Fix: zero out data structure before using it
[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 _GNU_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <paths.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <inttypes.h>
31 #include <sys/mman.h>
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
35 #include <sys/stat.h>
36 #include <sys/types.h>
37 #include <sys/wait.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <config.h>
41
42 #include <common/common.h>
43 #include <common/compat/socket.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/config/config.h>
50
51 #include "lttng-sessiond.h"
52 #include "buffer-registry.h"
53 #include "channel.h"
54 #include "cmd.h"
55 #include "consumer.h"
56 #include "context.h"
57 #include "event.h"
58 #include "kernel.h"
59 #include "kernel-consumer.h"
60 #include "modprobe.h"
61 #include "shm.h"
62 #include "ust-ctl.h"
63 #include "ust-consumer.h"
64 #include "utils.h"
65 #include "fd-limit.h"
66 #include "health-sessiond.h"
67 #include "testpoint.h"
68 #include "ust-thread.h"
69 #include "jul-thread.h"
70
71 #define CONSUMERD_FILE "lttng-consumerd"
72
73 const char *progname;
74 static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
75 static int tracing_group_name_override;
76 static char *opt_pidfile;
77 static int opt_sig_parent;
78 static int opt_verbose_consumer;
79 static int opt_daemon;
80 static int opt_no_kernel;
81 static int is_root; /* Set to 1 if the daemon is running as root */
82 static pid_t ppid; /* Parent PID for --sig-parent option */
83 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
84 static char *rundir;
85
86 /* Set to 1 when a SIGUSR1 signal is received. */
87 static int recv_child_signal;
88
89 /*
90 * Consumer daemon specific control data. Every value not initialized here is
91 * set to 0 by the static definition.
92 */
93 static struct consumer_data kconsumer_data = {
94 .type = LTTNG_CONSUMER_KERNEL,
95 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
96 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
97 .err_sock = -1,
98 .cmd_sock = -1,
99 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
100 .lock = PTHREAD_MUTEX_INITIALIZER,
101 .cond = PTHREAD_COND_INITIALIZER,
102 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
103 };
104 static struct consumer_data ustconsumer64_data = {
105 .type = LTTNG_CONSUMER64_UST,
106 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
107 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
108 .err_sock = -1,
109 .cmd_sock = -1,
110 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
111 .lock = PTHREAD_MUTEX_INITIALIZER,
112 .cond = PTHREAD_COND_INITIALIZER,
113 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
114 };
115 static struct consumer_data ustconsumer32_data = {
116 .type = LTTNG_CONSUMER32_UST,
117 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
118 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
119 .err_sock = -1,
120 .cmd_sock = -1,
121 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
122 .lock = PTHREAD_MUTEX_INITIALIZER,
123 .cond = PTHREAD_COND_INITIALIZER,
124 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
125 };
126
127 /* Command line options */
128 static const struct option long_options[] = {
129 { "client-sock", 1, 0, 'c' },
130 { "apps-sock", 1, 0, 'a' },
131 { "kconsumerd-cmd-sock", 1, 0, 'C' },
132 { "kconsumerd-err-sock", 1, 0, 'E' },
133 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
134 { "ustconsumerd32-err-sock", 1, 0, 'H' },
135 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
136 { "ustconsumerd64-err-sock", 1, 0, 'F' },
137 { "consumerd32-path", 1, 0, 'u' },
138 { "consumerd32-libdir", 1, 0, 'U' },
139 { "consumerd64-path", 1, 0, 't' },
140 { "consumerd64-libdir", 1, 0, 'T' },
141 { "daemonize", 0, 0, 'd' },
142 { "sig-parent", 0, 0, 'S' },
143 { "help", 0, 0, 'h' },
144 { "group", 1, 0, 'g' },
145 { "version", 0, 0, 'V' },
146 { "quiet", 0, 0, 'q' },
147 { "verbose", 0, 0, 'v' },
148 { "verbose-consumer", 0, 0, 'Z' },
149 { "no-kernel", 0, 0, 'N' },
150 { "pidfile", 1, 0, 'p' },
151 { "jul-tcp-port", 1, 0, 'J' },
152 { "config", 1, 0, 'f' },
153 { NULL, 0, 0, 0 }
154 };
155
156 /* Command line options to ignore from configuration file */
157 static const char *config_ignore_options[] = { "help", "version", "config" };
158
159 /* Shared between threads */
160 static int dispatch_thread_exit;
161
162 /* Global application Unix socket path */
163 static char apps_unix_sock_path[PATH_MAX];
164 /* Global client Unix socket path */
165 static char client_unix_sock_path[PATH_MAX];
166 /* global wait shm path for UST */
167 static char wait_shm_path[PATH_MAX];
168 /* Global health check unix path */
169 static char health_unix_sock_path[PATH_MAX];
170
171 /* Sockets and FDs */
172 static int client_sock = -1;
173 static int apps_sock = -1;
174 int kernel_tracer_fd = -1;
175 static int kernel_poll_pipe[2] = { -1, -1 };
176
177 /*
178 * Quit pipe for all threads. This permits a single cancellation point
179 * for all threads when receiving an event on the pipe.
180 */
181 static int thread_quit_pipe[2] = { -1, -1 };
182
183 /*
184 * This pipe is used to inform the thread managing application communication
185 * that a command is queued and ready to be processed.
186 */
187 static int apps_cmd_pipe[2] = { -1, -1 };
188
189 int apps_cmd_notify_pipe[2] = { -1, -1 };
190
191 /* Pthread, Mutexes and Semaphores */
192 static pthread_t apps_thread;
193 static pthread_t apps_notify_thread;
194 static pthread_t reg_apps_thread;
195 static pthread_t client_thread;
196 static pthread_t kernel_thread;
197 static pthread_t dispatch_thread;
198 static pthread_t health_thread;
199 static pthread_t ht_cleanup_thread;
200 static pthread_t jul_reg_thread;
201
202 /*
203 * UST registration command queue. This queue is tied with a futex and uses a N
204 * wakers / 1 waiter implemented and detailed in futex.c/.h
205 *
206 * The thread_manage_apps and thread_dispatch_ust_registration interact with
207 * this queue and the wait/wake scheme.
208 */
209 static struct ust_cmd_queue ust_cmd_queue;
210
211 /*
212 * Pointer initialized before thread creation.
213 *
214 * This points to the tracing session list containing the session count and a
215 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
216 * MUST NOT be taken if you call a public function in session.c.
217 *
218 * The lock is nested inside the structure: session_list_ptr->lock. Please use
219 * session_lock_list and session_unlock_list for lock acquisition.
220 */
221 static struct ltt_session_list *session_list_ptr;
222
223 int ust_consumerd64_fd = -1;
224 int ust_consumerd32_fd = -1;
225
226 static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
227 static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
228 static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
229 static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
230 static int consumerd32_bin_override;
231 static int consumerd64_bin_override;
232 static int consumerd32_libdir_override;
233 static int consumerd64_libdir_override;
234
235 static const char *module_proc_lttng = "/proc/lttng";
236
237 /*
238 * Consumer daemon state which is changed when spawning it, killing it or in
239 * case of a fatal error.
240 */
241 enum consumerd_state {
242 CONSUMER_STARTED = 1,
243 CONSUMER_STOPPED = 2,
244 CONSUMER_ERROR = 3,
245 };
246
247 /*
248 * This consumer daemon state is used to validate if a client command will be
249 * able to reach the consumer. If not, the client is informed. For instance,
250 * doing a "lttng start" when the consumer state is set to ERROR will return an
251 * error to the client.
252 *
253 * The following example shows a possible race condition of this scheme:
254 *
255 * consumer thread error happens
256 * client cmd arrives
257 * client cmd checks state -> still OK
258 * consumer thread exit, sets error
259 * client cmd try to talk to consumer
260 * ...
261 *
262 * However, since the consumer is a different daemon, we have no way of making
263 * sure the command will reach it safely even with this state flag. This is why
264 * we consider that up to the state validation during command processing, the
265 * command is safe. After that, we can not guarantee the correctness of the
266 * client request vis-a-vis the consumer.
267 */
268 static enum consumerd_state ust_consumerd_state;
269 static enum consumerd_state kernel_consumerd_state;
270
271 /*
272 * Socket timeout for receiving and sending in seconds.
273 */
274 static int app_socket_timeout;
275
276 /* Set in main() with the current page size. */
277 long page_size;
278
279 /* Application health monitoring */
280 struct health_app *health_sessiond;
281
282 /* JUL TCP port for registration. Used by the JUL thread. */
283 unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
284
285 const char * const config_section_name = "sessiond";
286
287 static
288 void setup_consumerd_path(void)
289 {
290 const char *bin, *libdir;
291
292 /*
293 * Allow INSTALL_BIN_PATH to be used as a target path for the
294 * native architecture size consumer if CONFIG_CONSUMER*_PATH
295 * has not been defined.
296 */
297 #if (CAA_BITS_PER_LONG == 32)
298 if (!consumerd32_bin[0]) {
299 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
300 }
301 if (!consumerd32_libdir[0]) {
302 consumerd32_libdir = INSTALL_LIB_PATH;
303 }
304 #elif (CAA_BITS_PER_LONG == 64)
305 if (!consumerd64_bin[0]) {
306 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
307 }
308 if (!consumerd64_libdir[0]) {
309 consumerd64_libdir = INSTALL_LIB_PATH;
310 }
311 #else
312 #error "Unknown bitness"
313 #endif
314
315 /*
316 * runtime env. var. overrides the build default.
317 */
318 bin = getenv("LTTNG_CONSUMERD32_BIN");
319 if (bin) {
320 consumerd32_bin = bin;
321 }
322 bin = getenv("LTTNG_CONSUMERD64_BIN");
323 if (bin) {
324 consumerd64_bin = bin;
325 }
326 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
327 if (libdir) {
328 consumerd32_libdir = libdir;
329 }
330 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
331 if (libdir) {
332 consumerd64_libdir = libdir;
333 }
334 }
335
336 /*
337 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
338 */
339 int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
340 {
341 int ret;
342
343 assert(events);
344
345 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
346 if (ret < 0) {
347 goto error;
348 }
349
350 /* Add quit pipe */
351 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
352 if (ret < 0) {
353 goto error;
354 }
355
356 return 0;
357
358 error:
359 return ret;
360 }
361
362 /*
363 * Check if the thread quit pipe was triggered.
364 *
365 * Return 1 if it was triggered else 0;
366 */
367 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
368 {
369 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
370 return 1;
371 }
372
373 return 0;
374 }
375
376 /*
377 * Init thread quit pipe.
378 *
379 * Return -1 on error or 0 if all pipes are created.
380 */
381 static int init_thread_quit_pipe(void)
382 {
383 int ret, i;
384
385 ret = pipe(thread_quit_pipe);
386 if (ret < 0) {
387 PERROR("thread quit pipe");
388 goto error;
389 }
390
391 for (i = 0; i < 2; i++) {
392 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
393 if (ret < 0) {
394 PERROR("fcntl");
395 goto error;
396 }
397 }
398
399 error:
400 return ret;
401 }
402
403 /*
404 * Stop all threads by closing the thread quit pipe.
