Initialize command header to zero
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/compat/getenv.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/session-config.h>
51
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
54 #include "channel.h"
55 #include "cmd.h"
56 #include "consumer.h"
57 #include "context.h"
58 #include "event.h"
59 #include "kernel.h"
60 #include "kernel-consumer.h"
61 #include "modprobe.h"
62 #include "shm.h"
63 #include "ust-ctl.h"
64 #include "ust-consumer.h"
65 #include "utils.h"
66 #include "fd-limit.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "agent-thread.h"
71 #include "save.h"
72 #include "load-session-thread.h"
73 #include "syscall.h"
74 #include "agent.h"
75
76 #define CONSUMERD_FILE "lttng-consumerd"
77
78 const char *progname;
79 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
80 static int tracing_group_name_override;
81 static char *opt_pidfile;
82 static int opt_sig_parent;
83 static int opt_verbose_consumer;
84 static int opt_daemon, opt_background;
85 static int opt_no_kernel;
86 static char *opt_load_session_path;
87 static pid_t ppid; /* Parent PID for --sig-parent option */
88 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
89 static char *rundir;
90 static int lockfile_fd = -1;
91
92 /* Set to 1 when a SIGUSR1 signal is received. */
93 static int recv_child_signal;
94
95 /*
96 * Consumer daemon specific control data. Every value not initialized here is
97 * set to 0 by the static definition.
98 */
99 static struct consumer_data kconsumer_data = {
100 .type = LTTNG_CONSUMER_KERNEL,
101 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
103 .err_sock = -1,
104 .cmd_sock = -1,
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
109 };
110 static struct consumer_data ustconsumer64_data = {
111 .type = LTTNG_CONSUMER64_UST,
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
114 .err_sock = -1,
115 .cmd_sock = -1,
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
120 };
121 static struct consumer_data ustconsumer32_data = {
122 .type = LTTNG_CONSUMER32_UST,
123 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
124 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
125 .err_sock = -1,
126 .cmd_sock = -1,
127 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
128 .lock = PTHREAD_MUTEX_INITIALIZER,
129 .cond = PTHREAD_COND_INITIALIZER,
130 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
131 };
132
133 /* Command line options */
134 static const struct option long_options[] = {
135 { "client-sock", required_argument, 0, 'c' },
136 { "apps-sock", required_argument, 0, 'a' },
137 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
138 { "kconsumerd-err-sock", required_argument, 0, '\0' },
139 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
143 { "consumerd32-path", required_argument, 0, '\0' },
144 { "consumerd32-libdir", required_argument, 0, '\0' },
145 { "consumerd64-path", required_argument, 0, '\0' },
146 { "consumerd64-libdir", required_argument, 0, '\0' },
147 { "daemonize", no_argument, 0, 'd' },
148 { "background", no_argument, 0, 'b' },
149 { "sig-parent", no_argument, 0, 'S' },
150 { "help", no_argument, 0, 'h' },
151 { "group", required_argument, 0, 'g' },
152 { "version", no_argument, 0, 'V' },
153 { "quiet", no_argument, 0, 'q' },
154 { "verbose", no_argument, 0, 'v' },
155 { "verbose-consumer", no_argument, 0, '\0' },
156 { "no-kernel", no_argument, 0, '\0' },
157 { "pidfile", required_argument, 0, 'p' },
158 { "agent-tcp-port", required_argument, 0, '\0' },
159 { "config", required_argument, 0, 'f' },
160 { "load", required_argument, 0, 'l' },
161 { "kmod-probes", required_argument, 0, '\0' },
162 { "extra-kmod-probes", required_argument, 0, '\0' },
163 { NULL, 0, 0, 0 }
164 };
165
166 /* Command line options to ignore from configuration file */
167 static const char *config_ignore_options[] = { "help", "version", "config" };
168
169 /* Shared between threads */
170 static int dispatch_thread_exit;
171
172 /* Global application Unix socket path */
173 static char apps_unix_sock_path[PATH_MAX];
174 /* Global client Unix socket path */
175 static char client_unix_sock_path[PATH_MAX];
176 /* global wait shm path for UST */
177 static char wait_shm_path[PATH_MAX];
178 /* Global health check unix path */
179 static char health_unix_sock_path[PATH_MAX];
180
181 /* Sockets and FDs */
182 static int client_sock = -1;
183 static int apps_sock = -1;
184 int kernel_tracer_fd = -1;
185 static int kernel_poll_pipe[2] = { -1, -1 };
186
187 /*
188 * Quit pipe for all threads. This permits a single cancellation point
189 * for all threads when receiving an event on the pipe.
190 */
191 static int thread_quit_pipe[2] = { -1, -1 };
192 static int ht_cleanup_quit_pipe[2] = { -1, -1 };
193
194 /*
195 * This pipe is used to inform the thread managing application communication
196 * that a command is queued and ready to be processed.
197 */
198 static int apps_cmd_pipe[2] = { -1, -1 };
199
200 int apps_cmd_notify_pipe[2] = { -1, -1 };
201
202 /* Pthread, Mutexes and Semaphores */
203 static pthread_t apps_thread;
204 static pthread_t apps_notify_thread;
205 static pthread_t reg_apps_thread;
206 static pthread_t client_thread;
207 static pthread_t kernel_thread;
208 static pthread_t dispatch_thread;
209 static pthread_t health_thread;
210 static pthread_t ht_cleanup_thread;
211 static pthread_t agent_reg_thread;
212 static pthread_t load_session_thread;
213
214 /*
215 * UST registration command queue. This queue is tied with a futex and uses a N
216 * wakers / 1 waiter implemented and detailed in futex.c/.h
217 *
218 * The thread_registration_apps and thread_dispatch_ust_registration uses this
219 * queue along with the wait/wake scheme. The thread_manage_apps receives down
220 * the line new application socket and monitors it for any I/O error or clean
221 * close that triggers an unregistration of the application.
222 */
223 static struct ust_cmd_queue ust_cmd_queue;
224
225 /*
226 * Pointer initialized before thread creation.
227 *
228 * This points to the tracing session list containing the session count and a
229 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
230 * MUST NOT be taken if you call a public function in session.c.
231 *
232 * The lock is nested inside the structure: session_list_ptr->lock. Please use
233 * session_lock_list and session_unlock_list for lock acquisition.
234 */
235 static struct ltt_session_list *session_list_ptr;
236
237 int ust_consumerd64_fd = -1;
238 int ust_consumerd32_fd = -1;
239
240 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
241 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
242 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
243 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
244 static int consumerd32_bin_override;
245 static int consumerd64_bin_override;
246 static int consumerd32_libdir_override;
247 static int consumerd64_libdir_override;
248
249 static const char *module_proc_lttng = "/proc/lttng";
250
251 /*
252 * Consumer daemon state which is changed when spawning it, killing it or in
253 * case of a fatal error.
254 */
255 enum consumerd_state {
256 CONSUMER_STARTED = 1,
257 CONSUMER_STOPPED = 2,
258 CONSUMER_ERROR = 3,
259 };
260
261 /*
262 * This consumer daemon state is used to validate if a client command will be
263 * able to reach the consumer. If not, the client is informed. For instance,
264 * doing a "lttng start" when the consumer state is set to ERROR will return an
265 * error to the client.
266 *
267 * The following example shows a possible race condition of this scheme:
268 *
269 * consumer thread error happens
270 * client cmd arrives
271 * client cmd checks state -> still OK
272 * consumer thread exit, sets error
273 * client cmd try to talk to consumer
274 * ...
275 *
276 * However, since the consumer is a different daemon, we have no way of making
277 * sure the command will reach it safely even with this state flag. This is why
278 * we consider that up to the state validation during command processing, the
279 * command is safe. After that, we can not guarantee the correctness of the
280 * client request vis-a-vis the consumer.
281 */
282 static enum consumerd_state ust_consumerd_state;
283 static enum consumerd_state kernel_consumerd_state;
284
285 /*
286 * Socket timeout for receiving and sending in seconds.
287 */
288 static int app_socket_timeout;
289
290 /* Set in main() with the current page size. */
291 long page_size;
292
293 /* Application health monitoring */
294 struct health_app *health_sessiond;
295
296 /* Agent TCP port for registration. Used by the agent thread. */
297 unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
298
299 /* Am I root or not. */
300 int is_root; /* Set to 1 if the daemon is running as root */
301
302 const char * const config_section_name = "sessiond";
303
304 /* Load session thread information to operate. */
305 struct load_session_thread_data *load_info;
306
307 /* Global hash tables */
308 struct lttng_ht *agent_apps_ht_by_sock = NULL;
309
310 /*
311 * Whether sessiond is ready for commands/health check requests.
312 * NR_LTTNG_SESSIOND_READY must match the number of calls to
313 * sessiond_notify_ready().
314 */
315 #define NR_LTTNG_SESSIOND_READY 3
316 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
317
318 /* Notify parents that we are ready for cmd and health check */
319 LTTNG_HIDDEN
320 void sessiond_notify_ready(void)
321 {
322 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
323 /*
324 * Notify parent pid that we are ready to accept command
325 * for client side. This ppid is the one from the
326 * external process that spawned us.
327 */
328 if (opt_sig_parent) {
329 kill(ppid, SIGUSR1);
330 }
331
332 /*
333 * Notify the parent of the fork() process that we are
334 * ready.
335 */
336 if (opt_daemon || opt_background) {
337 kill(child_ppid, SIGUSR1);
338 }
339 }
340 }
341
342 static
343 void setup_consumerd_path(void)
344 {
345 const char *bin, *libdir;
346
347 /*
348 * Allow INSTALL_BIN_PATH to be used as a target path for the
349 * native architecture size consumer if CONFIG_CONSUMER*_PATH
350 * has not been defined.
351 */
352 #if (CAA_BITS_PER_LONG == 32)
353 if (!consumerd32_bin[0]) {
354 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
355 }
356 if (!consumerd32_libdir[0]) {
357 consumerd32_libdir = INSTALL_LIB_PATH;
358 }
359 #elif (CAA_BITS_PER_LONG == 64)
360 if (!consumerd64_bin[0]) {
361 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
362 }
363 if (!consumerd64_libdir[0]) {
364 consumerd64_libdir = INSTALL_LIB_PATH;
365 }
366 #else
367 #error "Unknown bitness"
368 #endif
369
370 /*
371 * runtime env. var. overrides the build default.
372 */
373 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
374 if (bin) {
375 consumerd32_bin = bin;
376 }
377 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
378 if (bin) {
379 consumerd64_bin = bin;
380 }
381 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
382 if (libdir) {
383 consumerd32_libdir = libdir;
384 }
385 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
386 if (libdir) {
387 consumerd64_libdir = libdir;
388 }
389 }
390
391 static
392 int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
393 int *a_pipe)
394 {
395 int ret;
396
397 assert(events);
398
399 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
400 if (ret < 0) {
401 goto error;
402 }
403
404 /* Add quit pipe */
405 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
406 if (ret < 0) {
407 goto error;
408 }
409
410 return 0;
411
412 error:
413 return ret;
414 }
415
416 /*
417 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
418 */
419 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
420 {
421 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
422 }
423
424 /*
425 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
426 */
427 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
428 size_t size)
429 {
430 return __sessiond_set_thread_pollset(events, size,
431 ht_cleanup_quit_pipe);
432 }
433
434 static
435 int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
436 {
437 if (fd == a_pipe && (events & LPOLLIN)) {
438 return 1;
439 }
440 return 0;
441 }
442
443 /*
444 * Check if the thread quit pipe was triggered.
