centralize sessiond config option handling
[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
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
55 #include "channel.h"
56 #include "cmd.h"
57 #include "consumer.h"
58 #include "context.h"
59 #include "event.h"
60 #include "kernel.h"
61 #include "kernel-consumer.h"
62 #include "modprobe.h"
63 #include "shm.h"
64 #include "ust-ctl.h"
65 #include "ust-consumer.h"
66 #include "utils.h"
67 #include "fd-limit.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
72 #include "save.h"
73 #include "load-session-thread.h"
74 #include "notification-thread.h"
75 #include "notification-thread-commands.h"
76 #include "syscall.h"
77 #include "agent.h"
78 #include "ht-cleanup.h"
79 #include "sessiond-config.h"
80
81 static const char *help_msg =
82 #ifdef LTTNG_EMBED_HELP
83 #include <lttng-sessiond.8.h>
84 #else
85 NULL
86 #endif
87 ;
88
89 const char *progname;
90 static pid_t ppid; /* Parent PID for --sig-parent option */
91 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
92 static int lockfile_fd = -1;
93
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal;
96
97 /*
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
100 */
101 static struct consumer_data kconsumer_data = {
102 .type = LTTNG_CONSUMER_KERNEL,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .channel_monitor_pipe = -1,
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
110 };
111 static struct consumer_data ustconsumer64_data = {
112 .type = LTTNG_CONSUMER64_UST,
113 .err_sock = -1,
114 .cmd_sock = -1,
115 .channel_monitor_pipe = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121 static struct consumer_data ustconsumer32_data = {
122 .type = LTTNG_CONSUMER32_UST,
123 .err_sock = -1,
124 .cmd_sock = -1,
125 .channel_monitor_pipe = -1,
126 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
127 .lock = PTHREAD_MUTEX_INITIALIZER,
128 .cond = PTHREAD_COND_INITIALIZER,
129 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
130 };
131
132 /* Command line options */
133 static const struct option long_options[] = {
134 { "client-sock", required_argument, 0, 'c' },
135 { "apps-sock", required_argument, 0, 'a' },
136 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
137 { "kconsumerd-err-sock", required_argument, 0, '\0' },
138 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
139 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
140 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
141 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
142 { "consumerd32-path", required_argument, 0, '\0' },
143 { "consumerd32-libdir", required_argument, 0, '\0' },
144 { "consumerd64-path", required_argument, 0, '\0' },
145 { "consumerd64-libdir", required_argument, 0, '\0' },
146 { "daemonize", no_argument, 0, 'd' },
147 { "background", no_argument, 0, 'b' },
148 { "sig-parent", no_argument, 0, 'S' },
149 { "help", no_argument, 0, 'h' },
150 { "group", required_argument, 0, 'g' },
151 { "version", no_argument, 0, 'V' },
152 { "quiet", no_argument, 0, 'q' },
153 { "verbose", no_argument, 0, 'v' },
154 { "verbose-consumer", no_argument, 0, '\0' },
155 { "no-kernel", no_argument, 0, '\0' },
156 { "pidfile", required_argument, 0, 'p' },
157 { "agent-tcp-port", required_argument, 0, '\0' },
158 { "config", required_argument, 0, 'f' },
159 { "load", required_argument, 0, 'l' },
160 { "kmod-probes", required_argument, 0, '\0' },
161 { "extra-kmod-probes", required_argument, 0, '\0' },
162 { NULL, 0, 0, 0 }
163 };
164
165 struct sessiond_config config;
166
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options[] = { "help", "version", "config" };
169
170 /* Shared between threads */
171 static int dispatch_thread_exit;
172
173 /* Sockets and FDs */
174 static int client_sock = -1;
175 static int apps_sock = -1;
176 int kernel_tracer_fd = -1;
177 static int kernel_poll_pipe[2] = { -1, -1 };
178
179 /*
180 * Quit pipe for all threads. This permits a single cancellation point
181 * for all threads when receiving an event on the pipe.
182 */
183 static int thread_quit_pipe[2] = { -1, -1 };
184
185 /*
186 * This pipe is used to inform the thread managing application communication
187 * that a command is queued and ready to be processed.
188 */
189 static int apps_cmd_pipe[2] = { -1, -1 };
190
191 int apps_cmd_notify_pipe[2] = { -1, -1 };
192
193 /* Pthread, Mutexes and Semaphores */
194 static pthread_t apps_thread;
195 static pthread_t apps_notify_thread;
196 static pthread_t reg_apps_thread;
197 static pthread_t client_thread;
198 static pthread_t kernel_thread;
199 static pthread_t dispatch_thread;
200 static pthread_t health_thread;
201 static pthread_t ht_cleanup_thread;
202 static pthread_t agent_reg_thread;
203 static pthread_t load_session_thread;
204 static pthread_t notification_thread;
205
206 /*
207 * UST registration command queue. This queue is tied with a futex and uses a N
208 * wakers / 1 waiter implemented and detailed in futex.c/.h
209 *
210 * The thread_registration_apps and thread_dispatch_ust_registration uses this
211 * queue along with the wait/wake scheme. The thread_manage_apps receives down
212 * the line new application socket and monitors it for any I/O error or clean
213 * close that triggers an unregistration of the application.
214 */
215 static struct ust_cmd_queue ust_cmd_queue;
216
217 /*
218 * Pointer initialized before thread creation.
219 *
220 * This points to the tracing session list containing the session count and a
221 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
222 * MUST NOT be taken if you call a public function in session.c.
223 *
224 * The lock is nested inside the structure: session_list_ptr->lock. Please use
225 * session_lock_list and session_unlock_list for lock acquisition.
226 */
227 static struct ltt_session_list *session_list_ptr;
228
229 int ust_consumerd64_fd = -1;
230 int ust_consumerd32_fd = -1;
231
232 static const char *module_proc_lttng = "/proc/lttng";
233
234 /*
235 * Consumer daemon state which is changed when spawning it, killing it or in
236 * case of a fatal error.
237 */
238 enum consumerd_state {
239 CONSUMER_STARTED = 1,
240 CONSUMER_STOPPED = 2,
241 CONSUMER_ERROR = 3,
242 };
243
244 /*
245 * This consumer daemon state is used to validate if a client command will be
246 * able to reach the consumer. If not, the client is informed. For instance,
247 * doing a "lttng start" when the consumer state is set to ERROR will return an
248 * error to the client.
249 *
250 * The following example shows a possible race condition of this scheme:
251 *
252 * consumer thread error happens
253 * client cmd arrives
254 * client cmd checks state -> still OK
255 * consumer thread exit, sets error
256 * client cmd try to talk to consumer
257 * ...
258 *
259 * However, since the consumer is a different daemon, we have no way of making
260 * sure the command will reach it safely even with this state flag. This is why
261 * we consider that up to the state validation during command processing, the
262 * command is safe. After that, we can not guarantee the correctness of the
263 * client request vis-a-vis the consumer.
264 */
265 static enum consumerd_state ust_consumerd_state;
266 static enum consumerd_state kernel_consumerd_state;
267
268 /* Set in main() with the current page size. */
269 long page_size;
270
271 /* Application health monitoring */
272 struct health_app *health_sessiond;
273
274 /* Am I root or not. */
275 int is_root; /* Set to 1 if the daemon is running as root */
276
277 const char * const config_section_name = "sessiond";
278
279 /* Load session thread information to operate. */
280 struct load_session_thread_data *load_info;
281
282 /* Notification thread handle. */
283 struct notification_thread_handle *notification_thread_handle;
284
285 /* Global hash tables */
286 struct lttng_ht *agent_apps_ht_by_sock = NULL;
287
288 /*
289 * Whether sessiond is ready for commands/notification channel/health check
290 * requests.
291 * NR_LTTNG_SESSIOND_READY must match the number of calls to
292 * sessiond_notify_ready().
293 */
294 #define NR_LTTNG_SESSIOND_READY 4
295 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
296
297 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
298 {
299 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
300 }
301
302 /* Notify parents that we are ready for cmd and health check */
303 LTTNG_HIDDEN
304 void sessiond_notify_ready(void)
305 {
306 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
307 /*
308 * Notify parent pid that we are ready to accept command
309 * for client side. This ppid is the one from the
310 * external process that spawned us.
311 */
312 if (config.sig_parent) {
313 kill(ppid, SIGUSR1);
314 }
315
316 /*
317 * Notify the parent of the fork() process that we are
318 * ready.
319 */
320 if (config.daemonize || config.background) {
321 kill(child_ppid, SIGUSR1);
322 }
323 }
324 }
325
326 static
327 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
328 int *a_pipe)
329 {
330 int ret;
331
332 assert(events);
333
334 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
335 if (ret < 0) {
336 goto error;
337 }
338
339 /* Add quit pipe */
340 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
341 if (ret < 0) {
342 goto error;
343 }
344
345 return 0;
346
347 error:
348 return ret;
349 }
350
351 /*
352 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
353 */
354 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
355 {
356 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
357 }
358
359 /*
360 * Init thread quit pipe.
361 *
362 * Return -1 on error or 0 if all pipes are created.
363 */
364 static int __init_thread_quit_pipe(int *a_pipe)
365 {
366 int ret, i;
367
368 ret = pipe(a_pipe);
369 if (ret < 0) {
370 PERROR("thread quit pipe");
371 goto error;
372 }
373
374 for (i = 0; i < 2; i++) {
375 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
376 if (ret < 0) {
377 PERROR("fcntl");
378 goto error;
379 }
380 }
381
382 error:
383 return ret;
384 }
385
386 static int init_thread_quit_pipe(void)
387 {
388 return __init_thread_quit_pipe(thread_quit_pipe);
389 }
390
391 /*
392 * Stop all threads by closing the thread quit pipe.
393 */
394 static void stop_threads(void)
395 {
396 int ret;
397
398 /* Stopping all threads */
399 DBG("Terminating all threads");
400 ret = notify_thread_pipe(thread_quit_pipe[1]);
401 if (ret < 0) {
402 ERR("write error on thread quit pipe");
403 }
404
405 /* Dispatch thread */
406 CMM_STORE_SHARED(dispatch_thread_exit, 1);
407 futex_nto1_wake(&ust_cmd_queue.futex);
408 }
409
410 /*
411 * Close every consumer sockets.
412 */
413 static void close_consumer_sockets(void)
414 {
415 int ret;
416
417 if (kconsumer_data.err_sock >= 0) {
418 ret = close(kconsumer_data.err_sock);
419 if (ret < 0) {
420 PERROR("kernel consumer err_sock close");
421 }
422 }
423 if (ustconsumer32_data.err_sock >= 0) {
424 ret = close(ustconsumer32_data.err_sock);
425 if (ret < 0) {
426 PERROR("UST consumerd32 err_sock close");
427 }
428 }
429 if (ustconsumer64_data.err_sock >= 0) {
430 ret = close(ustconsumer64_data.err_sock);
431 if (ret < 0) {
432 PERROR("UST consumerd64 err_sock close");
433 }
434 }
435 if (kconsumer_data.cmd_sock >= 0) {
436 ret = close(kconsumer_data.cmd_sock);
437 if (ret < 0) {
438 PERROR("kernel consumer cmd_sock close");
439 }
440 }
441 if (ustconsumer32_data.cmd_sock >= 0) {
442 ret = close(ustconsumer32_data.cmd_sock);
443 if (ret < 0) {
444 PERROR("UST consumerd32 cmd_sock close");
445 }
446 }
447 if (ustconsumer64_data.cmd_sock >= 0) {
448 ret = close(ustconsumer64_data.cmd_sock);
449 if (ret < 0) {
450 PERROR("UST consumerd64 cmd_sock close");
451 }
452 }
453 if (kconsumer_data.channel_monitor_pipe >= 0) {
454 ret = close(kconsumer_data.channel_monitor_pipe);
455 if (ret < 0) {
456 PERROR("kernel consumer channel monitor pipe close");
457 }
458 }
459 if (ustconsumer32_data.channel_monitor_pipe >= 0) {
460 ret = close(ustconsumer32_data.channel_monitor_pipe);
461 if (ret < 0) {
462 PERROR("UST consumerd32 channel monitor pipe close");
463 }
464 }
465 if (ustconsumer64_data.channel_monitor_pipe >= 0) {
466 ret = close(ustconsumer64_data.channel_monitor_pipe);
467 if (ret < 0) {
468 PERROR("UST consumerd64 channel monitor pipe close");
469 }
470 }
471 }
472
473 /*
474 * Wait on consumer process termination.
