Pass lttng_event struct to the set_filter API call
[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>
fac6795d 4 *
d14d33bf
AM
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
91d76f53 8 *
d14d33bf
AM
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
91d76f53 13 *
d14d33bf
AM
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
fac6795d
DG
17 */
18
19#define _GNU_SOURCE
fac6795d
DG
20#include <getopt.h>
21#include <grp.h>
22#include <limits.h>
23#include <pthread.h>
24#include <signal.h>
25#include <stdio.h>
26#include <stdlib.h>
27#include <string.h>
0fdd1e2c 28#include <sys/mman.h>
b73401da 29#include <sys/mount.h>
1e307fab 30#include <sys/resource.h>
fac6795d
DG
31#include <sys/socket.h>
32#include <sys/stat.h>
33#include <sys/types.h>
0fdd1e2c 34#include <sys/wait.h>
5c827ce0 35#include <urcu/uatomic.h>
fac6795d 36#include <unistd.h>
3bd1e081 37#include <config.h>
fac6795d 38
990570ed 39#include <common/common.h>
0ba98ebc 40#include <common/compat/poll.h>
d27c42b8 41#include <common/compat/socket.h>
db758600
DG
42#include <common/defaults.h>
43#include <common/kernel-consumer/kernel-consumer.h>
50c8f484 44#include <common/futex.h>
00e2e675 45#include <common/relayd/relayd.h>
81b86775 46#include <common/utils.h>
fac6795d 47
10a8a223 48#include "lttng-sessiond.h"
54d01ffb 49#include "channel.h"
2f77fc4b 50#include "cmd.h"
00e2e675 51#include "consumer.h"
099e26bd 52#include "context.h"
54d01ffb 53#include "event.h"
4771f025 54#include "kernel.h"
f1e16794 55#include "kernel-consumer.h"
096102bd 56#include "modprobe.h"
0fdd1e2c 57#include "shm.h"
1e307fab 58#include "ust-ctl.h"
00e2e675 59#include "ust-consumer.h"
8e68d1c8 60#include "utils.h"
4063050c 61#include "fd-limit.h"
53a80697 62#include "filter.h"
44a5e5eb 63#include "health.h"
8ac94142 64#include "testpoint.h"
fac6795d 65
ebaeda94
MD
66#define CONSUMERD_FILE "lttng-consumerd"
67
75462a81 68/* Const values */
686204ab 69const char default_home_dir[] = DEFAULT_HOME_DIR;
bbccc3d2 70const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
686204ab
MD
71const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
72const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
73
fac6795d
DG
74const char *progname;
75const char *opt_tracing_group;
5b8719f5 76static int opt_sig_parent;
97e19046 77static int opt_verbose_consumer;
fac6795d 78static int opt_daemon;
4fba7219 79static int opt_no_kernel;
fac6795d 80static int is_root; /* Set to 1 if the daemon is running as root */
1d4b027a 81static pid_t ppid; /* Parent PID for --sig-parent option */
67e40797 82static char *rundir;
3bd1e081 83
a23ec3a7
DG
84/*
85 * Consumer daemon specific control data. Every value not initialized here is
86 * set to 0 by the static definition.
87 */
3bd1e081
MD
88static struct consumer_data kconsumer_data = {
89 .type = LTTNG_CONSUMER_KERNEL,
60922cb0
DG
90 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
91 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
03550b58
MD
92 .err_sock = -1,
93 .cmd_sock = -1,
173af62f
DG
94 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
95 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
96 .cond = PTHREAD_COND_INITIALIZER,
97 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081 98};
7753dea8
MD
99static struct consumer_data ustconsumer64_data = {
100 .type = LTTNG_CONSUMER64_UST,
60922cb0
DG
101 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
102 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
03550b58
MD
103 .err_sock = -1,
104 .cmd_sock = -1,
173af62f
DG
105 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
106 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
107 .cond = PTHREAD_COND_INITIALIZER,
108 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
7753dea8
MD
109};
110static struct consumer_data ustconsumer32_data = {
111 .type = LTTNG_CONSUMER32_UST,
60922cb0
DG
112 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
113 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
03550b58
MD
114 .err_sock = -1,
115 .cmd_sock = -1,
173af62f
DG
116 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
117 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
118 .cond = PTHREAD_COND_INITIALIZER,
119 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081
MD
120};
121
26c9d55e 122/* Shared between threads */
099e26bd 123static int dispatch_thread_exit;
fac6795d 124
54d01ffb
DG
125/* Global application Unix socket path */
126static char apps_unix_sock_path[PATH_MAX];
127/* Global client Unix socket path */
128static char client_unix_sock_path[PATH_MAX];
54d01ffb
DG
129/* global wait shm path for UST */
130static char wait_shm_path[PATH_MAX];
44a5e5eb
DG
131/* Global health check unix path */
132static char health_unix_sock_path[PATH_MAX];
fac6795d 133
1d4b027a 134/* Sockets and FDs */
a4b35e07
MD
135static int client_sock = -1;
136static int apps_sock = -1;
2f77fc4b 137int kernel_tracer_fd = -1;
76d7553f 138static int kernel_poll_pipe[2] = { -1, -1 };
1d4b027a 139
273ea72c
DG
140/*
141 * Quit pipe for all threads. This permits a single cancellation point
142 * for all threads when receiving an event on the pipe.
143 */
76d7553f 144static int thread_quit_pipe[2] = { -1, -1 };
273ea72c 145
099e26bd
DG
146/*
147 * This pipe is used to inform the thread managing application communication
148 * that a command is queued and ready to be processed.
149 */
76d7553f 150static int apps_cmd_pipe[2] = { -1, -1 };
099e26bd 151
1d4b027a 152/* Pthread, Mutexes and Semaphores */
1d4b027a 153static pthread_t apps_thread;
099e26bd 154static pthread_t reg_apps_thread;
1d4b027a 155static pthread_t client_thread;
7a485870 156static pthread_t kernel_thread;
099e26bd 157static pthread_t dispatch_thread;
44a5e5eb 158static pthread_t health_thread;
5eb91c98 159
099e26bd
DG
160/*
161 * UST registration command queue. This queue is tied with a futex and uses a N
162 * wakers / 1 waiter implemented and detailed in futex.c/.h
163 *
164 * The thread_manage_apps and thread_dispatch_ust_registration interact with
165 * this queue and the wait/wake scheme.
166 */
167static struct ust_cmd_queue ust_cmd_queue;
168
b5541356
DG
169/*
170 * Pointer initialized before thread creation.
171 *
172 * This points to the tracing session list containing the session count and a
173 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
174 * MUST NOT be taken if you call a public function in session.c.
04ea676f 175 *
d063d709 176 * The lock is nested inside the structure: session_list_ptr->lock. Please use
54d01ffb 177 * session_lock_list and session_unlock_list for lock acquisition.
b5541356
DG
178 */
179static struct ltt_session_list *session_list_ptr;
180
7753dea8
MD
181int ust_consumerd64_fd = -1;
182int ust_consumerd32_fd = -1;
183
fb6f1fa2
YB
184static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
185static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
186static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
187static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
fb09408a 188
2f77fc4b
DG
189static const char *module_proc_lttng = "/proc/lttng";
190
5c827ce0
DG
191/*
192 * Consumer daemon state which is changed when spawning it, killing it or in
193 * case of a fatal error.
194 */
195enum consumerd_state {
196 CONSUMER_STARTED = 1,
197 CONSUMER_STOPPED = 2,
198 CONSUMER_ERROR = 3,
199};
200
201/*
202 * This consumer daemon state is used to validate if a client command will be
203 * able to reach the consumer. If not, the client is informed. For instance,
204 * doing a "lttng start" when the consumer state is set to ERROR will return an
205 * error to the client.
206 *
207 * The following example shows a possible race condition of this scheme:
208 *
209 * consumer thread error happens
210 * client cmd arrives
211 * client cmd checks state -> still OK
212 * consumer thread exit, sets error
213 * client cmd try to talk to consumer
214 * ...
215 *
216 * However, since the consumer is a different daemon, we have no way of making
217 * sure the command will reach it safely even with this state flag. This is why
218 * we consider that up to the state validation during command processing, the
219 * command is safe. After that, we can not guarantee the correctness of the
220 * client request vis-a-vis the consumer.
221 */
222static enum consumerd_state ust_consumerd_state;
223static enum consumerd_state kernel_consumerd_state;
224
44a5e5eb
DG
225/* Used for the health monitoring of the session daemon. See health.h */
226struct health_state health_thread_cmd;
139ac872 227struct health_state health_thread_app_manage;
44a5e5eb
DG
228struct health_state health_thread_app_reg;
229struct health_state health_thread_kernel;
230
ae9e45b3
DG
231/*
232 * Socket timeout for receiving and sending in seconds.
233 */
234static int app_socket_timeout;
235
fb09408a 236static
7753dea8 237void setup_consumerd_path(void)
fb09408a 238{
fc7a59ce 239 const char *bin, *libdir;
fb09408a 240
7753dea8
MD
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
7753dea8 245 */
ebaeda94 246#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 256 }
ebaeda94
MD
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
259 }
260#else
261#error "Unknown bitness"
262#endif
263
fb09408a
MD
264 /*
265 * runtime env. var. overrides the build default.
266 */
fc7a59ce
AM
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
7753dea8 270 }
fc7a59ce
AM
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
ebaeda94 274 }
72f579ee 275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
72f579ee 279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
280 if (libdir) {
281 consumerd64_libdir = libdir;
fb09408a
MD
282 }
283}
284
5eb91c98
DG
285/*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288static int create_thread_poll_set(struct lttng_poll_event *events,
289 unsigned int size)
290{
291 int ret;
292
293 if (events == NULL || size == 0) {
294 ret = -1;
295 goto error;
296 }
297
298 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
299 if (ret < 0) {
300 goto error;
301 }
302
303 /* Add quit pipe */
304 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
305 if (ret < 0) {
306 goto error;
307 }
308
309 return 0;
310
311error:
312 return ret;
313}
314
315/*
316 * Check if the thread quit pipe was triggered.
317 *
318 * Return 1 if it was triggered else 0;
319 */
320static int check_thread_quit_pipe(int fd, uint32_t events)
321{
322 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
323 return 1;
324 }
325
326 return 0;
327}
328
0fdd1e2c
DG
329/*
330 * Return group ID of the tracing group or -1 if not found.
331 */
996b65c8
MD
332static gid_t allowed_group(void)
333{
334 struct group *grp;
335
274e143b
MD
336 if (opt_tracing_group) {
337 grp = getgrnam(opt_tracing_group);
338 } else {
339 grp = getgrnam(default_tracing_group);
340 }
996b65c8
MD
341 if (!grp) {
342 return -1;
343 } else {
344 return grp->gr_gid;
345 }
346}
347
273ea72c 348/*
5eb91c98 349 * Init thread quit pipe.
273ea72c
DG
350 *
351 * Return -1 on error or 0 if all pipes are created.
352 */
353static int init_thread_quit_pipe(void)
354{
730389d9 355 int ret, i;
273ea72c 356
730389d9 357 ret = pipe(thread_quit_pipe);
273ea72c 358 if (ret < 0) {
730389d9 359 PERROR("thread quit pipe");
273ea72c
DG
360 goto error;
361 }
362
730389d9
DG
363 for (i = 0; i < 2; i++) {
364 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
365 if (ret < 0) {
366 PERROR("fcntl");
367 goto error;
368 }
369 }
370
273ea72c
DG
371error:
372 return ret;
373}
374
099e26bd
DG
375/*
376 * Stop all threads by closing the thread quit pipe.
377 */
cf3af59e
MD
378static void stop_threads(void)
379{
5eb91c98
DG
380 int ret;
381
cf3af59e
MD
382 /* Stopping all threads */
383 DBG("Terminating all threads");
54d01ffb 384 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
385 if (ret < 0) {
386 ERR("write error on thread quit pipe");
387 }
388
099e26bd 389 /* Dispatch thread */
26c9d55e 390 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 391 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
392}
393
fac6795d 394/*
d063d709 395 * Cleanup the daemon
fac6795d 396 */
cf3af59e 397static void cleanup(void)
fac6795d 398{
ef599319 399 int ret;
6620da75 400 char *cmd = NULL;
af9737e9 401 struct ltt_session *sess, *stmp;
fac6795d 402
1d4b027a 403 DBG("Cleaning up");
e07ae692 404
2f77fc4b
DG
405 /* First thing first, stop all threads */
406 utils_close_pipe(thread_quit_pipe);
407
67e40797
DG
408 DBG("Removing %s directory", rundir);
409 ret = asprintf(&cmd, "rm -rf %s", rundir);
410 if (ret < 0) {
411 ERR("asprintf failed. Something is really wrong!");
412 }
5461b305 413
67e40797
DG
414 /* Remove lttng run directory */
415 ret = system(cmd);
416 if (ret < 0) {
417 ERR("Unable to clean %s", rundir);
1d4b027a 418 }
67e40797 419 free(cmd);
bd69add2 420 free(rundir);
5461b305 421
99bab54f 422 DBG("Cleaning up all sessions");
fac6795d 423
b5541356 424 /* Destroy session list mutex */
273ea72c
DG
425 if (session_list_ptr != NULL) {
426 pthread_mutex_destroy(&session_list_ptr->lock);
427
428 /* Cleanup ALL session */
54d01ffb
DG
429 cds_list_for_each_entry_safe(sess, stmp,
430 &session_list_ptr->head, list) {
2f77fc4b 431 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
432 }
433 }
434
099e26bd 435 DBG("Closing all UST sockets");
56fff090 436 ust_app_clean_list();
099e26bd 437
4fba7219
DG
438 if (is_root && !opt_no_kernel) {
439 DBG2("Closing kernel fd");
a4b35e07 440 if (kernel_tracer_fd >= 0) {
76d7553f
MD
441 ret = close(kernel_tracer_fd);
442 if (ret) {
443 PERROR("close");
444 }
a4b35e07 445 }
2f50c8a3 446 DBG("Unloading kernel modules");
096102bd 447 modprobe_remove_lttng_all();
2f50c8a3 448 }
2f77fc4b 449
421cb601 450 /* <fun> */
f56a39af 451 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
452 "Matthew, BEET driven development works!%c[%dm",
453 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
454 /* </fun> */
fac6795d
DG
455}
456
e065084a 457/*
d063d709 458 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 459 *
d063d709 460 * Return lttcomm error code.
e065084a
DG
461 */
462static int send_unix_sock(int sock, void *buf, size_t len)
463{
464 /* Check valid length */
465 if (len <= 0) {
466 return -1;
467 }
468
469 return lttcomm_send_unix_sock(sock, buf, len);
470}
471
5461b305 472/*
d063d709 473 * Free memory of a command context structure.
