Rename consumer socket fd to fd_ptr
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <inttypes.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <config.h>
39
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
50 #include "channel.h"
51 #include "cmd.h"
52 #include "consumer.h"
53 #include "context.h"
54 #include "event.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "modprobe.h"
58 #include "shm.h"
59 #include "ust-ctl.h"
60 #include "ust-consumer.h"
61 #include "utils.h"
62 #include "fd-limit.h"
63 #include "health.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66
67 #define CONSUMERD_FILE "lttng-consumerd"
68
69 /* Const values */
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71
72 const char *progname;
73 const char *opt_tracing_group;
74 static const char *opt_pidfile;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 int apps_cmd_notify_pipe[2] = { -1, -1 };
152
153 /* Pthread, Mutexes and Semaphores */
154 static pthread_t apps_thread;
155 static pthread_t apps_notify_thread;
156 static pthread_t reg_apps_thread;
157 static pthread_t client_thread;
158 static pthread_t kernel_thread;
159 static pthread_t dispatch_thread;
160 static pthread_t health_thread;
161 static pthread_t ht_cleanup_thread;
162
163 /*
164 * UST registration command queue. This queue is tied with a futex and uses a N
165 * wakers / 1 waiter implemented and detailed in futex.c/.h
166 *
167 * The thread_manage_apps and thread_dispatch_ust_registration interact with
168 * this queue and the wait/wake scheme.
169 */
170 static struct ust_cmd_queue ust_cmd_queue;
171
172 /*
173 * Pointer initialized before thread creation.
174 *
175 * This points to the tracing session list containing the session count and a
176 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
177 * MUST NOT be taken if you call a public function in session.c.
178 *
179 * The lock is nested inside the structure: session_list_ptr->lock. Please use
180 * session_lock_list and session_unlock_list for lock acquisition.
181 */
182 static struct ltt_session_list *session_list_ptr;
183
184 int ust_consumerd64_fd = -1;
185 int ust_consumerd32_fd = -1;
186
187 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
188 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
189 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
190 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
191
192 static const char *module_proc_lttng = "/proc/lttng";
193
194 /*
195 * Consumer daemon state which is changed when spawning it, killing it or in
196 * case of a fatal error.
197 */
198 enum consumerd_state {
199 CONSUMER_STARTED = 1,
200 CONSUMER_STOPPED = 2,
201 CONSUMER_ERROR = 3,
202 };
203
204 /*
205 * This consumer daemon state is used to validate if a client command will be
206 * able to reach the consumer. If not, the client is informed. For instance,
207 * doing a "lttng start" when the consumer state is set to ERROR will return an
208 * error to the client.
209 *
210 * The following example shows a possible race condition of this scheme:
211 *
212 * consumer thread error happens
213 * client cmd arrives
214 * client cmd checks state -> still OK
215 * consumer thread exit, sets error
216 * client cmd try to talk to consumer
217 * ...
218 *
219 * However, since the consumer is a different daemon, we have no way of making
220 * sure the command will reach it safely even with this state flag. This is why
221 * we consider that up to the state validation during command processing, the
222 * command is safe. After that, we can not guarantee the correctness of the
223 * client request vis-a-vis the consumer.
224 */
225 static enum consumerd_state ust_consumerd_state;
226 static enum consumerd_state kernel_consumerd_state;
227
228 /*
229 * Socket timeout for receiving and sending in seconds.
230 */
231 static int app_socket_timeout;
232
233 /* Set in main() with the current page size. */
234 long page_size;
235
236 static
237 void setup_consumerd_path(void)
238 {
239 const char *bin, *libdir;
240
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
245 */
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
256 }
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
259 }
260 #else
261 #error "Unknown bitness"
262 #endif
263
264 /*
265 * runtime env. var. overrides the build default.
266 */
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
270 }
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
274 }
275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
280 if (libdir) {
281 consumerd64_libdir = libdir;
282 }
283 }
284
285 /*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
289 {
290 int ret;
291
292 assert(events);
293
294 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
295 if (ret < 0) {
296 goto error;
297 }
298
299 /* Add quit pipe */
300 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
301 if (ret < 0) {
302 goto error;
303 }
304
305 return 0;
306
307 error:
308 return ret;
309 }
310
311 /*
312 * Check if the thread quit pipe was triggered.
313 *
314 * Return 1 if it was triggered else 0;
315 */
316 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
317 {
318 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
319 return 1;
320 }
321
322 return 0;
323 }
324
325 /*
326 * Return group ID of the tracing group or -1 if not found.
327 */
328 static gid_t allowed_group(void)
329 {
330 struct group *grp;
331
332 if (opt_tracing_group) {
333 grp = getgrnam(opt_tracing_group);
334 } else {
335 grp = getgrnam(default_tracing_group);
336 }
337 if (!grp) {
338 return -1;
339 } else {
340 return grp->gr_gid;
341 }
342 }
343
344 /*
345 * Init thread quit pipe.
346 *
347 * Return -1 on error or 0 if all pipes are created.
348 */
349 static int init_thread_quit_pipe(void)
350 {
351 int ret, i;
352
353 ret = pipe(thread_quit_pipe);
354 if (ret < 0) {
355 PERROR("thread quit pipe");
356 goto error;
357 }
358
359 for (i = 0; i < 2; i++) {
360 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
361 if (ret < 0) {
362 PERROR("fcntl");
363 goto error;
364 }
365 }
366
367 error:
368 return ret;
369 }
370
371 /*
372 * Stop all threads by closing the thread quit pipe.
373 */
374 static void stop_threads(void)
375 {
376 int ret;
377
378 /* Stopping all threads */
379 DBG("Terminating all threads");
380 ret = notify_thread_pipe(thread_quit_pipe[1]);
381 if (ret < 0) {
382 ERR("write error on thread quit pipe");
383 }
384
385 /* Dispatch thread */
386 CMM_STORE_SHARED(dispatch_thread_exit, 1);
387 futex_nto1_wake(&ust_cmd_queue.futex);
388 }
389
390 /*
391 * Close every consumer sockets.
392 */
393 static void close_consumer_sockets(void)
394 {
395 int ret;
396
397 if (kconsumer_data.err_sock >= 0) {
398 ret = close(kconsumer_data.err_sock);
399 if (ret < 0) {
400 PERROR("kernel consumer err_sock close");
401 }
402 }
403 if (ustconsumer32_data.err_sock >= 0) {
404 ret = close(ustconsumer32_data.err_sock);
405 if (ret < 0) {
406 PERROR("UST consumerd32 err_sock close");
407 }
408 }
409 if (ustconsumer64_data.err_sock >= 0) {
410 ret = close(ustconsumer64_data.err_sock);
411 if (ret < 0) {
412 PERROR("UST consumerd64 err_sock close");
413 }
414 }
415 if (kconsumer_data.cmd_sock >= 0) {
416 ret = close(kconsumer_data.cmd_sock);
417 if (ret < 0) {
418 PERROR("kernel consumer cmd_sock close");
419 }
420 }
421 if (ustconsumer32_data.cmd_sock >= 0) {
422 ret = close(ustconsumer32_data.cmd_sock);
423 if (ret < 0) {
424 PERROR("UST consumerd32 cmd_sock close");
425 }
426 }
427 if (ustconsumer64_data.cmd_sock >= 0) {
428 ret = close(ustconsumer64_data.cmd_sock);
429 if (ret < 0) {
430 PERROR("UST consumerd64 cmd_sock close");
431 }
432 }
433 }
434
435 /*
436 * Cleanup the daemon
437 */
438 static void cleanup(void)
439 {
440 int ret;
441 char *cmd = NULL;
442 struct ltt_session *sess, *stmp;
443
444 DBG("Cleaning up");
445
446 /*
447 * Close the thread quit pipe. It has already done its job,
448 * since we are now called.
449 */
450 utils_close_pipe(thread_quit_pipe);
451
452 /*
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
455 */
456 if (opt_pidfile) {
457 ret = remove(opt_pidfile);
458 if (ret < 0) {
459 PERROR("remove pidfile %s", opt_pidfile);
460 }
461 }
462
463 DBG("Removing %s directory", rundir);
464 ret = asprintf(&cmd, "rm -rf %s", rundir);
465 if (ret < 0) {
466 ERR("asprintf failed. Something is really wrong!");
467 }
468
469 /* Remove lttng run directory */
470 ret = system(cmd);
471 if (ret < 0) {
472 ERR("Unable to clean %s", rundir);
473 }
474 free(cmd);
475 free(rundir);
476
477 DBG("Cleaning up all sessions");
478
479 /* Destroy session list mutex */
480 if (session_list_ptr != NULL) {
481 pthread_mutex_destroy(&session_list_ptr->lock);
482
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess, stmp,
485 &session_list_ptr->head, list) {
486 cmd_destroy_session(sess, kernel_poll_pipe[1]);
487 }
488 }
489
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
493
494 if (is_root && !opt_no_kernel) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd >= 0) {
497 ret = close(kernel_tracer_fd);
498 if (ret) {
499 PERROR("close");
500 }
501 }
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
504 }
505
506 close_consumer_sockets();
507
508 /* <fun> */
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
512 /* </fun> */
513 }
514
515 /*
516 * Send data on a unix socket using the liblttsessiondcomm API.
517 *
518 * Return lttcomm error code.
519 */
520 static int send_unix_sock(int sock, void *buf, size_t len)
521 {
522 /* Check valid length */
523 if (len == 0) {
524 return -1;
525 }
526
527 return lttcomm_send_unix_sock(sock, buf, len);
528 }
529
530 /*
531 * Free memory of a command context structure.
532 */
533 static void clean_command_ctx(struct command_ctx **cmd_ctx)
534 {
535 DBG("Clean command context structure");
536 if (*cmd_ctx) {
537 if ((*cmd_ctx)->llm) {
538 free((*cmd_ctx)->llm);
539 }
540 if ((*cmd_ctx)->lsm) {
541 free((*cmd_ctx)->lsm);
542 }
543 free(*cmd_ctx);
544 *cmd_ctx = NULL;
545 }
546 }
547
548 /*
549 * Notify UST applications using the shm mmap futex.
550 */
551 static int notify_ust_apps(int active)
552 {
553 char *wait_shm_mmap;
554
555 DBG("Notifying applications of session daemon state: %d", active);
556
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
559 if (wait_shm_mmap == NULL) {
560 goto error;
561 }
562
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap, active);
565
566 /* Apps notified successfully */
567 return 0;
568
569 error:
570 return -1;
571 }
572
573 /*
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
576 * structure.
577 *
578 * Return total size of the buffer pointed by buf.
579 */
580 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
581 {
582 int ret, buf_size;
583
584 buf_size = size;
585
586 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
587 if (cmd_ctx->llm == NULL) {
588 PERROR("zmalloc");
589 ret = -ENOMEM;
590 goto error;
591 }
592
593 /* Copy common data */
594 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
595 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
596
597 cmd_ctx->llm->data_size = size;
598 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
599
600 return buf_size;
601
602 error:
603 return ret;
604 }
605
606 /*
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
609 */
610 static int update_kernel_poll(struct lttng_poll_event *events)
611 {
612 int ret;
613 struct ltt_session *session;
614 struct ltt_kernel_channel *channel;
615
616 DBG("Updating kernel poll set");
617
618 session_lock_list();
619 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
620 session_lock(session);
621 if (session->kernel_session == NULL) {
622 session_unlock(session);
623 continue;
624 }
625
626 cds_list_for_each_entry(channel,
627 &session->kernel_session->channel_list.head, list) {
628 /* Add channel fd to the kernel poll set */
629 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
630 if (ret < 0) {
631 session_unlock(session);
632 goto error;
633 }
634 DBG("Channel fd %d added to kernel set", channel->fd);
635 }
636 session_unlock(session);
637 }
638 session_unlock_list();
639
640 return 0;
641
642 error:
643 session_unlock_list();
644 return -1;
645 }
646
647 /*
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
650 *
651 * Useful for CPU hotplug feature.
652 */
653 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
654 {
655 int ret = 0;
656 struct ltt_session *session;
657 struct ltt_kernel_session *ksess;
658 struct ltt_kernel_channel *channel;
659
660 DBG("Updating kernel streams for channel fd %d", fd);
661
662 session_lock_list();
663 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
664 session_lock(session);
665 if (session->kernel_session == NULL) {
666 session_unlock(session);
667 continue;
668 }
669 ksess = session->kernel_session;
670
671 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
672 if (channel->fd == fd) {
673 DBG("Channel found, updating kernel streams");
674 ret = kernel_open_channel_stream(channel);
675 if (ret < 0) {
676 goto error;
677 }
678 /* Update the stream global counter */
679 ksess->stream_count_global += ret;
680
681 /*
682 * Have we already sent fds to the consumer? If yes, it means
683 * that tracing is started so it is safe to send our updated
684 * stream fds.
685 */
686 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
687 struct lttng_ht_iter iter;
688 struct consumer_socket *socket;
689
690 rcu_read_lock();
691 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
692 &iter.iter, socket, node.node) {
693 pthread_mutex_lock(socket->lock);
694 ret = kernel_consumer_send_channel_stream(socket,
695 channel, ksess,
696 session->output_traces ? 1 : 0);
697 pthread_mutex_unlock(socket->lock);
698 if (ret < 0) {
699 rcu_read_unlock();
700 goto error;
701 }
702 }
703 rcu_read_unlock();
704 }
705 goto error;
706 }
707 }
708 session_unlock(session);
709 }
710 session_unlock_list();
711 return ret;
712
713 error:
714 session_unlock(session);
715 session_unlock_list();
716 return ret;
717 }
718
719 /*
720 * For each tracing session, update newly registered apps. The session list
721 * lock MUST be acquired before calling this.
722 */
723 static void update_ust_app(int app_sock)
724 {
725 struct ltt_session *sess, *stmp;
726
727 /* Consumer is in an ERROR state. Stop any application update. */
728 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
729 /* Stop the update process since the consumer is dead. */
730 return;
731 }
732
733 /* For all tracing session(s) */
734 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
735 session_lock(sess);
736 if (sess->ust_session) {
737 ust_app_global_update(sess->ust_session, app_sock);
738 }
739 session_unlock(sess);
740 }
741 }
742
743 /*
744 * This thread manage event coming from the kernel.
745 *
746 * Features supported in this thread:
747 * -) CPU Hotplug
748 */
749 static void *thread_manage_kernel(void *data)
750 {
751 int ret, i, pollfd, update_poll_flag = 1, err = -1;
752 uint32_t revents, nb_fd;
753 char tmp;
754 struct lttng_poll_event events;
755
756 DBG("[thread] Thread manage kernel started");
757
758 health_register(HEALTH_TYPE_KERNEL);
759
760 /*
761 * This first step of the while is to clean this structure which could free
762 * non NULL pointers so initialize it before the loop.
763 */
764 lttng_poll_init(&events);
765
766 if (testpoint(thread_manage_kernel)) {
767 goto error_testpoint;
768 }
769
770 health_code_update();
771
772 if (testpoint(thread_manage_kernel_before_loop)) {
773 goto error_testpoint;
774 }
775
776 while (1) {
777 health_code_update();
778
779 if (update_poll_flag == 1) {
780 /* Clean events object. We are about to populate it again. */
781 lttng_poll_clean(&events);
782
783 ret = sessiond_set_thread_pollset(&events, 2);
784 if (ret < 0) {
785 goto error_poll_create;
786 }
787
788 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
789 if (ret < 0) {
790 goto error;
791 }
792
793 /* This will add the available kernel channel if any. */
794 ret = update_kernel_poll(&events);
795 if (ret < 0) {
796 goto error;
797 }
798 update_poll_flag = 0;
799 }
800
801 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
802
803 /* Poll infinite value of time */
804 restart:
805 health_poll_entry();
806 ret = lttng_poll_wait(&events, -1);
807 health_poll_exit();
808 if (ret < 0) {
809 /*
810 * Restart interrupted system call.
