e5a22460fc3c6fe2342bf82d015d6da84a85ffd2
[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 /* 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 /*
232 * Socket timeout for receiving and sending in seconds.
233 */
234 static int app_socket_timeout;
235
236 static
237 void setup_consumerd_path(void)
238 {
239 const char *bin, *libdir;
240
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
245 */
246 #if (CAA_BITS_PER_LONG == 32)
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253 #elif (CAA_BITS_PER_LONG == 64)
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
256 }
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
259 }
260 #else
261 #error "Unknown bitness"
262 #endif
263
264 /*
265 * runtime env. var. overrides the build default.
266 */
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
270 }
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
274 }
275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
280 if (libdir) {
281 consumerd64_libdir = libdir;
282 }
283 }
284
285 /*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288 static int create_thread_poll_set(struct lttng_poll_event *events,
289 unsigned int size)
290 {
291 int ret;
292
293 if (events == NULL || size == 0) {
294 ret = -1;
295 goto error;
296 }
297
298 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
299 if (ret < 0) {
300 goto error;
301 }
302
303 /* Add quit pipe */
304 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
305 if (ret < 0) {
306 goto error;
307 }
308
309 return 0;
310
311 error:
312 return ret;
313 }
314
315 /*
316 * Check if the thread quit pipe was triggered.
317 *
318 * Return 1 if it was triggered else 0;
319 */
320 static int check_thread_quit_pipe(int fd, uint32_t events)
321 {
322 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
323 return 1;
324 }
325
326 return 0;
327 }
328
329 /*
330 * Return group ID of the tracing group or -1 if not found.
331 */
332 static gid_t allowed_group(void)
333 {
334 struct group *grp;
335
336 if (opt_tracing_group) {
337 grp = getgrnam(opt_tracing_group);
338 } else {
339 grp = getgrnam(default_tracing_group);
340 }
341 if (!grp) {
342 return -1;
343 } else {
344 return grp->gr_gid;
345 }
346 }
347
348 /*
349 * Init thread quit pipe.
350 *
351 * Return -1 on error or 0 if all pipes are created.
352 */
353 static int init_thread_quit_pipe(void)
354 {
355 int ret, i;
356
357 ret = pipe(thread_quit_pipe);
358 if (ret < 0) {
359 PERROR("thread quit pipe");
360 goto error;
361 }
362
363 for (i = 0; i < 2; i++) {
364 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
365 if (ret < 0) {
366 PERROR("fcntl");
367 goto error;
368 }
369 }
370
371 error:
372 return ret;
373 }
374
375 /*
376 * Stop all threads by closing the thread quit pipe.
377 */
378 static void stop_threads(void)
379 {
380 int ret;
381
382 /* Stopping all threads */
383 DBG("Terminating all threads");
384 ret = notify_thread_pipe(thread_quit_pipe[1]);
385 if (ret < 0) {
386 ERR("write error on thread quit pipe");
387 }
388
389 /* Dispatch thread */
390 CMM_STORE_SHARED(dispatch_thread_exit, 1);
391 futex_nto1_wake(&ust_cmd_queue.futex);
392 }
393
394 /*
395 * Cleanup the daemon
396 */
397 static void cleanup(void)
398 {
399 int ret;
400 char *cmd = NULL;
401 struct ltt_session *sess, *stmp;
402
403 DBG("Cleaning up");
404
405 /* First thing first, stop all threads */
406 utils_close_pipe(thread_quit_pipe);
407
408 /*
409 * If opt_pidfile is undefined, the default file will be wiped when
410 * removing the rundir.
411 */
412 if (opt_pidfile) {
413 ret = remove(opt_pidfile);
414 if (ret < 0) {
415 PERROR("remove pidfile %s", opt_pidfile);
416 }
417 }
418
419 DBG("Removing %s directory", rundir);
420 ret = asprintf(&cmd, "rm -rf %s", rundir);
421 if (ret < 0) {
422 ERR("asprintf failed. Something is really wrong!");
423 }
424
425 /* Remove lttng run directory */
426 ret = system(cmd);
427 if (ret < 0) {
428 ERR("Unable to clean %s", rundir);
429 }
430 free(cmd);
431 free(rundir);
432
433 DBG("Cleaning up all sessions");
434
435 /* Destroy session list mutex */
436 if (session_list_ptr != NULL) {
437 pthread_mutex_destroy(&session_list_ptr->lock);
438
439 /* Cleanup ALL session */
440 cds_list_for_each_entry_safe(sess, stmp,
441 &session_list_ptr->head, list) {
442 cmd_destroy_session(sess, kernel_poll_pipe[1]);
443 }
444 }
445
446 DBG("Closing all UST sockets");
447 ust_app_clean_list();
448
449 if (is_root && !opt_no_kernel) {
450 DBG2("Closing kernel fd");
451 if (kernel_tracer_fd >= 0) {
452 ret = close(kernel_tracer_fd);
453 if (ret) {
454 PERROR("close");
455 }
456 }
457 DBG("Unloading kernel modules");
458 modprobe_remove_lttng_all();
459 }
460
461 /* <fun> */
462 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
463 "Matthew, BEET driven development works!%c[%dm",
464 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
465 /* </fun> */
466 }
467
468 /*
469 * Send data on a unix socket using the liblttsessiondcomm API.
470 *
471 * Return lttcomm error code.
472 */
473 static int send_unix_sock(int sock, void *buf, size_t len)
474 {
475 /* Check valid length */
476 if (len == 0) {
477 return -1;
478 }
479
480 return lttcomm_send_unix_sock(sock, buf, len);
481 }
482
483 /*
484 * Free memory of a command context structure.
485 */
486 static void clean_command_ctx(struct command_ctx **cmd_ctx)
487 {
488 DBG("Clean command context structure");
489 if (*cmd_ctx) {
490 if ((*cmd_ctx)->llm) {
491 free((*cmd_ctx)->llm);
492 }
493 if ((*cmd_ctx)->lsm) {
494 free((*cmd_ctx)->lsm);
495 }
496 free(*cmd_ctx);
497 *cmd_ctx = NULL;
498 }
499 }
500
501 /*
502 * Notify UST applications using the shm mmap futex.
503 */
504 static int notify_ust_apps(int active)
505 {
506 char *wait_shm_mmap;
507
508 DBG("Notifying applications of session daemon state: %d", active);
509
510 /* See shm.c for this call implying mmap, shm and futex calls */
511 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
512 if (wait_shm_mmap == NULL) {
513 goto error;
514 }
515
516 /* Wake waiting process */
517 futex_wait_update((int32_t *) wait_shm_mmap, active);
518
519 /* Apps notified successfully */
520 return 0;
521
522 error:
523 return -1;
524 }
525
526 /*
527 * Setup the outgoing data buffer for the response (llm) by allocating the
528 * right amount of memory and copying the original information from the lsm
529 * structure.
530 *
531 * Return total size of the buffer pointed by buf.
532 */
533 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
534 {
535 int ret, buf_size;
536
537 buf_size = size;
538
539 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
540 if (cmd_ctx->llm == NULL) {
541 PERROR("zmalloc");
542 ret = -ENOMEM;
543 goto error;
544 }
545
546 /* Copy common data */
547 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
548 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
549
550 cmd_ctx->llm->data_size = size;
551 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
552
553 return buf_size;
554
555 error:
556 return ret;
557 }
558
559 /*
560 * Update the kernel poll set of all channel fd available over all tracing
561 * session. Add the wakeup pipe at the end of the set.
562 */
563 static int update_kernel_poll(struct lttng_poll_event *events)
564 {
565 int ret;
566 struct ltt_session *session;
567 struct ltt_kernel_channel *channel;
568
569 DBG("Updating kernel poll set");
570
571 session_lock_list();
572 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
573 session_lock(session);
574 if (session->kernel_session == NULL) {
575 session_unlock(session);
576 continue;
577 }
578
579 cds_list_for_each_entry(channel,
580 &session->kernel_session->channel_list.head, list) {
581 /* Add channel fd to the kernel poll set */
582 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
583 if (ret < 0) {
584 session_unlock(session);
585 goto error;
586 }
587 DBG("Channel fd %d added to kernel set", channel->fd);
588 }
589 session_unlock(session);
590 }
591 session_unlock_list();
592
593 return 0;
594
595 error:
596 session_unlock_list();
597 return -1;
598 }
599
600 /*
601 * Find the channel fd from 'fd' over all tracing session. When found, check
602 * for new channel stream and send those stream fds to the kernel consumer.
603 *
604 * Useful for CPU hotplug feature.
605 */
606 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
607 {
608 int ret = 0;
609 struct ltt_session *session;
610 struct ltt_kernel_session *ksess;
611 struct ltt_kernel_channel *channel;
612
613 DBG("Updating kernel streams for channel fd %d", fd);
614
615 session_lock_list();
616 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
617 session_lock(session);
618 if (session->kernel_session == NULL) {
619 session_unlock(session);
620 continue;
621 }
622 ksess = session->kernel_session;
623
624 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
625 if (channel->fd == fd) {
626 DBG("Channel found, updating kernel streams");
627 ret = kernel_open_channel_stream(channel);
628 if (ret < 0) {
629 goto error;
630 }
631
632 /*
633 * Have we already sent fds to the consumer? If yes, it means
634 * that tracing is started so it is safe to send our updated
635 * stream fds.
636 */
637 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
638 struct lttng_ht_iter iter;
639 struct consumer_socket *socket;
640
641
642 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
643 &iter.iter, socket, node.node) {
644 /* Code flow error */
645 assert(socket->fd >= 0);
646
647 pthread_mutex_lock(socket->lock);
648 ret = kernel_consumer_send_channel_stream(socket,
649 channel, ksess);
650 pthread_mutex_unlock(socket->lock);
651 if (ret < 0) {
652 goto error;
653 }
654 }
655 }
656 goto error;
657 }
658 }
659 session_unlock(session);
660 }
661 session_unlock_list();
662 return ret;
663
664 error:
665 session_unlock(session);
666 session_unlock_list();
667 return ret;
668 }
669
670 /*
671 * For each tracing session, update newly registered apps.
672 */
673 static void update_ust_app(int app_sock)
674 {
675 struct ltt_session *sess, *stmp;
676
677 session_lock_list();
678
679 /* For all tracing session(s) */
680 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
681 session_lock(sess);
682 if (sess->ust_session) {
683 ust_app_global_update(sess->ust_session, app_sock);
684 }
685 session_unlock(sess);
686 }
687
688 session_unlock_list();
689 }
690
691 /*
692 * This thread manage event coming from the kernel.
