Fix: remove consumer health poll update on startup
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <urcu/uatomic.h>
36 #include <unistd.h>
37 #include <config.h>
38
39 #include <common/common.h>
40 #include <common/compat/poll.h>
41 #include <common/compat/socket.h>
42 #include <common/defaults.h>
43 #include <common/kernel-consumer/kernel-consumer.h>
44 #include <common/futex.h>
45 #include <common/relayd/relayd.h>
46 #include <common/utils.h>
47
48 #include "lttng-sessiond.h"
49 #include "channel.h"
50 #include "cmd.h"
51 #include "consumer.h"
52 #include "context.h"
53 #include "event.h"
54 #include "kernel.h"
55 #include "kernel-consumer.h"
56 #include "modprobe.h"
57 #include "shm.h"
58 #include "ust-ctl.h"
59 #include "ust-consumer.h"
60 #include "utils.h"
61 #include "fd-limit.h"
62 #include "health.h"
63 #include "testpoint.h"
64
65 #define CONSUMERD_FILE "lttng-consumerd"
66
67 /* Const values */
68 const char default_home_dir[] = DEFAULT_HOME_DIR;
69 const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
70 const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
71 const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
72
73 const char *progname;
74 const char *opt_tracing_group;
75 static const char *opt_pidfile;
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 /*
226 * Socket timeout for receiving and sending in seconds.
227 */
228 static int app_socket_timeout;
229
230 static
231 void setup_consumerd_path(void)
232 {
233 const char *bin, *libdir;
234
235 /*
236 * Allow INSTALL_BIN_PATH to be used as a target path for the
237 * native architecture size consumer if CONFIG_CONSUMER*_PATH
238 * has not been defined.
239 */
240 #if (CAA_BITS_PER_LONG == 32)
241 if (!consumerd32_bin[0]) {
242 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
243 }
244 if (!consumerd32_libdir[0]) {
245 consumerd32_libdir = INSTALL_LIB_PATH;
246 }
247 #elif (CAA_BITS_PER_LONG == 64)
248 if (!consumerd64_bin[0]) {
249 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
250 }
251 if (!consumerd64_libdir[0]) {
252 consumerd64_libdir = INSTALL_LIB_PATH;
253 }
254 #else
255 #error "Unknown bitness"
256 #endif
257
258 /*
259 * runtime env. var. overrides the build default.
260 */
261 bin = getenv("LTTNG_CONSUMERD32_BIN");
262 if (bin) {
263 consumerd32_bin = bin;
264 }
265 bin = getenv("LTTNG_CONSUMERD64_BIN");
266 if (bin) {
267 consumerd64_bin = bin;
268 }
269 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
270 if (libdir) {
271 consumerd32_libdir = libdir;
272 }
273 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
274 if (libdir) {
275 consumerd64_libdir = libdir;
276 }
277 }
278
279 /*
280 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
281 */
282 static int create_thread_poll_set(struct lttng_poll_event *events,
283 unsigned int size)
284 {
285 int ret;
286
287 if (events == NULL || size == 0) {
288 ret = -1;
289 goto error;
290 }
291
292 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
293 if (ret < 0) {
294 goto error;
295 }
296
297 /* Add quit pipe */
298 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
299 if (ret < 0) {
300 goto error;
301 }
302
303 return 0;
304
305 error:
306 return ret;
307 }
308
309 /*
310 * Check if the thread quit pipe was triggered.
311 *
312 * Return 1 if it was triggered else 0;
313 */
314 static int check_thread_quit_pipe(int fd, uint32_t events)
315 {
316 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
317 return 1;
318 }
319
320 return 0;
321 }
322
323 /*
324 * Return group ID of the tracing group or -1 if not found.
325 */
326 static gid_t allowed_group(void)
327 {
328 struct group *grp;
329
330 if (opt_tracing_group) {
331 grp = getgrnam(opt_tracing_group);
332 } else {
333 grp = getgrnam(default_tracing_group);
334 }
335 if (!grp) {
336 return -1;
337 } else {
338 return grp->gr_gid;
339 }
340 }
341
342 /*
343 * Init thread quit pipe.
344 *
345 * Return -1 on error or 0 if all pipes are created.
346 */
347 static int init_thread_quit_pipe(void)
348 {
349 int ret, i;
350
351 ret = pipe(thread_quit_pipe);
352 if (ret < 0) {
353 PERROR("thread quit pipe");
354 goto error;
355 }
356
357 for (i = 0; i < 2; i++) {
358 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
359 if (ret < 0) {
360 PERROR("fcntl");
361 goto error;
362 }
363 }
364
365 error:
366 return ret;
367 }
368
369 /*
370 * Stop all threads by closing the thread quit pipe.
371 */
372 static void stop_threads(void)
373 {
374 int ret;
375
376 /* Stopping all threads */
377 DBG("Terminating all threads");
378 ret = notify_thread_pipe(thread_quit_pipe[1]);
379 if (ret < 0) {
380 ERR("write error on thread quit pipe");
381 }
382
383 /* Dispatch thread */
384 CMM_STORE_SHARED(dispatch_thread_exit, 1);
385 futex_nto1_wake(&ust_cmd_queue.futex);
386 }
387
388 /*
389 * Cleanup the daemon
390 */
391 static void cleanup(void)
392 {
393 int ret;
394 char *cmd = NULL;
395 struct ltt_session *sess, *stmp;
396
397 DBG("Cleaning up");
398
399 /* First thing first, stop all threads */
400 utils_close_pipe(thread_quit_pipe);
401
402 /*
403 * If opt_pidfile is undefined, the default file will be wiped when
404 * removing the rundir.
405 */
406 if (opt_pidfile) {
407 ret = remove(opt_pidfile);
408 if (ret < 0) {
409 PERROR("remove pidfile %s", opt_pidfile);
410 }
411 }
412
413 DBG("Removing %s directory", rundir);
414 ret = asprintf(&cmd, "rm -rf %s", rundir);
415 if (ret < 0) {
416 ERR("asprintf failed. Something is really wrong!");
417 }
418
419 /* Remove lttng run directory */
420 ret = system(cmd);
421 if (ret < 0) {
422 ERR("Unable to clean %s", rundir);
423 }
424 free(cmd);
425 free(rundir);
426
427 DBG("Cleaning up all sessions");
428
429 /* Destroy session list mutex */
430 if (session_list_ptr != NULL) {
431 pthread_mutex_destroy(&session_list_ptr->lock);
432
433 /* Cleanup ALL session */
434 cds_list_for_each_entry_safe(sess, stmp,
435 &session_list_ptr->head, list) {
436 cmd_destroy_session(sess, kernel_poll_pipe[1]);
437 }
438 }
439
440 DBG("Closing all UST sockets");
441 ust_app_clean_list();
442
443 if (is_root && !opt_no_kernel) {
444 DBG2("Closing kernel fd");
445 if (kernel_tracer_fd >= 0) {
446 ret = close(kernel_tracer_fd);
447 if (ret) {
448 PERROR("close");
449 }
450 }
451 DBG("Unloading kernel modules");
452 modprobe_remove_lttng_all();
453 }
454
455 /* <fun> */
456 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
457 "Matthew, BEET driven development works!%c[%dm",
458 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
459 /* </fun> */
460 }
461
462 /*
463 * Send data on a unix socket using the liblttsessiondcomm API.
464 *
465 * Return lttcomm error code.
466 */
467 static int send_unix_sock(int sock, void *buf, size_t len)
468 {
469 /* Check valid length */
470 if (len == 0) {
471 return -1;
472 }
473
474 return lttcomm_send_unix_sock(sock, buf, len);
475 }
476
477 /*
478 * Free memory of a command context structure.
479 */
480 static void clean_command_ctx(struct command_ctx **cmd_ctx)
481 {
482 DBG("Clean command context structure");
483 if (*cmd_ctx) {
484 if ((*cmd_ctx)->llm) {
485 free((*cmd_ctx)->llm);
486 }
487 if ((*cmd_ctx)->lsm) {
488 free((*cmd_ctx)->lsm);
489 }
490 free(*cmd_ctx);
491 *cmd_ctx = NULL;
492 }
493 }
494
495 /*
496 * Notify UST applications using the shm mmap futex.
497 */
498 static int notify_ust_apps(int active)
499 {
500 char *wait_shm_mmap;
501
502 DBG("Notifying applications of session daemon state: %d", active);
503
504 /* See shm.c for this call implying mmap, shm and futex calls */
505 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
506 if (wait_shm_mmap == NULL) {
507 goto error;
508 }
509
510 /* Wake waiting process */
511 futex_wait_update((int32_t *) wait_shm_mmap, active);
512
513 /* Apps notified successfully */
514 return 0;
515
516 error:
517 return -1;
518 }
519
520 /*
521 * Setup the outgoing data buffer for the response (llm) by allocating the
522 * right amount of memory and copying the original information from the lsm
523 * structure.
524 *
525 * Return total size of the buffer pointed by buf.
526 */
527 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
528 {
529 int ret, buf_size;
530
531 buf_size = size;
532
533 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
534 if (cmd_ctx->llm == NULL) {
535 PERROR("zmalloc");
536 ret = -ENOMEM;
537 goto error;
538 }
539
540 /* Copy common data */
541 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
542 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
543
544 cmd_ctx->llm->data_size = size;
545 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
546
547 return buf_size;
548
549 error:
550 return ret;
551 }
552
553 /*
554 * Update the kernel poll set of all channel fd available over all tracing
555 * session. Add the wakeup pipe at the end of the set.
556 */
557 static int update_kernel_poll(struct lttng_poll_event *events)
558 {
559 int ret;
560 struct ltt_session *session;
561 struct ltt_kernel_channel *channel;
562
563 DBG("Updating kernel poll set");
564
565 session_lock_list();
566 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
567 session_lock(session);
568 if (session->kernel_session == NULL) {
569 session_unlock(session);
570 continue;
571 }
572
573 cds_list_for_each_entry(channel,
574 &session->kernel_session->channel_list.head, list) {
575 /* Add channel fd to the kernel poll set */
576 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
577 if (ret < 0) {
578 session_unlock(session);
579 goto error;
580 }
581 DBG("Channel fd %d added to kernel set", channel->fd);
582 }
583 session_unlock(session);
584 }
585 session_unlock_list();
586
587 return 0;
588
589 error:
590 session_unlock_list();
591 return -1;
592 }
593
594 /*
595 * Find the channel fd from 'fd' over all tracing session. When found, check
596 * for new channel stream and send those stream fds to the kernel consumer.
597 *
598 * Useful for CPU hotplug feature.
