Add application context support to lttng-ctl lttng_add_context
[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
6c1c0768 20#define _LGPL_SOURCE
fac6795d
DG
21#include <getopt.h>
22#include <grp.h>
23#include <limits.h>
0bb7724a 24#include <paths.h>
fac6795d
DG
25#include <pthread.h>
26#include <signal.h>
27#include <stdio.h>
28#include <stdlib.h>
29#include <string.h>
331744e3 30#include <inttypes.h>
0fdd1e2c 31#include <sys/mman.h>
b73401da 32#include <sys/mount.h>
1e307fab 33#include <sys/resource.h>
fac6795d
DG
34#include <sys/socket.h>
35#include <sys/stat.h>
36#include <sys/types.h>
0fdd1e2c 37#include <sys/wait.h>
5c827ce0 38#include <urcu/uatomic.h>
fac6795d
DG
39#include <unistd.h>
40
990570ed 41#include <common/common.h>
d27c42b8 42#include <common/compat/socket.h>
e8fa9fb0 43#include <common/compat/getenv.h>
db758600
DG
44#include <common/defaults.h>
45#include <common/kernel-consumer/kernel-consumer.h>
50c8f484 46#include <common/futex.h>
00e2e675 47#include <common/relayd/relayd.h>
81b86775 48#include <common/utils.h>
3ccdf997 49#include <common/daemonize.h>
f40ef1d5 50#include <common/config/session-config.h>
fac6795d 51
10a8a223 52#include "lttng-sessiond.h"
7972aab2 53#include "buffer-registry.h"
54d01ffb 54#include "channel.h"
2f77fc4b 55#include "cmd.h"
00e2e675 56#include "consumer.h"
099e26bd 57#include "context.h"
54d01ffb 58#include "event.h"
4771f025 59#include "kernel.h"
f1e16794 60#include "kernel-consumer.h"
096102bd 61#include "modprobe.h"
0fdd1e2c 62#include "shm.h"
1e307fab 63#include "ust-ctl.h"
00e2e675 64#include "ust-consumer.h"
8e68d1c8 65#include "utils.h"
4063050c 66#include "fd-limit.h"
8782cc74 67#include "health-sessiond.h"
8ac94142 68#include "testpoint.h"
d0b96690 69#include "ust-thread.h"
022d91ba 70#include "agent-thread.h"
fb198a11 71#include "save.h"
ef367a93 72#include "load-session-thread.h"
834978fd 73#include "syscall.h"
7c1d2758 74#include "agent.h"
fac6795d 75
ebaeda94
MD
76#define CONSUMERD_FILE "lttng-consumerd"
77
fac6795d 78const char *progname;
6c71277b 79static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
26296c48
JG
80static int tracing_group_name_override;
81static char *opt_pidfile;
5b8719f5 82static int opt_sig_parent;
97e19046 83static int opt_verbose_consumer;
72dd7491 84static int opt_daemon, opt_background;
4fba7219 85static int opt_no_kernel;
ef367a93 86static char *opt_load_session_path;
1d4b027a 87static pid_t ppid; /* Parent PID for --sig-parent option */
0bb7724a 88static pid_t child_ppid; /* Internal parent PID use with daemonize. */
67e40797 89static char *rundir;
c9cb3e7d 90static int lockfile_fd = -1;
3bd1e081 91
0bb7724a
DG
92/* Set to 1 when a SIGUSR1 signal is received. */
93static int recv_child_signal;
94
a23ec3a7
DG
95/*
96 * Consumer daemon specific control data. Every value not initialized here is
97 * set to 0 by the static definition.
98 */
3bd1e081
MD
99static struct consumer_data kconsumer_data = {
100 .type = LTTNG_CONSUMER_KERNEL,
60922cb0
DG
101 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
03550b58
MD
103 .err_sock = -1,
104 .cmd_sock = -1,
173af62f
DG
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081 109};
7753dea8
MD
110static struct consumer_data ustconsumer64_data = {
111 .type = LTTNG_CONSUMER64_UST,
60922cb0
DG
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
03550b58
MD
114 .err_sock = -1,
115 .cmd_sock = -1,
173af62f
DG
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
7753dea8
MD
120};
121static struct consumer_data ustconsumer32_data = {
122 .type = LTTNG_CONSUMER32_UST,
60922cb0
DG
123 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
124 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
03550b58
MD
125 .err_sock = -1,
126 .cmd_sock = -1,
173af62f
DG
127 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
128 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
129 .cond = PTHREAD_COND_INITIALIZER,
130 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081
MD
131};
132
26296c48
JG
133/* Command line options */
134static const struct option long_options[] = {
0f5ea17c
JG
135 { "client-sock", required_argument, 0, 'c' },
136 { "apps-sock", required_argument, 0, 'a' },
137 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
138 { "kconsumerd-err-sock", required_argument, 0, '\0' },
139 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
140 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
143 { "consumerd32-path", required_argument, 0, '\0' },
144 { "consumerd32-libdir", required_argument, 0, '\0' },
145 { "consumerd64-path", required_argument, 0, '\0' },
146 { "consumerd64-libdir", required_argument, 0, '\0' },
147 { "daemonize", no_argument, 0, 'd' },
148 { "background", no_argument, 0, 'b' },
149 { "sig-parent", no_argument, 0, 'S' },
150 { "help", no_argument, 0, 'h' },
151 { "group", required_argument, 0, 'g' },
152 { "version", no_argument, 0, 'V' },
153 { "quiet", no_argument, 0, 'q' },
154 { "verbose", no_argument, 0, 'v' },
155 { "verbose-consumer", no_argument, 0, '\0' },
156 { "no-kernel", no_argument, 0, '\0' },
157 { "pidfile", required_argument, 0, 'p' },
158 { "agent-tcp-port", required_argument, 0, '\0' },
159 { "config", required_argument, 0, 'f' },
160 { "load", required_argument, 0, 'l' },
161 { "kmod-probes", required_argument, 0, '\0' },
162 { "extra-kmod-probes", required_argument, 0, '\0' },
26296c48
JG
163 { NULL, 0, 0, 0 }
164};
165
166/* Command line options to ignore from configuration file */
167static const char *config_ignore_options[] = { "help", "version", "config" };
168
26c9d55e 169/* Shared between threads */
099e26bd 170static int dispatch_thread_exit;
fac6795d 171
54d01ffb
DG
172/* Global application Unix socket path */
173static char apps_unix_sock_path[PATH_MAX];
174/* Global client Unix socket path */
175static char client_unix_sock_path[PATH_MAX];
54d01ffb
DG
176/* global wait shm path for UST */
177static char wait_shm_path[PATH_MAX];
44a5e5eb
DG
178/* Global health check unix path */
179static char health_unix_sock_path[PATH_MAX];
fac6795d 180
1d4b027a 181/* Sockets and FDs */
a4b35e07
MD
182static int client_sock = -1;
183static int apps_sock = -1;
2f77fc4b 184int kernel_tracer_fd = -1;
76d7553f 185static int kernel_poll_pipe[2] = { -1, -1 };
1d4b027a 186
273ea72c
DG
187/*
188 * Quit pipe for all threads. This permits a single cancellation point
189 * for all threads when receiving an event on the pipe.
190 */
76d7553f 191static int thread_quit_pipe[2] = { -1, -1 };
4a15001e 192static int ht_cleanup_quit_pipe[2] = { -1, -1 };
273ea72c 193
099e26bd
DG
194/*
195 * This pipe is used to inform the thread managing application communication
196 * that a command is queued and ready to be processed.
197 */
76d7553f 198static int apps_cmd_pipe[2] = { -1, -1 };
099e26bd 199
d0b96690
DG
200int apps_cmd_notify_pipe[2] = { -1, -1 };
201
1d4b027a 202/* Pthread, Mutexes and Semaphores */
1d4b027a 203static pthread_t apps_thread;
d0b96690 204static pthread_t apps_notify_thread;
099e26bd 205static pthread_t reg_apps_thread;
1d4b027a 206static pthread_t client_thread;
7a485870 207static pthread_t kernel_thread;
099e26bd 208static pthread_t dispatch_thread;
44a5e5eb 209static pthread_t health_thread;
0b2dc8df 210static pthread_t ht_cleanup_thread;
022d91ba 211static pthread_t agent_reg_thread;
ef367a93 212static pthread_t load_session_thread;
5eb91c98 213
099e26bd
DG
214/*
215 * UST registration command queue. This queue is tied with a futex and uses a N
216 * wakers / 1 waiter implemented and detailed in futex.c/.h
217 *
b22c5da8
DG
218 * The thread_registration_apps and thread_dispatch_ust_registration uses this
219 * queue along with the wait/wake scheme. The thread_manage_apps receives down
220 * the line new application socket and monitors it for any I/O error or clean
221 * close that triggers an unregistration of the application.
099e26bd
DG
222 */
223static struct ust_cmd_queue ust_cmd_queue;
224
b5541356
DG
225/*
226 * Pointer initialized before thread creation.
227 *
228 * This points to the tracing session list containing the session count and a
229 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
230 * MUST NOT be taken if you call a public function in session.c.
04ea676f 231 *
d063d709 232 * The lock is nested inside the structure: session_list_ptr->lock. Please use
54d01ffb 233 * session_lock_list and session_unlock_list for lock acquisition.
b5541356
DG
234 */
235static struct ltt_session_list *session_list_ptr;
236
7753dea8
MD
237int ust_consumerd64_fd = -1;
238int ust_consumerd32_fd = -1;
239
fb6f1fa2
YB
240static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
241static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
242static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
243static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
26296c48
JG
244static int consumerd32_bin_override;
245static int consumerd64_bin_override;
246static int consumerd32_libdir_override;
247static int consumerd64_libdir_override;
fb09408a 248
2f77fc4b
DG
249static const char *module_proc_lttng = "/proc/lttng";
250
5c827ce0
DG
251/*
252 * Consumer daemon state which is changed when spawning it, killing it or in
253 * case of a fatal error.
254 */
255enum consumerd_state {
256 CONSUMER_STARTED = 1,
257 CONSUMER_STOPPED = 2,
258 CONSUMER_ERROR = 3,
259};
260
261/*
262 * This consumer daemon state is used to validate if a client command will be
263 * able to reach the consumer. If not, the client is informed. For instance,
264 * doing a "lttng start" when the consumer state is set to ERROR will return an
265 * error to the client.
266 *
267 * The following example shows a possible race condition of this scheme:
268 *
269 * consumer thread error happens
270 * client cmd arrives
271 * client cmd checks state -> still OK
272 * consumer thread exit, sets error
273 * client cmd try to talk to consumer
274 * ...
275 *
276 * However, since the consumer is a different daemon, we have no way of making
277 * sure the command will reach it safely even with this state flag. This is why
278 * we consider that up to the state validation during command processing, the
279 * command is safe. After that, we can not guarantee the correctness of the
280 * client request vis-a-vis the consumer.
281 */
282static enum consumerd_state ust_consumerd_state;
283static enum consumerd_state kernel_consumerd_state;
284
ae9e45b3
DG
285/*
286 * Socket timeout for receiving and sending in seconds.
287 */
288static int app_socket_timeout;
289
12744796
DG
290/* Set in main() with the current page size. */
291long page_size;
292
8782cc74
MD
293/* Application health monitoring */
294struct health_app *health_sessiond;
295
022d91ba
DG
296/* Agent TCP port for registration. Used by the agent thread. */
297unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
4d076222 298
f43f95a9
DG
299/* Am I root or not. */
300int is_root; /* Set to 1 if the daemon is running as root */
301
26296c48
JG
302const char * const config_section_name = "sessiond";
303
ef367a93
JG
304/* Load session thread information to operate. */
305struct load_session_thread_data *load_info;
306
7c1d2758
JG
307/* Global hash tables */
308struct lttng_ht *agent_apps_ht_by_sock = NULL;
309
97bc1426
MD
310/*
311 * Whether sessiond is ready for commands/health check requests.
312 * NR_LTTNG_SESSIOND_READY must match the number of calls to
ef367a93 313 * sessiond_notify_ready().
97bc1426 314 */
ef367a93 315#define NR_LTTNG_SESSIOND_READY 3
97bc1426
MD
316int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
317
318/* Notify parents that we are ready for cmd and health check */
ef367a93
JG
319LTTNG_HIDDEN
320void sessiond_notify_ready(void)
97bc1426
MD
321{
322 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
323 /*
324 * Notify parent pid that we are ready to accept command
325 * for client side. This ppid is the one from the
326 * external process that spawned us.
327 */
328 if (opt_sig_parent) {
329 kill(ppid, SIGUSR1);
330 }
331
332 /*
333 * Notify the parent of the fork() process that we are
334 * ready.
335 */
72dd7491 336 if (opt_daemon || opt_background) {
97bc1426
MD
337 kill(child_ppid, SIGUSR1);
338 }
339 }
340}
341
fb09408a 342static
7753dea8 343void setup_consumerd_path(void)
fb09408a 344{
fc7a59ce 345 const char *bin, *libdir;
fb09408a 346
7753dea8
MD
347 /*
348 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
349 * native architecture size consumer if CONFIG_CONSUMER*_PATH
350 * has not been defined.
7753dea8 351 */
ebaeda94 352#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
353 if (!consumerd32_bin[0]) {
354 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
355 }
356 if (!consumerd32_libdir[0]) {
357 consumerd32_libdir = INSTALL_LIB_PATH;
358 }
359#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
360 if (!consumerd64_bin[0]) {
361 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 362 }
ebaeda94
MD
363 if (!consumerd64_libdir[0]) {
364 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
365 }
366#else
367#error "Unknown bitness"
368#endif
369
fb09408a
MD
370 /*
371 * runtime env. var. overrides the build default.
372 */
e8fa9fb0 373 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
fc7a59ce
AM
374 if (bin) {
375 consumerd32_bin = bin;
7753dea8 376 }
e8fa9fb0 377 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
fc7a59ce
AM
378 if (bin) {
379 consumerd64_bin = bin;
ebaeda94 380 }
e8fa9fb0 381 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
382 if (libdir) {
383 consumerd32_libdir = libdir;
384 }
e8fa9fb0 385 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
386 if (libdir) {
387 consumerd64_libdir = libdir;
fb09408a
MD
388 }
389}
390
4a15001e
MD
391static
392int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
393 int *a_pipe)
5eb91c98
DG
394{
395 int ret;
396
d0b96690 397 assert(events);
5eb91c98
DG
398
399 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
400 if (ret < 0) {
401 goto error;
402 }
403
404 /* Add quit pipe */
4a15001e 405 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
5eb91c98
DG
406 if (ret < 0) {
407 goto error;
408 }
409
410 return 0;
411
412error:
413 return ret;
414}
415
4a15001e
MD
416/*
417 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
418 */
419int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
420{
421 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
422}
423
424/*
425 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
426 */
427int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
428 size_t size)
429{
430 return __sessiond_set_thread_pollset(events, size,
431 ht_cleanup_quit_pipe);
432}
433
434static
435int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
436{
437 if (fd == a_pipe && (events & LPOLLIN)) {
438 return 1;
439 }
440 return 0;
441}
442
5eb91c98
DG
443/*
444 * Check if the thread quit pipe was triggered.
445 *
446 * Return 1 if it was triggered else 0;
447 */
d0b96690 448int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
5eb91c98 449{
4a15001e
MD
450 return __sessiond_check_thread_quit_pipe(fd, events,
451 thread_quit_pipe[0]);
452}
5eb91c98 453
4a15001e
MD
454/*
455 * Check if the ht_cleanup thread quit pipe was triggered.
456 *
457 * Return 1 if it was triggered else 0;
458 */
459int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
460{
461 return __sessiond_check_thread_quit_pipe(fd, events,
462 ht_cleanup_quit_pipe[0]);
5eb91c98
DG
463}
464
273ea72c 465/*
5eb91c98 466 * Init thread quit pipe.
273ea72c
DG
467 *
468 * Return -1 on error or 0 if all pipes are created.
469 */
4a15001e 470static int __init_thread_quit_pipe(int *a_pipe)
273ea72c 471{
730389d9 472 int ret, i;
273ea72c 473
4a15001e 474 ret = pipe(a_pipe);
273ea72c 475 if (ret < 0) {
730389d9 476 PERROR("thread quit pipe");
273ea72c
DG
477 goto error;
478 }
479
730389d9 480 for (i = 0; i < 2; i++) {
4a15001e 481 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
730389d9
DG
482 if (ret < 0) {
483 PERROR("fcntl");
484 goto error;
485 }
486 }
487
273ea72c
DG
488error:
489 return ret;
490}
491
4a15001e
MD
492static int init_thread_quit_pipe(void)
493{
494 return __init_thread_quit_pipe(thread_quit_pipe);
495}
496
497static int init_ht_cleanup_quit_pipe(void)
498{
499 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
500}
501
099e26bd
DG
502/*
503 * Stop all threads by closing the thread quit pipe.
504 */
cf3af59e
MD
505static void stop_threads(void)
506{
5eb91c98
DG
507 int ret;
508
cf3af59e
MD
509 /* Stopping all threads */
510 DBG("Terminating all threads");
54d01ffb 511 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
512 if (ret < 0) {
513 ERR("write error on thread quit pipe");
514 }
515
099e26bd 516 /* Dispatch thread */
26c9d55e 517 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 518 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
519}
520
e975f9f8
DG
521/*
522 * Close every consumer sockets.
523 */
524static void close_consumer_sockets(void)
525{
526 int ret;
527
528 if (kconsumer_data.err_sock >= 0) {
529 ret = close(kconsumer_data.err_sock);
530 if (ret < 0) {
531 PERROR("kernel consumer err_sock close");
532 }
533 }
534 if (ustconsumer32_data.err_sock >= 0) {
535 ret = close(ustconsumer32_data.err_sock);
536 if (ret < 0) {
a76cbd9f 537 PERROR("UST consumerd32 err_sock close");
e975f9f8
DG
538 }
539 }
540 if (ustconsumer64_data.err_sock >= 0) {
541 ret = close(ustconsumer64_data.err_sock);
542 if (ret < 0) {
a76cbd9f 543 PERROR("UST consumerd64 err_sock close");
e975f9f8
DG
544 }
545 }
546 if (kconsumer_data.cmd_sock >= 0) {
547 ret = close(kconsumer_data.cmd_sock);
548 if (ret < 0) {
549 PERROR("kernel consumer cmd_sock close");
550 }
551 }
552 if (ustconsumer32_data.cmd_sock >= 0) {
553 ret = close(ustconsumer32_data.cmd_sock);
554 if (ret < 0) {
a76cbd9f 555 PERROR("UST consumerd32 cmd_sock close");
e975f9f8
DG
556 }
557 }
558 if (ustconsumer64_data.cmd_sock >= 0) {
559 ret = close(ustconsumer64_data.cmd_sock);
560 if (ret < 0) {
a76cbd9f 561 PERROR("UST consumerd64 cmd_sock close");
e975f9f8
DG
562 }
563 }
564}
565
c9cb3e7d
JG
566/*
567 * Generate the full lock file path using the rundir.
568 *
569 * Return the snprintf() return value thus a negative value is an error.
570 */
571static int generate_lock_file_path(char *path, size_t len)
572{
573 int ret;
574
575 assert(path);
576 assert(rundir);
577
578 /* Build lockfile path from rundir. */
579 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
580 if (ret < 0) {
581 PERROR("snprintf lockfile path");
582 }
583
584 return ret;
585}
586
4e4714cb
JR
587/*
588 * Wait on consumer process termination.
589 *
590 * Need to be called with the consumer data lock held or from a context
591 * ensuring no concurrent access to data (e.g: cleanup).
592 */
593static void wait_consumer(struct consumer_data *consumer_data)
594{
595 pid_t ret;
596 int status;
597
598 if (consumer_data->pid <= 0) {
599 return;
600 }
601
602 DBG("Waiting for complete teardown of consumerd (PID: %d)",
603 consumer_data->pid);
604 ret = waitpid(consumer_data->pid, &status, 0);
605 if (ret == -1) {
606 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
607 }
608 if (!WIFEXITED(status)) {
609 ERR("consumerd termination with error: %d",
610 WEXITSTATUS(ret));
611 }
612 consumer_data->pid = 0;
613}
614
fac6795d 615/*
4a15001e 616 * Cleanup the session daemon's data structures.
fac6795d 617 */
4a15001e 618static void sessiond_cleanup(void)
fac6795d 619{
ef599319 620 int ret;
af9737e9 621 struct ltt_session *sess, *stmp;
8c6c56c2 622 char path[PATH_MAX];
fac6795d 623
4a15001e 624 DBG("Cleanup sessiond");
e07ae692 625
4e449f3f
MD
626 /*
627 * Close the thread quit pipe. It has already done its job,
628 * since we are now called.
629 */
2f77fc4b
DG
630 utils_close_pipe(thread_quit_pipe);
631
35f90c40
DG
632 /*
633 * If opt_pidfile is undefined, the default file will be wiped when
634 * removing the rundir.
635 */
636 if (opt_pidfile) {
637 ret = remove(opt_pidfile);
638 if (ret < 0) {
639 PERROR("remove pidfile %s", opt_pidfile);
640 }
641 }
642
8c6c56c2
MD
643 DBG("Removing sessiond and consumerd content of directory %s", rundir);
644
645 /* sessiond */
646 snprintf(path, PATH_MAX,
647 "%s/%s",
648 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
649 DBG("Removing %s", path);
650 (void) unlink(path);
651
cd9290dd 652 snprintf(path, PATH_MAX, "%s/%s", rundir,
022d91ba 653 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
cd9290dd
DG
654 DBG("Removing %s", path);
655 (void) unlink(path);
656
8c6c56c2
MD
657 /* kconsumerd */
658 snprintf(path, PATH_MAX,
659 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
660 rundir);
661 DBG("Removing %s", path);
662 (void) unlink(path);
663
664 snprintf(path, PATH_MAX,
665 DEFAULT_KCONSUMERD_PATH,
666 rundir);
667 DBG("Removing directory %s", path);
668 (void) rmdir(path);
669
670 /* ust consumerd 32 */
671 snprintf(path, PATH_MAX,
672 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
673 rundir);
674 DBG("Removing %s", path);
675 (void) unlink(path);
676
677 snprintf(path, PATH_MAX,
678 DEFAULT_USTCONSUMERD32_PATH,
679 rundir);
680 DBG("Removing directory %s", path);
681 (void) rmdir(path);
682
683 /* ust consumerd 64 */
684 snprintf(path, PATH_MAX,
685 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
686 rundir);
687 DBG("Removing %s", path);
688 (void) unlink(path);
689
690 snprintf(path, PATH_MAX,
691 DEFAULT_USTCONSUMERD64_PATH,
692 rundir);
693 DBG("Removing directory %s", path);
694 (void) rmdir(path);
5461b305 695
99bab54f 696 DBG("Cleaning up all sessions");
fac6795d 697
b5541356 698 /* Destroy session list mutex */
273ea72c
DG
699 if (session_list_ptr != NULL) {
700 pthread_mutex_destroy(&session_list_ptr->lock);
701
702 /* Cleanup ALL session */
54d01ffb
DG
703 cds_list_for_each_entry_safe(sess, stmp,
704 &session_list_ptr->head, list) {
2f77fc4b 705 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
706 }
707 }
708
4e4714cb
JR
709 wait_consumer(&kconsumer_data);
710 wait_consumer(&ustconsumer64_data);
711 wait_consumer(&ustconsumer32_data);
712
6a4e4039
JG
713 DBG("Cleaning up all agent apps");
714 agent_app_ht_clean();
715
099e26bd 716 DBG("Closing all UST sockets");
56fff090 717 ust_app_clean_list();
7972aab2 718 buffer_reg_destroy_registries();
099e26bd 719
4fba7219
DG
720 if (is_root && !opt_no_kernel) {
721 DBG2("Closing kernel fd");
a4b35e07 722 if (kernel_tracer_fd >= 0) {
76d7553f
MD
723 ret = close(kernel_tracer_fd);
724 if (ret) {
725 PERROR("close");
726 }
a4b35e07 727 }
2f50c8a3 728 DBG("Unloading kernel modules");
096102bd 729 modprobe_remove_lttng_all();
834978fd 730 free(syscall_table);
2f50c8a3 731 }
2f77fc4b 732
e975f9f8
DG
733 close_consumer_sockets();
734
ef367a93
JG
735 if (load_info) {
736 load_session_destroy_data(load_info);
737 free(load_info);
738 }
739
c9cb3e7d
JG
740 /*
741 * Cleanup lock file by deleting it and finaly closing it which will
742 * release the file system lock.
743 */
744 if (lockfile_fd >= 0) {
745 char lockfile_path[PATH_MAX];
746
4a15001e
MD
747 ret = generate_lock_file_path(lockfile_path,
748 sizeof(lockfile_path));
c9cb3e7d
JG
749 if (ret > 0) {
750 ret = remove(lockfile_path);
751 if (ret < 0) {
752 PERROR("remove lock file");
753 }
754 ret = close(lockfile_fd);
755 if (ret < 0) {
756 PERROR("close lock file");
757 }
758 }
759 }
760
761 /*
762 * We do NOT rmdir rundir because there are other processes
763 * using it, for instance lttng-relayd, which can start in
764 * parallel with this teardown.
765 */
766
767 free(rundir);
4a15001e
MD
768}
769
770/*
771 * Cleanup the daemon's option data structures.
772 */
773static void sessiond_cleanup_options(void)
774{
775 DBG("Cleaning up options");
776
777 /*
778 * If the override option is set, the pointer points to a *non* const
779 * thus freeing it even though the variable type is set to const.
