Fix: double RCU unlock on event_agent_disable_all
[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>
26296c48 4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
fac6795d 5 *
d14d33bf
AM
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
91d76f53 9 *
d14d33bf
AM
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
91d76f53 14 *
d14d33bf
AM
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
fac6795d
DG
18 */
19
20#define _GNU_SOURCE
6c1c0768 21#define _LGPL_SOURCE
fac6795d
DG
22#include <getopt.h>
23#include <grp.h>
24#include <limits.h>
0bb7724a 25#include <paths.h>
fac6795d
DG
26#include <pthread.h>
27#include <signal.h>
28#include <stdio.h>
29#include <stdlib.h>
30#include <string.h>
331744e3 31#include <inttypes.h>
0fdd1e2c 32#include <sys/mman.h>
b73401da 33#include <sys/mount.h>
1e307fab 34#include <sys/resource.h>
fac6795d
DG
35#include <sys/socket.h>
36#include <sys/stat.h>
37#include <sys/types.h>
0fdd1e2c 38#include <sys/wait.h>
5c827ce0 39#include <urcu/uatomic.h>
fac6795d 40#include <unistd.h>
3bd1e081 41#include <config.h>
fac6795d 42
990570ed 43#include <common/common.h>
d27c42b8 44#include <common/compat/socket.h>
e8fa9fb0 45#include <common/compat/getenv.h>
db758600
DG
46#include <common/defaults.h>
47#include <common/kernel-consumer/kernel-consumer.h>
50c8f484 48#include <common/futex.h>
00e2e675 49#include <common/relayd/relayd.h>
81b86775 50#include <common/utils.h>
3ccdf997 51#include <common/daemonize.h>
26296c48 52#include <common/config/config.h>
fac6795d 53
10a8a223 54#include "lttng-sessiond.h"
7972aab2 55#include "buffer-registry.h"
54d01ffb 56#include "channel.h"
2f77fc4b 57#include "cmd.h"
00e2e675 58#include "consumer.h"
099e26bd 59#include "context.h"
54d01ffb 60#include "event.h"
4771f025 61#include "kernel.h"
f1e16794 62#include "kernel-consumer.h"
096102bd 63#include "modprobe.h"
0fdd1e2c 64#include "shm.h"
1e307fab 65#include "ust-ctl.h"
00e2e675 66#include "ust-consumer.h"
8e68d1c8 67#include "utils.h"
4063050c 68#include "fd-limit.h"
8782cc74 69#include "health-sessiond.h"
8ac94142 70#include "testpoint.h"
d0b96690 71#include "ust-thread.h"
022d91ba 72#include "agent-thread.h"
fb198a11 73#include "save.h"
ef367a93 74#include "load-session-thread.h"
834978fd 75#include "syscall.h"
7c1d2758 76#include "agent.h"
fac6795d 77
ebaeda94
MD
78#define CONSUMERD_FILE "lttng-consumerd"
79
fac6795d 80const char *progname;
6c71277b 81static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
26296c48
JG
82static int tracing_group_name_override;
83static char *opt_pidfile;
5b8719f5 84static int opt_sig_parent;
97e19046 85static int opt_verbose_consumer;
72dd7491 86static int opt_daemon, opt_background;
4fba7219 87static int opt_no_kernel;
ef367a93 88static char *opt_load_session_path;
1d4b027a 89static pid_t ppid; /* Parent PID for --sig-parent option */
0bb7724a 90static pid_t child_ppid; /* Internal parent PID use with daemonize. */
67e40797 91static char *rundir;
c9cb3e7d 92static int lockfile_fd = -1;
3bd1e081 93
0bb7724a
DG
94/* Set to 1 when a SIGUSR1 signal is received. */
95static int recv_child_signal;
96
a23ec3a7
DG
97/*
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
100 */
3bd1e081
MD
101static struct consumer_data kconsumer_data = {
102 .type = LTTNG_CONSUMER_KERNEL,
60922cb0
DG
103 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
104 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_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,
3bd1e081 111};
7753dea8
MD
112static struct consumer_data ustconsumer64_data = {
113 .type = LTTNG_CONSUMER64_UST,
60922cb0
DG
114 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
115 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_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,
7753dea8
MD
122};
123static struct consumer_data ustconsumer32_data = {
124 .type = LTTNG_CONSUMER32_UST,
60922cb0
DG
125 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
126 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
03550b58
MD
127 .err_sock = -1,
128 .cmd_sock = -1,
173af62f
DG
129 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
130 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
131 .cond = PTHREAD_COND_INITIALIZER,
132 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081
MD
133};
134
26296c48
JG
135/* Command line options */
136static const struct option long_options[] = {
0f5ea17c
JG
137 { "client-sock", required_argument, 0, 'c' },
138 { "apps-sock", required_argument, 0, 'a' },
139 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
140 { "kconsumerd-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
144 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
145 { "consumerd32-path", required_argument, 0, '\0' },
146 { "consumerd32-libdir", required_argument, 0, '\0' },
147 { "consumerd64-path", required_argument, 0, '\0' },
148 { "consumerd64-libdir", required_argument, 0, '\0' },
149 { "daemonize", no_argument, 0, 'd' },
150 { "background", no_argument, 0, 'b' },
151 { "sig-parent", no_argument, 0, 'S' },
152 { "help", no_argument, 0, 'h' },
153 { "group", required_argument, 0, 'g' },
154 { "version", no_argument, 0, 'V' },
155 { "quiet", no_argument, 0, 'q' },
156 { "verbose", no_argument, 0, 'v' },
157 { "verbose-consumer", no_argument, 0, '\0' },
158 { "no-kernel", no_argument, 0, '\0' },
159 { "pidfile", required_argument, 0, 'p' },
160 { "agent-tcp-port", required_argument, 0, '\0' },
161 { "config", required_argument, 0, 'f' },
162 { "load", required_argument, 0, 'l' },
163 { "kmod-probes", required_argument, 0, '\0' },
164 { "extra-kmod-probes", required_argument, 0, '\0' },
26296c48
JG
165 { NULL, 0, 0, 0 }
166};
167
168/* Command line options to ignore from configuration file */
169static const char *config_ignore_options[] = { "help", "version", "config" };
170
26c9d55e 171/* Shared between threads */
099e26bd 172static int dispatch_thread_exit;
fac6795d 173
54d01ffb
DG
174/* Global application Unix socket path */
175static char apps_unix_sock_path[PATH_MAX];
176/* Global client Unix socket path */
177static char client_unix_sock_path[PATH_MAX];
54d01ffb
DG
178/* global wait shm path for UST */
179static char wait_shm_path[PATH_MAX];
44a5e5eb
DG
180/* Global health check unix path */
181static char health_unix_sock_path[PATH_MAX];
fac6795d 182
1d4b027a 183/* Sockets and FDs */
a4b35e07
MD
184static int client_sock = -1;
185static int apps_sock = -1;
2f77fc4b 186int kernel_tracer_fd = -1;
76d7553f 187static int kernel_poll_pipe[2] = { -1, -1 };
1d4b027a 188
273ea72c
DG
189/*
190 * Quit pipe for all threads. This permits a single cancellation point
191 * for all threads when receiving an event on the pipe.
192 */
76d7553f 193static int thread_quit_pipe[2] = { -1, -1 };
4a15001e 194static int ht_cleanup_quit_pipe[2] = { -1, -1 };
273ea72c 195
099e26bd
DG
196/*
197 * This pipe is used to inform the thread managing application communication
198 * that a command is queued and ready to be processed.
199 */
76d7553f 200static int apps_cmd_pipe[2] = { -1, -1 };
099e26bd 201
d0b96690
DG
202int apps_cmd_notify_pipe[2] = { -1, -1 };
203
1d4b027a 204/* Pthread, Mutexes and Semaphores */
1d4b027a 205static pthread_t apps_thread;
d0b96690 206static pthread_t apps_notify_thread;
099e26bd 207static pthread_t reg_apps_thread;
1d4b027a 208static pthread_t client_thread;
7a485870 209static pthread_t kernel_thread;
099e26bd 210static pthread_t dispatch_thread;
44a5e5eb 211static pthread_t health_thread;
0b2dc8df 212static pthread_t ht_cleanup_thread;
022d91ba 213static pthread_t agent_reg_thread;
ef367a93 214static pthread_t load_session_thread;
5eb91c98 215
099e26bd
DG
216/*
217 * UST registration command queue. This queue is tied with a futex and uses a N
218 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 *
b22c5da8
DG
220 * The thread_registration_apps and thread_dispatch_ust_registration uses this
221 * queue along with the wait/wake scheme. The thread_manage_apps receives down
222 * the line new application socket and monitors it for any I/O error or clean
223 * close that triggers an unregistration of the application.
099e26bd
DG
224 */
225static struct ust_cmd_queue ust_cmd_queue;
226
b5541356
DG
227/*
228 * Pointer initialized before thread creation.
229 *
230 * This points to the tracing session list containing the session count and a
231 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
232 * MUST NOT be taken if you call a public function in session.c.
04ea676f 233 *
d063d709 234 * The lock is nested inside the structure: session_list_ptr->lock. Please use
54d01ffb 235 * session_lock_list and session_unlock_list for lock acquisition.
b5541356
DG
236 */
237static struct ltt_session_list *session_list_ptr;
238
7753dea8
MD
239int ust_consumerd64_fd = -1;
240int ust_consumerd32_fd = -1;
241
fb6f1fa2
YB
242static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
243static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
244static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
245static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
26296c48
JG
246static int consumerd32_bin_override;
247static int consumerd64_bin_override;
248static int consumerd32_libdir_override;
249static int consumerd64_libdir_override;
fb09408a 250
2f77fc4b
DG
251static const char *module_proc_lttng = "/proc/lttng";
252
5c827ce0
DG
253/*
254 * Consumer daemon state which is changed when spawning it, killing it or in
255 * case of a fatal error.
256 */
257enum consumerd_state {
258 CONSUMER_STARTED = 1,
259 CONSUMER_STOPPED = 2,
260 CONSUMER_ERROR = 3,
261};
262
263/*
264 * This consumer daemon state is used to validate if a client command will be
265 * able to reach the consumer. If not, the client is informed. For instance,
266 * doing a "lttng start" when the consumer state is set to ERROR will return an
267 * error to the client.
268 *
269 * The following example shows a possible race condition of this scheme:
270 *
271 * consumer thread error happens
272 * client cmd arrives
273 * client cmd checks state -> still OK
274 * consumer thread exit, sets error
275 * client cmd try to talk to consumer
276 * ...
277 *
278 * However, since the consumer is a different daemon, we have no way of making
279 * sure the command will reach it safely even with this state flag. This is why
280 * we consider that up to the state validation during command processing, the
281 * command is safe. After that, we can not guarantee the correctness of the
282 * client request vis-a-vis the consumer.
283 */
284static enum consumerd_state ust_consumerd_state;
285static enum consumerd_state kernel_consumerd_state;
286
ae9e45b3
DG
287/*
288 * Socket timeout for receiving and sending in seconds.
289 */
290static int app_socket_timeout;
291
12744796
DG
292/* Set in main() with the current page size. */
293long page_size;
294
8782cc74
MD
295/* Application health monitoring */
296struct health_app *health_sessiond;
297
022d91ba
DG
298/* Agent TCP port for registration. Used by the agent thread. */
299unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
4d076222 300
f43f95a9
DG
301/* Am I root or not. */
302int is_root; /* Set to 1 if the daemon is running as root */
303
26296c48
JG
304const char * const config_section_name = "sessiond";
305
ef367a93
JG
306/* Load session thread information to operate. */
307struct load_session_thread_data *load_info;
308
7c1d2758
JG
309/* Global hash tables */
310struct lttng_ht *agent_apps_ht_by_sock = NULL;
311
97bc1426
MD
312/*
313 * Whether sessiond is ready for commands/health check requests.
314 * NR_LTTNG_SESSIOND_READY must match the number of calls to
ef367a93 315 * sessiond_notify_ready().
97bc1426 316 */
ef367a93 317#define NR_LTTNG_SESSIOND_READY 3
97bc1426
MD
318int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
319
320/* Notify parents that we are ready for cmd and health check */
ef367a93
JG
321LTTNG_HIDDEN
322void sessiond_notify_ready(void)
97bc1426
MD
323{
324 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
325 /*
326 * Notify parent pid that we are ready to accept command
327 * for client side. This ppid is the one from the
328 * external process that spawned us.
329 */
330 if (opt_sig_parent) {
331 kill(ppid, SIGUSR1);
332 }
333
334 /*
335 * Notify the parent of the fork() process that we are
336 * ready.
337 */
72dd7491 338 if (opt_daemon || opt_background) {
97bc1426
MD
339 kill(child_ppid, SIGUSR1);
340 }
341 }
342}
343
fb09408a 344static
7753dea8 345void setup_consumerd_path(void)
fb09408a 346{
fc7a59ce 347 const char *bin, *libdir;
fb09408a 348
7753dea8
MD
349 /*
350 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
351 * native architecture size consumer if CONFIG_CONSUMER*_PATH
352 * has not been defined.
7753dea8 353 */
ebaeda94 354#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
355 if (!consumerd32_bin[0]) {
356 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
357 }
358 if (!consumerd32_libdir[0]) {
359 consumerd32_libdir = INSTALL_LIB_PATH;
360 }
361#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
362 if (!consumerd64_bin[0]) {
363 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 364 }
ebaeda94
MD
365 if (!consumerd64_libdir[0]) {
366 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
367 }
368#else
369#error "Unknown bitness"
370#endif
371
fb09408a
MD
372 /*
373 * runtime env. var. overrides the build default.
374 */
e8fa9fb0 375 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
fc7a59ce
AM
376 if (bin) {
377 consumerd32_bin = bin;
7753dea8 378 }
e8fa9fb0 379 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
fc7a59ce
AM
380 if (bin) {
381 consumerd64_bin = bin;
ebaeda94 382 }
e8fa9fb0 383 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
384 if (libdir) {
385 consumerd32_libdir = libdir;
386 }
e8fa9fb0 387 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
388 if (libdir) {
389 consumerd64_libdir = libdir;
fb09408a
MD
390 }
391}
392
4a15001e
MD
393static
394int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
395 int *a_pipe)
5eb91c98
DG
396{
397 int ret;
398
d0b96690 399 assert(events);
5eb91c98
DG
400
401 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
402 if (ret < 0) {
403 goto error;
404 }
405
406 /* Add quit pipe */
4a15001e 407 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
5eb91c98
DG
408 if (ret < 0) {
409 goto error;
410 }
411
412 return 0;
413
414error:
415 return ret;
416}
417
4a15001e
MD
418/*
419 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
420 */
421int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
422{
423 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
424}
425
426/*
427 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
428 */
429int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
430 size_t size)
431{
432 return __sessiond_set_thread_pollset(events, size,
433 ht_cleanup_quit_pipe);
434}
435
436static
437int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
438{
439 if (fd == a_pipe && (events & LPOLLIN)) {
440 return 1;
441 }
442 return 0;
443}
444
5eb91c98
DG
445/*
446 * Check if the thread quit pipe was triggered.
447 *
448 * Return 1 if it was triggered else 0;
449 */
d0b96690 450int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
5eb91c98 451{
4a15001e
MD
452 return __sessiond_check_thread_quit_pipe(fd, events,
453 thread_quit_pipe[0]);
454}
5eb91c98 455
4a15001e
MD
456/*
457 * Check if the ht_cleanup thread quit pipe was triggered.
458 *
459 * Return 1 if it was triggered else 0;
460 */
461int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
462{
463 return __sessiond_check_thread_quit_pipe(fd, events,
464 ht_cleanup_quit_pipe[0]);
5eb91c98
DG
465}
466
273ea72c 467/*
5eb91c98 468 * Init thread quit pipe.
273ea72c
DG
469 *
470 * Return -1 on error or 0 if all pipes are created.
471 */
4a15001e 472static int __init_thread_quit_pipe(int *a_pipe)
273ea72c 473{
730389d9 474 int ret, i;
273ea72c 475
4a15001e 476 ret = pipe(a_pipe);
273ea72c 477 if (ret < 0) {
730389d9 478 PERROR("thread quit pipe");
273ea72c
DG
479 goto error;
480 }
481
730389d9 482 for (i = 0; i < 2; i++) {
4a15001e 483 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
730389d9
DG
484 if (ret < 0) {
485 PERROR("fcntl");
486 goto error;
487 }
488 }
489
273ea72c
DG
490error:
491 return ret;
492}
493
4a15001e
MD
494static int init_thread_quit_pipe(void)
495{
496 return __init_thread_quit_pipe(thread_quit_pipe);
497}
498
499static int init_ht_cleanup_quit_pipe(void)
500{
501 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
502}
503
099e26bd
DG
504/*
505 * Stop all threads by closing the thread quit pipe.
506 */
cf3af59e
MD
507static void stop_threads(void)
508{
5eb91c98
DG
509 int ret;
510
cf3af59e
MD
511 /* Stopping all threads */
512 DBG("Terminating all threads");
54d01ffb 513 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
514 if (ret < 0) {
515 ERR("write error on thread quit pipe");
516 }
517
099e26bd 518 /* Dispatch thread */
26c9d55e 519 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 520 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
521}
522
e975f9f8
DG
523/*
524 * Close every consumer sockets.
525 */
526static void close_consumer_sockets(void)
527{
528 int ret;
529
530 if (kconsumer_data.err_sock >= 0) {
531 ret = close(kconsumer_data.err_sock);
532 if (ret < 0) {
533 PERROR("kernel consumer err_sock close");
534 }
535 }
536 if (ustconsumer32_data.err_sock >= 0) {
537 ret = close(ustconsumer32_data.err_sock);
538 if (ret < 0) {
a76cbd9f 539 PERROR("UST consumerd32 err_sock close");
e975f9f8
DG
540 }
541 }
542 if (ustconsumer64_data.err_sock >= 0) {
543 ret = close(ustconsumer64_data.err_sock);
544 if (ret < 0) {
a76cbd9f 545 PERROR("UST consumerd64 err_sock close");
e975f9f8
DG
546 }
547 }
548 if (kconsumer_data.cmd_sock >= 0) {
549 ret = close(kconsumer_data.cmd_sock);
550 if (ret < 0) {
551 PERROR("kernel consumer cmd_sock close");
552 }
553 }
554 if (ustconsumer32_data.cmd_sock >= 0) {
555 ret = close(ustconsumer32_data.cmd_sock);
556 if (ret < 0) {
a76cbd9f 557 PERROR("UST consumerd32 cmd_sock close");
e975f9f8
DG
558 }
559 }
560 if (ustconsumer64_data.cmd_sock >= 0) {
561 ret = close(ustconsumer64_data.cmd_sock);
562 if (ret < 0) {
a76cbd9f 563 PERROR("UST consumerd64 cmd_sock close");
e975f9f8
DG
564 }
565 }
566}
567
c9cb3e7d
JG
568/*
569 * Generate the full lock file path using the rundir.
570 *
571 * Return the snprintf() return value thus a negative value is an error.
572 */
573static int generate_lock_file_path(char *path, size_t len)
574{
575 int ret;
576
577 assert(path);
578 assert(rundir);
579
580 /* Build lockfile path from rundir. */
581 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
582 if (ret < 0) {
583 PERROR("snprintf lockfile path");
584 }
585
586 return ret;
587}
588
4e4714cb
JR
589/*
590 * Wait on consumer process termination.
591 *
592 * Need to be called with the consumer data lock held or from a context
593 * ensuring no concurrent access to data (e.g: cleanup).
594 */
595static void wait_consumer(struct consumer_data *consumer_data)
596{
597 pid_t ret;
598 int status;
599
600 if (consumer_data->pid <= 0) {
601 return;
602 }
603
604 DBG("Waiting for complete teardown of consumerd (PID: %d)",
605 consumer_data->pid);
606 ret = waitpid(consumer_data->pid, &status, 0);
607 if (ret == -1) {
608 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
609 }
610 if (!WIFEXITED(status)) {
611 ERR("consumerd termination with error: %d",
612 WEXITSTATUS(ret));
613 }
614 consumer_data->pid = 0;
615}
616
fac6795d 617/*
4a15001e 618 * Cleanup the session daemon's data structures.
fac6795d 619 */
4a15001e 620static void sessiond_cleanup(void)
fac6795d 621{
ef599319 622 int ret;
af9737e9 623 struct ltt_session *sess, *stmp;
8c6c56c2 624 char path[PATH_MAX];
fac6795d 625
4a15001e 626 DBG("Cleanup sessiond");
e07ae692 627
4e449f3f
MD
628 /*
629 * Close the thread quit pipe. It has already done its job,
630 * since we are now called.
631 */
2f77fc4b
DG
632 utils_close_pipe(thread_quit_pipe);
633
35f90c40
DG
634 /*
635 * If opt_pidfile is undefined, the default file will be wiped when
636 * removing the rundir.
637 */
638 if (opt_pidfile) {
639 ret = remove(opt_pidfile);
640 if (ret < 0) {
641 PERROR("remove pidfile %s", opt_pidfile);
642 }
643 }
644
8c6c56c2
MD
645 DBG("Removing sessiond and consumerd content of directory %s", rundir);
646
647 /* sessiond */
648 snprintf(path, PATH_MAX,
649 "%s/%s",
650 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
651 DBG("Removing %s", path);
652 (void) unlink(path);
653
cd9290dd 654 snprintf(path, PATH_MAX, "%s/%s", rundir,
022d91ba 655 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
cd9290dd
DG
656 DBG("Removing %s", path);
657 (void) unlink(path);
658
8c6c56c2
MD
659 /* kconsumerd */
660 snprintf(path, PATH_MAX,
661 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
662 rundir);
663 DBG("Removing %s", path);
664 (void) unlink(path);
665
666 snprintf(path, PATH_MAX,
667 DEFAULT_KCONSUMERD_PATH,
668 rundir);
669 DBG("Removing directory %s", path);
670 (void) rmdir(path);
671
672 /* ust consumerd 32 */
673 snprintf(path, PATH_MAX,
674 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
675 rundir);
676 DBG("Removing %s", path);
677 (void) unlink(path);
678
679 snprintf(path, PATH_MAX,
680 DEFAULT_USTCONSUMERD32_PATH,
681 rundir);
682 DBG("Removing directory %s", path);
683 (void) rmdir(path);
684
685 /* ust consumerd 64 */
686 snprintf(path, PATH_MAX,
687 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
688 rundir);
689 DBG("Removing %s", path);
690 (void) unlink(path);
691
692 snprintf(path, PATH_MAX,
693 DEFAULT_USTCONSUMERD64_PATH,
694 rundir);
695 DBG("Removing directory %s", path);
696 (void) rmdir(path);
5461b305 697
99bab54f 698 DBG("Cleaning up all sessions");
fac6795d 699
b5541356 700 /* Destroy session list mutex */
273ea72c
DG
701 if (session_list_ptr != NULL) {
702 pthread_mutex_destroy(&session_list_ptr->lock);
703
704 /* Cleanup ALL session */
54d01ffb
DG
705 cds_list_for_each_entry_safe(sess, stmp,
706 &session_list_ptr->head, list) {
2f77fc4b 707 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
708 }
709 }
710
4e4714cb
JR
711 wait_consumer(&kconsumer_data);
712 wait_consumer(&ustconsumer64_data);
713 wait_consumer(&ustconsumer32_data);
714
6a4e4039
JG
715 DBG("Cleaning up all agent apps");
716 agent_app_ht_clean();
717
099e26bd 718 DBG("Closing all UST sockets");
56fff090 719 ust_app_clean_list();
7972aab2 720 buffer_reg_destroy_registries();
099e26bd 721
4fba7219
DG
722 if (is_root && !opt_no_kernel) {
723 DBG2("Closing kernel fd");
a4b35e07 724 if (kernel_tracer_fd >= 0) {
76d7553f
MD
725 ret = close(kernel_tracer_fd);
726 if (ret) {
727 PERROR("close");
728 }
a4b35e07 729 }
2f50c8a3 730 DBG("Unloading kernel modules");
096102bd 731 modprobe_remove_lttng_all();
834978fd 732 free(syscall_table);
2f50c8a3 733 }
2f77fc4b 734
e975f9f8
DG
735 close_consumer_sockets();
736
ef367a93
JG
737 if (load_info) {
738 load_session_destroy_data(load_info);
739 free(load_info);
740 }
741
c9cb3e7d
JG
742 /*
743 * Cleanup lock file by deleting it and finaly closing it which will
744 * release the file system lock.
745 */
746 if (lockfile_fd >= 0) {
747 char lockfile_path[PATH_MAX];
748
4a15001e
MD
749 ret = generate_lock_file_path(lockfile_path,
750 sizeof(lockfile_path));
c9cb3e7d
JG
751 if (ret > 0) {
752 ret = remove(lockfile_path);
753 if (ret < 0) {
754 PERROR("remove lock file");
755 }
756 ret = close(lockfile_fd);
757 if (ret < 0) {
758 PERROR("close lock file");
759 }
760 }
761 }
762
763 /*
764 * We do NOT rmdir rundir because there are other processes
765 * using it, for instance lttng-relayd, which can start in
766 * parallel with this teardown.
767 */
768
769 free(rundir);
4a15001e
MD
770}
771
772/*
773 * Cleanup the daemon's option data structures.
774 */
775static void sessiond_cleanup_options(void)
776{
777 DBG("Cleaning up options");
778
779 /*
780 * If the override option is set, the pointer points to a *non* const
781 * thus freeing it even though the variable type is set to const.
