Fix: grab more than one packet for snapshots
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
CommitLineData
826d496d
MD
1/*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
0fdd1e2c 3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
26296c48 4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
fac6795d 5 *
d14d33bf
AM
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
91d76f53 9 *
d14d33bf
AM
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
91d76f53 14 *
d14d33bf
AM
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
fac6795d
DG
18 */
19
20#define _GNU_SOURCE
6c1c0768 21#define _LGPL_SOURCE
fac6795d
DG
22#include <getopt.h>
23#include <grp.h>
24#include <limits.h>
0bb7724a 25#include <paths.h>
fac6795d
DG
26#include <pthread.h>
27#include <signal.h>
28#include <stdio.h>
29#include <stdlib.h>
30#include <string.h>
331744e3 31#include <inttypes.h>
0fdd1e2c 32#include <sys/mman.h>
b73401da 33#include <sys/mount.h>
1e307fab 34#include <sys/resource.h>
fac6795d
DG
35#include <sys/socket.h>
36#include <sys/stat.h>
37#include <sys/types.h>
0fdd1e2c 38#include <sys/wait.h>
5c827ce0 39#include <urcu/uatomic.h>
fac6795d 40#include <unistd.h>
3bd1e081 41#include <config.h>
fac6795d 42
990570ed 43#include <common/common.h>
d27c42b8 44#include <common/compat/socket.h>
db758600
DG
45#include <common/defaults.h>
46#include <common/kernel-consumer/kernel-consumer.h>
50c8f484 47#include <common/futex.h>
00e2e675 48#include <common/relayd/relayd.h>
81b86775 49#include <common/utils.h>
3ccdf997 50#include <common/daemonize.h>
26296c48 51#include <common/config/config.h>
fac6795d 52
10a8a223 53#include "lttng-sessiond.h"
7972aab2 54#include "buffer-registry.h"
54d01ffb 55#include "channel.h"
2f77fc4b 56#include "cmd.h"
00e2e675 57#include "consumer.h"
099e26bd 58#include "context.h"
54d01ffb 59#include "event.h"
4771f025 60#include "kernel.h"
f1e16794 61#include "kernel-consumer.h"
096102bd 62#include "modprobe.h"
0fdd1e2c 63#include "shm.h"
1e307fab 64#include "ust-ctl.h"
00e2e675 65#include "ust-consumer.h"
8e68d1c8 66#include "utils.h"
4063050c 67#include "fd-limit.h"
8782cc74 68#include "health-sessiond.h"
8ac94142 69#include "testpoint.h"
d0b96690 70#include "ust-thread.h"
022d91ba 71#include "agent-thread.h"
fb198a11 72#include "save.h"
ef367a93 73#include "load-session-thread.h"
834978fd 74#include "syscall.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[] = {
135 { "client-sock", 1, 0, 'c' },
136 { "apps-sock", 1, 0, 'a' },
137 { "kconsumerd-cmd-sock", 1, 0, 'C' },
138 { "kconsumerd-err-sock", 1, 0, 'E' },
139 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
140 { "ustconsumerd32-err-sock", 1, 0, 'H' },
141 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
142 { "ustconsumerd64-err-sock", 1, 0, 'F' },
143 { "consumerd32-path", 1, 0, 'u' },
144 { "consumerd32-libdir", 1, 0, 'U' },
145 { "consumerd64-path", 1, 0, 't' },
146 { "consumerd64-libdir", 1, 0, 'T' },
147 { "daemonize", 0, 0, 'd' },
72dd7491 148 { "background", 0, 0, 'b' },
26296c48
JG
149 { "sig-parent", 0, 0, 'S' },
150 { "help", 0, 0, 'h' },
151 { "group", 1, 0, 'g' },
152 { "version", 0, 0, 'V' },
153 { "quiet", 0, 0, 'q' },
154 { "verbose", 0, 0, 'v' },
155 { "verbose-consumer", 0, 0, 'Z' },
156 { "no-kernel", 0, 0, 'N' },
157 { "pidfile", 1, 0, 'p' },
1b2ef7fe 158 { "agent-tcp-port", 1, 0, 'J' },
26296c48 159 { "config", 1, 0, 'f' },
ef367a93 160 { "load", 1, 0, 'l' },
fbb9748b 161 { "kmod-probes", 1, 0, 'P' },
c9d42407 162 { "extra-kmod-probes", 1, 0, 'e' },
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
97bc1426
MD
307/*
308 * Whether sessiond is ready for commands/health check requests.
309 * NR_LTTNG_SESSIOND_READY must match the number of calls to
ef367a93 310 * sessiond_notify_ready().
97bc1426 311 */
ef367a93 312#define NR_LTTNG_SESSIOND_READY 3
97bc1426
MD
313int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
314
315/* Notify parents that we are ready for cmd and health check */
ef367a93
JG
316LTTNG_HIDDEN
317void sessiond_notify_ready(void)
97bc1426
MD
318{
319 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
320 /*
321 * Notify parent pid that we are ready to accept command
322 * for client side. This ppid is the one from the
323 * external process that spawned us.
324 */
325 if (opt_sig_parent) {
326 kill(ppid, SIGUSR1);
327 }
328
329 /*
330 * Notify the parent of the fork() process that we are
331 * ready.
332 */
72dd7491 333 if (opt_daemon || opt_background) {
97bc1426
MD
334 kill(child_ppid, SIGUSR1);
335 }
336 }
337}
338
fb09408a 339static
7753dea8 340void setup_consumerd_path(void)
fb09408a 341{
fc7a59ce 342 const char *bin, *libdir;
fb09408a 343
7753dea8
MD
344 /*
345 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
346 * native architecture size consumer if CONFIG_CONSUMER*_PATH
347 * has not been defined.
7753dea8 348 */
ebaeda94 349#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
350 if (!consumerd32_bin[0]) {
351 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
352 }
353 if (!consumerd32_libdir[0]) {
354 consumerd32_libdir = INSTALL_LIB_PATH;
355 }
356#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
357 if (!consumerd64_bin[0]) {
358 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 359 }
ebaeda94
MD
360 if (!consumerd64_libdir[0]) {
361 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
362 }
363#else
364#error "Unknown bitness"
365#endif
366
fb09408a
MD
367 /*
368 * runtime env. var. overrides the build default.
369 */
fc7a59ce
AM
370 bin = getenv("LTTNG_CONSUMERD32_BIN");
371 if (bin) {
372 consumerd32_bin = bin;
7753dea8 373 }
fc7a59ce
AM
374 bin = getenv("LTTNG_CONSUMERD64_BIN");
375 if (bin) {
376 consumerd64_bin = bin;
ebaeda94 377 }
72f579ee 378 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
379 if (libdir) {
380 consumerd32_libdir = libdir;
381 }
72f579ee 382 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
383 if (libdir) {
384 consumerd64_libdir = libdir;
fb09408a
MD
385 }
386}
387
4a15001e
MD
388static
389int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
390 int *a_pipe)
5eb91c98
DG
391{
392 int ret;
393
d0b96690 394 assert(events);
5eb91c98
DG
395
396 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
397 if (ret < 0) {
398 goto error;
399 }
400
401 /* Add quit pipe */
4a15001e 402 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
5eb91c98
DG
403 if (ret < 0) {
404 goto error;
405 }
406
407 return 0;
408
409error:
410 return ret;
411}
412
4a15001e
MD
413/*
414 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
415 */
416int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
417{
418 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
419}
420
421/*
422 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
423 */
424int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event *events,
425 size_t size)
426{
427 return __sessiond_set_thread_pollset(events, size,
428 ht_cleanup_quit_pipe);
429}
430
431static
432int __sessiond_check_thread_quit_pipe(int fd, uint32_t events, int a_pipe)
433{
434 if (fd == a_pipe && (events & LPOLLIN)) {
435 return 1;
436 }
437 return 0;
438}
439
5eb91c98
DG
440/*
441 * Check if the thread quit pipe was triggered.
442 *
443 * Return 1 if it was triggered else 0;
444 */
d0b96690 445int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
5eb91c98 446{
4a15001e
MD
447 return __sessiond_check_thread_quit_pipe(fd, events,
448 thread_quit_pipe[0]);
449}
5eb91c98 450
4a15001e
MD
451/*
452 * Check if the ht_cleanup thread quit pipe was triggered.
453 *
454 * Return 1 if it was triggered else 0;
455 */
456int sessiond_check_ht_cleanup_quit(int fd, uint32_t events)
457{
458 return __sessiond_check_thread_quit_pipe(fd, events,
459 ht_cleanup_quit_pipe[0]);
5eb91c98
DG
460}
461
273ea72c 462/*
5eb91c98 463 * Init thread quit pipe.
273ea72c
DG
464 *
465 * Return -1 on error or 0 if all pipes are created.
466 */
4a15001e 467static int __init_thread_quit_pipe(int *a_pipe)
273ea72c 468{
730389d9 469 int ret, i;
273ea72c 470
4a15001e 471 ret = pipe(a_pipe);
273ea72c 472 if (ret < 0) {
730389d9 473 PERROR("thread quit pipe");
273ea72c
DG
474 goto error;
475 }
476
730389d9 477 for (i = 0; i < 2; i++) {
4a15001e 478 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
730389d9
DG
479 if (ret < 0) {
480 PERROR("fcntl");
481 goto error;
482 }
483 }
484
273ea72c
DG
485error:
486 return ret;
487}
488
4a15001e
MD
489static int init_thread_quit_pipe(void)
490{
491 return __init_thread_quit_pipe(thread_quit_pipe);
492}
493
494static int init_ht_cleanup_quit_pipe(void)
495{
496 return __init_thread_quit_pipe(ht_cleanup_quit_pipe);
497}
498
099e26bd
DG
499/*
500 * Stop all threads by closing the thread quit pipe.
501 */
cf3af59e
MD
502static void stop_threads(void)
503{
5eb91c98
DG
504 int ret;
505
cf3af59e
MD
506 /* Stopping all threads */
507 DBG("Terminating all threads");
54d01ffb 508 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
509 if (ret < 0) {
510 ERR("write error on thread quit pipe");
511 }
512
099e26bd 513 /* Dispatch thread */
26c9d55e 514 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 515 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
516}
517
e975f9f8
DG
518/*
519 * Close every consumer sockets.
520 */
521static void close_consumer_sockets(void)
522{
523 int ret;
524
525 if (kconsumer_data.err_sock >= 0) {
526 ret = close(kconsumer_data.err_sock);
527 if (ret < 0) {
528 PERROR("kernel consumer err_sock close");
529 }
530 }
531 if (ustconsumer32_data.err_sock >= 0) {
532 ret = close(ustconsumer32_data.err_sock);
533 if (ret < 0) {
a76cbd9f 534 PERROR("UST consumerd32 err_sock close");
e975f9f8
DG
535 }
536 }
537 if (ustconsumer64_data.err_sock >= 0) {
538 ret = close(ustconsumer64_data.err_sock);
539 if (ret < 0) {
a76cbd9f 540 PERROR("UST consumerd64 err_sock close");
e975f9f8
DG
541 }
542 }
543 if (kconsumer_data.cmd_sock >= 0) {
544 ret = close(kconsumer_data.cmd_sock);
545 if (ret < 0) {
546 PERROR("kernel consumer cmd_sock close");
547 }
548 }
549 if (ustconsumer32_data.cmd_sock >= 0) {
550 ret = close(ustconsumer32_data.cmd_sock);
551 if (ret < 0) {
a76cbd9f 552 PERROR("UST consumerd32 cmd_sock close");
e975f9f8
DG
553 }
554 }
555 if (ustconsumer64_data.cmd_sock >= 0) {
556 ret = close(ustconsumer64_data.cmd_sock);
557 if (ret < 0) {
a76cbd9f 558 PERROR("UST consumerd64 cmd_sock close");
e975f9f8
DG
559 }
560 }
561}
562
c9cb3e7d
JG
563/*
564 * Generate the full lock file path using the rundir.
565 *
566 * Return the snprintf() return value thus a negative value is an error.
567 */
568static int generate_lock_file_path(char *path, size_t len)
569{
570 int ret;
571
572 assert(path);
573 assert(rundir);
574
575 /* Build lockfile path from rundir. */
576 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
577 if (ret < 0) {
578 PERROR("snprintf lockfile path");
579 }
580
581 return ret;
582}
583
fac6795d 584/*
4a15001e 585 * Cleanup the session daemon's data structures.
fac6795d 586 */
4a15001e 587static void sessiond_cleanup(void)
fac6795d 588{
ef599319 589 int ret;
af9737e9 590 struct ltt_session *sess, *stmp;
8c6c56c2 591 char path[PATH_MAX];
fac6795d 592
4a15001e 593 DBG("Cleanup sessiond");
e07ae692 594
4e449f3f
MD
595 /*
596 * Close the thread quit pipe. It has already done its job,
597 * since we are now called.
598 */
2f77fc4b
DG
599 utils_close_pipe(thread_quit_pipe);
600
35f90c40
DG
601 /*
602 * If opt_pidfile is undefined, the default file will be wiped when
603 * removing the rundir.
604 */
605 if (opt_pidfile) {
606 ret = remove(opt_pidfile);
607 if (ret < 0) {
608 PERROR("remove pidfile %s", opt_pidfile);
609 }
610 }
611
8c6c56c2
MD
612 DBG("Removing sessiond and consumerd content of directory %s", rundir);
613
614 /* sessiond */
615 snprintf(path, PATH_MAX,
616 "%s/%s",
617 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
618 DBG("Removing %s", path);
619 (void) unlink(path);
620
cd9290dd 621 snprintf(path, PATH_MAX, "%s/%s", rundir,
022d91ba 622 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
cd9290dd
DG
623 DBG("Removing %s", path);
624 (void) unlink(path);
625
8c6c56c2
MD
626 /* kconsumerd */
627 snprintf(path, PATH_MAX,
628 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
629 rundir);
630 DBG("Removing %s", path);
631 (void) unlink(path);
632
633 snprintf(path, PATH_MAX,
634 DEFAULT_KCONSUMERD_PATH,
635 rundir);
636 DBG("Removing directory %s", path);
637 (void) rmdir(path);
638
639 /* ust consumerd 32 */
640 snprintf(path, PATH_MAX,
641 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
642 rundir);
643 DBG("Removing %s", path);
644 (void) unlink(path);
645
646 snprintf(path, PATH_MAX,
647 DEFAULT_USTCONSUMERD32_PATH,
648 rundir);
649 DBG("Removing directory %s", path);
650 (void) rmdir(path);
651
652 /* ust consumerd 64 */
653 snprintf(path, PATH_MAX,
654 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
655 rundir);
656 DBG("Removing %s", path);
657 (void) unlink(path);
658
659 snprintf(path, PATH_MAX,
660 DEFAULT_USTCONSUMERD64_PATH,
661 rundir);
662 DBG("Removing directory %s", path);
663 (void) rmdir(path);
5461b305 664
99bab54f 665 DBG("Cleaning up all sessions");
fac6795d 666
b5541356 667 /* Destroy session list mutex */
273ea72c
DG
668 if (session_list_ptr != NULL) {
669 pthread_mutex_destroy(&session_list_ptr->lock);
670
671 /* Cleanup ALL session */
54d01ffb
DG
672 cds_list_for_each_entry_safe(sess, stmp,
673 &session_list_ptr->head, list) {
2f77fc4b 674 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
675 }
676 }
677
099e26bd 678 DBG("Closing all UST sockets");
56fff090 679 ust_app_clean_list();
7972aab2 680 buffer_reg_destroy_registries();
099e26bd 681
4fba7219
DG
682 if (is_root && !opt_no_kernel) {
683 DBG2("Closing kernel fd");
a4b35e07 684 if (kernel_tracer_fd >= 0) {
76d7553f
MD
685 ret = close(kernel_tracer_fd);
686 if (ret) {
687 PERROR("close");
688 }
a4b35e07 689 }
2f50c8a3 690 DBG("Unloading kernel modules");
096102bd 691 modprobe_remove_lttng_all();
834978fd 692 free(syscall_table);
2f50c8a3 693 }
2f77fc4b 694
e975f9f8
DG
695 close_consumer_sockets();
696
ef367a93
JG
697 if (load_info) {
698 load_session_destroy_data(load_info);
699 free(load_info);
700 }
701
c9cb3e7d
JG
702 /*
703 * Cleanup lock file by deleting it and finaly closing it which will
704 * release the file system lock.
705 */
706 if (lockfile_fd >= 0) {
707 char lockfile_path[PATH_MAX];
708
4a15001e
MD
709 ret = generate_lock_file_path(lockfile_path,
710 sizeof(lockfile_path));
c9cb3e7d
JG
711 if (ret > 0) {
712 ret = remove(lockfile_path);
713 if (ret < 0) {
714 PERROR("remove lock file");
715 }
716 ret = close(lockfile_fd);
717 if (ret < 0) {
718 PERROR("close lock file");
719 }
720 }
721 }
722
723 /*
724 * We do NOT rmdir rundir because there are other processes
725 * using it, for instance lttng-relayd, which can start in
726 * parallel with this teardown.
727 */
728
729 free(rundir);
4a15001e
MD
730}
731
732/*
733 * Cleanup the daemon's option data structures.
734 */
735static void sessiond_cleanup_options(void)
736{
737 DBG("Cleaning up options");
738
739 /*
740 * If the override option is set, the pointer points to a *non* const
741 * thus freeing it even though the variable type is set to const.
742 */
743 if (tracing_group_name_override) {
744 free((void *) tracing_group_name);
745 }
746 if (consumerd32_bin_override) {
747 free((void *) consumerd32_bin);
748 }
749 if (consumerd64_bin_override) {
750 free((void *) consumerd64_bin);
751 }
752 if (consumerd32_libdir_override) {
753 free((void *) consumerd32_libdir);
754 }
755 if (consumerd64_libdir_override) {
756 free((void *) consumerd64_libdir);
757 }
758
759 free(opt_pidfile);
760 free(opt_load_session_path);
761 free(kmod_probes_list);
762 free(kmod_extra_probes_list);
c9cb3e7d 763
421cb601 764 /* <fun> */
f56a39af 765 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
766 "Matthew, BEET driven development works!%c[%dm",
767 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
768 /* </fun> */
fac6795d
DG
769}
770
e065084a 771/*
d063d709 772 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 773 *
d063d709 774 * Return lttcomm error code.
e065084a
DG
775 */
776static int send_unix_sock(int sock, void *buf, size_t len)
777{
778 /* Check valid length */
c617c0c6 779 if (len == 0) {
e065084a
DG
780 return -1;
781 }
782
783 return lttcomm_send_unix_sock(sock, buf, len);
784}
785
5461b305 786/*
d063d709 787 * Free memory of a command context structure.
5461b305 788 */
a2fb29a5 789static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 790{
a2fb29a5
DG
791 DBG("Clean command context structure");
792 if (*cmd_ctx) {
793 if ((*cmd_ctx)->llm) {
794 free((*cmd_ctx)->llm);
5461b305 795 }
a2fb29a5
DG
796 if ((*cmd_ctx)->lsm) {
797 free((*cmd_ctx)->lsm);
5461b305 798 }
a2fb29a5
DG
799 free(*cmd_ctx);
800 *cmd_ctx = NULL;
5461b305
DG
801 }
802}
803
fac6795d 804/*
0fdd1e2c 805 * Notify UST applications using the shm mmap futex.
fac6795d 806 */
0fdd1e2c 807static int notify_ust_apps(int active)
fac6795d 808{
0fdd1e2c 809 char *wait_shm_mmap;
fac6795d 810
0fdd1e2c 811 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 812
0fdd1e2c
DG
813 /* See shm.c for this call implying mmap, shm and futex calls */
814 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
815 if (wait_shm_mmap == NULL) {
fac6795d
DG
816 goto error;
817 }
818
0fdd1e2c
DG
819 /* Wake waiting process */
820 futex_wait_update((int32_t *) wait_shm_mmap, active);
821
822 /* Apps notified successfully */
823 return 0;
fac6795d
DG
824
825error:
0fdd1e2c 826 return -1;
fac6795d
DG
827}
828
e065084a 829/*
d063d709
DG
830 * Setup the outgoing data buffer for the response (llm) by allocating the
831 * right amount of memory and copying the original information from the lsm
832 * structure.
ca95a216 833 *
d063d709 834 * Return total size of the buffer pointed by buf.
ca95a216 835 */
5461b305 836static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 837{
f3ed775e 838 int ret, buf_size;
ca95a216 839
f3ed775e 840 buf_size = size;
5461b305 841
ba7f0ae5 842 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 843 if (cmd_ctx->llm == NULL) {
76d7553f 844 PERROR("zmalloc");
5461b305 845 ret = -ENOMEM;
ca95a216
DG
846 goto error;
847 }
848
5461b305
DG
849 /* Copy common data */
850 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 851 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 852
5461b305
DG
853 cmd_ctx->llm->data_size = size;
854 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
855
ca95a216
DG
856 return buf_size;
857
858error:
859 return ret;
860}
861
7a485870 862/*
5eb91c98 863 * Update the kernel poll set of all channel fd available over all tracing
d063d709 864 * session. Add the wakeup pipe at the end of the set.
7a485870 865 */
5eb91c98 866static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 867{
5eb91c98 868 int ret;
7a485870
DG
869 struct ltt_session *session;
870 struct ltt_kernel_channel *channel;
871
5eb91c98 872 DBG("Updating kernel poll set");
7a485870 873
54d01ffb 874 session_lock_list();
b5541356 875 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 876 session_lock(session);
7a485870 877 if (session->kernel_session == NULL) {
54d01ffb 878 session_unlock(session);
7a485870
DG
879 continue;
880 }
7a485870 881
54d01ffb
DG
882 cds_list_for_each_entry(channel,
883 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
884 /* Add channel fd to the kernel poll set */
885 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
886 if (ret < 0) {
54d01ffb 887 session_unlock(session);
5eb91c98
DG
888 goto error;
889 }
890 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 891 }
54d01ffb 892 session_unlock(session);
7a485870 893 }
54d01ffb 894 session_unlock_list();
7a485870 895
5eb91c98 896 return 0;
7a485870
DG
897
898error:
54d01ffb 899 session_unlock_list();
7a485870
DG
900 return -1;
901}
902
903/*
54d01ffb 904 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 905 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 906 *
d063d709 907 * Useful for CPU hotplug feature.
7a485870 908 */
2bdd86d4 909static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
910{
911 int ret = 0;
912 struct ltt_session *session;
173af62f 913 struct ltt_kernel_session *ksess;
7a485870
DG
914 struct ltt_kernel_channel *channel;
915
916 DBG("Updating kernel streams for channel fd %d", fd);
917
54d01ffb 918 session_lock_list();
b5541356 919 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 920 session_lock(session);
7a485870 921 if (session->kernel_session == NULL) {
54d01ffb 922 session_unlock(session);
7a485870
DG
923 continue;
924 }
173af62f 925 ksess = session->kernel_session;
d9800920 926
4a15001e
MD
927 cds_list_for_each_entry(channel,
928 &ksess->channel_list.head, list) {
929 struct lttng_ht_iter iter;
930 struct consumer_socket *socket;
d9800920 931
4a15001e
MD
932 if (channel->fd != fd) {
933 continue;
934 }
935 DBG("Channel found, updating kernel streams");
936 ret = kernel_open_channel_stream(channel);
937 if (ret < 0) {
938 goto error;
939 }
940 /* Update the stream global counter */
941 ksess->stream_count_global += ret;
942
943 /*
944 * Have we already sent fds to the consumer? If yes, it
945 * means that tracing is started so it is safe to send
946 * our updated stream fds.
947 */
948 if (ksess->consumer_fds_sent != 1
949 || ksess->consumer == NULL) {
950 ret = -1;
951 goto error;
952 }
953
954 rcu_read_lock();
955 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
956 &iter.iter, socket, node.node) {
957 pthread_mutex_lock(socket->lock);
958 ret = kernel_consumer_send_channel_stream(socket,
959 channel, ksess,
960 session->output_traces ? 1 : 0);
961 pthread_mutex_unlock(socket->lock);
962 if (ret < 0) {
e7fe706f 963 rcu_read_unlock();
4a15001e 964 goto error;
7a485870 965 }
7a485870 966 }
4a15001e 967 rcu_read_unlock();
7a485870 968 }
54d01ffb 969 session_unlock(session);
7a485870 970 }
54d01ffb 971 session_unlock_list();
b3c750d2 972 return ret;
7a485870 973
b3c750d2 974error:
54d01ffb
DG
975 session_unlock(session);
976 session_unlock_list();
7a485870
DG
977 return ret;
978}
979
487cf67c 980/*
ffe60014
DG
981 * For each tracing session, update newly registered apps. The session list
982 * lock MUST be acquired before calling this.
487cf67c
DG
983 */
984static void update_ust_app(int app_sock)
985{
986 struct ltt_session *sess, *stmp;
987
fdadac08
DG
988 /* Consumer is in an ERROR state. Stop any application update. */
989 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
990 /* Stop the update process since the consumer is dead. */
991 return;
992 }
993
487cf67c
DG
994 /* For all tracing session(s) */
995 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
4ee14516 996 session_lock(sess);
421cb601
DG
997 if (sess->ust_session) {
998 ust_app_global_update(sess->ust_session, app_sock);
999 }
4ee14516 1000 session_unlock(sess);
487cf67c
DG
1001 }
1002}
1003
7a485870 1004/*
d063d709 1005 * This thread manage event coming from the kernel.
7a485870 1006 *
d063d709
DG
1007 * Features supported in this thread:
1008 * -) CPU Hotplug
7a485870
DG
1009 */
1010static void *thread_manage_kernel(void *data)
1011{
139ac872 1012 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 1013 uint32_t revents, nb_fd;
7a485870 1014 char tmp;
5eb91c98 1015 struct lttng_poll_event events;
7a485870 1016
6993eeb3 1017 DBG("[thread] Thread manage kernel started");
7a485870 1018
6c71277b 1019 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
927ca06a 1020
d5d63bf1
DG
1021 /*
1022 * This first step of the while is to clean this structure which could free
6d737ce4 1023 * non NULL pointers so initialize it before the loop.
d5d63bf1 1024 */
6d737ce4 1025 lttng_poll_init(&events);
d5d63bf1 1026
e547b070 1027 if (testpoint(sessiond_thread_manage_kernel)) {
6993eeb3
CB
1028 goto error_testpoint;
1029 }
8ac94142 1030
840cb59c 1031 health_code_update();
44a5e5eb 1032
e547b070 1033 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
d21b0d71 1034 goto error_testpoint;
6993eeb3
CB
1035 }
1036
7a485870 1037 while (1) {
840cb59c 1038 health_code_update();
44a5e5eb 1039
7a485870 1040 if (update_poll_flag == 1) {
d21b0d71
DG
1041 /* Clean events object. We are about to populate it again. */
1042 lttng_poll_clean(&events);
1043
d0b96690 1044 ret = sessiond_set_thread_pollset(&events, 2);
d21b0d71
DG
1045 if (ret < 0) {
1046 goto error_poll_create;
1047 }
1048
1049 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1050 if (ret < 0) {
1051 goto error;
1052 }
5f822d0a 1053
d21b0d71 1054 /* This will add the available kernel channel if any. */
5eb91c98
DG
1055 ret = update_kernel_poll(&events);
1056 if (ret < 0) {
7a485870
DG
1057 goto error;
1058 }
1059 update_poll_flag = 0;
1060 }
1061
7fa2082e 1062 DBG("Thread kernel polling");
7a485870
DG
1063
1064 /* Poll infinite value of time */
88f2b785 1065 restart:
a78af745 1066 health_poll_entry();
5eb91c98 1067 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1068 DBG("Thread kernel return from poll on %d fds",
1069 LTTNG_POLL_GETNB(&events));
a78af745 1070 health_poll_exit();
7a485870 1071 if (ret < 0) {
88f2b785
MD
1072 /*
1073 * Restart interrupted system call.
1074 */
1075 if (errno == EINTR) {
1076 goto restart;
1077 }
7a485870
DG
1078 goto error;
1079 } else if (ret == 0) {
1080 /* Should not happen since timeout is infinite */
85611738
DG
1081 ERR("Return value of poll is 0 with an infinite timeout.\n"
1082 "This should not have happened! Continuing...");
7a485870
DG
1083 continue;
1084 }
1085
0d9c5d77
DG
1086 nb_fd = ret;
1087
5eb91c98
DG
1088 for (i = 0; i < nb_fd; i++) {
1089 /* Fetch once the poll data */
1090 revents = LTTNG_POLL_GETEV(&events, i);
1091 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 1092
840cb59c 1093 health_code_update();
44a5e5eb 1094
fd20dac9
MD
1095 if (!revents) {
1096 /* No activity for this FD (poll implementation). */
1097 continue;
1098 }
1099
5eb91c98 1100 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1101 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1102 if (ret) {
139ac872
MD
1103 err = 0;
1104 goto exit;
5eb91c98 1105 }
7a485870 1106
5eb91c98
DG
1107 /* Check for data on kernel pipe */
1108 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
6cd525e8
MD
1109 (void) lttng_read(kernel_poll_pipe[0],
1110 &tmp, 1);
f921c78f
DG
1111 /*
1112 * Ret value is useless here, if this pipe gets any actions an
1113 * update is required anyway.
1114 */
5eb91c98
DG
1115 update_poll_flag = 1;
1116 continue;
1117 } else {
1118 /*
1119 * New CPU detected by the kernel. Adding kernel stream to
1120 * kernel session and updating the kernel consumer
1121 */
1122 if (revents & LPOLLIN) {
2bdd86d4 1123 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
1124 if (ret < 0) {
1125 continue;
1126 }
1127 break;
1128 /*
1129 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1130 * and unregister kernel stream at this point.
1131 */
7a485870 1132 }
7a485870
DG
1133 }
1134 }
1135 }
1136
139ac872 1137exit:
7a485870 1138error:
5eb91c98 1139 lttng_poll_clean(&events);
76d7553f 1140error_poll_create:
6993eeb3 1141error_testpoint:
6620da75
DG
1142 utils_close_pipe(kernel_poll_pipe);
1143 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 1144 if (err) {
840cb59c 1145 health_error();
139ac872 1146 ERR("Health error occurred in %s", __func__);
6620da75
DG
1147 WARN("Kernel thread died unexpectedly. "
1148 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 1149 }
8782cc74 1150 health_unregister(health_sessiond);
76d7553f 1151 DBG("Kernel thread dying");
7a485870
DG
1152 return NULL;
1153}
1154
a23ec3a7
DG
1155/*
1156 * Signal pthread condition of the consumer data that the thread.
1157 */
1158static void signal_consumer_condition(struct consumer_data *data, int state)
1159{
1160 pthread_mutex_lock(&data->cond_mutex);
1161
1162 /*
1163 * The state is set before signaling. It can be any value, it's the waiter
1164 * job to correctly interpret this condition variable associated to the
1165 * consumer pthread_cond.
1166 *
1167 * A value of 0 means that the corresponding thread of the consumer data
1168 * was not started. 1 indicates that the thread has started and is ready
1169 * for action. A negative value means that there was an error during the
1170 * thread bootstrap.
1171 */
1172 data->consumer_thread_is_ready = state;
1173 (void) pthread_cond_signal(&data->cond);
1174
1175 pthread_mutex_unlock(&data->cond_mutex);
1176}
1177
1d4b027a 1178/*
3bd1e081 1179 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 1180 */
3bd1e081 1181static void *thread_manage_consumer(void *data)
1d4b027a 1182{
42fc1d0b 1183 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
5eb91c98 1184 uint32_t revents, nb_fd;
1d4b027a 1185 enum lttcomm_return_code code;
5eb91c98 1186 struct lttng_poll_event events;
3bd1e081 1187 struct consumer_data *consumer_data = data;
1d4b027a 1188
3bd1e081 1189 DBG("[thread] Manage consumer started");
1d4b027a 1190
6c71277b 1191 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
927ca06a 1192
855060f8 1193 health_code_update();
9449cc75 1194
5eb91c98 1195 /*
331744e3
JD
1196 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1197 * metadata_sock. Nothing more will be added to this poll set.
5eb91c98 1198 */
331744e3 1199 ret = sessiond_set_thread_pollset(&events, 3);
5eb91c98 1200 if (ret < 0) {
76d7553f 1201 goto error_poll;
5eb91c98 1202 }
273ea72c 1203
edb8b045
DG
1204 /*
1205 * The error socket here is already in a listening state which was done
1206 * just before spawning this thread to avoid a race between the consumer
1207 * daemon exec trying to connect and the listen() call.
1208 */
3bd1e081 1209 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
1210 if (ret < 0) {
1211 goto error;
1212 }
1213
840cb59c 1214 health_code_update();
44a5e5eb 1215
331744e3 1216 /* Infinite blocking call, waiting for transmission */
88f2b785 1217restart:
a78af745 1218 health_poll_entry();
8ac94142 1219
e547b070 1220 if (testpoint(sessiond_thread_manage_consumer)) {
6993eeb3
CB
1221 goto error;
1222 }
8ac94142 1223
5eb91c98 1224 ret = lttng_poll_wait(&events, -1);
a78af745 1225 health_poll_exit();
273ea72c 1226 if (ret < 0) {
88f2b785
MD
1227 /*
1228 * Restart interrupted system call.
1229 */
1230 if (errno == EINTR) {
1231 goto restart;
1232 }
273ea72c
DG
1233 goto error;
1234 }
1235
0d9c5d77
DG
1236 nb_fd = ret;
1237
5eb91c98
DG
1238 for (i = 0; i < nb_fd; i++) {
1239 /* Fetch once the poll data */
1240 revents = LTTNG_POLL_GETEV(&events, i);
1241 pollfd = LTTNG_POLL_GETFD(&events, i);
1242
840cb59c 1243 health_code_update();
44a5e5eb 1244
fd20dac9
MD
1245 if (!revents) {
1246 /* No activity for this FD (poll implementation). */
1247 continue;
1248 }
1249
5eb91c98 1250 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1251 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1252 if (ret) {
139ac872
MD
1253 err = 0;
1254 goto exit;
5eb91c98
DG
1255 }
1256
1257 /* Event on the registration socket */
3bd1e081 1258 if (pollfd == consumer_data->err_sock) {
5eb91c98 1259 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1260 ERR("consumer err socket poll error");
5eb91c98
DG
1261 goto error;
1262 }
1263 }
273ea72c
DG
1264 }
1265
3bd1e081 1266 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
1267 if (sock < 0) {
1268 goto error;
1269 }
1270
b662582b
DG
1271 /*
1272 * Set the CLOEXEC flag. Return code is useless because either way, the
1273 * show must go on.
1274 */
1275 (void) utils_set_fd_cloexec(sock);
1276
840cb59c 1277 health_code_update();
44a5e5eb 1278
3bd1e081 1279 DBG2("Receiving code from consumer err_sock");
ee0b0061 1280
712ea556 1281 /* Getting status code from kconsumerd */
54d01ffb
DG
1282 ret = lttcomm_recv_unix_sock(sock, &code,
1283 sizeof(enum lttcomm_return_code));
1d4b027a
DG
1284 if (ret <= 0) {
1285 goto error;
1286 }
1287
840cb59c 1288 health_code_update();
f73fabfd 1289 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
331744e3 1290 /* Connect both socket, command and metadata. */
3bd1e081
MD
1291 consumer_data->cmd_sock =
1292 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
4ce514c4 1293 consumer_data->metadata_fd =
331744e3 1294 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
92db7cdc
DG
1295 if (consumer_data->cmd_sock < 0
1296 || consumer_data->metadata_fd < 0) {
331744e3 1297 PERROR("consumer connect cmd socket");
a23ec3a7
DG
1298 /* On error, signal condition and quit. */
1299 signal_consumer_condition(consumer_data, -1);
1d4b027a
DG
1300 goto error;
1301 }
9363801e 1302 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
331744e3
JD
1303 /* Create metadata socket lock. */
1304 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1305 if (consumer_data->metadata_sock.lock == NULL) {
1306 PERROR("zmalloc pthread mutex");
1307 ret = -1;
1308 goto error;
1309 }
1310 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1311
a23ec3a7 1312 signal_consumer_condition(consumer_data, 1);
331744e3
JD
1313 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1314 DBG("Consumer metadata socket ready (fd: %d)",
4ce514c4 1315 consumer_data->metadata_fd);
1d4b027a 1316 } else {
3bd1e081 1317 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
1318 lttcomm_get_readable_code(-code));
1319 goto error;
1320 }
1321
331744e3 1322 /* Remove the consumerd error sock since we've established a connexion */
3bd1e081 1323 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1324 if (ret < 0) {
72079cae
DG
1325 goto error;
1326 }
1327
331744e3 1328 /* Add new accepted error socket. */
5eb91c98
DG
1329 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1330 if (ret < 0) {
72079cae 1331 goto error;
5eb91c98
DG
1332 }
1333
331744e3 1334 /* Add metadata socket that is successfully connected. */
4ce514c4 1335 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
331744e3
JD
1336 LPOLLIN | LPOLLRDHUP);
1337 if (ret < 0) {
1338 goto error;
1339 }
1340
840cb59c 1341 health_code_update();
44a5e5eb 1342
331744e3 1343 /* Infinite blocking call, waiting for transmission */
88f2b785 1344restart_poll:
331744e3 1345 while (1) {
42fc1d0b
DG
1346 health_code_update();
1347
1348 /* Exit the thread because the thread quit pipe has been triggered. */
1349 if (should_quit) {
1350 /* Not a health error. */
1351 err = 0;
1352 goto exit;
1353 }
1354
331744e3
JD
1355 health_poll_entry();
1356 ret = lttng_poll_wait(&events, -1);
1357 health_poll_exit();
1358 if (ret < 0) {
1359 /*
1360 * Restart interrupted system call.
1361 */
1362 if (errno == EINTR) {
1363 goto restart_poll;
1364 }
1365 goto error;
88f2b785 1366 }
72079cae 1367
331744e3 1368 nb_fd = ret;
0d9c5d77 1369
331744e3
JD
1370 for (i = 0; i < nb_fd; i++) {
1371 /* Fetch once the poll data */
1372 revents = LTTNG_POLL_GETEV(&events, i);
1373 pollfd = LTTNG_POLL_GETFD(&events, i);
5eb91c98 1374
331744e3 1375 health_code_update();
44a5e5eb 1376
fd20dac9
MD
1377 if (!revents) {
1378 /* No activity for this FD (poll implementation). */
1379 continue;
1380 }
1381
42fc1d0b
DG
1382 /*
1383 * Thread quit pipe has been triggered, flag that we should stop
1384 * but continue the current loop to handle potential data from
1385 * consumer.
1386 */
1387 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1388
331744e3
JD
1389 if (pollfd == sock) {
1390 /* Event on the consumerd socket */
1391 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1392 ERR("consumer err socket second poll error");
1393 goto error;
1394 }
1395 health_code_update();
1396 /* Wait for any kconsumerd error */
1397 ret = lttcomm_recv_unix_sock(sock, &code,
1398 sizeof(enum lttcomm_return_code));
1399 if (ret <= 0) {
1400 ERR("consumer closed the command socket");
1401 goto error;
1402 }
1403
1404 ERR("consumer return code : %s",
1405 lttcomm_get_readable_code(-code));
1406
1407 goto exit;
4ce514c4 1408 } else if (pollfd == consumer_data->metadata_fd) {
331744e3
JD
1409 /* UST metadata requests */
1410 ret = ust_consumer_metadata_request(
1411 &consumer_data->metadata_sock);
1412 if (ret < 0) {
1413 ERR("Handling metadata request");
1414 goto error;
1415 }
5eb91c98 1416 }
42fc1d0b 1417 /* No need for an else branch all FDs are tested prior. */
5eb91c98 1418 }
331744e3 1419 health_code_update();
5eb91c98
DG
1420 }
1421
139ac872 1422exit:
1d4b027a 1423error:
fdadac08
DG
1424 /*
1425 * We lock here because we are about to close the sockets and some other
92db7cdc
DG
1426 * thread might be using them so get exclusive access which will abort all
1427 * other consumer command by other threads.
fdadac08
DG
1428 */
1429 pthread_mutex_lock(&consumer_data->lock);
1430
5c827ce0
DG
1431 /* Immediately set the consumerd state to stopped */
1432 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1433 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1434 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1435 consumer_data->type == LTTNG_CONSUMER32_UST) {
1436 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1437 } else {
1438 /* Code flow error... */
1439 assert(0);
1440 }
1441
76d7553f
MD
1442 if (consumer_data->err_sock >= 0) {
1443 ret = close(consumer_data->err_sock);
1444 if (ret) {
1445 PERROR("close");
1446 }
a76cbd9f 1447 consumer_data->err_sock = -1;
76d7553f
MD
1448 }
1449 if (consumer_data->cmd_sock >= 0) {
1450 ret = close(consumer_data->cmd_sock);
1451 if (ret) {
1452 PERROR("close");
1453 }
a76cbd9f 1454 consumer_data->cmd_sock = -1;
76d7553f 1455 }
96544455
SS
1456 if (consumer_data->metadata_sock.fd_ptr &&
1457 *consumer_data->metadata_sock.fd_ptr >= 0) {
9363801e 1458 ret = close(*consumer_data->metadata_sock.fd_ptr);
331744e3
JD
1459 if (ret) {
1460 PERROR("close");
1461 }
1462 }
76d7553f
MD
1463 if (sock >= 0) {
1464 ret = close(sock);
1465 if (ret) {
1466 PERROR("close");
1467 }
1468 }
273ea72c 1469
3bd1e081
MD
1470 unlink(consumer_data->err_unix_sock_path);
1471 unlink(consumer_data->cmd_unix_sock_path);
1472 consumer_data->pid = 0;
fdadac08 1473 pthread_mutex_unlock(&consumer_data->lock);
92db7cdc 1474
fdadac08 1475 /* Cleanup metadata socket mutex. */
96544455
SS
1476 if (consumer_data->metadata_sock.lock) {
1477 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1478 free(consumer_data->metadata_sock.lock);
1479 }
5eb91c98 1480 lttng_poll_clean(&events);
76d7553f 1481error_poll:
139ac872 1482 if (err) {
840cb59c 1483 health_error();
139ac872
MD
1484 ERR("Health error occurred in %s", __func__);
1485 }
8782cc74 1486 health_unregister(health_sessiond);
76d7553f 1487 DBG("consumer thread cleanup completed");
0177d773 1488
5eb91c98 1489 return NULL;
099e26bd
DG
1490}
1491
099e26bd
DG
1492/*
1493 * This thread manage application communication.
1d4b027a
DG
1494 */
1495static void *thread_manage_apps(void *data)
099e26bd 1496{
139ac872 1497 int i, ret, pollfd, err = -1;
6cd525e8 1498 ssize_t size_ret;
5eb91c98 1499 uint32_t revents, nb_fd;
5eb91c98 1500 struct lttng_poll_event events;
099e26bd
DG
1501
1502 DBG("[thread] Manage application started");
1503
f6a9efaa
DG
1504 rcu_register_thread();
1505 rcu_thread_online();
1506
6c71277b 1507 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
927ca06a 1508
e547b070 1509 if (testpoint(sessiond_thread_manage_apps)) {
6993eeb3
CB
1510 goto error_testpoint;
1511 }
1512
840cb59c 1513 health_code_update();
44a5e5eb 1514
d0b96690 1515 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1516 if (ret < 0) {
76d7553f 1517 goto error_poll_create;
5eb91c98 1518 }
099e26bd 1519
5eb91c98
DG
1520 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1521 if (ret < 0) {
1522 goto error;
1523 }
099e26bd 1524
e547b070 1525 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
6993eeb3
CB
1526 goto error;
1527 }
8ac94142 1528
840cb59c 1529 health_code_update();
44a5e5eb 1530
5eb91c98 1531 while (1) {
7fa2082e 1532 DBG("Apps thread polling");
099e26bd
DG
1533
1534 /* Inifinite blocking call, waiting for transmission */
88f2b785 1535 restart:
a78af745 1536 health_poll_entry();
5eb91c98 1537 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1538 DBG("Apps thread return from poll on %d fds",
1539 LTTNG_POLL_GETNB(&events));
a78af745 1540 health_poll_exit();
099e26bd 1541 if (ret < 0) {
88f2b785
MD
1542 /*
1543 * Restart interrupted system call.
1544 */
1545 if (errno == EINTR) {
1546 goto restart;
1547 }
099e26bd
DG
1548 goto error;
1549 }
1550
0d9c5d77
DG
1551 nb_fd = ret;
1552
5eb91c98
DG
1553 for (i = 0; i < nb_fd; i++) {
1554 /* Fetch once the poll data */
1555 revents = LTTNG_POLL_GETEV(&events, i);
1556 pollfd = LTTNG_POLL_GETFD(&events, i);
1557
840cb59c 1558 health_code_update();
44a5e5eb 1559
fd20dac9
MD
1560 if (!revents) {
1561 /* No activity for this FD (poll implementation). */
1562 continue;
1563 }
1564
5eb91c98 1565 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1566 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1567 if (ret) {
139ac872
MD
1568 err = 0;
1569 goto exit;
5eb91c98 1570 }
099e26bd 1571
5eb91c98
DG
1572 /* Inspect the apps cmd pipe */
1573 if (pollfd == apps_cmd_pipe[0]) {
1574 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1575 ERR("Apps command pipe error");
0177d773 1576 goto error;
5eb91c98 1577 } else if (revents & LPOLLIN) {
d0b96690
DG
1578 int sock;
1579
5eb91c98 1580 /* Empty pipe */
6cd525e8
MD
1581 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1582 if (size_ret < sizeof(sock)) {
76d7553f 1583 PERROR("read apps cmd pipe");
5eb91c98
DG
1584 goto error;
1585 }
099e26bd 1586
840cb59c 1587 health_code_update();
44a5e5eb 1588
ffe60014 1589 /*
d0b96690
DG
1590 * We only monitor the error events of the socket. This
1591 * thread does not handle any incoming data from UST
1592 * (POLLIN).
ffe60014 1593 */
d0b96690
DG
1594 ret = lttng_poll_add(&events, sock,
1595 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1596 if (ret < 0) {
5eb91c98 1597 goto error;
e0c7ec2b 1598 }
acc7b41b 1599
d0b96690 1600 DBG("Apps with sock %d added to poll set", sock);
0177d773 1601 }
5eb91c98
DG
1602 } else {
1603 /*
54d01ffb
DG
1604 * At this point, we know that a registered application made
1605 * the event at poll_wait.
5eb91c98
DG
1606 */
1607 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1608 /* Removing from the poll set */
1609 ret = lttng_poll_del(&events, pollfd);
1610 if (ret < 0) {
1611 goto error;
1612 }
099e26bd 1613
b9d9b220 1614 /* Socket closed on remote end. */
56fff090 1615 ust_app_unregister(pollfd);
5eb91c98 1616 }
099e26bd 1617 }
44a5e5eb 1618
840cb59c 1619 health_code_update();
099e26bd 1620 }
099e26bd
DG
1621 }
1622
139ac872 1623exit:
099e26bd 1624error:
5eb91c98 1625 lttng_poll_clean(&events);
76d7553f 1626error_poll_create:
6993eeb3 1627error_testpoint:
6620da75
DG
1628 utils_close_pipe(apps_cmd_pipe);
1629 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1630
1631 /*
1632 * We don't clean the UST app hash table here since already registered
1633 * applications can still be controlled so let them be until the session
1634 * daemon dies or the applications stop.
1635 */
1636
139ac872 1637 if (err) {
840cb59c 1638 health_error();
139ac872
MD
1639 ERR("Health error occurred in %s", __func__);
1640 }
8782cc74 1641 health_unregister(health_sessiond);
76d7553f 1642 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1643 rcu_thread_offline();
1644 rcu_unregister_thread();
099e26bd
DG
1645 return NULL;
1646}
1647
d0b96690 1648/*
d88aee68
DG
1649 * Send a socket to a thread This is called from the dispatch UST registration
1650 * thread once all sockets are set for the application.
d0b96690 1651 *
b85dc84c
DG
1652 * The sock value can be invalid, we don't really care, the thread will handle
1653 * it and make the necessary cleanup if so.
1654 *
d0b96690
DG
1655 * On success, return 0 else a negative value being the errno message of the
1656 * write().
1657 */
d88aee68 1658static int send_socket_to_thread(int fd, int sock)
d0b96690 1659{
6cd525e8 1660 ssize_t ret;
d0b96690 1661
b85dc84c
DG
1662 /*
1663 * It's possible that the FD is set as invalid with -1 concurrently just
1664 * before calling this function being a shutdown state of the thread.
1665 */
1666 if (fd < 0) {
1667 ret = -EBADF;
1668 goto error;
1669 }
d0b96690 1670
6cd525e8
MD
1671 ret = lttng_write(fd, &sock, sizeof(sock));
1672 if (ret < sizeof(sock)) {
d88aee68 1673 PERROR("write apps pipe %d", fd);
d0b96690
DG
1674 if (ret < 0) {
1675 ret = -errno;
1676 }
1677 goto error;
1678 }
1679
1680 /* All good. Don't send back the write positive ret value. */
1681 ret = 0;
1682error:
6cd525e8 1683 return (int) ret;
d0b96690
DG
1684}
1685
f45e313d
DG
1686/*
1687 * Sanitize the wait queue of the dispatch registration thread meaning removing
1688 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1689 * notify socket is never received.
1690 */
1691static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1692{
1693 int ret, nb_fd = 0, i;
1694 unsigned int fd_added = 0;
1695 struct lttng_poll_event events;
1696 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1697
1698 assert(wait_queue);
1699
1700 lttng_poll_init(&events);
1701
1702 /* Just skip everything for an empty queue. */
1703 if (!wait_queue->count) {
1704 goto end;
1705 }
1706
1707 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1708 if (ret < 0) {
1709 goto error_create;
1710 }
1711
1712 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1713 &wait_queue->head, head) {
1714 assert(wait_node->app);
1715 ret = lttng_poll_add(&events, wait_node->app->sock,
1716 LPOLLHUP | LPOLLERR);
1717 if (ret < 0) {
1718 goto error;
1719 }
1720
1721 fd_added = 1;
1722 }
1723
1724 if (!fd_added) {
1725 goto end;
1726 }
1727
1728 /*
1729 * Poll but don't block so we can quickly identify the faulty events and
1730 * clean them afterwards from the wait queue.
1731 */
1732 ret = lttng_poll_wait(&events, 0);
1733 if (ret < 0) {
1734 goto error;
1735 }
1736 nb_fd = ret;
1737
1738 for (i = 0; i < nb_fd; i++) {
1739 /* Get faulty FD. */
1740 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1741 int pollfd = LTTNG_POLL_GETFD(&events, i);
1742
fd20dac9
MD
1743 if (!revents) {
1744 /* No activity for this FD (poll implementation). */
1745 continue;
1746 }
1747
f45e313d
DG
1748 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1749 &wait_queue->head, head) {
1750 if (pollfd == wait_node->app->sock &&
1751 (revents & (LPOLLHUP | LPOLLERR))) {
1752 cds_list_del(&wait_node->head);
1753 wait_queue->count--;
1754 ust_app_destroy(wait_node->app);
1755 free(wait_node);
1756 break;
1757 }
1758 }
1759 }
1760
1761 if (nb_fd > 0) {
1762 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1763 }
1764
1765end:
1766 lttng_poll_clean(&events);
1767 return;
1768
1769error:
1770 lttng_poll_clean(&events);
1771error_create:
1772 ERR("Unable to sanitize wait queue");
1773 return;
1774}
1775
099e26bd
DG
1776/*
1777 * Dispatch request from the registration threads to the application
1778 * communication thread.
1779 */
1780static void *thread_dispatch_ust_registration(void *data)
1781{
12e2b881 1782 int ret, err = -1;
8bdee6e2 1783 struct cds_wfcq_node *node;
099e26bd 1784 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1785 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1786 struct ust_reg_wait_queue wait_queue = {
1787 .count = 0,
1788 };
d0b96690 1789
6c71277b 1790 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
12e2b881 1791
9ad42ec1
MD
1792 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1793 goto error_testpoint;
1794 }
1795
12e2b881
MD
1796 health_code_update();
1797
f45e313d 1798 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1799
1800 DBG("[thread] Dispatch UST command started");
1801
26c9d55e 1802 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1803 health_code_update();
1804
099e26bd
DG
1805 /* Atomically prepare the queue futex */
1806 futex_nto1_prepare(&ust_cmd_queue.futex);
1807
1808 do {
d0b96690 1809 struct ust_app *app = NULL;
7972aab2 1810 ust_cmd = NULL;
d0b96690 1811
f45e313d
DG
1812 /*
1813 * Make sure we don't have node(s) that have hung up before receiving
1814 * the notify socket. This is to clean the list in order to avoid
1815 * memory leaks from notify socket that are never seen.
1816 */
1817 sanitize_wait_queue(&wait_queue);
1818
12e2b881 1819 health_code_update();
099e26bd 1820 /* Dequeue command for registration */
8bdee6e2 1821 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
099e26bd 1822 if (node == NULL) {
00a17c97 1823 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1824 /* Continue thread execution */
1825 break;
1826 }
1827
1828 ust_cmd = caa_container_of(node, struct ust_command, node);
1829
2f50c8a3
DG
1830 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1831 " gid:%d sock:%d name:%s (version %d.%d)",
1832 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1833 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1834 ust_cmd->sock, ust_cmd->reg_msg.name,
1835 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1836
1837 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1838 wait_node = zmalloc(sizeof(*wait_node));
1839 if (!wait_node) {
1840 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1841 ret = close(ust_cmd->sock);
1842 if (ret < 0) {
1843 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1844 }
51dec90d 1845 lttng_fd_put(LTTNG_FD_APPS, 1);
7972aab2 1846 free(ust_cmd);
d0b96690
DG
1847 goto error;
1848 }
1849 CDS_INIT_LIST_HEAD(&wait_node->head);
1850
1851 /* Create application object if socket is CMD. */
1852 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1853 ust_cmd->sock);
1854 if (!wait_node->app) {
1855 ret = close(ust_cmd->sock);
1856 if (ret < 0) {
1857 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1858 }
51dec90d 1859 lttng_fd_put(LTTNG_FD_APPS, 1);
d88aee68 1860 free(wait_node);
7972aab2 1861 free(ust_cmd);
d0b96690
DG
1862 continue;
1863 }
1864 /*
1865 * Add application to the wait queue so we can set the notify
1866 * socket before putting this object in the global ht.
1867 */
f45e313d
DG
1868 cds_list_add(&wait_node->head, &wait_queue.head);
1869 wait_queue.count++;
d0b96690 1870
7972aab2 1871 free(ust_cmd);
d0b96690
DG
1872 /*
1873 * We have to continue here since we don't have the notify
1874 * socket and the application MUST be added to the hash table
1875 * only at that moment.
1876 */
1877 continue;
1878 } else {
1879 /*
1880 * Look for the application in the local wait queue and set the
1881 * notify socket if found.
1882 */
d88aee68 1883 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1884 &wait_queue.head, head) {
12e2b881 1885 health_code_update();
d0b96690
DG
1886 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1887 wait_node->app->notify_sock = ust_cmd->sock;
1888 cds_list_del(&wait_node->head);
f45e313d 1889 wait_queue.count--;
d0b96690
DG
1890 app = wait_node->app;
1891 free(wait_node);
1892 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1893 break;
1894 }
1895 }
020d7f60
DG
1896
1897 /*
1898 * With no application at this stage the received socket is
1899 * basically useless so close it before we free the cmd data
1900 * structure for good.
1901 */
1902 if (!app) {
1903 ret = close(ust_cmd->sock);
1904 if (ret < 0) {
1905 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1906 }
51dec90d 1907 lttng_fd_put(LTTNG_FD_APPS, 1);
020d7f60 1908 }
7972aab2 1909 free(ust_cmd);
d0b96690
DG
1910 }
1911
1912 if (app) {
d0b96690
DG
1913 /*
1914 * @session_lock_list
1915 *
1916 * Lock the global session list so from the register up to the
1917 * registration done message, no thread can see the application
1918 * and change its state.
1919 */
1920 session_lock_list();
1921 rcu_read_lock();
d88aee68 1922
d0b96690
DG
1923 /*
1924 * Add application to the global hash table. This needs to be
1925 * done before the update to the UST registry can locate the
1926 * application.
1927 */
1928 ust_app_add(app);
d88aee68
DG
1929
1930 /* Set app version. This call will print an error if needed. */
1931 (void) ust_app_version(app);
1932
1933 /* Send notify socket through the notify pipe. */
1934 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1935 app->notify_sock);
1936 if (ret < 0) {
1937 rcu_read_unlock();
1938 session_unlock_list();
b85dc84c
DG
1939 /*
1940 * No notify thread, stop the UST tracing. However, this is
1941 * not an internal error of the this thread thus setting
1942 * the health error code to a normal exit.
1943 */
1944 err = 0;
d88aee68 1945 goto error;
6620da75 1946 }
d88aee68 1947
d0b96690
DG
1948 /*
1949 * Update newly registered application with the tracing
1950 * registry info already enabled information.
1951 */
1952 update_ust_app(app->sock);
d88aee68
DG
1953
1954 /*
1955 * Don't care about return value. Let the manage apps threads
1956 * handle app unregistration upon socket close.
1957 */
1958 (void) ust_app_register_done(app->sock);
1959
1960 /*
1961 * Even if the application socket has been closed, send the app
1962 * to the thread and unregistration will take place at that
1963 * place.
1964 */
1965 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 1966 if (ret < 0) {
d88aee68
DG
1967 rcu_read_unlock();
1968 session_unlock_list();
b85dc84c
DG
1969 /*
1970 * No apps. thread, stop the UST tracing. However, this is
1971 * not an internal error of the this thread thus setting
1972 * the health error code to a normal exit.
1973 */
1974 err = 0;
d88aee68 1975 goto error;
d0b96690 1976 }
d88aee68 1977
d0b96690
DG
1978 rcu_read_unlock();
1979 session_unlock_list();
099e26bd 1980 }
099e26bd
DG
1981 } while (node != NULL);
1982
12e2b881 1983 health_poll_entry();
099e26bd
DG
1984 /* Futex wait on queue. Blocking call on futex() */
1985 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 1986 health_poll_exit();
099e26bd 1987 }
12e2b881
MD
1988 /* Normal exit, no error */
1989 err = 0;
099e26bd
DG
1990
1991error:
d88aee68
DG
1992 /* Clean up wait queue. */
1993 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1994 &wait_queue.head, head) {
d88aee68 1995 cds_list_del(&wait_node->head);
f45e313d 1996 wait_queue.count--;
d88aee68
DG
1997 free(wait_node);
1998 }
1999
9ad42ec1 2000error_testpoint:
099e26bd 2001 DBG("Dispatch thread dying");
12e2b881
MD
2002 if (err) {
2003 health_error();
2004 ERR("Health error occurred in %s", __func__);
2005 }
8782cc74 2006 health_unregister(health_sessiond);
099e26bd
DG
2007 return NULL;
2008}
2009
2010/*
2011 * This thread manage application registration.
2012 */
2013static void *thread_registration_apps(void *data)
1d4b027a 2014{
139ac872 2015 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
2016 uint32_t revents, nb_fd;
2017 struct lttng_poll_event events;
099e26bd
DG
2018 /*
2019 * Get allocated in this thread, enqueued to a global queue, dequeued and
2020 * freed in the manage apps thread.
2021 */
2022 struct ust_command *ust_cmd = NULL;
1d4b027a 2023
099e26bd 2024 DBG("[thread] Manage application registration started");
1d4b027a 2025
6c71277b 2026 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
927ca06a 2027
e547b070 2028 if (testpoint(sessiond_thread_registration_apps)) {
6993eeb3
CB
2029 goto error_testpoint;
2030 }
8ac94142 2031
1d4b027a
DG
2032 ret = lttcomm_listen_unix_sock(apps_sock);
2033 if (ret < 0) {
76d7553f 2034 goto error_listen;
1d4b027a
DG
2035 }
2036
5eb91c98
DG
2037 /*
2038 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2039 * more will be added to this poll set.
2040 */
d0b96690 2041 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 2042 if (ret < 0) {
76d7553f 2043 goto error_create_poll;
5eb91c98 2044 }
273ea72c 2045
5eb91c98
DG
2046 /* Add the application registration socket */
2047 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2048 if (ret < 0) {
76d7553f 2049 goto error_poll_add;
5eb91c98 2050 }
273ea72c 2051
1d4b027a 2052 /* Notify all applications to register */
0fdd1e2c
DG
2053 ret = notify_ust_apps(1);
2054 if (ret < 0) {
2055 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
2056 "Execution continues but there might be problem for already\n"
2057 "running applications that wishes to register.");
0fdd1e2c 2058 }
1d4b027a
DG
2059
2060 while (1) {
2061 DBG("Accepting application registration");
273ea72c
DG
2062
2063 /* Inifinite blocking call, waiting for transmission */
88f2b785 2064 restart:
a78af745 2065 health_poll_entry();
5eb91c98 2066 ret = lttng_poll_wait(&events, -1);
a78af745 2067 health_poll_exit();
273ea72c 2068 if (ret < 0) {
88f2b785
MD
2069 /*
2070 * Restart interrupted system call.
2071 */
2072 if (errno == EINTR) {
2073 goto restart;
2074 }
273ea72c
DG
2075 goto error;
2076 }
2077
0d9c5d77
DG
2078 nb_fd = ret;
2079
5eb91c98 2080 for (i = 0; i < nb_fd; i++) {
840cb59c 2081 health_code_update();
139ac872 2082
5eb91c98
DG
2083 /* Fetch once the poll data */
2084 revents = LTTNG_POLL_GETEV(&events, i);
2085 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 2086
fd20dac9
MD
2087 if (!revents) {
2088 /* No activity for this FD (poll implementation). */
2089 continue;
2090 }
2091
5eb91c98 2092 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 2093 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 2094 if (ret) {
139ac872
MD
2095 err = 0;
2096 goto exit;
90014c57 2097 }
1d4b027a 2098
5eb91c98
DG
2099 /* Event on the registration socket */
2100 if (pollfd == apps_sock) {
2101 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2102 ERR("Register apps socket poll error");
2103 goto error;
2104 } else if (revents & LPOLLIN) {
2105 sock = lttcomm_accept_unix_sock(apps_sock);
2106 if (sock < 0) {
2107 goto error;
2108 }
099e26bd 2109
16c5c8fa
DG
2110 /*
2111 * Set socket timeout for both receiving and ending.
2112 * app_socket_timeout is in seconds, whereas
2113 * lttcomm_setsockopt_rcv_timeout and
2114 * lttcomm_setsockopt_snd_timeout expect msec as
2115 * parameter.
2116 */
2117 (void) lttcomm_setsockopt_rcv_timeout(sock,
2118 app_socket_timeout * 1000);
2119 (void) lttcomm_setsockopt_snd_timeout(sock,
2120 app_socket_timeout * 1000);
2121
b662582b
DG
2122 /*
2123 * Set the CLOEXEC flag. Return code is useless because
2124 * either way, the show must go on.
2125 */
2126 (void) utils_set_fd_cloexec(sock);
2127
5eb91c98 2128 /* Create UST registration command for enqueuing */
ba7f0ae5 2129 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 2130 if (ust_cmd == NULL) {
76d7553f 2131 PERROR("ust command zmalloc");
5eb91c98
DG
2132 goto error;
2133 }
1d4b027a 2134
5eb91c98
DG
2135 /*
2136 * Using message-based transmissions to ensure we don't
2137 * have to deal with partially received messages.
2138 */
4063050c
MD
2139 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2140 if (ret < 0) {
2141 ERR("Exhausted file descriptors allowed for applications.");
2142 free(ust_cmd);
2143 ret = close(sock);
2144 if (ret) {
2145 PERROR("close");
2146 }
2147 sock = -1;
2148 continue;
2149 }
d88aee68 2150
840cb59c 2151 health_code_update();
d0b96690
DG
2152 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2153 if (ret < 0) {
5eb91c98 2154 free(ust_cmd);
d0b96690 2155 /* Close socket of the application. */
76d7553f
MD
2156 ret = close(sock);
2157 if (ret) {
2158 PERROR("close");
2159 }
4063050c 2160 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 2161 sock = -1;
5eb91c98
DG
2162 continue;
2163 }
840cb59c 2164 health_code_update();
099e26bd 2165
5eb91c98 2166 ust_cmd->sock = sock;
34a2494f 2167 sock = -1;
099e26bd 2168
5eb91c98
DG
2169 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2170 " gid:%d sock:%d name:%s (version %d.%d)",
2171 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2172 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2173 ust_cmd->sock, ust_cmd->reg_msg.name,
2174 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 2175
5eb91c98
DG
2176 /*
2177 * Lock free enqueue the registration request. The red pill
54d01ffb 2178 * has been taken! This apps will be part of the *system*.
5eb91c98 2179 */
8bdee6e2 2180 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
5eb91c98
DG
2181
2182 /*
2183 * Wake the registration queue futex. Implicit memory
8bdee6e2 2184 * barrier with the exchange in cds_wfcq_enqueue.
5eb91c98
DG
2185 */
2186 futex_nto1_wake(&ust_cmd_queue.futex);
2187 }
2188 }
90014c57 2189 }
1d4b027a
DG
2190 }
2191
139ac872 2192exit:
1d4b027a 2193error:
0fdd1e2c
DG
2194 /* Notify that the registration thread is gone */
2195 notify_ust_apps(0);
2196
a4b35e07 2197 if (apps_sock >= 0) {
76d7553f
MD
2198 ret = close(apps_sock);
2199 if (ret) {
2200 PERROR("close");
2201 }
a4b35e07 2202 }
46c3f085 2203 if (sock >= 0) {
76d7553f
MD
2204 ret = close(sock);
2205 if (ret) {
2206 PERROR("close");
2207 }
4063050c 2208 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 2209 }
273ea72c 2210 unlink(apps_unix_sock_path);
0fdd1e2c 2211
76d7553f 2212error_poll_add:
5eb91c98 2213 lttng_poll_clean(&events);
76d7553f
MD
2214error_listen:
2215error_create_poll:
6993eeb3 2216error_testpoint:
76d7553f 2217 DBG("UST Registration thread cleanup complete");
9ad42ec1
MD
2218 if (err) {
2219 health_error();
2220 ERR("Health error occurred in %s", __func__);
2221 }
8782cc74 2222 health_unregister(health_sessiond);
5eb91c98 2223
1d4b027a
DG
2224 return NULL;
2225}
2226
8c0faa1d 2227/*
3bd1e081 2228 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 2229 * exec or it will fails.
8c0faa1d 2230 */
3bd1e081 2231static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 2232{
a23ec3a7 2233 int ret, clock_ret;
ee0b0061
DG
2234 struct timespec timeout;
2235
a23ec3a7
DG
2236 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2237 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 2238
a23ec3a7
DG
2239 /* Setup pthread condition */
2240 ret = pthread_condattr_init(&consumer_data->condattr);
4a15001e 2241 if (ret) {
a23ec3a7
DG
2242 errno = ret;
2243 PERROR("pthread_condattr_init consumer data");
2244 goto error;
2245 }
2246
2247 /*
2248 * Set the monotonic clock in order to make sure we DO NOT jump in time
2249 * between the clock_gettime() call and the timedwait call. See bug #324
2250 * for a more details and how we noticed it.
2251 */
2252 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
4a15001e 2253 if (ret) {
a23ec3a7
DG
2254 errno = ret;
2255 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
2256 goto error;
2257 }
8c0faa1d 2258
a23ec3a7 2259 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
4a15001e 2260 if (ret) {
a23ec3a7
DG
2261 errno = ret;
2262 PERROR("pthread_cond_init consumer data");
2263 goto error;
2264 }
2265
2266 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2267 consumer_data);
4a15001e
MD
2268 if (ret) {
2269 errno = ret;
3bd1e081 2270 PERROR("pthread_create consumer");
ee0b0061 2271 ret = -1;
8c0faa1d
DG
2272 goto error;
2273 }
2274
a23ec3a7
DG
2275 /* We are about to wait on a pthread condition */
2276 pthread_mutex_lock(&consumer_data->cond_mutex);
2277
ee0b0061 2278 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
2279 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2280 /*
2281 * Set the timeout for the condition timed wait even if the clock gettime
2282 * call fails since we might loop on that call and we want to avoid to
2283 * increment the timeout too many times.
2284 */
2285 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2286
2287 /*
2288 * The following loop COULD be skipped in some conditions so this is why we
2289 * set ret to 0 in order to make sure at least one round of the loop is
2290 * done.
2291 */
2292 ret = 0;
2293
2294 /*
2295 * Loop until the condition is reached or when a timeout is reached. Note
2296 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2297 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2298 * possible. This loop does not take any chances and works with both of
2299 * them.
2300 */
2301 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2302 if (clock_ret < 0) {
2303 PERROR("clock_gettime spawn consumer");
2304 /* Infinite wait for the consumerd thread to be ready */
2305 ret = pthread_cond_wait(&consumer_data->cond,
2306 &consumer_data->cond_mutex);
2307 } else {
2308 ret = pthread_cond_timedwait(&consumer_data->cond,
2309 &consumer_data->cond_mutex, &timeout);
2310 }
ee0b0061 2311 }
8c0faa1d 2312
a23ec3a7
DG
2313 /* Release the pthread condition */
2314 pthread_mutex_unlock(&consumer_data->cond_mutex);
2315
2316 if (ret != 0) {
2317 errno = ret;
2318 if (ret == ETIMEDOUT) {
4282f9a3
DG
2319 int pth_ret;
2320
ee0b0061
DG
2321 /*
2322 * Call has timed out so we kill the kconsumerd_thread and return
2323 * an error.
2324 */
a23ec3a7
DG
2325 ERR("Condition timed out. The consumer thread was never ready."
2326 " Killing it");
4282f9a3
DG
2327 pth_ret = pthread_cancel(consumer_data->thread);
2328 if (pth_ret < 0) {
3bd1e081 2329 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
2330 }
2331 } else {
a23ec3a7 2332 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061 2333 }
4282f9a3
DG
2334 /* Caller is expecting a negative value on failure. */
2335 ret = -1;
ee0b0061
DG
2336 goto error;
2337 }
2338
3bd1e081
MD
2339 pthread_mutex_lock(&consumer_data->pid_mutex);
2340 if (consumer_data->pid == 0) {
a23ec3a7 2341 ERR("Consumerd did not start");
3bd1e081 2342 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2343 goto error;
2344 }
3bd1e081 2345 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2346
8c0faa1d
DG
2347 return 0;
2348
2349error:
2350 return ret;
2351}
2352
d9800920 2353/*
3bd1e081 2354 * Join consumer thread
d9800920 2355 */
3bd1e081 2356static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2357{
2358 void *status;
cf3af59e 2359
e8209f6b
DG
2360 /* Consumer pid must be a real one. */
2361 if (consumer_data->pid > 0) {
c617c0c6 2362 int ret;
3bd1e081 2363 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2364 if (ret) {
4a15001e 2365 PERROR("Error killing consumer daemon");
cf3af59e
MD
2366 return ret;
2367 }
3bd1e081 2368 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2369 } else {
2370 return 0;
2371 }
2372}
2373
8c0faa1d 2374/*
3bd1e081 2375 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2376 *
d063d709 2377 * Return pid if successful else -1.
8c0faa1d 2378 */
3bd1e081 2379static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2380{
2381 int ret;
2382 pid_t pid;
94c55f17 2383 const char *consumer_to_use;
53086306 2384 const char *verbosity;
94c55f17 2385 struct stat st;
8c0faa1d 2386
3bd1e081 2387 DBG("Spawning consumerd");
c49dc785 2388
8c0faa1d
DG
2389 pid = fork();
2390 if (pid == 0) {
2391 /*
3bd1e081 2392 * Exec consumerd.
8c0faa1d 2393 */
daee5345 2394 if (opt_verbose_consumer) {
53086306 2395 verbosity = "--verbose";
4421f712 2396 } else if (lttng_opt_quiet) {
53086306 2397 verbosity = "--quiet";
4421f712
DG
2398 } else {
2399 verbosity = "";
53086306 2400 }
4421f712 2401
3bd1e081
MD
2402 switch (consumer_data->type) {
2403 case LTTNG_CONSUMER_KERNEL:
94c55f17 2404 /*
c7704d57
DG
2405 * Find out which consumerd to execute. We will first try the
2406 * 64-bit path, then the sessiond's installation directory, and
2407 * fallback on the 32-bit one,
94c55f17 2408 */
63a799e8
AM
2409 DBG3("Looking for a kernel consumer at these locations:");
2410 DBG3(" 1) %s", consumerd64_bin);
2411 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2412 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2413 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2414 DBG3("Found location #1");
94c55f17 2415 consumer_to_use = consumerd64_bin;
94c55f17 2416 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2417 DBG3("Found location #2");
94c55f17 2418 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2419 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2420 DBG3("Found location #3");
eb1e0bd4 2421 consumer_to_use = consumerd32_bin;
94c55f17 2422 } else {
63a799e8 2423 DBG("Could not find any valid consumerd executable");
4282f9a3 2424 ret = -EINVAL;
94c55f17
AM
2425 break;
2426 }
2427 DBG("Using kernel consumer at: %s", consumer_to_use);
4282f9a3 2428 ret = execl(consumer_to_use,
94c55f17
AM
2429 "lttng-consumerd", verbosity, "-k",
2430 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2431 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2432 "--group", tracing_group_name,
94c55f17 2433 NULL);
3bd1e081 2434 break;
7753dea8
MD
2435 case LTTNG_CONSUMER64_UST:
2436 {
b1e0b6b6 2437 char *tmpnew = NULL;
8f4905da
MD
2438
2439 if (consumerd64_libdir[0] != '\0') {
2440 char *tmp;
2441 size_t tmplen;
2442
2443 tmp = getenv("LD_LIBRARY_PATH");
2444 if (!tmp) {
2445 tmp = "";
2446 }
2447 tmplen = strlen("LD_LIBRARY_PATH=")
2448 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2449 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2450 if (!tmpnew) {
2451 ret = -ENOMEM;
2452 goto error;
2453 }
2454 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2455 strcat(tmpnew, consumerd64_libdir);
2456 if (tmp[0] != '\0') {
2457 strcat(tmpnew, ":");
2458 strcat(tmpnew, tmp);
2459 }
2460 ret = putenv(tmpnew);
2461 if (ret) {
2462 ret = -errno;
c6f76da9 2463 free(tmpnew);
8f4905da
MD
2464 goto error;
2465 }
2466 }
94c55f17 2467 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2468 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2469 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2470 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2471 "--group", tracing_group_name,
7753dea8 2472 NULL);
8f4905da
MD
2473 if (consumerd64_libdir[0] != '\0') {
2474 free(tmpnew);
2475 }
3bd1e081 2476 break;
7753dea8
MD
2477 }
2478 case LTTNG_CONSUMER32_UST:
2479 {
937dde8e 2480 char *tmpnew = NULL;
8f4905da
MD
2481
2482 if (consumerd32_libdir[0] != '\0') {
2483 char *tmp;
2484 size_t tmplen;
2485
2486 tmp = getenv("LD_LIBRARY_PATH");
2487 if (!tmp) {
2488 tmp = "";
2489 }
2490 tmplen = strlen("LD_LIBRARY_PATH=")
2491 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2492 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2493 if (!tmpnew) {
2494 ret = -ENOMEM;
2495 goto error;
2496 }
2497 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2498 strcat(tmpnew, consumerd32_libdir);
2499 if (tmp[0] != '\0') {
2500 strcat(tmpnew, ":");
2501 strcat(tmpnew, tmp);
2502 }
2503 ret = putenv(tmpnew);
2504 if (ret) {
2505 ret = -errno;
c6f76da9 2506 free(tmpnew);
8f4905da
MD
2507 goto error;
2508 }
2509 }
94c55f17 2510 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2511 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2512 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2513 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2514 "--group", tracing_group_name,
7753dea8 2515 NULL);
8f4905da
MD
2516 if (consumerd32_libdir[0] != '\0') {
2517 free(tmpnew);
2518 }
7753dea8
MD
2519 break;
2520 }
3bd1e081 2521 default:
76d7553f 2522 PERROR("unknown consumer type");
3bd1e081
MD
2523 exit(EXIT_FAILURE);
2524 }
8c0faa1d 2525 if (errno != 0) {
4282f9a3 2526 PERROR("Consumer execl()");
8c0faa1d 2527 }
4282f9a3 2528 /* Reaching this point, we got a failure on our execl(). */
8c0faa1d
DG
2529 exit(EXIT_FAILURE);
2530 } else if (pid > 0) {
2531 ret = pid;
8c0faa1d 2532 } else {
76d7553f 2533 PERROR("start consumer fork");
8c0faa1d 2534 ret = -errno;
8c0faa1d 2535 }
8f4905da 2536error:
8c0faa1d
DG
2537 return ret;
2538}
2539
693bd40b 2540/*
3bd1e081 2541 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2542 */
3bd1e081 2543static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2544{
c617c0c6 2545 int ret;
edb8b045
DG
2546
2547 /*
2548 * Set the listen() state on the socket since there is a possible race
2549 * between the exec() of the consumer daemon and this call if place in the
2550 * consumer thread. See bug #366 for more details.
2551 */
2552 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2553 if (ret < 0) {
2554 goto error;
2555 }
693bd40b 2556
3bd1e081
MD
2557 pthread_mutex_lock(&consumer_data->pid_mutex);
2558 if (consumer_data->pid != 0) {
2559 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2560 goto end;
2561 }
693bd40b 2562
3bd1e081 2563 ret = spawn_consumerd(consumer_data);
c49dc785 2564 if (ret < 0) {
3bd1e081
MD
2565 ERR("Spawning consumerd failed");
2566 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2567 goto error;
693bd40b 2568 }
c49dc785 2569
3bd1e081
MD
2570 /* Setting up the consumer_data pid */
2571 consumer_data->pid = ret;
48842b30 2572 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2573 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2574
3bd1e081
MD
2575 DBG2("Spawning consumer control thread");
2576 ret = spawn_consumer_thread(consumer_data);
693bd40b 2577 if (ret < 0) {
3bd1e081 2578 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2579 goto error;
2580 }
2581
c49dc785 2582end:
693bd40b
DG
2583 return 0;
2584
2585error:
331744e3 2586 /* Cleanup already created sockets on error. */
edb8b045 2587 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2588 int err;
2589
edb8b045
DG
2590 err = close(consumer_data->err_sock);
2591 if (err < 0) {
2592 PERROR("close consumer data error socket");
2593 }
2594 }
693bd40b
DG
2595 return ret;
2596}
2597
b73401da 2598/*
096102bd 2599 * Setup necessary data for kernel tracer action.
b73401da 2600 */
096102bd 2601static int init_kernel_tracer(void)
b73401da
DG
2602{
2603 int ret;
b73401da 2604
096102bd
DG
2605 /* Modprobe lttng kernel modules */
2606 ret = modprobe_lttng_control();
b73401da 2607 if (ret < 0) {
b73401da
DG
2608 goto error;
2609 }
2610
096102bd
DG
2611 /* Open debugfs lttng */
2612 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2613 if (kernel_tracer_fd < 0) {
2614 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2615 ret = -1;
2616 goto error_open;
54d01ffb
DG
2617 }
2618
2f77fc4b
DG
2619 /* Validate kernel version */
2620 ret = kernel_validate_version(kernel_tracer_fd);
2621 if (ret < 0) {
2622 goto error_version;
b551a063 2623 }
54d01ffb 2624
2f77fc4b
DG
2625 ret = modprobe_lttng_data();
2626 if (ret < 0) {
2627 goto error_modules;
54d01ffb
DG
2628 }
2629
2f77fc4b
DG
2630 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2631 return 0;
2632
2633error_version:
2634 modprobe_remove_lttng_control();
2635 ret = close(kernel_tracer_fd);
2636 if (ret) {
2637 PERROR("close");
b551a063 2638 }
2f77fc4b 2639 kernel_tracer_fd = -1;
f73fabfd 2640 return LTTNG_ERR_KERN_VERSION;
b551a063 2641
2f77fc4b
DG
2642error_modules:
2643 ret = close(kernel_tracer_fd);
2644 if (ret) {
2645 PERROR("close");
b551a063 2646 }
54d01ffb 2647
2f77fc4b
DG
2648error_open:
2649 modprobe_remove_lttng_control();
54d01ffb
DG
2650
2651error:
2f77fc4b
DG
2652 WARN("No kernel tracer available");
2653 kernel_tracer_fd = -1;
2654 if (!is_root) {
f73fabfd 2655 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2656 } else {
f73fabfd 2657 return LTTNG_ERR_KERN_NA;
2f77fc4b 2658 }
54d01ffb
DG
2659}
2660
2f77fc4b 2661
54d01ffb 2662/*
2f77fc4b
DG
2663 * Copy consumer output from the tracing session to the domain session. The
2664 * function also applies the right modification on a per domain basis for the
2665 * trace files destination directory.
36b588ed
MD
2666 *
2667 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2668 */
2f77fc4b 2669static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2670{
2671 int ret;
2f77fc4b
DG
2672 const char *dir_name;
2673 struct consumer_output *consumer;
2674
2675 assert(session);
2676 assert(session->consumer);
54d01ffb 2677
b551a063
DG
2678 switch (domain) {
2679 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2680 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2681 /*
2682 * XXX: We should audit the session creation and what this function
2683 * does "extra" in order to avoid a destroy since this function is used
2684 * in the domain session creation (kernel and ust) only. Same for UST
2685 * domain.
2686 */
2687 if (session->kernel_session->consumer) {
2688 consumer_destroy_output(session->kernel_session->consumer);
2689 }
2f77fc4b
DG
2690 session->kernel_session->consumer =
2691 consumer_copy_output(session->consumer);
2692 /* Ease our life a bit for the next part */
2693 consumer = session->kernel_session->consumer;
2694 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063 2695 break;
f20baf8e 2696 case LTTNG_DOMAIN_JUL:
5cdb6027 2697 case LTTNG_DOMAIN_LOG4J:
0e115563 2698 case LTTNG_DOMAIN_PYTHON:
b551a063 2699 case LTTNG_DOMAIN_UST:
2f77fc4b 2700 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2701 if (session->ust_session->consumer) {
2702 consumer_destroy_output(session->ust_session->consumer);
2703 }
2f77fc4b
DG
2704 session->ust_session->consumer =
2705 consumer_copy_output(session->consumer);
2706 /* Ease our life a bit for the next part */
2707 consumer = session->ust_session->consumer;
2708 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2709 break;
2710 default:
f73fabfd 2711 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2712 goto error;
2713 }
2714
2f77fc4b 2715 /* Append correct directory to subdir */
c30ce0b3
CB
2716 strncat(consumer->subdir, dir_name,
2717 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2718 DBG3("Copy session consumer subdir %s", consumer->subdir);
2719
f73fabfd 2720 ret = LTTNG_OK;
54d01ffb
DG
2721
2722error:
2723 return ret;
2724}
2725
00e2e675 2726/*
2f77fc4b 2727 * Create an UST session and add it to the session ust list.
36b588ed
MD
2728 *
2729 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2730 */
2f77fc4b
DG
2731static int create_ust_session(struct ltt_session *session,
2732 struct lttng_domain *domain)
00e2e675
DG
2733{
2734 int ret;
2f77fc4b 2735 struct ltt_ust_session *lus = NULL;
00e2e675 2736
a4b92340 2737 assert(session);
2f77fc4b
DG
2738 assert(domain);
2739 assert(session->consumer);
a4b92340 2740
2f77fc4b 2741 switch (domain->type) {
f20baf8e 2742 case LTTNG_DOMAIN_JUL:
5cdb6027 2743 case LTTNG_DOMAIN_LOG4J:
0e115563 2744 case LTTNG_DOMAIN_PYTHON:
2f77fc4b
DG
2745 case LTTNG_DOMAIN_UST:
2746 break;
2747 default:
2748 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2749 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2750 goto error;
2751 }
2752
2f77fc4b
DG
2753 DBG("Creating UST session");
2754
dec56f6c 2755 lus = trace_ust_create_session(session->id);
2f77fc4b 2756 if (lus == NULL) {
f73fabfd 2757 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2758 goto error;
2759 }
2760
2f77fc4b
DG
2761 lus->uid = session->uid;
2762 lus->gid = session->gid;
2bba9e53 2763 lus->output_traces = session->output_traces;
27babd3a 2764 lus->snapshot_mode = session->snapshot_mode;
ecc48a90 2765 lus->live_timer_interval = session->live_timer;
2f77fc4b 2766 session->ust_session = lus;
00e2e675 2767
2f77fc4b
DG
2768 /* Copy session output to the newly created UST session */
2769 ret = copy_session_consumer(domain->type, session);
f73fabfd 2770 if (ret != LTTNG_OK) {
00e2e675
DG
2771 goto error;
2772 }
2773
f73fabfd 2774 return LTTNG_OK;
00e2e675
DG
2775
2776error:
2f77fc4b
DG
2777 free(lus);
2778 session->ust_session = NULL;
00e2e675
DG
2779 return ret;
2780}
2781
2782/*
2f77fc4b 2783 * Create a kernel tracer session then create the default channel.
00e2e675 2784 */
2f77fc4b 2785static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2786{
2787 int ret;
a4b92340 2788
2f77fc4b 2789 DBG("Creating kernel session");
00e2e675 2790
2f77fc4b
DG
2791 ret = kernel_create_session(session, kernel_tracer_fd);
2792 if (ret < 0) {
f73fabfd 2793 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2794 goto error;
2795 }
2796
2f77fc4b
DG
2797 /* Code flow safety */
2798 assert(session->kernel_session);
2799
2800 /* Copy session output to the newly created Kernel session */
2801 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2802 if (ret != LTTNG_OK) {
a4b92340
DG
2803 goto error;
2804 }
2805
2f77fc4b
DG
2806 /* Create directory(ies) on local filesystem. */
2807 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2808 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2809 ret = run_as_mkdir_recursive(
2810 session->kernel_session->consumer->dst.trace_path,
2811 S_IRWXU | S_IRWXG, session->uid, session->gid);
2812 if (ret < 0) {
2813 if (ret != -EEXIST) {
2814 ERR("Trace directory creation error");
00e2e675
DG
2815 goto error;
2816 }
00e2e675 2817 }
2f77fc4b 2818 }
00e2e675 2819
2f77fc4b
DG
2820 session->kernel_session->uid = session->uid;
2821 session->kernel_session->gid = session->gid;
2bba9e53 2822 session->kernel_session->output_traces = session->output_traces;
27babd3a 2823 session->kernel_session->snapshot_mode = session->snapshot_mode;
00e2e675 2824
f73fabfd 2825 return LTTNG_OK;
00e2e675 2826
2f77fc4b
DG
2827error:
2828 trace_kernel_destroy_session(session->kernel_session);
2829 session->kernel_session = NULL;
2830 return ret;
2831}
00e2e675 2832
2f77fc4b
DG
2833/*
2834 * Count number of session permitted by uid/gid.
2835 */
2836static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2837{
2838 unsigned int i = 0;
2839 struct ltt_session *session;
07424f16 2840
2f77fc4b
DG
2841 DBG("Counting number of available session for UID %d GID %d",
2842 uid, gid);
2843 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2844 /*
2f77fc4b 2845 * Only list the sessions the user can control.
00e2e675 2846 */
2f77fc4b
DG
2847 if (!session_access_ok(session, uid, gid)) {
2848 continue;
2849 }
2850 i++;
a4b92340 2851 }
2f77fc4b 2852 return i;
00e2e675
DG
2853}
2854
54d01ffb
DG
2855/*
2856 * Process the command requested by the lttng client within the command
2857 * context structure. This function make sure that the return structure (llm)
2858 * is set and ready for transmission before returning.
2859 *
2860 * Return any error encountered or 0 for success.
53a80697
MD
2861 *
2862 * "sock" is only used for special-case var. len data.
36b588ed
MD
2863 *
2864 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2865 */
53a80697
MD
2866static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2867 int *sock_error)
54d01ffb 2868{
f73fabfd 2869 int ret = LTTNG_OK;
44d3bd01 2870 int need_tracing_session = 1;
2e09ba09 2871 int need_domain;
54d01ffb
DG
2872
2873 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2874
53a80697
MD
2875 *sock_error = 0;
2876
2e09ba09
MD
2877 switch (cmd_ctx->lsm->cmd_type) {
2878 case LTTNG_CREATE_SESSION:
27babd3a 2879 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2880 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09
MD
2881 case LTTNG_DESTROY_SESSION:
2882 case LTTNG_LIST_SESSIONS:
2883 case LTTNG_LIST_DOMAINS:
2884 case LTTNG_START_TRACE:
2885 case LTTNG_STOP_TRACE:
6d805429 2886 case LTTNG_DATA_PENDING:
da3c9ec1
DG
2887 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2888 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2889 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2890 case LTTNG_SNAPSHOT_RECORD:
fb198a11 2891 case LTTNG_SAVE_SESSION:
2e09ba09 2892 need_domain = 0;
3aace903 2893 break;
2e09ba09
MD
2894 default:
2895 need_domain = 1;
2896 }
2897
2898 if (opt_no_kernel && need_domain
2899 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2900 if (!is_root) {
f73fabfd 2901 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2902 } else {
f73fabfd 2903 ret = LTTNG_ERR_KERN_NA;
531d29f9 2904 }
4fba7219
DG
2905 goto error;
2906 }
2907
8d3113b2
DG
2908 /* Deny register consumer if we already have a spawned consumer. */
2909 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2910 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2911 if (kconsumer_data.pid > 0) {
f73fabfd 2912 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2913 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2914 goto error;
2915 }
2916 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2917 }
2918
54d01ffb
DG
2919 /*
2920 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2921 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2922 * command.
2923 */
2924 switch(cmd_ctx->lsm->cmd_type) {
2925 case LTTNG_LIST_SESSIONS:
2926 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2927 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2928 case LTTNG_LIST_DOMAINS:
2929 case LTTNG_LIST_CHANNELS:
2930 case LTTNG_LIST_EVENTS:
834978fd 2931 case LTTNG_LIST_SYSCALLS:
54d01ffb
DG
2932 break;
2933 default:
2934 /* Setup lttng message with no payload */
2935 ret = setup_lttng_msg(cmd_ctx, 0);
2936 if (ret < 0) {
2937 /* This label does not try to unlock the session */
2938 goto init_setup_error;
2939 }
2940 }
2941
2942 /* Commands that DO NOT need a session. */
2943 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2944 case LTTNG_CREATE_SESSION:
27babd3a 2945 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2946 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09 2947 case LTTNG_CALIBRATE:
54d01ffb
DG
2948 case LTTNG_LIST_SESSIONS:
2949 case LTTNG_LIST_TRACEPOINTS:
834978fd 2950 case LTTNG_LIST_SYSCALLS:
f37d259d 2951 case LTTNG_LIST_TRACEPOINT_FIELDS:
fb198a11 2952 case LTTNG_SAVE_SESSION:
44d3bd01 2953 need_tracing_session = 0;
54d01ffb
DG
2954 break;
2955 default:
2956 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2957 /*
2958 * We keep the session list lock across _all_ commands
2959 * for now, because the per-session lock does not
2960 * handle teardown properly.
2961 */
74babd95 2962 session_lock_list();
54d01ffb
DG
2963 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2964 if (cmd_ctx->session == NULL) {
bba2d65f 2965 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2966 goto error;
2967 } else {
2968 /* Acquire lock for the session */
2969 session_lock(cmd_ctx->session);
2970 }
2971 break;
2972 }
b389abbe 2973
5f3ecf22
DG
2974 /*
2975 * Commands that need a valid session but should NOT create one if none
2976 * exists. Instead of creating one and destroying it when the command is
2977 * handled, process that right before so we save some round trip in useless
2978 * code path.
2979 */
2980 switch (cmd_ctx->lsm->cmd_type) {
2981 case LTTNG_DISABLE_CHANNEL:
2982 case LTTNG_DISABLE_EVENT:
5f3ecf22
DG
2983 switch (cmd_ctx->lsm->domain.type) {
2984 case LTTNG_DOMAIN_KERNEL:
2985 if (!cmd_ctx->session->kernel_session) {
2986 ret = LTTNG_ERR_NO_CHANNEL;
2987 goto error;
2988 }
2989 break;
2990 case LTTNG_DOMAIN_JUL:
5cdb6027 2991 case LTTNG_DOMAIN_LOG4J:
0e115563 2992 case LTTNG_DOMAIN_PYTHON:
5f3ecf22
DG
2993 case LTTNG_DOMAIN_UST:
2994 if (!cmd_ctx->session->ust_session) {
2995 ret = LTTNG_ERR_NO_CHANNEL;
2996 goto error;
2997 }
2998 break;
2999 default:
3000 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
3001 goto error;
3002 }
3003 default:
3004 break;
3005 }
3006
2e09ba09
MD
3007 if (!need_domain) {
3008 goto skip_domain;
3009 }
a4b92340 3010
54d01ffb
DG
3011 /*
3012 * Check domain type for specific "pre-action".
3013 */
3014 switch (cmd_ctx->lsm->domain.type) {
3015 case LTTNG_DOMAIN_KERNEL:
d1f1c568 3016 if (!is_root) {
f73fabfd 3017 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
3018 goto error;
3019 }
3020
54d01ffb 3021 /* Kernel tracer check */
a4b35e07 3022 if (kernel_tracer_fd == -1) {
54d01ffb 3023 /* Basically, load kernel tracer modules */
096102bd
DG
3024 ret = init_kernel_tracer();
3025 if (ret != 0) {
54d01ffb
DG
3026 goto error;
3027 }
3028 }
5eb91c98 3029
5c827ce0
DG
3030 /* Consumer is in an ERROR state. Report back to client */
3031 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 3032 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
3033 goto error;
3034 }
3035
54d01ffb 3036 /* Need a session for kernel command */
44d3bd01 3037 if (need_tracing_session) {
54d01ffb 3038 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 3039 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 3040 if (ret < 0) {
f73fabfd 3041 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
3042 goto error;
3043 }
b389abbe 3044 }
7d29a247 3045
54d01ffb 3046 /* Start the kernel consumer daemon */
3bd1e081
MD
3047 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3048 if (kconsumer_data.pid == 0 &&
785d2d0d 3049 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3bd1e081
MD
3050 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3051 ret = start_consumerd(&kconsumer_data);
7d29a247 3052 if (ret < 0) {
f73fabfd 3053 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 3054 goto error;
950131af 3055 }
5c827ce0 3056 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
3057 } else {
3058 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 3059 }
173af62f 3060
a4b92340
DG
3061 /*
3062 * The consumer was just spawned so we need to add the socket to
3063 * the consumer output of the session if exist.
3064 */
3065 ret = consumer_create_socket(&kconsumer_data,
3066 cmd_ctx->session->kernel_session->consumer);
3067 if (ret < 0) {
3068 goto error;
173af62f 3069 }
0d0c377a 3070 }
5c827ce0 3071
54d01ffb 3072 break;
b9dfb167 3073 case LTTNG_DOMAIN_JUL:
5cdb6027 3074 case LTTNG_DOMAIN_LOG4J:
0e115563 3075 case LTTNG_DOMAIN_PYTHON:
2bdd86d4 3076 case LTTNG_DOMAIN_UST:
44d3bd01 3077 {
b51ec5b4
MD
3078 if (!ust_app_supported()) {
3079 ret = LTTNG_ERR_NO_UST;
3080 goto error;
3081 }
5c827ce0
DG
3082 /* Consumer is in an ERROR state. Report back to client */
3083 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 3084 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
3085 goto error;
3086 }
3087
44d3bd01 3088 if (need_tracing_session) {
a4b92340 3089 /* Create UST session if none exist. */
f6a9efaa 3090 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 3091 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 3092 &cmd_ctx->lsm->domain);
f73fabfd 3093 if (ret != LTTNG_OK) {
44d3bd01
DG
3094 goto error;
3095 }
3096 }
00e2e675 3097
7753dea8
MD
3098 /* Start the UST consumer daemons */
3099 /* 64-bit */
3100 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 3101 if (consumerd64_bin[0] != '\0' &&
7753dea8 3102 ustconsumer64_data.pid == 0 &&
785d2d0d 3103 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {