Fix: consumer health state
[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 "filter.h"
63 #include "health.h"
64 #include "testpoint.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 /*
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
87 */
88 static struct consumer_data kconsumer_data = {
89 .type = LTTNG_CONSUMER_KERNEL,
90 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
91 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
92 .err_sock = -1,
93 .cmd_sock = -1,
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
98 };
99 static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer32_data = {
111 .type = LTTNG_CONSUMER32_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121
122 /* Shared between threads */
123 static int dispatch_thread_exit;
124
125 /* Global application Unix socket path */
126 static char apps_unix_sock_path[PATH_MAX];
127 /* Global client Unix socket path */
128 static char client_unix_sock_path[PATH_MAX];
129 /* global wait shm path for UST */
130 static char wait_shm_path[PATH_MAX];
131 /* Global health check unix path */
132 static char health_unix_sock_path[PATH_MAX];
133
134 /* Sockets and FDs */
135 static int client_sock = -1;
136 static int apps_sock = -1;
137 int kernel_tracer_fd = -1;
138 static int kernel_poll_pipe[2] = { -1, -1 };
139
140 /*
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
143 */
144 static int thread_quit_pipe[2] = { -1, -1 };
145
146 /*
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
149 */
150 static int apps_cmd_pipe[2] = { -1, -1 };
151
152 /* Pthread, Mutexes and Semaphores */
153 static pthread_t apps_thread;
154 static pthread_t reg_apps_thread;
155 static pthread_t client_thread;
156 static pthread_t kernel_thread;
157 static pthread_t dispatch_thread;
158 static pthread_t health_thread;
159
160 /*
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 *
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
166 */
167 static struct ust_cmd_queue ust_cmd_queue;
168
169 /*
170 * Pointer initialized before thread creation.
171 *
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
175 *
176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
177 * session_lock_list and session_unlock_list for lock acquisition.
178 */
179 static struct ltt_session_list *session_list_ptr;
180
181 int ust_consumerd64_fd = -1;
182 int ust_consumerd32_fd = -1;
183
184 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
185 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
186 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
187 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
188
189 static const char *module_proc_lttng = "/proc/lttng";
190
191 /*
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
194 */
195 enum consumerd_state {
196 CONSUMER_STARTED = 1,
197 CONSUMER_STOPPED = 2,
198 CONSUMER_ERROR = 3,
199 };
200
201 /*
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
206 *
207 * The following example shows a possible race condition of this scheme:
208 *
209 * consumer thread error happens
210 * client cmd arrives
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
214 * ...
215 *
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
221 */
222 static enum consumerd_state ust_consumerd_state;
223 static enum consumerd_state kernel_consumerd_state;
224
225 /* Used for the health monitoring of the session daemon. See health.h */
226 struct health_state health_thread_cmd;
227 struct health_state health_thread_app_manage;
228 struct health_state health_thread_app_reg;
229 struct health_state health_thread_kernel;
230
231 static
232 void setup_consumerd_path(void)
233 {
234 const char *bin, *libdir;
235
236 /*
237 * Allow INSTALL_BIN_PATH to be used as a target path for the
238 * native architecture size consumer if CONFIG_CONSUMER*_PATH
239 * has not been defined.
240 */
241 #if (CAA_BITS_PER_LONG == 32)
242 if (!consumerd32_bin[0]) {
243 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
244 }
245 if (!consumerd32_libdir[0]) {
246 consumerd32_libdir = INSTALL_LIB_PATH;
247 }
248 #elif (CAA_BITS_PER_LONG == 64)
249 if (!consumerd64_bin[0]) {
250 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
251 }
252 if (!consumerd64_libdir[0]) {
253 consumerd64_libdir = INSTALL_LIB_PATH;
254 }
255 #else
256 #error "Unknown bitness"
257 #endif
258
259 /*
260 * runtime env. var. overrides the build default.
261 */
262 bin = getenv("LTTNG_CONSUMERD32_BIN");
263 if (bin) {
264 consumerd32_bin = bin;
265 }
266 bin = getenv("LTTNG_CONSUMERD64_BIN");
267 if (bin) {
268 consumerd64_bin = bin;
269 }
270 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
271 if (libdir) {
272 consumerd32_libdir = libdir;
273 }
274 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
275 if (libdir) {
276 consumerd64_libdir = libdir;
277 }
278 }
279
280 /*
281 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
282 */
283 static int create_thread_poll_set(struct lttng_poll_event *events,
284 unsigned int size)
285 {
286 int ret;
287
288 if (events == NULL || size == 0) {
289 ret = -1;
290 goto error;
291 }
292
293 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
294 if (ret < 0) {
295 goto error;
296 }
297
298 /* Add quit pipe */
299 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
300 if (ret < 0) {
301 goto error;
302 }
303
304 return 0;
305
306 error:
307 return ret;
308 }
309
310 /*
311 * Check if the thread quit pipe was triggered.
312 *
313 * Return 1 if it was triggered else 0;
314 */
315 static int check_thread_quit_pipe(int fd, uint32_t events)
316 {
317 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
318 return 1;
319 }
320
321 return 0;
322 }
323
324 /*
325 * Return group ID of the tracing group or -1 if not found.
326 */
327 static gid_t allowed_group(void)
328 {
329 struct group *grp;
330
331 if (opt_tracing_group) {
332 grp = getgrnam(opt_tracing_group);
333 } else {
334 grp = getgrnam(default_tracing_group);
335 }
336 if (!grp) {
337 return -1;
338 } else {
339 return grp->gr_gid;
340 }
341 }
342
343 /*
344 * Init thread quit pipe.
345 *
346 * Return -1 on error or 0 if all pipes are created.
347 */
348 static int init_thread_quit_pipe(void)
349 {
350 int ret, i;
351
352 ret = pipe(thread_quit_pipe);
353 if (ret < 0) {
354 PERROR("thread quit pipe");
355 goto error;
356 }
357
358 for (i = 0; i < 2; i++) {
359 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
360 if (ret < 0) {
361 PERROR("fcntl");
362 goto error;
363 }
364 }
365
366 error:
367 return ret;
368 }
369
370 /*
371 * Stop all threads by closing the thread quit pipe.
372 */
373 static void stop_threads(void)
374 {
375 int ret;
376
377 /* Stopping all threads */
378 DBG("Terminating all threads");
379 ret = notify_thread_pipe(thread_quit_pipe[1]);
380 if (ret < 0) {
381 ERR("write error on thread quit pipe");
382 }
383
384 /* Dispatch thread */
385 CMM_STORE_SHARED(dispatch_thread_exit, 1);
386 futex_nto1_wake(&ust_cmd_queue.futex);
387 }
388
389 /*
390 * Cleanup the daemon
391 */
392 static void cleanup(void)
393 {
394 int ret;
395 char *cmd;
396 struct ltt_session *sess, *stmp;
397
398 DBG("Cleaning up");
399
400 /* First thing first, stop all threads */
401 utils_close_pipe(thread_quit_pipe);
402
403 DBG("Removing %s directory", rundir);
404 ret = asprintf(&cmd, "rm -rf %s", rundir);
405 if (ret < 0) {
406 ERR("asprintf failed. Something is really wrong!");
407 }
408
409 /* Remove lttng run directory */
410 ret = system(cmd);
411 if (ret < 0) {
412 ERR("Unable to clean %s", rundir);
413 }
414 free(cmd);
415 free(rundir);
416
417 DBG("Cleaning up all sessions");
418
419 /* Destroy session list mutex */
420 if (session_list_ptr != NULL) {
421 pthread_mutex_destroy(&session_list_ptr->lock);
422
423 /* Cleanup ALL session */
424 cds_list_for_each_entry_safe(sess, stmp,
425 &session_list_ptr->head, list) {
426 cmd_destroy_session(sess, kernel_poll_pipe[1]);
427 }
428 }
429
430 DBG("Closing all UST sockets");
431 ust_app_clean_list();
432
433 if (is_root && !opt_no_kernel) {
434 DBG2("Closing kernel fd");
435 if (kernel_tracer_fd >= 0) {
436 ret = close(kernel_tracer_fd);
437 if (ret) {
438 PERROR("close");
439 }
440 }
441 DBG("Unloading kernel modules");
442 modprobe_remove_lttng_all();
443 }
444
445 utils_close_pipe(kernel_poll_pipe);
446 utils_close_pipe(apps_cmd_pipe);
447
448 /* <fun> */
449 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
450 "Matthew, BEET driven development works!%c[%dm",
451 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
452 /* </fun> */
453 }
454
455 /*
456 * Send data on a unix socket using the liblttsessiondcomm API.
457 *
458 * Return lttcomm error code.
459 */
460 static int send_unix_sock(int sock, void *buf, size_t len)
461 {
462 /* Check valid length */
463 if (len <= 0) {
464 return -1;
465 }
466
467 return lttcomm_send_unix_sock(sock, buf, len);
468 }
469
470 /*
471 * Free memory of a command context structure.
472 */
473 static void clean_command_ctx(struct command_ctx **cmd_ctx)
474 {
475 DBG("Clean command context structure");
476 if (*cmd_ctx) {
477 if ((*cmd_ctx)->llm) {
478 free((*cmd_ctx)->llm);
479 }
480 if ((*cmd_ctx)->lsm) {
481 free((*cmd_ctx)->lsm);
482 }
483 free(*cmd_ctx);
484 *cmd_ctx = NULL;
485 }
486 }
487
488 /*
489 * Notify UST applications using the shm mmap futex.
490 */
491 static int notify_ust_apps(int active)
492 {
493 char *wait_shm_mmap;
494
495 DBG("Notifying applications of session daemon state: %d", active);
496
497 /* See shm.c for this call implying mmap, shm and futex calls */
498 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
499 if (wait_shm_mmap == NULL) {
500 goto error;
501 }
502
503 /* Wake waiting process */
504 futex_wait_update((int32_t *) wait_shm_mmap, active);
505
506 /* Apps notified successfully */
507 return 0;
508
509 error:
510 return -1;
511 }
512
513 /*
514 * Setup the outgoing data buffer for the response (llm) by allocating the
515 * right amount of memory and copying the original information from the lsm
516 * structure.
517 *
518 * Return total size of the buffer pointed by buf.
519 */
520 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
521 {
522 int ret, buf_size;
523
524 buf_size = size;
525
526 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
527 if (cmd_ctx->llm == NULL) {
528 PERROR("zmalloc");
529 ret = -ENOMEM;
530 goto error;
531 }
532
533 /* Copy common data */
534 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
535 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
536
537 cmd_ctx->llm->data_size = size;
538 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
539
540 return buf_size;
541
542 error:
543 return ret;
544 }
545
546 /*
547 * Update the kernel poll set of all channel fd available over all tracing
548 * session. Add the wakeup pipe at the end of the set.
