1d047fd85240582854f5a6c50744b23ce7ac4cb1
[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 .metadata_sock.fd = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .metadata_sock.fd = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer32_data = {
112 .type = LTTNG_CONSUMER32_UST,
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
115 .err_sock = -1,
116 .cmd_sock = -1,
117 .metadata_sock.fd = -1,
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
122 };
123
124 /* Shared between threads */
125 static int dispatch_thread_exit;
126
127 /* Global application Unix socket path */
128 static char apps_unix_sock_path[PATH_MAX];
129 /* Global client Unix socket path */
130 static char client_unix_sock_path[PATH_MAX];
131 /* global wait shm path for UST */
132 static char wait_shm_path[PATH_MAX];
133 /* Global health check unix path */
134 static char health_unix_sock_path[PATH_MAX];
135
136 /* Sockets and FDs */
137 static int client_sock = -1;
138 static int apps_sock = -1;
139 int kernel_tracer_fd = -1;
140 static int kernel_poll_pipe[2] = { -1, -1 };
141
142 /*
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
145 */
146 static int thread_quit_pipe[2] = { -1, -1 };
147
148 /*
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
151 */
152 static int apps_cmd_pipe[2] = { -1, -1 };
153
154 int apps_cmd_notify_pipe[2] = { -1, -1 };
155
156 /* Pthread, Mutexes and Semaphores */
157 static pthread_t apps_thread;
158 static pthread_t apps_notify_thread;
159 static pthread_t reg_apps_thread;
160 static pthread_t client_thread;
161 static pthread_t kernel_thread;
162 static pthread_t dispatch_thread;
163 static pthread_t health_thread;
164 static pthread_t ht_cleanup_thread;
165
166 /*
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
169 *
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
172 */
173 static struct ust_cmd_queue ust_cmd_queue;
174
175 /*
176 * Pointer initialized before thread creation.
177 *
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
181 *
182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
183 * session_lock_list and session_unlock_list for lock acquisition.
184 */
185 static struct ltt_session_list *session_list_ptr;
186
187 int ust_consumerd64_fd = -1;
188 int ust_consumerd32_fd = -1;
189
190 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
191 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
192 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
193 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
194
195 static const char *module_proc_lttng = "/proc/lttng";
196
197 /*
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
200 */
201 enum consumerd_state {
202 CONSUMER_STARTED = 1,
203 CONSUMER_STOPPED = 2,
204 CONSUMER_ERROR = 3,
205 };
206
207 /*
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
212 *
213 * The following example shows a possible race condition of this scheme:
214 *
215 * consumer thread error happens
216 * client cmd arrives
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
220 * ...
221 *
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
227 */
228 static enum consumerd_state ust_consumerd_state;
229 static enum consumerd_state kernel_consumerd_state;
230
231 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 /* Set in main() with the current page size. */
237 long page_size;
238
239 static
240 void setup_consumerd_path(void)
241 {
242 const char *bin, *libdir;
243
244 /*
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
248 */
249 #if (CAA_BITS_PER_LONG == 32)
250 if (!consumerd32_bin[0]) {
251 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
252 }
253 if (!consumerd32_libdir[0]) {
254 consumerd32_libdir = INSTALL_LIB_PATH;
255 }
256 #elif (CAA_BITS_PER_LONG == 64)
257 if (!consumerd64_bin[0]) {
258 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
259 }
260 if (!consumerd64_libdir[0]) {
261 consumerd64_libdir = INSTALL_LIB_PATH;
262 }
263 #else
264 #error "Unknown bitness"
265 #endif
266
267 /*
268 * runtime env. var. overrides the build default.
269 */
270 bin = getenv("LTTNG_CONSUMERD32_BIN");
271 if (bin) {
272 consumerd32_bin = bin;
273 }
274 bin = getenv("LTTNG_CONSUMERD64_BIN");
275 if (bin) {
276 consumerd64_bin = bin;
277 }
278 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
279 if (libdir) {
280 consumerd32_libdir = libdir;
281 }
282 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
283 if (libdir) {
284 consumerd64_libdir = libdir;
285 }
286 }
287
288 /*
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
290 */
291 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
292 {
293 int ret;
294
295 assert(events);
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310 error:
311 return ret;
312 }
313
314 /*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
319 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326 }
327
328 /*
329 * Return group ID of the tracing group or -1 if not found.
330 */
331 static gid_t allowed_group(void)
332 {
333 struct group *grp;
334
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345 }
346
347 /*
348 * Init thread quit pipe.
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352 static int init_thread_quit_pipe(void)
353 {
354 int ret, i;
355
356 ret = pipe(thread_quit_pipe);
357 if (ret < 0) {
358 PERROR("thread quit pipe");
359 goto error;
360 }
361
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
370 error:
371 return ret;
372 }
373
374 /*
375 * Stop all threads by closing the thread quit pipe.
376 */
377 static void stop_threads(void)
378 {
379 int ret;
380
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret = notify_thread_pipe(thread_quit_pipe[1]);
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
390 futex_nto1_wake(&ust_cmd_queue.futex);
391 }
392
393 /*
394 * Close every consumer sockets.
395 */
396 static void close_consumer_sockets(void)
397 {
398 int ret;
399
400 if (kconsumer_data.err_sock >= 0) {
401 ret = close(kconsumer_data.err_sock);
402 if (ret < 0) {
403 PERROR("kernel consumer err_sock close");
404 }
405 }
406 if (ustconsumer32_data.err_sock >= 0) {
407 ret = close(ustconsumer32_data.err_sock);
408 if (ret < 0) {
409 PERROR("UST consumerd32 err_sock close");
410 }
411 }
412 if (ustconsumer64_data.err_sock >= 0) {
413 ret = close(ustconsumer64_data.err_sock);
414 if (ret < 0) {
415 PERROR("UST consumerd64 err_sock close");
416 }
417 }
418 if (kconsumer_data.cmd_sock >= 0) {
419 ret = close(kconsumer_data.cmd_sock);
420 if (ret < 0) {
421 PERROR("kernel consumer cmd_sock close");
422 }
423 }
424 if (ustconsumer32_data.cmd_sock >= 0) {
425 ret = close(ustconsumer32_data.cmd_sock);
426 if (ret < 0) {
427 PERROR("UST consumerd32 cmd_sock close");
428 }
429 }
430 if (ustconsumer64_data.cmd_sock >= 0) {
431 ret = close(ustconsumer64_data.cmd_sock);
432 if (ret < 0) {
433 PERROR("UST consumerd64 cmd_sock close");
434 }
435 }
436 }
437
438 /*
439 * Cleanup the daemon
440 */
441 static void cleanup(void)
442 {
443 int ret;
444 char *cmd = NULL;
445 struct ltt_session *sess, *stmp;
446
447 DBG("Cleaning up");
448
449 /* First thing first, stop all threads */
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
679 /*
680 * Have we already sent fds to the consumer? If yes, it means
681 * that tracing is started so it is safe to send our updated
682 * stream fds.
683 */
684 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
685 struct lttng_ht_iter iter;
686 struct consumer_socket *socket;
687
688 rcu_read_lock();
689 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
690 &iter.iter, socket, node.node) {
691 /* Code flow error */
692 assert(socket->fd >= 0);
693
694 pthread_mutex_lock(socket->lock);
695 ret = kernel_consumer_send_channel_stream(socket,
696 channel, ksess,
697 session->output_traces ? 1 : 0);
698 pthread_mutex_unlock(socket->lock);
699 if (ret < 0) {
700 rcu_read_unlock();
701 goto error;
702 }
703 }
704 rcu_read_unlock();
705 }
706 goto error;
707 }
708 }
709 session_unlock(session);
710 }
711 session_unlock_list();
712 return ret;
713
714 error:
715 session_unlock(session);
716 session_unlock_list();
717 return ret;
718 }
719
720 /*
721 * For each tracing session, update newly registered apps. The session list
722 * lock MUST be acquired before calling this.
723 */
724 static void update_ust_app(int app_sock)
725 {
726 struct ltt_session *sess, *stmp;
727
728 /* For all tracing session(s) */
729 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
730 session_lock(sess);
731 if (sess->ust_session) {
732 ust_app_global_update(sess->ust_session, app_sock);
733 }
734 session_unlock(sess);
735 }
736 }
737
738 /*
739 * This thread manage event coming from the kernel.
740 *
741 * Features supported in this thread:
742 * -) CPU Hotplug
743 */
744 static void *thread_manage_kernel(void *data)
745 {
746 int ret, i, pollfd, update_poll_flag = 1, err = -1;
747 uint32_t revents, nb_fd;
748 char tmp;
749 struct lttng_poll_event events;
750
751 DBG("[thread] Thread manage kernel started");
752
753 health_register(HEALTH_TYPE_KERNEL);
754
755 /*
756 * This first step of the while is to clean this structure which could free
757 * non NULL pointers so initialize it before the loop.
758 */
759 lttng_poll_init(&events);
760
761 if (testpoint(thread_manage_kernel)) {
762 goto error_testpoint;
763 }
764
765 health_code_update();
766
767 if (testpoint(thread_manage_kernel_before_loop)) {
768 goto error_testpoint;
769 }
770
771 while (1) {
772 health_code_update();
773
774 if (update_poll_flag == 1) {
775 /* Clean events object. We are about to populate it again. */
776 lttng_poll_clean(&events);
777
778 ret = sessiond_set_thread_pollset(&events, 2);
779 if (ret < 0) {
780 goto error_poll_create;
781 }
782
783 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
784 if (ret < 0) {
785 goto error;
786 }
787
788 /* This will add the available kernel channel if any. */
789 ret = update_kernel_poll(&events);
790 if (ret < 0) {
791 goto error;
792 }
793 update_poll_flag = 0;
794 }
795
796 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
797
798 /* Poll infinite value of time */
799 restart:
800 health_poll_entry();
801 ret = lttng_poll_wait(&events, -1);
802 health_poll_exit();
803 if (ret < 0) {
804 /*
805 * Restart interrupted system call.
806 */
807 if (errno == EINTR) {
808 goto restart;
809 }
810 goto error;
811 } else if (ret == 0) {
812 /* Should not happen since timeout is infinite */
813 ERR("Return value of poll is 0 with an infinite timeout.\n"
814 "This should not have happened! Continuing...");
815 continue;
816 }
817
818 nb_fd = ret;
819
820 for (i = 0; i < nb_fd; i++) {
821 /* Fetch once the poll data */
822 revents = LTTNG_POLL_GETEV(&events, i);
823 pollfd = LTTNG_POLL_GETFD(&events, i);
824
825 health_code_update();
826
827 /* Thread quit pipe has been closed. Killing thread. */
828 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
829 if (ret) {
830 err = 0;
831 goto exit;
832 }
833
834 /* Check for data on kernel pipe */
835 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
836 do {
837 ret = read(kernel_poll_pipe[0], &tmp, 1);
838 } while (ret < 0 && errno == EINTR);
839 /*
840 * Ret value is useless here, if this pipe gets any actions an
841 * update is required anyway.
842 */
843 update_poll_flag = 1;
844 continue;
845 } else {
846 /*
847 * New CPU detected by the kernel. Adding kernel stream to
848 * kernel session and updating the kernel consumer
849 */
850 if (revents & LPOLLIN) {
851 ret = update_kernel_stream(&kconsumer_data, pollfd);
852 if (ret < 0) {
853 continue;
854 }
855 break;
856 /*
857 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
858 * and unregister kernel stream at this point.
859 */
860 }
861 }
862 }
863 }
864
865 exit:
866 error:
867 lttng_poll_clean(&events);
868 error_poll_create:
869 error_testpoint:
870 utils_close_pipe(kernel_poll_pipe);
871 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
872 if (err) {
873 health_error();
874 ERR("Health error occurred in %s", __func__);
875 WARN("Kernel thread died unexpectedly. "
876 "Kernel tracing can continue but CPU hotplug is disabled.");
877 }
878 health_unregister();
879 DBG("Kernel thread dying");
880 return NULL;
881 }
882
883 /*
884 * Signal pthread condition of the consumer data that the thread.
885 */
886 static void signal_consumer_condition(struct consumer_data *data, int state)
887 {
888 pthread_mutex_lock(&data->cond_mutex);
889
890 /*
891 * The state is set before signaling. It can be any value, it's the waiter
892 * job to correctly interpret this condition variable associated to the
893 * consumer pthread_cond.
894 *
895 * A value of 0 means that the corresponding thread of the consumer data
896 * was not started. 1 indicates that the thread has started and is ready
897 * for action. A negative value means that there was an error during the
898 * thread bootstrap.
899 */
900 data->consumer_thread_is_ready = state;
901 (void) pthread_cond_signal(&data->cond);
902
903 pthread_mutex_unlock(&data->cond_mutex);
904 }
905
906 /*
907 * This thread manage the consumer error sent back to the session daemon.
908 */
909 static void *thread_manage_consumer(void *data)
910 {
911 int sock = -1, i, ret, pollfd, err = -1;
912 uint32_t revents, nb_fd;
913 enum lttcomm_return_code code;
914 struct lttng_poll_event events;
915 struct consumer_data *consumer_data = data;
916
917 DBG("[thread] Manage consumer started");
918
919 health_register(HEALTH_TYPE_CONSUMER);
920
921 health_code_update();
922
923 /*
924 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
925 * metadata_sock. Nothing more will be added to this poll set.
926 */
927 ret = sessiond_set_thread_pollset(&events, 3);
928 if (ret < 0) {
929 goto error_poll;
930 }
931
932 /*
933 * The error socket here is already in a listening state which was done
934 * just before spawning this thread to avoid a race between the consumer
935 * daemon exec trying to connect and the listen() call.
936 */
937 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
938 if (ret < 0) {
939 goto error;
940 }
941
942 health_code_update();
943
944 /* Infinite blocking call, waiting for transmission */
945 restart:
946 health_poll_entry();
947
948 if (testpoint(thread_manage_consumer)) {
949 goto error;
950 }
951
952 ret = lttng_poll_wait(&events, -1);
953 health_poll_exit();
954 if (ret < 0) {
955 /*
956 * Restart interrupted system call.
957 */
958 if (errno == EINTR) {
959 goto restart;
960 }
961 goto error;
962 }
963
964 nb_fd = ret;
965
966 for (i = 0; i < nb_fd; i++) {
967 /* Fetch once the poll data */
968 revents = LTTNG_POLL_GETEV(&events, i);
969 pollfd = LTTNG_POLL_GETFD(&events, i);
970
971 health_code_update();
972
973 /* Thread quit pipe has been closed. Killing thread. */
974 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
975 if (ret) {
976 err = 0;
977 goto exit;
978 }
979
980 /* Event on the registration socket */
981 if (pollfd == consumer_data->err_sock) {
982 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
983 ERR("consumer err socket poll error");
984 goto error;
985 }
986 }
987 }
988
989 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
990 if (sock < 0) {
991 goto error;
992 }
993
994 /*
995 * Set the CLOEXEC flag. Return code is useless because either way, the
996 * show must go on.
