Use consumer fd reference in consumer socket obj
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <inttypes.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <config.h>
39
40 #include <common/common.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "buffer-registry.h"
50 #include "channel.h"
51 #include "cmd.h"
52 #include "consumer.h"
53 #include "context.h"
54 #include "event.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "modprobe.h"
58 #include "shm.h"
59 #include "ust-ctl.h"
60 #include "ust-consumer.h"
61 #include "utils.h"
62 #include "fd-limit.h"
63 #include "health.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66
67 #define CONSUMERD_FILE "lttng-consumerd"
68
69 /* Const values */
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71
72 const char *progname;
73 const char *opt_tracing_group;
74 static const char *opt_pidfile;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 int apps_cmd_notify_pipe[2] = { -1, -1 };
152
153 /* Pthread, Mutexes and Semaphores */
154 static pthread_t apps_thread;
155 static pthread_t apps_notify_thread;
156 static pthread_t reg_apps_thread;
157 static pthread_t client_thread;
158 static pthread_t kernel_thread;
159 static pthread_t dispatch_thread;
160 static pthread_t health_thread;
161 static pthread_t ht_cleanup_thread;
162
163 /*
164 * UST registration command queue. This queue is tied with a futex and uses a N
165 * wakers / 1 waiter implemented and detailed in futex.c/.h
166 *
167 * The thread_manage_apps and thread_dispatch_ust_registration interact with
168 * this queue and the wait/wake scheme.
169 */
170 static struct ust_cmd_queue ust_cmd_queue;
171
172 /*
173 * Pointer initialized before thread creation.
174 *
175 * This points to the tracing session list containing the session count and a
176 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
177 * MUST NOT be taken if you call a public function in session.c.
178 *
179 * The lock is nested inside the structure: session_list_ptr->lock. Please use
180 * session_lock_list and session_unlock_list for lock acquisition.
181 */
182 static struct ltt_session_list *session_list_ptr;
183
184 int ust_consumerd64_fd = -1;
185 int ust_consumerd32_fd = -1;
186
187 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
188 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
189 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
190 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
191
192 static const char *module_proc_lttng = "/proc/lttng";
193
194 /*
195 * Consumer daemon state which is changed when spawning it, killing it or in
196 * case of a fatal error.
197 */
198 enum consumerd_state {
199 CONSUMER_STARTED = 1,
200 CONSUMER_STOPPED = 2,
201 CONSUMER_ERROR = 3,
202 };
203
204 /*
205 * This consumer daemon state is used to validate if a client command will be
206 * able to reach the consumer. If not, the client is informed. For instance,
207 * doing a "lttng start" when the consumer state is set to ERROR will return an
208 * error to the client.
209 *
210 * The following example shows a possible race condition of this scheme:
211 *
212 * consumer thread error happens
213 * client cmd arrives
214 * client cmd checks state -> still OK
215 * consumer thread exit, sets error
216 * client cmd try to talk to consumer
217 * ...
218 *
219 * However, since the consumer is a different daemon, we have no way of making
220 * sure the command will reach it safely even with this state flag. This is why
221 * we consider that up to the state validation during command processing, the
222 * command is safe. After that, we can not guarantee the correctness of the
223 * client request vis-a-vis the consumer.
224 */
225 static enum consumerd_state ust_consumerd_state;
226 static enum consumerd_state kernel_consumerd_state;
227
228 /*
229 * Socket timeout for receiving and sending in seconds.
230 */
231 static int app_socket_timeout;
232
233 /* Set in main() with the current page size. */
234 long page_size;
235
236 static
237 void setup_consumerd_path(void)
238 {
239 const char *bin, *libdir;
240
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
245 */
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
256 }
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
259 }
260 #else
261 #error "Unknown bitness"
262 #endif
263
264 /*
265 * runtime env. var. overrides the build default.
266 */
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
270 }
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
274 }
275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
280 if (libdir) {
281 consumerd64_libdir = libdir;
282 }
283 }
284
285 /*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
289 {
290 int ret;
291
292 assert(events);
293
294 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
295 if (ret < 0) {
296 goto error;
297 }
298
299 /* Add quit pipe */
300 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
301 if (ret < 0) {
302 goto error;
303 }
304
305 return 0;
306
307 error:
308 return ret;
309 }
310
311 /*
312 * Check if the thread quit pipe was triggered.
313 *
314 * Return 1 if it was triggered else 0;
315 */
316 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
317 {
318 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
319 return 1;
320 }
321
322 return 0;
323 }
324
325 /*
326 * Return group ID of the tracing group or -1 if not found.
327 */
328 static gid_t allowed_group(void)
329 {
330 struct group *grp;
331
332 if (opt_tracing_group) {
333 grp = getgrnam(opt_tracing_group);
334 } else {
335 grp = getgrnam(default_tracing_group);
336 }
337 if (!grp) {
338 return -1;
339 } else {
340 return grp->gr_gid;
341 }
342 }
343
344 /*
345 * Init thread quit pipe.
346 *
347 * Return -1 on error or 0 if all pipes are created.
348 */
349 static int init_thread_quit_pipe(void)
350 {
351 int ret, i;
352
353 ret = pipe(thread_quit_pipe);
354 if (ret < 0) {
355 PERROR("thread quit pipe");
356 goto error;
357 }
358
359 for (i = 0; i < 2; i++) {
360 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
361 if (ret < 0) {
362 PERROR("fcntl");
363 goto error;
364 }
365 }
366
367 error:
368 return ret;
369 }
370
371 /*
372 * Stop all threads by closing the thread quit pipe.
373 */
374 static void stop_threads(void)
375 {
376 int ret;
377
378 /* Stopping all threads */
379 DBG("Terminating all threads");
380 ret = notify_thread_pipe(thread_quit_pipe[1]);
381 if (ret < 0) {
382 ERR("write error on thread quit pipe");
383 }
384
385 /* Dispatch thread */
386 CMM_STORE_SHARED(dispatch_thread_exit, 1);
387 futex_nto1_wake(&ust_cmd_queue.futex);
388 }
389
390 /*
391 * Close every consumer sockets.
392 */
393 static void close_consumer_sockets(void)
394 {
395 int ret;
396
397 if (kconsumer_data.err_sock >= 0) {
398 ret = close(kconsumer_data.err_sock);
399 if (ret < 0) {
400 PERROR("kernel consumer err_sock close");
401 }
402 }
403 if (ustconsumer32_data.err_sock >= 0) {
404 ret = close(ustconsumer32_data.err_sock);
405 if (ret < 0) {
406 PERROR("UST consumerd32 err_sock close");
407 }
408 }
409 if (ustconsumer64_data.err_sock >= 0) {
410 ret = close(ustconsumer64_data.err_sock);
411 if (ret < 0) {
412 PERROR("UST consumerd64 err_sock close");
413 }
414 }
415 if (kconsumer_data.cmd_sock >= 0) {
416 ret = close(kconsumer_data.cmd_sock);
417 if (ret < 0) {
418 PERROR("kernel consumer cmd_sock close");
419 }
420 }
421 if (ustconsumer32_data.cmd_sock >= 0) {
422 ret = close(ustconsumer32_data.cmd_sock);
423 if (ret < 0) {
424 PERROR("UST consumerd32 cmd_sock close");
425 }
426 }
427 if (ustconsumer64_data.cmd_sock >= 0) {
428 ret = close(ustconsumer64_data.cmd_sock);
429 if (ret < 0) {
430 PERROR("UST consumerd64 cmd_sock close");
431 }
432 }
433 }
434
435 /*
436 * Cleanup the daemon
437 */
438 static void cleanup(void)
439 {
440 int ret;
441 char *cmd = NULL;
442 struct ltt_session *sess, *stmp;
443
444 DBG("Cleaning up");
445
446 /*
447 * Close the thread quit pipe. It has already done its job,
448 * since we are now called.
449 */
450 utils_close_pipe(thread_quit_pipe);
451
452 /*
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
455 */
456 if (opt_pidfile) {
457 ret = remove(opt_pidfile);
458 if (ret < 0) {
459 PERROR("remove pidfile %s", opt_pidfile);
460 }
461 }
462
463 DBG("Removing %s directory", rundir);
464 ret = asprintf(&cmd, "rm -rf %s", rundir);
465 if (ret < 0) {
466 ERR("asprintf failed. Something is really wrong!");
467 }
468
469 /* Remove lttng run directory */
470 ret = system(cmd);
471 if (ret < 0) {
472 ERR("Unable to clean %s", rundir);
473 }
474 free(cmd);
475 free(rundir);
476
477 DBG("Cleaning up all sessions");
478
479 /* Destroy session list mutex */
480 if (session_list_ptr != NULL) {
481 pthread_mutex_destroy(&session_list_ptr->lock);
482
483 /* Cleanup ALL session */
484 cds_list_for_each_entry_safe(sess, stmp,
485 &session_list_ptr->head, list) {
486 cmd_destroy_session(sess, kernel_poll_pipe[1]);
487 }
488 }
489
490 DBG("Closing all UST sockets");
491 ust_app_clean_list();
492 buffer_reg_destroy_registries();
493
494 if (is_root && !opt_no_kernel) {
495 DBG2("Closing kernel fd");
496 if (kernel_tracer_fd >= 0) {
497 ret = close(kernel_tracer_fd);
498 if (ret) {
499 PERROR("close");
500 }
501 }
502 DBG("Unloading kernel modules");
503 modprobe_remove_lttng_all();
504 }
505
506 close_consumer_sockets();
507
508 /* <fun> */
509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
512 /* </fun> */
513 }
514
515 /*
516 * Send data on a unix socket using the liblttsessiondcomm API.
517 *
518 * Return lttcomm error code.
519 */
520 static int send_unix_sock(int sock, void *buf, size_t len)
521 {
522 /* Check valid length */
523 if (len == 0) {
524 return -1;
525 }
526
527 return lttcomm_send_unix_sock(sock, buf, len);
528 }
529
530 /*
531 * Free memory of a command context structure.
532 */
533 static void clean_command_ctx(struct command_ctx **cmd_ctx)
534 {
535 DBG("Clean command context structure");
536 if (*cmd_ctx) {
537 if ((*cmd_ctx)->llm) {
538 free((*cmd_ctx)->llm);
539 }
540 if ((*cmd_ctx)->lsm) {
541 free((*cmd_ctx)->lsm);
542 }
543 free(*cmd_ctx);
544 *cmd_ctx = NULL;
545 }
546 }
547
548 /*
549 * Notify UST applications using the shm mmap futex.
550 */
551 static int notify_ust_apps(int active)
552 {
553 char *wait_shm_mmap;
554
555 DBG("Notifying applications of session daemon state: %d", active);
556
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
559 if (wait_shm_mmap == NULL) {
560 goto error;
561 }
562
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap, active);
565
566 /* Apps notified successfully */
567 return 0;
568
569 error:
570 return -1;
571 }
572
573 /*
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
576 * structure.
577 *
578 * Return total size of the buffer pointed by buf.
579 */
580 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
581 {
582 int ret, buf_size;
583
584 buf_size = size;
585
586 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
587 if (cmd_ctx->llm == NULL) {
588 PERROR("zmalloc");
589 ret = -ENOMEM;
590 goto error;
591 }
592
593 /* Copy common data */
594 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
595 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
596
597 cmd_ctx->llm->data_size = size;
598 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
599
600 return buf_size;
601
602 error:
603 return ret;
604 }
605
606 /*
607 * Update the kernel poll set of all channel fd available over all tracing
608 * session. Add the wakeup pipe at the end of the set.
609 */
610 static int update_kernel_poll(struct lttng_poll_event *events)
611 {
612 int ret;
613 struct ltt_session *session;
614 struct ltt_kernel_channel *channel;
615
616 DBG("Updating kernel poll set");
617
618 session_lock_list();
619 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
620 session_lock(session);
621 if (session->kernel_session == NULL) {
622 session_unlock(session);
623 continue;
624 }
625
626 cds_list_for_each_entry(channel,
627 &session->kernel_session->channel_list.head, list) {
628 /* Add channel fd to the kernel poll set */
629 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
630 if (ret < 0) {
631 session_unlock(session);
632 goto error;
633 }
634 DBG("Channel fd %d added to kernel set", channel->fd);
635 }
636 session_unlock(session);
637 }
638 session_unlock_list();
639
640 return 0;
641
642 error:
643 session_unlock_list();
644 return -1;
645 }
646
647 /*
648 * Find the channel fd from 'fd' over all tracing session. When found, check
649 * for new channel stream and send those stream fds to the kernel consumer.
650 *
651 * Useful for CPU hotplug feature.
652 */
653 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
654 {
655 int ret = 0;
656 struct ltt_session *session;
657 struct ltt_kernel_session *ksess;
658 struct ltt_kernel_channel *channel;
659
660 DBG("Updating kernel streams for channel fd %d", fd);
661
662 session_lock_list();
663 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
664 session_lock(session);
665 if (session->kernel_session == NULL) {
666 session_unlock(session);
667 continue;
668 }
669 ksess = session->kernel_session;
670
671 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
672 if (channel->fd == fd) {
673 DBG("Channel found, updating kernel streams");
674 ret = kernel_open_channel_stream(channel);
675 if (ret < 0) {
676 goto error;
677 }
678 /* Update the stream global counter */
679 ksess->stream_count_global += ret;
680
681 /*
682 * Have we already sent fds to the consumer? If yes, it means
683 * that tracing is started so it is safe to send our updated
684 * stream fds.
685 */
686 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
687 struct lttng_ht_iter iter;
688 struct consumer_socket *socket;
689
690 rcu_read_lock();
691 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
692 &iter.iter, socket, node.node) {
693 /* Code flow error */
694 assert(socket->fd);
695
696 pthread_mutex_lock(socket->lock);
697 ret = kernel_consumer_send_channel_stream(socket,
698 channel, ksess,
699 session->output_traces ? 1 : 0);
700 pthread_mutex_unlock(socket->lock);
701 if (ret < 0) {
702 rcu_read_unlock();
703 goto error;
704 }
705 }
706 rcu_read_unlock();
707 }
708 goto error;
709 }
710 }
711 session_unlock(session);
712 }
713 session_unlock_list();
714 return ret;
715
716 error:
717 session_unlock(session);
718 session_unlock_list();
719 return ret;
720 }
721
722 /*
723 * For each tracing session, update newly registered apps. The session list
724 * lock MUST be acquired before calling this.
725 */
726 static void update_ust_app(int app_sock)
727 {
728 struct ltt_session *sess, *stmp;
729
730 /* For all tracing session(s) */
731 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
732 session_lock(sess);
733 if (sess->ust_session) {
734 ust_app_global_update(sess->ust_session, app_sock);
735 }
736 session_unlock(sess);
737 }
738 }
739
740 /*
741 * This thread manage event coming from the kernel.
742 *
743 * Features supported in this thread:
744 * -) CPU Hotplug
745 */
746 static void *thread_manage_kernel(void *data)
747 {
748 int ret, i, pollfd, update_poll_flag = 1, err = -1;
749 uint32_t revents, nb_fd;
750 char tmp;
751 struct lttng_poll_event events;
752
753 DBG("[thread] Thread manage kernel started");
754
755 health_register(HEALTH_TYPE_KERNEL);
756
757 /*
758 * This first step of the while is to clean this structure which could free
759 * non NULL pointers so initialize it before the loop.
760 */
761 lttng_poll_init(&events);
762
763 if (testpoint(thread_manage_kernel)) {
764 goto error_testpoint;
765 }
766
767 health_code_update();
768
769 if (testpoint(thread_manage_kernel_before_loop)) {
770 goto error_testpoint;
771 }
772
773 while (1) {
774 health_code_update();
775
776 if (update_poll_flag == 1) {
777 /* Clean events object. We are about to populate it again. */
778 lttng_poll_clean(&events);
779
780 ret = sessiond_set_thread_pollset(&events, 2);
781 if (ret < 0) {
782 goto error_poll_create;
783 }
784
785 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
786 if (ret < 0) {
787 goto error;
788 }
789
790 /* This will add the available kernel channel if any. */
791 ret = update_kernel_poll(&events);
792 if (ret < 0) {
793 goto error;
794 }
795 update_poll_flag = 0;
796 }
797
798 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
799
800 /* Poll infinite value of time */
801 restart:
802 health_poll_entry();
803 ret = lttng_poll_wait(&events, -1);
804 health_poll_exit();
805 if (ret < 0) {
806 /*
807 * Restart interrupted system call.
