f01ca2e84b1f6e63bc0070d73d15fa06c6b01043
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <paths.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <inttypes.h>
30 #include <sys/mman.h>
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
34 #include <sys/stat.h>
35 #include <sys/types.h>
36 #include <sys/wait.h>
37 #include <urcu/uatomic.h>
38 #include <unistd.h>
39 #include <config.h>
40
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/defaults.h>
44 #include <common/kernel-consumer/kernel-consumer.h>
45 #include <common/futex.h>
46 #include <common/relayd/relayd.h>
47 #include <common/utils.h>
48 #include <common/daemonize.h>
49
50 #include "lttng-sessiond.h"
51 #include "buffer-registry.h"
52 #include "channel.h"
53 #include "cmd.h"
54 #include "consumer.h"
55 #include "context.h"
56 #include "event.h"
57 #include "kernel.h"
58 #include "kernel-consumer.h"
59 #include "modprobe.h"
60 #include "shm.h"
61 #include "ust-ctl.h"
62 #include "ust-consumer.h"
63 #include "utils.h"
64 #include "fd-limit.h"
65 #include "health-sessiond.h"
66 #include "testpoint.h"
67 #include "ust-thread.h"
68 #include "jul-thread.h"
69
70 #define CONSUMERD_FILE "lttng-consumerd"
71
72 const char *progname;
73 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
74 static const char *opt_pidfile;
75 static int opt_sig_parent;
76 static int opt_verbose_consumer;
77 static int opt_daemon, opt_background;
78 static int opt_no_kernel;
79 static int is_root; /* Set to 1 if the daemon is running as root */
80 static pid_t ppid; /* Parent PID for --sig-parent option */
81 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
82 static char *rundir;
83
84 /* Set to 1 when a SIGUSR1 signal is received. */
85 static int recv_child_signal;
86
87 /*
88 * Consumer daemon specific control data. Every value not initialized here is
89 * set to 0 by the static definition.
90 */
91 static struct consumer_data kconsumer_data = {
92 .type = LTTNG_CONSUMER_KERNEL,
93 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
94 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
95 .err_sock = -1,
96 .cmd_sock = -1,
97 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
98 .lock = PTHREAD_MUTEX_INITIALIZER,
99 .cond = PTHREAD_COND_INITIALIZER,
100 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
101 };
102 static struct consumer_data ustconsumer64_data = {
103 .type = LTTNG_CONSUMER64_UST,
104 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
105 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
106 .err_sock = -1,
107 .cmd_sock = -1,
108 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
109 .lock = PTHREAD_MUTEX_INITIALIZER,
110 .cond = PTHREAD_COND_INITIALIZER,
111 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
112 };
113 static struct consumer_data ustconsumer32_data = {
114 .type = LTTNG_CONSUMER32_UST,
115 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
116 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
117 .err_sock = -1,
118 .cmd_sock = -1,
119 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
120 .lock = PTHREAD_MUTEX_INITIALIZER,
121 .cond = PTHREAD_COND_INITIALIZER,
122 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
123 };
124
125 /* Shared between threads */
126 static int dispatch_thread_exit;
127
128 /* Global application Unix socket path */
129 static char apps_unix_sock_path[PATH_MAX];
130 /* Global client Unix socket path */
131 static char client_unix_sock_path[PATH_MAX];
132 /* global wait shm path for UST */
133 static char wait_shm_path[PATH_MAX];
134 /* Global health check unix path */
135 static char health_unix_sock_path[PATH_MAX];
136
137 /* Sockets and FDs */
138 static int client_sock = -1;
139 static int apps_sock = -1;
140 int kernel_tracer_fd = -1;
141 static int kernel_poll_pipe[2] = { -1, -1 };
142
143 /*
144 * Quit pipe for all threads. This permits a single cancellation point
145 * for all threads when receiving an event on the pipe.
146 */
147 static int thread_quit_pipe[2] = { -1, -1 };
148
149 /*
150 * This pipe is used to inform the thread managing application communication
151 * that a command is queued and ready to be processed.
152 */
153 static int apps_cmd_pipe[2] = { -1, -1 };
154
155 int apps_cmd_notify_pipe[2] = { -1, -1 };
156
157 /* Pthread, Mutexes and Semaphores */
158 static pthread_t apps_thread;
159 static pthread_t apps_notify_thread;
160 static pthread_t reg_apps_thread;
161 static pthread_t client_thread;
162 static pthread_t kernel_thread;
163 static pthread_t dispatch_thread;
164 static pthread_t health_thread;
165 static pthread_t ht_cleanup_thread;
166 static pthread_t jul_reg_thread;
167
168 /*
169 * UST registration command queue. This queue is tied with a futex and uses a N
170 * wakers / 1 waiter implemented and detailed in futex.c/.h
171 *
172 * The thread_manage_apps and thread_dispatch_ust_registration interact with
173 * this queue and the wait/wake scheme.
174 */
175 static struct ust_cmd_queue ust_cmd_queue;
176
177 /*
178 * Pointer initialized before thread creation.
179 *
180 * This points to the tracing session list containing the session count and a
181 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
182 * MUST NOT be taken if you call a public function in session.c.
183 *
184 * The lock is nested inside the structure: session_list_ptr->lock. Please use
185 * session_lock_list and session_unlock_list for lock acquisition.
186 */
187 static struct ltt_session_list *session_list_ptr;
188
189 int ust_consumerd64_fd = -1;
190 int ust_consumerd32_fd = -1;
191
192 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
193 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
194 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
195 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
196
197 static const char *module_proc_lttng = "/proc/lttng";
198
199 /*
200 * Consumer daemon state which is changed when spawning it, killing it or in
201 * case of a fatal error.
202 */
203 enum consumerd_state {
204 CONSUMER_STARTED = 1,
205 CONSUMER_STOPPED = 2,
206 CONSUMER_ERROR = 3,
207 };
208
209 /*
210 * This consumer daemon state is used to validate if a client command will be
211 * able to reach the consumer. If not, the client is informed. For instance,
212 * doing a "lttng start" when the consumer state is set to ERROR will return an
213 * error to the client.
214 *
215 * The following example shows a possible race condition of this scheme:
216 *
217 * consumer thread error happens
218 * client cmd arrives
219 * client cmd checks state -> still OK
220 * consumer thread exit, sets error
221 * client cmd try to talk to consumer
222 * ...
223 *
224 * However, since the consumer is a different daemon, we have no way of making
225 * sure the command will reach it safely even with this state flag. This is why
226 * we consider that up to the state validation during command processing, the
227 * command is safe. After that, we can not guarantee the correctness of the
228 * client request vis-a-vis the consumer.
229 */
230 static enum consumerd_state ust_consumerd_state;
231 static enum consumerd_state kernel_consumerd_state;
232
233 /*
234 * Socket timeout for receiving and sending in seconds.
235 */
236 static int app_socket_timeout;
237
238 /* Set in main() with the current page size. */
239 long page_size;
240
241 /* Application health monitoring */
242 struct health_app *health_sessiond;
243
244 /* JUL TCP port for registration. Used by the JUL thread. */
245 unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
246
247 /*
248 * Whether sessiond is ready for commands/health check requests.
249 * NR_LTTNG_SESSIOND_READY must match the number of calls to
250 * lttng_sessiond_notify_ready().
251 */
252 #define NR_LTTNG_SESSIOND_READY 2
253 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
254
255 /* Notify parents that we are ready for cmd and health check */
256 static
257 void lttng_sessiond_notify_ready(void)
258 {
259 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
260 /*
261 * Notify parent pid that we are ready to accept command
262 * for client side. This ppid is the one from the
263 * external process that spawned us.
264 */
265 if (opt_sig_parent) {
266 kill(ppid, SIGUSR1);
267 }
268
269 /*
270 * Notify the parent of the fork() process that we are
271 * ready.
272 */
273 if (opt_daemon || opt_background) {
274 kill(child_ppid, SIGUSR1);
275 }
276 }
277 }
278
279 static
280 void setup_consumerd_path(void)
281 {
282 const char *bin, *libdir;
283
284 /*
285 * Allow INSTALL_BIN_PATH to be used as a target path for the
286 * native architecture size consumer if CONFIG_CONSUMER*_PATH
287 * has not been defined.
288 */
289 #if (CAA_BITS_PER_LONG == 32)
290 if (!consumerd32_bin[0]) {
291 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
292 }
293 if (!consumerd32_libdir[0]) {
294 consumerd32_libdir = INSTALL_LIB_PATH;
295 }
296 #elif (CAA_BITS_PER_LONG == 64)
297 if (!consumerd64_bin[0]) {
298 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
299 }
300 if (!consumerd64_libdir[0]) {
301 consumerd64_libdir = INSTALL_LIB_PATH;
302 }
303 #else
304 #error "Unknown bitness"
305 #endif
306
307 /*
308 * runtime env. var. overrides the build default.
309 */
310 bin = getenv("LTTNG_CONSUMERD32_BIN");
311 if (bin) {
312 consumerd32_bin = bin;
313 }
314 bin = getenv("LTTNG_CONSUMERD64_BIN");
315 if (bin) {
316 consumerd64_bin = bin;
317 }
318 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
319 if (libdir) {
320 consumerd32_libdir = libdir;
321 }
322 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
323 if (libdir) {
324 consumerd64_libdir = libdir;
325 }
326 }
327
328 /*
329 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
330 */
331 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
332 {
333 int ret;
334
335 assert(events);
336
337 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
338 if (ret < 0) {
339 goto error;
340 }
341
342 /* Add quit pipe */
343 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
344 if (ret < 0) {
345 goto error;
346 }
347
348 return 0;
349
350 error:
351 return ret;
352 }
353
354 /*
355 * Check if the thread quit pipe was triggered.
356 *
357 * Return 1 if it was triggered else 0;
358 */
359 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
360 {
361 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
362 return 1;
363 }
364
365 return 0;
366 }
367
368 /*
369 * Init thread quit pipe.
370 *
371 * Return -1 on error or 0 if all pipes are created.
372 */
373 static int init_thread_quit_pipe(void)
374 {
375 int ret, i;
376
377 ret = pipe(thread_quit_pipe);
378 if (ret < 0) {
379 PERROR("thread quit pipe");
380 goto error;
381 }
382
383 for (i = 0; i < 2; i++) {
384 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
385 if (ret < 0) {
386 PERROR("fcntl");
387 goto error;
388 }
389 }
390
391 error:
392 return ret;
393 }
394
395 /*
396 * Stop all threads by closing the thread quit pipe.
397 */
398 static void stop_threads(void)
399 {
400 int ret;
401
402 /* Stopping all threads */
403 DBG("Terminating all threads");
404 ret = notify_thread_pipe(thread_quit_pipe[1]);
405 if (ret < 0) {
406 ERR("write error on thread quit pipe");
407 }
408
409 /* Dispatch thread */
410 CMM_STORE_SHARED(dispatch_thread_exit, 1);
411 futex_nto1_wake(&ust_cmd_queue.futex);
412 }
413
414 /*
415 * Close every consumer sockets.
416 */
417 static void close_consumer_sockets(void)
418 {
419 int ret;
420
421 if (kconsumer_data.err_sock >= 0) {
422 ret = close(kconsumer_data.err_sock);
423 if (ret < 0) {
424 PERROR("kernel consumer err_sock close");
425 }
426 }
427 if (ustconsumer32_data.err_sock >= 0) {
428 ret = close(ustconsumer32_data.err_sock);
429 if (ret < 0) {
430 PERROR("UST consumerd32 err_sock close");
431 }
432 }
433 if (ustconsumer64_data.err_sock >= 0) {
434 ret = close(ustconsumer64_data.err_sock);
435 if (ret < 0) {
436 PERROR("UST consumerd64 err_sock close");
437 }
438 }
439 if (kconsumer_data.cmd_sock >= 0) {
440 ret = close(kconsumer_data.cmd_sock);
441 if (ret < 0) {
442 PERROR("kernel consumer cmd_sock close");
443 }
444 }
445 if (ustconsumer32_data.cmd_sock >= 0) {
446 ret = close(ustconsumer32_data.cmd_sock);
447 if (ret < 0) {
448 PERROR("UST consumerd32 cmd_sock close");
449 }
450 }
451 if (ustconsumer64_data.cmd_sock >= 0) {
452 ret = close(ustconsumer64_data.cmd_sock);
453 if (ret < 0) {
454 PERROR("UST consumerd64 cmd_sock close");
455 }
456 }
457 }
458
459 /*
460 * Cleanup the daemon
461 */
462 static void cleanup(void)
463 {
464 int ret;
465 struct ltt_session *sess, *stmp;
466 char path[PATH_MAX];
467
468 DBG("Cleaning up");
469
470 /*
471 * Close the thread quit pipe. It has already done its job,
472 * since we are now called.
473 */
474 utils_close_pipe(thread_quit_pipe);
475
476 /*
477 * If opt_pidfile is undefined, the default file will be wiped when
478 * removing the rundir.
479 */
480 if (opt_pidfile) {
481 ret = remove(opt_pidfile);
482 if (ret < 0) {
483 PERROR("remove pidfile %s", opt_pidfile);
484 }
485 }
486
487 DBG("Removing sessiond and consumerd content of directory %s", rundir);
488
489 /* sessiond */
490 snprintf(path, PATH_MAX,
491 "%s/%s",
492 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
493 DBG("Removing %s", path);
494 (void) unlink(path);
495
496 snprintf(path, PATH_MAX, "%s/%s", rundir,
497 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
498 DBG("Removing %s", path);
499 (void) unlink(path);
500
501 /* kconsumerd */
502 snprintf(path, PATH_MAX,
503 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
504 rundir);
505 DBG("Removing %s", path);
506 (void) unlink(path);
507
508 snprintf(path, PATH_MAX,
509 DEFAULT_KCONSUMERD_PATH,
510 rundir);
511 DBG("Removing directory %s", path);
512 (void) rmdir(path);
513
514 /* ust consumerd 32 */
515 snprintf(path, PATH_MAX,
516 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
517 rundir);
518 DBG("Removing %s", path);
519 (void) unlink(path);
520
521 snprintf(path, PATH_MAX,
522 DEFAULT_USTCONSUMERD32_PATH,
523 rundir);
524 DBG("Removing directory %s", path);
525 (void) rmdir(path);
526
527 /* ust consumerd 64 */
528 snprintf(path, PATH_MAX,
529 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
530 rundir);
531 DBG("Removing %s", path);
532 (void) unlink(path);
533
534 snprintf(path, PATH_MAX,
535 DEFAULT_USTCONSUMERD64_PATH,
536 rundir);
537 DBG("Removing directory %s", path);
538 (void) rmdir(path);
539
540 /*
541 * We do NOT rmdir rundir because there are other processes
542 * using it, for instance lttng-relayd, which can start in
543 * parallel with this teardown.
544 */
545
546 free(rundir);
547
548 DBG("Cleaning up all sessions");
549
550 /* Destroy session list mutex */
551 if (session_list_ptr != NULL) {
552 pthread_mutex_destroy(&session_list_ptr->lock);
553
554 /* Cleanup ALL session */
555 cds_list_for_each_entry_safe(sess, stmp,
556 &session_list_ptr->head, list) {
557 cmd_destroy_session(sess, kernel_poll_pipe[1]);
558 }
559 }
560
561 DBG("Closing all UST sockets");
562 ust_app_clean_list();
563 buffer_reg_destroy_registries();
564
565 if (is_root && !opt_no_kernel) {
566 DBG2("Closing kernel fd");
567 if (kernel_tracer_fd >= 0) {
568 ret = close(kernel_tracer_fd);
569 if (ret) {
570 PERROR("close");
571 }
572 }
573 DBG("Unloading kernel modules");
574 modprobe_remove_lttng_all();
575 }
576
577 close_consumer_sockets();
578
579 /* <fun> */
580 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
581 "Matthew, BEET driven development works!%c[%dm",
582 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
583 /* </fun> */
584 }
585
586 /*
587 * Send data on a unix socket using the liblttsessiondcomm API.
588 *
589 * Return lttcomm error code.
590 */
591 static int send_unix_sock(int sock, void *buf, size_t len)
592 {
593 /* Check valid length */
594 if (len == 0) {
595 return -1;
596 }
597
598 return lttcomm_send_unix_sock(sock, buf, len);
599 }
600
601 /*
602 * Free memory of a command context structure.
603 */
604 static void clean_command_ctx(struct command_ctx **cmd_ctx)
605 {
606 DBG("Clean command context structure");
607 if (*cmd_ctx) {
608 if ((*cmd_ctx)->llm) {
609 free((*cmd_ctx)->llm);
610 }
611 if ((*cmd_ctx)->lsm) {
612 free((*cmd_ctx)->lsm);
613 }
614 free(*cmd_ctx);
615 *cmd_ctx = NULL;
616 }
617 }
618
619 /*
620 * Notify UST applications using the shm mmap futex.
621 */
622 static int notify_ust_apps(int active)
623 {
624 char *wait_shm_mmap;
625
626 DBG("Notifying applications of session daemon state: %d", active);
627
628 /* See shm.c for this call implying mmap, shm and futex calls */
629 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
630 if (wait_shm_mmap == NULL) {
631 goto error;
632 }
633
634 /* Wake waiting process */
635 futex_wait_update((int32_t *) wait_shm_mmap, active);
636
637 /* Apps notified successfully */
638 return 0;
639
640 error:
641 return -1;
642 }
643
644 /*
645 * Setup the outgoing data buffer for the response (llm) by allocating the
646 * right amount of memory and copying the original information from the lsm
647 * structure.
