Add lttng-error.h containing every API err. code
[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 <semaphore.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <urcu/uatomic.h>
37 #include <unistd.h>
38 #include <config.h>
39
40 #include <common/common.h>
41 #include <common/compat/poll.h>
42 #include <common/compat/socket.h>
43 #include <common/defaults.h>
44 #include <common/kernel-consumer/kernel-consumer.h>
45 #include <common/futex.h>
46 #include <common/relayd/relayd.h>
47 #include <common/utils.h>
48
49 #include "lttng-sessiond.h"
50 #include "channel.h"
51 #include "cmd.h"
52 #include "consumer.h"
53 #include "context.h"
54 #include "event.h"
55 #include "kernel.h"
56 #include "kernel-consumer.h"
57 #include "modprobe.h"
58 #include "shm.h"
59 #include "ust-ctl.h"
60 #include "ust-consumer.h"
61 #include "utils.h"
62 #include "fd-limit.h"
63 #include "filter.h"
64 #include "health.h"
65
66 #define CONSUMERD_FILE "lttng-consumerd"
67
68 /* Const values */
69 const char default_home_dir[] = DEFAULT_HOME_DIR;
70 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
71 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
72 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
73
74 const char *progname;
75 const char *opt_tracing_group;
76 static int opt_sig_parent;
77 static int opt_verbose_consumer;
78 static int opt_daemon;
79 static int opt_no_kernel;
80 static int is_root; /* Set to 1 if the daemon is running as root */
81 static pid_t ppid; /* Parent PID for --sig-parent option */
82 static char *rundir;
83
84 /* Consumer daemon specific control data */
85 static struct consumer_data kconsumer_data = {
86 .type = LTTNG_CONSUMER_KERNEL,
87 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
88 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
89 .err_sock = -1,
90 .cmd_sock = -1,
91 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
92 .lock = PTHREAD_MUTEX_INITIALIZER,
93 };
94 static struct consumer_data ustconsumer64_data = {
95 .type = LTTNG_CONSUMER64_UST,
96 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
97 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
98 .err_sock = -1,
99 .cmd_sock = -1,
100 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
101 .lock = PTHREAD_MUTEX_INITIALIZER,
102 };
103 static struct consumer_data ustconsumer32_data = {
104 .type = LTTNG_CONSUMER32_UST,
105 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
106 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
107 .err_sock = -1,
108 .cmd_sock = -1,
109 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
110 .lock = PTHREAD_MUTEX_INITIALIZER,
111 };
112
113 /* Shared between threads */
114 static int dispatch_thread_exit;
115
116 /* Global application Unix socket path */
117 static char apps_unix_sock_path[PATH_MAX];
118 /* Global client Unix socket path */
119 static char client_unix_sock_path[PATH_MAX];
120 /* global wait shm path for UST */
121 static char wait_shm_path[PATH_MAX];
122 /* Global health check unix path */
123 static char health_unix_sock_path[PATH_MAX];
124
125 /* Sockets and FDs */
126 static int client_sock = -1;
127 static int apps_sock = -1;
128 int kernel_tracer_fd = -1;
129 static int kernel_poll_pipe[2] = { -1, -1 };
130
131 /*
132 * Quit pipe for all threads. This permits a single cancellation point
133 * for all threads when receiving an event on the pipe.
134 */
135 static int thread_quit_pipe[2] = { -1, -1 };
136
137 /*
138 * This pipe is used to inform the thread managing application communication
139 * that a command is queued and ready to be processed.
140 */
141 static int apps_cmd_pipe[2] = { -1, -1 };
142
143 /* Pthread, Mutexes and Semaphores */
144 static pthread_t apps_thread;
145 static pthread_t reg_apps_thread;
146 static pthread_t client_thread;
147 static pthread_t kernel_thread;
148 static pthread_t dispatch_thread;
149 static pthread_t health_thread;
150
151 /*
152 * UST registration command queue. This queue is tied with a futex and uses a N
153 * wakers / 1 waiter implemented and detailed in futex.c/.h
154 *
155 * The thread_manage_apps and thread_dispatch_ust_registration interact with
156 * this queue and the wait/wake scheme.
157 */
158 static struct ust_cmd_queue ust_cmd_queue;
159
160 /*
161 * Pointer initialized before thread creation.
162 *
163 * This points to the tracing session list containing the session count and a
164 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
165 * MUST NOT be taken if you call a public function in session.c.
166 *
167 * The lock is nested inside the structure: session_list_ptr->lock. Please use
168 * session_lock_list and session_unlock_list for lock acquisition.
169 */
170 static struct ltt_session_list *session_list_ptr;
171
172 int ust_consumerd64_fd = -1;
173 int ust_consumerd32_fd = -1;
174
175 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
176 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
177 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
178 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
179
180 static const char *module_proc_lttng = "/proc/lttng";
181
182 /*
183 * Consumer daemon state which is changed when spawning it, killing it or in
184 * case of a fatal error.
185 */
186 enum consumerd_state {
187 CONSUMER_STARTED = 1,
188 CONSUMER_STOPPED = 2,
189 CONSUMER_ERROR = 3,
190 };
191
192 /*
193 * This consumer daemon state is used to validate if a client command will be
194 * able to reach the consumer. If not, the client is informed. For instance,
195 * doing a "lttng start" when the consumer state is set to ERROR will return an
196 * error to the client.
197 *
198 * The following example shows a possible race condition of this scheme:
199 *
200 * consumer thread error happens
201 * client cmd arrives
202 * client cmd checks state -> still OK
203 * consumer thread exit, sets error
204 * client cmd try to talk to consumer
205 * ...
206 *
207 * However, since the consumer is a different daemon, we have no way of making
208 * sure the command will reach it safely even with this state flag. This is why
209 * we consider that up to the state validation during command processing, the
210 * command is safe. After that, we can not guarantee the correctness of the
211 * client request vis-a-vis the consumer.
212 */
213 static enum consumerd_state ust_consumerd_state;
214 static enum consumerd_state kernel_consumerd_state;
215
216 /* Used for the health monitoring of the session daemon. See health.h */
217 struct health_state health_thread_cmd;
218 struct health_state health_thread_app_manage;
219 struct health_state health_thread_app_reg;
220 struct health_state health_thread_kernel;
221
222 static
223 void setup_consumerd_path(void)
224 {
225 const char *bin, *libdir;
226
227 /*
228 * Allow INSTALL_BIN_PATH to be used as a target path for the
229 * native architecture size consumer if CONFIG_CONSUMER*_PATH
230 * has not been defined.
231 */
232 #if (CAA_BITS_PER_LONG == 32)
233 if (!consumerd32_bin[0]) {
234 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
235 }
236 if (!consumerd32_libdir[0]) {
237 consumerd32_libdir = INSTALL_LIB_PATH;
238 }
239 #elif (CAA_BITS_PER_LONG == 64)
240 if (!consumerd64_bin[0]) {
241 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
242 }
243 if (!consumerd64_libdir[0]) {
244 consumerd64_libdir = INSTALL_LIB_PATH;
245 }
246 #else
247 #error "Unknown bitness"
248 #endif
249
250 /*
251 * runtime env. var. overrides the build default.
252 */
253 bin = getenv("LTTNG_CONSUMERD32_BIN");
254 if (bin) {
255 consumerd32_bin = bin;
256 }
257 bin = getenv("LTTNG_CONSUMERD64_BIN");
258 if (bin) {
259 consumerd64_bin = bin;
260 }
261 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
262 if (libdir) {
263 consumerd32_libdir = libdir;
264 }
265 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
266 if (libdir) {
267 consumerd64_libdir = libdir;
268 }
269 }
270
271 /*
272 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
273 */
274 static int create_thread_poll_set(struct lttng_poll_event *events,
275 unsigned int size)
276 {
277 int ret;
278
279 if (events == NULL || size == 0) {
280 ret = -1;
281 goto error;
282 }
283
284 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
285 if (ret < 0) {
286 goto error;
287 }
288
289 /* Add quit pipe */
290 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
291 if (ret < 0) {
292 goto error;
293 }
294
295 return 0;
296
297 error:
298 return ret;
299 }
300
301 /*
302 * Check if the thread quit pipe was triggered.
303 *
304 * Return 1 if it was triggered else 0;
305 */
306 static int check_thread_quit_pipe(int fd, uint32_t events)
307 {
308 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
309 return 1;
310 }
311
312 return 0;
313 }
314
315 /*
316 * Return group ID of the tracing group or -1 if not found.
317 */
318 static gid_t allowed_group(void)
319 {
320 struct group *grp;
321
322 if (opt_tracing_group) {
323 grp = getgrnam(opt_tracing_group);
324 } else {
325 grp = getgrnam(default_tracing_group);
326 }
327 if (!grp) {
328 return -1;
329 } else {
330 return grp->gr_gid;
331 }
332 }
333
334 /*
335 * Init thread quit pipe.
336 *
337 * Return -1 on error or 0 if all pipes are created.
338 */
339 static int init_thread_quit_pipe(void)
340 {
341 int ret, i;
342
343 ret = pipe(thread_quit_pipe);
344 if (ret < 0) {
345 PERROR("thread quit pipe");
346 goto error;
347 }
348
349 for (i = 0; i < 2; i++) {
350 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
351 if (ret < 0) {
352 PERROR("fcntl");
353 goto error;
354 }
355 }
356
357 error:
358 return ret;
359 }
360
361 /*
362 * Stop all threads by closing the thread quit pipe.
363 */
364 static void stop_threads(void)
365 {
366 int ret;
367
368 /* Stopping all threads */
369 DBG("Terminating all threads");
370 ret = notify_thread_pipe(thread_quit_pipe[1]);
371 if (ret < 0) {
372 ERR("write error on thread quit pipe");
373 }
374
375 /* Dispatch thread */
376 CMM_STORE_SHARED(dispatch_thread_exit, 1);
377 futex_nto1_wake(&ust_cmd_queue.futex);
378 }
379
380 /*
381 * Cleanup the daemon
382 */
383 static void cleanup(void)
384 {
385 int ret;
386 char *cmd;
387 struct ltt_session *sess, *stmp;
388
389 DBG("Cleaning up");
390
391 /* First thing first, stop all threads */
392 utils_close_pipe(thread_quit_pipe);
393
394 DBG("Removing %s directory", rundir);
395 ret = asprintf(&cmd, "rm -rf %s", rundir);
396 if (ret < 0) {
397 ERR("asprintf failed. Something is really wrong!");
398 }
399
400 /* Remove lttng run directory */
401 ret = system(cmd);
402 if (ret < 0) {
403 ERR("Unable to clean %s", rundir);
404 }
405 free(cmd);
406
407 DBG("Cleaning up all sessions");
408
409 /* Destroy session list mutex */
410 if (session_list_ptr != NULL) {
411 pthread_mutex_destroy(&session_list_ptr->lock);
412
413 /* Cleanup ALL session */
414 cds_list_for_each_entry_safe(sess, stmp,
415 &session_list_ptr->head, list) {
416 cmd_destroy_session(sess, kernel_poll_pipe[1]);
417 }
418 }
419
420 DBG("Closing all UST sockets");
421 ust_app_clean_list();
422
423 if (is_root && !opt_no_kernel) {
424 DBG2("Closing kernel fd");
425 if (kernel_tracer_fd >= 0) {
426 ret = close(kernel_tracer_fd);
427 if (ret) {
428 PERROR("close");
429 }
430 }
431 DBG("Unloading kernel modules");
432 modprobe_remove_lttng_all();
433 }
434
435 utils_close_pipe(kernel_poll_pipe);
436 utils_close_pipe(apps_cmd_pipe);
437
438 /* <fun> */
439 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
440 "Matthew, BEET driven development works!%c[%dm",
441 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
442 /* </fun> */
443 }
444
445 /*
446 * Send data on a unix socket using the liblttsessiondcomm API.