599 */
600 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
601 {
602 int ret = 0;
603 struct ltt_session *session;
604 struct ltt_kernel_session *ksess;
605 struct ltt_kernel_channel *channel;
606
607 DBG("Updating kernel streams for channel fd %d", fd);
608
609 session_lock_list();
610 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
611 session_lock(session);
612 if (session->kernel_session == NULL) {
613 session_unlock(session);
614 continue;
615 }
616 ksess = session->kernel_session;
617
618 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
619 if (channel->fd == fd) {
620 DBG("Channel found, updating kernel streams");
621 ret = kernel_open_channel_stream(channel);
622 if (ret < 0) {
623 goto error;
624 }
625
626 /*
627 * Have we already sent fds to the consumer? If yes, it means
628 * that tracing is started so it is safe to send our updated
629 * stream fds.
630 */
631 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
632 struct lttng_ht_iter iter;
633 struct consumer_socket *socket;
634
635 rcu_read_lock();
636 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
637 &iter.iter, socket, node.node) {
638 /* Code flow error */
639 assert(socket->fd >= 0);
640
641 pthread_mutex_lock(socket->lock);
642 ret = kernel_consumer_send_channel_stream(socket,
643 channel, ksess);
644 pthread_mutex_unlock(socket->lock);
645 if (ret < 0) {
646 rcu_read_unlock();
647 goto error;
648 }
649 }
650 rcu_read_unlock();
651 }
652 goto error;
653 }
654 }
655 session_unlock(session);
656 }
657 session_unlock_list();
658 return ret;
659
660 error:
661 session_unlock(session);
662 session_unlock_list();
663 return ret;
664 }
665
666 /*
667 * For each tracing session, update newly registered apps.
668 */
669 static void update_ust_app(int app_sock)
670 {
671 struct ltt_session *sess, *stmp;
672
673 session_lock_list();
674
675 /* For all tracing session(s) */
676 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
677 session_lock(sess);
678 if (sess->ust_session) {
679 ust_app_global_update(sess->ust_session, app_sock);
680 }
681 session_unlock(sess);
682 }
683
684 session_unlock_list();
685 }
686
687 /*
688 * This thread manage event coming from the kernel.
689 *
690 * Features supported in this thread:
691 * -) CPU Hotplug
692 */
693 static void *thread_manage_kernel(void *data)
694 {
695 int ret, i, pollfd, update_poll_flag = 1, err = -1;
696 uint32_t revents, nb_fd;
697 char tmp;
698 struct lttng_poll_event events;
699
700 DBG("[thread] Thread manage kernel started");
701
702 health_register(HEALTH_TYPE_KERNEL);
703
704 /*
705 * This first step of the while is to clean this structure which could free
706 * non NULL pointers so zero it before the loop.
707 */
708 memset(&events, 0, sizeof(events));
709
710 if (testpoint(thread_manage_kernel)) {
711 goto error_testpoint;
712 }
713
714 health_code_update();
715
716 if (testpoint(thread_manage_kernel_before_loop)) {
717 goto error_testpoint;
718 }
719
720 while (1) {
721 health_code_update();
722
723 if (update_poll_flag == 1) {
724 /* Clean events object. We are about to populate it again. */
725 lttng_poll_clean(&events);
726
727 ret = create_thread_poll_set(&events, 2);
728 if (ret < 0) {
729 goto error_poll_create;
730 }
731
732 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
733 if (ret < 0) {
734 goto error;
735 }
736
737 /* This will add the available kernel channel if any. */
738 ret = update_kernel_poll(&events);
739 if (ret < 0) {
740 goto error;
741 }
742 update_poll_flag = 0;
743 }
744
745 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
746
747 /* Poll infinite value of time */
748 restart:
749 health_poll_entry();
750 ret = lttng_poll_wait(&events, -1);
751 health_poll_exit();
752 if (ret < 0) {
753 /*
754 * Restart interrupted system call.
755 */
756 if (errno == EINTR) {
757 goto restart;
758 }
759 goto error;
760 } else if (ret == 0) {
761 /* Should not happen since timeout is infinite */
762 ERR("Return value of poll is 0 with an infinite timeout.\n"
763 "This should not have happened! Continuing...");
764 continue;
765 }
766
767 nb_fd = ret;
768
769 for (i = 0; i < nb_fd; i++) {
770 /* Fetch once the poll data */
771 revents = LTTNG_POLL_GETEV(&events, i);
772 pollfd = LTTNG_POLL_GETFD(&events, i);
773
774 health_code_update();
775
776 /* Thread quit pipe has been closed. Killing thread. */
777 ret = check_thread_quit_pipe(pollfd, revents);
778 if (ret) {
779 err = 0;
780 goto exit;
781 }
782
783 /* Check for data on kernel pipe */
784 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
785 do {
786 ret = read(kernel_poll_pipe[0], &tmp, 1);
787 } while (ret < 0 && errno == EINTR);
788 /*
789 * Ret value is useless here, if this pipe gets any actions an
790 * update is required anyway.
791 */
792 update_poll_flag = 1;
793 continue;
794 } else {
795 /*
796 * New CPU detected by the kernel. Adding kernel stream to
797 * kernel session and updating the kernel consumer
798 */
799 if (revents & LPOLLIN) {
800 ret = update_kernel_stream(&kconsumer_data, pollfd);
801 if (ret < 0) {
802 continue;
803 }
804 break;
805 /*
806 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
807 * and unregister kernel stream at this point.
808 */
809 }
810 }
811 }
812 }
813
814 exit:
815 error:
816 lttng_poll_clean(&events);
817 error_poll_create:
818 error_testpoint:
819 utils_close_pipe(kernel_poll_pipe);
820 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
821 if (err) {
822 health_error();
823 ERR("Health error occurred in %s", __func__);
824 WARN("Kernel thread died unexpectedly. "
825 "Kernel tracing can continue but CPU hotplug is disabled.");
826 }
827 health_unregister();
828 DBG("Kernel thread dying");
829 return NULL;
830 }
831
832 /*
833 * Signal pthread condition of the consumer data that the thread.
834 */
835 static void signal_consumer_condition(struct consumer_data *data, int state)
836 {
837 pthread_mutex_lock(&data->cond_mutex);
838
839 /*
840 * The state is set before signaling. It can be any value, it's the waiter
841 * job to correctly interpret this condition variable associated to the
842 * consumer pthread_cond.
843 *
844 * A value of 0 means that the corresponding thread of the consumer data
845 * was not started. 1 indicates that the thread has started and is ready
846 * for action. A negative value means that there was an error during the
847 * thread bootstrap.
848 */
849 data->consumer_thread_is_ready = state;
850 (void) pthread_cond_signal(&data->cond);
851
852 pthread_mutex_unlock(&data->cond_mutex);
853 }
854
855 /*
856 * This thread manage the consumer error sent back to the session daemon.
857 */
858 static void *thread_manage_consumer(void *data)
859 {
860 int sock = -1, i, ret, pollfd, err = -1;
861 uint32_t revents, nb_fd;
862 enum lttcomm_return_code code;
863 struct lttng_poll_event events;
864 struct consumer_data *consumer_data = data;
865
866 DBG("[thread] Manage consumer started");
867
868 health_register(HEALTH_TYPE_CONSUMER);
869
870 health_code_update();
871
872 /*
873 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
874 * Nothing more will be added to this poll set.
875 */
876 ret = create_thread_poll_set(&events, 2);
877 if (ret < 0) {
878 goto error_poll;
879 }
880
881 /*
882 * The error socket here is already in a listening state which was done
883 * just before spawning this thread to avoid a race between the consumer
884 * daemon exec trying to connect and the listen() call.
885 */
886 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
887 if (ret < 0) {
888 goto error;
889 }
890
891 health_code_update();
892
893 /* Inifinite blocking call, waiting for transmission */
894 restart:
895 health_poll_entry();
896
897 if (testpoint(thread_manage_consumer)) {
898 goto error;
899 }
900
901 ret = lttng_poll_wait(&events, -1);
902 health_poll_exit();
903 if (ret < 0) {
904 /*
905 * Restart interrupted system call.
906 */
907 if (errno == EINTR) {
908 goto restart;
909 }
910 goto error;
911 }
912
913 nb_fd = ret;
914
915 for (i = 0; i < nb_fd; i++) {
916 /* Fetch once the poll data */
917 revents = LTTNG_POLL_GETEV(&events, i);
918 pollfd = LTTNG_POLL_GETFD(&events, i);
919
920 health_code_update();
921
922 /* Thread quit pipe has been closed. Killing thread. */
923 ret = check_thread_quit_pipe(pollfd, revents);
924 if (ret) {
925 err = 0;
926 goto exit;
927 }
928
929 /* Event on the registration socket */
930 if (pollfd == consumer_data->err_sock) {
931 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
932 ERR("consumer err socket poll error");
933 goto error;
934 }
935 }
936 }
937
938 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
939 if (sock < 0) {
940 goto error;
941 }
942
943 /*
944 * Set the CLOEXEC flag. Return code is useless because either way, the
945 * show must go on.
946 */
947 (void) utils_set_fd_cloexec(sock);
948
949 health_code_update();
950
951 DBG2("Receiving code from consumer err_sock");
952
953 /* Getting status code from kconsumerd */
954 ret = lttcomm_recv_unix_sock(sock, &code,
955 sizeof(enum lttcomm_return_code));
956 if (ret <= 0) {
957 goto error;
958 }
959
960 health_code_update();
961
962 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
963 consumer_data->cmd_sock =
964 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
965 if (consumer_data->cmd_sock < 0) {
966 /* On error, signal condition and quit. */
967 signal_consumer_condition(consumer_data, -1);
968 PERROR("consumer connect");
969 goto error;
970 }
971 signal_consumer_condition(consumer_data, 1);
972 DBG("Consumer command socket ready");
973 } else {
974 ERR("consumer error when waiting for SOCK_READY : %s",
975 lttcomm_get_readable_code(-code));
976 goto error;
977 }
978
979 /* Remove the kconsumerd error sock since we've established a connexion */
980 ret = lttng_poll_del(&events, consumer_data->err_sock);
981 if (ret < 0) {
982 goto error;
983 }
984
985 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
986 if (ret < 0) {
987 goto error;
988 }
989
990 health_code_update();
991
992 /* Inifinite blocking call, waiting for transmission */
993 restart_poll:
994 health_poll_entry();
995 ret = lttng_poll_wait(&events, -1);
996 health_poll_exit();
997 if (ret < 0) {
998 /*
999 * Restart interrupted system call.