997 */
998 (void) utils_set_fd_cloexec(sock);
999
1000 health_code_update();
1001
1002 DBG2("Receiving code from consumer err_sock");
1003
1004 /* Getting status code from kconsumerd */
1005 ret = lttcomm_recv_unix_sock(sock, &code,
1006 sizeof(enum lttcomm_return_code));
1007 if (ret <= 0) {
1008 goto error;
1009 }
1010
1011 health_code_update();
1012
1013 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1014 /* Connect both socket, command and metadata. */
1015 consumer_data->cmd_sock =
1016 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1017 consumer_data->metadata_sock.fd =
1018 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1019 if (consumer_data->cmd_sock < 0 ||
1020 consumer_data->metadata_sock.fd < 0) {
1021 PERROR("consumer connect cmd socket");
1022 /* On error, signal condition and quit. */
1023 signal_consumer_condition(consumer_data, -1);
1024 goto error;
1025 }
1026 /* Create metadata socket lock. */
1027 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1028 if (consumer_data->metadata_sock.lock == NULL) {
1029 PERROR("zmalloc pthread mutex");
1030 ret = -1;
1031 goto error;
1032 }
1033 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1034
1035 signal_consumer_condition(consumer_data, 1);
1036 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1037 DBG("Consumer metadata socket ready (fd: %d)",
1038 consumer_data->metadata_sock.fd);
1039 } else {
1040 ERR("consumer error when waiting for SOCK_READY : %s",
1041 lttcomm_get_readable_code(-code));
1042 goto error;
1043 }
1044
1045 /* Remove the consumerd error sock since we've established a connexion */
1046 ret = lttng_poll_del(&events, consumer_data->err_sock);
1047 if (ret < 0) {
1048 goto error;
1049 }
1050
1051 /* Add new accepted error socket. */
1052 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1053 if (ret < 0) {
1054 goto error;
1055 }
1056
1057 /* Add metadata socket that is successfully connected. */
1058 ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
1059 LPOLLIN | LPOLLRDHUP);
1060 if (ret < 0) {
1061 goto error;
1062 }
1063
1064 health_code_update();
1065
1066 /* Infinite blocking call, waiting for transmission */
1067 restart_poll:
1068 while (1) {
1069 health_poll_entry();
1070 ret = lttng_poll_wait(&events, -1);
1071 health_poll_exit();
1072 if (ret < 0) {
1073 /*
1074 * Restart interrupted system call.
1075 */
1076 if (errno == EINTR) {
1077 goto restart_poll;
1078 }
1079 goto error;
1080 }
1081
1082 nb_fd = ret;
1083
1084 for (i = 0; i < nb_fd; i++) {
1085 /* Fetch once the poll data */
1086 revents = LTTNG_POLL_GETEV(&events, i);
1087 pollfd = LTTNG_POLL_GETFD(&events, i);
1088
1089 health_code_update();
1090
1091 /* Thread quit pipe has been closed. Killing thread. */
1092 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1093 if (ret) {
1094 err = 0;
1095 goto exit;
1096 }
1097
1098 if (pollfd == sock) {
1099 /* Event on the consumerd socket */
1100 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1101 ERR("consumer err socket second poll error");
1102 goto error;
1103 }
1104 health_code_update();
1105 /* Wait for any kconsumerd error */
1106 ret = lttcomm_recv_unix_sock(sock, &code,
1107 sizeof(enum lttcomm_return_code));
1108 if (ret <= 0) {
1109 ERR("consumer closed the command socket");
1110 goto error;
1111 }
1112
1113 ERR("consumer return code : %s",
1114 lttcomm_get_readable_code(-code));
1115
1116 goto exit;
1117 } else if (pollfd == consumer_data->metadata_sock.fd) {
1118 /* UST metadata requests */
1119 ret = ust_consumer_metadata_request(
1120 &consumer_data->metadata_sock);
1121 if (ret < 0) {
1122 ERR("Handling metadata request");
1123 goto error;
1124 }
1125 break;
1126 } else {
1127 ERR("Unknown pollfd");
1128 goto error;
1129 }
1130 }
1131 health_code_update();
1132 }
1133
1134 exit:
1135 error:
1136 /* Immediately set the consumerd state to stopped */
1137 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1138 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1139 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1140 consumer_data->type == LTTNG_CONSUMER32_UST) {
1141 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1142 } else {
1143 /* Code flow error... */
1144 assert(0);
1145 }
1146
1147 if (consumer_data->err_sock >= 0) {
1148 ret = close(consumer_data->err_sock);
1149 if (ret) {
1150 PERROR("close");
1151 }
1152 consumer_data->err_sock = -1;
1153 }
1154 if (consumer_data->cmd_sock >= 0) {
1155 ret = close(consumer_data->cmd_sock);
1156 if (ret) {
1157 PERROR("close");
1158 }
1159 consumer_data->cmd_sock = -1;
1160 }
1161 if (consumer_data->metadata_sock.fd >= 0) {
1162 ret = close(consumer_data->metadata_sock.fd);
1163 if (ret) {
1164 PERROR("close");
1165 }
1166 }
1167 /* Cleanup metadata socket mutex. */
1168 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1169 free(consumer_data->metadata_sock.lock);
1170
1171 if (sock >= 0) {
1172 ret = close(sock);
1173 if (ret) {
1174 PERROR("close");
1175 }
1176 }
1177
1178 unlink(consumer_data->err_unix_sock_path);
1179 unlink(consumer_data->cmd_unix_sock_path);
1180 consumer_data->pid = 0;
1181
1182 lttng_poll_clean(&events);
1183 error_poll:
1184 if (err) {
1185 health_error();
1186 ERR("Health error occurred in %s", __func__);
1187 }
1188 health_unregister();
1189 DBG("consumer thread cleanup completed");
1190
1191 return NULL;
1192 }
1193
1194 /*
1195 * This thread manage application communication.
1196 */
1197 static void *thread_manage_apps(void *data)
1198 {
1199 int i, ret, pollfd, err = -1;
1200 uint32_t revents, nb_fd;
1201 struct lttng_poll_event events;
1202
1203 DBG("[thread] Manage application started");
1204
1205 rcu_register_thread();
1206 rcu_thread_online();
1207
1208 health_register(HEALTH_TYPE_APP_MANAGE);
1209
1210 if (testpoint(thread_manage_apps)) {
1211 goto error_testpoint;
1212 }
1213
1214 health_code_update();
1215
1216 ret = sessiond_set_thread_pollset(&events, 2);
1217 if (ret < 0) {
1218 goto error_poll_create;
1219 }
1220
1221 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1222 if (ret < 0) {
1223 goto error;
1224 }
1225
1226 if (testpoint(thread_manage_apps_before_loop)) {
1227 goto error;
1228 }
1229
1230 health_code_update();
1231
1232 while (1) {
1233 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1234
1235 /* Inifinite blocking call, waiting for transmission */
1236 restart:
1237 health_poll_entry();
1238 ret = lttng_poll_wait(&events, -1);
1239 health_poll_exit();
1240 if (ret < 0) {
1241 /*
1242 * Restart interrupted system call.
1243 */
1244 if (errno == EINTR) {
1245 goto restart;
1246 }
1247 goto error;
1248 }
1249
1250 nb_fd = ret;
1251
1252 for (i = 0; i < nb_fd; i++) {
1253 /* Fetch once the poll data */
1254 revents = LTTNG_POLL_GETEV(&events, i);
1255 pollfd = LTTNG_POLL_GETFD(&events, i);
1256
1257 health_code_update();
1258
1259 /* Thread quit pipe has been closed. Killing thread. */
1260 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1261 if (ret) {
1262 err = 0;
1263 goto exit;
1264 }
1265
1266 /* Inspect the apps cmd pipe */
1267 if (pollfd == apps_cmd_pipe[0]) {
1268 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1269 ERR("Apps command pipe error");
1270 goto error;
1271 } else if (revents & LPOLLIN) {
1272 int sock;
1273
1274 /* Empty pipe */
1275 do {
1276 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1277 } while (ret < 0 && errno == EINTR);
1278 if (ret < 0 || ret < sizeof(sock)) {
1279 PERROR("read apps cmd pipe");
1280 goto error;
1281 }
1282
1283 health_code_update();
1284
1285 /*
1286 * We only monitor the error events of the socket. This
1287 * thread does not handle any incoming data from UST
1288 * (POLLIN).
1289 */
1290 ret = lttng_poll_add(&events, sock,
1291 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1292 if (ret < 0) {
1293 goto error;
1294 }
1295
1296 /* Set socket timeout for both receiving and ending */
1297 (void) lttcomm_setsockopt_rcv_timeout(sock,
1298 app_socket_timeout);
1299 (void) lttcomm_setsockopt_snd_timeout(sock,
1300 app_socket_timeout);
1301
1302 DBG("Apps with sock %d added to poll set", sock);
1303
1304 health_code_update();
1305
1306 break;
1307 }
1308 } else {
1309 /*
1310 * At this point, we know that a registered application made
1311 * the event at poll_wait.
1312 */
1313 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1314 /* Removing from the poll set */
1315 ret = lttng_poll_del(&events, pollfd);
1316 if (ret < 0) {
1317 goto error;
1318 }
1319
1320 /* Socket closed on remote end. */
1321 ust_app_unregister(pollfd);
1322 break;
1323 }
1324 }
1325
1326 health_code_update();
1327 }
1328 }
1329
1330 exit:
1331 error:
1332 lttng_poll_clean(&events);
1333 error_poll_create:
1334 error_testpoint:
1335 utils_close_pipe(apps_cmd_pipe);
1336 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1337
1338 /*
1339 * We don't clean the UST app hash table here since already registered
1340 * applications can still be controlled so let them be until the session
1341 * daemon dies or the applications stop.
1342 */
1343
1344 if (err) {
1345 health_error();
1346 ERR("Health error occurred in %s", __func__);
1347 }
1348 health_unregister();
1349 DBG("Application communication apps thread cleanup complete");
1350 rcu_thread_offline();
1351 rcu_unregister_thread();
1352 return NULL;
1353 }
1354
1355 /*
1356 * Send a socket to a thread This is called from the dispatch UST registration
1357 * thread once all sockets are set for the application.
1358 *
1359 * On success, return 0 else a negative value being the errno message of the
1360 * write().
1361 */
1362 static int send_socket_to_thread(int fd, int sock)
1363 {
1364 int ret;
1365
1366 /* Sockets MUST be set or else this should not have been called. */
1367 assert(fd >= 0);
1368 assert(sock >= 0);
1369
1370 do {
1371 ret = write(fd, &sock, sizeof(sock));
1372 } while (ret < 0 && errno == EINTR);
1373 if (ret < 0 || ret != sizeof(sock)) {
1374 PERROR("write apps pipe %d", fd);
1375 if (ret < 0) {
1376 ret = -errno;
1377 }
1378 goto error;
1379 }
1380
1381 /* All good. Don't send back the write positive ret value. */
1382 ret = 0;
1383 error:
1384 return ret;
1385 }
1386
1387 /*
1388 * Sanitize the wait queue of the dispatch registration thread meaning removing
1389 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1390 * notify socket is never received.
1391 */
1392 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1393 {
1394 int ret, nb_fd = 0, i;
1395 unsigned int fd_added = 0;
1396 struct lttng_poll_event events;
1397 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1398
1399 assert(wait_queue);
1400
1401 lttng_poll_init(&events);
1402
1403 /* Just skip everything for an empty queue. */
1404 if (!wait_queue->count) {
1405 goto end;
1406 }
1407
1408 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1409 if (ret < 0) {
1410 goto error_create;
1411 }
1412
1413 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1414 &wait_queue->head, head) {
1415 assert(wait_node->app);
1416 ret = lttng_poll_add(&events, wait_node->app->sock,
1417 LPOLLHUP | LPOLLERR);
1418 if (ret < 0) {
1419 goto error;
1420 }
1421
1422 fd_added = 1;
1423 }
1424
1425 if (!fd_added) {
1426 goto end;
1427 }
1428
1429 /*
1430 * Poll but don't block so we can quickly identify the faulty events and
1431 * clean them afterwards from the wait queue.
1432 */
1433 ret = lttng_poll_wait(&events, 0);
1434 if (ret < 0) {
1435 goto error;
1436 }
1437 nb_fd = ret;
1438
1439 for (i = 0; i < nb_fd; i++) {
1440 /* Get faulty FD. */
1441 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1442 int pollfd = LTTNG_POLL_GETFD(&events, i);
1443
1444 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1445 &wait_queue->head, head) {
1446 if (pollfd == wait_node->app->sock &&
1447 (revents & (LPOLLHUP | LPOLLERR))) {
1448 cds_list_del(&wait_node->head);
1449 wait_queue->count--;
1450 ust_app_destroy(wait_node->app);
1451 free(wait_node);
1452 break;
1453 }
1454 }
1455 }
1456
1457 if (nb_fd > 0) {
1458 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1459 }
1460
1461 end:
1462 lttng_poll_clean(&events);
1463 return;
1464
1465 error:
1466 lttng_poll_clean(&events);
1467 error_create:
1468 ERR("Unable to sanitize wait queue");
1469 return;
1470 }
1471
1472 /*
1473 * Dispatch request from the registration threads to the application
1474 * communication thread.
1475 */
1476 static void *thread_dispatch_ust_registration(void *data)
1477 {
1478 int ret, err = -1;
1479 struct cds_wfq_node *node;
1480 struct ust_command *ust_cmd = NULL;
1481 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1482 struct ust_reg_wait_queue wait_queue = {
1483 .count = 0,
1484 };
1485
1486 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1487
1488 health_code_update();
1489
1490 CDS_INIT_LIST_HEAD(&wait_queue.head);
1491
1492 DBG("[thread] Dispatch UST command started");
1493
1494 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1495 health_code_update();
1496
1497 /* Atomically prepare the queue futex */
1498 futex_nto1_prepare(&ust_cmd_queue.futex);
1499
1500 do {
1501 struct ust_app *app = NULL;
1502 ust_cmd = NULL;
1503
1504 /*
1505 * Make sure we don't have node(s) that have hung up before receiving
1506 * the notify socket. This is to clean the list in order to avoid
1507 * memory leaks from notify socket that are never seen.
1508 */
1509 sanitize_wait_queue(&wait_queue);
1510
1511 health_code_update();
1512 /* Dequeue command for registration */
1513 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1514 if (node == NULL) {
1515 DBG("Woken up but nothing in the UST command queue");
1516 /* Continue thread execution */
1517 break;
1518 }
1519
1520 ust_cmd = caa_container_of(node, struct ust_command, node);
1521
1522 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1523 " gid:%d sock:%d name:%s (version %d.%d)",
1524 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1525 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1526 ust_cmd->sock, ust_cmd->reg_msg.name,
1527 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1528
1529 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1530 wait_node = zmalloc(sizeof(*wait_node));
1531 if (!wait_node) {
1532 PERROR("zmalloc wait_node dispatch");
1533 ret = close(ust_cmd->sock);
1534 if (ret < 0) {
1535 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1536 }
1537 lttng_fd_put(1, LTTNG_FD_APPS);
1538 free(ust_cmd);
1539 goto error;
1540 }
1541 CDS_INIT_LIST_HEAD(&wait_node->head);
1542
1543 /* Create application object if socket is CMD. */
1544 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1545 ust_cmd->sock);
1546 if (!wait_node->app) {
1547 ret = close(ust_cmd->sock);
1548 if (ret < 0) {
1549 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1550 }
1551 lttng_fd_put(1, LTTNG_FD_APPS);
1552 free(wait_node);
1553 free(ust_cmd);
1554 continue;
1555 }
1556 /*
1557 * Add application to the wait queue so we can set the notify
1558 * socket before putting this object in the global ht.
