Fix: sessiond write() to handle EINTR
[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 <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/uatomic.h>
36 #include <unistd.h>
37 #include <config.h>
38
39 #include <common/common.h>
40 #include <common/compat/poll.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 "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "kernel.h"
55 #include "kernel-consumer.h"
56 #include "modprobe.h"
57 #include "shm.h"
58 #include "ust-ctl.h"
59 #include "ust-consumer.h"
60 #include "utils.h"
61 #include "fd-limit.h"
62 #include "health.h"
63 #include "testpoint.h"
64
65 #define CONSUMERD_FILE "lttng-consumerd"
66
67 /* Const values */
68 const char default_home_dir[] = DEFAULT_HOME_DIR;
69 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
70 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
71 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
72
73 const char *progname;
74 const char *opt_tracing_group;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static char *rundir;
82
83 /*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
87 static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
91 .err_sock = -1,
92 .cmd_sock = -1,
93 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
94 .lock = PTHREAD_MUTEX_INITIALIZER,
95 .cond = PTHREAD_COND_INITIALIZER,
96 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
97 };
98 static struct consumer_data ustconsumer64_data = {
99 .type = LTTNG_CONSUMER64_UST,
100 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
101 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
102 .err_sock = -1,
103 .cmd_sock = -1,
104 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
105 .lock = PTHREAD_MUTEX_INITIALIZER,
106 .cond = PTHREAD_COND_INITIALIZER,
107 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
108 };
109 static struct consumer_data ustconsumer32_data = {
110 .type = LTTNG_CONSUMER32_UST,
111 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
112 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
116 .lock = PTHREAD_MUTEX_INITIALIZER,
117 .cond = PTHREAD_COND_INITIALIZER,
118 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
119 };
120
121 /* Shared between threads */
122 static int dispatch_thread_exit;
123
124 /* Global application Unix socket path */
125 static char apps_unix_sock_path[PATH_MAX];
126 /* Global client Unix socket path */
127 static char client_unix_sock_path[PATH_MAX];
128 /* global wait shm path for UST */
129 static char wait_shm_path[PATH_MAX];
130 /* Global health check unix path */
131 static char health_unix_sock_path[PATH_MAX];
132
133 /* Sockets and FDs */
134 static int client_sock = -1;
135 static int apps_sock = -1;
136 int kernel_tracer_fd = -1;
137 static int kernel_poll_pipe[2] = { -1, -1 };
138
139 /*
140 * Quit pipe for all threads. This permits a single cancellation point
141 * for all threads when receiving an event on the pipe.
142 */
143 static int thread_quit_pipe[2] = { -1, -1 };
144
145 /*
146 * This pipe is used to inform the thread managing application communication
147 * that a command is queued and ready to be processed.
148 */
149 static int apps_cmd_pipe[2] = { -1, -1 };
150
151 /* Pthread, Mutexes and Semaphores */
152 static pthread_t apps_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
159 /*
160 * UST registration command queue. This queue is tied with a futex and uses a N
161 * wakers / 1 waiter implemented and detailed in futex.c/.h
162 *
163 * The thread_manage_apps and thread_dispatch_ust_registration interact with
164 * this queue and the wait/wake scheme.
165 */
166 static struct ust_cmd_queue ust_cmd_queue;
167
168 /*
169 * Pointer initialized before thread creation.
170 *
171 * This points to the tracing session list containing the session count and a
172 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
173 * MUST NOT be taken if you call a public function in session.c.
174 *
175 * The lock is nested inside the structure: session_list_ptr->lock. Please use
176 * session_lock_list and session_unlock_list for lock acquisition.
177 */
178 static struct ltt_session_list *session_list_ptr;
179
180 int ust_consumerd64_fd = -1;
181 int ust_consumerd32_fd = -1;
182
183 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
184 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
185 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
186 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
187
188 static const char *module_proc_lttng = "/proc/lttng";
189
190 /*
191 * Consumer daemon state which is changed when spawning it, killing it or in
192 * case of a fatal error.
193 */
194 enum consumerd_state {
195 CONSUMER_STARTED = 1,
196 CONSUMER_STOPPED = 2,
197 CONSUMER_ERROR = 3,
198 };
199
200 /*
201 * This consumer daemon state is used to validate if a client command will be
202 * able to reach the consumer. If not, the client is informed. For instance,
203 * doing a "lttng start" when the consumer state is set to ERROR will return an
204 * error to the client.
205 *
206 * The following example shows a possible race condition of this scheme:
207 *
208 * consumer thread error happens
209 * client cmd arrives
210 * client cmd checks state -> still OK
211 * consumer thread exit, sets error
212 * client cmd try to talk to consumer
213 * ...
214 *
215 * However, since the consumer is a different daemon, we have no way of making
216 * sure the command will reach it safely even with this state flag. This is why
217 * we consider that up to the state validation during command processing, the
218 * command is safe. After that, we can not guarantee the correctness of the
219 * client request vis-a-vis the consumer.
220 */
221 static enum consumerd_state ust_consumerd_state;
222 static enum consumerd_state kernel_consumerd_state;
223
224 /* Used for the health monitoring of the session daemon. See health.h */
225 struct health_state health_thread_cmd;
226 struct health_state health_thread_app_manage;
227 struct health_state health_thread_app_reg;
228 struct health_state health_thread_kernel;
229
230 /*
231 * Socket timeout for receiving and sending in seconds.
232 */
233 static int app_socket_timeout;
234
235 static
236 void setup_consumerd_path(void)
237 {
238 const char *bin, *libdir;
239
240 /*
241 * Allow INSTALL_BIN_PATH to be used as a target path for the
242 * native architecture size consumer if CONFIG_CONSUMER*_PATH
243 * has not been defined.
244 */
245 #if (CAA_BITS_PER_LONG == 32)
246 if (!consumerd32_bin[0]) {
247 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
248 }
249 if (!consumerd32_libdir[0]) {
250 consumerd32_libdir = INSTALL_LIB_PATH;
251 }
252 #elif (CAA_BITS_PER_LONG == 64)
253 if (!consumerd64_bin[0]) {
254 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
255 }
256 if (!consumerd64_libdir[0]) {
257 consumerd64_libdir = INSTALL_LIB_PATH;
258 }
259 #else
260 #error "Unknown bitness"
261 #endif
262
263 /*
264 * runtime env. var. overrides the build default.
265 */
266 bin = getenv("LTTNG_CONSUMERD32_BIN");
267 if (bin) {
268 consumerd32_bin = bin;
269 }
270 bin = getenv("LTTNG_CONSUMERD64_BIN");
271 if (bin) {
272 consumerd64_bin = bin;
273 }
274 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
275 if (libdir) {
276 consumerd32_libdir = libdir;
277 }
278 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
279 if (libdir) {
280 consumerd64_libdir = libdir;
281 }
282 }
283
284 /*
285 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
286 */
287 static int create_thread_poll_set(struct lttng_poll_event *events,
288 unsigned int size)
289 {
290 int ret;
291
292 if (events == NULL || size == 0) {
293 ret = -1;
294 goto error;
295 }
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310 error:
311 return ret;
312 }
313
314 /*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
319 static int check_thread_quit_pipe(int fd, uint32_t events)
320 {
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326 }
327
328 /*
329 * Return group ID of the tracing group or -1 if not found.
330 */
331 static gid_t allowed_group(void)
332 {
333 struct group *grp;
334
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345 }
346
347 /*
348 * Init thread quit pipe.
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352 static int init_thread_quit_pipe(void)
353 {
354 int ret, i;
355
356 ret = pipe(thread_quit_pipe);
357 if (ret < 0) {
358 PERROR("thread quit pipe");
359 goto error;
360 }
361
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
370 error:
371 return ret;
372 }
373
374 /*
375 * Stop all threads by closing the thread quit pipe.
376 */
377 static void stop_threads(void)
378 {
379 int ret;
380
381 /* Stopping all threads */
382 DBG("Terminating all threads");
383 ret = notify_thread_pipe(thread_quit_pipe[1]);
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
388 /* Dispatch thread */
389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
390 futex_nto1_wake(&ust_cmd_queue.futex);
391 }
392
393 /*
394 * Cleanup the daemon
395 */
396 static void cleanup(void)
397 {
398 int ret;
399 char *cmd = NULL;
400 struct ltt_session *sess, *stmp;
401
402 DBG("Cleaning up");
403
404 /* First thing first, stop all threads */
405 utils_close_pipe(thread_quit_pipe);
406
407 DBG("Removing %s directory", rundir);
408 ret = asprintf(&cmd, "rm -rf %s", rundir);
409 if (ret < 0) {
410 ERR("asprintf failed. Something is really wrong!");
411 }
412
413 /* Remove lttng run directory */
414 ret = system(cmd);
415 if (ret < 0) {
416 ERR("Unable to clean %s", rundir);
417 }
418 free(cmd);
419 free(rundir);
420
421 DBG("Cleaning up all sessions");
422
423 /* Destroy session list mutex */
424 if (session_list_ptr != NULL) {
425 pthread_mutex_destroy(&session_list_ptr->lock);
426
427 /* Cleanup ALL session */
428 cds_list_for_each_entry_safe(sess, stmp,
429 &session_list_ptr->head, list) {
430 cmd_destroy_session(sess, kernel_poll_pipe[1]);
431 }
432 }
433
434 DBG("Closing all UST sockets");
435 ust_app_clean_list();
436
437 if (is_root && !opt_no_kernel) {
438 DBG2("Closing kernel fd");
439 if (kernel_tracer_fd >= 0) {
440 ret = close(kernel_tracer_fd);
441 if (ret) {
442 PERROR("close");
443 }
444 }
445 DBG("Unloading kernel modules");
446 modprobe_remove_lttng_all();
447 }
448
449 /* <fun> */
450 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
451 "Matthew, BEET driven development works!%c[%dm",
452 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
453 /* </fun> */
454 }
455
456 /*
457 * Send data on a unix socket using the liblttsessiondcomm API.
458 *
459 * Return lttcomm error code.
460 */
461 static int send_unix_sock(int sock, void *buf, size_t len)
462 {
463 /* Check valid length */
464 if (len == 0) {
465 return -1;
466 }
467
468 return lttcomm_send_unix_sock(sock, buf, len);
469 }
470
471 /*
472 * Free memory of a command context structure.
473 */
474 static void clean_command_ctx(struct command_ctx **cmd_ctx)
475 {
476 DBG("Clean command context structure");
477 if (*cmd_ctx) {
478 if ((*cmd_ctx)->llm) {
479 free((*cmd_ctx)->llm);
480 }
481 if ((*cmd_ctx)->lsm) {
482 free((*cmd_ctx)->lsm);
483 }
484 free(*cmd_ctx);
485 *cmd_ctx = NULL;
486 }
487 }
488
489 /*
490 * Notify UST applications using the shm mmap futex.
491 */
492 static int notify_ust_apps(int active)
493 {
494 char *wait_shm_mmap;
495
496 DBG("Notifying applications of session daemon state: %d", active);
497
498 /* See shm.c for this call implying mmap, shm and futex calls */
499 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
500 if (wait_shm_mmap == NULL) {
501 goto error;
502 }
503
504 /* Wake waiting process */
505 futex_wait_update((int32_t *) wait_shm_mmap, active);
506
507 /* Apps notified successfully */
508 return 0;
509
510 error:
511 return -1;
512 }
513
514 /*
515 * Setup the outgoing data buffer for the response (llm) by allocating the
516 * right amount of memory and copying the original information from the lsm
517 * structure.
518 *
519 * Return total size of the buffer pointed by buf.
520 */
521 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
522 {
523 int ret, buf_size;
524
525 buf_size = size;
526
527 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
528 if (cmd_ctx->llm == NULL) {
529 PERROR("zmalloc");
530 ret = -ENOMEM;
531 goto error;
532 }
533
534 /* Copy common data */
535 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
536 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
537
538 cmd_ctx->llm->data_size = size;
539 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
540
541 return buf_size;
542
543 error:
544 return ret;
545 }
546
547 /*
548 * Update the kernel poll set of all channel fd available over all tracing
549 * session. Add the wakeup pipe at the end of the set.
550 */
551 static int update_kernel_poll(struct lttng_poll_event *events)
552 {
553 int ret;
554 struct ltt_session *session;
555 struct ltt_kernel_channel *channel;
556
557 DBG("Updating kernel poll set");
558
559 session_lock_list();
560 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
561 session_lock(session);
562 if (session->kernel_session == NULL) {
563 session_unlock(session);
564 continue;
565 }
566
567 cds_list_for_each_entry(channel,
568 &session->kernel_session->channel_list.head, list) {
569 /* Add channel fd to the kernel poll set */
570 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
571 if (ret < 0) {
572 session_unlock(session);
573 goto error;
574 }
575 DBG("Channel fd %d added to kernel set", channel->fd);
576 }
577 session_unlock(session);
578 }
579 session_unlock_list();
580
581 return 0;
582
583 error:
584 session_unlock_list();
585 return -1;
586 }
587
588 /*
589 * Find the channel fd from 'fd' over all tracing session. When found, check
590 * for new channel stream and send those stream fds to the kernel consumer.
591 *
592 * Useful for CPU hotplug feature.
593 */
594 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
595 {
596 int ret = 0;
597 struct ltt_session *session;
598 struct ltt_kernel_session *ksess;
599 struct ltt_kernel_channel *channel;
600
601 DBG("Updating kernel streams for channel fd %d", fd);
602
603 session_lock_list();
604 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
605 session_lock(session);
606 if (session->kernel_session == NULL) {
607 session_unlock(session);
608 continue;
609 }
610 ksess = session->kernel_session;
611
612 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
613 if (channel->fd == fd) {
614 DBG("Channel found, updating kernel streams");
615 ret = kernel_open_channel_stream(channel);
616 if (ret < 0) {
617 goto error;
618 }
619
620 /*
621 * Have we already sent fds to the consumer? If yes, it means
622 * that tracing is started so it is safe to send our updated
623 * stream fds.
624 */
625 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
626 struct lttng_ht_iter iter;
627 struct consumer_socket *socket;
628
629
630 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
631 &iter.iter, socket, node.node) {
632 /* Code flow error */
633 assert(socket->fd >= 0);
634
635 pthread_mutex_lock(socket->lock);
636 ret = kernel_consumer_send_channel_stream(socket,
637 channel, ksess);
638 pthread_mutex_unlock(socket->lock);
639 if (ret < 0) {
640 goto error;
641 }
642 }
643 }
644 goto error;
645 }
646 }
647 session_unlock(session);
648 }
649 session_unlock_list();
650 return ret;
651
652 error:
653 session_unlock(session);
654 session_unlock_list();
655 return ret;
656 }
657
658 /*
659 * For each tracing session, update newly registered apps.