648 *
649 * Return total size of the buffer pointed by buf.
650 */
651 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
652 {
653 int ret, buf_size;
654
655 buf_size = size;
656
657 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
658 if (cmd_ctx->llm == NULL) {
659 PERROR("zmalloc");
660 ret = -ENOMEM;
661 goto error;
662 }
663
664 /* Copy common data */
665 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
666 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
667
668 cmd_ctx->llm->data_size = size;
669 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
670
671 return buf_size;
672
673 error:
674 return ret;
675 }
676
677 /*
678 * Update the kernel poll set of all channel fd available over all tracing
679 * session. Add the wakeup pipe at the end of the set.
680 */
681 static int update_kernel_poll(struct lttng_poll_event *events)
682 {
683 int ret;
684 struct ltt_session *session;
685 struct ltt_kernel_channel *channel;
686
687 DBG("Updating kernel poll set");
688
689 session_lock_list();
690 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
691 session_lock(session);
692 if (session->kernel_session == NULL) {
693 session_unlock(session);
694 continue;
695 }
696
697 cds_list_for_each_entry(channel,
698 &session->kernel_session->channel_list.head, list) {
699 /* Add channel fd to the kernel poll set */
700 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
701 if (ret < 0) {
702 session_unlock(session);
703 goto error;
704 }
705 DBG("Channel fd %d added to kernel set", channel->fd);
706 }
707 session_unlock(session);
708 }
709 session_unlock_list();
710
711 return 0;
712
713 error:
714 session_unlock_list();
715 return -1;
716 }
717
718 /*
719 * Find the channel fd from 'fd' over all tracing session. When found, check
720 * for new channel stream and send those stream fds to the kernel consumer.
721 *
722 * Useful for CPU hotplug feature.
723 */
724 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
725 {
726 int ret = 0;
727 struct ltt_session *session;
728 struct ltt_kernel_session *ksess;
729 struct ltt_kernel_channel *channel;
730
731 DBG("Updating kernel streams for channel fd %d", fd);
732
733 session_lock_list();
734 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
735 session_lock(session);
736 if (session->kernel_session == NULL) {
737 session_unlock(session);
738 continue;
739 }
740 ksess = session->kernel_session;
741
742 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
743 if (channel->fd == fd) {
744 DBG("Channel found, updating kernel streams");
745 ret = kernel_open_channel_stream(channel);
746 if (ret < 0) {
747 goto error;
748 }
749 /* Update the stream global counter */
750 ksess->stream_count_global += ret;
751
752 /*
753 * Have we already sent fds to the consumer? If yes, it means
754 * that tracing is started so it is safe to send our updated
755 * stream fds.
756 */
757 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
758 struct lttng_ht_iter iter;
759 struct consumer_socket *socket;
760
761 rcu_read_lock();
762 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
763 &iter.iter, socket, node.node) {
764 pthread_mutex_lock(socket->lock);
765 ret = kernel_consumer_send_channel_stream(socket,
766 channel, ksess,
767 session->output_traces ? 1 : 0);
768 pthread_mutex_unlock(socket->lock);
769 if (ret < 0) {
770 rcu_read_unlock();
771 goto error;
772 }
773 }
774 rcu_read_unlock();
775 }
776 goto error;
777 }
778 }
779 session_unlock(session);
780 }
781 session_unlock_list();
782 return ret;
783
784 error:
785 session_unlock(session);
786 session_unlock_list();
787 return ret;
788 }
789
790 /*
791 * For each tracing session, update newly registered apps. The session list
792 * lock MUST be acquired before calling this.
793 */
794 static void update_ust_app(int app_sock)
795 {
796 struct ltt_session *sess, *stmp;
797
798 /* Consumer is in an ERROR state. Stop any application update. */
799 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
800 /* Stop the update process since the consumer is dead. */
801 return;
802 }
803
804 /* For all tracing session(s) */
805 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
806 session_lock(sess);
807 if (sess->ust_session) {
808 ust_app_global_update(sess->ust_session, app_sock);
809 }
810 session_unlock(sess);
811 }
812 }
813
814 /*
815 * This thread manage event coming from the kernel.
816 *
817 * Features supported in this thread:
818 * -) CPU Hotplug
819 */
820 static void *thread_manage_kernel(void *data)
821 {
822 int ret, i, pollfd, update_poll_flag = 1, err = -1;
823 uint32_t revents, nb_fd;
824 char tmp;
825 struct lttng_poll_event events;
826
827 DBG("[thread] Thread manage kernel started");
828
829 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
830
831 /*
832 * This first step of the while is to clean this structure which could free
833 * non NULL pointers so initialize it before the loop.
834 */
835 lttng_poll_init(&events);
836
837 if (testpoint(sessiond_thread_manage_kernel)) {
838 goto error_testpoint;
839 }
840
841 health_code_update();
842
843 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
844 goto error_testpoint;
845 }
846
847 while (1) {
848 health_code_update();
849
850 if (update_poll_flag == 1) {
851 /* Clean events object. We are about to populate it again. */
852 lttng_poll_clean(&events);
853
854 ret = sessiond_set_thread_pollset(&events, 2);
855 if (ret < 0) {
856 goto error_poll_create;
857 }
858
859 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
860 if (ret < 0) {
861 goto error;
862 }
863
864 /* This will add the available kernel channel if any. */
865 ret = update_kernel_poll(&events);
866 if (ret < 0) {
867 goto error;
868 }
869 update_poll_flag = 0;
870 }
871
872 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
873
874 /* Poll infinite value of time */
875 restart:
876 health_poll_entry();
877 ret = lttng_poll_wait(&events, -1);
878 health_poll_exit();
879 if (ret < 0) {
880 /*
881 * Restart interrupted system call.
882 */
883 if (errno == EINTR) {
884 goto restart;
885 }
886 goto error;
887 } else if (ret == 0) {
888 /* Should not happen since timeout is infinite */
889 ERR("Return value of poll is 0 with an infinite timeout.\n"
890 "This should not have happened! Continuing...");
891 continue;
892 }
893
894 nb_fd = ret;
895
896 for (i = 0; i < nb_fd; i++) {
897 /* Fetch once the poll data */
898 revents = LTTNG_POLL_GETEV(&events, i);
899 pollfd = LTTNG_POLL_GETFD(&events, i);
900
901 health_code_update();
902
903 /* Thread quit pipe has been closed. Killing thread. */
904 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
905 if (ret) {
906 err = 0;
907 goto exit;
908 }
909
910 /* Check for data on kernel pipe */
911 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
912 (void) lttng_read(kernel_poll_pipe[0],
913 &tmp, 1);
914 /*
915 * Ret value is useless here, if this pipe gets any actions an
916 * update is required anyway.
917 */
918 update_poll_flag = 1;
919 continue;
920 } else {
921 /*
922 * New CPU detected by the kernel. Adding kernel stream to
923 * kernel session and updating the kernel consumer
924 */
925 if (revents & LPOLLIN) {
926 ret = update_kernel_stream(&kconsumer_data, pollfd);
927 if (ret < 0) {
928 continue;
929 }
930 break;
931 /*
932 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
933 * and unregister kernel stream at this point.
934 */
935 }
936 }
937 }
938 }
939
940 exit:
941 error:
942 lttng_poll_clean(&events);
943 error_poll_create:
944 error_testpoint:
945 utils_close_pipe(kernel_poll_pipe);
946 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
947 if (err) {
948 health_error();
949 ERR("Health error occurred in %s", __func__);
950 WARN("Kernel thread died unexpectedly. "
951 "Kernel tracing can continue but CPU hotplug is disabled.");
952 }
953 health_unregister(health_sessiond);
954 DBG("Kernel thread dying");
955 return NULL;
956 }
957
958 /*
959 * Signal pthread condition of the consumer data that the thread.
960 */
961 static void signal_consumer_condition(struct consumer_data *data, int state)
962 {
963 pthread_mutex_lock(&data->cond_mutex);
964
965 /*
966 * The state is set before signaling. It can be any value, it's the waiter
967 * job to correctly interpret this condition variable associated to the
968 * consumer pthread_cond.
969 *
970 * A value of 0 means that the corresponding thread of the consumer data
971 * was not started. 1 indicates that the thread has started and is ready
972 * for action. A negative value means that there was an error during the
973 * thread bootstrap.
974 */
975 data->consumer_thread_is_ready = state;
976 (void) pthread_cond_signal(&data->cond);
977
978 pthread_mutex_unlock(&data->cond_mutex);
979 }
980
981 /*
982 * This thread manage the consumer error sent back to the session daemon.
983 */
984 static void *thread_manage_consumer(void *data)
985 {
986 int sock = -1, i, ret, pollfd, err = -1;
987 uint32_t revents, nb_fd;
988 enum lttcomm_return_code code;
989 struct lttng_poll_event events;
990 struct consumer_data *consumer_data = data;
991
992 DBG("[thread] Manage consumer started");
993
994 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
995
996 health_code_update();
997
998 /*
999 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1000 * metadata_sock. Nothing more will be added to this poll set.
1001 */
1002 ret = sessiond_set_thread_pollset(&events, 3);
1003 if (ret < 0) {
1004 goto error_poll;
1005 }
1006
1007 /*
1008 * The error socket here is already in a listening state which was done
1009 * just before spawning this thread to avoid a race between the consumer
1010 * daemon exec trying to connect and the listen() call.
1011 */
1012 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1013 if (ret < 0) {
1014 goto error;
1015 }
1016
1017 health_code_update();
1018
1019 /* Infinite blocking call, waiting for transmission */
1020 restart:
1021 health_poll_entry();
1022
1023 if (testpoint(sessiond_thread_manage_consumer)) {
1024 goto error;
1025 }
1026
1027 ret = lttng_poll_wait(&events, -1);
1028 health_poll_exit();
1029 if (ret < 0) {
1030 /*
1031 * Restart interrupted system call.
1032 */
1033 if (errno == EINTR) {
1034 goto restart;
1035 }
1036 goto error;
1037 }
1038
1039 nb_fd = ret;
1040
1041 for (i = 0; i < nb_fd; i++) {
1042 /* Fetch once the poll data */
1043 revents = LTTNG_POLL_GETEV(&events, i);
1044 pollfd = LTTNG_POLL_GETFD(&events, i);
1045
1046 health_code_update();
1047
1048 /* Thread quit pipe has been closed. Killing thread. */
1049 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1050 if (ret) {
1051 err = 0;
1052 goto exit;
1053 }
1054
1055 /* Event on the registration socket */
1056 if (pollfd == consumer_data->err_sock) {
1057 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1058 ERR("consumer err socket poll error");
1059 goto error;
1060 }
1061 }
1062 }
1063
1064 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1065 if (sock < 0) {
1066 goto error;
1067 }
1068
1069 /*
1070 * Set the CLOEXEC flag. Return code is useless because either way, the
1071 * show must go on.
1072 */
1073 (void) utils_set_fd_cloexec(sock);
1074
1075 health_code_update();
1076
1077 DBG2("Receiving code from consumer err_sock");
1078
1079 /* Getting status code from kconsumerd */
1080 ret = lttcomm_recv_unix_sock(sock, &code,
1081 sizeof(enum lttcomm_return_code));
1082 if (ret <= 0) {
1083 goto error;
1084 }
1085
1086 health_code_update();
1087
1088 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1089 /* Connect both socket, command and metadata. */
1090 consumer_data->cmd_sock =
1091 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1092 consumer_data->metadata_fd =
1093 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1094 if (consumer_data->cmd_sock < 0
1095 || consumer_data->metadata_fd < 0) {
1096 PERROR("consumer connect cmd socket");
1097 /* On error, signal condition and quit. */
1098 signal_consumer_condition(consumer_data, -1);
1099 goto error;
1100 }
1101 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1102 /* Create metadata socket lock. */
1103 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1104 if (consumer_data->metadata_sock.lock == NULL) {
1105 PERROR("zmalloc pthread mutex");
1106 ret = -1;
1107 goto error;
1108 }
1109 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1110
1111 signal_consumer_condition(consumer_data, 1);
1112 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1113 DBG("Consumer metadata socket ready (fd: %d)",
1114 consumer_data->metadata_fd);
1115 } else {
1116 ERR("consumer error when waiting for SOCK_READY : %s",
1117 lttcomm_get_readable_code(-code));
1118 goto error;
1119 }
1120
1121 /* Remove the consumerd error sock since we've established a connexion */
1122 ret = lttng_poll_del(&events, consumer_data->err_sock);
1123 if (ret < 0) {
1124 goto error;
1125 }
1126
1127 /* Add new accepted error socket. */
1128 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1129 if (ret < 0) {
1130 goto error;
1131 }
1132
1133 /* Add metadata socket that is successfully connected. */
1134 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1135 LPOLLIN | LPOLLRDHUP);
1136 if (ret < 0) {
1137 goto error;
1138 }
1139
1140 health_code_update();
1141
1142 /* Infinite blocking call, waiting for transmission */
1143 restart_poll:
1144 while (1) {
1145 health_poll_entry();
1146 ret = lttng_poll_wait(&events, -1);
1147 health_poll_exit();
1148 if (ret < 0) {
1149 /*
1150 * Restart interrupted system call.
1151 */
1152 if (errno == EINTR) {
1153 goto restart_poll;
1154 }
1155 goto error;
1156 }
1157
1158 nb_fd = ret;
1159
1160 for (i = 0; i < nb_fd; i++) {
1161 /* Fetch once the poll data */
1162 revents = LTTNG_POLL_GETEV(&events, i);
1163 pollfd = LTTNG_POLL_GETFD(&events, i);
1164
1165 health_code_update();
1166
1167 /* Thread quit pipe has been closed. Killing thread. */
1168 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1169 if (ret) {
1170 err = 0;
1171 goto exit;
1172 }
1173
1174 if (pollfd == sock) {
1175 /* Event on the consumerd socket */
1176 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1177 ERR("consumer err socket second poll error");
1178 goto error;
1179 }
1180 health_code_update();
1181 /* Wait for any kconsumerd error */
1182 ret = lttcomm_recv_unix_sock(sock, &code,
1183 sizeof(enum lttcomm_return_code));
1184 if (ret <= 0) {
1185 ERR("consumer closed the command socket");
1186 goto error;
1187 }
1188
1189 ERR("consumer return code : %s",
1190 lttcomm_get_readable_code(-code));
1191
1192 goto exit;
1193 } else if (pollfd == consumer_data->metadata_fd) {
1194 /* UST metadata requests */
1195 ret = ust_consumer_metadata_request(
1196 &consumer_data->metadata_sock);
1197 if (ret < 0) {
1198 ERR("Handling metadata request");
1199 goto error;
1200 }
1201 break;
1202 } else {
1203 ERR("Unknown pollfd");
1204 goto error;
1205 }
1206 }
1207 health_code_update();
1208 }
1209
1210 exit:
1211 error:
1212 /*
1213 * We lock here because we are about to close the sockets and some other
1214 * thread might be using them so get exclusive access which will abort all
1215 * other consumer command by other threads.
1216 */
1217 pthread_mutex_lock(&consumer_data->lock);
1218
1219 /* Immediately set the consumerd state to stopped */
1220 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1221 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1222 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1223 consumer_data->type == LTTNG_CONSUMER32_UST) {
1224 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1225 } else {
1226 /* Code flow error... */
1227 assert(0);
1228 }
1229
1230 if (consumer_data->err_sock >= 0) {
1231 ret = close(consumer_data->err_sock);
1232 if (ret) {
1233 PERROR("close");
1234 }
1235 consumer_data->err_sock = -1;
1236 }
1237 if (consumer_data->cmd_sock >= 0) {
1238 ret = close(consumer_data->cmd_sock);
1239 if (ret) {
1240 PERROR("close");
1241 }
1242 consumer_data->cmd_sock = -1;
1243 }
1244 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1245 ret = close(*consumer_data->metadata_sock.fd_ptr);
1246 if (ret) {
1247 PERROR("close");
1248 }
1249 }
1250
1251 if (sock >= 0) {
1252 ret = close(sock);
1253 if (ret) {
1254 PERROR("close");
1255 }
1256 }
1257
1258 unlink(consumer_data->err_unix_sock_path);
1259 unlink(consumer_data->cmd_unix_sock_path);
1260 consumer_data->pid = 0;
1261 pthread_mutex_unlock(&consumer_data->lock);
1262
1263 /* Cleanup metadata socket mutex. */
1264 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1265 free(consumer_data->metadata_sock.lock);
1266
1267 lttng_poll_clean(&events);
1268 error_poll:
1269 if (err) {
1270 health_error();
1271 ERR("Health error occurred in %s", __func__);
1272 }
1273 health_unregister(health_sessiond);
1274 DBG("consumer thread cleanup completed");
1275
1276 return NULL;
1277 }
1278
1279 /*
1280 * This thread manage application communication.
1281 */
1282 static void *thread_manage_apps(void *data)
1283 {
1284 int i, ret, pollfd, err = -1;
1285 ssize_t size_ret;
1286 uint32_t revents, nb_fd;
1287 struct lttng_poll_event events;
1288
1289 DBG("[thread] Manage application started");
1290
1291 rcu_register_thread();
1292 rcu_thread_online();
1293
1294 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1295
1296 if (testpoint(sessiond_thread_manage_apps)) {
1297 goto error_testpoint;
1298 }
1299
1300 health_code_update();
1301
1302 ret = sessiond_set_thread_pollset(&events, 2);
1303 if (ret < 0) {
1304 goto error_poll_create;
1305 }
1306
1307 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1308 if (ret < 0) {
1309 goto error;
1310 }
1311
1312 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1313 goto error;
1314 }
1315
1316 health_code_update();
1317
1318 while (1) {
1319 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1320
1321 /* Inifinite blocking call, waiting for transmission */
1322 restart:
1323 health_poll_entry();
1324 ret = lttng_poll_wait(&events, -1);
1325 health_poll_exit();
1326 if (ret < 0) {
1327 /*
1328 * Restart interrupted system call.