780 */
781 if (tracing_group_name_override) {
782 free((void *) tracing_group_name);
783 }
784 if (consumerd32_bin_override) {
785 free((void *) consumerd32_bin);
786 }
787 if (consumerd64_bin_override) {
788 free((void *) consumerd64_bin);
789 }
790 if (consumerd32_libdir_override) {
791 free((void *) consumerd32_libdir);
792 }
793 if (consumerd64_libdir_override) {
794 free((void *) consumerd64_libdir);
795 }
796
797 free(opt_pidfile);
798 free(opt_load_session_path);
799 free(kmod_probes_list);
800 free(kmod_extra_probes_list);
c9cb3e7d 801
7567352f
MD
802 run_as_destroy_worker();
803
421cb601 804 /* <fun> */
f56a39af 805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 /* </fun> */
fac6795d
DG
809}
810
e065084a 811/*
d063d709 812 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 813 *
d063d709 814 * Return lttcomm error code.
e065084a
DG
815 */
816static int send_unix_sock(int sock, void *buf, size_t len)
817{
818 /* Check valid length */
c617c0c6 819 if (len == 0) {
e065084a
DG
820 return -1;
821 }
822
823 return lttcomm_send_unix_sock(sock, buf, len);
824}
825
5461b305 826/*
d063d709 827 * Free memory of a command context structure.
5461b305 828 */
a2fb29a5 829static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 830{
a2fb29a5
DG
831 DBG("Clean command context structure");
832 if (*cmd_ctx) {
833 if ((*cmd_ctx)->llm) {
834 free((*cmd_ctx)->llm);
5461b305 835 }
a2fb29a5
DG
836 if ((*cmd_ctx)->lsm) {
837 free((*cmd_ctx)->lsm);
5461b305 838 }
a2fb29a5
DG
839 free(*cmd_ctx);
840 *cmd_ctx = NULL;
5461b305
DG
841 }
842}
843
fac6795d 844/*
0fdd1e2c 845 * Notify UST applications using the shm mmap futex.
fac6795d 846 */
0fdd1e2c 847static int notify_ust_apps(int active)
fac6795d 848{
0fdd1e2c 849 char *wait_shm_mmap;
fac6795d 850
0fdd1e2c 851 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 852
0fdd1e2c
DG
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
855 if (wait_shm_mmap == NULL) {
fac6795d
DG
856 goto error;
857 }
858
0fdd1e2c
DG
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap, active);
861
862 /* Apps notified successfully */
863 return 0;
fac6795d
DG
864
865error:
0fdd1e2c 866 return -1;
fac6795d
DG
867}
868
e065084a 869/*
d063d709
DG
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
872 * structure.
ca95a216 873 *
d063d709 874 * Return total size of the buffer pointed by buf.
ca95a216 875 */
5461b305 876static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 877{
f3ed775e 878 int ret, buf_size;
ca95a216 879
f3ed775e 880 buf_size = size;
5461b305 881
ba7f0ae5 882 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 883 if (cmd_ctx->llm == NULL) {
76d7553f 884 PERROR("zmalloc");
5461b305 885 ret = -ENOMEM;
ca95a216
DG
886 goto error;
887 }
888
5461b305
DG
889 /* Copy common data */
890 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 891 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 892
5461b305
DG
893 cmd_ctx->llm->data_size = size;
894 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
895
ca95a216
DG
896 return buf_size;
897
898error:
899 return ret;
900}
901
7a485870 902/*
5eb91c98 903 * Update the kernel poll set of all channel fd available over all tracing
d063d709 904 * session. Add the wakeup pipe at the end of the set.
7a485870 905 */
5eb91c98 906static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 907{
5eb91c98 908 int ret;
7a485870
DG
909 struct ltt_session *session;
910 struct ltt_kernel_channel *channel;
911
5eb91c98 912 DBG("Updating kernel poll set");
7a485870 913
54d01ffb 914 session_lock_list();
b5541356 915 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 916 session_lock(session);
7a485870 917 if (session->kernel_session == NULL) {
54d01ffb 918 session_unlock(session);
7a485870
DG
919 continue;
920 }
7a485870 921
54d01ffb
DG
922 cds_list_for_each_entry(channel,
923 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
924 /* Add channel fd to the kernel poll set */
925 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
926 if (ret < 0) {
54d01ffb 927 session_unlock(session);
5eb91c98
DG
928 goto error;
929 }
930 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 931 }
54d01ffb 932 session_unlock(session);
7a485870 933 }
54d01ffb 934 session_unlock_list();
7a485870 935
5eb91c98 936 return 0;
7a485870
DG
937
938error:
54d01ffb 939 session_unlock_list();
7a485870
DG
940 return -1;
941}
942
943/*
54d01ffb 944 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 945 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 946 *
d063d709 947 * Useful for CPU hotplug feature.
7a485870 948 */
2bdd86d4 949static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
950{
951 int ret = 0;
952 struct ltt_session *session;
173af62f 953 struct ltt_kernel_session *ksess;
7a485870
DG
954 struct ltt_kernel_channel *channel;
955
956 DBG("Updating kernel streams for channel fd %d", fd);
957
54d01ffb 958 session_lock_list();
b5541356 959 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 960 session_lock(session);
7a485870 961 if (session->kernel_session == NULL) {
54d01ffb 962 session_unlock(session);
7a485870
DG
963 continue;
964 }
173af62f 965 ksess = session->kernel_session;
d9800920 966
4a15001e
MD
967 cds_list_for_each_entry(channel,
968 &ksess->channel_list.head, list) {
969 struct lttng_ht_iter iter;
970 struct consumer_socket *socket;
d9800920 971
4a15001e
MD
972 if (channel->fd != fd) {
973 continue;
974 }
975 DBG("Channel found, updating kernel streams");
976 ret = kernel_open_channel_stream(channel);
977 if (ret < 0) {
978 goto error;
979 }
980 /* Update the stream global counter */
981 ksess->stream_count_global += ret;
982
983 /*
984 * Have we already sent fds to the consumer? If yes, it
985 * means that tracing is started so it is safe to send
986 * our updated stream fds.
987 */
988 if (ksess->consumer_fds_sent != 1
989 || ksess->consumer == NULL) {
990 ret = -1;
991 goto error;
992 }
993
994 rcu_read_lock();
995 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
996 &iter.iter, socket, node.node) {
997 pthread_mutex_lock(socket->lock);
998 ret = kernel_consumer_send_channel_stream(socket,
999 channel, ksess,
1000 session->output_traces ? 1 : 0);
1001 pthread_mutex_unlock(socket->lock);
1002 if (ret < 0) {
e7fe706f 1003 rcu_read_unlock();
4a15001e 1004 goto error;
7a485870 1005 }
7a485870 1006 }
4a15001e 1007 rcu_read_unlock();
7a485870 1008 }
54d01ffb 1009 session_unlock(session);
7a485870 1010 }
54d01ffb 1011 session_unlock_list();
b3c750d2 1012 return ret;
7a485870 1013
b3c750d2 1014error:
54d01ffb
DG
1015 session_unlock(session);
1016 session_unlock_list();
7a485870
DG
1017 return ret;
1018}
1019
487cf67c 1020/*
ffe60014
DG
1021 * For each tracing session, update newly registered apps. The session list
1022 * lock MUST be acquired before calling this.
487cf67c
DG
1023 */
1024static void update_ust_app(int app_sock)
1025{
1026 struct ltt_session *sess, *stmp;
1027
fdadac08
DG
1028 /* Consumer is in an ERROR state. Stop any application update. */
1029 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1030 /* Stop the update process since the consumer is dead. */
1031 return;
1032 }
1033
487cf67c
DG
1034 /* For all tracing session(s) */
1035 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
a9ad0c8f
MD
1036 struct ust_app *app;
1037
4ee14516 1038 session_lock(sess);
a9ad0c8f
MD
1039 if (!sess->ust_session) {
1040 goto unlock_session;
1041 }
1042
1043 rcu_read_lock();
1044 assert(app_sock >= 0);
1045 app = ust_app_find_by_sock(app_sock);
1046 if (app == NULL) {
1047 /*
1048 * Application can be unregistered before so
1049 * this is possible hence simply stopping the
1050 * update.
1051 */
1052 DBG3("UST app update failed to find app sock %d",
1053 app_sock);
1054 goto unlock_rcu;
421cb601 1055 }
a9ad0c8f
MD
1056 ust_app_global_update(sess->ust_session, app);
1057 unlock_rcu:
1058 rcu_read_unlock();
1059 unlock_session:
4ee14516 1060 session_unlock(sess);
487cf67c
DG
1061 }
1062}
1063
7a485870 1064/*
d063d709 1065 * This thread manage event coming from the kernel.
7a485870 1066 *
d063d709
DG
1067 * Features supported in this thread:
1068 * -) CPU Hotplug
7a485870
DG
1069 */
1070static void *thread_manage_kernel(void *data)
1071{
139ac872 1072 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 1073 uint32_t revents, nb_fd;
7a485870 1074 char tmp;
5eb91c98 1075 struct lttng_poll_event events;
7a485870 1076
6993eeb3 1077 DBG("[thread] Thread manage kernel started");
7a485870 1078
6c71277b 1079 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
927ca06a 1080
d5d63bf1
DG
1081 /*
1082 * This first step of the while is to clean this structure which could free
6d737ce4 1083 * non NULL pointers so initialize it before the loop.
d5d63bf1 1084 */
6d737ce4 1085 lttng_poll_init(&events);
d5d63bf1 1086
e547b070 1087 if (testpoint(sessiond_thread_manage_kernel)) {
6993eeb3
CB
1088 goto error_testpoint;
1089 }
8ac94142 1090
840cb59c 1091 health_code_update();
44a5e5eb 1092
e547b070 1093 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
d21b0d71 1094 goto error_testpoint;
6993eeb3
CB
1095 }
1096
7a485870 1097 while (1) {
840cb59c 1098 health_code_update();
44a5e5eb 1099
7a485870 1100 if (update_poll_flag == 1) {
d21b0d71
DG
1101 /* Clean events object. We are about to populate it again. */
1102 lttng_poll_clean(&events);
1103
d0b96690 1104 ret = sessiond_set_thread_pollset(&events, 2);
d21b0d71
DG
1105 if (ret < 0) {
1106 goto error_poll_create;
1107 }
1108
1109 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1110 if (ret < 0) {
1111 goto error;
1112 }
5f822d0a 1113
d21b0d71 1114 /* This will add the available kernel channel if any. */
5eb91c98
DG
1115 ret = update_kernel_poll(&events);
1116 if (ret < 0) {
7a485870
DG
1117 goto error;
1118 }
1119 update_poll_flag = 0;
1120 }
1121
7fa2082e 1122 DBG("Thread kernel polling");
7a485870
DG
1123
1124 /* Poll infinite value of time */
88f2b785 1125 restart:
a78af745 1126 health_poll_entry();
5eb91c98 1127 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1128 DBG("Thread kernel return from poll on %d fds",
1129 LTTNG_POLL_GETNB(&events));
a78af745 1130 health_poll_exit();
7a485870 1131 if (ret < 0) {
88f2b785
MD
1132 /*
1133 * Restart interrupted system call.
1134 */
1135 if (errno == EINTR) {
1136 goto restart;
1137 }
7a485870
DG
1138 goto error;
1139 } else if (ret == 0) {
1140 /* Should not happen since timeout is infinite */
85611738
DG
1141 ERR("Return value of poll is 0 with an infinite timeout.\n"
1142 "This should not have happened! Continuing...");
7a485870
DG
1143 continue;
1144 }
1145
0d9c5d77
DG
1146 nb_fd = ret;
1147
5eb91c98
DG
1148 for (i = 0; i < nb_fd; i++) {
1149 /* Fetch once the poll data */
1150 revents = LTTNG_POLL_GETEV(&events, i);
1151 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 1152
840cb59c 1153 health_code_update();
44a5e5eb 1154
fd20dac9
MD
1155 if (!revents) {
1156 /* No activity for this FD (poll implementation). */
1157 continue;
1158 }
1159
5eb91c98 1160 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1161 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1162 if (ret) {
139ac872
MD
1163 err = 0;
1164 goto exit;
5eb91c98 1165 }
7a485870 1166
5eb91c98 1167 /* Check for data on kernel pipe */
03e43155
MD
1168 if (revents & LPOLLIN) {
1169 if (pollfd == kernel_poll_pipe[0]) {
1170 (void) lttng_read(kernel_poll_pipe[0],
1171 &tmp, 1);
1172 /*
1173 * Ret value is useless here, if this pipe gets any actions an
1174 * update is required anyway.
1175 */
1176 update_poll_flag = 1;
1177 continue;
1178 } else {
1179 /*
1180 * New CPU detected by the kernel. Adding kernel stream to
1181 * kernel session and updating the kernel consumer
1182 */
2bdd86d4 1183 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
1184 if (ret < 0) {
1185 continue;
1186 }
1187 break;
7a485870 1188 }
03e43155
MD
1189 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1190 update_poll_flag = 1;
1191 continue;
1192 } else {
1193 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1194 goto error;
7a485870
DG
1195 }
1196 }
1197 }
1198
139ac872 1199exit:
7a485870 1200error:
5eb91c98 1201 lttng_poll_clean(&events);
76d7553f 1202error_poll_create:
6993eeb3 1203error_testpoint:
6620da75
DG
1204 utils_close_pipe(kernel_poll_pipe);
1205 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 1206 if (err) {
840cb59c 1207 health_error();
139ac872 1208 ERR("Health error occurred in %s", __func__);
6620da75
DG
1209 WARN("Kernel thread died unexpectedly. "
1210 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 1211 }
8782cc74 1212 health_unregister(health_sessiond);
76d7553f 1213 DBG("Kernel thread dying");
7a485870
DG
1214 return NULL;
1215}
1216
a23ec3a7
DG
1217/*
1218 * Signal pthread condition of the consumer data that the thread.
