Fix: Wrong fd used by kernel_wait_quiescent
[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;
1d4b027a 400 char *cmd;
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
ef599319 450 utils_close_pipe(kernel_poll_pipe);
ef599319 451 utils_close_pipe(apps_cmd_pipe);
421cb601
DG
452
453 /* <fun> */
f56a39af 454 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
455 "Matthew, BEET driven development works!%c[%dm",
456 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
457 /* </fun> */
fac6795d
DG
458}
459
e065084a 460/*
d063d709 461 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 462 *
d063d709 463 * Return lttcomm error code.
e065084a
DG
464 */
465static int send_unix_sock(int sock, void *buf, size_t len)
466{
467 /* Check valid length */
468 if (len <= 0) {
469 return -1;
470 }
471
472 return lttcomm_send_unix_sock(sock, buf, len);
473}
474
5461b305 475/*
d063d709 476 * Free memory of a command context structure.
5461b305 477 */
a2fb29a5 478static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 479{
a2fb29a5
DG
480 DBG("Clean command context structure");
481 if (*cmd_ctx) {
482 if ((*cmd_ctx)->llm) {
483 free((*cmd_ctx)->llm);
5461b305 484 }
a2fb29a5
DG
485 if ((*cmd_ctx)->lsm) {
486 free((*cmd_ctx)->lsm);
5461b305 487 }
a2fb29a5
DG
488 free(*cmd_ctx);
489 *cmd_ctx = NULL;
5461b305
DG
490 }
491}
492
fac6795d 493/*
0fdd1e2c 494 * Notify UST applications using the shm mmap futex.
fac6795d 495 */
0fdd1e2c 496static int notify_ust_apps(int active)
fac6795d 497{
0fdd1e2c 498 char *wait_shm_mmap;
fac6795d 499
0fdd1e2c 500 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 501
0fdd1e2c
DG
502 /* See shm.c for this call implying mmap, shm and futex calls */
503 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
504 if (wait_shm_mmap == NULL) {
fac6795d
DG
505 goto error;
506 }
507
0fdd1e2c
DG
508 /* Wake waiting process */
509 futex_wait_update((int32_t *) wait_shm_mmap, active);
510
511 /* Apps notified successfully */
512 return 0;
fac6795d
DG
513
514error:
0fdd1e2c 515 return -1;
fac6795d
DG
516}
517
e065084a 518/*
d063d709
DG
519 * Setup the outgoing data buffer for the response (llm) by allocating the
520 * right amount of memory and copying the original information from the lsm
521 * structure.
ca95a216 522 *
d063d709 523 * Return total size of the buffer pointed by buf.
ca95a216 524 */
5461b305 525static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 526{
f3ed775e 527 int ret, buf_size;
ca95a216 528
f3ed775e 529 buf_size = size;
5461b305 530
ba7f0ae5 531 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 532 if (cmd_ctx->llm == NULL) {
76d7553f 533 PERROR("zmalloc");
5461b305 534 ret = -ENOMEM;
ca95a216
DG
535 goto error;
536 }
537
5461b305
DG
538 /* Copy common data */
539 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 540 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 541
5461b305
DG
542 cmd_ctx->llm->data_size = size;
543 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
544
ca95a216
DG
545 return buf_size;
546
547error:
548 return ret;
549}
550
7a485870 551/*
5eb91c98 552 * Update the kernel poll set of all channel fd available over all tracing
d063d709 553 * session. Add the wakeup pipe at the end of the set.
7a485870 554 */
5eb91c98 555static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 556{
5eb91c98 557 int ret;
7a485870
DG
558 struct ltt_session *session;
559 struct ltt_kernel_channel *channel;
560
5eb91c98 561 DBG("Updating kernel poll set");
7a485870 562
54d01ffb 563 session_lock_list();
b5541356 564 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 565 session_lock(session);
7a485870 566 if (session->kernel_session == NULL) {
54d01ffb 567 session_unlock(session);
7a485870
DG
568 continue;
569 }
7a485870 570
54d01ffb
DG
571 cds_list_for_each_entry(channel,
572 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
573 /* Add channel fd to the kernel poll set */
574 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
575 if (ret < 0) {
54d01ffb 576 session_unlock(session);
5eb91c98
DG
577 goto error;
578 }
579 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 580 }
54d01ffb 581 session_unlock(session);
7a485870 582 }
54d01ffb 583 session_unlock_list();
7a485870 584
5eb91c98 585 return 0;
7a485870
DG
586
587error:
54d01ffb 588 session_unlock_list();
7a485870
DG
589 return -1;
590}
591
592/*
54d01ffb 593 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 594 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 595 *
d063d709 596 * Useful for CPU hotplug feature.
7a485870 597 */
2bdd86d4 598static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
599{
600 int ret = 0;
601 struct ltt_session *session;
173af62f 602 struct ltt_kernel_session *ksess;
7a485870
DG
603 struct ltt_kernel_channel *channel;
604
605 DBG("Updating kernel streams for channel fd %d", fd);
606
54d01ffb 607 session_lock_list();
b5541356 608 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 609 session_lock(session);
7a485870 610 if (session->kernel_session == NULL) {
54d01ffb 611 session_unlock(session);
7a485870
DG
612 continue;
613 }
173af62f 614 ksess = session->kernel_session;
d9800920 615
173af62f 616 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
7a485870
DG
617 if (channel->fd == fd) {
618 DBG("Channel found, updating kernel streams");
619 ret = kernel_open_channel_stream(channel);
620 if (ret < 0) {
b3c750d2 621 goto error;
7a485870 622 }
d9800920 623
7a485870 624 /*
5eb91c98
DG
625 * Have we already sent fds to the consumer? If yes, it means
626 * that tracing is started so it is safe to send our updated
627 * stream fds.
7a485870 628 */
173af62f
DG
629 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
630 struct lttng_ht_iter iter;
631 struct consumer_socket *socket;
632
633
634 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
635 &iter.iter, socket, node.node) {
636 /* Code flow error */
637 assert(socket->fd >= 0);
638
639 pthread_mutex_lock(socket->lock);
640 ret = kernel_consumer_send_channel_stream(socket->fd,
641 channel, ksess);
642 pthread_mutex_unlock(socket->lock);
643 if (ret < 0) {
644 goto error;
645 }
7a485870
DG
646 }
647 }
b3c750d2 648 goto error;
7a485870
DG
649 }
650 }
54d01ffb 651 session_unlock(session);
7a485870 652 }
54d01ffb 653 session_unlock_list();
b3c750d2 654 return ret;
7a485870 655
b3c750d2 656error:
54d01ffb
DG
657 session_unlock(session);
658 session_unlock_list();
7a485870
DG
659 return ret;
660}
661
487cf67c
DG
662/*
663 * For each tracing session, update newly registered apps.
664 */
665static void update_ust_app(int app_sock)
666{
667 struct ltt_session *sess, *stmp;
668
4ee14516
DG
669 session_lock_list();
670
487cf67c
DG
671 /* For all tracing session(s) */
672 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
4ee14516 673 session_lock(sess);
421cb601
DG
674 if (sess->ust_session) {
675 ust_app_global_update(sess->ust_session, app_sock);
676 }
4ee14516 677 session_unlock(sess);
487cf67c 678 }
4ee14516
DG
679
680 session_unlock_list();
487cf67c
DG
681}
682
7a485870 683/*
d063d709 684 * This thread manage event coming from the kernel.
7a485870 685 *
d063d709
DG
686 * Features supported in this thread:
687 * -) CPU Hotplug
7a485870
DG
688 */
689static void *thread_manage_kernel(void *data)
690{
139ac872 691 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 692 uint32_t revents, nb_fd;
7a485870 693 char tmp;
5eb91c98 694 struct lttng_poll_event events;
7a485870
DG
695
696 DBG("Thread manage kernel started");
697
8ac94142
CB
698 testpoint(thread_manage_kernel);
699
44a5e5eb
DG
700 health_code_update(&health_thread_kernel);
701
8ac94142
CB
702 testpoint(thread_manage_kernel_before_loop);
703
5eb91c98
DG
704 ret = create_thread_poll_set(&events, 2);
705 if (ret < 0) {
76d7553f 706 goto error_poll_create;
5eb91c98
DG
707 }
708
709 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
710 if (ret < 0) {
711 goto error;
712 }
713
7a485870 714 while (1) {
44a5e5eb
DG
715 health_code_update(&health_thread_kernel);
716
7a485870 717 if (update_poll_flag == 1) {
5f822d0a
DG
718 /*
719 * Reset number of fd in the poll set. Always 2 since there is the thread
720 * quit pipe and the kernel pipe.
721 */
722 events.nb_fd = 2;
723
5eb91c98
DG
724 ret = update_kernel_poll(&events);
725 if (ret < 0) {
7a485870
DG
726 goto error;
727 }
728 update_poll_flag = 0;
729 }
730
5eb91c98
DG
731 nb_fd = LTTNG_POLL_GETNB(&events);
732
733 DBG("Thread kernel polling on %d fds", nb_fd);
734
735 /* Zeroed the poll events */
736 lttng_poll_reset(&events);
7a485870
DG
737
738 /* Poll infinite value of time */
88f2b785 739 restart:
44a5e5eb 740 health_poll_update(&health_thread_kernel);
5eb91c98 741 ret = lttng_poll_wait(&events, -1);
44a5e5eb 742 health_poll_update(&health_thread_kernel);
7a485870 743 if (ret < 0) {
88f2b785
MD
744 /*
745 * Restart interrupted system call.
746 */
747 if (errno == EINTR) {
748 goto restart;
749 }
7a485870
DG
750 goto error;
751 } else if (ret == 0) {
752 /* Should not happen since timeout is infinite */
85611738
DG
753 ERR("Return value of poll is 0 with an infinite timeout.\n"
754 "This should not have happened! Continuing...");
7a485870
DG
755 continue;
756 }
757
5eb91c98
DG
758 for (i = 0; i < nb_fd; i++) {
759 /* Fetch once the poll data */
760 revents = LTTNG_POLL_GETEV(&events, i);
761 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 762
44a5e5eb
DG
763 health_code_update(&health_thread_kernel);
764
5eb91c98
DG
765 /* Thread quit pipe has been closed. Killing thread. */
766 ret = check_thread_quit_pipe(pollfd, revents);
767 if (ret) {
139ac872
MD
768 err = 0;
769 goto exit;
5eb91c98 770 }
7a485870 771
5eb91c98
DG
772 /* Check for data on kernel pipe */
773 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
774 ret = read(kernel_poll_pipe[0], &tmp, 1);
775 update_poll_flag = 1;
776 continue;
777 } else {
778 /*
779 * New CPU detected by the kernel. Adding kernel stream to
780 * kernel session and updating the kernel consumer
781 */
782 if (revents & LPOLLIN) {
2bdd86d4 783 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
784 if (ret < 0) {
785 continue;
786 }
787 break;
788 /*
789 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
790 * and unregister kernel stream at this point.
791 */
7a485870 792 }
7a485870
DG
793 }
794 }
795 }
796
139ac872 797exit:
7a485870 798error:
5eb91c98 799 lttng_poll_clean(&events);
76d7553f 800error_poll_create:
139ac872
MD
801 if (err) {
802 health_error(&health_thread_kernel);
803 ERR("Health error occurred in %s", __func__);
804 }
805 health_exit(&health_thread_kernel);
76d7553f 806 DBG("Kernel thread dying");
7a485870
DG
807 return NULL;
808}
809
a23ec3a7
DG
810/*
811 * Signal pthread condition of the consumer data that the thread.
