Clean-up: fix comment type in lttng-consumerd.c
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
CommitLineData
826d496d
MD
1/*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
0fdd1e2c 3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
26296c48 4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
fac6795d 5 *
d14d33bf
AM
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
91d76f53 9 *
d14d33bf
AM
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
91d76f53 14 *
d14d33bf
AM
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
fac6795d
DG
18 */
19
6c1c0768 20#define _LGPL_SOURCE
fac6795d
DG
21#include <getopt.h>
22#include <grp.h>
23#include <limits.h>
0bb7724a 24#include <paths.h>
fac6795d
DG
25#include <pthread.h>
26#include <signal.h>
27#include <stdio.h>
28#include <stdlib.h>
29#include <string.h>
331744e3 30#include <inttypes.h>
0fdd1e2c 31#include <sys/mman.h>
b73401da 32#include <sys/mount.h>
1e307fab 33#include <sys/resource.h>
fac6795d
DG
34#include <sys/socket.h>
35#include <sys/stat.h>
36#include <sys/types.h>
0fdd1e2c 37#include <sys/wait.h>
5c827ce0 38#include <urcu/uatomic.h>
fac6795d 39#include <unistd.h>
4f0b90ee 40#include <ctype.h>
fac6795d 41
990570ed 42#include <common/common.h>
d27c42b8 43#include <common/compat/socket.h>
e8fa9fb0 44#include <common/compat/getenv.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>
f40ef1d5 51#include <common/config/session-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"
7c1d2758 75#include "agent.h"
5e97de00 76#include "ht-cleanup.h"
fac6795d 77
ebaeda94
MD
78#define CONSUMERD_FILE "lttng-consumerd"
79
fac6795d 80const char *progname;
6c71277b 81static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
26296c48
JG
82static int tracing_group_name_override;
83static char *opt_pidfile;
5b8719f5 84static int opt_sig_parent;
97e19046 85static int opt_verbose_consumer;
72dd7491 86static int opt_daemon, opt_background;
4fba7219 87static int opt_no_kernel;
ef367a93 88static char *opt_load_session_path;
1d4b027a 89static pid_t ppid; /* Parent PID for --sig-parent option */
0bb7724a 90static pid_t child_ppid; /* Internal parent PID use with daemonize. */
67e40797 91static char *rundir;
c9cb3e7d 92static int lockfile_fd = -1;
3bd1e081 93
0bb7724a
DG
94/* Set to 1 when a SIGUSR1 signal is received. */
95static int recv_child_signal;
96
a23ec3a7
DG
97/*
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
100 */
3bd1e081
MD
101static struct consumer_data kconsumer_data = {
102 .type = LTTNG_CONSUMER_KERNEL,
60922cb0
DG
103 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
104 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
03550b58
MD
105 .err_sock = -1,
106 .cmd_sock = -1,
173af62f
DG
107 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
108 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
109 .cond = PTHREAD_COND_INITIALIZER,
110 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081 111};
7753dea8
MD
112static struct consumer_data ustconsumer64_data = {
113 .type = LTTNG_CONSUMER64_UST,
60922cb0
DG
114 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
115 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
03550b58
MD
116 .err_sock = -1,
117 .cmd_sock = -1,
173af62f
DG
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
7753dea8
MD
122};
123static struct consumer_data ustconsumer32_data = {
124 .type = LTTNG_CONSUMER32_UST,
60922cb0
DG
125 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
126 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
03550b58
MD
127 .err_sock = -1,
128 .cmd_sock = -1,
173af62f
DG
129 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
130 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
131 .cond = PTHREAD_COND_INITIALIZER,
132 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081
MD
133};
134
26296c48
JG
135/* Command line options */
136static const struct option long_options[] = {
0f5ea17c
JG
137 { "client-sock", required_argument, 0, 'c' },
138 { "apps-sock", required_argument, 0, 'a' },
139 { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
140 { "kconsumerd-err-sock", required_argument, 0, '\0' },
141 { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
142 { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
143 { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
144 { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
145 { "consumerd32-path", required_argument, 0, '\0' },
146 { "consumerd32-libdir", required_argument, 0, '\0' },
147 { "consumerd64-path", required_argument, 0, '\0' },
148 { "consumerd64-libdir", required_argument, 0, '\0' },
149 { "daemonize", no_argument, 0, 'd' },
150 { "background", no_argument, 0, 'b' },
151 { "sig-parent", no_argument, 0, 'S' },
152 { "help", no_argument, 0, 'h' },
153 { "group", required_argument, 0, 'g' },
154 { "version", no_argument, 0, 'V' },
155 { "quiet", no_argument, 0, 'q' },
156 { "verbose", no_argument, 0, 'v' },
157 { "verbose-consumer", no_argument, 0, '\0' },
158 { "no-kernel", no_argument, 0, '\0' },
159 { "pidfile", required_argument, 0, 'p' },
160 { "agent-tcp-port", required_argument, 0, '\0' },
161 { "config", required_argument, 0, 'f' },
162 { "load", required_argument, 0, 'l' },
163 { "kmod-probes", required_argument, 0, '\0' },
164 { "extra-kmod-probes", required_argument, 0, '\0' },
26296c48
JG
165 { NULL, 0, 0, 0 }
166};
167
168/* Command line options to ignore from configuration file */
169static const char *config_ignore_options[] = { "help", "version", "config" };
170
26c9d55e 171/* Shared between threads */
099e26bd 172static int dispatch_thread_exit;
fac6795d 173
54d01ffb
DG
174/* Global application Unix socket path */
175static char apps_unix_sock_path[PATH_MAX];
176/* Global client Unix socket path */
177static char client_unix_sock_path[PATH_MAX];
54d01ffb
DG
178/* global wait shm path for UST */
179static char wait_shm_path[PATH_MAX];
44a5e5eb
DG
180/* Global health check unix path */
181static char health_unix_sock_path[PATH_MAX];
fac6795d 182
1d4b027a 183/* Sockets and FDs */
a4b35e07
MD
184static int client_sock = -1;
185static int apps_sock = -1;
2f77fc4b 186int kernel_tracer_fd = -1;
76d7553f 187static int kernel_poll_pipe[2] = { -1, -1 };
1d4b027a 188
273ea72c
DG
189/*
190 * Quit pipe for all threads. This permits a single cancellation point
191 * for all threads when receiving an event on the pipe.
192 */
76d7553f 193static int thread_quit_pipe[2] = { -1, -1 };
273ea72c 194
099e26bd
DG
195/*
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
198 */
76d7553f 199static int apps_cmd_pipe[2] = { -1, -1 };
099e26bd 200
d0b96690
DG
201int apps_cmd_notify_pipe[2] = { -1, -1 };
202
1d4b027a 203/* Pthread, Mutexes and Semaphores */
1d4b027a 204static pthread_t apps_thread;
d0b96690 205static pthread_t apps_notify_thread;
099e26bd 206static pthread_t reg_apps_thread;
1d4b027a 207static pthread_t client_thread;
7a485870 208static pthread_t kernel_thread;
099e26bd 209static pthread_t dispatch_thread;
44a5e5eb 210static pthread_t health_thread;
0b2dc8df 211static pthread_t ht_cleanup_thread;
022d91ba 212static pthread_t agent_reg_thread;
ef367a93 213static pthread_t load_session_thread;
5eb91c98 214
099e26bd
DG
215/*
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 *
b22c5da8
DG
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
099e26bd
DG
223 */
224static struct ust_cmd_queue ust_cmd_queue;
225
b5541356
DG
226/*
227 * Pointer initialized before thread creation.
228 *
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
04ea676f 232 *
d063d709 233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
54d01ffb 234 * session_lock_list and session_unlock_list for lock acquisition.
b5541356
DG
235 */
236static struct ltt_session_list *session_list_ptr;
237
7753dea8
MD
238int ust_consumerd64_fd = -1;
239int ust_consumerd32_fd = -1;
240
fb6f1fa2
YB
241static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
242static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
243static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
244static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
26296c48
JG
245static int consumerd32_bin_override;
246static int consumerd64_bin_override;
247static int consumerd32_libdir_override;
248static int consumerd64_libdir_override;
fb09408a 249
2f77fc4b
DG
250static const char *module_proc_lttng = "/proc/lttng";
251
5c827ce0
DG
252/*
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
255 */
256enum consumerd_state {
257 CONSUMER_STARTED = 1,
258 CONSUMER_STOPPED = 2,
259 CONSUMER_ERROR = 3,
260};
261
262/*
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
267 *
268 * The following example shows a possible race condition of this scheme:
269 *
270 * consumer thread error happens
271 * client cmd arrives
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
275 * ...
276 *
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
282 */
283static enum consumerd_state ust_consumerd_state;
284static enum consumerd_state kernel_consumerd_state;
285
ae9e45b3
DG
286/*
287 * Socket timeout for receiving and sending in seconds.
288 */
289static int app_socket_timeout;
290
12744796
DG
291/* Set in main() with the current page size. */
292long page_size;
293
8782cc74
MD
294/* Application health monitoring */
295struct health_app *health_sessiond;
296
022d91ba
DG
297/* Agent TCP port for registration. Used by the agent thread. */
298unsigned int agent_tcp_port = DEFAULT_AGENT_TCP_PORT;
4d076222 299
f43f95a9
DG
300/* Am I root or not. */
301int is_root; /* Set to 1 if the daemon is running as root */
302
26296c48
JG
303const char * const config_section_name = "sessiond";
304
ef367a93
JG
305/* Load session thread information to operate. */
306struct load_session_thread_data *load_info;
307
7c1d2758
JG
308/* Global hash tables */
309struct lttng_ht *agent_apps_ht_by_sock = NULL;
310
97bc1426
MD
311/*
312 * Whether sessiond is ready for commands/health check requests.
313 * NR_LTTNG_SESSIOND_READY must match the number of calls to
ef367a93 314 * sessiond_notify_ready().
97bc1426 315 */
ef367a93 316#define NR_LTTNG_SESSIOND_READY 3
97bc1426
MD
317int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
318
5e97de00
JG
319int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
320{
321 return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
322}
323
97bc1426 324/* Notify parents that we are ready for cmd and health check */
ef367a93
JG
325LTTNG_HIDDEN
326void sessiond_notify_ready(void)
97bc1426
MD
327{
328 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
329 /*
330 * Notify parent pid that we are ready to accept command
331 * for client side. This ppid is the one from the
332 * external process that spawned us.
333 */
334 if (opt_sig_parent) {
335 kill(ppid, SIGUSR1);
336 }
337
338 /*
339 * Notify the parent of the fork() process that we are
340 * ready.
341 */
72dd7491 342 if (opt_daemon || opt_background) {
97bc1426
MD
343 kill(child_ppid, SIGUSR1);
344 }
345 }
346}
347
fb09408a 348static
7753dea8 349void setup_consumerd_path(void)
fb09408a 350{
fc7a59ce 351 const char *bin, *libdir;
fb09408a 352
7753dea8
MD
353 /*
354 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
355 * native architecture size consumer if CONFIG_CONSUMER*_PATH
356 * has not been defined.
7753dea8 357 */
ebaeda94 358#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
359 if (!consumerd32_bin[0]) {
360 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
361 }
362 if (!consumerd32_libdir[0]) {
363 consumerd32_libdir = INSTALL_LIB_PATH;
364 }
365#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
366 if (!consumerd64_bin[0]) {
367 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 368 }
ebaeda94
MD
369 if (!consumerd64_libdir[0]) {
370 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
371 }
372#else
373#error "Unknown bitness"
374#endif
375
fb09408a
MD
376 /*
377 * runtime env. var. overrides the build default.
378 */
e8fa9fb0 379 bin = lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
fc7a59ce
AM
380 if (bin) {
381 consumerd32_bin = bin;
7753dea8 382 }
e8fa9fb0 383 bin = lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
fc7a59ce
AM
384 if (bin) {
385 consumerd64_bin = bin;
ebaeda94 386 }
e8fa9fb0 387 libdir = lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
388 if (libdir) {
389 consumerd32_libdir = libdir;
390 }
e8fa9fb0 391 libdir = lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
392 if (libdir) {
393 consumerd64_libdir = libdir;
fb09408a
MD
394 }
395}
396
4a15001e
MD
397static
398int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
399 int *a_pipe)
5eb91c98
DG
400{
401 int ret;
402
d0b96690 403 assert(events);
5eb91c98
DG
404
405 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
406 if (ret < 0) {
407 goto error;
408 }
409
410 /* Add quit pipe */
4a15001e 411 ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
5eb91c98
DG
412 if (ret < 0) {
413 goto error;
414 }
415
416 return 0;
417
418error:
419 return ret;
420}
421
4a15001e
MD
422/*
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 */
425int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
426{
427 return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
428}
429
273ea72c 430/*
5eb91c98 431 * Init thread quit pipe.
273ea72c
DG
432 *
433 * Return -1 on error or 0 if all pipes are created.
