Wait for the destruction of sessions before tearing down main thread
[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 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <ctype.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
52 #include <common/dynamic-buffer.h>
53 #include <lttng/userspace-probe-internal.h>
54 #include <lttng/event-internal.h>
55
56 #include "lttng-sessiond.h"
57 #include "buffer-registry.h"
58 #include "channel.h"
59 #include "cmd.h"
60 #include "consumer.h"
61 #include "context.h"
62 #include "event.h"
63 #include "kernel.h"
64 #include "kernel-consumer.h"
65 #include "modprobe.h"
66 #include "shm.h"
67 #include "ust-ctl.h"
68 #include "ust-consumer.h"
69 #include "utils.h"
70 #include "fd-limit.h"
71 #include "health-sessiond.h"
72 #include "testpoint.h"
73 #include "ust-thread.h"
74 #include "agent-thread.h"
75 #include "save.h"
76 #include "load-session-thread.h"
77 #include "notification-thread.h"
78 #include "notification-thread-commands.h"
79 #include "rotation-thread.h"
80 #include "lttng-syscall.h"
81 #include "agent.h"
82 #include "ht-cleanup.h"
83 #include "sessiond-config.h"
84 #include "timer.h"
85
86 static const char *help_msg =
87 #ifdef LTTNG_EMBED_HELP
88 #include <lttng-sessiond.8.h>
89 #else
90 NULL
91 #endif
92 ;
93
94 const char *progname;
95 static int lockfile_fd = -1;
96
97 /* Set to 1 when a SIGUSR1 signal is received. */
98 static int recv_child_signal;
99
100 static struct lttng_kernel_tracer_version kernel_tracer_version;
101 static struct lttng_kernel_tracer_abi_version kernel_tracer_abi_version;
102
103 /*
104 * Consumer daemon specific control data. Every value not initialized here is
105 * set to 0 by the static definition.
106 */
107 static struct consumer_data kconsumer_data = {
108 .type = LTTNG_CONSUMER_KERNEL,
109 .err_sock = -1,
110 .cmd_sock = -1,
111 .channel_monitor_pipe = -1,
112 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
113 .lock = PTHREAD_MUTEX_INITIALIZER,
114 .cond = PTHREAD_COND_INITIALIZER,
115 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
116 };
117 static struct consumer_data ustconsumer64_data = {
118 .type = LTTNG_CONSUMER64_UST,
119 .err_sock = -1,
120 .cmd_sock = -1,
121 .channel_monitor_pipe = -1,
122 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
123 .lock = PTHREAD_MUTEX_INITIALIZER,
124 .cond = PTHREAD_COND_INITIALIZER,
125 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
126 };
127 static struct consumer_data ustconsumer32_data = {
128 .type = LTTNG_CONSUMER32_UST,
129 .err_sock = -1,
130 .cmd_sock = -1,
131 .channel_monitor_pipe = -1,
132 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
133 .lock = PTHREAD_MUTEX_INITIALIZER,
134 .cond = PTHREAD_COND_INITIALIZER,
135 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
136 };
137
138 /* Command line options */
139 static const struct option long_options[] = {
140 { "client-sock", required_argument, 0, 'c' },
141 { "apps-sock", required_argument, 0, 'a' },
142 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
143 { "kconsumerd-err-sock", required_argument, 0, '\0' },
144 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
145 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
146 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
147 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
148 { "consumerd32-path", required_argument, 0, '\0' },
149 { "consumerd32-libdir", required_argument, 0, '\0' },
150 { "consumerd64-path", required_argument, 0, '\0' },
151 { "consumerd64-libdir", required_argument, 0, '\0' },
152 { "daemonize", no_argument, 0, 'd' },
153 { "background", no_argument, 0, 'b' },
154 { "sig-parent", no_argument, 0, 'S' },
155 { "help", no_argument, 0, 'h' },
156 { "group", required_argument, 0, 'g' },
157 { "version", no_argument, 0, 'V' },
158 { "quiet", no_argument, 0, 'q' },
159 { "verbose", no_argument, 0, 'v' },
160 { "verbose-consumer", no_argument, 0, '\0' },
161 { "no-kernel", no_argument, 0, '\0' },
162 { "pidfile", required_argument, 0, 'p' },
163 { "agent-tcp-port", required_argument, 0, '\0' },
164 { "config", required_argument, 0, 'f' },
165 { "load", required_argument, 0, 'l' },
166 { "kmod-probes", required_argument, 0, '\0' },
167 { "extra-kmod-probes", required_argument, 0, '\0' },
168 { NULL, 0, 0, 0 }
169 };
170
171 /* Command line options to ignore from configuration file */
172 static const char *config_ignore_options[] = { "help", "version", "config" };
173
174 /* Shared between threads */
175 static int dispatch_thread_exit;
176
177 /* Sockets and FDs */
178 static int client_sock = -1;
179 static int apps_sock = -1;
180
181 /*
182 * This pipe is used to inform the thread managing application communication
183 * that a command is queued and ready to be processed.
184 */
185 static int apps_cmd_pipe[2] = { -1, -1 };
186
187 /* Pthread, Mutexes and Semaphores */
188 static pthread_t apps_thread;
189 static pthread_t apps_notify_thread;
190 static pthread_t reg_apps_thread;
191 static pthread_t client_thread;
192 static pthread_t kernel_thread;
193 static pthread_t dispatch_thread;
194 static pthread_t health_thread;
195 static pthread_t ht_cleanup_thread;
196 static pthread_t agent_reg_thread;
197 static pthread_t load_session_thread;
198 static pthread_t notification_thread;
199 static pthread_t rotation_thread;
200 static pthread_t timer_thread;
201
202 /*
203 * UST registration command queue. This queue is tied with a futex and uses a N
204 * wakers / 1 waiter implemented and detailed in futex.c/.h
205 *
206 * The thread_registration_apps and thread_dispatch_ust_registration uses this
207 * queue along with the wait/wake scheme. The thread_manage_apps receives down
208 * the line new application socket and monitors it for any I/O error or clean
209 * close that triggers an unregistration of the application.
210 */
211 static struct ust_cmd_queue ust_cmd_queue;
212
213 static const char *module_proc_lttng = "/proc/lttng";
214
215 /*
216 * Consumer daemon state which is changed when spawning it, killing it or in
217 * case of a fatal error.
218 */
219 enum consumerd_state {
220 CONSUMER_STARTED = 1,
221 CONSUMER_STOPPED = 2,
222 CONSUMER_ERROR = 3,
223 };
224
225 /*
226 * This consumer daemon state is used to validate if a client command will be
227 * able to reach the consumer. If not, the client is informed. For instance,
228 * doing a "lttng start" when the consumer state is set to ERROR will return an
229 * error to the client.
230 *
231 * The following example shows a possible race condition of this scheme:
232 *
233 * consumer thread error happens
234 * client cmd arrives
235 * client cmd checks state -> still OK
236 * consumer thread exit, sets error
237 * client cmd try to talk to consumer
238 * ...
239 *
240 * However, since the consumer is a different daemon, we have no way of making
241 * sure the command will reach it safely even with this state flag. This is why
242 * we consider that up to the state validation during command processing, the
243 * command is safe. After that, we can not guarantee the correctness of the
244 * client request vis-a-vis the consumer.
245 */
246 static enum consumerd_state ust_consumerd_state;
247 static enum consumerd_state kernel_consumerd_state;
248
249 /* Load session thread information to operate. */
250 static struct load_session_thread_data *load_info;
251
252 /*
253 * Section name to look for in the daemon configuration file.
254 */
255 static const char * const config_section_name = "sessiond";
256
257 /* Am I root or not. Set to 1 if the daemon is running as root */
258 static int is_root;
259
260 /* Rotation thread handle. */
261 static struct rotation_thread_handle *rotation_thread_handle;
262
263 /*
264 * Stop all threads by closing the thread quit pipe.
265 */
266 static void stop_threads(void)
267 {
268 int ret;
269
270 /* Stopping all threads */
271 DBG("Terminating all threads");
272 ret = sessiond_notify_quit_pipe();
273 if (ret < 0) {
274 ERR("write error on thread quit pipe");
275 }
276
277 /* Dispatch thread */
278 CMM_STORE_SHARED(dispatch_thread_exit, 1);
279 futex_nto1_wake(&ust_cmd_queue.futex);
280 }
281
282 /*
283 * Close every consumer sockets.
284 */
285 static void close_consumer_sockets(void)
286 {
287 int ret;
288
289 if (kconsumer_data.err_sock >= 0) {
290 ret = close(kconsumer_data.err_sock);
291 if (ret < 0) {
292 PERROR("kernel consumer err_sock close");
293 }
294 }
295 if (ustconsumer32_data.err_sock >= 0) {
296 ret = close(ustconsumer32_data.err_sock);
297 if (ret < 0) {
298 PERROR("UST consumerd32 err_sock close");
299 }
300 }
301 if (ustconsumer64_data.err_sock >= 0) {
302 ret = close(ustconsumer64_data.err_sock);
303 if (ret < 0) {
304 PERROR("UST consumerd64 err_sock close");
305 }
306 }
307 if (kconsumer_data.cmd_sock >= 0) {
308 ret = close(kconsumer_data.cmd_sock);
309 if (ret < 0) {
310 PERROR("kernel consumer cmd_sock close");
311 }
312 }
313 if (ustconsumer32_data.cmd_sock >= 0) {
314 ret = close(ustconsumer32_data.cmd_sock);
315 if (ret < 0) {
316 PERROR("UST consumerd32 cmd_sock close");
317 }
318 }
319 if (ustconsumer64_data.cmd_sock >= 0) {
320 ret = close(ustconsumer64_data.cmd_sock);
321 if (ret < 0) {
322 PERROR("UST consumerd64 cmd_sock close");
323 }
324 }
325 if (kconsumer_data.channel_monitor_pipe >= 0) {
326 ret = close(kconsumer_data.channel_monitor_pipe);
327 if (ret < 0) {
328 PERROR("kernel consumer channel monitor pipe close");
329 }
330 }
331 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
332 ret = close(ustconsumer32_data.channel_monitor_pipe);
333 if (ret < 0) {
334 PERROR("UST consumerd32 channel monitor pipe close");
335 }
336 }
337 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
338 ret = close(ustconsumer64_data.channel_monitor_pipe);
339 if (ret < 0) {
340 PERROR("UST consumerd64 channel monitor pipe close");
341 }
342 }
343 }
344
345 /*
346 * Wait on consumer process termination.
347 *
348 * Need to be called with the consumer data lock held or from a context
349 * ensuring no concurrent access to data (e.g: cleanup).
350 */
351 static void wait_consumer(struct consumer_data *consumer_data)
352 {
353 pid_t ret;
354 int status;
355
356 if (consumer_data->pid <= 0) {
357 return;
358 }
359
360 DBG("Waiting for complete teardown of consumerd (PID: %d)",
361 consumer_data->pid);
362 ret = waitpid(consumer_data->pid, &status, 0);
363 if (ret == -1) {
364 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
365 } else if (!WIFEXITED(status)) {
366 ERR("consumerd termination with error: %d",
367 WEXITSTATUS(ret));
368 }
369 consumer_data->pid = 0;
370 }
371
372 /*
373 * Cleanup the session daemon's data structures.
374 */
375 static void sessiond_cleanup(void)
376 {
377 int ret;
378 struct ltt_session_list *session_list = session_get_list();
379
380 DBG("Cleanup sessiond");
381
382 /*
383 * Close the thread quit pipe. It has already done its job,
384 * since we are now called.
385 */
386 sessiond_close_quit_pipe();
387
388 ret = remove(config.pid_file_path.value);
389 if (ret < 0) {
390 PERROR("remove pidfile %s", config.pid_file_path.value);
391 }
392
393 DBG("Removing sessiond and consumerd content of directory %s",
394 config.rundir.value);
395
396 /* sessiond */
397 DBG("Removing %s", config.pid_file_path.value);
398 (void) unlink(config.pid_file_path.value);
399
400 DBG("Removing %s", config.agent_port_file_path.value);
401 (void) unlink(config.agent_port_file_path.value);
402
403 /* kconsumerd */
404 DBG("Removing %s", kconsumer_data.err_unix_sock_path);
405 (void) unlink(kconsumer_data.err_unix_sock_path);
406
407 DBG("Removing directory %s", config.kconsumerd_path.value);
408 (void) rmdir(config.kconsumerd_path.value);
409
410 /* ust consumerd 32 */
411 DBG("Removing %s", config.consumerd32_err_unix_sock_path.value);
412 (void) unlink(config.consumerd32_err_unix_sock_path.value);
413
414 DBG("Removing directory %s", config.consumerd32_path.value);
415 (void) rmdir(config.consumerd32_path.value);
416
417 /* ust consumerd 64 */
418 DBG("Removing %s", config.consumerd64_err_unix_sock_path.value);
419 (void) unlink(config.consumerd64_err_unix_sock_path.value);
420
421 DBG("Removing directory %s", config.consumerd64_path.value);
422 (void) rmdir(config.consumerd64_path.value);
423
424 pthread_mutex_destroy(&session_list->lock);
425
426 wait_consumer(&kconsumer_data);
427 wait_consumer(&ustconsumer64_data);
428 wait_consumer(&ustconsumer32_data);
429
430 DBG("Cleaning up all agent apps");
431 agent_app_ht_clean();
432
433 DBG("Closing all UST sockets");
434 ust_app_clean_list();
435 buffer_reg_destroy_registries();
436
437 if (is_root && !config.no_kernel) {
438 DBG2("Closing kernel fd");
439 if (kernel_tracer_fd >= 0) {
440 ret = close(kernel_tracer_fd);
441 if (ret) {
442 PERROR("close");
443 }
444 }
445 DBG("Unloading kernel modules");
446 modprobe_remove_lttng_all();
447 free(syscall_table);
448 }
449
450 close_consumer_sockets();
451
452 if (load_info) {
453 load_session_destroy_data(load_info);
454 free(load_info);
455 }
456
457 /*
458 * We do NOT rmdir rundir because there are other processes
459 * using it, for instance lttng-relayd, which can start in
460 * parallel with this teardown.
461 */
462 }
463
464 /*
465 * Cleanup the daemon's option data structures.
466 */
467 static void sessiond_cleanup_options(void)
468 {
469 DBG("Cleaning up options");
470
471 sessiond_config_fini(&config);
472
473 run_as_destroy_worker();
474 }
475
476 /*
477 * Send data on a unix socket using the liblttsessiondcomm API.
478 *
479 * Return lttcomm error code.
480 */
481 static int send_unix_sock(int sock, void *buf, size_t len)
482 {
483 /* Check valid length */
484 if (len == 0) {
485 return -1;
486 }
487
488 return lttcomm_send_unix_sock(sock, buf, len);
489 }
490
491 /*
492 * Free memory of a command context structure.
493 */
494 static void clean_command_ctx(struct command_ctx **cmd_ctx)
495 {
496 DBG("Clean command context structure");
497 if (*cmd_ctx) {
498 if ((*cmd_ctx)->llm) {
499 free((*cmd_ctx)->llm);
500 }
501 if ((*cmd_ctx)->lsm) {
502 free((*cmd_ctx)->lsm);
503 }
504 free(*cmd_ctx);
505 *cmd_ctx = NULL;
506 }
507 }
508
509 /*
510 * Notify UST applications using the shm mmap futex.
511 */
512 static int notify_ust_apps(int active)
513 {
514 char *wait_shm_mmap;
515
516 DBG("Notifying applications of session daemon state: %d", active);
517
518 /* See shm.c for this call implying mmap, shm and futex calls */
519 wait_shm_mmap = shm_ust_get_mmap(config.wait_shm_path.value, is_root);
520 if (wait_shm_mmap == NULL) {
521 goto error;
522 }
523
524 /* Wake waiting process */
525 futex_wait_update((int32_t *) wait_shm_mmap, active);
526
527 /* Apps notified successfully */
528 return 0;
529
530 error:
531 return -1;
532 }
533
534 /*
535 * Setup the outgoing data buffer for the response (llm) by allocating the
536 * right amount of memory and copying the original information from the lsm
537 * structure.
538 *
539 * Return 0 on success, negative value on error.
540 */
541 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
542 const void *payload_buf, size_t payload_len,
543 const void *cmd_header_buf, size_t cmd_header_len)
544 {
545 int ret = 0;
546 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
547 const size_t cmd_header_offset = header_len;
548 const size_t payload_offset = cmd_header_offset + cmd_header_len;
549 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
550
551 cmd_ctx->llm = zmalloc(total_msg_size);
552
553 if (cmd_ctx->llm == NULL) {
554 PERROR("zmalloc");
555 ret = -ENOMEM;
556 goto end;
557 }
558
559 /* Copy common data */
560 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
561 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
562 cmd_ctx->llm->cmd_header_size = cmd_header_len;
563 cmd_ctx->llm->data_size = payload_len;
564 cmd_ctx->lttng_msg_size = total_msg_size;
565
566 /* Copy command header */
567 if (cmd_header_len) {
568 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
569 cmd_header_len);
570 }
571
572 /* Copy payload */
573 if (payload_len) {
574 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
575 payload_len);
576 }
577
578 end:
579 return ret;
580 }
581
582 /*
583 * Version of setup_lttng_msg() without command header.
584 */
585 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
586 void *payload_buf, size_t payload_len)
587 {
588 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
589 }
590 /*
591 * Update the kernel poll set of all channel fd available over all tracing
592 * session. Add the wakeup pipe at the end of the set.
593 */
594 static int update_kernel_poll(struct lttng_poll_event *events)
595 {
596 int ret;
597 struct ltt_kernel_channel *channel;
598 struct ltt_session *session;
599 const struct ltt_session_list *session_list = session_get_list();
600
601 DBG("Updating kernel poll set");
602
603 session_lock_list();
604 cds_list_for_each_entry(session, &session_list->head, list) {
605 if (!session_get(session)) {
606 continue;
607 }
608 session_lock(session);
609 if (session->kernel_session == NULL) {
610 session_unlock(session);
611 session_put(session);
612 continue;
613 }
614
615 cds_list_for_each_entry(channel,
616 &session->kernel_session->channel_list.head, list) {
617 /* Add channel fd to the kernel poll set */
618 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
619 if (ret < 0) {
620 session_unlock(session);
621 session_put(session);
622 goto error;
623 }
624 DBG("Channel fd %d added to kernel set", channel->fd);
625 }
626 session_unlock(session);
627 }
628 session_unlock_list();
629
630 return 0;
631
632 error:
633 session_unlock_list();
634 return -1;
635 }
636
637 /*
638 * Find the channel fd from 'fd' over all tracing session. When found, check
639 * for new channel stream and send those stream fds to the kernel consumer.
640 *
641 * Useful for CPU hotplug feature.
642 */
643 static int update_kernel_stream(int fd)
644 {
645 int ret = 0;
646 struct ltt_session *session;
647 struct ltt_kernel_session *ksess;
648 struct ltt_kernel_channel *channel;
649 const struct ltt_session_list *session_list = session_get_list();
650
651 DBG("Updating kernel streams for channel fd %d", fd);
652
653 session_lock_list();
654 cds_list_for_each_entry(session, &session_list->head, list) {
655 if (!session_get(session)) {
656 continue;
657 }
658 session_lock(session);
659 if (session->kernel_session == NULL) {
660 session_unlock(session);
661 session_put(session);
662 continue;
663 }
664 ksess = session->kernel_session;
665
666 cds_list_for_each_entry(channel,
667 &ksess->channel_list.head, list) {
668 struct lttng_ht_iter iter;
669 struct consumer_socket *socket;
670
671 if (channel->fd != fd) {
672 continue;
673 }
674 DBG("Channel found, updating kernel streams");
675 ret = kernel_open_channel_stream(channel);
676 if (ret < 0) {
677 goto error;
678 }
679 /* Update the stream global counter */
680 ksess->stream_count_global += ret;
681
682 /*
683 * Have we already sent fds to the consumer? If yes, it
684 * means that tracing is started so it is safe to send
685 * our updated stream fds.
686 */
687 if (ksess->consumer_fds_sent != 1
688 || ksess->consumer == NULL) {
689 ret = -1;
690 goto error;
691 }
692
693 rcu_read_lock();
694 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
695 &iter.iter, socket, node.node) {
696 pthread_mutex_lock(socket->lock);
697 ret = kernel_consumer_send_channel_streams(socket,
698 channel, ksess,
699 session->output_traces ? 1 : 0);
700 pthread_mutex_unlock(socket->lock);
701 if (ret < 0) {
702 rcu_read_unlock();
703 goto error;
704 }
705 }
706 rcu_read_unlock();
707 }
708 session_unlock(session);
709 session_put(session);
710 }
711 session_unlock_list();
712 return ret;
713
714 error:
715 session_unlock(session);
716 session_put(session);
717 session_unlock_list();
718 return ret;
719 }
720
721 /*
722 * For each tracing session, update newly registered apps. The session list
723 * lock MUST be acquired before calling this.
724 */
725 static void update_ust_app(int app_sock)
726 {
727 struct ltt_session *sess, *stmp;
728 const struct ltt_session_list *session_list = session_get_list();
729
730 /* Consumer is in an ERROR state. Stop any application update. */
731 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
732 /* Stop the update process since the consumer is dead. */
733 return;
734 }
735
736 /* For all tracing session(s) */
737 cds_list_for_each_entry_safe(sess, stmp, &session_list->head, list) {
738 struct ust_app *app;
739
740 if (!session_get(sess)) {
741 continue;
742 }
743 session_lock(sess);
744 if (!sess->ust_session) {
745 goto unlock_session;
746 }
747
748 rcu_read_lock();
749 assert(app_sock >= 0);
750 app = ust_app_find_by_sock(app_sock);
751 if (app == NULL) {
752 /*
753 * Application can be unregistered before so
754 * this is possible hence simply stopping the
755 * update.
756 */
757 DBG3("UST app update failed to find app sock %d",
758 app_sock);
759 goto unlock_rcu;
760 }
761 ust_app_global_update(sess->ust_session, app);
762 unlock_rcu:
763 rcu_read_unlock();
764 unlock_session:
765 session_unlock(sess);
766 session_put(sess);
767 }
768 }
769
770 /*
771 * This thread manage event coming from the kernel.
772 *
773 * Features supported in this thread:
774 * -) CPU Hotplug
775 */
776 static void *thread_manage_kernel(void *data)
777 {
778 int ret, i, pollfd, update_poll_flag = 1, err = -1;
779 uint32_t revents, nb_fd;
780 char tmp;
781 struct lttng_poll_event events;
782
783 DBG("[thread] Thread manage kernel started");
784
785 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
786
787 /*
788 * This first step of the while is to clean this structure which could free
789 * non NULL pointers so initialize it before the loop.
790 */
791 lttng_poll_init(&events);
792
793 if (testpoint(sessiond_thread_manage_kernel)) {
794 goto error_testpoint;
795 }
796
797 health_code_update();
798
799 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
800 goto error_testpoint;
801 }
802
803 while (1) {
804 health_code_update();
805
806 if (update_poll_flag == 1) {
807 /* Clean events object. We are about to populate it again. */
808 lttng_poll_clean(&events);
809
810 ret = sessiond_set_thread_pollset(&events, 2);
811 if (ret < 0) {
812 goto error_poll_create;
813 }
814
815 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
816 if (ret < 0) {
817 goto error;
818 }
819
820 /* This will add the available kernel channel if any. */
821 ret = update_kernel_poll(&events);
822 if (ret < 0) {
823 goto error;
824 }
825 update_poll_flag = 0;
826 }
827
828 DBG("Thread kernel polling");
829
830 /* Poll infinite value of time */
831 restart:
832 health_poll_entry();
833 ret = lttng_poll_wait(&events, -1);
834 DBG("Thread kernel return from poll on %d fds",
835 LTTNG_POLL_GETNB(&events));
836 health_poll_exit();
837 if (ret < 0) {
838 /*
839 * Restart interrupted system call.
