Fix: don't skip chmod if tracing group is not found
[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 /*
450 * Close the thread quit pipe. It has already done its job,
451 * since we are now called.
452 */
453 utils_close_pipe(thread_quit_pipe);
454
455 /*
456 * If opt_pidfile is undefined, the default file will be wiped when
457 * removing the rundir.
458 */
459 if (opt_pidfile) {
460 ret = remove(opt_pidfile);
461 if (ret < 0) {
462 PERROR("remove pidfile %s", opt_pidfile);
463 }
464 }
465
466 DBG("Removing %s directory", rundir);
467 ret = asprintf(&cmd, "rm -rf %s", rundir);
468 if (ret < 0) {
469 ERR("asprintf failed. Something is really wrong!");
470 }
471
472 /* Remove lttng run directory */
473 ret = system(cmd);
474 if (ret < 0) {
475 ERR("Unable to clean %s", rundir);
476 }
477 free(cmd);
478 free(rundir);
479
480 DBG("Cleaning up all sessions");
481
482 /* Destroy session list mutex */
483 if (session_list_ptr != NULL) {
484 pthread_mutex_destroy(&session_list_ptr->lock);
485
486 /* Cleanup ALL session */
487 cds_list_for_each_entry_safe(sess, stmp,
488 &session_list_ptr->head, list) {
489 cmd_destroy_session(sess, kernel_poll_pipe[1]);
490 }
491 }
492
493 DBG("Closing all UST sockets");
494 ust_app_clean_list();
495 buffer_reg_destroy_registries();
496
497 if (is_root && !opt_no_kernel) {
498 DBG2("Closing kernel fd");
499 if (kernel_tracer_fd >= 0) {
500 ret = close(kernel_tracer_fd);
501 if (ret) {
502 PERROR("close");
503 }
504 }
505 DBG("Unloading kernel modules");
506 modprobe_remove_lttng_all();
507 }
508
509 close_consumer_sockets();
510
511 /* <fun> */
512 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
513 "Matthew, BEET driven development works!%c[%dm",
514 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
515 /* </fun> */
516 }
517
518 /*
519 * Send data on a unix socket using the liblttsessiondcomm API.
520 *
521 * Return lttcomm error code.
522 */
523 static int send_unix_sock(int sock, void *buf, size_t len)
524 {
525 /* Check valid length */
526 if (len == 0) {
527 return -1;
528 }
529
530 return lttcomm_send_unix_sock(sock, buf, len);
531 }
532
533 /*
534 * Free memory of a command context structure.
535 */
536 static void clean_command_ctx(struct command_ctx **cmd_ctx)
537 {
538 DBG("Clean command context structure");
539 if (*cmd_ctx) {
540 if ((*cmd_ctx)->llm) {
541 free((*cmd_ctx)->llm);
542 }
543 if ((*cmd_ctx)->lsm) {
544 free((*cmd_ctx)->lsm);
545 }
546 free(*cmd_ctx);
547 *cmd_ctx = NULL;
548 }
549 }
550
551 /*
552 * Notify UST applications using the shm mmap futex.
553 */
554 static int notify_ust_apps(int active)
555 {
556 char *wait_shm_mmap;
557
558 DBG("Notifying applications of session daemon state: %d", active);
559
560 /* See shm.c for this call implying mmap, shm and futex calls */
561 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
562 if (wait_shm_mmap == NULL) {
563 goto error;
564 }
565
566 /* Wake waiting process */
567 futex_wait_update((int32_t *) wait_shm_mmap, active);
568
569 /* Apps notified successfully */
570 return 0;
571
572 error:
573 return -1;
574 }
575
576 /*
577 * Setup the outgoing data buffer for the response (llm) by allocating the
578 * right amount of memory and copying the original information from the lsm
579 * structure.
580 *
581 * Return total size of the buffer pointed by buf.
582 */
583 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
584 {
585 int ret, buf_size;
586
587 buf_size = size;
588
589 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
590 if (cmd_ctx->llm == NULL) {
591 PERROR("zmalloc");
592 ret = -ENOMEM;
593 goto error;
594 }
595
596 /* Copy common data */
597 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
598 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
599
600 cmd_ctx->llm->data_size = size;
601 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
602
603 return buf_size;
604
605 error:
606 return ret;
607 }
608
609 /*
610 * Update the kernel poll set of all channel fd available over all tracing
611 * session. Add the wakeup pipe at the end of the set.
612 */
613 static int update_kernel_poll(struct lttng_poll_event *events)
614 {
615 int ret;
616 struct ltt_session *session;
617 struct ltt_kernel_channel *channel;
618
619 DBG("Updating kernel poll set");
620
621 session_lock_list();
622 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
623 session_lock(session);
624 if (session->kernel_session == NULL) {
625 session_unlock(session);
626 continue;
627 }
628
629 cds_list_for_each_entry(channel,
630 &session->kernel_session->channel_list.head, list) {
631 /* Add channel fd to the kernel poll set */
632 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
633 if (ret < 0) {
634 session_unlock(session);
635 goto error;
636 }
637 DBG("Channel fd %d added to kernel set", channel->fd);
638 }
639 session_unlock(session);
640 }
641 session_unlock_list();
642
643 return 0;
644
645 error:
646 session_unlock_list();
647 return -1;
648 }
649
650 /*
651 * Find the channel fd from 'fd' over all tracing session. When found, check
652 * for new channel stream and send those stream fds to the kernel consumer.
653 *
654 * Useful for CPU hotplug feature.
655 */
656 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
657 {
658 int ret = 0;
659 struct ltt_session *session;
660 struct ltt_kernel_session *ksess;
661 struct ltt_kernel_channel *channel;
662
663 DBG("Updating kernel streams for channel fd %d", fd);
664
665 session_lock_list();
666 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
667 session_lock(session);
668 if (session->kernel_session == NULL) {
669 session_unlock(session);
670 continue;
671 }
672 ksess = session->kernel_session;
673
674 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
675 if (channel->fd == fd) {
676 DBG("Channel found, updating kernel streams");
677 ret = kernel_open_channel_stream(channel);
678 if (ret < 0) {
679 goto error;
680 }
681 /* Update the stream global counter */
682 ksess->stream_count_global += ret;
683
684 /*
685 * Have we already sent fds to the consumer? If yes, it means
686 * that tracing is started so it is safe to send our updated
687 * stream fds.
688 */
689 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
690 struct lttng_ht_iter iter;
691 struct consumer_socket *socket;
692
693 rcu_read_lock();
694 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
695 &iter.iter, socket, node.node) {
696 /* Code flow error */
697 assert(socket->fd >= 0);
698
699 pthread_mutex_lock(socket->lock);
700 ret = kernel_consumer_send_channel_stream(socket,
701 channel, ksess,
702 session->output_traces ? 1 : 0);
703 pthread_mutex_unlock(socket->lock);
704 if (ret < 0) {
705 rcu_read_unlock();
706 goto error;
707 }
708 }
709 rcu_read_unlock();
710 }
711 goto error;
712 }
713 }
714 session_unlock(session);
715 }
716 session_unlock_list();
717 return ret;
718
719 error:
720 session_unlock(session);
721 session_unlock_list();
722 return ret;
723 }
724
725 /*
726 * For each tracing session, update newly registered apps. The session list
727 * lock MUST be acquired before calling this.
728 */
729 static void update_ust_app(int app_sock)
730 {
731 struct ltt_session *sess, *stmp;
732
733 /* For all tracing session(s) */
734 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
735 session_lock(sess);
736 if (sess->ust_session) {
737 ust_app_global_update(sess->ust_session, app_sock);
738 }
739 session_unlock(sess);
740 }
741 }
742
743 /*
744 * This thread manage event coming from the kernel.
745 *
746 * Features supported in this thread:
747 * -) CPU Hotplug
748 */
749 static void *thread_manage_kernel(void *data)
750 {
751 int ret, i, pollfd, update_poll_flag = 1, err = -1;
752 uint32_t revents, nb_fd;
753 char tmp;
754 struct lttng_poll_event events;
755
756 DBG("[thread] Thread manage kernel started");
757
758 health_register(HEALTH_TYPE_KERNEL);
759
760 /*
761 * This first step of the while is to clean this structure which could free
762 * non NULL pointers so initialize it before the loop.
763 */
764 lttng_poll_init(&events);
765
766 if (testpoint(thread_manage_kernel)) {
767 goto error_testpoint;
768 }
769
770 health_code_update();
771
772 if (testpoint(thread_manage_kernel_before_loop)) {
773 goto error_testpoint;
774 }
775
776 while (1) {
777 health_code_update();
778
779 if (update_poll_flag == 1) {
780 /* Clean events object. We are about to populate it again. */
781 lttng_poll_clean(&events);
782
783 ret = sessiond_set_thread_pollset(&events, 2);
784 if (ret < 0) {
785 goto error_poll_create;
786 }
787
788 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
789 if (ret < 0) {
790 goto error;
791 }
792
793 /* This will add the available kernel channel if any. */
794 ret = update_kernel_poll(&events);
795 if (ret < 0) {
796 goto error;
797 }
798 update_poll_flag = 0;
799 }
800
801 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
802
803 /* Poll infinite value of time */
804 restart:
805 health_poll_entry();
806 ret = lttng_poll_wait(&events, -1);
807 health_poll_exit();
808 if (ret < 0) {
809 /*
810 * Restart interrupted system call.
811 */
812 if (errno == EINTR) {
813 goto restart;
814 }
815 goto error;
816 } else if (ret == 0) {
817 /* Should not happen since timeout is infinite */
818 ERR("Return value of poll is 0 with an infinite timeout.\n"
819 "This should not have happened! Continuing...");
820 continue;
821 }
822
823 nb_fd = ret;
824
825 for (i = 0; i < nb_fd; i++) {
826 /* Fetch once the poll data */
827 revents = LTTNG_POLL_GETEV(&events, i);
828 pollfd = LTTNG_POLL_GETFD(&events, i);
829
830 health_code_update();
831
832 /* Thread quit pipe has been closed. Killing thread. */
833 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
834 if (ret) {
835 err = 0;
836 goto exit;
837 }
838
839 /* Check for data on kernel pipe */
840 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
841 do {
842 ret = read(kernel_poll_pipe[0], &tmp, 1);
843 } while (ret < 0 && errno == EINTR);
844 /*
845 * Ret value is useless here, if this pipe gets any actions an
846 * update is required anyway.
847 */
848 update_poll_flag = 1;
849 continue;
850 } else {
851 /*
852 * New CPU detected by the kernel. Adding kernel stream to
853 * kernel session and updating the kernel consumer
854 */
855 if (revents & LPOLLIN) {
856 ret = update_kernel_stream(&kconsumer_data, pollfd);
857 if (ret < 0) {
858 continue;
859 }
860 break;
861 /*
862 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
863 * and unregister kernel stream at this point.
864 */
865 }
866 }
867 }
868 }
869
870 exit:
871 error:
872 lttng_poll_clean(&events);
873 error_poll_create:
874 error_testpoint:
875 utils_close_pipe(kernel_poll_pipe);
876 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
877 if (err) {
878 health_error();
879 ERR("Health error occurred in %s", __func__);
880 WARN("Kernel thread died unexpectedly. "
881 "Kernel tracing can continue but CPU hotplug is disabled.");
882 }
883 health_unregister();
884 DBG("Kernel thread dying");
885 return NULL;
886 }
887
888 /*
889 * Signal pthread condition of the consumer data that the thread.
890 */
891 static void signal_consumer_condition(struct consumer_data *data, int state)
892 {
893 pthread_mutex_lock(&data->cond_mutex);
894
895 /*
896 * The state is set before signaling. It can be any value, it's the waiter
897 * job to correctly interpret this condition variable associated to the
898 * consumer pthread_cond.
899 *
900 * A value of 0 means that the corresponding thread of the consumer data
901 * was not started. 1 indicates that the thread has started and is ready
902 * for action. A negative value means that there was an error during the
903 * thread bootstrap.
904 */
905 data->consumer_thread_is_ready = state;
906 (void) pthread_cond_signal(&data->cond);
907
908 pthread_mutex_unlock(&data->cond_mutex);
909 }
910
911 /*
912 * This thread manage the consumer error sent back to the session daemon.
913 */
914 static void *thread_manage_consumer(void *data)
915 {
916 int sock = -1, i, ret, pollfd, err = -1;
917 uint32_t revents, nb_fd;
918 enum lttcomm_return_code code;
919 struct lttng_poll_event events;
920 struct consumer_data *consumer_data = data;
921
922 DBG("[thread] Manage consumer started");
923
924 health_register(HEALTH_TYPE_CONSUMER);
925
926 health_code_update();
927
928 /*
929 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
930 * metadata_sock. Nothing more will be added to this poll set.
931 */
932 ret = sessiond_set_thread_pollset(&events, 3);
933 if (ret < 0) {
934 goto error_poll;
935 }
936
937 /*
938 * The error socket here is already in a listening state which was done
939 * just before spawning this thread to avoid a race between the consumer
940 * daemon exec trying to connect and the listen() call.
941 */
942 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
943 if (ret < 0) {
944 goto error;
945 }
946
947 health_code_update();
948
949 /* Infinite blocking call, waiting for transmission */
950 restart:
951 health_poll_entry();
952
953 if (testpoint(thread_manage_consumer)) {
954 goto error;
955 }
956
957 ret = lttng_poll_wait(&events, -1);
958 health_poll_exit();
959 if (ret < 0) {
960 /*
961 * Restart interrupted system call.
962 */
963 if (errno == EINTR) {
964 goto restart;
965 }
966 goto error;
967 }
968
969 nb_fd = ret;
970
971 for (i = 0; i < nb_fd; i++) {
972 /* Fetch once the poll data */
973 revents = LTTNG_POLL_GETEV(&events, i);
974 pollfd = LTTNG_POLL_GETFD(&events, i);
975
976 health_code_update();
977
978 /* Thread quit pipe has been closed. Killing thread. */
979 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
980 if (ret) {
981 err = 0;
982 goto exit;
983 }
984
985 /* Event on the registration socket */
986 if (pollfd == consumer_data->err_sock) {
987 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
988 ERR("consumer err socket poll error");
989 goto error;
990 }
991 }
992 }
993
994 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
995 if (sock < 0) {
996 goto error;
997 }
998
999 /*
1000 * Set the CLOEXEC flag. Return code is useless because either way, the
1001 * show must go on.