808 */
809 if (errno == EINTR) {
810 goto restart;
811 }
812 goto error;
813 } else if (ret == 0) {
814 /* Should not happen since timeout is infinite */
815 ERR("Return value of poll is 0 with an infinite timeout.\n"
816 "This should not have happened! Continuing...");
817 continue;
818 }
819
820 nb_fd = ret;
821
822 for (i = 0; i < nb_fd; i++) {
823 /* Fetch once the poll data */
824 revents = LTTNG_POLL_GETEV(&events, i);
825 pollfd = LTTNG_POLL_GETFD(&events, i);
826
827 health_code_update();
828
829 /* Thread quit pipe has been closed. Killing thread. */
830 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
831 if (ret) {
832 err = 0;
833 goto exit;
834 }
835
836 /* Check for data on kernel pipe */
837 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
838 do {
839 ret = read(kernel_poll_pipe[0], &tmp, 1);
840 } while (ret < 0 && errno == EINTR);
841 /*
842 * Ret value is useless here, if this pipe gets any actions an
843 * update is required anyway.
844 */
845 update_poll_flag = 1;
846 continue;
847 } else {
848 /*
849 * New CPU detected by the kernel. Adding kernel stream to
850 * kernel session and updating the kernel consumer
851 */
852 if (revents & LPOLLIN) {
853 ret = update_kernel_stream(&kconsumer_data, pollfd);
854 if (ret < 0) {
855 continue;
856 }
857 break;
858 /*
859 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
860 * and unregister kernel stream at this point.
861 */
862 }
863 }
864 }
865 }
866
867 exit:
868 error:
869 lttng_poll_clean(&events);
870 error_poll_create:
871 error_testpoint:
872 utils_close_pipe(kernel_poll_pipe);
873 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
874 if (err) {
875 health_error();
876 ERR("Health error occurred in %s", __func__);
877 WARN("Kernel thread died unexpectedly. "
878 "Kernel tracing can continue but CPU hotplug is disabled.");
879 }
880 health_unregister();
881 DBG("Kernel thread dying");
882 return NULL;
883 }
884
885 /*
886 * Signal pthread condition of the consumer data that the thread.
887 */
888 static void signal_consumer_condition(struct consumer_data *data, int state)
889 {
890 pthread_mutex_lock(&data->cond_mutex);
891
892 /*
893 * The state is set before signaling. It can be any value, it's the waiter
894 * job to correctly interpret this condition variable associated to the
895 * consumer pthread_cond.
896 *
897 * A value of 0 means that the corresponding thread of the consumer data
898 * was not started. 1 indicates that the thread has started and is ready
899 * for action. A negative value means that there was an error during the
900 * thread bootstrap.
901 */
902 data->consumer_thread_is_ready = state;
903 (void) pthread_cond_signal(&data->cond);
904
905 pthread_mutex_unlock(&data->cond_mutex);
906 }
907
908 /*
909 * This thread manage the consumer error sent back to the session daemon.
910 */
911 static void *thread_manage_consumer(void *data)
912 {
913 int sock = -1, i, ret, pollfd, err = -1;
914 uint32_t revents, nb_fd;
915 enum lttcomm_return_code code;
916 struct lttng_poll_event events;
917 struct consumer_data *consumer_data = data;
918
919 DBG("[thread] Manage consumer started");
920
921 health_register(HEALTH_TYPE_CONSUMER);
922
923 health_code_update();
924
925 /*
926 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
927 * metadata_sock. Nothing more will be added to this poll set.
928 */
929 ret = sessiond_set_thread_pollset(&events, 3);
930 if (ret < 0) {
931 goto error_poll;
932 }
933
934 /*
935 * The error socket here is already in a listening state which was done
936 * just before spawning this thread to avoid a race between the consumer
937 * daemon exec trying to connect and the listen() call.
938 */
939 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
940 if (ret < 0) {
941 goto error;
942 }
943
944 health_code_update();
945
946 /* Infinite blocking call, waiting for transmission */
947 restart:
948 health_poll_entry();
949
950 if (testpoint(thread_manage_consumer)) {
951 goto error;
952 }
953
954 ret = lttng_poll_wait(&events, -1);
955 health_poll_exit();
956 if (ret < 0) {
957 /*
958 * Restart interrupted system call.
959 */
960 if (errno == EINTR) {
961 goto restart;
962 }
963 goto error;
964 }
965
966 nb_fd = ret;
967
968 for (i = 0; i < nb_fd; i++) {
969 /* Fetch once the poll data */
970 revents = LTTNG_POLL_GETEV(&events, i);
971 pollfd = LTTNG_POLL_GETFD(&events, i);
972
973 health_code_update();
974
975 /* Thread quit pipe has been closed. Killing thread. */
976 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
977 if (ret) {
978 err = 0;
979 goto exit;
980 }
981
982 /* Event on the registration socket */
983 if (pollfd == consumer_data->err_sock) {
984 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
985 ERR("consumer err socket poll error");
986 goto error;
987 }
988 }
989 }
990
991 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
992 if (sock < 0) {
993 goto error;
994 }
995
996 /*
997 * Set the CLOEXEC flag. Return code is useless because either way, the
998 * show must go on.
999 */
1000 (void) utils_set_fd_cloexec(sock);
1001
1002 health_code_update();
1003
1004 DBG2("Receiving code from consumer err_sock");
1005
1006 /* Getting status code from kconsumerd */
1007 ret = lttcomm_recv_unix_sock(sock, &code,
1008 sizeof(enum lttcomm_return_code));
1009 if (ret <= 0) {
1010 goto error;
1011 }
1012
1013 health_code_update();
1014
1015 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1016 /* Connect both socket, command and metadata. */
1017 consumer_data->cmd_sock =
1018 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1019 consumer_data->metadata_fd =
1020 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1021 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1022 PERROR("consumer connect cmd socket");
1023 /* On error, signal condition and quit. */
1024 signal_consumer_condition(consumer_data, -1);
1025 goto error;
1026 }
1027 consumer_data->metadata_sock.fd = &consumer_data->metadata_fd;
1028 /* Create metadata socket lock. */
1029 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1030 if (consumer_data->metadata_sock.lock == NULL) {
1031 PERROR("zmalloc pthread mutex");
1032 ret = -1;
1033 goto error;
1034 }
1035 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1036
1037 signal_consumer_condition(consumer_data, 1);
1038 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1039 DBG("Consumer metadata socket ready (fd: %d)",
1040 consumer_data->metadata_fd);
1041 } else {
1042 ERR("consumer error when waiting for SOCK_READY : %s",
1043 lttcomm_get_readable_code(-code));
1044 goto error;
1045 }
1046
1047 /* Remove the consumerd error sock since we've established a connexion */
1048 ret = lttng_poll_del(&events, consumer_data->err_sock);
1049 if (ret < 0) {
1050 goto error;
1051 }
1052
1053 /* Add new accepted error socket. */
1054 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1055 if (ret < 0) {
1056 goto error;
1057 }
1058
1059 /* Add metadata socket that is successfully connected. */
1060 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1061 LPOLLIN | LPOLLRDHUP);
1062 if (ret < 0) {
1063 goto error;
1064 }
1065
1066 health_code_update();
1067
1068 /* Infinite blocking call, waiting for transmission */
1069 restart_poll:
1070 while (1) {
1071 health_poll_entry();
1072 ret = lttng_poll_wait(&events, -1);
1073 health_poll_exit();
1074 if (ret < 0) {
1075 /*
1076 * Restart interrupted system call.
1077 */
1078 if (errno == EINTR) {
1079 goto restart_poll;
1080 }
1081 goto error;
1082 }
1083
1084 nb_fd = ret;
1085
1086 for (i = 0; i < nb_fd; i++) {
1087 /* Fetch once the poll data */
1088 revents = LTTNG_POLL_GETEV(&events, i);
1089 pollfd = LTTNG_POLL_GETFD(&events, i);
1090
1091 health_code_update();
1092
1093 /* Thread quit pipe has been closed. Killing thread. */
1094 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1095 if (ret) {
1096 err = 0;
1097 goto exit;
1098 }
1099
1100 if (pollfd == sock) {
1101 /* Event on the consumerd socket */
1102 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1103 ERR("consumer err socket second poll error");
1104 goto error;
1105 }
1106 health_code_update();
1107 /* Wait for any kconsumerd error */
1108 ret = lttcomm_recv_unix_sock(sock, &code,
1109 sizeof(enum lttcomm_return_code));
1110 if (ret <= 0) {
1111 ERR("consumer closed the command socket");
1112 goto error;
1113 }
1114
1115 ERR("consumer return code : %s",
1116 lttcomm_get_readable_code(-code));
1117
1118 goto exit;
1119 } else if (pollfd == consumer_data->metadata_fd) {
1120 /* UST metadata requests */
1121 ret = ust_consumer_metadata_request(
1122 &consumer_data->metadata_sock);
1123 if (ret < 0) {
1124 ERR("Handling metadata request");
1125 goto error;
1126 }
1127 break;
1128 } else {
1129 ERR("Unknown pollfd");
1130 goto error;
1131 }
1132 }
1133 health_code_update();
1134 }
1135
1136 exit:
1137 error:
1138 /* Immediately set the consumerd state to stopped */
1139 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1140 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1141 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1142 consumer_data->type == LTTNG_CONSUMER32_UST) {
1143 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1144 } else {
1145 /* Code flow error... */
1146 assert(0);
1147 }
1148
1149 if (consumer_data->err_sock >= 0) {
1150 ret = close(consumer_data->err_sock);
1151 if (ret) {
1152 PERROR("close");
1153 }
1154 consumer_data->err_sock = -1;
1155 }
1156 if (consumer_data->cmd_sock >= 0) {
1157 ret = close(consumer_data->cmd_sock);
1158 if (ret) {
1159 PERROR("close");
1160 }
1161 consumer_data->cmd_sock = -1;
1162 }
1163 if (*consumer_data->metadata_sock.fd >= 0) {
1164 ret = close(*consumer_data->metadata_sock.fd);
1165 if (ret) {
1166 PERROR("close");
1167 }
1168 }
1169 /* Cleanup metadata socket mutex. */
1170 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1171 free(consumer_data->metadata_sock.lock);
1172
1173 if (sock >= 0) {
1174 ret = close(sock);
1175 if (ret) {
1176 PERROR("close");
1177 }
1178 }
1179
1180 unlink(consumer_data->err_unix_sock_path);
1181 unlink(consumer_data->cmd_unix_sock_path);
1182 consumer_data->pid = 0;
1183
1184 lttng_poll_clean(&events);
1185 error_poll:
1186 if (err) {
1187 health_error();
1188 ERR("Health error occurred in %s", __func__);
1189 }
1190 health_unregister();
1191 DBG("consumer thread cleanup completed");
1192
1193 return NULL;
1194 }
1195
1196 /*
1197 * This thread manage application communication.
1198 */
1199 static void *thread_manage_apps(void *data)
1200 {
1201 int i, ret, pollfd, err = -1;
1202 uint32_t revents, nb_fd;
1203 struct lttng_poll_event events;
1204
1205 DBG("[thread] Manage application started");
1206
1207 rcu_register_thread();
1208 rcu_thread_online();
1209
1210 health_register(HEALTH_TYPE_APP_MANAGE);
1211
1212 if (testpoint(thread_manage_apps)) {
1213 goto error_testpoint;
1214 }
1215
1216 health_code_update();
1217
1218 ret = sessiond_set_thread_pollset(&events, 2);
1219 if (ret < 0) {
1220 goto error_poll_create;
1221 }
1222
1223 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1224 if (ret < 0) {
1225 goto error;
1226 }
1227
1228 if (testpoint(thread_manage_apps_before_loop)) {
1229 goto error;
1230 }
1231
1232 health_code_update();
1233
1234 while (1) {
1235 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1236
1237 /* Inifinite blocking call, waiting for transmission */
1238 restart:
1239 health_poll_entry();
1240 ret = lttng_poll_wait(&events, -1);
1241 health_poll_exit();
1242 if (ret < 0) {
1243 /*
1244 * Restart interrupted system call.
1245 */
1246 if (errno == EINTR) {
1247 goto restart;
1248 }
1249 goto error;
1250 }
1251
1252 nb_fd = ret;
1253
1254 for (i = 0; i < nb_fd; i++) {
1255 /* Fetch once the poll data */
1256 revents = LTTNG_POLL_GETEV(&events, i);
1257 pollfd = LTTNG_POLL_GETFD(&events, i);
1258
1259 health_code_update();
1260
1261 /* Thread quit pipe has been closed. Killing thread. */
1262 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1263 if (ret) {
1264 err = 0;
1265 goto exit;
1266 }
1267
1268 /* Inspect the apps cmd pipe */
1269 if (pollfd == apps_cmd_pipe[0]) {
1270 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1271 ERR("Apps command pipe error");
1272 goto error;
1273 } else if (revents & LPOLLIN) {
1274 int sock;
1275
1276 /* Empty pipe */
1277 do {
1278 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1279 } while (ret < 0 && errno == EINTR);
1280 if (ret < 0 || ret < sizeof(sock)) {
1281 PERROR("read apps cmd pipe");
1282 goto error;
1283 }
1284
1285 health_code_update();
1286
1287 /*
1288 * We only monitor the error events of the socket. This
1289 * thread does not handle any incoming data from UST
1290 * (POLLIN).
1291 */
1292 ret = lttng_poll_add(&events, sock,
1293 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1294 if (ret < 0) {
1295 goto error;
1296 }
1297
1298 /*
1299 * Set socket timeout for both receiving and ending.
1300 * app_socket_timeout is in seconds, whereas
1301 * lttcomm_setsockopt_rcv_timeout and
1302 * lttcomm_setsockopt_snd_timeout expect msec as
1303 * parameter.
1304 */
1305 (void) lttcomm_setsockopt_rcv_timeout(sock,
1306 app_socket_timeout * 1000);
1307 (void) lttcomm_setsockopt_snd_timeout(sock,
1308 app_socket_timeout * 1000);
1309
1310 DBG("Apps with sock %d added to poll set", sock);
1311
1312 health_code_update();
1313
1314 break;
1315 }
1316 } else {
1317 /*
1318 * At this point, we know that a registered application made
1319 * the event at poll_wait.
1320 */
1321 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1322 /* Removing from the poll set */
1323 ret = lttng_poll_del(&events, pollfd);
1324 if (ret < 0) {
1325 goto error;
1326 }
1327
1328 /* Socket closed on remote end. */
1329 ust_app_unregister(pollfd);
1330 break;
1331 }
1332 }
1333
1334 health_code_update();
1335 }
1336 }
1337
1338 exit:
1339 error:
1340 lttng_poll_clean(&events);
1341 error_poll_create:
1342 error_testpoint:
1343 utils_close_pipe(apps_cmd_pipe);
1344 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1345
1346 /*
1347 * We don't clean the UST app hash table here since already registered
1348 * applications can still be controlled so let them be until the session
1349 * daemon dies or the applications stop.
1350 */
1351
1352 if (err) {
1353 health_error();
1354 ERR("Health error occurred in %s", __func__);
1355 }
1356 health_unregister();
1357 DBG("Application communication apps thread cleanup complete");
1358 rcu_thread_offline();
1359 rcu_unregister_thread();
1360 return NULL;
1361 }
1362
1363 /*
1364 * Send a socket to a thread This is called from the dispatch UST registration
1365 * thread once all sockets are set for the application.
1366 *
1367 * On success, return 0 else a negative value being the errno message of the
1368 * write().
1369 */
1370 static int send_socket_to_thread(int fd, int sock)
1371 {
1372 int ret;
1373
1374 /* Sockets MUST be set or else this should not have been called. */
1375 assert(fd >= 0);
1376 assert(sock >= 0);
1377
1378 do {
1379 ret = write(fd, &sock, sizeof(sock));
1380 } while (ret < 0 && errno == EINTR);
1381 if (ret < 0 || ret != sizeof(sock)) {
1382 PERROR("write apps pipe %d", fd);
1383 if (ret < 0) {
1384 ret = -errno;
1385 }
1386 goto error;
1387 }
1388
1389 /* All good. Don't send back the write positive ret value. */
1390 ret = 0;
1391 error:
1392 return ret;
1393 }
1394
1395 /*
1396 * Sanitize the wait queue of the dispatch registration thread meaning removing
1397 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1398 * notify socket is never received.
1399 */
1400 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1401 {
1402 int ret, nb_fd = 0, i;
1403 unsigned int fd_added = 0;
1404 struct lttng_poll_event events;
1405 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1406
1407 assert(wait_queue);
1408
1409 lttng_poll_init(&events);
1410
1411 /* Just skip everything for an empty queue. */
1412 if (!wait_queue->count) {
1413 goto end;
1414 }
1415
1416 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1417 if (ret < 0) {
1418 goto error_create;
1419 }
1420
1421 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1422 &wait_queue->head, head) {
1423 assert(wait_node->app);
1424 ret = lttng_poll_add(&events, wait_node->app->sock,
1425 LPOLLHUP | LPOLLERR);
1426 if (ret < 0) {
1427 goto error;
1428 }
1429
1430 fd_added = 1;
1431 }
1432
1433 if (!fd_added) {
1434 goto end;
1435 }
1436
1437 /*
1438 * Poll but don't block so we can quickly identify the faulty events and
1439 * clean them afterwards from the wait queue.
1440 */
1441 ret = lttng_poll_wait(&events, 0);
1442 if (ret < 0) {
1443 goto error;
1444 }
1445 nb_fd = ret;
1446
1447 for (i = 0; i < nb_fd; i++) {
1448 /* Get faulty FD. */
1449 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1450 int pollfd = LTTNG_POLL_GETFD(&events, i);
1451
1452 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1453 &wait_queue->head, head) {
1454 if (pollfd == wait_node->app->sock &&
1455 (revents & (LPOLLHUP | LPOLLERR))) {
1456 cds_list_del(&wait_node->head);
1457 wait_queue->count--;
1458 ust_app_destroy(wait_node->app);
1459 free(wait_node);
1460 break;
1461 }
1462 }
1463 }
1464
1465 if (nb_fd > 0) {
1466 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1467 }
1468
1469 end:
1470 lttng_poll_clean(&events);
1471 return;
1472
1473 error:
1474 lttng_poll_clean(&events);
1475 error_create:
1476 ERR("Unable to sanitize wait queue");
1477 return;
1478 }
1479
1480 /*
1481 * Dispatch request from the registration threads to the application
1482 * communication thread.
1483 */
1484 static void *thread_dispatch_ust_registration(void *data)
1485 {
1486 int ret, err = -1;
1487 struct cds_wfq_node *node;
1488 struct ust_command *ust_cmd = NULL;
1489 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1490 struct ust_reg_wait_queue wait_queue = {
1491 .count = 0,
1492 };
1493
1494 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1495
1496 health_code_update();
1497
1498 CDS_INIT_LIST_HEAD(&wait_queue.head);
1499
1500 DBG("[thread] Dispatch UST command started");
1501
1502 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1503 health_code_update();
1504
1505 /* Atomically prepare the queue futex */
1506 futex_nto1_prepare(&ust_cmd_queue.futex);
1507
1508 do {
1509 struct ust_app *app = NULL;
1510 ust_cmd = NULL;
1511
1512 /*
1513 * Make sure we don't have node(s) that have hung up before receiving
1514 * the notify socket. This is to clean the list in order to avoid
1515 * memory leaks from notify socket that are never seen.
1516 */
1517 sanitize_wait_queue(&wait_queue);
1518
1519 health_code_update();
1520 /* Dequeue command for registration */
1521 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1522 if (node == NULL) {
1523 DBG("Woken up but nothing in the UST command queue");
1524 /* Continue thread execution */
1525 break;
1526 }
1527
1528 ust_cmd = caa_container_of(node, struct ust_command, node);
1529
1530 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1531 " gid:%d sock:%d name:%s (version %d.%d)",
1532 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1533 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1534 ust_cmd->sock, ust_cmd->reg_msg.name,
1535 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1536
1537 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1538 wait_node = zmalloc(sizeof(*wait_node));
1539 if (!wait_node) {
1540 PERROR("zmalloc wait_node dispatch");
1541 ret = close(ust_cmd->sock);
1542 if (ret < 0) {
1543 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1544 }
1545 lttng_fd_put(1, LTTNG_FD_APPS);
1546 free(ust_cmd);
1547 goto error;
1548 }
1549 CDS_INIT_LIST_HEAD(&wait_node->head);
1550
1551 /* Create application object if socket is CMD. */
1552 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1553 ust_cmd->sock);
1554 if (!wait_node->app) {
1555 ret = close(ust_cmd->sock);
1556 if (ret < 0) {
1557 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1558 }
1559 lttng_fd_put(1, LTTNG_FD_APPS);
1560 free(wait_node);
1561 free(ust_cmd);
1562 continue;
1563 }
1564 /*
1565 * Add application to the wait queue so we can set the notify
1566 * socket before putting this object in the global ht.
1567 */
1568 cds_list_add(&wait_node->head, &wait_queue.head);
1569 wait_queue.count++;
1570
1571 free(ust_cmd);
1572 /*
1573 * We have to continue here since we don't have the notify
1574 * socket and the application MUST be added to the hash table
1575 * only at that moment.
1576 */
1577 continue;
1578 } else {
1579 /*
1580 * Look for the application in the local wait queue and set the
1581 * notify socket if found.
1582 */
1583 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1584 &wait_queue.head, head) {
1585 health_code_update();
1586 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1587 wait_node->app->notify_sock = ust_cmd->sock;
1588 cds_list_del(&wait_node->head);
1589 wait_queue.count--;
1590 app = wait_node->app;
1591 free(wait_node);
1592 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1593 break;
1594 }
1595 }
1596
1597 /*
1598 * With no application at this stage the received socket is
1599 * basically useless so close it before we free the cmd data
1600 * structure for good.
1601 */
1602 if (!app) {
1603 ret = close(ust_cmd->sock);
1604 if (ret < 0) {
1605 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1606 }
1607 lttng_fd_put(1, LTTNG_FD_APPS);
1608 }
1609 free(ust_cmd);
1610 }
1611
1612 if (app) {
1613 /*
1614 * @session_lock_list
1615 *
1616 * Lock the global session list so from the register up to the
1617 * registration done message, no thread can see the application
1618 * and change its state.
1619 */
1620 session_lock_list();
1621 rcu_read_lock();
1622
1623 /*
1624 * Add application to the global hash table. This needs to be
1625 * done before the update to the UST registry can locate the
1626 * application.
1627 */
1628 ust_app_add(app);
1629
1630 /* Set app version. This call will print an error if needed. */
1631 (void) ust_app_version(app);
1632
1633 /* Send notify socket through the notify pipe. */
1634 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1635 app->notify_sock);
1636 if (ret < 0) {
1637 rcu_read_unlock();
1638 session_unlock_list();
1639 /* No notify thread, stop the UST tracing. */
1640 goto error;
1641 }
1642
1643 /*
1644 * Update newly registered application with the tracing
1645 * registry info already enabled information.
1646 */
1647 update_ust_app(app->sock);
1648
1649 /*
1650 * Don't care about return value. Let the manage apps threads
1651 * handle app unregistration upon socket close.
1652 */
1653 (void) ust_app_register_done(app->sock);
1654
1655 /*
1656 * Even if the application socket has been closed, send the app
1657 * to the thread and unregistration will take place at that
1658 * place.
1659 */
1660 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1661 if (ret < 0) {
1662 rcu_read_unlock();
1663 session_unlock_list();
1664 /* No apps. thread, stop the UST tracing. */
1665 goto error;
1666 }
1667
1668 rcu_read_unlock();
1669 session_unlock_list();
1670 }
1671 } while (node != NULL);
1672
1673 health_poll_entry();
1674 /* Futex wait on queue. Blocking call on futex() */
1675 futex_nto1_wait(&ust_cmd_queue.futex);
1676 health_poll_exit();
1677 }
1678 /* Normal exit, no error */
1679 err = 0;
1680
1681 error:
1682 /* Clean up wait queue. */
1683 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1684 &wait_queue.head, head) {
1685 cds_list_del(&wait_node->head);
1686 wait_queue.count--;
1687 free(wait_node);
1688 }
1689
1690 DBG("Dispatch thread dying");
1691 if (err) {
1692 health_error();
1693 ERR("Health error occurred in %s", __func__);
1694 }
1695 health_unregister();
1696 return NULL;
1697 }
1698
1699 /*
1700 * This thread manage application registration.
1701 */
1702 static void *thread_registration_apps(void *data)
1703 {
1704 int sock = -1, i, ret, pollfd, err = -1;
1705 uint32_t revents, nb_fd;
1706 struct lttng_poll_event events;
1707 /*
1708 * Get allocated in this thread, enqueued to a global queue, dequeued and
1709 * freed in the manage apps thread.
1710 */
1711 struct ust_command *ust_cmd = NULL;
1712
1713 DBG("[thread] Manage application registration started");
1714
1715 health_register(HEALTH_TYPE_APP_REG);
1716
1717 if (testpoint(thread_registration_apps)) {
1718 goto error_testpoint;
1719 }
1720
1721 ret = lttcomm_listen_unix_sock(apps_sock);
1722 if (ret < 0) {
1723 goto error_listen;
1724 }
1725
1726 /*
1727 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1728 * more will be added to this poll set.
1729 */
1730 ret = sessiond_set_thread_pollset(&events, 2);
1731 if (ret < 0) {
1732 goto error_create_poll;
1733 }
1734
1735 /* Add the application registration socket */
1736 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1737 if (ret < 0) {
1738 goto error_poll_add;
1739 }
1740
1741 /* Notify all applications to register */
1742 ret = notify_ust_apps(1);
1743 if (ret < 0) {
1744 ERR("Failed to notify applications or create the wait shared memory.\n"
1745 "Execution continues but there might be problem for already\n"
1746 "running applications that wishes to register.");
1747 }
1748
1749 while (1) {
1750 DBG("Accepting application registration");
1751
1752 /* Inifinite blocking call, waiting for transmission */
1753 restart:
1754 health_poll_entry();
1755 ret = lttng_poll_wait(&events, -1);
1756 health_poll_exit();
1757 if (ret < 0) {
1758 /*
1759 * Restart interrupted system call.
1760 */
1761 if (errno == EINTR) {
1762 goto restart;
1763 }
1764 goto error;
1765 }
1766
1767 nb_fd = ret;
1768
1769 for (i = 0; i < nb_fd; i++) {
1770 health_code_update();
1771
1772 /* Fetch once the poll data */
1773 revents = LTTNG_POLL_GETEV(&events, i);
1774 pollfd = LTTNG_POLL_GETFD(&events, i);
1775
1776 /* Thread quit pipe has been closed. Killing thread. */
1777 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1778 if (ret) {
1779 err = 0;
1780 goto exit;
1781 }
1782
1783 /* Event on the registration socket */
1784 if (pollfd == apps_sock) {
1785 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1786 ERR("Register apps socket poll error");
1787 goto error;
1788 } else if (revents & LPOLLIN) {
1789 sock = lttcomm_accept_unix_sock(apps_sock);
1790 if (sock < 0) {
1791 goto error;
1792 }
1793
1794 /*
1795 * Set the CLOEXEC flag. Return code is useless because
1796 * either way, the show must go on.
1797 */
1798 (void) utils_set_fd_cloexec(sock);
1799
1800 /* Create UST registration command for enqueuing */
1801 ust_cmd = zmalloc(sizeof(struct ust_command));
1802 if (ust_cmd == NULL) {
1803 PERROR("ust command zmalloc");
1804 goto error;
1805 }
1806
1807 /*
1808 * Using message-based transmissions to ensure we don't
1809 * have to deal with partially received messages.
1810 */
1811 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1812 if (ret < 0) {
1813 ERR("Exhausted file descriptors allowed for applications.");
1814 free(ust_cmd);
1815 ret = close(sock);
1816 if (ret) {
1817 PERROR("close");
1818 }
1819 sock = -1;
1820 continue;
1821 }
1822
1823 health_code_update();
1824 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1825 if (ret < 0) {
1826 free(ust_cmd);
1827 /* Close socket of the application. */
1828 ret = close(sock);
1829 if (ret) {
1830 PERROR("close");
1831 }
1832 lttng_fd_put(LTTNG_FD_APPS, 1);
1833 sock = -1;
1834 continue;
1835 }
1836 health_code_update();
1837
1838 ust_cmd->sock = sock;
1839 sock = -1;
1840
1841 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1842 " gid:%d sock:%d name:%s (version %d.%d)",
1843 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1844 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1845 ust_cmd->sock, ust_cmd->reg_msg.name,
1846 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1847
1848 /*
1849 * Lock free enqueue the registration request. The red pill
1850 * has been taken! This apps will be part of the *system*.
1851 */
1852 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1853
1854 /*
1855 * Wake the registration queue futex. Implicit memory
1856 * barrier with the exchange in cds_wfq_enqueue.
1857 */
1858 futex_nto1_wake(&ust_cmd_queue.futex);
1859 }
1860 }
1861 }
1862 }
1863
1864 exit:
1865 error:
1866 if (err) {
1867 health_error();
1868 ERR("Health error occurred in %s", __func__);
1869 }
1870
1871 /* Notify that the registration thread is gone */
1872 notify_ust_apps(0);
1873
1874 if (apps_sock >= 0) {
1875 ret = close(apps_sock);
1876 if (ret) {
1877 PERROR("close");
1878 }
1879 }
1880 if (sock >= 0) {
1881 ret = close(sock);
1882 if (ret) {
1883 PERROR("close");
1884 }
1885 lttng_fd_put(LTTNG_FD_APPS, 1);
1886 }
1887 unlink(apps_unix_sock_path);
1888
1889 error_poll_add:
1890 lttng_poll_clean(&events);
1891 error_listen:
1892 error_create_poll:
1893 error_testpoint:
1894 DBG("UST Registration thread cleanup complete");
1895 health_unregister();
1896
1897 return NULL;
1898 }
1899
1900 /*
1901 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1902 * exec or it will fails.
1903 */
1904 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1905 {
1906 int ret, clock_ret;
1907 struct timespec timeout;
1908
1909 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1910 consumer_data->consumer_thread_is_ready = 0;
1911
1912 /* Setup pthread condition */
1913 ret = pthread_condattr_init(&consumer_data->condattr);
1914 if (ret != 0) {
1915 errno = ret;
1916 PERROR("pthread_condattr_init consumer data");
1917 goto error;
1918 }
1919
1920 /*
1921 * Set the monotonic clock in order to make sure we DO NOT jump in time
1922 * between the clock_gettime() call and the timedwait call. See bug #324
1923 * for a more details and how we noticed it.
1924 */
1925 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1926 if (ret != 0) {
1927 errno = ret;
1928 PERROR("pthread_condattr_setclock consumer data");
1929 goto error;
1930 }
1931
1932 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1933 if (ret != 0) {
1934 errno = ret;
1935 PERROR("pthread_cond_init consumer data");
1936 goto error;
1937 }
1938
1939 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1940 consumer_data);
1941 if (ret != 0) {
1942 PERROR("pthread_create consumer");
1943 ret = -1;
1944 goto error;
1945 }
1946
1947 /* We are about to wait on a pthread condition */
1948 pthread_mutex_lock(&consumer_data->cond_mutex);
1949
1950 /* Get time for sem_timedwait absolute timeout */
1951 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1952 /*
1953 * Set the timeout for the condition timed wait even if the clock gettime
1954 * call fails since we might loop on that call and we want to avoid to
1955 * increment the timeout too many times.