405 */
406 static void stop_threads(void)
407 {
408 int ret;
409
410 /* Stopping all threads */
411 DBG("Terminating all threads");
412 ret = notify_thread_pipe(thread_quit_pipe[1]);
413 if (ret < 0) {
414 ERR("write error on thread quit pipe");
415 }
416
417 /* Dispatch thread */
418 CMM_STORE_SHARED(dispatch_thread_exit, 1);
419 futex_nto1_wake(&ust_cmd_queue.futex);
420 }
421
422 /*
423 * Close every consumer sockets.
424 */
425 static void close_consumer_sockets(void)
426 {
427 int ret;
428
429 if (kconsumer_data.err_sock >= 0) {
430 ret = close(kconsumer_data.err_sock);
431 if (ret < 0) {
432 PERROR("kernel consumer err_sock close");
433 }
434 }
435 if (ustconsumer32_data.err_sock >= 0) {
436 ret = close(ustconsumer32_data.err_sock);
437 if (ret < 0) {
438 PERROR("UST consumerd32 err_sock close");
439 }
440 }
441 if (ustconsumer64_data.err_sock >= 0) {
442 ret = close(ustconsumer64_data.err_sock);
443 if (ret < 0) {
444 PERROR("UST consumerd64 err_sock close");
445 }
446 }
447 if (kconsumer_data.cmd_sock >= 0) {
448 ret = close(kconsumer_data.cmd_sock);
449 if (ret < 0) {
450 PERROR("kernel consumer cmd_sock close");
451 }
452 }
453 if (ustconsumer32_data.cmd_sock >= 0) {
454 ret = close(ustconsumer32_data.cmd_sock);
455 if (ret < 0) {
456 PERROR("UST consumerd32 cmd_sock close");
457 }
458 }
459 if (ustconsumer64_data.cmd_sock >= 0) {
460 ret = close(ustconsumer64_data.cmd_sock);
461 if (ret < 0) {
462 PERROR("UST consumerd64 cmd_sock close");
463 }
464 }
465 }
466
467 /*
468 * Cleanup the daemon
469 */
470 static void cleanup(void)
471 {
472 int ret;
473 struct ltt_session *sess, *stmp;
474 char path[PATH_MAX];
475
476 DBG("Cleaning up");
477
478 /*
479 * Close the thread quit pipe. It has already done its job,
480 * since we are now called.
481 */
482 utils_close_pipe(thread_quit_pipe);
483
484 /*
485 * If opt_pidfile is undefined, the default file will be wiped when
486 * removing the rundir.
487 */
488 if (opt_pidfile) {
489 ret = remove(opt_pidfile);
490 if (ret < 0) {
491 PERROR("remove pidfile %s", opt_pidfile);
492 }
493 }
494
495 DBG("Removing sessiond and consumerd content of directory %s", rundir);
496
497 /* sessiond */
498 snprintf(path, PATH_MAX,
499 "%s/%s",
500 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
501 DBG("Removing %s", path);
502 (void) unlink(path);
503
504 snprintf(path, PATH_MAX, "%s/%s", rundir,
505 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
506 DBG("Removing %s", path);
507 (void) unlink(path);
508
509 /* kconsumerd */
510 snprintf(path, PATH_MAX,
511 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
512 rundir);
513 DBG("Removing %s", path);
514 (void) unlink(path);
515
516 snprintf(path, PATH_MAX,
517 DEFAULT_KCONSUMERD_PATH,
518 rundir);
519 DBG("Removing directory %s", path);
520 (void) rmdir(path);
521
522 /* ust consumerd 32 */
523 snprintf(path, PATH_MAX,
524 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
525 rundir);
526 DBG("Removing %s", path);
527 (void) unlink(path);
528
529 snprintf(path, PATH_MAX,
530 DEFAULT_USTCONSUMERD32_PATH,
531 rundir);
532 DBG("Removing directory %s", path);
533 (void) rmdir(path);
534
535 /* ust consumerd 64 */
536 snprintf(path, PATH_MAX,
537 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
538 rundir);
539 DBG("Removing %s", path);
540 (void) unlink(path);
541
542 snprintf(path, PATH_MAX,
543 DEFAULT_USTCONSUMERD64_PATH,
544 rundir);
545 DBG("Removing directory %s", path);
546 (void) rmdir(path);
547
548 /*
549 * We do NOT rmdir rundir because there are other processes
550 * using it, for instance lttng-relayd, which can start in
551 * parallel with this teardown.
552 */
553
554 free(rundir);
555
556 DBG("Cleaning up all sessions");
557
558 /* Destroy session list mutex */
559 if (session_list_ptr != NULL) {
560 pthread_mutex_destroy(&session_list_ptr->lock);
561
562 /* Cleanup ALL session */
563 cds_list_for_each_entry_safe(sess, stmp,
564 &session_list_ptr->head, list) {
565 cmd_destroy_session(sess, kernel_poll_pipe[1]);
566 }
567 }
568
569 DBG("Closing all UST sockets");
570 ust_app_clean_list();
571 buffer_reg_destroy_registries();
572
573 if (is_root && !opt_no_kernel) {
574 DBG2("Closing kernel fd");
575 if (kernel_tracer_fd >= 0) {
576 ret = close(kernel_tracer_fd);
577 if (ret) {
578 PERROR("close");
579 }
580 }
581 DBG("Unloading kernel modules");
582 modprobe_remove_lttng_all();
583 }
584
585 close_consumer_sockets();
586
587 /*
588 * If the override option is set, the pointer points to a *non* const thus
589 * freeing it even though the variable type is set to const.
590 */
591 if (tracing_group_name_override) {
592 free((void *) tracing_group_name);
593 }
594 if (consumerd32_bin_override) {
595 free((void *) consumerd32_bin);
596 }
597 if (consumerd64_bin_override) {
598 free((void *) consumerd64_bin);
599 }
600 if (consumerd32_libdir_override) {
601 free((void *) consumerd32_libdir);
602 }
603 if (consumerd64_libdir_override) {
604 free((void *) consumerd64_libdir);
605 }
606
607 if (opt_pidfile) {
608 free(opt_pidfile);
609 }
610
611 /* <fun> */
612 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
613 "Matthew, BEET driven development works!%c[%dm",
614 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
615 /* </fun> */
616 }
617
618 /*
619 * Send data on a unix socket using the liblttsessiondcomm API.
620 *
621 * Return lttcomm error code.
622 */
623 static int send_unix_sock(int sock, void *buf, size_t len)
624 {
625 /* Check valid length */
626 if (len == 0) {
627 return -1;
628 }
629
630 return lttcomm_send_unix_sock(sock, buf, len);
631 }
632
633 /*
634 * Free memory of a command context structure.
635 */
636 static void clean_command_ctx(struct command_ctx **cmd_ctx)
637 {
638 DBG("Clean command context structure");
639 if (*cmd_ctx) {
640 if ((*cmd_ctx)->llm) {
641 free((*cmd_ctx)->llm);
642 }
643 if ((*cmd_ctx)->lsm) {
644 free((*cmd_ctx)->lsm);
645 }
646 free(*cmd_ctx);
647 *cmd_ctx = NULL;
648 }
649 }
650
651 /*
652 * Notify UST applications using the shm mmap futex.
653 */
654 static int notify_ust_apps(int active)
655 {
656 char *wait_shm_mmap;
657
658 DBG("Notifying applications of session daemon state: %d", active);
659
660 /* See shm.c for this call implying mmap, shm and futex calls */
661 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
662 if (wait_shm_mmap == NULL) {
663 goto error;
664 }
665
666 /* Wake waiting process */
667 futex_wait_update((int32_t *) wait_shm_mmap, active);
668
669 /* Apps notified successfully */
670 return 0;
671
672 error:
673 return -1;
674 }
675
676 /*
677 * Setup the outgoing data buffer for the response (llm) by allocating the
678 * right amount of memory and copying the original information from the lsm
679 * structure.
680 *
681 * Return total size of the buffer pointed by buf.
682 */
683 static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
684 {
685 int ret, buf_size;
686
687 buf_size = size;
688
689 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
690 if (cmd_ctx->llm == NULL) {
691 PERROR("zmalloc");
692 ret = -ENOMEM;
693 goto error;
694 }
695
696 /* Copy common data */
697 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
698 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
699
700 cmd_ctx->llm->data_size = size;
701 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
702
703 return buf_size;
704
705 error:
706 return ret;
707 }
708
709 /*
710 * Update the kernel poll set of all channel fd available over all tracing
711 * session. Add the wakeup pipe at the end of the set.
712 */
713 static int update_kernel_poll(struct lttng_poll_event *events)
714 {
715 int ret;
716 struct ltt_session *session;
717 struct ltt_kernel_channel *channel;
718
719 DBG("Updating kernel poll set");
720
721 session_lock_list();
722 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
723 session_lock(session);
724 if (session->kernel_session == NULL) {
725 session_unlock(session);
726 continue;
727 }
728
729 cds_list_for_each_entry(channel,
730 &session->kernel_session->channel_list.head, list) {
731 /* Add channel fd to the kernel poll set */
732 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
733 if (ret < 0) {
734 session_unlock(session);
735 goto error;
736 }
737 DBG("Channel fd %d added to kernel set", channel->fd);
738 }
739 session_unlock(session);
740 }
741 session_unlock_list();
742
743 return 0;
744
745 error:
746 session_unlock_list();
747 return -1;
748 }
749
750 /*
751 * Find the channel fd from 'fd' over all tracing session. When found, check
752 * for new channel stream and send those stream fds to the kernel consumer.
753 *
754 * Useful for CPU hotplug feature.
755 */
756 static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
757 {
758 int ret = 0;
759 struct ltt_session *session;
760 struct ltt_kernel_session *ksess;
761 struct ltt_kernel_channel *channel;
762
763 DBG("Updating kernel streams for channel fd %d", fd);
764
765 session_lock_list();
766 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
767 session_lock(session);
768 if (session->kernel_session == NULL) {
769 session_unlock(session);
770 continue;
771 }
772 ksess = session->kernel_session;
773
774 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
775 if (channel->fd == fd) {
776 DBG("Channel found, updating kernel streams");
777 ret = kernel_open_channel_stream(channel);
778 if (ret < 0) {
779 goto error;
780 }
781 /* Update the stream global counter */
782 ksess->stream_count_global += ret;
783
784 /*
785 * Have we already sent fds to the consumer? If yes, it means
786 * that tracing is started so it is safe to send our updated
787 * stream fds.
788 */
789 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
790 struct lttng_ht_iter iter;
791 struct consumer_socket *socket;
792
793 rcu_read_lock();
794 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
795 &iter.iter, socket, node.node) {
796 pthread_mutex_lock(socket->lock);
797 ret = kernel_consumer_send_channel_stream(socket,
798 channel, ksess,
799 session->output_traces ? 1 : 0);
800 pthread_mutex_unlock(socket->lock);
801 if (ret < 0) {
802 rcu_read_unlock();
803 goto error;
804 }
805 }
806 rcu_read_unlock();
807 }
808 goto error;
809 }
810 }
811 session_unlock(session);
812 }
813 session_unlock_list();
814 return ret;
815
816 error:
817 session_unlock(session);
818 session_unlock_list();
819 return ret;
820 }
821
822 /*
823 * For each tracing session, update newly registered apps. The session list
824 * lock MUST be acquired before calling this.
825 */
826 static void update_ust_app(int app_sock)
827 {
828 struct ltt_session *sess, *stmp;
829
830 /* Consumer is in an ERROR state. Stop any application update. */
831 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
832 /* Stop the update process since the consumer is dead. */
833 return;
834 }
835
836 /* For all tracing session(s) */
837 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
838 session_lock(sess);
839 if (sess->ust_session) {
840 ust_app_global_update(sess->ust_session, app_sock);
841 }
842 session_unlock(sess);
843 }
844 }
845
846 /*
847 * This thread manage event coming from the kernel.