782 */
783 if (tracing_group_name_override) {
784 free((void *) tracing_group_name);
785 }
786 if (consumerd32_bin_override) {
787 free((void *) consumerd32_bin);
788 }
789 if (consumerd64_bin_override) {
790 free((void *) consumerd64_bin);
791 }
792 if (consumerd32_libdir_override) {
793 free((void *) consumerd32_libdir);
794 }
795 if (consumerd64_libdir_override) {
796 free((void *) consumerd64_libdir);
797 }
798
799 free(opt_pidfile);
800 free(opt_load_session_path);
801 free(kmod_probes_list);
802 free(kmod_extra_probes_list);
c9cb3e7d 803
421cb601 804 /* <fun> */
f56a39af 805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 /* </fun> */
fac6795d
DG
809}
810
e065084a 811/*
d063d709 812 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 813 *
d063d709 814 * Return lttcomm error code.
e065084a
DG
815 */
816static int send_unix_sock(int sock, void *buf, size_t len)
817{
818 /* Check valid length */
c617c0c6 819 if (len == 0) {
e065084a
DG
820 return -1;
821 }
822
823 return lttcomm_send_unix_sock(sock, buf, len);
824}
825
5461b305 826/*
d063d709 827 * Free memory of a command context structure.
5461b305 828 */
a2fb29a5 829static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 830{
a2fb29a5
DG
831 DBG("Clean command context structure");
832 if (*cmd_ctx) {
833 if ((*cmd_ctx)->llm) {
834 free((*cmd_ctx)->llm);
5461b305 835 }
a2fb29a5
DG
836 if ((*cmd_ctx)->lsm) {
837 free((*cmd_ctx)->lsm);
5461b305 838 }
a2fb29a5
DG
839 free(*cmd_ctx);
840 *cmd_ctx = NULL;
5461b305
DG
841 }
842}
843
fac6795d 844/*
0fdd1e2c 845 * Notify UST applications using the shm mmap futex.
fac6795d 846 */
0fdd1e2c 847static int notify_ust_apps(int active)
fac6795d 848{
0fdd1e2c 849 char *wait_shm_mmap;
fac6795d 850
0fdd1e2c 851 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 852
0fdd1e2c
DG
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
855 if (wait_shm_mmap == NULL) {
fac6795d
DG
856 goto error;
857 }
858
0fdd1e2c
DG
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap, active);
861
862 /* Apps notified successfully */
863 return 0;
fac6795d
DG
864
865error:
0fdd1e2c 866 return -1;
fac6795d
DG
867}
868
e065084a 869/*
d063d709
DG
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
872 * structure.
ca95a216 873 *
d063d709 874 * Return total size of the buffer pointed by buf.
ca95a216 875 */
5461b305 876static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 877{
f3ed775e 878 int ret, buf_size;
ca95a216 879
f3ed775e 880 buf_size = size;
5461b305 881
ba7f0ae5 882 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 883 if (cmd_ctx->llm == NULL) {
76d7553f 884 PERROR("zmalloc");
5461b305 885 ret = -ENOMEM;
ca95a216
DG
886 goto error;
887 }
888
5461b305
DG
889 /* Copy common data */
890 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 891 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 892
5461b305
DG
893 cmd_ctx->llm->data_size = size;
894 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
895
ca95a216
DG
896 return buf_size;
897
898error:
899 return ret;
900}
901
7a485870 902/*
5eb91c98 903 * Update the kernel poll set of all channel fd available over all tracing
d063d709 904 * session. Add the wakeup pipe at the end of the set.
7a485870 905 */
5eb91c98 906static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 907{
5eb91c98 908 int ret;
7a485870
DG
909 struct ltt_session *session;
910 struct ltt_kernel_channel *channel;
911
5eb91c98 912 DBG("Updating kernel poll set");
7a485870 913
54d01ffb 914 session_lock_list();
b5541356 915 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 916 session_lock(session);
7a485870 917 if (session->kernel_session == NULL) {
54d01ffb 918 session_unlock(session);
7a485870
DG
919 continue;
920 }
7a485870 921
54d01ffb
DG
922 cds_list_for_each_entry(channel,
923 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
924 /* Add channel fd to the kernel poll set */
925 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
926 if (ret < 0) {
54d01ffb 927 session_unlock(session);
5eb91c98
DG
928 goto error;
929 }
930 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 931 }
54d01ffb 932 session_unlock(session);
7a485870 933 }
54d01ffb 934 session_unlock_list();
7a485870 935
5eb91c98 936 return 0;
7a485870
DG
937
938error:
54d01ffb 939 session_unlock_list();
7a485870
DG
940 return -1;
941}
942
943/*
54d01ffb 944 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 945 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 946 *
d063d709 947 * Useful for CPU hotplug feature.
7a485870 948 */
2bdd86d4 949static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
950{
951 int ret = 0;
952 struct ltt_session *session;
173af62f 953 struct ltt_kernel_session *ksess;
7a485870
DG
954 struct ltt_kernel_channel *channel;
955
956 DBG("Updating kernel streams for channel fd %d", fd);
957
54d01ffb 958 session_lock_list();
b5541356 959 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 960 session_lock(session);
7a485870 961 if (session->kernel_session == NULL) {
54d01ffb 962 session_unlock(session);
7a485870
DG
963 continue;
964 }
173af62f 965 ksess = session->kernel_session;
d9800920 966
4a15001e
MD
967 cds_list_for_each_entry(channel,
968 &ksess->channel_list.head, list) {
969 struct lttng_ht_iter iter;
970 struct consumer_socket *socket;
d9800920 971
4a15001e
MD
972 if (channel->fd != fd) {
973 continue;
974 }
975 DBG("Channel found, updating kernel streams");
976 ret = kernel_open_channel_stream(channel);
977 if (ret < 0) {
978 goto error;
979 }
980 /* Update the stream global counter */
981 ksess->stream_count_global += ret;
982
983 /*
984 * Have we already sent fds to the consumer? If yes, it
985 * means that tracing is started so it is safe to send
986 * our updated stream fds.
987 */
988 if (ksess->consumer_fds_sent != 1
989 || ksess->consumer == NULL) {
990 ret = -1;
991 goto error;
992 }
993
994 rcu_read_lock();
995 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
996 &iter.iter, socket, node.node) {
997 pthread_mutex_lock(socket->lock);
998 ret = kernel_consumer_send_channel_stream(socket,
999 channel, ksess,
1000 session->output_traces ? 1 : 0);
1001 pthread_mutex_unlock(socket->lock);
1002 if (ret < 0) {
e7fe706f 1003 rcu_read_unlock();
4a15001e 1004 goto error;
7a485870 1005 }
7a485870 1006 }
4a15001e 1007 rcu_read_unlock();
7a485870 1008 }
54d01ffb 1009 session_unlock(session);
7a485870 1010 }
54d01ffb 1011 session_unlock_list();
b3c750d2 1012 return ret;
7a485870 1013
b3c750d2 1014error:
54d01ffb
DG
1015 session_unlock(session);
1016 session_unlock_list();
7a485870
DG
1017 return ret;
1018}
1019
487cf67c 1020/*
ffe60014
DG
1021 * For each tracing session, update newly registered apps. The session list
1022 * lock MUST be acquired before calling this.
487cf67c
DG
1023 */
1024static void update_ust_app(int app_sock)
1025{
1026 struct ltt_session *sess, *stmp;
1027
fdadac08
DG
1028 /* Consumer is in an ERROR state. Stop any application update. */
1029 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1030 /* Stop the update process since the consumer is dead. */
1031 return;
1032 }
1033
487cf67c
DG
1034 /* For all tracing session(s) */
1035 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
a9ad0c8f
MD
1036 struct ust_app *app;
1037
4ee14516 1038 session_lock(sess);
a9ad0c8f
MD
1039 if (!sess->ust_session) {
1040 goto unlock_session;
1041 }
1042
1043 rcu_read_lock();
1044 assert(app_sock >= 0);
1045 app = ust_app_find_by_sock(app_sock);
1046 if (app == NULL) {
1047 /*
1048 * Application can be unregistered before so
1049 * this is possible hence simply stopping the
1050 * update.
1051 */
1052 DBG3("UST app update failed to find app sock %d",
1053 app_sock);
1054 goto unlock_rcu;
421cb601 1055 }
a9ad0c8f
MD
1056 ust_app_global_update(sess->ust_session, app);
1057 unlock_rcu:
1058 rcu_read_unlock();
1059 unlock_session:
4ee14516 1060 session_unlock(sess);
487cf67c
DG
1061 }
1062}
1063
7a485870 1064/*
d063d709 1065 * This thread manage event coming from the kernel.
7a485870 1066 *
d063d709
DG
1067 * Features supported in this thread:
1068 * -) CPU Hotplug
7a485870
DG
1069 */
1070static void *thread_manage_kernel(void *data)
1071{
139ac872 1072 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 1073 uint32_t revents, nb_fd;
7a485870 1074 char tmp;
5eb91c98 1075 struct lttng_poll_event events;
7a485870 1076
6993eeb3 1077 DBG("[thread] Thread manage kernel started");
7a485870 1078
6c71277b 1079 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
927ca06a 1080
d5d63bf1
DG
1081 /*
1082 * This first step of the while is to clean this structure which could free
6d737ce4 1083 * non NULL pointers so initialize it before the loop.
d5d63bf1 1084 */
6d737ce4 1085 lttng_poll_init(&events);
d5d63bf1 1086
e547b070 1087 if (testpoint(sessiond_thread_manage_kernel)) {
6993eeb3
CB
1088 goto error_testpoint;
1089 }
8ac94142 1090
840cb59c 1091 health_code_update();
44a5e5eb 1092
e547b070 1093 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
d21b0d71 1094 goto error_testpoint;
6993eeb3
CB
1095 }
1096
7a485870 1097 while (1) {
840cb59c 1098 health_code_update();
44a5e5eb 1099
7a485870 1100 if (update_poll_flag == 1) {
d21b0d71
DG
1101 /* Clean events object. We are about to populate it again. */
1102 lttng_poll_clean(&events);
1103
d0b96690 1104 ret = sessiond_set_thread_pollset(&events, 2);
d21b0d71
DG
1105 if (ret < 0) {
1106 goto error_poll_create;
1107 }
1108
1109 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1110 if (ret < 0) {
1111 goto error;
1112 }
5f822d0a 1113
d21b0d71 1114 /* This will add the available kernel channel if any. */
5eb91c98
DG
1115 ret = update_kernel_poll(&events);
1116 if (ret < 0) {
7a485870
DG
1117 goto error;
1118 }
1119 update_poll_flag = 0;
1120 }
1121
7fa2082e 1122 DBG("Thread kernel polling");
7a485870
DG
1123
1124 /* Poll infinite value of time */
88f2b785 1125 restart:
a78af745 1126 health_poll_entry();
5eb91c98 1127 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1128 DBG("Thread kernel return from poll on %d fds",
1129 LTTNG_POLL_GETNB(&events));
a78af745 1130 health_poll_exit();
7a485870 1131 if (ret < 0) {
88f2b785
MD
1132 /*
1133 * Restart interrupted system call.
1134 */
1135 if (errno == EINTR) {
1136 goto restart;
1137 }
7a485870
DG
1138 goto error;
1139 } else if (ret == 0) {
1140 /* Should not happen since timeout is infinite */
85611738
DG
1141 ERR("Return value of poll is 0 with an infinite timeout.\n"
1142 "This should not have happened! Continuing...");
7a485870
DG
1143 continue;
1144 }
1145
0d9c5d77
DG
1146 nb_fd = ret;
1147
5eb91c98
DG
1148 for (i = 0; i < nb_fd; i++) {
1149 /* Fetch once the poll data */
1150 revents = LTTNG_POLL_GETEV(&events, i);
1151 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 1152
840cb59c 1153 health_code_update();
44a5e5eb 1154
fd20dac9
MD
1155 if (!revents) {
1156 /* No activity for this FD (poll implementation). */
1157 continue;
1158 }
1159
5eb91c98 1160 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1161 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1162 if (ret) {
139ac872
MD
1163 err = 0;
1164 goto exit;
5eb91c98 1165 }
7a485870 1166
5eb91c98
DG
1167 /* Check for data on kernel pipe */
1168 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
6cd525e8
MD
1169 (void) lttng_read(kernel_poll_pipe[0],
1170 &tmp, 1);
f921c78f
DG
1171 /*
1172 * Ret value is useless here, if this pipe gets any actions an
1173 * update is required anyway.
1174 */
5eb91c98
DG
1175 update_poll_flag = 1;
1176 continue;
1177 } else {
1178 /*
1179 * New CPU detected by the kernel. Adding kernel stream to
1180 * kernel session and updating the kernel consumer
1181 */
1182 if (revents & LPOLLIN) {
2bdd86d4 1183 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
1184 if (ret < 0) {
1185 continue;
1186 }
1187 break;
1188 /*
1189 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1190 * and unregister kernel stream at this point.
1191 */
7a485870 1192 }
7a485870
DG
1193 }
1194 }
1195 }
1196
139ac872 1197exit:
7a485870 1198error:
5eb91c98 1199 lttng_poll_clean(&events);
76d7553f 1200error_poll_create:
6993eeb3 1201error_testpoint:
6620da75
DG
1202 utils_close_pipe(kernel_poll_pipe);
1203 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 1204 if (err) {
840cb59c 1205 health_error();
139ac872 1206 ERR("Health error occurred in %s", __func__);
6620da75
DG
1207 WARN("Kernel thread died unexpectedly. "
1208 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 1209 }
8782cc74 1210 health_unregister(health_sessiond);
76d7553f 1211 DBG("Kernel thread dying");
7a485870
DG
1212 return NULL;
1213}
1214
a23ec3a7
DG
1215/*
1216 * Signal pthread condition of the consumer data that the thread.