475 *
476 * Need to be called with the consumer data lock held or from a context
477 * ensuring no concurrent access to data (e.g: cleanup).
478 */
479 static void wait_consumer(struct consumer_data *consumer_data)
480 {
481 pid_t ret;
482 int status;
483
484 if (consumer_data->pid <= 0) {
485 return;
486 }
487
488 DBG("Waiting for complete teardown of consumerd (PID: %d)",
489 consumer_data->pid);
490 ret = waitpid(consumer_data->pid, &status, 0);
491 if (ret == -1) {
492 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
493 } else if (!WIFEXITED(status)) {
494 ERR("consumerd termination with error: %d",
495 WEXITSTATUS(ret));
496 }
497 consumer_data->pid = 0;
498 }
499
500 /*
501 * Cleanup the session daemon's data structures.
502 */
503 static void sessiond_cleanup(void)
504 {
505 int ret;
506 struct ltt_session *sess, *stmp;
507
508 DBG("Cleanup sessiond");
509
510 /*
511 * Close the thread quit pipe. It has already done its job,
512 * since we are now called.
513 */
514 utils_close_pipe(thread_quit_pipe);
515
516 /*
517 * If config.pid_file_path.value is undefined, the default file will be
518 * wiped when removing the rundir.
519 */
520 if (config.pid_file_path.value) {
521 ret = remove(config.pid_file_path.value);
522 if (ret < 0) {
523 PERROR("remove pidfile %s", config.pid_file_path.value);
524 }
525 }
526
527 DBG("Removing sessiond and consumerd content of directory %s",
528 config.rundir.value);
529
530 /* sessiond */
531 DBG("Removing %s", config.pid_file_path.value);
532 (void) unlink(config.pid_file_path.value);
533
534 DBG("Removing %s", config.agent_port_file_path.value);
535 (void) unlink(config.agent_port_file_path.value);
536
537 /* kconsumerd */
538 DBG("Removing %s", kconsumer_data.err_unix_sock_path);
539 (void) unlink(kconsumer_data.err_unix_sock_path);
540
541 DBG("Removing directory %s", config.kconsumerd_path.value);
542 (void) rmdir(config.kconsumerd_path.value);
543
544 /* ust consumerd 32 */
545 DBG("Removing %s", config.consumerd32_err_unix_sock_path.value);
546 (void) unlink(config.consumerd32_err_unix_sock_path.value);
547
548 DBG("Removing directory %s", config.consumerd32_path.value);
549 (void) rmdir(config.consumerd32_path.value);
550
551 /* ust consumerd 64 */
552 DBG("Removing %s", config.consumerd64_err_unix_sock_path.value);
553 (void) unlink(config.consumerd64_err_unix_sock_path.value);
554
555 DBG("Removing directory %s", config.consumerd64_path.value);
556 (void) rmdir(config.consumerd64_path.value);
557
558 DBG("Cleaning up all sessions");
559
560 /* Destroy session list mutex */
561 if (session_list_ptr != NULL) {
562 pthread_mutex_destroy(&session_list_ptr->lock);
563
564 /* Cleanup ALL session */
565 cds_list_for_each_entry_safe(sess, stmp,
566 &session_list_ptr->head, list) {
567 cmd_destroy_session(sess, kernel_poll_pipe[1]);
568 }
569 }
570
571 wait_consumer(&kconsumer_data);
572 wait_consumer(&ustconsumer64_data);
573 wait_consumer(&ustconsumer32_data);
574
575 DBG("Cleaning up all agent apps");
576 agent_app_ht_clean();
577
578 DBG("Closing all UST sockets");
579 ust_app_clean_list();
580 buffer_reg_destroy_registries();
581
582 if (is_root && !config.no_kernel) {
583 DBG2("Closing kernel fd");
584 if (kernel_tracer_fd >= 0) {
585 ret = close(kernel_tracer_fd);
586 if (ret) {
587 PERROR("close");
588 }
589 }
590 DBG("Unloading kernel modules");
591 modprobe_remove_lttng_all();
592 free(syscall_table);
593 }
594
595 close_consumer_sockets();
596
597 if (load_info) {
598 load_session_destroy_data(load_info);
599 free(load_info);
600 }
601
602 /*
603 * Cleanup lock file by deleting it and finaly closing it which will
604 * release the file system lock.
605 */
606 if (lockfile_fd >= 0) {
607 ret = remove(config.lock_file_path.value);
608 if (ret < 0) {
609 PERROR("remove lock file");
610 }
611 ret = close(lockfile_fd);
612 if (ret < 0) {
613 PERROR("close lock file");
614 }
615 }
616
617 /*
618 * We do NOT rmdir rundir because there are other processes
619 * using it, for instance lttng-relayd, which can start in
620 * parallel with this teardown.
621 */
622 }
623
624 /*
625 * Cleanup the daemon's option data structures.
626 */
627 static void sessiond_cleanup_options(void)
628 {
629 DBG("Cleaning up options");
630
631 sessiond_config_fini(&config);
632
633 run_as_destroy_worker();
634 }
635
636 /*
637 * Send data on a unix socket using the liblttsessiondcomm API.
638 *
639 * Return lttcomm error code.
640 */
641 static int send_unix_sock(int sock, void *buf, size_t len)
642 {
643 /* Check valid length */
644 if (len == 0) {
645 return -1;
646 }
647
648 return lttcomm_send_unix_sock(sock, buf, len);
649 }
650
651 /*
652 * Free memory of a command context structure.
653 */
654 static void clean_command_ctx(struct command_ctx **cmd_ctx)
655 {
656 DBG("Clean command context structure");
657 if (*cmd_ctx) {
658 if ((*cmd_ctx)->llm) {
659 free((*cmd_ctx)->llm);
660 }
661 if ((*cmd_ctx)->lsm) {
662 free((*cmd_ctx)->lsm);
663 }
664 free(*cmd_ctx);
665 *cmd_ctx = NULL;
666 }
667 }
668
669 /*
670 * Notify UST applications using the shm mmap futex.
671 */
672 static int notify_ust_apps(int active)
673 {
674 char *wait_shm_mmap;
675
676 DBG("Notifying applications of session daemon state: %d", active);
677
678 /* See shm.c for this call implying mmap, shm and futex calls */
679 wait_shm_mmap = shm_ust_get_mmap(config.wait_shm_path.value, is_root);
680 if (wait_shm_mmap == NULL) {
681 goto error;
682 }
683
684 /* Wake waiting process */
685 futex_wait_update((int32_t *) wait_shm_mmap, active);
686
687 /* Apps notified successfully */
688 return 0;
689
690 error:
691 return -1;
692 }
693
694 /*
695 * Setup the outgoing data buffer for the response (llm) by allocating the
696 * right amount of memory and copying the original information from the lsm
697 * structure.
698 *
699 * Return 0 on success, negative value on error.
700 */
701 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
702 const void *payload_buf, size_t payload_len,
703 const void *cmd_header_buf, size_t cmd_header_len)
704 {
705 int ret = 0;
706 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
707 const size_t cmd_header_offset = header_len;
708 const size_t payload_offset = cmd_header_offset + cmd_header_len;
709 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
710
711 cmd_ctx->llm = zmalloc(total_msg_size);
712
713 if (cmd_ctx->llm == NULL) {
714 PERROR("zmalloc");
715 ret = -ENOMEM;
716 goto end;
717 }
718
719 /* Copy common data */
720 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
721 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
722 cmd_ctx->llm->cmd_header_size = cmd_header_len;
723 cmd_ctx->llm->data_size = payload_len;
724 cmd_ctx->lttng_msg_size = total_msg_size;
725
726 /* Copy command header */
727 if (cmd_header_len) {
728 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
729 cmd_header_len);
730 }
731
732 /* Copy payload */
733 if (payload_len) {
734 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
735 payload_len);
736 }
737
738 end:
739 return ret;
740 }
741
742 /*
743 * Version of setup_lttng_msg() without command header.
744 */
745 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
746 void *payload_buf, size_t payload_len)
747 {
748 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
749 }
750 /*
751 * Update the kernel poll set of all channel fd available over all tracing
752 * session. Add the wakeup pipe at the end of the set.
753 */
754 static int update_kernel_poll(struct lttng_poll_event *events)
755 {
756 int ret;
757 struct ltt_session *session;
758 struct ltt_kernel_channel *channel;
759
760 DBG("Updating kernel poll set");
761
762 session_lock_list();
763 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
764 session_lock(session);
765 if (session->kernel_session == NULL) {
766 session_unlock(session);
767 continue;
768 }
769
770 cds_list_for_each_entry(channel,
771 &session->kernel_session->channel_list.head, list) {
772 /* Add channel fd to the kernel poll set */
773 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
774 if (ret < 0) {
775 session_unlock(session);
776 goto error;
777 }
778 DBG("Channel fd %d added to kernel set", channel->fd);
779 }
780 session_unlock(session);
781 }
782 session_unlock_list();
783
784 return 0;
785
786 error:
787 session_unlock_list();
788 return -1;
789 }
790
791 /*
792 * Find the channel fd from 'fd' over all tracing session. When found, check
793 * for new channel stream and send those stream fds to the kernel consumer.
794 *
795 * Useful for CPU hotplug feature.
796 */
797 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
798 {
799 int ret = 0;
800 struct ltt_session *session;
801 struct ltt_kernel_session *ksess;
802 struct ltt_kernel_channel *channel;
803
804 DBG("Updating kernel streams for channel fd %d", fd);
805
806 session_lock_list();
807 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
808 session_lock(session);
809 if (session->kernel_session == NULL) {
810 session_unlock(session);
811 continue;
812 }
813 ksess = session->kernel_session;
814
815 cds_list_for_each_entry(channel,
816 &ksess->channel_list.head, list) {
817 struct lttng_ht_iter iter;
818 struct consumer_socket *socket;
819
820 if (channel->fd != fd) {
821 continue;
822 }
823 DBG("Channel found, updating kernel streams");
824 ret = kernel_open_channel_stream(channel);
825 if (ret < 0) {
826 goto error;
827 }
828 /* Update the stream global counter */
829 ksess->stream_count_global += ret;
830
831 /*
832 * Have we already sent fds to the consumer? If yes, it
833 * means that tracing is started so it is safe to send
834 * our updated stream fds.