811 */
812 if (errno == EINTR) {
813 goto restart;
814 }
815 goto error;
816 } else if (ret == 0) {
817 /* Should not happen since timeout is infinite */
818 ERR("Return value of poll is 0 with an infinite timeout.\n"
819 "This should not have happened! Continuing...");
820 continue;
821 }
822
823 nb_fd = ret;
824
825 for (i = 0; i < nb_fd; i++) {
826 /* Fetch once the poll data */
827 revents = LTTNG_POLL_GETEV(&events, i);
828 pollfd = LTTNG_POLL_GETFD(&events, i);
829
830 health_code_update();
831
832 /* Thread quit pipe has been closed. Killing thread. */
833 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
834 if (ret) {
835 err = 0;
836 goto exit;
837 }
838
839 /* Check for data on kernel pipe */
840 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
841 do {
842 ret = read(kernel_poll_pipe[0], &tmp, 1);
843 } while (ret < 0 && errno == EINTR);
844 /*
845 * Ret value is useless here, if this pipe gets any actions an
846 * update is required anyway.
847 */
848 update_poll_flag = 1;
849 continue;
850 } else {
851 /*
852 * New CPU detected by the kernel. Adding kernel stream to
853 * kernel session and updating the kernel consumer
854 */
855 if (revents & LPOLLIN) {
856 ret = update_kernel_stream(&kconsumer_data, pollfd);
857 if (ret < 0) {
858 continue;
859 }
860 break;
861 /*
862 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
863 * and unregister kernel stream at this point.
864 */
865 }
866 }
867 }
868 }
869
870 exit:
871 error:
872 lttng_poll_clean(&events);
873 error_poll_create:
874 error_testpoint:
875 utils_close_pipe(kernel_poll_pipe);
876 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
877 if (err) {
878 health_error();
879 ERR("Health error occurred in %s", __func__);
880 WARN("Kernel thread died unexpectedly. "
881 "Kernel tracing can continue but CPU hotplug is disabled.");
882 }
883 health_unregister();
884 DBG("Kernel thread dying");
885 return NULL;
886 }
887
888 /*
889 * Signal pthread condition of the consumer data that the thread.
890 */
891 static void signal_consumer_condition(struct consumer_data *data, int state)
892 {
893 pthread_mutex_lock(&data->cond_mutex);
894
895 /*
896 * The state is set before signaling. It can be any value, it's the waiter
897 * job to correctly interpret this condition variable associated to the
898 * consumer pthread_cond.
899 *
900 * A value of 0 means that the corresponding thread of the consumer data
901 * was not started. 1 indicates that the thread has started and is ready
902 * for action. A negative value means that there was an error during the
903 * thread bootstrap.
904 */
905 data->consumer_thread_is_ready = state;
906 (void) pthread_cond_signal(&data->cond);
907
908 pthread_mutex_unlock(&data->cond_mutex);
909 }
910
911 /*
912 * This thread manage the consumer error sent back to the session daemon.
913 */
914 static void *thread_manage_consumer(void *data)
915 {
916 int sock = -1, i, ret, pollfd, err = -1;
917 uint32_t revents, nb_fd;
918 enum lttcomm_return_code code;
919 struct lttng_poll_event events;
920 struct consumer_data *consumer_data = data;
921
922 DBG("[thread] Manage consumer started");
923
924 health_register(HEALTH_TYPE_CONSUMER);
925
926 health_code_update();
927
928 /*
929 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
930 * metadata_sock. Nothing more will be added to this poll set.
931 */
932 ret = sessiond_set_thread_pollset(&events, 3);
933 if (ret < 0) {
934 goto error_poll;
935 }
936
937 /*
938 * The error socket here is already in a listening state which was done
939 * just before spawning this thread to avoid a race between the consumer
940 * daemon exec trying to connect and the listen() call.
941 */
942 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
943 if (ret < 0) {
944 goto error;
945 }
946
947 health_code_update();
948
949 /* Infinite blocking call, waiting for transmission */
950 restart:
951 health_poll_entry();
952
953 if (testpoint(thread_manage_consumer)) {
954 goto error;
955 }
956
957 ret = lttng_poll_wait(&events, -1);
958 health_poll_exit();
959 if (ret < 0) {
960 /*
961 * Restart interrupted system call.
962 */
963 if (errno == EINTR) {
964 goto restart;
965 }
966 goto error;
967 }
968
969 nb_fd = ret;
970
971 for (i = 0; i < nb_fd; i++) {
972 /* Fetch once the poll data */
973 revents = LTTNG_POLL_GETEV(&events, i);
974 pollfd = LTTNG_POLL_GETFD(&events, i);
975
976 health_code_update();
977
978 /* Thread quit pipe has been closed. Killing thread. */
979 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
980 if (ret) {
981 err = 0;
982 goto exit;
983 }
984
985 /* Event on the registration socket */
986 if (pollfd == consumer_data->err_sock) {
987 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
988 ERR("consumer err socket poll error");
989 goto error;
990 }
991 }
992 }
993
994 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
995 if (sock < 0) {
996 goto error;
997 }
998
999 /*
1000 * Set the CLOEXEC flag. Return code is useless because either way, the
1001 * show must go on.
1002 */
1003 (void) utils_set_fd_cloexec(sock);
1004
1005 health_code_update();
1006
1007 DBG2("Receiving code from consumer err_sock");
1008
1009 /* Getting status code from kconsumerd */
1010 ret = lttcomm_recv_unix_sock(sock, &code,
1011 sizeof(enum lttcomm_return_code));
1012 if (ret <= 0) {
1013 goto error;
1014 }
1015
1016 health_code_update();
1017
1018 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1019 /* Connect both socket, command and metadata. */
1020 consumer_data->cmd_sock =
1021 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1022 consumer_data->metadata_fd =
1023 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1024 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1025 PERROR("consumer connect cmd socket");
1026 /* On error, signal condition and quit. */
1027 signal_consumer_condition(consumer_data, -1);
1028 goto error;
1029 }
1030 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
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_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_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_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 /*
1142 * We lock here because we are about to close the sockets and some other
1143 * thread might be using them so wait before we are exclusive which will
1144 * abort all other consumer command by other threads.
1145 */
1146 pthread_mutex_lock(&consumer_data->lock);
1147
1148 /* Immediately set the consumerd state to stopped */
1149 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1150 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1151 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1152 consumer_data->type == LTTNG_CONSUMER32_UST) {
1153 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1154 } else {
1155 /* Code flow error... */
1156 assert(0);
1157 }
1158
1159 if (consumer_data->err_sock >= 0) {
1160 ret = close(consumer_data->err_sock);
1161 if (ret) {
1162 PERROR("close");
1163 }
1164 consumer_data->err_sock = -1;
1165 }
1166 if (consumer_data->cmd_sock >= 0) {
1167 ret = close(consumer_data->cmd_sock);
1168 if (ret) {
1169 PERROR("close");
1170 }
1171 consumer_data->cmd_sock = -1;
1172 }
1173 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1174 ret = close(*consumer_data->metadata_sock.fd_ptr);
1175 if (ret) {
1176 PERROR("close");
1177 }
1178 }
1179
1180 if (sock >= 0) {
1181 ret = close(sock);
1182 if (ret) {
1183 PERROR("close");
1184 }
1185 }
1186
1187 unlink(consumer_data->err_unix_sock_path);
1188 unlink(consumer_data->cmd_unix_sock_path);
1189 consumer_data->pid = 0;
1190 pthread_mutex_unlock(&consumer_data->lock);
1191 /* Cleanup metadata socket mutex. */
1192 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1193 free(consumer_data->metadata_sock.lock);
1194
1195 lttng_poll_clean(&events);
1196 error_poll:
1197 if (err) {
1198 health_error();
1199 ERR("Health error occurred in %s", __func__);
1200 }
1201 health_unregister();
1202 DBG("consumer thread cleanup completed");
1203
1204 return NULL;
1205 }
1206
1207 /*
1208 * This thread manage application communication.
1209 */
1210 static void *thread_manage_apps(void *data)
1211 {
1212 int i, ret, pollfd, err = -1;
1213 uint32_t revents, nb_fd;
1214 struct lttng_poll_event events;
1215
1216 DBG("[thread] Manage application started");
1217
1218 rcu_register_thread();
1219 rcu_thread_online();
1220
1221 health_register(HEALTH_TYPE_APP_MANAGE);
1222
1223 if (testpoint(thread_manage_apps)) {
1224 goto error_testpoint;
1225 }
1226
1227 health_code_update();
1228
1229 ret = sessiond_set_thread_pollset(&events, 2);
1230 if (ret < 0) {
1231 goto error_poll_create;
1232 }
1233
1234 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1235 if (ret < 0) {
1236 goto error;
1237 }
1238
1239 if (testpoint(thread_manage_apps_before_loop)) {
1240 goto error;
1241 }
1242
1243 health_code_update();
1244
1245 while (1) {
1246 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1247
1248 /* Inifinite blocking call, waiting for transmission */
1249 restart:
1250 health_poll_entry();
1251 ret = lttng_poll_wait(&events, -1);
1252 health_poll_exit();
1253 if (ret < 0) {
1254 /*
1255 * Restart interrupted system call.
1256 */
1257 if (errno == EINTR) {
1258 goto restart;
1259 }
1260 goto error;
1261 }
1262
1263 nb_fd = ret;
1264
1265 for (i = 0; i < nb_fd; i++) {
1266 /* Fetch once the poll data */
1267 revents = LTTNG_POLL_GETEV(&events, i);
1268 pollfd = LTTNG_POLL_GETFD(&events, i);
1269
1270 health_code_update();
1271
1272 /* Thread quit pipe has been closed. Killing thread. */
1273 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1274 if (ret) {
1275 err = 0;
1276 goto exit;
1277 }
1278
1279 /* Inspect the apps cmd pipe */
1280 if (pollfd == apps_cmd_pipe[0]) {
1281 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1282 ERR("Apps command pipe error");
1283 goto error;
1284 } else if (revents & LPOLLIN) {
1285 int sock;
1286
1287 /* Empty pipe */
1288 do {
1289 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1290 } while (ret < 0 && errno == EINTR);
1291 if (ret < 0 || ret < sizeof(sock)) {
1292 PERROR("read apps cmd pipe");
1293 goto error;
1294 }
1295
1296 health_code_update();
1297
1298 /*
1299 * We only monitor the error events of the socket. This
1300 * thread does not handle any incoming data from UST
1301 * (POLLIN).
1302 */
1303 ret = lttng_poll_add(&events, sock,
1304 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1305 if (ret < 0) {
1306 goto error;
1307 }
1308
1309 /*
1310 * Set socket timeout for both receiving and ending.
1311 * app_socket_timeout is in seconds, whereas
1312 * lttcomm_setsockopt_rcv_timeout and
1313 * lttcomm_setsockopt_snd_timeout expect msec as
1314 * parameter.
1315 */
1316 (void) lttcomm_setsockopt_rcv_timeout(sock,
1317 app_socket_timeout * 1000);
1318 (void) lttcomm_setsockopt_snd_timeout(sock,
1319 app_socket_timeout * 1000);
1320
1321 DBG("Apps with sock %d added to poll set", sock);
1322
1323 health_code_update();
1324
1325 break;
1326 }
1327 } else {
1328 /*
1329 * At this point, we know that a registered application made
1330 * the event at poll_wait.
1331 */
1332 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1333 /* Removing from the poll set */
1334 ret = lttng_poll_del(&events, pollfd);
1335 if (ret < 0) {
1336 goto error;
1337 }
1338
1339 /* Socket closed on remote end. */
1340 ust_app_unregister(pollfd);
1341 break;
1342 }
1343 }
1344
1345 health_code_update();
1346 }
1347 }
1348
1349 exit:
1350 error:
1351 lttng_poll_clean(&events);
1352 error_poll_create:
1353 error_testpoint:
1354 utils_close_pipe(apps_cmd_pipe);
1355 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1356
1357 /*
1358 * We don't clean the UST app hash table here since already registered
1359 * applications can still be controlled so let them be until the session
1360 * daemon dies or the applications stop.
1361 */
1362
1363 if (err) {
1364 health_error();
1365 ERR("Health error occurred in %s", __func__);
1366 }
1367 health_unregister();
1368 DBG("Application communication apps thread cleanup complete");
1369 rcu_thread_offline();
1370 rcu_unregister_thread();
1371 return NULL;
1372 }
1373
1374 /*
1375 * Send a socket to a thread This is called from the dispatch UST registration
1376 * thread once all sockets are set for the application.
1377 *
1378 * On success, return 0 else a negative value being the errno message of the
1379 * write().
1380 */
1381 static int send_socket_to_thread(int fd, int sock)
1382 {
1383 int ret;
1384
1385 /* Sockets MUST be set or else this should not have been called. */
1386 assert(fd >= 0);
1387 assert(sock >= 0);
1388
1389 do {
1390 ret = write(fd, &sock, sizeof(sock));
1391 } while (ret < 0 && errno == EINTR);
1392 if (ret < 0 || ret != sizeof(sock)) {
1393 PERROR("write apps pipe %d", fd);
1394 if (ret < 0) {
1395 ret = -errno;
1396 }
1397 goto error;
1398 }
1399
1400 /* All good. Don't send back the write positive ret value. */
1401 ret = 0;
1402 error:
1403 return ret;
1404 }
1405
1406 /*
1407 * Sanitize the wait queue of the dispatch registration thread meaning removing
1408 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1409 * notify socket is never received.
1410 */
1411 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1412 {
1413 int ret, nb_fd = 0, i;
1414 unsigned int fd_added = 0;
1415 struct lttng_poll_event events;
1416 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1417
1418 assert(wait_queue);
1419
1420 lttng_poll_init(&events);
1421
1422 /* Just skip everything for an empty queue. */
1423 if (!wait_queue->count) {
1424 goto end;
1425 }
1426
1427 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1428 if (ret < 0) {
1429 goto error_create;
1430 }
1431
1432 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1433 &wait_queue->head, head) {
1434 assert(wait_node->app);
1435 ret = lttng_poll_add(&events, wait_node->app->sock,
1436 LPOLLHUP | LPOLLERR);
1437 if (ret < 0) {
1438 goto error;
1439 }
1440
1441 fd_added = 1;
1442 }
1443
1444 if (!fd_added) {
1445 goto end;
1446 }
1447
1448 /*
1449 * Poll but don't block so we can quickly identify the faulty events and
1450 * clean them afterwards from the wait queue.
1451 */
1452 ret = lttng_poll_wait(&events, 0);
1453 if (ret < 0) {
1454 goto error;
1455 }
1456 nb_fd = ret;
1457
1458 for (i = 0; i < nb_fd; i++) {
1459 /* Get faulty FD. */
1460 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1461 int pollfd = LTTNG_POLL_GETFD(&events, i);
1462
1463 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1464 &wait_queue->head, head) {
1465 if (pollfd == wait_node->app->sock &&
1466 (revents & (LPOLLHUP | LPOLLERR))) {
1467 cds_list_del(&wait_node->head);
1468 wait_queue->count--;
1469 ust_app_destroy(wait_node->app);
1470 free(wait_node);
1471 break;
1472 }
1473 }
1474 }
1475
1476 if (nb_fd > 0) {
1477 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1478 }
1479
1480 end:
1481 lttng_poll_clean(&events);
1482 return;
1483
1484 error:
1485 lttng_poll_clean(&events);
1486 error_create:
1487 ERR("Unable to sanitize wait queue");
1488 return;
1489 }
1490
1491 /*
1492 * Dispatch request from the registration threads to the application
1493 * communication thread.