693 *
694 * Features supported in this thread:
695 * -) CPU Hotplug
696 */
697 static void *thread_manage_kernel(void *data)
698 {
699 int ret, i, pollfd, update_poll_flag = 1, err = -1;
700 uint32_t revents, nb_fd;
701 char tmp;
702 struct lttng_poll_event events;
703
704 DBG("[thread] Thread manage kernel started");
705
706 /*
707 * This first step of the while is to clean this structure which could free
708 * non NULL pointers so zero it before the loop.
709 */
710 memset(&events, 0, sizeof(events));
711
712 if (testpoint(thread_manage_kernel)) {
713 goto error_testpoint;
714 }
715
716 health_code_update(&health_thread_kernel);
717
718 if (testpoint(thread_manage_kernel_before_loop)) {
719 goto error_testpoint;
720 }
721
722 while (1) {
723 health_code_update(&health_thread_kernel);
724
725 if (update_poll_flag == 1) {
726 /* Clean events object. We are about to populate it again. */
727 lttng_poll_clean(&events);
728
729 ret = create_thread_poll_set(&events, 2);
730 if (ret < 0) {
731 goto error_poll_create;
732 }
733
734 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
735 if (ret < 0) {
736 goto error;
737 }
738
739 /* This will add the available kernel channel if any. */
740 ret = update_kernel_poll(&events);
741 if (ret < 0) {
742 goto error;
743 }
744 update_poll_flag = 0;
745 }
746
747 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
748
749 /* Poll infinite value of time */
750 restart:
751 health_poll_update(&health_thread_kernel);
752 ret = lttng_poll_wait(&events, -1);
753 health_poll_update(&health_thread_kernel);
754 if (ret < 0) {
755 /*
756 * Restart interrupted system call.
757 */
758 if (errno == EINTR) {
759 goto restart;
760 }
761 goto error;
762 } else if (ret == 0) {
763 /* Should not happen since timeout is infinite */
764 ERR("Return value of poll is 0 with an infinite timeout.\n"
765 "This should not have happened! Continuing...");
766 continue;
767 }
768
769 nb_fd = ret;
770
771 for (i = 0; i < nb_fd; i++) {
772 /* Fetch once the poll data */
773 revents = LTTNG_POLL_GETEV(&events, i);
774 pollfd = LTTNG_POLL_GETFD(&events, i);
775
776 health_code_update(&health_thread_kernel);
777
778 /* Thread quit pipe has been closed. Killing thread. */
779 ret = check_thread_quit_pipe(pollfd, revents);
780 if (ret) {
781 err = 0;
782 goto exit;
783 }
784
785 /* Check for data on kernel pipe */
786 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
787 do {
788 ret = read(kernel_poll_pipe[0], &tmp, 1);
789 } while (ret < 0 && errno == EINTR);
790 /*
791 * Ret value is useless here, if this pipe gets any actions an
792 * update is required anyway.
793 */
794 update_poll_flag = 1;
795 continue;
796 } else {
797 /*
798 * New CPU detected by the kernel. Adding kernel stream to
799 * kernel session and updating the kernel consumer
800 */
801 if (revents & LPOLLIN) {
802 ret = update_kernel_stream(&kconsumer_data, pollfd);
803 if (ret < 0) {
804 continue;
805 }
806 break;
807 /*
808 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
809 * and unregister kernel stream at this point.
810 */
811 }
812 }
813 }
814 }
815
816 exit:
817 error:
818 lttng_poll_clean(&events);
819 error_poll_create:
820 error_testpoint:
821 utils_close_pipe(kernel_poll_pipe);
822 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
823 if (err) {
824 health_error(&health_thread_kernel);
825 ERR("Health error occurred in %s", __func__);
826 WARN("Kernel thread died unexpectedly. "
827 "Kernel tracing can continue but CPU hotplug is disabled.");
828 }
829 health_exit(&health_thread_kernel);
830 DBG("Kernel thread dying");
831 return NULL;
832 }
833
834 /*
835 * Signal pthread condition of the consumer data that the thread.
836 */
837 static void signal_consumer_condition(struct consumer_data *data, int state)
838 {
839 pthread_mutex_lock(&data->cond_mutex);
840
841 /*
842 * The state is set before signaling. It can be any value, it's the waiter
843 * job to correctly interpret this condition variable associated to the
844 * consumer pthread_cond.
845 *
846 * A value of 0 means that the corresponding thread of the consumer data
847 * was not started. 1 indicates that the thread has started and is ready
848 * for action. A negative value means that there was an error during the
849 * thread bootstrap.
850 */
851 data->consumer_thread_is_ready = state;
852 (void) pthread_cond_signal(&data->cond);
853
854 pthread_mutex_unlock(&data->cond_mutex);
855 }
856
857 /*
858 * This thread manage the consumer error sent back to the session daemon.
859 */
860 static void *thread_manage_consumer(void *data)
861 {
862 int sock = -1, i, ret, pollfd, err = -1;
863 uint32_t revents, nb_fd;
864 enum lttcomm_return_code code;
865 struct lttng_poll_event events;
866 struct consumer_data *consumer_data = data;
867
868 DBG("[thread] Manage consumer started");
869
870 /*
871 * Since the consumer thread can be spawned at any moment in time, we init
872 * the health to a poll status (1, which is a valid health over time).
873 * When the thread starts, we update here the health to a "code" path being
874 * an even value so this thread, when reaching a poll wait, does not
875 * trigger an error with an even value.
876 *
877 * Here is the use case we avoid.
878 *
879 * +1: the first poll update during initialization (main())
880 * +2 * x: multiple code update once in this thread.
881 * +1: poll wait in this thread (being a good health state).
882 * == even number which after the wait period shows as a bad health.
883 *
884 * In a nutshell, the following poll update to the health state brings back
885 * the state to an even value meaning a code path.
886 */
887 health_poll_update(&consumer_data->health);
888
889 /*
890 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
891 * Nothing more will be added to this poll set.
892 */
893 ret = create_thread_poll_set(&events, 2);
894 if (ret < 0) {
895 goto error_poll;
896 }
897
898 /*
899 * The error socket here is already in a listening state which was done
900 * just before spawning this thread to avoid a race between the consumer
901 * daemon exec trying to connect and the listen() call.
902 */
903 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
904 if (ret < 0) {
905 goto error;
906 }
907
908 health_code_update(&consumer_data->health);
909
910 /* Inifinite blocking call, waiting for transmission */
911 restart:
912 health_poll_update(&consumer_data->health);
913
914 if (testpoint(thread_manage_consumer)) {
915 goto error;
916 }
917
918 ret = lttng_poll_wait(&events, -1);
919 health_poll_update(&consumer_data->health);
920 if (ret < 0) {
921 /*
922 * Restart interrupted system call.
923 */
924 if (errno == EINTR) {
925 goto restart;
926 }
927 goto error;
928 }
929
930 nb_fd = ret;
931
932 for (i = 0; i < nb_fd; i++) {
933 /* Fetch once the poll data */
934 revents = LTTNG_POLL_GETEV(&events, i);
935 pollfd = LTTNG_POLL_GETFD(&events, i);
936
937 health_code_update(&consumer_data->health);
938
939 /* Thread quit pipe has been closed. Killing thread. */
940 ret = check_thread_quit_pipe(pollfd, revents);
941 if (ret) {
942 err = 0;
943 goto exit;
944 }
945
946 /* Event on the registration socket */
947 if (pollfd == consumer_data->err_sock) {
948 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
949 ERR("consumer err socket poll error");
950 goto error;
951 }
952 }
953 }
954
955 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
956 if (sock < 0) {
957 goto error;
958 }
959
960 /*
961 * Set the CLOEXEC flag. Return code is useless because either way, the
962 * show must go on.
963 */
964 (void) utils_set_fd_cloexec(sock);
965
966 health_code_update(&consumer_data->health);
967
968 DBG2("Receiving code from consumer err_sock");
969
970 /* Getting status code from kconsumerd */
971 ret = lttcomm_recv_unix_sock(sock, &code,
972 sizeof(enum lttcomm_return_code));
973 if (ret <= 0) {
974 goto error;
975 }
976
977 health_code_update(&consumer_data->health);
978
979 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
980 consumer_data->cmd_sock =
981 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
982 if (consumer_data->cmd_sock < 0) {
983 /* On error, signal condition and quit. */
984 signal_consumer_condition(consumer_data, -1);
985 PERROR("consumer connect");
986 goto error;
987 }
988 signal_consumer_condition(consumer_data, 1);
989 DBG("Consumer command socket ready");
990 } else {
991 ERR("consumer error when waiting for SOCK_READY : %s",
992 lttcomm_get_readable_code(-code));
993 goto error;
994 }
995
996 /* Remove the kconsumerd error sock since we've established a connexion */
997 ret = lttng_poll_del(&events, consumer_data->err_sock);
998 if (ret < 0) {
999 goto error;
1000 }
1001
1002 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1003 if (ret < 0) {
1004 goto error;
1005 }
1006
1007 health_code_update(&consumer_data->health);
1008
1009 /* Inifinite blocking call, waiting for transmission */
1010 restart_poll:
1011 health_poll_update(&consumer_data->health);
1012 ret = lttng_poll_wait(&events, -1);
1013 health_poll_update(&consumer_data->health);
1014 if (ret < 0) {
1015 /*
1016 * Restart interrupted system call.
