Fix: app cmd leak on sessiond exit
[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
fac6795d 589/*
4a15001e 590 * Cleanup the session daemon's data structures.
fac6795d 591 */
4a15001e 592static void sessiond_cleanup(void)
fac6795d 593{
ef599319 594 int ret;
af9737e9 595 struct ltt_session *sess, *stmp;
8c6c56c2 596 char path[PATH_MAX];
fac6795d 597
4a15001e 598 DBG("Cleanup sessiond");
e07ae692 599
4e449f3f
MD
600 /*
601 * Close the thread quit pipe. It has already done its job,
602 * since we are now called.
603 */
2f77fc4b
DG
604 utils_close_pipe(thread_quit_pipe);
605
35f90c40
DG
606 /*
607 * If opt_pidfile is undefined, the default file will be wiped when
608 * removing the rundir.
609 */
610 if (opt_pidfile) {
611 ret = remove(opt_pidfile);
612 if (ret < 0) {
613 PERROR("remove pidfile %s", opt_pidfile);
614 }
615 }
616
8c6c56c2
MD
617 DBG("Removing sessiond and consumerd content of directory %s", rundir);
618
619 /* sessiond */
620 snprintf(path, PATH_MAX,
621 "%s/%s",
622 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
623 DBG("Removing %s", path);
624 (void) unlink(path);
625
cd9290dd 626 snprintf(path, PATH_MAX, "%s/%s", rundir,
022d91ba 627 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
cd9290dd
DG
628 DBG("Removing %s", path);
629 (void) unlink(path);
630
8c6c56c2
MD
631 /* kconsumerd */
632 snprintf(path, PATH_MAX,
633 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
634 rundir);
635 DBG("Removing %s", path);
636 (void) unlink(path);
637
638 snprintf(path, PATH_MAX,
639 DEFAULT_KCONSUMERD_PATH,
640 rundir);
641 DBG("Removing directory %s", path);
642 (void) rmdir(path);
643
644 /* ust consumerd 32 */
645 snprintf(path, PATH_MAX,
646 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
647 rundir);
648 DBG("Removing %s", path);
649 (void) unlink(path);
650
651 snprintf(path, PATH_MAX,
652 DEFAULT_USTCONSUMERD32_PATH,
653 rundir);
654 DBG("Removing directory %s", path);
655 (void) rmdir(path);
656
657 /* ust consumerd 64 */
658 snprintf(path, PATH_MAX,
659 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
660 rundir);
661 DBG("Removing %s", path);
662 (void) unlink(path);
663
664 snprintf(path, PATH_MAX,
665 DEFAULT_USTCONSUMERD64_PATH,
666 rundir);
667 DBG("Removing directory %s", path);
668 (void) rmdir(path);
5461b305 669
99bab54f 670 DBG("Cleaning up all sessions");
fac6795d 671
b5541356 672 /* Destroy session list mutex */
273ea72c
DG
673 if (session_list_ptr != NULL) {
674 pthread_mutex_destroy(&session_list_ptr->lock);
675
676 /* Cleanup ALL session */
54d01ffb
DG
677 cds_list_for_each_entry_safe(sess, stmp,
678 &session_list_ptr->head, list) {
2f77fc4b 679 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
680 }
681 }
682
6a4e4039
JG
683 DBG("Cleaning up all agent apps");
684 agent_app_ht_clean();
685
099e26bd 686 DBG("Closing all UST sockets");
56fff090 687 ust_app_clean_list();
7972aab2 688 buffer_reg_destroy_registries();
099e26bd 689
4fba7219
DG
690 if (is_root && !opt_no_kernel) {
691 DBG2("Closing kernel fd");
a4b35e07 692 if (kernel_tracer_fd >= 0) {
76d7553f
MD
693 ret = close(kernel_tracer_fd);
694 if (ret) {
695 PERROR("close");
696 }
a4b35e07 697 }
2f50c8a3 698 DBG("Unloading kernel modules");
096102bd 699 modprobe_remove_lttng_all();
834978fd 700 free(syscall_table);
2f50c8a3 701 }
2f77fc4b 702
e975f9f8
DG
703 close_consumer_sockets();
704
ef367a93
JG
705 if (load_info) {
706 load_session_destroy_data(load_info);
707 free(load_info);
708 }
709
c9cb3e7d
JG
710 /*
711 * Cleanup lock file by deleting it and finaly closing it which will
712 * release the file system lock.
713 */
714 if (lockfile_fd >= 0) {
715 char lockfile_path[PATH_MAX];
716
4a15001e
MD
717 ret = generate_lock_file_path(lockfile_path,
718 sizeof(lockfile_path));
c9cb3e7d
JG
719 if (ret > 0) {
720 ret = remove(lockfile_path);
721 if (ret < 0) {
722 PERROR("remove lock file");
723 }
724 ret = close(lockfile_fd);
725 if (ret < 0) {
726 PERROR("close lock file");
727 }
728 }
729 }
730
731 /*
732 * We do NOT rmdir rundir because there are other processes
733 * using it, for instance lttng-relayd, which can start in
734 * parallel with this teardown.
735 */
736
737 free(rundir);
4a15001e
MD
738}
739
740/*
741 * Cleanup the daemon's option data structures.
742 */
743static void sessiond_cleanup_options(void)
744{
745 DBG("Cleaning up options");
746
747 /*
748 * If the override option is set, the pointer points to a *non* const
749 * thus freeing it even though the variable type is set to const.
750 */
751 if (tracing_group_name_override) {
752 free((void *) tracing_group_name);
753 }
754 if (consumerd32_bin_override) {
755 free((void *) consumerd32_bin);
756 }
757 if (consumerd64_bin_override) {
758 free((void *) consumerd64_bin);
759 }
760 if (consumerd32_libdir_override) {
761 free((void *) consumerd32_libdir);
762 }
763 if (consumerd64_libdir_override) {
764 free((void *) consumerd64_libdir);
765 }
766
767 free(opt_pidfile);
768 free(opt_load_session_path);
769 free(kmod_probes_list);
770 free(kmod_extra_probes_list);
c9cb3e7d 771
421cb601 772 /* <fun> */
f56a39af 773 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
774 "Matthew, BEET driven development works!%c[%dm",
775 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
776 /* </fun> */
fac6795d
DG
777}
778
e065084a 779/*
d063d709 780 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 781 *
d063d709 782 * Return lttcomm error code.
e065084a
DG
783 */
784static int send_unix_sock(int sock, void *buf, size_t len)
785{
786 /* Check valid length */
c617c0c6 787 if (len == 0) {
e065084a
DG
788 return -1;
789 }
790
791 return lttcomm_send_unix_sock(sock, buf, len);
792}
793
5461b305 794/*
d063d709 795 * Free memory of a command context structure.
5461b305 796 */
a2fb29a5 797static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 798{
a2fb29a5
DG
799 DBG("Clean command context structure");
800 if (*cmd_ctx) {
801 if ((*cmd_ctx)->llm) {
802 free((*cmd_ctx)->llm);
5461b305 803 }
a2fb29a5
DG
804 if ((*cmd_ctx)->lsm) {
805 free((*cmd_ctx)->lsm);
5461b305 806 }
a2fb29a5
DG
807 free(*cmd_ctx);
808 *cmd_ctx = NULL;
5461b305
DG
809 }
810}
811
fac6795d 812/*
0fdd1e2c 813 * Notify UST applications using the shm mmap futex.
fac6795d 814 */
0fdd1e2c 815static int notify_ust_apps(int active)
fac6795d 816{
0fdd1e2c 817 char *wait_shm_mmap;
fac6795d 818
0fdd1e2c 819 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 820
0fdd1e2c
DG
821 /* See shm.c for this call implying mmap, shm and futex calls */
822 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
823 if (wait_shm_mmap == NULL) {
fac6795d
DG
824 goto error;
825 }
826
0fdd1e2c
DG
827 /* Wake waiting process */
828 futex_wait_update((int32_t *) wait_shm_mmap, active);
829
830 /* Apps notified successfully */
831 return 0;
fac6795d
DG
832
833error:
0fdd1e2c 834 return -1;
fac6795d
DG
835}
836
e065084a 837/*
d063d709
DG
838 * Setup the outgoing data buffer for the response (llm) by allocating the
839 * right amount of memory and copying the original information from the lsm
840 * structure.
ca95a216 841 *
d063d709 842 * Return total size of the buffer pointed by buf.
ca95a216 843 */
5461b305 844static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 845{
f3ed775e 846 int ret, buf_size;
ca95a216 847
f3ed775e 848 buf_size = size;
5461b305 849
ba7f0ae5 850 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 851 if (cmd_ctx->llm == NULL) {
76d7553f 852 PERROR("zmalloc");
5461b305 853 ret = -ENOMEM;
ca95a216
DG
854 goto error;
855 }
856
5461b305
DG
857 /* Copy common data */
858 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 859 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 860
5461b305
DG
861 cmd_ctx->llm->data_size = size;
862 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
863
ca95a216
DG
864 return buf_size;
865
866error:
867 return ret;
868}
869
7a485870 870/*
5eb91c98 871 * Update the kernel poll set of all channel fd available over all tracing
d063d709 872 * session. Add the wakeup pipe at the end of the set.
7a485870 873 */
5eb91c98 874static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 875{
5eb91c98 876 int ret;
7a485870
DG
877 struct ltt_session *session;
878 struct ltt_kernel_channel *channel;
879
5eb91c98 880 DBG("Updating kernel poll set");
7a485870 881
54d01ffb 882 session_lock_list();
b5541356 883 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 884 session_lock(session);
7a485870 885 if (session->kernel_session == NULL) {
54d01ffb 886 session_unlock(session);
7a485870
DG
887 continue;
888 }
7a485870 889
54d01ffb
DG
890 cds_list_for_each_entry(channel,
891 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
892 /* Add channel fd to the kernel poll set */
893 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
894 if (ret < 0) {
54d01ffb 895 session_unlock(session);
5eb91c98
DG
896 goto error;
897 }
898 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 899 }
54d01ffb 900 session_unlock(session);
7a485870 901 }
54d01ffb 902 session_unlock_list();
7a485870 903
5eb91c98 904 return 0;
7a485870
DG
905
906error:
54d01ffb 907 session_unlock_list();
7a485870
DG
908 return -1;
909}
910
911/*
54d01ffb 912 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 913 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 914 *
d063d709 915 * Useful for CPU hotplug feature.