660 */
661 static void update_ust_app(int app_sock)
662 {
663 struct ltt_session *sess, *stmp;
664
665 session_lock_list();
666
667 /* For all tracing session(s) */
668 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
669 session_lock(sess);
670 if (sess->ust_session) {
671 ust_app_global_update(sess->ust_session, app_sock);
672 }
673 session_unlock(sess);
674 }
675
676 session_unlock_list();
677 }
678
679 /*
680 * This thread manage event coming from the kernel.
681 *
682 * Features supported in this thread:
683 * -) CPU Hotplug
684 */
685 static void *thread_manage_kernel(void *data)
686 {
687 int ret, i, pollfd, update_poll_flag = 1, err = -1;
688 uint32_t revents, nb_fd;
689 char tmp;
690 struct lttng_poll_event events;
691
692 DBG("[thread] Thread manage kernel started");
693
694 /*
695 * This first step of the while is to clean this structure which could free
696 * non NULL pointers so zero it before the loop.
697 */
698 memset(&events, 0, sizeof(events));
699
700 if (testpoint(thread_manage_kernel)) {
701 goto error_testpoint;
702 }
703
704 health_code_update(&health_thread_kernel);
705
706 if (testpoint(thread_manage_kernel_before_loop)) {
707 goto error_testpoint;
708 }
709
710 while (1) {
711 health_code_update(&health_thread_kernel);
712
713 if (update_poll_flag == 1) {
714 /* Clean events object. We are about to populate it again. */
715 lttng_poll_clean(&events);
716
717 ret = create_thread_poll_set(&events, 2);
718 if (ret < 0) {
719 goto error_poll_create;
720 }
721
722 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
723 if (ret < 0) {
724 goto error;
725 }
726
727 /* This will add the available kernel channel if any. */
728 ret = update_kernel_poll(&events);
729 if (ret < 0) {
730 goto error;
731 }
732 update_poll_flag = 0;
733 }
734
735 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
736
737 /* Poll infinite value of time */
738 restart:
739 health_poll_update(&health_thread_kernel);
740 ret = lttng_poll_wait(&events, -1);
741 health_poll_update(&health_thread_kernel);
742 if (ret < 0) {
743 /*
744 * Restart interrupted system call.
745 */
746 if (errno == EINTR) {
747 goto restart;
748 }
749 goto error;
750 } else if (ret == 0) {
751 /* Should not happen since timeout is infinite */
752 ERR("Return value of poll is 0 with an infinite timeout.\n"
753 "This should not have happened! Continuing...");
754 continue;
755 }
756
757 nb_fd = ret;
758
759 for (i = 0; i < nb_fd; i++) {
760 /* Fetch once the poll data */
761 revents = LTTNG_POLL_GETEV(&events, i);
762 pollfd = LTTNG_POLL_GETFD(&events, i);
763
764 health_code_update(&health_thread_kernel);
765
766 /* Thread quit pipe has been closed. Killing thread. */
767 ret = check_thread_quit_pipe(pollfd, revents);
768 if (ret) {
769 err = 0;
770 goto exit;
771 }
772
773 /* Check for data on kernel pipe */
774 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
775 do {
776 ret = read(kernel_poll_pipe[0], &tmp, 1);
777 } while (ret < 0 && errno == EINTR);
778 /*
779 * Ret value is useless here, if this pipe gets any actions an
780 * update is required anyway.
781 */
782 update_poll_flag = 1;
783 continue;
784 } else {
785 /*
786 * New CPU detected by the kernel. Adding kernel stream to
787 * kernel session and updating the kernel consumer
788 */
789 if (revents & LPOLLIN) {
790 ret = update_kernel_stream(&kconsumer_data, pollfd);
791 if (ret < 0) {
792 continue;
793 }
794 break;
795 /*
796 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
797 * and unregister kernel stream at this point.
798 */
799 }
800 }
801 }
802 }
803
804 exit:
805 error:
806 lttng_poll_clean(&events);
807 error_poll_create:
808 error_testpoint:
809 utils_close_pipe(kernel_poll_pipe);
810 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
811 if (err) {
812 health_error(&health_thread_kernel);
813 ERR("Health error occurred in %s", __func__);
814 WARN("Kernel thread died unexpectedly. "
815 "Kernel tracing can continue but CPU hotplug is disabled.");
816 }
817 health_exit(&health_thread_kernel);
818 DBG("Kernel thread dying");
819 return NULL;
820 }
821
822 /*
823 * Signal pthread condition of the consumer data that the thread.
824 */
825 static void signal_consumer_condition(struct consumer_data *data, int state)
826 {
827 pthread_mutex_lock(&data->cond_mutex);
828
829 /*
830 * The state is set before signaling. It can be any value, it's the waiter
831 * job to correctly interpret this condition variable associated to the
832 * consumer pthread_cond.
833 *
834 * A value of 0 means that the corresponding thread of the consumer data
835 * was not started. 1 indicates that the thread has started and is ready
836 * for action. A negative value means that there was an error during the
837 * thread bootstrap.
838 */
839 data->consumer_thread_is_ready = state;
840 (void) pthread_cond_signal(&data->cond);
841
842 pthread_mutex_unlock(&data->cond_mutex);
843 }
844
845 /*
846 * This thread manage the consumer error sent back to the session daemon.
847 */
848 static void *thread_manage_consumer(void *data)
849 {
850 int sock = -1, i, ret, pollfd, err = -1;
851 uint32_t revents, nb_fd;
852 enum lttcomm_return_code code;
853 struct lttng_poll_event events;
854 struct consumer_data *consumer_data = data;
855
856 DBG("[thread] Manage consumer started");
857
858 /*
859 * Since the consumer thread can be spawned at any moment in time, we init
860 * the health to a poll status (1, which is a valid health over time).
861 * When the thread starts, we update here the health to a "code" path being
862 * an even value so this thread, when reaching a poll wait, does not
863 * trigger an error with an even value.
864 *
865 * Here is the use case we avoid.
866 *
867 * +1: the first poll update during initialization (main())
868 * +2 * x: multiple code update once in this thread.
869 * +1: poll wait in this thread (being a good health state).
870 * == even number which after the wait period shows as a bad health.
871 *
872 * In a nutshell, the following poll update to the health state brings back
873 * the state to an even value meaning a code path.
874 */
875 health_poll_update(&consumer_data->health);
876
877 /*
878 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
879 * Nothing more will be added to this poll set.
880 */
881 ret = create_thread_poll_set(&events, 2);
882 if (ret < 0) {
883 goto error_poll;
884 }
885
886 /*
887 * The error socket here is already in a listening state which was done
888 * just before spawning this thread to avoid a race between the consumer
889 * daemon exec trying to connect and the listen() call.
890 */
891 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
892 if (ret < 0) {
893 goto error;
894 }
895
896 health_code_update(&consumer_data->health);
897
898 /* Inifinite blocking call, waiting for transmission */
899 restart:
900 health_poll_update(&consumer_data->health);
901
902 if (testpoint(thread_manage_consumer)) {
903 goto error;
904 }
905
906 ret = lttng_poll_wait(&events, -1);
907 health_poll_update(&consumer_data->health);
908 if (ret < 0) {
909 /*
910 * Restart interrupted system call.
911 */
912 if (errno == EINTR) {
913 goto restart;
914 }
915 goto error;
916 }
917
918 nb_fd = ret;
919
920 for (i = 0; i < nb_fd; i++) {
921 /* Fetch once the poll data */
922 revents = LTTNG_POLL_GETEV(&events, i);
923 pollfd = LTTNG_POLL_GETFD(&events, i);
924
925 health_code_update(&consumer_data->health);
926
927 /* Thread quit pipe has been closed. Killing thread. */
928 ret = check_thread_quit_pipe(pollfd, revents);
929 if (ret) {
930 err = 0;
931 goto exit;
932 }
933
934 /* Event on the registration socket */
935 if (pollfd == consumer_data->err_sock) {
936 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
937 ERR("consumer err socket poll error");
938 goto error;
939 }
940 }
941 }
942
943 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
944 if (sock < 0) {
945 goto error;
946 }
947
948 /*
949 * Set the CLOEXEC flag. Return code is useless because either way, the
950 * show must go on.
951 */
952 (void) utils_set_fd_cloexec(sock);
953
954 health_code_update(&consumer_data->health);
955
956 DBG2("Receiving code from consumer err_sock");
957
958 /* Getting status code from kconsumerd */
959 ret = lttcomm_recv_unix_sock(sock, &code,
960 sizeof(enum lttcomm_return_code));
961 if (ret <= 0) {
962 goto error;
963 }
964
965 health_code_update(&consumer_data->health);
966
967 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
968 consumer_data->cmd_sock =
969 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
970 if (consumer_data->cmd_sock < 0) {
971 /* On error, signal condition and quit. */
972 signal_consumer_condition(consumer_data, -1);
973 PERROR("consumer connect");
974 goto error;
975 }
976 signal_consumer_condition(consumer_data, 1);
977 DBG("Consumer command socket ready");
978 } else {
979 ERR("consumer error when waiting for SOCK_READY : %s",
980 lttcomm_get_readable_code(-code));
981 goto error;
982 }
983
984 /* Remove the kconsumerd error sock since we've established a connexion */
985 ret = lttng_poll_del(&events, consumer_data->err_sock);
986 if (ret < 0) {
987 goto error;
988 }
989
990 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
991 if (ret < 0) {
992 goto error;
993 }
994
995 health_code_update(&consumer_data->health);
996
997 /* Inifinite blocking call, waiting for transmission */
998 restart_poll:
999 health_poll_update(&consumer_data->health);
1000 ret = lttng_poll_wait(&events, -1);
1001 health_poll_update(&consumer_data->health);
1002 if (ret < 0) {
1003 /*
1004 * Restart interrupted system call.
1005 */
1006 if (errno == EINTR) {
1007 goto restart_poll;
1008 }
1009 goto error;
1010 }
1011
1012 nb_fd = ret;
1013
1014 for (i = 0; i < nb_fd; i++) {
1015 /* Fetch once the poll data */
1016 revents = LTTNG_POLL_GETEV(&events, i);
1017 pollfd = LTTNG_POLL_GETFD(&events, i);
1018
1019 health_code_update(&consumer_data->health);
1020
1021 /* Thread quit pipe has been closed. Killing thread. */
1022 ret = check_thread_quit_pipe(pollfd, revents);
1023 if (ret) {
1024 err = 0;
1025 goto exit;
1026 }
1027
1028 /* Event on the kconsumerd socket */
1029 if (pollfd == sock) {
1030 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1031 ERR("consumer err socket second poll error");
1032 goto error;
1033 }
1034 }
1035 }
1036
1037 health_code_update(&consumer_data->health);
1038
1039 /* Wait for any kconsumerd error */
1040 ret = lttcomm_recv_unix_sock(sock, &code,
1041 sizeof(enum lttcomm_return_code));
1042 if (ret <= 0) {
1043 ERR("consumer closed the command socket");
1044 goto error;
1045 }
1046
1047 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1048
1049 exit:
1050 error:
1051 /* Immediately set the consumerd state to stopped */
1052 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1053 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1054 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1055 consumer_data->type == LTTNG_CONSUMER32_UST) {
1056 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1057 } else {
1058 /* Code flow error... */
1059 assert(0);
1060 }
1061
1062 if (consumer_data->err_sock >= 0) {
1063 ret = close(consumer_data->err_sock);
1064 if (ret) {
1065 PERROR("close");
1066 }
1067 }
1068 if (consumer_data->cmd_sock >= 0) {
1069 ret = close(consumer_data->cmd_sock);
1070 if (ret) {
1071 PERROR("close");
1072 }
1073 }
1074 if (sock >= 0) {
1075 ret = close(sock);
1076 if (ret) {
1077 PERROR("close");
1078 }
1079 }
1080
1081 unlink(consumer_data->err_unix_sock_path);
1082 unlink(consumer_data->cmd_unix_sock_path);
1083 consumer_data->pid = 0;
1084
1085 lttng_poll_clean(&events);
1086 error_poll:
1087 if (err) {
1088 health_error(&consumer_data->health);
1089 ERR("Health error occurred in %s", __func__);
1090 }
1091 health_exit(&consumer_data->health);
1092 DBG("consumer thread cleanup completed");
1093
1094 return NULL;
1095 }
1096
1097 /*
1098 * This thread manage application communication.
1099 */
1100 static void *thread_manage_apps(void *data)
1101 {
1102 int i, ret, pollfd, err = -1;
1103 uint32_t revents, nb_fd;
1104 struct ust_command ust_cmd;
1105 struct lttng_poll_event events;
1106
1107 DBG("[thread] Manage application started");
1108
1109 rcu_register_thread();
1110 rcu_thread_online();
1111
1112 if (testpoint(thread_manage_apps)) {
1113 goto error_testpoint;
1114 }
1115
1116 health_code_update(&health_thread_app_manage);
1117
1118 ret = create_thread_poll_set(&events, 2);
1119 if (ret < 0) {
1120 goto error_poll_create;
1121 }
1122
1123 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1124 if (ret < 0) {
1125 goto error;
1126 }
1127
1128 if (testpoint(thread_manage_apps_before_loop)) {
1129 goto error;
1130 }
1131
1132 health_code_update(&health_thread_app_manage);
1133
1134 while (1) {
1135 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1136
1137 /* Inifinite blocking call, waiting for transmission */
1138 restart:
1139 health_poll_update(&health_thread_app_manage);
1140 ret = lttng_poll_wait(&events, -1);
1141 health_poll_update(&health_thread_app_manage);
1142 if (ret < 0) {
1143 /*
1144 * Restart interrupted system call.
1145 */
1146 if (errno == EINTR) {
1147 goto restart;
1148 }
1149 goto error;
1150 }
1151
1152 nb_fd = ret;
1153
1154 for (i = 0; i < nb_fd; i++) {
1155 /* Fetch once the poll data */
1156 revents = LTTNG_POLL_GETEV(&events, i);
1157 pollfd = LTTNG_POLL_GETFD(&events, i);
1158
1159 health_code_update(&health_thread_app_manage);
1160
1161 /* Thread quit pipe has been closed. Killing thread. */
1162 ret = check_thread_quit_pipe(pollfd, revents);
1163 if (ret) {
1164 err = 0;
1165 goto exit;
1166 }
1167
1168 /* Inspect the apps cmd pipe */
1169 if (pollfd == apps_cmd_pipe[0]) {
1170 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1171 ERR("Apps command pipe error");
1172 goto error;
1173 } else if (revents & LPOLLIN) {
1174 /* Empty pipe */
1175 do {
1176 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1177 } while (ret < 0 && errno == EINTR);
1178 if (ret < 0 || ret < sizeof(ust_cmd)) {
1179 PERROR("read apps cmd pipe");
1180 goto error;
1181 }
1182
1183 health_code_update(&health_thread_app_manage);
1184
1185 /* Register applicaton to the session daemon */
1186 ret = ust_app_register(&ust_cmd.reg_msg,
1187 ust_cmd.sock);
1188 if (ret == -ENOMEM) {
1189 goto error;
1190 } else if (ret < 0) {
1191 break;
1192 }
1193
1194 health_code_update(&health_thread_app_manage);
1195
1196 /*
1197 * Validate UST version compatibility.