1000 */
1001 if (errno == EINTR) {
1002 goto restart_poll;
1003 }
1004 goto error;
1005 }
1006
1007 nb_fd = ret;
1008
1009 for (i = 0; i < nb_fd; i++) {
1010 /* Fetch once the poll data */
1011 revents = LTTNG_POLL_GETEV(&events, i);
1012 pollfd = LTTNG_POLL_GETFD(&events, i);
1013
1014 health_code_update();
1015
1016 /* Thread quit pipe has been closed. Killing thread. */
1017 ret = check_thread_quit_pipe(pollfd, revents);
1018 if (ret) {
1019 err = 0;
1020 goto exit;
1021 }
1022
1023 /* Event on the kconsumerd socket */
1024 if (pollfd == sock) {
1025 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1026 ERR("consumer err socket second poll error");
1027 goto error;
1028 }
1029 }
1030 }
1031
1032 health_code_update();
1033
1034 /* Wait for any kconsumerd error */
1035 ret = lttcomm_recv_unix_sock(sock, &code,
1036 sizeof(enum lttcomm_return_code));
1037 if (ret <= 0) {
1038 ERR("consumer closed the command socket");
1039 goto error;
1040 }
1041
1042 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1043
1044 exit:
1045 error:
1046 /* Immediately set the consumerd state to stopped */
1047 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1048 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1049 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1050 consumer_data->type == LTTNG_CONSUMER32_UST) {
1051 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1052 } else {
1053 /* Code flow error... */
1054 assert(0);
1055 }
1056
1057 if (consumer_data->err_sock >= 0) {
1058 ret = close(consumer_data->err_sock);
1059 if (ret) {
1060 PERROR("close");
1061 }
1062 }
1063 if (consumer_data->cmd_sock >= 0) {
1064 ret = close(consumer_data->cmd_sock);
1065 if (ret) {
1066 PERROR("close");
1067 }
1068 }
1069 if (sock >= 0) {
1070 ret = close(sock);
1071 if (ret) {
1072 PERROR("close");
1073 }
1074 }
1075
1076 unlink(consumer_data->err_unix_sock_path);
1077 unlink(consumer_data->cmd_unix_sock_path);
1078 consumer_data->pid = 0;
1079
1080 lttng_poll_clean(&events);
1081 error_poll:
1082 if (err) {
1083 health_error();
1084 ERR("Health error occurred in %s", __func__);
1085 }
1086 health_unregister();
1087 DBG("consumer thread cleanup completed");
1088
1089 return NULL;
1090 }
1091
1092 /*
1093 * This thread manage application communication.
1094 */
1095 static void *thread_manage_apps(void *data)
1096 {
1097 int i, ret, pollfd, err = -1;
1098 uint32_t revents, nb_fd;
1099 struct ust_command ust_cmd;
1100 struct lttng_poll_event events;
1101
1102 DBG("[thread] Manage application started");
1103
1104 rcu_register_thread();
1105 rcu_thread_online();
1106
1107 health_register(HEALTH_TYPE_APP_MANAGE);
1108
1109 if (testpoint(thread_manage_apps)) {
1110 goto error_testpoint;
1111 }
1112
1113 health_code_update();
1114
1115 ret = create_thread_poll_set(&events, 2);
1116 if (ret < 0) {
1117 goto error_poll_create;
1118 }
1119
1120 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1121 if (ret < 0) {
1122 goto error;
1123 }
1124
1125 if (testpoint(thread_manage_apps_before_loop)) {
1126 goto error;
1127 }
1128
1129 health_code_update();
1130
1131 while (1) {
1132 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1133
1134 /* Inifinite blocking call, waiting for transmission */
1135 restart:
1136 health_poll_entry();
1137 ret = lttng_poll_wait(&events, -1);
1138 health_poll_exit();
1139 if (ret < 0) {
1140 /*
1141 * Restart interrupted system call.
1142 */
1143 if (errno == EINTR) {
1144 goto restart;
1145 }
1146 goto error;
1147 }
1148
1149 nb_fd = ret;
1150
1151 for (i = 0; i < nb_fd; i++) {
1152 /* Fetch once the poll data */
1153 revents = LTTNG_POLL_GETEV(&events, i);
1154 pollfd = LTTNG_POLL_GETFD(&events, i);
1155
1156 health_code_update();
1157
1158 /* Thread quit pipe has been closed. Killing thread. */
1159 ret = check_thread_quit_pipe(pollfd, revents);
1160 if (ret) {
1161 err = 0;
1162 goto exit;
1163 }
1164
1165 /* Inspect the apps cmd pipe */
1166 if (pollfd == apps_cmd_pipe[0]) {
1167 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1168 ERR("Apps command pipe error");
1169 goto error;
1170 } else if (revents & LPOLLIN) {
1171 /* Empty pipe */
1172 do {
1173 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1174 } while (ret < 0 && errno == EINTR);
1175 if (ret < 0 || ret < sizeof(ust_cmd)) {
1176 PERROR("read apps cmd pipe");
1177 goto error;
1178 }
1179
1180 health_code_update();
1181
1182 /* Register applicaton to the session daemon */
1183 ret = ust_app_register(&ust_cmd.reg_msg,
1184 ust_cmd.sock);
1185 if (ret == -ENOMEM) {
1186 goto error;
1187 } else if (ret < 0) {
1188 break;
1189 }
1190
1191 health_code_update();
1192
1193 /*
1194 * Validate UST version compatibility.
1195 */
1196 ret = ust_app_validate_version(ust_cmd.sock);
1197 if (ret >= 0) {
1198 /*
1199 * Add channel(s) and event(s) to newly registered apps
1200 * from lttng global UST domain.
1201 */
1202 update_ust_app(ust_cmd.sock);
1203 }
1204
1205 health_code_update();
1206
1207 ret = ust_app_register_done(ust_cmd.sock);
1208 if (ret < 0) {
1209 /*
1210 * If the registration is not possible, we simply
1211 * unregister the apps and continue
1212 */
1213 ust_app_unregister(ust_cmd.sock);
1214 } else {
1215 /*
1216 * We only monitor the error events of the socket. This
1217 * thread does not handle any incoming data from UST
1218 * (POLLIN).
1219 */
1220 ret = lttng_poll_add(&events, ust_cmd.sock,
1221 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1222 if (ret < 0) {
1223 goto error;
1224 }
1225
1226 /* Set socket timeout for both receiving and ending */
1227 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1228 app_socket_timeout);
1229 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1230 app_socket_timeout);
1231
1232 DBG("Apps with sock %d added to poll set",
1233 ust_cmd.sock);
1234 }
1235
1236 health_code_update();
1237
1238 break;
1239 }
1240 } else {
1241 /*
1242 * At this point, we know that a registered application made
1243 * the event at poll_wait.
1244 */
1245 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1246 /* Removing from the poll set */
1247 ret = lttng_poll_del(&events, pollfd);
1248 if (ret < 0) {
1249 goto error;
1250 }
1251
1252 /* Socket closed on remote end. */
1253 ust_app_unregister(pollfd);
1254 break;
1255 }
1256 }
1257
1258 health_code_update();
1259 }
1260 }
1261
1262 exit:
1263 error:
1264 lttng_poll_clean(&events);
1265 error_poll_create:
1266 error_testpoint:
1267 utils_close_pipe(apps_cmd_pipe);
1268 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1269
1270 /*
1271 * We don't clean the UST app hash table here since already registered
1272 * applications can still be controlled so let them be until the session
1273 * daemon dies or the applications stop.
1274 */
1275
1276 if (err) {
1277 health_error();
1278 ERR("Health error occurred in %s", __func__);
1279 }
1280 health_unregister();
1281 DBG("Application communication apps thread cleanup complete");
1282 rcu_thread_offline();
1283 rcu_unregister_thread();
1284 return NULL;
1285 }
1286
1287 /*
1288 * Dispatch request from the registration threads to the application
1289 * communication thread.
1290 */
1291 static void *thread_dispatch_ust_registration(void *data)
1292 {
1293 int ret;
1294 struct cds_wfq_node *node;
1295 struct ust_command *ust_cmd = NULL;
1296
1297 DBG("[thread] Dispatch UST command started");
1298
1299 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1300 /* Atomically prepare the queue futex */
1301 futex_nto1_prepare(&ust_cmd_queue.futex);
1302
1303 do {
1304 /* Dequeue command for registration */
1305 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1306 if (node == NULL) {
1307 DBG("Woken up but nothing in the UST command queue");
1308 /* Continue thread execution */
1309 break;
1310 }
1311
1312 ust_cmd = caa_container_of(node, struct ust_command, node);
1313
1314 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1315 " gid:%d sock:%d name:%s (version %d.%d)",
1316 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1317 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1318 ust_cmd->sock, ust_cmd->reg_msg.name,
1319 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1320 /*
1321 * Inform apps thread of the new application registration. This
1322 * call is blocking so we can be assured that the data will be read
1323 * at some point in time or wait to the end of the world :)
1324 */
1325 if (apps_cmd_pipe[1] >= 0) {
1326 do {
1327 ret = write(apps_cmd_pipe[1], ust_cmd,
1328 sizeof(struct ust_command));
1329 } while (ret < 0 && errno == EINTR);
1330 if (ret < 0 || ret != sizeof(struct ust_command)) {
1331 PERROR("write apps cmd pipe");
1332 if (errno == EBADF) {
1333 /*
1334 * We can't inform the application thread to process
1335 * registration. We will exit or else application
1336 * registration will not occur and tracing will never
1337 * start.
1338 */
1339 goto error;
1340 }
1341 }
1342 } else {
1343 /* Application manager thread is not available. */
1344 ret = close(ust_cmd->sock);
1345 if (ret < 0) {
1346 PERROR("close ust_cmd sock");
1347 }
1348 }
1349 free(ust_cmd);
1350 } while (node != NULL);
1351
1352 /* Futex wait on queue. Blocking call on futex() */
1353 futex_nto1_wait(&ust_cmd_queue.futex);
1354 }
1355
1356 error:
1357 DBG("Dispatch thread dying");
1358 return NULL;
1359 }
1360
1361 /*
1362 * This thread manage application registration.
1363 */
1364 static void *thread_registration_apps(void *data)
1365 {
1366 int sock = -1, i, ret, pollfd, err = -1;
1367 uint32_t revents, nb_fd;
1368 struct lttng_poll_event events;
1369 /*
1370 * Get allocated in this thread, enqueued to a global queue, dequeued and
1371 * freed in the manage apps thread.
