Fix: remove unused code in session daemon
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
CommitLineData
826d496d
MD
1/*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
0fdd1e2c 3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
fac6795d 4 *
d14d33bf
AM
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
91d76f53 8 *
d14d33bf
AM
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
91d76f53 13 *
d14d33bf
AM
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
fac6795d
DG
17 */
18
19#define _GNU_SOURCE
fac6795d
DG
20#include <getopt.h>
21#include <grp.h>
22#include <limits.h>
0bb7724a 23#include <paths.h>
fac6795d
DG
24#include <pthread.h>
25#include <signal.h>
26#include <stdio.h>
27#include <stdlib.h>
28#include <string.h>
331744e3 29#include <inttypes.h>
0fdd1e2c 30#include <sys/mman.h>
b73401da 31#include <sys/mount.h>
1e307fab 32#include <sys/resource.h>
fac6795d
DG
33#include <sys/socket.h>
34#include <sys/stat.h>
35#include <sys/types.h>
0fdd1e2c 36#include <sys/wait.h>
5c827ce0 37#include <urcu/uatomic.h>
fac6795d 38#include <unistd.h>
3bd1e081 39#include <config.h>
fac6795d 40
990570ed 41#include <common/common.h>
d27c42b8 42#include <common/compat/socket.h>
db758600
DG
43#include <common/defaults.h>
44#include <common/kernel-consumer/kernel-consumer.h>
50c8f484 45#include <common/futex.h>
00e2e675 46#include <common/relayd/relayd.h>
81b86775 47#include <common/utils.h>
fac6795d 48
10a8a223 49#include "lttng-sessiond.h"
7972aab2 50#include "buffer-registry.h"
54d01ffb 51#include "channel.h"
2f77fc4b 52#include "cmd.h"
00e2e675 53#include "consumer.h"
099e26bd 54#include "context.h"
54d01ffb 55#include "event.h"
4771f025 56#include "kernel.h"
f1e16794 57#include "kernel-consumer.h"
096102bd 58#include "modprobe.h"
0fdd1e2c 59#include "shm.h"
1e307fab 60#include "ust-ctl.h"
00e2e675 61#include "ust-consumer.h"
8e68d1c8 62#include "utils.h"
4063050c 63#include "fd-limit.h"
8782cc74 64#include "health-sessiond.h"
8ac94142 65#include "testpoint.h"
d0b96690 66#include "ust-thread.h"
4d076222 67#include "jul-thread.h"
fac6795d 68
ebaeda94
MD
69#define CONSUMERD_FILE "lttng-consumerd"
70
fac6795d 71const char *progname;
6c71277b 72static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
35f90c40 73static const char *opt_pidfile;
5b8719f5 74static int opt_sig_parent;
97e19046 75static int opt_verbose_consumer;
fac6795d 76static int opt_daemon;
4fba7219 77static int opt_no_kernel;
fac6795d 78static int is_root; /* Set to 1 if the daemon is running as root */
1d4b027a 79static pid_t ppid; /* Parent PID for --sig-parent option */
0bb7724a 80static pid_t child_ppid; /* Internal parent PID use with daemonize. */
67e40797 81static char *rundir;
3bd1e081 82
0bb7724a
DG
83/* Set to 1 when a SIGUSR1 signal is received. */
84static int recv_child_signal;
85
a23ec3a7
DG
86/*
87 * Consumer daemon specific control data. Every value not initialized here is
88 * set to 0 by the static definition.
89 */
3bd1e081
MD
90static struct consumer_data kconsumer_data = {
91 .type = LTTNG_CONSUMER_KERNEL,
60922cb0
DG
92 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
93 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
03550b58
MD
94 .err_sock = -1,
95 .cmd_sock = -1,
173af62f
DG
96 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
97 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
98 .cond = PTHREAD_COND_INITIALIZER,
99 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081 100};
7753dea8
MD
101static struct consumer_data ustconsumer64_data = {
102 .type = LTTNG_CONSUMER64_UST,
60922cb0
DG
103 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
104 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
03550b58
MD
105 .err_sock = -1,
106 .cmd_sock = -1,
173af62f
DG
107 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
108 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
109 .cond = PTHREAD_COND_INITIALIZER,
110 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
7753dea8
MD
111};
112static struct consumer_data ustconsumer32_data = {
113 .type = LTTNG_CONSUMER32_UST,
60922cb0
DG
114 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
115 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
03550b58
MD
116 .err_sock = -1,
117 .cmd_sock = -1,
173af62f
DG
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081
MD
122};
123
26c9d55e 124/* Shared between threads */
099e26bd 125static int dispatch_thread_exit;
fac6795d 126
54d01ffb
DG
127/* Global application Unix socket path */
128static char apps_unix_sock_path[PATH_MAX];
129/* Global client Unix socket path */
130static char client_unix_sock_path[PATH_MAX];
54d01ffb
DG
131/* global wait shm path for UST */
132static char wait_shm_path[PATH_MAX];
44a5e5eb
DG
133/* Global health check unix path */
134static char health_unix_sock_path[PATH_MAX];
fac6795d 135
1d4b027a 136/* Sockets and FDs */
a4b35e07
MD
137static int client_sock = -1;
138static int apps_sock = -1;
2f77fc4b 139int kernel_tracer_fd = -1;
76d7553f 140static int kernel_poll_pipe[2] = { -1, -1 };
1d4b027a 141
273ea72c
DG
142/*
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
145 */
76d7553f 146static int thread_quit_pipe[2] = { -1, -1 };
273ea72c 147
099e26bd
DG
148/*
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
151 */
76d7553f 152static int apps_cmd_pipe[2] = { -1, -1 };
099e26bd 153
d0b96690
DG
154int apps_cmd_notify_pipe[2] = { -1, -1 };
155
1d4b027a 156/* Pthread, Mutexes and Semaphores */
1d4b027a 157static pthread_t apps_thread;
d0b96690 158static pthread_t apps_notify_thread;
099e26bd 159static pthread_t reg_apps_thread;
1d4b027a 160static pthread_t client_thread;
7a485870 161static pthread_t kernel_thread;
099e26bd 162static pthread_t dispatch_thread;
44a5e5eb 163static pthread_t health_thread;
0b2dc8df 164static pthread_t ht_cleanup_thread;
4d076222 165static pthread_t jul_reg_thread;
5eb91c98 166
099e26bd
DG
167/*
168 * UST registration command queue. This queue is tied with a futex and uses a N
169 * wakers / 1 waiter implemented and detailed in futex.c/.h
170 *
171 * The thread_manage_apps and thread_dispatch_ust_registration interact with
172 * this queue and the wait/wake scheme.
173 */
174static struct ust_cmd_queue ust_cmd_queue;
175
b5541356
DG
176/*
177 * Pointer initialized before thread creation.
178 *
179 * This points to the tracing session list containing the session count and a
180 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
181 * MUST NOT be taken if you call a public function in session.c.
04ea676f 182 *
d063d709 183 * The lock is nested inside the structure: session_list_ptr->lock. Please use
54d01ffb 184 * session_lock_list and session_unlock_list for lock acquisition.
b5541356
DG
185 */
186static struct ltt_session_list *session_list_ptr;
187
7753dea8
MD
188int ust_consumerd64_fd = -1;
189int ust_consumerd32_fd = -1;
190
fb6f1fa2
YB
191static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
192static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
193static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
194static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
fb09408a 195
2f77fc4b
DG
196static const char *module_proc_lttng = "/proc/lttng";
197
5c827ce0
DG
198/*
199 * Consumer daemon state which is changed when spawning it, killing it or in
200 * case of a fatal error.
201 */
202enum consumerd_state {
203 CONSUMER_STARTED = 1,
204 CONSUMER_STOPPED = 2,
205 CONSUMER_ERROR = 3,
206};
207
208/*
209 * This consumer daemon state is used to validate if a client command will be
210 * able to reach the consumer. If not, the client is informed. For instance,
211 * doing a "lttng start" when the consumer state is set to ERROR will return an
212 * error to the client.
213 *
214 * The following example shows a possible race condition of this scheme:
215 *
216 * consumer thread error happens
217 * client cmd arrives
218 * client cmd checks state -> still OK
219 * consumer thread exit, sets error
220 * client cmd try to talk to consumer
221 * ...
222 *
223 * However, since the consumer is a different daemon, we have no way of making
224 * sure the command will reach it safely even with this state flag. This is why
225 * we consider that up to the state validation during command processing, the
226 * command is safe. After that, we can not guarantee the correctness of the
227 * client request vis-a-vis the consumer.
228 */
229static enum consumerd_state ust_consumerd_state;
230static enum consumerd_state kernel_consumerd_state;
231
ae9e45b3
DG
232/*
233 * Socket timeout for receiving and sending in seconds.
234 */
235static int app_socket_timeout;
236
12744796
DG
237/* Set in main() with the current page size. */
238long page_size;
239
8782cc74
MD
240/* Application health monitoring */
241struct health_app *health_sessiond;
242
4d076222
DG
243/* JUL TCP port for registration. Used by the JUL thread. */
244unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
245
fb09408a 246static
7753dea8 247void setup_consumerd_path(void)
fb09408a 248{
fc7a59ce 249 const char *bin, *libdir;
fb09408a 250
7753dea8
MD
251 /*
252 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
253 * native architecture size consumer if CONFIG_CONSUMER*_PATH
254 * has not been defined.
7753dea8 255 */
ebaeda94 256#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
257 if (!consumerd32_bin[0]) {
258 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
259 }
260 if (!consumerd32_libdir[0]) {
261 consumerd32_libdir = INSTALL_LIB_PATH;
262 }
263#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
264 if (!consumerd64_bin[0]) {
265 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 266 }
ebaeda94
MD
267 if (!consumerd64_libdir[0]) {
268 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
269 }
270#else
271#error "Unknown bitness"
272#endif
273
fb09408a
MD
274 /*
275 * runtime env. var. overrides the build default.
276 */
fc7a59ce
AM
277 bin = getenv("LTTNG_CONSUMERD32_BIN");
278 if (bin) {
279 consumerd32_bin = bin;
7753dea8 280 }
fc7a59ce
AM
281 bin = getenv("LTTNG_CONSUMERD64_BIN");
282 if (bin) {
283 consumerd64_bin = bin;
ebaeda94 284 }
72f579ee 285 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
286 if (libdir) {
287 consumerd32_libdir = libdir;
288 }
72f579ee 289 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
290 if (libdir) {
291 consumerd64_libdir = libdir;
fb09408a
MD
292 }
293}
294
5eb91c98
DG
295/*
296 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
297 */
d0b96690 298int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
5eb91c98
DG
299{
300 int ret;
301
d0b96690 302 assert(events);
5eb91c98
DG
303
304 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
305 if (ret < 0) {
306 goto error;
307 }
308
309 /* Add quit pipe */
d0b96690 310 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
5eb91c98
DG
311 if (ret < 0) {
312 goto error;
313 }
314
315 return 0;
316
317error:
318 return ret;
319}
320
321/*
322 * Check if the thread quit pipe was triggered.
323 *
324 * Return 1 if it was triggered else 0;
325 */
d0b96690 326int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
5eb91c98
DG
327{
328 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
329 return 1;
330 }
331
332 return 0;
333}
334
273ea72c 335/*
5eb91c98 336 * Init thread quit pipe.
273ea72c
DG
337 *
338 * Return -1 on error or 0 if all pipes are created.
339 */
340static int init_thread_quit_pipe(void)
341{
730389d9 342 int ret, i;
273ea72c 343
730389d9 344 ret = pipe(thread_quit_pipe);
273ea72c 345 if (ret < 0) {
730389d9 346 PERROR("thread quit pipe");
273ea72c
DG
347 goto error;
348 }
349
730389d9
DG
350 for (i = 0; i < 2; i++) {
351 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
352 if (ret < 0) {
353 PERROR("fcntl");
354 goto error;
355 }
356 }
357
273ea72c
DG
358error:
359 return ret;
360}
361
099e26bd
DG
362/*
363 * Stop all threads by closing the thread quit pipe.
364 */
cf3af59e
MD
365static void stop_threads(void)
366{
5eb91c98
DG
367 int ret;
368
cf3af59e
MD
369 /* Stopping all threads */
370 DBG("Terminating all threads");
54d01ffb 371 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
372 if (ret < 0) {
373 ERR("write error on thread quit pipe");
374 }
375
099e26bd 376 /* Dispatch thread */
26c9d55e 377 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 378 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
379}
380
e975f9f8
DG
381/*
382 * Close every consumer sockets.
383 */
384static void close_consumer_sockets(void)
385{
386 int ret;
387
388 if (kconsumer_data.err_sock >= 0) {
389 ret = close(kconsumer_data.err_sock);
390 if (ret < 0) {
391 PERROR("kernel consumer err_sock close");
392 }
393 }
394 if (ustconsumer32_data.err_sock >= 0) {
395 ret = close(ustconsumer32_data.err_sock);
396 if (ret < 0) {
a76cbd9f 397 PERROR("UST consumerd32 err_sock close");
e975f9f8
DG
398 }
399 }
400 if (ustconsumer64_data.err_sock >= 0) {
401 ret = close(ustconsumer64_data.err_sock);
402 if (ret < 0) {
a76cbd9f 403 PERROR("UST consumerd64 err_sock close");
e975f9f8
DG
404 }
405 }
406 if (kconsumer_data.cmd_sock >= 0) {
407 ret = close(kconsumer_data.cmd_sock);
408 if (ret < 0) {
409 PERROR("kernel consumer cmd_sock close");
410 }
411 }
412 if (ustconsumer32_data.cmd_sock >= 0) {
413 ret = close(ustconsumer32_data.cmd_sock);
414 if (ret < 0) {
a76cbd9f 415 PERROR("UST consumerd32 cmd_sock close");
e975f9f8
DG
416 }
417 }
418 if (ustconsumer64_data.cmd_sock >= 0) {
419 ret = close(ustconsumer64_data.cmd_sock);
420 if (ret < 0) {
a76cbd9f 421 PERROR("UST consumerd64 cmd_sock close");
e975f9f8
DG
422 }
423 }
424}
425
fac6795d 426/*
d063d709 427 * Cleanup the daemon
fac6795d 428 */
cf3af59e 429static void cleanup(void)
fac6795d 430{
ef599319 431 int ret;
af9737e9 432 struct ltt_session *sess, *stmp;
8c6c56c2 433 char path[PATH_MAX];
fac6795d 434
1d4b027a 435 DBG("Cleaning up");
e07ae692 436
4e449f3f
MD
437 /*
438 * Close the thread quit pipe. It has already done its job,
439 * since we are now called.
440 */
2f77fc4b
DG
441 utils_close_pipe(thread_quit_pipe);
442
35f90c40
DG
443 /*
444 * If opt_pidfile is undefined, the default file will be wiped when
445 * removing the rundir.
446 */
447 if (opt_pidfile) {
448 ret = remove(opt_pidfile);
449 if (ret < 0) {
450 PERROR("remove pidfile %s", opt_pidfile);
451 }
452 }
453
8c6c56c2
MD
454 DBG("Removing sessiond and consumerd content of directory %s", rundir);
455
456 /* sessiond */
457 snprintf(path, PATH_MAX,
458 "%s/%s",
459 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
460 DBG("Removing %s", path);
461 (void) unlink(path);
462
cd9290dd
DG
463 snprintf(path, PATH_MAX, "%s/%s", rundir,
464 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
465 DBG("Removing %s", path);
466 (void) unlink(path);
467
8c6c56c2
MD
468 /* kconsumerd */
469 snprintf(path, PATH_MAX,
470 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
471 rundir);
472 DBG("Removing %s", path);
473 (void) unlink(path);
474
475 snprintf(path, PATH_MAX,
476 DEFAULT_KCONSUMERD_PATH,
477 rundir);
478 DBG("Removing directory %s", path);
479 (void) rmdir(path);
480
481 /* ust consumerd 32 */
482 snprintf(path, PATH_MAX,
483 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
484 rundir);
485 DBG("Removing %s", path);
486 (void) unlink(path);
487
488 snprintf(path, PATH_MAX,
489 DEFAULT_USTCONSUMERD32_PATH,
490 rundir);
491 DBG("Removing directory %s", path);
492 (void) rmdir(path);
493
494 /* ust consumerd 64 */
495 snprintf(path, PATH_MAX,
496 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
497 rundir);
498 DBG("Removing %s", path);
499 (void) unlink(path);
500
501 snprintf(path, PATH_MAX,
502 DEFAULT_USTCONSUMERD64_PATH,
503 rundir);
504 DBG("Removing directory %s", path);
505 (void) rmdir(path);
5461b305 506
dcbcae3e
MD
507 /*
508 * We do NOT rmdir rundir because there are other processes
509 * using it, for instance lttng-relayd, which can start in
510 * parallel with this teardown.
511 */
512
bd69add2 513 free(rundir);
5461b305 514
99bab54f 515 DBG("Cleaning up all sessions");
fac6795d 516
b5541356 517 /* Destroy session list mutex */
273ea72c
DG
518 if (session_list_ptr != NULL) {
519 pthread_mutex_destroy(&session_list_ptr->lock);
520
521 /* Cleanup ALL session */
54d01ffb
DG
522 cds_list_for_each_entry_safe(sess, stmp,
523 &session_list_ptr->head, list) {
2f77fc4b 524 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
525 }
526 }
527
099e26bd 528 DBG("Closing all UST sockets");
56fff090 529 ust_app_clean_list();
7972aab2 530 buffer_reg_destroy_registries();
099e26bd 531
4fba7219
DG
532 if (is_root && !opt_no_kernel) {
533 DBG2("Closing kernel fd");
a4b35e07 534 if (kernel_tracer_fd >= 0) {
76d7553f
MD
535 ret = close(kernel_tracer_fd);
536 if (ret) {
537 PERROR("close");
538 }
a4b35e07 539 }
2f50c8a3 540 DBG("Unloading kernel modules");
096102bd 541 modprobe_remove_lttng_all();
2f50c8a3 542 }
2f77fc4b 543
e975f9f8
DG
544 close_consumer_sockets();
545
421cb601 546 /* <fun> */
f56a39af 547 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
548 "Matthew, BEET driven development works!%c[%dm",
549 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
550 /* </fun> */
fac6795d
DG
551}
552
e065084a 553/*
d063d709 554 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 555 *
d063d709 556 * Return lttcomm error code.
e065084a
DG
557 */
558static int send_unix_sock(int sock, void *buf, size_t len)
559{
560 /* Check valid length */
c617c0c6 561 if (len == 0) {
e065084a
DG
562 return -1;
563 }
564
565 return lttcomm_send_unix_sock(sock, buf, len);
566}
567
5461b305 568/*
d063d709 569 * Free memory of a command context structure.
5461b305 570 */
a2fb29a5 571static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 572{
a2fb29a5
DG
573 DBG("Clean command context structure");
574 if (*cmd_ctx) {
575 if ((*cmd_ctx)->llm) {
576 free((*cmd_ctx)->llm);
5461b305 577 }
a2fb29a5
DG
578 if ((*cmd_ctx)->lsm) {
579 free((*cmd_ctx)->lsm);
5461b305 580 }
a2fb29a5
DG
581 free(*cmd_ctx);
582 *cmd_ctx = NULL;
5461b305
DG
583 }
584}
585
fac6795d 586/*
0fdd1e2c 587 * Notify UST applications using the shm mmap futex.
fac6795d 588 */
0fdd1e2c 589static int notify_ust_apps(int active)
fac6795d 590{
0fdd1e2c 591 char *wait_shm_mmap;
fac6795d 592
0fdd1e2c 593 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 594
0fdd1e2c
DG
595 /* See shm.c for this call implying mmap, shm and futex calls */
596 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
597 if (wait_shm_mmap == NULL) {
fac6795d
DG
598 goto error;
599 }
600
0fdd1e2c
DG
601 /* Wake waiting process */
602 futex_wait_update((int32_t *) wait_shm_mmap, active);
603
604 /* Apps notified successfully */
605 return 0;
fac6795d
DG
606
607error:
0fdd1e2c 608 return -1;
fac6795d
DG
609}
610
e065084a 611/*
d063d709
DG
612 * Setup the outgoing data buffer for the response (llm) by allocating the
613 * right amount of memory and copying the original information from the lsm
614 * structure.
ca95a216 615 *
d063d709 616 * Return total size of the buffer pointed by buf.
ca95a216 617 */
5461b305 618static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 619{
f3ed775e 620 int ret, buf_size;
ca95a216 621
f3ed775e 622 buf_size = size;
5461b305 623
ba7f0ae5 624 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 625 if (cmd_ctx->llm == NULL) {
76d7553f 626 PERROR("zmalloc");
5461b305 627 ret = -ENOMEM;
ca95a216
DG
628 goto error;
629 }
630
5461b305
DG
631 /* Copy common data */
632 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 633 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 634
5461b305
DG
635 cmd_ctx->llm->data_size = size;
636 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
637
ca95a216
DG
638 return buf_size;
639
640error:
641 return ret;
642}
643
7a485870 644/*
5eb91c98 645 * Update the kernel poll set of all channel fd available over all tracing
d063d709 646 * session. Add the wakeup pipe at the end of the set.