1329 */
1330 if (errno == EINTR) {
1331 goto restart;
1332 }
1333 goto error;
1334 }
1335
1336 nb_fd = ret;
1337
1338 for (i = 0; i < nb_fd; i++) {
1339 /* Fetch once the poll data */
1340 revents = LTTNG_POLL_GETEV(&events, i);
1341 pollfd = LTTNG_POLL_GETFD(&events, i);
1342
1343 health_code_update();
1344
1345 /* Thread quit pipe has been closed. Killing thread. */
1346 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1347 if (ret) {
1348 err = 0;
1349 goto exit;
1350 }
1351
1352 /* Inspect the apps cmd pipe */
1353 if (pollfd == apps_cmd_pipe[0]) {
1354 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1355 ERR("Apps command pipe error");
1356 goto error;
1357 } else if (revents & LPOLLIN) {
1358 int sock;
1359
1360 /* Empty pipe */
1361 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1362 if (size_ret < sizeof(sock)) {
1363 PERROR("read apps cmd pipe");
1364 goto error;
1365 }
1366
1367 health_code_update();
1368
1369 /*
1370 * We only monitor the error events of the socket. This
1371 * thread does not handle any incoming data from UST
1372 * (POLLIN).
1373 */
1374 ret = lttng_poll_add(&events, sock,
1375 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1376 if (ret < 0) {
1377 goto error;
1378 }
1379
1380 DBG("Apps with sock %d added to poll set", sock);
1381 }
1382 } else {
1383 /*
1384 * At this point, we know that a registered application made
1385 * the event at poll_wait.
1386 */
1387 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1388 /* Removing from the poll set */
1389 ret = lttng_poll_del(&events, pollfd);
1390 if (ret < 0) {
1391 goto error;
1392 }
1393
1394 /* Socket closed on remote end. */
1395 ust_app_unregister(pollfd);
1396 }
1397 }
1398
1399 health_code_update();
1400 }
1401 }
1402
1403 exit:
1404 error:
1405 lttng_poll_clean(&events);
1406 error_poll_create:
1407 error_testpoint:
1408 utils_close_pipe(apps_cmd_pipe);
1409 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1410
1411 /*
1412 * We don't clean the UST app hash table here since already registered
1413 * applications can still be controlled so let them be until the session
1414 * daemon dies or the applications stop.
1415 */
1416
1417 if (err) {
1418 health_error();
1419 ERR("Health error occurred in %s", __func__);
1420 }
1421 health_unregister(health_sessiond);
1422 DBG("Application communication apps thread cleanup complete");
1423 rcu_thread_offline();
1424 rcu_unregister_thread();
1425 return NULL;
1426 }
1427
1428 /*
1429 * Send a socket to a thread This is called from the dispatch UST registration
1430 * thread once all sockets are set for the application.
1431 *
1432 * The sock value can be invalid, we don't really care, the thread will handle
1433 * it and make the necessary cleanup if so.
1434 *
1435 * On success, return 0 else a negative value being the errno message of the
1436 * write().
1437 */
1438 static int send_socket_to_thread(int fd, int sock)
1439 {
1440 ssize_t ret;
1441
1442 /*
1443 * It's possible that the FD is set as invalid with -1 concurrently just
1444 * before calling this function being a shutdown state of the thread.
1445 */
1446 if (fd < 0) {
1447 ret = -EBADF;
1448 goto error;
1449 }
1450
1451 ret = lttng_write(fd, &sock, sizeof(sock));
1452 if (ret < sizeof(sock)) {
1453 PERROR("write apps pipe %d", fd);
1454 if (ret < 0) {
1455 ret = -errno;
1456 }
1457 goto error;
1458 }
1459
1460 /* All good. Don't send back the write positive ret value. */
1461 ret = 0;
1462 error:
1463 return (int) ret;
1464 }
1465
1466 /*
1467 * Sanitize the wait queue of the dispatch registration thread meaning removing
1468 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1469 * notify socket is never received.
1470 */
1471 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1472 {
1473 int ret, nb_fd = 0, i;
1474 unsigned int fd_added = 0;
1475 struct lttng_poll_event events;
1476 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1477
1478 assert(wait_queue);
1479
1480 lttng_poll_init(&events);
1481
1482 /* Just skip everything for an empty queue. */
1483 if (!wait_queue->count) {
1484 goto end;
1485 }
1486
1487 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1488 if (ret < 0) {
1489 goto error_create;
1490 }
1491
1492 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1493 &wait_queue->head, head) {
1494 assert(wait_node->app);
1495 ret = lttng_poll_add(&events, wait_node->app->sock,
1496 LPOLLHUP | LPOLLERR);
1497 if (ret < 0) {
1498 goto error;
1499 }
1500
1501 fd_added = 1;
1502 }
1503
1504 if (!fd_added) {
1505 goto end;
1506 }
1507
1508 /*
1509 * Poll but don't block so we can quickly identify the faulty events and
1510 * clean them afterwards from the wait queue.
1511 */
1512 ret = lttng_poll_wait(&events, 0);
1513 if (ret < 0) {
1514 goto error;
1515 }
1516 nb_fd = ret;
1517
1518 for (i = 0; i < nb_fd; i++) {
1519 /* Get faulty FD. */
1520 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1521 int pollfd = LTTNG_POLL_GETFD(&events, i);
1522
1523 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1524 &wait_queue->head, head) {
1525 if (pollfd == wait_node->app->sock &&
1526 (revents & (LPOLLHUP | LPOLLERR))) {
1527 cds_list_del(&wait_node->head);
1528 wait_queue->count--;
1529 ust_app_destroy(wait_node->app);
1530 free(wait_node);
1531 break;
1532 }
1533 }
1534 }
1535
1536 if (nb_fd > 0) {
1537 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1538 }
1539
1540 end:
1541 lttng_poll_clean(&events);
1542 return;
1543
1544 error:
1545 lttng_poll_clean(&events);
1546 error_create:
1547 ERR("Unable to sanitize wait queue");
1548 return;
1549 }
1550
1551 /*
1552 * Dispatch request from the registration threads to the application
1553 * communication thread.
1554 */
1555 static void *thread_dispatch_ust_registration(void *data)
1556 {
1557 int ret, err = -1;
1558 struct cds_wfq_node *node;
1559 struct ust_command *ust_cmd = NULL;
1560 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1561 struct ust_reg_wait_queue wait_queue = {
1562 .count = 0,
1563 };
1564
1565 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1566
1567 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1568 goto error_testpoint;
1569 }
1570
1571 health_code_update();
1572
1573 CDS_INIT_LIST_HEAD(&wait_queue.head);
1574
1575 DBG("[thread] Dispatch UST command started");
1576
1577 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1578 health_code_update();
1579
1580 /* Atomically prepare the queue futex */
1581 futex_nto1_prepare(&ust_cmd_queue.futex);
1582
1583 do {
1584 struct ust_app *app = NULL;
1585 ust_cmd = NULL;
1586
1587 /*
1588 * Make sure we don't have node(s) that have hung up before receiving
1589 * the notify socket. This is to clean the list in order to avoid
1590 * memory leaks from notify socket that are never seen.
1591 */
1592 sanitize_wait_queue(&wait_queue);
1593
1594 health_code_update();
1595 /* Dequeue command for registration */
1596 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1597 if (node == NULL) {
1598 DBG("Woken up but nothing in the UST command queue");
1599 /* Continue thread execution */
1600 break;
1601 }
1602
1603 ust_cmd = caa_container_of(node, struct ust_command, node);
1604
1605 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1606 " gid:%d sock:%d name:%s (version %d.%d)",
1607 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1608 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1609 ust_cmd->sock, ust_cmd->reg_msg.name,
1610 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1611
1612 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1613 wait_node = zmalloc(sizeof(*wait_node));
1614 if (!wait_node) {
1615 PERROR("zmalloc wait_node dispatch");
1616 ret = close(ust_cmd->sock);
1617 if (ret < 0) {
1618 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1619 }
1620 lttng_fd_put(LTTNG_FD_APPS, 1);
1621 free(ust_cmd);
1622 goto error;
1623 }
1624 CDS_INIT_LIST_HEAD(&wait_node->head);
1625
1626 /* Create application object if socket is CMD. */
1627 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1628 ust_cmd->sock);
1629 if (!wait_node->app) {
1630 ret = close(ust_cmd->sock);
1631 if (ret < 0) {
1632 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1633 }
1634 lttng_fd_put(LTTNG_FD_APPS, 1);
1635 free(wait_node);
1636 free(ust_cmd);
1637 continue;
1638 }
1639 /*
1640 * Add application to the wait queue so we can set the notify
1641 * socket before putting this object in the global ht.
1642 */
1643 cds_list_add(&wait_node->head, &wait_queue.head);
1644 wait_queue.count++;
1645
1646 free(ust_cmd);
1647 /*
1648 * We have to continue here since we don't have the notify
1649 * socket and the application MUST be added to the hash table
1650 * only at that moment.
1651 */
1652 continue;
1653 } else {
1654 /*
1655 * Look for the application in the local wait queue and set the
1656 * notify socket if found.
1657 */
1658 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1659 &wait_queue.head, head) {
1660 health_code_update();
1661 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1662 wait_node->app->notify_sock = ust_cmd->sock;
1663 cds_list_del(&wait_node->head);
1664 wait_queue.count--;
1665 app = wait_node->app;
1666 free(wait_node);
1667 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1668 break;
1669 }
1670 }
1671
1672 /*
1673 * With no application at this stage the received socket is
1674 * basically useless so close it before we free the cmd data
1675 * structure for good.
1676 */
1677 if (!app) {
1678 ret = close(ust_cmd->sock);
1679 if (ret < 0) {
1680 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1681 }
1682 lttng_fd_put(LTTNG_FD_APPS, 1);
1683 }
1684 free(ust_cmd);
1685 }
1686
1687 if (app) {
1688 /*
1689 * @session_lock_list
1690 *
1691 * Lock the global session list so from the register up to the
1692 * registration done message, no thread can see the application
1693 * and change its state.
1694 */
1695 session_lock_list();
1696 rcu_read_lock();
1697
1698 /*
1699 * Add application to the global hash table. This needs to be
1700 * done before the update to the UST registry can locate the
1701 * application.
1702 */
1703 ust_app_add(app);
1704
1705 /* Set app version. This call will print an error if needed. */
1706 (void) ust_app_version(app);
1707
1708 /* Send notify socket through the notify pipe. */
1709 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1710 app->notify_sock);
1711 if (ret < 0) {
1712 rcu_read_unlock();
1713 session_unlock_list();
1714 /*
1715 * No notify thread, stop the UST tracing. However, this is
1716 * not an internal error of the this thread thus setting
1717 * the health error code to a normal exit.
1718 */
1719 err = 0;
1720 goto error;
1721 }
1722
1723 /*
1724 * Update newly registered application with the tracing
1725 * registry info already enabled information.
1726 */
1727 update_ust_app(app->sock);
1728
1729 /*
1730 * Don't care about return value. Let the manage apps threads
1731 * handle app unregistration upon socket close.
1732 */
1733 (void) ust_app_register_done(app->sock);
1734
1735 /*
1736 * Even if the application socket has been closed, send the app
1737 * to the thread and unregistration will take place at that
1738 * place.
1739 */
1740 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1741 if (ret < 0) {
1742 rcu_read_unlock();
1743 session_unlock_list();
1744 /*
1745 * No apps. thread, stop the UST tracing. However, this is
1746 * not an internal error of the this thread thus setting
1747 * the health error code to a normal exit.
1748 */
1749 err = 0;
1750 goto error;
1751 }
1752
1753 rcu_read_unlock();
1754 session_unlock_list();
1755 }
1756 } while (node != NULL);
1757
1758 health_poll_entry();
1759 /* Futex wait on queue. Blocking call on futex() */
1760 futex_nto1_wait(&ust_cmd_queue.futex);
1761 health_poll_exit();
1762 }
1763 /* Normal exit, no error */
1764 err = 0;
1765
1766 error:
1767 /* Clean up wait queue. */
1768 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1769 &wait_queue.head, head) {
1770 cds_list_del(&wait_node->head);
1771 wait_queue.count--;
1772 free(wait_node);
1773 }
1774
1775 error_testpoint:
1776 DBG("Dispatch thread dying");
1777 if (err) {
1778 health_error();
1779 ERR("Health error occurred in %s", __func__);
1780 }
1781 health_unregister(health_sessiond);
1782 return NULL;
1783 }
1784
1785 /*
1786 * This thread manage application registration.
1787 */
1788 static void *thread_registration_apps(void *data)
1789 {
1790 int sock = -1, i, ret, pollfd, err = -1;
1791 uint32_t revents, nb_fd;
1792 struct lttng_poll_event events;
1793 /*
1794 * Get allocated in this thread, enqueued to a global queue, dequeued and
1795 * freed in the manage apps thread.
1796 */
1797 struct ust_command *ust_cmd = NULL;
1798
1799 DBG("[thread] Manage application registration started");
1800
1801 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1802
1803 if (testpoint(sessiond_thread_registration_apps)) {
1804 goto error_testpoint;
1805 }
1806
1807 ret = lttcomm_listen_unix_sock(apps_sock);
1808 if (ret < 0) {
1809 goto error_listen;
1810 }
1811
1812 /*
1813 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1814 * more will be added to this poll set.
1815 */
1816 ret = sessiond_set_thread_pollset(&events, 2);
1817 if (ret < 0) {
1818 goto error_create_poll;
1819 }
1820
1821 /* Add the application registration socket */
1822 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1823 if (ret < 0) {
1824 goto error_poll_add;
1825 }
1826
1827 /* Notify all applications to register */
1828 ret = notify_ust_apps(1);
1829 if (ret < 0) {
1830 ERR("Failed to notify applications or create the wait shared memory.\n"
1831 "Execution continues but there might be problem for already\n"
1832 "running applications that wishes to register.");
1833 }
1834
1835 while (1) {
1836 DBG("Accepting application registration");
1837
1838 /* Inifinite blocking call, waiting for transmission */
1839 restart:
1840 health_poll_entry();
1841 ret = lttng_poll_wait(&events, -1);
1842 health_poll_exit();
1843 if (ret < 0) {
1844 /*
1845 * Restart interrupted system call.
1846 */
1847 if (errno == EINTR) {
1848 goto restart;
1849 }
1850 goto error;
1851 }
1852
1853 nb_fd = ret;
1854
1855 for (i = 0; i < nb_fd; i++) {
1856 health_code_update();
1857
1858 /* Fetch once the poll data */
1859 revents = LTTNG_POLL_GETEV(&events, i);
1860 pollfd = LTTNG_POLL_GETFD(&events, i);
1861
1862 /* Thread quit pipe has been closed. Killing thread. */
1863 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1864 if (ret) {
1865 err = 0;
1866 goto exit;
1867 }
1868
1869 /* Event on the registration socket */
1870 if (pollfd == apps_sock) {
1871 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1872 ERR("Register apps socket poll error");
1873 goto error;
1874 } else if (revents & LPOLLIN) {
1875 sock = lttcomm_accept_unix_sock(apps_sock);
1876 if (sock < 0) {
1877 goto error;
1878 }
1879
1880 /*
1881 * Set socket timeout for both receiving and ending.
1882 * app_socket_timeout is in seconds, whereas
1883 * lttcomm_setsockopt_rcv_timeout and
1884 * lttcomm_setsockopt_snd_timeout expect msec as
1885 * parameter.
1886 */
1887 (void) lttcomm_setsockopt_rcv_timeout(sock,
1888 app_socket_timeout * 1000);
1889 (void) lttcomm_setsockopt_snd_timeout(sock,
1890 app_socket_timeout * 1000);
1891
1892 /*
1893 * Set the CLOEXEC flag. Return code is useless because
1894 * either way, the show must go on.
1895 */
1896 (void) utils_set_fd_cloexec(sock);
1897
1898 /* Create UST registration command for enqueuing */
1899 ust_cmd = zmalloc(sizeof(struct ust_command));
1900 if (ust_cmd == NULL) {
1901 PERROR("ust command zmalloc");
1902 goto error;
1903 }
1904
1905 /*
1906 * Using message-based transmissions to ensure we don't
1907 * have to deal with partially received messages.
1908 */
1909 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1910 if (ret < 0) {
1911 ERR("Exhausted file descriptors allowed for applications.");
1912 free(ust_cmd);
1913 ret = close(sock);
1914 if (ret) {
1915 PERROR("close");
1916 }
1917 sock = -1;
1918 continue;
1919 }
1920
1921 health_code_update();
1922 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1923 if (ret < 0) {
1924 free(ust_cmd);
1925 /* Close socket of the application. */
1926 ret = close(sock);
1927 if (ret) {
1928 PERROR("close");
1929 }
1930 lttng_fd_put(LTTNG_FD_APPS, 1);
1931 sock = -1;
1932 continue;
1933 }
1934 health_code_update();
1935
1936 ust_cmd->sock = sock;
1937 sock = -1;
1938
1939 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1940 " gid:%d sock:%d name:%s (version %d.%d)",
1941 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1942 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1943 ust_cmd->sock, ust_cmd->reg_msg.name,
1944 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1945
1946 /*
1947 * Lock free enqueue the registration request. The red pill
1948 * has been taken! This apps will be part of the *system*.