1219 */
1220static void signal_consumer_condition(struct consumer_data *data, int state)
1221{
1222 pthread_mutex_lock(&data->cond_mutex);
1223
1224 /*
1225 * The state is set before signaling. It can be any value, it's the waiter
1226 * job to correctly interpret this condition variable associated to the
1227 * consumer pthread_cond.
1228 *
1229 * A value of 0 means that the corresponding thread of the consumer data
1230 * was not started. 1 indicates that the thread has started and is ready
1231 * for action. A negative value means that there was an error during the
1232 * thread bootstrap.
1233 */
1234 data->consumer_thread_is_ready = state;
1235 (void) pthread_cond_signal(&data->cond);
1236
1237 pthread_mutex_unlock(&data->cond_mutex);
1238}
1239
1d4b027a 1240/*
3bd1e081 1241 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 1242 */
3bd1e081 1243static void *thread_manage_consumer(void *data)
1d4b027a 1244{
42fc1d0b 1245 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
5eb91c98 1246 uint32_t revents, nb_fd;
1d4b027a 1247 enum lttcomm_return_code code;
5eb91c98 1248 struct lttng_poll_event events;
3bd1e081 1249 struct consumer_data *consumer_data = data;
1d4b027a 1250
3bd1e081 1251 DBG("[thread] Manage consumer started");
1d4b027a 1252
34c1e15a
MD
1253 rcu_register_thread();
1254 rcu_thread_online();
1255
6c71277b 1256 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
927ca06a 1257
855060f8 1258 health_code_update();
9449cc75 1259
5eb91c98 1260 /*
331744e3
JD
1261 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1262 * metadata_sock. Nothing more will be added to this poll set.
5eb91c98 1263 */
331744e3 1264 ret = sessiond_set_thread_pollset(&events, 3);
5eb91c98 1265 if (ret < 0) {
76d7553f 1266 goto error_poll;
5eb91c98 1267 }
273ea72c 1268
edb8b045
DG
1269 /*
1270 * The error socket here is already in a listening state which was done
1271 * just before spawning this thread to avoid a race between the consumer
1272 * daemon exec trying to connect and the listen() call.
1273 */
3bd1e081 1274 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
1275 if (ret < 0) {
1276 goto error;
1277 }
1278
840cb59c 1279 health_code_update();
44a5e5eb 1280
331744e3 1281 /* Infinite blocking call, waiting for transmission */
88f2b785 1282restart:
a78af745 1283 health_poll_entry();
8ac94142 1284
e547b070 1285 if (testpoint(sessiond_thread_manage_consumer)) {
6993eeb3
CB
1286 goto error;
1287 }
8ac94142 1288
5eb91c98 1289 ret = lttng_poll_wait(&events, -1);
a78af745 1290 health_poll_exit();
273ea72c 1291 if (ret < 0) {
88f2b785
MD
1292 /*
1293 * Restart interrupted system call.
1294 */
1295 if (errno == EINTR) {
1296 goto restart;
1297 }
273ea72c
DG
1298 goto error;
1299 }
1300
0d9c5d77
DG
1301 nb_fd = ret;
1302
5eb91c98
DG
1303 for (i = 0; i < nb_fd; i++) {
1304 /* Fetch once the poll data */
1305 revents = LTTNG_POLL_GETEV(&events, i);
1306 pollfd = LTTNG_POLL_GETFD(&events, i);
1307
840cb59c 1308 health_code_update();
44a5e5eb 1309
fd20dac9
MD
1310 if (!revents) {
1311 /* No activity for this FD (poll implementation). */
1312 continue;
1313 }
1314
5eb91c98 1315 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1316 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1317 if (ret) {
139ac872
MD
1318 err = 0;
1319 goto exit;
5eb91c98
DG
1320 }
1321
1322 /* Event on the registration socket */
3bd1e081 1323 if (pollfd == consumer_data->err_sock) {
03e43155
MD
1324 if (revents & LPOLLIN) {
1325 continue;
1326 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1327 ERR("consumer err socket poll error");
5eb91c98 1328 goto error;
03e43155
MD
1329 } else {
1330 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1331 goto error;
5eb91c98
DG
1332 }
1333 }
273ea72c
DG
1334 }
1335
3bd1e081 1336 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
1337 if (sock < 0) {
1338 goto error;
1339 }
1340
b662582b
DG
1341 /*
1342 * Set the CLOEXEC flag. Return code is useless because either way, the
1343 * show must go on.
1344 */
1345 (void) utils_set_fd_cloexec(sock);
1346
840cb59c 1347 health_code_update();
44a5e5eb 1348
3bd1e081 1349 DBG2("Receiving code from consumer err_sock");
ee0b0061 1350
712ea556 1351 /* Getting status code from kconsumerd */
54d01ffb
DG
1352 ret = lttcomm_recv_unix_sock(sock, &code,
1353 sizeof(enum lttcomm_return_code));
1d4b027a
DG
1354 if (ret <= 0) {
1355 goto error;
1356 }
1357
840cb59c 1358 health_code_update();
f73fabfd 1359 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
331744e3 1360 /* Connect both socket, command and metadata. */
3bd1e081
MD
1361 consumer_data->cmd_sock =
1362 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
4ce514c4 1363 consumer_data->metadata_fd =
331744e3 1364 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
92db7cdc
DG
1365 if (consumer_data->cmd_sock < 0
1366 || consumer_data->metadata_fd < 0) {
331744e3 1367 PERROR("consumer connect cmd socket");
a23ec3a7
DG
1368 /* On error, signal condition and quit. */
1369 signal_consumer_condition(consumer_data, -1);
1d4b027a
DG
1370 goto error;
1371 }
9363801e 1372 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
331744e3
JD
1373 /* Create metadata socket lock. */
1374 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1375 if (consumer_data->metadata_sock.lock == NULL) {
1376 PERROR("zmalloc pthread mutex");
1377 ret = -1;
1378 goto error;
1379 }
1380 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1381
a23ec3a7 1382 signal_consumer_condition(consumer_data, 1);
331744e3
JD
1383 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1384 DBG("Consumer metadata socket ready (fd: %d)",
4ce514c4 1385 consumer_data->metadata_fd);
1d4b027a 1386 } else {
3bd1e081 1387 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
1388 lttcomm_get_readable_code(-code));
1389 goto error;
1390 }
1391
331744e3 1392 /* Remove the consumerd error sock since we've established a connexion */
3bd1e081 1393 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1394 if (ret < 0) {
72079cae
DG
1395 goto error;
1396 }
1397
331744e3 1398 /* Add new accepted error socket. */
5eb91c98
DG
1399 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1400 if (ret < 0) {
72079cae 1401 goto error;
5eb91c98
DG
1402 }
1403
331744e3 1404 /* Add metadata socket that is successfully connected. */
4ce514c4 1405 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
331744e3
JD
1406 LPOLLIN | LPOLLRDHUP);
1407 if (ret < 0) {
1408 goto error;
1409 }
1410
840cb59c 1411 health_code_update();
44a5e5eb 1412
331744e3 1413 /* Infinite blocking call, waiting for transmission */
88f2b785 1414restart_poll:
331744e3 1415 while (1) {
42fc1d0b
DG
1416 health_code_update();
1417
1418 /* Exit the thread because the thread quit pipe has been triggered. */
1419 if (should_quit) {
1420 /* Not a health error. */
1421 err = 0;
1422 goto exit;
1423 }
1424
331744e3
JD
1425 health_poll_entry();
1426 ret = lttng_poll_wait(&events, -1);
1427 health_poll_exit();
1428 if (ret < 0) {
1429 /*
1430 * Restart interrupted system call.
1431 */
1432 if (errno == EINTR) {
1433 goto restart_poll;
1434 }
1435 goto error;
88f2b785 1436 }
72079cae 1437
331744e3 1438 nb_fd = ret;
0d9c5d77 1439
331744e3
JD
1440 for (i = 0; i < nb_fd; i++) {
1441 /* Fetch once the poll data */
1442 revents = LTTNG_POLL_GETEV(&events, i);
1443 pollfd = LTTNG_POLL_GETFD(&events, i);
5eb91c98 1444
331744e3 1445 health_code_update();
44a5e5eb 1446
fd20dac9
MD
1447 if (!revents) {
1448 /* No activity for this FD (poll implementation). */
1449 continue;
1450 }
1451
42fc1d0b
DG
1452 /*
1453 * Thread quit pipe has been triggered, flag that we should stop
1454 * but continue the current loop to handle potential data from
1455 * consumer.
1456 */
1457 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1458
331744e3
JD
1459 if (pollfd == sock) {
1460 /* Event on the consumerd socket */
03e43155
MD
1461 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1462 && !(revents & LPOLLIN)) {
331744e3
JD
1463 ERR("consumer err socket second poll error");
1464 goto error;
1465 }
1466 health_code_update();
1467 /* Wait for any kconsumerd error */
1468 ret = lttcomm_recv_unix_sock(sock, &code,
1469 sizeof(enum lttcomm_return_code));
1470 if (ret <= 0) {
1471 ERR("consumer closed the command socket");
1472 goto error;
1473 }
1474
1475 ERR("consumer return code : %s",
1476 lttcomm_get_readable_code(-code));
1477
1478 goto exit;
4ce514c4 1479 } else if (pollfd == consumer_data->metadata_fd) {
03e43155
MD
1480 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1481 && !(revents & LPOLLIN)) {
1482 ERR("consumer err metadata socket second poll error");
1483 goto error;
1484 }
331744e3
JD
1485 /* UST metadata requests */
1486 ret = ust_consumer_metadata_request(
1487 &consumer_data->metadata_sock);
1488 if (ret < 0) {
1489 ERR("Handling metadata request");
1490 goto error;
1491 }
5eb91c98 1492 }
42fc1d0b 1493 /* No need for an else branch all FDs are tested prior. */
5eb91c98 1494 }
331744e3 1495 health_code_update();
5eb91c98
DG
1496 }
1497
139ac872 1498exit:
1d4b027a 1499error:
fdadac08
DG
1500 /*
1501 * We lock here because we are about to close the sockets and some other
92db7cdc
DG
1502 * thread might be using them so get exclusive access which will abort all
1503 * other consumer command by other threads.
fdadac08
DG
1504 */
1505 pthread_mutex_lock(&consumer_data->lock);
1506
5c827ce0
DG
1507 /* Immediately set the consumerd state to stopped */
1508 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1509 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1510 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1511 consumer_data->type == LTTNG_CONSUMER32_UST) {
1512 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1513 } else {
1514 /* Code flow error... */
1515 assert(0);
1516 }
1517
76d7553f
MD
1518 if (consumer_data->err_sock >= 0) {
1519 ret = close(consumer_data->err_sock);
1520 if (ret) {
1521 PERROR("close");
1522 }
a76cbd9f 1523 consumer_data->err_sock = -1;
76d7553f
MD
1524 }
1525 if (consumer_data->cmd_sock >= 0) {
1526 ret = close(consumer_data->cmd_sock);
1527 if (ret) {
1528 PERROR("close");
1529 }
a76cbd9f 1530 consumer_data->cmd_sock = -1;
76d7553f 1531 }
96544455
SS
1532 if (consumer_data->metadata_sock.fd_ptr &&
1533 *consumer_data->metadata_sock.fd_ptr >= 0) {
9363801e 1534 ret = close(*consumer_data->metadata_sock.fd_ptr);
331744e3
JD
1535 if (ret) {
1536 PERROR("close");
1537 }
1538 }
76d7553f
MD
1539 if (sock >= 0) {
1540 ret = close(sock);
1541 if (ret) {
1542 PERROR("close");
1543 }
1544 }
273ea72c 1545
3bd1e081
MD
1546 unlink(consumer_data->err_unix_sock_path);
1547 unlink(consumer_data->cmd_unix_sock_path);
fdadac08 1548 pthread_mutex_unlock(&consumer_data->lock);
92db7cdc 1549
fdadac08 1550 /* Cleanup metadata socket mutex. */
96544455
SS
1551 if (consumer_data->metadata_sock.lock) {
1552 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1553 free(consumer_data->metadata_sock.lock);
1554 }
5eb91c98 1555 lttng_poll_clean(&events);
76d7553f 1556error_poll:
139ac872 1557 if (err) {
840cb59c 1558 health_error();
139ac872
MD
1559 ERR("Health error occurred in %s", __func__);
1560 }
8782cc74 1561 health_unregister(health_sessiond);
76d7553f 1562 DBG("consumer thread cleanup completed");
0177d773 1563
34c1e15a
MD
1564 rcu_thread_offline();
1565 rcu_unregister_thread();
1566
5eb91c98 1567 return NULL;
099e26bd
DG
1568}
1569
099e26bd
DG
1570/*
1571 * This thread manage application communication.
1d4b027a
DG
1572 */
1573static void *thread_manage_apps(void *data)
099e26bd 1574{
139ac872 1575 int i, ret, pollfd, err = -1;
6cd525e8 1576 ssize_t size_ret;
5eb91c98 1577 uint32_t revents, nb_fd;
5eb91c98 1578 struct lttng_poll_event events;
099e26bd
DG
1579
1580 DBG("[thread] Manage application started");
1581
f6a9efaa
DG
1582 rcu_register_thread();
1583 rcu_thread_online();
1584
6c71277b 1585 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
927ca06a 1586
e547b070 1587 if (testpoint(sessiond_thread_manage_apps)) {
6993eeb3
CB
1588 goto error_testpoint;
1589 }
1590
840cb59c 1591 health_code_update();
44a5e5eb 1592
d0b96690 1593 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1594 if (ret < 0) {
76d7553f 1595 goto error_poll_create;
5eb91c98 1596 }
099e26bd 1597
5eb91c98
DG
1598 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1599 if (ret < 0) {
1600 goto error;
1601 }
099e26bd 1602
e547b070 1603 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
6993eeb3
CB
1604 goto error;
1605 }
8ac94142 1606
840cb59c 1607 health_code_update();
44a5e5eb 1608
5eb91c98 1609 while (1) {
7fa2082e 1610 DBG("Apps thread polling");
099e26bd
DG
1611
1612 /* Inifinite blocking call, waiting for transmission */
88f2b785 1613 restart:
a78af745 1614 health_poll_entry();
5eb91c98 1615 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1616 DBG("Apps thread return from poll on %d fds",
1617 LTTNG_POLL_GETNB(&events));
a78af745 1618 health_poll_exit();
099e26bd 1619 if (ret < 0) {
88f2b785
MD
1620 /*
1621 * Restart interrupted system call.
1622 */
1623 if (errno == EINTR) {
1624 goto restart;
1625 }
099e26bd
DG
1626 goto error;
1627 }
1628
0d9c5d77
DG
1629 nb_fd = ret;
1630
5eb91c98
DG
1631 for (i = 0; i < nb_fd; i++) {
1632 /* Fetch once the poll data */
1633 revents = LTTNG_POLL_GETEV(&events, i);
1634 pollfd = LTTNG_POLL_GETFD(&events, i);
1635
840cb59c 1636 health_code_update();
44a5e5eb 1637
fd20dac9
MD
1638 if (!revents) {
1639 /* No activity for this FD (poll implementation). */
1640 continue;
1641 }
1642
5eb91c98 1643 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1644 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1645 if (ret) {
139ac872
MD
1646 err = 0;
1647 goto exit;
5eb91c98 1648 }
099e26bd 1649
5eb91c98
DG
1650 /* Inspect the apps cmd pipe */
1651 if (pollfd == apps_cmd_pipe[0]) {
03e43155 1652 if (revents & LPOLLIN) {
d0b96690
DG
1653 int sock;
1654
5eb91c98 1655 /* Empty pipe */
6cd525e8
MD
1656 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1657 if (size_ret < sizeof(sock)) {
76d7553f 1658 PERROR("read apps cmd pipe");
5eb91c98
DG
1659 goto error;
1660 }
099e26bd 1661
840cb59c 1662 health_code_update();
44a5e5eb 1663
ffe60014 1664 /*
03e43155
MD
1665 * Since this is a command socket (write then read),
1666 * we only monitor the error events of the socket.
ffe60014 1667 */
d0b96690
DG
1668 ret = lttng_poll_add(&events, sock,
1669 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1670 if (ret < 0) {
5eb91c98 1671 goto error;
e0c7ec2b 1672 }
acc7b41b 1673
d0b96690 1674 DBG("Apps with sock %d added to poll set", sock);
03e43155
MD
1675 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1676 ERR("Apps command pipe error");
1677 goto error;
1678 } else {
1679 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1680 goto error;
0177d773 1681 }
5eb91c98
DG
1682 } else {
1683 /*
54d01ffb
DG
1684 * At this point, we know that a registered application made
1685 * the event at poll_wait.
5eb91c98
DG
1686 */
1687 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1688 /* Removing from the poll set */
1689 ret = lttng_poll_del(&events, pollfd);
1690 if (ret < 0) {
1691 goto error;
1692 }
099e26bd 1693
b9d9b220 1694 /* Socket closed on remote end. */
56fff090 1695 ust_app_unregister(pollfd);
03e43155
MD
1696 } else {
1697 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1698 goto error;
5eb91c98 1699 }
099e26bd 1700 }
44a5e5eb 1701
840cb59c 1702 health_code_update();
099e26bd 1703 }
099e26bd
DG
1704 }
1705
139ac872 1706exit:
099e26bd 1707error:
5eb91c98 1708 lttng_poll_clean(&events);
76d7553f 1709error_poll_create:
6993eeb3 1710error_testpoint:
6620da75
DG
1711 utils_close_pipe(apps_cmd_pipe);
1712 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1713
1714 /*
1715 * We don't clean the UST app hash table here since already registered
1716 * applications can still be controlled so let them be until the session
1717 * daemon dies or the applications stop.
1718 */
1719
139ac872 1720 if (err) {
840cb59c 1721 health_error();
139ac872
MD
1722 ERR("Health error occurred in %s", __func__);
1723 }
8782cc74 1724 health_unregister(health_sessiond);
76d7553f 1725 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1726 rcu_thread_offline();
1727 rcu_unregister_thread();
099e26bd
DG
1728 return NULL;
1729}
1730
d0b96690 1731/*
d88aee68
DG
1732 * Send a socket to a thread This is called from the dispatch UST registration
1733 * thread once all sockets are set for the application.
d0b96690 1734 *
b85dc84c
DG
1735 * The sock value can be invalid, we don't really care, the thread will handle
1736 * it and make the necessary cleanup if so.
1737 *
d0b96690
DG
1738 * On success, return 0 else a negative value being the errno message of the
1739 * write().
1740 */
d88aee68 1741static int send_socket_to_thread(int fd, int sock)
d0b96690 1742{
6cd525e8 1743 ssize_t ret;
d0b96690 1744
b85dc84c
DG
1745 /*
1746 * It's possible that the FD is set as invalid with -1 concurrently just
1747 * before calling this function being a shutdown state of the thread.
1748 */
1749 if (fd < 0) {
1750 ret = -EBADF;
1751 goto error;
1752 }
d0b96690 1753
6cd525e8
MD
1754 ret = lttng_write(fd, &sock, sizeof(sock));
1755 if (ret < sizeof(sock)) {
d88aee68 1756 PERROR("write apps pipe %d", fd);
d0b96690
DG
1757 if (ret < 0) {
1758 ret = -errno;
1759 }
1760 goto error;
1761 }
1762
1763 /* All good. Don't send back the write positive ret value. */
1764 ret = 0;
1765error:
6cd525e8 1766 return (int) ret;
d0b96690
DG
1767}
1768
f45e313d
DG
1769/*
1770 * Sanitize the wait queue of the dispatch registration thread meaning removing
1771 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1772 * notify socket is never received.
1773 */
1774static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1775{
1776 int ret, nb_fd = 0, i;
1777 unsigned int fd_added = 0;
1778 struct lttng_poll_event events;
1779 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1780
1781 assert(wait_queue);
1782
1783 lttng_poll_init(&events);
1784
1785 /* Just skip everything for an empty queue. */
1786 if (!wait_queue->count) {
1787 goto end;
1788 }
1789
1790 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1791 if (ret < 0) {
1792 goto error_create;
1793 }
1794
1795 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1796 &wait_queue->head, head) {
1797 assert(wait_node->app);
1798 ret = lttng_poll_add(&events, wait_node->app->sock,
1799 LPOLLHUP | LPOLLERR);
1800 if (ret < 0) {
1801 goto error;
1802 }
1803
1804 fd_added = 1;
1805 }
1806
1807 if (!fd_added) {
1808 goto end;
1809 }
1810
1811 /*
1812 * Poll but don't block so we can quickly identify the faulty events and
1813 * clean them afterwards from the wait queue.
1814 */
1815 ret = lttng_poll_wait(&events, 0);
1816 if (ret < 0) {
1817 goto error;
1818 }
1819 nb_fd = ret;
1820
1821 for (i = 0; i < nb_fd; i++) {
1822 /* Get faulty FD. */
1823 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1824 int pollfd = LTTNG_POLL_GETFD(&events, i);
1825
fd20dac9
MD
1826 if (!revents) {
1827 /* No activity for this FD (poll implementation). */
1828 continue;
1829 }
1830
f45e313d
DG
1831 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1832 &wait_queue->head, head) {
1833 if (pollfd == wait_node->app->sock &&
1834 (revents & (LPOLLHUP | LPOLLERR))) {
1835 cds_list_del(&wait_node->head);
1836 wait_queue->count--;
1837 ust_app_destroy(wait_node->app);
1838 free(wait_node);
48b40bcf
JG
1839 /*
1840 * Silence warning of use-after-free in
1841 * cds_list_for_each_entry_safe which uses
1842 * __typeof__(*wait_node).
1843 */
1844 wait_node = NULL;
f45e313d 1845 break;
03e43155
MD
1846 } else {
1847 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1848 goto error;
f45e313d
DG
1849 }
1850 }
1851 }
1852
1853 if (nb_fd > 0) {
1854 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1855 }
1856
1857end:
1858 lttng_poll_clean(&events);
1859 return;
1860
1861error:
1862 lttng_poll_clean(&events);
1863error_create:
1864 ERR("Unable to sanitize wait queue");
1865 return;
1866}
1867
099e26bd
DG
1868/*
1869 * Dispatch request from the registration threads to the application
1870 * communication thread.
1871 */
1872static void *thread_dispatch_ust_registration(void *data)
1873{
12e2b881 1874 int ret, err = -1;
8bdee6e2 1875 struct cds_wfcq_node *node;
099e26bd 1876 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1877 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1878 struct ust_reg_wait_queue wait_queue = {
1879 .count = 0,
1880 };
d0b96690 1881
6c71277b 1882 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
12e2b881 1883
9ad42ec1
MD
1884 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1885 goto error_testpoint;
1886 }
1887
12e2b881
MD
1888 health_code_update();
1889
f45e313d 1890 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1891
1892 DBG("[thread] Dispatch UST command started");
1893
26c9d55e 1894 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1895 health_code_update();
1896
099e26bd
DG
1897 /* Atomically prepare the queue futex */
1898 futex_nto1_prepare(&ust_cmd_queue.futex);
1899
1900 do {
d0b96690 1901 struct ust_app *app = NULL;
7972aab2 1902 ust_cmd = NULL;
d0b96690 1903
f45e313d
DG
1904 /*
1905 * Make sure we don't have node(s) that have hung up before receiving
1906 * the notify socket. This is to clean the list in order to avoid
1907 * memory leaks from notify socket that are never seen.
1908 */
1909 sanitize_wait_queue(&wait_queue);
1910
12e2b881 1911 health_code_update();
099e26bd 1912 /* Dequeue command for registration */
8bdee6e2 1913 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
099e26bd 1914 if (node == NULL) {
00a17c97 1915 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1916 /* Continue thread execution */
1917 break;
1918 }
1919
1920 ust_cmd = caa_container_of(node, struct ust_command, node);
1921
2f50c8a3
DG
1922 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1923 " gid:%d sock:%d name:%s (version %d.%d)",
1924 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1925 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1926 ust_cmd->sock, ust_cmd->reg_msg.name,
1927 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1928
1929 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1930 wait_node = zmalloc(sizeof(*wait_node));
1931 if (!wait_node) {
1932 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1933 ret = close(ust_cmd->sock);
1934 if (ret < 0) {
1935 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1936 }
51dec90d 1937 lttng_fd_put(LTTNG_FD_APPS, 1);
7972aab2 1938 free(ust_cmd);
d0b96690
DG
1939 goto error;
1940 }
1941 CDS_INIT_LIST_HEAD(&wait_node->head);
1942
1943 /* Create application object if socket is CMD. */
1944 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1945 ust_cmd->sock);
1946 if (!wait_node->app) {
1947 ret = close(ust_cmd->sock);
1948 if (ret < 0) {
1949 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1950 }
51dec90d 1951 lttng_fd_put(LTTNG_FD_APPS, 1);
d88aee68 1952 free(wait_node);
7972aab2 1953 free(ust_cmd);
d0b96690
DG
1954 continue;
1955 }
1956 /*
1957 * Add application to the wait queue so we can set the notify
1958 * socket before putting this object in the global ht.
1959 */
f45e313d
DG
1960 cds_list_add(&wait_node->head, &wait_queue.head);
1961 wait_queue.count++;
d0b96690 1962
7972aab2 1963 free(ust_cmd);
d0b96690
DG
1964 /*
1965 * We have to continue here since we don't have the notify
1966 * socket and the application MUST be added to the hash table
1967 * only at that moment.
1968 */
1969 continue;
1970 } else {
1971 /*
1972 * Look for the application in the local wait queue and set the
1973 * notify socket if found.
1974 */
d88aee68 1975 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1976 &wait_queue.head, head) {
12e2b881 1977 health_code_update();
d0b96690
DG
1978 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1979 wait_node->app->notify_sock = ust_cmd->sock;
1980 cds_list_del(&wait_node->head);
f45e313d 1981 wait_queue.count--;
d0b96690
DG
1982 app = wait_node->app;
1983 free(wait_node);
1984 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1985 break;
1986 }
1987 }
020d7f60
DG
1988
1989 /*
1990 * With no application at this stage the received socket is
1991 * basically useless so close it before we free the cmd data
1992 * structure for good.
1993 */
1994 if (!app) {
1995 ret = close(ust_cmd->sock);
1996 if (ret < 0) {
1997 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1998 }
51dec90d 1999 lttng_fd_put(LTTNG_FD_APPS, 1);
020d7f60 2000 }
7972aab2 2001 free(ust_cmd);
d0b96690
DG
2002 }
2003
2004 if (app) {
d0b96690
DG
2005 /*
2006 * @session_lock_list
2007 *
2008 * Lock the global session list so from the register up to the
2009 * registration done message, no thread can see the application
2010 * and change its state.
2011 */
2012 session_lock_list();
2013 rcu_read_lock();
d88aee68 2014
d0b96690
DG
2015 /*
2016 * Add application to the global hash table. This needs to be
2017 * done before the update to the UST registry can locate the
2018 * application.
2019 */
2020 ust_app_add(app);
d88aee68
DG
2021
2022 /* Set app version. This call will print an error if needed. */
2023 (void) ust_app_version(app);
2024
2025 /* Send notify socket through the notify pipe. */
2026 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2027 app->notify_sock);
2028 if (ret < 0) {
2029 rcu_read_unlock();
2030 session_unlock_list();
b85dc84c
DG
2031 /*
2032 * No notify thread, stop the UST tracing. However, this is
2033 * not an internal error of the this thread thus setting
2034 * the health error code to a normal exit.
2035 */
2036 err = 0;
d88aee68 2037 goto error;
6620da75 2038 }
d88aee68 2039
d0b96690
DG
2040 /*
2041 * Update newly registered application with the tracing
2042 * registry info already enabled information.
2043 */
2044 update_ust_app(app->sock);
d88aee68
DG
2045
2046 /*
2047 * Don't care about return value. Let the manage apps threads
2048 * handle app unregistration upon socket close.
2049 */
fb45065e 2050 (void) ust_app_register_done(app);
d88aee68
DG
2051
2052 /*
2053 * Even if the application socket has been closed, send the app
2054 * to the thread and unregistration will take place at that
2055 * place.
2056 */
2057 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 2058 if (ret < 0) {
d88aee68
DG
2059 rcu_read_unlock();
2060 session_unlock_list();
b85dc84c
DG
2061 /*
2062 * No apps. thread, stop the UST tracing. However, this is
2063 * not an internal error of the this thread thus setting
2064 * the health error code to a normal exit.
2065 */
2066 err = 0;
d88aee68 2067 goto error;
d0b96690 2068 }
d88aee68 2069
d0b96690
DG
2070 rcu_read_unlock();
2071 session_unlock_list();
099e26bd 2072 }
099e26bd
DG
2073 } while (node != NULL);
2074
12e2b881 2075 health_poll_entry();
099e26bd
DG
2076 /* Futex wait on queue. Blocking call on futex() */
2077 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 2078 health_poll_exit();
099e26bd 2079 }
12e2b881
MD
2080 /* Normal exit, no error */
2081 err = 0;
099e26bd
DG
2082
2083error:
d88aee68
DG
2084 /* Clean up wait queue. */
2085 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 2086 &wait_queue.head, head) {
d88aee68 2087 cds_list_del(&wait_node->head);
f45e313d 2088 wait_queue.count--;
d88aee68
DG
2089 free(wait_node);
2090 }
2091
772b8f4d
MD
2092 /* Empty command queue. */
2093 for (;;) {
2094 /* Dequeue command for registration */
2095 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2096 if (node == NULL) {
2097 break;
2098 }
2099 ust_cmd = caa_container_of(node, struct ust_command, node);
2100 ret = close(ust_cmd->sock);
2101 if (ret < 0) {
2102 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2103 }
2104 lttng_fd_put(LTTNG_FD_APPS, 1);
2105 free(ust_cmd);
2106 }
2107
9ad42ec1 2108error_testpoint:
099e26bd 2109 DBG("Dispatch thread dying");
12e2b881
MD
2110 if (err) {
2111 health_error();
2112 ERR("Health error occurred in %s", __func__);
2113 }
8782cc74 2114 health_unregister(health_sessiond);
099e26bd
DG
2115 return NULL;
2116}
2117
2118/*
2119 * This thread manage application registration.
2120 */
2121static void *thread_registration_apps(void *data)
1d4b027a 2122{
139ac872 2123 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
2124 uint32_t revents, nb_fd;
2125 struct lttng_poll_event events;
099e26bd
DG
2126 /*
2127 * Get allocated in this thread, enqueued to a global queue, dequeued and
2128 * freed in the manage apps thread.
2129 */
2130 struct ust_command *ust_cmd = NULL;
1d4b027a 2131
099e26bd 2132 DBG("[thread] Manage application registration started");
1d4b027a 2133
6c71277b 2134 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
927ca06a 2135
e547b070 2136 if (testpoint(sessiond_thread_registration_apps)) {
6993eeb3
CB
2137 goto error_testpoint;
2138 }
8ac94142 2139
1d4b027a
DG
2140 ret = lttcomm_listen_unix_sock(apps_sock);
2141 if (ret < 0) {
76d7553f 2142 goto error_listen;
1d4b027a
DG
2143 }
2144
5eb91c98
DG
2145 /*
2146 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2147 * more will be added to this poll set.
2148 */
d0b96690 2149 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 2150 if (ret < 0) {
76d7553f 2151 goto error_create_poll;
5eb91c98 2152 }
273ea72c 2153
5eb91c98
DG
2154 /* Add the application registration socket */
2155 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2156 if (ret < 0) {
76d7553f 2157 goto error_poll_add;
5eb91c98 2158 }
273ea72c 2159
1d4b027a 2160 /* Notify all applications to register */
0fdd1e2c
DG
2161 ret = notify_ust_apps(1);
2162 if (ret < 0) {
2163 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
2164 "Execution continues but there might be problem for already\n"
2165 "running applications that wishes to register.");
0fdd1e2c 2166 }
1d4b027a
DG
2167
2168 while (1) {
2169 DBG("Accepting application registration");
273ea72c
DG
2170
2171 /* Inifinite blocking call, waiting for transmission */
88f2b785 2172 restart:
a78af745 2173 health_poll_entry();
5eb91c98 2174 ret = lttng_poll_wait(&events, -1);
a78af745 2175 health_poll_exit();
273ea72c 2176 if (ret < 0) {
88f2b785
MD
2177 /*
2178 * Restart interrupted system call.
2179 */
2180 if (errno == EINTR) {
2181 goto restart;
2182 }
273ea72c
DG
2183 goto error;
2184 }
2185
0d9c5d77
DG
2186 nb_fd = ret;
2187
5eb91c98 2188 for (i = 0; i < nb_fd; i++) {
840cb59c 2189 health_code_update();
139ac872 2190
5eb91c98
DG
2191 /* Fetch once the poll data */
2192 revents = LTTNG_POLL_GETEV(&events, i);
2193 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 2194
fd20dac9
MD
2195 if (!revents) {
2196 /* No activity for this FD (poll implementation). */
2197 continue;
2198 }
2199
5eb91c98 2200 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 2201 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 2202 if (ret) {
139ac872
MD
2203 err = 0;
2204 goto exit;
90014c57 2205 }
1d4b027a 2206
5eb91c98
DG
2207 /* Event on the registration socket */
2208 if (pollfd == apps_sock) {
03e43155 2209 if (revents & LPOLLIN) {
5eb91c98
DG
2210 sock = lttcomm_accept_unix_sock(apps_sock);
2211 if (sock < 0) {
2212 goto error;
2213 }
099e26bd 2214
16c5c8fa
DG
2215 /*
2216 * Set socket timeout for both receiving and ending.
2217 * app_socket_timeout is in seconds, whereas
2218 * lttcomm_setsockopt_rcv_timeout and
2219 * lttcomm_setsockopt_snd_timeout expect msec as
2220 * parameter.
2221 */
2222 (void) lttcomm_setsockopt_rcv_timeout(sock,
2223 app_socket_timeout * 1000);
2224 (void) lttcomm_setsockopt_snd_timeout(sock,
2225 app_socket_timeout * 1000);
2226
b662582b
DG
2227 /*
2228 * Set the CLOEXEC flag. Return code is useless because
2229 * either way, the show must go on.
2230 */
2231 (void) utils_set_fd_cloexec(sock);
2232
5eb91c98 2233 /* Create UST registration command for enqueuing */
ba7f0ae5 2234 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 2235 if (ust_cmd == NULL) {
76d7553f 2236 PERROR("ust command zmalloc");
41ed8e47
MD
2237 ret = close(sock);
2238 if (ret) {
2239 PERROR("close");
2240 }
5eb91c98
DG
2241 goto error;
2242 }
1d4b027a 2243
5eb91c98
DG
2244 /*
2245 * Using message-based transmissions to ensure we don't
2246 * have to deal with partially received messages.
2247 */
4063050c
MD
2248 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2249 if (ret < 0) {
2250 ERR("Exhausted file descriptors allowed for applications.");
2251 free(ust_cmd);
2252 ret = close(sock);
2253 if (ret) {
2254 PERROR("close");
2255 }
2256 sock = -1;
2257 continue;
2258 }
d88aee68 2259
840cb59c 2260 health_code_update();
d0b96690
DG
2261 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2262 if (ret < 0) {
5eb91c98 2263 free(ust_cmd);
d0b96690 2264 /* Close socket of the application. */
76d7553f
MD
2265 ret = close(sock);
2266 if (ret) {
2267 PERROR("close");
2268 }
4063050c 2269 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 2270 sock = -1;
5eb91c98
DG
2271 continue;
2272 }
840cb59c 2273 health_code_update();
099e26bd 2274
5eb91c98 2275 ust_cmd->sock = sock;
34a2494f 2276 sock = -1;
099e26bd 2277
5eb91c98
DG
2278 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2279 " gid:%d sock:%d name:%s (version %d.%d)",
2280 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2281 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2282 ust_cmd->sock, ust_cmd->reg_msg.name,
2283 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 2284
5eb91c98
DG
2285 /*
2286 * Lock free enqueue the registration request. The red pill
54d01ffb 2287 * has been taken! This apps will be part of the *system*.
5eb91c98 2288 */
8bdee6e2 2289 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
5eb91c98
DG
2290
2291 /*
2292 * Wake the registration queue futex. Implicit memory
8bdee6e2 2293 * barrier with the exchange in cds_wfcq_enqueue.
5eb91c98
DG
2294 */
2295 futex_nto1_wake(&ust_cmd_queue.futex);
03e43155
MD
2296 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2297 ERR("Register apps socket poll error");
2298 goto error;
2299 } else {
2300 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2301 goto error;
5eb91c98
DG
2302 }
2303 }
90014c57 2304 }
1d4b027a
DG
2305 }
2306
139ac872 2307exit:
1d4b027a 2308error:
0fdd1e2c
DG
2309 /* Notify that the registration thread is gone */
2310 notify_ust_apps(0);
2311
a4b35e07 2312 if (apps_sock >= 0) {
76d7553f
MD
2313 ret = close(apps_sock);
2314 if (ret) {
2315 PERROR("close");
2316 }
a4b35e07 2317 }
46c3f085 2318 if (sock >= 0) {
76d7553f
MD
2319 ret = close(sock);
2320 if (ret) {
2321 PERROR("close");
2322 }
4063050c 2323 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 2324 }
273ea72c 2325 unlink(apps_unix_sock_path);
0fdd1e2c 2326
76d7553f 2327error_poll_add:
5eb91c98 2328 lttng_poll_clean(&events);
76d7553f
MD
2329error_listen:
2330error_create_poll:
6993eeb3 2331error_testpoint:
76d7553f 2332 DBG("UST Registration thread cleanup complete");
9ad42ec1
MD
2333 if (err) {
2334 health_error();
2335 ERR("Health error occurred in %s", __func__);
2336 }
8782cc74 2337 health_unregister(health_sessiond);
5eb91c98 2338
1d4b027a
DG
2339 return NULL;
2340}
2341
8c0faa1d 2342/*
3bd1e081 2343 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 2344 * exec or it will fails.
8c0faa1d 2345 */
3bd1e081 2346static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 2347{
a23ec3a7 2348 int ret, clock_ret;
ee0b0061
DG
2349 struct timespec timeout;
2350
a23ec3a7
DG
2351 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2352 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 2353
a23ec3a7
DG
2354 /* Setup pthread condition */
2355 ret = pthread_condattr_init(&consumer_data->condattr);
4a15001e 2356 if (ret) {
a23ec3a7
DG
2357 errno = ret;
2358 PERROR("pthread_condattr_init consumer data");
2359 goto error;
2360 }
2361
2362 /*
2363 * Set the monotonic clock in order to make sure we DO NOT jump in time
2364 * between the clock_gettime() call and the timedwait call. See bug #324
2365 * for a more details and how we noticed it.
2366 */
2367 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
4a15001e 2368 if (ret) {
a23ec3a7
DG
2369 errno = ret;
2370 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
2371 goto error;
2372 }
8c0faa1d 2373
a23ec3a7 2374 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
4a15001e 2375 if (ret) {
a23ec3a7
DG
2376 errno = ret;
2377 PERROR("pthread_cond_init consumer data");
2378 goto error;
2379 }
2380
2381 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2382 consumer_data);
4a15001e
MD
2383 if (ret) {
2384 errno = ret;
3bd1e081 2385 PERROR("pthread_create consumer");
ee0b0061 2386 ret = -1;
8c0faa1d
DG
2387 goto error;
2388 }
2389
a23ec3a7
DG
2390 /* We are about to wait on a pthread condition */
2391 pthread_mutex_lock(&consumer_data->cond_mutex);
2392
ee0b0061 2393 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
2394 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2395 /*
2396 * Set the timeout for the condition timed wait even if the clock gettime
2397 * call fails since we might loop on that call and we want to avoid to
2398 * increment the timeout too many times.
2399 */
2400 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2401
2402 /*
2403 * The following loop COULD be skipped in some conditions so this is why we
2404 * set ret to 0 in order to make sure at least one round of the loop is
2405 * done.
2406 */
2407 ret = 0;
2408
2409 /*
2410 * Loop until the condition is reached or when a timeout is reached. Note
2411 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2412 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2413 * possible. This loop does not take any chances and works with both of
2414 * them.
2415 */
2416 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2417 if (clock_ret < 0) {
2418 PERROR("clock_gettime spawn consumer");
2419 /* Infinite wait for the consumerd thread to be ready */
2420 ret = pthread_cond_wait(&consumer_data->cond,
2421 &consumer_data->cond_mutex);
2422 } else {
2423 ret = pthread_cond_timedwait(&consumer_data->cond,
2424 &consumer_data->cond_mutex, &timeout);
2425 }
ee0b0061 2426 }
8c0faa1d 2427
a23ec3a7
DG
2428 /* Release the pthread condition */
2429 pthread_mutex_unlock(&consumer_data->cond_mutex);
2430
2431 if (ret != 0) {
2432 errno = ret;
2433 if (ret == ETIMEDOUT) {
4282f9a3
DG
2434 int pth_ret;
2435
ee0b0061
DG
2436 /*
2437 * Call has timed out so we kill the kconsumerd_thread and return
2438 * an error.
2439 */
a23ec3a7
DG
2440 ERR("Condition timed out. The consumer thread was never ready."
2441 " Killing it");
4282f9a3
DG
2442 pth_ret = pthread_cancel(consumer_data->thread);
2443 if (pth_ret < 0) {
3bd1e081 2444 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
2445 }
2446 } else {
a23ec3a7 2447 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061 2448 }
4282f9a3
DG
2449 /* Caller is expecting a negative value on failure. */
2450 ret = -1;
ee0b0061
DG
2451 goto error;
2452 }
2453
3bd1e081
MD
2454 pthread_mutex_lock(&consumer_data->pid_mutex);
2455 if (consumer_data->pid == 0) {
a23ec3a7 2456 ERR("Consumerd did not start");
3bd1e081 2457 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2458 goto error;
2459 }
3bd1e081 2460 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2461
8c0faa1d
DG
2462 return 0;
2463
2464error:
2465 return ret;
2466}
2467
d9800920 2468/*
3bd1e081 2469 * Join consumer thread
d9800920 2470 */
3bd1e081 2471static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2472{
2473 void *status;
cf3af59e 2474
e8209f6b
DG
2475 /* Consumer pid must be a real one. */
2476 if (consumer_data->pid > 0) {
c617c0c6 2477 int ret;
3bd1e081 2478 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2479 if (ret) {
4a15001e 2480 PERROR("Error killing consumer daemon");
cf3af59e
MD
2481 return ret;
2482 }
3bd1e081 2483 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2484 } else {
2485 return 0;
2486 }
2487}
2488
8c0faa1d 2489/*
3bd1e081 2490 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2491 *
d063d709 2492 * Return pid if successful else -1.
8c0faa1d 2493 */
3bd1e081 2494static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2495{
2496 int ret;
2497 pid_t pid;
94c55f17 2498 const char *consumer_to_use;
53086306 2499 const char *verbosity;
94c55f17 2500 struct stat st;
8c0faa1d 2501
3bd1e081 2502 DBG("Spawning consumerd");
c49dc785 2503
8c0faa1d
DG
2504 pid = fork();
2505 if (pid == 0) {
2506 /*
3bd1e081 2507 * Exec consumerd.
8c0faa1d 2508 */
daee5345 2509 if (opt_verbose_consumer) {
53086306 2510 verbosity = "--verbose";
4421f712 2511 } else if (lttng_opt_quiet) {
53086306 2512 verbosity = "--quiet";
4421f712
DG
2513 } else {
2514 verbosity = "";
53086306 2515 }
4421f712 2516
3bd1e081
MD
2517 switch (consumer_data->type) {
2518 case LTTNG_CONSUMER_KERNEL:
94c55f17 2519 /*
c7704d57
DG
2520 * Find out which consumerd to execute. We will first try the
2521 * 64-bit path, then the sessiond's installation directory, and
2522 * fallback on the 32-bit one,
94c55f17 2523 */
63a799e8
AM
2524 DBG3("Looking for a kernel consumer at these locations:");
2525 DBG3(" 1) %s", consumerd64_bin);
2526 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2527 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2528 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2529 DBG3("Found location #1");
94c55f17 2530 consumer_to_use = consumerd64_bin;
94c55f17 2531 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2532 DBG3("Found location #2");
94c55f17 2533 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2534 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2535 DBG3("Found location #3");
eb1e0bd4 2536 consumer_to_use = consumerd32_bin;
94c55f17 2537 } else {
63a799e8 2538 DBG("Could not find any valid consumerd executable");
4282f9a3 2539 ret = -EINVAL;
94c55f17
AM
2540 break;
2541 }
2542 DBG("Using kernel consumer at: %s", consumer_to_use);
4282f9a3 2543 ret = execl(consumer_to_use,
94c55f17
AM
2544 "lttng-consumerd", verbosity, "-k",
2545 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2546 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2547 "--group", tracing_group_name,
94c55f17 2548 NULL);
3bd1e081 2549 break;
7753dea8
MD
2550 case LTTNG_CONSUMER64_UST:
2551 {
b1e0b6b6 2552 char *tmpnew = NULL;
8f4905da
MD
2553
2554 if (consumerd64_libdir[0] != '\0') {
2555 char *tmp;
2556 size_t tmplen;
2557
e8fa9fb0 2558 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
8f4905da
MD
2559 if (!tmp) {
2560 tmp = "";
2561 }
2562 tmplen = strlen("LD_LIBRARY_PATH=")
2563 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2564 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2565 if (!tmpnew) {
2566 ret = -ENOMEM;
2567 goto error;
2568 }
2569 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2570 strcat(tmpnew, consumerd64_libdir);
2571 if (tmp[0] != '\0') {
2572 strcat(tmpnew, ":");
2573 strcat(tmpnew, tmp);
2574 }
2575 ret = putenv(tmpnew);
2576 if (ret) {
2577 ret = -errno;
c6f76da9 2578 free(tmpnew);
8f4905da
MD
2579 goto error;
2580 }
2581 }
94c55f17 2582 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2583 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2584 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2585 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2586 "--group", tracing_group_name,
7753dea8 2587 NULL);
8f4905da
MD
2588 if (consumerd64_libdir[0] != '\0') {
2589 free(tmpnew);
2590 }
3bd1e081 2591 break;
7753dea8
MD
2592 }
2593 case LTTNG_CONSUMER32_UST:
2594 {
937dde8e 2595 char *tmpnew = NULL;
8f4905da
MD
2596
2597 if (consumerd32_libdir[0] != '\0') {
2598 char *tmp;
2599 size_t tmplen;
2600
e8fa9fb0 2601 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
8f4905da
MD
2602 if (!tmp) {
2603 tmp = "";
2604 }
2605 tmplen = strlen("LD_LIBRARY_PATH=")
2606 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2607 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2608 if (!tmpnew) {
2609 ret = -ENOMEM;
2610 goto error;
2611 }
2612 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2613 strcat(tmpnew, consumerd32_libdir);
2614 if (tmp[0] != '\0') {
2615 strcat(tmpnew, ":");
2616 strcat(tmpnew, tmp);
2617 }
2618 ret = putenv(tmpnew);
2619 if (ret) {
2620 ret = -errno;
c6f76da9 2621 free(tmpnew);
8f4905da
MD
2622 goto error;
2623 }
2624 }
94c55f17 2625 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2626 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2627 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2628 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2629 "--group", tracing_group_name,
7753dea8 2630 NULL);
8f4905da
MD
2631 if (consumerd32_libdir[0] != '\0') {
2632 free(tmpnew);
2633 }
7753dea8
MD
2634 break;
2635 }
3bd1e081 2636 default:
76d7553f 2637 PERROR("unknown consumer type");
3bd1e081
MD
2638 exit(EXIT_FAILURE);
2639 }
8c0faa1d 2640 if (errno != 0) {
4282f9a3 2641 PERROR("Consumer execl()");
8c0faa1d 2642 }
4282f9a3 2643 /* Reaching this point, we got a failure on our execl(). */
8c0faa1d
DG
2644 exit(EXIT_FAILURE);
2645 } else if (pid > 0) {
2646 ret = pid;
8c0faa1d 2647 } else {
76d7553f 2648 PERROR("start consumer fork");
8c0faa1d 2649 ret = -errno;
8c0faa1d 2650 }
8f4905da 2651error:
8c0faa1d
DG
2652 return ret;
2653}
2654
693bd40b 2655/*
3bd1e081 2656 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2657 */
3bd1e081 2658static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2659{
c617c0c6 2660 int ret;
edb8b045
DG
2661
2662 /*
2663 * Set the listen() state on the socket since there is a possible race
2664 * between the exec() of the consumer daemon and this call if place in the
2665 * consumer thread. See bug #366 for more details.
2666 */
2667 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2668 if (ret < 0) {
2669 goto error;
2670 }
693bd40b 2671
3bd1e081
MD
2672 pthread_mutex_lock(&consumer_data->pid_mutex);
2673 if (consumer_data->pid != 0) {
2674 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2675 goto end;
2676 }
693bd40b 2677
3bd1e081 2678 ret = spawn_consumerd(consumer_data);
c49dc785 2679 if (ret < 0) {
3bd1e081
MD
2680 ERR("Spawning consumerd failed");
2681 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2682 goto error;
693bd40b 2683 }
c49dc785 2684
3bd1e081
MD
2685 /* Setting up the consumer_data pid */
2686 consumer_data->pid = ret;
48842b30 2687 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2688 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2689
3bd1e081
MD
2690 DBG2("Spawning consumer control thread");
2691 ret = spawn_consumer_thread(consumer_data);
693bd40b 2692 if (ret < 0) {
3bd1e081 2693 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2694 goto error;
2695 }
2696
c49dc785 2697end:
693bd40b
DG
2698 return 0;
2699
2700error:
331744e3 2701 /* Cleanup already created sockets on error. */
edb8b045 2702 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2703 int err;
2704
edb8b045
DG
2705 err = close(consumer_data->err_sock);
2706 if (err < 0) {
2707 PERROR("close consumer data error socket");
2708 }
2709 }
693bd40b
DG
2710 return ret;
2711}
2712
b73401da 2713/*
096102bd 2714 * Setup necessary data for kernel tracer action.
b73401da 2715 */
096102bd 2716static int init_kernel_tracer(void)
b73401da
DG
2717{
2718 int ret;
b73401da 2719
096102bd
DG
2720 /* Modprobe lttng kernel modules */
2721 ret = modprobe_lttng_control();
b73401da 2722 if (ret < 0) {
b73401da
DG
2723 goto error;
2724 }
2725
096102bd
DG
2726 /* Open debugfs lttng */
2727 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2728 if (kernel_tracer_fd < 0) {
2729 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2730 ret = -1;
2731 goto error_open;
54d01ffb
DG
2732 }
2733
2f77fc4b
DG
2734 /* Validate kernel version */
2735 ret = kernel_validate_version(kernel_tracer_fd);
2736 if (ret < 0) {
2737 goto error_version;
b551a063 2738 }
54d01ffb 2739
2f77fc4b
DG
2740 ret = modprobe_lttng_data();
2741 if (ret < 0) {
2742 goto error_modules;
54d01ffb
DG
2743 }
2744
2f77fc4b
DG
2745 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2746 return 0;
2747
2748error_version:
2749 modprobe_remove_lttng_control();
2750 ret = close(kernel_tracer_fd);
2751 if (ret) {
2752 PERROR("close");
b551a063 2753 }
2f77fc4b 2754 kernel_tracer_fd = -1;
f73fabfd 2755 return LTTNG_ERR_KERN_VERSION;
b551a063 2756
2f77fc4b
DG
2757error_modules:
2758 ret = close(kernel_tracer_fd);
2759 if (ret) {
2760 PERROR("close");
b551a063 2761 }
54d01ffb 2762
2f77fc4b
DG
2763error_open:
2764 modprobe_remove_lttng_control();
54d01ffb
DG
2765
2766error:
2f77fc4b
DG
2767 WARN("No kernel tracer available");
2768 kernel_tracer_fd = -1;
2769 if (!is_root) {
f73fabfd 2770 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2771 } else {
f73fabfd 2772 return LTTNG_ERR_KERN_NA;
2f77fc4b 2773 }
54d01ffb
DG
2774}
2775
2f77fc4b 2776
54d01ffb 2777/*
2f77fc4b
DG
2778 * Copy consumer output from the tracing session to the domain session. The
2779 * function also applies the right modification on a per domain basis for the
2780 * trace files destination directory.