1559 */
1560 cds_list_add(&wait_node->head, &wait_queue.head);
1561 wait_queue.count++;
1562
1563 free(ust_cmd);
1564 /*
1565 * We have to continue here since we don't have the notify
1566 * socket and the application MUST be added to the hash table
1567 * only at that moment.
1568 */
1569 continue;
1570 } else {
1571 /*
1572 * Look for the application in the local wait queue and set the
1573 * notify socket if found.
1574 */
1575 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1576 &wait_queue.head, head) {
1577 health_code_update();
1578 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1579 wait_node->app->notify_sock = ust_cmd->sock;
1580 cds_list_del(&wait_node->head);
1581 wait_queue.count--;
1582 app = wait_node->app;
1583 free(wait_node);
1584 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1585 break;
1586 }
1587 }
1588
1589 /*
1590 * With no application at this stage the received socket is
1591 * basically useless so close it before we free the cmd data
1592 * structure for good.
1593 */
1594 if (!app) {
1595 ret = close(ust_cmd->sock);
1596 if (ret < 0) {
1597 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1598 }
1599 lttng_fd_put(1, LTTNG_FD_APPS);
1600 }
1601 free(ust_cmd);
1602 }
1603
1604 if (app) {
1605 /*
1606 * @session_lock_list
1607 *
1608 * Lock the global session list so from the register up to the
1609 * registration done message, no thread can see the application
1610 * and change its state.
1611 */
1612 session_lock_list();
1613 rcu_read_lock();
1614
1615 /*
1616 * Add application to the global hash table. This needs to be
1617 * done before the update to the UST registry can locate the
1618 * application.
1619 */
1620 ust_app_add(app);
1621
1622 /* Set app version. This call will print an error if needed. */
1623 (void) ust_app_version(app);
1624
1625 /* Send notify socket through the notify pipe. */
1626 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1627 app->notify_sock);
1628 if (ret < 0) {
1629 rcu_read_unlock();
1630 session_unlock_list();
1631 /* No notify thread, stop the UST tracing. */
1632 goto error;
1633 }
1634
1635 /*
1636 * Update newly registered application with the tracing
1637 * registry info already enabled information.
1638 */
1639 update_ust_app(app->sock);
1640
1641 /*
1642 * Don't care about return value. Let the manage apps threads
1643 * handle app unregistration upon socket close.
1644 */
1645 (void) ust_app_register_done(app->sock);
1646
1647 /*
1648 * Even if the application socket has been closed, send the app
1649 * to the thread and unregistration will take place at that
1650 * place.
1651 */
1652 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1653 if (ret < 0) {
1654 rcu_read_unlock();
1655 session_unlock_list();
1656 /* No apps. thread, stop the UST tracing. */
1657 goto error;
1658 }
1659
1660 rcu_read_unlock();
1661 session_unlock_list();
1662 }
1663 } while (node != NULL);
1664
1665 health_poll_entry();
1666 /* Futex wait on queue. Blocking call on futex() */
1667 futex_nto1_wait(&ust_cmd_queue.futex);
1668 health_poll_exit();
1669 }
1670 /* Normal exit, no error */
1671 err = 0;
1672
1673 error:
1674 /* Clean up wait queue. */
1675 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1676 &wait_queue.head, head) {
1677 cds_list_del(&wait_node->head);
1678 wait_queue.count--;
1679 free(wait_node);
1680 }
1681
1682 DBG("Dispatch thread dying");
1683 if (err) {
1684 health_error();
1685 ERR("Health error occurred in %s", __func__);
1686 }
1687 health_unregister();
1688 return NULL;
1689 }
1690
1691 /*
1692 * This thread manage application registration.
1693 */
1694 static void *thread_registration_apps(void *data)
1695 {
1696 int sock = -1, i, ret, pollfd, err = -1;
1697 uint32_t revents, nb_fd;
1698 struct lttng_poll_event events;
1699 /*
1700 * Get allocated in this thread, enqueued to a global queue, dequeued and
1701 * freed in the manage apps thread.
1702 */
1703 struct ust_command *ust_cmd = NULL;
1704
1705 DBG("[thread] Manage application registration started");
1706
1707 health_register(HEALTH_TYPE_APP_REG);
1708
1709 if (testpoint(thread_registration_apps)) {
1710 goto error_testpoint;
1711 }
1712
1713 ret = lttcomm_listen_unix_sock(apps_sock);
1714 if (ret < 0) {
1715 goto error_listen;
1716 }
1717
1718 /*
1719 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1720 * more will be added to this poll set.
1721 */
1722 ret = sessiond_set_thread_pollset(&events, 2);
1723 if (ret < 0) {
1724 goto error_create_poll;
1725 }
1726
1727 /* Add the application registration socket */
1728 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1729 if (ret < 0) {
1730 goto error_poll_add;
1731 }
1732
1733 /* Notify all applications to register */
1734 ret = notify_ust_apps(1);
1735 if (ret < 0) {
1736 ERR("Failed to notify applications or create the wait shared memory.\n"
1737 "Execution continues but there might be problem for already\n"
1738 "running applications that wishes to register.");
1739 }
1740
1741 while (1) {
1742 DBG("Accepting application registration");
1743
1744 /* Inifinite blocking call, waiting for transmission */
1745 restart:
1746 health_poll_entry();
1747 ret = lttng_poll_wait(&events, -1);
1748 health_poll_exit();
1749 if (ret < 0) {
1750 /*
1751 * Restart interrupted system call.
1752 */
1753 if (errno == EINTR) {
1754 goto restart;
1755 }
1756 goto error;
1757 }
1758
1759 nb_fd = ret;
1760
1761 for (i = 0; i < nb_fd; i++) {
1762 health_code_update();
1763
1764 /* Fetch once the poll data */
1765 revents = LTTNG_POLL_GETEV(&events, i);
1766 pollfd = LTTNG_POLL_GETFD(&events, i);
1767
1768 /* Thread quit pipe has been closed. Killing thread. */
1769 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1770 if (ret) {
1771 err = 0;
1772 goto exit;
1773 }
1774
1775 /* Event on the registration socket */
1776 if (pollfd == apps_sock) {
1777 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1778 ERR("Register apps socket poll error");
1779 goto error;
1780 } else if (revents & LPOLLIN) {
1781 sock = lttcomm_accept_unix_sock(apps_sock);
1782 if (sock < 0) {
1783 goto error;
1784 }
1785
1786 /*
1787 * Set the CLOEXEC flag. Return code is useless because
1788 * either way, the show must go on.
1789 */
1790 (void) utils_set_fd_cloexec(sock);
1791
1792 /* Create UST registration command for enqueuing */
1793 ust_cmd = zmalloc(sizeof(struct ust_command));
1794 if (ust_cmd == NULL) {
1795 PERROR("ust command zmalloc");
1796 goto error;
1797 }
1798
1799 /*
1800 * Using message-based transmissions to ensure we don't
1801 * have to deal with partially received messages.
1802 */
1803 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1804 if (ret < 0) {
1805 ERR("Exhausted file descriptors allowed for applications.");
1806 free(ust_cmd);
1807 ret = close(sock);
1808 if (ret) {
1809 PERROR("close");
1810 }
1811 sock = -1;
1812 continue;
1813 }
1814
1815 health_code_update();
1816 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1817 if (ret < 0) {
1818 free(ust_cmd);
1819 /* Close socket of the application. */
1820 ret = close(sock);
1821 if (ret) {
1822 PERROR("close");
1823 }
1824 lttng_fd_put(LTTNG_FD_APPS, 1);
1825 sock = -1;
1826 continue;
1827 }
1828 health_code_update();
1829
1830 ust_cmd->sock = sock;
1831 sock = -1;
1832
1833 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1834 " gid:%d sock:%d name:%s (version %d.%d)",
1835 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1836 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1837 ust_cmd->sock, ust_cmd->reg_msg.name,
1838 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1839
1840 /*
1841 * Lock free enqueue the registration request. The red pill
1842 * has been taken! This apps will be part of the *system*.
1843 */
1844 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1845
1846 /*
1847 * Wake the registration queue futex. Implicit memory
1848 * barrier with the exchange in cds_wfq_enqueue.
1849 */
1850 futex_nto1_wake(&ust_cmd_queue.futex);
1851 }
1852 }
1853 }
1854 }
1855
1856 exit:
1857 error:
1858 if (err) {
1859 health_error();
1860 ERR("Health error occurred in %s", __func__);
1861 }
1862
1863 /* Notify that the registration thread is gone */
1864 notify_ust_apps(0);
1865
1866 if (apps_sock >= 0) {
1867 ret = close(apps_sock);
1868 if (ret) {
1869 PERROR("close");
1870 }
1871 }
1872 if (sock >= 0) {
1873 ret = close(sock);
1874 if (ret) {
1875 PERROR("close");
1876 }
1877 lttng_fd_put(LTTNG_FD_APPS, 1);
1878 }
1879 unlink(apps_unix_sock_path);
1880
1881 error_poll_add:
1882 lttng_poll_clean(&events);
1883 error_listen:
1884 error_create_poll:
1885 error_testpoint:
1886 DBG("UST Registration thread cleanup complete");
1887 health_unregister();
1888
1889 return NULL;
1890 }
1891
1892 /*
1893 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1894 * exec or it will fails.
1895 */
1896 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1897 {
1898 int ret, clock_ret;
1899 struct timespec timeout;
1900
1901 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1902 consumer_data->consumer_thread_is_ready = 0;
1903
1904 /* Setup pthread condition */
1905 ret = pthread_condattr_init(&consumer_data->condattr);
1906 if (ret != 0) {
1907 errno = ret;
1908 PERROR("pthread_condattr_init consumer data");
1909 goto error;
1910 }
1911
1912 /*
1913 * Set the monotonic clock in order to make sure we DO NOT jump in time
1914 * between the clock_gettime() call and the timedwait call. See bug #324
1915 * for a more details and how we noticed it.
1916 */
1917 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1918 if (ret != 0) {
1919 errno = ret;
1920 PERROR("pthread_condattr_setclock consumer data");
1921 goto error;
1922 }
1923
1924 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1925 if (ret != 0) {
1926 errno = ret;
1927 PERROR("pthread_cond_init consumer data");
1928 goto error;
1929 }
1930
1931 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1932 consumer_data);
1933 if (ret != 0) {
1934 PERROR("pthread_create consumer");
1935 ret = -1;
1936 goto error;
1937 }
1938
1939 /* We are about to wait on a pthread condition */
1940 pthread_mutex_lock(&consumer_data->cond_mutex);
1941
1942 /* Get time for sem_timedwait absolute timeout */
1943 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1944 /*
1945 * Set the timeout for the condition timed wait even if the clock gettime
1946 * call fails since we might loop on that call and we want to avoid to
1947 * increment the timeout too many times.
1948 */
1949 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1950
1951 /*
1952 * The following loop COULD be skipped in some conditions so this is why we
1953 * set ret to 0 in order to make sure at least one round of the loop is
1954 * done.
1955 */
1956 ret = 0;
1957
1958 /*
1959 * Loop until the condition is reached or when a timeout is reached. Note
1960 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1961 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1962 * possible. This loop does not take any chances and works with both of
1963 * them.
1964 */
1965 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1966 if (clock_ret < 0) {
1967 PERROR("clock_gettime spawn consumer");
1968 /* Infinite wait for the consumerd thread to be ready */
1969 ret = pthread_cond_wait(&consumer_data->cond,
1970 &consumer_data->cond_mutex);
1971 } else {
1972 ret = pthread_cond_timedwait(&consumer_data->cond,
1973 &consumer_data->cond_mutex, &timeout);
1974 }
1975 }
1976
1977 /* Release the pthread condition */
1978 pthread_mutex_unlock(&consumer_data->cond_mutex);
1979
1980 if (ret != 0) {
1981 errno = ret;
1982 if (ret == ETIMEDOUT) {
1983 /*
1984 * Call has timed out so we kill the kconsumerd_thread and return
1985 * an error.
1986 */
1987 ERR("Condition timed out. The consumer thread was never ready."
1988 " Killing it");
1989 ret = pthread_cancel(consumer_data->thread);
1990 if (ret < 0) {
1991 PERROR("pthread_cancel consumer thread");
1992 }
1993 } else {
1994 PERROR("pthread_cond_wait failed consumer thread");
1995 }
1996 goto error;
1997 }
1998
1999 pthread_mutex_lock(&consumer_data->pid_mutex);
2000 if (consumer_data->pid == 0) {
2001 ERR("Consumerd did not start");
2002 pthread_mutex_unlock(&consumer_data->pid_mutex);
2003 goto error;
2004 }
2005 pthread_mutex_unlock(&consumer_data->pid_mutex);
2006
2007 return 0;
2008
2009 error:
2010 return ret;
2011 }
2012
2013 /*
2014 * Join consumer thread
2015 */
2016 static int join_consumer_thread(struct consumer_data *consumer_data)
2017 {
2018 void *status;
2019
2020 /* Consumer pid must be a real one. */
2021 if (consumer_data->pid > 0) {
2022 int ret;
2023 ret = kill(consumer_data->pid, SIGTERM);
2024 if (ret) {
2025 ERR("Error killing consumer daemon");
2026 return ret;
2027 }
2028 return pthread_join(consumer_data->thread, &status);
2029 } else {
2030 return 0;
2031 }
2032 }
2033
2034 /*
2035 * Fork and exec a consumer daemon (consumerd).
2036 *
2037 * Return pid if successful else -1.
2038 */
2039 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2040 {
2041 int ret;
2042 pid_t pid;
2043 const char *consumer_to_use;
2044 const char *verbosity;
2045 struct stat st;
2046
2047 DBG("Spawning consumerd");
2048
2049 pid = fork();
2050 if (pid == 0) {
2051 /*
2052 * Exec consumerd.