434 */
4a15001e 435static int __init_thread_quit_pipe(int *a_pipe)
273ea72c 436{
730389d9 437 int ret, i;
273ea72c 438
4a15001e 439 ret = pipe(a_pipe);
273ea72c 440 if (ret < 0) {
730389d9 441 PERROR("thread quit pipe");
273ea72c
DG
442 goto error;
443 }
444
730389d9 445 for (i = 0; i < 2; i++) {
4a15001e 446 ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
730389d9
DG
447 if (ret < 0) {
448 PERROR("fcntl");
449 goto error;
450 }
451 }
452
273ea72c
DG
453error:
454 return ret;
455}
456
4a15001e
MD
457static int init_thread_quit_pipe(void)
458{
459 return __init_thread_quit_pipe(thread_quit_pipe);
460}
461
099e26bd
DG
462/*
463 * Stop all threads by closing the thread quit pipe.
464 */
cf3af59e
MD
465static void stop_threads(void)
466{
5eb91c98
DG
467 int ret;
468
cf3af59e
MD
469 /* Stopping all threads */
470 DBG("Terminating all threads");
54d01ffb 471 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
472 if (ret < 0) {
473 ERR("write error on thread quit pipe");
474 }
475
099e26bd 476 /* Dispatch thread */
26c9d55e 477 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 478 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
479}
480
e975f9f8
DG
481/*
482 * Close every consumer sockets.
483 */
484static void close_consumer_sockets(void)
485{
486 int ret;
487
488 if (kconsumer_data.err_sock >= 0) {
489 ret = close(kconsumer_data.err_sock);
490 if (ret < 0) {
491 PERROR("kernel consumer err_sock close");
492 }
493 }
494 if (ustconsumer32_data.err_sock >= 0) {
495 ret = close(ustconsumer32_data.err_sock);
496 if (ret < 0) {
a76cbd9f 497 PERROR("UST consumerd32 err_sock close");
e975f9f8
DG
498 }
499 }
500 if (ustconsumer64_data.err_sock >= 0) {
501 ret = close(ustconsumer64_data.err_sock);
502 if (ret < 0) {
a76cbd9f 503 PERROR("UST consumerd64 err_sock close");
e975f9f8
DG
504 }
505 }
506 if (kconsumer_data.cmd_sock >= 0) {
507 ret = close(kconsumer_data.cmd_sock);
508 if (ret < 0) {
509 PERROR("kernel consumer cmd_sock close");
510 }
511 }
512 if (ustconsumer32_data.cmd_sock >= 0) {
513 ret = close(ustconsumer32_data.cmd_sock);
514 if (ret < 0) {
a76cbd9f 515 PERROR("UST consumerd32 cmd_sock close");
e975f9f8
DG
516 }
517 }
518 if (ustconsumer64_data.cmd_sock >= 0) {
519 ret = close(ustconsumer64_data.cmd_sock);
520 if (ret < 0) {
a76cbd9f 521 PERROR("UST consumerd64 cmd_sock close");
e975f9f8
DG
522 }
523 }
524}
525
c9cb3e7d
JG
526/*
527 * Generate the full lock file path using the rundir.
528 *
529 * Return the snprintf() return value thus a negative value is an error.
530 */
531static int generate_lock_file_path(char *path, size_t len)
532{
533 int ret;
534
535 assert(path);
536 assert(rundir);
537
538 /* Build lockfile path from rundir. */
539 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
540 if (ret < 0) {
541 PERROR("snprintf lockfile path");
542 }
543
544 return ret;
545}
546
4e4714cb
JR
547/*
548 * Wait on consumer process termination.
549 *
550 * Need to be called with the consumer data lock held or from a context
551 * ensuring no concurrent access to data (e.g: cleanup).
552 */
553static void wait_consumer(struct consumer_data *consumer_data)
554{
555 pid_t ret;
556 int status;
557
558 if (consumer_data->pid <= 0) {
559 return;
560 }
561
562 DBG("Waiting for complete teardown of consumerd (PID: %d)",
563 consumer_data->pid);
564 ret = waitpid(consumer_data->pid, &status, 0);
565 if (ret == -1) {
566 PERROR("consumerd waitpid pid: %d", consumer_data->pid)
567 }
568 if (!WIFEXITED(status)) {
569 ERR("consumerd termination with error: %d",
570 WEXITSTATUS(ret));
571 }
572 consumer_data->pid = 0;
573}
574
fac6795d 575/*
4a15001e 576 * Cleanup the session daemon's data structures.
fac6795d 577 */
4a15001e 578static void sessiond_cleanup(void)
fac6795d 579{
ef599319 580 int ret;
af9737e9 581 struct ltt_session *sess, *stmp;
8c6c56c2 582 char path[PATH_MAX];
fac6795d 583
4a15001e 584 DBG("Cleanup sessiond");
e07ae692 585
4e449f3f
MD
586 /*
587 * Close the thread quit pipe. It has already done its job,
588 * since we are now called.
589 */
2f77fc4b
DG
590 utils_close_pipe(thread_quit_pipe);
591
35f90c40
DG
592 /*
593 * If opt_pidfile is undefined, the default file will be wiped when
594 * removing the rundir.
595 */
596 if (opt_pidfile) {
597 ret = remove(opt_pidfile);
598 if (ret < 0) {
599 PERROR("remove pidfile %s", opt_pidfile);
600 }
601 }
602
8c6c56c2
MD
603 DBG("Removing sessiond and consumerd content of directory %s", rundir);
604
605 /* sessiond */
606 snprintf(path, PATH_MAX,
607 "%s/%s",
608 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
609 DBG("Removing %s", path);
610 (void) unlink(path);
611
cd9290dd 612 snprintf(path, PATH_MAX, "%s/%s", rundir,
022d91ba 613 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE);
cd9290dd
DG
614 DBG("Removing %s", path);
615 (void) unlink(path);
616
8c6c56c2
MD
617 /* kconsumerd */
618 snprintf(path, PATH_MAX,
619 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
620 rundir);
621 DBG("Removing %s", path);
622 (void) unlink(path);
623
624 snprintf(path, PATH_MAX,
625 DEFAULT_KCONSUMERD_PATH,
626 rundir);
627 DBG("Removing directory %s", path);
628 (void) rmdir(path);
629
630 /* ust consumerd 32 */
631 snprintf(path, PATH_MAX,
632 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
633 rundir);
634 DBG("Removing %s", path);
635 (void) unlink(path);
636
637 snprintf(path, PATH_MAX,
638 DEFAULT_USTCONSUMERD32_PATH,
639 rundir);
640 DBG("Removing directory %s", path);
641 (void) rmdir(path);
642
643 /* ust consumerd 64 */
644 snprintf(path, PATH_MAX,
645 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
646 rundir);
647 DBG("Removing %s", path);
648 (void) unlink(path);
649
650 snprintf(path, PATH_MAX,
651 DEFAULT_USTCONSUMERD64_PATH,
652 rundir);
653 DBG("Removing directory %s", path);
654 (void) rmdir(path);
5461b305 655
99bab54f 656 DBG("Cleaning up all sessions");
fac6795d 657
b5541356 658 /* Destroy session list mutex */
273ea72c
DG
659 if (session_list_ptr != NULL) {
660 pthread_mutex_destroy(&session_list_ptr->lock);
661
662 /* Cleanup ALL session */
54d01ffb
DG
663 cds_list_for_each_entry_safe(sess, stmp,
664 &session_list_ptr->head, list) {
2f77fc4b 665 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
666 }
667 }
668
4e4714cb
JR
669 wait_consumer(&kconsumer_data);
670 wait_consumer(&ustconsumer64_data);
671 wait_consumer(&ustconsumer32_data);
672
6a4e4039
JG
673 DBG("Cleaning up all agent apps");
674 agent_app_ht_clean();
675
099e26bd 676 DBG("Closing all UST sockets");
56fff090 677 ust_app_clean_list();
7972aab2 678 buffer_reg_destroy_registries();
099e26bd 679
4fba7219
DG
680 if (is_root && !opt_no_kernel) {
681 DBG2("Closing kernel fd");
a4b35e07 682 if (kernel_tracer_fd >= 0) {
76d7553f
MD
683 ret = close(kernel_tracer_fd);
684 if (ret) {
685 PERROR("close");
686 }
a4b35e07 687 }
2f50c8a3 688 DBG("Unloading kernel modules");
096102bd 689 modprobe_remove_lttng_all();
834978fd 690 free(syscall_table);
2f50c8a3 691 }
2f77fc4b 692
e975f9f8
DG
693 close_consumer_sockets();
694
ef367a93
JG
695 if (load_info) {
696 load_session_destroy_data(load_info);
697 free(load_info);
698 }
699
c9cb3e7d
JG
700 /*
701 * Cleanup lock file by deleting it and finaly closing it which will
702 * release the file system lock.
703 */
704 if (lockfile_fd >= 0) {
705 char lockfile_path[PATH_MAX];
706
4a15001e
MD
707 ret = generate_lock_file_path(lockfile_path,
708 sizeof(lockfile_path));
c9cb3e7d
JG
709 if (ret > 0) {
710 ret = remove(lockfile_path);
711 if (ret < 0) {
712 PERROR("remove lock file");
713 }
714 ret = close(lockfile_fd);
715 if (ret < 0) {
716 PERROR("close lock file");
717 }
718 }
719 }
720
721 /*
722 * We do NOT rmdir rundir because there are other processes
723 * using it, for instance lttng-relayd, which can start in
724 * parallel with this teardown.
725 */
726
727 free(rundir);
4a15001e
MD
728}
729
730/*
731 * Cleanup the daemon's option data structures.
732 */
733static void sessiond_cleanup_options(void)
734{
735 DBG("Cleaning up options");
736
737 /*
738 * If the override option is set, the pointer points to a *non* const
739 * thus freeing it even though the variable type is set to const.
740 */
741 if (tracing_group_name_override) {
742 free((void *) tracing_group_name);
743 }
744 if (consumerd32_bin_override) {
745 free((void *) consumerd32_bin);
746 }
747 if (consumerd64_bin_override) {
748 free((void *) consumerd64_bin);
749 }
750 if (consumerd32_libdir_override) {
751 free((void *) consumerd32_libdir);
752 }
753 if (consumerd64_libdir_override) {
754 free((void *) consumerd64_libdir);
755 }
756
757 free(opt_pidfile);
758 free(opt_load_session_path);
759 free(kmod_probes_list);
760 free(kmod_extra_probes_list);
c9cb3e7d 761
7567352f 762 run_as_destroy_worker();
fac6795d
DG
763}
764
e065084a 765/*
d063d709 766 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 767 *
d063d709 768 * Return lttcomm error code.
e065084a
DG
769 */
770static int send_unix_sock(int sock, void *buf, size_t len)
771{
772 /* Check valid length */
c617c0c6 773 if (len == 0) {
e065084a
DG
774 return -1;
775 }
776
777 return lttcomm_send_unix_sock(sock, buf, len);
778}
779
5461b305 780/*
d063d709 781 * Free memory of a command context structure.
5461b305 782 */
a2fb29a5 783static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 784{
a2fb29a5
DG
785 DBG("Clean command context structure");
786 if (*cmd_ctx) {
787 if ((*cmd_ctx)->llm) {
788 free((*cmd_ctx)->llm);
5461b305 789 }
a2fb29a5
DG
790 if ((*cmd_ctx)->lsm) {
791 free((*cmd_ctx)->lsm);
5461b305 792 }
a2fb29a5
DG
793 free(*cmd_ctx);
794 *cmd_ctx = NULL;
5461b305
DG
795 }
796}
797
fac6795d 798/*
0fdd1e2c 799 * Notify UST applications using the shm mmap futex.
fac6795d 800 */
0fdd1e2c 801static int notify_ust_apps(int active)
fac6795d 802{
0fdd1e2c 803 char *wait_shm_mmap;
fac6795d 804
0fdd1e2c 805 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 806
0fdd1e2c
DG
807 /* See shm.c for this call implying mmap, shm and futex calls */
808 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
809 if (wait_shm_mmap == NULL) {
fac6795d
DG
810 goto error;
811 }
812
0fdd1e2c
DG
813 /* Wake waiting process */
814 futex_wait_update((int32_t *) wait_shm_mmap, active);
815
816 /* Apps notified successfully */
817 return 0;
fac6795d
DG
818
819error:
0fdd1e2c 820 return -1;
fac6795d
DG
821}
822
e065084a 823/*
d063d709
DG
824 * Setup the outgoing data buffer for the response (llm) by allocating the
825 * right amount of memory and copying the original information from the lsm
826 * structure.
ca95a216 827 *
6e10c9b9 828 * Return 0 on success, negative value on error.
ca95a216 829 */
6e10c9b9
PP
830static int setup_lttng_msg(struct command_ctx *cmd_ctx,
831 const void *payload_buf, size_t payload_len,
832 const void *cmd_header_buf, size_t cmd_header_len)
ca95a216 833{
6e10c9b9
PP
834 int ret = 0;
835 const size_t header_len = sizeof(struct lttcomm_lttng_msg);
836 const size_t cmd_header_offset = header_len;
837 const size_t payload_offset = cmd_header_offset + cmd_header_len;
838 const size_t total_msg_size = header_len + cmd_header_len + payload_len;
ca95a216 839
6e10c9b9 840 cmd_ctx->llm = zmalloc(total_msg_size);
5461b305 841
5461b305 842 if (cmd_ctx->llm == NULL) {
76d7553f 843 PERROR("zmalloc");
5461b305 844 ret = -ENOMEM;
6e10c9b9 845 goto end;
ca95a216
DG
846 }
847
5461b305
DG
848 /* Copy common data */
849 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 850 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
6e10c9b9
PP
851 cmd_ctx->llm->cmd_header_size = cmd_header_len;
852 cmd_ctx->llm->data_size = payload_len;
853 cmd_ctx->lttng_msg_size = total_msg_size;
5461b305 854
6e10c9b9 855 /* Copy command header */
b4e3ceb9
PP
856 if (cmd_header_len) {
857 memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
858 cmd_header_len);
859 }
5461b305 860
6e10c9b9 861 /* Copy payload */
b4e3ceb9
PP
862 if (payload_len) {
863 memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
864 payload_len);
865 }
ca95a216 866
6e10c9b9 867end:
ca95a216
DG
868 return ret;
869}
870
6e10c9b9
PP
871/*
872 * Version of setup_lttng_msg() without command header.