1372 */
1373 struct ust_command *ust_cmd = NULL;
1374
1375 DBG("[thread] Manage application registration started");
1376
1377 health_register(HEALTH_TYPE_APP_REG);
1378
1379 if (testpoint(thread_registration_apps)) {
1380 goto error_testpoint;
1381 }
1382
1383 ret = lttcomm_listen_unix_sock(apps_sock);
1384 if (ret < 0) {
1385 goto error_listen;
1386 }
1387
1388 /*
1389 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1390 * more will be added to this poll set.
1391 */
1392 ret = create_thread_poll_set(&events, 2);
1393 if (ret < 0) {
1394 goto error_create_poll;
1395 }
1396
1397 /* Add the application registration socket */
1398 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1399 if (ret < 0) {
1400 goto error_poll_add;
1401 }
1402
1403 /* Notify all applications to register */
1404 ret = notify_ust_apps(1);
1405 if (ret < 0) {
1406 ERR("Failed to notify applications or create the wait shared memory.\n"
1407 "Execution continues but there might be problem for already\n"
1408 "running applications that wishes to register.");
1409 }
1410
1411 while (1) {
1412 DBG("Accepting application registration");
1413
1414 /* Inifinite blocking call, waiting for transmission */
1415 restart:
1416 health_poll_entry();
1417 ret = lttng_poll_wait(&events, -1);
1418 health_poll_exit();
1419 if (ret < 0) {
1420 /*
1421 * Restart interrupted system call.
1422 */
1423 if (errno == EINTR) {
1424 goto restart;
1425 }
1426 goto error;
1427 }
1428
1429 nb_fd = ret;
1430
1431 for (i = 0; i < nb_fd; i++) {
1432 health_code_update();
1433
1434 /* Fetch once the poll data */
1435 revents = LTTNG_POLL_GETEV(&events, i);
1436 pollfd = LTTNG_POLL_GETFD(&events, i);
1437
1438 /* Thread quit pipe has been closed. Killing thread. */
1439 ret = check_thread_quit_pipe(pollfd, revents);
1440 if (ret) {
1441 err = 0;
1442 goto exit;
1443 }
1444
1445 /* Event on the registration socket */
1446 if (pollfd == apps_sock) {
1447 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1448 ERR("Register apps socket poll error");
1449 goto error;
1450 } else if (revents & LPOLLIN) {
1451 sock = lttcomm_accept_unix_sock(apps_sock);
1452 if (sock < 0) {
1453 goto error;
1454 }
1455
1456 /*
1457 * Set the CLOEXEC flag. Return code is useless because
1458 * either way, the show must go on.
1459 */
1460 (void) utils_set_fd_cloexec(sock);
1461
1462 /* Create UST registration command for enqueuing */
1463 ust_cmd = zmalloc(sizeof(struct ust_command));
1464 if (ust_cmd == NULL) {
1465 PERROR("ust command zmalloc");
1466 goto error;
1467 }
1468
1469 /*
1470 * Using message-based transmissions to ensure we don't
1471 * have to deal with partially received messages.
1472 */
1473 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1474 if (ret < 0) {
1475 ERR("Exhausted file descriptors allowed for applications.");
1476 free(ust_cmd);
1477 ret = close(sock);
1478 if (ret) {
1479 PERROR("close");
1480 }
1481 sock = -1;
1482 continue;
1483 }
1484 health_code_update();
1485 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1486 sizeof(struct ust_register_msg));
1487 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1488 if (ret < 0) {
1489 PERROR("lttcomm_recv_unix_sock register apps");
1490 } else {
1491 ERR("Wrong size received on apps register");
1492 }
1493 free(ust_cmd);
1494 ret = close(sock);
1495 if (ret) {
1496 PERROR("close");
1497 }
1498 lttng_fd_put(LTTNG_FD_APPS, 1);
1499 sock = -1;
1500 continue;
1501 }
1502 health_code_update();
1503
1504 ust_cmd->sock = sock;
1505 sock = -1;
1506
1507 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1508 " gid:%d sock:%d name:%s (version %d.%d)",
1509 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1510 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1511 ust_cmd->sock, ust_cmd->reg_msg.name,
1512 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1513
1514 /*
1515 * Lock free enqueue the registration request. The red pill
1516 * has been taken! This apps will be part of the *system*.
1517 */
1518 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1519
1520 /*
1521 * Wake the registration queue futex. Implicit memory
1522 * barrier with the exchange in cds_wfq_enqueue.
1523 */
1524 futex_nto1_wake(&ust_cmd_queue.futex);
1525 }
1526 }
1527 }
1528 }
1529
1530 exit:
1531 error:
1532 if (err) {
1533 health_error();
1534 ERR("Health error occurred in %s", __func__);
1535 }
1536
1537 /* Notify that the registration thread is gone */
1538 notify_ust_apps(0);
1539
1540 if (apps_sock >= 0) {
1541 ret = close(apps_sock);
1542 if (ret) {
1543 PERROR("close");
1544 }
1545 }
1546 if (sock >= 0) {
1547 ret = close(sock);
1548 if (ret) {
1549 PERROR("close");
1550 }
1551 lttng_fd_put(LTTNG_FD_APPS, 1);
1552 }
1553 unlink(apps_unix_sock_path);
1554
1555 error_poll_add:
1556 lttng_poll_clean(&events);
1557 error_listen:
1558 error_create_poll:
1559 error_testpoint:
1560 DBG("UST Registration thread cleanup complete");
1561 health_unregister();
1562
1563 return NULL;
1564 }
1565
1566 /*
1567 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1568 * exec or it will fails.
1569 */
1570 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1571 {
1572 int ret, clock_ret;
1573 struct timespec timeout;
1574
1575 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1576 consumer_data->consumer_thread_is_ready = 0;
1577
1578 /* Setup pthread condition */
1579 ret = pthread_condattr_init(&consumer_data->condattr);
1580 if (ret != 0) {
1581 errno = ret;
1582 PERROR("pthread_condattr_init consumer data");
1583 goto error;
1584 }
1585
1586 /*
1587 * Set the monotonic clock in order to make sure we DO NOT jump in time
1588 * between the clock_gettime() call and the timedwait call. See bug #324
1589 * for a more details and how we noticed it.
1590 */
1591 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1592 if (ret != 0) {
1593 errno = ret;
1594 PERROR("pthread_condattr_setclock consumer data");
1595 goto error;
1596 }
1597
1598 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1599 if (ret != 0) {
1600 errno = ret;
1601 PERROR("pthread_cond_init consumer data");
1602 goto error;
1603 }
1604
1605 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1606 consumer_data);
1607 if (ret != 0) {
1608 PERROR("pthread_create consumer");
1609 ret = -1;
1610 goto error;
1611 }
1612
1613 /* We are about to wait on a pthread condition */
1614 pthread_mutex_lock(&consumer_data->cond_mutex);
1615
1616 /* Get time for sem_timedwait absolute timeout */
1617 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1618 /*
1619 * Set the timeout for the condition timed wait even if the clock gettime
1620 * call fails since we might loop on that call and we want to avoid to
1621 * increment the timeout too many times.
1622 */
1623 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1624
1625 /*
1626 * The following loop COULD be skipped in some conditions so this is why we
1627 * set ret to 0 in order to make sure at least one round of the loop is
1628 * done.
1629 */
1630 ret = 0;
1631
1632 /*
1633 * Loop until the condition is reached or when a timeout is reached. Note
1634 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1635 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1636 * possible. This loop does not take any chances and works with both of
1637 * them.
1638 */
1639 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1640 if (clock_ret < 0) {
1641 PERROR("clock_gettime spawn consumer");
1642 /* Infinite wait for the consumerd thread to be ready */
1643 ret = pthread_cond_wait(&consumer_data->cond,
1644 &consumer_data->cond_mutex);
1645 } else {
1646 ret = pthread_cond_timedwait(&consumer_data->cond,
1647 &consumer_data->cond_mutex, &timeout);
1648 }
1649 }
1650
1651 /* Release the pthread condition */
1652 pthread_mutex_unlock(&consumer_data->cond_mutex);
1653
1654 if (ret != 0) {
1655 errno = ret;
1656 if (ret == ETIMEDOUT) {
1657 /*
1658 * Call has timed out so we kill the kconsumerd_thread and return
1659 * an error.
1660 */
1661 ERR("Condition timed out. The consumer thread was never ready."
1662 " Killing it");
1663 ret = pthread_cancel(consumer_data->thread);
1664 if (ret < 0) {
1665 PERROR("pthread_cancel consumer thread");
1666 }
1667 } else {
1668 PERROR("pthread_cond_wait failed consumer thread");
1669 }
1670 goto error;
1671 }
1672
1673 pthread_mutex_lock(&consumer_data->pid_mutex);
1674 if (consumer_data->pid == 0) {
1675 ERR("Consumerd did not start");
1676 pthread_mutex_unlock(&consumer_data->pid_mutex);
1677 goto error;
1678 }
1679 pthread_mutex_unlock(&consumer_data->pid_mutex);
1680
1681 return 0;
1682
1683 error:
1684 return ret;
1685 }
1686
1687 /*
1688 * Join consumer thread
1689 */
1690 static int join_consumer_thread(struct consumer_data *consumer_data)
1691 {
1692 void *status;
1693
1694 /* Consumer pid must be a real one. */
1695 if (consumer_data->pid > 0) {
1696 int ret;
1697 ret = kill(consumer_data->pid, SIGTERM);
1698 if (ret) {
1699 ERR("Error killing consumer daemon");
1700 return ret;
1701 }
1702 return pthread_join(consumer_data->thread, &status);
1703 } else {
1704 return 0;
1705 }
1706 }
1707
1708 /*
1709 * Fork and exec a consumer daemon (consumerd).
1710 *
1711 * Return pid if successful else -1.
1712 */
1713 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1714 {
1715 int ret;
1716 pid_t pid;
1717 const char *consumer_to_use;
1718 const char *verbosity;
1719 struct stat st;
1720
1721 DBG("Spawning consumerd");
1722
1723 pid = fork();
1724 if (pid == 0) {
1725 /*
1726 * Exec consumerd.