1956 */
1957 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1958
1959 /*
1960 * The following loop COULD be skipped in some conditions so this is why we
1961 * set ret to 0 in order to make sure at least one round of the loop is
1962 * done.
1963 */
1964 ret = 0;
1965
1966 /*
1967 * Loop until the condition is reached or when a timeout is reached. Note
1968 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1969 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1970 * possible. This loop does not take any chances and works with both of
1971 * them.
1972 */
1973 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1974 if (clock_ret < 0) {
1975 PERROR("clock_gettime spawn consumer");
1976 /* Infinite wait for the consumerd thread to be ready */
1977 ret = pthread_cond_wait(&consumer_data->cond,
1978 &consumer_data->cond_mutex);
1979 } else {
1980 ret = pthread_cond_timedwait(&consumer_data->cond,
1981 &consumer_data->cond_mutex, &timeout);
1982 }
1983 }
1984
1985 /* Release the pthread condition */
1986 pthread_mutex_unlock(&consumer_data->cond_mutex);
1987
1988 if (ret != 0) {
1989 errno = ret;
1990 if (ret == ETIMEDOUT) {
1991 /*
1992 * Call has timed out so we kill the kconsumerd_thread and return
1993 * an error.
1994 */
1995 ERR("Condition timed out. The consumer thread was never ready."
1996 " Killing it");
1997 ret = pthread_cancel(consumer_data->thread);
1998 if (ret < 0) {
1999 PERROR("pthread_cancel consumer thread");
2000 }
2001 } else {
2002 PERROR("pthread_cond_wait failed consumer thread");
2003 }
2004 goto error;
2005 }
2006
2007 pthread_mutex_lock(&consumer_data->pid_mutex);
2008 if (consumer_data->pid == 0) {
2009 ERR("Consumerd did not start");
2010 pthread_mutex_unlock(&consumer_data->pid_mutex);
2011 goto error;
2012 }
2013 pthread_mutex_unlock(&consumer_data->pid_mutex);
2014
2015 return 0;
2016
2017 error:
2018 return ret;
2019 }
2020
2021 /*
2022 * Join consumer thread
2023 */
2024 static int join_consumer_thread(struct consumer_data *consumer_data)
2025 {
2026 void *status;
2027
2028 /* Consumer pid must be a real one. */
2029 if (consumer_data->pid > 0) {
2030 int ret;
2031 ret = kill(consumer_data->pid, SIGTERM);
2032 if (ret) {
2033 ERR("Error killing consumer daemon");
2034 return ret;
2035 }
2036 return pthread_join(consumer_data->thread, &status);
2037 } else {
2038 return 0;
2039 }
2040 }
2041
2042 /*
2043 * Fork and exec a consumer daemon (consumerd).
2044 *
2045 * Return pid if successful else -1.
2046 */
2047 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2048 {
2049 int ret;
2050 pid_t pid;
2051 const char *consumer_to_use;
2052 const char *verbosity;
2053 struct stat st;
2054
2055 DBG("Spawning consumerd");
2056
2057 pid = fork();
2058 if (pid == 0) {
2059 /*
2060 * Exec consumerd.
2061 */
2062 if (opt_verbose_consumer) {
2063 verbosity = "--verbose";
2064 } else {
2065 verbosity = "--quiet";
2066 }
2067 switch (consumer_data->type) {
2068 case LTTNG_CONSUMER_KERNEL:
2069 /*
2070 * Find out which consumerd to execute. We will first try the
2071 * 64-bit path, then the sessiond's installation directory, and
2072 * fallback on the 32-bit one,
2073 */
2074 DBG3("Looking for a kernel consumer at these locations:");
2075 DBG3(" 1) %s", consumerd64_bin);
2076 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2077 DBG3(" 3) %s", consumerd32_bin);
2078 if (stat(consumerd64_bin, &st) == 0) {
2079 DBG3("Found location #1");
2080 consumer_to_use = consumerd64_bin;
2081 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2082 DBG3("Found location #2");
2083 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2084 } else if (stat(consumerd32_bin, &st) == 0) {
2085 DBG3("Found location #3");
2086 consumer_to_use = consumerd32_bin;
2087 } else {
2088 DBG("Could not find any valid consumerd executable");
2089 break;
2090 }
2091 DBG("Using kernel consumer at: %s", consumer_to_use);
2092 execl(consumer_to_use,
2093 "lttng-consumerd", verbosity, "-k",
2094 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2095 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2096 NULL);
2097 break;
2098 case LTTNG_CONSUMER64_UST:
2099 {
2100 char *tmpnew = NULL;
2101
2102 if (consumerd64_libdir[0] != '\0') {
2103 char *tmp;
2104 size_t tmplen;
2105
2106 tmp = getenv("LD_LIBRARY_PATH");
2107 if (!tmp) {
2108 tmp = "";
2109 }
2110 tmplen = strlen("LD_LIBRARY_PATH=")
2111 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2112 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2113 if (!tmpnew) {
2114 ret = -ENOMEM;
2115 goto error;
2116 }
2117 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2118 strcat(tmpnew, consumerd64_libdir);
2119 if (tmp[0] != '\0') {
2120 strcat(tmpnew, ":");
2121 strcat(tmpnew, tmp);
2122 }
2123 ret = putenv(tmpnew);
2124 if (ret) {
2125 ret = -errno;
2126 free(tmpnew);
2127 goto error;
2128 }
2129 }
2130 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2131 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2132 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2133 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2134 NULL);
2135 if (consumerd64_libdir[0] != '\0') {
2136 free(tmpnew);
2137 }
2138 if (ret) {
2139 goto error;
2140 }
2141 break;
2142 }
2143 case LTTNG_CONSUMER32_UST:
2144 {
2145 char *tmpnew = NULL;
2146
2147 if (consumerd32_libdir[0] != '\0') {
2148 char *tmp;
2149 size_t tmplen;
2150
2151 tmp = getenv("LD_LIBRARY_PATH");
2152 if (!tmp) {
2153 tmp = "";
2154 }
2155 tmplen = strlen("LD_LIBRARY_PATH=")
2156 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2157 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2158 if (!tmpnew) {
2159 ret = -ENOMEM;
2160 goto error;
2161 }
2162 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2163 strcat(tmpnew, consumerd32_libdir);
2164 if (tmp[0] != '\0') {
2165 strcat(tmpnew, ":");
2166 strcat(tmpnew, tmp);
2167 }
2168 ret = putenv(tmpnew);
2169 if (ret) {
2170 ret = -errno;
2171 free(tmpnew);
2172 goto error;
2173 }
2174 }
2175 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2176 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2177 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2178 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2179 NULL);
2180 if (consumerd32_libdir[0] != '\0') {
2181 free(tmpnew);
2182 }
2183 if (ret) {
2184 goto error;
2185 }
2186 break;
2187 }
2188 default:
2189 PERROR("unknown consumer type");
2190 exit(EXIT_FAILURE);
2191 }
2192 if (errno != 0) {
2193 PERROR("kernel start consumer exec");
2194 }
2195 exit(EXIT_FAILURE);
2196 } else if (pid > 0) {
2197 ret = pid;
2198 } else {
2199 PERROR("start consumer fork");
2200 ret = -errno;
2201 }
2202 error:
2203 return ret;
2204 }
2205
2206 /*
2207 * Spawn the consumerd daemon and session daemon thread.
2208 */
2209 static int start_consumerd(struct consumer_data *consumer_data)
2210 {
2211 int ret;
2212
2213 /*
2214 * Set the listen() state on the socket since there is a possible race
2215 * between the exec() of the consumer daemon and this call if place in the
2216 * consumer thread. See bug #366 for more details.
2217 */
2218 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2219 if (ret < 0) {
2220 goto error;
2221 }
2222
2223 pthread_mutex_lock(&consumer_data->pid_mutex);
2224 if (consumer_data->pid != 0) {
2225 pthread_mutex_unlock(&consumer_data->pid_mutex);
2226 goto end;
2227 }
2228
2229 ret = spawn_consumerd(consumer_data);
2230 if (ret < 0) {
2231 ERR("Spawning consumerd failed");
2232 pthread_mutex_unlock(&consumer_data->pid_mutex);
2233 goto error;
2234 }
2235
2236 /* Setting up the consumer_data pid */
2237 consumer_data->pid = ret;
2238 DBG2("Consumer pid %d", consumer_data->pid);
2239 pthread_mutex_unlock(&consumer_data->pid_mutex);
2240
2241 DBG2("Spawning consumer control thread");
2242 ret = spawn_consumer_thread(consumer_data);
2243 if (ret < 0) {
2244 ERR("Fatal error spawning consumer control thread");
2245 goto error;
2246 }
2247
2248 end:
2249 return 0;
2250
2251 error:
2252 /* Cleanup already created sockets on error. */
2253 if (consumer_data->err_sock >= 0) {
2254 int err;
2255
2256 err = close(consumer_data->err_sock);
2257 if (err < 0) {
2258 PERROR("close consumer data error socket");
2259 }
2260 }
2261 return ret;
2262 }
2263
2264 /*
2265 * Compute health status of each consumer. If one of them is zero (bad
2266 * state), we return 0.
2267 */
2268 static int check_consumer_health(void)
2269 {
2270 int ret;
2271
2272 ret = health_check_state(HEALTH_TYPE_CONSUMER);
2273
2274 DBG3("Health consumer check %d", ret);
2275
2276 return ret;
2277 }
2278
2279 /*
2280 * Setup necessary data for kernel tracer action.
2281 */
2282 static int init_kernel_tracer(void)
2283 {
2284 int ret;
2285
2286 /* Modprobe lttng kernel modules */
2287 ret = modprobe_lttng_control();
2288 if (ret < 0) {
2289 goto error;
2290 }
2291
2292 /* Open debugfs lttng */
2293 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2294 if (kernel_tracer_fd < 0) {
2295 DBG("Failed to open %s", module_proc_lttng);
2296 ret = -1;
2297 goto error_open;
2298 }
2299
2300 /* Validate kernel version */
2301 ret = kernel_validate_version(kernel_tracer_fd);
2302 if (ret < 0) {
2303 goto error_version;
2304 }
2305
2306 ret = modprobe_lttng_data();
2307 if (ret < 0) {
2308 goto error_modules;
2309 }
2310
2311 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2312 return 0;
2313
2314 error_version:
2315 modprobe_remove_lttng_control();
2316 ret = close(kernel_tracer_fd);
2317 if (ret) {
2318 PERROR("close");
2319 }
2320 kernel_tracer_fd = -1;
2321 return LTTNG_ERR_KERN_VERSION;
2322
2323 error_modules:
2324 ret = close(kernel_tracer_fd);
2325 if (ret) {
2326 PERROR("close");
2327 }
2328
2329 error_open:
2330 modprobe_remove_lttng_control();
2331
2332 error:
2333 WARN("No kernel tracer available");
2334 kernel_tracer_fd = -1;
2335 if (!is_root) {
2336 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2337 } else {
2338 return LTTNG_ERR_KERN_NA;
2339 }
2340 }
2341
2342
2343 /*
2344 * Copy consumer output from the tracing session to the domain session. The
2345 * function also applies the right modification on a per domain basis for the
2346 * trace files destination directory.
2347 *
2348 * Should *NOT* be called with RCU read-side lock held.
2349 */
2350 static int copy_session_consumer(int domain, struct ltt_session *session)
2351 {
2352 int ret;
2353 const char *dir_name;
2354 struct consumer_output *consumer;
2355
2356 assert(session);
2357 assert(session->consumer);
2358
2359 switch (domain) {
2360 case LTTNG_DOMAIN_KERNEL:
2361 DBG3("Copying tracing session consumer output in kernel session");
2362 /*
2363 * XXX: We should audit the session creation and what this function
2364 * does "extra" in order to avoid a destroy since this function is used
2365 * in the domain session creation (kernel and ust) only. Same for UST
2366 * domain.
2367 */
2368 if (session->kernel_session->consumer) {
2369 consumer_destroy_output(session->kernel_session->consumer);
2370 }
2371 session->kernel_session->consumer =
2372 consumer_copy_output(session->consumer);
2373 /* Ease our life a bit for the next part */
2374 consumer = session->kernel_session->consumer;
2375 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2376 break;
2377 case LTTNG_DOMAIN_UST:
2378 DBG3("Copying tracing session consumer output in UST session");
2379 if (session->ust_session->consumer) {
2380 consumer_destroy_output(session->ust_session->consumer);
2381 }
2382 session->ust_session->consumer =
2383 consumer_copy_output(session->consumer);
2384 /* Ease our life a bit for the next part */
2385 consumer = session->ust_session->consumer;
2386 dir_name = DEFAULT_UST_TRACE_DIR;
2387 break;
2388 default:
2389 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2390 goto error;
2391 }
2392
2393 /* Append correct directory to subdir */
2394 strncat(consumer->subdir, dir_name,
2395 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2396 DBG3("Copy session consumer subdir %s", consumer->subdir);
2397
2398 ret = LTTNG_OK;
2399
2400 error:
2401 return ret;
2402 }
2403
2404 /*
2405 * Create an UST session and add it to the session ust list.
2406 *
2407 * Should *NOT* be called with RCU read-side lock held.
2408 */
2409 static int create_ust_session(struct ltt_session *session,
2410 struct lttng_domain *domain)
2411 {
2412 int ret;
2413 struct ltt_ust_session *lus = NULL;
2414
2415 assert(session);
2416 assert(domain);
2417 assert(session->consumer);
2418
2419 switch (domain->type) {
2420 case LTTNG_DOMAIN_UST:
2421 break;
2422 default:
2423 ERR("Unknown UST domain on create session %d", domain->type);
2424 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2425 goto error;
2426 }
2427
2428 DBG("Creating UST session");
2429
2430 lus = trace_ust_create_session(session->id);
2431 if (lus == NULL) {
2432 ret = LTTNG_ERR_UST_SESS_FAIL;
2433 goto error;
2434 }
2435
2436 lus->uid = session->uid;
2437 lus->gid = session->gid;
2438 lus->output_traces = session->output_traces;
2439 lus->snapshot_mode = session->snapshot_mode;
2440 session->ust_session = lus;
2441
2442 /* Copy session output to the newly created UST session */
2443 ret = copy_session_consumer(domain->type, session);
2444 if (ret != LTTNG_OK) {
2445 goto error;
2446 }
2447
2448 return LTTNG_OK;
2449
2450 error:
2451 free(lus);
2452 session->ust_session = NULL;
2453 return ret;
2454 }
2455
2456 /*
2457 * Create a kernel tracer session then create the default channel.
2458 */
2459 static int create_kernel_session(struct ltt_session *session)
2460 {
2461 int ret;
2462
2463 DBG("Creating kernel session");
2464
2465 ret = kernel_create_session(session, kernel_tracer_fd);
2466 if (ret < 0) {
2467 ret = LTTNG_ERR_KERN_SESS_FAIL;
2468 goto error;
2469 }
2470
2471 /* Code flow safety */
2472 assert(session->kernel_session);
2473
2474 /* Copy session output to the newly created Kernel session */
2475 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2476 if (ret != LTTNG_OK) {
2477 goto error;
2478 }
2479
2480 /* Create directory(ies) on local filesystem. */
2481 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2482 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2483 ret = run_as_mkdir_recursive(
2484 session->kernel_session->consumer->dst.trace_path,
2485 S_IRWXU | S_IRWXG, session->uid, session->gid);
2486 if (ret < 0) {
2487 if (ret != -EEXIST) {
2488 ERR("Trace directory creation error");
2489 goto error;
2490 }
2491 }
2492 }
2493
2494 session->kernel_session->uid = session->uid;
2495 session->kernel_session->gid = session->gid;
2496 session->kernel_session->output_traces = session->output_traces;
2497 session->kernel_session->snapshot_mode = session->snapshot_mode;
2498
2499 return LTTNG_OK;
2500
2501 error:
2502 trace_kernel_destroy_session(session->kernel_session);
2503 session->kernel_session = NULL;
2504 return ret;
2505 }
2506
2507 /*
2508 * Count number of session permitted by uid/gid.