7a485870 647 */
5eb91c98 648static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 649{
5eb91c98 650 int ret;
7a485870
DG
651 struct ltt_session *session;
652 struct ltt_kernel_channel *channel;
653
5eb91c98 654 DBG("Updating kernel poll set");
7a485870 655
54d01ffb 656 session_lock_list();
b5541356 657 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 658 session_lock(session);
7a485870 659 if (session->kernel_session == NULL) {
54d01ffb 660 session_unlock(session);
7a485870
DG
661 continue;
662 }
7a485870 663
54d01ffb
DG
664 cds_list_for_each_entry(channel,
665 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
666 /* Add channel fd to the kernel poll set */
667 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
668 if (ret < 0) {
54d01ffb 669 session_unlock(session);
5eb91c98
DG
670 goto error;
671 }
672 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 673 }
54d01ffb 674 session_unlock(session);
7a485870 675 }
54d01ffb 676 session_unlock_list();
7a485870 677
5eb91c98 678 return 0;
7a485870
DG
679
680error:
54d01ffb 681 session_unlock_list();
7a485870
DG
682 return -1;
683}
684
685/*
54d01ffb 686 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 687 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 688 *
d063d709 689 * Useful for CPU hotplug feature.
7a485870 690 */
2bdd86d4 691static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
692{
693 int ret = 0;
694 struct ltt_session *session;
173af62f 695 struct ltt_kernel_session *ksess;
7a485870
DG
696 struct ltt_kernel_channel *channel;
697
698 DBG("Updating kernel streams for channel fd %d", fd);
699
54d01ffb 700 session_lock_list();
b5541356 701 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 702 session_lock(session);
7a485870 703 if (session->kernel_session == NULL) {
54d01ffb 704 session_unlock(session);
7a485870
DG
705 continue;
706 }
173af62f 707 ksess = session->kernel_session;
d9800920 708
173af62f 709 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
7a485870
DG
710 if (channel->fd == fd) {
711 DBG("Channel found, updating kernel streams");
712 ret = kernel_open_channel_stream(channel);
713 if (ret < 0) {
b3c750d2 714 goto error;
7a485870 715 }
5c786ded
JD
716 /* Update the stream global counter */
717 ksess->stream_count_global += ret;
d9800920 718
7a485870 719 /*
5eb91c98
DG
720 * Have we already sent fds to the consumer? If yes, it means
721 * that tracing is started so it is safe to send our updated
722 * stream fds.
7a485870 723 */
173af62f
DG
724 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
725 struct lttng_ht_iter iter;
726 struct consumer_socket *socket;
727
e7fe706f 728 rcu_read_lock();
173af62f
DG
729 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
730 &iter.iter, socket, node.node) {
173af62f 731 pthread_mutex_lock(socket->lock);
f50f23d9 732 ret = kernel_consumer_send_channel_stream(socket,
2bba9e53
DG
733 channel, ksess,
734 session->output_traces ? 1 : 0);
173af62f
DG
735 pthread_mutex_unlock(socket->lock);
736 if (ret < 0) {
e7fe706f 737 rcu_read_unlock();
173af62f
DG
738 goto error;
739 }
7a485870 740 }
e7fe706f 741 rcu_read_unlock();
7a485870 742 }
b3c750d2 743 goto error;
7a485870
DG
744 }
745 }
54d01ffb 746 session_unlock(session);
7a485870 747 }
54d01ffb 748 session_unlock_list();
b3c750d2 749 return ret;
7a485870 750
b3c750d2 751error:
54d01ffb
DG
752 session_unlock(session);
753 session_unlock_list();
7a485870
DG
754 return ret;
755}
756
487cf67c 757/*
ffe60014
DG
758 * For each tracing session, update newly registered apps. The session list
759 * lock MUST be acquired before calling this.
487cf67c
DG
760 */
761static void update_ust_app(int app_sock)
762{
763 struct ltt_session *sess, *stmp;
764
fdadac08
DG
765 /* Consumer is in an ERROR state. Stop any application update. */
766 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
767 /* Stop the update process since the consumer is dead. */
768 return;
769 }
770
487cf67c
DG
771 /* For all tracing session(s) */
772 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
4ee14516 773 session_lock(sess);
421cb601
DG
774 if (sess->ust_session) {
775 ust_app_global_update(sess->ust_session, app_sock);
776 }
4ee14516 777 session_unlock(sess);
487cf67c
DG
778 }
779}
780
7a485870 781/*
d063d709 782 * This thread manage event coming from the kernel.
7a485870 783 *
d063d709
DG
784 * Features supported in this thread:
785 * -) CPU Hotplug
7a485870
DG
786 */
787static void *thread_manage_kernel(void *data)
788{
139ac872 789 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 790 uint32_t revents, nb_fd;
7a485870 791 char tmp;
5eb91c98 792 struct lttng_poll_event events;
7a485870 793
6993eeb3 794 DBG("[thread] Thread manage kernel started");
7a485870 795
6c71277b 796 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
927ca06a 797
d5d63bf1
DG
798 /*
799 * This first step of the while is to clean this structure which could free
6d737ce4 800 * non NULL pointers so initialize it before the loop.
d5d63bf1 801 */
6d737ce4 802 lttng_poll_init(&events);
d5d63bf1 803
6993eeb3
CB
804 if (testpoint(thread_manage_kernel)) {
805 goto error_testpoint;
806 }
8ac94142 807
840cb59c 808 health_code_update();
44a5e5eb 809
6993eeb3 810 if (testpoint(thread_manage_kernel_before_loop)) {
d21b0d71 811 goto error_testpoint;
6993eeb3
CB
812 }
813
7a485870 814 while (1) {
840cb59c 815 health_code_update();
44a5e5eb 816
7a485870 817 if (update_poll_flag == 1) {
d21b0d71
DG
818 /* Clean events object. We are about to populate it again. */
819 lttng_poll_clean(&events);
820
d0b96690 821 ret = sessiond_set_thread_pollset(&events, 2);
d21b0d71
DG
822 if (ret < 0) {
823 goto error_poll_create;
824 }
825
826 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
827 if (ret < 0) {
828 goto error;
829 }
5f822d0a 830
d21b0d71 831 /* This will add the available kernel channel if any. */
5eb91c98
DG
832 ret = update_kernel_poll(&events);
833 if (ret < 0) {
7a485870
DG
834 goto error;
835 }
836 update_poll_flag = 0;
837 }
838
d21b0d71 839 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
7a485870
DG
840
841 /* Poll infinite value of time */
88f2b785 842 restart:
a78af745 843 health_poll_entry();
5eb91c98 844 ret = lttng_poll_wait(&events, -1);
a78af745 845 health_poll_exit();
7a485870 846 if (ret < 0) {
88f2b785
MD
847 /*
848 * Restart interrupted system call.
849 */
850 if (errno == EINTR) {
851 goto restart;
852 }
7a485870
DG
853 goto error;
854 } else if (ret == 0) {
855 /* Should not happen since timeout is infinite */
85611738
DG
856 ERR("Return value of poll is 0 with an infinite timeout.\n"
857 "This should not have happened! Continuing...");
7a485870
DG
858 continue;
859 }
860
0d9c5d77
DG
861 nb_fd = ret;
862
5eb91c98
DG
863 for (i = 0; i < nb_fd; i++) {
864 /* Fetch once the poll data */
865 revents = LTTNG_POLL_GETEV(&events, i);
866 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 867
840cb59c 868 health_code_update();
44a5e5eb 869
5eb91c98 870 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 871 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 872 if (ret) {
139ac872
MD
873 err = 0;
874 goto exit;
5eb91c98 875 }
7a485870 876
5eb91c98
DG
877 /* Check for data on kernel pipe */
878 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
6cd525e8
MD
879 (void) lttng_read(kernel_poll_pipe[0],
880 &tmp, 1);
f921c78f
DG
881 /*
882 * Ret value is useless here, if this pipe gets any actions an
883 * update is required anyway.
884 */
5eb91c98
DG
885 update_poll_flag = 1;
886 continue;
887 } else {
888 /*
889 * New CPU detected by the kernel. Adding kernel stream to
890 * kernel session and updating the kernel consumer
891 */
892 if (revents & LPOLLIN) {
2bdd86d4 893 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
894 if (ret < 0) {
895 continue;
896 }
897 break;
898 /*
899 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
900 * and unregister kernel stream at this point.
901 */
7a485870 902 }
7a485870
DG
903 }
904 }
905 }
906
139ac872 907exit:
7a485870 908error:
5eb91c98 909 lttng_poll_clean(&events);
76d7553f 910error_poll_create:
6993eeb3 911error_testpoint:
6620da75
DG
912 utils_close_pipe(kernel_poll_pipe);
913 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 914 if (err) {
840cb59c 915 health_error();
139ac872 916 ERR("Health error occurred in %s", __func__);
6620da75
DG
917 WARN("Kernel thread died unexpectedly. "
918 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 919 }
8782cc74 920 health_unregister(health_sessiond);
76d7553f 921 DBG("Kernel thread dying");
7a485870
DG
922 return NULL;
923}
924
a23ec3a7
DG
925/*
926 * Signal pthread condition of the consumer data that the thread.
927 */
928static void signal_consumer_condition(struct consumer_data *data, int state)
929{
930 pthread_mutex_lock(&data->cond_mutex);
931
932 /*
933 * The state is set before signaling. It can be any value, it's the waiter
934 * job to correctly interpret this condition variable associated to the
935 * consumer pthread_cond.
936 *
937 * A value of 0 means that the corresponding thread of the consumer data
938 * was not started. 1 indicates that the thread has started and is ready
939 * for action. A negative value means that there was an error during the
940 * thread bootstrap.
941 */
942 data->consumer_thread_is_ready = state;
943 (void) pthread_cond_signal(&data->cond);
944
945 pthread_mutex_unlock(&data->cond_mutex);
946}
947
1d4b027a 948/*
3bd1e081 949 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 950 */
3bd1e081 951static void *thread_manage_consumer(void *data)
1d4b027a 952{
139ac872 953 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98 954 uint32_t revents, nb_fd;
1d4b027a 955 enum lttcomm_return_code code;
5eb91c98 956 struct lttng_poll_event events;
3bd1e081 957 struct consumer_data *consumer_data = data;
1d4b027a 958
3bd1e081 959 DBG("[thread] Manage consumer started");
1d4b027a 960
6c71277b 961 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
927ca06a 962
855060f8 963 health_code_update();
9449cc75 964
5eb91c98 965 /*
331744e3
JD
966 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
967 * metadata_sock. Nothing more will be added to this poll set.
5eb91c98 968 */
331744e3 969 ret = sessiond_set_thread_pollset(&events, 3);
5eb91c98 970 if (ret < 0) {
76d7553f 971 goto error_poll;
5eb91c98 972 }
273ea72c 973
edb8b045
DG
974 /*
975 * The error socket here is already in a listening state which was done
976 * just before spawning this thread to avoid a race between the consumer
977 * daemon exec trying to connect and the listen() call.
978 */
3bd1e081 979 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
980 if (ret < 0) {
981 goto error;
982 }
983
840cb59c 984 health_code_update();
44a5e5eb 985
331744e3 986 /* Infinite blocking call, waiting for transmission */
88f2b785 987restart:
a78af745 988 health_poll_entry();
8ac94142 989
6993eeb3
CB
990 if (testpoint(thread_manage_consumer)) {
991 goto error;
992 }
8ac94142 993
5eb91c98 994 ret = lttng_poll_wait(&events, -1);
a78af745 995 health_poll_exit();
273ea72c 996 if (ret < 0) {
88f2b785
MD
997 /*
998 * Restart interrupted system call.
999 */
1000 if (errno == EINTR) {
1001 goto restart;
1002 }
273ea72c
DG
1003 goto error;
1004 }
1005
0d9c5d77
DG
1006 nb_fd = ret;
1007
5eb91c98
DG
1008 for (i = 0; i < nb_fd; i++) {
1009 /* Fetch once the poll data */
1010 revents = LTTNG_POLL_GETEV(&events, i);
1011 pollfd = LTTNG_POLL_GETFD(&events, i);
1012
840cb59c 1013 health_code_update();
44a5e5eb 1014
5eb91c98 1015 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1016 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1017 if (ret) {
139ac872
MD
1018 err = 0;
1019 goto exit;
5eb91c98
DG
1020 }
1021
1022 /* Event on the registration socket */
3bd1e081 1023 if (pollfd == consumer_data->err_sock) {
5eb91c98 1024 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1025 ERR("consumer err socket poll error");
5eb91c98
DG
1026 goto error;
1027 }
1028 }
273ea72c
DG
1029 }
1030
3bd1e081 1031 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
1032 if (sock < 0) {
1033 goto error;
1034 }
1035
b662582b
DG
1036 /*
1037 * Set the CLOEXEC flag. Return code is useless because either way, the
1038 * show must go on.
1039 */
1040 (void) utils_set_fd_cloexec(sock);
1041
840cb59c 1042 health_code_update();
44a5e5eb 1043
3bd1e081 1044 DBG2("Receiving code from consumer err_sock");
ee0b0061 1045
712ea556 1046 /* Getting status code from kconsumerd */
54d01ffb
DG
1047 ret = lttcomm_recv_unix_sock(sock, &code,
1048 sizeof(enum lttcomm_return_code));
1d4b027a
DG
1049 if (ret <= 0) {
1050 goto error;
1051 }
1052
840cb59c 1053 health_code_update();
44a5e5eb 1054
f73fabfd 1055 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
331744e3 1056 /* Connect both socket, command and metadata. */
3bd1e081
MD
1057 consumer_data->cmd_sock =
1058 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
4ce514c4 1059 consumer_data->metadata_fd =
331744e3 1060 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
92db7cdc
DG
1061 if (consumer_data->cmd_sock < 0
1062 || consumer_data->metadata_fd < 0) {
331744e3 1063 PERROR("consumer connect cmd socket");
a23ec3a7
DG
1064 /* On error, signal condition and quit. */
1065 signal_consumer_condition(consumer_data, -1);
1d4b027a
DG
1066 goto error;
1067 }
9363801e 1068 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
331744e3
JD
1069 /* Create metadata socket lock. */
1070 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1071 if (consumer_data->metadata_sock.lock == NULL) {
1072 PERROR("zmalloc pthread mutex");
1073 ret = -1;
1074 goto error;
1075 }
1076 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1077
a23ec3a7 1078 signal_consumer_condition(consumer_data, 1);
331744e3
JD
1079 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1080 DBG("Consumer metadata socket ready (fd: %d)",
4ce514c4 1081 consumer_data->metadata_fd);
1d4b027a 1082 } else {
3bd1e081 1083 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
1084 lttcomm_get_readable_code(-code));
1085 goto error;
1086 }
1087
331744e3 1088 /* Remove the consumerd error sock since we've established a connexion */
3bd1e081 1089 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1090 if (ret < 0) {
72079cae
DG
1091 goto error;
1092 }
1093
331744e3 1094 /* Add new accepted error socket. */
5eb91c98
DG
1095 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1096 if (ret < 0) {
72079cae 1097 goto error;
5eb91c98
DG
1098 }
1099
331744e3 1100 /* Add metadata socket that is successfully connected. */
4ce514c4 1101 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
331744e3
JD
1102 LPOLLIN | LPOLLRDHUP);
1103 if (ret < 0) {
1104 goto error;
1105 }
1106
840cb59c 1107 health_code_update();
44a5e5eb 1108
331744e3 1109 /* Infinite blocking call, waiting for transmission */
88f2b785 1110restart_poll:
331744e3
JD
1111 while (1) {
1112 health_poll_entry();
1113 ret = lttng_poll_wait(&events, -1);
1114 health_poll_exit();
1115 if (ret < 0) {
1116 /*
1117 * Restart interrupted system call.
1118 */
1119 if (errno == EINTR) {
1120 goto restart_poll;
1121 }
1122 goto error;
88f2b785 1123 }
72079cae 1124
331744e3 1125 nb_fd = ret;
0d9c5d77 1126
331744e3
JD
1127 for (i = 0; i < nb_fd; i++) {
1128 /* Fetch once the poll data */
1129 revents = LTTNG_POLL_GETEV(&events, i);
1130 pollfd = LTTNG_POLL_GETFD(&events, i);
5eb91c98 1131
331744e3 1132 health_code_update();
44a5e5eb 1133
331744e3
JD
1134 /* Thread quit pipe has been closed. Killing thread. */
1135 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1136 if (ret) {
1137 err = 0;
1138 goto exit;
1139 }
5eb91c98 1140
331744e3
JD
1141 if (pollfd == sock) {
1142 /* Event on the consumerd socket */
1143 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1144 ERR("consumer err socket second poll error");
1145 goto error;
1146 }
1147 health_code_update();
1148 /* Wait for any kconsumerd error */
1149 ret = lttcomm_recv_unix_sock(sock, &code,
1150 sizeof(enum lttcomm_return_code));
1151 if (ret <= 0) {
1152 ERR("consumer closed the command socket");
1153 goto error;
1154 }
1155
1156 ERR("consumer return code : %s",
1157 lttcomm_get_readable_code(-code));
1158
1159 goto exit;
4ce514c4 1160 } else if (pollfd == consumer_data->metadata_fd) {
331744e3
JD
1161 /* UST metadata requests */
1162 ret = ust_consumer_metadata_request(
1163 &consumer_data->metadata_sock);
1164 if (ret < 0) {
1165 ERR("Handling metadata request");
1166 goto error;
1167 }
1168 break;
1169 } else {
1170 ERR("Unknown pollfd");
5eb91c98
DG
1171 goto error;
1172 }
1173 }
331744e3 1174 health_code_update();
5eb91c98
DG
1175 }
1176
139ac872 1177exit:
1d4b027a 1178error:
fdadac08
DG
1179 /*
1180 * We lock here because we are about to close the sockets and some other
92db7cdc
DG
1181 * thread might be using them so get exclusive access which will abort all
1182 * other consumer command by other threads.
fdadac08
DG
1183 */
1184 pthread_mutex_lock(&consumer_data->lock);
1185
5c827ce0
DG
1186 /* Immediately set the consumerd state to stopped */
1187 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1188 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1189 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1190 consumer_data->type == LTTNG_CONSUMER32_UST) {
1191 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1192 } else {
1193 /* Code flow error... */
1194 assert(0);
1195 }
1196
76d7553f
MD
1197 if (consumer_data->err_sock >= 0) {
1198 ret = close(consumer_data->err_sock);
1199 if (ret) {
1200 PERROR("close");
1201 }
a76cbd9f 1202 consumer_data->err_sock = -1;
76d7553f
MD
1203 }
1204 if (consumer_data->cmd_sock >= 0) {
1205 ret = close(consumer_data->cmd_sock);
1206 if (ret) {
1207 PERROR("close");
1208 }
a76cbd9f 1209 consumer_data->cmd_sock = -1;
76d7553f 1210 }
9363801e
DG
1211 if (*consumer_data->metadata_sock.fd_ptr >= 0) {
1212 ret = close(*consumer_data->metadata_sock.fd_ptr);
331744e3
JD
1213 if (ret) {
1214 PERROR("close");
1215 }
1216 }
331744e3 1217
76d7553f
MD
1218 if (sock >= 0) {
1219 ret = close(sock);
1220 if (ret) {
1221 PERROR("close");
1222 }
1223 }
273ea72c 1224
3bd1e081
MD
1225 unlink(consumer_data->err_unix_sock_path);
1226 unlink(consumer_data->cmd_unix_sock_path);
1227 consumer_data->pid = 0;
fdadac08 1228 pthread_mutex_unlock(&consumer_data->lock);
92db7cdc 1229
fdadac08
DG
1230 /* Cleanup metadata socket mutex. */
1231 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1232 free(consumer_data->metadata_sock.lock);
1d4b027a 1233
5eb91c98 1234 lttng_poll_clean(&events);
76d7553f 1235error_poll:
139ac872 1236 if (err) {
840cb59c 1237 health_error();
139ac872
MD
1238 ERR("Health error occurred in %s", __func__);
1239 }
8782cc74 1240 health_unregister(health_sessiond);
76d7553f 1241 DBG("consumer thread cleanup completed");
0177d773 1242
5eb91c98 1243 return NULL;
099e26bd
DG
1244}
1245
099e26bd
DG
1246/*
1247 * This thread manage application communication.
1d4b027a
DG
1248 */
1249static void *thread_manage_apps(void *data)
099e26bd 1250{
139ac872 1251 int i, ret, pollfd, err = -1;
6cd525e8 1252 ssize_t size_ret;
5eb91c98 1253 uint32_t revents, nb_fd;
5eb91c98 1254 struct lttng_poll_event events;
099e26bd
DG
1255
1256 DBG("[thread] Manage application started");
1257
f6a9efaa
DG
1258 rcu_register_thread();
1259 rcu_thread_online();
1260
6c71277b 1261 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
927ca06a 1262
6993eeb3
CB
1263 if (testpoint(thread_manage_apps)) {
1264 goto error_testpoint;
1265 }
1266
840cb59c 1267 health_code_update();
44a5e5eb 1268
d0b96690 1269 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1270 if (ret < 0) {
76d7553f 1271 goto error_poll_create;
5eb91c98 1272 }
099e26bd 1273
5eb91c98
DG
1274 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1275 if (ret < 0) {
1276 goto error;
1277 }
099e26bd 1278
6993eeb3
CB
1279 if (testpoint(thread_manage_apps_before_loop)) {
1280 goto error;
1281 }
8ac94142 1282
840cb59c 1283 health_code_update();
44a5e5eb 1284
5eb91c98 1285 while (1) {
d21b0d71 1286 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
099e26bd
DG
1287
1288 /* Inifinite blocking call, waiting for transmission */
88f2b785 1289 restart:
a78af745 1290 health_poll_entry();
5eb91c98 1291 ret = lttng_poll_wait(&events, -1);
a78af745 1292 health_poll_exit();
099e26bd 1293 if (ret < 0) {
88f2b785
MD
1294 /*
1295 * Restart interrupted system call.