5461b305 474 */
a2fb29a5 475static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 476{
a2fb29a5
DG
477 DBG("Clean command context structure");
478 if (*cmd_ctx) {
479 if ((*cmd_ctx)->llm) {
480 free((*cmd_ctx)->llm);
5461b305 481 }
a2fb29a5
DG
482 if ((*cmd_ctx)->lsm) {
483 free((*cmd_ctx)->lsm);
5461b305 484 }
a2fb29a5
DG
485 free(*cmd_ctx);
486 *cmd_ctx = NULL;
5461b305
DG
487 }
488}
489
fac6795d 490/*
0fdd1e2c 491 * Notify UST applications using the shm mmap futex.
fac6795d 492 */
0fdd1e2c 493static int notify_ust_apps(int active)
fac6795d 494{
0fdd1e2c 495 char *wait_shm_mmap;
fac6795d 496
0fdd1e2c 497 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 498
0fdd1e2c
DG
499 /* See shm.c for this call implying mmap, shm and futex calls */
500 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
501 if (wait_shm_mmap == NULL) {
fac6795d
DG
502 goto error;
503 }
504
0fdd1e2c
DG
505 /* Wake waiting process */
506 futex_wait_update((int32_t *) wait_shm_mmap, active);
507
508 /* Apps notified successfully */
509 return 0;
fac6795d
DG
510
511error:
0fdd1e2c 512 return -1;
fac6795d
DG
513}
514
e065084a 515/*
d063d709
DG
516 * Setup the outgoing data buffer for the response (llm) by allocating the
517 * right amount of memory and copying the original information from the lsm
518 * structure.
ca95a216 519 *
d063d709 520 * Return total size of the buffer pointed by buf.
ca95a216 521 */
5461b305 522static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 523{
f3ed775e 524 int ret, buf_size;
ca95a216 525
f3ed775e 526 buf_size = size;
5461b305 527
ba7f0ae5 528 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 529 if (cmd_ctx->llm == NULL) {
76d7553f 530 PERROR("zmalloc");
5461b305 531 ret = -ENOMEM;
ca95a216
DG
532 goto error;
533 }
534
5461b305
DG
535 /* Copy common data */
536 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 537 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 538
5461b305
DG
539 cmd_ctx->llm->data_size = size;
540 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
541
ca95a216
DG
542 return buf_size;
543
544error:
545 return ret;
546}
547
7a485870 548/*
5eb91c98 549 * Update the kernel poll set of all channel fd available over all tracing
d063d709 550 * session. Add the wakeup pipe at the end of the set.
7a485870 551 */
5eb91c98 552static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 553{
5eb91c98 554 int ret;
7a485870
DG
555 struct ltt_session *session;
556 struct ltt_kernel_channel *channel;
557
5eb91c98 558 DBG("Updating kernel poll set");
7a485870 559
54d01ffb 560 session_lock_list();
b5541356 561 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 562 session_lock(session);
7a485870 563 if (session->kernel_session == NULL) {
54d01ffb 564 session_unlock(session);
7a485870
DG
565 continue;
566 }
7a485870 567
54d01ffb
DG
568 cds_list_for_each_entry(channel,
569 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
570 /* Add channel fd to the kernel poll set */
571 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
572 if (ret < 0) {
54d01ffb 573 session_unlock(session);
5eb91c98
DG
574 goto error;
575 }
576 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 577 }
54d01ffb 578 session_unlock(session);
7a485870 579 }
54d01ffb 580 session_unlock_list();
7a485870 581
5eb91c98 582 return 0;
7a485870
DG
583
584error:
54d01ffb 585 session_unlock_list();
7a485870
DG
586 return -1;
587}
588
589/*
54d01ffb 590 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 591 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 592 *
d063d709 593 * Useful for CPU hotplug feature.
7a485870 594 */
2bdd86d4 595static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
596{
597 int ret = 0;
598 struct ltt_session *session;
173af62f 599 struct ltt_kernel_session *ksess;
7a485870
DG
600 struct ltt_kernel_channel *channel;
601
602 DBG("Updating kernel streams for channel fd %d", fd);
603
54d01ffb 604 session_lock_list();
b5541356 605 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 606 session_lock(session);
7a485870 607 if (session->kernel_session == NULL) {
54d01ffb 608 session_unlock(session);
7a485870
DG
609 continue;
610 }
173af62f 611 ksess = session->kernel_session;
d9800920 612
173af62f 613 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
7a485870
DG
614 if (channel->fd == fd) {
615 DBG("Channel found, updating kernel streams");
616 ret = kernel_open_channel_stream(channel);
617 if (ret < 0) {
b3c750d2 618 goto error;
7a485870 619 }
d9800920 620
7a485870 621 /*
5eb91c98
DG
622 * Have we already sent fds to the consumer? If yes, it means
623 * that tracing is started so it is safe to send our updated
624 * stream fds.
7a485870 625 */
173af62f
DG
626 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
627 struct lttng_ht_iter iter;
628 struct consumer_socket *socket;
629
630
631 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
632 &iter.iter, socket, node.node) {
633 /* Code flow error */
634 assert(socket->fd >= 0);
635
636 pthread_mutex_lock(socket->lock);
637 ret = kernel_consumer_send_channel_stream(socket->fd,
638 channel, ksess);
639 pthread_mutex_unlock(socket->lock);
640 if (ret < 0) {
641 goto error;
642 }
7a485870
DG
643 }
644 }
b3c750d2 645 goto error;
7a485870
DG
646 }
647 }
54d01ffb 648 session_unlock(session);
7a485870 649 }
54d01ffb 650 session_unlock_list();
b3c750d2 651 return ret;
7a485870 652
b3c750d2 653error:
54d01ffb
DG
654 session_unlock(session);
655 session_unlock_list();
7a485870
DG
656 return ret;
657}
658
487cf67c
DG
659/*
660 * For each tracing session, update newly registered apps.
661 */
662static void update_ust_app(int app_sock)
663{
664 struct ltt_session *sess, *stmp;
665
4ee14516
DG
666 session_lock_list();
667
487cf67c
DG
668 /* For all tracing session(s) */
669 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
4ee14516 670 session_lock(sess);
421cb601
DG
671 if (sess->ust_session) {
672 ust_app_global_update(sess->ust_session, app_sock);
673 }
4ee14516 674 session_unlock(sess);
487cf67c 675 }
4ee14516
DG
676
677 session_unlock_list();
487cf67c
DG
678}
679
7a485870 680/*
d063d709 681 * This thread manage event coming from the kernel.
7a485870 682 *
d063d709
DG
683 * Features supported in this thread:
684 * -) CPU Hotplug
7a485870
DG
685 */
686static void *thread_manage_kernel(void *data)
687{
139ac872 688 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 689 uint32_t revents, nb_fd;
7a485870 690 char tmp;
5eb91c98 691 struct lttng_poll_event events;
7a485870
DG
692
693 DBG("Thread manage kernel started");
694
8ac94142
CB
695 testpoint(thread_manage_kernel);
696
44a5e5eb
DG
697 health_code_update(&health_thread_kernel);
698
8ac94142
CB
699 testpoint(thread_manage_kernel_before_loop);
700
5eb91c98
DG
701 ret = create_thread_poll_set(&events, 2);
702 if (ret < 0) {
76d7553f 703 goto error_poll_create;
5eb91c98
DG
704 }
705
706 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
707 if (ret < 0) {
708 goto error;
709 }
710
7a485870 711 while (1) {
44a5e5eb
DG
712 health_code_update(&health_thread_kernel);
713
7a485870 714 if (update_poll_flag == 1) {
5f822d0a
DG
715 /*
716 * Reset number of fd in the poll set. Always 2 since there is the thread
717 * quit pipe and the kernel pipe.
718 */
719 events.nb_fd = 2;
720
5eb91c98
DG
721 ret = update_kernel_poll(&events);
722 if (ret < 0) {
7a485870
DG
723 goto error;
724 }
725 update_poll_flag = 0;
726 }
727
5eb91c98
DG
728 nb_fd = LTTNG_POLL_GETNB(&events);
729
730 DBG("Thread kernel polling on %d fds", nb_fd);
731
732 /* Zeroed the poll events */
733 lttng_poll_reset(&events);
7a485870
DG
734
735 /* Poll infinite value of time */
88f2b785 736 restart:
44a5e5eb 737 health_poll_update(&health_thread_kernel);
5eb91c98 738 ret = lttng_poll_wait(&events, -1);
44a5e5eb 739 health_poll_update(&health_thread_kernel);
7a485870 740 if (ret < 0) {
88f2b785
MD
741 /*
742 * Restart interrupted system call.
743 */
744 if (errno == EINTR) {
745 goto restart;
746 }
7a485870
DG
747 goto error;
748 } else if (ret == 0) {
749 /* Should not happen since timeout is infinite */
85611738
DG
750 ERR("Return value of poll is 0 with an infinite timeout.\n"
751 "This should not have happened! Continuing...");
7a485870
DG
752 continue;
753 }
754
5eb91c98
DG
755 for (i = 0; i < nb_fd; i++) {
756 /* Fetch once the poll data */
757 revents = LTTNG_POLL_GETEV(&events, i);
758 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 759
44a5e5eb
DG
760 health_code_update(&health_thread_kernel);
761
5eb91c98
DG
762 /* Thread quit pipe has been closed. Killing thread. */
763 ret = check_thread_quit_pipe(pollfd, revents);
764 if (ret) {
139ac872
MD
765 err = 0;
766 goto exit;
5eb91c98 767 }
7a485870 768
5eb91c98
DG
769 /* Check for data on kernel pipe */
770 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
771 ret = read(kernel_poll_pipe[0], &tmp, 1);
772 update_poll_flag = 1;
773 continue;
774 } else {
775 /*
776 * New CPU detected by the kernel. Adding kernel stream to
777 * kernel session and updating the kernel consumer
778 */
779 if (revents & LPOLLIN) {
2bdd86d4 780 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
781 if (ret < 0) {
782 continue;
783 }
784 break;
785 /*
786 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
787 * and unregister kernel stream at this point.
788 */
7a485870 789 }
7a485870
DG
790 }
791 }
792 }
793
139ac872 794exit:
7a485870 795error:
5eb91c98 796 lttng_poll_clean(&events);
76d7553f 797error_poll_create:
6620da75
DG
798 utils_close_pipe(kernel_poll_pipe);
799 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872
MD
800 if (err) {
801 health_error(&health_thread_kernel);
802 ERR("Health error occurred in %s", __func__);
6620da75
DG
803 WARN("Kernel thread died unexpectedly. "
804 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872
MD
805 }
806 health_exit(&health_thread_kernel);
76d7553f 807 DBG("Kernel thread dying");
7a485870
DG
808 return NULL;
809}
810
a23ec3a7
DG
811/*
812 * Signal pthread condition of the consumer data that the thread.
813 */
814static void signal_consumer_condition(struct consumer_data *data, int state)
815{
816 pthread_mutex_lock(&data->cond_mutex);
817
818 /*
819 * The state is set before signaling. It can be any value, it's the waiter
820 * job to correctly interpret this condition variable associated to the
821 * consumer pthread_cond.
822 *
823 * A value of 0 means that the corresponding thread of the consumer data
824 * was not started. 1 indicates that the thread has started and is ready
825 * for action. A negative value means that there was an error during the
826 * thread bootstrap.