1217 */
1218static void signal_consumer_condition(struct consumer_data *data, int state)
1219{
1220 pthread_mutex_lock(&data->cond_mutex);
1221
1222 /*
1223 * The state is set before signaling. It can be any value, it's the waiter
1224 * job to correctly interpret this condition variable associated to the
1225 * consumer pthread_cond.
1226 *
1227 * A value of 0 means that the corresponding thread of the consumer data
1228 * was not started. 1 indicates that the thread has started and is ready
1229 * for action. A negative value means that there was an error during the
1230 * thread bootstrap.
1231 */
1232 data->consumer_thread_is_ready = state;
1233 (void) pthread_cond_signal(&data->cond);
1234
1235 pthread_mutex_unlock(&data->cond_mutex);
1236}
1237
1d4b027a 1238/*
3bd1e081 1239 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 1240 */
3bd1e081 1241static void *thread_manage_consumer(void *data)
1d4b027a 1242{
42fc1d0b 1243 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
5eb91c98 1244 uint32_t revents, nb_fd;
1d4b027a 1245 enum lttcomm_return_code code;
5eb91c98 1246 struct lttng_poll_event events;
3bd1e081 1247 struct consumer_data *consumer_data = data;
1d4b027a 1248
3bd1e081 1249 DBG("[thread] Manage consumer started");
1d4b027a 1250
34c1e15a
MD
1251 rcu_register_thread();
1252 rcu_thread_online();
1253
6c71277b 1254 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
927ca06a 1255
855060f8 1256 health_code_update();
9449cc75 1257
5eb91c98 1258 /*
331744e3
JD
1259 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1260 * metadata_sock. Nothing more will be added to this poll set.
5eb91c98 1261 */
331744e3 1262 ret = sessiond_set_thread_pollset(&events, 3);
5eb91c98 1263 if (ret < 0) {
76d7553f 1264 goto error_poll;
5eb91c98 1265 }
273ea72c 1266
edb8b045
DG
1267 /*
1268 * The error socket here is already in a listening state which was done
1269 * just before spawning this thread to avoid a race between the consumer
1270 * daemon exec trying to connect and the listen() call.
1271 */
3bd1e081 1272 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
1273 if (ret < 0) {
1274 goto error;
1275 }
1276
840cb59c 1277 health_code_update();
44a5e5eb 1278
331744e3 1279 /* Infinite blocking call, waiting for transmission */
88f2b785 1280restart:
a78af745 1281 health_poll_entry();
8ac94142 1282
e547b070 1283 if (testpoint(sessiond_thread_manage_consumer)) {
6993eeb3
CB
1284 goto error;
1285 }
8ac94142 1286
5eb91c98 1287 ret = lttng_poll_wait(&events, -1);
a78af745 1288 health_poll_exit();
273ea72c 1289 if (ret < 0) {
88f2b785
MD
1290 /*
1291 * Restart interrupted system call.
1292 */
1293 if (errno == EINTR) {
1294 goto restart;
1295 }
273ea72c
DG
1296 goto error;
1297 }
1298
0d9c5d77
DG
1299 nb_fd = ret;
1300
5eb91c98
DG
1301 for (i = 0; i < nb_fd; i++) {
1302 /* Fetch once the poll data */
1303 revents = LTTNG_POLL_GETEV(&events, i);
1304 pollfd = LTTNG_POLL_GETFD(&events, i);
1305
840cb59c 1306 health_code_update();
44a5e5eb 1307
fd20dac9
MD
1308 if (!revents) {
1309 /* No activity for this FD (poll implementation). */
1310 continue;
1311 }
1312
5eb91c98 1313 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1314 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1315 if (ret) {
139ac872
MD
1316 err = 0;
1317 goto exit;
5eb91c98
DG
1318 }
1319
1320 /* Event on the registration socket */
3bd1e081 1321 if (pollfd == consumer_data->err_sock) {
5eb91c98 1322 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1323 ERR("consumer err socket poll error");
5eb91c98
DG
1324 goto error;
1325 }
1326 }
273ea72c
DG
1327 }
1328
3bd1e081 1329 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
1330 if (sock < 0) {
1331 goto error;
1332 }
1333
b662582b
DG
1334 /*
1335 * Set the CLOEXEC flag. Return code is useless because either way, the
1336 * show must go on.
1337 */
1338 (void) utils_set_fd_cloexec(sock);
1339
840cb59c 1340 health_code_update();
44a5e5eb 1341
3bd1e081 1342 DBG2("Receiving code from consumer err_sock");
ee0b0061 1343
712ea556 1344 /* Getting status code from kconsumerd */
54d01ffb
DG
1345 ret = lttcomm_recv_unix_sock(sock, &code,
1346 sizeof(enum lttcomm_return_code));
1d4b027a
DG
1347 if (ret <= 0) {
1348 goto error;
1349 }
1350
840cb59c 1351 health_code_update();
f73fabfd 1352 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
331744e3 1353 /* Connect both socket, command and metadata. */
3bd1e081
MD
1354 consumer_data->cmd_sock =
1355 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
4ce514c4 1356 consumer_data->metadata_fd =
331744e3 1357 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
92db7cdc
DG
1358 if (consumer_data->cmd_sock < 0
1359 || consumer_data->metadata_fd < 0) {
331744e3 1360 PERROR("consumer connect cmd socket");
a23ec3a7
DG
1361 /* On error, signal condition and quit. */
1362 signal_consumer_condition(consumer_data, -1);
1d4b027a
DG
1363 goto error;
1364 }
9363801e 1365 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
331744e3
JD
1366 /* Create metadata socket lock. */
1367 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1368 if (consumer_data->metadata_sock.lock == NULL) {
1369 PERROR("zmalloc pthread mutex");
1370 ret = -1;
1371 goto error;
1372 }
1373 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1374
a23ec3a7 1375 signal_consumer_condition(consumer_data, 1);
331744e3
JD
1376 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1377 DBG("Consumer metadata socket ready (fd: %d)",
4ce514c4 1378 consumer_data->metadata_fd);
1d4b027a 1379 } else {
3bd1e081 1380 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
1381 lttcomm_get_readable_code(-code));
1382 goto error;
1383 }
1384
331744e3 1385 /* Remove the consumerd error sock since we've established a connexion */
3bd1e081 1386 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1387 if (ret < 0) {
72079cae
DG
1388 goto error;
1389 }
1390
331744e3 1391 /* Add new accepted error socket. */
5eb91c98
DG
1392 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1393 if (ret < 0) {
72079cae 1394 goto error;
5eb91c98
DG
1395 }
1396
331744e3 1397 /* Add metadata socket that is successfully connected. */
4ce514c4 1398 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
331744e3
JD
1399 LPOLLIN | LPOLLRDHUP);
1400 if (ret < 0) {
1401 goto error;
1402 }
1403
840cb59c 1404 health_code_update();
44a5e5eb 1405
331744e3 1406 /* Infinite blocking call, waiting for transmission */
88f2b785 1407restart_poll:
331744e3 1408 while (1) {
42fc1d0b
DG
1409 health_code_update();
1410
1411 /* Exit the thread because the thread quit pipe has been triggered. */
1412 if (should_quit) {
1413 /* Not a health error. */
1414 err = 0;
1415 goto exit;
1416 }
1417
331744e3
JD
1418 health_poll_entry();
1419 ret = lttng_poll_wait(&events, -1);
1420 health_poll_exit();
1421 if (ret < 0) {
1422 /*
1423 * Restart interrupted system call.
1424 */
1425 if (errno == EINTR) {
1426 goto restart_poll;
1427 }
1428 goto error;
88f2b785 1429 }
72079cae 1430
331744e3 1431 nb_fd = ret;
0d9c5d77 1432
331744e3
JD
1433 for (i = 0; i < nb_fd; i++) {
1434 /* Fetch once the poll data */
1435 revents = LTTNG_POLL_GETEV(&events, i);
1436 pollfd = LTTNG_POLL_GETFD(&events, i);
5eb91c98 1437
331744e3 1438 health_code_update();
44a5e5eb 1439
fd20dac9
MD
1440 if (!revents) {
1441 /* No activity for this FD (poll implementation). */
1442 continue;
1443 }
1444
42fc1d0b
DG
1445 /*
1446 * Thread quit pipe has been triggered, flag that we should stop
1447 * but continue the current loop to handle potential data from
1448 * consumer.
1449 */
1450 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1451
331744e3
JD
1452 if (pollfd == sock) {
1453 /* Event on the consumerd socket */
1454 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1455 ERR("consumer err socket second poll error");
1456 goto error;
1457 }
1458 health_code_update();
1459 /* Wait for any kconsumerd error */
1460 ret = lttcomm_recv_unix_sock(sock, &code,
1461 sizeof(enum lttcomm_return_code));
1462 if (ret <= 0) {
1463 ERR("consumer closed the command socket");
1464 goto error;
1465 }
1466
1467 ERR("consumer return code : %s",
1468 lttcomm_get_readable_code(-code));
1469
1470 goto exit;
4ce514c4 1471 } else if (pollfd == consumer_data->metadata_fd) {
331744e3
JD
1472 /* UST metadata requests */
1473 ret = ust_consumer_metadata_request(
1474 &consumer_data->metadata_sock);
1475 if (ret < 0) {
1476 ERR("Handling metadata request");
1477 goto error;
1478 }
5eb91c98 1479 }
42fc1d0b 1480 /* No need for an else branch all FDs are tested prior. */
5eb91c98 1481 }
331744e3 1482 health_code_update();
5eb91c98
DG
1483 }
1484
139ac872 1485exit:
1d4b027a 1486error:
fdadac08
DG
1487 /*
1488 * We lock here because we are about to close the sockets and some other
92db7cdc
DG
1489 * thread might be using them so get exclusive access which will abort all
1490 * other consumer command by other threads.
fdadac08
DG
1491 */
1492 pthread_mutex_lock(&consumer_data->lock);
1493
5c827ce0
DG
1494 /* Immediately set the consumerd state to stopped */
1495 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1496 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1497 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1498 consumer_data->type == LTTNG_CONSUMER32_UST) {
1499 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1500 } else {
1501 /* Code flow error... */
1502 assert(0);
1503 }
1504
76d7553f
MD
1505 if (consumer_data->err_sock >= 0) {
1506 ret = close(consumer_data->err_sock);
1507 if (ret) {
1508 PERROR("close");
1509 }
a76cbd9f 1510 consumer_data->err_sock = -1;
76d7553f
MD
1511 }
1512 if (consumer_data->cmd_sock >= 0) {
1513 ret = close(consumer_data->cmd_sock);
1514 if (ret) {
1515 PERROR("close");
1516 }
a76cbd9f 1517 consumer_data->cmd_sock = -1;
76d7553f 1518 }
96544455
SS
1519 if (consumer_data->metadata_sock.fd_ptr &&
1520 *consumer_data->metadata_sock.fd_ptr >= 0) {
9363801e 1521 ret = close(*consumer_data->metadata_sock.fd_ptr);
331744e3
JD
1522 if (ret) {
1523 PERROR("close");
1524 }
1525 }
76d7553f
MD
1526 if (sock >= 0) {
1527 ret = close(sock);
1528 if (ret) {
1529 PERROR("close");
1530 }
1531 }
273ea72c 1532
3bd1e081
MD
1533 unlink(consumer_data->err_unix_sock_path);
1534 unlink(consumer_data->cmd_unix_sock_path);
fdadac08 1535 pthread_mutex_unlock(&consumer_data->lock);
92db7cdc 1536
fdadac08 1537 /* Cleanup metadata socket mutex. */
96544455
SS
1538 if (consumer_data->metadata_sock.lock) {
1539 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1540 free(consumer_data->metadata_sock.lock);
1541 }
5eb91c98 1542 lttng_poll_clean(&events);
76d7553f 1543error_poll:
139ac872 1544 if (err) {
840cb59c 1545 health_error();
139ac872
MD
1546 ERR("Health error occurred in %s", __func__);
1547 }
8782cc74 1548 health_unregister(health_sessiond);
76d7553f 1549 DBG("consumer thread cleanup completed");
0177d773 1550
34c1e15a
MD
1551 rcu_thread_offline();
1552 rcu_unregister_thread();
1553
5eb91c98 1554 return NULL;
099e26bd
DG
1555}
1556
099e26bd
DG
1557/*
1558 * This thread manage application communication.