1296 */
1297 if (errno == EINTR) {
1298 goto restart;
1299 }
099e26bd
DG
1300 goto error;
1301 }
1302
0d9c5d77
DG
1303 nb_fd = ret;
1304
5eb91c98
DG
1305 for (i = 0; i < nb_fd; i++) {
1306 /* Fetch once the poll data */
1307 revents = LTTNG_POLL_GETEV(&events, i);
1308 pollfd = LTTNG_POLL_GETFD(&events, i);
1309
840cb59c 1310 health_code_update();
44a5e5eb 1311
5eb91c98 1312 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1313 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1314 if (ret) {
139ac872
MD
1315 err = 0;
1316 goto exit;
5eb91c98 1317 }
099e26bd 1318
5eb91c98
DG
1319 /* Inspect the apps cmd pipe */
1320 if (pollfd == apps_cmd_pipe[0]) {
1321 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1322 ERR("Apps command pipe error");
0177d773 1323 goto error;
5eb91c98 1324 } else if (revents & LPOLLIN) {
d0b96690
DG
1325 int sock;
1326
5eb91c98 1327 /* Empty pipe */
6cd525e8
MD
1328 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1329 if (size_ret < sizeof(sock)) {
76d7553f 1330 PERROR("read apps cmd pipe");
5eb91c98
DG
1331 goto error;
1332 }
099e26bd 1333
840cb59c 1334 health_code_update();
44a5e5eb 1335
ffe60014 1336 /*
d0b96690
DG
1337 * We only monitor the error events of the socket. This
1338 * thread does not handle any incoming data from UST
1339 * (POLLIN).
ffe60014 1340 */
d0b96690
DG
1341 ret = lttng_poll_add(&events, sock,
1342 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1343 if (ret < 0) {
5eb91c98 1344 goto error;
e0c7ec2b 1345 }
acc7b41b 1346
d0b96690 1347 DBG("Apps with sock %d added to poll set", sock);
487cf67c 1348
840cb59c 1349 health_code_update();
44a5e5eb 1350
5eb91c98 1351 break;
0177d773 1352 }
5eb91c98
DG
1353 } else {
1354 /*
54d01ffb
DG
1355 * At this point, we know that a registered application made
1356 * the event at poll_wait.
5eb91c98
DG
1357 */
1358 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1359 /* Removing from the poll set */
1360 ret = lttng_poll_del(&events, pollfd);
1361 if (ret < 0) {
1362 goto error;
1363 }
099e26bd 1364
b9d9b220 1365 /* Socket closed on remote end. */
56fff090 1366 ust_app_unregister(pollfd);
5eb91c98
DG
1367 break;
1368 }
099e26bd 1369 }
44a5e5eb 1370
840cb59c 1371 health_code_update();
099e26bd 1372 }
099e26bd
DG
1373 }
1374
139ac872 1375exit:
099e26bd 1376error:
5eb91c98 1377 lttng_poll_clean(&events);
76d7553f 1378error_poll_create:
6993eeb3 1379error_testpoint:
6620da75
DG
1380 utils_close_pipe(apps_cmd_pipe);
1381 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1382
1383 /*
1384 * We don't clean the UST app hash table here since already registered
1385 * applications can still be controlled so let them be until the session
1386 * daemon dies or the applications stop.
1387 */
1388
139ac872 1389 if (err) {
840cb59c 1390 health_error();
139ac872
MD
1391 ERR("Health error occurred in %s", __func__);
1392 }
8782cc74 1393 health_unregister(health_sessiond);
76d7553f 1394 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1395 rcu_thread_offline();
1396 rcu_unregister_thread();
099e26bd
DG
1397 return NULL;
1398}
1399
d0b96690 1400/*
d88aee68
DG
1401 * Send a socket to a thread This is called from the dispatch UST registration
1402 * thread once all sockets are set for the application.
d0b96690 1403 *
b85dc84c
DG
1404 * The sock value can be invalid, we don't really care, the thread will handle
1405 * it and make the necessary cleanup if so.
1406 *
d0b96690
DG
1407 * On success, return 0 else a negative value being the errno message of the
1408 * write().
1409 */
d88aee68 1410static int send_socket_to_thread(int fd, int sock)
d0b96690 1411{
6cd525e8 1412 ssize_t ret;
d0b96690 1413
b85dc84c
DG
1414 /*
1415 * It's possible that the FD is set as invalid with -1 concurrently just
1416 * before calling this function being a shutdown state of the thread.
1417 */
1418 if (fd < 0) {
1419 ret = -EBADF;
1420 goto error;
1421 }
d0b96690 1422
6cd525e8
MD
1423 ret = lttng_write(fd, &sock, sizeof(sock));
1424 if (ret < sizeof(sock)) {
d88aee68 1425 PERROR("write apps pipe %d", fd);
d0b96690
DG
1426 if (ret < 0) {
1427 ret = -errno;
1428 }
1429 goto error;
1430 }
1431
1432 /* All good. Don't send back the write positive ret value. */
1433 ret = 0;
1434error:
6cd525e8 1435 return (int) ret;
d0b96690
DG
1436}
1437
f45e313d
DG
1438/*
1439 * Sanitize the wait queue of the dispatch registration thread meaning removing
1440 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1441 * notify socket is never received.
1442 */
1443static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1444{
1445 int ret, nb_fd = 0, i;
1446 unsigned int fd_added = 0;
1447 struct lttng_poll_event events;
1448 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1449
1450 assert(wait_queue);
1451
1452 lttng_poll_init(&events);
1453
1454 /* Just skip everything for an empty queue. */
1455 if (!wait_queue->count) {
1456 goto end;
1457 }
1458
1459 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1460 if (ret < 0) {
1461 goto error_create;
1462 }
1463
1464 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1465 &wait_queue->head, head) {
1466 assert(wait_node->app);
1467 ret = lttng_poll_add(&events, wait_node->app->sock,
1468 LPOLLHUP | LPOLLERR);
1469 if (ret < 0) {
1470 goto error;
1471 }
1472
1473 fd_added = 1;
1474 }
1475
1476 if (!fd_added) {
1477 goto end;
1478 }
1479
1480 /*
1481 * Poll but don't block so we can quickly identify the faulty events and
1482 * clean them afterwards from the wait queue.
1483 */
1484 ret = lttng_poll_wait(&events, 0);
1485 if (ret < 0) {
1486 goto error;
1487 }
1488 nb_fd = ret;
1489
1490 for (i = 0; i < nb_fd; i++) {
1491 /* Get faulty FD. */
1492 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1493 int pollfd = LTTNG_POLL_GETFD(&events, i);
1494
1495 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1496 &wait_queue->head, head) {
1497 if (pollfd == wait_node->app->sock &&
1498 (revents & (LPOLLHUP | LPOLLERR))) {
1499 cds_list_del(&wait_node->head);
1500 wait_queue->count--;
1501 ust_app_destroy(wait_node->app);
1502 free(wait_node);
1503 break;
1504 }
1505 }
1506 }
1507
1508 if (nb_fd > 0) {
1509 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1510 }
1511
1512end:
1513 lttng_poll_clean(&events);
1514 return;
1515
1516error:
1517 lttng_poll_clean(&events);
1518error_create:
1519 ERR("Unable to sanitize wait queue");
1520 return;
1521}
1522
099e26bd
DG
1523/*
1524 * Dispatch request from the registration threads to the application
1525 * communication thread.
1526 */
1527static void *thread_dispatch_ust_registration(void *data)
1528{
12e2b881 1529 int ret, err = -1;
099e26bd
DG
1530 struct cds_wfq_node *node;
1531 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1532 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1533 struct ust_reg_wait_queue wait_queue = {
1534 .count = 0,
1535 };
d0b96690 1536
6c71277b 1537 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
12e2b881
MD
1538
1539 health_code_update();
1540
f45e313d 1541 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1542
1543 DBG("[thread] Dispatch UST command started");
1544
26c9d55e 1545 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1546 health_code_update();
1547
099e26bd
DG
1548 /* Atomically prepare the queue futex */
1549 futex_nto1_prepare(&ust_cmd_queue.futex);
1550
1551 do {
d0b96690 1552 struct ust_app *app = NULL;
7972aab2 1553 ust_cmd = NULL;
d0b96690 1554
f45e313d
DG
1555 /*
1556 * Make sure we don't have node(s) that have hung up before receiving
1557 * the notify socket. This is to clean the list in order to avoid
1558 * memory leaks from notify socket that are never seen.
1559 */
1560 sanitize_wait_queue(&wait_queue);
1561
12e2b881 1562 health_code_update();
099e26bd
DG
1563 /* Dequeue command for registration */
1564 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1565 if (node == NULL) {
00a17c97 1566 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1567 /* Continue thread execution */
1568 break;
1569 }
1570
1571 ust_cmd = caa_container_of(node, struct ust_command, node);
1572
2f50c8a3
DG
1573 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1574 " gid:%d sock:%d name:%s (version %d.%d)",
1575 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1576 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1577 ust_cmd->sock, ust_cmd->reg_msg.name,
1578 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1579
1580 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1581 wait_node = zmalloc(sizeof(*wait_node));
1582 if (!wait_node) {
1583 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1584 ret = close(ust_cmd->sock);
1585 if (ret < 0) {
1586 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1587 }
51dec90d 1588 lttng_fd_put(LTTNG_FD_APPS, 1);
7972aab2 1589 free(ust_cmd);
d0b96690
DG
1590 goto error;
1591 }
1592 CDS_INIT_LIST_HEAD(&wait_node->head);
1593
1594 /* Create application object if socket is CMD. */
1595 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1596 ust_cmd->sock);
1597 if (!wait_node->app) {
1598 ret = close(ust_cmd->sock);
1599 if (ret < 0) {
1600 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1601 }
51dec90d 1602 lttng_fd_put(LTTNG_FD_APPS, 1);
d88aee68 1603 free(wait_node);
7972aab2 1604 free(ust_cmd);
d0b96690
DG
1605 continue;
1606 }
1607 /*
1608 * Add application to the wait queue so we can set the notify
1609 * socket before putting this object in the global ht.
1610 */
f45e313d
DG
1611 cds_list_add(&wait_node->head, &wait_queue.head);
1612 wait_queue.count++;
d0b96690 1613
7972aab2 1614 free(ust_cmd);
d0b96690
DG
1615 /*
1616 * We have to continue here since we don't have the notify
1617 * socket and the application MUST be added to the hash table
1618 * only at that moment.
1619 */
1620 continue;
1621 } else {
1622 /*
1623 * Look for the application in the local wait queue and set the
1624 * notify socket if found.
1625 */
d88aee68 1626 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1627 &wait_queue.head, head) {
12e2b881 1628 health_code_update();
d0b96690
DG
1629 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1630 wait_node->app->notify_sock = ust_cmd->sock;
1631 cds_list_del(&wait_node->head);
f45e313d 1632 wait_queue.count--;
d0b96690
DG
1633 app = wait_node->app;
1634 free(wait_node);
1635 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1636 break;
1637 }
1638 }
020d7f60
DG
1639
1640 /*
1641 * With no application at this stage the received socket is
1642 * basically useless so close it before we free the cmd data
1643 * structure for good.
1644 */
1645 if (!app) {
1646 ret = close(ust_cmd->sock);
1647 if (ret < 0) {
1648 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1649 }
51dec90d 1650 lttng_fd_put(LTTNG_FD_APPS, 1);
020d7f60 1651 }
7972aab2 1652 free(ust_cmd);
d0b96690
DG
1653 }
1654
1655 if (app) {
d0b96690
DG
1656 /*
1657 * @session_lock_list
1658 *
1659 * Lock the global session list so from the register up to the
1660 * registration done message, no thread can see the application
1661 * and change its state.
1662 */
1663 session_lock_list();
1664 rcu_read_lock();
d88aee68 1665
d0b96690
DG
1666 /*
1667 * Add application to the global hash table. This needs to be
1668 * done before the update to the UST registry can locate the
1669 * application.
1670 */
1671 ust_app_add(app);
d88aee68
DG
1672
1673 /* Set app version. This call will print an error if needed. */
1674 (void) ust_app_version(app);
1675
1676 /* Send notify socket through the notify pipe. */
1677 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1678 app->notify_sock);
1679 if (ret < 0) {
1680 rcu_read_unlock();
1681 session_unlock_list();
b85dc84c
DG
1682 /*
1683 * No notify thread, stop the UST tracing. However, this is
1684 * not an internal error of the this thread thus setting
1685 * the health error code to a normal exit.
1686 */
1687 err = 0;
d88aee68 1688 goto error;
6620da75 1689 }
d88aee68 1690
d0b96690
DG
1691 /*
1692 * Update newly registered application with the tracing
1693 * registry info already enabled information.
1694 */
1695 update_ust_app(app->sock);
d88aee68
DG
1696
1697 /*
1698 * Don't care about return value. Let the manage apps threads
1699 * handle app unregistration upon socket close.
1700 */
1701 (void) ust_app_register_done(app->sock);
1702
1703 /*
1704 * Even if the application socket has been closed, send the app
1705 * to the thread and unregistration will take place at that
1706 * place.
1707 */
1708 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 1709 if (ret < 0) {
d88aee68
DG
1710 rcu_read_unlock();
1711 session_unlock_list();
b85dc84c
DG
1712 /*
1713 * No apps. thread, stop the UST tracing. However, this is
1714 * not an internal error of the this thread thus setting
1715 * the health error code to a normal exit.
1716 */
1717 err = 0;
d88aee68 1718 goto error;
d0b96690 1719 }
d88aee68 1720
d0b96690
DG
1721 rcu_read_unlock();
1722 session_unlock_list();
099e26bd 1723 }
099e26bd
DG
1724 } while (node != NULL);
1725
12e2b881 1726 health_poll_entry();
099e26bd
DG
1727 /* Futex wait on queue. Blocking call on futex() */
1728 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 1729 health_poll_exit();
099e26bd 1730 }
12e2b881
MD
1731 /* Normal exit, no error */
1732 err = 0;
099e26bd
DG
1733
1734error:
d88aee68
DG
1735 /* Clean up wait queue. */
1736 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1737 &wait_queue.head, head) {
d88aee68 1738 cds_list_del(&wait_node->head);
f45e313d 1739 wait_queue.count--;
d88aee68
DG
1740 free(wait_node);
1741 }
1742
099e26bd 1743 DBG("Dispatch thread dying");
12e2b881
MD
1744 if (err) {
1745 health_error();
1746 ERR("Health error occurred in %s", __func__);
1747 }
8782cc74 1748 health_unregister(health_sessiond);
099e26bd
DG
1749 return NULL;
1750}
1751
1752/*
1753 * This thread manage application registration.
1754 */
1755static void *thread_registration_apps(void *data)
1d4b027a 1756{
139ac872 1757 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
1758 uint32_t revents, nb_fd;
1759 struct lttng_poll_event events;
099e26bd
DG
1760 /*
1761 * Get allocated in this thread, enqueued to a global queue, dequeued and
1762 * freed in the manage apps thread.
1763 */
1764 struct ust_command *ust_cmd = NULL;
1d4b027a 1765
099e26bd 1766 DBG("[thread] Manage application registration started");
1d4b027a 1767
6c71277b 1768 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
927ca06a 1769
6993eeb3
CB
1770 if (testpoint(thread_registration_apps)) {
1771 goto error_testpoint;
1772 }
8ac94142 1773
1d4b027a
DG
1774 ret = lttcomm_listen_unix_sock(apps_sock);
1775 if (ret < 0) {
76d7553f 1776 goto error_listen;
1d4b027a
DG
1777 }
1778
5eb91c98
DG
1779 /*
1780 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1781 * more will be added to this poll set.
1782 */
d0b96690 1783 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1784 if (ret < 0) {
76d7553f 1785 goto error_create_poll;
5eb91c98 1786 }
273ea72c 1787
5eb91c98
DG
1788 /* Add the application registration socket */
1789 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1790 if (ret < 0) {
76d7553f 1791 goto error_poll_add;
5eb91c98 1792 }
273ea72c 1793
1d4b027a 1794 /* Notify all applications to register */
0fdd1e2c
DG
1795 ret = notify_ust_apps(1);
1796 if (ret < 0) {
1797 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
1798 "Execution continues but there might be problem for already\n"
1799 "running applications that wishes to register.");
0fdd1e2c 1800 }
1d4b027a
DG
1801
1802 while (1) {
1803 DBG("Accepting application registration");
273ea72c
DG
1804
1805 /* Inifinite blocking call, waiting for transmission */
88f2b785 1806 restart:
a78af745 1807 health_poll_entry();
5eb91c98 1808 ret = lttng_poll_wait(&events, -1);
a78af745 1809 health_poll_exit();
273ea72c 1810 if (ret < 0) {
88f2b785
MD
1811 /*
1812 * Restart interrupted system call.
1813 */
1814 if (errno == EINTR) {
1815 goto restart;
1816 }
273ea72c
DG
1817 goto error;
1818 }
1819
0d9c5d77
DG
1820 nb_fd = ret;
1821
5eb91c98 1822 for (i = 0; i < nb_fd; i++) {
840cb59c 1823 health_code_update();
139ac872 1824
5eb91c98
DG
1825 /* Fetch once the poll data */
1826 revents = LTTNG_POLL_GETEV(&events, i);
1827 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 1828
5eb91c98 1829 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1830 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1831 if (ret) {
139ac872
MD
1832 err = 0;
1833 goto exit;
90014c57 1834 }
1d4b027a 1835
5eb91c98
DG
1836 /* Event on the registration socket */
1837 if (pollfd == apps_sock) {
1838 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1839 ERR("Register apps socket poll error");
1840 goto error;
1841 } else if (revents & LPOLLIN) {
1842 sock = lttcomm_accept_unix_sock(apps_sock);
1843 if (sock < 0) {
1844 goto error;
1845 }
099e26bd 1846
16c5c8fa
DG
1847 /*
1848 * Set socket timeout for both receiving and ending.
1849 * app_socket_timeout is in seconds, whereas
1850 * lttcomm_setsockopt_rcv_timeout and
1851 * lttcomm_setsockopt_snd_timeout expect msec as
1852 * parameter.
1853 */
1854 (void) lttcomm_setsockopt_rcv_timeout(sock,
1855 app_socket_timeout * 1000);
1856 (void) lttcomm_setsockopt_snd_timeout(sock,
1857 app_socket_timeout * 1000);
1858
b662582b
DG
1859 /*
1860 * Set the CLOEXEC flag. Return code is useless because
1861 * either way, the show must go on.
1862 */
1863 (void) utils_set_fd_cloexec(sock);
1864
5eb91c98 1865 /* Create UST registration command for enqueuing */
ba7f0ae5 1866 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 1867 if (ust_cmd == NULL) {
76d7553f 1868 PERROR("ust command zmalloc");
5eb91c98
DG
1869 goto error;
1870 }
1d4b027a 1871
5eb91c98
DG
1872 /*
1873 * Using message-based transmissions to ensure we don't
1874 * have to deal with partially received messages.
1875 */
4063050c
MD
1876 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1877 if (ret < 0) {
1878 ERR("Exhausted file descriptors allowed for applications.");
1879 free(ust_cmd);
1880 ret = close(sock);
1881 if (ret) {
1882 PERROR("close");
1883 }
1884 sock = -1;
1885 continue;
1886 }
d88aee68 1887
840cb59c 1888 health_code_update();
d0b96690
DG
1889 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1890 if (ret < 0) {
5eb91c98 1891 free(ust_cmd);
d0b96690 1892 /* Close socket of the application. */
76d7553f
MD
1893 ret = close(sock);
1894 if (ret) {
1895 PERROR("close");
1896 }
4063050c 1897 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 1898 sock = -1;
5eb91c98
DG
1899 continue;
1900 }
840cb59c 1901 health_code_update();
099e26bd 1902
5eb91c98 1903 ust_cmd->sock = sock;
34a2494f 1904 sock = -1;
099e26bd 1905
5eb91c98
DG
1906 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1907 " gid:%d sock:%d name:%s (version %d.%d)",
1908 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1909 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1910 ust_cmd->sock, ust_cmd->reg_msg.name,
1911 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 1912
5eb91c98
DG
1913 /*
1914 * Lock free enqueue the registration request. The red pill
54d01ffb 1915 * has been taken! This apps will be part of the *system*.
5eb91c98
DG
1916 */
1917 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1918
1919 /*
1920 * Wake the registration queue futex. Implicit memory
1921 * barrier with the exchange in cds_wfq_enqueue.
1922 */
1923 futex_nto1_wake(&ust_cmd_queue.futex);
1924 }
1925 }
90014c57 1926 }
1d4b027a
DG
1927 }
1928
139ac872 1929exit:
1d4b027a 1930error:
139ac872 1931 if (err) {
840cb59c 1932 health_error();
139ac872
MD
1933 ERR("Health error occurred in %s", __func__);
1934 }
139ac872 1935
0fdd1e2c
DG
1936 /* Notify that the registration thread is gone */
1937 notify_ust_apps(0);
1938
a4b35e07 1939 if (apps_sock >= 0) {
76d7553f
MD
1940 ret = close(apps_sock);
1941 if (ret) {
1942 PERROR("close");
1943 }
a4b35e07 1944 }
46c3f085 1945 if (sock >= 0) {
76d7553f
MD
1946 ret = close(sock);
1947 if (ret) {
1948 PERROR("close");
1949 }
4063050c 1950 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 1951 }
273ea72c 1952 unlink(apps_unix_sock_path);
0fdd1e2c 1953
76d7553f 1954error_poll_add:
5eb91c98 1955 lttng_poll_clean(&events);
76d7553f
MD
1956error_listen:
1957error_create_poll:
6993eeb3 1958error_testpoint:
76d7553f 1959 DBG("UST Registration thread cleanup complete");
8782cc74 1960 health_unregister(health_sessiond);
5eb91c98 1961
1d4b027a
DG
1962 return NULL;
1963}
1964
8c0faa1d 1965/*
3bd1e081 1966 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 1967 * exec or it will fails.