827 */
828 data->consumer_thread_is_ready = state;
829 (void) pthread_cond_signal(&data->cond);
830
831 pthread_mutex_unlock(&data->cond_mutex);
832}
833
1d4b027a 834/*
3bd1e081 835 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 836 */
3bd1e081 837static void *thread_manage_consumer(void *data)
1d4b027a 838{
139ac872 839 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98 840 uint32_t revents, nb_fd;
1d4b027a 841 enum lttcomm_return_code code;
5eb91c98 842 struct lttng_poll_event events;
3bd1e081 843 struct consumer_data *consumer_data = data;
1d4b027a 844
3bd1e081 845 DBG("[thread] Manage consumer started");
1d4b027a 846
9449cc75
DG
847 /*
848 * Since the consumer thread can be spawned at any moment in time, we init
849 * the health to a poll status (1, which is a valid health over time).
850 * When the thread starts, we update here the health to a "code" path being
851 * an even value so this thread, when reaching a poll wait, does not
852 * trigger an error with an even value.
853 *
854 * Here is the use case we avoid.
855 *
856 * +1: the first poll update during initialization (main())
857 * +2 * x: multiple code update once in this thread.
858 * +1: poll wait in this thread (being a good health state).
859 * == even number which after the wait period shows as a bad health.
860 *
861 * In a nutshell, the following poll update to the health state brings back
862 * the state to an even value meaning a code path.
863 */
864 health_poll_update(&consumer_data->health);
865
5eb91c98
DG
866 /*
867 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
868 * Nothing more will be added to this poll set.
869 */
870 ret = create_thread_poll_set(&events, 2);
871 if (ret < 0) {
76d7553f 872 goto error_poll;
5eb91c98 873 }
273ea72c 874
edb8b045
DG
875 /*
876 * The error socket here is already in a listening state which was done
877 * just before spawning this thread to avoid a race between the consumer
878 * daemon exec trying to connect and the listen() call.
879 */
3bd1e081 880 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
881 if (ret < 0) {
882 goto error;
883 }
884
885 nb_fd = LTTNG_POLL_GETNB(&events);
273ea72c 886
44a5e5eb
DG
887 health_code_update(&consumer_data->health);
888
273ea72c 889 /* Inifinite blocking call, waiting for transmission */
88f2b785 890restart:
44a5e5eb 891 health_poll_update(&consumer_data->health);
8ac94142
CB
892
893 testpoint(thread_manage_consumer);
894
5eb91c98 895 ret = lttng_poll_wait(&events, -1);
44a5e5eb 896 health_poll_update(&consumer_data->health);
273ea72c 897 if (ret < 0) {
88f2b785
MD
898 /*
899 * Restart interrupted system call.
900 */
901 if (errno == EINTR) {
902 goto restart;
903 }
273ea72c
DG
904 goto error;
905 }
906
5eb91c98
DG
907 for (i = 0; i < nb_fd; i++) {
908 /* Fetch once the poll data */
909 revents = LTTNG_POLL_GETEV(&events, i);
910 pollfd = LTTNG_POLL_GETFD(&events, i);
911
44a5e5eb
DG
912 health_code_update(&consumer_data->health);
913
5eb91c98
DG
914 /* Thread quit pipe has been closed. Killing thread. */
915 ret = check_thread_quit_pipe(pollfd, revents);
916 if (ret) {
139ac872
MD
917 err = 0;
918 goto exit;
5eb91c98
DG
919 }
920
921 /* Event on the registration socket */
3bd1e081 922 if (pollfd == consumer_data->err_sock) {
5eb91c98 923 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 924 ERR("consumer err socket poll error");
5eb91c98
DG
925 goto error;
926 }
927 }
273ea72c
DG
928 }
929
3bd1e081 930 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
931 if (sock < 0) {
932 goto error;
933 }
934
b662582b
DG
935 /*
936 * Set the CLOEXEC flag. Return code is useless because either way, the
937 * show must go on.
938 */
939 (void) utils_set_fd_cloexec(sock);
940
44a5e5eb
DG
941 health_code_update(&consumer_data->health);
942
3bd1e081 943 DBG2("Receiving code from consumer err_sock");
ee0b0061 944
712ea556 945 /* Getting status code from kconsumerd */
54d01ffb
DG
946 ret = lttcomm_recv_unix_sock(sock, &code,
947 sizeof(enum lttcomm_return_code));
1d4b027a
DG
948 if (ret <= 0) {
949 goto error;
950 }
951
44a5e5eb
DG
952 health_code_update(&consumer_data->health);
953
f73fabfd 954 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
3bd1e081
MD
955 consumer_data->cmd_sock =
956 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
957 if (consumer_data->cmd_sock < 0) {
a23ec3a7
DG
958 /* On error, signal condition and quit. */
959 signal_consumer_condition(consumer_data, -1);
48842b30 960 PERROR("consumer connect");
1d4b027a
DG
961 goto error;
962 }
a23ec3a7
DG
963 signal_consumer_condition(consumer_data, 1);
964 DBG("Consumer command socket ready");
1d4b027a 965 } else {
3bd1e081 966 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
967 lttcomm_get_readable_code(-code));
968 goto error;
969 }
970
54d01ffb 971 /* Remove the kconsumerd error sock since we've established a connexion */
3bd1e081 972 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 973 if (ret < 0) {
72079cae
DG
974 goto error;
975 }
976
5eb91c98
DG
977 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
978 if (ret < 0) {
72079cae 979 goto error;
5eb91c98
DG
980 }
981
44a5e5eb
DG
982 health_code_update(&consumer_data->health);
983
5eb91c98
DG
984 /* Update number of fd */
985 nb_fd = LTTNG_POLL_GETNB(&events);
986
987 /* Inifinite blocking call, waiting for transmission */
88f2b785 988restart_poll:
44a5e5eb 989 health_poll_update(&consumer_data->health);
5eb91c98 990 ret = lttng_poll_wait(&events, -1);
44a5e5eb 991 health_poll_update(&consumer_data->health);
5eb91c98 992 if (ret < 0) {
88f2b785
MD
993 /*
994 * Restart interrupted system call.
995 */
996 if (errno == EINTR) {
997 goto restart_poll;
998 }
72079cae
DG
999 goto error;
1000 }
1001
5eb91c98
DG
1002 for (i = 0; i < nb_fd; i++) {
1003 /* Fetch once the poll data */
1004 revents = LTTNG_POLL_GETEV(&events, i);
1005 pollfd = LTTNG_POLL_GETFD(&events, i);
1006
44a5e5eb
DG
1007 health_code_update(&consumer_data->health);
1008
5eb91c98
DG
1009 /* Thread quit pipe has been closed. Killing thread. */
1010 ret = check_thread_quit_pipe(pollfd, revents);
1011 if (ret) {
139ac872
MD
1012 err = 0;
1013 goto exit;
5eb91c98
DG
1014 }
1015
1016 /* Event on the kconsumerd socket */
1017 if (pollfd == sock) {
1018 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1019 ERR("consumer err socket second poll error");
5eb91c98
DG
1020 goto error;
1021 }
1022 }
1023 }
1024
44a5e5eb
DG
1025 health_code_update(&consumer_data->health);
1026
712ea556 1027 /* Wait for any kconsumerd error */
54d01ffb
DG
1028 ret = lttcomm_recv_unix_sock(sock, &code,
1029 sizeof(enum lttcomm_return_code));
712ea556 1030 if (ret <= 0) {
3bd1e081 1031 ERR("consumer closed the command socket");
712ea556 1032 goto error;
6f61d021 1033 }
1d4b027a 1034
3bd1e081 1035 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
712ea556 1036
139ac872 1037exit:
1d4b027a 1038error:
5c827ce0
DG
1039 /* Immediately set the consumerd state to stopped */
1040 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1041 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1042 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1043 consumer_data->type == LTTNG_CONSUMER32_UST) {
1044 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1045 } else {
1046 /* Code flow error... */
1047 assert(0);
1048 }
1049
76d7553f
MD
1050 if (consumer_data->err_sock >= 0) {
1051 ret = close(consumer_data->err_sock);
1052 if (ret) {
1053 PERROR("close");
1054 }
1055 }
1056 if (consumer_data->cmd_sock >= 0) {
1057 ret = close(consumer_data->cmd_sock);
1058 if (ret) {
1059 PERROR("close");
1060 }
1061 }
1062 if (sock >= 0) {
1063 ret = close(sock);
1064 if (ret) {
1065 PERROR("close");
1066 }
1067 }
273ea72c 1068
3bd1e081
MD
1069 unlink(consumer_data->err_unix_sock_path);
1070 unlink(consumer_data->cmd_unix_sock_path);
1071 consumer_data->pid = 0;
1d4b027a 1072
5eb91c98 1073 lttng_poll_clean(&events);
76d7553f 1074error_poll:
139ac872
MD
1075 if (err) {
1076 health_error(&consumer_data->health);
1077 ERR("Health error occurred in %s", __func__);
1078 }
1079 health_exit(&consumer_data->health);
76d7553f 1080 DBG("consumer thread cleanup completed");
0177d773 1081
5eb91c98 1082 return NULL;
099e26bd
DG
1083}
1084
099e26bd
DG
1085/*
1086 * This thread manage application communication.
1d4b027a
DG
1087 */
1088static void *thread_manage_apps(void *data)
099e26bd 1089{
139ac872 1090 int i, ret, pollfd, err = -1;
5eb91c98 1091 uint32_t revents, nb_fd;
099e26bd 1092 struct ust_command ust_cmd;
5eb91c98 1093 struct lttng_poll_event events;
099e26bd
DG
1094
1095 DBG("[thread] Manage application started");
1096
8ac94142
CB
1097 testpoint(thread_manage_apps);
1098
f6a9efaa
DG
1099 rcu_register_thread();
1100 rcu_thread_online();
1101
139ac872 1102 health_code_update(&health_thread_app_manage);
44a5e5eb 1103
5eb91c98
DG
1104 ret = create_thread_poll_set(&events, 2);
1105 if (ret < 0) {
76d7553f 1106 goto error_poll_create;
5eb91c98 1107 }
099e26bd 1108
5eb91c98
DG
1109 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1110 if (ret < 0) {
1111 goto error;
1112 }
099e26bd 1113
8ac94142
CB
1114 testpoint(thread_manage_apps_before_loop);
1115
139ac872 1116 health_code_update(&health_thread_app_manage);
44a5e5eb 1117
5eb91c98
DG
1118 while (1) {
1119 /* Zeroed the events structure */
1120 lttng_poll_reset(&events);
0177d773 1121
5eb91c98 1122 nb_fd = LTTNG_POLL_GETNB(&events);
099e26bd
DG
1123
1124 DBG("Apps thread polling on %d fds", nb_fd);
1125
1126 /* Inifinite blocking call, waiting for transmission */
88f2b785 1127 restart:
139ac872 1128 health_poll_update(&health_thread_app_manage);
5eb91c98 1129 ret = lttng_poll_wait(&events, -1);
139ac872 1130 health_poll_update(&health_thread_app_manage);
099e26bd 1131 if (ret < 0) {
88f2b785
MD
1132 /*
1133 * Restart interrupted system call.
1134 */
1135 if (errno == EINTR) {
1136 goto restart;
1137 }
099e26bd
DG
1138 goto error;
1139 }
1140
5eb91c98
DG
1141 for (i = 0; i < nb_fd; i++) {
1142 /* Fetch once the poll data */
1143 revents = LTTNG_POLL_GETEV(&events, i);
1144 pollfd = LTTNG_POLL_GETFD(&events, i);
1145
139ac872 1146 health_code_update(&health_thread_app_manage);
44a5e5eb 1147
5eb91c98
DG
1148 /* Thread quit pipe has been closed. Killing thread. */
1149 ret = check_thread_quit_pipe(pollfd, revents);
1150 if (ret) {
139ac872
MD
1151 err = 0;
1152 goto exit;
5eb91c98 1153 }
099e26bd 1154
5eb91c98
DG
1155 /* Inspect the apps cmd pipe */
1156 if (pollfd == apps_cmd_pipe[0]) {
1157 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1158 ERR("Apps command pipe error");
0177d773 1159 goto error;
5eb91c98
DG
1160 } else if (revents & LPOLLIN) {
1161 /* Empty pipe */
1162 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1163 if (ret < 0 || ret < sizeof(ust_cmd)) {
76d7553f 1164 PERROR("read apps cmd pipe");
5eb91c98
DG
1165 goto error;
1166 }
099e26bd 1167
139ac872 1168 health_code_update(&health_thread_app_manage);
44a5e5eb 1169
5eb91c98 1170 /* Register applicaton to the session daemon */
56fff090 1171 ret = ust_app_register(&ust_cmd.reg_msg,
54d01ffb 1172 ust_cmd.sock);
88ff5b7f 1173 if (ret == -ENOMEM) {
5eb91c98 1174 goto error;
88ff5b7f
MD
1175 } else if (ret < 0) {
1176 break;
5eb91c98
DG
1177 }
1178
139ac872 1179 health_code_update(&health_thread_app_manage);
44a5e5eb 1180
acc7b41b 1181 /*
e0c7ec2b 1182 * Validate UST version compatibility.
acc7b41b 1183 */
e0c7ec2b
DG
1184 ret = ust_app_validate_version(ust_cmd.sock);
1185 if (ret >= 0) {
1186 /*
1187 * Add channel(s) and event(s) to newly registered apps
1188 * from lttng global UST domain.