1198 */
1199 ret = ust_app_validate_version(ust_cmd.sock);
1200 if (ret >= 0) {
1201 /*
1202 * Add channel(s) and event(s) to newly registered apps
1203 * from lttng global UST domain.
1204 */
1205 update_ust_app(ust_cmd.sock);
1206 }
1207
1208 health_code_update(&health_thread_app_manage);
1209
1210 ret = ust_app_register_done(ust_cmd.sock);
1211 if (ret < 0) {
1212 /*
1213 * If the registration is not possible, we simply
1214 * unregister the apps and continue
1215 */
1216 ust_app_unregister(ust_cmd.sock);
1217 } else {
1218 /*
1219 * We only monitor the error events of the socket. This
1220 * thread does not handle any incoming data from UST
1221 * (POLLIN).
1222 */
1223 ret = lttng_poll_add(&events, ust_cmd.sock,
1224 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1225 if (ret < 0) {
1226 goto error;
1227 }
1228
1229 /* Set socket timeout for both receiving and ending */
1230 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1231 app_socket_timeout);
1232 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1233 app_socket_timeout);
1234
1235 DBG("Apps with sock %d added to poll set",
1236 ust_cmd.sock);
1237 }
1238
1239 health_code_update(&health_thread_app_manage);
1240
1241 break;
1242 }
1243 } else {
1244 /*
1245 * At this point, we know that a registered application made
1246 * the event at poll_wait.
1247 */
1248 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1249 /* Removing from the poll set */
1250 ret = lttng_poll_del(&events, pollfd);
1251 if (ret < 0) {
1252 goto error;
1253 }
1254
1255 /* Socket closed on remote end. */
1256 ust_app_unregister(pollfd);
1257 break;
1258 }
1259 }
1260
1261 health_code_update(&health_thread_app_manage);
1262 }
1263 }
1264
1265 exit:
1266 error:
1267 lttng_poll_clean(&events);
1268 error_poll_create:
1269 error_testpoint:
1270 utils_close_pipe(apps_cmd_pipe);
1271 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1272
1273 /*
1274 * We don't clean the UST app hash table here since already registered
1275 * applications can still be controlled so let them be until the session
1276 * daemon dies or the applications stop.
1277 */
1278
1279 if (err) {
1280 health_error(&health_thread_app_manage);
1281 ERR("Health error occurred in %s", __func__);
1282 }
1283 health_exit(&health_thread_app_manage);
1284 DBG("Application communication apps thread cleanup complete");
1285 rcu_thread_offline();
1286 rcu_unregister_thread();
1287 return NULL;
1288 }
1289
1290 /*
1291 * Dispatch request from the registration threads to the application
1292 * communication thread.
1293 */
1294 static void *thread_dispatch_ust_registration(void *data)
1295 {
1296 int ret;
1297 struct cds_wfq_node *node;
1298 struct ust_command *ust_cmd = NULL;
1299
1300 DBG("[thread] Dispatch UST command started");
1301
1302 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1303 /* Atomically prepare the queue futex */
1304 futex_nto1_prepare(&ust_cmd_queue.futex);
1305
1306 do {
1307 /* Dequeue command for registration */
1308 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1309 if (node == NULL) {
1310 DBG("Woken up but nothing in the UST command queue");
1311 /* Continue thread execution */
1312 break;
1313 }
1314
1315 ust_cmd = caa_container_of(node, struct ust_command, node);
1316
1317 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1318 " gid:%d sock:%d name:%s (version %d.%d)",
1319 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1320 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1321 ust_cmd->sock, ust_cmd->reg_msg.name,
1322 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1323 /*
1324 * Inform apps thread of the new application registration. This
1325 * call is blocking so we can be assured that the data will be read
1326 * at some point in time or wait to the end of the world :)
1327 */
1328 if (apps_cmd_pipe[1] >= 0) {
1329 do {
1330 ret = write(apps_cmd_pipe[1], ust_cmd,
1331 sizeof(struct ust_command));
1332 } while (ret < 0 && errno == EINTR);
1333 if (ret < 0) {
1334 PERROR("write apps cmd pipe");
1335 if (errno == EBADF) {
1336 /*
1337 * We can't inform the application thread to process
1338 * registration. We will exit or else application
1339 * registration will not occur and tracing will never
1340 * start.
1341 */
1342 goto error;
1343 }
1344 }
1345 } else {
1346 /* Application manager thread is not available. */
1347 ret = close(ust_cmd->sock);
1348 if (ret < 0) {
1349 PERROR("close ust_cmd sock");
1350 }
1351 }
1352 free(ust_cmd);
1353 } while (node != NULL);
1354
1355 /* Futex wait on queue. Blocking call on futex() */
1356 futex_nto1_wait(&ust_cmd_queue.futex);
1357 }
1358
1359 error:
1360 DBG("Dispatch thread dying");
1361 return NULL;
1362 }
1363
1364 /*
1365 * This thread manage application registration.
1366 */
1367 static void *thread_registration_apps(void *data)
1368 {
1369 int sock = -1, i, ret, pollfd, err = -1;
1370 uint32_t revents, nb_fd;
1371 struct lttng_poll_event events;
1372 /*
1373 * Get allocated in this thread, enqueued to a global queue, dequeued and
1374 * freed in the manage apps thread.
1375 */
1376 struct ust_command *ust_cmd = NULL;
1377
1378 DBG("[thread] Manage application registration started");
1379
1380 if (testpoint(thread_registration_apps)) {
1381 goto error_testpoint;
1382 }
1383
1384 ret = lttcomm_listen_unix_sock(apps_sock);
1385 if (ret < 0) {
1386 goto error_listen;
1387 }
1388
1389 /*
1390 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1391 * more will be added to this poll set.
1392 */
1393 ret = create_thread_poll_set(&events, 2);
1394 if (ret < 0) {
1395 goto error_create_poll;
1396 }
1397
1398 /* Add the application registration socket */
1399 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1400 if (ret < 0) {
1401 goto error_poll_add;
1402 }
1403
1404 /* Notify all applications to register */
1405 ret = notify_ust_apps(1);
1406 if (ret < 0) {
1407 ERR("Failed to notify applications or create the wait shared memory.\n"
1408 "Execution continues but there might be problem for already\n"
1409 "running applications that wishes to register.");
1410 }
1411
1412 while (1) {
1413 DBG("Accepting application registration");
1414
1415 /* Inifinite blocking call, waiting for transmission */
1416 restart:
1417 health_poll_update(&health_thread_app_reg);
1418 ret = lttng_poll_wait(&events, -1);
1419 health_poll_update(&health_thread_app_reg);
1420 if (ret < 0) {
1421 /*
1422 * Restart interrupted system call.
1423 */
1424 if (errno == EINTR) {
1425 goto restart;
1426 }
1427 goto error;
1428 }
1429
1430 nb_fd = ret;
1431
1432 for (i = 0; i < nb_fd; i++) {
1433 health_code_update(&health_thread_app_reg);
1434
1435 /* Fetch once the poll data */
1436 revents = LTTNG_POLL_GETEV(&events, i);
1437 pollfd = LTTNG_POLL_GETFD(&events, i);
1438
1439 /* Thread quit pipe has been closed. Killing thread. */
1440 ret = check_thread_quit_pipe(pollfd, revents);
1441 if (ret) {
1442 err = 0;
1443 goto exit;
1444 }
1445
1446 /* Event on the registration socket */
1447 if (pollfd == apps_sock) {
1448 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1449 ERR("Register apps socket poll error");
1450 goto error;
1451 } else if (revents & LPOLLIN) {
1452 sock = lttcomm_accept_unix_sock(apps_sock);
1453 if (sock < 0) {
1454 goto error;
1455 }
1456
1457 /*
1458 * Set the CLOEXEC flag. Return code is useless because
1459 * either way, the show must go on.
1460 */
1461 (void) utils_set_fd_cloexec(sock);
1462
1463 /* Create UST registration command for enqueuing */
1464 ust_cmd = zmalloc(sizeof(struct ust_command));
1465 if (ust_cmd == NULL) {
1466 PERROR("ust command zmalloc");
1467 goto error;
1468 }
1469
1470 /*
1471 * Using message-based transmissions to ensure we don't
1472 * have to deal with partially received messages.
1473 */
1474 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1475 if (ret < 0) {
1476 ERR("Exhausted file descriptors allowed for applications.");
1477 free(ust_cmd);
1478 ret = close(sock);
1479 if (ret) {
1480 PERROR("close");
1481 }
1482 sock = -1;
1483 continue;
1484 }
1485 health_code_update(&health_thread_app_reg);
1486 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1487 sizeof(struct ust_register_msg));
1488 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1489 if (ret < 0) {
1490 PERROR("lttcomm_recv_unix_sock register apps");
1491 } else {
1492 ERR("Wrong size received on apps register");
1493 }
1494 free(ust_cmd);
1495 ret = close(sock);
1496 if (ret) {
1497 PERROR("close");
1498 }
1499 lttng_fd_put(LTTNG_FD_APPS, 1);
1500 sock = -1;
1501 continue;
1502 }
1503 health_code_update(&health_thread_app_reg);
1504
1505 ust_cmd->sock = sock;
1506 sock = -1;
1507
1508 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1509 " gid:%d sock:%d name:%s (version %d.%d)",
1510 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1511 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1512 ust_cmd->sock, ust_cmd->reg_msg.name,
1513 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1514
1515 /*
1516 * Lock free enqueue the registration request. The red pill
1517 * has been taken! This apps will be part of the *system*.
1518 */
1519 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1520
1521 /*
1522 * Wake the registration queue futex. Implicit memory
1523 * barrier with the exchange in cds_wfq_enqueue.
1524 */
1525 futex_nto1_wake(&ust_cmd_queue.futex);
1526 }
1527 }
1528 }
1529 }
1530
1531 exit:
1532 error:
1533 if (err) {
1534 health_error(&health_thread_app_reg);
1535 ERR("Health error occurred in %s", __func__);
1536 }
1537
1538 /* Notify that the registration thread is gone */
1539 notify_ust_apps(0);
1540
1541 if (apps_sock >= 0) {
1542 ret = close(apps_sock);
1543 if (ret) {
1544 PERROR("close");
1545 }
1546 }
1547 if (sock >= 0) {
1548 ret = close(sock);
1549 if (ret) {
1550 PERROR("close");
1551 }
1552 lttng_fd_put(LTTNG_FD_APPS, 1);
1553 }
1554 unlink(apps_unix_sock_path);
1555
1556 error_poll_add:
1557 lttng_poll_clean(&events);
1558 error_listen:
1559 error_create_poll:
1560 error_testpoint:
1561 DBG("UST Registration thread cleanup complete");
1562 health_exit(&health_thread_app_reg);
1563
1564 return NULL;
1565 }
1566
1567 /*
1568 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1569 * exec or it will fails.
1570 */
1571 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1572 {
1573 int ret, clock_ret;
1574 struct timespec timeout;
1575
1576 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1577 consumer_data->consumer_thread_is_ready = 0;
1578
1579 /* Setup pthread condition */
1580 ret = pthread_condattr_init(&consumer_data->condattr);
1581 if (ret != 0) {
1582 errno = ret;
1583 PERROR("pthread_condattr_init consumer data");
1584 goto error;
1585 }
1586
1587 /*
1588 * Set the monotonic clock in order to make sure we DO NOT jump in time
1589 * between the clock_gettime() call and the timedwait call. See bug #324
1590 * for a more details and how we noticed it.
1591 */
1592 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1593 if (ret != 0) {
1594 errno = ret;
1595 PERROR("pthread_condattr_setclock consumer data");
1596 goto error;
1597 }
1598
1599 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1600 if (ret != 0) {
1601 errno = ret;
1602 PERROR("pthread_cond_init consumer data");
1603 goto error;
1604 }
1605
1606 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1607 consumer_data);
1608 if (ret != 0) {
1609 PERROR("pthread_create consumer");
1610 ret = -1;
1611 goto error;
1612 }
1613
1614 /* We are about to wait on a pthread condition */
1615 pthread_mutex_lock(&consumer_data->cond_mutex);
1616
1617 /* Get time for sem_timedwait absolute timeout */
1618 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1619 /*
1620 * Set the timeout for the condition timed wait even if the clock gettime
1621 * call fails since we might loop on that call and we want to avoid to
1622 * increment the timeout too many times.
1623 */
1624 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1625
1626 /*
1627 * The following loop COULD be skipped in some conditions so this is why we
1628 * set ret to 0 in order to make sure at least one round of the loop is
1629 * done.
1630 */
1631 ret = 0;
1632
1633 /*
1634 * Loop until the condition is reached or when a timeout is reached. Note
1635 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1636 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1637 * possible. This loop does not take any chances and works with both of
1638 * them.
1639 */
1640 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1641 if (clock_ret < 0) {
1642 PERROR("clock_gettime spawn consumer");
1643 /* Infinite wait for the consumerd thread to be ready */
1644 ret = pthread_cond_wait(&consumer_data->cond,
1645 &consumer_data->cond_mutex);
1646 } else {
1647 ret = pthread_cond_timedwait(&consumer_data->cond,
1648 &consumer_data->cond_mutex, &timeout);
1649 }
1650 }
1651
1652 /* Release the pthread condition */
1653 pthread_mutex_unlock(&consumer_data->cond_mutex);
1654
1655 if (ret != 0) {
1656 errno = ret;
1657 if (ret == ETIMEDOUT) {
1658 /*
1659 * Call has timed out so we kill the kconsumerd_thread and return
1660 * an error.