1d4b027a
DG
1559 */
1560static void *thread_manage_apps(void *data)
099e26bd 1561{
139ac872 1562 int i, ret, pollfd, err = -1;
6cd525e8 1563 ssize_t size_ret;
5eb91c98 1564 uint32_t revents, nb_fd;
5eb91c98 1565 struct lttng_poll_event events;
099e26bd
DG
1566
1567 DBG("[thread] Manage application started");
1568
f6a9efaa
DG
1569 rcu_register_thread();
1570 rcu_thread_online();
1571
6c71277b 1572 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
927ca06a 1573
e547b070 1574 if (testpoint(sessiond_thread_manage_apps)) {
6993eeb3
CB
1575 goto error_testpoint;
1576 }
1577
840cb59c 1578 health_code_update();
44a5e5eb 1579
d0b96690 1580 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1581 if (ret < 0) {
76d7553f 1582 goto error_poll_create;
5eb91c98 1583 }
099e26bd 1584
5eb91c98
DG
1585 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1586 if (ret < 0) {
1587 goto error;
1588 }
099e26bd 1589
e547b070 1590 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
6993eeb3
CB
1591 goto error;
1592 }
8ac94142 1593
840cb59c 1594 health_code_update();
44a5e5eb 1595
5eb91c98 1596 while (1) {
7fa2082e 1597 DBG("Apps thread polling");
099e26bd
DG
1598
1599 /* Inifinite blocking call, waiting for transmission */
88f2b785 1600 restart:
a78af745 1601 health_poll_entry();
5eb91c98 1602 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1603 DBG("Apps thread return from poll on %d fds",
1604 LTTNG_POLL_GETNB(&events));
a78af745 1605 health_poll_exit();
099e26bd 1606 if (ret < 0) {
88f2b785
MD
1607 /*
1608 * Restart interrupted system call.
1609 */
1610 if (errno == EINTR) {
1611 goto restart;
1612 }
099e26bd
DG
1613 goto error;
1614 }
1615
0d9c5d77
DG
1616 nb_fd = ret;
1617
5eb91c98
DG
1618 for (i = 0; i < nb_fd; i++) {
1619 /* Fetch once the poll data */
1620 revents = LTTNG_POLL_GETEV(&events, i);
1621 pollfd = LTTNG_POLL_GETFD(&events, i);
1622
840cb59c 1623 health_code_update();
44a5e5eb 1624
fd20dac9
MD
1625 if (!revents) {
1626 /* No activity for this FD (poll implementation). */
1627 continue;
1628 }
1629
5eb91c98 1630 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1631 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1632 if (ret) {
139ac872
MD
1633 err = 0;
1634 goto exit;
5eb91c98 1635 }
099e26bd 1636
5eb91c98
DG
1637 /* Inspect the apps cmd pipe */
1638 if (pollfd == apps_cmd_pipe[0]) {
1639 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1640 ERR("Apps command pipe error");
0177d773 1641 goto error;
5eb91c98 1642 } else if (revents & LPOLLIN) {
d0b96690
DG
1643 int sock;
1644
5eb91c98 1645 /* Empty pipe */
6cd525e8
MD
1646 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1647 if (size_ret < sizeof(sock)) {
76d7553f 1648 PERROR("read apps cmd pipe");
5eb91c98
DG
1649 goto error;
1650 }
099e26bd 1651
840cb59c 1652 health_code_update();
44a5e5eb 1653
ffe60014 1654 /*
d0b96690
DG
1655 * We only monitor the error events of the socket. This
1656 * thread does not handle any incoming data from UST
1657 * (POLLIN).
ffe60014 1658 */
d0b96690
DG
1659 ret = lttng_poll_add(&events, sock,
1660 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1661 if (ret < 0) {
5eb91c98 1662 goto error;
e0c7ec2b 1663 }
acc7b41b 1664
d0b96690 1665 DBG("Apps with sock %d added to poll set", sock);
0177d773 1666 }
5eb91c98
DG
1667 } else {
1668 /*
54d01ffb
DG
1669 * At this point, we know that a registered application made
1670 * the event at poll_wait.
5eb91c98
DG
1671 */
1672 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1673 /* Removing from the poll set */
1674 ret = lttng_poll_del(&events, pollfd);
1675 if (ret < 0) {
1676 goto error;
1677 }
099e26bd 1678
b9d9b220 1679 /* Socket closed on remote end. */
56fff090 1680 ust_app_unregister(pollfd);
5eb91c98 1681 }
099e26bd 1682 }
44a5e5eb 1683
840cb59c 1684 health_code_update();
099e26bd 1685 }
099e26bd
DG
1686 }
1687
139ac872 1688exit:
099e26bd 1689error:
5eb91c98 1690 lttng_poll_clean(&events);
76d7553f 1691error_poll_create:
6993eeb3 1692error_testpoint:
6620da75
DG
1693 utils_close_pipe(apps_cmd_pipe);
1694 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1695
1696 /*
1697 * We don't clean the UST app hash table here since already registered
1698 * applications can still be controlled so let them be until the session
1699 * daemon dies or the applications stop.
1700 */
1701
139ac872 1702 if (err) {
840cb59c 1703 health_error();
139ac872
MD
1704 ERR("Health error occurred in %s", __func__);
1705 }
8782cc74 1706 health_unregister(health_sessiond);
76d7553f 1707 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1708 rcu_thread_offline();
1709 rcu_unregister_thread();
099e26bd
DG
1710 return NULL;
1711}
1712
d0b96690 1713/*
d88aee68
DG
1714 * Send a socket to a thread This is called from the dispatch UST registration
1715 * thread once all sockets are set for the application.
d0b96690 1716 *
b85dc84c
DG
1717 * The sock value can be invalid, we don't really care, the thread will handle
1718 * it and make the necessary cleanup if so.
1719 *
d0b96690
DG
1720 * On success, return 0 else a negative value being the errno message of the
1721 * write().
1722 */
d88aee68 1723static int send_socket_to_thread(int fd, int sock)
d0b96690 1724{
6cd525e8 1725 ssize_t ret;
d0b96690 1726
b85dc84c
DG
1727 /*
1728 * It's possible that the FD is set as invalid with -1 concurrently just
1729 * before calling this function being a shutdown state of the thread.
1730 */
1731 if (fd < 0) {
1732 ret = -EBADF;
1733 goto error;
1734 }
d0b96690 1735
6cd525e8
MD
1736 ret = lttng_write(fd, &sock, sizeof(sock));
1737 if (ret < sizeof(sock)) {
d88aee68 1738 PERROR("write apps pipe %d", fd);
d0b96690
DG
1739 if (ret < 0) {
1740 ret = -errno;
1741 }
1742 goto error;
1743 }
1744
1745 /* All good. Don't send back the write positive ret value. */
1746 ret = 0;
1747error:
6cd525e8 1748 return (int) ret;
d0b96690
DG
1749}
1750
f45e313d
DG
1751/*
1752 * Sanitize the wait queue of the dispatch registration thread meaning removing
1753 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1754 * notify socket is never received.
1755 */
1756static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1757{
1758 int ret, nb_fd = 0, i;
1759 unsigned int fd_added = 0;
1760 struct lttng_poll_event events;
1761 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1762
1763 assert(wait_queue);
1764
1765 lttng_poll_init(&events);
1766
1767 /* Just skip everything for an empty queue. */
1768 if (!wait_queue->count) {
1769 goto end;
1770 }
1771
1772 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1773 if (ret < 0) {
1774 goto error_create;
1775 }
1776
1777 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1778 &wait_queue->head, head) {
1779 assert(wait_node->app);
1780 ret = lttng_poll_add(&events, wait_node->app->sock,
1781 LPOLLHUP | LPOLLERR);
1782 if (ret < 0) {
1783 goto error;
1784 }
1785
1786 fd_added = 1;
1787 }
1788
1789 if (!fd_added) {
1790 goto end;
1791 }
1792
1793 /*
1794 * Poll but don't block so we can quickly identify the faulty events and
1795 * clean them afterwards from the wait queue.
1796 */
1797 ret = lttng_poll_wait(&events, 0);
1798 if (ret < 0) {
1799 goto error;
1800 }
1801 nb_fd = ret;
1802
1803 for (i = 0; i < nb_fd; i++) {
1804 /* Get faulty FD. */
1805 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1806 int pollfd = LTTNG_POLL_GETFD(&events, i);
1807
fd20dac9
MD
1808 if (!revents) {
1809 /* No activity for this FD (poll implementation). */
1810 continue;
1811 }
1812
f45e313d
DG
1813 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1814 &wait_queue->head, head) {
1815 if (pollfd == wait_node->app->sock &&
1816 (revents & (LPOLLHUP | LPOLLERR))) {
1817 cds_list_del(&wait_node->head);
1818 wait_queue->count--;
1819 ust_app_destroy(wait_node->app);
1820 free(wait_node);
48b40bcf
JG
1821 /*
1822 * Silence warning of use-after-free in
1823 * cds_list_for_each_entry_safe which uses
1824 * __typeof__(*wait_node).
1825 */
1826 wait_node = NULL;
f45e313d
DG
1827 break;
1828 }
1829 }
1830 }
1831
1832 if (nb_fd > 0) {
1833 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1834 }
1835
1836end:
1837 lttng_poll_clean(&events);
1838 return;
1839
1840error:
1841 lttng_poll_clean(&events);
1842error_create:
1843 ERR("Unable to sanitize wait queue");
1844 return;
1845}
1846
099e26bd
DG
1847/*
1848 * Dispatch request from the registration threads to the application
1849 * communication thread.
1850 */
1851static void *thread_dispatch_ust_registration(void *data)
1852{
12e2b881 1853 int ret, err = -1;
8bdee6e2 1854 struct cds_wfcq_node *node;
099e26bd 1855 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1856 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1857 struct ust_reg_wait_queue wait_queue = {
1858 .count = 0,
1859 };
d0b96690 1860
6c71277b 1861 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
12e2b881 1862
9ad42ec1
MD
1863 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1864 goto error_testpoint;
1865 }
1866
12e2b881
MD
1867 health_code_update();
1868
f45e313d 1869 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1870
1871 DBG("[thread] Dispatch UST command started");
1872
26c9d55e 1873 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1874 health_code_update();
1875
099e26bd
DG
1876 /* Atomically prepare the queue futex */
1877 futex_nto1_prepare(&ust_cmd_queue.futex);
1878
1879 do {
d0b96690 1880 struct ust_app *app = NULL;
7972aab2 1881 ust_cmd = NULL;
d0b96690 1882
f45e313d
DG
1883 /*
1884 * Make sure we don't have node(s) that have hung up before receiving
1885 * the notify socket. This is to clean the list in order to avoid
1886 * memory leaks from notify socket that are never seen.
1887 */
1888 sanitize_wait_queue(&wait_queue);
1889
12e2b881 1890 health_code_update();
099e26bd 1891 /* Dequeue command for registration */
8bdee6e2 1892 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
099e26bd 1893 if (node == NULL) {
00a17c97 1894 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1895 /* Continue thread execution */
1896 break;
1897 }
1898
1899 ust_cmd = caa_container_of(node, struct ust_command, node);
1900
2f50c8a3
DG
1901 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1902 " gid:%d sock:%d name:%s (version %d.%d)",
1903 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1904 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1905 ust_cmd->sock, ust_cmd->reg_msg.name,
1906 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1907
1908 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1909 wait_node = zmalloc(sizeof(*wait_node));
1910 if (!wait_node) {
1911 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1912 ret = close(ust_cmd->sock);
1913 if (ret < 0) {
1914 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1915 }
51dec90d 1916 lttng_fd_put(LTTNG_FD_APPS, 1);
7972aab2 1917 free(ust_cmd);
d0b96690
DG
1918 goto error;
1919 }
1920 CDS_INIT_LIST_HEAD(&wait_node->head);
1921
1922 /* Create application object if socket is CMD. */
1923 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1924 ust_cmd->sock);
1925 if (!wait_node->app) {
1926 ret = close(ust_cmd->sock);
1927 if (ret < 0) {
1928 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1929 }
51dec90d 1930 lttng_fd_put(LTTNG_FD_APPS, 1);
d88aee68 1931 free(wait_node);
7972aab2 1932 free(ust_cmd);
d0b96690
DG
1933 continue;
1934 }
1935 /*
1936 * Add application to the wait queue so we can set the notify
1937 * socket before putting this object in the global ht.
1938 */
f45e313d
DG
1939 cds_list_add(&wait_node->head, &wait_queue.head);
1940 wait_queue.count++;
d0b96690 1941
7972aab2 1942 free(ust_cmd);
d0b96690
DG
1943 /*
1944 * We have to continue here since we don't have the notify
1945 * socket and the application MUST be added to the hash table
1946 * only at that moment.
1947 */
1948 continue;
1949 } else {
1950 /*
1951 * Look for the application in the local wait queue and set the
1952 * notify socket if found.
1953 */
d88aee68 1954 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1955 &wait_queue.head, head) {
12e2b881 1956 health_code_update();
d0b96690
DG
1957 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1958 wait_node->app->notify_sock = ust_cmd->sock;
1959 cds_list_del(&wait_node->head);
f45e313d 1960 wait_queue.count--;
d0b96690
DG
1961 app = wait_node->app;
1962 free(wait_node);
1963 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1964 break;
1965 }
1966 }
020d7f60
DG
1967
1968 /*
1969 * With no application at this stage the received socket is
1970 * basically useless so close it before we free the cmd data
1971 * structure for good.
1972 */
1973 if (!app) {
1974 ret = close(ust_cmd->sock);
1975 if (ret < 0) {
1976 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1977 }
51dec90d 1978 lttng_fd_put(LTTNG_FD_APPS, 1);
020d7f60 1979 }
7972aab2 1980 free(ust_cmd);
d0b96690
DG
1981 }
1982
1983 if (app) {
d0b96690
DG
1984 /*
1985 * @session_lock_list
1986 *
1987 * Lock the global session list so from the register up to the
1988 * registration done message, no thread can see the application
1989 * and change its state.