445 *
446 * Return 1 if it was triggered else 0;
447 */
448 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
449 {
450 return __sessiond_check_thread_quit_pipe(fd, events,
451 thread_quit_pipe[0]);
452 }
453
454 /*
455 * Check if the ht_cleanup thread quit pipe was triggered.
456 *
457 * Return 1 if it was triggered else 0;
458 */
459 int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
460 {
461 return __sessiond_check_thread_quit_pipe(fd, events,
462 ht_cleanup_quit_pipe[0]);
463 }
464
465 /*
466 * Init thread quit pipe.
467 *
468 * Return -1 on error or 0 if all pipes are created.
469 */
470 static int __init_thread_quit_pipe(int *a_pipe)
471 {
472 int ret, i;
473
474 ret = pipe(a_pipe);
475 if (ret < 0) {
476 PERROR("thread quit pipe");
477 goto error;
478 }
479
480 for (i = 0; i < 2; i++) {
481 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
482 if (ret < 0) {
483 PERROR("fcntl");
484 goto error;
485 }
486 }
487
488 error:
489 return ret;
490 }
491
492 static int init_thread_quit_pipe(void)
493 {
494 return __init_thread_quit_pipe(thread_quit_pipe);
495 }
496
497 static int init_ht_cleanup_quit_pipe(void)
498 {
499 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
500 }
501
502 /*
503 * Stop all threads by closing the thread quit pipe.
504 */
505 static void stop_threads(void)
506 {
507 int ret;
508
509 /* Stopping all threads */
510 DBG("Terminating all threads");
511 ret = notify_thread_pipe(thread_quit_pipe[1]);
512 if (ret < 0) {
513 ERR("write error on thread quit pipe");
514 }
515
516 /* Dispatch thread */
517 CMM_STORE_SHARED(dispatch_thread_exit, 1);
518 futex_nto1_wake(&ust_cmd_queue.futex);
519 }
520
521 /*
522 * Close every consumer sockets.
523 */
524 static void close_consumer_sockets(void)
525 {
526 int ret;
527
528 if (kconsumer_data.err_sock >= 0) {
529 ret = close(kconsumer_data.err_sock);
530 if (ret < 0) {
531 PERROR("kernel consumer err_sock close");
532 }
533 }
534 if (ustconsumer32_data.err_sock >= 0) {
535 ret = close(ustconsumer32_data.err_sock);
536 if (ret < 0) {
537 PERROR("UST consumerd32 err_sock close");
538 }
539 }
540 if (ustconsumer64_data.err_sock >= 0) {
541 ret = close(ustconsumer64_data.err_sock);
542 if (ret < 0) {
543 PERROR("UST consumerd64 err_sock close");
544 }
545 }
546 if (kconsumer_data.cmd_sock >= 0) {
547 ret = close(kconsumer_data.cmd_sock);
548 if (ret < 0) {
549 PERROR("kernel consumer cmd_sock close");
550 }
551 }
552 if (ustconsumer32_data.cmd_sock >= 0) {
553 ret = close(ustconsumer32_data.cmd_sock);
554 if (ret < 0) {
555 PERROR("UST consumerd32 cmd_sock close");
556 }
557 }
558 if (ustconsumer64_data.cmd_sock >= 0) {
559 ret = close(ustconsumer64_data.cmd_sock);
560 if (ret < 0) {
561 PERROR("UST consumerd64 cmd_sock close");
562 }
563 }
564 }
565
566 /*
567 * Generate the full lock file path using the rundir.
568 *
569 * Return the snprintf() return value thus a negative value is an error.
570 */
571 static int generate_lock_file_path(char *path, size_t len)
572 {
573 int ret;
574
575 assert(path);
576 assert(rundir);
577
578 /* Build lockfile path from rundir. */
579 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
580 if (ret < 0) {
581 PERROR("snprintf lockfile path");
582 }
583
584 return ret;
585 }
586
587 /*
588 * Wait on consumer process termination.
589 *
590 * Need to be called with the consumer data lock held or from a context
591 * ensuring no concurrent access to data (e.g: cleanup).
592 */
593 static void wait_consumer(struct consumer_data *consumer_data)
594 {
595 pid_t ret;
596 int status;
597
598 if (consumer_data->pid <= 0) {
599 return;
600 }
601
602 DBG("Waiting for complete teardown of consumerd (PID: %d)",
603 consumer_data->pid);
604 ret = waitpid(consumer_data->pid, &status, 0);
605 if (ret == -1) {
606 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
607 }
608 if (!WIFEXITED(status)) {
609 ERR("consumerd termination with error: %d",
610 WEXITSTATUS(ret));
611 }
612 consumer_data->pid = 0;
613 }
614
615 /*
616 * Cleanup the session daemon's data structures.
617 */
618 static void sessiond_cleanup(void)
619 {
620 int ret;
621 struct ltt_session *sess, *stmp;
622 char path[PATH_MAX];
623
624 DBG("Cleanup sessiond");
625
626 /*
627 * Close the thread quit pipe. It has already done its job,
628 * since we are now called.
629 */
630 utils_close_pipe(thread_quit_pipe);
631
632 /*
633 * If opt_pidfile is undefined, the default file will be wiped when
634 * removing the rundir.
635 */
636 if (opt_pidfile) {
637 ret = remove(opt_pidfile);
638 if (ret < 0) {
639 PERROR("remove pidfile %s", opt_pidfile);
640 }
641 }
642
643 DBG("Removing sessiond and consumerd content of directory %s", rundir);
644
645 /* sessiond */
646 snprintf(path, PATH_MAX,
647 "%s/%s",
648 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
649 DBG("Removing %s", path);
650 (void) unlink(path);
651
652 snprintf(path, PATH_MAX, "%s/%s", rundir,
653 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
654 DBG("Removing %s", path);
655 (void) unlink(path);
656
657 /* kconsumerd */
658 snprintf(path, PATH_MAX,
659 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
660 rundir);
661 DBG("Removing %s", path);
662 (void) unlink(path);
663
664 snprintf(path, PATH_MAX,
665 DEFAULT_KCONSUMERD_PATH,
666 rundir);
667 DBG("Removing directory %s", path);
668 (void) rmdir(path);
669
670 /* ust consumerd 32 */
671 snprintf(path, PATH_MAX,
672 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
673 rundir);
674 DBG("Removing %s", path);
675 (void) unlink(path);
676
677 snprintf(path, PATH_MAX,
678 DEFAULT_USTCONSUMERD32_PATH,
679 rundir);
680 DBG("Removing directory %s", path);
681 (void) rmdir(path);
682
683 /* ust consumerd 64 */
684 snprintf(path, PATH_MAX,
685 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
686 rundir);
687 DBG("Removing %s", path);
688 (void) unlink(path);
689
690 snprintf(path, PATH_MAX,
691 DEFAULT_USTCONSUMERD64_PATH,
692 rundir);
693 DBG("Removing directory %s", path);
694 (void) rmdir(path);
695
696 DBG("Cleaning up all sessions");
697
698 /* Destroy session list mutex */
699 if (session_list_ptr != NULL) {
700 pthread_mutex_destroy(&session_list_ptr->lock);
701
702 /* Cleanup ALL session */
703 cds_list_for_each_entry_safe(sess, stmp,
704 &session_list_ptr->head, list) {
705 cmd_destroy_session(sess, kernel_poll_pipe[1]);
706 }
707 }
708
709 wait_consumer(&kconsumer_data);
710 wait_consumer(&ustconsumer64_data);
711 wait_consumer(&ustconsumer32_data);
712
713 DBG("Cleaning up all agent apps");
714 agent_app_ht_clean();
715
716 DBG("Closing all UST sockets");
717 ust_app_clean_list();
718 buffer_reg_destroy_registries();
719
720 if (is_root && !opt_no_kernel) {
721 DBG2("Closing kernel fd");
722 if (kernel_tracer_fd >= 0) {
723 ret = close(kernel_tracer_fd);
724 if (ret) {
725 PERROR("close");
726 }
727 }
728 DBG("Unloading kernel modules");
729 modprobe_remove_lttng_all();
730 free(syscall_table);
731 }
732
733 close_consumer_sockets();
734
735 if (load_info) {
736 load_session_destroy_data(load_info);
737 free(load_info);
738 }
739
740 /*
741 * Cleanup lock file by deleting it and finaly closing it which will
742 * release the file system lock.
743 */
744 if (lockfile_fd >= 0) {
745 char lockfile_path[PATH_MAX];
746
747 ret = generate_lock_file_path(lockfile_path,
748 sizeof(lockfile_path));
749 if (ret > 0) {
750 ret = remove(lockfile_path);
751 if (ret < 0) {
752 PERROR("remove lock file");
753 }
754 ret = close(lockfile_fd);
755 if (ret < 0) {
756 PERROR("close lock file");
757 }
758 }
759 }
760
761 /*
762 * We do NOT rmdir rundir because there are other processes
763 * using it, for instance lttng-relayd, which can start in
764 * parallel with this teardown.
765 */
766
767 free(rundir);
768 }
769
770 /*
771 * Cleanup the daemon's option data structures.
772 */
773 static void sessiond_cleanup_options(void)
774 {
775 DBG("Cleaning up options");
776
777 /*
778 * If the override option is set, the pointer points to a *non* const
779 * thus freeing it even though the variable type is set to const.
780 */
781 if (tracing_group_name_override) {
782 free((void *) tracing_group_name);
783 }
784 if (consumerd32_bin_override) {
785 free((void *) consumerd32_bin);
786 }
787 if (consumerd64_bin_override) {
788 free((void *) consumerd64_bin);
789 }
790 if (consumerd32_libdir_override) {
791 free((void *) consumerd32_libdir);
792 }
793 if (consumerd64_libdir_override) {
794 free((void *) consumerd64_libdir);
795 }
796
797 free(opt_pidfile);
798 free(opt_load_session_path);
799 free(kmod_probes_list);
800 free(kmod_extra_probes_list);
801
802 run_as_destroy_worker();
803
804 /* <fun> */
805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 /* </fun> */
809 }
810
811 /*
812 * Send data on a unix socket using the liblttsessiondcomm API.
813 *
814 * Return lttcomm error code.
815 */
816 static int send_unix_sock(int sock, void *buf, size_t len)
817 {
818 /* Check valid length */
819 if (len == 0) {
820 return -1;
821 }
822
823 return lttcomm_send_unix_sock(sock, buf, len);
824 }
825
826 /*
827 * Free memory of a command context structure.
828 */
829 static void clean_command_ctx(struct command_ctx **cmd_ctx)
830 {
831 DBG("Clean command context structure");
832 if (*cmd_ctx) {
833 if ((*cmd_ctx)->llm) {
834 free((*cmd_ctx)->llm);
835 }
836 if ((*cmd_ctx)->lsm) {
837 free((*cmd_ctx)->lsm);
838 }
839 free(*cmd_ctx);
840 *cmd_ctx = NULL;
841 }
842 }
843
844 /*
845 * Notify UST applications using the shm mmap futex.
846 */
847 static int notify_ust_apps(int active)
848 {
849 char *wait_shm_mmap;
850
851 DBG("Notifying applications of session daemon state: %d", active);
852
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
855 if (wait_shm_mmap == NULL) {
856 goto error;
857 }
858
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap, active);
861
862 /* Apps notified successfully */
863 return 0;
864
865 error:
866 return -1;
867 }
868
869 /*
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
872 * structure.