812 */
813static void signal_consumer_condition(struct consumer_data *data, int state)
814{
815 pthread_mutex_lock(&data->cond_mutex);
816
817 /*
818 * The state is set before signaling. It can be any value, it's the waiter
819 * job to correctly interpret this condition variable associated to the
820 * consumer pthread_cond.
821 *
822 * A value of 0 means that the corresponding thread of the consumer data
823 * was not started. 1 indicates that the thread has started and is ready
824 * for action. A negative value means that there was an error during the
825 * thread bootstrap.
826 */
827 data->consumer_thread_is_ready = state;
828 (void) pthread_cond_signal(&data->cond);
829
830 pthread_mutex_unlock(&data->cond_mutex);
831}
832
1d4b027a 833/*
3bd1e081 834 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 835 */
3bd1e081 836static void *thread_manage_consumer(void *data)
1d4b027a 837{
139ac872 838 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98 839 uint32_t revents, nb_fd;
1d4b027a 840 enum lttcomm_return_code code;
5eb91c98 841 struct lttng_poll_event events;
3bd1e081 842 struct consumer_data *consumer_data = data;
1d4b027a 843
3bd1e081 844 DBG("[thread] Manage consumer started");
1d4b027a 845
9449cc75
DG
846 /*
847 * Since the consumer thread can be spawned at any moment in time, we init
848 * the health to a poll status (1, which is a valid health over time).
849 * When the thread starts, we update here the health to a "code" path being
850 * an even value so this thread, when reaching a poll wait, does not
851 * trigger an error with an even value.
852 *
853 * Here is the use case we avoid.
854 *
855 * +1: the first poll update during initialization (main())
856 * +2 * x: multiple code update once in this thread.
857 * +1: poll wait in this thread (being a good health state).
858 * == even number which after the wait period shows as a bad health.
859 *
860 * In a nutshell, the following poll update to the health state brings back
861 * the state to an even value meaning a code path.
862 */
863 health_poll_update(&consumer_data->health);
864
5eb91c98
DG
865 /*
866 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
867 * Nothing more will be added to this poll set.
868 */
869 ret = create_thread_poll_set(&events, 2);
870 if (ret < 0) {
76d7553f 871 goto error_poll;
5eb91c98 872 }
273ea72c 873
edb8b045
DG
874 /*
875 * The error socket here is already in a listening state which was done
876 * just before spawning this thread to avoid a race between the consumer
877 * daemon exec trying to connect and the listen() call.
878 */
3bd1e081 879 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
880 if (ret < 0) {
881 goto error;
882 }
883
884 nb_fd = LTTNG_POLL_GETNB(&events);
273ea72c 885
44a5e5eb
DG
886 health_code_update(&consumer_data->health);
887
273ea72c 888 /* Inifinite blocking call, waiting for transmission */
88f2b785 889restart:
44a5e5eb 890 health_poll_update(&consumer_data->health);
8ac94142
CB
891
892 testpoint(thread_manage_consumer);
893
5eb91c98 894 ret = lttng_poll_wait(&events, -1);
44a5e5eb 895 health_poll_update(&consumer_data->health);
273ea72c 896 if (ret < 0) {
88f2b785
MD
897 /*
898 * Restart interrupted system call.
899 */
900 if (errno == EINTR) {
901 goto restart;
902 }
273ea72c
DG
903 goto error;
904 }
905
5eb91c98
DG
906 for (i = 0; i < nb_fd; i++) {
907 /* Fetch once the poll data */
908 revents = LTTNG_POLL_GETEV(&events, i);
909 pollfd = LTTNG_POLL_GETFD(&events, i);
910
44a5e5eb
DG
911 health_code_update(&consumer_data->health);
912
5eb91c98
DG
913 /* Thread quit pipe has been closed. Killing thread. */
914 ret = check_thread_quit_pipe(pollfd, revents);
915 if (ret) {
139ac872
MD
916 err = 0;
917 goto exit;
5eb91c98
DG
918 }
919
920 /* Event on the registration socket */
3bd1e081 921 if (pollfd == consumer_data->err_sock) {
5eb91c98 922 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 923 ERR("consumer err socket poll error");
5eb91c98
DG
924 goto error;
925 }
926 }
273ea72c
DG
927 }
928
3bd1e081 929 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
930 if (sock < 0) {
931 goto error;
932 }
933
b662582b
DG
934 /*
935 * Set the CLOEXEC flag. Return code is useless because either way, the
936 * show must go on.
937 */
938 (void) utils_set_fd_cloexec(sock);
939
44a5e5eb
DG
940 health_code_update(&consumer_data->health);
941
3bd1e081 942 DBG2("Receiving code from consumer err_sock");
ee0b0061 943
712ea556 944 /* Getting status code from kconsumerd */
54d01ffb
DG
945 ret = lttcomm_recv_unix_sock(sock, &code,
946 sizeof(enum lttcomm_return_code));
1d4b027a
DG
947 if (ret <= 0) {
948 goto error;
949 }
950
44a5e5eb
DG
951 health_code_update(&consumer_data->health);
952
f73fabfd 953 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
3bd1e081
MD
954 consumer_data->cmd_sock =
955 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
956 if (consumer_data->cmd_sock < 0) {
a23ec3a7
DG
957 /* On error, signal condition and quit. */
958 signal_consumer_condition(consumer_data, -1);
48842b30 959 PERROR("consumer connect");
1d4b027a
DG
960 goto error;
961 }
a23ec3a7
DG
962 signal_consumer_condition(consumer_data, 1);
963 DBG("Consumer command socket ready");
1d4b027a 964 } else {
3bd1e081 965 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
966 lttcomm_get_readable_code(-code));
967 goto error;
968 }
969
54d01ffb 970 /* Remove the kconsumerd error sock since we've established a connexion */
3bd1e081 971 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 972 if (ret < 0) {
72079cae
DG
973 goto error;
974 }
975
5eb91c98
DG
976 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
977 if (ret < 0) {
72079cae 978 goto error;
5eb91c98
DG
979 }
980
44a5e5eb
DG
981 health_code_update(&consumer_data->health);
982
5eb91c98
DG
983 /* Update number of fd */
984 nb_fd = LTTNG_POLL_GETNB(&events);
985
986 /* Inifinite blocking call, waiting for transmission */
88f2b785 987restart_poll:
44a5e5eb 988 health_poll_update(&consumer_data->health);
5eb91c98 989 ret = lttng_poll_wait(&events, -1);
44a5e5eb 990 health_poll_update(&consumer_data->health);
5eb91c98 991 if (ret < 0) {
88f2b785
MD
992 /*
993 * Restart interrupted system call.
994 */
995 if (errno == EINTR) {
996 goto restart_poll;
997 }
72079cae
DG
998 goto error;
999 }
1000
5eb91c98
DG
1001 for (i = 0; i < nb_fd; i++) {
1002 /* Fetch once the poll data */
1003 revents = LTTNG_POLL_GETEV(&events, i);
1004 pollfd = LTTNG_POLL_GETFD(&events, i);
1005
44a5e5eb
DG
1006 health_code_update(&consumer_data->health);
1007
5eb91c98
DG
1008 /* Thread quit pipe has been closed. Killing thread. */
1009 ret = check_thread_quit_pipe(pollfd, revents);
1010 if (ret) {
139ac872
MD
1011 err = 0;
1012 goto exit;
5eb91c98
DG
1013 }
1014
1015 /* Event on the kconsumerd socket */
1016 if (pollfd == sock) {
1017 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1018 ERR("consumer err socket second poll error");
5eb91c98
DG
1019 goto error;
1020 }
1021 }
1022 }
1023
44a5e5eb
DG
1024 health_code_update(&consumer_data->health);
1025
712ea556 1026 /* Wait for any kconsumerd error */
54d01ffb
DG
1027 ret = lttcomm_recv_unix_sock(sock, &code,
1028 sizeof(enum lttcomm_return_code));
712ea556 1029 if (ret <= 0) {
3bd1e081 1030 ERR("consumer closed the command socket");
712ea556 1031 goto error;
6f61d021 1032 }
1d4b027a 1033
3bd1e081 1034 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
712ea556 1035
139ac872 1036exit:
1d4b027a 1037error:
5c827ce0
DG
1038 /* Immediately set the consumerd state to stopped */
1039 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1040 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1041 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1042 consumer_data->type == LTTNG_CONSUMER32_UST) {
1043 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1044 } else {
1045 /* Code flow error... */
1046 assert(0);
1047 }
1048
76d7553f
MD
1049 if (consumer_data->err_sock >= 0) {
1050 ret = close(consumer_data->err_sock);
1051 if (ret) {
1052 PERROR("close");
1053 }
1054 }
1055 if (consumer_data->cmd_sock >= 0) {
1056 ret = close(consumer_data->cmd_sock);
1057 if (ret) {
1058 PERROR("close");
1059 }
1060 }
1061 if (sock >= 0) {
1062 ret = close(sock);
1063 if (ret) {
1064 PERROR("close");
1065 }
1066 }
273ea72c 1067
3bd1e081
MD
1068 unlink(consumer_data->err_unix_sock_path);
1069 unlink(consumer_data->cmd_unix_sock_path);
1070 consumer_data->pid = 0;
1d4b027a 1071
5eb91c98 1072 lttng_poll_clean(&events);
76d7553f 1073error_poll:
139ac872
MD
1074 if (err) {
1075 health_error(&consumer_data->health);
1076 ERR("Health error occurred in %s", __func__);
1077 }
1078 health_exit(&consumer_data->health);
76d7553f 1079 DBG("consumer thread cleanup completed");
0177d773 1080
5eb91c98 1081 return NULL;
099e26bd
DG
1082}
1083
099e26bd
DG
1084/*
1085 * This thread manage application communication.
1d4b027a
DG
1086 */
1087static void *thread_manage_apps(void *data)
099e26bd 1088{
139ac872 1089 int i, ret, pollfd, err = -1;
5eb91c98 1090 uint32_t revents, nb_fd;
099e26bd 1091 struct ust_command ust_cmd;
5eb91c98 1092 struct lttng_poll_event events;
099e26bd
DG
1093
1094 DBG("[thread] Manage application started");
1095
8ac94142
CB
1096 testpoint(thread_manage_apps);
1097
f6a9efaa
DG
1098 rcu_register_thread();
1099 rcu_thread_online();
1100
139ac872 1101 health_code_update(&health_thread_app_manage);
44a5e5eb 1102
5eb91c98
DG
1103 ret = create_thread_poll_set(&events, 2);
1104 if (ret < 0) {
76d7553f 1105 goto error_poll_create;
5eb91c98 1106 }
099e26bd 1107
5eb91c98
DG
1108 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1109 if (ret < 0) {
1110 goto error;
1111 }
099e26bd 1112
8ac94142
CB
1113 testpoint(thread_manage_apps_before_loop);
1114
139ac872 1115 health_code_update(&health_thread_app_manage);
44a5e5eb 1116
5eb91c98
DG
1117 while (1) {
1118 /* Zeroed the events structure */
1119 lttng_poll_reset(&events);
0177d773 1120
5eb91c98 1121 nb_fd = LTTNG_POLL_GETNB(&events);
099e26bd
DG
1122
1123 DBG("Apps thread polling on %d fds", nb_fd);
1124
1125 /* Inifinite blocking call, waiting for transmission */
88f2b785 1126 restart:
139ac872 1127 health_poll_update(&health_thread_app_manage);
5eb91c98 1128 ret = lttng_poll_wait(&events, -1);
139ac872 1129 health_poll_update(&health_thread_app_manage);
099e26bd 1130 if (ret < 0) {
88f2b785
MD
1131 /*
1132 * Restart interrupted system call.