1727 */
1728 if (opt_verbose_consumer) {
1729 verbosity = "--verbose";
1730 } else {
1731 verbosity = "--quiet";
1732 }
1733 switch (consumer_data->type) {
1734 case LTTNG_CONSUMER_KERNEL:
1735 /*
1736 * Find out which consumerd to execute. We will first try the
1737 * 64-bit path, then the sessiond's installation directory, and
1738 * fallback on the 32-bit one,
1739 */
1740 DBG3("Looking for a kernel consumer at these locations:");
1741 DBG3(" 1) %s", consumerd64_bin);
1742 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1743 DBG3(" 3) %s", consumerd32_bin);
1744 if (stat(consumerd64_bin, &st) == 0) {
1745 DBG3("Found location #1");
1746 consumer_to_use = consumerd64_bin;
1747 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1748 DBG3("Found location #2");
1749 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1750 } else if (stat(consumerd32_bin, &st) == 0) {
1751 DBG3("Found location #3");
1752 consumer_to_use = consumerd32_bin;
1753 } else {
1754 DBG("Could not find any valid consumerd executable");
1755 break;
1756 }
1757 DBG("Using kernel consumer at: %s", consumer_to_use);
1758 execl(consumer_to_use,
1759 "lttng-consumerd", verbosity, "-k",
1760 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1761 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1762 NULL);
1763 break;
1764 case LTTNG_CONSUMER64_UST:
1765 {
1766 char *tmpnew = NULL;
1767
1768 if (consumerd64_libdir[0] != '\0') {
1769 char *tmp;
1770 size_t tmplen;
1771
1772 tmp = getenv("LD_LIBRARY_PATH");
1773 if (!tmp) {
1774 tmp = "";
1775 }
1776 tmplen = strlen("LD_LIBRARY_PATH=")
1777 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1778 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1779 if (!tmpnew) {
1780 ret = -ENOMEM;
1781 goto error;
1782 }
1783 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1784 strcat(tmpnew, consumerd64_libdir);
1785 if (tmp[0] != '\0') {
1786 strcat(tmpnew, ":");
1787 strcat(tmpnew, tmp);
1788 }
1789 ret = putenv(tmpnew);
1790 if (ret) {
1791 ret = -errno;
1792 goto error;
1793 }
1794 }
1795 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1796 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1797 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1798 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1799 NULL);
1800 if (consumerd64_libdir[0] != '\0') {
1801 free(tmpnew);
1802 }
1803 if (ret) {
1804 goto error;
1805 }
1806 break;
1807 }
1808 case LTTNG_CONSUMER32_UST:
1809 {
1810 char *tmpnew = NULL;
1811
1812 if (consumerd32_libdir[0] != '\0') {
1813 char *tmp;
1814 size_t tmplen;
1815
1816 tmp = getenv("LD_LIBRARY_PATH");
1817 if (!tmp) {
1818 tmp = "";
1819 }
1820 tmplen = strlen("LD_LIBRARY_PATH=")
1821 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1822 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1823 if (!tmpnew) {
1824 ret = -ENOMEM;
1825 goto error;
1826 }
1827 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1828 strcat(tmpnew, consumerd32_libdir);
1829 if (tmp[0] != '\0') {
1830 strcat(tmpnew, ":");
1831 strcat(tmpnew, tmp);
1832 }
1833 ret = putenv(tmpnew);
1834 if (ret) {
1835 ret = -errno;
1836 goto error;
1837 }
1838 }
1839 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1840 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1841 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1842 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1843 NULL);
1844 if (consumerd32_libdir[0] != '\0') {
1845 free(tmpnew);
1846 }
1847 if (ret) {
1848 goto error;
1849 }
1850 break;
1851 }
1852 default:
1853 PERROR("unknown consumer type");
1854 exit(EXIT_FAILURE);
1855 }
1856 if (errno != 0) {
1857 PERROR("kernel start consumer exec");
1858 }
1859 exit(EXIT_FAILURE);
1860 } else if (pid > 0) {
1861 ret = pid;
1862 } else {
1863 PERROR("start consumer fork");
1864 ret = -errno;
1865 }
1866 error:
1867 return ret;
1868 }
1869
1870 /*
1871 * Spawn the consumerd daemon and session daemon thread.
1872 */
1873 static int start_consumerd(struct consumer_data *consumer_data)
1874 {
1875 int ret;
1876
1877 /*
1878 * Set the listen() state on the socket since there is a possible race
1879 * between the exec() of the consumer daemon and this call if place in the
1880 * consumer thread. See bug #366 for more details.
1881 */
1882 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1883 if (ret < 0) {
1884 goto error;
1885 }
1886
1887 pthread_mutex_lock(&consumer_data->pid_mutex);
1888 if (consumer_data->pid != 0) {
1889 pthread_mutex_unlock(&consumer_data->pid_mutex);
1890 goto end;
1891 }
1892
1893 ret = spawn_consumerd(consumer_data);
1894 if (ret < 0) {
1895 ERR("Spawning consumerd failed");
1896 pthread_mutex_unlock(&consumer_data->pid_mutex);
1897 goto error;
1898 }
1899
1900 /* Setting up the consumer_data pid */
1901 consumer_data->pid = ret;
1902 DBG2("Consumer pid %d", consumer_data->pid);
1903 pthread_mutex_unlock(&consumer_data->pid_mutex);
1904
1905 DBG2("Spawning consumer control thread");
1906 ret = spawn_consumer_thread(consumer_data);
1907 if (ret < 0) {
1908 ERR("Fatal error spawning consumer control thread");
1909 goto error;
1910 }
1911
1912 end:
1913 return 0;
1914
1915 error:
1916 /* Cleanup already created socket on error. */
1917 if (consumer_data->err_sock >= 0) {
1918 int err;
1919
1920 err = close(consumer_data->err_sock);
1921 if (err < 0) {
1922 PERROR("close consumer data error socket");
1923 }
1924 }
1925 return ret;
1926 }
1927
1928 /*
1929 * Compute health status of each consumer. If one of them is zero (bad
1930 * state), we return 0.
1931 */
1932 static int check_consumer_health(void)
1933 {
1934 int ret;
1935
1936 ret = health_check_state(HEALTH_TYPE_CONSUMER);
1937
1938 DBG3("Health consumer check %d", ret);
1939
1940 return ret;
1941 }
1942
1943 /*
1944 * Setup necessary data for kernel tracer action.
1945 */
1946 static int init_kernel_tracer(void)
1947 {
1948 int ret;
1949
1950 /* Modprobe lttng kernel modules */
1951 ret = modprobe_lttng_control();
1952 if (ret < 0) {
1953 goto error;
1954 }
1955
1956 /* Open debugfs lttng */
1957 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1958 if (kernel_tracer_fd < 0) {
1959 DBG("Failed to open %s", module_proc_lttng);
1960 ret = -1;
1961 goto error_open;
1962 }
1963
1964 /* Validate kernel version */
1965 ret = kernel_validate_version(kernel_tracer_fd);
1966 if (ret < 0) {
1967 goto error_version;
1968 }
1969
1970 ret = modprobe_lttng_data();
1971 if (ret < 0) {
1972 goto error_modules;
1973 }
1974
1975 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1976 return 0;
1977
1978 error_version:
1979 modprobe_remove_lttng_control();
1980 ret = close(kernel_tracer_fd);
1981 if (ret) {
1982 PERROR("close");
1983 }
1984 kernel_tracer_fd = -1;
1985 return LTTNG_ERR_KERN_VERSION;
1986
1987 error_modules:
1988 ret = close(kernel_tracer_fd);
1989 if (ret) {
1990 PERROR("close");
1991 }
1992
1993 error_open:
1994 modprobe_remove_lttng_control();
1995
1996 error:
1997 WARN("No kernel tracer available");
1998 kernel_tracer_fd = -1;
1999 if (!is_root) {
2000 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2001 } else {
2002 return LTTNG_ERR_KERN_NA;
2003 }
2004 }
2005
2006
2007 /*
2008 * Copy consumer output from the tracing session to the domain session. The
2009 * function also applies the right modification on a per domain basis for the
2010 * trace files destination directory.
2011 */
2012 static int copy_session_consumer(int domain, struct ltt_session *session)
2013 {
2014 int ret;
2015 const char *dir_name;
2016 struct consumer_output *consumer;
2017
2018 assert(session);
2019 assert(session->consumer);
2020
2021 switch (domain) {
2022 case LTTNG_DOMAIN_KERNEL:
2023 DBG3("Copying tracing session consumer output in kernel session");
2024 /*
2025 * XXX: We should audit the session creation and what this function
2026 * does "extra" in order to avoid a destroy since this function is used
2027 * in the domain session creation (kernel and ust) only. Same for UST
2028 * domain.
2029 */
2030 if (session->kernel_session->consumer) {
2031 consumer_destroy_output(session->kernel_session->consumer);
2032 }
2033 session->kernel_session->consumer =
2034 consumer_copy_output(session->consumer);
2035 /* Ease our life a bit for the next part */
2036 consumer = session->kernel_session->consumer;
2037 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2038 break;
2039 case LTTNG_DOMAIN_UST:
2040 DBG3("Copying tracing session consumer output in UST session");
2041 if (session->ust_session->consumer) {
2042 consumer_destroy_output(session->ust_session->consumer);
2043 }
2044 session->ust_session->consumer =
2045 consumer_copy_output(session->consumer);
2046 /* Ease our life a bit for the next part */
2047 consumer = session->ust_session->consumer;
2048 dir_name = DEFAULT_UST_TRACE_DIR;
2049 break;
2050 default:
2051 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2052 goto error;
2053 }
2054
2055 /* Append correct directory to subdir */
2056 strncat(consumer->subdir, dir_name,
2057 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2058 DBG3("Copy session consumer subdir %s", consumer->subdir);
2059
2060 ret = LTTNG_OK;
2061
2062 error:
2063 return ret;
2064 }
2065
2066 /*
2067 * Create an UST session and add it to the session ust list.
2068 */
2069 static int create_ust_session(struct ltt_session *session,
2070 struct lttng_domain *domain)
2071 {
2072 int ret;
2073 struct ltt_ust_session *lus = NULL;
2074
2075 assert(session);
2076 assert(domain);
2077 assert(session->consumer);
2078
2079 switch (domain->type) {
2080 case LTTNG_DOMAIN_UST:
2081 break;
2082 default:
2083 ERR("Unknown UST domain on create session %d", domain->type);
2084 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2085 goto error;
2086 }
2087
2088 DBG("Creating UST session");
2089
2090 lus = trace_ust_create_session(session->path, session->id);
2091 if (lus == NULL) {
2092 ret = LTTNG_ERR_UST_SESS_FAIL;
2093 goto error;
2094 }
2095
2096 lus->uid = session->uid;
2097 lus->gid = session->gid;
2098 session->ust_session = lus;
2099
2100 /* Copy session output to the newly created UST session */
2101 ret = copy_session_consumer(domain->type, session);
2102 if (ret != LTTNG_OK) {
2103 goto error;
2104 }
2105
2106 return LTTNG_OK;
2107
2108 error:
2109 free(lus);
2110 session->ust_session = NULL;
2111 return ret;
2112 }
2113
2114 /*
2115 * Create a kernel tracer session then create the default channel.
