Add -j/--jul to lttng UI and ABI
[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-sessiond.h"
64 #include "testpoint.h"
65 #include "ust-thread.h"
66
67 #define CONSUMERD_FILE "lttng-consumerd"
68
69 const char *progname;
70 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
71 static const char *opt_pidfile;
72 static int opt_sig_parent;
73 static int opt_verbose_consumer;
74 static int opt_daemon;
75 static int opt_no_kernel;
76 static int is_root; /* Set to 1 if the daemon is running as root */
77 static pid_t ppid; /* Parent PID for --sig-parent option */
78 static char *rundir;
79
80 /*
81 * Consumer daemon specific control data. Every value not initialized here is
82 * set to 0 by the static definition.
83 */
84 static struct consumer_data kconsumer_data = {
85 .type = LTTNG_CONSUMER_KERNEL,
86 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
87 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
88 .err_sock = -1,
89 .cmd_sock = -1,
90 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
91 .lock = PTHREAD_MUTEX_INITIALIZER,
92 .cond = PTHREAD_COND_INITIALIZER,
93 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
94 };
95 static struct consumer_data ustconsumer64_data = {
96 .type = LTTNG_CONSUMER64_UST,
97 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
98 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
99 .err_sock = -1,
100 .cmd_sock = -1,
101 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
102 .lock = PTHREAD_MUTEX_INITIALIZER,
103 .cond = PTHREAD_COND_INITIALIZER,
104 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
105 };
106 static struct consumer_data ustconsumer32_data = {
107 .type = LTTNG_CONSUMER32_UST,
108 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
109 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
110 .err_sock = -1,
111 .cmd_sock = -1,
112 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
113 .lock = PTHREAD_MUTEX_INITIALIZER,
114 .cond = PTHREAD_COND_INITIALIZER,
115 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
116 };
117
118 /* Shared between threads */
119 static int dispatch_thread_exit;
120
121 /* Global application Unix socket path */
122 static char apps_unix_sock_path[PATH_MAX];
123 /* Global client Unix socket path */
124 static char client_unix_sock_path[PATH_MAX];
125 /* global wait shm path for UST */
126 static char wait_shm_path[PATH_MAX];
127 /* Global health check unix path */
128 static char health_unix_sock_path[PATH_MAX];
129
130 /* Sockets and FDs */
131 static int client_sock = -1;
132 static int apps_sock = -1;
133 int kernel_tracer_fd = -1;
134 static int kernel_poll_pipe[2] = { -1, -1 };
135
136 /*
137 * Quit pipe for all threads. This permits a single cancellation point
138 * for all threads when receiving an event on the pipe.
139 */
140 static int thread_quit_pipe[2] = { -1, -1 };
141
142 /*
143 * This pipe is used to inform the thread managing application communication
144 * that a command is queued and ready to be processed.
145 */
146 static int apps_cmd_pipe[2] = { -1, -1 };
147
148 int apps_cmd_notify_pipe[2] = { -1, -1 };
149
150 /* Pthread, Mutexes and Semaphores */
151 static pthread_t apps_thread;
152 static pthread_t apps_notify_thread;
153 static pthread_t reg_apps_thread;
154 static pthread_t client_thread;
155 static pthread_t kernel_thread;
156 static pthread_t dispatch_thread;
157 static pthread_t health_thread;
158 static pthread_t ht_cleanup_thread;
159
160 /*
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 *
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
166 */
167 static struct ust_cmd_queue ust_cmd_queue;
168
169 /*
170 * Pointer initialized before thread creation.
171 *
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
175 *
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
178 */
179 static struct ltt_session_list *session_list_ptr;
180
181 int ust_consumerd64_fd = -1;
182 int ust_consumerd32_fd = -1;
183
184 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
185 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
186 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
187 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
188
189 static const char *module_proc_lttng = "/proc/lttng";
190
191 /*
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
194 */
195 enum consumerd_state {
196 CONSUMER_STARTED = 1,
197 CONSUMER_STOPPED = 2,
198 CONSUMER_ERROR = 3,
199 };
200
201 /*
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
206 *
207 * The following example shows a possible race condition of this scheme:
208 *
209 * consumer thread error happens
210 * client cmd arrives
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
214 * ...
215 *
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
221 */
222 static enum consumerd_state ust_consumerd_state;
223 static enum consumerd_state kernel_consumerd_state;
224
225 /*
226 * Socket timeout for receiving and sending in seconds.
227 */
228 static int app_socket_timeout;
229
230 /* Set in main() with the current page size. */
231 long page_size;
232
233 /* Application health monitoring */
234 struct health_app *health_sessiond;
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 * Init thread quit pipe.
327 *
328 * Return -1 on error or 0 if all pipes are created.
329 */
330 static int init_thread_quit_pipe(void)
331 {
332 int ret, i;
333
334 ret = pipe(thread_quit_pipe);
335 if (ret < 0) {
336 PERROR("thread quit pipe");
337 goto error;
338 }
339
340 for (i = 0; i < 2; i++) {
341 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
342 if (ret < 0) {
343 PERROR("fcntl");
344 goto error;
345 }
346 }
347
348 error:
349 return ret;
350 }
351
352 /*
353 * Stop all threads by closing the thread quit pipe.
354 */
355 static void stop_threads(void)
356 {
357 int ret;
358
359 /* Stopping all threads */
360 DBG("Terminating all threads");
361 ret = notify_thread_pipe(thread_quit_pipe[1]);
362 if (ret < 0) {
363 ERR("write error on thread quit pipe");
364 }
365
366 /* Dispatch thread */
367 CMM_STORE_SHARED(dispatch_thread_exit, 1);
368 futex_nto1_wake(&ust_cmd_queue.futex);
369 }
370
371 /*
372 * Close every consumer sockets.
373 */
374 static void close_consumer_sockets(void)
375 {
376 int ret;
377
378 if (kconsumer_data.err_sock >= 0) {
379 ret = close(kconsumer_data.err_sock);
380 if (ret < 0) {
381 PERROR("kernel consumer err_sock close");
382 }
383 }
384 if (ustconsumer32_data.err_sock >= 0) {
385 ret = close(ustconsumer32_data.err_sock);
386 if (ret < 0) {
387 PERROR("UST consumerd32 err_sock close");
388 }
389 }
390 if (ustconsumer64_data.err_sock >= 0) {
391 ret = close(ustconsumer64_data.err_sock);
392 if (ret < 0) {
393 PERROR("UST consumerd64 err_sock close");
394 }
395 }
396 if (kconsumer_data.cmd_sock >= 0) {
397 ret = close(kconsumer_data.cmd_sock);
398 if (ret < 0) {
399 PERROR("kernel consumer cmd_sock close");
400 }
401 }
402 if (ustconsumer32_data.cmd_sock >= 0) {
403 ret = close(ustconsumer32_data.cmd_sock);
404 if (ret < 0) {
405 PERROR("UST consumerd32 cmd_sock close");
406 }
407 }
408 if (ustconsumer64_data.cmd_sock >= 0) {
409 ret = close(ustconsumer64_data.cmd_sock);
410 if (ret < 0) {
411 PERROR("UST consumerd64 cmd_sock close");
412 }
413 }
414 }
415
416 /*
417 * Cleanup the daemon
418 */
419 static void cleanup(void)
420 {
421 int ret;
422 struct ltt_session *sess, *stmp;
423 char path[PATH_MAX];
424
425 DBG("Cleaning up");
426
427 /*
428 * Close the thread quit pipe. It has already done its job,
429 * since we are now called.
430 */
431 utils_close_pipe(thread_quit_pipe);
432
433 /*
434 * If opt_pidfile is undefined, the default file will be wiped when
435 * removing the rundir.
436 */
437 if (opt_pidfile) {
438 ret = remove(opt_pidfile);
439 if (ret < 0) {
440 PERROR("remove pidfile %s", opt_pidfile);
441 }
442 }
443
444 DBG("Removing sessiond and consumerd content of directory %s", rundir);
445
446 /* sessiond */
447 snprintf(path, PATH_MAX,
448 "%s/%s",
449 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
450 DBG("Removing %s", path);
451 (void) unlink(path);
452
453 /* kconsumerd */
454 snprintf(path, PATH_MAX,
455 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
456 rundir);
457 DBG("Removing %s", path);
458 (void) unlink(path);
459
460 snprintf(path, PATH_MAX,
461 DEFAULT_KCONSUMERD_PATH,
462 rundir);
463 DBG("Removing directory %s", path);
464 (void) rmdir(path);
465
466 /* ust consumerd 32 */
467 snprintf(path, PATH_MAX,
468 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
469 rundir);
470 DBG("Removing %s", path);
471 (void) unlink(path);
472
473 snprintf(path, PATH_MAX,
474 DEFAULT_USTCONSUMERD32_PATH,
475 rundir);
476 DBG("Removing directory %s", path);
477 (void) rmdir(path);
478
479 /* ust consumerd 64 */
480 snprintf(path, PATH_MAX,
481 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
482 rundir);
483 DBG("Removing %s", path);
484 (void) unlink(path);
485
486 snprintf(path, PATH_MAX,
487 DEFAULT_USTCONSUMERD64_PATH,
488 rundir);
489 DBG("Removing directory %s", path);
490 (void) rmdir(path);
491
492 free(rundir);
493
494 DBG("Cleaning up all sessions");
495
496 /* Destroy session list mutex */
497 if (session_list_ptr != NULL) {
498 pthread_mutex_destroy(&session_list_ptr->lock);
499
500 /* Cleanup ALL session */
501 cds_list_for_each_entry_safe(sess, stmp,
502 &session_list_ptr->head, list) {
503 cmd_destroy_session(sess, kernel_poll_pipe[1]);
504 }
505 }
506
507 DBG("Closing all UST sockets");
508 ust_app_clean_list();
509 buffer_reg_destroy_registries();
510
511 if (is_root && !opt_no_kernel) {
512 DBG2("Closing kernel fd");
513 if (kernel_tracer_fd >= 0) {
514 ret = close(kernel_tracer_fd);
515 if (ret) {
516 PERROR("close");
517 }
518 }
519 DBG("Unloading kernel modules");
520 modprobe_remove_lttng_all();
521 }
522
523 close_consumer_sockets();
524
525 /* <fun> */
526 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
527 "Matthew, BEET driven development works!%c[%dm",
528 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
529 /* </fun> */
530 }
531
532 /*
533 * Send data on a unix socket using the liblttsessiondcomm API.
534 *
535 * Return lttcomm error code.
536 */
537 static int send_unix_sock(int sock, void *buf, size_t len)
538 {
539 /* Check valid length */
540 if (len == 0) {
541 return -1;
542 }
543
544 return lttcomm_send_unix_sock(sock, buf, len);
545 }
546
547 /*
548 * Free memory of a command context structure.
549 */
550 static void clean_command_ctx(struct command_ctx **cmd_ctx)
551 {
552 DBG("Clean command context structure");
553 if (*cmd_ctx) {
554 if ((*cmd_ctx)->llm) {
555 free((*cmd_ctx)->llm);
556 }
557 if ((*cmd_ctx)->lsm) {
558 free((*cmd_ctx)->lsm);
559 }
560 free(*cmd_ctx);
561 *cmd_ctx = NULL;
562 }
563 }
564
565 /*
566 * Notify UST applications using the shm mmap futex.
567 */
568 static int notify_ust_apps(int active)
569 {
570 char *wait_shm_mmap;
571
572 DBG("Notifying applications of session daemon state: %d", active);
573
574 /* See shm.c for this call implying mmap, shm and futex calls */
575 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
576 if (wait_shm_mmap == NULL) {
577 goto error;
578 }
579
580 /* Wake waiting process */
581 futex_wait_update((int32_t *) wait_shm_mmap, active);
582
583 /* Apps notified successfully */
584 return 0;
585
586 error:
587 return -1;
588 }
589
590 /*
591 * Setup the outgoing data buffer for the response (llm) by allocating the
592 * right amount of memory and copying the original information from the lsm
593 * structure.
594 *
595 * Return total size of the buffer pointed by buf.
596 */
597 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
598 {
599 int ret, buf_size;
600
601 buf_size = size;
602
603 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
604 if (cmd_ctx->llm == NULL) {
605 PERROR("zmalloc");
606 ret = -ENOMEM;
607 goto error;
608 }
609
610 /* Copy common data */
611 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
612 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
613
614 cmd_ctx->llm->data_size = size;
615 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
616
617 return buf_size;
618
619 error:
620 return ret;
621 }
622
623 /*
624 * Update the kernel poll set of all channel fd available over all tracing
625 * session. Add the wakeup pipe at the end of the set.
626 */
627 static int update_kernel_poll(struct lttng_poll_event *events)
628 {
629 int ret;
630 struct ltt_session *session;
631 struct ltt_kernel_channel *channel;
632
633 DBG("Updating kernel poll set");
634
635 session_lock_list();
636 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
637 session_lock(session);
638 if (session->kernel_session == NULL) {
639 session_unlock(session);
640 continue;
641 }
642
643 cds_list_for_each_entry(channel,
644 &session->kernel_session->channel_list.head, list) {
645 /* Add channel fd to the kernel poll set */
646 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
647 if (ret < 0) {
648 session_unlock(session);
649 goto error;
650 }
651 DBG("Channel fd %d added to kernel set", channel->fd);
652 }
653 session_unlock(session);
654 }
655 session_unlock_list();
656
657 return 0;
658
659 error:
660 session_unlock_list();
661 return -1;
662 }
663
664 /*
665 * Find the channel fd from 'fd' over all tracing session. When found, check
666 * for new channel stream and send those stream fds to the kernel consumer.
667 *
668 * Useful for CPU hotplug feature.
669 */
670 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
671 {
672 int ret = 0;
673 struct ltt_session *session;
674 struct ltt_kernel_session *ksess;
675 struct ltt_kernel_channel *channel;
676
677 DBG("Updating kernel streams for channel fd %d", fd);
678
679 session_lock_list();
680 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
681 session_lock(session);
682 if (session->kernel_session == NULL) {
683 session_unlock(session);
684 continue;
685 }
686 ksess = session->kernel_session;
687
688 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
689 if (channel->fd == fd) {
690 DBG("Channel found, updating kernel streams");
691 ret = kernel_open_channel_stream(channel);
692 if (ret < 0) {
693 goto error;
694 }
695 /* Update the stream global counter */
696 ksess->stream_count_global += ret;
697
698 /*
699 * Have we already sent fds to the consumer? If yes, it means
700 * that tracing is started so it is safe to send our updated
701 * stream fds.
702 */
703 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
704 struct lttng_ht_iter iter;
705 struct consumer_socket *socket;
706
707 rcu_read_lock();
708 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
709 &iter.iter, socket, node.node) {
710 pthread_mutex_lock(socket->lock);
711 ret = kernel_consumer_send_channel_stream(socket,
712 channel, ksess,
713 session->output_traces ? 1 : 0);
714 pthread_mutex_unlock(socket->lock);
715 if (ret < 0) {
716 rcu_read_unlock();
717 goto error;
718 }
719 }
720 rcu_read_unlock();
721 }
722 goto error;
723 }
724 }
725 session_unlock(session);
726 }
727 session_unlock_list();
728 return ret;
729
730 error:
731 session_unlock(session);
732 session_unlock_list();
733 return ret;
734 }
735
736 /*
737 * For each tracing session, update newly registered apps. The session list
738 * lock MUST be acquired before calling this.
739 */
740 static void update_ust_app(int app_sock)
741 {
742 struct ltt_session *sess, *stmp;
743
744 /* Consumer is in an ERROR state. Stop any application update. */
745 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
746 /* Stop the update process since the consumer is dead. */
747 return;
748 }
749
750 /* For all tracing session(s) */
751 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
752 session_lock(sess);
753 if (sess->ust_session) {
754 ust_app_global_update(sess->ust_session, app_sock);
755 }
756 session_unlock(sess);
757 }
758 }
759
760 /*
761 * This thread manage event coming from the kernel.
762 *
763 * Features supported in this thread:
764 * -) CPU Hotplug
765 */
766 static void *thread_manage_kernel(void *data)
767 {
768 int ret, i, pollfd, update_poll_flag = 1, err = -1;
769 uint32_t revents, nb_fd;
770 char tmp;
771 struct lttng_poll_event events;
772
773 DBG("[thread] Thread manage kernel started");
774
775 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
776
777 /*
778 * This first step of the while is to clean this structure which could free
779 * non NULL pointers so initialize it before the loop.