1002 */
1003 (void) utils_set_fd_cloexec(sock);
1004
1005 health_code_update();
1006
1007 DBG2("Receiving code from consumer err_sock");
1008
1009 /* Getting status code from kconsumerd */
1010 ret = lttcomm_recv_unix_sock(sock, &code,
1011 sizeof(enum lttcomm_return_code));
1012 if (ret <= 0) {
1013 goto error;
1014 }
1015
1016 health_code_update();
1017
1018 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1019 /* Connect both socket, command and metadata. */
1020 consumer_data->cmd_sock =
1021 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1022 consumer_data->metadata_sock.fd =
1023 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1024 if (consumer_data->cmd_sock < 0 ||
1025 consumer_data->metadata_sock.fd < 0) {
1026 PERROR("consumer connect cmd socket");
1027 /* On error, signal condition and quit. */
1028 signal_consumer_condition(consumer_data, -1);
1029 goto error;
1030 }
1031 /* Create metadata socket lock. */
1032 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1033 if (consumer_data->metadata_sock.lock == NULL) {
1034 PERROR("zmalloc pthread mutex");
1035 ret = -1;
1036 goto error;
1037 }
1038 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1039
1040 signal_consumer_condition(consumer_data, 1);
1041 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1042 DBG("Consumer metadata socket ready (fd: %d)",
1043 consumer_data->metadata_sock.fd);
1044 } else {
1045 ERR("consumer error when waiting for SOCK_READY : %s",
1046 lttcomm_get_readable_code(-code));
1047 goto error;
1048 }
1049
1050 /* Remove the consumerd error sock since we've established a connexion */
1051 ret = lttng_poll_del(&events, consumer_data->err_sock);
1052 if (ret < 0) {
1053 goto error;
1054 }
1055
1056 /* Add new accepted error socket. */
1057 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1058 if (ret < 0) {
1059 goto error;
1060 }
1061
1062 /* Add metadata socket that is successfully connected. */
1063 ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
1064 LPOLLIN | LPOLLRDHUP);
1065 if (ret < 0) {
1066 goto error;
1067 }
1068
1069 health_code_update();
1070
1071 /* Infinite blocking call, waiting for transmission */
1072 restart_poll:
1073 while (1) {
1074 health_poll_entry();
1075 ret = lttng_poll_wait(&events, -1);
1076 health_poll_exit();
1077 if (ret < 0) {
1078 /*
1079 * Restart interrupted system call.
1080 */
1081 if (errno == EINTR) {
1082 goto restart_poll;
1083 }
1084 goto error;
1085 }
1086
1087 nb_fd = ret;
1088
1089 for (i = 0; i < nb_fd; i++) {
1090 /* Fetch once the poll data */
1091 revents = LTTNG_POLL_GETEV(&events, i);
1092 pollfd = LTTNG_POLL_GETFD(&events, i);
1093
1094 health_code_update();
1095
1096 /* Thread quit pipe has been closed. Killing thread. */
1097 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1098 if (ret) {
1099 err = 0;
1100 goto exit;
1101 }
1102
1103 if (pollfd == sock) {
1104 /* Event on the consumerd socket */
1105 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1106 ERR("consumer err socket second poll error");
1107 goto error;
1108 }
1109 health_code_update();
1110 /* Wait for any kconsumerd error */
1111 ret = lttcomm_recv_unix_sock(sock, &code,
1112 sizeof(enum lttcomm_return_code));
1113 if (ret <= 0) {
1114 ERR("consumer closed the command socket");
1115 goto error;
1116 }
1117
1118 ERR("consumer return code : %s",
1119 lttcomm_get_readable_code(-code));
1120
1121 goto exit;
1122 } else if (pollfd == consumer_data->metadata_sock.fd) {
1123 /* UST metadata requests */
1124 ret = ust_consumer_metadata_request(
1125 &consumer_data->metadata_sock);
1126 if (ret < 0) {
1127 ERR("Handling metadata request");
1128 goto error;
1129 }
1130 break;
1131 } else {
1132 ERR("Unknown pollfd");
1133 goto error;
1134 }
1135 }
1136 health_code_update();
1137 }
1138
1139 exit:
1140 error:
1141 /* Immediately set the consumerd state to stopped */
1142 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1143 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1144 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1145 consumer_data->type == LTTNG_CONSUMER32_UST) {
1146 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1147 } else {
1148 /* Code flow error... */
1149 assert(0);
1150 }
1151
1152 if (consumer_data->err_sock >= 0) {
1153 ret = close(consumer_data->err_sock);
1154 if (ret) {
1155 PERROR("close");
1156 }
1157 consumer_data->err_sock = -1;
1158 }
1159 if (consumer_data->cmd_sock >= 0) {
1160 ret = close(consumer_data->cmd_sock);
1161 if (ret) {
1162 PERROR("close");
1163 }
1164 consumer_data->cmd_sock = -1;
1165 }
1166 if (consumer_data->metadata_sock.fd >= 0) {
1167 ret = close(consumer_data->metadata_sock.fd);
1168 if (ret) {
1169 PERROR("close");
1170 }
1171 }
1172 /* Cleanup metadata socket mutex. */
1173 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1174 free(consumer_data->metadata_sock.lock);
1175
1176 if (sock >= 0) {
1177 ret = close(sock);
1178 if (ret) {
1179 PERROR("close");
1180 }
1181 }
1182
1183 unlink(consumer_data->err_unix_sock_path);
1184 unlink(consumer_data->cmd_unix_sock_path);
1185 consumer_data->pid = 0;
1186
1187 lttng_poll_clean(&events);
1188 error_poll:
1189 if (err) {
1190 health_error();
1191 ERR("Health error occurred in %s", __func__);
1192 }
1193 health_unregister();
1194 DBG("consumer thread cleanup completed");
1195
1196 return NULL;
1197 }
1198
1199 /*
1200 * This thread manage application communication.
1201 */
1202 static void *thread_manage_apps(void *data)
1203 {
1204 int i, ret, pollfd, err = -1;
1205 uint32_t revents, nb_fd;
1206 struct lttng_poll_event events;
1207
1208 DBG("[thread] Manage application started");
1209
1210 rcu_register_thread();
1211 rcu_thread_online();
1212
1213 health_register(HEALTH_TYPE_APP_MANAGE);
1214
1215 if (testpoint(thread_manage_apps)) {
1216 goto error_testpoint;
1217 }
1218
1219 health_code_update();
1220
1221 ret = sessiond_set_thread_pollset(&events, 2);
1222 if (ret < 0) {
1223 goto error_poll_create;
1224 }
1225
1226 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1227 if (ret < 0) {
1228 goto error;
1229 }
1230
1231 if (testpoint(thread_manage_apps_before_loop)) {
1232 goto error;
1233 }
1234
1235 health_code_update();
1236
1237 while (1) {
1238 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1239
1240 /* Inifinite blocking call, waiting for transmission */
1241 restart:
1242 health_poll_entry();
1243 ret = lttng_poll_wait(&events, -1);
1244 health_poll_exit();
1245 if (ret < 0) {
1246 /*
1247 * Restart interrupted system call.
1248 */
1249 if (errno == EINTR) {
1250 goto restart;
1251 }
1252 goto error;
1253 }
1254
1255 nb_fd = ret;
1256
1257 for (i = 0; i < nb_fd; i++) {
1258 /* Fetch once the poll data */
1259 revents = LTTNG_POLL_GETEV(&events, i);
1260 pollfd = LTTNG_POLL_GETFD(&events, i);
1261
1262 health_code_update();
1263
1264 /* Thread quit pipe has been closed. Killing thread. */
1265 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1266 if (ret) {
1267 err = 0;
1268 goto exit;
1269 }
1270
1271 /* Inspect the apps cmd pipe */
1272 if (pollfd == apps_cmd_pipe[0]) {
1273 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1274 ERR("Apps command pipe error");
1275 goto error;
1276 } else if (revents & LPOLLIN) {
1277 int sock;
1278
1279 /* Empty pipe */
1280 do {
1281 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1282 } while (ret < 0 && errno == EINTR);
1283 if (ret < 0 || ret < sizeof(sock)) {
1284 PERROR("read apps cmd pipe");
1285 goto error;
1286 }
1287
1288 health_code_update();
1289
1290 /*
1291 * We only monitor the error events of the socket. This
1292 * thread does not handle any incoming data from UST
1293 * (POLLIN).
1294 */
1295 ret = lttng_poll_add(&events, sock,
1296 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1297 if (ret < 0) {
1298 goto error;
1299 }
1300
1301 /*
1302 * Set socket timeout for both receiving and ending.
1303 * app_socket_timeout is in seconds, whereas
1304 * lttcomm_setsockopt_rcv_timeout and
1305 * lttcomm_setsockopt_snd_timeout expect msec as
1306 * parameter.
1307 */
1308 (void) lttcomm_setsockopt_rcv_timeout(sock,
1309 app_socket_timeout * 1000);
1310 (void) lttcomm_setsockopt_snd_timeout(sock,
1311 app_socket_timeout * 1000);
1312
1313 DBG("Apps with sock %d added to poll set", sock);
1314
1315 health_code_update();
1316
1317 break;
1318 }
1319 } else {
1320 /*
1321 * At this point, we know that a registered application made
1322 * the event at poll_wait.
1323 */
1324 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1325 /* Removing from the poll set */
1326 ret = lttng_poll_del(&events, pollfd);
1327 if (ret < 0) {
1328 goto error;
1329 }
1330
1331 /* Socket closed on remote end. */
1332 ust_app_unregister(pollfd);
1333 break;
1334 }
1335 }
1336
1337 health_code_update();
1338 }
1339 }
1340
1341 exit:
1342 error:
1343 lttng_poll_clean(&events);
1344 error_poll_create:
1345 error_testpoint:
1346 utils_close_pipe(apps_cmd_pipe);
1347 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1348
1349 /*
1350 * We don't clean the UST app hash table here since already registered
1351 * applications can still be controlled so let them be until the session
1352 * daemon dies or the applications stop.
1353 */
1354
1355 if (err) {
1356 health_error();
1357 ERR("Health error occurred in %s", __func__);
1358 }
1359 health_unregister();
1360 DBG("Application communication apps thread cleanup complete");
1361 rcu_thread_offline();
1362 rcu_unregister_thread();
1363 return NULL;
1364 }
1365
1366 /*
1367 * Send a socket to a thread This is called from the dispatch UST registration
1368 * thread once all sockets are set for the application.
1369 *
1370 * On success, return 0 else a negative value being the errno message of the
1371 * write().
1372 */
1373 static int send_socket_to_thread(int fd, int sock)
1374 {
1375 int ret;
1376
1377 /* Sockets MUST be set or else this should not have been called. */
1378 assert(fd >= 0);
1379 assert(sock >= 0);
1380
1381 do {
1382 ret = write(fd, &sock, sizeof(sock));
1383 } while (ret < 0 && errno == EINTR);
1384 if (ret < 0 || ret != sizeof(sock)) {
1385 PERROR("write apps pipe %d", fd);
1386 if (ret < 0) {
1387 ret = -errno;
1388 }
1389 goto error;
1390 }
1391
1392 /* All good. Don't send back the write positive ret value. */
1393 ret = 0;
1394 error:
1395 return ret;
1396 }
1397
1398 /*
1399 * Sanitize the wait queue of the dispatch registration thread meaning removing
1400 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1401 * notify socket is never received.
1402 */
1403 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1404 {
1405 int ret, nb_fd = 0, i;
1406 unsigned int fd_added = 0;
1407 struct lttng_poll_event events;
1408 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1409
1410 assert(wait_queue);
1411
1412 lttng_poll_init(&events);
1413
1414 /* Just skip everything for an empty queue. */
1415 if (!wait_queue->count) {
1416 goto end;
1417 }
1418
1419 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1420 if (ret < 0) {
1421 goto error_create;
1422 }
1423
1424 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1425 &wait_queue->head, head) {
1426 assert(wait_node->app);
1427 ret = lttng_poll_add(&events, wait_node->app->sock,
1428 LPOLLHUP | LPOLLERR);
1429 if (ret < 0) {
1430 goto error;
1431 }
1432
1433 fd_added = 1;
1434 }
1435
1436 if (!fd_added) {
1437 goto end;
1438 }
1439
1440 /*
1441 * Poll but don't block so we can quickly identify the faulty events and
1442 * clean them afterwards from the wait queue.
1443 */
1444 ret = lttng_poll_wait(&events, 0);
1445 if (ret < 0) {
1446 goto error;
1447 }
1448 nb_fd = ret;
1449
1450 for (i = 0; i < nb_fd; i++) {
1451 /* Get faulty FD. */
1452 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1453 int pollfd = LTTNG_POLL_GETFD(&events, i);
1454
1455 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1456 &wait_queue->head, head) {
1457 if (pollfd == wait_node->app->sock &&
1458 (revents & (LPOLLHUP | LPOLLERR))) {
1459 cds_list_del(&wait_node->head);
1460 wait_queue->count--;
1461 ust_app_destroy(wait_node->app);
1462 free(wait_node);
1463 break;
1464 }
1465 }
1466 }
1467
1468 if (nb_fd > 0) {
1469 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1470 }
1471
1472 end:
1473 lttng_poll_clean(&events);
1474 return;
1475
1476 error:
1477 lttng_poll_clean(&events);
1478 error_create:
1479 ERR("Unable to sanitize wait queue");
1480 return;
1481 }
1482
1483 /*
1484 * Dispatch request from the registration threads to the application
1485 * communication thread.
1486 */
1487 static void *thread_dispatch_ust_registration(void *data)
1488 {
1489 int ret, err = -1;
1490 struct cds_wfq_node *node;
1491 struct ust_command *ust_cmd = NULL;
1492 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1493 struct ust_reg_wait_queue wait_queue = {
1494 .count = 0,
1495 };
1496
1497 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1498
1499 health_code_update();
1500
1501 CDS_INIT_LIST_HEAD(&wait_queue.head);
1502
1503 DBG("[thread] Dispatch UST command started");
1504
1505 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1506 health_code_update();
1507
1508 /* Atomically prepare the queue futex */
1509 futex_nto1_prepare(&ust_cmd_queue.futex);
1510
1511 do {
1512 struct ust_app *app = NULL;
1513 ust_cmd = NULL;
1514
1515 /*
1516 * Make sure we don't have node(s) that have hung up before receiving
1517 * the notify socket. This is to clean the list in order to avoid
1518 * memory leaks from notify socket that are never seen.
1519 */
1520 sanitize_wait_queue(&wait_queue);
1521
1522 health_code_update();
1523 /* Dequeue command for registration */
1524 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1525 if (node == NULL) {
1526 DBG("Woken up but nothing in the UST command queue");
1527 /* Continue thread execution */
1528 break;
1529 }
1530
1531 ust_cmd = caa_container_of(node, struct ust_command, node);
1532
1533 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1534 " gid:%d sock:%d name:%s (version %d.%d)",
1535 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1536 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1537 ust_cmd->sock, ust_cmd->reg_msg.name,
1538 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1539
1540 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1541 wait_node = zmalloc(sizeof(*wait_node));
1542 if (!wait_node) {
1543 PERROR("zmalloc wait_node dispatch");
1544 ret = close(ust_cmd->sock);
1545 if (ret < 0) {
1546 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1547 }
1548 lttng_fd_put(1, LTTNG_FD_APPS);
1549 free(ust_cmd);
1550 goto error;
1551 }
1552 CDS_INIT_LIST_HEAD(&wait_node->head);
1553
1554 /* Create application object if socket is CMD. */
1555 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1556 ust_cmd->sock);
1557 if (!wait_node->app) {
1558 ret = close(ust_cmd->sock);
1559 if (ret < 0) {
1560 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1561 }
1562 lttng_fd_put(1, LTTNG_FD_APPS);
1563 free(wait_node);
1564 free(ust_cmd);
1565 continue;
1566 }
1567 /*
1568 * Add application to the wait queue so we can set the notify
1569 * socket before putting this object in the global ht.
