero (bad
2279 * state), we return 0.
2280 */
2281 static int check_consumer_health(void)
2282 {
2283 int ret;
2284
2285 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2286
2287 DBG3("Health consumer check %d", ret);
2288
2289 return ret;
2290 }
2291
2292 /*
2293 * Setup necessary data for kernel tracer action.
2294 */
2295 static int init_kernel_tracer(void)
2296 {
2297 int ret;
2298
2299 /* Modprobe lttng kernel modules */
2300 ret = modprobe_lttng_control();
2301 if (ret < 0) {
2302 goto error;
2303 }
2304
2305 /* Open debugfs lttng */
2306 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2307 if (kernel_tracer_fd < 0) {
2308 DBG("Failed to open %s", module_proc_lttng);
2309 ret = -1;
2310 goto error_open;
2311 }
2312
2313 /* Validate kernel version */
2314 ret = kernel_validate_version(kernel_tracer_fd);
2315 if (ret < 0) {
2316 goto error_version;
2317 }
2318
2319 ret = modprobe_lttng_data();
2320 if (ret < 0) {
2321 goto error_modules;
2322 }
2323
2324 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2325 return 0;
2326
2327 error_version:
2328 modprobe_remove_lttng_control();
2329 ret = close(kernel_tracer_fd);
2330 if (ret) {
2331 PERROR("close");
2332 }
2333 kernel_tracer_fd = -1;
2334 return LTTNG_ERR_KERN_VERSION;
2335
2336 error_modules:
2337 ret = close(kernel_tracer_fd);
2338 if (ret) {
2339 PERROR("close");
2340 }
2341
2342 error_open:
2343 modprobe_remove_lttng_control();
2344
2345 error:
2346 WARN("No kernel tracer available");
2347 kernel_tracer_fd = -1;
2348 if (!is_root) {
2349 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2350 } else {
2351 return LTTNG_ERR_KERN_NA;
2352 }
2353 }
2354
2355
2356 /*
2357 * Copy consumer output from the tracing session to the domain session. The
2358 * function also applies the right modification on a per domain basis for the
2359 * trace files destination directory.
2360 *
2361 * Should *NOT* be called with RCU read-side lock held.
2362 */
2363 static int copy_session_consumer(int domain, struct ltt_session *session)
2364 {
2365 int ret;
2366 const char *dir_name;
2367 struct consumer_output *consumer;
2368
2369 assert(session);
2370 assert(session->consumer);
2371
2372 switch (domain) {
2373 case LTTNG_DOMAIN_KERNEL:
2374 DBG3("Copying tracing session consumer output in kernel session");
2375 /*
2376 * XXX: We should audit the session creation and what this function
2377 * does "extra" in order to avoid a destroy since this function is used
2378 * in the domain session creation (kernel and ust) only. Same for UST
2379 * domain.
2380 */
2381 if (session->kernel_session->consumer) {
2382 consumer_destroy_output(session->kernel_session->consumer);
2383 }
2384 session->kernel_session->consumer =
2385 consumer_copy_output(session->consumer);
2386 /* Ease our life a bit for the next part */
2387 consumer = session->kernel_session->consumer;
2388 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2389 break;
2390 case LTTNG_DOMAIN_UST:
2391 DBG3("Copying tracing session consumer output in UST session");
2392 if (session->ust_session->consumer) {
2393 consumer_destroy_output(session->ust_session->consumer);
2394 }
2395 session->ust_session->consumer =
2396 consumer_copy_output(session->consumer);
2397 /* Ease our life a bit for the next part */
2398 consumer = session->ust_session->consumer;
2399 dir_name = DEFAULT_UST_TRACE_DIR;
2400 break;
2401 default:
2402 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2403 goto error;
2404 }
2405
2406 /* Append correct directory to subdir */
2407 strncat(consumer->subdir, dir_name,
2408 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2409 DBG3("Copy session consumer subdir %s", consumer->subdir);
2410
2411 ret = LTTNG_OK;
2412
2413 error:
2414 return ret;
2415 }
2416
2417 /*
2418 * Create an UST session and add it to the session ust list.
2419 *
2420 * Should *NOT* be called with RCU read-side lock held.
2421 */
2422 static int create_ust_session(struct ltt_session *session,
2423 struct lttng_domain *domain)
2424 {
2425 int ret;
2426 struct ltt_ust_session *lus = NULL;
2427
2428 assert(session);
2429 assert(domain);
2430 assert(session->consumer);
2431
2432 switch (domain->type) {
2433 case LTTNG_DOMAIN_UST:
2434 break;
2435 default:
2436 ERR("Unknown UST domain on create session %d", domain->type);
2437 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2438 goto error;
2439 }
2440
2441 DBG("Creating UST session");
2442
2443 lus = trace_ust_create_session(session->id);
2444 if (lus == NULL) {
2445 ret = LTTNG_ERR_UST_SESS_FAIL;
2446 goto error;
2447 }
2448
2449 lus->uid = session->uid;
2450 lus->gid = session->gid;
2451 lus->output_traces = session->output_traces;
2452 lus->snapshot_mode = session->snapshot_mode;
2453 session->ust_session = lus;
2454
2455 /* Copy session output to the newly created UST session */
2456 ret = copy_session_consumer(domain->type, session);
2457 if (ret != LTTNG_OK) {
2458 goto error;
2459 }
2460
2461 return LTTNG_OK;
2462
2463 error:
2464 free(lus);
2465 session->ust_session = NULL;
2466 return ret;
2467 }
2468
2469 /*
2470 * Create a kernel tracer session then create the default channel.
2471 */
2472 static int create_kernel_session(struct ltt_session *session)
2473 {
2474 int ret;
2475
2476 DBG("Creating kernel session");
2477
2478 ret = kernel_create_session(session, kernel_tracer_fd);
2479 if (ret < 0) {
2480 ret = LTTNG_ERR_KERN_SESS_FAIL;
2481 goto error;
2482 }
2483
2484 /* Code flow safety */
2485 assert(session->kernel_session);
2486
2487 /* Copy session output to the newly created Kernel session */
2488 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2489 if (ret != LTTNG_OK) {
2490 goto error;
2491 }
2492
2493 /* Create directory(ies) on local filesystem. */
2494 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2495 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2496 ret = run_as_mkdir_recursive(
2497 session->kernel_session->consumer->dst.trace_path,
2498 S_IRWXU | S_IRWXG, session->uid, session->gid);
2499 if (ret < 0) {
2500 if (ret != -EEXIST) {
2501 ERR("Trace directory creation error");
2502 goto error;
2503 }
2504 }
2505 }
2506
2507 session->kernel_session->uid = session->uid;
2508 session->kernel_session->gid = session->gid;
2509 session->kernel_session->output_traces = session->output_traces;
2510 session->kernel_session->snapshot_mode = session->snapshot_mode;
2511
2512 return LTTNG_OK;
2513
2514 error:
2515 trace_kernel_destroy_session(session->kernel_session);
2516 session->kernel_session = NULL;
2517 return ret;
2518 }
2519
2520 /*
2521 * Count number of session permitted by uid/gid.
2522 */
2523 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2524 {
2525 unsigned int i = 0;
2526 struct ltt_session *session;
2527
2528 DBG("Counting number of available session for UID %d GID %d",
2529 uid, gid);
2530 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2531 /*
2532 * Only list the sessions the user can control.
2533 */
2534 if (!session_access_ok(session, uid, gid)) {
2535 continue;
2536 }
2537 i++;
2538 }
2539 return i;
2540 }
2541
2542 /*
2543 * Process the command requested by the lttng client within the command
2544 * context structure. This function make sure that the return structure (llm)
2545 * is set and ready for transmission before returning.
2546 *
2547 * Return any error encountered or 0 for success.
2548 *
2549 * "sock" is only used for special-case var. len data.
2550 *
2551 * Should *NOT* be called with RCU read-side lock held.
2552 */
2553 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2554 int *sock_error)
2555 {
2556 int ret = LTTNG_OK;
2557 int need_tracing_session = 1;
2558 int need_domain;
2559
2560 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2561
2562 *sock_error = 0;
2563
2564 switch (cmd_ctx->lsm->cmd_type) {
2565 case LTTNG_CREATE_SESSION:
2566 case LTTNG_CREATE_SESSION_SNAPSHOT:
2567 case LTTNG_DESTROY_SESSION:
2568 case LTTNG_LIST_SESSIONS:
2569 case LTTNG_LIST_DOMAINS:
2570 case LTTNG_START_TRACE:
2571 case LTTNG_STOP_TRACE:
2572 case LTTNG_DATA_PENDING:
2573 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2574 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2575 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2576 case LTTNG_SNAPSHOT_RECORD:
2577 need_domain = 0;
2578 break;
2579 default:
2580 need_domain = 1;
2581 }
2582
2583 if (opt_no_kernel && need_domain
2584 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2585 if (!is_root) {
2586 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2587 } else {
2588 ret = LTTNG_ERR_KERN_NA;
2589 }
2590 goto error;
2591 }
2592
2593 /* Deny register consumer if we already have a spawned consumer. */
2594 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2595 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2596 if (kconsumer_data.pid > 0) {
2597 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2598 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2599 goto error;
2600 }
2601 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2602 }
2603
2604 /*
2605 * Check for command that don't needs to allocate a returned payload. We do
2606 * this here so we don't have to make the call for no payload at each
2607 * command.