780 */
781 lttng_poll_init(&events);
782
783 if (testpoint(thread_manage_kernel)) {
784 goto error_testpoint;
785 }
786
787 health_code_update();
788
789 if (testpoint(thread_manage_kernel_before_loop)) {
790 goto error_testpoint;
791 }
792
793 while (1) {
794 health_code_update();
795
796 if (update_poll_flag == 1) {
797 /* Clean events object. We are about to populate it again. */
798 lttng_poll_clean(&events);
799
800 ret = sessiond_set_thread_pollset(&events, 2);
801 if (ret < 0) {
802 goto error_poll_create;
803 }
804
805 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
806 if (ret < 0) {
807 goto error;
808 }
809
810 /* This will add the available kernel channel if any. */
811 ret = update_kernel_poll(&events);
812 if (ret < 0) {
813 goto error;
814 }
815 update_poll_flag = 0;
816 }
817
818 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
819
820 /* Poll infinite value of time */
821 restart:
822 health_poll_entry();
823 ret = lttng_poll_wait(&events, -1);
824 health_poll_exit();
825 if (ret < 0) {
826 /*
827 * Restart interrupted system call.
828 */
829 if (errno == EINTR) {
830 goto restart;
831 }
832 goto error;
833 } else if (ret == 0) {
834 /* Should not happen since timeout is infinite */
835 ERR("Return value of poll is 0 with an infinite timeout.\n"
836 "This should not have happened! Continuing...");
837 continue;
838 }
839
840 nb_fd = ret;
841
842 for (i = 0; i < nb_fd; i++) {
843 /* Fetch once the poll data */
844 revents = LTTNG_POLL_GETEV(&events, i);
845 pollfd = LTTNG_POLL_GETFD(&events, i);
846
847 health_code_update();
848
849 /* Thread quit pipe has been closed. Killing thread. */
850 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
851 if (ret) {
852 err = 0;
853 goto exit;
854 }
855
856 /* Check for data on kernel pipe */
857 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
858 do {
859 ret = read(kernel_poll_pipe[0], &tmp, 1);
860 } while (ret < 0 && errno == EINTR);
861 /*
862 * Ret value is useless here, if this pipe gets any actions an
863 * update is required anyway.
864 */
865 update_poll_flag = 1;
866 continue;
867 } else {
868 /*
869 * New CPU detected by the kernel. Adding kernel stream to
870 * kernel session and updating the kernel consumer
871 */
872 if (revents & LPOLLIN) {
873 ret = update_kernel_stream(&kconsumer_data, pollfd);
874 if (ret < 0) {
875 continue;
876 }
877 break;
878 /*
879 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
880 * and unregister kernel stream at this point.
881 */
882 }
883 }
884 }
885 }
886
887 exit:
888 error:
889 lttng_poll_clean(&events);
890 error_poll_create:
891 error_testpoint:
892 utils_close_pipe(kernel_poll_pipe);
893 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
894 if (err) {
895 health_error();
896 ERR("Health error occurred in %s", __func__);
897 WARN("Kernel thread died unexpectedly. "
898 "Kernel tracing can continue but CPU hotplug is disabled.");
899 }
900 health_unregister(health_sessiond);
901 DBG("Kernel thread dying");
902 return NULL;
903 }
904
905 /*
906 * Signal pthread condition of the consumer data that the thread.
907 */
908 static void signal_consumer_condition(struct consumer_data *data, int state)
909 {
910 pthread_mutex_lock(&data->cond_mutex);
911
912 /*
913 * The state is set before signaling. It can be any value, it's the waiter
914 * job to correctly interpret this condition variable associated to the
915 * consumer pthread_cond.
916 *
917 * A value of 0 means that the corresponding thread of the consumer data
918 * was not started. 1 indicates that the thread has started and is ready
919 * for action. A negative value means that there was an error during the
920 * thread bootstrap.
921 */
922 data->consumer_thread_is_ready = state;
923 (void) pthread_cond_signal(&data->cond);
924
925 pthread_mutex_unlock(&data->cond_mutex);
926 }
927
928 /*
929 * This thread manage the consumer error sent back to the session daemon.
930 */
931 static void *thread_manage_consumer(void *data)
932 {
933 int sock = -1, i, ret, pollfd, err = -1;
934 uint32_t revents, nb_fd;
935 enum lttcomm_return_code code;
936 struct lttng_poll_event events;
937 struct consumer_data *consumer_data = data;
938
939 DBG("[thread] Manage consumer started");
940
941 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
942
943 health_code_update();
944
945 /*
946 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
947 * metadata_sock. Nothing more will be added to this poll set.
948 */
949 ret = sessiond_set_thread_pollset(&events, 3);
950 if (ret < 0) {
951 goto error_poll;
952 }
953
954 /*
955 * The error socket here is already in a listening state which was done
956 * just before spawning this thread to avoid a race between the consumer
957 * daemon exec trying to connect and the listen() call.
958 */
959 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
960 if (ret < 0) {
961 goto error;
962 }
963
964 health_code_update();
965
966 /* Infinite blocking call, waiting for transmission */
967 restart:
968 health_poll_entry();
969
970 if (testpoint(thread_manage_consumer)) {
971 goto error;
972 }
973
974 ret = lttng_poll_wait(&events, -1);
975 health_poll_exit();
976 if (ret < 0) {
977 /*
978 * Restart interrupted system call.
979 */
980 if (errno == EINTR) {
981 goto restart;
982 }
983 goto error;
984 }
985
986 nb_fd = ret;
987
988 for (i = 0; i < nb_fd; i++) {
989 /* Fetch once the poll data */
990 revents = LTTNG_POLL_GETEV(&events, i);
991 pollfd = LTTNG_POLL_GETFD(&events, i);
992
993 health_code_update();
994
995 /* Thread quit pipe has been closed. Killing thread. */
996 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
997 if (ret) {
998 err = 0;
999 goto exit;
1000 }
1001
1002 /* Event on the registration socket */
1003 if (pollfd == consumer_data->err_sock) {
1004 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1005 ERR("consumer err socket poll error");
1006 goto error;
1007 }
1008 }
1009 }
1010
1011 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1012 if (sock < 0) {
1013 goto error;
1014 }
1015
1016 /*
1017 * Set the CLOEXEC flag. Return code is useless because either way, the
1018 * show must go on.
1019 */
1020 (void) utils_set_fd_cloexec(sock);
1021
1022 health_code_update();
1023
1024 DBG2("Receiving code from consumer err_sock");
1025
1026 /* Getting status code from kconsumerd */
1027 ret = lttcomm_recv_unix_sock(sock, &code,
1028 sizeof(enum lttcomm_return_code));
1029 if (ret <= 0) {
1030 goto error;
1031 }
1032
1033 health_code_update();
1034
1035 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1036 /* Connect both socket, command and metadata. */
1037 consumer_data->cmd_sock =
1038 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1039 consumer_data->metadata_fd =
1040 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1041 if (consumer_data->cmd_sock < 0
1042 || consumer_data->metadata_fd < 0) {
1043 PERROR("consumer connect cmd socket");
1044 /* On error, signal condition and quit. */
1045 signal_consumer_condition(consumer_data, -1);
1046 goto error;
1047 }
1048 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1049 /* Create metadata socket lock. */
1050 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1051 if (consumer_data->metadata_sock.lock == NULL) {
1052 PERROR("zmalloc pthread mutex");
1053 ret = -1;
1054 goto error;
1055 }
1056 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1057
1058 signal_consumer_condition(consumer_data, 1);
1059 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1060 DBG("Consumer metadata socket ready (fd: %d)",
1061 consumer_data->metadata_fd);
1062 } else {
1063 ERR("consumer error when waiting for SOCK_READY : %s",
1064 lttcomm_get_readable_code(-code));
1065 goto error;
1066 }
1067
1068 /* Remove the consumerd error sock since we've established a connexion */
1069 ret = lttng_poll_del(&events, consumer_data->err_sock);
1070 if (ret < 0) {
1071 goto error;
1072 }
1073
1074 /* Add new accepted error socket. */
1075 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1076 if (ret < 0) {
1077 goto error;
1078 }
1079
1080 /* Add metadata socket that is successfully connected. */
1081 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1082 LPOLLIN | LPOLLRDHUP);
1083 if (ret < 0) {
1084 goto error;
1085 }
1086
1087 health_code_update();
1088
1089 /* Infinite blocking call, waiting for transmission */
1090 restart_poll:
1091 while (1) {
1092 health_poll_entry();
1093 ret = lttng_poll_wait(&events, -1);
1094 health_poll_exit();
1095 if (ret < 0) {
1096 /*
1097 * Restart interrupted system call.
1098 */
1099 if (errno == EINTR) {
1100 goto restart_poll;
1101 }
1102 goto error;
1103 }
1104
1105 nb_fd = ret;
1106
1107 for (i = 0; i < nb_fd; i++) {
1108 /* Fetch once the poll data */
1109 revents = LTTNG_POLL_GETEV(&events, i);
1110 pollfd = LTTNG_POLL_GETFD(&events, i);
1111
1112 health_code_update();
1113
1114 /* Thread quit pipe has been closed. Killing thread. */
1115 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1116 if (ret) {
1117 err = 0;
1118 goto exit;
1119 }
1120
1121 if (pollfd == sock) {
1122 /* Event on the consumerd socket */
1123 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1124 ERR("consumer err socket second poll error");
1125 goto error;
1126 }
1127 health_code_update();
1128 /* Wait for any kconsumerd error */
1129 ret = lttcomm_recv_unix_sock(sock, &code,
1130 sizeof(enum lttcomm_return_code));
1131 if (ret <= 0) {
1132 ERR("consumer closed the command socket");
1133 goto error;
1134 }
1135
1136 ERR("consumer return code : %s",
1137 lttcomm_get_readable_code(-code));
1138
1139 goto exit;
1140 } else if (pollfd == consumer_data->metadata_fd) {
1141 /* UST metadata requests */
1142 ret = ust_consumer_metadata_request(
1143 &consumer_data->metadata_sock);
1144 if (ret < 0) {
1145 ERR("Handling metadata request");
1146 goto error;
1147 }
1148 break;
1149 } else {
1150 ERR("Unknown pollfd");
1151 goto error;
1152 }
1153 }
1154 health_code_update();
1155 }
1156
1157 exit:
1158 error:
1159 /*
1160 * We lock here because we are about to close the sockets and some other
1161 * thread might be using them so get exclusive access which will abort all
1162 * other consumer command by other threads.
1163 */
1164 pthread_mutex_lock(&consumer_data->lock);
1165
1166 /* Immediately set the consumerd state to stopped */
1167 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1168 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1169 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1170 consumer_data->type == LTTNG_CONSUMER32_UST) {
1171 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1172 } else {
1173 /* Code flow error... */
1174 assert(0);
1175 }
1176
1177 if (consumer_data->err_sock >= 0) {
1178 ret = close(consumer_data->err_sock);
1179 if (ret) {
1180 PERROR("close");
1181 }
1182 consumer_data->err_sock = -1;
1183 }
1184 if (consumer_data->cmd_sock >= 0) {
1185 ret = close(consumer_data->cmd_sock);
1186 if (ret) {
1187 PERROR("close");
1188 }
1189 consumer_data->cmd_sock = -1;
1190 }
1191 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1192 ret = close(*consumer_data->metadata_sock.fd_ptr);
1193 if (ret) {
1194 PERROR("close");
1195 }
1196 }
1197
1198 if (sock >= 0) {
1199 ret = close(sock);
1200 if (ret) {
1201 PERROR("close");
1202 }
1203 }
1204
1205 unlink(consumer_data->err_unix_sock_path);
1206 unlink(consumer_data->cmd_unix_sock_path);
1207 consumer_data->pid = 0;
1208 pthread_mutex_unlock(&consumer_data->lock);
1209
1210 /* Cleanup metadata socket mutex. */
1211 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1212 free(consumer_data->metadata_sock.lock);
1213
1214 lttng_poll_clean(&events);
1215 error_poll:
1216 if (err) {
1217 health_error();
1218 ERR("Health error occurred in %s", __func__);
1219 }
1220 health_unregister(health_sessiond);
1221 DBG("consumer thread cleanup completed");
1222
1223 return NULL;
1224 }
1225
1226 /*
1227 * This thread manage application communication.
1228 */
1229 static void *thread_manage_apps(void *data)
1230 {
1231 int i, ret, pollfd, err = -1;
1232 uint32_t revents, nb_fd;
1233 struct lttng_poll_event events;
1234
1235 DBG("[thread] Manage application started");
1236
1237 rcu_register_thread();
1238 rcu_thread_online();
1239
1240 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1241
1242 if (testpoint(thread_manage_apps)) {
1243 goto error_testpoint;
1244 }
1245
1246 health_code_update();
1247
1248 ret = sessiond_set_thread_pollset(&events, 2);
1249 if (ret < 0) {
1250 goto error_poll_create;
1251 }
1252
1253 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1254 if (ret < 0) {
1255 goto error;
1256 }
1257
1258 if (testpoint(thread_manage_apps_before_loop)) {
1259 goto error;
1260 }
1261
1262 health_code_update();
1263
1264 while (1) {
1265 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1266
1267 /* Inifinite blocking call, waiting for transmission */
1268 restart:
1269 health_poll_entry();
1270 ret = lttng_poll_wait(&events, -1);
1271 health_poll_exit();
1272 if (ret < 0) {
1273 /*
1274 * Restart interrupted system call.
1275 */
1276 if (errno == EINTR) {
1277 goto restart;
1278 }
1279 goto error;
1280 }
1281
1282 nb_fd = ret;
1283
1284 for (i = 0; i < nb_fd; i++) {
1285 /* Fetch once the poll data */
1286 revents = LTTNG_POLL_GETEV(&events, i);
1287 pollfd = LTTNG_POLL_GETFD(&events, i);
1288
1289 health_code_update();
1290
1291 /* Thread quit pipe has been closed. Killing thread. */
1292 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1293 if (ret) {
1294 err = 0;
1295 goto exit;
1296 }
1297
1298 /* Inspect the apps cmd pipe */
1299 if (pollfd == apps_cmd_pipe[0]) {
1300 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1301 ERR("Apps command pipe error");
1302 goto error;
1303 } else if (revents & LPOLLIN) {
1304 int sock;
1305
1306 /* Empty pipe */
1307 do {
1308 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
1309 } while (ret < 0 && errno == EINTR);
1310 if (ret < 0 || ret < sizeof(sock)) {
1311 PERROR("read apps cmd pipe");
1312 goto error;
1313 }
1314
1315 health_code_update();
1316
1317 /*
1318 * We only monitor the error events of the socket. This
1319 * thread does not handle any incoming data from UST
1320 * (POLLIN).
1321 */
1322 ret = lttng_poll_add(&events, sock,
1323 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1324 if (ret < 0) {
1325 goto error;
1326 }
1327
1328 DBG("Apps with sock %d added to poll set", sock);
1329
1330 health_code_update();
1331
1332 break;
1333 }
1334 } else {
1335 /*
1336 * At this point, we know that a registered application made
1337 * the event at poll_wait.
1338 */
1339 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1340 /* Removing from the poll set */
1341 ret = lttng_poll_del(&events, pollfd);
1342 if (ret < 0) {
1343 goto error;
1344 }
1345
1346 /* Socket closed on remote end. */
1347 ust_app_unregister(pollfd);
1348 break;
1349 }
1350 }
1351
1352 health_code_update();
1353 }
1354 }
1355
1356 exit:
1357 error:
1358 lttng_poll_clean(&events);
1359 error_poll_create:
1360 error_testpoint:
1361 utils_close_pipe(apps_cmd_pipe);
1362 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1363
1364 /*
1365 * We don't clean the UST app hash table here since already registered
1366 * applications can still be controlled so let them be until the session
1367 * daemon dies or the applications stop.
1368 */
1369
1370 if (err) {
1371 health_error();
1372 ERR("Health error occurred in %s", __func__);
1373 }
1374 health_unregister(health_sessiond);
1375 DBG("Application communication apps thread cleanup complete");
1376 rcu_thread_offline();
1377 rcu_unregister_thread();
1378 return NULL;
1379 }
1380
1381 /*
1382 * Send a socket to a thread This is called from the dispatch UST registration
1383 * thread once all sockets are set for the application.
1384 *
1385 * The sock value can be invalid, we don't really care, the thread will handle
1386 * it and make the necessary cleanup if so.
1387 *
1388 * On success, return 0 else a negative value being the errno message of the
1389 * write().
1390 */
1391 static int send_socket_to_thread(int fd, int sock)
1392 {
1393 int ret;
1394
1395 /*
1396 * It's possible that the FD is set as invalid with -1 concurrently just
1397 * before calling this function being a shutdown state of the thread.
1398 */
1399 if (fd < 0) {
1400 ret = -EBADF;
1401 goto error;
1402 }
1403
1404 do {
1405 ret = write(fd, &sock, sizeof(sock));
1406 } while (ret < 0 && errno == EINTR);
1407 if (ret < 0 || ret != sizeof(sock)) {
1408 PERROR("write apps pipe %d", fd);
1409 if (ret < 0) {
1410 ret = -errno;
1411 }
1412 goto error;
1413 }
1414
1415 /* All good. Don't send back the write positive ret value. */
1416 ret = 0;
1417 error:
1418 return ret;
1419 }
1420
1421 /*
1422 * Sanitize the wait queue of the dispatch registration thread meaning removing
1423 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1424 * notify socket is never received.
1425 */
1426 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1427 {
1428 int ret, nb_fd = 0, i;
1429 unsigned int fd_added = 0;
1430 struct lttng_poll_event events;
1431 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1432
1433 assert(wait_queue);
1434
1435 lttng_poll_init(&events);
1436
1437 /* Just skip everything for an empty queue. */
1438 if (!wait_queue->count) {
1439 goto end;
1440 }
1441
1442 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1443 if (ret < 0) {
1444 goto error_create;
1445 }
1446
1447 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1448 &wait_queue->head, head) {
1449 assert(wait_node->app);
1450 ret = lttng_poll_add(&events, wait_node->app->sock,
1451 LPOLLHUP | LPOLLERR);
1452 if (ret < 0) {
1453 goto error;
1454 }
1455
1456 fd_added = 1;
1457 }
1458
1459 if (!fd_added) {
1460 goto end;
1461 }
1462
1463 /*
1464 * Poll but don't block so we can quickly identify the faulty events and
1465 * clean them afterwards from the wait queue.
1466 */
1467 ret = lttng_poll_wait(&events, 0);
1468 if (ret < 0) {
1469 goto error;
1470 }
1471 nb_fd = ret;
1472
1473 for (i = 0; i < nb_fd; i++) {
1474 /* Get faulty FD. */
1475 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1476 int pollfd = LTTNG_POLL_GETFD(&events, i);
1477
1478 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1479 &wait_queue->head, head) {
1480 if (pollfd == wait_node->app->sock &&
1481 (revents & (LPOLLHUP | LPOLLERR))) {
1482 cds_list_del(&wait_node->head);
1483 wait_queue->count--;
1484 ust_app_destroy(wait_node->app);
1485 free(wait_node);
1486 break;
1487 }
1488 }
1489 }
1490
1491 if (nb_fd > 0) {
1492 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1493 }
1494
1495 end:
1496 lttng_poll_clean(&events);
1497 return;
1498
1499 error:
1500 lttng_poll_clean(&events);
1501 error_create:
1502 ERR("Unable to sanitize wait queue");
1503 return;
1504 }
1505
1506 /*
1507 * Dispatch request from the registration threads to the application
1508 * communication thread.
1509 */
1510 static void *thread_dispatch_ust_registration(void *data)
1511 {
1512 int ret, err = -1;
1513 struct cds_wfq_node *node;
1514 struct ust_command *ust_cmd = NULL;
1515 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1516 struct ust_reg_wait_queue wait_queue = {
1517 .count = 0,
1518 };
1519
1520 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1521
1522 health_code_update();
1523
1524 CDS_INIT_LIST_HEAD(&wait_queue.head);
1525
1526 DBG("[thread] Dispatch UST command started");
1527
1528 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1529 health_code_update();
1530
1531 /* Atomically prepare the queue futex */
1532 futex_nto1_prepare(&ust_cmd_queue.futex);
1533
1534 do {
1535 struct ust_app *app = NULL;
1536 ust_cmd = NULL;
1537
1538 /*
1539 * Make sure we don't have node(s) that have hung up before receiving
1540 * the notify socket. This is to clean the list in order to avoid
1541 * memory leaks from notify socket that are never seen.
1542 */
1543 sanitize_wait_queue(&wait_queue);
1544
1545 health_code_update();
1546 /* Dequeue command for registration */
1547 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1548 if (node == NULL) {
1549 DBG("Woken up but nothing in the UST command queue");
1550 /* Continue thread execution */
1551 break;
1552 }
1553
1554 ust_cmd = caa_container_of(node, struct ust_command, node);
1555
1556 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1557 " gid:%d sock:%d name:%s (version %d.%d)",
1558 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1559 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1560 ust_cmd->sock, ust_cmd->reg_msg.name,
1561 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1562
1563 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1564 wait_node = zmalloc(sizeof(*wait_node));
1565 if (!wait_node) {
1566 PERROR("zmalloc wait_node dispatch");
1567 ret = close(ust_cmd->sock);
1568 if (ret < 0) {
1569 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1570 }
1571 lttng_fd_put(1, LTTNG_FD_APPS);
1572 free(ust_cmd);
1573 goto error;
1574 }
1575 CDS_INIT_LIST_HEAD(&wait_node->head);
1576
1577 /* Create application object if socket is CMD. */
1578 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1579 ust_cmd->sock);
1580 if (!wait_node->app) {
1581 ret = close(ust_cmd->sock);
1582 if (ret < 0) {
1583 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1584 }
1585 lttng_fd_put(1, LTTNG_FD_APPS);
1586 free(wait_node);
1587 free(ust_cmd);
1588 continue;
1589 }
1590 /*
1591 * Add application to the wait queue so we can set the notify
1592 * socket before putting this object in the global ht.
1593 */
1594 cds_list_add(&wait_node->head, &wait_queue.head);
1595 wait_queue.count++;
1596
1597 free(ust_cmd);
1598 /*
1599 * We have to continue here since we don't have the notify
1600 * socket and the application MUST be added to the hash table
1601 * only at that moment.
1602 */
1603 continue;
1604 } else {
1605 /*
1606 * Look for the application in the local wait queue and set the
1607 * notify socket if found.
1608 */
1609 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1610 &wait_queue.head, head) {
1611 health_code_update();
1612 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1613 wait_node->app->notify_sock = ust_cmd->sock;
1614 cds_list_del(&wait_node->head);
1615 wait_queue.count--;
1616 app = wait_node->app;
1617 free(wait_node);
1618 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1619 break;
1620 }
1621 }
1622
1623 /*
1624 * With no application at this stage the received socket is
1625 * basically useless so close it before we free the cmd data
1626 * structure for good.
1627 */
1628 if (!app) {
1629 ret = close(ust_cmd->sock);
1630 if (ret < 0) {
1631 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1632 }
1633 lttng_fd_put(1, LTTNG_FD_APPS);
1634 }
1635 free(ust_cmd);
1636 }
1637
1638 if (app) {
1639 /*
1640 * @session_lock_list
1641 *
1642 * Lock the global session list so from the register up to the
1643 * registration done message, no thread can see the application
1644 * and change its state.
1645 */
1646 session_lock_list();
1647 rcu_read_lock();
1648
1649 /*
1650 * Add application to the global hash table. This needs to be
1651 * done before the update to the UST registry can locate the
1652 * application.
1653 */
1654 ust_app_add(app);
1655
1656 /* Set app version. This call will print an error if needed. */
1657 (void) ust_app_version(app);
1658
1659 /* Send notify socket through the notify pipe. */
1660 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1661 app->notify_sock);
1662 if (ret < 0) {
1663 rcu_read_unlock();
1664 session_unlock_list();
1665 /*
1666 * No notify thread, stop the UST tracing. However, this is
1667 * not an internal error of the this thread thus setting
1668 * the health error code to a normal exit.
1669 */
1670 err = 0;
1671 goto error;
1672 }
1673
1674 /*
1675 * Update newly registered application with the tracing
1676 * registry info already enabled information.
1677 */
1678 update_ust_app(app->sock);
1679
1680 /*
1681 * Don't care about return value. Let the manage apps threads
1682 * handle app unregistration upon socket close.
1683 */
1684 (void) ust_app_register_done(app->sock);
1685
1686 /*
1687 * Even if the application socket has been closed, send the app
1688 * to the thread and unregistration will take place at that
1689 * place.
1690 */
1691 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1692 if (ret < 0) {
1693 rcu_read_unlock();
1694 session_unlock_list();
1695 /*
1696 * No apps. thread, stop the UST tracing. However, this is
1697 * not an internal error of the this thread thus setting
1698 * the health error code to a normal exit.
1699 */
1700 err = 0;
1701 goto error;
1702 }
1703
1704 rcu_read_unlock();
1705 session_unlock_list();
1706 }
1707 } while (node != NULL);
1708
1709 health_poll_entry();
1710 /* Futex wait on queue. Blocking call on futex() */
1711 futex_nto1_wait(&ust_cmd_queue.futex);
1712 health_poll_exit();
1713 }
1714 /* Normal exit, no error */
1715 err = 0;
1716
1717 error:
1718 /* Clean up wait queue. */
1719 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1720 &wait_queue.head, head) {
1721 cds_list_del(&wait_node->head);
1722 wait_queue.count--;
1723 free(wait_node);
1724 }
1725
1726 DBG("Dispatch thread dying");
1727 if (err) {
1728 health_error();
1729 ERR("Health error occurred in %s", __func__);
1730 }
1731 health_unregister(health_sessiond);
1732 return NULL;
1733 }
1734
1735 /*
1736 * This thread manage application registration.
1737 */
1738 static void *thread_registration_apps(void *data)
1739 {
1740 int sock = -1, i, ret, pollfd, err = -1;
1741 uint32_t revents, nb_fd;
1742 struct lttng_poll_event events;
1743 /*
1744 * Get allocated in this thread, enqueued to a global queue, dequeued and
1745 * freed in the manage apps thread.
1746 */
1747 struct ust_command *ust_cmd = NULL;
1748
1749 DBG("[thread] Manage application registration started");
1750
1751 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1752
1753 if (testpoint(thread_registration_apps)) {
1754 goto error_testpoint;
1755 }
1756
1757 ret = lttcomm_listen_unix_sock(apps_sock);
1758 if (ret < 0) {
1759 goto error_listen;
1760 }
1761
1762 /*
1763 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1764 * more will be added to this poll set.
1765 */
1766 ret = sessiond_set_thread_pollset(&events, 2);
1767 if (ret < 0) {
1768 goto error_create_poll;
1769 }
1770
1771 /* Add the application registration socket */
1772 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1773 if (ret < 0) {
1774 goto error_poll_add;
1775 }
1776
1777 /* Notify all applications to register */
1778 ret = notify_ust_apps(1);
1779 if (ret < 0) {
1780 ERR("Failed to notify applications or create the wait shared memory.\n"
1781 "Execution continues but there might be problem for already\n"
1782 "running applications that wishes to register.");
1783 }
1784
1785 while (1) {
1786 DBG("Accepting application registration");
1787
1788 /* Inifinite blocking call, waiting for transmission */
1789 restart:
1790 health_poll_entry();
1791 ret = lttng_poll_wait(&events, -1);
1792 health_poll_exit();
1793 if (ret < 0) {
1794 /*
1795 * Restart interrupted system call.
1796 */
1797 if (errno == EINTR) {
1798 goto restart;
1799 }
1800 goto error;
1801 }
1802
1803 nb_fd = ret;
1804
1805 for (i = 0; i < nb_fd; i++) {
1806 health_code_update();
1807
1808 /* Fetch once the poll data */
1809 revents = LTTNG_POLL_GETEV(&events, i);
1810 pollfd = LTTNG_POLL_GETFD(&events, i);
1811
1812 /* Thread quit pipe has been closed. Killing thread. */
1813 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1814 if (ret) {
1815 err = 0;
1816 goto exit;
1817 }
1818
1819 /* Event on the registration socket */
1820 if (pollfd == apps_sock) {
1821 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1822 ERR("Register apps socket poll error");
1823 goto error;
1824 } else if (revents & LPOLLIN) {
1825 sock = lttcomm_accept_unix_sock(apps_sock);
1826 if (sock < 0) {
1827 goto error;
1828 }
1829
1830 /*
1831 * Set socket timeout for both receiving and ending.
1832 * app_socket_timeout is in seconds, whereas
1833 * lttcomm_setsockopt_rcv_timeout and
1834 * lttcomm_setsockopt_snd_timeout expect msec as
1835 * parameter.
1836 */
1837 (void) lttcomm_setsockopt_rcv_timeout(sock,
1838 app_socket_timeout * 1000);
1839 (void) lttcomm_setsockopt_snd_timeout(sock,
1840 app_socket_timeout * 1000);
1841
1842 /*
1843 * Set the CLOEXEC flag. Return code is useless because
1844 * either way, the show must go on.
1845 */
1846 (void) utils_set_fd_cloexec(sock);
1847
1848 /* Create UST registration command for enqueuing */
1849 ust_cmd = zmalloc(sizeof(struct ust_command));
1850 if (ust_cmd == NULL) {
1851 PERROR("ust command zmalloc");
1852 goto error;
1853 }
1854
1855 /*
1856 * Using message-based transmissions to ensure we don't
1857 * have to deal with partially received messages.
1858 */
1859 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1860 if (ret < 0) {
1861 ERR("Exhausted file descriptors allowed for applications.");
1862 free(ust_cmd);
1863 ret = close(sock);
1864 if (ret) {
1865 PERROR("close");
1866 }
1867 sock = -1;
1868 continue;
1869 }
1870
1871 health_code_update();
1872 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1873 if (ret < 0) {
1874 free(ust_cmd);
1875 /* Close socket of the application. */
1876 ret = close(sock);
1877 if (ret) {
1878 PERROR("close");
1879 }
1880 lttng_fd_put(LTTNG_FD_APPS, 1);
1881 sock = -1;
1882 continue;
1883 }
1884 health_code_update();
1885
1886 ust_cmd->sock = sock;
1887 sock = -1;
1888
1889 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1890 " gid:%d sock:%d name:%s (version %d.%d)",
1891 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1892 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1893 ust_cmd->sock, ust_cmd->reg_msg.name,
1894 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1895
1896 /*
1897 * Lock free enqueue the registration request. The red pill
1898 * has been taken! This apps will be part of the *system*.
1899 */
1900 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1901
1902 /*
1903 * Wake the registration queue futex. Implicit memory
1904 * barrier with the exchange in cds_wfq_enqueue.
1905 */
1906 futex_nto1_wake(&ust_cmd_queue.futex);
1907 }
1908 }
1909 }
1910 }
1911
1912 exit:
1913 error:
1914 if (err) {
1915 health_error();
1916 ERR("Health error occurred in %s", __func__);
1917 }
1918
1919 /* Notify that the registration thread is gone */
1920 notify_ust_apps(0);
1921
1922 if (apps_sock >= 0) {
1923 ret = close(apps_sock);
1924 if (ret) {
1925 PERROR("close");
1926 }
1927 }
1928 if (sock >= 0) {
1929 ret = close(sock);
1930 if (ret) {
1931 PERROR("close");
1932 }
1933 lttng_fd_put(LTTNG_FD_APPS, 1);
1934 }
1935 unlink(apps_unix_sock_path);
1936
1937 error_poll_add:
1938 lttng_poll_clean(&events);
1939 error_listen:
1940 error_create_poll:
1941 error_testpoint:
1942 DBG("UST Registration thread cleanup complete");
1943 health_unregister(health_sessiond);
1944
1945 return NULL;
1946 }
1947
1948 /*
1949 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1950 * exec or it will fails.
1951 */
1952 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1953 {
1954 int ret, clock_ret;
1955 struct timespec timeout;
1956
1957 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1958 consumer_data->consumer_thread_is_ready = 0;
1959
1960 /* Setup pthread condition */
1961 ret = pthread_condattr_init(&consumer_data->condattr);
1962 if (ret != 0) {
1963 errno = ret;
1964 PERROR("pthread_condattr_init consumer data");
1965 goto error;
1966 }
1967
1968 /*
1969 * Set the monotonic clock in order to make sure we DO NOT jump in time
1970 * between the clock_gettime() call and the timedwait call. See bug #324
1971 * for a more details and how we noticed it.
1972 */
1973 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1974 if (ret != 0) {
1975 errno = ret;
1976 PERROR("pthread_condattr_setclock consumer data");
1977 goto error;
1978 }
1979
1980 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1981 if (ret != 0) {
1982 errno = ret;
1983 PERROR("pthread_cond_init consumer data");
1984 goto error;
1985 }
1986
1987 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1988 consumer_data);
1989 if (ret != 0) {
1990 PERROR("pthread_create consumer");
1991 ret = -1;
1992 goto error;
1993 }
1994
1995 /* We are about to wait on a pthread condition */
1996 pthread_mutex_lock(&consumer_data->cond_mutex);
1997
1998 /* Get time for sem_timedwait absolute timeout */
1999 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2000 /*
2001 * Set the timeout for the condition timed wait even if the clock gettime
2002 * call fails since we might loop on that call and we want to avoid to
2003 * increment the timeout too many times.
2004 */
2005 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2006
2007 /*
2008 * The following loop COULD be skipped in some conditions so this is why we
2009 * set ret to 0 in order to make sure at least one round of the loop is
2010 * done.
2011 */
2012 ret = 0;
2013
2014 /*
2015 * Loop until the condition is reached or when a timeout is reached. Note
2016 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2017 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2018 * possible. This loop does not take any chances and works with both of
2019 * them.
2020 */
2021 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2022 if (clock_ret < 0) {
2023 PERROR("clock_gettime spawn consumer");
2024 /* Infinite wait for the consumerd thread to be ready */
2025 ret = pthread_cond_wait(&consumer_data->cond,
2026 &consumer_data->cond_mutex);
2027 } else {
2028 ret = pthread_cond_timedwait(&consumer_data->cond,
2029 &consumer_data->cond_mutex, &timeout);
2030 }
2031 }
2032
2033 /* Release the pthread condition */
2034 pthread_mutex_unlock(&consumer_data->cond_mutex);
2035
2036 if (ret != 0) {
2037 errno = ret;
2038 if (ret == ETIMEDOUT) {
2039 /*
2040 * Call has timed out so we kill the kconsumerd_thread and return
2041 * an error.
2042 */
2043 ERR("Condition timed out. The consumer thread was never ready."
2044 " Killing it");
2045 ret = pthread_cancel(consumer_data->thread);
2046 if (ret < 0) {
2047 PERROR("pthread_cancel consumer thread");
2048 }
2049 } else {
2050 PERROR("pthread_cond_wait failed consumer thread");
2051 }
2052 goto error;
2053 }
2054
2055 pthread_mutex_lock(&consumer_data->pid_mutex);
2056 if (consumer_data->pid == 0) {
2057 ERR("Consumerd did not start");
2058 pthread_mutex_unlock(&consumer_data->pid_mutex);
2059 goto error;
2060 }
2061 pthread_mutex_unlock(&consumer_data->pid_mutex);
2062
2063 return 0;
2064
2065 error:
2066 return ret;
2067 }
2068
2069 /*
2070 * Join consumer thread
2071 */
2072 static int join_consumer_thread(struct consumer_data *consumer_data)
2073 {
2074 void *status;
2075
2076 /* Consumer pid must be a real one. */
2077 if (consumer_data->pid > 0) {
2078 int ret;
2079 ret = kill(consumer_data->pid, SIGTERM);
2080 if (ret) {
2081 ERR("Error killing consumer daemon");
2082 return ret;
2083 }
2084 return pthread_join(consumer_data->thread, &status);
2085 } else {
2086 return 0;
2087 }
2088 }
2089
2090 /*
2091 * Fork and exec a consumer daemon (consumerd).
2092 *
2093 * Return pid if successful else -1.
2094 */
2095 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2096 {
2097 int ret;
2098 pid_t pid;
2099 const char *consumer_to_use;
2100 const char *verbosity;
2101 struct stat st;
2102
2103 DBG("Spawning consumerd");
2104
2105 pid = fork();
2106 if (pid == 0) {
2107 /*
2108 * Exec consumerd.
2109 */
2110 if (opt_verbose_consumer) {
2111 verbosity = "--verbose";
2112 } else {
2113 verbosity = "--quiet";
2114 }
2115 switch (consumer_data->type) {
2116 case LTTNG_CONSUMER_KERNEL:
2117 /*
2118 * Find out which consumerd to execute. We will first try the
2119 * 64-bit path, then the sessiond's installation directory, and
2120 * fallback on the 32-bit one,
2121 */
2122 DBG3("Looking for a kernel consumer at these locations:");
2123 DBG3(" 1) %s", consumerd64_bin);
2124 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2125 DBG3(" 3) %s", consumerd32_bin);
2126 if (stat(consumerd64_bin, &st) == 0) {
2127 DBG3("Found location #1");
2128 consumer_to_use = consumerd64_bin;
2129 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2130 DBG3("Found location #2");
2131 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2132 } else if (stat(consumerd32_bin, &st) == 0) {
2133 DBG3("Found location #3");
2134 consumer_to_use = consumerd32_bin;
2135 } else {
2136 DBG("Could not find any valid consumerd executable");
2137 break;
2138 }
2139 DBG("Using kernel consumer at: %s", consumer_to_use);
2140 execl(consumer_to_use,
2141 "lttng-consumerd", verbosity, "-k",
2142 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2143 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2144 "--group", tracing_group_name,
2145 NULL);
2146 break;
2147 case LTTNG_CONSUMER64_UST:
2148 {
2149 char *tmpnew = NULL;
2150
2151 if (consumerd64_libdir[0] != '\0') {
2152 char *tmp;
2153 size_t tmplen;
2154
2155 tmp = getenv("LD_LIBRARY_PATH");
2156 if (!tmp) {
2157 tmp = "";
2158 }
2159 tmplen = strlen("LD_LIBRARY_PATH=")
2160 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2161 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2162 if (!tmpnew) {
2163 ret = -ENOMEM;
2164 goto error;
2165 }
2166 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2167 strcat(tmpnew, consumerd64_libdir);
2168 if (tmp[0] != '\0') {
2169 strcat(tmpnew, ":");
2170 strcat(tmpnew, tmp);
2171 }
2172 ret = putenv(tmpnew);
2173 if (ret) {
2174 ret = -errno;
2175 free(tmpnew);
2176 goto error;
2177 }
2178 }
2179 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2180 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2181 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2182 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2183 "--group", tracing_group_name,
2184 NULL);
2185 if (consumerd64_libdir[0] != '\0') {
2186 free(tmpnew);
2187 }
2188 if (ret) {
2189 goto error;
2190 }
2191 break;
2192 }
2193 case LTTNG_CONSUMER32_UST:
2194 {
2195 char *tmpnew = NULL;
2196
2197 if (consumerd32_libdir[0] != '\0') {
2198 char *tmp;
2199 size_t tmplen;
2200
2201 tmp = getenv("LD_LIBRARY_PATH");
2202 if (!tmp) {
2203 tmp = "";
2204 }
2205 tmplen = strlen("LD_LIBRARY_PATH=")
2206 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2207 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2208 if (!tmpnew) {
2209 ret = -ENOMEM;
2210 goto error;
2211 }
2212 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2213 strcat(tmpnew, consumerd32_libdir);
2214 if (tmp[0] != '\0') {
2215 strcat(tmpnew, ":");
2216 strcat(tmpnew, tmp);
2217 }
2218 ret = putenv(tmpnew);
2219 if (ret) {
2220 ret = -errno;
2221 free(tmpnew);
2222 goto error;
2223 }
2224 }
2225 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2226 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2227 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2228 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2229 "--group", tracing_group_name,
2230 NULL);
2231 if (consumerd32_libdir[0] != '\0') {
2232 free(tmpnew);
2233 }
2234 if (ret) {
2235 goto error;
2236 }
2237 break;
2238 }
2239 default:
2240 PERROR("unknown consumer type");
2241 exit(EXIT_FAILURE);
2242 }
2243 if (errno != 0) {
2244 PERROR("kernel start consumer exec");
2245 }
2246 exit(EXIT_FAILURE);
2247 } else if (pid > 0) {
2248 ret = pid;
2249 } else {
2250 PERROR("start consumer fork");
2251 ret = -errno;
2252 }
2253 error:
2254 return ret;
2255 }
2256
2257 /*
2258 * Spawn the consumerd daemon and session daemon thread.
2259 */
2260 static int start_consumerd(struct consumer_data *consumer_data)
2261 {
2262 int ret;
2263
2264 /*
2265 * Set the listen() state on the socket since there is a possible race
2266 * between the exec() of the consumer daemon and this call if place in the
2267 * consumer thread. See bug #366 for more details.
2268 */
2269 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2270 if (ret < 0) {
2271 goto error;
2272 }
2273
2274 pthread_mutex_lock(&consumer_data->pid_mutex);
2275 if (consumer_data->pid != 0) {
2276 pthread_mutex_unlock(&consumer_data->pid_mutex);
2277 goto end;
2278 }
2279
2280 ret = spawn_consumerd(consumer_data);
2281 if (ret < 0) {
2282 ERR("Spawning consumerd failed");
2283 pthread_mutex_unlock(&consumer_data->pid_mutex);
2284 goto error;
2285 }
2286
2287 /* Setting up the consumer_data pid */
2288 consumer_data->pid = ret;
2289 DBG2("Consumer pid %d", consumer_data->pid);
2290 pthread_mutex_unlock(&consumer_data->pid_mutex);
2291
2292 DBG2("Spawning consumer control thread");
2293 ret = spawn_consumer_thread(consumer_data);
2294 if (ret < 0) {
2295 ERR("Fatal error spawning consumer control thread");
2296 goto error;
2297 }
2298
2299 end:
2300 return 0;
2301
2302 error:
2303 /* Cleanup already created sockets on error. */
2304 if (consumer_data->err_sock >= 0) {
2305 int err;
2306
2307 err = close(consumer_data->err_sock);
2308 if (err < 0) {
2309 PERROR("close consumer data error socket");
2310 }
2311 }
2312 return ret;
2313 }
2314
2315 /*
2316 * Setup necessary data for kernel tracer action.
2317 */
2318 static int init_kernel_tracer(void)
2319 {
2320 int ret;
2321
2322 /* Modprobe lttng kernel modules */
2323 ret = modprobe_lttng_control();
2324 if (ret < 0) {
2325 goto error;
2326 }
2327
2328 /* Open debugfs lttng */
2329 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2330 if (kernel_tracer_fd < 0) {
2331 DBG("Failed to open %s", module_proc_lttng);
2332 ret = -1;
2333 goto error_open;
2334 }
2335
2336 /* Validate kernel version */
2337 ret = kernel_validate_version(kernel_tracer_fd);
2338 if (ret < 0) {
2339 goto error_version;
2340 }
2341
2342 ret = modprobe_lttng_data();
2343 if (ret < 0) {
2344 goto error_modules;
2345 }
2346
2347 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2348 return 0;
2349
2350 error_version:
2351 modprobe_remove_lttng_control();
2352 ret = close(kernel_tracer_fd);
2353 if (ret) {
2354 PERROR("close");
2355 }
2356 kernel_tracer_fd = -1;
2357 return LTTNG_ERR_KERN_VERSION;
2358
2359 error_modules:
2360 ret = close(kernel_tracer_fd);
2361 if (ret) {
2362 PERROR("close");
2363 }
2364
2365 error_open:
2366 modprobe_remove_lttng_control();
2367
2368 error:
2369 WARN("No kernel tracer available");
2370 kernel_tracer_fd = -1;
2371 if (!is_root) {
2372 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2373 } else {
2374 return LTTNG_ERR_KERN_NA;
2375 }
2376 }
2377
2378
2379 /*
2380 * Copy consumer output from the tracing session to the domain session. The
2381 * function also applies the right modification on a per domain basis for the
2382 * trace files destination directory.
2383 *
2384 * Should *NOT* be called with RCU read-side lock held.
2385 */
2386 static int copy_session_consumer(int domain, struct ltt_session *session)
2387 {
2388 int ret;
2389 const char *dir_name;
2390 struct consumer_output *consumer;
2391
2392 assert(session);
2393 assert(session->consumer);
2394
2395 switch (domain) {
2396 case LTTNG_DOMAIN_KERNEL:
2397 DBG3("Copying tracing session consumer output in kernel session");
2398 /*
2399 * XXX: We should audit the session creation and what this function
2400 * does "extra" in order to avoid a destroy since this function is used
2401 * in the domain session creation (kernel and ust) only. Same for UST
2402 * domain.
2403 */
2404 if (session->kernel_session->consumer) {
2405 consumer_destroy_output(session->kernel_session->consumer);
2406 }
2407 session->kernel_session->consumer =
2408 consumer_copy_output(session->consumer);
2409 /* Ease our life a bit for the next part */
2410 consumer = session->kernel_session->consumer;
2411 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2412 break;
2413 case LTTNG_DOMAIN_UST:
2414 DBG3("Copying tracing session consumer output in UST session");
2415 if (session->ust_session->consumer) {
2416 consumer_destroy_output(session->ust_session->consumer);
2417 }
2418 session->ust_session->consumer =
2419 consumer_copy_output(session->consumer);
2420 /* Ease our life a bit for the next part */
2421 consumer = session->ust_session->consumer;
2422 dir_name = DEFAULT_UST_TRACE_DIR;
2423 break;
2424 default:
2425 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2426 goto error;
2427 }
2428
2429 /* Append correct directory to subdir */
2430 strncat(consumer->subdir, dir_name,
2431 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2432 DBG3("Copy session consumer subdir %s", consumer->subdir);
2433
2434 ret = LTTNG_OK;
2435
2436 error:
2437 return ret;
2438 }
2439
2440 /*
2441 * Create an UST session and add it to the session ust list.
2442 *
2443 * Should *NOT* be called with RCU read-side lock held.
2444 */
2445 static int create_ust_session(struct ltt_session *session,
2446 struct lttng_domain *domain)
2447 {
2448 int ret;
2449 struct ltt_ust_session *lus = NULL;
2450
2451 assert(session);
2452 assert(domain);
2453 assert(session->consumer);
2454
2455 switch (domain->type) {
2456 case LTTNG_DOMAIN_UST:
2457 break;
2458 default:
2459 ERR("Unknown UST domain on create session %d", domain->type);
2460 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2461 goto error;
2462 }
2463
2464 DBG("Creating UST session");
2465
2466 lus = trace_ust_create_session(session->id);
2467 if (lus == NULL) {
2468 ret = LTTNG_ERR_UST_SESS_FAIL;
2469 goto error;
2470 }
2471
2472 lus->uid = session->uid;
2473 lus->gid = session->gid;
2474 lus->output_traces = session->output_traces;
2475 lus->snapshot_mode = session->snapshot_mode;
2476 lus->live_timer_interval = session->live_timer;
2477 session->ust_session = lus;
2478
2479 /* Copy session output to the newly created UST session */
2480 ret = copy_session_consumer(domain->type, session);
2481 if (ret != LTTNG_OK) {
2482 goto error;
2483 }
2484
2485 return LTTNG_OK;
2486
2487 error:
2488 free(lus);
2489 session->ust_session = NULL;
2490 return ret;
2491 }
2492
2493 /*
2494 * Create a kernel tracer session then create the default channel.
2495 */
2496 static int create_kernel_session(struct ltt_session *session)
2497 {
2498 int ret;
2499
2500 DBG("Creating kernel session");
2501
2502 ret = kernel_create_session(session, kernel_tracer_fd);
2503 if (ret < 0) {
2504 ret = LTTNG_ERR_KERN_SESS_FAIL;
2505 goto error;
2506 }
2507
2508 /* Code flow safety */
2509 assert(session->kernel_session);
2510
2511 /* Copy session output to the newly created Kernel session */
2512 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2513 if (ret != LTTNG_OK) {
2514 goto error;
2515 }
2516
2517 /* Create directory(ies) on local filesystem. */
2518 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2519 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2520 ret = run_as_mkdir_recursive(
2521 session->kernel_session->consumer->dst.trace_path,
2522 S_IRWXU | S_IRWXG, session->uid, session->gid);
2523 if (ret < 0) {
2524 if (ret != -EEXIST) {
2525 ERR("Trace directory creation error");
2526 goto error;
2527 }
2528 }
2529 }
2530
2531 session->kernel_session->uid = session->uid;
2532 session->kernel_session->gid = session->gid;
2533 session->kernel_session->output_traces = session->output_traces;
2534 session->kernel_session->snapshot_mode = session->snapshot_mode;
2535
2536 return LTTNG_OK;
2537
2538 error:
2539 trace_kernel_destroy_session(session->kernel_session);
2540 session->kernel_session = NULL;
2541 return ret;
2542 }
2543
2544 /*
2545 * Count number of session permitted by uid/gid.