848 *
849 * Features supported in this thread:
850 * -) CPU Hotplug
851 */
852 static void *thread_manage_kernel(void *data)
853 {
854 int ret, i, pollfd, update_poll_flag = 1, err = -1;
855 uint32_t revents, nb_fd;
856 char tmp;
857 struct lttng_poll_event events;
858
859 DBG("[thread] Thread manage kernel started");
860
861 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
862
863 /*
864 * This first step of the while is to clean this structure which could free
865 * non NULL pointers so initialize it before the loop.
866 */
867 lttng_poll_init(&events);
868
869 if (testpoint(thread_manage_kernel)) {
870 goto error_testpoint;
871 }
872
873 health_code_update();
874
875 if (testpoint(thread_manage_kernel_before_loop)) {
876 goto error_testpoint;
877 }
878
879 while (1) {
880 health_code_update();
881
882 if (update_poll_flag == 1) {
883 /* Clean events object. We are about to populate it again. */
884 lttng_poll_clean(&events);
885
886 ret = sessiond_set_thread_pollset(&events, 2);
887 if (ret < 0) {
888 goto error_poll_create;
889 }
890
891 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
892 if (ret < 0) {
893 goto error;
894 }
895
896 /* This will add the available kernel channel if any. */
897 ret = update_kernel_poll(&events);
898 if (ret < 0) {
899 goto error;
900 }
901 update_poll_flag = 0;
902 }
903
904 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
905
906 /* Poll infinite value of time */
907 restart:
908 health_poll_entry();
909 ret = lttng_poll_wait(&events, -1);
910 health_poll_exit();
911 if (ret < 0) {
912 /*
913 * Restart interrupted system call.
914 */
915 if (errno == EINTR) {
916 goto restart;
917 }
918 goto error;
919 } else if (ret == 0) {
920 /* Should not happen since timeout is infinite */
921 ERR("Return value of poll is 0 with an infinite timeout.\n"
922 "This should not have happened! Continuing...");
923 continue;
924 }
925
926 nb_fd = ret;
927
928 for (i = 0; i < nb_fd; i++) {
929 /* Fetch once the poll data */
930 revents = LTTNG_POLL_GETEV(&events, i);
931 pollfd = LTTNG_POLL_GETFD(&events, i);
932
933 health_code_update();
934
935 /* Thread quit pipe has been closed. Killing thread. */
936 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
937 if (ret) {
938 err = 0;
939 goto exit;
940 }
941
942 /* Check for data on kernel pipe */
943 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
944 (void) lttng_read(kernel_poll_pipe[0],
945 &tmp, 1);
946 /*
947 * Ret value is useless here, if this pipe gets any actions an
948 * update is required anyway.
949 */
950 update_poll_flag = 1;
951 continue;
952 } else {
953 /*
954 * New CPU detected by the kernel. Adding kernel stream to
955 * kernel session and updating the kernel consumer
956 */
957 if (revents & LPOLLIN) {
958 ret = update_kernel_stream(&kconsumer_data, pollfd);
959 if (ret < 0) {
960 continue;
961 }
962 break;
963 /*
964 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
965 * and unregister kernel stream at this point.
966 */
967 }
968 }
969 }
970 }
971
972 exit:
973 error:
974 lttng_poll_clean(&events);
975 error_poll_create:
976 error_testpoint:
977 utils_close_pipe(kernel_poll_pipe);
978 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
979 if (err) {
980 health_error();
981 ERR("Health error occurred in %s", __func__);
982 WARN("Kernel thread died unexpectedly. "
983 "Kernel tracing can continue but CPU hotplug is disabled.");
984 }
985 health_unregister(health_sessiond);
986 DBG("Kernel thread dying");
987 return NULL;
988 }
989
990 /*
991 * Signal pthread condition of the consumer data that the thread.
992 */
993 static void signal_consumer_condition(struct consumer_data *data, int state)
994 {
995 pthread_mutex_lock(&data->cond_mutex);
996
997 /*
998 * The state is set before signaling. It can be any value, it's the waiter
999 * job to correctly interpret this condition variable associated to the
1000 * consumer pthread_cond.
1001 *
1002 * A value of 0 means that the corresponding thread of the consumer data
1003 * was not started. 1 indicates that the thread has started and is ready
1004 * for action. A negative value means that there was an error during the
1005 * thread bootstrap.
1006 */
1007 data->consumer_thread_is_ready = state;
1008 (void) pthread_cond_signal(&data->cond);
1009
1010 pthread_mutex_unlock(&data->cond_mutex);
1011 }
1012
1013 /*
1014 * This thread manage the consumer error sent back to the session daemon.
1015 */
1016 static void *thread_manage_consumer(void *data)
1017 {
1018 int sock = -1, i, ret, pollfd, err = -1;
1019 uint32_t revents, nb_fd;
1020 enum lttcomm_return_code code;
1021 struct lttng_poll_event events;
1022 struct consumer_data *consumer_data = data;
1023
1024 DBG("[thread] Manage consumer started");
1025
1026 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
1027
1028 health_code_update();
1029
1030 /*
1031 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1032 * metadata_sock. Nothing more will be added to this poll set.
1033 */
1034 ret = sessiond_set_thread_pollset(&events, 3);
1035 if (ret < 0) {
1036 goto error_poll;
1037 }
1038
1039 /*
1040 * The error socket here is already in a listening state which was done
1041 * just before spawning this thread to avoid a race between the consumer
1042 * daemon exec trying to connect and the listen() call.
1043 */
1044 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
1045 if (ret < 0) {
1046 goto error;
1047 }
1048
1049 health_code_update();
1050
1051 /* Infinite blocking call, waiting for transmission */
1052 restart:
1053 health_poll_entry();
1054
1055 if (testpoint(thread_manage_consumer)) {
1056 goto error;
1057 }
1058
1059 ret = lttng_poll_wait(&events, -1);
1060 health_poll_exit();
1061 if (ret < 0) {
1062 /*
1063 * Restart interrupted system call.
1064 */
1065 if (errno == EINTR) {
1066 goto restart;
1067 }
1068 goto error;
1069 }
1070
1071 nb_fd = ret;
1072
1073 for (i = 0; i < nb_fd; i++) {
1074 /* Fetch once the poll data */
1075 revents = LTTNG_POLL_GETEV(&events, i);
1076 pollfd = LTTNG_POLL_GETFD(&events, i);
1077
1078 health_code_update();
1079
1080 /* Thread quit pipe has been closed. Killing thread. */
1081 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1082 if (ret) {
1083 err = 0;
1084 goto exit;
1085 }
1086
1087 /* Event on the registration socket */
1088 if (pollfd == consumer_data->err_sock) {
1089 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1090 ERR("consumer err socket poll error");
1091 goto error;
1092 }
1093 }
1094 }
1095
1096 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1097 if (sock < 0) {
1098 goto error;
1099 }
1100
1101 /*
1102 * Set the CLOEXEC flag. Return code is useless because either way, the
1103 * show must go on.
1104 */
1105 (void) utils_set_fd_cloexec(sock);
1106
1107 health_code_update();
1108
1109 DBG2("Receiving code from consumer err_sock");
1110
1111 /* Getting status code from kconsumerd */
1112 ret = lttcomm_recv_unix_sock(sock, &code,
1113 sizeof(enum lttcomm_return_code));
1114 if (ret <= 0) {
1115 goto error;
1116 }
1117
1118 health_code_update();
1119
1120 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
1121 /* Connect both socket, command and metadata. */
1122 consumer_data->cmd_sock =
1123 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1124 consumer_data->metadata_fd =
1125 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1126 if (consumer_data->cmd_sock < 0
1127 || consumer_data->metadata_fd < 0) {
1128 PERROR("consumer connect cmd socket");
1129 /* On error, signal condition and quit. */
1130 signal_consumer_condition(consumer_data, -1);
1131 goto error;
1132 }
1133 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
1134 /* Create metadata socket lock. */
1135 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1136 if (consumer_data->metadata_sock.lock == NULL) {
1137 PERROR("zmalloc pthread mutex");
1138 ret = -1;
1139 goto error;
1140 }
1141 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1142
1143 signal_consumer_condition(consumer_data, 1);
1144 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1145 DBG("Consumer metadata socket ready (fd: %d)",
1146 consumer_data->metadata_fd);
1147 } else {
1148 ERR("consumer error when waiting for SOCK_READY : %s",
1149 lttcomm_get_readable_code(-code));
1150 goto error;
1151 }
1152
1153 /* Remove the consumerd error sock since we've established a connexion */
1154 ret = lttng_poll_del(&events, consumer_data->err_sock);
1155 if (ret < 0) {
1156 goto error;
1157 }
1158
1159 /* Add new accepted error socket. */
1160 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1161 if (ret < 0) {
1162 goto error;
1163 }
1164
1165 /* Add metadata socket that is successfully connected. */
1166 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
1167 LPOLLIN | LPOLLRDHUP);
1168 if (ret < 0) {
1169 goto error;
1170 }
1171
1172 health_code_update();
1173
1174 /* Infinite blocking call, waiting for transmission */
1175 restart_poll:
1176 while (1) {
1177 health_poll_entry();
1178 ret = lttng_poll_wait(&events, -1);
1179 health_poll_exit();
1180 if (ret < 0) {
1181 /*
1182 * Restart interrupted system call.
1183 */
1184 if (errno == EINTR) {
1185 goto restart_poll;
1186 }
1187 goto error;
1188 }
1189
1190 nb_fd = ret;
1191
1192 for (i = 0; i < nb_fd; i++) {
1193 /* Fetch once the poll data */
1194 revents = LTTNG_POLL_GETEV(&events, i);
1195 pollfd = LTTNG_POLL_GETFD(&events, i);
1196
1197 health_code_update();
1198
1199 /* Thread quit pipe has been closed. Killing thread. */
1200 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1201 if (ret) {
1202 err = 0;
1203 goto exit;
1204 }
1205
1206 if (pollfd == sock) {
1207 /* Event on the consumerd socket */
1208 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1209 ERR("consumer err socket second poll error");
1210 goto error;
1211 }
1212 health_code_update();
1213 /* Wait for any kconsumerd error */
1214 ret = lttcomm_recv_unix_sock(sock, &code,
1215 sizeof(enum lttcomm_return_code));
1216 if (ret <= 0) {
1217 ERR("consumer closed the command socket");
1218 goto error;
1219 }
1220
1221 ERR("consumer return code : %s",
1222 lttcomm_get_readable_code(-code));
1223
1224 goto exit;
1225 } else if (pollfd == consumer_data->metadata_fd) {
1226 /* UST metadata requests */
1227 ret = ust_consumer_metadata_request(
1228 &consumer_data->metadata_sock);
1229 if (ret < 0) {
1230 ERR("Handling metadata request");
1231 goto error;
1232 }
1233 break;
1234 } else {
1235 ERR("Unknown pollfd");
1236 goto error;
1237 }
1238 }
1239 health_code_update();
1240 }
1241
1242 exit:
1243 error:
1244 /*
1245 * We lock here because we are about to close the sockets and some other
1246 * thread might be using them so get exclusive access which will abort all
1247 * other consumer command by other threads.