549 */
550 static int update_kernel_poll(struct lttng_poll_event *events)
551 {
552 int ret;
553 struct ltt_session *session;
554 struct ltt_kernel_channel *channel;
555
556 DBG("Updating kernel poll set");
557
558 session_lock_list();
559 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
560 session_lock(session);
561 if (session->kernel_session == NULL) {
562 session_unlock(session);
563 continue;
564 }
565
566 cds_list_for_each_entry(channel,
567 &session->kernel_session->channel_list.head, list) {
568 /* Add channel fd to the kernel poll set */
569 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
570 if (ret < 0) {
571 session_unlock(session);
572 goto error;
573 }
574 DBG("Channel fd %d added to kernel set", channel->fd);
575 }
576 session_unlock(session);
577 }
578 session_unlock_list();
579
580 return 0;
581
582 error:
583 session_unlock_list();
584 return -1;
585 }
586
587 /*
588 * Find the channel fd from 'fd' over all tracing session. When found, check
589 * for new channel stream and send those stream fds to the kernel consumer.
590 *
591 * Useful for CPU hotplug feature.
592 */
593 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
594 {
595 int ret = 0;
596 struct ltt_session *session;
597 struct ltt_kernel_session *ksess;
598 struct ltt_kernel_channel *channel;
599
600 DBG("Updating kernel streams for channel fd %d", fd);
601
602 session_lock_list();
603 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
604 session_lock(session);
605 if (session->kernel_session == NULL) {
606 session_unlock(session);
607 continue;
608 }
609 ksess = session->kernel_session;
610
611 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
612 if (channel->fd == fd) {
613 DBG("Channel found, updating kernel streams");
614 ret = kernel_open_channel_stream(channel);
615 if (ret < 0) {
616 goto error;
617 }
618
619 /*
620 * Have we already sent fds to the consumer? If yes, it means
621 * that tracing is started so it is safe to send our updated
622 * stream fds.
623 */
624 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
625 struct lttng_ht_iter iter;
626 struct consumer_socket *socket;
627
628
629 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
630 &iter.iter, socket, node.node) {
631 /* Code flow error */
632 assert(socket->fd >= 0);
633
634 pthread_mutex_lock(socket->lock);
635 ret = kernel_consumer_send_channel_stream(socket->fd,
636 channel, ksess);
637 pthread_mutex_unlock(socket->lock);
638 if (ret < 0) {
639 goto error;
640 }
641 }
642 }
643 goto error;
644 }
645 }
646 session_unlock(session);
647 }
648 session_unlock_list();
649 return ret;
650
651 error:
652 session_unlock(session);
653 session_unlock_list();
654 return ret;
655 }
656
657 /*
658 * For each tracing session, update newly registered apps.
659 */
660 static void update_ust_app(int app_sock)
661 {
662 struct ltt_session *sess, *stmp;
663
664 session_lock_list();
665
666 /* For all tracing session(s) */
667 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
668 session_lock(sess);
669 if (sess->ust_session) {
670 ust_app_global_update(sess->ust_session, app_sock);
671 }
672 session_unlock(sess);
673 }
674
675 session_unlock_list();
676 }
677
678 /*
679 * This thread manage event coming from the kernel.
680 *
681 * Features supported in this thread:
682 * -) CPU Hotplug
683 */
684 static void *thread_manage_kernel(void *data)
685 {
686 int ret, i, pollfd, update_poll_flag = 1, err = -1;
687 uint32_t revents, nb_fd;
688 char tmp;
689 struct lttng_poll_event events;
690
691 DBG("Thread manage kernel started");
692
693 testpoint(thread_manage_kernel);
694
695 health_code_update(&health_thread_kernel);
696
697 testpoint(thread_manage_kernel_before_loop);
698
699 ret = create_thread_poll_set(&events, 2);
700 if (ret < 0) {
701 goto error_poll_create;
702 }
703
704 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
705 if (ret < 0) {
706 goto error;
707 }
708
709 while (1) {
710 health_code_update(&health_thread_kernel);
711
712 if (update_poll_flag == 1) {
713 /*
714 * Reset number of fd in the poll set. Always 2 since there is the thread
715 * quit pipe and the kernel pipe.
716 */
717 events.nb_fd = 2;
718
719 ret = update_kernel_poll(&events);
720 if (ret < 0) {
721 goto error;
722 }
723 update_poll_flag = 0;
724 }
725
726 nb_fd = LTTNG_POLL_GETNB(&events);
727
728 DBG("Thread kernel polling on %d fds", nb_fd);
729
730 /* Zeroed the poll events */
731 lttng_poll_reset(&events);
732
733 /* Poll infinite value of time */
734 restart:
735 health_poll_update(&health_thread_kernel);
736 ret = lttng_poll_wait(&events, -1);
737 health_poll_update(&health_thread_kernel);
738 if (ret < 0) {
739 /*
740 * Restart interrupted system call.
741 */
742 if (errno == EINTR) {
743 goto restart;
744 }
745 goto error;
746 } else if (ret == 0) {
747 /* Should not happen since timeout is infinite */
748 ERR("Return value of poll is 0 with an infinite timeout.\n"
749 "This should not have happened! Continuing...");
750 continue;
751 }
752
753 for (i = 0; i < nb_fd; i++) {
754 /* Fetch once the poll data */
755 revents = LTTNG_POLL_GETEV(&events, i);
756 pollfd = LTTNG_POLL_GETFD(&events, i);
757
758 health_code_update(&health_thread_kernel);
759
760 /* Thread quit pipe has been closed. Killing thread. */
761 ret = check_thread_quit_pipe(pollfd, revents);
762 if (ret) {
763 err = 0;
764 goto exit;
765 }
766
767 /* Check for data on kernel pipe */
768 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
769 ret = read(kernel_poll_pipe[0], &tmp, 1);
770 update_poll_flag = 1;
771 continue;
772 } else {
773 /*
774 * New CPU detected by the kernel. Adding kernel stream to
775 * kernel session and updating the kernel consumer
776 */
777 if (revents & LPOLLIN) {
778 ret = update_kernel_stream(&kconsumer_data, pollfd);
779 if (ret < 0) {
780 continue;
781 }
782 break;
783 /*
784 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
785 * and unregister kernel stream at this point.
786 */
787 }
788 }
789 }
790 }
791
792 exit:
793 error:
794 lttng_poll_clean(&events);
795 error_poll_create:
796 if (err) {
797 health_error(&health_thread_kernel);
798 ERR("Health error occurred in %s", __func__);
799 }
800 health_exit(&health_thread_kernel);
801 DBG("Kernel thread dying");
802 return NULL;
803 }
804
805 /*
806 * Signal pthread condition of the consumer data that the thread.
807 */
808 static void signal_consumer_condition(struct consumer_data *data, int state)
809 {
810 pthread_mutex_lock(&data->cond_mutex);
811
812 /*
813 * The state is set before signaling. It can be any value, it's the waiter
814 * job to correctly interpret this condition variable associated to the
815 * consumer pthread_cond.
816 *
817 * A value of 0 means that the corresponding thread of the consumer data
818 * was not started. 1 indicates that the thread has started and is ready
819 * for action. A negative value means that there was an error during the
820 * thread bootstrap.
821 */
822 data->consumer_thread_is_ready = state;
823 (void) pthread_cond_signal(&data->cond);
824
825 pthread_mutex_unlock(&data->cond_mutex);
826 }
827
828 /*
829 * This thread manage the consumer error sent back to the session daemon.
830 */
831 static void *thread_manage_consumer(void *data)
832 {
833 int sock = -1, i, ret, pollfd, err = -1;
834 uint32_t revents, nb_fd;
835 enum lttcomm_return_code code;
836 struct lttng_poll_event events;
837 struct consumer_data *consumer_data = data;
838
839 DBG("[thread] Manage consumer started");
840
841 /*
842 * Since the consumer thread can be spawned at any moment in time, we init
843 * the health to a poll status (1, which is a valid health over time).
844 * When the thread starts, we update here the health to a "code" path being
845 * an even value so this thread, when reaching a poll wait, does not
846 * trigger an error with an even value.
847 *
848 * Here is the use case we avoid.
849 *
850 * +1: the first poll update during initialization (main())
851 * +2 * x: multiple code update once in this thread.
852 * +1: poll wait in this thread (being a good health state).
853 * == even number which after the wait period shows as a bad health.
854 *
855 * In a nutshell, the following poll update to the health state brings back
856 * the state to an even value meaning a code path.
857 */
858 health_poll_update(&consumer_data->health);
859
860 health_code_update(&consumer_data->health);
861
862 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
863 if (ret < 0) {
864 goto error_listen;
865 }
866
867 /*
868 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
869 * Nothing more will be added to this poll set.
870 */
871 ret = create_thread_poll_set(&events, 2);
872 if (ret < 0) {
873 goto error_poll;
874 }
875
876 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
877 if (ret < 0) {
878 goto error;
879 }
880
881 nb_fd = LTTNG_POLL_GETNB(&events);
882
883 health_code_update(&consumer_data->health);
884
885 /* Inifinite blocking call, waiting for transmission */
886 restart:
887 health_poll_update(&consumer_data->health);
888
889 testpoint(thread_manage_consumer);
890
891 ret = lttng_poll_wait(&events, -1);
892 health_poll_update(&consumer_data->health);
893 if (ret < 0) {
894 /*
895 * Restart interrupted system call.
896 */
897 if (errno == EINTR) {
898 goto restart;
899 }
900 goto error;
901 }
902
903 for (i = 0; i < nb_fd; i++) {
904 /* Fetch once the poll data */
905 revents = LTTNG_POLL_GETEV(&events, i);
906 pollfd = LTTNG_POLL_GETFD(&events, i);
907
908 health_code_update(&consumer_data->health);
909
910 /* Thread quit pipe has been closed. Killing thread. */
911 ret = check_thread_quit_pipe(pollfd, revents);
912 if (ret) {
913 err = 0;
914 goto exit;
915 }
916
917 /* Event on the registration socket */
918 if (pollfd == consumer_data->err_sock) {
919 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
920 ERR("consumer err socket poll error");
921 goto error;
922 }
923 }
924 }
925
926 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
927 if (sock < 0) {
928 goto error;
929 }
930
931 health_code_update(&consumer_data->health);
932
933 DBG2("Receiving code from consumer err_sock");
934
935 /* Getting status code from kconsumerd */
936 ret = lttcomm_recv_unix_sock(sock, &code,
937 sizeof(enum lttcomm_return_code));
938 if (ret <= 0) {
939 goto error;
940 }
941
942 health_code_update(&consumer_data->health);
943
944 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
945 consumer_data->cmd_sock =
946 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
947 if (consumer_data->cmd_sock < 0) {
948 /* On error, signal condition and quit. */
949 signal_consumer_condition(consumer_data, -1);
950 PERROR("consumer connect");
951 goto error;
952 }
953 signal_consumer_condition(consumer_data, 1);
954 DBG("Consumer command socket ready");
955 } else {
956 ERR("consumer error when waiting for SOCK_READY : %s",
957 lttcomm_get_readable_code(-code));
958 goto error;
959 }
960
961 /* Remove the kconsumerd error sock since we've established a connexion */
962 ret = lttng_poll_del(&events, consumer_data->err_sock);
963 if (ret < 0) {
964 goto error;
965 }
966
967 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
968 if (ret < 0) {
969 goto error;
970 }
971
972 health_code_update(&consumer_data->health);
973
974 /* Update number of fd */
975 nb_fd = LTTNG_POLL_GETNB(&events);
976
977 /* Inifinite blocking call, waiting for transmission */
978 restart_poll:
979 health_poll_update(&consumer_data->health);
980 ret = lttng_poll_wait(&events, -1);
981 health_poll_update(&consumer_data->health);
982 if (ret < 0) {
983 /*
984 * Restart interrupted system call.