2053 */
2054 if (opt_verbose_consumer) {
2055 verbosity = "--verbose";
2056 } else {
2057 verbosity = "--quiet";
2058 }
2059 switch (consumer_data->type) {
2060 case LTTNG_CONSUMER_KERNEL:
2061 /*
2062 * Find out which consumerd to execute. We will first try the
2063 * 64-bit path, then the sessiond's installation directory, and
2064 * fallback on the 32-bit one,
2065 */
2066 DBG3("Looking for a kernel consumer at these locations:");
2067 DBG3(" 1) %s", consumerd64_bin);
2068 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2069 DBG3(" 3) %s", consumerd32_bin);
2070 if (stat(consumerd64_bin, &st) == 0) {
2071 DBG3("Found location #1");
2072 consumer_to_use = consumerd64_bin;
2073 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2074 DBG3("Found location #2");
2075 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2076 } else if (stat(consumerd32_bin, &st) == 0) {
2077 DBG3("Found location #3");
2078 consumer_to_use = consumerd32_bin;
2079 } else {
2080 DBG("Could not find any valid consumerd executable");
2081 break;
2082 }
2083 DBG("Using kernel consumer at: %s", consumer_to_use);
2084 execl(consumer_to_use,
2085 "lttng-consumerd", verbosity, "-k",
2086 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2087 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2088 NULL);
2089 break;
2090 case LTTNG_CONSUMER64_UST:
2091 {
2092 char *tmpnew = NULL;
2093
2094 if (consumerd64_libdir[0] != '\0') {
2095 char *tmp;
2096 size_t tmplen;
2097
2098 tmp = getenv("LD_LIBRARY_PATH");
2099 if (!tmp) {
2100 tmp = "";
2101 }
2102 tmplen = strlen("LD_LIBRARY_PATH=")
2103 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2104 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2105 if (!tmpnew) {
2106 ret = -ENOMEM;
2107 goto error;
2108 }
2109 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2110 strcat(tmpnew, consumerd64_libdir);
2111 if (tmp[0] != '\0') {
2112 strcat(tmpnew, ":");
2113 strcat(tmpnew, tmp);
2114 }
2115 ret = putenv(tmpnew);
2116 if (ret) {
2117 ret = -errno;
2118 free(tmpnew);
2119 goto error;
2120 }
2121 }
2122 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2123 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2124 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2125 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2126 NULL);
2127 if (consumerd64_libdir[0] != '\0') {
2128 free(tmpnew);
2129 }
2130 if (ret) {
2131 goto error;
2132 }
2133 break;
2134 }
2135 case LTTNG_CONSUMER32_UST:
2136 {
2137 char *tmpnew = NULL;
2138
2139 if (consumerd32_libdir[0] != '\0') {
2140 char *tmp;
2141 size_t tmplen;
2142
2143 tmp = getenv("LD_LIBRARY_PATH");
2144 if (!tmp) {
2145 tmp = "";
2146 }
2147 tmplen = strlen("LD_LIBRARY_PATH=")
2148 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2149 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2150 if (!tmpnew) {
2151 ret = -ENOMEM;
2152 goto error;
2153 }
2154 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2155 strcat(tmpnew, consumerd32_libdir);
2156 if (tmp[0] != '\0') {
2157 strcat(tmpnew, ":");
2158 strcat(tmpnew, tmp);
2159 }
2160 ret = putenv(tmpnew);
2161 if (ret) {
2162 ret = -errno;
2163 free(tmpnew);
2164 goto error;
2165 }
2166 }
2167 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2168 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2169 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2170 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2171 NULL);
2172 if (consumerd32_libdir[0] != '\0') {
2173 free(tmpnew);
2174 }
2175 if (ret) {
2176 goto error;
2177 }
2178 break;
2179 }
2180 default:
2181 PERROR("unknown consumer type");
2182 exit(EXIT_FAILURE);
2183 }
2184 if (errno != 0) {
2185 PERROR("kernel start consumer exec");
2186 }
2187 exit(EXIT_FAILURE);
2188 } else if (pid > 0) {
2189 ret = pid;
2190 } else {
2191 PERROR("start consumer fork");
2192 ret = -errno;
2193 }
2194 error:
2195 return ret;
2196 }
2197
2198 /*
2199 * Spawn the consumerd daemon and session daemon thread.
2200 */
2201 static int start_consumerd(struct consumer_data *consumer_data)
2202 {
2203 int ret;
2204
2205 /*
2206 * Set the listen() state on the socket since there is a possible race
2207 * between the exec() of the consumer daemon and this call if place in the
2208 * consumer thread. See bug #366 for more details.
2209 */
2210 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2211 if (ret < 0) {
2212 goto error;
2213 }
2214
2215 pthread_mutex_lock(&consumer_data->pid_mutex);
2216 if (consumer_data->pid != 0) {
2217 pthread_mutex_unlock(&consumer_data->pid_mutex);
2218 goto end;
2219 }
2220
2221 ret = spawn_consumerd(consumer_data);
2222 if (ret < 0) {
2223 ERR("Spawning consumerd failed");
2224 pthread_mutex_unlock(&consumer_data->pid_mutex);
2225 goto error;
2226 }
2227
2228 /* Setting up the consumer_data pid */
2229 consumer_data->pid = ret;
2230 DBG2("Consumer pid %d", consumer_data->pid);
2231 pthread_mutex_unlock(&consumer_data->pid_mutex);
2232
2233 DBG2("Spawning consumer control thread");
2234 ret = spawn_consumer_thread(consumer_data);
2235 if (ret < 0) {
2236 ERR("Fatal error spawning consumer control thread");
2237 goto error;
2238 }
2239
2240 end:
2241 return 0;
2242
2243 error:
2244 /* Cleanup already created sockets on error. */
2245 if (consumer_data->err_sock >= 0) {
2246 int err;
2247
2248 err = close(consumer_data->err_sock);
2249 if (err < 0) {
2250 PERROR("close consumer data error socket");
2251 }
2252 }
2253 return ret;
2254 }
2255
2256 /*
2257 * Compute health status of each consumer. If one of them is zero (bad
2258 * state), we return 0.
2259 */
2260 static int check_consumer_health(void)
2261 {
2262 int ret;
2263
2264 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2265
2266 DBG3("Health consumer check %d", ret);
2267
2268 return ret;
2269 }
2270
2271 /*
2272 * Setup necessary data for kernel tracer action.
2273 */
2274 static int init_kernel_tracer(void)
2275 {
2276 int ret;
2277
2278 /* Modprobe lttng kernel modules */
2279 ret = modprobe_lttng_control();
2280 if (ret < 0) {
2281 goto error;
2282 }
2283
2284 /* Open debugfs lttng */
2285 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2286 if (kernel_tracer_fd < 0) {
2287 DBG("Failed to open %s", module_proc_lttng);
2288 ret = -1;
2289 goto error_open;
2290 }
2291
2292 /* Validate kernel version */
2293 ret = kernel_validate_version(kernel_tracer_fd);
2294 if (ret < 0) {
2295 goto error_version;
2296 }
2297
2298 ret = modprobe_lttng_data();
2299 if (ret < 0) {
2300 goto error_modules;
2301 }
2302
2303 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2304 return 0;
2305
2306 error_version:
2307 modprobe_remove_lttng_control();
2308 ret = close(kernel_tracer_fd);
2309 if (ret) {
2310 PERROR("close");
2311 }
2312 kernel_tracer_fd = -1;
2313 return LTTNG_ERR_KERN_VERSION;
2314
2315 error_modules:
2316 ret = close(kernel_tracer_fd);
2317 if (ret) {
2318 PERROR("close");
2319 }
2320
2321 error_open:
2322 modprobe_remove_lttng_control();
2323
2324 error:
2325 WARN("No kernel tracer available");
2326 kernel_tracer_fd = -1;
2327 if (!is_root) {
2328 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2329 } else {
2330 return LTTNG_ERR_KERN_NA;
2331 }
2332 }
2333
2334
2335 /*
2336 * Copy consumer output from the tracing session to the domain session. The
2337 * function also applies the right modification on a per domain basis for the
2338 * trace files destination directory.
2339 *
2340 * Should *NOT* be called with RCU read-side lock held.
2341 */
2342 static int copy_session_consumer(int domain, struct ltt_session *session)
2343 {
2344 int ret;
2345 const char *dir_name;
2346 struct consumer_output *consumer;
2347
2348 assert(session);
2349 assert(session->consumer);
2350
2351 switch (domain) {
2352 case LTTNG_DOMAIN_KERNEL:
2353 DBG3("Copying tracing session consumer output in kernel session");
2354 /*
2355 * XXX: We should audit the session creation and what this function
2356 * does "extra" in order to avoid a destroy since this function is used
2357 * in the domain session creation (kernel and ust) only. Same for UST
2358 * domain.
2359 */
2360 if (session->kernel_session->consumer) {
2361 consumer_destroy_output(session->kernel_session->consumer);
2362 }
2363 session->kernel_session->consumer =
2364 consumer_copy_output(session->consumer);
2365 /* Ease our life a bit for the next part */
2366 consumer = session->kernel_session->consumer;
2367 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2368 break;
2369 case LTTNG_DOMAIN_UST:
2370 DBG3("Copying tracing session consumer output in UST session");
2371 if (session->ust_session->consumer) {
2372 consumer_destroy_output(session->ust_session->consumer);
2373 }
2374 session->ust_session->consumer =
2375 consumer_copy_output(session->consumer);
2376 /* Ease our life a bit for the next part */
2377 consumer = session->ust_session->consumer;
2378 dir_name = DEFAULT_UST_TRACE_DIR;
2379 break;
2380 default:
2381 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2382 goto error;
2383 }
2384
2385 /* Append correct directory to subdir */
2386 strncat(consumer->subdir, dir_name,
2387 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2388 DBG3("Copy session consumer subdir %s", consumer->subdir);
2389
2390 ret = LTTNG_OK;
2391
2392 error:
2393 return ret;
2394 }
2395
2396 /*
2397 * Create an UST session and add it to the session ust list.
2398 *
2399 * Should *NOT* be called with RCU read-side lock held.
2400 */
2401 static int create_ust_session(struct ltt_session *session,
2402 struct lttng_domain *domain)
2403 {
2404 int ret;
2405 struct ltt_ust_session *lus = NULL;
2406
2407 assert(session);
2408 assert(domain);
2409 assert(session->consumer);
2410
2411 switch (domain->type) {
2412 case LTTNG_DOMAIN_UST:
2413 break;
2414 default:
2415 ERR("Unknown UST domain on create session %d", domain->type);
2416 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2417 goto error;
2418 }
2419
2420 DBG("Creating UST session");
2421
2422 lus = trace_ust_create_session(session->id);
2423 if (lus == NULL) {
2424 ret = LTTNG_ERR_UST_SESS_FAIL;
2425 goto error;
2426 }
2427
2428 lus->uid = session->uid;
2429 lus->gid = session->gid;
2430 lus->output_traces = session->output_traces;
2431 lus->snapshot_mode = session->snapshot_mode;
2432 session->ust_session = lus;
2433
2434 /* Copy session output to the newly created UST session */
2435 ret = copy_session_consumer(domain->type, session);
2436 if (ret != LTTNG_OK) {
2437 goto error;
2438 }
2439
2440 return LTTNG_OK;
2441
2442 error:
2443 free(lus);
2444 session->ust_session = NULL;
2445 return ret;
2446 }
2447
2448 /*
2449 * Create a kernel tracer session then create the default channel.
2450 */
2451 static int create_kernel_session(struct ltt_session *session)
2452 {
2453 int ret;
2454
2455 DBG("Creating kernel session");
2456
2457 ret = kernel_create_session(session, kernel_tracer_fd);
2458 if (ret < 0) {
2459 ret = LTTNG_ERR_KERN_SESS_FAIL;
2460 goto error;
2461 }
2462
2463 /* Code flow safety */
2464 assert(session->kernel_session);
2465
2466 /* Copy session output to the newly created Kernel session */
2467 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2468 if (ret != LTTNG_OK) {
2469 goto error;
2470 }
2471
2472 /* Create directory(ies) on local filesystem. */
2473 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2474 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2475 ret = run_as_mkdir_recursive(
2476 session->kernel_session->consumer->dst.trace_path,
2477 S_IRWXU | S_IRWXG, session->uid, session->gid);
2478 if (ret < 0) {
2479 if (ret != -EEXIST) {
2480 ERR("Trace directory creation error");
2481 goto error;
2482 }
2483 }
2484 }
2485
2486 session->kernel_session->uid = session->uid;
2487 session->kernel_session->gid = session->gid;
2488 session->kernel_session->output_traces = session->output_traces;
2489 session->kernel_session->snapshot_mode = session->snapshot_mode;
2490
2491 return LTTNG_OK;
2492
2493 error:
2494 trace_kernel_destroy_session(session->kernel_session);
2495 session->kernel_session = NULL;
2496 return ret;
2497 }
2498
2499 /*
2500 * Count number of session permitted by uid/gid.
2501 */
2502 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2503 {
2504 unsigned int i = 0;
2505 struct ltt_session *session;
2506
2507 DBG("Counting number of available session for UID %d GID %d",
2508 uid, gid);
2509 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2510 /*
2511 * Only list the sessions the user can control.
2512 */
2513 if (!session_access_ok(session, uid, gid)) {
2514 continue;
2515 }
2516 i++;
2517 }
2518 return i;
2519 }
2520
2521 /*
2522 * Process the command requested by the lttng client within the command
2523 * context structure. This function make sure that the return structure (llm)
2524 * is set and ready for transmission before returning.
2525 *
2526 * Return any error encountered or 0 for success.
2527 *
2528 * "sock" is only used for special-case var. len data.
2529 *
2530 * Should *NOT* be called with RCU read-side lock held.
2531 */
2532 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2533 int *sock_error)
2534 {
2535 int ret = LTTNG_OK;
2536 int need_tracing_session = 1;
2537 int need_domain;
2538
2539 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2540
2541 *sock_error = 0;
2542
2543 switch (cmd_ctx->lsm->cmd_type) {
2544 case LTTNG_CREATE_SESSION:
2545 case LTTNG_CREATE_SESSION_SNAPSHOT:
2546 case LTTNG_DESTROY_SESSION:
2547 case LTTNG_LIST_SESSIONS:
2548 case LTTNG_LIST_DOMAINS:
2549 case LTTNG_START_TRACE:
2550 case LTTNG_STOP_TRACE:
2551 case LTTNG_DATA_PENDING:
2552 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2553 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2554 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2555 case LTTNG_SNAPSHOT_RECORD:
2556 need_domain = 0;
2557 break;
2558 default:
2559 need_domain = 1;
2560 }
2561
2562 if (opt_no_kernel && need_domain
2563 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2564 if (!is_root) {
2565 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2566 } else {
2567 ret = LTTNG_ERR_KERN_NA;
2568 }
2569 goto error;
2570 }
2571
2572 /* Deny register consumer if we already have a spawned consumer. */
2573 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2574 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2575 if (kconsumer_data.pid > 0) {
2576 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2577 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2578 goto error;
2579 }
2580 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2581 }
2582
2583 /*
2584 * Check for command that don't needs to allocate a returned payload. We do
2585 * this here so we don't have to make the call for no payload at each
2586 * command.
2587 */
2588 switch(cmd_ctx->lsm->cmd_type) {
2589 case LTTNG_LIST_SESSIONS:
2590 case LTTNG_LIST_TRACEPOINTS:
2591 case LTTNG_LIST_TRACEPOINT_FIELDS:
2592 case LTTNG_LIST_DOMAINS:
2593 case LTTNG_LIST_CHANNELS:
2594 case LTTNG_LIST_EVENTS:
2595 break;
2596 default:
2597 /* Setup lttng message with no payload */
2598 ret = setup_lttng_msg(cmd_ctx, 0);
2599 if (ret < 0) {
2600 /* This label does not try to unlock the session */
2601 goto init_setup_error;
2602 }
2603 }
2604
2605 /* Commands that DO NOT need a session. */
2606 switch (cmd_ctx->lsm->cmd_type) {
2607 case LTTNG_CREATE_SESSION:
2608 case LTTNG_CREATE_SESSION_SNAPSHOT:
2609 case LTTNG_CALIBRATE:
2610 case LTTNG_LIST_SESSIONS:
2611 case LTTNG_LIST_TRACEPOINTS:
2612 case LTTNG_LIST_TRACEPOINT_FIELDS:
2613 need_tracing_session = 0;
2614 break;
2615 default:
2616 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2617 /*
2618 * We keep the session list lock across _all_ commands
2619 * for now, because the per-session lock does not
2620 * handle teardown properly.
2621 */
2622 session_lock_list();
2623 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2624 if (cmd_ctx->session == NULL) {
2625 ret = LTTNG_ERR_SESS_NOT_FOUND;
2626 goto error;
2627 } else {
2628 /* Acquire lock for the session */
2629 session_lock(cmd_ctx->session);
2630 }
2631 break;
2632 }
2633
2634 if (!need_domain) {
2635 goto skip_domain;
2636 }
2637
2638 /*
2639 * Check domain type for specific "pre-action".
2640 */
2641 switch (cmd_ctx->lsm->domain.type) {
2642 case LTTNG_DOMAIN_KERNEL:
2643 if (!is_root) {
2644 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2645 goto error;
2646 }
2647
2648 /* Kernel tracer check */
2649 if (kernel_tracer_fd == -1) {
2650 /* Basically, load kernel tracer modules */
2651 ret = init_kernel_tracer();
2652 if (ret != 0) {
2653 goto error;
2654 }
2655 }
2656
2657 /* Consumer is in an ERROR state. Report back to client */
2658 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2659 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2660 goto error;
2661 }
2662
2663 /* Need a session for kernel command */
2664 if (need_tracing_session) {
2665 if (cmd_ctx->session->kernel_session == NULL) {
2666 ret = create_kernel_session(cmd_ctx->session);
2667 if (ret < 0) {
2668 ret = LTTNG_ERR_KERN_SESS_FAIL;
2669 goto error;
2670 }
2671 }
2672
2673 /* Start the kernel consumer daemon */
2674 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2675 if (kconsumer_data.pid == 0 &&
2676 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2677 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2678 ret = start_consumerd(&kconsumer_data);
2679 if (ret < 0) {
2680 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2681 goto error;
2682 }
2683 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2684 } else {
2685 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2686 }
2687
2688 /*
2689 * The consumer was just spawned so we need to add the socket to
2690 * the consumer output of the session if exist.
2691 */
2692 ret = consumer_create_socket(&kconsumer_data,
2693 cmd_ctx->session->kernel_session->consumer);
2694 if (ret < 0) {
2695 goto error;
2696 }
2697 }
2698
2699 break;
2700 case LTTNG_DOMAIN_UST:
2701 {
2702 /* Consumer is in an ERROR state. Report back to client */
2703 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2704 ret = LTTNG_ERR_NO_USTCONSUMERD;
2705 goto error;
2706 }
2707
2708 if (need_tracing_session) {
2709 /* Create UST session if none exist. */
2710 if (cmd_ctx->session->ust_session == NULL) {
2711 ret = create_ust_session(cmd_ctx->session,
2712 &cmd_ctx->lsm->domain);
2713 if (ret != LTTNG_OK) {
2714 goto error;
2715 }
2716 }
2717
2718 /* Start the UST consumer daemons */
2719 /* 64-bit */
2720 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2721 if (consumerd64_bin[0] != '\0' &&
2722 ustconsumer64_data.pid == 0 &&
2723 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2724 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2725 ret = start_consumerd(&ustconsumer64_data);
2726 if (ret < 0) {
2727 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2728 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2729 goto error;
2730 }
2731
2732 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2733 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2734 } else {
2735 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2736 }
2737
2738 /*
2739 * Setup socket for consumer 64 bit. No need for atomic access
2740 * since it was set above and can ONLY be set in this thread.
2741 */
2742 ret = consumer_create_socket(&ustconsumer64_data,
2743 cmd_ctx->session->ust_session->consumer);
2744 if (ret < 0) {
2745 goto error;
2746 }
2747
2748 /* 32-bit */
2749 if (consumerd32_bin[0] != '\0' &&
2750 ustconsumer32_data.pid == 0 &&
2751 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2752 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2753 ret = start_consumerd(&ustconsumer32_data);
2754 if (ret < 0) {
2755 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2756 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2757 goto error;
2758 }
2759
2760 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2761 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2762 } else {
2763 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2764 }
2765
2766 /*
2767 * Setup socket for consumer 64 bit. No need for atomic access
2768 * since it was set above and can ONLY be set in this thread.
2769 */
2770 ret = consumer_create_socket(&ustconsumer32_data,
2771 cmd_ctx->session->ust_session->consumer);
2772 if (ret < 0) {
2773 goto error;
2774 }
2775 }
2776 break;
2777 }
2778 default:
2779 break;
2780 }
2781 skip_domain:
2782
2783 /* Validate consumer daemon state when start/stop trace command */
2784 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2785 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2786 switch (cmd_ctx->lsm->domain.type) {
2787 case LTTNG_DOMAIN_UST:
2788 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2789 ret = LTTNG_ERR_NO_USTCONSUMERD;
2790 goto error;
2791 }
2792 break;
2793 case LTTNG_DOMAIN_KERNEL:
2794 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2795 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2796 goto error;
2797 }
2798 break;
2799 }
2800 }
2801
2802 /*
2803 * Check that the UID or GID match that of the tracing session.
2804 * The root user can interact with all sessions.
2805 */
2806 if (need_tracing_session) {
2807 if (!session_access_ok(cmd_ctx->session,
2808 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2809 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2810 ret = LTTNG_ERR_EPERM;
2811 goto error;
2812 }
2813 }
2814
2815 /*
2816 * Send relayd information to consumer as soon as we have a domain and a
2817 * session defined.
2818 */
2819 if (cmd_ctx->session && need_domain) {
2820 /*
2821 * Setup relayd if not done yet. If the relayd information was already
2822 * sent to the consumer, this call will gracefully return.
2823 */
2824 ret = cmd_setup_relayd(cmd_ctx->session);
2825 if (ret != LTTNG_OK) {
2826 goto error;
2827 }
2828 }
2829
2830 /* Process by command type */
2831 switch (cmd_ctx->lsm->cmd_type) {
2832 case LTTNG_ADD_CONTEXT:
2833 {
2834 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2835 cmd_ctx->lsm->u.context.channel_name,
2836 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2837 break;
2838 }
2839 case LTTNG_DISABLE_CHANNEL:
2840 {
2841 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2842 cmd_ctx->lsm->u.disable.channel_name);
2843 break;
2844 }
2845 case LTTNG_DISABLE_EVENT:
2846 {
2847 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2848 cmd_ctx->lsm->u.disable.channel_name,
2849 cmd_ctx->lsm->u.disable.name);
2850 break;
2851 }
2852 case LTTNG_DISABLE_ALL_EVENT:
2853 {
2854 DBG("Disabling all events");
2855
2856 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2857 cmd_ctx->lsm->u.disable.channel_name);
2858 break;
2859 }
2860 case LTTNG_ENABLE_CHANNEL:
2861 {
2862 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2863 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2864 break;
2865 }
2866 case LTTNG_ENABLE_EVENT:
2867 {
2868 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2869 cmd_ctx->lsm->u.enable.channel_name,
2870 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2871 break;
2872 }
2873 case LTTNG_ENABLE_ALL_EVENT:
2874 {
2875 DBG("Enabling all events");
2876
2877 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2878 cmd_ctx->lsm->u.enable.channel_name,
2879 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2880 break;
2881 }
2882 case LTTNG_LIST_TRACEPOINTS:
2883 {
2884 struct lttng_event *events;
2885 ssize_t nb_events;
2886
2887 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2888 if (nb_events < 0) {
2889 /* Return value is a negative lttng_error_code. */
2890 ret = -nb_events;
2891 goto error;
2892 }
2893
2894 /*
2895 * Setup lttng message with payload size set to the event list size in
2896 * bytes and then copy list into the llm payload.
2897 */
2898 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2899 if (ret < 0) {
2900 free(events);
2901 goto setup_error;
2902 }
2903
2904 /* Copy event list into message payload */
2905 memcpy(cmd_ctx->llm->payload, events,
2906 sizeof(struct lttng_event) * nb_events);
2907
2908 free(events);
2909
2910 ret = LTTNG_OK;
2911 break;
2912 }
2913 case LTTNG_LIST_TRACEPOINT_FIELDS:
2914 {
2915 struct lttng_event_field *fields;
2916 ssize_t nb_fields;
2917
2918 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2919 &fields);
2920 if (nb_fields < 0) {
2921 /* Return value is a negative lttng_error_code. */
2922 ret = -nb_fields;
2923 goto error;
2924 }
2925
2926 /*
2927 * Setup lttng message with payload size set to the event list size in
2928 * bytes and then copy list into the llm payload.
2929 */
2930 ret = setup_lttng_msg(cmd_ctx,
2931 sizeof(struct lttng_event_field) * nb_fields);
2932 if (ret < 0) {
2933 free(fields);
2934 goto setup_error;
2935 }
2936
2937 /* Copy event list into message payload */
2938 memcpy(cmd_ctx->llm->payload, fields,
2939 sizeof(struct lttng_event_field) * nb_fields);
2940
2941 free(fields);
2942
2943 ret = LTTNG_OK;
2944 break;
2945 }
2946 case LTTNG_SET_CONSUMER_URI:
2947 {
2948 size_t nb_uri, len;
2949 struct lttng_uri *uris;
2950
2951 nb_uri = cmd_ctx->lsm->u.uri.size;
2952 len = nb_uri * sizeof(struct lttng_uri);
2953
2954 if (nb_uri == 0) {
2955 ret = LTTNG_ERR_INVALID;
2956 goto error;
2957 }
2958
2959 uris = zmalloc(len);
2960 if (uris == NULL) {
2961 ret = LTTNG_ERR_FATAL;
2962 goto error;
2963 }
2964
2965 /* Receive variable len data */
2966 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2967 ret = lttcomm_recv_unix_sock(sock, uris, len);
2968 if (ret <= 0) {
2969 DBG("No URIs received from client... continuing");
2970 *sock_error = 1;
2971 ret = LTTNG_ERR_SESSION_FAIL;
2972 free(uris);
2973 goto error;
2974 }
2975
2976 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2977 nb_uri, uris);
2978 if (ret != LTTNG_OK) {
2979 free(uris);
2980 goto error;
2981 }
2982
2983 /*
2984 * XXX: 0 means that this URI should be applied on the session. Should
2985 * be a DOMAIN enuam.