2116 */
2117 static int create_kernel_session(struct ltt_session *session)
2118 {
2119 int ret;
2120
2121 DBG("Creating kernel session");
2122
2123 ret = kernel_create_session(session, kernel_tracer_fd);
2124 if (ret < 0) {
2125 ret = LTTNG_ERR_KERN_SESS_FAIL;
2126 goto error;
2127 }
2128
2129 /* Code flow safety */
2130 assert(session->kernel_session);
2131
2132 /* Copy session output to the newly created Kernel session */
2133 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2134 if (ret != LTTNG_OK) {
2135 goto error;
2136 }
2137
2138 /* Create directory(ies) on local filesystem. */
2139 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2140 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2141 ret = run_as_mkdir_recursive(
2142 session->kernel_session->consumer->dst.trace_path,
2143 S_IRWXU | S_IRWXG, session->uid, session->gid);
2144 if (ret < 0) {
2145 if (ret != -EEXIST) {
2146 ERR("Trace directory creation error");
2147 goto error;
2148 }
2149 }
2150 }
2151
2152 session->kernel_session->uid = session->uid;
2153 session->kernel_session->gid = session->gid;
2154
2155 return LTTNG_OK;
2156
2157 error:
2158 trace_kernel_destroy_session(session->kernel_session);
2159 session->kernel_session = NULL;
2160 return ret;
2161 }
2162
2163 /*
2164 * Count number of session permitted by uid/gid.
2165 */
2166 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2167 {
2168 unsigned int i = 0;
2169 struct ltt_session *session;
2170
2171 DBG("Counting number of available session for UID %d GID %d",
2172 uid, gid);
2173 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2174 /*
2175 * Only list the sessions the user can control.
2176 */
2177 if (!session_access_ok(session, uid, gid)) {
2178 continue;
2179 }
2180 i++;
2181 }
2182 return i;
2183 }
2184
2185 /*
2186 * Process the command requested by the lttng client within the command
2187 * context structure. This function make sure that the return structure (llm)
2188 * is set and ready for transmission before returning.
2189 *
2190 * Return any error encountered or 0 for success.
2191 *
2192 * "sock" is only used for special-case var. len data.
2193 */
2194 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2195 int *sock_error)
2196 {
2197 int ret = LTTNG_OK;
2198 int need_tracing_session = 1;
2199 int need_domain;
2200
2201 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2202
2203 *sock_error = 0;
2204
2205 switch (cmd_ctx->lsm->cmd_type) {
2206 case LTTNG_CREATE_SESSION:
2207 case LTTNG_DESTROY_SESSION:
2208 case LTTNG_LIST_SESSIONS:
2209 case LTTNG_LIST_DOMAINS:
2210 case LTTNG_START_TRACE:
2211 case LTTNG_STOP_TRACE:
2212 case LTTNG_DATA_PENDING:
2213 need_domain = 0;
2214 break;
2215 default:
2216 need_domain = 1;
2217 }
2218
2219 if (opt_no_kernel && need_domain
2220 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2221 if (!is_root) {
2222 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2223 } else {
2224 ret = LTTNG_ERR_KERN_NA;
2225 }
2226 goto error;
2227 }
2228
2229 /* Deny register consumer if we already have a spawned consumer. */
2230 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2231 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2232 if (kconsumer_data.pid > 0) {
2233 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2234 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2235 goto error;
2236 }
2237 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2238 }
2239
2240 /*
2241 * Check for command that don't needs to allocate a returned payload. We do
2242 * this here so we don't have to make the call for no payload at each
2243 * command.
2244 */
2245 switch(cmd_ctx->lsm->cmd_type) {
2246 case LTTNG_LIST_SESSIONS:
2247 case LTTNG_LIST_TRACEPOINTS:
2248 case LTTNG_LIST_TRACEPOINT_FIELDS:
2249 case LTTNG_LIST_DOMAINS:
2250 case LTTNG_LIST_CHANNELS:
2251 case LTTNG_LIST_EVENTS:
2252 break;
2253 default:
2254 /* Setup lttng message with no payload */
2255 ret = setup_lttng_msg(cmd_ctx, 0);
2256 if (ret < 0) {
2257 /* This label does not try to unlock the session */
2258 goto init_setup_error;
2259 }
2260 }
2261
2262 /* Commands that DO NOT need a session. */
2263 switch (cmd_ctx->lsm->cmd_type) {
2264 case LTTNG_CREATE_SESSION:
2265 case LTTNG_CALIBRATE:
2266 case LTTNG_LIST_SESSIONS:
2267 case LTTNG_LIST_TRACEPOINTS:
2268 case LTTNG_LIST_TRACEPOINT_FIELDS:
2269 need_tracing_session = 0;
2270 break;
2271 default:
2272 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2273 /*
2274 * We keep the session list lock across _all_ commands
2275 * for now, because the per-session lock does not
2276 * handle teardown properly.
2277 */
2278 session_lock_list();
2279 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2280 if (cmd_ctx->session == NULL) {
2281 if (cmd_ctx->lsm->session.name != NULL) {
2282 ret = LTTNG_ERR_SESS_NOT_FOUND;
2283 } else {
2284 /* If no session name specified */
2285 ret = LTTNG_ERR_SELECT_SESS;
2286 }
2287 goto error;
2288 } else {
2289 /* Acquire lock for the session */
2290 session_lock(cmd_ctx->session);
2291 }
2292 break;
2293 }
2294
2295 if (!need_domain) {
2296 goto skip_domain;
2297 }
2298
2299 /*
2300 * Check domain type for specific "pre-action".
2301 */
2302 switch (cmd_ctx->lsm->domain.type) {
2303 case LTTNG_DOMAIN_KERNEL:
2304 if (!is_root) {
2305 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2306 goto error;
2307 }
2308
2309 /* Kernel tracer check */
2310 if (kernel_tracer_fd == -1) {
2311 /* Basically, load kernel tracer modules */
2312 ret = init_kernel_tracer();
2313 if (ret != 0) {
2314 goto error;
2315 }
2316 }
2317
2318 /* Consumer is in an ERROR state. Report back to client */
2319 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2320 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2321 goto error;
2322 }
2323
2324 /* Need a session for kernel command */
2325 if (need_tracing_session) {
2326 if (cmd_ctx->session->kernel_session == NULL) {
2327 ret = create_kernel_session(cmd_ctx->session);
2328 if (ret < 0) {
2329 ret = LTTNG_ERR_KERN_SESS_FAIL;
2330 goto error;
2331 }
2332 }
2333
2334 /* Start the kernel consumer daemon */
2335 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2336 if (kconsumer_data.pid == 0 &&
2337 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2338 cmd_ctx->session->start_consumer) {
2339 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2340 ret = start_consumerd(&kconsumer_data);
2341 if (ret < 0) {
2342 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2343 goto error;
2344 }
2345 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2346 } else {
2347 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2348 }
2349
2350 /*
2351 * The consumer was just spawned so we need to add the socket to
2352 * the consumer output of the session if exist.
2353 */
2354 ret = consumer_create_socket(&kconsumer_data,
2355 cmd_ctx->session->kernel_session->consumer);
2356 if (ret < 0) {
2357 goto error;
2358 }
2359 }
2360
2361 break;
2362 case LTTNG_DOMAIN_UST:
2363 {
2364 /* Consumer is in an ERROR state. Report back to client */
2365 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2366 ret = LTTNG_ERR_NO_USTCONSUMERD;
2367 goto error;
2368 }
2369
2370 if (need_tracing_session) {
2371 /* Create UST session if none exist. */
2372 if (cmd_ctx->session->ust_session == NULL) {
2373 ret = create_ust_session(cmd_ctx->session,
2374 &cmd_ctx->lsm->domain);
2375 if (ret != LTTNG_OK) {
2376 goto error;
2377 }
2378 }
2379
2380 /* Start the UST consumer daemons */
2381 /* 64-bit */
2382 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2383 if (consumerd64_bin[0] != '\0' &&
2384 ustconsumer64_data.pid == 0 &&
2385 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2386 cmd_ctx->session->start_consumer) {
2387 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2388 ret = start_consumerd(&ustconsumer64_data);
2389 if (ret < 0) {
2390 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2391 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2392 goto error;
2393 }
2394
2395 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2396 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2397 } else {
2398 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2399 }
2400
2401 /*
2402 * Setup socket for consumer 64 bit. No need for atomic access
2403 * since it was set above and can ONLY be set in this thread.
2404 */
2405 ret = consumer_create_socket(&ustconsumer64_data,
2406 cmd_ctx->session->ust_session->consumer);
2407 if (ret < 0) {
2408 goto error;
2409 }
2410
2411 /* 32-bit */
2412 if (consumerd32_bin[0] != '\0' &&
2413 ustconsumer32_data.pid == 0 &&
2414 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2415 cmd_ctx->session->start_consumer) {
2416 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2417 ret = start_consumerd(&ustconsumer32_data);
2418 if (ret < 0) {
2419 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2420 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2421 goto error;
2422 }
2423
2424 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2425 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2426 } else {
2427 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2428 }
2429
2430 /*
2431 * Setup socket for consumer 64 bit. No need for atomic access
2432 * since it was set above and can ONLY be set in this thread.
2433 */
2434 ret = consumer_create_socket(&ustconsumer32_data,
2435 cmd_ctx->session->ust_session->consumer);
2436 if (ret < 0) {
2437 goto error;
2438 }
2439 }
2440 break;
2441 }
2442 default:
2443 break;
2444 }
2445 skip_domain:
2446
2447 /* Validate consumer daemon state when start/stop trace command */
2448 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2449 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2450 switch (cmd_ctx->lsm->domain.type) {
2451 case LTTNG_DOMAIN_UST:
2452 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2453 ret = LTTNG_ERR_NO_USTCONSUMERD;
2454 goto error;
2455 }
2456 break;
2457 case LTTNG_DOMAIN_KERNEL:
2458 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2459 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2460 goto error;
2461 }
2462 break;
2463 }
2464 }
2465
2466 /*
2467 * Check that the UID or GID match that of the tracing session.
2468 * The root user can interact with all sessions.