1570 */
1571 cds_list_add(&wait_node->head, &wait_queue.head);
1572 wait_queue.count++;
1573
1574 free(ust_cmd);
1575 /*
1576 * We have to continue here since we don't have the notify
1577 * socket and the application MUST be added to the hash table
1578 * only at that moment.
1579 */
1580 continue;
1581 } else {
1582 /*
1583 * Look for the application in the local wait queue and set the
1584 * notify socket if found.
1585 */
1586 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1587 &wait_queue.head, head) {
1588 health_code_update();
1589 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1590 wait_node->app->notify_sock = ust_cmd->sock;
1591 cds_list_del(&wait_node->head);
1592 wait_queue.count--;
1593 app = wait_node->app;
1594 free(wait_node);
1595 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1596 break;
1597 }
1598 }
1599
1600 /*
1601 * With no application at this stage the received socket is
1602 * basically useless so close it before we free the cmd data
1603 * structure for good.
1604 */
1605 if (!app) {
1606 ret = close(ust_cmd->sock);
1607 if (ret < 0) {
1608 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1609 }
1610 lttng_fd_put(1, LTTNG_FD_APPS);
1611 }
1612 free(ust_cmd);
1613 }
1614
1615 if (app) {
1616 /*
1617 * @session_lock_list
1618 *
1619 * Lock the global session list so from the register up to the
1620 * registration done message, no thread can see the application
1621 * and change its state.
1622 */
1623 session_lock_list();
1624 rcu_read_lock();
1625
1626 /*
1627 * Add application to the global hash table. This needs to be
1628 * done before the update to the UST registry can locate the
1629 * application.
1630 */
1631 ust_app_add(app);
1632
1633 /* Set app version. This call will print an error if needed. */
1634 (void) ust_app_version(app);
1635
1636 /* Send notify socket through the notify pipe. */
1637 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1638 app->notify_sock);
1639 if (ret < 0) {
1640 rcu_read_unlock();
1641 session_unlock_list();
1642 /* No notify thread, stop the UST tracing. */
1643 goto error;
1644 }
1645
1646 /*
1647 * Update newly registered application with the tracing
1648 * registry info already enabled information.
1649 */
1650 update_ust_app(app->sock);
1651
1652 /*
1653 * Don't care about return value. Let the manage apps threads
1654 * handle app unregistration upon socket close.
1655 */
1656 (void) ust_app_register_done(app->sock);
1657
1658 /*
1659 * Even if the application socket has been closed, send the app
1660 * to the thread and unregistration will take place at that
1661 * place.
1662 */
1663 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1664 if (ret < 0) {
1665 rcu_read_unlock();
1666 session_unlock_list();
1667 /* No apps. thread, stop the UST tracing. */
1668 goto error;
1669 }
1670
1671 rcu_read_unlock();
1672 session_unlock_list();
1673 }
1674 } while (node != NULL);
1675
1676 health_poll_entry();
1677 /* Futex wait on queue. Blocking call on futex() */
1678 futex_nto1_wait(&ust_cmd_queue.futex);
1679 health_poll_exit();
1680 }
1681 /* Normal exit, no error */
1682 err = 0;
1683
1684 error:
1685 /* Clean up wait queue. */
1686 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1687 &wait_queue.head, head) {
1688 cds_list_del(&wait_node->head);
1689 wait_queue.count--;
1690 free(wait_node);
1691 }
1692
1693 DBG("Dispatch thread dying");
1694 if (err) {
1695 health_error();
1696 ERR("Health error occurred in %s", __func__);
1697 }
1698 health_unregister();
1699 return NULL;
1700 }
1701
1702 /*
1703 * This thread manage application registration.
1704 */
1705 static void *thread_registration_apps(void *data)
1706 {
1707 int sock = -1, i, ret, pollfd, err = -1;
1708 uint32_t revents, nb_fd;
1709 struct lttng_poll_event events;
1710 /*
1711 * Get allocated in this thread, enqueued to a global queue, dequeued and
1712 * freed in the manage apps thread.
1713 */
1714 struct ust_command *ust_cmd = NULL;
1715
1716 DBG("[thread] Manage application registration started");
1717
1718 health_register(HEALTH_TYPE_APP_REG);
1719
1720 if (testpoint(thread_registration_apps)) {
1721 goto error_testpoint;
1722 }
1723
1724 ret = lttcomm_listen_unix_sock(apps_sock);
1725 if (ret < 0) {
1726 goto error_listen;
1727 }
1728
1729 /*
1730 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1731 * more will be added to this poll set.
1732 */
1733 ret = sessiond_set_thread_pollset(&events, 2);
1734 if (ret < 0) {
1735 goto error_create_poll;
1736 }
1737
1738 /* Add the application registration socket */
1739 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1740 if (ret < 0) {
1741 goto error_poll_add;
1742 }
1743
1744 /* Notify all applications to register */
1745 ret = notify_ust_apps(1);
1746 if (ret < 0) {
1747 ERR("Failed to notify applications or create the wait shared memory.\n"
1748 "Execution continues but there might be problem for already\n"
1749 "running applications that wishes to register.");
1750 }
1751
1752 while (1) {
1753 DBG("Accepting application registration");
1754
1755 /* Inifinite blocking call, waiting for transmission */
1756 restart:
1757 health_poll_entry();
1758 ret = lttng_poll_wait(&events, -1);
1759 health_poll_exit();
1760 if (ret < 0) {
1761 /*
1762 * Restart interrupted system call.
1763 */
1764 if (errno == EINTR) {
1765 goto restart;
1766 }
1767 goto error;
1768 }
1769
1770 nb_fd = ret;
1771
1772 for (i = 0; i < nb_fd; i++) {
1773 health_code_update();
1774
1775 /* Fetch once the poll data */
1776 revents = LTTNG_POLL_GETEV(&events, i);
1777 pollfd = LTTNG_POLL_GETFD(&events, i);
1778
1779 /* Thread quit pipe has been closed. Killing thread. */
1780 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1781 if (ret) {
1782 err = 0;
1783 goto exit;
1784 }
1785
1786 /* Event on the registration socket */
1787 if (pollfd == apps_sock) {
1788 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1789 ERR("Register apps socket poll error");
1790 goto error;
1791 } else if (revents & LPOLLIN) {
1792 sock = lttcomm_accept_unix_sock(apps_sock);
1793 if (sock < 0) {
1794 goto error;
1795 }
1796
1797 /*
1798 * Set the CLOEXEC flag. Return code is useless because
1799 * either way, the show must go on.
1800 */
1801 (void) utils_set_fd_cloexec(sock);
1802
1803 /* Create UST registration command for enqueuing */
1804 ust_cmd = zmalloc(sizeof(struct ust_command));
1805 if (ust_cmd == NULL) {
1806 PERROR("ust command zmalloc");
1807 goto error;
1808 }
1809
1810 /*
1811 * Using message-based transmissions to ensure we don't
1812 * have to deal with partially received messages.
1813 */
1814 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1815 if (ret < 0) {
1816 ERR("Exhausted file descriptors allowed for applications.");
1817 free(ust_cmd);
1818 ret = close(sock);
1819 if (ret) {
1820 PERROR("close");
1821 }
1822 sock = -1;
1823 continue;
1824 }
1825
1826 health_code_update();
1827 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1828 if (ret < 0) {
1829 free(ust_cmd);
1830 /* Close socket of the application. */
1831 ret = close(sock);
1832 if (ret) {
1833 PERROR("close");
1834 }
1835 lttng_fd_put(LTTNG_FD_APPS, 1);
1836 sock = -1;
1837 continue;
1838 }
1839 health_code_update();
1840
1841 ust_cmd->sock = sock;
1842 sock = -1;
1843
1844 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1845 " gid:%d sock:%d name:%s (version %d.%d)",
1846 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1847 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1848 ust_cmd->sock, ust_cmd->reg_msg.name,
1849 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1850
1851 /*
1852 * Lock free enqueue the registration request. The red pill
1853 * has been taken! This apps will be part of the *system*.
1854 */
1855 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1856
1857 /*
1858 * Wake the registration queue futex. Implicit memory
1859 * barrier with the exchange in cds_wfq_enqueue.
1860 */
1861 futex_nto1_wake(&ust_cmd_queue.futex);
1862 }
1863 }
1864 }
1865 }
1866
1867 exit:
1868 error:
1869 if (err) {
1870 health_error();
1871 ERR("Health error occurred in %s", __func__);
1872 }
1873
1874 /* Notify that the registration thread is gone */
1875 notify_ust_apps(0);
1876
1877 if (apps_sock >= 0) {
1878 ret = close(apps_sock);
1879 if (ret) {
1880 PERROR("close");
1881 }
1882 }
1883 if (sock >= 0) {
1884 ret = close(sock);
1885 if (ret) {
1886 PERROR("close");
1887 }
1888 lttng_fd_put(LTTNG_FD_APPS, 1);
1889 }
1890 unlink(apps_unix_sock_path);
1891
1892 error_poll_add:
1893 lttng_poll_clean(&events);
1894 error_listen:
1895 error_create_poll:
1896 error_testpoint:
1897 DBG("UST Registration thread cleanup complete");
1898 health_unregister();
1899
1900 return NULL;
1901 }
1902
1903 /*
1904 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1905 * exec or it will fails.
1906 */
1907 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1908 {
1909 int ret, clock_ret;
1910 struct timespec timeout;
1911
1912 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1913 consumer_data->consumer_thread_is_ready = 0;
1914
1915 /* Setup pthread condition */
1916 ret = pthread_condattr_init(&consumer_data->condattr);
1917 if (ret != 0) {
1918 errno = ret;
1919 PERROR("pthread_condattr_init consumer data");
1920 goto error;
1921 }
1922
1923 /*
1924 * Set the monotonic clock in order to make sure we DO NOT jump in time
1925 * between the clock_gettime() call and the timedwait call. See bug #324
1926 * for a more details and how we noticed it.
1927 */
1928 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1929 if (ret != 0) {
1930 errno = ret;
1931 PERROR("pthread_condattr_setclock consumer data");
1932 goto error;
1933 }
1934
1935 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1936 if (ret != 0) {
1937 errno = ret;
1938 PERROR("pthread_cond_init consumer data");
1939 goto error;
1940 }
1941
1942 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1943 consumer_data);
1944 if (ret != 0) {
1945 PERROR("pthread_create consumer");
1946 ret = -1;
1947 goto error;
1948 }
1949
1950 /* We are about to wait on a pthread condition */
1951 pthread_mutex_lock(&consumer_data->cond_mutex);
1952
1953 /* Get time for sem_timedwait absolute timeout */
1954 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1955 /*
1956 * Set the timeout for the condition timed wait even if the clock gettime
1957 * call fails since we might loop on that call and we want to avoid to
1958 * increment the timeout too many times.
1959 */
1960 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1961
1962 /*
1963 * The following loop COULD be skipped in some conditions so this is why we
1964 * set ret to 0 in order to make sure at least one round of the loop is
1965 * done.
1966 */
1967 ret = 0;
1968
1969 /*
1970 * Loop until the condition is reached or when a timeout is reached. Note
1971 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1972 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1973 * possible. This loop does not take any chances and works with both of
1974 * them.
1975 */
1976 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1977 if (clock_ret < 0) {
1978 PERROR("clock_gettime spawn consumer");
1979 /* Infinite wait for the consumerd thread to be ready */
1980 ret = pthread_cond_wait(&consumer_data->cond,
1981 &consumer_data->cond_mutex);
1982 } else {
1983 ret = pthread_cond_timedwait(&consumer_data->cond,
1984 &consumer_data->cond_mutex, &timeout);
1985 }
1986 }
1987
1988 /* Release the pthread condition */
1989 pthread_mutex_unlock(&consumer_data->cond_mutex);
1990
1991 if (ret != 0) {
1992 errno = ret;
1993 if (ret == ETIMEDOUT) {
1994 /*
1995 * Call has timed out so we kill the kconsumerd_thread and return
1996 * an error.
1997 */
1998 ERR("Condition timed out. The consumer thread was never ready."
1999 " Killing it");
2000 ret = pthread_cancel(consumer_data->thread);
2001 if (ret < 0) {
2002 PERROR("pthread_cancel consumer thread");
2003 }
2004 } else {
2005 PERROR("pthread_cond_wait failed consumer thread");
2006 }
2007 goto error;
2008 }
2009
2010 pthread_mutex_lock(&consumer_data->pid_mutex);
2011 if (consumer_data->pid == 0) {
2012 ERR("Consumerd did not start");
2013 pthread_mutex_unlock(&consumer_data->pid_mutex);
2014 goto error;
2015 }
2016 pthread_mutex_unlock(&consumer_data->pid_mutex);
2017
2018 return 0;
2019
2020 error:
2021 return ret;
2022 }
2023
2024 /*
2025 * Join consumer thread
2026 */
2027 static int join_consumer_thread(struct consumer_data *consumer_data)
2028 {
2029 void *status;
2030
2031 /* Consumer pid must be a real one. */
2032 if (consumer_data->pid > 0) {
2033 int ret;
2034 ret = kill(consumer_data->pid, SIGTERM);
2035 if (ret) {
2036 ERR("Error killing consumer daemon");
2037 return ret;
2038 }
2039 return pthread_join(consumer_data->thread, &status);
2040 } else {
2041 return 0;
2042 }
2043 }
2044
2045 /*
2046 * Fork and exec a consumer daemon (consumerd).
2047 *
2048 * Return pid if successful else -1.
2049 */
2050 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2051 {
2052 int ret;
2053 pid_t pid;
2054 const char *consumer_to_use;
2055 const char *verbosity;
2056 struct stat st;
2057
2058 DBG("Spawning consumerd");
2059
2060 pid = fork();
2061 if (pid == 0) {
2062 /*
2063 * Exec consumerd.