1494 */
1495 static void *thread_dispatch_ust_registration(void *data)
1496 {
1497 int ret, err = -1;
1498 struct cds_wfq_node *node;
1499 struct ust_command *ust_cmd = NULL;
1500 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1501 struct ust_reg_wait_queue wait_queue = {
1502 .count = 0,
1503 };
1504
1505 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1506
1507 health_code_update();
1508
1509 CDS_INIT_LIST_HEAD(&wait_queue.head);
1510
1511 DBG("[thread] Dispatch UST command started");
1512
1513 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1514 health_code_update();
1515
1516 /* Atomically prepare the queue futex */
1517 futex_nto1_prepare(&ust_cmd_queue.futex);
1518
1519 do {
1520 struct ust_app *app = NULL;
1521 ust_cmd = NULL;
1522
1523 /*
1524 * Make sure we don't have node(s) that have hung up before receiving
1525 * the notify socket. This is to clean the list in order to avoid
1526 * memory leaks from notify socket that are never seen.
1527 */
1528 sanitize_wait_queue(&wait_queue);
1529
1530 health_code_update();
1531 /* Dequeue command for registration */
1532 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1533 if (node == NULL) {
1534 DBG("Woken up but nothing in the UST command queue");
1535 /* Continue thread execution */
1536 break;
1537 }
1538
1539 ust_cmd = caa_container_of(node, struct ust_command, node);
1540
1541 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1542 " gid:%d sock:%d name:%s (version %d.%d)",
1543 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1544 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1545 ust_cmd->sock, ust_cmd->reg_msg.name,
1546 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1547
1548 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1549 wait_node = zmalloc(sizeof(*wait_node));
1550 if (!wait_node) {
1551 PERROR("zmalloc wait_node dispatch");
1552 ret = close(ust_cmd->sock);
1553 if (ret < 0) {
1554 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1555 }
1556 lttng_fd_put(1, LTTNG_FD_APPS);
1557 free(ust_cmd);
1558 goto error;
1559 }
1560 CDS_INIT_LIST_HEAD(&wait_node->head);
1561
1562 /* Create application object if socket is CMD. */
1563 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1564 ust_cmd->sock);
1565 if (!wait_node->app) {
1566 ret = close(ust_cmd->sock);
1567 if (ret < 0) {
1568 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1569 }
1570 lttng_fd_put(1, LTTNG_FD_APPS);
1571 free(wait_node);
1572 free(ust_cmd);
1573 continue;
1574 }
1575 /*
1576 * Add application to the wait queue so we can set the notify
1577 * socket before putting this object in the global ht.
1578 */
1579 cds_list_add(&wait_node->head, &wait_queue.head);
1580 wait_queue.count++;
1581
1582 free(ust_cmd);
1583 /*
1584 * We have to continue here since we don't have the notify
1585 * socket and the application MUST be added to the hash table
1586 * only at that moment.
1587 */
1588 continue;
1589 } else {
1590 /*
1591 * Look for the application in the local wait queue and set the
1592 * notify socket if found.
1593 */
1594 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1595 &wait_queue.head, head) {
1596 health_code_update();
1597 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1598 wait_node->app->notify_sock = ust_cmd->sock;
1599 cds_list_del(&wait_node->head);
1600 wait_queue.count--;
1601 app = wait_node->app;
1602 free(wait_node);
1603 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1604 break;
1605 }
1606 }
1607
1608 /*
1609 * With no application at this stage the received socket is
1610 * basically useless so close it before we free the cmd data
1611 * structure for good.
1612 */
1613 if (!app) {
1614 ret = close(ust_cmd->sock);
1615 if (ret < 0) {
1616 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1617 }
1618 lttng_fd_put(1, LTTNG_FD_APPS);
1619 }
1620 free(ust_cmd);
1621 }
1622
1623 if (app) {
1624 /*
1625 * @session_lock_list
1626 *
1627 * Lock the global session list so from the register up to the
1628 * registration done message, no thread can see the application
1629 * and change its state.
1630 */
1631 session_lock_list();
1632 rcu_read_lock();
1633
1634 /*
1635 * Add application to the global hash table. This needs to be
1636 * done before the update to the UST registry can locate the
1637 * application.
1638 */
1639 ust_app_add(app);
1640
1641 /* Set app version. This call will print an error if needed. */
1642 (void) ust_app_version(app);
1643
1644 /* Send notify socket through the notify pipe. */
1645 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1646 app->notify_sock);
1647 if (ret < 0) {
1648 rcu_read_unlock();
1649 session_unlock_list();
1650 /* No notify thread, stop the UST tracing. */
1651 goto error;
1652 }
1653
1654 /*
1655 * Update newly registered application with the tracing
1656 * registry info already enabled information.
1657 */
1658 update_ust_app(app->sock);
1659
1660 /*
1661 * Don't care about return value. Let the manage apps threads
1662 * handle app unregistration upon socket close.
1663 */
1664 (void) ust_app_register_done(app->sock);
1665
1666 /*
1667 * Even if the application socket has been closed, send the app
1668 * to the thread and unregistration will take place at that
1669 * place.
1670 */
1671 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1672 if (ret < 0) {
1673 rcu_read_unlock();
1674 session_unlock_list();
1675 /* No apps. thread, stop the UST tracing. */
1676 goto error;
1677 }
1678
1679 rcu_read_unlock();
1680 session_unlock_list();
1681 }
1682 } while (node != NULL);
1683
1684 health_poll_entry();
1685 /* Futex wait on queue. Blocking call on futex() */
1686 futex_nto1_wait(&ust_cmd_queue.futex);
1687 health_poll_exit();
1688 }
1689 /* Normal exit, no error */
1690 err = 0;
1691
1692 error:
1693 /* Clean up wait queue. */
1694 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1695 &wait_queue.head, head) {
1696 cds_list_del(&wait_node->head);
1697 wait_queue.count--;
1698 free(wait_node);
1699 }
1700
1701 DBG("Dispatch thread dying");
1702 if (err) {
1703 health_error();
1704 ERR("Health error occurred in %s", __func__);
1705 }
1706 health_unregister();
1707 return NULL;
1708 }
1709
1710 /*
1711 * This thread manage application registration.
1712 */
1713 static void *thread_registration_apps(void *data)
1714 {
1715 int sock = -1, i, ret, pollfd, err = -1;
1716 uint32_t revents, nb_fd;
1717 struct lttng_poll_event events;
1718 /*
1719 * Get allocated in this thread, enqueued to a global queue, dequeued and
1720 * freed in the manage apps thread.
1721 */
1722 struct ust_command *ust_cmd = NULL;
1723
1724 DBG("[thread] Manage application registration started");
1725
1726 health_register(HEALTH_TYPE_APP_REG);
1727
1728 if (testpoint(thread_registration_apps)) {
1729 goto error_testpoint;
1730 }
1731
1732 ret = lttcomm_listen_unix_sock(apps_sock);
1733 if (ret < 0) {
1734 goto error_listen;
1735 }
1736
1737 /*
1738 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1739 * more will be added to this poll set.
1740 */
1741 ret = sessiond_set_thread_pollset(&events, 2);
1742 if (ret < 0) {
1743 goto error_create_poll;
1744 }
1745
1746 /* Add the application registration socket */
1747 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1748 if (ret < 0) {
1749 goto error_poll_add;
1750 }
1751
1752 /* Notify all applications to register */
1753 ret = notify_ust_apps(1);
1754 if (ret < 0) {
1755 ERR("Failed to notify applications or create the wait shared memory.\n"
1756 "Execution continues but there might be problem for already\n"
1757 "running applications that wishes to register.");
1758 }
1759
1760 while (1) {
1761 DBG("Accepting application registration");
1762
1763 /* Inifinite blocking call, waiting for transmission */
1764 restart:
1765 health_poll_entry();
1766 ret = lttng_poll_wait(&events, -1);
1767 health_poll_exit();
1768 if (ret < 0) {
1769 /*
1770 * Restart interrupted system call.
1771 */
1772 if (errno == EINTR) {
1773 goto restart;
1774 }
1775 goto error;
1776 }
1777
1778 nb_fd = ret;
1779
1780 for (i = 0; i < nb_fd; i++) {
1781 health_code_update();
1782
1783 /* Fetch once the poll data */
1784 revents = LTTNG_POLL_GETEV(&events, i);
1785 pollfd = LTTNG_POLL_GETFD(&events, i);
1786
1787 /* Thread quit pipe has been closed. Killing thread. */
1788 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1789 if (ret) {
1790 err = 0;
1791 goto exit;
1792 }
1793
1794 /* Event on the registration socket */
1795 if (pollfd == apps_sock) {
1796 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1797 ERR("Register apps socket poll error");
1798 goto error;
1799 } else if (revents & LPOLLIN) {
1800 sock = lttcomm_accept_unix_sock(apps_sock);
1801 if (sock < 0) {
1802 goto error;
1803 }
1804
1805 /*
1806 * Set the CLOEXEC flag. Return code is useless because
1807 * either way, the show must go on.
1808 */
1809 (void) utils_set_fd_cloexec(sock);
1810
1811 /* Create UST registration command for enqueuing */
1812 ust_cmd = zmalloc(sizeof(struct ust_command));
1813 if (ust_cmd == NULL) {
1814 PERROR("ust command zmalloc");
1815 goto error;
1816 }
1817
1818 /*
1819 * Using message-based transmissions to ensure we don't
1820 * have to deal with partially received messages.
1821 */
1822 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1823 if (ret < 0) {
1824 ERR("Exhausted file descriptors allowed for applications.");
1825 free(ust_cmd);
1826 ret = close(sock);
1827 if (ret) {
1828 PERROR("close");
1829 }
1830 sock = -1;
1831 continue;
1832 }
1833
1834 health_code_update();
1835 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1836 if (ret < 0) {
1837 free(ust_cmd);
1838 /* Close socket of the application. */
1839 ret = close(sock);
1840 if (ret) {
1841 PERROR("close");
1842 }
1843 lttng_fd_put(LTTNG_FD_APPS, 1);
1844 sock = -1;
1845 continue;
1846 }
1847 health_code_update();
1848
1849 ust_cmd->sock = sock;
1850 sock = -1;
1851
1852 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1853 " gid:%d sock:%d name:%s (version %d.%d)",
1854 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1855 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1856 ust_cmd->sock, ust_cmd->reg_msg.name,
1857 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1858
1859 /*
1860 * Lock free enqueue the registration request. The red pill
1861 * has been taken! This apps will be part of the *system*.
1862 */
1863 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1864
1865 /*
1866 * Wake the registration queue futex. Implicit memory
1867 * barrier with the exchange in cds_wfq_enqueue.
1868 */
1869 futex_nto1_wake(&ust_cmd_queue.futex);
1870 }
1871 }
1872 }
1873 }
1874
1875 exit:
1876 error:
1877 if (err) {
1878 health_error();
1879 ERR("Health error occurred in %s", __func__);
1880 }
1881
1882 /* Notify that the registration thread is gone */
1883 notify_ust_apps(0);
1884
1885 if (apps_sock >= 0) {
1886 ret = close(apps_sock);
1887 if (ret) {
1888 PERROR("close");
1889 }
1890 }
1891 if (sock >= 0) {
1892 ret = close(sock);
1893 if (ret) {
1894 PERROR("close");
1895 }
1896 lttng_fd_put(LTTNG_FD_APPS, 1);
1897 }
1898 unlink(apps_unix_sock_path);
1899
1900 error_poll_add:
1901 lttng_poll_clean(&events);
1902 error_listen:
1903 error_create_poll:
1904 error_testpoint:
1905 DBG("UST Registration thread cleanup complete");
1906 health_unregister();
1907
1908 return NULL;
1909 }
1910
1911 /*
1912 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1913 * exec or it will fails.
1914 */
1915 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1916 {
1917 int ret, clock_ret;
1918 struct timespec timeout;
1919
1920 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1921 consumer_data->consumer_thread_is_ready = 0;
1922
1923 /* Setup pthread condition */
1924 ret = pthread_condattr_init(&consumer_data->condattr);
1925 if (ret != 0) {
1926 errno = ret;
1927 PERROR("pthread_condattr_init consumer data");
1928 goto error;
1929 }
1930
1931 /*
1932 * Set the monotonic clock in order to make sure we DO NOT jump in time
1933 * between the clock_gettime() call and the timedwait call. See bug #324
1934 * for a more details and how we noticed it.
1935 */
1936 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1937 if (ret != 0) {
1938 errno = ret;
1939 PERROR("pthread_condattr_setclock consumer data");
1940 goto error;
1941 }
1942
1943 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1944 if (ret != 0) {
1945 errno = ret;
1946 PERROR("pthread_cond_init consumer data");
1947 goto error;
1948 }
1949
1950 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1951 consumer_data);
1952 if (ret != 0) {
1953 PERROR("pthread_create consumer");
1954 ret = -1;
1955 goto error;
1956 }
1957
1958 /* We are about to wait on a pthread condition */
1959 pthread_mutex_lock(&consumer_data->cond_mutex);
1960
1961 /* Get time for sem_timedwait absolute timeout */
1962 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1963 /*
1964 * Set the timeout for the condition timed wait even if the clock gettime
1965 * call fails since we might loop on that call and we want to avoid to
1966 * increment the timeout too many times.
1967 */
1968 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1969
1970 /*
1971 * The following loop COULD be skipped in some conditions so this is why we
1972 * set ret to 0 in order to make sure at least one round of the loop is
1973 * done.
1974 */
1975 ret = 0;
1976
1977 /*
1978 * Loop until the condition is reached or when a timeout is reached. Note
1979 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1980 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1981 * possible. This loop does not take any chances and works with both of
1982 * them.
1983 */
1984 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1985 if (clock_ret < 0) {
1986 PERROR("clock_gettime spawn consumer");
1987 /* Infinite wait for the consumerd thread to be ready */
1988 ret = pthread_cond_wait(&consumer_data->cond,
1989 &consumer_data->cond_mutex);
1990 } else {
1991 ret = pthread_cond_timedwait(&consumer_data->cond,
1992 &consumer_data->cond_mutex, &timeout);
1993 }
1994 }
1995
1996 /* Release the pthread condition */
1997 pthread_mutex_unlock(&consumer_data->cond_mutex);
1998
1999 if (ret != 0) {
2000 errno = ret;
2001 if (ret == ETIMEDOUT) {
2002 /*
2003 * Call has timed out so we kill the kconsumerd_thread and return
2004 * an error.