1017 */
1018 if (errno == EINTR) {
1019 goto restart_poll;
1020 }
1021 goto error;
1022 }
1023
1024 nb_fd = ret;
1025
1026 for (i = 0; i < nb_fd; i++) {
1027 /* Fetch once the poll data */
1028 revents = LTTNG_POLL_GETEV(&events, i);
1029 pollfd = LTTNG_POLL_GETFD(&events, i);
1030
1031 health_code_update(&consumer_data->health);
1032
1033 /* Thread quit pipe has been closed. Killing thread. */
1034 ret = check_thread_quit_pipe(pollfd, revents);
1035 if (ret) {
1036 err = 0;
1037 goto exit;
1038 }
1039
1040 /* Event on the kconsumerd socket */
1041 if (pollfd == sock) {
1042 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1043 ERR("consumer err socket second poll error");
1044 goto error;
1045 }
1046 }
1047 }
1048
1049 health_code_update(&consumer_data->health);
1050
1051 /* Wait for any kconsumerd error */
1052 ret = lttcomm_recv_unix_sock(sock, &code,
1053 sizeof(enum lttcomm_return_code));
1054 if (ret <= 0) {
1055 ERR("consumer closed the command socket");
1056 goto error;
1057 }
1058
1059 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
1060
1061 exit:
1062 error:
1063 /* Immediately set the consumerd state to stopped */
1064 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1065 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1066 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1067 consumer_data->type == LTTNG_CONSUMER32_UST) {
1068 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1069 } else {
1070 /* Code flow error... */
1071 assert(0);
1072 }
1073
1074 if (consumer_data->err_sock >= 0) {
1075 ret = close(consumer_data->err_sock);
1076 if (ret) {
1077 PERROR("close");
1078 }
1079 }
1080 if (consumer_data->cmd_sock >= 0) {
1081 ret = close(consumer_data->cmd_sock);
1082 if (ret) {
1083 PERROR("close");
1084 }
1085 }
1086 if (sock >= 0) {
1087 ret = close(sock);
1088 if (ret) {
1089 PERROR("close");
1090 }
1091 }
1092
1093 unlink(consumer_data->err_unix_sock_path);
1094 unlink(consumer_data->cmd_unix_sock_path);
1095 consumer_data->pid = 0;
1096
1097 lttng_poll_clean(&events);
1098 error_poll:
1099 if (err) {
1100 health_error(&consumer_data->health);
1101 ERR("Health error occurred in %s", __func__);
1102 }
1103 health_exit(&consumer_data->health);
1104 DBG("consumer thread cleanup completed");
1105
1106 return NULL;
1107 }
1108
1109 /*
1110 * This thread manage application communication.
1111 */
1112 static void *thread_manage_apps(void *data)
1113 {
1114 int i, ret, pollfd, err = -1;
1115 uint32_t revents, nb_fd;
1116 struct ust_command ust_cmd;
1117 struct lttng_poll_event events;
1118
1119 DBG("[thread] Manage application started");
1120
1121 rcu_register_thread();
1122 rcu_thread_online();
1123
1124 if (testpoint(thread_manage_apps)) {
1125 goto error_testpoint;
1126 }
1127
1128 health_code_update(&health_thread_app_manage);
1129
1130 ret = create_thread_poll_set(&events, 2);
1131 if (ret < 0) {
1132 goto error_poll_create;
1133 }
1134
1135 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1136 if (ret < 0) {
1137 goto error;
1138 }
1139
1140 if (testpoint(thread_manage_apps_before_loop)) {
1141 goto error;
1142 }
1143
1144 health_code_update(&health_thread_app_manage);
1145
1146 while (1) {
1147 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1148
1149 /* Inifinite blocking call, waiting for transmission */
1150 restart:
1151 health_poll_update(&health_thread_app_manage);
1152 ret = lttng_poll_wait(&events, -1);
1153 health_poll_update(&health_thread_app_manage);
1154 if (ret < 0) {
1155 /*
1156 * Restart interrupted system call.
1157 */
1158 if (errno == EINTR) {
1159 goto restart;
1160 }
1161 goto error;
1162 }
1163
1164 nb_fd = ret;
1165
1166 for (i = 0; i < nb_fd; i++) {
1167 /* Fetch once the poll data */
1168 revents = LTTNG_POLL_GETEV(&events, i);
1169 pollfd = LTTNG_POLL_GETFD(&events, i);
1170
1171 health_code_update(&health_thread_app_manage);
1172
1173 /* Thread quit pipe has been closed. Killing thread. */
1174 ret = check_thread_quit_pipe(pollfd, revents);
1175 if (ret) {
1176 err = 0;
1177 goto exit;
1178 }
1179
1180 /* Inspect the apps cmd pipe */
1181 if (pollfd == apps_cmd_pipe[0]) {
1182 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1183 ERR("Apps command pipe error");
1184 goto error;
1185 } else if (revents & LPOLLIN) {
1186 /* Empty pipe */
1187 do {
1188 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1189 } while (ret < 0 && errno == EINTR);
1190 if (ret < 0 || ret < sizeof(ust_cmd)) {
1191 PERROR("read apps cmd pipe");
1192 goto error;
1193 }
1194
1195 health_code_update(&health_thread_app_manage);
1196
1197 /* Register applicaton to the session daemon */
1198 ret = ust_app_register(&ust_cmd.reg_msg,
1199 ust_cmd.sock);
1200 if (ret == -ENOMEM) {
1201 goto error;
1202 } else if (ret < 0) {
1203 break;
1204 }
1205
1206 health_code_update(&health_thread_app_manage);
1207
1208 /*
1209 * Validate UST version compatibility.
1210 */
1211 ret = ust_app_validate_version(ust_cmd.sock);
1212 if (ret >= 0) {
1213 /*
1214 * Add channel(s) and event(s) to newly registered apps
1215 * from lttng global UST domain.
1216 */
1217 update_ust_app(ust_cmd.sock);
1218 }
1219
1220 health_code_update(&health_thread_app_manage);
1221
1222 ret = ust_app_register_done(ust_cmd.sock);
1223 if (ret < 0) {
1224 /*
1225 * If the registration is not possible, we simply
1226 * unregister the apps and continue
1227 */
1228 ust_app_unregister(ust_cmd.sock);
1229 } else {
1230 /*
1231 * We only monitor the error events of the socket. This
1232 * thread does not handle any incoming data from UST
1233 * (POLLIN).
1234 */
1235 ret = lttng_poll_add(&events, ust_cmd.sock,
1236 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
1237 if (ret < 0) {
1238 goto error;
1239 }
1240
1241 /* Set socket timeout for both receiving and ending */
1242 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1243 app_socket_timeout);
1244 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1245 app_socket_timeout);
1246
1247 DBG("Apps with sock %d added to poll set",
1248 ust_cmd.sock);
1249 }
1250
1251 health_code_update(&health_thread_app_manage);
1252
1253 break;
1254 }
1255 } else {
1256 /*
1257 * At this point, we know that a registered application made
1258 * the event at poll_wait.
1259 */
1260 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1261 /* Removing from the poll set */
1262 ret = lttng_poll_del(&events, pollfd);
1263 if (ret < 0) {
1264 goto error;
1265 }
1266
1267 /* Socket closed on remote end. */
1268 ust_app_unregister(pollfd);
1269 break;
1270 }
1271 }
1272
1273 health_code_update(&health_thread_app_manage);
1274 }
1275 }
1276
1277 exit:
1278 error:
1279 lttng_poll_clean(&events);
1280 error_poll_create:
1281 error_testpoint:
1282 utils_close_pipe(apps_cmd_pipe);
1283 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1284
1285 /*
1286 * We don't clean the UST app hash table here since already registered
1287 * applications can still be controlled so let them be until the session
1288 * daemon dies or the applications stop.
1289 */
1290
1291 if (err) {
1292 health_error(&health_thread_app_manage);
1293 ERR("Health error occurred in %s", __func__);
1294 }
1295 health_exit(&health_thread_app_manage);
1296 DBG("Application communication apps thread cleanup complete");
1297 rcu_thread_offline();
1298 rcu_unregister_thread();
1299 return NULL;
1300 }
1301
1302 /*
1303 * Dispatch request from the registration threads to the application
1304 * communication thread.
1305 */
1306 static void *thread_dispatch_ust_registration(void *data)
1307 {
1308 int ret;
1309 struct cds_wfq_node *node;
1310 struct ust_command *ust_cmd = NULL;
1311
1312 DBG("[thread] Dispatch UST command started");
1313
1314 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1315 /* Atomically prepare the queue futex */
1316 futex_nto1_prepare(&ust_cmd_queue.futex);
1317
1318 do {
1319 /* Dequeue command for registration */
1320 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1321 if (node == NULL) {
1322 DBG("Woken up but nothing in the UST command queue");
1323 /* Continue thread execution */
1324 break;
1325 }
1326
1327 ust_cmd = caa_container_of(node, struct ust_command, node);
1328
1329 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1330 " gid:%d sock:%d name:%s (version %d.%d)",
1331 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1332 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1333 ust_cmd->sock, ust_cmd->reg_msg.name,
1334 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1335 /*
1336 * Inform apps thread of the new application registration. This
1337 * call is blocking so we can be assured that the data will be read
1338 * at some point in time or wait to the end of the world :)
1339 */
1340 if (apps_cmd_pipe[1] >= 0) {
1341 do {
1342 ret = write(apps_cmd_pipe[1], ust_cmd,
1343 sizeof(struct ust_command));
1344 } while (ret < 0 && errno == EINTR);
1345 if (ret < 0 || ret != sizeof(struct ust_command)) {
1346 PERROR("write apps cmd pipe");
1347 if (errno == EBADF) {
1348 /*
1349 * We can't inform the application thread to process
1350 * registration. We will exit or else application
1351 * registration will not occur and tracing will never
1352 * start.
1353 */
1354 goto error;
1355 }
1356 }
1357 } else {
1358 /* Application manager thread is not available. */
1359 ret = close(ust_cmd->sock);
1360 if (ret < 0) {
1361 PERROR("close ust_cmd sock");
1362 }
1363 }
1364 free(ust_cmd);
1365 } while (node != NULL);
1366
1367 /* Futex wait on queue. Blocking call on futex() */
1368 futex_nto1_wait(&ust_cmd_queue.futex);
1369 }
1370
1371 error:
1372 DBG("Dispatch thread dying");
1373 return NULL;
1374 }
1375
1376 /*
1377 * This thread manage application registration.
1378 */
1379 static void *thread_registration_apps(void *data)
1380 {
1381 int sock = -1, i, ret, pollfd, err = -1;
1382 uint32_t revents, nb_fd;
1383 struct lttng_poll_event events;
1384 /*
1385 * Get allocated in this thread, enqueued to a global queue, dequeued and
1386 * freed in the manage apps thread.
1387 */
1388 struct ust_command *ust_cmd = NULL;
1389
1390 DBG("[thread] Manage application registration started");
1391
1392 if (testpoint(thread_registration_apps)) {
1393 goto error_testpoint;
1394 }
1395
1396 ret = lttcomm_listen_unix_sock(apps_sock);
1397 if (ret < 0) {
1398 goto error_listen;
1399 }
1400
1401 /*
1402 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1403 * more will be added to this poll set.