8c0faa1d 1968 */
3bd1e081 1969static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 1970{
a23ec3a7 1971 int ret, clock_ret;
ee0b0061
DG
1972 struct timespec timeout;
1973
a23ec3a7
DG
1974 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1975 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 1976
a23ec3a7
DG
1977 /* Setup pthread condition */
1978 ret = pthread_condattr_init(&consumer_data->condattr);
1979 if (ret != 0) {
1980 errno = ret;
1981 PERROR("pthread_condattr_init consumer data");
1982 goto error;
1983 }
1984
1985 /*
1986 * Set the monotonic clock in order to make sure we DO NOT jump in time
1987 * between the clock_gettime() call and the timedwait call. See bug #324
1988 * for a more details and how we noticed it.
1989 */
1990 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1991 if (ret != 0) {
1992 errno = ret;
1993 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
1994 goto error;
1995 }
8c0faa1d 1996
a23ec3a7
DG
1997 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1998 if (ret != 0) {
1999 errno = ret;
2000 PERROR("pthread_cond_init consumer data");
2001 goto error;
2002 }
2003
2004 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2005 consumer_data);
8c0faa1d 2006 if (ret != 0) {
3bd1e081 2007 PERROR("pthread_create consumer");
ee0b0061 2008 ret = -1;
8c0faa1d
DG
2009 goto error;
2010 }
2011
a23ec3a7
DG
2012 /* We are about to wait on a pthread condition */
2013 pthread_mutex_lock(&consumer_data->cond_mutex);
2014
ee0b0061 2015 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
2016 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2017 /*
2018 * Set the timeout for the condition timed wait even if the clock gettime
2019 * call fails since we might loop on that call and we want to avoid to
2020 * increment the timeout too many times.
2021 */
2022 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2023
2024 /*
2025 * The following loop COULD be skipped in some conditions so this is why we
2026 * set ret to 0 in order to make sure at least one round of the loop is
2027 * done.
2028 */
2029 ret = 0;
2030
2031 /*
2032 * Loop until the condition is reached or when a timeout is reached. Note
2033 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2034 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2035 * possible. This loop does not take any chances and works with both of
2036 * them.
2037 */
2038 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2039 if (clock_ret < 0) {
2040 PERROR("clock_gettime spawn consumer");
2041 /* Infinite wait for the consumerd thread to be ready */
2042 ret = pthread_cond_wait(&consumer_data->cond,
2043 &consumer_data->cond_mutex);
2044 } else {
2045 ret = pthread_cond_timedwait(&consumer_data->cond,
2046 &consumer_data->cond_mutex, &timeout);
2047 }
ee0b0061 2048 }
8c0faa1d 2049
a23ec3a7
DG
2050 /* Release the pthread condition */
2051 pthread_mutex_unlock(&consumer_data->cond_mutex);
2052
2053 if (ret != 0) {
2054 errno = ret;
2055 if (ret == ETIMEDOUT) {
ee0b0061
DG
2056 /*
2057 * Call has timed out so we kill the kconsumerd_thread and return
2058 * an error.
2059 */
a23ec3a7
DG
2060 ERR("Condition timed out. The consumer thread was never ready."
2061 " Killing it");
3bd1e081 2062 ret = pthread_cancel(consumer_data->thread);
ee0b0061 2063 if (ret < 0) {
3bd1e081 2064 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
2065 }
2066 } else {
a23ec3a7 2067 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061
DG
2068 }
2069 goto error;
2070 }
2071
3bd1e081
MD
2072 pthread_mutex_lock(&consumer_data->pid_mutex);
2073 if (consumer_data->pid == 0) {
a23ec3a7 2074 ERR("Consumerd did not start");
3bd1e081 2075 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2076 goto error;
2077 }
3bd1e081 2078 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2079
8c0faa1d
DG
2080 return 0;
2081
2082error:
2083 return ret;
2084}
2085
d9800920 2086/*
3bd1e081 2087 * Join consumer thread
d9800920 2088 */
3bd1e081 2089static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2090{
2091 void *status;
cf3af59e 2092
e8209f6b
DG
2093 /* Consumer pid must be a real one. */
2094 if (consumer_data->pid > 0) {
c617c0c6 2095 int ret;
3bd1e081 2096 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2097 if (ret) {
3bd1e081 2098 ERR("Error killing consumer daemon");
cf3af59e
MD
2099 return ret;
2100 }
3bd1e081 2101 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2102 } else {
2103 return 0;
2104 }
2105}
2106
8c0faa1d 2107/*
3bd1e081 2108 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2109 *
d063d709 2110 * Return pid if successful else -1.
8c0faa1d 2111 */
3bd1e081 2112static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2113{
2114 int ret;
2115 pid_t pid;
94c55f17 2116 const char *consumer_to_use;
53086306 2117 const char *verbosity;
94c55f17 2118 struct stat st;
8c0faa1d 2119
3bd1e081 2120 DBG("Spawning consumerd");
c49dc785 2121
8c0faa1d
DG
2122 pid = fork();
2123 if (pid == 0) {
2124 /*
3bd1e081 2125 * Exec consumerd.
8c0faa1d 2126 */
daee5345 2127 if (opt_verbose_consumer) {
53086306
DG
2128 verbosity = "--verbose";
2129 } else {
2130 verbosity = "--quiet";
2131 }
3bd1e081
MD
2132 switch (consumer_data->type) {
2133 case LTTNG_CONSUMER_KERNEL:
94c55f17 2134 /*
c7704d57
DG
2135 * Find out which consumerd to execute. We will first try the
2136 * 64-bit path, then the sessiond's installation directory, and
2137 * fallback on the 32-bit one,
94c55f17 2138 */
63a799e8
AM
2139 DBG3("Looking for a kernel consumer at these locations:");
2140 DBG3(" 1) %s", consumerd64_bin);
2141 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2142 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2143 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2144 DBG3("Found location #1");
94c55f17 2145 consumer_to_use = consumerd64_bin;
94c55f17 2146 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2147 DBG3("Found location #2");
94c55f17 2148 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2149 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2150 DBG3("Found location #3");
eb1e0bd4 2151 consumer_to_use = consumerd32_bin;
94c55f17 2152 } else {
63a799e8 2153 DBG("Could not find any valid consumerd executable");
94c55f17
AM
2154 break;
2155 }
2156 DBG("Using kernel consumer at: %s", consumer_to_use);
2157 execl(consumer_to_use,
2158 "lttng-consumerd", verbosity, "-k",
2159 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2160 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2161 "--group", tracing_group_name,
94c55f17 2162 NULL);
3bd1e081 2163 break;
7753dea8
MD
2164 case LTTNG_CONSUMER64_UST:
2165 {
b1e0b6b6 2166 char *tmpnew = NULL;
8f4905da
MD
2167
2168 if (consumerd64_libdir[0] != '\0') {
2169 char *tmp;
2170 size_t tmplen;
2171
2172 tmp = getenv("LD_LIBRARY_PATH");
2173 if (!tmp) {
2174 tmp = "";
2175 }
2176 tmplen = strlen("LD_LIBRARY_PATH=")
2177 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2178 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2179 if (!tmpnew) {
2180 ret = -ENOMEM;
2181 goto error;
2182 }
2183 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2184 strcat(tmpnew, consumerd64_libdir);
2185 if (tmp[0] != '\0') {
2186 strcat(tmpnew, ":");
2187 strcat(tmpnew, tmp);
2188 }
2189 ret = putenv(tmpnew);
2190 if (ret) {
2191 ret = -errno;
c6f76da9 2192 free(tmpnew);
8f4905da
MD
2193 goto error;
2194 }
2195 }
94c55f17 2196 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2197 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2198 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2199 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2200 "--group", tracing_group_name,
7753dea8 2201 NULL);
8f4905da
MD
2202 if (consumerd64_libdir[0] != '\0') {
2203 free(tmpnew);
2204 }
2205 if (ret) {
2206 goto error;
2207 }
3bd1e081 2208 break;
7753dea8
MD
2209 }
2210 case LTTNG_CONSUMER32_UST:
2211 {
937dde8e 2212 char *tmpnew = NULL;
8f4905da
MD
2213
2214 if (consumerd32_libdir[0] != '\0') {
2215 char *tmp;
2216 size_t tmplen;
2217
2218 tmp = getenv("LD_LIBRARY_PATH");
2219 if (!tmp) {
2220 tmp = "";
2221 }
2222 tmplen = strlen("LD_LIBRARY_PATH=")
2223 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2224 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2225 if (!tmpnew) {
2226 ret = -ENOMEM;
2227 goto error;
2228 }
2229 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2230 strcat(tmpnew, consumerd32_libdir);
2231 if (tmp[0] != '\0') {
2232 strcat(tmpnew, ":");
2233 strcat(tmpnew, tmp);
2234 }
2235 ret = putenv(tmpnew);
2236 if (ret) {
2237 ret = -errno;
c6f76da9 2238 free(tmpnew);
8f4905da
MD
2239 goto error;
2240 }
2241 }
94c55f17 2242 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2243 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2244 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2245 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2246 "--group", tracing_group_name,
7753dea8 2247 NULL);
8f4905da
MD
2248 if (consumerd32_libdir[0] != '\0') {
2249 free(tmpnew);
2250 }
2251 if (ret) {
2252 goto error;
2253 }
7753dea8
MD
2254 break;
2255 }
3bd1e081 2256 default:
76d7553f 2257 PERROR("unknown consumer type");
3bd1e081
MD
2258 exit(EXIT_FAILURE);
2259 }
8c0faa1d 2260 if (errno != 0) {
76d7553f 2261 PERROR("kernel start consumer exec");
8c0faa1d
DG
2262 }
2263 exit(EXIT_FAILURE);
2264 } else if (pid > 0) {
2265 ret = pid;
8c0faa1d 2266 } else {
76d7553f 2267 PERROR("start consumer fork");
8c0faa1d 2268 ret = -errno;
8c0faa1d 2269 }
8f4905da 2270error:
8c0faa1d
DG
2271 return ret;
2272}
2273
693bd40b 2274/*
3bd1e081 2275 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2276 */
3bd1e081 2277static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2278{
c617c0c6 2279 int ret;
edb8b045
DG
2280
2281 /*
2282 * Set the listen() state on the socket since there is a possible race
2283 * between the exec() of the consumer daemon and this call if place in the
2284 * consumer thread. See bug #366 for more details.
2285 */
2286 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2287 if (ret < 0) {
2288 goto error;
2289 }
693bd40b 2290
3bd1e081
MD
2291 pthread_mutex_lock(&consumer_data->pid_mutex);
2292 if (consumer_data->pid != 0) {
2293 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2294 goto end;
2295 }
693bd40b 2296
3bd1e081 2297 ret = spawn_consumerd(consumer_data);
c49dc785 2298 if (ret < 0) {
3bd1e081
MD
2299 ERR("Spawning consumerd failed");
2300 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2301 goto error;
693bd40b 2302 }
c49dc785 2303
3bd1e081
MD
2304 /* Setting up the consumer_data pid */
2305 consumer_data->pid = ret;
48842b30 2306 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2307 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2308
3bd1e081
MD
2309 DBG2("Spawning consumer control thread");
2310 ret = spawn_consumer_thread(consumer_data);
693bd40b 2311 if (ret < 0) {
3bd1e081 2312 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2313 goto error;
2314 }
2315
c49dc785 2316end:
693bd40b
DG
2317 return 0;
2318
2319error:
331744e3 2320 /* Cleanup already created sockets on error. */
edb8b045 2321 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2322 int err;
2323
edb8b045
DG
2324 err = close(consumer_data->err_sock);
2325 if (err < 0) {
2326 PERROR("close consumer data error socket");
2327 }
2328 }
693bd40b
DG
2329 return ret;
2330}
2331
b73401da 2332/*
096102bd 2333 * Setup necessary data for kernel tracer action.
b73401da 2334 */
096102bd 2335static int init_kernel_tracer(void)
b73401da
DG
2336{
2337 int ret;
b73401da 2338
096102bd
DG
2339 /* Modprobe lttng kernel modules */
2340 ret = modprobe_lttng_control();
b73401da 2341 if (ret < 0) {
b73401da
DG
2342 goto error;
2343 }
2344
096102bd
DG
2345 /* Open debugfs lttng */
2346 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2347 if (kernel_tracer_fd < 0) {
2348 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2349 ret = -1;
2350 goto error_open;
54d01ffb
DG
2351 }
2352
2f77fc4b
DG
2353 /* Validate kernel version */
2354 ret = kernel_validate_version(kernel_tracer_fd);
2355 if (ret < 0) {
2356 goto error_version;
b551a063 2357 }
54d01ffb 2358
2f77fc4b
DG
2359 ret = modprobe_lttng_data();
2360 if (ret < 0) {
2361 goto error_modules;
54d01ffb
DG
2362 }
2363
2f77fc4b
DG
2364 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2365 return 0;
2366
2367error_version:
2368 modprobe_remove_lttng_control();
2369 ret = close(kernel_tracer_fd);
2370 if (ret) {
2371 PERROR("close");
b551a063 2372 }
2f77fc4b 2373 kernel_tracer_fd = -1;
f73fabfd 2374 return LTTNG_ERR_KERN_VERSION;
b551a063 2375
2f77fc4b
DG
2376error_modules:
2377 ret = close(kernel_tracer_fd);
2378 if (ret) {
2379 PERROR("close");
b551a063 2380 }
54d01ffb 2381
2f77fc4b
DG
2382error_open:
2383 modprobe_remove_lttng_control();
54d01ffb
DG
2384
2385error:
2f77fc4b
DG
2386 WARN("No kernel tracer available");
2387 kernel_tracer_fd = -1;
2388 if (!is_root) {
f73fabfd 2389 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2390 } else {
f73fabfd 2391 return LTTNG_ERR_KERN_NA;
2f77fc4b 2392 }
54d01ffb
DG
2393}
2394
2f77fc4b 2395
54d01ffb 2396/*
2f77fc4b
DG
2397 * Copy consumer output from the tracing session to the domain session. The
2398 * function also applies the right modification on a per domain basis for the
2399 * trace files destination directory.
36b588ed
MD
2400 *
2401 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2402 */
2f77fc4b 2403static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2404{
2405 int ret;
2f77fc4b
DG
2406 const char *dir_name;
2407 struct consumer_output *consumer;
2408
2409 assert(session);
2410 assert(session->consumer);
54d01ffb 2411
b551a063
DG
2412 switch (domain) {
2413 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2414 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2415 /*
2416 * XXX: We should audit the session creation and what this function
2417 * does "extra" in order to avoid a destroy since this function is used
2418 * in the domain session creation (kernel and ust) only. Same for UST
2419 * domain.
2420 */
2421 if (session->kernel_session->consumer) {
2422 consumer_destroy_output(session->kernel_session->consumer);
2423 }
2f77fc4b
DG
2424 session->kernel_session->consumer =
2425 consumer_copy_output(session->consumer);
2426 /* Ease our life a bit for the next part */
2427 consumer = session->kernel_session->consumer;
2428 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063 2429 break;
f20baf8e 2430 case LTTNG_DOMAIN_JUL:
b551a063 2431 case LTTNG_DOMAIN_UST:
2f77fc4b 2432 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2433 if (session->ust_session->consumer) {
2434 consumer_destroy_output(session->ust_session->consumer);
2435 }
2f77fc4b
DG
2436 session->ust_session->consumer =
2437 consumer_copy_output(session->consumer);
2438 /* Ease our life a bit for the next part */
2439 consumer = session->ust_session->consumer;
2440 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2441 break;
2442 default:
f73fabfd 2443 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2444 goto error;
2445 }
2446
2f77fc4b 2447 /* Append correct directory to subdir */
c30ce0b3
CB
2448 strncat(consumer->subdir, dir_name,
2449 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2450 DBG3("Copy session consumer subdir %s", consumer->subdir);
2451
f73fabfd 2452 ret = LTTNG_OK;
54d01ffb
DG
2453
2454error:
2455 return ret;
2456}
2457
00e2e675 2458/*
2f77fc4b 2459 * Create an UST session and add it to the session ust list.
36b588ed
MD
2460 *
2461 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2462 */
2f77fc4b
DG
2463static int create_ust_session(struct ltt_session *session,
2464 struct lttng_domain *domain)
00e2e675
DG
2465{
2466 int ret;
2f77fc4b 2467 struct ltt_ust_session *lus = NULL;
00e2e675 2468
a4b92340 2469 assert(session);
2f77fc4b
DG
2470 assert(domain);
2471 assert(session->consumer);
a4b92340 2472
2f77fc4b 2473 switch (domain->type) {
f20baf8e 2474 case LTTNG_DOMAIN_JUL:
2f77fc4b
DG
2475 case LTTNG_DOMAIN_UST:
2476 break;
2477 default:
2478 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2479 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2480 goto error;
2481 }
2482
2f77fc4b
DG
2483 DBG("Creating UST session");
2484
dec56f6c 2485 lus = trace_ust_create_session(session->id);
2f77fc4b 2486 if (lus == NULL) {
f73fabfd 2487 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2488 goto error;
2489 }
2490
2f77fc4b
DG
2491 lus->uid = session->uid;
2492 lus->gid = session->gid;
2bba9e53 2493 lus->output_traces = session->output_traces;
27babd3a 2494 lus->snapshot_mode = session->snapshot_mode;
ecc48a90 2495 lus->live_timer_interval = session->live_timer;
2f77fc4b 2496 session->ust_session = lus;
00e2e675 2497
2f77fc4b
DG
2498 /* Copy session output to the newly created UST session */
2499 ret = copy_session_consumer(domain->type, session);
f73fabfd 2500 if (ret != LTTNG_OK) {
00e2e675
DG
2501 goto error;
2502 }
2503
f73fabfd 2504 return LTTNG_OK;
00e2e675
DG
2505
2506error:
2f77fc4b
DG
2507 free(lus);
2508 session->ust_session = NULL;
00e2e675
DG
2509 return ret;
2510}
2511
2512/*
2f77fc4b 2513 * Create a kernel tracer session then create the default channel.
00e2e675 2514 */
2f77fc4b 2515static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2516{
2517 int ret;
a4b92340 2518
2f77fc4b 2519 DBG("Creating kernel session");
00e2e675 2520
2f77fc4b
DG
2521 ret = kernel_create_session(session, kernel_tracer_fd);
2522 if (ret < 0) {
f73fabfd 2523 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2524 goto error;
2525 }
2526
2f77fc4b
DG
2527 /* Code flow safety */
2528 assert(session->kernel_session);
2529
2530 /* Copy session output to the newly created Kernel session */
2531 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2532 if (ret != LTTNG_OK) {
a4b92340
DG
2533 goto error;
2534 }
2535
2f77fc4b
DG
2536 /* Create directory(ies) on local filesystem. */
2537 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2538 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2539 ret = run_as_mkdir_recursive(
2540 session->kernel_session->consumer->dst.trace_path,
2541 S_IRWXU | S_IRWXG, session->uid, session->gid);
2542 if (ret < 0) {
2543 if (ret != -EEXIST) {
2544 ERR("Trace directory creation error");
00e2e675
DG
2545 goto error;
2546 }
00e2e675 2547 }
2f77fc4b 2548 }
00e2e675 2549
2f77fc4b
DG
2550 session->kernel_session->uid = session->uid;
2551 session->kernel_session->gid = session->gid;
2bba9e53 2552 session->kernel_session->output_traces = session->output_traces;
27babd3a 2553 session->kernel_session->snapshot_mode = session->snapshot_mode;
00e2e675 2554
f73fabfd 2555 return LTTNG_OK;
00e2e675 2556
2f77fc4b
DG
2557error:
2558 trace_kernel_destroy_session(session->kernel_session);
2559 session->kernel_session = NULL;
2560 return ret;
2561}
00e2e675 2562
2f77fc4b
DG
2563/*
2564 * Count number of session permitted by uid/gid.
2565 */
2566static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2567{
2568 unsigned int i = 0;
2569 struct ltt_session *session;
07424f16 2570
2f77fc4b
DG
2571 DBG("Counting number of available session for UID %d GID %d",
2572 uid, gid);
2573 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2574 /*
2f77fc4b 2575 * Only list the sessions the user can control.
00e2e675 2576 */
2f77fc4b
DG
2577 if (!session_access_ok(session, uid, gid)) {
2578 continue;
2579 }
2580 i++;
a4b92340 2581 }
2f77fc4b 2582 return i;
00e2e675
DG
2583}
2584
54d01ffb
DG
2585/*
2586 * Process the command requested by the lttng client within the command
2587 * context structure. This function make sure that the return structure (llm)
2588 * is set and ready for transmission before returning.
2589 *
2590 * Return any error encountered or 0 for success.
53a80697
MD
2591 *
2592 * "sock" is only used for special-case var. len data.
