Deprecate enable/disable-consumer
[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"
44a5e5eb 62#include "health.h"
8ac94142 63#include "testpoint.h"
fac6795d 64
ebaeda94
MD
65#define CONSUMERD_FILE "lttng-consumerd"
66
75462a81 67/* Const values */
686204ab 68const char default_home_dir[] = DEFAULT_HOME_DIR;
bbccc3d2 69const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
686204ab
MD
70const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR;
71const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE;
72
fac6795d
DG
73const char *progname;
74const char *opt_tracing_group;
35f90c40 75static const char *opt_pidfile;
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
ae9e45b3
DG
225/*
226 * Socket timeout for receiving and sending in seconds.
227 */
228static int app_socket_timeout;
229
fb09408a 230static
7753dea8 231void setup_consumerd_path(void)
fb09408a 232{
fc7a59ce 233 const char *bin, *libdir;
fb09408a 234
7753dea8
MD
235 /*
236 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
237 * native architecture size consumer if CONFIG_CONSUMER*_PATH
238 * has not been defined.
7753dea8 239 */
ebaeda94 240#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
241 if (!consumerd32_bin[0]) {
242 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
243 }
244 if (!consumerd32_libdir[0]) {
245 consumerd32_libdir = INSTALL_LIB_PATH;
246 }
247#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
248 if (!consumerd64_bin[0]) {
249 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 250 }
ebaeda94
MD
251 if (!consumerd64_libdir[0]) {
252 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
253 }
254#else
255#error "Unknown bitness"
256#endif
257
fb09408a
MD
258 /*
259 * runtime env. var. overrides the build default.
260 */
fc7a59ce
AM
261 bin = getenv("LTTNG_CONSUMERD32_BIN");
262 if (bin) {
263 consumerd32_bin = bin;
7753dea8 264 }
fc7a59ce
AM
265 bin = getenv("LTTNG_CONSUMERD64_BIN");
266 if (bin) {
267 consumerd64_bin = bin;
ebaeda94 268 }
72f579ee 269 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
270 if (libdir) {
271 consumerd32_libdir = libdir;
272 }
72f579ee 273 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
274 if (libdir) {
275 consumerd64_libdir = libdir;
fb09408a
MD
276 }
277}
278
5eb91c98
DG
279/*
280 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
281 */
282static int create_thread_poll_set(struct lttng_poll_event *events,
283 unsigned int size)
284{
285 int ret;
286
287 if (events == NULL || size == 0) {
288 ret = -1;
289 goto error;
290 }
291
292 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
293 if (ret < 0) {
294 goto error;
295 }
296
297 /* Add quit pipe */
298 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
299 if (ret < 0) {
300 goto error;
301 }
302
303 return 0;
304
305error:
306 return ret;
307}
308
309/*
310 * Check if the thread quit pipe was triggered.
311 *
312 * Return 1 if it was triggered else 0;
313 */
314static int check_thread_quit_pipe(int fd, uint32_t events)
315{
316 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
317 return 1;
318 }
319
320 return 0;
321}
322
0fdd1e2c
DG
323/*
324 * Return group ID of the tracing group or -1 if not found.
325 */
996b65c8
MD
326static gid_t allowed_group(void)
327{
328 struct group *grp;
329
274e143b
MD
330 if (opt_tracing_group) {
331 grp = getgrnam(opt_tracing_group);
332 } else {
333 grp = getgrnam(default_tracing_group);
334 }
996b65c8
MD
335 if (!grp) {
336 return -1;
337 } else {
338 return grp->gr_gid;
339 }
340}
341
273ea72c 342/*
5eb91c98 343 * Init thread quit pipe.
273ea72c
DG
344 *
345 * Return -1 on error or 0 if all pipes are created.
346 */
347static int init_thread_quit_pipe(void)
348{
730389d9 349 int ret, i;
273ea72c 350
730389d9 351 ret = pipe(thread_quit_pipe);
273ea72c 352 if (ret < 0) {
730389d9 353 PERROR("thread quit pipe");
273ea72c
DG
354 goto error;
355 }
356
730389d9
DG
357 for (i = 0; i < 2; i++) {
358 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
359 if (ret < 0) {
360 PERROR("fcntl");
361 goto error;
362 }
363 }
364
273ea72c
DG
365error:
366 return ret;
367}
368
099e26bd
DG
369/*
370 * Stop all threads by closing the thread quit pipe.
371 */
cf3af59e
MD
372static void stop_threads(void)
373{
5eb91c98
DG
374 int ret;
375
cf3af59e
MD
376 /* Stopping all threads */
377 DBG("Terminating all threads");
54d01ffb 378 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
379 if (ret < 0) {
380 ERR("write error on thread quit pipe");
381 }
382
099e26bd 383 /* Dispatch thread */
26c9d55e 384 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 385 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
386}
387
fac6795d 388/*
d063d709 389 * Cleanup the daemon
fac6795d 390 */
cf3af59e 391static void cleanup(void)
fac6795d 392{
ef599319 393 int ret;
6620da75 394 char *cmd = NULL;
af9737e9 395 struct ltt_session *sess, *stmp;
fac6795d 396
1d4b027a 397 DBG("Cleaning up");
e07ae692 398
2f77fc4b
DG
399 /* First thing first, stop all threads */
400 utils_close_pipe(thread_quit_pipe);
401
35f90c40
DG
402 /*
403 * If opt_pidfile is undefined, the default file will be wiped when
404 * removing the rundir.
405 */
406 if (opt_pidfile) {
407 ret = remove(opt_pidfile);
408 if (ret < 0) {
409 PERROR("remove pidfile %s", opt_pidfile);
410 }
411 }
412
67e40797
DG
413 DBG("Removing %s directory", rundir);
414 ret = asprintf(&cmd, "rm -rf %s", rundir);
415 if (ret < 0) {
416 ERR("asprintf failed. Something is really wrong!");
417 }
5461b305 418
67e40797
DG
419 /* Remove lttng run directory */
420 ret = system(cmd);
421 if (ret < 0) {
422 ERR("Unable to clean %s", rundir);
1d4b027a 423 }
67e40797 424 free(cmd);
bd69add2 425 free(rundir);
5461b305 426
99bab54f 427 DBG("Cleaning up all sessions");
fac6795d 428
b5541356 429 /* Destroy session list mutex */
273ea72c
DG
430 if (session_list_ptr != NULL) {
431 pthread_mutex_destroy(&session_list_ptr->lock);
432
433 /* Cleanup ALL session */
54d01ffb
DG
434 cds_list_for_each_entry_safe(sess, stmp,
435 &session_list_ptr->head, list) {
2f77fc4b 436 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
437 }
438 }
439
099e26bd 440 DBG("Closing all UST sockets");
56fff090 441 ust_app_clean_list();
099e26bd 442
4fba7219
DG
443 if (is_root && !opt_no_kernel) {
444 DBG2("Closing kernel fd");
a4b35e07 445 if (kernel_tracer_fd >= 0) {
76d7553f
MD
446 ret = close(kernel_tracer_fd);
447 if (ret) {
448 PERROR("close");
449 }
a4b35e07 450 }
2f50c8a3 451 DBG("Unloading kernel modules");
096102bd 452 modprobe_remove_lttng_all();
2f50c8a3 453 }
2f77fc4b 454
421cb601 455 /* <fun> */
f56a39af 456 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
457 "Matthew, BEET driven development works!%c[%dm",
458 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
459 /* </fun> */
fac6795d
DG
460}
461
e065084a 462/*
d063d709 463 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 464 *
d063d709 465 * Return lttcomm error code.
e065084a
DG
466 */
467static int send_unix_sock(int sock, void *buf, size_t len)
468{
469 /* Check valid length */
c617c0c6 470 if (len == 0) {
e065084a
DG
471 return -1;
472 }
473
474 return lttcomm_send_unix_sock(sock, buf, len);
475}
476
5461b305 477/*
d063d709 478 * Free memory of a command context structure.
5461b305 479 */
a2fb29a5 480static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 481{
a2fb29a5
DG
482 DBG("Clean command context structure");
483 if (*cmd_ctx) {
484 if ((*cmd_ctx)->llm) {
485 free((*cmd_ctx)->llm);
5461b305 486 }
a2fb29a5
DG
487 if ((*cmd_ctx)->lsm) {
488 free((*cmd_ctx)->lsm);
5461b305 489 }
a2fb29a5
DG
490 free(*cmd_ctx);
491 *cmd_ctx = NULL;
5461b305
DG
492 }
493}
494
fac6795d 495/*
0fdd1e2c 496 * Notify UST applications using the shm mmap futex.
fac6795d 497 */
0fdd1e2c 498static int notify_ust_apps(int active)
fac6795d 499{
0fdd1e2c 500 char *wait_shm_mmap;
fac6795d 501
0fdd1e2c 502 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 503
0fdd1e2c
DG
504 /* See shm.c for this call implying mmap, shm and futex calls */
505 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
506 if (wait_shm_mmap == NULL) {
fac6795d
DG
507 goto error;
508 }
509
0fdd1e2c
DG
510 /* Wake waiting process */
511 futex_wait_update((int32_t *) wait_shm_mmap, active);
512
513 /* Apps notified successfully */
514 return 0;
fac6795d
DG
515
516error:
0fdd1e2c 517 return -1;
fac6795d
DG
518}
519
e065084a 520/*
d063d709
DG
521 * Setup the outgoing data buffer for the response (llm) by allocating the
522 * right amount of memory and copying the original information from the lsm
523 * structure.
ca95a216 524 *
d063d709 525 * Return total size of the buffer pointed by buf.
ca95a216 526 */
5461b305 527static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 528{
f3ed775e 529 int ret, buf_size;
ca95a216 530
f3ed775e 531 buf_size = size;
5461b305 532
ba7f0ae5 533 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 534 if (cmd_ctx->llm == NULL) {
76d7553f 535 PERROR("zmalloc");
5461b305 536 ret = -ENOMEM;
ca95a216
DG
537 goto error;
538 }
539
5461b305
DG
540 /* Copy common data */
541 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 542 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 543
5461b305
DG
544 cmd_ctx->llm->data_size = size;
545 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
546
ca95a216
DG
547 return buf_size;
548
549error:
550 return ret;
551}
552
7a485870 553/*
5eb91c98 554 * Update the kernel poll set of all channel fd available over all tracing
d063d709 555 * session. Add the wakeup pipe at the end of the set.
7a485870 556 */
5eb91c98 557static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 558{
5eb91c98 559 int ret;
7a485870
DG
560 struct ltt_session *session;
561 struct ltt_kernel_channel *channel;
562
5eb91c98 563 DBG("Updating kernel poll set");
7a485870 564
54d01ffb 565 session_lock_list();
b5541356 566 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 567 session_lock(session);
7a485870 568 if (session->kernel_session == NULL) {
54d01ffb 569 session_unlock(session);
7a485870
DG
570 continue;
571 }
7a485870 572
54d01ffb
DG
573 cds_list_for_each_entry(channel,
574 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
575 /* Add channel fd to the kernel poll set */
576 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
577 if (ret < 0) {
54d01ffb 578 session_unlock(session);
5eb91c98
DG
579 goto error;
580 }
581 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 582 }
54d01ffb 583 session_unlock(session);
7a485870 584 }
54d01ffb 585 session_unlock_list();
7a485870 586
5eb91c98 587 return 0;
7a485870
DG
588
589error:
54d01ffb 590 session_unlock_list();
7a485870
DG
591 return -1;
592}
593
594/*
54d01ffb 595 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 596 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 597 *
d063d709 598 * Useful for CPU hotplug feature.