2064 */
2065 if (opt_verbose_consumer) {
2066 verbosity = "--verbose";
2067 } else {
2068 verbosity = "--quiet";
2069 }
2070 switch (consumer_data->type) {
2071 case LTTNG_CONSUMER_KERNEL:
2072 /*
2073 * Find out which consumerd to execute. We will first try the
2074 * 64-bit path, then the sessiond's installation directory, and
2075 * fallback on the 32-bit one,
2076 */
2077 DBG3("Looking for a kernel consumer at these locations:");
2078 DBG3(" 1) %s", consumerd64_bin);
2079 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2080 DBG3(" 3) %s", consumerd32_bin);
2081 if (stat(consumerd64_bin, &st) == 0) {
2082 DBG3("Found location #1");
2083 consumer_to_use = consumerd64_bin;
2084 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2085 DBG3("Found location #2");
2086 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2087 } else if (stat(consumerd32_bin, &st) == 0) {
2088 DBG3("Found location #3");
2089 consumer_to_use = consumerd32_bin;
2090 } else {
2091 DBG("Could not find any valid consumerd executable");
2092 break;
2093 }
2094 DBG("Using kernel consumer at: %s", consumer_to_use);
2095 execl(consumer_to_use,
2096 "lttng-consumerd", verbosity, "-k",
2097 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2098 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2099 NULL);
2100 break;
2101 case LTTNG_CONSUMER64_UST:
2102 {
2103 char *tmpnew = NULL;
2104
2105 if (consumerd64_libdir[0] != '\0') {
2106 char *tmp;
2107 size_t tmplen;
2108
2109 tmp = getenv("LD_LIBRARY_PATH");
2110 if (!tmp) {
2111 tmp = "";
2112 }
2113 tmplen = strlen("LD_LIBRARY_PATH=")
2114 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2115 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2116 if (!tmpnew) {
2117 ret = -ENOMEM;
2118 goto error;
2119 }
2120 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2121 strcat(tmpnew, consumerd64_libdir);
2122 if (tmp[0] != '\0') {
2123 strcat(tmpnew, ":");
2124 strcat(tmpnew, tmp);
2125 }
2126 ret = putenv(tmpnew);
2127 if (ret) {
2128 ret = -errno;
2129 free(tmpnew);
2130 goto error;
2131 }
2132 }
2133 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2134 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2135 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2136 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2137 NULL);
2138 if (consumerd64_libdir[0] != '\0') {
2139 free(tmpnew);
2140 }
2141 if (ret) {
2142 goto error;
2143 }
2144 break;
2145 }
2146 case LTTNG_CONSUMER32_UST:
2147 {
2148 char *tmpnew = NULL;
2149
2150 if (consumerd32_libdir[0] != '\0') {
2151 char *tmp;
2152 size_t tmplen;
2153
2154 tmp = getenv("LD_LIBRARY_PATH");
2155 if (!tmp) {
2156 tmp = "";
2157 }
2158 tmplen = strlen("LD_LIBRARY_PATH=")
2159 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2160 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2161 if (!tmpnew) {
2162 ret = -ENOMEM;
2163 goto error;
2164 }
2165 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2166 strcat(tmpnew, consumerd32_libdir);
2167 if (tmp[0] != '\0') {
2168 strcat(tmpnew, ":");
2169 strcat(tmpnew, tmp);
2170 }
2171 ret = putenv(tmpnew);
2172 if (ret) {
2173 ret = -errno;
2174 free(tmpnew);
2175 goto error;
2176 }
2177 }
2178 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2179 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2180 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2181 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2182 NULL);
2183 if (consumerd32_libdir[0] != '\0') {
2184 free(tmpnew);
2185 }
2186 if (ret) {
2187 goto error;
2188 }
2189 break;
2190 }
2191 default:
2192 PERROR("unknown consumer type");
2193 exit(EXIT_FAILURE);
2194 }
2195 if (errno != 0) {
2196 PERROR("kernel start consumer exec");
2197 }
2198 exit(EXIT_FAILURE);
2199 } else if (pid > 0) {
2200 ret = pid;
2201 } else {
2202 PERROR("start consumer fork");
2203 ret = -errno;
2204 }
2205 error:
2206 return ret;
2207 }
2208
2209 /*
2210 * Spawn the consumerd daemon and session daemon thread.
2211 */
2212 static int start_consumerd(struct consumer_data *consumer_data)
2213 {
2214 int ret;
2215
2216 /*
2217 * Set the listen() state on the socket since there is a possible race
2218 * between the exec() of the consumer daemon and this call if place in the
2219 * consumer thread. See bug #366 for more details.
2220 */
2221 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2222 if (ret < 0) {
2223 goto error;
2224 }
2225
2226 pthread_mutex_lock(&consumer_data->pid_mutex);
2227 if (consumer_data->pid != 0) {
2228 pthread_mutex_unlock(&consumer_data->pid_mutex);
2229 goto end;
2230 }
2231
2232 ret = spawn_consumerd(consumer_data);
2233 if (ret < 0) {
2234 ERR("Spawning consumerd failed");
2235 pthread_mutex_unlock(&consumer_data->pid_mutex);
2236 goto error;
2237 }
2238
2239 /* Setting up the consumer_data pid */
2240 consumer_data->pid = ret;
2241 DBG2("Consumer pid %d", consumer_data->pid);
2242 pthread_mutex_unlock(&consumer_data->pid_mutex);
2243
2244 DBG2("Spawning consumer control thread");
2245 ret = spawn_consumer_thread(consumer_data);
2246 if (ret < 0) {
2247 ERR("Fatal error spawning consumer control thread");
2248 goto error;
2249 }
2250
2251 end:
2252 return 0;
2253
2254 error:
2255 /* Cleanup already created sockets on error. */
2256 if (consumer_data->err_sock >= 0) {
2257 int err;
2258
2259 err = close(consumer_data->err_sock);
2260 if (err < 0) {
2261 PERROR("close consumer data error socket");
2262 }
2263 }
2264 return ret;
2265 }
2266
2267 /*
2268 * Compute health status of each consumer. If one of them is zero (bad
2269 * state), we return 0.
2270 */
2271 static int check_consumer_health(void)
2272 {
2273 int ret;
2274
2275 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2276
2277 DBG3("Health consumer check %d", ret);
2278
2279 return ret;
2280 }
2281
2282 /*
2283 * Setup necessary data for kernel tracer action.
2284 */
2285 static int init_kernel_tracer(void)
2286 {
2287 int ret;
2288
2289 /* Modprobe lttng kernel modules */
2290 ret = modprobe_lttng_control();
2291 if (ret < 0) {
2292 goto error;
2293 }
2294
2295 /* Open debugfs lttng */
2296 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2297 if (kernel_tracer_fd < 0) {
2298 DBG("Failed to open %s", module_proc_lttng);
2299 ret = -1;
2300 goto error_open;
2301 }
2302
2303 /* Validate kernel version */
2304 ret = kernel_validate_version(kernel_tracer_fd);
2305 if (ret < 0) {
2306 goto error_version;
2307 }
2308
2309 ret = modprobe_lttng_data();
2310 if (ret < 0) {
2311 goto error_modules;
2312 }
2313
2314 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2315 return 0;
2316
2317 error_version:
2318 modprobe_remove_lttng_control();
2319 ret = close(kernel_tracer_fd);
2320 if (ret) {
2321 PERROR("close");
2322 }
2323 kernel_tracer_fd = -1;
2324 return LTTNG_ERR_KERN_VERSION;
2325
2326 error_modules:
2327 ret = close(kernel_tracer_fd);
2328 if (ret) {
2329 PERROR("close");
2330 }
2331
2332 error_open:
2333 modprobe_remove_lttng_control();
2334
2335 error:
2336 WARN("No kernel tracer available");
2337 kernel_tracer_fd = -1;
2338 if (!is_root) {
2339 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2340 } else {
2341 return LTTNG_ERR_KERN_NA;
2342 }
2343 }
2344
2345
2346 /*
2347 * Copy consumer output from the tracing session to the domain session. The
2348 * function also applies the right modification on a per domain basis for the
2349 * trace files destination directory.
2350 *
2351 * Should *NOT* be called with RCU read-side lock held.
2352 */
2353 static int copy_session_consumer(int domain, struct ltt_session *session)
2354 {
2355 int ret;
2356 const char *dir_name;
2357 struct consumer_output *consumer;
2358
2359 assert(session);
2360 assert(session->consumer);
2361
2362 switch (domain) {
2363 case LTTNG_DOMAIN_KERNEL:
2364 DBG3("Copying tracing session consumer output in kernel session");
2365 /*
2366 * XXX: We should audit the session creation and what this function
2367 * does "extra" in order to avoid a destroy since this function is used
2368 * in the domain session creation (kernel and ust) only. Same for UST
2369 * domain.
2370 */
2371 if (session->kernel_session->consumer) {
2372 consumer_destroy_output(session->kernel_session->consumer);
2373 }
2374 session->kernel_session->consumer =
2375 consumer_copy_output(session->consumer);
2376 /* Ease our life a bit for the next part */
2377 consumer = session->kernel_session->consumer;
2378 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2379 break;
2380 case LTTNG_DOMAIN_UST:
2381 DBG3("Copying tracing session consumer output in UST session");
2382 if (session->ust_session->consumer) {
2383 consumer_destroy_output(session->ust_session->consumer);
2384 }
2385 session->ust_session->consumer =
2386 consumer_copy_output(session->consumer);
2387 /* Ease our life a bit for the next part */
2388 consumer = session->ust_session->consumer;
2389 dir_name = DEFAULT_UST_TRACE_DIR;
2390 break;
2391 default:
2392 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2393 goto error;
2394 }
2395
2396 /* Append correct directory to subdir */
2397 strncat(consumer->subdir, dir_name,
2398 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2399 DBG3("Copy session consumer subdir %s", consumer->subdir);
2400
2401 ret = LTTNG_OK;
2402
2403 error:
2404 return ret;
2405 }
2406
2407 /*
2408 * Create an UST session and add it to the session ust list.
2409 *
2410 * Should *NOT* be called with RCU read-side lock held.
2411 */
2412 static int create_ust_session(struct ltt_session *session,
2413 struct lttng_domain *domain)
2414 {
2415 int ret;
2416 struct ltt_ust_session *lus = NULL;
2417
2418 assert(session);
2419 assert(domain);
2420 assert(session->consumer);
2421
2422 switch (domain->type) {
2423 case LTTNG_DOMAIN_UST:
2424 break;
2425 default:
2426 ERR("Unknown UST domain on create session %d", domain->type);
2427 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2428 goto error;
2429 }
2430
2431 DBG("Creating UST session");
2432
2433 lus = trace_ust_create_session(session->id);
2434 if (lus == NULL) {
2435 ret = LTTNG_ERR_UST_SESS_FAIL;
2436 goto error;
2437 }
2438
2439 lus->uid = session->uid;
2440 lus->gid = session->gid;
2441 lus->output_traces = session->output_traces;
2442 lus->snapshot_mode = session->snapshot_mode;
2443 session->ust_session = lus;
2444
2445 /* Copy session output to the newly created UST session */
2446 ret = copy_session_consumer(domain->type, session);
2447 if (ret != LTTNG_OK) {
2448 goto error;
2449 }
2450
2451 return LTTNG_OK;
2452
2453 error:
2454 free(lus);
2455 session->ust_session = NULL;
2456 return ret;
2457 }
2458
2459 /*
2460 * Create a kernel tracer session then create the default channel.
2461 */
2462 static int create_kernel_session(struct ltt_session *session)
2463 {
2464 int ret;
2465
2466 DBG("Creating kernel session");
2467
2468 ret = kernel_create_session(session, kernel_tracer_fd);
2469 if (ret < 0) {
2470 ret = LTTNG_ERR_KERN_SESS_FAIL;
2471 goto error;
2472 }
2473
2474 /* Code flow safety */
2475 assert(session->kernel_session);
2476
2477 /* Copy session output to the newly created Kernel session */
2478 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2479 if (ret != LTTNG_OK) {
2480 goto error;
2481 }
2482
2483 /* Create directory(ies) on local filesystem. */
2484 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2485 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2486 ret = run_as_mkdir_recursive(
2487 session->kernel_session->consumer->dst.trace_path,
2488 S_IRWXU | S_IRWXG, session->uid, session->gid);
2489 if (ret < 0) {
2490 if (ret != -EEXIST) {
2491 ERR("Trace directory creation error");
2492 goto error;
2493 }
2494 }
2495 }
2496
2497 session->kernel_session->uid = session->uid;
2498 session->kernel_session->gid = session->gid;
2499 session->kernel_session->output_traces = session->output_traces;
2500 session->kernel_session->snapshot_mode = session->snapshot_mode;
2501
2502 return LTTNG_OK;
2503
2504 error:
2505 trace_kernel_destroy_session(session->kernel_session);
2506 session->kernel_session = NULL;
2507 return ret;
2508 }
2509
2510 /*
2511 * Count number of session permitted by uid/gid.
2512 */
2513 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2514 {
2515 unsigned int i = 0;
2516 struct ltt_session *session;
2517
2518 DBG("Counting number of available session for UID %d GID %d",
2519 uid, gid);
2520 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2521 /*
2522 * Only list the sessions the user can control.
2523 */
2524 if (!session_access_ok(session, uid, gid)) {
2525 continue;
2526 }
2527 i++;
2528 }
2529 return i;
2530 }
2531
2532 /*
2533 * Process the command requested by the lttng client within the command
2534 * context structure. This function make sure that the return structure (llm)
2535 * is set and ready for transmission before returning.
2536 *
2537 * Return any error encountered or 0 for success.
2538 *
2539 * "sock" is only used for special-case var. len data.
2540 *
2541 * Should *NOT* be called with RCU read-side lock held.
2542 */
2543 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2544 int *sock_error)
2545 {
2546 int ret = LTTNG_OK;
2547 int need_tracing_session = 1;
2548 int need_domain;
2549
2550 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2551
2552 *sock_error = 0;
2553
2554 switch (cmd_ctx->lsm->cmd_type) {
2555 case LTTNG_CREATE_SESSION:
2556 case LTTNG_CREATE_SESSION_SNAPSHOT:
2557 case LTTNG_DESTROY_SESSION:
2558 case LTTNG_LIST_SESSIONS:
2559 case LTTNG_LIST_DOMAINS:
2560 case LTTNG_START_TRACE:
2561 case LTTNG_STOP_TRACE:
2562 case LTTNG_DATA_PENDING:
2563 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2564 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2565 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2566 case LTTNG_SNAPSHOT_RECORD:
2567 need_domain = 0;
2568 break;
2569 default:
2570 need_domain = 1;
2571 }
2572
2573 if (opt_no_kernel && need_domain
2574 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2575 if (!is_root) {
2576 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2577 } else {
2578 ret = LTTNG_ERR_KERN_NA;
2579 }
2580 goto error;
2581 }
2582
2583 /* Deny register consumer if we already have a spawned consumer. */
2584 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2585 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2586 if (kconsumer_data.pid > 0) {
2587 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2588 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2589 goto error;
2590 }
2591 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2592 }
2593
2594 /*
2595 * Check for command that don't needs to allocate a returned payload. We do
2596 * this here so we don't have to make the call for no payload at each
2597 * command.
2598 */
2599 switch(cmd_ctx->lsm->cmd_type) {
2600 case LTTNG_LIST_SESSIONS:
2601 case LTTNG_LIST_TRACEPOINTS:
2602 case LTTNG_LIST_TRACEPOINT_FIELDS:
2603 case LTTNG_LIST_DOMAINS:
2604 case LTTNG_LIST_CHANNELS:
2605 case LTTNG_LIST_EVENTS:
2606 break;
2607 default:
2608 /* Setup lttng message with no payload */
2609 ret = setup_lttng_msg(cmd_ctx, 0);
2610 if (ret < 0) {
2611 /* This label does not try to unlock the session */
2612 goto init_setup_error;
2613 }
2614 }
2615
2616 /* Commands that DO NOT need a session. */
2617 switch (cmd_ctx->lsm->cmd_type) {
2618 case LTTNG_CREATE_SESSION:
2619 case LTTNG_CREATE_SESSION_SNAPSHOT:
2620 case LTTNG_CALIBRATE:
2621 case LTTNG_LIST_SESSIONS:
2622 case LTTNG_LIST_TRACEPOINTS:
2623 case LTTNG_LIST_TRACEPOINT_FIELDS:
2624 need_tracing_session = 0;
2625 break;
2626 default:
2627 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2628 /*
2629 * We keep the session list lock across _all_ commands
2630 * for now, because the per-session lock does not
2631 * handle teardown properly.