2469 */
2470 if (need_tracing_session) {
2471 if (!session_access_ok(cmd_ctx->session,
2472 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2473 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2474 ret = LTTNG_ERR_EPERM;
2475 goto error;
2476 }
2477 }
2478
2479 /* Process by command type */
2480 switch (cmd_ctx->lsm->cmd_type) {
2481 case LTTNG_ADD_CONTEXT:
2482 {
2483 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2484 cmd_ctx->lsm->u.context.channel_name,
2485 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2486 break;
2487 }
2488 case LTTNG_DISABLE_CHANNEL:
2489 {
2490 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2491 cmd_ctx->lsm->u.disable.channel_name);
2492 break;
2493 }
2494 case LTTNG_DISABLE_EVENT:
2495 {
2496 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2497 cmd_ctx->lsm->u.disable.channel_name,
2498 cmd_ctx->lsm->u.disable.name);
2499 break;
2500 }
2501 case LTTNG_DISABLE_ALL_EVENT:
2502 {
2503 DBG("Disabling all events");
2504
2505 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2506 cmd_ctx->lsm->u.disable.channel_name);
2507 break;
2508 }
2509 case LTTNG_DISABLE_CONSUMER:
2510 {
2511 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2512 break;
2513 }
2514 case LTTNG_ENABLE_CHANNEL:
2515 {
2516 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2517 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2518 break;
2519 }
2520 case LTTNG_ENABLE_CONSUMER:
2521 {
2522 /*
2523 * XXX: 0 means that this URI should be applied on the session. Should
2524 * be a DOMAIN enuam.
2525 */
2526 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2527 if (ret != LTTNG_OK) {
2528 goto error;
2529 }
2530
2531 if (cmd_ctx->lsm->domain.type == 0) {
2532 /* Add the URI for the UST session if a consumer is present. */
2533 if (cmd_ctx->session->ust_session &&
2534 cmd_ctx->session->ust_session->consumer) {
2535 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2536 } else if (cmd_ctx->session->kernel_session &&
2537 cmd_ctx->session->kernel_session->consumer) {
2538 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2539 cmd_ctx->session);
2540 }
2541 }
2542 break;
2543 }
2544 case LTTNG_ENABLE_EVENT:
2545 {
2546 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2547 cmd_ctx->lsm->u.enable.channel_name,
2548 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2549 break;
2550 }
2551 case LTTNG_ENABLE_ALL_EVENT:
2552 {
2553 DBG("Enabling all events");
2554
2555 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2556 cmd_ctx->lsm->u.enable.channel_name,
2557 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2558 break;
2559 }
2560 case LTTNG_LIST_TRACEPOINTS:
2561 {
2562 struct lttng_event *events;
2563 ssize_t nb_events;
2564
2565 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2566 if (nb_events < 0) {
2567 /* Return value is a negative lttng_error_code. */
2568 ret = -nb_events;
2569 goto error;
2570 }
2571
2572 /*
2573 * Setup lttng message with payload size set to the event list size in
2574 * bytes and then copy list into the llm payload.
2575 */
2576 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2577 if (ret < 0) {
2578 free(events);
2579 goto setup_error;
2580 }
2581
2582 /* Copy event list into message payload */
2583 memcpy(cmd_ctx->llm->payload, events,
2584 sizeof(struct lttng_event) * nb_events);
2585
2586 free(events);
2587
2588 ret = LTTNG_OK;
2589 break;
2590 }
2591 case LTTNG_LIST_TRACEPOINT_FIELDS:
2592 {
2593 struct lttng_event_field *fields;
2594 ssize_t nb_fields;
2595
2596 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2597 &fields);
2598 if (nb_fields < 0) {
2599 /* Return value is a negative lttng_error_code. */
2600 ret = -nb_fields;
2601 goto error;
2602 }
2603
2604 /*
2605 * Setup lttng message with payload size set to the event list size in
2606 * bytes and then copy list into the llm payload.
2607 */
2608 ret = setup_lttng_msg(cmd_ctx,
2609 sizeof(struct lttng_event_field) * nb_fields);
2610 if (ret < 0) {
2611 free(fields);
2612 goto setup_error;
2613 }
2614
2615 /* Copy event list into message payload */
2616 memcpy(cmd_ctx->llm->payload, fields,
2617 sizeof(struct lttng_event_field) * nb_fields);
2618
2619 free(fields);
2620
2621 ret = LTTNG_OK;
2622 break;
2623 }
2624 case LTTNG_SET_CONSUMER_URI:
2625 {
2626 size_t nb_uri, len;
2627 struct lttng_uri *uris;
2628
2629 nb_uri = cmd_ctx->lsm->u.uri.size;
2630 len = nb_uri * sizeof(struct lttng_uri);
2631
2632 if (nb_uri == 0) {
2633 ret = LTTNG_ERR_INVALID;
2634 goto error;
2635 }
2636
2637 uris = zmalloc(len);
2638 if (uris == NULL) {
2639 ret = LTTNG_ERR_FATAL;
2640 goto error;
2641 }
2642
2643 /* Receive variable len data */
2644 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2645 ret = lttcomm_recv_unix_sock(sock, uris, len);
2646 if (ret <= 0) {
2647 DBG("No URIs received from client... continuing");
2648 *sock_error = 1;
2649 ret = LTTNG_ERR_SESSION_FAIL;
2650 free(uris);
2651 goto error;
2652 }
2653
2654 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2655 nb_uri, uris);
2656 if (ret != LTTNG_OK) {
2657 free(uris);
2658 goto error;
2659 }
2660
2661 /*
2662 * XXX: 0 means that this URI should be applied on the session. Should
2663 * be a DOMAIN enuam.
2664 */
2665 if (cmd_ctx->lsm->domain.type == 0) {
2666 /* Add the URI for the UST session if a consumer is present. */
2667 if (cmd_ctx->session->ust_session &&
2668 cmd_ctx->session->ust_session->consumer) {
2669 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2670 nb_uri, uris);
2671 } else if (cmd_ctx->session->kernel_session &&
2672 cmd_ctx->session->kernel_session->consumer) {
2673 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2674 cmd_ctx->session, nb_uri, uris);
2675 }
2676 }
2677
2678 free(uris);
2679
2680 break;
2681 }
2682 case LTTNG_START_TRACE:
2683 {
2684 ret = cmd_start_trace(cmd_ctx->session);
2685 break;
2686 }
2687 case LTTNG_STOP_TRACE:
2688 {
2689 ret = cmd_stop_trace(cmd_ctx->session);
2690 break;
2691 }
2692 case LTTNG_CREATE_SESSION:
2693 {
2694 size_t nb_uri, len;
2695 struct lttng_uri *uris = NULL;
2696
2697 nb_uri = cmd_ctx->lsm->u.uri.size;
2698 len = nb_uri * sizeof(struct lttng_uri);
2699
2700 if (nb_uri > 0) {
2701 uris = zmalloc(len);
2702 if (uris == NULL) {
2703 ret = LTTNG_ERR_FATAL;
2704 goto error;
2705 }
2706
2707 /* Receive variable len data */
2708 DBG("Waiting for %zu URIs from client ...", nb_uri);
2709 ret = lttcomm_recv_unix_sock(sock, uris, len);
2710 if (ret <= 0) {
2711 DBG("No URIs received from client... continuing");
2712 *sock_error = 1;
2713 ret = LTTNG_ERR_SESSION_FAIL;
2714 free(uris);
2715 goto error;
2716 }
2717
2718 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2719 DBG("Creating session with ONE network URI is a bad call");
2720 ret = LTTNG_ERR_SESSION_FAIL;
2721 free(uris);
2722 goto error;
2723 }
2724 }
2725
2726 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2727 &cmd_ctx->creds);
2728
2729 free(uris);
2730
2731 break;
2732 }
2733 case LTTNG_DESTROY_SESSION:
2734 {
2735 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2736
2737 /* Set session to NULL so we do not unlock it after free. */
2738 cmd_ctx->session = NULL;
2739 break;
2740 }
2741 case LTTNG_LIST_DOMAINS:
2742 {
2743 ssize_t nb_dom;
2744 struct lttng_domain *domains;
2745
2746 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2747 if (nb_dom < 0) {
2748 /* Return value is a negative lttng_error_code. */
2749 ret = -nb_dom;
2750 goto error;
2751 }
2752
2753 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2754 if (ret < 0) {
2755 goto setup_error;
2756 }
2757
2758 /* Copy event list into message payload */
2759 memcpy(cmd_ctx->llm->payload, domains,
2760 nb_dom * sizeof(struct lttng_domain));
2761
2762 free(domains);
2763
2764 ret = LTTNG_OK;
2765 break;
2766 }
2767 case LTTNG_LIST_CHANNELS:
2768 {
2769 int nb_chan;
2770 struct lttng_channel *channels;
2771
2772 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2773 cmd_ctx->session, &channels);
2774 if (nb_chan < 0) {
2775 /* Return value is a negative lttng_error_code. */
2776 ret = -nb_chan;
2777 goto error;
2778 }
2779
2780 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2781 if (ret < 0) {
2782 goto setup_error;
2783 }
2784
2785 /* Copy event list into message payload */
2786 memcpy(cmd_ctx->llm->payload, channels,
2787 nb_chan * sizeof(struct lttng_channel));
2788
2789 free(channels);
2790
2791 ret = LTTNG_OK;
2792 break;
2793 }
2794 case LTTNG_LIST_EVENTS:
2795 {
2796 ssize_t nb_event;
2797 struct lttng_event *events = NULL;
2798
2799 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2800 cmd_ctx->lsm->u.list.channel_name, &events);
2801 if (nb_event < 0) {
2802 /* Return value is a negative lttng_error_code. */
2803 ret = -nb_event;
2804 goto error;
2805 }
2806
2807 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2808 if (ret < 0) {
2809 goto setup_error;
2810 }
2811
2812 /* Copy event list into message payload */
2813 memcpy(cmd_ctx->llm->payload, events,
2814 nb_event * sizeof(struct lttng_event));
2815
2816 free(events);
2817
2818 ret = LTTNG_OK;
2819 break;
2820 }
2821 case LTTNG_LIST_SESSIONS:
2822 {
2823 unsigned int nr_sessions;
2824
2825 session_lock_list();
2826 nr_sessions = lttng_sessions_count(
2827 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2828 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2829
2830 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2831 if (ret < 0) {
2832 session_unlock_list();
2833 goto setup_error;
2834 }
2835
2836 /* Filled the session array */
2837 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2838 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2839 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2840
2841 session_unlock_list();
2842
2843 ret = LTTNG_OK;
2844 break;
2845 }
2846 case LTTNG_CALIBRATE:
2847 {
2848 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2849 &cmd_ctx->lsm->u.calibrate);
2850 break;
2851 }
2852 case LTTNG_REGISTER_CONSUMER:
2853 {
2854 struct consumer_data *cdata;
2855
2856 switch (cmd_ctx->lsm->domain.type) {
2857 case LTTNG_DOMAIN_KERNEL:
2858 cdata = &kconsumer_data;
2859 break;
2860 default:
2861 ret = LTTNG_ERR_UND;
2862 goto error;
2863 }
2864
2865 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2866 cmd_ctx->lsm->u.reg.path, cdata);
2867 break;
2868 }
2869 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2870 {
2871 struct lttng_filter_bytecode *bytecode;
2872
2873 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2874 ret = LTTNG_ERR_FILTER_INVAL;
2875 goto error;
2876 }
2877 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2878 ret = LTTNG_ERR_FILTER_INVAL;
2879 goto error;
2880 }
2881 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2882 if (!bytecode) {
2883 ret = LTTNG_ERR_FILTER_NOMEM;
2884 goto error;
2885 }
2886 /* Receive var. len. data */
2887 DBG("Receiving var len data from client ...");
2888 ret = lttcomm_recv_unix_sock(sock, bytecode,
2889 cmd_ctx->lsm->u.enable.bytecode_len);
2890 if (ret <= 0) {
2891 DBG("Nothing recv() from client var len data... continuing");
2892 *sock_error = 1;
2893 ret = LTTNG_ERR_FILTER_INVAL;
2894 goto error;
2895 }
2896
2897 if (bytecode->len + sizeof(*bytecode)
2898 != cmd_ctx->lsm->u.enable.bytecode_len) {
2899 free(bytecode);
2900 ret = LTTNG_ERR_FILTER_INVAL;
2901 goto error;
2902 }
2903
2904 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2905 cmd_ctx->lsm->u.enable.channel_name,
2906 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2907 break;
2908 }
2909 case LTTNG_DATA_PENDING:
2910 {
2911 ret = cmd_data_pending(cmd_ctx->session);
2912 break;
2913 }
2914 default:
2915 ret = LTTNG_ERR_UND;
2916 break;
2917 }
2918
2919 error:
2920 if (cmd_ctx->llm == NULL) {
2921 DBG("Missing llm structure. Allocating one.");
2922 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2923 goto setup_error;
2924 }
2925 }
2926 /* Set return code */
2927 cmd_ctx->llm->ret_code = ret;
2928 setup_error:
2929 if (cmd_ctx->session) {
2930 session_unlock(cmd_ctx->session);
2931 }
2932 if (need_tracing_session) {
2933 session_unlock_list();
2934 }
2935 init_setup_error:
2936 return ret;
2937 }
2938
2939 /*
2940 * Thread managing health check socket.