873 */
874static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
875 void *payload_buf, size_t payload_len)
876{
877 return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
878}
7a485870 879/*
5eb91c98 880 * Update the kernel poll set of all channel fd available over all tracing
d063d709 881 * session. Add the wakeup pipe at the end of the set.
7a485870 882 */
5eb91c98 883static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 884{
5eb91c98 885 int ret;
7a485870
DG
886 struct ltt_session *session;
887 struct ltt_kernel_channel *channel;
888
5eb91c98 889 DBG("Updating kernel poll set");
7a485870 890
54d01ffb 891 session_lock_list();
b5541356 892 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 893 session_lock(session);
7a485870 894 if (session->kernel_session == NULL) {
54d01ffb 895 session_unlock(session);
7a485870
DG
896 continue;
897 }
7a485870 898
54d01ffb
DG
899 cds_list_for_each_entry(channel,
900 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
901 /* Add channel fd to the kernel poll set */
902 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
903 if (ret < 0) {
54d01ffb 904 session_unlock(session);
5eb91c98
DG
905 goto error;
906 }
907 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 908 }
54d01ffb 909 session_unlock(session);
7a485870 910 }
54d01ffb 911 session_unlock_list();
7a485870 912
5eb91c98 913 return 0;
7a485870
DG
914
915error:
54d01ffb 916 session_unlock_list();
7a485870
DG
917 return -1;
918}
919
920/*
54d01ffb 921 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 922 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 923 *
d063d709 924 * Useful for CPU hotplug feature.
7a485870 925 */
2bdd86d4 926static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
927{
928 int ret = 0;
929 struct ltt_session *session;
173af62f 930 struct ltt_kernel_session *ksess;
7a485870
DG
931 struct ltt_kernel_channel *channel;
932
933 DBG("Updating kernel streams for channel fd %d", fd);
934
54d01ffb 935 session_lock_list();
b5541356 936 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 937 session_lock(session);
7a485870 938 if (session->kernel_session == NULL) {
54d01ffb 939 session_unlock(session);
7a485870
DG
940 continue;
941 }
173af62f 942 ksess = session->kernel_session;
d9800920 943
4a15001e
MD
944 cds_list_for_each_entry(channel,
945 &ksess->channel_list.head, list) {
946 struct lttng_ht_iter iter;
947 struct consumer_socket *socket;
d9800920 948
4a15001e
MD
949 if (channel->fd != fd) {
950 continue;
951 }
952 DBG("Channel found, updating kernel streams");
953 ret = kernel_open_channel_stream(channel);
954 if (ret < 0) {
955 goto error;
956 }
957 /* Update the stream global counter */
958 ksess->stream_count_global += ret;
959
960 /*
961 * Have we already sent fds to the consumer? If yes, it
962 * means that tracing is started so it is safe to send
963 * our updated stream fds.
964 */
965 if (ksess->consumer_fds_sent != 1
966 || ksess->consumer == NULL) {
967 ret = -1;
968 goto error;
969 }
970
971 rcu_read_lock();
972 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
973 &iter.iter, socket, node.node) {
974 pthread_mutex_lock(socket->lock);
975 ret = kernel_consumer_send_channel_stream(socket,
976 channel, ksess,
977 session->output_traces ? 1 : 0);
978 pthread_mutex_unlock(socket->lock);
979 if (ret < 0) {
e7fe706f 980 rcu_read_unlock();
4a15001e 981 goto error;
7a485870 982 }
7a485870 983 }
4a15001e 984 rcu_read_unlock();
7a485870 985 }
54d01ffb 986 session_unlock(session);
7a485870 987 }
54d01ffb 988 session_unlock_list();
b3c750d2 989 return ret;
7a485870 990
b3c750d2 991error:
54d01ffb
DG
992 session_unlock(session);
993 session_unlock_list();
7a485870
DG
994 return ret;
995}
996
487cf67c 997/*
ffe60014
DG
998 * For each tracing session, update newly registered apps. The session list
999 * lock MUST be acquired before calling this.
487cf67c
DG
1000 */
1001static void update_ust_app(int app_sock)
1002{
1003 struct ltt_session *sess, *stmp;
1004
fdadac08
DG
1005 /* Consumer is in an ERROR state. Stop any application update. */
1006 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
1007 /* Stop the update process since the consumer is dead. */
1008 return;
1009 }
1010
487cf67c
DG
1011 /* For all tracing session(s) */
1012 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
a9ad0c8f
MD
1013 struct ust_app *app;
1014
4ee14516 1015 session_lock(sess);
a9ad0c8f
MD
1016 if (!sess->ust_session) {
1017 goto unlock_session;
1018 }
1019
1020 rcu_read_lock();
1021 assert(app_sock >= 0);
1022 app = ust_app_find_by_sock(app_sock);
1023 if (app == NULL) {
1024 /*
1025 * Application can be unregistered before so
1026 * this is possible hence simply stopping the
1027 * update.
1028 */
1029 DBG3("UST app update failed to find app sock %d",
1030 app_sock);
1031 goto unlock_rcu;
421cb601 1032 }
a9ad0c8f
MD
1033 ust_app_global_update(sess->ust_session, app);
1034 unlock_rcu:
1035 rcu_read_unlock();
1036 unlock_session:
4ee14516 1037 session_unlock(sess);
487cf67c
DG
1038 }
1039}
1040
7a485870 1041/*
d063d709 1042 * This thread manage event coming from the kernel.
7a485870 1043 *
d063d709
DG
1044 * Features supported in this thread:
1045 * -) CPU Hotplug
7a485870
DG
1046 */
1047static void *thread_manage_kernel(void *data)
1048{
139ac872 1049 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 1050 uint32_t revents, nb_fd;
7a485870 1051 char tmp;
5eb91c98 1052 struct lttng_poll_event events;
7a485870 1053
6993eeb3 1054 DBG("[thread] Thread manage kernel started");
7a485870 1055
6c71277b 1056 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
927ca06a 1057
d5d63bf1
DG
1058 /*
1059 * This first step of the while is to clean this structure which could free
6d737ce4 1060 * non NULL pointers so initialize it before the loop.
d5d63bf1 1061 */
6d737ce4 1062 lttng_poll_init(&events);
d5d63bf1 1063
e547b070 1064 if (testpoint(sessiond_thread_manage_kernel)) {
6993eeb3
CB
1065 goto error_testpoint;
1066 }
8ac94142 1067
840cb59c 1068 health_code_update();
44a5e5eb 1069
e547b070 1070 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
d21b0d71 1071 goto error_testpoint;
6993eeb3
CB
1072 }
1073
7a485870 1074 while (1) {
840cb59c 1075 health_code_update();
44a5e5eb 1076
7a485870 1077 if (update_poll_flag == 1) {
d21b0d71
DG
1078 /* Clean events object. We are about to populate it again. */
1079 lttng_poll_clean(&events);
1080
d0b96690 1081 ret = sessiond_set_thread_pollset(&events, 2);
d21b0d71
DG
1082 if (ret < 0) {
1083 goto error_poll_create;
1084 }
1085
1086 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
1087 if (ret < 0) {
1088 goto error;
1089 }
5f822d0a 1090
d21b0d71 1091 /* This will add the available kernel channel if any. */
5eb91c98
DG
1092 ret = update_kernel_poll(&events);
1093 if (ret < 0) {
7a485870
DG
1094 goto error;
1095 }
1096 update_poll_flag = 0;
1097 }
1098
7fa2082e 1099 DBG("Thread kernel polling");
7a485870
DG
1100
1101 /* Poll infinite value of time */
88f2b785 1102 restart:
a78af745 1103 health_poll_entry();
5eb91c98 1104 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1105 DBG("Thread kernel return from poll on %d fds",
1106 LTTNG_POLL_GETNB(&events));
a78af745 1107 health_poll_exit();
7a485870 1108 if (ret < 0) {
88f2b785
MD
1109 /*
1110 * Restart interrupted system call.
1111 */
1112 if (errno == EINTR) {
1113 goto restart;
1114 }
7a485870
DG
1115 goto error;
1116 } else if (ret == 0) {
1117 /* Should not happen since timeout is infinite */
85611738
DG
1118 ERR("Return value of poll is 0 with an infinite timeout.\n"
1119 "This should not have happened! Continuing...");
7a485870
DG
1120 continue;
1121 }
1122
0d9c5d77
DG
1123 nb_fd = ret;
1124
5eb91c98
DG
1125 for (i = 0; i < nb_fd; i++) {
1126 /* Fetch once the poll data */
1127 revents = LTTNG_POLL_GETEV(&events, i);
1128 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 1129
840cb59c 1130 health_code_update();
44a5e5eb 1131
fd20dac9
MD
1132 if (!revents) {
1133 /* No activity for this FD (poll implementation). */
1134 continue;
1135 }
1136
5eb91c98 1137 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1138 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1139 if (ret) {
139ac872
MD
1140 err = 0;
1141 goto exit;
5eb91c98 1142 }
7a485870 1143
5eb91c98 1144 /* Check for data on kernel pipe */
03e43155
MD
1145 if (revents & LPOLLIN) {
1146 if (pollfd == kernel_poll_pipe[0]) {
1147 (void) lttng_read(kernel_poll_pipe[0],
1148 &tmp, 1);
1149 /*
1150 * Ret value is useless here, if this pipe gets any actions an
1151 * update is required anyway.
1152 */
1153 update_poll_flag = 1;
1154 continue;
1155 } else {
1156 /*
1157 * New CPU detected by the kernel. Adding kernel stream to
1158 * kernel session and updating the kernel consumer
1159 */
2bdd86d4 1160 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
1161 if (ret < 0) {
1162 continue;
1163 }
1164 break;
7a485870 1165 }
03e43155
MD
1166 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1167 update_poll_flag = 1;
1168 continue;
1169 } else {
1170 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1171 goto error;
7a485870
DG
1172 }
1173 }
1174 }
1175
139ac872 1176exit:
7a485870 1177error:
5eb91c98 1178 lttng_poll_clean(&events);
76d7553f 1179error_poll_create:
6993eeb3 1180error_testpoint:
6620da75
DG
1181 utils_close_pipe(kernel_poll_pipe);
1182 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 1183 if (err) {
840cb59c 1184 health_error();
139ac872 1185 ERR("Health error occurred in %s", __func__);
6620da75
DG
1186 WARN("Kernel thread died unexpectedly. "
1187 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 1188 }
8782cc74 1189 health_unregister(health_sessiond);
76d7553f 1190 DBG("Kernel thread dying");
7a485870
DG
1191 return NULL;
1192}
1193
a23ec3a7
DG
1194/*
1195 * Signal pthread condition of the consumer data that the thread.
1196 */
1197static void signal_consumer_condition(struct consumer_data *data, int state)
1198{
1199 pthread_mutex_lock(&data->cond_mutex);
1200
1201 /*
1202 * The state is set before signaling. It can be any value, it's the waiter
1203 * job to correctly interpret this condition variable associated to the
1204 * consumer pthread_cond.
1205 *
1206 * A value of 0 means that the corresponding thread of the consumer data
1207 * was not started. 1 indicates that the thread has started and is ready
1208 * for action. A negative value means that there was an error during the
1209 * thread bootstrap.
1210 */
1211 data->consumer_thread_is_ready = state;
1212 (void) pthread_cond_signal(&data->cond);
1213
1214 pthread_mutex_unlock(&data->cond_mutex);
1215}
1216
1d4b027a 1217/*
3bd1e081 1218 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 1219 */
3bd1e081 1220static void *thread_manage_consumer(void *data)
1d4b027a 1221{
42fc1d0b 1222 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
5eb91c98 1223 uint32_t revents, nb_fd;
1d4b027a 1224 enum lttcomm_return_code code;
5eb91c98 1225 struct lttng_poll_event events;
3bd1e081 1226 struct consumer_data *consumer_data = data;
1d4b027a 1227
3bd1e081 1228 DBG("[thread] Manage consumer started");
1d4b027a 1229
34c1e15a
MD
1230 rcu_register_thread();
1231 rcu_thread_online();
1232
6c71277b 1233 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
927ca06a 1234
855060f8 1235 health_code_update();
9449cc75 1236
5eb91c98 1237 /*
331744e3
JD
1238 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1239 * metadata_sock. Nothing more will be added to this poll set.