1133 */
1134 if (errno == EINTR) {
1135 goto restart;
1136 }
099e26bd
DG
1137 goto error;
1138 }
1139
5eb91c98
DG
1140 for (i = 0; i < nb_fd; i++) {
1141 /* Fetch once the poll data */
1142 revents = LTTNG_POLL_GETEV(&events, i);
1143 pollfd = LTTNG_POLL_GETFD(&events, i);
1144
139ac872 1145 health_code_update(&health_thread_app_manage);
44a5e5eb 1146
5eb91c98
DG
1147 /* Thread quit pipe has been closed. Killing thread. */
1148 ret = check_thread_quit_pipe(pollfd, revents);
1149 if (ret) {
139ac872
MD
1150 err = 0;
1151 goto exit;
5eb91c98 1152 }
099e26bd 1153
5eb91c98
DG
1154 /* Inspect the apps cmd pipe */
1155 if (pollfd == apps_cmd_pipe[0]) {
1156 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1157 ERR("Apps command pipe error");
0177d773 1158 goto error;
5eb91c98
DG
1159 } else if (revents & LPOLLIN) {
1160 /* Empty pipe */
1161 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1162 if (ret < 0 || ret < sizeof(ust_cmd)) {
76d7553f 1163 PERROR("read apps cmd pipe");
5eb91c98
DG
1164 goto error;
1165 }
099e26bd 1166
139ac872 1167 health_code_update(&health_thread_app_manage);
44a5e5eb 1168
5eb91c98 1169 /* Register applicaton to the session daemon */
56fff090 1170 ret = ust_app_register(&ust_cmd.reg_msg,
54d01ffb 1171 ust_cmd.sock);
88ff5b7f 1172 if (ret == -ENOMEM) {
5eb91c98 1173 goto error;
88ff5b7f
MD
1174 } else if (ret < 0) {
1175 break;
5eb91c98
DG
1176 }
1177
139ac872 1178 health_code_update(&health_thread_app_manage);
44a5e5eb 1179
acc7b41b 1180 /*
e0c7ec2b 1181 * Validate UST version compatibility.
acc7b41b 1182 */
e0c7ec2b
DG
1183 ret = ust_app_validate_version(ust_cmd.sock);
1184 if (ret >= 0) {
1185 /*
1186 * Add channel(s) and event(s) to newly registered apps
1187 * from lttng global UST domain.
1188 */
1189 update_ust_app(ust_cmd.sock);
1190 }
acc7b41b 1191
139ac872 1192 health_code_update(&health_thread_app_manage);
44a5e5eb 1193
f2ca2e25 1194 ret = ust_app_register_done(ust_cmd.sock);
5eb91c98
DG
1195 if (ret < 0) {
1196 /*
1197 * If the registration is not possible, we simply
1198 * unregister the apps and continue
1199 */
56fff090 1200 ust_app_unregister(ust_cmd.sock);
5eb91c98
DG
1201 } else {
1202 /*
622c9ecf
DG
1203 * We only monitor the error events of the socket. This
1204 * thread does not handle any incoming data from UST
1205 * (POLLIN).
5eb91c98 1206 */
622c9ecf
DG
1207 ret = lttng_poll_add(&events, ust_cmd.sock,
1208 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
5eb91c98
DG
1209 if (ret < 0) {
1210 goto error;
1211 }
1212
ae9e45b3
DG
1213 /* Set socket timeout for both receiving and ending */
1214 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1215 app_socket_timeout);
1216 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1217 app_socket_timeout);
1218
54d01ffb
DG
1219 DBG("Apps with sock %d added to poll set",
1220 ust_cmd.sock);
5eb91c98 1221 }
487cf67c 1222
139ac872 1223 health_code_update(&health_thread_app_manage);
44a5e5eb 1224
5eb91c98 1225 break;
0177d773 1226 }
5eb91c98
DG
1227 } else {
1228 /*
54d01ffb
DG
1229 * At this point, we know that a registered application made
1230 * the event at poll_wait.
5eb91c98
DG
1231 */
1232 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1233 /* Removing from the poll set */
1234 ret = lttng_poll_del(&events, pollfd);
1235 if (ret < 0) {
1236 goto error;
1237 }
099e26bd 1238
b9d9b220 1239 /* Socket closed on remote end. */
56fff090 1240 ust_app_unregister(pollfd);
5eb91c98
DG
1241 break;
1242 }
099e26bd 1243 }
44a5e5eb 1244
139ac872 1245 health_code_update(&health_thread_app_manage);
099e26bd 1246 }
099e26bd
DG
1247 }
1248
139ac872 1249exit:
099e26bd 1250error:
5eb91c98 1251 lttng_poll_clean(&events);
76d7553f 1252error_poll_create:
139ac872
MD
1253 if (err) {
1254 health_error(&health_thread_app_manage);
1255 ERR("Health error occurred in %s", __func__);
1256 }
1257 health_exit(&health_thread_app_manage);
76d7553f 1258 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1259 rcu_thread_offline();
1260 rcu_unregister_thread();
099e26bd
DG
1261 return NULL;
1262}
1263
1264/*
1265 * Dispatch request from the registration threads to the application
1266 * communication thread.
1267 */
1268static void *thread_dispatch_ust_registration(void *data)
1269{
1270 int ret;
1271 struct cds_wfq_node *node;
1272 struct ust_command *ust_cmd = NULL;
1273
1274 DBG("[thread] Dispatch UST command started");
1275
26c9d55e 1276 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
099e26bd
DG
1277 /* Atomically prepare the queue futex */
1278 futex_nto1_prepare(&ust_cmd_queue.futex);
1279
1280 do {
1281 /* Dequeue command for registration */
1282 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1283 if (node == NULL) {
00a17c97 1284 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1285 /* Continue thread execution */
1286 break;
1287 }
1288
1289 ust_cmd = caa_container_of(node, struct ust_command, node);
1290
2f50c8a3
DG
1291 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1292 " gid:%d sock:%d name:%s (version %d.%d)",
1293 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1294 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1295 ust_cmd->sock, ust_cmd->reg_msg.name,
1296 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
099e26bd
DG
1297 /*
1298 * Inform apps thread of the new application registration. This
1299 * call is blocking so we can be assured that the data will be read
1300 * at some point in time or wait to the end of the world :)
1301 */
1302 ret = write(apps_cmd_pipe[1], ust_cmd,
1303 sizeof(struct ust_command));
1304 if (ret < 0) {
76d7553f 1305 PERROR("write apps cmd pipe");
099e26bd
DG
1306 if (errno == EBADF) {
1307 /*
1308 * We can't inform the application thread to process
1309 * registration. We will exit or else application
1310 * registration will not occur and tracing will never
1311 * start.
1312 */
1313 goto error;
1314 }
1315 }
1316 free(ust_cmd);
1317 } while (node != NULL);
1318
1319 /* Futex wait on queue. Blocking call on futex() */
1320 futex_nto1_wait(&ust_cmd_queue.futex);
1321 }
1322
1323error:
1324 DBG("Dispatch thread dying");
1325 return NULL;
1326}
1327
1328/*
1329 * This thread manage application registration.
1330 */
1331static void *thread_registration_apps(void *data)
1d4b027a 1332{
139ac872 1333 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
1334 uint32_t revents, nb_fd;
1335 struct lttng_poll_event events;
099e26bd
DG
1336 /*
1337 * Get allocated in this thread, enqueued to a global queue, dequeued and
1338 * freed in the manage apps thread.
1339 */
1340 struct ust_command *ust_cmd = NULL;
1d4b027a 1341
099e26bd 1342 DBG("[thread] Manage application registration started");
1d4b027a 1343
8ac94142
CB
1344 testpoint(thread_registration_apps);
1345
1d4b027a
DG
1346 ret = lttcomm_listen_unix_sock(apps_sock);
1347 if (ret < 0) {
76d7553f 1348 goto error_listen;
1d4b027a
DG
1349 }
1350
5eb91c98
DG
1351 /*
1352 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1353 * more will be added to this poll set.
1354 */
1355 ret = create_thread_poll_set(&events, 2);
1356 if (ret < 0) {
76d7553f 1357 goto error_create_poll;
5eb91c98 1358 }
273ea72c 1359
5eb91c98
DG
1360 /* Add the application registration socket */
1361 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1362 if (ret < 0) {
76d7553f 1363 goto error_poll_add;
5eb91c98 1364 }
273ea72c 1365
1d4b027a 1366 /* Notify all applications to register */
0fdd1e2c
DG
1367 ret = notify_ust_apps(1);
1368 if (ret < 0) {
1369 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
1370 "Execution continues but there might be problem for already\n"
1371 "running applications that wishes to register.");
0fdd1e2c 1372 }
1d4b027a
DG
1373
1374 while (1) {
1375 DBG("Accepting application registration");
273ea72c 1376
5eb91c98
DG
1377 nb_fd = LTTNG_POLL_GETNB(&events);
1378
273ea72c 1379 /* Inifinite blocking call, waiting for transmission */
88f2b785 1380 restart:
139ac872 1381 health_poll_update(&health_thread_app_reg);
5eb91c98 1382 ret = lttng_poll_wait(&events, -1);
139ac872 1383 health_poll_update(&health_thread_app_reg);
273ea72c 1384 if (ret < 0) {
88f2b785
MD
1385 /*
1386 * Restart interrupted system call.
1387 */
1388 if (errno == EINTR) {
1389 goto restart;
1390 }
273ea72c
DG
1391 goto error;
1392 }
1393
5eb91c98 1394 for (i = 0; i < nb_fd; i++) {
139ac872
MD
1395 health_code_update(&health_thread_app_reg);
1396
5eb91c98
DG
1397 /* Fetch once the poll data */
1398 revents = LTTNG_POLL_GETEV(&events, i);
1399 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 1400
5eb91c98
DG
1401 /* Thread quit pipe has been closed. Killing thread. */
1402 ret = check_thread_quit_pipe(pollfd, revents);
1403 if (ret) {
139ac872
MD
1404 err = 0;
1405 goto exit;
90014c57 1406 }
1d4b027a 1407
5eb91c98
DG
1408 /* Event on the registration socket */
1409 if (pollfd == apps_sock) {
1410 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1411 ERR("Register apps socket poll error");
1412 goto error;
1413 } else if (revents & LPOLLIN) {
1414 sock = lttcomm_accept_unix_sock(apps_sock);
1415 if (sock < 0) {
1416 goto error;
1417 }
099e26bd 1418
b662582b
DG
1419 /*
1420 * Set the CLOEXEC flag. Return code is useless because
1421 * either way, the show must go on.
1422 */
1423 (void) utils_set_fd_cloexec(sock);
1424
5eb91c98 1425 /* Create UST registration command for enqueuing */
ba7f0ae5 1426 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 1427 if (ust_cmd == NULL) {
76d7553f 1428 PERROR("ust command zmalloc");
5eb91c98
DG
1429 goto error;
1430 }
1d4b027a 1431
5eb91c98
DG
1432 /*
1433 * Using message-based transmissions to ensure we don't
1434 * have to deal with partially received messages.
1435 */
4063050c
MD
1436 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1437 if (ret < 0) {
1438 ERR("Exhausted file descriptors allowed for applications.");
1439 free(ust_cmd);
1440 ret = close(sock);
1441 if (ret) {
1442 PERROR("close");
1443 }
1444 sock = -1;
1445 continue;
1446 }
139ac872 1447 health_code_update(&health_thread_app_reg);
5eb91c98
DG
1448 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1449 sizeof(struct ust_register_msg));
1450 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1451 if (ret < 0) {
76d7553f 1452 PERROR("lttcomm_recv_unix_sock register apps");
5eb91c98
DG
1453 } else {
1454 ERR("Wrong size received on apps register");
1455 }
1456 free(ust_cmd);
76d7553f
MD
1457 ret = close(sock);
1458 if (ret) {
1459 PERROR("close");
1460 }
4063050c 1461 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 1462 sock = -1;
5eb91c98
DG
1463 continue;
1464 }
139ac872 1465 health_code_update(&health_thread_app_reg);
099e26bd 1466
5eb91c98 1467 ust_cmd->sock = sock;
34a2494f 1468 sock = -1;
099e26bd 1469
5eb91c98
DG
1470 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1471 " gid:%d sock:%d name:%s (version %d.%d)",
1472 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1473 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1474 ust_cmd->sock, ust_cmd->reg_msg.name,
1475 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 1476
5eb91c98
DG
1477 /*
1478 * Lock free enqueue the registration request. The red pill
54d01ffb 1479 * has been taken! This apps will be part of the *system*.
5eb91c98
DG
1480 */
1481 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1482
1483 /*
1484 * Wake the registration queue futex. Implicit memory
1485 * barrier with the exchange in cds_wfq_enqueue.
1486 */
1487 futex_nto1_wake(&ust_cmd_queue.futex);
1488 }
1489 }
90014c57 1490 }
1d4b027a
DG
1491 }
1492
139ac872 1493exit:
1d4b027a 1494error:
139ac872
MD
1495 if (err) {
1496 health_error(&health_thread_app_reg);
1497 ERR("Health error occurred in %s", __func__);
1498 }
1499 health_exit(&health_thread_app_reg);
1500
0fdd1e2c
DG
1501 /* Notify that the registration thread is gone */
1502 notify_ust_apps(0);
1503
a4b35e07 1504 if (apps_sock >= 0) {
76d7553f
MD
1505 ret = close(apps_sock);
1506 if (ret) {
1507 PERROR("close");
1508 }
a4b35e07 1509 }
46c3f085 1510 if (sock >= 0) {
76d7553f
MD
1511 ret = close(sock);
1512 if (ret) {
1513 PERROR("close");
1514 }
4063050c 1515 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 1516 }
273ea72c 1517 unlink(apps_unix_sock_path);
0fdd1e2c 1518
76d7553f 1519error_poll_add:
5eb91c98 1520 lttng_poll_clean(&events);
76d7553f
MD
1521error_listen:
1522error_create_poll:
1523 DBG("UST Registration thread cleanup complete");
5eb91c98 1524
1d4b027a
DG
1525 return NULL;
1526}
1527
8c0faa1d 1528/*
3bd1e081 1529 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 1530 * exec or it will fails.
8c0faa1d 1531 */
3bd1e081 1532static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 1533{
a23ec3a7 1534 int ret, clock_ret;
ee0b0061
DG
1535 struct timespec timeout;
1536
a23ec3a7
DG
1537 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1538 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 1539
a23ec3a7
DG
1540 /* Setup pthread condition */
1541 ret = pthread_condattr_init(&consumer_data->condattr);
1542 if (ret != 0) {
1543 errno = ret;
1544 PERROR("pthread_condattr_init consumer data");
1545 goto error;
1546 }
1547
1548 /*
1549 * Set the monotonic clock in order to make sure we DO NOT jump in time
1550 * between the clock_gettime() call and the timedwait call. See bug #324
1551 * for a more details and how we noticed it.
1552 */
1553 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1554 if (ret != 0) {
1555 errno = ret;
1556 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
1557 goto error;
1558 }
8c0faa1d 1559
a23ec3a7
DG
1560 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1561 if (ret != 0) {
1562 errno = ret;
1563 PERROR("pthread_cond_init consumer data");
1564 goto error;
1565 }
1566
1567 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1568 consumer_data);
8c0faa1d 1569 if (ret != 0) {
3bd1e081 1570 PERROR("pthread_create consumer");
ee0b0061 1571 ret = -1;
8c0faa1d
DG
1572 goto error;
1573 }
1574
a23ec3a7
DG
1575 /* We are about to wait on a pthread condition */
1576 pthread_mutex_lock(&consumer_data->cond_mutex);
1577
ee0b0061 1578 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
1579 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1580 /*
1581 * Set the timeout for the condition timed wait even if the clock gettime
1582 * call fails since we might loop on that call and we want to avoid to
1583 * increment the timeout too many times.
1584 */
1585 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1586
1587 /*
1588 * The following loop COULD be skipped in some conditions so this is why we
1589 * set ret to 0 in order to make sure at least one round of the loop is
1590 * done.
1591 */
1592 ret = 0;
1593
1594 /*
1595 * Loop until the condition is reached or when a timeout is reached. Note
1596 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1597 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1598 * possible. This loop does not take any chances and works with both of
1599 * them.
1600 */
1601 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1602 if (clock_ret < 0) {
1603 PERROR("clock_gettime spawn consumer");
1604 /* Infinite wait for the consumerd thread to be ready */
1605 ret = pthread_cond_wait(&consumer_data->cond,
1606 &consumer_data->cond_mutex);
1607 } else {
1608 ret = pthread_cond_timedwait(&consumer_data->cond,
1609 &consumer_data->cond_mutex, &timeout);
1610 }
ee0b0061 1611 }
8c0faa1d 1612
a23ec3a7
DG
1613 /* Release the pthread condition */
1614 pthread_mutex_unlock(&consumer_data->cond_mutex);
1615
1616 if (ret != 0) {
1617 errno = ret;
1618 if (ret == ETIMEDOUT) {
ee0b0061
DG
1619 /*
1620 * Call has timed out so we kill the kconsumerd_thread and return
1621 * an error.
1622 */
a23ec3a7
DG
1623 ERR("Condition timed out. The consumer thread was never ready."
1624 " Killing it");
3bd1e081 1625 ret = pthread_cancel(consumer_data->thread);
ee0b0061 1626 if (ret < 0) {
3bd1e081 1627 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
1628 }
1629 } else {
a23ec3a7 1630 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061
DG
1631 }
1632 goto error;
1633 }
1634
3bd1e081
MD
1635 pthread_mutex_lock(&consumer_data->pid_mutex);
1636 if (consumer_data->pid == 0) {
a23ec3a7 1637 ERR("Consumerd did not start");
3bd1e081 1638 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
1639 goto error;
1640 }
3bd1e081 1641 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 1642
8c0faa1d
DG
1643 return 0;
1644
1645error:
1646 return ret;
1647}
1648
d9800920 1649/*
3bd1e081 1650 * Join consumer thread
d9800920 1651 */
3bd1e081 1652static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
1653{
1654 void *status;
1655 int ret;
1656
e8209f6b
DG
1657 /* Consumer pid must be a real one. */
1658 if (consumer_data->pid > 0) {
3bd1e081 1659 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 1660 if (ret) {
3bd1e081 1661 ERR("Error killing consumer daemon");
cf3af59e
MD
1662 return ret;
1663 }
3bd1e081 1664 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
1665 } else {
1666 return 0;
1667 }
1668}
1669
8c0faa1d 1670/*
3bd1e081 1671 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 1672 *
d063d709 1673 * Return pid if successful else -1.
8c0faa1d 1674 */
3bd1e081 1675static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
1676{
1677 int ret;
1678 pid_t pid;
94c55f17 1679 const char *consumer_to_use;
53086306 1680 const char *verbosity;
94c55f17 1681 struct stat st;
8c0faa1d 1682
3bd1e081 1683 DBG("Spawning consumerd");
c49dc785 1684
8c0faa1d
DG
1685 pid = fork();
1686 if (pid == 0) {
1687 /*
3bd1e081 1688 * Exec consumerd.
8c0faa1d 1689 */
daee5345 1690 if (opt_verbose_consumer) {
53086306
DG
1691 verbosity = "--verbose";
1692 } else {
1693 verbosity = "--quiet";
1694 }
3bd1e081
MD
1695 switch (consumer_data->type) {
1696 case LTTNG_CONSUMER_KERNEL:
94c55f17 1697 /*
c7704d57
DG
1698 * Find out which consumerd to execute. We will first try the
1699 * 64-bit path, then the sessiond's installation directory, and
1700 * fallback on the 32-bit one,
94c55f17 1701 */
63a799e8
AM
1702 DBG3("Looking for a kernel consumer at these locations:");
1703 DBG3(" 1) %s", consumerd64_bin);
1704 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1705 DBG3(" 3) %s", consumerd32_bin);
94c55f17 1706 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 1707 DBG3("Found location #1");
94c55f17 1708 consumer_to_use = consumerd64_bin;
94c55f17 1709 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 1710 DBG3("Found location #2");
94c55f17 1711 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 1712 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 1713 DBG3("Found location #3");
eb1e0bd4 1714 consumer_to_use = consumerd32_bin;
94c55f17 1715 } else {
63a799e8 1716 DBG("Could not find any valid consumerd executable");
94c55f17
AM
1717 break;
1718 }
1719 DBG("Using kernel consumer at: %s", consumer_to_use);
1720 execl(consumer_to_use,
1721 "lttng-consumerd", verbosity, "-k",
1722 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1723 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1724 NULL);
3bd1e081 1725 break;
7753dea8
MD
1726 case LTTNG_CONSUMER64_UST:
1727 {
b1e0b6b6 1728 char *tmpnew = NULL;
8f4905da
MD
1729
1730 if (consumerd64_libdir[0] != '\0') {
1731 char *tmp;
1732 size_t tmplen;
1733
1734 tmp = getenv("LD_LIBRARY_PATH");
1735 if (!tmp) {
1736 tmp = "";
1737 }
1738 tmplen = strlen("LD_LIBRARY_PATH=")
1739 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1740 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1741 if (!tmpnew) {
1742 ret = -ENOMEM;
1743 goto error;
1744 }
1745 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1746 strcat(tmpnew, consumerd64_libdir);
1747 if (tmp[0] != '\0') {
1748 strcat(tmpnew, ":");
1749 strcat(tmpnew, tmp);
1750 }
1751 ret = putenv(tmpnew);
1752 if (ret) {
1753 ret = -errno;
1754 goto error;
1755 }
1756 }
94c55f17 1757 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 1758 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
1759 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1760 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1761 NULL);
8f4905da
MD
1762 if (consumerd64_libdir[0] != '\0') {
1763 free(tmpnew);
1764 }
1765 if (ret) {
1766 goto error;
1767 }
3bd1e081 1768 break;
7753dea8
MD
1769 }
1770 case LTTNG_CONSUMER32_UST:
1771 {
937dde8e 1772 char *tmpnew = NULL;
8f4905da
MD
1773
1774 if (consumerd32_libdir[0] != '\0') {
1775 char *tmp;
1776 size_t tmplen;
1777
1778 tmp = getenv("LD_LIBRARY_PATH");
1779 if (!tmp) {
1780 tmp = "";
1781 }
1782 tmplen = strlen("LD_LIBRARY_PATH=")
1783 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1784 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1785 if (!tmpnew) {
1786 ret = -ENOMEM;
1787 goto error;
1788 }
1789 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1790 strcat(tmpnew, consumerd32_libdir);
1791 if (tmp[0] != '\0') {
1792 strcat(tmpnew, ":");
1793 strcat(tmpnew, tmp);
1794 }
1795 ret = putenv(tmpnew);
1796 if (ret) {
1797 ret = -errno;
1798 goto error;
1799 }
1800 }
94c55f17 1801 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 1802 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
1803 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1804 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1805 NULL);
8f4905da
MD
1806 if (consumerd32_libdir[0] != '\0') {
1807 free(tmpnew);
1808 }
1809 if (ret) {
1810 goto error;
1811 }
7753dea8
MD
1812 break;
1813 }
3bd1e081 1814 default:
76d7553f 1815 PERROR("unknown consumer type");
3bd1e081
MD
1816 exit(EXIT_FAILURE);
1817 }
8c0faa1d 1818 if (errno != 0) {
76d7553f 1819 PERROR("kernel start consumer exec");
8c0faa1d
DG
1820 }
1821 exit(EXIT_FAILURE);
1822 } else if (pid > 0) {
1823 ret = pid;
8c0faa1d 1824 } else {
76d7553f 1825 PERROR("start consumer fork");
8c0faa1d 1826 ret = -errno;
8c0faa1d 1827 }
8f4905da 1828error:
8c0faa1d
DG
1829 return ret;
1830}
1831
693bd40b 1832/*
3bd1e081 1833 * Spawn the consumerd daemon and session daemon thread.