873 *
874 * Return 0 on success, negative value on error.
875 */
876 static int setup_lttng_msg(struct command_ctx *cmd_ctx,
877 const void *payload_buf, size_t payload_len,
878 const void *cmd_header_buf, size_t cmd_header_len)
879 {
880 int ret = 0;
881 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
882 const size_t cmd_header_offset = header_len;
883 const size_t payload_offset = cmd_header_offset + cmd_header_len;
884 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
885
886 cmd_ctx->llm = zmalloc(total_msg_size);
887
888 if (cmd_ctx->llm == NULL) {
889 PERROR("zmalloc");
890 ret = -ENOMEM;
891 goto end;
892 }
893
894 /* Copy common data */
895 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
896 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
897 cmd_ctx->llm->cmd_header_size = cmd_header_len;
898 cmd_ctx->llm->data_size = payload_len;
899 cmd_ctx->lttng_msg_size = total_msg_size;
900
901 /* Copy command header */
902 if (cmd_header_len) {
903 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
904 cmd_header_len);
905 }
906
907 /* Copy payload */
908 if (payload_len) {
909 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
910 payload_len);
911 }
912
913 end:
914 return ret;
915 }
916
917 /*
918 * Version of setup_lttng_msg() without command header.
919 */
920 static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
921 void *payload_buf, size_t payload_len)
922 {
923 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
924 }
925 /*
926 * Update the kernel poll set of all channel fd available over all tracing
927 * session. Add the wakeup pipe at the end of the set.
928 */
929 static int update_kernel_poll(struct lttng_poll_event *events)
930 {
931 int ret;
932 struct ltt_session *session;
933 struct ltt_kernel_channel *channel;
934
935 DBG("Updating kernel poll set");
936
937 session_lock_list();
938 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
939 session_lock(session);
940 if (session->kernel_session == NULL) {
941 session_unlock(session);
942 continue;
943 }
944
945 cds_list_for_each_entry(channel,
946 &session->kernel_session->channel_list.head, list) {
947 /* Add channel fd to the kernel poll set */
948 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
949 if (ret < 0) {
950 session_unlock(session);
951 goto error;
952 }
953 DBG("Channel fd %d added to kernel set", channel->fd);
954 }
955 session_unlock(session);
956 }
957 session_unlock_list();
958
959 return 0;
960
961 error:
962 session_unlock_list();
963 return -1;
964 }
965
966 /*
967 * Find the channel fd from 'fd' over all tracing session. When found, check
968 * for new channel stream and send those stream fds to the kernel consumer.
969 *
970 * Useful for CPU hotplug feature.
971 */
972 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
973 {
974 int ret = 0;
975 struct ltt_session *session;
976 struct ltt_kernel_session *ksess;
977 struct ltt_kernel_channel *channel;
978
979 DBG("Updating kernel streams for channel fd %d", fd);
980
981 session_lock_list();
982 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
983 session_lock(session);
984 if (session->kernel_session == NULL) {
985 session_unlock(session);
986 continue;
987 }
988 ksess = session->kernel_session;
989
990 cds_list_for_each_entry(channel,
991 &ksess->channel_list.head, list) {
992 struct lttng_ht_iter iter;
993 struct consumer_socket *socket;
994
995 if (channel->fd != fd) {
996 continue;
997 }
998 DBG("Channel found, updating kernel streams");
999 ret = kernel_open_channel_stream(channel);
1000 if (ret < 0) {
1001 goto error;
1002 }
1003 /* Update the stream global counter */
1004 ksess->stream_count_global += ret;
1005
1006 /*
1007 * Have we already sent fds to the consumer? If yes, it
1008 * means that tracing is started so it is safe to send
1009 * our updated stream fds.
1010 */
1011 if (ksess->consumer_fds_sent != 1
1012 || ksess->consumer == NULL) {
1013 ret = -1;
1014 goto error;
1015 }
1016
1017 rcu_read_lock();
1018 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
1019 &iter.iter, socket, node.node) {
1020 pthread_mutex_lock(socket->lock);
1021 ret = kernel_consumer_send_channel_stream(socket,
1022 channel, ksess,
1023 session->output_traces ? 1 : 0);
1024 pthread_mutex_unlock(socket->lock);
1025 if (ret < 0) {
1026 rcu_read_unlock();
1027 goto error;
1028 }
1029 }
1030 rcu_read_unlock();
1031 }
1032 session_unlock(session);
1033 }
1034 session_unlock_list();
1035 return ret;
1036
1037 error:
1038 session_unlock(session);
1039 session_unlock_list();
1040 return ret;
1041 }
1042
1043 /*
1044 * For each tracing session, update newly registered apps. The session list
1045 * lock MUST be acquired before calling this.
1046 */
1047 static void update_ust_app(int app_sock)
1048 {
1049 struct ltt_session *sess, *stmp;
1050
1051 /* Consumer is in an ERROR state. Stop any application update. */
1052 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1053 /* Stop the update process since the consumer is dead. */
1054 return;
1055 }
1056
1057 /* For all tracing session(s) */
1058 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
1059 struct ust_app *app;
1060
1061 session_lock(sess);
1062 if (!sess->ust_session) {
1063 goto unlock_session;
1064 }
1065
1066 rcu_read_lock();
1067 assert(app_sock >= 0);
1068 app = ust_app_find_by_sock(app_sock);
1069 if (app == NULL) {
1070 /*
1071 * Application can be unregistered before so
1072 * this is possible hence simply stopping the
1073 * update.
1074 */
1075 DBG3("UST app update failed to find app sock %d",
1076 app_sock);
1077 goto unlock_rcu;
1078 }
1079 ust_app_global_update(sess->ust_session, app);
1080 unlock_rcu:
1081 rcu_read_unlock();
1082 unlock_session:
1083 session_unlock(sess);
1084 }
1085 }
1086
1087 /*
1088 * This thread manage event coming from the kernel.
1089 *
1090 * Features supported in this thread:
1091 * -) CPU Hotplug
1092 */
1093 static void *thread_manage_kernel(void *data)
1094 {
1095 int ret, i, pollfd, update_poll_flag = 1, err = -1;
1096 uint32_t revents, nb_fd;
1097 char tmp;
1098 struct lttng_poll_event events;
1099
1100 DBG("[thread] Thread manage kernel started");
1101
1102 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
1103
1104 /*
1105 * This first step of the while is to clean this structure which could free
1106 * non NULL pointers so initialize it before the loop.
1107 */
1108 lttng_poll_init(&events);
1109
1110 if (testpoint(sessiond_thread_manage_kernel)) {
1111 goto error_testpoint;
1112 }
1113
1114 health_code_update();
1115
1116 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
1117 goto error_testpoint;
1118 }
1119
1120 while (1) {
1121 health_code_update();
1122
1123 if (update_poll_flag == 1) {
1124 /* Clean events object. We are about to populate it again. */
1125 lttng_poll_clean(&events);
1126
1127 ret = sessiond_set_thread_pollset(&events, 2);
1128 if (ret < 0) {
1129 goto error_poll_create;
1130 }
1131
1132 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1133 if (ret < 0) {
1134 goto error;
1135 }
1136
1137 /* This will add the available kernel channel if any. */
1138 ret = update_kernel_poll(&events);
1139 if (ret < 0) {
1140 goto error;
1141 }
1142 update_poll_flag = 0;
1143 }
1144
1145 DBG("Thread kernel polling");
1146
1147 /* Poll infinite value of time */
1148 restart:
1149 health_poll_entry();
1150 ret = lttng_poll_wait(&events, -1);
1151 DBG("Thread kernel return from poll on %d fds",
1152 LTTNG_POLL_GETNB(&events));
1153 health_poll_exit();
1154 if (ret < 0) {
1155 /*
1156 * Restart interrupted system call.
1157 */
1158 if (errno == EINTR) {
1159 goto restart;
1160 }
1161 goto error;
1162 } else if (ret == 0) {
1163 /* Should not happen since timeout is infinite */
1164 ERR("Return value of poll is 0 with an infinite timeout.\n"
1165 "This should not have happened! Continuing...");
1166 continue;
1167 }
1168
1169 nb_fd = ret;
1170
1171 for (i = 0; i < nb_fd; i++) {
1172 /* Fetch once the poll data */
1173 revents = LTTNG_POLL_GETEV(&events, i);
1174 pollfd = LTTNG_POLL_GETFD(&events, i);
1175
1176 health_code_update();
1177
1178 if (!revents) {
1179 /* No activity for this FD (poll implementation). */
1180 continue;
1181 }
1182
1183 /* Thread quit pipe has been closed. Killing thread. */
1184 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1185 if (ret) {
1186 err = 0;
1187 goto exit;
1188 }
1189
1190 /* Check for data on kernel pipe */
1191 if (revents & LPOLLIN) {
1192 if (pollfd == kernel_poll_pipe[0]) {
1193 (void) lttng_read(kernel_poll_pipe[0],
1194 &tmp, 1);
1195 /*
1196 * Ret value is useless here, if this pipe gets any actions an
1197 * update is required anyway.
1198 */
1199 update_poll_flag = 1;
1200 continue;
1201 } else {
1202 /*
1203 * New CPU detected by the kernel. Adding kernel stream to
1204 * kernel session and updating the kernel consumer
1205 */
1206 ret = update_kernel_stream(&kconsumer_data, pollfd);
1207 if (ret < 0) {
1208 continue;
1209 }
1210 break;
1211 }
1212 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1213 update_poll_flag = 1;
1214 continue;
1215 } else {
1216 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1217 goto error;
1218 }
1219 }
1220 }
1221
1222 exit:
1223 error:
1224 lttng_poll_clean(&events);
1225 error_poll_create:
1226 error_testpoint:
1227 utils_close_pipe(kernel_poll_pipe);
1228 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
1229 if (err) {
1230 health_error();
1231 ERR("Health error occurred in %s", __func__);
1232 WARN("Kernel thread died unexpectedly. "
1233 "Kernel tracing can continue but CPU hotplug is disabled.");
1234 }
1235 health_unregister(health_sessiond);
1236 DBG("Kernel thread dying");
1237 return NULL;
1238 }
1239
1240 /*
1241 * Signal pthread condition of the consumer data that the thread.
1242 */
1243 static void signal_consumer_condition(struct consumer_data *data, int state)
1244 {
1245 pthread_mutex_lock(&data->cond_mutex);
1246
1247 /*
1248 * The state is set before signaling. It can be any value, it's the waiter
1249 * job to correctly interpret this condition variable associated to the
1250 * consumer pthread_cond.
1251 *
1252 * A value of 0 means that the corresponding thread of the consumer data
1253 * was not started. 1 indicates that the thread has started and is ready
1254 * for action. A negative value means that there was an error during the
1255 * thread bootstrap.
1256 */
1257 data->consumer_thread_is_ready = state;
1258 (void) pthread_cond_signal(&data->cond);
1259
1260 pthread_mutex_unlock(&data->cond_mutex);
1261 }
1262
1263 /*
1264 * This thread manage the consumer error sent back to the session daemon.