2005 */
2006 ERR("Condition timed out. The consumer thread was never ready."
2007 " Killing it");
2008 ret = pthread_cancel(consumer_data->thread);
2009 if (ret < 0) {
2010 PERROR("pthread_cancel consumer thread");
2011 }
2012 } else {
2013 PERROR("pthread_cond_wait failed consumer thread");
2014 }
2015 goto error;
2016 }
2017
2018 pthread_mutex_lock(&consumer_data->pid_mutex);
2019 if (consumer_data->pid == 0) {
2020 ERR("Consumerd did not start");
2021 pthread_mutex_unlock(&consumer_data->pid_mutex);
2022 goto error;
2023 }
2024 pthread_mutex_unlock(&consumer_data->pid_mutex);
2025
2026 return 0;
2027
2028 error:
2029 return ret;
2030 }
2031
2032 /*
2033 * Join consumer thread
2034 */
2035 static int join_consumer_thread(struct consumer_data *consumer_data)
2036 {
2037 void *status;
2038
2039 /* Consumer pid must be a real one. */
2040 if (consumer_data->pid > 0) {
2041 int ret;
2042 ret = kill(consumer_data->pid, SIGTERM);
2043 if (ret) {
2044 ERR("Error killing consumer daemon");
2045 return ret;
2046 }
2047 return pthread_join(consumer_data->thread, &status);
2048 } else {
2049 return 0;
2050 }
2051 }
2052
2053 /*
2054 * Fork and exec a consumer daemon (consumerd).
2055 *
2056 * Return pid if successful else -1.
2057 */
2058 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2059 {
2060 int ret;
2061 pid_t pid;
2062 const char *consumer_to_use;
2063 const char *verbosity;
2064 struct stat st;
2065
2066 DBG("Spawning consumerd");
2067
2068 pid = fork();
2069 if (pid == 0) {
2070 /*
2071 * Exec consumerd.
2072 */
2073 if (opt_verbose_consumer) {
2074 verbosity = "--verbose";
2075 } else {
2076 verbosity = "--quiet";
2077 }
2078 switch (consumer_data->type) {
2079 case LTTNG_CONSUMER_KERNEL:
2080 /*
2081 * Find out which consumerd to execute. We will first try the
2082 * 64-bit path, then the sessiond's installation directory, and
2083 * fallback on the 32-bit one,
2084 */
2085 DBG3("Looking for a kernel consumer at these locations:");
2086 DBG3(" 1) %s", consumerd64_bin);
2087 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2088 DBG3(" 3) %s", consumerd32_bin);
2089 if (stat(consumerd64_bin, &st) == 0) {
2090 DBG3("Found location #1");
2091 consumer_to_use = consumerd64_bin;
2092 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2093 DBG3("Found location #2");
2094 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2095 } else if (stat(consumerd32_bin, &st) == 0) {
2096 DBG3("Found location #3");
2097 consumer_to_use = consumerd32_bin;
2098 } else {
2099 DBG("Could not find any valid consumerd executable");
2100 break;
2101 }
2102 DBG("Using kernel consumer at: %s", consumer_to_use);
2103 execl(consumer_to_use,
2104 "lttng-consumerd", verbosity, "-k",
2105 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2106 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2107 NULL);
2108 break;
2109 case LTTNG_CONSUMER64_UST:
2110 {
2111 char *tmpnew = NULL;
2112
2113 if (consumerd64_libdir[0] != '\0') {
2114 char *tmp;
2115 size_t tmplen;
2116
2117 tmp = getenv("LD_LIBRARY_PATH");
2118 if (!tmp) {
2119 tmp = "";
2120 }
2121 tmplen = strlen("LD_LIBRARY_PATH=")
2122 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2123 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2124 if (!tmpnew) {
2125 ret = -ENOMEM;
2126 goto error;
2127 }
2128 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2129 strcat(tmpnew, consumerd64_libdir);
2130 if (tmp[0] != '\0') {
2131 strcat(tmpnew, ":");
2132 strcat(tmpnew, tmp);
2133 }
2134 ret = putenv(tmpnew);
2135 if (ret) {
2136 ret = -errno;
2137 free(tmpnew);
2138 goto error;
2139 }
2140 }
2141 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2142 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2143 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2144 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2145 NULL);
2146 if (consumerd64_libdir[0] != '\0') {
2147 free(tmpnew);
2148 }
2149 if (ret) {
2150 goto error;
2151 }
2152 break;
2153 }
2154 case LTTNG_CONSUMER32_UST:
2155 {
2156 char *tmpnew = NULL;
2157
2158 if (consumerd32_libdir[0] != '\0') {
2159 char *tmp;
2160 size_t tmplen;
2161
2162 tmp = getenv("LD_LIBRARY_PATH");
2163 if (!tmp) {
2164 tmp = "";
2165 }
2166 tmplen = strlen("LD_LIBRARY_PATH=")
2167 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2168 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2169 if (!tmpnew) {
2170 ret = -ENOMEM;
2171 goto error;
2172 }
2173 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2174 strcat(tmpnew, consumerd32_libdir);
2175 if (tmp[0] != '\0') {
2176 strcat(tmpnew, ":");
2177 strcat(tmpnew, tmp);
2178 }
2179 ret = putenv(tmpnew);
2180 if (ret) {
2181 ret = -errno;
2182 free(tmpnew);
2183 goto error;
2184 }
2185 }
2186 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2187 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2188 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2189 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2190 NULL);
2191 if (consumerd32_libdir[0] != '\0') {
2192 free(tmpnew);
2193 }
2194 if (ret) {
2195 goto error;
2196 }
2197 break;
2198 }
2199 default:
2200 PERROR("unknown consumer type");
2201 exit(EXIT_FAILURE);
2202 }
2203 if (errno != 0) {
2204 PERROR("kernel start consumer exec");
2205 }
2206 exit(EXIT_FAILURE);
2207 } else if (pid > 0) {
2208 ret = pid;
2209 } else {
2210 PERROR("start consumer fork");
2211 ret = -errno;
2212 }
2213 error:
2214 return ret;
2215 }
2216
2217 /*
2218 * Spawn the consumerd daemon and session daemon thread.
2219 */
2220 static int start_consumerd(struct consumer_data *consumer_data)
2221 {
2222 int ret;
2223
2224 /*
2225 * Set the listen() state on the socket since there is a possible race
2226 * between the exec() of the consumer daemon and this call if place in the
2227 * consumer thread. See bug #366 for more details.
2228 */
2229 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2230 if (ret < 0) {
2231 goto error;
2232 }
2233
2234 pthread_mutex_lock(&consumer_data->pid_mutex);
2235 if (consumer_data->pid != 0) {
2236 pthread_mutex_unlock(&consumer_data->pid_mutex);
2237 goto end;
2238 }
2239
2240 ret = spawn_consumerd(consumer_data);
2241 if (ret < 0) {
2242 ERR("Spawning consumerd failed");
2243 pthread_mutex_unlock(&consumer_data->pid_mutex);
2244 goto error;
2245 }
2246
2247 /* Setting up the consumer_data pid */
2248 consumer_data->pid = ret;
2249 DBG2("Consumer pid %d", consumer_data->pid);
2250 pthread_mutex_unlock(&consumer_data->pid_mutex);
2251
2252 DBG2("Spawning consumer control thread");
2253 ret = spawn_consumer_thread(consumer_data);
2254 if (ret < 0) {
2255 ERR("Fatal error spawning consumer control thread");
2256 goto error;
2257 }
2258
2259 end:
2260 return 0;
2261
2262 error:
2263 /* Cleanup already created sockets on error. */
2264 if (consumer_data->err_sock >= 0) {
2265 int err;
2266
2267 err = close(consumer_data->err_sock);
2268 if (err < 0) {
2269 PERROR("close consumer data error socket");
2270 }
2271 }
2272 return ret;
2273 }
2274
2275 /*
2276 * Compute health status of each consumer. If one of them is zero (bad
2277 * state), we return 0.
2278 */
2279 static int check_consumer_health(void)
2280 {
2281 int ret;
2282
2283 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2284
2285 DBG3("Health consumer check %d", ret);
2286
2287 return ret;
2288 }
2289
2290 /*
2291 * Setup necessary data for kernel tracer action.
2292 */
2293 static int init_kernel_tracer(void)
2294 {
2295 int ret;
2296
2297 /* Modprobe lttng kernel modules */
2298 ret = modprobe_lttng_control();
2299 if (ret < 0) {
2300 goto error;
2301 }
2302
2303 /* Open debugfs lttng */
2304 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2305 if (kernel_tracer_fd < 0) {
2306 DBG("Failed to open %s", module_proc_lttng);
2307 ret = -1;
2308 goto error_open;
2309 }
2310
2311 /* Validate kernel version */
2312 ret = kernel_validate_version(kernel_tracer_fd);
2313 if (ret < 0) {
2314 goto error_version;
2315 }
2316
2317 ret = modprobe_lttng_data();
2318 if (ret < 0) {
2319 goto error_modules;
2320 }
2321
2322 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2323 return 0;
2324
2325 error_version:
2326 modprobe_remove_lttng_control();
2327 ret = close(kernel_tracer_fd);
2328 if (ret) {
2329 PERROR("close");
2330 }
2331 kernel_tracer_fd = -1;
2332 return LTTNG_ERR_KERN_VERSION;
2333
2334 error_modules:
2335 ret = close(kernel_tracer_fd);
2336 if (ret) {
2337 PERROR("close");
2338 }
2339
2340 error_open:
2341 modprobe_remove_lttng_control();
2342
2343 error:
2344 WARN("No kernel tracer available");
2345 kernel_tracer_fd = -1;
2346 if (!is_root) {
2347 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2348 } else {
2349 return LTTNG_ERR_KERN_NA;
2350 }
2351 }
2352
2353
2354 /*
2355 * Copy consumer output from the tracing session to the domain session. The
2356 * function also applies the right modification on a per domain basis for the
2357 * trace files destination directory.
2358 *
2359 * Should *NOT* be called with RCU read-side lock held.
2360 */
2361 static int copy_session_consumer(int domain, struct ltt_session *session)
2362 {
2363 int ret;
2364 const char *dir_name;
2365 struct consumer_output *consumer;
2366
2367 assert(session);
2368 assert(session->consumer);
2369
2370 switch (domain) {
2371 case LTTNG_DOMAIN_KERNEL:
2372 DBG3("Copying tracing session consumer output in kernel session");
2373 /*
2374 * XXX: We should audit the session creation and what this function
2375 * does "extra" in order to avoid a destroy since this function is used
2376 * in the domain session creation (kernel and ust) only. Same for UST
2377 * domain.
2378 */
2379 if (session->kernel_session->consumer) {
2380 consumer_destroy_output(session->kernel_session->consumer);
2381 }
2382 session->kernel_session->consumer =
2383 consumer_copy_output(session->consumer);
2384 /* Ease our life a bit for the next part */
2385 consumer = session->kernel_session->consumer;
2386 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2387 break;
2388 case LTTNG_DOMAIN_UST:
2389 DBG3("Copying tracing session consumer output in UST session");
2390 if (session->ust_session->consumer) {
2391 consumer_destroy_output(session->ust_session->consumer);
2392 }
2393 session->ust_session->consumer =
2394 consumer_copy_output(session->consumer);
2395 /* Ease our life a bit for the next part */
2396 consumer = session->ust_session->consumer;
2397 dir_name = DEFAULT_UST_TRACE_DIR;
2398 break;
2399 default:
2400 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2401 goto error;
2402 }
2403
2404 /* Append correct directory to subdir */
2405 strncat(consumer->subdir, dir_name,
2406 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2407 DBG3("Copy session consumer subdir %s", consumer->subdir);
2408
2409 ret = LTTNG_OK;
2410
2411 error:
2412 return ret;
2413 }
2414
2415 /*
2416 * Create an UST session and add it to the session ust list.
2417 *
2418 * Should *NOT* be called with RCU read-side lock held.
2419 */
2420 static int create_ust_session(struct ltt_session *session,
2421 struct lttng_domain *domain)
2422 {
2423 int ret;
2424 struct ltt_ust_session *lus = NULL;
2425
2426 assert(session);
2427 assert(domain);
2428 assert(session->consumer);
2429
2430 switch (domain->type) {
2431 case LTTNG_DOMAIN_UST:
2432 break;
2433 default:
2434 ERR("Unknown UST domain on create session %d", domain->type);
2435 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2436 goto error;
2437 }
2438
2439 DBG("Creating UST session");
2440
2441 lus = trace_ust_create_session(session->id);
2442 if (lus == NULL) {
2443 ret = LTTNG_ERR_UST_SESS_FAIL;
2444 goto error;
2445 }
2446
2447 lus->uid = session->uid;
2448 lus->gid = session->gid;
2449 lus->output_traces = session->output_traces;
2450 lus->snapshot_mode = session->snapshot_mode;
2451 session->ust_session = lus;
2452
2453 /* Copy session output to the newly created UST session */
2454 ret = copy_session_consumer(domain->type, session);
2455 if (ret != LTTNG_OK) {
2456 goto error;
2457 }
2458
2459 return LTTNG_OK;
2460
2461 error:
2462 free(lus);
2463 session->ust_session = NULL;
2464 return ret;
2465 }
2466
2467 /*
2468 * Create a kernel tracer session then create the default channel.
2469 */
2470 static int create_kernel_session(struct ltt_session *session)
2471 {
2472 int ret;
2473
2474 DBG("Creating kernel session");
2475
2476 ret = kernel_create_session(session, kernel_tracer_fd);
2477 if (ret < 0) {
2478 ret = LTTNG_ERR_KERN_SESS_FAIL;
2479 goto error;
2480 }
2481
2482 /* Code flow safety */
2483 assert(session->kernel_session);
2484
2485 /* Copy session output to the newly created Kernel session */
2486 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2487 if (ret != LTTNG_OK) {
2488 goto error;
2489 }
2490
2491 /* Create directory(ies) on local filesystem. */
2492 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2493 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2494 ret = run_as_mkdir_recursive(
2495 session->kernel_session->consumer->dst.trace_path,
2496 S_IRWXU | S_IRWXG, session->uid, session->gid);
2497 if (ret < 0) {
2498 if (ret != -EEXIST) {
2499 ERR("Trace directory creation error");
2500 goto error;
2501 }
2502 }
2503 }
2504
2505 session->kernel_session->uid = session->uid;
2506 session->kernel_session->gid = session->gid;
2507 session->kernel_session->output_traces = session->output_traces;
2508 session->kernel_session->snapshot_mode = session->snapshot_mode;
2509
2510 return LTTNG_OK;
2511
2512 error:
2513 trace_kernel_destroy_session(session->kernel_session);
2514 session->kernel_session = NULL;
2515 return ret;
2516 }
2517
2518 /*
2519 * Count number of session permitted by uid/gid.
2520 */
2521 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2522 {
2523 unsigned int i = 0;
2524 struct ltt_session *session;
2525
2526 DBG("Counting number of available session for UID %d GID %d",
2527 uid, gid);
2528 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2529 /*
2530 * Only list the sessions the user can control.
2531 */
2532 if (!session_access_ok(session, uid, gid)) {
2533 continue;
2534 }
2535 i++;
2536 }
2537 return i;
2538 }
2539
2540 /*
2541 * Process the command requested by the lttng client within the command
2542 * context structure. This function make sure that the return structure (llm)
2543 * is set and ready for transmission before returning.
2544 *
2545 * Return any error encountered or 0 for success.
2546 *
2547 * "sock" is only used for special-case var. len data.
2548 *
2549 * Should *NOT* be called with RCU read-side lock held.