1949 */
1950 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1951
1952 /*
1953 * Wake the registration queue futex. Implicit memory
1954 * barrier with the exchange in cds_wfq_enqueue.
1955 */
1956 futex_nto1_wake(&ust_cmd_queue.futex);
1957 }
1958 }
1959 }
1960 }
1961
1962 exit:
1963 error:
1964 /* Notify that the registration thread is gone */
1965 notify_ust_apps(0);
1966
1967 if (apps_sock >= 0) {
1968 ret = close(apps_sock);
1969 if (ret) {
1970 PERROR("close");
1971 }
1972 }
1973 if (sock >= 0) {
1974 ret = close(sock);
1975 if (ret) {
1976 PERROR("close");
1977 }
1978 lttng_fd_put(LTTNG_FD_APPS, 1);
1979 }
1980 unlink(apps_unix_sock_path);
1981
1982 error_poll_add:
1983 lttng_poll_clean(&events);
1984 error_listen:
1985 error_create_poll:
1986 error_testpoint:
1987 DBG("UST Registration thread cleanup complete");
1988 if (err) {
1989 health_error();
1990 ERR("Health error occurred in %s", __func__);
1991 }
1992 health_unregister(health_sessiond);
1993
1994 return NULL;
1995 }
1996
1997 /*
1998 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
1999 * exec or it will fails.
2000 */
2001 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2002 {
2003 int ret, clock_ret;
2004 struct timespec timeout;
2005
2006 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2007 consumer_data->consumer_thread_is_ready = 0;
2008
2009 /* Setup pthread condition */
2010 ret = pthread_condattr_init(&consumer_data->condattr);
2011 if (ret != 0) {
2012 errno = ret;
2013 PERROR("pthread_condattr_init consumer data");
2014 goto error;
2015 }
2016
2017 /*
2018 * Set the monotonic clock in order to make sure we DO NOT jump in time
2019 * between the clock_gettime() call and the timedwait call. See bug #324
2020 * for a more details and how we noticed it.
2021 */
2022 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2023 if (ret != 0) {
2024 errno = ret;
2025 PERROR("pthread_condattr_setclock consumer data");
2026 goto error;
2027 }
2028
2029 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2030 if (ret != 0) {
2031 errno = ret;
2032 PERROR("pthread_cond_init consumer data");
2033 goto error;
2034 }
2035
2036 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2037 consumer_data);
2038 if (ret != 0) {
2039 PERROR("pthread_create consumer");
2040 ret = -1;
2041 goto error;
2042 }
2043
2044 /* We are about to wait on a pthread condition */
2045 pthread_mutex_lock(&consumer_data->cond_mutex);
2046
2047 /* Get time for sem_timedwait absolute timeout */
2048 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2049 /*
2050 * Set the timeout for the condition timed wait even if the clock gettime
2051 * call fails since we might loop on that call and we want to avoid to
2052 * increment the timeout too many times.
2053 */
2054 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2055
2056 /*
2057 * The following loop COULD be skipped in some conditions so this is why we
2058 * set ret to 0 in order to make sure at least one round of the loop is
2059 * done.
2060 */
2061 ret = 0;
2062
2063 /*
2064 * Loop until the condition is reached or when a timeout is reached. Note
2065 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2066 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2067 * possible. This loop does not take any chances and works with both of
2068 * them.
2069 */
2070 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2071 if (clock_ret < 0) {
2072 PERROR("clock_gettime spawn consumer");
2073 /* Infinite wait for the consumerd thread to be ready */
2074 ret = pthread_cond_wait(&consumer_data->cond,
2075 &consumer_data->cond_mutex);
2076 } else {
2077 ret = pthread_cond_timedwait(&consumer_data->cond,
2078 &consumer_data->cond_mutex, &timeout);
2079 }
2080 }
2081
2082 /* Release the pthread condition */
2083 pthread_mutex_unlock(&consumer_data->cond_mutex);
2084
2085 if (ret != 0) {
2086 errno = ret;
2087 if (ret == ETIMEDOUT) {
2088 /*
2089 * Call has timed out so we kill the kconsumerd_thread and return
2090 * an error.
2091 */
2092 ERR("Condition timed out. The consumer thread was never ready."
2093 " Killing it");
2094 ret = pthread_cancel(consumer_data->thread);
2095 if (ret < 0) {
2096 PERROR("pthread_cancel consumer thread");
2097 }
2098 } else {
2099 PERROR("pthread_cond_wait failed consumer thread");
2100 }
2101 goto error;
2102 }
2103
2104 pthread_mutex_lock(&consumer_data->pid_mutex);
2105 if (consumer_data->pid == 0) {
2106 ERR("Consumerd did not start");
2107 pthread_mutex_unlock(&consumer_data->pid_mutex);
2108 goto error;
2109 }
2110 pthread_mutex_unlock(&consumer_data->pid_mutex);
2111
2112 return 0;
2113
2114 error:
2115 return ret;
2116 }
2117
2118 /*
2119 * Join consumer thread
2120 */
2121 static int join_consumer_thread(struct consumer_data *consumer_data)
2122 {
2123 void *status;
2124
2125 /* Consumer pid must be a real one. */
2126 if (consumer_data->pid > 0) {
2127 int ret;
2128 ret = kill(consumer_data->pid, SIGTERM);
2129 if (ret) {
2130 ERR("Error killing consumer daemon");
2131 return ret;
2132 }
2133 return pthread_join(consumer_data->thread, &status);
2134 } else {
2135 return 0;
2136 }
2137 }
2138
2139 /*
2140 * Fork and exec a consumer daemon (consumerd).
2141 *
2142 * Return pid if successful else -1.
2143 */
2144 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2145 {
2146 int ret;
2147 pid_t pid;
2148 const char *consumer_to_use;
2149 const char *verbosity;
2150 struct stat st;
2151
2152 DBG("Spawning consumerd");
2153
2154 pid = fork();
2155 if (pid == 0) {
2156 /*
2157 * Exec consumerd.
2158 */
2159 if (opt_verbose_consumer) {
2160 verbosity = "--verbose";
2161 } else {
2162 verbosity = "--quiet";
2163 }
2164 switch (consumer_data->type) {
2165 case LTTNG_CONSUMER_KERNEL:
2166 /*
2167 * Find out which consumerd to execute. We will first try the
2168 * 64-bit path, then the sessiond's installation directory, and
2169 * fallback on the 32-bit one,
2170 */
2171 DBG3("Looking for a kernel consumer at these locations:");
2172 DBG3(" 1) %s", consumerd64_bin);
2173 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2174 DBG3(" 3) %s", consumerd32_bin);
2175 if (stat(consumerd64_bin, &st) == 0) {
2176 DBG3("Found location #1");
2177 consumer_to_use = consumerd64_bin;
2178 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2179 DBG3("Found location #2");
2180 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2181 } else if (stat(consumerd32_bin, &st) == 0) {
2182 DBG3("Found location #3");
2183 consumer_to_use = consumerd32_bin;
2184 } else {
2185 DBG("Could not find any valid consumerd executable");
2186 break;
2187 }
2188 DBG("Using kernel consumer at: %s", consumer_to_use);
2189 execl(consumer_to_use,
2190 "lttng-consumerd", verbosity, "-k",
2191 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2192 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2193 "--group", tracing_group_name,
2194 NULL);
2195 break;
2196 case LTTNG_CONSUMER64_UST:
2197 {
2198 char *tmpnew = NULL;
2199
2200 if (consumerd64_libdir[0] != '\0') {
2201 char *tmp;
2202 size_t tmplen;
2203
2204 tmp = getenv("LD_LIBRARY_PATH");
2205 if (!tmp) {
2206 tmp = "";
2207 }
2208 tmplen = strlen("LD_LIBRARY_PATH=")
2209 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2210 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2211 if (!tmpnew) {
2212 ret = -ENOMEM;
2213 goto error;
2214 }
2215 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2216 strcat(tmpnew, consumerd64_libdir);
2217 if (tmp[0] != '\0') {
2218 strcat(tmpnew, ":");
2219 strcat(tmpnew, tmp);
2220 }
2221 ret = putenv(tmpnew);
2222 if (ret) {
2223 ret = -errno;
2224 free(tmpnew);
2225 goto error;
2226 }
2227 }
2228 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2229 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2230 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2231 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2232 "--group", tracing_group_name,
2233 NULL);
2234 if (consumerd64_libdir[0] != '\0') {
2235 free(tmpnew);
2236 }
2237 if (ret) {
2238 goto error;
2239 }
2240 break;
2241 }
2242 case LTTNG_CONSUMER32_UST:
2243 {
2244 char *tmpnew = NULL;
2245
2246 if (consumerd32_libdir[0] != '\0') {
2247 char *tmp;
2248 size_t tmplen;
2249
2250 tmp = getenv("LD_LIBRARY_PATH");
2251 if (!tmp) {
2252 tmp = "";
2253 }
2254 tmplen = strlen("LD_LIBRARY_PATH=")
2255 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2256 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2257 if (!tmpnew) {
2258 ret = -ENOMEM;
2259 goto error;
2260 }
2261 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2262 strcat(tmpnew, consumerd32_libdir);
2263 if (tmp[0] != '\0') {
2264 strcat(tmpnew, ":");
2265 strcat(tmpnew, tmp);
2266 }
2267 ret = putenv(tmpnew);
2268 if (ret) {
2269 ret = -errno;
2270 free(tmpnew);
2271 goto error;
2272 }
2273 }
2274 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2275 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2276 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2277 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2278 "--group", tracing_group_name,
2279 NULL);
2280 if (consumerd32_libdir[0] != '\0') {
2281 free(tmpnew);
2282 }
2283 if (ret) {
2284 goto error;
2285 }
2286 break;
2287 }
2288 default:
2289 PERROR("unknown consumer type");
2290 exit(EXIT_FAILURE);
2291 }
2292 if (errno != 0) {
2293 PERROR("kernel start consumer exec");
2294 }
2295 exit(EXIT_FAILURE);
2296 } else if (pid > 0) {
2297 ret = pid;
2298 } else {
2299 PERROR("start consumer fork");
2300 ret = -errno;
2301 }
2302 error:
2303 return ret;
2304 }
2305
2306 /*
2307 * Spawn the consumerd daemon and session daemon thread.
2308 */
2309 static int start_consumerd(struct consumer_data *consumer_data)
2310 {
2311 int ret;
2312
2313 /*
2314 * Set the listen() state on the socket since there is a possible race
2315 * between the exec() of the consumer daemon and this call if place in the
2316 * consumer thread. See bug #366 for more details.
2317 */
2318 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2319 if (ret < 0) {
2320 goto error;
2321 }
2322
2323 pthread_mutex_lock(&consumer_data->pid_mutex);
2324 if (consumer_data->pid != 0) {
2325 pthread_mutex_unlock(&consumer_data->pid_mutex);
2326 goto end;
2327 }
2328
2329 ret = spawn_consumerd(consumer_data);
2330 if (ret < 0) {
2331 ERR("Spawning consumerd failed");
2332 pthread_mutex_unlock(&consumer_data->pid_mutex);
2333 goto error;
2334 }
2335
2336 /* Setting up the consumer_data pid */
2337 consumer_data->pid = ret;
2338 DBG2("Consumer pid %d", consumer_data->pid);
2339 pthread_mutex_unlock(&consumer_data->pid_mutex);
2340
2341 DBG2("Spawning consumer control thread");
2342 ret = spawn_consumer_thread(consumer_data);
2343 if (ret < 0) {
2344 ERR("Fatal error spawning consumer control thread");
2345 goto error;
2346 }
2347
2348 end:
2349 return 0;
2350
2351 error:
2352 /* Cleanup already created sockets on error. */
2353 if (consumer_data->err_sock >= 0) {
2354 int err;
2355
2356 err = close(consumer_data->err_sock);
2357 if (err < 0) {
2358 PERROR("close consumer data error socket");
2359 }
2360 }
2361 return ret;
2362 }
2363
2364 /*
2365 * Setup necessary data for kernel tracer action.
2366 */
2367 static int init_kernel_tracer(void)
2368 {
2369 int ret;
2370
2371 /* Modprobe lttng kernel modules */
2372 ret = modprobe_lttng_control();
2373 if (ret < 0) {
2374 goto error;
2375 }
2376
2377 /* Open debugfs lttng */
2378 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2379 if (kernel_tracer_fd < 0) {
2380 DBG("Failed to open %s", module_proc_lttng);
2381 ret = -1;
2382 goto error_open;
2383 }
2384
2385 /* Validate kernel version */
2386 ret = kernel_validate_version(kernel_tracer_fd);
2387 if (ret < 0) {
2388 goto error_version;
2389 }
2390
2391 ret = modprobe_lttng_data();
2392 if (ret < 0) {
2393 goto error_modules;
2394 }
2395
2396 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2397 return 0;
2398
2399 error_version:
2400 modprobe_remove_lttng_control();
2401 ret = close(kernel_tracer_fd);
2402 if (ret) {
2403 PERROR("close");
2404 }
2405 kernel_tracer_fd = -1;
2406 return LTTNG_ERR_KERN_VERSION;
2407
2408 error_modules:
2409 ret = close(kernel_tracer_fd);
2410 if (ret) {
2411 PERROR("close");
2412 }
2413
2414 error_open:
2415 modprobe_remove_lttng_control();
2416
2417 error:
2418 WARN("No kernel tracer available");
2419 kernel_tracer_fd = -1;
2420 if (!is_root) {
2421 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2422 } else {
2423 return LTTNG_ERR_KERN_NA;
2424 }
2425 }
2426
2427
2428 /*
2429 * Copy consumer output from the tracing session to the domain session. The
2430 * function also applies the right modification on a per domain basis for the
2431 * trace files destination directory.
2432 *
2433 * Should *NOT* be called with RCU read-side lock held.
2434 */
2435 static int copy_session_consumer(int domain, struct ltt_session *session)
2436 {
2437 int ret;
2438 const char *dir_name;
2439 struct consumer_output *consumer;
2440
2441 assert(session);
2442 assert(session->consumer);
2443
2444 switch (domain) {
2445 case LTTNG_DOMAIN_KERNEL:
2446 DBG3("Copying tracing session consumer output in kernel session");
2447 /*
2448 * XXX: We should audit the session creation and what this function
2449 * does "extra" in order to avoid a destroy since this function is used
2450 * in the domain session creation (kernel and ust) only. Same for UST
2451 * domain.
2452 */
2453 if (session->kernel_session->consumer) {
2454 consumer_destroy_output(session->kernel_session->consumer);
2455 }
2456 session->kernel_session->consumer =
2457 consumer_copy_output(session->consumer);
2458 /* Ease our life a bit for the next part */
2459 consumer = session->kernel_session->consumer;
2460 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2461 break;
2462 case LTTNG_DOMAIN_JUL:
2463 case LTTNG_DOMAIN_UST:
2464 DBG3("Copying tracing session consumer output in UST session");
2465 if (session->ust_session->consumer) {
2466 consumer_destroy_output(session->ust_session->consumer);
2467 }
2468 session->ust_session->consumer =
2469 consumer_copy_output(session->consumer);
2470 /* Ease our life a bit for the next part */
2471 consumer = session->ust_session->consumer;
2472 dir_name = DEFAULT_UST_TRACE_DIR;
2473 break;
2474 default:
2475 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2476 goto error;
2477 }
2478
2479 /* Append correct directory to subdir */
2480 strncat(consumer->subdir, dir_name,
2481 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2482 DBG3("Copy session consumer subdir %s", consumer->subdir);
2483
2484 ret = LTTNG_OK;
2485
2486 error:
2487 return ret;
2488 }
2489
2490 /*
2491 * Create an UST session and add it to the session ust list.
2492 *
2493 * Should *NOT* be called with RCU read-side lock held.
2494 */
2495 static int create_ust_session(struct ltt_session *session,
2496 struct lttng_domain *domain)
2497 {
2498 int ret;
2499 struct ltt_ust_session *lus = NULL;
2500
2501 assert(session);
2502 assert(domain);
2503 assert(session->consumer);
2504
2505 switch (domain->type) {
2506 case LTTNG_DOMAIN_JUL:
2507 case LTTNG_DOMAIN_UST:
2508 break;
2509 default:
2510 ERR("Unknown UST domain on create session %d", domain->type);
2511 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2512 goto error;
2513 }
2514
2515 DBG("Creating UST session");
2516
2517 lus = trace_ust_create_session(session->id);
2518 if (lus == NULL) {
2519 ret = LTTNG_ERR_UST_SESS_FAIL;
2520 goto error;
2521 }
2522
2523 lus->uid = session->uid;
2524 lus->gid = session->gid;
2525 lus->output_traces = session->output_traces;
2526 lus->snapshot_mode = session->snapshot_mode;
2527 lus->live_timer_interval = session->live_timer;
2528 session->ust_session = lus;
2529
2530 /* Copy session output to the newly created UST session */
2531 ret = copy_session_consumer(domain->type, session);
2532 if (ret != LTTNG_OK) {
2533 goto error;
2534 }
2535
2536 return LTTNG_OK;
2537
2538 error:
2539 free(lus);
2540 session->ust_session = NULL;
2541 return ret;
2542 }
2543
2544 /*
2545 * Create a kernel tracer session then create the default channel.