2608 */
2609 switch(cmd_ctx->lsm->cmd_type) {
2610 case LTTNG_LIST_SESSIONS:
2611 case LTTNG_LIST_TRACEPOINTS:
2612 case LTTNG_LIST_TRACEPOINT_FIELDS:
2613 case LTTNG_LIST_DOMAINS:
2614 case LTTNG_LIST_CHANNELS:
2615 case LTTNG_LIST_EVENTS:
2616 break;
2617 default:
2618 /* Setup lttng message with no payload */
2619 ret = setup_lttng_msg(cmd_ctx, 0);
2620 if (ret < 0) {
2621 /* This label does not try to unlock the session */
2622 goto init_setup_error;
2623 }
2624 }
2625
2626 /* Commands that DO NOT need a session. */
2627 switch (cmd_ctx->lsm->cmd_type) {
2628 case LTTNG_CREATE_SESSION:
2629 case LTTNG_CREATE_SESSION_SNAPSHOT:
2630 case LTTNG_CALIBRATE:
2631 case LTTNG_LIST_SESSIONS:
2632 case LTTNG_LIST_TRACEPOINTS:
2633 case LTTNG_LIST_TRACEPOINT_FIELDS:
2634 need_tracing_session = 0;
2635 break;
2636 default:
2637 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2638 /*
2639 * We keep the session list lock across _all_ commands
2640 * for now, because the per-session lock does not
2641 * handle teardown properly.
2642 */
2643 session_lock_list();
2644 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2645 if (cmd_ctx->session == NULL) {
2646 ret = LTTNG_ERR_SESS_NOT_FOUND;
2647 goto error;
2648 } else {
2649 /* Acquire lock for the session */
2650 session_lock(cmd_ctx->session);
2651 }
2652 break;
2653 }
2654
2655 if (!need_domain) {
2656 goto skip_domain;
2657 }
2658
2659 /*
2660 * Check domain type for specific "pre-action".
2661 */
2662 switch (cmd_ctx->lsm->domain.type) {
2663 case LTTNG_DOMAIN_KERNEL:
2664 if (!is_root) {
2665 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2666 goto error;
2667 }
2668
2669 /* Kernel tracer check */
2670 if (kernel_tracer_fd == -1) {
2671 /* Basically, load kernel tracer modules */
2672 ret = init_kernel_tracer();
2673 if (ret != 0) {
2674 goto error;
2675 }
2676 }
2677
2678 /* Consumer is in an ERROR state. Report back to client */
2679 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2680 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2681 goto error;
2682 }
2683
2684 /* Need a session for kernel command */
2685 if (need_tracing_session) {
2686 if (cmd_ctx->session->kernel_session == NULL) {
2687 ret = create_kernel_session(cmd_ctx->session);
2688 if (ret < 0) {
2689 ret = LTTNG_ERR_KERN_SESS_FAIL;
2690 goto error;
2691 }
2692 }
2693
2694 /* Start the kernel consumer daemon */
2695 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2696 if (kconsumer_data.pid == 0 &&
2697 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2698 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2699 ret = start_consumerd(&kconsumer_data);
2700 if (ret < 0) {
2701 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2702 goto error;
2703 }
2704 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2705 } else {
2706 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2707 }
2708
2709 /*
2710 * The consumer was just spawned so we need to add the socket to
2711 * the consumer output of the session if exist.
2712 */
2713 ret = consumer_create_socket(&kconsumer_data,
2714 cmd_ctx->session->kernel_session->consumer);
2715 if (ret < 0) {
2716 goto error;
2717 }
2718 }
2719
2720 break;
2721 case LTTNG_DOMAIN_UST:
2722 {
2723 if (!ust_app_supported()) {
2724 ret = LTTNG_ERR_NO_UST;
2725 goto error;
2726 }
2727 /* Consumer is in an ERROR state. Report back to client */
2728 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2729 ret = LTTNG_ERR_NO_USTCONSUMERD;
2730 goto error;
2731 }
2732
2733 if (need_tracing_session) {
2734 /* Create UST session if none exist. */
2735 if (cmd_ctx->session->ust_session == NULL) {
2736 ret = create_ust_session(cmd_ctx->session,
2737 &cmd_ctx->lsm->domain);
2738 if (ret != LTTNG_OK) {
2739 goto error;
2740 }
2741 }
2742
2743 /* Start the UST consumer daemons */
2744 /* 64-bit */
2745 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2746 if (consumerd64_bin[0] != '\0' &&
2747 ustconsumer64_data.pid == 0 &&
2748 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2749 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2750 ret = start_consumerd(&ustconsumer64_data);
2751 if (ret < 0) {
2752 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2753 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2754 goto error;
2755 }
2756
2757 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2758 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2759 } else {
2760 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2761 }
2762
2763 /*
2764 * Setup socket for consumer 64 bit. No need for atomic access
2765 * since it was set above and can ONLY be set in this thread.
2766 */
2767 ret = consumer_create_socket(&ustconsumer64_data,
2768 cmd_ctx->session->ust_session->consumer);
2769 if (ret < 0) {
2770 goto error;
2771 }
2772
2773 /* 32-bit */
2774 if (consumerd32_bin[0] != '\0' &&
2775 ustconsumer32_data.pid == 0 &&
2776 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2777 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2778 ret = start_consumerd(&ustconsumer32_data);
2779 if (ret < 0) {
2780 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2781 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2782 goto error;
2783 }
2784
2785 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2786 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2787 } else {
2788 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2789 }
2790
2791 /*
2792 * Setup socket for consumer 64 bit. No need for atomic access
2793 * since it was set above and can ONLY be set in this thread.
2794 */
2795 ret = consumer_create_socket(&ustconsumer32_data,
2796 cmd_ctx->session->ust_session->consumer);
2797 if (ret < 0) {
2798 goto error;
2799 }
2800 }
2801 break;
2802 }
2803 default:
2804 break;
2805 }
2806 skip_domain:
2807
2808 /* Validate consumer daemon state when start/stop trace command */
2809 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2810 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2811 switch (cmd_ctx->lsm->domain.type) {
2812 case LTTNG_DOMAIN_UST:
2813 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2814 ret = LTTNG_ERR_NO_USTCONSUMERD;
2815 goto error;
2816 }
2817 break;
2818 case LTTNG_DOMAIN_KERNEL:
2819 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2820 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2821 goto error;
2822 }
2823 break;
2824 }
2825 }
2826
2827 /*
2828 * Check that the UID or GID match that of the tracing session.
2829 * The root user can interact with all sessions.
2830 */
2831 if (need_tracing_session) {
2832 if (!session_access_ok(cmd_ctx->session,
2833 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2834 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2835 ret = LTTNG_ERR_EPERM;
2836 goto error;
2837 }
2838 }
2839
2840 /*
2841 * Send relayd information to consumer as soon as we have a domain and a
2842 * session defined.
2843 */
2844 if (cmd_ctx->session && need_domain) {
2845 /*
2846 * Setup relayd if not done yet. If the relayd information was already
2847 * sent to the consumer, this call will gracefully return.
2848 */
2849 ret = cmd_setup_relayd(cmd_ctx->session);
2850 if (ret != LTTNG_OK) {
2851 goto error;
2852 }
2853 }
2854
2855 /* Process by command type */
2856 switch (cmd_ctx->lsm->cmd_type) {
2857 case LTTNG_ADD_CONTEXT:
2858 {
2859 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2860 cmd_ctx->lsm->u.context.channel_name,
2861 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2862 break;
2863 }
2864 case LTTNG_DISABLE_CHANNEL:
2865 {
2866 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2867 cmd_ctx->lsm->u.disable.channel_name);
2868 break;
2869 }
2870 case LTTNG_DISABLE_EVENT:
2871 {
2872 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2873 cmd_ctx->lsm->u.disable.channel_name,
2874 cmd_ctx->lsm->u.disable.name);
2875 break;
2876 }
2877 case LTTNG_DISABLE_ALL_EVENT:
2878 {
2879 DBG("Disabling all events");
2880
2881 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2882 cmd_ctx->lsm->u.disable.channel_name);
2883 break;
2884 }
2885 case LTTNG_ENABLE_CHANNEL:
2886 {
2887 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2888 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2889 break;
2890 }
2891 case LTTNG_ENABLE_EVENT:
2892 {
2893 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2894 cmd_ctx->lsm->u.enable.channel_name,
2895 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2896 break;
2897 }
2898 case LTTNG_ENABLE_ALL_EVENT:
2899 {
2900 DBG("Enabling all events");
2901
2902 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2903 cmd_ctx->lsm->u.enable.channel_name,
2904 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2905 break;
2906 }
2907 case LTTNG_LIST_TRACEPOINTS:
2908 {
2909 struct lttng_event *events;
2910 ssize_t nb_events;
2911
2912 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2913 if (nb_events < 0) {
2914 /* Return value is a negative lttng_error_code. */
2915 ret = -nb_events;
2916 goto error;
2917 }
2918
2919 /*
2920 * Setup lttng message with payload size set to the event list size in
2921 * bytes and then copy list into the llm payload.