840 */
841 if (errno == EINTR) {
842 goto restart;
843 }
844 goto error;
845 } else if (ret == 0) {
846 /* Should not happen since timeout is infinite */
847 ERR("Return value of poll is 0 with an infinite timeout.\n"
848 "This should not have happened! Continuing...");
849 continue;
850 }
851
852 nb_fd = ret;
853
854 for (i = 0; i < nb_fd; i++) {
855 /* Fetch once the poll data */
856 revents = LTTNG_POLL_GETEV(&events, i);
857 pollfd = LTTNG_POLL_GETFD(&events, i);
858
859 health_code_update();
860
861 if (!revents) {
862 /* No activity for this FD (poll implementation). */
863 continue;
864 }
865
866 /* Thread quit pipe has been closed. Killing thread. */
867 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
868 if (ret) {
869 err = 0;
870 goto exit;
871 }
872
873 /* Check for data on kernel pipe */
874 if (revents & LPOLLIN) {
875 if (pollfd == kernel_poll_pipe[0]) {
876 (void) lttng_read(kernel_poll_pipe[0],
877 &tmp, 1);
878 /*
879 * Ret value is useless here, if this pipe gets any actions an
880 * update is required anyway.
881 */
882 update_poll_flag = 1;
883 continue;
884 } else {
885 /*
886 * New CPU detected by the kernel. Adding kernel stream to
887 * kernel session and updating the kernel consumer
888 */
889 ret = update_kernel_stream(pollfd);
890 if (ret < 0) {
891 continue;
892 }
893 break;
894 }
895 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
896 update_poll_flag = 1;
897 continue;
898 } else {
899 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
900 goto error;
901 }
902 }
903 }
904
905 exit:
906 error:
907 lttng_poll_clean(&events);
908 error_poll_create:
909 error_testpoint:
910 utils_close_pipe(kernel_poll_pipe);
911 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
912 if (err) {
913 health_error();
914 ERR("Health error occurred in %s", __func__);
915 WARN("Kernel thread died unexpectedly. "
916 "Kernel tracing can continue but CPU hotplug is disabled.");
917 }
918 health_unregister(health_sessiond);
919 DBG("Kernel thread dying");
920 return NULL;
921 }
922
923 /*
924 * Signal pthread condition of the consumer data that the thread.
925 */
926 static void signal_consumer_condition(struct consumer_data *data, int state)
927 {
928 pthread_mutex_lock(&data->cond_mutex);
929
930 /*
931 * The state is set before signaling. It can be any value, it's the waiter
932 * job to correctly interpret this condition variable associated to the
933 * consumer pthread_cond.
934 *
935 * A value of 0 means that the corresponding thread of the consumer data
936 * was not started. 1 indicates that the thread has started and is ready
937 * for action. A negative value means that there was an error during the
938 * thread bootstrap.
939 */
940 data->consumer_thread_is_ready = state;
941 (void) pthread_cond_signal(&data->cond);
942
943 pthread_mutex_unlock(&data->cond_mutex);
944 }
945
946 /*
947 * This thread manage the consumer error sent back to the session daemon.
948 */
949 static void *thread_manage_consumer(void *data)
950 {
951 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
952 uint32_t revents, nb_fd;
953 enum lttcomm_return_code code;
954 struct lttng_poll_event events;
955 struct consumer_data *consumer_data = data;
956 struct consumer_socket *cmd_socket_wrapper = NULL;
957
958 DBG("[thread] Manage consumer started");
959
960 rcu_register_thread();
961 rcu_thread_online();
962
963 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
964
965 health_code_update();
966
967 /*
968 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
969 * metadata_sock. Nothing more will be added to this poll set.
970 */
971 ret = sessiond_set_thread_pollset(&events, 3);
972 if (ret < 0) {
973 goto error_poll;
974 }
975
976 /*
977 * The error socket here is already in a listening state which was done
978 * just before spawning this thread to avoid a race between the consumer
979 * daemon exec trying to connect and the listen() call.
980 */
981 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
982 if (ret < 0) {
983 goto error;
984 }
985
986 health_code_update();
987
988 /* Infinite blocking call, waiting for transmission */
989 restart:
990 health_poll_entry();
991
992 if (testpoint(sessiond_thread_manage_consumer)) {
993 goto error;
994 }
995
996 ret = lttng_poll_wait(&events, -1);
997 health_poll_exit();
998 if (ret < 0) {
999 /*
1000 * Restart interrupted system call.
1001 */
1002 if (errno == EINTR) {
1003 goto restart;
1004 }
1005 goto error;
1006 }
1007
1008 nb_fd = ret;
1009
1010 for (i = 0; i < nb_fd; i++) {
1011 /* Fetch once the poll data */
1012 revents = LTTNG_POLL_GETEV(&events, i);
1013 pollfd = LTTNG_POLL_GETFD(&events, i);
1014
1015 health_code_update();
1016
1017 if (!revents) {
1018 /* No activity for this FD (poll implementation). */
1019 continue;
1020 }
1021
1022 /* Thread quit pipe has been closed. Killing thread. */
1023 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1024 if (ret) {
1025 err = 0;
1026 goto exit;
1027 }
1028
1029 /* Event on the registration socket */
1030 if (pollfd == consumer_data->err_sock) {
1031 if (revents & LPOLLIN) {
1032 continue;
1033 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1034 ERR("consumer err socket poll error");
1035 goto error;
1036 } else {
1037 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1038 goto error;
1039 }
1040 }
1041 }
1042
1043 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1044 if (sock < 0) {
1045 goto error;
1046 }
1047
1048 /*
1049 * Set the CLOEXEC flag. Return code is useless because either way, the
1050 * show must go on.
1051 */
1052 (void) utils_set_fd_cloexec(sock);
1053
1054 health_code_update();
1055
1056 DBG2("Receiving code from consumer err_sock");
1057
1058 /* Getting status code from kconsumerd */
1059 ret = lttcomm_recv_unix_sock(sock, &code,
1060 sizeof(enum lttcomm_return_code));
1061 if (ret <= 0) {
1062 goto error;
1063 }
1064
1065 health_code_update();
1066 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1067 ERR("consumer error when waiting for SOCK_READY : %s",
1068 lttcomm_get_readable_code(-code));
1069 goto error;
1070 }
1071
1072 /* Connect both command and metadata sockets. */
1073 consumer_data->cmd_sock =
1074 lttcomm_connect_unix_sock(
1075 consumer_data->cmd_unix_sock_path);
1076 consumer_data->metadata_fd =
1077 lttcomm_connect_unix_sock(
1078 consumer_data->cmd_unix_sock_path);
1079 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1080 PERROR("consumer connect cmd socket");
1081 /* On error, signal condition and quit. */
1082 signal_consumer_condition(consumer_data, -1);
1083 goto error;
1084 }
1085
1086 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1087
1088 /* Create metadata socket lock. */
1089 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1090 if (consumer_data->metadata_sock.lock == NULL) {
1091 PERROR("zmalloc pthread mutex");
1092 goto error;
1093 }
1094 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1095
1096 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1097 DBG("Consumer metadata socket ready (fd: %d)",
1098 consumer_data->metadata_fd);
1099
1100 /*
1101 * Remove the consumerd error sock since we've established a connection.
1102 */
1103 ret = lttng_poll_del(&events, consumer_data->err_sock);
1104 if (ret < 0) {
1105 goto error;
1106 }
1107
1108 /* Add new accepted error socket. */
1109 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1110 if (ret < 0) {
1111 goto error;
1112 }
1113
1114 /* Add metadata socket that is successfully connected. */
1115 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1116 LPOLLIN | LPOLLRDHUP);
1117 if (ret < 0) {
1118 goto error;
1119 }
1120
1121 health_code_update();
1122
1123 /*
1124 * Transfer the write-end of the channel monitoring and rotate pipe
1125 * to the consumer by issuing a SET_CHANNEL_MONITOR_PIPE command.
1126 */
1127 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1128 if (!cmd_socket_wrapper) {
1129 goto error;
1130 }
1131 cmd_socket_wrapper->lock = &consumer_data->lock;
1132
1133 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1134 consumer_data->channel_monitor_pipe);
1135 if (ret) {
1136 goto error;
1137 }
1138
1139 /* Discard the socket wrapper as it is no longer needed. */
1140 consumer_destroy_socket(cmd_socket_wrapper);
1141 cmd_socket_wrapper = NULL;
1142
1143 /* The thread is completely initialized, signal that it is ready. */
1144 signal_consumer_condition(consumer_data, 1);
1145
1146 /* Infinite blocking call, waiting for transmission */
1147 restart_poll:
1148 while (1) {
1149 health_code_update();
1150
1151 /* Exit the thread because the thread quit pipe has been triggered. */
1152 if (should_quit) {
1153 /* Not a health error. */
1154 err = 0;
1155 goto exit;
1156 }
1157
1158 health_poll_entry();
1159 ret = lttng_poll_wait(&events, -1);
1160 health_poll_exit();
1161 if (ret < 0) {
1162 /*
1163 * Restart interrupted system call.
1164 */
1165 if (errno == EINTR) {
1166 goto restart_poll;
1167 }
1168 goto error;
1169 }
1170
1171 nb_fd = ret;
1172
1173 for (i = 0; i < nb_fd; i++) {
1174 /* Fetch once the poll data */
1175 revents = LTTNG_POLL_GETEV(&events, i);
1176 pollfd = LTTNG_POLL_GETFD(&events, i);
1177
1178 health_code_update();
1179
1180 if (!revents) {
1181 /* No activity for this FD (poll implementation). */
1182 continue;
1183 }
1184
1185 /*
1186 * Thread quit pipe has been triggered, flag that we should stop
1187 * but continue the current loop to handle potential data from
1188 * consumer.
1189 */
1190 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1191
1192 if (pollfd == sock) {
1193 /* Event on the consumerd socket */
1194 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1195 && !(revents & LPOLLIN)) {
1196 ERR("consumer err socket second poll error");
1197 goto error;
1198 }
1199 health_code_update();
1200 /* Wait for any kconsumerd error */
1201 ret = lttcomm_recv_unix_sock(sock, &code,
1202 sizeof(enum lttcomm_return_code));
1203 if (ret <= 0) {
1204 ERR("consumer closed the command socket");
1205 goto error;
1206 }
1207
1208 ERR("consumer return code : %s",
1209 lttcomm_get_readable_code(-code));
1210
1211 goto exit;
1212 } else if (pollfd == consumer_data->metadata_fd) {
1213 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1214 && !(revents & LPOLLIN)) {
1215 ERR("consumer err metadata socket second poll error");
1216 goto error;
1217 }
1218 /* UST metadata requests */
1219 ret = ust_consumer_metadata_request(
1220 &consumer_data->metadata_sock);
1221 if (ret < 0) {
1222 ERR("Handling metadata request");
1223 goto error;
1224 }
1225 }
1226 /* No need for an else branch all FDs are tested prior. */
1227 }
1228 health_code_update();
1229 }
1230
1231 exit:
1232 error:
1233 /*
1234 * We lock here because we are about to close the sockets and some other
1235 * thread might be using them so get exclusive access which will abort all
1236 * other consumer command by other threads.
1237 */
1238 pthread_mutex_lock(&consumer_data->lock);
1239
1240 /* Immediately set the consumerd state to stopped */
1241 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1242 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1243 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1244 consumer_data->type == LTTNG_CONSUMER32_UST) {
1245 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1246 } else {
1247 /* Code flow error... */
1248 assert(0);
1249 }
1250
1251 if (consumer_data->err_sock >= 0) {
1252 ret = close(consumer_data->err_sock);
1253 if (ret) {
1254 PERROR("close");
1255 }
1256 consumer_data->err_sock = -1;
1257 }
1258 if (consumer_data->cmd_sock >= 0) {
1259 ret = close(consumer_data->cmd_sock);
1260 if (ret) {
1261 PERROR("close");
1262 }
1263 consumer_data->cmd_sock = -1;
1264 }
1265 if (consumer_data->metadata_sock.fd_ptr &&
1266 *consumer_data->metadata_sock.fd_ptr >= 0) {
1267 ret = close(*consumer_data->metadata_sock.fd_ptr);
1268 if (ret) {
1269 PERROR("close");
1270 }
1271 }
1272 if (sock >= 0) {
1273 ret = close(sock);
1274 if (ret) {
1275 PERROR("close");
1276 }
1277 }
1278
1279 unlink(consumer_data->err_unix_sock_path);
1280 unlink(consumer_data->cmd_unix_sock_path);
1281 pthread_mutex_unlock(&consumer_data->lock);
1282
1283 /* Cleanup metadata socket mutex. */
1284 if (consumer_data->metadata_sock.lock) {
1285 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1286 free(consumer_data->metadata_sock.lock);
1287 }
1288 lttng_poll_clean(&events);
1289
1290 if (cmd_socket_wrapper) {
1291 consumer_destroy_socket(cmd_socket_wrapper);
1292 }
1293 error_poll:
1294 if (err) {
1295 health_error();
1296 ERR("Health error occurred in %s", __func__);
1297 }
1298 health_unregister(health_sessiond);
1299 DBG("consumer thread cleanup completed");
1300
1301 rcu_thread_offline();
1302 rcu_unregister_thread();
1303
1304 return NULL;
1305 }
1306
1307 /*
1308 * This thread receives application command sockets (FDs) on the
1309 * apps_cmd_pipe and waits (polls) on them until they are closed
1310 * or an error occurs.
1311 *
1312 * At that point, it flushes the data (tracing and metadata) associated
1313 * with this application and tears down ust app sessions and other
1314 * associated data structures through ust_app_unregister().
1315 *
1316 * Note that this thread never sends commands to the applications
1317 * through the command sockets; it merely listens for hang-ups
1318 * and errors on those sockets and cleans-up as they occur.
1319 */
1320 static void *thread_manage_apps(void *data)
1321 {
1322 int i, ret, pollfd, err = -1;
1323 ssize_t size_ret;
1324 uint32_t revents, nb_fd;
1325 struct lttng_poll_event events;
1326
1327 DBG("[thread] Manage application started");
1328
1329 rcu_register_thread();
1330 rcu_thread_online();
1331
1332 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1333
1334 if (testpoint(sessiond_thread_manage_apps)) {
1335 goto error_testpoint;
1336 }
1337
1338 health_code_update();
1339
1340 ret = sessiond_set_thread_pollset(&events, 2);
1341 if (ret < 0) {
1342 goto error_poll_create;
1343 }
1344
1345 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1346 if (ret < 0) {
1347 goto error;
1348 }
1349
1350 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1351 goto error;
1352 }
1353
1354 health_code_update();
1355
1356 while (1) {
1357 DBG("Apps thread polling");
1358
1359 /* Inifinite blocking call, waiting for transmission */
1360 restart:
1361 health_poll_entry();
1362 ret = lttng_poll_wait(&events, -1);
1363 DBG("Apps thread return from poll on %d fds",
1364 LTTNG_POLL_GETNB(&events));
1365 health_poll_exit();
1366 if (ret < 0) {
1367 /*
1368 * Restart interrupted system call.
1369 */
1370 if (errno == EINTR) {
1371 goto restart;
1372 }
1373 goto error;
1374 }
1375
1376 nb_fd = ret;
1377
1378 for (i = 0; i < nb_fd; i++) {
1379 /* Fetch once the poll data */
1380 revents = LTTNG_POLL_GETEV(&events, i);
1381 pollfd = LTTNG_POLL_GETFD(&events, i);
1382
1383 health_code_update();
1384
1385 if (!revents) {
1386 /* No activity for this FD (poll implementation). */
1387 continue;
1388 }
1389
1390 /* Thread quit pipe has been closed. Killing thread. */
1391 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1392 if (ret) {
1393 err = 0;
1394 goto exit;
1395 }
1396
1397 /* Inspect the apps cmd pipe */
1398 if (pollfd == apps_cmd_pipe[0]) {
1399 if (revents & LPOLLIN) {
1400 int sock;
1401
1402 /* Empty pipe */
1403 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1404 if (size_ret < sizeof(sock)) {
1405 PERROR("read apps cmd pipe");
1406 goto error;
1407 }
1408
1409 health_code_update();
1410
1411 /*
1412 * Since this is a command socket (write then read),
1413 * we only monitor the error events of the socket.
1414 */
1415 ret = lttng_poll_add(&events, sock,
1416 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1417 if (ret < 0) {
1418 goto error;
1419 }
1420
1421 DBG("Apps with sock %d added to poll set", sock);
1422 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1423 ERR("Apps command pipe error");
1424 goto error;
1425 } else {
1426 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1427 goto error;
1428 }
1429 } else {
1430 /*
1431 * At this point, we know that a registered application made
1432 * the event at poll_wait.
1433 */
1434 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1435 /* Removing from the poll set */
1436 ret = lttng_poll_del(&events, pollfd);
1437 if (ret < 0) {
1438 goto error;
1439 }
1440
1441 /* Socket closed on remote end. */
1442 ust_app_unregister(pollfd);
1443 } else {
1444 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1445 goto error;
1446 }
1447 }
1448
1449 health_code_update();
1450 }
1451 }
1452
1453 exit:
1454 error:
1455 lttng_poll_clean(&events);
1456 error_poll_create:
1457 error_testpoint:
1458 utils_close_pipe(apps_cmd_pipe);
1459 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1460
1461 /*
1462 * We don't clean the UST app hash table here since already registered
1463 * applications can still be controlled so let them be until the session
1464 * daemon dies or the applications stop.
1465 */
1466
1467 if (err) {
1468 health_error();
1469 ERR("Health error occurred in %s", __func__);
1470 }
1471 health_unregister(health_sessiond);
1472 DBG("Application communication apps thread cleanup complete");
1473 rcu_thread_offline();
1474 rcu_unregister_thread();
1475 return NULL;
1476 }
1477
1478 /*
1479 * Send a socket to a thread This is called from the dispatch UST registration
1480 * thread once all sockets are set for the application.
1481 *
1482 * The sock value can be invalid, we don't really care, the thread will handle
1483 * it and make the necessary cleanup if so.
1484 *
1485 * On success, return 0 else a negative value being the errno message of the
1486 * write().
1487 */
1488 static int send_socket_to_thread(int fd, int sock)
1489 {
1490 ssize_t ret;
1491
1492 /*
1493 * It's possible that the FD is set as invalid with -1 concurrently just
1494 * before calling this function being a shutdown state of the thread.
1495 */
1496 if (fd < 0) {
1497 ret = -EBADF;
1498 goto error;
1499 }
1500
1501 ret = lttng_write(fd, &sock, sizeof(sock));
1502 if (ret < sizeof(sock)) {
1503 PERROR("write apps pipe %d", fd);
1504 if (ret < 0) {
1505 ret = -errno;
1506 }
1507 goto error;
1508 }
1509
1510 /* All good. Don't send back the write positive ret value. */
1511 ret = 0;
1512 error:
1513 return (int) ret;
1514 }
1515
1516 /*
1517 * Sanitize the wait queue of the dispatch registration thread meaning removing
1518 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1519 * notify socket is never received.
1520 */
1521 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1522 {
1523 int ret, nb_fd = 0, i;
1524 unsigned int fd_added = 0;
1525 struct lttng_poll_event events;
1526 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1527
1528 assert(wait_queue);
1529
1530 lttng_poll_init(&events);
1531
1532 /* Just skip everything for an empty queue. */
1533 if (!wait_queue->count) {
1534 goto end;
1535 }
1536
1537 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1538 if (ret < 0) {
1539 goto error_create;
1540 }
1541
1542 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1543 &wait_queue->head, head) {
1544 assert(wait_node->app);
1545 ret = lttng_poll_add(&events, wait_node->app->sock,
1546 LPOLLHUP | LPOLLERR);
1547 if (ret < 0) {
1548 goto error;
1549 }
1550
1551 fd_added = 1;
1552 }
1553
1554 if (!fd_added) {
1555 goto end;
1556 }
1557
1558 /*
1559 * Poll but don't block so we can quickly identify the faulty events and
1560 * clean them afterwards from the wait queue.
1561 */
1562 ret = lttng_poll_wait(&events, 0);
1563 if (ret < 0) {
1564 goto error;
1565 }
1566 nb_fd = ret;
1567
1568 for (i = 0; i < nb_fd; i++) {
1569 /* Get faulty FD. */
1570 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1571 int pollfd = LTTNG_POLL_GETFD(&events, i);
1572
1573 if (!revents) {
1574 /* No activity for this FD (poll implementation). */
1575 continue;
1576 }
1577
1578 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1579 &wait_queue->head, head) {
1580 if (pollfd == wait_node->app->sock &&
1581 (revents & (LPOLLHUP | LPOLLERR))) {
1582 cds_list_del(&wait_node->head);
1583 wait_queue->count--;
1584 ust_app_destroy(wait_node->app);
1585 free(wait_node);
1586 /*
1587 * Silence warning of use-after-free in
1588 * cds_list_for_each_entry_safe which uses
1589 * __typeof__(*wait_node).
1590 */
1591 wait_node = NULL;
1592 break;
1593 } else {
1594 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1595 goto error;
1596 }
1597 }
1598 }
1599
1600 if (nb_fd > 0) {
1601 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1602 }
1603
1604 end:
1605 lttng_poll_clean(&events);
1606 return;
1607
1608 error:
1609 lttng_poll_clean(&events);
1610 error_create:
1611 ERR("Unable to sanitize wait queue");
1612 return;
1613 }
1614
1615 /*
1616 * Dispatch request from the registration threads to the application
1617 * communication thread.
1618 */
1619 static void *thread_dispatch_ust_registration(void *data)
1620 {
1621 int ret, err = -1;
1622 struct cds_wfcq_node *node;
1623 struct ust_command *ust_cmd = NULL;
1624 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1625 struct ust_reg_wait_queue wait_queue = {
1626 .count = 0,
1627 };
1628
1629 rcu_register_thread();
1630
1631 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1632
1633 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1634 goto error_testpoint;
1635 }
1636
1637 health_code_update();
1638
1639 CDS_INIT_LIST_HEAD(&wait_queue.head);
1640
1641 DBG("[thread] Dispatch UST command started");
1642
1643 for (;;) {
1644 health_code_update();
1645
1646 /* Atomically prepare the queue futex */
1647 futex_nto1_prepare(&ust_cmd_queue.futex);
1648
1649 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1650 break;
1651 }
1652
1653 do {
1654 struct ust_app *app = NULL;
1655 ust_cmd = NULL;
1656
1657 /*
1658 * Make sure we don't have node(s) that have hung up before receiving
1659 * the notify socket. This is to clean the list in order to avoid
1660 * memory leaks from notify socket that are never seen.
1661 */
1662 sanitize_wait_queue(&wait_queue);
1663
1664 health_code_update();
1665 /* Dequeue command for registration */
1666 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1667 if (node == NULL) {
1668 DBG("Woken up but nothing in the UST command queue");
1669 /* Continue thread execution */
1670 break;
1671 }
1672
1673 ust_cmd = caa_container_of(node, struct ust_command, node);
1674
1675 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1676 " gid:%d sock:%d name:%s (version %d.%d)",
1677 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1678 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1679 ust_cmd->sock, ust_cmd->reg_msg.name,
1680 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1681
1682 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1683 wait_node = zmalloc(sizeof(*wait_node));
1684 if (!wait_node) {
1685 PERROR("zmalloc wait_node dispatch");
1686 ret = close(ust_cmd->sock);
1687 if (ret < 0) {
1688 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1689 }
1690 lttng_fd_put(LTTNG_FD_APPS, 1);
1691 free(ust_cmd);
1692 goto error;
1693 }
1694 CDS_INIT_LIST_HEAD(&wait_node->head);
1695
1696 /* Create application object if socket is CMD. */
1697 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1698 ust_cmd->sock);
1699 if (!wait_node->app) {
1700 ret = close(ust_cmd->sock);
1701 if (ret < 0) {
1702 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1703 }
1704 lttng_fd_put(LTTNG_FD_APPS, 1);
1705 free(wait_node);
1706 free(ust_cmd);
1707 continue;
1708 }
1709 /*
1710 * Add application to the wait queue so we can set the notify
1711 * socket before putting this object in the global ht.