693bd40b 1834 */
3bd1e081 1835static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 1836{
edb8b045
DG
1837 int ret, err;
1838
1839 /*
1840 * Set the listen() state on the socket since there is a possible race
1841 * between the exec() of the consumer daemon and this call if place in the
1842 * consumer thread. See bug #366 for more details.
1843 */
1844 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1845 if (ret < 0) {
1846 goto error;
1847 }
693bd40b 1848
3bd1e081
MD
1849 pthread_mutex_lock(&consumer_data->pid_mutex);
1850 if (consumer_data->pid != 0) {
1851 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
1852 goto end;
1853 }
693bd40b 1854
3bd1e081 1855 ret = spawn_consumerd(consumer_data);
c49dc785 1856 if (ret < 0) {
3bd1e081
MD
1857 ERR("Spawning consumerd failed");
1858 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 1859 goto error;
693bd40b 1860 }
c49dc785 1861
3bd1e081
MD
1862 /* Setting up the consumer_data pid */
1863 consumer_data->pid = ret;
48842b30 1864 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 1865 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 1866
3bd1e081
MD
1867 DBG2("Spawning consumer control thread");
1868 ret = spawn_consumer_thread(consumer_data);
693bd40b 1869 if (ret < 0) {
3bd1e081 1870 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
1871 goto error;
1872 }
1873
c49dc785 1874end:
693bd40b
DG
1875 return 0;
1876
1877error:
edb8b045
DG
1878 /* Cleanup already created socket on error. */
1879 if (consumer_data->err_sock >= 0) {
1880 err = close(consumer_data->err_sock);
1881 if (err < 0) {
1882 PERROR("close consumer data error socket");
1883 }
1884 }
693bd40b
DG
1885 return ret;
1886}
1887
44a5e5eb 1888/*
139ac872
MD
1889 * Compute health status of each consumer. If one of them is zero (bad
1890 * state), we return 0.
44a5e5eb
DG
1891 */
1892static int check_consumer_health(void)
1893{
1894 int ret;
1895
139ac872
MD
1896 ret = health_check_state(&kconsumer_data.health) &&
1897 health_check_state(&ustconsumer32_data.health) &&
44a5e5eb
DG
1898 health_check_state(&ustconsumer64_data.health);
1899
1900 DBG3("Health consumer check %d", ret);
1901
1902 return ret;
1903}
1904
b73401da 1905/*
096102bd 1906 * Setup necessary data for kernel tracer action.
b73401da 1907 */
096102bd 1908static int init_kernel_tracer(void)
b73401da
DG
1909{
1910 int ret;
b73401da 1911
096102bd
DG
1912 /* Modprobe lttng kernel modules */
1913 ret = modprobe_lttng_control();
b73401da 1914 if (ret < 0) {
b73401da
DG
1915 goto error;
1916 }
1917
096102bd
DG
1918 /* Open debugfs lttng */
1919 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1920 if (kernel_tracer_fd < 0) {
1921 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
1922 ret = -1;
1923 goto error_open;
54d01ffb
DG
1924 }
1925
2f77fc4b
DG
1926 /* Validate kernel version */
1927 ret = kernel_validate_version(kernel_tracer_fd);
1928 if (ret < 0) {
1929 goto error_version;
b551a063 1930 }
54d01ffb 1931
2f77fc4b
DG
1932 ret = modprobe_lttng_data();
1933 if (ret < 0) {
1934 goto error_modules;
54d01ffb
DG
1935 }
1936
2f77fc4b
DG
1937 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1938 return 0;
1939
1940error_version:
1941 modprobe_remove_lttng_control();
1942 ret = close(kernel_tracer_fd);
1943 if (ret) {
1944 PERROR("close");
b551a063 1945 }
2f77fc4b 1946 kernel_tracer_fd = -1;
f73fabfd 1947 return LTTNG_ERR_KERN_VERSION;
b551a063 1948
2f77fc4b
DG
1949error_modules:
1950 ret = close(kernel_tracer_fd);
1951 if (ret) {
1952 PERROR("close");
b551a063 1953 }
54d01ffb 1954
2f77fc4b
DG
1955error_open:
1956 modprobe_remove_lttng_control();
54d01ffb
DG
1957
1958error:
2f77fc4b
DG
1959 WARN("No kernel tracer available");
1960 kernel_tracer_fd = -1;
1961 if (!is_root) {
f73fabfd 1962 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 1963 } else {
f73fabfd 1964 return LTTNG_ERR_KERN_NA;
2f77fc4b 1965 }
54d01ffb
DG
1966}
1967
2f77fc4b 1968
54d01ffb 1969/*
2f77fc4b
DG
1970 * Copy consumer output from the tracing session to the domain session. The
1971 * function also applies the right modification on a per domain basis for the
1972 * trace files destination directory.
54d01ffb 1973 */
2f77fc4b 1974static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
1975{
1976 int ret;
2f77fc4b
DG
1977 const char *dir_name;
1978 struct consumer_output *consumer;
1979
1980 assert(session);
1981 assert(session->consumer);
54d01ffb 1982
b551a063
DG
1983 switch (domain) {
1984 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 1985 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
1986 /*
1987 * XXX: We should audit the session creation and what this function
1988 * does "extra" in order to avoid a destroy since this function is used
1989 * in the domain session creation (kernel and ust) only. Same for UST
1990 * domain.
1991 */
1992 if (session->kernel_session->consumer) {
1993 consumer_destroy_output(session->kernel_session->consumer);
1994 }
2f77fc4b
DG
1995 session->kernel_session->consumer =
1996 consumer_copy_output(session->consumer);
1997 /* Ease our life a bit for the next part */
1998 consumer = session->kernel_session->consumer;
1999 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063
DG
2000 break;
2001 case LTTNG_DOMAIN_UST:
2f77fc4b 2002 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2003 if (session->ust_session->consumer) {
2004 consumer_destroy_output(session->ust_session->consumer);
2005 }
2f77fc4b
DG
2006 session->ust_session->consumer =
2007 consumer_copy_output(session->consumer);
2008 /* Ease our life a bit for the next part */
2009 consumer = session->ust_session->consumer;
2010 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2011 break;
2012 default:
f73fabfd 2013 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2014 goto error;
2015 }
2016
2f77fc4b 2017 /* Append correct directory to subdir */
c30ce0b3
CB
2018 strncat(consumer->subdir, dir_name,
2019 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2020 DBG3("Copy session consumer subdir %s", consumer->subdir);
2021
f73fabfd 2022 ret = LTTNG_OK;
54d01ffb
DG
2023
2024error:
2025 return ret;
2026}
2027
00e2e675 2028/*
2f77fc4b 2029 * Create an UST session and add it to the session ust list.
00e2e675 2030 */
2f77fc4b
DG
2031static int create_ust_session(struct ltt_session *session,
2032 struct lttng_domain *domain)
00e2e675
DG
2033{
2034 int ret;
2f77fc4b 2035 struct ltt_ust_session *lus = NULL;
00e2e675 2036
a4b92340 2037 assert(session);
2f77fc4b
DG
2038 assert(domain);
2039 assert(session->consumer);
a4b92340 2040
2f77fc4b
DG
2041 switch (domain->type) {
2042 case LTTNG_DOMAIN_UST:
2043 break;
2044 default:
2045 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2046 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2047 goto error;
2048 }
2049
2f77fc4b
DG
2050 DBG("Creating UST session");
2051
2052 lus = trace_ust_create_session(session->path, session->id, domain);
2053 if (lus == NULL) {
f73fabfd 2054 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2055 goto error;
2056 }
2057
2f77fc4b
DG
2058 lus->uid = session->uid;
2059 lus->gid = session->gid;
2060 session->ust_session = lus;
00e2e675 2061
2f77fc4b
DG
2062 /* Copy session output to the newly created UST session */
2063 ret = copy_session_consumer(domain->type, session);
f73fabfd 2064 if (ret != LTTNG_OK) {
00e2e675
DG
2065 goto error;
2066 }
2067
f73fabfd 2068 return LTTNG_OK;
00e2e675
DG
2069
2070error:
2f77fc4b
DG
2071 free(lus);
2072 session->ust_session = NULL;
00e2e675
DG
2073 return ret;
2074}
2075
2076/*
2f77fc4b 2077 * Create a kernel tracer session then create the default channel.
00e2e675 2078 */
2f77fc4b 2079static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2080{
2081 int ret;
a4b92340 2082
2f77fc4b 2083 DBG("Creating kernel session");
00e2e675 2084
2f77fc4b
DG
2085 ret = kernel_create_session(session, kernel_tracer_fd);
2086 if (ret < 0) {
f73fabfd 2087 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2088 goto error;
2089 }
2090
2f77fc4b
DG
2091 /* Code flow safety */
2092 assert(session->kernel_session);
2093
2094 /* Copy session output to the newly created Kernel session */
2095 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2096 if (ret != LTTNG_OK) {
a4b92340
DG
2097 goto error;
2098 }
2099
2f77fc4b
DG
2100 /* Create directory(ies) on local filesystem. */
2101 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2102 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2103 ret = run_as_mkdir_recursive(
2104 session->kernel_session->consumer->dst.trace_path,
2105 S_IRWXU | S_IRWXG, session->uid, session->gid);
2106 if (ret < 0) {
2107 if (ret != -EEXIST) {
2108 ERR("Trace directory creation error");
00e2e675
DG
2109 goto error;
2110 }
00e2e675 2111 }
2f77fc4b 2112 }
00e2e675 2113
2f77fc4b
DG
2114 session->kernel_session->uid = session->uid;
2115 session->kernel_session->gid = session->gid;
00e2e675 2116
f73fabfd 2117 return LTTNG_OK;
00e2e675 2118
2f77fc4b
DG
2119error:
2120 trace_kernel_destroy_session(session->kernel_session);
2121 session->kernel_session = NULL;
2122 return ret;
2123}
00e2e675 2124
2f77fc4b
DG
2125/*
2126 * Count number of session permitted by uid/gid.
2127 */
2128static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2129{
2130 unsigned int i = 0;
2131 struct ltt_session *session;
07424f16 2132
2f77fc4b
DG
2133 DBG("Counting number of available session for UID %d GID %d",
2134 uid, gid);
2135 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2136 /*
2f77fc4b 2137 * Only list the sessions the user can control.