1661 */
1662 ERR("Condition timed out. The consumer thread was never ready."
1663 " Killing it");
1664 ret = pthread_cancel(consumer_data->thread);
1665 if (ret < 0) {
1666 PERROR("pthread_cancel consumer thread");
1667 }
1668 } else {
1669 PERROR("pthread_cond_wait failed consumer thread");
1670 }
1671 goto error;
1672 }
1673
1674 pthread_mutex_lock(&consumer_data->pid_mutex);
1675 if (consumer_data->pid == 0) {
1676 ERR("Consumerd did not start");
1677 pthread_mutex_unlock(&consumer_data->pid_mutex);
1678 goto error;
1679 }
1680 pthread_mutex_unlock(&consumer_data->pid_mutex);
1681
1682 return 0;
1683
1684 error:
1685 return ret;
1686 }
1687
1688 /*
1689 * Join consumer thread
1690 */
1691 static int join_consumer_thread(struct consumer_data *consumer_data)
1692 {
1693 void *status;
1694
1695 /* Consumer pid must be a real one. */
1696 if (consumer_data->pid > 0) {
1697 int ret;
1698 ret = kill(consumer_data->pid, SIGTERM);
1699 if (ret) {
1700 ERR("Error killing consumer daemon");
1701 return ret;
1702 }
1703 return pthread_join(consumer_data->thread, &status);
1704 } else {
1705 return 0;
1706 }
1707 }
1708
1709 /*
1710 * Fork and exec a consumer daemon (consumerd).
1711 *
1712 * Return pid if successful else -1.
1713 */
1714 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1715 {
1716 int ret;
1717 pid_t pid;
1718 const char *consumer_to_use;
1719 const char *verbosity;
1720 struct stat st;
1721
1722 DBG("Spawning consumerd");
1723
1724 pid = fork();
1725 if (pid == 0) {
1726 /*
1727 * Exec consumerd.
1728 */
1729 if (opt_verbose_consumer) {
1730 verbosity = "--verbose";
1731 } else {
1732 verbosity = "--quiet";
1733 }
1734 switch (consumer_data->type) {
1735 case LTTNG_CONSUMER_KERNEL:
1736 /*
1737 * Find out which consumerd to execute. We will first try the
1738 * 64-bit path, then the sessiond's installation directory, and
1739 * fallback on the 32-bit one,
1740 */
1741 DBG3("Looking for a kernel consumer at these locations:");
1742 DBG3(" 1) %s", consumerd64_bin);
1743 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1744 DBG3(" 3) %s", consumerd32_bin);
1745 if (stat(consumerd64_bin, &st) == 0) {
1746 DBG3("Found location #1");
1747 consumer_to_use = consumerd64_bin;
1748 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1749 DBG3("Found location #2");
1750 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1751 } else if (stat(consumerd32_bin, &st) == 0) {
1752 DBG3("Found location #3");
1753 consumer_to_use = consumerd32_bin;
1754 } else {
1755 DBG("Could not find any valid consumerd executable");
1756 break;
1757 }
1758 DBG("Using kernel consumer at: %s", consumer_to_use);
1759 execl(consumer_to_use,
1760 "lttng-consumerd", verbosity, "-k",
1761 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1762 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1763 NULL);
1764 break;
1765 case LTTNG_CONSUMER64_UST:
1766 {
1767 char *tmpnew = NULL;
1768
1769 if (consumerd64_libdir[0] != '\0') {
1770 char *tmp;
1771 size_t tmplen;
1772
1773 tmp = getenv("LD_LIBRARY_PATH");
1774 if (!tmp) {
1775 tmp = "";
1776 }
1777 tmplen = strlen("LD_LIBRARY_PATH=")
1778 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1779 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1780 if (!tmpnew) {
1781 ret = -ENOMEM;
1782 goto error;
1783 }
1784 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1785 strcat(tmpnew, consumerd64_libdir);
1786 if (tmp[0] != '\0') {
1787 strcat(tmpnew, ":");
1788 strcat(tmpnew, tmp);
1789 }
1790 ret = putenv(tmpnew);
1791 if (ret) {
1792 ret = -errno;
1793 goto error;
1794 }
1795 }
1796 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1797 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1798 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1799 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1800 NULL);
1801 if (consumerd64_libdir[0] != '\0') {
1802 free(tmpnew);
1803 }
1804 if (ret) {
1805 goto error;
1806 }
1807 break;
1808 }
1809 case LTTNG_CONSUMER32_UST:
1810 {
1811 char *tmpnew = NULL;
1812
1813 if (consumerd32_libdir[0] != '\0') {
1814 char *tmp;
1815 size_t tmplen;
1816
1817 tmp = getenv("LD_LIBRARY_PATH");
1818 if (!tmp) {
1819 tmp = "";
1820 }
1821 tmplen = strlen("LD_LIBRARY_PATH=")
1822 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1823 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1824 if (!tmpnew) {
1825 ret = -ENOMEM;
1826 goto error;
1827 }
1828 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1829 strcat(tmpnew, consumerd32_libdir);
1830 if (tmp[0] != '\0') {
1831 strcat(tmpnew, ":");
1832 strcat(tmpnew, tmp);
1833 }
1834 ret = putenv(tmpnew);
1835 if (ret) {
1836 ret = -errno;
1837 goto error;
1838 }
1839 }
1840 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1841 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1842 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1843 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1844 NULL);
1845 if (consumerd32_libdir[0] != '\0') {
1846 free(tmpnew);
1847 }
1848 if (ret) {
1849 goto error;
1850 }
1851 break;
1852 }
1853 default:
1854 PERROR("unknown consumer type");
1855 exit(EXIT_FAILURE);
1856 }
1857 if (errno != 0) {
1858 PERROR("kernel start consumer exec");
1859 }
1860 exit(EXIT_FAILURE);
1861 } else if (pid > 0) {
1862 ret = pid;
1863 } else {
1864 PERROR("start consumer fork");
1865 ret = -errno;
1866 }
1867 error:
1868 return ret;
1869 }
1870
1871 /*
1872 * Spawn the consumerd daemon and session daemon thread.
1873 */
1874 static int start_consumerd(struct consumer_data *consumer_data)
1875 {
1876 int ret;
1877
1878 /*
1879 * Set the listen() state on the socket since there is a possible race
1880 * between the exec() of the consumer daemon and this call if place in the
1881 * consumer thread. See bug #366 for more details.
1882 */
1883 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1884 if (ret < 0) {
1885 goto error;
1886 }
1887
1888 pthread_mutex_lock(&consumer_data->pid_mutex);
1889 if (consumer_data->pid != 0) {
1890 pthread_mutex_unlock(&consumer_data->pid_mutex);
1891 goto end;
1892 }
1893
1894 ret = spawn_consumerd(consumer_data);
1895 if (ret < 0) {
1896 ERR("Spawning consumerd failed");
1897 pthread_mutex_unlock(&consumer_data->pid_mutex);
1898 goto error;
1899 }
1900
1901 /* Setting up the consumer_data pid */
1902 consumer_data->pid = ret;
1903 DBG2("Consumer pid %d", consumer_data->pid);
1904 pthread_mutex_unlock(&consumer_data->pid_mutex);
1905
1906 DBG2("Spawning consumer control thread");
1907 ret = spawn_consumer_thread(consumer_data);
1908 if (ret < 0) {
1909 ERR("Fatal error spawning consumer control thread");
1910 goto error;
1911 }
1912
1913 end:
1914 return 0;
1915
1916 error:
1917 /* Cleanup already created socket on error. */
1918 if (consumer_data->err_sock >= 0) {
1919 int err;
1920
1921 err = close(consumer_data->err_sock);
1922 if (err < 0) {
1923 PERROR("close consumer data error socket");
1924 }
1925 }
1926 return ret;
1927 }
1928
1929 /*
1930 * Compute health status of each consumer. If one of them is zero (bad
1931 * state), we return 0.
1932 */
1933 static int check_consumer_health(void)
1934 {
1935 int ret;
1936
1937 ret = health_check_state(&kconsumer_data.health) &&
1938 health_check_state(&ustconsumer32_data.health) &&
1939 health_check_state(&ustconsumer64_data.health);
1940
1941 DBG3("Health consumer check %d", ret);
1942
1943 return ret;
1944 }
1945
1946 /*
1947 * Setup necessary data for kernel tracer action.
1948 */
1949 static int init_kernel_tracer(void)
1950 {
1951 int ret;
1952
1953 /* Modprobe lttng kernel modules */
1954 ret = modprobe_lttng_control();
1955 if (ret < 0) {
1956 goto error;
1957 }
1958
1959 /* Open debugfs lttng */
1960 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1961 if (kernel_tracer_fd < 0) {
1962 DBG("Failed to open %s", module_proc_lttng);
1963 ret = -1;
1964 goto error_open;
1965 }
1966
1967 /* Validate kernel version */
1968 ret = kernel_validate_version(kernel_tracer_fd);
1969 if (ret < 0) {
1970 goto error_version;
1971 }
1972
1973 ret = modprobe_lttng_data();
1974 if (ret < 0) {
1975 goto error_modules;
1976 }
1977
1978 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1979 return 0;
1980
1981 error_version:
1982 modprobe_remove_lttng_control();
1983 ret = close(kernel_tracer_fd);
1984 if (ret) {
1985 PERROR("close");
1986 }
1987 kernel_tracer_fd = -1;
1988 return LTTNG_ERR_KERN_VERSION;
1989
1990 error_modules:
1991 ret = close(kernel_tracer_fd);
1992 if (ret) {
1993 PERROR("close");
1994 }
1995
1996 error_open:
1997 modprobe_remove_lttng_control();
1998
1999 error:
2000 WARN("No kernel tracer available");
2001 kernel_tracer_fd = -1;
2002 if (!is_root) {
2003 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2004 } else {
2005 return LTTNG_ERR_KERN_NA;
2006 }
2007 }
2008
2009
2010 /*
2011 * Copy consumer output from the tracing session to the domain session. The
2012 * function also applies the right modification on a per domain basis for the
2013 * trace files destination directory.
2014 */
2015 static int copy_session_consumer(int domain, struct ltt_session *session)
2016 {
2017 int ret;
2018 const char *dir_name;
2019 struct consumer_output *consumer;
2020
2021 assert(session);
2022 assert(session->consumer);
2023
2024 switch (domain) {
2025 case LTTNG_DOMAIN_KERNEL:
2026 DBG3("Copying tracing session consumer output in kernel session");
2027 /*
2028 * XXX: We should audit the session creation and what this function
2029 * does "extra" in order to avoid a destroy since this function is used
2030 * in the domain session creation (kernel and ust) only. Same for UST
2031 * domain.
2032 */
2033 if (session->kernel_session->consumer) {
2034 consumer_destroy_output(session->kernel_session->consumer);
2035 }
2036 session->kernel_session->consumer =
2037 consumer_copy_output(session->consumer);
2038 /* Ease our life a bit for the next part */
2039 consumer = session->kernel_session->consumer;
2040 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2041 break;
2042 case LTTNG_DOMAIN_UST:
2043 DBG3("Copying tracing session consumer output in UST session");
2044 if (session->ust_session->consumer) {
2045 consumer_destroy_output(session->ust_session->consumer);
2046 }
2047 session->ust_session->consumer =
2048 consumer_copy_output(session->consumer);
2049 /* Ease our life a bit for the next part */
2050 consumer = session->ust_session->consumer;
2051 dir_name = DEFAULT_UST_TRACE_DIR;
2052 break;
2053 default:
2054 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2055 goto error;
2056 }
2057
2058 /* Append correct directory to subdir */
2059 strncat(consumer->subdir, dir_name,
2060 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2061 DBG3("Copy session consumer subdir %s", consumer->subdir);
2062
2063 ret = LTTNG_OK;
2064
2065 error:
2066 return ret;
2067 }
2068
2069 /*
2070 * Create an UST session and add it to the session ust list.
2071 */
2072 static int create_ust_session(struct ltt_session *session,
2073 struct lttng_domain *domain)
2074 {
2075 int ret;
2076 struct ltt_ust_session *lus = NULL;
2077
2078 assert(session);
2079 assert(domain);
2080 assert(session->consumer);
2081
2082 switch (domain->type) {
2083 case LTTNG_DOMAIN_UST:
2084 break;
2085 default:
2086 ERR("Unknown UST domain on create session %d", domain->type);
2087 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2088 goto error;
2089 }
2090
2091 DBG("Creating UST session");
2092
2093 lus = trace_ust_create_session(session->path, session->id, domain);
2094 if (lus == NULL) {
2095 ret = LTTNG_ERR_UST_SESS_FAIL;
2096 goto error;
2097 }
2098
2099 lus->uid = session->uid;
2100 lus->gid = session->gid;
2101 session->ust_session = lus;
2102
2103 /* Copy session output to the newly created UST session */
2104 ret = copy_session_consumer(domain->type, session);
2105 if (ret != LTTNG_OK) {
2106 goto error;
2107 }
2108
2109 return LTTNG_OK;
2110
2111 error:
2112 free(lus);
2113 session->ust_session = NULL;
2114 return ret;
2115 }
2116
2117 /*
2118 * Create a kernel tracer session then create the default channel.
2119 */
2120 static int create_kernel_session(struct ltt_session *session)
2121 {
2122 int ret;
2123
2124 DBG("Creating kernel session");
2125
2126 ret = kernel_create_session(session, kernel_tracer_fd);
2127 if (ret < 0) {
2128 ret = LTTNG_ERR_KERN_SESS_FAIL;
2129 goto error;
2130 }
2131
2132 /* Code flow safety */
2133 assert(session->kernel_session);
2134
2135 /* Copy session output to the newly created Kernel session */
2136 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2137 if (ret != LTTNG_OK) {
2138 goto error;
2139 }
2140
2141 /* Create directory(ies) on local filesystem. */
2142 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2143 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2144 ret = run_as_mkdir_recursive(
2145 session->kernel_session->consumer->dst.trace_path,
2146 S_IRWXU | S_IRWXG, session->uid, session->gid);
2147 if (ret < 0) {
2148 if (ret != -EEXIST) {
2149 ERR("Trace directory creation error");
2150 goto error;
2151 }
2152 }
2153 }
2154
2155 session->kernel_session->uid = session->uid;
2156 session->kernel_session->gid = session->gid;
2157
2158 return LTTNG_OK;
2159
2160 error:
2161 trace_kernel_destroy_session(session->kernel_session);
2162 session->kernel_session = NULL;
2163 return ret;
2164 }
2165
2166 /*
2167 * Count number of session permitted by uid/gid.
2168 */
2169 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2170 {
2171 unsigned int i = 0;
2172 struct ltt_session *session;
2173
2174 DBG("Counting number of available session for UID %d GID %d",
2175 uid, gid);
2176 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2177 /*
2178 * Only list the sessions the user can control.
2179 */
2180 if (!session_access_ok(session, uid, gid)) {
2181 continue;
2182 }
2183 i++;
2184 }
2185 return i;
2186 }
2187
2188 /*
2189 * Process the command requested by the lttng client within the command
2190 * context structure. This function make sure that the return structure (llm)
2191 * is set and ready for transmission before returning.
2192 *
2193 * Return any error encountered or 0 for success.
2194 *
2195 * "sock" is only used for special-case var. len data.
2196 */
2197 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2198 int *sock_error)
2199 {
2200 int ret = LTTNG_OK;
2201 int need_tracing_session = 1;
2202 int need_domain;
2203
2204 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2205
2206 *sock_error = 0;
2207
2208 switch (cmd_ctx->lsm->cmd_type) {
2209 case LTTNG_CREATE_SESSION:
2210 case LTTNG_DESTROY_SESSION:
2211 case LTTNG_LIST_SESSIONS:
2212 case LTTNG_LIST_DOMAINS:
2213 case LTTNG_START_TRACE:
2214 case LTTNG_STOP_TRACE:
2215 case LTTNG_DATA_PENDING:
2216 need_domain = 0;
2217 break;
2218 default:
2219 need_domain = 1;
2220 }
2221
2222 if (opt_no_kernel && need_domain
2223 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2224 if (!is_root) {
2225 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2226 } else {
2227 ret = LTTNG_ERR_KERN_NA;
2228 }
2229 goto error;
2230 }
2231
2232 /* Deny register consumer if we already have a spawned consumer. */
2233 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2234 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2235 if (kconsumer_data.pid > 0) {
2236 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2237 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2238 goto error;
2239 }
2240 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2241 }
2242
2243 /*
2244 * Check for command that don't needs to allocate a returned payload. We do
2245 * this here so we don't have to make the call for no payload at each
2246 * command.
2247 */
2248 switch(cmd_ctx->lsm->cmd_type) {
2249 case LTTNG_LIST_SESSIONS:
2250 case LTTNG_LIST_TRACEPOINTS:
2251 case LTTNG_LIST_TRACEPOINT_FIELDS:
2252 case LTTNG_LIST_DOMAINS:
2253 case LTTNG_LIST_CHANNELS:
2254 case LTTNG_LIST_EVENTS:
2255 break;
2256 default:
2257 /* Setup lttng message with no payload */
2258 ret = setup_lttng_msg(cmd_ctx, 0);
2259 if (ret < 0) {
2260 /* This label does not try to unlock the session */
2261 goto init_setup_error;
2262 }
2263 }
2264
2265 /* Commands that DO NOT need a session. */
2266 switch (cmd_ctx->lsm->cmd_type) {
2267 case LTTNG_CREATE_SESSION:
2268 case LTTNG_CALIBRATE:
2269 case LTTNG_LIST_SESSIONS:
2270 case LTTNG_LIST_TRACEPOINTS:
2271 case LTTNG_LIST_TRACEPOINT_FIELDS:
2272 need_tracing_session = 0;
2273 break;
2274 default:
2275 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2276 /*
2277 * We keep the session list lock across _all_ commands
2278 * for now, because the per-session lock does not
2279 * handle teardown properly.
2280 */
2281 session_lock_list();
2282 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2283 if (cmd_ctx->session == NULL) {
2284 if (cmd_ctx->lsm->session.name != NULL) {
2285 ret = LTTNG_ERR_SESS_NOT_FOUND;
2286 } else {
2287 /* If no session name specified */
2288 ret = LTTNG_ERR_SELECT_SESS;
2289 }
2290 goto error;
2291 } else {
2292 /* Acquire lock for the session */
2293 session_lock(cmd_ctx->session);
2294 }
2295 break;
2296 }
2297
2298 if (!need_domain) {
2299 goto skip_domain;
2300 }
2301
2302 /*
2303 * Check domain type for specific "pre-action".
2304 */
2305 switch (cmd_ctx->lsm->domain.type) {
2306 case LTTNG_DOMAIN_KERNEL:
2307 if (!is_root) {
2308 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2309 goto error;
2310 }
2311
2312 /* Kernel tracer check */
2313 if (kernel_tracer_fd == -1) {
2314 /* Basically, load kernel tracer modules */
2315 ret = init_kernel_tracer();
2316 if (ret != 0) {
2317 goto error;
2318 }
2319 }
2320
2321 /* Consumer is in an ERROR state. Report back to client */
2322 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2323 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2324 goto error;
2325 }
2326
2327 /* Need a session for kernel command */
2328 if (need_tracing_session) {
2329 if (cmd_ctx->session->kernel_session == NULL) {
2330 ret = create_kernel_session(cmd_ctx->session);
2331 if (ret < 0) {
2332 ret = LTTNG_ERR_KERN_SESS_FAIL;
2333 goto error;
2334 }
2335 }
2336
2337 /* Start the kernel consumer daemon */
2338 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2339 if (kconsumer_data.pid == 0 &&
2340 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2341 cmd_ctx->session->start_consumer) {
2342 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2343 ret = start_consumerd(&kconsumer_data);
2344 if (ret < 0) {
2345 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2346 goto error;
2347 }
2348 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2349 } else {
2350 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2351 }
2352
2353 /*
2354 * The consumer was just spawned so we need to add the socket to
2355 * the consumer output of the session if exist.
2356 */
2357 ret = consumer_create_socket(&kconsumer_data,
2358 cmd_ctx->session->kernel_session->consumer);
2359 if (ret < 0) {
2360 goto error;
2361 }
2362 }
2363
2364 break;
2365 case LTTNG_DOMAIN_UST:
2366 {
2367 /* Consumer is in an ERROR state. Report back to client */
2368 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2369 ret = LTTNG_ERR_NO_USTCONSUMERD;
2370 goto error;
2371 }
2372
2373 if (need_tracing_session) {
2374 /* Create UST session if none exist. */
2375 if (cmd_ctx->session->ust_session == NULL) {
2376 ret = create_ust_session(cmd_ctx->session,
2377 &cmd_ctx->lsm->domain);
2378 if (ret != LTTNG_OK) {
2379 goto error;
2380 }
2381 }
2382
2383 /* Start the UST consumer daemons */
2384 /* 64-bit */
2385 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2386 if (consumerd64_bin[0] != '\0' &&
2387 ustconsumer64_data.pid == 0 &&
2388 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2389 cmd_ctx->session->start_consumer) {
2390 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2391 ret = start_consumerd(&ustconsumer64_data);
2392 if (ret < 0) {
2393 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2394 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2395 goto error;
2396 }
2397
2398 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2399 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2400 } else {
2401 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2402 }
2403
2404 /*
2405 * Setup socket for consumer 64 bit. No need for atomic access
2406 * since it was set above and can ONLY be set in this thread.
2407 */
2408 ret = consumer_create_socket(&ustconsumer64_data,
2409 cmd_ctx->session->ust_session->consumer);
2410 if (ret < 0) {
2411 goto error;
2412 }
2413
2414 /* 32-bit */
2415 if (consumerd32_bin[0] != '\0' &&
2416 ustconsumer32_data.pid == 0 &&
2417 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2418 cmd_ctx->session->start_consumer) {
2419 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2420 ret = start_consumerd(&ustconsumer32_data);
2421 if (ret < 0) {
2422 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2423 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2424 goto error;
2425 }
2426
2427 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2428 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2429 } else {
2430 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2431 }
2432
2433 /*
2434 * Setup socket for consumer 64 bit. No need for atomic access
2435 * since it was set above and can ONLY be set in this thread.
2436 */
2437 ret = consumer_create_socket(&ustconsumer32_data,
2438 cmd_ctx->session->ust_session->consumer);
2439 if (ret < 0) {
2440 goto error;
2441 }
2442 }
2443 break;
2444 }
2445 default:
2446 break;
2447 }
2448 skip_domain:
2449
2450 /* Validate consumer daemon state when start/stop trace command */
2451 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2452 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2453 switch (cmd_ctx->lsm->domain.type) {
2454 case LTTNG_DOMAIN_UST:
2455 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2456 ret = LTTNG_ERR_NO_USTCONSUMERD;
2457 goto error;
2458 }
2459 break;
2460 case LTTNG_DOMAIN_KERNEL:
2461 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2462 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2463 goto error;
2464 }
2465 break;
2466 }
2467 }
2468
2469 /*
2470 * Check that the UID or GID match that of the tracing session.
2471 * The root user can interact with all sessions.
2472 */
2473 if (need_tracing_session) {
2474 if (!session_access_ok(cmd_ctx->session,
2475 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2476 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2477 ret = LTTNG_ERR_EPERM;
2478 goto error;
2479 }
2480 }
2481
2482 /* Process by command type */
2483 switch (cmd_ctx->lsm->cmd_type) {
2484 case LTTNG_ADD_CONTEXT:
2485 {
2486 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2487 cmd_ctx->lsm->u.context.channel_name,
2488 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2489 break;
2490 }
2491 case LTTNG_DISABLE_CHANNEL:
2492 {
2493 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2494 cmd_ctx->lsm->u.disable.channel_name);
2495 break;
2496 }
2497 case LTTNG_DISABLE_EVENT:
2498 {
2499 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2500 cmd_ctx->lsm->u.disable.channel_name,
2501 cmd_ctx->lsm->u.disable.name);
2502 break;
2503 }
2504 case LTTNG_DISABLE_ALL_EVENT:
2505 {
2506 DBG("Disabling all events");
2507
2508 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2509 cmd_ctx->lsm->u.disable.channel_name);
2510 break;
2511 }
2512 case LTTNG_DISABLE_CONSUMER:
2513 {
2514 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2515 break;
2516 }
2517 case LTTNG_ENABLE_CHANNEL:
2518 {
2519 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2520 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2521 break;
2522 }
2523 case LTTNG_ENABLE_CONSUMER:
2524 {
2525 /*
2526 * XXX: 0 means that this URI should be applied on the session. Should
2527 * be a DOMAIN enuam.
2528 */
2529 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2530 if (ret != LTTNG_OK) {
2531 goto error;
2532 }
2533
2534 if (cmd_ctx->lsm->domain.type == 0) {
2535 /* Add the URI for the UST session if a consumer is present. */
2536 if (cmd_ctx->session->ust_session &&
2537 cmd_ctx->session->ust_session->consumer) {
2538 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2539 } else if (cmd_ctx->session->kernel_session &&
2540 cmd_ctx->session->kernel_session->consumer) {
2541 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2542 cmd_ctx->session);
2543 }
2544 }
2545 break;
2546 }
2547 case LTTNG_ENABLE_EVENT:
2548 {
2549 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2550 cmd_ctx->lsm->u.enable.channel_name,
2551 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2552 break;
2553 }
2554 case LTTNG_ENABLE_ALL_EVENT:
2555 {
2556 DBG("Enabling all events");
2557
2558 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2559 cmd_ctx->lsm->u.enable.channel_name,
2560 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2561 break;
2562 }
2563 case LTTNG_LIST_TRACEPOINTS:
2564 {
2565 struct lttng_event *events;
2566 ssize_t nb_events;
2567
2568 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2569 if (nb_events < 0) {
2570 /* Return value is a negative lttng_error_code. */
2571 ret = -nb_events;
2572 goto error;
2573 }
2574
2575 /*
2576 * Setup lttng message with payload size set to the event list size in
2577 * bytes and then copy list into the llm payload.
2578 */
2579 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2580 if (ret < 0) {
2581 free(events);
2582 goto setup_error;
2583 }
2584
2585 /* Copy event list into message payload */
2586 memcpy(cmd_ctx->llm->payload, events,
2587 sizeof(struct lttng_event) * nb_events);
2588
2589 free(events);
2590
2591 ret = LTTNG_OK;
2592 break;
2593 }
2594 case LTTNG_LIST_TRACEPOINT_FIELDS:
2595 {
2596 struct lttng_event_field *fields;
2597 ssize_t nb_fields;
2598
2599 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2600 &fields);
2601 if (nb_fields < 0) {
2602 /* Return value is a negative lttng_error_code. */
2603 ret = -nb_fields;
2604 goto error;
2605 }
2606
2607 /*
2608 * Setup lttng message with payload size set to the event list size in
2609 * bytes and then copy list into the llm payload.
2610 */
2611 ret = setup_lttng_msg(cmd_ctx,
2612 sizeof(struct lttng_event_field) * nb_fields);
2613 if (ret < 0) {
2614 free(fields);
2615 goto setup_error;
2616 }
2617
2618 /* Copy event list into message payload */
2619 memcpy(cmd_ctx->llm->payload, fields,
2620 sizeof(struct lttng_event_field) * nb_fields);
2621
2622 free(fields);
2623
2624 ret = LTTNG_OK;
2625 break;
2626 }
2627 case LTTNG_SET_CONSUMER_URI:
2628 {
2629 size_t nb_uri, len;
2630 struct lttng_uri *uris;
2631
2632 nb_uri = cmd_ctx->lsm->u.uri.size;
2633 len = nb_uri * sizeof(struct lttng_uri);
2634
2635 if (nb_uri == 0) {
2636 ret = LTTNG_ERR_INVALID;
2637 goto error;
2638 }
2639
2640 uris = zmalloc(len);
2641 if (uris == NULL) {
2642 ret = LTTNG_ERR_FATAL;
2643 goto error;
2644 }
2645
2646 /* Receive variable len data */
2647 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2648 ret = lttcomm_recv_unix_sock(sock, uris, len);
2649 if (ret <= 0) {
2650 DBG("No URIs received from client... continuing");
2651 *sock_error = 1;
2652 ret = LTTNG_ERR_SESSION_FAIL;
2653 free(uris);
2654 goto error;
2655 }
2656
2657 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2658 nb_uri, uris);
2659 if (ret != LTTNG_OK) {
2660 free(uris);
2661 goto error;
2662 }
2663
2664 /*
2665 * XXX: 0 means that this URI should be applied on the session. Should
2666 * be a DOMAIN enuam.