1990 */
1991 session_lock_list();
1992 rcu_read_lock();
d88aee68 1993
d0b96690
DG
1994 /*
1995 * Add application to the global hash table. This needs to be
1996 * done before the update to the UST registry can locate the
1997 * application.
1998 */
1999 ust_app_add(app);
d88aee68
DG
2000
2001 /* Set app version. This call will print an error if needed. */
2002 (void) ust_app_version(app);
2003
2004 /* Send notify socket through the notify pipe. */
2005 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2006 app->notify_sock);
2007 if (ret < 0) {
2008 rcu_read_unlock();
2009 session_unlock_list();
b85dc84c
DG
2010 /*
2011 * No notify thread, stop the UST tracing. However, this is
2012 * not an internal error of the this thread thus setting
2013 * the health error code to a normal exit.
2014 */
2015 err = 0;
d88aee68 2016 goto error;
6620da75 2017 }
d88aee68 2018
d0b96690
DG
2019 /*
2020 * Update newly registered application with the tracing
2021 * registry info already enabled information.
2022 */
2023 update_ust_app(app->sock);
d88aee68
DG
2024
2025 /*
2026 * Don't care about return value. Let the manage apps threads
2027 * handle app unregistration upon socket close.
2028 */
fb45065e 2029 (void) ust_app_register_done(app);
d88aee68
DG
2030
2031 /*
2032 * Even if the application socket has been closed, send the app
2033 * to the thread and unregistration will take place at that
2034 * place.
2035 */
2036 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 2037 if (ret < 0) {
d88aee68
DG
2038 rcu_read_unlock();
2039 session_unlock_list();
b85dc84c
DG
2040 /*
2041 * No apps. thread, stop the UST tracing. However, this is
2042 * not an internal error of the this thread thus setting
2043 * the health error code to a normal exit.
2044 */
2045 err = 0;
d88aee68 2046 goto error;
d0b96690 2047 }
d88aee68 2048
d0b96690
DG
2049 rcu_read_unlock();
2050 session_unlock_list();
099e26bd 2051 }
099e26bd
DG
2052 } while (node != NULL);
2053
12e2b881 2054 health_poll_entry();
099e26bd
DG
2055 /* Futex wait on queue. Blocking call on futex() */
2056 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 2057 health_poll_exit();
099e26bd 2058 }
12e2b881
MD
2059 /* Normal exit, no error */
2060 err = 0;
099e26bd
DG
2061
2062error:
d88aee68
DG
2063 /* Clean up wait queue. */
2064 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 2065 &wait_queue.head, head) {
d88aee68 2066 cds_list_del(&wait_node->head);
f45e313d 2067 wait_queue.count--;
d88aee68
DG
2068 free(wait_node);
2069 }
2070
772b8f4d
MD
2071 /* Empty command queue. */
2072 for (;;) {
2073 /* Dequeue command for registration */
2074 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2075 if (node == NULL) {
2076 break;
2077 }
2078 ust_cmd = caa_container_of(node, struct ust_command, node);
2079 ret = close(ust_cmd->sock);
2080 if (ret < 0) {
2081 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2082 }
2083 lttng_fd_put(LTTNG_FD_APPS, 1);
2084 free(ust_cmd);
2085 }
2086
9ad42ec1 2087error_testpoint:
099e26bd 2088 DBG("Dispatch thread dying");
12e2b881
MD
2089 if (err) {
2090 health_error();
2091 ERR("Health error occurred in %s", __func__);
2092 }
8782cc74 2093 health_unregister(health_sessiond);
099e26bd
DG
2094 return NULL;
2095}
2096
2097/*
2098 * This thread manage application registration.
2099 */
2100static void *thread_registration_apps(void *data)
1d4b027a 2101{
139ac872 2102 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
2103 uint32_t revents, nb_fd;
2104 struct lttng_poll_event events;
099e26bd
DG
2105 /*
2106 * Get allocated in this thread, enqueued to a global queue, dequeued and
2107 * freed in the manage apps thread.
2108 */
2109 struct ust_command *ust_cmd = NULL;
1d4b027a 2110
099e26bd 2111 DBG("[thread] Manage application registration started");
1d4b027a 2112
6c71277b 2113 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
927ca06a 2114
e547b070 2115 if (testpoint(sessiond_thread_registration_apps)) {
6993eeb3
CB
2116 goto error_testpoint;
2117 }
8ac94142 2118
1d4b027a
DG
2119 ret = lttcomm_listen_unix_sock(apps_sock);
2120 if (ret < 0) {
76d7553f 2121 goto error_listen;
1d4b027a
DG
2122 }
2123
5eb91c98
DG
2124 /*
2125 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2126 * more will be added to this poll set.
2127 */
d0b96690 2128 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 2129 if (ret < 0) {
76d7553f 2130 goto error_create_poll;
5eb91c98 2131 }
273ea72c 2132
5eb91c98
DG
2133 /* Add the application registration socket */
2134 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2135 if (ret < 0) {
76d7553f 2136 goto error_poll_add;
5eb91c98 2137 }
273ea72c 2138
1d4b027a 2139 /* Notify all applications to register */
0fdd1e2c
DG
2140 ret = notify_ust_apps(1);
2141 if (ret < 0) {
2142 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
2143 "Execution continues but there might be problem for already\n"
2144 "running applications that wishes to register.");
0fdd1e2c 2145 }
1d4b027a
DG
2146
2147 while (1) {
2148 DBG("Accepting application registration");
273ea72c
DG
2149
2150 /* Inifinite blocking call, waiting for transmission */
88f2b785 2151 restart:
a78af745 2152 health_poll_entry();
5eb91c98 2153 ret = lttng_poll_wait(&events, -1);
a78af745 2154 health_poll_exit();
273ea72c 2155 if (ret < 0) {
88f2b785
MD
2156 /*
2157 * Restart interrupted system call.
2158 */
2159 if (errno == EINTR) {
2160 goto restart;
2161 }
273ea72c
DG
2162 goto error;
2163 }
2164
0d9c5d77
DG
2165 nb_fd = ret;
2166
5eb91c98 2167 for (i = 0; i < nb_fd; i++) {
840cb59c 2168 health_code_update();
139ac872 2169
5eb91c98
DG
2170 /* Fetch once the poll data */
2171 revents = LTTNG_POLL_GETEV(&events, i);
2172 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 2173
fd20dac9
MD
2174 if (!revents) {
2175 /* No activity for this FD (poll implementation). */
2176 continue;
2177 }
2178
5eb91c98 2179 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 2180 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 2181 if (ret) {
139ac872
MD
2182 err = 0;
2183 goto exit;
90014c57 2184 }
1d4b027a 2185
5eb91c98
DG
2186 /* Event on the registration socket */
2187 if (pollfd == apps_sock) {
2188 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2189 ERR("Register apps socket poll error");
2190 goto error;
2191 } else if (revents & LPOLLIN) {
2192 sock = lttcomm_accept_unix_sock(apps_sock);
2193 if (sock < 0) {
2194 goto error;
2195 }
099e26bd 2196
16c5c8fa
DG
2197 /*
2198 * Set socket timeout for both receiving and ending.
2199 * app_socket_timeout is in seconds, whereas
2200 * lttcomm_setsockopt_rcv_timeout and
2201 * lttcomm_setsockopt_snd_timeout expect msec as
2202 * parameter.
2203 */
2204 (void) lttcomm_setsockopt_rcv_timeout(sock,
2205 app_socket_timeout * 1000);
2206 (void) lttcomm_setsockopt_snd_timeout(sock,
2207 app_socket_timeout * 1000);
2208
b662582b
DG
2209 /*
2210 * Set the CLOEXEC flag. Return code is useless because
2211 * either way, the show must go on.
2212 */
2213 (void) utils_set_fd_cloexec(sock);
2214
5eb91c98 2215 /* Create UST registration command for enqueuing */
ba7f0ae5 2216 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 2217 if (ust_cmd == NULL) {
76d7553f 2218 PERROR("ust command zmalloc");
41ed8e47
MD
2219 ret = close(sock);
2220 if (ret) {
2221 PERROR("close");
2222 }
5eb91c98
DG
2223 goto error;
2224 }
1d4b027a 2225
5eb91c98
DG
2226 /*
2227 * Using message-based transmissions to ensure we don't
2228 * have to deal with partially received messages.
2229 */
4063050c
MD
2230 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2231 if (ret < 0) {
2232 ERR("Exhausted file descriptors allowed for applications.");
2233 free(ust_cmd);
2234 ret = close(sock);
2235 if (ret) {
2236 PERROR("close");
2237 }
2238 sock = -1;
2239 continue;
2240 }
d88aee68 2241
840cb59c 2242 health_code_update();
d0b96690
DG
2243 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2244 if (ret < 0) {
5eb91c98 2245 free(ust_cmd);
d0b96690 2246 /* Close socket of the application. */
76d7553f
MD
2247 ret = close(sock);
2248 if (ret) {
2249 PERROR("close");
2250 }
4063050c 2251 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 2252 sock = -1;
5eb91c98
DG
2253 continue;
2254 }
840cb59c 2255 health_code_update();
099e26bd 2256
5eb91c98 2257 ust_cmd->sock = sock;
34a2494f 2258 sock = -1;
099e26bd 2259
5eb91c98
DG
2260 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2261 " gid:%d sock:%d name:%s (version %d.%d)",
2262 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2263 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2264 ust_cmd->sock, ust_cmd->reg_msg.name,
2265 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 2266
5eb91c98
DG
2267 /*
2268 * Lock free enqueue the registration request. The red pill
54d01ffb 2269 * has been taken! This apps will be part of the *system*.
5eb91c98 2270 */
8bdee6e2 2271 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
5eb91c98
DG
2272
2273 /*
2274 * Wake the registration queue futex. Implicit memory
8bdee6e2 2275 * barrier with the exchange in cds_wfcq_enqueue.
5eb91c98
DG
2276 */
2277 futex_nto1_wake(&ust_cmd_queue.futex);
2278 }
2279 }
90014c57 2280 }
1d4b027a
DG
2281 }
2282
139ac872 2283exit:
1d4b027a 2284error:
0fdd1e2c
DG
2285 /* Notify that the registration thread is gone */
2286 notify_ust_apps(0);
2287
a4b35e07 2288 if (apps_sock >= 0) {
76d7553f
MD
2289 ret = close(apps_sock);
2290 if (ret) {
2291 PERROR("close");
2292 }
a4b35e07 2293 }
46c3f085 2294 if (sock >= 0) {
76d7553f
MD
2295 ret = close(sock);
2296 if (ret) {
2297 PERROR("close");
2298 }
4063050c 2299 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 2300 }
273ea72c 2301 unlink(apps_unix_sock_path);
0fdd1e2c 2302
76d7553f 2303error_poll_add:
5eb91c98 2304 lttng_poll_clean(&events);
76d7553f
MD
2305error_listen:
2306error_create_poll:
6993eeb3 2307error_testpoint:
76d7553f 2308 DBG("UST Registration thread cleanup complete");
9ad42ec1
MD
2309 if (err) {
2310 health_error();
2311 ERR("Health error occurred in %s", __func__);
2312 }
8782cc74 2313 health_unregister(health_sessiond);
5eb91c98 2314
1d4b027a
DG
2315 return NULL;
2316}
2317
8c0faa1d 2318/*
3bd1e081 2319 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 2320 * exec or it will fails.
8c0faa1d 2321 */
3bd1e081 2322static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 2323{
a23ec3a7 2324 int ret, clock_ret;
ee0b0061
DG
2325 struct timespec timeout;
2326
a23ec3a7
DG
2327 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2328 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 2329
a23ec3a7
DG
2330 /* Setup pthread condition */
2331 ret = pthread_condattr_init(&consumer_data->condattr);
4a15001e 2332 if (ret) {
a23ec3a7
DG
2333 errno = ret;
2334 PERROR("pthread_condattr_init consumer data");
2335 goto error;
2336 }
2337
2338 /*
2339 * Set the monotonic clock in order to make sure we DO NOT jump in time
2340 * between the clock_gettime() call and the timedwait call. See bug #324
2341 * for a more details and how we noticed it.
2342 */
2343 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
4a15001e 2344 if (ret) {
a23ec3a7
DG
2345 errno = ret;
2346 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
2347 goto error;
2348 }
8c0faa1d 2349
a23ec3a7 2350 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
4a15001e 2351 if (ret) {
a23ec3a7
DG
2352 errno = ret;
2353 PERROR("pthread_cond_init consumer data");
2354 goto error;
2355 }
2356
2357 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2358 consumer_data);
4a15001e
MD
2359 if (ret) {
2360 errno = ret;
3bd1e081 2361 PERROR("pthread_create consumer");
ee0b0061 2362 ret = -1;
8c0faa1d
DG
2363 goto error;
2364 }
2365
a23ec3a7
DG
2366 /* We are about to wait on a pthread condition */
2367 pthread_mutex_lock(&consumer_data->cond_mutex);
2368
ee0b0061 2369 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
2370 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2371 /*
2372 * Set the timeout for the condition timed wait even if the clock gettime
2373 * call fails since we might loop on that call and we want to avoid to
2374 * increment the timeout too many times.