2922 */
2923 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2924 if (ret < 0) {
2925 free(events);
2926 goto setup_error;
2927 }
2928
2929 /* Copy event list into message payload */
2930 memcpy(cmd_ctx->llm->payload, events,
2931 sizeof(struct lttng_event) * nb_events);
2932
2933 free(events);
2934
2935 ret = LTTNG_OK;
2936 break;
2937 }
2938 case LTTNG_LIST_TRACEPOINT_FIELDS:
2939 {
2940 struct lttng_event_field *fields;
2941 ssize_t nb_fields;
2942
2943 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2944 &fields);
2945 if (nb_fields < 0) {
2946 /* Return value is a negative lttng_error_code. */
2947 ret = -nb_fields;
2948 goto error;
2949 }
2950
2951 /*
2952 * Setup lttng message with payload size set to the event list size in
2953 * bytes and then copy list into the llm payload.
2954 */
2955 ret = setup_lttng_msg(cmd_ctx,
2956 sizeof(struct lttng_event_field) * nb_fields);
2957 if (ret < 0) {
2958 free(fields);
2959 goto setup_error;
2960 }
2961
2962 /* Copy event list into message payload */
2963 memcpy(cmd_ctx->llm->payload, fields,
2964 sizeof(struct lttng_event_field) * nb_fields);
2965
2966 free(fields);
2967
2968 ret = LTTNG_OK;
2969 break;
2970 }
2971 case LTTNG_SET_CONSUMER_URI:
2972 {
2973 size_t nb_uri, len;
2974 struct lttng_uri *uris;
2975
2976 nb_uri = cmd_ctx->lsm->u.uri.size;
2977 len = nb_uri * sizeof(struct lttng_uri);
2978
2979 if (nb_uri == 0) {
2980 ret = LTTNG_ERR_INVALID;
2981 goto error;
2982 }
2983
2984 uris = zmalloc(len);
2985 if (uris == NULL) {
2986 ret = LTTNG_ERR_FATAL;
2987 goto error;
2988 }
2989
2990 /* Receive variable len data */
2991 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2992 ret = lttcomm_recv_unix_sock(sock, uris, len);
2993 if (ret <= 0) {
2994 DBG("No URIs received from client... continuing");
2995 *sock_error = 1;
2996 ret = LTTNG_ERR_SESSION_FAIL;
2997 free(uris);
2998 goto error;
2999 }
3000
3001 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3002 nb_uri, uris);
3003 if (ret != LTTNG_OK) {
3004 free(uris);
3005 goto error;
3006 }
3007
3008 /*
3009 * XXX: 0 means that this URI should be applied on the session. Should
3010 * be a DOMAIN enuam.
3011 */
3012 if (cmd_ctx->lsm->domain.type == 0) {
3013 /* Add the URI for the UST session if a consumer is present. */
3014 if (cmd_ctx->session->ust_session &&
3015 cmd_ctx->session->ust_session->consumer) {
3016 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3017 nb_uri, uris);
3018 } else if (cmd_ctx->session->kernel_session &&
3019 cmd_ctx->session->kernel_session->consumer) {
3020 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3021 cmd_ctx->session, nb_uri, uris);
3022 }
3023 }
3024
3025 free(uris);
3026
3027 break;
3028 }
3029 case LTTNG_START_TRACE:
3030 {
3031 ret = cmd_start_trace(cmd_ctx->session);
3032 break;
3033 }
3034 case LTTNG_STOP_TRACE:
3035 {
3036 ret = cmd_stop_trace(cmd_ctx->session);
3037 break;
3038 }
3039 case LTTNG_CREATE_SESSION:
3040 {
3041 size_t nb_uri, len;
3042 struct lttng_uri *uris = NULL;
3043
3044 nb_uri = cmd_ctx->lsm->u.uri.size;
3045 len = nb_uri * sizeof(struct lttng_uri);
3046
3047 if (nb_uri > 0) {
3048 uris = zmalloc(len);
3049 if (uris == NULL) {
3050 ret = LTTNG_ERR_FATAL;
3051 goto error;
3052 }
3053
3054 /* Receive variable len data */
3055 DBG("Waiting for %zu URIs from client ...", nb_uri);
3056 ret = lttcomm_recv_unix_sock(sock, uris, len);
3057 if (ret <= 0) {
3058 DBG("No URIs received from client... continuing");
3059 *sock_error = 1;
3060 ret = LTTNG_ERR_SESSION_FAIL;
3061 free(uris);
3062 goto error;
3063 }
3064
3065 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3066 DBG("Creating session with ONE network URI is a bad call");
3067 ret = LTTNG_ERR_SESSION_FAIL;
3068 free(uris);
3069 goto error;
3070 }
3071 }
3072
3073 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3074 &cmd_ctx->creds);
3075
3076 free(uris);
3077
3078 break;
3079 }
3080 case LTTNG_DESTROY_SESSION:
3081 {
3082 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3083
3084 /* Set session to NULL so we do not unlock it after free. */
3085 cmd_ctx->session = NULL;
3086 break;
3087 }
3088 case LTTNG_LIST_DOMAINS:
3089 {
3090 ssize_t nb_dom;
3091 struct lttng_domain *domains;
3092
3093 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3094 if (nb_dom < 0) {
3095 /* Return value is a negative lttng_error_code. */
3096 ret = -nb_dom;
3097 goto error;
3098 }
3099
3100 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3101 if (ret < 0) {
3102 free(domains);
3103 goto setup_error;
3104 }
3105
3106 /* Copy event list into message payload */
3107 memcpy(cmd_ctx->llm->payload, domains,
3108 nb_dom * sizeof(struct lttng_domain));
3109
3110 free(domains);
3111
3112 ret = LTTNG_OK;
3113 break;
3114 }
3115 case LTTNG_LIST_CHANNELS:
3116 {
3117 int nb_chan;
3118 struct lttng_channel *channels;
3119
3120 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3121 cmd_ctx->session, &channels);
3122 if (nb_chan < 0) {
3123 /* Return value is a negative lttng_error_code. */
3124 ret = -nb_chan;
3125 goto error;
3126 }
3127
3128 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3129 if (ret < 0) {
3130 free(channels);
3131 goto setup_error;
3132 }
3133
3134 /* Copy event list into message payload */
3135 memcpy(cmd_ctx->llm->payload, channels,
3136 nb_chan * sizeof(struct lttng_channel));
3137
3138 free(channels);
3139
3140 ret = LTTNG_OK;
3141 break;
3142 }
3143 case LTTNG_LIST_EVENTS:
3144 {
3145 ssize_t nb_event;
3146 struct lttng_event *events = NULL;
3147
3148 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3149 cmd_ctx->lsm->u.list.channel_name, &events);
3150 if (nb_event < 0) {
3151 /* Return value is a negative lttng_error_code. */
3152 ret = -nb_event;
3153 goto error;
3154 }
3155
3156 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3157 if (ret < 0) {
3158 free(events);
3159 goto setup_error;
3160 }
3161
3162 /* Copy event list into message payload */
3163 memcpy(cmd_ctx->llm->payload, events,
3164 nb_event * sizeof(struct lttng_event));
3165
3166 free(events);
3167
3168 ret = LTTNG_OK;
3169 break;
3170 }
3171 case LTTNG_LIST_SESSIONS:
3172 {
3173 unsigned int nr_sessions;
3174
3175 session_lock_list();
3176 nr_sessions = lttng_sessions_count(
3177 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3178 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3179
3180 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3181 if (ret < 0) {
3182 session_unlock_list();
3183 goto setup_error;
3184 }
3185
3186 /* Filled the session array */
3187 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3188 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3189 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3190
3191 session_unlock_list();
3192
3193 ret = LTTNG_OK;
3194 break;
3195 }
3196 case LTTNG_CALIBRATE:
3197 {
3198 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3199 &cmd_ctx->lsm->u.calibrate);
3200 break;
3201 }
3202 case LTTNG_REGISTER_CONSUMER:
3203 {
3204 struct consumer_data *cdata;
3205
3206 switch (cmd_ctx->lsm->domain.type) {
3207 case LTTNG_DOMAIN_KERNEL:
3208 cdata = &kconsumer_data;
3209 break;
3210 default:
3211 ret = LTTNG_ERR_UND;
3212 goto error;
3213 }
3214
3215 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3216 cmd_ctx->lsm->u.reg.path, cdata);
3217 break;
3218 }
3219 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3220 {
3221 struct lttng_filter_bytecode *bytecode;
3222
3223 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3224 ret = LTTNG_ERR_FILTER_INVAL;
3225 goto error;
3226 }
3227 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3228 ret = LTTNG_ERR_FILTER_INVAL;
3229 goto error;
3230 }
3231 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3232 if (!bytecode) {
3233 ret = LTTNG_ERR_FILTER_NOMEM;
3234 goto error;
3235 }
3236 /* Receive var. len. data */
3237 DBG("Receiving var len data from client ...");
3238 ret = lttcomm_recv_unix_sock(sock, bytecode,
3239 cmd_ctx->lsm->u.enable.bytecode_len);
3240 if (ret <= 0) {
3241 DBG("Nothing recv() from client var len data... continuing");
3242 *sock_error = 1;
3243 ret = LTTNG_ERR_FILTER_INVAL;
3244 goto error;