835 */
836 if (ksess->consumer_fds_sent != 1
837 || ksess->consumer == NULL) {
838 ret = -1;
839 goto error;
840 }
841
842 rcu_read_lock();
843 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
844 &iter.iter, socket, node.node) {
845 pthread_mutex_lock(socket->lock);
846 ret = kernel_consumer_send_channel_stream(socket,
847 channel, ksess,
848 session->output_traces ? 1 : 0);
849 pthread_mutex_unlock(socket->lock);
850 if (ret < 0) {
851 rcu_read_unlock();
852 goto error;
853 }
854 }
855 rcu_read_unlock();
856 }
857 session_unlock(session);
858 }
859 session_unlock_list();
860 return ret;
861
862 error:
863 session_unlock(session);
864 session_unlock_list();
865 return ret;
866 }
867
868 /*
869 * For each tracing session, update newly registered apps. The session list
870 * lock MUST be acquired before calling this.
871 */
872 static void update_ust_app(int app_sock)
873 {
874 struct ltt_session *sess, *stmp;
875
876 /* Consumer is in an ERROR state. Stop any application update. */
877 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
878 /* Stop the update process since the consumer is dead. */
879 return;
880 }
881
882 /* For all tracing session(s) */
883 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
884 struct ust_app *app;
885
886 session_lock(sess);
887 if (!sess->ust_session) {
888 goto unlock_session;
889 }
890
891 rcu_read_lock();
892 assert(app_sock >= 0);
893 app = ust_app_find_by_sock(app_sock);
894 if (app == NULL) {
895 /*
896 * Application can be unregistered before so
897 * this is possible hence simply stopping the
898 * update.
899 */
900 DBG3("UST app update failed to find app sock %d",
901 app_sock);
902 goto unlock_rcu;
903 }
904 ust_app_global_update(sess->ust_session, app);
905 unlock_rcu:
906 rcu_read_unlock();
907 unlock_session:
908 session_unlock(sess);
909 }
910 }
911
912 /*
913 * This thread manage event coming from the kernel.
914 *
915 * Features supported in this thread:
916 * -) CPU Hotplug
917 */
918 static void *thread_manage_kernel(void *data)
919 {
920 int ret, i, pollfd, update_poll_flag = 1, err = -1;
921 uint32_t revents, nb_fd;
922 char tmp;
923 struct lttng_poll_event events;
924
925 DBG("[thread] Thread manage kernel started");
926
927 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
928
929 /*
930 * This first step of the while is to clean this structure which could free
931 * non NULL pointers so initialize it before the loop.
932 */
933 lttng_poll_init(&events);
934
935 if (testpoint(sessiond_thread_manage_kernel)) {
936 goto error_testpoint;
937 }
938
939 health_code_update();
940
941 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
942 goto error_testpoint;
943 }
944
945 while (1) {
946 health_code_update();
947
948 if (update_poll_flag == 1) {
949 /* Clean events object. We are about to populate it again. */
950 lttng_poll_clean(&events);
951
952 ret = sessiond_set_thread_pollset(&events, 2);
953 if (ret < 0) {
954 goto error_poll_create;
955 }
956
957 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
958 if (ret < 0) {
959 goto error;
960 }
961
962 /* This will add the available kernel channel if any. */
963 ret = update_kernel_poll(&events);
964 if (ret < 0) {
965 goto error;
966 }
967 update_poll_flag = 0;
968 }
969
970 DBG("Thread kernel polling");
971
972 /* Poll infinite value of time */
973 restart:
974 health_poll_entry();
975 ret = lttng_poll_wait(&events, -1);
976 DBG("Thread kernel return from poll on %d fds",
977 LTTNG_POLL_GETNB(&events));
978 health_poll_exit();
979 if (ret < 0) {
980 /*
981 * Restart interrupted system call.
982 */
983 if (errno == EINTR) {
984 goto restart;
985 }
986 goto error;
987 } else if (ret == 0) {
988 /* Should not happen since timeout is infinite */
989 ERR("Return value of poll is 0 with an infinite timeout.\n"
990 "This should not have happened! Continuing...");
991 continue;
992 }
993
994 nb_fd = ret;
995
996 for (i = 0; i < nb_fd; i++) {
997 /* Fetch once the poll data */
998 revents = LTTNG_POLL_GETEV(&events, i);
999 pollfd = LTTNG_POLL_GETFD(&events, i);
1000
1001 health_code_update();
1002
1003 if (!revents) {
1004 /* No activity for this FD (poll implementation). */
1005 continue;
1006 }
1007
1008 /* Thread quit pipe has been closed. Killing thread. */
1009 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1010 if (ret) {
1011 err = 0;
1012 goto exit;
1013 }
1014
1015 /* Check for data on kernel pipe */
1016 if (revents & LPOLLIN) {
1017 if (pollfd == kernel_poll_pipe[0]) {
1018 (void) lttng_read(kernel_poll_pipe[0],
1019 &tmp, 1);
1020 /*
1021 * Ret value is useless here, if this pipe gets any actions an
1022 * update is required anyway.
1023 */
1024 update_poll_flag = 1;
1025 continue;
1026 } else {
1027 /*
1028 * New CPU detected by the kernel. Adding kernel stream to
1029 * kernel session and updating the kernel consumer
1030 */
1031 ret = update_kernel_stream(&kconsumer_data, pollfd);
1032 if (ret < 0) {
1033 continue;
1034 }
1035 break;
1036 }
1037 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1038 update_poll_flag = 1;
1039 continue;
1040 } else {
1041 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1042 goto error;
1043 }
1044 }
1045 }
1046
1047 exit:
1048 error:
1049 lttng_poll_clean(&events);
1050 error_poll_create:
1051 error_testpoint:
1052 utils_close_pipe(kernel_poll_pipe);
1053 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1054 if (err) {
1055 health_error();
1056 ERR("Health error occurred in %s", __func__);
1057 WARN("Kernel thread died unexpectedly. "
1058 "Kernel tracing can continue but CPU hotplug is disabled.");
1059 }
1060 health_unregister(health_sessiond);
1061 DBG("Kernel thread dying");
1062 return NULL;
1063 }
1064
1065 /*
1066 * Signal pthread condition of the consumer data that the thread.
1067 */
1068 static void signal_consumer_condition(struct consumer_data *data, int state)
1069 {
1070 pthread_mutex_lock(&data->cond_mutex);
1071
1072 /*
1073 * The state is set before signaling. It can be any value, it's the waiter
1074 * job to correctly interpret this condition variable associated to the
1075 * consumer pthread_cond.
1076 *
1077 * A value of 0 means that the corresponding thread of the consumer data
1078 * was not started. 1 indicates that the thread has started and is ready
1079 * for action. A negative value means that there was an error during the
1080 * thread bootstrap.
1081 */
1082 data->consumer_thread_is_ready = state;
1083 (void) pthread_cond_signal(&data->cond);
1084
1085 pthread_mutex_unlock(&data->cond_mutex);
1086 }
1087
1088 /*
1089 * This thread manage the consumer error sent back to the session daemon.
1090 */
1091 static void *thread_manage_consumer(void *data)
1092 {
1093 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1094 uint32_t revents, nb_fd;
1095 enum lttcomm_return_code code;
1096 struct lttng_poll_event events;
1097 struct consumer_data *consumer_data = data;
1098 struct consumer_socket *cmd_socket_wrapper = NULL;
1099
1100 DBG("[thread] Manage consumer started");
1101
1102 rcu_register_thread();
1103 rcu_thread_online();
1104
1105 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1106
1107 health_code_update();
1108
1109 /*
1110 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1111 * metadata_sock. Nothing more will be added to this poll set.
1112 */
1113 ret = sessiond_set_thread_pollset(&events, 3);
1114 if (ret < 0) {
1115 goto error_poll;
1116 }
1117
1118 /*
1119 * The error socket here is already in a listening state which was done
1120 * just before spawning this thread to avoid a race between the consumer
1121 * daemon exec trying to connect and the listen() call.
1122 */
1123 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1124 if (ret < 0) {
1125 goto error;
1126 }
1127
1128 health_code_update();
1129
1130 /* Infinite blocking call, waiting for transmission */
1131 restart:
1132 health_poll_entry();
1133
1134 if (testpoint(sessiond_thread_manage_consumer)) {
1135 goto error;
1136 }
1137
1138 ret = lttng_poll_wait(&events, -1);
1139 health_poll_exit();
1140 if (ret < 0) {
1141 /*
1142 * Restart interrupted system call.
1143 */
1144 if (errno == EINTR) {
1145 goto restart;
1146 }
1147 goto error;
1148 }
1149
1150 nb_fd = ret;
1151
1152 for (i = 0; i < nb_fd; i++) {
1153 /* Fetch once the poll data */
1154 revents = LTTNG_POLL_GETEV(&events, i);
1155 pollfd = LTTNG_POLL_GETFD(&events, i);
1156
1157 health_code_update();
1158
1159 if (!revents) {
1160 /* No activity for this FD (poll implementation). */
1161 continue;
1162 }
1163
1164 /* Thread quit pipe has been closed. Killing thread. */
1165 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1166 if (ret) {
1167 err = 0;
1168 goto exit;
1169 }
1170
1171 /* Event on the registration socket */
1172 if (pollfd == consumer_data->err_sock) {
1173 if (revents & LPOLLIN) {
1174 continue;
1175 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1176 ERR("consumer err socket poll error");
1177 goto error;
1178 } else {
1179 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1180 goto error;
1181 }
1182 }
1183 }
1184
1185 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1186 if (sock < 0) {
1187 goto error;
1188 }
1189
1190 /*
1191 * Set the CLOEXEC flag. Return code is useless because either way, the
1192 * show must go on.
1193 */
1194 (void) utils_set_fd_cloexec(sock);
1195
1196 health_code_update();
1197
1198 DBG2("Receiving code from consumer err_sock");
1199
1200 /* Getting status code from kconsumerd */
1201 ret = lttcomm_recv_unix_sock(sock, &code,
1202 sizeof(enum lttcomm_return_code));
1203 if (ret <= 0) {
1204 goto error;
1205 }
1206
1207 health_code_update();
1208 if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1209 ERR("consumer error when waiting for SOCK_READY : %s",
1210 lttcomm_get_readable_code(-code));
1211 goto error;
1212 }
1213
1214 /* Connect both command and metadata sockets. */
1215 consumer_data->cmd_sock =
1216 lttcomm_connect_unix_sock(
1217 consumer_data->cmd_unix_sock_path);
1218 consumer_data->metadata_fd =
1219 lttcomm_connect_unix_sock(
1220 consumer_data->cmd_unix_sock_path);
1221 if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) {
1222 PERROR("consumer connect cmd socket");
1223 /* On error, signal condition and quit. */
1224 signal_consumer_condition(consumer_data, -1);
1225 goto error;
1226 }
1227
1228 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1229
1230 /* Create metadata socket lock. */
1231 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1232 if (consumer_data->metadata_sock.lock == NULL) {
1233 PERROR("zmalloc pthread mutex");
1234 goto error;
1235 }
1236 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1237
1238 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1239 DBG("Consumer metadata socket ready (fd: %d)",
1240 consumer_data->metadata_fd);
1241
1242 /*
1243 * Remove the consumerd error sock since we've established a connection.