1248 */
1249 pthread_mutex_lock(&consumer_data->lock);
1250
1251 /* Immediately set the consumerd state to stopped */
1252 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1253 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1254 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1255 consumer_data->type == LTTNG_CONSUMER32_UST) {
1256 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1257 } else {
1258 /* Code flow error... */
1259 assert(0);
1260 }
1261
1262 if (consumer_data->err_sock >= 0) {
1263 ret = close(consumer_data->err_sock);
1264 if (ret) {
1265 PERROR("close");
1266 }
1267 consumer_data->err_sock = -1;
1268 }
1269 if (consumer_data->cmd_sock >= 0) {
1270 ret = close(consumer_data->cmd_sock);
1271 if (ret) {
1272 PERROR("close");
1273 }
1274 consumer_data->cmd_sock = -1;
1275 }
1276 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1277 ret = close(*consumer_data->metadata_sock.fd_ptr);
1278 if (ret) {
1279 PERROR("close");
1280 }
1281 }
1282
1283 if (sock >= 0) {
1284 ret = close(sock);
1285 if (ret) {
1286 PERROR("close");
1287 }
1288 }
1289
1290 unlink(consumer_data->err_unix_sock_path);
1291 unlink(consumer_data->cmd_unix_sock_path);
1292 consumer_data->pid = 0;
1293 pthread_mutex_unlock(&consumer_data->lock);
1294
1295 /* Cleanup metadata socket mutex. */
1296 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1297 free(consumer_data->metadata_sock.lock);
1298
1299 lttng_poll_clean(&events);
1300 error_poll:
1301 if (err) {
1302 health_error();
1303 ERR("Health error occurred in %s", __func__);
1304 }
1305 health_unregister(health_sessiond);
1306 DBG("consumer thread cleanup completed");
1307
1308 return NULL;
1309 }
1310
1311 /*
1312 * This thread manage application communication.
1313 */
1314 static void *thread_manage_apps(void *data)
1315 {
1316 int i, ret, pollfd, err = -1;
1317 ssize_t size_ret;
1318 uint32_t revents, nb_fd;
1319 struct lttng_poll_event events;
1320
1321 DBG("[thread] Manage application started");
1322
1323 rcu_register_thread();
1324 rcu_thread_online();
1325
1326 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
1327
1328 if (testpoint(thread_manage_apps)) {
1329 goto error_testpoint;
1330 }
1331
1332 health_code_update();
1333
1334 ret = sessiond_set_thread_pollset(&events, 2);
1335 if (ret < 0) {
1336 goto error_poll_create;
1337 }
1338
1339 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1340 if (ret < 0) {
1341 goto error;
1342 }
1343
1344 if (testpoint(thread_manage_apps_before_loop)) {
1345 goto error;
1346 }
1347
1348 health_code_update();
1349
1350 while (1) {
1351 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
1352
1353 /* Inifinite blocking call, waiting for transmission */
1354 restart:
1355 health_poll_entry();
1356 ret = lttng_poll_wait(&events, -1);
1357 health_poll_exit();
1358 if (ret < 0) {
1359 /*
1360 * Restart interrupted system call.
1361 */
1362 if (errno == EINTR) {
1363 goto restart;
1364 }
1365 goto error;
1366 }
1367
1368 nb_fd = ret;
1369
1370 for (i = 0; i < nb_fd; i++) {
1371 /* Fetch once the poll data */
1372 revents = LTTNG_POLL_GETEV(&events, i);
1373 pollfd = LTTNG_POLL_GETFD(&events, i);
1374
1375 health_code_update();
1376
1377 /* Thread quit pipe has been closed. Killing thread. */
1378 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1379 if (ret) {
1380 err = 0;
1381 goto exit;
1382 }
1383
1384 /* Inspect the apps cmd pipe */
1385 if (pollfd == apps_cmd_pipe[0]) {
1386 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1387 ERR("Apps command pipe error");
1388 goto error;
1389 } else if (revents & LPOLLIN) {
1390 int sock;
1391
1392 /* Empty pipe */
1393 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1394 if (size_ret < sizeof(sock)) {
1395 PERROR("read apps cmd pipe");
1396 goto error;
1397 }
1398
1399 health_code_update();
1400
1401 /*
1402 * We only monitor the error events of the socket. This
1403 * thread does not handle any incoming data from UST
1404 * (POLLIN).
1405 */
1406 ret = lttng_poll_add(&events, sock,
1407 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1408 if (ret < 0) {
1409 goto error;
1410 }
1411
1412 DBG("Apps with sock %d added to poll set", sock);
1413 }
1414 } else {
1415 /*
1416 * At this point, we know that a registered application made
1417 * the event at poll_wait.
1418 */
1419 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1420 /* Removing from the poll set */
1421 ret = lttng_poll_del(&events, pollfd);
1422 if (ret < 0) {
1423 goto error;
1424 }
1425
1426 /* Socket closed on remote end. */
1427 ust_app_unregister(pollfd);
1428 }
1429 }
1430
1431 health_code_update();
1432 }
1433 }
1434
1435 exit:
1436 error:
1437 lttng_poll_clean(&events);
1438 error_poll_create:
1439 error_testpoint:
1440 utils_close_pipe(apps_cmd_pipe);
1441 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1442
1443 /*
1444 * We don't clean the UST app hash table here since already registered
1445 * applications can still be controlled so let them be until the session
1446 * daemon dies or the applications stop.
1447 */
1448
1449 if (err) {
1450 health_error();
1451 ERR("Health error occurred in %s", __func__);
1452 }
1453 health_unregister(health_sessiond);
1454 DBG("Application communication apps thread cleanup complete");
1455 rcu_thread_offline();
1456 rcu_unregister_thread();
1457 return NULL;
1458 }
1459
1460 /*
1461 * Send a socket to a thread This is called from the dispatch UST registration
1462 * thread once all sockets are set for the application.
1463 *
1464 * The sock value can be invalid, we don't really care, the thread will handle
1465 * it and make the necessary cleanup if so.
1466 *
1467 * On success, return 0 else a negative value being the errno message of the
1468 * write().
1469 */
1470 static int send_socket_to_thread(int fd, int sock)
1471 {
1472 ssize_t ret;
1473
1474 /*
1475 * It's possible that the FD is set as invalid with -1 concurrently just
1476 * before calling this function being a shutdown state of the thread.
1477 */
1478 if (fd < 0) {
1479 ret = -EBADF;
1480 goto error;
1481 }
1482
1483 ret = lttng_write(fd, &sock, sizeof(sock));
1484 if (ret < sizeof(sock)) {
1485 PERROR("write apps pipe %d", fd);
1486 if (ret < 0) {
1487 ret = -errno;
1488 }
1489 goto error;
1490 }
1491
1492 /* All good. Don't send back the write positive ret value. */
1493 ret = 0;
1494 error:
1495 return (int) ret;
1496 }
1497
1498 /*
1499 * Sanitize the wait queue of the dispatch registration thread meaning removing
1500 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1501 * notify socket is never received.
1502 */
1503 static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1504 {
1505 int ret, nb_fd = 0, i;
1506 unsigned int fd_added = 0;
1507 struct lttng_poll_event events;
1508 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1509
1510 assert(wait_queue);
1511
1512 lttng_poll_init(&events);
1513
1514 /* Just skip everything for an empty queue. */
1515 if (!wait_queue->count) {
1516 goto end;
1517 }
1518
1519 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1520 if (ret < 0) {
1521 goto error_create;
1522 }
1523
1524 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1525 &wait_queue->head, head) {
1526 assert(wait_node->app);
1527 ret = lttng_poll_add(&events, wait_node->app->sock,
1528 LPOLLHUP | LPOLLERR);
1529 if (ret < 0) {
1530 goto error;
1531 }
1532
1533 fd_added = 1;
1534 }
1535
1536 if (!fd_added) {
1537 goto end;
1538 }
1539
1540 /*
1541 * Poll but don't block so we can quickly identify the faulty events and
1542 * clean them afterwards from the wait queue.
1543 */
1544 ret = lttng_poll_wait(&events, 0);
1545 if (ret < 0) {
1546 goto error;
1547 }
1548 nb_fd = ret;
1549
1550 for (i = 0; i < nb_fd; i++) {
1551 /* Get faulty FD. */
1552 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1553 int pollfd = LTTNG_POLL_GETFD(&events, i);
1554
1555 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1556 &wait_queue->head, head) {
1557 if (pollfd == wait_node->app->sock &&
1558 (revents & (LPOLLHUP | LPOLLERR))) {
1559 cds_list_del(&wait_node->head);
1560 wait_queue->count--;
1561 ust_app_destroy(wait_node->app);
1562 free(wait_node);
1563 break;
1564 }
1565 }
1566 }
1567
1568 if (nb_fd > 0) {
1569 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1570 }
1571
1572 end:
1573 lttng_poll_clean(&events);
1574 return;
1575
1576 error:
1577 lttng_poll_clean(&events);
1578 error_create:
1579 ERR("Unable to sanitize wait queue");
1580 return;
1581 }
1582
1583 /*
1584 * Dispatch request from the registration threads to the application
1585 * communication thread.
1586 */
1587 static void *thread_dispatch_ust_registration(void *data)
1588 {
1589 int ret, err = -1;
1590 struct cds_wfq_node *node;
1591 struct ust_command *ust_cmd = NULL;
1592 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1593 struct ust_reg_wait_queue wait_queue = {
1594 .count = 0,
1595 };
1596
1597 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
1598
1599 health_code_update();
1600
1601 CDS_INIT_LIST_HEAD(&wait_queue.head);
1602
1603 DBG("[thread] Dispatch UST command started");
1604
1605 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
1606 health_code_update();
1607
1608 /* Atomically prepare the queue futex */
1609 futex_nto1_prepare(&ust_cmd_queue.futex);
1610
1611 do {
1612 struct ust_app *app = NULL;
1613 ust_cmd = NULL;
1614
1615 /*
1616 * Make sure we don't have node(s) that have hung up before receiving
1617 * the notify socket. This is to clean the list in order to avoid
1618 * memory leaks from notify socket that are never seen.
1619 */
1620 sanitize_wait_queue(&wait_queue);
1621
1622 health_code_update();
1623 /* Dequeue command for registration */
1624 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1625 if (node == NULL) {
1626 DBG("Woken up but nothing in the UST command queue");
1627 /* Continue thread execution */
1628 break;
1629 }
1630
1631 ust_cmd = caa_container_of(node, struct ust_command, node);
1632
1633 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1634 " gid:%d sock:%d name:%s (version %d.%d)",
1635 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1636 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1637 ust_cmd->sock, ust_cmd->reg_msg.name,
1638 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1639
1640 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1641 wait_node = zmalloc(sizeof(*wait_node));
1642 if (!wait_node) {
1643 PERROR("zmalloc wait_node dispatch");
1644 ret = close(ust_cmd->sock);
1645 if (ret < 0) {
1646 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1647 }
1648 lttng_fd_put(LTTNG_FD_APPS, 1);
1649 free(ust_cmd);
1650 goto error;
1651 }
1652 CDS_INIT_LIST_HEAD(&wait_node->head);
1653
1654 /* Create application object if socket is CMD. */
1655 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1656 ust_cmd->sock);
1657 if (!wait_node->app) {
1658 ret = close(ust_cmd->sock);
1659 if (ret < 0) {
1660 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1661 }
1662 lttng_fd_put(LTTNG_FD_APPS, 1);
1663 free(wait_node);
1664 free(ust_cmd);
1665 continue;
1666 }
1667 /*
1668 * Add application to the wait queue so we can set the notify
1669 * socket before putting this object in the global ht.