2546 */
2547 static int create_kernel_session(struct ltt_session *session)
2548 {
2549 int ret;
2550
2551 DBG("Creating kernel session");
2552
2553 ret = kernel_create_session(session, kernel_tracer_fd);
2554 if (ret < 0) {
2555 ret = LTTNG_ERR_KERN_SESS_FAIL;
2556 goto error;
2557 }
2558
2559 /* Code flow safety */
2560 assert(session->kernel_session);
2561
2562 /* Copy session output to the newly created Kernel session */
2563 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2564 if (ret != LTTNG_OK) {
2565 goto error;
2566 }
2567
2568 /* Create directory(ies) on local filesystem. */
2569 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2570 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2571 ret = run_as_mkdir_recursive(
2572 session->kernel_session->consumer->dst.trace_path,
2573 S_IRWXU | S_IRWXG, session->uid, session->gid);
2574 if (ret < 0) {
2575 if (ret != -EEXIST) {
2576 ERR("Trace directory creation error");
2577 goto error;
2578 }
2579 }
2580 }
2581
2582 session->kernel_session->uid = session->uid;
2583 session->kernel_session->gid = session->gid;
2584 session->kernel_session->output_traces = session->output_traces;
2585 session->kernel_session->snapshot_mode = session->snapshot_mode;
2586
2587 return LTTNG_OK;
2588
2589 error:
2590 trace_kernel_destroy_session(session->kernel_session);
2591 session->kernel_session = NULL;
2592 return ret;
2593 }
2594
2595 /*
2596 * Count number of session permitted by uid/gid.
2597 */
2598 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2599 {
2600 unsigned int i = 0;
2601 struct ltt_session *session;
2602
2603 DBG("Counting number of available session for UID %d GID %d",
2604 uid, gid);
2605 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2606 /*
2607 * Only list the sessions the user can control.
2608 */
2609 if (!session_access_ok(session, uid, gid)) {
2610 continue;
2611 }
2612 i++;
2613 }
2614 return i;
2615 }
2616
2617 /*
2618 * Process the command requested by the lttng client within the command
2619 * context structure. This function make sure that the return structure (llm)
2620 * is set and ready for transmission before returning.
2621 *
2622 * Return any error encountered or 0 for success.
2623 *
2624 * "sock" is only used for special-case var. len data.
2625 *
2626 * Should *NOT* be called with RCU read-side lock held.
2627 */
2628 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2629 int *sock_error)
2630 {
2631 int ret = LTTNG_OK;
2632 int need_tracing_session = 1;
2633 int need_domain;
2634
2635 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2636
2637 *sock_error = 0;
2638
2639 switch (cmd_ctx->lsm->cmd_type) {
2640 case LTTNG_CREATE_SESSION:
2641 case LTTNG_CREATE_SESSION_SNAPSHOT:
2642 case LTTNG_CREATE_SESSION_LIVE:
2643 case LTTNG_DESTROY_SESSION:
2644 case LTTNG_LIST_SESSIONS:
2645 case LTTNG_LIST_DOMAINS:
2646 case LTTNG_START_TRACE:
2647 case LTTNG_STOP_TRACE:
2648 case LTTNG_DATA_PENDING:
2649 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2650 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2651 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2652 case LTTNG_SNAPSHOT_RECORD:
2653 need_domain = 0;
2654 break;
2655 default:
2656 need_domain = 1;
2657 }
2658
2659 if (opt_no_kernel && need_domain
2660 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2661 if (!is_root) {
2662 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2663 } else {
2664 ret = LTTNG_ERR_KERN_NA;
2665 }
2666 goto error;
2667 }
2668
2669 /* Deny register consumer if we already have a spawned consumer. */
2670 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2671 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2672 if (kconsumer_data.pid > 0) {
2673 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2674 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2675 goto error;
2676 }
2677 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2678 }
2679
2680 /*
2681 * Check for command that don't needs to allocate a returned payload. We do
2682 * this here so we don't have to make the call for no payload at each
2683 * command.
2684 */
2685 switch(cmd_ctx->lsm->cmd_type) {
2686 case LTTNG_LIST_SESSIONS:
2687 case LTTNG_LIST_TRACEPOINTS:
2688 case LTTNG_LIST_TRACEPOINT_FIELDS:
2689 case LTTNG_LIST_DOMAINS:
2690 case LTTNG_LIST_CHANNELS:
2691 case LTTNG_LIST_EVENTS:
2692 break;
2693 default:
2694 /* Setup lttng message with no payload */
2695 ret = setup_lttng_msg(cmd_ctx, 0);
2696 if (ret < 0) {
2697 /* This label does not try to unlock the session */
2698 goto init_setup_error;
2699 }
2700 }
2701
2702 /* Commands that DO NOT need a session. */
2703 switch (cmd_ctx->lsm->cmd_type) {
2704 case LTTNG_CREATE_SESSION:
2705 case LTTNG_CREATE_SESSION_SNAPSHOT:
2706 case LTTNG_CREATE_SESSION_LIVE:
2707 case LTTNG_CALIBRATE:
2708 case LTTNG_LIST_SESSIONS:
2709 case LTTNG_LIST_TRACEPOINTS:
2710 case LTTNG_LIST_TRACEPOINT_FIELDS:
2711 need_tracing_session = 0;
2712 break;
2713 default:
2714 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2715 /*
2716 * We keep the session list lock across _all_ commands
2717 * for now, because the per-session lock does not
2718 * handle teardown properly.
2719 */
2720 session_lock_list();
2721 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2722 if (cmd_ctx->session == NULL) {
2723 ret = LTTNG_ERR_SESS_NOT_FOUND;
2724 goto error;
2725 } else {
2726 /* Acquire lock for the session */
2727 session_lock(cmd_ctx->session);
2728 }
2729 break;
2730 }
2731
2732 if (!need_domain) {
2733 goto skip_domain;
2734 }
2735
2736 /*
2737 * Check domain type for specific "pre-action".
2738 */
2739 switch (cmd_ctx->lsm->domain.type) {
2740 case LTTNG_DOMAIN_KERNEL:
2741 if (!is_root) {
2742 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2743 goto error;
2744 }
2745
2746 /* Kernel tracer check */
2747 if (kernel_tracer_fd == -1) {
2748 /* Basically, load kernel tracer modules */
2749 ret = init_kernel_tracer();
2750 if (ret != 0) {
2751 goto error;
2752 }
2753 }
2754
2755 /* Consumer is in an ERROR state. Report back to client */
2756 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2757 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2758 goto error;
2759 }
2760
2761 /* Need a session for kernel command */
2762 if (need_tracing_session) {
2763 if (cmd_ctx->session->kernel_session == NULL) {
2764 ret = create_kernel_session(cmd_ctx->session);
2765 if (ret < 0) {
2766 ret = LTTNG_ERR_KERN_SESS_FAIL;
2767 goto error;
2768 }
2769 }
2770
2771 /* Start the kernel consumer daemon */
2772 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2773 if (kconsumer_data.pid == 0 &&
2774 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2775 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2776 ret = start_consumerd(&kconsumer_data);
2777 if (ret < 0) {
2778 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2779 goto error;
2780 }
2781 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2782 } else {
2783 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2784 }
2785
2786 /*
2787 * The consumer was just spawned so we need to add the socket to
2788 * the consumer output of the session if exist.
2789 */
2790 ret = consumer_create_socket(&kconsumer_data,
2791 cmd_ctx->session->kernel_session->consumer);
2792 if (ret < 0) {
2793 goto error;
2794 }
2795 }
2796
2797 break;
2798 case LTTNG_DOMAIN_JUL:
2799 case LTTNG_DOMAIN_UST:
2800 {
2801 if (!ust_app_supported()) {
2802 ret = LTTNG_ERR_NO_UST;
2803 goto error;
2804 }
2805 /* Consumer is in an ERROR state. Report back to client */
2806 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2807 ret = LTTNG_ERR_NO_USTCONSUMERD;
2808 goto error;
2809 }
2810
2811 if (need_tracing_session) {
2812 /* Create UST session if none exist. */
2813 if (cmd_ctx->session->ust_session == NULL) {
2814 ret = create_ust_session(cmd_ctx->session,
2815 &cmd_ctx->lsm->domain);
2816 if (ret != LTTNG_OK) {
2817 goto error;
2818 }
2819 }
2820
2821 /* Start the UST consumer daemons */
2822 /* 64-bit */
2823 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2824 if (consumerd64_bin[0] != '\0' &&
2825 ustconsumer64_data.pid == 0 &&
2826 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2827 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2828 ret = start_consumerd(&ustconsumer64_data);
2829 if (ret < 0) {
2830 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2831 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2832 goto error;
2833 }
2834
2835 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2836 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2837 } else {
2838 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2839 }
2840
2841 /*
2842 * Setup socket for consumer 64 bit. No need for atomic access
2843 * since it was set above and can ONLY be set in this thread.
2844 */
2845 ret = consumer_create_socket(&ustconsumer64_data,
2846 cmd_ctx->session->ust_session->consumer);
2847 if (ret < 0) {
2848 goto error;
2849 }
2850
2851 /* 32-bit */
2852 if (consumerd32_bin[0] != '\0' &&
2853 ustconsumer32_data.pid == 0 &&
2854 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2855 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2856 ret = start_consumerd(&ustconsumer32_data);
2857 if (ret < 0) {
2858 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2859 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2860 goto error;
2861 }
2862
2863 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2864 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2865 } else {
2866 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2867 }
2868
2869 /*
2870 * Setup socket for consumer 64 bit. No need for atomic access
2871 * since it was set above and can ONLY be set in this thread.
2872 */
2873 ret = consumer_create_socket(&ustconsumer32_data,
2874 cmd_ctx->session->ust_session->consumer);
2875 if (ret < 0) {
2876 goto error;
2877 }
2878 }
2879 break;
2880 }
2881 default:
2882 break;
2883 }
2884 skip_domain:
2885
2886 /* Validate consumer daemon state when start/stop trace command */
2887 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2888 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2889 switch (cmd_ctx->lsm->domain.type) {
2890 case LTTNG_DOMAIN_JUL:
2891 case LTTNG_DOMAIN_UST:
2892 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2893 ret = LTTNG_ERR_NO_USTCONSUMERD;
2894 goto error;
2895 }
2896 break;
2897 case LTTNG_DOMAIN_KERNEL:
2898 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2899 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2900 goto error;
2901 }
2902 break;
2903 }
2904 }
2905
2906 /*
2907 * Check that the UID or GID match that of the tracing session.
2908 * The root user can interact with all sessions.
2909 */
2910 if (need_tracing_session) {
2911 if (!session_access_ok(cmd_ctx->session,
2912 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2913 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2914 ret = LTTNG_ERR_EPERM;
2915 goto error;
2916 }
2917 }
2918
2919 /*
2920 * Send relayd information to consumer as soon as we have a domain and a
2921 * session defined.
2922 */
2923 if (cmd_ctx->session && need_domain) {
2924 /*
2925 * Setup relayd if not done yet. If the relayd information was already
2926 * sent to the consumer, this call will gracefully return.
2927 */
2928 ret = cmd_setup_relayd(cmd_ctx->session);
2929 if (ret != LTTNG_OK) {
2930 goto error;
2931 }
2932 }
2933
2934 /* Process by command type */
2935 switch (cmd_ctx->lsm->cmd_type) {
2936 case LTTNG_ADD_CONTEXT:
2937 {
2938 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2939 cmd_ctx->lsm->u.context.channel_name,
2940 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2941 break;
2942 }
2943 case LTTNG_DISABLE_CHANNEL:
2944 {
2945 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2946 cmd_ctx->lsm->u.disable.channel_name);
2947 break;
2948 }
2949 case LTTNG_DISABLE_EVENT:
2950 {
2951 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2952 cmd_ctx->lsm->u.disable.channel_name,
2953 cmd_ctx->lsm->u.disable.name);
2954 break;
2955 }
2956 case LTTNG_DISABLE_ALL_EVENT:
2957 {
2958 DBG("Disabling all events");
2959
2960 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2961 cmd_ctx->lsm->u.disable.channel_name);
2962 break;
2963 }
2964 case LTTNG_ENABLE_CHANNEL:
2965 {
2966 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2967 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2968 break;
2969 }
2970 case LTTNG_ENABLE_EVENT:
2971 {
2972 struct lttng_event_exclusion *exclusion = NULL;
2973 struct lttng_filter_bytecode *bytecode = NULL;
2974
2975 /* Handle exclusion events and receive it from the client. */
2976 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
2977 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
2978
2979 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
2980 (count * LTTNG_SYMBOL_NAME_LEN));
2981 if (!exclusion) {
2982 ret = LTTNG_ERR_EXCLUSION_NOMEM;
2983 goto error;
2984 }
2985
2986 DBG("Receiving var len exclusion event list from client ...");
2987 exclusion->count = count;
2988 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
2989 count * LTTNG_SYMBOL_NAME_LEN);
2990 if (ret <= 0) {
2991 DBG("Nothing recv() from client var len data... continuing");
2992 *sock_error = 1;
2993 free(exclusion);
2994 ret = LTTNG_ERR_EXCLUSION_INVAL;
2995 goto error;
2996 }
2997 }
2998
2999 /* Handle filter and get bytecode from client. */
3000 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3001 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3002
3003 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3004 ret = LTTNG_ERR_FILTER_INVAL;
3005 free(exclusion);
3006 goto error;
3007 }
3008
3009 bytecode = zmalloc(bytecode_len);
3010 if (!bytecode) {
3011 free(exclusion);
3012 ret = LTTNG_ERR_FILTER_NOMEM;
3013 goto error;
3014 }
3015
3016 /* Receive var. len. data */
3017 DBG("Receiving var len filter's bytecode from client ...");
3018 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3019 if (ret <= 0) {
3020 DBG("Nothing recv() from client car len data... continuing");
3021 *sock_error = 1;
3022 free(bytecode);
3023 free(exclusion);
3024 ret = LTTNG_ERR_FILTER_INVAL;
3025 goto error;
3026 }
3027
3028 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3029 free(bytecode);
3030 free(exclusion);
3031 ret = LTTNG_ERR_FILTER_INVAL;
3032 goto error;
3033 }
3034 }
3035
3036 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3037 cmd_ctx->lsm->u.enable.channel_name,
3038 &cmd_ctx->lsm->u.enable.event, bytecode, exclusion,
3039 kernel_poll_pipe[1]);
3040 break;
3041 }
3042 case LTTNG_ENABLE_ALL_EVENT:
3043 {
3044 DBG("Enabling all events");
3045
3046 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
3047 cmd_ctx->lsm->u.enable.channel_name,
3048 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
3049 break;
3050 }
3051 case LTTNG_LIST_TRACEPOINTS:
3052 {
3053 struct lttng_event *events;
3054 ssize_t nb_events;
3055
3056 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3057 if (nb_events < 0) {
3058 /* Return value is a negative lttng_error_code. */
3059 ret = -nb_events;
3060 goto error;
3061 }
3062
3063 /*
3064 * Setup lttng message with payload size set to the event list size in
3065 * bytes and then copy list into the llm payload.
3066 */
3067 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3068 if (ret < 0) {
3069 free(events);
3070 goto setup_error;
3071 }
3072
3073 /* Copy event list into message payload */
3074 memcpy(cmd_ctx->llm->payload, events,
3075 sizeof(struct lttng_event) * nb_events);
3076
3077 free(events);
3078
3079 ret = LTTNG_OK;
3080 break;
3081 }
3082 case LTTNG_LIST_TRACEPOINT_FIELDS:
3083 {
3084 struct lttng_event_field *fields;
3085 ssize_t nb_fields;
3086
3087 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3088 &fields);
3089 if (nb_fields < 0) {
3090 /* Return value is a negative lttng_error_code. */
3091 ret = -nb_fields;
3092 goto error;
3093 }
3094
3095 /*
3096 * Setup lttng message with payload size set to the event list size in
3097 * bytes and then copy list into the llm payload.
3098 */
3099 ret = setup_lttng_msg(cmd_ctx,
3100 sizeof(struct lttng_event_field) * nb_fields);
3101 if (ret < 0) {
3102 free(fields);
3103 goto setup_error;
3104 }
3105
3106 /* Copy event list into message payload */
3107 memcpy(cmd_ctx->llm->payload, fields,
3108 sizeof(struct lttng_event_field) * nb_fields);
3109
3110 free(fields);
3111
3112 ret = LTTNG_OK;
3113 break;
3114 }
3115 case LTTNG_SET_CONSUMER_URI:
3116 {
3117 size_t nb_uri, len;
3118 struct lttng_uri *uris;
3119
3120 nb_uri = cmd_ctx->lsm->u.uri.size;
3121 len = nb_uri * sizeof(struct lttng_uri);
3122
3123 if (nb_uri == 0) {
3124 ret = LTTNG_ERR_INVALID;
3125 goto error;
3126 }
3127
3128 uris = zmalloc(len);
3129 if (uris == NULL) {
3130 ret = LTTNG_ERR_FATAL;
3131 goto error;
3132 }
3133
3134 /* Receive variable len data */
3135 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3136 ret = lttcomm_recv_unix_sock(sock, uris, len);
3137 if (ret <= 0) {
3138 DBG("No URIs received from client... continuing");
3139 *sock_error = 1;
3140 ret = LTTNG_ERR_SESSION_FAIL;
3141 free(uris);
3142 goto error;
3143 }
3144
3145 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3146 nb_uri, uris);
3147 if (ret != LTTNG_OK) {
3148 free(uris);
3149 goto error;
3150 }
3151
3152 /*
3153 * XXX: 0 means that this URI should be applied on the session. Should
3154 * be a DOMAIN enuam.
3155 */
3156 if (cmd_ctx->lsm->domain.type == 0) {
3157 /* Add the URI for the UST session if a consumer is present. */
3158 if (cmd_ctx->session->ust_session &&
3159 cmd_ctx->session->ust_session->consumer) {
3160 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3161 nb_uri, uris);
3162 } else if (cmd_ctx->session->kernel_session &&
3163 cmd_ctx->session->kernel_session->consumer) {
3164 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3165 cmd_ctx->session, nb_uri, uris);
3166 }
3167 }
3168
3169 free(uris);
3170
3171 break;
3172 }
3173 case LTTNG_START_TRACE:
3174 {
3175 ret = cmd_start_trace(cmd_ctx->session);
3176 break;
3177 }
3178 case LTTNG_STOP_TRACE:
3179 {
3180 ret = cmd_stop_trace(cmd_ctx->session);
3181 break;
3182 }
3183 case LTTNG_CREATE_SESSION:
3184 {
3185 size_t nb_uri, len;
3186 struct lttng_uri *uris = NULL;
3187
3188 nb_uri = cmd_ctx->lsm->u.uri.size;
3189 len = nb_uri * sizeof(struct lttng_uri);
3190
3191 if (nb_uri > 0) {
3192 uris = zmalloc(len);
3193 if (uris == NULL) {
3194 ret = LTTNG_ERR_FATAL;
3195 goto error;
3196 }
3197
3198 /* Receive variable len data */
3199 DBG("Waiting for %zu URIs from client ...", nb_uri);
3200 ret = lttcomm_recv_unix_sock(sock, uris, len);
3201 if (ret <= 0) {
3202 DBG("No URIs received from client... continuing");
3203 *sock_error = 1;
3204 ret = LTTNG_ERR_SESSION_FAIL;
3205 free(uris);
3206 goto error;
3207 }
3208
3209 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3210 DBG("Creating session with ONE network URI is a bad call");
3211 ret = LTTNG_ERR_SESSION_FAIL;
3212 free(uris);
3213 goto error;
3214 }
3215 }
3216
3217 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3218 &cmd_ctx->creds, 0);
3219
3220 free(uris);
3221
3222 break;
3223 }
3224 case LTTNG_DESTROY_SESSION:
3225 {
3226 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
3227
3228 /* Set session to NULL so we do not unlock it after free. */
3229 cmd_ctx->session = NULL;
3230 break;
3231 }
3232 case LTTNG_LIST_DOMAINS:
3233 {
3234 ssize_t nb_dom;
3235 struct lttng_domain *domains;
3236
3237 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3238 if (nb_dom < 0) {
3239 /* Return value is a negative lttng_error_code. */
3240 ret = -nb_dom;
3241 goto error;
3242 }
3243
3244 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
3245 if (ret < 0) {
3246 free(domains);
3247 goto setup_error;
3248 }
3249
3250 /* Copy event list into message payload */
3251 memcpy(cmd_ctx->llm->payload, domains,
3252 nb_dom * sizeof(struct lttng_domain));
3253
3254 free(domains);
3255
3256 ret = LTTNG_OK;
3257 break;
3258 }
3259 case LTTNG_LIST_CHANNELS:
3260 {
3261 int nb_chan;
3262 struct lttng_channel *channels;
3263
3264 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3265 cmd_ctx->session, &channels);
3266 if (nb_chan < 0) {
3267 /* Return value is a negative lttng_error_code. */
3268 ret = -nb_chan;
3269 goto error;
3270 }
3271
3272 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
3273 if (ret < 0) {
3274 free(channels);
3275 goto setup_error;
3276 }
3277
3278 /* Copy event list into message payload */
3279 memcpy(cmd_ctx->llm->payload, channels,
3280 nb_chan * sizeof(struct lttng_channel));
3281
3282 free(channels);
3283
3284 ret = LTTNG_OK;
3285 break;
3286 }
3287 case LTTNG_LIST_EVENTS:
3288 {
3289 ssize_t nb_event;
3290 struct lttng_event *events = NULL;
3291
3292 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3293 cmd_ctx->lsm->u.list.channel_name, &events);
3294 if (nb_event < 0) {
3295 /* Return value is a negative lttng_error_code. */
3296 ret = -nb_event;
3297 goto error;
3298 }
3299
3300 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
3301 if (ret < 0) {
3302 free(events);
3303 goto setup_error;
3304 }
3305
3306 /* Copy event list into message payload */
3307 memcpy(cmd_ctx->llm->payload, events,
3308 nb_event * sizeof(struct lttng_event));
3309
3310 free(events);
3311
3312 ret = LTTNG_OK;
3313 break;
3314 }
3315 case LTTNG_LIST_SESSIONS:
3316 {
3317 unsigned int nr_sessions;
3318
3319 session_lock_list();
3320 nr_sessions = lttng_sessions_count(
3321 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3322 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3323
3324 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
3325 if (ret < 0) {
3326 session_unlock_list();
3327 goto setup_error;
3328 }
3329
3330 /* Filled the session array */
3331 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
3332 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3333 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
3334
3335 session_unlock_list();
3336
3337 ret = LTTNG_OK;
3338 break;
3339 }
3340 case LTTNG_CALIBRATE:
3341 {
3342 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3343 &cmd_ctx->lsm->u.calibrate);
3344 break;
3345 }
3346 case LTTNG_REGISTER_CONSUMER:
3347 {
3348 struct consumer_data *cdata;
3349
3350 switch (cmd_ctx->lsm->domain.type) {
3351 case LTTNG_DOMAIN_KERNEL:
3352 cdata = &kconsumer_data;
3353 break;
3354 default:
3355 ret = LTTNG_ERR_UND;
3356 goto error;
3357 }
3358
3359 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3360 cmd_ctx->lsm->u.reg.path, cdata);
3361 break;
3362 }
3363 case LTTNG_DATA_PENDING:
3364 {
3365 ret = cmd_data_pending(cmd_ctx->session);
3366 break;
3367 }
3368 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3369 {
3370 struct lttcomm_lttng_output_id reply;
3371
3372 ret = cmd_snapshot_add_output(cmd_ctx->session,
3373 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3374 if (ret != LTTNG_OK) {
3375 goto error;
3376 }
3377
3378 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3379 if (ret < 0) {
3380 goto setup_error;
3381 }
3382
3383 /* Copy output list into message payload */
3384 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3385 ret = LTTNG_OK;
3386 break;
3387 }
3388 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3389 {
3390 ret = cmd_snapshot_del_output(cmd_ctx->session,
3391 &cmd_ctx->lsm->u.snapshot_output.output);
3392 break;
3393 }
3394 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3395 {
3396 ssize_t nb_output;
3397 struct lttng_snapshot_output *outputs = NULL;
3398
3399 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3400 if (nb_output < 0) {
3401 ret = -nb_output;
3402 goto error;
3403 }
3404
3405 ret = setup_lttng_msg(cmd_ctx,
3406 nb_output * sizeof(struct lttng_snapshot_output));
3407 if (ret < 0) {
3408 free(outputs);
3409 goto setup_error;
3410 }
3411
3412 if (outputs) {
3413 /* Copy output list into message payload */
3414 memcpy(cmd_ctx->llm->payload, outputs,
3415 nb_output * sizeof(struct lttng_snapshot_output));
3416 free(outputs);
3417 }
3418
3419 ret = LTTNG_OK;
3420 break;
3421 }
3422 case LTTNG_SNAPSHOT_RECORD:
3423 {
3424 ret = cmd_snapshot_record(cmd_ctx->session,
3425 &cmd_ctx->lsm->u.snapshot_record.output,
3426 cmd_ctx->lsm->u.snapshot_record.wait);
3427 break;
3428 }
3429 case LTTNG_CREATE_SESSION_SNAPSHOT:
3430 {
3431 size_t nb_uri, len;
3432 struct lttng_uri *uris = NULL;
3433
3434 nb_uri = cmd_ctx->lsm->u.uri.size;
3435 len = nb_uri * sizeof(struct lttng_uri);
3436
3437 if (nb_uri > 0) {
3438 uris = zmalloc(len);
3439 if (uris == NULL) {
3440 ret = LTTNG_ERR_FATAL;
3441 goto error;
3442 }
3443
3444 /* Receive variable len data */
3445 DBG("Waiting for %zu URIs from client ...", nb_uri);
3446 ret = lttcomm_recv_unix_sock(sock, uris, len);
3447 if (ret <= 0) {
3448 DBG("No URIs received from client... continuing");
3449 *sock_error = 1;
3450 ret = LTTNG_ERR_SESSION_FAIL;
3451 free(uris);
3452 goto error;
3453 }
3454
3455 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3456 DBG("Creating session with ONE network URI is a bad call");
3457 ret = LTTNG_ERR_SESSION_FAIL;
3458 free(uris);
3459 goto error;
3460 }
3461 }
3462
3463 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3464 nb_uri, &cmd_ctx->creds);
3465 free(uris);
3466 break;
3467 }
3468 case LTTNG_CREATE_SESSION_LIVE:
3469 {
3470 size_t nb_uri, len;
3471 struct lttng_uri *uris = NULL;
3472
3473 nb_uri = cmd_ctx->lsm->u.uri.size;
3474 len = nb_uri * sizeof(struct lttng_uri);
3475
3476 if (nb_uri > 0) {
3477 uris = zmalloc(len);
3478 if (uris == NULL) {
3479 ret = LTTNG_ERR_FATAL;
3480 goto error;
3481 }
3482
3483 /* Receive variable len data */
3484 DBG("Waiting for %zu URIs from client ...", nb_uri);
3485 ret = lttcomm_recv_unix_sock(sock, uris, len);
3486 if (ret <= 0) {
3487 DBG("No URIs received from client... continuing");
3488 *sock_error = 1;
3489 ret = LTTNG_ERR_SESSION_FAIL;
3490 free(uris);
3491 goto error;
3492 }
3493
3494 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3495 DBG("Creating session with ONE network URI is a bad call");
3496 ret = LTTNG_ERR_SESSION_FAIL;
3497 free(uris);
3498 goto error;
3499 }
3500 }
3501
3502 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3503 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3504 free(uris);
3505 break;
3506 }
3507 default:
3508 ret = LTTNG_ERR_UND;
3509 break;
3510 }
3511
3512 error:
3513 if (cmd_ctx->llm == NULL) {
3514 DBG("Missing llm structure. Allocating one.");
3515 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
3516 goto setup_error;
3517 }
3518 }
3519 /* Set return code */
3520 cmd_ctx->llm->ret_code = ret;
3521 setup_error:
3522 if (cmd_ctx->session) {
3523 session_unlock(cmd_ctx->session);
3524 }
3525 if (need_tracing_session) {
3526 session_unlock_list();
3527 }
3528 init_setup_error:
3529 return ret;
3530 }
3531
3532 /*
3533 * Thread managing health check socket.
3534 */
3535 static void *thread_manage_health(void *data)
3536 {
3537 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
3538 uint32_t revents, nb_fd;
3539 struct lttng_poll_event events;
3540 struct health_comm_msg msg;
3541 struct health_comm_reply reply;
3542
3543 DBG("[thread] Manage health check started");
3544
3545 rcu_register_thread();
3546
3547 /* We might hit an error path before this is created. */
3548 lttng_poll_init(&events);
3549
3550 /* Create unix socket */
3551 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3552 if (sock < 0) {
3553 ERR("Unable to create health check Unix socket");
3554 ret = -1;
3555 goto error;
3556 }
3557
3558 if (is_root) {
3559 /* lttng health client socket path permissions */
3560 ret = chown(health_unix_sock_path, 0,
3561 utils_get_group_id(tracing_group_name));
3562 if (ret < 0) {
3563 ERR("Unable to set group on %s", health_unix_sock_path);
3564 PERROR("chown");
3565 ret = -1;
3566 goto error;
3567 }
3568
3569 ret = chmod(health_unix_sock_path,
3570 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3571 if (ret < 0) {
3572 ERR("Unable to set permissions on %s", health_unix_sock_path);
3573 PERROR("chmod");
3574 ret = -1;
3575 goto error;
3576 }
3577 }
3578
3579 /*
3580 * Set the CLOEXEC flag. Return code is useless because either way, the
3581 * show must go on.
3582 */
3583 (void) utils_set_fd_cloexec(sock);
3584
3585 ret = lttcomm_listen_unix_sock(sock);
3586 if (ret < 0) {
3587 goto error;
3588 }
3589
3590 /*
3591 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3592 * more will be added to this poll set.
3593 */
3594 ret = sessiond_set_thread_pollset(&events, 2);
3595 if (ret < 0) {
3596 goto error;
3597 }
3598
3599 /* Add the application registration socket */
3600 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3601 if (ret < 0) {
3602 goto error;
3603 }
3604
3605 lttng_sessiond_notify_ready();
3606
3607 while (1) {
3608 DBG("Health check ready");
3609
3610 /* Inifinite blocking call, waiting for transmission */
3611 restart:
3612 ret = lttng_poll_wait(&events, -1);
3613 if (ret < 0) {
3614 /*
3615 * Restart interrupted system call.
3616 */
3617 if (errno == EINTR) {
3618 goto restart;
3619 }
3620 goto error;
3621 }
3622
3623 nb_fd = ret;
3624
3625 for (i = 0; i < nb_fd; i++) {
3626 /* Fetch once the poll data */
3627 revents = LTTNG_POLL_GETEV(&events, i);
3628 pollfd = LTTNG_POLL_GETFD(&events, i);
3629
3630 /* Thread quit pipe has been closed. Killing thread. */
3631 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3632 if (ret) {
3633 err = 0;
3634 goto exit;
3635 }
3636
3637 /* Event on the registration socket */
3638 if (pollfd == sock) {
3639 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3640 ERR("Health socket poll error");
3641 goto error;
3642 }
3643 }
3644 }
3645
3646 new_sock = lttcomm_accept_unix_sock(sock);
3647 if (new_sock < 0) {
3648 goto error;
3649 }
3650
3651 /*
3652 * Set the CLOEXEC flag. Return code is useless because either way, the
3653 * show must go on.
3654 */
3655 (void) utils_set_fd_cloexec(new_sock);
3656
3657 DBG("Receiving data from client for health...");
3658 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3659 if (ret <= 0) {
3660 DBG("Nothing recv() from client... continuing");
3661 ret = close(new_sock);
3662 if (ret) {
3663 PERROR("close");
3664 }
3665 new_sock = -1;
3666 continue;
3667 }
3668
3669 rcu_thread_online();
3670
3671 reply.ret_code = 0;
3672 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3673 /*
3674 * health_check_state returns 0 if health is
3675 * bad.
3676 */
3677 if (!health_check_state(health_sessiond, i)) {
3678 reply.ret_code |= 1ULL << i;
3679 }
3680 }
3681
3682 DBG2("Health check return value %" PRIx64, reply.ret_code);
3683
3684 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3685 if (ret < 0) {
3686 ERR("Failed to send health data back to client");
3687 }
3688
3689 /* End of transmission */
3690 ret = close(new_sock);
3691 if (ret) {
3692 PERROR("close");
3693 }
3694 new_sock = -1;
3695 }
3696
3697 exit:
3698 error:
3699 if (err) {
3700 ERR("Health error occurred in %s", __func__);
3701 }
3702 DBG("Health check thread dying");
3703 unlink(health_unix_sock_path);
3704 if (sock >= 0) {
3705 ret = close(sock);
3706 if (ret) {
3707 PERROR("close");
3708 }
3709 }
3710
3711 lttng_poll_clean(&events);
3712
3713 rcu_unregister_thread();
3714 return NULL;
3715 }
3716
3717 /*
3718 * This thread manage all clients request using the unix client socket for
3719 * communication.
3720 */
3721 static void *thread_manage_clients(void *data)
3722 {
3723 int sock = -1, ret, i, pollfd, err = -1;
3724 int sock_error;
3725 uint32_t revents, nb_fd;
3726 struct command_ctx *cmd_ctx = NULL;
3727 struct lttng_poll_event events;
3728
3729 DBG("[thread] Manage client started");
3730
3731 rcu_register_thread();
3732
3733 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
3734
3735 health_code_update();
3736
3737 ret = lttcomm_listen_unix_sock(client_sock);
3738 if (ret < 0) {
3739 goto error_listen;
3740 }
3741
3742 /*
3743 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3744 * more will be added to this poll set.
3745 */
3746 ret = sessiond_set_thread_pollset(&events, 2);
3747 if (ret < 0) {
3748 goto error_create_poll;
3749 }
3750
3751 /* Add the application registration socket */
3752 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3753 if (ret < 0) {
3754 goto error;
3755 }
3756
3757 lttng_sessiond_notify_ready();
3758
3759 /* This testpoint is after we signal readiness to the parent. */
3760 if (testpoint(sessiond_thread_manage_clients)) {
3761 goto error;
3762 }
3763
3764 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
3765 goto error;
3766 }
3767
3768 health_code_update();
3769
3770 while (1) {
3771 DBG("Accepting client command ...");
3772
3773 /* Inifinite blocking call, waiting for transmission */
3774 restart:
3775 health_poll_entry();
3776 ret = lttng_poll_wait(&events, -1);
3777 health_poll_exit();
3778 if (ret < 0) {
3779 /*
3780 * Restart interrupted system call.
3781 */
3782 if (errno == EINTR) {
3783 goto restart;
3784 }
3785 goto error;
3786 }
3787
3788 nb_fd = ret;
3789
3790 for (i = 0; i < nb_fd; i++) {
3791 /* Fetch once the poll data */
3792 revents = LTTNG_POLL_GETEV(&events, i);
3793 pollfd = LTTNG_POLL_GETFD(&events, i);
3794
3795 health_code_update();
3796
3797 /* Thread quit pipe has been closed. Killing thread. */
3798 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
3799 if (ret) {
3800 err = 0;
3801 goto exit;
3802 }
3803
3804 /* Event on the registration socket */
3805 if (pollfd == client_sock) {
3806 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3807 ERR("Client socket poll error");
3808 goto error;
3809 }
3810 }
3811 }
3812
3813 DBG("Wait for client response");
3814
3815 health_code_update();
3816
3817 sock = lttcomm_accept_unix_sock(client_sock);
3818 if (sock < 0) {
3819 goto error;
3820 }
3821
3822 /*
3823 * Set the CLOEXEC flag. Return code is useless because either way, the
3824 * show must go on.
3825 */
3826 (void) utils_set_fd_cloexec(sock);
3827
3828 /* Set socket option for credentials retrieval */
3829 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3830 if (ret < 0) {
3831 goto error;
3832 }
3833
3834 /* Allocate context command to process the client request */
3835 cmd_ctx = zmalloc(sizeof(struct command_ctx));
3836 if (cmd_ctx == NULL) {
3837 PERROR("zmalloc cmd_ctx");
3838 goto error;
3839 }
3840
3841 /* Allocate data buffer for reception */
3842 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
3843 if (cmd_ctx->lsm == NULL) {
3844 PERROR("zmalloc cmd_ctx->lsm");
3845 goto error;
3846 }
3847
3848 cmd_ctx->llm = NULL;
3849 cmd_ctx->session = NULL;
3850
3851 health_code_update();
3852
3853 /*
3854 * Data is received from the lttng client. The struct
3855 * lttcomm_session_msg (lsm) contains the command and data request of
3856 * the client.
3857 */
3858 DBG("Receiving data from client ...");
3859 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3860 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
3861 if (ret <= 0) {
3862 DBG("Nothing recv() from client... continuing");
3863 ret = close(sock);
3864 if (ret) {
3865 PERROR("close");
3866 }
3867 sock = -1;
3868 clean_command_ctx(&cmd_ctx);
3869 continue;
3870 }
3871
3872 health_code_update();
3873
3874 // TODO: Validate cmd_ctx including sanity check for
3875 // security purpose.
3876
3877 rcu_thread_online();
3878 /*
3879 * This function dispatch the work to the kernel or userspace tracer
3880 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3881 * informations for the client. The command context struct contains
3882 * everything this function may needs.
3883 */
3884 ret = process_client_msg(cmd_ctx, sock, &sock_error);
3885 rcu_thread_offline();
3886 if (ret < 0) {
3887 ret = close(sock);
3888 if (ret) {
3889 PERROR("close");
3890 }
3891 sock = -1;
3892 /*
3893 * TODO: Inform client somehow of the fatal error. At
3894 * this point, ret < 0 means that a zmalloc failed
3895 * (ENOMEM). Error detected but still accept
3896 * command, unless a socket error has been
3897 * detected.
3898 */
3899 clean_command_ctx(&cmd_ctx);
3900 continue;
3901 }
3902
3903 health_code_update();
3904
3905 DBG("Sending response (size: %d, retcode: %s)",
3906 cmd_ctx->lttng_msg_size,
3907 lttng_strerror(-cmd_ctx->llm->ret_code));
3908 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
3909 if (ret < 0) {
3910 ERR("Failed to send data back to client");
3911 }
3912
3913 /* End of transmission */
3914 ret = close(sock);
3915 if (ret) {
3916 PERROR("close");
3917 }
3918 sock = -1;
3919
3920 clean_command_ctx(&cmd_ctx);
3921
3922 health_code_update();
3923 }
3924
3925 exit:
3926 error:
3927 if (sock >= 0) {
3928 ret = close(sock);
3929 if (ret) {
3930 PERROR("close");
3931 }
3932 }
3933
3934 lttng_poll_clean(&events);
3935 clean_command_ctx(&cmd_ctx);
3936
3937 error_listen:
3938 error_create_poll:
3939 unlink(client_unix_sock_path);
3940 if (client_sock >= 0) {
3941 ret = close(client_sock);
3942 if (ret) {
3943 PERROR("close");
3944 }
3945 }
3946
3947 if (err) {
3948 health_error();
3949 ERR("Health error occurred in %s", __func__);
3950 }
3951
3952 health_unregister(health_sessiond);
3953
3954 DBG("Client thread dying");
3955
3956 rcu_unregister_thread();
3957 return NULL;
3958 }
3959
3960
3961 /*
3962 * usage function on stderr
3963 */
3964 static void usage(void)
3965 {
3966 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
3967 fprintf(stderr, " -h, --help Display this usage.\n");
3968 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3969 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3970 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3971 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
3972 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3973 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3974 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3975 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
3976 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3977 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3978 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3979 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
3980 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3981 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
3982 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3983 fprintf(stderr, " -V, --version Show version number.\n");
3984 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
3985 fprintf(stderr, " -q, --quiet No output at all.\n");
3986 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
3987 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3988 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
3989 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
3990 fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
3991 }
3992
3993 /*
3994 * daemon argument parsing
3995 */
3996 static int parse_args(int argc, char **argv)
3997 {
3998 int c;
3999
4000 static struct option long_options[] = {
4001 { "client-sock", 1, 0, 'c' },
4002 { "apps-sock", 1, 0, 'a' },
4003 { "kconsumerd-cmd-sock", 1, 0, 'C' },
4004 { "kconsumerd-err-sock", 1, 0, 'E' },
4005 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
4006 { "ustconsumerd32-err-sock", 1, 0, 'H' },
4007 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
4008 { "ustconsumerd64-err-sock", 1, 0, 'F' },
4009 { "consumerd32-path", 1, 0, 'u' },
4010 { "consumerd32-libdir", 1, 0, 'U' },
4011 { "consumerd64-path", 1, 0, 't' },
4012 { "consumerd64-libdir", 1, 0, 'T' },
4013 { "daemonize", 0, 0, 'd' },
4014 { "sig-parent", 0, 0, 'S' },
4015 { "help", 0, 0, 'h' },
4016 { "group", 1, 0, 'g' },
4017 { "version", 0, 0, 'V' },
4018 { "quiet", 0, 0, 'q' },
4019 { "verbose", 0, 0, 'v' },
4020 { "verbose-consumer", 0, 0, 'Z' },
4021 { "no-kernel", 0, 0, 'N' },
4022 { "pidfile", 1, 0, 'p' },
4023 { "jul-tcp-port", 1, 0, 'J' },
4024 { NULL, 0, 0, 0 }
4025 };
4026
4027 while (1) {
4028 int option_index = 0;
4029 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
4030 long_options, &option_index);
4031 if (c == -1) {
4032 break;
4033 }
4034
4035 switch (c) {
4036 case 0:
4037 fprintf(stderr, "option %s", long_options[option_index].name);
4038 if (optarg) {
4039 fprintf(stderr, " with arg %s\n", optarg);
4040 }
4041 break;
4042 case 'c':
4043 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
4044 break;
4045 case 'a':
4046 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
4047 break;
4048 case 'd':
4049 opt_daemon = 1;
4050 break;
4051 case 'g':
4052 tracing_group_name = optarg;
4053 break;
4054 case 'h':
4055 usage();
4056 exit(EXIT_FAILURE);
4057 case 'V':
4058 fprintf(stdout, "%s\n", VERSION);
4059 exit(EXIT_SUCCESS);
4060 case 'S':
4061 opt_sig_parent = 1;
4062 break;
4063 case 'E':
4064 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4065 break;
4066 case 'C':
4067 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4068 break;
4069 case 'F':
4070 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4071 break;
4072 case 'D':
4073 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4074 break;
4075 case 'H':
4076 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4077 break;
4078 case 'G':
4079 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4080 break;
4081 case 'N':
4082 opt_no_kernel = 1;
4083 break;
4084 case 'q':
4085 lttng_opt_quiet = 1;
4086 break;
4087 case 'v':
4088 /* Verbose level can increase using multiple -v */
4089 lttng_opt_verbose += 1;
4090 break;
4091 case 'Z':
4092 opt_verbose_consumer += 1;
4093 break;
4094 case 'u':
4095 consumerd32_bin= optarg;
4096 break;
4097 case 'U':
4098 consumerd32_libdir = optarg;
4099 break;
4100 case 't':
4101 consumerd64_bin = optarg;
4102 break;
4103 case 'T':
4104 consumerd64_libdir = optarg;
4105 break;
4106 case 'p':
4107 opt_pidfile = optarg;
4108 break;
4109 case 'J': /* JUL TCP port. */
4110 {
4111 unsigned long v;
4112
4113 errno = 0;
4114 v = strtoul(optarg, NULL, 0);
4115 if (errno != 0 || !isdigit(optarg[0])) {
4116 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg);
4117 return -1;
4118 }
4119 if (v == 0 || v >= 65535) {
4120 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg);
4121 return -1;
4122 }
4123 jul_tcp_port = (uint32_t) v;
4124 DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
4125 break;
4126 }
4127 default:
4128 /* Unknown option or other error.
4129 * Error is printed by getopt, just return */
4130 return -1;
4131 }
4132 }
4133
4134 return 0;
4135 }
4136
4137 /*
4138 * Creates the two needed socket by the daemon.
4139 * apps_sock - The communication socket for all UST apps.
4140 * client_sock - The communication of the cli tool (lttng).
4141 */
4142 static int init_daemon_socket(void)
4143 {
4144 int ret = 0;
4145 mode_t old_umask;
4146
4147 old_umask = umask(0);
4148
4149 /* Create client tool unix socket */
4150 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4151 if (client_sock < 0) {
4152 ERR("Create unix sock failed: %s", client_unix_sock_path);
4153 ret = -1;
4154 goto end;
4155 }
4156
4157 /* Set the cloexec flag */
4158 ret = utils_set_fd_cloexec(client_sock);
4159 if (ret < 0) {
4160 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4161 "Continuing but note that the consumer daemon will have a "
4162 "reference to this socket on exec()", client_sock);
4163 }
4164
4165 /* File permission MUST be 660 */
4166 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4167 if (ret < 0) {
4168 ERR("Set file permissions failed: %s", client_unix_sock_path);
4169 PERROR("chmod");
4170 goto end;
4171 }
4172
4173 /* Create the application unix socket */
4174 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4175 if (apps_sock < 0) {
4176 ERR("Create unix sock failed: %s", apps_unix_sock_path);
4177 ret = -1;
4178 goto end;
4179 }
4180
4181 /* Set the cloexec flag */
4182 ret = utils_set_fd_cloexec(apps_sock);
4183 if (ret < 0) {
4184 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4185 "Continuing but note that the consumer daemon will have a "
4186 "reference to this socket on exec()", apps_sock);
4187 }
4188
4189 /* File permission MUST be 666 */
4190 ret = chmod(apps_unix_sock_path,
4191 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
4192 if (ret < 0) {
4193 ERR("Set file permissions failed: %s", apps_unix_sock_path);
4194 PERROR("chmod");
4195 goto end;
4196 }
4197
4198 DBG3("Session daemon client socket %d and application socket %d created",
4199 client_sock, apps_sock);
4200
4201 end:
4202 umask(old_umask);
4203 return ret;
4204 }
4205
4206 /*
4207 * Check if the global socket is available, and if a daemon is answering at the
4208 * other side. If yes, error is returned.
4209 */
4210 static int check_existing_daemon(void)
4211 {
4212 /* Is there anybody out there ? */
4213 if (lttng_session_daemon_alive()) {
4214 return -EEXIST;
4215 }
4216
4217 return 0;
4218 }
4219
4220 /*
4221 * Set the tracing group gid onto the client socket.
4222 *
4223 * Race window between mkdir and chown is OK because we are going from more
4224 * permissive (root.root) to less permissive (root.tracing).
4225 */
4226 static int set_permissions(char *rundir)
4227 {
4228 int ret;
4229 gid_t gid;
4230
4231 gid = utils_get_group_id(tracing_group_name);
4232
4233 /* Set lttng run dir */
4234 ret = chown(rundir, 0, gid);
4235 if (ret < 0) {
4236 ERR("Unable to set group on %s", rundir);
4237 PERROR("chown");
4238 }
4239
4240 /*
4241 * Ensure all applications and tracing group can search the run
4242 * dir. Allow everyone to read the directory, since it does not
4243 * buy us anything to hide its content.
4244 */
4245 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
4246 if (ret < 0) {
4247 ERR("Unable to set permissions on %s", rundir);
4248 PERROR("chmod");
4249 }
4250
4251 /* lttng client socket path */
4252 ret = chown(client_unix_sock_path, 0, gid);
4253 if (ret < 0) {
4254 ERR("Unable to set group on %s", client_unix_sock_path);
4255 PERROR("chown");
4256 }
4257
4258 /* kconsumer error socket path */
4259 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
4260 if (ret < 0) {
4261 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
4262 PERROR("chown");
4263 }
4264
4265 /* 64-bit ustconsumer error socket path */
4266 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
4267 if (ret < 0) {
4268 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
4269 PERROR("chown");
4270 }
4271
4272 /* 32-bit ustconsumer compat32 error socket path */
4273 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
4274 if (ret < 0) {
4275 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
4276 PERROR("chown");
4277 }
4278
4279 DBG("All permissions are set");
4280
4281 return ret;
4282 }
4283
4284 /*
4285 * Create the lttng run directory needed for all global sockets and pipe.
4286 */
4287 static int create_lttng_rundir(const char *rundir)
4288 {
4289 int ret;
4290
4291 DBG3("Creating LTTng run directory: %s", rundir);
4292
4293 ret = mkdir(rundir, S_IRWXU);
4294 if (ret < 0) {
4295 if (errno != EEXIST) {
4296 ERR("Unable to create %s", rundir);
4297 goto error;
4298 } else {
4299 ret = 0;
4300 }
4301 }
4302
4303 error:
4304 return ret;
4305 }
4306
4307 /*
4308 * Setup sockets and directory needed by the kconsumerd communication with the
4309 * session daemon.
4310 */
4311 static int set_consumer_sockets(struct consumer_data *consumer_data,
4312 const char *rundir)
4313 {
4314 int ret;
4315 char path[PATH_MAX];
4316
4317 switch (consumer_data->type) {
4318 case LTTNG_CONSUMER_KERNEL:
4319 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
4320 break;
4321 case LTTNG_CONSUMER64_UST:
4322 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
4323 break;
4324 case LTTNG_CONSUMER32_UST:
4325 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
4326 break;
4327 default:
4328 ERR("Consumer type unknown");
4329 ret = -EINVAL;
4330 goto error;
4331 }
4332
4333 DBG2("Creating consumer directory: %s", path);
4334
4335 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
4336 if (ret < 0) {
4337 if (errno != EEXIST) {
4338 PERROR("mkdir");
4339 ERR("Failed to create %s", path);
4340 goto error;
4341 }
4342 ret = -1;
4343 }
4344 if (is_root) {
4345 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4346 if (ret < 0) {
4347 ERR("Unable to set group on %s", path);
4348 PERROR("chown");
4349 goto error;
4350 }
4351 }
4352
4353 /* Create the kconsumerd error unix socket */
4354 consumer_data->err_sock =
4355 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4356 if (consumer_data->err_sock < 0) {
4357 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
4358 ret = -1;
4359 goto error;
4360 }
4361
4362 /*
4363 * Set the CLOEXEC flag. Return code is useless because either way, the
4364 * show must go on.
4365 */
4366 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4367 if (ret < 0) {
4368 PERROR("utils_set_fd_cloexec");
4369 /* continue anyway */
4370 }
4371
4372 /* File permission MUST be 660 */
4373 ret = chmod(consumer_data->err_unix_sock_path,
4374 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4375 if (ret < 0) {
4376 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
4377 PERROR("chmod");
4378 goto error;
4379 }
4380
4381 error:
4382 return ret;
4383 }
4384
4385 /*
4386 * Signal handler for the daemon
4387 *
4388 * Simply stop all worker threads, leaving main() return gracefully after
4389 * joining all threads and calling cleanup().
4390 */
4391 static void sighandler(int sig)
4392 {
4393 switch (sig) {
4394 case SIGPIPE:
4395 DBG("SIGPIPE caught");
4396 return;
4397 case SIGINT:
4398 DBG("SIGINT caught");
4399 stop_threads();
4400 break;
4401 case SIGTERM:
4402 DBG("SIGTERM caught");
4403 stop_threads();
4404 break;
4405 case SIGUSR1:
4406 CMM_STORE_SHARED(recv_child_signal, 1);
4407 break;
4408 default:
4409 break;
4410 }
4411 }
4412
4413 /*
4414 * Setup signal handler for :
4415 * SIGINT, SIGTERM, SIGPIPE
4416 */
4417 static int set_signal_handler(void)
4418 {
4419 int ret = 0;
4420 struct sigaction sa;
4421 sigset_t sigset;
4422
4423 if ((ret = sigemptyset(&sigset)) < 0) {
4424 PERROR("sigemptyset");
4425 return ret;
4426 }
4427
4428 sa.sa_handler = sighandler;
4429 sa.sa_mask = sigset;
4430 sa.sa_flags = 0;
4431 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
4432 PERROR("sigaction");
4433 return ret;
4434 }
4435
4436 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
4437 PERROR("sigaction");
4438 return ret;
4439 }
4440
4441 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
4442 PERROR("sigaction");
4443 return ret;
4444 }
4445
4446 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4447 PERROR("sigaction");
4448 return ret;
4449 }
4450
4451 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4452
4453 return ret;
4454 }
4455
4456 /*
4457 * Set open files limit to unlimited. This daemon can open a large number of
4458 * file descriptors in order to consumer multiple kernel traces.