1404 */
1405 ret = create_thread_poll_set(&events, 2);
1406 if (ret < 0) {
1407 goto error_create_poll;
1408 }
1409
1410 /* Add the application registration socket */
1411 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1412 if (ret < 0) {
1413 goto error_poll_add;
1414 }
1415
1416 /* Notify all applications to register */
1417 ret = notify_ust_apps(1);
1418 if (ret < 0) {
1419 ERR("Failed to notify applications or create the wait shared memory.\n"
1420 "Execution continues but there might be problem for already\n"
1421 "running applications that wishes to register.");
1422 }
1423
1424 while (1) {
1425 DBG("Accepting application registration");
1426
1427 /* Inifinite blocking call, waiting for transmission */
1428 restart:
1429 health_poll_update(&health_thread_app_reg);
1430 ret = lttng_poll_wait(&events, -1);
1431 health_poll_update(&health_thread_app_reg);
1432 if (ret < 0) {
1433 /*
1434 * Restart interrupted system call.
1435 */
1436 if (errno == EINTR) {
1437 goto restart;
1438 }
1439 goto error;
1440 }
1441
1442 nb_fd = ret;
1443
1444 for (i = 0; i < nb_fd; i++) {
1445 health_code_update(&health_thread_app_reg);
1446
1447 /* Fetch once the poll data */
1448 revents = LTTNG_POLL_GETEV(&events, i);
1449 pollfd = LTTNG_POLL_GETFD(&events, i);
1450
1451 /* Thread quit pipe has been closed. Killing thread. */
1452 ret = check_thread_quit_pipe(pollfd, revents);
1453 if (ret) {
1454 err = 0;
1455 goto exit;
1456 }
1457
1458 /* Event on the registration socket */
1459 if (pollfd == apps_sock) {
1460 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1461 ERR("Register apps socket poll error");
1462 goto error;
1463 } else if (revents & LPOLLIN) {
1464 sock = lttcomm_accept_unix_sock(apps_sock);
1465 if (sock < 0) {
1466 goto error;
1467 }
1468
1469 /*
1470 * Set the CLOEXEC flag. Return code is useless because
1471 * either way, the show must go on.
1472 */
1473 (void) utils_set_fd_cloexec(sock);
1474
1475 /* Create UST registration command for enqueuing */
1476 ust_cmd = zmalloc(sizeof(struct ust_command));
1477 if (ust_cmd == NULL) {
1478 PERROR("ust command zmalloc");
1479 goto error;
1480 }
1481
1482 /*
1483 * Using message-based transmissions to ensure we don't
1484 * have to deal with partially received messages.
1485 */
1486 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1487 if (ret < 0) {
1488 ERR("Exhausted file descriptors allowed for applications.");
1489 free(ust_cmd);
1490 ret = close(sock);
1491 if (ret) {
1492 PERROR("close");
1493 }
1494 sock = -1;
1495 continue;
1496 }
1497 health_code_update(&health_thread_app_reg);
1498 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1499 sizeof(struct ust_register_msg));
1500 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1501 if (ret < 0) {
1502 PERROR("lttcomm_recv_unix_sock register apps");
1503 } else {
1504 ERR("Wrong size received on apps register");
1505 }
1506 free(ust_cmd);
1507 ret = close(sock);
1508 if (ret) {
1509 PERROR("close");
1510 }
1511 lttng_fd_put(LTTNG_FD_APPS, 1);
1512 sock = -1;
1513 continue;
1514 }
1515 health_code_update(&health_thread_app_reg);
1516
1517 ust_cmd->sock = sock;
1518 sock = -1;
1519
1520 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1521 " gid:%d sock:%d name:%s (version %d.%d)",
1522 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1523 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1524 ust_cmd->sock, ust_cmd->reg_msg.name,
1525 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1526
1527 /*
1528 * Lock free enqueue the registration request. The red pill
1529 * has been taken! This apps will be part of the *system*.
1530 */
1531 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1532
1533 /*
1534 * Wake the registration queue futex. Implicit memory
1535 * barrier with the exchange in cds_wfq_enqueue.
1536 */
1537 futex_nto1_wake(&ust_cmd_queue.futex);
1538 }
1539 }
1540 }
1541 }
1542
1543 exit:
1544 error:
1545 if (err) {
1546 health_error(&health_thread_app_reg);
1547 ERR("Health error occurred in %s", __func__);
1548 }
1549
1550 /* Notify that the registration thread is gone */
1551 notify_ust_apps(0);
1552
1553 if (apps_sock >= 0) {
1554 ret = close(apps_sock);
1555 if (ret) {
1556 PERROR("close");
1557 }
1558 }
1559 if (sock >= 0) {
1560 ret = close(sock);
1561 if (ret) {
1562 PERROR("close");
1563 }
1564 lttng_fd_put(LTTNG_FD_APPS, 1);
1565 }
1566 unlink(apps_unix_sock_path);
1567
1568 error_poll_add:
1569 lttng_poll_clean(&events);
1570 error_listen:
1571 error_create_poll:
1572 error_testpoint:
1573 DBG("UST Registration thread cleanup complete");
1574 health_exit(&health_thread_app_reg);
1575
1576 return NULL;
1577 }
1578
1579 /*
1580 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1581 * exec or it will fails.
1582 */
1583 static int spawn_consumer_thread(struct consumer_data *consumer_data)
1584 {
1585 int ret, clock_ret;
1586 struct timespec timeout;
1587
1588 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1589 consumer_data->consumer_thread_is_ready = 0;
1590
1591 /* Setup pthread condition */
1592 ret = pthread_condattr_init(&consumer_data->condattr);
1593 if (ret != 0) {
1594 errno = ret;
1595 PERROR("pthread_condattr_init consumer data");
1596 goto error;
1597 }
1598
1599 /*
1600 * Set the monotonic clock in order to make sure we DO NOT jump in time
1601 * between the clock_gettime() call and the timedwait call. See bug #324
1602 * for a more details and how we noticed it.
1603 */
1604 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1605 if (ret != 0) {
1606 errno = ret;
1607 PERROR("pthread_condattr_setclock consumer data");
1608 goto error;
1609 }
1610
1611 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1612 if (ret != 0) {
1613 errno = ret;
1614 PERROR("pthread_cond_init consumer data");
1615 goto error;
1616 }
1617
1618 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1619 consumer_data);
1620 if (ret != 0) {
1621 PERROR("pthread_create consumer");
1622 ret = -1;
1623 goto error;
1624 }
1625
1626 /* We are about to wait on a pthread condition */
1627 pthread_mutex_lock(&consumer_data->cond_mutex);
1628
1629 /* Get time for sem_timedwait absolute timeout */
1630 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1631 /*
1632 * Set the timeout for the condition timed wait even if the clock gettime
1633 * call fails since we might loop on that call and we want to avoid to
1634 * increment the timeout too many times.
1635 */
1636 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1637
1638 /*
1639 * The following loop COULD be skipped in some conditions so this is why we
1640 * set ret to 0 in order to make sure at least one round of the loop is
1641 * done.
1642 */
1643 ret = 0;
1644
1645 /*
1646 * Loop until the condition is reached or when a timeout is reached. Note
1647 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1648 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1649 * possible. This loop does not take any chances and works with both of
1650 * them.
1651 */
1652 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1653 if (clock_ret < 0) {
1654 PERROR("clock_gettime spawn consumer");
1655 /* Infinite wait for the consumerd thread to be ready */
1656 ret = pthread_cond_wait(&consumer_data->cond,
1657 &consumer_data->cond_mutex);
1658 } else {
1659 ret = pthread_cond_timedwait(&consumer_data->cond,
1660 &consumer_data->cond_mutex, &timeout);
1661 }
1662 }
1663
1664 /* Release the pthread condition */
1665 pthread_mutex_unlock(&consumer_data->cond_mutex);
1666
1667 if (ret != 0) {
1668 errno = ret;
1669 if (ret == ETIMEDOUT) {
1670 /*
1671 * Call has timed out so we kill the kconsumerd_thread and return
1672 * an error.
1673 */
1674 ERR("Condition timed out. The consumer thread was never ready."
1675 " Killing it");
1676 ret = pthread_cancel(consumer_data->thread);
1677 if (ret < 0) {
1678 PERROR("pthread_cancel consumer thread");
1679 }
1680 } else {
1681 PERROR("pthread_cond_wait failed consumer thread");
1682 }
1683 goto error;
1684 }
1685
1686 pthread_mutex_lock(&consumer_data->pid_mutex);
1687 if (consumer_data->pid == 0) {
1688 ERR("Consumerd did not start");
1689 pthread_mutex_unlock(&consumer_data->pid_mutex);
1690 goto error;
1691 }
1692 pthread_mutex_unlock(&consumer_data->pid_mutex);
1693
1694 return 0;
1695
1696 error:
1697 return ret;
1698 }
1699
1700 /*
1701 * Join consumer thread
1702 */
1703 static int join_consumer_thread(struct consumer_data *consumer_data)
1704 {
1705 void *status;
1706
1707 /* Consumer pid must be a real one. */
1708 if (consumer_data->pid > 0) {
1709 int ret;
1710 ret = kill(consumer_data->pid, SIGTERM);
1711 if (ret) {
1712 ERR("Error killing consumer daemon");
1713 return ret;
1714 }
1715 return pthread_join(consumer_data->thread, &status);
1716 } else {
1717 return 0;
1718 }
1719 }
1720
1721 /*
1722 * Fork and exec a consumer daemon (consumerd).
1723 *
1724 * Return pid if successful else -1.
1725 */
1726 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
1727 {
1728 int ret;
1729 pid_t pid;
1730 const char *consumer_to_use;
1731 const char *verbosity;
1732 struct stat st;
1733
1734 DBG("Spawning consumerd");
1735
1736 pid = fork();
1737 if (pid == 0) {
1738 /*
1739 * Exec consumerd.