2375 */
2376 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2377
2378 /*
2379 * The following loop COULD be skipped in some conditions so this is why we
2380 * set ret to 0 in order to make sure at least one round of the loop is
2381 * done.
2382 */
2383 ret = 0;
2384
2385 /*
2386 * Loop until the condition is reached or when a timeout is reached. Note
2387 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2388 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2389 * possible. This loop does not take any chances and works with both of
2390 * them.
2391 */
2392 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2393 if (clock_ret < 0) {
2394 PERROR("clock_gettime spawn consumer");
2395 /* Infinite wait for the consumerd thread to be ready */
2396 ret = pthread_cond_wait(&consumer_data->cond,
2397 &consumer_data->cond_mutex);
2398 } else {
2399 ret = pthread_cond_timedwait(&consumer_data->cond,
2400 &consumer_data->cond_mutex, &timeout);
2401 }
ee0b0061 2402 }
8c0faa1d 2403
a23ec3a7
DG
2404 /* Release the pthread condition */
2405 pthread_mutex_unlock(&consumer_data->cond_mutex);
2406
2407 if (ret != 0) {
2408 errno = ret;
2409 if (ret == ETIMEDOUT) {
4282f9a3
DG
2410 int pth_ret;
2411
ee0b0061
DG
2412 /*
2413 * Call has timed out so we kill the kconsumerd_thread and return
2414 * an error.
2415 */
a23ec3a7
DG
2416 ERR("Condition timed out. The consumer thread was never ready."
2417 " Killing it");
4282f9a3
DG
2418 pth_ret = pthread_cancel(consumer_data->thread);
2419 if (pth_ret < 0) {
3bd1e081 2420 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
2421 }
2422 } else {
a23ec3a7 2423 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061 2424 }
4282f9a3
DG
2425 /* Caller is expecting a negative value on failure. */
2426 ret = -1;
ee0b0061
DG
2427 goto error;
2428 }
2429
3bd1e081
MD
2430 pthread_mutex_lock(&consumer_data->pid_mutex);
2431 if (consumer_data->pid == 0) {
a23ec3a7 2432 ERR("Consumerd did not start");
3bd1e081 2433 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2434 goto error;
2435 }
3bd1e081 2436 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2437
8c0faa1d
DG
2438 return 0;
2439
2440error:
2441 return ret;
2442}
2443
d9800920 2444/*
3bd1e081 2445 * Join consumer thread
d9800920 2446 */
3bd1e081 2447static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2448{
2449 void *status;
cf3af59e 2450
e8209f6b
DG
2451 /* Consumer pid must be a real one. */
2452 if (consumer_data->pid > 0) {
c617c0c6 2453 int ret;
3bd1e081 2454 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2455 if (ret) {
4a15001e 2456 PERROR("Error killing consumer daemon");
cf3af59e
MD
2457 return ret;
2458 }
3bd1e081 2459 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2460 } else {
2461 return 0;
2462 }
2463}
2464
8c0faa1d 2465/*
3bd1e081 2466 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2467 *
d063d709 2468 * Return pid if successful else -1.
8c0faa1d 2469 */
3bd1e081 2470static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2471{
2472 int ret;
2473 pid_t pid;
94c55f17 2474 const char *consumer_to_use;
53086306 2475 const char *verbosity;
94c55f17 2476 struct stat st;
8c0faa1d 2477
3bd1e081 2478 DBG("Spawning consumerd");
c49dc785 2479
8c0faa1d
DG
2480 pid = fork();
2481 if (pid == 0) {
2482 /*
3bd1e081 2483 * Exec consumerd.
8c0faa1d 2484 */
daee5345 2485 if (opt_verbose_consumer) {
53086306 2486 verbosity = "--verbose";
4421f712 2487 } else if (lttng_opt_quiet) {
53086306 2488 verbosity = "--quiet";
4421f712
DG
2489 } else {
2490 verbosity = "";
53086306 2491 }
4421f712 2492
3bd1e081
MD
2493 switch (consumer_data->type) {
2494 case LTTNG_CONSUMER_KERNEL:
94c55f17 2495 /*
c7704d57
DG
2496 * Find out which consumerd to execute. We will first try the
2497 * 64-bit path, then the sessiond's installation directory, and
2498 * fallback on the 32-bit one,
94c55f17 2499 */
63a799e8
AM
2500 DBG3("Looking for a kernel consumer at these locations:");
2501 DBG3(" 1) %s", consumerd64_bin);
2502 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2503 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2504 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2505 DBG3("Found location #1");
94c55f17 2506 consumer_to_use = consumerd64_bin;
94c55f17 2507 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2508 DBG3("Found location #2");
94c55f17 2509 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2510 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2511 DBG3("Found location #3");
eb1e0bd4 2512 consumer_to_use = consumerd32_bin;
94c55f17 2513 } else {
63a799e8 2514 DBG("Could not find any valid consumerd executable");
4282f9a3 2515 ret = -EINVAL;
94c55f17
AM
2516 break;
2517 }
2518 DBG("Using kernel consumer at: %s", consumer_to_use);
4282f9a3 2519 ret = execl(consumer_to_use,
94c55f17
AM
2520 "lttng-consumerd", verbosity, "-k",
2521 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2522 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2523 "--group", tracing_group_name,
94c55f17 2524 NULL);
3bd1e081 2525 break;
7753dea8
MD
2526 case LTTNG_CONSUMER64_UST:
2527 {
b1e0b6b6 2528 char *tmpnew = NULL;
8f4905da
MD
2529
2530 if (consumerd64_libdir[0] != '\0') {
2531 char *tmp;
2532 size_t tmplen;
2533
e8fa9fb0 2534 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
8f4905da
MD
2535 if (!tmp) {
2536 tmp = "";
2537 }
2538 tmplen = strlen("LD_LIBRARY_PATH=")
2539 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2540 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2541 if (!tmpnew) {
2542 ret = -ENOMEM;
2543 goto error;
2544 }
2545 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2546 strcat(tmpnew, consumerd64_libdir);
2547 if (tmp[0] != '\0') {
2548 strcat(tmpnew, ":");
2549 strcat(tmpnew, tmp);
2550 }
2551 ret = putenv(tmpnew);
2552 if (ret) {
2553 ret = -errno;
c6f76da9 2554 free(tmpnew);
8f4905da
MD
2555 goto error;
2556 }
2557 }
94c55f17 2558 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2559 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2560 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2561 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2562 "--group", tracing_group_name,
7753dea8 2563 NULL);
8f4905da
MD
2564 if (consumerd64_libdir[0] != '\0') {
2565 free(tmpnew);
2566 }
3bd1e081 2567 break;
7753dea8
MD
2568 }
2569 case LTTNG_CONSUMER32_UST:
2570 {
937dde8e 2571 char *tmpnew = NULL;
8f4905da
MD
2572
2573 if (consumerd32_libdir[0] != '\0') {
2574 char *tmp;
2575 size_t tmplen;
2576
e8fa9fb0 2577 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
8f4905da
MD
2578 if (!tmp) {
2579 tmp = "";
2580 }
2581 tmplen = strlen("LD_LIBRARY_PATH=")
2582 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2583 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2584 if (!tmpnew) {
2585 ret = -ENOMEM;
2586 goto error;
2587 }
2588 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2589 strcat(tmpnew, consumerd32_libdir);
2590 if (tmp[0] != '\0') {
2591 strcat(tmpnew, ":");
2592 strcat(tmpnew, tmp);
2593 }
2594 ret = putenv(tmpnew);
2595 if (ret) {
2596 ret = -errno;
c6f76da9 2597 free(tmpnew);
8f4905da
MD
2598 goto error;
2599 }
2600 }
94c55f17 2601 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2602 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2603 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2604 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2605 "--group", tracing_group_name,
7753dea8 2606 NULL);
8f4905da
MD
2607 if (consumerd32_libdir[0] != '\0') {
2608 free(tmpnew);
2609 }
7753dea8
MD
2610 break;
2611 }
3bd1e081 2612 default:
76d7553f 2613 PERROR("unknown consumer type");
3bd1e081
MD
2614 exit(EXIT_FAILURE);
2615 }
8c0faa1d 2616 if (errno != 0) {
4282f9a3 2617 PERROR("Consumer execl()");
8c0faa1d 2618 }
4282f9a3 2619 /* Reaching this point, we got a failure on our execl(). */
8c0faa1d
DG
2620 exit(EXIT_FAILURE);
2621 } else if (pid > 0) {
2622 ret = pid;
8c0faa1d 2623 } else {
76d7553f 2624 PERROR("start consumer fork");
8c0faa1d 2625 ret = -errno;
8c0faa1d 2626 }
8f4905da 2627error:
8c0faa1d
DG
2628 return ret;
2629}
2630
693bd40b 2631/*
3bd1e081 2632 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2633 */
3bd1e081 2634static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2635{
c617c0c6 2636 int ret;
edb8b045
DG
2637
2638 /*
2639 * Set the listen() state on the socket since there is a possible race
2640 * between the exec() of the consumer daemon and this call if place in the
2641 * consumer thread. See bug #366 for more details.
2642 */
2643 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2644 if (ret < 0) {
2645 goto error;
2646 }
693bd40b 2647
3bd1e081
MD
2648 pthread_mutex_lock(&consumer_data->pid_mutex);
2649 if (consumer_data->pid != 0) {
2650 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2651 goto end;
2652 }
693bd40b 2653
3bd1e081 2654 ret = spawn_consumerd(consumer_data);
c49dc785 2655 if (ret < 0) {
3bd1e081
MD
2656 ERR("Spawning consumerd failed");
2657 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2658 goto error;
693bd40b 2659 }
c49dc785 2660
3bd1e081
MD
2661 /* Setting up the consumer_data pid */
2662 consumer_data->pid = ret;
48842b30 2663 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2664 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2665
3bd1e081
MD
2666 DBG2("Spawning consumer control thread");
2667 ret = spawn_consumer_thread(consumer_data);
693bd40b 2668 if (ret < 0) {
3bd1e081 2669 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2670 goto error;
2671 }
2672
c49dc785 2673end:
693bd40b
DG
2674 return 0;
2675
2676error:
331744e3 2677 /* Cleanup already created sockets on error. */
edb8b045 2678 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2679 int err;
2680
edb8b045
DG
2681 err = close(consumer_data->err_sock);
2682 if (err < 0) {
2683 PERROR("close consumer data error socket");
2684 }
2685 }
693bd40b
DG
2686 return ret;
2687}
2688
b73401da 2689/*
096102bd 2690 * Setup necessary data for kernel tracer action.
b73401da 2691 */
096102bd 2692static int init_kernel_tracer(void)
b73401da
DG
2693{
2694 int ret;
b73401da 2695
096102bd
DG
2696 /* Modprobe lttng kernel modules */
2697 ret = modprobe_lttng_control();
b73401da 2698 if (ret < 0) {
b73401da
DG
2699 goto error;
2700 }
2701
096102bd
DG
2702 /* Open debugfs lttng */
2703 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2704 if (kernel_tracer_fd < 0) {
2705 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2706 ret = -1;
2707 goto error_open;
54d01ffb
DG
2708 }
2709
2f77fc4b
DG
2710 /* Validate kernel version */
2711 ret = kernel_validate_version(kernel_tracer_fd);
2712 if (ret < 0) {
2713 goto error_version;
b551a063 2714 }
54d01ffb 2715
2f77fc4b
DG
2716 ret = modprobe_lttng_data();
2717 if (ret < 0) {
2718 goto error_modules;
54d01ffb
DG
2719 }
2720
2f77fc4b
DG
2721 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2722 return 0;
2723
2724error_version:
2725 modprobe_remove_lttng_control();
2726 ret = close(kernel_tracer_fd);
2727 if (ret) {
2728 PERROR("close");
b551a063 2729 }
2f77fc4b 2730 kernel_tracer_fd = -1;
f73fabfd 2731 return LTTNG_ERR_KERN_VERSION;
b551a063 2732
2f77fc4b
DG
2733error_modules:
2734 ret = close(kernel_tracer_fd);
2735 if (ret) {
2736 PERROR("close");
b551a063 2737 }
54d01ffb 2738
2f77fc4b
DG
2739error_open:
2740 modprobe_remove_lttng_control();
54d01ffb
DG
2741
2742error:
2f77fc4b
DG
2743 WARN("No kernel tracer available");
2744 kernel_tracer_fd = -1;
2745 if (!is_root) {
f73fabfd 2746 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2747 } else {
f73fabfd 2748 return LTTNG_ERR_KERN_NA;
2f77fc4b 2749 }
54d01ffb
DG
2750}
2751
2f77fc4b 2752
54d01ffb 2753/*
2f77fc4b
DG
2754 * Copy consumer output from the tracing session to the domain session. The
2755 * function also applies the right modification on a per domain basis for the
2756 * trace files destination directory.