7a485870 916 */
2bdd86d4 917static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
918{
919 int ret = 0;
920 struct ltt_session *session;
173af62f 921 struct ltt_kernel_session *ksess;
7a485870
DG
922 struct ltt_kernel_channel *channel;
923
924 DBG("Updating kernel streams for channel fd %d", fd);
925
54d01ffb 926 session_lock_list();
b5541356 927 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 928 session_lock(session);
7a485870 929 if (session->kernel_session == NULL) {
54d01ffb 930 session_unlock(session);
7a485870
DG
931 continue;
932 }
173af62f 933 ksess = session->kernel_session;
d9800920 934
4a15001e
MD
935 cds_list_for_each_entry(channel,
936 &ksess->channel_list.head, list) {
937 struct lttng_ht_iter iter;
938 struct consumer_socket *socket;
d9800920 939
4a15001e
MD
940 if (channel->fd != fd) {
941 continue;
942 }
943 DBG("Channel found, updating kernel streams");
944 ret = kernel_open_channel_stream(channel);
945 if (ret < 0) {
946 goto error;
947 }
948 /* Update the stream global counter */
949 ksess->stream_count_global += ret;
950
951 /*
952 * Have we already sent fds to the consumer? If yes, it
953 * means that tracing is started so it is safe to send
954 * our updated stream fds.
955 */
956 if (ksess->consumer_fds_sent != 1
957 || ksess->consumer == NULL) {
958 ret = -1;
959 goto error;
960 }
961
962 rcu_read_lock();
963 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
964 &iter.iter, socket, node.node) {
965 pthread_mutex_lock(socket->lock);
966 ret = kernel_consumer_send_channel_stream(socket,
967 channel, ksess,
968 session->output_traces ? 1 : 0);
969 pthread_mutex_unlock(socket->lock);
970 if (ret < 0) {
e7fe706f 971 rcu_read_unlock();
4a15001e 972 goto error;
7a485870 973 }
7a485870 974 }
4a15001e 975 rcu_read_unlock();
7a485870 976 }
54d01ffb 977 session_unlock(session);
7a485870 978 }
54d01ffb 979 session_unlock_list();
b3c750d2 980 return ret;
7a485870 981
b3c750d2 982error:
54d01ffb
DG
983 session_unlock(session);
984 session_unlock_list();
7a485870
DG
985 return ret;
986}
987
487cf67c 988/*
ffe60014
DG
989 * For each tracing session, update newly registered apps. The session list
990 * lock MUST be acquired before calling this.
487cf67c
DG
991 */
992static void update_ust_app(int app_sock)
993{
994 struct ltt_session *sess, *stmp;
995
fdadac08
DG
996 /* Consumer is in an ERROR state. Stop any application update. */
997 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
998 /* Stop the update process since the consumer is dead. */
999 return;
1000 }
1001
487cf67c
DG
1002 /* For all tracing session(s) */
1003 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
a9ad0c8f
MD
1004 struct ust_app *app;
1005
4ee14516 1006 session_lock(sess);
a9ad0c8f
MD
1007 if (!sess->ust_session) {
1008 goto unlock_session;
1009 }
1010
1011 rcu_read_lock();
1012 assert(app_sock >= 0);
1013 app = ust_app_find_by_sock(app_sock);
1014 if (app == NULL) {
1015 /*
1016 * Application can be unregistered before so
1017 * this is possible hence simply stopping the
1018 * update.
1019 */
1020 DBG3("UST app update failed to find app sock %d",
1021 app_sock);
1022 goto unlock_rcu;
421cb601 1023 }
a9ad0c8f
MD
1024 ust_app_global_update(sess->ust_session, app);
1025 unlock_rcu:
1026 rcu_read_unlock();
1027 unlock_session:
4ee14516 1028 session_unlock(sess);
487cf67c
DG
1029 }
1030}
1031
7a485870 1032/*
d063d709 1033 * This thread manage event coming from the kernel.
7a485870 1034 *
d063d709
DG
1035 * Features supported in this thread:
1036 * -) CPU Hotplug
7a485870
DG
1037 */
1038static void *thread_manage_kernel(void *data)
1039{
139ac872 1040 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 1041 uint32_t revents, nb_fd;
7a485870 1042 char tmp;
5eb91c98 1043 struct lttng_poll_event events;
7a485870 1044
6993eeb3 1045 DBG("[thread] Thread manage kernel started");
7a485870 1046
6c71277b 1047 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
927ca06a 1048
d5d63bf1
DG
1049 /*
1050 * This first step of the while is to clean this structure which could free
6d737ce4 1051 * non NULL pointers so initialize it before the loop.
d5d63bf1 1052 */
6d737ce4 1053 lttng_poll_init(&events);
d5d63bf1 1054
e547b070 1055 if (testpoint(sessiond_thread_manage_kernel)) {
6993eeb3
CB
1056 goto error_testpoint;
1057 }
8ac94142 1058
840cb59c 1059 health_code_update();
44a5e5eb 1060
e547b070 1061 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
d21b0d71 1062 goto error_testpoint;
6993eeb3
CB
1063 }
1064
7a485870 1065 while (1) {
840cb59c 1066 health_code_update();
44a5e5eb 1067
7a485870 1068 if (update_poll_flag == 1) {
d21b0d71
DG
1069 /* Clean events object. We are about to populate it again. */
1070 lttng_poll_clean(&events);
1071
d0b96690 1072 ret = sessiond_set_thread_pollset(&events, 2);
d21b0d71
DG
1073 if (ret < 0) {
1074 goto error_poll_create;
1075 }
1076
1077 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1078 if (ret < 0) {
1079 goto error;
1080 }
5f822d0a 1081
d21b0d71 1082 /* This will add the available kernel channel if any. */
5eb91c98
DG
1083 ret = update_kernel_poll(&events);
1084 if (ret < 0) {
7a485870
DG
1085 goto error;
1086 }
1087 update_poll_flag = 0;
1088 }
1089
7fa2082e 1090 DBG("Thread kernel polling");
7a485870
DG
1091
1092 /* Poll infinite value of time */
88f2b785 1093 restart:
a78af745 1094 health_poll_entry();
5eb91c98 1095 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1096 DBG("Thread kernel return from poll on %d fds",
1097 LTTNG_POLL_GETNB(&events));
a78af745 1098 health_poll_exit();
7a485870 1099 if (ret < 0) {
88f2b785
MD
1100 /*
1101 * Restart interrupted system call.
1102 */
1103 if (errno == EINTR) {
1104 goto restart;
1105 }
7a485870
DG
1106 goto error;
1107 } else if (ret == 0) {
1108 /* Should not happen since timeout is infinite */
85611738
DG
1109 ERR("Return value of poll is 0 with an infinite timeout.\n"
1110 "This should not have happened! Continuing...");
7a485870
DG
1111 continue;
1112 }
1113
0d9c5d77
DG
1114 nb_fd = ret;
1115
5eb91c98
DG
1116 for (i = 0; i < nb_fd; i++) {
1117 /* Fetch once the poll data */
1118 revents = LTTNG_POLL_GETEV(&events, i);
1119 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 1120
840cb59c 1121 health_code_update();
44a5e5eb 1122
fd20dac9
MD
1123 if (!revents) {
1124 /* No activity for this FD (poll implementation). */
1125 continue;
1126 }
1127
5eb91c98 1128 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1129 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1130 if (ret) {
139ac872
MD
1131 err = 0;
1132 goto exit;
5eb91c98 1133 }
7a485870 1134
5eb91c98
DG
1135 /* Check for data on kernel pipe */
1136 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
6cd525e8
MD
1137 (void) lttng_read(kernel_poll_pipe[0],
1138 &tmp, 1);
f921c78f
DG
1139 /*
1140 * Ret value is useless here, if this pipe gets any actions an
1141 * update is required anyway.
1142 */
5eb91c98
DG
1143 update_poll_flag = 1;
1144 continue;
1145 } else {
1146 /*
1147 * New CPU detected by the kernel. Adding kernel stream to
1148 * kernel session and updating the kernel consumer
1149 */
1150 if (revents & LPOLLIN) {
2bdd86d4 1151 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
1152 if (ret < 0) {
1153 continue;
1154 }
1155 break;
1156 /*
1157 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1158 * and unregister kernel stream at this point.
1159 */
7a485870 1160 }
7a485870
DG
1161 }
1162 }
1163 }
1164
139ac872 1165exit:
7a485870 1166error:
5eb91c98 1167 lttng_poll_clean(&events);
76d7553f 1168error_poll_create:
6993eeb3 1169error_testpoint:
6620da75
DG
1170 utils_close_pipe(kernel_poll_pipe);
1171 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 1172 if (err) {
840cb59c 1173 health_error();
139ac872 1174 ERR("Health error occurred in %s", __func__);
6620da75
DG
1175 WARN("Kernel thread died unexpectedly. "
1176 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 1177 }
8782cc74 1178 health_unregister(health_sessiond);
76d7553f 1179 DBG("Kernel thread dying");
7a485870
DG
1180 return NULL;
1181}
1182
a23ec3a7
DG
1183/*
1184 * Signal pthread condition of the consumer data that the thread.
1185 */
1186static void signal_consumer_condition(struct consumer_data *data, int state)
1187{
1188 pthread_mutex_lock(&data->cond_mutex);
1189
1190 /*
1191 * The state is set before signaling. It can be any value, it's the waiter
1192 * job to correctly interpret this condition variable associated to the
1193 * consumer pthread_cond.
1194 *
1195 * A value of 0 means that the corresponding thread of the consumer data
1196 * was not started. 1 indicates that the thread has started and is ready
1197 * for action. A negative value means that there was an error during the
1198 * thread bootstrap.
1199 */
1200 data->consumer_thread_is_ready = state;
1201 (void) pthread_cond_signal(&data->cond);
1202
1203 pthread_mutex_unlock(&data->cond_mutex);
1204}
1205
1d4b027a 1206/*
3bd1e081 1207 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 1208 */
3bd1e081 1209static void *thread_manage_consumer(void *data)
1d4b027a 1210{
42fc1d0b 1211 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
5eb91c98 1212 uint32_t revents, nb_fd;
1d4b027a 1213 enum lttcomm_return_code code;
5eb91c98 1214 struct lttng_poll_event events;
3bd1e081 1215 struct consumer_data *consumer_data = data;
1d4b027a 1216
3bd1e081 1217 DBG("[thread] Manage consumer started");
1d4b027a 1218
34c1e15a
MD
1219 rcu_register_thread();
1220 rcu_thread_online();
1221
6c71277b 1222 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
927ca06a 1223
855060f8 1224 health_code_update();
9449cc75 1225
5eb91c98 1226 /*
331744e3
JD
1227 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1228 * metadata_sock. Nothing more will be added to this poll set.
5eb91c98 1229 */
331744e3 1230 ret = sessiond_set_thread_pollset(&events, 3);
5eb91c98 1231 if (ret < 0) {
76d7553f 1232 goto error_poll;
5eb91c98 1233 }
273ea72c 1234
edb8b045
DG
1235 /*
1236 * The error socket here is already in a listening state which was done
1237 * just before spawning this thread to avoid a race between the consumer
1238 * daemon exec trying to connect and the listen() call.