2632 */
2633 session_lock_list();
2634 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2635 if (cmd_ctx->session == NULL) {
2636 ret = LTTNG_ERR_SESS_NOT_FOUND;
2637 goto error;
2638 } else {
2639 /* Acquire lock for the session */
2640 session_lock(cmd_ctx->session);
2641 }
2642 break;
2643 }
2644
2645 if (!need_domain) {
2646 goto skip_domain;
2647 }
2648
2649 /*
2650 * Check domain type for specific "pre-action".
2651 */
2652 switch (cmd_ctx->lsm->domain.type) {
2653 case LTTNG_DOMAIN_KERNEL:
2654 if (!is_root) {
2655 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2656 goto error;
2657 }
2658
2659 /* Kernel tracer check */
2660 if (kernel_tracer_fd == -1) {
2661 /* Basically, load kernel tracer modules */
2662 ret = init_kernel_tracer();
2663 if (ret != 0) {
2664 goto error;
2665 }
2666 }
2667
2668 /* Consumer is in an ERROR state. Report back to client */
2669 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2670 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2671 goto error;
2672 }
2673
2674 /* Need a session for kernel command */
2675 if (need_tracing_session) {
2676 if (cmd_ctx->session->kernel_session == NULL) {
2677 ret = create_kernel_session(cmd_ctx->session);
2678 if (ret < 0) {
2679 ret = LTTNG_ERR_KERN_SESS_FAIL;
2680 goto error;
2681 }
2682 }
2683
2684 /* Start the kernel consumer daemon */
2685 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2686 if (kconsumer_data.pid == 0 &&
2687 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2688 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2689 ret = start_consumerd(&kconsumer_data);
2690 if (ret < 0) {
2691 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2692 goto error;
2693 }
2694 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2695 } else {
2696 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2697 }
2698
2699 /*
2700 * The consumer was just spawned so we need to add the socket to
2701 * the consumer output of the session if exist.
2702 */
2703 ret = consumer_create_socket(&kconsumer_data,
2704 cmd_ctx->session->kernel_session->consumer);
2705 if (ret < 0) {
2706 goto error;
2707 }
2708 }
2709
2710 break;
2711 case LTTNG_DOMAIN_UST:
2712 {
2713 /* Consumer is in an ERROR state. Report back to client */
2714 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2715 ret = LTTNG_ERR_NO_USTCONSUMERD;
2716 goto error;
2717 }
2718
2719 if (need_tracing_session) {
2720 /* Create UST session if none exist. */
2721 if (cmd_ctx->session->ust_session == NULL) {
2722 ret = create_ust_session(cmd_ctx->session,
2723 &cmd_ctx->lsm->domain);
2724 if (ret != LTTNG_OK) {
2725 goto error;
2726 }
2727 }
2728
2729 /* Start the UST consumer daemons */
2730 /* 64-bit */
2731 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2732 if (consumerd64_bin[0] != '\0' &&
2733 ustconsumer64_data.pid == 0 &&
2734 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2735 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2736 ret = start_consumerd(&ustconsumer64_data);
2737 if (ret < 0) {
2738 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2739 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2740 goto error;
2741 }
2742
2743 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2744 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2745 } else {
2746 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2747 }
2748
2749 /*
2750 * Setup socket for consumer 64 bit. No need for atomic access
2751 * since it was set above and can ONLY be set in this thread.
2752 */
2753 ret = consumer_create_socket(&ustconsumer64_data,
2754 cmd_ctx->session->ust_session->consumer);
2755 if (ret < 0) {
2756 goto error;
2757 }
2758
2759 /* 32-bit */
2760 if (consumerd32_bin[0] != '\0' &&
2761 ustconsumer32_data.pid == 0 &&
2762 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2763 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2764 ret = start_consumerd(&ustconsumer32_data);
2765 if (ret < 0) {
2766 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2767 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2768 goto error;
2769 }
2770
2771 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2772 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2773 } else {
2774 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2775 }
2776
2777 /*
2778 * Setup socket for consumer 64 bit. No need for atomic access
2779 * since it was set above and can ONLY be set in this thread.
2780 */
2781 ret = consumer_create_socket(&ustconsumer32_data,
2782 cmd_ctx->session->ust_session->consumer);
2783 if (ret < 0) {
2784 goto error;
2785 }
2786 }
2787 break;
2788 }
2789 default:
2790 break;
2791 }
2792 skip_domain:
2793
2794 /* Validate consumer daemon state when start/stop trace command */
2795 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2796 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2797 switch (cmd_ctx->lsm->domain.type) {
2798 case LTTNG_DOMAIN_UST:
2799 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2800 ret = LTTNG_ERR_NO_USTCONSUMERD;
2801 goto error;
2802 }
2803 break;
2804 case LTTNG_DOMAIN_KERNEL:
2805 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2806 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2807 goto error;
2808 }
2809 break;
2810 }
2811 }
2812
2813 /*
2814 * Check that the UID or GID match that of the tracing session.
2815 * The root user can interact with all sessions.
2816 */
2817 if (need_tracing_session) {
2818 if (!session_access_ok(cmd_ctx->session,
2819 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2820 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2821 ret = LTTNG_ERR_EPERM;
2822 goto error;
2823 }
2824 }
2825
2826 /*
2827 * Send relayd information to consumer as soon as we have a domain and a
2828 * session defined.
2829 */
2830 if (cmd_ctx->session && need_domain) {
2831 /*
2832 * Setup relayd if not done yet. If the relayd information was already
2833 * sent to the consumer, this call will gracefully return.
2834 */
2835 ret = cmd_setup_relayd(cmd_ctx->session);
2836 if (ret != LTTNG_OK) {
2837 goto error;
2838 }
2839 }
2840
2841 /* Process by command type */
2842 switch (cmd_ctx->lsm->cmd_type) {
2843 case LTTNG_ADD_CONTEXT:
2844 {
2845 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2846 cmd_ctx->lsm->u.context.channel_name,
2847 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2848 break;
2849 }
2850 case LTTNG_DISABLE_CHANNEL:
2851 {
2852 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2853 cmd_ctx->lsm->u.disable.channel_name);
2854 break;
2855 }
2856 case LTTNG_DISABLE_EVENT:
2857 {
2858 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2859 cmd_ctx->lsm->u.disable.channel_name,
2860 cmd_ctx->lsm->u.disable.name);
2861 break;
2862 }
2863 case LTTNG_DISABLE_ALL_EVENT:
2864 {
2865 DBG("Disabling all events");
2866
2867 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2868 cmd_ctx->lsm->u.disable.channel_name);
2869 break;
2870 }
2871 case LTTNG_ENABLE_CHANNEL:
2872 {
2873 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2874 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2875 break;
2876 }
2877 case LTTNG_ENABLE_EVENT:
2878 {
2879 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2880 cmd_ctx->lsm->u.enable.channel_name,
2881 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2882 break;
2883 }
2884 case LTTNG_ENABLE_ALL_EVENT:
2885 {
2886 DBG("Enabling all events");
2887
2888 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2889 cmd_ctx->lsm->u.enable.channel_name,
2890 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2891 break;
2892 }
2893 case LTTNG_LIST_TRACEPOINTS:
2894 {
2895 struct lttng_event *events;
2896 ssize_t nb_events;
2897
2898 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2899 if (nb_events < 0) {
2900 /* Return value is a negative lttng_error_code. */
2901 ret = -nb_events;
2902 goto error;
2903 }
2904
2905 /*
2906 * Setup lttng message with payload size set to the event list size in
2907 * bytes and then copy list into the llm payload.
2908 */
2909 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2910 if (ret < 0) {
2911 free(events);
2912 goto setup_error;
2913 }
2914
2915 /* Copy event list into message payload */
2916 memcpy(cmd_ctx->llm->payload, events,
2917 sizeof(struct lttng_event) * nb_events);
2918
2919 free(events);
2920
2921 ret = LTTNG_OK;
2922 break;
2923 }
2924 case LTTNG_LIST_TRACEPOINT_FIELDS:
2925 {
2926 struct lttng_event_field *fields;
2927 ssize_t nb_fields;
2928
2929 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2930 &fields);
2931 if (nb_fields < 0) {
2932 /* Return value is a negative lttng_error_code. */
2933 ret = -nb_fields;
2934 goto error;
2935 }
2936
2937 /*
2938 * Setup lttng message with payload size set to the event list size in
2939 * bytes and then copy list into the llm payload.
2940 */
2941 ret = setup_lttng_msg(cmd_ctx,
2942 sizeof(struct lttng_event_field) * nb_fields);
2943 if (ret < 0) {
2944 free(fields);
2945 goto setup_error;
2946 }
2947
2948 /* Copy event list into message payload */
2949 memcpy(cmd_ctx->llm->payload, fields,
2950 sizeof(struct lttng_event_field) * nb_fields);
2951
2952 free(fields);
2953
2954 ret = LTTNG_OK;
2955 break;
2956 }
2957 case LTTNG_SET_CONSUMER_URI:
2958 {
2959 size_t nb_uri, len;
2960 struct lttng_uri *uris;
2961
2962 nb_uri = cmd_ctx->lsm->u.uri.size;
2963 len = nb_uri * sizeof(struct lttng_uri);
2964
2965 if (nb_uri == 0) {
2966 ret = LTTNG_ERR_INVALID;
2967 goto error;
2968 }
2969
2970 uris = zmalloc(len);
2971 if (uris == NULL) {
2972 ret = LTTNG_ERR_FATAL;
2973 goto error;
2974 }
2975
2976 /* Receive variable len data */
2977 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2978 ret = lttcomm_recv_unix_sock(sock, uris, len);
2979 if (ret <= 0) {
2980 DBG("No URIs received from client... continuing");
2981 *sock_error = 1;
2982 ret = LTTNG_ERR_SESSION_FAIL;
2983 free(uris);
2984 goto error;
2985 }
2986
2987 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2988 nb_uri, uris);
2989 if (ret != LTTNG_OK) {
2990 free(uris);
2991 goto error;
2992 }
2993
2994 /*
2995 * XXX: 0 means that this URI should be applied on the session. Should
2996 * be a DOMAIN enuam.