1712 */
1713 cds_list_add(&wait_node->head, &wait_queue.head);
1714 wait_queue.count++;
1715
1716 free(ust_cmd);
1717 /*
1718 * We have to continue here since we don't have the notify
1719 * socket and the application MUST be added to the hash table
1720 * only at that moment.
1721 */
1722 continue;
1723 } else {
1724 /*
1725 * Look for the application in the local wait queue and set the
1726 * notify socket if found.
1727 */
1728 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1729 &wait_queue.head, head) {
1730 health_code_update();
1731 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1732 wait_node->app->notify_sock = ust_cmd->sock;
1733 cds_list_del(&wait_node->head);
1734 wait_queue.count--;
1735 app = wait_node->app;
1736 free(wait_node);
1737 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1738 break;
1739 }
1740 }
1741
1742 /*
1743 * With no application at this stage the received socket is
1744 * basically useless so close it before we free the cmd data
1745 * structure for good.
1746 */
1747 if (!app) {
1748 ret = close(ust_cmd->sock);
1749 if (ret < 0) {
1750 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1751 }
1752 lttng_fd_put(LTTNG_FD_APPS, 1);
1753 }
1754 free(ust_cmd);
1755 }
1756
1757 if (app) {
1758 /*
1759 * @session_lock_list
1760 *
1761 * Lock the global session list so from the register up to the
1762 * registration done message, no thread can see the application
1763 * and change its state.
1764 */
1765 session_lock_list();
1766 rcu_read_lock();
1767
1768 /*
1769 * Add application to the global hash table. This needs to be
1770 * done before the update to the UST registry can locate the
1771 * application.
1772 */
1773 ust_app_add(app);
1774
1775 /* Set app version. This call will print an error if needed. */
1776 (void) ust_app_version(app);
1777
1778 /* Send notify socket through the notify pipe. */
1779 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1780 app->notify_sock);
1781 if (ret < 0) {
1782 rcu_read_unlock();
1783 session_unlock_list();
1784 /*
1785 * No notify thread, stop the UST tracing. However, this is
1786 * not an internal error of the this thread thus setting
1787 * the health error code to a normal exit.
1788 */
1789 err = 0;
1790 goto error;
1791 }
1792
1793 /*
1794 * Update newly registered application with the tracing
1795 * registry info already enabled information.
1796 */
1797 update_ust_app(app->sock);
1798
1799 /*
1800 * Don't care about return value. Let the manage apps threads
1801 * handle app unregistration upon socket close.
1802 */
1803 (void) ust_app_register_done(app);
1804
1805 /*
1806 * Even if the application socket has been closed, send the app
1807 * to the thread and unregistration will take place at that
1808 * place.
1809 */
1810 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1811 if (ret < 0) {
1812 rcu_read_unlock();
1813 session_unlock_list();
1814 /*
1815 * No apps. thread, stop the UST tracing. However, this is
1816 * not an internal error of the this thread thus setting
1817 * the health error code to a normal exit.
1818 */
1819 err = 0;
1820 goto error;
1821 }
1822
1823 rcu_read_unlock();
1824 session_unlock_list();
1825 }
1826 } while (node != NULL);
1827
1828 health_poll_entry();
1829 /* Futex wait on queue. Blocking call on futex() */
1830 futex_nto1_wait(&ust_cmd_queue.futex);
1831 health_poll_exit();
1832 }
1833 /* Normal exit, no error */
1834 err = 0;
1835
1836 error:
1837 /* Clean up wait queue. */
1838 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1839 &wait_queue.head, head) {
1840 cds_list_del(&wait_node->head);
1841 wait_queue.count--;
1842 free(wait_node);
1843 }
1844
1845 /* Empty command queue. */
1846 for (;;) {
1847 /* Dequeue command for registration */
1848 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1849 if (node == NULL) {
1850 break;
1851 }
1852 ust_cmd = caa_container_of(node, struct ust_command, node);
1853 ret = close(ust_cmd->sock);
1854 if (ret < 0) {
1855 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
1856 }
1857 lttng_fd_put(LTTNG_FD_APPS, 1);
1858 free(ust_cmd);
1859 }
1860
1861 error_testpoint:
1862 DBG("Dispatch thread dying");
1863 if (err) {
1864 health_error();
1865 ERR("Health error occurred in %s", __func__);
1866 }
1867 health_unregister(health_sessiond);
1868 rcu_unregister_thread();
1869 return NULL;
1870 }
1871
1872 /*
1873 * This thread manage application registration.
1874 */
1875 static void *thread_registration_apps(void *data)
1876 {
1877 int sock = -1, i, ret, pollfd, err = -1;
1878 uint32_t revents, nb_fd;
1879 struct lttng_poll_event events;
1880 /*
1881 * Get allocated in this thread, enqueued to a global queue, dequeued and
1882 * freed in the manage apps thread.
1883 */
1884 struct ust_command *ust_cmd = NULL;
1885
1886 DBG("[thread] Manage application registration started");
1887
1888 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1889
1890 if (testpoint(sessiond_thread_registration_apps)) {
1891 goto error_testpoint;
1892 }
1893
1894 ret = lttcomm_listen_unix_sock(apps_sock);
1895 if (ret < 0) {
1896 goto error_listen;
1897 }
1898
1899 /*
1900 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1901 * more will be added to this poll set.
1902 */
1903 ret = sessiond_set_thread_pollset(&events, 2);
1904 if (ret < 0) {
1905 goto error_create_poll;
1906 }
1907
1908 /* Add the application registration socket */
1909 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1910 if (ret < 0) {
1911 goto error_poll_add;
1912 }
1913
1914 /* Notify all applications to register */
1915 ret = notify_ust_apps(1);
1916 if (ret < 0) {
1917 ERR("Failed to notify applications or create the wait shared memory.\n"
1918 "Execution continues but there might be problem for already\n"
1919 "running applications that wishes to register.");
1920 }
1921
1922 while (1) {
1923 DBG("Accepting application registration");
1924
1925 /* Inifinite blocking call, waiting for transmission */
1926 restart:
1927 health_poll_entry();
1928 ret = lttng_poll_wait(&events, -1);
1929 health_poll_exit();
1930 if (ret < 0) {
1931 /*
1932 * Restart interrupted system call.
1933 */
1934 if (errno == EINTR) {
1935 goto restart;
1936 }
1937 goto error;
1938 }
1939
1940 nb_fd = ret;
1941
1942 for (i = 0; i < nb_fd; i++) {
1943 health_code_update();
1944
1945 /* Fetch once the poll data */
1946 revents = LTTNG_POLL_GETEV(&events, i);
1947 pollfd = LTTNG_POLL_GETFD(&events, i);
1948
1949 if (!revents) {
1950 /* No activity for this FD (poll implementation). */
1951 continue;
1952 }
1953
1954 /* Thread quit pipe has been closed. Killing thread. */
1955 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1956 if (ret) {
1957 err = 0;
1958 goto exit;
1959 }
1960
1961 /* Event on the registration socket */
1962 if (pollfd == apps_sock) {
1963 if (revents & LPOLLIN) {
1964 sock = lttcomm_accept_unix_sock(apps_sock);
1965 if (sock < 0) {
1966 goto error;
1967 }
1968
1969 /*
1970 * Set socket timeout for both receiving and ending.
1971 * app_socket_timeout is in seconds, whereas
1972 * lttcomm_setsockopt_rcv_timeout and
1973 * lttcomm_setsockopt_snd_timeout expect msec as
1974 * parameter.
1975 */
1976 if (config.app_socket_timeout >= 0) {
1977 (void) lttcomm_setsockopt_rcv_timeout(sock,
1978 config.app_socket_timeout * 1000);
1979 (void) lttcomm_setsockopt_snd_timeout(sock,
1980 config.app_socket_timeout * 1000);
1981 }
1982
1983 /*
1984 * Set the CLOEXEC flag. Return code is useless because
1985 * either way, the show must go on.
1986 */
1987 (void) utils_set_fd_cloexec(sock);
1988
1989 /* Create UST registration command for enqueuing */
1990 ust_cmd = zmalloc(sizeof(struct ust_command));
1991 if (ust_cmd == NULL) {
1992 PERROR("ust command zmalloc");
1993 ret = close(sock);
1994 if (ret) {
1995 PERROR("close");
1996 }
1997 goto error;
1998 }
1999
2000 /*
2001 * Using message-based transmissions to ensure we don't
2002 * have to deal with partially received messages.
2003 */
2004 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2005 if (ret < 0) {
2006 ERR("Exhausted file descriptors allowed for applications.");
2007 free(ust_cmd);
2008 ret = close(sock);
2009 if (ret) {
2010 PERROR("close");
2011 }
2012 sock = -1;
2013 continue;
2014 }
2015
2016 health_code_update();
2017 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2018 if (ret < 0) {
2019 free(ust_cmd);
2020 /* Close socket of the application. */
2021 ret = close(sock);
2022 if (ret) {
2023 PERROR("close");
2024 }
2025 lttng_fd_put(LTTNG_FD_APPS, 1);
2026 sock = -1;
2027 continue;
2028 }
2029 health_code_update();
2030
2031 ust_cmd->sock = sock;
2032 sock = -1;
2033
2034 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2035 " gid:%d sock:%d name:%s (version %d.%d)",
2036 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2037 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2038 ust_cmd->sock, ust_cmd->reg_msg.name,
2039 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2040
2041 /*
2042 * Lock free enqueue the registration request. The red pill
2043 * has been taken! This apps will be part of the *system*.
2044 */
2045 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2046
2047 /*
2048 * Wake the registration queue futex. Implicit memory
2049 * barrier with the exchange in cds_wfcq_enqueue.
2050 */
2051 futex_nto1_wake(&ust_cmd_queue.futex);
2052 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2053 ERR("Register apps socket poll error");
2054 goto error;
2055 } else {
2056 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2057 goto error;
2058 }
2059 }
2060 }
2061 }
2062
2063 exit:
2064 error:
2065 /* Notify that the registration thread is gone */
2066 notify_ust_apps(0);
2067
2068 if (apps_sock >= 0) {
2069 ret = close(apps_sock);
2070 if (ret) {
2071 PERROR("close");
2072 }
2073 }
2074 if (sock >= 0) {
2075 ret = close(sock);
2076 if (ret) {
2077 PERROR("close");
2078 }
2079 lttng_fd_put(LTTNG_FD_APPS, 1);
2080 }
2081 unlink(config.apps_unix_sock_path.value);
2082
2083 error_poll_add:
2084 lttng_poll_clean(&events);
2085 error_listen:
2086 error_create_poll:
2087 error_testpoint:
2088 DBG("UST Registration thread cleanup complete");
2089 if (err) {
2090 health_error();
2091 ERR("Health error occurred in %s", __func__);
2092 }
2093 health_unregister(health_sessiond);
2094
2095 return NULL;
2096 }
2097
2098 /*
2099 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2100 * exec or it will fails.
2101 */
2102 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2103 {
2104 int ret, clock_ret;
2105 struct timespec timeout;
2106
2107 /*
2108 * Make sure we set the readiness flag to 0 because we are NOT ready.
2109 * This access to consumer_thread_is_ready does not need to be
2110 * protected by consumer_data.cond_mutex (yet) since the consumer
2111 * management thread has not been started at this point.
2112 */
2113 consumer_data->consumer_thread_is_ready = 0;
2114
2115 /* Setup pthread condition */
2116 ret = pthread_condattr_init(&consumer_data->condattr);
2117 if (ret) {
2118 errno = ret;
2119 PERROR("pthread_condattr_init consumer data");
2120 goto error;
2121 }
2122
2123 /*
2124 * Set the monotonic clock in order to make sure we DO NOT jump in time
2125 * between the clock_gettime() call and the timedwait call. See bug #324
2126 * for a more details and how we noticed it.
2127 */
2128 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2129 if (ret) {
2130 errno = ret;
2131 PERROR("pthread_condattr_setclock consumer data");
2132 goto error;
2133 }
2134
2135 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2136 if (ret) {
2137 errno = ret;
2138 PERROR("pthread_cond_init consumer data");
2139 goto error;
2140 }
2141
2142 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2143 thread_manage_consumer, consumer_data);
2144 if (ret) {
2145 errno = ret;
2146 PERROR("pthread_create consumer");
2147 ret = -1;
2148 goto error;
2149 }
2150
2151 /* We are about to wait on a pthread condition */
2152 pthread_mutex_lock(&consumer_data->cond_mutex);
2153
2154 /* Get time for sem_timedwait absolute timeout */
2155 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2156 /*
2157 * Set the timeout for the condition timed wait even if the clock gettime
2158 * call fails since we might loop on that call and we want to avoid to
2159 * increment the timeout too many times.
2160 */
2161 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2162
2163 /*
2164 * The following loop COULD be skipped in some conditions so this is why we
2165 * set ret to 0 in order to make sure at least one round of the loop is
2166 * done.
2167 */
2168 ret = 0;
2169
2170 /*
2171 * Loop until the condition is reached or when a timeout is reached. Note
2172 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2173 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2174 * possible. This loop does not take any chances and works with both of
2175 * them.
2176 */
2177 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2178 if (clock_ret < 0) {
2179 PERROR("clock_gettime spawn consumer");
2180 /* Infinite wait for the consumerd thread to be ready */
2181 ret = pthread_cond_wait(&consumer_data->cond,
2182 &consumer_data->cond_mutex);
2183 } else {
2184 ret = pthread_cond_timedwait(&consumer_data->cond,
2185 &consumer_data->cond_mutex, &timeout);
2186 }
2187 }
2188
2189 /* Release the pthread condition */
2190 pthread_mutex_unlock(&consumer_data->cond_mutex);
2191
2192 if (ret != 0) {
2193 errno = ret;
2194 if (ret == ETIMEDOUT) {
2195 int pth_ret;
2196
2197 /*
2198 * Call has timed out so we kill the kconsumerd_thread and return
2199 * an error.
2200 */
2201 ERR("Condition timed out. The consumer thread was never ready."
2202 " Killing it");
2203 pth_ret = pthread_cancel(consumer_data->thread);
2204 if (pth_ret < 0) {
2205 PERROR("pthread_cancel consumer thread");
2206 }
2207 } else {
2208 PERROR("pthread_cond_wait failed consumer thread");
2209 }
2210 /* Caller is expecting a negative value on failure. */
2211 ret = -1;
2212 goto error;
2213 }
2214
2215 pthread_mutex_lock(&consumer_data->pid_mutex);
2216 if (consumer_data->pid == 0) {
2217 ERR("Consumerd did not start");
2218 pthread_mutex_unlock(&consumer_data->pid_mutex);
2219 goto error;
2220 }
2221 pthread_mutex_unlock(&consumer_data->pid_mutex);
2222
2223 return 0;
2224
2225 error:
2226 return ret;
2227 }
2228
2229 /*
2230 * Join consumer thread
2231 */
2232 static int join_consumer_thread(struct consumer_data *consumer_data)
2233 {
2234 void *status;
2235
2236 /* Consumer pid must be a real one. */
2237 if (consumer_data->pid > 0) {
2238 int ret;
2239 ret = kill(consumer_data->pid, SIGTERM);
2240 if (ret) {
2241 PERROR("Error killing consumer daemon");
2242 return ret;
2243 }
2244 return pthread_join(consumer_data->thread, &status);
2245 } else {
2246 return 0;
2247 }
2248 }
2249
2250 /*
2251 * Fork and exec a consumer daemon (consumerd).
2252 *
2253 * Return pid if successful else -1.
2254 */
2255 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2256 {
2257 int ret;
2258 pid_t pid;
2259 const char *consumer_to_use;
2260 const char *verbosity;
2261 struct stat st;
2262
2263 DBG("Spawning consumerd");
2264
2265 pid = fork();
2266 if (pid == 0) {
2267 /*
2268 * Exec consumerd.
2269 */
2270 if (config.verbose_consumer) {
2271 verbosity = "--verbose";
2272 } else if (lttng_opt_quiet) {
2273 verbosity = "--quiet";
2274 } else {
2275 verbosity = "";
2276 }
2277
2278 switch (consumer_data->type) {
2279 case LTTNG_CONSUMER_KERNEL:
2280 /*
2281 * Find out which consumerd to execute. We will first try the
2282 * 64-bit path, then the sessiond's installation directory, and
2283 * fallback on the 32-bit one,
2284 */
2285 DBG3("Looking for a kernel consumer at these locations:");
2286 DBG3(" 1) %s", config.consumerd64_bin_path.value ? : "NULL");
2287 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
2288 DBG3(" 3) %s", config.consumerd32_bin_path.value ? : "NULL");
2289 if (stat(config.consumerd64_bin_path.value, &st) == 0) {
2290 DBG3("Found location #1");
2291 consumer_to_use = config.consumerd64_bin_path.value;
2292 } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
2293 DBG3("Found location #2");
2294 consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
2295 } else if (config.consumerd32_bin_path.value &&
2296 stat(config.consumerd32_bin_path.value, &st) == 0) {
2297 DBG3("Found location #3");
2298 consumer_to_use = config.consumerd32_bin_path.value;
2299 } else {
2300 DBG("Could not find any valid consumerd executable");
2301 ret = -EINVAL;
2302 goto error;
2303 }
2304 DBG("Using kernel consumer at: %s", consumer_to_use);
2305 (void) execl(consumer_to_use,
2306 "lttng-consumerd", verbosity, "-k",
2307 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2308 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2309 "--group", config.tracing_group_name.value,
2310 NULL);
2311 break;
2312 case LTTNG_CONSUMER64_UST:
2313 {
2314 if (config.consumerd64_lib_dir.value) {
2315 char *tmp;
2316 size_t tmplen;
2317 char *tmpnew;
2318
2319 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2320 if (!tmp) {
2321 tmp = "";
2322 }
2323 tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2324 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2325 if (!tmpnew) {
2326 ret = -ENOMEM;
2327 goto error;
2328 }
2329 strcat(tmpnew, config.consumerd64_lib_dir.value);
2330 if (tmp[0] != '\0') {
2331 strcat(tmpnew, ":");
2332 strcat(tmpnew, tmp);
2333 }
2334 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2335 free(tmpnew);
2336 if (ret) {
2337 ret = -errno;
2338 goto error;
2339 }
2340 }
2341 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2342 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2343 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2344 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2345 "--group", config.tracing_group_name.value,
2346 NULL);
2347 break;
2348 }
2349 case LTTNG_CONSUMER32_UST:
2350 {
2351 if (config.consumerd32_lib_dir.value) {
2352 char *tmp;
2353 size_t tmplen;
2354 char *tmpnew;
2355
2356 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2357 if (!tmp) {
2358 tmp = "";
2359 }
2360 tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2361 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2362 if (!tmpnew) {
2363 ret = -ENOMEM;
2364 goto error;
2365 }
2366 strcat(tmpnew, config.consumerd32_lib_dir.value);
2367 if (tmp[0] != '\0') {
2368 strcat(tmpnew, ":");
2369 strcat(tmpnew, tmp);
2370 }
2371 ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
2372 free(tmpnew);
2373 if (ret) {
2374 ret = -errno;
2375 goto error;
2376 }
2377 }
2378 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2379 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2380 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2381 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2382 "--group", config.tracing_group_name.value,
2383 NULL);
2384 break;
2385 }
2386 default:
2387 ERR("unknown consumer type");
2388 errno = 0;
2389 }
2390 if (errno != 0) {
2391 PERROR("Consumer execl()");
2392 }
2393 /* Reaching this point, we got a failure on our execl(). */
2394 exit(EXIT_FAILURE);
2395 } else if (pid > 0) {
2396 ret = pid;
2397 } else {
2398 PERROR("start consumer fork");
2399 ret = -errno;
2400 }
2401 error:
2402 return ret;
2403 }
2404
2405 /*
2406 * Spawn the consumerd daemon and session daemon thread.
2407 */
2408 static int start_consumerd(struct consumer_data *consumer_data)
2409 {
2410 int ret;
2411
2412 /*
2413 * Set the listen() state on the socket since there is a possible race
2414 * between the exec() of the consumer daemon and this call if place in the
2415 * consumer thread. See bug #366 for more details.
2416 */
2417 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2418 if (ret < 0) {
2419 goto error;
2420 }
2421
2422 pthread_mutex_lock(&consumer_data->pid_mutex);
2423 if (consumer_data->pid != 0) {
2424 pthread_mutex_unlock(&consumer_data->pid_mutex);
2425 goto end;
2426 }
2427
2428 ret = spawn_consumerd(consumer_data);
2429 if (ret < 0) {
2430 ERR("Spawning consumerd failed");
2431 pthread_mutex_unlock(&consumer_data->pid_mutex);
2432 goto error;
2433 }
2434
2435 /* Setting up the consumer_data pid */
2436 consumer_data->pid = ret;
2437 DBG2("Consumer pid %d", consumer_data->pid);
2438 pthread_mutex_unlock(&consumer_data->pid_mutex);
2439
2440 DBG2("Spawning consumer control thread");
2441 ret = spawn_consumer_thread(consumer_data);
2442 if (ret < 0) {
2443 ERR("Fatal error spawning consumer control thread");
2444 goto error;
2445 }
2446
2447 end:
2448 return 0;
2449
2450 error:
2451 /* Cleanup already created sockets on error. */
2452 if (consumer_data->err_sock >= 0) {
2453 int err;
2454
2455 err = close(consumer_data->err_sock);
2456 if (err < 0) {
2457 PERROR("close consumer data error socket");
2458 }
2459 }
2460 return ret;
2461 }
2462
2463 /*
2464 * Setup necessary data for kernel tracer action.
2465 */
2466 static int init_kernel_tracer(void)
2467 {
2468 int ret;
2469
2470 /* Modprobe lttng kernel modules */
2471 ret = modprobe_lttng_control();
2472 if (ret < 0) {
2473 goto error;
2474 }
2475
2476 /* Open debugfs lttng */
2477 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2478 if (kernel_tracer_fd < 0) {
2479 DBG("Failed to open %s", module_proc_lttng);
2480 goto error_open;
2481 }
2482
2483 /* Validate kernel version */
2484 ret = kernel_validate_version(kernel_tracer_fd, &kernel_tracer_version,
2485 &kernel_tracer_abi_version);
2486 if (ret < 0) {
2487 goto error_version;
2488 }
2489
2490 ret = modprobe_lttng_data();
2491 if (ret < 0) {
2492 goto error_modules;
2493 }
2494
2495 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2496 kernel_tracer_fd);
2497 if (ret < 0) {
2498 goto error_modules;
2499 }
2500
2501 if (ret < 1) {
2502 WARN("Kernel tracer does not support buffer monitoring. "
2503 "The monitoring timer of channels in the kernel domain "
2504 "will be set to 0 (disabled).");
2505 }
2506
2507 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2508 return 0;
2509
2510 error_version:
2511 modprobe_remove_lttng_control();
2512 ret = close(kernel_tracer_fd);
2513 if (ret) {
2514 PERROR("close");
2515 }
2516 kernel_tracer_fd = -1;
2517 return LTTNG_ERR_KERN_VERSION;
2518
2519 error_modules:
2520 ret = close(kernel_tracer_fd);
2521 if (ret) {
2522 PERROR("close");
2523 }
2524
2525 error_open:
2526 modprobe_remove_lttng_control();
2527
2528 error:
2529 WARN("No kernel tracer available");
2530 kernel_tracer_fd = -1;
2531 if (!is_root) {
2532 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2533 } else {
2534 return LTTNG_ERR_KERN_NA;
2535 }
2536 }
2537
2538
2539 /*
2540 * Copy consumer output from the tracing session to the domain session. The
2541 * function also applies the right modification on a per domain basis for the
2542 * trace files destination directory.