3245 }
3246
3247 if (bytecode->len + sizeof(*bytecode)
3248 != cmd_ctx->lsm->u.enable.bytecode_len) {
3249 free(bytecode);
3250 ret = LTTNG_ERR_FILTER_INVAL;
3251 goto error;
3252 }
3253
3254 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3255 cmd_ctx->lsm->u.enable.channel_name,
3256 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3257 break;
3258 }
3259 case LTTNG_DATA_PENDING:
3260 {
3261 ret = cmd_data_pending(cmd_ctx->session);
3262 break;
3263 }
3264 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3265 {
3266 struct lttcomm_lttng_output_id reply;
3267
3268 ret = cmd_snapshot_add_output(cmd_ctx->session,
3269 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3270 if (ret != LTTNG_OK) {
3271 goto error;
3272 }
3273
3274 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3275 if (ret < 0) {
3276 goto setup_error;
3277 }
3278
3279 /* Copy output list into message payload */
3280 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3281 ret = LTTNG_OK;
3282 break;
3283 }
3284 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3285 {
3286 ret = cmd_snapshot_del_output(cmd_ctx->session,
3287 &cmd_ctx->lsm->u.snapshot_output.output);
3288 break;
3289 }
3290 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3291 {
3292 ssize_t nb_output;
3293 struct lttng_snapshot_output *outputs = NULL;
3294
3295 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3296 if (nb_output < 0) {
3297 ret = -nb_output;
3298 goto error;
3299 }
3300
3301 ret = setup_lttng_msg(cmd_ctx,
3302 nb_output * sizeof(struct lttng_snapshot_output));
3303 if (ret < 0) {
3304 free(outputs);
3305 goto setup_error;
3306 }
3307
3308 if (outputs) {
3309 /* Copy output list into message payload */
3310 memcpy(cmd_ctx->llm->payload, outputs,
3311 nb_output * sizeof(struct lttng_snapshot_output));
3312 free(outputs);
3313 }
3314
3315 ret = LTTNG_OK;
3316 break;
3317 }
3318 case LTTNG_SNAPSHOT_RECORD:
3319 {
3320 ret = cmd_snapshot_record(cmd_ctx->session,
3321 &cmd_ctx->lsm->u.snapshot_record.output,
3322 cmd_ctx->lsm->u.snapshot_record.wait);
3323 break;
3324 }
3325 case LTTNG_CREATE_SESSION_SNAPSHOT:
3326 {
3327 size_t nb_uri, len;
3328 struct lttng_uri *uris = NULL;
3329
3330 nb_uri = cmd_ctx->lsm->u.uri.size;
3331 len = nb_uri * sizeof(struct lttng_uri);
3332
3333 if (nb_uri > 0) {
3334 uris = zmalloc(len);
3335 if (uris == NULL) {
3336 ret = LTTNG_ERR_FATAL;
3337 goto error;
3338 }
3339
3340 /* Receive variable len data */
3341 DBG("Waiting for %zu URIs from client ...", nb_uri);
3342 ret = lttcomm_recv_unix_sock(sock, uris, len);
3343 if (ret <= 0) {
3344 DBG("No URIs received from client... continuing");
3345 *sock_error = 1;
3346 ret = LTTNG_ERR_SESSION_FAIL;
3347 free(uris);
3348 goto error;
3349 }
3350
3351 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3352 DBG("Creating session with ONE network URI is a bad call");
3353 ret = LTTNG_ERR_SESSION_FAIL;
3354 free(uris);
3355 goto error;
3356 }
3357 }
3358
3359 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3360 nb_uri, &cmd_ctx->creds);
3361 free(uris);
3362 break;
3363 }
3364 default:
3365 ret = LTTNG_ERR_UND;
3366 break;
3367 }
3368
3369 error:
3370 if (cmd_ctx->llm == NULL) {
3371 DBG("Missing llm structure. Allocating one.");
3372 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3373 goto setup_error;
3374 }
3375 }
3376 /* Set return code */
3377 cmd_ctx->llm->ret_code = ret;
3378 setup_error:
3379 if (cmd_ctx->session) {
3380 session_unlock(cmd_ctx->session);
3381 }
3382 if (need_tracing_session) {
3383 session_unlock_list();
3384 }
3385 init_setup_error:
3386 return ret;
3387 }
3388
3389 /*
3390 * Thread managing health check socket.
3391 */
3392 static void *thread_manage_health(void *data)
3393 {
3394 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3395 uint32_t revents, nb_fd;
3396 struct lttng_poll_event events;
3397 struct lttcomm_health_msg msg;
3398 struct lttcomm_health_data reply;
3399
3400 DBG("[thread] Manage health check started");
3401
3402 rcu_register_thread();
3403
3404 /* We might hit an error path before this is created. */
3405 lttng_poll_init(&events);
3406
3407 /* Create unix socket */
3408 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3409 if (sock < 0) {
3410 ERR("Unable to create health check Unix socket");
3411 ret = -1;
3412 goto error;
3413 }
3414
3415 /*
3416 * Set the CLOEXEC flag. Return code is useless because either way, the
3417 * show must go on.
3418 */
3419 (void) utils_set_fd_cloexec(sock);
3420
3421 ret = lttcomm_listen_unix_sock(sock);
3422 if (ret < 0) {
3423 goto error;
3424 }
3425
3426 /*
3427 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3428 * more will be added to this poll set.
3429 */
3430 ret = sessiond_set_thread_pollset(&events, 2);
3431 if (ret < 0) {
3432 goto error;
3433 }
3434
3435 /* Add the application registration socket */
3436 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3437 if (ret < 0) {
3438 goto error;
3439 }
3440
3441 while (1) {
3442 DBG("Health check ready");
3443
3444 /* Inifinite blocking call, waiting for transmission */
3445 restart:
3446 ret = lttng_poll_wait(&events, -1);
3447 if (ret < 0) {
3448 /*
3449 * Restart interrupted system call.
3450 */
3451 if (errno == EINTR) {
3452 goto restart;
3453 }
3454 goto error;
3455 }
3456
3457 nb_fd = ret;
3458
3459 for (i = 0; i < nb_fd; i++) {
3460 /* Fetch once the poll data */
3461 revents = LTTNG_POLL_GETEV(&events, i);
3462 pollfd = LTTNG_POLL_GETFD(&events, i);
3463
3464 /* Thread quit pipe has been closed. Killing thread. */
3465 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3466 if (ret) {
3467 err = 0;
3468 goto exit;
3469 }
3470
3471 /* Event on the registration socket */
3472 if (pollfd == sock) {
3473 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3474 ERR("Health socket poll error");
3475 goto error;
3476 }
3477 }
3478 }
3479
3480 new_sock = lttcomm_accept_unix_sock(sock);
3481 if (new_sock < 0) {
3482 goto error;
3483 }
3484
3485 /*
3486 * Set the CLOEXEC flag. Return code is useless because either way, the
3487 * show must go on.
3488 */
3489 (void) utils_set_fd_cloexec(new_sock);
3490
3491 DBG("Receiving data from client for health...");
3492 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3493 if (ret <= 0) {
3494 DBG("Nothing recv() from client... continuing");
3495 ret = close(new_sock);
3496 if (ret) {
3497 PERROR("close");
3498 }
3499 new_sock = -1;
3500 continue;
3501 }
3502
3503 rcu_thread_online();
3504
3505 switch (msg.component) {
3506 case LTTNG_HEALTH_CMD:
3507 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3508 break;
3509 case LTTNG_HEALTH_APP_MANAGE:
3510 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3511 break;
3512 case LTTNG_HEALTH_APP_REG:
3513 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3514 break;
3515 case LTTNG_HEALTH_KERNEL:
3516 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3517 break;
3518 case LTTNG_HEALTH_CONSUMER:
3519 reply.ret_code = check_consumer_health();
3520 break;
3521 case LTTNG_HEALTH_HT_CLEANUP:
3522 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3523 break;
3524 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3525 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3526 break;
3527 case LTTNG_HEALTH_APP_REG_DISPATCH:
3528 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3529 break;
3530 case LTTNG_HEALTH_ALL:
3531 reply.ret_code =
3532 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3533 health_check_state(HEALTH_TYPE_APP_REG) &&
3534 health_check_state(HEALTH_TYPE_CMD) &&
3535 health_check_state(HEALTH_TYPE_KERNEL) &&
3536 check_consumer_health() &&
3537 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3538 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3539 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3540 break;
3541 default:
3542 reply.ret_code = LTTNG_ERR_UND;
3543 break;
3544 }
3545
3546 /*
3547 * Flip ret value since 0 is a success and 1 indicates a bad health for
3548 * the client where in the sessiond it is the opposite. Again, this is
3549 * just to make things easier for us poor developer which enjoy a lot
3550 * lazyness.