1244 */
1245 ret = lttng_poll_del(&events, consumer_data->err_sock);
1246 if (ret < 0) {
1247 goto error;
1248 }
1249
1250 /* Add new accepted error socket. */
1251 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1252 if (ret < 0) {
1253 goto error;
1254 }
1255
1256 /* Add metadata socket that is successfully connected. */
1257 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1258 LPOLLIN | LPOLLRDHUP);
1259 if (ret < 0) {
1260 goto error;
1261 }
1262
1263 health_code_update();
1264
1265 /*
1266 * Transfer the write-end of the channel monitoring pipe to the
1267 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1268 */
1269 cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock);
1270 if (!cmd_socket_wrapper) {
1271 goto error;
1272 }
1273
1274 ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper,
1275 consumer_data->channel_monitor_pipe);
1276 if (ret) {
1277 goto error;
1278 }
1279 /* Discard the socket wrapper as it is no longer needed. */
1280 consumer_destroy_socket(cmd_socket_wrapper);
1281 cmd_socket_wrapper = NULL;
1282
1283 /* The thread is completely initialized, signal that it is ready. */
1284 signal_consumer_condition(consumer_data, 1);
1285
1286 /* Infinite blocking call, waiting for transmission */
1287 restart_poll:
1288 while (1) {
1289 health_code_update();
1290
1291 /* Exit the thread because the thread quit pipe has been triggered. */
1292 if (should_quit) {
1293 /* Not a health error. */
1294 err = 0;
1295 goto exit;
1296 }
1297
1298 health_poll_entry();
1299 ret = lttng_poll_wait(&events, -1);
1300 health_poll_exit();
1301 if (ret < 0) {
1302 /*
1303 * Restart interrupted system call.
1304 */
1305 if (errno == EINTR) {
1306 goto restart_poll;
1307 }
1308 goto error;
1309 }
1310
1311 nb_fd = ret;
1312
1313 for (i = 0; i < nb_fd; i++) {
1314 /* Fetch once the poll data */
1315 revents = LTTNG_POLL_GETEV(&events, i);
1316 pollfd = LTTNG_POLL_GETFD(&events, i);
1317
1318 health_code_update();
1319
1320 if (!revents) {
1321 /* No activity for this FD (poll implementation). */
1322 continue;
1323 }
1324
1325 /*
1326 * Thread quit pipe has been triggered, flag that we should stop
1327 * but continue the current loop to handle potential data from
1328 * consumer.
1329 */
1330 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1331
1332 if (pollfd == sock) {
1333 /* Event on the consumerd socket */
1334 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1335 && !(revents & LPOLLIN)) {
1336 ERR("consumer err socket second poll error");
1337 goto error;
1338 }
1339 health_code_update();
1340 /* Wait for any kconsumerd error */
1341 ret = lttcomm_recv_unix_sock(sock, &code,
1342 sizeof(enum lttcomm_return_code));
1343 if (ret <= 0) {
1344 ERR("consumer closed the command socket");
1345 goto error;
1346 }
1347
1348 ERR("consumer return code : %s",
1349 lttcomm_get_readable_code(-code));
1350
1351 goto exit;
1352 } else if (pollfd == consumer_data->metadata_fd) {
1353 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1354 && !(revents & LPOLLIN)) {
1355 ERR("consumer err metadata socket second poll error");
1356 goto error;
1357 }
1358 /* UST metadata requests */
1359 ret = ust_consumer_metadata_request(
1360 &consumer_data->metadata_sock);
1361 if (ret < 0) {
1362 ERR("Handling metadata request");
1363 goto error;
1364 }
1365 }
1366 /* No need for an else branch all FDs are tested prior. */
1367 }
1368 health_code_update();
1369 }
1370
1371 exit:
1372 error:
1373 /*
1374 * We lock here because we are about to close the sockets and some other
1375 * thread might be using them so get exclusive access which will abort all
1376 * other consumer command by other threads.
1377 */
1378 pthread_mutex_lock(&consumer_data->lock);
1379
1380 /* Immediately set the consumerd state to stopped */
1381 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1382 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1383 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1384 consumer_data->type == LTTNG_CONSUMER32_UST) {
1385 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1386 } else {
1387 /* Code flow error... */
1388 assert(0);
1389 }
1390
1391 if (consumer_data->err_sock >= 0) {
1392 ret = close(consumer_data->err_sock);
1393 if (ret) {
1394 PERROR("close");
1395 }
1396 consumer_data->err_sock = -1;
1397 }
1398 if (consumer_data->cmd_sock >= 0) {
1399 ret = close(consumer_data->cmd_sock);
1400 if (ret) {
1401 PERROR("close");
1402 }
1403 consumer_data->cmd_sock = -1;
1404 }
1405 if (consumer_data->metadata_sock.fd_ptr &&
1406 *consumer_data->metadata_sock.fd_ptr >= 0) {
1407 ret = close(*consumer_data->metadata_sock.fd_ptr);
1408 if (ret) {
1409 PERROR("close");
1410 }
1411 }
1412 if (sock >= 0) {
1413 ret = close(sock);
1414 if (ret) {
1415 PERROR("close");
1416 }
1417 }
1418
1419 unlink(consumer_data->err_unix_sock_path);
1420 unlink(consumer_data->cmd_unix_sock_path);
1421 pthread_mutex_unlock(&consumer_data->lock);
1422
1423 /* Cleanup metadata socket mutex. */
1424 if (consumer_data->metadata_sock.lock) {
1425 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1426 free(consumer_data->metadata_sock.lock);
1427 }
1428 lttng_poll_clean(&events);
1429
1430 if (cmd_socket_wrapper) {
1431 consumer_destroy_socket(cmd_socket_wrapper);
1432 }
1433 error_poll:
1434 if (err) {
1435 health_error();
1436 ERR("Health error occurred in %s", __func__);
1437 }
1438 health_unregister(health_sessiond);
1439 DBG("consumer thread cleanup completed");
1440
1441 rcu_thread_offline();
1442 rcu_unregister_thread();
1443
1444 return NULL;
1445 }
1446
1447 /*
1448 * This thread manage application communication.
1449 */
1450 static void *thread_manage_apps(void *data)
1451 {
1452 int i, ret, pollfd, err = -1;
1453 ssize_t size_ret;
1454 uint32_t revents, nb_fd;
1455 struct lttng_poll_event events;
1456
1457 DBG("[thread] Manage application started");
1458
1459 rcu_register_thread();
1460 rcu_thread_online();
1461
1462 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1463
1464 if (testpoint(sessiond_thread_manage_apps)) {
1465 goto error_testpoint;
1466 }
1467
1468 health_code_update();
1469
1470 ret = sessiond_set_thread_pollset(&events, 2);
1471 if (ret < 0) {
1472 goto error_poll_create;
1473 }
1474
1475 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1476 if (ret < 0) {
1477 goto error;
1478 }
1479
1480 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1481 goto error;
1482 }
1483
1484 health_code_update();
1485
1486 while (1) {
1487 DBG("Apps thread polling");
1488
1489 /* Inifinite blocking call, waiting for transmission */
1490 restart:
1491 health_poll_entry();
1492 ret = lttng_poll_wait(&events, -1);
1493 DBG("Apps thread return from poll on %d fds",
1494 LTTNG_POLL_GETNB(&events));
1495 health_poll_exit();
1496 if (ret < 0) {
1497 /*
1498 * Restart interrupted system call.
1499 */
1500 if (errno == EINTR) {
1501 goto restart;
1502 }
1503 goto error;
1504 }
1505
1506 nb_fd = ret;
1507
1508 for (i = 0; i < nb_fd; i++) {
1509 /* Fetch once the poll data */
1510 revents = LTTNG_POLL_GETEV(&events, i);
1511 pollfd = LTTNG_POLL_GETFD(&events, i);
1512
1513 health_code_update();
1514
1515 if (!revents) {
1516 /* No activity for this FD (poll implementation). */
1517 continue;
1518 }
1519
1520 /* Thread quit pipe has been closed. Killing thread. */
1521 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1522 if (ret) {
1523 err = 0;
1524 goto exit;
1525 }
1526
1527 /* Inspect the apps cmd pipe */
1528 if (pollfd == apps_cmd_pipe[0]) {
1529 if (revents & LPOLLIN) {
1530 int sock;
1531
1532 /* Empty pipe */
1533 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1534 if (size_ret < sizeof(sock)) {
1535 PERROR("read apps cmd pipe");
1536 goto error;
1537 }
1538
1539 health_code_update();
1540
1541 /*
1542 * Since this is a command socket (write then read),
1543 * we only monitor the error events of the socket.
1544 */
1545 ret = lttng_poll_add(&events, sock,
1546 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1547 if (ret < 0) {
1548 goto error;
1549 }
1550
1551 DBG("Apps with sock %d added to poll set", sock);
1552 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1553 ERR("Apps command pipe error");
1554 goto error;
1555 } else {
1556 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1557 goto error;
1558 }
1559 } else {
1560 /*
1561 * At this point, we know that a registered application made
1562 * the event at poll_wait.
1563 */
1564 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1565 /* Removing from the poll set */
1566 ret = lttng_poll_del(&events, pollfd);
1567 if (ret < 0) {
1568 goto error;
1569 }
1570
1571 /* Socket closed on remote end. */
1572 ust_app_unregister(pollfd);
1573 } else {
1574 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1575 goto error;
1576 }
1577 }
1578
1579 health_code_update();
1580 }
1581 }
1582
1583 exit:
1584 error:
1585 lttng_poll_clean(&events);
1586 error_poll_create:
1587 error_testpoint:
1588 utils_close_pipe(apps_cmd_pipe);
1589 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1590
1591 /*
1592 * We don't clean the UST app hash table here since already registered
1593 * applications can still be controlled so let them be until the session
1594 * daemon dies or the applications stop.
1595 */
1596
1597 if (err) {
1598 health_error();
1599 ERR("Health error occurred in %s", __func__);
1600 }
1601 health_unregister(health_sessiond);
1602 DBG("Application communication apps thread cleanup complete");
1603 rcu_thread_offline();
1604 rcu_unregister_thread();
1605 return NULL;
1606 }
1607
1608 /*
1609 * Send a socket to a thread This is called from the dispatch UST registration
1610 * thread once all sockets are set for the application.
1611 *
1612 * The sock value can be invalid, we don't really care, the thread will handle
1613 * it and make the necessary cleanup if so.
1614 *
1615 * On success, return 0 else a negative value being the errno message of the
1616 * write().
1617 */
1618 static int send_socket_to_thread(int fd, int sock)
1619 {
1620 ssize_t ret;
1621
1622 /*
1623 * It's possible that the FD is set as invalid with -1 concurrently just
1624 * before calling this function being a shutdown state of the thread.
1625 */
1626 if (fd < 0) {
1627 ret = -EBADF;
1628 goto error;
1629 }
1630
1631 ret = lttng_write(fd, &sock, sizeof(sock));
1632 if (ret < sizeof(sock)) {
1633 PERROR("write apps pipe %d", fd);
1634 if (ret < 0) {
1635 ret = -errno;
1636 }
1637 goto error;
1638 }
1639
1640 /* All good. Don't send back the write positive ret value. */
1641 ret = 0;
1642 error:
1643 return (int) ret;
1644 }
1645
1646 /*
1647 * Sanitize the wait queue of the dispatch registration thread meaning removing
1648 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1649 * notify socket is never received.
1650 */
1651 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1652 {
1653 int ret, nb_fd = 0, i;
1654 unsigned int fd_added = 0;
1655 struct lttng_poll_event events;
1656 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1657
1658 assert(wait_queue);
1659
1660 lttng_poll_init(&events);
1661
1662 /* Just skip everything for an empty queue. */
1663 if (!wait_queue->count) {
1664 goto end;
1665 }
1666
1667 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1668 if (ret < 0) {
1669 goto error_create;
1670 }
1671
1672 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1673 &wait_queue->head, head) {
1674 assert(wait_node->app);
1675 ret = lttng_poll_add(&events, wait_node->app->sock,
1676 LPOLLHUP | LPOLLERR);
1677 if (ret < 0) {
1678 goto error;
1679 }
1680
1681 fd_added = 1;
1682 }
1683
1684 if (!fd_added) {
1685 goto end;
1686 }
1687
1688 /*
1689 * Poll but don't block so we can quickly identify the faulty events and
1690 * clean them afterwards from the wait queue.