5eb91c98 1240 */
331744e3 1241 ret = sessiond_set_thread_pollset(&events, 3);
5eb91c98 1242 if (ret < 0) {
76d7553f 1243 goto error_poll;
5eb91c98 1244 }
273ea72c 1245
edb8b045
DG
1246 /*
1247 * The error socket here is already in a listening state which was done
1248 * just before spawning this thread to avoid a race between the consumer
1249 * daemon exec trying to connect and the listen() call.
1250 */
3bd1e081 1251 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
1252 if (ret < 0) {
1253 goto error;
1254 }
1255
840cb59c 1256 health_code_update();
44a5e5eb 1257
331744e3 1258 /* Infinite blocking call, waiting for transmission */
88f2b785 1259restart:
a78af745 1260 health_poll_entry();
8ac94142 1261
e547b070 1262 if (testpoint(sessiond_thread_manage_consumer)) {
6993eeb3
CB
1263 goto error;
1264 }
8ac94142 1265
5eb91c98 1266 ret = lttng_poll_wait(&events, -1);
a78af745 1267 health_poll_exit();
273ea72c 1268 if (ret < 0) {
88f2b785
MD
1269 /*
1270 * Restart interrupted system call.
1271 */
1272 if (errno == EINTR) {
1273 goto restart;
1274 }
273ea72c
DG
1275 goto error;
1276 }
1277
0d9c5d77
DG
1278 nb_fd = ret;
1279
5eb91c98
DG
1280 for (i = 0; i < nb_fd; i++) {
1281 /* Fetch once the poll data */
1282 revents = LTTNG_POLL_GETEV(&events, i);
1283 pollfd = LTTNG_POLL_GETFD(&events, i);
1284
840cb59c 1285 health_code_update();
44a5e5eb 1286
fd20dac9
MD
1287 if (!revents) {
1288 /* No activity for this FD (poll implementation). */
1289 continue;
1290 }
1291
5eb91c98 1292 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1293 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1294 if (ret) {
139ac872
MD
1295 err = 0;
1296 goto exit;
5eb91c98
DG
1297 }
1298
1299 /* Event on the registration socket */
3bd1e081 1300 if (pollfd == consumer_data->err_sock) {
03e43155
MD
1301 if (revents & LPOLLIN) {
1302 continue;
1303 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1304 ERR("consumer err socket poll error");
5eb91c98 1305 goto error;
03e43155
MD
1306 } else {
1307 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1308 goto error;
5eb91c98
DG
1309 }
1310 }
273ea72c
DG
1311 }
1312
3bd1e081 1313 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
1314 if (sock < 0) {
1315 goto error;
1316 }
1317
b662582b
DG
1318 /*
1319 * Set the CLOEXEC flag. Return code is useless because either way, the
1320 * show must go on.
1321 */
1322 (void) utils_set_fd_cloexec(sock);
1323
840cb59c 1324 health_code_update();
44a5e5eb 1325
3bd1e081 1326 DBG2("Receiving code from consumer err_sock");
ee0b0061 1327
712ea556 1328 /* Getting status code from kconsumerd */
54d01ffb
DG
1329 ret = lttcomm_recv_unix_sock(sock, &code,
1330 sizeof(enum lttcomm_return_code));
1d4b027a
DG
1331 if (ret <= 0) {
1332 goto error;
1333 }
1334
840cb59c 1335 health_code_update();
f73fabfd 1336 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
331744e3 1337 /* Connect both socket, command and metadata. */
3bd1e081
MD
1338 consumer_data->cmd_sock =
1339 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
4ce514c4 1340 consumer_data->metadata_fd =
331744e3 1341 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
92db7cdc
DG
1342 if (consumer_data->cmd_sock < 0
1343 || consumer_data->metadata_fd < 0) {
331744e3 1344 PERROR("consumer connect cmd socket");
a23ec3a7
DG
1345 /* On error, signal condition and quit. */
1346 signal_consumer_condition(consumer_data, -1);
1d4b027a
DG
1347 goto error;
1348 }
9363801e 1349 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
331744e3
JD
1350 /* Create metadata socket lock. */
1351 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1352 if (consumer_data->metadata_sock.lock == NULL) {
1353 PERROR("zmalloc pthread mutex");
331744e3
JD
1354 goto error;
1355 }
1356 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1357
a23ec3a7 1358 signal_consumer_condition(consumer_data, 1);
331744e3
JD
1359 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1360 DBG("Consumer metadata socket ready (fd: %d)",
4ce514c4 1361 consumer_data->metadata_fd);
1d4b027a 1362 } else {
3bd1e081 1363 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
1364 lttcomm_get_readable_code(-code));
1365 goto error;
1366 }
1367
331744e3 1368 /* Remove the consumerd error sock since we've established a connexion */
3bd1e081 1369 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1370 if (ret < 0) {
72079cae
DG
1371 goto error;
1372 }
1373
331744e3 1374 /* Add new accepted error socket. */
5eb91c98
DG
1375 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1376 if (ret < 0) {
72079cae 1377 goto error;
5eb91c98
DG
1378 }
1379
331744e3 1380 /* Add metadata socket that is successfully connected. */
4ce514c4 1381 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
331744e3
JD
1382 LPOLLIN | LPOLLRDHUP);
1383 if (ret < 0) {
1384 goto error;
1385 }
1386
840cb59c 1387 health_code_update();
44a5e5eb 1388
331744e3 1389 /* Infinite blocking call, waiting for transmission */
88f2b785 1390restart_poll:
331744e3 1391 while (1) {
42fc1d0b
DG
1392 health_code_update();
1393
1394 /* Exit the thread because the thread quit pipe has been triggered. */
1395 if (should_quit) {
1396 /* Not a health error. */
1397 err = 0;
1398 goto exit;
1399 }
1400
331744e3
JD
1401 health_poll_entry();
1402 ret = lttng_poll_wait(&events, -1);
1403 health_poll_exit();
1404 if (ret < 0) {
1405 /*
1406 * Restart interrupted system call.
1407 */
1408 if (errno == EINTR) {
1409 goto restart_poll;
1410 }
1411 goto error;
88f2b785 1412 }
72079cae 1413
331744e3 1414 nb_fd = ret;
0d9c5d77 1415
331744e3
JD
1416 for (i = 0; i < nb_fd; i++) {
1417 /* Fetch once the poll data */
1418 revents = LTTNG_POLL_GETEV(&events, i);
1419 pollfd = LTTNG_POLL_GETFD(&events, i);
5eb91c98 1420
331744e3 1421 health_code_update();
44a5e5eb 1422
fd20dac9
MD
1423 if (!revents) {
1424 /* No activity for this FD (poll implementation). */
1425 continue;
1426 }
1427
42fc1d0b
DG
1428 /*
1429 * Thread quit pipe has been triggered, flag that we should stop
1430 * but continue the current loop to handle potential data from
1431 * consumer.
1432 */
1433 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1434
331744e3
JD
1435 if (pollfd == sock) {
1436 /* Event on the consumerd socket */
03e43155
MD
1437 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1438 && !(revents & LPOLLIN)) {
331744e3
JD
1439 ERR("consumer err socket second poll error");
1440 goto error;
1441 }
1442 health_code_update();
1443 /* Wait for any kconsumerd error */
1444 ret = lttcomm_recv_unix_sock(sock, &code,
1445 sizeof(enum lttcomm_return_code));
1446 if (ret <= 0) {
1447 ERR("consumer closed the command socket");
1448 goto error;
1449 }
1450
1451 ERR("consumer return code : %s",
1452 lttcomm_get_readable_code(-code));
1453
1454 goto exit;
4ce514c4 1455 } else if (pollfd == consumer_data->metadata_fd) {
03e43155
MD
1456 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
1457 && !(revents & LPOLLIN)) {
1458 ERR("consumer err metadata socket second poll error");
1459 goto error;
1460 }
331744e3
JD
1461 /* UST metadata requests */
1462 ret = ust_consumer_metadata_request(
1463 &consumer_data->metadata_sock);
1464 if (ret < 0) {
1465 ERR("Handling metadata request");
1466 goto error;
1467 }
5eb91c98 1468 }
42fc1d0b 1469 /* No need for an else branch all FDs are tested prior. */
5eb91c98 1470 }
331744e3 1471 health_code_update();
5eb91c98
DG
1472 }
1473
139ac872 1474exit:
1d4b027a 1475error:
fdadac08
DG
1476 /*
1477 * We lock here because we are about to close the sockets and some other
92db7cdc
DG
1478 * thread might be using them so get exclusive access which will abort all
1479 * other consumer command by other threads.
fdadac08
DG
1480 */
1481 pthread_mutex_lock(&consumer_data->lock);
1482
5c827ce0
DG
1483 /* Immediately set the consumerd state to stopped */
1484 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1485 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1486 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1487 consumer_data->type == LTTNG_CONSUMER32_UST) {
1488 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1489 } else {
1490 /* Code flow error... */
1491 assert(0);
1492 }
1493
76d7553f
MD
1494 if (consumer_data->err_sock >= 0) {
1495 ret = close(consumer_data->err_sock);
1496 if (ret) {
1497 PERROR("close");
1498 }
a76cbd9f 1499 consumer_data->err_sock = -1;
76d7553f
MD
1500 }
1501 if (consumer_data->cmd_sock >= 0) {
1502 ret = close(consumer_data->cmd_sock);
1503 if (ret) {
1504 PERROR("close");
1505 }
a76cbd9f 1506 consumer_data->cmd_sock = -1;
76d7553f 1507 }
96544455
SS
1508 if (consumer_data->metadata_sock.fd_ptr &&
1509 *consumer_data->metadata_sock.fd_ptr >= 0) {
9363801e 1510 ret = close(*consumer_data->metadata_sock.fd_ptr);
331744e3
JD
1511 if (ret) {
1512 PERROR("close");
1513 }
1514 }
76d7553f
MD
1515 if (sock >= 0) {
1516 ret = close(sock);
1517 if (ret) {
1518 PERROR("close");
1519 }
1520 }
273ea72c 1521
3bd1e081
MD
1522 unlink(consumer_data->err_unix_sock_path);
1523 unlink(consumer_data->cmd_unix_sock_path);
fdadac08 1524 pthread_mutex_unlock(&consumer_data->lock);
92db7cdc 1525
fdadac08 1526 /* Cleanup metadata socket mutex. */
96544455
SS
1527 if (consumer_data->metadata_sock.lock) {
1528 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1529 free(consumer_data->metadata_sock.lock);
1530 }
5eb91c98 1531 lttng_poll_clean(&events);
76d7553f 1532error_poll:
139ac872 1533 if (err) {
840cb59c 1534 health_error();
139ac872
MD
1535 ERR("Health error occurred in %s", __func__);
1536 }
8782cc74 1537 health_unregister(health_sessiond);
76d7553f 1538 DBG("consumer thread cleanup completed");
0177d773 1539
34c1e15a
MD
1540 rcu_thread_offline();
1541 rcu_unregister_thread();
1542
5eb91c98 1543 return NULL;
099e26bd
DG
1544}
1545
099e26bd
DG
1546/*
1547 * This thread manage application communication.
1d4b027a
DG
1548 */
1549static void *thread_manage_apps(void *data)
099e26bd 1550{
139ac872 1551 int i, ret, pollfd, err = -1;
6cd525e8 1552 ssize_t size_ret;
5eb91c98 1553 uint32_t revents, nb_fd;
5eb91c98 1554 struct lttng_poll_event events;
099e26bd
DG
1555
1556 DBG("[thread] Manage application started");
1557
f6a9efaa
DG
1558 rcu_register_thread();
1559 rcu_thread_online();
1560
6c71277b 1561 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
927ca06a 1562
e547b070 1563 if (testpoint(sessiond_thread_manage_apps)) {
6993eeb3
CB
1564 goto error_testpoint;
1565 }
1566
840cb59c 1567 health_code_update();
44a5e5eb 1568
d0b96690 1569 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1570 if (ret < 0) {
76d7553f 1571 goto error_poll_create;
5eb91c98 1572 }
099e26bd 1573
5eb91c98
DG
1574 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1575 if (ret < 0) {
1576 goto error;
1577 }
099e26bd 1578
e547b070 1579 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
6993eeb3
CB
1580 goto error;
1581 }
8ac94142 1582
840cb59c 1583 health_code_update();
44a5e5eb 1584
5eb91c98 1585 while (1) {
7fa2082e 1586 DBG("Apps thread polling");
099e26bd
DG
1587
1588 /* Inifinite blocking call, waiting for transmission */
88f2b785 1589 restart:
a78af745 1590 health_poll_entry();
5eb91c98 1591 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
1592 DBG("Apps thread return from poll on %d fds",
1593 LTTNG_POLL_GETNB(&events));
a78af745 1594 health_poll_exit();
099e26bd 1595 if (ret < 0) {
88f2b785
MD
1596 /*
1597 * Restart interrupted system call.
1598 */
1599 if (errno == EINTR) {
1600 goto restart;
1601 }
099e26bd
DG
1602 goto error;
1603 }
1604
0d9c5d77
DG
1605 nb_fd = ret;
1606
5eb91c98
DG
1607 for (i = 0; i < nb_fd; i++) {
1608 /* Fetch once the poll data */
1609 revents = LTTNG_POLL_GETEV(&events, i);
1610 pollfd = LTTNG_POLL_GETFD(&events, i);
1611
840cb59c 1612 health_code_update();
44a5e5eb 1613
fd20dac9
MD
1614 if (!revents) {
1615 /* No activity for this FD (poll implementation). */
1616 continue;
1617 }
1618
5eb91c98 1619 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1620 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1621 if (ret) {
139ac872
MD
1622 err = 0;
1623 goto exit;
5eb91c98 1624 }
099e26bd 1625
5eb91c98
DG
1626 /* Inspect the apps cmd pipe */
1627 if (pollfd == apps_cmd_pipe[0]) {
03e43155 1628 if (revents & LPOLLIN) {
d0b96690
DG
1629 int sock;
1630
5eb91c98 1631 /* Empty pipe */
6cd525e8
MD
1632 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1633 if (size_ret < sizeof(sock)) {
76d7553f 1634 PERROR("read apps cmd pipe");
5eb91c98
DG
1635 goto error;
1636 }
099e26bd 1637
840cb59c 1638 health_code_update();
44a5e5eb 1639
ffe60014 1640 /*
03e43155
MD
1641 * Since this is a command socket (write then read),
1642 * we only monitor the error events of the socket.
ffe60014 1643 */
d0b96690
DG
1644 ret = lttng_poll_add(&events, sock,
1645 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1646 if (ret < 0) {
5eb91c98 1647 goto error;
e0c7ec2b 1648 }
acc7b41b 1649
d0b96690 1650 DBG("Apps with sock %d added to poll set", sock);
03e43155
MD
1651 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1652 ERR("Apps command pipe error");
1653 goto error;
1654 } else {
1655 ERR("Unknown poll events %u for sock %d", revents, pollfd);
1656 goto error;
0177d773 1657 }
5eb91c98
DG
1658 } else {
1659 /*
54d01ffb
DG
1660 * At this point, we know that a registered application made
1661 * the event at poll_wait.
5eb91c98
DG
1662 */
1663 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1664 /* Removing from the poll set */
1665 ret = lttng_poll_del(&events, pollfd);
1666 if (ret < 0) {
1667 goto error;
1668 }
099e26bd 1669
b9d9b220 1670 /* Socket closed on remote end. */
56fff090 1671 ust_app_unregister(pollfd);
03e43155
MD
1672 } else {
1673 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1674 goto error;
5eb91c98 1675 }
099e26bd 1676 }
44a5e5eb 1677
840cb59c 1678 health_code_update();
099e26bd 1679 }
099e26bd
DG
1680 }
1681
139ac872 1682exit:
099e26bd 1683error:
5eb91c98 1684 lttng_poll_clean(&events);
76d7553f 1685error_poll_create:
6993eeb3 1686error_testpoint:
6620da75
DG
1687 utils_close_pipe(apps_cmd_pipe);
1688 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1689
1690 /*
1691 * We don't clean the UST app hash table here since already registered
1692 * applications can still be controlled so let them be until the session
1693 * daemon dies or the applications stop.
1694 */
1695
139ac872 1696 if (err) {
840cb59c 1697 health_error();
139ac872
MD
1698 ERR("Health error occurred in %s", __func__);
1699 }
8782cc74 1700 health_unregister(health_sessiond);
76d7553f 1701 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1702 rcu_thread_offline();
1703 rcu_unregister_thread();
099e26bd
DG
1704 return NULL;
1705}
1706
d0b96690 1707/*
d88aee68
DG
1708 * Send a socket to a thread This is called from the dispatch UST registration
1709 * thread once all sockets are set for the application.
d0b96690 1710 *
b85dc84c
DG
1711 * The sock value can be invalid, we don't really care, the thread will handle
1712 * it and make the necessary cleanup if so.
1713 *
d0b96690
DG
1714 * On success, return 0 else a negative value being the errno message of the
1715 * write().
1716 */
d88aee68 1717static int send_socket_to_thread(int fd, int sock)
d0b96690 1718{
6cd525e8 1719 ssize_t ret;
d0b96690 1720
b85dc84c
DG
1721 /*
1722 * It's possible that the FD is set as invalid with -1 concurrently just
1723 * before calling this function being a shutdown state of the thread.
1724 */
1725 if (fd < 0) {
1726 ret = -EBADF;
1727 goto error;
1728 }
d0b96690 1729
6cd525e8
MD
1730 ret = lttng_write(fd, &sock, sizeof(sock));
1731 if (ret < sizeof(sock)) {
d88aee68 1732 PERROR("write apps pipe %d", fd);
d0b96690
DG
1733 if (ret < 0) {
1734 ret = -errno;
1735 }
1736 goto error;
1737 }
1738
1739 /* All good. Don't send back the write positive ret value. */
1740 ret = 0;
1741error:
6cd525e8 1742 return (int) ret;
d0b96690
DG
1743}
1744
f45e313d
DG
1745/*
1746 * Sanitize the wait queue of the dispatch registration thread meaning removing
1747 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1748 * notify socket is never received.
1749 */
1750static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1751{
1752 int ret, nb_fd = 0, i;
1753 unsigned int fd_added = 0;
1754 struct lttng_poll_event events;
1755 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1756
1757 assert(wait_queue);
1758
1759 lttng_poll_init(&events);
1760
1761 /* Just skip everything for an empty queue. */
1762 if (!wait_queue->count) {
1763 goto end;
1764 }
1765
1766 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1767 if (ret < 0) {
1768 goto error_create;
1769 }
1770
1771 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1772 &wait_queue->head, head) {
1773 assert(wait_node->app);
1774 ret = lttng_poll_add(&events, wait_node->app->sock,
1775 LPOLLHUP | LPOLLERR);
1776 if (ret < 0) {
1777 goto error;
1778 }
1779
1780 fd_added = 1;
1781 }
1782
1783 if (!fd_added) {
1784 goto end;
1785 }
1786
1787 /*
1788 * Poll but don't block so we can quickly identify the faulty events and
1789 * clean them afterwards from the wait queue.
1790 */
1791 ret = lttng_poll_wait(&events, 0);
1792 if (ret < 0) {
1793 goto error;
1794 }
1795 nb_fd = ret;
1796
1797 for (i = 0; i < nb_fd; i++) {
1798 /* Get faulty FD. */
1799 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1800 int pollfd = LTTNG_POLL_GETFD(&events, i);
1801
fd20dac9
MD
1802 if (!revents) {
1803 /* No activity for this FD (poll implementation). */
1804 continue;
1805 }
1806
f45e313d
DG
1807 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1808 &wait_queue->head, head) {
1809 if (pollfd == wait_node->app->sock &&
1810 (revents & (LPOLLHUP | LPOLLERR))) {
1811 cds_list_del(&wait_node->head);
1812 wait_queue->count--;
1813 ust_app_destroy(wait_node->app);
1814 free(wait_node);
48b40bcf
JG
1815 /*
1816 * Silence warning of use-after-free in
1817 * cds_list_for_each_entry_safe which uses
1818 * __typeof__(*wait_node).
1819 */
1820 wait_node = NULL;
f45e313d 1821 break;
03e43155
MD
1822 } else {
1823 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1824 goto error;
f45e313d
DG
1825 }
1826 }
1827 }
1828
1829 if (nb_fd > 0) {
1830 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1831 }
1832
1833end:
1834 lttng_poll_clean(&events);
1835 return;
1836
1837error:
1838 lttng_poll_clean(&events);
1839error_create:
1840 ERR("Unable to sanitize wait queue");
1841 return;
1842}
1843
099e26bd
DG
1844/*
1845 * Dispatch request from the registration threads to the application
1846 * communication thread.
1847 */
1848static void *thread_dispatch_ust_registration(void *data)
1849{
12e2b881 1850 int ret, err = -1;
8bdee6e2 1851 struct cds_wfcq_node *node;
099e26bd 1852 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1853 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1854 struct ust_reg_wait_queue wait_queue = {
1855 .count = 0,
1856 };
d0b96690 1857
967e3668
MD
1858 rcu_register_thread();
1859
6c71277b 1860 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
12e2b881 1861
9ad42ec1
MD
1862 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1863 goto error_testpoint;
1864 }
1865
12e2b881
MD
1866 health_code_update();
1867
f45e313d 1868 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1869
1870 DBG("[thread] Dispatch UST command started");
1871
26c9d55e 1872 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1873 health_code_update();
1874
099e26bd
DG
1875 /* Atomically prepare the queue futex */
1876 futex_nto1_prepare(&ust_cmd_queue.futex);
1877
1878 do {
d0b96690 1879 struct ust_app *app = NULL;
7972aab2 1880 ust_cmd = NULL;
d0b96690 1881
f45e313d
DG
1882 /*
1883 * Make sure we don't have node(s) that have hung up before receiving
1884 * the notify socket. This is to clean the list in order to avoid
1885 * memory leaks from notify socket that are never seen.
1886 */
1887 sanitize_wait_queue(&wait_queue);
1888
12e2b881 1889 health_code_update();
099e26bd 1890 /* Dequeue command for registration */
8bdee6e2 1891 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
099e26bd 1892 if (node == NULL) {
00a17c97 1893 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1894 /* Continue thread execution */
1895 break;
1896 }
1897
1898 ust_cmd = caa_container_of(node, struct ust_command, node);
1899
2f50c8a3
DG
1900 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1901 " gid:%d sock:%d name:%s (version %d.%d)",
1902 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1903 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1904 ust_cmd->sock, ust_cmd->reg_msg.name,
1905 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1906
1907 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1908 wait_node = zmalloc(sizeof(*wait_node));
1909 if (!wait_node) {
1910 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1911 ret = close(ust_cmd->sock);
1912 if (ret < 0) {
1913 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1914 }
51dec90d 1915 lttng_fd_put(LTTNG_FD_APPS, 1);
7972aab2 1916 free(ust_cmd);
d0b96690
DG
1917 goto error;
1918 }
1919 CDS_INIT_LIST_HEAD(&wait_node->head);
1920
1921 /* Create application object if socket is CMD. */
1922 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1923 ust_cmd->sock);
1924 if (!wait_node->app) {
1925 ret = close(ust_cmd->sock);
1926 if (ret < 0) {
1927 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1928 }
51dec90d 1929 lttng_fd_put(LTTNG_FD_APPS, 1);
d88aee68 1930 free(wait_node);
7972aab2 1931 free(ust_cmd);
d0b96690
DG
1932 continue;
1933 }
1934 /*
1935 * Add application to the wait queue so we can set the notify
1936 * socket before putting this object in the global ht.
1937 */
f45e313d
DG
1938 cds_list_add(&wait_node->head, &wait_queue.head);
1939 wait_queue.count++;
d0b96690 1940
7972aab2 1941 free(ust_cmd);
d0b96690
DG
1942 /*
1943 * We have to continue here since we don't have the notify
1944 * socket and the application MUST be added to the hash table
1945 * only at that moment.
1946 */
1947 continue;
1948 } else {
1949 /*
1950 * Look for the application in the local wait queue and set the
1951 * notify socket if found.
1952 */
d88aee68 1953 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1954 &wait_queue.head, head) {
12e2b881 1955 health_code_update();
d0b96690
DG
1956 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1957 wait_node->app->notify_sock = ust_cmd->sock;
1958 cds_list_del(&wait_node->head);
f45e313d 1959 wait_queue.count--;
d0b96690
DG
1960 app = wait_node->app;
1961 free(wait_node);
1962 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1963 break;
1964 }
1965 }
020d7f60
DG
1966
1967 /*
1968 * With no application at this stage the received socket is
1969 * basically useless so close it before we free the cmd data
1970 * structure for good.
1971 */
1972 if (!app) {
1973 ret = close(ust_cmd->sock);
1974 if (ret < 0) {
1975 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1976 }
51dec90d 1977 lttng_fd_put(LTTNG_FD_APPS, 1);
020d7f60 1978 }
7972aab2 1979 free(ust_cmd);
d0b96690
DG
1980 }
1981
1982 if (app) {
d0b96690
DG
1983 /*
1984 * @session_lock_list
1985 *
1986 * Lock the global session list so from the register up to the
1987 * registration done message, no thread can see the application
1988 * and change its state.
1989 */
1990 session_lock_list();
1991 rcu_read_lock();
d88aee68 1992
d0b96690
DG
1993 /*
1994 * Add application to the global hash table. This needs to be
1995 * done before the update to the UST registry can locate the
1996 * application.
1997 */
1998 ust_app_add(app);
d88aee68
DG
1999
2000 /* Set app version. This call will print an error if needed. */
2001 (void) ust_app_version(app);
2002
2003 /* Send notify socket through the notify pipe. */
2004 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
2005 app->notify_sock);
2006 if (ret < 0) {
2007 rcu_read_unlock();
2008 session_unlock_list();
b85dc84c
DG
2009 /*
2010 * No notify thread, stop the UST tracing. However, this is
2011 * not an internal error of the this thread thus setting
2012 * the health error code to a normal exit.
2013 */
2014 err = 0;
d88aee68 2015 goto error;
6620da75 2016 }
d88aee68 2017
d0b96690
DG
2018 /*
2019 * Update newly registered application with the tracing
2020 * registry info already enabled information.
2021 */
2022 update_ust_app(app->sock);
d88aee68
DG
2023
2024 /*
2025 * Don't care about return value. Let the manage apps threads
2026 * handle app unregistration upon socket close.
2027 */
fb45065e 2028 (void) ust_app_register_done(app);
d88aee68
DG
2029
2030 /*
2031 * Even if the application socket has been closed, send the app
2032 * to the thread and unregistration will take place at that
2033 * place.
2034 */
2035 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 2036 if (ret < 0) {
d88aee68
DG
2037 rcu_read_unlock();
2038 session_unlock_list();
b85dc84c
DG
2039 /*
2040 * No apps. thread, stop the UST tracing. However, this is
2041 * not an internal error of the this thread thus setting
2042 * the health error code to a normal exit.
2043 */
2044 err = 0;
d88aee68 2045 goto error;
d0b96690 2046 }
d88aee68 2047
d0b96690
DG
2048 rcu_read_unlock();
2049 session_unlock_list();
099e26bd 2050 }
099e26bd
DG
2051 } while (node != NULL);
2052
12e2b881 2053 health_poll_entry();
099e26bd
DG
2054 /* Futex wait on queue. Blocking call on futex() */
2055 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 2056 health_poll_exit();
099e26bd 2057 }
12e2b881
MD
2058 /* Normal exit, no error */
2059 err = 0;
099e26bd
DG
2060
2061error:
d88aee68
DG
2062 /* Clean up wait queue. */
2063 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 2064 &wait_queue.head, head) {
d88aee68 2065 cds_list_del(&wait_node->head);
f45e313d 2066 wait_queue.count--;
d88aee68
DG
2067 free(wait_node);
2068 }
2069
772b8f4d
MD
2070 /* Empty command queue. */
2071 for (;;) {
2072 /* Dequeue command for registration */
2073 node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
2074 if (node == NULL) {
2075 break;
2076 }
2077 ust_cmd = caa_container_of(node, struct ust_command, node);
2078 ret = close(ust_cmd->sock);
2079 if (ret < 0) {
2080 PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
2081 }
2082 lttng_fd_put(LTTNG_FD_APPS, 1);
2083 free(ust_cmd);
2084 }
2085
9ad42ec1 2086error_testpoint:
099e26bd 2087 DBG("Dispatch thread dying");
12e2b881
MD
2088 if (err) {
2089 health_error();
2090 ERR("Health error occurred in %s", __func__);
2091 }
8782cc74 2092 health_unregister(health_sessiond);
967e3668 2093 rcu_unregister_thread();
099e26bd
DG
2094 return NULL;
2095}
2096
2097/*
2098 * This thread manage application registration.
2099 */
2100static void *thread_registration_apps(void *data)
1d4b027a 2101{
139ac872 2102 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
2103 uint32_t revents, nb_fd;
2104 struct lttng_poll_event events;
099e26bd
DG
2105 /*
2106 * Get allocated in this thread, enqueued to a global queue, dequeued and
2107 * freed in the manage apps thread.
2108 */
2109 struct ust_command *ust_cmd = NULL;
1d4b027a 2110
099e26bd 2111 DBG("[thread] Manage application registration started");
1d4b027a 2112
6c71277b 2113 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
927ca06a 2114
e547b070 2115 if (testpoint(sessiond_thread_registration_apps)) {
6993eeb3
CB
2116 goto error_testpoint;
2117 }
8ac94142 2118
1d4b027a
DG
2119 ret = lttcomm_listen_unix_sock(apps_sock);
2120 if (ret < 0) {
76d7553f 2121 goto error_listen;
1d4b027a
DG
2122 }
2123
5eb91c98
DG
2124 /*
2125 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2126 * more will be added to this poll set.
2127 */
d0b96690 2128 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 2129 if (ret < 0) {
76d7553f 2130 goto error_create_poll;
5eb91c98 2131 }
273ea72c 2132
5eb91c98
DG
2133 /* Add the application registration socket */
2134 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
2135 if (ret < 0) {
76d7553f 2136 goto error_poll_add;
5eb91c98 2137 }
273ea72c 2138
1d4b027a 2139 /* Notify all applications to register */
0fdd1e2c
DG
2140 ret = notify_ust_apps(1);
2141 if (ret < 0) {
2142 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
2143 "Execution continues but there might be problem for already\n"
2144 "running applications that wishes to register.");
0fdd1e2c 2145 }
1d4b027a
DG
2146
2147 while (1) {
2148 DBG("Accepting application registration");
273ea72c
DG
2149
2150 /* Inifinite blocking call, waiting for transmission */
88f2b785 2151 restart:
a78af745 2152 health_poll_entry();
5eb91c98 2153 ret = lttng_poll_wait(&events, -1);
a78af745 2154 health_poll_exit();
273ea72c 2155 if (ret < 0) {
88f2b785
MD
2156 /*
2157 * Restart interrupted system call.
2158 */
2159 if (errno == EINTR) {
2160 goto restart;
2161 }
273ea72c
DG
2162 goto error;
2163 }
2164
0d9c5d77
DG
2165 nb_fd = ret;
2166
5eb91c98 2167 for (i = 0; i < nb_fd; i++) {
840cb59c 2168 health_code_update();
139ac872 2169
5eb91c98
DG
2170 /* Fetch once the poll data */
2171 revents = LTTNG_POLL_GETEV(&events, i);
2172 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 2173
fd20dac9
MD
2174 if (!revents) {
2175 /* No activity for this FD (poll implementation). */
2176 continue;
2177 }
2178
5eb91c98 2179 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 2180 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 2181 if (ret) {
139ac872
MD
2182 err = 0;
2183 goto exit;
90014c57 2184 }
1d4b027a 2185
5eb91c98
DG
2186 /* Event on the registration socket */
2187 if (pollfd == apps_sock) {
03e43155 2188 if (revents & LPOLLIN) {
5eb91c98
DG
2189 sock = lttcomm_accept_unix_sock(apps_sock);
2190 if (sock < 0) {
2191 goto error;
2192 }
099e26bd 2193
16c5c8fa
DG
2194 /*
2195 * Set socket timeout for both receiving and ending.
2196 * app_socket_timeout is in seconds, whereas
2197 * lttcomm_setsockopt_rcv_timeout and
2198 * lttcomm_setsockopt_snd_timeout expect msec as
2199 * parameter.
2200 */
28ce0ff2
JG
2201 if (app_socket_timeout >= 0) {
2202 (void) lttcomm_setsockopt_rcv_timeout(sock,
2203 app_socket_timeout * 1000);
2204 (void) lttcomm_setsockopt_snd_timeout(sock,
2205 app_socket_timeout * 1000);
2206 }
16c5c8fa 2207
b662582b
DG
2208 /*
2209 * Set the CLOEXEC flag. Return code is useless because
2210 * either way, the show must go on.
2211 */
2212 (void) utils_set_fd_cloexec(sock);
2213
5eb91c98 2214 /* Create UST registration command for enqueuing */
ba7f0ae5 2215 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 2216 if (ust_cmd == NULL) {
76d7553f 2217 PERROR("ust command zmalloc");
41ed8e47
MD
2218 ret = close(sock);
2219 if (ret) {
2220 PERROR("close");
2221 }
5eb91c98
DG
2222 goto error;
2223 }
1d4b027a 2224
5eb91c98
DG
2225 /*
2226 * Using message-based transmissions to ensure we don't
2227 * have to deal with partially received messages.
2228 */
4063050c
MD
2229 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
2230 if (ret < 0) {
2231 ERR("Exhausted file descriptors allowed for applications.");
2232 free(ust_cmd);
2233 ret = close(sock);
2234 if (ret) {
2235 PERROR("close");
2236 }
2237 sock = -1;
2238 continue;
2239 }
d88aee68 2240
840cb59c 2241 health_code_update();
d0b96690
DG
2242 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2243 if (ret < 0) {
5eb91c98 2244 free(ust_cmd);
d0b96690 2245 /* Close socket of the application. */
76d7553f
MD
2246 ret = close(sock);
2247 if (ret) {
2248 PERROR("close");
2249 }
4063050c 2250 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 2251 sock = -1;
5eb91c98
DG
2252 continue;
2253 }
840cb59c 2254 health_code_update();
099e26bd 2255
5eb91c98 2256 ust_cmd->sock = sock;
34a2494f 2257 sock = -1;
099e26bd 2258
5eb91c98
DG
2259 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2260 " gid:%d sock:%d name:%s (version %d.%d)",
2261 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2262 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2263 ust_cmd->sock, ust_cmd->reg_msg.name,
2264 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 2265
5eb91c98
DG
2266 /*
2267 * Lock free enqueue the registration request. The red pill
54d01ffb 2268 * has been taken! This apps will be part of the *system*.
5eb91c98 2269 */
8bdee6e2 2270 cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);
5eb91c98
DG
2271
2272 /*
2273 * Wake the registration queue futex. Implicit memory
8bdee6e2 2274 * barrier with the exchange in cds_wfcq_enqueue.
5eb91c98
DG
2275 */
2276 futex_nto1_wake(&ust_cmd_queue.futex);
03e43155
MD
2277 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2278 ERR("Register apps socket poll error");
2279 goto error;
2280 } else {
2281 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2282 goto error;
5eb91c98
DG
2283 }
2284 }
90014c57 2285 }
1d4b027a
DG
2286 }
2287
139ac872 2288exit:
1d4b027a 2289error:
0fdd1e2c
DG
2290 /* Notify that the registration thread is gone */
2291 notify_ust_apps(0);
2292
a4b35e07 2293 if (apps_sock >= 0) {
76d7553f
MD
2294 ret = close(apps_sock);
2295 if (ret) {
2296 PERROR("close");
2297 }
a4b35e07 2298 }
46c3f085 2299 if (sock >= 0) {
76d7553f
MD
2300 ret = close(sock);
2301 if (ret) {
2302 PERROR("close");
2303 }
4063050c 2304 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 2305 }
273ea72c 2306 unlink(apps_unix_sock_path);
0fdd1e2c 2307
76d7553f 2308error_poll_add:
5eb91c98 2309 lttng_poll_clean(&events);
76d7553f
MD
2310error_listen:
2311error_create_poll:
6993eeb3 2312error_testpoint:
76d7553f 2313 DBG("UST Registration thread cleanup complete");
9ad42ec1
MD
2314 if (err) {
2315 health_error();
2316 ERR("Health error occurred in %s", __func__);
2317 }
8782cc74 2318 health_unregister(health_sessiond);
5eb91c98 2319
1d4b027a
DG
2320 return NULL;
2321}
2322
8c0faa1d 2323/*
3bd1e081 2324 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 2325 * exec or it will fails.
8c0faa1d 2326 */
3bd1e081 2327static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 2328{
a23ec3a7 2329 int ret, clock_ret;
ee0b0061
DG
2330 struct timespec timeout;
2331
13a7bce3
JG
2332 /*
2333 * Make sure we set the readiness flag to 0 because we are NOT ready.
2334 * This access to consumer_thread_is_ready does not need to be
2335 * protected by consumer_data.cond_mutex (yet) since the consumer
2336 * management thread has not been started at this point.
2337 */
a23ec3a7 2338 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 2339
a23ec3a7
DG
2340 /* Setup pthread condition */
2341 ret = pthread_condattr_init(&consumer_data->condattr);
4a15001e 2342 if (ret) {
a23ec3a7
DG
2343 errno = ret;
2344 PERROR("pthread_condattr_init consumer data");
2345 goto error;
2346 }
2347
2348 /*
2349 * Set the monotonic clock in order to make sure we DO NOT jump in time
2350 * between the clock_gettime() call and the timedwait call. See bug #324
2351 * for a more details and how we noticed it.
2352 */
2353 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
4a15001e 2354 if (ret) {
a23ec3a7
DG
2355 errno = ret;
2356 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
2357 goto error;
2358 }
8c0faa1d 2359
a23ec3a7 2360 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
4a15001e 2361 if (ret) {
a23ec3a7
DG
2362 errno = ret;
2363 PERROR("pthread_cond_init consumer data");
2364 goto error;
2365 }
2366
1a1a34b4
MJ
2367 ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
2368 thread_manage_consumer, consumer_data);
4a15001e
MD
2369 if (ret) {
2370 errno = ret;
3bd1e081 2371 PERROR("pthread_create consumer");
ee0b0061 2372 ret = -1;
8c0faa1d
DG
2373 goto error;
2374 }
2375
a23ec3a7
DG
2376 /* We are about to wait on a pthread condition */
2377 pthread_mutex_lock(&consumer_data->cond_mutex);
2378
ee0b0061 2379 /* Get time for sem_timedwait absolute timeout */
389fbf04 2380 clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
a23ec3a7
DG
2381 /*
2382 * Set the timeout for the condition timed wait even if the clock gettime
2383 * call fails since we might loop on that call and we want to avoid to
2384 * increment the timeout too many times.