447 *
448 * Return lttcomm error code.
449 */
450 static int send_unix_sock(int sock, void *buf, size_t len)
451 {
452 /* Check valid length */
453 if (len <= 0) {
454 return -1;
455 }
456
457 return lttcomm_send_unix_sock(sock, buf, len);
458 }
459
460 /*
461 * Free memory of a command context structure.
462 */
463 static void clean_command_ctx(struct command_ctx **cmd_ctx)
464 {
465 DBG("Clean command context structure");
466 if (*cmd_ctx) {
467 if ((*cmd_ctx)->llm) {
468 free((*cmd_ctx)->llm);
469 }
470 if ((*cmd_ctx)->lsm) {
471 free((*cmd_ctx)->lsm);
472 }
473 free(*cmd_ctx);
474 *cmd_ctx = NULL;
475 }
476 }
477
478 /*
479 * Notify UST applications using the shm mmap futex.
480 */
481 static int notify_ust_apps(int active)
482 {
483 char *wait_shm_mmap;
484
485 DBG("Notifying applications of session daemon state: %d", active);
486
487 /* See shm.c for this call implying mmap, shm and futex calls */
488 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
489 if (wait_shm_mmap == NULL) {
490 goto error;
491 }
492
493 /* Wake waiting process */
494 futex_wait_update((int32_t *) wait_shm_mmap, active);
495
496 /* Apps notified successfully */
497 return 0;
498
499 error:
500 return -1;
501 }
502
503 /*
504 * Setup the outgoing data buffer for the response (llm) by allocating the
505 * right amount of memory and copying the original information from the lsm
506 * structure.
507 *
508 * Return total size of the buffer pointed by buf.
509 */
510 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
511 {
512 int ret, buf_size;
513
514 buf_size = size;
515
516 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
517 if (cmd_ctx->llm == NULL) {
518 PERROR("zmalloc");
519 ret = -ENOMEM;
520 goto error;
521 }
522
523 /* Copy common data */
524 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
525 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
526
527 cmd_ctx->llm->data_size = size;
528 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
529
530 return buf_size;
531
532 error:
533 return ret;
534 }
535
536 /*
537 * Update the kernel poll set of all channel fd available over all tracing
538 * session. Add the wakeup pipe at the end of the set.
539 */
540 static int update_kernel_poll(struct lttng_poll_event *events)
541 {
542 int ret;
543 struct ltt_session *session;
544 struct ltt_kernel_channel *channel;
545
546 DBG("Updating kernel poll set");
547
548 session_lock_list();
549 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
550 session_lock(session);
551 if (session->kernel_session == NULL) {
552 session_unlock(session);
553 continue;
554 }
555
556 cds_list_for_each_entry(channel,
557 &session->kernel_session->channel_list.head, list) {
558 /* Add channel fd to the kernel poll set */
559 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
560 if (ret < 0) {
561 session_unlock(session);
562 goto error;
563 }
564 DBG("Channel fd %d added to kernel set", channel->fd);
565 }
566 session_unlock(session);
567 }
568 session_unlock_list();
569
570 return 0;
571
572 error:
573 session_unlock_list();
574 return -1;
575 }
576
577 /*
578 * Find the channel fd from 'fd' over all tracing session. When found, check
579 * for new channel stream and send those stream fds to the kernel consumer.
580 *
581 * Useful for CPU hotplug feature.
582 */
583 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
584 {
585 int ret = 0;
586 struct ltt_session *session;
587 struct ltt_kernel_session *ksess;
588 struct ltt_kernel_channel *channel;
589
590 DBG("Updating kernel streams for channel fd %d", fd);
591
592 session_lock_list();
593 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
594 session_lock(session);
595 if (session->kernel_session == NULL) {
596 session_unlock(session);
597 continue;
598 }
599 ksess = session->kernel_session;
600
601 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
602 if (channel->fd == fd) {
603 DBG("Channel found, updating kernel streams");
604 ret = kernel_open_channel_stream(channel);
605 if (ret < 0) {
606 goto error;
607 }
608
609 /*
610 * Have we already sent fds to the consumer? If yes, it means
611 * that tracing is started so it is safe to send our updated
612 * stream fds.
613 */
614 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
615 struct lttng_ht_iter iter;
616 struct consumer_socket *socket;
617
618
619 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
620 &iter.iter, socket, node.node) {
621 /* Code flow error */
622 assert(socket->fd >= 0);
623
624 pthread_mutex_lock(socket->lock);
625 ret = kernel_consumer_send_channel_stream(socket->fd,
626 channel, ksess);
627 pthread_mutex_unlock(socket->lock);
628 if (ret < 0) {
629 goto error;
630 }
631 }
632 }
633 goto error;
634 }
635 }
636 session_unlock(session);
637 }
638 session_unlock_list();
639 return ret;
640
641 error:
642 session_unlock(session);
643 session_unlock_list();
644 return ret;
645 }
646
647 /*
648 * For each tracing session, update newly registered apps.
649 */
650 static void update_ust_app(int app_sock)
651 {
652 struct ltt_session *sess, *stmp;
653
654 session_lock_list();
655
656 /* For all tracing session(s) */
657 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
658 session_lock(sess);
659 if (sess->ust_session) {
660 ust_app_global_update(sess->ust_session, app_sock);
661 }
662 session_unlock(sess);
663 }
664
665 session_unlock_list();
666 }
667
668 /*
669 * This thread manage event coming from the kernel.
670 *
671 * Features supported in this thread:
672 * -) CPU Hotplug
673 */
674 static void *thread_manage_kernel(void *data)
675 {
676 int ret, i, pollfd, update_poll_flag = 1, err = -1;
677 uint32_t revents, nb_fd;
678 char tmp;
679 struct lttng_poll_event events;
680
681 DBG("Thread manage kernel started");
682
683 health_code_update(&health_thread_kernel);
684
685 ret = create_thread_poll_set(&events, 2);
686 if (ret < 0) {
687 goto error_poll_create;
688 }
689
690 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
691 if (ret < 0) {
692 goto error;
693 }
694
695 while (1) {
696 health_code_update(&health_thread_kernel);
697
698 if (update_poll_flag == 1) {
699 /*
700 * Reset number of fd in the poll set. Always 2 since there is the thread
701 * quit pipe and the kernel pipe.
702 */
703 events.nb_fd = 2;
704
705 ret = update_kernel_poll(&events);
706 if (ret < 0) {
707 goto error;
708 }
709 update_poll_flag = 0;
710 }
711
712 nb_fd = LTTNG_POLL_GETNB(&events);
713
714 DBG("Thread kernel polling on %d fds", nb_fd);
715
716 /* Zeroed the poll events */
717 lttng_poll_reset(&events);
718
719 /* Poll infinite value of time */
720 restart:
721 health_poll_update(&health_thread_kernel);
722 ret = lttng_poll_wait(&events, -1);
723 health_poll_update(&health_thread_kernel);
724 if (ret < 0) {
725 /*
726 * Restart interrupted system call.
727 */
728 if (errno == EINTR) {
729 goto restart;
730 }
731 goto error;
732 } else if (ret == 0) {
733 /* Should not happen since timeout is infinite */
734 ERR("Return value of poll is 0 with an infinite timeout.\n"
735 "This should not have happened! Continuing...");
736 continue;
737 }
738
739 for (i = 0; i < nb_fd; i++) {
740 /* Fetch once the poll data */
741 revents = LTTNG_POLL_GETEV(&events, i);
742 pollfd = LTTNG_POLL_GETFD(&events, i);
743
744 health_code_update(&health_thread_kernel);
745
746 /* Thread quit pipe has been closed. Killing thread. */
747 ret = check_thread_quit_pipe(pollfd, revents);
748 if (ret) {
749 err = 0;
750 goto exit;
751 }
752
753 /* Check for data on kernel pipe */
754 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
755 ret = read(kernel_poll_pipe[0], &tmp, 1);
756 update_poll_flag = 1;
757 continue;
758 } else {
759 /*
760 * New CPU detected by the kernel. Adding kernel stream to
761 * kernel session and updating the kernel consumer
762 */
763 if (revents & LPOLLIN) {
764 ret = update_kernel_stream(&kconsumer_data, pollfd);
765 if (ret < 0) {
766 continue;
767 }
768 break;
769 /*
770 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
771 * and unregister kernel stream at this point.
772 */
773 }
774 }
775 }
776 }
777
778 exit:
779 error:
780 lttng_poll_clean(&events);
781 error_poll_create:
782 if (err) {
783 health_error(&health_thread_kernel);
784 ERR("Health error occurred in %s", __func__);
785 }
786 health_exit(&health_thread_kernel);
787 DBG("Kernel thread dying");
788 return NULL;
789 }
790
791 /*
792 * This thread manage the consumer error sent back to the session daemon.
793 */
794 static void *thread_manage_consumer(void *data)
795 {
796 int sock = -1, i, ret, pollfd, err = -1;
797 uint32_t revents, nb_fd;
798 enum lttcomm_return_code code;
799 struct lttng_poll_event events;
800 struct consumer_data *consumer_data = data;
801
802 DBG("[thread] Manage consumer started");
803
804 health_code_update(&consumer_data->health);
805
806 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
807 if (ret < 0) {
808 goto error_listen;
809 }
810
811 /*
812 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
813 * Nothing more will be added to this poll set.
814 */
815 ret = create_thread_poll_set(&events, 2);
816 if (ret < 0) {
817 goto error_poll;
818 }
819
820 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
821 if (ret < 0) {
822 goto error;
823 }
824
825 nb_fd = LTTNG_POLL_GETNB(&events);
826
827 health_code_update(&consumer_data->health);
828
829 /* Inifinite blocking call, waiting for transmission */
830 restart:
831 health_poll_update(&consumer_data->health);
832 ret = lttng_poll_wait(&events, -1);
833 health_poll_update(&consumer_data->health);
834 if (ret < 0) {
835 /*
836 * Restart interrupted system call.