7a485870 599 */
2bdd86d4 600static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
601{
602 int ret = 0;
603 struct ltt_session *session;
173af62f 604 struct ltt_kernel_session *ksess;
7a485870
DG
605 struct ltt_kernel_channel *channel;
606
607 DBG("Updating kernel streams for channel fd %d", fd);
608
54d01ffb 609 session_lock_list();
b5541356 610 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 611 session_lock(session);
7a485870 612 if (session->kernel_session == NULL) {
54d01ffb 613 session_unlock(session);
7a485870
DG
614 continue;
615 }
173af62f 616 ksess = session->kernel_session;
d9800920 617
173af62f 618 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
7a485870
DG
619 if (channel->fd == fd) {
620 DBG("Channel found, updating kernel streams");
621 ret = kernel_open_channel_stream(channel);
622 if (ret < 0) {
b3c750d2 623 goto error;
7a485870 624 }
d9800920 625
7a485870 626 /*
5eb91c98
DG
627 * Have we already sent fds to the consumer? If yes, it means
628 * that tracing is started so it is safe to send our updated
629 * stream fds.
7a485870 630 */
173af62f
DG
631 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
632 struct lttng_ht_iter iter;
633 struct consumer_socket *socket;
634
e7fe706f 635 rcu_read_lock();
173af62f
DG
636 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
637 &iter.iter, socket, node.node) {
638 /* Code flow error */
639 assert(socket->fd >= 0);
640
641 pthread_mutex_lock(socket->lock);
f50f23d9 642 ret = kernel_consumer_send_channel_stream(socket,
173af62f
DG
643 channel, ksess);
644 pthread_mutex_unlock(socket->lock);
645 if (ret < 0) {
e7fe706f 646 rcu_read_unlock();
173af62f
DG
647 goto error;
648 }
7a485870 649 }
e7fe706f 650 rcu_read_unlock();
7a485870 651 }
b3c750d2 652 goto error;
7a485870
DG
653 }
654 }
54d01ffb 655 session_unlock(session);
7a485870 656 }
54d01ffb 657 session_unlock_list();
b3c750d2 658 return ret;
7a485870 659
b3c750d2 660error:
54d01ffb
DG
661 session_unlock(session);
662 session_unlock_list();
7a485870
DG
663 return ret;
664}
665
487cf67c 666/*
ffe60014
DG
667 * For each tracing session, update newly registered apps. The session list
668 * lock MUST be acquired before calling this.
487cf67c
DG
669 */
670static void update_ust_app(int app_sock)
671{
672 struct ltt_session *sess, *stmp;
673
674 /* For all tracing session(s) */
675 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
4ee14516 676 session_lock(sess);
421cb601
DG
677 if (sess->ust_session) {
678 ust_app_global_update(sess->ust_session, app_sock);
679 }
4ee14516 680 session_unlock(sess);
487cf67c
DG
681 }
682}
683
7a485870 684/*
d063d709 685 * This thread manage event coming from the kernel.
7a485870 686 *
d063d709
DG
687 * Features supported in this thread:
688 * -) CPU Hotplug
7a485870
DG
689 */
690static void *thread_manage_kernel(void *data)
691{
139ac872 692 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 693 uint32_t revents, nb_fd;
7a485870 694 char tmp;
5eb91c98 695 struct lttng_poll_event events;
7a485870 696
6993eeb3 697 DBG("[thread] Thread manage kernel started");
7a485870 698
927ca06a
DG
699 health_register(HEALTH_TYPE_KERNEL);
700
d5d63bf1
DG
701 /*
702 * This first step of the while is to clean this structure which could free
703 * non NULL pointers so zero it before the loop.
704 */
705 memset(&events, 0, sizeof(events));
706
6993eeb3
CB
707 if (testpoint(thread_manage_kernel)) {
708 goto error_testpoint;
709 }
8ac94142 710
840cb59c 711 health_code_update();
44a5e5eb 712
6993eeb3 713 if (testpoint(thread_manage_kernel_before_loop)) {
d21b0d71 714 goto error_testpoint;
6993eeb3
CB
715 }
716
7a485870 717 while (1) {
840cb59c 718 health_code_update();
44a5e5eb 719
7a485870 720 if (update_poll_flag == 1) {
d21b0d71
DG
721 /* Clean events object. We are about to populate it again. */
722 lttng_poll_clean(&events);
723
724 ret = create_thread_poll_set(&events, 2);
725 if (ret < 0) {
726 goto error_poll_create;
727 }
728
729 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
730 if (ret < 0) {
731 goto error;
732 }
5f822d0a 733
d21b0d71 734 /* This will add the available kernel channel if any. */
5eb91c98
DG
735 ret = update_kernel_poll(&events);
736 if (ret < 0) {
7a485870
DG
737 goto error;
738 }
739 update_poll_flag = 0;
740 }
741
d21b0d71 742 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
7a485870
DG
743
744 /* Poll infinite value of time */
88f2b785 745 restart:
a78af745 746 health_poll_entry();
5eb91c98 747 ret = lttng_poll_wait(&events, -1);
a78af745 748 health_poll_exit();
7a485870 749 if (ret < 0) {
88f2b785
MD
750 /*
751 * Restart interrupted system call.
752 */
753 if (errno == EINTR) {
754 goto restart;
755 }
7a485870
DG
756 goto error;
757 } else if (ret == 0) {
758 /* Should not happen since timeout is infinite */
85611738
DG
759 ERR("Return value of poll is 0 with an infinite timeout.\n"
760 "This should not have happened! Continuing...");
7a485870
DG
761 continue;
762 }
763
0d9c5d77
DG
764 nb_fd = ret;
765
5eb91c98
DG
766 for (i = 0; i < nb_fd; i++) {
767 /* Fetch once the poll data */
768 revents = LTTNG_POLL_GETEV(&events, i);
769 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 770
840cb59c 771 health_code_update();
44a5e5eb 772
5eb91c98
DG
773 /* Thread quit pipe has been closed. Killing thread. */
774 ret = check_thread_quit_pipe(pollfd, revents);
775 if (ret) {
139ac872
MD
776 err = 0;
777 goto exit;
5eb91c98 778 }
7a485870 779
5eb91c98
DG
780 /* Check for data on kernel pipe */
781 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
f921c78f
DG
782 do {
783 ret = read(kernel_poll_pipe[0], &tmp, 1);
784 } while (ret < 0 && errno == EINTR);
785 /*
786 * Ret value is useless here, if this pipe gets any actions an
787 * update is required anyway.
788 */
5eb91c98
DG
789 update_poll_flag = 1;
790 continue;
791 } else {
792 /*
793 * New CPU detected by the kernel. Adding kernel stream to
794 * kernel session and updating the kernel consumer
795 */
796 if (revents & LPOLLIN) {
2bdd86d4 797 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
798 if (ret < 0) {
799 continue;
800 }
801 break;
802 /*
803 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
804 * and unregister kernel stream at this point.
805 */
7a485870 806 }
7a485870
DG
807 }
808 }
809 }
810
139ac872 811exit:
7a485870 812error:
5eb91c98 813 lttng_poll_clean(&events);
76d7553f 814error_poll_create:
6993eeb3 815error_testpoint:
6620da75
DG
816 utils_close_pipe(kernel_poll_pipe);
817 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 818 if (err) {
840cb59c 819 health_error();
139ac872 820 ERR("Health error occurred in %s", __func__);
6620da75
DG
821 WARN("Kernel thread died unexpectedly. "
822 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 823 }
927ca06a 824 health_unregister();
76d7553f 825 DBG("Kernel thread dying");
7a485870
DG
826 return NULL;
827}
828
a23ec3a7
DG
829/*
830 * Signal pthread condition of the consumer data that the thread.
831 */
832static void signal_consumer_condition(struct consumer_data *data, int state)
833{
834 pthread_mutex_lock(&data->cond_mutex);
835
836 /*
837 * The state is set before signaling. It can be any value, it's the waiter
838 * job to correctly interpret this condition variable associated to the
839 * consumer pthread_cond.
840 *
841 * A value of 0 means that the corresponding thread of the consumer data
842 * was not started. 1 indicates that the thread has started and is ready
843 * for action. A negative value means that there was an error during the
844 * thread bootstrap.
845 */
846 data->consumer_thread_is_ready = state;
847 (void) pthread_cond_signal(&data->cond);
848
849 pthread_mutex_unlock(&data->cond_mutex);
850}
851
1d4b027a 852/*
3bd1e081 853 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 854 */
3bd1e081 855static void *thread_manage_consumer(void *data)
1d4b027a 856{
139ac872 857 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98 858 uint32_t revents, nb_fd;
1d4b027a 859 enum lttcomm_return_code code;
5eb91c98 860 struct lttng_poll_event events;
3bd1e081 861 struct consumer_data *consumer_data = data;
1d4b027a 862
3bd1e081 863 DBG("[thread] Manage consumer started");
1d4b027a 864
927ca06a
DG
865 health_register(HEALTH_TYPE_CONSUMER);
866
855060f8 867 health_code_update();
9449cc75 868
5eb91c98
DG
869 /*
870 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
871 * Nothing more will be added to this poll set.
872 */
873 ret = create_thread_poll_set(&events, 2);
874 if (ret < 0) {
76d7553f 875 goto error_poll;
5eb91c98 876 }
273ea72c 877
edb8b045
DG
878 /*
879 * The error socket here is already in a listening state which was done
880 * just before spawning this thread to avoid a race between the consumer
881 * daemon exec trying to connect and the listen() call.
882 */
3bd1e081 883 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
884 if (ret < 0) {
885 goto error;
886 }
887
840cb59c 888 health_code_update();
44a5e5eb 889
273ea72c 890 /* Inifinite blocking call, waiting for transmission */
88f2b785 891restart:
a78af745 892 health_poll_entry();
8ac94142 893
6993eeb3
CB
894 if (testpoint(thread_manage_consumer)) {
895 goto error;
896 }
8ac94142 897
5eb91c98 898 ret = lttng_poll_wait(&events, -1);
a78af745 899 health_poll_exit();
273ea72c 900 if (ret < 0) {
88f2b785
MD
901 /*
902 * Restart interrupted system call.
903 */
904 if (errno == EINTR) {
905 goto restart;
906 }
273ea72c
DG
907 goto error;
908 }
909
0d9c5d77
DG
910 nb_fd = ret;
911
5eb91c98
DG
912 for (i = 0; i < nb_fd; i++) {
913 /* Fetch once the poll data */
914 revents = LTTNG_POLL_GETEV(&events, i);
915 pollfd = LTTNG_POLL_GETFD(&events, i);
916
840cb59c 917 health_code_update();
44a5e5eb 918
5eb91c98
DG
919 /* Thread quit pipe has been closed. Killing thread. */
920 ret = check_thread_quit_pipe(pollfd, revents);
921 if (ret) {
139ac872
MD
922 err = 0;
923 goto exit;
5eb91c98
DG
924 }
925
926 /* Event on the registration socket */
3bd1e081 927 if (pollfd == consumer_data->err_sock) {
5eb91c98 928 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 929 ERR("consumer err socket poll error");
5eb91c98
DG
930 goto error;
931 }
932 }
273ea72c
DG
933 }
934
3bd1e081 935 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
936 if (sock < 0) {
937 goto error;
938 }
939
b662582b
DG
940 /*
941 * Set the CLOEXEC flag. Return code is useless because either way, the
942 * show must go on.
943 */
944 (void) utils_set_fd_cloexec(sock);
945
840cb59c 946 health_code_update();
44a5e5eb 947
3bd1e081 948 DBG2("Receiving code from consumer err_sock");
ee0b0061 949
712ea556 950 /* Getting status code from kconsumerd */
54d01ffb
DG
951 ret = lttcomm_recv_unix_sock(sock, &code,
952 sizeof(enum lttcomm_return_code));
1d4b027a
DG
953 if (ret <= 0) {
954 goto error;
955 }
956
840cb59c 957 health_code_update();
44a5e5eb 958
f73fabfd 959 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
3bd1e081
MD
960 consumer_data->cmd_sock =
961 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
962 if (consumer_data->cmd_sock < 0) {
a23ec3a7
DG
963 /* On error, signal condition and quit. */
964 signal_consumer_condition(consumer_data, -1);
48842b30 965 PERROR("consumer connect");
1d4b027a
DG
966 goto error;
967 }
a23ec3a7
DG
968 signal_consumer_condition(consumer_data, 1);
969 DBG("Consumer command socket ready");
1d4b027a 970 } else {
3bd1e081 971 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
972 lttcomm_get_readable_code(-code));
973 goto error;
974 }
975
54d01ffb 976 /* Remove the kconsumerd error sock since we've established a connexion */
3bd1e081 977 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 978 if (ret < 0) {
72079cae
DG
979 goto error;
980 }
981
5eb91c98
DG
982 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
983 if (ret < 0) {
72079cae 984 goto error;
5eb91c98
DG
985 }
986
840cb59c 987 health_code_update();
44a5e5eb 988
5eb91c98 989 /* Inifinite blocking call, waiting for transmission */
88f2b785 990restart_poll:
a78af745 991 health_poll_entry();
5eb91c98 992 ret = lttng_poll_wait(&events, -1);
a78af745 993 health_poll_exit();
5eb91c98 994 if (ret < 0) {
88f2b785
MD
995 /*
996 * Restart interrupted system call.