2550 */
2551 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2552 int *sock_error)
2553 {
2554 int ret = LTTNG_OK;
2555 int need_tracing_session = 1;
2556 int need_domain;
2557
2558 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2559
2560 *sock_error = 0;
2561
2562 switch (cmd_ctx->lsm->cmd_type) {
2563 case LTTNG_CREATE_SESSION:
2564 case LTTNG_CREATE_SESSION_SNAPSHOT:
2565 case LTTNG_DESTROY_SESSION:
2566 case LTTNG_LIST_SESSIONS:
2567 case LTTNG_LIST_DOMAINS:
2568 case LTTNG_START_TRACE:
2569 case LTTNG_STOP_TRACE:
2570 case LTTNG_DATA_PENDING:
2571 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2572 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2573 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2574 case LTTNG_SNAPSHOT_RECORD:
2575 need_domain = 0;
2576 break;
2577 default:
2578 need_domain = 1;
2579 }
2580
2581 if (opt_no_kernel && need_domain
2582 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2583 if (!is_root) {
2584 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2585 } else {
2586 ret = LTTNG_ERR_KERN_NA;
2587 }
2588 goto error;
2589 }
2590
2591 /* Deny register consumer if we already have a spawned consumer. */
2592 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2593 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2594 if (kconsumer_data.pid > 0) {
2595 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2596 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2597 goto error;
2598 }
2599 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2600 }
2601
2602 /*
2603 * Check for command that don't needs to allocate a returned payload. We do
2604 * this here so we don't have to make the call for no payload at each
2605 * command.
2606 */
2607 switch(cmd_ctx->lsm->cmd_type) {
2608 case LTTNG_LIST_SESSIONS:
2609 case LTTNG_LIST_TRACEPOINTS:
2610 case LTTNG_LIST_TRACEPOINT_FIELDS:
2611 case LTTNG_LIST_DOMAINS:
2612 case LTTNG_LIST_CHANNELS:
2613 case LTTNG_LIST_EVENTS:
2614 break;
2615 default:
2616 /* Setup lttng message with no payload */
2617 ret = setup_lttng_msg(cmd_ctx, 0);
2618 if (ret < 0) {
2619 /* This label does not try to unlock the session */
2620 goto init_setup_error;
2621 }
2622 }
2623
2624 /* Commands that DO NOT need a session. */
2625 switch (cmd_ctx->lsm->cmd_type) {
2626 case LTTNG_CREATE_SESSION:
2627 case LTTNG_CREATE_SESSION_SNAPSHOT:
2628 case LTTNG_CALIBRATE:
2629 case LTTNG_LIST_SESSIONS:
2630 case LTTNG_LIST_TRACEPOINTS:
2631 case LTTNG_LIST_TRACEPOINT_FIELDS:
2632 need_tracing_session = 0;
2633 break;
2634 default:
2635 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2636 /*
2637 * We keep the session list lock across _all_ commands
2638 * for now, because the per-session lock does not
2639 * handle teardown properly.
2640 */
2641 session_lock_list();
2642 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2643 if (cmd_ctx->session == NULL) {
2644 ret = LTTNG_ERR_SESS_NOT_FOUND;
2645 goto error;
2646 } else {
2647 /* Acquire lock for the session */
2648 session_lock(cmd_ctx->session);
2649 }
2650 break;
2651 }
2652
2653 if (!need_domain) {
2654 goto skip_domain;
2655 }
2656
2657 /*
2658 * Check domain type for specific "pre-action".
2659 */
2660 switch (cmd_ctx->lsm->domain.type) {
2661 case LTTNG_DOMAIN_KERNEL:
2662 if (!is_root) {
2663 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2664 goto error;
2665 }
2666
2667 /* Kernel tracer check */
2668 if (kernel_tracer_fd == -1) {
2669 /* Basically, load kernel tracer modules */
2670 ret = init_kernel_tracer();
2671 if (ret != 0) {
2672 goto error;
2673 }
2674 }
2675
2676 /* Consumer is in an ERROR state. Report back to client */
2677 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2678 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2679 goto error;
2680 }
2681
2682 /* Need a session for kernel command */
2683 if (need_tracing_session) {
2684 if (cmd_ctx->session->kernel_session == NULL) {
2685 ret = create_kernel_session(cmd_ctx->session);
2686 if (ret < 0) {
2687 ret = LTTNG_ERR_KERN_SESS_FAIL;
2688 goto error;
2689 }
2690 }
2691
2692 /* Start the kernel consumer daemon */
2693 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2694 if (kconsumer_data.pid == 0 &&
2695 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2696 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2697 ret = start_consumerd(&kconsumer_data);
2698 if (ret < 0) {
2699 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2700 goto error;
2701 }
2702 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2703 } else {
2704 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2705 }
2706
2707 /*
2708 * The consumer was just spawned so we need to add the socket to
2709 * the consumer output of the session if exist.
2710 */
2711 ret = consumer_create_socket(&kconsumer_data,
2712 cmd_ctx->session->kernel_session->consumer);
2713 if (ret < 0) {
2714 goto error;
2715 }
2716 }
2717
2718 break;
2719 case LTTNG_DOMAIN_UST:
2720 {
2721 if (!ust_app_supported()) {
2722 ret = LTTNG_ERR_NO_UST;
2723 goto error;
2724 }
2725 /* Consumer is in an ERROR state. Report back to client */
2726 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2727 ret = LTTNG_ERR_NO_USTCONSUMERD;
2728 goto error;
2729 }
2730
2731 if (need_tracing_session) {
2732 /* Create UST session if none exist. */
2733 if (cmd_ctx->session->ust_session == NULL) {
2734 ret = create_ust_session(cmd_ctx->session,
2735 &cmd_ctx->lsm->domain);
2736 if (ret != LTTNG_OK) {
2737 goto error;
2738 }
2739 }
2740
2741 /* Start the UST consumer daemons */
2742 /* 64-bit */
2743 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2744 if (consumerd64_bin[0] != '\0' &&
2745 ustconsumer64_data.pid == 0 &&
2746 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2747 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2748 ret = start_consumerd(&ustconsumer64_data);
2749 if (ret < 0) {
2750 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2751 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2752 goto error;
2753 }
2754
2755 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2756 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2757 } else {
2758 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2759 }
2760
2761 /*
2762 * Setup socket for consumer 64 bit. No need for atomic access
2763 * since it was set above and can ONLY be set in this thread.
2764 */
2765 ret = consumer_create_socket(&ustconsumer64_data,
2766 cmd_ctx->session->ust_session->consumer);
2767 if (ret < 0) {
2768 goto error;
2769 }
2770
2771 /* 32-bit */
2772 if (consumerd32_bin[0] != '\0' &&
2773 ustconsumer32_data.pid == 0 &&
2774 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2775 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2776 ret = start_consumerd(&ustconsumer32_data);
2777 if (ret < 0) {
2778 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2779 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2780 goto error;
2781 }
2782
2783 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2784 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2785 } else {
2786 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2787 }
2788
2789 /*
2790 * Setup socket for consumer 64 bit. No need for atomic access
2791 * since it was set above and can ONLY be set in this thread.
2792 */
2793 ret = consumer_create_socket(&ustconsumer32_data,
2794 cmd_ctx->session->ust_session->consumer);
2795 if (ret < 0) {
2796 goto error;
2797 }
2798 }
2799 break;
2800 }
2801 default:
2802 break;
2803 }
2804 skip_domain:
2805
2806 /* Validate consumer daemon state when start/stop trace command */
2807 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2808 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2809 switch (cmd_ctx->lsm->domain.type) {
2810 case LTTNG_DOMAIN_UST:
2811 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2812 ret = LTTNG_ERR_NO_USTCONSUMERD;
2813 goto error;
2814 }
2815 break;
2816 case LTTNG_DOMAIN_KERNEL:
2817 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2818 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2819 goto error;
2820 }
2821 break;
2822 }
2823 }
2824
2825 /*
2826 * Check that the UID or GID match that of the tracing session.
2827 * The root user can interact with all sessions.
2828 */
2829 if (need_tracing_session) {
2830 if (!session_access_ok(cmd_ctx->session,
2831 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2832 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2833 ret = LTTNG_ERR_EPERM;
2834 goto error;
2835 }
2836 }
2837
2838 /*
2839 * Send relayd information to consumer as soon as we have a domain and a
2840 * session defined.
2841 */
2842 if (cmd_ctx->session && need_domain) {
2843 /*
2844 * Setup relayd if not done yet. If the relayd information was already
2845 * sent to the consumer, this call will gracefully return.
2846 */
2847 ret = cmd_setup_relayd(cmd_ctx->session);
2848 if (ret != LTTNG_OK) {
2849 goto error;
2850 }
2851 }
2852
2853 /* Process by command type */
2854 switch (cmd_ctx->lsm->cmd_type) {
2855 case LTTNG_ADD_CONTEXT:
2856 {
2857 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2858 cmd_ctx->lsm->u.context.channel_name,
2859 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2860 break;
2861 }
2862 case LTTNG_DISABLE_CHANNEL:
2863 {
2864 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2865 cmd_ctx->lsm->u.disable.channel_name);
2866 break;
2867 }
2868 case LTTNG_DISABLE_EVENT:
2869 {
2870 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2871 cmd_ctx->lsm->u.disable.channel_name,
2872 cmd_ctx->lsm->u.disable.name);
2873 break;
2874 }
2875 case LTTNG_DISABLE_ALL_EVENT:
2876 {
2877 DBG("Disabling all events");
2878
2879 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2880 cmd_ctx->lsm->u.disable.channel_name);
2881 break;
2882 }
2883 case LTTNG_ENABLE_CHANNEL:
2884 {
2885 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2886 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2887 break;
2888 }
2889 case LTTNG_ENABLE_EVENT:
2890 {
2891 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2892 cmd_ctx->lsm->u.enable.channel_name,
2893 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2894 break;
2895 }
2896 case LTTNG_ENABLE_ALL_EVENT:
2897 {
2898 DBG("Enabling all events");
2899
2900 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2901 cmd_ctx->lsm->u.enable.channel_name,
2902 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2903 break;
2904 }
2905 case LTTNG_LIST_TRACEPOINTS:
2906 {
2907 struct lttng_event *events;
2908 ssize_t nb_events;
2909
2910 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2911 if (nb_events < 0) {
2912 /* Return value is a negative lttng_error_code. */
2913 ret = -nb_events;
2914 goto error;
2915 }
2916
2917 /*
2918 * Setup lttng message with payload size set to the event list size in
2919 * bytes and then copy list into the llm payload.
2920 */
2921 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2922 if (ret < 0) {
2923 free(events);
2924 goto setup_error;
2925 }
2926
2927 /* Copy event list into message payload */
2928 memcpy(cmd_ctx->llm->payload, events,
2929 sizeof(struct lttng_event) * nb_events);
2930
2931 free(events);
2932
2933 ret = LTTNG_OK;
2934 break;
2935 }
2936 case LTTNG_LIST_TRACEPOINT_FIELDS:
2937 {
2938 struct lttng_event_field *fields;
2939 ssize_t nb_fields;
2940
2941 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2942 &fields);
2943 if (nb_fields < 0) {
2944 /* Return value is a negative lttng_error_code. */
2945 ret = -nb_fields;
2946 goto error;
2947 }
2948
2949 /*
2950 * Setup lttng message with payload size set to the event list size in
2951 * bytes and then copy list into the llm payload.
2952 */
2953 ret = setup_lttng_msg(cmd_ctx,
2954 sizeof(struct lttng_event_field) * nb_fields);
2955 if (ret < 0) {
2956 free(fields);
2957 goto setup_error;
2958 }
2959
2960 /* Copy event list into message payload */
2961 memcpy(cmd_ctx->llm->payload, fields,
2962 sizeof(struct lttng_event_field) * nb_fields);
2963
2964 free(fields);
2965
2966 ret = LTTNG_OK;
2967 break;
2968 }
2969 case LTTNG_SET_CONSUMER_URI:
2970 {
2971 size_t nb_uri, len;
2972 struct lttng_uri *uris;
2973
2974 nb_uri = cmd_ctx->lsm->u.uri.size;
2975 len = nb_uri * sizeof(struct lttng_uri);
2976
2977 if (nb_uri == 0) {
2978 ret = LTTNG_ERR_INVALID;
2979 goto error;
2980 }
2981
2982 uris = zmalloc(len);
2983 if (uris == NULL) {
2984 ret = LTTNG_ERR_FATAL;
2985 goto error;
2986 }
2987
2988 /* Receive variable len data */
2989 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2990 ret = lttcomm_recv_unix_sock(sock, uris, len);
2991 if (ret <= 0) {
2992 DBG("No URIs received from client... continuing");
2993 *sock_error = 1;
2994 ret = LTTNG_ERR_SESSION_FAIL;
2995 free(uris);
2996 goto error;
2997 }
2998
2999 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3000 nb_uri, uris);
3001 if (ret != LTTNG_OK) {
3002 free(uris);
3003 goto error;
3004 }
3005
3006 /*
3007 * XXX: 0 means that this URI should be applied on the session. Should
3008 * be a DOMAIN enuam.