1189 */
1190 update_ust_app(ust_cmd.sock);
1191 }
acc7b41b 1192
139ac872 1193 health_code_update(&health_thread_app_manage);
44a5e5eb 1194
f2ca2e25 1195 ret = ust_app_register_done(ust_cmd.sock);
5eb91c98
DG
1196 if (ret < 0) {
1197 /*
1198 * If the registration is not possible, we simply
1199 * unregister the apps and continue
1200 */
56fff090 1201 ust_app_unregister(ust_cmd.sock);
5eb91c98
DG
1202 } else {
1203 /*
622c9ecf
DG
1204 * We only monitor the error events of the socket. This
1205 * thread does not handle any incoming data from UST
1206 * (POLLIN).
5eb91c98 1207 */
622c9ecf
DG
1208 ret = lttng_poll_add(&events, ust_cmd.sock,
1209 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
5eb91c98
DG
1210 if (ret < 0) {
1211 goto error;
1212 }
1213
ae9e45b3
DG
1214 /* Set socket timeout for both receiving and ending */
1215 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1216 app_socket_timeout);
1217 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1218 app_socket_timeout);
1219
54d01ffb
DG
1220 DBG("Apps with sock %d added to poll set",
1221 ust_cmd.sock);
5eb91c98 1222 }
487cf67c 1223
139ac872 1224 health_code_update(&health_thread_app_manage);
44a5e5eb 1225
5eb91c98 1226 break;
0177d773 1227 }
5eb91c98
DG
1228 } else {
1229 /*
54d01ffb
DG
1230 * At this point, we know that a registered application made
1231 * the event at poll_wait.
5eb91c98
DG
1232 */
1233 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1234 /* Removing from the poll set */
1235 ret = lttng_poll_del(&events, pollfd);
1236 if (ret < 0) {
1237 goto error;
1238 }
099e26bd 1239
b9d9b220 1240 /* Socket closed on remote end. */
56fff090 1241 ust_app_unregister(pollfd);
5eb91c98
DG
1242 break;
1243 }
099e26bd 1244 }
44a5e5eb 1245
139ac872 1246 health_code_update(&health_thread_app_manage);
099e26bd 1247 }
099e26bd
DG
1248 }
1249
139ac872 1250exit:
099e26bd 1251error:
5eb91c98 1252 lttng_poll_clean(&events);
76d7553f 1253error_poll_create:
6620da75
DG
1254 utils_close_pipe(apps_cmd_pipe);
1255 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1256
1257 /*
1258 * We don't clean the UST app hash table here since already registered
1259 * applications can still be controlled so let them be until the session
1260 * daemon dies or the applications stop.
1261 */
1262
139ac872
MD
1263 if (err) {
1264 health_error(&health_thread_app_manage);
1265 ERR("Health error occurred in %s", __func__);
1266 }
1267 health_exit(&health_thread_app_manage);
76d7553f 1268 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1269 rcu_thread_offline();
1270 rcu_unregister_thread();
099e26bd
DG
1271 return NULL;
1272}
1273
1274/*
1275 * Dispatch request from the registration threads to the application
1276 * communication thread.
1277 */
1278static void *thread_dispatch_ust_registration(void *data)
1279{
1280 int ret;
1281 struct cds_wfq_node *node;
1282 struct ust_command *ust_cmd = NULL;
1283
1284 DBG("[thread] Dispatch UST command started");
1285
26c9d55e 1286 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
099e26bd
DG
1287 /* Atomically prepare the queue futex */
1288 futex_nto1_prepare(&ust_cmd_queue.futex);
1289
1290 do {
1291 /* Dequeue command for registration */
1292 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1293 if (node == NULL) {
00a17c97 1294 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1295 /* Continue thread execution */
1296 break;
1297 }
1298
1299 ust_cmd = caa_container_of(node, struct ust_command, node);
1300
2f50c8a3
DG
1301 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1302 " gid:%d sock:%d name:%s (version %d.%d)",
1303 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1304 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1305 ust_cmd->sock, ust_cmd->reg_msg.name,
1306 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
099e26bd
DG
1307 /*
1308 * Inform apps thread of the new application registration. This
1309 * call is blocking so we can be assured that the data will be read
1310 * at some point in time or wait to the end of the world :)
1311 */
6620da75
DG
1312 if (apps_cmd_pipe[1] >= 0) {
1313 ret = write(apps_cmd_pipe[1], ust_cmd,
1314 sizeof(struct ust_command));
1315 if (ret < 0) {
1316 PERROR("write apps cmd pipe");
1317 if (errno == EBADF) {
1318 /*
1319 * We can't inform the application thread to process
1320 * registration. We will exit or else application
1321 * registration will not occur and tracing will never
1322 * start.
1323 */
1324 goto error;
1325 }
1326 }
1327 } else {
1328 /* Application manager thread is not available. */
1329 ret = close(ust_cmd->sock);
1330 if (ret < 0) {
1331 PERROR("close ust_cmd sock");
099e26bd
DG
1332 }
1333 }
1334 free(ust_cmd);
1335 } while (node != NULL);
1336
1337 /* Futex wait on queue. Blocking call on futex() */
1338 futex_nto1_wait(&ust_cmd_queue.futex);
1339 }
1340
1341error:
1342 DBG("Dispatch thread dying");
1343 return NULL;
1344}
1345
1346/*
1347 * This thread manage application registration.
1348 */
1349static void *thread_registration_apps(void *data)
1d4b027a 1350{
139ac872 1351 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
1352 uint32_t revents, nb_fd;
1353 struct lttng_poll_event events;
099e26bd
DG
1354 /*
1355 * Get allocated in this thread, enqueued to a global queue, dequeued and
1356 * freed in the manage apps thread.
1357 */
1358 struct ust_command *ust_cmd = NULL;
1d4b027a 1359
099e26bd 1360 DBG("[thread] Manage application registration started");
1d4b027a 1361
8ac94142
CB
1362 testpoint(thread_registration_apps);
1363
1d4b027a
DG
1364 ret = lttcomm_listen_unix_sock(apps_sock);
1365 if (ret < 0) {
76d7553f 1366 goto error_listen;
1d4b027a
DG
1367 }
1368
5eb91c98
DG
1369 /*
1370 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1371 * more will be added to this poll set.
1372 */
1373 ret = create_thread_poll_set(&events, 2);
1374 if (ret < 0) {
76d7553f 1375 goto error_create_poll;
5eb91c98 1376 }
273ea72c 1377
5eb91c98
DG
1378 /* Add the application registration socket */
1379 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1380 if (ret < 0) {
76d7553f 1381 goto error_poll_add;
5eb91c98 1382 }
273ea72c 1383
1d4b027a 1384 /* Notify all applications to register */
0fdd1e2c
DG
1385 ret = notify_ust_apps(1);
1386 if (ret < 0) {
1387 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
1388 "Execution continues but there might be problem for already\n"
1389 "running applications that wishes to register.");
0fdd1e2c 1390 }
1d4b027a
DG
1391
1392 while (1) {
1393 DBG("Accepting application registration");
273ea72c 1394
5eb91c98
DG
1395 nb_fd = LTTNG_POLL_GETNB(&events);
1396
273ea72c 1397 /* Inifinite blocking call, waiting for transmission */
88f2b785 1398 restart:
139ac872 1399 health_poll_update(&health_thread_app_reg);
5eb91c98 1400 ret = lttng_poll_wait(&events, -1);
139ac872 1401 health_poll_update(&health_thread_app_reg);
273ea72c 1402 if (ret < 0) {
88f2b785
MD
1403 /*
1404 * Restart interrupted system call.
1405 */
1406 if (errno == EINTR) {
1407 goto restart;
1408 }
273ea72c
DG
1409 goto error;
1410 }
1411
5eb91c98 1412 for (i = 0; i < nb_fd; i++) {
139ac872
MD
1413 health_code_update(&health_thread_app_reg);
1414
5eb91c98
DG
1415 /* Fetch once the poll data */
1416 revents = LTTNG_POLL_GETEV(&events, i);
1417 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 1418
5eb91c98
DG
1419 /* Thread quit pipe has been closed. Killing thread. */
1420 ret = check_thread_quit_pipe(pollfd, revents);
1421 if (ret) {
139ac872
MD
1422 err = 0;
1423 goto exit;
90014c57 1424 }
1d4b027a 1425
5eb91c98
DG
1426 /* Event on the registration socket */
1427 if (pollfd == apps_sock) {
1428 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1429 ERR("Register apps socket poll error");
1430 goto error;
1431 } else if (revents & LPOLLIN) {
1432 sock = lttcomm_accept_unix_sock(apps_sock);
1433 if (sock < 0) {
1434 goto error;
1435 }
099e26bd 1436
b662582b
DG
1437 /*
1438 * Set the CLOEXEC flag. Return code is useless because
1439 * either way, the show must go on.
1440 */
1441 (void) utils_set_fd_cloexec(sock);
1442
5eb91c98 1443 /* Create UST registration command for enqueuing */
ba7f0ae5 1444 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 1445 if (ust_cmd == NULL) {
76d7553f 1446 PERROR("ust command zmalloc");
5eb91c98
DG
1447 goto error;
1448 }
1d4b027a 1449
5eb91c98
DG
1450 /*
1451 * Using message-based transmissions to ensure we don't
1452 * have to deal with partially received messages.
1453 */
4063050c
MD
1454 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1455 if (ret < 0) {
1456 ERR("Exhausted file descriptors allowed for applications.");
1457 free(ust_cmd);
1458 ret = close(sock);
1459 if (ret) {
1460 PERROR("close");
1461 }
1462 sock = -1;
1463 continue;
1464 }
139ac872 1465 health_code_update(&health_thread_app_reg);
5eb91c98
DG
1466 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1467 sizeof(struct ust_register_msg));
1468 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1469 if (ret < 0) {
76d7553f 1470 PERROR("lttcomm_recv_unix_sock register apps");
5eb91c98
DG
1471 } else {
1472 ERR("Wrong size received on apps register");
1473 }
1474 free(ust_cmd);
76d7553f
MD
1475 ret = close(sock);
1476 if (ret) {
1477 PERROR("close");
1478 }
4063050c 1479 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 1480 sock = -1;
5eb91c98
DG
1481 continue;
1482 }
139ac872 1483 health_code_update(&health_thread_app_reg);
099e26bd 1484
5eb91c98 1485 ust_cmd->sock = sock;
34a2494f 1486 sock = -1;
099e26bd 1487
5eb91c98
DG
1488 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1489 " gid:%d sock:%d name:%s (version %d.%d)",
1490 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1491 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1492 ust_cmd->sock, ust_cmd->reg_msg.name,
1493 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 1494
5eb91c98
DG
1495 /*
1496 * Lock free enqueue the registration request. The red pill
54d01ffb 1497 * has been taken! This apps will be part of the *system*.
5eb91c98
DG
1498 */
1499 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1500
1501 /*
1502 * Wake the registration queue futex. Implicit memory
1503 * barrier with the exchange in cds_wfq_enqueue.
1504 */
1505 futex_nto1_wake(&ust_cmd_queue.futex);
1506 }
1507 }
90014c57 1508 }
1d4b027a
DG
1509 }
1510
139ac872 1511exit:
1d4b027a 1512error:
139ac872
MD
1513 if (err) {
1514 health_error(&health_thread_app_reg);
1515 ERR("Health error occurred in %s", __func__);
1516 }
1517 health_exit(&health_thread_app_reg);
1518
0fdd1e2c
DG
1519 /* Notify that the registration thread is gone */
1520 notify_ust_apps(0);
1521
a4b35e07 1522 if (apps_sock >= 0) {
76d7553f
MD
1523 ret = close(apps_sock);
1524 if (ret) {
1525 PERROR("close");
1526 }
a4b35e07 1527 }
46c3f085 1528 if (sock >= 0) {
76d7553f
MD
1529 ret = close(sock);
1530 if (ret) {
1531 PERROR("close");
1532 }
4063050c 1533 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 1534 }
273ea72c 1535 unlink(apps_unix_sock_path);
0fdd1e2c 1536
76d7553f 1537error_poll_add:
5eb91c98 1538 lttng_poll_clean(&events);
76d7553f
MD
1539error_listen:
1540error_create_poll:
1541 DBG("UST Registration thread cleanup complete");
5eb91c98 1542
1d4b027a
DG
1543 return NULL;
1544}
1545
8c0faa1d 1546/*
3bd1e081 1547 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 1548 * exec or it will fails.
8c0faa1d 1549 */
3bd1e081 1550static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 1551{
a23ec3a7 1552 int ret, clock_ret;
ee0b0061
DG
1553 struct timespec timeout;
1554
a23ec3a7
DG
1555 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1556 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 1557
a23ec3a7
DG
1558 /* Setup pthread condition */
1559 ret = pthread_condattr_init(&consumer_data->condattr);
1560 if (ret != 0) {
1561 errno = ret;
1562 PERROR("pthread_condattr_init consumer data");
1563 goto error;
1564 }
1565
1566 /*
1567 * Set the monotonic clock in order to make sure we DO NOT jump in time
1568 * between the clock_gettime() call and the timedwait call. See bug #324
1569 * for a more details and how we noticed it.