1670 */
1671 cds_list_add(&wait_node->head, &wait_queue.head);
1672 wait_queue.count++;
1673
1674 free(ust_cmd);
1675 /*
1676 * We have to continue here since we don't have the notify
1677 * socket and the application MUST be added to the hash table
1678 * only at that moment.
1679 */
1680 continue;
1681 } else {
1682 /*
1683 * Look for the application in the local wait queue and set the
1684 * notify socket if found.
1685 */
1686 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1687 &wait_queue.head, head) {
1688 health_code_update();
1689 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1690 wait_node->app->notify_sock = ust_cmd->sock;
1691 cds_list_del(&wait_node->head);
1692 wait_queue.count--;
1693 app = wait_node->app;
1694 free(wait_node);
1695 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1696 break;
1697 }
1698 }
1699
1700 /*
1701 * With no application at this stage the received socket is
1702 * basically useless so close it before we free the cmd data
1703 * structure for good.
1704 */
1705 if (!app) {
1706 ret = close(ust_cmd->sock);
1707 if (ret < 0) {
1708 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1709 }
1710 lttng_fd_put(LTTNG_FD_APPS, 1);
1711 }
1712 free(ust_cmd);
1713 }
1714
1715 if (app) {
1716 /*
1717 * @session_lock_list
1718 *
1719 * Lock the global session list so from the register up to the
1720 * registration done message, no thread can see the application
1721 * and change its state.
1722 */
1723 session_lock_list();
1724 rcu_read_lock();
1725
1726 /*
1727 * Add application to the global hash table. This needs to be
1728 * done before the update to the UST registry can locate the
1729 * application.
1730 */
1731 ust_app_add(app);
1732
1733 /* Set app version. This call will print an error if needed. */
1734 (void) ust_app_version(app);
1735
1736 /* Send notify socket through the notify pipe. */
1737 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1738 app->notify_sock);
1739 if (ret < 0) {
1740 rcu_read_unlock();
1741 session_unlock_list();
1742 /*
1743 * No notify thread, stop the UST tracing. However, this is
1744 * not an internal error of the this thread thus setting
1745 * the health error code to a normal exit.
1746 */
1747 err = 0;
1748 goto error;
1749 }
1750
1751 /*
1752 * Update newly registered application with the tracing
1753 * registry info already enabled information.
1754 */
1755 update_ust_app(app->sock);
1756
1757 /*
1758 * Don't care about return value. Let the manage apps threads
1759 * handle app unregistration upon socket close.
1760 */
1761 (void) ust_app_register_done(app->sock);
1762
1763 /*
1764 * Even if the application socket has been closed, send the app
1765 * to the thread and unregistration will take place at that
1766 * place.
1767 */
1768 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
1769 if (ret < 0) {
1770 rcu_read_unlock();
1771 session_unlock_list();
1772 /*
1773 * No apps. thread, stop the UST tracing. However, this is
1774 * not an internal error of the this thread thus setting
1775 * the health error code to a normal exit.
1776 */
1777 err = 0;
1778 goto error;
1779 }
1780
1781 rcu_read_unlock();
1782 session_unlock_list();
1783 }
1784 } while (node != NULL);
1785
1786 health_poll_entry();
1787 /* Futex wait on queue. Blocking call on futex() */
1788 futex_nto1_wait(&ust_cmd_queue.futex);
1789 health_poll_exit();
1790 }
1791 /* Normal exit, no error */
1792 err = 0;
1793
1794 error:
1795 /* Clean up wait queue. */
1796 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1797 &wait_queue.head, head) {
1798 cds_list_del(&wait_node->head);
1799 wait_queue.count--;
1800 free(wait_node);
1801 }
1802
1803 DBG("Dispatch thread dying");
1804 if (err) {
1805 health_error();
1806 ERR("Health error occurred in %s", __func__);
1807 }
1808 health_unregister(health_sessiond);
1809 return NULL;
1810 }
1811
1812 /*
1813 * This thread manage application registration.
1814 */
1815 static void *thread_registration_apps(void *data)
1816 {
1817 int sock = -1, i, ret, pollfd, err = -1;
1818 uint32_t revents, nb_fd;
1819 struct lttng_poll_event events;
1820 /*
1821 * Get allocated in this thread, enqueued to a global queue, dequeued and
1822 * freed in the manage apps thread.
1823 */
1824 struct ust_command *ust_cmd = NULL;
1825
1826 DBG("[thread] Manage application registration started");
1827
1828 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
1829
1830 if (testpoint(thread_registration_apps)) {
1831 goto error_testpoint;
1832 }
1833
1834 ret = lttcomm_listen_unix_sock(apps_sock);
1835 if (ret < 0) {
1836 goto error_listen;
1837 }
1838
1839 /*
1840 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1841 * more will be added to this poll set.
1842 */
1843 ret = sessiond_set_thread_pollset(&events, 2);
1844 if (ret < 0) {
1845 goto error_create_poll;
1846 }
1847
1848 /* Add the application registration socket */
1849 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1850 if (ret < 0) {
1851 goto error_poll_add;
1852 }
1853
1854 /* Notify all applications to register */
1855 ret = notify_ust_apps(1);
1856 if (ret < 0) {
1857 ERR("Failed to notify applications or create the wait shared memory.\n"
1858 "Execution continues but there might be problem for already\n"
1859 "running applications that wishes to register.");
1860 }
1861
1862 while (1) {
1863 DBG("Accepting application registration");
1864
1865 /* Inifinite blocking call, waiting for transmission */
1866 restart:
1867 health_poll_entry();
1868 ret = lttng_poll_wait(&events, -1);
1869 health_poll_exit();
1870 if (ret < 0) {
1871 /*
1872 * Restart interrupted system call.
1873 */
1874 if (errno == EINTR) {
1875 goto restart;
1876 }
1877 goto error;
1878 }
1879
1880 nb_fd = ret;
1881
1882 for (i = 0; i < nb_fd; i++) {
1883 health_code_update();
1884
1885 /* Fetch once the poll data */
1886 revents = LTTNG_POLL_GETEV(&events, i);
1887 pollfd = LTTNG_POLL_GETFD(&events, i);
1888
1889 /* Thread quit pipe has been closed. Killing thread. */
1890 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1891 if (ret) {
1892 err = 0;
1893 goto exit;
1894 }
1895
1896 /* Event on the registration socket */
1897 if (pollfd == apps_sock) {
1898 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1899 ERR("Register apps socket poll error");
1900 goto error;
1901 } else if (revents & LPOLLIN) {
1902 sock = lttcomm_accept_unix_sock(apps_sock);
1903 if (sock < 0) {
1904 goto error;
1905 }
1906
1907 /*
1908 * Set socket timeout for both receiving and ending.
1909 * app_socket_timeout is in seconds, whereas
1910 * lttcomm_setsockopt_rcv_timeout and
1911 * lttcomm_setsockopt_snd_timeout expect msec as
1912 * parameter.
1913 */
1914 (void) lttcomm_setsockopt_rcv_timeout(sock,
1915 app_socket_timeout * 1000);
1916 (void) lttcomm_setsockopt_snd_timeout(sock,
1917 app_socket_timeout * 1000);
1918
1919 /*
1920 * Set the CLOEXEC flag. Return code is useless because
1921 * either way, the show must go on.
1922 */
1923 (void) utils_set_fd_cloexec(sock);
1924
1925 /* Create UST registration command for enqueuing */
1926 ust_cmd = zmalloc(sizeof(struct ust_command));
1927 if (ust_cmd == NULL) {
1928 PERROR("ust command zmalloc");
1929 goto error;
1930 }
1931
1932 /*
1933 * Using message-based transmissions to ensure we don't
1934 * have to deal with partially received messages.
1935 */
1936 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1937 if (ret < 0) {
1938 ERR("Exhausted file descriptors allowed for applications.");
1939 free(ust_cmd);
1940 ret = close(sock);
1941 if (ret) {
1942 PERROR("close");
1943 }
1944 sock = -1;
1945 continue;
1946 }
1947
1948 health_code_update();
1949 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1950 if (ret < 0) {
1951 free(ust_cmd);
1952 /* Close socket of the application. */
1953 ret = close(sock);
1954 if (ret) {
1955 PERROR("close");
1956 }
1957 lttng_fd_put(LTTNG_FD_APPS, 1);
1958 sock = -1;
1959 continue;
1960 }
1961 health_code_update();
1962
1963 ust_cmd->sock = sock;
1964 sock = -1;
1965
1966 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1967 " gid:%d sock:%d name:%s (version %d.%d)",
1968 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1969 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1970 ust_cmd->sock, ust_cmd->reg_msg.name,
1971 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
1972
1973 /*
1974 * Lock free enqueue the registration request. The red pill
1975 * has been taken! This apps will be part of the *system*.
1976 */
1977 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1978
1979 /*
1980 * Wake the registration queue futex. Implicit memory
1981 * barrier with the exchange in cds_wfq_enqueue.
1982 */
1983 futex_nto1_wake(&ust_cmd_queue.futex);
1984 }
1985 }
1986 }
1987 }
1988
1989 exit:
1990 error:
1991 if (err) {
1992 health_error();
1993 ERR("Health error occurred in %s", __func__);
1994 }
1995
1996 /* Notify that the registration thread is gone */
1997 notify_ust_apps(0);
1998
1999 if (apps_sock >= 0) {
2000 ret = close(apps_sock);
2001 if (ret) {
2002 PERROR("close");
2003 }
2004 }
2005 if (sock >= 0) {
2006 ret = close(sock);
2007 if (ret) {
2008 PERROR("close");
2009 }
2010 lttng_fd_put(LTTNG_FD_APPS, 1);
2011 }
2012 unlink(apps_unix_sock_path);
2013
2014 error_poll_add:
2015 lttng_poll_clean(&events);
2016 error_listen:
2017 error_create_poll:
2018 error_testpoint:
2019 DBG("UST Registration thread cleanup complete");
2020 health_unregister(health_sessiond);
2021
2022 return NULL;
2023 }
2024
2025 /*
2026 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2027 * exec or it will fails.
2028 */
2029 static int spawn_consumer_thread(struct consumer_data *consumer_data)
2030 {
2031 int ret, clock_ret;
2032 struct timespec timeout;
2033
2034 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2035 consumer_data->consumer_thread_is_ready = 0;
2036
2037 /* Setup pthread condition */
2038 ret = pthread_condattr_init(&consumer_data->condattr);
2039 if (ret != 0) {
2040 errno = ret;
2041 PERROR("pthread_condattr_init consumer data");
2042 goto error;
2043 }
2044
2045 /*
2046 * Set the monotonic clock in order to make sure we DO NOT jump in time
2047 * between the clock_gettime() call and the timedwait call. See bug #324
2048 * for a more details and how we noticed it.