3009 */
3010 if (cmd_ctx->lsm->domain.type == 0) {
3011 /* Add the URI for the UST session if a consumer is present. */
3012 if (cmd_ctx->session->ust_session &&
3013 cmd_ctx->session->ust_session->consumer) {
3014 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3015 nb_uri, uris);
3016 } else if (cmd_ctx->session->kernel_session &&
3017 cmd_ctx->session->kernel_session->consumer) {
3018 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3019 cmd_ctx->session, nb_uri, uris);
3020 }
3021 }
3022
3023 free(uris);
3024
3025 break;
3026 }
3027 case LTTNG_START_TRACE:
3028 {
3029 ret = cmd_start_trace(cmd_ctx->session);
3030 break;
3031 }
3032 case LTTNG_STOP_TRACE:
3033 {
3034 ret = cmd_stop_trace(cmd_ctx->session);
3035 break;
3036 }
3037 case LTTNG_CREATE_SESSION:
3038 {
3039 size_t nb_uri, len;
3040 struct lttng_uri *uris = NULL;
3041
3042 nb_uri = cmd_ctx->lsm->u.uri.size;
3043 len = nb_uri * sizeof(struct lttng_uri);
3044
3045 if (nb_uri > 0) {
3046 uris = zmalloc(len);
3047 if (uris == NULL) {
3048 ret = LTTNG_ERR_FATAL;
3049 goto error;
3050 }
3051
3052 /* Receive variable len data */
3053 DBG("Waiting for %zu URIs from client ...", nb_uri);
3054 ret = lttcomm_recv_unix_sock(sock, uris, len);
3055 if (ret <= 0) {
3056 DBG("No URIs received from client... continuing");
3057 *sock_error = 1;
3058 ret = LTTNG_ERR_SESSION_FAIL;
3059 free(uris);
3060 goto error;
3061 }
3062
3063 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3064 DBG("Creating session with ONE network URI is a bad call");
3065 ret = LTTNG_ERR_SESSION_FAIL;
3066 free(uris);
3067 goto error;
3068 }
3069 }
3070
3071 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3072 &cmd_ctx->creds);
3073
3074 free(uris);
3075
3076 break;
3077 }
3078 case LTTNG_DESTROY_SESSION:
3079 {
3080 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3081
3082 /* Set session to NULL so we do not unlock it after free. */
3083 cmd_ctx->session = NULL;
3084 break;
3085 }
3086 case LTTNG_LIST_DOMAINS:
3087 {
3088 ssize_t nb_dom;
3089 struct lttng_domain *domains;
3090
3091 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3092 if (nb_dom < 0) {
3093 /* Return value is a negative lttng_error_code. */
3094 ret = -nb_dom;
3095 goto error;
3096 }
3097
3098 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3099 if (ret < 0) {
3100 free(domains);
3101 goto setup_error;
3102 }
3103
3104 /* Copy event list into message payload */
3105 memcpy(cmd_ctx->llm->payload, domains,
3106 nb_dom * sizeof(struct lttng_domain));
3107
3108 free(domains);
3109
3110 ret = LTTNG_OK;
3111 break;
3112 }
3113 case LTTNG_LIST_CHANNELS:
3114 {
3115 int nb_chan;
3116 struct lttng_channel *channels;
3117
3118 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3119 cmd_ctx->session, &channels);
3120 if (nb_chan < 0) {
3121 /* Return value is a negative lttng_error_code. */
3122 ret = -nb_chan;
3123 goto error;
3124 }
3125
3126 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3127 if (ret < 0) {
3128 free(channels);
3129 goto setup_error;
3130 }
3131
3132 /* Copy event list into message payload */
3133 memcpy(cmd_ctx->llm->payload, channels,
3134 nb_chan * sizeof(struct lttng_channel));
3135
3136 free(channels);
3137
3138 ret = LTTNG_OK;
3139 break;
3140 }
3141 case LTTNG_LIST_EVENTS:
3142 {
3143 ssize_t nb_event;
3144 struct lttng_event *events = NULL;
3145
3146 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3147 cmd_ctx->lsm->u.list.channel_name, &events);
3148 if (nb_event < 0) {
3149 /* Return value is a negative lttng_error_code. */
3150 ret = -nb_event;
3151 goto error;
3152 }
3153
3154 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3155 if (ret < 0) {
3156 free(events);
3157 goto setup_error;
3158 }
3159
3160 /* Copy event list into message payload */
3161 memcpy(cmd_ctx->llm->payload, events,
3162 nb_event * sizeof(struct lttng_event));
3163
3164 free(events);
3165
3166 ret = LTTNG_OK;
3167 break;
3168 }
3169 case LTTNG_LIST_SESSIONS:
3170 {
3171 unsigned int nr_sessions;
3172
3173 session_lock_list();
3174 nr_sessions = lttng_sessions_count(
3175 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3176 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3177
3178 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3179 if (ret < 0) {
3180 session_unlock_list();
3181 goto setup_error;
3182 }
3183
3184 /* Filled the session array */
3185 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3186 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3187 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3188
3189 session_unlock_list();
3190
3191 ret = LTTNG_OK;
3192 break;
3193 }
3194 case LTTNG_CALIBRATE:
3195 {
3196 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3197 &cmd_ctx->lsm->u.calibrate);
3198 break;
3199 }
3200 case LTTNG_REGISTER_CONSUMER:
3201 {
3202 struct consumer_data *cdata;
3203
3204 switch (cmd_ctx->lsm->domain.type) {
3205 case LTTNG_DOMAIN_KERNEL:
3206 cdata = &kconsumer_data;
3207 break;
3208 default:
3209 ret = LTTNG_ERR_UND;
3210 goto error;
3211 }
3212
3213 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3214 cmd_ctx->lsm->u.reg.path, cdata);
3215 break;
3216 }
3217 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3218 {
3219 struct lttng_filter_bytecode *bytecode;
3220
3221 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3222 ret = LTTNG_ERR_FILTER_INVAL;
3223 goto error;
3224 }
3225 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3226 ret = LTTNG_ERR_FILTER_INVAL;
3227 goto error;
3228 }
3229 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3230 if (!bytecode) {
3231 ret = LTTNG_ERR_FILTER_NOMEM;
3232 goto error;
3233 }
3234 /* Receive var. len. data */
3235 DBG("Receiving var len data from client ...");
3236 ret = lttcomm_recv_unix_sock(sock, bytecode,
3237 cmd_ctx->lsm->u.enable.bytecode_len);
3238 if (ret <= 0) {
3239 DBG("Nothing recv() from client var len data... continuing");
3240 *sock_error = 1;
3241 ret = LTTNG_ERR_FILTER_INVAL;
3242 goto error;
3243 }
3244
3245 if (bytecode->len + sizeof(*bytecode)
3246 != cmd_ctx->lsm->u.enable.bytecode_len) {
3247 free(bytecode);
3248 ret = LTTNG_ERR_FILTER_INVAL;
3249 goto error;
3250 }
3251
3252 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3253 cmd_ctx->lsm->u.enable.channel_name,
3254 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3255 break;
3256 }
3257 case LTTNG_DATA_PENDING:
3258 {
3259 ret = cmd_data_pending(cmd_ctx->session);
3260 break;
3261 }
3262 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3263 {
3264 struct lttcomm_lttng_output_id reply;
3265
3266 ret = cmd_snapshot_add_output(cmd_ctx->session,
3267 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3268 if (ret != LTTNG_OK) {
3269 goto error;
3270 }
3271
3272 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3273 if (ret < 0) {
3274 goto setup_error;
3275 }
3276
3277 /* Copy output list into message payload */
3278 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3279 ret = LTTNG_OK;
3280 break;
3281 }
3282 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3283 {
3284 ret = cmd_snapshot_del_output(cmd_ctx->session,
3285 &cmd_ctx->lsm->u.snapshot_output.output);
3286 break;
3287 }
3288 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3289 {
3290 ssize_t nb_output;
3291 struct lttng_snapshot_output *outputs = NULL;
3292
3293 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3294 if (nb_output < 0) {
3295 ret = -nb_output;
3296 goto error;
3297 }
3298
3299 ret = setup_lttng_msg(cmd_ctx,
3300 nb_output * sizeof(struct lttng_snapshot_output));
3301 if (ret < 0) {
3302 free(outputs);
3303 goto setup_error;
3304 }
3305
3306 if (outputs) {
3307 /* Copy output list into message payload */
3308 memcpy(cmd_ctx->llm->payload, outputs,
3309 nb_output * sizeof(struct lttng_snapshot_output));
3310 free(outputs);
3311 }
3312
3313 ret = LTTNG_OK;
3314 break;
3315 }
3316 case LTTNG_SNAPSHOT_RECORD:
3317 {
3318 ret = cmd_snapshot_record(cmd_ctx->session,
3319 &cmd_ctx->lsm->u.snapshot_record.output,
3320 cmd_ctx->lsm->u.snapshot_record.wait);
3321 break;
3322 }
3323 case LTTNG_CREATE_SESSION_SNAPSHOT:
3324 {
3325 size_t nb_uri, len;
3326 struct lttng_uri *uris = NULL;
3327
3328 nb_uri = cmd_ctx->lsm->u.uri.size;
3329 len = nb_uri * sizeof(struct lttng_uri);
3330
3331 if (nb_uri > 0) {
3332 uris = zmalloc(len);
3333 if (uris == NULL) {
3334 ret = LTTNG_ERR_FATAL;
3335 goto error;
3336 }
3337
3338 /* Receive variable len data */
3339 DBG("Waiting for %zu URIs from client ...", nb_uri);
3340 ret = lttcomm_recv_unix_sock(sock, uris, len);
3341 if (ret <= 0) {
3342 DBG("No URIs received from client... continuing");
3343 *sock_error = 1;
3344 ret = LTTNG_ERR_SESSION_FAIL;
3345 free(uris);
3346 goto error;
3347 }
3348
3349 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3350 DBG("Creating session with ONE network URI is a bad call");
3351 ret = LTTNG_ERR_SESSION_FAIL;
3352 free(uris);
3353 goto error;
3354 }
3355 }
3356
3357 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3358 nb_uri, &cmd_ctx->creds);
3359 free(uris);
3360 break;
3361 }
3362 default:
3363 ret = LTTNG_ERR_UND;
3364 break;
3365 }
3366
3367 error:
3368 if (cmd_ctx->llm == NULL) {
3369 DBG("Missing llm structure. Allocating one.");
3370 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3371 goto setup_error;
3372 }
3373 }
3374 /* Set return code */
3375 cmd_ctx->llm->ret_code = ret;
3376 setup_error:
3377 if (cmd_ctx->session) {
3378 session_unlock(cmd_ctx->session);
3379 }
3380 if (need_tracing_session) {
3381 session_unlock_list();
3382 }
3383 init_setup_error:
3384 return ret;
3385 }
3386
3387 /*
3388 * Thread managing health check socket.
3389 */
3390 static void *thread_manage_health(void *data)
3391 {
3392 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3393 uint32_t revents, nb_fd;
3394 struct lttng_poll_event events;
3395 struct lttcomm_health_msg msg;
3396 struct lttcomm_health_data reply;
3397
3398 DBG("[thread] Manage health check started");
3399
3400 rcu_register_thread();
3401
3402 /* We might hit an error path before this is created. */
3403 lttng_poll_init(&events);
3404
3405 /* Create unix socket */
3406 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3407 if (sock < 0) {
3408 ERR("Unable to create health check Unix socket");
3409 ret = -1;
3410 goto error;
3411 }
3412
3413 /*
3414 * Set the CLOEXEC flag. Return code is useless because either way, the
3415 * show must go on.
3416 */
3417 (void) utils_set_fd_cloexec(sock);
3418
3419 ret = lttcomm_listen_unix_sock(sock);
3420 if (ret < 0) {
3421 goto error;
3422 }
3423
3424 /*
3425 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3426 * more will be added to this poll set.
3427 */
3428 ret = sessiond_set_thread_pollset(&events, 2);
3429 if (ret < 0) {
3430 goto error;
3431 }
3432
3433 /* Add the application registration socket */
3434 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3435 if (ret < 0) {
3436 goto error;
3437 }
3438
3439 while (1) {
3440 DBG("Health check ready");
3441
3442 /* Inifinite blocking call, waiting for transmission */
3443 restart:
3444 ret = lttng_poll_wait(&events, -1);
3445 if (ret < 0) {
3446 /*
3447 * Restart interrupted system call.
3448 */
3449 if (errno == EINTR) {
3450 goto restart;
3451 }
3452 goto error;
3453 }
3454
3455 nb_fd = ret;
3456
3457 for (i = 0; i < nb_fd; i++) {
3458 /* Fetch once the poll data */
3459 revents = LTTNG_POLL_GETEV(&events, i);
3460 pollfd = LTTNG_POLL_GETFD(&events, i);
3461
3462 /* Thread quit pipe has been closed. Killing thread. */
3463 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3464 if (ret) {
3465 err = 0;
3466 goto exit;
3467 }
3468
3469 /* Event on the registration socket */
3470 if (pollfd == sock) {
3471 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3472 ERR("Health socket poll error");
3473 goto error;
3474 }
3475 }
3476 }
3477
3478 new_sock = lttcomm_accept_unix_sock(sock);
3479 if (new_sock < 0) {
3480 goto error;
3481 }
3482
3483 /*
3484 * Set the CLOEXEC flag. Return code is useless because either way, the
3485 * show must go on.
3486 */
3487 (void) utils_set_fd_cloexec(new_sock);
3488
3489 DBG("Receiving data from client for health...");
3490 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3491 if (ret <= 0) {
3492 DBG("Nothing recv() from client... continuing");
3493 ret = close(new_sock);
3494 if (ret) {
3495 PERROR("close");
3496 }
3497 new_sock = -1;
3498 continue;
3499 }
3500
3501 rcu_thread_online();
3502
3503 switch (msg.component) {
3504 case LTTNG_HEALTH_CMD:
3505 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3506 break;
3507 case LTTNG_HEALTH_APP_MANAGE:
3508 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3509 break;
3510 case LTTNG_HEALTH_APP_REG:
3511 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3512 break;
3513 case LTTNG_HEALTH_KERNEL:
3514 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3515 break;
3516 case LTTNG_HEALTH_CONSUMER:
3517 reply.ret_code = check_consumer_health();
3518 break;
3519 case LTTNG_HEALTH_HT_CLEANUP:
3520 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3521 break;
3522 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3523 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3524 break;
3525 case LTTNG_HEALTH_APP_REG_DISPATCH:
3526 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3527 break;
3528 case LTTNG_HEALTH_ALL:
3529 reply.ret_code =
3530 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3531 health_check_state(HEALTH_TYPE_APP_REG) &&
3532 health_check_state(HEALTH_TYPE_CMD) &&
3533 health_check_state(HEALTH_TYPE_KERNEL) &&
3534 check_consumer_health() &&
3535 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3536 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3537 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3538 break;
3539 default:
3540 reply.ret_code = LTTNG_ERR_UND;
3541 break;
3542 }
3543
3544 /*
3545 * Flip ret value since 0 is a success and 1 indicates a bad health for
3546 * the client where in the sessiond it is the opposite. Again, this is
3547 * just to make things easier for us poor developer which enjoy a lot
3548 * lazyness.
3549 */
3550 if (reply.ret_code == 0 || reply.ret_code == 1) {
3551 reply.ret_code = !reply.ret_code;
3552 }
3553
3554 DBG2("Health check return value %d", reply.ret_code);
3555
3556 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3557 if (ret < 0) {
3558 ERR("Failed to send health data back to client");
3559 }
3560
3561 /* End of transmission */
3562 ret = close(new_sock);
3563 if (ret) {
3564 PERROR("close");
3565 }
3566 new_sock = -1;
3567 }
3568
3569 exit:
3570 error:
3571 if (err) {
3572 ERR("Health error occurred in %s", __func__);
3573 }
3574 DBG("Health check thread dying");
3575 unlink(health_unix_sock_path);
3576 if (sock >= 0) {
3577 ret = close(sock);
3578 if (ret) {
3579 PERROR("close");
3580 }
3581 }
3582
3583 lttng_poll_clean(&events);
3584
3585 rcu_unregister_thread();
3586 return NULL;
3587 }
3588
3589 /*
3590 * This thread manage all clients request using the unix client socket for
3591 * communication.
3592 */
3593 static void *thread_manage_clients(void *data)
3594 {
3595 int sock = -1, ret, i, pollfd, err = -1;
3596 int sock_error;
3597 uint32_t revents, nb_fd;
3598 struct command_ctx *cmd_ctx = NULL;
3599 struct lttng_poll_event events;
3600
3601 DBG("[thread] Manage client started");
3602
3603 rcu_register_thread();
3604
3605 health_register(HEALTH_TYPE_CMD);
3606
3607 if (testpoint(thread_manage_clients)) {
3608 goto error_testpoint;
3609 }
3610
3611 health_code_update();
3612
3613 ret = lttcomm_listen_unix_sock(client_sock);
3614 if (ret < 0) {
3615 goto error_listen;
3616 }
3617
3618 /*
3619 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3620 * more will be added to this poll set.
3621 */
3622 ret = sessiond_set_thread_pollset(&events, 2);
3623 if (ret < 0) {
3624 goto error_create_poll;
3625 }
3626
3627 /* Add the application registration socket */
3628 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3629 if (ret < 0) {
3630 goto error;
3631 }
3632
3633 /*
3634 * Notify parent pid that we are ready to accept command for client side.
3635 */
3636 if (opt_sig_parent) {
3637 kill(ppid, SIGUSR1);
3638 }
3639
3640 if (testpoint(thread_manage_clients_before_loop)) {
3641 goto error;
3642 }
3643
3644 health_code_update();
3645
3646 while (1) {
3647 DBG("Accepting client command ...");
3648
3649 /* Inifinite blocking call, waiting for transmission */
3650 restart:
3651 health_poll_entry();
3652 ret = lttng_poll_wait(&events, -1);
3653 health_poll_exit();
3654 if (ret < 0) {
3655 /*
3656 * Restart interrupted system call.