1740 */
1741 if (opt_verbose_consumer) {
1742 verbosity = "--verbose";
1743 } else {
1744 verbosity = "--quiet";
1745 }
1746 switch (consumer_data->type) {
1747 case LTTNG_CONSUMER_KERNEL:
1748 /*
1749 * Find out which consumerd to execute. We will first try the
1750 * 64-bit path, then the sessiond's installation directory, and
1751 * fallback on the 32-bit one,
1752 */
1753 DBG3("Looking for a kernel consumer at these locations:");
1754 DBG3(" 1) %s", consumerd64_bin);
1755 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1756 DBG3(" 3) %s", consumerd32_bin);
1757 if (stat(consumerd64_bin, &st) == 0) {
1758 DBG3("Found location #1");
1759 consumer_to_use = consumerd64_bin;
1760 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
1761 DBG3("Found location #2");
1762 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
1763 } else if (stat(consumerd32_bin, &st) == 0) {
1764 DBG3("Found location #3");
1765 consumer_to_use = consumerd32_bin;
1766 } else {
1767 DBG("Could not find any valid consumerd executable");
1768 break;
1769 }
1770 DBG("Using kernel consumer at: %s", consumer_to_use);
1771 execl(consumer_to_use,
1772 "lttng-consumerd", verbosity, "-k",
1773 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1774 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1775 NULL);
1776 break;
1777 case LTTNG_CONSUMER64_UST:
1778 {
1779 char *tmpnew = NULL;
1780
1781 if (consumerd64_libdir[0] != '\0') {
1782 char *tmp;
1783 size_t tmplen;
1784
1785 tmp = getenv("LD_LIBRARY_PATH");
1786 if (!tmp) {
1787 tmp = "";
1788 }
1789 tmplen = strlen("LD_LIBRARY_PATH=")
1790 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1791 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1792 if (!tmpnew) {
1793 ret = -ENOMEM;
1794 goto error;
1795 }
1796 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1797 strcat(tmpnew, consumerd64_libdir);
1798 if (tmp[0] != '\0') {
1799 strcat(tmpnew, ":");
1800 strcat(tmpnew, tmp);
1801 }
1802 ret = putenv(tmpnew);
1803 if (ret) {
1804 ret = -errno;
1805 goto error;
1806 }
1807 }
1808 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
1809 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
1810 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1811 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1812 NULL);
1813 if (consumerd64_libdir[0] != '\0') {
1814 free(tmpnew);
1815 }
1816 if (ret) {
1817 goto error;
1818 }
1819 break;
1820 }
1821 case LTTNG_CONSUMER32_UST:
1822 {
1823 char *tmpnew = NULL;
1824
1825 if (consumerd32_libdir[0] != '\0') {
1826 char *tmp;
1827 size_t tmplen;
1828
1829 tmp = getenv("LD_LIBRARY_PATH");
1830 if (!tmp) {
1831 tmp = "";
1832 }
1833 tmplen = strlen("LD_LIBRARY_PATH=")
1834 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1835 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1836 if (!tmpnew) {
1837 ret = -ENOMEM;
1838 goto error;
1839 }
1840 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1841 strcat(tmpnew, consumerd32_libdir);
1842 if (tmp[0] != '\0') {
1843 strcat(tmpnew, ":");
1844 strcat(tmpnew, tmp);
1845 }
1846 ret = putenv(tmpnew);
1847 if (ret) {
1848 ret = -errno;
1849 goto error;
1850 }
1851 }
1852 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
1853 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
1854 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1855 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1856 NULL);
1857 if (consumerd32_libdir[0] != '\0') {
1858 free(tmpnew);
1859 }
1860 if (ret) {
1861 goto error;
1862 }
1863 break;
1864 }
1865 default:
1866 PERROR("unknown consumer type");
1867 exit(EXIT_FAILURE);
1868 }
1869 if (errno != 0) {
1870 PERROR("kernel start consumer exec");
1871 }
1872 exit(EXIT_FAILURE);
1873 } else if (pid > 0) {
1874 ret = pid;
1875 } else {
1876 PERROR("start consumer fork");
1877 ret = -errno;
1878 }
1879 error:
1880 return ret;
1881 }
1882
1883 /*
1884 * Spawn the consumerd daemon and session daemon thread.
1885 */
1886 static int start_consumerd(struct consumer_data *consumer_data)
1887 {
1888 int ret;
1889
1890 /*
1891 * Set the listen() state on the socket since there is a possible race
1892 * between the exec() of the consumer daemon and this call if place in the
1893 * consumer thread. See bug #366 for more details.
1894 */
1895 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1896 if (ret < 0) {
1897 goto error;
1898 }
1899
1900 pthread_mutex_lock(&consumer_data->pid_mutex);
1901 if (consumer_data->pid != 0) {
1902 pthread_mutex_unlock(&consumer_data->pid_mutex);
1903 goto end;
1904 }
1905
1906 ret = spawn_consumerd(consumer_data);
1907 if (ret < 0) {
1908 ERR("Spawning consumerd failed");
1909 pthread_mutex_unlock(&consumer_data->pid_mutex);
1910 goto error;
1911 }
1912
1913 /* Setting up the consumer_data pid */
1914 consumer_data->pid = ret;
1915 DBG2("Consumer pid %d", consumer_data->pid);
1916 pthread_mutex_unlock(&consumer_data->pid_mutex);
1917
1918 DBG2("Spawning consumer control thread");
1919 ret = spawn_consumer_thread(consumer_data);
1920 if (ret < 0) {
1921 ERR("Fatal error spawning consumer control thread");
1922 goto error;
1923 }
1924
1925 end:
1926 return 0;
1927
1928 error:
1929 /* Cleanup already created socket on error. */
1930 if (consumer_data->err_sock >= 0) {
1931 int err;
1932
1933 err = close(consumer_data->err_sock);
1934 if (err < 0) {
1935 PERROR("close consumer data error socket");
1936 }
1937 }
1938 return ret;
1939 }
1940
1941 /*
1942 * Compute health status of each consumer. If one of them is zero (bad
1943 * state), we return 0.
1944 */
1945 static int check_consumer_health(void)
1946 {
1947 int ret;
1948
1949 ret = health_check_state(&kconsumer_data.health) &&
1950 health_check_state(&ustconsumer32_data.health) &&
1951 health_check_state(&ustconsumer64_data.health);
1952
1953 DBG3("Health consumer check %d", ret);
1954
1955 return ret;
1956 }
1957
1958 /*
1959 * Setup necessary data for kernel tracer action.
1960 */
1961 static int init_kernel_tracer(void)
1962 {
1963 int ret;
1964
1965 /* Modprobe lttng kernel modules */
1966 ret = modprobe_lttng_control();
1967 if (ret < 0) {
1968 goto error;
1969 }
1970
1971 /* Open debugfs lttng */
1972 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1973 if (kernel_tracer_fd < 0) {
1974 DBG("Failed to open %s", module_proc_lttng);
1975 ret = -1;
1976 goto error_open;
1977 }
1978
1979 /* Validate kernel version */
1980 ret = kernel_validate_version(kernel_tracer_fd);
1981 if (ret < 0) {
1982 goto error_version;
1983 }
1984
1985 ret = modprobe_lttng_data();
1986 if (ret < 0) {
1987 goto error_modules;
1988 }
1989
1990 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1991 return 0;
1992
1993 error_version:
1994 modprobe_remove_lttng_control();
1995 ret = close(kernel_tracer_fd);
1996 if (ret) {
1997 PERROR("close");
1998 }
1999 kernel_tracer_fd = -1;
2000 return LTTNG_ERR_KERN_VERSION;
2001
2002 error_modules:
2003 ret = close(kernel_tracer_fd);
2004 if (ret) {
2005 PERROR("close");
2006 }
2007
2008 error_open:
2009 modprobe_remove_lttng_control();
2010
2011 error:
2012 WARN("No kernel tracer available");
2013 kernel_tracer_fd = -1;
2014 if (!is_root) {
2015 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2016 } else {
2017 return LTTNG_ERR_KERN_NA;
2018 }
2019 }
2020
2021
2022 /*
2023 * Copy consumer output from the tracing session to the domain session. The
2024 * function also applies the right modification on a per domain basis for the
2025 * trace files destination directory.
2026 */
2027 static int copy_session_consumer(int domain, struct ltt_session *session)
2028 {
2029 int ret;
2030 const char *dir_name;
2031 struct consumer_output *consumer;
2032
2033 assert(session);
2034 assert(session->consumer);
2035
2036 switch (domain) {
2037 case LTTNG_DOMAIN_KERNEL:
2038 DBG3("Copying tracing session consumer output in kernel session");
2039 /*
2040 * XXX: We should audit the session creation and what this function
2041 * does "extra" in order to avoid a destroy since this function is used
2042 * in the domain session creation (kernel and ust) only. Same for UST
2043 * domain.
2044 */
2045 if (session->kernel_session->consumer) {
2046 consumer_destroy_output(session->kernel_session->consumer);
2047 }
2048 session->kernel_session->consumer =
2049 consumer_copy_output(session->consumer);
2050 /* Ease our life a bit for the next part */
2051 consumer = session->kernel_session->consumer;
2052 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2053 break;
2054 case LTTNG_DOMAIN_UST:
2055 DBG3("Copying tracing session consumer output in UST session");
2056 if (session->ust_session->consumer) {
2057 consumer_destroy_output(session->ust_session->consumer);
2058 }
2059 session->ust_session->consumer =
2060 consumer_copy_output(session->consumer);
2061 /* Ease our life a bit for the next part */
2062 consumer = session->ust_session->consumer;
2063 dir_name = DEFAULT_UST_TRACE_DIR;
2064 break;
2065 default:
2066 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2067 goto error;
2068 }
2069
2070 /* Append correct directory to subdir */
2071 strncat(consumer->subdir, dir_name,
2072 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2073 DBG3("Copy session consumer subdir %s", consumer->subdir);
2074
2075 ret = LTTNG_OK;
2076
2077 error:
2078 return ret;
2079 }
2080
2081 /*
2082 * Create an UST session and add it to the session ust list.
2083 */
2084 static int create_ust_session(struct ltt_session *session,
2085 struct lttng_domain *domain)
2086 {
2087 int ret;
2088 struct ltt_ust_session *lus = NULL;
2089
2090 assert(session);
2091 assert(domain);
2092 assert(session->consumer);
2093
2094 switch (domain->type) {
2095 case LTTNG_DOMAIN_UST:
2096 break;
2097 default:
2098 ERR("Unknown UST domain on create session %d", domain->type);
2099 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2100 goto error;
2101 }
2102
2103 DBG("Creating UST session");
2104
2105 lus = trace_ust_create_session(session->path, session->id, domain);
2106 if (lus == NULL) {
2107 ret = LTTNG_ERR_UST_SESS_FAIL;
2108 goto error;
2109 }
2110
2111 lus->uid = session->uid;
2112 lus->gid = session->gid;
2113 session->ust_session = lus;
2114
2115 /* Copy session output to the newly created UST session */
2116 ret = copy_session_consumer(domain->type, session);
2117 if (ret != LTTNG_OK) {
2118 goto error;
2119 }
2120
2121 return LTTNG_OK;
2122
2123 error:
2124 free(lus);
2125 session->ust_session = NULL;
2126 return ret;
2127 }
2128
2129 /*
2130 * Create a kernel tracer session then create the default channel.