985 */
986 if (errno == EINTR) {
987 goto restart_poll;
988 }
989 goto error;
990 }
991
992 for (i = 0; i < nb_fd; i++) {
993 /* Fetch once the poll data */
994 revents = LTTNG_POLL_GETEV(&events, i);
995 pollfd = LTTNG_POLL_GETFD(&events, i);
996
997 health_code_update(&consumer_data->health);
998
999 /* Thread quit pipe has been closed. Killing thread. */
1000 ret = check_thread_quit_pipe(pollfd, revents);
1001 if (ret) {
1002 err = 0;
1003 goto exit;
1004 }
1005
1006 /* Event on the kconsumerd socket */
1007 if (pollfd == sock) {
1008 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1009 ERR("consumer err socket second poll error");
1010 goto error;
1011 }
1012 }
1013 }
1014
1015 health_code_update(&consumer_data->health);
1016
1017 /* Wait for any kconsumerd error */
1018 ret = lttcomm_recv_unix_sock(sock, &code,
1019 sizeof(enum lttcomm_return_code));
1020 if (ret <= 0) {
1021 ERR("consumer closed the command socket");
1022 goto error;
1023 }
1024
1025 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1026
1027 exit:
1028 error:
1029 /* Immediately set the consumerd state to stopped */
1030 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1031 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1032 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1033 consumer_data->type == LTTNG_CONSUMER32_UST) {
1034 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1035 } else {
1036 /* Code flow error... */
1037 assert(0);
1038 }
1039
1040 if (consumer_data->err_sock >= 0) {
1041 ret = close(consumer_data->err_sock);
1042 if (ret) {
1043 PERROR("close");
1044 }
1045 }
1046 if (consumer_data->cmd_sock >= 0) {
1047 ret = close(consumer_data->cmd_sock);
1048 if (ret) {
1049 PERROR("close");
1050 }
1051 }
1052 if (sock >= 0) {
1053 ret = close(sock);
1054 if (ret) {
1055 PERROR("close");
1056 }
1057 }
1058
1059 unlink(consumer_data->err_unix_sock_path);
1060 unlink(consumer_data->cmd_unix_sock_path);
1061 consumer_data->pid = 0;
1062
1063 lttng_poll_clean(&events);
1064 error_poll:
1065 error_listen:
1066 if (err) {
1067 health_error(&consumer_data->health);
1068 ERR("Health error occurred in %s", __func__);
1069 }
1070 health_exit(&consumer_data->health);
1071 DBG("consumer thread cleanup completed");
1072
1073 return NULL;
1074 }
1075
1076 /*
1077 * This thread manage application communication.
1078 */
1079 static void *thread_manage_apps(void *data)
1080 {
1081 int i, ret, pollfd, err = -1;
1082 uint32_t revents, nb_fd;
1083 struct ust_command ust_cmd;
1084 struct lttng_poll_event events;
1085
1086 DBG("[thread] Manage application started");
1087
1088 testpoint(thread_manage_apps);
1089
1090 rcu_register_thread();
1091 rcu_thread_online();
1092
1093 health_code_update(&health_thread_app_manage);
1094
1095 ret = create_thread_poll_set(&events, 2);
1096 if (ret < 0) {
1097 goto error_poll_create;
1098 }
1099
1100 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1101 if (ret < 0) {
1102 goto error;
1103 }
1104
1105 testpoint(thread_manage_apps_before_loop);
1106
1107 health_code_update(&health_thread_app_manage);
1108
1109 while (1) {
1110 /* Zeroed the events structure */
1111 lttng_poll_reset(&events);
1112
1113 nb_fd = LTTNG_POLL_GETNB(&events);
1114
1115 DBG("Apps thread polling on %d fds", nb_fd);
1116
1117 /* Inifinite blocking call, waiting for transmission */
1118 restart:
1119 health_poll_update(&health_thread_app_manage);
1120 ret = lttng_poll_wait(&events, -1);
1121 health_poll_update(&health_thread_app_manage);
1122 if (ret < 0) {
1123 /*
1124 * Restart interrupted system call.
1125 */
1126 if (errno == EINTR) {
1127 goto restart;
1128 }
1129 goto error;
1130 }
1131
1132 for (i = 0; i < nb_fd; i++) {
1133 /* Fetch once the poll data */
1134 revents = LTTNG_POLL_GETEV(&events, i);
1135 pollfd = LTTNG_POLL_GETFD(&events, i);
1136
1137 health_code_update(&health_thread_app_manage);
1138
1139 /* Thread quit pipe has been closed. Killing thread. */
1140 ret = check_thread_quit_pipe(pollfd, revents);
1141 if (ret) {
1142 err = 0;
1143 goto exit;
1144 }
1145
1146 /* Inspect the apps cmd pipe */
1147 if (pollfd == apps_cmd_pipe[0]) {
1148 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1149 ERR("Apps command pipe error");
1150 goto error;
1151 } else if (revents & LPOLLIN) {
1152 /* Empty pipe */
1153 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1154 if (ret < 0 || ret < sizeof(ust_cmd)) {
1155 PERROR("read apps cmd pipe");
1156 goto error;
1157 }
1158
1159 health_code_update(&health_thread_app_manage);
1160
1161 /* Register applicaton to the session daemon */
1162 ret = ust_app_register(&ust_cmd.reg_msg,
1163 ust_cmd.sock);
1164 if (ret == -ENOMEM) {
1165 goto error;
1166 } else if (ret < 0) {
1167 break;
1168 }
1169
1170 health_code_update(&health_thread_app_manage);
1171
1172 /*
1173 * Validate UST version compatibility.
1174 */
1175 ret = ust_app_validate_version(ust_cmd.sock);
1176 if (ret >= 0) {
1177 /*
1178 * Add channel(s) and event(s) to newly registered apps
1179 * from lttng global UST domain.
1180 */
1181 update_ust_app(ust_cmd.sock);
1182 }
1183
1184 health_code_update(&health_thread_app_manage);
1185
1186 ret = ust_app_register_done(ust_cmd.sock);
1187 if (ret < 0) {
1188 /*
1189 * If the registration is not possible, we simply
1190 * unregister the apps and continue
1191 */
1192 ust_app_unregister(ust_cmd.sock);
1193 } else {
1194 /*
1195 * We just need here to monitor the close of the UST
1196 * socket and poll set monitor those by default.
1197 * Listen on POLLIN (even if we never expect any
1198 * data) to ensure that hangup wakes us.
1199 */
1200 ret = lttng_poll_add(&events, ust_cmd.sock, LPOLLIN);
1201 if (ret < 0) {
1202 goto error;
1203 }
1204
1205 DBG("Apps with sock %d added to poll set",
1206 ust_cmd.sock);
1207 }
1208
1209 health_code_update(&health_thread_app_manage);
1210
1211 break;
1212 }
1213 } else {
1214 /*
1215 * At this point, we know that a registered application made
1216 * the event at poll_wait.
1217 */
1218 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1219 /* Removing from the poll set */
1220 ret = lttng_poll_del(&events, pollfd);
1221 if (ret < 0) {
1222 goto error;
1223 }
1224
1225 /* Socket closed on remote end. */
1226 ust_app_unregister(pollfd);
1227 break;
1228 }
1229 }
1230
1231 health_code_update(&health_thread_app_manage);
1232 }
1233 }
1234
1235 exit:
1236 error:
1237 lttng_poll_clean(&events);
1238 error_poll_create:
1239 if (err) {
1240 health_error(&health_thread_app_manage);
1241 ERR("Health error occurred in %s", __func__);
1242 }
1243 health_exit(&health_thread_app_manage);
1244 DBG("Application communication apps thread cleanup complete");
1245 rcu_thread_offline();
1246 rcu_unregister_thread();
1247 return NULL;
1248 }
1249
1250 /*
1251 * Dispatch request from the registration threads to the application
1252 * communication thread.
1253 */
1254 static void *thread_dispatch_ust_registration(void *data)
1255 {
1256 int ret;
1257 struct cds_wfq_node *node;
1258 struct ust_command *ust_cmd = NULL;
1259
1260 DBG("[thread] Dispatch UST command started");
1261
1262 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1263 /* Atomically prepare the queue futex */
1264 futex_nto1_prepare(&ust_cmd_queue.futex);
1265
1266 do {
1267 /* Dequeue command for registration */
1268 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1269 if (node == NULL) {
1270 DBG("Woken up but nothing in the UST command queue");
1271 /* Continue thread execution */
1272 break;
1273 }
1274
1275 ust_cmd = caa_container_of(node, struct ust_command, node);
1276
1277 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1278 " gid:%d sock:%d name:%s (version %d.%d)",
1279 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1280 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1281 ust_cmd->sock, ust_cmd->reg_msg.name,
1282 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1283 /*
1284 * Inform apps thread of the new application registration. This
1285 * call is blocking so we can be assured that the data will be read
1286 * at some point in time or wait to the end of the world :)
1287 */
1288 ret = write(apps_cmd_pipe[1], ust_cmd,
1289 sizeof(struct ust_command));
1290 if (ret < 0) {
1291 PERROR("write apps cmd pipe");
1292 if (errno == EBADF) {
1293 /*
1294 * We can't inform the application thread to process
1295 * registration. We will exit or else application
1296 * registration will not occur and tracing will never
1297 * start.
1298 */
1299 goto error;
1300 }
1301 }
1302 free(ust_cmd);
1303 } while (node != NULL);
1304
1305 /* Futex wait on queue. Blocking call on futex() */
1306 futex_nto1_wait(&ust_cmd_queue.futex);
1307 }
1308
1309 error:
1310 DBG("Dispatch thread dying");
1311 return NULL;
1312 }
1313
1314 /*
1315 * This thread manage application registration.
1316 */
1317 static void *thread_registration_apps(void *data)
1318 {
1319 int sock = -1, i, ret, pollfd, err = -1;
1320 uint32_t revents, nb_fd;
1321 struct lttng_poll_event events;
1322 /*
1323 * Get allocated in this thread, enqueued to a global queue, dequeued and
1324 * freed in the manage apps thread.
1325 */
1326 struct ust_command *ust_cmd = NULL;
1327
1328 DBG("[thread] Manage application registration started");
1329
1330 testpoint(thread_registration_apps);
1331
1332 ret = lttcomm_listen_unix_sock(apps_sock);
1333 if (ret < 0) {
1334 goto error_listen;
1335 }
1336
1337 /*
1338 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1339 * more will be added to this poll set.