3657 */
3658 if (errno == EINTR) {
3659 goto restart;
3660 }
3661 goto error;
3662 }
3663
3664 nb_fd = ret;
3665
3666 for (i = 0; i < nb_fd; i++) {
3667 /* Fetch once the poll data */
3668 revents = LTTNG_POLL_GETEV(&events, i);
3669 pollfd = LTTNG_POLL_GETFD(&events, i);
3670
3671 health_code_update();
3672
3673 /* Thread quit pipe has been closed. Killing thread. */
3674 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3675 if (ret) {
3676 err = 0;
3677 goto exit;
3678 }
3679
3680 /* Event on the registration socket */
3681 if (pollfd == client_sock) {
3682 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3683 ERR("Client socket poll error");
3684 goto error;
3685 }
3686 }
3687 }
3688
3689 DBG("Wait for client response");
3690
3691 health_code_update();
3692
3693 sock = lttcomm_accept_unix_sock(client_sock);
3694 if (sock < 0) {
3695 goto error;
3696 }
3697
3698 /*
3699 * Set the CLOEXEC flag. Return code is useless because either way, the
3700 * show must go on.
3701 */
3702 (void) utils_set_fd_cloexec(sock);
3703
3704 /* Set socket option for credentials retrieval */
3705 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3706 if (ret < 0) {
3707 goto error;
3708 }
3709
3710 /* Allocate context command to process the client request */
3711 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3712 if (cmd_ctx == NULL) {
3713 PERROR("zmalloc cmd_ctx");
3714 goto error;
3715 }
3716
3717 /* Allocate data buffer for reception */
3718 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3719 if (cmd_ctx->lsm == NULL) {
3720 PERROR("zmalloc cmd_ctx->lsm");
3721 goto error;
3722 }
3723
3724 cmd_ctx->llm = NULL;
3725 cmd_ctx->session = NULL;
3726
3727 health_code_update();
3728
3729 /*
3730 * Data is received from the lttng client. The struct
3731 * lttcomm_session_msg (lsm) contains the command and data request of
3732 * the client.
3733 */
3734 DBG("Receiving data from client ...");
3735 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3736 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3737 if (ret <= 0) {
3738 DBG("Nothing recv() from client... continuing");
3739 ret = close(sock);
3740 if (ret) {
3741 PERROR("close");
3742 }
3743 sock = -1;
3744 clean_command_ctx(&cmd_ctx);
3745 continue;
3746 }
3747
3748 health_code_update();
3749
3750 // TODO: Validate cmd_ctx including sanity check for
3751 // security purpose.
3752
3753 rcu_thread_online();
3754 /*
3755 * This function dispatch the work to the kernel or userspace tracer
3756 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3757 * informations for the client. The command context struct contains
3758 * everything this function may needs.
3759 */
3760 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3761 rcu_thread_offline();
3762 if (ret < 0) {
3763 ret = close(sock);
3764 if (ret) {
3765 PERROR("close");
3766 }
3767 sock = -1;
3768 /*
3769 * TODO: Inform client somehow of the fatal error. At
3770 * this point, ret < 0 means that a zmalloc failed
3771 * (ENOMEM). Error detected but still accept
3772 * command, unless a socket error has been
3773 * detected.
3774 */
3775 clean_command_ctx(&cmd_ctx);
3776 continue;
3777 }
3778
3779 health_code_update();
3780
3781 DBG("Sending response (size: %d, retcode: %s)",
3782 cmd_ctx->lttng_msg_size,
3783 lttng_strerror(-cmd_ctx->llm->ret_code));
3784 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3785 if (ret < 0) {
3786 ERR("Failed to send data back to client");
3787 }
3788
3789 /* End of transmission */
3790 ret = close(sock);
3791 if (ret) {
3792 PERROR("close");
3793 }
3794 sock = -1;
3795
3796 clean_command_ctx(&cmd_ctx);
3797
3798 health_code_update();
3799 }
3800
3801 exit:
3802 error:
3803 if (sock >= 0) {
3804 ret = close(sock);
3805 if (ret) {
3806 PERROR("close");
3807 }
3808 }
3809
3810 lttng_poll_clean(&events);
3811 clean_command_ctx(&cmd_ctx);
3812
3813 error_listen:
3814 error_create_poll:
3815 error_testpoint:
3816 unlink(client_unix_sock_path);
3817 if (client_sock >= 0) {
3818 ret = close(client_sock);
3819 if (ret) {
3820 PERROR("close");
3821 }
3822 }
3823
3824 if (err) {
3825 health_error();
3826 ERR("Health error occurred in %s", __func__);
3827 }
3828
3829 health_unregister();
3830
3831 DBG("Client thread dying");
3832
3833 rcu_unregister_thread();
3834 return NULL;
3835 }
3836
3837
3838 /*
3839 * usage function on stderr
3840 */
3841 static void usage(void)
3842 {
3843 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3844 fprintf(stderr, " -h, --help Display this usage.\n");
3845 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3846 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3847 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3848 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3849 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3850 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3851 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3852 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3853 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3854 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3855 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3856 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3857 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3858 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3859 fprintf(stderr, " -V, --version Show version number.\n");
3860 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3861 fprintf(stderr, " -q, --quiet No output at all.\n");
3862 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3863 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3864 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3865 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3866 }
3867
3868 /*
3869 * daemon argument parsing
3870 */
3871 static int parse_args(int argc, char **argv)
3872 {
3873 int c;
3874
3875 static struct option long_options[] = {
3876 { "client-sock", 1, 0, 'c' },
3877 { "apps-sock", 1, 0, 'a' },
3878 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3879 { "kconsumerd-err-sock", 1, 0, 'E' },
3880 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3881 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3882 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3883 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3884 { "consumerd32-path", 1, 0, 'u' },
3885 { "consumerd32-libdir", 1, 0, 'U' },
3886 { "consumerd64-path", 1, 0, 't' },
3887 { "consumerd64-libdir", 1, 0, 'T' },
3888 { "daemonize", 0, 0, 'd' },
3889 { "sig-parent", 0, 0, 'S' },
3890 { "help", 0, 0, 'h' },
3891 { "group", 1, 0, 'g' },
3892 { "version", 0, 0, 'V' },
3893 { "quiet", 0, 0, 'q' },
3894 { "verbose", 0, 0, 'v' },
3895 { "verbose-consumer", 0, 0, 'Z' },
3896 { "no-kernel", 0, 0, 'N' },
3897 { "pidfile", 1, 0, 'p' },
3898 { NULL, 0, 0, 0 }
3899 };
3900
3901 while (1) {
3902 int option_index = 0;
3903 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3904 long_options, &option_index);
3905 if (c == -1) {
3906 break;
3907 }
3908
3909 switch (c) {
3910 case 0:
3911 fprintf(stderr, "option %s", long_options[option_index].name);
3912 if (optarg) {
3913 fprintf(stderr, " with arg %s\n", optarg);
3914 }
3915 break;
3916 case 'c':
3917 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3918 break;
3919 case 'a':
3920 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3921 break;
3922 case 'd':
3923 opt_daemon = 1;
3924 break;
3925 case 'g':
3926 opt_tracing_group = optarg;
3927 break;
3928 case 'h':
3929 usage();
3930 exit(EXIT_FAILURE);
3931 case 'V':
3932 fprintf(stdout, "%s\n", VERSION);
3933 exit(EXIT_SUCCESS);
3934 case 'S':
3935 opt_sig_parent = 1;
3936 break;
3937 case 'E':
3938 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3939 break;
3940 case 'C':
3941 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3942 break;
3943 case 'F':
3944 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3945 break;
3946 case 'D':
3947 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3948 break;
3949 case 'H':
3950 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3951 break;
3952 case 'G':
3953 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3954 break;
3955 case 'N':
3956 opt_no_kernel = 1;
3957 break;
3958 case 'q':
3959 lttng_opt_quiet = 1;
3960 break;
3961 case 'v':
3962 /* Verbose level can increase using multiple -v */
3963 lttng_opt_verbose += 1;
3964 break;
3965 case 'Z':
3966 opt_verbose_consumer += 1;
3967 break;
3968 case 'u':
3969 consumerd32_bin= optarg;
3970 break;
3971 case 'U':
3972 consumerd32_libdir = optarg;
3973 break;
3974 case 't':
3975 consumerd64_bin = optarg;
3976 break;
3977 case 'T':
3978 consumerd64_libdir = optarg;
3979 break;
3980 case 'p':
3981 opt_pidfile = optarg;
3982 break;
3983 default:
3984 /* Unknown option or other error.
3985 * Error is printed by getopt, just return */
3986 return -1;
3987 }
3988 }
3989
3990 return 0;
3991 }
3992
3993 /*
3994 * Creates the two needed socket by the daemon.
3995 * apps_sock - The communication socket for all UST apps.
3996 * client_sock - The communication of the cli tool (lttng).
3997 */
3998 static int init_daemon_socket(void)
3999 {
4000 int ret = 0;
4001 mode_t old_umask;
4002
4003 old_umask = umask(0);
4004
4005 /* Create client tool unix socket */
4006 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4007 if (client_sock < 0) {
4008 ERR("Create unix sock failed: %s", client_unix_sock_path);
4009 ret = -1;
4010 goto end;
4011 }
4012
4013 /* Set the cloexec flag */
4014 ret = utils_set_fd_cloexec(client_sock);
4015 if (ret < 0) {
4016 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4017 "Continuing but note that the consumer daemon will have a "
4018 "reference to this socket on exec()", client_sock);
4019 }
4020
4021 /* File permission MUST be 660 */
4022 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4023 if (ret < 0) {
4024 ERR("Set file permissions failed: %s", client_unix_sock_path);
4025 PERROR("chmod");
4026 goto end;
4027 }
4028
4029 /* Create the application unix socket */
4030 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4031 if (apps_sock < 0) {
4032 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4033 ret = -1;
4034 goto end;
4035 }
4036
4037 /* Set the cloexec flag */
4038 ret = utils_set_fd_cloexec(apps_sock);
4039 if (ret < 0) {
4040 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4041 "Continuing but note that the consumer daemon will have a "
4042 "reference to this socket on exec()", apps_sock);
4043 }
4044
4045 /* File permission MUST be 666 */
4046 ret = chmod(apps_unix_sock_path,
4047 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4048 if (ret < 0) {
4049 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4050 PERROR("chmod");
4051 goto end;
4052 }
4053
4054 DBG3("Session daemon client socket %d and application socket %d created",
4055 client_sock, apps_sock);
4056
4057 end:
4058 umask(old_umask);
4059 return ret;
4060 }
4061
4062 /*
4063 * Check if the global socket is available, and if a daemon is answering at the
4064 * other side. If yes, error is returned.
4065 */
4066 static int check_existing_daemon(void)
4067 {
4068 /* Is there anybody out there ? */
4069 if (lttng_session_daemon_alive()) {
4070 return -EEXIST;
4071 }
4072
4073 return 0;
4074 }
4075
4076 /*
4077 * Set the tracing group gid onto the client socket.
4078 *
4079 * Race window between mkdir and chown is OK because we are going from more
4080 * permissive (root.root) to less permissive (root.tracing).
4081 */
4082 static int set_permissions(char *rundir)
4083 {
4084 int ret;
4085 gid_t gid;
4086
4087 ret = allowed_group();
4088 if (ret < 0) {
4089 WARN("No tracing group detected");
4090 /* Setting gid to 0 if no tracing group is found */
4091 gid = 0;
4092 } else {
4093 gid = ret;
4094 }
4095
4096 /* Set lttng run dir */
4097 ret = chown(rundir, 0, gid);
4098 if (ret < 0) {
4099 ERR("Unable to set group on %s", rundir);
4100 PERROR("chown");
4101 }
4102
4103 /* Ensure all applications and tracing group can search the run dir */
4104 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
4105 if (ret < 0) {
4106 ERR("Unable to set permissions on %s", rundir);
4107 PERROR("chmod");
4108 }
4109
4110 /* lttng client socket path */
4111 ret = chown(client_unix_sock_path, 0, gid);
4112 if (ret < 0) {
4113 ERR("Unable to set group on %s", client_unix_sock_path);
4114 PERROR("chown");
4115 }
4116
4117 /* kconsumer error socket path */
4118 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4119 if (ret < 0) {
4120 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4121 PERROR("chown");
4122 }
4123
4124 /* 64-bit ustconsumer error socket path */
4125 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4126 if (ret < 0) {
4127 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4128 PERROR("chown");
4129 }
4130
4131 /* 32-bit ustconsumer compat32 error socket path */
4132 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4133 if (ret < 0) {
4134 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4135 PERROR("chown");
4136 }
4137
4138 DBG("All permissions are set");
4139
4140 return ret;
4141 }
4142
4143 /*
4144 * Create the lttng run directory needed for all global sockets and pipe.
4145 */
4146 static int create_lttng_rundir(const char *rundir)
4147 {
4148 int ret;
4149
4150 DBG3("Creating LTTng run directory: %s", rundir);
4151
4152 ret = mkdir(rundir, S_IRWXU);
4153 if (ret < 0) {
4154 if (errno != EEXIST) {
4155 ERR("Unable to create %s", rundir);
4156 goto error;
4157 } else {
4158 ret = 0;
4159 }
4160 }
4161
4162 error:
4163 return ret;
4164 }
4165
4166 /*
4167 * Setup sockets and directory needed by the kconsumerd communication with the
4168 * session daemon.
4169 */
4170 static int set_consumer_sockets(struct consumer_data *consumer_data,
4171 const char *rundir)
4172 {
4173 int ret;
4174 char path[PATH_MAX];
4175
4176 switch (consumer_data->type) {
4177 case LTTNG_CONSUMER_KERNEL:
4178 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4179 break;
4180 case LTTNG_CONSUMER64_UST:
4181 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4182 break;
4183 case LTTNG_CONSUMER32_UST:
4184 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4185 break;
4186 default:
4187 ERR("Consumer type unknown");
4188 ret = -EINVAL;
4189 goto error;
4190 }
4191
4192 DBG2("Creating consumer directory: %s", path);
4193
4194 ret = mkdir(path, S_IRWXU);
4195 if (ret < 0) {
4196 if (errno != EEXIST) {
4197 PERROR("mkdir");
4198 ERR("Failed to create %s", path);
4199 goto error;
4200 }
4201 ret = -1;
4202 }
4203
4204 /* Create the kconsumerd error unix socket */
4205 consumer_data->err_sock =
4206 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4207 if (consumer_data->err_sock < 0) {
4208 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4209 ret = -1;
4210 goto error;
4211 }
4212
4213 /*
4214 * Set the CLOEXEC flag. Return code is useless because either way, the
4215 * show must go on.
4216 */
4217 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4218 if (ret < 0) {
4219 PERROR("utils_set_fd_cloexec");
4220 /* continue anyway */
4221 }
4222
4223 /* File permission MUST be 660 */
4224 ret = chmod(consumer_data->err_unix_sock_path,
4225 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4226 if (ret < 0) {
4227 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4228 PERROR("chmod");
4229 goto error;
4230 }
4231
4232 error:
4233 return ret;
4234 }
4235
4236 /*
4237 * Signal handler for the daemon
4238 *
4239 * Simply stop all worker threads, leaving main() return gracefully after
4240 * joining all threads and calling cleanup().
4241 */
4242 static void sighandler(int sig)
4243 {
4244 switch (sig) {
4245 case SIGPIPE:
4246 DBG("SIGPIPE caught");
4247 return;
4248 case SIGINT:
4249 DBG("SIGINT caught");
4250 stop_threads();
4251 break;
4252 case SIGTERM:
4253 DBG("SIGTERM caught");
4254 stop_threads();
4255 break;
4256 default:
4257 break;
4258 }
4259 }
4260
4261 /*
4262 * Setup signal handler for :
4263 * SIGINT, SIGTERM, SIGPIPE
4264 */
4265 static int set_signal_handler(void)
4266 {
4267 int ret = 0;
4268 struct sigaction sa;
4269 sigset_t sigset;
4270
4271 if ((ret = sigemptyset(&sigset)) < 0) {
4272 PERROR("sigemptyset");
4273 return ret;
4274 }
4275
4276 sa.sa_handler = sighandler;
4277 sa.sa_mask = sigset;
4278 sa.sa_flags = 0;
4279 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4280 PERROR("sigaction");
4281 return ret;
4282 }
4283
4284 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4285 PERROR("sigaction");
4286 return ret;
4287 }
4288
4289 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4290 PERROR("sigaction");
4291 return ret;
4292 }
4293
4294 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4295
4296 return ret;
4297 }
4298
4299 /*
4300 * Set open files limit to unlimited. This daemon can open a large number of
4301 * file descriptors in order to consumer multiple kernel traces.