1570 */
1571 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1572 if (ret != 0) {
1573 errno = ret;
1574 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
1575 goto error;
1576 }
8c0faa1d 1577
a23ec3a7
DG
1578 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1579 if (ret != 0) {
1580 errno = ret;
1581 PERROR("pthread_cond_init consumer data");
1582 goto error;
1583 }
1584
1585 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1586 consumer_data);
8c0faa1d 1587 if (ret != 0) {
3bd1e081 1588 PERROR("pthread_create consumer");
ee0b0061 1589 ret = -1;
8c0faa1d
DG
1590 goto error;
1591 }
1592
a23ec3a7
DG
1593 /* We are about to wait on a pthread condition */
1594 pthread_mutex_lock(&consumer_data->cond_mutex);
1595
ee0b0061 1596 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
1597 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1598 /*
1599 * Set the timeout for the condition timed wait even if the clock gettime
1600 * call fails since we might loop on that call and we want to avoid to
1601 * increment the timeout too many times.
1602 */
1603 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1604
1605 /*
1606 * The following loop COULD be skipped in some conditions so this is why we
1607 * set ret to 0 in order to make sure at least one round of the loop is
1608 * done.
1609 */
1610 ret = 0;
1611
1612 /*
1613 * Loop until the condition is reached or when a timeout is reached. Note
1614 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1615 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1616 * possible. This loop does not take any chances and works with both of
1617 * them.
1618 */
1619 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1620 if (clock_ret < 0) {
1621 PERROR("clock_gettime spawn consumer");
1622 /* Infinite wait for the consumerd thread to be ready */
1623 ret = pthread_cond_wait(&consumer_data->cond,
1624 &consumer_data->cond_mutex);
1625 } else {
1626 ret = pthread_cond_timedwait(&consumer_data->cond,
1627 &consumer_data->cond_mutex, &timeout);
1628 }
ee0b0061 1629 }
8c0faa1d 1630
a23ec3a7
DG
1631 /* Release the pthread condition */
1632 pthread_mutex_unlock(&consumer_data->cond_mutex);
1633
1634 if (ret != 0) {
1635 errno = ret;
1636 if (ret == ETIMEDOUT) {
ee0b0061
DG
1637 /*
1638 * Call has timed out so we kill the kconsumerd_thread and return
1639 * an error.
1640 */
a23ec3a7
DG
1641 ERR("Condition timed out. The consumer thread was never ready."
1642 " Killing it");
3bd1e081 1643 ret = pthread_cancel(consumer_data->thread);
ee0b0061 1644 if (ret < 0) {
3bd1e081 1645 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
1646 }
1647 } else {
a23ec3a7 1648 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061
DG
1649 }
1650 goto error;
1651 }
1652
3bd1e081
MD
1653 pthread_mutex_lock(&consumer_data->pid_mutex);
1654 if (consumer_data->pid == 0) {
a23ec3a7 1655 ERR("Consumerd did not start");
3bd1e081 1656 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
1657 goto error;
1658 }
3bd1e081 1659 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 1660
8c0faa1d
DG
1661 return 0;
1662
1663error:
1664 return ret;
1665}
1666
d9800920 1667/*
3bd1e081 1668 * Join consumer thread
d9800920 1669 */
3bd1e081 1670static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
1671{
1672 void *status;
1673 int ret;
1674
e8209f6b
DG
1675 /* Consumer pid must be a real one. */
1676 if (consumer_data->pid > 0) {
3bd1e081 1677 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 1678 if (ret) {
3bd1e081 1679 ERR("Error killing consumer daemon");
cf3af59e
MD
1680 return ret;
1681 }
3bd1e081 1682 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
1683 } else {
1684 return 0;
1685 }
1686}
1687
8c0faa1d 1688/*
3bd1e081 1689 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 1690 *
d063d709 1691 * Return pid if successful else -1.
8c0faa1d 1692 */
3bd1e081 1693static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
1694{
1695 int ret;
1696 pid_t pid;
94c55f17 1697 const char *consumer_to_use;
53086306 1698 const char *verbosity;
94c55f17 1699 struct stat st;
8c0faa1d 1700
3bd1e081 1701 DBG("Spawning consumerd");
c49dc785 1702
8c0faa1d
DG
1703 pid = fork();
1704 if (pid == 0) {
1705 /*
3bd1e081 1706 * Exec consumerd.
8c0faa1d 1707 */
daee5345 1708 if (opt_verbose_consumer) {
53086306
DG
1709 verbosity = "--verbose";
1710 } else {
1711 verbosity = "--quiet";
1712 }
3bd1e081
MD
1713 switch (consumer_data->type) {
1714 case LTTNG_CONSUMER_KERNEL:
94c55f17 1715 /*
c7704d57
DG
1716 * Find out which consumerd to execute. We will first try the
1717 * 64-bit path, then the sessiond's installation directory, and
1718 * fallback on the 32-bit one,
94c55f17 1719 */
63a799e8
AM
1720 DBG3("Looking for a kernel consumer at these locations:");
1721 DBG3(" 1) %s", consumerd64_bin);
1722 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1723 DBG3(" 3) %s", consumerd32_bin);
94c55f17 1724 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 1725 DBG3("Found location #1");
94c55f17 1726 consumer_to_use = consumerd64_bin;
94c55f17 1727 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 1728 DBG3("Found location #2");
94c55f17 1729 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 1730 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 1731 DBG3("Found location #3");
eb1e0bd4 1732 consumer_to_use = consumerd32_bin;
94c55f17 1733 } else {
63a799e8 1734 DBG("Could not find any valid consumerd executable");
94c55f17
AM
1735 break;
1736 }
1737 DBG("Using kernel consumer at: %s", consumer_to_use);
1738 execl(consumer_to_use,
1739 "lttng-consumerd", verbosity, "-k",
1740 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1741 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1742 NULL);
3bd1e081 1743 break;
7753dea8
MD
1744 case LTTNG_CONSUMER64_UST:
1745 {
b1e0b6b6 1746 char *tmpnew = NULL;
8f4905da
MD
1747
1748 if (consumerd64_libdir[0] != '\0') {
1749 char *tmp;
1750 size_t tmplen;
1751
1752 tmp = getenv("LD_LIBRARY_PATH");
1753 if (!tmp) {
1754 tmp = "";
1755 }
1756 tmplen = strlen("LD_LIBRARY_PATH=")
1757 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1758 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1759 if (!tmpnew) {
1760 ret = -ENOMEM;
1761 goto error;
1762 }
1763 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1764 strcat(tmpnew, consumerd64_libdir);
1765 if (tmp[0] != '\0') {
1766 strcat(tmpnew, ":");
1767 strcat(tmpnew, tmp);
1768 }
1769 ret = putenv(tmpnew);
1770 if (ret) {
1771 ret = -errno;
1772 goto error;
1773 }
1774 }
94c55f17 1775 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 1776 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
1777 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1778 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1779 NULL);
8f4905da
MD
1780 if (consumerd64_libdir[0] != '\0') {
1781 free(tmpnew);
1782 }
1783 if (ret) {
1784 goto error;
1785 }
3bd1e081 1786 break;
7753dea8
MD
1787 }
1788 case LTTNG_CONSUMER32_UST:
1789 {
937dde8e 1790 char *tmpnew = NULL;
8f4905da
MD
1791
1792 if (consumerd32_libdir[0] != '\0') {
1793 char *tmp;
1794 size_t tmplen;
1795
1796 tmp = getenv("LD_LIBRARY_PATH");
1797 if (!tmp) {
1798 tmp = "";
1799 }
1800 tmplen = strlen("LD_LIBRARY_PATH=")
1801 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1802 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1803 if (!tmpnew) {
1804 ret = -ENOMEM;
1805 goto error;
1806 }
1807 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1808 strcat(tmpnew, consumerd32_libdir);
1809 if (tmp[0] != '\0') {
1810 strcat(tmpnew, ":");
1811 strcat(tmpnew, tmp);
1812 }
1813 ret = putenv(tmpnew);
1814 if (ret) {
1815 ret = -errno;
1816 goto error;
1817 }
1818 }
94c55f17 1819 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 1820 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
1821 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1822 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1823 NULL);
8f4905da
MD
1824 if (consumerd32_libdir[0] != '\0') {
1825 free(tmpnew);
1826 }
1827 if (ret) {
1828 goto error;
1829 }
7753dea8
MD
1830 break;
1831 }
3bd1e081 1832 default:
76d7553f 1833 PERROR("unknown consumer type");
3bd1e081
MD
1834 exit(EXIT_FAILURE);
1835 }
8c0faa1d 1836 if (errno != 0) {
76d7553f 1837 PERROR("kernel start consumer exec");
8c0faa1d
DG
1838 }
1839 exit(EXIT_FAILURE);
1840 } else if (pid > 0) {
1841 ret = pid;
8c0faa1d 1842 } else {
76d7553f 1843 PERROR("start consumer fork");
8c0faa1d 1844 ret = -errno;
8c0faa1d 1845 }
8f4905da 1846error:
8c0faa1d
DG
1847 return ret;
1848}
1849
693bd40b 1850/*
3bd1e081 1851 * Spawn the consumerd daemon and session daemon thread.
693bd40b 1852 */
3bd1e081 1853static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 1854{
edb8b045
DG
1855 int ret, err;
1856
1857 /*
1858 * Set the listen() state on the socket since there is a possible race
1859 * between the exec() of the consumer daemon and this call if place in the
1860 * consumer thread. See bug #366 for more details.
1861 */
1862 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1863 if (ret < 0) {
1864 goto error;
1865 }
693bd40b 1866
3bd1e081
MD
1867 pthread_mutex_lock(&consumer_data->pid_mutex);
1868 if (consumer_data->pid != 0) {
1869 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
1870 goto end;
1871 }
693bd40b 1872
3bd1e081 1873 ret = spawn_consumerd(consumer_data);
c49dc785 1874 if (ret < 0) {
3bd1e081
MD
1875 ERR("Spawning consumerd failed");
1876 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 1877 goto error;
693bd40b 1878 }
c49dc785 1879
3bd1e081
MD
1880 /* Setting up the consumer_data pid */
1881 consumer_data->pid = ret;
48842b30 1882 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 1883 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 1884
3bd1e081
MD
1885 DBG2("Spawning consumer control thread");
1886 ret = spawn_consumer_thread(consumer_data);
693bd40b 1887 if (ret < 0) {
3bd1e081 1888 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
1889 goto error;
1890 }
1891
c49dc785 1892end:
693bd40b
DG
1893 return 0;
1894
1895error:
edb8b045
DG
1896 /* Cleanup already created socket on error. */
1897 if (consumer_data->err_sock >= 0) {
1898 err = close(consumer_data->err_sock);
1899 if (err < 0) {
1900 PERROR("close consumer data error socket");
1901 }
1902 }
693bd40b
DG
1903 return ret;
1904}
1905
44a5e5eb 1906/*
139ac872
MD
1907 * Compute health status of each consumer. If one of them is zero (bad
1908 * state), we return 0.
44a5e5eb
DG
1909 */
1910static int check_consumer_health(void)
1911{
1912 int ret;
1913
139ac872
MD
1914 ret = health_check_state(&kconsumer_data.health) &&
1915 health_check_state(&ustconsumer32_data.health) &&
44a5e5eb
DG
1916 health_check_state(&ustconsumer64_data.health);
1917
1918 DBG3("Health consumer check %d", ret);
1919
1920 return ret;
1921}
1922
b73401da 1923/*
096102bd 1924 * Setup necessary data for kernel tracer action.
b73401da 1925 */
096102bd 1926static int init_kernel_tracer(void)
b73401da
DG
1927{
1928 int ret;
b73401da 1929
096102bd
DG
1930 /* Modprobe lttng kernel modules */
1931 ret = modprobe_lttng_control();
b73401da 1932 if (ret < 0) {
b73401da
DG
1933 goto error;
1934 }
1935
096102bd
DG
1936 /* Open debugfs lttng */
1937 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1938 if (kernel_tracer_fd < 0) {
1939 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
1940 ret = -1;
1941 goto error_open;
54d01ffb
DG
1942 }
1943
2f77fc4b
DG
1944 /* Validate kernel version */
1945 ret = kernel_validate_version(kernel_tracer_fd);
1946 if (ret < 0) {
1947 goto error_version;
b551a063 1948 }
54d01ffb 1949
2f77fc4b
DG
1950 ret = modprobe_lttng_data();
1951 if (ret < 0) {
1952 goto error_modules;
54d01ffb
DG
1953 }
1954
2f77fc4b
DG
1955 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1956 return 0;
1957
1958error_version:
1959 modprobe_remove_lttng_control();
1960 ret = close(kernel_tracer_fd);
1961 if (ret) {
1962 PERROR("close");
b551a063 1963 }
2f77fc4b 1964 kernel_tracer_fd = -1;
f73fabfd 1965 return LTTNG_ERR_KERN_VERSION;
b551a063 1966
2f77fc4b
DG
1967error_modules:
1968 ret = close(kernel_tracer_fd);
1969 if (ret) {
1970 PERROR("close");
b551a063 1971 }
54d01ffb 1972
2f77fc4b
DG
1973error_open:
1974 modprobe_remove_lttng_control();
54d01ffb
DG
1975
1976error:
2f77fc4b
DG
1977 WARN("No kernel tracer available");
1978 kernel_tracer_fd = -1;
1979 if (!is_root) {
f73fabfd 1980 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 1981 } else {
f73fabfd 1982 return LTTNG_ERR_KERN_NA;
2f77fc4b 1983 }
54d01ffb
DG
1984}
1985
2f77fc4b 1986
54d01ffb 1987/*
2f77fc4b
DG
1988 * Copy consumer output from the tracing session to the domain session. The
1989 * function also applies the right modification on a per domain basis for the
1990 * trace files destination directory.
