Fix: add missing rcu read side lock
[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
44a5e5eb
DG
225/* Used for the health monitoring of the session daemon. See health.h */
226struct health_state health_thread_cmd;
139ac872 227struct health_state health_thread_app_manage;
44a5e5eb
DG
228struct health_state health_thread_app_reg;
229struct health_state health_thread_kernel;
230
ae9e45b3
DG
231/*
232 * Socket timeout for receiving and sending in seconds.
233 */
234static int app_socket_timeout;
235
fb09408a 236static
7753dea8 237void setup_consumerd_path(void)
fb09408a 238{
fc7a59ce 239 const char *bin, *libdir;
fb09408a 240
7753dea8
MD
241 /*
242 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
243 * native architecture size consumer if CONFIG_CONSUMER*_PATH
244 * has not been defined.
7753dea8 245 */
ebaeda94 246#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
247 if (!consumerd32_bin[0]) {
248 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
249 }
250 if (!consumerd32_libdir[0]) {
251 consumerd32_libdir = INSTALL_LIB_PATH;
252 }
253#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
254 if (!consumerd64_bin[0]) {
255 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 256 }
ebaeda94
MD
257 if (!consumerd64_libdir[0]) {
258 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
259 }
260#else
261#error "Unknown bitness"
262#endif
263
fb09408a
MD
264 /*
265 * runtime env. var. overrides the build default.
266 */
fc7a59ce
AM
267 bin = getenv("LTTNG_CONSUMERD32_BIN");
268 if (bin) {
269 consumerd32_bin = bin;
7753dea8 270 }
fc7a59ce
AM
271 bin = getenv("LTTNG_CONSUMERD64_BIN");
272 if (bin) {
273 consumerd64_bin = bin;
ebaeda94 274 }
72f579ee 275 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
276 if (libdir) {
277 consumerd32_libdir = libdir;
278 }
72f579ee 279 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
280 if (libdir) {
281 consumerd64_libdir = libdir;
fb09408a
MD
282 }
283}
284
5eb91c98
DG
285/*
286 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
287 */
288static int create_thread_poll_set(struct lttng_poll_event *events,
289 unsigned int size)
290{
291 int ret;
292
293 if (events == NULL || size == 0) {
294 ret = -1;
295 goto error;
296 }
297
298 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
299 if (ret < 0) {
300 goto error;
301 }
302
303 /* Add quit pipe */
304 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
305 if (ret < 0) {
306 goto error;
307 }
308
309 return 0;
310
311error:
312 return ret;
313}
314
315/*
316 * Check if the thread quit pipe was triggered.
317 *
318 * Return 1 if it was triggered else 0;
319 */
320static int check_thread_quit_pipe(int fd, uint32_t events)
321{
322 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
323 return 1;
324 }
325
326 return 0;
327}
328
0fdd1e2c
DG
329/*
330 * Return group ID of the tracing group or -1 if not found.
331 */
996b65c8
MD
332static gid_t allowed_group(void)
333{
334 struct group *grp;
335
274e143b
MD
336 if (opt_tracing_group) {
337 grp = getgrnam(opt_tracing_group);
338 } else {
339 grp = getgrnam(default_tracing_group);
340 }
996b65c8
MD
341 if (!grp) {
342 return -1;
343 } else {
344 return grp->gr_gid;
345 }
346}
347
273ea72c 348/*
5eb91c98 349 * Init thread quit pipe.
273ea72c
DG
350 *
351 * Return -1 on error or 0 if all pipes are created.
352 */
353static int init_thread_quit_pipe(void)
354{
730389d9 355 int ret, i;
273ea72c 356
730389d9 357 ret = pipe(thread_quit_pipe);
273ea72c 358 if (ret < 0) {
730389d9 359 PERROR("thread quit pipe");
273ea72c
DG
360 goto error;
361 }
362
730389d9
DG
363 for (i = 0; i < 2; i++) {
364 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
365 if (ret < 0) {
366 PERROR("fcntl");
367 goto error;
368 }
369 }
370
273ea72c
DG
371error:
372 return ret;
373}
374
099e26bd
DG
375/*
376 * Stop all threads by closing the thread quit pipe.
377 */
cf3af59e
MD
378static void stop_threads(void)
379{
5eb91c98
DG
380 int ret;
381
cf3af59e
MD
382 /* Stopping all threads */
383 DBG("Terminating all threads");
54d01ffb 384 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
385 if (ret < 0) {
386 ERR("write error on thread quit pipe");
387 }
388
099e26bd 389 /* Dispatch thread */
26c9d55e 390 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 391 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
392}
393
fac6795d 394/*
d063d709 395 * Cleanup the daemon
fac6795d 396 */
cf3af59e 397static void cleanup(void)
fac6795d 398{
ef599319 399 int ret;
6620da75 400 char *cmd = NULL;
af9737e9 401 struct ltt_session *sess, *stmp;
fac6795d 402
1d4b027a 403 DBG("Cleaning up");
e07ae692 404
2f77fc4b
DG
405 /* First thing first, stop all threads */
406 utils_close_pipe(thread_quit_pipe);
407
35f90c40
DG
408 /*
409 * If opt_pidfile is undefined, the default file will be wiped when
410 * removing the rundir.
411 */
412 if (opt_pidfile) {
413 ret = remove(opt_pidfile);
414 if (ret < 0) {
415 PERROR("remove pidfile %s", opt_pidfile);
416 }
417 }
418
67e40797
DG
419 DBG("Removing %s directory", rundir);
420 ret = asprintf(&cmd, "rm -rf %s", rundir);
421 if (ret < 0) {
422 ERR("asprintf failed. Something is really wrong!");
423 }
5461b305 424
67e40797
DG
425 /* Remove lttng run directory */
426 ret = system(cmd);
427 if (ret < 0) {
428 ERR("Unable to clean %s", rundir);
1d4b027a 429 }
67e40797 430 free(cmd);
bd69add2 431 free(rundir);
5461b305 432
99bab54f 433 DBG("Cleaning up all sessions");
fac6795d 434
b5541356 435 /* Destroy session list mutex */
273ea72c
DG
436 if (session_list_ptr != NULL) {
437 pthread_mutex_destroy(&session_list_ptr->lock);
438
439 /* Cleanup ALL session */
54d01ffb
DG
440 cds_list_for_each_entry_safe(sess, stmp,
441 &session_list_ptr->head, list) {
2f77fc4b 442 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
443 }
444 }
445
099e26bd 446 DBG("Closing all UST sockets");
56fff090 447 ust_app_clean_list();
099e26bd 448
4fba7219
DG
449 if (is_root && !opt_no_kernel) {
450 DBG2("Closing kernel fd");
a4b35e07 451 if (kernel_tracer_fd >= 0) {
76d7553f
MD
452 ret = close(kernel_tracer_fd);
453 if (ret) {
454 PERROR("close");
455 }
a4b35e07 456 }
2f50c8a3 457 DBG("Unloading kernel modules");
096102bd 458 modprobe_remove_lttng_all();
2f50c8a3 459 }
2f77fc4b 460
421cb601 461 /* <fun> */
f56a39af 462 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
463 "Matthew, BEET driven development works!%c[%dm",
464 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
465 /* </fun> */
fac6795d
DG
466}
467
e065084a 468/*
d063d709 469 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 470 *
d063d709 471 * Return lttcomm error code.
e065084a
DG
472 */
473static int send_unix_sock(int sock, void *buf, size_t len)
474{
475 /* Check valid length */
c617c0c6 476 if (len == 0) {
e065084a
DG
477 return -1;
478 }
479
480 return lttcomm_send_unix_sock(sock, buf, len);
481}
482
5461b305 483/*
d063d709 484 * Free memory of a command context structure.
5461b305 485 */
a2fb29a5 486static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 487{
a2fb29a5
DG
488 DBG("Clean command context structure");
489 if (*cmd_ctx) {
490 if ((*cmd_ctx)->llm) {
491 free((*cmd_ctx)->llm);
5461b305 492 }
a2fb29a5
DG
493 if ((*cmd_ctx)->lsm) {
494 free((*cmd_ctx)->lsm);
5461b305 495 }
a2fb29a5
DG
496 free(*cmd_ctx);
497 *cmd_ctx = NULL;
5461b305
DG
498 }
499}
500
fac6795d 501/*
0fdd1e2c 502 * Notify UST applications using the shm mmap futex.
fac6795d 503 */
0fdd1e2c 504static int notify_ust_apps(int active)
fac6795d 505{
0fdd1e2c 506 char *wait_shm_mmap;
fac6795d 507
0fdd1e2c 508 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 509
0fdd1e2c
DG
510 /* See shm.c for this call implying mmap, shm and futex calls */
511 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
512 if (wait_shm_mmap == NULL) {
fac6795d
DG
513 goto error;
514 }
515
0fdd1e2c
DG
516 /* Wake waiting process */
517 futex_wait_update((int32_t *) wait_shm_mmap, active);
518
519 /* Apps notified successfully */
520 return 0;
fac6795d
DG
521
522error:
0fdd1e2c 523 return -1;
fac6795d
DG
524}
525
e065084a 526/*
d063d709
DG
527 * Setup the outgoing data buffer for the response (llm) by allocating the
528 * right amount of memory and copying the original information from the lsm
529 * structure.
ca95a216 530 *
d063d709 531 * Return total size of the buffer pointed by buf.
ca95a216 532 */
5461b305 533static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 534{
f3ed775e 535 int ret, buf_size;
ca95a216 536
f3ed775e 537 buf_size = size;
5461b305 538
ba7f0ae5 539 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 540 if (cmd_ctx->llm == NULL) {
76d7553f 541 PERROR("zmalloc");
5461b305 542 ret = -ENOMEM;
ca95a216
DG
543 goto error;
544 }
545
5461b305
DG
546 /* Copy common data */
547 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 548 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 549
5461b305
DG
550 cmd_ctx->llm->data_size = size;
551 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
552
ca95a216
DG
553 return buf_size;
554
555error:
556 return ret;
557}
558
7a485870 559/*
5eb91c98 560 * Update the kernel poll set of all channel fd available over all tracing
d063d709 561 * session. Add the wakeup pipe at the end of the set.
7a485870 562 */
5eb91c98 563static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 564{
5eb91c98 565 int ret;
7a485870
DG
566 struct ltt_session *session;
567 struct ltt_kernel_channel *channel;
568
5eb91c98 569 DBG("Updating kernel poll set");
7a485870 570
54d01ffb 571 session_lock_list();
b5541356 572 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 573 session_lock(session);
7a485870 574 if (session->kernel_session == NULL) {
54d01ffb 575 session_unlock(session);
7a485870
DG
576 continue;
577 }
7a485870 578
54d01ffb
DG
579 cds_list_for_each_entry(channel,
580 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
581 /* Add channel fd to the kernel poll set */
582 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
583 if (ret < 0) {
54d01ffb 584 session_unlock(session);
5eb91c98
DG
585 goto error;
586 }
587 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 588 }
54d01ffb 589 session_unlock(session);
7a485870 590 }
54d01ffb 591 session_unlock_list();
7a485870 592
5eb91c98 593 return 0;
7a485870
DG
594
595error:
54d01ffb 596 session_unlock_list();
7a485870
DG
597 return -1;
598}
599
600/*
54d01ffb 601 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 602 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 603 *
d063d709 604 * Useful for CPU hotplug feature.
7a485870 605 */
2bdd86d4 606static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
607{
608 int ret = 0;
609 struct ltt_session *session;
173af62f 610 struct ltt_kernel_session *ksess;
7a485870
DG
611 struct ltt_kernel_channel *channel;
612
613 DBG("Updating kernel streams for channel fd %d", fd);
614
54d01ffb 615 session_lock_list();
b5541356 616 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 617 session_lock(session);
7a485870 618 if (session->kernel_session == NULL) {
54d01ffb 619 session_unlock(session);
7a485870
DG
620 continue;
621 }
173af62f 622 ksess = session->kernel_session;
d9800920 623
173af62f 624 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
7a485870
DG
625 if (channel->fd == fd) {
626 DBG("Channel found, updating kernel streams");
627 ret = kernel_open_channel_stream(channel);
628 if (ret < 0) {
b3c750d2 629 goto error;
7a485870 630 }
d9800920 631
7a485870 632 /*
5eb91c98
DG
633 * Have we already sent fds to the consumer? If yes, it means
634 * that tracing is started so it is safe to send our updated
635 * stream fds.
7a485870 636 */
173af62f
DG
637 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
638 struct lttng_ht_iter iter;
639 struct consumer_socket *socket;
640
e7fe706f 641 rcu_read_lock();
173af62f
DG
642 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
643 &iter.iter, socket, node.node) {
644 /* Code flow error */
645 assert(socket->fd >= 0);
646
647 pthread_mutex_lock(socket->lock);
f50f23d9 648 ret = kernel_consumer_send_channel_stream(socket,
173af62f
DG
649 channel, ksess);
650 pthread_mutex_unlock(socket->lock);
651 if (ret < 0) {
e7fe706f 652 rcu_read_unlock();
173af62f
DG
653 goto error;
654 }
7a485870 655 }
e7fe706f 656 rcu_read_unlock();
7a485870 657 }
b3c750d2 658 goto error;
7a485870
DG
659 }
660 }
54d01ffb 661 session_unlock(session);
7a485870 662 }
54d01ffb 663 session_unlock_list();
b3c750d2 664 return ret;
7a485870 665
b3c750d2 666error:
54d01ffb
DG
667 session_unlock(session);
668 session_unlock_list();
7a485870
DG
669 return ret;
670}
671
487cf67c
DG
672/*
673 * For each tracing session, update newly registered apps.
674 */
675static void update_ust_app(int app_sock)
676{
677 struct ltt_session *sess, *stmp;
678
4ee14516
DG
679 session_lock_list();
680
487cf67c
DG
681 /* For all tracing session(s) */
682 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
4ee14516 683 session_lock(sess);
421cb601
DG
684 if (sess->ust_session) {
685 ust_app_global_update(sess->ust_session, app_sock);
686 }
4ee14516 687 session_unlock(sess);
487cf67c 688 }
4ee14516
DG
689
690 session_unlock_list();
487cf67c
DG
691}
692
7a485870 693/*
d063d709 694 * This thread manage event coming from the kernel.
7a485870 695 *
d063d709
DG
696 * Features supported in this thread:
697 * -) CPU Hotplug
7a485870
DG
698 */
699static void *thread_manage_kernel(void *data)
700{
139ac872 701 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 702 uint32_t revents, nb_fd;
7a485870 703 char tmp;
5eb91c98 704 struct lttng_poll_event events;
7a485870 705
6993eeb3 706 DBG("[thread] Thread manage kernel started");
7a485870 707
d5d63bf1
DG
708 /*
709 * This first step of the while is to clean this structure which could free
710 * non NULL pointers so zero it before the loop.
711 */
712 memset(&events, 0, sizeof(events));
713
6993eeb3
CB
714 if (testpoint(thread_manage_kernel)) {
715 goto error_testpoint;
716 }
8ac94142 717
44a5e5eb
DG
718 health_code_update(&health_thread_kernel);
719
6993eeb3 720 if (testpoint(thread_manage_kernel_before_loop)) {
d21b0d71 721 goto error_testpoint;
6993eeb3
CB
722 }
723
7a485870 724 while (1) {
44a5e5eb
DG
725 health_code_update(&health_thread_kernel);
726
7a485870 727 if (update_poll_flag == 1) {
d21b0d71
DG
728 /* Clean events object. We are about to populate it again. */
729 lttng_poll_clean(&events);
730
731 ret = create_thread_poll_set(&events, 2);
732 if (ret < 0) {
733 goto error_poll_create;
734 }
735
736 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
737 if (ret < 0) {
738 goto error;
739 }
5f822d0a 740
d21b0d71 741 /* This will add the available kernel channel if any. */
5eb91c98
DG
742 ret = update_kernel_poll(&events);
743 if (ret < 0) {
7a485870
DG
744 goto error;
745 }
746 update_poll_flag = 0;
747 }
748
d21b0d71 749 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
7a485870
DG
750
751 /* Poll infinite value of time */
88f2b785 752 restart:
44a5e5eb 753 health_poll_update(&health_thread_kernel);
5eb91c98 754 ret = lttng_poll_wait(&events, -1);
44a5e5eb 755 health_poll_update(&health_thread_kernel);
7a485870 756 if (ret < 0) {
88f2b785
MD
757 /*
758 * Restart interrupted system call.
759 */
760 if (errno == EINTR) {
761 goto restart;
762 }
7a485870
DG
763 goto error;
764 } else if (ret == 0) {
765 /* Should not happen since timeout is infinite */
85611738
DG
766 ERR("Return value of poll is 0 with an infinite timeout.\n"
767 "This should not have happened! Continuing...");
7a485870
DG
768 continue;
769 }
770
0d9c5d77
DG
771 nb_fd = ret;
772
5eb91c98
DG
773 for (i = 0; i < nb_fd; i++) {
774 /* Fetch once the poll data */
775 revents = LTTNG_POLL_GETEV(&events, i);
776 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 777
44a5e5eb
DG
778 health_code_update(&health_thread_kernel);
779
5eb91c98
DG
780 /* Thread quit pipe has been closed. Killing thread. */
781 ret = check_thread_quit_pipe(pollfd, revents);
782 if (ret) {
139ac872
MD
783 err = 0;
784 goto exit;
5eb91c98 785 }
7a485870 786
5eb91c98
DG
787 /* Check for data on kernel pipe */
788 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
f921c78f
DG
789 do {
790 ret = read(kernel_poll_pipe[0], &tmp, 1);
791 } while (ret < 0 && errno == EINTR);
792 /*
793 * Ret value is useless here, if this pipe gets any actions an
794 * update is required anyway.
795 */
5eb91c98
DG
796 update_poll_flag = 1;
797 continue;
798 } else {
799 /*
800 * New CPU detected by the kernel. Adding kernel stream to
801 * kernel session and updating the kernel consumer
802 */
803 if (revents & LPOLLIN) {
2bdd86d4 804 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
805 if (ret < 0) {
806 continue;
807 }
808 break;
809 /*
810 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
811 * and unregister kernel stream at this point.
812 */
7a485870 813 }
7a485870
DG
814 }
815 }
816 }
817
139ac872 818exit:
7a485870 819error:
5eb91c98 820 lttng_poll_clean(&events);
76d7553f 821error_poll_create:
6993eeb3 822error_testpoint:
6620da75
DG
823 utils_close_pipe(kernel_poll_pipe);
824 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872
MD
825 if (err) {
826 health_error(&health_thread_kernel);
827 ERR("Health error occurred in %s", __func__);
6620da75
DG
828 WARN("Kernel thread died unexpectedly. "
829 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872
MD
830 }
831 health_exit(&health_thread_kernel);
76d7553f 832 DBG("Kernel thread dying");
7a485870
DG
833 return NULL;
834}
835
a23ec3a7
DG
836/*
837 * Signal pthread condition of the consumer data that the thread.
838 */
839static void signal_consumer_condition(struct consumer_data *data, int state)
840{
841 pthread_mutex_lock(&data->cond_mutex);
842
843 /*
844 * The state is set before signaling. It can be any value, it's the waiter
845 * job to correctly interpret this condition variable associated to the
846 * consumer pthread_cond.
847 *
848 * A value of 0 means that the corresponding thread of the consumer data
849 * was not started. 1 indicates that the thread has started and is ready
850 * for action. A negative value means that there was an error during the
851 * thread bootstrap.
852 */
853 data->consumer_thread_is_ready = state;
854 (void) pthread_cond_signal(&data->cond);
855
856 pthread_mutex_unlock(&data->cond_mutex);
857}
858
1d4b027a 859/*
3bd1e081 860 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 861 */
3bd1e081 862static void *thread_manage_consumer(void *data)
1d4b027a 863{
139ac872 864 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98 865 uint32_t revents, nb_fd;
1d4b027a 866 enum lttcomm_return_code code;
5eb91c98 867 struct lttng_poll_event events;
3bd1e081 868 struct consumer_data *consumer_data = data;
1d4b027a 869
3bd1e081 870 DBG("[thread] Manage consumer started");
1d4b027a 871
9449cc75
DG
872 /*
873 * Since the consumer thread can be spawned at any moment in time, we init
874 * the health to a poll status (1, which is a valid health over time).
875 * When the thread starts, we update here the health to a "code" path being
876 * an even value so this thread, when reaching a poll wait, does not
877 * trigger an error with an even value.
878 *
879 * Here is the use case we avoid.
880 *
881 * +1: the first poll update during initialization (main())
882 * +2 * x: multiple code update once in this thread.
883 * +1: poll wait in this thread (being a good health state).
884 * == even number which after the wait period shows as a bad health.
885 *
886 * In a nutshell, the following poll update to the health state brings back
887 * the state to an even value meaning a code path.
888 */
889 health_poll_update(&consumer_data->health);
890
5eb91c98
DG
891 /*
892 * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock.
893 * Nothing more will be added to this poll set.
894 */
895 ret = create_thread_poll_set(&events, 2);
896 if (ret < 0) {
76d7553f 897 goto error_poll;
5eb91c98 898 }
273ea72c 899
edb8b045
DG
900 /*
901 * The error socket here is already in a listening state which was done
902 * just before spawning this thread to avoid a race between the consumer
903 * daemon exec trying to connect and the listen() call.
904 */
3bd1e081 905 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
906 if (ret < 0) {
907 goto error;
908 }
909
44a5e5eb
DG
910 health_code_update(&consumer_data->health);
911
273ea72c 912 /* Inifinite blocking call, waiting for transmission */
88f2b785 913restart:
44a5e5eb 914 health_poll_update(&consumer_data->health);
8ac94142 915
6993eeb3
CB
916 if (testpoint(thread_manage_consumer)) {
917 goto error;
918 }
8ac94142 919
5eb91c98 920 ret = lttng_poll_wait(&events, -1);
44a5e5eb 921 health_poll_update(&consumer_data->health);
273ea72c 922 if (ret < 0) {
88f2b785
MD
923 /*
924 * Restart interrupted system call.
925 */
926 if (errno == EINTR) {
927 goto restart;
928 }
273ea72c
DG
929 goto error;
930 }
931
0d9c5d77
DG
932 nb_fd = ret;
933
5eb91c98
DG
934 for (i = 0; i < nb_fd; i++) {
935 /* Fetch once the poll data */
936 revents = LTTNG_POLL_GETEV(&events, i);
937 pollfd = LTTNG_POLL_GETFD(&events, i);
938
44a5e5eb
DG
939 health_code_update(&consumer_data->health);
940
5eb91c98
DG
941 /* Thread quit pipe has been closed. Killing thread. */
942 ret = check_thread_quit_pipe(pollfd, revents);
943 if (ret) {
139ac872
MD
944 err = 0;
945 goto exit;
5eb91c98
DG
946 }
947
948 /* Event on the registration socket */
3bd1e081 949 if (pollfd == consumer_data->err_sock) {
5eb91c98 950 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 951 ERR("consumer err socket poll error");
5eb91c98
DG
952 goto error;
953 }
954 }
273ea72c
DG
955 }
956
3bd1e081 957 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
958 if (sock < 0) {
959 goto error;
960 }
961
b662582b
DG
962 /*
963 * Set the CLOEXEC flag. Return code is useless because either way, the
964 * show must go on.
965 */
966 (void) utils_set_fd_cloexec(sock);
967
44a5e5eb
DG
968 health_code_update(&consumer_data->health);
969
3bd1e081 970 DBG2("Receiving code from consumer err_sock");
ee0b0061 971
712ea556 972 /* Getting status code from kconsumerd */
54d01ffb
DG
973 ret = lttcomm_recv_unix_sock(sock, &code,
974 sizeof(enum lttcomm_return_code));
1d4b027a
DG
975 if (ret <= 0) {
976 goto error;
977 }
978
44a5e5eb
DG
979 health_code_update(&consumer_data->health);
980
f73fabfd 981 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
3bd1e081
MD
982 consumer_data->cmd_sock =
983 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
984 if (consumer_data->cmd_sock < 0) {
a23ec3a7
DG
985 /* On error, signal condition and quit. */
986 signal_consumer_condition(consumer_data, -1);
48842b30 987 PERROR("consumer connect");
1d4b027a
DG
988 goto error;
989 }
a23ec3a7
DG
990 signal_consumer_condition(consumer_data, 1);
991 DBG("Consumer command socket ready");
1d4b027a 992 } else {
3bd1e081 993 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
994 lttcomm_get_readable_code(-code));
995 goto error;
996 }
997
54d01ffb 998 /* Remove the kconsumerd error sock since we've established a connexion */
3bd1e081 999 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1000 if (ret < 0) {
72079cae
DG
1001 goto error;
1002 }
1003
5eb91c98
DG
1004 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1005 if (ret < 0) {
72079cae 1006 goto error;
5eb91c98
DG
1007 }
1008
44a5e5eb
DG
1009 health_code_update(&consumer_data->health);
1010
5eb91c98 1011 /* Inifinite blocking call, waiting for transmission */
88f2b785 1012restart_poll:
44a5e5eb 1013 health_poll_update(&consumer_data->health);
5eb91c98 1014 ret = lttng_poll_wait(&events, -1);
44a5e5eb 1015 health_poll_update(&consumer_data->health);
5eb91c98 1016 if (ret < 0) {
88f2b785
MD
1017 /*
1018 * Restart interrupted system call.
1019 */
1020 if (errno == EINTR) {
1021 goto restart_poll;
1022 }
72079cae
DG
1023 goto error;
1024 }
1025
0d9c5d77
DG
1026 nb_fd = ret;
1027
5eb91c98
DG
1028 for (i = 0; i < nb_fd; i++) {
1029 /* Fetch once the poll data */
1030 revents = LTTNG_POLL_GETEV(&events, i);
1031 pollfd = LTTNG_POLL_GETFD(&events, i);
1032
44a5e5eb
DG
1033 health_code_update(&consumer_data->health);
1034
5eb91c98
DG
1035 /* Thread quit pipe has been closed. Killing thread. */
1036 ret = check_thread_quit_pipe(pollfd, revents);
1037 if (ret) {
139ac872
MD
1038 err = 0;
1039 goto exit;
5eb91c98
DG
1040 }
1041
1042 /* Event on the kconsumerd socket */
1043 if (pollfd == sock) {
1044 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1045 ERR("consumer err socket second poll error");
5eb91c98
DG
1046 goto error;
1047 }
1048 }
1049 }
1050
44a5e5eb
DG
1051 health_code_update(&consumer_data->health);
1052
712ea556 1053 /* Wait for any kconsumerd error */
54d01ffb
DG
1054 ret = lttcomm_recv_unix_sock(sock, &code,
1055 sizeof(enum lttcomm_return_code));
712ea556 1056 if (ret <= 0) {
3bd1e081 1057 ERR("consumer closed the command socket");
712ea556 1058 goto error;
6f61d021 1059 }
1d4b027a 1060
3bd1e081 1061 ERR("consumer return code : %s", lttcomm_get_readable_code(-code));
712ea556 1062
139ac872 1063exit:
1d4b027a 1064error:
5c827ce0
DG
1065 /* Immediately set the consumerd state to stopped */
1066 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1067 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1068 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1069 consumer_data->type == LTTNG_CONSUMER32_UST) {
1070 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1071 } else {
1072 /* Code flow error... */
1073 assert(0);
1074 }
1075
76d7553f
MD
1076 if (consumer_data->err_sock >= 0) {
1077 ret = close(consumer_data->err_sock);
1078 if (ret) {
1079 PERROR("close");
1080 }
1081 }
1082 if (consumer_data->cmd_sock >= 0) {
1083 ret = close(consumer_data->cmd_sock);
1084 if (ret) {
1085 PERROR("close");
1086 }
1087 }
1088 if (sock >= 0) {
1089 ret = close(sock);
1090 if (ret) {
1091 PERROR("close");
1092 }
1093 }
273ea72c 1094
3bd1e081
MD
1095 unlink(consumer_data->err_unix_sock_path);
1096 unlink(consumer_data->cmd_unix_sock_path);
1097 consumer_data->pid = 0;
1d4b027a 1098
5eb91c98 1099 lttng_poll_clean(&events);
76d7553f 1100error_poll:
139ac872
MD
1101 if (err) {
1102 health_error(&consumer_data->health);
1103 ERR("Health error occurred in %s", __func__);
1104 }
1105 health_exit(&consumer_data->health);
76d7553f 1106 DBG("consumer thread cleanup completed");
0177d773 1107
5eb91c98 1108 return NULL;
099e26bd
DG
1109}
1110
099e26bd
DG
1111/*
1112 * This thread manage application communication.
1d4b027a
DG
1113 */
1114static void *thread_manage_apps(void *data)
099e26bd 1115{
139ac872 1116 int i, ret, pollfd, err = -1;
5eb91c98 1117 uint32_t revents, nb_fd;
099e26bd 1118 struct ust_command ust_cmd;
5eb91c98 1119 struct lttng_poll_event events;
099e26bd
DG
1120
1121 DBG("[thread] Manage application started");
1122
f6a9efaa
DG
1123 rcu_register_thread();
1124 rcu_thread_online();
1125
6993eeb3
CB
1126 if (testpoint(thread_manage_apps)) {
1127 goto error_testpoint;
1128 }
1129
139ac872 1130 health_code_update(&health_thread_app_manage);
44a5e5eb 1131
5eb91c98
DG
1132 ret = create_thread_poll_set(&events, 2);
1133 if (ret < 0) {
76d7553f 1134 goto error_poll_create;
5eb91c98 1135 }
099e26bd 1136
5eb91c98
DG
1137 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1138 if (ret < 0) {
1139 goto error;
1140 }
099e26bd 1141
6993eeb3
CB
1142 if (testpoint(thread_manage_apps_before_loop)) {
1143 goto error;
1144 }
8ac94142 1145
139ac872 1146 health_code_update(&health_thread_app_manage);
44a5e5eb 1147
5eb91c98 1148 while (1) {
d21b0d71 1149 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
099e26bd
DG
1150
1151 /* Inifinite blocking call, waiting for transmission */
88f2b785 1152 restart:
139ac872 1153 health_poll_update(&health_thread_app_manage);
5eb91c98 1154 ret = lttng_poll_wait(&events, -1);
139ac872 1155 health_poll_update(&health_thread_app_manage);
099e26bd 1156 if (ret < 0) {
88f2b785
MD
1157 /*
1158 * Restart interrupted system call.
1159 */
1160 if (errno == EINTR) {
1161 goto restart;
1162 }
099e26bd
DG
1163 goto error;
1164 }
1165
0d9c5d77
DG
1166 nb_fd = ret;
1167
5eb91c98
DG
1168 for (i = 0; i < nb_fd; i++) {
1169 /* Fetch once the poll data */
1170 revents = LTTNG_POLL_GETEV(&events, i);
1171 pollfd = LTTNG_POLL_GETFD(&events, i);
1172
139ac872 1173 health_code_update(&health_thread_app_manage);
44a5e5eb 1174
5eb91c98
DG
1175 /* Thread quit pipe has been closed. Killing thread. */
1176 ret = check_thread_quit_pipe(pollfd, revents);
1177 if (ret) {
139ac872
MD
1178 err = 0;
1179 goto exit;
5eb91c98 1180 }
099e26bd 1181
5eb91c98
DG
1182 /* Inspect the apps cmd pipe */
1183 if (pollfd == apps_cmd_pipe[0]) {
1184 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1185 ERR("Apps command pipe error");
0177d773 1186 goto error;
5eb91c98
DG
1187 } else if (revents & LPOLLIN) {
1188 /* Empty pipe */
f921c78f
DG
1189 do {
1190 ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd));
1191 } while (ret < 0 && errno == EINTR);
5eb91c98 1192 if (ret < 0 || ret < sizeof(ust_cmd)) {
76d7553f 1193 PERROR("read apps cmd pipe");
5eb91c98
DG
1194 goto error;
1195 }
099e26bd 1196
139ac872 1197 health_code_update(&health_thread_app_manage);
44a5e5eb 1198
5eb91c98 1199 /* Register applicaton to the session daemon */
56fff090 1200 ret = ust_app_register(&ust_cmd.reg_msg,
54d01ffb 1201 ust_cmd.sock);
88ff5b7f 1202 if (ret == -ENOMEM) {
5eb91c98 1203 goto error;
88ff5b7f
MD
1204 } else if (ret < 0) {
1205 break;
5eb91c98
DG
1206 }
1207
139ac872 1208 health_code_update(&health_thread_app_manage);
44a5e5eb 1209
acc7b41b 1210 /*
e0c7ec2b 1211 * Validate UST version compatibility.
acc7b41b 1212 */
e0c7ec2b
DG
1213 ret = ust_app_validate_version(ust_cmd.sock);
1214 if (ret >= 0) {
1215 /*
1216 * Add channel(s) and event(s) to newly registered apps
1217 * from lttng global UST domain.
1218 */
1219 update_ust_app(ust_cmd.sock);
1220 }
acc7b41b 1221
139ac872 1222 health_code_update(&health_thread_app_manage);
44a5e5eb 1223
f2ca2e25 1224 ret = ust_app_register_done(ust_cmd.sock);
5eb91c98
DG
1225 if (ret < 0) {
1226 /*
1227 * If the registration is not possible, we simply
1228 * unregister the apps and continue
1229 */
56fff090 1230 ust_app_unregister(ust_cmd.sock);
5eb91c98
DG
1231 } else {
1232 /*
622c9ecf
DG
1233 * We only monitor the error events of the socket. This
1234 * thread does not handle any incoming data from UST
1235 * (POLLIN).
5eb91c98 1236 */
622c9ecf
DG
1237 ret = lttng_poll_add(&events, ust_cmd.sock,
1238 LPOLLERR & LPOLLHUP & LPOLLRDHUP);
5eb91c98
DG
1239 if (ret < 0) {
1240 goto error;
1241 }
1242
ae9e45b3
DG
1243 /* Set socket timeout for both receiving and ending */
1244 (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock,
1245 app_socket_timeout);
1246 (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock,
1247 app_socket_timeout);
1248
54d01ffb
DG
1249 DBG("Apps with sock %d added to poll set",
1250 ust_cmd.sock);
5eb91c98 1251 }
487cf67c 1252
139ac872 1253 health_code_update(&health_thread_app_manage);
44a5e5eb 1254
5eb91c98 1255 break;
0177d773 1256 }
5eb91c98
DG
1257 } else {
1258 /*
54d01ffb
DG
1259 * At this point, we know that a registered application made
1260 * the event at poll_wait.
5eb91c98
DG
1261 */
1262 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1263 /* Removing from the poll set */
1264 ret = lttng_poll_del(&events, pollfd);
1265 if (ret < 0) {
1266 goto error;
1267 }
099e26bd 1268
b9d9b220 1269 /* Socket closed on remote end. */
56fff090 1270 ust_app_unregister(pollfd);
5eb91c98
DG
1271 break;
1272 }
099e26bd 1273 }
44a5e5eb 1274
139ac872 1275 health_code_update(&health_thread_app_manage);
099e26bd 1276 }
099e26bd
DG
1277 }
1278
139ac872 1279exit:
099e26bd 1280error:
5eb91c98 1281 lttng_poll_clean(&events);
76d7553f 1282error_poll_create:
6993eeb3 1283error_testpoint:
6620da75
DG
1284 utils_close_pipe(apps_cmd_pipe);
1285 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1286
1287 /*
1288 * We don't clean the UST app hash table here since already registered
1289 * applications can still be controlled so let them be until the session
1290 * daemon dies or the applications stop.
1291 */
1292
139ac872
MD
1293 if (err) {
1294 health_error(&health_thread_app_manage);
1295 ERR("Health error occurred in %s", __func__);
1296 }
1297 health_exit(&health_thread_app_manage);
76d7553f 1298 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1299 rcu_thread_offline();
1300 rcu_unregister_thread();
099e26bd
DG
1301 return NULL;
1302}
1303
1304/*
1305 * Dispatch request from the registration threads to the application
1306 * communication thread.
1307 */
1308static void *thread_dispatch_ust_registration(void *data)
1309{
1310 int ret;
1311 struct cds_wfq_node *node;
1312 struct ust_command *ust_cmd = NULL;
1313
1314 DBG("[thread] Dispatch UST command started");
1315
26c9d55e 1316 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
099e26bd
DG
1317 /* Atomically prepare the queue futex */
1318 futex_nto1_prepare(&ust_cmd_queue.futex);
1319
1320 do {
1321 /* Dequeue command for registration */
1322 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1323 if (node == NULL) {
00a17c97 1324 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1325 /* Continue thread execution */
1326 break;
1327 }
1328
1329 ust_cmd = caa_container_of(node, struct ust_command, node);
1330
2f50c8a3
DG
1331 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1332 " gid:%d sock:%d name:%s (version %d.%d)",
1333 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1334 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1335 ust_cmd->sock, ust_cmd->reg_msg.name,
1336 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
099e26bd
DG
1337 /*
1338 * Inform apps thread of the new application registration. This
1339 * call is blocking so we can be assured that the data will be read
1340 * at some point in time or wait to the end of the world :)
1341 */
6620da75 1342 if (apps_cmd_pipe[1] >= 0) {
fbb3b395
DG
1343 do {
1344 ret = write(apps_cmd_pipe[1], ust_cmd,
1345 sizeof(struct ust_command));
1346 } while (ret < 0 && errno == EINTR);
4cec016f 1347 if (ret < 0 || ret != sizeof(struct ust_command)) {
6620da75
DG
1348 PERROR("write apps cmd pipe");
1349 if (errno == EBADF) {
1350 /*
1351 * We can't inform the application thread to process
1352 * registration. We will exit or else application
1353 * registration will not occur and tracing will never
1354 * start.
1355 */
1356 goto error;
1357 }
1358 }
1359 } else {
1360 /* Application manager thread is not available. */
1361 ret = close(ust_cmd->sock);
1362 if (ret < 0) {
1363 PERROR("close ust_cmd sock");
099e26bd
DG
1364 }
1365 }
1366 free(ust_cmd);
1367 } while (node != NULL);
1368
1369 /* Futex wait on queue. Blocking call on futex() */
1370 futex_nto1_wait(&ust_cmd_queue.futex);
1371 }
1372
1373error:
1374 DBG("Dispatch thread dying");
1375 return NULL;
1376}
1377
1378/*
1379 * This thread manage application registration.
1380 */
1381static void *thread_registration_apps(void *data)
1d4b027a 1382{
139ac872 1383 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
1384 uint32_t revents, nb_fd;
1385 struct lttng_poll_event events;
099e26bd
DG
1386 /*
1387 * Get allocated in this thread, enqueued to a global queue, dequeued and
1388 * freed in the manage apps thread.
1389 */
1390 struct ust_command *ust_cmd = NULL;
1d4b027a 1391
099e26bd 1392 DBG("[thread] Manage application registration started");
1d4b027a 1393
6993eeb3
CB
1394 if (testpoint(thread_registration_apps)) {
1395 goto error_testpoint;
1396 }
8ac94142 1397
1d4b027a
DG
1398 ret = lttcomm_listen_unix_sock(apps_sock);
1399 if (ret < 0) {
76d7553f 1400 goto error_listen;
1d4b027a
DG
1401 }
1402
5eb91c98
DG
1403 /*
1404 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1405 * more will be added to this poll set.
1406 */
1407 ret = create_thread_poll_set(&events, 2);
1408 if (ret < 0) {
76d7553f 1409 goto error_create_poll;
5eb91c98 1410 }
273ea72c 1411
5eb91c98
DG
1412 /* Add the application registration socket */
1413 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1414 if (ret < 0) {
76d7553f 1415 goto error_poll_add;
5eb91c98 1416 }
273ea72c 1417
1d4b027a 1418 /* Notify all applications to register */
0fdd1e2c
DG
1419 ret = notify_ust_apps(1);
1420 if (ret < 0) {
1421 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
1422 "Execution continues but there might be problem for already\n"
1423 "running applications that wishes to register.");
0fdd1e2c 1424 }
1d4b027a
DG
1425
1426 while (1) {
1427 DBG("Accepting application registration");
273ea72c
DG
1428
1429 /* Inifinite blocking call, waiting for transmission */
88f2b785 1430 restart:
139ac872 1431 health_poll_update(&health_thread_app_reg);
5eb91c98 1432 ret = lttng_poll_wait(&events, -1);
139ac872 1433 health_poll_update(&health_thread_app_reg);
273ea72c 1434 if (ret < 0) {
88f2b785
MD
1435 /*
1436 * Restart interrupted system call.
1437 */
1438 if (errno == EINTR) {
1439 goto restart;
1440 }
273ea72c
DG
1441 goto error;
1442 }
1443
0d9c5d77
DG
1444 nb_fd = ret;
1445
5eb91c98 1446 for (i = 0; i < nb_fd; i++) {
139ac872
MD
1447 health_code_update(&health_thread_app_reg);
1448
5eb91c98
DG
1449 /* Fetch once the poll data */
1450 revents = LTTNG_POLL_GETEV(&events, i);
1451 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 1452
5eb91c98
DG
1453 /* Thread quit pipe has been closed. Killing thread. */
1454 ret = check_thread_quit_pipe(pollfd, revents);
1455 if (ret) {
139ac872
MD
1456 err = 0;
1457 goto exit;
90014c57 1458 }
1d4b027a 1459
5eb91c98
DG
1460 /* Event on the registration socket */
1461 if (pollfd == apps_sock) {
1462 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1463 ERR("Register apps socket poll error");
1464 goto error;
1465 } else if (revents & LPOLLIN) {
1466 sock = lttcomm_accept_unix_sock(apps_sock);
1467 if (sock < 0) {
1468 goto error;
1469 }
099e26bd 1470
b662582b
DG
1471 /*
1472 * Set the CLOEXEC flag. Return code is useless because
1473 * either way, the show must go on.
1474 */
1475 (void) utils_set_fd_cloexec(sock);
1476
5eb91c98 1477 /* Create UST registration command for enqueuing */
ba7f0ae5 1478 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 1479 if (ust_cmd == NULL) {
76d7553f 1480 PERROR("ust command zmalloc");
5eb91c98
DG
1481 goto error;
1482 }
1d4b027a 1483
5eb91c98
DG
1484 /*
1485 * Using message-based transmissions to ensure we don't
1486 * have to deal with partially received messages.
1487 */
4063050c
MD
1488 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1489 if (ret < 0) {
1490 ERR("Exhausted file descriptors allowed for applications.");
1491 free(ust_cmd);
1492 ret = close(sock);
1493 if (ret) {
1494 PERROR("close");
1495 }
1496 sock = -1;
1497 continue;
1498 }
139ac872 1499 health_code_update(&health_thread_app_reg);
5eb91c98
DG
1500 ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg,
1501 sizeof(struct ust_register_msg));
1502 if (ret < 0 || ret < sizeof(struct ust_register_msg)) {
1503 if (ret < 0) {
76d7553f 1504 PERROR("lttcomm_recv_unix_sock register apps");
5eb91c98
DG
1505 } else {
1506 ERR("Wrong size received on apps register");
1507 }
1508 free(ust_cmd);
76d7553f
MD
1509 ret = close(sock);
1510 if (ret) {
1511 PERROR("close");
1512 }
4063050c 1513 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 1514 sock = -1;
5eb91c98
DG
1515 continue;
1516 }
139ac872 1517 health_code_update(&health_thread_app_reg);
099e26bd 1518
5eb91c98 1519 ust_cmd->sock = sock;
34a2494f 1520 sock = -1;
099e26bd 1521
5eb91c98
DG
1522 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1523 " gid:%d sock:%d name:%s (version %d.%d)",
1524 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1525 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1526 ust_cmd->sock, ust_cmd->reg_msg.name,
1527 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 1528
5eb91c98
DG
1529 /*
1530 * Lock free enqueue the registration request. The red pill
54d01ffb 1531 * has been taken! This apps will be part of the *system*.
5eb91c98
DG
1532 */
1533 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1534
1535 /*
1536 * Wake the registration queue futex. Implicit memory
1537 * barrier with the exchange in cds_wfq_enqueue.
1538 */
1539 futex_nto1_wake(&ust_cmd_queue.futex);
1540 }
1541 }
90014c57 1542 }
1d4b027a
DG
1543 }
1544
139ac872 1545exit:
1d4b027a 1546error:
139ac872
MD
1547 if (err) {
1548 health_error(&health_thread_app_reg);
1549 ERR("Health error occurred in %s", __func__);
1550 }
139ac872 1551
0fdd1e2c
DG
1552 /* Notify that the registration thread is gone */
1553 notify_ust_apps(0);
1554
a4b35e07 1555 if (apps_sock >= 0) {
76d7553f
MD
1556 ret = close(apps_sock);
1557 if (ret) {
1558 PERROR("close");
1559 }
a4b35e07 1560 }
46c3f085 1561 if (sock >= 0) {
76d7553f
MD
1562 ret = close(sock);
1563 if (ret) {
1564 PERROR("close");
1565 }
4063050c 1566 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 1567 }
273ea72c 1568 unlink(apps_unix_sock_path);
0fdd1e2c 1569
76d7553f 1570error_poll_add:
5eb91c98 1571 lttng_poll_clean(&events);
76d7553f
MD
1572error_listen:
1573error_create_poll:
6993eeb3 1574error_testpoint:
76d7553f 1575 DBG("UST Registration thread cleanup complete");
152413a0 1576 health_exit(&health_thread_app_reg);
5eb91c98 1577
1d4b027a
DG
1578 return NULL;
1579}
1580
8c0faa1d 1581/*
3bd1e081 1582 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 1583 * exec or it will fails.
8c0faa1d 1584 */
3bd1e081 1585static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 1586{
a23ec3a7 1587 int ret, clock_ret;
ee0b0061
DG
1588 struct timespec timeout;
1589
a23ec3a7
DG
1590 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1591 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 1592
a23ec3a7
DG
1593 /* Setup pthread condition */
1594 ret = pthread_condattr_init(&consumer_data->condattr);
1595 if (ret != 0) {
1596 errno = ret;
1597 PERROR("pthread_condattr_init consumer data");
1598 goto error;
1599 }
1600
1601 /*
1602 * Set the monotonic clock in order to make sure we DO NOT jump in time
1603 * between the clock_gettime() call and the timedwait call. See bug #324
1604 * for a more details and how we noticed it.
1605 */
1606 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1607 if (ret != 0) {
1608 errno = ret;
1609 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
1610 goto error;
1611 }
8c0faa1d 1612
a23ec3a7
DG
1613 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1614 if (ret != 0) {
1615 errno = ret;
1616 PERROR("pthread_cond_init consumer data");
1617 goto error;
1618 }
1619
1620 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1621 consumer_data);
8c0faa1d 1622 if (ret != 0) {
3bd1e081 1623 PERROR("pthread_create consumer");
ee0b0061 1624 ret = -1;
8c0faa1d
DG
1625 goto error;
1626 }
1627
a23ec3a7
DG
1628 /* We are about to wait on a pthread condition */
1629 pthread_mutex_lock(&consumer_data->cond_mutex);
1630
ee0b0061 1631 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
1632 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1633 /*
1634 * Set the timeout for the condition timed wait even if the clock gettime
1635 * call fails since we might loop on that call and we want to avoid to
1636 * increment the timeout too many times.
1637 */
1638 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1639
1640 /*
1641 * The following loop COULD be skipped in some conditions so this is why we
1642 * set ret to 0 in order to make sure at least one round of the loop is
1643 * done.
1644 */
1645 ret = 0;
1646
1647 /*
1648 * Loop until the condition is reached or when a timeout is reached. Note
1649 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1650 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1651 * possible. This loop does not take any chances and works with both of
1652 * them.
1653 */
1654 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1655 if (clock_ret < 0) {
1656 PERROR("clock_gettime spawn consumer");
1657 /* Infinite wait for the consumerd thread to be ready */
1658 ret = pthread_cond_wait(&consumer_data->cond,
1659 &consumer_data->cond_mutex);
1660 } else {
1661 ret = pthread_cond_timedwait(&consumer_data->cond,
1662 &consumer_data->cond_mutex, &timeout);
1663 }
ee0b0061 1664 }
8c0faa1d 1665
a23ec3a7
DG
1666 /* Release the pthread condition */
1667 pthread_mutex_unlock(&consumer_data->cond_mutex);
1668
1669 if (ret != 0) {
1670 errno = ret;
1671 if (ret == ETIMEDOUT) {
ee0b0061
DG
1672 /*
1673 * Call has timed out so we kill the kconsumerd_thread and return
1674 * an error.
1675 */
a23ec3a7
DG
1676 ERR("Condition timed out. The consumer thread was never ready."
1677 " Killing it");
3bd1e081 1678 ret = pthread_cancel(consumer_data->thread);
ee0b0061 1679 if (ret < 0) {
3bd1e081 1680 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
1681 }
1682 } else {
a23ec3a7 1683 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061
DG
1684 }
1685 goto error;
1686 }
1687
3bd1e081
MD
1688 pthread_mutex_lock(&consumer_data->pid_mutex);
1689 if (consumer_data->pid == 0) {
a23ec3a7 1690 ERR("Consumerd did not start");
3bd1e081 1691 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
1692 goto error;
1693 }
3bd1e081 1694 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 1695
8c0faa1d
DG
1696 return 0;
1697
1698error:
1699 return ret;
1700}
1701
d9800920 1702/*
3bd1e081 1703 * Join consumer thread
d9800920 1704 */
3bd1e081 1705static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
1706{
1707 void *status;
cf3af59e 1708
e8209f6b
DG
1709 /* Consumer pid must be a real one. */
1710 if (consumer_data->pid > 0) {
c617c0c6 1711 int ret;
3bd1e081 1712 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 1713 if (ret) {
3bd1e081 1714 ERR("Error killing consumer daemon");
cf3af59e
MD
1715 return ret;
1716 }
3bd1e081 1717 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
1718 } else {
1719 return 0;
1720 }
1721}
1722
8c0faa1d 1723/*
3bd1e081 1724 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 1725 *
d063d709 1726 * Return pid if successful else -1.
8c0faa1d 1727 */
3bd1e081 1728static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
1729{
1730 int ret;
1731 pid_t pid;
94c55f17 1732 const char *consumer_to_use;
53086306 1733 const char *verbosity;
94c55f17 1734 struct stat st;
8c0faa1d 1735
3bd1e081 1736 DBG("Spawning consumerd");
c49dc785 1737
8c0faa1d
DG
1738 pid = fork();
1739 if (pid == 0) {
1740 /*
3bd1e081 1741 * Exec consumerd.
8c0faa1d 1742 */
daee5345 1743 if (opt_verbose_consumer) {
53086306
DG
1744 verbosity = "--verbose";
1745 } else {
1746 verbosity = "--quiet";
1747 }
3bd1e081
MD
1748 switch (consumer_data->type) {
1749 case LTTNG_CONSUMER_KERNEL:
94c55f17 1750 /*
c7704d57
DG
1751 * Find out which consumerd to execute. We will first try the
1752 * 64-bit path, then the sessiond's installation directory, and
1753 * fallback on the 32-bit one,
94c55f17 1754 */
63a799e8
AM
1755 DBG3("Looking for a kernel consumer at these locations:");
1756 DBG3(" 1) %s", consumerd64_bin);
1757 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
1758 DBG3(" 3) %s", consumerd32_bin);
94c55f17 1759 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 1760 DBG3("Found location #1");
94c55f17 1761 consumer_to_use = consumerd64_bin;
94c55f17 1762 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 1763 DBG3("Found location #2");
94c55f17 1764 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 1765 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 1766 DBG3("Found location #3");
eb1e0bd4 1767 consumer_to_use = consumerd32_bin;
94c55f17 1768 } else {
63a799e8 1769 DBG("Could not find any valid consumerd executable");
94c55f17
AM
1770 break;
1771 }
1772 DBG("Using kernel consumer at: %s", consumer_to_use);
1773 execl(consumer_to_use,
1774 "lttng-consumerd", verbosity, "-k",
1775 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1776 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1777 NULL);
3bd1e081 1778 break;
7753dea8
MD
1779 case LTTNG_CONSUMER64_UST:
1780 {
b1e0b6b6 1781 char *tmpnew = NULL;
8f4905da
MD
1782
1783 if (consumerd64_libdir[0] != '\0') {
1784 char *tmp;
1785 size_t tmplen;
1786
1787 tmp = getenv("LD_LIBRARY_PATH");
1788 if (!tmp) {
1789 tmp = "";
1790 }
1791 tmplen = strlen("LD_LIBRARY_PATH=")
1792 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
1793 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1794 if (!tmpnew) {
1795 ret = -ENOMEM;
1796 goto error;
1797 }
1798 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1799 strcat(tmpnew, consumerd64_libdir);
1800 if (tmp[0] != '\0') {
1801 strcat(tmpnew, ":");
1802 strcat(tmpnew, tmp);
1803 }
1804 ret = putenv(tmpnew);
1805 if (ret) {
1806 ret = -errno;
1807 goto error;
1808 }
1809 }
94c55f17 1810 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 1811 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
1812 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1813 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1814 NULL);
8f4905da
MD
1815 if (consumerd64_libdir[0] != '\0') {
1816 free(tmpnew);
1817 }
1818 if (ret) {
1819 goto error;
1820 }
3bd1e081 1821 break;
7753dea8
MD
1822 }
1823 case LTTNG_CONSUMER32_UST:
1824 {
937dde8e 1825 char *tmpnew = NULL;
8f4905da
MD
1826
1827 if (consumerd32_libdir[0] != '\0') {
1828 char *tmp;
1829 size_t tmplen;
1830
1831 tmp = getenv("LD_LIBRARY_PATH");
1832 if (!tmp) {
1833 tmp = "";
1834 }
1835 tmplen = strlen("LD_LIBRARY_PATH=")
1836 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
1837 tmpnew = zmalloc(tmplen + 1 /* \0 */);
1838 if (!tmpnew) {
1839 ret = -ENOMEM;
1840 goto error;
1841 }
1842 strcpy(tmpnew, "LD_LIBRARY_PATH=");
1843 strcat(tmpnew, consumerd32_libdir);
1844 if (tmp[0] != '\0') {
1845 strcat(tmpnew, ":");
1846 strcat(tmpnew, tmp);
1847 }
1848 ret = putenv(tmpnew);
1849 if (ret) {
1850 ret = -errno;
1851 goto error;
1852 }
1853 }
94c55f17 1854 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 1855 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
1856 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
1857 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
1858 NULL);
8f4905da
MD
1859 if (consumerd32_libdir[0] != '\0') {
1860 free(tmpnew);
1861 }
1862 if (ret) {
1863 goto error;
1864 }
7753dea8
MD
1865 break;
1866 }
3bd1e081 1867 default:
76d7553f 1868 PERROR("unknown consumer type");
3bd1e081
MD
1869 exit(EXIT_FAILURE);
1870 }
8c0faa1d 1871 if (errno != 0) {
76d7553f 1872 PERROR("kernel start consumer exec");
8c0faa1d
DG
1873 }
1874 exit(EXIT_FAILURE);
1875 } else if (pid > 0) {
1876 ret = pid;
8c0faa1d 1877 } else {
76d7553f 1878 PERROR("start consumer fork");
8c0faa1d 1879 ret = -errno;
8c0faa1d 1880 }
8f4905da 1881error:
8c0faa1d
DG
1882 return ret;
1883}
1884
693bd40b 1885/*
3bd1e081 1886 * Spawn the consumerd daemon and session daemon thread.
693bd40b 1887 */
3bd1e081 1888static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 1889{
c617c0c6 1890 int ret;
edb8b045
DG
1891
1892 /*
1893 * Set the listen() state on the socket since there is a possible race
1894 * between the exec() of the consumer daemon and this call if place in the
1895 * consumer thread. See bug #366 for more details.
1896 */
1897 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
1898 if (ret < 0) {
1899 goto error;
1900 }
693bd40b 1901
3bd1e081
MD
1902 pthread_mutex_lock(&consumer_data->pid_mutex);
1903 if (consumer_data->pid != 0) {
1904 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
1905 goto end;
1906 }
693bd40b 1907
3bd1e081 1908 ret = spawn_consumerd(consumer_data);
c49dc785 1909 if (ret < 0) {
3bd1e081
MD
1910 ERR("Spawning consumerd failed");
1911 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 1912 goto error;
693bd40b 1913 }
c49dc785 1914
3bd1e081
MD
1915 /* Setting up the consumer_data pid */
1916 consumer_data->pid = ret;
48842b30 1917 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 1918 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 1919
3bd1e081
MD
1920 DBG2("Spawning consumer control thread");
1921 ret = spawn_consumer_thread(consumer_data);
693bd40b 1922 if (ret < 0) {
3bd1e081 1923 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
1924 goto error;
1925 }
1926
c49dc785 1927end:
693bd40b
DG
1928 return 0;
1929
1930error:
edb8b045
DG
1931 /* Cleanup already created socket on error. */
1932 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
1933 int err;
1934
edb8b045
DG
1935 err = close(consumer_data->err_sock);
1936 if (err < 0) {
1937 PERROR("close consumer data error socket");
1938 }
1939 }
693bd40b
DG
1940 return ret;
1941}
1942
44a5e5eb 1943/*
139ac872
MD
1944 * Compute health status of each consumer. If one of them is zero (bad
1945 * state), we return 0.
44a5e5eb
DG
1946 */
1947static int check_consumer_health(void)
1948{
1949 int ret;
1950
139ac872
MD
1951 ret = health_check_state(&kconsumer_data.health) &&
1952 health_check_state(&ustconsumer32_data.health) &&
44a5e5eb
DG
1953 health_check_state(&ustconsumer64_data.health);
1954
1955 DBG3("Health consumer check %d", ret);
1956
1957 return ret;
1958}
1959
b73401da 1960/*
096102bd 1961 * Setup necessary data for kernel tracer action.
b73401da 1962 */
096102bd 1963static int init_kernel_tracer(void)
b73401da
DG
1964{
1965 int ret;
b73401da 1966
096102bd
DG
1967 /* Modprobe lttng kernel modules */
1968 ret = modprobe_lttng_control();
b73401da 1969 if (ret < 0) {
b73401da
DG
1970 goto error;
1971 }
1972
096102bd
DG
1973 /* Open debugfs lttng */
1974 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1975 if (kernel_tracer_fd < 0) {
1976 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
1977 ret = -1;
1978 goto error_open;
54d01ffb
DG
1979 }
1980
2f77fc4b
DG
1981 /* Validate kernel version */
1982 ret = kernel_validate_version(kernel_tracer_fd);
1983 if (ret < 0) {
1984 goto error_version;
b551a063 1985 }
54d01ffb 1986
2f77fc4b
DG
1987 ret = modprobe_lttng_data();
1988 if (ret < 0) {
1989 goto error_modules;
54d01ffb
DG
1990 }
1991
2f77fc4b
DG
1992 DBG("Kernel tracer fd %d", kernel_tracer_fd);
1993 return 0;
1994
1995error_version:
1996 modprobe_remove_lttng_control();
1997 ret = close(kernel_tracer_fd);
1998 if (ret) {
1999 PERROR("close");
b551a063 2000 }
2f77fc4b 2001 kernel_tracer_fd = -1;
f73fabfd 2002 return LTTNG_ERR_KERN_VERSION;
b551a063 2003
2f77fc4b
DG
2004error_modules:
2005 ret = close(kernel_tracer_fd);
2006 if (ret) {
2007 PERROR("close");
b551a063 2008 }
54d01ffb 2009
2f77fc4b
DG
2010error_open:
2011 modprobe_remove_lttng_control();
54d01ffb
DG
2012
2013error:
2f77fc4b
DG
2014 WARN("No kernel tracer available");
2015 kernel_tracer_fd = -1;
2016 if (!is_root) {
f73fabfd 2017 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2018 } else {
f73fabfd 2019 return LTTNG_ERR_KERN_NA;
2f77fc4b 2020 }
54d01ffb
DG
2021}
2022
2f77fc4b 2023
54d01ffb 2024/*
2f77fc4b
DG
2025 * Copy consumer output from the tracing session to the domain session. The
2026 * function also applies the right modification on a per domain basis for the
2027 * trace files destination directory.
54d01ffb 2028 */
2f77fc4b 2029static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2030{
2031 int ret;
2f77fc4b
DG
2032 const char *dir_name;
2033 struct consumer_output *consumer;
2034
2035 assert(session);
2036 assert(session->consumer);
54d01ffb 2037
b551a063
DG
2038 switch (domain) {
2039 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2040 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2041 /*
2042 * XXX: We should audit the session creation and what this function
2043 * does "extra" in order to avoid a destroy since this function is used
2044 * in the domain session creation (kernel and ust) only. Same for UST
2045 * domain.
2046 */
2047 if (session->kernel_session->consumer) {
2048 consumer_destroy_output(session->kernel_session->consumer);
2049 }
2f77fc4b
DG
2050 session->kernel_session->consumer =
2051 consumer_copy_output(session->consumer);
2052 /* Ease our life a bit for the next part */
2053 consumer = session->kernel_session->consumer;
2054 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063
DG
2055 break;
2056 case LTTNG_DOMAIN_UST:
2f77fc4b 2057 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2058 if (session->ust_session->consumer) {
2059 consumer_destroy_output(session->ust_session->consumer);
2060 }
2f77fc4b
DG
2061 session->ust_session->consumer =
2062 consumer_copy_output(session->consumer);
2063 /* Ease our life a bit for the next part */
2064 consumer = session->ust_session->consumer;
2065 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2066 break;
2067 default:
f73fabfd 2068 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2069 goto error;
2070 }
2071
2f77fc4b 2072 /* Append correct directory to subdir */
c30ce0b3
CB
2073 strncat(consumer->subdir, dir_name,
2074 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2075 DBG3("Copy session consumer subdir %s", consumer->subdir);
2076
f73fabfd 2077 ret = LTTNG_OK;
54d01ffb
DG
2078
2079error:
2080 return ret;
2081}
2082
00e2e675 2083/*
2f77fc4b 2084 * Create an UST session and add it to the session ust list.
00e2e675 2085 */
2f77fc4b
DG
2086static int create_ust_session(struct ltt_session *session,
2087 struct lttng_domain *domain)
00e2e675
DG
2088{
2089 int ret;
2f77fc4b 2090 struct ltt_ust_session *lus = NULL;
00e2e675 2091
a4b92340 2092 assert(session);
2f77fc4b
DG
2093 assert(domain);
2094 assert(session->consumer);
a4b92340 2095
2f77fc4b
DG
2096 switch (domain->type) {
2097 case LTTNG_DOMAIN_UST:
2098 break;
2099 default:
2100 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2101 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2102 goto error;
2103 }
2104
2f77fc4b
DG
2105 DBG("Creating UST session");
2106
2107 lus = trace_ust_create_session(session->path, session->id, domain);
2108 if (lus == NULL) {
f73fabfd 2109 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2110 goto error;
2111 }
2112
2f77fc4b
DG
2113 lus->uid = session->uid;
2114 lus->gid = session->gid;
2115 session->ust_session = lus;
00e2e675 2116
2f77fc4b
DG
2117 /* Copy session output to the newly created UST session */
2118 ret = copy_session_consumer(domain->type, session);
f73fabfd 2119 if (ret != LTTNG_OK) {
00e2e675
DG
2120 goto error;
2121 }
2122
f73fabfd 2123 return LTTNG_OK;
00e2e675
DG
2124
2125error:
2f77fc4b
DG
2126 free(lus);
2127 session->ust_session = NULL;
00e2e675
DG
2128 return ret;
2129}
2130
2131/*
2f77fc4b 2132 * Create a kernel tracer session then create the default channel.
00e2e675 2133 */
2f77fc4b 2134static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2135{
2136 int ret;
a4b92340 2137
2f77fc4b 2138 DBG("Creating kernel session");
00e2e675 2139
2f77fc4b
DG
2140 ret = kernel_create_session(session, kernel_tracer_fd);
2141 if (ret < 0) {
f73fabfd 2142 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2143 goto error;
2144 }
2145
2f77fc4b
DG
2146 /* Code flow safety */
2147 assert(session->kernel_session);
2148
2149 /* Copy session output to the newly created Kernel session */
2150 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2151 if (ret != LTTNG_OK) {
a4b92340
DG
2152 goto error;
2153 }
2154
2f77fc4b
DG
2155 /* Create directory(ies) on local filesystem. */
2156 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2157 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2158 ret = run_as_mkdir_recursive(
2159 session->kernel_session->consumer->dst.trace_path,
2160 S_IRWXU | S_IRWXG, session->uid, session->gid);
2161 if (ret < 0) {
2162 if (ret != -EEXIST) {
2163 ERR("Trace directory creation error");
00e2e675
DG
2164 goto error;
2165 }
00e2e675 2166 }
2f77fc4b 2167 }
00e2e675 2168
2f77fc4b
DG
2169 session->kernel_session->uid = session->uid;
2170 session->kernel_session->gid = session->gid;
00e2e675 2171
f73fabfd 2172 return LTTNG_OK;
00e2e675 2173
2f77fc4b
DG
2174error:
2175 trace_kernel_destroy_session(session->kernel_session);
2176 session->kernel_session = NULL;
2177 return ret;
2178}
00e2e675 2179
2f77fc4b
DG
2180/*
2181 * Count number of session permitted by uid/gid.
2182 */
2183static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2184{
2185 unsigned int i = 0;
2186 struct ltt_session *session;
07424f16 2187
2f77fc4b
DG
2188 DBG("Counting number of available session for UID %d GID %d",
2189 uid, gid);
2190 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2191 /*
2f77fc4b 2192 * Only list the sessions the user can control.
00e2e675 2193 */
2f77fc4b
DG
2194 if (!session_access_ok(session, uid, gid)) {
2195 continue;
2196 }
2197 i++;
a4b92340 2198 }
2f77fc4b 2199 return i;
00e2e675
DG
2200}
2201
54d01ffb
DG
2202/*
2203 * Process the command requested by the lttng client within the command
2204 * context structure. This function make sure that the return structure (llm)
2205 * is set and ready for transmission before returning.
2206 *
2207 * Return any error encountered or 0 for success.
53a80697
MD
2208 *
2209 * "sock" is only used for special-case var. len data.
54d01ffb 2210 */
53a80697
MD
2211static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2212 int *sock_error)
54d01ffb 2213{
f73fabfd 2214 int ret = LTTNG_OK;
44d3bd01 2215 int need_tracing_session = 1;
2e09ba09 2216 int need_domain;
54d01ffb
DG
2217
2218 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2219
53a80697
MD
2220 *sock_error = 0;
2221
2e09ba09
MD
2222 switch (cmd_ctx->lsm->cmd_type) {
2223 case LTTNG_CREATE_SESSION:
2224 case LTTNG_DESTROY_SESSION:
2225 case LTTNG_LIST_SESSIONS:
2226 case LTTNG_LIST_DOMAINS:
2227 case LTTNG_START_TRACE:
2228 case LTTNG_STOP_TRACE:
6d805429 2229 case LTTNG_DATA_PENDING:
2e09ba09 2230 need_domain = 0;
3aace903 2231 break;
2e09ba09
MD
2232 default:
2233 need_domain = 1;
2234 }
2235
2236 if (opt_no_kernel && need_domain
2237 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2238 if (!is_root) {
f73fabfd 2239 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2240 } else {
f73fabfd 2241 ret = LTTNG_ERR_KERN_NA;
531d29f9 2242 }
4fba7219
DG
2243 goto error;
2244 }
2245
8d3113b2
DG
2246 /* Deny register consumer if we already have a spawned consumer. */
2247 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2248 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2249 if (kconsumer_data.pid > 0) {
f73fabfd 2250 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2251 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2252 goto error;
2253 }
2254 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2255 }
2256
54d01ffb
DG
2257 /*
2258 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2259 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2260 * command.
2261 */
2262 switch(cmd_ctx->lsm->cmd_type) {
2263 case LTTNG_LIST_SESSIONS:
2264 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2265 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2266 case LTTNG_LIST_DOMAINS:
2267 case LTTNG_LIST_CHANNELS:
2268 case LTTNG_LIST_EVENTS:
2269 break;
2270 default:
2271 /* Setup lttng message with no payload */
2272 ret = setup_lttng_msg(cmd_ctx, 0);
2273 if (ret < 0) {
2274 /* This label does not try to unlock the session */
2275 goto init_setup_error;
2276 }
2277 }
2278
2279 /* Commands that DO NOT need a session. */
2280 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2281 case LTTNG_CREATE_SESSION:
2e09ba09 2282 case LTTNG_CALIBRATE:
54d01ffb
DG
2283 case LTTNG_LIST_SESSIONS:
2284 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2285 case LTTNG_LIST_TRACEPOINT_FIELDS:
44d3bd01 2286 need_tracing_session = 0;
54d01ffb
DG
2287 break;
2288 default:
2289 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2290 /*
2291 * We keep the session list lock across _all_ commands
2292 * for now, because the per-session lock does not
2293 * handle teardown properly.
2294 */
74babd95 2295 session_lock_list();
54d01ffb
DG
2296 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2297 if (cmd_ctx->session == NULL) {
2298 if (cmd_ctx->lsm->session.name != NULL) {
f73fabfd 2299 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2300 } else {
2301 /* If no session name specified */
f73fabfd 2302 ret = LTTNG_ERR_SELECT_SESS;
54d01ffb
DG
2303 }
2304 goto error;
2305 } else {
2306 /* Acquire lock for the session */
2307 session_lock(cmd_ctx->session);
2308 }
2309 break;
2310 }
b389abbe 2311
2e09ba09
MD
2312 if (!need_domain) {
2313 goto skip_domain;
2314 }
a4b92340 2315
54d01ffb
DG
2316 /*
2317 * Check domain type for specific "pre-action".
2318 */
2319 switch (cmd_ctx->lsm->domain.type) {
2320 case LTTNG_DOMAIN_KERNEL:
d1f1c568 2321 if (!is_root) {
f73fabfd 2322 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
2323 goto error;
2324 }
2325
54d01ffb 2326 /* Kernel tracer check */
a4b35e07 2327 if (kernel_tracer_fd == -1) {
54d01ffb 2328 /* Basically, load kernel tracer modules */
096102bd
DG
2329 ret = init_kernel_tracer();
2330 if (ret != 0) {
54d01ffb
DG
2331 goto error;
2332 }
2333 }
5eb91c98 2334
5c827ce0
DG
2335 /* Consumer is in an ERROR state. Report back to client */
2336 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2337 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2338 goto error;
2339 }
2340
54d01ffb 2341 /* Need a session for kernel command */
44d3bd01 2342 if (need_tracing_session) {
54d01ffb 2343 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 2344 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 2345 if (ret < 0) {
f73fabfd 2346 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
2347 goto error;
2348 }
b389abbe 2349 }
7d29a247 2350
54d01ffb 2351 /* Start the kernel consumer daemon */
3bd1e081
MD
2352 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2353 if (kconsumer_data.pid == 0 &&
a4b92340
DG
2354 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2355 cmd_ctx->session->start_consumer) {
3bd1e081
MD
2356 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2357 ret = start_consumerd(&kconsumer_data);
7d29a247 2358 if (ret < 0) {
f73fabfd 2359 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 2360 goto error;
950131af 2361 }
5c827ce0 2362 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
2363 } else {
2364 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 2365 }
173af62f 2366
a4b92340
DG
2367 /*
2368 * The consumer was just spawned so we need to add the socket to
2369 * the consumer output of the session if exist.
2370 */
2371 ret = consumer_create_socket(&kconsumer_data,
2372 cmd_ctx->session->kernel_session->consumer);
2373 if (ret < 0) {
2374 goto error;
173af62f 2375 }
0d0c377a 2376 }
5c827ce0 2377
54d01ffb 2378 break;
2bdd86d4 2379 case LTTNG_DOMAIN_UST:
44d3bd01 2380 {
5c827ce0
DG
2381 /* Consumer is in an ERROR state. Report back to client */
2382 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2383 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2384 goto error;
2385 }
2386
44d3bd01 2387 if (need_tracing_session) {
a4b92340 2388 /* Create UST session if none exist. */
f6a9efaa 2389 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 2390 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 2391 &cmd_ctx->lsm->domain);
f73fabfd 2392 if (ret != LTTNG_OK) {
44d3bd01
DG
2393 goto error;
2394 }
2395 }
00e2e675 2396
7753dea8
MD
2397 /* Start the UST consumer daemons */
2398 /* 64-bit */
2399 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 2400 if (consumerd64_bin[0] != '\0' &&
7753dea8 2401 ustconsumer64_data.pid == 0 &&
a4b92340
DG
2402 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2403 cmd_ctx->session->start_consumer) {
7753dea8
MD
2404 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2405 ret = start_consumerd(&ustconsumer64_data);
2bdd86d4 2406 if (ret < 0) {
f73fabfd 2407 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
173af62f 2408 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2bdd86d4
MD
2409 goto error;
2410 }
48842b30 2411
173af62f 2412 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
5c827ce0 2413 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3ff2ecac 2414 } else {
7753dea8
MD
2415 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2416 }
173af62f
DG
2417
2418 /*
2419 * Setup socket for consumer 64 bit. No need for atomic access
2420 * since it was set above and can ONLY be set in this thread.
2421 */
a4b92340
DG
2422 ret = consumer_create_socket(&ustconsumer64_data,
2423 cmd_ctx->session->ust_session->consumer);
2424 if (ret < 0) {
2425 goto error;
173af62f
DG
2426 }
2427
7753dea8 2428 /* 32-bit */
fc7a59ce 2429 if (consumerd32_bin[0] != '\0' &&
7753dea8 2430 ustconsumer32_data.pid == 0 &&
a4b92340
DG
2431 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER &&
2432 cmd_ctx->session->start_consumer) {
7753dea8
MD
2433 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2434 ret = start_consumerd(&ustconsumer32_data);
2435 if (ret < 0) {
f73fabfd 2436 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
173af62f 2437 uatomic_set(&ust_consumerd32_fd, -EINVAL);
7753dea8
MD
2438 goto error;
2439 }
5c827ce0 2440
173af62f 2441 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
5c827ce0 2442 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
7753dea8
MD
2443 } else {
2444 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2bdd86d4 2445 }
173af62f
DG
2446
2447 /*
2448 * Setup socket for consumer 64 bit. No need for atomic access
2449 * since it was set above and can ONLY be set in this thread.
2450 */
a4b92340
DG
2451 ret = consumer_create_socket(&ustconsumer32_data,
2452 cmd_ctx->session->ust_session->consumer);
2453 if (ret < 0) {
2454 goto error;
173af62f 2455 }
44d3bd01
DG
2456 }
2457 break;
48842b30 2458 }
54d01ffb 2459 default:
54d01ffb
DG
2460 break;
2461 }
2e09ba09 2462skip_domain:
33a2b854 2463
5c827ce0
DG
2464 /* Validate consumer daemon state when start/stop trace command */
2465 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2466 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2467 switch (cmd_ctx->lsm->domain.type) {
2468 case LTTNG_DOMAIN_UST:
2469 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2470 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2471 goto error;
2472 }
2473 break;
2474 case LTTNG_DOMAIN_KERNEL:
2475 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2476 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2477 goto error;
2478 }
2479 break;
2480 }
2481 }
2482
8e0af1b4
MD
2483 /*
2484 * Check that the UID or GID match that of the tracing session.
2485 * The root user can interact with all sessions.
2486 */
2487 if (need_tracing_session) {
2488 if (!session_access_ok(cmd_ctx->session,
730389d9
DG
2489 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2490 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
f73fabfd 2491 ret = LTTNG_ERR_EPERM;
8e0af1b4
MD
2492 goto error;
2493 }
2494 }
2495
54d01ffb
DG
2496 /* Process by command type */
2497 switch (cmd_ctx->lsm->cmd_type) {
2498 case LTTNG_ADD_CONTEXT:
2499 {
2500 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2501 cmd_ctx->lsm->u.context.channel_name,
979e618e 2502 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
54d01ffb
DG
2503 break;
2504 }
2505 case LTTNG_DISABLE_CHANNEL:
2506 {
2507 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2508 cmd_ctx->lsm->u.disable.channel_name);
2509 break;
2510 }
2511 case LTTNG_DISABLE_EVENT:
2512 {
2513 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2514 cmd_ctx->lsm->u.disable.channel_name,
2515 cmd_ctx->lsm->u.disable.name);
33a2b854
DG
2516 break;
2517 }
54d01ffb
DG
2518 case LTTNG_DISABLE_ALL_EVENT:
2519 {
2bdd86d4 2520 DBG("Disabling all events");
54d01ffb
DG
2521
2522 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2523 cmd_ctx->lsm->u.disable.channel_name);
2524 break;
2525 }
00e2e675
DG
2526 case LTTNG_DISABLE_CONSUMER:
2527 {
2528 ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
2529 break;
2530 }
54d01ffb
DG
2531 case LTTNG_ENABLE_CHANNEL:
2532 {
aea829b3 2533 ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 2534 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
54d01ffb
DG
2535 break;
2536 }
00e2e675
DG
2537 case LTTNG_ENABLE_CONSUMER:
2538 {
a4b92340
DG
2539 /*
2540 * XXX: 0 means that this URI should be applied on the session. Should
2541 * be a DOMAIN enuam.
2542 */
00e2e675 2543 ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session);
f73fabfd 2544 if (ret != LTTNG_OK) {
a4b92340
DG
2545 goto error;
2546 }
2547
2548 if (cmd_ctx->lsm->domain.type == 0) {
2549 /* Add the URI for the UST session if a consumer is present. */
2550 if (cmd_ctx->session->ust_session &&
2551 cmd_ctx->session->ust_session->consumer) {
2552 ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session);
2553 } else if (cmd_ctx->session->kernel_session &&
2554 cmd_ctx->session->kernel_session->consumer) {
2555 ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL,
2556 cmd_ctx->session);
2557 }
2558 }
00e2e675
DG
2559 break;
2560 }
54d01ffb
DG
2561 case LTTNG_ENABLE_EVENT:
2562 {
2563 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2564 cmd_ctx->lsm->u.enable.channel_name,
025faf73 2565 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
2566 break;
2567 }
2568 case LTTNG_ENABLE_ALL_EVENT:
2569 {
2bdd86d4 2570 DBG("Enabling all events");
54d01ffb
DG
2571
2572 ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
8c9ae521 2573 cmd_ctx->lsm->u.enable.channel_name,
025faf73 2574 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
2575 break;
2576 }
052da939 2577 case LTTNG_LIST_TRACEPOINTS:
2ef84c95 2578 {
9f19cc17 2579 struct lttng_event *events;
54d01ffb 2580 ssize_t nb_events;
052da939 2581
54d01ffb
DG
2582 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2583 if (nb_events < 0) {
f73fabfd 2584 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2585 ret = -nb_events;
2586 goto error;
2ef84c95
DG
2587 }
2588
2589 /*
2590 * Setup lttng message with payload size set to the event list size in
2591 * bytes and then copy list into the llm payload.
2592 */
052da939 2593 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2ef84c95 2594 if (ret < 0) {
052da939 2595 free(events);
2ef84c95
DG
2596 goto setup_error;
2597 }
2598
2599 /* Copy event list into message payload */
9f19cc17 2600 memcpy(cmd_ctx->llm->payload, events,
052da939 2601 sizeof(struct lttng_event) * nb_events);
2ef84c95 2602
9f19cc17 2603 free(events);
2ef84c95 2604
f73fabfd 2605 ret = LTTNG_OK;
2ef84c95
DG
2606 break;
2607 }
f37d259d
MD
2608 case LTTNG_LIST_TRACEPOINT_FIELDS:
2609 {
2610 struct lttng_event_field *fields;
2611 ssize_t nb_fields;
2612
a4b92340
DG
2613 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2614 &fields);
f37d259d 2615 if (nb_fields < 0) {
f73fabfd 2616 /* Return value is a negative lttng_error_code. */
f37d259d
MD
2617 ret = -nb_fields;
2618 goto error;
2619 }
2620
2621 /*
2622 * Setup lttng message with payload size set to the event list size in
2623 * bytes and then copy list into the llm payload.
2624 */
a4b92340
DG
2625 ret = setup_lttng_msg(cmd_ctx,
2626 sizeof(struct lttng_event_field) * nb_fields);
f37d259d
MD
2627 if (ret < 0) {
2628 free(fields);
2629 goto setup_error;
2630 }
2631
2632 /* Copy event list into message payload */
2633 memcpy(cmd_ctx->llm->payload, fields,
2634 sizeof(struct lttng_event_field) * nb_fields);
2635
2636 free(fields);
2637
f73fabfd 2638 ret = LTTNG_OK;
f37d259d
MD
2639 break;
2640 }
00e2e675
DG
2641 case LTTNG_SET_CONSUMER_URI:
2642 {
a4b92340
DG
2643 size_t nb_uri, len;
2644 struct lttng_uri *uris;
2645
2646 nb_uri = cmd_ctx->lsm->u.uri.size;
2647 len = nb_uri * sizeof(struct lttng_uri);
2648
2649 if (nb_uri == 0) {
f73fabfd 2650 ret = LTTNG_ERR_INVALID;
a4b92340
DG
2651 goto error;
2652 }
2653
2654 uris = zmalloc(len);
2655 if (uris == NULL) {
f73fabfd 2656 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2657 goto error;
2658 }
2659
2660 /* Receive variable len data */
77c7c900 2661 DBG("Receiving %zu URI(s) from client ...", nb_uri);
a4b92340
DG
2662 ret = lttcomm_recv_unix_sock(sock, uris, len);
2663 if (ret <= 0) {
2664 DBG("No URIs received from client... continuing");
2665 *sock_error = 1;
f73fabfd 2666 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2667 free(uris);
a4b92340
DG
2668 goto error;
2669 }
2670
00e2e675 2671 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
a4b92340 2672 nb_uri, uris);
f73fabfd 2673 if (ret != LTTNG_OK) {
b1d41407 2674 free(uris);
a4b92340
DG
2675 goto error;
2676 }
2677
2678 /*
2679 * XXX: 0 means that this URI should be applied on the session. Should
2680 * be a DOMAIN enuam.
2681 */
2682 if (cmd_ctx->lsm->domain.type == 0) {
2683 /* Add the URI for the UST session if a consumer is present. */
2684 if (cmd_ctx->session->ust_session &&
2685 cmd_ctx->session->ust_session->consumer) {
2686 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2687 nb_uri, uris);
2688 } else if (cmd_ctx->session->kernel_session &&
2689 cmd_ctx->session->kernel_session->consumer) {
2690 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2691 cmd_ctx->session, nb_uri, uris);
2692 }
2693 }
2694
b1d41407
DG
2695 free(uris);
2696
00e2e675
DG
2697 break;
2698 }
f3ed775e 2699 case LTTNG_START_TRACE:
8c0faa1d 2700 {
54d01ffb 2701 ret = cmd_start_trace(cmd_ctx->session);
8c0faa1d
DG
2702 break;
2703 }
f3ed775e 2704 case LTTNG_STOP_TRACE:
8c0faa1d 2705 {
54d01ffb 2706 ret = cmd_stop_trace(cmd_ctx->session);
8c0faa1d
DG
2707 break;
2708 }
5e16da05
MD
2709 case LTTNG_CREATE_SESSION:
2710 {
a4b92340
DG
2711 size_t nb_uri, len;
2712 struct lttng_uri *uris = NULL;
2713
2714 nb_uri = cmd_ctx->lsm->u.uri.size;
2715 len = nb_uri * sizeof(struct lttng_uri);
2716
2717 if (nb_uri > 0) {
2718 uris = zmalloc(len);
2719 if (uris == NULL) {
f73fabfd 2720 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2721 goto error;
2722 }
2723
2724 /* Receive variable len data */
77c7c900 2725 DBG("Waiting for %zu URIs from client ...", nb_uri);
a4b92340
DG
2726 ret = lttcomm_recv_unix_sock(sock, uris, len);
2727 if (ret <= 0) {
2728 DBG("No URIs received from client... continuing");
2729 *sock_error = 1;
f73fabfd 2730 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2731 free(uris);
a4b92340
DG
2732 goto error;
2733 }
2734
2735 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
2736 DBG("Creating session with ONE network URI is a bad call");
f73fabfd 2737 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2738 free(uris);
a4b92340
DG
2739 goto error;
2740 }
2741 }
2742
2743 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
2744 &cmd_ctx->creds);
2745
b1d41407
DG
2746 free(uris);
2747
00e2e675
DG
2748 break;
2749 }
5e16da05
MD
2750 case LTTNG_DESTROY_SESSION:
2751 {
2f77fc4b 2752 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
a4b92340
DG
2753
2754 /* Set session to NULL so we do not unlock it after free. */
256a5576 2755 cmd_ctx->session = NULL;
5461b305 2756 break;
5e16da05 2757 }
9f19cc17 2758 case LTTNG_LIST_DOMAINS:
5e16da05 2759 {
54d01ffb
DG
2760 ssize_t nb_dom;
2761 struct lttng_domain *domains;
5461b305 2762
54d01ffb
DG
2763 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
2764 if (nb_dom < 0) {
f73fabfd 2765 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2766 ret = -nb_dom;
2767 goto error;
ce3d728c 2768 }
520ff687 2769
54d01ffb 2770 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
5461b305
DG
2771 if (ret < 0) {
2772 goto setup_error;
520ff687 2773 }
57167058 2774
54d01ffb
DG
2775 /* Copy event list into message payload */
2776 memcpy(cmd_ctx->llm->payload, domains,
2777 nb_dom * sizeof(struct lttng_domain));
2778
2779 free(domains);
5e16da05 2780
f73fabfd 2781 ret = LTTNG_OK;
5e16da05
MD
2782 break;
2783 }
9f19cc17 2784 case LTTNG_LIST_CHANNELS:
5e16da05 2785 {
6775595e 2786 int nb_chan;
54d01ffb
DG
2787 struct lttng_channel *channels;
2788
b551a063
DG
2789 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
2790 cmd_ctx->session, &channels);
54d01ffb 2791 if (nb_chan < 0) {
f73fabfd 2792 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2793 ret = -nb_chan;
2794 goto error;
5461b305 2795 }
ca95a216 2796
54d01ffb 2797 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
5461b305
DG
2798 if (ret < 0) {
2799 goto setup_error;
2800 }
9f19cc17 2801
54d01ffb
DG
2802 /* Copy event list into message payload */
2803 memcpy(cmd_ctx->llm->payload, channels,
2804 nb_chan * sizeof(struct lttng_channel));
2805
2806 free(channels);
9f19cc17 2807
f73fabfd 2808 ret = LTTNG_OK;
5461b305 2809 break;
5e16da05 2810 }
9f19cc17 2811 case LTTNG_LIST_EVENTS:
5e16da05 2812 {
b551a063 2813 ssize_t nb_event;
684d34d2 2814 struct lttng_event *events = NULL;
9f19cc17 2815
b551a063 2816 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
54d01ffb
DG
2817 cmd_ctx->lsm->u.list.channel_name, &events);
2818 if (nb_event < 0) {
f73fabfd 2819 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2820 ret = -nb_event;
2821 goto error;
5461b305 2822 }
ca95a216 2823
54d01ffb 2824 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
5461b305
DG
2825 if (ret < 0) {
2826 goto setup_error;
2827 }
9f19cc17 2828
54d01ffb
DG
2829 /* Copy event list into message payload */
2830 memcpy(cmd_ctx->llm->payload, events,
2831 nb_event * sizeof(struct lttng_event));
9f19cc17 2832
54d01ffb 2833 free(events);
9f19cc17 2834
f73fabfd 2835 ret = LTTNG_OK;
5461b305 2836 break;
5e16da05
MD
2837 }
2838 case LTTNG_LIST_SESSIONS:
2839 {
8e0af1b4 2840 unsigned int nr_sessions;
5461b305 2841
8e0af1b4 2842 session_lock_list();
730389d9
DG
2843 nr_sessions = lttng_sessions_count(
2844 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2845 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
d32fb093 2846
8e0af1b4 2847 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
5461b305 2848 if (ret < 0) {
54d01ffb 2849 session_unlock_list();
5461b305 2850 goto setup_error;
e065084a 2851 }
5e16da05 2852
6c9cc2ab 2853 /* Filled the session array */
2f77fc4b 2854 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
730389d9
DG
2855 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2856 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
6c9cc2ab 2857
54d01ffb 2858 session_unlock_list();
5e16da05 2859
f73fabfd 2860 ret = LTTNG_OK;
5e16da05
MD
2861 break;
2862 }
d0254c7c
MD
2863 case LTTNG_CALIBRATE:
2864 {
54d01ffb
DG
2865 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
2866 &cmd_ctx->lsm->u.calibrate);
d0254c7c
MD
2867 break;
2868 }
d9800920
DG
2869 case LTTNG_REGISTER_CONSUMER:
2870 {
2f77fc4b
DG
2871 struct consumer_data *cdata;
2872
2873 switch (cmd_ctx->lsm->domain.type) {
2874 case LTTNG_DOMAIN_KERNEL:
2875 cdata = &kconsumer_data;
2876 break;
2877 default:
f73fabfd 2878 ret = LTTNG_ERR_UND;
2f77fc4b
DG
2879 goto error;
2880 }
2881
54d01ffb 2882 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 2883 cmd_ctx->lsm->u.reg.path, cdata);
d9800920
DG
2884 break;
2885 }
025faf73 2886 case LTTNG_ENABLE_EVENT_WITH_FILTER:
53a80697
MD
2887 {
2888 struct lttng_filter_bytecode *bytecode;
2889
025faf73 2890 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
f73fabfd 2891 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2892 goto error;
2893 }
52df2401
MD
2894 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
2895 ret = LTTNG_ERR_FILTER_INVAL;
2896 goto error;
2897 }
025faf73 2898 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
53a80697 2899 if (!bytecode) {
f73fabfd 2900 ret = LTTNG_ERR_FILTER_NOMEM;
53a80697
MD
2901 goto error;
2902 }
2903 /* Receive var. len. data */
2904 DBG("Receiving var len data from client ...");
2905 ret = lttcomm_recv_unix_sock(sock, bytecode,
025faf73 2906 cmd_ctx->lsm->u.enable.bytecode_len);
53a80697
MD
2907 if (ret <= 0) {
2908 DBG("Nothing recv() from client var len data... continuing");
2909 *sock_error = 1;
f73fabfd 2910 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2911 goto error;
2912 }
2913
2914 if (bytecode->len + sizeof(*bytecode)
025faf73 2915 != cmd_ctx->lsm->u.enable.bytecode_len) {
53a80697 2916 free(bytecode);
f73fabfd 2917 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
2918 goto error;
2919 }
2920
025faf73
DG
2921 ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2922 cmd_ctx->lsm->u.enable.channel_name,
2923 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
53a80697
MD
2924 break;
2925 }
6d805429 2926 case LTTNG_DATA_PENDING:
806e2684 2927 {
6d805429 2928 ret = cmd_data_pending(cmd_ctx->session);
806e2684
DG
2929 break;
2930 }
5e16da05 2931 default:
f73fabfd 2932 ret = LTTNG_ERR_UND;
5461b305 2933 break;
fac6795d
DG
2934 }
2935
5461b305 2936error:
5461b305
DG
2937 if (cmd_ctx->llm == NULL) {
2938 DBG("Missing llm structure. Allocating one.");
894be886 2939 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
5461b305
DG
2940 goto setup_error;
2941 }
2942 }
54d01ffb 2943 /* Set return code */
5461b305 2944 cmd_ctx->llm->ret_code = ret;
5461b305 2945setup_error:
b5541356 2946 if (cmd_ctx->session) {
54d01ffb 2947 session_unlock(cmd_ctx->session);
b5541356 2948 }
256a5576
MD
2949 if (need_tracing_session) {
2950 session_unlock_list();
2951 }
54d01ffb 2952init_setup_error:
8028d920 2953 return ret;
fac6795d
DG
2954}
2955
44a5e5eb
DG
2956/*
2957 * Thread managing health check socket.
2958 */
2959static void *thread_manage_health(void *data)
2960{
eb4a2943 2961 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
44a5e5eb
DG
2962 uint32_t revents, nb_fd;
2963 struct lttng_poll_event events;
2964 struct lttcomm_health_msg msg;
2965 struct lttcomm_health_data reply;
2966
2967 DBG("[thread] Manage health check started");
2968
2969 rcu_register_thread();
2970
2971 /* Create unix socket */
2972 sock = lttcomm_create_unix_sock(health_unix_sock_path);
2973 if (sock < 0) {
2974 ERR("Unable to create health check Unix socket");
2975 ret = -1;
2976 goto error;
2977 }
2978
b662582b
DG
2979 /*
2980 * Set the CLOEXEC flag. Return code is useless because either way, the
2981 * show must go on.
2982 */
2983 (void) utils_set_fd_cloexec(sock);
2984
44a5e5eb
DG
2985 ret = lttcomm_listen_unix_sock(sock);
2986 if (ret < 0) {
2987 goto error;
2988 }
2989
2990 /*
2991 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
2992 * more will be added to this poll set.
2993 */
2994 ret = create_thread_poll_set(&events, 2);
2995 if (ret < 0) {
2996 goto error;
2997 }
2998
2999 /* Add the application registration socket */
3000 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3001 if (ret < 0) {
3002 goto error;
3003 }
3004
3005 while (1) {
3006 DBG("Health check ready");
3007
44a5e5eb
DG
3008 /* Inifinite blocking call, waiting for transmission */
3009restart:
3010 ret = lttng_poll_wait(&events, -1);
3011 if (ret < 0) {
3012 /*
3013 * Restart interrupted system call.
3014 */
3015 if (errno == EINTR) {
3016 goto restart;
3017 }
3018 goto error;
3019 }
3020
0d9c5d77
DG
3021 nb_fd = ret;
3022
44a5e5eb
DG
3023 for (i = 0; i < nb_fd; i++) {
3024 /* Fetch once the poll data */
3025 revents = LTTNG_POLL_GETEV(&events, i);
3026 pollfd = LTTNG_POLL_GETFD(&events, i);
3027
3028 /* Thread quit pipe has been closed. Killing thread. */
3029 ret = check_thread_quit_pipe(pollfd, revents);
3030 if (ret) {
139ac872
MD
3031 err = 0;
3032 goto exit;
44a5e5eb
DG
3033 }
3034
3035 /* Event on the registration socket */
3036 if (pollfd == sock) {
3037 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3038 ERR("Health socket poll error");
3039 goto error;
3040 }
3041 }
3042 }
3043
3044 new_sock = lttcomm_accept_unix_sock(sock);
3045 if (new_sock < 0) {
3046 goto error;
3047 }
3048
b662582b
DG
3049 /*
3050 * Set the CLOEXEC flag. Return code is useless because either way, the
3051 * show must go on.
3052 */
3053 (void) utils_set_fd_cloexec(new_sock);
3054
44a5e5eb
DG
3055 DBG("Receiving data from client for health...");
3056 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3057 if (ret <= 0) {
3058 DBG("Nothing recv() from client... continuing");
3059 ret = close(new_sock);
3060 if (ret) {
3061 PERROR("close");
3062 }
3063 new_sock = -1;
3064 continue;
3065 }
3066
3067 rcu_thread_online();
3068
3069 switch (msg.component) {
3070 case LTTNG_HEALTH_CMD:
3071 reply.ret_code = health_check_state(&health_thread_cmd);
3072 break;
139ac872
MD
3073 case LTTNG_HEALTH_APP_MANAGE:
3074 reply.ret_code = health_check_state(&health_thread_app_manage);
3075 break;
44a5e5eb
DG
3076 case LTTNG_HEALTH_APP_REG:
3077 reply.ret_code = health_check_state(&health_thread_app_reg);
3078 break;
3079 case LTTNG_HEALTH_KERNEL:
3080 reply.ret_code = health_check_state(&health_thread_kernel);
3081 break;
3082 case LTTNG_HEALTH_CONSUMER:
3083 reply.ret_code = check_consumer_health();
3084 break;
3085 case LTTNG_HEALTH_ALL:
44a5e5eb 3086 reply.ret_code =
139ac872
MD
3087 health_check_state(&health_thread_app_manage) &&
3088 health_check_state(&health_thread_app_reg) &&
3089 health_check_state(&health_thread_cmd) &&
3090 health_check_state(&health_thread_kernel) &&
3091 check_consumer_health();
44a5e5eb
DG
3092 break;
3093 default:
f73fabfd 3094 reply.ret_code = LTTNG_ERR_UND;
44a5e5eb
DG
3095 break;
3096 }
3097
3098 /*
3099 * Flip ret value since 0 is a success and 1 indicates a bad health for
3100 * the client where in the sessiond it is the opposite. Again, this is
3101 * just to make things easier for us poor developer which enjoy a lot
3102 * lazyness.
3103 */
3104 if (reply.ret_code == 0 || reply.ret_code == 1) {
3105 reply.ret_code = !reply.ret_code;
3106 }
3107
3108 DBG2("Health check return value %d", reply.ret_code);
3109
3110 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3111 if (ret < 0) {
3112 ERR("Failed to send health data back to client");
3113 }