2667 */
2668 if (cmd_ctx->lsm->domain.type == 0) {
2669 /* Add the URI for the UST session if a consumer is present. */
2670 if (cmd_ctx->session->ust_session &&
2671 cmd_ctx->session->ust_session->consumer) {
2672 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2673 nb_uri, uris);
2674 } else if (cmd_ctx->session->kernel_session &&
2675 cmd_ctx->session->kernel_session->consumer) {
2676 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2677 cmd_ctx->session, nb_uri, uris);
2678 }
2679 }
2680
2681 free(uris);
2682
2683 break;
2684 }
2685 case LTTNG_START_TRACE:
2686 {
2687 ret = cmd_start_trace(cmd_ctx->session);
2688 break;
2689 }
2690 case LTTNG_STOP_TRACE:
2691 {
2692 ret = cmd_stop_trace(cmd_ctx->session);
2693 break;
2694 }
2695 case LTTNG_CREATE_SESSION:
2696 {
2697 size_t nb_uri, len;
2698 struct lttng_uri *uris = NULL;
2699
2700 nb_uri = cmd_ctx->lsm->u.uri.size;
2701 len = nb_uri * sizeof(struct lttng_uri);
2702
2703 if (nb_uri > 0) {
2704 uris = zmalloc(len);
2705 if (uris == NULL) {
2706 ret = LTTNG_ERR_FATAL;
2707 goto error;
2708 }
2709
2710 /* Receive variable len data */
2711 DBG("Waiting for %zu URIs from client ...", nb_uri);
2712 ret = lttcomm_recv_unix_sock(sock, uris, len);
2713 if (ret <= 0) {
2714 DBG("No URIs received from client... continuing");
2715 *sock_error = 1;
2716 ret = LTTNG_ERR_SESSION_FAIL;
2717 free(uris);
2718 goto error;
2719 }
2720
2721 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2722 DBG("Creating session with ONE network URI is a bad call");
2723 ret = LTTNG_ERR_SESSION_FAIL;
2724 free(uris);
2725 goto error;
2726 }
2727 }
2728
2729 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2730 &cmd_ctx->creds);
2731
2732 free(uris);
2733
2734 break;
2735 }
2736 case LTTNG_DESTROY_SESSION:
2737 {
2738 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2739
2740 /* Set session to NULL so we do not unlock it after free. */
2741 cmd_ctx->session = NULL;
2742 break;
2743 }
2744 case LTTNG_LIST_DOMAINS:
2745 {
2746 ssize_t nb_dom;
2747 struct lttng_domain *domains;
2748
2749 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2750 if (nb_dom < 0) {
2751 /* Return value is a negative lttng_error_code. */
2752 ret = -nb_dom;
2753 goto error;
2754 }
2755
2756 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2757 if (ret < 0) {
2758 goto setup_error;
2759 }
2760
2761 /* Copy event list into message payload */
2762 memcpy(cmd_ctx->llm->payload, domains,
2763 nb_dom * sizeof(struct lttng_domain));
2764
2765 free(domains);
2766
2767 ret = LTTNG_OK;
2768 break;
2769 }
2770 case LTTNG_LIST_CHANNELS:
2771 {
2772 int nb_chan;
2773 struct lttng_channel *channels;
2774
2775 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2776 cmd_ctx->session, &channels);
2777 if (nb_chan < 0) {
2778 /* Return value is a negative lttng_error_code. */
2779 ret = -nb_chan;
2780 goto error;
2781 }
2782
2783 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2784 if (ret < 0) {
2785 goto setup_error;
2786 }
2787
2788 /* Copy event list into message payload */
2789 memcpy(cmd_ctx->llm->payload, channels,
2790 nb_chan * sizeof(struct lttng_channel));
2791
2792 free(channels);
2793
2794 ret = LTTNG_OK;
2795 break;
2796 }
2797 case LTTNG_LIST_EVENTS:
2798 {
2799 ssize_t nb_event;
2800 struct lttng_event *events = NULL;
2801
2802 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2803 cmd_ctx->lsm->u.list.channel_name, &events);
2804 if (nb_event < 0) {
2805 /* Return value is a negative lttng_error_code. */
2806 ret = -nb_event;
2807 goto error;
2808 }
2809
2810 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2811 if (ret < 0) {
2812 goto setup_error;
2813 }
2814
2815 /* Copy event list into message payload */
2816 memcpy(cmd_ctx->llm->payload, events,
2817 nb_event * sizeof(struct lttng_event));
2818
2819 free(events);
2820
2821 ret = LTTNG_OK;
2822 break;
2823 }
2824 case LTTNG_LIST_SESSIONS:
2825 {
2826 unsigned int nr_sessions;
2827
2828 session_lock_list();
2829 nr_sessions = lttng_sessions_count(
2830 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2831 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2832
2833 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2834 if (ret < 0) {
2835 session_unlock_list();
2836 goto setup_error;
2837 }
2838
2839 /* Filled the session array */
2840 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2841 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2842 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2843
2844 session_unlock_list();
2845
2846 ret = LTTNG_OK;
2847 break;
2848 }
2849 case LTTNG_CALIBRATE:
2850 {
2851 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2852 &cmd_ctx->lsm->u.calibrate);
2853 break;
2854 }
2855 case LTTNG_REGISTER_CONSUMER:
2856 {
2857 struct consumer_data *cdata;
2858
2859 switch (cmd_ctx->lsm->domain.type) {
2860 case LTTNG_DOMAIN_KERNEL:
2861 cdata = &kconsumer_data;
2862 break;
2863 default:
2864 ret = LTTNG_ERR_UND;
2865 goto error;
2866 }
2867
2868 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2869 cmd_ctx->lsm->u.reg.path, cdata);
2870 break;
2871 }
2872 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2873 {
2874 struct lttng_filter_bytecode *bytecode;
2875
2876 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2877 ret = LTTNG_ERR_FILTER_INVAL;
2878 goto error;
2879 }
2880 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2881 ret = LTTNG_ERR_FILTER_INVAL;
2882 goto error;
2883 }
2884 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2885 if (!bytecode) {
2886 ret = LTTNG_ERR_FILTER_NOMEM;
2887 goto error;
2888 }
2889 /* Receive var. len. data */
2890 DBG("Receiving var len data from client ...");
2891 ret = lttcomm_recv_unix_sock(sock, bytecode,
2892 cmd_ctx->lsm->u.enable.bytecode_len);
2893 if (ret <= 0) {
2894 DBG("Nothing recv() from client var len data... continuing");
2895 *sock_error = 1;
2896 ret = LTTNG_ERR_FILTER_INVAL;
2897 goto error;
2898 }
2899
2900 if (bytecode->len + sizeof(*bytecode)
2901 != cmd_ctx->lsm->u.enable.bytecode_len) {
2902 free(bytecode);
2903 ret = LTTNG_ERR_FILTER_INVAL;
2904 goto error;
2905 }
2906
2907 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2908 cmd_ctx->lsm->u.enable.channel_name,
2909 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2910 break;
2911 }
2912 case LTTNG_DATA_PENDING:
2913 {
2914 ret = cmd_data_pending(cmd_ctx->session);
2915 break;
2916 }
2917 default:
2918 ret = LTTNG_ERR_UND;
2919 break;
2920 }
2921
2922 error:
2923 if (cmd_ctx->llm == NULL) {
2924 DBG("Missing llm structure. Allocating one.");
2925 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2926 goto setup_error;
2927 }
2928 }
2929 /* Set return code */
2930 cmd_ctx->llm->ret_code = ret;
2931 setup_error:
2932 if (cmd_ctx->session) {
2933 session_unlock(cmd_ctx->session);
2934 }
2935 if (need_tracing_session) {
2936 session_unlock_list();
2937 }
2938 init_setup_error:
2939 return ret;
2940 }
2941
2942 /*
2943 * Thread managing health check socket.
2944 */
2945 static void *thread_manage_health(void *data)
2946 {
2947 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2948 uint32_t revents, nb_fd;
2949 struct lttng_poll_event events;
2950 struct lttcomm_health_msg msg;
2951 struct lttcomm_health_data reply;
2952
2953 DBG("[thread] Manage health check started");
2954
2955 rcu_register_thread();
2956
2957 /* Create unix socket */
2958 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2959 if (sock < 0) {
2960 ERR("Unable to create health check Unix socket");
2961 ret = -1;
2962 goto error;
2963 }
2964
2965 /*
2966 * Set the CLOEXEC flag. Return code is useless because either way, the
2967 * show must go on.
2968 */
2969 (void) utils_set_fd_cloexec(sock);
2970
2971 ret = lttcomm_listen_unix_sock(sock);
2972 if (ret < 0) {
2973 goto error;
2974 }
2975
2976 /*
2977 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2978 * more will be added to this poll set.
2979 */
2980 ret = create_thread_poll_set(&events, 2);
2981 if (ret < 0) {
2982 goto error;
2983 }
2984
2985 /* Add the application registration socket */
2986 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2987 if (ret < 0) {
2988 goto error;
2989 }
2990
2991 while (1) {
2992 DBG("Health check ready");
2993
2994 /* Inifinite blocking call, waiting for transmission */
2995 restart:
2996 ret = lttng_poll_wait(&events, -1);
2997 if (ret < 0) {
2998 /*
2999 * Restart interrupted system call.
3000 */
3001 if (errno == EINTR) {
3002 goto restart;
3003 }
3004 goto error;
3005 }
3006
3007 nb_fd = ret;
3008
3009 for (i = 0; i < nb_fd; i++) {
3010 /* Fetch once the poll data */
3011 revents = LTTNG_POLL_GETEV(&events, i);
3012 pollfd = LTTNG_POLL_GETFD(&events, i);
3013
3014 /* Thread quit pipe has been closed. Killing thread. */
3015 ret = check_thread_quit_pipe(pollfd, revents);
3016 if (ret) {
3017 err = 0;
3018 goto exit;
3019 }
3020
3021 /* Event on the registration socket */
3022 if (pollfd == sock) {
3023 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3024 ERR("Health socket poll error");
3025 goto error;
3026 }
3027 }
3028 }
3029
3030 new_sock = lttcomm_accept_unix_sock(sock);
3031 if (new_sock < 0) {
3032 goto error;
3033 }
3034
3035 /*
3036 * Set the CLOEXEC flag. Return code is useless because either way, the
3037 * show must go on.
3038 */
3039 (void) utils_set_fd_cloexec(new_sock);
3040
3041 DBG("Receiving data from client for health...");
3042 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3043 if (ret <= 0) {
3044 DBG("Nothing recv() from client... continuing");
3045 ret = close(new_sock);
3046 if (ret) {
3047 PERROR("close");
3048 }
3049 new_sock = -1;
3050 continue;
3051 }
3052
3053 rcu_thread_online();
3054
3055 switch (msg.component) {
3056 case LTTNG_HEALTH_CMD:
3057 reply.ret_code = health_check_state(&health_thread_cmd);
3058 break;
3059 case LTTNG_HEALTH_APP_MANAGE:
3060 reply.ret_code = health_check_state(&health_thread_app_manage);
3061 break;
3062 case LTTNG_HEALTH_APP_REG:
3063 reply.ret_code = health_check_state(&health_thread_app_reg);
3064 break;
3065 case LTTNG_HEALTH_KERNEL:
3066 reply.ret_code = health_check_state(&health_thread_kernel);
3067 break;
3068 case LTTNG_HEALTH_CONSUMER:
3069 reply.ret_code = check_consumer_health();
3070 break;
3071 case LTTNG_HEALTH_ALL:
3072 reply.ret_code =
3073 health_check_state(&health_thread_app_manage) &&
3074 health_check_state(&health_thread_app_reg) &&
3075 health_check_state(&health_thread_cmd) &&
3076 health_check_state(&health_thread_kernel) &&
3077 check_consumer_health();
3078 break;
3079 default:
3080 reply.ret_code = LTTNG_ERR_UND;
3081 break;
3082 }
3083
3084 /*
3085 * Flip ret value since 0 is a success and 1 indicates a bad health for
3086 * the client where in the sessiond it is the opposite. Again, this is
3087 * just to make things easier for us poor developer which enjoy a lot
3088 * lazyness.
3089 */
3090 if (reply.ret_code == 0 || reply.ret_code == 1) {
3091 reply.ret_code = !reply.ret_code;
3092 }
3093
3094 DBG2("Health check return value %d", reply.ret_code);
3095
3096 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3097 if (ret < 0) {
3098 ERR("Failed to send health data back to client");
3099 }
3100
3101 /* End of transmission */
3102 ret = close(new_sock);
3103 if (ret) {
3104 PERROR("close");
3105 }
3106 new_sock = -1;
3107 }
3108
3109 exit:
3110 error:
3111 if (err) {
3112 ERR("Health error occurred in %s", __func__);
3113 }
3114 DBG("Health check thread dying");
3115 unlink(health_unix_sock_path);
3116 if (sock >= 0) {
3117 ret = close(sock);
3118 if (ret) {
3119 PERROR("close");
3120 }
3121 }
3122 if (new_sock >= 0) {
3123 ret = close(new_sock);
3124 if (ret) {
3125 PERROR("close");
3126 }
3127 }
3128
3129 lttng_poll_clean(&events);
3130
3131 rcu_unregister_thread();
3132 return NULL;
3133 }
3134
3135 /*
3136 * This thread manage all clients request using the unix client socket for
3137 * communication.
3138 */
3139 static void *thread_manage_clients(void *data)
3140 {
3141 int sock = -1, ret, i, pollfd, err = -1;
3142 int sock_error;
3143 uint32_t revents, nb_fd;
3144 struct command_ctx *cmd_ctx = NULL;
3145 struct lttng_poll_event events;
3146
3147 DBG("[thread] Manage client started");
3148
3149 rcu_register_thread();
3150
3151 if (testpoint(thread_manage_clients)) {
3152 goto error_testpoint;
3153 }
3154
3155 health_code_update(&health_thread_cmd);
3156
3157 ret = lttcomm_listen_unix_sock(client_sock);
3158 if (ret < 0) {
3159 goto error_listen;
3160 }
3161
3162 /*
3163 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3164 * more will be added to this poll set.
3165 */
3166 ret = create_thread_poll_set(&events, 2);
3167 if (ret < 0) {
3168 goto error_create_poll;
3169 }
3170
3171 /* Add the application registration socket */
3172 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3173 if (ret < 0) {
3174 goto error;
3175 }
3176
3177 /*
3178 * Notify parent pid that we are ready to accept command for client side.
3179 */
3180 if (opt_sig_parent) {
3181 kill(ppid, SIGUSR1);
3182 }
3183
3184 if (testpoint(thread_manage_clients_before_loop)) {
3185 goto error;
3186 }
3187
3188 health_code_update(&health_thread_cmd);
3189
3190 while (1) {
3191 DBG("Accepting client command ...");
3192
3193 /* Inifinite blocking call, waiting for transmission */
3194 restart:
3195 health_poll_update(&health_thread_cmd);
3196 ret = lttng_poll_wait(&events, -1);
3197 health_poll_update(&health_thread_cmd);
3198 if (ret < 0) {
3199 /*
3200 * Restart interrupted system call.
3201 */
3202 if (errno == EINTR) {
3203 goto restart;
3204 }
3205 goto error;
3206 }
3207
3208 nb_fd = ret;
3209
3210 for (i = 0; i < nb_fd; i++) {
3211 /* Fetch once the poll data */
3212 revents = LTTNG_POLL_GETEV(&events, i);
3213 pollfd = LTTNG_POLL_GETFD(&events, i);
3214
3215 health_code_update(&health_thread_cmd);
3216
3217 /* Thread quit pipe has been closed. Killing thread. */
3218 ret = check_thread_quit_pipe(pollfd, revents);
3219 if (ret) {
3220 err = 0;
3221 goto exit;
3222 }
3223
3224 /* Event on the registration socket */
3225 if (pollfd == client_sock) {
3226 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3227 ERR("Client socket poll error");
3228 goto error;
3229 }
3230 }
3231 }
3232
3233 DBG("Wait for client response");
3234
3235 health_code_update(&health_thread_cmd);
3236
3237 sock = lttcomm_accept_unix_sock(client_sock);
3238 if (sock < 0) {
3239 goto error;
3240 }
3241
3242 /*
3243 * Set the CLOEXEC flag. Return code is useless because either way, the
3244 * show must go on.
3245 */
3246 (void) utils_set_fd_cloexec(sock);
3247
3248 /* Set socket option for credentials retrieval */
3249 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3250 if (ret < 0) {
3251 goto error;
3252 }
3253
3254 /* Allocate context command to process the client request */
3255 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3256 if (cmd_ctx == NULL) {
3257 PERROR("zmalloc cmd_ctx");
3258 goto error;
3259 }
3260
3261 /* Allocate data buffer for reception */
3262 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3263 if (cmd_ctx->lsm == NULL) {
3264 PERROR("zmalloc cmd_ctx->lsm");
3265 goto error;
3266 }
3267
3268 cmd_ctx->llm = NULL;
3269 cmd_ctx->session = NULL;
3270
3271 health_code_update(&health_thread_cmd);
3272
3273 /*
3274 * Data is received from the lttng client. The struct
3275 * lttcomm_session_msg (lsm) contains the command and data request of
3276 * the client.
3277 */
3278 DBG("Receiving data from client ...");
3279 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3280 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3281 if (ret <= 0) {
3282 DBG("Nothing recv() from client... continuing");
3283 ret = close(sock);
3284 if (ret) {
3285 PERROR("close");
3286 }
3287 sock = -1;
3288 clean_command_ctx(&cmd_ctx);
3289 continue;
3290 }
3291
3292 health_code_update(&health_thread_cmd);
3293
3294 // TODO: Validate cmd_ctx including sanity check for
3295 // security purpose.
3296
3297 rcu_thread_online();
3298 /*
3299 * This function dispatch the work to the kernel or userspace tracer
3300 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3301 * informations for the client. The command context struct contains
3302 * everything this function may needs.
3303 */
3304 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3305 rcu_thread_offline();
3306 if (ret < 0) {
3307 if (sock_error) {
3308 ret = close(sock);
3309 if (ret) {
3310 PERROR("close");
3311 }
3312 sock = -1;
3313 }
3314 /*
3315 * TODO: Inform client somehow of the fatal error. At
3316 * this point, ret < 0 means that a zmalloc failed
3317 * (ENOMEM). Error detected but still accept
3318 * command, unless a socket error has been
3319 * detected.
3320 */
3321 clean_command_ctx(&cmd_ctx);
3322 continue;
3323 }
3324
3325 health_code_update(&health_thread_cmd);
3326
3327 DBG("Sending response (size: %d, retcode: %s)",
3328 cmd_ctx->lttng_msg_size,
3329 lttng_strerror(-cmd_ctx->llm->ret_code));
3330 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3331 if (ret < 0) {
3332 ERR("Failed to send data back to client");
3333 }
3334
3335 /* End of transmission */
3336 ret = close(sock);
3337 if (ret) {
3338 PERROR("close");
3339 }
3340 sock = -1;
3341
3342 clean_command_ctx(&cmd_ctx);
3343
3344 health_code_update(&health_thread_cmd);
3345 }
3346
3347 exit:
3348 error:
3349 if (sock >= 0) {
3350 ret = close(sock);
3351 if (ret) {
3352 PERROR("close");
3353 }
3354 }
3355
3356 lttng_poll_clean(&events);
3357 clean_command_ctx(&cmd_ctx);
3358
3359 error_listen:
3360 error_create_poll:
3361 error_testpoint:
3362 unlink(client_unix_sock_path);
3363 if (client_sock >= 0) {
3364 ret = close(client_sock);
3365 if (ret) {
3366 PERROR("close");
3367 }
3368 }
3369
3370 if (err) {
3371 health_error(&health_thread_cmd);
3372 ERR("Health error occurred in %s", __func__);
3373 }
3374
3375 health_exit(&health_thread_cmd);
3376
3377 DBG("Client thread dying");
3378
3379 rcu_unregister_thread();
3380 return NULL;
3381 }
3382
3383
3384 /*
3385 * usage function on stderr
3386 */
3387 static void usage(void)
3388 {
3389 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3390 fprintf(stderr, " -h, --help Display this usage.\n");
3391 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3392 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3393 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3394 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3395 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3396 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3397 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3398 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3399 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3400 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3401 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3402 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3403 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3404 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3405 fprintf(stderr, " -V, --version Show version number.\n");
3406 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3407 fprintf(stderr, " -q, --quiet No output at all.\n");
3408 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3409 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3410 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3411 }
3412
3413 /*
3414 * daemon argument parsing
3415 */
3416 static int parse_args(int argc, char **argv)
3417 {
3418 int c;
3419
3420 static struct option long_options[] = {
3421 { "client-sock", 1, 0, 'c' },
3422 { "apps-sock", 1, 0, 'a' },
3423 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3424 { "kconsumerd-err-sock", 1, 0, 'E' },
3425 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3426 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3427 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3428 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3429 { "consumerd32-path", 1, 0, 'u' },
3430 { "consumerd32-libdir", 1, 0, 'U' },
3431 { "consumerd64-path", 1, 0, 't' },
3432 { "consumerd64-libdir", 1, 0, 'T' },
3433 { "daemonize", 0, 0, 'd' },
3434 { "sig-parent", 0, 0, 'S' },
3435 { "help", 0, 0, 'h' },
3436 { "group", 1, 0, 'g' },
3437 { "version", 0, 0, 'V' },
3438 { "quiet", 0, 0, 'q' },
3439 { "verbose", 0, 0, 'v' },
3440 { "verbose-consumer", 0, 0, 'Z' },
3441 { "no-kernel", 0, 0, 'N' },
3442 { NULL, 0, 0, 0 }
3443 };
3444
3445 while (1) {
3446 int option_index = 0;
3447 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3448 long_options, &option_index);
3449 if (c == -1) {
3450 break;
3451 }
3452
3453 switch (c) {
3454 case 0:
3455 fprintf(stderr, "option %s", long_options[option_index].name);
3456 if (optarg) {
3457 fprintf(stderr, " with arg %s\n", optarg);
3458 }
3459 break;
3460 case 'c':
3461 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3462 break;
3463 case 'a':
3464 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3465 break;
3466 case 'd':
3467 opt_daemon = 1;
3468 break;
3469 case 'g':
3470 opt_tracing_group = optarg;
3471 break;
3472 case 'h':
3473 usage();
3474 exit(EXIT_FAILURE);
3475 case 'V':
3476 fprintf(stdout, "%s\n", VERSION);
3477 exit(EXIT_SUCCESS);
3478 case 'S':
3479 opt_sig_parent = 1;
3480 break;
3481 case 'E':
3482 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3483 break;
3484 case 'C':
3485 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3486 break;
3487 case 'F':
3488 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3489 break;
3490 case 'D':
3491 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3492 break;
3493 case 'H':
3494 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3495 break;
3496 case 'G':
3497 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3498 break;
3499 case 'N':
3500 opt_no_kernel = 1;
3501 break;
3502 case 'q':
3503 lttng_opt_quiet = 1;
3504 break;
3505 case 'v':
3506 /* Verbose level can increase using multiple -v */
3507 lttng_opt_verbose += 1;
3508 break;
3509 case 'Z':
3510 opt_verbose_consumer += 1;
3511 break;
3512 case 'u':
3513 consumerd32_bin= optarg;
3514 break;
3515 case 'U':
3516 consumerd32_libdir = optarg;
3517 break;
3518 case 't':
3519 consumerd64_bin = optarg;
3520 break;
3521 case 'T':
3522 consumerd64_libdir = optarg;
3523 break;
3524 default:
3525 /* Unknown option or other error.
3526 * Error is printed by getopt, just return */
3527 return -1;
3528 }
3529 }
3530
3531 return 0;
3532 }
3533
3534 /*
3535 * Creates the two needed socket by the daemon.
3536 * apps_sock - The communication socket for all UST apps.
3537 * client_sock - The communication of the cli tool (lttng).
3538 */
3539 static int init_daemon_socket(void)
3540 {
3541 int ret = 0;
3542 mode_t old_umask;
3543
3544 old_umask = umask(0);
3545
3546 /* Create client tool unix socket */
3547 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3548 if (client_sock < 0) {
3549 ERR("Create unix sock failed: %s", client_unix_sock_path);
3550 ret = -1;
3551 goto end;
3552 }
3553
3554 /* Set the cloexec flag */
3555 ret = utils_set_fd_cloexec(client_sock);
3556 if (ret < 0) {
3557 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3558 "Continuing but note that the consumer daemon will have a "
3559 "reference to this socket on exec()", client_sock);
3560 }
3561
3562 /* File permission MUST be 660 */
3563 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3564 if (ret < 0) {
3565 ERR("Set file permissions failed: %s", client_unix_sock_path);
3566 PERROR("chmod");
3567 goto end;
3568 }
3569
3570 /* Create the application unix socket */
3571 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3572 if (apps_sock < 0) {
3573 ERR("Create unix sock failed: %s", apps_unix_sock_path);
3574 ret = -1;
3575 goto end;
3576 }
3577
3578 /* Set the cloexec flag */
3579 ret = utils_set_fd_cloexec(apps_sock);
3580 if (ret < 0) {
3581 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3582 "Continuing but note that the consumer daemon will have a "
3583 "reference to this socket on exec()", apps_sock);
3584 }
3585
3586 /* File permission MUST be 666 */
3587 ret = chmod(apps_unix_sock_path,
3588 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
3589 if (ret < 0) {
3590 ERR("Set file permissions failed: %s", apps_unix_sock_path);
3591 PERROR("chmod");
3592 goto end;
3593 }
3594
3595 DBG3("Session daemon client socket %d and application socket %d created",
3596 client_sock, apps_sock);
3597
3598 end:
3599 umask(old_umask);
3600 return ret;
3601 }
3602
3603 /*
3604 * Check if the global socket is available, and if a daemon is answering at the
3605 * other side. If yes, error is returned.
3606 */
3607 static int check_existing_daemon(void)
3608 {
3609 /* Is there anybody out there ? */
3610 if (lttng_session_daemon_alive()) {
3611 return -EEXIST;
3612 }
3613
3614 return 0;
3615 }
3616
3617 /*
3618 * Set the tracing group gid onto the client socket.
3619 *
3620 * Race window between mkdir and chown is OK because we are going from more
3621 * permissive (root.root) to less permissive (root.tracing).
3622 */
3623 static int set_permissions(char *rundir)
3624 {
3625 int ret;
3626 gid_t gid;
3627
3628 ret = allowed_group();
3629 if (ret < 0) {
3630 WARN("No tracing group detected");
3631 ret = 0;
3632 goto end;
3633 }
3634
3635 gid = ret;
3636
3637 /* Set lttng run dir */
3638 ret = chown(rundir, 0, gid);
3639 if (ret < 0) {
3640 ERR("Unable to set group on %s", rundir);
3641 PERROR("chown");
3642 }
3643
3644 /* Ensure tracing group can search the run dir */
3645 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
3646 if (ret < 0) {
3647 ERR("Unable to set permissions on %s", rundir);
3648 PERROR("chmod");
3649 }
3650
3651 /* lttng client socket path */
3652 ret = chown(client_unix_sock_path, 0, gid);
3653 if (ret < 0) {
3654 ERR("Unable to set group on %s", client_unix_sock_path);
3655 PERROR("chown");
3656 }
3657
3658 /* kconsumer error socket path */
3659 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
3660 if (ret < 0) {
3661 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
3662 PERROR("chown");
3663 }
3664
3665 /* 64-bit ustconsumer error socket path */
3666 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3667 if (ret < 0) {
3668 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
3669 PERROR("chown");
3670 }
3671
3672 /* 32-bit ustconsumer compat32 error socket path */
3673 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3674 if (ret < 0) {
3675 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
3676 PERROR("chown");
3677 }
3678
3679 DBG("All permissions are set");
3680
3681 end:
3682 return ret;
3683 }
3684
3685 /*
3686 * Create the lttng run directory needed for all global sockets and pipe.
3687 */
3688 static int create_lttng_rundir(const char *rundir)
3689 {
3690 int ret;
3691
3692 DBG3("Creating LTTng run directory: %s", rundir);
3693
3694 ret = mkdir(rundir, S_IRWXU);
3695 if (ret < 0) {
3696 if (errno != EEXIST) {
3697 ERR("Unable to create %s", rundir);
3698 goto error;
3699 } else {
3700 ret = 0;
3701 }
3702 }
3703
3704 error:
3705 return ret;
3706 }
3707
3708 /*
3709 * Setup sockets and directory needed by the kconsumerd communication with the
3710 * session daemon.
3711 */
3712 static int set_consumer_sockets(struct consumer_data *consumer_data,
3713 const char *rundir)
3714 {
3715 int ret;
3716 char path[PATH_MAX];
3717
3718 switch (consumer_data->type) {
3719 case LTTNG_CONSUMER_KERNEL:
3720 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
3721 break;
3722 case LTTNG_CONSUMER64_UST:
3723 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
3724 break;
3725 case LTTNG_CONSUMER32_UST:
3726 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
3727 break;
3728 default:
3729 ERR("Consumer type unknown");
3730 ret = -EINVAL;
3731 goto error;
3732 }
3733
3734 DBG2("Creating consumer directory: %s", path);
3735
3736 ret = mkdir(path, S_IRWXU);
3737 if (ret < 0) {
3738 if (errno != EEXIST) {
3739 PERROR("mkdir");
3740 ERR("Failed to create %s", path);
3741 goto error;
3742 }
3743 ret = -1;
3744 }
3745
3746 /* Create the kconsumerd error unix socket */
3747 consumer_data->err_sock =
3748 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
3749 if (consumer_data->err_sock < 0) {
3750 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
3751 ret = -1;
3752 goto error;
3753 }
3754
3755 /* File permission MUST be 660 */
3756 ret = chmod(consumer_data->err_unix_sock_path,
3757 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3758 if (ret < 0) {
3759 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
3760 PERROR("chmod");
3761 goto error;
3762 }
3763
3764 error:
3765 return ret;
3766 }
3767
3768 /*
3769 * Signal handler for the daemon
3770 *
3771 * Simply stop all worker threads, leaving main() return gracefully after
3772 * joining all threads and calling cleanup().