54d01ffb 1991 */
2f77fc4b 1992static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
1993{
1994 int ret;
2f77fc4b
DG
1995 const char *dir_name;
1996 struct consumer_output *consumer;
1997
1998 assert(session);
1999 assert(session->consumer);
54d01ffb 2000
b551a063
DG
2001 switch (domain) {
2002 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2003 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2004 /*
2005 * XXX: We should audit the session creation and what this function
2006 * does "extra" in order to avoid a destroy since this function is used
2007 * in the domain session creation (kernel and ust) only. Same for UST
2008 * domain.
2009 */
2010 if (session->kernel_session->consumer) {
2011 consumer_destroy_output(session->kernel_session->consumer);
2012 }
2f77fc4b
DG
2013 session->kernel_session->consumer =
2014 consumer_copy_output(session->consumer);
2015 /* Ease our life a bit for the next part */
2016 consumer = session->kernel_session->consumer;
2017 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063
DG
2018 break;
2019 case LTTNG_DOMAIN_UST:
2f77fc4b 2020 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2021 if (session->ust_session->consumer) {
2022 consumer_destroy_output(session->ust_session->consumer);
2023 }
2f77fc4b
DG
2024 session->ust_session->consumer =
2025 consumer_copy_output(session->consumer);
2026 /* Ease our life a bit for the next part */
2027 consumer = session->ust_session->consumer;
2028 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2029 break;
2030 default:
f73fabfd 2031 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2032 goto error;
2033 }
2034
2f77fc4b 2035 /* Append correct directory to subdir */
c30ce0b3
CB
2036 strncat(consumer->subdir, dir_name,
2037 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2038 DBG3("Copy session consumer subdir %s", consumer->subdir);
2039
f73fabfd 2040 ret = LTTNG_OK;
54d01ffb
DG
2041
2042error:
2043 return ret;
2044}
2045
00e2e675 2046/*
2f77fc4b 2047 * Create an UST session and add it to the session ust list.
00e2e675 2048 */
2f77fc4b
DG
2049static int create_ust_session(struct ltt_session *session,
2050 struct lttng_domain *domain)
00e2e675
DG
2051{
2052 int ret;
2f77fc4b 2053 struct ltt_ust_session *lus = NULL;
00e2e675 2054
a4b92340 2055 assert(session);
2f77fc4b
DG
2056 assert(domain);
2057 assert(session->consumer);
a4b92340 2058
2f77fc4b
DG
2059 switch (domain->type) {
2060 case LTTNG_DOMAIN_UST:
2061 break;
2062 default:
2063 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2064 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2065 goto error;
2066 }
2067
2f77fc4b
DG
2068 DBG("Creating UST session");
2069
2070 lus = trace_ust_create_session(session->path, session->id, domain);
2071 if (lus == NULL) {
f73fabfd 2072 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2073 goto error;
2074 }
2075
2f77fc4b
DG
2076 lus->uid = session->uid;
2077 lus->gid = session->gid;
2078 session->ust_session = lus;
00e2e675 2079
2f77fc4b
DG
2080 /* Copy session output to the newly created UST session */
2081 ret = copy_session_consumer(domain->type, session);
f73fabfd 2082 if (ret != LTTNG_OK) {
00e2e675
DG
2083 goto error;
2084 }
2085
f73fabfd 2086 return LTTNG_OK;
00e2e675
DG
2087
2088error:
2f77fc4b
DG
2089 free(lus);
2090 session->ust_session = NULL;
00e2e675
DG
2091 return ret;
2092}
2093
2094/*
2f77fc4b 2095 * Create a kernel tracer session then create the default channel.
00e2e675 2096 */
2f77fc4b 2097static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2098{
2099 int ret;
a4b92340 2100
2f77fc4b 2101 DBG("Creating kernel session");
00e2e675 2102
2f77fc4b
DG
2103 ret = kernel_create_session(session, kernel_tracer_fd);
2104 if (ret < 0) {
f73fabfd 2105 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2106 goto error;
2107 }
2108
2f77fc4b
DG
2109 /* Code flow safety */
2110 assert(session->kernel_session);
2111
2112 /* Copy session output to the newly created Kernel session */
2113 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2114 if (ret != LTTNG_OK) {
a4b92340
DG
2115 goto error;
2116 }
2117
2f77fc4b
DG
2118 /* Create directory(ies) on local filesystem. */
2119 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2120 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2121 ret = run_as_mkdir_recursive(
2122 session->kernel_session->consumer->dst.trace_path,
2123 S_IRWXU | S_IRWXG, session->uid, session->gid);
2124 if (ret < 0) {
2125 if (ret != -EEXIST) {
2126 ERR("Trace directory creation error");
00e2e675
DG
2127 goto error;
2128 }
00e2e675 2129 }
2f77fc4b 2130 }
00e2e675 2131
2f77fc4b
DG
2132 session->kernel_session->uid = session->uid;
2133 session->kernel_session->gid = session->gid;
00e2e675 2134
f73fabfd 2135 return LTTNG_OK;
00e2e675 2136
2f77fc4b
DG
2137error:
2138 trace_kernel_destroy_session(session->kernel_session);
2139 session->kernel_session = NULL;
2140 return ret;
2141}
00e2e675 2142
2f77fc4b
DG
2143/*
2144 * Count number of session permitted by uid/gid.
2145 */
2146static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2147{
2148 unsigned int i = 0;
2149 struct ltt_session *session;
07424f16 2150
2f77fc4b
DG
2151 DBG("Counting number of available session for UID %d GID %d",
2152 uid, gid);
2153 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2154 /*
2f77fc4b 2155 * Only list the sessions the user can control.
00e2e675 2156 */
2f77fc4b
DG
2157 if (!session_access_ok(session, uid, gid)) {
2158 continue;
2159 }
2160 i++;
a4b92340 2161 }
2f77fc4b 2162 return i;
00e2e675
DG
2163}
2164
54d01ffb
DG
2165/*
2166 * Process the command requested by the lttng client within the command
2167 * context structure. This function make sure that the return structure (llm)
2168 * is set and ready for transmission before returning.
2169 *
2170 * Return any error encountered or 0 for success.
53a80697
MD
2171 *
2172 * "sock" is only used for special-case var. len data.
54d01ffb 2173 */
53a80697
MD
2174static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2175 int *sock_error)
54d01ffb 2176{
f73fabfd 2177 int ret = LTTNG_OK;
44d3bd01 2178 int need_tracing_session = 1;
2e09ba09 2179 int need_domain;
54d01ffb
DG
2180
2181 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2182
53a80697
MD
2183 *sock_error = 0;
2184
2e09ba09
MD
2185 switch (cmd_ctx->lsm->cmd_type) {
2186 case LTTNG_CREATE_SESSION:
2187 case LTTNG_DESTROY_SESSION:
2188 case LTTNG_LIST_SESSIONS:
2189 case LTTNG_LIST_DOMAINS:
2190 case LTTNG_START_TRACE:
2191 case LTTNG_STOP_TRACE:
6d805429 2192 case LTTNG_DATA_PENDING:
2e09ba09 2193 need_domain = 0;
3aace903 2194 break;
2e09ba09
MD
2195 default:
2196 need_domain = 1;
2197 }
2198
2199 if (opt_no_kernel && need_domain
2200 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2201 if (!is_root) {
f73fabfd 2202 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2203 } else {
f73fabfd 2204 ret = LTTNG_ERR_KERN_NA;
531d29f9 2205 }
4fba7219
DG
2206 goto error;
2207 }
2208
8d3113b2
DG
2209 /* Deny register consumer if we already have a spawned consumer. */
2210 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2211 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2212 if (kconsumer_data.pid > 0) {
f73fabfd 2213 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2214 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2215 goto error;
2216 }
2217 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2218 }
2219
54d01ffb
DG
2220 /*
2221 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2222 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2223 * command.
2224 */
2225 switch(cmd_ctx->lsm->cmd_type) {
2226 case LTTNG_LIST_SESSIONS:
2227 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2228 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2229 case LTTNG_LIST_DOMAINS:
2230 case LTTNG_LIST_CHANNELS:
2231 case LTTNG_LIST_EVENTS:
2232 break;
2233 default:
2234 /* Setup lttng message with no payload */
2235 ret = setup_lttng_msg(cmd_ctx, 0);
2236 if (ret < 0) {
2237 /* This label does not try to unlock the session */
2238 goto init_setup_error;
2239 }
2240 }
2241
2242 /* Commands that DO NOT need a session. */
2243 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2244 case LTTNG_CREATE_SESSION:
2e09ba09 2245 case LTTNG_CALIBRATE:
54d01ffb
DG
2246 case LTTNG_LIST_SESSIONS:
2247 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2248 case LTTNG_LIST_TRACEPOINT_FIELDS:
44d3bd01 2249 need_tracing_session = 0;
54d01ffb
DG
2250 break;
2251 default:
2252 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2253 /*
2254 * We keep the session list lock across _all_ commands
2255 * for now, because the per-session lock does not
2256 * handle teardown properly.
2257 */
74babd95 2258 session_lock_list();
54d01ffb
DG
2259 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2260 if (cmd_ctx->session == NULL) {
2261 if (cmd_ctx->lsm->session.name != NULL) {
f73fabfd 2262 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2263 } else {
2264 /* If no session name specified */
f73fabfd 2265 ret = LTTNG_ERR_SELECT_SESS;
54d01ffb
DG
2266 }
2267 goto error;
2268 } else {
2269 /* Acquire lock for the session */
2270 session_lock(cmd_ctx->session);
2271 }
2272 break;
2273 }
b389abbe 2274
2e09ba09
MD
2275 if (!need_domain) {
2276 goto skip_domain;
2277 }
a4b92340 2278
54d01ffb
DG
2279 /*
2280 * Check domain type for specific "pre-action".
2281 */
2282 switch (cmd_ctx->lsm->domain.type) {
2283 case LTTNG_DOMAIN_KERNEL:
d1f1c568 2284 if (!is_root) {
f73fabfd 2285 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
2286 goto error;
2287 }
2288
54d01ffb 2289 /* Kernel tracer check */
a4b35e07 2290 if (kernel_tracer_fd == -1) {
54d01ffb 2291 /* Basically, load kernel tracer modules */
096102bd
DG
2292 ret = init_kernel_tracer();
2293 if (ret != 0) {
54d01ffb
DG
2294 goto error;
2295 }
2296 }
5eb91c98 2297
5c827ce0
DG
2298 /* Consumer is in an ERROR state. Report back to client */
2299 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2300 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2301 goto error;
2302 }
2303
54d01ffb 2304 /* Need a session for kernel command */
44d3bd01 2305 if (need_tracing_session) {
54d01ffb 2306 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 2307 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 2308 if (ret < 0) {
f73fabfd 2309 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
2310 goto error;
2311 }
b389abbe 2312 }
7d29a247 2313
54d01ffb 2314 /* Start the kernel consumer daemon */
3bd1e081
MD
2315 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2316 if (kconsumer_data.pid == 0 &&
a4b92340
DG
2317 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2318 cmd_ctx->session->start_consumer) {
3bd1e081
MD
2319 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2320 ret = start_consumerd(&kconsumer_data);
7d29a247 2321 if (ret < 0) {
f73fabfd 2322 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 2323 goto error;
950131af 2324 }
5c827ce0 2325 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
2326 } else {
2327 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 2328 }
173af62f 2329
a4b92340
DG
2330 /*
2331 * The consumer was just spawned so we need to add the socket to
2332 * the consumer output of the session if exist.
2333 */
2334 ret = consumer_create_socket(&kconsumer_data,
2335 cmd_ctx->session->kernel_session->consumer);
2336 if (ret < 0) {
2337 goto error;
173af62f 2338 }
0d0c377a 2339 }
5c827ce0 2340
54d01ffb 2341 break;
2bdd86d4 2342 case LTTNG_DOMAIN_UST:
44d3bd01 2343 {
5c827ce0
DG
2344 /* Consumer is in an ERROR state. Report back to client */
2345 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2346 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2347 goto error;
2348 }
2349
44d3bd01 2350 if (need_tracing_session) {
a4b92340 2351 /* Create UST session if none exist. */
f6a9efaa 2352 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 2353 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 2354 &cmd_ctx->lsm->domain);
f73fabfd 2355 if (ret != LTTNG_OK) {
44d3bd01
DG
2356 goto error;
2357 }
2358 }
00e2e675 2359
7753dea8
MD
2360 /* Start the UST consumer daemons */
2361 /* 64-bit */
2362 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 2363 if (consumerd64_bin[0] != '\0' &&
7753dea8 2364 ustconsumer64_data.pid == 0 &&
a4b92340
DG
2365 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2366 cmd_ctx->session->start_consumer) {
7753dea8
MD
2367 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2368 ret = start_consumerd(&ustconsumer64_data);
2bdd86d4 2369 if (ret < 0) {
f73fabfd 2370 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
173af62f 2371 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2bdd86d4
MD
2372 goto error;
2373 }
48842b30 2374
173af62f 2375 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
5c827ce0 2376 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3ff2ecac 2377 } else {
7753dea8
MD
2378 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2379 }
173af62f
DG
2380
2381 /*
2382 * Setup socket for consumer 64 bit. No need for atomic access
2383 * since it was set above and can ONLY be set in this thread.