1265 */
1266 static void *thread_manage_consumer(void *data)
1267 {
1268 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
1269 uint32_t revents, nb_fd;
1270 enum lttcomm_return_code code;
1271 struct lttng_poll_event events;
1272 struct consumer_data *consumer_data = data;
1273
1274 DBG("[thread] Manage consumer started");
1275
1276 rcu_register_thread();
1277 rcu_thread_online();
1278
1279 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1280
1281 health_code_update();
1282
1283 /*
1284 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1285 * metadata_sock. Nothing more will be added to this poll set.
1286 */
1287 ret = sessiond_set_thread_pollset(&events, 3);
1288 if (ret < 0) {
1289 goto error_poll;
1290 }
1291
1292 /*
1293 * The error socket here is already in a listening state which was done
1294 * just before spawning this thread to avoid a race between the consumer
1295 * daemon exec trying to connect and the listen() call.
1296 */
1297 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1298 if (ret < 0) {
1299 goto error;
1300 }
1301
1302 health_code_update();
1303
1304 /* Infinite blocking call, waiting for transmission */
1305 restart:
1306 health_poll_entry();
1307
1308 if (testpoint(sessiond_thread_manage_consumer)) {
1309 goto error;
1310 }
1311
1312 ret = lttng_poll_wait(&events, -1);
1313 health_poll_exit();
1314 if (ret < 0) {
1315 /*
1316 * Restart interrupted system call.
1317 */
1318 if (errno == EINTR) {
1319 goto restart;
1320 }
1321 goto error;
1322 }
1323
1324 nb_fd = ret;
1325
1326 for (i = 0; i < nb_fd; i++) {
1327 /* Fetch once the poll data */
1328 revents = LTTNG_POLL_GETEV(&events, i);
1329 pollfd = LTTNG_POLL_GETFD(&events, i);
1330
1331 health_code_update();
1332
1333 if (!revents) {
1334 /* No activity for this FD (poll implementation). */
1335 continue;
1336 }
1337
1338 /* Thread quit pipe has been closed. Killing thread. */
1339 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1340 if (ret) {
1341 err = 0;
1342 goto exit;
1343 }
1344
1345 /* Event on the registration socket */
1346 if (pollfd == consumer_data->err_sock) {
1347 if (revents & LPOLLIN) {
1348 continue;
1349 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1350 ERR("consumer err socket poll error");
1351 goto error;
1352 } else {
1353 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1354 goto error;
1355 }
1356 }
1357 }
1358
1359 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1360 if (sock < 0) {
1361 goto error;
1362 }
1363
1364 /*
1365 * Set the CLOEXEC flag. Return code is useless because either way, the
1366 * show must go on.
1367 */
1368 (void) utils_set_fd_cloexec(sock);
1369
1370 health_code_update();
1371
1372 DBG2("Receiving code from consumer err_sock");
1373
1374 /* Getting status code from kconsumerd */
1375 ret = lttcomm_recv_unix_sock(sock, &code,
1376 sizeof(enum lttcomm_return_code));
1377 if (ret <= 0) {
1378 goto error;
1379 }
1380
1381 health_code_update();
1382 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1383 /* Connect both socket, command and metadata. */
1384 consumer_data->cmd_sock =
1385 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1386 consumer_data->metadata_fd =
1387 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1388 if (consumer_data->cmd_sock < 0
1389 || consumer_data->metadata_fd < 0) {
1390 PERROR("consumer connect cmd socket");
1391 /* On error, signal condition and quit. */
1392 signal_consumer_condition(consumer_data, -1);
1393 goto error;
1394 }
1395 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1396 /* Create metadata socket lock. */
1397 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1398 if (consumer_data->metadata_sock.lock == NULL) {
1399 PERROR("zmalloc pthread mutex");
1400 ret = -1;
1401 goto error;
1402 }
1403 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1404
1405 signal_consumer_condition(consumer_data, 1);
1406 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1407 DBG("Consumer metadata socket ready (fd: %d)",
1408 consumer_data->metadata_fd);
1409 } else {
1410 ERR("consumer error when waiting for SOCK_READY : %s",
1411 lttcomm_get_readable_code(-code));
1412 goto error;
1413 }
1414
1415 /* Remove the consumerd error sock since we've established a connexion */
1416 ret = lttng_poll_del(&events, consumer_data->err_sock);
1417 if (ret < 0) {
1418 goto error;
1419 }
1420
1421 /* Add new accepted error socket. */
1422 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1423 if (ret < 0) {
1424 goto error;
1425 }
1426
1427 /* Add metadata socket that is successfully connected. */
1428 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1429 LPOLLIN | LPOLLRDHUP);
1430 if (ret < 0) {
1431 goto error;
1432 }
1433
1434 health_code_update();
1435
1436 /* Infinite blocking call, waiting for transmission */
1437 restart_poll:
1438 while (1) {
1439 health_code_update();
1440
1441 /* Exit the thread because the thread quit pipe has been triggered. */
1442 if (should_quit) {
1443 /* Not a health error. */
1444 err = 0;
1445 goto exit;
1446 }
1447
1448 health_poll_entry();
1449 ret = lttng_poll_wait(&events, -1);
1450 health_poll_exit();
1451 if (ret < 0) {
1452 /*
1453 * Restart interrupted system call.
1454 */
1455 if (errno == EINTR) {
1456 goto restart_poll;
1457 }
1458 goto error;
1459 }
1460
1461 nb_fd = ret;
1462
1463 for (i = 0; i < nb_fd; i++) {
1464 /* Fetch once the poll data */
1465 revents = LTTNG_POLL_GETEV(&events, i);
1466 pollfd = LTTNG_POLL_GETFD(&events, i);
1467
1468 health_code_update();
1469
1470 if (!revents) {
1471 /* No activity for this FD (poll implementation). */
1472 continue;
1473 }
1474
1475 /*
1476 * Thread quit pipe has been triggered, flag that we should stop
1477 * but continue the current loop to handle potential data from
1478 * consumer.
1479 */
1480 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
1481
1482 if (pollfd == sock) {
1483 /* Event on the consumerd socket */
1484 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1485 && !(revents & LPOLLIN)) {
1486 ERR("consumer err socket second poll error");
1487 goto error;
1488 }
1489 health_code_update();
1490 /* Wait for any kconsumerd error */
1491 ret = lttcomm_recv_unix_sock(sock, &code,
1492 sizeof(enum lttcomm_return_code));
1493 if (ret <= 0) {
1494 ERR("consumer closed the command socket");
1495 goto error;
1496 }
1497
1498 ERR("consumer return code : %s",
1499 lttcomm_get_readable_code(-code));
1500
1501 goto exit;
1502 } else if (pollfd == consumer_data->metadata_fd) {
1503 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1504 && !(revents & LPOLLIN)) {
1505 ERR("consumer err metadata socket second poll error");
1506 goto error;
1507 }
1508 /* UST metadata requests */
1509 ret = ust_consumer_metadata_request(
1510 &consumer_data->metadata_sock);
1511 if (ret < 0) {
1512 ERR("Handling metadata request");
1513 goto error;
1514 }
1515 }
1516 /* No need for an else branch all FDs are tested prior. */
1517 }
1518 health_code_update();
1519 }
1520
1521 exit:
1522 error:
1523 /*
1524 * We lock here because we are about to close the sockets and some other
1525 * thread might be using them so get exclusive access which will abort all
1526 * other consumer command by other threads.
1527 */
1528 pthread_mutex_lock(&consumer_data->lock);
1529
1530 /* Immediately set the consumerd state to stopped */
1531 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1532 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1533 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1534 consumer_data->type == LTTNG_CONSUMER32_UST) {
1535 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1536 } else {
1537 /* Code flow error... */
1538 assert(0);
1539 }
1540
1541 if (consumer_data->err_sock >= 0) {
1542 ret = close(consumer_data->err_sock);
1543 if (ret) {
1544 PERROR("close");
1545 }
1546 consumer_data->err_sock = -1;
1547 }
1548 if (consumer_data->cmd_sock >= 0) {
1549 ret = close(consumer_data->cmd_sock);
1550 if (ret) {
1551 PERROR("close");
1552 }
1553 consumer_data->cmd_sock = -1;
1554 }
1555 if (consumer_data->metadata_sock.fd_ptr &&
1556 *consumer_data->metadata_sock.fd_ptr >= 0) {
1557 ret = close(*consumer_data->metadata_sock.fd_ptr);
1558 if (ret) {
1559 PERROR("close");
1560 }
1561 }
1562 if (sock >= 0) {
1563 ret = close(sock);
1564 if (ret) {
1565 PERROR("close");
1566 }
1567 }
1568
1569 unlink(consumer_data->err_unix_sock_path);
1570 unlink(consumer_data->cmd_unix_sock_path);
1571 pthread_mutex_unlock(&consumer_data->lock);
1572
1573 /* Cleanup metadata socket mutex. */
1574 if (consumer_data->metadata_sock.lock) {
1575 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1576 free(consumer_data->metadata_sock.lock);
1577 }
1578 lttng_poll_clean(&events);
1579 error_poll:
1580 if (err) {
1581 health_error();
1582 ERR("Health error occurred in %s", __func__);
1583 }
1584 health_unregister(health_sessiond);
1585 DBG("consumer thread cleanup completed");
1586
1587 rcu_thread_offline();
1588 rcu_unregister_thread();
1589
1590 return NULL;
1591 }
1592
1593 /*
1594 * This thread manage application communication.
1595 */
1596 static void *thread_manage_apps(void *data)
1597 {
1598 int i, ret, pollfd, err = -1;
1599 ssize_t size_ret;
1600 uint32_t revents, nb_fd;
1601 struct lttng_poll_event events;
1602
1603 DBG("[thread] Manage application started");
1604
1605 rcu_register_thread();
1606 rcu_thread_online();
1607
1608 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1609
1610 if (testpoint(sessiond_thread_manage_apps)) {
1611 goto error_testpoint;
1612 }
1613
1614 health_code_update();
1615
1616 ret = sessiond_set_thread_pollset(&events, 2);
1617 if (ret < 0) {
1618 goto error_poll_create;
1619 }
1620
1621 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1622 if (ret < 0) {
1623 goto error;
1624 }
1625
1626 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
1627 goto error;
1628 }
1629
1630 health_code_update();
1631
1632 while (1) {
1633 DBG("Apps thread polling");
1634
1635 /* Inifinite blocking call, waiting for transmission */
1636 restart:
1637 health_poll_entry();
1638 ret = lttng_poll_wait(&events, -1);
1639 DBG("Apps thread return from poll on %d fds",
1640 LTTNG_POLL_GETNB(&events));
1641 health_poll_exit();
1642 if (ret < 0) {
1643 /*
1644 * Restart interrupted system call.
1645 */
1646 if (errno == EINTR) {
1647 goto restart;
1648 }
1649 goto error;
1650 }
1651
1652 nb_fd = ret;
1653
1654 for (i = 0; i < nb_fd; i++) {
1655 /* Fetch once the poll data */
1656 revents = LTTNG_POLL_GETEV(&events, i);
1657 pollfd = LTTNG_POLL_GETFD(&events, i);
1658
1659 health_code_update();
1660
1661 if (!revents) {
1662 /* No activity for this FD (poll implementation). */
1663 continue;
1664 }
1665
1666 /* Thread quit pipe has been closed. Killing thread. */
1667 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1668 if (ret) {
1669 err = 0;
1670 goto exit;
1671 }
1672
1673 /* Inspect the apps cmd pipe */
1674 if (pollfd == apps_cmd_pipe[0]) {
1675 if (revents & LPOLLIN) {
1676 int sock;
1677
1678 /* Empty pipe */
1679 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1680 if (size_ret < sizeof(sock)) {
1681 PERROR("read apps cmd pipe");
1682 goto error;
1683 }
1684
1685 health_code_update();
1686
1687 /*
1688 * Since this is a command socket (write then read),
1689 * we only monitor the error events of the socket.