36b588ed
MD
2757 *
2758 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2759 */
2f77fc4b 2760static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2761{
2762 int ret;
2f77fc4b
DG
2763 const char *dir_name;
2764 struct consumer_output *consumer;
2765
2766 assert(session);
2767 assert(session->consumer);
54d01ffb 2768
b551a063
DG
2769 switch (domain) {
2770 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2771 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2772 /*
2773 * XXX: We should audit the session creation and what this function
2774 * does "extra" in order to avoid a destroy since this function is used
2775 * in the domain session creation (kernel and ust) only. Same for UST
2776 * domain.
2777 */
2778 if (session->kernel_session->consumer) {
6addfa37 2779 consumer_output_put(session->kernel_session->consumer);
09a90bcd 2780 }
2f77fc4b
DG
2781 session->kernel_session->consumer =
2782 consumer_copy_output(session->consumer);
2783 /* Ease our life a bit for the next part */
2784 consumer = session->kernel_session->consumer;
2785 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063 2786 break;
f20baf8e 2787 case LTTNG_DOMAIN_JUL:
5cdb6027 2788 case LTTNG_DOMAIN_LOG4J:
0e115563 2789 case LTTNG_DOMAIN_PYTHON:
b551a063 2790 case LTTNG_DOMAIN_UST:
2f77fc4b 2791 DBG3("Copying tracing session consumer output in UST session");
09a90bcd 2792 if (session->ust_session->consumer) {
6addfa37 2793 consumer_output_put(session->ust_session->consumer);
09a90bcd 2794 }
2f77fc4b
DG
2795 session->ust_session->consumer =
2796 consumer_copy_output(session->consumer);
2797 /* Ease our life a bit for the next part */
2798 consumer = session->ust_session->consumer;
2799 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2800 break;
2801 default:
f73fabfd 2802 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2803 goto error;
2804 }
2805
2f77fc4b 2806 /* Append correct directory to subdir */
c30ce0b3
CB
2807 strncat(consumer->subdir, dir_name,
2808 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2809 DBG3("Copy session consumer subdir %s", consumer->subdir);
2810
f73fabfd 2811 ret = LTTNG_OK;
54d01ffb
DG
2812
2813error:
2814 return ret;
2815}
2816
00e2e675 2817/*
2f77fc4b 2818 * Create an UST session and add it to the session ust list.
36b588ed
MD
2819 *
2820 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2821 */
2f77fc4b
DG
2822static int create_ust_session(struct ltt_session *session,
2823 struct lttng_domain *domain)
00e2e675
DG
2824{
2825 int ret;
2f77fc4b 2826 struct ltt_ust_session *lus = NULL;
00e2e675 2827
a4b92340 2828 assert(session);
2f77fc4b
DG
2829 assert(domain);
2830 assert(session->consumer);
a4b92340 2831
2f77fc4b 2832 switch (domain->type) {
f20baf8e 2833 case LTTNG_DOMAIN_JUL:
5cdb6027 2834 case LTTNG_DOMAIN_LOG4J:
0e115563 2835 case LTTNG_DOMAIN_PYTHON:
2f77fc4b
DG
2836 case LTTNG_DOMAIN_UST:
2837 break;
2838 default:
2839 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2840 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2841 goto error;
2842 }
2843
2f77fc4b
DG
2844 DBG("Creating UST session");
2845
dec56f6c 2846 lus = trace_ust_create_session(session->id);
2f77fc4b 2847 if (lus == NULL) {
f73fabfd 2848 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2849 goto error;
2850 }
2851
2f77fc4b
DG
2852 lus->uid = session->uid;
2853 lus->gid = session->gid;
2bba9e53 2854 lus->output_traces = session->output_traces;
27babd3a 2855 lus->snapshot_mode = session->snapshot_mode;
ecc48a90 2856 lus->live_timer_interval = session->live_timer;
2f77fc4b 2857 session->ust_session = lus;
d7ba1388 2858 if (session->shm_path[0]) {
3d071855
MD
2859 strncpy(lus->root_shm_path, session->shm_path,
2860 sizeof(lus->root_shm_path));
2861 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
d7ba1388
MD
2862 strncpy(lus->shm_path, session->shm_path,
2863 sizeof(lus->shm_path));
2864 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2865 strncat(lus->shm_path, "/ust",
2866 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2867 }
2f77fc4b
DG
2868 /* Copy session output to the newly created UST session */
2869 ret = copy_session_consumer(domain->type, session);
f73fabfd 2870 if (ret != LTTNG_OK) {
00e2e675
DG
2871 goto error;
2872 }
2873
f73fabfd 2874 return LTTNG_OK;
00e2e675
DG
2875
2876error:
2f77fc4b
DG
2877 free(lus);
2878 session->ust_session = NULL;
00e2e675
DG
2879 return ret;
2880}
2881
2882/*
2f77fc4b 2883 * Create a kernel tracer session then create the default channel.
00e2e675 2884 */
2f77fc4b 2885static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2886{
2887 int ret;
a4b92340 2888
2f77fc4b 2889 DBG("Creating kernel session");
00e2e675 2890
2f77fc4b
DG
2891 ret = kernel_create_session(session, kernel_tracer_fd);
2892 if (ret < 0) {
f73fabfd 2893 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2894 goto error;
2895 }
2896
2f77fc4b
DG
2897 /* Code flow safety */
2898 assert(session->kernel_session);
2899
2900 /* Copy session output to the newly created Kernel session */
2901 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2902 if (ret != LTTNG_OK) {
a4b92340
DG
2903 goto error;
2904 }
2905
2f77fc4b
DG
2906 /* Create directory(ies) on local filesystem. */
2907 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2908 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2909 ret = run_as_mkdir_recursive(
2910 session->kernel_session->consumer->dst.trace_path,
2911 S_IRWXU | S_IRWXG, session->uid, session->gid);
2912 if (ret < 0) {
df5b86c8 2913 if (errno != EEXIST) {
2f77fc4b 2914 ERR("Trace directory creation error");
00e2e675
DG
2915 goto error;
2916 }
00e2e675 2917 }
2f77fc4b 2918 }
00e2e675 2919
2f77fc4b
DG
2920 session->kernel_session->uid = session->uid;
2921 session->kernel_session->gid = session->gid;
2bba9e53 2922 session->kernel_session->output_traces = session->output_traces;
27babd3a 2923 session->kernel_session->snapshot_mode = session->snapshot_mode;
00e2e675 2924
f73fabfd 2925 return LTTNG_OK;
00e2e675 2926
2f77fc4b
DG
2927error:
2928 trace_kernel_destroy_session(session->kernel_session);
2929 session->kernel_session = NULL;
2930 return ret;
2931}
00e2e675 2932
2f77fc4b
DG
2933/*
2934 * Count number of session permitted by uid/gid.
2935 */
2936static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2937{
2938 unsigned int i = 0;
2939 struct ltt_session *session;
07424f16 2940
2f77fc4b
DG
2941 DBG("Counting number of available session for UID %d GID %d",
2942 uid, gid);
2943 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2944 /*
2f77fc4b 2945 * Only list the sessions the user can control.
00e2e675 2946 */
2f77fc4b
DG
2947 if (!session_access_ok(session, uid, gid)) {
2948 continue;
2949 }
2950 i++;
a4b92340 2951 }
2f77fc4b 2952 return i;
00e2e675
DG
2953}
2954
54d01ffb
DG
2955/*
2956 * Process the command requested by the lttng client within the command
2957 * context structure. This function make sure that the return structure (llm)
2958 * is set and ready for transmission before returning.
2959 *
2960 * Return any error encountered or 0 for success.
53a80697
MD
2961 *
2962 * "sock" is only used for special-case var. len data.
36b588ed
MD
2963 *
2964 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2965 */
53a80697
MD
2966static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2967 int *sock_error)
54d01ffb 2968{
f73fabfd 2969 int ret = LTTNG_OK;
44d3bd01 2970 int need_tracing_session = 1;
2e09ba09 2971 int need_domain;
54d01ffb
DG
2972
2973 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2974
3745d315
MD
2975 assert(!rcu_read_ongoing());
2976
53a80697
MD
2977 *sock_error = 0;
2978
2e09ba09
MD
2979 switch (cmd_ctx->lsm->cmd_type) {
2980 case LTTNG_CREATE_SESSION:
27babd3a 2981 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2982 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09
MD
2983 case LTTNG_DESTROY_SESSION:
2984 case LTTNG_LIST_SESSIONS:
2985 case LTTNG_LIST_DOMAINS:
2986 case LTTNG_START_TRACE:
2987 case LTTNG_STOP_TRACE:
6d805429 2988 case LTTNG_DATA_PENDING:
da3c9ec1
DG
2989 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2990 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2991 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2992 case LTTNG_SNAPSHOT_RECORD:
fb198a11 2993 case LTTNG_SAVE_SESSION:
d7ba1388 2994 case LTTNG_SET_SESSION_SHM_PATH:
2e09ba09 2995 need_domain = 0;
3aace903 2996 break;
2e09ba09
MD
2997 default:
2998 need_domain = 1;
2999 }
3000
3001 if (opt_no_kernel && need_domain
3002 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 3003 if (!is_root) {
f73fabfd 3004 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 3005 } else {
f73fabfd 3006 ret = LTTNG_ERR_KERN_NA;
531d29f9 3007 }
4fba7219
DG
3008 goto error;
3009 }
3010
8d3113b2
DG
3011 /* Deny register consumer if we already have a spawned consumer. */
3012 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3013 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3014 if (kconsumer_data.pid > 0) {
f73fabfd 3015 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 3016 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
3017 goto error;
3018 }
3019 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3020 }
3021
54d01ffb
DG
3022 /*
3023 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 3024 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
3025 * command.
3026 */
3027 switch(cmd_ctx->lsm->cmd_type) {
3028 case LTTNG_LIST_SESSIONS:
3029 case LTTNG_LIST_TRACEPOINTS:
f37d259d 3030 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
3031 case LTTNG_LIST_DOMAINS:
3032 case LTTNG_LIST_CHANNELS:
3033 case LTTNG_LIST_EVENTS:
834978fd 3034 case LTTNG_LIST_SYSCALLS:
a5dfbb9d 3035 case LTTNG_LIST_TRACKER_PIDS:
54d01ffb
DG
3036 break;
3037 default:
3038 /* Setup lttng message with no payload */
3039 ret = setup_lttng_msg(cmd_ctx, 0);
3040 if (ret < 0) {
3041 /* This label does not try to unlock the session */
3042 goto init_setup_error;
3043 }
3044 }
3045
3046 /* Commands that DO NOT need a session. */
3047 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 3048 case LTTNG_CREATE_SESSION:
27babd3a 3049 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 3050 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09 3051 case LTTNG_CALIBRATE:
54d01ffb
DG
3052 case LTTNG_LIST_SESSIONS:
3053 case LTTNG_LIST_TRACEPOINTS:
834978fd 3054 case LTTNG_LIST_SYSCALLS:
f37d259d 3055 case LTTNG_LIST_TRACEPOINT_FIELDS:
fb198a11 3056 case LTTNG_SAVE_SESSION:
44d3bd01 3057 need_tracing_session = 0;
54d01ffb
DG
3058 break;
3059 default:
3060 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
3061 /*
3062 * We keep the session list lock across _all_ commands
3063 * for now, because the per-session lock does not
3064 * handle teardown properly.
3065 */
74babd95 3066 session_lock_list();
54d01ffb
DG
3067 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3068 if (cmd_ctx->session == NULL) {
bba2d65f 3069 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
3070 goto error;
3071 } else {
3072 /* Acquire lock for the session */
3073 session_lock(cmd_ctx->session);
3074 }
3075 break;
3076 }
b389abbe 3077
5f3ecf22
DG
3078 /*
3079 * Commands that need a valid session but should NOT create one if none
3080 * exists. Instead of creating one and destroying it when the command is
3081 * handled, process that right before so we save some round trip in useless
3082 * code path.
3083 */
3084 switch (cmd_ctx->lsm->cmd_type) {
3085 case LTTNG_DISABLE_CHANNEL:
3086 case LTTNG_DISABLE_EVENT:
5f3ecf22
DG
3087 switch (cmd_ctx->lsm->domain.type) {
3088 case LTTNG_DOMAIN_KERNEL:
3089 if (!cmd_ctx->session->kernel_session) {
3090 ret = LTTNG_ERR_NO_CHANNEL;
3091 goto error;
3092 }
3093 break;
3094 case LTTNG_DOMAIN_JUL:
5cdb6027 3095 case LTTNG_DOMAIN_LOG4J:
0e115563 3096 case LTTNG_DOMAIN_PYTHON:
5f3ecf22
DG
3097 case LTTNG_DOMAIN_UST:
3098 if (!cmd_ctx->session->ust_session) {
3099 ret = LTTNG_ERR_NO_CHANNEL;
3100 goto error;
3101 }
3102 break;
3103 default:
3104 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3105 goto error;
3106 }
3107 default:
3108 break;
3109 }
3110
2e09ba09
MD
3111 if (!need_domain) {
3112 goto skip_domain;
3113 }
a4b92340 3114
54d01ffb
DG
3115 /*
3116 * Check domain type for specific "pre-action".
3117 */
3118 switch (cmd_ctx->lsm->domain.type) {
3119 case LTTNG_DOMAIN_KERNEL:
d1f1c568 3120 if (!is_root) {
f73fabfd 3121 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
3122