2384 */
a4b92340
DG
2385 ret = consumer_create_socket(&ustconsumer64_data,
2386 cmd_ctx->session->ust_session->consumer);
2387 if (ret < 0) {
2388 goto error;
173af62f
DG
2389 }
2390
7753dea8 2391 /* 32-bit */
fc7a59ce 2392 if (consumerd32_bin[0] != '\0' &&
7753dea8 2393 ustconsumer32_data.pid == 0 &&
a4b92340
DG
2394 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2395 cmd_ctx->session->start_consumer) {
7753dea8
MD
2396 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2397 ret = start_consumerd(&ustconsumer32_data);
2398 if (ret < 0) {
f73fabfd 2399 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
173af62f 2400 uatomic_set(&ust_consumerd32_fd, -EINVAL);
7753dea8
MD
2401 goto error;
2402 }
5c827ce0 2403
173af62f 2404 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
5c827ce0 2405 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
7753dea8
MD
2406 } else {
2407 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2bdd86d4 2408 }
173af62f
DG
2409
2410 /*
2411 * Setup socket for consumer 64 bit. No need for atomic access
2412 * since it was set above and can ONLY be set in this thread.
2413 */
a4b92340
DG
2414 ret = consumer_create_socket(&ustconsumer32_data,
2415 cmd_ctx->session->ust_session->consumer);
2416 if (ret < 0) {
2417 goto error;
173af62f 2418 }
44d3bd01
DG
2419 }
2420 break;
48842b30 2421 }
54d01ffb 2422 default:
54d01ffb
DG
2423 break;
2424 }
2e09ba09 2425skip_domain:
33a2b854 2426
5c827ce0
DG
2427 /* Validate consumer daemon state when start/stop trace command */
2428 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2429 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2430 switch (cmd_ctx->lsm->domain.type) {
2431 case LTTNG_DOMAIN_UST:
2432 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2433 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2434 goto error;
2435 }
2436 break;
2437 case LTTNG_DOMAIN_KERNEL:
2438 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2439 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2440 goto error;
2441 }
2442 break;
2443 }
2444 }
2445
8e0af1b4
MD
2446 /*
2447 * Check that the UID or GID match that of the tracing session.
2448 * The root user can interact with all sessions.
2449 */
2450 if (need_tracing_session) {
2451 if (!session_access_ok(cmd_ctx->session,
730389d9
DG
2452 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2453 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
f73fabfd 2454 ret = LTTNG_ERR_EPERM;
8e0af1b4
MD
2455 goto error;
2456 }
2457 }
2458
54d01ffb
DG
2459 /* Process by command type */
2460 switch (cmd_ctx->lsm->cmd_type) {
2461 case LTTNG_ADD_CONTEXT:
2462 {
2463 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2464 cmd_ctx->lsm->u.context.channel_name,
979e618e 2465 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
54d01ffb
DG
2466 break;
2467 }
2468 case LTTNG_DISABLE_CHANNEL:
2469 {
2470 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2471 cmd_ctx->lsm->u.disable.channel_name);
2472 break;
2473 }
2474 case LTTNG_DISABLE_EVENT:
2475 {
2476 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2477 cmd_ctx->lsm->u.disable.channel_name,
2478 cmd_ctx->lsm->u.disable.name);
33a2b854
DG
2479 break;
2480 }
54d01ffb
DG
2481 case LTTNG_DISABLE_ALL_EVENT:
2482 {
2bdd86d4 2483 DBG("Disabling all events");
54d01ffb
DG
2484
2485 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2486 cmd_ctx->lsm->u.disable.channel_name);
2487 break;
2488 }
00e2e675
DG
2489 case LTTNG_DISABLE_CONSUMER:
2490 {
2491 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2492 break;
2493 }
54d01ffb
DG
2494 case LTTNG_ENABLE_CHANNEL:
2495 {
aea829b3 2496 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 2497 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
54d01ffb
DG
2498 break;
2499 }
00e2e675
DG
2500 case LTTNG_ENABLE_CONSUMER:
2501 {
a4b92340
DG
2502 /*
2503 * XXX: 0 means that this URI should be applied on the session. Should
2504 * be a DOMAIN enuam.
2505 */
00e2e675 2506 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
f73fabfd 2507 if (ret != LTTNG_OK) {
a4b92340
DG
2508 goto error;
2509 }
2510
2511 if (cmd_ctx->lsm->domain.type == 0) {
2512 /* Add the URI for the UST session if a consumer is present. */
2513 if (cmd_ctx->session->ust_session &&
2514 cmd_ctx->session->ust_session->consumer) {
2515 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2516 } else if (cmd_ctx->session->kernel_session &&
2517 cmd_ctx->session->kernel_session->consumer) {
2518 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2519 cmd_ctx->session);
2520 }
2521 }
00e2e675
DG
2522 break;
2523 }
54d01ffb
DG
2524 case LTTNG_ENABLE_EVENT:
2525 {
2526 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2527 cmd_ctx->lsm->u.enable.channel_name,
2f77fc4b 2528 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
54d01ffb
DG
2529 break;
2530 }
2531 case LTTNG_ENABLE_ALL_EVENT:
2532 {
2bdd86d4 2533 DBG("Enabling all events");
54d01ffb
DG
2534
2535 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
8c9ae521 2536 cmd_ctx->lsm->u.enable.channel_name,
2f77fc4b 2537 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
54d01ffb
DG
2538 break;
2539 }
052da939 2540 case LTTNG_LIST_TRACEPOINTS:
2ef84c95 2541 {
9f19cc17 2542 struct lttng_event *events;
54d01ffb 2543 ssize_t nb_events;
052da939 2544
54d01ffb
DG
2545 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2546 if (nb_events < 0) {
f73fabfd 2547 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2548 ret = -nb_events;
2549 goto error;
2ef84c95
DG
2550 }
2551
2552 /*
2553 * Setup lttng message with payload size set to the event list size in
2554 * bytes and then copy list into the llm payload.
2555 */
052da939 2556 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2ef84c95 2557 if (ret < 0) {
052da939 2558 free(events);
2ef84c95
DG
2559 goto setup_error;
2560 }
2561
2562 /* Copy event list into message payload */
9f19cc17 2563 memcpy(cmd_ctx->llm->payload, events,
052da939 2564 sizeof(struct lttng_event) * nb_events);
2ef84c95 2565
9f19cc17 2566 free(events);
2ef84c95 2567
f73fabfd 2568 ret = LTTNG_OK;
2ef84c95
DG
2569 break;
2570 }
f37d259d
MD
2571 case LTTNG_LIST_TRACEPOINT_FIELDS:
2572 {
2573 struct lttng_event_field *fields;
2574 ssize_t nb_fields;
2575
a4b92340
DG
2576 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2577 &fields);
f37d259d 2578 if (nb_fields < 0) {
f73fabfd 2579 /* Return value is a negative lttng_error_code. */
f37d259d
MD
2580 ret = -nb_fields;
2581 goto error;
2582 }
2583
2584 /*
2585 * Setup lttng message with payload size set to the event list size in
2586 * bytes and then copy list into the llm payload.
2587 */
a4b92340
DG
2588 ret = setup_lttng_msg(cmd_ctx,
2589 sizeof(struct lttng_event_field) * nb_fields);
f37d259d
MD
2590 if (ret < 0) {
2591 free(fields);
2592 goto setup_error;
2593 }
2594
2595 /* Copy event list into message payload */
2596 memcpy(cmd_ctx->llm->payload, fields,
2597 sizeof(struct lttng_event_field) * nb_fields);
2598
2599 free(fields);
2600
f73fabfd 2601 ret = LTTNG_OK;
f37d259d
MD
2602 break;
2603 }
00e2e675
DG
2604 case LTTNG_SET_CONSUMER_URI:
2605 {
a4b92340
DG
2606 size_t nb_uri, len;
2607 struct lttng_uri *uris;
2608
2609 nb_uri = cmd_ctx->lsm->u.uri.size;
2610 len = nb_uri * sizeof(struct lttng_uri);
2611
2612 if (nb_uri == 0) {
f73fabfd 2613 ret = LTTNG_ERR_INVALID;
a4b92340
DG
2614 goto error;
2615 }
2616
2617 uris = zmalloc(len);
2618 if (uris == NULL) {
f73fabfd 2619 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2620 goto error;
2621 }
2622
2623 /* Receive variable len data */
77c7c900 2624 DBG("Receiving %zu URI(s) from client ...", nb_uri);
a4b92340
DG
2625 ret = lttcomm_recv_unix_sock(sock, uris, len);
2626 if (ret <= 0) {
2627 DBG("No URIs received from client... continuing");
2628 *sock_error = 1;
f73fabfd 2629 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2630 free(uris);
a4b92340
DG
2631 goto error;
2632 }
2633
00e2e675 2634 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
a4b92340 2635 nb_uri, uris);
f73fabfd 2636 if (ret != LTTNG_OK) {
b1d41407 2637 free(uris);
a4b92340
DG
2638 goto error;
2639 }
2640
2641 /*
2642 * XXX: 0 means that this URI should be applied on the session. Should
2643 * be a DOMAIN enuam.