2385 */
2386 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2387
2388 /*
2389 * The following loop COULD be skipped in some conditions so this is why we
2390 * set ret to 0 in order to make sure at least one round of the loop is
2391 * done.
2392 */
2393 ret = 0;
2394
2395 /*
2396 * Loop until the condition is reached or when a timeout is reached. Note
2397 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2398 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2399 * possible. This loop does not take any chances and works with both of
2400 * them.
2401 */
2402 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2403 if (clock_ret < 0) {
2404 PERROR("clock_gettime spawn consumer");
2405 /* Infinite wait for the consumerd thread to be ready */
2406 ret = pthread_cond_wait(&consumer_data->cond,
2407 &consumer_data->cond_mutex);
2408 } else {
2409 ret = pthread_cond_timedwait(&consumer_data->cond,
2410 &consumer_data->cond_mutex, &timeout);
2411 }
ee0b0061 2412 }
8c0faa1d 2413
a23ec3a7
DG
2414 /* Release the pthread condition */
2415 pthread_mutex_unlock(&consumer_data->cond_mutex);
2416
2417 if (ret != 0) {
2418 errno = ret;
2419 if (ret == ETIMEDOUT) {
4282f9a3
DG
2420 int pth_ret;
2421
ee0b0061
DG
2422 /*
2423 * Call has timed out so we kill the kconsumerd_thread and return
2424 * an error.
2425 */
a23ec3a7
DG
2426 ERR("Condition timed out. The consumer thread was never ready."
2427 " Killing it");
4282f9a3
DG
2428 pth_ret = pthread_cancel(consumer_data->thread);
2429 if (pth_ret < 0) {
3bd1e081 2430 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
2431 }
2432 } else {
a23ec3a7 2433 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061 2434 }
4282f9a3
DG
2435 /* Caller is expecting a negative value on failure. */
2436 ret = -1;
ee0b0061
DG
2437 goto error;
2438 }
2439
3bd1e081
MD
2440 pthread_mutex_lock(&consumer_data->pid_mutex);
2441 if (consumer_data->pid == 0) {
a23ec3a7 2442 ERR("Consumerd did not start");
3bd1e081 2443 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2444 goto error;
2445 }
3bd1e081 2446 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2447
8c0faa1d
DG
2448 return 0;
2449
2450error:
2451 return ret;
2452}
2453
d9800920 2454/*
3bd1e081 2455 * Join consumer thread
d9800920 2456 */
3bd1e081 2457static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2458{
2459 void *status;
cf3af59e 2460
e8209f6b
DG
2461 /* Consumer pid must be a real one. */
2462 if (consumer_data->pid > 0) {
c617c0c6 2463 int ret;
3bd1e081 2464 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2465 if (ret) {
4a15001e 2466 PERROR("Error killing consumer daemon");
cf3af59e
MD
2467 return ret;
2468 }
3bd1e081 2469 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2470 } else {
2471 return 0;
2472 }
2473}
2474
8c0faa1d 2475/*
3bd1e081 2476 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2477 *
d063d709 2478 * Return pid if successful else -1.
8c0faa1d 2479 */
3bd1e081 2480static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2481{
2482 int ret;
2483 pid_t pid;
94c55f17 2484 const char *consumer_to_use;
53086306 2485 const char *verbosity;
94c55f17 2486 struct stat st;
8c0faa1d 2487
3bd1e081 2488 DBG("Spawning consumerd");
c49dc785 2489
8c0faa1d
DG
2490 pid = fork();
2491 if (pid == 0) {
2492 /*
3bd1e081 2493 * Exec consumerd.
8c0faa1d 2494 */
daee5345 2495 if (opt_verbose_consumer) {
53086306 2496 verbosity = "--verbose";
4421f712 2497 } else if (lttng_opt_quiet) {
53086306 2498 verbosity = "--quiet";
4421f712
DG
2499 } else {
2500 verbosity = "";
53086306 2501 }
4421f712 2502
3bd1e081
MD
2503 switch (consumer_data->type) {
2504 case LTTNG_CONSUMER_KERNEL:
94c55f17 2505 /*
c7704d57
DG
2506 * Find out which consumerd to execute. We will first try the
2507 * 64-bit path, then the sessiond's installation directory, and
2508 * fallback on the 32-bit one,
94c55f17 2509 */
63a799e8
AM
2510 DBG3("Looking for a kernel consumer at these locations:");
2511 DBG3(" 1) %s", consumerd64_bin);
2512 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2513 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2514 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2515 DBG3("Found location #1");
94c55f17 2516 consumer_to_use = consumerd64_bin;
94c55f17 2517 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2518 DBG3("Found location #2");
94c55f17 2519 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2520 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2521 DBG3("Found location #3");
eb1e0bd4 2522 consumer_to_use = consumerd32_bin;
94c55f17 2523 } else {
63a799e8 2524 DBG("Could not find any valid consumerd executable");
4282f9a3 2525 ret = -EINVAL;
94c55f17
AM
2526 break;
2527 }
2528 DBG("Using kernel consumer at: %s", consumer_to_use);
4282f9a3 2529 ret = execl(consumer_to_use,
94c55f17
AM
2530 "lttng-consumerd", verbosity, "-k",
2531 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2532 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2533 "--group", tracing_group_name,
94c55f17 2534 NULL);
3bd1e081 2535 break;
7753dea8
MD
2536 case LTTNG_CONSUMER64_UST:
2537 {
b1e0b6b6 2538 char *tmpnew = NULL;
8f4905da
MD
2539
2540 if (consumerd64_libdir[0] != '\0') {
2541 char *tmp;
2542 size_t tmplen;
2543
e8fa9fb0 2544 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
8f4905da
MD
2545 if (!tmp) {
2546 tmp = "";
2547 }
2548 tmplen = strlen("LD_LIBRARY_PATH=")
2549 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2550 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2551 if (!tmpnew) {
2552 ret = -ENOMEM;
2553 goto error;
2554 }
2555 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2556 strcat(tmpnew, consumerd64_libdir);
2557 if (tmp[0] != '\0') {
2558 strcat(tmpnew, ":");
2559 strcat(tmpnew, tmp);
2560 }
2561 ret = putenv(tmpnew);
2562 if (ret) {
2563 ret = -errno;
c6f76da9 2564 free(tmpnew);
8f4905da
MD
2565 goto error;
2566 }
2567 }
94c55f17 2568 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2569 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2570 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2571 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2572 "--group", tracing_group_name,
7753dea8 2573 NULL);
8f4905da
MD
2574 if (consumerd64_libdir[0] != '\0') {
2575 free(tmpnew);
2576 }
3bd1e081 2577 break;
7753dea8
MD
2578 }
2579 case LTTNG_CONSUMER32_UST:
2580 {
937dde8e 2581 char *tmpnew = NULL;
8f4905da
MD
2582
2583 if (consumerd32_libdir[0] != '\0') {
2584 char *tmp;
2585 size_t tmplen;
2586
e8fa9fb0 2587 tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
8f4905da
MD
2588 if (!tmp) {
2589 tmp = "";
2590 }
2591 tmplen = strlen("LD_LIBRARY_PATH=")
2592 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2593 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2594 if (!tmpnew) {
2595 ret = -ENOMEM;
2596 goto error;
2597 }
2598 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2599 strcat(tmpnew, consumerd32_libdir);
2600 if (tmp[0] != '\0') {
2601 strcat(tmpnew, ":");
2602 strcat(tmpnew, tmp);
2603 }
2604 ret = putenv(tmpnew);
2605 if (ret) {
2606 ret = -errno;
c6f76da9 2607 free(tmpnew);
8f4905da
MD
2608 goto error;
2609 }
2610 }
94c55f17 2611 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2612 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2613 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2614 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2615 "--group", tracing_group_name,
7753dea8 2616 NULL);
8f4905da
MD
2617 if (consumerd32_libdir[0] != '\0') {
2618 free(tmpnew);
2619 }
7753dea8
MD
2620 break;
2621 }
3bd1e081 2622 default:
76d7553f 2623 PERROR("unknown consumer type");
3bd1e081
MD
2624 exit(EXIT_FAILURE);
2625 }
8c0faa1d 2626 if (errno != 0) {
4282f9a3 2627 PERROR("Consumer execl()");
8c0faa1d 2628 }
4282f9a3 2629 /* Reaching this point, we got a failure on our execl(). */
8c0faa1d
DG
2630 exit(EXIT_FAILURE);
2631 } else if (pid > 0) {
2632 ret = pid;
8c0faa1d 2633 } else {
76d7553f 2634 PERROR("start consumer fork");
8c0faa1d 2635 ret = -errno;
8c0faa1d 2636 }
8f4905da 2637error:
8c0faa1d
DG
2638 return ret;
2639}
2640
693bd40b 2641/*
3bd1e081 2642 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2643 */
3bd1e081 2644static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2645{
c617c0c6 2646 int ret;
edb8b045
DG
2647
2648 /*
2649 * Set the listen() state on the socket since there is a possible race
2650 * between the exec() of the consumer daemon and this call if place in the
2651 * consumer thread. See bug #366 for more details.
2652 */
2653 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2654 if (ret < 0) {
2655 goto error;
2656 }
693bd40b 2657
3bd1e081
MD
2658 pthread_mutex_lock(&consumer_data->pid_mutex);
2659 if (consumer_data->pid != 0) {
2660 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2661 goto end;
2662 }
693bd40b 2663
3bd1e081 2664 ret = spawn_consumerd(consumer_data);
c49dc785 2665 if (ret < 0) {
3bd1e081
MD
2666 ERR("Spawning consumerd failed");
2667 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2668 goto error;
693bd40b 2669 }
c49dc785 2670
3bd1e081
MD
2671 /* Setting up the consumer_data pid */
2672 consumer_data->pid = ret;
48842b30 2673 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2674 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2675
3bd1e081
MD
2676 DBG2("Spawning consumer control thread");
2677 ret = spawn_consumer_thread(consumer_data);
693bd40b 2678 if (ret < 0) {
3bd1e081 2679 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2680 goto error;
2681 }
2682
c49dc785 2683end:
693bd40b
DG
2684 return 0;
2685
2686error:
331744e3 2687 /* Cleanup already created sockets on error. */
edb8b045 2688 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2689 int err;
2690
edb8b045
DG
2691 err = close(consumer_data->err_sock);
2692 if (err < 0) {
2693 PERROR("close consumer data error socket");
2694 }
2695 }
693bd40b
DG
2696 return ret;
2697}
2698
b73401da 2699/*
096102bd 2700 * Setup necessary data for kernel tracer action.
b73401da 2701 */
096102bd 2702static int init_kernel_tracer(void)
b73401da
DG
2703{
2704 int ret;
b73401da 2705
096102bd
DG
2706 /* Modprobe lttng kernel modules */
2707 ret = modprobe_lttng_control();
b73401da 2708 if (ret < 0) {
b73401da
DG
2709 goto error;
2710 }
2711
096102bd
DG
2712 /* Open debugfs lttng */
2713 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2714 if (kernel_tracer_fd < 0) {
2715 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2716 ret = -1;
2717 goto error_open;
54d01ffb
DG
2718 }
2719
2f77fc4b
DG
2720 /* Validate kernel version */
2721 ret = kernel_validate_version(kernel_tracer_fd);
2722 if (ret < 0) {
2723 goto error_version;
b551a063 2724 }
54d01ffb 2725
2f77fc4b
DG
2726 ret = modprobe_lttng_data();
2727 if (ret < 0) {
2728 goto error_modules;
54d01ffb
DG
2729 }
2730
2f77fc4b
DG
2731 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2732 return 0;
2733
2734error_version:
2735 modprobe_remove_lttng_control();
2736 ret = close(kernel_tracer_fd);
2737 if (ret) {
2738 PERROR("close");
b551a063 2739 }
2f77fc4b 2740 kernel_tracer_fd = -1;
f73fabfd 2741 return LTTNG_ERR_KERN_VERSION;
b551a063 2742
2f77fc4b
DG
2743error_modules:
2744 ret = close(kernel_tracer_fd);
2745 if (ret) {
2746 PERROR("close");
b551a063 2747 }
54d01ffb 2748
2f77fc4b
DG
2749error_open:
2750 modprobe_remove_lttng_control();
54d01ffb
DG
2751
2752error:
2f77fc4b
DG
2753 WARN("No kernel tracer available");
2754 kernel_tracer_fd = -1;
2755 if (!is_root) {
f73fabfd 2756 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2757 } else {
f73fabfd 2758 return LTTNG_ERR_KERN_NA;
2f77fc4b 2759 }
54d01ffb
DG
2760}
2761
2f77fc4b 2762
54d01ffb 2763/*
2f77fc4b
DG
2764 * Copy consumer output from the tracing session to the domain session. The
2765 * function also applies the right modification on a per domain basis for the
2766 * trace files destination directory.