3551 */
3552 if (reply.ret_code == 0 || reply.ret_code == 1) {
3553 reply.ret_code = !reply.ret_code;
3554 }
3555
3556 DBG2("Health check return value %d", reply.ret_code);
3557
3558 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3559 if (ret < 0) {
3560 ERR("Failed to send health data back to client");
3561 }
3562
3563 /* End of transmission */
3564 ret = close(new_sock);
3565 if (ret) {
3566 PERROR("close");
3567 }
3568 new_sock = -1;
3569 }
3570
3571 exit:
3572 error:
3573 if (err) {
3574 ERR("Health error occurred in %s", __func__);
3575 }
3576 DBG("Health check thread dying");
3577 unlink(health_unix_sock_path);
3578 if (sock >= 0) {
3579 ret = close(sock);
3580 if (ret) {
3581 PERROR("close");
3582 }
3583 }
3584
3585 lttng_poll_clean(&events);
3586
3587 rcu_unregister_thread();
3588 return NULL;
3589 }
3590
3591 /*
3592 * This thread manage all clients request using the unix client socket for
3593 * communication.
3594 */
3595 static void *thread_manage_clients(void *data)
3596 {
3597 int sock = -1, ret, i, pollfd, err = -1;
3598 int sock_error;
3599 uint32_t revents, nb_fd;
3600 struct command_ctx *cmd_ctx = NULL;
3601 struct lttng_poll_event events;
3602
3603 DBG("[thread] Manage client started");
3604
3605 rcu_register_thread();
3606
3607 health_register(HEALTH_TYPE_CMD);
3608
3609 if (testpoint(thread_manage_clients)) {
3610 goto error_testpoint;
3611 }
3612
3613 health_code_update();
3614
3615 ret = lttcomm_listen_unix_sock(client_sock);
3616 if (ret < 0) {
3617 goto error_listen;
3618 }
3619
3620 /*
3621 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3622 * more will be added to this poll set.
3623 */
3624 ret = sessiond_set_thread_pollset(&events, 2);
3625 if (ret < 0) {
3626 goto error_create_poll;
3627 }
3628
3629 /* Add the application registration socket */
3630 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3631 if (ret < 0) {
3632 goto error;
3633 }
3634
3635 /*
3636 * Notify parent pid that we are ready to accept command for client side.
3637 */
3638 if (opt_sig_parent) {
3639 kill(ppid, SIGUSR1);
3640 }
3641
3642 if (testpoint(thread_manage_clients_before_loop)) {
3643 goto error;
3644 }
3645
3646 health_code_update();
3647
3648 while (1) {
3649 DBG("Accepting client command ...");
3650
3651 /* Inifinite blocking call, waiting for transmission */
3652 restart:
3653 health_poll_entry();
3654 ret = lttng_poll_wait(&events, -1);
3655 health_poll_exit();
3656 if (ret < 0) {
3657 /*
3658 * Restart interrupted system call.
3659 */
3660 if (errno == EINTR) {
3661 goto restart;
3662 }
3663 goto error;
3664 }
3665
3666 nb_fd = ret;
3667
3668 for (i = 0; i < nb_fd; i++) {
3669 /* Fetch once the poll data */
3670 revents = LTTNG_POLL_GETEV(&events, i);
3671 pollfd = LTTNG_POLL_GETFD(&events, i);
3672
3673 health_code_update();
3674
3675 /* Thread quit pipe has been closed. Killing thread. */
3676 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3677 if (ret) {
3678 err = 0;
3679 goto exit;
3680 }
3681
3682 /* Event on the registration socket */
3683 if (pollfd == client_sock) {
3684 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3685 ERR("Client socket poll error");
3686 goto error;
3687 }
3688 }
3689 }
3690
3691 DBG("Wait for client response");
3692
3693 health_code_update();
3694
3695 sock = lttcomm_accept_unix_sock(client_sock);
3696 if (sock < 0) {
3697 goto error;
3698 }
3699
3700 /*
3701 * Set the CLOEXEC flag. Return code is useless because either way, the
3702 * show must go on.
3703 */
3704 (void) utils_set_fd_cloexec(sock);
3705
3706 /* Set socket option for credentials retrieval */
3707 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3708 if (ret < 0) {
3709 goto error;
3710 }
3711
3712 /* Allocate context command to process the client request */
3713 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3714 if (cmd_ctx == NULL) {
3715 PERROR("zmalloc cmd_ctx");
3716 goto error;
3717 }
3718
3719 /* Allocate data buffer for reception */
3720 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3721 if (cmd_ctx->lsm == NULL) {
3722 PERROR("zmalloc cmd_ctx->lsm");
3723 goto error;
3724 }
3725
3726 cmd_ctx->llm = NULL;
3727 cmd_ctx->session = NULL;
3728
3729 health_code_update();
3730
3731 /*
3732 * Data is received from the lttng client. The struct
3733 * lttcomm_session_msg (lsm) contains the command and data request of
3734 * the client.
3735 */
3736 DBG("Receiving data from client ...");
3737 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3738 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3739 if (ret <= 0) {
3740 DBG("Nothing recv() from client... continuing");
3741 ret = close(sock);
3742 if (ret) {
3743 PERROR("close");
3744 }
3745 sock = -1;
3746 clean_command_ctx(&cmd_ctx);
3747 continue;
3748 }
3749
3750 health_code_update();
3751
3752 // TODO: Validate cmd_ctx including sanity check for
3753 // security purpose.
3754
3755 rcu_thread_online();
3756 /*
3757 * This function dispatch the work to the kernel or userspace tracer
3758 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3759 * informations for the client. The command context struct contains
3760 * everything this function may needs.
3761 */
3762 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3763 rcu_thread_offline();
3764 if (ret < 0) {
3765 ret = close(sock);
3766 if (ret) {
3767 PERROR("close");
3768 }
3769 sock = -1;
3770 /*
3771 * TODO: Inform client somehow of the fatal error. At
3772 * this point, ret < 0 means that a zmalloc failed
3773 * (ENOMEM). Error detected but still accept
3774 * command, unless a socket error has been
3775 * detected.