2049 */
2050 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2051 if (ret != 0) {
2052 errno = ret;
2053 PERROR("pthread_condattr_setclock consumer data");
2054 goto error;
2055 }
2056
2057 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2058 if (ret != 0) {
2059 errno = ret;
2060 PERROR("pthread_cond_init consumer data");
2061 goto error;
2062 }
2063
2064 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2065 consumer_data);
2066 if (ret != 0) {
2067 PERROR("pthread_create consumer");
2068 ret = -1;
2069 goto error;
2070 }
2071
2072 /* We are about to wait on a pthread condition */
2073 pthread_mutex_lock(&consumer_data->cond_mutex);
2074
2075 /* Get time for sem_timedwait absolute timeout */
2076 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2077 /*
2078 * Set the timeout for the condition timed wait even if the clock gettime
2079 * call fails since we might loop on that call and we want to avoid to
2080 * increment the timeout too many times.
2081 */
2082 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2083
2084 /*
2085 * The following loop COULD be skipped in some conditions so this is why we
2086 * set ret to 0 in order to make sure at least one round of the loop is
2087 * done.
2088 */
2089 ret = 0;
2090
2091 /*
2092 * Loop until the condition is reached or when a timeout is reached. Note
2093 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2094 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2095 * possible. This loop does not take any chances and works with both of
2096 * them.
2097 */
2098 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2099 if (clock_ret < 0) {
2100 PERROR("clock_gettime spawn consumer");
2101 /* Infinite wait for the consumerd thread to be ready */
2102 ret = pthread_cond_wait(&consumer_data->cond,
2103 &consumer_data->cond_mutex);
2104 } else {
2105 ret = pthread_cond_timedwait(&consumer_data->cond,
2106 &consumer_data->cond_mutex, &timeout);
2107 }
2108 }
2109
2110 /* Release the pthread condition */
2111 pthread_mutex_unlock(&consumer_data->cond_mutex);
2112
2113 if (ret != 0) {
2114 errno = ret;
2115 if (ret == ETIMEDOUT) {
2116 /*
2117 * Call has timed out so we kill the kconsumerd_thread and return
2118 * an error.
2119 */
2120 ERR("Condition timed out. The consumer thread was never ready."
2121 " Killing it");
2122 ret = pthread_cancel(consumer_data->thread);
2123 if (ret < 0) {
2124 PERROR("pthread_cancel consumer thread");
2125 }
2126 } else {
2127 PERROR("pthread_cond_wait failed consumer thread");
2128 }
2129 goto error;
2130 }
2131
2132 pthread_mutex_lock(&consumer_data->pid_mutex);
2133 if (consumer_data->pid == 0) {
2134 ERR("Consumerd did not start");
2135 pthread_mutex_unlock(&consumer_data->pid_mutex);
2136 goto error;
2137 }
2138 pthread_mutex_unlock(&consumer_data->pid_mutex);
2139
2140 return 0;
2141
2142 error:
2143 return ret;
2144 }
2145
2146 /*
2147 * Join consumer thread
2148 */
2149 static int join_consumer_thread(struct consumer_data *consumer_data)
2150 {
2151 void *status;
2152
2153 /* Consumer pid must be a real one. */
2154 if (consumer_data->pid > 0) {
2155 int ret;
2156 ret = kill(consumer_data->pid, SIGTERM);
2157 if (ret) {
2158 ERR("Error killing consumer daemon");
2159 return ret;
2160 }
2161 return pthread_join(consumer_data->thread, &status);
2162 } else {
2163 return 0;
2164 }
2165 }
2166
2167 /*
2168 * Fork and exec a consumer daemon (consumerd).
2169 *
2170 * Return pid if successful else -1.
2171 */
2172 static pid_t spawn_consumerd(struct consumer_data *consumer_data)
2173 {
2174 int ret;
2175 pid_t pid;
2176 const char *consumer_to_use;
2177 const char *verbosity;
2178 struct stat st;
2179
2180 DBG("Spawning consumerd");
2181
2182 pid = fork();
2183 if (pid == 0) {
2184 /*
2185 * Exec consumerd.
2186 */
2187 if (opt_verbose_consumer) {
2188 verbosity = "--verbose";
2189 } else {
2190 verbosity = "--quiet";
2191 }
2192 switch (consumer_data->type) {
2193 case LTTNG_CONSUMER_KERNEL:
2194 /*
2195 * Find out which consumerd to execute. We will first try the
2196 * 64-bit path, then the sessiond's installation directory, and
2197 * fallback on the 32-bit one,
2198 */
2199 DBG3("Looking for a kernel consumer at these locations:");
2200 DBG3(" 1) %s", consumerd64_bin);
2201 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2202 DBG3(" 3) %s", consumerd32_bin);
2203 if (stat(consumerd64_bin, &st) == 0) {
2204 DBG3("Found location #1");
2205 consumer_to_use = consumerd64_bin;
2206 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
2207 DBG3("Found location #2");
2208 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
2209 } else if (stat(consumerd32_bin, &st) == 0) {
2210 DBG3("Found location #3");
2211 consumer_to_use = consumerd32_bin;
2212 } else {
2213 DBG("Could not find any valid consumerd executable");
2214 break;
2215 }
2216 DBG("Using kernel consumer at: %s", consumer_to_use);
2217 execl(consumer_to_use,
2218 "lttng-consumerd", verbosity, "-k",
2219 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2220 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2221 "--group", tracing_group_name,
2222 NULL);
2223 break;
2224 case LTTNG_CONSUMER64_UST:
2225 {
2226 char *tmpnew = NULL;
2227
2228 if (consumerd64_libdir[0] != '\0') {
2229 char *tmp;
2230 size_t tmplen;
2231
2232 tmp = getenv("LD_LIBRARY_PATH");
2233 if (!tmp) {
2234 tmp = "";
2235 }
2236 tmplen = strlen("LD_LIBRARY_PATH=")
2237 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2238 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2239 if (!tmpnew) {
2240 ret = -ENOMEM;
2241 goto error;
2242 }
2243 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2244 strcat(tmpnew, consumerd64_libdir);
2245 if (tmp[0] != '\0') {
2246 strcat(tmpnew, ":");
2247 strcat(tmpnew, tmp);
2248 }
2249 ret = putenv(tmpnew);
2250 if (ret) {
2251 ret = -errno;
2252 free(tmpnew);
2253 goto error;
2254 }
2255 }
2256 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
2257 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
2258 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2259 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2260 "--group", tracing_group_name,
2261 NULL);
2262 if (consumerd64_libdir[0] != '\0') {
2263 free(tmpnew);
2264 }
2265 if (ret) {
2266 goto error;
2267 }
2268 break;
2269 }
2270 case LTTNG_CONSUMER32_UST:
2271 {
2272 char *tmpnew = NULL;
2273
2274 if (consumerd32_libdir[0] != '\0') {
2275 char *tmp;
2276 size_t tmplen;
2277
2278 tmp = getenv("LD_LIBRARY_PATH");
2279 if (!tmp) {
2280 tmp = "";
2281 }
2282 tmplen = strlen("LD_LIBRARY_PATH=")
2283 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2284 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2285 if (!tmpnew) {
2286 ret = -ENOMEM;
2287 goto error;
2288 }
2289 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2290 strcat(tmpnew, consumerd32_libdir);
2291 if (tmp[0] != '\0') {
2292 strcat(tmpnew, ":");
2293 strcat(tmpnew, tmp);
2294 }
2295 ret = putenv(tmpnew);
2296 if (ret) {
2297 ret = -errno;
2298 free(tmpnew);
2299 goto error;
2300 }
2301 }
2302 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
2303 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
2304 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2305 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2306 "--group", tracing_group_name,
2307 NULL);
2308 if (consumerd32_libdir[0] != '\0') {
2309 free(tmpnew);
2310 }
2311 if (ret) {
2312 goto error;
2313 }
2314 break;
2315 }
2316 default:
2317 PERROR("unknown consumer type");
2318 exit(EXIT_FAILURE);
2319 }
2320 if (errno != 0) {
2321 PERROR("kernel start consumer exec");
2322 }
2323 exit(EXIT_FAILURE);
2324 } else if (pid > 0) {
2325 ret = pid;
2326 } else {
2327 PERROR("start consumer fork");
2328 ret = -errno;
2329 }
2330 error:
2331 return ret;
2332 }
2333
2334 /*
2335 * Spawn the consumerd daemon and session daemon thread.
2336 */
2337 static int start_consumerd(struct consumer_data *consumer_data)
2338 {
2339 int ret;
2340
2341 /*
2342 * Set the listen() state on the socket since there is a possible race
2343 * between the exec() of the consumer daemon and this call if place in the
2344 * consumer thread. See bug #366 for more details.
2345 */
2346 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2347 if (ret < 0) {
2348 goto error;
2349 }
2350
2351 pthread_mutex_lock(&consumer_data->pid_mutex);
2352 if (consumer_data->pid != 0) {
2353 pthread_mutex_unlock(&consumer_data->pid_mutex);
2354 goto end;
2355 }
2356
2357 ret = spawn_consumerd(consumer_data);
2358 if (ret < 0) {
2359 ERR("Spawning consumerd failed");
2360 pthread_mutex_unlock(&consumer_data->pid_mutex);
2361 goto error;
2362 }
2363
2364 /* Setting up the consumer_data pid */
2365 consumer_data->pid = ret;
2366 DBG2("Consumer pid %d", consumer_data->pid);
2367 pthread_mutex_unlock(&consumer_data->pid_mutex);
2368
2369 DBG2("Spawning consumer control thread");
2370 ret = spawn_consumer_thread(consumer_data);
2371 if (ret < 0) {
2372 ERR("Fatal error spawning consumer control thread");
2373 goto error;
2374 }
2375
2376 end:
2377 return 0;
2378
2379 error:
2380 /* Cleanup already created sockets on error. */
2381 if (consumer_data->err_sock >= 0) {
2382 int err;
2383
2384 err = close(consumer_data->err_sock);
2385 if (err < 0) {
2386 PERROR("close consumer data error socket");
2387 }
2388 }
2389 return ret;
2390 }
2391
2392 /*
2393 * Setup necessary data for kernel tracer action.
2394 */
2395 static int init_kernel_tracer(void)
2396 {
2397 int ret;
2398
2399 /* Modprobe lttng kernel modules */
2400 ret = modprobe_lttng_control();
2401 if (ret < 0) {
2402 goto error;
2403 }
2404
2405 /* Open debugfs lttng */
2406 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2407 if (kernel_tracer_fd < 0) {
2408 DBG("Failed to open %s", module_proc_lttng);
2409 ret = -1;
2410 goto error_open;
2411 }
2412
2413 /* Validate kernel version */
2414 ret = kernel_validate_version(kernel_tracer_fd);
2415 if (ret < 0) {
2416 goto error_version;
2417 }
2418
2419 ret = modprobe_lttng_data();
2420 if (ret < 0) {
2421 goto error_modules;
2422 }
2423
2424 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2425 return 0;
2426
2427 error_version:
2428 modprobe_remove_lttng_control();
2429 ret = close(kernel_tracer_fd);
2430 if (ret) {
2431 PERROR("close");
2432 }
2433 kernel_tracer_fd = -1;
2434 return LTTNG_ERR_KERN_VERSION;
2435
2436 error_modules:
2437 ret = close(kernel_tracer_fd);
2438 if (ret) {
2439 PERROR("close");
2440 }
2441
2442 error_open:
2443 modprobe_remove_lttng_control();
2444
2445 error:
2446 WARN("No kernel tracer available");
2447 kernel_tracer_fd = -1;
2448 if (!is_root) {
2449 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2450 } else {
2451 return LTTNG_ERR_KERN_NA;
2452 }
2453 }
2454
2455
2456 /*
2457 * Copy consumer output from the tracing session to the domain session. The
2458 * function also applies the right modification on a per domain basis for the
2459 * trace files destination directory.
2460 *
2461 * Should *NOT* be called with RCU read-side lock held.
2462 */
2463 static int copy_session_consumer(int domain, struct ltt_session *session)
2464 {
2465 int ret;
2466 const char *dir_name;
2467 struct consumer_output *consumer;
2468
2469 assert(session);
2470 assert(session->consumer);
2471
2472 switch (domain) {
2473 case LTTNG_DOMAIN_KERNEL:
2474 DBG3("Copying tracing session consumer output in kernel session");
2475 /*
2476 * XXX: We should audit the session creation and what this function
2477 * does "extra" in order to avoid a destroy since this function is used
2478 * in the domain session creation (kernel and ust) only. Same for UST
2479 * domain.
2480 */
2481 if (session->kernel_session->consumer) {
2482 consumer_destroy_output(session->kernel_session->consumer);
2483 }
2484 session->kernel_session->consumer =
2485 consumer_copy_output(session->consumer);
2486 /* Ease our life a bit for the next part */
2487 consumer = session->kernel_session->consumer;
2488 dir_name = DEFAULT_KERNEL_TRACE_DIR;
2489 break;
2490 case LTTNG_DOMAIN_JUL:
2491 case LTTNG_DOMAIN_UST:
2492 DBG3("Copying tracing session consumer output in UST session");
2493 if (session->ust_session->consumer) {
2494 consumer_destroy_output(session->ust_session->consumer);
2495 }
2496 session->ust_session->consumer =
2497 consumer_copy_output(session->consumer);
2498 /* Ease our life a bit for the next part */
2499 consumer = session->ust_session->consumer;
2500 dir_name = DEFAULT_UST_TRACE_DIR;
2501 break;
2502 default:
2503 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2504 goto error;
2505 }
2506
2507 /* Append correct directory to subdir */
2508 strncat(consumer->subdir, dir_name,
2509 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2510 DBG3("Copy session consumer subdir %s", consumer->subdir);
2511
2512 ret = LTTNG_OK;
2513
2514 error:
2515 return ret;
2516 }
2517
2518 /*
2519 * Create an UST session and add it to the session ust list.
2520 *
2521 * Should *NOT* be called with RCU read-side lock held.
2522 */
2523 static int create_ust_session(struct ltt_session *session,
2524 struct lttng_domain *domain)
2525 {
2526 int ret;
2527 struct ltt_ust_session *lus = NULL;
2528
2529 assert(session);
2530 assert(domain);
2531 assert(session->consumer);
2532
2533 switch (domain->type) {
2534 case LTTNG_DOMAIN_JUL:
2535 case LTTNG_DOMAIN_UST:
2536 break;
2537 default:
2538 ERR("Unknown UST domain on create session %d", domain->type);
2539 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
2540 goto error;
2541 }
2542
2543 DBG("Creating UST session");
2544
2545 lus = trace_ust_create_session(session->id);
2546 if (lus == NULL) {
2547 ret = LTTNG_ERR_UST_SESS_FAIL;
2548 goto error;
2549 }
2550
2551 lus->uid = session->uid;
2552 lus->gid = session->gid;
2553 lus->output_traces = session->output_traces;
2554 lus->snapshot_mode = session->snapshot_mode;
2555 lus->live_timer_interval = session->live_timer;
2556 session->ust_session = lus;
2557
2558 /* Copy session output to the newly created UST session */
2559 ret = copy_session_consumer(domain->type, session);
2560 if (ret != LTTNG_OK) {
2561 goto error;
2562 }
2563
2564 return LTTNG_OK;
2565
2566 error:
2567 free(lus);
2568 session->ust_session = NULL;
2569 return ret;
2570 }
2571
2572 /*
2573 * Create a kernel tracer session then create the default channel.
2574 */
2575 static int create_kernel_session(struct ltt_session *session)
2576 {
2577 int ret;
2578
2579 DBG("Creating kernel session");
2580
2581 ret = kernel_create_session(session, kernel_tracer_fd);
2582 if (ret < 0) {
2583 ret = LTTNG_ERR_KERN_SESS_FAIL;
2584 goto error;
2585 }
2586
2587 /* Code flow safety */
2588 assert(session->kernel_session);
2589
2590 /* Copy session output to the newly created Kernel session */
2591 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
2592 if (ret != LTTNG_OK) {
2593 goto error;
2594 }
2595
2596 /* Create directory(ies) on local filesystem. */
2597 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2598 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2599 ret = run_as_mkdir_recursive(
2600 session->kernel_session->consumer->dst.trace_path,
2601 S_IRWXU | S_IRWXG, session->uid, session->gid);
2602 if (ret < 0) {
2603 if (ret != -EEXIST) {
2604 ERR("Trace directory creation error");
2605 goto error;
2606 }
2607 }
2608 }
2609
2610 session->kernel_session->uid = session->uid;
2611 session->kernel_session->gid = session->gid;
2612 session->kernel_session->output_traces = session->output_traces;
2613 session->kernel_session->snapshot_mode = session->snapshot_mode;
2614
2615 return LTTNG_OK;
2616
2617 error:
2618 trace_kernel_destroy_session(session->kernel_session);
2619 session->kernel_session = NULL;
2620 return ret;
2621 }
2622
2623 /*
2624 * Count number of session permitted by uid/gid.
2625 */
2626 static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2627 {
2628 unsigned int i = 0;
2629 struct ltt_session *session;
2630
2631 DBG("Counting number of available session for UID %d GID %d",
2632 uid, gid);
2633 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
2634 /*
2635 * Only list the sessions the user can control.
2636 */
2637 if (!session_access_ok(session, uid, gid)) {
2638 continue;
2639 }
2640 i++;
2641 }
2642 return i;
2643 }
2644
2645 /*
2646 * Process the command requested by the lttng client within the command
2647 * context structure. This function make sure that the return structure (llm)
2648 * is set and ready for transmission before returning.
2649 *
2650 * Return any error encountered or 0 for success.
2651 *
2652 * "sock" is only used for special-case var. len data.
2653 *
2654 * Should *NOT* be called with RCU read-side lock held.
2655 */
2656 static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2657 int *sock_error)
2658 {
2659 int ret = LTTNG_OK;
2660 int need_tracing_session = 1;
2661 int need_domain;
2662
2663 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2664
2665 *sock_error = 0;
2666
2667 switch (cmd_ctx->lsm->cmd_type) {
2668 case LTTNG_CREATE_SESSION:
2669 case LTTNG_CREATE_SESSION_SNAPSHOT:
2670 case LTTNG_CREATE_SESSION_LIVE:
2671 case LTTNG_DESTROY_SESSION:
2672 case LTTNG_LIST_SESSIONS:
2673 case LTTNG_LIST_DOMAINS:
2674 case LTTNG_START_TRACE:
2675 case LTTNG_STOP_TRACE:
2676 case LTTNG_DATA_PENDING:
2677 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2678 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2679 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2680 case LTTNG_SNAPSHOT_RECORD:
2681 need_domain = 0;
2682 break;
2683 default:
2684 need_domain = 1;
2685 }
2686
2687 if (opt_no_kernel && need_domain
2688 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
2689 if (!is_root) {
2690 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2691 } else {
2692 ret = LTTNG_ERR_KERN_NA;
2693 }
2694 goto error;
2695 }
2696
2697 /* Deny register consumer if we already have a spawned consumer. */
2698 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2699 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2700 if (kconsumer_data.pid > 0) {
2701 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2702 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2703 goto error;
2704 }
2705 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2706 }
2707
2708 /*
2709 * Check for command that don't needs to allocate a returned payload. We do
2710 * this here so we don't have to make the call for no payload at each
2711 * command.
2712 */
2713 switch(cmd_ctx->lsm->cmd_type) {
2714 case LTTNG_LIST_SESSIONS:
2715 case LTTNG_LIST_TRACEPOINTS:
2716 case LTTNG_LIST_TRACEPOINT_FIELDS:
2717 case LTTNG_LIST_DOMAINS:
2718 case LTTNG_LIST_CHANNELS:
2719 case LTTNG_LIST_EVENTS:
2720 break;
2721 default:
2722 /* Setup lttng message with no payload */
2723 ret = setup_lttng_msg(cmd_ctx, 0);
2724 if (ret < 0) {
2725 /* This label does not try to unlock the session */
2726 goto init_setup_error;
2727 }
2728 }
2729
2730 /* Commands that DO NOT need a session. */
2731 switch (cmd_ctx->lsm->cmd_type) {
2732 case LTTNG_CREATE_SESSION:
2733 case LTTNG_CREATE_SESSION_SNAPSHOT:
2734 case LTTNG_CREATE_SESSION_LIVE:
2735 case LTTNG_CALIBRATE:
2736 case LTTNG_LIST_SESSIONS:
2737 case LTTNG_LIST_TRACEPOINTS:
2738 case LTTNG_LIST_TRACEPOINT_FIELDS:
2739 need_tracing_session = 0;
2740 break;
2741 default:
2742 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
2743 /*
2744 * We keep the session list lock across _all_ commands
2745 * for now, because the per-session lock does not
2746 * handle teardown properly.
2747 */
2748 session_lock_list();
2749 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2750 if (cmd_ctx->session == NULL) {
2751 ret = LTTNG_ERR_SESS_NOT_FOUND;
2752 goto error;
2753 } else {
2754 /* Acquire lock for the session */
2755 session_lock(cmd_ctx->session);
2756 }
2757 break;
2758 }
2759
2760 if (!need_domain) {
2761 goto skip_domain;
2762 }
2763
2764 /*
2765 * Check domain type for specific "pre-action".
2766 */
2767 switch (cmd_ctx->lsm->domain.type) {
2768 case LTTNG_DOMAIN_KERNEL:
2769 if (!is_root) {
2770 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
2771 goto error;
2772 }
2773
2774 /* Kernel tracer check */
2775 if (kernel_tracer_fd == -1) {
2776 /* Basically, load kernel tracer modules */
2777 ret = init_kernel_tracer();
2778 if (ret != 0) {
2779 goto error;
2780 }
2781 }
2782
2783 /* Consumer is in an ERROR state. Report back to client */
2784 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
2785 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2786 goto error;
2787 }
2788
2789 /* Need a session for kernel command */
2790 if (need_tracing_session) {
2791 if (cmd_ctx->session->kernel_session == NULL) {
2792 ret = create_kernel_session(cmd_ctx->session);
2793 if (ret < 0) {
2794 ret = LTTNG_ERR_KERN_SESS_FAIL;
2795 goto error;
2796 }
2797 }
2798
2799 /* Start the kernel consumer daemon */
2800 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2801 if (kconsumer_data.pid == 0 &&
2802 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2803 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2804 ret = start_consumerd(&kconsumer_data);
2805 if (ret < 0) {
2806 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
2807 goto error;
2808 }
2809 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
2810 } else {
2811 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2812 }
2813
2814 /*
2815 * The consumer was just spawned so we need to add the socket to
2816 * the consumer output of the session if exist.