00e2e675 2138 */
2f77fc4b
DG
2139 if (!session_access_ok(session, uid, gid)) {
2140 continue;
2141 }
2142 i++;
a4b92340 2143 }
2f77fc4b 2144 return i;
00e2e675
DG
2145}
2146
54d01ffb
DG
2147/*
2148 * Process the command requested by the lttng client within the command
2149 * context structure. This function make sure that the return structure (llm)
2150 * is set and ready for transmission before returning.
2151 *
2152 * Return any error encountered or 0 for success.
53a80697
MD
2153 *
2154 * "sock" is only used for special-case var. len data.
54d01ffb 2155 */
53a80697
MD
2156static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2157 int *sock_error)
54d01ffb 2158{
f73fabfd 2159 int ret = LTTNG_OK;
44d3bd01 2160 int need_tracing_session = 1;
2e09ba09 2161 int need_domain;
54d01ffb
DG
2162
2163 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2164
53a80697
MD
2165 *sock_error = 0;
2166
2e09ba09
MD
2167 switch (cmd_ctx->lsm->cmd_type) {
2168 case LTTNG_CREATE_SESSION:
2169 case LTTNG_DESTROY_SESSION:
2170 case LTTNG_LIST_SESSIONS:
2171 case LTTNG_LIST_DOMAINS:
2172 case LTTNG_START_TRACE:
2173 case LTTNG_STOP_TRACE:
6d805429 2174 case LTTNG_DATA_PENDING:
2e09ba09 2175 need_domain = 0;
3aace903 2176 break;
2e09ba09
MD
2177 default:
2178 need_domain = 1;
2179 }
2180
2181 if (opt_no_kernel && need_domain
2182 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2183 if (!is_root) {
f73fabfd 2184 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2185 } else {
f73fabfd 2186 ret = LTTNG_ERR_KERN_NA;
531d29f9 2187 }
4fba7219
DG
2188 goto error;
2189 }
2190
8d3113b2
DG
2191 /* Deny register consumer if we already have a spawned consumer. */
2192 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2193 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2194 if (kconsumer_data.pid > 0) {
f73fabfd 2195 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2196 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2197 goto error;
2198 }
2199 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2200 }
2201
54d01ffb
DG
2202 /*
2203 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2204 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2205 * command.
2206 */
2207 switch(cmd_ctx->lsm->cmd_type) {
2208 case LTTNG_LIST_SESSIONS:
2209 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2210 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2211 case LTTNG_LIST_DOMAINS:
2212 case LTTNG_LIST_CHANNELS:
2213 case LTTNG_LIST_EVENTS:
2214 break;
2215 default:
2216 /* Setup lttng message with no payload */
2217 ret = setup_lttng_msg(cmd_ctx, 0);
2218 if (ret < 0) {
2219 /* This label does not try to unlock the session */
2220 goto init_setup_error;
2221 }
2222 }
2223
2224 /* Commands that DO NOT need a session. */
2225 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2226 case LTTNG_CREATE_SESSION:
2e09ba09 2227 case LTTNG_CALIBRATE:
54d01ffb
DG
2228 case LTTNG_LIST_SESSIONS:
2229 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2230 case LTTNG_LIST_TRACEPOINT_FIELDS:
44d3bd01 2231 need_tracing_session = 0;
54d01ffb
DG
2232 break;
2233 default:
2234 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2235 /*
2236 * We keep the session list lock across _all_ commands
2237 * for now, because the per-session lock does not
2238 * handle teardown properly.
2239 */
74babd95 2240 session_lock_list();
54d01ffb
DG
2241 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2242 if (cmd_ctx->session == NULL) {
2243 if (cmd_ctx->lsm->session.name != NULL) {
f73fabfd 2244 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2245 } else {
2246 /* If no session name specified */
f73fabfd 2247 ret = LTTNG_ERR_SELECT_SESS;
54d01ffb
DG
2248 }
2249 goto error;
2250 } else {
2251 /* Acquire lock for the session */
2252 session_lock(cmd_ctx->session);
2253 }
2254 break;
2255 }
b389abbe 2256
2e09ba09
MD
2257 if (!need_domain) {
2258 goto skip_domain;
2259 }
a4b92340 2260
54d01ffb
DG
2261 /*
2262 * Check domain type for specific "pre-action".
2263 */
2264 switch (cmd_ctx->lsm->domain.type) {
2265 case LTTNG_DOMAIN_KERNEL:
d1f1c568 2266 if (!is_root) {
f73fabfd 2267 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
2268 goto error;
2269 }
2270
54d01ffb 2271 /* Kernel tracer check */
a4b35e07 2272 if (kernel_tracer_fd == -1) {
54d01ffb 2273 /* Basically, load kernel tracer modules */
096102bd
DG
2274 ret = init_kernel_tracer();
2275 if (ret != 0) {
54d01ffb
DG
2276 goto error;
2277 }
2278 }
5eb91c98 2279
5c827ce0
DG
2280 /* Consumer is in an ERROR state. Report back to client */
2281 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2282 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2283 goto error;
2284 }
2285
54d01ffb 2286 /* Need a session for kernel command */
44d3bd01 2287 if (need_tracing_session) {
54d01ffb 2288 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 2289 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 2290 if (ret < 0) {
f73fabfd 2291 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
2292 goto error;
2293 }
b389abbe 2294 }
7d29a247 2295
54d01ffb 2296 /* Start the kernel consumer daemon */
3bd1e081
MD
2297 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2298 if (kconsumer_data.pid == 0 &&
a4b92340
DG
2299 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2300 cmd_ctx->session->start_consumer) {
3bd1e081
MD
2301 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2302 ret = start_consumerd(&kconsumer_data);
7d29a247 2303 if (ret < 0) {
f73fabfd 2304 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 2305 goto error;
950131af 2306 }
5c827ce0 2307 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
2308 } else {
2309 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 2310 }
173af62f 2311
a4b92340
DG
2312 /*
2313 * The consumer was just spawned so we need to add the socket to
2314 * the consumer output of the session if exist.
2315 */
2316 ret = consumer_create_socket(&kconsumer_data,
2317 cmd_ctx->session->kernel_session->consumer);
2318 if (ret < 0) {
2319 goto error;
173af62f 2320 }
0d0c377a 2321 }
5c827ce0 2322
54d01ffb 2323 break;
2bdd86d4 2324 case LTTNG_DOMAIN_UST:
44d3bd01 2325 {
5c827ce0
DG
2326 /* Consumer is in an ERROR state. Report back to client */
2327 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2328 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2329 goto error;
2330 }
2331
44d3bd01 2332 if (need_tracing_session) {
a4b92340 2333 /* Create UST session if none exist. */
f6a9efaa 2334 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 2335 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 2336 &cmd_ctx->lsm->domain);
f73fabfd 2337 if (ret != LTTNG_OK) {
44d3bd01
DG
2338 goto error;
2339 }
2340 }
00e2e675 2341
7753dea8
MD
2342 /* Start the UST consumer daemons */
2343 /* 64-bit */
2344 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 2345 if (consumerd64_bin[0] != '\0' &&
7753dea8 2346 ustconsumer64_data.pid == 0 &&
a4b92340
DG
2347 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2348 cmd_ctx->session->start_consumer) {
7753dea8
MD
2349 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2350 ret = start_consumerd(&ustconsumer64_data);
2bdd86d4 2351 if (ret < 0) {
f73fabfd 2352 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
173af62f 2353 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2bdd86d4
MD
2354 goto error;
2355 }
48842b30 2356
173af62f 2357 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
5c827ce0 2358 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3ff2ecac 2359 } else {
7753dea8
MD
2360 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2361 }
173af62f
DG
2362
2363 /*
2364 * Setup socket for consumer 64 bit. No need for atomic access
2365 * since it was set above and can ONLY be set in this thread.
2366 */
a4b92340
DG
2367 ret = consumer_create_socket(&ustconsumer64_data,
2368 cmd_ctx->session->ust_session->consumer);
2369 if (ret < 0) {
2370 goto error;
173af62f
DG
2371 }
2372
7753dea8 2373 /* 32-bit */
fc7a59ce 2374 if (consumerd32_bin[0] != '\0' &&
7753dea8 2375 ustconsumer32_data.pid == 0 &&
a4b92340
DG
2376 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2377 cmd_ctx->session->start_consumer) {
7753dea8
MD
2378 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2379 ret = start_consumerd(&ustconsumer32_data);
2380 if (ret < 0) {
f73fabfd 2381 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
173af62f 2382 uatomic_set(&ust_consumerd32_fd, -EINVAL);
7753dea8
MD
2383 goto error;
2384 }
5c827ce0 2385
173af62f 2386 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
5c827ce0 2387 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
7753dea8
MD
2388 } else {
2389 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2bdd86d4 2390 }
173af62f
DG
2391
2392 /*
2393 * Setup socket for consumer 64 bit. No need for atomic access
2394 * since it was set above and can ONLY be set in this thread.
2395 */
a4b92340
DG
2396 ret = consumer_create_socket(&ustconsumer32_data,
2397 cmd_ctx->session->ust_session->consumer);
2398 if (ret < 0) {
2399 goto error;
173af62f 2400 }
44d3bd01
DG
2401 }
2402 break;
48842b30 2403 }
54d01ffb 2404 default:
54d01ffb
DG
2405 break;
2406 }
2e09ba09 2407skip_domain:
33a2b854 2408
5c827ce0
DG
2409 /* Validate consumer daemon state when start/stop trace command */
2410 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2411 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2412 switch (cmd_ctx->lsm->domain.type) {
2413 case LTTNG_DOMAIN_UST:
2414 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2415 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2416 goto error;
2417 }
2418 break;
2419 case LTTNG_DOMAIN_KERNEL:
2420 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2421 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2422 goto error;
2423 }
2424 break;
2425 }
2426 }
2427
8e0af1b4
MD
2428 /*
2429 * Check that the UID or GID match that of the tracing session.
2430 * The root user can interact with all sessions.
2431 */
2432 if (need_tracing_session) {
2433 if (!session_access_ok(cmd_ctx->session,
730389d9
DG
2434 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2435 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
f73fabfd 2436 ret = LTTNG_ERR_EPERM;
8e0af1b4
MD
2437 goto error;
2438 }
2439 }
2440
54d01ffb
DG
2441 /* Process by command type */
2442 switch (cmd_ctx->lsm->cmd_type) {
2443 case LTTNG_ADD_CONTEXT:
2444 {
2445 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2446 cmd_ctx->lsm->u.context.channel_name,
2447 cmd_ctx->lsm->u.context.event_name,
979e618e 2448 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
54d01ffb
DG
2449 break;
2450 }
2451 case LTTNG_DISABLE_CHANNEL:
2452 {
2453 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2454 cmd_ctx->lsm->u.disable.channel_name);
2455 break;
2456 }
2457 case LTTNG_DISABLE_EVENT:
2458 {
2459 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2460 cmd_ctx->lsm->u.disable.channel_name,
2461 cmd_ctx->lsm->u.disable.name);
33a2b854
DG
2462 break;
2463 }
54d01ffb
DG
2464 case LTTNG_DISABLE_ALL_EVENT:
2465 {
2bdd86d4 2466 DBG("Disabling all events");
54d01ffb
DG
2467
2468 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2469 cmd_ctx->lsm->u.disable.channel_name);
2470 break;
2471 }
00e2e675
DG
2472 case LTTNG_DISABLE_CONSUMER:
2473 {
2474 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2475 break;
2476 }
54d01ffb
DG
2477 case LTTNG_ENABLE_CHANNEL:
2478 {
aea829b3 2479 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 2480 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
54d01ffb
DG
2481 break;
2482 }
00e2e675
DG
2483 case LTTNG_ENABLE_CONSUMER:
2484 {
a4b92340
DG
2485 /*
2486 * XXX: 0 means that this URI should be applied on the session. Should
2487 * be a DOMAIN enuam.