997 */
998 if (errno == EINTR) {
999 goto restart_poll;
1000 }
72079cae
DG
1001 goto error;
1002 }
1003
0d9c5d77
DG
1004 nb_fd = ret;
1005
5eb91c98
DG
1006 for (i = 0; i < nb_fd; i++) {
1007 /* Fetch once the poll data */
1008 revents = LTTNG_POLL_GETEV(&events, i);
1009 pollfd = LTTNG_POLL_GETFD(&events, i);
1010
840cb59c 1011 health_code_update();
44a5e5eb 1012
5eb91c98
DG
1013 /* Thread quit pipe has been closed. Killing thread. */
1014 ret = check_thread_quit_pipe(pollfd, revents);
1015 if (ret) {
139ac872
MD
1016 err = 0;
1017 goto exit;
5eb91c98
DG
1018 }
1019
1020 /* Event on the kconsumerd socket */
1021 if (pollfd == sock) {
1022 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1023 ERR("consumer err socket second poll error");
5eb91c98
DG
1024 goto error;
1025 }
1026 }
1027 }
1028
840cb59c 1029 health_code_update();
44a5e5eb 1030
712ea556 1031 /* Wait for any kconsumerd error */
54d01ffb
DG
1032 ret = lttcomm_recv_unix_sock(sock, &code,
1033 sizeof(enum lttcomm_return_code));
712ea556 1034 if (ret <= 0) {
3bd1e081 1035 ERR("consumer closed the command socket");
712ea556 1036 goto error;
6f61d021 1037 }
1d4b027a 1038
3bd1e081 1039 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
712ea556 1040
139ac872 1041exit:
1d4b027a 1042error:
5c827ce0
DG
1043 /* Immediately set the consumerd state to stopped */
1044 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1045 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1046 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1047 consumer_data->type == LTTNG_CONSUMER32_UST) {
1048 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1049 } else {
1050 /* Code flow error... */
1051 assert(0);
1052 }
1053
76d7553f
MD
1054 if (consumer_data->err_sock >= 0) {
1055 ret = close(consumer_data->err_sock);
1056 if (ret) {
1057 PERROR("close");
1058 }
1059 }
1060 if (consumer_data->cmd_sock >= 0) {
1061 ret = close(consumer_data->cmd_sock);
1062 if (ret) {
1063 PERROR("close");
1064 }
1065 }
1066 if (sock >= 0) {
1067 ret = close(sock);
1068 if (ret) {
1069 PERROR("close");
1070 }
1071 }
273ea72c 1072
3bd1e081
MD
1073 unlink(consumer_data->err_unix_sock_path);
1074 unlink(consumer_data->cmd_unix_sock_path);
1075 consumer_data->pid = 0;
1d4b027a 1076
5eb91c98 1077 lttng_poll_clean(&events);
76d7553f 1078error_poll:
139ac872 1079 if (err) {
840cb59c 1080 health_error();
139ac872
MD
1081 ERR("Health error occurred in %s", __func__);
1082 }
927ca06a 1083 health_unregister();
76d7553f 1084 DBG("consumer thread cleanup completed");
0177d773 1085
5eb91c98 1086 return NULL;
099e26bd
DG
1087}
1088
099e26bd
DG
1089/*
1090 * This thread manage application communication.
1d4b027a
DG
1091 */
1092static void *thread_manage_apps(void *data)
099e26bd 1093{
139ac872 1094 int i, ret, pollfd, err = -1;
5eb91c98 1095 uint32_t revents, nb_fd;
099e26bd 1096 struct ust_command ust_cmd;
5eb91c98 1097 struct lttng_poll_event events;
099e26bd
DG
1098
1099 DBG("[thread] Manage application started");
1100
f6a9efaa
DG
1101 rcu_register_thread();
1102 rcu_thread_online();
1103
927ca06a
DG
1104 health_register(HEALTH_TYPE_APP_MANAGE);
1105
6993eeb3
CB
1106 if (testpoint(thread_manage_apps)) {
1107 goto error_testpoint;
1108 }
1109
840cb59c 1110 health_code_update();
44a5e5eb 1111
5eb91c98
DG
1112 ret = create_thread_poll_set(&events, 2);
1113 if (ret < 0) {
76d7553f 1114 goto error_poll_create;
5eb91c98 1115 }
099e26bd 1116
5eb91c98
DG
1117 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1118 if (ret < 0) {
1119 goto error;
1120 }
099e26bd 1121
6993eeb3
CB
1122 if (testpoint(thread_manage_apps_before_loop)) {
1123 goto error;
1124 }
8ac94142 1125
840cb59c 1126 health_code_update();
44a5e5eb 1127
5eb91c98 1128 while (1) {
d21b0d71 1129 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
099e26bd
DG
1130
1131 /* Inifinite blocking call, waiting for transmission */
88f2b785 1132 restart:
a78af745 1133 health_poll_entry();
5eb91c98 1134 ret = lttng_poll_wait(&events, -1);
a78af745 1135 health_poll_exit();
099e26bd 1136 if (ret < 0) {
88f2b785
MD
1137 /*
1138 * Restart interrupted system call.
1139 */
1140 if (errno == EINTR) {
1141 goto restart;
1142 }
099e26bd
DG
1143 goto error;
1144 }
1145
0d9c5d77
DG
1146 nb_fd = ret;
1147
5eb91c98
DG
1148 for (i = 0; i < nb_fd; i++) {
1149 /* Fetch once the poll data */
1150 revents = LTTNG_POLL_GETEV(&events, i);
1151 pollfd = LTTNG_POLL_GETFD(&events, i);
1152
840cb59c 1153 health_code_update();
44a5e5eb 1154
5eb91c98
DG
1155 /* Thread quit pipe has been closed. Killing thread. */
1156 ret = check_thread_quit_pipe(pollfd, revents);
1157 if (ret) {
139ac872
MD
1158 err = 0;
1159 goto exit;
5eb91c98 1160 }
099e26bd 1161
5eb91c98
DG
1162 /* Inspect the apps cmd pipe */
1163 if (pollfd == apps_cmd_pipe[0]) {
1164 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1165 ERR("Apps command pipe error");
0177d773 1166 goto error;
5eb91c98
DG
1167 } else if (revents & LPOLLIN) {
1168 /* Empty pipe */
f921c78f
DG
1169 do {
1170 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1171 } while (ret < 0 && errno == EINTR);
5eb91c98 1172 if (ret < 0 || ret < sizeof(ust_cmd)) {
76d7553f 1173 PERROR("read apps cmd pipe");
5eb91c98
DG
1174 goto error;
1175 }
099e26bd 1176
840cb59c 1177 health_code_update();
44a5e5eb 1178
ffe60014
DG
1179 /*
1180 * @session_lock
1181 * Lock the global session list so from the register up to
1182 * the registration done message, no thread can see the
1183 * application and change its state.
1184 */
1185 session_lock_list();
1186
5eb91c98 1187 /* Register applicaton to the session daemon */
56fff090 1188 ret = ust_app_register(&ust_cmd.reg_msg,
54d01ffb 1189 ust_cmd.sock);
88ff5b7f 1190 if (ret == -ENOMEM) {
ffe60014 1191 session_unlock_list();
5eb91c98 1192 goto error;
88ff5b7f 1193 } else if (ret < 0) {
ffe60014 1194 session_unlock_list();
88ff5b7f 1195 break;
5eb91c98
DG
1196 }
1197
840cb59c 1198 health_code_update();
44a5e5eb 1199
acc7b41b 1200 /*
e0c7ec2b 1201 * Validate UST version compatibility.
acc7b41b 1202 */
e0c7ec2b
DG
1203 ret = ust_app_validate_version(ust_cmd.sock);
1204 if (ret >= 0) {
1205 /*
1206 * Add channel(s) and event(s) to newly registered apps
1207 * from lttng global UST domain.
1208 */
1209 update_ust_app(ust_cmd.sock);
1210 }
acc7b41b 1211
840cb59c 1212 health_code_update();
44a5e5eb 1213
f2ca2e25 1214 ret = ust_app_register_done(ust_cmd.sock);
5eb91c98
DG
1215 if (ret < 0) {
1216 /*
1217 * If the registration is not possible, we simply
1218 * unregister the apps and continue
1219 */
56fff090 1220 ust_app_unregister(ust_cmd.sock);
5eb91c98
DG
1221 } else {
1222 /*
622c9ecf
DG
1223 * We only monitor the error events of the socket. This
1224 * thread does not handle any incoming data from UST
1225 * (POLLIN).
5eb91c98 1226 */
622c9ecf
DG
1227 ret = lttng_poll_add(&events, ust_cmd.sock,
1228 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
5eb91c98 1229 if (ret < 0) {
ffe60014 1230 session_unlock_list();
5eb91c98
DG
1231 goto error;
1232 }
1233
ae9e45b3
DG
1234 /* Set socket timeout for both receiving and ending */
1235 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1236 app_socket_timeout);
1237 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1238 app_socket_timeout);
1239
54d01ffb
DG
1240 DBG("Apps with sock %d added to poll set",
1241 ust_cmd.sock);
5eb91c98 1242 }
ffe60014 1243 session_unlock_list();
487cf67c 1244
840cb59c 1245 health_code_update();
44a5e5eb 1246
5eb91c98 1247 break;
0177d773 1248 }
5eb91c98
DG
1249 } else {
1250 /*
54d01ffb
DG
1251 * At this point, we know that a registered application made
1252 * the event at poll_wait.
5eb91c98
DG
1253 */
1254 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1255 /* Removing from the poll set */
1256 ret = lttng_poll_del(&events, pollfd);
1257 if (ret < 0) {
1258 goto error;
1259 }
099e26bd 1260
b9d9b220 1261 /* Socket closed on remote end. */
56fff090 1262 ust_app_unregister(pollfd);
5eb91c98
DG
1263 break;
1264 }
099e26bd 1265 }
44a5e5eb 1266
840cb59c 1267 health_code_update();
099e26bd 1268 }
099e26bd
DG
1269 }
1270
139ac872 1271exit:
099e26bd 1272error:
5eb91c98 1273 lttng_poll_clean(&events);
76d7553f 1274error_poll_create:
6993eeb3 1275error_testpoint:
6620da75
DG
1276 utils_close_pipe(apps_cmd_pipe);
1277 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1278
1279 /*
1280 * We don't clean the UST app hash table here since already registered
1281 * applications can still be controlled so let them be until the session
1282 * daemon dies or the applications stop.
1283 */
1284
139ac872 1285 if (err) {
840cb59c 1286 health_error();
139ac872
MD
1287 ERR("Health error occurred in %s", __func__);
1288 }
927ca06a 1289 health_unregister();
76d7553f 1290 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1291 rcu_thread_offline();
1292 rcu_unregister_thread();
099e26bd
DG
1293 return NULL;
1294}
1295
1296/*
1297 * Dispatch request from the registration threads to the application
1298 * communication thread.