1239 */
3bd1e081 1240 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
1241 if (ret < 0) {
1242 goto error;
1243 }
1244
840cb59c 1245 health_code_update();
44a5e5eb 1246
331744e3 1247 /* Infinite blocking call, waiting for transmission */
88f2b785 1248restart:
a78af745 1249 health_poll_entry();
8ac94142 1250
e547b070 1251 if (testpoint(sessiond_thread_manage_consumer)) {
6993eeb3
CB
1252 goto error;
1253 }
8ac94142 1254
5eb91c98 1255 ret = lttng_poll_wait(&events, -1);
a78af745 1256 health_poll_exit();
273ea72c 1257 if (ret < 0) {
88f2b785
MD
1258 /*
1259 * Restart interrupted system call.
1260 */
1261 if (errno == EINTR) {
1262 goto restart;
1263 }
273ea72c
DG
1264 goto error;
1265 }
1266
0d9c5d77
DG
1267 nb_fd = ret;
1268
5eb91c98
DG
1269 for (i = 0; i < nb_fd; i++) {
1270 /* Fetch once the poll data */
1271 revents = LTTNG_POLL_GETEV(&events, i);
1272 pollfd = LTTNG_POLL_GETFD(&events, i);
1273
840cb59c 1274 health_code_update();
44a5e5eb 1275
fd20dac9
MD
1276 if (!revents) {
1277 /* No activity for this FD (poll implementation). */
1278 continue;
1279 }
1280
5eb91c98 1281 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1282 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1283 if (ret) {
139ac872
MD
1284 err = 0;
1285 goto exit;
5eb91c98
DG
1286 }
1287
1288 /* Event on the registration socket */
3bd1e081 1289 if (pollfd == consumer_data->err_sock) {
5eb91c98 1290 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1291 ERR("consumer err socket poll error");
5eb91c98
DG
1292 goto error;
1293 }
1294 }
273ea72c
DG
1295 }
1296
3bd1e081 1297 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
1298 if (sock < 0) {
1299 goto error;
1300 }
1301
b662582b
DG
1302 /*
1303 * Set the CLOEXEC flag. Return code is useless because either way, the
1304 * show must go on.
1305 */
1306 (void) utils_set_fd_cloexec(sock);
1307
840cb59c 1308 health_code_update();
44a5e5eb 1309
3bd1e081 1310 DBG2("Receiving code from consumer err_sock");
ee0b0061 1311
712ea556 1312 /* Getting status code from kconsumerd */
54d01ffb
DG
1313 ret = lttcomm_recv_unix_sock(sock, &code,
1314 sizeof(enum lttcomm_return_code));
1d4b027a
DG
1315 if (ret <= 0) {
1316 goto error;
1317 }
1318
840cb59c 1319 health_code_update();
f73fabfd 1320 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
331744e3 1321 /* Connect both socket, command and metadata. */
3bd1e081
MD
1322 consumer_data->cmd_sock =
1323 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
4ce514c4 1324 consumer_data->metadata_fd =
331744e3 1325 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
92db7cdc
DG
1326 if (consumer_data->cmd_sock < 0
1327 || consumer_data->metadata_fd < 0) {
331744e3 1328 PERROR("consumer connect cmd socket");
a23ec3a7
DG
1329 /* On error, signal condition and quit. */
1330 signal_consumer_condition(consumer_data, -1);
1d4b027a
DG
1331 goto error;
1332 }
9363801e 1333 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
331744e3
JD
1334 /* Create metadata socket lock. */
1335 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1336 if (consumer_data->metadata_sock.lock == NULL) {
1337 PERROR("zmalloc pthread mutex");
1338 ret = -1;
1339 goto error;
1340 }
1341 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1342
a23ec3a7 1343 signal_consumer_condition(consumer_data, 1);
331744e3
JD
1344 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1345 DBG("Consumer metadata socket ready (fd: %d)",
4ce514c4 1346 consumer_data->metadata_fd);
1d4b027a 1347 } else {
3bd1e081 1348 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
1349 lttcomm_get_readable_code(-code));
1350 goto error;
1351 }
1352
331744e3 1353 /* Remove the consumerd error sock since we've established a connexion */
3bd1e081 1354 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1355 if (ret < 0) {
72079cae
DG
1356 goto error;
1357 }
1358
331744e3 1359 /* Add new accepted error socket. */
5eb91c98
DG
1360 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1361 if (ret < 0) {
72079cae 1362 goto error;
5eb91c98
DG
1363 }
1364
331744e3 1365 /* Add metadata socket that is successfully connected. */
4ce514c4 1366 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
331744e3
JD
1367 LPOLLIN | LPOLLRDHUP);
1368 if (ret < 0) {
1369 goto error;
1370 }
1371
840cb59c 1372 health_code_update();
44a5e5eb 1373
331744e3 1374 /* Infinite blocking call, waiting for transmission */
88f2b785 1375restart_poll:
331744e3 1376 while (1) {
42fc1d0b
DG
1377 health_code_update();
1378
1379 /* Exit the thread because the thread quit pipe has been triggered. */
1380 if (should_quit) {
1381 /* Not a health error. */
1382 err = 0;
1383 goto exit;
1384 }
1385
331744e3
JD
1386 health_poll_entry();
1387 ret = lttng_poll_wait(&events, -1);
1388 health_poll_exit();
1389 if (ret < 0) {
1390 /*
1391 * Restart interrupted system call.
1392 */
1393 if (errno == EINTR) {
1394 goto restart_poll;
1395 }
1396 goto error;
88f2b785 1397 }
72079cae 1398
331744e3 1399 nb_fd = ret;
0d9c5d77 1400
331744e3
JD
1401 for (i = 0; i < nb_fd; i++) {
1402 /* Fetch once the poll data */
1403 revents = LTTNG_POLL_GETEV(&events, i);
1404 pollfd = LTTNG_POLL_GETFD(&events, i);
5eb91c98 1405
331744e3 1406 health_code_update();
44a5e5eb 1407
fd20dac9
MD
1408 if (!revents) {
1409 /* No activity for this FD (poll implementation). */
1410 continue;
1411 }
1412
42fc1d0b
DG
1413 /*
1414 * Thread quit pipe has been triggered, flag that we should stop
1415 * but continue the current loop to handle potential data from
1416 * consumer.
1417 */
1418 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1419
331744e3
JD
1420 if (pollfd == sock) {
1421 /* Event on the consumerd socket */
1422 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1423 ERR("consumer err socket second poll error");
1424 goto error;
1425 }
1426 health_code_update();
1427 /* Wait for any kconsumerd error */
1428 ret = lttcomm_recv_unix_sock(sock, &code,
1429 sizeof(enum lttcomm_return_code));
1430 if (ret <= 0) {
1431 ERR("consumer closed the command socket");
1432 goto error;
1433 }
1434
1435 ERR("consumer return code : %s",
1436 lttcomm_get_readable_code(-code));
1437
1438 goto exit;
4ce514c4 1439 } else if (pollfd == consumer_data->metadata_fd) {
331744e3
JD
1440 /* UST metadata requests */
1441 ret = ust_consumer_metadata_request(
1442 &consumer_data->metadata_sock);
1443 if (ret < 0) {
1444 ERR("Handling metadata request");
1445 goto error;
1446 }
5eb91c98 1447 }
42fc1d0b 1448 /* No need for an else branch all FDs are tested prior. */
5eb91c98 1449 }
331744e3 1450 health_code_update();
5eb91c98
DG
1451 }
1452
139ac872 1453exit:
1d4b027a 1454error:
fdadac08
DG
1455 /*
1456 * We lock here because we are about to close the sockets and some other
92db7cdc
DG
1457 * thread might be using them so get exclusive access which will abort all
1458 * other consumer command by other threads.
fdadac08
DG
1459 */
1460 pthread_mutex_lock(&consumer_data->lock);
1461
5c827ce0
DG
1462 /* Immediately set the consumerd state to stopped */
1463 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1464 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1465 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1466 consumer_data->type == LTTNG_CONSUMER32_UST) {
1467 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1468 } else {
1469 /* Code flow error... */
1470 assert(0);
1471 }
1472
76d7553f
MD
1473 if (consumer_data->err_sock >= 0) {
1474 ret = close(consumer_data->err_sock);
1475 if (ret) {
1476 PERROR("close");
1477 }
a76cbd9f 1478 consumer_data->err_sock = -1;
76d7553f
MD
1479 }
1480 if (consumer_data->cmd_sock >= 0) {
1481 ret = close(consumer_data->cmd_sock);
1482 if (ret) {
1483 PERROR("close");
1484 }
a76cbd9f 1485 consumer_data->cmd_sock = -1;
76d7553f 1486 }
96544455
SS
1487 if (consumer_data->metadata_sock.fd_ptr &&
1488 *consumer_data->metadata_sock.fd_ptr >= 0) {
9363801e 1489 ret = close(*consumer_data->metadata_sock.fd_ptr);
331744e3
JD
1490 if (ret) {
1491 PERROR("close");
1492 }
1493 }
76d7553f
MD
1494 if (sock >= 0) {
1495 ret = close(sock);
1496 if (ret) {
1497 PERROR("close");
1498 }
1499 }
273ea72c 1500
3bd1e081
MD
1501 unlink(consumer_data->err_unix_sock_path);
1502 unlink(consumer_data->cmd_unix_sock_path);
1503 consumer_data->pid = 0;
fdadac08 1504 pthread_mutex_unlock(&consumer_data->lock);
92db7cdc 1505
fdadac08 1506 /* Cleanup metadata socket mutex. */
96544455
SS
1507 if (consumer_data->metadata_sock.lock) {
1508 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1509 free(consumer_data->metadata_sock.lock);
1510 }
5eb91c98 1511 lttng_poll_clean(&events);
76d7553f 1512error_poll:
139ac872 1513 if (err) {
840cb59c 1514 health_error();
139ac872
MD
1515 ERR("Health error occurred in %s", __func__);
1516 }
8782cc74 1517 health_unregister(health_sessiond);
76d7553f 1518 DBG("consumer thread cleanup completed");
0177d773 1519
34c1e15a
MD
1520 rcu_thread_offline();
1521 rcu_unregister_thread();
1522
5eb91c98 1523 return NULL;
099e26bd
DG
1524}
1525
099e26bd
DG
1526/*
1527 * This thread manage application communication.