4302 */
4303 static void set_ulimit(void)
4304 {
4305 int ret;
4306 struct rlimit lim;
4307
4308 /* The kernel does not allowed an infinite limit for open files */
4309 lim.rlim_cur = 65535;
4310 lim.rlim_max = 65535;
4311
4312 ret = setrlimit(RLIMIT_NOFILE, &lim);
4313 if (ret < 0) {
4314 PERROR("failed to set open files limit");
4315 }
4316 }
4317
4318 /*
4319 * Write pidfile using the rundir and opt_pidfile.
4320 */
4321 static void write_pidfile(void)
4322 {
4323 int ret;
4324 char pidfile_path[PATH_MAX];
4325
4326 assert(rundir);
4327
4328 if (opt_pidfile) {
4329 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4330 } else {
4331 /* Build pidfile path from rundir and opt_pidfile. */
4332 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4333 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4334 if (ret < 0) {
4335 PERROR("snprintf pidfile path");
4336 goto error;
4337 }
4338 }
4339
4340 /*
4341 * Create pid file in rundir. Return value is of no importance. The
4342 * execution will continue even though we are not able to write the file.
4343 */
4344 (void) utils_create_pid_file(getpid(), pidfile_path);
4345
4346 error:
4347 return;
4348 }
4349
4350 /*
4351 * main
4352 */
4353 int main(int argc, char **argv)
4354 {
4355 int ret = 0;
4356 void *status;
4357 const char *home_path, *env_app_timeout;
4358
4359 init_kernel_workarounds();
4360
4361 rcu_register_thread();
4362
4363 setup_consumerd_path();
4364
4365 page_size = sysconf(_SC_PAGESIZE);
4366 if (page_size < 0) {
4367 PERROR("sysconf _SC_PAGESIZE");
4368 page_size = LONG_MAX;
4369 WARN("Fallback page size to %ld", page_size);
4370 }
4371
4372 /* Parse arguments */
4373 progname = argv[0];
4374 if ((ret = parse_args(argc, argv)) < 0) {
4375 goto error;
4376 }
4377
4378 /* Daemonize */
4379 if (opt_daemon) {
4380 int i;
4381
4382 /*
4383 * fork
4384 * child: setsid, close FD 0, 1, 2, chdir /
4385 * parent: exit (if fork is successful)
4386 */
4387 ret = daemon(0, 0);
4388 if (ret < 0) {
4389 PERROR("daemon");
4390 goto error;
4391 }
4392 /*
4393 * We are in the child. Make sure all other file
4394 * descriptors are closed, in case we are called with
4395 * more opened file descriptors than the standard ones.
4396 */
4397 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4398 (void) close(i);
4399 }
4400 }
4401
4402 /* Create thread quit pipe */
4403 if ((ret = init_thread_quit_pipe()) < 0) {
4404 goto error;
4405 }
4406
4407 /* Check if daemon is UID = 0 */
4408 is_root = !getuid();
4409
4410 if (is_root) {
4411 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4412
4413 /* Create global run dir with root access */
4414 ret = create_lttng_rundir(rundir);
4415 if (ret < 0) {
4416 goto error;
4417 }
4418
4419 if (strlen(apps_unix_sock_path) == 0) {
4420 snprintf(apps_unix_sock_path, PATH_MAX,
4421 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4422 }
4423
4424 if (strlen(client_unix_sock_path) == 0) {
4425 snprintf(client_unix_sock_path, PATH_MAX,
4426 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4427 }
4428
4429 /* Set global SHM for ust */
4430 if (strlen(wait_shm_path) == 0) {
4431 snprintf(wait_shm_path, PATH_MAX,
4432 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4433 }
4434
4435 if (strlen(health_unix_sock_path) == 0) {
4436 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4437 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4438 }
4439
4440 /* Setup kernel consumerd path */
4441 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4442 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4443 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4444 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4445
4446 DBG2("Kernel consumer err path: %s",
4447 kconsumer_data.err_unix_sock_path);
4448 DBG2("Kernel consumer cmd path: %s",
4449 kconsumer_data.cmd_unix_sock_path);
4450 } else {
4451 home_path = utils_get_home_dir();
4452 if (home_path == NULL) {
4453 /* TODO: Add --socket PATH option */
4454 ERR("Can't get HOME directory for sockets creation.");
4455 ret = -EPERM;
4456 goto error;
4457 }
4458
4459 /*
4460 * Create rundir from home path. This will create something like
4461 * $HOME/.lttng
4462 */
4463 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4464 if (ret < 0) {
4465 ret = -ENOMEM;
4466 goto error;
4467 }
4468
4469 ret = create_lttng_rundir(rundir);
4470 if (ret < 0) {
4471 goto error;
4472 }
4473
4474 if (strlen(apps_unix_sock_path) == 0) {
4475 snprintf(apps_unix_sock_path, PATH_MAX,
4476 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4477 }
4478
4479 /* Set the cli tool unix socket path */
4480 if (strlen(client_unix_sock_path) == 0) {
4481 snprintf(client_unix_sock_path, PATH_MAX,
4482 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4483 }
4484
4485 /* Set global SHM for ust */
4486 if (strlen(wait_shm_path) == 0) {
4487 snprintf(wait_shm_path, PATH_MAX,
4488 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4489 }
4490
4491 /* Set health check Unix path */
4492 if (strlen(health_unix_sock_path) == 0) {
4493 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4494 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4495 }
4496 }
4497
4498 /* Set consumer initial state */
4499 kernel_consumerd_state = CONSUMER_STOPPED;
4500 ust_consumerd_state = CONSUMER_STOPPED;
4501
4502 DBG("Client socket path %s", client_unix_sock_path);
4503 DBG("Application socket path %s", apps_unix_sock_path);
4504 DBG("Application wait path %s", wait_shm_path);
4505 DBG("LTTng run directory path: %s", rundir);
4506
4507 /* 32 bits consumerd path setup */
4508 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4509 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4510 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4511 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4512
4513 DBG2("UST consumer 32 bits err path: %s",
4514 ustconsumer32_data.err_unix_sock_path);
4515 DBG2("UST consumer 32 bits cmd path: %s",
4516 ustconsumer32_data.cmd_unix_sock_path);
4517
4518 /* 64 bits consumerd path setup */
4519 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4520 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4521 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4522 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4523
4524 DBG2("UST consumer 64 bits err path: %s",
4525 ustconsumer64_data.err_unix_sock_path);
4526 DBG2("UST consumer 64 bits cmd path: %s",
4527 ustconsumer64_data.cmd_unix_sock_path);
4528
4529 /*
4530 * See if daemon already exist.
4531 */
4532 if ((ret = check_existing_daemon()) < 0) {
4533 ERR("Already running daemon.\n");
4534 /*
4535 * We do not goto exit because we must not cleanup()
4536 * because a daemon is already running.
4537 */
4538 goto error;
4539 }
4540
4541 /*
4542 * Init UST app hash table. Alloc hash table before this point since
4543 * cleanup() can get called after that point.
4544 */
4545 ust_app_ht_alloc();
4546
4547 /* After this point, we can safely call cleanup() with "goto exit" */
4548
4549 /*
4550 * These actions must be executed as root. We do that *after* setting up
4551 * the sockets path because we MUST make the check for another daemon using
4552 * those paths *before* trying to set the kernel consumer sockets and init
4553 * kernel tracer.
4554 */
4555 if (is_root) {
4556 ret = set_consumer_sockets(&kconsumer_data, rundir);
4557 if (ret < 0) {
4558 goto exit;
4559 }
4560
4561 /* Setup kernel tracer */
4562 if (!opt_no_kernel) {
4563 init_kernel_tracer();
4564 }
4565
4566 /* Set ulimit for open files */
4567 set_ulimit();
4568 }
4569 /* init lttng_fd tracking must be done after set_ulimit. */
4570 lttng_fd_init();
4571
4572 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4573 if (ret < 0) {
4574 goto exit;
4575 }
4576
4577 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4578 if (ret < 0) {
4579 goto exit;
4580 }
4581
4582 if ((ret = set_signal_handler()) < 0) {
4583 goto exit;
4584 }
4585
4586 /* Setup the needed unix socket */
4587 if ((ret = init_daemon_socket()) < 0) {
4588 goto exit;
4589 }
4590
4591 /* Set credentials to socket */
4592 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4593 goto exit;
4594 }
4595
4596 /* Get parent pid if -S, --sig-parent is specified. */
4597 if (opt_sig_parent) {
4598 ppid = getppid();
4599 }
4600
4601 /* Setup the kernel pipe for waking up the kernel thread */
4602 if (is_root && !opt_no_kernel) {
4603 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4604 goto exit;
4605 }
4606 }
4607
4608 /* Setup the thread ht_cleanup communication pipe. */
4609 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4610 goto exit;
4611 }
4612
4613 /* Setup the thread apps communication pipe. */
4614 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4615 goto exit;
4616 }
4617
4618 /* Setup the thread apps notify communication pipe. */
4619 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4620 goto exit;
4621 }
4622
4623 /* Initialize global buffer per UID and PID registry. */
4624 buffer_reg_init_uid_registry();
4625 buffer_reg_init_pid_registry();
4626
4627 /* Init UST command queue. */
4628 cds_wfq_init(&ust_cmd_queue.queue);
4629
4630 /*
4631 * Get session list pointer. This pointer MUST NOT be free(). This list is
4632 * statically declared in session.c
4633 */
4634 session_list_ptr = session_get_list();
4635
4636 /* Set up max poll set size */
4637 lttng_poll_set_max_size();
4638
4639 cmd_init();
4640
4641 /* Check for the application socket timeout env variable. */
4642 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4643 if (env_app_timeout) {
4644 app_socket_timeout = atoi(env_app_timeout);
4645 } else {
4646 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4647 }
4648
4649 write_pidfile();
4650
4651 /* Initialize communication library */
4652 lttcomm_init();
4653
4654 /* Create thread to manage the client socket */
4655 ret = pthread_create(&ht_cleanup_thread, NULL,
4656 thread_ht_cleanup, (void *) NULL);
4657 if (ret != 0) {
4658 PERROR("pthread_create ht_cleanup");
4659 goto exit_ht_cleanup;
4660 }
4661
4662 /* Create thread to manage the client socket */
4663 ret = pthread_create(&health_thread, NULL,
4664 thread_manage_health, (void *) NULL);
4665 if (ret != 0) {
4666 PERROR("pthread_create health");
4667 goto exit_health;
4668 }
4669
4670 /* Create thread to manage the client socket */
4671 ret = pthread_create(&client_thread, NULL,
4672 thread_manage_clients, (void *) NULL);
4673 if (ret != 0) {
4674 PERROR("pthread_create clients");
4675 goto exit_client;
4676 }
4677
4678 /* Create thread to dispatch registration */
4679 ret = pthread_create(&dispatch_thread, NULL,
4680 thread_dispatch_ust_registration, (void *) NULL);
4681 if (ret != 0) {
4682 PERROR("pthread_create dispatch");
4683 goto exit_dispatch;
4684 }
4685
4686 /* Create thread to manage application registration. */
4687 ret = pthread_create(&reg_apps_thread, NULL,
4688 thread_registration_apps, (void *) NULL);
4689 if (ret != 0) {
4690 PERROR("pthread_create registration");
4691 goto exit_reg_apps;
4692 }
4693
4694 /* Create thread to manage application socket */
4695 ret = pthread_create(&apps_thread, NULL,
4696 thread_manage_apps, (void *) NULL);
4697 if (ret != 0) {
4698 PERROR("pthread_create apps");
4699 goto exit_apps;
4700 }
4701
4702 /* Create thread to manage application notify socket */
4703 ret = pthread_create(&apps_notify_thread, NULL,
4704 ust_thread_manage_notify, (void *) NULL);
4705 if (ret != 0) {
4706 PERROR("pthread_create apps");
4707 goto exit_apps;
4708 }
4709
4710 /* Don't start this thread if kernel tracing is not requested nor root */
4711 if (is_root && !opt_no_kernel) {
4712 /* Create kernel thread to manage kernel event */
4713 ret = pthread_create(&kernel_thread, NULL,
4714 thread_manage_kernel, (void *) NULL);
4715 if (ret != 0) {
4716 PERROR("pthread_create kernel");
4717 goto exit_kernel;
4718 }
4719
4720 ret = pthread_join(kernel_thread, &status);
4721 if (ret != 0) {
4722 PERROR("pthread_join");
4723 goto error; /* join error, exit without cleanup */
4724 }
4725 }
4726
4727 exit_kernel:
4728 ret = pthread_join(apps_thread, &status);
4729 if (ret != 0) {
4730 PERROR("pthread_join");
4731 goto error; /* join error, exit without cleanup */
4732 }
4733
4734 exit_apps:
4735 ret = pthread_join(reg_apps_thread, &status);
4736 if (ret != 0) {
4737 PERROR("pthread_join");
4738 goto error; /* join error, exit without cleanup */
4739 }
4740
4741 exit_reg_apps:
4742 ret = pthread_join(dispatch_thread, &status);
4743 if (ret != 0) {
4744 PERROR("pthread_join");
4745 goto error; /* join error, exit without cleanup */
4746 }
4747
4748 exit_dispatch:
4749 ret = pthread_join(client_thread, &status);
4750 if (ret != 0) {
4751 PERROR("pthread_join");
4752 goto error; /* join error, exit without cleanup */
4753 }
4754
4755 ret = join_consumer_thread(&kconsumer_data);
4756 if (ret != 0) {
4757 PERROR("join_consumer");
4758 goto error; /* join error, exit without cleanup */
4759 }
4760
4761 ret = join_consumer_thread(&ustconsumer32_data);
4762 if (ret != 0) {
4763 PERROR("join_consumer ust32");
4764 goto error; /* join error, exit without cleanup */
4765 }
4766
4767 ret = join_consumer_thread(&ustconsumer64_data);
4768 if (ret != 0) {
4769 PERROR("join_consumer ust64");
4770 goto error; /* join error, exit without cleanup */
4771 }
4772
4773 exit_client:
4774 ret = pthread_join(health_thread, &status);
4775 if (ret != 0) {
4776 PERROR("pthread_join health thread");
4777 goto error; /* join error, exit without cleanup */
4778 }
4779
4780 exit_health:
4781 ret = pthread_join(ht_cleanup_thread, &status);
4782 if (ret != 0) {
4783 PERROR("pthread_join ht cleanup thread");
4784 goto error; /* join error, exit without cleanup */
4785 }
4786 exit_ht_cleanup:
4787 exit:
4788 /*
4789 * cleanup() is called when no other thread is running.
4790 */
4791 rcu_thread_online();
4792 cleanup();
4793 rcu_thread_offline();
4794 rcu_unregister_thread();
4795 if (!ret) {
4796 exit(EXIT_SUCCESS);
4797 }
4798 error:
4799 exit(EXIT_FAILURE);
4800 }
This page took 0.182932 seconds and 5 git commands to generate.