2986 */
2987 if (cmd_ctx->lsm->domain.type == 0) {
2988 /* Add the URI for the UST session if a consumer is present. */
2989 if (cmd_ctx->session->ust_session &&
2990 cmd_ctx->session->ust_session->consumer) {
2991 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2992 nb_uri, uris);
2993 } else if (cmd_ctx->session->kernel_session &&
2994 cmd_ctx->session->kernel_session->consumer) {
2995 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2996 cmd_ctx->session, nb_uri, uris);
2997 }
2998 }
2999
3000 free(uris);
3001
3002 break;
3003 }
3004 case LTTNG_START_TRACE:
3005 {
3006 ret = cmd_start_trace(cmd_ctx->session);
3007 break;
3008 }
3009 case LTTNG_STOP_TRACE:
3010 {
3011 ret = cmd_stop_trace(cmd_ctx->session);
3012 break;
3013 }
3014 case LTTNG_CREATE_SESSION:
3015 {
3016 size_t nb_uri, len;
3017 struct lttng_uri *uris = NULL;
3018
3019 nb_uri = cmd_ctx->lsm->u.uri.size;
3020 len = nb_uri * sizeof(struct lttng_uri);
3021
3022 if (nb_uri > 0) {
3023 uris = zmalloc(len);
3024 if (uris == NULL) {
3025 ret = LTTNG_ERR_FATAL;
3026 goto error;
3027 }
3028
3029 /* Receive variable len data */
3030 DBG("Waiting for %zu URIs from client ...", nb_uri);
3031 ret = lttcomm_recv_unix_sock(sock, uris, len);
3032 if (ret <= 0) {
3033 DBG("No URIs received from client... continuing");
3034 *sock_error = 1;
3035 ret = LTTNG_ERR_SESSION_FAIL;
3036 free(uris);
3037 goto error;
3038 }
3039
3040 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3041 DBG("Creating session with ONE network URI is a bad call");
3042 ret = LTTNG_ERR_SESSION_FAIL;
3043 free(uris);
3044 goto error;
3045 }
3046 }
3047
3048 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3049 &cmd_ctx->creds);
3050
3051 free(uris);
3052
3053 break;
3054 }
3055 case LTTNG_DESTROY_SESSION:
3056 {
3057 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3058
3059 /* Set session to NULL so we do not unlock it after free. */
3060 cmd_ctx->session = NULL;
3061 break;
3062 }
3063 case LTTNG_LIST_DOMAINS:
3064 {
3065 ssize_t nb_dom;
3066 struct lttng_domain *domains;
3067
3068 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3069 if (nb_dom < 0) {
3070 /* Return value is a negative lttng_error_code. */
3071 ret = -nb_dom;
3072 goto error;
3073 }
3074
3075 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3076 if (ret < 0) {
3077 free(domains);
3078 goto setup_error;
3079 }
3080
3081 /* Copy event list into message payload */
3082 memcpy(cmd_ctx->llm->payload, domains,
3083 nb_dom * sizeof(struct lttng_domain));
3084
3085 free(domains);
3086
3087 ret = LTTNG_OK;
3088 break;
3089 }
3090 case LTTNG_LIST_CHANNELS:
3091 {
3092 int nb_chan;
3093 struct lttng_channel *channels;
3094
3095 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3096 cmd_ctx->session, &channels);
3097 if (nb_chan < 0) {
3098 /* Return value is a negative lttng_error_code. */
3099 ret = -nb_chan;
3100 goto error;
3101 }
3102
3103 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3104 if (ret < 0) {
3105 free(channels);
3106 goto setup_error;
3107 }
3108
3109 /* Copy event list into message payload */
3110 memcpy(cmd_ctx->llm->payload, channels,
3111 nb_chan * sizeof(struct lttng_channel));
3112
3113 free(channels);
3114
3115 ret = LTTNG_OK;
3116 break;
3117 }
3118 case LTTNG_LIST_EVENTS:
3119 {
3120 ssize_t nb_event;
3121 struct lttng_event *events = NULL;
3122
3123 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3124 cmd_ctx->lsm->u.list.channel_name, &events);
3125 if (nb_event < 0) {
3126 /* Return value is a negative lttng_error_code. */
3127 ret = -nb_event;
3128 goto error;
3129 }
3130
3131 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3132 if (ret < 0) {
3133 free(events);
3134 goto setup_error;
3135 }
3136
3137 /* Copy event list into message payload */
3138 memcpy(cmd_ctx->llm->payload, events,
3139 nb_event * sizeof(struct lttng_event));
3140
3141 free(events);
3142
3143 ret = LTTNG_OK;
3144 break;
3145 }
3146 case LTTNG_LIST_SESSIONS:
3147 {
3148 unsigned int nr_sessions;
3149
3150 session_lock_list();
3151 nr_sessions = lttng_sessions_count(
3152 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3153 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3154
3155 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3156 if (ret < 0) {
3157 session_unlock_list();
3158 goto setup_error;
3159 }
3160
3161 /* Filled the session array */
3162 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3163 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3164 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3165
3166 session_unlock_list();
3167
3168 ret = LTTNG_OK;
3169 break;
3170 }
3171 case LTTNG_CALIBRATE:
3172 {
3173 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3174 &cmd_ctx->lsm->u.calibrate);
3175 break;
3176 }
3177 case LTTNG_REGISTER_CONSUMER:
3178 {
3179 struct consumer_data *cdata;
3180
3181 switch (cmd_ctx->lsm->domain.type) {
3182 case LTTNG_DOMAIN_KERNEL:
3183 cdata = &kconsumer_data;
3184 break;
3185 default:
3186 ret = LTTNG_ERR_UND;
3187 goto error;
3188 }
3189
3190 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3191 cmd_ctx->lsm->u.reg.path, cdata);
3192 break;
3193 }
3194 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3195 {
3196 struct lttng_filter_bytecode *bytecode;
3197
3198 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3199 ret = LTTNG_ERR_FILTER_INVAL;
3200 goto error;
3201 }
3202 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3203 ret = LTTNG_ERR_FILTER_INVAL;
3204 goto error;
3205 }
3206 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3207 if (!bytecode) {
3208 ret = LTTNG_ERR_FILTER_NOMEM;
3209 goto error;
3210 }
3211 /* Receive var. len. data */
3212 DBG("Receiving var len data from client ...");
3213 ret = lttcomm_recv_unix_sock(sock, bytecode,
3214 cmd_ctx->lsm->u.enable.bytecode_len);
3215 if (ret <= 0) {
3216 DBG("Nothing recv() from client var len data... continuing");
3217 *sock_error = 1;
3218 ret = LTTNG_ERR_FILTER_INVAL;
3219 goto error;
3220 }
3221
3222 if (bytecode->len + sizeof(*bytecode)
3223 != cmd_ctx->lsm->u.enable.bytecode_len) {
3224 free(bytecode);
3225 ret = LTTNG_ERR_FILTER_INVAL;
3226 goto error;
3227 }
3228
3229 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3230 cmd_ctx->lsm->u.enable.channel_name,
3231 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3232 break;
3233 }
3234 case LTTNG_DATA_PENDING:
3235 {
3236 ret = cmd_data_pending(cmd_ctx->session);
3237 break;
3238 }
3239 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3240 {
3241 struct lttcomm_lttng_output_id reply;
3242
3243 ret = cmd_snapshot_add_output(cmd_ctx->session,
3244 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3245 if (ret != LTTNG_OK) {
3246 goto error;
3247 }
3248
3249 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3250 if (ret < 0) {
3251 goto setup_error;
3252 }
3253
3254 /* Copy output list into message payload */
3255 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3256 ret = LTTNG_OK;
3257 break;
3258 }
3259 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3260 {
3261 ret = cmd_snapshot_del_output(cmd_ctx->session,
3262 &cmd_ctx->lsm->u.snapshot_output.output);
3263 break;
3264 }
3265 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3266 {
3267 ssize_t nb_output;
3268 struct lttng_snapshot_output *outputs = NULL;
3269
3270 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3271 if (nb_output < 0) {
3272 ret = -nb_output;
3273 goto error;
3274 }
3275
3276 ret = setup_lttng_msg(cmd_ctx,
3277 nb_output * sizeof(struct lttng_snapshot_output));
3278 if (ret < 0) {
3279 free(outputs);
3280 goto setup_error;
3281 }
3282
3283 if (outputs) {
3284 /* Copy output list into message payload */
3285 memcpy(cmd_ctx->llm->payload, outputs,
3286 nb_output * sizeof(struct lttng_snapshot_output));
3287 free(outputs);
3288 }
3289
3290 ret = LTTNG_OK;
3291 break;
3292 }
3293 case LTTNG_SNAPSHOT_RECORD:
3294 {
3295 ret = cmd_snapshot_record(cmd_ctx->session,
3296 &cmd_ctx->lsm->u.snapshot_record.output,
3297 cmd_ctx->lsm->u.snapshot_record.wait);
3298 break;
3299 }
3300 case LTTNG_CREATE_SESSION_SNAPSHOT:
3301 {
3302 size_t nb_uri, len;
3303 struct lttng_uri *uris = NULL;
3304
3305 nb_uri = cmd_ctx->lsm->u.uri.size;
3306 len = nb_uri * sizeof(struct lttng_uri);
3307
3308 if (nb_uri > 0) {
3309 uris = zmalloc(len);
3310 if (uris == NULL) {
3311 ret = LTTNG_ERR_FATAL;
3312 goto error;
3313 }
3314
3315 /* Receive variable len data */
3316 DBG("Waiting for %zu URIs from client ...", nb_uri);
3317 ret = lttcomm_recv_unix_sock(sock, uris, len);
3318 if (ret <= 0) {
3319 DBG("No URIs received from client... continuing");
3320 *sock_error = 1;
3321 ret = LTTNG_ERR_SESSION_FAIL;
3322 free(uris);
3323 goto error;
3324 }
3325
3326 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3327 DBG("Creating session with ONE network URI is a bad call");
3328 ret = LTTNG_ERR_SESSION_FAIL;
3329 free(uris);
3330 goto error;
3331 }
3332 }
3333
3334 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3335 nb_uri, &cmd_ctx->creds);
3336 free(uris);
3337 break;
3338 }
3339 default:
3340 ret = LTTNG_ERR_UND;
3341 break;
3342 }
3343
3344 error:
3345 if (cmd_ctx->llm == NULL) {
3346 DBG("Missing llm structure. Allocating one.");
3347 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3348 goto setup_error;
3349 }
3350 }
3351 /* Set return code */
3352 cmd_ctx->llm->ret_code = ret;
3353 setup_error:
3354 if (cmd_ctx->session) {
3355 session_unlock(cmd_ctx->session);
3356 }
3357 if (need_tracing_session) {
3358 session_unlock_list();
3359 }
3360 init_setup_error:
3361 return ret;
3362 }
3363
3364 /*
3365 * Thread managing health check socket.
3366 */
3367 static void *thread_manage_health(void *data)
3368 {
3369 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3370 uint32_t revents, nb_fd;
3371 struct lttng_poll_event events;
3372 struct lttcomm_health_msg msg;
3373 struct lttcomm_health_data reply;
3374
3375 DBG("[thread] Manage health check started");
3376
3377 rcu_register_thread();
3378
3379 /* We might hit an error path before this is created. */
3380 lttng_poll_init(&events);
3381
3382 /* Create unix socket */
3383 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3384 if (sock < 0) {
3385 ERR("Unable to create health check Unix socket");
3386 ret = -1;
3387 goto error;
3388 }
3389
3390 /*
3391 * Set the CLOEXEC flag. Return code is useless because either way, the
3392 * show must go on.
3393 */
3394 (void) utils_set_fd_cloexec(sock);
3395
3396 ret = lttcomm_listen_unix_sock(sock);
3397 if (ret < 0) {
3398 goto error;
3399 }
3400
3401 /*
3402 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3403 * more will be added to this poll set.
3404 */
3405 ret = sessiond_set_thread_pollset(&events, 2);
3406 if (ret < 0) {
3407 goto error;
3408 }
3409
3410 /* Add the application registration socket */
3411 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3412 if (ret < 0) {
3413 goto error;
3414 }
3415
3416 while (1) {
3417 DBG("Health check ready");
3418
3419 /* Inifinite blocking call, waiting for transmission */
3420 restart:
3421 ret = lttng_poll_wait(&events, -1);
3422 if (ret < 0) {
3423 /*
3424 * Restart interrupted system call.
3425 */
3426 if (errno == EINTR) {
3427 goto restart;
3428 }
3429 goto error;
3430 }
3431
3432 nb_fd = ret;
3433
3434 for (i = 0; i < nb_fd; i++) {
3435 /* Fetch once the poll data */
3436 revents = LTTNG_POLL_GETEV(&events, i);
3437 pollfd = LTTNG_POLL_GETFD(&events, i);
3438
3439 /* Thread quit pipe has been closed. Killing thread. */
3440 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3441 if (ret) {
3442 err = 0;
3443 goto exit;
3444 }
3445
3446 /* Event on the registration socket */
3447 if (pollfd == sock) {
3448 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3449 ERR("Health socket poll error");
3450 goto error;
3451 }
3452 }
3453 }
3454
3455 new_sock = lttcomm_accept_unix_sock(sock);
3456 if (new_sock < 0) {
3457 goto error;
3458 }
3459
3460 /*
3461 * Set the CLOEXEC flag. Return code is useless because either way, the
3462 * show must go on.
3463 */
3464 (void) utils_set_fd_cloexec(new_sock);
3465
3466 DBG("Receiving data from client for health...");
3467 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3468 if (ret <= 0) {
3469 DBG("Nothing recv() from client... continuing");
3470 ret = close(new_sock);
3471 if (ret) {
3472 PERROR("close");
3473 }
3474 new_sock = -1;
3475 continue;
3476 }
3477
3478 rcu_thread_online();
3479
3480 switch (msg.component) {
3481 case LTTNG_HEALTH_CMD:
3482 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3483 break;
3484 case LTTNG_HEALTH_APP_MANAGE:
3485 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3486 break;
3487 case LTTNG_HEALTH_APP_REG:
3488 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3489 break;
3490 case LTTNG_HEALTH_KERNEL:
3491 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3492 break;
3493 case LTTNG_HEALTH_CONSUMER:
3494 reply.ret_code = check_consumer_health();
3495 break;
3496 case LTTNG_HEALTH_HT_CLEANUP:
3497 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3498 break;
3499 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3500 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3501 break;
3502 case LTTNG_HEALTH_APP_REG_DISPATCH:
3503 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3504 break;
3505 case LTTNG_HEALTH_ALL:
3506 reply.ret_code =
3507 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3508 health_check_state(HEALTH_TYPE_APP_REG) &&
3509 health_check_state(HEALTH_TYPE_CMD) &&
3510 health_check_state(HEALTH_TYPE_KERNEL) &&
3511 check_consumer_health() &&
3512 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3513 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3514 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3515 break;
3516 default:
3517 reply.ret_code = LTTNG_ERR_UND;
3518 break;
3519 }
3520
3521 /*
3522 * Flip ret value since 0 is a success and 1 indicates a bad health for
3523 * the client where in the sessiond it is the opposite. Again, this is
3524 * just to make things easier for us poor developer which enjoy a lot
3525 * lazyness.
3526 */
3527 if (reply.ret_code == 0 || reply.ret_code == 1) {
3528 reply.ret_code = !reply.ret_code;
3529 }
3530
3531 DBG2("Health check return value %d", reply.ret_code);
3532
3533 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3534 if (ret < 0) {
3535 ERR("Failed to send health data back to client");
3536 }
3537
3538 /* End of transmission */
3539 ret = close(new_sock);
3540 if (ret) {
3541 PERROR("close");
3542 }
3543 new_sock = -1;
3544 }
3545
3546 exit:
3547 error:
3548 if (err) {
3549 ERR("Health error occurred in %s", __func__);
3550 }
3551 DBG("Health check thread dying");
3552 unlink(health_unix_sock_path);
3553 if (sock >= 0) {
3554 ret = close(sock);
3555 if (ret) {
3556 PERROR("close");
3557 }
3558 }
3559
3560 lttng_poll_clean(&events);
3561
3562 rcu_unregister_thread();
3563 return NULL;
3564 }
3565
3566 /*
3567 * This thread manage all clients request using the unix client socket for
3568 * communication.
3569 */
3570 static void *thread_manage_clients(void *data)
3571 {
3572 int sock = -1, ret, i, pollfd, err = -1;
3573 int sock_error;
3574 uint32_t revents, nb_fd;
3575 struct command_ctx *cmd_ctx = NULL;
3576 struct lttng_poll_event events;
3577
3578 DBG("[thread] Manage client started");
3579
3580 rcu_register_thread();
3581
3582 health_register(HEALTH_TYPE_CMD);
3583
3584 if (testpoint(thread_manage_clients)) {
3585 goto error_testpoint;
3586 }
3587
3588 health_code_update();
3589
3590 ret = lttcomm_listen_unix_sock(client_sock);
3591 if (ret < 0) {
3592 goto error_listen;
3593 }
3594
3595 /*
3596 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3597 * more will be added to this poll set.
3598 */
3599 ret = sessiond_set_thread_pollset(&events, 2);
3600 if (ret < 0) {
3601 goto error_create_poll;
3602 }
3603
3604 /* Add the application registration socket */
3605 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3606 if (ret < 0) {
3607 goto error;
3608 }
3609
3610 /*
3611 * Notify parent pid that we are ready to accept command for client side.
3612 */
3613 if (opt_sig_parent) {
3614 kill(ppid, SIGUSR1);
3615 }
3616
3617 if (testpoint(thread_manage_clients_before_loop)) {
3618 goto error;
3619 }
3620
3621 health_code_update();
3622
3623 while (1) {
3624 DBG("Accepting client command ...");
3625
3626 /* Inifinite blocking call, waiting for transmission */
3627 restart:
3628 health_poll_entry();
3629 ret = lttng_poll_wait(&events, -1);
3630 health_poll_exit();
3631 if (ret < 0) {
3632 /*
3633 * Restart interrupted system call.
3634 */
3635 if (errno == EINTR) {
3636 goto restart;
3637 }
3638 goto error;
3639 }
3640
3641 nb_fd = ret;
3642
3643 for (i = 0; i < nb_fd; i++) {
3644 /* Fetch once the poll data */
3645 revents = LTTNG_POLL_GETEV(&events, i);
3646 pollfd = LTTNG_POLL_GETFD(&events, i);
3647
3648 health_code_update();
3649
3650 /* Thread quit pipe has been closed. Killing thread. */
3651 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3652 if (ret) {
3653 err = 0;
3654 goto exit;
3655 }
3656
3657 /* Event on the registration socket */
3658 if (pollfd == client_sock) {
3659 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3660 ERR("Client socket poll error");
3661 goto error;
3662 }
3663 }
3664 }
3665
3666 DBG("Wait for client response");
3667
3668 health_code_update();
3669
3670 sock = lttcomm_accept_unix_sock(client_sock);
3671 if (sock < 0) {
3672 goto error;
3673 }
3674
3675 /*
3676 * Set the CLOEXEC flag. Return code is useless because either way, the
3677 * show must go on.