1299 */
1300static void *thread_dispatch_ust_registration(void *data)
1301{
1302 int ret;
1303 struct cds_wfq_node *node;
1304 struct ust_command *ust_cmd = NULL;
1305
1306 DBG("[thread] Dispatch UST command started");
1307
26c9d55e 1308 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
099e26bd
DG
1309 /* Atomically prepare the queue futex */
1310 futex_nto1_prepare(&ust_cmd_queue.futex);
1311
1312 do {
1313 /* Dequeue command for registration */
1314 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1315 if (node == NULL) {
00a17c97 1316 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1317 /* Continue thread execution */
1318 break;
1319 }
1320
1321 ust_cmd = caa_container_of(node, struct ust_command, node);
1322
2f50c8a3
DG
1323 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1324 " gid:%d sock:%d name:%s (version %d.%d)",
1325 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1326 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1327 ust_cmd->sock, ust_cmd->reg_msg.name,
1328 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
099e26bd
DG
1329 /*
1330 * Inform apps thread of the new application registration. This
1331 * call is blocking so we can be assured that the data will be read
1332 * at some point in time or wait to the end of the world :)
1333 */
6620da75 1334 if (apps_cmd_pipe[1] >= 0) {
fbb3b395
DG
1335 do {
1336 ret = write(apps_cmd_pipe[1], ust_cmd,
1337 sizeof(struct ust_command));
1338 } while (ret < 0 && errno == EINTR);
4cec016f 1339 if (ret < 0 || ret != sizeof(struct ust_command)) {
6620da75
DG
1340 PERROR("write apps cmd pipe");
1341 if (errno == EBADF) {
1342 /*
1343 * We can't inform the application thread to process
1344 * registration. We will exit or else application
1345 * registration will not occur and tracing will never
1346 * start.
1347 */
1348 goto error;
1349 }
1350 }
1351 } else {
1352 /* Application manager thread is not available. */
1353 ret = close(ust_cmd->sock);
1354 if (ret < 0) {
1355 PERROR("close ust_cmd sock");
099e26bd
DG
1356 }
1357 }
1358 free(ust_cmd);
1359 } while (node != NULL);
1360
1361 /* Futex wait on queue. Blocking call on futex() */
1362 futex_nto1_wait(&ust_cmd_queue.futex);
1363 }
1364
1365error:
1366 DBG("Dispatch thread dying");
1367 return NULL;
1368}
1369
1370/*
1371 * This thread manage application registration.
1372 */
1373static void *thread_registration_apps(void *data)
1d4b027a 1374{
139ac872 1375 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
1376 uint32_t revents, nb_fd;
1377 struct lttng_poll_event events;
099e26bd
DG
1378 /*
1379 * Get allocated in this thread, enqueued to a global queue, dequeued and
1380 * freed in the manage apps thread.
1381 */
1382 struct ust_command *ust_cmd = NULL;
1d4b027a 1383
099e26bd 1384 DBG("[thread] Manage application registration started");
1d4b027a 1385
927ca06a
DG
1386 health_register(HEALTH_TYPE_APP_REG);
1387
6993eeb3
CB
1388 if (testpoint(thread_registration_apps)) {
1389 goto error_testpoint;
1390 }
8ac94142 1391
1d4b027a
DG
1392 ret = lttcomm_listen_unix_sock(apps_sock);
1393 if (ret < 0) {
76d7553f 1394 goto error_listen;
1d4b027a
DG
1395 }
1396
5eb91c98
DG
1397 /*
1398 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1399 * more will be added to this poll set.
1400 */
1401 ret = create_thread_poll_set(&events, 2);
1402 if (ret < 0) {
76d7553f 1403 goto error_create_poll;
5eb91c98 1404 }
273ea72c 1405
5eb91c98
DG
1406 /* Add the application registration socket */
1407 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1408 if (ret < 0) {
76d7553f 1409 goto error_poll_add;
5eb91c98 1410 }
273ea72c 1411
1d4b027a 1412 /* Notify all applications to register */
0fdd1e2c
DG
1413 ret = notify_ust_apps(1);
1414 if (ret < 0) {
1415 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
1416 "Execution continues but there might be problem for already\n"
1417 "running applications that wishes to register.");
0fdd1e2c 1418 }
1d4b027a
DG
1419
1420 while (1) {
1421 DBG("Accepting application registration");
273ea72c
DG
1422
1423 /* Inifinite blocking call, waiting for transmission */
88f2b785 1424 restart:
a78af745 1425 health_poll_entry();
5eb91c98 1426 ret = lttng_poll_wait(&events, -1);
a78af745 1427 health_poll_exit();
273ea72c 1428 if (ret < 0) {
88f2b785
MD
1429 /*
1430 * Restart interrupted system call.
1431 */
1432 if (errno == EINTR) {
1433 goto restart;
1434 }
273ea72c
DG
1435 goto error;
1436 }
1437
0d9c5d77
DG
1438 nb_fd = ret;
1439
5eb91c98 1440 for (i = 0; i < nb_fd; i++) {
840cb59c 1441 health_code_update();
139ac872 1442
5eb91c98
DG
1443 /* Fetch once the poll data */
1444 revents = LTTNG_POLL_GETEV(&events, i);
1445 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 1446
5eb91c98
DG
1447 /* Thread quit pipe has been closed. Killing thread. */
1448 ret = check_thread_quit_pipe(pollfd, revents);
1449 if (ret) {
139ac872
MD
1450 err = 0;
1451 goto exit;
90014c57 1452 }
1d4b027a 1453
5eb91c98
DG
1454 /* Event on the registration socket */
1455 if (pollfd == apps_sock) {
1456 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1457 ERR("Register apps socket poll error");
1458 goto error;
1459 } else if (revents & LPOLLIN) {
1460 sock = lttcomm_accept_unix_sock(apps_sock);
1461 if (sock < 0) {
1462 goto error;
1463 }
099e26bd 1464
b662582b
DG
1465 /*
1466 * Set the CLOEXEC flag. Return code is useless because
1467 * either way, the show must go on.
1468 */
1469 (void) utils_set_fd_cloexec(sock);
1470
5eb91c98 1471 /* Create UST registration command for enqueuing */
ba7f0ae5 1472 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 1473 if (ust_cmd == NULL) {
76d7553f 1474 PERROR("ust command zmalloc");
5eb91c98
DG
1475 goto error;
1476 }
1d4b027a 1477
5eb91c98
DG
1478 /*
1479 * Using message-based transmissions to ensure we don't
1480 * have to deal with partially received messages.
1481 */
4063050c
MD
1482 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1483 if (ret < 0) {
1484 ERR("Exhausted file descriptors allowed for applications.");
1485 free(ust_cmd);
1486 ret = close(sock);
1487 if (ret) {
1488 PERROR("close");
1489 }
1490 sock = -1;
1491 continue;
1492 }
840cb59c 1493 health_code_update();
5eb91c98
DG
1494 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1495 sizeof(struct ust_register_msg));
1496 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1497 if (ret < 0) {
76d7553f 1498 PERROR("lttcomm_recv_unix_sock register apps");
5eb91c98
DG
1499 } else {
1500 ERR("Wrong size received on apps register");
1501 }
1502 free(ust_cmd);
76d7553f
MD
1503 ret = close(sock);
1504 if (ret) {
1505 PERROR("close");
1506 }
4063050c 1507 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 1508 sock = -1;
5eb91c98
DG
1509 continue;
1510 }
840cb59c 1511 health_code_update();
099e26bd 1512
5eb91c98 1513 ust_cmd->sock = sock;
34a2494f 1514 sock = -1;
099e26bd 1515
5eb91c98
DG
1516 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1517 " gid:%d sock:%d name:%s (version %d.%d)",
1518 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1519 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1520 ust_cmd->sock, ust_cmd->reg_msg.name,
1521 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 1522
5eb91c98
DG
1523 /*
1524 * Lock free enqueue the registration request. The red pill
54d01ffb 1525 * has been taken! This apps will be part of the *system*.
5eb91c98
DG
1526 */
1527 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1528
1529 /*
1530 * Wake the registration queue futex. Implicit memory
1531 * barrier with the exchange in cds_wfq_enqueue.
1532 */
1533 futex_nto1_wake(&ust_cmd_queue.futex);
1534 }
1535 }
90014c57 1536 }
1d4b027a
DG
1537 }
1538
139ac872 1539exit:
1d4b027a 1540error:
139ac872 1541 if (err) {
840cb59c 1542 health_error();
139ac872
MD
1543 ERR("Health error occurred in %s", __func__);
1544 }
139ac872 1545
0fdd1e2c
DG
1546 /* Notify that the registration thread is gone */
1547 notify_ust_apps(0);
1548
a4b35e07 1549 if (apps_sock >= 0) {
76d7553f
MD
1550 ret = close(apps_sock);
1551 if (ret) {
1552 PERROR("close");
1553 }
a4b35e07 1554 }
46c3f085 1555 if (sock >= 0) {
76d7553f
MD
1556 ret = close(sock);
1557 if (ret) {
1558 PERROR("close");
1559 }
4063050c 1560 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 1561 }
273ea72c 1562 unlink(apps_unix_sock_path);
0fdd1e2c 1563
76d7553f 1564error_poll_add:
5eb91c98 1565 lttng_poll_clean(&events);
76d7553f
MD
1566error_listen:
1567error_create_poll:
6993eeb3 1568error_testpoint:
76d7553f 1569 DBG("UST Registration thread cleanup complete");
927ca06a 1570 health_unregister();
5eb91c98 1571
1d4b027a
DG
1572 return NULL;
1573}
1574
8c0faa1d 1575/*
3bd1e081 1576 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 1577 * exec or it will fails.
8c0faa1d 1578 */
3bd1e081 1579static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 1580{
a23ec3a7 1581 int ret, clock_ret;
ee0b0061
DG
1582 struct timespec timeout;
1583
a23ec3a7
DG
1584 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1585 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 1586
a23ec3a7
DG
1587 /* Setup pthread condition */
1588 ret = pthread_condattr_init(&consumer_data->condattr);
1589 if (ret != 0) {
1590 errno = ret;
1591 PERROR("pthread_condattr_init consumer data");
1592 goto error;
1593 }
1594
1595 /*
1596 * Set the monotonic clock in order to make sure we DO NOT jump in time
1597 * between the clock_gettime() call and the timedwait call. See bug #324
1598 * for a more details and how we noticed it.
1599 */
1600 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1601 if (ret != 0) {
1602 errno = ret;
1603 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
1604 goto error;
1605 }
8c0faa1d 1606
a23ec3a7
DG
1607 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1608 if (ret != 0) {
1609 errno = ret;
1610 PERROR("pthread_cond_init consumer data");
1611 goto error;
1612 }
1613
1614 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1615 consumer_data);
8c0faa1d 1616 if (ret != 0) {
3bd1e081 1617 PERROR("pthread_create consumer");
ee0b0061 1618 ret = -1;
8c0faa1d
DG
1619 goto error;
1620 }
1621
a23ec3a7
DG
1622 /* We are about to wait on a pthread condition */
1623 pthread_mutex_lock(&consumer_data->cond_mutex);
1624
ee0b0061 1625 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
1626 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1627 /*
1628 * Set the timeout for the condition timed wait even if the clock gettime
1629 * call fails since we might loop on that call and we want to avoid to
1630 * increment the timeout too many times.
1631 */
1632 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1633
1634 /*
1635 * The following loop COULD be skipped in some conditions so this is why we
1636 * set ret to 0 in order to make sure at least one round of the loop is
1637 * done.
1638 */
1639 ret = 0;
1640
1641 /*
1642 * Loop until the condition is reached or when a timeout is reached. Note
1643 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1644 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1645 * possible. This loop does not take any chances and works with both of
1646 * them.
1647 */
1648 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1649 if (clock_ret < 0) {
1650 PERROR("clock_gettime spawn consumer");
1651 /* Infinite wait for the consumerd thread to be ready */
1652 ret = pthread_cond_wait(&consumer_data->cond,
1653 &consumer_data->cond_mutex);
1654 } else {
1655 ret = pthread_cond_timedwait(&consumer_data->cond,
1656 &consumer_data->cond_mutex, &timeout);
1657 }
ee0b0061 1658 }
8c0faa1d 1659
a23ec3a7
DG
1660 /* Release the pthread condition */
1661 pthread_mutex_unlock(&consumer_data->cond_mutex);
1662
1663 if (ret != 0) {
1664 errno = ret;
1665 if (ret == ETIMEDOUT) {
ee0b0061
DG
1666 /*
1667 * Call has timed out so we kill the kconsumerd_thread and return
1668 * an error.
1669 */
a23ec3a7
DG
1670 ERR("Condition timed out. The consumer thread was never ready."
1671 " Killing it");
3bd1e081 1672 ret = pthread_cancel(consumer_data->thread);
ee0b0061 1673 if (ret < 0) {
3bd1e081 1674 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
1675 }
1676 } else {
a23ec3a7 1677 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061
DG
1678 }
1679 goto error;
1680 }
1681
3bd1e081
MD
1682 pthread_mutex_lock(&consumer_data->pid_mutex);
1683 if (consumer_data->pid == 0) {
a23ec3a7 1684 ERR("Consumerd did not start");
3bd1e081 1685 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
1686 goto error;
1687 }
3bd1e081 1688 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 1689
8c0faa1d
DG
1690 return 0;
1691
1692error:
1693 return ret;
1694}
1695
d9800920 1696/*
3bd1e081 1697 * Join consumer thread
d9800920 1698 */
3bd1e081 1699static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
1700{
1701 void *status;
cf3af59e 1702
e8209f6b
DG
1703 /* Consumer pid must be a real one. */
1704 if (consumer_data->pid > 0) {
c617c0c6 1705 int ret;
3bd1e081 1706 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 1707 if (ret) {
3bd1e081 1708 ERR("Error killing consumer daemon");
cf3af59e
MD
1709 return ret;
1710 }
3bd1e081 1711 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
1712 } else {
1713 return 0;
1714 }
1715}
1716
8c0faa1d 1717/*
3bd1e081 1718 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 1719 *
d063d709 1720 * Return pid if successful else -1.
8c0faa1d 1721 */
3bd1e081 1722static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
1723{
1724 int ret;
1725 pid_t pid;
94c55f17 1726 const char *consumer_to_use;
53086306 1727 const char *verbosity;
94c55f17 1728 struct stat st;
8c0faa1d 1729
3bd1e081 1730 DBG("Spawning consumerd");
c49dc785 1731
8c0faa1d
DG
1732 pid = fork();
1733 if (pid == 0) {
1734 /*
3bd1e081 1735 * Exec consumerd.
8c0faa1d 1736 */
daee5345 1737 if (opt_verbose_consumer) {
53086306
DG
1738 verbosity = "--verbose";
1739 } else {
1740 verbosity = "--quiet";
1741 }
3bd1e081
MD
1742 switch (consumer_data->type) {
1743 case LTTNG_CONSUMER_KERNEL:
94c55f17 1744 /*
c7704d57
DG
1745 * Find out which consumerd to execute. We will first try the
1746 * 64-bit path, then the sessiond's installation directory, and
1747 * fallback on the 32-bit one,
94c55f17 1748 */
63a799e8
AM
1749 DBG3("Looking for a kernel consumer at these locations:");
1750 DBG3(" 1) %s", consumerd64_bin);
1751 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1752 DBG3(" 3) %s", consumerd32_bin);
94c55f17 1753 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 1754 DBG3("Found location #1");
94c55f17 1755 consumer_to_use = consumerd64_bin;
94c55f17 1756 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 1757 DBG3("Found location #2");
94c55f17 1758 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 1759 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 1760 DBG3("Found location #3");
eb1e0bd4 1761 consumer_to_use = consumerd32_bin;
94c55f17 1762 } else {
63a799e8 1763 DBG("Could not find any valid consumerd executable");
94c55f17
AM
1764 break;
1765 }
1766 DBG("Using kernel consumer at: %s", consumer_to_use);
1767 execl(consumer_to_use,
1768 "lttng-consumerd", verbosity, "-k",
1769 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1770 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1771 NULL);
3bd1e081 1772 break;
7753dea8
MD
1773 case LTTNG_CONSUMER64_UST:
1774 {
b1e0b6b6 1775 char *tmpnew = NULL;
8f4905da
MD
1776
1777 if (consumerd64_libdir[0] != '\0') {
1778 char *tmp;
1779 size_t tmplen;
1780
1781 tmp = getenv("LD_LIBRARY_PATH");
1782 if (!tmp) {
1783 tmp = "";
1784 }
1785 tmplen = strlen("LD_LIBRARY_PATH=")
1786 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1787 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1788 if (!tmpnew) {
1789 ret = -ENOMEM;
1790 goto error;
1791 }
1792 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1793 strcat(tmpnew, consumerd64_libdir);
1794 if (tmp[0] != '\0') {
1795 strcat(tmpnew, ":");
1796 strcat(tmpnew, tmp);
1797 }
1798 ret = putenv(tmpnew);
1799 if (ret) {
1800 ret = -errno;
1801 goto error;
1802 }
1803 }
94c55f17 1804 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 1805 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
1806 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1807 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1808 NULL);
8f4905da
MD
1809 if (consumerd64_libdir[0] != '\0') {
1810 free(tmpnew);
1811 }
1812 if (ret) {
1813 goto error;
1814 }
3bd1e081 1815 break;
7753dea8
MD
1816 }
1817 case LTTNG_CONSUMER32_UST:
1818 {
937dde8e 1819 char *tmpnew = NULL;
8f4905da
MD
1820
1821 if (consumerd32_libdir[0] != '\0') {
1822 char *tmp;
1823 size_t tmplen;
1824
1825 tmp = getenv("LD_LIBRARY_PATH");
1826 if (!tmp) {
1827 tmp = "";
1828 }
1829 tmplen = strlen("LD_LIBRARY_PATH=")
1830 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1831 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1832 if (!tmpnew) {
1833 ret = -ENOMEM;
1834 goto error;
1835 }
1836 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1837 strcat(tmpnew, consumerd32_libdir);
1838 if (tmp[0] != '\0') {
1839 strcat(tmpnew, ":");
1840 strcat(tmpnew, tmp);
1841 }
1842 ret = putenv(tmpnew);
1843 if (ret) {
1844 ret = -errno;
1845 goto error;
1846 }
1847 }
94c55f17 1848 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 1849 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
1850 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1851 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1852 NULL);
8f4905da
MD
1853 if (consumerd32_libdir[0] != '\0') {
1854 free(tmpnew);
1855 }
1856 if (ret) {
1857 goto error;
1858 }
7753dea8
MD
1859 break;
1860 }
3bd1e081 1861 default:
76d7553f 1862 PERROR("unknown consumer type");
3bd1e081
MD
1863 exit(EXIT_FAILURE);
1864 }
8c0faa1d 1865 if (errno != 0) {
76d7553f 1866 PERROR("kernel start consumer exec");
8c0faa1d
DG
1867 }
1868 exit(EXIT_FAILURE);
1869 } else if (pid > 0) {
1870 ret = pid;
8c0faa1d 1871 } else {
76d7553f 1872 PERROR("start consumer fork");
8c0faa1d 1873 ret = -errno;
8c0faa1d 1874 }
8f4905da 1875error:
8c0faa1d
DG
1876 return ret;
1877}
1878
693bd40b 1879/*
3bd1e081 1880 * Spawn the consumerd daemon and session daemon thread.
693bd40b 1881 */
3bd1e081 1882static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 1883{
c617c0c6 1884 int ret;
edb8b045
DG
1885
1886 /*
1887 * Set the listen() state on the socket since there is a possible race
1888 * between the exec() of the consumer daemon and this call if place in the
1889 * consumer thread. See bug #366 for more details.
1890 */
1891 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1892 if (ret < 0) {
1893 goto error;
1894 }
693bd40b 1895
3bd1e081
MD
1896 pthread_mutex_lock(&consumer_data->pid_mutex);
1897 if (consumer_data->pid != 0) {
1898 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
1899 goto end;
1900 }
693bd40b 1901
3bd1e081 1902 ret = spawn_consumerd(consumer_data);
c49dc785 1903 if (ret < 0) {
3bd1e081
MD
1904 ERR("Spawning consumerd failed");
1905 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 1906 goto error;
693bd40b 1907 }
c49dc785 1908
3bd1e081
MD
1909 /* Setting up the consumer_data pid */
1910 consumer_data->pid = ret;
48842b30 1911 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 1912 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 1913
3bd1e081
MD
1914 DBG2("Spawning consumer control thread");
1915 ret = spawn_consumer_thread(consumer_data);
693bd40b 1916 if (ret < 0) {
3bd1e081 1917 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
1918 goto error;
1919 }
1920
c49dc785 1921end:
693bd40b
DG
1922 return 0;
1923
1924error:
edb8b045
DG
1925 /* Cleanup already created socket on error. */
1926 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
1927 int err;
1928
edb8b045
DG
1929 err = close(consumer_data->err_sock);
1930 if (err < 0) {
1931 PERROR("close consumer data error socket");
1932 }
1933 }
693bd40b
DG
1934 return ret;
1935}
1936
44a5e5eb 1937/*
139ac872
MD
1938 * Compute health status of each consumer. If one of them is zero (bad
1939 * state), we return 0.
44a5e5eb
DG
1940 */
1941static int check_consumer_health(void)
1942{
1943 int ret;
1944
927ca06a 1945 ret = health_check_state(HEALTH_TYPE_CONSUMER);
44a5e5eb
DG
1946
1947 DBG3("Health consumer check %d", ret);
1948
1949 return ret;
1950}
1951
b73401da 1952/*
096102bd 1953 * Setup necessary data for kernel tracer action.
b73401da 1954 */
096102bd 1955static int init_kernel_tracer(void)
b73401da
DG
1956{
1957 int ret;
b73401da 1958
096102bd
DG
1959 /* Modprobe lttng kernel modules */
1960 ret = modprobe_lttng_control();
b73401da 1961 if (ret < 0) {
b73401da
DG
1962 goto error;
1963 }
1964
096102bd
DG
1965 /* Open debugfs lttng */
1966 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1967 if (kernel_tracer_fd < 0) {
1968 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
1969 ret = -1;
1970 goto error_open;
54d01ffb
DG
1971 }
1972
2f77fc4b
DG
1973 /* Validate kernel version */
1974 ret = kernel_validate_version(kernel_tracer_fd);
1975 if (ret < 0) {
1976 goto error_version;
b551a063 1977 }
54d01ffb 1978
2f77fc4b
DG
1979 ret = modprobe_lttng_data();
1980 if (ret < 0) {
1981 goto error_modules;
54d01ffb
DG
1982 }
1983
2f77fc4b
DG
1984 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1985 return 0;
1986
1987error_version:
1988 modprobe_remove_lttng_control();
1989 ret = close(kernel_tracer_fd);
1990 if (ret) {
1991 PERROR("close");
b551a063 1992 }
2f77fc4b 1993 kernel_tracer_fd = -1;
f73fabfd 1994 return LTTNG_ERR_KERN_VERSION;
b551a063 1995
2f77fc4b
DG
1996error_modules:
1997 ret = close(kernel_tracer_fd);
1998 if (ret) {
1999 PERROR("close");
b551a063 2000 }
54d01ffb 2001
2f77fc4b
DG
2002error_open:
2003 modprobe_remove_lttng_control();
54d01ffb
DG
2004
2005error:
2f77fc4b
DG
2006 WARN("No kernel tracer available");
2007 kernel_tracer_fd = -1;
2008 if (!is_root) {
f73fabfd 2009 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2010 } else {
f73fabfd 2011 return LTTNG_ERR_KERN_NA;
2f77fc4b 2012 }
54d01ffb
DG
2013}
2014
2f77fc4b 2015
54d01ffb 2016/*
2f77fc4b
DG
2017 * Copy consumer output from the tracing session to the domain session. The
2018 * function also applies the right modification on a per domain basis for the
2019 * trace files destination directory.
54d01ffb 2020 */
2f77fc4b 2021static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2022{
2023 int ret;
2f77fc4b
DG
2024 const char *dir_name;
2025 struct consumer_output *consumer;
2026
2027 assert(session);
2028 assert(session->consumer);
54d01ffb 2029
b551a063
DG
2030 switch (domain) {
2031 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2032 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2033 /*
2034 * XXX: We should audit the session creation and what this function
2035 * does "extra" in order to avoid a destroy since this function is used
2036 * in the domain session creation (kernel and ust) only. Same for UST
2037 * domain.