2488 */
00e2e675 2489 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
f73fabfd 2490 if (ret != LTTNG_OK) {
a4b92340
DG
2491 goto error;
2492 }
2493
2494 if (cmd_ctx->lsm->domain.type == 0) {
2495 /* Add the URI for the UST session if a consumer is present. */
2496 if (cmd_ctx->session->ust_session &&
2497 cmd_ctx->session->ust_session->consumer) {
2498 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2499 } else if (cmd_ctx->session->kernel_session &&
2500 cmd_ctx->session->kernel_session->consumer) {
2501 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2502 cmd_ctx->session);
2503 }
2504 }
00e2e675
DG
2505 break;
2506 }
54d01ffb
DG
2507 case LTTNG_ENABLE_EVENT:
2508 {
2509 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2510 cmd_ctx->lsm->u.enable.channel_name,
2f77fc4b 2511 &cmd_ctx->lsm->u.enable.event, kernel_poll_pipe[1]);
54d01ffb
DG
2512 break;
2513 }
2514 case LTTNG_ENABLE_ALL_EVENT:
2515 {
2bdd86d4 2516 DBG("Enabling all events");
54d01ffb
DG
2517
2518 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
8c9ae521 2519 cmd_ctx->lsm->u.enable.channel_name,
2f77fc4b 2520 cmd_ctx->lsm->u.enable.event.type, kernel_poll_pipe[1]);
54d01ffb
DG
2521 break;
2522 }
052da939 2523 case LTTNG_LIST_TRACEPOINTS:
2ef84c95 2524 {
9f19cc17 2525 struct lttng_event *events;
54d01ffb 2526 ssize_t nb_events;
052da939 2527
54d01ffb
DG
2528 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2529 if (nb_events < 0) {
f73fabfd 2530 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2531 ret = -nb_events;
2532 goto error;
2ef84c95
DG
2533 }
2534
2535 /*
2536 * Setup lttng message with payload size set to the event list size in
2537 * bytes and then copy list into the llm payload.
2538 */
052da939 2539 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2ef84c95 2540 if (ret < 0) {
052da939 2541 free(events);
2ef84c95
DG
2542 goto setup_error;
2543 }
2544
2545 /* Copy event list into message payload */
9f19cc17 2546 memcpy(cmd_ctx->llm->payload, events,
052da939 2547 sizeof(struct lttng_event) * nb_events);
2ef84c95 2548
9f19cc17 2549 free(events);
2ef84c95 2550
f73fabfd 2551 ret = LTTNG_OK;
2ef84c95
DG
2552 break;
2553 }
f37d259d
MD
2554 case LTTNG_LIST_TRACEPOINT_FIELDS:
2555 {
2556 struct lttng_event_field *fields;
2557 ssize_t nb_fields;
2558
a4b92340
DG
2559 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2560 &fields);
f37d259d 2561 if (nb_fields < 0) {
f73fabfd 2562 /* Return value is a negative lttng_error_code. */
f37d259d
MD
2563 ret = -nb_fields;
2564 goto error;
2565 }
2566
2567 /*
2568 * Setup lttng message with payload size set to the event list size in
2569 * bytes and then copy list into the llm payload.
2570 */
a4b92340
DG
2571 ret = setup_lttng_msg(cmd_ctx,
2572 sizeof(struct lttng_event_field) * nb_fields);
f37d259d
MD
2573 if (ret < 0) {
2574 free(fields);
2575 goto setup_error;
2576 }
2577
2578 /* Copy event list into message payload */
2579 memcpy(cmd_ctx->llm->payload, fields,
2580 sizeof(struct lttng_event_field) * nb_fields);
2581
2582 free(fields);
2583
f73fabfd 2584 ret = LTTNG_OK;
f37d259d
MD
2585 break;
2586 }
00e2e675
DG
2587 case LTTNG_SET_CONSUMER_URI:
2588 {
a4b92340
DG
2589 size_t nb_uri, len;
2590 struct lttng_uri *uris;
2591
2592 nb_uri = cmd_ctx->lsm->u.uri.size;
2593 len = nb_uri * sizeof(struct lttng_uri);
2594
2595 if (nb_uri == 0) {
f73fabfd 2596 ret = LTTNG_ERR_INVALID;
a4b92340
DG
2597 goto error;
2598 }
2599
2600 uris = zmalloc(len);
2601 if (uris == NULL) {
f73fabfd 2602 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2603 goto error;
2604 }
2605
2606 /* Receive variable len data */
77c7c900 2607 DBG("Receiving %zu URI(s) from client ...", nb_uri);
a4b92340
DG
2608 ret = lttcomm_recv_unix_sock(sock, uris, len);
2609 if (ret <= 0) {
2610 DBG("No URIs received from client... continuing");
2611 *sock_error = 1;
f73fabfd 2612 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2613 free(uris);
a4b92340
DG
2614 goto error;
2615 }
2616
00e2e675 2617 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
a4b92340 2618 nb_uri, uris);
f73fabfd 2619 if (ret != LTTNG_OK) {
b1d41407 2620 free(uris);
a4b92340
DG
2621 goto error;
2622 }
2623
2624 /*
2625 * XXX: 0 means that this URI should be applied on the session. Should
2626 * be a DOMAIN enuam.
2627 */
2628 if (cmd_ctx->lsm->domain.type == 0) {
2629 /* Add the URI for the UST session if a consumer is present. */
2630 if (cmd_ctx->session->ust_session &&
2631 cmd_ctx->session->ust_session->consumer) {
2632 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2633 nb_uri, uris);
2634 } else if (cmd_ctx->session->kernel_session &&
2635 cmd_ctx->session->kernel_session->consumer) {
2636 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2637 cmd_ctx->session, nb_uri, uris);
2638 }
2639 }
2640
b1d41407
DG
2641 free(uris);
2642
00e2e675
DG
2643 break;
2644 }
f3ed775e 2645 case LTTNG_START_TRACE:
8c0faa1d 2646 {
54d01ffb 2647 ret = cmd_start_trace(cmd_ctx->session);
8c0faa1d
DG
2648 break;
2649 }
f3ed775e 2650 case LTTNG_STOP_TRACE:
8c0faa1d 2651 {
54d01ffb 2652 ret = cmd_stop_trace(cmd_ctx->session);
8c0faa1d
DG
2653 break;
2654 }
5e16da05
MD
2655 case LTTNG_CREATE_SESSION:
2656 {
a4b92340
DG
2657 size_t nb_uri, len;
2658 struct lttng_uri *uris = NULL;
2659
2660 nb_uri = cmd_ctx->lsm->u.uri.size;
2661 len = nb_uri * sizeof(struct lttng_uri);
2662
2663 if (nb_uri > 0) {
2664 uris = zmalloc(len);
2665 if (uris == NULL) {
f73fabfd 2666 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2667 goto error;
2668 }
2669
2670 /* Receive variable len data */
77c7c900 2671 DBG("Waiting for %zu URIs from client ...", nb_uri);
a4b92340
DG
2672 ret = lttcomm_recv_unix_sock(sock, uris, len);
2673 if (ret <= 0) {
2674 DBG("No URIs received from client... continuing");
2675 *sock_error = 1;
f73fabfd 2676 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2677 free(uris);
a4b92340
DG
2678 goto error;
2679 }
2680
2681 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2682 DBG("Creating session with ONE network URI is a bad call");
f73fabfd 2683 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2684 free(uris);
a4b92340
DG
2685 goto error;
2686 }
2687 }
2688
2689 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2690 &cmd_ctx->creds);
2691
b1d41407
DG
2692 free(uris);
2693
00e2e675
DG
2694 break;
2695 }
5e16da05
MD
2696 case LTTNG_DESTROY_SESSION:
2697 {
2f77fc4b 2698 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
a4b92340
DG
2699
2700 /* Set session to NULL so we do not unlock it after free. */
256a5576 2701 cmd_ctx->session = NULL;
5461b305 2702 break;
5e16da05 2703 }
9f19cc17 2704 case LTTNG_LIST_DOMAINS:
5e16da05 2705 {
54d01ffb
DG
2706 ssize_t nb_dom;
2707 struct lttng_domain *domains;
5461b305 2708
54d01ffb
DG
2709 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2710 if (nb_dom < 0) {
f73fabfd 2711 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2712 ret = -nb_dom;
2713 goto error;
ce3d728c 2714 }
520ff687 2715
54d01ffb 2716 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
5461b305
DG
2717 if (ret < 0) {
2718 goto setup_error;
520ff687 2719 }
57167058 2720
54d01ffb
DG
2721 /* Copy event list into message payload */
2722 memcpy(cmd_ctx->llm->payload, domains,
2723 nb_dom * sizeof(struct lttng_domain));
2724
2725 free(domains);
5e16da05 2726
f73fabfd 2727 ret = LTTNG_OK;
5e16da05
MD
2728 break;
2729 }
9f19cc17 2730 case LTTNG_LIST_CHANNELS:
5e16da05 2731 {
6775595e 2732 int nb_chan;
54d01ffb
DG
2733 struct lttng_channel *channels;
2734
b551a063
DG
2735 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2736 cmd_ctx->session, &channels);
54d01ffb 2737 if (nb_chan < 0) {
f73fabfd 2738 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2739 ret = -nb_chan;
2740 goto error;
5461b305 2741 }
ca95a216 2742
54d01ffb 2743 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
5461b305
DG
2744 if (ret < 0) {
2745 goto setup_error;
2746 }
9f19cc17 2747
54d01ffb
DG
2748 /* Copy event list into message payload */
2749 memcpy(cmd_ctx->llm->payload, channels,
2750 nb_chan * sizeof(struct lttng_channel));
2751
2752 free(channels);
9f19cc17 2753
f73fabfd 2754 ret = LTTNG_OK;
5461b305 2755 break;
5e16da05 2756 }
9f19cc17 2757 case LTTNG_LIST_EVENTS:
5e16da05 2758 {
b551a063 2759 ssize_t nb_event;
684d34d2 2760 struct lttng_event *events = NULL;
9f19cc17 2761
b551a063 2762 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
54d01ffb
DG
2763 cmd_ctx->lsm->u.list.channel_name, &events);
2764 if (nb_event < 0) {
f73fabfd 2765 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2766 ret = -nb_event;
2767 goto error;
5461b305 2768 }
ca95a216 2769
54d01ffb 2770 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
5461b305
DG
2771 if (ret < 0) {
2772 goto setup_error;
2773 }
9f19cc17 2774
54d01ffb
DG
2775 /* Copy event list into message payload */
2776 memcpy(cmd_ctx->llm->payload, events,
2777 nb_event * sizeof(struct lttng_event));
9f19cc17 2778
54d01ffb 2779 free(events);
9f19cc17 2780
f73fabfd 2781 ret = LTTNG_OK;
5461b305 2782 break;
5e16da05
MD
2783 }
2784 case LTTNG_LIST_SESSIONS:
2785 {
8e0af1b4 2786 unsigned int nr_sessions;
5461b305 2787
8e0af1b4 2788 session_lock_list();
730389d9
DG
2789 nr_sessions = lttng_sessions_count(
2790 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2791 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
d32fb093 2792
8e0af1b4 2793 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
5461b305 2794 if (ret < 0) {
54d01ffb 2795 session_unlock_list();
5461b305 2796 goto setup_error;
e065084a 2797 }
5e16da05 2798
6c9cc2ab 2799 /* Filled the session array */