1d4b027a
DG
1528 */
1529static void *thread_manage_apps(void *data)
099e26bd 1530{
139ac872 1531 int i, ret, pollfd, err = -1;
6cd525e8 1532 ssize_t size_ret;
5eb91c98 1533 uint32_t revents, nb_fd;
5eb91c98 1534 struct lttng_poll_event events;
099e26bd
DG
1535
1536 DBG("[thread] Manage application started");
1537
f6a9efaa
DG
1538 rcu_register_thread();
1539 rcu_thread_online();
1540
6c71277b 1541 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
927ca06a 1542
e547b070 1543 if (testpoint(sessiond_thread_manage_apps)) {
6993eeb3
CB
1544 goto error_testpoint;
1545 }
1546
840cb59c 1547 health_code_update();
44a5e5eb 1548
d0b96690 1549 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1550 if (ret < 0) {
76d7553f 1551 goto error_poll_create;
5eb91c98 1552 }
099e26bd 1553
5eb91c98
DG
1554 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1555 if (ret < 0) {
1556 goto error;
1557 }
099e26bd 1558
e547b070 1559 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
6993eeb3
CB
1560 goto error;
1561 }
8ac94142 1562
840cb59c 1563 health_code_update();
44a5e5eb 1564
5eb91c98 1565 while (1) {
7fa2082e 1566 DBG("Apps thread polling");
099e26bd
DG
1567
1568 /* Inifinite blocking call, waiting for transmission */
88f2b785 1569 restart:
a78af745 1570 health_poll_entry();
5eb91c98 1571 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1572 DBG("Apps thread return from poll on %d fds",
1573 LTTNG_POLL_GETNB(&events));
a78af745 1574 health_poll_exit();
099e26bd 1575 if (ret < 0) {
88f2b785
MD
1576 /*
1577 * Restart interrupted system call.
1578 */
1579 if (errno == EINTR) {
1580 goto restart;
1581 }
099e26bd
DG
1582 goto error;
1583 }
1584
0d9c5d77
DG
1585 nb_fd = ret;
1586
5eb91c98
DG
1587 for (i = 0; i < nb_fd; i++) {
1588 /* Fetch once the poll data */
1589 revents = LTTNG_POLL_GETEV(&events, i);
1590 pollfd = LTTNG_POLL_GETFD(&events, i);
1591
840cb59c 1592 health_code_update();
44a5e5eb 1593
fd20dac9
MD
1594 if (!revents) {
1595 /* No activity for this FD (poll implementation). */
1596 continue;
1597 }
1598
5eb91c98 1599 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1600 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1601 if (ret) {
139ac872
MD
1602 err = 0;
1603 goto exit;
5eb91c98 1604 }
099e26bd 1605
5eb91c98
DG
1606 /* Inspect the apps cmd pipe */
1607 if (pollfd == apps_cmd_pipe[0]) {
1608 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1609 ERR("Apps command pipe error");
0177d773 1610 goto error;
5eb91c98 1611 } else if (revents & LPOLLIN) {
d0b96690
DG
1612 int sock;
1613
5eb91c98 1614 /* Empty pipe */
6cd525e8
MD
1615 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1616 if (size_ret < sizeof(sock)) {
76d7553f 1617 PERROR("read apps cmd pipe");
5eb91c98
DG
1618 goto error;
1619 }
099e26bd 1620
840cb59c 1621 health_code_update();
44a5e5eb 1622
ffe60014 1623 /*
d0b96690
DG
1624 * We only monitor the error events of the socket. This
1625 * thread does not handle any incoming data from UST
1626 * (POLLIN).
ffe60014 1627 */
d0b96690
DG
1628 ret = lttng_poll_add(&events, sock,
1629 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1630 if (ret < 0) {
5eb91c98 1631 goto error;
e0c7ec2b 1632 }
acc7b41b 1633
d0b96690 1634 DBG("Apps with sock %d added to poll set", sock);
0177d773 1635 }
5eb91c98
DG
1636 } else {
1637 /*
54d01ffb
DG
1638 * At this point, we know that a registered application made
1639 * the event at poll_wait.
5eb91c98
DG
1640 */
1641 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1642 /* Removing from the poll set */
1643 ret = lttng_poll_del(&events, pollfd);
1644 if (ret < 0) {
1645 goto error;
1646 }
099e26bd 1647
b9d9b220 1648 /* Socket closed on remote end. */
56fff090 1649 ust_app_unregister(pollfd);
5eb91c98 1650 }
099e26bd 1651 }
44a5e5eb 1652
840cb59c 1653 health_code_update();
099e26bd 1654 }
099e26bd
DG
1655 }
1656
139ac872 1657exit:
099e26bd 1658error:
5eb91c98 1659 lttng_poll_clean(&events);
76d7553f 1660error_poll_create:
6993eeb3 1661error_testpoint:
6620da75
DG
1662 utils_close_pipe(apps_cmd_pipe);
1663 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1664
1665 /*
1666 * We don't clean the UST app hash table here since already registered
1667 * applications can still be controlled so let them be until the session
1668 * daemon dies or the applications stop.
1669 */
1670
139ac872 1671 if (err) {
840cb59c 1672 health_error();
139ac872
MD
1673 ERR("Health error occurred in %s", __func__);
1674 }
8782cc74 1675 health_unregister(health_sessiond);
76d7553f 1676 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1677 rcu_thread_offline();
1678 rcu_unregister_thread();
099e26bd
DG
1679 return NULL;
1680}
1681
d0b96690 1682/*
d88aee68
DG
1683 * Send a socket to a thread This is called from the dispatch UST registration
1684 * thread once all sockets are set for the application.
d0b96690 1685 *
b85dc84c
DG
1686 * The sock value can be invalid, we don't really care, the thread will handle
1687 * it and make the necessary cleanup if so.
1688 *
d0b96690
DG
1689 * On success, return 0 else a negative value being the errno message of the
1690 * write().
1691 */
d88aee68 1692static int send_socket_to_thread(int fd, int sock)
d0b96690 1693{
6cd525e8 1694 ssize_t ret;
d0b96690 1695
b85dc84c
DG
1696 /*
1697 * It's possible that the FD is set as invalid with -1 concurrently just
1698 * before calling this function being a shutdown state of the thread.
1699 */
1700 if (fd < 0) {
1701 ret = -EBADF;
1702 goto error;
1703 }
d0b96690 1704
6cd525e8
MD
1705 ret = lttng_write(fd, &sock, sizeof(sock));
1706 if (ret < sizeof(sock)) {
d88aee68 1707 PERROR("write apps pipe %d", fd);
d0b96690
DG
1708 if (ret < 0) {
1709 ret = -errno;
1710 }
1711 goto error;
1712 }
1713
1714 /* All good. Don't send back the write positive ret value. */
1715 ret = 0;
1716error:
6cd525e8 1717 return (int) ret;
d0b96690
DG
1718}
1719
f45e313d
DG
1720/*
1721 * Sanitize the wait queue of the dispatch registration thread meaning removing
1722 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1723 * notify socket is never received.
1724 */
1725static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1726{
1727 int ret, nb_fd = 0, i;
1728 unsigned int fd_added = 0;
1729 struct lttng_poll_event events;
1730 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1731
1732 assert(wait_queue);
1733
1734 lttng_poll_init(&events);
1735
1736 /* Just skip everything for an empty queue. */
1737 if (!wait_queue->count) {
1738 goto end;
1739 }
1740
1741 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1742 if (ret < 0) {
1743 goto error_create;
1744 }
1745
1746 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1747 &wait_queue->head, head) {
1748 assert(wait_node->app);
1749 ret = lttng_poll_add(&events, wait_node->app->sock,
1750 LPOLLHUP | LPOLLERR);
1751 if (ret < 0) {
1752 goto error;
1753 }
1754
1755 fd_added = 1;
1756 }
1757
1758 if (!fd_added) {
1759 goto end;
1760 }
1761
1762 /*
1763 * Poll but don't block so we can quickly identify the faulty events and
1764 * clean them afterwards from the wait queue.
1765 */
1766 ret = lttng_poll_wait(&events, 0);
1767 if (ret < 0) {
1768 goto error;
1769 }
1770 nb_fd = ret;
1771
1772 for (i = 0; i < nb_fd; i++) {
1773 /* Get faulty FD. */
1774 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1775 int pollfd = LTTNG_POLL_GETFD(&events, i);
1776
fd20dac9
MD
1777 if (!revents) {
1778 /* No activity for this FD (poll implementation). */
1779 continue;
1780 }
1781
f45e313d
DG
1782 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1783 &wait_queue->head, head) {
1784 if (pollfd == wait_node->app->sock &&
1785 (revents & (LPOLLHUP | LPOLLERR))) {
1786 cds_list_del(&wait_node->head);
1787 wait_queue->count--;
1788 ust_app_destroy(wait_node->app);
1789 free(wait_node);
1790 break;
1791 }
1792 }
1793 }
1794
1795 if (nb_fd > 0) {
1796 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1797 }
1798
1799end:
1800 lttng_poll_clean(&events);
1801 return;
1802
1803error:
1804 lttng_poll_clean(&events);
1805error_create:
1806 ERR("Unable to sanitize wait queue");
1807 return;
1808}
1809
099e26bd
DG
1810/*
1811 * Dispatch request from the registration threads to the application
1812 * communication thread.
1813 */
1814static void *thread_dispatch_ust_registration(void *data)
1815{
12e2b881 1816 int ret, err = -1;
8bdee6e2 1817 struct cds_wfcq_node *node;
099e26bd 1818 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1819 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1820 struct ust_reg_wait_queue wait_queue = {
1821 .count = 0,
1822 };
d0b96690 1823
6c71277b 1824 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
12e2b881 1825
9ad42ec1
MD
1826 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1827 goto error_testpoint;
1828 }
1829
12e2b881
MD
1830 health_code_update();
1831
f45e313d 1832 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1833
1834 DBG("[thread] Dispatch UST command started");
1835
26c9d55e 1836 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1837 health_code_update();
1838
099e26bd
DG
1839 /* Atomically prepare the queue futex */
1840 futex_nto1_prepare(&ust_cmd_queue.futex);
1841
1842 do {
d0b96690 1843 struct ust_app *app = NULL;
7972aab2 1844 ust_cmd = NULL;
d0b96690 1845
f45e313d
DG
1846 /*
1847 * Make sure we don't have node(s) that have hung up before receiving
1848 * the notify socket. This is to clean the list in order to avoid
1849 * memory leaks from notify socket that are never seen.