1690 */
1691 ret = lttng_poll_add(&events, sock,
1692 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1693 if (ret < 0) {
1694 goto error;
1695 }
1696
1697 DBG("Apps with sock %d added to poll set", sock);
1698 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1699 ERR("Apps command pipe error");
1700 goto error;
1701 } else {
1702 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1703 goto error;
1704 }
1705 } else {
1706 /*
1707 * At this point, we know that a registered application made
1708 * the event at poll_wait.
1709 */
1710 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1711 /* Removing from the poll set */
1712 ret = lttng_poll_del(&events, pollfd);
1713 if (ret < 0) {
1714 goto error;
1715 }
1716
1717 /* Socket closed on remote end. */
1718 ust_app_unregister(pollfd);
1719 } else {
1720 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1721 goto error;
1722 }
1723 }
1724
1725 health_code_update();
1726 }
1727 }
1728
1729 exit:
1730 error:
1731 lttng_poll_clean(&events);
1732 error_poll_create:
1733 error_testpoint:
1734 utils_close_pipe(apps_cmd_pipe);
1735 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1736
1737 /*
1738 * We don't clean the UST app hash table here since already registered
1739 * applications can still be controlled so let them be until the session
1740 * daemon dies or the applications stop.
1741 */
1742
1743 if (err) {
1744 health_error();
1745 ERR("Health error occurred in %s", __func__);
1746 }
1747 health_unregister(health_sessiond);
1748 DBG("Application communication apps thread cleanup complete");
1749 rcu_thread_offline();
1750 rcu_unregister_thread();
1751 return NULL;
1752 }
1753
1754 /*
1755 * Send a socket to a thread This is called from the dispatch UST registration
1756 * thread once all sockets are set for the application.
1757 *
1758 * The sock value can be invalid, we don't really care, the thread will handle
1759 * it and make the necessary cleanup if so.
1760 *
1761 * On success, return 0 else a negative value being the errno message of the
1762 * write().
1763 */
1764 static int send_socket_to_thread(int fd, int sock)
1765 {
1766 ssize_t ret;
1767
1768 /*
1769 * It's possible that the FD is set as invalid with -1 concurrently just
1770 * before calling this function being a shutdown state of the thread.
1771 */
1772 if (fd < 0) {
1773 ret = -EBADF;
1774 goto error;
1775 }
1776
1777 ret = lttng_write(fd, &sock, sizeof(sock));
1778 if (ret < sizeof(sock)) {
1779 PERROR("write apps pipe %d", fd);
1780 if (ret < 0) {
1781 ret = -errno;
1782 }
1783 goto error;
1784 }
1785
1786 /* All good. Don't send back the write positive ret value. */
1787 ret = 0;
1788 error:
1789 return (int) ret;
1790 }
1791
1792 /*
1793 * Sanitize the wait queue of the dispatch registration thread meaning removing
1794 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1795 * notify socket is never received.
1796 */
1797 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1798 {
1799 int ret, nb_fd = 0, i;
1800 unsigned int fd_added = 0;
1801 struct lttng_poll_event events;
1802 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1803
1804 assert(wait_queue);
1805
1806 lttng_poll_init(&events);
1807
1808 /* Just skip everything for an empty queue. */
1809 if (!wait_queue->count) {
1810 goto end;
1811 }
1812
1813 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1814 if (ret < 0) {
1815 goto error_create;
1816 }
1817
1818 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1819 &wait_queue->head, head) {
1820 assert(wait_node->app);
1821 ret = lttng_poll_add(&events, wait_node->app->sock,
1822 LPOLLHUP | LPOLLERR);
1823 if (ret < 0) {
1824 goto error;
1825 }
1826
1827 fd_added = 1;
1828 }
1829
1830 if (!fd_added) {
1831 goto end;
1832 }
1833
1834 /*
1835 * Poll but don't block so we can quickly identify the faulty events and
1836 * clean them afterwards from the wait queue.
1837 */
1838 ret = lttng_poll_wait(&events, 0);
1839 if (ret < 0) {
1840 goto error;
1841 }
1842 nb_fd = ret;
1843
1844 for (i = 0; i < nb_fd; i++) {
1845 /* Get faulty FD. */
1846 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1847 int pollfd = LTTNG_POLL_GETFD(&events, i);
1848
1849 if (!revents) {
1850 /* No activity for this FD (poll implementation). */
1851 continue;
1852 }
1853
1854 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1855 &wait_queue->head, head) {
1856 if (pollfd == wait_node->app->sock &&
1857 (revents & (LPOLLHUP | LPOLLERR))) {
1858 cds_list_del(&wait_node->head);
1859 wait_queue->count--;
1860 ust_app_destroy(wait_node->app);
1861 free(wait_node);
1862 /*
1863 * Silence warning of use-after-free in
1864 * cds_list_for_each_entry_safe which uses
1865 * __typeof__(*wait_node).
1866 */
1867 wait_node = NULL;
1868 break;
1869 } else {
1870 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1871 goto error;
1872 }
1873 }
1874 }
1875
1876 if (nb_fd > 0) {
1877 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1878 }
1879
1880 end:
1881 lttng_poll_clean(&events);
1882 return;
1883
1884 error:
1885 lttng_poll_clean(&events);
1886 error_create:
1887 ERR("Unable to sanitize wait queue");
1888 return;
1889 }
1890
1891 /*
1892 * Dispatch request from the registration threads to the application
1893 * communication thread.
1894 */
1895 static void *thread_dispatch_ust_registration(void *data)
1896 {
1897 int ret, err = -1;
1898 struct cds_wfcq_node *node;
1899 struct ust_command *ust_cmd = NULL;
1900 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1901 struct ust_reg_wait_queue wait_queue = {
1902 .count = 0,
1903 };
1904
1905 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1906
1907 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1908 goto error_testpoint;
1909 }
1910
1911 health_code_update();
1912
1913 CDS_INIT_LIST_HEAD(&wait_queue.head);
1914
1915 DBG("[thread] Dispatch UST command started");
1916
1917 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1918 health_code_update();
1919
1920 /* Atomically prepare the queue futex */
1921 futex_nto1_prepare(&ust_cmd_queue.futex);
1922
1923 do {
1924 struct ust_app *app = NULL;
1925 ust_cmd = NULL;
1926
1927 /*
1928 * Make sure we don't have node(s) that have hung up before receiving
1929 * the notify socket. This is to clean the list in order to avoid
1930 * memory leaks from notify socket that are never seen.
1931 */
1932 sanitize_wait_queue(&wait_queue);
1933
1934 health_code_update();
1935 /* Dequeue command for registration */
1936 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
1937 if (node == NULL) {
1938 DBG("Woken up but nothing in the UST command queue");
1939 /* Continue thread execution */
1940 break;
1941 }
1942
1943 ust_cmd = caa_container_of(node, struct ust_command, node);
1944
1945 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1946 " gid:%d sock:%d name:%s (version %d.%d)",
1947 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1948 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1949 ust_cmd->sock, ust_cmd->reg_msg.name,
1950 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1951
1952 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1953 wait_node = zmalloc(sizeof(*wait_node));
1954 if (!wait_node) {
1955 PERROR("zmalloc wait_node dispatch");
1956 ret = close(ust_cmd->sock);
1957 if (ret < 0) {
1958 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1959 }
1960 lttng_fd_put(LTTNG_FD_APPS, 1);
1961 free(ust_cmd);
1962 goto error;
1963 }
1964 CDS_INIT_LIST_HEAD(&wait_node->head);
1965
1966 /* Create application object if socket is CMD. */
1967 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1968 ust_cmd->sock);
1969 if (!wait_node->app) {
1970 ret = close(ust_cmd->sock);
1971 if (ret < 0) {
1972 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1973 }
1974 lttng_fd_put(LTTNG_FD_APPS, 1);
1975 free(wait_node);
1976 free(ust_cmd);
1977 continue;
1978 }
1979 /*
1980 * Add application to the wait queue so we can set the notify
1981 * socket before putting this object in the global ht.
1982 */
1983 cds_list_add(&wait_node->head, &wait_queue.head);
1984 wait_queue.count++;
1985
1986 free(ust_cmd);
1987 /*
1988 * We have to continue here since we don't have the notify
1989 * socket and the application MUST be added to the hash table
1990 * only at that moment.
1991 */
1992 continue;
1993 } else {
1994 /*
1995 * Look for the application in the local wait queue and set the
1996 * notify socket if found.
1997 */
1998 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1999 &wait_queue.head, head) {
2000 health_code_update();
2001 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
2002 wait_node->app->notify_sock = ust_cmd->sock;
2003 cds_list_del(&wait_node->head);
2004 wait_queue.count--;
2005 app = wait_node->app;
2006 free(wait_node);
2007 DBG3("UST app notify socket %d is set", ust_cmd->sock);
2008 break;
2009 }
2010 }
2011
2012 /*
2013 * With no application at this stage the received socket is
2014 * basically useless so close it before we free the cmd data
2015 * structure for good.
2016 */
2017 if (!app) {
2018 ret = close(ust_cmd->sock);
2019 if (ret < 0) {
2020 PERROR("close ust sock dispatch %d", ust_cmd->sock);
2021 }
2022 lttng_fd_put(LTTNG_FD_APPS, 1);
2023 }
2024 free(ust_cmd);
2025 }
2026
2027 if (app) {
2028 /*
2029 * @session_lock_list
2030 *
2031 * Lock the global session list so from the register up to the
2032 * registration done message, no thread can see the application
2033 * and change its state.
2034 */
2035 session_lock_list();
2036 rcu_read_lock();
2037
2038 /*
2039 * Add application to the global hash table. This needs to be
2040 * done before the update to the UST registry can locate the
2041 * application.
2042 */
2043 ust_app_add(app);
2044
2045 /* Set app version. This call will print an error if needed. */
2046 (void) ust_app_version(app);
2047
2048 /* Send notify socket through the notify pipe. */
2049 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2050 app->notify_sock);
2051 if (ret < 0) {
2052 rcu_read_unlock();
2053 session_unlock_list();
2054 /*
2055 * No notify thread, stop the UST tracing. However, this is
2056 * not an internal error of the this thread thus setting
2057 * the health error code to a normal exit.
2058 */
2059 err = 0;
2060 goto error;
2061 }
2062
2063 /*
2064 * Update newly registered application with the tracing
2065 * registry info already enabled information.
2066 */
2067 update_ust_app(app->sock);
2068
2069 /*
2070 * Don't care about return value. Let the manage apps threads
2071 * handle app unregistration upon socket close.
2072 */
2073 (void) ust_app_register_done(app);
2074
2075 /*
2076 * Even if the application socket has been closed, send the app
2077 * to the thread and unregistration will take place at that
2078 * place.
2079 */
2080 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
2081 if (ret < 0) {
2082 rcu_read_unlock();
2083 session_unlock_list();
2084 /*
2085 * No apps. thread, stop the UST tracing. However, this is
2086 * not an internal error of the this thread thus setting
2087 * the health error code to a normal exit.