2038 */
2039 if (session->kernel_session->consumer) {
2040 consumer_destroy_output(session->kernel_session->consumer);
2041 }
2f77fc4b
DG
2042 session->kernel_session->consumer =
2043 consumer_copy_output(session->consumer);
2044 /* Ease our life a bit for the next part */
2045 consumer = session->kernel_session->consumer;
2046 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063
DG
2047 break;
2048 case LTTNG_DOMAIN_UST:
2f77fc4b 2049 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2050 if (session->ust_session->consumer) {
2051 consumer_destroy_output(session->ust_session->consumer);
2052 }
2f77fc4b
DG
2053 session->ust_session->consumer =
2054 consumer_copy_output(session->consumer);
2055 /* Ease our life a bit for the next part */
2056 consumer = session->ust_session->consumer;
2057 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2058 break;
2059 default:
f73fabfd 2060 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2061 goto error;
2062 }
2063
2f77fc4b 2064 /* Append correct directory to subdir */
c30ce0b3
CB
2065 strncat(consumer->subdir, dir_name,
2066 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2067 DBG3("Copy session consumer subdir %s", consumer->subdir);
2068
f73fabfd 2069 ret = LTTNG_OK;
54d01ffb
DG
2070
2071error:
2072 return ret;
2073}
2074
00e2e675 2075/*
2f77fc4b 2076 * Create an UST session and add it to the session ust list.
00e2e675 2077 */
2f77fc4b
DG
2078static int create_ust_session(struct ltt_session *session,
2079 struct lttng_domain *domain)
00e2e675
DG
2080{
2081 int ret;
2f77fc4b 2082 struct ltt_ust_session *lus = NULL;
00e2e675 2083
a4b92340 2084 assert(session);
2f77fc4b
DG
2085 assert(domain);
2086 assert(session->consumer);
a4b92340 2087
2f77fc4b
DG
2088 switch (domain->type) {
2089 case LTTNG_DOMAIN_UST:
2090 break;
2091 default:
2092 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2093 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2094 goto error;
2095 }
2096
2f77fc4b
DG
2097 DBG("Creating UST session");
2098
fdd9eb17 2099 lus = trace_ust_create_session(session->path, session->id);
2f77fc4b 2100 if (lus == NULL) {
f73fabfd 2101 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2102 goto error;
2103 }
2104
2f77fc4b
DG
2105 lus->uid = session->uid;
2106 lus->gid = session->gid;
2107 session->ust_session = lus;
00e2e675 2108
2f77fc4b
DG
2109 /* Copy session output to the newly created UST session */
2110 ret = copy_session_consumer(domain->type, session);
f73fabfd 2111 if (ret != LTTNG_OK) {
00e2e675
DG
2112 goto error;
2113 }
2114
f73fabfd 2115 return LTTNG_OK;
00e2e675
DG
2116
2117error:
2f77fc4b
DG
2118 free(lus);
2119 session->ust_session = NULL;
00e2e675
DG
2120 return ret;
2121}
2122
2123/*
2f77fc4b 2124 * Create a kernel tracer session then create the default channel.
00e2e675 2125 */
2f77fc4b 2126static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2127{
2128 int ret;
a4b92340 2129
2f77fc4b 2130 DBG("Creating kernel session");
00e2e675 2131
2f77fc4b
DG
2132 ret = kernel_create_session(session, kernel_tracer_fd);
2133 if (ret < 0) {
f73fabfd 2134 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2135 goto error;
2136 }
2137
2f77fc4b
DG
2138 /* Code flow safety */
2139 assert(session->kernel_session);
2140
2141 /* Copy session output to the newly created Kernel session */
2142 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2143 if (ret != LTTNG_OK) {
a4b92340
DG
2144 goto error;
2145 }
2146
2f77fc4b
DG
2147 /* Create directory(ies) on local filesystem. */
2148 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2149 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2150 ret = run_as_mkdir_recursive(
2151 session->kernel_session->consumer->dst.trace_path,
2152 S_IRWXU | S_IRWXG, session->uid, session->gid);
2153 if (ret < 0) {
2154 if (ret != -EEXIST) {
2155 ERR("Trace directory creation error");
00e2e675
DG
2156 goto error;
2157 }
00e2e675 2158 }
2f77fc4b 2159 }
00e2e675 2160
2f77fc4b
DG
2161 session->kernel_session->uid = session->uid;
2162 session->kernel_session->gid = session->gid;
00e2e675 2163
f73fabfd 2164 return LTTNG_OK;
00e2e675 2165
2f77fc4b
DG
2166error:
2167 trace_kernel_destroy_session(session->kernel_session);
2168 session->kernel_session = NULL;
2169 return ret;
2170}
00e2e675 2171
2f77fc4b
DG
2172/*
2173 * Count number of session permitted by uid/gid.
2174 */
2175static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2176{
2177 unsigned int i = 0;
2178 struct ltt_session *session;
07424f16 2179
2f77fc4b
DG
2180 DBG("Counting number of available session for UID %d GID %d",
2181 uid, gid);
2182 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2183 /*
2f77fc4b 2184 * Only list the sessions the user can control.
00e2e675 2185 */
2f77fc4b
DG
2186 if (!session_access_ok(session, uid, gid)) {
2187 continue;
2188 }
2189 i++;
a4b92340 2190 }
2f77fc4b 2191 return i;
00e2e675
DG
2192}
2193
54d01ffb
DG
2194/*
2195 * Process the command requested by the lttng client within the command
2196 * context structure. This function make sure that the return structure (llm)
2197 * is set and ready for transmission before returning.
2198 *
2199 * Return any error encountered or 0 for success.
53a80697
MD
2200 *
2201 * "sock" is only used for special-case var. len data.
54d01ffb 2202 */
53a80697
MD
2203static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2204 int *sock_error)
54d01ffb 2205{
f73fabfd 2206 int ret = LTTNG_OK;
44d3bd01 2207 int need_tracing_session = 1;
2e09ba09 2208 int need_domain;
54d01ffb
DG
2209
2210 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2211
53a80697
MD
2212 *sock_error = 0;
2213
2e09ba09
MD
2214 switch (cmd_ctx->lsm->cmd_type) {
2215 case LTTNG_CREATE_SESSION:
2216 case LTTNG_DESTROY_SESSION:
2217 case LTTNG_LIST_SESSIONS:
2218 case LTTNG_LIST_DOMAINS:
2219 case LTTNG_START_TRACE:
2220 case LTTNG_STOP_TRACE:
6d805429 2221 case LTTNG_DATA_PENDING:
2e09ba09 2222 need_domain = 0;
3aace903 2223 break;
2e09ba09
MD
2224 default:
2225 need_domain = 1;
2226 }
2227
2228 if (opt_no_kernel && need_domain
2229 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2230 if (!is_root) {
f73fabfd 2231 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2232 } else {
f73fabfd 2233 ret = LTTNG_ERR_KERN_NA;
531d29f9 2234 }
4fba7219
DG
2235 goto error;
2236 }
2237
8d3113b2
DG
2238 /* Deny register consumer if we already have a spawned consumer. */
2239 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2240 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2241 if (kconsumer_data.pid > 0) {
f73fabfd 2242 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2243 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2244 goto error;
2245 }
2246 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2247 }
2248
54d01ffb
DG
2249 /*
2250 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2251 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2252 * command.
2253 */
2254 switch(cmd_ctx->lsm->cmd_type) {
2255 case LTTNG_LIST_SESSIONS:
2256 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2257 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2258 case LTTNG_LIST_DOMAINS:
2259 case LTTNG_LIST_CHANNELS:
2260 case LTTNG_LIST_EVENTS:
2261 break;
2262 default:
2263 /* Setup lttng message with no payload */
2264 ret = setup_lttng_msg(cmd_ctx, 0);
2265 if (ret < 0) {
2266 /* This label does not try to unlock the session */
2267 goto init_setup_error;
2268 }
2269 }
2270
2271 /* Commands that DO NOT need a session. */
2272 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2273 case LTTNG_CREATE_SESSION:
2e09ba09 2274 case LTTNG_CALIBRATE:
54d01ffb
DG
2275 case LTTNG_LIST_SESSIONS:
2276 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2277 case LTTNG_LIST_TRACEPOINT_FIELDS:
44d3bd01 2278 need_tracing_session = 0;
54d01ffb
DG
2279 break;
2280 default:
2281 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2282 /*
2283 * We keep the session list lock across _all_ commands
2284 * for now, because the per-session lock does not
2285 * handle teardown properly.
2286 */
74babd95 2287 session_lock_list();
54d01ffb
DG
2288 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2289 if (cmd_ctx->session == NULL) {
2290 if (cmd_ctx->lsm->session.name != NULL) {
f73fabfd 2291 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2292 } else {
2293 /* If no session name specified */
f73fabfd 2294 ret = LTTNG_ERR_SELECT_SESS;
54d01ffb
DG
2295 }
2296 goto error;
2297 } else {
2298 /* Acquire lock for the session */
2299 session_lock(cmd_ctx->session);
2300 }
2301 break;
2302 }
b389abbe 2303
2e09ba09
MD
2304 if (!need_domain) {
2305 goto skip_domain;
2306 }
a4b92340 2307
54d01ffb
DG
2308 /*
2309 * Check domain type for specific "pre-action".
2310 */
2311 switch (cmd_ctx->lsm->domain.type) {
2312 case LTTNG_DOMAIN_KERNEL:
d1f1c568 2313 if (!is_root) {
f73fabfd 2314 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
2315 goto error;
2316 }
2317
54d01ffb 2318 /* Kernel tracer check */
a4b35e07 2319 if (kernel_tracer_fd == -1) {
54d01ffb 2320 /* Basically, load kernel tracer modules */
096102bd
DG
2321 ret = init_kernel_tracer();
2322 if (ret != 0) {
54d01ffb
DG
2323 goto error;
2324 }
2325 }
5eb91c98 2326
5c827ce0
DG
2327 /* Consumer is in an ERROR state. Report back to client */
2328 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2329 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2330 goto error;
2331 }
2332
54d01ffb 2333 /* Need a session for kernel command */
44d3bd01 2334 if (need_tracing_session) {
54d01ffb 2335 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 2336 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 2337 if (ret < 0) {
f73fabfd 2338 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
2339 goto error;
2340 }
b389abbe 2341 }
7d29a247 2342
54d01ffb 2343 /* Start the kernel consumer daemon */
3bd1e081
MD
2344 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2345 if (kconsumer_data.pid == 0 &&
785d2d0d 2346 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3bd1e081
MD
2347 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2348 ret = start_consumerd(&kconsumer_data);
7d29a247 2349 if (ret < 0) {
f73fabfd 2350 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 2351 goto error;
950131af 2352 }
5c827ce0 2353 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
2354 } else {
2355 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 2356 }
173af62f 2357
a4b92340
DG
2358 /*
2359 * The consumer was just spawned so we need to add the socket to
2360 * the consumer output of the session if exist.
2361 */
2362 ret = consumer_create_socket(&kconsumer_data,
2363 cmd_ctx->session->kernel_session->consumer);
2364 if (ret < 0) {
2365 goto error;
173af62f 2366 }
0d0c377a 2367 }
5c827ce0 2368
54d01ffb 2369 break;
2bdd86d4 2370 case LTTNG_DOMAIN_UST:
44d3bd01 2371 {
5c827ce0
DG
2372 /* Consumer is in an ERROR state. Report back to client */
2373 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2374 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2375 goto error;
2376 }
2377
44d3bd01 2378 if (need_tracing_session) {
a4b92340 2379 /* Create UST session if none exist. */
f6a9efaa 2380 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 2381 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 2382 &cmd_ctx->lsm->domain);
f73fabfd 2383 if (ret != LTTNG_OK) {
44d3bd01
DG
2384 goto error;
2385 }
2386 }
00e2e675 2387
7753dea8
MD
2388 /* Start the UST consumer daemons */
2389 /* 64-bit */
2390 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 2391 if (consumerd64_bin[0] != '\0' &&
7753dea8 2392 ustconsumer64_data.pid == 0 &&
785d2d0d 2393 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2394 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2395 ret = start_consumerd(&ustconsumer64_data);
2bdd86d4 2396 if (ret < 0) {
f73fabfd 2397 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
173af62f 2398 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2bdd86d4
MD
2399 goto error;
2400 }
48842b30 2401
173af62f 2402 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
5c827ce0 2403 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3ff2ecac 2404 } else {
7753dea8
MD
2405 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2406 }
173af62f
DG
2407
2408 /*
2409 * Setup socket for consumer 64 bit. No need for atomic access
2410 * since it was set above and can ONLY be set in this thread.
2411 */
a4b92340
DG
2412 ret = consumer_create_socket(&ustconsumer64_data,
2413 cmd_ctx->session->ust_session->consumer);
2414 if (ret < 0) {
2415 goto error;
173af62f
DG
2416 }
2417
7753dea8 2418 /* 32-bit */
fc7a59ce 2419 if (consumerd32_bin[0] != '\0' &&
7753dea8 2420 ustconsumer32_data.pid == 0 &&
785d2d0d 2421 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2422 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2423 ret = start_consumerd(&ustconsumer32_data);
2424 if (ret < 0) {
f73fabfd 2425 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
173af62f 2426 uatomic_set(&ust_consumerd32_fd, -EINVAL);
7753dea8
MD
2427 goto error;
2428 }
5c827ce0 2429
173af62f 2430 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
5c827ce0 2431 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
7753dea8
MD
2432 } else {
2433 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2bdd86d4 2434 }
173af62f
DG
2435
2436 /*
2437 * Setup socket for consumer 64 bit. No need for atomic access
2438 * since it was set above and can ONLY be set in this thread.
2439 */
a4b92340
DG
2440 ret = consumer_create_socket(&ustconsumer32_data,
2441 cmd_ctx->session->ust_session->consumer);
2442 if (ret < 0) {
2443 goto error;
173af62f 2444 }
44d3bd01
DG
2445 }
2446 break;
48842b30 2447 }
54d01ffb 2448 default:
54d01ffb
DG
2449 break;
2450 }
2e09ba09 2451skip_domain:
33a2b854 2452
5c827ce0
DG
2453 /* Validate consumer daemon state when start/stop trace command */
2454 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2455 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2456 switch (cmd_ctx->lsm->domain.type) {
2457 case LTTNG_DOMAIN_UST:
2458 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2459 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2460 goto error;
2461 }
2462 break;
2463 case LTTNG_DOMAIN_KERNEL:
2464 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2465 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2466 goto error;
2467 }
2468 break;
2469 }
2470 }
2471
8e0af1b4
MD
2472 /*
2473 * Check that the UID or GID match that of the tracing session.
2474 * The root user can interact with all sessions.
2475 */
2476 if (need_tracing_session) {
2477 if (!session_access_ok(cmd_ctx->session,
730389d9
DG
2478 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2479 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
f73fabfd 2480 ret = LTTNG_ERR_EPERM;
8e0af1b4
MD
2481 goto error;
2482 }
2483 }
2484
ffe60014
DG
2485 /*
2486 * Send relayd information to consumer as soon as we have a domain and a
2487 * session defined.
2488 */
2489 if (cmd_ctx->session && need_domain) {
2490 /*
2491 * Setup relayd if not done yet. If the relayd information was already
2492 * sent to the consumer, this call will gracefully return.
2493 */
2494 ret = cmd_setup_relayd(cmd_ctx->session);
2495 if (ret != LTTNG_OK) {
2496 goto error;
2497 }
2498 }
2499
54d01ffb
DG
2500 /* Process by command type */
2501 switch (cmd_ctx->lsm->cmd_type) {
2502 case LTTNG_ADD_CONTEXT:
2503 {
2504 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2505 cmd_ctx->lsm->u.context.channel_name,
979e618e 2506 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
54d01ffb
DG
2507 break;
2508 }
2509 case LTTNG_DISABLE_CHANNEL:
2510 {
2511 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2512 cmd_ctx->lsm->u.disable.channel_name);
2513 break;
2514 }
2515 case LTTNG_DISABLE_EVENT:
2516 {
2517 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2518 cmd_ctx->lsm->u.disable.channel_name,
2519 cmd_ctx->lsm->u.disable.name);
33a2b854
DG
2520 break;
2521 }
54d01ffb
DG
2522 case LTTNG_DISABLE_ALL_EVENT:
2523 {
2bdd86d4 2524 DBG("Disabling all events");
54d01ffb
DG
2525
2526 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2527 cmd_ctx->lsm->u.disable.channel_name);
2528 break;
2529 }
2530 case LTTNG_ENABLE_CHANNEL:
2531 {
aea829b3 2532 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 2533 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
54d01ffb
DG
2534 break;
2535 }
2536 case LTTNG_ENABLE_EVENT:
2537 {
2538 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2539 cmd_ctx->lsm->u.enable.channel_name,
025faf73 2540 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
2541 break;
2542 }
2543 case LTTNG_ENABLE_ALL_EVENT:
2544 {
2bdd86d4 2545 DBG("Enabling all events");
54d01ffb
DG
2546
2547 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
8c9ae521 2548 cmd_ctx->lsm->u.enable.channel_name,
025faf73 2549 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
2550 break;
2551 }
052da939 2552 case LTTNG_LIST_TRACEPOINTS:
2ef84c95 2553 {
9f19cc17 2554 struct lttng_event *events;
54d01ffb 2555 ssize_t nb_events;
052da939 2556
54d01ffb
DG
2557 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2558 if (nb_events < 0) {
f73fabfd 2559 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2560 ret = -nb_events;
2561 goto error;
2ef84c95
DG
2562 }
2563
2564 /*
2565 * Setup lttng message with payload size set to the event list size in
2566 * bytes and then copy list into the llm payload.
2567 */
052da939 2568 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2ef84c95 2569 if (ret < 0) {
052da939 2570 free(events);
2ef84c95
DG
2571 goto setup_error;
2572 }
2573
2574 /* Copy event list into message payload */
9f19cc17 2575 memcpy(cmd_ctx->llm->payload, events,
052da939 2576 sizeof(struct lttng_event) * nb_events);
2ef84c95 2577
9f19cc17 2578 free(events);
2ef84c95 2579
f73fabfd 2580 ret = LTTNG_OK;
2ef84c95
DG
2581 break;
2582 }
f37d259d
MD
2583 case LTTNG_LIST_TRACEPOINT_FIELDS:
2584 {
2585 struct lttng_event_field *fields;
2586 ssize_t nb_fields;
2587
a4b92340
DG
2588 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2589 &fields);
f37d259d 2590 if (nb_fields < 0) {
f73fabfd 2591 /* Return value is a negative lttng_error_code. */
f37d259d
MD
2592 ret = -nb_fields;
2593 goto error;
2594 }
2595
2596 /*
2597 * Setup lttng message with payload size set to the event list size in
2598 * bytes and then copy list into the llm payload.
2599 */
a4b92340
DG
2600 ret = setup_lttng_msg(cmd_ctx,
2601 sizeof(struct lttng_event_field) * nb_fields);
f37d259d
MD
2602 if (ret < 0) {
2603 free(fields);
2604 goto setup_error;
2605 }
2606
2607 /* Copy event list into message payload */
2608 memcpy(cmd_ctx->llm->payload, fields,
2609 sizeof(struct lttng_event_field) * nb_fields);
2610
2611 free(fields);
2612
f73fabfd 2613 ret = LTTNG_OK;
f37d259d
MD
2614 break;
2615 }
00e2e675
DG
2616 case LTTNG_SET_CONSUMER_URI:
2617 {
a4b92340
DG
2618 size_t nb_uri, len;
2619 struct lttng_uri *uris;
2620
2621 nb_uri = cmd_ctx->lsm->u.uri.size;
2622 len = nb_uri * sizeof(struct lttng_uri);
2623
2624 if (nb_uri == 0) {
f73fabfd 2625 ret = LTTNG_ERR_INVALID;
a4b92340
DG
2626 goto error;
2627 }
2628
2629 uris = zmalloc(len);
2630 if (uris == NULL) {
f73fabfd 2631 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2632 goto error;
2633 }
2634
2635 /* Receive variable len data */
77c7c900 2636 DBG("Receiving %zu URI(s) from client ...", nb_uri);
a4b92340
DG
2637 ret = lttcomm_recv_unix_sock(sock, uris, len);
2638 if (ret <= 0) {
2639 DBG("No URIs received from client... continuing");
2640 *sock_error = 1;
f73fabfd 2641 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2642 free(uris);
a4b92340
DG
2643 goto error;
2644 }
2645
00e2e675 2646 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
a4b92340 2647 nb_uri, uris);
f73fabfd 2648 if (ret != LTTNG_OK) {
b1d41407 2649 free(uris);
a4b92340
DG
2650 goto error;
2651 }
2652
2653 /*
2654 * XXX: 0 means that this URI should be applied on the session. Should
2655 * be a DOMAIN enuam.