2997 */
2998 if (cmd_ctx->lsm->domain.type == 0) {
2999 /* Add the URI for the UST session if a consumer is present. */
3000 if (cmd_ctx->session->ust_session &&
3001 cmd_ctx->session->ust_session->consumer) {
3002 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3003 nb_uri, uris);
3004 } else if (cmd_ctx->session->kernel_session &&
3005 cmd_ctx->session->kernel_session->consumer) {
3006 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3007 cmd_ctx->session, nb_uri, uris);
3008 }
3009 }
3010
3011 free(uris);
3012
3013 break;
3014 }
3015 case LTTNG_START_TRACE:
3016 {
3017 ret = cmd_start_trace(cmd_ctx->session);
3018 break;
3019 }
3020 case LTTNG_STOP_TRACE:
3021 {
3022 ret = cmd_stop_trace(cmd_ctx->session);
3023 break;
3024 }
3025 case LTTNG_CREATE_SESSION:
3026 {
3027 size_t nb_uri, len;
3028 struct lttng_uri *uris = NULL;
3029
3030 nb_uri = cmd_ctx->lsm->u.uri.size;
3031 len = nb_uri * sizeof(struct lttng_uri);
3032
3033 if (nb_uri > 0) {
3034 uris = zmalloc(len);
3035 if (uris == NULL) {
3036 ret = LTTNG_ERR_FATAL;
3037 goto error;
3038 }
3039
3040 /* Receive variable len data */
3041 DBG("Waiting for %zu URIs from client ...", nb_uri);
3042 ret = lttcomm_recv_unix_sock(sock, uris, len);
3043 if (ret <= 0) {
3044 DBG("No URIs received from client... continuing");
3045 *sock_error = 1;
3046 ret = LTTNG_ERR_SESSION_FAIL;
3047 free(uris);
3048 goto error;
3049 }
3050
3051 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3052 DBG("Creating session with ONE network URI is a bad call");
3053 ret = LTTNG_ERR_SESSION_FAIL;
3054 free(uris);
3055 goto error;
3056 }
3057 }
3058
3059 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3060 &cmd_ctx->creds);
3061
3062 free(uris);
3063
3064 break;
3065 }
3066 case LTTNG_DESTROY_SESSION:
3067 {
3068 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3069
3070 /* Set session to NULL so we do not unlock it after free. */
3071 cmd_ctx->session = NULL;
3072 break;
3073 }
3074 case LTTNG_LIST_DOMAINS:
3075 {
3076 ssize_t nb_dom;
3077 struct lttng_domain *domains;
3078
3079 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3080 if (nb_dom < 0) {
3081 /* Return value is a negative lttng_error_code. */
3082 ret = -nb_dom;
3083 goto error;
3084 }
3085
3086 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3087 if (ret < 0) {
3088 free(domains);
3089 goto setup_error;
3090 }
3091
3092 /* Copy event list into message payload */
3093 memcpy(cmd_ctx->llm->payload, domains,
3094 nb_dom * sizeof(struct lttng_domain));
3095
3096 free(domains);
3097
3098 ret = LTTNG_OK;
3099 break;
3100 }
3101 case LTTNG_LIST_CHANNELS:
3102 {
3103 int nb_chan;
3104 struct lttng_channel *channels;
3105
3106 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3107 cmd_ctx->session, &channels);
3108 if (nb_chan < 0) {
3109 /* Return value is a negative lttng_error_code. */
3110 ret = -nb_chan;
3111 goto error;
3112 }
3113
3114 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3115 if (ret < 0) {
3116 free(channels);
3117 goto setup_error;
3118 }
3119
3120 /* Copy event list into message payload */
3121 memcpy(cmd_ctx->llm->payload, channels,
3122 nb_chan * sizeof(struct lttng_channel));
3123
3124 free(channels);
3125
3126 ret = LTTNG_OK;
3127 break;
3128 }
3129 case LTTNG_LIST_EVENTS:
3130 {
3131 ssize_t nb_event;
3132 struct lttng_event *events = NULL;
3133
3134 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3135 cmd_ctx->lsm->u.list.channel_name, &events);
3136 if (nb_event < 0) {
3137 /* Return value is a negative lttng_error_code. */
3138 ret = -nb_event;
3139 goto error;
3140 }
3141
3142 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3143 if (ret < 0) {
3144 free(events);
3145 goto setup_error;
3146 }
3147
3148 /* Copy event list into message payload */
3149 memcpy(cmd_ctx->llm->payload, events,
3150 nb_event * sizeof(struct lttng_event));
3151
3152 free(events);
3153
3154 ret = LTTNG_OK;
3155 break;
3156 }
3157 case LTTNG_LIST_SESSIONS:
3158 {
3159 unsigned int nr_sessions;
3160
3161 session_lock_list();
3162 nr_sessions = lttng_sessions_count(
3163 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3164 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3165
3166 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3167 if (ret < 0) {
3168 session_unlock_list();
3169 goto setup_error;
3170 }
3171
3172 /* Filled the session array */
3173 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3174 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3175 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3176
3177 session_unlock_list();
3178
3179 ret = LTTNG_OK;
3180 break;
3181 }
3182 case LTTNG_CALIBRATE:
3183 {
3184 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3185 &cmd_ctx->lsm->u.calibrate);
3186 break;
3187 }
3188 case LTTNG_REGISTER_CONSUMER:
3189 {
3190 struct consumer_data *cdata;
3191
3192 switch (cmd_ctx->lsm->domain.type) {
3193 case LTTNG_DOMAIN_KERNEL:
3194 cdata = &kconsumer_data;
3195 break;
3196 default:
3197 ret = LTTNG_ERR_UND;
3198 goto error;
3199 }
3200
3201 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3202 cmd_ctx->lsm->u.reg.path, cdata);
3203 break;
3204 }
3205 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3206 {
3207 struct lttng_filter_bytecode *bytecode;
3208
3209 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3210 ret = LTTNG_ERR_FILTER_INVAL;
3211 goto error;
3212 }
3213 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3214 ret = LTTNG_ERR_FILTER_INVAL;
3215 goto error;
3216 }
3217 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3218 if (!bytecode) {
3219 ret = LTTNG_ERR_FILTER_NOMEM;
3220 goto error;
3221 }
3222 /* Receive var. len. data */
3223 DBG("Receiving var len data from client ...");
3224 ret = lttcomm_recv_unix_sock(sock, bytecode,
3225 cmd_ctx->lsm->u.enable.bytecode_len);
3226 if (ret <= 0) {
3227 DBG("Nothing recv() from client var len data... continuing");
3228 *sock_error = 1;
3229 ret = LTTNG_ERR_FILTER_INVAL;
3230 goto error;
3231 }
3232
3233 if (bytecode->len + sizeof(*bytecode)
3234 != cmd_ctx->lsm->u.enable.bytecode_len) {
3235 free(bytecode);
3236 ret = LTTNG_ERR_FILTER_INVAL;
3237 goto error;
3238 }
3239
3240 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3241 cmd_ctx->lsm->u.enable.channel_name,
3242 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3243 break;
3244 }
3245 case LTTNG_DATA_PENDING:
3246 {
3247 ret = cmd_data_pending(cmd_ctx->session);
3248 break;
3249 }
3250 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3251 {
3252 struct lttcomm_lttng_output_id reply;
3253
3254 ret = cmd_snapshot_add_output(cmd_ctx->session,
3255 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3256 if (ret != LTTNG_OK) {
3257 goto error;
3258 }
3259
3260 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3261 if (ret < 0) {
3262 goto setup_error;
3263 }
3264
3265 /* Copy output list into message payload */
3266 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3267 ret = LTTNG_OK;
3268 break;
3269 }
3270 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3271 {
3272 ret = cmd_snapshot_del_output(cmd_ctx->session,
3273 &cmd_ctx->lsm->u.snapshot_output.output);
3274 break;
3275 }
3276 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3277 {
3278 ssize_t nb_output;
3279 struct lttng_snapshot_output *outputs = NULL;
3280
3281 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3282 if (nb_output < 0) {
3283 ret = -nb_output;
3284 goto error;
3285 }
3286
3287 ret = setup_lttng_msg(cmd_ctx,
3288 nb_output * sizeof(struct lttng_snapshot_output));
3289 if (ret < 0) {
3290 free(outputs);
3291 goto setup_error;
3292 }
3293
3294 if (outputs) {
3295 /* Copy output list into message payload */
3296 memcpy(cmd_ctx->llm->payload, outputs,
3297 nb_output * sizeof(struct lttng_snapshot_output));
3298 free(outputs);
3299 }
3300
3301 ret = LTTNG_OK;
3302 break;
3303 }
3304 case LTTNG_SNAPSHOT_RECORD:
3305 {
3306 ret = cmd_snapshot_record(cmd_ctx->session,
3307 &cmd_ctx->lsm->u.snapshot_record.output,
3308 cmd_ctx->lsm->u.snapshot_record.wait);
3309 break;
3310 }
3311 case LTTNG_CREATE_SESSION_SNAPSHOT:
3312 {
3313 size_t nb_uri, len;
3314 struct lttng_uri *uris = NULL;
3315
3316 nb_uri = cmd_ctx->lsm->u.uri.size;
3317 len = nb_uri * sizeof(struct lttng_uri);
3318
3319 if (nb_uri > 0) {
3320 uris = zmalloc(len);
3321 if (uris == NULL) {
3322 ret = LTTNG_ERR_FATAL;
3323 goto error;
3324 }
3325
3326 /* Receive variable len data */
3327 DBG("Waiting for %zu URIs from client ...", nb_uri);
3328 ret = lttcomm_recv_unix_sock(sock, uris, len);
3329 if (ret <= 0) {
3330 DBG("No URIs received from client... continuing");
3331 *sock_error = 1;
3332 ret = LTTNG_ERR_SESSION_FAIL;
3333 free(uris);
3334 goto error;
3335 }
3336
3337 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3338 DBG("Creating session with ONE network URI is a bad call");
3339 ret = LTTNG_ERR_SESSION_FAIL;
3340 free(uris);
3341 goto error;
3342 }
3343 }
3344
3345 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3346 nb_uri, &cmd_ctx->creds);
3347 free(uris);
3348 break;
3349 }
3350 default:
3351 ret = LTTNG_ERR_UND;
3352 break;
3353 }
3354
3355 error:
3356 if (cmd_ctx->llm == NULL) {
3357 DBG("Missing llm structure. Allocating one.");
3358 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3359 goto setup_error;
3360 }
3361 }
3362 /* Set return code */
3363 cmd_ctx->llm->ret_code = ret;
3364 setup_error:
3365 if (cmd_ctx->session) {
3366 session_unlock(cmd_ctx->session);
3367 }
3368 if (need_tracing_session) {
3369 session_unlock_list();
3370 }
3371 init_setup_error:
3372 return ret;
3373 }
3374
3375 /*
3376 * Thread managing health check socket.
3377 */
3378 static void *thread_manage_health(void *data)
3379 {
3380 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3381 uint32_t revents, nb_fd;
3382 struct lttng_poll_event events;
3383 struct lttcomm_health_msg msg;
3384 struct lttcomm_health_data reply;
3385
3386 DBG("[thread] Manage health check started");
3387
3388 rcu_register_thread();
3389
3390 /* We might hit an error path before this is created. */
3391 lttng_poll_init(&events);
3392
3393 /* Create unix socket */
3394 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3395 if (sock < 0) {
3396 ERR("Unable to create health check Unix socket");
3397 ret = -1;
3398 goto error;
3399 }
3400
3401 /*
3402 * Set the CLOEXEC flag. Return code is useless because either way, the
3403 * show must go on.
3404 */
3405 (void) utils_set_fd_cloexec(sock);
3406
3407 ret = lttcomm_listen_unix_sock(sock);
3408 if (ret < 0) {
3409 goto error;
3410 }
3411
3412 /*
3413 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3414 * more will be added to this poll set.
3415 */
3416 ret = sessiond_set_thread_pollset(&events, 2);
3417 if (ret < 0) {
3418 goto error;
3419 }
3420
3421 /* Add the application registration socket */
3422 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3423 if (ret < 0) {
3424 goto error;
3425 }
3426
3427 while (1) {
3428 DBG("Health check ready");
3429
3430 /* Inifinite blocking call, waiting for transmission */
3431 restart:
3432 ret = lttng_poll_wait(&events, -1);
3433 if (ret < 0) {
3434 /*
3435 * Restart interrupted system call.
3436 */
3437 if (errno == EINTR) {
3438 goto restart;
3439 }
3440 goto error;
3441 }
3442
3443 nb_fd = ret;
3444
3445 for (i = 0; i < nb_fd; i++) {
3446 /* Fetch once the poll data */
3447 revents = LTTNG_POLL_GETEV(&events, i);
3448 pollfd = LTTNG_POLL_GETFD(&events, i);
3449
3450 /* Thread quit pipe has been closed. Killing thread. */
3451 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3452 if (ret) {
3453 err = 0;
3454 goto exit;
3455 }
3456
3457 /* Event on the registration socket */
3458 if (pollfd == sock) {
3459 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3460 ERR("Health socket poll error");
3461 goto error;
3462 }
3463 }
3464 }
3465
3466 new_sock = lttcomm_accept_unix_sock(sock);
3467 if (new_sock < 0) {
3468 goto error;
3469 }
3470
3471 /*
3472 * Set the CLOEXEC flag. Return code is useless because either way, the
3473 * show must go on.
3474 */
3475 (void) utils_set_fd_cloexec(new_sock);
3476
3477 DBG("Receiving data from client for health...");
3478 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3479 if (ret <= 0) {
3480 DBG("Nothing recv() from client... continuing");
3481 ret = close(new_sock);
3482 if (ret) {
3483 PERROR("close");
3484 }
3485 new_sock = -1;
3486 continue;
3487 }
3488
3489 rcu_thread_online();
3490
3491 switch (msg.component) {
3492 case LTTNG_HEALTH_CMD:
3493 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3494 break;
3495 case LTTNG_HEALTH_APP_MANAGE:
3496 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3497 break;
3498 case LTTNG_HEALTH_APP_REG:
3499 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3500 break;
3501 case LTTNG_HEALTH_KERNEL:
3502 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3503 break;
3504 case LTTNG_HEALTH_CONSUMER:
3505 reply.ret_code = check_consumer_health();
3506 break;
3507 case LTTNG_HEALTH_HT_CLEANUP:
3508 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3509 break;
3510 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3511 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3512 break;
3513 case LTTNG_HEALTH_APP_REG_DISPATCH:
3514 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3515 break;
3516 case LTTNG_HEALTH_ALL:
3517 reply.ret_code =
3518 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3519 health_check_state(HEALTH_TYPE_APP_REG) &&
3520 health_check_state(HEALTH_TYPE_CMD) &&
3521 health_check_state(HEALTH_TYPE_KERNEL) &&
3522 check_consumer_health() &&
3523 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3524 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3525 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3526 break;
3527 default:
3528 reply.ret_code = LTTNG_ERR_UND;
3529 break;
3530 }
3531
3532 /*
3533 * Flip ret value since 0 is a success and 1 indicates a bad health for
3534 * the client where in the sessiond it is the opposite. Again, this is
3535 * just to make things easier for us poor developer which enjoy a lot
3536 * lazyness.
3537 */
3538 if (reply.ret_code == 0 || reply.ret_code == 1) {
3539 reply.ret_code = !reply.ret_code;
3540 }
3541
3542 DBG2("Health check return value %d", reply.ret_code);
3543
3544 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3545 if (ret < 0) {
3546 ERR("Failed to send health data back to client");
3547 }
3548
3549 /* End of transmission */
3550 ret = close(new_sock);
3551 if (ret) {
3552 PERROR("close");
3553 }
3554 new_sock = -1;
3555 }
3556
3557 exit:
3558 error:
3559 if (err) {
3560 ERR("Health error occurred in %s", __func__);
3561 }
3562 DBG("Health check thread dying");
3563 unlink(health_unix_sock_path);
3564 if (sock >= 0) {
3565 ret = close(sock);
3566 if (ret) {
3567 PERROR("close");
3568 }
3569 }
3570
3571 lttng_poll_clean(&events);
3572
3573 rcu_unregister_thread();
3574 return NULL;
3575 }
3576
3577 /*
3578 * This thread manage all clients request using the unix client socket for
3579 * communication.
3580 */
3581 static void *thread_manage_clients(void *data)
3582 {
3583 int sock = -1, ret, i, pollfd, err = -1;
3584 int sock_error;
3585 uint32_t revents, nb_fd;
3586 struct command_ctx *cmd_ctx = NULL;
3587 struct lttng_poll_event events;
3588
3589 DBG("[thread] Manage client started");
3590
3591 rcu_register_thread();
3592
3593 health_register(HEALTH_TYPE_CMD);
3594
3595 if (testpoint(thread_manage_clients)) {
3596 goto error_testpoint;
3597 }
3598
3599 health_code_update();
3600
3601 ret = lttcomm_listen_unix_sock(client_sock);
3602 if (ret < 0) {
3603 goto error_listen;
3604 }
3605
3606 /*
3607 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3608 * more will be added to this poll set.
3609 */
3610 ret = sessiond_set_thread_pollset(&events, 2);
3611 if (ret < 0) {
3612 goto error_create_poll;
3613 }
3614
3615 /* Add the application registration socket */
3616 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3617 if (ret < 0) {
3618 goto error;
3619 }
3620
3621 /*
3622 * Notify parent pid that we are ready to accept command for client side.
3623 */
3624 if (opt_sig_parent) {
3625 kill(ppid, SIGUSR1);
3626 }
3627
3628 if (testpoint(thread_manage_clients_before_loop)) {
3629 goto error;
3630 }
3631
3632 health_code_update();
3633
3634 while (1) {
3635 DBG("Accepting client command ...");
3636
3637 /* Inifinite blocking call, waiting for transmission */
3638 restart:
3639 health_poll_entry();
3640 ret = lttng_poll_wait(&events, -1);
3641 health_poll_exit();
3642 if (ret < 0) {
3643 /*
3644 * Restart interrupted system call.
3645 */
3646 if (errno == EINTR) {
3647 goto restart;
3648 }
3649 goto error;
3650 }
3651
3652 nb_fd = ret;
3653
3654 for (i = 0; i < nb_fd; i++) {
3655 /* Fetch once the poll data */
3656 revents = LTTNG_POLL_GETEV(&events, i);
3657 pollfd = LTTNG_POLL_GETFD(&events, i);
3658
3659 health_code_update();
3660
3661 /* Thread quit pipe has been closed. Killing thread. */
3662 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3663 if (ret) {
3664 err = 0;
3665 goto exit;
3666 }
3667
3668 /* Event on the registration socket */
3669 if (pollfd == client_sock) {
3670 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3671 ERR("Client socket poll error");
3672 goto error;
3673 }
3674 }
3675 }
3676
3677 DBG("Wait for client response");
3678
3679 health_code_update();
3680
3681 sock = lttcomm_accept_unix_sock(client_sock);
3682 if (sock < 0) {
3683 goto error;
3684 }
3685
3686 /*
3687 * Set the CLOEXEC flag. Return code is useless because either way, the
3688 * show must go on.