36b588ed
MD
2781 *
2782 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2783 */
2f77fc4b 2784static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2785{
2786 int ret;
2f77fc4b
DG
2787 const char *dir_name;
2788 struct consumer_output *consumer;
2789
2790 assert(session);
2791 assert(session->consumer);
54d01ffb 2792
b551a063
DG
2793 switch (domain) {
2794 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2795 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2796 /*
2797 * XXX: We should audit the session creation and what this function
2798 * does "extra" in order to avoid a destroy since this function is used
2799 * in the domain session creation (kernel and ust) only. Same for UST
2800 * domain.
2801 */
2802 if (session->kernel_session->consumer) {
6addfa37 2803 consumer_output_put(session->kernel_session->consumer);
09a90bcd 2804 }
2f77fc4b
DG
2805 session->kernel_session->consumer =
2806 consumer_copy_output(session->consumer);
2807 /* Ease our life a bit for the next part */
2808 consumer = session->kernel_session->consumer;
2809 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063 2810 break;
f20baf8e 2811 case LTTNG_DOMAIN_JUL:
5cdb6027 2812 case LTTNG_DOMAIN_LOG4J:
0e115563 2813 case LTTNG_DOMAIN_PYTHON:
b551a063 2814 case LTTNG_DOMAIN_UST:
2f77fc4b 2815 DBG3("Copying tracing session consumer output in UST session");
09a90bcd 2816 if (session->ust_session->consumer) {
6addfa37 2817 consumer_output_put(session->ust_session->consumer);
09a90bcd 2818 }
2f77fc4b
DG
2819 session->ust_session->consumer =
2820 consumer_copy_output(session->consumer);
2821 /* Ease our life a bit for the next part */
2822 consumer = session->ust_session->consumer;
2823 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2824 break;
2825 default:
f73fabfd 2826 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2827 goto error;
2828 }
2829
2f77fc4b 2830 /* Append correct directory to subdir */
c30ce0b3
CB
2831 strncat(consumer->subdir, dir_name,
2832 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2833 DBG3("Copy session consumer subdir %s", consumer->subdir);
2834
f73fabfd 2835 ret = LTTNG_OK;
54d01ffb
DG
2836
2837error:
2838 return ret;
2839}
2840
00e2e675 2841/*
2f77fc4b 2842 * Create an UST session and add it to the session ust list.
36b588ed
MD
2843 *
2844 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2845 */
2f77fc4b
DG
2846static int create_ust_session(struct ltt_session *session,
2847 struct lttng_domain *domain)
00e2e675
DG
2848{
2849 int ret;
2f77fc4b 2850 struct ltt_ust_session *lus = NULL;
00e2e675 2851
a4b92340 2852 assert(session);
2f77fc4b
DG
2853 assert(domain);
2854 assert(session->consumer);
a4b92340 2855
2f77fc4b 2856 switch (domain->type) {
f20baf8e 2857 case LTTNG_DOMAIN_JUL:
5cdb6027 2858 case LTTNG_DOMAIN_LOG4J:
0e115563 2859 case LTTNG_DOMAIN_PYTHON:
2f77fc4b
DG
2860 case LTTNG_DOMAIN_UST:
2861 break;
2862 default:
2863 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2864 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2865 goto error;
2866 }
2867
2f77fc4b
DG
2868 DBG("Creating UST session");
2869
dec56f6c 2870 lus = trace_ust_create_session(session->id);
2f77fc4b 2871 if (lus == NULL) {
f73fabfd 2872 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2873 goto error;
2874 }
2875
2f77fc4b
DG
2876 lus->uid = session->uid;
2877 lus->gid = session->gid;
2bba9e53 2878 lus->output_traces = session->output_traces;
27babd3a 2879 lus->snapshot_mode = session->snapshot_mode;
ecc48a90 2880 lus->live_timer_interval = session->live_timer;
2f77fc4b 2881 session->ust_session = lus;
d7ba1388 2882 if (session->shm_path[0]) {
3d071855
MD
2883 strncpy(lus->root_shm_path, session->shm_path,
2884 sizeof(lus->root_shm_path));
2885 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
d7ba1388
MD
2886 strncpy(lus->shm_path, session->shm_path,
2887 sizeof(lus->shm_path));
2888 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2889 strncat(lus->shm_path, "/ust",
2890 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2891 }
2f77fc4b
DG
2892 /* Copy session output to the newly created UST session */
2893 ret = copy_session_consumer(domain->type, session);
f73fabfd 2894 if (ret != LTTNG_OK) {
00e2e675
DG
2895 goto error;
2896 }
2897
f73fabfd 2898 return LTTNG_OK;
00e2e675
DG
2899
2900error:
2f77fc4b
DG
2901 free(lus);
2902 session->ust_session = NULL;
00e2e675
DG
2903 return ret;
2904}
2905
2906/*
2f77fc4b 2907 * Create a kernel tracer session then create the default channel.
00e2e675 2908 */
2f77fc4b 2909static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2910{
2911 int ret;
a4b92340 2912
2f77fc4b 2913 DBG("Creating kernel session");
00e2e675 2914
2f77fc4b
DG
2915 ret = kernel_create_session(session, kernel_tracer_fd);
2916 if (ret < 0) {
f73fabfd 2917 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2918 goto error;
2919 }
2920
2f77fc4b
DG
2921 /* Code flow safety */
2922 assert(session->kernel_session);
2923
2924 /* Copy session output to the newly created Kernel session */
2925 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2926 if (ret != LTTNG_OK) {
a4b92340
DG
2927 goto error;
2928 }
2929
2f77fc4b
DG
2930 /* Create directory(ies) on local filesystem. */
2931 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2932 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2933 ret = run_as_mkdir_recursive(
2934 session->kernel_session->consumer->dst.trace_path,
2935 S_IRWXU | S_IRWXG, session->uid, session->gid);
2936 if (ret < 0) {
df5b86c8 2937 if (errno != EEXIST) {
2f77fc4b 2938 ERR("Trace directory creation error");
00e2e675
DG
2939 goto error;
2940 }
00e2e675 2941 }
2f77fc4b 2942 }
00e2e675 2943
2f77fc4b
DG
2944 session->kernel_session->uid = session->uid;
2945 session->kernel_session->gid = session->gid;
2bba9e53 2946 session->kernel_session->output_traces = session->output_traces;
27babd3a 2947 session->kernel_session->snapshot_mode = session->snapshot_mode;
00e2e675 2948
f73fabfd 2949 return LTTNG_OK;
00e2e675 2950
2f77fc4b
DG
2951error:
2952 trace_kernel_destroy_session(session->kernel_session);
2953 session->kernel_session = NULL;
2954 return ret;
2955}
00e2e675 2956
2f77fc4b
DG
2957/*
2958 * Count number of session permitted by uid/gid.
2959 */
2960static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2961{
2962 unsigned int i = 0;
2963 struct ltt_session *session;
07424f16 2964
2f77fc4b
DG
2965 DBG("Counting number of available session for UID %d GID %d",
2966 uid, gid);
2967 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2968 /*
2f77fc4b 2969 * Only list the sessions the user can control.
00e2e675 2970 */
2f77fc4b
DG
2971 if (!session_access_ok(session, uid, gid)) {
2972 continue;
2973 }
2974 i++;
a4b92340 2975 }
2f77fc4b 2976 return i;
00e2e675
DG
2977}
2978
54d01ffb
DG
2979/*
2980 * Process the command requested by the lttng client within the command
2981 * context structure. This function make sure that the return structure (llm)
2982 * is set and ready for transmission before returning.
2983 *
2984 * Return any error encountered or 0 for success.
53a80697
MD
2985 *
2986 * "sock" is only used for special-case var. len data.
36b588ed
MD
2987 *
2988 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2989 */
53a80697
MD
2990static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2991 int *sock_error)
54d01ffb 2992{
f73fabfd 2993 int ret = LTTNG_OK;
44d3bd01 2994 int need_tracing_session = 1;
2e09ba09 2995 int need_domain;
54d01ffb
DG
2996
2997 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2998
3745d315
MD
2999 assert(!rcu_read_ongoing());
3000
53a80697
MD
3001 *sock_error = 0;
3002
2e09ba09
MD
3003 switch (cmd_ctx->lsm->cmd_type) {
3004 case LTTNG_CREATE_SESSION:
27babd3a 3005 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 3006 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09
MD
3007 case LTTNG_DESTROY_SESSION:
3008 case LTTNG_LIST_SESSIONS:
3009 case LTTNG_LIST_DOMAINS:
3010 case LTTNG_START_TRACE:
3011 case LTTNG_STOP_TRACE:
6d805429 3012 case LTTNG_DATA_PENDING:
da3c9ec1
DG
3013 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3014 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3015 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3016 case LTTNG_SNAPSHOT_RECORD:
fb198a11 3017 case LTTNG_SAVE_SESSION:
d7ba1388 3018 case LTTNG_SET_SESSION_SHM_PATH:
2e09ba09 3019 need_domain = 0;
3aace903 3020 break;
2e09ba09
MD
3021 default:
3022 need_domain = 1;
3023 }
3024
3025 if (opt_no_kernel && need_domain
3026 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 3027 if (!is_root) {
f73fabfd 3028 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 3029 } else {
f73fabfd 3030 ret = LTTNG_ERR_KERN_NA;
531d29f9 3031 }
4fba7219
DG
3032 goto error;
3033 }
3034
8d3113b2
DG
3035 /* Deny register consumer if we already have a spawned consumer. */
3036 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3037 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3038 if (kconsumer_data.pid > 0) {
f73fabfd 3039 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 3040 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
3041 goto error;
3042 }
3043 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3044 }
3045
54d01ffb
DG
3046 /*
3047 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 3048 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
3049 * command.
3050 */
3051 switch(cmd_ctx->lsm->cmd_type) {
3052 case LTTNG_LIST_SESSIONS:
3053 case LTTNG_LIST_TRACEPOINTS:
f37d259d 3054 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
3055 case LTTNG_LIST_DOMAINS:
3056 case LTTNG_LIST_CHANNELS:
3057 case LTTNG_LIST_EVENTS:
834978fd 3058 case LTTNG_LIST_SYSCALLS:
a5dfbb9d 3059 case LTTNG_LIST_TRACKER_PIDS:
54d01ffb
DG
3060 break;
3061 default:
3062 /* Setup lttng message with no payload */
3063 ret = setup_lttng_msg(cmd_ctx, 0);
3064 if (ret < 0) {
3065 /* This label does not try to unlock the session */
3066 goto init_setup_error;
3067 }
3068 }
3069
3070 /* Commands that DO NOT need a session. */
3071 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 3072 case LTTNG_CREATE_SESSION:
27babd3a 3073 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 3074 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09 3075 case LTTNG_CALIBRATE:
54d01ffb
DG
3076 case LTTNG_LIST_SESSIONS:
3077 case LTTNG_LIST_TRACEPOINTS:
834978fd 3078 case LTTNG_LIST_SYSCALLS:
f37d259d 3079 case LTTNG_LIST_TRACEPOINT_FIELDS:
fb198a11 3080 case LTTNG_SAVE_SESSION:
44d3bd01 3081 need_tracing_session = 0;
54d01ffb
DG
3082 break;
3083 default:
3084 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
3085 /*
3086 * We keep the session list lock across _all_ commands
3087 * for now, because the per-session lock does not
3088 * handle teardown properly.
3089 */
74babd95 3090 session_lock_list();
54d01ffb
DG
3091 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3092 if (cmd_ctx->session == NULL) {
bba2d65f 3093 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
3094 goto error;
3095 } else {
3096 /* Acquire lock for the session */
3097 session_lock(cmd_ctx->session);
3098 }
3099 break;
3100 }
b389abbe 3101
5f3ecf22
DG
3102 /*
3103 * Commands that need a valid session but should NOT create one if none
3104 * exists. Instead of creating one and destroying it when the command is
3105 * handled, process that right before so we save some round trip in useless
3106 * code path.
3107 */
3108 switch (cmd_ctx->lsm->cmd_type) {
3109 case LTTNG_DISABLE_CHANNEL:
3110 case LTTNG_DISABLE_EVENT:
5f3ecf22
DG
3111 switch (cmd_ctx->lsm->domain.type) {
3112 case LTTNG_DOMAIN_KERNEL:
3113 if (!cmd_ctx->session->kernel_session) {
3114 ret = LTTNG_ERR_NO_CHANNEL;
3115 goto error;
3116 }
3117 break;