2131 */
2132 static int create_kernel_session(struct ltt_session *session)
2133 {
2134 int ret;
2135
2136 DBG("Creating kernel session");
2137
2138 ret = kernel_create_session(session, kernel_tracer_fd);
2139 if (ret < 0) {
2140 ret = LTTNG_ERR_KERN_SESS_FAIL;
2141 goto error;
2142 }
2143
2144 /* Code flow safety */
2145 assert(session->kernel_session);
2146
2147 /* Copy session output to the newly created Kernel session */
2148 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2149 if (ret != LTTNG_OK) {
2150 goto error;
2151 }
2152
2153 /* Create directory(ies) on local filesystem. */
2154 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2155 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2156 ret = run_as_mkdir_recursive(
2157 session->kernel_session->consumer->dst.trace_path,
2158 S_IRWXU | S_IRWXG, session->uid, session->gid);
2159 if (ret < 0) {
2160 if (ret != -EEXIST) {
2161 ERR("Trace directory creation error");
2162 goto error;
2163 }
2164 }
2165 }
2166
2167 session->kernel_session->uid = session->uid;
2168 session->kernel_session->gid = session->gid;
2169
2170 return LTTNG_OK;
2171
2172 error:
2173 trace_kernel_destroy_session(session->kernel_session);
2174 session->kernel_session = NULL;
2175 return ret;
2176 }
2177
2178 /*
2179 * Count number of session permitted by uid/gid.
2180 */
2181 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2182 {
2183 unsigned int i = 0;
2184 struct ltt_session *session;
2185
2186 DBG("Counting number of available session for UID %d GID %d",
2187 uid, gid);
2188 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2189 /*
2190 * Only list the sessions the user can control.
2191 */
2192 if (!session_access_ok(session, uid, gid)) {
2193 continue;
2194 }
2195 i++;
2196 }
2197 return i;
2198 }
2199
2200 /*
2201 * Process the command requested by the lttng client within the command
2202 * context structure. This function make sure that the return structure (llm)
2203 * is set and ready for transmission before returning.
2204 *
2205 * Return any error encountered or 0 for success.
2206 *
2207 * "sock" is only used for special-case var. len data.
2208 */
2209 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2210 int *sock_error)
2211 {
2212 int ret = LTTNG_OK;
2213 int need_tracing_session = 1;
2214 int need_domain;
2215
2216 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2217
2218 *sock_error = 0;
2219
2220 switch (cmd_ctx->lsm->cmd_type) {
2221 case LTTNG_CREATE_SESSION:
2222 case LTTNG_DESTROY_SESSION:
2223 case LTTNG_LIST_SESSIONS:
2224 case LTTNG_LIST_DOMAINS:
2225 case LTTNG_START_TRACE:
2226 case LTTNG_STOP_TRACE:
2227 case LTTNG_DATA_PENDING:
2228 need_domain = 0;
2229 break;
2230 default:
2231 need_domain = 1;
2232 }
2233
2234 if (opt_no_kernel && need_domain
2235 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2236 if (!is_root) {
2237 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2238 } else {
2239 ret = LTTNG_ERR_KERN_NA;
2240 }
2241 goto error;
2242 }
2243
2244 /* Deny register consumer if we already have a spawned consumer. */
2245 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2246 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2247 if (kconsumer_data.pid > 0) {
2248 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2249 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2250 goto error;
2251 }
2252 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2253 }
2254
2255 /*
2256 * Check for command that don't needs to allocate a returned payload. We do
2257 * this here so we don't have to make the call for no payload at each
2258 * command.
2259 */
2260 switch(cmd_ctx->lsm->cmd_type) {
2261 case LTTNG_LIST_SESSIONS:
2262 case LTTNG_LIST_TRACEPOINTS:
2263 case LTTNG_LIST_TRACEPOINT_FIELDS:
2264 case LTTNG_LIST_DOMAINS:
2265 case LTTNG_LIST_CHANNELS:
2266 case LTTNG_LIST_EVENTS:
2267 break;
2268 default:
2269 /* Setup lttng message with no payload */
2270 ret = setup_lttng_msg(cmd_ctx, 0);
2271 if (ret < 0) {
2272 /* This label does not try to unlock the session */
2273 goto init_setup_error;
2274 }
2275 }
2276
2277 /* Commands that DO NOT need a session. */
2278 switch (cmd_ctx->lsm->cmd_type) {
2279 case LTTNG_CREATE_SESSION:
2280 case LTTNG_CALIBRATE:
2281 case LTTNG_LIST_SESSIONS:
2282 case LTTNG_LIST_TRACEPOINTS:
2283 case LTTNG_LIST_TRACEPOINT_FIELDS:
2284 need_tracing_session = 0;
2285 break;
2286 default:
2287 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2288 /*
2289 * We keep the session list lock across _all_ commands
2290 * for now, because the per-session lock does not
2291 * handle teardown properly.
2292 */
2293 session_lock_list();
2294 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2295 if (cmd_ctx->session == NULL) {
2296 if (cmd_ctx->lsm->session.name != NULL) {
2297 ret = LTTNG_ERR_SESS_NOT_FOUND;
2298 } else {
2299 /* If no session name specified */
2300 ret = LTTNG_ERR_SELECT_SESS;
2301 }
2302 goto error;
2303 } else {
2304 /* Acquire lock for the session */
2305 session_lock(cmd_ctx->session);
2306 }
2307 break;
2308 }
2309
2310 if (!need_domain) {
2311 goto skip_domain;
2312 }
2313
2314 /*
2315 * Check domain type for specific "pre-action".
2316 */
2317 switch (cmd_ctx->lsm->domain.type) {
2318 case LTTNG_DOMAIN_KERNEL:
2319 if (!is_root) {
2320 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2321 goto error;
2322 }
2323
2324 /* Kernel tracer check */
2325 if (kernel_tracer_fd == -1) {
2326 /* Basically, load kernel tracer modules */
2327 ret = init_kernel_tracer();
2328 if (ret != 0) {
2329 goto error;
2330 }
2331 }
2332
2333 /* Consumer is in an ERROR state. Report back to client */
2334 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2335 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2336 goto error;
2337 }
2338
2339 /* Need a session for kernel command */
2340 if (need_tracing_session) {
2341 if (cmd_ctx->session->kernel_session == NULL) {
2342 ret = create_kernel_session(cmd_ctx->session);
2343 if (ret < 0) {
2344 ret = LTTNG_ERR_KERN_SESS_FAIL;
2345 goto error;
2346 }
2347 }
2348
2349 /* Start the kernel consumer daemon */
2350 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2351 if (kconsumer_data.pid == 0 &&
2352 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2353 cmd_ctx->session->start_consumer) {
2354 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2355 ret = start_consumerd(&kconsumer_data);
2356 if (ret < 0) {
2357 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2358 goto error;
2359 }
2360 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2361 } else {
2362 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2363 }
2364
2365 /*
2366 * The consumer was just spawned so we need to add the socket to
2367 * the consumer output of the session if exist.
2368 */
2369 ret = consumer_create_socket(&kconsumer_data,
2370 cmd_ctx->session->kernel_session->consumer);
2371 if (ret < 0) {
2372 goto error;
2373 }
2374 }
2375
2376 break;
2377 case LTTNG_DOMAIN_UST:
2378 {
2379 /* Consumer is in an ERROR state. Report back to client */
2380 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2381 ret = LTTNG_ERR_NO_USTCONSUMERD;
2382 goto error;
2383 }
2384
2385 if (need_tracing_session) {
2386 /* Create UST session if none exist. */
2387 if (cmd_ctx->session->ust_session == NULL) {
2388 ret = create_ust_session(cmd_ctx->session,
2389 &cmd_ctx->lsm->domain);
2390 if (ret != LTTNG_OK) {
2391 goto error;
2392 }
2393 }
2394
2395 /* Start the UST consumer daemons */
2396 /* 64-bit */
2397 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2398 if (consumerd64_bin[0] != '\0' &&
2399 ustconsumer64_data.pid == 0 &&
2400 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2401 cmd_ctx->session->start_consumer) {
2402 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2403 ret = start_consumerd(&ustconsumer64_data);
2404 if (ret < 0) {
2405 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2406 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2407 goto error;
2408 }
2409
2410 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2411 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2412 } else {
2413 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2414 }
2415
2416 /*
2417 * Setup socket for consumer 64 bit. No need for atomic access
2418 * since it was set above and can ONLY be set in this thread.
2419 */
2420 ret = consumer_create_socket(&ustconsumer64_data,
2421 cmd_ctx->session->ust_session->consumer);
2422 if (ret < 0) {
2423 goto error;
2424 }
2425
2426 /* 32-bit */
2427 if (consumerd32_bin[0] != '\0' &&
2428 ustconsumer32_data.pid == 0 &&
2429 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2430 cmd_ctx->session->start_consumer) {
2431 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2432 ret = start_consumerd(&ustconsumer32_data);
2433 if (ret < 0) {
2434 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2435 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2436 goto error;
2437 }
2438
2439 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2440 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2441 } else {
2442 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2443 }
2444
2445 /*
2446 * Setup socket for consumer 64 bit. No need for atomic access
2447 * since it was set above and can ONLY be set in this thread.
2448 */
2449 ret = consumer_create_socket(&ustconsumer32_data,
2450 cmd_ctx->session->ust_session->consumer);
2451 if (ret < 0) {
2452 goto error;
2453 }
2454 }
2455 break;
2456 }
2457 default:
2458 break;
2459 }
2460 skip_domain:
2461
2462 /* Validate consumer daemon state when start/stop trace command */
2463 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2464 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2465 switch (cmd_ctx->lsm->domain.type) {
2466 case LTTNG_DOMAIN_UST:
2467 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2468 ret = LTTNG_ERR_NO_USTCONSUMERD;
2469 goto error;
2470 }
2471 break;
2472 case LTTNG_DOMAIN_KERNEL:
2473 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2474 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2475 goto error;
2476 }
2477 break;
2478 }
2479 }
2480
2481 /*
2482 * Check that the UID or GID match that of the tracing session.
2483 * The root user can interact with all sessions.