3689 */
3690 (void) utils_set_fd_cloexec(sock);
3691
3692 /* Set socket option for credentials retrieval */
3693 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3694 if (ret < 0) {
3695 goto error;
3696 }
3697
3698 /* Allocate context command to process the client request */
3699 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3700 if (cmd_ctx == NULL) {
3701 PERROR("zmalloc cmd_ctx");
3702 goto error;
3703 }
3704
3705 /* Allocate data buffer for reception */
3706 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3707 if (cmd_ctx->lsm == NULL) {
3708 PERROR("zmalloc cmd_ctx->lsm");
3709 goto error;
3710 }
3711
3712 cmd_ctx->llm = NULL;
3713 cmd_ctx->session = NULL;
3714
3715 health_code_update();
3716
3717 /*
3718 * Data is received from the lttng client. The struct
3719 * lttcomm_session_msg (lsm) contains the command and data request of
3720 * the client.
3721 */
3722 DBG("Receiving data from client ...");
3723 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3724 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3725 if (ret <= 0) {
3726 DBG("Nothing recv() from client... continuing");
3727 ret = close(sock);
3728 if (ret) {
3729 PERROR("close");
3730 }
3731 sock = -1;
3732 clean_command_ctx(&cmd_ctx);
3733 continue;
3734 }
3735
3736 health_code_update();
3737
3738 // TODO: Validate cmd_ctx including sanity check for
3739 // security purpose.
3740
3741 rcu_thread_online();
3742 /*
3743 * This function dispatch the work to the kernel or userspace tracer
3744 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3745 * informations for the client. The command context struct contains
3746 * everything this function may needs.
3747 */
3748 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3749 rcu_thread_offline();
3750 if (ret < 0) {
3751 ret = close(sock);
3752 if (ret) {
3753 PERROR("close");
3754 }
3755 sock = -1;
3756 /*
3757 * TODO: Inform client somehow of the fatal error. At
3758 * this point, ret < 0 means that a zmalloc failed
3759 * (ENOMEM). Error detected but still accept
3760 * command, unless a socket error has been
3761 * detected.
3762 */
3763 clean_command_ctx(&cmd_ctx);
3764 continue;
3765 }
3766
3767 health_code_update();
3768
3769 DBG("Sending response (size: %d, retcode: %s)",
3770 cmd_ctx->lttng_msg_size,
3771 lttng_strerror(-cmd_ctx->llm->ret_code));
3772 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3773 if (ret < 0) {
3774 ERR("Failed to send data back to client");
3775 }
3776
3777 /* End of transmission */
3778 ret = close(sock);
3779 if (ret) {
3780 PERROR("close");
3781 }
3782 sock = -1;
3783
3784 clean_command_ctx(&cmd_ctx);
3785
3786 health_code_update();
3787 }
3788
3789 exit:
3790 error:
3791 if (sock >= 0) {
3792 ret = close(sock);
3793 if (ret) {
3794 PERROR("close");
3795 }
3796 }
3797
3798 lttng_poll_clean(&events);
3799 clean_command_ctx(&cmd_ctx);
3800
3801 error_listen:
3802 error_create_poll:
3803 error_testpoint:
3804 unlink(client_unix_sock_path);
3805 if (client_sock >= 0) {
3806 ret = close(client_sock);
3807 if (ret) {
3808 PERROR("close");
3809 }
3810 }
3811
3812 if (err) {
3813 health_error();
3814 ERR("Health error occurred in %s", __func__);
3815 }
3816
3817 health_unregister();
3818
3819 DBG("Client thread dying");
3820
3821 rcu_unregister_thread();
3822 return NULL;
3823 }
3824
3825
3826 /*
3827 * usage function on stderr
3828 */
3829 static void usage(void)
3830 {
3831 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3832 fprintf(stderr, " -h, --help Display this usage.\n");
3833 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3834 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3835 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3836 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3837 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3838 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3839 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3840 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3841 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3842 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3843 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3844 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3845 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3846 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3847 fprintf(stderr, " -V, --version Show version number.\n");
3848 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3849 fprintf(stderr, " -q, --quiet No output at all.\n");
3850 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3851 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3852 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3853 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3854 }
3855
3856 /*
3857 * daemon argument parsing
3858 */
3859 static int parse_args(int argc, char **argv)
3860 {
3861 int c;
3862
3863 static struct option long_options[] = {
3864 { "client-sock", 1, 0, 'c' },
3865 { "apps-sock", 1, 0, 'a' },
3866 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3867 { "kconsumerd-err-sock", 1, 0, 'E' },
3868 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3869 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3870 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3871 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3872 { "consumerd32-path", 1, 0, 'u' },
3873 { "consumerd32-libdir", 1, 0, 'U' },
3874 { "consumerd64-path", 1, 0, 't' },
3875 { "consumerd64-libdir", 1, 0, 'T' },
3876 { "daemonize", 0, 0, 'd' },
3877 { "sig-parent", 0, 0, 'S' },
3878 { "help", 0, 0, 'h' },
3879 { "group", 1, 0, 'g' },
3880 { "version", 0, 0, 'V' },
3881 { "quiet", 0, 0, 'q' },
3882 { "verbose", 0, 0, 'v' },
3883 { "verbose-consumer", 0, 0, 'Z' },
3884 { "no-kernel", 0, 0, 'N' },
3885 { "pidfile", 1, 0, 'p' },
3886 { NULL, 0, 0, 0 }
3887 };
3888
3889 while (1) {
3890 int option_index = 0;
3891 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3892 long_options, &option_index);
3893 if (c == -1) {
3894 break;
3895 }
3896
3897 switch (c) {
3898 case 0:
3899 fprintf(stderr, "option %s", long_options[option_index].name);
3900 if (optarg) {
3901 fprintf(stderr, " with arg %s\n", optarg);
3902 }
3903 break;
3904 case 'c':
3905 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3906 break;
3907 case 'a':
3908 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3909 break;
3910 case 'd':
3911 opt_daemon = 1;
3912 break;
3913 case 'g':
3914 opt_tracing_group = optarg;
3915 break;
3916 case 'h':
3917 usage();
3918 exit(EXIT_FAILURE);
3919 case 'V':
3920 fprintf(stdout, "%s\n", VERSION);
3921 exit(EXIT_SUCCESS);
3922 case 'S':
3923 opt_sig_parent = 1;
3924 break;
3925 case 'E':
3926 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3927 break;
3928 case 'C':
3929 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3930 break;
3931 case 'F':
3932 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3933 break;
3934 case 'D':
3935 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3936 break;
3937 case 'H':
3938 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3939 break;
3940 case 'G':
3941 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3942 break;
3943 case 'N':
3944 opt_no_kernel = 1;
3945 break;
3946 case 'q':
3947 lttng_opt_quiet = 1;
3948 break;
3949 case 'v':
3950 /* Verbose level can increase using multiple -v */
3951 lttng_opt_verbose += 1;
3952 break;
3953 case 'Z':
3954 opt_verbose_consumer += 1;
3955 break;
3956 case 'u':
3957 consumerd32_bin= optarg;
3958 break;
3959 case 'U':
3960 consumerd32_libdir = optarg;
3961 break;
3962 case 't':
3963 consumerd64_bin = optarg;
3964 break;
3965 case 'T':
3966 consumerd64_libdir = optarg;
3967 break;
3968 case 'p':
3969 opt_pidfile = optarg;
3970 break;
3971 default:
3972 /* Unknown option or other error.
3973 * Error is printed by getopt, just return */
3974 return -1;
3975 }
3976 }
3977
3978 return 0;
3979 }
3980
3981 /*
3982 * Creates the two needed socket by the daemon.
3983 * apps_sock - The communication socket for all UST apps.
3984 * client_sock - The communication of the cli tool (lttng).
3985 */
3986 static int init_daemon_socket(void)
3987 {
3988 int ret = 0;
3989 mode_t old_umask;
3990
3991 old_umask = umask(0);
3992
3993 /* Create client tool unix socket */
3994 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3995 if (client_sock < 0) {
3996 ERR("Create unix sock failed: %s", client_unix_sock_path);
3997 ret = -1;
3998 goto end;
3999 }
4000
4001 /* Set the cloexec flag */
4002 ret = utils_set_fd_cloexec(client_sock);
4003 if (ret < 0) {
4004 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4005 "Continuing but note that the consumer daemon will have a "
4006 "reference to this socket on exec()", client_sock);
4007 }
4008
4009 /* File permission MUST be 660 */
4010 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4011 if (ret < 0) {
4012 ERR("Set file permissions failed: %s", client_unix_sock_path);
4013 PERROR("chmod");
4014 goto end;
4015 }
4016
4017 /* Create the application unix socket */
4018 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4019 if (apps_sock < 0) {
4020 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4021 ret = -1;
4022 goto end;
4023 }
4024
4025 /* Set the cloexec flag */
4026 ret = utils_set_fd_cloexec(apps_sock);
4027 if (ret < 0) {
4028 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4029 "Continuing but note that the consumer daemon will have a "
4030 "reference to this socket on exec()", apps_sock);
4031 }
4032
4033 /* File permission MUST be 666 */
4034 ret = chmod(apps_unix_sock_path,
4035 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4036 if (ret < 0) {
4037 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4038 PERROR("chmod");
4039 goto end;
4040 }
4041
4042 DBG3("Session daemon client socket %d and application socket %d created",
4043 client_sock, apps_sock);
4044
4045 end:
4046 umask(old_umask);
4047 return ret;
4048 }
4049
4050 /*
4051 * Check if the global socket is available, and if a daemon is answering at the
4052 * other side. If yes, error is returned.
4053 */
4054 static int check_existing_daemon(void)
4055 {
4056 /* Is there anybody out there ? */
4057 if (lttng_session_daemon_alive()) {
4058 return -EEXIST;
4059 }
4060
4061 return 0;
4062 }
4063
4064 /*
4065 * Set the tracing group gid onto the client socket.
4066 *
4067 * Race window between mkdir and chown is OK because we are going from more
4068 * permissive (root.root) to less permissive (root.tracing).
4069 */
4070 static int set_permissions(char *rundir)
4071 {
4072 int ret;
4073 gid_t gid;
4074
4075 ret = allowed_group();
4076 if (ret < 0) {
4077 WARN("No tracing group detected");
4078 /* Setting gid to 0 if no tracing group is found */
4079 gid = 0;
4080 } else {
4081 gid = ret;
4082 }
4083
4084 /* Set lttng run dir */
4085 ret = chown(rundir, 0, gid);
4086 if (ret < 0) {
4087 ERR("Unable to set group on %s", rundir);
4088 PERROR("chown");
4089 }
4090
4091 /* Ensure all applications and tracing group can search the run dir */
4092 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
4093 if (ret < 0) {
4094 ERR("Unable to set permissions on %s", rundir);
4095 PERROR("chmod");
4096 }
4097
4098 /* lttng client socket path */
4099 ret = chown(client_unix_sock_path, 0, gid);
4100 if (ret < 0) {
4101 ERR("Unable to set group on %s", client_unix_sock_path);
4102 PERROR("chown");
4103 }
4104
4105 /* kconsumer error socket path */
4106 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4107 if (ret < 0) {
4108 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4109 PERROR("chown");
4110 }
4111
4112 /* 64-bit ustconsumer error socket path */
4113 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4114 if (ret < 0) {
4115 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4116 PERROR("chown");
4117 }
4118
4119 /* 32-bit ustconsumer compat32 error socket path */
4120 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4121 if (ret < 0) {
4122 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4123 PERROR("chown");
4124 }
4125
4126 DBG("All permissions are set");
4127
4128 return ret;
4129 }
4130
4131 /*
4132 * Create the lttng run directory needed for all global sockets and pipe.
4133 */
4134 static int create_lttng_rundir(const char *rundir)
4135 {
4136 int ret;
4137
4138 DBG3("Creating LTTng run directory: %s", rundir);
4139
4140 ret = mkdir(rundir, S_IRWXU);
4141 if (ret < 0) {
4142 if (errno != EEXIST) {
4143 ERR("Unable to create %s", rundir);
4144 goto error;
4145 } else {
4146 ret = 0;
4147 }
4148 }
4149
4150 error:
4151 return ret;
4152 }
4153
4154 /*
4155 * Setup sockets and directory needed by the kconsumerd communication with the
4156 * session daemon.
4157 */
4158 static int set_consumer_sockets(struct consumer_data *consumer_data,
4159 const char *rundir)
4160 {
4161 int ret;
4162 char path[PATH_MAX];
4163
4164 switch (consumer_data->type) {
4165 case LTTNG_CONSUMER_KERNEL:
4166 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4167 break;
4168 case LTTNG_CONSUMER64_UST:
4169 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4170 break;
4171 case LTTNG_CONSUMER32_UST:
4172 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4173 break;
4174 default:
4175 ERR("Consumer type unknown");
4176 ret = -EINVAL;
4177 goto error;
4178 }
4179
4180 DBG2("Creating consumer directory: %s", path);
4181
4182 ret = mkdir(path, S_IRWXU);
4183 if (ret < 0) {
4184 if (errno != EEXIST) {
4185 PERROR("mkdir");
4186 ERR("Failed to create %s", path);
4187 goto error;
4188 }
4189 ret = -1;
4190 }
4191
4192 /* Create the kconsumerd error unix socket */
4193 consumer_data->err_sock =
4194 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4195 if (consumer_data->err_sock < 0) {
4196 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4197 ret = -1;
4198 goto error;
4199 }
4200
4201 /*
4202 * Set the CLOEXEC flag. Return code is useless because either way, the
4203 * show must go on.
4204 */
4205 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4206 if (ret < 0) {
4207 PERROR("utils_set_fd_cloexec");
4208 /* continue anyway */
4209 }
4210
4211 /* File permission MUST be 660 */
4212 ret = chmod(consumer_data->err_unix_sock_path,
4213 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4214 if (ret < 0) {
4215 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4216 PERROR("chmod");
4217 goto error;
4218 }
4219
4220 error:
4221 return ret;
4222 }
4223
4224 /*
4225 * Signal handler for the daemon
4226 *
4227 * Simply stop all worker threads, leaving main() return gracefully after
4228 * joining all threads and calling cleanup().
4229 */
4230 static void sighandler(int sig)
4231 {
4232 switch (sig) {
4233 case SIGPIPE:
4234 DBG("SIGPIPE caught");
4235 return;
4236 case SIGINT:
4237 DBG("SIGINT caught");
4238 stop_threads();
4239 break;
4240 case SIGTERM:
4241 DBG("SIGTERM caught");
4242 stop_threads();
4243 break;
4244 default:
4245 break;
4246 }
4247 }
4248
4249 /*
4250 * Setup signal handler for :
4251 * SIGINT, SIGTERM, SIGPIPE
4252 */
4253 static int set_signal_handler(void)
4254 {
4255 int ret = 0;
4256 struct sigaction sa;
4257 sigset_t sigset;
4258
4259 if ((ret = sigemptyset(&sigset)) < 0) {
4260 PERROR("sigemptyset");
4261 return ret;
4262 }
4263
4264 sa.sa_handler = sighandler;
4265 sa.sa_mask = sigset;
4266 sa.sa_flags = 0;
4267 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4268 PERROR("sigaction");
4269 return ret;
4270 }
4271
4272 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4273 PERROR("sigaction");
4274 return ret;
4275 }
4276
4277 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4278 PERROR("sigaction");
4279 return ret;
4280 }
4281
4282 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4283
4284 return ret;
4285 }
4286
4287 /*
4288 * Set open files limit to unlimited. This daemon can open a large number of
4289 * file descriptors in order to consumer multiple kernel traces.