2546 */
2547 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2548 {
2549 unsigned int i = 0;
2550 struct ltt_session *session;
2551
2552 DBG("Counting number of available session for UID %d GID %d",
2553 uid, gid);
2554 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2555 /*
2556 * Only list the sessions the user can control.
2557 */
2558 if (!session_access_ok(session, uid, gid)) {
2559 continue;
2560 }
2561 i++;
2562 }
2563 return i;
2564 }
2565
2566 /*
2567 * Process the command requested by the lttng client within the command
2568 * context structure. This function make sure that the return structure (llm)
2569 * is set and ready for transmission before returning.
2570 *
2571 * Return any error encountered or 0 for success.
2572 *
2573 * "sock" is only used for special-case var. len data.
2574 *
2575 * Should *NOT* be called with RCU read-side lock held.
2576 */
2577 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2578 int *sock_error)
2579 {
2580 int ret = LTTNG_OK;
2581 int need_tracing_session = 1;
2582 int need_domain;
2583
2584 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2585
2586 *sock_error = 0;
2587
2588 switch (cmd_ctx->lsm->cmd_type) {
2589 case LTTNG_CREATE_SESSION:
2590 case LTTNG_CREATE_SESSION_SNAPSHOT:
2591 case LTTNG_CREATE_SESSION_LIVE:
2592 case LTTNG_DESTROY_SESSION:
2593 case LTTNG_LIST_SESSIONS:
2594 case LTTNG_LIST_DOMAINS:
2595 case LTTNG_START_TRACE:
2596 case LTTNG_STOP_TRACE:
2597 case LTTNG_DATA_PENDING:
2598 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2599 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2600 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2601 case LTTNG_SNAPSHOT_RECORD:
2602 need_domain = 0;
2603 break;
2604 default:
2605 need_domain = 1;
2606 }
2607
2608 if (opt_no_kernel && need_domain
2609 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2610 if (!is_root) {
2611 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2612 } else {
2613 ret = LTTNG_ERR_KERN_NA;
2614 }
2615 goto error;
2616 }
2617
2618 /* Deny register consumer if we already have a spawned consumer. */
2619 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2620 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2621 if (kconsumer_data.pid > 0) {
2622 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2623 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2624 goto error;
2625 }
2626 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2627 }
2628
2629 /*
2630 * Check for command that don't needs to allocate a returned payload. We do
2631 * this here so we don't have to make the call for no payload at each
2632 * command.
2633 */
2634 switch(cmd_ctx->lsm->cmd_type) {
2635 case LTTNG_LIST_SESSIONS:
2636 case LTTNG_LIST_TRACEPOINTS:
2637 case LTTNG_LIST_TRACEPOINT_FIELDS:
2638 case LTTNG_LIST_DOMAINS:
2639 case LTTNG_LIST_CHANNELS:
2640 case LTTNG_LIST_EVENTS:
2641 break;
2642 default:
2643 /* Setup lttng message with no payload */
2644 ret = setup_lttng_msg(cmd_ctx, 0);
2645 if (ret < 0) {
2646 /* This label does not try to unlock the session */
2647 goto init_setup_error;
2648 }
2649 }
2650
2651 /* Commands that DO NOT need a session. */
2652 switch (cmd_ctx->lsm->cmd_type) {
2653 case LTTNG_CREATE_SESSION:
2654 case LTTNG_CREATE_SESSION_SNAPSHOT:
2655 case LTTNG_CREATE_SESSION_LIVE:
2656 case LTTNG_CALIBRATE:
2657 case LTTNG_LIST_SESSIONS:
2658 case LTTNG_LIST_TRACEPOINTS:
2659 case LTTNG_LIST_TRACEPOINT_FIELDS:
2660 need_tracing_session = 0;
2661 break;
2662 default:
2663 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2664 /*
2665 * We keep the session list lock across _all_ commands
2666 * for now, because the per-session lock does not
2667 * handle teardown properly.
2668 */
2669 session_lock_list();
2670 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2671 if (cmd_ctx->session == NULL) {
2672 ret = LTTNG_ERR_SESS_NOT_FOUND;
2673 goto error;
2674 } else {
2675 /* Acquire lock for the session */
2676 session_lock(cmd_ctx->session);
2677 }
2678 break;
2679 }
2680
2681 if (!need_domain) {
2682 goto skip_domain;
2683 }
2684
2685 /*
2686 * Check domain type for specific "pre-action".
2687 */
2688 switch (cmd_ctx->lsm->domain.type) {
2689 case LTTNG_DOMAIN_KERNEL:
2690 if (!is_root) {
2691 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2692 goto error;
2693 }
2694
2695 /* Kernel tracer check */
2696 if (kernel_tracer_fd == -1) {
2697 /* Basically, load kernel tracer modules */
2698 ret = init_kernel_tracer();
2699 if (ret != 0) {
2700 goto error;
2701 }
2702 }
2703
2704 /* Consumer is in an ERROR state. Report back to client */
2705 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2706 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2707 goto error;
2708 }
2709
2710 /* Need a session for kernel command */
2711 if (need_tracing_session) {
2712 if (cmd_ctx->session->kernel_session == NULL) {
2713 ret = create_kernel_session(cmd_ctx->session);
2714 if (ret < 0) {
2715 ret = LTTNG_ERR_KERN_SESS_FAIL;
2716 goto error;
2717 }
2718 }
2719
2720 /* Start the kernel consumer daemon */
2721 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2722 if (kconsumer_data.pid == 0 &&
2723 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2724 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2725 ret = start_consumerd(&kconsumer_data);
2726 if (ret < 0) {
2727 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2728 goto error;
2729 }
2730 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2731 } else {
2732 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2733 }
2734
2735 /*
2736 * The consumer was just spawned so we need to add the socket to
2737 * the consumer output of the session if exist.
2738 */
2739 ret = consumer_create_socket(&kconsumer_data,
2740 cmd_ctx->session->kernel_session->consumer);
2741 if (ret < 0) {
2742 goto error;
2743 }
2744 }
2745
2746 break;
2747 case LTTNG_DOMAIN_JUL:
2748 {
2749 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2750 goto error;
2751 }
2752 case LTTNG_DOMAIN_UST:
2753 {
2754 if (!ust_app_supported()) {
2755 ret = LTTNG_ERR_NO_UST;
2756 goto error;
2757 }
2758 /* Consumer is in an ERROR state. Report back to client */
2759 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2760 ret = LTTNG_ERR_NO_USTCONSUMERD;
2761 goto error;
2762 }
2763
2764 if (need_tracing_session) {
2765 /* Create UST session if none exist. */
2766 if (cmd_ctx->session->ust_session == NULL) {
2767 ret = create_ust_session(cmd_ctx->session,
2768 &cmd_ctx->lsm->domain);
2769 if (ret != LTTNG_OK) {
2770 goto error;
2771 }
2772 }
2773
2774 /* Start the UST consumer daemons */
2775 /* 64-bit */
2776 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2777 if (consumerd64_bin[0] != '\0' &&
2778 ustconsumer64_data.pid == 0 &&
2779 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2780 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2781 ret = start_consumerd(&ustconsumer64_data);
2782 if (ret < 0) {
2783 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2784 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2785 goto error;
2786 }
2787
2788 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2789 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2790 } else {
2791 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2792 }
2793
2794 /*
2795 * Setup socket for consumer 64 bit. No need for atomic access
2796 * since it was set above and can ONLY be set in this thread.
2797 */
2798 ret = consumer_create_socket(&ustconsumer64_data,
2799 cmd_ctx->session->ust_session->consumer);
2800 if (ret < 0) {
2801 goto error;
2802 }
2803
2804 /* 32-bit */
2805 if (consumerd32_bin[0] != '\0' &&
2806 ustconsumer32_data.pid == 0 &&
2807 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2808 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2809 ret = start_consumerd(&ustconsumer32_data);
2810 if (ret < 0) {
2811 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2812 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2813 goto error;
2814 }
2815
2816 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2817 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2818 } else {
2819 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2820 }
2821
2822 /*
2823 * Setup socket for consumer 64 bit. No need for atomic access
2824 * since it was set above and can ONLY be set in this thread.
2825 */
2826 ret = consumer_create_socket(&ustconsumer32_data,
2827 cmd_ctx->session->ust_session->consumer);
2828 if (ret < 0) {
2829 goto error;
2830 }
2831 }
2832 break;
2833 }
2834 default:
2835 break;
2836 }
2837 skip_domain:
2838
2839 /* Validate consumer daemon state when start/stop trace command */
2840 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2841 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2842 switch (cmd_ctx->lsm->domain.type) {
2843 case LTTNG_DOMAIN_JUL:
2844 case LTTNG_DOMAIN_UST:
2845 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2846 ret = LTTNG_ERR_NO_USTCONSUMERD;
2847 goto error;
2848 }
2849 break;
2850 case LTTNG_DOMAIN_KERNEL:
2851 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2852 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2853 goto error;
2854 }
2855 break;
2856 }
2857 }
2858
2859 /*
2860 * Check that the UID or GID match that of the tracing session.
2861 * The root user can interact with all sessions.
2862 */
2863 if (need_tracing_session) {
2864 if (!session_access_ok(cmd_ctx->session,
2865 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2866 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2867 ret = LTTNG_ERR_EPERM;
2868 goto error;
2869 }
2870 }
2871
2872 /*
2873 * Send relayd information to consumer as soon as we have a domain and a
2874 * session defined.
2875 */
2876 if (cmd_ctx->session && need_domain) {
2877 /*
2878 * Setup relayd if not done yet. If the relayd information was already
2879 * sent to the consumer, this call will gracefully return.
2880 */
2881 ret = cmd_setup_relayd(cmd_ctx->session);
2882 if (ret != LTTNG_OK) {
2883 goto error;
2884 }
2885 }
2886
2887 /* Process by command type */
2888 switch (cmd_ctx->lsm->cmd_type) {
2889 case LTTNG_ADD_CONTEXT:
2890 {
2891 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2892 cmd_ctx->lsm->u.context.channel_name,
2893 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2894 break;
2895 }
2896 case LTTNG_DISABLE_CHANNEL:
2897 {
2898 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2899 cmd_ctx->lsm->u.disable.channel_name);
2900 break;
2901 }
2902 case LTTNG_DISABLE_EVENT:
2903 {
2904 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2905 cmd_ctx->lsm->u.disable.channel_name,
2906 cmd_ctx->lsm->u.disable.name);
2907 break;
2908 }
2909 case LTTNG_DISABLE_ALL_EVENT:
2910 {
2911 DBG("Disabling all events");
2912
2913 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2914 cmd_ctx->lsm->u.disable.channel_name);
2915 break;
2916 }
2917 case LTTNG_ENABLE_CHANNEL:
2918 {
2919 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2920 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2921 break;
2922 }
2923 case LTTNG_ENABLE_EVENT:
2924 {
2925 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
2926 cmd_ctx->lsm->u.enable.channel_name,
2927 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2928 break;
2929 }
2930 case LTTNG_ENABLE_ALL_EVENT:
2931 {
2932 DBG("Enabling all events");
2933
2934 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
2935 cmd_ctx->lsm->u.enable.channel_name,
2936 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2937 break;
2938 }
2939 case LTTNG_LIST_TRACEPOINTS:
2940 {
2941 struct lttng_event *events;
2942 ssize_t nb_events;
2943
2944 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2945 if (nb_events < 0) {
2946 /* Return value is a negative lttng_error_code. */
2947 ret = -nb_events;
2948 goto error;
2949 }
2950
2951 /*
2952 * Setup lttng message with payload size set to the event list size in
2953 * bytes and then copy list into the llm payload.
2954 */
2955 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2956 if (ret < 0) {
2957 free(events);
2958 goto setup_error;
2959 }
2960
2961 /* Copy event list into message payload */
2962 memcpy(cmd_ctx->llm->payload, events,
2963 sizeof(struct lttng_event) * nb_events);
2964
2965 free(events);
2966
2967 ret = LTTNG_OK;
2968 break;
2969 }
2970 case LTTNG_LIST_TRACEPOINT_FIELDS:
2971 {
2972 struct lttng_event_field *fields;
2973 ssize_t nb_fields;
2974
2975 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2976 &fields);
2977 if (nb_fields < 0) {
2978 /* Return value is a negative lttng_error_code. */
2979 ret = -nb_fields;
2980 goto error;
2981 }
2982
2983 /*
2984 * Setup lttng message with payload size set to the event list size in
2985 * bytes and then copy list into the llm payload.
2986 */
2987 ret = setup_lttng_msg(cmd_ctx,
2988 sizeof(struct lttng_event_field) * nb_fields);
2989 if (ret < 0) {
2990 free(fields);
2991 goto setup_error;
2992 }
2993
2994 /* Copy event list into message payload */
2995 memcpy(cmd_ctx->llm->payload, fields,
2996 sizeof(struct lttng_event_field) * nb_fields);
2997
2998 free(fields);
2999
3000 ret = LTTNG_OK;
3001 break;
3002 }
3003 case LTTNG_SET_CONSUMER_URI:
3004 {
3005 size_t nb_uri, len;
3006 struct lttng_uri *uris;
3007
3008 nb_uri = cmd_ctx->lsm->u.uri.size;
3009 len = nb_uri * sizeof(struct lttng_uri);
3010
3011 if (nb_uri == 0) {
3012 ret = LTTNG_ERR_INVALID;
3013 goto error;
3014 }
3015
3016 uris = zmalloc(len);
3017 if (uris == NULL) {
3018 ret = LTTNG_ERR_FATAL;
3019 goto error;
3020 }
3021
3022 /* Receive variable len data */
3023 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3024 ret = lttcomm_recv_unix_sock(sock, uris, len);
3025 if (ret <= 0) {
3026 DBG("No URIs received from client... continuing");
3027 *sock_error = 1;
3028 ret = LTTNG_ERR_SESSION_FAIL;
3029 free(uris);
3030 goto error;
3031 }
3032
3033 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3034 nb_uri, uris);
3035 if (ret != LTTNG_OK) {
3036 free(uris);
3037 goto error;
3038 }
3039
3040 /*
3041 * XXX: 0 means that this URI should be applied on the session. Should
3042 * be a DOMAIN enuam.