2817 */
2818 ret = consumer_create_socket(&kconsumer_data,
2819 cmd_ctx->session->kernel_session->consumer);
2820 if (ret < 0) {
2821 goto error;
2822 }
2823 }
2824
2825 break;
2826 case LTTNG_DOMAIN_JUL:
2827 case LTTNG_DOMAIN_UST:
2828 {
2829 if (!ust_app_supported()) {
2830 ret = LTTNG_ERR_NO_UST;
2831 goto error;
2832 }
2833 /* Consumer is in an ERROR state. Report back to client */
2834 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
2835 ret = LTTNG_ERR_NO_USTCONSUMERD;
2836 goto error;
2837 }
2838
2839 if (need_tracing_session) {
2840 /* Create UST session if none exist. */
2841 if (cmd_ctx->session->ust_session == NULL) {
2842 ret = create_ust_session(cmd_ctx->session,
2843 &cmd_ctx->lsm->domain);
2844 if (ret != LTTNG_OK) {
2845 goto error;
2846 }
2847 }
2848
2849 /* Start the UST consumer daemons */
2850 /* 64-bit */
2851 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
2852 if (consumerd64_bin[0] != '\0' &&
2853 ustconsumer64_data.pid == 0 &&
2854 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2855 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2856 ret = start_consumerd(&ustconsumer64_data);
2857 if (ret < 0) {
2858 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
2859 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2860 goto error;
2861 }
2862
2863 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
2864 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2865 } else {
2866 pthread_mutex_unlock(&ustconsumer64_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(&ustconsumer64_data,
2874 cmd_ctx->session->ust_session->consumer);
2875 if (ret < 0) {
2876 goto error;
2877 }
2878
2879 /* 32-bit */
2880 if (consumerd32_bin[0] != '\0' &&
2881 ustconsumer32_data.pid == 0 &&
2882 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
2883 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2884 ret = start_consumerd(&ustconsumer32_data);
2885 if (ret < 0) {
2886 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
2887 uatomic_set(&ust_consumerd32_fd, -EINVAL);
2888 goto error;
2889 }
2890
2891 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
2892 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
2893 } else {
2894 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2895 }
2896
2897 /*
2898 * Setup socket for consumer 64 bit. No need for atomic access
2899 * since it was set above and can ONLY be set in this thread.
2900 */
2901 ret = consumer_create_socket(&ustconsumer32_data,
2902 cmd_ctx->session->ust_session->consumer);
2903 if (ret < 0) {
2904 goto error;
2905 }
2906 }
2907 break;
2908 }
2909 default:
2910 break;
2911 }
2912 skip_domain:
2913
2914 /* Validate consumer daemon state when start/stop trace command */
2915 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2916 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2917 switch (cmd_ctx->lsm->domain.type) {
2918 case LTTNG_DOMAIN_JUL:
2919 case LTTNG_DOMAIN_UST:
2920 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
2921 ret = LTTNG_ERR_NO_USTCONSUMERD;
2922 goto error;
2923 }
2924 break;
2925 case LTTNG_DOMAIN_KERNEL:
2926 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
2927 ret = LTTNG_ERR_NO_KERNCONSUMERD;
2928 goto error;
2929 }
2930 break;
2931 }
2932 }
2933
2934 /*
2935 * Check that the UID or GID match that of the tracing session.
2936 * The root user can interact with all sessions.
2937 */
2938 if (need_tracing_session) {
2939 if (!session_access_ok(cmd_ctx->session,
2940 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2941 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
2942 ret = LTTNG_ERR_EPERM;
2943 goto error;
2944 }
2945 }
2946
2947 /*
2948 * Send relayd information to consumer as soon as we have a domain and a
2949 * session defined.
2950 */
2951 if (cmd_ctx->session && need_domain) {
2952 /*
2953 * Setup relayd if not done yet. If the relayd information was already
2954 * sent to the consumer, this call will gracefully return.
2955 */
2956 ret = cmd_setup_relayd(cmd_ctx->session);
2957 if (ret != LTTNG_OK) {
2958 goto error;
2959 }
2960 }
2961
2962 /* Process by command type */
2963 switch (cmd_ctx->lsm->cmd_type) {
2964 case LTTNG_ADD_CONTEXT:
2965 {
2966 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2967 cmd_ctx->lsm->u.context.channel_name,
2968 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
2969 break;
2970 }
2971 case LTTNG_DISABLE_CHANNEL:
2972 {
2973 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2974 cmd_ctx->lsm->u.disable.channel_name);
2975 break;
2976 }
2977 case LTTNG_DISABLE_EVENT:
2978 {
2979 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2980 cmd_ctx->lsm->u.disable.channel_name,
2981 cmd_ctx->lsm->u.disable.name);
2982 break;
2983 }
2984 case LTTNG_DISABLE_ALL_EVENT:
2985 {
2986 DBG("Disabling all events");
2987
2988 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2989 cmd_ctx->lsm->u.disable.channel_name);
2990 break;
2991 }
2992 case LTTNG_ENABLE_CHANNEL:
2993 {
2994 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2995 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
2996 break;
2997 }
2998 case LTTNG_ENABLE_EVENT:
2999 {
3000 struct lttng_event_exclusion *exclusion = NULL;
3001 struct lttng_filter_bytecode *bytecode = NULL;
3002
3003 /* Handle exclusion events and receive it from the client. */
3004 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3005 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3006
3007 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3008 (count * LTTNG_SYMBOL_NAME_LEN));
3009 if (!exclusion) {
3010 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3011 goto error;
3012 }
3013
3014 DBG("Receiving var len exclusion event list from client ...");
3015 exclusion->count = count;
3016 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3017 count * LTTNG_SYMBOL_NAME_LEN);
3018 if (ret <= 0) {
3019 DBG("Nothing recv() from client var len data... continuing");
3020 *sock_error = 1;
3021 free(exclusion);
3022 ret = LTTNG_ERR_EXCLUSION_INVAL;
3023 goto error;
3024 }
3025 }
3026
3027 /* Handle filter and get bytecode from client. */
3028 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3029 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3030
3031 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3032 ret = LTTNG_ERR_FILTER_INVAL;
3033 free(exclusion);
3034 goto error;
3035 }
3036
3037 bytecode = zmalloc(bytecode_len);
3038 if (!bytecode) {
3039 free(exclusion);
3040 ret = LTTNG_ERR_FILTER_NOMEM;
3041 goto error;
3042 }
3043
3044 /* Receive var. len. data */
3045 DBG("Receiving var len filter's bytecode from client ...");
3046 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3047 if (ret <= 0) {
3048 DBG("Nothing recv() from client car len data... continuing");
3049 *sock_error = 1;
3050 free(bytecode);
3051 free(exclusion);
3052 ret = LTTNG_ERR_FILTER_INVAL;
3053 goto error;
3054 }
3055
3056 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3057 free(bytecode);
3058 free(exclusion);
3059 ret = LTTNG_ERR_FILTER_INVAL;
3060 goto error;
3061 }
3062 }
3063
3064 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3065 cmd_ctx->lsm->u.enable.channel_name,
3066 &cmd_ctx->lsm->u.enable.event, bytecode, exclusion,
3067 kernel_poll_pipe[1]);
3068 break;
3069 }
3070 case LTTNG_ENABLE_ALL_EVENT:
3071 {
3072 DBG("Enabling all events");
3073
3074 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
3075 cmd_ctx->lsm->u.enable.channel_name,
3076 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
3077 break;
3078 }
3079 case LTTNG_LIST_TRACEPOINTS:
3080 {
3081 struct lttng_event *events;
3082 ssize_t nb_events;
3083
3084 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3085 if (nb_events < 0) {
3086 /* Return value is a negative lttng_error_code. */
3087 ret = -nb_events;
3088 goto error;
3089 }
3090
3091 /*
3092 * Setup lttng message with payload size set to the event list size in
3093 * bytes and then copy list into the llm payload.
3094 */
3095 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
3096 if (ret < 0) {
3097 free(events);
3098 goto setup_error;
3099 }
3100
3101 /* Copy event list into message payload */
3102 memcpy(cmd_ctx->llm->payload, events,
3103 sizeof(struct lttng_event) * nb_events);
3104
3105 free(events);
3106
3107 ret = LTTNG_OK;
3108 break;
3109 }
3110 case LTTNG_LIST_TRACEPOINT_FIELDS:
3111 {
3112 struct lttng_event_field *fields;
3113 ssize_t nb_fields;
3114
3115 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3116 &fields);
3117 if (nb_fields < 0) {
3118 /* Return value is a negative lttng_error_code. */
3119 ret = -nb_fields;
3120 goto error;
3121 }
3122
3123 /*
3124 * Setup lttng message with payload size set to the event list size in
3125 * bytes and then copy list into the llm payload.
3126 */
3127 ret = setup_lttng_msg(cmd_ctx,
3128 sizeof(struct lttng_event_field) * nb_fields);
3129 if (ret < 0) {
3130 free(fields);
3131 goto setup_error;
3132 }
3133
3134 /* Copy event list into message payload */
3135 memcpy(cmd_ctx->llm->payload, fields,
3136 sizeof(struct lttng_event_field) * nb_fields);
3137
3138 free(fields);
3139
3140 ret = LTTNG_OK;
3141 break;
3142 }
3143 case LTTNG_SET_CONSUMER_URI:
3144 {
3145 size_t nb_uri, len;
3146 struct lttng_uri *uris;
3147
3148 nb_uri = cmd_ctx->lsm->u.uri.size;
3149 len = nb_uri * sizeof(struct lttng_uri);
3150
3151 if (nb_uri == 0) {
3152 ret = LTTNG_ERR_INVALID;
3153 goto error;
3154 }
3155
3156 uris = zmalloc(len);
3157 if (uris == NULL) {
3158 ret = LTTNG_ERR_FATAL;
3159 goto error;
3160 }
3161
3162 /* Receive variable len data */
3163 DBG("Receiving %zu URI(s) from client ...", nb_uri);
3164 ret = lttcomm_recv_unix_sock(sock, uris, len);
3165 if (ret <= 0) {
3166 DBG("No URIs received from client... continuing");
3167 *sock_error = 1;
3168 ret = LTTNG_ERR_SESSION_FAIL;
3169 free(uris);
3170 goto error;
3171 }
3172
3173 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
3174 nb_uri, uris);
3175 if (ret != LTTNG_OK) {
3176 free(uris);
3177 goto error;
3178 }
3179
3180 /*
3181 * XXX: 0 means that this URI should be applied on the session. Should
3182 * be a DOMAIN enuam.
3183 */
3184 if (cmd_ctx->lsm->domain.type == 0) {
3185 /* Add the URI for the UST session if a consumer is present. */
3186 if (cmd_ctx->session->ust_session &&
3187 cmd_ctx->session->ust_session->consumer) {
3188 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3189 nb_uri, uris);
3190 } else if (cmd_ctx->session->kernel_session &&
3191 cmd_ctx->session->kernel_session->consumer) {
3192 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3193 cmd_ctx->session, nb_uri, uris);
3194 }
3195 }
3196
3197 free(uris);
3198
3199 break;
3200 }
3201 case LTTNG_START_TRACE:
3202 {
3203 ret = cmd_start_trace(cmd_ctx->session);
3204 break;
3205 }
3206 case LTTNG_STOP_TRACE:
3207 {
3208 ret = cmd_stop_trace(cmd_ctx->session);
3209 break;
3210 }
3211 case LTTNG_CREATE_SESSION:
3212 {
3213 size_t nb_uri, len;
3214 struct lttng_uri *uris = NULL;
3215
3216 nb_uri = cmd_ctx->lsm->u.uri.size;
3217 len = nb_uri * sizeof(struct lttng_uri);
3218