3678 */
3679 (void) utils_set_fd_cloexec(sock);
3680
3681 /* Set socket option for credentials retrieval */
3682 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3683 if (ret < 0) {
3684 goto error;
3685 }
3686
3687 /* Allocate context command to process the client request */
3688 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3689 if (cmd_ctx == NULL) {
3690 PERROR("zmalloc cmd_ctx");
3691 goto error;
3692 }
3693
3694 /* Allocate data buffer for reception */
3695 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3696 if (cmd_ctx->lsm == NULL) {
3697 PERROR("zmalloc cmd_ctx->lsm");
3698 goto error;
3699 }
3700
3701 cmd_ctx->llm = NULL;
3702 cmd_ctx->session = NULL;
3703
3704 health_code_update();
3705
3706 /*
3707 * Data is received from the lttng client. The struct
3708 * lttcomm_session_msg (lsm) contains the command and data request of
3709 * the client.
3710 */
3711 DBG("Receiving data from client ...");
3712 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3713 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3714 if (ret <= 0) {
3715 DBG("Nothing recv() from client... continuing");
3716 ret = close(sock);
3717 if (ret) {
3718 PERROR("close");
3719 }
3720 sock = -1;
3721 clean_command_ctx(&cmd_ctx);
3722 continue;
3723 }
3724
3725 health_code_update();
3726
3727 // TODO: Validate cmd_ctx including sanity check for
3728 // security purpose.
3729
3730 rcu_thread_online();
3731 /*
3732 * This function dispatch the work to the kernel or userspace tracer
3733 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3734 * informations for the client. The command context struct contains
3735 * everything this function may needs.
3736 */
3737 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3738 rcu_thread_offline();
3739 if (ret < 0) {
3740 ret = close(sock);
3741 if (ret) {
3742 PERROR("close");
3743 }
3744 sock = -1;
3745 /*
3746 * TODO: Inform client somehow of the fatal error. At
3747 * this point, ret < 0 means that a zmalloc failed
3748 * (ENOMEM). Error detected but still accept
3749 * command, unless a socket error has been
3750 * detected.
3751 */
3752 clean_command_ctx(&cmd_ctx);
3753 continue;
3754 }
3755
3756 health_code_update();
3757
3758 DBG("Sending response (size: %d, retcode: %s)",
3759 cmd_ctx->lttng_msg_size,
3760 lttng_strerror(-cmd_ctx->llm->ret_code));
3761 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3762 if (ret < 0) {
3763 ERR("Failed to send data back to client");
3764 }
3765
3766 /* End of transmission */
3767 ret = close(sock);
3768 if (ret) {
3769 PERROR("close");
3770 }
3771 sock = -1;
3772
3773 clean_command_ctx(&cmd_ctx);
3774
3775 health_code_update();
3776 }
3777
3778 exit:
3779 error:
3780 if (sock >= 0) {
3781 ret = close(sock);
3782 if (ret) {
3783 PERROR("close");
3784 }
3785 }
3786
3787 lttng_poll_clean(&events);
3788 clean_command_ctx(&cmd_ctx);
3789
3790 error_listen:
3791 error_create_poll:
3792 error_testpoint:
3793 unlink(client_unix_sock_path);
3794 if (client_sock >= 0) {
3795 ret = close(client_sock);
3796 if (ret) {
3797 PERROR("close");
3798 }
3799 }
3800
3801 if (err) {
3802 health_error();
3803 ERR("Health error occurred in %s", __func__);
3804 }
3805
3806 health_unregister();
3807
3808 DBG("Client thread dying");
3809
3810 rcu_unregister_thread();
3811 return NULL;
3812 }
3813
3814
3815 /*
3816 * usage function on stderr
3817 */
3818 static void usage(void)
3819 {
3820 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3821 fprintf(stderr, " -h, --help Display this usage.\n");
3822 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3823 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3824 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3825 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3826 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3827 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3828 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3829 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3830 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3831 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3832 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3833 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3834 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3835 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3836 fprintf(stderr, " -V, --version Show version number.\n");
3837 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3838 fprintf(stderr, " -q, --quiet No output at all.\n");
3839 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3840 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3841 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3842 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3843 }
3844
3845 /*
3846 * daemon argument parsing
3847 */
3848 static int parse_args(int argc, char **argv)
3849 {
3850 int c;
3851
3852 static struct option long_options[] = {
3853 { "client-sock", 1, 0, 'c' },
3854 { "apps-sock", 1, 0, 'a' },
3855 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3856 { "kconsumerd-err-sock", 1, 0, 'E' },
3857 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3858 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3859 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3860 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3861 { "consumerd32-path", 1, 0, 'u' },
3862 { "consumerd32-libdir", 1, 0, 'U' },
3863 { "consumerd64-path", 1, 0, 't' },
3864 { "consumerd64-libdir", 1, 0, 'T' },
3865 { "daemonize", 0, 0, 'd' },
3866 { "sig-parent", 0, 0, 'S' },
3867 { "help", 0, 0, 'h' },
3868 { "group", 1, 0, 'g' },
3869 { "version", 0, 0, 'V' },
3870 { "quiet", 0, 0, 'q' },
3871 { "verbose", 0, 0, 'v' },
3872 { "verbose-consumer", 0, 0, 'Z' },
3873 { "no-kernel", 0, 0, 'N' },
3874 { "pidfile", 1, 0, 'p' },
3875 { NULL, 0, 0, 0 }
3876 };
3877
3878 while (1) {
3879 int option_index = 0;
3880 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3881 long_options, &option_index);
3882 if (c == -1) {
3883 break;
3884 }
3885
3886 switch (c) {
3887 case 0:
3888 fprintf(stderr, "option %s", long_options[option_index].name);
3889 if (optarg) {
3890 fprintf(stderr, " with arg %s\n", optarg);
3891 }
3892 break;
3893 case 'c':
3894 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3895 break;
3896 case 'a':
3897 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3898 break;
3899 case 'd':
3900 opt_daemon = 1;
3901 break;
3902 case 'g':
3903 opt_tracing_group = optarg;
3904 break;
3905 case 'h':
3906 usage();
3907 exit(EXIT_FAILURE);
3908 case 'V':
3909 fprintf(stdout, "%s\n", VERSION);
3910 exit(EXIT_SUCCESS);
3911 case 'S':
3912 opt_sig_parent = 1;
3913 break;
3914 case 'E':
3915 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3916 break;
3917 case 'C':
3918 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3919 break;
3920 case 'F':
3921 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3922 break;
3923 case 'D':
3924 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3925 break;
3926 case 'H':
3927 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3928 break;
3929 case 'G':
3930 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3931 break;
3932 case 'N':
3933 opt_no_kernel = 1;
3934 break;
3935 case 'q':
3936 lttng_opt_quiet = 1;
3937 break;
3938 case 'v':
3939 /* Verbose level can increase using multiple -v */
3940 lttng_opt_verbose += 1;
3941 break;
3942 case 'Z':
3943 opt_verbose_consumer += 1;
3944 break;
3945 case 'u':
3946 consumerd32_bin= optarg;
3947 break;
3948 case 'U':
3949 consumerd32_libdir = optarg;
3950 break;
3951 case 't':
3952 consumerd64_bin = optarg;
3953 break;
3954 case 'T':
3955 consumerd64_libdir = optarg;
3956 break;
3957 case 'p':
3958 opt_pidfile = optarg;
3959 break;
3960 default:
3961 /* Unknown option or other error.
3962 * Error is printed by getopt, just return */
3963 return -1;
3964 }
3965 }
3966
3967 return 0;
3968 }
3969
3970 /*
3971 * Creates the two needed socket by the daemon.
3972 * apps_sock - The communication socket for all UST apps.
3973 * client_sock - The communication of the cli tool (lttng).
3974 */
3975 static int init_daemon_socket(void)
3976 {
3977 int ret = 0;
3978 mode_t old_umask;
3979
3980 old_umask = umask(0);
3981
3982 /* Create client tool unix socket */
3983 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3984 if (client_sock < 0) {
3985 ERR("Create unix sock failed: %s", client_unix_sock_path);
3986 ret = -1;
3987 goto end;
3988 }
3989
3990 /* Set the cloexec flag */
3991 ret = utils_set_fd_cloexec(client_sock);
3992 if (ret < 0) {
3993 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3994 "Continuing but note that the consumer daemon will have a "
3995 "reference to this socket on exec()", client_sock);
3996 }
3997
3998 /* File permission MUST be 660 */
3999 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4000 if (ret < 0) {
4001 ERR("Set file permissions failed: %s", client_unix_sock_path);
4002 PERROR("chmod");
4003 goto end;
4004 }
4005
4006 /* Create the application unix socket */
4007 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4008 if (apps_sock < 0) {
4009 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4010 ret = -1;
4011 goto end;
4012 }
4013
4014 /* Set the cloexec flag */
4015 ret = utils_set_fd_cloexec(apps_sock);
4016 if (ret < 0) {
4017 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4018 "Continuing but note that the consumer daemon will have a "
4019 "reference to this socket on exec()", apps_sock);
4020 }
4021
4022 /* File permission MUST be 666 */
4023 ret = chmod(apps_unix_sock_path,
4024 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4025 if (ret < 0) {
4026 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4027 PERROR("chmod");
4028 goto end;
4029 }
4030
4031 DBG3("Session daemon client socket %d and application socket %d created",
4032 client_sock, apps_sock);
4033
4034 end:
4035 umask(old_umask);
4036 return ret;
4037 }
4038
4039 /*
4040 * Check if the global socket is available, and if a daemon is answering at the
4041 * other side. If yes, error is returned.
4042 */
4043 static int check_existing_daemon(void)
4044 {
4045 /* Is there anybody out there ? */
4046 if (lttng_session_daemon_alive()) {
4047 return -EEXIST;
4048 }
4049
4050 return 0;
4051 }
4052
4053 /*
4054 * Set the tracing group gid onto the client socket.
4055 *
4056 * Race window between mkdir and chown is OK because we are going from more
4057 * permissive (root.root) to less permissive (root.tracing).
4058 */
4059 static int set_permissions(char *rundir)
4060 {
4061 int ret;
4062 gid_t gid;
4063
4064 ret = allowed_group();
4065 if (ret < 0) {
4066 WARN("No tracing group detected");
4067 ret = 0;
4068 goto end;
4069 }
4070
4071 gid = ret;
4072
4073 /* Set lttng run dir */
4074 ret = chown(rundir, 0, gid);
4075 if (ret < 0) {
4076 ERR("Unable to set group on %s", rundir);
4077 PERROR("chown");
4078 }
4079
4080 /* Ensure tracing group can search the run dir */
4081 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
4082 if (ret < 0) {
4083 ERR("Unable to set permissions on %s", rundir);
4084 PERROR("chmod");
4085 }
4086
4087 /* lttng client socket path */
4088 ret = chown(client_unix_sock_path, 0, gid);
4089 if (ret < 0) {
4090 ERR("Unable to set group on %s", client_unix_sock_path);
4091 PERROR("chown");
4092 }
4093
4094 /* kconsumer error socket path */
4095 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4096 if (ret < 0) {
4097 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4098 PERROR("chown");
4099 }
4100
4101 /* 64-bit ustconsumer error socket path */
4102 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4103 if (ret < 0) {
4104 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4105 PERROR("chown");
4106 }
4107
4108 /* 32-bit ustconsumer compat32 error socket path */
4109 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4110 if (ret < 0) {
4111 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4112 PERROR("chown");
4113 }
4114
4115 DBG("All permissions are set");
4116
4117 end:
4118 return ret;
4119 }
4120
4121 /*
4122 * Create the lttng run directory needed for all global sockets and pipe.
4123 */
4124 static int create_lttng_rundir(const char *rundir)
4125 {
4126 int ret;
4127
4128 DBG3("Creating LTTng run directory: %s", rundir);
4129
4130 ret = mkdir(rundir, S_IRWXU);
4131 if (ret < 0) {
4132 if (errno != EEXIST) {
4133 ERR("Unable to create %s", rundir);
4134 goto error;
4135 } else {
4136 ret = 0;
4137 }
4138 }
4139
4140 error:
4141 return ret;
4142 }
4143
4144 /*
4145 * Setup sockets and directory needed by the kconsumerd communication with the
4146 * session daemon.
4147 */
4148 static int set_consumer_sockets(struct consumer_data *consumer_data,
4149 const char *rundir)
4150 {
4151 int ret;
4152 char path[PATH_MAX];
4153
4154 switch (consumer_data->type) {
4155 case LTTNG_CONSUMER_KERNEL:
4156 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4157 break;
4158 case LTTNG_CONSUMER64_UST:
4159 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4160 break;
4161 case LTTNG_CONSUMER32_UST:
4162 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4163 break;
4164 default:
4165 ERR("Consumer type unknown");
4166 ret = -EINVAL;
4167 goto error;
4168 }
4169
4170 DBG2("Creating consumer directory: %s", path);
4171
4172 ret = mkdir(path, S_IRWXU);
4173 if (ret < 0) {
4174 if (errno != EEXIST) {
4175 PERROR("mkdir");
4176 ERR("Failed to create %s", path);
4177 goto error;
4178 }
4179 ret = -1;
4180 }
4181
4182 /* Create the kconsumerd error unix socket */
4183 consumer_data->err_sock =
4184 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4185 if (consumer_data->err_sock < 0) {
4186 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4187 ret = -1;
4188 goto error;
4189 }
4190
4191 /*
4192 * Set the CLOEXEC flag. Return code is useless because either way, the
4193 * show must go on.
4194 */
4195 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4196 if (ret < 0) {
4197 PERROR("utils_set_fd_cloexec");
4198 /* continue anyway */
4199 }
4200
4201 /* File permission MUST be 660 */
4202 ret = chmod(consumer_data->err_unix_sock_path,
4203 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4204 if (ret < 0) {
4205 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4206 PERROR("chmod");
4207 goto error;
4208 }
4209
4210 error:
4211 return ret;
4212 }
4213
4214 /*
4215 * Signal handler for the daemon
4216 *
4217 * Simply stop all worker threads, leaving main() return gracefully after
4218 * joining all threads and calling cleanup().
4219 */
4220 static void sighandler(int sig)
4221 {
4222 switch (sig) {
4223 case SIGPIPE:
4224 DBG("SIGPIPE caught");
4225 return;
4226 case SIGINT:
4227 DBG("SIGINT caught");
4228 stop_threads();
4229 break;
4230 case SIGTERM:
4231 DBG("SIGTERM caught");
4232 stop_threads();
4233 break;
4234 default:
4235 break;
4236 }
4237 }
4238
4239 /*
4240 * Setup signal handler for :
4241 * SIGINT, SIGTERM, SIGPIPE
4242 */
4243 static int set_signal_handler(void)
4244 {
4245 int ret = 0;
4246 struct sigaction sa;
4247 sigset_t sigset;
4248
4249 if ((ret = sigemptyset(&sigset)) < 0) {
4250 PERROR("sigemptyset");
4251 return ret;
4252 }
4253
4254 sa.sa_handler = sighandler;
4255 sa.sa_mask = sigset;
4256 sa.sa_flags = 0;
4257 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4258 PERROR("sigaction");
4259 return ret;
4260 }
4261
4262 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4263 PERROR("sigaction");
4264 return ret;
4265 }
4266
4267 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4268 PERROR("sigaction");
4269 return ret;
4270 }
4271
4272 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4273
4274 return ret;
4275 }
4276
4277 /*
4278 * Set open files limit to unlimited. This daemon can open a large number of
4279 * file descriptors in order to consumer multiple kernel traces.