3776 */
3777 clean_command_ctx(&cmd_ctx);
3778 continue;
3779 }
3780
3781 health_code_update();
3782
3783 DBG("Sending response (size: %d, retcode: %s)",
3784 cmd_ctx->lttng_msg_size,
3785 lttng_strerror(-cmd_ctx->llm->ret_code));
3786 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3787 if (ret < 0) {
3788 ERR("Failed to send data back to client");
3789 }
3790
3791 /* End of transmission */
3792 ret = close(sock);
3793 if (ret) {
3794 PERROR("close");
3795 }
3796 sock = -1;
3797
3798 clean_command_ctx(&cmd_ctx);
3799
3800 health_code_update();
3801 }
3802
3803 exit:
3804 error:
3805 if (sock >= 0) {
3806 ret = close(sock);
3807 if (ret) {
3808 PERROR("close");
3809 }
3810 }
3811
3812 lttng_poll_clean(&events);
3813 clean_command_ctx(&cmd_ctx);
3814
3815 error_listen:
3816 error_create_poll:
3817 error_testpoint:
3818 unlink(client_unix_sock_path);
3819 if (client_sock >= 0) {
3820 ret = close(client_sock);
3821 if (ret) {
3822 PERROR("close");
3823 }
3824 }
3825
3826 if (err) {
3827 health_error();
3828 ERR("Health error occurred in %s", __func__);
3829 }
3830
3831 health_unregister();
3832
3833 DBG("Client thread dying");
3834
3835 rcu_unregister_thread();
3836 return NULL;
3837 }
3838
3839
3840 /*
3841 * usage function on stderr
3842 */
3843 static void usage(void)
3844 {
3845 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3846 fprintf(stderr, " -h, --help Display this usage.\n");
3847 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3848 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3849 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3850 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3851 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3852 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3853 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3854 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3855 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3856 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3857 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3858 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3859 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3860 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3861 fprintf(stderr, " -V, --version Show version number.\n");
3862 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3863 fprintf(stderr, " -q, --quiet No output at all.\n");
3864 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3865 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3866 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3867 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3868 }
3869
3870 /*
3871 * daemon argument parsing
3872 */
3873 static int parse_args(int argc, char **argv)
3874 {
3875 int c;
3876
3877 static struct option long_options[] = {
3878 { "client-sock", 1, 0, 'c' },
3879 { "apps-sock", 1, 0, 'a' },
3880 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3881 { "kconsumerd-err-sock", 1, 0, 'E' },
3882 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3883 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3884 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3885 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3886 { "consumerd32-path", 1, 0, 'u' },
3887 { "consumerd32-libdir", 1, 0, 'U' },
3888 { "consumerd64-path", 1, 0, 't' },
3889 { "consumerd64-libdir", 1, 0, 'T' },
3890 { "daemonize", 0, 0, 'd' },
3891 { "sig-parent", 0, 0, 'S' },
3892 { "help", 0, 0, 'h' },
3893 { "group", 1, 0, 'g' },
3894 { "version", 0, 0, 'V' },
3895 { "quiet", 0, 0, 'q' },
3896 { "verbose", 0, 0, 'v' },
3897 { "verbose-consumer", 0, 0, 'Z' },
3898 { "no-kernel", 0, 0, 'N' },
3899 { "pidfile", 1, 0, 'p' },
3900 { NULL, 0, 0, 0 }
3901 };
3902
3903 while (1) {
3904 int option_index = 0;
3905 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3906 long_options, &option_index);
3907 if (c == -1) {
3908 break;
3909 }
3910
3911 switch (c) {
3912 case 0:
3913 fprintf(stderr, "option %s", long_options[option_index].name);
3914 if (optarg) {
3915 fprintf(stderr, " with arg %s\n", optarg);
3916 }
3917 break;
3918 case 'c':
3919 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3920 break;
3921 case 'a':
3922 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3923 break;
3924 case 'd':
3925 opt_daemon = 1;
3926 break;
3927 case 'g':
3928 opt_tracing_group = optarg;
3929 break;
3930 case 'h':
3931 usage();
3932 exit(EXIT_FAILURE);
3933 case 'V':
3934 fprintf(stdout, "%s\n", VERSION);
3935 exit(EXIT_SUCCESS);
3936 case 'S':
3937 opt_sig_parent = 1;
3938 break;
3939 case 'E':
3940 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3941 break;
3942 case 'C':
3943 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3944 break;
3945 case 'F':
3946 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3947 break;
3948 case 'D':
3949 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3950 break;
3951 case 'H':
3952 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3953 break;
3954 case 'G':
3955 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3956 break;
3957 case 'N':
3958 opt_no_kernel = 1;
3959 break;
3960 case 'q':
3961 lttng_opt_quiet = 1;
3962 break;
3963 case 'v':
3964 /* Verbose level can increase using multiple -v */
3965 lttng_opt_verbose += 1;
3966 break;
3967 case 'Z':
3968 opt_verbose_consumer += 1;
3969 break;
3970 case 'u':
3971 consumerd32_bin= optarg;
3972 break;
3973 case 'U':
3974 consumerd32_libdir = optarg;
3975 break;
3976 case 't':
3977 consumerd64_bin = optarg;
3978 break;
3979 case 'T':
3980 consumerd64_libdir = optarg;
3981 break;
3982 case 'p':
3983 opt_pidfile = optarg;
3984 break;
3985 default:
3986 /* Unknown option or other error.
3987 * Error is printed by getopt, just return */
3988 return -1;
3989 }
3990 }
3991
3992 return 0;
3993 }
3994
3995 /*
3996 * Creates the two needed socket by the daemon.
3997 * apps_sock - The communication socket for all UST apps.
3998 * client_sock - The communication of the cli tool (lttng).
3999 */
4000 static int init_daemon_socket(void)
4001 {
4002 int ret = 0;
4003 mode_t old_umask;
4004
4005 old_umask = umask(0);
4006
4007 /* Create client tool unix socket */
4008 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4009 if (client_sock < 0) {
4010 ERR("Create unix sock failed: %s", client_unix_sock_path);
4011 ret = -1;
4012 goto end;
4013 }
4014
4015 /* Set the cloexec flag */
4016 ret = utils_set_fd_cloexec(client_sock);
4017 if (ret < 0) {
4018 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4019 "Continuing but note that the consumer daemon will have a "
4020 "reference to this socket on exec()", client_sock);
4021 }
4022
4023 /* File permission MUST be 660 */
4024 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4025 if (ret < 0) {
4026 ERR("Set file permissions failed: %s", client_unix_sock_path);
4027 PERROR("chmod");
4028 goto end;
4029 }
4030
4031 /* Create the application unix socket */
4032 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4033 if (apps_sock < 0) {
4034 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4035 ret = -1;
4036 goto end;
4037 }
4038
4039 /* Set the cloexec flag */
4040 ret = utils_set_fd_cloexec(apps_sock);
4041 if (ret < 0) {
4042 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4043 "Continuing but note that the consumer daemon will have a "
4044 "reference to this socket on exec()", apps_sock);
4045 }
4046
4047 /* File permission MUST be 666 */
4048 ret = chmod(apps_unix_sock_path,
4049 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4050 if (ret < 0) {
4051 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4052 PERROR("chmod");
4053 goto end;
4054 }
4055
4056 DBG3("Session daemon client socket %d and application socket %d created",
4057 client_sock, apps_sock);
4058
4059 end:
4060 umask(old_umask);
4061 return ret;
4062 }
4063
4064 /*
4065 * Check if the global socket is available, and if a daemon is answering at the
4066 * other side. If yes, error is returned.
4067 */
4068 static int check_existing_daemon(void)
4069 {
4070 /* Is there anybody out there ? */
4071 if (lttng_session_daemon_alive()) {
4072 return -EEXIST;
4073 }
4074
4075 return 0;
4076 }
4077
4078 /*
4079 * Set the tracing group gid onto the client socket.
4080 *
4081 * Race window between mkdir and chown is OK because we are going from more
4082 * permissive (root.root) to less permissive (root.tracing).
4083 */
4084 static int set_permissions(char *rundir)
4085 {
4086 int ret;
4087 gid_t gid;
4088
4089 ret = allowed_group();
4090 if (ret < 0) {
4091 WARN("No tracing group detected");
4092 /* Setting gid to 0 if no tracing group is found */
4093 gid = 0;
4094 } else {
4095 gid = ret;
4096 }
4097
4098 /* Set lttng run dir */
4099 ret = chown(rundir, 0, gid);
4100 if (ret < 0) {
4101 ERR("Unable to set group on %s", rundir);
4102 PERROR("chown");
4103 }
4104
4105 /* Ensure all applications and tracing group can search the run dir */
4106 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
4107 if (ret < 0) {
4108 ERR("Unable to set permissions on %s", rundir);
4109 PERROR("chmod");
4110 }
4111
4112 /* lttng client socket path */
4113 ret = chown(client_unix_sock_path, 0, gid);
4114 if (ret < 0) {
4115 ERR("Unable to set group on %s", client_unix_sock_path);
4116 PERROR("chown");
4117 }
4118
4119 /* kconsumer error socket path */
4120 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4121 if (ret < 0) {
4122 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4123 PERROR("chown");
4124 }
4125
4126 /* 64-bit ustconsumer error socket path */
4127 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4128 if (ret < 0) {
4129 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4130 PERROR("chown");
4131 }
4132
4133 /* 32-bit ustconsumer compat32 error socket path */
4134 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4135 if (ret < 0) {
4136 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4137 PERROR("chown");
4138 }
4139
4140 DBG("All permissions are set");
4141
4142 return ret;
4143 }
4144
4145 /*
4146 * Create the lttng run directory needed for all global sockets and pipe.
4147 */
4148 static int create_lttng_rundir(const char *rundir)
4149 {
4150 int ret;
4151
4152 DBG3("Creating LTTng run directory: %s", rundir);
4153
4154 ret = mkdir(rundir, S_IRWXU);
4155 if (ret < 0) {
4156 if (errno != EEXIST) {
4157 ERR("Unable to create %s", rundir);
4158 goto error;
4159 } else {
4160 ret = 0;
4161 }
4162 }
4163
4164 error:
4165 return ret;
4166 }
4167
4168 /*
4169 * Setup sockets and directory needed by the kconsumerd communication with the
4170 * session daemon.