837 */
838 if (errno == EINTR) {
839 goto restart;
840 }
841 goto error;
842 }
843
844 for (i = 0; i < nb_fd; i++) {
845 /* Fetch once the poll data */
846 revents = LTTNG_POLL_GETEV(&events, i);
847 pollfd = LTTNG_POLL_GETFD(&events, i);
848
849 health_code_update(&consumer_data->health);
850
851 /* Thread quit pipe has been closed. Killing thread. */
852 ret = check_thread_quit_pipe(pollfd, revents);
853 if (ret) {
854 err = 0;
855 goto exit;
856 }
857
858 /* Event on the registration socket */
859 if (pollfd == consumer_data->err_sock) {
860 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
861 ERR("consumer err socket poll error");
862 goto error;
863 }
864 }
865 }
866
867 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
868 if (sock < 0) {
869 goto error;
870 }
871
872 health_code_update(&consumer_data->health);
873
874 DBG2("Receiving code from consumer err_sock");
875
876 /* Getting status code from kconsumerd */
877 ret = lttcomm_recv_unix_sock(sock, &code,
878 sizeof(enum lttcomm_return_code));
879 if (ret <= 0) {
880 goto error;
881 }
882
883 health_code_update(&consumer_data->health);
884
885 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
886 consumer_data->cmd_sock =
887 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
888 if (consumer_data->cmd_sock < 0) {
889 sem_post(&consumer_data->sem);
890 PERROR("consumer connect");
891 goto error;
892 }
893 /* Signal condition to tell that the kconsumerd is ready */
894 sem_post(&consumer_data->sem);
895 DBG("consumer command socket ready");
896 } else {
897 ERR("consumer error when waiting for SOCK_READY : %s",
898 lttcomm_get_readable_code(-code));
899 goto error;
900 }
901
902 /* Remove the kconsumerd error sock since we've established a connexion */
903 ret = lttng_poll_del(&events, consumer_data->err_sock);
904 if (ret < 0) {
905 goto error;
906 }
907
908 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
909 if (ret < 0) {
910 goto error;
911 }
912
913 health_code_update(&consumer_data->health);
914
915 /* Update number of fd */
916 nb_fd = LTTNG_POLL_GETNB(&events);
917
918 /* Inifinite blocking call, waiting for transmission */
919 restart_poll:
920 health_poll_update(&consumer_data->health);
921 ret = lttng_poll_wait(&events, -1);
922 health_poll_update(&consumer_data->health);
923 if (ret < 0) {
924 /*
925 * Restart interrupted system call.
926 */
927 if (errno == EINTR) {
928 goto restart_poll;
929 }
930 goto error;
931 }
932
933 for (i = 0; i < nb_fd; i++) {
934 /* Fetch once the poll data */
935 revents = LTTNG_POLL_GETEV(&events, i);
936 pollfd = LTTNG_POLL_GETFD(&events, i);
937
938 health_code_update(&consumer_data->health);
939
940 /* Thread quit pipe has been closed. Killing thread. */
941 ret = check_thread_quit_pipe(pollfd, revents);
942 if (ret) {
943 err = 0;
944 goto exit;
945 }
946
947 /* Event on the kconsumerd socket */
948 if (pollfd == sock) {
949 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
950 ERR("consumer err socket second poll error");
951 goto error;
952 }
953 }
954 }
955
956 health_code_update(&consumer_data->health);
957
958 /* Wait for any kconsumerd error */
959 ret = lttcomm_recv_unix_sock(sock, &code,
960 sizeof(enum lttcomm_return_code));
961 if (ret <= 0) {
962 ERR("consumer closed the command socket");
963 goto error;
964 }
965
966 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
967
968 exit:
969 error:
970 /* Immediately set the consumerd state to stopped */
971 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
972 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
973 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
974 consumer_data->type == LTTNG_CONSUMER32_UST) {
975 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
976 } else {
977 /* Code flow error... */
978 assert(0);
979 }
980
981 if (consumer_data->err_sock >= 0) {
982 ret = close(consumer_data->err_sock);
983 if (ret) {
984 PERROR("close");
985 }
986 }
987 if (consumer_data->cmd_sock >= 0) {
988 ret = close(consumer_data->cmd_sock);
989 if (ret) {
990 PERROR("close");
991 }
992 }
993 if (sock >= 0) {
994 ret = close(sock);
995 if (ret) {
996 PERROR("close");
997 }
998 }
999
1000 unlink(consumer_data->err_unix_sock_path);
1001 unlink(consumer_data->cmd_unix_sock_path);
1002 consumer_data->pid = 0;
1003
1004 lttng_poll_clean(&events);
1005 error_poll:
1006 error_listen:
1007 if (err) {
1008 health_error(&consumer_data->health);
1009 ERR("Health error occurred in %s", __func__);
1010 }
1011 health_exit(&consumer_data->health);
1012 DBG("consumer thread cleanup completed");
1013
1014 return NULL;
1015 }
1016
1017 /*
1018 * This thread manage application communication.
1019 */
1020 static void *thread_manage_apps(void *data)
1021 {
1022 int i, ret, pollfd, err = -1;
1023 uint32_t revents, nb_fd;
1024 struct ust_command ust_cmd;
1025 struct lttng_poll_event events;
1026
1027 DBG("[thread] Manage application started");
1028
1029 rcu_register_thread();
1030 rcu_thread_online();
1031
1032 health_code_update(&health_thread_app_manage);
1033
1034 ret = create_thread_poll_set(&events, 2);
1035 if (ret < 0) {
1036 goto error_poll_create;
1037 }
1038
1039 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1040 if (ret < 0) {
1041 goto error;
1042 }
1043
1044 health_code_update(&health_thread_app_manage);
1045
1046 while (1) {
1047 /* Zeroed the events structure */
1048 lttng_poll_reset(&events);
1049
1050 nb_fd = LTTNG_POLL_GETNB(&events);
1051
1052 DBG("Apps thread polling on %d fds", nb_fd);
1053
1054 /* Inifinite blocking call, waiting for transmission */
1055 restart:
1056 health_poll_update(&health_thread_app_manage);
1057 ret = lttng_poll_wait(&events, -1);
1058 health_poll_update(&health_thread_app_manage);
1059 if (ret < 0) {
1060 /*
1061 * Restart interrupted system call.
1062 */
1063 if (errno == EINTR) {
1064 goto restart;
1065 }
1066 goto error;
1067 }
1068
1069 for (i = 0; i < nb_fd; i++) {
1070 /* Fetch once the poll data */
1071 revents = LTTNG_POLL_GETEV(&events, i);
1072 pollfd = LTTNG_POLL_GETFD(&events, i);
1073
1074 health_code_update(&health_thread_app_manage);
1075
1076 /* Thread quit pipe has been closed. Killing thread. */
1077 ret = check_thread_quit_pipe(pollfd, revents);
1078 if (ret) {
1079 err = 0;
1080 goto exit;
1081 }
1082
1083 /* Inspect the apps cmd pipe */
1084 if (pollfd == apps_cmd_pipe[0]) {
1085 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1086 ERR("Apps command pipe error");
1087 goto error;
1088 } else if (revents & LPOLLIN) {
1089 /* Empty pipe */
1090 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1091 if (ret < 0 || ret < sizeof(ust_cmd)) {
1092 PERROR("read apps cmd pipe");
1093 goto error;
1094 }
1095
1096 health_code_update(&health_thread_app_manage);
1097
1098 /* Register applicaton to the session daemon */
1099 ret = ust_app_register(&ust_cmd.reg_msg,
1100 ust_cmd.sock);
1101 if (ret == -ENOMEM) {
1102 goto error;
1103 } else if (ret < 0) {
1104 break;
1105 }
1106
1107 health_code_update(&health_thread_app_manage);
1108
1109 /*
1110 * Validate UST version compatibility.
1111 */
1112 ret = ust_app_validate_version(ust_cmd.sock);
1113 if (ret >= 0) {
1114 /*
1115 * Add channel(s) and event(s) to newly registered apps
1116 * from lttng global UST domain.
1117 */
1118 update_ust_app(ust_cmd.sock);
1119 }
1120
1121 health_code_update(&health_thread_app_manage);
1122
1123 ret = ust_app_register_done(ust_cmd.sock);
1124 if (ret < 0) {
1125 /*
1126 * If the registration is not possible, we simply
1127 * unregister the apps and continue
1128 */
1129 ust_app_unregister(ust_cmd.sock);
1130 } else {
1131 /*
1132 * We just need here to monitor the close of the UST
1133 * socket and poll set monitor those by default.
1134 * Listen on POLLIN (even if we never expect any
1135 * data) to ensure that hangup wakes us.
1136 */
1137 ret = lttng_poll_add(&events, ust_cmd.sock, LPOLLIN);
1138 if (ret < 0) {
1139 goto error;
1140 }
1141
1142 DBG("Apps with sock %d added to poll set",
1143 ust_cmd.sock);
1144 }
1145
1146 health_code_update(&health_thread_app_manage);
1147
1148 break;
1149 }
1150 } else {
1151 /*
1152 * At this point, we know that a registered application made
1153 * the event at poll_wait.
1154 */
1155 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1156 /* Removing from the poll set */
1157 ret = lttng_poll_del(&events, pollfd);
1158 if (ret < 0) {
1159 goto error;
1160 }
1161
1162 /* Socket closed on remote end. */
1163 ust_app_unregister(pollfd);
1164 break;
1165 }
1166 }
1167
1168 health_code_update(&health_thread_app_manage);
1169 }
1170 }
1171
1172 exit:
1173 error:
1174 lttng_poll_clean(&events);
1175 error_poll_create:
1176 if (err) {
1177 health_error(&health_thread_app_manage);
1178 ERR("Health error occurred in %s", __func__);
1179 }
1180 health_exit(&health_thread_app_manage);
1181 DBG("Application communication apps thread cleanup complete");
1182 rcu_thread_offline();
1183 rcu_unregister_thread();
1184 return NULL;
1185 }
1186
1187 /*
1188 * Dispatch request from the registration threads to the application
1189 * communication thread.
1190 */
1191 static void *thread_dispatch_ust_registration(void *data)
1192 {
1193 int ret;
1194 struct cds_wfq_node *node;
1195 struct ust_command *ust_cmd = NULL;
1196
1197 DBG("[thread] Dispatch UST command started");
1198
1199 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1200 /* Atomically prepare the queue futex */
1201 futex_nto1_prepare(&ust_cmd_queue.futex);
1202
1203 do {
1204 /* Dequeue command for registration */
1205 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1206 if (node == NULL) {
1207 DBG("Woken up but nothing in the UST command queue");
1208 /* Continue thread execution */
1209 break;
1210 }
1211
1212 ust_cmd = caa_container_of(node, struct ust_command, node);
1213
1214 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1215 " gid:%d sock:%d name:%s (version %d.%d)",
1216 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1217 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1218 ust_cmd->sock, ust_cmd->reg_msg.name,
1219 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1220 /*
1221 * Inform apps thread of the new application registration. This
1222 * call is blocking so we can be assured that the data will be read
1223 * at some point in time or wait to the end of the world :)
1224 */
1225 ret = write(apps_cmd_pipe[1], ust_cmd,
1226 sizeof(struct ust_command));
1227 if (ret < 0) {
1228 PERROR("write apps cmd pipe");
1229 if (errno == EBADF) {
1230 /*
1231 * We can't inform the application thread to process
1232 * registration. We will exit or else application
1233 * registration will not occur and tracing will never
1234 * start.
1235 */
1236 goto error;
1237 }
1238 }
1239 free(ust_cmd);
1240 } while (node != NULL);
1241
1242 /* Futex wait on queue. Blocking call on futex() */
1243 futex_nto1_wait(&ust_cmd_queue.futex);
1244 }
1245
1246 error:
1247 DBG("Dispatch thread dying");
1248 return NULL;
1249 }
1250
1251 /*
1252 * This thread manage application registration.
1253 */
1254 static void *thread_registration_apps(void *data)
1255 {
1256 int sock = -1, i, ret, pollfd, err = -1;
1257 uint32_t revents, nb_fd;
1258 struct lttng_poll_event events;
1259 /*
1260 * Get allocated in this thread, enqueued to a global queue, dequeued and
1261 * freed in the manage apps thread.
1262 */
1263 struct ust_command *ust_cmd = NULL;
1264
1265 DBG("[thread] Manage application registration started");
1266
1267 ret = lttcomm_listen_unix_sock(apps_sock);
1268 if (ret < 0) {
1269 goto error_listen;
1270 }
1271
1272 /*
1273 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1274 * more will be added to this poll set.
1275 */
1276 ret = create_thread_poll_set(&events, 2);
1277 if (ret < 0) {
1278 goto error_create_poll;
1279 }
1280
1281 /* Add the application registration socket */
1282 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1283 if (ret < 0) {
1284 goto error_poll_add;
1285 }
1286
1287 /* Notify all applications to register */
1288 ret = notify_ust_apps(1);
1289 if (ret < 0) {
1290 ERR("Failed to notify applications or create the wait shared memory.\n"
1291 "Execution continues but there might be problem for already\n"
1292 "running applications that wishes to register.");
1293 }
1294
1295 while (1) {
1296 DBG("Accepting application registration");
1297
1298 nb_fd = LTTNG_POLL_GETNB(&events);
1299
1300 /* Inifinite blocking call, waiting for transmission */
1301 restart:
1302 health_poll_update(&health_thread_app_reg);
1303 ret = lttng_poll_wait(&events, -1);
1304 health_poll_update(&health_thread_app_reg);
1305 if (ret < 0) {
1306 /*
1307 * Restart interrupted system call.
1308 */
1309 if (errno == EINTR) {
1310 goto restart;
1311 }
1312 goto error;
1313 }
1314
1315 for (i = 0; i < nb_fd; i++) {
1316 health_code_update(&health_thread_app_reg);
1317
1318 /* Fetch once the poll data */
1319 revents = LTTNG_POLL_GETEV(&events, i);
1320 pollfd = LTTNG_POLL_GETFD(&events, i);
1321
1322 /* Thread quit pipe has been closed. Killing thread. */
1323 ret = check_thread_quit_pipe(pollfd, revents);
1324 if (ret) {
1325 err = 0;
1326 goto exit;
1327 }
1328
1329 /* Event on the registration socket */
1330 if (pollfd == apps_sock) {
1331 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1332 ERR("Register apps socket poll error");
1333 goto error;
1334 } else if (revents & LPOLLIN) {
1335 sock = lttcomm_accept_unix_sock(apps_sock);
1336 if (sock < 0) {
1337 goto error;
1338 }
1339
1340 /* Create UST registration command for enqueuing */
1341 ust_cmd = zmalloc(sizeof(struct ust_command));
1342 if (ust_cmd == NULL) {
1343 PERROR("ust command zmalloc");
1344 goto error;
1345 }
1346
1347 /*
1348 * Using message-based transmissions to ensure we don't
1349 * have to deal with partially received messages.
1350 */
1351 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1352 if (ret < 0) {
1353 ERR("Exhausted file descriptors allowed for applications.");
1354 free(ust_cmd);
1355 ret = close(sock);
1356 if (ret) {
1357 PERROR("close");
1358 }
1359 sock = -1;
1360 continue;
1361 }
1362 health_code_update(&health_thread_app_reg);
1363 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1364 sizeof(struct ust_register_msg));
1365 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1366 if (ret < 0) {
1367 PERROR("lttcomm_recv_unix_sock register apps");
1368 } else {
1369 ERR("Wrong size received on apps register");
1370 }
1371 free(ust_cmd);
1372 ret = close(sock);
1373 if (ret) {
1374 PERROR("close");
1375 }
1376 lttng_fd_put(LTTNG_FD_APPS, 1);
1377 sock = -1;
1378 continue;
1379 }
1380 health_code_update(&health_thread_app_reg);
1381
1382 ust_cmd->sock = sock;
1383 sock = -1;
1384
1385 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1386 " gid:%d sock:%d name:%s (version %d.%d)",
1387 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1388 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1389 ust_cmd->sock, ust_cmd->reg_msg.name,
1390 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1391
1392 /*
1393 * Lock free enqueue the registration request. The red pill
1394 * has been taken! This apps will be part of the *system*.
1395 */
1396 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1397
1398 /*
1399 * Wake the registration queue futex. Implicit memory
1400 * barrier with the exchange in cds_wfq_enqueue.
1401 */
1402 futex_nto1_wake(&ust_cmd_queue.futex);
1403 }
1404 }
1405 }
1406 }
1407
1408 exit:
1409 error:
1410 if (err) {
1411 health_error(&health_thread_app_reg);
1412 ERR("Health error occurred in %s", __func__);
1413 }
1414 health_exit(&health_thread_app_reg);
1415
1416 /* Notify that the registration thread is gone */
1417 notify_ust_apps(0);
1418
1419 if (apps_sock >= 0) {
1420 ret = close(apps_sock);
1421 if (ret) {
1422 PERROR("close");
1423 }
1424 }
1425 if (sock >= 0) {
1426 ret = close(sock);
1427 if (ret) {
1428 PERROR("close");
1429 }
1430 lttng_fd_put(LTTNG_FD_APPS, 1);
1431 }
1432 unlink(apps_unix_sock_path);
1433
1434 error_poll_add:
1435 lttng_poll_clean(&events);
1436 error_listen:
1437 error_create_poll:
1438 DBG("UST Registration thread cleanup complete");
1439
1440 return NULL;
1441 }
1442
1443 /*
1444 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1445 * exec or it will fails.
1446 */
1447 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1448 {
1449 int ret;
1450 struct timespec timeout;
1451
1452 timeout.tv_sec = DEFAULT_SEM_WAIT_TIMEOUT;
1453 timeout.tv_nsec = 0;
1454
1455 /* Setup semaphore */
1456 ret = sem_init(&consumer_data->sem, 0, 0);
1457 if (ret < 0) {
1458 PERROR("sem_init consumer semaphore");
1459 goto error;
1460 }
1461
1462 ret = pthread_create(&consumer_data->thread, NULL,
1463 thread_manage_consumer, consumer_data);
1464 if (ret != 0) {
1465 PERROR("pthread_create consumer");
1466 ret = -1;
1467 goto error;
1468 }
1469
1470 /* Get time for sem_timedwait absolute timeout */
1471 ret = clock_gettime(CLOCK_REALTIME, &timeout);
1472 if (ret < 0) {
1473 PERROR("clock_gettime spawn consumer");
1474 /* Infinite wait for the kconsumerd thread to be ready */
1475 ret = sem_wait(&consumer_data->sem);
1476 } else {
1477 /* Normal timeout if the gettime was successful */
1478 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1479 ret = sem_timedwait(&consumer_data->sem, &timeout);
1480 }
1481
1482 if (ret < 0) {
1483 if (errno == ETIMEDOUT) {
1484 /*
1485 * Call has timed out so we kill the kconsumerd_thread and return
1486 * an error.
1487 */
1488 ERR("The consumer thread was never ready. Killing it");
1489 ret = pthread_cancel(consumer_data->thread);
1490 if (ret < 0) {
1491 PERROR("pthread_cancel consumer thread");
1492 }
1493 } else {
1494 PERROR("semaphore wait failed consumer thread");
1495 }
1496 goto error;
1497 }
1498
1499 pthread_mutex_lock(&consumer_data->pid_mutex);
1500 if (consumer_data->pid == 0) {
1501 ERR("Kconsumerd did not start");
1502 pthread_mutex_unlock(&consumer_data->pid_mutex);
1503 goto error;
1504 }
1505 pthread_mutex_unlock(&consumer_data->pid_mutex);
1506
1507 return 0;
1508
1509 error:
1510 return ret;
1511 }
1512
1513 /*
1514 * Join consumer thread
1515 */
1516 static int join_consumer_thread(struct consumer_data *consumer_data)
1517 {
1518 void *status;
1519 int ret;
1520
1521 /* Consumer pid must be a real one. */
1522 if (consumer_data->pid > 0) {
1523 ret = kill(consumer_data->pid, SIGTERM);
1524 if (ret) {
1525 ERR("Error killing consumer daemon");
1526 return ret;
1527 }
1528 return pthread_join(consumer_data->thread, &status);
1529 } else {
1530 return 0;
1531 }
1532 }
1533
1534 /*
1535 * Fork and exec a consumer daemon (consumerd).
1536 *
1537 * Return pid if successful else -1.
1538 */
1539 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1540 {
1541 int ret;
1542 pid_t pid;
1543 const char *consumer_to_use;
1544 const char *verbosity;
1545 struct stat st;
1546
1547 DBG("Spawning consumerd");
1548
1549 pid = fork();
1550 if (pid == 0) {
1551 /*
1552 * Exec consumerd.
1553 */
1554 if (opt_verbose_consumer) {
1555 verbosity = "--verbose";
1556 } else {
1557 verbosity = "--quiet";
1558 }
1559 switch (consumer_data->type) {
1560 case LTTNG_CONSUMER_KERNEL:
1561 /*
1562 * Find out which consumerd to execute. We will first try the
1563 * 64-bit path, then the sessiond's installation directory, and
1564 * fallback on the 32-bit one,
1565 */
1566 DBG3("Looking for a kernel consumer at these locations:");
1567 DBG3(" 1) %s", consumerd64_bin);
1568 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1569 DBG3(" 3) %s", consumerd32_bin);
1570 if (stat(consumerd64_bin, &st) == 0) {
1571 DBG3("Found location #1");
1572 consumer_to_use = consumerd64_bin;
1573 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1574 DBG3("Found location #2");
1575 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1576 } else if (stat(consumerd32_bin, &st) == 0) {
1577 DBG3("Found location #3");
1578 consumer_to_use = consumerd32_bin;
1579 } else {
1580 DBG("Could not find any valid consumerd executable");
1581 break;
1582 }
1583 DBG("Using kernel consumer at: %s", consumer_to_use);
1584 execl(consumer_to_use,
1585 "lttng-consumerd", verbosity, "-k",
1586 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1587 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1588 NULL);
1589 break;
1590 case LTTNG_CONSUMER64_UST:
1591 {
1592 char *tmpnew = NULL;
1593
1594 if (consumerd64_libdir[0] != '\0') {
1595 char *tmp;
1596 size_t tmplen;
1597
1598 tmp = getenv("LD_LIBRARY_PATH");
1599 if (!tmp) {
1600 tmp = "";
1601 }
1602 tmplen = strlen("LD_LIBRARY_PATH=")
1603 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1604 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1605 if (!tmpnew) {
1606 ret = -ENOMEM;
1607 goto error;
1608 }
1609 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1610 strcat(tmpnew, consumerd64_libdir);
1611 if (tmp[0] != '\0') {
1612 strcat(tmpnew, ":");
1613 strcat(tmpnew, tmp);
1614 }
1615 ret = putenv(tmpnew);
1616 if (ret) {
1617 ret = -errno;
1618 goto error;
1619 }
1620 }
1621 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1622 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1623 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1624 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1625 NULL);
1626 if (consumerd64_libdir[0] != '\0') {
1627 free(tmpnew);
1628 }
1629 if (ret) {
1630 goto error;
1631 }
1632 break;
1633 }
1634 case LTTNG_CONSUMER32_UST:
1635 {
1636 char *tmpnew = NULL;
1637
1638 if (consumerd32_libdir[0] != '\0') {
1639 char *tmp;
1640 size_t tmplen;
1641
1642 tmp = getenv("LD_LIBRARY_PATH");
1643 if (!tmp) {
1644 tmp = "";
1645 }
1646 tmplen = strlen("LD_LIBRARY_PATH=")
1647 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1648 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1649 if (!tmpnew) {
1650 ret = -ENOMEM;
1651 goto error;
1652 }
1653 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1654 strcat(tmpnew, consumerd32_libdir);
1655 if (tmp[0] != '\0') {
1656 strcat(tmpnew, ":");
1657 strcat(tmpnew, tmp);
1658 }
1659 ret = putenv(tmpnew);
1660 if (ret) {
1661 ret = -errno;
1662 goto error;
1663 }
1664 }
1665 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1666 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1667 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1668 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1669 NULL);
1670 if (consumerd32_libdir[0] != '\0') {
1671 free(tmpnew);
1672 }
1673 if (ret) {
1674 goto error;
1675 }
1676 break;
1677 }
1678 default:
1679 PERROR("unknown consumer type");
1680 exit(EXIT_FAILURE);
1681 }
1682 if (errno != 0) {
1683 PERROR("kernel start consumer exec");
1684 }
1685 exit(EXIT_FAILURE);
1686 } else if (pid > 0) {
1687 ret = pid;
1688 } else {
1689 PERROR("start consumer fork");
1690 ret = -errno;
1691 }
1692 error:
1693 return ret;
1694 }
1695
1696 /*
1697 * Spawn the consumerd daemon and session daemon thread.
1698 */
1699 static int start_consumerd(struct consumer_data *consumer_data)
1700 {
1701 int ret;
1702
1703 pthread_mutex_lock(&consumer_data->pid_mutex);
1704 if (consumer_data->pid != 0) {
1705 pthread_mutex_unlock(&consumer_data->pid_mutex);
1706 goto end;
1707 }
1708
1709 ret = spawn_consumerd(consumer_data);
1710 if (ret < 0) {
1711 ERR("Spawning consumerd failed");
1712 pthread_mutex_unlock(&consumer_data->pid_mutex);
1713 goto error;
1714 }
1715
1716 /* Setting up the consumer_data pid */
1717 consumer_data->pid = ret;
1718 DBG2("Consumer pid %d", consumer_data->pid);
1719 pthread_mutex_unlock(&consumer_data->pid_mutex);
1720
1721 DBG2("Spawning consumer control thread");
1722 ret = spawn_consumer_thread(consumer_data);
1723 if (ret < 0) {
1724 ERR("Fatal error spawning consumer control thread");
1725 goto error;
1726 }
1727
1728 end:
1729 return 0;
1730
1731 error:
1732 return ret;
1733 }
1734
1735 /*
1736 * Compute health status of each consumer. If one of them is zero (bad
1737 * state), we return 0.
1738 */
1739 static int check_consumer_health(void)
1740 {
1741 int ret;
1742
1743 ret = health_check_state(&kconsumer_data.health) &&
1744 health_check_state(&ustconsumer32_data.health) &&
1745 health_check_state(&ustconsumer64_data.health);
1746
1747 DBG3("Health consumer check %d", ret);
1748
1749 return ret;
1750 }
1751
1752 /*
1753 * Setup necessary data for kernel tracer action.
1754 */
1755 static int init_kernel_tracer(void)
1756 {
1757 int ret;
1758
1759 /* Modprobe lttng kernel modules */
1760 ret = modprobe_lttng_control();
1761 if (ret < 0) {
1762 goto error;
1763 }
1764
1765 /* Open debugfs lttng */
1766 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1767 if (kernel_tracer_fd < 0) {
1768 DBG("Failed to open %s", module_proc_lttng);
1769 ret = -1;
1770 goto error_open;
1771 }
1772
1773 /* Validate kernel version */
1774 ret = kernel_validate_version(kernel_tracer_fd);
1775 if (ret < 0) {
1776 goto error_version;
1777 }
1778
1779 ret = modprobe_lttng_data();
1780 if (ret < 0) {
1781 goto error_modules;
1782 }
1783
1784 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1785 return 0;
1786
1787 error_version:
1788 modprobe_remove_lttng_control();
1789 ret = close(kernel_tracer_fd);
1790 if (ret) {
1791 PERROR("close");
1792 }
1793 kernel_tracer_fd = -1;
1794 return LTTNG_ERR_KERN_VERSION;
1795
1796 error_modules:
1797 ret = close(kernel_tracer_fd);
1798 if (ret) {
1799 PERROR("close");
1800 }
1801
1802 error_open:
1803 modprobe_remove_lttng_control();
1804
1805 error:
1806 WARN("No kernel tracer available");
1807 kernel_tracer_fd = -1;
1808 if (!is_root) {
1809 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1810 } else {
1811 return LTTNG_ERR_KERN_NA;
1812 }
1813 }
1814
1815
1816 /*
1817 * Copy consumer output from the tracing session to the domain session. The
1818 * function also applies the right modification on a per domain basis for the
1819 * trace files destination directory.
1820 */
1821 static int copy_session_consumer(int domain, struct ltt_session *session)
1822 {
1823 int ret;
1824 const char *dir_name;
1825 struct consumer_output *consumer;
1826
1827 assert(session);
1828 assert(session->consumer);
1829
1830 switch (domain) {
1831 case LTTNG_DOMAIN_KERNEL:
1832 DBG3("Copying tracing session consumer output in kernel session");
1833 session->kernel_session->consumer =
1834 consumer_copy_output(session->consumer);
1835 /* Ease our life a bit for the next part */
1836 consumer = session->kernel_session->consumer;
1837 dir_name = DEFAULT_KERNEL_TRACE_DIR;
1838 break;
1839 case LTTNG_DOMAIN_UST:
1840 DBG3("Copying tracing session consumer output in UST session");
1841 session->ust_session->consumer =
1842 consumer_copy_output(session->consumer);
1843 /* Ease our life a bit for the next part */
1844 consumer = session->ust_session->consumer;
1845 dir_name = DEFAULT_UST_TRACE_DIR;
1846 break;
1847 default:
1848 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1849 goto error;
1850 }
1851
1852 /* Append correct directory to subdir */
1853 strncat(consumer->subdir, dir_name,
1854 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
1855 DBG3("Copy session consumer subdir %s", consumer->subdir);
1856
1857 ret = LTTNG_OK;
1858
1859 error:
1860 return ret;
1861 }
1862
1863 /*
1864 * Create an UST session and add it to the session ust list.
1865 */
1866 static int create_ust_session(struct ltt_session *session,
1867 struct lttng_domain *domain)
1868 {
1869 int ret;
1870 struct ltt_ust_session *lus = NULL;
1871
1872 assert(session);
1873 assert(domain);
1874 assert(session->consumer);
1875
1876 switch (domain->type) {
1877 case LTTNG_DOMAIN_UST:
1878 break;
1879 default:
1880 ERR("Unknown UST domain on create session %d", domain->type);
1881 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1882 goto error;
1883 }
1884
1885 DBG("Creating UST session");
1886
1887 lus = trace_ust_create_session(session->path, session->id, domain);
1888 if (lus == NULL) {
1889 ret = LTTNG_ERR_UST_SESS_FAIL;
1890 goto error;
1891 }
1892
1893 lus->uid = session->uid;
1894 lus->gid = session->gid;
1895 session->ust_session = lus;
1896
1897 /* Copy session output to the newly created UST session */
1898 ret = copy_session_consumer(domain->type, session);
1899 if (ret != LTTNG_OK) {
1900 goto error;
1901 }
1902
1903 return LTTNG_OK;
1904
1905 error:
1906 free(lus);
1907 session->ust_session = NULL;
1908 return ret;
1909 }
1910
1911 /*
1912 * Create a kernel tracer session then create the default channel.
1913 */
1914 static int create_kernel_session(struct ltt_session *session)
1915 {
1916 int ret;
1917
1918 DBG("Creating kernel session");
1919
1920 ret = kernel_create_session(session, kernel_tracer_fd);
1921 if (ret < 0) {
1922 ret = LTTNG_ERR_KERN_SESS_FAIL;
1923 goto error;
1924 }
1925
1926 /* Code flow safety */
1927 assert(session->kernel_session);
1928
1929 /* Copy session output to the newly created Kernel session */
1930 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
1931 if (ret != LTTNG_OK) {
1932 goto error;
1933 }
1934
1935 /* Create directory(ies) on local filesystem. */
1936 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
1937 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
1938 ret = run_as_mkdir_recursive(
1939 session->kernel_session->consumer->dst.trace_path,
1940 S_IRWXU | S_IRWXG, session->uid, session->gid);
1941 if (ret < 0) {
1942 if (ret != -EEXIST) {
1943 ERR("Trace directory creation error");
1944 goto error;
1945 }
1946 }
1947 }
1948
1949 session->kernel_session->uid = session->uid;
1950 session->kernel_session->gid = session->gid;
1951
1952 return LTTNG_OK;
1953
1954 error:
1955 trace_kernel_destroy_session(session->kernel_session);
1956 session->kernel_session = NULL;
1957 return ret;
1958 }
1959
1960 /*
1961 * Count number of session permitted by uid/gid.
1962 */
1963 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
1964 {
1965 unsigned int i = 0;
1966 struct ltt_session *session;
1967
1968 DBG("Counting number of available session for UID %d GID %d",
1969 uid, gid);
1970 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
1971 /*
1972 * Only list the sessions the user can control.
1973 */
1974 if (!session_access_ok(session, uid, gid)) {
1975 continue;
1976 }
1977 i++;
1978 }
1979 return i;
1980 }
1981
1982 /*
1983 * Process the command requested by the lttng client within the command
1984 * context structure. This function make sure that the return structure (llm)
1985 * is set and ready for transmission before returning.
1986 *
1987 * Return any error encountered or 0 for success.
1988 *
1989 * "sock" is only used for special-case var. len data.
1990 */
1991 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
1992 int *sock_error)
1993 {
1994 int ret = LTTNG_OK;
1995 int need_tracing_session = 1;
1996 int need_domain;
1997
1998 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
1999
2000 *sock_error = 0;
2001
2002 switch (cmd_ctx->lsm->cmd_type) {
2003 case LTTNG_CREATE_SESSION:
2004 case LTTNG_DESTROY_SESSION:
2005 case LTTNG_LIST_SESSIONS:
2006 case LTTNG_LIST_DOMAINS:
2007 case LTTNG_START_TRACE:
2008 case LTTNG_STOP_TRACE:
2009 need_domain = 0;
2010 break;
2011 default:
2012 need_domain = 1;
2013 }
2014
2015 if (opt_no_kernel && need_domain
2016 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2017 if (!is_root) {
2018 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2019 } else {
2020 ret = LTTNG_ERR_KERN_NA;
2021 }
2022 goto error;
2023 }
2024
2025 /* Deny register consumer if we already have a spawned consumer. */
2026 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2027 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2028 if (kconsumer_data.pid > 0) {
2029 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2030 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2031 goto error;
2032 }
2033 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2034 }
2035
2036 /*
2037 * Check for command that don't needs to allocate a returned payload. We do
2038 * this here so we don't have to make the call for no payload at each
2039 * command.
2040 */
2041 switch(cmd_ctx->lsm->cmd_type) {
2042 case LTTNG_LIST_SESSIONS:
2043 case LTTNG_LIST_TRACEPOINTS:
2044 case LTTNG_LIST_TRACEPOINT_FIELDS:
2045 case LTTNG_LIST_DOMAINS:
2046 case LTTNG_LIST_CHANNELS:
2047 case LTTNG_LIST_EVENTS:
2048 break;
2049 default:
2050 /* Setup lttng message with no payload */
2051 ret = setup_lttng_msg(cmd_ctx, 0);
2052 if (ret < 0) {
2053 /* This label does not try to unlock the session */
2054 goto init_setup_error;
2055 }
2056 }
2057
2058 /* Commands that DO NOT need a session. */
2059 switch (cmd_ctx->lsm->cmd_type) {
2060 case LTTNG_CREATE_SESSION:
2061 case LTTNG_CALIBRATE:
2062 case LTTNG_LIST_SESSIONS:
2063 case LTTNG_LIST_TRACEPOINTS:
2064 case LTTNG_LIST_TRACEPOINT_FIELDS:
2065 need_tracing_session = 0;
2066 break;
2067 default:
2068 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2069 /*
2070 * We keep the session list lock across _all_ commands
2071 * for now, because the per-session lock does not
2072 * handle teardown properly.
2073 */
2074 session_lock_list();
2075 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2076 if (cmd_ctx->session == NULL) {
2077 if (cmd_ctx->lsm->session.name != NULL) {
2078 ret = LTTNG_ERR_SESS_NOT_FOUND;
2079 } else {
2080 /* If no session name specified */
2081 ret = LTTNG_ERR_SELECT_SESS;
2082 }
2083 goto error;
2084 } else {
2085 /* Acquire lock for the session */
2086 session_lock(cmd_ctx->session);
2087 }
2088 break;
2089 }
2090
2091 if (!need_domain) {
2092 goto skip_domain;
2093 }
2094
2095 /*
2096 * Check domain type for specific "pre-action".
2097 */
2098 switch (cmd_ctx->lsm->domain.type) {
2099 case LTTNG_DOMAIN_KERNEL:
2100 if (!is_root) {
2101 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2102 goto error;
2103 }
2104
2105 /* Kernel tracer check */
2106 if (kernel_tracer_fd == -1) {
2107 /* Basically, load kernel tracer modules */
2108 ret = init_kernel_tracer();
2109 if (ret != 0) {
2110 goto error;
2111 }
2112 }
2113
2114 /* Consumer is in an ERROR state. Report back to client */
2115 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2116 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2117 goto error;
2118 }
2119
2120 /* Need a session for kernel command */
2121 if (need_tracing_session) {
2122 if (cmd_ctx->session->kernel_session == NULL) {
2123 ret = create_kernel_session(cmd_ctx->session);
2124 if (ret < 0) {
2125 ret = LTTNG_ERR_KERN_SESS_FAIL;
2126 goto error;
2127 }
2128 }
2129
2130 /* Start the kernel consumer daemon */
2131 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2132 if (kconsumer_data.pid == 0 &&
2133 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2134 cmd_ctx->session->start_consumer) {
2135 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2136 ret = start_consumerd(&kconsumer_data);
2137 if (ret < 0) {
2138 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2139 goto error;
2140 }
2141 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2142 } else {
2143 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2144 }
2145
2146 /*
2147 * The consumer was just spawned so we need to add the socket to
2148 * the consumer output of the session if exist.
2149 */
2150 ret = consumer_create_socket(&kconsumer_data,
2151 cmd_ctx->session->kernel_session->consumer);
2152 if (ret < 0) {
2153 goto error;
2154 }
2155 }
2156
2157 break;
2158 case LTTNG_DOMAIN_UST:
2159 {
2160 /* Consumer is in an ERROR state. Report back to client */
2161 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2162 ret = LTTNG_ERR_NO_USTCONSUMERD;
2163 goto error;
2164 }
2165
2166 if (need_tracing_session) {
2167 /* Create UST session if none exist. */
2168 if (cmd_ctx->session->ust_session == NULL) {
2169 ret = create_ust_session(cmd_ctx->session,
2170 &cmd_ctx->lsm->domain);
2171 if (ret != LTTNG_OK) {
2172 goto error;
2173 }
2174 }
2175
2176 /* Start the UST consumer daemons */
2177 /* 64-bit */
2178 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2179 if (consumerd64_bin[0] != '\0' &&
2180 ustconsumer64_data.pid == 0 &&
2181 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2182 cmd_ctx->session->start_consumer) {
2183 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2184 ret = start_consumerd(&ustconsumer64_data);
2185 if (ret < 0) {
2186 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2187 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2188 goto error;
2189 }
2190
2191 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2192 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2193 } else {
2194 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2195 }
2196
2197 /*
2198 * Setup socket for consumer 64 bit. No need for atomic access
2199 * since it was set above and can ONLY be set in this thread.
2200 */
2201 ret = consumer_create_socket(&ustconsumer64_data,
2202 cmd_ctx->session->ust_session->consumer);
2203 if (ret < 0) {
2204 goto error;
2205 }
2206
2207 /* 32-bit */
2208 if (consumerd32_bin[0] != '\0' &&
2209 ustconsumer32_data.pid == 0 &&
2210 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2211 cmd_ctx->session->start_consumer) {
2212 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2213 ret = start_consumerd(&ustconsumer32_data);
2214 if (ret < 0) {
2215 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2216 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2217 goto error;
2218 }
2219
2220 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2221 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2222 } else {
2223 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2224 }
2225
2226 /*
2227 * Setup socket for consumer 64 bit. No need for atomic access
2228 * since it was set above and can ONLY be set in this thread.
2229 */
2230 ret = consumer_create_socket(&ustconsumer32_data,
2231 cmd_ctx->session->ust_session->consumer);
2232 if (ret < 0) {
2233 goto error;
2234 }
2235 }
2236 break;
2237 }
2238 default:
2239 break;
2240 }
2241 skip_domain:
2242
2243 /* Validate consumer daemon state when start/stop trace command */
2244 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2245 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2246 switch (cmd_ctx->lsm->domain.type) {
2247 case LTTNG_DOMAIN_UST:
2248 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2249 ret = LTTNG_ERR_NO_USTCONSUMERD;
2250 goto error;
2251 }
2252 break;
2253 case LTTNG_DOMAIN_KERNEL:
2254 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2255 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2256 goto error;
2257 }
2258 break;
2259 }
2260 }
2261
2262 /*
2263 * Check that the UID or GID match that of the tracing session.
2264 * The root user can interact with all sessions.
2265 */
2266 if (need_tracing_session) {
2267 if (!session_access_ok(cmd_ctx->session,
2268 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2269 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2270 ret = LTTNG_ERR_EPERM;
2271 goto error;
2272 }
2273 }
2274
2275 /* Process by command type */
2276 switch (cmd_ctx->lsm->cmd_type) {
2277 case LTTNG_ADD_CONTEXT:
2278 {
2279 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2280 cmd_ctx->lsm->u.context.channel_name,
2281 cmd_ctx->lsm->u.context.event_name,
2282 &cmd_ctx->lsm->u.context.ctx);
2283 break;
2284 }
2285 case LTTNG_DISABLE_CHANNEL:
2286 {
2287 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2288 cmd_ctx->lsm->u.disable.channel_name);
2289 break;
2290 }
2291 case LTTNG_DISABLE_EVENT:
2292 {
2293 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2294 cmd_ctx->lsm->u.disable.channel_name,
2295 cmd_ctx->lsm->u.disable.name);
2296 break;
2297 }
2298 case LTTNG_DISABLE_ALL_EVENT:
2299 {
2300 DBG("Disabling all events");
2301
2302 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2303 cmd_ctx->lsm->u.disable.channel_name);
2304 break;
2305 }
2306 case LTTNG_DISABLE_CONSUMER:
2307 {
2308 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2309 break;
2310 }
2311 case LTTNG_ENABLE_CHANNEL:
2312 {
2313 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2314 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2315 break;
2316 }
2317 case LTTNG_ENABLE_CONSUMER:
2318 {
2319 /*
2320 * XXX: 0 means that this URI should be applied on the session. Should
2321 * be a DOMAIN enuam.
2322 */
2323 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2324 if (ret != LTTNG_OK) {
2325 goto error;
2326 }
2327
2328 if (cmd_ctx->lsm->domain.type == 0) {
2329 /* Add the URI for the UST session if a consumer is present. */
2330 if (cmd_ctx->session->ust_session &&
2331 cmd_ctx->session->ust_session->consumer) {
2332 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2333 } else if (cmd_ctx->session->kernel_session &&
2334 cmd_ctx->session->kernel_session->consumer) {
2335 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2336 cmd_ctx->session);
2337 }
2338 }
2339 break;
2340 }
2341 case LTTNG_ENABLE_EVENT:
2342 {
2343 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2344 cmd_ctx->lsm->u.enable.channel_name,
2345 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
2346 break;
2347 }
2348 case LTTNG_ENABLE_ALL_EVENT:
2349 {
2350 DBG("Enabling all events");
2351
2352 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2353 cmd_ctx->lsm->u.enable.channel_name,
2354 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
2355 break;
2356 }
2357 case LTTNG_LIST_TRACEPOINTS:
2358 {
2359 struct lttng_event *events;
2360 ssize_t nb_events;
2361
2362 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2363 if (nb_events < 0) {
2364 /* Return value is a negative lttng_error_code. */
2365 ret = -nb_events;
2366 goto error;
2367 }
2368
2369 /*
2370 * Setup lttng message with payload size set to the event list size in
2371 * bytes and then copy list into the llm payload.
2372 */
2373 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2374 if (ret < 0) {
2375 free(events);
2376 goto setup_error;
2377 }
2378
2379 /* Copy event list into message payload */
2380 memcpy(cmd_ctx->llm->payload, events,
2381 sizeof(struct lttng_event) * nb_events);
2382
2383 free(events);
2384
2385 ret = LTTNG_OK;
2386 break;
2387 }
2388 case LTTNG_LIST_TRACEPOINT_FIELDS:
2389 {
2390 struct lttng_event_field *fields;
2391 ssize_t nb_fields;
2392
2393 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2394 &fields);
2395 if (nb_fields < 0) {
2396 /* Return value is a negative lttng_error_code. */
2397 ret = -nb_fields;
2398 goto error;
2399 }
2400
2401 /*
2402 * Setup lttng message with payload size set to the event list size in
2403 * bytes and then copy list into the llm payload.