36b588ed
MD
2593 *
2594 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2595 */
53a80697
MD
2596static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2597 int *sock_error)
54d01ffb 2598{
f73fabfd 2599 int ret = LTTNG_OK;
44d3bd01 2600 int need_tracing_session = 1;
2e09ba09 2601 int need_domain;
54d01ffb
DG
2602
2603 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2604
53a80697
MD
2605 *sock_error = 0;
2606
2e09ba09
MD
2607 switch (cmd_ctx->lsm->cmd_type) {
2608 case LTTNG_CREATE_SESSION:
27babd3a 2609 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2610 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09
MD
2611 case LTTNG_DESTROY_SESSION:
2612 case LTTNG_LIST_SESSIONS:
2613 case LTTNG_LIST_DOMAINS:
2614 case LTTNG_START_TRACE:
2615 case LTTNG_STOP_TRACE:
6d805429 2616 case LTTNG_DATA_PENDING:
da3c9ec1
DG
2617 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2618 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2619 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2620 case LTTNG_SNAPSHOT_RECORD:
2e09ba09 2621 need_domain = 0;
3aace903 2622 break;
2e09ba09
MD
2623 default:
2624 need_domain = 1;
2625 }
2626
2627 if (opt_no_kernel && need_domain
2628 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2629 if (!is_root) {
f73fabfd 2630 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2631 } else {
f73fabfd 2632 ret = LTTNG_ERR_KERN_NA;
531d29f9 2633 }
4fba7219
DG
2634 goto error;
2635 }
2636
8d3113b2
DG
2637 /* Deny register consumer if we already have a spawned consumer. */
2638 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2639 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2640 if (kconsumer_data.pid > 0) {
f73fabfd 2641 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2642 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2643 goto error;
2644 }
2645 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2646 }
2647
54d01ffb
DG
2648 /*
2649 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2650 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2651 * command.
2652 */
2653 switch(cmd_ctx->lsm->cmd_type) {
2654 case LTTNG_LIST_SESSIONS:
2655 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2656 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2657 case LTTNG_LIST_DOMAINS:
2658 case LTTNG_LIST_CHANNELS:
2659 case LTTNG_LIST_EVENTS:
2660 break;
2661 default:
2662 /* Setup lttng message with no payload */
2663 ret = setup_lttng_msg(cmd_ctx, 0);
2664 if (ret < 0) {
2665 /* This label does not try to unlock the session */
2666 goto init_setup_error;
2667 }
2668 }
2669
2670 /* Commands that DO NOT need a session. */
2671 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2672 case LTTNG_CREATE_SESSION:
27babd3a 2673 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2674 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09 2675 case LTTNG_CALIBRATE:
54d01ffb
DG
2676 case LTTNG_LIST_SESSIONS:
2677 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2678 case LTTNG_LIST_TRACEPOINT_FIELDS:
44d3bd01 2679 need_tracing_session = 0;
54d01ffb
DG
2680 break;
2681 default:
2682 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2683 /*
2684 * We keep the session list lock across _all_ commands
2685 * for now, because the per-session lock does not
2686 * handle teardown properly.
2687 */
74babd95 2688 session_lock_list();
54d01ffb
DG
2689 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2690 if (cmd_ctx->session == NULL) {
bba2d65f 2691 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2692 goto error;
2693 } else {
2694 /* Acquire lock for the session */
2695 session_lock(cmd_ctx->session);
2696 }
2697 break;
2698 }
b389abbe 2699
2e09ba09
MD
2700 if (!need_domain) {
2701 goto skip_domain;
2702 }
a4b92340 2703
54d01ffb
DG
2704 /*
2705 * Check domain type for specific "pre-action".
2706 */
2707 switch (cmd_ctx->lsm->domain.type) {
2708 case LTTNG_DOMAIN_KERNEL:
d1f1c568 2709 if (!is_root) {
f73fabfd 2710 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
2711 goto error;
2712 }
2713
54d01ffb 2714 /* Kernel tracer check */
a4b35e07 2715 if (kernel_tracer_fd == -1) {
54d01ffb 2716 /* Basically, load kernel tracer modules */
096102bd
DG
2717 ret = init_kernel_tracer();
2718 if (ret != 0) {
54d01ffb
DG
2719 goto error;
2720 }
2721 }
5eb91c98 2722
5c827ce0
DG
2723 /* Consumer is in an ERROR state. Report back to client */
2724 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2725 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2726 goto error;
2727 }
2728
54d01ffb 2729 /* Need a session for kernel command */
44d3bd01 2730 if (need_tracing_session) {
54d01ffb 2731 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 2732 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 2733 if (ret < 0) {
f73fabfd 2734 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
2735 goto error;
2736 }
b389abbe 2737 }
7d29a247 2738
54d01ffb 2739 /* Start the kernel consumer daemon */
3bd1e081
MD
2740 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2741 if (kconsumer_data.pid == 0 &&
785d2d0d 2742 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3bd1e081
MD
2743 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2744 ret = start_consumerd(&kconsumer_data);
7d29a247 2745 if (ret < 0) {
f73fabfd 2746 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 2747 goto error;
950131af 2748 }
5c827ce0 2749 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
2750 } else {
2751 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 2752 }
173af62f 2753
a4b92340
DG
2754 /*
2755 * The consumer was just spawned so we need to add the socket to
2756 * the consumer output of the session if exist.
2757 */
2758 ret = consumer_create_socket(&kconsumer_data,
2759 cmd_ctx->session->kernel_session->consumer);
2760 if (ret < 0) {
2761 goto error;
173af62f 2762 }
0d0c377a 2763 }
5c827ce0 2764
54d01ffb 2765 break;
b9dfb167 2766 case LTTNG_DOMAIN_JUL:
2bdd86d4 2767 case LTTNG_DOMAIN_UST:
44d3bd01 2768 {
b51ec5b4
MD
2769 if (!ust_app_supported()) {
2770 ret = LTTNG_ERR_NO_UST;
2771 goto error;
2772 }
5c827ce0
DG
2773 /* Consumer is in an ERROR state. Report back to client */
2774 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2775 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2776 goto error;
2777 }
2778
44d3bd01 2779 if (need_tracing_session) {
a4b92340 2780 /* Create UST session if none exist. */
f6a9efaa 2781 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 2782 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 2783 &cmd_ctx->lsm->domain);
f73fabfd 2784 if (ret != LTTNG_OK) {
44d3bd01
DG
2785 goto error;
2786 }
2787 }
00e2e675 2788
7753dea8
MD
2789 /* Start the UST consumer daemons */
2790 /* 64-bit */
2791 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 2792 if (consumerd64_bin[0] != '\0' &&
7753dea8 2793 ustconsumer64_data.pid == 0 &&
785d2d0d 2794 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2795 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2796 ret = start_consumerd(&ustconsumer64_data);
2bdd86d4 2797 if (ret < 0) {
f73fabfd 2798 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
173af62f 2799 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2bdd86d4
MD
2800 goto error;
2801 }
48842b30 2802
173af62f 2803 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
5c827ce0 2804 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3ff2ecac 2805 } else {
7753dea8
MD
2806 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2807 }
173af62f
DG
2808
2809 /*
2810 * Setup socket for consumer 64 bit. No need for atomic access
2811 * since it was set above and can ONLY be set in this thread.
2812 */
a4b92340
DG
2813 ret = consumer_create_socket(&ustconsumer64_data,
2814 cmd_ctx->session->ust_session->consumer);
2815 if (ret < 0) {
2816 goto error;
173af62f
DG
2817 }
2818
7753dea8 2819 /* 32-bit */
fc7a59ce 2820 if (consumerd32_bin[0] != '\0' &&
7753dea8 2821 ustconsumer32_data.pid == 0 &&
785d2d0d 2822 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2823 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2824 ret = start_consumerd(&ustconsumer32_data);
2825 if (ret < 0) {
f73fabfd 2826 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
173af62f 2827 uatomic_set(&ust_consumerd32_fd, -EINVAL);
7753dea8
MD
2828 goto error;
2829 }
5c827ce0 2830
173af62f 2831 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
5c827ce0 2832 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
7753dea8
MD
2833 } else {
2834 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2bdd86d4 2835 }
173af62f
DG
2836
2837 /*
2838 * Setup socket for consumer 64 bit. No need for atomic access
2839 * since it was set above and can ONLY be set in this thread.
2840 */
a4b92340
DG
2841 ret = consumer_create_socket(&ustconsumer32_data,
2842 cmd_ctx->session->ust_session->consumer);
2843 if (ret < 0) {
2844 goto error;
173af62f 2845 }
44d3bd01
DG
2846 }
2847 break;
48842b30 2848 }
54d01ffb 2849 default:
54d01ffb
DG
2850 break;
2851 }
2e09ba09 2852skip_domain:
33a2b854 2853
5c827ce0
DG
2854 /* Validate consumer daemon state when start/stop trace command */
2855 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2856 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2857 switch (cmd_ctx->lsm->domain.type) {
b9dfb167 2858 case LTTNG_DOMAIN_JUL:
5c827ce0
DG
2859 case LTTNG_DOMAIN_UST:
2860 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2861 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2862 goto error;
2863 }
2864 break;
2865 case LTTNG_DOMAIN_KERNEL:
2866 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2867 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2868 goto error;
2869 }
2870 break;
2871 }
2872 }
2873
8e0af1b4
MD
2874 /*
2875 * Check that the UID or GID match that of the tracing session.
2876 * The root user can interact with all sessions.
2877 */
2878 if (need_tracing_session) {
2879 if (!session_access_ok(cmd_ctx->session,
730389d9
DG
2880 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2881 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
f73fabfd 2882 ret = LTTNG_ERR_EPERM;
8e0af1b4
MD
2883 goto error;
2884 }
2885 }
2886
ffe60014
DG
2887 /*
2888 * Send relayd information to consumer as soon as we have a domain and a
2889 * session defined.
2890 */
2891 if (cmd_ctx->session && need_domain) {
2892 /*
2893 * Setup relayd if not done yet. If the relayd information was already
2894 * sent to the consumer, this call will gracefully return.
2895 */
2896 ret = cmd_setup_relayd(cmd_ctx->session);
2897 if (ret != LTTNG_OK) {
2898 goto error;
2899 }
2900 }
2901
54d01ffb
DG
2902 /* Process by command type */
2903 switch (cmd_ctx->lsm->cmd_type) {
2904 case LTTNG_ADD_CONTEXT:
2905 {
2906 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2907 cmd_ctx->lsm->u.context.channel_name,
979e618e 2908 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
54d01ffb
DG
2909 break;
2910 }
2911 case LTTNG_DISABLE_CHANNEL:
2912 {
2913 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2914 cmd_ctx->lsm->u.disable.channel_name);
2915 break;
2916 }
2917 case LTTNG_DISABLE_EVENT:
2918 {
2919 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2920 cmd_ctx->lsm->u.disable.channel_name,
2921 cmd_ctx->lsm->u.disable.name);
33a2b854
DG
2922 break;
2923 }
54d01ffb
DG
2924 case LTTNG_DISABLE_ALL_EVENT:
2925 {
2bdd86d4 2926 DBG("Disabling all events");
54d01ffb
DG
2927
2928 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2929 cmd_ctx->lsm->u.disable.channel_name);
2930 break;
2931 }
2932 case LTTNG_ENABLE_CHANNEL:
2933 {
7972aab2 2934 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2f77fc4b 2935 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
54d01ffb
DG
2936 break;
2937 }
2938 case LTTNG_ENABLE_EVENT:
2939 {
882ef835
JI
2940 struct lttng_event_exclusion *exclusion = NULL;
2941 struct lttng_filter_bytecode *bytecode = NULL;
2942
67676bd8
DG
2943 /* Handle exclusion events and receive it from the client. */
2944 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
2945 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
2946
2947 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
2948 (count * LTTNG_SYMBOL_NAME_LEN));
2949 if (!exclusion) {
2950 ret = LTTNG_ERR_EXCLUSION_NOMEM;
2951 goto error;
882ef835 2952 }
882ef835 2953
67676bd8
DG
2954 DBG("Receiving var len exclusion event list from client ...");
2955 exclusion->count = count;
2956 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
2957 count * LTTNG_SYMBOL_NAME_LEN);
2958 if (ret <= 0) {
2959 DBG("Nothing recv() from client var len data... continuing");
2960 *sock_error = 1;
2961 free(exclusion);
2962 ret = LTTNG_ERR_EXCLUSION_INVAL;
2963 goto error;
2964 }
2965 }
2966
2967 /* Handle filter and get bytecode from client. */
2968 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
2969 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
2970
2971 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
2972 ret = LTTNG_ERR_FILTER_INVAL;
2973 free(exclusion);
2974 goto error;
2975 }
2976
2977 bytecode = zmalloc(bytecode_len);
2978 if (!bytecode) {
2979 free(exclusion);
2980 ret = LTTNG_ERR_FILTER_NOMEM;
2981 goto error;
882ef835
JI
2982 }
2983
67676bd8
DG
2984 /* Receive var. len. data */
2985 DBG("Receiving var len filter's bytecode from client ...");
2986 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
2987 if (ret <= 0) {
2988 DBG("Nothing recv() from client car len data... continuing");
2989 *sock_error = 1;
2990 free(bytecode);
2991 free(exclusion);
2992 ret = LTTNG_ERR_FILTER_INVAL;
2993 goto error;
2994 }
2995
2996 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
2997 free(bytecode);
2998 free(exclusion);
2999 ret = LTTNG_ERR_FILTER_INVAL;
3000 goto error;
3001 }
882ef835 3002 }
67676bd8
DG
3003
3004 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3005 cmd_ctx->lsm->u.enable.channel_name,
3006 &cmd_ctx->lsm->u.enable.event, bytecode, exclusion,
3007 kernel_poll_pipe[1]);
54d01ffb
DG
3008 break;
3009 }
3010 case LTTNG_ENABLE_ALL_EVENT:
3011 {
2bdd86d4 3012 DBG("Enabling all events");
54d01ffb 3013
7972aab2 3014 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
8c9ae521 3015 cmd_ctx->lsm->u.enable.channel_name,
025faf73 3016 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
3017 break;
3018 }
052da939 3019 case LTTNG_LIST_TRACEPOINTS:
2ef84c95 3020 {
9f19cc17 3021 struct lttng_event *events;
54d01ffb 3022 ssize_t nb_events;
052da939 3023
54d01ffb
DG
3024 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
3025 if (nb_events < 0) {
f73fabfd 3026 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3027 ret = -nb_events;
3028 goto error;
2ef84c95
DG
3029 }
3030
3031 /*
3032 * Setup lttng message with payload size set to the event list size in
3033 * bytes and then copy list into the llm payload.
3034 */
052da939 3035 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2ef84c95 3036 if (ret < 0) {
052da939 3037 free(events);
2ef84c95
DG
3038 goto setup_error;
3039 }
3040
3041 /* Copy event list into message payload */
9f19cc17 3042 memcpy(cmd_ctx->llm->payload, events,
052da939 3043 sizeof(struct lttng_event) * nb_events);
2ef84c95 3044
9f19cc17 3045 free(events);
2ef84c95 3046
f73fabfd 3047 ret = LTTNG_OK;
2ef84c95
DG
3048 break;
3049 }
f37d259d
MD
3050 case LTTNG_LIST_TRACEPOINT_FIELDS:
3051 {
3052 struct lttng_event_field *fields;
3053 ssize_t nb_fields;
3054
a4b92340
DG
3055 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3056 &fields);
f37d259d 3057 if (nb_fields < 0) {
f73fabfd 3058 /* Return value is a negative lttng_error_code. */
f37d259d
MD
3059 ret = -nb_fields;
3060 goto error;
3061 }
3062
3063 /*
3064 * Setup lttng message with payload size set to the event list size in
3065 * bytes and then copy list into the llm payload.
3066 */
a4b92340
DG
3067 ret = setup_lttng_msg(cmd_ctx,
3068 sizeof(struct lttng_event_field) * nb_fields);
f37d259d
MD
3069 if (ret < 0) {
3070 free(fields);
3071 goto setup_error;
3072 }
3073
3074 /* Copy event list into message payload */
3075 memcpy(cmd_ctx->llm->payload, fields,
3076 sizeof(struct lttng_event_field) * nb_fields);
3077
3078 free(fields);
3079
f73fabfd 3080 ret = LTTNG_OK;
f37d259d
MD
3081 break;
3082 }
00e2e675
DG
3083 case LTTNG_SET_CONSUMER_URI:
3084 {
a4b92340
DG
3085 size_t nb_uri, len;
3086 struct lttng_uri *uris;
3087
3088 nb_uri = cmd_ctx->lsm->u.uri.size;
3089 len = nb_uri * sizeof(struct lttng_uri);
3090
3091 if (nb_uri == 0) {
f73fabfd 3092 ret = LTTNG_ERR_INVALID;
a4b92340
DG
3093 goto error;
3094 }
3095
3096 uris = zmalloc(len);
3097 if (uris == NULL) {
f73fabfd 3098 ret = LTTNG_ERR_FATAL;
a4b92340
DG
3099 goto error;
3100 }
3101
3102 /* Receive variable len data */
77c7c900 3103 DBG("Receiving %zu URI(s) from client ...", nb_uri);
a4b92340
DG
3104 ret = lttcomm_recv_unix_sock(sock, uris, len);
3105 if (ret <= 0) {
3106 DBG("No URIs received from client... continuing");
3107 *sock_error = 1;
f73fabfd 3108 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3109 free(uris);
a4b92340
DG
3110 goto error;
3111 }
3112
00e2e675 3113 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
a4b92340 3114 nb_uri, uris);
f73fabfd 3115 if (ret != LTTNG_OK) {
b1d41407 3116 free(uris);
a4b92340
DG
3117 goto error;
3118 }
3119
3120 /*
3121 * XXX: 0 means that this URI should be applied on the session. Should
3122 * be a DOMAIN enuam.
3123 */
3124 if (cmd_ctx->lsm->domain.type == 0) {
3125 /* Add the URI for the UST session if a consumer is present. */
3126 if (cmd_ctx->session->ust_session &&
3127 cmd_ctx->session->ust_session->consumer) {
3128 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3129 nb_uri, uris);
3130 } else if (cmd_ctx->session->kernel_session &&
3131 cmd_ctx->session->kernel_session->consumer) {
3132 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3133 cmd_ctx->session, nb_uri, uris);
3134 }
3135 }
3136
b1d41407
DG
3137 free(uris);
3138
00e2e675
DG
3139 break;
3140 }
f3ed775e 3141 case LTTNG_START_TRACE:
8c0faa1d 3142 {
54d01ffb 3143 ret = cmd_start_trace(cmd_ctx->session);
8c0faa1d
DG
3144 break;
3145 }
f3ed775e 3146 case LTTNG_STOP_TRACE:
8c0faa1d 3147 {
54d01ffb 3148 ret = cmd_stop_trace(cmd_ctx->session);
8c0faa1d
DG
3149 break;
3150 }
5e16da05
MD
3151 case LTTNG_CREATE_SESSION:
3152 {
a4b92340
DG
3153 size_t nb_uri, len;
3154 struct lttng_uri *uris = NULL;
3155
3156 nb_uri = cmd_ctx->lsm->u.uri.size;
3157 len = nb_uri * sizeof(struct lttng_uri);
3158
3159 if (nb_uri > 0) {
3160 uris = zmalloc(len);
3161 if (uris == NULL) {
f73fabfd 3162 ret = LTTNG_ERR_FATAL;
a4b92340
DG
3163 goto error;
3164 }
3165
3166 /* Receive variable len data */
77c7c900 3167 DBG("Waiting for %zu URIs from client ...", nb_uri);
a4b92340
DG
3168 ret = lttcomm_recv_unix_sock(sock, uris, len);
3169 if (ret <= 0) {
3170 DBG("No URIs received from client... continuing");
3171 *sock_error = 1;
f73fabfd 3172 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3173 free(uris);
a4b92340
DG
3174 goto error;
3175 }
3176
3177 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3178 DBG("Creating session with ONE network URI is a bad call");
f73fabfd 3179 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3180 free(uris);
a4b92340
DG
3181 goto error;
3182 }
3183 }
3184
3185 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
ecc48a90 3186 &cmd_ctx->creds, 0);
a4b92340 3187
b1d41407
DG
3188 free(uris);
3189
00e2e675
DG
3190 break;
3191 }
5e16da05
MD
3192 case LTTNG_DESTROY_SESSION:
3193 {
2f77fc4b 3194 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
a4b92340
DG
3195
3196 /* Set session to NULL so we do not unlock it after free. */
256a5576 3197 cmd_ctx->session = NULL;
5461b305 3198 break;
5e16da05 3199 }
9f19cc17 3200 case LTTNG_LIST_DOMAINS:
5e16da05 3201 {
54d01ffb
DG
3202 ssize_t nb_dom;
3203 struct lttng_domain *domains;
5461b305 3204
54d01ffb
DG
3205 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3206 if (nb_dom < 0) {
f73fabfd 3207 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3208 ret = -nb_dom;
3209 goto error;
ce3d728c 3210 }
520ff687 3211
54d01ffb 3212 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
5461b305 3213 if (ret < 0) {
ae9c20bc 3214 free(domains);
5461b305 3215 goto setup_error;
520ff687 3216 }
57167058 3217
54d01ffb
DG
3218 /* Copy event list into message payload */
3219 memcpy(cmd_ctx->llm->payload, domains,
3220 nb_dom * sizeof(struct lttng_domain));
3221
3222 free(domains);
5e16da05 3223
f73fabfd 3224 ret = LTTNG_OK;
5e16da05
MD
3225 break;
3226 }
9f19cc17 3227 case LTTNG_LIST_CHANNELS:
5e16da05 3228 {
6775595e 3229 int nb_chan;
54d01ffb
DG
3230 struct lttng_channel *channels;
3231
b551a063
DG
3232 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3233 cmd_ctx->session, &channels);
54d01ffb 3234 if (nb_chan < 0) {
f73fabfd 3235 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3236 ret = -nb_chan;
3237 goto error;
5461b305 3238 }
ca95a216 3239
54d01ffb 3240 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
5461b305 3241 if (ret < 0) {
ae9c20bc 3242 free(channels);
5461b305
DG
3243 goto setup_error;
3244 }
9f19cc17 3245
54d01ffb
DG
3246 /* Copy event list into message payload */
3247 memcpy(cmd_ctx->llm->payload, channels,
3248 nb_chan * sizeof(struct lttng_channel));
3249
3250 free(channels);
9f19cc17 3251
f73fabfd 3252 ret = LTTNG_OK;
5461b305 3253 break;
5e16da05 3254 }
9f19cc17 3255 case LTTNG_LIST_EVENTS:
5e16da05 3256 {
b551a063 3257 ssize_t nb_event;
684d34d2 3258 struct lttng_event *events = NULL;
9f19cc17 3259
b551a063 3260 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
54d01ffb
DG
3261 cmd_ctx->lsm->u.list.channel_name, &events);
3262 if (nb_event < 0) {
f73fabfd 3263 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3264 ret = -nb_event;
3265 goto error;
5461b305 3266 }
ca95a216 3267
54d01ffb 3268 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
5461b305 3269 if (ret < 0) {
ae9c20bc 3270 free(events);
5461b305
DG
3271 goto setup_error;
3272 }
9f19cc17 3273
54d01ffb
DG
3274 /* Copy event list into message payload */
3275 memcpy(cmd_ctx->llm->payload, events,
3276 nb_event * sizeof(struct lttng_event));
9f19cc17 3277
54d01ffb 3278 free(events);
9f19cc17 3279
f73fabfd 3280 ret = LTTNG_OK;
5461b305 3281 break;
5e16da05
MD
3282 }
3283 case LTTNG_LIST_SESSIONS:
3284 {
8e0af1b4 3285 unsigned int nr_sessions;
5461b305 3286
8e0af1b4 3287 session_lock_list();
730389d9
DG
3288 nr_sessions = lttng_sessions_count(
3289 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3290 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
d32fb093 3291
8e0af1b4 3292 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
5461b305 3293 if (ret < 0) {
54d01ffb 3294 session_unlock_list();
5461b305 3295 goto setup_error;
e065084a 3296 }
5e16da05 3297
6c9cc2ab 3298 /* Filled the session array */
2f77fc4b 3299 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
730389d9
DG
3300 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3301 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
6c9cc2ab 3302
54d01ffb 3303 session_unlock_list();
5e16da05 3304
f73fabfd 3305 ret = LTTNG_OK;
5e16da05
MD
3306 break;
3307 }
d0254c7c
MD
3308 case LTTNG_CALIBRATE:
3309 {
54d01ffb
DG
3310 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3311 &cmd_ctx->lsm->u.calibrate);
d0254c7c
MD
3312 break;
3313 }
d9800920
DG
3314 case LTTNG_REGISTER_CONSUMER:
3315 {
2f77fc4b
DG
3316 struct consumer_data *cdata;
3317
3318 switch (cmd_ctx->lsm->domain.type) {
3319 case LTTNG_DOMAIN_KERNEL:
3320 cdata = &kconsumer_data;
3321 break;
3322 default:
f73fabfd 3323 ret = LTTNG_ERR_UND;
2f77fc4b
DG
3324 goto error;
3325 }
3326
54d01ffb 3327 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 3328 cmd_ctx->lsm->u.reg.path, cdata);
d9800920
DG
3329 break;
3330 }
6d805429 3331 case LTTNG_DATA_PENDING:
806e2684 3332 {
6d805429 3333 ret = cmd_data_pending(cmd_ctx->session);
806e2684
DG
3334 break;
3335 }
da3c9ec1
DG
3336 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3337 {
6dc3064a
DG
3338 struct lttcomm_lttng_output_id reply;
3339
3340 ret = cmd_snapshot_add_output(cmd_ctx->session,
3341 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3342 if (ret != LTTNG_OK) {
3343 goto error;
3344 }
3345
3346 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3347 if (ret < 0) {
3348 goto setup_error;
3349 }
3350
3351 /* Copy output list into message payload */
3352 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3353 ret = LTTNG_OK;
da3c9ec1
DG
3354 break;
3355 }
3356 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3357 {
6dc3064a
DG
3358 ret = cmd_snapshot_del_output(cmd_ctx->session,
3359 &cmd_ctx->lsm->u.snapshot_output.output);
da3c9ec1
DG
3360 break;
3361 }
3362 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3363 {
6dc3064a
DG
3364 ssize_t nb_output;
3365 struct lttng_snapshot_output *outputs = NULL;
3366
3367 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3368 if (nb_output < 0) {
3369 ret = -nb_output;
3370 goto error;
3371 }
3372
3373 ret = setup_lttng_msg(cmd_ctx,
3374 nb_output * sizeof(struct lttng_snapshot_output));
3375 if (ret < 0) {
3376 free(outputs);
3377 goto setup_error;
3378 }
3379
3380 if (outputs) {
3381 /* Copy output list into message payload */
3382 memcpy(cmd_ctx->llm->payload, outputs,
3383 nb_output * sizeof(struct lttng_snapshot_output));
3384 free(outputs);
3385 }
3386
3387 ret = LTTNG_OK;
da3c9ec1
DG
3388 break;
3389 }
3390 case LTTNG_SNAPSHOT_RECORD:
3391 {
6dc3064a
DG
3392 ret = cmd_snapshot_record(cmd_ctx->session,
3393 &cmd_ctx->lsm->u.snapshot_record.output,
3394 cmd_ctx->lsm->u.snapshot_record.wait);
da3c9ec1
DG
3395 break;
3396 }
27babd3a
DG
3397 case LTTNG_CREATE_SESSION_SNAPSHOT:
3398 {
3399 size_t nb_uri, len;
3400 struct lttng_uri *uris = NULL;
3401
3402 nb_uri = cmd_ctx->lsm->u.uri.size;
3403 len = nb_uri * sizeof(struct lttng_uri);
3404
3405 if (nb_uri > 0) {
3406 uris = zmalloc(len);
3407 if (uris == NULL) {
3408 ret = LTTNG_ERR_FATAL;
3409 goto error;
3410 }
3411
3412 /* Receive variable len data */
3413 DBG("Waiting for %zu URIs from client ...", nb_uri);
3414 ret = lttcomm_recv_unix_sock(sock, uris, len);
3415 if (ret <= 0) {
3416 DBG("No URIs received from client... continuing");
3417 *sock_error = 1;
3418 ret = LTTNG_ERR_SESSION_FAIL;
3419 free(uris);
3420 goto error;
3421 }
3422
3423 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3424 DBG("Creating session with ONE network URI is a bad call");
3425 ret = LTTNG_ERR_SESSION_FAIL;
3426 free(uris);
3427 goto error;
3428 }
3429 }
3430
3431 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3432 nb_uri, &cmd_ctx->creds);
3433 free(uris);
3434 break;
3435 }
ecc48a90
JD
3436 case LTTNG_CREATE_SESSION_LIVE:
3437 {
3438 size_t nb_uri, len;
3439 struct lttng_uri *uris = NULL;
3440
3441 nb_uri = cmd_ctx->lsm->u.uri.size;
3442 len = nb_uri * sizeof(struct lttng_uri);
3443
3444 if (nb_uri > 0) {
3445 uris = zmalloc(len);
3446 if (uris == NULL) {
3447 ret = LTTNG_ERR_FATAL;
3448 goto error;
3449 }
3450
3451 /* Receive variable len data */
3452 DBG("Waiting for %zu URIs from client ...", nb_uri);
3453 ret = lttcomm_recv_unix_sock(sock, uris, len);
3454 if (ret <= 0) {
3455 DBG("No URIs received from client... continuing");
3456 *sock_error = 1;
3457 ret = LTTNG_ERR_SESSION_FAIL;
3458 free(uris);
3459 goto error;
3460 }
3461
3462 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3463 DBG("Creating session with ONE network URI is a bad call");
3464 ret = LTTNG_ERR_SESSION_FAIL;
3465 free(uris);
3466 goto error;
3467 }
3468 }
3469
3470 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3471 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3472 free(uris);
3473 break;
3474 }
5e16da05 3475 default:
f73fabfd 3476 ret = LTTNG_ERR_UND;
5461b305 3477 break;
fac6795d
DG
3478 }
3479
5461b305 3480error:
5461b305
DG
3481 if (cmd_ctx->llm == NULL) {
3482 DBG("Missing llm structure. Allocating one.");
894be886 3483 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
5461b305
DG
3484 goto setup_error;
3485 }
3486 }
54d01ffb 3487 /* Set return code */
5461b305 3488 cmd_ctx->llm->ret_code = ret;
5461b305 3489setup_error:
b5541356 3490 if (cmd_ctx->session) {
54d01ffb 3491 session_unlock(cmd_ctx->session);
b5541356 3492 }
256a5576
MD
3493 if (need_tracing_session) {
3494 session_unlock_list();
3495 }
54d01ffb 3496init_setup_error:
8028d920 3497 return ret;
fac6795d
DG
3498}
3499
44a5e5eb
DG
3500/*
3501 * Thread managing health check socket.
3502 */
3503static void *thread_manage_health(void *data)
3504{
eb4a2943 3505 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
44a5e5eb
DG
3506 uint32_t revents, nb_fd;
3507 struct lttng_poll_event events;
0c89d795
MD
3508 struct health_comm_msg msg;
3509 struct health_comm_reply reply;
44a5e5eb
DG
3510
3511 DBG("[thread] Manage health check started");
3512
3513 rcu_register_thread();
3514
6d737ce4
DG
3515 /* We might hit an error path before this is created. */
3516 lttng_poll_init(&events);
3cc04881 3517
44a5e5eb
DG
3518 /* Create unix socket */
3519 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3520 if (sock < 0) {
3521 ERR("Unable to create health check Unix socket");
3522 ret = -1;
3523 goto error;
3524 }
3525
6c71277b
MD
3526 if (is_root) {
3527 /* lttng health client socket path permissions */
3528 ret = chown(health_unix_sock_path, 0,
3529 utils_get_group_id(tracing_group_name));
3530 if (ret < 0) {
3531 ERR("Unable to set group on %s", health_unix_sock_path);
3532 PERROR("chown");
3533 ret = -1;
3534 goto error;
3535 }
3536
3537 ret = chmod(health_unix_sock_path,
3538 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3539 if (ret < 0) {
3540 ERR("Unable to set permissions on %s", health_unix_sock_path);
3541 PERROR("chmod");
3542 ret = -1;
3543 goto error;
3544 }
3545 }
3546
b662582b
DG
3547 /*
3548 * Set the CLOEXEC flag. Return code is useless because either way, the
3549 * show must go on.
3550 */
3551 (void) utils_set_fd_cloexec(sock);
3552
44a5e5eb
DG
3553 ret = lttcomm_listen_unix_sock(sock);
3554 if (ret < 0) {
3555 goto error;
3556 }
3557
3558 /*
3559 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3560 * more will be added to this poll set.
3561 */
d0b96690 3562 ret = sessiond_set_thread_pollset(&events, 2);
44a5e5eb
DG
3563 if (ret < 0) {
3564 goto error;
3565 }
3566
3567 /* Add the application registration socket */
3568 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3569 if (ret < 0) {
3570 goto error;
3571 }
3572
3573 while (1) {
3574 DBG("Health check ready");
3575
44a5e5eb
DG
3576 /* Inifinite blocking call, waiting for transmission */
3577restart:
3578 ret = lttng_poll_wait(&events, -1);
3579 if (ret < 0) {
3580 /*
3581 * Restart interrupted system call.
3582 */
3583 if (errno == EINTR) {
3584 goto restart;
3585 }
3586 goto error;
3587 }
3588
0d9c5d77
DG
3589 nb_fd = ret;
3590
44a5e5eb
DG
3591 for (i = 0; i < nb_fd; i++) {
3592 /* Fetch once the poll data */
3593 revents = LTTNG_POLL_GETEV(&events, i);
3594 pollfd = LTTNG_POLL_GETFD(&events, i);
3595
3596 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 3597 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
44a5e5eb 3598 if (ret) {
139ac872
MD
3599 err = 0;
3600 goto exit;
44a5e5eb
DG
3601 }
3602
3603 /* Event on the registration socket */
3604 if (pollfd == sock) {
3605 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3606 ERR("Health socket poll error");
3607 goto error;
3608 }
3609 }
3610 }
3611
3612 new_sock = lttcomm_accept_unix_sock(sock);
3613 if (new_sock < 0) {
3614 goto error;
3615 }
3616
b662582b
DG
3617 /*
3618 * Set the CLOEXEC flag. Return code is useless because either way, the
3619 * show must go on.
3620 */
3621 (void) utils_set_fd_cloexec(new_sock);
3622
44a5e5eb
DG
3623 DBG("Receiving data from client for health...");
3624 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3625 if (ret <= 0) {
3626 DBG("Nothing recv() from client... continuing");
3627 ret = close(new_sock);
3628 if (ret) {
3629 PERROR("close");
3630 }
3631 new_sock = -1;
3632 continue;
3633 }
3634
3635 rcu_thread_online();
3636
6c71277b
MD
3637 reply.ret_code = 0;
3638 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3639 /*
3640 * health_check_state returns 0 if health is
3641 * bad.
3642 */
3643 if (!health_check_state(health_sessiond, i)) {
3644 reply.ret_code |= 1ULL << i;
3645 }
44a5e5eb
DG
3646 }
3647
6c71277b 3648 DBG2("Health check return value %" PRIx64, reply.ret_code);
44a5e5eb
DG
3649
3650 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3651 if (ret < 0) {
3652 ERR("Failed to send health data back to client");
3653 }
3654
3655 /* End of transmission */
3656 ret = close(new_sock);
3657 if (ret) {
3658 PERROR("close");
3659 }
3660 new_sock = -1;
3661 }
3662
139ac872 3663exit:
44a5e5eb 3664error:
139ac872
MD
3665 if (err) {
3666 ERR("Health error occurred in %s", __func__);
3667 }
44a5e5eb
DG
3668 DBG("Health check thread dying");
3669 unlink(health_unix_sock_path);
3670 if (sock >= 0) {
3671 ret = close(sock);
3672 if (ret) {
3673 PERROR("close");
3674 }
3675 }
44a5e5eb
DG
3676
3677 lttng_poll_clean(&events);
3678
3679 rcu_unregister_thread();
3680 return NULL;
3681}
3682
1d4b027a 3683/*
d063d709
DG
3684 * This thread manage all clients request using the unix client socket for
3685 * communication.
1d4b027a
DG
3686 */
3687static void *thread_manage_clients(void *data)
3688{
139ac872 3689 int sock = -1, ret, i, pollfd, err = -1;
53a80697 3690 int sock_error;
5eb91c98 3691 uint32_t revents, nb_fd;
273ea72c 3692 struct command_ctx *cmd_ctx = NULL;
5eb91c98 3693 struct lttng_poll_event events;
1d4b027a
DG
3694
3695 DBG("[thread] Manage client started");
3696
f6a9efaa
DG
3697 rcu_register_thread();
3698
6c71277b 3699 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
927ca06a 3700
6993eeb3
CB
3701 if (testpoint(thread_manage_clients)) {
3702 goto error_testpoint;
3703 }
3704
840cb59c 3705 health_code_update();
44a5e5eb 3706
1d4b027a
DG
3707 ret = lttcomm_listen_unix_sock(client_sock);
3708 if (ret < 0) {
35df2755 3709 goto error_listen;
1d4b027a
DG
3710 }
3711
5eb91c98
DG
3712 /*
3713 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3714 * more will be added to this poll set.
3715 */
d0b96690 3716 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 3717 if (ret < 0) {
35df2755 3718 goto error_create_poll;
5eb91c98 3719 }
273ea72c 3720
5eb91c98
DG
3721 /* Add the application registration socket */
3722 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3723 if (ret < 0) {
3724 goto error;
3725 }
273ea72c 3726
bbd973c2
DG
3727 /*
3728 * Notify parent pid that we are ready to accept command for client side.
0bb7724a 3729 * This ppid is the one from the external process that spawned us.
1d4b027a
DG
3730 */
3731 if (opt_sig_parent) {
8db8d1dc 3732 kill(ppid, SIGUSR1);
1d4b027a
DG
3733 }
3734
0bb7724a
DG
3735 /* Notify the parent of the fork() process that we are ready. */
3736 if (opt_daemon) {
3737 kill(child_ppid, SIGUSR1);
3738 }
3739
6993eeb3
CB
3740 if (testpoint(thread_manage_clients_before_loop)) {
3741 goto error;
3742 }
8ac94142 3743
840cb59c 3744 health_code_update();
44a5e5eb 3745
1d4b027a 3746 while (1) {
1d4b027a 3747 DBG("Accepting client command ...");
273ea72c
DG
3748
3749 /* Inifinite blocking call, waiting for transmission */
88f2b785 3750 restart:
a78af745 3751 health_poll_entry();
5eb91c98 3752 ret = lttng_poll_wait(&events, -1);
a78af745 3753 health_poll_exit();
273ea72c 3754 if (ret < 0) {
88f2b785
MD
3755 /*
3756 * Restart interrupted system call.
3757 */
3758 if (errno == EINTR) {
3759 goto restart;
3760 }
273ea72c
DG
3761 goto error;
3762 }
3763
0d9c5d77
DG
3764 nb_fd = ret;
3765
5eb91c98
DG
3766 for (i = 0; i < nb_fd; i++) {
3767 /* Fetch once the poll data */
3768 revents = LTTNG_POLL_GETEV(&events, i);
3769 pollfd = LTTNG_POLL_GETFD(&events, i);
3770
840cb59c 3771 health_code_update();
44a5e5eb 3772
5eb91c98 3773 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 3774 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 3775 if (ret) {
139ac872
MD
3776 err = 0;
3777 goto exit;
5eb91c98
DG
3778 }
3779
3780 /* Event on the registration socket */
3781 if (pollfd == client_sock) {
3782 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3783 ERR("Client socket poll error");
3784 goto error;
3785 }
3786 }
3787 }
3788
3789 DBG("Wait for client response");
3790
840cb59c 3791 health_code_update();
44a5e5eb 3792
5eb91c98
DG
3793 sock = lttcomm_accept_unix_sock(client_sock);
3794 if (sock < 0) {
273ea72c 3795 goto error;
273ea72c
DG
3796 }
3797
b662582b
DG
3798 /*
3799 * Set the CLOEXEC flag. Return code is useless because either way, the
3800 * show must go on.
3801 */
3802 (void) utils_set_fd_cloexec(sock);
3803
be040666
DG
3804 /* Set socket option for credentials retrieval */
3805 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3806 if (ret < 0) {
3807 goto error;
3808 }
3809
5eb91c98 3810 /* Allocate context command to process the client request */
ba7f0ae5 3811 cmd_ctx = zmalloc(sizeof(struct command_ctx));
5eb91c98 3812 if (cmd_ctx == NULL) {
76d7553f 3813 PERROR("zmalloc cmd_ctx");
1d4b027a 3814 goto error;
5eb91c98 3815 }
1d4b027a 3816
5eb91c98 3817 /* Allocate data buffer for reception */
ba7f0ae5 3818 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
5eb91c98 3819 if (cmd_ctx->lsm == NULL) {
76d7553f 3820 PERROR("zmalloc cmd_ctx->lsm");
5eb91c98
DG
3821 goto error;
3822 }
1d4b027a 3823
5eb91c98
DG
3824 cmd_ctx->llm = NULL;
3825 cmd_ctx->session = NULL;
1d4b027a 3826
840cb59c 3827 health_code_update();
44a5e5eb 3828
5eb91c98
DG
3829 /*
3830 * Data is received from the lttng client. The struct
3831 * lttcomm_session_msg (lsm) contains the command and data request of
3832 * the client.
3833 */
3834 DBG("Receiving data from client ...");
be040666
DG
3835 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3836 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
5eb91c98
DG
3837 if (ret <= 0) {
3838 DBG("Nothing recv() from client... continuing");
76d7553f
MD
3839 ret = close(sock);
3840 if (ret) {
3841 PERROR("close");
3842 }
3843 sock = -1;
a2c0da86 3844 clean_command_ctx(&cmd_ctx);
5eb91c98
DG
3845 continue;
3846 }
1d4b027a 3847
840cb59c 3848 health_code_update();
44a5e5eb 3849
5eb91c98
DG
3850 // TODO: Validate cmd_ctx including sanity check for
3851 // security purpose.
f7776ea7 3852
f6a9efaa 3853 rcu_thread_online();
5eb91c98
DG
3854 /*
3855 * This function dispatch the work to the kernel or userspace tracer
3856 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3857 * informations for the client. The command context struct contains
3858 * everything this function may needs.
3859 */
53a80697 3860 ret = process_client_msg(cmd_ctx, sock, &sock_error);
f6a9efaa 3861 rcu_thread_offline();
5eb91c98 3862 if (ret < 0) {
36134aa1
DG
3863 ret = close(sock);
3864 if (ret) {
3865 PERROR("close");
53a80697 3866 }
36134aa1 3867 sock = -1;
bbd973c2 3868 /*
5eb91c98 3869 * TODO: Inform client somehow of the fatal error. At
ba7f0ae5 3870 * this point, ret < 0 means that a zmalloc failed
53a80697
MD
3871 * (ENOMEM). Error detected but still accept
3872 * command, unless a socket error has been
3873 * detected.
bbd973c2 3874 */
bbd973c2 3875 clean_command_ctx(&cmd_ctx);
5eb91c98
DG
3876 continue;
3877 }
d6e4fca4 3878
840cb59c 3879 health_code_update();
44a5e5eb 3880
54d01ffb
DG
3881 DBG("Sending response (size: %d, retcode: %s)",
3882 cmd_ctx->lttng_msg_size,
9a745bc7 3883 lttng_strerror(-cmd_ctx->llm->ret_code));
54d01ffb 3884 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
5eb91c98
DG
3885 if (ret < 0) {
3886 ERR("Failed to send data back to client");
bbd973c2 3887 }
1d4b027a 3888
5eb91c98 3889 /* End of transmission */
76d7553f
MD
3890 ret = close(sock);
3891 if (ret) {
3892 PERROR("close");
3893 }
3894 sock = -1;
be040666
DG
3895
3896 clean_command_ctx(&cmd_ctx);
44a5e5eb 3897
840cb59c 3898 health_code_update();
273ea72c
DG
3899 }
3900
139ac872 3901exit:
5eb91c98 3902error:
35df2755
CB
3903 if (sock >= 0) {
3904 ret = close(sock);
3905 if (ret) {
3906 PERROR("close");
3907 }
139ac872 3908 }
44a5e5eb 3909
35df2755
CB
3910 lttng_poll_clean(&events);
3911 clean_command_ctx(&cmd_ctx);
3912
3913error_listen:
3914error_create_poll:
6993eeb3 3915error_testpoint:
273ea72c 3916 unlink(client_unix_sock_path);
76d7553f
MD
3917 if (client_sock >= 0) {
3918 ret = close(client_sock);
3919 if (ret) {
3920 PERROR("close");
3921 }
a4b35e07 3922 }
35df2755
CB
3923
3924 if (err) {
840cb59c 3925 health_error();
35df2755 3926 ERR("Health error occurred in %s", __func__);
a4b35e07 3927 }
273ea72c 3928
8782cc74 3929 health_unregister(health_sessiond);
35df2755
CB
3930
3931 DBG("Client thread dying");
f6a9efaa
DG
3932
3933 rcu_unregister_thread();
1d4b027a
DG
3934 return NULL;
3935}
3936
3937
fac6795d
DG
3938/*
3939 * usage function on stderr
3940 */
3941static void usage(void)
3942{
b716ce68 3943 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
d6f42150
DG
3944 fprintf(stderr, " -h, --help Display this usage.\n");
3945 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3946 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3947 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3948 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
7753dea8
MD
3949 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3950 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3951 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3952 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
ebaeda94
MD
3953 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3954 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3955 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3956 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
d6f42150
DG
3957 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3958 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3959 fprintf(stderr, " -V, --version Show version number.\n");
7fe6d2c5 3960 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
d6f42150
DG
3961 fprintf(stderr, " -q, --quiet No output at all.\n");
3962 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
35f90c40 3963 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3bd1e081 3964 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4fba7219 3965 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4d076222 3966 fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
fac6795d
DG
3967}
3968
3969/*
3970 * daemon argument parsing
3971 */
3972static int parse_args(int argc, char **argv)
3973{
3974 int c;
3975
3976 static struct option long_options[] = {
3977 { "client-sock", 1, 0, 'c' },
3978 { "apps-sock", 1, 0, 'a' },
3bd1e081
MD
3979 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3980 { "kconsumerd-err-sock", 1, 0, 'E' },
7753dea8
MD
3981 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3982 { "ustconsumerd32-err-sock", 1, 0, 'H' },
ebaeda94
MD
3983 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3984 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3985 { "consumerd32-path", 1, 0, 'u' },
3986 { "consumerd32-libdir", 1, 0, 'U' },
3987 { "consumerd64-path", 1, 0, 't' },
3988 { "consumerd64-libdir", 1, 0, 'T' },
fac6795d 3989 { "daemonize", 0, 0, 'd' },
5b8719f5 3990 { "sig-parent", 0, 0, 'S' },
fac6795d
DG
3991 { "help", 0, 0, 'h' },
3992 { "group", 1, 0, 'g' },
3993 { "version", 0, 0, 'V' },
75462a81 3994 { "quiet", 0, 0, 'q' },
3f9947db 3995 { "verbose", 0, 0, 'v' },
3bd1e081 3996 { "verbose-consumer", 0, 0, 'Z' },
4fba7219 3997 { "no-kernel", 0, 0, 'N' },
35f90c40 3998 { "pidfile", 1, 0, 'p' },
4d076222 3999 { "jul-tcp-port", 1, 0, 'J' },
fac6795d
DG
4000 { NULL, 0, 0, 0 }
4001 };
4002
4003 while (1) {
4004 int option_index = 0;
4d076222 4005 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:",
54d01ffb 4006 long_options, &option_index);
fac6795d
DG
4007 if (c == -1) {
4008 break;
4009 }
4010
4011 switch (c) {
4012 case 0:
4013 fprintf(stderr, "option %s", long_options[option_index].name);
4014 if (optarg) {
4015 fprintf(stderr, " with arg %s\n", optarg);
4016 }
4017 break;
b716ce68 4018 case 'c':
fac6795d
DG
4019 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
4020 break;
4021 case 'a':
4022 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
4023 break;
4024 case 'd':
4025 opt_daemon = 1;
4026 break;
4027 case 'g':
6c71277b 4028 tracing_group_name = optarg;
fac6795d
DG
4029 break;
4030 case 'h':
4031 usage();
4032 exit(EXIT_FAILURE);
4033 case 'V':
4034 fprintf(stdout, "%s\n", VERSION);
4035 exit(EXIT_SUCCESS);
5b8719f5
DG
4036 case 'S':
4037 opt_sig_parent = 1;
4038 break;
d6f42150 4039 case 'E':
3bd1e081 4040 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
d6f42150
DG
4041 break;
4042 case 'C':
3bd1e081
MD
4043 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4044 break;
4045 case 'F':
7753dea8 4046 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3bd1e081
MD
4047 break;
4048 case 'D':
7753dea8
MD
4049 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4050 break;
4051 case 'H':
4052 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4053 break;
4054 case 'G':
4055 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
d6f42150 4056 break;
4fba7219
DG
4057 case 'N':
4058 opt_no_kernel = 1;
4059 break;
75462a81 4060 case 'q':
97e19046 4061 lttng_opt_quiet = 1;
75462a81 4062 break;
3f9947db 4063 case 'v':
53086306 4064 /* Verbose level can increase using multiple -v */
97e19046 4065 lttng_opt_verbose += 1;
3f9947db 4066 break;
31f73cc9 4067 case 'Z':
3bd1e081 4068 opt_verbose_consumer += 1;
31f73cc9 4069 break;
fb09408a 4070 case 'u':
fc7a59ce 4071 consumerd32_bin= optarg;
ebaeda94
MD
4072 break;
4073 case 'U':
4074 consumerd32_libdir = optarg;
7753dea8
MD
4075 break;
4076 case 't':
fc7a59ce 4077 consumerd64_bin = optarg;
ebaeda94
MD
4078 break;
4079 case 'T':
4080 consumerd64_libdir = optarg;
fb09408a 4081 break;
35f90c40
DG
4082 case 'p':
4083 opt_pidfile = optarg;
4084 break;
4d076222
DG
4085 case 'J': /* JUL TCP port. */
4086 {
4087 unsigned long v;
4088
4089 errno = 0;
4090 v = strtoul(optarg, NULL, 0);
4091 if (errno != 0 || !isdigit(optarg[0])) {
4092 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg);
4093 return -1;
4094 }
4095 if (v == 0 || v >= 65535) {
4096 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg);
4097 return -1;
4098 }
4099 jul_tcp_port = (uint32_t) v;
4100 DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
4101 break;
4102 }
fac6795d
DG
4103 default:
4104 /* Unknown option or other error.
4105 * Error is printed by getopt, just return */
4106 return -1;
4107 }
4108 }
4109
4110 return 0;
4111}
4112
4113/*
d063d709 4114 * Creates the two needed socket by the daemon.
d6f42150
DG
4115 * apps_sock - The communication socket for all UST apps.
4116 * client_sock - The communication of the cli tool (lttng).
fac6795d 4117 */
cf3af59e 4118static int init_daemon_socket(void)
fac6795d
DG
4119{
4120 int ret = 0;
4121 mode_t old_umask;
4122
4123 old_umask = umask(0);
4124
4125 /* Create client tool unix socket */
d6f42150
DG
4126 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4127 if (client_sock < 0) {
4128 ERR("Create unix sock failed: %s", client_unix_sock_path);
fac6795d
DG
4129 ret = -1;
4130 goto end;
4131 }
4132
b662582b
DG
4133 /* Set the cloexec flag */
4134 ret = utils_set_fd_cloexec(client_sock);
4135 if (ret < 0) {
4136 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4137 "Continuing but note that the consumer daemon will have a "
4138 "reference to this socket on exec()", client_sock);
4139 }
4140
fac6795d
DG
4141 /* File permission MUST be 660 */
4142 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4143 if (ret < 0) {
d6f42150 4144 ERR("Set file permissions failed: %s", client_unix_sock_path);
76d7553f 4145 PERROR("chmod");
fac6795d
DG
4146 goto end;
4147 }
4148
4149 /* Create the application unix socket */
d6f42150
DG
4150 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4151 if (apps_sock < 0) {
4152 ERR("Create unix sock failed: %s", apps_unix_sock_path);
fac6795d
DG
4153 ret = -1;
4154 goto end;
4155 }
4156
b662582b
DG
4157 /* Set the cloexec flag */
4158 ret = utils_set_fd_cloexec(apps_sock);
4159 if (ret < 0) {
4160 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4161 "Continuing but note that the consumer daemon will have a "
4162 "reference to this socket on exec()", apps_sock);
4163 }
4164
d6f42150 4165 /* File permission MUST be 666 */
54d01ffb
DG
4166 ret = chmod(apps_unix_sock_path,
4167 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
fac6795d 4168 if (ret < 0) {
d6f42150 4169 ERR("Set file permissions failed: %s", apps_unix_sock_path);
76d7553f 4170 PERROR("chmod");
fac6795d
DG
4171 goto end;
4172 }
4173
b662582b
DG
4174 DBG3("Session daemon client socket %d and application socket %d created",
4175 client_sock, apps_sock);
4176
fac6795d
DG
4177end:
4178 umask(old_umask);
4179 return ret;
4180}
4181
4182/*
54d01ffb
DG
4183 * Check if the global socket is available, and if a daemon is answering at the
4184 * other side. If yes, error is returned.
fac6795d 4185 */
cf3af59e 4186static int check_existing_daemon(void)
fac6795d 4187{
7d8234d9 4188 /* Is there anybody out there ? */
099e26bd 4189 if (lttng_session_daemon_alive()) {
7d8234d9 4190 return -EEXIST;
099e26bd 4191 }
b09c7c76
DG
4192
4193 return 0;
fac6795d
DG
4194}
4195
fac6795d 4196/*
d063d709 4197 * Set the tracing group gid onto the client socket.
5e16da05 4198 *
d063d709 4199 * Race window between mkdir and chown is OK because we are going from more
d1613cf5 4200 * permissive (root.root) to less permissive (root.tracing).
fac6795d 4201 */
be040666 4202static int set_permissions(char *rundir)
fac6795d
DG
4203{
4204 int ret;
996b65c8 4205 gid_t gid;
fac6795d 4206
6c71277b 4207 gid = utils_get_group_id(tracing_group_name);
fac6795d 4208
d6f42150 4209 /* Set lttng run dir */
be040666 4210 ret = chown(rundir, 0, gid);
d6f42150 4211 if (ret < 0) {
be040666 4212 ERR("Unable to set group on %s", rundir);
76d7553f 4213 PERROR("chown");
d6f42150
DG
4214 }
4215
6c71277b
MD
4216 /*
4217 * Ensure all applications and tracing group can search the run
4218 * dir. Allow everyone to read the directory, since it does not
4219 * buy us anything to hide its content.
4220 */
4221 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
d1613cf5
JN
4222 if (ret < 0) {
4223 ERR("Unable to set permissions on %s", rundir);
76d7553f 4224 PERROR("chmod");
d1613cf5
JN
4225 }
4226
d6f42150 4227 /* lttng client socket path */
996b65c8 4228 ret = chown(client_unix_sock_path, 0, gid);
fac6795d 4229 if (ret < 0) {
d6f42150 4230 ERR("Unable to set group on %s", client_unix_sock_path);
76d7553f 4231 PERROR("chown");
d6f42150
DG
4232 }
4233
3bd1e081 4234 /* kconsumer error socket path */
6c71277b 4235 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
d6f42150 4236 if (ret < 0) {
3bd1e081 4237 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
76d7553f 4238 PERROR("chown");
3bd1e081
MD
4239 }
4240
7753dea8 4241 /* 64-bit ustconsumer error socket path */
6c71277b 4242 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
7753dea8
MD
4243 if (ret < 0) {
4244 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
76d7553f 4245 PERROR("chown");
7753dea8
MD
4246 }
4247
4248 /* 32-bit ustconsumer compat32 error socket path */
6c71277b 4249 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
3bd1e081 4250 if (ret < 0) {
7753dea8 4251 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
76d7553f 4252 PERROR("chown");
fac6795d
DG
4253 }
4254
d6f42150 4255 DBG("All permissions are set");
e07ae692 4256
fac6795d
DG
4257 return ret;
4258}
4259
d6f42150 4260/*
d063d709 4261 * Create the lttng run directory needed for all global sockets and pipe.
d6f42150 4262 */
67e40797 4263static int create_lttng_rundir(const char *rundir)
d6f42150
DG
4264{
4265 int ret;
4266
67e40797
DG
4267 DBG3("Creating LTTng run directory: %s", rundir);
4268
d1613cf5 4269 ret = mkdir(rundir, S_IRWXU);
d6f42150 4270 if (ret < 0) {
b1f11e69 4271 if (errno != EEXIST) {
67e40797 4272 ERR("Unable to create %s", rundir);
b1f11e69
DG
4273 goto error;
4274 } else {
4275 ret = 0;
4276 }
d6f42150
DG
4277 }
4278
4279error:
4280 return ret;
4281}
4282
4283/*
d063d709
DG
4284 * Setup sockets and directory needed by the kconsumerd communication with the
4285 * session daemon.
d6f42150 4286 */
67e40797
DG
4287static int set_consumer_sockets(struct consumer_data *consumer_data,
4288 const char *rundir)
d6f42150
DG
4289{
4290 int ret;
67e40797 4291 char path[PATH_MAX];
d6f42150 4292
6c71277b 4293 switch (consumer_data->type) {
7753dea8 4294 case LTTNG_CONSUMER_KERNEL:
60922cb0 4295 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
7753dea8
MD
4296 break;
4297 case LTTNG_CONSUMER64_UST:
60922cb0 4298 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
7753dea8
MD
4299 break;
4300 case LTTNG_CONSUMER32_UST:
60922cb0 4301 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
7753dea8
MD
4302 break;
4303 default:
4304 ERR("Consumer type unknown");
4305 ret = -EINVAL;
4306 goto error;
d6f42150
DG
4307 }
4308
67e40797
DG
4309 DBG2("Creating consumer directory: %s", path);
4310
6c71277b 4311 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
d6f42150 4312 if (ret < 0) {
6beb2242 4313 if (errno != EEXIST) {
f11e84c2 4314 PERROR("mkdir");
3bd1e081 4315 ERR("Failed to create %s", path);
6beb2242
DG
4316 goto error;
4317 }
f11e84c2 4318 ret = -1;
d6f42150 4319 }
6c71277b
MD
4320 if (is_root) {
4321 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4322 if (ret < 0) {
4323 ERR("Unable to set group on %s", path);
4324 PERROR("chown");
4325 goto error;
4326 }
4327 }
d6f42150
DG
4328
4329 /* Create the kconsumerd error unix socket */
3bd1e081
MD
4330 consumer_data->err_sock =
4331 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4332 if (consumer_data->err_sock < 0) {
4333 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
d6f42150
DG
4334 ret = -1;
4335 goto error;
4336 }
4337
a24f05ab
MD
4338 /*
4339 * Set the CLOEXEC flag. Return code is useless because either way, the
4340 * show must go on.
4341 */
4342 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4343 if (ret < 0) {
4344 PERROR("utils_set_fd_cloexec");
4345 /* continue anyway */
4346 }
4347
d6f42150 4348 /* File permission MUST be 660 */
3bd1e081 4349 ret = chmod(consumer_data->err_unix_sock_path,
54d01ffb 4350 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
d6f42150 4351 if (ret < 0) {
3bd1e081 4352 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
67e40797 4353 PERROR("chmod");
d6f42150
DG
4354 goto error;
4355 }
4356
4357error:
4358 return ret;
4359}
4360
fac6795d 4361/*
d063d709 4362 * Signal handler for the daemon
cf3af59e 4363 *
54d01ffb
DG
4364 * Simply stop all worker threads, leaving main() return gracefully after
4365 * joining all threads and calling cleanup().
fac6795d
DG
4366 */
4367static void sighandler(int sig)
4368{
4369 switch (sig) {
cf3af59e 4370 case SIGPIPE:
af87c45a 4371 DBG("SIGPIPE caught");
cf3af59e
MD
4372 return;
4373 case SIGINT:
af87c45a 4374 DBG("SIGINT caught");
cf3af59e
MD
4375 stop_threads();
4376 break;
4377 case SIGTERM:
af87c45a 4378 DBG("SIGTERM caught");
cf3af59e
MD
4379 stop_threads();
4380 break;
0bb7724a
DG
4381 case SIGUSR1:
4382 CMM_STORE_SHARED(recv_child_signal, 1);
4383 break;
cf3af59e
MD
4384 default:
4385 break;
fac6795d 4386 }
fac6795d
DG
4387}
4388
4389/*
d063d709 4390 * Setup signal handler for :
1d4b027a 4391 * SIGINT, SIGTERM, SIGPIPE
fac6795d 4392 */
1d4b027a 4393static int set_signal_handler(void)
fac6795d 4394{
1d4b027a
DG
4395 int ret = 0;
4396 struct sigaction sa;
4397 sigset_t sigset;
fac6795d 4398
1d4b027a 4399 if ((ret = sigemptyset(&sigset)) < 0) {
76d7553f 4400 PERROR("sigemptyset");
1d4b027a
DG
4401 return ret;
4402 }
d6f42150 4403
1d4b027a
DG
4404 sa.sa_handler = sighandler;
4405 sa.sa_mask = sigset;
4406 sa.sa_flags = 0;
4407 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
76d7553f 4408 PERROR("sigaction");
1d4b027a 4409 return ret;
d6f42150
DG
4410 }
4411
1d4b027a 4412 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
76d7553f 4413 PERROR("sigaction");
1d4b027a 4414 return ret;
d6f42150 4415 }
aaf26714 4416
1d4b027a 4417 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
76d7553f 4418 PERROR("sigaction");
1d4b027a 4419 return ret;
8c0faa1d
DG
4420 }
4421
0bb7724a
DG
4422 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4423 PERROR("sigaction");
4424 return ret;
4425 }
4426
4427 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
1d4b027a
DG
4428
4429 return ret;
fac6795d
DG
4430}
4431
f3ed775e 4432/*
d063d709
DG
4433 * Set open files limit to unlimited. This daemon can open a large number of
4434 * file descriptors in order to consumer multiple kernel traces.
f3ed775e
DG
4435 */
4436static void set_ulimit(void)
4437{
4438 int ret;
4439 struct rlimit lim;
4440
a88df331 4441 /* The kernel does not allowed an infinite limit for open files */
f3ed775e
DG
4442 lim.rlim_cur = 65535;
4443 lim.rlim_max = 65535;
4444
4445 ret = setrlimit(RLIMIT_NOFILE, &lim);
4446 if (ret < 0) {
76d7553f 4447 PERROR("failed to set open files limit");
f3ed775e
DG
4448 }
4449}
4450
35f90c40
DG
4451/*
4452 * Write pidfile using the rundir and opt_pidfile.
4453 */
4454static void write_pidfile(void)
4455{
4456 int ret;
4457 char pidfile_path[PATH_MAX];
4458
4459 assert(rundir);
4460
4461 if (opt_pidfile) {
4462 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4463 } else {
4464 /* Build pidfile path from rundir and opt_pidfile. */
4465 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4466 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4467 if (ret < 0) {
4468 PERROR("snprintf pidfile path");
4469 goto error;
4470 }
4471 }
4472
4473 /*
4474 * Create pid file in rundir. Return value is of no importance. The
4475 * execution will continue even though we are not able to write the file.
4476 */
4477 (void) utils_create_pid_file(getpid(), pidfile_path);
4478
4479error:
4480 return;
4481}
4482
cd9290dd
DG
4483/*
4484 * Write JUL TCP port using the rundir.
4485 */
4486static void write_julport(void)
4487{
4488 int ret;
4489 char path[PATH_MAX];
4490
4491 assert(rundir);
4492
4493 ret = snprintf(path, sizeof(path), "%s/"
4494 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
4495 if (ret < 0) {
4496 PERROR("snprintf julport path");
4497 goto error;
4498 }
4499
4500 /*
4501 * Create TCP JUL port file in rundir. Return value is of no importance.
4502 * The execution will continue even though we are not able to write the
4503 * file.