1691 */
1692 ret = lttng_poll_wait(&events, 0);
1693 if (ret < 0) {
1694 goto error;
1695 }
1696 nb_fd = ret;
1697
1698 for (i = 0; i < nb_fd; i++) {
1699 /* Get faulty FD. */
1700 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1701 int pollfd = LTTNG_POLL_GETFD(&events, i);
1702
1703 if (!revents) {
1704 /* No activity for this FD (poll implementation). */
1705 continue;
1706 }
1707
1708 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1709 &wait_queue->head, head) {
1710 if (pollfd == wait_node->app->sock &&
1711 (revents & (LPOLLHUP | LPOLLERR))) {
1712 cds_list_del(&wait_node->head);
1713 wait_queue->count--;
1714 ust_app_destroy(wait_node->app);
1715 free(wait_node);
1716 /*
1717 * Silence warning of use-after-free in
1718 * cds_list_for_each_entry_safe which uses
1719 * __typeof__(*wait_node).
1720 */
1721 wait_node = NULL;
1722 break;
1723 } else {
1724 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1725 goto error;
1726 }
1727 }
1728 }
1729
1730 if (nb_fd > 0) {
1731 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1732 }
1733
1734 end:
1735 lttng_poll_clean(&events);
1736 return;
1737
1738 error:
1739 lttng_poll_clean(&events);
1740 error_create:
1741 ERR("Unable to sanitize wait queue");
1742 return;
1743 }
1744
1745 /*
1746 * Dispatch request from the registration threads to the application
1747 * communication thread.
1748 */
1749 static void *thread_dispatch_ust_registration(void *data)
1750 {
1751 int ret, err = -1;
1752 struct cds_wfcq_node *node;
1753 struct ust_command *ust_cmd = NULL;
1754 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1755 struct ust_reg_wait_queue wait_queue = {
1756 .count = 0,
1757 };
1758
1759 rcu_register_thread();
1760
1761 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1762
1763 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1764 goto error_testpoint;
1765 }
1766
1767 health_code_update();
1768
1769 CDS_INIT_LIST_HEAD(&wait_queue.head);
1770
1771 DBG("[thread] Dispatch UST command started");
1772
1773 for (;;) {
1774 health_code_update();
1775
1776 /* Atomically prepare the queue futex */
1777 futex_nto1_prepare(&ust_cmd_queue.futex);
1778
1779 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1780 break;
1781 }
1782
1783 do {
1784 struct ust_app *app = NULL;
1785 ust_cmd = NULL;
1786
1787 /*
1788 * Make sure we don't have node(s) that have hung up before receiving
1789 * the notify socket. This is to clean the list in order to avoid
1790 * memory leaks from notify socket that are never seen.
1791 */
1792 sanitize_wait_queue(&wait_queue);
1793
1794 health_code_update();
1795 /* Dequeue command for registration */
1796 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1797 if (node == NULL) {
1798 DBG("Woken up but nothing in the UST command queue");
1799 /* Continue thread execution */
1800 break;
1801 }
1802
1803 ust_cmd = caa_container_of(node, struct ust_command, node);
1804
1805 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1806 " gid:%d sock:%d name:%s (version %d.%d)",
1807 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1808 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1809 ust_cmd->sock, ust_cmd->reg_msg.name,
1810 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1811
1812 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1813 wait_node = zmalloc(sizeof(*wait_node));
1814 if (!wait_node) {
1815 PERROR("zmalloc wait_node dispatch");
1816 ret = close(ust_cmd->sock);
1817 if (ret < 0) {
1818 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1819 }
1820 lttng_fd_put(LTTNG_FD_APPS, 1);
1821 free(ust_cmd);
1822 goto error;
1823 }
1824 CDS_INIT_LIST_HEAD(&wait_node->head);
1825
1826 /* Create application object if socket is CMD. */
1827 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1828 ust_cmd->sock);
1829 if (!wait_node->app) {
1830 ret = close(ust_cmd->sock);
1831 if (ret < 0) {
1832 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1833 }
1834 lttng_fd_put(LTTNG_FD_APPS, 1);
1835 free(wait_node);
1836 free(ust_cmd);
1837 continue;
1838 }
1839 /*
1840 * Add application to the wait queue so we can set the notify
1841 * socket before putting this object in the global ht.
1842 */
1843 cds_list_add(&wait_node->head, &wait_queue.head);
1844 wait_queue.count++;
1845
1846 free(ust_cmd);
1847 /*
1848 * We have to continue here since we don't have the notify
1849 * socket and the application MUST be added to the hash table
1850 * only at that moment.
1851 */
1852 continue;
1853 } else {
1854 /*
1855 * Look for the application in the local wait queue and set the
1856 * notify socket if found.
1857 */
1858 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1859 &wait_queue.head, head) {
1860 health_code_update();
1861 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1862 wait_node->app->notify_sock = ust_cmd->sock;
1863 cds_list_del(&wait_node->head);
1864 wait_queue.count--;
1865 app = wait_node->app;
1866 free(wait_node);
1867 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1868 break;
1869 }
1870 }
1871
1872 /*
1873 * With no application at this stage the received socket is
1874 * basically useless so close it before we free the cmd data
1875 * structure for good.
1876 */
1877 if (!app) {
1878 ret = close(ust_cmd->sock);
1879 if (ret < 0) {
1880 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1881 }
1882 lttng_fd_put(LTTNG_FD_APPS, 1);
1883 }
1884 free(ust_cmd);
1885 }
1886
1887 if (app) {
1888 /*
1889 * @session_lock_list
1890 *
1891 * Lock the global session list so from the register up to the
1892 * registration done message, no thread can see the application
1893 * and change its state.
1894 */
1895 session_lock_list();
1896 rcu_read_lock();
1897
1898 /*
1899 * Add application to the global hash table. This needs to be
1900 * done before the update to the UST registry can locate the
1901 * application.
1902 */
1903 ust_app_add(app);
1904
1905 /* Set app version. This call will print an error if needed. */
1906 (void) ust_app_version(app);
1907
1908 /* Send notify socket through the notify pipe. */
1909 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1910 app->notify_sock);
1911 if (ret < 0) {
1912 rcu_read_unlock();
1913 session_unlock_list();
1914 /*
1915 * No notify thread, stop the UST tracing. However, this is
1916 * not an internal error of the this thread thus setting
1917 * the health error code to a normal exit.
1918 */
1919 err = 0;
1920 goto error;
1921 }
1922
1923 /*
1924 * Update newly registered application with the tracing
1925 * registry info already enabled information.
1926 */
1927 update_ust_app(app->sock);
1928
1929 /*
1930 * Don't care about return value. Let the manage apps threads
1931 * handle app unregistration upon socket close.
1932 */
1933 (void) ust_app_register_done(app);
1934
1935 /*
1936 * Even if the application socket has been closed, send the app
1937 * to the thread and unregistration will take place at that
1938 * place.
1939 */
1940 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1941 if (ret < 0) {
1942 rcu_read_unlock();
1943 session_unlock_list();
1944 /*
1945 * No apps. thread, stop the UST tracing. However, this is
1946 * not an internal error of the this thread thus setting
1947 * the health error code to a normal exit.
1948 */
1949 err = 0;
1950 goto error;
1951 }
1952
1953 rcu_read_unlock();
1954 session_unlock_list();
1955 }
1956 } while (node != NULL);
1957
1958 health_poll_entry();
1959 /* Futex wait on queue. Blocking call on futex() */
1960 futex_nto1_wait(&ust_cmd_queue.futex);
1961 health_poll_exit();
1962 }
1963 /* Normal exit, no error */
1964 err = 0;
1965
1966 error:
1967 /* Clean up wait queue. */
1968 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1969 &wait_queue.head, head) {
1970 cds_list_del(&wait_node->head);
1971 wait_queue.count--;
1972 free(wait_node);
1973 }
1974
1975 /* Empty command queue. */
1976 for (;;) {
1977 /* Dequeue command for registration */
1978 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1979 if (node == NULL) {
1980 break;
1981 }
1982 ust_cmd = caa_container_of(node, struct ust_command, node);
1983 ret = close(ust_cmd->sock);
1984 if (ret < 0) {
1985 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
1986 }
1987 lttng_fd_put(LTTNG_FD_APPS, 1);
1988 free(ust_cmd);
1989 }
1990
1991 error_testpoint:
1992 DBG("Dispatch thread dying");
1993 if (err) {
1994 health_error();
1995 ERR("Health error occurred in %s", __func__);
1996 }
1997 health_unregister(health_sessiond);
1998 rcu_unregister_thread();
1999 return NULL;
2000 }
2001
2002 /*
2003 * This thread manage application registration.
2004 */
2005 static void *thread_registration_apps(void *data)
2006 {
2007 int sock = -1, i, ret, pollfd, err = -1;
2008 uint32_t revents, nb_fd;
2009 struct lttng_poll_event events;
2010 /*
2011 * Get allocated in this thread, enqueued to a global queue, dequeued and
2012 * freed in the manage apps thread.
2013 */
2014 struct ust_command *ust_cmd = NULL;
2015
2016 DBG("[thread] Manage application registration started");
2017
2018 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2019
2020 if (testpoint(sessiond_thread_registration_apps)) {
2021 goto error_testpoint;
2022 }
2023
2024 ret = lttcomm_listen_unix_sock(apps_sock);
2025 if (ret < 0) {
2026 goto error_listen;
2027 }
2028
2029 /*
2030 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2031 * more will be added to this poll set.
2032 */
2033 ret = sessiond_set_thread_pollset(&events, 2);
2034 if (ret < 0) {
2035 goto error_create_poll;
2036 }
2037
2038 /* Add the application registration socket */
2039 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2040 if (ret < 0) {
2041 goto error_poll_add;
2042 }
2043
2044 /* Notify all applications to register */
2045 ret = notify_ust_apps(1);
2046 if (ret < 0) {
2047 ERR("Failed to notify applications or create the wait shared memory.\n"
2048 "Execution continues but there might be problem for already\n"
2049 "running applications that wishes to register.");
2050 }
2051
2052 while (1) {
2053 DBG("Accepting application registration");
2054
2055 /* Inifinite blocking call, waiting for transmission */
2056 restart:
2057 health_poll_entry();
2058 ret = lttng_poll_wait(&events, -1);
2059 health_poll_exit();
2060 if (ret < 0) {
2061 /*
2062 * Restart interrupted system call.
2063 */
2064 if (errno == EINTR) {
2065 goto restart;
2066 }
2067 goto error;
2068 }
2069
2070 nb_fd = ret;
2071
2072 for (i = 0; i < nb_fd; i++) {
2073 health_code_update();
2074
2075 /* Fetch once the poll data */
2076 revents = LTTNG_POLL_GETEV(&events, i);
2077 pollfd = LTTNG_POLL_GETFD(&events, i);
2078
2079 if (!revents) {
2080 /* No activity for this FD (poll implementation). */
2081 continue;
2082 }
2083
2084 /* Thread quit pipe has been closed. Killing thread. */
2085 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2086 if (ret) {
2087 err = 0;
2088 goto exit;
2089 }
2090
2091 /* Event on the registration socket */
2092 if (pollfd == apps_sock) {
2093 if (revents & LPOLLIN) {
2094 sock = lttcomm_accept_unix_sock(apps_sock);
2095 if (sock < 0) {
2096 goto error;
2097 }
2098
2099 /*
2100 * Set socket timeout for both receiving and ending.