1850 */
1851 sanitize_wait_queue(&wait_queue);
1852
12e2b881 1853 health_code_update();
099e26bd 1854 /* Dequeue command for registration */
8bdee6e2 1855 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
099e26bd 1856 if (node == NULL) {
00a17c97 1857 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1858 /* Continue thread execution */
1859 break;
1860 }
1861
1862 ust_cmd = caa_container_of(node, struct ust_command, node);
1863
2f50c8a3
DG
1864 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1865 " gid:%d sock:%d name:%s (version %d.%d)",
1866 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1867 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1868 ust_cmd->sock, ust_cmd->reg_msg.name,
1869 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1870
1871 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1872 wait_node = zmalloc(sizeof(*wait_node));
1873 if (!wait_node) {
1874 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1875 ret = close(ust_cmd->sock);
1876 if (ret < 0) {
1877 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1878 }
51dec90d 1879 lttng_fd_put(LTTNG_FD_APPS, 1);
7972aab2 1880 free(ust_cmd);
d0b96690
DG
1881 goto error;
1882 }
1883 CDS_INIT_LIST_HEAD(&wait_node->head);
1884
1885 /* Create application object if socket is CMD. */
1886 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1887 ust_cmd->sock);
1888 if (!wait_node->app) {
1889 ret = close(ust_cmd->sock);
1890 if (ret < 0) {
1891 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1892 }
51dec90d 1893 lttng_fd_put(LTTNG_FD_APPS, 1);
d88aee68 1894 free(wait_node);
7972aab2 1895 free(ust_cmd);
d0b96690
DG
1896 continue;
1897 }
1898 /*
1899 * Add application to the wait queue so we can set the notify
1900 * socket before putting this object in the global ht.
1901 */
f45e313d
DG
1902 cds_list_add(&wait_node->head, &wait_queue.head);
1903 wait_queue.count++;
d0b96690 1904
7972aab2 1905 free(ust_cmd);
d0b96690
DG
1906 /*
1907 * We have to continue here since we don't have the notify
1908 * socket and the application MUST be added to the hash table
1909 * only at that moment.
1910 */
1911 continue;
1912 } else {
1913 /*
1914 * Look for the application in the local wait queue and set the
1915 * notify socket if found.
1916 */
d88aee68 1917 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1918 &wait_queue.head, head) {
12e2b881 1919 health_code_update();
d0b96690
DG
1920 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1921 wait_node->app->notify_sock = ust_cmd->sock;
1922 cds_list_del(&wait_node->head);
f45e313d 1923 wait_queue.count--;
d0b96690
DG
1924 app = wait_node->app;
1925 free(wait_node);
1926 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1927 break;
1928 }
1929 }
020d7f60
DG
1930
1931 /*
1932 * With no application at this stage the received socket is
1933 * basically useless so close it before we free the cmd data
1934 * structure for good.
1935 */
1936 if (!app) {
1937 ret = close(ust_cmd->sock);
1938 if (ret < 0) {
1939 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1940 }
51dec90d 1941 lttng_fd_put(LTTNG_FD_APPS, 1);
020d7f60 1942 }
7972aab2 1943 free(ust_cmd);
d0b96690
DG
1944 }
1945
1946 if (app) {
d0b96690
DG
1947 /*
1948 * @session_lock_list
1949 *
1950 * Lock the global session list so from the register up to the
1951 * registration done message, no thread can see the application
1952 * and change its state.
1953 */
1954 session_lock_list();
1955 rcu_read_lock();
d88aee68 1956
d0b96690
DG
1957 /*
1958 * Add application to the global hash table. This needs to be
1959 * done before the update to the UST registry can locate the
1960 * application.
1961 */
1962 ust_app_add(app);
d88aee68
DG
1963
1964 /* Set app version. This call will print an error if needed. */
1965 (void) ust_app_version(app);
1966
1967 /* Send notify socket through the notify pipe. */
1968 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1969 app->notify_sock);
1970 if (ret < 0) {
1971 rcu_read_unlock();
1972 session_unlock_list();
b85dc84c
DG
1973 /*
1974 * No notify thread, stop the UST tracing. However, this is
1975 * not an internal error of the this thread thus setting
1976 * the health error code to a normal exit.
1977 */
1978 err = 0;
d88aee68 1979 goto error;
6620da75 1980 }
d88aee68 1981
d0b96690
DG
1982 /*
1983 * Update newly registered application with the tracing
1984 * registry info already enabled information.
1985 */
1986 update_ust_app(app->sock);
d88aee68
DG
1987
1988 /*
1989 * Don't care about return value. Let the manage apps threads
1990 * handle app unregistration upon socket close.
1991 */
1992 (void) ust_app_register_done(app->sock);
1993
1994 /*
1995 * Even if the application socket has been closed, send the app
1996 * to the thread and unregistration will take place at that
1997 * place.
1998 */
1999 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 2000 if (ret < 0) {
d88aee68
DG
2001 rcu_read_unlock();
2002 session_unlock_list();
b85dc84c
DG
2003 /*
2004 * No apps. thread, stop the UST tracing. However, this is
2005 * not an internal error of the this thread thus setting
2006 * the health error code to a normal exit.
2007 */
2008 err = 0;
d88aee68 2009 goto error;
d0b96690 2010 }
d88aee68 2011
d0b96690
DG
2012 rcu_read_unlock();
2013 session_unlock_list();
099e26bd 2014 }
099e26bd
DG
2015 } while (node != NULL);
2016
12e2b881 2017 health_poll_entry();
099e26bd
DG
2018 /* Futex wait on queue. Blocking call on futex() */
2019 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 2020 health_poll_exit();
099e26bd 2021 }
12e2b881
MD
2022 /* Normal exit, no error */
2023 err = 0;
099e26bd
DG
2024
2025error:
d88aee68
DG
2026 /* Clean up wait queue. */
2027 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 2028 &wait_queue.head, head) {
d88aee68 2029 cds_list_del(&wait_node->head);
f45e313d 2030 wait_queue.count--;
d88aee68
DG
2031 free(wait_node);
2032 }
2033
772b8f4d
MD
2034 /* Empty command queue. */
2035 for (;;) {
2036 /* Dequeue command for registration */
2037 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2038 if (node == NULL) {
2039 break;
2040 }
2041 ust_cmd = caa_container_of(node, struct ust_command, node);
2042 ret = close(ust_cmd->sock);
2043 if (ret < 0) {
2044 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2045 }
2046 lttng_fd_put(LTTNG_FD_APPS, 1);
2047 free(ust_cmd);
2048 }
2049
9ad42ec1 2050error_testpoint:
099e26bd 2051 DBG("Dispatch thread dying");
12e2b881
MD
2052 if (err) {
2053 health_error();
2054 ERR("Health error occurred in %s", __func__);
2055 }
8782cc74 2056 health_unregister(health_sessiond);
099e26bd
DG
2057 return NULL;
2058}
2059
2060/*
2061 * This thread manage application registration.
2062 */
2063static void *thread_registration_apps(void *data)
1d4b027a 2064{
139ac872 2065 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
2066 uint32_t revents, nb_fd;
2067 struct lttng_poll_event events;
099e26bd
DG
2068 /*
2069 * Get allocated in this thread, enqueued to a global queue, dequeued and
2070 * freed in the manage apps thread.
2071 */
2072 struct ust_command *ust_cmd = NULL;
1d4b027a 2073
099e26bd 2074 DBG("[thread] Manage application registration started");
1d4b027a 2075
6c71277b 2076 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
927ca06a 2077
e547b070 2078 if (testpoint(sessiond_thread_registration_apps)) {
6993eeb3
CB
2079 goto error_testpoint;
2080 }
8ac94142 2081
1d4b027a
DG
2082 ret = lttcomm_listen_unix_sock(apps_sock);
2083 if (ret < 0) {
76d7553f 2084 goto error_listen;
1d4b027a
DG
2085 }
2086
5eb91c98
DG
2087 /*
2088 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2089 * more will be added to this poll set.
2090 */
d0b96690 2091 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 2092 if (ret < 0) {
76d7553f 2093 goto error_create_poll;
5eb91c98 2094 }
273ea72c 2095
5eb91c98
DG
2096 /* Add the application registration socket */
2097 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2098 if (ret < 0) {
76d7553f 2099 goto error_poll_add;
5eb91c98 2100 }
273ea72c 2101
1d4b027a 2102 /* Notify all applications to register */
0fdd1e2c
DG
2103 ret = notify_ust_apps(1);
2104 if (ret < 0) {
2105 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
2106 "Execution continues but there might be problem for already\n"
2107 "running applications that wishes to register.");
0fdd1e2c 2108 }
1d4b027a
DG
2109
2110 while (1) {
2111 DBG("Accepting application registration");
273ea72c
DG
2112
2113 /* Inifinite blocking call, waiting for transmission */
88f2b785 2114 restart:
a78af745 2115 health_poll_entry();
5eb91c98 2116 ret = lttng_poll_wait(&events, -1);
a78af745 2117 health_poll_exit();
273ea72c 2118 if (ret < 0) {
88f2b785
MD
2119 /*
2120 * Restart interrupted system call.
2121 */
2122 if (errno == EINTR) {
2123 goto restart;
2124 }
273ea72c
DG
2125 goto error;
2126 }
2127
0d9c5d77
DG
2128 nb_fd = ret;
2129
5eb91c98 2130 for (i = 0; i < nb_fd; i++) {
840cb59c 2131 health_code_update();
139ac872 2132
5eb91c98
DG
2133 /* Fetch once the poll data */
2134 revents = LTTNG_POLL_GETEV(&events, i);
2135 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 2136
fd20dac9
MD
2137 if (!revents) {
2138 /* No activity for this FD (poll implementation). */
2139 continue;
2140 }
2141
5eb91c98 2142 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 2143 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 2144 if (ret) {
139ac872
MD
2145 err = 0;
2146 goto exit;
90014c57 2147 }
1d4b027a 2148
5eb91c98
DG
2149 /* Event on the registration socket */
2150 if (pollfd == apps_sock) {
2151 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2152 ERR("Register apps socket poll error");
2153 goto error;
2154 } else if (revents & LPOLLIN) {
2155 sock = lttcomm_accept_unix_sock(apps_sock);
2156 if (sock < 0) {
2157 goto error;
2158 }
099e26bd 2159
16c5c8fa
DG
2160 /*
2161 * Set socket timeout for both receiving and ending.
2162 * app_socket_timeout is in seconds, whereas
2163 * lttcomm_setsockopt_rcv_timeout and
2164 * lttcomm_setsockopt_snd_timeout expect msec as
2165 * parameter.
2166 */
2167 (void) lttcomm_setsockopt_rcv_timeout(sock,
2168 app_socket_timeout * 1000);
2169 (void) lttcomm_setsockopt_snd_timeout(sock,
2170 app_socket_timeout * 1000);
2171
b662582b
DG
2172 /*
2173 * Set the CLOEXEC flag. Return code is useless because
2174 * either way, the show must go on.
2175 */
2176 (void) utils_set_fd_cloexec(sock);
2177
5eb91c98 2178 /* Create UST registration command for enqueuing */
ba7f0ae5 2179 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 2180 if (ust_cmd == NULL) {
76d7553f 2181 PERROR("ust command zmalloc");
41ed8e47
MD
2182 ret = close(sock);
2183 if (ret) {
2184 PERROR("close");
2185 }
5eb91c98
DG
2186 goto error;
2187 }
1d4b027a 2188
5eb91c98
DG
2189 /*
2190 * Using message-based transmissions to ensure we don't
2191 * have to deal with partially received messages.