4290 */
4291 static void set_ulimit(void)
4292 {
4293 int ret;
4294 struct rlimit lim;
4295
4296 /* The kernel does not allowed an infinite limit for open files */
4297 lim.rlim_cur = 65535;
4298 lim.rlim_max = 65535;
4299
4300 ret = setrlimit(RLIMIT_NOFILE, &lim);
4301 if (ret < 0) {
4302 PERROR("failed to set open files limit");
4303 }
4304 }
4305
4306 /*
4307 * Write pidfile using the rundir and opt_pidfile.
4308 */
4309 static void write_pidfile(void)
4310 {
4311 int ret;
4312 char pidfile_path[PATH_MAX];
4313
4314 assert(rundir);
4315
4316 if (opt_pidfile) {
4317 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4318 } else {
4319 /* Build pidfile path from rundir and opt_pidfile. */
4320 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4321 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4322 if (ret < 0) {
4323 PERROR("snprintf pidfile path");
4324 goto error;
4325 }
4326 }
4327
4328 /*
4329 * Create pid file in rundir. Return value is of no importance. The
4330 * execution will continue even though we are not able to write the file.
4331 */
4332 (void) utils_create_pid_file(getpid(), pidfile_path);
4333
4334 error:
4335 return;
4336 }
4337
4338 /*
4339 * main
4340 */
4341 int main(int argc, char **argv)
4342 {
4343 int ret = 0;
4344 void *status;
4345 const char *home_path, *env_app_timeout;
4346
4347 init_kernel_workarounds();
4348
4349 rcu_register_thread();
4350
4351 setup_consumerd_path();
4352
4353 page_size = sysconf(_SC_PAGESIZE);
4354 if (page_size < 0) {
4355 PERROR("sysconf _SC_PAGESIZE");
4356 page_size = LONG_MAX;
4357 WARN("Fallback page size to %ld", page_size);
4358 }
4359
4360 /* Parse arguments */
4361 progname = argv[0];
4362 if ((ret = parse_args(argc, argv)) < 0) {
4363 goto error;
4364 }
4365
4366 /* Daemonize */
4367 if (opt_daemon) {
4368 int i;
4369
4370 /*
4371 * fork
4372 * child: setsid, close FD 0, 1, 2, chdir /
4373 * parent: exit (if fork is successful)
4374 */
4375 ret = daemon(0, 0);
4376 if (ret < 0) {
4377 PERROR("daemon");
4378 goto error;
4379 }
4380 /*
4381 * We are in the child. Make sure all other file
4382 * descriptors are closed, in case we are called with
4383 * more opened file descriptors than the standard ones.
4384 */
4385 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4386 (void) close(i);
4387 }
4388 }
4389
4390 /* Create thread quit pipe */
4391 if ((ret = init_thread_quit_pipe()) < 0) {
4392 goto error;
4393 }
4394
4395 /* Check if daemon is UID = 0 */
4396 is_root = !getuid();
4397
4398 if (is_root) {
4399 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4400
4401 /* Create global run dir with root access */
4402 ret = create_lttng_rundir(rundir);
4403 if (ret < 0) {
4404 goto error;
4405 }
4406
4407 if (strlen(apps_unix_sock_path) == 0) {
4408 snprintf(apps_unix_sock_path, PATH_MAX,
4409 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4410 }
4411
4412 if (strlen(client_unix_sock_path) == 0) {
4413 snprintf(client_unix_sock_path, PATH_MAX,
4414 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4415 }
4416
4417 /* Set global SHM for ust */
4418 if (strlen(wait_shm_path) == 0) {
4419 snprintf(wait_shm_path, PATH_MAX,
4420 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4421 }
4422
4423 if (strlen(health_unix_sock_path) == 0) {
4424 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4425 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4426 }
4427
4428 /* Setup kernel consumerd path */
4429 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4430 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4431 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4432 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4433
4434 DBG2("Kernel consumer err path: %s",
4435 kconsumer_data.err_unix_sock_path);
4436 DBG2("Kernel consumer cmd path: %s",
4437 kconsumer_data.cmd_unix_sock_path);
4438 } else {
4439 home_path = utils_get_home_dir();
4440 if (home_path == NULL) {
4441 /* TODO: Add --socket PATH option */
4442 ERR("Can't get HOME directory for sockets creation.");
4443 ret = -EPERM;
4444 goto error;
4445 }
4446
4447 /*
4448 * Create rundir from home path. This will create something like
4449 * $HOME/.lttng
4450 */
4451 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4452 if (ret < 0) {
4453 ret = -ENOMEM;
4454 goto error;
4455 }
4456
4457 ret = create_lttng_rundir(rundir);
4458 if (ret < 0) {
4459 goto error;
4460 }
4461
4462 if (strlen(apps_unix_sock_path) == 0) {
4463 snprintf(apps_unix_sock_path, PATH_MAX,
4464 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4465 }
4466
4467 /* Set the cli tool unix socket path */
4468 if (strlen(client_unix_sock_path) == 0) {
4469 snprintf(client_unix_sock_path, PATH_MAX,
4470 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4471 }
4472
4473 /* Set global SHM for ust */
4474 if (strlen(wait_shm_path) == 0) {
4475 snprintf(wait_shm_path, PATH_MAX,
4476 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4477 }
4478
4479 /* Set health check Unix path */
4480 if (strlen(health_unix_sock_path) == 0) {
4481 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4482 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4483 }
4484 }
4485
4486 /* Set consumer initial state */
4487 kernel_consumerd_state = CONSUMER_STOPPED;
4488 ust_consumerd_state = CONSUMER_STOPPED;
4489
4490 DBG("Client socket path %s", client_unix_sock_path);
4491 DBG("Application socket path %s", apps_unix_sock_path);
4492 DBG("Application wait path %s", wait_shm_path);
4493 DBG("LTTng run directory path: %s", rundir);
4494
4495 /* 32 bits consumerd path setup */
4496 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4497 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4498 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4499 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4500
4501 DBG2("UST consumer 32 bits err path: %s",
4502 ustconsumer32_data.err_unix_sock_path);
4503 DBG2("UST consumer 32 bits cmd path: %s",
4504 ustconsumer32_data.cmd_unix_sock_path);
4505
4506 /* 64 bits consumerd path setup */
4507 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4508 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4509 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4510 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4511
4512 DBG2("UST consumer 64 bits err path: %s",
4513 ustconsumer64_data.err_unix_sock_path);
4514 DBG2("UST consumer 64 bits cmd path: %s",
4515 ustconsumer64_data.cmd_unix_sock_path);
4516
4517 /*
4518 * See if daemon already exist.
4519 */
4520 if ((ret = check_existing_daemon()) < 0) {
4521 ERR("Already running daemon.\n");
4522 /*
4523 * We do not goto exit because we must not cleanup()
4524 * because a daemon is already running.
4525 */
4526 goto error;
4527 }
4528
4529 /*
4530 * Init UST app hash table. Alloc hash table before this point since
4531 * cleanup() can get called after that point.
4532 */
4533 ust_app_ht_alloc();
4534
4535 /* After this point, we can safely call cleanup() with "goto exit" */
4536
4537 /*
4538 * These actions must be executed as root. We do that *after* setting up
4539 * the sockets path because we MUST make the check for another daemon using
4540 * those paths *before* trying to set the kernel consumer sockets and init
4541 * kernel tracer.
4542 */
4543 if (is_root) {
4544 ret = set_consumer_sockets(&kconsumer_data, rundir);
4545 if (ret < 0) {
4546 goto exit;
4547 }
4548
4549 /* Setup kernel tracer */
4550 if (!opt_no_kernel) {
4551 init_kernel_tracer();
4552 }
4553
4554 /* Set ulimit for open files */
4555 set_ulimit();
4556 }
4557 /* init lttng_fd tracking must be done after set_ulimit. */
4558 lttng_fd_init();
4559
4560 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4561 if (ret < 0) {
4562 goto exit;
4563 }
4564
4565 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4566 if (ret < 0) {
4567 goto exit;
4568 }
4569
4570 if ((ret = set_signal_handler()) < 0) {
4571 goto exit;
4572 }
4573
4574 /* Setup the needed unix socket */
4575 if ((ret = init_daemon_socket()) < 0) {
4576 goto exit;
4577 }
4578
4579 /* Set credentials to socket */
4580 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4581 goto exit;
4582 }
4583
4584 /* Get parent pid if -S, --sig-parent is specified. */
4585 if (opt_sig_parent) {
4586 ppid = getppid();
4587 }
4588
4589 /* Setup the kernel pipe for waking up the kernel thread */
4590 if (is_root && !opt_no_kernel) {
4591 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4592 goto exit;
4593 }
4594 }
4595
4596 /* Setup the thread ht_cleanup communication pipe. */
4597 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4598 goto exit;
4599 }
4600
4601 /* Setup the thread apps communication pipe. */
4602 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4603 goto exit;
4604 }
4605
4606 /* Setup the thread apps notify communication pipe. */
4607 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4608 goto exit;
4609 }
4610
4611 /* Initialize global buffer per UID and PID registry. */
4612 buffer_reg_init_uid_registry();
4613 buffer_reg_init_pid_registry();
4614
4615 /* Init UST command queue. */
4616 cds_wfq_init(&ust_cmd_queue.queue);
4617
4618 /*
4619 * Get session list pointer. This pointer MUST NOT be free(). This list is
4620 * statically declared in session.c
4621 */
4622 session_list_ptr = session_get_list();
4623
4624 /* Set up max poll set size */
4625 lttng_poll_set_max_size();
4626
4627 cmd_init();
4628
4629 /* Check for the application socket timeout env variable. */
4630 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4631 if (env_app_timeout) {
4632 app_socket_timeout = atoi(env_app_timeout);
4633 } else {
4634 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4635 }
4636
4637 write_pidfile();
4638
4639 /* Initialize communication library */
4640 lttcomm_init();
4641
4642 /* Create thread to manage the client socket */
4643 ret = pthread_create(&ht_cleanup_thread, NULL,
4644 thread_ht_cleanup, (void *) NULL);
4645 if (ret != 0) {
4646 PERROR("pthread_create ht_cleanup");
4647 goto exit_ht_cleanup;
4648 }
4649
4650 /* Create thread to manage the client socket */
4651 ret = pthread_create(&health_thread, NULL,
4652 thread_manage_health, (void *) NULL);
4653 if (ret != 0) {
4654 PERROR("pthread_create health");
4655 goto exit_health;
4656 }
4657
4658 /* Create thread to manage the client socket */
4659 ret = pthread_create(&client_thread, NULL,
4660 thread_manage_clients, (void *) NULL);
4661 if (ret != 0) {
4662 PERROR("pthread_create clients");
4663 goto exit_client;
4664 }
4665
4666 /* Create thread to dispatch registration */
4667 ret = pthread_create(&dispatch_thread, NULL,
4668 thread_dispatch_ust_registration, (void *) NULL);
4669 if (ret != 0) {
4670 PERROR("pthread_create dispatch");
4671 goto exit_dispatch;
4672 }
4673
4674 /* Create thread to manage application registration. */
4675 ret = pthread_create(&reg_apps_thread, NULL,
4676 thread_registration_apps, (void *) NULL);
4677 if (ret != 0) {
4678 PERROR("pthread_create registration");
4679 goto exit_reg_apps;
4680 }
4681
4682 /* Create thread to manage application socket */
4683 ret = pthread_create(&apps_thread, NULL,
4684 thread_manage_apps, (void *) NULL);
4685 if (ret != 0) {
4686 PERROR("pthread_create apps");
4687 goto exit_apps;
4688 }
4689
4690 /* Create thread to manage application notify socket */
4691 ret = pthread_create(&apps_notify_thread, NULL,
4692 ust_thread_manage_notify, (void *) NULL);
4693 if (ret != 0) {
4694 PERROR("pthread_create apps");
4695 goto exit_apps;
4696 }
4697
4698 /* Don't start this thread if kernel tracing is not requested nor root */
4699 if (is_root && !opt_no_kernel) {
4700 /* Create kernel thread to manage kernel event */
4701 ret = pthread_create(&kernel_thread, NULL,
4702 thread_manage_kernel, (void *) NULL);
4703 if (ret != 0) {
4704 PERROR("pthread_create kernel");
4705 goto exit_kernel;
4706 }
4707
4708 ret = pthread_join(kernel_thread, &status);
4709 if (ret != 0) {
4710 PERROR("pthread_join");
4711 goto error; /* join error, exit without cleanup */
4712 }
4713 }
4714
4715 exit_kernel:
4716 ret = pthread_join(apps_thread, &status);
4717 if (ret != 0) {
4718 PERROR("pthread_join");
4719 goto error; /* join error, exit without cleanup */
4720 }
4721
4722 exit_apps:
4723 ret = pthread_join(reg_apps_thread, &status);
4724 if (ret != 0) {
4725 PERROR("pthread_join");
4726 goto error; /* join error, exit without cleanup */
4727 }
4728
4729 exit_reg_apps:
4730 ret = pthread_join(dispatch_thread, &status);
4731 if (ret != 0) {
4732 PERROR("pthread_join");
4733 goto error; /* join error, exit without cleanup */
4734 }
4735
4736 exit_dispatch:
4737 ret = pthread_join(client_thread, &status);
4738 if (ret != 0) {
4739 PERROR("pthread_join");
4740 goto error; /* join error, exit without cleanup */
4741 }
4742
4743 ret = join_consumer_thread(&kconsumer_data);
4744 if (ret != 0) {
4745 PERROR("join_consumer");
4746 goto error; /* join error, exit without cleanup */
4747 }
4748
4749 ret = join_consumer_thread(&ustconsumer32_data);
4750 if (ret != 0) {
4751 PERROR("join_consumer ust32");
4752 goto error; /* join error, exit without cleanup */
4753 }
4754
4755 ret = join_consumer_thread(&ustconsumer64_data);
4756 if (ret != 0) {
4757 PERROR("join_consumer ust64");
4758 goto error; /* join error, exit without cleanup */
4759 }
4760
4761 exit_client:
4762 ret = pthread_join(health_thread, &status);
4763 if (ret != 0) {
4764 PERROR("pthread_join health thread");
4765 goto error; /* join error, exit without cleanup */
4766 }
4767
4768 exit_health:
4769 ret = pthread_join(ht_cleanup_thread, &status);
4770 if (ret != 0) {
4771 PERROR("pthread_join ht cleanup thread");
4772 goto error; /* join error, exit without cleanup */
4773 }
4774 exit_ht_cleanup:
4775 exit:
4776 /*
4777 * cleanup() is called when no other thread is running.
4778 */
4779 rcu_thread_online();
4780 cleanup();
4781 rcu_thread_offline();
4782 rcu_unregister_thread();
4783 if (!ret) {
4784 exit(EXIT_SUCCESS);
4785 }
4786 error:
4787 exit(EXIT_FAILURE);
4788 }
This page took 0.175078 seconds and 5 git commands to generate.