2088 */
2089 err = 0;
2090 goto error;
2091 }
2092
2093 rcu_read_unlock();
2094 session_unlock_list();
2095 }
2096 } while (node != NULL);
2097
2098 health_poll_entry();
2099 /* Futex wait on queue. Blocking call on futex() */
2100 futex_nto1_wait(&ust_cmd_queue.futex);
2101 health_poll_exit();
2102 }
2103 /* Normal exit, no error */
2104 err = 0;
2105
2106 error:
2107 /* Clean up wait queue. */
2108 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
2109 &wait_queue.head, head) {
2110 cds_list_del(&wait_node->head);
2111 wait_queue.count--;
2112 free(wait_node);
2113 }
2114
2115 /* Empty command queue. */
2116 for (;;) {
2117 /* Dequeue command for registration */
2118 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2119 if (node == NULL) {
2120 break;
2121 }
2122 ust_cmd = caa_container_of(node, struct ust_command, node);
2123 ret = close(ust_cmd->sock);
2124 if (ret < 0) {
2125 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2126 }
2127 lttng_fd_put(LTTNG_FD_APPS, 1);
2128 free(ust_cmd);
2129 }
2130
2131 error_testpoint:
2132 DBG("Dispatch thread dying");
2133 if (err) {
2134 health_error();
2135 ERR("Health error occurred in %s", __func__);
2136 }
2137 health_unregister(health_sessiond);
2138 return NULL;
2139 }
2140
2141 /*
2142 * This thread manage application registration.
2143 */
2144 static void *thread_registration_apps(void *data)
2145 {
2146 int sock = -1, i, ret, pollfd, err = -1;
2147 uint32_t revents, nb_fd;
2148 struct lttng_poll_event events;
2149 /*
2150 * Get allocated in this thread, enqueued to a global queue, dequeued and
2151 * freed in the manage apps thread.
2152 */
2153 struct ust_command *ust_cmd = NULL;
2154
2155 DBG("[thread] Manage application registration started");
2156
2157 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
2158
2159 if (testpoint(sessiond_thread_registration_apps)) {
2160 goto error_testpoint;
2161 }
2162
2163 ret = lttcomm_listen_unix_sock(apps_sock);
2164 if (ret < 0) {
2165 goto error_listen;
2166 }
2167
2168 /*
2169 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2170 * more will be added to this poll set.
2171 */
2172 ret = sessiond_set_thread_pollset(&events, 2);
2173 if (ret < 0) {
2174 goto error_create_poll;
2175 }
2176
2177 /* Add the application registration socket */
2178 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2179 if (ret < 0) {
2180 goto error_poll_add;
2181 }
2182
2183 /* Notify all applications to register */
2184 ret = notify_ust_apps(1);
2185 if (ret < 0) {
2186 ERR("Failed to notify applications or create the wait shared memory.\n"
2187 "Execution continues but there might be problem for already\n"
2188 "running applications that wishes to register.");
2189 }
2190
2191 while (1) {
2192 DBG("Accepting application registration");
2193
2194 /* Inifinite blocking call, waiting for transmission */
2195 restart:
2196 health_poll_entry();
2197 ret = lttng_poll_wait(&events, -1);
2198 health_poll_exit();
2199 if (ret < 0) {
2200 /*
2201 * Restart interrupted system call.
2202 */
2203 if (errno == EINTR) {
2204 goto restart;
2205 }
2206 goto error;
2207 }
2208
2209 nb_fd = ret;
2210
2211 for (i = 0; i < nb_fd; i++) {
2212 health_code_update();
2213
2214 /* Fetch once the poll data */
2215 revents = LTTNG_POLL_GETEV(&events, i);
2216 pollfd = LTTNG_POLL_GETFD(&events, i);
2217
2218 if (!revents) {
2219 /* No activity for this FD (poll implementation). */
2220 continue;
2221 }
2222
2223 /* Thread quit pipe has been closed. Killing thread. */
2224 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
2225 if (ret) {
2226 err = 0;
2227 goto exit;
2228 }
2229
2230 /* Event on the registration socket */
2231 if (pollfd == apps_sock) {
2232 if (revents & LPOLLIN) {
2233 sock = lttcomm_accept_unix_sock(apps_sock);
2234 if (sock < 0) {
2235 goto error;
2236 }
2237
2238 /*
2239 * Set socket timeout for both receiving and ending.
2240 * app_socket_timeout is in seconds, whereas
2241 * lttcomm_setsockopt_rcv_timeout and
2242 * lttcomm_setsockopt_snd_timeout expect msec as
2243 * parameter.
2244 */
2245 (void) lttcomm_setsockopt_rcv_timeout(sock,
2246 app_socket_timeout * 1000);
2247 (void) lttcomm_setsockopt_snd_timeout(sock,
2248 app_socket_timeout * 1000);
2249
2250 /*
2251 * Set the CLOEXEC flag. Return code is useless because
2252 * either way, the show must go on.
2253 */
2254 (void) utils_set_fd_cloexec(sock);
2255
2256 /* Create UST registration command for enqueuing */
2257 ust_cmd = zmalloc(sizeof(struct ust_command));
2258 if (ust_cmd == NULL) {
2259 PERROR("ust command zmalloc");
2260 ret = close(sock);
2261 if (ret) {
2262 PERROR("close");
2263 }
2264 goto error;
2265 }
2266
2267 /*
2268 * Using message-based transmissions to ensure we don't
2269 * have to deal with partially received messages.
2270 */
2271 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2272 if (ret < 0) {
2273 ERR("Exhausted file descriptors allowed for applications.");
2274 free(ust_cmd);
2275 ret = close(sock);
2276 if (ret) {
2277 PERROR("close");
2278 }
2279 sock = -1;
2280 continue;
2281 }
2282
2283 health_code_update();
2284 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2285 if (ret < 0) {
2286 free(ust_cmd);
2287 /* Close socket of the application. */
2288 ret = close(sock);
2289 if (ret) {
2290 PERROR("close");
2291 }
2292 lttng_fd_put(LTTNG_FD_APPS, 1);
2293 sock = -1;
2294 continue;
2295 }
2296 health_code_update();
2297
2298 ust_cmd->sock = sock;
2299 sock = -1;
2300
2301 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2302 " gid:%d sock:%d name:%s (version %d.%d)",
2303 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2304 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2305 ust_cmd->sock, ust_cmd->reg_msg.name,
2306 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
2307
2308 /*
2309 * Lock free enqueue the registration request. The red pill
2310 * has been taken! This apps will be part of the *system*.
2311 */
2312 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
2313
2314 /*
2315 * Wake the registration queue futex. Implicit memory
2316 * barrier with the exchange in cds_wfcq_enqueue.
2317 */
2318 futex_nto1_wake(&ust_cmd_queue.futex);
2319 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2320 ERR("Register apps socket poll error");
2321 goto error;
2322 } else {
2323 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2324 goto error;
2325 }
2326 }
2327 }
2328 }
2329
2330 exit:
2331 error:
2332 /* Notify that the registration thread is gone */
2333 notify_ust_apps(0);
2334
2335 if (apps_sock >= 0) {
2336 ret = close(apps_sock);
2337 if (ret) {
2338 PERROR("close");
2339 }
2340 }
2341 if (sock >= 0) {
2342 ret = close(sock);
2343 if (ret) {
2344 PERROR("close");
2345 }
2346 lttng_fd_put(LTTNG_FD_APPS, 1);
2347 }
2348 unlink(apps_unix_sock_path);
2349
2350 error_poll_add:
2351 lttng_poll_clean(&events);
2352 error_listen:
2353 error_create_poll:
2354 error_testpoint:
2355 DBG("UST Registration thread cleanup complete");
2356 if (err) {
2357 health_error();
2358 ERR("Health error occurred in %s", __func__);
2359 }
2360 health_unregister(health_sessiond);
2361
2362 return NULL;
2363 }
2364
2365 /*
2366 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2367 * exec or it will fails.
2368 */
2369 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2370 {
2371 int ret, clock_ret;
2372 struct timespec timeout;
2373
2374 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2375 consumer_data->consumer_thread_is_ready = 0;
2376
2377 /* Setup pthread condition */
2378 ret = pthread_condattr_init(&consumer_data->condattr);
2379 if (ret) {
2380 errno = ret;
2381 PERROR("pthread_condattr_init consumer data");
2382 goto error;
2383 }
2384
2385 /*
2386 * Set the monotonic clock in order to make sure we DO NOT jump in time
2387 * between the clock_gettime() call and the timedwait call. See bug #324
2388 * for a more details and how we noticed it.
2389 */
2390 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2391 if (ret) {
2392 errno = ret;
2393 PERROR("pthread_condattr_setclock consumer data");
2394 goto error;
2395 }
2396
2397 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2398 if (ret) {
2399 errno = ret;
2400 PERROR("pthread_cond_init consumer data");
2401 goto error;
2402 }
2403
2404 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2405 consumer_data);
2406 if (ret) {
2407 errno = ret;
2408 PERROR("pthread_create consumer");
2409 ret = -1;
2410 goto error;
2411 }
2412
2413 /* We are about to wait on a pthread condition */
2414 pthread_mutex_lock(&consumer_data->cond_mutex);
2415
2416 /* Get time for sem_timedwait absolute timeout */
2417 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2418 /*
2419 * Set the timeout for the condition timed wait even if the clock gettime
2420 * call fails since we might loop on that call and we want to avoid to
2421 * increment the timeout too many times.
2422 */
2423 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2424
2425 /*
2426 * The following loop COULD be skipped in some conditions so this is why we
2427 * set ret to 0 in order to make sure at least one round of the loop is
2428 * done.
2429 */
2430 ret = 0;
2431
2432 /*
2433 * Loop until the condition is reached or when a timeout is reached. Note
2434 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2435 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2436 * possible. This loop does not take any chances and works with both of
2437 * them.
2438 */
2439 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2440 if (clock_ret < 0) {
2441 PERROR("clock_gettime spawn consumer");
2442 /* Infinite wait for the consumerd thread to be ready */
2443 ret = pthread_cond_wait(&consumer_data->cond,
2444 &consumer_data->cond_mutex);
2445 } else {
2446 ret = pthread_cond_timedwait(&consumer_data->cond,
2447 &consumer_data->cond_mutex, &timeout);
2448 }
2449 }
2450
2451 /* Release the pthread condition */
2452 pthread_mutex_unlock(&consumer_data->cond_mutex);
2453
2454 if (ret != 0) {
2455 errno = ret;
2456 if (ret == ETIMEDOUT) {
2457 int pth_ret;
2458
2459 /*
2460 * Call has timed out so we kill the kconsumerd_thread and return
2461 * an error.