2656 */
2657 if (cmd_ctx->lsm->domain.type == 0) {
2658 /* Add the URI for the UST session if a consumer is present. */
2659 if (cmd_ctx->session->ust_session &&
2660 cmd_ctx->session->ust_session->consumer) {
2661 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2662 nb_uri, uris);
2663 } else if (cmd_ctx->session->kernel_session &&
2664 cmd_ctx->session->kernel_session->consumer) {
2665 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2666 cmd_ctx->session, nb_uri, uris);
2667 }
2668 }
2669
b1d41407
DG
2670 free(uris);
2671
00e2e675
DG
2672 break;
2673 }
f3ed775e 2674 case LTTNG_START_TRACE:
8c0faa1d 2675 {
54d01ffb 2676 ret = cmd_start_trace(cmd_ctx->session);
8c0faa1d
DG
2677 break;
2678 }
f3ed775e 2679 case LTTNG_STOP_TRACE:
8c0faa1d 2680 {
54d01ffb 2681 ret = cmd_stop_trace(cmd_ctx->session);
8c0faa1d
DG
2682 break;
2683 }
5e16da05
MD
2684 case LTTNG_CREATE_SESSION:
2685 {
a4b92340
DG
2686 size_t nb_uri, len;
2687 struct lttng_uri *uris = NULL;
2688
2689 nb_uri = cmd_ctx->lsm->u.uri.size;
2690 len = nb_uri * sizeof(struct lttng_uri);
2691
2692 if (nb_uri > 0) {
2693 uris = zmalloc(len);
2694 if (uris == NULL) {
f73fabfd 2695 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2696 goto error;
2697 }
2698
2699 /* Receive variable len data */
77c7c900 2700 DBG("Waiting for %zu URIs from client ...", nb_uri);
a4b92340
DG
2701 ret = lttcomm_recv_unix_sock(sock, uris, len);
2702 if (ret <= 0) {
2703 DBG("No URIs received from client... continuing");
2704 *sock_error = 1;
f73fabfd 2705 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2706 free(uris);
a4b92340
DG
2707 goto error;
2708 }
2709
2710 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2711 DBG("Creating session with ONE network URI is a bad call");
f73fabfd 2712 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2713 free(uris);
a4b92340
DG
2714 goto error;
2715 }
2716 }
2717
2718 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2719 &cmd_ctx->creds);
2720
b1d41407
DG
2721 free(uris);
2722
00e2e675
DG
2723 break;
2724 }
5e16da05
MD
2725 case LTTNG_DESTROY_SESSION:
2726 {
2f77fc4b 2727 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
a4b92340
DG
2728
2729 /* Set session to NULL so we do not unlock it after free. */
256a5576 2730 cmd_ctx->session = NULL;
5461b305 2731 break;
5e16da05 2732 }
9f19cc17 2733 case LTTNG_LIST_DOMAINS:
5e16da05 2734 {
54d01ffb
DG
2735 ssize_t nb_dom;
2736 struct lttng_domain *domains;
5461b305 2737
54d01ffb
DG
2738 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2739 if (nb_dom < 0) {
f73fabfd 2740 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2741 ret = -nb_dom;
2742 goto error;
ce3d728c 2743 }
520ff687 2744
54d01ffb 2745 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
5461b305
DG
2746 if (ret < 0) {
2747 goto setup_error;
520ff687 2748 }
57167058 2749
54d01ffb
DG
2750 /* Copy event list into message payload */
2751 memcpy(cmd_ctx->llm->payload, domains,
2752 nb_dom * sizeof(struct lttng_domain));
2753
2754 free(domains);
5e16da05 2755
f73fabfd 2756 ret = LTTNG_OK;
5e16da05
MD
2757 break;
2758 }
9f19cc17 2759 case LTTNG_LIST_CHANNELS:
5e16da05 2760 {
6775595e 2761 int nb_chan;
54d01ffb
DG
2762 struct lttng_channel *channels;
2763
b551a063
DG
2764 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2765 cmd_ctx->session, &channels);
54d01ffb 2766 if (nb_chan < 0) {
f73fabfd 2767 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2768 ret = -nb_chan;
2769 goto error;
5461b305 2770 }
ca95a216 2771
54d01ffb 2772 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
5461b305
DG
2773 if (ret < 0) {
2774 goto setup_error;
2775 }
9f19cc17 2776
54d01ffb
DG
2777 /* Copy event list into message payload */
2778 memcpy(cmd_ctx->llm->payload, channels,
2779 nb_chan * sizeof(struct lttng_channel));
2780
2781 free(channels);
9f19cc17 2782
f73fabfd 2783 ret = LTTNG_OK;
5461b305 2784 break;
5e16da05 2785 }
9f19cc17 2786 case LTTNG_LIST_EVENTS:
5e16da05 2787 {
b551a063 2788 ssize_t nb_event;
684d34d2 2789 struct lttng_event *events = NULL;
9f19cc17 2790
b551a063 2791 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
54d01ffb
DG
2792 cmd_ctx->lsm->u.list.channel_name, &events);
2793 if (nb_event < 0) {
f73fabfd 2794 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2795 ret = -nb_event;
2796 goto error;
5461b305 2797 }
ca95a216 2798
54d01ffb 2799 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
5461b305
DG
2800 if (ret < 0) {
2801 goto setup_error;
2802 }
9f19cc17 2803
54d01ffb
DG
2804 /* Copy event list into message payload */
2805 memcpy(cmd_ctx->llm->payload, events,
2806 nb_event * sizeof(struct lttng_event));
9f19cc17 2807
54d01ffb 2808 free(events);
9f19cc17 2809
f73fabfd 2810 ret = LTTNG_OK;
5461b305 2811 break;
5e16da05
MD
2812 }
2813 case LTTNG_LIST_SESSIONS:
2814 {
8e0af1b4 2815 unsigned int nr_sessions;
5461b305 2816
8e0af1b4 2817 session_lock_list();
730389d9
DG
2818 nr_sessions = lttng_sessions_count(
2819 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2820 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
d32fb093 2821
8e0af1b4 2822 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
5461b305 2823 if (ret < 0) {
54d01ffb 2824 session_unlock_list();
5461b305 2825 goto setup_error;
e065084a 2826 }
5e16da05 2827
6c9cc2ab 2828 /* Filled the session array */
2f77fc4b 2829 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
730389d9
DG
2830 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2831 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
6c9cc2ab 2832
54d01ffb 2833 session_unlock_list();
5e16da05 2834
f73fabfd 2835 ret = LTTNG_OK;
5e16da05
MD
2836 break;
2837 }
d0254c7c
MD
2838 case LTTNG_CALIBRATE:
2839 {
54d01ffb
DG
2840 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2841 &cmd_ctx->lsm->u.calibrate);
d0254c7c
MD
2842 break;
2843 }
d9800920
DG
2844 case LTTNG_REGISTER_CONSUMER:
2845 {
2f77fc4b
DG
2846 struct consumer_data *cdata;
2847
2848 switch (cmd_ctx->lsm->domain.type) {
2849 case LTTNG_DOMAIN_KERNEL:
2850 cdata = &kconsumer_data;
2851 break;
2852 default:
f73fabfd 2853 ret = LTTNG_ERR_UND;
2f77fc4b
DG
2854 goto error;
2855 }
2856
54d01ffb 2857 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 2858 cmd_ctx->lsm->u.reg.path, cdata);
d9800920
DG
2859 break;
2860 }
025faf73 2861 case LTTNG_ENABLE_EVENT_WITH_FILTER:
53a80697
MD
2862 {
2863 struct lttng_filter_bytecode *bytecode;
2864
025faf73 2865 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
f73fabfd 2866 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2867 goto error;
2868 }
52df2401
MD
2869 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2870 ret = LTTNG_ERR_FILTER_INVAL;
2871 goto error;
2872 }
025faf73 2873 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
53a80697 2874 if (!bytecode) {
f73fabfd 2875 ret = LTTNG_ERR_FILTER_NOMEM;
53a80697
MD
2876 goto error;
2877 }
2878 /* Receive var. len. data */
2879 DBG("Receiving var len data from client ...");
2880 ret = lttcomm_recv_unix_sock(sock, bytecode,
025faf73 2881 cmd_ctx->lsm->u.enable.bytecode_len);
53a80697
MD
2882 if (ret <= 0) {
2883 DBG("Nothing recv() from client var len data... continuing");
2884 *sock_error = 1;
f73fabfd 2885 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2886 goto error;
2887 }
2888
2889 if (bytecode->len + sizeof(*bytecode)
025faf73 2890 != cmd_ctx->lsm->u.enable.bytecode_len) {
53a80697 2891 free(bytecode);
f73fabfd 2892 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2893 goto error;
2894 }
2895
025faf73
DG
2896 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2897 cmd_ctx->lsm->u.enable.channel_name,
2898 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
53a80697
MD
2899 break;
2900 }
6d805429 2901 case LTTNG_DATA_PENDING:
806e2684 2902 {
6d805429 2903 ret = cmd_data_pending(cmd_ctx->session);
806e2684
DG
2904 break;
2905 }
5e16da05 2906 default:
f73fabfd 2907 ret = LTTNG_ERR_UND;
5461b305 2908 break;
fac6795d
DG
2909 }
2910
5461b305 2911error:
5461b305
DG
2912 if (cmd_ctx->llm == NULL) {
2913 DBG("Missing llm structure. Allocating one.");
894be886 2914 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
5461b305
DG
2915 goto setup_error;
2916 }
2917 }
54d01ffb 2918 /* Set return code */
5461b305 2919 cmd_ctx->llm->ret_code = ret;
5461b305 2920setup_error:
b5541356 2921 if (cmd_ctx->session) {
54d01ffb 2922 session_unlock(cmd_ctx->session);
b5541356 2923 }
256a5576
MD
2924 if (need_tracing_session) {
2925 session_unlock_list();
2926 }
54d01ffb 2927init_setup_error:
8028d920 2928 return ret;
fac6795d
DG
2929}
2930
44a5e5eb
DG
2931/*
2932 * Thread managing health check socket.
2933 */
2934static void *thread_manage_health(void *data)
2935{
eb4a2943 2936 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
44a5e5eb
DG
2937 uint32_t revents, nb_fd;
2938 struct lttng_poll_event events;
2939 struct lttcomm_health_msg msg;
2940 struct lttcomm_health_data reply;
2941
2942 DBG("[thread] Manage health check started");
2943
2944 rcu_register_thread();
2945
2946 /* Create unix socket */
2947 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2948 if (sock < 0) {
2949 ERR("Unable to create health check Unix socket");
2950 ret = -1;
2951 goto error;
2952 }
2953
b662582b
DG
2954 /*
2955 * Set the CLOEXEC flag. Return code is useless because either way, the
2956 * show must go on.
2957 */
2958 (void) utils_set_fd_cloexec(sock);
2959
44a5e5eb
DG
2960 ret = lttcomm_listen_unix_sock(sock);
2961 if (ret < 0) {
2962 goto error;
2963 }
2964
2965 /*
2966 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2967 * more will be added to this poll set.
2968 */
2969 ret = create_thread_poll_set(&events, 2);
2970 if (ret < 0) {
2971 goto error;
2972 }
2973
2974 /* Add the application registration socket */
2975 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
2976 if (ret < 0) {
2977 goto error;
2978 }
2979
2980 while (1) {
2981 DBG("Health check ready");
2982
44a5e5eb
DG
2983 /* Inifinite blocking call, waiting for transmission */
2984restart:
2985 ret = lttng_poll_wait(&events, -1);
2986 if (ret < 0) {
2987 /*
2988 * Restart interrupted system call.
2989 */
2990 if (errno == EINTR) {
2991 goto restart;
2992 }
2993 goto error;
2994 }
2995
0d9c5d77
DG
2996 nb_fd = ret;
2997
44a5e5eb
DG
2998 for (i = 0; i < nb_fd; i++) {
2999 /* Fetch once the poll data */
3000 revents = LTTNG_POLL_GETEV(&events, i);
3001 pollfd = LTTNG_POLL_GETFD(&events, i);
3002
3003 /* Thread quit pipe has been closed. Killing thread. */
3004 ret = check_thread_quit_pipe(pollfd, revents);
3005 if (ret) {
139ac872
MD
3006 err = 0;
3007 goto exit;
44a5e5eb
DG
3008 }
3009
3010 /* Event on the registration socket */
3011 if (pollfd == sock) {
3012 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3013 ERR("Health socket poll error");
3014 goto error;
3015 }
3016 }
3017 }
3018
3019 new_sock = lttcomm_accept_unix_sock(sock);
3020 if (new_sock < 0) {
3021 goto error;
3022 }
3023
b662582b
DG
3024 /*
3025 * Set the CLOEXEC flag. Return code is useless because either way, the
3026 * show must go on.
3027 */
3028 (void) utils_set_fd_cloexec(new_sock);
3029
44a5e5eb
DG
3030 DBG("Receiving data from client for health...");
3031 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3032 if (ret <= 0) {
3033 DBG("Nothing recv() from client... continuing");
3034 ret = close(new_sock);
3035 if (ret) {
3036 PERROR("close");
3037 }
3038 new_sock = -1;
3039 continue;
3040 }
3041
3042 rcu_thread_online();
3043
3044 switch (msg.component) {
3045 case LTTNG_HEALTH_CMD:
927ca06a 3046 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
44a5e5eb 3047 break;
139ac872 3048 case LTTNG_HEALTH_APP_MANAGE:
927ca06a 3049 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
139ac872 3050 break;
44a5e5eb 3051 case LTTNG_HEALTH_APP_REG:
927ca06a 3052 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
44a5e5eb
DG
3053 break;
3054 case LTTNG_HEALTH_KERNEL:
927ca06a 3055 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
44a5e5eb
DG
3056 break;
3057 case LTTNG_HEALTH_CONSUMER:
3058 reply.ret_code = check_consumer_health();
3059 break;
3060 case LTTNG_HEALTH_ALL:
44a5e5eb 3061 reply.ret_code =
927ca06a
DG
3062 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3063 health_check_state(HEALTH_TYPE_APP_REG) &&
3064 health_check_state(HEALTH_TYPE_CMD) &&
3065 health_check_state(HEALTH_TYPE_KERNEL) &&
139ac872 3066 check_consumer_health();
44a5e5eb
DG
3067 break;
3068 default:
f73fabfd 3069 reply.ret_code = LTTNG_ERR_UND;
44a5e5eb
DG
3070 break;
3071 }
3072
3073 /*
3074 * Flip ret value since 0 is a success and 1 indicates a bad health for
3075 * the client where in the sessiond it is the opposite. Again, this is
3076 * just to make things easier for us poor developer which enjoy a lot
3077 * lazyness.
3078 */
3079 if (reply.ret_code == 0 || reply.ret_code == 1) {
3080 reply.ret_code = !reply.ret_code;
3081 }
3082
3083 DBG2("Health check return value %d", reply.ret_code);
3084
3085 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3086 if (ret < 0) {
3087 ERR("Failed to send health data back to client");
3088 }
3089
3090 /* End of transmission */
3091 ret = close(new_sock);
3092 if (ret) {
3093 PERROR("close");
3094 }
3095 new_sock = -1;
3096 }
3097
139ac872