2192 */
4063050c
MD
2193 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2194 if (ret < 0) {
2195 ERR("Exhausted file descriptors allowed for applications.");
2196 free(ust_cmd);
2197 ret = close(sock);
2198 if (ret) {
2199 PERROR("close");
2200 }
2201 sock = -1;
2202 continue;
2203 }
d88aee68 2204
840cb59c 2205 health_code_update();
d0b96690
DG
2206 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2207 if (ret < 0) {
5eb91c98 2208 free(ust_cmd);
d0b96690 2209 /* Close socket of the application. */
76d7553f
MD
2210 ret = close(sock);
2211 if (ret) {
2212 PERROR("close");
2213 }
4063050c 2214 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 2215 sock = -1;
5eb91c98
DG
2216 continue;
2217 }
840cb59c 2218 health_code_update();
099e26bd 2219
5eb91c98 2220 ust_cmd->sock = sock;
34a2494f 2221 sock = -1;
099e26bd 2222
5eb91c98
DG
2223 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2224 " gid:%d sock:%d name:%s (version %d.%d)",
2225 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2226 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2227 ust_cmd->sock, ust_cmd->reg_msg.name,
2228 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 2229
5eb91c98
DG
2230 /*
2231 * Lock free enqueue the registration request. The red pill
54d01ffb 2232 * has been taken! This apps will be part of the *system*.
5eb91c98 2233 */
8bdee6e2 2234 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
5eb91c98
DG
2235
2236 /*
2237 * Wake the registration queue futex. Implicit memory
8bdee6e2 2238 * barrier with the exchange in cds_wfcq_enqueue.
5eb91c98
DG
2239 */
2240 futex_nto1_wake(&ust_cmd_queue.futex);
2241 }
2242 }
90014c57 2243 }
1d4b027a
DG
2244 }
2245
139ac872 2246exit:
1d4b027a 2247error:
0fdd1e2c
DG
2248 /* Notify that the registration thread is gone */
2249 notify_ust_apps(0);
2250
a4b35e07 2251 if (apps_sock >= 0) {
76d7553f
MD
2252 ret = close(apps_sock);
2253 if (ret) {
2254 PERROR("close");
2255 }
a4b35e07 2256 }
46c3f085 2257 if (sock >= 0) {
76d7553f
MD
2258 ret = close(sock);
2259 if (ret) {
2260 PERROR("close");
2261 }
4063050c 2262 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 2263 }
273ea72c 2264 unlink(apps_unix_sock_path);
0fdd1e2c 2265
76d7553f 2266error_poll_add:
5eb91c98 2267 lttng_poll_clean(&events);
76d7553f
MD
2268error_listen:
2269error_create_poll:
6993eeb3 2270error_testpoint:
76d7553f 2271 DBG("UST Registration thread cleanup complete");
9ad42ec1
MD
2272 if (err) {
2273 health_error();
2274 ERR("Health error occurred in %s", __func__);
2275 }
8782cc74 2276 health_unregister(health_sessiond);
5eb91c98 2277
1d4b027a
DG
2278 return NULL;
2279}
2280
8c0faa1d 2281/*
3bd1e081 2282 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 2283 * exec or it will fails.
8c0faa1d 2284 */
3bd1e081 2285static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 2286{
a23ec3a7 2287 int ret, clock_ret;
ee0b0061
DG
2288 struct timespec timeout;
2289
a23ec3a7
DG
2290 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2291 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 2292
a23ec3a7
DG
2293 /* Setup pthread condition */
2294 ret = pthread_condattr_init(&consumer_data->condattr);
4a15001e 2295 if (ret) {
a23ec3a7
DG
2296 errno = ret;
2297 PERROR("pthread_condattr_init consumer data");
2298 goto error;
2299 }
2300
2301 /*
2302 * Set the monotonic clock in order to make sure we DO NOT jump in time
2303 * between the clock_gettime() call and the timedwait call. See bug #324
2304 * for a more details and how we noticed it.
2305 */
2306 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
4a15001e 2307 if (ret) {
a23ec3a7
DG
2308 errno = ret;
2309 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
2310 goto error;
2311 }
8c0faa1d 2312
a23ec3a7 2313 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
4a15001e 2314 if (ret) {
a23ec3a7
DG
2315 errno = ret;
2316 PERROR("pthread_cond_init consumer data");
2317 goto error;
2318 }
2319
2320 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2321 consumer_data);
4a15001e
MD
2322 if (ret) {
2323 errno = ret;
3bd1e081 2324 PERROR("pthread_create consumer");
ee0b0061 2325 ret = -1;
8c0faa1d
DG
2326 goto error;
2327 }
2328
a23ec3a7
DG
2329 /* We are about to wait on a pthread condition */
2330 pthread_mutex_lock(&consumer_data->cond_mutex);
2331
ee0b0061 2332 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
2333 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2334 /*
2335 * Set the timeout for the condition timed wait even if the clock gettime
2336 * call fails since we might loop on that call and we want to avoid to
2337 * increment the timeout too many times.
2338 */
2339 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2340
2341 /*
2342 * The following loop COULD be skipped in some conditions so this is why we
2343 * set ret to 0 in order to make sure at least one round of the loop is
2344 * done.
2345 */
2346 ret = 0;
2347
2348 /*
2349 * Loop until the condition is reached or when a timeout is reached. Note
2350 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2351 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2352 * possible. This loop does not take any chances and works with both of
2353 * them.
2354 */
2355 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2356 if (clock_ret < 0) {
2357 PERROR("clock_gettime spawn consumer");
2358 /* Infinite wait for the consumerd thread to be ready */
2359 ret = pthread_cond_wait(&consumer_data->cond,
2360 &consumer_data->cond_mutex);
2361 } else {
2362 ret = pthread_cond_timedwait(&consumer_data->cond,
2363 &consumer_data->cond_mutex, &timeout);
2364 }
ee0b0061 2365 }
8c0faa1d 2366
a23ec3a7
DG
2367 /* Release the pthread condition */
2368 pthread_mutex_unlock(&consumer_data->cond_mutex);
2369
2370 if (ret != 0) {
2371 errno = ret;
2372 if (ret == ETIMEDOUT) {
4282f9a3
DG
2373 int pth_ret;
2374
ee0b0061
DG
2375 /*
2376 * Call has timed out so we kill the kconsumerd_thread and return
2377 * an error.
2378 */
a23ec3a7
DG
2379 ERR("Condition timed out. The consumer thread was never ready."
2380 " Killing it");
4282f9a3
DG
2381 pth_ret = pthread_cancel(consumer_data->thread);
2382 if (pth_ret < 0) {
3bd1e081 2383 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
2384 }
2385 } else {
a23ec3a7 2386 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061 2387 }
4282f9a3
DG
2388 /* Caller is expecting a negative value on failure. */
2389 ret = -1;
ee0b0061
DG
2390 goto error;
2391 }
2392
3bd1e081
MD
2393 pthread_mutex_lock(&consumer_data->pid_mutex);
2394 if (consumer_data->pid == 0) {
a23ec3a7 2395 ERR("Consumerd did not start");
3bd1e081 2396 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2397 goto error;
2398 }
3bd1e081 2399 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2400
8c0faa1d
DG
2401 return 0;
2402
2403error:
2404 return ret;
2405}
2406
d9800920 2407/*
3bd1e081 2408 * Join consumer thread
d9800920 2409 */
3bd1e081 2410static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2411{
2412 void *status;
cf3af59e 2413
e8209f6b
DG
2414 /* Consumer pid must be a real one. */
2415 if (consumer_data->pid > 0) {
c617c0c6 2416 int ret;
3bd1e081 2417 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2418 if (ret) {
4a15001e 2419 PERROR("Error killing consumer daemon");
cf3af59e
MD
2420 return ret;
2421 }
3bd1e081 2422 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2423 } else {
2424 return 0;
2425 }
2426}
2427
8c0faa1d 2428/*
3bd1e081 2429 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2430 *
d063d709 2431 * Return pid if successful else -1.
8c0faa1d 2432 */
3bd1e081 2433static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2434{
2435 int ret;
2436 pid_t pid;
94c55f17 2437 const char *consumer_to_use;
53086306 2438 const char *verbosity;
94c55f17 2439 struct stat st;
8c0faa1d 2440
3bd1e081 2441 DBG("Spawning consumerd");
c49dc785 2442
8c0faa1d
DG
2443 pid = fork();
2444 if (pid == 0) {
2445 /*
3bd1e081 2446 * Exec consumerd.