4504 */
4505 (void) utils_create_pid_file(jul_tcp_port, path);
4506
4507error:
4508 return;
4509}
4510
0bb7724a
DG
4511/*
4512 * Daemonize this process by forking and making the parent wait for the child
4513 * to signal it indicating readiness. Once received, the parent successfully
4514 * quits.
4515 *
4516 * The child process undergoes the same action that daemon(3) does meaning
4517 * setsid, chdir, and dup /dev/null into 0, 1 and 2.
4518 *
4519 * Return 0 on success else -1 on error.
4520 */
4521static int daemonize(void)
4522{
4523 int ret;
4524 pid_t pid;
4525
4526 /* Get parent pid of this process. */
4527 child_ppid = getppid();
4528
4529 pid = fork();
4530 if (pid < 0) {
4531 PERROR("fork");
4532 goto error;
4533 } else if (pid == 0) {
4534 int fd;
4535 pid_t sid;
4536
4537 /* Child */
4538
4539 /*
4540 * Get the newly created parent pid so we can signal that process when
4541 * we are ready to operate.
4542 */
4543 child_ppid = getppid();
4544
4545 sid = setsid();
4546 if (sid < 0) {
4547 PERROR("setsid");
4548 goto error;
4549 }
4550
4551 /* Try to change directory to /. If we can't well at least notify. */
4552 ret = chdir("/");
4553 if (ret < 0) {
4554 PERROR("chdir");
4555 }
4556
4557 fd = open(_PATH_DEVNULL, O_RDWR, 0);
4558 if (fd < 0) {
4559 PERROR("open %s", _PATH_DEVNULL);
4560 /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */
4561 } else {
4562 (void) dup2(fd, STDIN_FILENO);
4563 (void) dup2(fd, STDOUT_FILENO);
4564 (void) dup2(fd, STDERR_FILENO);
4565 if (fd > 2) {
4566 ret = close(fd);
4567 if (ret < 0) {
4568 PERROR("close");
4569 }
4570 }
4571 }
4572 goto end;
4573 } else {
4574 /* Parent */
4575
4576 /*
4577 * Waiting for child to notify this parent that it can exit. Note that
4578 * sleep() is interrupted before the 1 second delay as soon as the
4579 * signal is received, so it will not cause visible delay for the
4580 * user.
4581 */
4582 while (!CMM_LOAD_SHARED(recv_child_signal)) {
077df0b3
DG
4583 int status;
4584 pid_t ret;
4585
4586 /*
4587 * Check if child exists without blocking. If so, we have to stop
4588 * this parent process and return an error.
4589 */
4590 ret = waitpid(pid, &status, WNOHANG);
4591 if (ret < 0 || (ret != 0 && WIFEXITED(status))) {
4592 /* The child exited somehow or was not valid. */
4593 goto error;
4594 }
0bb7724a
DG
4595 sleep(1);
4596 }
4597
4598 /*
4599 * From this point on, the parent can exit and the child is now an
4600 * operationnal session daemon ready to serve clients and applications.
4601 */
4602 exit(EXIT_SUCCESS);
4603 }
4604
4605end:
4606 return 0;
4607
4608error:
4609 return -1;
4610}
4611
fac6795d
DG
4612/*
4613 * main
4614 */
4615int main(int argc, char **argv)
4616{
fac6795d
DG
4617 int ret = 0;
4618 void *status;
ae9e45b3 4619 const char *home_path, *env_app_timeout;
fac6795d 4620
335a95b7
MD
4621 init_kernel_workarounds();
4622
f6a9efaa
DG
4623 rcu_register_thread();
4624
0bb7724a
DG
4625 if ((ret = set_signal_handler()) < 0) {
4626 goto error;
4627 }
4628
7753dea8 4629 setup_consumerd_path();
fb09408a 4630
12744796
DG
4631 page_size = sysconf(_SC_PAGESIZE);
4632 if (page_size < 0) {
4633 PERROR("sysconf _SC_PAGESIZE");
4634 page_size = LONG_MAX;
4635 WARN("Fallback page size to %ld", page_size);
4636 }
4637
fac6795d
DG
4638 /* Parse arguments */
4639 progname = argv[0];
c617c0c6 4640 if ((ret = parse_args(argc, argv)) < 0) {
cf3af59e 4641 goto error;
fac6795d
DG
4642 }
4643
4644 /* Daemonize */
4645 if (opt_daemon) {
ceed52b5
MD
4646 int i;
4647
0bb7724a 4648 ret = daemonize();
53094c05 4649 if (ret < 0) {
cf3af59e 4650 goto error;
53094c05 4651 }
0bb7724a 4652
ceed52b5 4653 /*
0bb7724a
DG
4654 * We are in the child. Make sure all other file descriptors are
4655 * closed, in case we are called with more opened file descriptors than
4656 * the standard ones.
ceed52b5
MD
4657 */
4658 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4659 (void) close(i);
4660 }
4661 }
4662
4663 /* Create thread quit pipe */
4664 if ((ret = init_thread_quit_pipe()) < 0) {
4665 goto error;
fac6795d
DG
4666 }
4667
4668 /* Check if daemon is UID = 0 */
4669 is_root = !getuid();
4670
fac6795d 4671 if (is_root) {
990570ed 4672 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
67e40797
DG
4673
4674 /* Create global run dir with root access */
4675 ret = create_lttng_rundir(rundir);
d6f42150 4676 if (ret < 0) {
cf3af59e 4677 goto error;
d6f42150
DG
4678 }
4679
fac6795d 4680 if (strlen(apps_unix_sock_path) == 0) {
d6f42150
DG
4681 snprintf(apps_unix_sock_path, PATH_MAX,
4682 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
fac6795d
DG
4683 }
4684
4685 if (strlen(client_unix_sock_path) == 0) {
d6f42150
DG
4686 snprintf(client_unix_sock_path, PATH_MAX,
4687 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4688 }
0fdd1e2c
DG
4689
4690 /* Set global SHM for ust */
4691 if (strlen(wait_shm_path) == 0) {
4692 snprintf(wait_shm_path, PATH_MAX,
4693 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4694 }
67e40797 4695
44a5e5eb
DG
4696 if (strlen(health_unix_sock_path) == 0) {
4697 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4698 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4699 }
4700
67e40797
DG
4701 /* Setup kernel consumerd path */
4702 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
60922cb0 4703 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
67e40797 4704 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4705 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
67e40797
DG
4706
4707 DBG2("Kernel consumer err path: %s",
4708 kconsumer_data.err_unix_sock_path);
4709 DBG2("Kernel consumer cmd path: %s",
4710 kconsumer_data.cmd_unix_sock_path);
fac6795d 4711 } else {
feb0f3e5 4712 home_path = utils_get_home_dir();
b082db07 4713 if (home_path == NULL) {
273ea72c
DG
4714 /* TODO: Add --socket PATH option */
4715 ERR("Can't get HOME directory for sockets creation.");
cf3af59e
MD
4716 ret = -EPERM;
4717 goto error;
b082db07
DG
4718 }
4719
67e40797
DG
4720 /*
4721 * Create rundir from home path. This will create something like
4722 * $HOME/.lttng
4723 */
990570ed 4724 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
67e40797
DG
4725 if (ret < 0) {
4726 ret = -ENOMEM;
4727 goto error;
4728 }
4729
4730 ret = create_lttng_rundir(rundir);
4731 if (ret < 0) {
4732 goto error;
4733 }
4734
fac6795d 4735 if (strlen(apps_unix_sock_path) == 0) {
d6f42150 4736 snprintf(apps_unix_sock_path, PATH_MAX,
b082db07 4737 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
fac6795d
DG
4738 }
4739
4740 /* Set the cli tool unix socket path */
4741 if (strlen(client_unix_sock_path) == 0) {
d6f42150 4742 snprintf(client_unix_sock_path, PATH_MAX,
b082db07 4743 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
fac6795d 4744 }
0fdd1e2c
DG
4745
4746 /* Set global SHM for ust */
4747 if (strlen(wait_shm_path) == 0) {
4748 snprintf(wait_shm_path, PATH_MAX,
d0b96690 4749 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
0fdd1e2c 4750 }
44a5e5eb
DG
4751
4752 /* Set health check Unix path */
4753 if (strlen(health_unix_sock_path) == 0) {
4754 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4755 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4756 }
fac6795d
DG
4757 }
4758
5c827ce0
DG
4759 /* Set consumer initial state */
4760 kernel_consumerd_state = CONSUMER_STOPPED;
4761 ust_consumerd_state = CONSUMER_STOPPED;
4762
847177cd
DG
4763 DBG("Client socket path %s", client_unix_sock_path);
4764 DBG("Application socket path %s", apps_unix_sock_path);
d0b96690 4765 DBG("Application wait path %s", wait_shm_path);
67e40797
DG
4766 DBG("LTTng run directory path: %s", rundir);
4767
4768 /* 32 bits consumerd path setup */
4769 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
60922cb0 4770 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
67e40797 4771 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4772 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
67e40797
DG
4773
4774 DBG2("UST consumer 32 bits err path: %s",
4775 ustconsumer32_data.err_unix_sock_path);
4776 DBG2("UST consumer 32 bits cmd path: %s",
4777 ustconsumer32_data.cmd_unix_sock_path);
4778
4779 /* 64 bits consumerd path setup */
4780 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
60922cb0 4781 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
67e40797 4782 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4783 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
67e40797
DG
4784
4785 DBG2("UST consumer 64 bits err path: %s",
4786 ustconsumer64_data.err_unix_sock_path);
4787 DBG2("UST consumer 64 bits cmd path: %s",
4788 ustconsumer64_data.cmd_unix_sock_path);
847177cd 4789
273ea72c 4790 /*
7d8234d9 4791 * See if daemon already exist.
fac6795d 4792 */
7d8234d9 4793 if ((ret = check_existing_daemon()) < 0) {
75462a81 4794 ERR("Already running daemon.\n");
273ea72c 4795 /*
cf3af59e
MD
4796 * We do not goto exit because we must not cleanup()
4797 * because a daemon is already running.
ab118b20 4798 */
cf3af59e 4799 goto error;
a88df331
DG
4800 }
4801
1427f9b2
DG
4802 /*
4803 * Init UST app hash table. Alloc hash table before this point since
4804 * cleanup() can get called after that point.
4805 */
4806 ust_app_ht_alloc();
4807
f20baf8e
DG
4808 /* Initialize JUL domain subsystem. */
4809 if ((ret = jul_init()) < 0) {
4810 /* ENOMEM at this point. */
4811 goto error;
4812 }
4813
54d01ffb 4814 /* After this point, we can safely call cleanup() with "goto exit" */
a88df331
DG
4815
4816 /*
4817 * These actions must be executed as root. We do that *after* setting up
4818 * the sockets path because we MUST make the check for another daemon using
4819 * those paths *before* trying to set the kernel consumer sockets and init
4820 * kernel tracer.
4821 */
4822 if (is_root) {
67e40797 4823 ret = set_consumer_sockets(&kconsumer_data, rundir);
7753dea8
MD
4824 if (ret < 0) {
4825 goto exit;
4826 }
4827
a88df331 4828 /* Setup kernel tracer */
4fba7219
DG
4829 if (!opt_no_kernel) {
4830 init_kernel_tracer();
4831 }
a88df331
DG
4832
4833 /* Set ulimit for open files */
4834 set_ulimit();
fac6795d 4835 }
4063050c
MD
4836 /* init lttng_fd tracking must be done after set_ulimit. */
4837 lttng_fd_init();
fac6795d 4838
67e40797
DG
4839 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4840 if (ret < 0) {
4841 goto exit;
4842 }
4843
4844 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4845 if (ret < 0) {
4846 goto exit;
4847 }
4848
d6f42150 4849 /* Setup the needed unix socket */
cf3af59e
MD
4850 if ((ret = init_daemon_socket()) < 0) {
4851 goto exit;
fac6795d
DG
4852 }
4853
4854 /* Set credentials to socket */
be040666 4855 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
cf3af59e 4856 goto exit;
fac6795d
DG
4857 }
4858
5b8719f5
DG
4859 /* Get parent pid if -S, --sig-parent is specified. */
4860 if (opt_sig_parent) {
4861 ppid = getppid();
4862 }
4863
7a485870 4864 /* Setup the kernel pipe for waking up the kernel thread */
6620da75
DG
4865 if (is_root && !opt_no_kernel) {
4866 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4867 goto exit;
4868 }
7a485870
DG
4869 }
4870
0b2dc8df
MD
4871 /* Setup the thread ht_cleanup communication pipe. */
4872 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4873 goto exit;
4874 }
4875
099e26bd 4876 /* Setup the thread apps communication pipe. */
ef599319 4877 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
099e26bd
DG
4878 goto exit;
4879 }
4880
d0b96690
DG
4881 /* Setup the thread apps notify communication pipe. */
4882 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4883 goto exit;
4884 }
4885
7972aab2
DG
4886 /* Initialize global buffer per UID and PID registry. */
4887 buffer_reg_init_uid_registry();
4888 buffer_reg_init_pid_registry();
4889
099e26bd
DG
4890 /* Init UST command queue. */
4891 cds_wfq_init(&ust_cmd_queue.queue);
4892
273ea72c 4893 /*
54d01ffb
DG
4894 * Get session list pointer. This pointer MUST NOT be free(). This list is
4895 * statically declared in session.c
273ea72c 4896 */
54d01ffb 4897 session_list_ptr = session_get_list();
b5541356 4898
5eb91c98
DG
4899 /* Set up max poll set size */
4900 lttng_poll_set_max_size();
4901
2f77fc4b 4902 cmd_init();
00e2e675 4903
ae9e45b3
DG
4904 /* Check for the application socket timeout env variable. */
4905 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4906 if (env_app_timeout) {
4907 app_socket_timeout = atoi(env_app_timeout);
4908 } else {
4909 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4910 }
4911
35f90c40 4912 write_pidfile();
cd9290dd 4913 write_julport();
35f90c40 4914
554831e7
MD
4915 /* Initialize communication library */
4916 lttcomm_init();
d831c249
DG
4917 /* This is to get the TCP timeout value. */
4918 lttcomm_inet_init();
554831e7 4919
67e05644
DG
4920 /*
4921 * Initialize the health check subsystem. This call should set the
4922 * appropriate time values.
4923 */
6c71277b 4924 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
8782cc74
MD
4925 if (!health_sessiond) {
4926 PERROR("health_app_create error");
4927 goto exit_health_sessiond_cleanup;
4928 }
67e05644 4929
d5b9fd2f 4930 /* Create thread to clean up RCU hash tables */
0b2dc8df
MD
4931 ret = pthread_create(&ht_cleanup_thread, NULL,
4932 thread_ht_cleanup, (void *) NULL);
4933 if (ret != 0) {
4934 PERROR("pthread_create ht_cleanup");
4935 goto exit_ht_cleanup;
4936 }
4937
d5b9fd2f 4938 /* Create health-check thread */
44a5e5eb
DG
4939 ret = pthread_create(&health_thread, NULL,
4940 thread_manage_health, (void *) NULL);
4941 if (ret != 0) {
4942 PERROR("pthread_create health");
4943 goto exit_health;
4944 }
4945
cf3af59e 4946 /* Create thread to manage the client socket */
099e26bd
DG
4947 ret = pthread_create(&client_thread, NULL,
4948 thread_manage_clients, (void *) NULL);
cf3af59e 4949 if (ret != 0) {
76d7553f 4950 PERROR("pthread_create clients");
cf3af59e
MD
4951 goto exit_client;
4952 }
fac6795d 4953
099e26bd
DG
4954 /* Create thread to dispatch registration */
4955 ret = pthread_create(&dispatch_thread, NULL,
4956 thread_dispatch_ust_registration, (void *) NULL);
4957 if (ret != 0) {
76d7553f 4958 PERROR("pthread_create dispatch");
099e26bd
DG
4959 goto exit_dispatch;
4960 }
4961
4962 /* Create thread to manage application registration. */
4963 ret = pthread_create(&reg_apps_thread, NULL,
4964 thread_registration_apps, (void *) NULL);
4965 if (ret != 0) {
76d7553f 4966 PERROR("pthread_create registration");
099e26bd
DG
4967 goto exit_reg_apps;
4968 }
4969
cf3af59e 4970 /* Create thread to manage application socket */
54d01ffb
DG
4971 ret = pthread_create(&apps_thread, NULL,
4972 thread_manage_apps, (void *) NULL);
cf3af59e 4973 if (ret != 0) {
d0b96690
DG
4974 PERROR("pthread_create apps");
4975 goto exit_apps;
4976 }
4977
4978 /* Create thread to manage application notify socket */
4979 ret = pthread_create(&apps_notify_thread, NULL,
4980 ust_thread_manage_notify, (void *) NULL);
4981 if (ret != 0) {
76d7553f 4982 PERROR("pthread_create apps");
9563b0ad 4983 goto exit_apps_notify;
cf3af59e 4984 }
fac6795d 4985
4d076222
DG
4986 /* Create JUL registration thread. */
4987 ret = pthread_create(&jul_reg_thread, NULL,
4988 jul_thread_manage_registration, (void *) NULL);
4989 if (ret != 0) {
4990 PERROR("pthread_create apps");
4991 goto exit_jul_reg;
4992 }
4993
6620da75
DG
4994 /* Don't start this thread if kernel tracing is not requested nor root */
4995 if (is_root && !opt_no_kernel) {
4996 /* Create kernel thread to manage kernel event */
4997 ret = pthread_create(&kernel_thread, NULL,
4998 thread_manage_kernel, (void *) NULL);
4999 if (ret != 0) {
5000 PERROR("pthread_create kernel");
5001 goto exit_kernel;
5002 }
7a485870 5003
6620da75
DG
5004 ret = pthread_join(kernel_thread, &status);
5005 if (ret != 0) {
5006 PERROR("pthread_join");
5007 goto error; /* join error, exit without cleanup */
5008 }
fac6795d
DG
5009 }
5010
cf3af59e 5011exit_kernel:
4d076222
DG
5012 ret = pthread_join(jul_reg_thread, &status);
5013 if (ret != 0) {
5014 PERROR("pthread_join JUL");
5015 goto error; /* join error, exit without cleanup */
5016 }
5017
5018exit_jul_reg:
9563b0ad
DG
5019 ret = pthread_join(apps_notify_thread, &status);
5020 if (ret != 0) {
5021 PERROR("pthread_join apps notify");
5022 goto error; /* join error, exit without cleanup */
5023 }
5024
5025exit_apps_notify:
cf3af59e
MD
5026 ret = pthread_join(apps_thread, &status);
5027 if (ret != 0) {
9563b0ad 5028 PERROR("pthread_join apps");
cf3af59e
MD
5029 goto error; /* join error, exit without cleanup */
5030 }
fac6795d 5031
4ccd8c8f 5032
cf3af59e 5033exit_apps:
099e26bd
DG
5034 ret = pthread_join(reg_apps_thread, &status);
5035 if (ret != 0) {
76d7553f 5036 PERROR("pthread_join");
099e26bd
DG
5037 goto error; /* join error, exit without cleanup */
5038 }
5039
5040exit_reg_apps:
5041 ret = pthread_join(dispatch_thread, &status);
5042 if (ret != 0) {
76d7553f 5043 PERROR("pthread_join");
099e26bd
DG
5044 goto error; /* join error, exit without cleanup */
5045 }
5046
5047exit_dispatch:
cf3af59e
MD
5048 ret = pthread_join(client_thread, &status);
5049 if (ret != 0) {
76d7553f 5050 PERROR("pthread_join");
cf3af59e
MD
5051 goto error; /* join error, exit without cleanup */
5052 }
5053
3bd1e081 5054 ret = join_consumer_thread(&kconsumer_data);
cf3af59e 5055 if (ret != 0) {
76d7553f 5056 PERROR("join_consumer");
cf3af59e
MD
5057 goto error; /* join error, exit without cleanup */
5058 }
a88df331 5059
06f525de
DG
5060 ret = join_consumer_thread(&ustconsumer32_data);
5061 if (ret != 0) {
5062 PERROR("join_consumer ust32");
5063 goto error; /* join error, exit without cleanup */
5064 }
5065
5066 ret = join_consumer_thread(&ustconsumer64_data);
5067 if (ret != 0) {
5068 PERROR("join_consumer ust64");
5069 goto error; /* join error, exit without cleanup */
5070 }
5071
cf3af59e 5072exit_client:
06f525de
DG
5073 ret = pthread_join(health_thread, &status);
5074 if (ret != 0) {
5075 PERROR("pthread_join health thread");
5076 goto error; /* join error, exit without cleanup */
5077 }
5078
44a5e5eb 5079exit_health:
0b2dc8df
MD
5080 ret = pthread_join(ht_cleanup_thread, &status);
5081 if (ret != 0) {
5082 PERROR("pthread_join ht cleanup thread");
5083 goto error; /* join error, exit without cleanup */
5084 }
5085exit_ht_cleanup:
8782cc74
MD
5086 health_app_destroy(health_sessiond);
5087exit_health_sessiond_cleanup:
a88df331 5088exit:
cf3af59e
MD
5089 /*
5090 * cleanup() is called when no other thread is running.
5091 */
f6a9efaa 5092 rcu_thread_online();
cf3af59e 5093 cleanup();
f6a9efaa
DG
5094 rcu_thread_offline();
5095 rcu_unregister_thread();
67e40797 5096 if (!ret) {
cf3af59e 5097 exit(EXIT_SUCCESS);
67e40797 5098 }
cf3af59e 5099error:
5e16da05 5100 exit(EXIT_FAILURE);
fac6795d 5101}
This page took 0.469183 seconds and 4 git commands to generate.