4459 */
4460 static void set_ulimit(void)
4461 {
4462 int ret;
4463 struct rlimit lim;
4464
4465 /* The kernel does not allowed an infinite limit for open files */
4466 lim.rlim_cur = 65535;
4467 lim.rlim_max = 65535;
4468
4469 ret = setrlimit(RLIMIT_NOFILE, &lim);
4470 if (ret < 0) {
4471 PERROR("failed to set open files limit");
4472 }
4473 }
4474
4475 /*
4476 * Write pidfile using the rundir and opt_pidfile.
4477 */
4478 static void write_pidfile(void)
4479 {
4480 int ret;
4481 char pidfile_path[PATH_MAX];
4482
4483 assert(rundir);
4484
4485 if (opt_pidfile) {
4486 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4487 } else {
4488 /* Build pidfile path from rundir and opt_pidfile. */
4489 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4490 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4491 if (ret < 0) {
4492 PERROR("snprintf pidfile path");
4493 goto error;
4494 }
4495 }
4496
4497 /*
4498 * Create pid file in rundir. Return value is of no importance. The
4499 * execution will continue even though we are not able to write the file.
4500 */
4501 (void) utils_create_pid_file(getpid(), pidfile_path);
4502
4503 error:
4504 return;
4505 }
4506
4507 /*
4508 * Write JUL TCP port using the rundir.
4509 */
4510 static void write_julport(void)
4511 {
4512 int ret;
4513 char path[PATH_MAX];
4514
4515 assert(rundir);
4516
4517 ret = snprintf(path, sizeof(path), "%s/"
4518 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
4519 if (ret < 0) {
4520 PERROR("snprintf julport path");
4521 goto error;
4522 }
4523
4524 /*
4525 * Create TCP JUL port file in rundir. Return value is of no importance.
4526 * The execution will continue even though we are not able to write the
4527 * file.
4528 */
4529 (void) utils_create_pid_file(jul_tcp_port, path);
4530
4531 error:
4532 return;
4533 }
4534
4535 /*
4536 * main
4537 */
4538 int main(int argc, char **argv)
4539 {
4540 int ret = 0;
4541 void *status;
4542 const char *home_path, *env_app_timeout;
4543
4544 init_kernel_workarounds();
4545
4546 rcu_register_thread();
4547
4548 if ((ret = set_signal_handler()) < 0) {
4549 goto error;
4550 }
4551
4552 setup_consumerd_path();
4553
4554 page_size = sysconf(_SC_PAGESIZE);
4555 if (page_size < 0) {
4556 PERROR("sysconf _SC_PAGESIZE");
4557 page_size = LONG_MAX;
4558 WARN("Fallback page size to %ld", page_size);
4559 }
4560
4561 /* Parse arguments */
4562 progname = argv[0];
4563 if ((ret = parse_args(argc, argv)) < 0) {
4564 goto error;
4565 }
4566
4567 /* Daemonize */
4568 if (opt_daemon || opt_background) {
4569 int i;
4570
4571 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
4572 !opt_background);
4573 if (ret < 0) {
4574 goto error;
4575 }
4576
4577 /*
4578 * We are in the child. Make sure all other file descriptors are
4579 * closed, in case we are called with more opened file descriptors than
4580 * the standard ones.
4581 */
4582 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4583 (void) close(i);
4584 }
4585 }
4586
4587 /* Create thread quit pipe */
4588 if ((ret = init_thread_quit_pipe()) < 0) {
4589 goto error;
4590 }
4591
4592 /* Check if daemon is UID = 0 */
4593 is_root = !getuid();
4594
4595 if (is_root) {
4596 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
4597
4598 /* Create global run dir with root access */
4599 ret = create_lttng_rundir(rundir);
4600 if (ret < 0) {
4601 goto error;
4602 }
4603
4604 if (strlen(apps_unix_sock_path) == 0) {
4605 snprintf(apps_unix_sock_path, PATH_MAX,
4606 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
4607 }
4608
4609 if (strlen(client_unix_sock_path) == 0) {
4610 snprintf(client_unix_sock_path, PATH_MAX,
4611 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4612 }
4613
4614 /* Set global SHM for ust */
4615 if (strlen(wait_shm_path) == 0) {
4616 snprintf(wait_shm_path, PATH_MAX,
4617 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4618 }
4619
4620 if (strlen(health_unix_sock_path) == 0) {
4621 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4622 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4623 }
4624
4625 /* Setup kernel consumerd path */
4626 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
4627 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
4628 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
4629 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
4630
4631 DBG2("Kernel consumer err path: %s",
4632 kconsumer_data.err_unix_sock_path);
4633 DBG2("Kernel consumer cmd path: %s",
4634 kconsumer_data.cmd_unix_sock_path);
4635 } else {
4636 home_path = utils_get_home_dir();
4637 if (home_path == NULL) {
4638 /* TODO: Add --socket PATH option */
4639 ERR("Can't get HOME directory for sockets creation.");
4640 ret = -EPERM;
4641 goto error;
4642 }
4643
4644 /*
4645 * Create rundir from home path. This will create something like
4646 * $HOME/.lttng
4647 */
4648 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
4649 if (ret < 0) {
4650 ret = -ENOMEM;
4651 goto error;
4652 }
4653
4654 ret = create_lttng_rundir(rundir);
4655 if (ret < 0) {
4656 goto error;
4657 }
4658
4659 if (strlen(apps_unix_sock_path) == 0) {
4660 snprintf(apps_unix_sock_path, PATH_MAX,
4661 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
4662 }
4663
4664 /* Set the cli tool unix socket path */
4665 if (strlen(client_unix_sock_path) == 0) {
4666 snprintf(client_unix_sock_path, PATH_MAX,
4667 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
4668 }
4669
4670 /* Set global SHM for ust */
4671 if (strlen(wait_shm_path) == 0) {
4672 snprintf(wait_shm_path, PATH_MAX,
4673 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
4674 }
4675
4676 /* Set health check Unix path */
4677 if (strlen(health_unix_sock_path) == 0) {
4678 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4679 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4680 }
4681 }
4682
4683 /* Set consumer initial state */
4684 kernel_consumerd_state = CONSUMER_STOPPED;
4685 ust_consumerd_state = CONSUMER_STOPPED;
4686
4687 DBG("Client socket path %s", client_unix_sock_path);
4688 DBG("Application socket path %s", apps_unix_sock_path);
4689 DBG("Application wait path %s", wait_shm_path);
4690 DBG("LTTng run directory path: %s", rundir);
4691
4692 /* 32 bits consumerd path setup */
4693 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
4694 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
4695 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
4696 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
4697
4698 DBG2("UST consumer 32 bits err path: %s",
4699 ustconsumer32_data.err_unix_sock_path);
4700 DBG2("UST consumer 32 bits cmd path: %s",
4701 ustconsumer32_data.cmd_unix_sock_path);
4702
4703 /* 64 bits consumerd path setup */
4704 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
4705 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
4706 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
4707 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
4708
4709 DBG2("UST consumer 64 bits err path: %s",
4710 ustconsumer64_data.err_unix_sock_path);
4711 DBG2("UST consumer 64 bits cmd path: %s",
4712 ustconsumer64_data.cmd_unix_sock_path);
4713
4714 /*
4715 * See if daemon already exist.
4716 */
4717 if ((ret = check_existing_daemon()) < 0) {
4718 ERR("Already running daemon.\n");
4719 /*
4720 * We do not goto exit because we must not cleanup()
4721 * because a daemon is already running.
4722 */
4723 goto error;
4724 }
4725
4726 /*
4727 * Init UST app hash table. Alloc hash table before this point since
4728 * cleanup() can get called after that point.
4729 */
4730 ust_app_ht_alloc();
4731
4732 /* Initialize JUL domain subsystem. */
4733 if ((ret = jul_init()) < 0) {
4734 /* ENOMEM at this point. */
4735 goto error;
4736 }
4737
4738 /* After this point, we can safely call cleanup() with "goto exit" */
4739
4740 /*
4741 * These actions must be executed as root. We do that *after* setting up
4742 * the sockets path because we MUST make the check for another daemon using
4743 * those paths *before* trying to set the kernel consumer sockets and init
4744 * kernel tracer.
4745 */
4746 if (is_root) {
4747 ret = set_consumer_sockets(&kconsumer_data, rundir);
4748 if (ret < 0) {
4749 goto exit;
4750 }
4751
4752 /* Setup kernel tracer */
4753 if (!opt_no_kernel) {
4754 init_kernel_tracer();
4755 }
4756
4757 /* Set ulimit for open files */
4758 set_ulimit();
4759 }
4760 /* init lttng_fd tracking must be done after set_ulimit. */
4761 lttng_fd_init();
4762
4763 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4764 if (ret < 0) {
4765 goto exit;
4766 }
4767
4768 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4769 if (ret < 0) {
4770 goto exit;
4771 }
4772
4773 /* Setup the needed unix socket */
4774 if ((ret = init_daemon_socket()) < 0) {
4775 goto exit;
4776 }
4777
4778 /* Set credentials to socket */
4779 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
4780 goto exit;
4781 }
4782
4783 /* Get parent pid if -S, --sig-parent is specified. */
4784 if (opt_sig_parent) {
4785 ppid = getppid();
4786 }
4787
4788 /* Setup the kernel pipe for waking up the kernel thread */
4789 if (is_root && !opt_no_kernel) {
4790 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4791 goto exit;
4792 }
4793 }
4794
4795 /* Setup the thread ht_cleanup communication pipe. */
4796 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4797 goto exit;
4798 }
4799
4800 /* Setup the thread apps communication pipe. */
4801 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
4802 goto exit;
4803 }
4804
4805 /* Setup the thread apps notify communication pipe. */
4806 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4807 goto exit;
4808 }
4809
4810 /* Initialize global buffer per UID and PID registry. */
4811 buffer_reg_init_uid_registry();
4812 buffer_reg_init_pid_registry();
4813
4814 /* Init UST command queue. */
4815 cds_wfq_init(&ust_cmd_queue.queue);
4816
4817 /*
4818 * Get session list pointer. This pointer MUST NOT be free(). This list is
4819 * statically declared in session.c
4820 */
4821 session_list_ptr = session_get_list();
4822
4823 /* Set up max poll set size */
4824 lttng_poll_set_max_size();
4825
4826 cmd_init();
4827
4828 /* Check for the application socket timeout env variable. */
4829 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4830 if (env_app_timeout) {
4831 app_socket_timeout = atoi(env_app_timeout);
4832 } else {
4833 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4834 }
4835
4836 write_pidfile();
4837 write_julport();
4838
4839 /* Initialize communication library */
4840 lttcomm_init();
4841 /* This is to get the TCP timeout value. */
4842 lttcomm_inet_init();
4843
4844 /*
4845 * Initialize the health check subsystem. This call should set the
4846 * appropriate time values.
4847 */
4848 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
4849 if (!health_sessiond) {
4850 PERROR("health_app_create error");
4851 goto exit_health_sessiond_cleanup;
4852 }
4853
4854 /* Create thread to clean up RCU hash tables */
4855 ret = pthread_create(&ht_cleanup_thread, NULL,
4856 thread_ht_cleanup, (void *) NULL);
4857 if (ret != 0) {
4858 PERROR("pthread_create ht_cleanup");
4859 goto exit_ht_cleanup;
4860 }
4861
4862 /* Create health-check thread */
4863 ret = pthread_create(&health_thread, NULL,
4864 thread_manage_health, (void *) NULL);
4865 if (ret != 0) {
4866 PERROR("pthread_create health");
4867 goto exit_health;
4868 }
4869
4870 /* Create thread to manage the client socket */
4871 ret = pthread_create(&client_thread, NULL,
4872 thread_manage_clients, (void *) NULL);
4873 if (ret != 0) {
4874 PERROR("pthread_create clients");
4875 goto exit_client;
4876 }
4877
4878 /* Create thread to dispatch registration */
4879 ret = pthread_create(&dispatch_thread, NULL,
4880 thread_dispatch_ust_registration, (void *) NULL);
4881 if (ret != 0) {
4882 PERROR("pthread_create dispatch");
4883 goto exit_dispatch;
4884 }
4885
4886 /* Create thread to manage application registration. */
4887 ret = pthread_create(&reg_apps_thread, NULL,
4888 thread_registration_apps, (void *) NULL);
4889 if (ret != 0) {
4890 PERROR("pthread_create registration");
4891 goto exit_reg_apps;
4892 }
4893
4894 /* Create thread to manage application socket */
4895 ret = pthread_create(&apps_thread, NULL,
4896 thread_manage_apps, (void *) NULL);
4897 if (ret != 0) {
4898 PERROR("pthread_create apps");
4899 goto exit_apps;
4900 }
4901
4902 /* Create thread to manage application notify socket */
4903 ret = pthread_create(&apps_notify_thread, NULL,
4904 ust_thread_manage_notify, (void *) NULL);
4905 if (ret != 0) {
4906 PERROR("pthread_create apps");
4907 goto exit_apps_notify;
4908 }
4909
4910 /* Create JUL registration thread. */
4911 ret = pthread_create(&jul_reg_thread, NULL,
4912 jul_thread_manage_registration, (void *) NULL);
4913 if (ret != 0) {
4914 PERROR("pthread_create apps");
4915 goto exit_jul_reg;
4916 }
4917
4918 /* Don't start this thread if kernel tracing is not requested nor root */
4919 if (is_root && !opt_no_kernel) {
4920 /* Create kernel thread to manage kernel event */
4921 ret = pthread_create(&kernel_thread, NULL,
4922 thread_manage_kernel, (void *) NULL);
4923 if (ret != 0) {
4924 PERROR("pthread_create kernel");
4925 goto exit_kernel;
4926 }
4927
4928 ret = pthread_join(kernel_thread, &status);
4929 if (ret != 0) {
4930 PERROR("pthread_join");
4931 goto error; /* join error, exit without cleanup */
4932 }
4933 }
4934
4935 exit_kernel:
4936 ret = pthread_join(jul_reg_thread, &status);
4937 if (ret != 0) {
4938 PERROR("pthread_join JUL");
4939 goto error; /* join error, exit without cleanup */
4940 }
4941
4942 exit_jul_reg:
4943 ret = pthread_join(apps_notify_thread, &status);
4944 if (ret != 0) {
4945 PERROR("pthread_join apps notify");
4946 goto error; /* join error, exit without cleanup */
4947 }
4948
4949 exit_apps_notify:
4950 ret = pthread_join(apps_thread, &status);
4951 if (ret != 0) {
4952 PERROR("pthread_join apps");
4953 goto error; /* join error, exit without cleanup */
4954 }
4955
4956
4957 exit_apps:
4958 ret = pthread_join(reg_apps_thread, &status);
4959 if (ret != 0) {
4960 PERROR("pthread_join");
4961 goto error; /* join error, exit without cleanup */
4962 }
4963
4964 exit_reg_apps:
4965 ret = pthread_join(dispatch_thread, &status);
4966 if (ret != 0) {
4967 PERROR("pthread_join");
4968 goto error; /* join error, exit without cleanup */
4969 }
4970
4971 exit_dispatch:
4972 ret = pthread_join(client_thread, &status);
4973 if (ret != 0) {
4974 PERROR("pthread_join");
4975 goto error; /* join error, exit without cleanup */
4976 }
4977
4978 ret = join_consumer_thread(&kconsumer_data);
4979 if (ret != 0) {
4980 PERROR("join_consumer");
4981 goto error; /* join error, exit without cleanup */
4982 }
4983
4984 ret = join_consumer_thread(&ustconsumer32_data);
4985 if (ret != 0) {
4986 PERROR("join_consumer ust32");
4987 goto error; /* join error, exit without cleanup */
4988 }
4989
4990 ret = join_consumer_thread(&ustconsumer64_data);
4991 if (ret != 0) {
4992 PERROR("join_consumer ust64");
4993 goto error; /* join error, exit without cleanup */
4994 }
4995
4996 exit_client:
4997 ret = pthread_join(health_thread, &status);
4998 if (ret != 0) {
4999 PERROR("pthread_join health thread");
5000 goto error; /* join error, exit without cleanup */
5001 }
5002
5003 exit_health:
5004 ret = pthread_join(ht_cleanup_thread, &status);
5005 if (ret != 0) {
5006 PERROR("pthread_join ht cleanup thread");
5007 goto error; /* join error, exit without cleanup */
5008 }
5009 exit_ht_cleanup:
5010 health_app_destroy(health_sessiond);
5011 exit_health_sessiond_cleanup:
5012 exit:
5013 /*
5014 * cleanup() is called when no other thread is running.
5015 */
5016 rcu_thread_online();
5017 cleanup();
5018 rcu_thread_offline();
5019 rcu_unregister_thread();
5020 if (!ret) {
5021 exit(EXIT_SUCCESS);
5022 }
5023 error:
5024 exit(EXIT_FAILURE);
5025 }
This page took 0.122974 seconds and 3 git commands to generate.