2101 * app_socket_timeout is in seconds, whereas
2102 * lttcomm_setsockopt_rcv_timeout and
2103 * lttcomm_setsockopt_snd_timeout expect msec as
2104 * parameter.
2105 */
2106 if (config.app_socket_timeout >= 0) {
2107 (void) lttcomm_setsockopt_rcv_timeout(sock,
2108 config.app_socket_timeout * 1000);
2109 (void) lttcomm_setsockopt_snd_timeout(sock,
2110 config.app_socket_timeout * 1000);
2111 }
2112
2113 /*
2114 * Set the CLOEXEC flag. Return code is useless because
2115 * either way, the show must go on.
2116 */
2117 (void) utils_set_fd_cloexec(sock);
2118
2119 /* Create UST registration command for enqueuing */
2120 ust_cmd = zmalloc(sizeof(struct ust_command));
2121 if (ust_cmd == NULL) {
2122 PERROR("ust command zmalloc");
2123 ret = close(sock);
2124 if (ret) {
2125 PERROR("close");
2126 }
2127 goto error;
2128 }
2129
2130 /*
2131 * Using message-based transmissions to ensure we don't
2132 * have to deal with partially received messages.
2133 */
2134 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2135 if (ret < 0) {
2136 ERR("Exhausted file descriptors allowed for applications.");
2137 free(ust_cmd);
2138 ret = close(sock);
2139 if (ret) {
2140 PERROR("close");
2141 }
2142 sock = -1;
2143 continue;
2144 }
2145
2146 health_code_update();
2147 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2148 if (ret < 0) {
2149 free(ust_cmd);
2150 /* Close socket of the application. */
2151 ret = close(sock);
2152 if (ret) {
2153 PERROR("close");
2154 }
2155 lttng_fd_put(LTTNG_FD_APPS, 1);
2156 sock = -1;
2157 continue;
2158 }
2159 health_code_update();
2160
2161 ust_cmd->sock = sock;
2162 sock = -1;
2163
2164 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2165 " gid:%d sock:%d name:%s (version %d.%d)",
2166 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2167 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2168 ust_cmd->sock, ust_cmd->reg_msg.name,
2169 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2170
2171 /*
2172 * Lock free enqueue the registration request. The red pill
2173 * has been taken! This apps will be part of the *system*.
2174 */
2175 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2176
2177 /*
2178 * Wake the registration queue futex. Implicit memory
2179 * barrier with the exchange in cds_wfcq_enqueue.
2180 */
2181 futex_nto1_wake(&ust_cmd_queue.futex);
2182 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2183 ERR("Register apps socket poll error");
2184 goto error;
2185 } else {
2186 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2187 goto error;
2188 }
2189 }
2190 }
2191 }
2192
2193 exit:
2194 error:
2195 /* Notify that the registration thread is gone */
2196 notify_ust_apps(0);
2197
2198 if (apps_sock >= 0) {
2199 ret = close(apps_sock);
2200 if (ret) {
2201 PERROR("close");
2202 }
2203 }
2204 if (sock >= 0) {
2205 ret = close(sock);
2206 if (ret) {
2207 PERROR("close");
2208 }
2209 lttng_fd_put(LTTNG_FD_APPS, 1);
2210 }
2211 unlink(config.apps_unix_sock_path.value);
2212
2213 error_poll_add:
2214 lttng_poll_clean(&events);
2215 error_listen:
2216 error_create_poll:
2217 error_testpoint:
2218 DBG("UST Registration thread cleanup complete");
2219 if (err) {
2220 health_error();
2221 ERR("Health error occurred in %s", __func__);
2222 }
2223 health_unregister(health_sessiond);
2224
2225 return NULL;
2226 }
2227
2228 /*
2229 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2230 * exec or it will fails.
2231 */
2232 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2233 {
2234 int ret, clock_ret;
2235 struct timespec timeout;
2236
2237 /*
2238 * Make sure we set the readiness flag to 0 because we are NOT ready.
2239 * This access to consumer_thread_is_ready does not need to be
2240 * protected by consumer_data.cond_mutex (yet) since the consumer
2241 * management thread has not been started at this point.
2242 */
2243 consumer_data->consumer_thread_is_ready = 0;
2244
2245 /* Setup pthread condition */
2246 ret = pthread_condattr_init(&consumer_data->condattr);
2247 if (ret) {
2248 errno = ret;
2249 PERROR("pthread_condattr_init consumer data");
2250 goto error;
2251 }
2252
2253 /*
2254 * Set the monotonic clock in order to make sure we DO NOT jump in time
2255 * between the clock_gettime() call and the timedwait call. See bug #324
2256 * for a more details and how we noticed it.
2257 */
2258 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2259 if (ret) {
2260 errno = ret;
2261 PERROR("pthread_condattr_setclock consumer data");
2262 goto error;
2263 }
2264
2265 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2266 if (ret) {
2267 errno = ret;
2268 PERROR("pthread_cond_init consumer data");
2269 goto error;
2270 }
2271
2272 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2273 thread_manage_consumer, consumer_data);
2274 if (ret) {
2275 errno = ret;
2276 PERROR("pthread_create consumer");
2277 ret = -1;
2278 goto error;
2279 }
2280
2281 /* We are about to wait on a pthread condition */
2282 pthread_mutex_lock(&consumer_data->cond_mutex);
2283
2284 /* Get time for sem_timedwait absolute timeout */
2285 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
2286 /*
2287 * Set the timeout for the condition timed wait even if the clock gettime
2288 * call fails since we might loop on that call and we want to avoid to
2289 * increment the timeout too many times.
2290 */
2291 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2292
2293 /*
2294 * The following loop COULD be skipped in some conditions so this is why we
2295 * set ret to 0 in order to make sure at least one round of the loop is
2296 * done.
2297 */
2298 ret = 0;
2299
2300 /*
2301 * Loop until the condition is reached or when a timeout is reached. Note
2302 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2303 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2304 * possible. This loop does not take any chances and works with both of
2305 * them.
2306 */
2307 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2308 if (clock_ret < 0) {
2309 PERROR("clock_gettime spawn consumer");
2310 /* Infinite wait for the consumerd thread to be ready */
2311 ret = pthread_cond_wait(&consumer_data->cond,
2312 &consumer_data->cond_mutex);
2313 } else {
2314 ret = pthread_cond_timedwait(&consumer_data->cond,
2315 &consumer_data->cond_mutex, &timeout);
2316 }
2317 }
2318
2319 /* Release the pthread condition */
2320 pthread_mutex_unlock(&consumer_data->cond_mutex);
2321
2322 if (ret != 0) {
2323 errno = ret;
2324 if (ret == ETIMEDOUT) {
2325 int pth_ret;
2326
2327 /*
2328 * Call has timed out so we kill the kconsumerd_thread and return
2329 * an error.
2330 */
2331 ERR("Condition timed out. The consumer thread was never ready."
2332 " Killing it");
2333 pth_ret = pthread_cancel(consumer_data->thread);
2334 if (pth_ret < 0) {
2335 PERROR("pthread_cancel consumer thread");
2336 }
2337 } else {
2338 PERROR("pthread_cond_wait failed consumer thread");
2339 }
2340 /* Caller is expecting a negative value on failure. */
2341 ret = -1;
2342 goto error;
2343 }
2344
2345 pthread_mutex_lock(&consumer_data->pid_mutex);
2346 if (consumer_data->pid == 0) {
2347 ERR("Consumerd did not start");
2348 pthread_mutex_unlock(&consumer_data->pid_mutex);
2349 goto error;
2350 }
2351 pthread_mutex_unlock(&consumer_data->pid_mutex);
2352
2353 return 0;
2354
2355 error:
2356 return ret;
2357 }
2358
2359 /*
2360 * Join consumer thread
2361 */
2362 static int join_consumer_thread(struct consumer_data *consumer_data)
2363 {
2364 void *status;
2365
2366 /* Consumer pid must be a real one. */
2367 if (consumer_data->pid > 0) {
2368 int ret;
2369 ret = kill(consumer_data->pid, SIGTERM);
2370 if (ret) {
2371 PERROR("Error killing consumer daemon");
2372 return ret;
2373 }
2374 return pthread_join(consumer_data->thread, &status);
2375 } else {
2376 return 0;
2377 }
2378 }
2379
2380 /*
2381 * Fork and exec a consumer daemon (consumerd).
2382 *
2383 * Return pid if successful else -1.
2384 */
2385 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2386 {
2387 int ret;
2388 pid_t pid;
2389 const char *consumer_to_use;
2390 const char *verbosity;
2391 struct stat st;
2392
2393 DBG("Spawning consumerd");
2394
2395 pid = fork();
2396 if (pid == 0) {
2397 /*
2398 * Exec consumerd.
2399 */
2400 if (config.verbose_consumer) {
2401 verbosity = "--verbose";
2402 } else if (lttng_opt_quiet) {
2403 verbosity = "--quiet";
2404 } else {
2405 verbosity = "";
2406 }
2407
2408 switch (consumer_data->type) {
2409 case LTTNG_CONSUMER_KERNEL:
2410 /*
2411 * Find out which consumerd to execute. We will first try the
2412 * 64-bit path, then the sessiond's installation directory, and
2413 * fallback on the 32-bit one,
2414 */
2415 DBG3("Looking for a kernel consumer at these locations:");
2416 DBG3(" 1) %s", config.consumerd64_bin_path.value);
2417 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
2418 DBG3(" 3) %s", config.consumerd32_bin_path.value);
2419 if (stat(config.consumerd64_bin_path.value, &st) == 0) {
2420 DBG3("Found location #1");
2421 consumer_to_use = config.consumerd64_bin_path.value;
2422 } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
2423 DBG3("Found location #2");
2424 consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
2425 } else if (stat(config.consumerd32_bin_path.value, &st) == 0) {
2426 DBG3("Found location #3");
2427 consumer_to_use = config.consumerd32_bin_path.value;
2428 } else {
2429 DBG("Could not find any valid consumerd executable");
2430 ret = -EINVAL;
2431 goto error;
2432 }
2433 DBG("Using kernel consumer at: %s", consumer_to_use);
2434 (void) execl(consumer_to_use,
2435 "lttng-consumerd", verbosity, "-k",
2436 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2437 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2438 "--group", config.tracing_group_name.value,
2439 NULL);
2440 break;
2441 case LTTNG_CONSUMER64_UST:
2442 {
2443 char *tmpnew = NULL;
2444
2445 if (config.consumerd64_lib_dir.value[0] != '\0') {
2446 char *tmp;
2447 size_t tmplen;
2448
2449 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2450 if (!tmp) {
2451 tmp = "";
2452 }
2453 tmplen = strlen("LD_LIBRARY_PATH=")
2454 + strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
2455 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2456 if (!tmpnew) {
2457 ret = -ENOMEM;
2458 goto error;
2459 }
2460 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2461 strcat(tmpnew, config.consumerd64_lib_dir.value);
2462 if (tmp[0] != '\0') {
2463 strcat(tmpnew, ":");
2464 strcat(tmpnew, tmp);
2465 }
2466 ret = putenv(tmpnew);
2467 if (ret) {
2468 ret = -errno;
2469 free(tmpnew);
2470 goto error;
2471 }
2472 }
2473 DBG("Using 64-bit UST consumer at: %s", config.consumerd64_bin_path.value);
2474 (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
2475 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2476 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2477 "--group", config.tracing_group_name.value,
2478 NULL);
2479 if (config.consumerd64_lib_dir.value[0] != '\0') {
2480 free(tmpnew);
2481 }
2482 break;
2483 }
2484 case LTTNG_CONSUMER32_UST:
2485 {
2486 char *tmpnew = NULL;
2487
2488 if (config.consumerd32_lib_dir.value[0] != '\0') {
2489 char *tmp;
2490 size_t tmplen;
2491
2492 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2493 if (!tmp) {
2494 tmp = "";
2495 }
2496 tmplen = strlen("LD_LIBRARY_PATH=")
2497 + strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
2498 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2499 if (!tmpnew) {
2500 ret = -ENOMEM;
2501 goto error;
2502 }
2503 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2504 strcat(tmpnew, config.consumerd32_lib_dir.value);
2505 if (tmp[0] != '\0') {
2506 strcat(tmpnew, ":");
2507 strcat(tmpnew, tmp);
2508 }
2509 ret = putenv(tmpnew);
2510 if (ret) {
2511 ret = -errno;
2512 free(tmpnew);
2513 goto error;
2514 }
2515 }
2516 DBG("Using 32-bit UST consumer at: %s", config.consumerd32_bin_path.value);
2517 (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
2518 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2519 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2520 "--group", config.tracing_group_name.value,
2521 NULL);
2522 if (config.consumerd32_lib_dir.value[0] != '\0') {
2523 free(tmpnew);
2524 }
2525 break;
2526 }
2527 default:
2528 PERROR("unknown consumer type");
2529 exit(EXIT_FAILURE);
2530 }
2531 if (errno != 0) {
2532 PERROR("Consumer execl()");
2533 }
2534 /* Reaching this point, we got a failure on our execl(). */
2535 exit(EXIT_FAILURE);
2536 } else if (pid > 0) {
2537 ret = pid;
2538 } else {
2539 PERROR("start consumer fork");
2540 ret = -errno;
2541 }
2542 error:
2543 return ret;
2544 }
2545
2546 /*
2547 * Spawn the consumerd daemon and session daemon thread.
2548 */
2549 static int start_consumerd(struct consumer_data *consumer_data)
2550 {
2551 int ret;
2552
2553 /*
2554 * Set the listen() state on the socket since there is a possible race
2555 * between the exec() of the consumer daemon and this call if place in the
2556 * consumer thread. See bug #366 for more details.
2557 */
2558 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2559 if (ret < 0) {
2560 goto error;
2561 }
2562
2563 pthread_mutex_lock(&consumer_data->pid_mutex);
2564 if (consumer_data->pid != 0) {
2565 pthread_mutex_unlock(&consumer_data->pid_mutex);
2566 goto end;
2567 }
2568
2569 ret = spawn_consumerd(consumer_data);
2570 if (ret < 0) {
2571 ERR("Spawning consumerd failed");
2572 pthread_mutex_unlock(&consumer_data->pid_mutex);
2573 goto error;
2574 }
2575
2576 /* Setting up the consumer_data pid */
2577 consumer_data->pid = ret;
2578 DBG2("Consumer pid %d", consumer_data->pid);
2579 pthread_mutex_unlock(&consumer_data->pid_mutex);
2580
2581 DBG2("Spawning consumer control thread");
2582 ret = spawn_consumer_thread(consumer_data);
2583 if (ret < 0) {
2584 ERR("Fatal error spawning consumer control thread");
2585 goto error;
2586 }
2587
2588 end:
2589 return 0;
2590
2591 error:
2592 /* Cleanup already created sockets on error. */
2593 if (consumer_data->err_sock >= 0) {
2594 int err;
2595
2596 err = close(consumer_data->err_sock);
2597 if (err < 0) {
2598 PERROR("close consumer data error socket");
2599 }
2600 }
2601 return ret;
2602 }
2603
2604 /*
2605 * Setup necessary data for kernel tracer action.
2606 */
2607 static int init_kernel_tracer(void)
2608 {
2609 int ret;
2610
2611 /* Modprobe lttng kernel modules */
2612 ret = modprobe_lttng_control();
2613 if (ret < 0) {
2614 goto error;
2615 }
2616
2617 /* Open debugfs lttng */
2618 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2619 if (kernel_tracer_fd < 0) {
2620 DBG("Failed to open %s", module_proc_lttng);
2621 goto error_open;
2622 }
2623
2624 /* Validate kernel version */
2625 ret = kernel_validate_version(kernel_tracer_fd);
2626 if (ret < 0) {
2627 goto error_version;
2628 }
2629
2630 ret = modprobe_lttng_data();
2631 if (ret < 0) {
2632 goto error_modules;
2633 }
2634
2635 ret = kernel_supports_ring_buffer_snapshot_sample_positions(
2636 kernel_tracer_fd);
2637 if (ret < 0) {
2638 goto error_modules;
2639 }
2640
2641 if (ret < 1) {
2642 WARN("Kernel tracer does not support buffer monitoring. "
2643 "The monitoring timer of channels in the kernel domain "
2644 "will be set to 0 (disabled).");
2645 }
2646
2647 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2648 return 0;
2649
2650 error_version:
2651 modprobe_remove_lttng_control();
2652 ret = close(kernel_tracer_fd);
2653 if (ret) {
2654 PERROR("close");
2655 }
2656 kernel_tracer_fd = -1;
2657 return LTTNG_ERR_KERN_VERSION;
2658
2659 error_modules:
2660 ret = close(kernel_tracer_fd);
2661 if (ret) {
2662 PERROR("close");
2663 }
2664
2665 error_open:
2666 modprobe_remove_lttng_control();
2667
2668 error:
2669 WARN("No kernel tracer available");
2670 kernel_tracer_fd = -1;
2671 if (!is_root) {
2672 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2673 } else {
2674 return LTTNG_ERR_KERN_NA;
2675 }
2676 }
2677
2678
2679 /*
2680 * Copy consumer output from the tracing session to the domain session. The
2681 * function also applies the right modification on a per domain basis for the
2682 * trace files destination directory.
2683 *
2684 * Should *NOT* be called with RCU read-side lock held.
2685 */
2686 static int copy_session_consumer(int domain, struct ltt_session *session)
2687 {
2688 int ret;
2689 const char *dir_name;
2690 struct consumer_output *consumer;
2691
2692 assert(session);
2693 assert(session->consumer);
2694
2695 switch (domain) {
2696 case LTTNG_DOMAIN_KERNEL:
2697 DBG3("Copying tracing session consumer output in kernel session");
2698 /*
2699 * XXX: We should audit the session creation and what this function
2700 * does "extra" in order to avoid a destroy since this function is used
2701 * in the domain session creation (kernel and ust) only. Same for UST
2702 * domain.
2703 */
2704 if (session->kernel_session->consumer) {
2705 consumer_output_put(session->kernel_session->consumer);
2706 }
2707 session->kernel_session->consumer =
2708 consumer_copy_output(session->consumer);
2709 /* Ease our life a bit for the next part */
2710 consumer = session->kernel_session->consumer;
2711 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2712 break;
2713 case LTTNG_DOMAIN_JUL:
2714 case LTTNG_DOMAIN_LOG4J:
2715 case LTTNG_DOMAIN_PYTHON:
2716 case LTTNG_DOMAIN_UST:
2717 DBG3("Copying tracing session consumer output in UST session");
2718 if (session->ust_session->consumer) {
2719 consumer_output_put(session->ust_session->consumer);
2720 }
2721 session->ust_session->consumer =
2722 consumer_copy_output(session->consumer);
2723 /* Ease our life a bit for the next part */
2724 consumer = session->ust_session->consumer;
2725 dir_name = DEFAULT_UST_TRACE_DIR;
2726 break;
2727 default:
2728 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2729 goto error;
2730 }
2731
2732 /* Append correct directory to subdir */
2733 strncat(consumer->subdir, dir_name,
2734 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2735 DBG3("Copy session consumer subdir %s", consumer->subdir);
2736
2737 ret = LTTNG_OK;
2738
2739 error:
2740 return ret;
2741 }
2742
2743 /*
2744 * Create an UST session and add it to the session ust list.
2745 *
2746 * Should *NOT* be called with RCU read-side lock held.
2747 */
2748 static int create_ust_session(struct ltt_session *session,
2749 struct lttng_domain *domain)
2750 {
2751 int ret;
2752 struct ltt_ust_session *lus = NULL;
2753
2754 assert(session);
2755 assert(domain);
2756 assert(session->consumer);
2757
2758 switch (domain->type) {
2759 case LTTNG_DOMAIN_JUL:
2760 case LTTNG_DOMAIN_LOG4J:
2761 case LTTNG_DOMAIN_PYTHON:
2762 case LTTNG_DOMAIN_UST:
2763 break;
2764 default:
2765 ERR("Unknown UST domain on create session %d", domain->type);
2766 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2767 goto error;
2768 }
2769
2770 DBG("Creating UST session");
2771
2772 lus = trace_ust_create_session(session->id);
2773 if (lus == NULL) {
2774 ret = LTTNG_ERR_UST_SESS_FAIL;
2775 goto error;
2776 }
2777
2778 lus->uid = session->uid;
2779 lus->gid = session->gid;
2780 lus->output_traces = session->output_traces;
2781 lus->snapshot_mode = session->snapshot_mode;
2782 lus->live_timer_interval = session->live_timer;
2783 session->ust_session = lus;
2784 if (session->shm_path[0]) {
2785 strncpy(lus->root_shm_path, session->shm_path,
2786 sizeof(lus->root_shm_path));
2787 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2788 strncpy(lus->shm_path, session->shm_path,
2789 sizeof(lus->shm_path));
2790 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2791 strncat(lus->shm_path, "/ust",
2792 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2793 }
2794 /* Copy session output to the newly created UST session */
2795 ret = copy_session_consumer(domain->type, session);
2796 if (ret != LTTNG_OK) {
2797 goto error;
2798 }
2799
2800 return LTTNG_OK;
2801
2802 error:
2803 free(lus);
2804 session->ust_session = NULL;
2805 return ret;
2806 }
2807
2808 /*
2809 * Create a kernel tracer session then create the default channel.
2810 */
2811 static int create_kernel_session(struct ltt_session *session)
2812 {
2813 int ret;
2814
2815 DBG("Creating kernel session");
2816
2817 ret = kernel_create_session(session, kernel_tracer_fd);
2818 if (ret < 0) {
2819 ret = LTTNG_ERR_KERN_SESS_FAIL;
2820 goto error;
2821 }
2822
2823 /* Code flow safety */
2824 assert(session->kernel_session);
2825
2826 /* Copy session output to the newly created Kernel session */
2827 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2828 if (ret != LTTNG_OK) {
2829 goto error;
2830 }
2831
2832 /* Create directory(ies) on local filesystem. */
2833 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2834 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2835 ret = run_as_mkdir_recursive(
2836 session->kernel_session->consumer->dst.trace_path,
2837 S_IRWXU | S_IRWXG, session->uid, session->gid);
2838 if (ret < 0) {
2839 if (errno != EEXIST) {
2840 ERR("Trace directory creation error");
2841 goto error;
2842 }
2843 }
2844 }
2845
2846 session->kernel_session->uid = session->uid;
2847 session->kernel_session->gid = session->gid;
2848 session->kernel_session->output_traces = session->output_traces;
2849 session->kernel_session->snapshot_mode = session->snapshot_mode;
2850
2851 return LTTNG_OK;
2852
2853 error:
2854 trace_kernel_destroy_session(session->kernel_session);
2855 session->kernel_session = NULL;
2856 return ret;
2857 }
2858
2859 /*
2860 * Count number of session permitted by uid/gid.