36b588ed
MD
2767 *
2768 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2769 */
2f77fc4b 2770static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2771{
2772 int ret;
2f77fc4b
DG
2773 const char *dir_name;
2774 struct consumer_output *consumer;
2775
2776 assert(session);
2777 assert(session->consumer);
54d01ffb 2778
b551a063
DG
2779 switch (domain) {
2780 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2781 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2782 /*
2783 * XXX: We should audit the session creation and what this function
2784 * does "extra" in order to avoid a destroy since this function is used
2785 * in the domain session creation (kernel and ust) only. Same for UST
2786 * domain.
2787 */
2788 if (session->kernel_session->consumer) {
6addfa37 2789 consumer_output_put(session->kernel_session->consumer);
09a90bcd 2790 }
2f77fc4b
DG
2791 session->kernel_session->consumer =
2792 consumer_copy_output(session->consumer);
2793 /* Ease our life a bit for the next part */
2794 consumer = session->kernel_session->consumer;
2795 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063 2796 break;
f20baf8e 2797 case LTTNG_DOMAIN_JUL:
5cdb6027 2798 case LTTNG_DOMAIN_LOG4J:
0e115563 2799 case LTTNG_DOMAIN_PYTHON:
b551a063 2800 case LTTNG_DOMAIN_UST:
2f77fc4b 2801 DBG3("Copying tracing session consumer output in UST session");
09a90bcd 2802 if (session->ust_session->consumer) {
6addfa37 2803 consumer_output_put(session->ust_session->consumer);
09a90bcd 2804 }
2f77fc4b
DG
2805 session->ust_session->consumer =
2806 consumer_copy_output(session->consumer);
2807 /* Ease our life a bit for the next part */
2808 consumer = session->ust_session->consumer;
2809 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2810 break;
2811 default:
f73fabfd 2812 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2813 goto error;
2814 }
2815
2f77fc4b 2816 /* Append correct directory to subdir */
c30ce0b3
CB
2817 strncat(consumer->subdir, dir_name,
2818 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2819 DBG3("Copy session consumer subdir %s", consumer->subdir);
2820
f73fabfd 2821 ret = LTTNG_OK;
54d01ffb
DG
2822
2823error:
2824 return ret;
2825}
2826
00e2e675 2827/*
2f77fc4b 2828 * Create an UST session and add it to the session ust list.
36b588ed
MD
2829 *
2830 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2831 */
2f77fc4b
DG
2832static int create_ust_session(struct ltt_session *session,
2833 struct lttng_domain *domain)
00e2e675
DG
2834{
2835 int ret;
2f77fc4b 2836 struct ltt_ust_session *lus = NULL;
00e2e675 2837
a4b92340 2838 assert(session);
2f77fc4b
DG
2839 assert(domain);
2840 assert(session->consumer);
a4b92340 2841
2f77fc4b 2842 switch (domain->type) {
f20baf8e 2843 case LTTNG_DOMAIN_JUL:
5cdb6027 2844 case LTTNG_DOMAIN_LOG4J:
0e115563 2845 case LTTNG_DOMAIN_PYTHON:
2f77fc4b
DG
2846 case LTTNG_DOMAIN_UST:
2847 break;
2848 default:
2849 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2850 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2851 goto error;
2852 }
2853
2f77fc4b
DG
2854 DBG("Creating UST session");
2855
dec56f6c 2856 lus = trace_ust_create_session(session->id);
2f77fc4b 2857 if (lus == NULL) {
f73fabfd 2858 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2859 goto error;
2860 }
2861
2f77fc4b
DG
2862 lus->uid = session->uid;
2863 lus->gid = session->gid;
2bba9e53 2864 lus->output_traces = session->output_traces;
27babd3a 2865 lus->snapshot_mode = session->snapshot_mode;
ecc48a90 2866 lus->live_timer_interval = session->live_timer;
2f77fc4b 2867 session->ust_session = lus;
d7ba1388 2868 if (session->shm_path[0]) {
3d071855
MD
2869 strncpy(lus->root_shm_path, session->shm_path,
2870 sizeof(lus->root_shm_path));
2871 lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
d7ba1388
MD
2872 strncpy(lus->shm_path, session->shm_path,
2873 sizeof(lus->shm_path));
2874 lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
2875 strncat(lus->shm_path, "/ust",
2876 sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
2877 }
2f77fc4b
DG
2878 /* Copy session output to the newly created UST session */
2879 ret = copy_session_consumer(domain->type, session);
f73fabfd 2880 if (ret != LTTNG_OK) {
00e2e675
DG
2881 goto error;
2882 }
2883
f73fabfd 2884 return LTTNG_OK;
00e2e675
DG
2885
2886error:
2f77fc4b
DG
2887 free(lus);
2888 session->ust_session = NULL;
00e2e675
DG
2889 return ret;
2890}
2891
2892/*
2f77fc4b 2893 * Create a kernel tracer session then create the default channel.
00e2e675 2894 */
2f77fc4b 2895static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2896{
2897 int ret;
a4b92340 2898
2f77fc4b 2899 DBG("Creating kernel session");
00e2e675 2900
2f77fc4b
DG
2901 ret = kernel_create_session(session, kernel_tracer_fd);
2902 if (ret < 0) {
f73fabfd 2903 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2904 goto error;
2905 }
2906
2f77fc4b
DG
2907 /* Code flow safety */
2908 assert(session->kernel_session);
2909
2910 /* Copy session output to the newly created Kernel session */
2911 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2912 if (ret != LTTNG_OK) {
a4b92340
DG
2913 goto error;
2914 }
2915
2f77fc4b
DG
2916 /* Create directory(ies) on local filesystem. */
2917 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2918 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2919 ret = run_as_mkdir_recursive(
2920 session->kernel_session->consumer->dst.trace_path,
2921 S_IRWXU | S_IRWXG, session->uid, session->gid);
2922 if (ret < 0) {
df5b86c8 2923 if (errno != EEXIST) {
2f77fc4b 2924 ERR("Trace directory creation error");
00e2e675
DG
2925 goto error;
2926 }
00e2e675 2927 }
2f77fc4b 2928 }
00e2e675 2929
2f77fc4b
DG
2930 session->kernel_session->uid = session->uid;
2931 session->kernel_session->gid = session->gid;
2bba9e53 2932 session->kernel_session->output_traces = session->output_traces;
27babd3a 2933 session->kernel_session->snapshot_mode = session->snapshot_mode;
00e2e675 2934
f73fabfd 2935 return LTTNG_OK;
00e2e675 2936
2f77fc4b
DG
2937error:
2938 trace_kernel_destroy_session(session->kernel_session);
2939 session->kernel_session = NULL;
2940 return ret;
2941}
00e2e675 2942
2f77fc4b
DG
2943/*
2944 * Count number of session permitted by uid/gid.
2945 */
2946static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2947{
2948 unsigned int i = 0;
2949 struct ltt_session *session;
07424f16 2950
2f77fc4b
DG
2951 DBG("Counting number of available session for UID %d GID %d",
2952 uid, gid);
2953 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2954 /*
2f77fc4b 2955 * Only list the sessions the user can control.
00e2e675 2956 */
2f77fc4b
DG
2957 if (!session_access_ok(session, uid, gid)) {
2958 continue;
2959 }
2960 i++;
a4b92340 2961 }
2f77fc4b 2962 return i;
00e2e675
DG
2963}
2964
54d01ffb
DG
2965/*
2966 * Process the command requested by the lttng client within the command
2967 * context structure. This function make sure that the return structure (llm)
2968 * is set and ready for transmission before returning.
2969 *
2970 * Return any error encountered or 0 for success.
53a80697
MD
2971 *
2972 * "sock" is only used for special-case var. len data.
36b588ed
MD
2973 *
2974 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2975 */
53a80697
MD
2976static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2977 int *sock_error)
54d01ffb 2978{
f73fabfd 2979 int ret = LTTNG_OK;
44d3bd01 2980 int need_tracing_session = 1;
2e09ba09 2981 int need_domain;
54d01ffb
DG
2982
2983 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2984
3745d315
MD
2985 assert(!rcu_read_ongoing());
2986
53a80697
MD
2987 *sock_error = 0;
2988
2e09ba09
MD
2989 switch (cmd_ctx->lsm->cmd_type) {
2990 case LTTNG_CREATE_SESSION:
27babd3a 2991 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2992 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09
MD
2993 case LTTNG_DESTROY_SESSION:
2994 case LTTNG_LIST_SESSIONS:
2995 case LTTNG_LIST_DOMAINS:
2996 case LTTNG_START_TRACE:
2997 case LTTNG_STOP_TRACE:
6d805429 2998 case LTTNG_DATA_PENDING:
da3c9ec1
DG
2999 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3000 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3001 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3002 case LTTNG_SNAPSHOT_RECORD:
fb198a11 3003 case LTTNG_SAVE_SESSION:
d7ba1388 3004 case LTTNG_SET_SESSION_SHM_PATH:
eded6438 3005 case LTTNG_REGENERATE_METADATA:
c2561365 3006 case LTTNG_REGENERATE_STATEDUMP:
2e09ba09 3007 need_domain = 0;
3aace903 3008 break;
2e09ba09
MD
3009 default:
3010 need_domain = 1;
3011 }
3012
3013 if (opt_no_kernel && need_domain
3014 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 3015 if (!is_root) {
f73fabfd 3016 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 3017 } else {
f73fabfd 3018 ret = LTTNG_ERR_KERN_NA;
531d29f9 3019 }
4fba7219
DG
3020 goto error;
3021 }
3022
8d3113b2
DG
3023 /* Deny register consumer if we already have a spawned consumer. */
3024 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
3025 pthread_mutex_lock(&kconsumer_data.pid_mutex);
3026 if (kconsumer_data.pid > 0) {
f73fabfd 3027 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 3028 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
3029 goto error;
3030 }
3031 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
3032 }
3033
54d01ffb
DG
3034 /*
3035 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 3036 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
3037 * command.
3038 */
3039 switch(cmd_ctx->lsm->cmd_type) {
3040 case LTTNG_LIST_SESSIONS:
3041 case LTTNG_LIST_TRACEPOINTS:
f37d259d 3042 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
3043 case LTTNG_LIST_DOMAINS:
3044 case LTTNG_LIST_CHANNELS:
3045 case LTTNG_LIST_EVENTS:
834978fd 3046 case LTTNG_LIST_SYSCALLS:
a5dfbb9d 3047 case LTTNG_LIST_TRACKER_PIDS:
5cd0780d 3048 case LTTNG_DATA_PENDING:
54d01ffb
DG
3049 break;
3050 default:
3051 /* Setup lttng message with no payload */
6e10c9b9 3052 ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
54d01ffb
DG
3053 if (ret < 0) {
3054 /* This label does not try to unlock the session */
3055 goto init_setup_error;
3056 }
3057 }
3058
3059 /* Commands that DO NOT need a session. */
3060 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 3061 case LTTNG_CREATE_SESSION:
27babd3a 3062 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 3063 case LTTNG_CREATE_SESSION_LIVE:
54d01ffb
DG
3064 case LTTNG_LIST_SESSIONS:
3065 case LTTNG_LIST_TRACEPOINTS:
834978fd 3066 case LTTNG_LIST_SYSCALLS:
f37d259d 3067 case LTTNG_LIST_TRACEPOINT_FIELDS:
fb198a11 3068 case LTTNG_SAVE_SESSION:
44d3bd01 3069 need_tracing_session = 0;
54d01ffb
DG
3070 break;
3071 default:
3072 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
3073 /*
3074 * We keep the session list lock across _all_ commands
3075 * for now, because the per-session lock does not
3076 * handle teardown properly.
3077 */
74babd95 3078 session_lock_list();
54d01ffb
DG
3079 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
3080 if (cmd_ctx->session == NULL) {
bba2d65f 3081 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
3082 goto error;
3083 } else {
3084 /* Acquire lock for the session */
3085 session_lock(cmd_ctx->session);
3086 }
3087 break;
3088 }
b389abbe 3089
5f3ecf22
DG
3090 /*
3091 * Commands that need a valid session but should NOT create one if none
3092 * exists. Instead of creating one and destroying it when the command is
3093 * handled, process that right before so we save some round trip in useless
3094 * code path.
3095 */
3096 switch (cmd_ctx->lsm->cmd_type) {
3097 case LTTNG_DISABLE_CHANNEL:
3098 case LTTNG_DISABLE_EVENT:
5f3ecf22
DG
3099 switch (cmd_ctx->lsm->domain.type) {
3100 case LTTNG_DOMAIN_KERNEL:
3101 if (!cmd_ctx->session->kernel_session) {
3102 ret = LTTNG_ERR_NO_CHANNEL;
3103</