2509 */
2510 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2511 {
2512 unsigned int i = 0;
2513 struct ltt_session *session;
2514
2515 DBG("Counting number of available session for UID %d GID %d",
2516 uid, gid);
2517 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2518 /*
2519 * Only list the sessions the user can control.
2520 */
2521 if (!session_access_ok(session, uid, gid)) {
2522 continue;
2523 }
2524 i++;
2525 }
2526 return i;
2527 }
2528
2529 /*
2530 * Process the command requested by the lttng client within the command
2531 * context structure. This function make sure that the return structure (llm)
2532 * is set and ready for transmission before returning.
2533 *
2534 * Return any error encountered or 0 for success.
2535 *
2536 * "sock" is only used for special-case var. len data.
2537 *
2538 * Should *NOT* be called with RCU read-side lock held.
2539 */
2540 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2541 int *sock_error)
2542 {
2543 int ret = LTTNG_OK;
2544 int need_tracing_session = 1;
2545 int need_domain;
2546
2547 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2548
2549 *sock_error = 0;
2550
2551 switch (cmd_ctx->lsm->cmd_type) {
2552 case LTTNG_CREATE_SESSION:
2553 case LTTNG_CREATE_SESSION_SNAPSHOT:
2554 case LTTNG_DESTROY_SESSION:
2555 case LTTNG_LIST_SESSIONS:
2556 case LTTNG_LIST_DOMAINS:
2557 case LTTNG_START_TRACE:
2558 case LTTNG_STOP_TRACE:
2559 case LTTNG_DATA_PENDING:
2560 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2561 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2562 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2563 case LTTNG_SNAPSHOT_RECORD:
2564 need_domain = 0;
2565 break;
2566 default:
2567 need_domain = 1;
2568 }
2569
2570 if (opt_no_kernel && need_domain
2571 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2572 if (!is_root) {
2573 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2574 } else {
2575 ret = LTTNG_ERR_KERN_NA;
2576 }
2577 goto error;
2578 }
2579
2580 /* Deny register consumer if we already have a spawned consumer. */
2581 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2582 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2583 if (kconsumer_data.pid > 0) {
2584 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2585 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2586 goto error;
2587 }
2588 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2589 }
2590
2591 /*
2592 * Check for command that don't needs to allocate a returned payload. We do
2593 * this here so we don't have to make the call for no payload at each
2594 * command.
2595 */
2596 switch(cmd_ctx->lsm->cmd_type) {
2597 case LTTNG_LIST_SESSIONS:
2598 case LTTNG_LIST_TRACEPOINTS:
2599 case LTTNG_LIST_TRACEPOINT_FIELDS:
2600 case LTTNG_LIST_DOMAINS:
2601 case LTTNG_LIST_CHANNELS:
2602 case LTTNG_LIST_EVENTS:
2603 break;
2604 default:
2605 /* Setup lttng message with no payload */
2606 ret = setup_lttng_msg(cmd_ctx, 0);
2607 if (ret < 0) {
2608 /* This label does not try to unlock the session */
2609 goto init_setup_error;
2610 }
2611 }
2612
2613 /* Commands that DO NOT need a session. */
2614 switch (cmd_ctx->lsm->cmd_type) {
2615 case LTTNG_CREATE_SESSION:
2616 case LTTNG_CREATE_SESSION_SNAPSHOT:
2617 case LTTNG_CALIBRATE:
2618 case LTTNG_LIST_SESSIONS:
2619 case LTTNG_LIST_TRACEPOINTS:
2620 case LTTNG_LIST_TRACEPOINT_FIELDS:
2621 need_tracing_session = 0;
2622 break;
2623 default:
2624 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2625 /*
2626 * We keep the session list lock across _all_ commands
2627 * for now, because the per-session lock does not
2628 * handle teardown properly.
2629 */
2630 session_lock_list();
2631 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2632 if (cmd_ctx->session == NULL) {
2633 ret = LTTNG_ERR_SESS_NOT_FOUND;
2634 goto error;
2635 } else {
2636 /* Acquire lock for the session */
2637 session_lock(cmd_ctx->session);
2638 }
2639 break;
2640 }
2641
2642 if (!need_domain) {
2643 goto skip_domain;
2644 }
2645
2646 /*
2647 * Check domain type for specific "pre-action".
2648 */
2649 switch (cmd_ctx->lsm->domain.type) {
2650 case LTTNG_DOMAIN_KERNEL:
2651 if (!is_root) {
2652 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2653 goto error;
2654 }
2655
2656 /* Kernel tracer check */
2657 if (kernel_tracer_fd == -1) {
2658 /* Basically, load kernel tracer modules */
2659 ret = init_kernel_tracer();
2660 if (ret != 0) {
2661 goto error;
2662 }
2663 }
2664
2665 /* Consumer is in an ERROR state. Report back to client */
2666 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2667 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2668 goto error;
2669 }
2670
2671 /* Need a session for kernel command */
2672 if (need_tracing_session) {
2673 if (cmd_ctx->session->kernel_session == NULL) {
2674 ret = create_kernel_session(cmd_ctx->session);
2675 if (ret < 0) {
2676 ret = LTTNG_ERR_KERN_SESS_FAIL;
2677 goto error;
2678 }
2679 }
2680
2681 /* Start the kernel consumer daemon */
2682 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2683 if (kconsumer_data.pid == 0 &&
2684 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2685 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2686 ret = start_consumerd(&kconsumer_data);
2687 if (ret < 0) {
2688 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2689 goto error;
2690 }
2691 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2692 } else {
2693 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2694 }
2695
2696 /*
2697 * The consumer was just spawned so we need to add the socket to
2698 * the consumer output of the session if exist.
2699 */
2700 ret = consumer_create_socket(&kconsumer_data,
2701 cmd_ctx->session->kernel_session->consumer);
2702 if (ret < 0) {
2703 goto error;
2704 }
2705 }
2706
2707 break;
2708 case LTTNG_DOMAIN_UST:
2709 {
2710 if (!ust_app_supported()) {
2711 ret = LTTNG_ERR_NO_UST;
2712 goto error;
2713 }
2714 /* Consumer is in an ERROR state. Report back to client */
2715 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2716 ret = LTTNG_ERR_NO_USTCONSUMERD;
2717 goto error;
2718 }
2719
2720 if (need_tracing_session) {
2721 /* Create UST session if none exist. */
2722 if (cmd_ctx->session->ust_session == NULL) {
2723 ret = create_ust_session(cmd_ctx->session,
2724 &cmd_ctx->lsm->domain);
2725 if (ret != LTTNG_OK) {
2726 goto error;
2727 }
2728 }
2729
2730 /* Start the UST consumer daemons */
2731 /* 64-bit */
2732 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2733 if (consumerd64_bin[0] != '\0' &&
2734 ustconsumer64_data.pid == 0 &&
2735 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2736 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2737 ret = start_consumerd(&ustconsumer64_data);
2738 if (ret < 0) {
2739 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2740 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2741 goto error;
2742 }
2743
2744 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2745 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2746 } else {
2747 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2748 }
2749
2750 /*
2751 * Setup socket for consumer 64 bit. No need for atomic access
2752 * since it was set above and can ONLY be set in this thread.
2753 */
2754 ret = consumer_create_socket(&ustconsumer64_data,
2755 cmd_ctx->session->ust_session->consumer);
2756 if (ret < 0) {
2757 goto error;
2758 }
2759
2760 /* 32-bit */
2761 if (consumerd32_bin[0] != '\0' &&
2762 ustconsumer32_data.pid == 0 &&
2763 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2764 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2765 ret = start_consumerd(&ustconsumer32_data);
2766 if (ret < 0) {
2767 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2768 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2769 goto error;
2770 }
2771
2772 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2773 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2774 } else {
2775 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2776 }
2777
2778 /*
2779 * Setup socket for consumer 64 bit. No need for atomic access
2780 * since it was set above and can ONLY be set in this thread.
2781 */
2782 ret = consumer_create_socket(&ustconsumer32_data,
2783 cmd_ctx->session->ust_session->consumer);
2784 if (ret < 0) {
2785 goto error;
2786 }
2787 }
2788 break;
2789 }
2790 default:
2791 break;
2792 }
2793 skip_domain:
2794
2795 /* Validate consumer daemon state when start/stop trace command */
2796 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2797 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2798 switch (cmd_ctx->lsm->domain.type) {
2799 case LTTNG_DOMAIN_UST:
2800 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2801 ret = LTTNG_ERR_NO_USTCONSUMERD;
2802 goto error;
2803 }
2804 break;
2805 case LTTNG_DOMAIN_KERNEL:
2806 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2807 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2808 goto error;
2809 }
2810 break;
2811 }
2812 }
2813
2814 /*
2815 * Check that the UID or GID match that of the tracing session.
2816 * The root user can interact with all sessions.
2817 */
2818 if (need_tracing_session) {
2819 if (!session_access_ok(cmd_ctx->session,
2820 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2821 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2822 ret = LTTNG_ERR_EPERM;
2823 goto error;
2824 }
2825 }
2826
2827 /*
2828 * Send relayd information to consumer as soon as we have a domain and a
2829 * session defined.
2830 */
2831 if (cmd_ctx->session && need_domain) {
2832 /*
2833 * Setup relayd if not done yet. If the relayd information was already
2834 * sent to the consumer, this call will gracefully return.
2835 */
2836 ret = cmd_setup_relayd(cmd_ctx->session);
2837 if (ret != LTTNG_OK) {
2838 goto error;
2839 }
2840 }
2841
2842 /* Process by command type */
2843 switch (cmd_ctx->lsm->cmd_type) {
2844 case LTTNG_ADD_CONTEXT:
2845 {
2846 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2847 cmd_ctx->lsm->u.context.channel_name,
2848 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2849 break;
2850 }
2851 case LTTNG_DISABLE_CHANNEL:
2852 {
2853 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2854 cmd_ctx->lsm->u.disable.channel_name);
2855 break;
2856 }
2857 case LTTNG_DISABLE_EVENT:
2858 {
2859 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2860 cmd_ctx->lsm->u.disable.channel_name,
2861 cmd_ctx->lsm->u.disable.name);
2862 break;
2863 }
2864 case LTTNG_DISABLE_ALL_EVENT:
2865 {
2866 DBG("Disabling all events");
2867
2868 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2869 cmd_ctx->lsm->u.disable.channel_name);
2870 break;
2871 }
2872 case LTTNG_ENABLE_CHANNEL:
2873 {
2874 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2875 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2876 break;
2877 }
2878 case LTTNG_ENABLE_EVENT:
2879 {
2880 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2881 cmd_ctx->lsm->u.enable.channel_name,
2882 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2883 break;
2884 }
2885 case LTTNG_ENABLE_ALL_EVENT:
2886 {
2887 DBG("Enabling all events");
2888
2889 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2890 cmd_ctx->lsm->u.enable.channel_name,
2891 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2892 break;
2893 }
2894 case LTTNG_LIST_TRACEPOINTS:
2895 {
2896 struct lttng_event *events;
2897 ssize_t nb_events;
2898
2899 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2900 if (nb_events < 0) {
2901 /* Return value is a negative lttng_error_code. */
2902 ret = -nb_events;
2903 goto error;
2904 }
2905
2906 /*
2907 * Setup lttng message with payload size set to the event list size in
2908 * bytes and then copy list into the llm payload.
2909 */
2910 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2911 if (ret < 0) {
2912 free(events);
2913 goto setup_error;
2914 }
2915
2916 /* Copy event list into message payload */
2917 memcpy(cmd_ctx->llm->payload, events,
2918 sizeof(struct lttng_event) * nb_events);
2919
2920 free(events);
2921
2922 ret = LTTNG_OK;
2923 break;
2924 }
2925 case LTTNG_LIST_TRACEPOINT_FIELDS:
2926 {
2927 struct lttng_event_field *fields;
2928 ssize_t nb_fields;
2929
2930 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2931 &fields);
2932 if (nb_fields < 0) {
2933 /* Return value is a negative lttng_error_code. */
2934 ret = -nb_fields;
2935 goto error;
2936 }
2937
2938 /*
2939 * Setup lttng message with payload size set to the event list size in
2940 * bytes and then copy list into the llm payload.
2941 */
2942 ret = setup_lttng_msg(cmd_ctx,
2943 sizeof(struct lttng_event_field) * nb_fields);
2944 if (ret < 0) {
2945 free(fields);
2946 goto setup_error;
2947 }
2948
2949 /* Copy event list into message payload */
2950 memcpy(cmd_ctx->llm->payload, fields,
2951 sizeof(struct lttng_event_field) * nb_fields);
2952
2953 free(fields);
2954
2955 ret = LTTNG_OK;
2956 break;
2957 }
2958 case LTTNG_SET_CONSUMER_URI:
2959 {
2960 size_t nb_uri, len;
2961 struct lttng_uri *uris;
2962
2963 nb_uri = cmd_ctx->lsm->u.uri.size;
2964 len = nb_uri * sizeof(struct lttng_uri);
2965
2966 if (nb_uri == 0) {
2967 ret = LTTNG_ERR_INVALID;
2968 goto error;
2969 }
2970
2971 uris = zmalloc(len);
2972 if (uris == NULL) {
2973 ret = LTTNG_ERR_FATAL;
2974 goto error;
2975 }
2976
2977 /* Receive variable len data */
2978 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2979 ret = lttcomm_recv_unix_sock(sock, uris, len);
2980 if (ret <= 0) {
2981 DBG("No URIs received from client... continuing");
2982 *sock_error = 1;
2983 ret = LTTNG_ERR_SESSION_FAIL;
2984 free(uris);
2985 goto error;
2986 }
2987
2988 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2989 nb_uri, uris);
2990 if (ret != LTTNG_OK) {
2991 free(uris);
2992 goto error;
2993 }
2994
2995 /*
2996 * XXX: 0 means that this URI should be applied on the session. Should
2997 * be a DOMAIN enuam.
2998 */
2999 if (cmd_ctx->lsm->domain.type == 0) {
3000 /* Add the URI for the UST session if a consumer is present. */
3001 if (cmd_ctx->session->ust_session &&
3002 cmd_ctx->session->ust_session->consumer) {
3003 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3004 nb_uri, uris);
3005 } else if (cmd_ctx->session->kernel_session &&
3006 cmd_ctx->session->kernel_session->consumer) {
3007 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3008 cmd_ctx->session, nb_uri, uris);
3009 }
3010 }
3011
3012 free(uris);
3013
3014 break;
3015 }
3016 case LTTNG_START_TRACE:
3017 {
3018 ret = cmd_start_trace(cmd_ctx->session);
3019 break;
3020 }
3021 case LTTNG_STOP_TRACE:
3022 {
3023 ret = cmd_stop_trace(cmd_ctx->session);
3024 break;
3025 }
3026 case LTTNG_CREATE_SESSION:
3027 {
3028 size_t nb_uri, len;
3029 struct lttng_uri *uris = NULL;
3030
3031 nb_uri = cmd_ctx->lsm->u.uri.size;
3032 len = nb_uri * sizeof(struct lttng_uri);
3033
3034 if (nb_uri > 0) {
3035 uris = zmalloc(len);
3036 if (uris == NULL) {
3037 ret = LTTNG_ERR_FATAL;
3038 goto error;
3039 }
3040
3041 /* Receive variable len data */
3042 DBG("Waiting for %zu URIs from client ...", nb_uri);
3043 ret = lttcomm_recv_unix_sock(sock, uris, len);
3044 if (ret <= 0) {
3045 DBG("No URIs received from client... continuing");
3046 *sock_error = 1;
3047 ret = LTTNG_ERR_SESSION_FAIL;
3048 free(uris);
3049 goto error;
3050 }
3051
3052 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3053 DBG("Creating session with ONE network URI is a bad call");
3054 ret = LTTNG_ERR_SESSION_FAIL;
3055 free(uris);
3056 goto error;
3057 }
3058 }
3059
3060 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3061 &cmd_ctx->creds);
3062
3063 free(uris);
3064
3065 break;
3066 }
3067 case LTTNG_DESTROY_SESSION:
3068 {
3069 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3070
3071 /* Set session to NULL so we do not unlock it after free. */
3072 cmd_ctx->session = NULL;
3073 break;
3074 }
3075 case LTTNG_LIST_DOMAINS:
3076 {
3077 ssize_t nb_dom;
3078 struct lttng_domain *domains;
3079
3080 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3081 if (nb_dom < 0) {
3082 /* Return value is a negative lttng_error_code. */
3083 ret = -nb_dom;
3084 goto error;
3085 }
3086
3087 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3088 if (ret < 0) {
3089 free(domains);
3090 goto setup_error;
3091 }
3092
3093 /* Copy event list into message payload */
3094 memcpy(cmd_ctx->llm->payload, domains,
3095 nb_dom * sizeof(struct lttng_domain));
3096
3097 free(domains);
3098
3099 ret = LTTNG_OK;
3100 break;
3101 }
3102 case LTTNG_LIST_CHANNELS:
3103 {
3104 int nb_chan;
3105 struct lttng_channel *channels;
3106
3107 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3108 cmd_ctx->session, &channels);
3109 if (nb_chan < 0) {
3110 /* Return value is a negative lttng_error_code. */
3111 ret = -nb_chan;
3112 goto error;
3113 }
3114
3115 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3116 if (ret < 0) {
3117 free(channels);
3118 goto setup_error;
3119 }
3120
3121 /* Copy event list into message payload */
3122 memcpy(cmd_ctx->llm->payload, channels,
3123 nb_chan * sizeof(struct lttng_channel));
3124
3125 free(channels);
3126
3127 ret = LTTNG_OK;
3128 break;
3129 }
3130 case LTTNG_LIST_EVENTS:
3131 {
3132 ssize_t nb_event;
3133 struct lttng_event *events = NULL;
3134
3135 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3136 cmd_ctx->lsm->u.list.channel_name, &events);
3137 if (nb_event < 0) {
3138 /* Return value is a negative lttng_error_code. */
3139 ret = -nb_event;
3140 goto error;
3141 }
3142
3143 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3144 if (ret < 0) {
3145 free(events);
3146 goto setup_error;
3147 }
3148
3149 /* Copy event list into message payload */
3150 memcpy(cmd_ctx->llm->payload, events,
3151 nb_event * sizeof(struct lttng_event));
3152
3153 free(events);
3154
3155 ret = LTTNG_OK;
3156 break;
3157 }
3158 case LTTNG_LIST_SESSIONS:
3159 {
3160 unsigned int nr_sessions;
3161
3162 session_lock_list();
3163 nr_sessions = lttng_sessions_count(
3164 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3165 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3166
3167 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3168 if (ret < 0) {
3169 session_unlock_list();
3170 goto setup_error;
3171 }
3172
3173 /* Filled the session array */
3174 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3175 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3176 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3177
3178 session_unlock_list();
3179
3180 ret = LTTNG_OK;
3181 break;
3182 }
3183 case LTTNG_CALIBRATE:
3184 {
3185 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3186 &cmd_ctx->lsm->u.calibrate);
3187 break;
3188 }
3189 case LTTNG_REGISTER_CONSUMER:
3190 {
3191 struct consumer_data *cdata;
3192
3193 switch (cmd_ctx->lsm->domain.type) {
3194 case LTTNG_DOMAIN_KERNEL:
3195 cdata = &kconsumer_data;
3196 break;
3197 default:
3198 ret = LTTNG_ERR_UND;
3199 goto error;
3200 }
3201
3202 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3203 cmd_ctx->lsm->u.reg.path, cdata);
3204 break;
3205 }
3206 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3207 {
3208 struct lttng_filter_bytecode *bytecode;
3209
3210 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3211 ret = LTTNG_ERR_FILTER_INVAL;
3212 goto error;
3213 }
3214 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3215 ret = LTTNG_ERR_FILTER_INVAL;
3216 goto error;
3217 }
3218 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3219 if (!bytecode) {
3220 ret = LTTNG_ERR_FILTER_NOMEM;
3221 goto error;
3222 }
3223 /* Receive var. len. data */
3224 DBG("Receiving var len data from client ...");
3225 ret = lttcomm_recv_unix_sock(sock, bytecode,
3226 cmd_ctx->lsm->u.enable.bytecode_len);
3227 if (ret <= 0) {
3228 DBG("Nothing recv() from client var len data... continuing");
3229 *sock_error = 1;
3230 ret = LTTNG_ERR_FILTER_INVAL;
3231 goto error;
3232 }
3233
3234 if (bytecode->len + sizeof(*bytecode)
3235 != cmd_ctx->lsm->u.enable.bytecode_len) {
3236 free(bytecode);
3237 ret = LTTNG_ERR_FILTER_INVAL;
3238 goto error;
3239 }
3240
3241 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3242 cmd_ctx->lsm->u.enable.channel_name,
3243 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3244 break;
3245 }
3246 case LTTNG_DATA_PENDING:
3247 {
3248 ret = cmd_data_pending(cmd_ctx->session);
3249 break;
3250 }
3251 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3252 {
3253 struct lttcomm_lttng_output_id reply;
3254
3255 ret = cmd_snapshot_add_output(cmd_ctx->session,
3256 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3257 if (ret != LTTNG_OK) {
3258 goto error;
3259 }
3260
3261 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3262 if (ret < 0) {
3263 goto setup_error;
3264 }
3265
3266 /* Copy output list into message payload */
3267 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3268 ret = LTTNG_OK;
3269 break;
3270 }
3271 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3272 {
3273 ret = cmd_snapshot_del_output(cmd_ctx->session,
3274 &cmd_ctx->lsm->u.snapshot_output.output);
3275 break;
3276 }
3277 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3278 {
3279 ssize_t nb_output;
3280 struct lttng_snapshot_output *outputs = NULL;
3281
3282 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3283 if (nb_output < 0) {
3284 ret = -nb_output;
3285 goto error;
3286 }
3287
3288 ret = setup_lttng_msg(cmd_ctx,
3289 nb_output * sizeof(struct lttng_snapshot_output));
3290 if (ret < 0) {
3291 free(outputs);
3292 goto setup_error;
3293 }
3294
3295 if (outputs) {
3296 /* Copy output list into message payload */
3297 memcpy(cmd_ctx->llm->payload, outputs,
3298 nb_output * sizeof(struct lttng_snapshot_output));
3299 free(outputs);
3300 }
3301
3302 ret = LTTNG_OK;
3303 break;
3304 }
3305 case LTTNG_SNAPSHOT_RECORD:
3306 {
3307 ret = cmd_snapshot_record(cmd_ctx->session,
3308 &cmd_ctx->lsm->u.snapshot_record.output,
3309 cmd_ctx->lsm->u.snapshot_record.wait);
3310 break;
3311 }
3312 case LTTNG_CREATE_SESSION_SNAPSHOT:
3313 {
3314 size_t nb_uri, len;
3315 struct lttng_uri *uris = NULL;
3316
3317 nb_uri = cmd_ctx->lsm->u.uri.size;
3318 len = nb_uri * sizeof(struct lttng_uri);
3319
3320 if (nb_uri > 0) {
3321 uris = zmalloc(len);
3322 if (uris == NULL) {
3323 ret = LTTNG_ERR_FATAL;
3324 goto error;
3325 }
3326
3327 /* Receive variable len data */
3328 DBG("Waiting for %zu URIs from client ...", nb_uri);
3329 ret = lttcomm_recv_unix_sock(sock, uris, len);
3330 if (ret <= 0) {
3331 DBG("No URIs received from client... continuing");
3332 *sock_error = 1;
3333 ret = LTTNG_ERR_SESSION_FAIL;
3334 free(uris);
3335 goto error;
3336 }
3337
3338 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3339 DBG("Creating session with ONE network URI is a bad call");
3340 ret = LTTNG_ERR_SESSION_FAIL;
3341 free(uris);
3342 goto error;
3343 }
3344 }
3345
3346 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3347 nb_uri, &cmd_ctx->creds);
3348 free(uris);
3349 break;
3350 }
3351 default:
3352 ret = LTTNG_ERR_UND;
3353 break;
3354 }
3355
3356 error:
3357 if (cmd_ctx->llm == NULL) {
3358 DBG("Missing llm structure. Allocating one.");
3359 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3360 goto setup_error;
3361 }
3362 }
3363 /* Set return code */
3364 cmd_ctx->llm->ret_code = ret;
3365 setup_error:
3366 if (cmd_ctx->session) {
3367 session_unlock(cmd_ctx->session);
3368 }
3369 if (need_tracing_session) {
3370 session_unlock_list();
3371 }
3372 init_setup_error:
3373 return ret;
3374 }
3375
3376 /*
3377 * Thread managing health check socket.
3378 */
3379 static void *thread_manage_health(void *data)
3380 {
3381 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3382 uint32_t revents, nb_fd;
3383 struct lttng_poll_event events;
3384 struct lttcomm_health_msg msg;
3385 struct lttcomm_health_data reply;
3386
3387 DBG("[thread] Manage health check started");
3388
3389 rcu_register_thread();
3390
3391 /* We might hit an error path before this is created. */
3392 lttng_poll_init(&events);
3393
3394 /* Create unix socket */
3395 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3396 if (sock < 0) {
3397 ERR("Unable to create health check Unix socket");
3398 ret = -1;
3399 goto error;
3400 }
3401
3402 /*
3403 * Set the CLOEXEC flag. Return code is useless because either way, the
3404 * show must go on.
3405 */
3406 (void) utils_set_fd_cloexec(sock);
3407
3408 ret = lttcomm_listen_unix_sock(sock);
3409 if (ret < 0) {
3410 goto error;
3411 }
3412
3413 /*
3414 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3415 * more will be added to this poll set.
3416 */
3417 ret = sessiond_set_thread_pollset(&events, 2);
3418 if (ret < 0) {
3419 goto error;
3420 }
3421
3422 /* Add the application registration socket */
3423 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3424 if (ret < 0) {
3425 goto error;
3426 }
3427
3428 while (1) {
3429 DBG("Health check ready");
3430
3431 /* Inifinite blocking call, waiting for transmission */
3432 restart:
3433 ret = lttng_poll_wait(&events, -1);
3434 if (ret < 0) {
3435 /*
3436 * Restart interrupted system call.
3437 */
3438 if (errno == EINTR) {
3439 goto restart;
3440 }
3441 goto error;
3442 }
3443
3444 nb_fd = ret;
3445
3446 for (i = 0; i < nb_fd; i++) {
3447 /* Fetch once the poll data */
3448 revents = LTTNG_POLL_GETEV(&events, i);
3449 pollfd = LTTNG_POLL_GETFD(&events, i);
3450
3451 /* Thread quit pipe has been closed. Killing thread. */
3452 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3453 if (ret) {
3454 err = 0;
3455 goto exit;
3456 }
3457
3458 /* Event on the registration socket */
3459 if (pollfd == sock) {
3460 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3461 ERR("Health socket poll error");
3462 goto error;
3463 }
3464 }
3465 }
3466
3467 new_sock = lttcomm_accept_unix_sock(sock);
3468 if (new_sock < 0) {
3469 goto error;
3470 }
3471
3472 /*
3473 * Set the CLOEXEC flag. Return code is useless because either way, the
3474 * show must go on.
3475 */
3476 (void) utils_set_fd_cloexec(new_sock);
3477
3478 DBG("Receiving data from client for health...");
3479 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3480 if (ret <= 0) {
3481 DBG("Nothing recv() from client... continuing");
3482 ret = close(new_sock);
3483 if (ret) {
3484 PERROR("close");
3485 }
3486 new_sock = -1;
3487 continue;
3488 }
3489
3490 rcu_thread_online();
3491
3492 switch (msg.component) {
3493 case LTTNG_HEALTH_CMD:
3494 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3495 break;
3496 case LTTNG_HEALTH_APP_MANAGE:
3497 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3498 break;
3499 case LTTNG_HEALTH_APP_REG:
3500 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3501 break;
3502 case LTTNG_HEALTH_KERNEL:
3503 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3504 break;
3505 case LTTNG_HEALTH_CONSUMER:
3506 reply.ret_code = check_consumer_health();
3507 break;
3508 case LTTNG_HEALTH_HT_CLEANUP:
3509 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3510 break;
3511 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3512 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3513 break;
3514 case LTTNG_HEALTH_APP_REG_DISPATCH:
3515 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3516 break;
3517 case LTTNG_HEALTH_ALL:
3518 reply.ret_code =
3519 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3520 health_check_state(HEALTH_TYPE_APP_REG) &&
3521 health_check_state(HEALTH_TYPE_CMD) &&
3522 health_check_state(HEALTH_TYPE_KERNEL) &&
3523 check_consumer_health() &&
3524 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
3525 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3526 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3527 break;
3528 default:
3529 reply.ret_code = LTTNG_ERR_UND;
3530 break;
3531 }
3532
3533 /*
3534 * Flip ret value since 0 is a success and 1 indicates a bad health for
3535 * the client where in the sessiond it is the opposite. Again, this is
3536 * just to make things easier for us poor developer which enjoy a lot
3537 * lazyness.
3538 */
3539 if (reply.ret_code == 0 || reply.ret_code == 1) {
3540 reply.ret_code = !reply.ret_code;
3541 }
3542
3543 DBG2("Health check return value %d", reply.ret_code);
3544
3545 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3546 if (ret < 0) {
3547 ERR("Failed to send health data back to client");
3548 }
3549
3550 /* End of transmission */
3551 ret = close(new_sock);
3552 if (ret) {
3553 PERROR("close");
3554 }
3555 new_sock = -1;
3556 }
3557
3558 exit:
3559 error:
3560 if (err) {
3561 ERR("Health error occurred in %s", __func__);
3562 }
3563 DBG("Health check thread dying");
3564 unlink(health_unix_sock_path);
3565 if (sock >= 0) {
3566 ret = close(sock);
3567 if (ret) {
3568 PERROR("close");
3569 }
3570 }
3571
3572 lttng_poll_clean(&events);
3573
3574 rcu_unregister_thread();
3575 return NULL;
3576 }
3577
3578 /*
3579 * This thread manage all clients request using the unix client socket for
3580 * communication.
3581 */
3582 static void *thread_manage_clients(void *data)
3583 {
3584 int sock = -1, ret, i, pollfd, err = -1;
3585 int sock_error;
3586 uint32_t revents, nb_fd;
3587 struct command_ctx *cmd_ctx = NULL;
3588 struct lttng_poll_event events;
3589
3590 DBG("[thread] Manage client started");
3591
3592 rcu_register_thread();
3593
3594 health_register(HEALTH_TYPE_CMD);
3595
3596 if (testpoint(thread_manage_clients)) {
3597 goto error_testpoint;
3598 }
3599
3600 health_code_update();
3601
3602 ret = lttcomm_listen_unix_sock(client_sock);
3603 if (ret < 0) {
3604 goto error_listen;
3605 }
3606
3607 /*
3608 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3609 * more will be added to this poll set.
3610 */
3611 ret = sessiond_set_thread_pollset(&events, 2);
3612 if (ret < 0) {
3613 goto error_create_poll;
3614 }
3615
3616 /* Add the application registration socket */
3617 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3618 if (ret < 0) {
3619 goto error;
3620 }
3621
3622 /*
3623 * Notify parent pid that we are ready to accept command for client side.
3624 */
3625 if (opt_sig_parent) {
3626 kill(ppid, SIGUSR1);
3627 }
3628
3629 if (testpoint(thread_manage_clients_before_loop)) {
3630 goto error;
3631 }
3632
3633 health_code_update();
3634
3635 while (1) {
3636 DBG("Accepting client command ...");
3637
3638 /* Inifinite blocking call, waiting for transmission */
3639 restart:
3640 health_poll_entry();
3641 ret = lttng_poll_wait(&events, -1);
3642 health_poll_exit();
3643 if (ret < 0) {
3644 /*
3645 * Restart interrupted system call.
3646 */
3647 if (errno == EINTR) {
3648 goto restart;
3649 }
3650 goto error;
3651 }
3652
3653 nb_fd = ret;
3654
3655 for (i = 0; i < nb_fd; i++) {
3656 /* Fetch once the poll data */
3657 revents = LTTNG_POLL_GETEV(&events, i);
3658 pollfd = LTTNG_POLL_GETFD(&events, i);
3659
3660 health_code_update();
3661
3662 /* Thread quit pipe has been closed. Killing thread. */
3663 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3664 if (ret) {
3665 err = 0;
3666 goto exit;
3667 }
3668
3669 /* Event on the registration socket */
3670 if (pollfd == client_sock) {
3671 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3672 ERR("Client socket poll error");
3673 goto error;
3674 }
3675 }
3676 }
3677
3678 DBG("Wait for client response");
3679
3680 health_code_update();
3681
3682 sock = lttcomm_accept_unix_sock(client_sock);
3683 if (sock < 0) {
3684 goto error;
3685 }
3686
3687 /*
3688 * Set the CLOEXEC flag. Return code is useless because either way, the
3689 * show must go on.
3690 */
3691 (void) utils_set_fd_cloexec(sock);
3692
3693 /* Set socket option for credentials retrieval */
3694 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3695 if (ret < 0) {
3696 goto error;
3697 }
3698
3699 /* Allocate context command to process the client request */
3700 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3701 if (cmd_ctx == NULL) {
3702 PERROR("zmalloc cmd_ctx");
3703 goto error;
3704 }
3705
3706 /* Allocate data buffer for reception */
3707 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3708 if (cmd_ctx->lsm == NULL) {
3709 PERROR("zmalloc cmd_ctx->lsm");
3710 goto error;
3711 }
3712
3713 cmd_ctx->llm = NULL;
3714 cmd_ctx->session = NULL;
3715
3716 health_code_update();
3717
3718 /*
3719 * Data is received from the lttng client. The struct
3720 * lttcomm_session_msg (lsm) contains the command and data request of
3721 * the client.
3722 */
3723 DBG("Receiving data from client ...");
3724 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3725 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3726 if (ret <= 0) {
3727 DBG("Nothing recv() from client... continuing");
3728 ret = close(sock);
3729 if (ret) {
3730 PERROR("close");
3731 }
3732 sock = -1;
3733 clean_command_ctx(&cmd_ctx);
3734 continue;
3735 }
3736
3737 health_code_update();
3738
3739 // TODO: Validate cmd_ctx including sanity check for
3740 // security purpose.
3741
3742 rcu_thread_online();
3743 /*
3744 * This function dispatch the work to the kernel or userspace tracer
3745 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3746 * informations for the client. The command context struct contains
3747 * everything this function may needs.
3748 */
3749 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3750 rcu_thread_offline();
3751 if (ret < 0) {
3752 ret = close(sock);
3753 if (ret) {
3754 PERROR("close");
3755 }
3756 sock = -1;
3757 /*
3758 * TODO: Inform client somehow of the fatal error. At
3759 * this point, ret < 0 means that a zmalloc failed
3760 * (ENOMEM). Error detected but still accept
3761 * command, unless a socket error has been
3762 * detected.
3763 */
3764 clean_command_ctx(&cmd_ctx);
3765 continue;
3766 }
3767
3768 health_code_update();
3769
3770 DBG("Sending response (size: %d, retcode: %s)",
3771 cmd_ctx->lttng_msg_size,
3772 lttng_strerror(-cmd_ctx->llm->ret_code));
3773 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3774 if (ret < 0) {
3775 ERR("Failed to send data back to client");
3776 }
3777
3778 /* End of transmission */
3779 ret = close(sock);
3780 if (ret) {
3781 PERROR("close");
3782 }
3783 sock = -1;
3784
3785 clean_command_ctx(&cmd_ctx);
3786
3787 health_code_update();
3788 }
3789
3790 exit:
3791 error:
3792 if (sock >= 0) {
3793 ret = close(sock);
3794 if (ret) {
3795 PERROR("close");
3796 }
3797 }
3798
3799 lttng_poll_clean(&events);
3800 clean_command_ctx(&cmd_ctx);
3801
3802 error_listen:
3803 error_create_poll:
3804 error_testpoint:
3805 unlink(client_unix_sock_path);
3806 if (client_sock >= 0) {
3807 ret = close(client_sock);
3808 if (ret) {
3809 PERROR("close");
3810 }
3811 }
3812
3813 if (err) {
3814 health_error();
3815 ERR("Health error occurred in %s", __func__);
3816 }
3817
3818 health_unregister();
3819
3820 DBG("Client thread dying");
3821
3822 rcu_unregister_thread();
3823 return NULL;
3824 }
3825
3826
3827 /*
3828 * usage function on stderr
3829 */
3830 static void usage(void)
3831 {
3832 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3833 fprintf(stderr, " -h, --help Display this usage.\n");
3834 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3835 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3836 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3837 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3838 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3839 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3840 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3841 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3842 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3843 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3844 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3845 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3846 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3847 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3848 fprintf(stderr, " -V, --version Show version number.\n");
3849 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3850 fprintf(stderr, " -q, --quiet No output at all.\n");
3851 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3852 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3853 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3854 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3855 }
3856
3857 /*
3858 * daemon argument parsing
3859 */
3860 static int parse_args(int argc, char **argv)
3861 {
3862 int c;
3863
3864 static struct option long_options[] = {
3865 { "client-sock", 1, 0, 'c' },
3866 { "apps-sock", 1, 0, 'a' },
3867 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3868 { "kconsumerd-err-sock", 1, 0, 'E' },
3869 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3870 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3871 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3872 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3873 { "consumerd32-path", 1, 0, 'u' },
3874 { "consumerd32-libdir", 1, 0, 'U' },
3875 { "consumerd64-path", 1, 0, 't' },
3876 { "consumerd64-libdir", 1, 0, 'T' },
3877 { "daemonize", 0, 0, 'd' },
3878 { "sig-parent", 0, 0, 'S' },
3879 { "help", 0, 0, 'h' },
3880 { "group", 1, 0, 'g' },
3881 { "version", 0, 0, 'V' },
3882 { "quiet", 0, 0, 'q' },
3883 { "verbose", 0, 0, 'v' },
3884 { "verbose-consumer", 0, 0, 'Z' },
3885 { "no-kernel", 0, 0, 'N' },
3886 { "pidfile", 1, 0, 'p' },
3887 { NULL, 0, 0, 0 }
3888 };
3889
3890 while (1) {
3891 int option_index = 0;
3892 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3893 long_options, &option_index);
3894 if (c == -1) {
3895 break;
3896 }
3897
3898 switch (c) {
3899 case 0:
3900 fprintf(stderr, "option %s", long_options[option_index].name);
3901 if (optarg) {
3902 fprintf(stderr, " with arg %s\n", optarg);
3903 }
3904 break;
3905 case 'c':
3906 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3907 break;
3908 case 'a':
3909 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3910 break;
3911 case 'd':
3912 opt_daemon = 1;
3913 break;
3914 case 'g':
3915 opt_tracing_group = optarg;
3916 break;
3917 case 'h':
3918 usage();
3919 exit(EXIT_FAILURE);
3920 case 'V':
3921 fprintf(stdout, "%s\n", VERSION);
3922 exit(EXIT_SUCCESS);
3923 case 'S':
3924 opt_sig_parent = 1;
3925 break;
3926 case 'E':
3927 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3928 break;
3929 case 'C':
3930 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3931 break;
3932 case 'F':
3933 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3934 break;
3935 case 'D':
3936 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3937 break;
3938 case 'H':
3939 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3940 break;
3941 case 'G':
3942 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3943 break;
3944 case 'N':
3945 opt_no_kernel = 1;
3946 break;
3947 case 'q':
3948 lttng_opt_quiet = 1;
3949 break;
3950 case 'v':
3951 /* Verbose level can increase using multiple -v */
3952 lttng_opt_verbose += 1;
3953 break;
3954 case 'Z':
3955 opt_verbose_consumer += 1;
3956 break;
3957 case 'u':
3958 consumerd32_bin= optarg;
3959 break;
3960 case 'U':
3961 consumerd32_libdir = optarg;
3962 break;
3963 case 't':
3964 consumerd64_bin = optarg;
3965 break;
3966 case 'T':
3967 consumerd64_libdir = optarg;
3968 break;
3969 case 'p':
3970 opt_pidfile = optarg;
3971 break;
3972 default:
3973 /* Unknown option or other error.
3974 * Error is printed by getopt, just return */
3975 return -1;
3976 }
3977 }
3978
3979 return 0;
3980 }
3981
3982 /*
3983 * Creates the two needed socket by the daemon.
3984 * apps_sock - The communication socket for all UST apps.
3985 * client_sock - The communication of the cli tool (lttng).
3986 */
3987 static int init_daemon_socket(void)
3988 {
3989 int ret = 0;
3990 mode_t old_umask;
3991
3992 old_umask = umask(0);
3993
3994 /* Create client tool unix socket */
3995 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3996 if (client_sock < 0) {
3997 ERR("Create unix sock failed: %s", client_unix_sock_path);
3998 ret = -1;
3999 goto end;
4000 }
4001
4002 /* Set the cloexec flag */
4003 ret = utils_set_fd_cloexec(client_sock);
4004 if (ret < 0) {
4005 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4006 "Continuing but note that the consumer daemon will have a "
4007 "reference to this socket on exec()", client_sock);
4008 }
4009
4010 /* File permission MUST be 660 */
4011 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4012 if (ret < 0) {
4013 ERR("Set file permissions failed: %s", client_unix_sock_path);
4014 PERROR("chmod");
4015 goto end;
4016 }
4017
4018 /* Create the application unix socket */
4019 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4020 if (apps_sock < 0) {
4021 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4022 ret = -1;
4023 goto end;
4024 }
4025
4026 /* Set the cloexec flag */
4027 ret = utils_set_fd_cloexec(apps_sock);
4028 if (ret < 0) {
4029 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4030 "Continuing but note that the consumer daemon will have a "
4031 "reference to this socket on exec()", apps_sock);
4032 }
4033
4034 /* File permission MUST be 666 */
4035 ret = chmod(apps_unix_sock_path,
4036 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4037 if (ret < 0) {
4038 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4039 PERROR("chmod");
4040 goto end;
4041 }
4042
4043 DBG3("Session daemon client socket %d and application socket %d created",
4044 client_sock, apps_sock);
4045
4046 end:
4047 umask(old_umask);
4048 return ret;
4049 }
4050
4051 /*
4052 * Check if the global socket is available, and if a daemon is answering at the
4053 * other side. If yes, error is returned.
4054 */
4055 static int check_existing_daemon(void)
4056 {
4057 /* Is there anybody out there ? */
4058 if (lttng_session_daemon_alive()) {
4059 return -EEXIST;
4060 }
4061
4062 return 0;
4063 }
4064
4065 /*
4066 * Set the tracing group gid onto the client socket.
4067 *
4068 * Race window between mkdir and chown is OK because we are going from more
4069 * permissive (root.root) to less permissive (root.tracing).
4070 */
4071 static int set_permissions(char *rundir)
4072 {
4073 int ret;
4074 gid_t gid;
4075
4076 ret = allowed_group();
4077 if (ret < 0) {
4078 WARN("No tracing group detected");
4079 /* Setting gid to 0 if no tracing group is found */
4080 gid = 0;
4081 } else {
4082 gid = ret;
4083 }
4084
4085 /* Set lttng run dir */
4086 ret = chown(rundir, 0, gid);
4087 if (ret < 0) {
4088 ERR("Unable to set group on %s", rundir);
4089 PERROR("chown");
4090 }
4091
4092 /* Ensure all applications and tracing group can search the run dir */
4093 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
4094 if (ret < 0) {
4095 ERR("Unable to set permissions on %s", rundir);
4096 PERROR("chmod");
4097 }
4098
4099 /* lttng client socket path */
4100 ret = chown(client_unix_sock_path, 0, gid);
4101 if (ret < 0) {
4102 ERR("Unable to set group on %s", client_unix_sock_path);
4103 PERROR("chown");
4104 }
4105
4106 /* kconsumer error socket path */
4107 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
4108 if (ret < 0) {
4109 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4110 PERROR("chown");
4111 }
4112
4113 /* 64-bit ustconsumer error socket path */
4114 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4115 if (ret < 0) {
4116 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4117 PERROR("chown");
4118 }
4119
4120 /* 32-bit ustconsumer compat32 error socket path */
4121 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
4122 if (ret < 0) {
4123 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4124 PERROR("chown");
4125 }
4126
4127 DBG("All permissions are set");
4128
4129 return ret;
4130 }
4131
4132 /*
4133 * Create the lttng run directory needed for all global sockets and pipe.
4134 */
4135 static int create_lttng_rundir(const char *rundir)
4136 {
4137 int ret;
4138
4139 DBG3("Creating LTTng run directory: %s", rundir);
4140
4141 ret = mkdir(rundir, S_IRWXU);
4142 if (ret < 0) {
4143 if (errno != EEXIST) {
4144 ERR("Unable to create %s", rundir);
4145 goto error;
4146 } else {
4147 ret = 0;
4148 }
4149 }
4150
4151 error:
4152 return ret;
4153 }
4154
4155 /*
4156 * Setup sockets and directory needed by the kconsumerd communication with the
4157 * session daemon.
4158 */
4159 static int set_consumer_sockets(struct consumer_data *consumer_data,
4160 const char *rundir)
4161 {
4162 int ret;
4163 char path[PATH_MAX];
4164
4165 switch (consumer_data->type) {
4166 case LTTNG_CONSUMER_KERNEL:
4167 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4168 break;
4169 case LTTNG_CONSUMER64_UST:
4170 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4171 break;
4172 case LTTNG_CONSUMER32_UST:
4173 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4174 break;
4175 default:
4176 ERR("Consumer type unknown");
4177 ret = -EINVAL;
4178 goto error;
4179 }
4180
4181 DBG2("Creating consumer directory: %s", path);
4182
4183 ret = mkdir(path, S_IRWXU);
4184 if (ret < 0) {
4185 if (errno != EEXIST) {
4186 PERROR("mkdir");
4187 ERR("Failed to create %s", path);
4188 goto error;
4189 }
4190 ret = -1;
4191 }
4192
4193 /* Create the kconsumerd error unix socket */
4194 consumer_data->err_sock =
4195 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4196 if (consumer_data->err_sock < 0) {
4197 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4198 ret = -1;
4199 goto error;
4200 }
4201
4202 /*
4203 * Set the CLOEXEC flag. Return code is useless because either way, the
4204 * show must go on.
4205 */
4206 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4207 if (ret < 0) {
4208 PERROR("utils_set_fd_cloexec");
4209 /* continue anyway */
4210 }
4211
4212 /* File permission MUST be 660 */
4213 ret = chmod(consumer_data->err_unix_sock_path,
4214 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4215 if (ret < 0) {
4216 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4217 PERROR("chmod");
4218 goto error;
4219 }
4220
4221 error:
4222 return ret;
4223 }
4224
4225 /*
4226 * Signal handler for the daemon
4227 *
4228 * Simply stop all worker threads, leaving main() return gracefully after
4229 * joining all threads and calling cleanup().
4230 */
4231 static void sighandler(int sig)
4232 {
4233 switch (sig) {
4234 case SIGPIPE:
4235 DBG("SIGPIPE caught");
4236 return;
4237 case SIGINT:
4238 DBG("SIGINT caught");
4239 stop_threads();
4240 break;
4241 case SIGTERM:
4242 DBG("SIGTERM caught");
4243 stop_threads();
4244 break;
4245 default:
4246 break;
4247 }
4248 }
4249
4250 /*
4251 * Setup signal handler for :
4252 * SIGINT, SIGTERM, SIGPIPE
4253 */
4254 static int set_signal_handler(void)
4255 {
4256 int ret = 0;
4257 struct sigaction sa;
4258 sigset_t sigset;
4259
4260 if ((ret = sigemptyset(&sigset)) < 0) {
4261 PERROR("sigemptyset");
4262 return ret;
4263 }
4264
4265 sa.sa_handler = sighandler;
4266 sa.sa_mask = sigset;
4267 sa.sa_flags = 0;
4268 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4269 PERROR("sigaction");
4270 return ret;
4271 }
4272
4273 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4274 PERROR("sigaction");
4275 return ret;
4276 }
4277
4278 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4279 PERROR("sigaction");
4280 return ret;
4281 }
4282
4283 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4284
4285 return ret;
4286 }
4287
4288 /*
4289 * Set open files limit to unlimited. This daemon can open a large number of
4290 * file descriptors in order to consumer multiple kernel traces.
4291 */
4292 static void set_ulimit(void)
4293 {
4294 int ret;
4295 struct rlimit lim;
4296
4297 /* The kernel does not allowed an infinite limit for open files */
4298 lim.rlim_cur = 65535;
4299 lim.rlim_max = 65535;
4300
4301 ret = setrlimit(RLIMIT_NOFILE, &lim);
4302 if (ret < 0) {
4303 PERROR("failed to set open files limit");
4304 }
4305 }
4306
4307 /*
4308 * Write pidfile using the rundir and opt_pidfile.
4309 */
4310 static void write_pidfile(void)
4311 {
4312 int ret;
4313 char pidfile_path[PATH_MAX];
4314
4315 assert(rundir);
4316
4317 if (opt_pidfile) {
4318 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4319 } else {
4320 /* Build pidfile path from rundir and opt_pidfile. */
4321 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4322 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4323 if (ret < 0) {
4324 PERROR("snprintf pidfile path");
4325 goto error;
4326 }
4327 }
4328
4329 /*
4330 * Create pid file in rundir. Return value is of no importance. The
4331 * execution will continue even though we are not able to write the file.
4332 */
4333 (void) utils_create_pid_file(getpid(), pidfile_path);
4334
4335 error:
4336 return;
4337 }
4338
4339 /*
4340 * main
4341 */
4342 int main(int argc, char **argv)
4343 {
4344 int ret = 0;
4345 void *status;
4346 const char *home_path, *env_app_timeout;
4347
4348 init_kernel_workarounds();
4349
4350 rcu_register_thread();
4351
4352 setup_consumerd_path();
4353
4354 page_size = sysconf(_SC_PAGESIZE);
4355 if (page_size < 0) {
4356 PERROR("sysconf _SC_PAGESIZE");
4357 page_size = LONG_MAX;
4358 WARN("Fallback page size to %ld", page_size);
4359 }
4360
4361 /* Parse arguments */
4362 progname = argv[0];
4363 if ((ret = parse_args(argc, argv)) < 0) {
4364 goto error;
4365 }
4366
4367 /* Daemonize */
4368 if (opt_daemon) {
4369 int i;
4370
4371 /*
4372 * fork
4373 * child: setsid, close FD 0, 1, 2, chdir /
4374 * parent: exit (if fork is successful)
4375 */
4376 ret = daemon(0, 0);
4377 if (ret < 0) {
4378 PERROR("daemon");
4379 goto error;
4380 }
4381 /*
4382 * We are in the child. Make sure all other file
4383 * descriptors are closed, in case we are called with
4384 * more opened file descriptors than the standard ones.
4385 */
4386 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4387 (void) close(i);
4388 }
4389 }
4390
4391 /* Create thread quit pipe */
4392 if ((ret = init_thread_quit_pipe()) < 0) {
4393 goto error;
4394 }
4395
4396 /* Check if daemon is UID = 0 */
4397 is_root = !getuid();
4398
4399 if (is_root) {
4400 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4401
4402 /* Create global run dir with root access */
4403 ret = create_lttng_rundir(rundir);
4404 if (ret < 0) {
4405 goto error;
4406 }
4407
4408 if (strlen(apps_unix_sock_path) == 0) {
4409 snprintf(apps_unix_sock_path, PATH_MAX,
4410 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4411 }
4412
4413 if (strlen(client_unix_sock_path) == 0) {
4414 snprintf(client_unix_sock_path, PATH_MAX,
4415 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4416 }
4417
4418 /* Set global SHM for ust */
4419 if (strlen(wait_shm_path) == 0) {
4420 snprintf(wait_shm_path, PATH_MAX,
4421 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4422 }
4423
4424 if (strlen(health_unix_sock_path) == 0) {
4425 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4426 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4427 }
4428
4429 /* Setup kernel consumerd path */
4430 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4431 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4432 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4433 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4434
4435 DBG2("Kernel consumer err path: %s",
4436 kconsumer_data.err_unix_sock_path);
4437 DBG2("Kernel consumer cmd path: %s",
4438 kconsumer_data.cmd_unix_sock_path);
4439 } else {
4440 home_path = utils_get_home_dir();
4441 if (home_path == NULL) {
4442 /* TODO: Add --socket PATH option */
4443 ERR("Can't get HOME directory for sockets creation.");
4444 ret = -EPERM;
4445 goto error;
4446 }
4447
4448 /*
4449 * Create rundir from home path. This will create something like
4450 * $HOME/.lttng
4451 */
4452 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4453 if (ret < 0) {
4454 ret = -ENOMEM;
4455 goto error;
4456 }
4457
4458 ret = create_lttng_rundir(rundir);
4459 if (ret < 0) {
4460 goto error;
4461 }
4462
4463 if (strlen(apps_unix_sock_path) == 0) {
4464 snprintf(apps_unix_sock_path, PATH_MAX,
4465 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4466 }
4467
4468 /* Set the cli tool unix socket path */
4469 if (strlen(client_unix_sock_path) == 0) {
4470 snprintf(client_unix_sock_path, PATH_MAX,
4471 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4472 }
4473
4474 /* Set global SHM for ust */
4475 if (strlen(wait_shm_path) == 0) {
4476 snprintf(wait_shm_path, PATH_MAX,
4477 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4478 }
4479
4480 /* Set health check Unix path */
4481 if (strlen(health_unix_sock_path) == 0) {
4482 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4483 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4484 }
4485 }
4486
4487 /* Set consumer initial state */
4488 kernel_consumerd_state = CONSUMER_STOPPED;
4489 ust_consumerd_state = CONSUMER_STOPPED;
4490
4491 DBG("Client socket path %s", client_unix_sock_path);
4492 DBG("Application socket path %s", apps_unix_sock_path);
4493 DBG("Application wait path %s", wait_shm_path);
4494 DBG("LTTng run directory path: %s", rundir);
4495
4496 /* 32 bits consumerd path setup */
4497 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4498 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4499 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4500 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4501
4502 DBG2("UST consumer 32 bits err path: %s",
4503 ustconsumer32_data.err_unix_sock_path);
4504 DBG2("UST consumer 32 bits cmd path: %s",
4505 ustconsumer32_data.cmd_unix_sock_path);
4506
4507 /* 64 bits consumerd path setup */
4508 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4509 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4510 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4511 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4512
4513 DBG2("UST consumer 64 bits err path: %s",
4514 ustconsumer64_data.err_unix_sock_path);
4515 DBG2("UST consumer 64 bits cmd path: %s",
4516 ustconsumer64_data.cmd_unix_sock_path);
4517
4518 /*
4519 * See if daemon already exist.
4520 */
4521 if ((ret = check_existing_daemon()) < 0) {
4522 ERR("Already running daemon.\n");
4523 /*
4524 * We do not goto exit because we must not cleanup()
4525 * because a daemon is already running.
4526 */
4527 goto error;
4528 }
4529
4530 /*
4531 * Init UST app hash table. Alloc hash table before this point since
4532 * cleanup() can get called after that point.
4533 */
4534 ust_app_ht_alloc();
4535
4536 /* After this point, we can safely call cleanup() with "goto exit" */
4537
4538 /*
4539 * These actions must be executed as root. We do that *after* setting up
4540 * the sockets path because we MUST make the check for another daemon using
4541 * those paths *before* trying to set the kernel consumer sockets and init
4542 * kernel tracer.
4543 */
4544 if (is_root) {
4545 ret = set_consumer_sockets(&kconsumer_data, rundir);
4546 if (ret < 0) {
4547 goto exit;
4548 }
4549
4550 /* Setup kernel tracer */
4551 if (!opt_no_kernel) {
4552 init_kernel_tracer();
4553 }
4554
4555 /* Set ulimit for open files */
4556 set_ulimit();
4557 }
4558 /* init lttng_fd tracking must be done after set_ulimit. */
4559 lttng_fd_init();
4560
4561 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4562 if (ret < 0) {
4563 goto exit;
4564 }
4565
4566 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4567 if (ret < 0) {
4568 goto exit;
4569 }
4570
4571 if ((ret = set_signal_handler()) < 0) {
4572 goto exit;
4573 }
4574
4575 /* Setup the needed unix socket */
4576 if ((ret = init_daemon_socket()) < 0) {
4577 goto exit;
4578 }
4579
4580 /* Set credentials to socket */
4581 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4582 goto exit;
4583 }
4584
4585 /* Get parent pid if -S, --sig-parent is specified. */
4586 if (opt_sig_parent) {
4587 ppid = getppid();
4588 }
4589
4590 /* Setup the kernel pipe for waking up the kernel thread */
4591 if (is_root && !opt_no_kernel) {
4592 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4593 goto exit;
4594 }
4595 }
4596
4597 /* Setup the thread ht_cleanup communication pipe. */
4598 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4599 goto exit;
4600 }
4601
4602 /* Setup the thread apps communication pipe. */
4603 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4604 goto exit;
4605 }
4606
4607 /* Setup the thread apps notify communication pipe. */
4608 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4609 goto exit;
4610 }
4611
4612 /* Initialize global buffer per UID and PID registry. */
4613 buffer_reg_init_uid_registry();
4614 buffer_reg_init_pid_registry();
4615
4616 /* Init UST command queue. */
4617 cds_wfq_init(&ust_cmd_queue.queue);
4618
4619 /*
4620 * Get session list pointer. This pointer MUST NOT be free(). This list is
4621 * statically declared in session.c
4622 */
4623 session_list_ptr = session_get_list();
4624
4625 /* Set up max poll set size */
4626 lttng_poll_set_max_size();
4627
4628 cmd_init();
4629
4630 /* Check for the application socket timeout env variable. */
4631 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4632 if (env_app_timeout) {
4633 app_socket_timeout = atoi(env_app_timeout);
4634 } else {
4635 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4636 }
4637
4638 write_pidfile();
4639
4640 /* Initialize communication library */
4641 lttcomm_init();
4642
4643 /* Create thread to manage the client socket */
4644 ret = pthread_create(&ht_cleanup_thread, NULL,
4645 thread_ht_cleanup, (void *) NULL);
4646 if (ret != 0) {
4647 PERROR("pthread_create ht_cleanup");
4648 goto exit_ht_cleanup;
4649 }
4650
4651 /* Create thread to manage the client socket */
4652 ret = pthread_create(&health_thread, NULL,
4653 thread_manage_health, (void *) NULL);
4654 if (ret != 0) {
4655 PERROR("pthread_create health");
4656 goto exit_health;
4657 }
4658
4659 /* Create thread to manage the client socket */
4660 ret = pthread_create(&client_thread, NULL,
4661 thread_manage_clients, (void *) NULL);
4662 if (ret != 0) {
4663 PERROR("pthread_create clients");
4664 goto exit_client;
4665 }
4666
4667 /* Create thread to dispatch registration */
4668 ret = pthread_create(&dispatch_thread, NULL,
4669 thread_dispatch_ust_registration, (void *) NULL);
4670 if (ret != 0) {
4671 PERROR("pthread_create dispatch");
4672 goto exit_dispatch;
4673 }
4674
4675 /* Create thread to manage application registration. */
4676 ret = pthread_create(&reg_apps_thread, NULL,
4677 thread_registration_apps, (void *) NULL);
4678 if (ret != 0) {
4679 PERROR("pthread_create registration");
4680 goto exit_reg_apps;
4681 }
4682
4683 /* Create thread to manage application socket */
4684 ret = pthread_create(&apps_thread, NULL,
4685 thread_manage_apps, (void *) NULL);
4686 if (ret != 0) {
4687 PERROR("pthread_create apps");
4688 goto exit_apps;
4689 }
4690
4691 /* Create thread to manage application notify socket */
4692 ret = pthread_create(&apps_notify_thread, NULL,
4693 ust_thread_manage_notify, (void *) NULL);
4694 if (ret != 0) {
4695 PERROR("pthread_create apps");
4696 goto exit_apps;
4697 }
4698
4699 /* Don't start this thread if kernel tracing is not requested nor root */
4700 if (is_root && !opt_no_kernel) {
4701 /* Create kernel thread to manage kernel event */
4702 ret = pthread_create(&kernel_thread, NULL,
4703 thread_manage_kernel, (void *) NULL);
4704 if (ret != 0) {
4705 PERROR("pthread_create kernel");
4706 goto exit_kernel;
4707 }
4708
4709 ret = pthread_join(kernel_thread, &status);
4710 if (ret != 0) {
4711 PERROR("pthread_join");
4712 goto error; /* join error, exit without cleanup */
4713 }
4714 }
4715
4716 exit_kernel:
4717 ret = pthread_join(apps_thread, &status);
4718 if (ret != 0) {
4719 PERROR("pthread_join");
4720 goto error; /* join error, exit without cleanup */
4721 }
4722
4723 exit_apps:
4724 ret = pthread_join(reg_apps_thread, &status);
4725 if (ret != 0) {
4726 PERROR("pthread_join");
4727 goto error; /* join error, exit without cleanup */
4728 }
4729
4730 exit_reg_apps:
4731 ret = pthread_join(dispatch_thread, &status);
4732 if (ret != 0) {
4733 PERROR("pthread_join");
4734 goto error; /* join error, exit without cleanup */
4735 }
4736
4737 exit_dispatch:
4738 ret = pthread_join(client_thread, &status);
4739 if (ret != 0) {
4740 PERROR("pthread_join");
4741 goto error; /* join error, exit without cleanup */
4742 }
4743
4744 ret = join_consumer_thread(&kconsumer_data);
4745 if (ret != 0) {
4746 PERROR("join_consumer");
4747 goto error; /* join error, exit without cleanup */
4748 }
4749
4750 ret = join_consumer_thread(&ustconsumer32_data);
4751 if (ret != 0) {
4752 PERROR("join_consumer ust32");
4753 goto error; /* join error, exit without cleanup */
4754 }
4755
4756 ret = join_consumer_thread(&ustconsumer64_data);
4757 if (ret != 0) {
4758 PERROR("join_consumer ust64");
4759 goto error; /* join error, exit without cleanup */
4760 }
4761
4762 exit_client:
4763 ret = pthread_join(health_thread, &status);
4764 if (ret != 0) {
4765 PERROR("pthread_join health thread");
4766 goto error; /* join error, exit without cleanup */
4767 }
4768
4769 exit_health:
4770 ret = pthread_join(ht_cleanup_thread, &status);
4771 if (ret != 0) {
4772 PERROR("pthread_join ht cleanup thread");
4773 goto error; /* join error, exit without cleanup */
4774 }
4775 exit_ht_cleanup:
4776 exit:
4777 /*
4778 * cleanup() is called when no other thread is running.
4779 */
4780 rcu_thread_online();
4781 cleanup();
4782 rcu_thread_offline();
4783 rcu_unregister_thread();
4784 if (!ret) {
4785 exit(EXIT_SUCCESS);
4786 }
4787 error:
4788 exit(EXIT_FAILURE);
4789 }
This page took 0.198448 seconds and 6 git commands to generate.