2861 */
2862 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2863 {
2864 unsigned int i = 0;
2865 struct ltt_session *session;
2866
2867 DBG("Counting number of available session for UID %d GID %d",
2868 uid, gid);
2869 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2870 /*
2871 * Only list the sessions the user can control.
2872 */
2873 if (!session_access_ok(session, uid, gid)) {
2874 continue;
2875 }
2876 i++;
2877 }
2878 return i;
2879 }
2880
2881 /*
2882 * Process the command requested by the lttng client within the command
2883 * context structure. This function make sure that the return structure (llm)
2884 * is set and ready for transmission before returning.
2885 *
2886 * Return any error encountered or 0 for success.
2887 *
2888 * "sock" is only used for special-case var. len data.
2889 *
2890 * Should *NOT* be called with RCU read-side lock held.
2891 */
2892 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2893 int *sock_error)
2894 {
2895 int ret = LTTNG_OK;
2896 int need_tracing_session = 1;
2897 int need_domain;
2898
2899 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2900
2901 assert(!rcu_read_ongoing());
2902
2903 *sock_error = 0;
2904
2905 switch (cmd_ctx->lsm->cmd_type) {
2906 case LTTNG_CREATE_SESSION:
2907 case LTTNG_CREATE_SESSION_SNAPSHOT:
2908 case LTTNG_CREATE_SESSION_LIVE:
2909 case LTTNG_DESTROY_SESSION:
2910 case LTTNG_LIST_SESSIONS:
2911 case LTTNG_LIST_DOMAINS:
2912 case LTTNG_START_TRACE:
2913 case LTTNG_STOP_TRACE:
2914 case LTTNG_DATA_PENDING:
2915 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2916 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2917 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2918 case LTTNG_SNAPSHOT_RECORD:
2919 case LTTNG_SAVE_SESSION:
2920 case LTTNG_SET_SESSION_SHM_PATH:
2921 case LTTNG_REGENERATE_METADATA:
2922 case LTTNG_REGENERATE_STATEDUMP:
2923 case LTTNG_REGISTER_TRIGGER:
2924 case LTTNG_UNREGISTER_TRIGGER:
2925 need_domain = 0;
2926 break;
2927 default:
2928 need_domain = 1;
2929 }
2930
2931 if (config.no_kernel && need_domain
2932 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2933 if (!is_root) {
2934 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2935 } else {
2936 ret = LTTNG_ERR_KERN_NA;
2937 }
2938 goto error;
2939 }
2940
2941 /* Deny register consumer if we already have a spawned consumer. */
2942 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2943 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2944 if (kconsumer_data.pid > 0) {
2945 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2946 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2947 goto error;
2948 }
2949 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2950 }
2951
2952 /*
2953 * Check for command that don't needs to allocate a returned payload. We do
2954 * this here so we don't have to make the call for no payload at each
2955 * command.
2956 */
2957 switch(cmd_ctx->lsm->cmd_type) {
2958 case LTTNG_LIST_SESSIONS:
2959 case LTTNG_LIST_TRACEPOINTS:
2960 case LTTNG_LIST_TRACEPOINT_FIELDS:
2961 case LTTNG_LIST_DOMAINS:
2962 case LTTNG_LIST_CHANNELS:
2963 case LTTNG_LIST_EVENTS:
2964 case LTTNG_LIST_SYSCALLS:
2965 case LTTNG_LIST_TRACKER_PIDS:
2966 case LTTNG_DATA_PENDING:
2967 break;
2968 default:
2969 /* Setup lttng message with no payload */
2970 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
2971 if (ret < 0) {
2972 /* This label does not try to unlock the session */
2973 goto init_setup_error;
2974 }
2975 }
2976
2977 /* Commands that DO NOT need a session. */
2978 switch (cmd_ctx->lsm->cmd_type) {
2979 case LTTNG_CREATE_SESSION:
2980 case LTTNG_CREATE_SESSION_SNAPSHOT:
2981 case LTTNG_CREATE_SESSION_LIVE:
2982 case LTTNG_LIST_SESSIONS:
2983 case LTTNG_LIST_TRACEPOINTS:
2984 case LTTNG_LIST_SYSCALLS:
2985 case LTTNG_LIST_TRACEPOINT_FIELDS:
2986 case LTTNG_SAVE_SESSION:
2987 case LTTNG_REGISTER_TRIGGER:
2988 case LTTNG_UNREGISTER_TRIGGER:
2989 need_tracing_session = 0;
2990 break;
2991 default:
2992 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2993 /*
2994 * We keep the session list lock across _all_ commands
2995 * for now, because the per-session lock does not
2996 * handle teardown properly.
2997 */
2998 session_lock_list();
2999 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3000 if (cmd_ctx->session == NULL) {
3001 ret = LTTNG_ERR_SESS_NOT_FOUND;
3002 goto error;
3003 } else {
3004 /* Acquire lock for the session */
3005 session_lock(cmd_ctx->session);
3006 }
3007 break;
3008 }
3009
3010 /*
3011 * Commands that need a valid session but should NOT create one if none
3012 * exists. Instead of creating one and destroying it when the command is
3013 * handled, process that right before so we save some round trip in useless
3014 * code path.
3015 */
3016 switch (cmd_ctx->lsm->cmd_type) {
3017 case LTTNG_DISABLE_CHANNEL:
3018 case LTTNG_DISABLE_EVENT:
3019 switch (cmd_ctx->lsm->domain.type) {
3020 case LTTNG_DOMAIN_KERNEL:
3021 if (!cmd_ctx->session->kernel_session) {
3022 ret = LTTNG_ERR_NO_CHANNEL;
3023 goto error;
3024 }
3025 break;
3026 case LTTNG_DOMAIN_JUL:
3027 case LTTNG_DOMAIN_LOG4J:
3028 case LTTNG_DOMAIN_PYTHON:
3029 case LTTNG_DOMAIN_UST:
3030 if (!cmd_ctx->session->ust_session) {
3031 ret = LTTNG_ERR_NO_CHANNEL;
3032 goto error;
3033 }
3034 break;
3035 default:
3036 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3037 goto error;
3038 }
3039 default:
3040 break;
3041 }
3042
3043 if (!need_domain) {
3044 goto skip_domain;
3045 }
3046
3047 /*
3048 * Check domain type for specific "pre-action".
3049 */
3050 switch (cmd_ctx->lsm->domain.type) {
3051 case LTTNG_DOMAIN_KERNEL:
3052 if (!is_root) {
3053 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3054 goto error;
3055 }
3056
3057 /* Kernel tracer check */
3058 if (kernel_tracer_fd == -1) {
3059 /* Basically, load kernel tracer modules */
3060 ret = init_kernel_tracer();
3061 if (ret != 0) {
3062 goto error;
3063 }
3064 }
3065
3066 /* Consumer is in an ERROR state. Report back to client */
3067 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3068 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3069 goto error;
3070 }
3071
3072 /* Need a session for kernel command */
3073 if (need_tracing_session) {
3074 if (cmd_ctx->session->kernel_session == NULL) {
3075 ret = create_kernel_session(cmd_ctx->session);
3076 if (ret < 0) {
3077 ret = LTTNG_ERR_KERN_SESS_FAIL;
3078 goto error;
3079 }
3080 }
3081
3082 /* Start the kernel consumer daemon */
3083 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3084 if (kconsumer_data.pid == 0 &&
3085 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3086 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3087 ret = start_consumerd(&kconsumer_data);
3088 if (ret < 0) {
3089 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3090 goto error;
3091 }
3092 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3093 } else {
3094 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3095 }
3096
3097 /*
3098 * The consumer was just spawned so we need to add the socket to
3099 * the consumer output of the session if exist.
3100 */
3101 ret = consumer_create_socket(&kconsumer_data,
3102 cmd_ctx->session->kernel_session->consumer);
3103 if (ret < 0) {
3104 goto error;
3105 }
3106 }
3107
3108 break;
3109 case LTTNG_DOMAIN_JUL:
3110 case LTTNG_DOMAIN_LOG4J:
3111 case LTTNG_DOMAIN_PYTHON:
3112 case LTTNG_DOMAIN_UST:
3113 {
3114 if (!ust_app_supported()) {
3115 ret = LTTNG_ERR_NO_UST;
3116 goto error;
3117 }
3118 /* Consumer is in an ERROR state. Report back to client */
3119 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
3120 ret = LTTNG_ERR_NO_USTCONSUMERD;
3121 goto error;
3122 }
3123
3124 if (need_tracing_session) {
3125 /* Create UST session if none exist. */
3126 if (cmd_ctx->session->ust_session == NULL) {
3127 ret = create_ust_session(cmd_ctx->session,
3128 &cmd_ctx->lsm->domain);
3129 if (ret != LTTNG_OK) {
3130 goto error;
3131 }
3132 }
3133
3134 /* Start the UST consumer daemons */
3135 /* 64-bit */
3136 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
3137 if (config.consumerd64_bin_path.value &&
3138 ustconsumer64_data.pid == 0 &&
3139 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3140 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3141 ret = start_consumerd(&ustconsumer64_data);
3142 if (ret < 0) {
3143 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
3144 uatomic_set(&ust_consumerd64_fd, -EINVAL);
3145 goto error;
3146 }
3147
3148 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
3149 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3150 } else {
3151 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
3152 }
3153
3154 /*
3155 * Setup socket for consumer 64 bit. No need for atomic access
3156 * since it was set above and can ONLY be set in this thread.
3157 */
3158 ret = consumer_create_socket(&ustconsumer64_data,
3159 cmd_ctx->session->ust_session->consumer);
3160 if (ret < 0) {
3161 goto error;
3162 }
3163
3164 /* 32-bit */
3165 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
3166 if (config.consumerd32_bin_path.value &&
3167 ustconsumer32_data.pid == 0 &&
3168 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3169 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3170 ret = start_consumerd(&ustconsumer32_data);
3171 if (ret < 0) {
3172 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
3173 uatomic_set(&ust_consumerd32_fd, -EINVAL);
3174 goto error;
3175 }
3176
3177 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
3178 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3179 } else {
3180 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
3181 }
3182
3183 /*
3184 * Setup socket for consumer 64 bit. No need for atomic access
3185 * since it was set above and can ONLY be set in this thread.
3186 */
3187 ret = consumer_create_socket(&ustconsumer32_data,
3188 cmd_ctx->session->ust_session->consumer);
3189 if (ret < 0) {
3190 goto error;
3191 }
3192 }
3193 break;
3194 }
3195 default:
3196 break;
3197 }
3198 skip_domain:
3199
3200 /* Validate consumer daemon state when start/stop trace command */
3201 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
3202 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
3203 switch (cmd_ctx->lsm->domain.type) {
3204 case LTTNG_DOMAIN_NONE:
3205 break;
3206 case LTTNG_DOMAIN_JUL:
3207 case LTTNG_DOMAIN_LOG4J:
3208 case LTTNG_DOMAIN_PYTHON:
3209 case LTTNG_DOMAIN_UST:
3210 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
3211 ret = LTTNG_ERR_NO_USTCONSUMERD;
3212 goto error;
3213 }
3214 break;
3215 case LTTNG_DOMAIN_KERNEL:
3216 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_S