4171 */
4172 static int set_consumer_sockets(struct consumer_data *consumer_data,
4173 const char *rundir)
4174 {
4175 int ret;
4176 char path[PATH_MAX];
4177
4178 switch (consumer_data->type) {
4179 case LTTNG_CONSUMER_KERNEL:
4180 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4181 break;
4182 case LTTNG_CONSUMER64_UST:
4183 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4184 break;
4185 case LTTNG_CONSUMER32_UST:
4186 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4187 break;
4188 default:
4189 ERR("Consumer type unknown");
4190 ret = -EINVAL;
4191 goto error;
4192 }
4193
4194 DBG2("Creating consumer directory: %s", path);
4195
4196 ret = mkdir(path, S_IRWXU);
4197 if (ret < 0) {
4198 if (errno != EEXIST) {
4199 PERROR("mkdir");
4200 ERR("Failed to create %s", path);
4201 goto error;
4202 }
4203 ret = -1;
4204 }
4205
4206 /* Create the kconsumerd error unix socket */
4207 consumer_data->err_sock =
4208 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4209 if (consumer_data->err_sock < 0) {
4210 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4211 ret = -1;
4212 goto error;
4213 }
4214
4215 /*
4216 * Set the CLOEXEC flag. Return code is useless because either way, the
4217 * show must go on.
4218 */
4219 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4220 if (ret < 0) {
4221 PERROR("utils_set_fd_cloexec");
4222 /* continue anyway */
4223 }
4224
4225 /* File permission MUST be 660 */
4226 ret = chmod(consumer_data->err_unix_sock_path,
4227 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4228 if (ret < 0) {
4229 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4230 PERROR("chmod");
4231 goto error;
4232 }
4233
4234 error:
4235 return ret;
4236 }
4237
4238 /*
4239 * Signal handler for the daemon
4240 *
4241 * Simply stop all worker threads, leaving main() return gracefully after
4242 * joining all threads and calling cleanup().
4243 */
4244 static void sighandler(int sig)
4245 {
4246 switch (sig) {
4247 case SIGPIPE:
4248 DBG("SIGPIPE caught");
4249 return;
4250 case SIGINT:
4251 DBG("SIGINT caught");
4252 stop_threads();
4253 break;
4254 case SIGTERM:
4255 DBG("SIGTERM caught");
4256 stop_threads();
4257 break;
4258 default:
4259 break;
4260 }
4261 }
4262
4263 /*
4264 * Setup signal handler for :
4265 * SIGINT, SIGTERM, SIGPIPE
4266 */
4267 static int set_signal_handler(void)
4268 {
4269 int ret = 0;
4270 struct sigaction sa;
4271 sigset_t sigset;
4272
4273 if ((ret = sigemptyset(&sigset)) < 0) {
4274 PERROR("sigemptyset");
4275 return ret;
4276 }
4277
4278 sa.sa_handler = sighandler;
4279 sa.sa_mask = sigset;
4280 sa.sa_flags = 0;
4281 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4282 PERROR("sigaction");
4283 return ret;
4284 }
4285
4286 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4287 PERROR("sigaction");
4288 return ret;
4289 }
4290
4291 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4292 PERROR("sigaction");
4293 return ret;
4294 }
4295
4296 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4297
4298 return ret;
4299 }
4300
4301 /*
4302 * Set open files limit to unlimited. This daemon can open a large number of
4303 * file descriptors in order to consumer multiple kernel traces.
4304 */
4305 static void set_ulimit(void)
4306 {
4307 int ret;
4308 struct rlimit lim;
4309
4310 /* The kernel does not allowed an infinite limit for open files */
4311 lim.rlim_cur = 65535;
4312 lim.rlim_max = 65535;
4313
4314 ret = setrlimit(RLIMIT_NOFILE, &lim);
4315 if (ret < 0) {
4316 PERROR("failed to set open files limit");
4317 }
4318 }
4319
4320 /*
4321 * Write pidfile using the rundir and opt_pidfile.
4322 */
4323 static void write_pidfile(void)
4324 {
4325 int ret;
4326 char pidfile_path[PATH_MAX];
4327
4328 assert(rundir);
4329
4330 if (opt_pidfile) {
4331 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4332 } else {
4333 /* Build pidfile path from rundir and opt_pidfile. */
4334 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4335 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4336 if (ret < 0) {
4337 PERROR("snprintf pidfile path");
4338 goto error;
4339 }
4340 }
4341
4342 /*
4343 * Create pid file in rundir. Return value is of no importance. The
4344 * execution will continue even though we are not able to write the file.
4345 */
4346 (void) utils_create_pid_file(getpid(), pidfile_path);
4347
4348 error:
4349 return;
4350 }
4351
4352 /*
4353 * main
4354 */
4355 int main(int argc, char **argv)
4356 {
4357 int ret = 0;
4358 void *status;
4359 const char *home_path, *env_app_timeout;
4360
4361 init_kernel_workarounds();
4362
4363 rcu_register_thread();
4364
4365 setup_consumerd_path();
4366
4367 page_size = sysconf(_SC_PAGESIZE);
4368 if (page_size < 0) {
4369 PERROR("sysconf _SC_PAGESIZE");
4370 page_size = LONG_MAX;
4371 WARN("Fallback page size to %ld", page_size);
4372 }
4373
4374 /* Parse arguments */
4375 progname = argv[0];
4376 if ((ret = parse_args(argc, argv)) < 0) {
4377 goto error;
4378 }
4379
4380 /* Daemonize */
4381 if (opt_daemon) {
4382 int i;
4383
4384 /*
4385 * fork
4386 * child: setsid, close FD 0, 1, 2, chdir /
4387 * parent: exit (if fork is successful)
4388 */
4389 ret = daemon(0, 0);
4390 if (ret < 0) {
4391 PERROR("daemon");
4392 goto error;
4393 }
4394 /*
4395 * We are in the child. Make sure all other file
4396 * descriptors are closed, in case we are called with
4397 * more opened file descriptors than the standard ones.
4398 */
4399 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4400 (void) close(i);
4401 }
4402 }
4403
4404 /* Create thread quit pipe */
4405 if ((ret = init_thread_quit_pipe()) < 0) {
4406 goto error;
4407 }
4408
4409 /* Check if daemon is UID = 0 */
4410 is_root = !getuid();
4411
4412 if (is_root) {
4413 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4414
4415 /* Create global run dir with root access */
4416 ret = create_lttng_rundir(rundir);
4417 if (ret < 0) {
4418 goto error;
4419 }
4420
4421 if (strlen(apps_unix_sock_path) == 0) {
4422 snprintf(apps_unix_sock_path, PATH_MAX,
4423 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4424 }
4425
4426 if (strlen(client_unix_sock_path) == 0) {
4427 snprintf(client_unix_sock_path, PATH_MAX,
4428 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4429 }
4430
4431 /* Set global SHM for ust */
4432 if (strlen(wait_shm_path) == 0) {
4433 snprintf(wait_shm_path, PATH_MAX,
4434 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4435 }
4436
4437 if (strlen(health_unix_sock_path) == 0) {
4438 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4439 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4440 }
4441
4442 /* Setup kernel consumerd path */
4443 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4444 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4445 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4446 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4447
4448 DBG2("Kernel consumer err path: %s",
4449 kconsumer_data.err_unix_sock_path);
4450 DBG2("Kernel consumer cmd path: %s",
4451 kconsumer_data.cmd_unix_sock_path);
4452 } else {
4453 home_path = utils_get_home_dir();
4454 if (home_path == NULL) {
4455 /* TODO: Add --socket PATH option */
4456 ERR("Can't get HOME directory for sockets creation.");
4457 ret = -EPERM;
4458 goto error;
4459 }
4460
4461 /*
4462 * Create rundir from home path. This will create something like
4463 * $HOME/.lttng
4464 */
4465 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4466 if (ret < 0) {
4467 ret = -ENOMEM;
4468 goto error;
4469 }
4470
4471 ret = create_lttng_rundir(rundir);
4472 if (ret < 0) {
4473 goto error;
4474 }
4475
4476 if (strlen(apps_unix_sock_path) == 0) {
4477 snprintf(apps_unix_sock_path, PATH_MAX,
4478 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4479 }
4480
4481 /* Set the cli tool unix socket path */
4482 if (strlen(client_unix_sock_path) == 0) {
4483 snprintf(client_unix_sock_path, PATH_MAX,
4484 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4485 }
4486
4487 /* Set global SHM for ust */
4488 if (strlen(wait_shm_path) == 0) {
4489 snprintf(wait_shm_path, PATH_MAX,
4490 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4491 }
4492
4493 /* Set health check Unix path */
4494 if (strlen(health_unix_sock_path) == 0) {
4495 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4496 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4497 }
4498 }
4499
4500 /* Set consumer initial state */
4501 kernel_consumerd_state = CONSUMER_STOPPED;
4502 ust_consumerd_state = CONSUMER_STOPPED;
4503
4504 DBG("Client socket path %s", client_unix_sock_path);
4505 DBG("Application socket path %s", apps_unix_sock_path);
4506 DBG("Application wait path %s", wait_shm_path);
4507 DBG("LTTng run directory path: %s", rundir);
4508
4509 /* 32 bits consumerd path setup */
4510 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4511 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4512 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4513 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4514
4515 DBG2("UST consumer 32 bits err path: %s",
4516 ustconsumer32_data.err_unix_sock_path);
4517 DBG2("UST consumer 32 bits cmd path: %s",
4518 ustconsumer32_data.cmd_unix_sock_path);
4519
4520 /* 64 bits consumerd path setup */
4521 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4522 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4523 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4524 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4525
4526 DBG2("UST consumer 64 bits err path: %s",
4527 ustconsumer64_data.err_unix_sock_path);
4528 DBG2("UST consumer 64 bits cmd path: %s",
4529 ustconsumer64_data.cmd_unix_sock_path);
4530
4531 /*
4532 * See if daemon already exist.