2f77fc4b 2800 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
730389d9
DG
2801 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2802 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
6c9cc2ab 2803
54d01ffb 2804 session_unlock_list();
5e16da05 2805
f73fabfd 2806 ret = LTTNG_OK;
5e16da05
MD
2807 break;
2808 }
d0254c7c
MD
2809 case LTTNG_CALIBRATE:
2810 {
54d01ffb
DG
2811 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2812 &cmd_ctx->lsm->u.calibrate);
d0254c7c
MD
2813 break;
2814 }
d9800920
DG
2815 case LTTNG_REGISTER_CONSUMER:
2816 {
2f77fc4b
DG
2817 struct consumer_data *cdata;
2818
2819 switch (cmd_ctx->lsm->domain.type) {
2820 case LTTNG_DOMAIN_KERNEL:
2821 cdata = &kconsumer_data;
2822 break;
2823 default:
f73fabfd 2824 ret = LTTNG_ERR_UND;
2f77fc4b
DG
2825 goto error;
2826 }
2827
54d01ffb 2828 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 2829 cmd_ctx->lsm->u.reg.path, cdata);
d9800920
DG
2830 break;
2831 }
53a80697
MD
2832 case LTTNG_SET_FILTER:
2833 {
2834 struct lttng_filter_bytecode *bytecode;
2835
d93c4f1f 2836 if (cmd_ctx->lsm->u.filter.bytecode_len > LTTNG_FILTER_MAX_LEN) {
f73fabfd 2837 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2838 goto error;
2839 }
2840 bytecode = zmalloc(cmd_ctx->lsm->u.filter.bytecode_len);
2841 if (!bytecode) {
f73fabfd 2842 ret = LTTNG_ERR_FILTER_NOMEM;
53a80697
MD
2843 goto error;
2844 }
2845 /* Receive var. len. data */
2846 DBG("Receiving var len data from client ...");
2847 ret = lttcomm_recv_unix_sock(sock, bytecode,
2848 cmd_ctx->lsm->u.filter.bytecode_len);
2849 if (ret <= 0) {
2850 DBG("Nothing recv() from client var len data... continuing");
2851 *sock_error = 1;
f73fabfd 2852 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2853 goto error;
2854 }
2855
2856 if (bytecode->len + sizeof(*bytecode)
2857 != cmd_ctx->lsm->u.filter.bytecode_len) {
2858 free(bytecode);
f73fabfd 2859 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2860 goto error;
2861 }
2862
2863 ret = cmd_set_filter(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2864 cmd_ctx->lsm->u.filter.channel_name,
2865 cmd_ctx->lsm->u.filter.event_name,
2866 bytecode);
2867 break;
2868 }
6d805429 2869 case LTTNG_DATA_PENDING:
806e2684 2870 {
6d805429 2871 ret = cmd_data_pending(cmd_ctx->session);
806e2684
DG
2872 break;
2873 }
5e16da05 2874 default:
f73fabfd 2875 ret = LTTNG_ERR_UND;
5461b305 2876 break;
fac6795d
DG
2877 }
2878
5461b305 2879error:
5461b305
DG
2880 if (cmd_ctx->llm == NULL) {
2881 DBG("Missing llm structure. Allocating one.");
894be886 2882 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
5461b305
DG
2883 goto setup_error;
2884 }
2885 }
54d01ffb 2886 /* Set return code */
5461b305 2887 cmd_ctx->llm->ret_code = ret;
5461b305 2888setup_error:
b5541356 2889 if (cmd_ctx->session) {
54d01ffb 2890 session_unlock(cmd_ctx->session);
b5541356 2891 }
256a5576
MD
2892 if (need_tracing_session) {
2893 session_unlock_list();
2894 }
54d01ffb 2895init_setup_error:
8028d920 2896 return ret;
fac6795d
DG
2897}
2898
44a5e5eb
DG
2899/*
2900 * Thread managing health check socket.
2901 */
2902static void *thread_manage_health(void *data)
2903{
eb4a2943 2904 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
44a5e5eb
DG
2905 uint32_t revents, nb_fd;
2906 struct lttng_poll_event events;
2907 struct lttcomm_health_msg msg;
2908 struct lttcomm_health_data reply;
2909
2910 DBG("[thread] Manage health check started");
2911
2912 rcu_register_thread();
2913
2914 /* Create unix socket */
2915 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2916 if (sock < 0) {
2917 ERR("Unable to create health check Unix socket");
2918 ret = -1;
2919 goto error;
2920 }
2921
b662582b
DG
2922 /*
2923 * Set the CLOEXEC flag. Return code is useless because either way, the
2924 * show must go on.
2925 */
2926 (void) utils_set_fd_cloexec(sock);
2927
44a5e5eb
DG
2928 ret = lttcomm_listen_unix_sock(sock);
2929 if (ret < 0) {
2930 goto error;
2931 }
2932
2933 /*
2934 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2935 * more will be added to this poll set.
2936 */
2937 ret = create_thread_poll_set(&events, 2);
2938 if (ret < 0) {
2939 goto error;
2940 }
2941
2942 /* Add the application registration socket */
2943 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2944 if (ret < 0) {
2945 goto error;
2946 }
2947
2948 while (1) {
2949 DBG("Health check ready");
2950
2951 nb_fd = LTTNG_POLL_GETNB(&events);
2952
2953 /* Inifinite blocking call, waiting for transmission */
2954restart:
2955 ret = lttng_poll_wait(&events, -1);
2956 if (ret < 0) {
2957 /*
2958 * Restart interrupted system call.
2959 */
2960 if (errno == EINTR) {
2961 goto restart;
2962 }
2963 goto error;
2964 }
2965
2966 for (i = 0; i < nb_fd; i++) {
2967 /* Fetch once the poll data */
2968 revents = LTTNG_POLL_GETEV(&events, i);
2969 pollfd = LTTNG_POLL_GETFD(&events, i);
2970
2971 /* Thread quit pipe has been closed. Killing thread. */
2972 ret = check_thread_quit_pipe(pollfd, revents);
2973 if (ret) {
139ac872
MD
2974 err = 0;
2975 goto exit;
44a5e5eb
DG
2976 }
2977
2978 /* Event on the registration socket */
2979 if (pollfd == sock) {
2980 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2981 ERR("Health socket poll error");
2982 goto error;
2983 }
2984 }
2985 }
2986
2987 new_sock = lttcomm_accept_unix_sock(sock);
2988 if (new_sock < 0) {
2989 goto error;
2990 }
2991
b662582b
DG
2992 /*
2993 * Set the CLOEXEC flag. Return code is useless because either way, the
2994 * show must go on.
2995 */
2996 (void) utils_set_fd_cloexec(new_sock);
2997
44a5e5eb
DG
2998 DBG("Receiving data from client for health...");
2999 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3000 if (ret <= 0) {
3001 DBG("Nothing recv() from client... continuing");
3002 ret = close(new_sock);
3003 if (ret) {
3004 PERROR("close");
3005 }
3006 new_sock = -1;
3007 continue;
3008 }
3009
3010 rcu_thread_online();
3011
3012 switch (msg.component) {
3013 case LTTNG_HEALTH_CMD:
3014 reply.ret_code = health_check_state(&health_thread_cmd);
3015 break;
139ac872
MD
3016 case LTTNG_HEALTH_APP_MANAGE:
3017 reply.ret_code = health_check_state(&health_thread_app_manage);
3018 break;
44a5e5eb
DG
3019 case LTTNG_HEALTH_APP_REG:
3020 reply.ret_code = health_check_state(&health_thread_app_reg);
3021 break;
3022 case LTTNG_HEALTH_KERNEL:
3023 reply.ret_code = health_check_state(&health_thread_kernel);
3024 break;
3025 case LTTNG_HEALTH_CONSUMER:
3026 reply.ret_code = check_consumer_health();
3027 break;
3028 case LTTNG_HEALTH_ALL:
44a5e5eb 3029 reply.ret_code =
139ac872
MD
3030 health_check_state(&health_thread_app_manage) &&
3031 health_check_state(&health_thread_app_reg) &&
3032 health_check_state(&health_thread_cmd) &&
3033 health_check_state(&health_thread_kernel) &&
3034 check_consumer_health();
44a5e5eb
DG
3035 break;
3036 default:
f73fabfd 3037 reply.ret_code = LTTNG_ERR_UND;
44a5e5eb
DG
3038 break;
3039 }
3040
3041 /*
3042 * Flip ret value since 0 is a success and 1 indicates a bad health for
3043 * the client where in the sessiond it is the opposite. Again, this is
3044 * just to make things easier for us poor developer which enjoy a lot
3045 * lazyness.
3046 */
3047 if (reply.ret_code == 0 || reply.ret_code == 1) {
3048 reply.ret_code = !reply.ret_code;
3049 }
3050
3051 DBG2("Health check return value %d", reply.ret_code);
3052
3053 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3054 if (ret < 0) {
3055 ERR("Failed to send health data back to client");
3056 }
3057
3058 /* End of transmission */
3059 ret = close(new_sock);
3060 if (ret) {
3061 PERROR("close");
3062 }
3063 new_sock = -1;
3064 }
3065
139ac872 3066exit:
44a5e5eb 3067error:
139ac872
MD
3068 if (err) {
3069 ERR("Health error occurred in %s", __func__);
3070 }
44a5e5eb
DG
3071 DBG("Health check thread dying");
3072 unlink(health_unix_sock_path);
3073 if (sock >= 0) {
3074 ret = close(sock);
3075 if (ret) {
3076 PERROR("close");
3077 }
3078 }
3079 if (new_sock >= 0) {
3080 ret = close(new_sock);
3081 if (ret) {
3082 PERROR("close");
3083 }
3084 }
3085
3086 lttng_poll_clean(&events);
3087
3088 rcu_unregister_thread();
3089 return NULL;
3090}
3091
1d4b027a 3092/*
d063d709
DG
3093 * This thread manage all clients request using the unix client socket for
3094 * communication.
1d4b027a
DG
3095 */
3096static void *thread_manage_clients(void *data)
3097{
139ac872 3098 int sock = -1, ret, i, pollfd, err = -1;
53a80697 3099 int sock_error;
5eb91c98 3100 uint32_t revents, nb_fd;
273ea72c 3101 struct command_ctx *cmd_ctx = NULL;
5eb91c98 3102 struct lttng_poll_event events;
1d4b027a
DG
3103
3104 DBG("[thread] Manage client started");
3105
8ac94142
CB
3106 testpoint(thread_manage_clients);
3107
f6a9efaa
DG
3108 rcu_register_thread();
3109
44a5e5eb
DG
3110 health_code_update(&health_thread_cmd);
3111
1d4b027a
DG
3112 ret = lttcomm_listen_unix_sock(client_sock);
3113 if (ret < 0) {
35df2755 3114 goto error_listen;
1d4b027a
DG
3115 }
3116
5eb91c98
DG
3117 /*
3118 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3119 * more will be added to this poll set.
3120 */
3121 ret = create_thread_poll_set(&events, 2);
3122 if (ret < 0) {
35df2755 3123 goto error_create_poll;
5eb91c98 3124 }
273ea72c 3125
5eb91c98
DG
3126 /* Add the application registration socket */
3127 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3128 if (ret < 0) {
3129 goto error;
3130 }
273ea72c 3131
bbd973c2
DG
3132 /*
3133 * Notify parent pid that we are ready to accept command for client side.
1d4b027a
DG
3134 */
3135 if (opt_sig_parent) {
8db8d1dc 3136 kill(ppid, SIGUSR1);
1d4b027a
DG
3137 }
3138
8ac94142
CB
3139<