2404 */
2405 ret = setup_lttng_msg(cmd_ctx,
2406 sizeof(struct lttng_event_field) * nb_fields);
2407 if (ret < 0) {
2408 free(fields);
2409 goto setup_error;
2410 }
2411
2412 /* Copy event list into message payload */
2413 memcpy(cmd_ctx->llm->payload, fields,
2414 sizeof(struct lttng_event_field) * nb_fields);
2415
2416 free(fields);
2417
2418 ret = LTTNG_OK;
2419 break;
2420 }
2421 case LTTNG_SET_CONSUMER_URI:
2422 {
2423 size_t nb_uri, len;
2424 struct lttng_uri *uris;
2425
2426 nb_uri = cmd_ctx->lsm->u.uri.size;
2427 len = nb_uri * sizeof(struct lttng_uri);
2428
2429 if (nb_uri == 0) {
2430 ret = LTTNG_ERR_INVALID;
2431 goto error;
2432 }
2433
2434 uris = zmalloc(len);
2435 if (uris == NULL) {
2436 ret = LTTNG_ERR_FATAL;
2437 goto error;
2438 }
2439
2440 /* Receive variable len data */
2441 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2442 ret = lttcomm_recv_unix_sock(sock, uris, len);
2443 if (ret <= 0) {
2444 DBG("No URIs received from client... continuing");
2445 *sock_error = 1;
2446 ret = LTTNG_ERR_SESSION_FAIL;
2447 goto error;
2448 }
2449
2450 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2451 nb_uri, uris);
2452 if (ret != LTTNG_OK) {
2453 goto error;
2454 }
2455
2456 /*
2457 * XXX: 0 means that this URI should be applied on the session. Should
2458 * be a DOMAIN enuam.
2459 */
2460 if (cmd_ctx->lsm->domain.type == 0) {
2461 /* Add the URI for the UST session if a consumer is present. */
2462 if (cmd_ctx->session->ust_session &&
2463 cmd_ctx->session->ust_session->consumer) {
2464 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2465 nb_uri, uris);
2466 } else if (cmd_ctx->session->kernel_session &&
2467 cmd_ctx->session->kernel_session->consumer) {
2468 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2469 cmd_ctx->session, nb_uri, uris);
2470 }
2471 }
2472
2473 break;
2474 }
2475 case LTTNG_START_TRACE:
2476 {
2477 ret = cmd_start_trace(cmd_ctx->session);
2478 break;
2479 }
2480 case LTTNG_STOP_TRACE:
2481 {
2482 ret = cmd_stop_trace(cmd_ctx->session);
2483 break;
2484 }
2485 case LTTNG_CREATE_SESSION:
2486 {
2487 size_t nb_uri, len;
2488 struct lttng_uri *uris = NULL;
2489
2490 nb_uri = cmd_ctx->lsm->u.uri.size;
2491 len = nb_uri * sizeof(struct lttng_uri);
2492
2493 if (nb_uri > 0) {
2494 uris = zmalloc(len);
2495 if (uris == NULL) {
2496 ret = LTTNG_ERR_FATAL;
2497 goto error;
2498 }
2499
2500 /* Receive variable len data */
2501 DBG("Waiting for %zu URIs from client ...", nb_uri);
2502 ret = lttcomm_recv_unix_sock(sock, uris, len);
2503 if (ret <= 0) {
2504 DBG("No URIs received from client... continuing");
2505 *sock_error = 1;
2506 ret = LTTNG_ERR_SESSION_FAIL;
2507 goto error;
2508 }
2509
2510 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2511 DBG("Creating session with ONE network URI is a bad call");
2512 ret = LTTNG_ERR_SESSION_FAIL;
2513 goto error;
2514 }
2515 }
2516
2517 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2518 &cmd_ctx->creds);
2519
2520 break;
2521 }
2522 case LTTNG_DESTROY_SESSION:
2523 {
2524 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2525
2526 /* Set session to NULL so we do not unlock it after free. */
2527 cmd_ctx->session = NULL;
2528 break;
2529 }
2530 case LTTNG_LIST_DOMAINS:
2531 {
2532 ssize_t nb_dom;
2533 struct lttng_domain *domains;
2534
2535 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2536 if (nb_dom < 0) {
2537 /* Return value is a negative lttng_error_code. */
2538 ret = -nb_dom;
2539 goto error;
2540 }
2541
2542 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2543 if (ret < 0) {
2544 goto setup_error;
2545 }
2546
2547 /* Copy event list into message payload */
2548 memcpy(cmd_ctx->llm->payload, domains,
2549 nb_dom * sizeof(struct lttng_domain));
2550
2551 free(domains);
2552
2553 ret = LTTNG_OK;
2554 break;
2555 }
2556 case LTTNG_LIST_CHANNELS:
2557 {
2558 int nb_chan;
2559 struct lttng_channel *channels;
2560
2561 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2562 cmd_ctx->session, &channels);
2563 if (nb_chan < 0) {
2564 /* Return value is a negative lttng_error_code. */
2565 ret = -nb_chan;
2566 goto error;
2567 }
2568
2569 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2570 if (ret < 0) {
2571 goto setup_error;
2572 }
2573
2574 /* Copy event list into message payload */
2575 memcpy(cmd_ctx->llm->payload, channels,
2576 nb_chan * sizeof(struct lttng_channel));
2577
2578 free(channels);
2579
2580 ret = LTTNG_OK;
2581 break;
2582 }
2583 case LTTNG_LIST_EVENTS:
2584 {
2585 ssize_t nb_event;
2586 struct lttng_event *events = NULL;
2587
2588 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2589 cmd_ctx->lsm->u.list.channel_name, &events);
2590 if (nb_event < 0) {
2591 /* Return value is a negative lttng_error_code. */
2592 ret = -nb_event;
2593 goto error;
2594 }
2595
2596 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2597 if (ret < 0) {
2598 goto setup_error;
2599 }
2600
2601 /* Copy event list into message payload */
2602 memcpy(cmd_ctx->llm->payload, events,
2603 nb_event * sizeof(struct lttng_event));
2604
2605 free(events);
2606
2607 ret = LTTNG_OK;
2608 break;
2609 }
2610 case LTTNG_LIST_SESSIONS:
2611 {
2612 unsigned int nr_sessions;
2613
2614 session_lock_list();
2615 nr_sessions = lttng_sessions_count(
2616 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2617 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2618
2619 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2620 if (ret < 0) {
2621 session_unlock_list();
2622 goto setup_error;
2623 }
2624
2625 /* Filled the session array */
2626 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2627 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2628 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2629
2630 session_unlock_list();
2631
2632 ret = LTTNG_OK;
2633 break;
2634 }
2635 case LTTNG_CALIBRATE:
2636 {
2637 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2638 &cmd_ctx->lsm->u.calibrate);
2639 break;
2640 }
2641 case LTTNG_REGISTER_CONSUMER:
2642 {
2643 struct consumer_data *cdata;
2644
2645 switch (cmd_ctx->lsm->domain.type) {
2646 case LTTNG_DOMAIN_KERNEL:
2647 cdata = &kconsumer_data;
2648 break;
2649 default:
2650 ret = LTTNG_ERR_UND;
2651 goto error;
2652 }
2653
2654 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2655 cmd_ctx->lsm->u.reg.path, cdata);
2656 break;
2657 }
2658 case LTTNG_SET_FILTER:
2659 {
2660 struct lttng_filter_bytecode *bytecode;
2661
2662 if (cmd_ctx->lsm->u.filter.bytecode_len > 65336) {
2663 ret = LTTNG_ERR_FILTER_INVAL;
2664 goto error;
2665 }
2666 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2667 if (!bytecode) {
2668 ret = LTTNG_ERR_FILTER_NOMEM;
2669 goto error;
2670 }
2671 /* Receive var. len. data */
2672 DBG("Receiving var len data from client ...");
2673 ret = lttcomm_recv_unix_sock(sock, bytecode,
2674 cmd_ctx->lsm->u.filter.bytecode_len);
2675 if (ret <= 0) {
2676 DBG("Nothing recv() from client var len data... continuing");
2677 *sock_error = 1;
2678 ret = LTTNG_ERR_FILTER_INVAL;
2679 goto error;
2680 }
2681
2682 if (bytecode->len + sizeof(*bytecode)
2683 != cmd_ctx->lsm->u.filter.bytecode_len) {
2684 free(bytecode);
2685 ret = LTTNG_ERR_FILTER_INVAL;
2686 goto error;
2687 }
2688
2689 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2690 cmd_ctx->lsm->u.filter.channel_name,
2691 cmd_ctx->lsm->u.filter.event_name,
2692 bytecode);
2693 break;
2694 }
2695 default:
2696 ret = LTTNG_ERR_UND;
2697 break;
2698 }
2699
2700 error:
2701 if (cmd_ctx->llm == NULL) {
2702 DBG("Missing llm structure. Allocating one.");
2703 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2704 goto setup_error;
2705 }
2706 }
2707 /* Set return code */
2708 cmd_ctx->llm->ret_code = ret;
2709 setup_error:
2710 if (cmd_ctx->session) {
2711 session_unlock(cmd_ctx->session);
2712 }
2713 if (need_tracing_session) {
2714 session_unlock_list();
2715 }
2716 init_setup_error:
2717 return ret;
2718 }
2719
2720 /*
2721 * Thread managing health check socket.
2722 */
2723 static void *thread_manage_health(void *data)
2724 {
2725 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2726 uint32_t revents, nb_fd;
2727 struct lttng_poll_event events;
2728 struct lttcomm_health_msg msg;
2729 struct lttcomm_health_data reply;
2730
2731 DBG("[thread] Manage health check started");
2732
2733 rcu_register_thread();
2734
2735 /* Create unix socket */
2736 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2737 if (sock < 0) {
2738 ERR("Unable to create health check Unix socket");
2739 ret = -1;
2740 goto error;
2741 }
2742
2743 ret = lttcomm_listen_unix_sock(sock);
2744 if (ret < 0) {
2745 goto error;
2746 }
2747
2748 /*
2749 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2750 * more will be added to this poll set.
2751 */
2752 ret = create_thread_poll_set(&events, 2);
2753 if (ret < 0) {
2754 goto error;
2755 }
2756
2757 /* Add the application registration socket */
2758 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2759 if (ret < 0) {
2760 goto error;
2761 }
2762
2763 while (1) {
2764 DBG("Health check ready");
2765
2766 nb_fd = LTTNG_POLL_GETNB(&events);
2767
2768 /* Inifinite blocking call, waiting for transmission */
2769 restart:
2770 ret = lttng_poll_wait(&events, -1);
2771 if (ret < 0) {
2772 /*
2773 * Restart interrupted system call.
2774 */
2775 if (errno == EINTR) {
2776 goto restart;
2777 }
2778 goto error;
2779 }
2780
2781 for (i = 0; i < nb_fd; i++) {
2782 /* Fetch once the poll data */
2783 revents = LTTNG_POLL_GETEV(&events, i);
2784 pollfd = LTTNG_POLL_GETFD(&events, i);
2785
2786 /* Thread quit pipe has been closed. Killing thread. */
2787 ret = check_thread_quit_pipe(pollfd, revents);
2788 if (ret) {
2789 err = 0;
2790 goto exit;
2791 }
2792
2793 /* Event on the registration socket */
2794 if (pollfd == sock) {
2795 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2796 ERR("Health socket poll error");
2797 goto error;
2798 }
2799 }
2800 }
2801
2802 new_sock = lttcomm_accept_unix_sock(sock);
2803 if (new_sock < 0) {
2804 goto error;
2805 }
2806
2807 DBG("Receiving data from client for health...");
2808 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
2809 if (ret <= 0) {
2810 DBG("Nothing recv() from client... continuing");
2811 ret = close(new_sock);
2812 if (ret) {
2813 PERROR("close");
2814 }
2815 new_sock = -1;
2816 continue;
2817 }
2818
2819 rcu_thread_online();
2820
2821 switch (msg.component) {
2822 case LTTNG_HEALTH_CMD:
2823 reply.ret_code = health_check_state(&health_thread_cmd);
2824 break;
2825 case LTTNG_HEALTH_APP_MANAGE:
2826 reply.ret_code = health_check_state(&health_thread_app_manage);
2827 break;
2828 case LTTNG_HEALTH_APP_REG:
2829 reply.ret_code = health_check_state(&health_thread_app_reg);
2830 break;
2831 case LTTNG_HEALTH_KERNEL:
2832 reply.ret_code = health_check_state(&health_thread_kernel);
2833 break;
2834 case LTTNG_HEALTH_CONSUMER:
2835 reply.ret_code = check_consumer_health();
2836 break;
2837 case LTTNG_HEALTH_ALL:
2838 reply.ret_code =
2839 health_check_state(&health_thread_app_manage) &&
2840 health_check_state(&health_thread_app_reg) &&
2841 health_check_state(&health_thread_cmd) &&
2842 health_check_state(&health_thread_kernel) &&
2843 check_consumer_health();
2844 break;
2845 default:
2846 reply.ret_code = LTTNG_ERR_UND;
2847 break;
2848 }
2849
2850 /*
2851 * Flip ret value since 0 is a success and 1 indicates a bad health for
2852 * the client where in the sessiond it is the opposite. Again, this is
2853 * just to make things easier for us poor developer which enjoy a lot
2854 * lazyness.
2855 */
2856 if (reply.ret_code == 0 || reply.ret_code == 1) {
2857 reply.ret_code = !reply.ret_code;
2858 }
2859
2860 DBG2("Health check return value %d", reply.ret_code);
2861
2862 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
2863 if (ret < 0) {
2864 ERR("Failed to send health data back to client");
2865 }
2866
2867 /* End of transmission */
2868 ret = close(new_sock);
2869 if (ret) {
2870 PERROR("close");
2871 }
2872 new_sock = -1;
2873 }
2874
2875 exit:
2876 error:
2877 if (err) {
2878 ERR("Health error occurred in %s", __func__);
2879 }
2880 DBG("Health check thread dying");
2881 unlink(health_unix_sock_path);
2882 if (sock >= 0) {
2883 ret = close(sock);
2884 if (ret) {
2885 PERROR("close");
2886 }
2887 }
2888 if (new_sock >= 0) {
2889 ret = close(new_sock);
2890 if (ret) {
2891 PERROR("close");
2892 }
2893 }
2894
2895 lttng_poll_clean(&events);
2896
2897 rcu_unregister_thread();
2898 return NULL;
2899 }
2900
2901 /*
2902 * This thread manage all clients request using the unix client socket for
2903 * communication.
2904 */
2905 static void *thread_manage_clients(void *data)
2906 {
2907 int sock = -1, ret, i, pollfd, err = -1;
2908 int sock_error;
2909 uint32_t revents, nb_fd;
2910 struct command_ctx *cmd_ctx = NULL;
2911 struct lttng_poll_event events;
2912
2913 DBG("[thread] Manage client started");
2914
2915 rcu_register_thread();
2916
2917 health_code_update(&health_thread_cmd);
2918
2919 ret = lttcomm_listen_unix_sock(client_sock);
2920 if (ret < 0) {
2921 goto error;
2922 }
2923
2924 /*
2925 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2926 * more will be added to this poll set.
2927 */
2928 ret = create_thread_poll_set(&events, 2);
2929 if (ret < 0) {
2930 goto error;
2931 }
2932
2933 /* Add the application registration socket */
2934 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
2935 if (ret < 0) {
2936 goto error;
2937 }
2938
2939 /*
2940 * Notify parent pid that we are ready to accept command for client side.
2941 */
2942 if (opt_sig_parent) {
2943 kill(ppid, SIGUSR1);
2944 }
2945
2946 health_code_update(&health_thread_cmd);
2947
2948 while (1) {
2949 DBG("Accepting client command ...");
2950
2951 nb_fd = LTTNG_POLL_GETNB(&events);
2952
2953 /* Inifinite blocking call, waiting for transmission */
2954 restart:
2955 health_poll_update(&health_thread_cmd);
2956 ret = lttng_poll_wait(&events, -1);
2957 health_poll_update(&health_thread_cmd);
2958 if (ret < 0) {
2959 /*
2960 * Restart interrupted system call.
2961 */
2962 if (errno == EINTR) {
2963 goto restart;
2964 }
2965 goto error;
2966 }
2967
2968 for (i = 0; i < nb_fd; i++) {
2969 /* Fetch once the poll data */
2970 revents = LTTNG_POLL_GETEV(&events, i);
2971 pollfd = LTTNG_POLL_GETFD(&events, i);
2972
2973 health_code_update(&health_thread_cmd);
2974
2975 /* Thread quit pipe has been closed. Killing thread. */
2976 ret = check_thread_quit_pipe(pollfd, revents);
2977 if (ret) {
2978 err = 0;
2979 goto exit;
2980 }
2981
2982 /* Event on the registration socket */
2983 if (pollfd == client_sock) {
2984 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2985 ERR("Client socket poll error");
2986 goto error;
2987 }
2988 }
2989 }
2990
2991 DBG("Wait for client response");
2992
2993 health_code_update(&health_thread_cmd);
2994
2995 sock = lttcomm_accept_unix_sock(client_sock);
2996 if (sock < 0) {
2997 goto error;
2998 }
2999
3000 /* Set socket option for credentials retrieval */
3001 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3002 if (ret < 0) {
3003 goto error;
3004 }
3005
3006 /* Allocate context command to process the client request */
3007 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3008 if (cmd_ctx == NULL) {
3009 PERROR("zmalloc cmd_ctx");
3010 goto error;
3011 }
3012
3013 /* Allocate data buffer for reception */
3014 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3015 if (cmd_ctx->lsm == NULL) {
3016 PERROR("zmalloc cmd_ctx->lsm");
3017 goto error;
3018 }
3019
3020 cmd_ctx->llm = NULL;
3021 cmd_ctx->session = NULL;
3022
3023 health_code_update(&health_thread_cmd);
3024
3025 /*
3026 * Data is received from the lttng client. The struct
3027 * lttcomm_session_msg (lsm) contains the command and data request of
3028 * the client.
3029 */
3030 DBG("Receiving data from client ...");
3031 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3032 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3033 if (ret <= 0) {
3034 DBG("Nothing recv() from client... continuing");
3035 ret = close(sock);
3036 if (ret) {
3037 PERROR("close");
3038 }
3039 sock = -1;
3040 clean_command_ctx(&cmd_ctx);
3041 continue;
3042 }
3043
3044 health_code_update(&health_thread_cmd);
3045
3046 // TODO: Validate cmd_ctx including sanity check for
3047 // security purpose.
3048
3049 rcu_thread_online();
3050 /*
3051 * This function dispatch the work to the kernel or userspace tracer
3052 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3053 * informations for the client. The command context struct contains
3054 * everything this function may needs.
3055 */
3056 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3057 rcu_thread_offline();
3058 if (ret < 0) {
3059 if (sock_error) {
3060 ret = close(sock);
3061 if (ret) {
3062 PERROR("close");
3063 }
3064 sock = -1;
3065 }
3066 /*
3067 * TODO: Inform client somehow of the fatal error. At
3068 * this point, ret < 0 means that a zmalloc failed
3069 * (ENOMEM). Error detected but still accept
3070 * command, unless a socket error has been
3071 * detected.
3072 */
3073 clean_command_ctx(&cmd_ctx);
3074 continue;
3075 }
3076
3077 health_code_update(&health_thread_cmd);
3078
3079 DBG("Sending response (size: %d, retcode: %s)",
3080 cmd_ctx->lttng_msg_size,
3081 lttng_strerror(-cmd_ctx->llm->ret_code));
3082 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3083 if (ret < 0) {
3084 ERR("Failed to send data back to client");
3085 }
3086
3087 /* End of transmission */
3088 ret = close(sock);
3089 if (ret) {
3090 PERROR("close");
3091 }
3092 sock = -1;
3093
3094 clean_command_ctx(&cmd_ctx);
3095
3096 health_code_update(&health_thread_cmd);
3097 }
3098
3099 exit:
3100 error:
3101 if (err) {
3102 health_error(&health_thread_cmd);
3103 ERR("Health error occurred in %s", __func__);
3104 }
3105 health_exit(&health_thread_cmd);
3106
3107 DBG("Client thread dying");
3108 unlink(client_unix_sock_path);
3109 if (client_sock >= 0) {
3110 ret = close(client_sock);
3111 if (ret) {
3112 PERROR("close");
3113 }
3114 }
3115 if (sock >= 0) {
3116 ret = close(sock);
3117 if (ret) {
3118 PERROR("close");
3119 }
3120 }
3121
3122 lttng_poll_clean(&events);
3123 clean_command_ctx(&cmd_ctx);
3124
3125 rcu_unregister_thread();
3126 return NULL;
3127 }
3128
3129
3130 /*
3131 * usage function on stderr
3132 */
3133 static void usage(void)
3134 {
3135 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3136 fprintf(stderr, " -h, --help Display this usage.\n");
3137 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3138 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3139 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3140 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3141 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3142 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3143 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3144 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3145 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3146 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3147 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3148 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3149 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3150 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3151 fprintf(stderr, " -V, --version Show version number.\n");
3152 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3153 fprintf(stderr, " -q, --quiet No output at all.\n");
3154 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3155 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3156 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3157 }
3158
3159 /*
3160 * daemon argument parsing
3161 */
3162 static int parse_args(int argc, char **argv)
3163 {
3164 int c;
3165
3166 static struct option long_options[] = {
3167 { "client-sock", 1, 0, 'c' },
3168 { "apps-sock", 1, 0, 'a' },
3169 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3170 { "kconsumerd-err-sock", 1, 0, 'E' },
3171 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3172 { "ustconsumerd32-err-sock", 1, 0, 'H' },
3173 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3174 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3175 { "consumerd32-path", 1, 0, 'u' },
3176 { "consumerd32-libdir", 1, 0, 'U' },
3177 { "consumerd64-path", 1, 0, 't' },
3178 { "consumerd64-libdir", 1, 0, 'T' },
3179 { "daemonize", 0, 0, 'd' },
3180 { "sig-parent", 0, 0, 'S' },
3181 { "help", 0, 0, 'h' },
3182 { "group", 1, 0, 'g' },
3183 { "version", 0, 0, 'V' },
3184 { "quiet", 0, 0, 'q' },
3185 { "verbose", 0, 0, 'v' },
3186 { "verbose-consumer", 0, 0, 'Z' },
3187 { "no-kernel", 0, 0, 'N' },
3188 { NULL, 0, 0, 0 }
3189 };
3190
3191 while (1) {
3192 int option_index = 0;
3193 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t",
3194 long_options, &option_index);
3195 if (c == -1) {
3196 break;
3197 }
3198
3199 switch (c) {