4280 */
4281 static void set_ulimit(void)
4282 {
4283 int ret;
4284 struct rlimit lim;
4285
4286 /* The kernel does not allowed an infinite limit for open files */
4287 lim.rlim_cur = 65535;
4288 lim.rlim_max = 65535;
4289
4290 ret = setrlimit(RLIMIT_NOFILE, &lim);
4291 if (ret < 0) {
4292 PERROR("failed to set open files limit");
4293 }
4294 }
4295
4296 /*
4297 * Write pidfile using the rundir and opt_pidfile.
4298 */
4299 static void write_pidfile(void)
4300 {
4301 int ret;
4302 char pidfile_path[PATH_MAX];
4303
4304 assert(rundir);
4305
4306 if (opt_pidfile) {
4307 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4308 } else {
4309 /* Build pidfile path from rundir and opt_pidfile. */
4310 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4311 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4312 if (ret < 0) {
4313 PERROR("snprintf pidfile path");
4314 goto error;
4315 }
4316 }
4317
4318 /*
4319 * Create pid file in rundir. Return value is of no importance. The
4320 * execution will continue even though we are not able to write the file.
4321 */
4322 (void) utils_create_pid_file(getpid(), pidfile_path);
4323
4324 error:
4325 return;
4326 }
4327
4328 /*
4329 * main
4330 */
4331 int main(int argc, char **argv)
4332 {
4333 int ret = 0;
4334 void *status;
4335 const char *home_path, *env_app_timeout;
4336
4337 init_kernel_workarounds();
4338
4339 rcu_register_thread();
4340
4341 setup_consumerd_path();
4342
4343 page_size = sysconf(_SC_PAGESIZE);
4344 if (page_size < 0) {
4345 PERROR("sysconf _SC_PAGESIZE");
4346 page_size = LONG_MAX;
4347 WARN("Fallback page size to %ld", page_size);
4348 }
4349
4350 /* Parse arguments */
4351 progname = argv[0];
4352 if ((ret = parse_args(argc, argv)) < 0) {
4353 goto error;
4354 }
4355
4356 /* Daemonize */
4357 if (opt_daemon) {
4358 int i;
4359
4360 /*
4361 * fork
4362 * child: setsid, close FD 0, 1, 2, chdir /
4363 * parent: exit (if fork is successful)
4364 */
4365 ret = daemon(0, 0);
4366 if (ret < 0) {
4367 PERROR("daemon");
4368 goto error;
4369 }
4370 /*
4371 * We are in the child. Make sure all other file
4372 * descriptors are closed, in case we are called with
4373 * more opened file descriptors than the standard ones.
4374 */
4375 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4376 (void) close(i);
4377 }
4378 }
4379
4380 /* Create thread quit pipe */
4381 if ((ret = init_thread_quit_pipe()) < 0) {
4382 goto error;
4383 }
4384
4385 /* Check if daemon is UID = 0 */
4386 is_root = !getuid();
4387
4388 if (is_root) {
4389 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4390
4391 /* Create global run dir with root access */
4392 ret = create_lttng_rundir(rundir);
4393 if (ret < 0) {
4394 goto error;
4395 }
4396
4397 if (strlen(apps_unix_sock_path) == 0) {
4398 snprintf(apps_unix_sock_path, PATH_MAX,
4399 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4400 }
4401
4402 if (strlen(client_unix_sock_path) == 0) {
4403 snprintf(client_unix_sock_path, PATH_MAX,
4404 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4405 }
4406
4407 /* Set global SHM for ust */
4408 if (strlen(wait_shm_path) == 0) {
4409 snprintf(wait_shm_path, PATH_MAX,
4410 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4411 }
4412
4413 if (strlen(health_unix_sock_path) == 0) {
4414 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4415 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4416 }
4417
4418 /* Setup kernel consumerd path */
4419 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4420 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4421 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4422 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4423
4424 DBG2("Kernel consumer err path: %s",
4425 kconsumer_data.err_unix_sock_path);
4426 DBG2("Kernel consumer cmd path: %s",
4427 kconsumer_data.cmd_unix_sock_path);
4428 } else {
4429 home_path = utils_get_home_dir();
4430 if (home_path == NULL) {
4431 /* TODO: Add --socket PATH option */
4432 ERR("Can't get HOME directory for sockets creation.");
4433 ret = -EPERM;
4434 goto error;
4435 }
4436
4437 /*
4438 * Create rundir from home path. This will create something like
4439 * $HOME/.lttng
4440 */
4441 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4442 if (ret < 0) {
4443 ret = -ENOMEM;
4444 goto error;
4445 }
4446
4447 ret = create_lttng_rundir(rundir);
4448 if (ret < 0) {
4449 goto error;
4450 }
4451
4452 if (strlen(apps_unix_sock_path) == 0) {
4453 snprintf(apps_unix_sock_path, PATH_MAX,
4454 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4455 }
4456
4457 /* Set the cli tool unix socket path */
4458 if (strlen(client_unix_sock_path) == 0) {
4459 snprintf(client_unix_sock_path, PATH_MAX,
4460 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4461 }
4462
4463 /* Set global SHM for ust */
4464 if (strlen(wait_shm_path) == 0) {
4465 snprintf(wait_shm_path, PATH_MAX,
4466 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4467 }
4468
4469 /* Set health check Unix path */
4470 if (strlen(health_unix_sock_path) == 0) {
4471 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4472 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4473 }
4474 }
4475
4476 /* Set consumer initial state */
4477 kernel_consumerd_state = CONSUMER_STOPPED;
4478 ust_consumerd_state = CONSUMER_STOPPED;
4479
4480 DBG("Client socket path %s", client_unix_sock_path);
4481 DBG("Application socket path %s", apps_unix_sock_path);
4482 DBG("Application wait path %s", wait_shm_path);
4483 DBG("LTTng run directory path: %s", rundir);
4484
4485 /* 32 bits consumerd path setup */
4486 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4487 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4488 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4489 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4490
4491 DBG2("UST consumer 32 bits err path: %s",
4492 ustconsumer32_data.err_unix_sock_path);
4493 DBG2("UST consumer 32 bits cmd path: %s",
4494 ustconsumer32_data.cmd_unix_sock_path);
4495
4496 /* 64 bits consumerd path setup */
4497 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4498 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4499 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4500 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4501
4502 DBG2("UST consumer 64 bits err path: %s",
4503 ustconsumer64_data.err_unix_sock_path);
4504 DBG2("UST consumer 64 bits cmd path: %s",
4505 ustconsumer64_data.cmd_unix_sock_path);
4506
4507 /*
4508 * See if daemon already exist.
4509 */
4510 if ((ret = check_existing_daemon()) < 0) {
4511 ERR("Already running daemon.\n");
4512 /*
4513 * We do not goto exit because we must not cleanup()
4514 * because a daemon is already running.
4515 */
4516 goto error;
4517 }
4518
4519 /*
4520 * Init UST app hash table. Alloc hash table before this point since
4521 * cleanup() can get called after that point.
4522 */
4523 ust_app_ht_alloc();
4524
4525 /* After this point, we can safely call cleanup() with "goto exit" */
4526
4527 /*
4528 * These actions must be executed as root. We do that *after* setting up
4529 * the sockets path because we MUST make the check for another daemon using
4530 * those paths *before* trying to set the kernel consumer sockets and init
4531 * kernel tracer.
4532 */
4533 if (is_root) {
4534 ret = set_consumer_sockets(&kconsumer_data, rundir);
4535 if (ret < 0) {
4536 goto exit;
4537 }
4538
4539 /* Setup kernel tracer */
4540 if (!opt_no_kernel) {
4541 init_kernel_tracer();
4542 }
4543
4544 /* Set ulimit for open files */
4545 set_ulimit();
4546 }
4547 /* init lttng_fd tracking must be done after set_ulimit. */
4548 lttng_fd_init();
4549
4550 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4551 if (ret < 0) {
4552 goto exit;
4553 }
4554
4555 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4556 if (ret < 0) {
4557 goto exit;
4558 }
4559
4560 if ((ret = set_signal_handler()) < 0) {
4561 goto exit;
4562 }
4563
4564 /* Setup the needed unix socket */
4565 if ((ret = init_daemon_socket()) < 0) {
4566 goto exit;
4567 }
4568
4569 /* Set credentials to socket */
4570 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4571 goto exit;
4572 }
4573
4574 /* Get parent pid if -S, --sig-parent is specified. */
4575 if (opt_sig_parent) {
4576 ppid = getppid();
4577 }
4578
4579 /* Setup the kernel pipe for waking up the kernel thread */
4580 if (is_root && !opt_no_kernel) {
4581 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4582 goto exit;
4583 }
4584 }
4585
4586 /* Setup the thread ht_cleanup communication pipe. */
4587 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4588 goto exit;
4589 }
4590
4591 /* Setup the thread apps communication pipe. */
4592 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4593 goto exit;
4594 }
4595
4596 /* Setup the thread apps notify communication pipe. */
4597 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4598 goto exit;
4599 }
4600
4601 /* Initialize global buffer per UID and PID registry. */
4602 buffer_reg_init_uid_registry();
4603 buffer_reg_init_pid_registry();
4604
4605 /* Init UST command queue. */
4606 cds_wfq_init(&ust_cmd_queue.queue);
4607
4608 /*
4609 * Get session list pointer. This pointer MUST NOT be free(). This list is
4610 * statically declared in session.c
4611 */
4612 session_list_ptr = session_get_list();
4613
4614 /* Set up max poll set size */
4615 lttng_poll_set_max_size();
4616
4617 cmd_init();
4618
4619 /* Check for the application socket timeout env variable. */
4620 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4621 if (env_app_timeout) {
4622 app_socket_timeout = atoi(env_app_timeout);
4623 } else {
4624 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4625 }
4626
4627 write_pidfile();
4628
4629 /* Create thread to manage the client socket */
4630 ret = pthread_create(&ht_cleanup_thread, NULL,
4631 thread_ht_cleanup, (void *) NULL);
4632 if (ret != 0) {
4633 PERROR("pthread_create ht_cleanup");
4634 goto exit_ht_cleanup;
4635 }
4636
4637 /* Create thread to manage the client socket */
4638 ret = pthread_create(&health_thread, NULL,
4639 thread_manage_health, (void *) NULL);
4640 if (ret != 0) {
4641 PERROR("pthread_create health");
4642 goto exit_health;
4643 }
4644
4645 /* Create thread to manage the client socket */
4646 ret = pthread_create(&client_thread, NULL,
4647 thread_manage_clients, (void *) NULL);
4648 if (ret != 0) {
4649 PERROR("pthread_create clients");
4650 goto exit_client;
4651 }
4652
4653 /* Create thread to dispatch registration */
4654 ret = pthread_create(&dispatch_thread, NULL,
4655 thread_dispatch_ust_registration, (void *) NULL);
4656 if (ret != 0) {
4657 PERROR("pthread_create dispatch");
4658 goto exit_dispatch;
4659 }
4660
4661 /* Create thread to manage application registration. */
4662 ret = pthread_create(&reg_apps_thread, NULL,
4663 thread_registration_apps, (void *) NULL);
4664 if (ret != 0) {
4665 PERROR("pthread_create registration");
4666 goto exit_reg_apps;
4667 }
4668
4669 /* Create thread to manage application socket */
4670 ret = pthread_create(&apps_thread, NULL,
4671 thread_manage_apps, (void *) NULL);
4672 if (ret != 0) {
4673 PERROR("pthread_create apps");
4674 goto exit_apps;
4675 }
4676
4677 /* Create thread to manage application notify socket */
4678 ret = pthread_create(&apps_notify_thread, NULL,
4679 ust_thread_manage_notify, (void *) NULL);
4680 if (ret != 0) {
4681 PERROR("pthread_create apps");
4682 goto exit_apps;
4683 }
4684
4685 /* Don't start this thread if kernel tracing is not requested nor root */
4686 if (is_root && !opt_no_kernel) {
4687 /* Create kernel thread to manage kernel event */
4688 ret = pthread_create(&kernel_thread, NULL,
4689 thread_manage_kernel, (void *) NULL);
4690 if (ret != 0) {
4691 PERROR("pthread_create kernel");
4692 goto exit_kernel;
4693 }
4694
4695 ret = pthread_join(kernel_thread, &status);
4696 if (ret != 0) {
4697 PERROR("pthread_join");
4698 goto error; /* join error, exit without cleanup */
4699 }
4700 }
4701
4702 exit_kernel:
4703 ret = pthread_join(apps_thread, &status);
4704 if (ret != 0) {
4705 PERROR("pthread_join");
4706 goto error; /* join error, exit without cleanup */
4707 }
4708
4709 exit_apps:
4710 ret = pthread_join(reg_apps_thread, &status);
4711 if (ret != 0) {
4712 PERROR("pthread_join");
4713 goto error; /* join error, exit without cleanup */
4714 }
4715
4716 exit_reg_apps:
4717 ret = pthread_join(dispatch_thread, &status);
4718 if (ret != 0) {
4719 PERROR("pthread_join");
4720 goto error; /* join error, exit without cleanup */
4721 }
4722
4723 exit_dispatch:
4724 ret = pthread_join(client_thread, &status);
4725 if (ret != 0) {
4726 PERROR("pthread_join");
4727 goto error; /* join error, exit without cleanup */
4728 }
4729
4730 ret = join_consumer_thread(&kconsumer_data);
4731 if (ret != 0) {
4732 PERROR("join_consumer");
4733 goto error; /* join error, exit without cleanup */
4734 }
4735
4736 ret = join_consumer_thread(&ustconsumer32_data);
4737 if (ret != 0) {
4738 PERROR("join_consumer ust32");
4739 goto error; /* join error, exit without cleanup */
4740 }
4741
4742 ret = join_consumer_thread(&ustconsumer64_data);
4743 if (ret != 0) {
4744 PERROR("join_consumer ust64");
4745 goto error; /* join error, exit without cleanup */
4746 }
4747
4748 exit_client:
4749 ret = pthread_join(health_thread, &status);
4750 if (ret != 0) {
4751 PERROR("pthread_join health thread");
4752 goto error; /* join error, exit without cleanup */
4753 }
4754
4755 exit_health:
4756 ret = pthread_join(ht_cleanup_thread, &status);
4757 if (ret != 0) {
4758 PERROR("pthread_join ht cleanup thread");
4759 goto error; /* join error, exit without cleanup */
4760 }
4761 exit_ht_cleanup:
4762 exit:
4763 /*
4764 * cleanup() is called when no other thread is running.
4765 */
4766 rcu_thread_online();
4767 cleanup();
4768 rcu_thread_offline();
4769 rcu_unregister_thread();
4770 if (!ret) {
4771 exit(EXIT_SUCCESS);
4772 }
4773 error:
4774 exit(EXIT_FAILURE);
4775 }
This page took 0.138541 seconds and 4 git commands to generate.