4533 */
4534 if ((ret = check_existing_daemon()) < 0) {
4535 ERR("Already running daemon.\n");
4536 /*
4537 * We do not goto exit because we must not cleanup()
4538 * because a daemon is already running.
4539 */
4540 goto error;
4541 }
4542
4543 /*
4544 * Init UST app hash table. Alloc hash table before this point since
4545 * cleanup() can get called after that point.
4546 */
4547 ust_app_ht_alloc();
4548
4549 /* After this point, we can safely call cleanup() with "goto exit" */
4550
4551 /*
4552 * These actions must be executed as root. We do that *after* setting up
4553 * the sockets path because we MUST make the check for another daemon using
4554 * those paths *before* trying to set the kernel consumer sockets and init
4555 * kernel tracer.
4556 */
4557 if (is_root) {
4558 ret = set_consumer_sockets(&kconsumer_data, rundir);
4559 if (ret < 0) {
4560 goto exit;
4561 }
4562
4563 /* Setup kernel tracer */
4564 if (!opt_no_kernel) {
4565 init_kernel_tracer();
4566 }
4567
4568 /* Set ulimit for open files */
4569 set_ulimit();
4570 }
4571 /* init lttng_fd tracking must be done after set_ulimit. */
4572 lttng_fd_init();
4573
4574 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4575 if (ret < 0) {
4576 goto exit;
4577 }
4578
4579 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4580 if (ret < 0) {
4581 goto exit;
4582 }
4583
4584 if ((ret = set_signal_handler()) < 0) {
4585 goto exit;
4586 }
4587
4588 /* Setup the needed unix socket */
4589 if ((ret = init_daemon_socket()) < 0) {
4590 goto exit;
4591 }
4592
4593 /* Set credentials to socket */
4594 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4595 goto exit;
4596 }
4597
4598 /* Get parent pid if -S, --sig-parent is specified. */
4599 if (opt_sig_parent) {
4600 ppid = getppid();
4601 }
4602
4603 /* Setup the kernel pipe for waking up the kernel thread */
4604 if (is_root && !opt_no_kernel) {
4605 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4606 goto exit;
4607 }
4608 }
4609
4610 /* Setup the thread ht_cleanup communication pipe. */
4611 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4612 goto exit;
4613 }
4614
4615 /* Setup the thread apps communication pipe. */
4616 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4617 goto exit;
4618 }
4619
4620 /* Setup the thread apps notify communication pipe. */
4621 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4622 goto exit;
4623 }
4624
4625 /* Initialize global buffer per UID and PID registry. */
4626 buffer_reg_init_uid_registry();
4627 buffer_reg_init_pid_registry();
4628
4629 /* Init UST command queue. */
4630 cds_wfq_init(&ust_cmd_queue.queue);
4631
4632 /*
4633 * Get session list pointer. This pointer MUST NOT be free(). This list is
4634 * statically declared in session.c
4635 */
4636 session_list_ptr = session_get_list();
4637
4638 /* Set up max poll set size */
4639 lttng_poll_set_max_size();
4640
4641 cmd_init();
4642
4643 /* Check for the application socket timeout env variable. */
4644 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4645 if (env_app_timeout) {
4646 app_socket_timeout = atoi(env_app_timeout);
4647 } else {
4648 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4649 }
4650
4651 write_pidfile();
4652
4653 /* Initialize communication library */
4654 lttcomm_init();
4655
4656 /* Create thread to manage the client socket */
4657 ret = pthread_create(&ht_cleanup_thread, NULL,
4658 thread_ht_cleanup, (void *) NULL);
4659 if (ret != 0) {
4660 PERROR("pthread_create ht_cleanup");
4661 goto exit_ht_cleanup;
4662 }
4663
4664 /* Create thread to manage the client socket */
4665 ret = pthread_create(&health_thread, NULL,
4666 thread_manage_health, (void *) NULL);
4667 if (ret != 0) {
4668 PERROR("pthread_create health");
4669 goto exit_health;
4670 }
4671
4672 /* Create thread to manage the client socket */
4673 ret = pthread_create(&client_thread, NULL,
4674 thread_manage_clients, (void *) NULL);
4675 if (ret != 0) {
4676 PERROR("pthread_create clients");
4677 goto exit_client;
4678 }
4679
4680 /* Create thread to dispatch registration */
4681 ret = pthread_create(&dispatch_thread, NULL,
4682 thread_dispatch_ust_registration, (void *) NULL);
4683 if (ret != 0) {
4684 PERROR("pthread_create dispatch");
4685 goto exit_dispatch;
4686 }
4687
4688 /* Create thread to manage application registration. */
4689 ret = pthread_create(&reg_apps_thread, NULL,
4690 thread_registration_apps, (void *) NULL);
4691 if (ret != 0) {
4692 PERROR("pthread_create registration");
4693 goto exit_reg_apps;
4694 }
4695
4696 /* Create thread to manage application socket */
4697 ret = pthread_create(&apps_thread, NULL,
4698 thread_manage_apps, (void *) NULL);
4699 if (ret != 0) {
4700 PERROR("pthread_create apps");
4701 goto exit_apps;
4702 }
4703
4704 /* Create thread to manage application notify socket */
4705 ret = pthread_create(&apps_notify_thread, NULL,
4706 ust_thread_manage_notify, (void *) NULL);
4707 if (ret != 0) {
4708 PERROR("pthread_create apps");
4709 goto exit_apps;
4710 }
4711
4712 /* Don't start this thread if kernel tracing is not requested nor root */
4713 if (is_root && !opt_no_kernel) {
4714 /* Create kernel thread to manage kernel event */
4715 ret = pthread_create(&kernel_thread, NULL,
4716 thread_manage_kernel, (void *) NULL);
4717 if (ret != 0) {
4718 PERROR("pthread_create kernel");
4719 goto exit_kernel;
4720 }
4721
4722 ret = pthread_join(kernel_thread, &status);
4723 if (ret != 0) {
4724 PERROR("pthread_join");
4725 goto error; /* join error, exit without cleanup */
4726 }
4727 }
4728
4729 exit_kernel:
4730 ret = pthread_join(apps_thread, &status);
4731 if (ret != 0) {
4732 PERROR("pthread_join");
4733 goto error; /* join error, exit without cleanup */
4734 }
4735
4736 exit_apps:
4737 ret = pthread_join(reg_apps_thread, &status);
4738 if (ret != 0) {
4739 PERROR("pthread_join");
4740 goto error; /* join error, exit without cleanup */
4741 }
4742
4743 exit_reg_apps:
4744 ret = pthread_join(dispatch_thread, &status);
4745 if (ret != 0) {
4746 PERROR("pthread_join");
4747 goto error; /* join error, exit without cleanup */
4748 }
4749
4750 exit_dispatch:
4751 ret = pthread_join(client_thread, &status);
4752 if (ret != 0) {
4753 PERROR("pthread_join");
4754 goto error; /* join error, exit without cleanup */
4755 }
4756
4757 ret = join_consumer_thread(&kconsumer_data);
4758 if (ret != 0) {
4759 PERROR("join_consumer");
4760 goto error; /* join error, exit without cleanup */
4761 }
4762
4763 ret = join_consumer_thread(&ustconsumer32_data);
4764 if (ret != 0) {
4765 PERROR("join_consumer ust32");
4766 goto error; /* join error, exit without cleanup */
4767 }
4768
4769 ret = join_consumer_thread(&ustconsumer64_data);
4770 if (ret != 0) {
4771 PERROR("join_consumer ust64");
4772 goto error; /* join error, exit without cleanup */
4773 }
4774
4775 exit_client:
4776 ret = pthread_join(health_thread, &status);
4777 if (ret != 0) {
4778 PERROR("pthread_join health thread");
4779 goto error; /* join error, exit without cleanup */
4780 }
4781
4782 exit_health:
4783 ret = pthread_join(ht_cleanup_thread, &status);
4784 if (ret != 0) {
4785 PERROR("pthread_join ht cleanup thread");
4786 goto error; /* join error, exit without cleanup */
4787 }
4788 exit_ht_cleanup:
4789 exit:
4790 /*
4791 * cleanup() is called when no other thread is running.
4792 */
4793 rcu_thread_online();
4794 cleanup();
4795 rcu_thread_offline();
4796 rcu_unregister_thread();
4797 if (!ret) {
4798 exit(EXIT_SUCCESS);
4799 }
4800 error:
4801 exit(EXIT_FAILURE);
4802 }
This page took 0.2219 seconds and 5 git commands to generate.