1340 */
1341 ret = create_thread_poll_set(&events, 2);
1342 if (ret < 0) {
1343 goto error_create_poll;
1344 }
1345
1346 /* Add the application registration socket */
1347 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1348 if (ret < 0) {
1349 goto error_poll_add;
1350 }
1351
1352 /* Notify all applications to register */
1353 ret = notify_ust_apps(1);
1354 if (ret < 0) {
1355 ERR("Failed to notify applications or create the wait shared memory.\n"
1356 "Execution continues but there might be problem for already\n"
1357 "running applications that wishes to register.");
1358 }
1359
1360 while (1) {
1361 DBG("Accepting application registration");
1362
1363 nb_fd = LTTNG_POLL_GETNB(&events);
1364
1365 /* Inifinite blocking call, waiting for transmission */
1366 restart:
1367 health_poll_update(&health_thread_app_reg);
1368 ret = lttng_poll_wait(&events, -1);
1369 health_poll_update(&health_thread_app_reg);
1370 if (ret < 0) {
1371 /*
1372 * Restart interrupted system call.
1373 */
1374 if (errno == EINTR) {
1375 goto restart;
1376 }
1377 goto error;
1378 }
1379
1380 for (i = 0; i < nb_fd; i++) {
1381 health_code_update(&health_thread_app_reg);
1382
1383 /* Fetch once the poll data */
1384 revents = LTTNG_POLL_GETEV(&events, i);
1385 pollfd = LTTNG_POLL_GETFD(&events, i);
1386
1387 /* Thread quit pipe has been closed. Killing thread. */
1388 ret = check_thread_quit_pipe(pollfd, revents);
1389 if (ret) {
1390 err = 0;
1391 goto exit;
1392 }
1393
1394 /* Event on the registration socket */
1395 if (pollfd == apps_sock) {
1396 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1397 ERR("Register apps socket poll error");
1398 goto error;
1399 } else if (revents & LPOLLIN) {
1400 sock = lttcomm_accept_unix_sock(apps_sock);
1401 if (sock < 0) {
1402 goto error;
1403 }
1404
1405 /* Create UST registration command for enqueuing */
1406 ust_cmd = zmalloc(sizeof(struct ust_command));
1407 if (ust_cmd == NULL) {
1408 PERROR("ust command zmalloc");
1409 goto error;
1410 }
1411
1412 /*
1413 * Using message-based transmissions to ensure we don't
1414 * have to deal with partially received messages.
1415 */
1416 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1417 if (ret < 0) {
1418 ERR("Exhausted file descriptors allowed for applications.");
1419 free(ust_cmd);
1420 ret = close(sock);
1421 if (ret) {
1422 PERROR("close");
1423 }
1424 sock = -1;
1425 continue;
1426 }
1427 health_code_update(&health_thread_app_reg);
1428 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1429 sizeof(struct ust_register_msg));
1430 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1431 if (ret < 0) {
1432 PERROR("lttcomm_recv_unix_sock register apps");
1433 } else {
1434 ERR("Wrong size received on apps register");
1435 }
1436 free(ust_cmd);
1437 ret = close(sock);
1438 if (ret) {
1439 PERROR("close");
1440 }
1441 lttng_fd_put(LTTNG_FD_APPS, 1);
1442 sock = -1;
1443 continue;
1444 }
1445 health_code_update(&health_thread_app_reg);
1446
1447 ust_cmd->sock = sock;
1448 sock = -1;
1449
1450 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1451 " gid:%d sock:%d name:%s (version %d.%d)",
1452 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1453 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1454 ust_cmd->sock, ust_cmd->reg_msg.name,
1455 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1456
1457 /*
1458 * Lock free enqueue the registration request. The red pill
1459 * has been taken! This apps will be part of the *system*.
1460 */
1461 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1462
1463 /*
1464 * Wake the registration queue futex. Implicit memory
1465 * barrier with the exchange in cds_wfq_enqueue.
1466 */
1467 futex_nto1_wake(&ust_cmd_queue.futex);
1468 }
1469 }
1470 }
1471 }
1472
1473 exit:
1474 error:
1475 if (err) {
1476 health_error(&health_thread_app_reg);
1477 ERR("Health error occurred in %s", __func__);
1478 }
1479 health_exit(&health_thread_app_reg);
1480
1481 /* Notify that the registration thread is gone */
1482 notify_ust_apps(0);
1483
1484 if (apps_sock >= 0) {
1485 ret = close(apps_sock);
1486 if (ret) {
1487 PERROR("close");
1488 }
1489 }
1490 if (sock >= 0) {
1491 ret = close(sock);
1492 if (ret) {
1493 PERROR("close");
1494 }
1495 lttng_fd_put(LTTNG_FD_APPS, 1);
1496 }
1497 unlink(apps_unix_sock_path);
1498
1499 error_poll_add:
1500 lttng_poll_clean(&events);
1501 error_listen:
1502 error_create_poll:
1503 DBG("UST Registration thread cleanup complete");
1504
1505 return NULL;
1506 }
1507
1508 /*
1509 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1510 * exec or it will fails.
1511 */
1512 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1513 {
1514 int ret, clock_ret;
1515 struct timespec timeout;
1516
1517 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1518 consumer_data->consumer_thread_is_ready = 0;
1519
1520 /* Setup pthread condition */
1521 ret = pthread_condattr_init(&consumer_data->condattr);
1522 if (ret != 0) {
1523 errno = ret;
1524 PERROR("pthread_condattr_init consumer data");
1525 goto error;
1526 }
1527
1528 /*
1529 * Set the monotonic clock in order to make sure we DO NOT jump in time
1530 * between the clock_gettime() call and the timedwait call. See bug #324
1531 * for a more details and how we noticed it.
1532 */
1533 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1534 if (ret != 0) {
1535 errno = ret;
1536 PERROR("pthread_condattr_setclock consumer data");
1537 goto error;
1538 }
1539
1540 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1541 if (ret != 0) {
1542 errno = ret;
1543 PERROR("pthread_cond_init consumer data");
1544 goto error;
1545 }
1546
1547 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1548 consumer_data);
1549 if (ret != 0) {
1550 PERROR("pthread_create consumer");
1551 ret = -1;
1552 goto error;
1553 }
1554
1555 /* We are about to wait on a pthread condition */
1556 pthread_mutex_lock(&consumer_data->cond_mutex);
1557
1558 /* Get time for sem_timedwait absolute timeout */
1559 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1560 /*
1561 * Set the timeout for the condition timed wait even if the clock gettime
1562 * call fails since we might loop on that call and we want to avoid to
1563 * increment the timeout too many times.
1564 */
1565 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1566
1567 /*
1568 * The following loop COULD be skipped in some conditions so this is why we
1569 * set ret to 0 in order to make sure at least one round of the loop is
1570 * done.
1571 */
1572 ret = 0;
1573
1574 /*
1575 * Loop until the condition is reached or when a timeout is reached. Note
1576 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1577 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1578 * possible. This loop does not take any chances and works with both of
1579 * them.
1580 */
1581 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1582 if (clock_ret < 0) {
1583 PERROR("clock_gettime spawn consumer");
1584 /* Infinite wait for the consumerd thread to be ready */
1585 ret = pthread_cond_wait(&consumer_data->cond,
1586 &consumer_data->cond_mutex);
1587 } else {
1588 ret = pthread_cond_timedwait(&consumer_data->cond,
1589 &consumer_data->cond_mutex, &timeout);
1590 }
1591 }
1592
1593 /* Release the pthread condition */
1594 pthread_mutex_unlock(&consumer_data->cond_mutex);
1595
1596 if (ret != 0) {
1597 errno = ret;
1598 if (ret == ETIMEDOUT) {
1599 /*
1600 * Call has timed out so we kill the kconsumerd_thread and return
1601 * an error.
1602 */
1603 ERR("Condition timed out. The consumer thread was never ready."
1604 " Killing it");
1605 ret = pthread_cancel(consumer_data->thread);
1606 if (ret < 0) {
1607 PERROR("pthread_cancel consumer thread");
1608 }
1609 } else {
1610 PERROR("pthread_cond_wait failed consumer thread");
1611 }
1612 goto error;
1613 }
1614
1615 pthread_mutex_lock(&consumer_data->pid_mutex);
1616 if (consumer_data->pid == 0) {
1617 ERR("Consumerd did not start");
1618 pthread_mutex_unlock(&consumer_data->pid_mutex);
1619 goto error;
1620 }
1621 pthread_mutex_unlock(&consumer_data->pid_mutex);
1622
1623 return 0;
1624
1625 error:
1626 return ret;
1627 }
1628
1629 /*
1630 * Join consumer thread
1631 */
1632 static int join_consumer_thread(struct consumer_data *consumer_data)
1633 {
1634 void *status;
1635 int ret;
1636
1637 /* Consumer pid must be a real one. */
1638 if (consumer_data->pid > 0) {
1639 ret = kill(consumer_data->pid, SIGTERM);
1640 if (ret) {
1641 ERR("Error killing consumer daemon");
1642 return ret;
1643 }
1644 return pthread_join(consumer_data->thread, &status);
1645 } else {
1646 return 0;
1647 }
1648 }
1649
1650 /*
1651 * Fork and exec a consumer daemon (consumerd).
1652 *
1653 * Return pid if successful else -1.
1654 */
1655 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1656 {
1657 int ret;
1658 pid_t pid;
1659 const char *consumer_to_use;
1660 const char *verbosity;
1661 struct stat st;
1662
1663 DBG("Spawning consumerd");
1664
1665 pid = fork();
1666 if (pid == 0) {
1667 /*
1668 * Exec consumerd.