8c0faa1d 2447 */
daee5345 2448 if (opt_verbose_consumer) {
53086306 2449 verbosity = "--verbose";
4421f712 2450 } else if (lttng_opt_quiet) {
53086306 2451 verbosity = "--quiet";
4421f712
DG
2452 } else {
2453 verbosity = "";
53086306 2454 }
4421f712 2455
3bd1e081
MD
2456 switch (consumer_data->type) {
2457 case LTTNG_CONSUMER_KERNEL:
94c55f17 2458 /*
c7704d57
DG
2459 * Find out which consumerd to execute. We will first try the
2460 * 64-bit path, then the sessiond's installation directory, and
2461 * fallback on the 32-bit one,
94c55f17 2462 */
63a799e8
AM
2463 DBG3("Looking for a kernel consumer at these locations:");
2464 DBG3(" 1) %s", consumerd64_bin);
2465 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2466 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2467 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2468 DBG3("Found location #1");
94c55f17 2469 consumer_to_use = consumerd64_bin;
94c55f17 2470 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2471 DBG3("Found location #2");
94c55f17 2472 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2473 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2474 DBG3("Found location #3");
eb1e0bd4 2475 consumer_to_use = consumerd32_bin;
94c55f17 2476 } else {
63a799e8 2477 DBG("Could not find any valid consumerd executable");
4282f9a3 2478 ret = -EINVAL;
94c55f17
AM
2479 break;
2480 }
2481 DBG("Using kernel consumer at: %s", consumer_to_use);
4282f9a3 2482 ret = execl(consumer_to_use,
94c55f17
AM
2483 "lttng-consumerd", verbosity, "-k",
2484 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2485 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2486 "--group", tracing_group_name,
94c55f17 2487 NULL);
3bd1e081 2488 break;
7753dea8
MD
2489 case LTTNG_CONSUMER64_UST:
2490 {
b1e0b6b6 2491 char *tmpnew = NULL;
8f4905da
MD
2492
2493 if (consumerd64_libdir[0] != '\0') {
2494 char *tmp;
2495 size_t tmplen;
2496
e8fa9fb0 2497 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
8f4905da
MD
2498 if (!tmp) {
2499 tmp = "";
2500 }
2501 tmplen = strlen("LD_LIBRARY_PATH=")
2502 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2503 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2504 if (!tmpnew) {
2505 ret = -ENOMEM;
2506 goto error;
2507 }
2508 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2509 strcat(tmpnew, consumerd64_libdir);
2510 if (tmp[0] != '\0') {
2511 strcat(tmpnew, ":");
2512 strcat(tmpnew, tmp);
2513 }
2514 ret = putenv(tmpnew);
2515 if (ret) {
2516 ret = -errno;
c6f76da9 2517 free(tmpnew);
8f4905da
MD
2518 goto error;
2519 }
2520 }
94c55f17 2521 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2522 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2523 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2524 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2525 "--group", tracing_group_name,
7753dea8 2526 NULL);
8f4905da
MD
2527 if (consumerd64_libdir[0] != '\0') {
2528 free(tmpnew);
2529 }
3bd1e081 2530 break;
7753dea8
MD
2531 }
2532 case LTTNG_CONSUMER32_UST:
2533 {
937dde8e 2534 char *tmpnew = NULL;
8f4905da
MD
2535
2536 if (consumerd32_libdir[0] != '\0') {
2537 char *tmp;
2538 size_t tmplen;
2539
e8fa9fb0 2540 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
8f4905da
MD
2541 if (!tmp) {
2542 tmp = "";
2543 }
2544 tmplen = strlen("LD_LIBRARY_PATH=")
2545 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2546 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2547 if (!tmpnew) {
2548 ret = -ENOMEM;
2549 goto error;
2550 }
2551 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2552 strcat(tmpnew, consumerd32_libdir);
2553 if (tmp[0] != '\0') {
2554 strcat(tmpnew, ":");
2555 strcat(tmpnew, tmp);
2556 }
2557 ret = putenv(tmpnew);
2558 if (ret) {
2559 ret = -errno;
c6f76da9 2560 free(tmpnew);
8f4905da
MD
2561 goto error;
2562 }
2563 }
94c55f17 2564 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2565 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2566 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2567 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2568 "--group", tracing_group_name,
7753dea8 2569 NULL);
8f4905da
MD
2570 if (consumerd32_libdir[0] != '\0') {
2571 free(tmpnew);
2572 }
7753dea8
MD
2573 break;
2574 }
3bd1e081 2575 default:
76d7553f 2576 PERROR("unknown consumer type");
3bd1e081
MD
2577 exit(EXIT_FAILURE);
2578 }
8c0faa1d 2579 if (errno != 0) {
4282f9a3 2580 PERROR("Consumer execl()");
8c0faa1d 2581 }
4282f9a3 2582 /* Reaching this point, we got a failure on our execl(). */
8c0faa1d
DG
2583 exit(EXIT_FAILURE);
2584 } else if (pid > 0) {
2585 ret = pid;
8c0faa1d 2586 } else {
76d7553f 2587 PERROR("start consumer fork");
8c0faa1d 2588 ret = -errno;
8c0faa1d 2589 }
8f4905da 2590error:
8c0faa1d
DG
2591 return ret;
2592}
2593
693bd40b 2594/*
3bd1e081 2595 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2596 */
3bd1e081 2597static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2598{
c617c0c6 2599 int ret;
edb8b045
DG
2600
2601 /*
2602 * Set the listen() state on the socket since there is a possible race
2603 * between the exec() of the consumer daemon and this call if place in the
2604 * consumer thread. See bug #366 for more details.
2605 */
2606 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2607 if (ret < 0) {
2608 goto error;
2609 }
693bd40b 2610
3bd1e081
MD
2611 pthread_mutex_lock(&consumer_data->pid_mutex);
2612 if (consumer_data->pid != 0) {
2613 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2614 goto end;
2615 }
693bd40b 2616
3bd1e081 2617 ret = spawn_consumerd(consumer_data);
c49dc785 2618 if (ret < 0) {
3bd1e081
MD
2619 ERR("Spawning consumerd failed");
2620 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2621 goto error;
693bd40b 2622 }
c49dc785 2623
3bd1e081
MD
2624 /* Setting up the consumer_data pid */
2625 consumer_data->pid = ret;
48842b30 2626 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2627 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2628
3bd1e081
MD
2629 DBG2("Spawning consumer control thread");
2630 ret = spawn_consumer_thread(consumer_data);
693bd40b 2631 if (ret < 0) {
3bd1e081 2632 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2633 goto error;
2634 }
2635
c49dc785 2636end:
693bd40b
DG
2637 return 0;
2638
2639error:
331744e3 2640 /* Cleanup already created sockets on error. */
edb8b045 2641 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2642 int err;
2643
edb8b045
DG
2644 err = close(consumer_data->err_sock);
2645 if (err < 0) {
2646 PERROR("close consumer data error socket");
2647 }
2648 }
693bd40b
DG
2649 return ret;
2650}
2651
b73401da 2652/*
096102bd 2653 * Setup necessary data for kernel tracer action.
b73401da 2654 */
096102bd 2655static int init_kernel_tracer(void)
b73401da
DG
2656{
2657 int ret;
b73401da 2658
096102bd
DG
2659 /* Modprobe lttng kernel modules */
2660 ret = modprobe_lttng_control();
b73401da 2661 if (ret < 0) {
b73401da
DG
2662 goto error;
2663 }
2664
096102bd
DG
2665 /* Open debugfs lttng */
2666 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2667 if (kernel_tracer_fd < 0) {
2668 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2669 ret = -1;
2670 goto error_open;
54d01ffb
DG
2671 }
2672
2f77fc4b
DG
2673 /* Validate kernel version */
2674 ret = kernel_validate_version(kernel_tracer_fd);
2675 if (ret < 0) {
2676 goto error_version;
b551a063 2677 }
54d01ffb 2678
2f77fc4b
DG
2679 ret = modprobe_lttng_data();
2680 if (ret < 0) {
2681 goto error_modules;
54d01ffb
DG
2682 }
2683
2f77fc4b
DG
2684 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2685 return 0;
2686
2687error_version:
2688 modprobe_remove_lttng_control();
2689 ret = close(kernel_tracer_fd);
2690 if (ret) {
2691 PERROR("close");
b551a063 2692 }
2f77fc4b 2693 kernel_tracer_fd = -1;
f73fabfd 2694 return LTTNG_ERR_KERN_VERSION;
b551a063 2695
2f77fc4b
DG
2696error_modules:
2697 ret = close(kernel_tracer_fd);
2698 if (ret) {
2699 PERROR("close");
b551a063 2700 }
54d01ffb 2701
2f77fc4b
DG
2702error_open:
2703 modprobe_remove_lttng_control();
54d01ffb
DG
2704
2705error:
2f77fc4b
DG
2706 WARN("No kernel tracer available");
2707 kernel_tracer_fd = -1;
2708 if (!is_root) {
f73fabfd 2709 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2710 } else {
f73fabfd 2711 return LTTNG_ERR_KERN_NA;
2f77fc4b 2712 }
54d01ffb
DG
2713}
2714
2f77fc4b 2715
54d01ffb 2716/*
2f77fc4b
DG
2717 * Copy consumer output from the tracing session to the domain session. The
2718 * function also applies the right modification on a per domain basis for the
2719 * trace files destination directory.
36b588ed
MD
2720 *
2721 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2722 */
2f77fc4b 2723static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2724{
2725 int ret;
2f77fc4b
DG
2726 const char *dir_name;
2727 struct consumer_output *consumer;
2728
2729 assert(session);
2730 assert(session->consumer);
54d01ffb 2731
b551a063
DG
2732 switch (domain) {
2733 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2734 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2735 /*
2736 * XXX: We should audit the session creation and what this function
2737 * does "extra" in order to avoid a destroy since this function is used
2738 * in the domain session creation (kernel and ust) only. Same for UST
2739 * domain.
2740 */
2741 if (session->kernel_session->consumer) {
6addfa37 2742 consumer_output_put(session->kernel_session->consumer);
09a90bcd 2743 }
2f77fc4b
DG
2744 session->kernel_session->consumer =
2745 consumer_copy_output(session->consumer);
2746 /* Ease our life a bit for the next part */
2747 consumer = session->kernel_session->consumer;
2748 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063 2749 break;
f20baf8e 2750 case LTTNG_DOMAIN_JUL:
5cdb6027 2751 case LTTNG_DOMAIN_LOG4J:
0e115563 2752 case LTTNG_DOMAIN_PYTHON:
b551a063 2753 case LTTNG_DOMAIN_UST:
2f77fc4b 2754 DBG3("Copying tracing session consumer output in UST session");
09a90bcd 2755 if (session->ust_session->consumer) {
6addfa37 2756 consumer_output_put(session->ust_session->consumer);
09a90bcd 2757 }
2f77fc4b
DG
2758 session->ust_session->consumer =
2759 consumer_copy_output(session->consumer);
2760 /* Ease our life a bit for the next part */
2761 consumer = session->ust_session->consumer;
2762 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2763 break;
2764 default:
f73fabfd 2765 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2766 goto error;
2767 }
2768
2f77fc4b 2769 /* Append correct directory to subdir */
c30ce0b3
CB
2770 strncat(consumer->subdir, dir_name,
2771 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2772 DBG3("Copy session consumer subdir %s", consumer->subdir);
2773
f73fabfd 2774 ret = LTTNG_OK;
54d01ffb
DG
2775
2776error:
2777 return ret;
2778}
2779
00e2e675 2780/*
2f77fc4b 2781 * Create an UST session and add it to the session ust list.
36b588ed
MD
2782 *
2783 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2784 */
2f77fc4b
DG
2785static int create_ust_session(struct ltt_session *session,
2786 struct lttng_domain *domain)
00e2e675
DG
2787{
2788 int ret;
2f77fc4b 2789 struct ltt_ust_session *lus = NULL;
00e2e675 2790
a4b92340 2791 assert(session);
2f77fc4b
DG
2792 assert(domain);
2793 assert(session->consumer);
a4b92340 2794
2f77fc4b 2795 switch (domain->type) {
f20baf8e 2796 case LTTNG_DOMAIN_JUL:
5cdb6027 2797 case LTTNG_DOMAIN_LOG4J:
0e115563 2798 case LTTNG_DOMAIN_PYTHON:
2f77fc4b
DG
2799 case LTTNG_DOMAIN_UST:
2800 break;
2801 default:
2802 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2803 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2804 goto error;
2805 }
2806
2f77fc4b
DG
2807 DBG("Creating UST session");
2808
dec56f6c 2809 lus = trace_ust_create_session(session->id);
2f77fc4b 2810 if (lus == NULL) {
f73fabfd 2811 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2812 goto error;
2813 }
2814
2f77fc4b
DG
2815 lus->uid = session->uid;
2816 lus->gid = session->gid;
2bba9e53 2817 lus->output_traces = session->output_traces;
27babd3a 2818 lus->snapshot_mode = session->snapshot_mode;
ecc48a90 2819 lus->live_timer_interval = session->live_timer;
2f77fc4b 2820 session->ust_session = lus;
d7ba1388 2821 if (session->shm_path[0]) {
3d071855
MD
2822 strncpy(lus->root_shm_path, session->shm_path,
2823 sizeof(lus->root_shm_path));
2824 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
d7ba1388
MD
2825 strncpy(lus->shm_path, session->shm_path,
2826 sizeof(lus->shm_path));
2827 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2828 strncat(lus->shm_path, "/ust",
2829 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2830 }
2f77fc4b
DG
2831 /* Copy session output to the newly created UST session */
2832 ret = copy_session_consumer(domain->type, session);
f73fabfd 2833 if (ret != LTTNG_OK) {
00e2e675
DG
2834 goto error;
2835 }
2836
f73fabfd 2837 return LTTNG_OK;
00e2e675
DG
2838
2839error:
2f77fc4b
DG
2840 free(lus);
2841 session->ust_session = NULL;
00e2e675
DG
2842 return ret;
2843}
2844
2845/*
2f77fc4b 2846 * Create a kernel tracer session then create the default channel.
00e2e675 2847 */
2f77fc4b 2848static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2849{
2850 int ret;
a4b92340 2851
2f77fc4b 2852 DBG("Creating kernel session");
00e2e675 2853
2f77fc4b
DG
2854 ret = kernel_create_session(session, kernel_tracer_fd);
2855 if (ret < 0) {
f73fabfd 2856 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2857 goto error;
2858 }
2859
2f77fc4b
DG
2860 /* Code flow safety */
2861 assert(session->kernel_session);
2862
2863 /* Copy session output to the newly created Kernel session */
2864 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2865 if (ret != LTTNG_OK) {
a4b92340
DG
2866 goto error;
2867 }
2868
2f77fc4b
DG
2869 /* Create directory(ies) on local filesystem. */
2870 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2871 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2872 ret = run_as_mkdir_recursive(
2873 session->kernel_session->consumer->dst.trace_path,
2874 S_IRWXU | S_IRWXG, session->uid, session->gid);
2875 if (ret < 0) {
df5b86c8 2876 if (errno != EEXIST) {
2f77fc4b 2877 ERR("Trace directory creation error");
00e2e675
DG
2878 goto error;
2879 }
00e2e675 2880 }
2f77fc4b 2881 }
00e2e675 2882
2f77fc4b
DG
2883 session->kernel_session->uid = session->uid;
2884 session->kernel_session->gid = session->gid;
2bba9e53 2885 session->kernel_session->output_traces = session->output_traces;
27babd3a 2886 session->kernel_session->snapshot_mode = session->snapshot_mode;
00e2e675 2887
f73fabfd 2888 return LTTNG_OK;
00e2e675 2889
2f77fc4b
DG
2890error:
2891 trace_kernel_destroy_session(session->kernel_session);
2892 session->kernel_session = NULL;
2893 return ret;
2894}
00e2e675 2895
2f77fc4b
DG
2896/*
2897 * Count number of session permitted by uid/gid.
2898 */
2899static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2900{
2901 unsigned int i = 0;
2902 struct ltt_session *session;
07424f16 2903
2f77fc4b
DG
2904 DBG("Counting number of available session for UID %d GID %d",
2905 uid, gid);
2906 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2907 /*
2f77fc4b 2908 * Only list the sessions the user can control.
00e2e675 2909 */
2f77fc4b
DG
2910 if (!session_access_ok(session, uid, gid)) {
2911 continue;
2912 }
2913 i++;
a4b92340 2914 }
2f77fc4b 2915 return i;
00e2e675
DG
2916}
2917
54d01ffb
DG
2918/*
2919 * Process the command requested by the lttng client within the command
2920 * context structure. This function make sure that the return structure (llm)
2921 * is set and ready for transmission before returning.
2922 *
2923 * Return any error encountered or 0 for success.
53a80697
MD
2924 *
2925 * "sock" is only used for special-case var. len data.
36b588ed
MD
2926 *
2927 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2928 */
53a80697
MD
2929static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2930 int *sock_error)
54d01ffb 2931{
f73fabfd 2932 int ret = LTTNG_OK;
44d3bd01 2933 int need_tracing_session = 1;
2e09ba09 2934 int need_domain;
54d01ffb
DG
2935
2936 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2937
53a80697
MD
2938 *sock_error = 0;
2939
2e09ba09
MD
2940 switch (cmd_ctx->lsm->cmd_type) {
2941 case LTTNG_CREATE_SESSION:
27babd3a 2942 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2943 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09
MD
2944 case LTTNG_DESTROY_SESSION:
2945 case LTTNG_LIST_SESSIONS:
2946 case LTTNG_LIST_DOMAINS:
2947 case LTTNG_START_TRACE:
2948 case LTTNG_STOP_TRACE:
6d805429 2949 case LTTNG_DATA_PENDING:
da3c9ec1
DG
2950 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2951 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2952 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2953 case LTTNG_SNAPSHOT_RECORD:
fb198a11 2954 case LTTNG_SAVE_SESSION:
d7ba1388 2955 case LTTNG_SET_SESSION_SHM_PATH:
2e09ba09 2956 need_domain = 0;
3aace903 2957 break;
2e09ba09
MD
2958 default:
2959 need_domain = 1;
2960 }
2961
2962 if (opt_no_kernel && need_domain
2963 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2964 if (!is_root) {
f73fabfd 2965 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2966 } else {
f73fabfd 2967 ret = LTTNG_ERR_KERN_NA;
531d29f9 2968 }
4fba7219
DG
2969 goto error;
2970 }
2971
8d3113b2
DG
2972 /* Deny register consumer if we already have a spawned consumer. */
2973 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2974 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2975 if (kconsumer_data.pid > 0) {
f73fabfd 2976 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2977 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2978 goto error;
2979 }
2980 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2981 }
2982
54d01ffb
DG
2983 /*
2984 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2985 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2986 * command.
2987 */
2988 switch(cmd_ctx->lsm->cmd_type) {
2989 case LTTNG_LIST_SESSIONS:
2990 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2991 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2992 case LTTNG_LIST_DOMAINS:
2993 case LTTNG_LIST_CHANNELS:
2994 case LTTNG_LIST_EVENTS:
834978fd 2995 case LTTNG_LIST_SYSCALLS:
a5dfbb9d 2996 case LTTNG_LIST_TRACKER_PIDS:
54d01ffb
DG
2997 break;
2998 default:
2999 /* Setup lttng message with no payload */
3000 ret = setup_lttng_msg(cmd_ctx, 0);
3001 if (ret < 0) {
3002 /* This label does not try to unlock the session */
3003 goto init_setup_error;
3004 }
3005 }
3006
3007 /* Commands that DO NOT need a session. */
3008 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 3009 case LTTNG_CREATE_SESSION:
27babd3a 3010 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 3011 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09 3012 case LTTNG_CALIBRATE:
54d01ffb
DG
3013 case LTTNG_LIST_SESSIONS:
3014 case LTTNG_LIST_TRACEPOINTS:
834978fd 3015 case LTTNG_LIST_SYSCALLS:
f37d259d 3016 case LTTNG_LIST_TRACEPOINT_FIELDS:
fb198a11 3017 case LTTNG_SAVE_SESSION:
44d3bd01 3018 need_tracing_session = 0;
54d01ffb
DG
3019 break;
3020 default:
3021 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
3022 /*
3023 * We keep the session list lock across _all_ commands
3024 * for now, because the per-session lock does not
3025 * handle teardown properly.
3026 */
74babd95 3027 session_lock_list();
54d01ffb
DG
3028 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3029 if (cmd_ctx->session == NULL) {
bba2d65f 3030 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
3031 goto error;
3032 } else {
3033 /* Acquire lock for the session */
3034 session_lock(cmd_ctx->session);
3035 }
3036 break;
3037 }
b389abbe 3038
5f3ecf22
DG
3039 /*
3040 * Commands that need a valid session but should NOT create one if none
3041 * exists. Instead of creating one and destroying it when the command is
3042 * handled, process that right before so we save some round trip in useless
3043 * code path.
3044 */
3045 switch (cmd_ctx->lsm->cmd_type) {
3046 case LTTNG_DISABLE_CHANNEL:
3047 case LTTNG_DISABLE_EVENT:
5f3ecf22
DG
3048 switch (cmd_ctx->lsm->domain.type) {
3049 case LTTNG_DOMAIN_KERNEL:
3050 if (!cmd_ctx->session->kernel_session) {
3051 ret = LTTNG_ERR_NO_CHANNEL;
3052 goto error;
3053 }
3054 break;
3055 case LTTNG_DOMAIN_JUL:
5cdb6027 3056 case LTTNG_DOMAIN_LOG4J:
0e115563 3057 case LTTNG_DOMAIN_PYTHON:
5f3ecf22
DG
3058 case LTTNG_DOMAIN_UST:
3059 if (!cmd_ctx->session->ust_session) {
3060 ret = LTTNG_ERR_NO_CHANNEL;
3061 goto error;
3062 }
3063 break;
3064 default:
3065 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3066 goto error;
3067 }
3068 default:
3069 break;
3070 }
3071
2e09ba09
MD
3072 if (!need_domain) {
3073 goto skip_domain;
3074 }
a4b92340 3075
54d01ffb
DG
3076 /*
3077 * Check domain type for specific "pre-action".
3078 */
3079 switch (cmd_ctx->lsm->domain.type) {
3080 case LTTNG_DOMAIN_KERNEL:
d1f1c568 3081 if (!is_root) {
f73fabfd 3082 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
3083 goto error;
3084 }
3085
54d01ffb 3086 /* Kernel tracer check */
a4b35e07 3087 if (kernel_tracer_fd == -1) {
54d01ffb 3088 /* Basically, load kernel tracer modules */
096102bd
DG
3089 ret = init_kernel_tracer();
3090 if (ret != 0) {
54d01ffb
DG
3091 goto error;
3092 }
3093 }
5eb91c98 3094
5c827ce0
DG
3095 /* Consumer is in an ERROR state. Report back to client */
3096 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 3097 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
3098 goto error;
3099 }
3100
54d01ffb 3101 /* Need a session for kernel command */
44d3bd01 3102 if (need_tracing_session) {
54d01ffb 3103 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 3104 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 3105 if (ret < 0) {
f73fabfd 3106 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
3107 goto error;
3108 }
b389abbe 3109 }
7d29a247 3110
54d01ffb 3111 /* Start the kernel consumer daemon */
3bd1e081
MD
3112 pthread_mutex_lock(&kconsumer_data.pid_mutex);