2644 */
2645 if (cmd_ctx->lsm->domain.type == 0) {
2646 /* Add the URI for the UST session if a consumer is present. */
2647 if (cmd_ctx->session->ust_session &&
2648 cmd_ctx->session->ust_session->consumer) {
2649 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2650 nb_uri, uris);
2651 } else if (cmd_ctx->session->kernel_session &&
2652 cmd_ctx->session->kernel_session->consumer) {
2653 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2654 cmd_ctx->session, nb_uri, uris);
2655 }
2656 }
2657
b1d41407
DG
2658 free(uris);
2659
00e2e675
DG
2660 break;
2661 }
f3ed775e 2662 case LTTNG_START_TRACE:
8c0faa1d 2663 {
54d01ffb 2664 ret = cmd_start_trace(cmd_ctx->session);
8c0faa1d
DG
2665 break;
2666 }
f3ed775e 2667 case LTTNG_STOP_TRACE:
8c0faa1d 2668 {
54d01ffb 2669 ret = cmd_stop_trace(cmd_ctx->session);
8c0faa1d
DG
2670 break;
2671 }
5e16da05
MD
2672 case LTTNG_CREATE_SESSION:
2673 {
a4b92340
DG
2674 size_t nb_uri, len;
2675 struct lttng_uri *uris = NULL;
2676
2677 nb_uri = cmd_ctx->lsm->u.uri.size;
2678 len = nb_uri * sizeof(struct lttng_uri);
2679
2680 if (nb_uri > 0) {
2681 uris = zmalloc(len);
2682 if (uris == NULL) {
f73fabfd 2683 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2684 goto error;
2685 }
2686
2687 /* Receive variable len data */
77c7c900 2688 DBG("Waiting for %zu URIs from client ...", nb_uri);
a4b92340
DG
2689 ret = lttcomm_recv_unix_sock(sock, uris, len);
2690 if (ret <= 0) {
2691 DBG("No URIs received from client... continuing");
2692 *sock_error = 1;
f73fabfd 2693 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2694 free(uris);
a4b92340
DG
2695 goto error;
2696 }
2697
2698 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2699 DBG("Creating session with ONE network URI is a bad call");
f73fabfd 2700 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2701 free(uris);
a4b92340
DG
2702 goto error;
2703 }
2704 }
2705
2706 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2707 &cmd_ctx->creds);
2708
b1d41407
DG
2709 free(uris);
2710
00e2e675
DG
2711 break;
2712 }
5e16da05
MD
2713 case LTTNG_DESTROY_SESSION:
2714 {
2f77fc4b 2715 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
a4b92340
DG
2716
2717 /* Set session to NULL so we do not unlock it after free. */
256a5576 2718 cmd_ctx->session = NULL;
5461b305 2719 break;
5e16da05 2720 }
9f19cc17 2721 case LTTNG_LIST_DOMAINS:
5e16da05 2722 {
54d01ffb
DG
2723 ssize_t nb_dom;
2724 struct lttng_domain *domains;
5461b305 2725
54d01ffb
DG
2726 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2727 if (nb_dom < 0) {
f73fabfd 2728 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2729 ret = -nb_dom;
2730 goto error;
ce3d728c 2731 }
520ff687 2732
54d01ffb 2733 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
5461b305
DG
2734 if (ret < 0) {
2735 goto setup_error;
520ff687 2736 }
57167058 2737
54d01ffb
DG
2738 /* Copy event list into message payload */
2739 memcpy(cmd_ctx->llm->payload, domains,
2740 nb_dom * sizeof(struct lttng_domain));
2741
2742 free(domains);
5e16da05 2743
f73fabfd 2744 ret = LTTNG_OK;
5e16da05
MD
2745 break;
2746 }
9f19cc17 2747 case LTTNG_LIST_CHANNELS:
5e16da05 2748 {
6775595e 2749 int nb_chan;
54d01ffb
DG
2750 struct lttng_channel *channels;
2751
b551a063
DG
2752 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2753 cmd_ctx->session, &channels);
54d01ffb 2754 if (nb_chan < 0) {
f73fabfd 2755 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2756 ret = -nb_chan;
2757 goto error;
5461b305 2758 }
ca95a216 2759
54d01ffb 2760 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
5461b305
DG
2761 if (ret < 0) {
2762 goto setup_error;
2763 }
9f19cc17 2764
54d01ffb
DG
2765 /* Copy event list into message payload */
2766 memcpy(cmd_ctx->llm->payload, channels,
2767 nb_chan * sizeof(struct lttng_channel));
2768
2769 free(channels);
9f19cc17 2770
f73fabfd 2771 ret = LTTNG_OK;
5461b305 2772 break;
5e16da05 2773 }
9f19cc17 2774 case LTTNG_LIST_EVENTS:
5e16da05 2775 {
b551a063 2776 ssize_t nb_event;
684d34d2 2777 struct lttng_event *events = NULL;
9f19cc17 2778
b551a063 2779 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
54d01ffb
DG
2780 cmd_ctx->lsm->u.list.channel_name, &events);
2781 if (nb_event < 0) {
f73fabfd 2782 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2783 ret = -nb_event;
2784 goto error;
5461b305 2785 }
ca95a216 2786
54d01ffb 2787 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
5461b305
DG
2788 if (ret < 0) {
2789 goto setup_error;
2790 }
9f19cc17 2791
54d01ffb
DG
2792 /* Copy event list into message payload */
2793 memcpy(cmd_ctx->llm->payload, events,
2794 nb_event * sizeof(struct lttng_event));
9f19cc17 2795
54d01ffb 2796 free(events);
9f19cc17 2797
f73fabfd 2798 ret = LTTNG_OK;
5461b305 2799 break;
5e16da05
MD
2800 }
2801 case LTTNG_LIST_SESSIONS:
2802 {
8e0af1b4 2803 unsigned int nr_sessions;
5461b305 2804
8e0af1b4 2805 session_lock_list();
730389d9
DG
2806 nr_sessions = lttng_sessions_count(
2807 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2808 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
d32fb093 2809
8e0af1b4 2810 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
5461b305 2811 if (ret < 0) {
54d01ffb 2812 session_unlock_list();
5461b305 2813 goto setup_error;
e065084a 2814 }
5e16da05 2815
6c9cc2ab 2816 /* Filled the session array */
2f77fc4b 2817 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
730389d9
DG
2818 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2819 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
6c9cc2ab 2820
54d01ffb 2821 session_unlock_list();
5e16da05 2822
f73fabfd 2823 ret = LTTNG_OK;
5e16da05
MD
2824 break;
2825 }
d0254c7c
MD
2826 case LTTNG_CALIBRATE:
2827 {
54d01ffb
DG
2828 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2829 &cmd_ctx->lsm->u.calibrate);
d0254c7c
MD
2830 break;
2831 }
d9800920
DG
2832 case LTTNG_REGISTER_CONSUMER:
2833 {
2f77fc4b
DG
2834 struct consumer_data *cdata;
2835
2836 switch (cmd_ctx->lsm->domain.type) {
2837 case LTTNG_DOMAIN_KERNEL:
2838 cdata = &kconsumer_data;
2839 break;
2840 default:
f73fabfd 2841 ret = LTTNG_ERR_UND;
2f77fc4b
DG
2842 goto error;
2843 }
2844
54d01ffb 2845 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 2846 cmd_ctx->lsm->u.reg.path, cdata);
d9800920
DG
2847 break;
2848 }
53a80697
MD
2849 case LTTNG_SET_FILTER:
2850 {
2851 struct lttng_filter_bytecode *bytecode;
2852
d93c4f1f 2853 if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) {
f73fabfd 2854 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2855 goto error;
2856 }
2857 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2858 if (!bytecode) {
f73fabfd 2859 ret = LTTNG_ERR_FILTER_NOMEM;
53a80697
MD
2860 goto error;
2861 }
2862 /* Receive var. len. data */
2863 DBG("Receiving var len data from client ...");
2864 ret = lttcomm_recv_unix_sock(sock, bytecode,
2865 cmd_ctx->lsm->u.filter.bytecode_len);
2866 if (ret <= 0) {
2867 DBG("Nothing recv() from client var len data... continuing");
2868 *sock_error = 1;
f73fabfd 2869 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2870 goto error;
2871 }
2872
2873 if (bytecode->len + sizeof(*bytecode)
2874 != cmd_ctx->lsm->u.filter.bytecode_len) {
2875 free(bytecode);
f73fabfd 2876 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2877 goto error;
2878 }
2879
2880 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2881 cmd_ctx->lsm->u.filter.channel_name,
178191b3 2882 &cmd_ctx->lsm->u.filter.event,
53a80697
MD
2883 bytecode);
2884 break;
2885 }
6d805429 2886 case LTTNG_DATA_PENDING:
806e2684 2887 {
6d805429 2888 ret = cmd_data_pending(cmd_ctx->session);
806e2684
DG
2889 break;
2890 }
5e16da05 2891 default:
f73fabfd 2892 ret = LTTNG_ERR_UND;
5461b305 2893 break;
fac6795d
DG
2894 }
2895
5461b305 2896error:
5461b305
DG
2897 if (cmd_ctx->llm == NULL) {
2898 DBG("Missing llm structure. Allocating one.");
894be886 2899 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
5461b305
DG
2900 goto setup_error;
2901 }
2902 }
54d01ffb 2903 /* Set return code */
5461b305 2904 cmd_ctx->llm->ret_code = ret;
5461b305 2905setup_error:
b5541356 2906 if (cmd_ctx->session) {
54d01ffb 2907 session_unlock(cmd_ctx->session);
b5541356 2908 }
256a5576
MD
2909 if (need_tracing_session) {
2910 session_unlock_list();
2911 }
54d01ffb 2912init_setup_error:
8028d920 2913 return ret;
fac6795d
DG
2914}
2915
44a5e5eb
DG
2916/*
2917 * Thread managing health check socket.
2918 */
2919static void *thread_manage_health(void *data)
2920{
eb4a2943 2921 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
44a5e5eb
DG
2922 uint32_t revents, nb_fd;
2923 struct lttng_poll_event events;
2924 struct lttcomm_health_msg msg;
2925 struct lttcomm_health_data reply;
2926
2927 DBG("[thread] Manage health check started");
2928
2929 rcu_register_thread();
2930
2931 /* Create unix socket */
2932 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2933 if (sock < 0) {
2934 ERR("Unable to create health check Unix socket");
2935 ret = -1;
2936 goto error;
2937 }
2938
b662582b
DG
2939 /*
2940 * Set the CLOEXEC flag. Return code is useless because either way, the
2941 * show must go on.
2942 */
2943 (void) utils_set_fd_cloexec(sock);
2944
44a5e5eb
DG
2945 ret = lttcomm_listen_unix_sock(sock);
2946 if (ret < 0) {
2947 goto error;
2948 }
2949
2950 /*
2951 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2952 * more will be added to this poll set.
2953 */
2954 ret = create_thread_poll_set(&events, 2);
2955 if (ret < 0) {
2956 goto error;
2957 }
2958
2959 /* Add the application registration socket */
2960 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2961 if (ret < 0) {
2962 goto error;
2963 }
2964
2965 while (1) {
2966 DBG("Health check ready");
2967
2968 nb_fd = LTTNG_POLL_GETNB(&events);
2969
2970 /* Inifinite blocking call, waiting for transmission */
2971restart:
2972 ret = lttng_poll_wait(&events, -1);
2973 if (ret < 0) {
2974 /*
2975 * Restart interrupted system call.
2976 */
2977 if (errno == EINTR) {
2978 goto restart;
2979 }
2980 goto error;
2981 }
2982
2983 for (i = 0; i < nb_fd; i++) {
2984 /* Fetch once the poll data */
2985 revents = LTTNG_POLL_GETEV(&events, i);
2986 pollfd = LTTNG_POLL_GETFD(&events, i);
2987
2988 /* Thread quit pipe has been closed. Killing thread. */
2989 ret = check_thread_quit_pipe(pollfd, revents);
2990 if (ret) {
139ac872
MD
2991 err = 0;
2992 goto exit;
44a5e5eb
DG
2993 }
2994
2995 /* Event on the registration socket */
2996 if (pollfd == sock) {
2997 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2998 ERR("Health socket poll error");
2999 goto error;
3000 }
3001 }
3002 }
3003
3004 new_sock = lttcomm_accept_unix_sock(sock);
3005 if (new_sock < 0) {
3006 goto error;
3007 }
3008
b662582b
DG
3009 /*
3010 * Set the CLOEXEC flag. Return code is useless because either way, the
3011 * show must go on.
3012 */
3013 (void) utils_set_fd_cloexec(new_sock);
3014
44a5e5eb
DG
3015 DBG("Receiving data from client for health...");
3016 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3017 if (ret <= 0) {
3018 DBG("Nothing recv() from client... continuing");
3019 ret = close(new_sock);
3020 if (ret) {
3021 PERROR("close");
3022 }
3023 new_sock = -1;
3024 continue;
3025 }
3026
3027 rcu_thread_online();
3028
3029 switch (msg.component) {
3030 case LTTNG_HEALTH_CMD:
3031 reply.ret_code = health_check_state(&health_thread_cmd);
3032 break;
139ac872
MD
3033 case LTTNG_HEALTH_APP_MANAGE:
3034 reply.ret_code = health_check_state(&health_thread_app_manage);
3035 break;
44a5e5eb
DG
3036 case LTTNG_HEALTH_APP_REG:
3037 reply.ret_code = health_check_state(&health_thread_app_reg);
3038 break;
3039 case LTTNG_HEALTH_KERNEL:
3040 reply.ret_code = health_check_state(&health_thread_kernel);
3041 break;
3042 case LTTNG_HEALTH_CONSUMER:
3043 reply.ret_code = check_consumer_health();
3044 break;
3045 case LTTNG_HEALTH_ALL:
44a5e5eb 3046 reply.ret_code =
139ac872
MD
3047 health_check_state(&health_thread_app_manage) &&
3048 health_check_state(&health_thread_app_reg) &&
3049 health_check_state(&health_thread_cmd) &&
3050 health_check_state(&health_thread_kernel) &&
3051 check_consumer_health();
44a5e5eb
DG
3052 break;
3053 default:
f73fabfd 3054 reply.ret_code = LTTNG_ERR_UND;
44a5e5eb
DG
3055 break;
3056 }
3057
3058 /*
3059 * Flip ret value since 0 is a success and 1 indicates a bad health for
3060 * the client where in the sessiond it is the opposite. Again, this is
3061 * just to make things easier for us poor developer which enjoy a lot
3062 * lazyness.
3063 */
3064 if (reply.ret_code == 0 || reply.ret_code == 1) {
3065 reply.ret_code = !reply.ret_code;
3066 }
3067
3068 DBG2("Health check return value %d", reply.ret_code);
3069
3070 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3071 if (ret < 0) {
3072 ERR("Failed to send health data back to client");
3073 }
3074
3075 /* End of transmission */
3076 ret = close(new_sock);
3077 if (ret) {
3078 PERROR("close");
3079 }
3080 new_sock = -1;
3081 }
3082
139ac872 3083exit:
44a5e5eb 3084error:
139ac872
MD
3085 if (err) {
3086 ERR("Health error occurred in %s", __func__);
3087 }
44a5e5eb
DG
3088 DBG("Health check thread dying");
3089 unlink(health_unix_sock_path);
3090 if (sock >= 0) {
3091 ret = close(sock);
3092 if (ret) {
3093 PERROR("close");
3094 }
3095 }
3096 if (new_sock >= 0) {
3097 ret = close(new_sock);
3098 if (ret) {
3099 PERROR("close");
3100 }
3101 }
3102
3103 lttng_poll_clean(&events);
3104
3105 rcu_unregister_thread();
3106 return NULL;
3107}
3108
1d4b027a 3109/*
d063d709
DG
3110 * This thread manage all clients request using the unix client socket for
3111 * communication.
1d4b027a
DG
3112 */
3113static void *thread_manage_clients(void *data)
3114{
139ac872 3115 int sock = -1, ret, i, pollfd, err = -1;
53a80697 3116 int sock_error;
5eb91c98 3117 uint32_t revents, nb_fd;
273ea72c 3118 struct command_ctx *cmd_ctx = NULL;
5eb91c98 3119 struct lttng_poll_event events;
1d4b027a
DG
3120
3121 DBG("[thread] Manage client started");
3122
8ac94142
CB
3123 testpoint(thread_manage_clients);
3124
f6a9efaa
DG
3125 rcu_register_thread();
3126
44a5e5eb
DG
3127 health_code_update(&health_thread_cmd);
3128
1d4b027a
DG
3129 ret = lttcomm_listen_unix_sock(client_sock);
3130 if (ret < 0) {
35df2755 3131 goto error_listen;
1d4b027a
DG
3132 }
3133
5eb91c98
DG
3134 /*
3135 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3136 * more will be added to this poll set.
3137 */
3138 ret = create_thread_poll_set(&events, 2);
3139 if (ret < 0) {
35df2755 3140 goto error_create_poll;