1669 */
1670 if (opt_verbose_consumer) {
1671 verbosity = "--verbose";
1672 } else {
1673 verbosity = "--quiet";
1674 }
1675 switch (consumer_data->type) {
1676 case LTTNG_CONSUMER_KERNEL:
1677 /*
1678 * Find out which consumerd to execute. We will first try the
1679 * 64-bit path, then the sessiond's installation directory, and
1680 * fallback on the 32-bit one,
1681 */
1682 DBG3("Looking for a kernel consumer at these locations:");
1683 DBG3(" 1) %s", consumerd64_bin);
1684 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1685 DBG3(" 3) %s", consumerd32_bin);
1686 if (stat(consumerd64_bin, &st) == 0) {
1687 DBG3("Found location #1");
1688 consumer_to_use = consumerd64_bin;
1689 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1690 DBG3("Found location #2");
1691 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1692 } else if (stat(consumerd32_bin, &st) == 0) {
1693 DBG3("Found location #3");
1694 consumer_to_use = consumerd32_bin;
1695 } else {
1696 DBG("Could not find any valid consumerd executable");
1697 break;
1698 }
1699 DBG("Using kernel consumer at: %s", consumer_to_use);
1700 execl(consumer_to_use,
1701 "lttng-consumerd", verbosity, "-k",
1702 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1703 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1704 NULL);
1705 break;
1706 case LTTNG_CONSUMER64_UST:
1707 {
1708 char *tmpnew = NULL;
1709
1710 if (consumerd64_libdir[0] != '\0') {
1711 char *tmp;
1712 size_t tmplen;
1713
1714 tmp = getenv("LD_LIBRARY_PATH");
1715 if (!tmp) {
1716 tmp = "";
1717 }
1718 tmplen = strlen("LD_LIBRARY_PATH=")
1719 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1720 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1721 if (!tmpnew) {
1722 ret = -ENOMEM;
1723 goto error;
1724 }
1725 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1726 strcat(tmpnew, consumerd64_libdir);
1727 if (tmp[0] != '\0') {
1728 strcat(tmpnew, ":");
1729 strcat(tmpnew, tmp);
1730 }
1731 ret = putenv(tmpnew);
1732 if (ret) {
1733 ret = -errno;
1734 goto error;
1735 }
1736 }
1737 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1738 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1739 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1740 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1741 NULL);
1742 if (consumerd64_libdir[0] != '\0') {
1743 free(tmpnew);
1744 }
1745 if (ret) {
1746 goto error;
1747 }
1748 break;
1749 }
1750 case LTTNG_CONSUMER32_UST:
1751 {
1752 char *tmpnew = NULL;
1753
1754 if (consumerd32_libdir[0] != '\0') {
1755 char *tmp;
1756 size_t tmplen;
1757
1758 tmp = getenv("LD_LIBRARY_PATH");
1759 if (!tmp) {
1760 tmp = "";
1761 }
1762 tmplen = strlen("LD_LIBRARY_PATH=")
1763 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1764 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1765 if (!tmpnew) {
1766 ret = -ENOMEM;
1767 goto error;
1768 }
1769 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1770 strcat(tmpnew, consumerd32_libdir);
1771 if (tmp[0] != '\0') {
1772 strcat(tmpnew, ":");
1773 strcat(tmpnew, tmp);
1774 }
1775 ret = putenv(tmpnew);
1776 if (ret) {
1777 ret = -errno;
1778 goto error;
1779 }
1780 }
1781 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1782 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1783 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1784 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1785 NULL);
1786 if (consumerd32_libdir[0] != '\0') {
1787 free(tmpnew);
1788 }
1789 if (ret) {
1790 goto error;
1791 }
1792 break;
1793 }
1794 default:
1795 PERROR("unknown consumer type");
1796 exit(EXIT_FAILURE);
1797 }
1798 if (errno != 0) {
1799 PERROR("kernel start consumer exec");
1800 }
1801 exit(EXIT_FAILURE);
1802 } else if (pid > 0) {
1803 ret = pid;
1804 } else {
1805 PERROR("start consumer fork");
1806 ret = -errno;
1807 }
1808 error:
1809 return ret;
1810 }
1811
1812 /*
1813 * Spawn the consumerd daemon and session daemon thread.
1814 */
1815 static int start_consumerd(struct consumer_data *consumer_data)
1816 {
1817 int ret;
1818
1819 pthread_mutex_lock(&consumer_data->pid_mutex);
1820 if (consumer_data->pid != 0) {
1821 pthread_mutex_unlock(&consumer_data->pid_mutex);
1822 goto end;
1823 }
1824
1825 ret = spawn_consumerd(consumer_data);
1826 if (ret < 0) {
1827 ERR("Spawning consumerd failed");
1828 pthread_mutex_unlock(&consumer_data->pid_mutex);
1829 goto error;
1830 }
1831
1832 /* Setting up the consumer_data pid */
1833 consumer_data->pid = ret;
1834 DBG2("Consumer pid %d", consumer_data->pid);
1835 pthread_mutex_unlock(&consumer_data->pid_mutex);
1836
1837 DBG2("Spawning consumer control thread");
1838 ret = spawn_consumer_thread(consumer_data);
1839 if (ret < 0) {
1840 ERR("Fatal error spawning consumer control thread");
1841 goto error;
1842 }
1843
1844 end:
1845 return 0;
1846
1847 error:
1848 return ret;
1849 }
1850
1851 /*
1852 * Compute health status of each consumer. If one of them is zero (bad
1853 * state), we return 0.
1854 */
1855 static int check_consumer_health(void)
1856 {
1857 int ret;
1858
1859 ret = health_check_state(&kconsumer_data.health) &&
1860 health_check_state(&ustconsumer32_data.health) &&
1861 health_check_state(&ustconsumer64_data.health);
1862
1863 DBG3("Health consumer check %d", ret);
1864
1865 return ret;
1866 }
1867
1868 /*
1869 * Setup necessary data for kernel tracer action.
1870 */
1871 static int init_kernel_tracer(void)
1872 {
1873 int ret;
1874
1875 /* Modprobe lttng kernel modules */
1876 ret = modprobe_lttng_control();
1877 if (ret < 0) {
1878 goto error;
1879 }
1880
1881 /* Open debugfs lttng */
1882 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1883 if (kernel_tracer_fd < 0) {
1884 DBG("Failed to open %s", module_proc_lttng);
1885 ret = -1;
1886 goto error_open;
1887 }
1888
1889 /* Validate kernel version */
1890 ret = kernel_validate_version(kernel_tracer_fd);
1891 if (ret < 0) {
1892 goto error_version;
1893 }
1894
1895 ret = modprobe_lttng_data();
1896 if (ret < 0) {
1897 goto error_modules;
1898 }
1899
1900 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1901 return 0;
1902
1903 error_version:
1904 modprobe_remove_lttng_control();
1905 ret = close(kernel_tracer_fd);
1906 if (ret) {
1907 PERROR("close");
1908 }
1909 kernel_tracer_fd = -1;
1910 return LTTNG_ERR_KERN_VERSION;
1911
1912 error_modules:
1913 ret = close(kernel_tracer_fd);
1914 if (ret) {
1915 PERROR("close");
1916 }
1917
1918 error_open:
1919 modprobe_remove_lttng_control();
1920
1921 error:
1922 WARN("No kernel tracer available");
1923 kernel_tracer_fd = -1;
1924 if (!is_root) {
1925 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1926 } else {
1927 return LTTNG_ERR_KERN_NA;
1928 }
1929 }
1930
1931
1932 /*
1933 * Copy consumer output from the tracing session to the domain session. The
1934 * function also applies the right modification on a per domain basis for the
1935 * trace files destination directory.
1936 */
1937 static int copy_session_consumer(int domain, struct ltt_session *session)
1938 {
1939 int ret;
1940 const char *dir_name;
1941 struct consumer_output *consumer;
1942
1943 assert(session);
1944 assert(session->consumer);
1945
1946 switch (domain) {
1947 case LTTNG_DOMAIN_KERNEL:
1948 DBG3("Copying tracing session consumer output in kernel session");
1949 /*
1950 * XXX: We should audit the session creation and what this function
1951 * does "extra" in order to avoid a destroy since this function is used
1952 * in the domain session creation (kernel and ust) only. Same for UST
1953 * domain.
1954 */
1955 if (session->kernel_session->consumer) {
1956 consumer_destroy_output(session->kernel_session->consumer);
1957 }
1958 session->kernel_session->consumer =
1959 consumer_copy_output(session->consumer);
1960 /* Ease our life a bit for the next part */
1961 consumer = session->kernel_session->consumer;
1962 dir_name = DEFAULT_KERNEL_TRACE_DIR;
1963 break;
1964 case LTTNG_DOMAIN_UST:
1965 DBG3("Copying tracing session consumer output in UST session");
1966 if (session->ust_session->consumer) {
1967 consumer_destroy_output(session->ust_session->consumer);
1968 }
1969 session->ust_session->consumer =
1970 consumer_copy_output(session->consumer);
1971 /* Ease our life a bit for the next part */
1972 consumer = session->ust_session->consumer;
1973 dir_name = DEFAULT_UST_TRACE_DIR;
1974 break;
1975 default:
1976 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
1977 goto error;
1978 }
1979
1980 /* Append correct directory to subdir */
1981 strncat(consumer->subdir, dir_name,
1982 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
1983 DBG3("Copy session consumer subdir %s", consumer->subdir);
1984
1985 ret = LTTNG_OK;
1986
1987 error:
1988 return ret;
1989 }
1990
1991 /*
1992 * Create an UST session and add it to the session ust list.
1993 */
1994 static int create_ust_session(struct ltt_session *session,
1995 struct lttng_domain *domain)
1996 {
1997 int ret;
1998 struct ltt_ust_session *lus = NULL;
1999
2000 assert(session);
2001 assert(domain);
2002 assert(session->consumer);
2003
2004 switch (domain->type) {
2005 case LTTNG_DOMAIN_UST:
2006 break;
2007 default:
2008 ERR("Unknown UST domain on create session %d", domain->type);
2009 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2010 goto error;
2011 }
2012
2013 DBG("Creating UST session");
2014
2015 lus = trace_ust_create_session(session->path, session->id, domain);
2016 if (lus == NULL) {
2017 ret = LTTNG_ERR_UST_SESS_FAIL;
2018 goto error;
2019 }
2020
2021 lus->uid = session->uid;
2022 lus->gid = session->gid;
2023 session->ust_session = lus;
2024
2025 /* Copy session output to the newly created UST session */
2026 ret = copy_session_consumer(domain->type, session);
2027 if (ret != LTTNG_OK) {
2028 goto error;
2029 }
2030
2031 return LTTNG_OK;
2032
2033 error:
2034 free(lus);
2035 session->ust_session = NULL;
2036 return ret;
2037 }
2038
2039 /*
2040 * Create a kernel tracer session then create the default channel.
2041 */
2042 static int create_kernel_session(struct ltt_session *session)
2043 {
2044 int ret;
2045
2046 DBG("Creating kernel session");
2047
2048 ret = kernel_create_session(session, kernel_tracer_fd);
2049 if (ret < 0) {
2050 ret = LTTNG_ERR_KERN_SESS_FAIL;
2051 goto error;
2052 }
2053
2054 /* Code flow safety */
2055 assert(session->kernel_session);
2056
2057 /* Copy session output to the newly created Kernel session */
2058 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2059 if (ret != LTTNG_OK) {
2060 goto error;
2061 }
2062
2063 /* Create directory(ies) on local filesystem. */
2064 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2065 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2066 ret = run_as_mkdir_recursive(
2067 session->kernel_session->consumer->dst.trace_path,
2068 S_IRWXU | S_IRWXG, session->uid, session->gid);
2069 if (ret < 0) {
2070 if (ret != -EEXIST) {
2071 ERR("Trace directory creation error");
2072 goto error;
2073 }
2074 }
2075 }
2076
2077 session->kernel_session->uid = session->uid;
2078 session->kernel_session->gid = session->gid;
2079
2080 return LTTNG_OK;
2081
2082 error:
2083 trace_kernel_destroy_session(session->kernel_session);
2084 session->kernel_session = NULL;
2085 return ret;
2086 }
2087
2088 /*
2089 * Count number of session permitted by uid/gid.