3773 */
3774 static void sighandler(int sig)
3775 {
3776 switch (sig) {
3777 case SIGPIPE:
3778 DBG("SIGPIPE caught");
3779 return;
3780 case SIGINT:
3781 DBG("SIGINT caught");
3782 stop_threads();
3783 break;
3784 case SIGTERM:
3785 DBG("SIGTERM caught");
3786 stop_threads();
3787 break;
3788 default:
3789 break;
3790 }
3791 }
3792
3793 /*
3794 * Setup signal handler for :
3795 * SIGINT, SIGTERM, SIGPIPE
3796 */
3797 static int set_signal_handler(void)
3798 {
3799 int ret = 0;
3800 struct sigaction sa;
3801 sigset_t sigset;
3802
3803 if ((ret = sigemptyset(&sigset)) < 0) {
3804 PERROR("sigemptyset");
3805 return ret;
3806 }
3807
3808 sa.sa_handler = sighandler;
3809 sa.sa_mask = sigset;
3810 sa.sa_flags = 0;
3811 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
3812 PERROR("sigaction");
3813 return ret;
3814 }
3815
3816 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
3817 PERROR("sigaction");
3818 return ret;
3819 }
3820
3821 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3822 PERROR("sigaction");
3823 return ret;
3824 }
3825
3826 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
3827
3828 return ret;
3829 }
3830
3831 /*
3832 * Set open files limit to unlimited. This daemon can open a large number of
3833 * file descriptors in order to consumer multiple kernel traces.
3834 */
3835 static void set_ulimit(void)
3836 {
3837 int ret;
3838 struct rlimit lim;
3839
3840 /* The kernel does not allowed an infinite limit for open files */
3841 lim.rlim_cur = 65535;
3842 lim.rlim_max = 65535;
3843
3844 ret = setrlimit(RLIMIT_NOFILE, &lim);
3845 if (ret < 0) {
3846 PERROR("failed to set open files limit");
3847 }
3848 }
3849
3850 /*
3851 * main
3852 */
3853 int main(int argc, char **argv)
3854 {
3855 int ret = 0;
3856 void *status;
3857 const char *home_path, *env_app_timeout;
3858
3859 init_kernel_workarounds();
3860
3861 rcu_register_thread();
3862
3863 setup_consumerd_path();
3864
3865 /* Parse arguments */
3866 progname = argv[0];
3867 if ((ret = parse_args(argc, argv)) < 0) {
3868 goto error;
3869 }
3870
3871 /* Daemonize */
3872 if (opt_daemon) {
3873 int i;
3874
3875 /*
3876 * fork
3877 * child: setsid, close FD 0, 1, 2, chdir /
3878 * parent: exit (if fork is successful)
3879 */
3880 ret = daemon(0, 0);
3881 if (ret < 0) {
3882 PERROR("daemon");
3883 goto error;
3884 }
3885 /*
3886 * We are in the child. Make sure all other file
3887 * descriptors are closed, in case we are called with
3888 * more opened file descriptors than the standard ones.
3889 */
3890 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3891 (void) close(i);
3892 }
3893 }
3894
3895 /* Create thread quit pipe */
3896 if ((ret = init_thread_quit_pipe()) < 0) {
3897 goto error;
3898 }
3899
3900 /* Check if daemon is UID = 0 */
3901 is_root = !getuid();
3902
3903 if (is_root) {
3904 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
3905
3906 /* Create global run dir with root access */
3907 ret = create_lttng_rundir(rundir);
3908 if (ret < 0) {
3909 goto error;
3910 }
3911
3912 if (strlen(apps_unix_sock_path) == 0) {
3913 snprintf(apps_unix_sock_path, PATH_MAX,
3914 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
3915 }
3916
3917 if (strlen(client_unix_sock_path) == 0) {
3918 snprintf(client_unix_sock_path, PATH_MAX,
3919 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
3920 }
3921
3922 /* Set global SHM for ust */
3923 if (strlen(wait_shm_path) == 0) {
3924 snprintf(wait_shm_path, PATH_MAX,
3925 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
3926 }
3927
3928 if (strlen(health_unix_sock_path) == 0) {
3929 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3930 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
3931 }
3932
3933 /* Setup kernel consumerd path */
3934 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
3935 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
3936 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
3937 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
3938
3939 DBG2("Kernel consumer err path: %s",
3940 kconsumer_data.err_unix_sock_path);
3941 DBG2("Kernel consumer cmd path: %s",
3942 kconsumer_data.cmd_unix_sock_path);
3943 } else {
3944 home_path = get_home_dir();
3945 if (home_path == NULL) {
3946 /* TODO: Add --socket PATH option */
3947 ERR("Can't get HOME directory for sockets creation.");
3948 ret = -EPERM;
3949 goto error;
3950 }
3951
3952 /*
3953 * Create rundir from home path. This will create something like
3954 * $HOME/.lttng
3955 */
3956 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
3957 if (ret < 0) {
3958 ret = -ENOMEM;
3959 goto error;
3960 }
3961
3962 ret = create_lttng_rundir(rundir);
3963 if (ret < 0) {
3964 goto error;
3965 }
3966
3967 if (strlen(apps_unix_sock_path) == 0) {
3968 snprintf(apps_unix_sock_path, PATH_MAX,
3969 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
3970 }
3971
3972 /* Set the cli tool unix socket path */
3973 if (strlen(client_unix_sock_path) == 0) {
3974 snprintf(client_unix_sock_path, PATH_MAX,
3975 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
3976 }
3977
3978 /* Set global SHM for ust */
3979 if (strlen(wait_shm_path) == 0) {
3980 snprintf(wait_shm_path, PATH_MAX,
3981 DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid());
3982 }
3983
3984 /* Set health check Unix path */
3985 if (strlen(health_unix_sock_path) == 0) {
3986 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
3987 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
3988 }
3989 }
3990
3991 /* Set consumer initial state */
3992 kernel_consumerd_state = CONSUMER_STOPPED;
3993 ust_consumerd_state = CONSUMER_STOPPED;
3994
3995 DBG("Client socket path %s", client_unix_sock_path);
3996 DBG("Application socket path %s", apps_unix_sock_path);
3997 DBG("LTTng run directory path: %s", rundir);
3998
3999 /* 32 bits consumerd path setup */
4000 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4001 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4002 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4003 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4004
4005 DBG2("UST consumer 32 bits err path: %s",
4006 ustconsumer32_data.err_unix_sock_path);
4007 DBG2("UST consumer 32 bits cmd path: %s",
4008 ustconsumer32_data.cmd_unix_sock_path);
4009
4010 /* 64 bits consumerd path setup */
4011 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4012 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4013 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4014 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4015
4016 DBG2("UST consumer 64 bits err path: %s",
4017 ustconsumer64_data.err_unix_sock_path);
4018 DBG2("UST consumer 64 bits cmd path: %s",
4019 ustconsumer64_data.cmd_unix_sock_path);
4020
4021 /*
4022 * See if daemon already exist.
4023 */
4024 if ((ret = check_existing_daemon()) < 0) {
4025 ERR("Already running daemon.\n");
4026 /*
4027 * We do not goto exit because we must not cleanup()
4028 * because a daemon is already running.
4029 */
4030 goto error;
4031 }
4032
4033 /*
4034 * Init UST app hash table. Alloc hash table before this point since
4035 * cleanup() can get called after that point.
4036 */
4037 ust_app_ht_alloc();
4038
4039 /* After this point, we can safely call cleanup() with "goto exit" */
4040
4041 /*
4042 * These actions must be executed as root. We do that *after* setting up
4043 * the sockets path because we MUST make the check for another daemon using
4044 * those paths *before* trying to set the kernel consumer sockets and init
4045 * kernel tracer.
4046 */
4047 if (is_root) {
4048 ret = set_consumer_sockets(&kconsumer_data, rundir);
4049 if (ret < 0) {
4050 goto exit;
4051 }
4052
4053 /* Setup kernel tracer */
4054 if (!opt_no_kernel) {
4055 init_kernel_tracer();
4056 }
4057
4058 /* Set ulimit for open files */
4059 set_ulimit();
4060 }
4061 /* init lttng_fd tracking must be done after set_ulimit. */
4062 lttng_fd_init();
4063
4064 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4065 if (ret < 0) {
4066 goto exit;
4067 }
4068
4069 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4070 if (ret < 0) {
4071 goto exit;
4072 }
4073
4074 if ((ret = set_signal_handler()) < 0) {
4075 goto exit;
4076 }
4077
4078 /* Setup the needed unix socket */
4079 if ((ret = init_daemon_socket()) < 0) {
4080 goto exit;
4081 }
4082
4083 /* Set credentials to socket */
4084 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4085 goto exit;
4086 }
4087
4088 /* Get parent pid if -S, --sig-parent is specified. */
4089 if (opt_sig_parent) {
4090 ppid = getppid();
4091 }
4092
4093 /* Setup the kernel pipe for waking up the kernel thread */
4094 if (is_root && !opt_no_kernel) {
4095 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4096 goto exit;
4097 }
4098 }
4099
4100 /* Setup the thread apps communication pipe. */
4101 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4102 goto exit;
4103 }
4104
4105 /* Init UST command queue. */
4106 cds_wfq_init(&ust_cmd_queue.queue);
4107
4108 /*
4109 * Get session list pointer. This pointer MUST NOT be free(). This list is
4110 * statically declared in session.c
4111 */
4112 session_list_ptr = session_get_list();
4113
4114 /* Set up max poll set size */
4115 lttng_poll_set_max_size();
4116
4117 cmd_init();
4118
4119 /* Init all health thread counters. */
4120 health_init(&health_thread_cmd);
4121 health_init(&health_thread_kernel);
4122 health_init(&health_thread_app_manage);
4123 health_init(&health_thread_app_reg);
4124
4125 /*
4126 * Init health counters of the consumer thread. We do a quick hack here to
4127 * the state of the consumer health is fine even if the thread is not
4128 * started. Once the thread starts, the health state is updated with a poll
4129 * value to set a health code path. This is simply to ease our life and has
4130 * no cost what so ever.
4131 */
4132 health_init(&kconsumer_data.health);
4133 health_poll_update(&kconsumer_data.health);
4134 health_init(&ustconsumer32_data.health);
4135 health_poll_update(&ustconsumer32_data.health);
4136 health_init(&ustconsumer64_data.health);
4137 health_poll_update(&ustconsumer64_data.health);
4138
4139 /* Check for the application socket timeout env variable. */
4140 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4141 if (env_app_timeout) {
4142 app_socket_timeout = atoi(env_app_timeout);
4143 } else {
4144 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4145 }
4146
4147 /* Create thread to manage the client socket */
4148 ret = pthread_create(&health_thread, NULL,
4149 thread_manage_health, (void *) NULL);
4150 if (ret != 0) {
4151 PERROR("pthread_create health");
4152 goto exit_health;
4153 }
4154
4155 /* Create thread to manage the client socket */
4156 ret = pthread_create(&client_thread, NULL,
4157 thread_manage_clients, (void *) NULL);
4158 if (ret != 0) {
4159 PERROR("pthread_create clients");
4160 goto exit_client;
4161 }
4162
4163 /* Create thread to dispatch registration */
4164 ret = pthread_create(&dispatch_thread, NULL,
4165 thread_dispatch_ust_registration, (void *) NULL);
4166 if (ret != 0) {
4167 PERROR("pthread_create dispatch");
4168 goto exit_dispatch;
4169 }
4170
4171 /* Create thread to manage application registration. */
4172 ret = pthread_create(&reg_apps_thread, NULL,
4173 thread_registration_apps, (void *) NULL);
4174 if (ret != 0) {
4175 PERROR("pthread_create registration");
4176 goto exit_reg_apps;
4177 }
4178
4179 /* Create thread to manage application socket */
4180 ret = pthread_create(&apps_thread, NULL,
4181 thread_manage_apps, (void *) NULL);
4182 if (ret != 0) {
4183 PERROR("pthread_create apps");
4184 goto exit_apps;
4185 }
4186
4187 /* Don't start this thread if kernel tracing is not requested nor root */
4188 if (is_root && !opt_no_kernel) {
4189 /* Create kernel thread to manage kernel event */
4190 ret = pthread_create(&kernel_thread, NULL,
4191 thread_manage_kernel, (void *) NULL);
4192 if (ret != 0) {
4193 PERROR("pthread_create kernel");
4194 goto exit_kernel;
4195 }
4196
4197 ret = pthread_join(kernel_thread, &status);
4198 if (ret != 0) {
4199 PERROR("pthread_join");
4200 goto error; /* join error, exit without cleanup */
4201 }
4202 }
4203
4204 exit_kernel:
4205 ret = pthread_join(apps_thread, &status);
4206 if (ret != 0) {
4207 PERROR("pthread_join");
4208 goto error; /* join error, exit without cleanup */
4209 }
4210
4211 exit_apps:
4212 ret = pthread_join(reg_apps_thread, &status);
4213 if (ret != 0) {
4214 PERROR("pthread_join");
4215 goto error; /* join error, exit without cleanup */
4216 }
4217
4218 exit_reg_apps:
4219 ret = pthread_join(dispatch_thread, &status);
4220 if (ret != 0) {
4221 PERROR("pthread_join");
4222 goto error; /* join error, exit without cleanup */
4223 }
4224
4225 exit_dispatch:
4226 ret = pthread_join(client_thread, &status);
4227 if (ret != 0) {
4228 PERROR("pthread_join");
4229 goto error; /* join error, exit without cleanup */
4230 }
4231
4232 ret = join_consumer_thread(&kconsumer_data);
4233 if (ret != 0) {
4234 PERROR("join_consumer");
4235 goto error; /* join error, exit without cleanup */
4236 }
4237
4238 ret = join_consumer_thread(&ustconsumer32_data);
4239 if (ret != 0) {
4240 PERROR("join_consumer ust32");
4241 goto error; /* join error, exit without cleanup */
4242 }
4243
4244 ret = join_consumer_thread(&ustconsumer64_data);
4245 if (ret != 0) {
4246 PERROR("join_consumer ust64");
4247 goto error; /* join error, exit without cleanup */
4248 }
4249
4250 exit_client:
4251 ret = pthread_join(health_thread, &status);
4252 if (ret != 0) {
4253 PERROR("pthread_join health thread");
4254 goto error; /* join error, exit without cleanup */
4255 }
4256
4257 exit_health:
4258 exit:
4259 /*
4260 * cleanup() is called when no other thread is running.
4261 */
4262 rcu_thread_online();
4263 cleanup();
4264 rcu_thread_offline();
4265 rcu_unregister_thread();
4266 if (!ret) {
4267 exit(EXIT_SUCCESS);
4268 }
4269 error:
4270 exit(EXIT_FAILURE);
4271 }
This page took 0.161734 seconds and 5 git commands to generate.