2543 *
2544 * Should *NOT* be called with RCU read-side lock held.
2545 */
2546 static int copy_session_consumer(int domain, struct ltt_session *session)
2547 {
2548 int ret;
2549 const char *dir_name;
2550 struct consumer_output *consumer;
2551
2552 assert(session);
2553 assert(session->consumer);
2554
2555 switch (domain) {
2556 case LTTNG_DOMAIN_KERNEL:
2557 DBG3("Copying tracing session consumer output in kernel session");
2558 /*
2559 * XXX: We should audit the session creation and what this function
2560 * does "extra" in order to avoid a destroy since this function is used
2561 * in the domain session creation (kernel and ust) only. Same for UST
2562 * domain.
2563 */
2564 if (session->kernel_session->consumer) {
2565 consumer_output_put(session->kernel_session->consumer);
2566 }
2567 session->kernel_session->consumer =
2568 consumer_copy_output(session->consumer);
2569 /* Ease our life a bit for the next part */
2570 consumer = session->kernel_session->consumer;
2571 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2572 break;
2573 case LTTNG_DOMAIN_JUL:
2574 case LTTNG_DOMAIN_LOG4J:
2575 case LTTNG_DOMAIN_PYTHON:
2576 case LTTNG_DOMAIN_UST:
2577 DBG3("Copying tracing session consumer output in UST session");
2578 if (session->ust_session->consumer) {
2579 consumer_output_put(session->ust_session->consumer);
2580 }
2581 session->ust_session->consumer =
2582 consumer_copy_output(session->consumer);
2583 /* Ease our life a bit for the next part */
2584 consumer = session->ust_session->consumer;
2585 dir_name = DEFAULT_UST_TRACE_DIR;
2586 break;
2587 default:
2588 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2589 goto error;
2590 }
2591
2592 /* Append correct directory to subdir */
2593 strncat(consumer->subdir, dir_name,
2594 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2595 DBG3("Copy session consumer subdir %s", consumer->subdir);
2596
2597 ret = LTTNG_OK;
2598
2599 error:
2600 return ret;
2601 }
2602
2603 /*
2604 * Create an UST session and add it to the session ust list.
2605 *
2606 * Should *NOT* be called with RCU read-side lock held.
2607 */
2608 static int create_ust_session(struct ltt_session *session,
2609 struct lttng_domain *domain)
2610 {
2611 int ret;
2612 struct ltt_ust_session *lus = NULL;
2613
2614 assert(session);
2615 assert(domain);
2616 assert(session->consumer);
2617
2618 switch (domain->type) {
2619 case LTTNG_DOMAIN_JUL:
2620 case LTTNG_DOMAIN_LOG4J:
2621 case LTTNG_DOMAIN_PYTHON:
2622 case LTTNG_DOMAIN_UST:
2623 break;
2624 default:
2625 ERR("Unknown UST domain on create session %d", domain->type);
2626 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2627 goto error;
2628 }
2629
2630 DBG("Creating UST session");
2631
2632 lus = trace_ust_create_session(session->id);
2633 if (lus == NULL) {
2634 ret = LTTNG_ERR_UST_SESS_FAIL;
2635 goto error;
2636 }
2637
2638 lus->uid = session->uid;
2639 lus->gid = session->gid;
2640 lus->output_traces = session->output_traces;
2641 lus->snapshot_mode = session->snapshot_mode;
2642 lus->live_timer_interval = session->live_timer;
2643 session->ust_session = lus;
2644 if (session->shm_path[0]) {
2645 strncpy(lus->root_shm_path, session->shm_path,
2646 sizeof(lus->root_shm_path));
2647 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2648 strncpy(lus->shm_path, session->shm_path,
2649 sizeof(lus->shm_path));
2650 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2651 strncat(lus->shm_path, "/ust",
2652 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2653 }
2654 /* Copy session output to the newly created UST session */
2655 ret = copy_session_consumer(domain->type, session);
2656 if (ret != LTTNG_OK) {
2657 goto error;
2658 }
2659
2660 return LTTNG_OK;
2661
2662 error:
2663 free(lus);
2664 session->ust_session = NULL;
2665 return ret;
2666 }
2667
2668 /*
2669 * Create a kernel tracer session then create the default channel.
2670 */
2671 static int create_kernel_session(struct ltt_session *session)
2672 {
2673 int ret;
2674
2675 DBG("Creating kernel session");
2676
2677 ret = kernel_create_session(session, kernel_tracer_fd);
2678 if (ret < 0) {
2679 ret = LTTNG_ERR_KERN_SESS_FAIL;
2680 goto error;
2681 }
2682
2683 /* Code flow safety */
2684 assert(session->kernel_session);
2685
2686 /* Copy session output to the newly created Kernel session */
2687 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2688 if (ret != LTTNG_OK) {
2689 goto error;
2690 }
2691
2692 session->kernel_session->uid = session->uid;
2693 session->kernel_session->gid = session->gid;
2694 session->kernel_session->output_traces = session->output_traces;
2695 session->kernel_session->snapshot_mode = session->snapshot_mode;
2696
2697 return LTTNG_OK;
2698
2699 error:
2700 trace_kernel_destroy_session(session->kernel_session);
2701 session->kernel_session = NULL;
2702 return ret;
2703 }
2704
2705 /*
2706 * Count number of session permitted by uid/gid.
2707 */
2708 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2709 {
2710 unsigned int i = 0;
2711 struct ltt_session *session;
2712 const struct ltt_session_list *session_list = session_get_list();
2713
2714 DBG("Counting number of available session for UID %d GID %d",
2715 uid, gid);
2716 cds_list_for_each_entry(session, &session_list->head, list) {
2717 if (!session_get(session)) {
2718 continue;
2719 }
2720 session_lock(session);
2721 /* Only count the sessions the user can control. */
2722 if (session_access_ok(session, uid, gid) &&
2723 !session->destroyed) {
2724 i++;
2725 }
2726 session_unlock(session);
2727 session_put(session);
2728 }
2729 return i;
2730 }
2731
2732 static int receive_userspace_probe(struct command_ctx *cmd_ctx, int sock,
2733 int *sock_error, struct lttng_event *event)
2734 {
2735 int fd, ret;
2736 struct lttng_userspace_probe_location *probe_location;
2737 const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
2738 struct lttng_dynamic_buffer probe_location_buffer;
2739 struct lttng_buffer_view buffer_view;
2740
2741 /*
2742 * Create a buffer to store the serialized version of the probe
2743 * location.
2744 */
2745 lttng_dynamic_buffer_init(&probe_location_buffer);
2746 ret = lttng_dynamic_buffer_set_size(&probe_location_buffer,
2747 cmd_ctx->lsm->u.enable.userspace_probe_location_len);
2748 if (ret) {
2749 ret = LTTNG_ERR_NOMEM;
2750 goto error;
2751 }
2752
2753 /*
2754 * Receive the probe location.
2755 */
2756 ret = lttcomm_recv_unix_sock(sock, probe_location_buffer.data,
2757 probe_location_buffer.size);
2758 if (ret <= 0) {
2759 DBG("Nothing recv() from client var len data... continuing");
2760 *sock_error = 1;
2761 lttng_dynamic_buffer_reset(&probe_location_buffer);
2762 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2763 goto error;
2764 }
2765
2766 buffer_view = lttng_buffer_view_from_dynamic_buffer(
2767 &probe_location_buffer, 0, probe_location_buffer.size);
2768
2769 /*
2770 * Extract the probe location from the serialized version.
2771 */
2772 ret = lttng_userspace_probe_location_create_from_buffer(
2773 &buffer_view, &probe_location);
2774 if (ret < 0) {
2775 WARN("Failed to create a userspace probe location from the received buffer");
2776 lttng_dynamic_buffer_reset( &probe_location_buffer);
2777 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2778 goto error;
2779 }
2780
2781 /*
2782 * Receive the file descriptor to the target binary from the client.
2783 */
2784 DBG("Receiving userspace probe target FD from client ...");
2785 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
2786 if (ret <= 0) {
2787 DBG("Nothing recv() from client userspace probe fd... continuing");
2788 *sock_error = 1;
2789 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2790 goto error;
2791 }
2792
2793 /*
2794 * Set the file descriptor received from the client through the unix
2795 * socket in the probe location.
2796 */
2797 lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
2798 if (!lookup) {
2799 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2800 goto error;
2801 }
2802
2803 /*
2804 * From the kernel tracer's perspective, all userspace probe event types
2805 * are all the same: a file and an offset.
2806 */
2807 switch (lttng_userspace_probe_location_lookup_method_get_type(lookup)) {
2808 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
2809 ret = lttng_userspace_probe_location_function_set_binary_fd(
2810 probe_location, fd);
2811 break;
2812 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
2813 ret = lttng_userspace_probe_location_tracepoint_set_binary_fd(
2814 probe_location, fd);
2815 break;
2816 default:
2817 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2818 goto error;
2819 }
2820
2821 if (ret) {
2822 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2823 goto error;
2824 }
2825
2826 /* Attach the probe location to the event. */
2827 ret = lttng_event_set_userspace_probe_location(event, probe_location);
2828 if (ret) {
2829 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
2830 goto error;
2831 }
2832
2833 lttng_dynamic_buffer_reset(&probe_location_buffer);
2834 error:
2835 return ret;
2836 }
2837
2838 /*
2839 * Check if the current kernel tracer supports the session rotation feature.
2840 * Return 1 if it does, 0 otherwise.
2841 */
2842 static int check_rotate_compatible(void)
2843 {
2844 int ret = 1;
2845
2846 if (kernel_tracer_version.major != 2 || kernel_tracer_version.minor < 11) {
2847 DBG("Kernel tracer version is not compatible with the rotation feature");
2848 ret = 0;
2849 }
2850
2851 return ret;
2852 }
2853
2854 /*
2855 * Process the command requested by the lttng client within the command
2856 * context structure. This function make sure that the return structure (llm)
2857 * is set and ready for transmission before returning.
2858 *
2859 * Return any error encountered or 0 for success.
2860 *
2861 * "sock" is only used for special-case var. len data.
2862 *
2863 * Should *NOT* be called with RCU read-side lock held.
2864 */
2865 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2866 int *sock_error)
2867 {
2868 int ret = LTTNG_OK;
2869 int need_tracing_session = 1;
2870 int need_domain;
2871
2872 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2873
2874 assert(!rcu_read_ongoing());
2875
2876 *sock_error = 0;
2877
2878 switch (cmd_ctx->lsm->cmd_type) {
2879 case LTTNG_CREATE_SESSION:
2880 case LTTNG_CREATE_SESSION_SNAPSHOT:
2881 case LTTNG_CREATE_SESSION_LIVE:
2882 case LTTNG_DESTROY_SESSION:
2883 case LTTNG_LIST_SESSIONS:
2884 case LTTNG_LIST_DOMAINS:
2885 case LTTNG_START_TRACE:
2886 case LTTNG_STOP_TRACE:
2887 case LTTNG_DATA_PENDING:
2888 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2889 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2890 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2891 case LTTNG_SNAPSHOT_RECORD:
2892 case LTTNG_SAVE_SESSION:
2893 case LTTNG_SET_SESSION_SHM_PATH:
2894 case LTTNG_REGENERATE_METADATA:
2895 case LTTNG_REGENERATE_STATEDUMP:
2896 case LTTNG_REGISTER_TRIGGER:
2897 case LTTNG_UNREGISTER_TRIGGER:
2898 case LTTNG_ROTATE_SESSION:
2899 case LTTNG_ROTATION_GET_INFO:
2900 case LTTNG_ROTATION_SET_SCHEDULE:
2901 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
2902 need_domain = 0;
2903 break;
2904 default:
2905 need_domain = 1;
2906 }
2907
2908 if (config.no_kernel && need_domain
2909 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2910 if (!is_root) {
2911 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2912 } else {
2913 ret = LTTNG_ERR_KERN_NA;
2914 }
2915 goto error;
2916 }
2917
2918 /* Deny register consumer if we already have a spawned consumer. */
2919 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2920 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2921 if (kconsumer_data.pid > 0) {
2922 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2923 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2924 goto error;
2925 }
2926 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2927 }
2928
2929 /*
2930 * Check for command that don't needs to allocate a returned payload. We do
2931 * this here so we don't have to make the call for no payload at each
2932 * command.
2933 */
2934 switch(cmd_ctx->lsm->cmd_type) {
2935 case LTTNG_LIST_SESSIONS:
2936 case LTTNG_LIST_TRACEPOINTS:
2937 case LTTNG_LIST_TRACEPOINT_FIELDS:
2938 case LTTNG_LIST_DOMAINS:
2939 case LTTNG_LIST_CHANNELS:
2940 case LTTNG_LIST_EVENTS:
2941 case LTTNG_LIST_SYSCALLS:
2942 case LTTNG_LIST_TRACKER_PIDS:
2943 case LTTNG_DATA_PENDING:
2944 case LTTNG_ROTATE_SESSION:
2945 case LTTNG_ROTATION_GET_INFO:
2946 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
2947 break;
2948 default:
2949 /* Setup lttng message with no payload */
2950 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
2951 if (ret < 0) {
2952 /* This label does not try to unlock the session */
2953 goto init_setup_error;
2954 }
2955 }
2956
2957 /* Commands that DO NOT need a session. */
2958 switch (cmd_ctx->lsm->cmd_type) {
2959 case LTTNG_CREATE_SESSION:
2960 case LTTNG_CREATE_SESSION_SNAPSHOT:
2961 case LTTNG_CREATE_SESSION_LIVE:
2962 case LTTNG_LIST_SESSIONS:
2963 case LTTNG_LIST_TRACEPOINTS:
2964 case LTTNG_LIST_SYSCALLS:
2965 case LTTNG_LIST_TRACEPOINT_FIELDS:
2966 case LTTNG_SAVE_SESSION:
2967 case LTTNG_REGISTER_TRIGGER:
2968 case LTTNG_UNREGISTER_TRIGGER:
2969 need_tracing_session = 0;
2970 break;
2971 default:
2972 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2973 /*
2974 * We keep the session list lock across _all_ commands
2975 * for now, because the per-session lock does not
2976 * handle teardown properly.
2977 */
2978 session_lock_list();
2979 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2980 if (cmd_ctx->session == NULL) {
2981 ret = LTTNG_ERR_SESS_NOT_FOUND;
2982 goto error;
2983 } else {
2984 /* Acquire lock for the session */
2985 session_lock(cmd_ctx->session);
2986 }
2987 break;
2988 }
2989
2990 /*
2991 * Commands that need a valid session but should NOT create one if none
2992 * exists. Instead of creating one and destroying it when the command is
2993 * handled, process that right before so we save some round trip in useless
2994 * code path.
2995 */
2996 switch (cmd_ctx->lsm->cmd_type) {
2997 case LTTNG_DISABLE_CHANNEL:
2998 case LTTNG_DISABLE_EVENT:
2999 switch (cmd_ctx->lsm->domain.type) {
3000 case LTTNG_DOMAIN_KERNEL:
3001 if (!cmd_ctx->session->kernel_session) {
3002 ret = LTTNG_ERR_NO_CHANNEL;
3003 goto error;
3004 }
3005 break;
3006 case LTTNG_DOMAIN_JUL:
3007 case LTTNG_DOMAIN_LOG4J:
3008 case LTTNG_DOMAIN_PYTHON:
3009 case LTTNG_DOMAIN_UST:
3010 if (!cmd_ctx->session->ust_session) {
3011 ret = LTTNG_ERR_NO_CHANNEL;
3012 goto error;
3013 }
3014 break;
3015 default:
3016 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3017 goto error;
3018 }
3019 default:
3020 break;
3021 }
3022
3023 if (!need_domain) {
3024 goto skip_domain;
3025 }
3026
3027 /*
3028 * Check domain type for specific "pre-action".
3029 */
3030 switch (cmd_ctx->lsm->domain.type) {
3031 case LTTNG_DOMAIN_KERNEL:
3032 if (!is_root) {
3033 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3034 goto error;
3035 }
3036
3037 /* Kernel tracer check */
3038 if (kernel_tracer_fd == -1) {
3039 /* Basically, load kernel tracer modules */
3040 ret = init_kernel_tracer();
3041 if (ret != 0) {
3042 goto error;
3043 }
3044 }
3045
3046 /* Consumer is in an ERROR state. Report back to client */
3047 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3048 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3049 goto error;
3050 }
3051
3052 /* Need a session for kernel command */
3053 if (need_tracing_session) {
3054 if (cmd_ctx->session->kernel_session == NULL) {
3055 ret = create_kernel_session(cmd_ctx->session);
3056 if (ret < 0) {
3057 ret = LTTNG_ERR_KERN_SESS_FAIL;
3058 goto error;
3059 }
3060 }
3061
3062 /* Start the kernel consumer daemon */
3063 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3064 if (kconsumer_data.pid == 0 &&
3065 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3066 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3067 ret = start_consumerd(&kconsumer_data);
3068 if (ret < 0) {
3069 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3070 goto error;
3071 }
3072 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3073 } else {
3074 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3075 }
3076
3077 /*
3078 * The consumer was just spawned so we need to add the socket to
3079 * the consumer output of the session if exist.
3080 */
3081 ret = consumer_create_socket(&kconsumer_data,
3082 cmd_ctx->session->kernel_session->consumer);
3083 if (ret < 0) {
3084 goto error;
3085 }
3086 }
3087
3088 break;
3089 case LTTNG_DOMAIN_JUL:
3090 case LTTNG_DOMAIN_LOG4J:
3091 case LTTNG_DOMAIN_PYTHON:
3092 case LTTNG_DOMAIN_UST:
3093 {
3094 if (!ust_app_supported()) {
3095 ret = LTTNG_ERR_NO_UST;
3096 goto error;
3097 }
3098 /* Consumer is in an ERROR state. Report back to client */
3099 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3100 ret = LTTNG_ERR_NO_USTCONSUMERD;
3101 goto error;
3102 }
3103
3104 if (need_tracing_session) {
3105 /* Create UST session if none exist. */
3106 if (cmd_ctx->session->ust_session == NULL) {
3107 ret = create_ust_session(cmd_ctx->session,
3108 &cmd_ctx->lsm->domain);
3109 if (ret != LTTNG_OK) {
3110 goto error;
3111 }
3112 }
3113
3114 /* Start the UST consumer daemons */
3115 /* 64-bit */
3116 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3117 if (config.consumerd64_bin_path.value &&
3118 ustconsumer64_data.pid == 0 &&
3119 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3120 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3121 ret = start_consumerd(&ustconsumer64_data);
3122 if (ret < 0) {
3123 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3124 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3125 goto error;
3126 }
3127
3128 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3129 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3130 } else {
3131 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3132 }
3133
3134 /*
3135 * Setup socket for consumer 64 bit. No need for atomic access
3136 * since it was set above and can ONLY be set in this thread.
3137 */
3138 ret = consumer_create_socket(&ustconsumer64_data,
3139 cmd_ctx->session->ust_session->consumer);
3140 if (ret < 0) {
3141 goto error;
3142 }
3143
3144 /* 32-bit */
3145 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3146 if (config.consumerd32_bin_path.value &&
3147 ustconsumer32_data.pid == 0 &&
3148 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3149 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3150 ret = start_consumerd(&ustconsumer32_data);
3151 if (ret < 0) {
3152 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3153 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3154 goto error;
3155 }
3156
3157 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3158 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3159 } else {
3160 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3161 }
3162
3163 /*
3164 * Setup socket for consumer 32 bit. No need for atomic access
3165 * since it was set above and can ONLY be set in this thread.
3166 */
3167 ret = consumer_create_socket(&ustconsumer32_data,
3168 cmd_ctx->session->ust_session->consumer);
3169 if (ret < 0) {
3170 goto error;
3171 }
3172 }
3173 break;
3174 }
3175 default:
3176 break;
3177 }
3178 skip_domain:
3179
3180 /* Validate consumer daemon state when start/stop trace command */
3181 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3182 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3183 switch (cmd_ctx->lsm->domain.type) {
3184 case LTTNG_DOMAIN_NONE:
3185 break;
3186 case LTTNG_DOMAIN_JUL:
3187 case LTTNG_DOMAIN_LOG4J:
3188 case LTTNG_DOMAIN_PYTHON:
3189 case LTTNG_DOMAIN_UST:
3190 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3191 ret = LTTNG_ERR_NO_USTCONSUMERD;
3192 goto error;
3193 }
3194 break;
3195 case LTTNG_DOMAIN_KERNEL:
3196 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
3197 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3198 goto error;
3199 }
3200 break;
3201 default:
3202 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3203 goto error;
3204 }
3205 }
3206
3207 /*
3208 * Check that the UID or GID match that of the tracing session.
3209 * The root user can interact with all sessions.
3210 */
3211 if (need_tracing_session) {
3212 if (!session_access_ok(cmd_ctx->session,
3213 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3214 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds)) ||
3215 cmd_ctx->session->destroyed) {
3216 ret = LTTNG_ERR_EPERM;
3217 goto error;
3218 }
3219 }
3220
3221 /*
3222 * Send relayd information to consumer as soon as we have a domain and a
3223 * session defined.
3224 */
3225 if (cmd_ctx->session && need_domain) {
3226 /*
3227 * Setup relayd if not done yet. If the relayd information was already
3228 * sent to the consumer, this call will gracefully return.
3229 */
3230 ret = cmd_setup_relayd(cmd_ctx->session);
3231 if (ret != LTTNG_OK) {
3232 goto error;
3233 }
3234 }
3235
3236 /* Process by command type */
3237 switch (cmd_ctx->lsm->cmd_type) {
3238 case LTTNG_ADD_CONTEXT:
3239 {
3240 /*
3241 * An LTTNG_ADD_CONTEXT command might have a supplementary
3242 * payload if the context being added is an application context.
3243 */
3244 if (cmd_ctx->lsm->u.context.ctx.ctx ==
3245 LTTNG_EVENT_CONTEXT_APP_CONTEXT) {
3246 char *provider_name = NULL, *context_name = NULL;
3247 size_t provider_name_len =
3248 cmd_ctx->lsm->u.context.provider_name_len;
3249 size_t context_name_len =
3250 cmd_ctx->lsm->u.context.context_name_len;
3251
3252 if (provider_name_len == 0 || context_name_len == 0) {
3253 /*
3254 * Application provider and context names MUST
3255 * be provided.
3256 */
3257 ret = -LTTNG_ERR_INVALID;
3258 goto error;
3259 }
3260
3261 provider_name = zmalloc(provider_name_len + 1);
3262 if (!provider_name) {
3263 ret = -LTTNG_ERR_NOMEM;
3264 goto error;
3265 }
3266 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name =
3267 provider_name;
3268
3269 context_name = zmalloc(context_name_len + 1);
3270 if (!context_name) {
3271 ret = -LTTNG_ERR_NOMEM;
3272 goto error_add_context;
3273 }
3274 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name =
3275 context_name;
3276
3277 ret = lttcomm_recv_unix_sock(sock, provider_name,
3278 provider_name_len);
3279 if (ret < 0) {
3280 goto error_add_context;
3281 }
3282
3283 ret = lttcomm_recv_unix_sock(sock, context_name,
3284 context_name_len);
3285 if (ret < 0) {
3286 goto error_add_context;
3287 }
3288 }
3289
3290 /*
3291 * cmd_add_context assumes ownership of the provider and context
3292 * names.