2090 */
2091 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2092 {
2093 unsigned int i = 0;
2094 struct ltt_session *session;
2095
2096 DBG("Counting number of available session for UID %d GID %d",
2097 uid, gid);
2098 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2099 /*
2100 * Only list the sessions the user can control.
2101 */
2102 if (!session_access_ok(session, uid, gid)) {
2103 continue;
2104 }
2105 i++;
2106 }
2107 return i;
2108 }
2109
2110 /*
2111 * Process the command requested by the lttng client within the command
2112 * context structure. This function make sure that the return structure (llm)
2113 * is set and ready for transmission before returning.
2114 *
2115 * Return any error encountered or 0 for success.
2116 *
2117 * "sock" is only used for special-case var. len data.
2118 */
2119 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2120 int *sock_error)
2121 {
2122 int ret = LTTNG_OK;
2123 int need_tracing_session = 1;
2124 int need_domain;
2125
2126 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2127
2128 *sock_error = 0;
2129
2130 switch (cmd_ctx->lsm->cmd_type) {
2131 case LTTNG_CREATE_SESSION:
2132 case LTTNG_DESTROY_SESSION:
2133 case LTTNG_LIST_SESSIONS:
2134 case LTTNG_LIST_DOMAINS:
2135 case LTTNG_START_TRACE:
2136 case LTTNG_STOP_TRACE:
2137 case LTTNG_DATA_AVAILABLE:
2138 need_domain = 0;
2139 break;
2140 default:
2141 need_domain = 1;
2142 }
2143
2144 if (opt_no_kernel && need_domain
2145 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2146 if (!is_root) {
2147 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2148 } else {
2149 ret = LTTNG_ERR_KERN_NA;
2150 }
2151 goto error;
2152 }
2153
2154 /* Deny register consumer if we already have a spawned consumer. */
2155 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2156 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2157 if (kconsumer_data.pid > 0) {
2158 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2159 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2160 goto error;
2161 }
2162 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2163 }
2164
2165 /*
2166 * Check for command that don't needs to allocate a returned payload. We do
2167 * this here so we don't have to make the call for no payload at each
2168 * command.
2169 */
2170 switch(cmd_ctx->lsm->cmd_type) {
2171 case LTTNG_LIST_SESSIONS:
2172 case LTTNG_LIST_TRACEPOINTS:
2173 case LTTNG_LIST_TRACEPOINT_FIELDS:
2174 case LTTNG_LIST_DOMAINS:
2175 case LTTNG_LIST_CHANNELS:
2176 case LTTNG_LIST_EVENTS:
2177 break;
2178 default:
2179 /* Setup lttng message with no payload */
2180 ret = setup_lttng_msg(cmd_ctx, 0);
2181 if (ret < 0) {
2182 /* This label does not try to unlock the session */
2183 goto init_setup_error;
2184 }
2185 }
2186
2187 /* Commands that DO NOT need a session. */
2188 switch (cmd_ctx->lsm->cmd_type) {
2189 case LTTNG_CREATE_SESSION:
2190 case LTTNG_CALIBRATE:
2191 case LTTNG_LIST_SESSIONS:
2192 case LTTNG_LIST_TRACEPOINTS:
2193 case LTTNG_LIST_TRACEPOINT_FIELDS:
2194 need_tracing_session = 0;
2195 break;
2196 default:
2197 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2198 /*
2199 * We keep the session list lock across _all_ commands
2200 * for now, because the per-session lock does not
2201 * handle teardown properly.
2202 */
2203 session_lock_list();
2204 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2205 if (cmd_ctx->session == NULL) {
2206 if (cmd_ctx->lsm->session.name != NULL) {
2207 ret = LTTNG_ERR_SESS_NOT_FOUND;
2208 } else {
2209 /* If no session name specified */
2210 ret = LTTNG_ERR_SELECT_SESS;
2211 }
2212 goto error;
2213 } else {
2214 /* Acquire lock for the session */
2215 session_lock(cmd_ctx->session);
2216 }
2217 break;
2218 }
2219
2220 if (!need_domain) {
2221 goto skip_domain;
2222 }
2223
2224 /*
2225 * Check domain type for specific "pre-action".
2226 */
2227 switch (cmd_ctx->lsm->domain.type) {
2228 case LTTNG_DOMAIN_KERNEL:
2229 if (!is_root) {
2230 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2231 goto error;
2232 }
2233
2234 /* Kernel tracer check */
2235 if (kernel_tracer_fd == -1) {
2236 /* Basically, load kernel tracer modules */
2237 ret = init_kernel_tracer();
2238 if (ret != 0) {
2239 goto error;
2240 }
2241 }
2242
2243 /* Consumer is in an ERROR state. Report back to client */
2244 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2245 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2246 goto error;
2247 }
2248
2249 /* Need a session for kernel command */
2250 if (need_tracing_session) {
2251 if (cmd_ctx->session->kernel_session == NULL) {
2252 ret = create_kernel_session(cmd_ctx->session);
2253 if (ret < 0) {
2254 ret = LTTNG_ERR_KERN_SESS_FAIL;
2255 goto error;
2256 }
2257 }
2258
2259 /* Start the kernel consumer daemon */
2260 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2261 if (kconsumer_data.pid == 0 &&
2262 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2263 cmd_ctx->session->start_consumer) {
2264 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2265 ret = start_consumerd(&kconsumer_data);
2266 if (ret < 0) {
2267 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2268 goto error;
2269 }
2270 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2271 } else {
2272 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2273 }
2274
2275 /*
2276 * The consumer was just spawned so we need to add the socket to
2277 * the consumer output of the session if exist.
2278 */
2279 ret = consumer_create_socket(&kconsumer_data,
2280 cmd_ctx->session->kernel_session->consumer);
2281 if (ret < 0) {
2282 goto error;
2283 }
2284 }
2285
2286 break;
2287 case LTTNG_DOMAIN_UST:
2288 {
2289 /* Consumer is in an ERROR state. Report back to client */
2290 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2291 ret = LTTNG_ERR_NO_USTCONSUMERD;
2292 goto error;
2293 }
2294
2295 if (need_tracing_session) {
2296 /* Create UST session if none exist. */
2297 if (cmd_ctx->session->ust_session == NULL) {
2298 ret = create_ust_session(cmd_ctx->session,
2299 &cmd_ctx->lsm->domain);
2300 if (ret != LTTNG_OK) {
2301 goto error;
2302 }
2303 }
2304
2305 /* Start the UST consumer daemons */
2306 /* 64-bit */
2307 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2308 if (consumerd64_bin[0] != '\0' &&
2309 ustconsumer64_data.pid == 0 &&
2310 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2311 cmd_ctx->session->start_consumer) {
2312 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2313 ret = start_consumerd(&ustconsumer64_data);
2314 if (ret < 0) {
2315 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2316 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2317 goto error;
2318 }
2319
2320 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2321 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2322 } else {
2323 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2324 }
2325
2326 /*
2327 * Setup socket for consumer 64 bit. No need for atomic access
2328 * since it was set above and can ONLY be set in this thread.
2329 */
2330 ret = consumer_create_socket(&ustconsumer64_data,
2331 cmd_ctx->session->ust_session->consumer);
2332 if (ret < 0) {
2333 goto error;
2334 }
2335
2336 /* 32-bit */
2337 if (consumerd32_bin[0] != '\0' &&
2338 ustconsumer32_data.pid == 0 &&
2339 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2340 cmd_ctx->session->start_consumer) {
2341 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2342 ret = start_consumerd(&ustconsumer32_data);
2343 if (ret < 0) {
2344 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2345 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2346 goto error;
2347 }
2348
2349 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2350 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2351 } else {
2352 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2353 }
2354
2355 /*
2356 * Setup socket for consumer 64 bit. No need for atomic access
2357 * since it was set above and can ONLY be set in this thread.
2358 */
2359 ret = consumer_create_socket(&ustconsumer32_data,
2360 cmd_ctx->session->ust_session->consumer);
2361 if (ret < 0) {
2362 goto error;
2363 }
2364 }
2365 break;
2366 }
2367 default:
2368 break;
2369 }
2370 skip_domain:
2371
2372 /* Validate consumer daemon state when start/stop trace command */
2373 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2374 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2375 switch (cmd_ctx->lsm->domain.type) {
2376 case LTTNG_DOMAIN_UST:
2377 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2378 ret = LTTNG_ERR_NO_USTCONSUMERD;
2379 goto error;
2380 }
2381 break;
2382 case LTTNG_DOMAIN_KERNEL:
2383 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2384 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2385 goto error;
2386 }
2387 break;
2388 }
2389 }
2390
2391 /*
2392 * Check that the UID or GID match that of the tracing session.
2393 * The root user can interact with all sessions.
2394 */
2395 if (need_tracing_session) {
2396 if (!session_access_ok(cmd_ctx->session,
2397 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2398 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2399 ret = LTTNG_ERR_EPERM;
2400 goto error;
2401 }
2402 }
2403
2404 /* Process by command type */
2405 switch (cmd_ctx->lsm->cmd_type) {
2406 case LTTNG_ADD_CONTEXT:
2407 {
2408 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2409 cmd_ctx->lsm->u.context.channel_name,
2410 cmd_ctx->lsm->u.context.event_name,
2411 &cmd_ctx->lsm->u.context.ctx);
2412 break;
2413 }
2414 case LTTNG_DISABLE_CHANNEL:
2415 {
2416 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2417 cmd_ctx->lsm->u.disable.channel_name);
2418 break;
2419 }
2420 case LTTNG_DISABLE_EVENT:
2421 {
2422 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2423 cmd_ctx->lsm->u.disable.channel_name,
2424 cmd_ctx->lsm->u.disable.name);
2425 break;
2426 }
2427 case LTTNG_DISABLE_ALL_EVENT:
2428 {
2429 DBG("Disabling all events");
2430
2431 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2432 cmd_ctx->lsm->u.disable.channel_name);
2433 break;
2434 }
2435 case LTTNG_DISABLE_CONSUMER:
2436 {
2437 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2438 break;
2439 }
2440 case LTTNG_ENABLE_CHANNEL:
2441 {
2442 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2443 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2444 break;
2445 }
2446 case LTTNG_ENABLE_CONSUMER:
2447 {
2448 /*
2449 * XXX: 0 means that this URI should be applied on the session. Should
2450 * be a DOMAIN enuam.
2451 */
2452 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2453 if (ret != LTTNG_OK) {
2454 goto error;
2455 }
2456
2457 if (cmd_ctx->lsm->domain.type == 0) {
2458 /* Add the URI for the UST session if a consumer is present. */
2459 if (cmd_ctx->session->ust_session &&
2460 cmd_ctx->session->ust_session->consumer) {
2461 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2462 } else if (cmd_ctx->session->kernel_session &&
2463 cmd_ctx->session->kernel_session->consumer) {
2464 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2465 cmd_ctx->session);
2466 }
2467 }
2468 break;
2469 }
2470 case LTTNG_ENABLE_EVENT:
2471 {
2472 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2473 cmd_ctx->lsm->u.enable.channel_name,
2474 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
2475 break;
2476 }
2477 case LTTNG_ENABLE_ALL_EVENT:
2478 {
2479 DBG("Enabling all events");
2480
2481 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2482 cmd_ctx->lsm->u.enable.channel_name,
2483 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
2484 break;
2485 }
2486 case LTTNG_LIST_TRACEPOINTS:
2487 {
2488 struct lttng_event *events;
2489 ssize_t nb_events;
2490
2491 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2492 if (nb_events < 0) {
2493 /* Return value is a negative lttng_error_code. */
2494 ret = -nb_events;
2495 goto error;
2496 }
2497
2498 /*
2499 * Setup lttng message with payload size set to the event list size in
2500 * bytes and then copy list into the llm payload.
2501 */
2502 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2503 if (ret < 0) {
2504 free(events);
2505 goto setup_error;
2506 }
2507
2508 /* Copy event list into message payload */
2509 memcpy(cmd_ctx->llm->payload, events,
2510 sizeof(struct lttng_event) * nb_events);
2511
2512 free(events);
2513
2514 ret = LTTNG_OK;
2515 break;
2516 }
2517 case LTTNG_LIST_TRACEPOINT_FIELDS:
2518 {
2519 struct lttng_event_field *fields;
2520 ssize_t nb_fields;
2521
2522 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2523 &fields);
2524 if (nb_fields < 0) {
2525 /* Return value is a negative lttng_error_code. */
2526 ret = -nb_fields;
2527 goto error;
2528 }
2529
2530 /*
2531 * Setup lttng message with payload size set to the event list size in
2532 * bytes and then copy list into the llm payload.
2533 */
2534 ret = setup_lttng_msg(cmd_ctx,
2535 sizeof(struct lttng_event_field) * nb_fields);
2536 if (ret < 0) {
2537 free(fields);
2538 goto setup_error;
2539 }
2540
2541 /* Copy event list into message payload */
2542 memcpy(cmd_ctx->llm->payload, fields,
2543 sizeof(struct lttng_event_field) * nb_fields);
2544
2545 free(fields);
2546
2547 ret = LTTNG_OK;
2548 break;
2549 }
2550 case LTTNG_SET_CONSUMER_URI:
2551 {
2552 size_t nb_uri, len;
2553 struct lttng_uri *uris;
2554
2555 nb_uri = cmd_ctx->lsm->u.uri.size;
2556 len = nb_uri * sizeof(struct lttng_uri);
2557
2558 if (nb_uri == 0) {
2559 ret = LTTNG_ERR_INVALID;
2560 goto error;
2561 }
2562
2563 uris = zmalloc(len);
2564 if (uris == NULL) {
2565 ret = LTTNG_ERR_FATAL;
2566 goto error;
2567 }
2568
2569 /* Receive variable len data */
2570 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2571 ret = lttcomm_recv_unix_sock(sock, uris, len);
2572 if (ret <= 0) {
2573 DBG("No URIs received from client... continuing");
2574 *sock_error = 1;
2575 ret = LTTNG_ERR_SESSION_FAIL;
2576 free(uris);
2577 goto error;
2578 }
2579
2580 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2581 nb_uri, uris);
2582 if (ret != LTTNG_OK) {
2583 free(uris);
2584 goto error;
2585 }
2586
2587 /*
2588 * XXX: 0 means that this URI should be applied on the session. Should
2589 * be a DOMAIN enuam.
2590 */
2591 if (cmd_ctx->lsm->domain.type == 0) {
2592 /* Add the URI for the UST session if a consumer is present. */
2593 if (cmd_ctx->session->ust_session &&
2594 cmd_ctx->session->ust_session->consumer) {
2595 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2596 nb_uri, uris);
2597 } else if (cmd_ctx->session->kernel_session &&
2598 cmd_ctx->session->kernel_session->consumer) {
2599 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2600 cmd_ctx->session, nb_uri, uris);
2601 }
2602 }
2603
2604 free(uris);
2605
2606 break;
2607 }
2608 case LTTNG_START_TRACE:
2609 {
2610 ret = cmd_start_trace(cmd_ctx->session);
2611 break;
2612 }
2613 case LTTNG_STOP_TRACE:
2614 {
2615 ret = cmd_stop_trace(cmd_ctx->session);
2616 break;
2617 }
2618 case LTTNG_CREATE_SESSION:
2619 {
2620 size_t nb_uri, len;
2621 struct lttng_uri *uris = NULL;
2622
2623 nb_uri = cmd_ctx->lsm->u.uri.size;
2624 len = nb_uri * sizeof(struct lttng_uri);
2625
2626 if (nb_uri > 0) {
2627 uris = zmalloc(len);
2628 if (uris == NULL) {
2629 ret = LTTNG_ERR_FATAL;
2630 goto error;
2631 }
2632
2633 /* Receive variable len data */
2634 DBG("Waiting for %zu URIs from client ...", nb_uri);
2635 ret = lttcomm_recv_unix_sock(sock, uris, len);
2636 if (ret <= 0) {
2637 DBG("No URIs received from client... continuing");
2638 *sock_error = 1;
2639 ret = LTTNG_ERR_SESSION_FAIL;
2640 free(uris);
2641 goto error;
2642 }
2643
2644 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2645 DBG("Creating session with ONE network URI is a bad call");
2646 ret = LTTNG_ERR_SESSION_FAIL;
2647 free(uris);
2648 goto error;
2649 }
2650 }
2651
2652 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2653 &cmd_ctx->creds);
2654
2655 free(uris);
2656
2657 break;
2658 }
2659 case LTTNG_DESTROY_SESSION:
2660 {
2661 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2662
2663 /* Set session to NULL so we do not unlock it after free. */
2664 cmd_ctx->session = NULL;
2665 break;
2666 }
2667 case LTTNG_LIST_DOMAINS:
2668 {
2669 ssize_t nb_dom;
2670 struct lttng_domain *domains;
2671
2672 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2673 if (nb_dom < 0) {
2674 /* Return value is a negative lttng_error_code. */
2675 ret = -nb_dom;
2676 goto error;
2677 }
2678
2679 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2680 if (ret < 0) {
2681 goto setup_error;
2682 }
2683
2684 /* Copy event list into message payload */
2685 memcpy(cmd_ctx->llm->payload, domains,
2686 nb_dom * sizeof(struct lttng_domain));
2687
2688 free(domains);
2689
2690 ret = LTTNG_OK;
2691 break;
2692 }
2693 case LTTNG_LIST_CHANNELS:
2694 {
2695 int nb_chan;
2696 struct lttng_channel *channels;
2697
2698 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2699 cmd_ctx->session, &channels);
2700 if (nb_chan < 0) {
2701 /* Return value is a negative lttng_error_code. */
2702 ret = -nb_chan;
2703 goto error;
2704 }
2705
2706 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2707 if (ret < 0) {
2708 goto setup_error;
2709 }
2710
2711 /* Copy event list into message payload */
2712 memcpy(cmd_ctx->llm->payload, channels,
2713 nb_chan * sizeof(struct lttng_channel));
2714
2715 free(channels);
2716
2717 ret = LTTNG_OK;
2718 break;
2719 }
2720 case LTTNG_LIST_EVENTS:
2721 {
2722 ssize_t nb_event;
2723 struct lttng_event *events = NULL;
2724
2725 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2726 cmd_ctx->lsm->u.list.channel_name, &events);
2727 if (nb_event < 0) {
2728 /* Return value is a negative lttng_error_code. */
2729 ret = -nb_event;
2730 goto error;
2731 }
2732
2733 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2734 if (ret < 0) {
2735 goto setup_error;
2736 }
2737
2738 /* Copy event list into message payload */
2739 memcpy(cmd_ctx->llm->payload, events,
2740 nb_event * sizeof(struct lttng_event));
2741
2742 free(events);
2743
2744 ret = LTTNG_OK;
2745 break;
2746 }
2747 case LTTNG_LIST_SESSIONS:
2748 {
2749 unsigned int nr_sessions;
2750
2751 session_lock_list();
2752 nr_sessions = lttng_sessions_count(
2753 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2754 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2755
2756 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2757 if (ret < 0) {
2758 session_unlock_list();
2759 goto setup_error;
2760 }
2761
2762 /* Filled the session array */
2763 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2764 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2765 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2766
2767 session_unlock_list();
2768
2769 ret = LTTNG_OK;
2770 break;
2771 }
2772 case LTTNG_CALIBRATE:
2773 {
2774 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2775 &cmd_ctx->lsm->u.calibrate);
2776 break;
2777 }
2778 case LTTNG_REGISTER_CONSUMER:
2779 {
2780 struct consumer_data *cdata;
2781
2782 switch (cmd_ctx->lsm->domain.type) {
2783 case LTTNG_DOMAIN_KERNEL:
2784 cdata = &kconsumer_data;
2785 break;
2786 default:
2787 ret = LTTNG_ERR_UND;
2788 goto error;
2789 }
2790
2791 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2792 cmd_ctx->lsm->u.reg.path, cdata);
2793 break;
2794 }
2795 case LTTNG_SET_FILTER:
2796 {
2797 struct lttng_filter_bytecode *bytecode;
2798
2799 if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2800 ret = LTTNG_ERR_FILTER_INVAL;
2801 goto error;
2802 }
2803 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2804 if (!bytecode) {
2805 ret = LTTNG_ERR_FILTER_NOMEM;
2806 goto error;
2807 }
2808 /* Receive var. len. data */
2809 DBG("Receiving var len data from client ...");
2810 ret = lttcomm_recv_unix_sock(sock, bytecode,
2811 cmd_ctx->lsm->u.filter.bytecode_len);
2812 if (ret <= 0) {
2813 DBG("Nothing recv() from client var len data... continuing");
2814 *sock_error = 1;
2815 ret = LTTNG_ERR_FILTER_INVAL;
2816 goto error;
2817 }
2818
2819 if (bytecode->len + sizeof(*bytecode)
2820 != cmd_ctx->lsm->u.filter.bytecode_len) {
2821 free(bytecode);
2822 ret = LTTNG_ERR_FILTER_INVAL;
2823 goto error;
2824 }
2825
2826 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2827 cmd_ctx->lsm->u.filter.channel_name,
2828 cmd_ctx->lsm->u.filter.event_name,
2829 bytecode);
2830 break;
2831 }
2832 case LTTNG_DATA_AVAILABLE:
2833 {
2834 ret = cmd_data_available(cmd_ctx->session);
2835 break;
2836 }
2837 default:
2838 ret = LTTNG_ERR_UND;
2839 break;
2840 }
2841
2842 error:
2843 if (cmd_ctx->llm == NULL) {
2844 DBG("Missing llm structure. Allocating one.");
2845 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2846 goto setup_error;
2847 }
2848 }
2849 /* Set return code */
2850 cmd_ctx->llm->ret_code = ret;
2851 setup_error:
2852 if (cmd_ctx->session) {
2853 session_unlock(cmd_ctx->session);
2854 }
2855 if (need_tracing_session) {
2856 session_unlock_list();
2857 }
2858 init_setup_error:
2859 return ret;
2860 }
2861
2862 /*
2863 * Thread managing health check socket.