3043 */
3044 if (cmd_ctx->lsm->domain.type == 0) {
3045 /* Add the URI for the UST session if a consumer is present. */
3046 if (cmd_ctx->session->ust_session &&
3047 cmd_ctx->session->ust_session->consumer) {
3048 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3049 nb_uri, uris);
3050 } else if (cmd_ctx->session->kernel_session &&
3051 cmd_ctx->session->kernel_session->consumer) {
3052 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3053 cmd_ctx->session, nb_uri, uris);
3054 }
3055 }
3056
3057 free(uris);
3058
3059 break;
3060 }
3061 case LTTNG_START_TRACE:
3062 {
3063 ret = cmd_start_trace(cmd_ctx->session);
3064 break;
3065 }
3066 case LTTNG_STOP_TRACE:
3067 {
3068 ret = cmd_stop_trace(cmd_ctx->session);
3069 break;
3070 }
3071 case LTTNG_CREATE_SESSION:
3072 {
3073 size_t nb_uri, len;
3074 struct lttng_uri *uris = NULL;
3075
3076 nb_uri = cmd_ctx->lsm->u.uri.size;
3077 len = nb_uri * sizeof(struct lttng_uri);
3078
3079 if (nb_uri > 0) {
3080 uris = zmalloc(len);
3081 if (uris == NULL) {
3082 ret = LTTNG_ERR_FATAL;
3083 goto error;
3084 }
3085
3086 /* Receive variable len data */
3087 DBG("Waiting for %zu URIs from client ...", nb_uri);
3088 ret = lttcomm_recv_unix_sock(sock, uris, len);
3089 if (ret <= 0) {
3090 DBG("No URIs received from client... continuing");
3091 *sock_error = 1;
3092 ret = LTTNG_ERR_SESSION_FAIL;
3093 free(uris);
3094 goto error;
3095 }
3096
3097 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3098 DBG("Creating session with ONE network URI is a bad call");
3099 ret = LTTNG_ERR_SESSION_FAIL;
3100 free(uris);
3101 goto error;
3102 }
3103 }
3104
3105 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3106 &cmd_ctx->creds, 0);
3107
3108 free(uris);
3109
3110 break;
3111 }
3112 case LTTNG_DESTROY_SESSION:
3113 {
3114 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3115
3116 /* Set session to NULL so we do not unlock it after free. */
3117 cmd_ctx->session = NULL;
3118 break;
3119 }
3120 case LTTNG_LIST_DOMAINS:
3121 {
3122 ssize_t nb_dom;
3123 struct lttng_domain *domains;
3124
3125 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3126 if (nb_dom < 0) {
3127 /* Return value is a negative lttng_error_code. */
3128 ret = -nb_dom;
3129 goto error;
3130 }
3131
3132 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3133 if (ret < 0) {
3134 free(domains);
3135 goto setup_error;
3136 }
3137
3138 /* Copy event list into message payload */
3139 memcpy(cmd_ctx->llm->payload, domains,
3140 nb_dom * sizeof(struct lttng_domain));
3141
3142 free(domains);
3143
3144 ret = LTTNG_OK;
3145 break;
3146 }
3147 case LTTNG_LIST_CHANNELS:
3148 {
3149 int nb_chan;
3150 struct lttng_channel *channels;
3151
3152 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3153 cmd_ctx->session, &channels);
3154 if (nb_chan < 0) {
3155 /* Return value is a negative lttng_error_code. */
3156 ret = -nb_chan;
3157 goto error;
3158 }
3159
3160 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3161 if (ret < 0) {
3162 free(channels);
3163 goto setup_error;
3164 }
3165
3166 /* Copy event list into message payload */
3167 memcpy(cmd_ctx->llm->payload, channels,
3168 nb_chan * sizeof(struct lttng_channel));
3169
3170 free(channels);
3171
3172 ret = LTTNG_OK;
3173 break;
3174 }
3175 case LTTNG_LIST_EVENTS:
3176 {
3177 ssize_t nb_event;
3178 struct lttng_event *events = NULL;
3179
3180 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3181 cmd_ctx->lsm->u.list.channel_name, &events);
3182 if (nb_event < 0) {
3183 /* Return value is a negative lttng_error_code. */
3184 ret = -nb_event;
3185 goto error;
3186 }
3187
3188 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3189 if (ret < 0) {
3190 free(events);
3191 goto setup_error;
3192 }
3193
3194 /* Copy event list into message payload */
3195 memcpy(cmd_ctx->llm->payload, events,
3196 nb_event * sizeof(struct lttng_event));
3197
3198 free(events);
3199
3200 ret = LTTNG_OK;
3201 break;
3202 }
3203 case LTTNG_LIST_SESSIONS:
3204 {
3205 unsigned int nr_sessions;
3206
3207 session_lock_list();
3208 nr_sessions = lttng_sessions_count(
3209 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3210 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3211
3212 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3213 if (ret < 0) {
3214 session_unlock_list();
3215 goto setup_error;
3216 }
3217
3218 /* Filled the session array */
3219 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3220 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3221 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3222
3223 session_unlock_list();
3224
3225 ret = LTTNG_OK;
3226 break;
3227 }
3228 case LTTNG_CALIBRATE:
3229 {
3230 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3231 &cmd_ctx->lsm->u.calibrate);
3232 break;
3233 }
3234 case LTTNG_REGISTER_CONSUMER:
3235 {
3236 struct consumer_data *cdata;
3237
3238 switch (cmd_ctx->lsm->domain.type) {
3239 case LTTNG_DOMAIN_KERNEL:
3240 cdata = &kconsumer_data;
3241 break;
3242 default:
3243 ret = LTTNG_ERR_UND;
3244 goto error;
3245 }
3246
3247 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3248 cmd_ctx->lsm->u.reg.path, cdata);
3249 break;
3250 }
3251 case LTTNG_ENABLE_EVENT_WITH_FILTER:
3252 {
3253 struct lttng_filter_bytecode *bytecode;
3254
3255 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
3256 ret = LTTNG_ERR_FILTER_INVAL;
3257 goto error;
3258 }
3259 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3260 ret = LTTNG_ERR_FILTER_INVAL;
3261 goto error;
3262 }
3263 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
3264 if (!bytecode) {
3265 ret = LTTNG_ERR_FILTER_NOMEM;
3266 goto error;
3267 }
3268 /* Receive var. len. data */
3269 DBG("Receiving var len data from client ...");
3270 ret = lttcomm_recv_unix_sock(sock, bytecode,
3271 cmd_ctx->lsm->u.enable.bytecode_len);
3272 if (ret <= 0) {
3273 DBG("Nothing recv() from client var len data... continuing");
3274 *sock_error = 1;
3275 ret = LTTNG_ERR_FILTER_INVAL;
3276 goto error;
3277 }
3278
3279 if (bytecode->len + sizeof(*bytecode)
3280 != cmd_ctx->lsm->u.enable.bytecode_len) {
3281 free(bytecode);
3282 ret = LTTNG_ERR_FILTER_INVAL;
3283 goto error;
3284 }
3285
3286 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3287 cmd_ctx->lsm->u.enable.channel_name,
3288 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
3289 break;
3290 }
3291 case LTTNG_DATA_PENDING:
3292 {
3293 ret = cmd_data_pending(cmd_ctx->session);
3294 break;
3295 }
3296 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3297 {
3298 struct lttcomm_lttng_output_id reply;
3299
3300 ret = cmd_snapshot_add_output(cmd_ctx->session,
3301 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3302 if (ret != LTTNG_OK) {
3303 goto error;
3304 }
3305
3306 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3307 if (ret < 0) {
3308 goto setup_error;
3309 }
3310
3311 /* Copy output list into message payload */
3312 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3313 ret = LTTNG_OK;
3314 break;
3315 }
3316 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3317 {
3318 ret = cmd_snapshot_del_output(cmd_ctx->session,
3319 &cmd_ctx->lsm->u.snapshot_output.output);
3320 break;
3321 }
3322 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3323 {
3324 ssize_t nb_output;
3325 struct lttng_snapshot_output *outputs = NULL;
3326
3327 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3328 if (nb_output < 0) {
3329 ret = -nb_output;
3330 goto error;
3331 }
3332
3333 ret = setup_lttng_msg(cmd_ctx,
3334 nb_output * sizeof(struct lttng_snapshot_output));
3335 if (ret < 0) {
3336 free(outputs);
3337 goto setup_error;
3338 }
3339
3340 if (outputs) {
3341 /* Copy output list into message payload */
3342 memcpy(cmd_ctx->llm->payload, outputs,
3343 nb_output * sizeof(struct lttng_snapshot_output));
3344 free(outputs);
3345 }
3346
3347 ret = LTTNG_OK;
3348 break;
3349 }
3350 case LTTNG_SNAPSHOT_RECORD:
3351 {
3352 ret = cmd_snapshot_record(cmd_ctx->session,
3353 &cmd_ctx->lsm->u.snapshot_record.output,
3354 cmd_ctx->lsm->u.snapshot_record.wait);
3355 break;
3356 }
3357 case LTTNG_CREATE_SESSION_SNAPSHOT:
3358 {
3359 size_t nb_uri, len;
3360 struct lttng_uri *uris = NULL;
3361
3362 nb_uri = cmd_ctx->lsm->u.uri.size;
3363 len = nb_uri * sizeof(struct lttng_uri);
3364
3365 if (nb_uri > 0) {
3366 uris = zmalloc(len);
3367 if (uris == NULL) {
3368 ret = LTTNG_ERR_FATAL;
3369 goto error;
3370 }
3371
3372 /* Receive variable len data */
3373 DBG("Waiting for %zu URIs from client ...", nb_uri);
3374 ret = lttcomm_recv_unix_sock(sock, uris, len);
3375 if (ret <= 0) {
3376 DBG("No URIs received from client... continuing");
3377 *sock_error = 1;
3378 ret = LTTNG_ERR_SESSION_FAIL;
3379 free(uris);
3380 goto error;
3381 }
3382
3383 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3384 DBG("Creating session with ONE network URI is a bad call");
3385 ret = LTTNG_ERR_SESSION_FAIL;
3386 free(uris);
3387 goto error;
3388 }
3389 }
3390
3391 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3392 nb_uri, &cmd_ctx->creds);
3393 free(uris);
3394 break;
3395 }
3396 case LTTNG_CREATE_SESSION_LIVE:
3397 {
3398 size_t nb_uri, len;
3399 struct lttng_uri *uris = NULL;
3400
3401 nb_uri = cmd_ctx->lsm->u.uri.size;
3402 len = nb_uri * sizeof(struct lttng_uri);
3403
3404 if (nb_uri > 0) {
3405 uris = zmalloc(len);
3406 if (uris == NULL) {
3407 ret = LTTNG_ERR_FATAL;
3408 goto error;
3409 }
3410
3411 /* Receive variable len data */
3412 DBG("Waiting for %zu URIs from client ...", nb_uri);
3413 ret = lttcomm_recv_unix_sock(sock, uris, len);
3414 if (ret <= 0) {
3415 DBG("No URIs received from client... continuing");
3416 *sock_error = 1;
3417 ret = LTTNG_ERR_SESSION_FAIL;
3418 free(uris);
3419 goto error;
3420 }
3421
3422 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3423 DBG("Creating session with ONE network URI is a bad call");
3424 ret = LTTNG_ERR_SESSION_FAIL;
3425 free(uris);
3426 goto error;
3427 }
3428 }
3429
3430 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3431 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3432 free(uris);
3433 break;
3434 }
3435 default:
3436 ret = LTTNG_ERR_UND;
3437 break;
3438 }
3439
3440 error:
3441 if (cmd_ctx->llm == NULL) {
3442 DBG("Missing llm structure. Allocating one.");
3443 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3444 goto setup_error;
3445 }
3446 }
3447 /* Set return code */
3448 cmd_ctx->llm->ret_code = ret;
3449 setup_error:
3450 if (cmd_ctx->session) {
3451 session_unlock(cmd_ctx->session);
3452 }
3453 if (need_tracing_session) {
3454 session_unlock_list();
3455 }
3456 init_setup_error:
3457 return ret;
3458 }
3459
3460 /*
3461 * Thread managing health check socket.
3462 */
3463 static void *thread_manage_health(void *data)
3464 {
3465 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3466 uint32_t revents, nb_fd;
3467 struct lttng_poll_event events;
3468 struct health_comm_msg msg;
3469 struct health_comm_reply reply;
3470
3471 DBG("[thread] Manage health check started");
3472
3473 rcu_register_thread();
3474
3475 /* We might hit an error path before this is created. */
3476 lttng_poll_init(&events);
3477
3478 /* Create unix socket */
3479 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3480 if (sock < 0) {
3481 ERR("Unable to create health check Unix socket");
3482 ret = -1;
3483 goto error;
3484 }
3485
3486 if (is_root) {
3487 /* lttng health client socket path permissions */
3488 ret = chown(health_unix_sock_path, 0,
3489 utils_get_group_id(tracing_group_name));
3490 if (ret < 0) {
3491 ERR("Unable to set group on %s", health_unix_sock_path);
3492 PERROR("chown");
3493 ret = -1;
3494 goto error;
3495 }
3496
3497 ret = chmod(health_unix_sock_path,
3498 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3499 if (ret < 0) {
3500 ERR("Unable to set permissions on %s", health_unix_sock_path);
3501 PERROR("chmod");
3502 ret = -1;
3503 goto error;
3504 }
3505 }
3506
3507 /*
3508 * Set the CLOEXEC flag. Return code is useless because either way, the
3509 * show must go on.
3510 */
3511 (void) utils_set_fd_cloexec(sock);
3512
3513 ret = lttcomm_listen_unix_sock(sock);
3514 if (ret < 0) {
3515 goto error;
3516 }
3517
3518 /*
3519 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3520 * more will be added to this poll set.
3521 */
3522 ret = sessiond_set_thread_pollset(&events, 2);
3523 if (ret < 0) {
3524 goto error;
3525 }
3526
3527 /* Add the application registration socket */
3528 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3529 if (ret < 0) {
3530 goto error;
3531 }
3532
3533 while (1) {
3534 DBG("Health check ready");
3535
3536 /* Inifinite blocking call, waiting for transmission */
3537 restart:
3538 ret = lttng_poll_wait(&events, -1);
3539 if (ret < 0) {
3540 /*
3541 * Restart interrupted system call.
3542 */
3543 if (errno == EINTR) {
3544 goto restart;
3545 }
3546 goto error;
3547 }
3548
3549 nb_fd = ret;
3550
3551 for (i = 0; i < nb_fd; i++) {
3552 /* Fetch once the poll data */
3553 revents = LTTNG_POLL_GETEV(&events, i);
3554 pollfd = LTTNG_POLL_GETFD(&events, i);
3555
3556 /* Thread quit pipe has been closed. Killing thread. */
3557 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3558 if (ret) {
3559 err = 0;
3560 goto exit;
3561 }
3562
3563 /* Event on the registration socket */
3564 if (pollfd == sock) {
3565 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3566 ERR("Health socket poll error");
3567 goto error;
3568 }
3569 }
3570 }
3571
3572 new_sock = lttcomm_accept_unix_sock(sock);
3573 if (new_sock < 0) {
3574 goto error;
3575 }
3576
3577 /*
3578 * Set the CLOEXEC flag. Return code is useless because either way, the
3579 * show must go on.
3580 */
3581 (void) utils_set_fd_cloexec(new_sock);
3582
3583 DBG("Receiving data from client for health...");
3584 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3585 if (ret <= 0) {
3586 DBG("Nothing recv() from client... continuing");
3587 ret = close(new_sock);
3588 if (ret) {
3589 PERROR("close");
3590 }
3591 new_sock = -1;
3592 continue;
3593 }
3594
3595 rcu_thread_online();
3596
3597 reply.ret_code = 0;
3598 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3599 /*
3600 * health_check_state returns 0 if health is
3601 * bad.
3602 */
3603 if (!health_check_state(health_sessiond, i)) {
3604 reply.ret_code |= 1ULL << i;
3605 }
3606 }
3607
3608 DBG2("Health check return value %" PRIx64, reply.ret_code);
3609
3610 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3611 if (ret < 0) {
3612 ERR("Failed to send health data back to client");
3613 }
3614
3615 /* End of transmission */
3616 ret = close(new_sock);
3617 if (ret) {
3618 PERROR("close");
3619 }
3620 new_sock = -1;
3621 }
3622
3623 exit:
3624 error:
3625 if (err) {
3626 ERR("Health error occurred in %s", __func__);
3627 }
3628 DBG("Health check thread dying");
3629 unlink(health_unix_sock_path);
3630 if (sock >= 0) {
3631 ret = close(sock);
3632 if (ret) {
3633 PERROR("close");
3634 }
3635 }
3636
3637 lttng_poll_clean(&events);
3638
3639 rcu_unregister_thread();
3640 return NULL;
3641 }
3642
3643 /*
3644 * This thread manage all clients request using the unix client socket for
3645 * communication.
3646 */
3647 static void *thread_manage_clients(void *data)
3648 {
3649 int sock = -1, ret, i, pollfd, err = -1;
3650 int sock_error;
3651 uint32_t revents, nb_fd;
3652 struct command_ctx *cmd_ctx = NULL;
3653 struct lttng_poll_event events;
3654
3655 DBG("[thread] Manage client started");
3656
3657 rcu_register_thread();
3658
3659 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3660
3661 if (testpoint(thread_manage_clients)) {
3662 goto error_testpoint;
3663 }
3664
3665 health_code_update();
3666
3667 ret = lttcomm_listen_unix_sock(client_sock);
3668 if (ret < 0) {
3669 goto error_listen;
3670 }
3671
3672 /*
3673 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3674 * more will be added to this poll set.
3675 */
3676 ret = sessiond_set_thread_pollset(&events, 2);
3677 if (ret < 0) {
3678 goto error_create_poll;
3679 }
3680
3681 /* Add the application registration socket */
3682 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3683 if (ret < 0) {
3684 goto error;
3685 }
3686
3687 /*
3688 * Notify parent pid that we are ready to accept command for client side.
3689 */
3690 if (opt_sig_parent) {
3691 kill(ppid, SIGUSR1);
3692 }
3693
3694 if (testpoint(thread_manage_clients_before_loop)) {
3695 goto error;
3696 }
3697
3698 health_code_update();
3699
3700 while (1) {
3701 DBG("Accepting client command ...");
3702
3703 /* Inifinite blocking call, waiting for transmission */
3704 restart:
3705 health_poll_entry();
3706 ret = lttng_poll_wait(&events, -1);
3707 health_poll_exit();
3708 if (ret < 0) {
3709 /*
3710 * Restart interrupted system call.
3711 */
3712 if (errno == EINTR) {
3713 goto restart;
3714 }
3715 goto error;
3716 }
3717
3718 nb_fd = ret;
3719
3720 for (i = 0; i < nb_fd; i++) {
3721 /* Fetch once the poll data */
3722 revents = LTTNG_POLL_GETEV(&events, i);
3723 pollfd = LTTNG_POLL_GETFD(&events, i);
3724
3725 health_code_update();
3726
3727 /* Thread quit pipe has been closed. Killing thread. */
3728 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3729 if (ret) {
3730 err = 0;
3731 goto exit;
3732 }
3733
3734 /* Event on the registration socket */
3735 if (pollfd == client_sock) {
3736 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3737 ERR("Client socket poll error");
3738 goto error;
3739 }
3740 }
3741 }
3742
3743 DBG("Wait for client response");
3744
3745 health_code_update();
3746
3747 sock = lttcomm_accept_unix_sock(client_sock);
3748 if (sock < 0) {
3749 goto error;
3750 }
3751
3752 /*
3753 * Set the CLOEXEC flag. Return code is useless because either way, the
3754 * show must go on.
3755 */
3756 (void) utils_set_fd_cloexec(sock);
3757
3758 /* Set socket option for credentials retrieval */
3759 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3760 if (ret < 0) {
3761 goto error;
3762 }
3763
3764 /* Allocate context command to process the client request */
3765 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3766 if (cmd_ctx == NULL) {
3767 PERROR("zmalloc cmd_ctx");
3768 goto error;
3769 }
3770
3771 /* Allocate data buffer for reception */
3772 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3773 if (cmd_ctx->lsm == NULL) {
3774 PERROR("zmalloc cmd_ctx->lsm");
3775 goto error;
3776 }
3777
3778 cmd_ctx->llm = NULL;
3779 cmd_ctx->session = NULL;
3780
3781 health_code_update();
3782
3783 /*
3784 * Data is received from the lttng client. The struct
3785 * lttcomm_session_msg (lsm) contains the command and data request of
3786 * the client.
3787 */
3788 DBG("Receiving data from client ...");
3789 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3790 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3791 if (ret <= 0) {
3792 DBG("Nothing recv() from client... continuing");
3793 ret = close(sock);
3794 if (ret) {
3795 PERROR("close");
3796 }
3797 sock = -1;
3798 clean_command_ctx(&cmd_ctx);
3799 continue;
3800 }
3801
3802 health_code_update();
3803
3804 // TODO: Validate cmd_ctx including sanity check for
3805 // security purpose.
3806
3807 rcu_thread_online();
3808 /*
3809 * This function dispatch the work to the kernel or userspace tracer
3810 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3811 * informations for the client. The command context struct contains
3812 * everything this function may needs.
3813 */
3814 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3815 rcu_thread_offline();
3816 if (ret < 0) {
3817 ret = close(sock);
3818 if (ret) {
3819 PERROR("close");
3820 }
3821 sock = -1;
3822 /*
3823 * TODO: Inform client somehow of the fatal error. At
3824 * this point, ret < 0 means that a zmalloc failed
3825 * (ENOMEM). Error detected but still accept
3826 * command, unless a socket error has been
3827 * detected.
3828 */
3829 clean_command_ctx(&cmd_ctx);
3830 continue;
3831 }
3832
3833 health_code_update();
3834
3835 DBG("Sending response (size: %d, retcode: %s)",
3836 cmd_ctx->lttng_msg_size,
3837 lttng_strerror(-cmd_ctx->llm->ret_code));
3838 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3839 if (ret < 0) {
3840 ERR("Failed to send data back to client");
3841 }
3842
3843 /* End of transmission */
3844 ret = close(sock);
3845 if (ret) {
3846 PERROR("close");
3847 }
3848 sock = -1;
3849
3850 clean_command_ctx(&cmd_ctx);
3851
3852 health_code_update();
3853 }
3854
3855 exit:
3856 error:
3857 if (sock >= 0) {
3858 ret = close(sock);
3859 if (ret) {
3860 PERROR("close");
3861 }
3862 }
3863
3864 lttng_poll_clean(&events);
3865 clean_command_ctx(&cmd_ctx);
3866
3867 error_listen:
3868 error_create_poll:
3869 error_testpoint:
3870 unlink(client_unix_sock_path);
3871 if (client_sock >= 0) {
3872 ret = close(client_sock);
3873 if (ret) {
3874 PERROR("close");
3875 }
3876 }
3877
3878 if (err) {
3879 health_error();
3880 ERR("Health error occurred in %s", __func__);
3881 }
3882
3883 health_unregister(health_sessiond);
3884
3885 DBG("Client thread dying");
3886
3887 rcu_unregister_thread();
3888 return NULL;
3889 }
3890
3891
3892 /*
3893 * usage function on stderr
3894 */
3895 static void usage(void)
3896 {
3897 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3898 fprintf(stderr, " -h, --help Display this usage.\n");
3899 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3900 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3901 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3902 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3903 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3904 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3905 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3906 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3907 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3908 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3909 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3910 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3911 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3912 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3913 fprintf(stderr, " -V, --version Show version number.\n");
3914 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3915 fprintf(stderr, " -q, --quiet No output at all.\n");
3916 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3917 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3918 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3919 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3920 }
3921
3922 /*
3923 * daemon argument parsing
3924 */
3925 static int parse_args(int argc, char **argv)
3926 {
3927 int c;
3928
3929 static struct option long_options[] = {
3930 { "client-sock", 1, 0, 'c' },
3931 { "apps-sock", 1, 0, 'a' },
3932 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3933 { "kconsumerd-err-sock", 1, 0, 'E' },
3934 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3935 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3936 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3937 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3938 { "consumerd32-path", 1, 0, 'u' },
3939 { "consumerd32-libdir", 1, 0, 'U' },
3940 { "consumerd64-path", 1, 0, 't' },
3941 { "consumerd64-libdir", 1, 0, 'T' },
3942 { "daemonize", 0, 0, 'd' },
3943 { "sig-parent", 0, 0, 'S' },
3944 { "help", 0, 0, 'h' },
3945 { "group", 1, 0, 'g' },
3946 { "version", 0, 0, 'V' },
3947 { "quiet", 0, 0, 'q' },
3948 { "verbose", 0, 0, 'v' },
3949 { "verbose-consumer", 0, 0, 'Z' },
3950 { "no-kernel", 0, 0, 'N' },
3951 { "pidfile", 1, 0, 'p' },
3952 { NULL, 0, 0, 0 }
3953 };
3954
3955 while (1) {
3956 int option_index = 0;
3957 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
3958 long_options, &option_index);
3959 if (c == -1) {
3960 break;
3961 }
3962
3963 switch (c) {
3964 case 0:
3965 fprintf(stderr, "option %s", long_options[option_index].name);
3966 if (optarg) {
3967 fprintf(stderr, " with arg %s\n", optarg);
3968 }
3969 break;
3970 case 'c':
3971 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3972 break;
3973 case 'a':
3974 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3975 break;
3976 case 'd':
3977 opt_daemon = 1;
3978 break;
3979 case 'g':
3980 tracing_group_name = optarg;
3981 break;
3982 case 'h':
3983 usage();
3984 exit(EXIT_FAILURE);
3985 case 'V':
3986 fprintf(stdout, "%s\n", VERSION);
3987 exit(EXIT_SUCCESS);
3988 case 'S':
3989 opt_sig_parent = 1;
3990 break;
3991 case 'E':
3992 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3993 break;
3994 case 'C':
3995 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3996 break;
3997 case 'F':
3998 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3999 break;
4000 case 'D':
4001 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4002 break;
4003 case 'H':
4004 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4005 break;
4006 case 'G':
4007 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4008 break;
4009 case 'N':
4010 opt_no_kernel = 1;
4011 break;
4012 case 'q':
4013 lttng_opt_quiet = 1;
4014 break;
4015 case 'v':
4016 /* Verbose level can increase using multiple -v */
4017 lttng_opt_verbose += 1;
4018 break;
4019 case 'Z':
4020 opt_verbose_consumer += 1;
4021 break;
4022 case 'u':
4023 consumerd32_bin= optarg;
4024 break;
4025 case 'U':
4026 consumerd32_libdir = optarg;
4027 break;
4028 case 't':
4029 consumerd64_bin = optarg;
4030 break;
4031 case 'T':
4032 consumerd64_libdir = optarg;
4033 break;
4034 case 'p':
4035 opt_pidfile = optarg;
4036 break;
4037 default:
4038 /* Unknown option or other error.
4039 * Error is printed by getopt, just return */
4040 return -1;
4041 }
4042 }
4043
4044 return 0;
4045 }
4046
4047 /*
4048 * Creates the two needed socket by the daemon.
4049 * apps_sock - The communication socket for all UST apps.
4050 * client_sock - The communication of the cli tool (lttng).
4051 */
4052 static int init_daemon_socket(void)
4053 {
4054 int ret = 0;
4055 mode_t old_umask;
4056
4057 old_umask = umask(0);
4058
4059 /* Create client tool unix socket */
4060 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4061 if (client_sock < 0) {
4062 ERR("Create unix sock failed: %s", client_unix_sock_path);
4063 ret = -1;
4064 goto end;
4065 }
4066
4067 /* Set the cloexec flag */
4068 ret = utils_set_fd_cloexec(client_sock);
4069 if (ret < 0) {
4070 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4071 "Continuing but note that the consumer daemon will have a "
4072 "reference to this socket on exec()", client_sock);
4073 }
4074
4075 /* File permission MUST be 660 */
4076 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4077 if (ret < 0) {
4078 ERR("Set file permissions failed: %s", client_unix_sock_path);
4079 PERROR("chmod");
4080 goto end;
4081 }
4082
4083 /* Create the application unix socket */
4084 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4085 if (apps_sock < 0) {
4086 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4087 ret = -1;
4088 goto end;
4089 }
4090
4091 /* Set the cloexec flag */
4092 ret = utils_set_fd_cloexec(apps_sock);
4093 if (ret < 0) {
4094 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4095 "Continuing but note that the consumer daemon will have a "
4096 "reference to this socket on exec()", apps_sock);
4097 }
4098
4099 /* File permission MUST be 666 */
4100 ret = chmod(apps_unix_sock_path,
4101 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4102 if (ret < 0) {
4103 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4104 PERROR("chmod");
4105 goto end;
4106 }
4107
4108 DBG3("Session daemon client socket %d and application socket %d created",
4109 client_sock, apps_sock);
4110
4111 end:
4112 umask(old_umask);
4113 return ret;
4114 }
4115
4116 /*
4117 * Check if the global socket is available, and if a daemon is answering at the
4118 * other side. If yes, error is returned.
4119 */
4120 static int check_existing_daemon(void)
4121 {
4122 /* Is there anybody out there ? */
4123 if (lttng_session_daemon_alive()) {
4124 return -EEXIST;
4125 }
4126
4127 return 0;
4128 }
4129
4130 /*
4131 * Set the tracing group gid onto the client socket.
4132 *
4133 * Race window between mkdir and chown is OK because we are going from more
4134 * permissive (root.root) to less permissive (root.tracing).
4135 */
4136 static int set_permissions(char *rundir)
4137 {
4138 int ret;
4139 gid_t gid;
4140
4141 gid = utils_get_group_id(tracing_group_name);
4142
4143 /* Set lttng run dir */
4144 ret = chown(rundir, 0, gid);
4145 if (ret < 0) {
4146 ERR("Unable to set group on %s", rundir);
4147 PERROR("chown");
4148 }
4149
4150 /*
4151 * Ensure all applications and tracing group can search the run
4152 * dir. Allow everyone to read the directory, since it does not
4153 * buy us anything to hide its content.
4154 */
4155 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4156 if (ret < 0) {
4157 ERR("Unable to set permissions on %s", rundir);
4158 PERROR("chmod");
4159 }
4160
4161 /* lttng client socket path */
4162 ret = chown(client_unix_sock_path, 0, gid);
4163 if (ret < 0) {
4164 ERR("Unable to set group on %s", client_unix_sock_path);
4165 PERROR("chown");
4166 }
4167
4168 /* kconsumer error socket path */
4169 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4170 if (ret < 0) {
4171 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4172 PERROR("chown");
4173 }
4174
4175 /* 64-bit ustconsumer error socket path */
4176 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4177 if (ret < 0) {
4178 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4179 PERROR("chown");
4180 }
4181
4182 /* 32-bit ustconsumer compat32 error socket path */
4183 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4184 if (ret < 0) {
4185 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4186 PERROR("chown");
4187 }
4188
4189 DBG("All permissions are set");
4190
4191 return ret;
4192 }
4193
4194 /*
4195 * Create the lttng run directory needed for all global sockets and pipe.
4196 */
4197 static int create_lttng_rundir(const char *rundir)
4198 {
4199 int ret;
4200
4201 DBG3("Creating LTTng run directory: %s", rundir);
4202
4203 ret = mkdir(rundir, S_IRWXU);
4204 if (ret < 0) {
4205 if (errno != EEXIST) {
4206 ERR("Unable to create %s", rundir);
4207 goto error;
4208 } else {
4209 ret = 0;
4210 }
4211 }
4212
4213 error:
4214 return ret;
4215 }
4216
4217 /*
4218 * Setup sockets and directory needed by the kconsumerd communication with the
4219 * session daemon.
4220 */
4221 static int set_consumer_sockets(struct consumer_data *consumer_data,
4222 const char *rundir)
4223 {
4224 int ret;
4225 char path[PATH_MAX];
4226
4227 switch (consumer_data->type) {
4228 case LTTNG_CONSUMER_KERNEL:
4229 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4230 break;
4231 case LTTNG_CONSUMER64_UST:
4232 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4233 break;
4234 case LTTNG_CONSUMER32_UST:
4235 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4236 break;
4237 default:
4238 ERR("Consumer type unknown");
4239 ret = -EINVAL;
4240 goto error;
4241 }
4242
4243 DBG2("Creating consumer directory: %s", path);
4244
4245 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4246 if (ret < 0) {
4247 if (errno != EEXIST) {
4248 PERROR("mkdir");
4249 ERR("Failed to create %s", path);
4250 goto error;
4251 }
4252 ret = -1;
4253 }
4254 if (is_root) {
4255 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4256 if (ret < 0) {
4257 ERR("Unable to set group on %s", path);
4258 PERROR("chown");
4259 goto error;
4260 }
4261 }
4262
4263 /* Create the kconsumerd error unix socket */
4264 consumer_data->err_sock =
4265 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4266 if (consumer_data->err_sock < 0) {
4267 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4268 ret = -1;
4269 goto error;
4270 }
4271
4272 /*
4273 * Set the CLOEXEC flag. Return code is useless because either way, the
4274 * show must go on.
4275 */
4276 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4277 if (ret < 0) {
4278 PERROR("utils_set_fd_cloexec");
4279 /* continue anyway */
4280 }
4281
4282 /* File permission MUST be 660 */
4283 ret = chmod(consumer_data->err_unix_sock_path,
4284 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4285 if (ret < 0) {
4286 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4287 PERROR("chmod");
4288 goto error;
4289 }
4290
4291 error:
4292 return ret;
4293 }
4294
4295 /*
4296 * Signal handler for the daemon
4297 *
4298 * Simply stop all worker threads, leaving main() return gracefully after
4299 * joining all threads and calling cleanup().
4300 */
4301 static void sighandler(int sig)
4302 {
4303 switch (sig) {
4304 case SIGPIPE:
4305 DBG("SIGPIPE caught");
4306 return;
4307 case SIGINT:
4308 DBG("SIGINT caught");
4309 stop_threads();
4310 break;
4311 case SIGTERM:
4312 DBG("SIGTERM caught");
4313 stop_threads();
4314 break;
4315 default:
4316 break;
4317 }
4318 }
4319
4320 /*
4321 * Setup signal handler for :
4322 * SIGINT, SIGTERM, SIGPIPE
4323 */
4324 static int set_signal_handler(void)
4325 {
4326 int ret = 0;
4327 struct sigaction sa;
4328 sigset_t sigset;
4329
4330 if ((ret = sigemptyset(&sigset)) < 0) {
4331 PERROR("sigemptyset");
4332 return ret;
4333 }
4334
4335 sa.sa_handler = sighandler;
4336 sa.sa_mask = sigset;
4337 sa.sa_flags = 0;
4338 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4339 PERROR("sigaction");
4340 return ret;
4341 }
4342
4343 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4344 PERROR("sigaction");
4345 return ret;
4346 }
4347
4348 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4349 PERROR("sigaction");
4350 return ret;
4351 }
4352
4353 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4354
4355 return ret;
4356 }
4357
4358 /*
4359 * Set open files limit to unlimited. This daemon can open a large number of
4360 * file descriptors in order to consumer multiple kernel traces.
4361 */
4362 static void set_ulimit(void)
4363 {
4364 int ret;
4365 struct rlimit lim;
4366
4367 /* The kernel does not allowed an infinite limit for open files */
4368 lim.rlim_cur = 65535;
4369 lim.rlim_max = 65535;
4370
4371 ret = setrlimit(RLIMIT_NOFILE, &lim);
4372 if (ret < 0) {
4373 PERROR("failed to set open files limit");
4374 }
4375 }
4376
4377 /*
4378 * Write pidfile using the rundir and opt_pidfile.
4379 */
4380 static void write_pidfile(void)
4381 {
4382 int ret;
4383 char pidfile_path[PATH_MAX];
4384
4385 assert(rundir);
4386
4387 if (opt_pidfile) {
4388 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4389 } else {
4390 /* Build pidfile path from rundir and opt_pidfile. */
4391 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4392 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4393 if (ret < 0) {
4394 PERROR("snprintf pidfile path");
4395 goto error;
4396 }
4397 }
4398
4399 /*
4400 * Create pid file in rundir. Return value is of no importance. The
4401 * execution will continue even though we are not able to write the file.
4402 */
4403 (void) utils_create_pid_file(getpid(), pidfile_path);
4404
4405 error:
4406 return;
4407 }
4408
4409 /*
4410 * main
4411 */
4412 int main(int argc, char **argv)
4413 {
4414 int ret = 0;
4415 void *status;
4416 const char *home_path, *env_app_timeout;
4417
4418 init_kernel_workarounds();
4419
4420 rcu_register_thread();
4421
4422 setup_consumerd_path();
4423
4424 page_size = sysconf(_SC_PAGESIZE);
4425 if (page_size < 0) {
4426 PERROR("sysconf _SC_PAGESIZE");
4427 page_size = LONG_MAX;
4428 WARN("Fallback page size to %ld", page_size);
4429 }
4430
4431 /* Parse arguments */
4432 progname = argv[0];
4433 if ((ret = parse_args(argc, argv)) < 0) {
4434 goto error;
4435 }
4436
4437 /* Daemonize */
4438 if (opt_daemon) {
4439 int i;
4440
4441 /*
4442 * fork
4443 * child: setsid, close FD 0, 1, 2, chdir /
4444 * parent: exit (if fork is successful)
4445 */
4446 ret = daemon(0, 0);
4447 if (ret < 0) {
4448 PERROR("daemon");
4449 goto error;
4450 }
4451 /*
4452 * We are in the child. Make sure all other file
4453 * descriptors are closed, in case we are called with
4454 * more opened file descriptors than the standard ones.
4455 */
4456 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4457 (void) close(i);
4458 }
4459 }
4460
4461 /* Create thread quit pipe */
4462 if ((ret = init_thread_quit_pipe()) < 0) {
4463 goto error;
4464 }
4465
4466 /* Check if daemon is UID = 0 */
4467 is_root = !getuid();
4468
4469 if (is_root) {
4470 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4471
4472 /* Create global run dir with root access */
4473 ret = create_lttng_rundir(rundir);
4474 if (ret < 0) {
4475 goto error;
4476 }
4477
4478 if (strlen(apps_unix_sock_path) == 0) {
4479 snprintf(apps_unix_sock_path, PATH_MAX,
4480 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4481 }
4482
4483 if (strlen(client_unix_sock_path) == 0) {
4484 snprintf(client_unix_sock_path, PATH_MAX,
4485 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4486 }
4487
4488 /* Set global SHM for ust */
4489 if (strlen(wait_shm_path) == 0) {
4490 snprintf(wait_shm_path, PATH_MAX,
4491 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4492 }
4493
4494 if (strlen(health_unix_sock_path) == 0) {
4495 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4496 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4497 }
4498
4499 /* Setup kernel consumerd path */
4500 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4501 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4502 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4503 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4504
4505 DBG2("Kernel consumer err path: %s",
4506 kconsumer_data.err_unix_sock_path);
4507 DBG2("Kernel consumer cmd path: %s",
4508 kconsumer_data.cmd_unix_sock_path);
4509 } else {
4510 home_path = utils_get_home_dir();
4511 if (home_path == NULL) {
4512 /* TODO: Add --socket PATH option */
4513 ERR("Can't get HOME directory for sockets creation.");
4514 ret = -EPERM;
4515 goto error;
4516 }
4517
4518 /*
4519 * Create rundir from home path. This will create something like
4520 * $HOME/.lttng
4521 */
4522 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4523 if (ret < 0) {
4524 ret = -ENOMEM;
4525 goto error;
4526 }
4527
4528 ret = create_lttng_rundir(rundir);
4529 if (ret < 0) {
4530 goto error;
4531 }
4532
4533 if (strlen(apps_unix_sock_path) == 0) {
4534 snprintf(apps_unix_sock_path, PATH_MAX,
4535 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4536 }
4537
4538 /* Set the cli tool unix socket path */
4539 if (strlen(client_unix_sock_path) == 0) {
4540 snprintf(client_unix_sock_path, PATH_MAX,
4541 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4542 }
4543
4544 /* Set global SHM for ust */
4545 if (strlen(wait_shm_path) == 0) {
4546 snprintf(wait_shm_path, PATH_MAX,
4547 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4548 }
4549
4550 /* Set health check Unix path */
4551 if (strlen(health_unix_sock_path) == 0) {
4552 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4553 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4554 }
4555 }
4556
4557 /* Set consumer initial state */
4558 kernel_consumerd_state = CONSUMER_STOPPED;
4559 ust_consumerd_state = CONSUMER_STOPPED;
4560
4561 DBG("Client socket path %s", client_unix_sock_path);
4562 DBG("Application socket path %s", apps_unix_sock_path);
4563 DBG("Application wait path %s", wait_shm_path);
4564 DBG("LTTng run directory path: %s", rundir);
4565
4566 /* 32 bits consumerd path setup */
4567 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4568 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4569 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4570 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4571
4572 DBG2("UST consumer 32 bits err path: %s",
4573 ustconsumer32_data.err_unix_sock_path);
4574 DBG2("UST consumer 32 bits cmd path: %s",
4575 ustconsumer32_data.cmd_unix_sock_path);
4576
4577 /* 64 bits consumerd path setup */
4578 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4579 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4580 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4581 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4582
4583 DBG2("UST consumer 64 bits err path: %s",
4584 ustconsumer64_data.err_unix_sock_path);
4585 DBG2("UST consumer 64 bits cmd path: %s",
4586 ustconsumer64_data.cmd_unix_sock_path);
4587
4588 /*
4589 * See if daemon already exist.
4590 */
4591 if ((ret = check_existing_daemon()) < 0) {
4592 ERR("Already running daemon.\n");
4593 /*
4594 * We do not goto exit because we must not cleanup()
4595 * because a daemon is already running.
4596 */
4597 goto error;
4598 }
4599
4600 /*
4601 * Init UST app hash table. Alloc hash table before this point since
4602 * cleanup() can get called after that point.
4603 */
4604 ust_app_ht_alloc();
4605
4606 /* After this point, we can safely call cleanup() with "goto exit" */
4607
4608 /*
4609 * These actions must be executed as root. We do that *after* setting up
4610 * the sockets path because we MUST make the check for another daemon using
4611 * those paths *before* trying to set the kernel consumer sockets and init
4612 * kernel tracer.
4613 */
4614 if (is_root) {
4615 ret = set_consumer_sockets(&kconsumer_data, rundir);
4616 if (ret < 0) {
4617 goto exit;
4618 }
4619
4620 /* Setup kernel tracer */
4621 if (!opt_no_kernel) {
4622 init_kernel_tracer();
4623 }
4624
4625 /* Set ulimit for open files */
4626 set_ulimit();
4627 }
4628 /* init lttng_fd tracking must be done after set_ulimit. */
4629 lttng_fd_init();
4630
4631 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4632 if (ret < 0) {
4633 goto exit;
4634 }
4635
4636 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4637 if (ret < 0) {
4638 goto exit;
4639 }
4640
4641 if ((ret = set_signal_handler()) < 0) {
4642 goto exit;
4643 }
4644
4645 /* Setup the needed unix socket */
4646 if ((ret = init_daemon_socket()) < 0) {
4647 goto exit;
4648 }
4649
4650 /* Set credentials to socket */
4651 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4652 goto exit;
4653 }
4654
4655 /* Get parent pid if -S, --sig-parent is specified. */
4656 if (opt_sig_parent) {
4657 ppid = getppid();
4658 }
4659
4660 /* Setup the kernel pipe for waking up the kernel thread */
4661 if (is_root && !opt_no_kernel) {
4662 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4663 goto exit;
4664 }
4665 }
4666
4667 /* Setup the thread ht_cleanup communication pipe. */
4668 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4669 goto exit;
4670 }
4671
4672 /* Setup the thread apps communication pipe. */
4673 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4674 goto exit;
4675 }
4676
4677 /* Setup the thread apps notify communication pipe. */
4678 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4679 goto exit;
4680 }
4681
4682 /* Initialize global buffer per UID and PID registry. */
4683 buffer_reg_init_uid_registry();
4684 buffer_reg_init_pid_registry();
4685
4686 /* Init UST command queue. */
4687 cds_wfq_init(&ust_cmd_queue.queue);
4688
4689 /*
4690 * Get session list pointer. This pointer MUST NOT be free(). This list is
4691 * statically declared in session.c
4692 */
4693 session_list_ptr = session_get_list();
4694
4695 /* Set up max poll set size */
4696 lttng_poll_set_max_size();
4697
4698 cmd_init();
4699
4700 /* Check for the application socket timeout env variable. */
4701 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4702 if (env_app_timeout) {
4703 app_socket_timeout = atoi(env_app_timeout);
4704 } else {
4705 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4706 }
4707
4708 write_pidfile();
4709
4710 /* Initialize communication library */
4711 lttcomm_init();
4712 /* This is to get the TCP timeout value. */
4713 lttcomm_inet_init();
4714
4715 /*
4716 * Initialize the health check subsystem. This call should set the
4717 * appropriate time values.
4718 */
4719 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
4720 if (!health_sessiond) {
4721 PERROR("health_app_create error");
4722 goto exit_health_sessiond_cleanup;
4723 }
4724
4725 /* Create thread to manage the client socket */
4726 ret = pthread_create(&ht_cleanup_thread, NULL,
4727 thread_ht_cleanup, (void *) NULL);
4728 if (ret != 0) {
4729 PERROR("pthread_create ht_cleanup");
4730 goto exit_ht_cleanup;
4731 }
4732
4733 /* Create thread to manage the client socket */
4734 ret = pthread_create(&health_thread, NULL,
4735 thread_manage_health, (void *) NULL);
4736 if (ret != 0) {
4737 PERROR("pthread_create health");
4738 goto exit_health;
4739 }
4740
4741 /* Create thread to manage the client socket */
4742 ret = pthread_create(&client_thread, NULL,
4743 thread_manage_clients, (void *) NULL);
4744 if (ret != 0) {
4745 PERROR("pthread_create clients");
4746 goto exit_client;
4747 }
4748
4749 /* Create thread to dispatch registration */
4750 ret = pthread_create(&dispatch_thread, NULL,
4751 thread_dispatch_ust_registration, (void *) NULL);
4752 if (ret != 0) {
4753 PERROR("pthread_create dispatch");
4754 goto exit_dispatch;
4755 }
4756
4757 /* Create thread to manage application registration. */
4758 ret = pthread_create(&reg_apps_thread, NULL,
4759 thread_registration_apps, (void *) NULL);
4760 if (ret != 0) {
4761 PERROR("pthread_create registration");
4762 goto exit_reg_apps;
4763 }
4764
4765 /* Create thread to manage application socket */
4766 ret = pthread_create(&apps_thread, NULL,
4767 thread_manage_apps, (void *) NULL);
4768 if (ret != 0) {
4769 PERROR("pthread_create apps");
4770 goto exit_apps;
4771 }
4772
4773 /* Create thread to manage application notify socket */
4774 ret = pthread_create(&apps_notify_thread, NULL,
4775 ust_thread_manage_notify, (void *) NULL);
4776 if (ret != 0) {
4777 PERROR("pthread_create apps");
4778 goto exit_apps_notify;
4779 }
4780
4781 /* Don't start this thread if kernel tracing is not requested nor root */
4782 if (is_root && !opt_no_kernel) {
4783 /* Create kernel thread to manage kernel event */
4784 ret = pthread_create(&kernel_thread, NULL,
4785 thread_manage_kernel, (void *) NULL);
4786 if (ret != 0) {
4787 PERROR("pthread_create kernel");
4788 goto exit_kernel;
4789 }
4790
4791 ret = pthread_join(kernel_thread, &status);
4792 if (ret != 0) {
4793 PERROR("pthread_join");
4794 goto error; /* join error, exit without cleanup */
4795 }
4796 }
4797
4798 exit_kernel:
4799 ret = pthread_join(apps_notify_thread, &status);
4800 if (ret != 0) {
4801 PERROR("pthread_join apps notify");
4802 goto error; /* join error, exit without cleanup */
4803 }
4804
4805 exit_apps_notify:
4806 ret = pthread_join(apps_thread, &status);
4807 if (ret != 0) {
4808 PERROR("pthread_join apps");
4809 goto error; /* join error, exit without cleanup */
4810 }
4811
4812
4813 exit_apps:
4814 ret = pthread_join(reg_apps_thread, &status);
4815 if (ret != 0) {
4816 PERROR("pthread_join");
4817 goto error; /* join error, exit without cleanup */
4818 }
4819
4820 exit_reg_apps:
4821 ret = pthread_join(dispatch_thread, &status);
4822 if (ret != 0) {
4823 PERROR("pthread_join");
4824 goto error; /* join error, exit without cleanup */
4825 }
4826
4827 exit_dispatch:
4828 ret = pthread_join(client_thread, &status);
4829 if (ret != 0) {
4830 PERROR("pthread_join");
4831 goto error; /* join error, exit without cleanup */
4832 }
4833
4834 ret = join_consumer_thread(&kconsumer_data);
4835 if (ret != 0) {
4836 PERROR("join_consumer");
4837 goto error; /* join error, exit without cleanup */
4838 }
4839
4840 ret = join_consumer_thread(&ustconsumer32_data);
4841 if (ret != 0) {
4842 PERROR("join_consumer ust32");
4843 goto error; /* join error, exit without cleanup */
4844 }
4845
4846 ret = join_consumer_thread(&ustconsumer64_data);
4847 if (ret != 0) {
4848 PERROR("join_consumer ust64");
4849 goto error; /* join error, exit without cleanup */
4850 }
4851
4852 exit_client:
4853 ret = pthread_join(health_thread, &status);
4854 if (ret != 0) {
4855 PERROR("pthread_join health thread");
4856 goto error; /* join error, exit without cleanup */
4857 }
4858
4859 exit_health:
4860 ret = pthread_join(ht_cleanup_thread, &status);
4861 if (ret != 0) {
4862 PERROR("pthread_join ht cleanup thread");
4863 goto error; /* join error, exit without cleanup */
4864 }
4865 exit_ht_cleanup:
4866 health_app_destroy(health_sessiond);
4867 exit_health_sessiond_cleanup:
4868 exit:
4869 /*
4870 * cleanup() is called when no other thread is running.
4871 */
4872 rcu_thread_online();
4873 cleanup();
4874 rcu_thread_offline();
4875 rcu_unregister_thread();
4876 if (!ret) {
4877 exit(EXIT_SUCCESS);
4878 }
4879 error:
4880 exit(EXIT_FAILURE);
4881 }
This page took 0.181127 seconds and 5 git commands to generate.