2484 */
2485 if (need_tracing_session) {
2486 if (!session_access_ok(cmd_ctx->session,
2487 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2488 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2489 ret = LTTNG_ERR_EPERM;
2490 goto error;
2491 }
2492 }
2493
2494 /* Process by command type */
2495 switch (cmd_ctx->lsm->cmd_type) {
2496 case LTTNG_ADD_CONTEXT:
2497 {
2498 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2499 cmd_ctx->lsm->u.context.channel_name,
2500 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2501 break;
2502 }
2503 case LTTNG_DISABLE_CHANNEL:
2504 {
2505 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2506 cmd_ctx->lsm->u.disable.channel_name);
2507 break;
2508 }
2509 case LTTNG_DISABLE_EVENT:
2510 {
2511 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2512 cmd_ctx->lsm->u.disable.channel_name,
2513 cmd_ctx->lsm->u.disable.name);
2514 break;
2515 }
2516 case LTTNG_DISABLE_ALL_EVENT:
2517 {
2518 DBG("Disabling all events");
2519
2520 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2521 cmd_ctx->lsm->u.disable.channel_name);
2522 break;
2523 }
2524 case LTTNG_DISABLE_CONSUMER:
2525 {
2526 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2527 break;
2528 }
2529 case LTTNG_ENABLE_CHANNEL:
2530 {
2531 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2532 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2533 break;
2534 }
2535 case LTTNG_ENABLE_CONSUMER:
2536 {
2537 /*
2538 * XXX: 0 means that this URI should be applied on the session. Should
2539 * be a DOMAIN enuam.
2540 */
2541 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2542 if (ret != LTTNG_OK) {
2543 goto error;
2544 }
2545
2546 if (cmd_ctx->lsm->domain.type == 0) {
2547 /* Add the URI for the UST session if a consumer is present. */
2548 if (cmd_ctx->session->ust_session &&
2549 cmd_ctx->session->ust_session->consumer) {
2550 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2551 } else if (cmd_ctx->session->kernel_session &&
2552 cmd_ctx->session->kernel_session->consumer) {
2553 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2554 cmd_ctx->session);
2555 }
2556 }
2557 break;
2558 }
2559 case LTTNG_ENABLE_EVENT:
2560 {
2561 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2562 cmd_ctx->lsm->u.enable.channel_name,
2563 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
2564 break;
2565 }
2566 case LTTNG_ENABLE_ALL_EVENT:
2567 {
2568 DBG("Enabling all events");
2569
2570 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2571 cmd_ctx->lsm->u.enable.channel_name,
2572 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
2573 break;
2574 }
2575 case LTTNG_LIST_TRACEPOINTS:
2576 {
2577 struct lttng_event *events;
2578 ssize_t nb_events;
2579
2580 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2581 if (nb_events < 0) {
2582 /* Return value is a negative lttng_error_code. */
2583 ret = -nb_events;
2584 goto error;
2585 }
2586
2587 /*
2588 * Setup lttng message with payload size set to the event list size in
2589 * bytes and then copy list into the llm payload.
2590 */
2591 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2592 if (ret < 0) {
2593 free(events);
2594 goto setup_error;
2595 }
2596
2597 /* Copy event list into message payload */
2598 memcpy(cmd_ctx->llm->payload, events,
2599 sizeof(struct lttng_event) * nb_events);
2600
2601 free(events);
2602
2603 ret = LTTNG_OK;
2604 break;
2605 }
2606 case LTTNG_LIST_TRACEPOINT_FIELDS:
2607 {
2608 struct lttng_event_field *fields;
2609 ssize_t nb_fields;
2610
2611 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2612 &fields);
2613 if (nb_fields < 0) {
2614 /* Return value is a negative lttng_error_code. */
2615 ret = -nb_fields;
2616 goto error;
2617 }
2618
2619 /*
2620 * Setup lttng message with payload size set to the event list size in
2621 * bytes and then copy list into the llm payload.
2622 */
2623 ret = setup_lttng_msg(cmd_ctx,
2624 sizeof(struct lttng_event_field) * nb_fields);
2625 if (ret < 0) {
2626 free(fields);
2627 goto setup_error;
2628 }
2629
2630 /* Copy event list into message payload */
2631 memcpy(cmd_ctx->llm->payload, fields,
2632 sizeof(struct lttng_event_field) * nb_fields);
2633
2634 free(fields);
2635
2636 ret = LTTNG_OK;
2637 break;
2638 }
2639 case LTTNG_SET_CONSUMER_URI:
2640 {
2641 size_t nb_uri, len;
2642 struct lttng_uri *uris;
2643
2644 nb_uri = cmd_ctx->lsm->u.uri.size;
2645 len = nb_uri * sizeof(struct lttng_uri);
2646
2647 if (nb_uri == 0) {
2648 ret = LTTNG_ERR_INVALID;
2649 goto error;
2650 }
2651
2652 uris = zmalloc(len);
2653 if (uris == NULL) {
2654 ret = LTTNG_ERR_FATAL;
2655 goto error;
2656 }
2657
2658 /* Receive variable len data */
2659 DBG("Receiving %zu URI(s) from client ...", nb_uri);
2660 ret = lttcomm_recv_unix_sock(sock, uris, len);
2661 if (ret <= 0) {
2662 DBG("No URIs received from client... continuing");
2663 *sock_error = 1;
2664 ret = LTTNG_ERR_SESSION_FAIL;
2665 free(uris);
2666 goto error;
2667 }
2668
2669 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2670 nb_uri, uris);
2671 if (ret != LTTNG_OK) {
2672 free(uris);
2673 goto error;
2674 }
2675
2676 /*
2677 * XXX: 0 means that this URI should be applied on the session. Should
2678 * be a DOMAIN enuam.
2679 */
2680 if (cmd_ctx->lsm->domain.type == 0) {
2681 /* Add the URI for the UST session if a consumer is present. */
2682 if (cmd_ctx->session->ust_session &&
2683 cmd_ctx->session->ust_session->consumer) {
2684 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2685 nb_uri, uris);
2686 } else if (cmd_ctx->session->kernel_session &&
2687 cmd_ctx->session->kernel_session->consumer) {
2688 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2689 cmd_ctx->session, nb_uri, uris);
2690 }
2691 }
2692
2693 free(uris);
2694
2695 break;
2696 }
2697 case LTTNG_START_TRACE:
2698 {
2699 ret = cmd_start_trace(cmd_ctx->session);
2700 break;
2701 }
2702 case LTTNG_STOP_TRACE:
2703 {
2704 ret = cmd_stop_trace(cmd_ctx->session);
2705 break;
2706 }
2707 case LTTNG_CREATE_SESSION:
2708 {
2709 size_t nb_uri, len;
2710 struct lttng_uri *uris = NULL;
2711
2712 nb_uri = cmd_ctx->lsm->u.uri.size;
2713 len = nb_uri * sizeof(struct lttng_uri);
2714
2715 if (nb_uri > 0) {
2716 uris = zmalloc(len);
2717 if (uris == NULL) {
2718 ret = LTTNG_ERR_FATAL;
2719 goto error;
2720 }
2721
2722 /* Receive variable len data */
2723 DBG("Waiting for %zu URIs from client ...", nb_uri);
2724 ret = lttcomm_recv_unix_sock(sock, uris, len);
2725 if (ret <= 0) {
2726 DBG("No URIs received from client... continuing");
2727 *sock_error = 1;
2728 ret = LTTNG_ERR_SESSION_FAIL;
2729 free(uris);
2730 goto error;
2731 }
2732
2733 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2734 DBG("Creating session with ONE network URI is a bad call");
2735 ret = LTTNG_ERR_SESSION_FAIL;
2736 free(uris);
2737 goto error;
2738 }
2739 }
2740
2741 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2742 &cmd_ctx->creds);
2743
2744 free(uris);
2745
2746 break;
2747 }
2748 case LTTNG_DESTROY_SESSION:
2749 {
2750 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
2751
2752 /* Set session to NULL so we do not unlock it after free. */
2753 cmd_ctx->session = NULL;
2754 break;
2755 }
2756 case LTTNG_LIST_DOMAINS:
2757 {
2758 ssize_t nb_dom;
2759 struct lttng_domain *domains;
2760
2761 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2762 if (nb_dom < 0) {
2763 /* Return value is a negative lttng_error_code. */
2764 ret = -nb_dom;
2765 goto error;
2766 }
2767
2768 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
2769 if (ret < 0) {
2770 goto setup_error;
2771 }
2772
2773 /* Copy event list into message payload */
2774 memcpy(cmd_ctx->llm->payload, domains,
2775 nb_dom * sizeof(struct lttng_domain));
2776
2777 free(domains);
2778
2779 ret = LTTNG_OK;
2780 break;
2781 }
2782 case LTTNG_LIST_CHANNELS:
2783 {
2784 int nb_chan;
2785 struct lttng_channel *channels;
2786
2787 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2788 cmd_ctx->session, &channels);
2789 if (nb_chan < 0) {
2790 /* Return value is a negative lttng_error_code. */
2791 ret = -nb_chan;
2792 goto error;
2793 }
2794
2795 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
2796 if (ret < 0) {
2797 goto setup_error;
2798 }
2799
2800 /* Copy event list into message payload */
2801 memcpy(cmd_ctx->llm->payload, channels,
2802 nb_chan * sizeof(struct lttng_channel));
2803
2804 free(channels);
2805
2806 ret = LTTNG_OK;
2807 break;
2808 }
2809 case LTTNG_LIST_EVENTS:
2810 {
2811 ssize_t nb_event;
2812 struct lttng_event *events = NULL;
2813
2814 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
2815 cmd_ctx->lsm->u.list.channel_name, &events);
2816 if (nb_event < 0) {
2817 /* Return value is a negative lttng_error_code. */
2818 ret = -nb_event;
2819 goto error;
2820 }
2821
2822 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
2823 if (ret < 0) {
2824 goto setup_error;
2825 }
2826
2827 /* Copy event list into message payload */
2828 memcpy(cmd_ctx->llm->payload, events,
2829 nb_event * sizeof(struct lttng_event));
2830
2831 free(events);
2832
2833 ret = LTTNG_OK;
2834 break;
2835 }
2836 case LTTNG_LIST_SESSIONS:
2837 {
2838 unsigned int nr_sessions;
2839
2840 session_lock_list();
2841 nr_sessions = lttng_sessions_count(
2842 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2843 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2844
2845 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
2846 if (ret < 0) {
2847 session_unlock_list();
2848 goto setup_error;
2849 }
2850
2851 /* Filled the session array */
2852 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
2853 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2854 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
2855
2856 session_unlock_list();
2857
2858 ret = LTTNG_OK;
2859 break;
2860 }
2861 case LTTNG_CALIBRATE:
2862 {
2863 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2864 &cmd_ctx->lsm->u.calibrate);
2865 break;
2866 }
2867 case LTTNG_REGISTER_CONSUMER:
2868 {
2869 struct consumer_data *cdata;
2870
2871 switch (cmd_ctx->lsm->domain.type) {
2872 case LTTNG_DOMAIN_KERNEL:
2873 cdata = &kconsumer_data;
2874 break;
2875 default:
2876 ret = LTTNG_ERR_UND;
2877 goto error;
2878 }
2879
2880 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2881 cmd_ctx->lsm->u.reg.path, cdata);
2882 break;
2883 }
2884 case LTTNG_ENABLE_EVENT_WITH_FILTER:
2885 {
2886 struct lttng_filter_bytecode *bytecode;
2887
2888 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
2889 ret = LTTNG_ERR_FILTER_INVAL;
2890 goto error;
2891 }
2892 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2893 ret = LTTNG_ERR_FILTER_INVAL;
2894 goto error;
2895 }
2896 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
2897 if (!bytecode) {
2898 ret = LTTNG_ERR_FILTER_NOMEM;
2899 goto error;
2900 }
2901 /* Receive var. len. data */
2902 DBG("Receiving var len data from client ...");
2903 ret = lttcomm_recv_unix_sock(sock, bytecode,
2904 cmd_ctx->lsm->u.enable.bytecode_len);
2905 if (ret <= 0) {
2906 DBG("Nothing recv() from client var len data... continuing");
2907 *sock_error = 1;
2908 ret = LTTNG_ERR_FILTER_INVAL;
2909 goto error;
2910 }
2911
2912 if (bytecode->len + sizeof(*bytecode)
2913 != cmd_ctx->lsm->u.enable.bytecode_len) {
2914 free(bytecode);
2915 ret = LTTNG_ERR_FILTER_INVAL;
2916 goto error;
2917 }
2918
2919 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2920 cmd_ctx->lsm->u.enable.channel_name,
2921 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
2922 break;
2923 }
2924 case LTTNG_DATA_PENDING:
2925 {
2926 ret = cmd_data_pending(cmd_ctx->session);
2927 break;
2928 }
2929 default:
2930 ret = LTTNG_ERR_UND;
2931 break;
2932 }
2933
2934 error:
2935 if (cmd_ctx->llm == NULL) {
2936 DBG("Missing llm structure. Allocating one.");
2937 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
2938 goto setup_error;
2939 }
2940 }
2941 /* Set return code */
2942 cmd_ctx->llm->ret_code = ret;
2943 setup_error:
2944 if (cmd_ctx->session) {
2945 session_unlock(cmd_ctx->session);
2946 }
2947 if (need_tracing_session) {
2948 session_unlock_list();
2949 }
2950 init_setup_error:
2951 return ret;
2952 }
2953
2954 /*
2955 * Thread managing health check socket.