3293 */
3294 ret = cmd_add_context(cmd_ctx->session,
3295 cmd_ctx->lsm->domain.type,
3296 cmd_ctx->lsm->u.context.channel_name,
3297 &cmd_ctx->lsm->u.context.ctx,
3298 kernel_poll_pipe[1]);
3299
3300 cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name = NULL;
3301 cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name = NULL;
3302 error_add_context:
3303 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.provider_name);
3304 free(cmd_ctx->lsm->u.context.ctx.u.app_ctx.ctx_name);
3305 if (ret < 0) {
3306 goto error;
3307 }
3308 break;
3309 }
3310 case LTTNG_DISABLE_CHANNEL:
3311 {
3312 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3313 cmd_ctx->lsm->u.disable.channel_name);
3314 break;
3315 }
3316 case LTTNG_DISABLE_EVENT:
3317 {
3318
3319 /*
3320 * FIXME: handle filter; for now we just receive the filter's
3321 * bytecode along with the filter expression which are sent by
3322 * liblttng-ctl and discard them.
3323 *
3324 * This fixes an issue where the client may block while sending
3325 * the filter payload and encounter an error because the session
3326 * daemon closes the socket without ever handling this data.
3327 */
3328 size_t count = cmd_ctx->lsm->u.disable.expression_len +
3329 cmd_ctx->lsm->u.disable.bytecode_len;
3330
3331 if (count) {
3332 char data[LTTNG_FILTER_MAX_LEN];
3333
3334 DBG("Discarding disable event command payload of size %zu", count);
3335 while (count) {
3336 ret = lttcomm_recv_unix_sock(sock, data,
3337 count > sizeof(data) ? sizeof(data) : count);
3338 if (ret < 0) {
3339 goto error;
3340 }
3341
3342 count -= (size_t) ret;
3343 }
3344 }
3345 /* FIXME: passing packed structure to non-packed pointer */
3346 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3347 cmd_ctx->lsm->u.disable.channel_name,
3348 &cmd_ctx->lsm->u.disable.event);
3349 break;
3350 }
3351 case LTTNG_ENABLE_CHANNEL:
3352 {
3353 cmd_ctx->lsm->u.channel.chan.attr.extended.ptr =
3354 (struct lttng_channel_extended *) &cmd_ctx->lsm->u.channel.extended;
3355 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
3356 &cmd_ctx->lsm->u.channel.chan,
3357 kernel_poll_pipe[1]);
3358 break;
3359 }
3360 case LTTNG_TRACK_PID:
3361 {
3362 ret = cmd_track_pid(cmd_ctx->session,
3363 cmd_ctx->lsm->domain.type,
3364 cmd_ctx->lsm->u.pid_tracker.pid);
3365 break;
3366 }
3367 case LTTNG_UNTRACK_PID:
3368 {
3369 ret = cmd_untrack_pid(cmd_ctx->session,
3370 cmd_ctx->lsm->domain.type,
3371 cmd_ctx->lsm->u.pid_tracker.pid);
3372 break;
3373 }
3374 case LTTNG_ENABLE_EVENT:
3375 {
3376 struct lttng_event *ev = NULL;
3377 struct lttng_event_exclusion *exclusion = NULL;
3378 struct lttng_filter_bytecode *bytecode = NULL;
3379 char *filter_expression = NULL;
3380
3381 /* Handle exclusion events and receive it from the client. */
3382 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3383 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3384
3385 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3386 (count * LTTNG_SYMBOL_NAME_LEN));
3387 if (!exclusion) {
3388 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3389 goto error;
3390 }
3391
3392 DBG("Receiving var len exclusion event list from client ...");
3393 exclusion->count = count;
3394 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3395 count * LTTNG_SYMBOL_NAME_LEN);
3396 if (ret <= 0) {
3397 DBG("Nothing recv() from client var len data... continuing");
3398 *sock_error = 1;
3399 free(exclusion);
3400 ret = LTTNG_ERR_EXCLUSION_INVAL;
3401 goto error;
3402 }
3403 }
3404
3405 /* Get filter expression from client. */
3406 if (cmd_ctx->lsm->u.enable.expression_len > 0) {
3407 size_t expression_len =
3408 cmd_ctx->lsm->u.enable.expression_len;
3409
3410 if (expression_len > LTTNG_FILTER_MAX_LEN) {
3411 ret = LTTNG_ERR_FILTER_INVAL;
3412 free(exclusion);
3413 goto error;
3414 }
3415
3416 filter_expression = zmalloc(expression_len);
3417 if (!filter_expression) {
3418 free(exclusion);
3419 ret = LTTNG_ERR_FILTER_NOMEM;
3420 goto error;
3421 }
3422
3423 /* Receive var. len. data */
3424 DBG("Receiving var len filter's expression from client ...");
3425 ret = lttcomm_recv_unix_sock(sock, filter_expression,
3426 expression_len);
3427 if (ret <= 0) {
3428 DBG("Nothing recv() from client var len data... continuing");
3429 *sock_error = 1;
3430 free(filter_expression);
3431 free(exclusion);
3432 ret = LTTNG_ERR_FILTER_INVAL;
3433 goto error;
3434 }
3435 }
3436
3437 /* Handle filter and get bytecode from client. */
3438 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3439 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3440
3441 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3442 ret = LTTNG_ERR_FILTER_INVAL;
3443 free(filter_expression);
3444 free(exclusion);
3445 goto error;
3446 }
3447
3448 bytecode = zmalloc(bytecode_len);
3449 if (!bytecode) {
3450 free(filter_expression);
3451 free(exclusion);
3452 ret = LTTNG_ERR_FILTER_NOMEM;
3453 goto error;
3454 }
3455
3456 /* Receive var. len. data */
3457 DBG("Receiving var len filter's bytecode from client ...");
3458 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3459 if (ret <= 0) {
3460 DBG("Nothing recv() from client var len data... continuing");
3461 *sock_error = 1;
3462 free(filter_expression);
3463 free(bytecode);
3464 free(exclusion);
3465 ret = LTTNG_ERR_FILTER_INVAL;
3466 goto error;
3467 }
3468
3469 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3470 free(filter_expression);
3471 free(bytecode);
3472 free(exclusion);
3473 ret = LTTNG_ERR_FILTER_INVAL;
3474 goto error;
3475 }
3476 }
3477
3478 ev = lttng_event_copy(&cmd_ctx->lsm->u.enable.event);
3479 if (!ev) {
3480 DBG("Failed to copy event: %s",
3481 cmd_ctx->lsm->u.enable.event.name);
3482 free(filter_expression);
3483 free(bytecode);
3484 free(exclusion);
3485 ret = LTTNG_ERR_NOMEM;
3486 goto error;
3487 }
3488
3489
3490 if (cmd_ctx->lsm->u.enable.userspace_probe_location_len > 0) {
3491 /* Expect a userspace probe description. */
3492 ret = receive_userspace_probe(cmd_ctx, sock, sock_error, ev);
3493 if (ret) {
3494 free(filter_expression);
3495 free(bytecode);
3496 free(exclusion);
3497 lttng_event_destroy(ev);
3498 goto error;
3499 }
3500 }
3501
3502 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3503 cmd_ctx->lsm->u.enable.channel_name,
3504 ev,
3505 filter_expression, bytecode, exclusion,
3506 kernel_poll_pipe[1]);
3507 lttng_event_destroy(ev);
3508 break;
3509 }
3510 case LTTNG_LIST_TRACEPOINTS:
3511 {
3512 struct lttng_event *events;
3513 ssize_t nb_events;
3514
3515 session_lock_list();
3516 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3517 session_unlock_list();
3518 if (nb_events < 0) {
3519 /* Return value is a negative lttng_error_code. */
3520 ret = -nb_events;
3521 goto error;
3522 }
3523
3524 /*
3525 * Setup lttng message with payload size set to the event list size in
3526 * bytes and then copy list into the llm payload.
3527 */
3528 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3529 sizeof(struct lttng_event) * nb_events);
3530 free(events);
3531
3532 if (ret < 0) {
3533 goto setup_error;
3534 }
3535
3536 ret = LTTNG_OK;
3537 break;
3538 }
3539 case LTTNG_LIST_TRACEPOINT_FIELDS:
3540 {
3541 struct lttng_event_field *fields;
3542 ssize_t nb_fields;
3543
3544 session_lock_list();
3545 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3546 &fields);
3547 session_unlock_list();
3548 if (nb_fields < 0) {
3549 /* Return value is a negative lttng_error_code. */
3550 ret = -nb_fields;
3551 goto error;
3552 }
3553
3554 /*
3555 * Setup lttng message with payload size set to the event list size in
3556 * bytes and then copy list into the llm payload.
3557 */
3558 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, fields,
3559 sizeof(struct lttng_event_field) * nb_fields);
3560 free(fields);
3561
3562 if (ret < 0) {
3563 goto setup_error;
3564 }
3565
3566 ret = LTTNG_OK;
3567 break;
3568 }
3569 case LTTNG_LIST_SYSCALLS:
3570 {
3571 struct lttng_event *events;
3572 ssize_t nb_events;
3573
3574 nb_events = cmd_list_syscalls(&events);
3575 if (nb_events < 0) {
3576 /* Return value is a negative lttng_error_code. */
3577 ret = -nb_events;
3578 goto error;
3579 }
3580
3581 /*
3582 * Setup lttng message with payload size set to the event list size in
3583 * bytes and then copy list into the llm payload.
3584 */
3585 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, events,
3586 sizeof(struct lttng_event) * nb_events);
3587 free(events);
3588
3589 if (ret < 0) {
3590 goto setup_error;
3591 }
3592
3593 ret = LTTNG_OK;
3594 break;
3595 }
3596 case LTTNG_LIST_TRACKER_PIDS:
3597 {
3598 int32_t *pids = NULL;
3599 ssize_t nr_pids;
3600
3601 nr_pids = cmd_list_tracker_pids(cmd_ctx->session,
3602 cmd_ctx->lsm->domain.type, &pids);
3603 if (nr_pids < 0) {
3604 /* Return value is a negative lttng_error_code. */
3605 ret = -nr_pids;
3606 goto error;
3607 }
3608
3609 /*
3610 * Setup lttng message with payload size set to the event list size in
3611 * bytes and then copy list into the llm payload.
3612 */
3613 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pids,
3614 sizeof(int32_t) * nr_pids);
3615 free(pids);
3616
3617 if (ret < 0) {
3618 goto setup_error;
3619 }
3620
3621 ret = LTTNG_OK;
3622 break;
3623 }
3624 case LTTNG_SET_CONSUMER_URI:
3625 {
3626 size_t nb_uri, len;
3627 struct lttng_uri *uris;
3628
3629 nb_uri = cmd_ctx->lsm->u.uri.size;
3630 len = nb_uri * sizeof(struct lttng_uri);
3631
3632 if (nb_uri == 0) {
3633 ret = LTTNG_ERR_INVALID;
3634 goto error;
3635 }
3636
3637 uris = zmalloc(len);
3638 if (uris == NULL) {
3639 ret = LTTNG_ERR_FATAL;
3640 goto error;
3641 }
3642
3643 /* Receive variable len data */
3644 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3645 ret = lttcomm_recv_unix_sock(sock, uris, len);
3646 if (ret <= 0) {
3647 DBG("No URIs received from client... continuing");
3648 *sock_error = 1;
3649 ret = LTTNG_ERR_SESSION_FAIL;
3650 free(uris);
3651 goto error;
3652 }
3653
3654 ret = cmd_set_consumer_uri(cmd_ctx->session, nb_uri, uris);
3655 free(uris);
3656 if (ret != LTTNG_OK) {
3657 goto error;
3658 }
3659
3660
3661 break;
3662 }
3663 case LTTNG_START_TRACE:
3664 {
3665 /*
3666 * On the first start, if we have a kernel session and we have
3667 * enabled time or size-based rotations, we have to make sure
3668 * the kernel tracer supports it.
3669 */
3670 if (!cmd_ctx->session->has_been_started && \
3671 cmd_ctx->session->kernel_session && \
3672 (cmd_ctx->session->rotate_timer_period || \
3673 cmd_ctx->session->rotate_size) && \
3674 !check_rotate_compatible()) {
3675 DBG("Kernel tracer version is not compatible with the rotation feature");
3676 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
3677 goto error;
3678 }
3679 ret = cmd_start_trace(cmd_ctx->session);
3680 break;
3681 }
3682 case LTTNG_STOP_TRACE:
3683 {
3684 ret = cmd_stop_trace(cmd_ctx->session);
3685 break;
3686 }
3687 case LTTNG_CREATE_SESSION:
3688 {
3689 size_t nb_uri, len;
3690 struct lttng_uri *uris = NULL;
3691
3692 nb_uri = cmd_ctx->lsm->u.uri.size;
3693 len = nb_uri * sizeof(struct lttng_uri);
3694
3695 if (nb_uri > 0) {
3696 uris = zmalloc(len);
3697 if (uris == NULL) {
3698 ret = LTTNG_ERR_FATAL;
3699 goto error;
3700 }
3701
3702 /* Receive variable len data */
3703 DBG("Waiting for %zu URIs from client ...", nb_uri);
3704 ret = lttcomm_recv_unix_sock(sock, uris, len);
3705 if (ret <= 0) {
3706 DBG("No URIs received from client... continuing");
3707 *sock_error = 1;
3708 ret = LTTNG_ERR_SESSION_FAIL;
3709 free(uris);
3710 goto error;
3711 }
3712
3713 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3714 DBG("Creating session with ONE network URI is a bad call");
3715 ret = LTTNG_ERR_SESSION_FAIL;
3716 free(uris);
3717 goto error;
3718 }
3719 }
3720
3721 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3722 &cmd_ctx->creds, 0);
3723
3724 free(uris);
3725
3726 break;
3727 }
3728 case LTTNG_DESTROY_SESSION:
3729 {
3730 ret = cmd_destroy_session(cmd_ctx->session,
3731 notification_thread_handle);
3732 break;
3733 }
3734 case LTTNG_LIST_DOMAINS:
3735 {
3736 ssize_t nb_dom;
3737 struct lttng_domain *domains = NULL;
3738
3739 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3740 if (nb_dom < 0) {
3741 /* Return value is a negative lttng_error_code. */
3742 ret = -nb_dom;
3743 goto error;
3744 }
3745
3746 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, domains,
3747 nb_dom * sizeof(struct lttng_domain));
3748 free(domains);
3749
3750 if (ret < 0) {
3751 goto setup_error;
3752 }
3753
3754 ret = LTTNG_OK;
3755 break;
3756 }
3757 case LTTNG_LIST_CHANNELS:
3758 {
3759 ssize_t payload_size;
3760 struct lttng_channel *channels = NULL;
3761
3762 payload_size = cmd_list_channels(cmd_ctx->lsm->domain.type,
3763 cmd_ctx->session, &channels);
3764 if (payload_size < 0) {
3765 /* Return value is a negative lttng_error_code. */
3766 ret = -payload_size;
3767 goto error;
3768 }
3769
3770 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, channels,
3771 payload_size);
3772 free(channels);
3773
3774 if (ret < 0) {
3775 goto setup_error;
3776 }
3777
3778 ret = LTTNG_OK;
3779 break;
3780 }
3781 case LTTNG_LIST_EVENTS:
3782 {
3783 ssize_t nb_event;
3784 struct lttng_event *events = NULL;
3785 struct lttcomm_event_command_header cmd_header;
3786 size_t total_size;
3787
3788 memset(&cmd_header, 0, sizeof(cmd_header));
3789 /* Extended infos are included at the end of events */
3790 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type,
3791 cmd_ctx->session, cmd_ctx->lsm->u.list.channel_name,
3792 &events, &total_size);
3793
3794 if (nb_event < 0) {
3795 /* Return value is a negative lttng_error_code. */
3796 ret = -nb_event;
3797 goto error;
3798 }
3799
3800 cmd_header.nb_events = nb_event;
3801 ret = setup_lttng_msg(cmd_ctx, events, total_size,
3802 &cmd_header, sizeof(cmd_header));
3803 free(events);
3804
3805 if (ret < 0) {
3806 goto setup_error;
3807 }
3808
3809 ret = LTTNG_OK;
3810 break;
3811 }
3812 case LTTNG_LIST_SESSIONS:
3813 {
3814 unsigned int nr_sessions;
3815 void *sessions_payload;
3816 size_t payload_len;
3817
3818 session_lock_list();
3819 nr_sessions = lttng_sessions_count(
3820 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3821 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3822 payload_len = sizeof(struct lttng_session) * nr_sessions;
3823 sessions_payload = zmalloc(payload_len);
3824
3825 if (!sessions_payload) {
3826 session_unlock_list();
3827 ret = -ENOMEM;
3828 goto setup_error;
3829 }
3830
3831 cmd_list_lttng_sessions(sessions_payload,
3832 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3833 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3834 session_unlock_list();
3835
3836 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, sessions_payload,
3837 payload_len);
3838 free(sessions_payload);
3839
3840 if (ret < 0) {
3841 goto setup_error;
3842 }
3843
3844 ret = LTTNG_OK;
3845 break;
3846 }
3847 case LTTNG_REGISTER_CONSUMER:
3848 {
3849 struct consumer_data *cdata;
3850
3851 switch (cmd_ctx->lsm->domain.type) {
3852 case LTTNG_DOMAIN_KERNEL:
3853 cdata = &kconsumer_data;
3854 break;
3855 default:
3856 ret = LTTNG_ERR_UND;
3857 goto error;
3858 }
3859
3860 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3861 cmd_ctx->lsm->u.reg.path, cdata);
3862 break;
3863 }
3864 case LTTNG_DATA_PENDING:
3865 {
3866 int pending_ret;
3867 uint8_t pending_ret_byte;
3868
3869 pending_ret = cmd_data_pending(cmd_ctx->session);
3870
3871 /*
3872 * FIXME
3873 *
3874 * This function may returns 0 or 1 to indicate whether or not
3875 * there is data pending. In case of error, it should return an
3876 * LTTNG_ERR code. However, some code paths may still return
3877 * a nondescript error code, which we handle by returning an
3878 * "unknown" error.
3879 */
3880 if (pending_ret == 0 || pending_ret == 1) {
3881 /*
3882 * ret will be set to LTTNG_OK at the end of
3883 * this function.
3884 */
3885 } else if (pending_ret < 0) {
3886 ret = LTTNG_ERR_UNK;
3887 goto setup_error;
3888 } else {
3889 ret = pending_ret;
3890 goto setup_error;
3891 }
3892
3893 pending_ret_byte = (uint8_t) pending_ret;
3894
3895 /* 1 byte to return whether or not data is pending */
3896 ret = setup_lttng_msg_no_cmd_header(cmd_ctx,
3897 &pending_ret_byte, 1);
3898
3899 if (ret < 0) {
3900 goto setup_error;
3901 }
3902
3903 ret = LTTNG_OK;
3904 break;
3905 }
3906 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3907 {
3908 struct lttcomm_lttng_output_id reply;
3909
3910 ret = cmd_snapshot_add_output(cmd_ctx->session,
3911 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3912 if (ret != LTTNG_OK) {
3913 goto error;
3914 }
3915
3916 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &reply,
3917 sizeof(reply));
3918 if (ret < 0) {
3919 goto setup_error;
3920 }
3921
3922 /* Copy output list into message payload */
3923 ret = LTTNG_OK;
3924 break;
3925 }
3926 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3927 {
3928 ret = cmd_snapshot_del_output(cmd_ctx->session,
3929 &cmd_ctx->lsm->u.snapshot_output.output);
3930 break;
3931 }
3932 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3933 {
3934 ssize_t nb_output;
3935 struct lttng_snapshot_output *outputs = NULL;
3936
3937 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3938 if (nb_output < 0) {
3939 ret = -nb_output;
3940 goto error;
3941 }
3942
3943 assert((nb_output > 0 && outputs) || nb_output == 0);
3944 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, outputs,
3945 nb_output * sizeof(struct lttng_snapshot_output));
3946 free(outputs);
3947
3948 if (ret < 0) {
3949 goto setup_error;
3950 }
3951
3952 ret = LTTNG_OK;
3953 break;
3954 }
3955 case LTTNG_SNAPSHOT_RECORD:
3956 {
3957 ret = cmd_snapshot_record(cmd_ctx->session,
3958 &cmd_ctx->lsm->u.snapshot_record.output,
3959 cmd_ctx->lsm->u.snapshot_record.wait);
3960 break;
3961 }
3962 case LTTNG_CREATE_SESSION_SNAPSHOT:
3963 {
3964 size_t nb_uri, len;
3965 struct lttng_uri *uris = NULL;
3966
3967 nb_uri = cmd_ctx->lsm->u.uri.size;
3968 len = nb_uri * sizeof(struct lttng_uri);
3969
3970 if (nb_uri > 0) {
3971 uris = zmalloc(len);
3972 if (uris == NULL) {
3973 ret = LTTNG_ERR_FATAL;
3974 goto error;
3975 }
3976
3977 /* Receive variable len data */
3978 DBG("Waiting for %zu URIs from client ...", nb_uri);
3979 ret = lttcomm_recv_unix_sock(sock, uris, len);
3980 if (ret <= 0) {
3981 DBG("No URIs received from client... continuing");
3982 *sock_error = 1;
3983 ret = LTTNG_ERR_SESSION_FAIL;
3984 free(uris);
3985 goto error;
3986 }
3987
3988 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3989 DBG("Creating session with ONE network URI is a bad call");
3990 ret = LTTNG_ERR_SESSION_FAIL;
3991 free(uris);
3992 goto error;
3993 }
3994 }
3995
3996 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3997 nb_uri, &cmd_ctx->creds);
3998 free(uris);
3999 break;
4000 }
4001 case LTTNG_CREATE_SESSION_LIVE:
4002 {
4003 size_t nb_uri, len;
4004 struct lttng_uri *uris = NULL;
4005
4006 nb_uri = cmd_ctx->lsm->u.uri.size;
4007 len = nb_uri * sizeof(struct lttng_uri);
4008
4009 if (nb_uri > 0) {
4010 uris = zmalloc(len);
4011 if (uris == NULL) {
4012 ret = LTTNG_ERR_FATAL;
4013 goto error;
4014 }
4015
4016 /* Receive variable len data */
4017 DBG("Waiting for %zu URIs from client ...", nb_uri);
4018 ret = lttcomm_recv_unix_sock(sock, uris, len);
4019 if (ret <= 0) {
4020 DBG("No URIs received from client... continuing");
4021 *sock_error = 1;
4022 ret = LTTNG_ERR_SESSION_FAIL;
4023 free(uris);
4024 goto error;
4025 }
4026
4027 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
4028 DBG("Creating session with ONE network URI is a bad call");
4029 ret = LTTNG_ERR_SESSION_FAIL;
4030 free(uris);
4031 goto error;
4032 }
4033 }
4034
4035 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
4036 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
4037 free(uris);
4038 break;
4039 }
4040 case LTTNG_SAVE_SESSION:
4041 {
4042 ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr,
4043 &cmd_ctx->creds);
4044 break;
4045 }
4046 case LTTNG_SET_SESSION_SHM_PATH:
4047 {
4048 ret = cmd_set_session_shm_path(cmd_ctx->session,
4049 cmd_ctx->lsm->u.set_shm_path.shm_path);
4050 break;
4051 }
4052 case LTTNG_REGENERATE_METADATA:
4053 {
4054 ret = cmd_regenerate_metadata(cmd_ctx->session);
4055 break;
4056 }
4057 case LTTNG_REGENERATE_STATEDUMP:
4058 {
4059 ret = cmd_regenerate_statedump(cmd_ctx->session);
4060 break;
4061 }
4062 case LTTNG_REGISTER_TRIGGER:
4063 {
4064 ret = cmd_register_trigger(cmd_ctx, sock,
4065 notification_thread_handle);
4066 break;
4067 }
4068 case LTTNG_UNREGISTER_TRIGGER:
4069 {
4070 ret = cmd_unregister_trigger(cmd_ctx, sock,
4071 notification_thread_handle);
4072 break;
4073 }
4074 case LTTNG_ROTATE_SESSION:
4075 {
4076 struct lttng_rotate_session_return rotate_return;
4077
4078 DBG("Client rotate session \"%s\"", cmd_ctx->session->name);
4079
4080 memset(&rotate_return, 0, sizeof(rotate_return));
4081 if (cmd_ctx->session->kernel_session && !check_rotate_compatible()) {
4082 DBG("Kernel tracer version is not compatible with the rotation feature");
4083 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
4084 goto error;
4085 }
4086
4087 ret = cmd_rotate_session(cmd_ctx->session, &rotate_return);
4088 if (ret < 0) {
4089 ret = -ret;
4090 goto error;
4091 }
4092
4093 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &rotate_return,
4094 sizeof(rotate_return));
4095 if (ret < 0) {
4096 ret = -ret;
4097 goto error;
4098 }
4099
4100 ret = LTTNG_OK;
4101 break;
4102 }
4103 case LTTNG_ROTATION_GET_INFO:
4104 {
4105 struct lttng_rotation_get_info_return get_info_return;
4106
4107 memset(&get_info_return, 0, sizeof(get_info_return));
4108 ret = cmd_rotate_get_info(cmd_ctx->session, &get_info_return,
4109 cmd_ctx->lsm->u.get_rotation_info.rotation_id);
4110 if (ret < 0) {
4111 ret = -ret;
4112 goto error;
4113 }
4114
4115 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &get_info_return,
4116 sizeof(get_info_return));
4117 if (ret < 0) {
4118 ret = -ret;
4119 goto error;
4120 }
4121
4122 ret = LTTNG_OK;
4123 break;
4124 }
4125 case LTTNG_ROTATION_SET_SCHEDULE:
4126 {
4127 bool set_schedule;
4128 enum lttng_rotation_schedule_type schedule_type;
4129 uint64_t value;
4130
4131 if (cmd_ctx->session->kernel_session && !check_rotate_compatible()) {
4132 DBG("Kernel tracer version does not support session rotations");
4133 ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
4134 goto error;
4135 }
4136
4137 set_schedule = cmd_ctx->lsm->u.rotation_set_schedule.set == 1;
4138 schedule_type = (enum lttng_rotation_schedule_type) cmd_ctx->lsm->u.rotation_set_schedule.type;
4139 value = cmd_ctx->lsm->u.rotation_set_schedule.value;
4140
4141 ret = cmd_rotation_set_schedule(cmd_ctx->session,
4142 set_schedule,
4143 schedule_type,
4144 value,
4145 notification_thread_handle);
4146 if (ret != LTTNG_OK) {
4147 goto error;
4148 }
4149
4150 break;
4151 }
4152 case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
4153 {
4154 struct lttng_session_list_schedules_return schedules = {
4155 .periodic.set = !!cmd_ctx->session->rotate_timer_period,
4156 .periodic.value = cmd_ctx->session->rotate_timer_period,
4157 .size.set = !!cmd_ctx->session->rotate_size,
4158 .size.value = cmd_ctx->session->rotate_size,
4159 };
4160
4161 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &schedules,
4162 sizeof(schedules));
4163 if (ret < 0) {
4164 ret = -ret;
4165 goto error;
4166 }
4167
4168 ret = LTTNG_OK;
4169 break;
4170 }
4171 default:
4172 ret = LTTNG_ERR_UND;
4173 break;
4174 }
4175
4176 error:
4177 if (cmd_ctx->llm == NULL) {
4178 DBG("Missing llm structure. Allocating one.");
4179 if (setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0) < 0) {
4180 goto setup_error;
4181 }
4182 }
4183 /* Set return code */
4184 cmd_ctx->llm->ret_code = ret;
4185 setup_error:
4186 if (cmd_ctx->session) {
4187 session_unlock(cmd_ctx->session);
4188 session_put(cmd_ctx->session);
4189 }
4190 if (need_tracing_session) {
4191 session_unlock_list();
4192 }
4193 init_setup_error:
4194 assert(!rcu_read_ongoing());
4195 return ret;
4196 }
4197
4198 /*
4199 * Thread managing health check socket.
4200 */
4201 static void *thread_manage_health(void *data)
4202 {
4203 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
4204 uint32_t revents, nb_fd;
4205 struct lttng_poll_event events;
4206 struct health_comm_msg msg;
4207 struct health_comm_reply reply;
4208
4209 DBG("[thread] Manage health check started");
4210
4211 rcu_register_thread();
4212
4213 /* We might hit an error path before this is created. */
4214 lttng_poll_init(&events);
4215
4216 /* Create unix socket */
4217 sock = lttcomm_create_unix_sock(config.health_unix_sock_path.value);
4218 if (sock < 0) {
4219 ERR("Unable to create health check Unix socket");
4220 goto error;
4221 }
4222
4223 if (is_root) {
4224 /* lttng health client socket path permissions */
4225 ret = chown(config.health_unix_sock_path.value, 0,
4226 utils_get_group_id(config.tracing_group_name.value));
4227 if (ret < 0) {
4228 ERR("Unable to set group on %s", config.health_unix_sock_path.value);
4229 PERROR("chown");
4230 goto error;
4231 }
4232
4233 ret = chmod(config.health_unix_sock_path.value,
4234 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4235 if (ret < 0) {
4236 ERR("Unable to set permissions on %s", config.health_unix_sock_path.value);
4237 PERROR("chmod");
4238 goto error;
4239 }
4240 }
4241
4242 /*
4243 * Set the CLOEXEC flag. Return code is useless because either way, the
4244 * show must go on.
4245 */
4246 (void) utils_set_fd_cloexec(sock);
4247
4248 ret = lttcomm_listen_unix_sock(sock);
4249 if (ret < 0) {
4250 goto error;
4251 }
4252
4253 /*
4254 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4255 * more will be added to this poll set.
4256 */
4257 ret = sessiond_set_thread_pollset(&events, 2);
4258 if (ret < 0) {
4259 goto error;
4260 }
4261
4262 /* Add the application registration socket */
4263 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
4264 if (ret < 0) {
4265 goto error;
4266 }
4267
4268 sessiond_notify_ready();
4269
4270 while (1) {
4271 DBG("Health check ready");
4272
4273 /* Inifinite blocking call, waiting for transmission */
4274 restart:
4275 ret = lttng_poll_wait(&events, -1);
4276 if (ret < 0) {
4277 /*
4278 * Restart interrupted system call.
4279 */
4280 if (errno == EINTR) {
4281 goto restart;
4282 }
4283 goto error;
4284 }
4285
4286 nb_fd = ret;
4287
4288 for (i = 0; i < nb_fd; i++) {
4289 /* Fetch once the poll data */
4290 revents = LTTNG_POLL_GETEV(&events, i);
4291 pollfd = LTTNG_POLL_GETFD(&events, i);
4292
4293 if (!revents) {
4294 /* No activity for this FD (poll implementation). */
4295 continue;
4296 }
4297
4298 /* Thread quit pipe has been closed. Killing thread. */
4299 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4300 if (ret) {
4301 err = 0;
4302 goto exit;
4303 }
4304
4305 /* Event on the registration socket */
4306 if (pollfd == sock) {
4307 if (revents & LPOLLIN) {
4308 continue;
4309 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4310 ERR("Health socket poll error");
4311 goto error;
4312 } else {
4313 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4314 goto error;
4315 }
4316 }
4317 }
4318
4319 new_sock = lttcomm_accept_unix_sock(sock);
4320 if (new_sock < 0) {
4321 goto error;
4322 }
4323
4324 /*
4325 * Set the CLOEXEC flag. Return code is useless because either way, the
4326 * show must go on.
4327 */
4328 (void) utils_set_fd_cloexec(new_sock);
4329
4330 DBG("Receiving data from client for health...");
4331 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
4332 if (ret <= 0) {
4333 DBG("Nothing recv() from client... continuing");
4334 ret = close(new_sock);
4335 if (ret) {
4336 PERROR("close");
4337 }
4338 continue;
4339 }
4340
4341 rcu_thread_online();
4342
4343 memset(&reply, 0, sizeof(reply));
4344 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
4345 /*
4346 * health_check_state returns 0 if health is
4347 * bad.
4348 */
4349 if (!health_check_state(health_sessiond, i)) {
4350 reply.ret_code |= 1ULL << i;
4351 }
4352 }
4353
4354 DBG2("Health check return value %" PRIx64, reply.ret_code);
4355
4356 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
4357 if (ret < 0) {
4358 ERR("Failed to send health data back to client");
4359 }
4360
4361 /* End of transmission */
4362 ret = close(new_sock);
4363 if (ret) {
4364 PERROR("close");
4365 }
4366 }
4367
4368 exit:
4369 error:
4370 if (err) {
4371 ERR("Health error occurred in %s", __func__);
4372 }
4373 DBG("Health check thread dying");
4374 unlink(config.health_unix_sock_path.value);
4375 if (sock >= 0) {
4376 ret = close(sock);
4377 if (ret) {
4378 PERROR("close");
4379 }
4380 }
4381
4382 lttng_poll_clean(&events);
4383 stop_threads();
4384 rcu_unregister_thread();
4385 return NULL;
4386 }
4387
4388 /*
4389 * This thread manage all clients request using the unix client socket for
4390 * communication.
4391 */
4392 static void *thread_manage_clients(void *data)
4393 {
4394 int sock = -1, ret, i, pollfd, err = -1;
4395 int sock_error;
4396 uint32_t revents, nb_fd;
4397 struct command_ctx *cmd_ctx = NULL;
4398 struct lttng_poll_event events;
4399
4400 DBG("[thread] Manage client started");
4401
4402 rcu_register_thread();
4403
4404 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
4405
4406 health_code_update();
4407
4408 ret = lttcomm_listen_unix_sock(client_sock);
4409 if (ret < 0) {
4410 goto error_listen;
4411 }
4412
4413 /*
4414 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4415 * more will be added to this poll set.
4416 */
4417 ret = sessiond_set_thread_pollset(&events, 2);
4418 if (ret < 0) {
4419 goto error_create_poll;
4420 }
4421
4422 /* Add the application registration socket */
4423 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
4424 if (ret < 0) {
4425 goto error;
4426 }
4427
4428 ret = sem_post(&load_info->message_thread_ready);
4429 if (ret) {
4430 PERROR("sem_post message_thread_ready");
4431 goto error;
4432 }
4433
4434 /*
4435 * Wait until all support threads are initialized before accepting
4436 * commands.
4437 */
4438 while (uatomic_read(&lttng_sessiond_ready) != 0) {
4439 /*
4440 * If a support thread failed to launch, it may signal that
4441 * we must exit and the sessiond would never be marked as
4442 * "ready".
4443 *
4444 * The timeout is set to 1ms, which serves as a way to
4445 * pace down this check.
4446 */
4447 ret = sessiond_wait_for_quit_pipe(1000);
4448 if (ret > 0) {
4449 goto exit;
4450 }
4451 }
4452 /*
4453 * This barrier is paired with the one in sessiond_notify_ready() to
4454 * ensure that loads accessing data initialized by the other threads,
4455 * on which this thread was waiting, are not performed before this point.
4456 *
4457 * Note that this could be a 'read' memory barrier, but a full barrier
4458 * is used in case the code changes. The performance implications of
4459 * this choice are minimal since this is a slow path.
4460 */
4461 cmm_smp_mb();
4462
4463 /* This testpoint is after we signal readiness to the parent. */
4464 if (testpoint(sessiond_thread_manage_clients)) {
4465 goto error;
4466 }
4467
4468 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
4469 goto error;
4470 }
4471
4472 health_code_update();
4473
4474 /* Set state as running. */
4475 sessiond_set_client_thread_state(true);
4476
4477 while (1) {
4478 const struct cmd_completion_handler *cmd_completion_handler;
4479
4480 DBG("Accepting client command ...");
4481
4482 /* Inifinite blocking call, waiting for transmission */
4483 restart:
4484 health_poll_entry();
4485 ret = lttng_poll_wait(&events, -1);
4486 health_poll_exit();
4487 if (ret < 0) {
4488 /*
4489 * Restart interrupted system call.
4490 */
4491 if (errno == EINTR) {
4492 goto restart;
4493 }
4494 goto error;
4495 }
4496
4497 nb_fd = ret;
4498
4499 for (i = 0; i < nb_fd; i++) {
4500 /* Fetch once the poll data */
4501 revents = LTTNG_POLL_GETEV(&events, i);
4502 pollfd = LTTNG_POLL_GETFD(&events, i);
4503
4504 health_code_update();
4505
4506 if (!revents) {
4507 /* No activity for this FD (poll implementation). */
4508 continue;
4509 }
4510
4511 /* Thread quit pipe has been closed. Killing thread. */
4512 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
4513 if (ret) {
4514 err = 0;
4515 goto exit;
4516 }
4517
4518 /* Event on the registration socket */
4519 if (pollfd == client_sock) {
4520 if (revents & LPOLLIN) {
4521 continue;
4522 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4523 ERR("Client socket poll error");
4524 goto error;
4525 } else {
4526 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4527 goto error;
4528 }
4529 }
4530 }
4531
4532 DBG("Wait for client response");
4533
4534 health_code_update();
4535
4536 sock = lttcomm_accept_unix_sock(client_sock);
4537 if (sock < 0) {
4538 goto error;
4539 }
4540
4541 /*
4542 * Set the CLOEXEC flag. Return code is useless because either way, the
4543 * show must go on.
4544 */
4545 (void) utils_set_fd_cloexec(sock);
4546
4547 /* Set socket option for credentials retrieval */
4548 ret = lttcomm_setsockopt_creds_unix_sock(sock);
4549 if (ret < 0) {
4550 goto error;
4551 }
4552
4553 /* Allocate context command to process the client request */
4554 cmd_ctx = zmalloc(sizeof(struct command_ctx));
4555 if (cmd_ctx == NULL) {
4556 PERROR("zmalloc cmd_ctx");
4557 goto error;
4558 }
4559
4560 /* Allocate data buffer for reception */
4561 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
4562 if (cmd_ctx->lsm == NULL) {
4563 PERROR("zmalloc cmd_ctx->lsm");
4564 goto error;
4565 }
4566
4567 cmd_ctx->llm = NULL;
4568 cmd_ctx->session = NULL;
4569
4570 health_code_update();
4571
4572 /*
4573 * Data is received from the lttng client. The struct
4574 * lttcomm_session_msg (lsm) contains the command and data request of
4575 * the client.
4576 */
4577 DBG("Receiving data from client ...");
4578 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
4579 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
4580 if (ret <= 0) {
4581 DBG("Nothing recv() from client... continuing");
4582 ret = close(sock);
4583 if (ret) {
4584 PERROR("close");
4585 }
4586 sock = -1;
4587 clean_command_ctx(&cmd_ctx);
4588 continue;
4589 }
4590
4591 health_code_update();
4592
4593 // TODO: Validate cmd_ctx including sanity check for
4594 // security purpose.
4595
4596 rcu_thread_online();
4597 /*
4598 * This function dispatch the work to the kernel or userspace tracer
4599 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4600 * informations for the client. The command context struct contains
4601 * everything this function may needs.
4602 */
4603 ret = process_client_msg(cmd_ctx, sock, &sock_error);
4604 rcu_thread_offline();
4605 if (ret < 0) {
4606 ret = close(sock);
4607 if (ret) {
4608 PERROR("close");
4609 }
4610 sock = -1;
4611 /*
4612 * TODO: Inform client somehow of the fatal error. At
4613 * this point, ret < 0 means that a zmalloc failed
4614 * (ENOMEM). Error detected but still accept
4615 * command, unless a socket error has been
4616 * detected.
4617 */
4618 clean_command_ctx(&cmd_ctx);
4619 continue;
4620 }
4621
4622 cmd_completion_handler = cmd_pop_completion_handler();
4623 if (cmd_completion_handler) {
4624 enum lttng_error_code completion_code;
4625
4626 completion_code = cmd_completion_handler->run(
4627 cmd_completion_handler->data);
4628 if (completion_code != LTTNG_OK) {
4629 clean_command_ctx(&cmd_ctx);
4630 continue;
4631 }
4632 }
4633
4634 health_code_update();
4635
4636 DBG("Sending response (size: %d, retcode: %s (%d))",
4637 cmd_ctx->lttng_msg_size,
4638 lttng_strerror(-cmd_ctx->llm->ret_code),
4639 cmd_ctx->llm->ret_code);
4640 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
4641 if (ret < 0) {
4642 ERR("Failed to send data back to client");
4643 }
4644
4645 /* End of transmission */
4646 ret = close(sock);
4647 if (ret) {
4648 PERROR("close");
4649 }
4650 sock = -1;
4651
4652 clean_command_ctx(&cmd_ctx);
4653
4654 health_code_update();
4655 }
4656
4657 exit:
4658 error:
4659 if (sock >= 0) {
4660 ret = close(sock);
4661 if (ret) {
4662 PERROR("close");
4663 }
4664 }
4665
4666 lttng_poll_clean(&events);
4667 clean_command_ctx(&cmd_ctx);
4668
4669 error_listen:
4670 error_create_poll:
4671 unlink(config.client_unix_sock_path.value);
4672 if (client_sock >= 0) {
4673 ret = close(client_sock);
4674 if (ret) {
4675 PERROR("close");
4676 }
4677 }
4678
4679 if (err) {
4680 health_error();
4681 ERR("Health error occurred in %s", __func__);
4682 }
4683
4684 health_unregister(health_sessiond);
4685
4686 DBG("Client thread dying");
4687
4688 rcu_unregister_thread();
4689
4690 /*
4691 * Since we are creating the consumer threads, we own them, so we need
4692 * to join them before our thread exits.
4693 */
4694 ret = join_consumer_thread(&kconsumer_data);
4695 if (ret) {
4696 errno = ret;
4697 PERROR("join_consumer");
4698 }
4699
4700 ret = join_consumer_thread(&ustconsumer32_data);
4701 if (ret) {
4702 errno = ret;
4703 PERROR("join_consumer ust32");
4704 }
4705
4706 ret = join_consumer_thread(&ustconsumer64_data);
4707 if (ret) {
4708 errno = ret;
4709 PERROR("join_consumer ust64");
4710 }
4711
4712 /* Set state as non-running. */
4713 sessiond_set_client_thread_state(false);
4714 return NULL;
4715 }
4716
4717 static int string_match(const char *str1, const char *str2)
4718 {
4719 return (str1 && str2) && !strcmp(str1, str2);
4720 }
4721
4722 /*
4723 * Take an option from the getopt output and set it in the right variable to be
4724 * used later.
4725 *
4726 * Return 0 on success else a negative value.
4727 */
4728 static int set_option(int opt, const char *arg, const char *optname)
4729 {
4730 int ret = 0;
4731
4732 if (string_match(optname, "client-sock") || opt == 'c') {
4733 if (!arg || *arg == '\0') {
4734 ret = -EINVAL;
4735 goto end;
4736 }
4737 if (lttng_is_setuid_setgid()) {
4738 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4739 "-c, --client-sock");
4740 } else {
4741 config_string_set(&config.client_unix_sock_path,
4742 strdup(arg));
4743 if (!config.client_unix_sock_path.value) {
4744 ret = -ENOMEM;
4745 PERROR("strdup");
4746 }
4747 }
4748 } else if (string_match(optname, "apps-sock") || opt == 'a') {
4749 if (!arg || *arg == '\0') {
4750 ret = -EINVAL;
4751 goto end;
4752 }
4753 if (lttng_is_setuid_setgid()) {
4754 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4755 "-a, --apps-sock");
4756 } else {
4757 config_string_set(&config.apps_unix_sock_path,
4758 strdup(arg));
4759 if (!config.apps_unix_sock_path.value) {
4760 ret = -ENOMEM;
4761 PERROR("strdup");
4762 }
4763 }
4764 } else if (string_match(optname, "daemonize") || opt == 'd') {
4765 config.daemonize = true;
4766 } else if (string_match(optname, "background") || opt == 'b') {
4767 config.background = true;
4768 } else if (string_match(optname, "group") || opt == 'g') {
4769 if (!arg || *arg == '\0') {
4770 ret = -EINVAL;
4771 goto end;
4772 }
4773 if (lttng_is_setuid_setgid()) {
4774 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4775 "-g, --group");
4776 } else {
4777 config_string_set(&config.tracing_group_name,
4778 strdup(arg));
4779 if (!config.tracing_group_name.value) {
4780 ret = -ENOMEM;
4781 PERROR("strdup");
4782 }
4783 }
4784 } else if (string_match(optname, "help") || opt == 'h') {
4785 ret = utils_show_help(8, "lttng-sessiond", help_msg);
4786 if (ret) {
4787 ERR("Cannot show --help for `lttng-sessiond`");
4788 perror("exec");
4789 }
4790 exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
4791 } else if (string_match(optname, "version") || opt == 'V') {
4792 fprintf(stdout, "%s\n", VERSION);
4793 exit(EXIT_SUCCESS);
4794 } else if (string_match(optname, "sig-parent") || opt == 'S') {
4795 config.sig_parent = true;
4796 } else if (string_match(optname, "kconsumerd-err-sock")) {
4797 if (!arg || *arg == '\0') {
4798 ret = -EINVAL;
4799 goto end;
4800 }
4801 if (lttng_is_setuid_setgid()) {
4802 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4803 "--kconsumerd-err-sock");
4804 } else {
4805 config_string_set(&config.kconsumerd_err_unix_sock_path,
4806 strdup(arg));
4807 if (!config.kconsumerd_err_unix_sock_path.value) {
4808 ret = -ENOMEM;
4809 PERROR("strdup");
4810 }
4811 }
4812 } else if (string_match(optname, "kconsumerd-cmd-sock")) {
4813 if (!arg || *arg == '\0') {
4814 ret = -EINVAL;
4815 goto end;
4816 }
4817 if (lttng_is_setuid_setgid()) {
4818 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4819 "--kconsumerd-cmd-sock");
4820 } else {
4821 config_string_set(&config.kconsumerd_cmd_unix_sock_path,
4822 strdup(arg));
4823 if (!config.kconsumerd_cmd_unix_sock_path.value) {
4824 ret = -ENOMEM;
4825 PERROR("strdup");
4826 }
4827 }
4828 } else if (string_match(optname, "ustconsumerd64-err-sock")) {
4829 if (!arg || *arg == '\0') {
4830 ret = -EINVAL;
4831 goto end;
4832 }
4833 if (lttng_is_setuid_setgid()) {
4834 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4835 "--ustconsumerd64-err-sock");
4836 } else {
4837 config_string_set(&config.consumerd64_err_unix_sock_path,
4838 strdup(arg));
4839 if (!config.consumerd64_err_unix_sock_path.value) {
4840 ret = -ENOMEM;
4841 PERROR("strdup");
4842 }
4843 }
4844 } else if (string_match(optname, "ustconsumerd64-cmd-sock")) {
4845 if (!arg || *arg == '\0') {
4846 ret = -EINVAL;
4847 goto end;
4848 }
4849 if (lttng_is_setuid_setgid()) {
4850 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4851 "--ustconsumerd64-cmd-sock");
4852 } else {
4853 config_string_set(&config.consumerd64_cmd_unix_sock_path,
4854 strdup(arg));
4855 if (!config.consumerd64_cmd_unix_sock_path.value) {
4856 ret = -ENOMEM;
4857 PERROR("strdup");
4858 }
4859 }
4860 } else if (string_match(optname, "ustconsumerd32-err-sock")) {
4861 if (!arg || *arg == '\0') {
4862 ret = -EINVAL;
4863 goto end;
4864 }
4865 if (lttng_is_setuid_setgid()) {
4866 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4867 "--ustconsumerd32-err-sock");
4868 } else {
4869 config_string_set(&config.consumerd32_err_unix_sock_path,
4870 strdup(arg));
4871 if (!config.consumerd32_err_unix_sock_path.value) {
4872 ret = -ENOMEM;
4873 PERROR("strdup");
4874 }
4875 }
4876 } else if (string_match(optname, "ustconsumerd32-cmd-sock")) {
4877 if (!arg || *arg == '\0') {
4878 ret = -EINVAL;
4879 goto end;
4880 }
4881 if (lttng_is_setuid_setgid()) {
4882 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4883 "--ustconsumerd32-cmd-sock");
4884 } else {
4885 config_string_set(&config.consumerd32_cmd_unix_sock_path,
4886 strdup(arg));
4887 if (!config.consumerd32_cmd_unix_sock_path.value) {
4888 ret = -ENOMEM;
4889 PERROR("strdup");
4890 }
4891 }
4892 } else if (string_match(optname, "no-kernel")) {
4893 config.no_kernel = true;
4894 } else if (string_match(optname, "quiet") || opt == 'q') {
4895 config.quiet = true;
4896 } else if (string_match(optname, "verbose") || opt == 'v') {
4897 /* Verbose level can increase using multiple -v */
4898 if (arg) {
4899 /* Value obtained from config file */
4900 config.verbose = config_parse_value(arg);
4901 } else {
4902 /* -v used on command line */
4903 config.verbose++;
4904 }
4905 /* Clamp value to [0, 3] */
4906 config.verbose = config.verbose < 0 ? 0 :
4907 (config.verbose <= 3 ? config.verbose : 3);
4908 } else if (string_match(optname, "verbose-consumer")) {
4909 if (arg) {
4910 config.verbose_consumer = config_parse_value(arg);
4911 } else {
4912 config.verbose_consumer++;
4913 }
4914 } else if (string_match(optname, "consumerd32-path")) {
4915 if (!arg || *arg == '\0') {
4916 ret = -EINVAL;
4917 goto end;
4918 }
4919 if (lttng_is_setuid_setgid()) {
4920 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4921 "--consumerd32-path");
4922 } else {
4923 config_string_set(&config.consumerd32_bin_path,
4924 strdup(arg));
4925 if (!config.consumerd32_bin_path.value) {
4926 PERROR("strdup");
4927 ret = -ENOMEM;
4928 }
4929 }
4930 } else if (string_match(optname, "consumerd32-libdir")) {
4931 if (!arg || *arg == '\0') {
4932 ret = -EINVAL;
4933 goto end;
4934 }
4935 if (lttng_is_setuid_setgid()) {
4936 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4937 "--consumerd32-libdir");
4938 } else {
4939 config_string_set(&config.consumerd32_lib_dir,
4940 strdup(arg));
4941 if (!config.consumerd32_lib_dir.value) {
4942 PERROR("strdup");
4943 ret = -ENOMEM;
4944 }
4945 }
4946 } else if (string_match(optname, "consumerd64-path")) {
4947 if (!arg || *arg == '\0') {
4948 ret = -EINVAL;
4949 goto end;
4950 }
4951 if (lttng_is_setuid_setgid()) {
4952 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4953 "--consumerd64-path");
4954 } else {
4955 config_string_set(&config.consumerd64_bin_path,
4956 strdup(arg));
4957 if (!config.consumerd64_bin_path.value) {
4958 PERROR("strdup");
4959 ret = -ENOMEM;
4960 }
4961 }
4962 } else if (string_match(optname, "consumerd64-libdir")) {
4963 if (!arg || *arg == '\0') {
4964 ret = -EINVAL;
4965 goto end;
4966 }
4967 if (lttng_is_setuid_setgid()) {
4968 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4969 "--consumerd64-libdir");
4970 } else {
4971 config_string_set(&config.consumerd64_lib_dir,
4972 strdup(arg));
4973 if (!config.consumerd64_lib_dir.value) {
4974 PERROR("strdup");
4975 ret = -ENOMEM;
4976 }
4977 }
4978 } else if (string_match(optname, "pidfile") || opt == 'p') {
4979 if (!arg || *arg == '\0') {
4980 ret = -EINVAL;
4981 goto end;
4982 }
4983 if (lttng_is_setuid_setgid()) {
4984 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4985 "-p, --pidfile");
4986 } else {
4987 config_string_set(&config.pid_file_path, strdup(arg));
4988 if (!config.pid_file_path.value) {
4989 PERROR("strdup");
4990 ret = -ENOMEM;
4991 }
4992 }
4993 } else if (string_match(optname, "agent-tcp-port")) {
4994 if (!arg || *arg == '\0') {
4995 ret = -EINVAL;
4996 goto end;
4997 }
4998 if (lttng_is_setuid_setgid()) {
4999 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5000 "--agent-tcp-port");
5001 } else {
5002 unsigned long v;
5003
5004 errno = 0;
5005 v = strtoul(arg, NULL, 0);
5006 if (errno != 0 || !isdigit(arg[0])) {
5007 ERR("Wrong value in --agent-tcp-port parameter: %s", arg);
5008 return -1;
5009 }
5010 if (v == 0 || v >= 65535) {
5011 ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
5012 return -1;
5013 }
5014 config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v;
5015 DBG3("Agent TCP port set to non default: %i", (int) v);
5016 }
5017 } else if (string_match(optname, "load") || opt == 'l') {
5018 if (!arg || *arg == '\0') {
5019 ret = -EINVAL;
5020 goto end;
5021 }
5022 if (lttng_is_setuid_setgid()) {
5023 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5024 "-l, --load");
5025 } else {
5026 config_string_set(&config.load_session_path, strdup(arg));
5027 if (!config.load_session_path.value) {
5028 PERROR("strdup");
5029 ret = -ENOMEM;
5030 }
5031 }
5032 } else if (string_match(optname, "kmod-probes")) {
5033 if (!arg || *arg == '\0') {
5034 ret = -EINVAL;
5035 goto end;
5036 }
5037 if (lttng_is_setuid_setgid()) {
5038 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5039 "--kmod-probes");
5040 } else {
5041 config_string_set(&config.kmod_probes_list, strdup(arg));
5042 if (!config.kmod_probes_list.value) {
5043 PERROR("strdup");
5044 ret = -ENOMEM;
5045 }
5046 }
5047 } else if (string_match(optname, "extra-kmod-probes")) {
5048 if (!arg || *arg == '\0') {
5049 ret = -EINVAL;
5050 goto end;
5051 }
5052 if (lttng_is_setuid_setgid()) {
5053 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5054 "--extra-kmod-probes");
5055 } else {
5056 config_string_set(&config.kmod_extra_probes_list,
5057 strdup(arg));
5058 if (!config.kmod_extra_probes_list.value) {
5059 PERROR("strdup");
5060 ret = -ENOMEM;
5061 }
5062 }
5063 } else if (string_match(optname, "config") || opt == 'f') {
5064 /* This is handled in set_options() thus silent skip. */
5065 goto end;
5066 } else {
5067 /* Unknown option or other error.
5068 * Error is printed by getopt, just return */
5069 ret = -1;
5070 }
5071
5072 end:
5073 if (ret == -EINVAL) {
5074 const char *opt_name = "unknown";
5075 int i;
5076
5077 for (i = 0; i < sizeof(long_options) / sizeof(struct option);
5078 i++) {
5079 if (opt == long_options[i].val) {
5080 opt_name = long_options[i].name;
5081 break;
5082 }
5083 }
5084
5085 WARN("Invalid argument provided for option \"%s\", using default value.",
5086 opt_name);
5087 }
5088
5089 return ret;
5090 }
5091
5092 /*
5093 * config_entry_handler_cb used to handle options read from a config file.
5094 * See config_entry_handler_cb comment in common/config/session-config.h for the
5095 * return value conventions.
5096 */
5097 static int config_entry_handler(const struct config_entry *entry, void *unused)
5098 {
5099 int ret = 0, i;
5100
5101 if (!entry || !entry->name || !entry->value) {
5102 ret = -EINVAL;
5103 goto end;
5104 }
5105
5106 /* Check if the option is to be ignored */
5107 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
5108 if (!strcmp(entry->name, config_ignore_options[i])) {
5109 goto end;
5110 }
5111 }
5112
5113 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1;
5114 i++) {
5115
5116 /* Ignore if not fully matched. */
5117 if (strcmp(entry->name, long_options[i].name)) {
5118 continue;
5119 }
5120
5121 /*
5122 * If the option takes no argument on the command line, we have to
5123 * check if the value is "true". We support non-zero numeric values,
5124 * true, on and yes.
5125 */
5126 if (!long_options[i].has_arg) {
5127 ret = config_parse_value(entry->value);
5128 if (ret <= 0) {
5129 if (ret) {
5130 WARN("Invalid configuration value \"%s\" for option %s",
5131 entry->value, entry->name);
5132 }
5133 /* False, skip boolean config option. */
5134 goto end;
5135 }
5136 }
5137
5138 ret = set_option(long_options[i].val, entry->value, entry->name);
5139 goto end;
5140 }
5141
5142 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
5143
5144 end:
5145 return ret;
5146 }
5147
5148 /*
5149 * daemon configuration loading and argument parsing
5150 */
5151 static int set_options(int argc, char **argv)
5152 {
5153 int ret = 0, c = 0, option_index = 0;
5154 int orig_optopt = optopt, orig_optind = optind;
5155 char *optstring;
5156 const char *config_path = NULL;
5157
5158 optstring = utils_generate_optstring(long_options,
5159 sizeof(long_options) / sizeof(struct option));
5160 if (!optstring) {
5161 ret = -ENOMEM;
5162 goto end;
5163 }
5164
5165 /* Check for the --config option */
5166 while ((c = getopt_long(argc, argv, optstring, long_options,
5167 &option_index)) != -1) {
5168 if (c == '?') {
5169 ret = -EINVAL;
5170 goto end;
5171 } else if (c != 'f') {
5172 /* if not equal to --config option. */
5173 continue;
5174 }
5175
5176 if (lttng_is_setuid_setgid()) {
5177 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5178 "-f, --config");
5179 } else {
5180 config_path = utils_expand_path(optarg);
5181 if (!config_path) {
5182 ERR("Failed to resolve path: %s", optarg);
5183 }
5184 }
5185 }
5186
5187 ret = config_get_section_entries(config_path, config_section_name,
5188 config_entry_handler, NULL);
5189 if (ret) {
5190 if (ret > 0) {
5191 ERR("Invalid configuration option at line %i", ret);
5192 ret = -1;
5193 }
5194 goto end;
5195 }
5196
5197 /* Reset getopt's global state */
5198 optopt = orig_optopt;
5199 optind = orig_optind;
5200 while (1) {
5201 option_index = -1;
5202 /*
5203 * getopt_long() will not set option_index if it encounters a
5204 * short option.
5205 */
5206 c = getopt_long(argc, argv, optstring, long_options,
5207 &option_index);
5208 if (c == -1) {
5209 break;
5210 }
5211
5212 /*
5213 * Pass NULL as the long option name if popt left the index
5214 * unset.
5215 */
5216 ret = set_option(c, optarg,
5217 option_index < 0 ? NULL :
5218 long_options[option_index].name);
5219 if (ret < 0) {
5220 break;
5221 }
5222 }
5223
5224 end:
5225 free(optstring);
5226 return ret;
5227 }
5228
5229 /*
5230 * Creates the two needed socket by the daemon.
5231 * apps_sock - The communication socket for all UST apps.
5232 * client_sock - The communication of the cli tool (lttng).
5233 */
5234 static int init_daemon_socket(void)
5235 {
5236 int ret = 0;
5237 mode_t old_umask;
5238
5239 old_umask = umask(0);
5240
5241 /* Create client tool unix socket */
5242 client_sock = lttcomm_create_unix_sock(config.client_unix_sock_path.value);
5243 if (client_sock < 0) {
5244 ERR("Create unix sock failed: %s", config.client_unix_sock_path.value);
5245 ret = -1;
5246 goto end;
5247 }
5248
5249 /* Set the cloexec flag */
5250 ret = utils_set_fd_cloexec(client_sock);
5251 if (ret < 0) {
5252 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5253 "Continuing but note that the consumer daemon will have a "
5254 "reference to this socket on exec()", client_sock);
5255 }
5256
5257 /* File permission MUST be 660 */
5258 ret = chmod(config.client_unix_sock_path.value, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5259 if (ret < 0) {
5260 ERR("Set file permissions failed: %s", config.client_unix_sock_path.value);
5261 PERROR("chmod");
5262 goto end;
5263 }
5264
5265 /* Create the application unix socket */
5266 apps_sock = lttcomm_create_unix_sock(config.apps_unix_sock_path.value);
5267 if (apps_sock < 0) {
5268 ERR("Create unix sock failed: %s", config.apps_unix_sock_path.value);
5269 ret = -1;
5270 goto end;
5271 }
5272
5273 /* Set the cloexec flag */
5274 ret = utils_set_fd_cloexec(apps_sock);
5275 if (ret < 0) {
5276 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5277 "Continuing but note that the consumer daemon will have a "
5278 "reference to this socket on exec()", apps_sock);
5279 }
5280
5281 /* File permission MUST be 666 */
5282 ret = chmod(config.apps_unix_sock_path.value,
5283 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
5284 if (ret < 0) {
5285 ERR("Set file permissions failed: %s", config.apps_unix_sock_path.value);
5286 PERROR("chmod");
5287 goto end;
5288 }
5289
5290 DBG3("Session daemon client socket %d and application socket %d created",
5291 client_sock, apps_sock);
5292
5293 end:
5294 umask(old_umask);
5295 return ret;
5296 }
5297
5298 /*
5299 * Create lockfile using the rundir and return its fd.
5300 */
5301 static int create_lockfile(void)
5302 {
5303 return utils_create_lock_file(config.lock_file_path.value);
5304 }
5305
5306 /*
5307 * Check if the global socket is available, and if a daemon is answering at the
5308 * other side. If yes, error is returned.
5309 *
5310 * Also attempts to create and hold the lock file.
5311 */
5312 static int check_existing_daemon(void)
5313 {
5314 int ret = 0;
5315
5316 /* Is there anybody out there ? */
5317 if (lttng_session_daemon_alive()) {
5318 ret = -EEXIST;
5319 goto end;
5320 }
5321
5322 lockfile_fd = create_lockfile();
5323 if (lockfile_fd < 0) {
5324 ret = -EEXIST;
5325 goto end;
5326 }
5327 end:
5328 return ret;
5329 }
5330
5331 static void sessiond_cleanup_lock_file(void)
5332 {
5333 int ret;
5334
5335 /*
5336 * Cleanup lock file by deleting it and finaly closing it which will
5337 * release the file system lock.
5338 */
5339 if (lockfile_fd >= 0) {
5340 ret = remove(config.lock_file_path.value);
5341 if (ret < 0) {
5342 PERROR("remove lock file");
5343 }
5344 ret = close(lockfile_fd);
5345 if (ret < 0) {
5346 PERROR("close lock file");
5347 }
5348 }
5349 }
5350
5351 /*
5352 * Set the tracing group gid onto the client socket.
5353 *
5354 * Race window between mkdir and chown is OK because we are going from more
5355 * permissive (root.root) to less permissive (root.tracing).
5356 */
5357 static int set_permissions(char *rundir)
5358 {
5359 int ret;
5360 gid_t gid;
5361
5362 gid = utils_get_group_id(config.tracing_group_name.value);
5363
5364 /* Set lttng run dir */
5365 ret = chown(rundir, 0, gid);
5366 if (ret < 0) {
5367 ERR("Unable to set group on %s", rundir);
5368 PERROR("chown");
5369 }
5370
5371 /*
5372 * Ensure all applications and tracing group can search the run
5373 * dir. Allow everyone to read the directory, since it does not
5374 * buy us anything to hide its content.
5375 */
5376 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
5377 if (ret < 0) {
5378 ERR("Unable to set permissions on %s", rundir);
5379 PERROR("chmod");
5380 }
5381
5382 /* lttng client socket path */
5383 ret = chown(config.client_unix_sock_path.value, 0, gid);
5384 if (ret < 0) {
5385 ERR("Unable to set group on %s", config.client_unix_sock_path.value);
5386 PERROR("chown");
5387 }
5388
5389 /* kconsumer error socket path */
5390 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
5391 if (ret < 0) {
5392 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
5393 PERROR("chown");
5394 }
5395
5396 /* 64-bit ustconsumer error socket path */
5397 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
5398 if (ret < 0) {
5399 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
5400 PERROR("chown");
5401 }
5402
5403 /* 32-bit ustconsumer compat32 error socket path */
5404 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
5405 if (ret < 0) {
5406 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
5407 PERROR("chown");
5408 }
5409
5410 DBG("All permissions are set");
5411
5412 return ret;
5413 }
5414
5415 /*
5416 * Create the lttng run directory needed for all global sockets and pipe.
5417 */
5418 static int create_lttng_rundir(void)
5419 {
5420 int ret;
5421
5422 DBG3("Creating LTTng run directory: %s", config.rundir.value);
5423
5424 ret = mkdir(config.rundir.value, S_IRWXU);
5425 if (ret < 0) {
5426 if (errno != EEXIST) {
5427 ERR("Unable to create %s", config.rundir.value);
5428 goto error;
5429 } else {
5430 ret = 0;
5431 }
5432 }
5433
5434 error:
5435 return ret;
5436 }
5437
5438 /*
5439 * Setup sockets and directory needed by the consumerds' communication with the
5440 * session daemon.
5441 */
5442 static int set_consumer_sockets(struct consumer_data *consumer_data)
5443 {
5444 int ret;
5445 char *path = NULL;
5446
5447 switch (consumer_data->type) {
5448 case LTTNG_CONSUMER_KERNEL:
5449 path = config.kconsumerd_path.value;
5450 break;
5451 case LTTNG_CONSUMER64_UST:
5452 path = config.consumerd64_path.value;
5453 break;
5454 case LTTNG_CONSUMER32_UST:
5455 path = config.consumerd32_path.value;
5456 break;
5457 default:
5458 ERR("Consumer type unknown");
5459 ret = -EINVAL;
5460 goto error;
5461 }
5462 assert(path);
5463
5464 DBG2("Creating consumer directory: %s", path);
5465
5466 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
5467 if (ret < 0 && errno != EEXIST) {
5468 PERROR("mkdir");
5469 ERR("Failed to create %s", path);
5470 goto error;
5471 }
5472 if (is_root) {
5473 ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value));
5474 if (ret < 0) {
5475 ERR("Unable to set group on %s", path);
5476 PERROR("chown");
5477 goto error;
5478 }
5479 }
5480
5481 /* Create the consumerd error unix socket */
5482 consumer_data->err_sock =
5483 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
5484 if (consumer_data->err_sock < 0) {
5485 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
5486 ret = -1;
5487 goto error;
5488 }
5489
5490 /*
5491 * Set the CLOEXEC flag. Return code is useless because either way, the
5492 * show must go on.
5493 */
5494 ret = utils_set_fd_cloexec(consumer_data->err_sock);
5495 if (ret < 0) {
5496 PERROR("utils_set_fd_cloexec");
5497 /* continue anyway */
5498 }
5499
5500 /* File permission MUST be 660 */
5501 ret = chmod(consumer_data->err_unix_sock_path,
5502 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
5503 if (ret < 0) {
5504 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
5505 PERROR("chmod");
5506 goto error;
5507 }
5508
5509 error:
5510 return ret;
5511 }
5512
5513 /*
5514 * Signal handler for the daemon
5515 *
5516 * Simply stop all worker threads, leaving main() return gracefully after
5517 * joining all threads and calling cleanup().
5518 */
5519 static void sighandler(int sig)
5520 {
5521 switch (sig) {
5522 case SIGINT:
5523 DBG("SIGINT caught");
5524 stop_threads();
5525 break;
5526 case SIGTERM:
5527 DBG("SIGTERM caught");
5528 stop_threads();
5529 break;
5530 case SIGUSR1:
5531 CMM_STORE_SHARED(recv_child_signal, 1);
5532 break;
5533 default:
5534 break;
5535 }
5536 }
5537
5538 /*
5539 * Setup signal handler for :
5540 * SIGINT, SIGTERM, SIGPIPE
5541 */
5542 static int set_signal_handler(void)
5543 {
5544 int ret = 0;
5545 struct sigaction sa;
5546 sigset_t sigset;
5547
5548 if ((ret = sigemptyset(&sigset)) < 0) {
5549 PERROR("sigemptyset");
5550 return ret;
5551 }
5552
5553 sa.sa_mask = sigset;
5554 sa.sa_flags = 0;
5555
5556 sa.sa_handler = sighandler;
5557 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
5558 PERROR("sigaction");
5559 return ret;
5560 }
5561
5562 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
5563 PERROR("sigaction");
5564 return ret;
5565 }
5566
5567 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
5568 PERROR("sigaction");
5569 return ret;
5570 }
5571
5572 sa.sa_handler = SIG_IGN;
5573 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
5574 PERROR("sigaction");
5575 return ret;
5576 }
5577
5578 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5579
5580 return ret;
5581 }
5582
5583 /*
5584 * Set open files limit to unlimited. This daemon can open a large number of
5585 * file descriptors in order to consume multiple kernel traces.
5586 */
5587 static void set_ulimit(void)
5588 {
5589 int ret;
5590 struct rlimit lim;
5591
5592 /* The kernel does not allow an infinite limit for open files */
5593 lim.rlim_cur = 65535;
5594 lim.rlim_max = 65535;
5595
5596 ret = setrlimit(RLIMIT_NOFILE, &lim);
5597 if (ret < 0) {
5598 PERROR("failed to set open files limit");
5599 }
5600 }
5601
5602 static int write_pidfile(void)
5603 {
5604 return utils_create_pid_file(getpid(), config.pid_file_path.value);
5605 }
5606
5607 static int set_clock_plugin_env(void)
5608 {
5609 int ret = 0;
5610 char *env_value = NULL;
5611
5612 if (!config.lttng_ust_clock_plugin.value) {
5613 goto end;
5614 }
5615
5616 ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s",
5617 config.lttng_ust_clock_plugin.value);
5618 if (ret < 0) {
5619 PERROR("asprintf");
5620 goto end;
5621 }
5622
5623 ret = putenv(env_value);
5624 if (ret) {
5625 free(env_value);
5626 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5627 goto end;
5628 }
5629
5630 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5631 config.lttng_ust_clock_plugin.value);
5632 end:
5633 return ret;
5634 }
5635
5636 static void destroy_all_sessions_and_wait(void)
5637 {
5638 struct ltt_session *session, *tmp;
5639 struct ltt_session_list *session_list;
5640
5641 session_list = session_get_list();
5642 DBG("Initiating destruction of all sessions");
5643
5644 if (!session_list) {
5645 return;
5646 }
5647
5648 /*
5649 * Ensure that the client thread is no longer accepting new commands,
5650 * which could cause new sessions to be created.
5651 */
5652 sessiond_wait_client_thread_stopped();
5653
5654 session_lock_list();
5655 /* Initiate the destruction of all sessions. */
5656 cds_list_for_each_entry_safe(session, tmp,
5657 &session_list->head, list) {
5658 if (!session_get(session)) {
5659 continue;
5660 }
5661
5662 session_lock(session);
5663 if (session->destroyed) {
5664 goto unlock_session;
5665 }
5666 (void) cmd_destroy_session(session,
5667 notification_thread_handle);
5668 unlock_session:
5669 session_unlock(session);
5670 session_put(session);
5671 }
5672 session_unlock_list();
5673
5674 /* Wait for the destruction of all sessions to complete. */
5675 DBG("Waiting for the destruction of all sessions to complete");
5676 session_list_wait_empty();
5677 DBG("Destruction of all sessions completed");
5678 }
5679
5680 /*
5681 * main
5682 */
5683 int main(int argc, char **argv)
5684 {
5685 int ret = 0, retval = 0;
5686 void *status;
5687 const char *env_app_timeout;
5688 struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
5689 *ust64_channel_monitor_pipe = NULL,
5690 *kernel_channel_monitor_pipe = NULL;
5691 bool notification_thread_launched = false;
5692 bool rotation_thread_launched = false;
5693 bool timer_thread_launched = false;
5694 struct timer_thread_parameters timer_thread_ctx;
5695 /* Queue of rotation jobs populated by the sessiond-timer. */
5696 struct rotation_thread_timer_queue *rotation_timer_queue = NULL;
5697 sem_t notification_thread_ready;
5698
5699 init_kernel_workarounds();
5700
5701 rcu_register_thread();
5702
5703 if (set_signal_handler()) {
5704 retval = -1;
5705 goto exit_set_signal_handler;
5706 }
5707
5708 if (timer_signal_init()) {
5709 retval = -1;
5710 goto exit_set_signal_handler;
5711 }
5712
5713 page_size = sysconf(_SC_PAGESIZE);
5714 if (page_size < 0) {
5715 PERROR("sysconf _SC_PAGESIZE");
5716 page_size = LONG_MAX;
5717 WARN("Fallback page size to %ld", page_size);
5718 }
5719
5720 ret = sessiond_config_init(&config);
5721 if (ret) {
5722 retval = -1;
5723 goto exit_set_signal_handler;
5724 }
5725
5726 /*
5727 * Init config from environment variables.
5728 * Command line option override env configuration per-doc. Do env first.
5729 */
5730 sessiond_config_apply_env_config(&config);
5731
5732 /*
5733 * Parse arguments and load the daemon configuration file.
5734 *
5735 * We have an exit_options exit path to free memory reserved by
5736 * set_options. This is needed because the rest of sessiond_cleanup()
5737 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5738 * depends on set_options.
5739 */
5740 progname = argv[0];
5741 if (set_options(argc, argv)) {
5742 retval = -1;
5743 goto exit_options;
5744 }
5745
5746 /*
5747 * Resolve all paths received as arguments, configuration option, or
5748 * through environment variable as absolute paths. This is necessary
5749 * since daemonizing causes the sessiond's current working directory
5750 * to '/'.
5751 */
5752 ret = sessiond_config_resolve_paths(&config);
5753 if (ret) {
5754 goto exit_options;
5755 }
5756
5757 /* Apply config. */
5758 lttng_opt_verbose = config.verbose;
5759 lttng_opt_quiet = config.quiet;
5760 kconsumer_data.err_unix_sock_path =
5761 config.kconsumerd_err_unix_sock_path.value;
5762 kconsumer_data.cmd_unix_sock_path =
5763 config.kconsumerd_cmd_unix_sock_path.value;
5764 ustconsumer32_data.err_unix_sock_path =
5765 config.consumerd32_err_unix_sock_path.value;
5766 ustconsumer32_data.cmd_unix_sock_path =
5767 config.consumerd32_cmd_unix_sock_path.value;
5768 ustconsumer64_data.err_unix_sock_path =
5769 config.consumerd64_err_unix_sock_path.value;
5770 ustconsumer64_data.cmd_unix_sock_path =
5771 config.consumerd64_cmd_unix_sock_path.value;
5772 set_clock_plugin_env();
5773
5774 sessiond_config_log(&config);
5775
5776 if (create_lttng_rundir()) {
5777 retval = -1;
5778 goto exit_options;
5779 }
5780
5781 /* Abort launch if a session daemon is already running. */
5782 if (check_existing_daemon()) {
5783 ERR("A session daemon is already running.");
5784 retval = -1;
5785 goto exit_options;
5786 }
5787
5788 /* Daemonize */
5789 if (config.daemonize || config.background) {
5790 int i;
5791
5792 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
5793 !config.background);
5794 if (ret < 0) {
5795 retval = -1;
5796 goto exit_options;
5797 }
5798
5799 /*
5800 * We are in the child. Make sure all other file descriptors are
5801 * closed, in case we are called with more opened file
5802 * descriptors than the standard ones and the lock file.
5803 */
5804 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
5805 if (i == lockfile_fd) {
5806 continue;
5807 }
5808 (void) close(i);
5809 }
5810 }
5811
5812 if (run_as_create_worker(argv[0]) < 0) {
5813 goto exit_create_run_as_worker_cleanup;
5814 }
5815
5816 /*
5817 * Starting from here, we can create threads. This needs to be after
5818 * lttng_daemonize due to RCU.
5819 */
5820
5821 /*
5822 * Initialize the health check subsystem. This call should set the
5823 * appropriate time values.
5824 */
5825 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
5826 if (!health_sessiond) {
5827 PERROR("health_app_create error");
5828 retval = -1;
5829 goto exit_health_sessiond_cleanup;
5830 }
5831
5832 /* Create thread to clean up RCU hash tables */
5833 if (init_ht_cleanup_thread(&ht_cleanup_thread)) {
5834 retval = -1;
5835 goto exit_ht_cleanup;
5836 }
5837
5838 /* Create thread quit pipe */
5839 if (sessiond_init_thread_quit_pipe()) {
5840 retval = -1;
5841 goto exit_init_data;
5842 }
5843
5844 /* Check if daemon is UID = 0 */
5845 is_root = !getuid();
5846 if (is_root) {
5847 /* Create global run dir with root access */
5848
5849 kernel_channel_monitor_pipe = lttng_pipe_open(0);
5850 if (!kernel_channel_monitor_pipe) {
5851 ERR("Failed to create kernel consumer channel monitor pipe");
5852 retval = -1;
5853 goto exit_init_data;
5854 }
5855 kconsumer_data.channel_monitor_pipe =
5856 lttng_pipe_release_writefd(
5857 kernel_channel_monitor_pipe);
5858 if (kconsumer_data.channel_monitor_pipe < 0) {
5859 retval = -1;
5860 goto exit_init_data;
5861 }
5862 }
5863
5864 /* Set consumer initial state */
5865 kernel_consumerd_state = CONSUMER_STOPPED;
5866 ust_consumerd_state = CONSUMER_STOPPED;
5867
5868 ust32_channel_monitor_pipe = lttng_pipe_open(0);
5869 if (!ust32_channel_monitor_pipe) {
5870 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5871 retval = -1;
5872 goto exit_init_data;
5873 }
5874 ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5875 ust32_channel_monitor_pipe);
5876 if (ustconsumer32_data.channel_monitor_pipe < 0) {
5877 retval = -1;
5878 goto exit_init_data;
5879 }
5880
5881 /*
5882 * The rotation_thread_timer_queue structure is shared between the
5883 * sessiond timer thread and the rotation thread. The main thread keeps
5884 * its ownership and destroys it when both threads have been joined.
5885 */
5886 rotation_timer_queue = rotation_thread_timer_queue_create();
5887 if (!rotation_timer_queue) {
5888 retval = -1;
5889 goto exit_init_data;
5890 }
5891 timer_thread_ctx.rotation_thread_job_queue = rotation_timer_queue;
5892
5893 ust64_channel_monitor_pipe = lttng_pipe_open(0);
5894 if (!ust64_channel_monitor_pipe) {
5895 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5896 retval = -1;
5897 goto exit_init_data;
5898 }
5899 ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd(
5900 ust64_channel_monitor_pipe);
5901 if (ustconsumer64_data.channel_monitor_pipe < 0) {
5902 retval = -1;
5903 goto exit_init_data;
5904 }
5905
5906 /*
5907 * Init UST app hash table. Alloc hash table before this point since
5908 * cleanup() can get called after that point.
5909 */
5910 if (ust_app_ht_alloc()) {
5911 ERR("Failed to allocate UST app hash table");
5912 retval = -1;
5913 goto exit_init_data;
5914 }
5915
5916 /*
5917 * Initialize agent app hash table. We allocate the hash table here
5918 * since cleanup() can get called after this point.
5919 */
5920 if (agent_app_ht_alloc()) {
5921 ERR("Failed to allocate Agent app hash table");
5922 retval = -1;
5923 goto exit_init_data;
5924 }
5925
5926 /*
5927 * These actions must be executed as root. We do that *after* setting up
5928 * the sockets path because we MUST make the check for another daemon using
5929 * those paths *before* trying to set the kernel consumer sockets and init
5930 * kernel tracer.
5931 */
5932 if (is_root) {
5933 if (set_consumer_sockets(&kconsumer_data)) {
5934 retval = -1;
5935 goto exit_init_data;
5936 }
5937
5938 /* Setup kernel tracer */
5939 if (!config.no_kernel) {
5940 init_kernel_tracer();
5941 if (kernel_tracer_fd >= 0) {
5942 ret = syscall_init_table();
5943 if (ret < 0) {
5944 ERR("Unable to populate syscall table. "
5945 "Syscall tracing won't work "
5946 "for this session daemon.");
5947 }
5948 }
5949 }
5950
5951 /* Set ulimit for open files */
5952 set_ulimit();
5953 }
5954 /* init lttng_fd tracking must be done after set_ulimit. */
5955 lttng_fd_init();
5956
5957 if (set_consumer_sockets(&ustconsumer64_data)) {
5958 retval = -1;
5959 goto exit_init_data;
5960 }
5961
5962 if (set_consumer_sockets(&ustconsumer32_data)) {
5963 retval = -1;
5964 goto exit_init_data;
5965 }
5966
5967 /* Setup the needed unix socket */
5968 if (init_daemon_socket()) {
5969 retval = -1;
5970 goto exit_init_data;
5971 }
5972
5973 /* Set credentials to socket */
5974 if (is_root && set_permissions(config.rundir.value)) {
5975 retval = -1;
5976 goto exit_init_data;
5977 }
5978
5979 /* Get parent pid if -S, --sig-parent is specified. */
5980 if (config.sig_parent) {
5981 ppid = getppid();
5982 }
5983
5984 /* Setup the kernel pipe for waking up the kernel thread */
5985 if (is_root && !config.no_kernel) {
5986 if (utils_create_pipe_cloexec(kernel_poll_pipe)) {
5987 retval = -1;
5988 goto exit_init_data;
5989 }
5990 }
5991
5992 /* Setup the thread apps communication pipe. */
5993 if (utils_create_pipe_cloexec(apps_cmd_pipe)) {
5994 retval = -1;
5995 goto exit_init_data;
5996 }
5997
5998 /* Setup the thread apps notify communication pipe. */
5999 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) {
6000 retval = -1;
6001 goto exit_init_data;
6002 }
6003
6004 /* Initialize global buffer per UID and PID registry. */
6005 buffer_reg_init_uid_registry();
6006 buffer_reg_init_pid_registry();
6007
6008 /* Init UST command queue. */
6009 cds_wfcq_init(&ust_cmd_queue.head, &ust_cmd_queue.tail);
6010
6011 cmd_init();
6012
6013 /* Check for the application socket timeout env variable. */
6014 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
6015 if (env_app_timeout) {
6016 config.app_socket_timeout = atoi(env_app_timeout);
6017 } else {
6018 config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
6019 }
6020
6021 ret = write_pidfile();
6022 if (ret) {
6023 ERR("Error in write_pidfile");
6024 retval = -1;
6025 goto exit_init_data;
6026 }
6027
6028 /* Initialize communication library */
6029 lttcomm_init();
6030 /* Initialize TCP timeout values */
6031 lttcomm_inet_init();
6032
6033 if (load_session_init_data(&load_info) < 0) {
6034 retval = -1;
6035 goto exit_init_data;
6036 }
6037 load_info->path = config.load_session_path.value;
6038
6039 /* Create health-check thread. */
6040 ret = pthread_create(&health_thread, default_pthread_attr(),
6041 thread_manage_health, (void *) NULL);
6042 if (ret) {
6043 errno = ret;
6044 PERROR("pthread_create health");
6045 retval = -1;
6046 goto exit_health;
6047 }
6048
6049 /*
6050 * The rotation thread needs the notification thread to be ready before
6051 * creating the rotate_notification_channel, so we use this semaphore as
6052 * a rendez-vous point.
6053 */
6054 sem_init(&notification_thread_ready, 0, 0);
6055
6056 /* notification_thread_data acquires the pipes' read side. */
6057 notification_thread_handle = notification_thread_handle_create(
6058 ust32_channel_monitor_pipe,
6059 ust64_channel_monitor_pipe,
6060 kernel_channel_monitor_pipe,
6061 &notification_thread_ready);
6062 if (!notification_thread_handle) {
6063 retval = -1;
6064 ERR("Failed to create notification thread shared data");
6065 stop_threads();
6066 goto exit_notification;
6067 }
6068
6069 /* Create notification thread. */
6070 ret = pthread_create(&notification_thread, default_pthread_attr(),
6071 thread_notification, notification_thread_handle);
6072 if (ret) {
6073 errno = ret;
6074 PERROR("pthread_create notification");
6075 retval = -1;
6076 stop_threads();
6077 goto exit_notification;
6078 }
6079 notification_thread_launched = true;
6080
6081 /* Create timer thread. */
6082 ret = pthread_create(&timer_thread, default_pthread_attr(),
6083 timer_thread_func, &timer_thread_ctx);
6084 if (ret) {
6085 errno = ret;
6086 PERROR("pthread_create timer");
6087 retval = -1;
6088 stop_threads();
6089 goto exit_notification;
6090 }
6091 timer_thread_launched = true;
6092
6093 /* rotation_thread_data acquires the pipes' read side. */
6094 rotation_thread_handle = rotation_thread_handle_create(
6095 rotation_timer_queue,
6096 notification_thread_handle,
6097 &notification_thread_ready);
6098 if (!rotation_thread_handle) {
6099 retval = -1;
6100 ERR("Failed to create rotation thread shared data");
6101 stop_threads();
6102 goto exit_rotation;
6103 }
6104
6105 /* Create rotation thread. */
6106 ret = pthread_create(&rotation_thread, default_pthread_attr(),
6107 thread_rotation, rotation_thread_handle);
6108 if (ret) {
6109 errno = ret;
6110 PERROR("pthread_create rotation");
6111 retval = -1;
6112 stop_threads();
6113 goto exit_rotation;
6114 }
6115 rotation_thread_launched = true;
6116
6117 /* Create thread to manage the client socket */
6118 ret = pthread_create(&client_thread, default_pthread_attr(),
6119 thread_manage_clients, (void *) NULL);
6120 if (ret) {
6121 errno = ret;
6122 PERROR("pthread_create clients");
6123 retval = -1;
6124 stop_threads();
6125 goto exit_client;
6126 }
6127
6128 /* Create thread to dispatch registration */
6129 ret = pthread_create(&dispatch_thread, default_pthread_attr(),
6130 thread_dispatch_ust_registration, (void *) NULL);
6131 if (ret) {
6132 errno = ret;
6133 PERROR("pthread_create dispatch");
6134 retval = -1;
6135 stop_threads();
6136 goto exit_dispatch;
6137 }
6138
6139 /* Create thread to manage application registration. */
6140 ret = pthread_create(&reg_apps_thread, default_pthread_attr(),
6141 thread_registration_apps, (void *) NULL);
6142 if (ret) {
6143 errno = ret;
6144 PERROR("pthread_create registration");
6145 retval = -1;
6146 stop_threads();
6147 goto exit_reg_apps;
6148 }
6149
6150 /* Create thread to manage application socket */
6151 ret = pthread_create(&apps_thread, default_pthread_attr(),
6152 thread_manage_apps, (void *) NULL);
6153 if (ret) {
6154 errno = ret;
6155 PERROR("pthread_create apps");
6156 retval = -1;
6157 stop_threads();
6158 goto exit_apps;
6159 }
6160
6161 /* Create thread to manage application notify socket */
6162 ret = pthread_create(&apps_notify_thread, default_pthread_attr(),
6163 ust_thread_manage_notify, (void *) NULL);
6164 if (ret) {
6165 errno = ret;
6166 PERROR("pthread_create notify");
6167 retval = -1;
6168 stop_threads();
6169 goto exit_apps_notify;
6170 }
6171
6172 /* Create agent registration thread. */
6173 ret = pthread_create(&agent_reg_thread, default_pthread_attr(),
6174 agent_thread_manage_registration, (void *) NULL);
6175 if (ret) {
6176 errno = ret;
6177 PERROR("pthread_create agent");
6178 retval = -1;
6179 stop_threads();
6180 goto exit_agent_reg;
6181 }
6182
6183 /* Don't start this thread if kernel tracing is not requested nor root */
6184 if (is_root && !config.no_kernel) {
6185 /* Create kernel thread to manage kernel event */
6186 ret = pthread_create(&kernel_thread, default_pthread_attr(),
6187 thread_manage_kernel, (void *) NULL);
6188 if (ret) {
6189 errno = ret;
6190 PERROR("pthread_create kernel");
6191 retval = -1;
6192 stop_threads();
6193 goto exit_kernel;
6194 }
6195 }
6196
6197 /* Create session loading thread. */
6198 ret = pthread_create(&load_session_thread, default_pthread_attr(),
6199 thread_load_session, load_info);
6200 if (ret) {
6201 errno = ret;
6202 PERROR("pthread_create load_session_thread");
6203 retval = -1;
6204 stop_threads();
6205 goto exit_load_session;
6206 }
6207
6208 /*
6209 * This is where we start awaiting program completion (e.g. through
6210 * signal that asks threads to teardown).
6211 */
6212
6213 ret = pthread_join(load_session_thread, &status);
6214 if (ret) {
6215 errno = ret;
6216 PERROR("pthread_join load_session_thread");
6217 retval = -1;
6218 }
6219
6220 /* Initiate teardown once activity occurs on the quit pipe. */
6221 sessiond_wait_for_quit_pipe(-1U);
6222 destroy_all_sessions_and_wait();
6223 exit_load_session:
6224
6225 if (is_root && !config.no_kernel) {
6226 ret = pthread_join(kernel_thread, &status);
6227 if (ret) {
6228 errno = ret;
6229 PERROR("pthread_join");
6230 retval = -1;
6231 }
6232 }
6233 exit_kernel:
6234
6235 ret = pthread_join(agent_reg_thread, &status);
6236 if (ret) {
6237 errno = ret;
6238 PERROR("pthread_join agent");
6239 retval = -1;
6240 }
6241 exit_agent_reg:
6242
6243 ret = pthread_join(apps_notify_thread, &status);
6244 if (ret) {
6245 errno = ret;
6246 PERROR("pthread_join apps notify");
6247 retval = -1;
6248 }
6249 exit_apps_notify:
6250
6251 ret = pthread_join(apps_thread, &status);
6252 if (ret) {
6253 errno = ret;
6254 PERROR("pthread_join apps");
6255 retval = -1;
6256 }
6257 exit_apps:
6258
6259 ret = pthread_join(reg_apps_thread, &status);
6260 if (ret) {
6261 errno = ret;
6262 PERROR("pthread_join");
6263 retval = -1;
6264 }
6265 exit_reg_apps:
6266
6267 /*
6268 * Join dispatch thread after joining reg_apps_thread to ensure
6269 * we don't leak applications in the queue.
6270 */
6271 ret = pthread_join(dispatch_thread, &status);
6272 if (ret) {
6273 errno = ret;
6274 PERROR("pthread_join");
6275 retval = -1;
6276 }
6277 exit_dispatch:
6278
6279 ret = pthread_join(client_thread, &status);
6280 if (ret) {
6281 errno = ret;
6282 PERROR("pthread_join");
6283 retval = -1;
6284 }
6285
6286 exit_client:
6287 exit_rotation:
6288 exit_notification:
6289 sem_destroy(&notification_thread_ready);
6290 ret = pthread_join(health_thread, &status);
6291 if (ret) {
6292 errno = ret;
6293 PERROR("pthread_join health thread");
6294 retval = -1;
6295 }
6296
6297 exit_health:
6298 exit_init_data:
6299 /*
6300 * Wait for all pending call_rcu work to complete before tearing
6301 * down data structures. call_rcu worker may be trying to
6302 * perform lookups in those structures.
6303 */
6304 rcu_barrier();
6305 /*
6306 * sessiond_cleanup() is called when no other thread is running, except
6307 * the ht_cleanup thread, which is needed to destroy the hash tables.
6308 */
6309 rcu_thread_online();
6310 sessiond_cleanup();
6311
6312 /*
6313 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6314 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6315 * the queue is empty before shutting down the clean-up thread.
6316 */
6317 rcu_barrier();
6318
6319 /*
6320 * The teardown of the notification system is performed after the
6321 * session daemon's teardown in order to allow it to be notified
6322 * of the active session and channels at the moment of the teardown.
6323 */
6324 if (notification_thread_handle) {
6325 if (notification_thread_launched) {
6326 notification_thread_command_quit(
6327 notification_thread_handle);
6328 ret = pthread_join(notification_thread, &status);
6329 if (ret) {
6330 errno = ret;
6331 PERROR("pthread_join notification thread");
6332 retval = -1;
6333 }
6334 }
6335 notification_thread_handle_destroy(notification_thread_handle);
6336 }
6337
6338 if (rotation_thread_handle) {
6339 if (rotation_thread_launched) {
6340 ret = pthread_join(rotation_thread, &status);
6341 if (ret) {
6342 errno = ret;
6343 PERROR("pthread_join rotation thread");
6344 retval = -1;
6345 }
6346 }
6347 rotation_thread_handle_destroy(rotation_thread_handle);
6348 }
6349
6350 if (timer_thread_launched) {
6351 timer_exit();
6352 ret = pthread_join(timer_thread, &status);
6353 if (ret) {
6354 errno = ret;
6355 PERROR("pthread_join timer thread");
6356 retval = -1;
6357 }
6358 }
6359
6360 /*
6361 * After the rotation and timer thread have quit, we can safely destroy
6362 * the rotation_timer_queue.
6363 */
6364 rotation_thread_timer_queue_destroy(rotation_timer_queue);
6365
6366 rcu_thread_offline();
6367 rcu_unregister_thread();
6368
6369 ret = fini_ht_cleanup_thread(&ht_cleanup_thread);
6370 if (ret) {
6371 retval = -1;
6372 }
6373 lttng_pipe_destroy(ust32_channel_monitor_pipe);
6374 lttng_pipe_destroy(ust64_channel_monitor_pipe);
6375 lttng_pipe_destroy(kernel_channel_monitor_pipe);
6376 exit_ht_cleanup:
6377
6378 health_app_destroy(health_sessiond);
6379 exit_health_sessiond_cleanup:
6380 exit_create_run_as_worker_cleanup:
6381
6382 exit_options:
6383 sessiond_cleanup_lock_file();
6384 sessiond_cleanup_options();
6385
6386 exit_set_signal_handler:
6387 if (!retval) {
6388 exit(EXIT_SUCCESS);
6389 } else {
6390 exit(EXIT_FAILURE);
6391 }
6392 }
This page took 0.219356 seconds and 6 git commands to generate.