2864 */
2865 static void *thread_manage_health(void *data)
2866 {
2867 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2868 uint32_t revents, nb_fd;
2869 struct lttng_poll_event events;
2870 struct lttcomm_health_msg msg;
2871 struct lttcomm_health_data reply;
2872
2873 DBG("[thread] Manage health check started");
2874
2875 rcu_register_thread();
2876
2877 /* Create unix socket */
2878 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2879 if (sock < 0) {
2880 ERR("Unable to create health check Unix socket");
2881 ret = -1;
2882 goto error;
2883 }
2884
2885 ret = lttcomm_listen_unix_sock(sock);
2886 if (ret < 0) {
2887 goto error;
2888 }
2889
2890 /*
2891 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2892 * more will be added to this poll set.
2893 */
2894 ret = create_thread_poll_set(&events, 2);
2895 if (ret < 0) {
2896 goto error;
2897 }
2898
2899 /* Add the application registration socket */
2900 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2901 if (ret < 0) {
2902 goto error;
2903 }
2904
2905 while (1) {
2906 DBG("Health check ready");
2907
2908 nb_fd = LTTNG_POLL_GETNB(&events);
2909
2910 /* Inifinite blocking call, waiting for transmission */
2911 restart:
2912 ret = lttng_poll_wait(&events, -1);
2913 if (ret < 0) {
2914 /*
2915 * Restart interrupted system call.
2916 */
2917 if (errno == EINTR) {
2918 goto restart;
2919 }
2920 goto error;
2921 }
2922
2923 for (i = 0; i < nb_fd; i++) {
2924 /* Fetch once the poll data */
2925 revents = LTTNG_POLL_GETEV(&events, i);
2926 pollfd = LTTNG_POLL_GETFD(&events, i);
2927
2928 /* Thread quit pipe has been closed. Killing thread. */
2929 ret = check_thread_quit_pipe(pollfd, revents);
2930 if (ret) {
2931 err = 0;
2932 goto exit;
2933 }
2934
2935 /* Event on the registration socket */
2936 if (pollfd == sock) {
2937 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2938 ERR("Health socket poll error");
2939 goto error;
2940 }
2941 }
2942 }
2943
2944 new_sock = lttcomm_accept_unix_sock(sock);
2945 if (new_sock < 0) {
2946 goto error;
2947 }
2948
2949 DBG("Receiving data from client for health...");
2950 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
2951 if (ret <= 0) {
2952 DBG("Nothing recv() from client... continuing");
2953 ret = close(new_sock);
2954 if (ret) {
2955 PERROR("close");
2956 }
2957 new_sock = -1;
2958 continue;
2959 }
2960
2961 rcu_thread_online();
2962
2963 switch (msg.component) {
2964 case LTTNG_HEALTH_CMD:
2965 reply.ret_code = health_check_state(&health_thread_cmd);
2966 break;
2967 case LTTNG_HEALTH_APP_MANAGE:
2968 reply.ret_code = health_check_state(&health_thread_app_manage);
2969 break;
2970 case LTTNG_HEALTH_APP_REG:
2971 reply.ret_code = health_check_state(&health_thread_app_reg);
2972 break;
2973 case LTTNG_HEALTH_KERNEL:
2974 reply.ret_code = health_check_state(&health_thread_kernel);
2975 break;
2976 case LTTNG_HEALTH_CONSUMER:
2977 reply.ret_code = check_consumer_health();
2978 break;
2979 case LTTNG_HEALTH_ALL:
2980 reply.ret_code =
2981 health_check_state(&health_thread_app_manage) &&
2982 health_check_state(&health_thread_app_reg) &&
2983 health_check_state(&health_thread_cmd) &&
2984 health_check_state(&health_thread_kernel) &&
2985 check_consumer_health();
2986 break;
2987 default:
2988 reply.ret_code = LTTNG_ERR_UND;
2989 break;
2990 }
2991
2992 /*
2993 * Flip ret value since 0 is a success and 1 indicates a bad health for
2994 * the client where in the sessiond it is the opposite. Again, this is
2995 * just to make things easier for us poor developer which enjoy a lot
2996 * lazyness.
2997 */
2998 if (reply.ret_code == 0 || reply.ret_code == 1) {
2999 reply.ret_code = !reply.ret_code;
3000 }
3001
3002 DBG2("Health check return value %d", reply.ret_code);
3003
3004 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3005 if (ret < 0) {
3006 ERR("Failed to send health data back to client");
3007 }
3008
3009 /* End of transmission */
3010 ret = close(new_sock);
3011 if (ret) {
3012 PERROR("close");
3013 }
3014 new_sock = -1;
3015 }
3016
3017 exit:
3018 error:
3019 if (err) {
3020 ERR("Health error occurred in %s", __func__);
3021 }
3022 DBG("Health check thread dying");
3023 unlink(health_unix_sock_path);
3024 if (sock >= 0) {
3025 ret = close(sock);
3026 if (ret) {
3027 PERROR("close");
3028 }
3029 }
3030 if (new_sock >= 0) {
3031 ret = close(new_sock);
3032 if (ret) {
3033 PERROR("close");
3034 }
3035 }
3036
3037 lttng_poll_clean(&events);
3038
3039 rcu_unregister_thread();
3040 return NULL;
3041 }
3042
3043 /*
3044 * This thread manage all clients request using the unix client socket for
3045 * communication.
3046 */
3047 static void *thread_manage_clients(void *data)
3048 {
3049 int sock = -1, ret, i, pollfd, err = -1;
3050 int sock_error;
3051 uint32_t revents, nb_fd;
3052 struct command_ctx *cmd_ctx = NULL;
3053 struct lttng_poll_event events;
3054
3055 DBG("[thread] Manage client started");
3056
3057 testpoint(thread_manage_clients);
3058
3059 rcu_register_thread();
3060
3061 health_code_update(&health_thread_cmd);
3062
3063 ret = lttcomm_listen_unix_sock(client_sock);
3064 if (ret < 0) {
3065 goto error;
3066 }
3067
3068 /*
3069 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3070 * more will be added to this poll set.
3071 */
3072 ret = create_thread_poll_set(&events, 2);
3073 if (ret < 0) {
3074 goto error;
3075 }
3076
3077 /* Add the application registration socket */
3078 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3079 if (ret < 0) {
3080 goto error;
3081 }
3082
3083 /*
3084 * Notify parent pid that we are ready to accept command for client side.
3085 */
3086 if (opt_sig_parent) {
3087 kill(ppid, SIGUSR1);
3088 }
3089
3090 testpoint(thread_manage_clients_before_loop);
3091
3092 health_code_update(&health_thread_cmd);
3093
3094 while (1) {
3095 DBG("Accepting client command ...");
3096
3097 nb_fd = LTTNG_POLL_GETNB(&events);
3098
3099 /* Inifinite blocking call, waiting for transmission */
3100 restart:
3101 health_poll_update(&health_thread_cmd);
3102 ret = lttng_poll_wait(&events, -1);
3103 health_poll_update(&health_thread_cmd);
3104 if (ret < 0) {
3105 /*
3106 * Restart interrupted system call.
3107 */
3108 if (errno == EINTR) {
3109 goto restart;
3110 }
3111 goto error;
3112 }
3113
3114 for (i = 0; i < nb_fd; i++) {
3115 /* Fetch once the poll data */
3116 revents = LTTNG_POLL_GETEV(&events, i);
3117 pollfd = LTTNG_POLL_GETFD(&events, i);
3118
3119 health_code_update(&health_thread_cmd);
3120
3121 /* Thread quit pipe has been closed. Killing thread. */
3122 ret = check_thread_quit_pipe(pollfd, revents);
3123 if (ret) {
3124 err = 0;
3125 goto exit;
3126 }
3127
3128 /* Event on the registration socket */
3129 if (pollfd == client_sock) {
3130 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3131 ERR("Client socket poll error");
3132 goto error;
3133 }
3134 }
3135 }
3136
3137 DBG("Wait for client response");
3138
3139 health_code_update(&health_thread_cmd);
3140
3141 sock = lttcomm_accept_unix_sock(client_sock);
3142 if (sock < 0) {
3143 goto error;
3144 }
3145
3146 /* Set socket option for credentials retrieval */
3147 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3148 if (ret < 0) {
3149 goto error;
3150 }
3151
3152 /* Allocate context command to process the client request */
3153 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3154 if (cmd_ctx == NULL) {
3155 PERROR("zmalloc cmd_ctx");
3156 goto error;
3157 }
3158
3159 /* Allocate data buffer for reception */
3160 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3161 if (cmd_ctx->lsm == NULL) {
3162 PERROR("zmalloc cmd_ctx->lsm");
3163 goto error;
3164 }
3165
3166 cmd_ctx->llm = NULL;
3167 cmd_ctx->session = NULL;
3168
3169 health_code_update(&health_thread_cmd);
3170
3171 /*
3172 * Data is received from the lttng client. The struct
3173 * lttcomm_session_msg (lsm) contains the command and data request of
3174 * the client.
3175 */
3176 DBG("Receiving data from client ...");
3177 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3178 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3179 if (ret <= 0) {
3180 DBG("Nothing recv() from client... continuing");
3181 ret = close(sock);
3182 if (ret) {
3183 PERROR("close");
3184 }
3185 sock = -1;
3186 clean_command_ctx(&cmd_ctx);
3187 continue;
3188 }
3189
3190 health_code_update(&health_thread_cmd);
3191
3192 // TODO: Validate cmd_ctx including sanity check for
3193 // security purpose.
3194
3195 rcu_thread_online();
3196 /*
3197 * This function dispatch the work to the kernel or userspace tracer
3198 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3199 * informations for the client. The command context struct contains
3200 * everything this function may needs.
3201 */
3202 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3203 rcu_thread_offline();
3204 if (ret < 0) {
3205 if (sock_error) {
3206 ret = close(sock);
3207 if (ret) {
3208 PERROR("close");
3209 }
3210 sock = -1;
3211 }
3212 /*
3213 * TODO: Inform client somehow of the fatal error. At
3214 * this point, ret < 0 means that a zmalloc failed
3215 * (ENOMEM). Error detected but still accept
3216 * command, unless a socket error has been
3217 * detected.
3218 */
3219 clean_command_ctx(&cmd_ctx);
3220 continue;
3221 }
3222
3223 health_code_update(&health_thread_cmd);
3224
3225 DBG("Sending response (size: %d, retcode: %s)",
3226 cmd_ctx->lttng_msg_size,
3227 lttng_strerror(-cmd_ctx->llm->ret_code));
3228 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3229 if (ret < 0) {
3230 ERR("Failed to send data back to client&