2941 */
2942 static void *thread_manage_health(void *data)
2943 {
2944 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2945 uint32_t revents, nb_fd;
2946 struct lttng_poll_event events;
2947 struct lttcomm_health_msg msg;
2948 struct lttcomm_health_data reply;
2949
2950 DBG("[thread] Manage health check started");
2951
2952 rcu_register_thread();
2953
2954 /* Create unix socket */
2955 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2956 if (sock < 0) {
2957 ERR("Unable to create health check Unix socket");
2958 ret = -1;
2959 goto error;
2960 }
2961
2962 /*
2963 * Set the CLOEXEC flag. Return code is useless because either way, the
2964 * show must go on.
2965 */
2966 (void) utils_set_fd_cloexec(sock);
2967
2968 ret = lttcomm_listen_unix_sock(sock);
2969 if (ret < 0) {
2970 goto error;
2971 }
2972
2973 /*
2974 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2975 * more will be added to this poll set.
2976 */
2977 ret = create_thread_poll_set(&events, 2);
2978 if (ret < 0) {
2979 goto error;
2980 }
2981
2982 /* Add the application registration socket */
2983 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2984 if (ret < 0) {
2985 goto error;
2986 }
2987
2988 while (1) {
2989 DBG("Health check ready");
2990
2991 /* Inifinite blocking call, waiting for transmission */
2992 restart:
2993 ret = lttng_poll_wait(&events, -1);
2994 if (ret < 0) {
2995 /*
2996 * Restart interrupted system call.
2997 */
2998 if (errno == EINTR) {
2999 goto restart;
3000 }
3001 goto error;
3002 }
3003
3004 nb_fd = ret;
3005
3006 for (i = 0; i < nb_fd; i++) {
3007 /* Fetch once the poll data */
3008 revents = LTTNG_POLL_GETEV(&events, i);
3009 pollfd = LTTNG_POLL_GETFD(&events, i);
3010
3011 /* Thread quit pipe has been closed. Killing thread. */
3012 ret = check_thread_quit_pipe(pollfd, revents);
3013 if (ret) {
3014 err = 0;
3015 goto exit;
3016 }
3017
3018 /* Event on the registration socket */
3019 if (pollfd == sock) {
3020 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3021 ERR("Health socket poll error");
3022 goto error;
3023 }
3024 }
3025 }
3026
3027 new_sock = lttcomm_accept_unix_sock(sock);
3028 if (new_sock < 0) {
3029 goto error;
3030 }
3031
3032 /*
3033 * Set the CLOEXEC flag. Return code is useless because either way, the
3034 * show must go on.
3035 */
3036 (void) utils_set_fd_cloexec(new_sock);
3037
3038 DBG("Receiving data from client for health...");
3039 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3040 if (ret <= 0) {
3041 DBG("Nothing recv() from client... continuing");
3042 ret = close(new_sock);
3043 if (ret) {
3044 PERROR("close");
3045 }
3046 new_sock = -1;
3047 continue;
3048 }
3049
3050 rcu_thread_online();
3051
3052 switch (msg.component) {
3053 case LTTNG_HEALTH_CMD:
3054 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
3055 break;
3056 case LTTNG_HEALTH_APP_MANAGE:
3057 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
3058 break;
3059 case LTTNG_HEALTH_APP_REG:
3060 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
3061 break;
3062 case LTTNG_HEALTH_KERNEL:
3063 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
3064 break;
3065 case LTTNG_HEALTH_CONSUMER:
3066 reply.ret_code = check_consumer_health();
3067 break;
3068 case LTTNG_HEALTH_ALL:
3069 reply.ret_code =
3070 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3071 health_check_state(HEALTH_TYPE_APP_REG) &&
3072 health_check_state(HEALTH_TYPE_CMD) &&
3073 health_check_state(HEALTH_TYPE_KERNEL) &&
3074 check_consumer_health();
3075 break;
3076 default:
3077 reply.ret_code = LTTNG_ERR_UND;
3078 break;
3079 }
3080
3081 /*
3082 * Flip ret value since 0 is a success and 1 indicates a bad health for
3083 * the client where in the sessiond it is the opposite. Again, this is
3084 * just to make things easier for us poor developer which enjoy a lot
3085 * lazyness.
3086 */
3087 if (reply.ret_code == 0 || reply.ret_code == 1) {
3088 reply.ret_code = !reply.ret_code;
3089 }
3090
3091 DBG2("Health check return value %d", reply.ret_code);
3092
3093 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3094 if (ret < 0) {
3095 ERR("Failed to send health data back to client");
3096 }
3097
3098 /* End of transmission */
3099 ret = close(new_sock);
3100 if (ret) {
3101 PERROR("close");
3102 }
3103 new_sock = -1;
3104 }
3105
3106 exit:
3107 error:
3108 if (err) {
3109 ERR("Health error occurred in %s", __func__);
3110 }
3111 DBG("Health check thread dying");
3112 unlink(health_unix_sock_path);
3113 if (sock >= 0) {
3114 ret = close(sock);
3115 if (ret) {
3116 PERROR("close");
3117 }
3118 }
3119 if (new_sock >= 0) {
3120 ret = close(new_sock);
3121 if (ret) {
3122 PERROR("close");
3123 }
3124 }
3125
3126 lttng_poll_clean(&events);
3127
3128 rcu_unregister_thread();
3129 return NULL;
3130 }
3131
3132 /*
3133 * This thread manage all clients request using the unix client socket for
3134 * communication.
3135 */
3136 static void *thread_manage_clients(void *data)
3137 {
3138 int sock = -1, ret, i, pollfd, err = -1;
3139 int sock_error;
3140 uint32_t revents, nb_fd;
3141 struct command_ctx *cmd_ctx = NULL;
3142 struct lttng_poll_event events;
3143
3144 DBG("[thread] Manage client started");
3145
3146 rcu_register_thread();
3147
3148 health_register(HEALTH_TYPE_CMD);
3149
3150 if (testpoint(thread_manage_clients)) {
3151 goto error_testpoint;
3152 }
3153
3154 health_code_update();
3155
3156 ret = lttcomm_listen_unix_sock(client_sock);
3157 if (ret < 0) {
3158 goto error_listen;
3159 }
3160
3161 /*
3162 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3163 * more will be added to this poll set.
3164 */
3165 ret = create_thread_poll_set(&events, 2);
3166 if (ret < 0) {
3167 goto error_create_poll;
3168 }
3169
3170 /* Add the application registration socket */
3171 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3172 if (ret < 0) {
3173 goto error;
3174 }
3175
3176 /*
3177 * Notify parent pid that we are ready to accept command for client side.
3178 */
3179 if (opt_sig_parent) {
3180 kill(ppid, SIGUSR1);
3181 }
3182
3183 if (testpoint(thread_manage_clients_before_loop)) {
3184 goto error;
3185 }
3186
3187 health_code_update();
3188
3189 while (1) {
3190 DBG("Accepting client command ...");
3191
3192 /* Inifinite blocking call, waiting for transmission */
3193 restart:
3194 health_poll_entry();
3195 ret = lttng_poll_wait(&events, -1);
3196 health_poll_exit();
3197 if (ret < 0) {
3198 /*
3199 * Restart interrupted system call.
3200 */
3201 if (errno == EINTR) {
3202 goto restart;
3203 }
3204 goto error;
3205 }
3206
3207 nb_fd = ret;
3208
3209 for (i = 0; i < nb_fd; i++) {
3210 /* Fetch once the poll data */
3211 revents = LTTNG_POLL_GETEV(&events, i);
3212 pollfd = LTTNG_POLL_GETFD(&events, i);
3213
3214 health_code_update();
3215
3216 /* Thread quit pipe has been closed. Killing thread. */
3217 ret = check_thread_quit_pipe(pollfd, revents);
3218 if (ret) {
3219 err = 0;
3220 goto exit;
3221 }
3222
3223 /* Event on the registration socket */
3224 if (pollfd == client_sock) {
3225 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3226 ERR("Client socket poll error");
3227 goto error;
3228 }
3229 }
3230 }
3231
3232 DBG("Wait for client response");
3233
3234 health_code_update();
3235
3236 sock = lttcomm_accept_unix_sock(client_sock);
3237 if (sock < 0) {
3238