2956 */
2957 static void *thread_manage_health(void *data)
2958 {
2959 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
2960 uint32_t revents, nb_fd;
2961 struct lttng_poll_event events;
2962 struct lttcomm_health_msg msg;
2963 struct lttcomm_health_data reply;
2964
2965 DBG("[thread] Manage health check started");
2966
2967 rcu_register_thread();
2968
2969 /* Create unix socket */
2970 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2971 if (sock < 0) {
2972 ERR("Unable to create health check Unix socket");
2973 ret = -1;
2974 goto error;
2975 }
2976
2977 /*
2978 * Set the CLOEXEC flag. Return code is useless because either way, the
2979 * show must go on.
2980 */
2981 (void) utils_set_fd_cloexec(sock);
2982
2983 ret = lttcomm_listen_unix_sock(sock);
2984 if (ret < 0) {
2985 goto error;
2986 }
2987
2988 /*
2989 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2990 * more will be added to this poll set.
2991 */
2992 ret = create_thread_poll_set(&events, 2);
2993 if (ret < 0) {
2994 goto error;
2995 }
2996
2997 /* Add the application registration socket */
2998 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2999 if (ret < 0) {
3000 goto error;
3001 }
3002
3003 while (1) {
3004 DBG("Health check ready");
3005
3006 /* Inifinite blocking call, waiting for transmission */
3007 restart:
3008 ret = lttng_poll_wait(&events, -1);
3009 if (ret < 0) {
3010 /*
3011 * Restart interrupted system call.
3012 */
3013 if (errno == EINTR) {
3014 goto restart;
3015 }
3016 goto error;
3017 }
3018
3019 nb_fd = ret;
3020
3021 for (i = 0; i < nb_fd; i++) {
3022 /* Fetch once the poll data */
3023 revents = LTTNG_POLL_GETEV(&events, i);
3024 pollfd = LTTNG_POLL_GETFD(&events, i);
3025
3026 /* Thread quit pipe has been closed. Killing thread. */
3027 ret = check_thread_quit_pipe(pollfd, revents);
3028 if (ret) {
3029 err = 0;
3030 goto exit;
3031 }
3032
3033 /* Event on the registration socket */
3034 if (pollfd == sock) {
3035 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3036 ERR("Health socket poll error");
3037 goto error;
3038 }
3039 }
3040 }
3041
3042 new_sock = lttcomm_accept_unix_sock(sock);
3043 if (new_sock < 0) {
3044 goto error;
3045 }
3046
3047 /*
3048 * Set the CLOEXEC flag. Return code is useless because either way, the
3049 * show must go on.
3050 */
3051 (void) utils_set_fd_cloexec(new_sock);
3052
3053 DBG("Receiving data from client for health...");
3054 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3055 if (ret <= 0) {
3056 DBG("Nothing recv() from client... continuing");
3057 ret = close(new_sock);
3058 if (ret) {
3059 PERROR("close");
3060 }
3061 new_sock = -1;
3062 continue;
3063 }
3064
3065 rcu_thread_online();
3066
3067 switch (msg.component) {
3068 case LTTNG_HEALTH_CMD:
3069 reply.ret_code = health_check_state(&health_thread_cmd);
3070 break;
3071 case LTTNG_HEALTH_APP_MANAGE:
3072 reply.ret_code = health_check_state(&health_thread_app_manage);
3073 break;
3074 case LTTNG_HEALTH_APP_REG:
3075 reply.ret_code = health_check_state(&health_thread_app_reg);
3076 break;
3077 case LTTNG_HEALTH_KERNEL:
3078 reply.ret_code = health_check_state(&health_thread_kernel);
3079 break;
3080 case LTTNG_HEALTH_CONSUMER:
3081 reply.ret_code = check_consumer_health();
3082 break;
3083 case LTTNG_HEALTH_ALL:
3084 reply.ret_code =
3085 health_check_state(&health_thread_app_manage) &&
3086 health_check_state(&health_thread_app_reg) &&
3087 health_check_state(&health_thread_cmd) &&
3088 health_check_state(&health_thread_kernel) &&
3089 check_consumer_health();
3090 break;
3091 default:
3092 reply.ret_code = LTTNG_ERR_UND;
3093 break;
3094 }
3095
3096 /*
3097 * Flip ret value since 0 is a success and 1 indicates a bad health for
3098 * the client where in the sessiond it is the opposite. Again, this is
3099 * just to make things easier for us poor developer which enjoy a lot
3100 * lazyness.
3101 */
3102 if (reply.ret_code == 0 || reply.ret_code == 1) {
3103 reply.ret_code = !reply.ret_code;
3104 }
3105
3106 DBG2("Health check return value %d", reply.ret_code);
3107
3108 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3109 if (ret < 0) {
3110 ERR("Failed to send health data back to client");
3111 }
3112
3113 /* End of transmission */
3114 ret = close(new_sock);
3115 if (ret) {
3116 PERROR("close");
3117 }
3118 new_sock = -1;
3119 }
3120
3121 exit:
3122 error:
3123 if (err) {
3124 ERR("Health error occurred in %s", __func__);
3125 }
3126 DBG("Health check thread dying");
3127 unlink(health_unix_sock_path);
3128 if (sock >= 0) {
3129 ret = close(sock);
3130 if (ret) {
3131 PERROR("close");
3132 }
3133 }
3134 if (new_sock >= 0) {
3135 ret = close(new_sock);
3136 if (ret) {
3137 PERROR("close");
3138 }
3139 }
3140
3141 lttng_poll_clean(&events);
3142
3143 rcu_unregister_thread();
3144 return NULL;
3145 }
3146
3147 /*
3148 * This thread manage all clients request using the unix client socket for
3149 * communication.
3150 */
3151 static void *thread_manage_clients(void *data)
3152 {
3153 int sock = -1, ret, i, pollfd, err = -1;
3154 int sock_error;
3155 uint32_t revents, nb_fd;
3156 struct command_ctx *cmd_ctx = NULL;
3157 struct lttng_poll_event events;
3158
3159 DBG("[thread] Manage client started");
3160
3161 rcu_register_thread();
3162
3163 if (testpoint(thread_manage_clients)) {
3164 goto error_testpoint;
3165 }
3166
3167 health_code_update(&health_thread_cmd);
3168
3169 ret = lttcomm_listen_unix_sock(client_sock);
3170 if (ret < 0) {
3171 goto error_listen;
3172 }
3173
3174 /*
3175 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3176 * more will be added to this poll set.
3177 */
3178 ret = create_thread_poll_set(&events, 2);
3179 if (ret < 0) {
3180 goto error_create_poll;
3181 }
3182
3183 /* Add the application registration socket */
3184 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3185 if (ret < 0) {
3186 goto error;
3187 }
3188
3189 /*
3190 * Notify parent pid that we are ready to accept command for client side.
3191 */
3192 if (opt_sig_parent) {
3193 kill(ppid, SIGUSR1);
3194 }
3195
3196 if (testpoint(thread_manage_clients_before_loop)) {
3197 goto error;
3198 }
3199
3200 health_code_update(&health_thread_cmd);
3201
3202 while (1) {
3203 DBG("Accepting client command ...");
3204
3205 /* Inifinite blocking call, waiting for transmission */
3206 restart:
3207 health_poll_update(&health_thread_cmd);
3208 ret = lttng_poll_wait(&events, -1);
3209 health_poll_update(&health_thread_cmd);
3210 if (ret < 0) {
3211 /*
3212 * Restart interrupted system call.
3213 */
3214 if (errno == EINTR) {
3215 goto restart;
3216 }
3217 goto error;
3218 }
3219
3220 nb_fd = ret;
3221
3222 for (i = 0; i < nb_fd; i++) {
3223 /* Fetch once the poll data */
3224 revents = LTTNG_POLL_GETEV(&events, i);
3225 pollfd = LTTNG_POLL_GETFD(&events, i);
3226
3227 health_code_update(&health_thread_cmd);
3228
3229 /* Thread quit pipe has been closed. Killing thread. */
3230 ret = check_thread_quit_pipe(pollfd, revents);
3231 if (ret) {
3232 err = 0;