2462 */
2463 ERR("Condition timed out. The consumer thread was never ready."
2464 " Killing it");
2465 pth_ret = pthread_cancel(consumer_data->thread);
2466 if (pth_ret < 0) {
2467 PERROR("pthread_cancel consumer thread");
2468 }
2469 } else {
2470 PERROR("pthread_cond_wait failed consumer thread");
2471 }
2472 /* Caller is expecting a negative value on failure. */
2473 ret = -1;
2474 goto error;
2475 }
2476
2477 pthread_mutex_lock(&consumer_data->pid_mutex);
2478 if (consumer_data->pid == 0) {
2479 ERR("Consumerd did not start");
2480 pthread_mutex_unlock(&consumer_data->pid_mutex);
2481 goto error;
2482 }
2483 pthread_mutex_unlock(&consumer_data->pid_mutex);
2484
2485 return 0;
2486
2487 error:
2488 return ret;
2489 }
2490
2491 /*
2492 * Join consumer thread
2493 */
2494 static int join_consumer_thread(struct consumer_data *consumer_data)
2495 {
2496 void *status;
2497
2498 /* Consumer pid must be a real one. */
2499 if (consumer_data->pid > 0) {
2500 int ret;
2501 ret = kill(consumer_data->pid, SIGTERM);
2502 if (ret) {
2503 PERROR("Error killing consumer daemon");
2504 return ret;
2505 }
2506 return pthread_join(consumer_data->thread, &status);
2507 } else {
2508 return 0;
2509 }
2510 }
2511
2512 /*
2513 * Fork and exec a consumer daemon (consumerd).
2514 *
2515 * Return pid if successful else -1.
2516 */
2517 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2518 {
2519 int ret;
2520 pid_t pid;
2521 const char *consumer_to_use;
2522 const char *verbosity;
2523 struct stat st;
2524
2525 DBG("Spawning consumerd");
2526
2527 pid = fork();
2528 if (pid == 0) {
2529 /*
2530 * Exec consumerd.
2531 */
2532 if (opt_verbose_consumer) {
2533 verbosity = "--verbose";
2534 } else if (lttng_opt_quiet) {
2535 verbosity = "--quiet";
2536 } else {
2537 verbosity = "";
2538 }
2539
2540 switch (consumer_data->type) {
2541 case LTTNG_CONSUMER_KERNEL:
2542 /*
2543 * Find out which consumerd to execute. We will first try the
2544 * 64-bit path, then the sessiond's installation directory, and
2545 * fallback on the 32-bit one,
2546 */
2547 DBG3("Looking for a kernel consumer at these locations:");
2548 DBG3(" 1) %s", consumerd64_bin);
2549 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2550 DBG3(" 3) %s", consumerd32_bin);
2551 if (stat(consumerd64_bin, &st) == 0) {
2552 DBG3("Found location #1");
2553 consumer_to_use = consumerd64_bin;
2554 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2555 DBG3("Found location #2");
2556 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2557 } else if (stat(consumerd32_bin, &st) == 0) {
2558 DBG3("Found location #3");
2559 consumer_to_use = consumerd32_bin;
2560 } else {
2561 DBG("Could not find any valid consumerd executable");
2562 ret = -EINVAL;
2563 break;
2564 }
2565 DBG("Using kernel consumer at: %s", consumer_to_use);
2566 ret = execl(consumer_to_use,
2567 "lttng-consumerd", verbosity, "-k",
2568 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2569 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2570 "--group", tracing_group_name,
2571 NULL);
2572 break;
2573 case LTTNG_CONSUMER64_UST:
2574 {
2575 char *tmpnew = NULL;
2576
2577 if (consumerd64_libdir[0] != '\0') {
2578 char *tmp;
2579 size_t tmplen;
2580
2581 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2582 if (!tmp) {
2583 tmp = "";
2584 }
2585 tmplen = strlen("LD_LIBRARY_PATH=")
2586 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2587 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2588 if (!tmpnew) {
2589 ret = -ENOMEM;
2590 goto error;
2591 }
2592 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2593 strcat(tmpnew, consumerd64_libdir);
2594 if (tmp[0] != '\0') {
2595 strcat(tmpnew, ":");
2596 strcat(tmpnew, tmp);
2597 }
2598 ret = putenv(tmpnew);
2599 if (ret) {
2600 ret = -errno;
2601 free(tmpnew);
2602 goto error;
2603 }
2604 }
2605 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2606 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2607 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2608 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2609 "--group", tracing_group_name,
2610 NULL);
2611 if (consumerd64_libdir[0] != '\0') {
2612 free(tmpnew);
2613 }
2614 break;
2615 }
2616 case LTTNG_CONSUMER32_UST:
2617 {
2618 char *tmpnew = NULL;
2619
2620 if (consumerd32_libdir[0] != '\0') {
2621 char *tmp;
2622 size_t tmplen;
2623
2624 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
2625 if (!tmp) {
2626 tmp = "";
2627 }
2628 tmplen = strlen("LD_LIBRARY_PATH=")
2629 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2630 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2631 if (!tmpnew) {
2632 ret = -ENOMEM;
2633 goto error;
2634 }
2635 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2636 strcat(tmpnew, consumerd32_libdir);
2637 if (tmp[0] != '\0') {
2638 strcat(tmpnew, ":");
2639 strcat(tmpnew, tmp);
2640 }
2641 ret = putenv(tmpnew);
2642 if (ret) {
2643 ret = -errno;
2644 free(tmpnew);
2645 goto error;
2646 }
2647 }
2648 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2649 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2650 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2651 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2652 "--group", tracing_group_name,
2653 NULL);
2654 if (consumerd32_libdir[0] != '\0') {
2655 free(tmpnew);
2656 }
2657 break;
2658 }
2659 default:
2660 PERROR("unknown consumer type");
2661 exit(EXIT_FAILURE);
2662 }
2663 if (errno != 0) {
2664 PERROR("Consumer execl()");
2665 }
2666 /* Reaching this point, we got a failure on our execl(). */
2667 exit(EXIT_FAILURE);
2668 } else if (pid > 0) {
2669 ret = pid;
2670 } else {
2671 PERROR("start consumer fork");
2672 ret = -errno;
2673 }
2674 error:
2675 return ret;
2676 }
2677
2678 /*
2679 * Spawn the consumerd daemon and session daemon thread.
2680 */
2681 static int start_consumerd(struct consumer_data *consumer_data)
2682 {
2683 int ret;
2684
2685 /*
2686 * Set the listen() state on the socket since there is a possible race
2687 * between the exec() of the consumer daemon and this call if place in the
2688 * consumer thread. See bug #366 for more details.
2689 */
2690 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2691 if (ret < 0) {
2692 goto error;
2693 }
2694
2695 pthread_mutex_lock(&consumer_data->pid_mutex);
2696 if (consumer_data->pid != 0) {
2697 pthread_mutex_unlock(&consumer_data->pid_mutex);
2698 goto end;
2699 }
2700
2701 ret = spawn_consumerd(consumer_data);
2702 if (ret < 0) {
2703 ERR("Spawning consumerd failed");
2704 pthread_mutex_unlock(&consumer_data->pid_mutex);
2705 goto error;
2706 }
2707
2708 /* Setting up the consumer_data pid */
2709 consumer_data->pid = ret;
2710 DBG2("Consumer pid %d", consumer_data->pid);
2711 pthread_mutex_unlock(&consumer_data->pid_mutex);
2712
2713 DBG2("Spawning consumer control thread");
2714 ret = spawn_consumer_thread(consumer_data);
2715 if (ret < 0) {
2716 ERR("Fatal error spawning consumer control thread");
2717 goto error;
2718 }
2719
2720 end:
2721 return 0;
2722
2723 error:
2724 /* Cleanup already created sockets on error. */
2725 if (consumer_data->err_sock >= 0) {
2726 int err;
2727
2728 err = close(consumer_data->err_sock);
2729 if (err < 0) {
2730 PERROR("close consumer data error socket");
2731 }
2732 }
2733 return ret;
2734 }
2735
2736 /*
2737 * Setup necessary data for kernel tracer action.
2738 */
2739 static int init_kernel_tracer(void)
2740 {
2741 int ret;
2742
2743 /* Modprobe lttng kernel modules */
2744 ret = modprobe_lttng_control();
2745 if (ret < 0) {
2746 goto error;
2747 }
2748
2749 /* Open debugfs lttng */
2750 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2751 if (kernel_tracer_fd < 0) {
2752 DBG("Failed to open %s", module_proc_lttng);
2753 ret = -1;
2754 goto error_open;
2755 }
2756
2757 /* Validate kernel version */
2758 ret = kernel_validate_version(kernel_tracer_fd);
2759 if (ret < 0) {
2760 goto error_version;
2761 }
2762
2763 ret = modprobe_lttng_data();
2764 if (ret < 0) {
2765 goto error_modules;
2766 }
2767
2768 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2769 return 0;
2770
2771 error_version:
2772 modprobe_remove_lttng_control();
2773 ret = close(kernel_tracer_fd);
2774 if (ret) {
2775 PERROR("close");
2776 }
2777 kernel_tracer_fd = -1;
2778 return LTTNG_ERR_KERN_VERSION;
2779
2780 error_modules:
2781 ret = close(kernel_tracer_fd);
2782 if (ret) {
2783 PERROR("close");
2784 }
2785
2786 error_open:
2787 modprobe_remove_lttng_control();
2788
2789 error:
2790 WARN("No kernel tracer available");
2791 kernel_tracer_fd = -1;
2792 if (!is_root) {
2793 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2794 } else {
2795 return LTTNG_ERR_KERN_NA;
2796 }
2797 }
2798
2799
2800 /*
2801 * Copy consumer output from the tracing session to the domain session. The
2802 * function also applies the right modification on a per domain basis for the
2803 * trace files destination directory.
2804 *
2805 * Should *NOT* be called with RCU read-side lock held.
2806 */
2807 static int copy_session_consumer(int domain, struct ltt_session *session)
2808 {
2809 int ret;
2810 const char *dir_name;
2811 struct consumer_output *consumer;
2812
2813 assert(session);
2814 assert(session->consumer);
2815
2816 switch (domain) {
2817 case LTTNG_DOMAIN_KERNEL:
2818 DBG3("Copying tracing session consumer output in kernel session");
2819 /*
2820 * XXX: We should audit the session creation and what this function
2821 * does "extra" in order to avoid a destroy since this function is used
2822 * in the domain session creation (kernel and ust) only. Same for UST
2823 * domain.
2824 */
2825 if (session->kernel_session->consumer) {
2826 consumer_output_put(session->kernel_session->consumer);
2827 }
2828 session->kernel_session->consumer =
2829 consumer_copy_output(session->consumer);
2830 /* Ease our life a bit for the next part */
2831 consumer = session->kernel_session->consumer;
2832 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2833 break;
2834 case LTTNG_DOMAIN_JUL:
2835 case LTTNG_DOMAIN_LOG4J:
2836 case LTTNG_DOMAIN_PYTHON:
2837 case LTTNG_DOMAIN_UST:
2838 DBG3("Copying tracing session consumer output in UST session");
2839 if (session->ust_session->consumer) {
2840 consumer_output_put(session->ust_session->consumer);
2841 }
2842 session->ust_session->consumer =
2843 consumer_copy_output(session->consumer);
2844 /* Ease our life a bit for the next part */
2845 consumer = session->ust_session->consumer;
2846 dir_name = DEFAULT_UST_TRACE_DIR;
2847 break;
2848 default:
2849 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2850 goto error;
2851 }
2852
2853 /* Append correct directory to subdir */
2854 strncat(consumer->subdir, dir_name,
2855 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2856 DBG3("Copy session consumer subdir %s", consumer->subdir);
2857
2858 ret = LTTNG_OK;
2859
2860 error:
2861 return ret;
2862 }
2863
2864 /*
2865 * Create an UST session and add it to the session ust list.
2866 *
2867 * Should *NOT* be called with RCU read-side lock held.
2868 */
2869 static int create_ust_session(struct ltt_session *session,
2870 struct lttng_domain *domain)
2871 {
2872 int ret;
2873 struct ltt_ust_session *lus = NULL;
2874
2875 assert(session);
2876 assert(domain);
2877 assert(session->consumer);
2878
2879 switch (domain->type) {
2880 case LTTNG_DOMAIN_JUL:
2881 case LTTNG_DOMAIN_LOG4J:
2882 case LTTNG_DOMAIN_PYTHON:
2883 case LTTNG_DOMAIN_UST:
2884 break;
2885 default:
2886 ERR("Unknown UST domain on create session %d", domain->type);
2887 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2888 goto error;
2889 }
2890
2891 DBG("Creating UST session");
2892
2893 lus = trace_ust_create_session(session->id);
2894 if (lus == NULL) {
2895 ret = LTTNG_ERR_UST_SESS_FAIL;
2896 goto error;
2897 }
2898
2899 lus->uid = session->uid;
2900 lus->gid = session->gid;
2901 lus->output_traces = session->output_traces;
2902 lus->snapshot_mode = session->snapshot_mode;
2903 lus->live_timer_interval = session->live_timer;
2904 session->ust_session = lus;
2905 if (session->shm_path[0]) {
2906 strncpy(lus->root_shm_path, session->shm_path,
2907 sizeof(lus->root_shm_path));
2908 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
2909 strncpy(lus->shm_path, session->shm_path,
2910 sizeof(lus->shm_path));
2911 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2912 strncat(lus->shm_path, "/ust",
2913 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2914 }
2915 /* Copy session output to the newly created UST session */
2916 ret = copy_session_consumer(domain->type, session);
2917 if (ret != LTTNG_OK) {
2918 goto error;
2919 }
2920
2921 return LTTNG_OK;
2922
2923 error:
2924 free(lus);
2925 session->ust_session = NULL;
2926 return ret;
2927 }
2928
2929 /*
2930 * Create a kernel tracer session then create the default channel.
2931 */
2932 static int create_kernel_session(struct ltt_session *session)
2933 {
2934 int ret;
2935
2936 DBG("Creating kernel session");
2937
2938 ret = kernel_create_session(session, kernel_tracer_fd);
2939 if (ret < 0) {
2940 ret = LTTNG_ERR_KERN_SESS_FAIL;
2941 goto error;
2942 }
2943
2944 /* Code flow safety */
2945 assert(session->kernel_session);
2946
2947 /* Copy session output to the newly created Kernel session */
2948 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2949 if (ret != LTTNG_OK) {
2950 goto error;
2951 }
2952
2953 /* Create directory(ies) on local filesystem. */
2954 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2955 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2956 ret = run_as_mkdir_recursive(
2957 session->kernel_session->consumer->dst.trace_path,
2958 S_IRWXU | S_IRWXG, session->uid, session->gid);
2959 if (ret < 0) {
2960 if (errno != EEXIST) {
2961 ERR("Trace directory creation error");
2962 goto error;
2963 }
2964 }
2965 }
2966
2967 session->kernel_session->uid = session->uid;
2968 session->kernel_session->gid = session->gid;
2969 session->kernel_session->output_traces = session->output_traces;
2970 session->kernel_session->snapshot_mode = session->snapshot_mode;
2971
2972 return LTTNG_OK;
2973
2974 error:
2975 trace_kernel_destroy_session(session->kernel_session);
2976 session->kernel_session = NULL;
2977 return ret;
2978 }
2979
2980 /*
2981 * Count number of session permitted by uid/gid.
2982 */
2983 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2984 {
2985 unsigned int i = 0;
2986 struct ltt_session *session;
2987
2988 DBG("Counting number of available session for UID %d GID %d",
2989 uid, gid);
2990 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2991 /*
2992 * Only list the sessions the user can control.
2993 */
2994 if (!session_access_ok(session, uid, gid)) {
2995 continue;
2996 }
2997 i++;
2998 }
2999 return i;
3000 }
3001
3002 /*
3003 * Process the command requested by the lttng client within the command
3004 * context structure. This function make sure that the return structure (llm)
3005 * is set and ready for transmission before returning.
3006 *
3007 * Return any error encountered or 0 for success.
3008 *
3009 * "sock" is only used for special-case var. len data.
3010 *
3011 * Should *NOT* be called with RCU read-side lock held.
3012 */
3013 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
3014 int *sock_error)
3015 {
3016 int ret = LTTNG_OK;
3017 int need_tracing_session = 1;
3018 int need_domain;
3019
3020 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
3021
3022 assert(!rcu_read_ongoing());
3023
3024 *sock_error = 0;
3025
3026 switch (cmd_ctx->lsm->cmd_type) {
3027 case LTTNG_CREATE_SESSION:
3028 case LTTNG_CREATE_SESSION_SNAPSHOT:
3029 case LTTNG_CREATE_SESSION_LIVE:
3030 case LTTNG_DESTROY_SESSION:
3031 case LTTNG_LIST_SESSIONS:
3032 case LTTNG_LIST_DOMAINS:
3033 case LTTNG_START_TRACE:
3034 case LTTNG_STOP_TRACE:
3035 case LTTNG_DATA_PENDING:
3036 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3037 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3038 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3039 case LTTNG_SNAPSHOT_RECORD:
3040 case LTTNG_SAVE_SESSION:
3041 case LTTNG_SET_SESSION_SHM_PATH:
3042 need_domain = 0;
3043 break;
3044 default:
3045 need_domain = 1;
3046 }
3047
3048 if (opt_no_kernel && need_domain
3049 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
3050 if (!is_root) {
3051 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3052 } else {
3053 ret = LTTNG_ERR_KERN_NA;
3054 }
3055 goto error;
3056 }
3057
3058 /* Deny register consumer if we already have a spawned consumer. */
3059 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3060 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3061 if (kconsumer_data.pid > 0) {
3062 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3063 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3064 goto error;
3065 }
3066 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3067 }
3068
3069 /*
3070 * Check for command that don't needs to allocate a returned payload. We do
3071 * this here so we don't have to make the call for no payload at each
3072 * command.
3073 */
3074 switch(cmd_ctx->lsm->cmd_type) {
3075 case LTTNG_LIST_SESSIONS:
3076 case LTTNG_LIST_TRACEPOINTS:
3077 case LTTNG_LIST_TRACEPOINT_FIELDS:
3078 case LTTNG_LIST_DOMAINS:
3079 case LTTNG_LIST_CHANNELS:
3080 case LTTNG_LIST_EVENTS:
3081 case LTTNG_LIST_SYSCALLS:
3082 case LTTNG_LIST_TRACKER_PIDS:
3083 break;
3084 default:
3085 /* Setup lttng message with no payload */
3086 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
3087 if (ret < 0) {
3088 /* This label does not try to unlock the session */
3089 goto init_setup_error;
3090 }
3091 }
3092
3093 /* Commands that DO NOT need a session. */
3094 switch (cmd_ctx->lsm->cmd_type) {
3095 case LTTNG_CREATE_SESSION:
3096 case LTTNG_CREATE_SESSION_SNAPSHOT:
3097 case LTTNG_CREATE_SESSION_LIVE:
3098 case LTTNG_CALIBRATE:
3099 case LTTNG_LIST_SESSIONS:
3100 case LTTNG_LIST_TRACEPOINTS:
3101 case LTTNG_LIST_SYSCALLS:
3102 case LTTNG_LIST_TRACEPOINT_FIELDS:
3103 case LTTNG_SAVE_SESSION:
3104 need_tracing_session = 0;
3105 break;
3106 default:
3107 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
3108 /*
3109 * We keep the session list lock across _all_ commands
3110 * for now, because the per-session lock does not
3111 * handle teardown properly.
3112 */
3113 session_lock_list();
3114 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3115 if (cmd_ctx->session == NULL) {
3116 ret = LTTNG_ERR_SESS_NOT_FOUND;
3117 goto error;
3118 } else {
3119 /* Acquire lock for the session */
3120 session_lock(cmd_ctx->session);
3121 }
3122 break;
3123 }
3124
3125 /*
3126 * Commands that need a valid session but should NOT create one if none
3127 * exists. Instead of creating one and destroying it when the command is
3128 * handled, process that right before so we save some round trip in useless
3129 * code path.
3130 */
3131 switch (cmd_ctx->lsm->cmd_type) {
3132 case LTTNG_DISABLE_CHANNEL:
3133 case LTTNG_DISABLE_EVENT:
3134 switch (cmd_ctx->lsm->domain.type) {
3135 case LTTNG_DOMAIN_KERNEL:
3136 if (!cmd_ctx->session->kernel_session) {
3137 ret = LTTNG_ERR_NO_CHANNEL;
3138 goto error;
3139 }
3140 break;
3141 case LTTNG_DOMAIN_JUL:
3142 case LTTNG_DOMAIN_LOG4J:
3143 case LTTNG_DOMAIN_PYTHON:
3144 case LTTNG_DOMAIN_UST:
3145 if (!cmd_ctx->session->ust_session) {
3146 ret = LTTNG_ERR_NO_CHANNEL;
3147 goto error;
3148 }
3149 break;
3150 default:
3151 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3152 goto error;
3153 }
3154 default:
3155 break;
3156 }
3157
3158 if (!need_domain) {
3159 goto skip_domain;
3160 }
3161
3162 /*
3163 * Check domain type for specific "pre-action".
3164 */
3165 switch (cmd_ctx->lsm->domain.type) {
3166 case LTTNG_DOMAIN_KERNEL:
3167 if (!is_root) {
3168 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
3169 goto error;
3170 }
3171
3172 /* Kernel tracer check */
3173 if (kernel_tracer_fd == -1) {
3174 /* Basically, load kernel tracer modules */
3175 ret = init_kernel_tracer();
3176 if (ret != 0) {
3177 goto error;
3178 }
3179 }
3180
3181 /* Consumer is in an ERROR state. Report back to client */
3182 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
3183 ret = LTTNG_ERR_NO_KERNCONSUMERD;
3184 goto error;
3185 }
3186
3187 /* Need a session for kernel command */
3188 if (need_tracing_session) {
3189 if (cmd_ctx->session->kernel_session == NULL) {
3190 ret = create_kernel_session(cmd_ctx->session);
3191 if (ret < 0) {
3192 ret = LTTNG_ERR_KERN_SESS_FAIL;
3193 goto error;
3194 }
3195 }
3196
3197 /* Start the kernel consumer daemon */
3198 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3199 if (kconsumer_data.pid == 0 &&
3200 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3201 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3202 ret = start_consumerd(&kconsumer_data);
3203 if (ret < 0) {
3204 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
3205 goto error;
3206 }
3207 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3208 } else {
3209 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3210 }
3211
3212 /*
3213 * The consumer was just spawned so we need to add the socket to
3214 * the consumer output of the session if exist.
3215 */
3216 ret = consumer_create_socket(&kconsumer_data,
3217 cmd_ctx->session->kernel_session->consumer);
3218 if (ret < 0) {
3219 goto error;
3220 }
3221 }
3222
3223 break;
3224 case LTTNG_DOMAIN_JUL:
3225 case LTTNG_DOMAIN_LOG4J:
3226 case LTTNG_DOMAIN_PYTHON:
3227 case LTTNG_DOMAIN_UST:
3228 {
3229 if (!ust_app_supported()) {
3230 ret = LTTNG_ERR_NO_UST;
3231 goto error;
3232 }
3233 /* Consumer is in an ERROR state. Report back to client */
3234 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {