Fix: arguments in the wrong order for fd-limit
[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>
331744e3 28#include <inttypes.h>
0fdd1e2c 29#include <sys/mman.h>
b73401da 30#include <sys/mount.h>
1e307fab 31#include <sys/resource.h>
fac6795d
DG
32#include <sys/socket.h>
33#include <sys/stat.h>
34#include <sys/types.h>
0fdd1e2c 35#include <sys/wait.h>
5c827ce0 36#include <urcu/uatomic.h>
fac6795d 37#include <unistd.h>
3bd1e081 38#include <config.h>
fac6795d 39
990570ed 40#include <common/common.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"
7972aab2 49#include "buffer-registry.h"
54d01ffb 50#include "channel.h"
2f77fc4b 51#include "cmd.h"
00e2e675 52#include "consumer.h"
099e26bd 53#include "context.h"
54d01ffb 54#include "event.h"
4771f025 55#include "kernel.h"
f1e16794 56#include "kernel-consumer.h"
096102bd 57#include "modprobe.h"
0fdd1e2c 58#include "shm.h"
1e307fab 59#include "ust-ctl.h"
00e2e675 60#include "ust-consumer.h"
8e68d1c8 61#include "utils.h"
4063050c 62#include "fd-limit.h"
44a5e5eb 63#include "health.h"
8ac94142 64#include "testpoint.h"
d0b96690 65#include "ust-thread.h"
fac6795d 66
ebaeda94
MD
67#define CONSUMERD_FILE "lttng-consumerd"
68
75462a81 69/* Const values */
bbccc3d2 70const char default_tracing_group[] = DEFAULT_TRACING_GROUP;
686204ab 71
fac6795d
DG
72const char *progname;
73const char *opt_tracing_group;
35f90c40 74static const char *opt_pidfile;
5b8719f5 75static int opt_sig_parent;
97e19046 76static int opt_verbose_consumer;
fac6795d 77static int opt_daemon;
4fba7219 78static int opt_no_kernel;
fac6795d 79static int is_root; /* Set to 1 if the daemon is running as root */
1d4b027a 80static pid_t ppid; /* Parent PID for --sig-parent option */
67e40797 81static char *rundir;
3bd1e081 82
a23ec3a7
DG
83/*
84 * Consumer daemon specific control data. Every value not initialized here is
85 * set to 0 by the static definition.
86 */
3bd1e081
MD
87static struct consumer_data kconsumer_data = {
88 .type = LTTNG_CONSUMER_KERNEL,
60922cb0
DG
89 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
90 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
03550b58
MD
91 .err_sock = -1,
92 .cmd_sock = -1,
331744e3 93 .metadata_sock.fd = -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,
331744e3 105 .metadata_sock.fd = -1,
173af62f
DG
106 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
107 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
108 .cond = PTHREAD_COND_INITIALIZER,
109 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
7753dea8
MD
110};
111static struct consumer_data ustconsumer32_data = {
112 .type = LTTNG_CONSUMER32_UST,
60922cb0
DG
113 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
114 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
03550b58
MD
115 .err_sock = -1,
116 .cmd_sock = -1,
331744e3 117 .metadata_sock.fd = -1,
173af62f
DG
118 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
119 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
120 .cond = PTHREAD_COND_INITIALIZER,
121 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081
MD
122};
123
26c9d55e 124/* Shared between threads */
099e26bd 125static int dispatch_thread_exit;
fac6795d 126
54d01ffb
DG
127/* Global application Unix socket path */
128static char apps_unix_sock_path[PATH_MAX];
129/* Global client Unix socket path */
130static char client_unix_sock_path[PATH_MAX];
54d01ffb
DG
131/* global wait shm path for UST */
132static char wait_shm_path[PATH_MAX];
44a5e5eb
DG
133/* Global health check unix path */
134static char health_unix_sock_path[PATH_MAX];
fac6795d 135
1d4b027a 136/* Sockets and FDs */
a4b35e07
MD
137static int client_sock = -1;
138static int apps_sock = -1;
2f77fc4b 139int kernel_tracer_fd = -1;
76d7553f 140static int kernel_poll_pipe[2] = { -1, -1 };
1d4b027a 141
273ea72c
DG
142/*
143 * Quit pipe for all threads. This permits a single cancellation point
144 * for all threads when receiving an event on the pipe.
145 */
76d7553f 146static int thread_quit_pipe[2] = { -1, -1 };
273ea72c 147
099e26bd
DG
148/*
149 * This pipe is used to inform the thread managing application communication
150 * that a command is queued and ready to be processed.
151 */
76d7553f 152static int apps_cmd_pipe[2] = { -1, -1 };
099e26bd 153
d0b96690
DG
154int apps_cmd_notify_pipe[2] = { -1, -1 };
155
1d4b027a 156/* Pthread, Mutexes and Semaphores */
1d4b027a 157static pthread_t apps_thread;
d0b96690 158static pthread_t apps_notify_thread;
099e26bd 159static pthread_t reg_apps_thread;
1d4b027a 160static pthread_t client_thread;
7a485870 161static pthread_t kernel_thread;
099e26bd 162static pthread_t dispatch_thread;
44a5e5eb 163static pthread_t health_thread;
0b2dc8df 164static pthread_t ht_cleanup_thread;
5eb91c98 165
099e26bd
DG
166/*
167 * UST registration command queue. This queue is tied with a futex and uses a N
168 * wakers / 1 waiter implemented and detailed in futex.c/.h
169 *
170 * The thread_manage_apps and thread_dispatch_ust_registration interact with
171 * this queue and the wait/wake scheme.
172 */
173static struct ust_cmd_queue ust_cmd_queue;
174
b5541356
DG
175/*
176 * Pointer initialized before thread creation.
177 *
178 * This points to the tracing session list containing the session count and a
179 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
180 * MUST NOT be taken if you call a public function in session.c.
04ea676f 181 *
d063d709 182 * The lock is nested inside the structure: session_list_ptr->lock. Please use
54d01ffb 183 * session_lock_list and session_unlock_list for lock acquisition.
b5541356
DG
184 */
185static struct ltt_session_list *session_list_ptr;
186
7753dea8
MD
187int ust_consumerd64_fd = -1;
188int ust_consumerd32_fd = -1;
189
fb6f1fa2
YB
190static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
191static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
192static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
193static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
fb09408a 194
2f77fc4b
DG
195static const char *module_proc_lttng = "/proc/lttng";
196
5c827ce0
DG
197/*
198 * Consumer daemon state which is changed when spawning it, killing it or in
199 * case of a fatal error.
200 */
201enum consumerd_state {
202 CONSUMER_STARTED = 1,
203 CONSUMER_STOPPED = 2,
204 CONSUMER_ERROR = 3,
205};
206
207/*
208 * This consumer daemon state is used to validate if a client command will be
209 * able to reach the consumer. If not, the client is informed. For instance,
210 * doing a "lttng start" when the consumer state is set to ERROR will return an
211 * error to the client.
212 *
213 * The following example shows a possible race condition of this scheme:
214 *
215 * consumer thread error happens
216 * client cmd arrives
217 * client cmd checks state -> still OK
218 * consumer thread exit, sets error
219 * client cmd try to talk to consumer
220 * ...
221 *
222 * However, since the consumer is a different daemon, we have no way of making
223 * sure the command will reach it safely even with this state flag. This is why
224 * we consider that up to the state validation during command processing, the
225 * command is safe. After that, we can not guarantee the correctness of the
226 * client request vis-a-vis the consumer.
227 */
228static enum consumerd_state ust_consumerd_state;
229static enum consumerd_state kernel_consumerd_state;
230
ae9e45b3
DG
231/*
232 * Socket timeout for receiving and sending in seconds.
233 */
234static int app_socket_timeout;
235
12744796
DG
236/* Set in main() with the current page size. */
237long page_size;
238
fb09408a 239static
7753dea8 240void setup_consumerd_path(void)
fb09408a 241{
fc7a59ce 242 const char *bin, *libdir;
fb09408a 243
7753dea8
MD
244 /*
245 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
246 * native architecture size consumer if CONFIG_CONSUMER*_PATH
247 * has not been defined.
7753dea8 248 */
ebaeda94 249#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
250 if (!consumerd32_bin[0]) {
251 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
252 }
253 if (!consumerd32_libdir[0]) {
254 consumerd32_libdir = INSTALL_LIB_PATH;
255 }
256#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
257 if (!consumerd64_bin[0]) {
258 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 259 }
ebaeda94
MD
260 if (!consumerd64_libdir[0]) {
261 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
262 }
263#else
264#error "Unknown bitness"
265#endif
266
fb09408a
MD
267 /*
268 * runtime env. var. overrides the build default.
269 */
fc7a59ce
AM
270 bin = getenv("LTTNG_CONSUMERD32_BIN");
271 if (bin) {
272 consumerd32_bin = bin;
7753dea8 273 }
fc7a59ce
AM
274 bin = getenv("LTTNG_CONSUMERD64_BIN");
275 if (bin) {
276 consumerd64_bin = bin;
ebaeda94 277 }
72f579ee 278 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
279 if (libdir) {
280 consumerd32_libdir = libdir;
281 }
72f579ee 282 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
283 if (libdir) {
284 consumerd64_libdir = libdir;
fb09408a
MD
285 }
286}
287
5eb91c98
DG
288/*
289 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
290 */
d0b96690 291int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
5eb91c98
DG
292{
293 int ret;
294
d0b96690 295 assert(events);
5eb91c98
DG
296
297 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
298 if (ret < 0) {
299 goto error;
300 }
301
302 /* Add quit pipe */
d0b96690 303 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
5eb91c98
DG
304 if (ret < 0) {
305 goto error;
306 }
307
308 return 0;
309
310error:
311 return ret;
312}
313
314/*
315 * Check if the thread quit pipe was triggered.
316 *
317 * Return 1 if it was triggered else 0;
318 */
d0b96690 319int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
5eb91c98
DG
320{
321 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
322 return 1;
323 }
324
325 return 0;
326}
327
0fdd1e2c
DG
328/*
329 * Return group ID of the tracing group or -1 if not found.
330 */
996b65c8
MD
331static gid_t allowed_group(void)
332{
333 struct group *grp;
334
274e143b
MD
335 if (opt_tracing_group) {
336 grp = getgrnam(opt_tracing_group);
337 } else {
338 grp = getgrnam(default_tracing_group);
339 }
996b65c8
MD
340 if (!grp) {
341 return -1;
342 } else {
343 return grp->gr_gid;
344 }
345}
346
273ea72c 347/*
5eb91c98 348 * Init thread quit pipe.
273ea72c
DG
349 *
350 * Return -1 on error or 0 if all pipes are created.
351 */
352static int init_thread_quit_pipe(void)
353{
730389d9 354 int ret, i;
273ea72c 355
730389d9 356 ret = pipe(thread_quit_pipe);
273ea72c 357 if (ret < 0) {
730389d9 358 PERROR("thread quit pipe");
273ea72c
DG
359 goto error;
360 }
361
730389d9
DG
362 for (i = 0; i < 2; i++) {
363 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
364 if (ret < 0) {
365 PERROR("fcntl");
366 goto error;
367 }
368 }
369
273ea72c
DG
370error:
371 return ret;
372}
373
099e26bd
DG
374/*
375 * Stop all threads by closing the thread quit pipe.
376 */
cf3af59e
MD
377static void stop_threads(void)
378{
5eb91c98
DG
379 int ret;
380
cf3af59e
MD
381 /* Stopping all threads */
382 DBG("Terminating all threads");
54d01ffb 383 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
384 if (ret < 0) {
385 ERR("write error on thread quit pipe");
386 }
387
099e26bd 388 /* Dispatch thread */
26c9d55e 389 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 390 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
391}
392
5c760329
DG
393/*
394 * Close every consumer sockets.
395 */
396static void close_consumer_sockets(void)
397{
398 int ret;
399
400 if (kconsumer_data.err_sock >= 0) {
401 ret = close(kconsumer_data.err_sock);
402 if (ret < 0) {
403 PERROR("kernel consumer err_sock close");
404 }
405 }
406 if (ustconsumer32_data.err_sock >= 0) {
407 ret = close(ustconsumer32_data.err_sock);
408 if (ret < 0) {
073fffee 409 PERROR("UST consumerd32 err_sock close");
5c760329
DG
410 }
411 }
412 if (ustconsumer64_data.err_sock >= 0) {
413 ret = close(ustconsumer64_data.err_sock);
414 if (ret < 0) {
073fffee 415 PERROR("UST consumerd64 err_sock close");
5c760329
DG
416 }
417 }
418 if (kconsumer_data.cmd_sock >= 0) {
419 ret = close(kconsumer_data.cmd_sock);
420 if (ret < 0) {
421 PERROR("kernel consumer cmd_sock close");
422 }
423 }
424 if (ustconsumer32_data.cmd_sock >= 0) {
425 ret = close(ustconsumer32_data.cmd_sock);
426 if (ret < 0) {
073fffee 427 PERROR("UST consumerd32 cmd_sock close");
5c760329
DG
428 }
429 }
430 if (ustconsumer64_data.cmd_sock >= 0) {
431 ret = close(ustconsumer64_data.cmd_sock);
432 if (ret < 0) {
073fffee 433 PERROR("UST consumerd64 cmd_sock close");
5c760329
DG
434 }
435 }
436}
437
fac6795d 438/*
d063d709 439 * Cleanup the daemon
fac6795d 440 */
cf3af59e 441static void cleanup(void)
fac6795d 442{
ef599319 443 int ret;
6620da75 444 char *cmd = NULL;
af9737e9 445 struct ltt_session *sess, *stmp;
fac6795d 446
1d4b027a 447 DBG("Cleaning up");
e07ae692 448
2f77fc4b
DG
449 /* First thing first, stop all threads */
450 utils_close_pipe(thread_quit_pipe);
451
35f90c40
DG
452 /*
453 * If opt_pidfile is undefined, the default file will be wiped when
454 * removing the rundir.
455 */
456 if (opt_pidfile) {
457 ret = remove(opt_pidfile);
458 if (ret < 0) {
459 PERROR("remove pidfile %s", opt_pidfile);
460 }
461 }
462
67e40797
DG
463 DBG("Removing %s directory", rundir);
464 ret = asprintf(&cmd, "rm -rf %s", rundir);
465 if (ret < 0) {
466 ERR("asprintf failed. Something is really wrong!");
467 }
5461b305 468
67e40797
DG
469 /* Remove lttng run directory */
470 ret = system(cmd);
471 if (ret < 0) {
472 ERR("Unable to clean %s", rundir);
1d4b027a 473 }
67e40797 474 free(cmd);
bd69add2 475 free(rundir);
5461b305 476
99bab54f 477 DBG("Cleaning up all sessions");
fac6795d 478
b5541356 479 /* Destroy session list mutex */
273ea72c
DG
480 if (session_list_ptr != NULL) {
481 pthread_mutex_destroy(&session_list_ptr->lock);
482
483 /* Cleanup ALL session */
54d01ffb
DG
484 cds_list_for_each_entry_safe(sess, stmp,
485 &session_list_ptr->head, list) {
2f77fc4b 486 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
487 }
488 }
489
099e26bd 490 DBG("Closing all UST sockets");
56fff090 491 ust_app_clean_list();
7972aab2 492 buffer_reg_destroy_registries();
099e26bd 493
4fba7219
DG
494 if (is_root && !opt_no_kernel) {
495 DBG2("Closing kernel fd");
a4b35e07 496 if (kernel_tracer_fd >= 0) {
76d7553f
MD
497 ret = close(kernel_tracer_fd);
498 if (ret) {
499 PERROR("close");
500 }
a4b35e07 501 }
2f50c8a3 502 DBG("Unloading kernel modules");
096102bd 503 modprobe_remove_lttng_all();
2f50c8a3 504 }
2f77fc4b 505
5c760329
DG
506 close_consumer_sockets();
507
421cb601 508 /* <fun> */
f56a39af 509 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
510 "Matthew, BEET driven development works!%c[%dm",
511 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
512 /* </fun> */
fac6795d
DG
513}
514
e065084a 515/*
d063d709 516 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 517 *
d063d709 518 * Return lttcomm error code.
e065084a
DG
519 */
520static int send_unix_sock(int sock, void *buf, size_t len)
521{
522 /* Check valid length */
c617c0c6 523 if (len == 0) {
e065084a
DG
524 return -1;
525 }
526
527 return lttcomm_send_unix_sock(sock, buf, len);
528}
529
5461b305 530/*
d063d709 531 * Free memory of a command context structure.
5461b305 532 */
a2fb29a5 533static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 534{
a2fb29a5
DG
535 DBG("Clean command context structure");
536 if (*cmd_ctx) {
537 if ((*cmd_ctx)->llm) {
538 free((*cmd_ctx)->llm);
5461b305 539 }
a2fb29a5
DG
540 if ((*cmd_ctx)->lsm) {
541 free((*cmd_ctx)->lsm);
5461b305 542 }
a2fb29a5
DG
543 free(*cmd_ctx);
544 *cmd_ctx = NULL;
5461b305
DG
545 }
546}
547
fac6795d 548/*
0fdd1e2c 549 * Notify UST applications using the shm mmap futex.
fac6795d 550 */
0fdd1e2c 551static int notify_ust_apps(int active)
fac6795d 552{
0fdd1e2c 553 char *wait_shm_mmap;
fac6795d 554
0fdd1e2c 555 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 556
0fdd1e2c
DG
557 /* See shm.c for this call implying mmap, shm and futex calls */
558 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
559 if (wait_shm_mmap == NULL) {
fac6795d
DG
560 goto error;
561 }
562
0fdd1e2c
DG
563 /* Wake waiting process */
564 futex_wait_update((int32_t *) wait_shm_mmap, active);
565
566 /* Apps notified successfully */
567 return 0;
fac6795d
DG
568
569error:
0fdd1e2c 570 return -1;
fac6795d
DG
571}
572
e065084a 573/*
d063d709
DG
574 * Setup the outgoing data buffer for the response (llm) by allocating the
575 * right amount of memory and copying the original information from the lsm
576 * structure.
ca95a216 577 *
d063d709 578 * Return total size of the buffer pointed by buf.
ca95a216 579 */
5461b305 580static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 581{
f3ed775e 582 int ret, buf_size;
ca95a216 583
f3ed775e 584 buf_size = size;
5461b305 585
ba7f0ae5 586 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 587 if (cmd_ctx->llm == NULL) {
76d7553f 588 PERROR("zmalloc");
5461b305 589 ret = -ENOMEM;
ca95a216
DG
590 goto error;
591 }
592
5461b305
DG
593 /* Copy common data */
594 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 595 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 596
5461b305
DG
597 cmd_ctx->llm->data_size = size;
598 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
599
ca95a216
DG
600 return buf_size;
601
602error:
603 return ret;
604}
605
7a485870 606/*
5eb91c98 607 * Update the kernel poll set of all channel fd available over all tracing
d063d709 608 * session. Add the wakeup pipe at the end of the set.
7a485870 609 */
5eb91c98 610static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 611{
5eb91c98 612 int ret;
7a485870
DG
613 struct ltt_session *session;
614 struct ltt_kernel_channel *channel;
615
5eb91c98 616 DBG("Updating kernel poll set");
7a485870 617
54d01ffb 618 session_lock_list();
b5541356 619 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 620 session_lock(session);
7a485870 621 if (session->kernel_session == NULL) {
54d01ffb 622 session_unlock(session);
7a485870
DG
623 continue;
624 }
7a485870 625
54d01ffb
DG
626 cds_list_for_each_entry(channel,
627 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
628 /* Add channel fd to the kernel poll set */
629 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
630 if (ret < 0) {
54d01ffb 631 session_unlock(session);
5eb91c98
DG
632 goto error;
633 }
634 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 635 }
54d01ffb 636 session_unlock(session);
7a485870 637 }
54d01ffb 638 session_unlock_list();
7a485870 639
5eb91c98 640 return 0;
7a485870
DG
641
642error:
54d01ffb 643 session_unlock_list();
7a485870
DG
644 return -1;
645}
646
647/*
54d01ffb 648 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 649 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 650 *
d063d709 651 * Useful for CPU hotplug feature.
7a485870 652 */
2bdd86d4 653static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
654{
655 int ret = 0;
656 struct ltt_session *session;
173af62f 657 struct ltt_kernel_session *ksess;
7a485870
DG
658 struct ltt_kernel_channel *channel;
659
660 DBG("Updating kernel streams for channel fd %d", fd);
661
54d01ffb 662 session_lock_list();
b5541356 663 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 664 session_lock(session);
7a485870 665 if (session->kernel_session == NULL) {
54d01ffb 666 session_unlock(session);
7a485870
DG
667 continue;
668 }
173af62f 669 ksess = session->kernel_session;
d9800920 670
173af62f 671 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
7a485870
DG
672 if (channel->fd == fd) {
673 DBG("Channel found, updating kernel streams");
674 ret = kernel_open_channel_stream(channel);
675 if (ret < 0) {
b3c750d2 676 goto error;
7a485870 677 }
d9800920 678
7a485870 679 /*
5eb91c98
DG
680 * Have we already sent fds to the consumer? If yes, it means
681 * that tracing is started so it is safe to send our updated
682 * stream fds.
7a485870 683 */
173af62f
DG
684 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
685 struct lttng_ht_iter iter;
686 struct consumer_socket *socket;
687
e7fe706f 688 rcu_read_lock();
173af62f
DG
689 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
690 &iter.iter, socket, node.node) {
691 /* Code flow error */
692 assert(socket->fd >= 0);
693
694 pthread_mutex_lock(socket->lock);
f50f23d9 695 ret = kernel_consumer_send_channel_stream(socket,
173af62f
DG
696 channel, ksess);
697 pthread_mutex_unlock(socket->lock);
698 if (ret < 0) {
e7fe706f 699 rcu_read_unlock();
173af62f
DG
700 goto error;
701 }
7a485870 702 }
e7fe706f 703 rcu_read_unlock();
7a485870 704 }
b3c750d2 705 goto error;
7a485870
DG
706 }
707 }
54d01ffb 708 session_unlock(session);
7a485870 709 }
54d01ffb 710 session_unlock_list();
b3c750d2 711 return ret;
7a485870 712
b3c750d2 713error:
54d01ffb
DG
714 session_unlock(session);
715 session_unlock_list();
7a485870
DG
716 return ret;
717}
718
487cf67c 719/*
ffe60014
DG
720 * For each tracing session, update newly registered apps. The session list
721 * lock MUST be acquired before calling this.
487cf67c
DG
722 */
723static void update_ust_app(int app_sock)
724{
725 struct ltt_session *sess, *stmp;
726
727 /* For all tracing session(s) */
728 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
4ee14516 729 session_lock(sess);
421cb601
DG
730 if (sess->ust_session) {
731 ust_app_global_update(sess->ust_session, app_sock);
732 }
4ee14516 733 session_unlock(sess);
487cf67c
DG
734 }
735}
736
7a485870 737/*
d063d709 738 * This thread manage event coming from the kernel.
7a485870 739 *
d063d709
DG
740 * Features supported in this thread:
741 * -) CPU Hotplug
7a485870
DG
742 */
743static void *thread_manage_kernel(void *data)
744{
139ac872 745 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 746 uint32_t revents, nb_fd;
7a485870 747 char tmp;
5eb91c98 748 struct lttng_poll_event events;
7a485870 749
6993eeb3 750 DBG("[thread] Thread manage kernel started");
7a485870 751
927ca06a
DG
752 health_register(HEALTH_TYPE_KERNEL);
753
d5d63bf1
DG
754 /*
755 * This first step of the while is to clean this structure which could free
6d737ce4 756 * non NULL pointers so initialize it before the loop.
d5d63bf1 757 */
6d737ce4 758 lttng_poll_init(&events);
d5d63bf1 759
6993eeb3
CB
760 if (testpoint(thread_manage_kernel)) {
761 goto error_testpoint;
762 }
8ac94142 763
840cb59c 764 health_code_update();
44a5e5eb 765
6993eeb3 766 if (testpoint(thread_manage_kernel_before_loop)) {
d21b0d71 767 goto error_testpoint;
6993eeb3
CB
768 }
769
7a485870 770 while (1) {
840cb59c 771 health_code_update();
44a5e5eb 772
7a485870 773 if (update_poll_flag == 1) {
d21b0d71
DG
774 /* Clean events object. We are about to populate it again. */
775 lttng_poll_clean(&events);
776
d0b96690 777 ret = sessiond_set_thread_pollset(&events, 2);
d21b0d71
DG
778 if (ret < 0) {
779 goto error_poll_create;
780 }
781
782 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
783 if (ret < 0) {
784 goto error;
785 }
5f822d0a 786
d21b0d71 787 /* This will add the available kernel channel if any. */
5eb91c98
DG
788 ret = update_kernel_poll(&events);
789 if (ret < 0) {
7a485870
DG
790 goto error;
791 }
792 update_poll_flag = 0;
793 }
794
d21b0d71 795 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events));
7a485870
DG
796
797 /* Poll infinite value of time */
88f2b785 798 restart:
a78af745 799 health_poll_entry();
5eb91c98 800 ret = lttng_poll_wait(&events, -1);
a78af745 801 health_poll_exit();
7a485870 802 if (ret < 0) {
88f2b785
MD
803 /*
804 * Restart interrupted system call.
805 */
806 if (errno == EINTR) {
807 goto restart;
808 }
7a485870
DG
809 goto error;
810 } else if (ret == 0) {
811 /* Should not happen since timeout is infinite */
85611738
DG
812 ERR("Return value of poll is 0 with an infinite timeout.\n"
813 "This should not have happened! Continuing...");
7a485870
DG
814 continue;
815 }
816
0d9c5d77
DG
817 nb_fd = ret;
818
5eb91c98
DG
819 for (i = 0; i < nb_fd; i++) {
820 /* Fetch once the poll data */
821 revents = LTTNG_POLL_GETEV(&events, i);
822 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 823
840cb59c 824 health_code_update();
44a5e5eb 825
5eb91c98 826 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 827 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 828 if (ret) {
139ac872
MD
829 err = 0;
830 goto exit;
5eb91c98 831 }
7a485870 832
5eb91c98
DG
833 /* Check for data on kernel pipe */
834 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
f921c78f
DG
835 do {
836 ret = read(kernel_poll_pipe[0], &tmp, 1);
837 } while (ret < 0 && errno == EINTR);
838 /*
839 * Ret value is useless here, if this pipe gets any actions an
840 * update is required anyway.
841 */
5eb91c98
DG
842 update_poll_flag = 1;
843 continue;
844 } else {
845 /*
846 * New CPU detected by the kernel. Adding kernel stream to
847 * kernel session and updating the kernel consumer
848 */
849 if (revents & LPOLLIN) {
2bdd86d4 850 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
851 if (ret < 0) {
852 continue;
853 }
854 break;
855 /*
856 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
857 * and unregister kernel stream at this point.
858 */
7a485870 859 }
7a485870
DG
860 }
861 }
862 }
863
139ac872 864exit:
7a485870 865error:
5eb91c98 866 lttng_poll_clean(&events);
76d7553f 867error_poll_create:
6993eeb3 868error_testpoint:
6620da75
DG
869 utils_close_pipe(kernel_poll_pipe);
870 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 871 if (err) {
840cb59c 872 health_error();
139ac872 873 ERR("Health error occurred in %s", __func__);
6620da75
DG
874 WARN("Kernel thread died unexpectedly. "
875 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 876 }
927ca06a 877 health_unregister();
76d7553f 878 DBG("Kernel thread dying");
7a485870
DG
879 return NULL;
880}
881
a23ec3a7
DG
882/*
883 * Signal pthread condition of the consumer data that the thread.
884 */
885static void signal_consumer_condition(struct consumer_data *data, int state)
886{
887 pthread_mutex_lock(&data->cond_mutex);
888
889 /*
890 * The state is set before signaling. It can be any value, it's the waiter
891 * job to correctly interpret this condition variable associated to the
892 * consumer pthread_cond.
893 *
894 * A value of 0 means that the corresponding thread of the consumer data
895 * was not started. 1 indicates that the thread has started and is ready
896 * for action. A negative value means that there was an error during the
897 * thread bootstrap.
898 */
899 data->consumer_thread_is_ready = state;
900 (void) pthread_cond_signal(&data->cond);
901
902 pthread_mutex_unlock(&data->cond_mutex);
903}
904
1d4b027a 905/*
3bd1e081 906 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 907 */
3bd1e081 908static void *thread_manage_consumer(void *data)
1d4b027a 909{
139ac872 910 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98 911 uint32_t revents, nb_fd;
1d4b027a 912 enum lttcomm_return_code code;
5eb91c98 913 struct lttng_poll_event events;
3bd1e081 914 struct consumer_data *consumer_data = data;
1d4b027a 915
3bd1e081 916 DBG("[thread] Manage consumer started");
1d4b027a 917
927ca06a
DG
918 health_register(HEALTH_TYPE_CONSUMER);
919
855060f8 920 health_code_update();
9449cc75 921
5eb91c98 922 /*
331744e3
JD
923 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
924 * metadata_sock. Nothing more will be added to this poll set.
5eb91c98 925 */
331744e3 926 ret = sessiond_set_thread_pollset(&events, 3);
5eb91c98 927 if (ret < 0) {
76d7553f 928 goto error_poll;
5eb91c98 929 }
273ea72c 930
edb8b045
DG
931 /*
932 * The error socket here is already in a listening state which was done
933 * just before spawning this thread to avoid a race between the consumer
934 * daemon exec trying to connect and the listen() call.
935 */
3bd1e081 936 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
937 if (ret < 0) {
938 goto error;
939 }
940
840cb59c 941 health_code_update();
44a5e5eb 942
331744e3 943 /* Infinite blocking call, waiting for transmission */
88f2b785 944restart:
a78af745 945 health_poll_entry();
8ac94142 946
6993eeb3
CB
947 if (testpoint(thread_manage_consumer)) {
948 goto error;
949 }
8ac94142 950
5eb91c98 951 ret = lttng_poll_wait(&events, -1);
a78af745 952 health_poll_exit();
273ea72c 953 if (ret < 0) {
88f2b785
MD
954 /*
955 * Restart interrupted system call.
956 */
957 if (errno == EINTR) {
958 goto restart;
959 }
273ea72c
DG
960 goto error;
961 }
962
0d9c5d77
DG
963 nb_fd = ret;
964
5eb91c98
DG
965 for (i = 0; i < nb_fd; i++) {
966 /* Fetch once the poll data */
967 revents = LTTNG_POLL_GETEV(&events, i);
968 pollfd = LTTNG_POLL_GETFD(&events, i);
969
840cb59c 970 health_code_update();
44a5e5eb 971
5eb91c98 972 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 973 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 974 if (ret) {
139ac872
MD
975 err = 0;
976 goto exit;
5eb91c98
DG
977 }
978
979 /* Event on the registration socket */
3bd1e081 980 if (pollfd == consumer_data->err_sock) {
5eb91c98 981 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 982 ERR("consumer err socket poll error");
5eb91c98
DG
983 goto error;
984 }
985 }
273ea72c
DG
986 }
987
3bd1e081 988 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
989 if (sock < 0) {
990 goto error;
991 }
992
b662582b
DG
993 /*
994 * Set the CLOEXEC flag. Return code is useless because either way, the
995 * show must go on.
996 */
997 (void) utils_set_fd_cloexec(sock);
998
840cb59c 999 health_code_update();
44a5e5eb 1000
3bd1e081 1001 DBG2("Receiving code from consumer err_sock");
ee0b0061 1002
712ea556 1003 /* Getting status code from kconsumerd */
54d01ffb
DG
1004 ret = lttcomm_recv_unix_sock(sock, &code,
1005 sizeof(enum lttcomm_return_code));
1d4b027a
DG
1006 if (ret <= 0) {
1007 goto error;
1008 }
1009
840cb59c 1010 health_code_update();
44a5e5eb 1011
f73fabfd 1012 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
331744e3 1013 /* Connect both socket, command and metadata. */
3bd1e081
MD
1014 consumer_data->cmd_sock =
1015 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
331744e3
JD
1016 consumer_data->metadata_sock.fd =
1017 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
1018 if (consumer_data->cmd_sock < 0 ||
1019 consumer_data->metadata_sock.fd < 0) {
1020 PERROR("consumer connect cmd socket");
a23ec3a7
DG
1021 /* On error, signal condition and quit. */
1022 signal_consumer_condition(consumer_data, -1);
1d4b027a
DG
1023 goto error;
1024 }
331744e3
JD
1025 /* Create metadata socket lock. */
1026 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1027 if (consumer_data->metadata_sock.lock == NULL) {
1028 PERROR("zmalloc pthread mutex");
1029 ret = -1;
1030 goto error;
1031 }
1032 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1033
a23ec3a7 1034 signal_consumer_condition(consumer_data, 1);
331744e3
JD
1035 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1036 DBG("Consumer metadata socket ready (fd: %d)",
1037 consumer_data->metadata_sock.fd);
1d4b027a 1038 } else {
3bd1e081 1039 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
1040 lttcomm_get_readable_code(-code));
1041 goto error;
1042 }
1043
331744e3 1044 /* Remove the consumerd error sock since we've established a connexion */
3bd1e081 1045 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1046 if (ret < 0) {
72079cae
DG
1047 goto error;
1048 }
1049
331744e3 1050 /* Add new accepted error socket. */
5eb91c98
DG
1051 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1052 if (ret < 0) {
72079cae 1053 goto error;
5eb91c98
DG
1054 }
1055
331744e3
JD
1056 /* Add metadata socket that is successfully connected. */
1057 ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd,
1058 LPOLLIN | LPOLLRDHUP);
1059 if (ret < 0) {
1060 goto error;
1061 }
1062
840cb59c 1063 health_code_update();
44a5e5eb 1064
331744e3 1065 /* Infinite blocking call, waiting for transmission */
88f2b785 1066restart_poll:
331744e3
JD
1067 while (1) {
1068 health_poll_entry();
1069 ret = lttng_poll_wait(&events, -1);
1070 health_poll_exit();
1071 if (ret < 0) {
1072 /*
1073 * Restart interrupted system call.
1074 */
1075 if (errno == EINTR) {
1076 goto restart_poll;
1077 }
1078 goto error;
88f2b785 1079 }
72079cae 1080
331744e3 1081 nb_fd = ret;
0d9c5d77 1082
331744e3
JD
1083 for (i = 0; i < nb_fd; i++) {
1084 /* Fetch once the poll data */
1085 revents = LTTNG_POLL_GETEV(&events, i);
1086 pollfd = LTTNG_POLL_GETFD(&events, i);
5eb91c98 1087
331744e3 1088 health_code_update();
44a5e5eb 1089
331744e3
JD
1090 /* Thread quit pipe has been closed. Killing thread. */
1091 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
1092 if (ret) {
1093 err = 0;
1094 goto exit;
1095 }
5eb91c98 1096
331744e3
JD
1097 if (pollfd == sock) {
1098 /* Event on the consumerd socket */
1099 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1100 ERR("consumer err socket second poll error");
1101 goto error;
1102 }
1103 health_code_update();
1104 /* Wait for any kconsumerd error */
1105 ret = lttcomm_recv_unix_sock(sock, &code,
1106 sizeof(enum lttcomm_return_code));
1107 if (ret <= 0) {
1108 ERR("consumer closed the command socket");
1109 goto error;
1110 }
1111
1112 ERR("consumer return code : %s",
1113 lttcomm_get_readable_code(-code));
1114
1115 goto exit;
1116 } else if (pollfd == consumer_data->metadata_sock.fd) {
1117 /* UST metadata requests */
1118 ret = ust_consumer_metadata_request(
1119 &consumer_data->metadata_sock);
1120 if (ret < 0) {
1121 ERR("Handling metadata request");
1122 goto error;
1123 }
1124 break;
1125 } else {
1126 ERR("Unknown pollfd");
5eb91c98
DG
1127 goto error;
1128 }
1129 }
331744e3 1130 health_code_update();
5eb91c98
DG
1131 }
1132
139ac872 1133exit:
1d4b027a 1134error:
5c827ce0
DG
1135 /* Immediately set the consumerd state to stopped */
1136 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1137 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1138 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1139 consumer_data->type == LTTNG_CONSUMER32_UST) {
1140 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1141 } else {
1142 /* Code flow error... */
1143 assert(0);
1144 }
1145
76d7553f
MD
1146 if (consumer_data->err_sock >= 0) {
1147 ret = close(consumer_data->err_sock);
1148 if (ret) {
1149 PERROR("close");
1150 }
073fffee 1151 consumer_data->err_sock = -1;
76d7553f
MD
1152 }
1153 if (consumer_data->cmd_sock >= 0) {
1154 ret = close(consumer_data->cmd_sock);
1155 if (ret) {
1156 PERROR("close");
1157 }
073fffee 1158 consumer_data->cmd_sock = -1;
76d7553f 1159 }
331744e3
JD
1160 if (consumer_data->metadata_sock.fd >= 0) {
1161 ret = close(consumer_data->metadata_sock.fd);
1162 if (ret) {
1163 PERROR("close");
1164 }
1165 }
1166 /* Cleanup metadata socket mutex. */
1167 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1168 free(consumer_data->metadata_sock.lock);
1169
76d7553f
MD
1170 if (sock >= 0) {
1171 ret = close(sock);
1172 if (ret) {
1173 PERROR("close");
1174 }
1175 }
273ea72c 1176
3bd1e081
MD
1177 unlink(consumer_data->err_unix_sock_path);
1178 unlink(consumer_data->cmd_unix_sock_path);
1179 consumer_data->pid = 0;
1d4b027a 1180
5eb91c98 1181 lttng_poll_clean(&events);
76d7553f 1182error_poll:
139ac872 1183 if (err) {
840cb59c 1184 health_error();
139ac872
MD
1185 ERR("Health error occurred in %s", __func__);
1186 }
927ca06a 1187 health_unregister();
76d7553f 1188 DBG("consumer thread cleanup completed");
0177d773 1189
5eb91c98 1190 return NULL;
099e26bd
DG
1191}
1192
099e26bd
DG
1193/*
1194 * This thread manage application communication.
1d4b027a
DG
1195 */
1196static void *thread_manage_apps(void *data)
099e26bd 1197{
139ac872 1198 int i, ret, pollfd, err = -1;
5eb91c98 1199 uint32_t revents, nb_fd;
5eb91c98 1200 struct lttng_poll_event events;
099e26bd
DG
1201
1202 DBG("[thread] Manage application started");
1203
f6a9efaa
DG
1204 rcu_register_thread();
1205 rcu_thread_online();
1206
927ca06a
DG
1207 health_register(HEALTH_TYPE_APP_MANAGE);
1208
6993eeb3
CB
1209 if (testpoint(thread_manage_apps)) {
1210 goto error_testpoint;
1211 }
1212
840cb59c 1213 health_code_update();
44a5e5eb 1214
d0b96690 1215 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1216 if (ret < 0) {
76d7553f 1217 goto error_poll_create;
5eb91c98 1218 }
099e26bd 1219
5eb91c98
DG
1220 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1221 if (ret < 0) {
1222 goto error;
1223 }
099e26bd 1224
6993eeb3
CB
1225 if (testpoint(thread_manage_apps_before_loop)) {
1226 goto error;
1227 }
8ac94142 1228
840cb59c 1229 health_code_update();
44a5e5eb 1230
5eb91c98 1231 while (1) {
d21b0d71 1232 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events));
099e26bd
DG
1233
1234 /* Inifinite blocking call, waiting for transmission */
88f2b785 1235 restart:
a78af745 1236 health_poll_entry();
5eb91c98 1237 ret = lttng_poll_wait(&events, -1);
a78af745 1238 health_poll_exit();
099e26bd 1239 if (ret < 0) {
88f2b785
MD
1240 /*
1241 * Restart interrupted system call.
1242 */
1243 if (errno == EINTR) {
1244 goto restart;
1245 }
099e26bd
DG
1246 goto error;
1247 }
1248
0d9c5d77
DG
1249 nb_fd = ret;
1250
5eb91c98
DG
1251 for (i = 0; i < nb_fd; i++) {
1252 /* Fetch once the poll data */
1253 revents = LTTNG_POLL_GETEV(&events, i);
1254 pollfd = LTTNG_POLL_GETFD(&events, i);
1255
840cb59c 1256 health_code_update();
44a5e5eb 1257
5eb91c98 1258 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1259 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1260 if (ret) {
139ac872
MD
1261 err = 0;
1262 goto exit;
5eb91c98 1263 }
099e26bd 1264
5eb91c98
DG
1265 /* Inspect the apps cmd pipe */
1266 if (pollfd == apps_cmd_pipe[0]) {
1267 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1268 ERR("Apps command pipe error");
0177d773 1269 goto error;
5eb91c98 1270 } else if (revents & LPOLLIN) {
d0b96690
DG
1271 int sock;
1272
5eb91c98 1273 /* Empty pipe */
f921c78f 1274 do {
d0b96690 1275 ret = read(apps_cmd_pipe[0], &sock, sizeof(sock));
f921c78f 1276 } while (ret < 0 && errno == EINTR);
d0b96690 1277 if (ret < 0 || ret < sizeof(sock)) {
76d7553f 1278 PERROR("read apps cmd pipe");
5eb91c98
DG
1279 goto error;
1280 }
099e26bd 1281
840cb59c 1282 health_code_update();
44a5e5eb 1283
ffe60014 1284 /*
d0b96690
DG
1285 * We only monitor the error events of the socket. This
1286 * thread does not handle any incoming data from UST
1287 * (POLLIN).
ffe60014 1288 */
d0b96690
DG
1289 ret = lttng_poll_add(&events, sock,
1290 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1291 if (ret < 0) {
5eb91c98 1292 goto error;
e0c7ec2b 1293 }
acc7b41b 1294
d0b96690 1295 DBG("Apps with sock %d added to poll set", sock);
487cf67c 1296
840cb59c 1297 health_code_update();
44a5e5eb 1298
5eb91c98 1299 break;
0177d773 1300 }
5eb91c98
DG
1301 } else {
1302 /*
54d01ffb
DG
1303 * At this point, we know that a registered application made
1304 * the event at poll_wait.
5eb91c98
DG
1305 */
1306 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1307 /* Removing from the poll set */
1308 ret = lttng_poll_del(&events, pollfd);
1309 if (ret < 0) {
1310 goto error;
1311 }
099e26bd 1312
b9d9b220 1313 /* Socket closed on remote end. */
56fff090 1314 ust_app_unregister(pollfd);
5eb91c98
DG
1315 break;
1316 }
099e26bd 1317 }
44a5e5eb 1318
840cb59c 1319 health_code_update();
099e26bd 1320 }
099e26bd
DG
1321 }
1322
139ac872 1323exit:
099e26bd 1324error:
5eb91c98 1325 lttng_poll_clean(&events);
76d7553f 1326error_poll_create:
6993eeb3 1327error_testpoint:
6620da75
DG
1328 utils_close_pipe(apps_cmd_pipe);
1329 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1330
1331 /*
1332 * We don't clean the UST app hash table here since already registered
1333 * applications can still be controlled so let them be until the session
1334 * daemon dies or the applications stop.
1335 */
1336
139ac872 1337 if (err) {
840cb59c 1338 health_error();
139ac872
MD
1339 ERR("Health error occurred in %s", __func__);
1340 }
927ca06a 1341 health_unregister();
76d7553f 1342 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1343 rcu_thread_offline();
1344 rcu_unregister_thread();
099e26bd
DG
1345 return NULL;
1346}
1347
d0b96690 1348/*
d88aee68
DG
1349 * Send a socket to a thread This is called from the dispatch UST registration
1350 * thread once all sockets are set for the application.
d0b96690 1351 *
a71aab04
DG
1352 * The sock value can be invalid, we don't really care, the thread will handle
1353 * it and make the necessary cleanup if so.
1354 *
d0b96690
DG
1355 * On success, return 0 else a negative value being the errno message of the
1356 * write().
1357 */
d88aee68 1358static int send_socket_to_thread(int fd, int sock)
d0b96690
DG
1359{
1360 int ret;
1361
a71aab04
DG
1362 /*
1363 * It's possible that the FD is set as invalid with -1 concurrently just
1364 * before calling this function being a shutdown state of the thread.
1365 */
1366 if (fd < 0) {
1367 ret = -EBADF;
1368 goto error;
1369 }
d0b96690
DG
1370
1371 do {
d88aee68 1372 ret = write(fd, &sock, sizeof(sock));
d0b96690 1373 } while (ret < 0 && errno == EINTR);
d88aee68
DG
1374 if (ret < 0 || ret != sizeof(sock)) {
1375 PERROR("write apps pipe %d", fd);
d0b96690
DG
1376 if (ret < 0) {
1377 ret = -errno;
1378 }
1379 goto error;
1380 }
1381
1382 /* All good. Don't send back the write positive ret value. */
1383 ret = 0;
1384error:
1385 return ret;
1386}
1387
f45e313d
DG
1388/*
1389 * Sanitize the wait queue of the dispatch registration thread meaning removing
1390 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1391 * notify socket is never received.
1392 */
1393static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1394{
1395 int ret, nb_fd = 0, i;
1396 unsigned int fd_added = 0;
1397 struct lttng_poll_event events;
1398 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1399
1400 assert(wait_queue);
1401
1402 lttng_poll_init(&events);
1403
1404 /* Just skip everything for an empty queue. */
1405 if (!wait_queue->count) {
1406 goto end;
1407 }
1408
1409 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1410 if (ret < 0) {
1411 goto error_create;
1412 }
1413
1414 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1415 &wait_queue->head, head) {
1416 assert(wait_node->app);
1417 ret = lttng_poll_add(&events, wait_node->app->sock,
1418 LPOLLHUP | LPOLLERR);
1419 if (ret < 0) {
1420 goto error;
1421 }
1422
1423 fd_added = 1;
1424 }
1425
1426 if (!fd_added) {
1427 goto end;
1428 }
1429
1430 /*
1431 * Poll but don't block so we can quickly identify the faulty events and
1432 * clean them afterwards from the wait queue.
1433 */
1434 ret = lttng_poll_wait(&events, 0);
1435 if (ret < 0) {
1436 goto error;
1437 }
1438 nb_fd = ret;
1439
1440 for (i = 0; i < nb_fd; i++) {
1441 /* Get faulty FD. */
1442 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1443 int pollfd = LTTNG_POLL_GETFD(&events, i);
1444
1445 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1446 &wait_queue->head, head) {
1447 if (pollfd == wait_node->app->sock &&
1448 (revents & (LPOLLHUP | LPOLLERR))) {
1449 cds_list_del(&wait_node->head);
1450 wait_queue->count--;
1451 ust_app_destroy(wait_node->app);
1452 free(wait_node);
1453 break;
1454 }
1455 }
1456 }
1457
1458 if (nb_fd > 0) {
1459 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1460 }
1461
1462end:
1463 lttng_poll_clean(&events);
1464 return;
1465
1466error:
1467 lttng_poll_clean(&events);
1468error_create:
1469 ERR("Unable to sanitize wait queue");
1470 return;
1471}
1472
099e26bd
DG
1473/*
1474 * Dispatch request from the registration threads to the application
1475 * communication thread.
1476 */
1477static void *thread_dispatch_ust_registration(void *data)
1478{
12e2b881 1479 int ret, err = -1;
099e26bd
DG
1480 struct cds_wfq_node *node;
1481 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1482 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1483 struct ust_reg_wait_queue wait_queue = {
1484 .count = 0,
1485 };
d0b96690 1486
12e2b881
MD
1487 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1488
1489 health_code_update();
1490
f45e313d 1491 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1492
1493 DBG("[thread] Dispatch UST command started");
1494
26c9d55e 1495 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1496 health_code_update();
1497
099e26bd
DG
1498 /* Atomically prepare the queue futex */
1499 futex_nto1_prepare(&ust_cmd_queue.futex);
1500
1501 do {
d0b96690 1502 struct ust_app *app = NULL;
7972aab2 1503 ust_cmd = NULL;
d0b96690 1504
f45e313d
DG
1505 /*
1506 * Make sure we don't have node(s) that have hung up before receiving
1507 * the notify socket. This is to clean the list in order to avoid
1508 * memory leaks from notify socket that are never seen.
1509 */
1510 sanitize_wait_queue(&wait_queue);
1511
12e2b881 1512 health_code_update();
099e26bd
DG
1513 /* Dequeue command for registration */
1514 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1515 if (node == NULL) {
00a17c97 1516 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1517 /* Continue thread execution */
1518 break;
1519 }
1520
1521 ust_cmd = caa_container_of(node, struct ust_command, node);
1522
2f50c8a3
DG
1523 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1524 " gid:%d sock:%d name:%s (version %d.%d)",
1525 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1526 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1527 ust_cmd->sock, ust_cmd->reg_msg.name,
1528 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1529
1530 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1531 wait_node = zmalloc(sizeof(*wait_node));
1532 if (!wait_node) {
1533 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1534 ret = close(ust_cmd->sock);
1535 if (ret < 0) {
1536 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1537 }
caa25971 1538 lttng_fd_put(LTTNG_FD_APPS, 1);
7972aab2 1539 free(ust_cmd);
d0b96690
DG
1540 goto error;
1541 }
1542 CDS_INIT_LIST_HEAD(&wait_node->head);
1543
1544 /* Create application object if socket is CMD. */
1545 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1546 ust_cmd->sock);
1547 if (!wait_node->app) {
1548 ret = close(ust_cmd->sock);
1549 if (ret < 0) {
1550 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1551 }
caa25971 1552 lttng_fd_put(LTTNG_FD_APPS, 1);
d88aee68 1553 free(wait_node);
7972aab2 1554 free(ust_cmd);
d0b96690
DG
1555 continue;
1556 }
1557 /*
1558 * Add application to the wait queue so we can set the notify
1559 * socket before putting this object in the global ht.
1560 */
f45e313d
DG
1561 cds_list_add(&wait_node->head, &wait_queue.head);
1562 wait_queue.count++;
d0b96690 1563
7972aab2 1564 free(ust_cmd);
d0b96690
DG
1565 /*
1566 * We have to continue here since we don't have the notify
1567 * socket and the application MUST be added to the hash table
1568 * only at that moment.
1569 */
1570 continue;
1571 } else {
1572 /*
1573 * Look for the application in the local wait queue and set the
1574 * notify socket if found.
1575 */
d88aee68 1576 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1577 &wait_queue.head, head) {
12e2b881 1578 health_code_update();
d0b96690
DG
1579 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1580 wait_node->app->notify_sock = ust_cmd->sock;
1581 cds_list_del(&wait_node->head);
f45e313d 1582 wait_queue.count--;
d0b96690
DG
1583 app = wait_node->app;
1584 free(wait_node);
1585 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1586 break;
1587 }
1588 }
020d7f60
DG
1589
1590 /*
1591 * With no application at this stage the received socket is
1592 * basically useless so close it before we free the cmd data
1593 * structure for good.
1594 */
1595 if (!app) {
1596 ret = close(ust_cmd->sock);
1597 if (ret < 0) {
1598 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1599 }
caa25971 1600 lttng_fd_put(LTTNG_FD_APPS, 1);
020d7f60 1601 }
7972aab2 1602 free(ust_cmd);
d0b96690
DG
1603 }
1604
1605 if (app) {
d0b96690
DG
1606 /*
1607 * @session_lock_list
1608 *
1609 * Lock the global session list so from the register up to the
1610 * registration done message, no thread can see the application
1611 * and change its state.
1612 */
1613 session_lock_list();
1614 rcu_read_lock();
d88aee68 1615
d0b96690
DG
1616 /*
1617 * Add application to the global hash table. This needs to be
1618 * done before the update to the UST registry can locate the
1619 * application.
1620 */
1621 ust_app_add(app);
d88aee68
DG
1622
1623 /* Set app version. This call will print an error if needed. */
1624 (void) ust_app_version(app);
1625
1626 /* Send notify socket through the notify pipe. */
1627 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1628 app->notify_sock);
1629 if (ret < 0) {
1630 rcu_read_unlock();
1631 session_unlock_list();
a71aab04
DG
1632 /*
1633 * No notify thread, stop the UST tracing. However, this is
1634 * not an internal error of the this thread thus setting
1635 * the health error code to a normal exit.
1636 */
1637 err = 0;
d88aee68 1638 goto error;
6620da75 1639 }
d88aee68 1640
d0b96690
DG
1641 /*
1642 * Update newly registered application with the tracing
1643 * registry info already enabled information.
1644 */
1645 update_ust_app(app->sock);
d88aee68
DG
1646
1647 /*
1648 * Don't care about return value. Let the manage apps threads
1649 * handle app unregistration upon socket close.
1650 */
1651 (void) ust_app_register_done(app->sock);
1652
1653 /*
1654 * Even if the application socket has been closed, send the app
1655 * to the thread and unregistration will take place at that
1656 * place.
1657 */
1658 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 1659 if (ret < 0) {
d88aee68
DG
1660 rcu_read_unlock();
1661 session_unlock_list();
a71aab04
DG
1662 /*
1663 * No apps. thread, stop the UST tracing. However, this is
1664 * not an internal error of the this thread thus setting
1665 * the health error code to a normal exit.
1666 */
1667 err = 0;
d88aee68 1668 goto error;
d0b96690 1669 }
d88aee68 1670
d0b96690
DG
1671 rcu_read_unlock();
1672 session_unlock_list();
099e26bd 1673 }
099e26bd
DG
1674 } while (node != NULL);
1675
12e2b881 1676 health_poll_entry();
099e26bd
DG
1677 /* Futex wait on queue. Blocking call on futex() */
1678 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 1679 health_poll_exit();
099e26bd 1680 }
12e2b881
MD
1681 /* Normal exit, no error */
1682 err = 0;
099e26bd
DG
1683
1684error:
d88aee68
DG
1685 /* Clean up wait queue. */
1686 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1687 &wait_queue.head, head) {
d88aee68 1688 cds_list_del(&wait_node->head);
f45e313d 1689 wait_queue.count--;
d88aee68
DG
1690 free(wait_node);
1691 }
1692
099e26bd 1693 DBG("Dispatch thread dying");
12e2b881
MD
1694 if (err) {
1695 health_error();
1696 ERR("Health error occurred in %s", __func__);
1697 }
1698 health_unregister();
099e26bd
DG
1699 return NULL;
1700}
1701
1702/*
1703 * This thread manage application registration.
1704 */
1705static void *thread_registration_apps(void *data)
1d4b027a 1706{
139ac872 1707 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
1708 uint32_t revents, nb_fd;
1709 struct lttng_poll_event events;
099e26bd
DG
1710 /*
1711 * Get allocated in this thread, enqueued to a global queue, dequeued and
1712 * freed in the manage apps thread.
1713 */
1714 struct ust_command *ust_cmd = NULL;
1d4b027a 1715
099e26bd 1716 DBG("[thread] Manage application registration started");
1d4b027a 1717
927ca06a
DG
1718 health_register(HEALTH_TYPE_APP_REG);
1719
6993eeb3
CB
1720 if (testpoint(thread_registration_apps)) {
1721 goto error_testpoint;
1722 }
8ac94142 1723
1d4b027a
DG
1724 ret = lttcomm_listen_unix_sock(apps_sock);
1725 if (ret < 0) {
76d7553f 1726 goto error_listen;
1d4b027a
DG
1727 }
1728
5eb91c98
DG
1729 /*
1730 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1731 * more will be added to this poll set.
1732 */
d0b96690 1733 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1734 if (ret < 0) {
76d7553f 1735 goto error_create_poll;
5eb91c98 1736 }
273ea72c 1737
5eb91c98
DG
1738 /* Add the application registration socket */
1739 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1740 if (ret < 0) {
76d7553f 1741 goto error_poll_add;
5eb91c98 1742 }
273ea72c 1743
1d4b027a 1744 /* Notify all applications to register */
0fdd1e2c
DG
1745 ret = notify_ust_apps(1);
1746 if (ret < 0) {
1747 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
1748 "Execution continues but there might be problem for already\n"
1749 "running applications that wishes to register.");
0fdd1e2c 1750 }
1d4b027a
DG
1751
1752 while (1) {
1753 DBG("Accepting application registration");
273ea72c
DG
1754
1755 /* Inifinite blocking call, waiting for transmission */
88f2b785 1756 restart:
a78af745 1757 health_poll_entry();
5eb91c98 1758 ret = lttng_poll_wait(&events, -1);
a78af745 1759 health_poll_exit();
273ea72c 1760 if (ret < 0) {
88f2b785
MD
1761 /*
1762 * Restart interrupted system call.
1763 */
1764 if (errno == EINTR) {
1765 goto restart;
1766 }
273ea72c
DG
1767 goto error;
1768 }
1769
0d9c5d77
DG
1770 nb_fd = ret;
1771
5eb91c98 1772 for (i = 0; i < nb_fd; i++) {
840cb59c 1773 health_code_update();
139ac872 1774
5eb91c98
DG
1775 /* Fetch once the poll data */
1776 revents = LTTNG_POLL_GETEV(&events, i);
1777 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 1778
5eb91c98 1779 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1780 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1781 if (ret) {
139ac872
MD
1782 err = 0;
1783 goto exit;
90014c57 1784 }
1d4b027a 1785
5eb91c98
DG
1786 /* Event on the registration socket */
1787 if (pollfd == apps_sock) {
1788 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1789 ERR("Register apps socket poll error");
1790 goto error;
1791 } else if (revents & LPOLLIN) {
1792 sock = lttcomm_accept_unix_sock(apps_sock);
1793 if (sock < 0) {
1794 goto error;
1795 }
099e26bd 1796
af5e22f2
DG
1797 /*
1798 * Set socket timeout for both receiving and ending.
1799 * app_socket_timeout is in seconds, whereas
1800 * lttcomm_setsockopt_rcv_timeout and
1801 * lttcomm_setsockopt_snd_timeout expect msec as
1802 * parameter.
1803 */
1804 (void) lttcomm_setsockopt_rcv_timeout(sock,
b51183ae 1805 app_socket_timeout);
af5e22f2 1806 (void) lttcomm_setsockopt_snd_timeout(sock,
b51183ae 1807 app_socket_timeout);
af5e22f2 1808
b662582b
DG
1809 /*
1810 * Set the CLOEXEC flag. Return code is useless because
1811 * either way, the show must go on.
1812 */
1813 (void) utils_set_fd_cloexec(sock);
1814
5eb91c98 1815 /* Create UST registration command for enqueuing */
ba7f0ae5 1816 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 1817 if (ust_cmd == NULL) {
76d7553f 1818 PERROR("ust command zmalloc");
5eb91c98
DG
1819 goto error;
1820 }
1d4b027a 1821
5eb91c98
DG
1822 /*
1823 * Using message-based transmissions to ensure we don't
1824 * have to deal with partially received messages.
1825 */
4063050c
MD
1826 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1827 if (ret < 0) {
1828 ERR("Exhausted file descriptors allowed for applications.");
1829 free(ust_cmd);
1830 ret = close(sock);
1831 if (ret) {
1832 PERROR("close");
1833 }
1834 sock = -1;
1835 continue;
1836 }
d88aee68 1837
840cb59c 1838 health_code_update();
d0b96690
DG
1839 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1840 if (ret < 0) {
5eb91c98 1841 free(ust_cmd);
d0b96690 1842 /* Close socket of the application. */
76d7553f
MD
1843 ret = close(sock);
1844 if (ret) {
1845 PERROR("close");
1846 }
4063050c 1847 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 1848 sock = -1;
5eb91c98
DG
1849 continue;
1850 }
840cb59c 1851 health_code_update();
099e26bd 1852
5eb91c98 1853 ust_cmd->sock = sock;
34a2494f 1854 sock = -1;
099e26bd 1855
5eb91c98
DG
1856 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1857 " gid:%d sock:%d name:%s (version %d.%d)",
1858 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1859 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1860 ust_cmd->sock, ust_cmd->reg_msg.name,
1861 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 1862
5eb91c98
DG
1863 /*
1864 * Lock free enqueue the registration request. The red pill
54d01ffb 1865 * has been taken! This apps will be part of the *system*.
5eb91c98
DG
1866 */
1867 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1868
1869 /*
1870 * Wake the registration queue futex. Implicit memory
1871 * barrier with the exchange in cds_wfq_enqueue.
1872 */
1873 futex_nto1_wake(&ust_cmd_queue.futex);
1874 }
1875 }
90014c57 1876 }
1d4b027a
DG
1877 }
1878
139ac872 1879exit:
1d4b027a 1880error:
139ac872 1881 if (err) {
840cb59c 1882 health_error();
139ac872
MD
1883 ERR("Health error occurred in %s", __func__);
1884 }
139ac872 1885
0fdd1e2c
DG
1886 /* Notify that the registration thread is gone */
1887 notify_ust_apps(0);
1888
a4b35e07 1889 if (apps_sock >= 0) {
76d7553f
MD
1890 ret = close(apps_sock);
1891 if (ret) {
1892 PERROR("close");
1893 }
a4b35e07 1894 }
46c3f085 1895 if (sock >= 0) {
76d7553f
MD
1896 ret = close(sock);
1897 if (ret) {
1898 PERROR("close");
1899 }
4063050c 1900 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 1901 }
273ea72c 1902 unlink(apps_unix_sock_path);
0fdd1e2c 1903
76d7553f 1904error_poll_add:
5eb91c98 1905 lttng_poll_clean(&events);
76d7553f
MD
1906error_listen:
1907error_create_poll:
6993eeb3 1908error_testpoint:
76d7553f 1909 DBG("UST Registration thread cleanup complete");
927ca06a 1910 health_unregister();
5eb91c98 1911
1d4b027a
DG
1912 return NULL;
1913}
1914
8c0faa1d 1915/*
3bd1e081 1916 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 1917 * exec or it will fails.
8c0faa1d 1918 */
3bd1e081 1919static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 1920{
a23ec3a7 1921 int ret, clock_ret;
ee0b0061
DG
1922 struct timespec timeout;
1923
a23ec3a7
DG
1924 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1925 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 1926
a23ec3a7
DG
1927 /* Setup pthread condition */
1928 ret = pthread_condattr_init(&consumer_data->condattr);
1929 if (ret != 0) {
1930 errno = ret;
1931 PERROR("pthread_condattr_init consumer data");
1932 goto error;
1933 }
1934
1935 /*
1936 * Set the monotonic clock in order to make sure we DO NOT jump in time
1937 * between the clock_gettime() call and the timedwait call. See bug #324
1938 * for a more details and how we noticed it.
1939 */
1940 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1941 if (ret != 0) {
1942 errno = ret;
1943 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
1944 goto error;
1945 }
8c0faa1d 1946
a23ec3a7
DG
1947 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1948 if (ret != 0) {
1949 errno = ret;
1950 PERROR("pthread_cond_init consumer data");
1951 goto error;
1952 }
1953
1954 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1955 consumer_data);
8c0faa1d 1956 if (ret != 0) {
3bd1e081 1957 PERROR("pthread_create consumer");
ee0b0061 1958 ret = -1;
8c0faa1d
DG
1959 goto error;
1960 }
1961
a23ec3a7
DG
1962 /* We are about to wait on a pthread condition */
1963 pthread_mutex_lock(&consumer_data->cond_mutex);
1964
ee0b0061 1965 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
1966 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1967 /*
1968 * Set the timeout for the condition timed wait even if the clock gettime
1969 * call fails since we might loop on that call and we want to avoid to
1970 * increment the timeout too many times.
1971 */
1972 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1973
1974 /*
1975 * The following loop COULD be skipped in some conditions so this is why we
1976 * set ret to 0 in order to make sure at least one round of the loop is
1977 * done.
1978 */
1979 ret = 0;
1980
1981 /*
1982 * Loop until the condition is reached or when a timeout is reached. Note
1983 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1984 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1985 * possible. This loop does not take any chances and works with both of
1986 * them.
1987 */
1988 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1989 if (clock_ret < 0) {
1990 PERROR("clock_gettime spawn consumer");
1991 /* Infinite wait for the consumerd thread to be ready */
1992 ret = pthread_cond_wait(&consumer_data->cond,
1993 &consumer_data->cond_mutex);
1994 } else {
1995 ret = pthread_cond_timedwait(&consumer_data->cond,
1996 &consumer_data->cond_mutex, &timeout);
1997 }
ee0b0061 1998 }
8c0faa1d 1999
a23ec3a7
DG
2000 /* Release the pthread condition */
2001 pthread_mutex_unlock(&consumer_data->cond_mutex);
2002
2003 if (ret != 0) {
2004 errno = ret;
2005 if (ret == ETIMEDOUT) {
ee0b0061
DG
2006 /*
2007 * Call has timed out so we kill the kconsumerd_thread and return
2008 * an error.
2009 */
a23ec3a7
DG
2010 ERR("Condition timed out. The consumer thread was never ready."
2011 " Killing it");
3bd1e081 2012 ret = pthread_cancel(consumer_data->thread);
ee0b0061 2013 if (ret < 0) {
3bd1e081 2014 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
2015 }
2016 } else {
a23ec3a7 2017 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061
DG
2018 }
2019 goto error;
2020 }
2021
3bd1e081
MD
2022 pthread_mutex_lock(&consumer_data->pid_mutex);
2023 if (consumer_data->pid == 0) {
a23ec3a7 2024 ERR("Consumerd did not start");
3bd1e081 2025 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2026 goto error;
2027 }
3bd1e081 2028 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2029
8c0faa1d
DG
2030 return 0;
2031
2032error:
2033 return ret;
2034}
2035
d9800920 2036/*
3bd1e081 2037 * Join consumer thread
d9800920 2038 */
3bd1e081 2039static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2040{
2041 void *status;
cf3af59e 2042
e8209f6b
DG
2043 /* Consumer pid must be a real one. */
2044 if (consumer_data->pid > 0) {
c617c0c6 2045 int ret;
3bd1e081 2046 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2047 if (ret) {
3bd1e081 2048 ERR("Error killing consumer daemon");
cf3af59e
MD
2049 return ret;
2050 }
3bd1e081 2051 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2052 } else {
2053 return 0;
2054 }
2055}
2056
8c0faa1d 2057/*
3bd1e081 2058 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2059 *
d063d709 2060 * Return pid if successful else -1.
8c0faa1d 2061 */
3bd1e081 2062static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2063{
2064 int ret;
2065 pid_t pid;
94c55f17 2066 const char *consumer_to_use;
53086306 2067 const char *verbosity;
94c55f17 2068 struct stat st;
8c0faa1d 2069
3bd1e081 2070 DBG("Spawning consumerd");
c49dc785 2071
8c0faa1d
DG
2072 pid = fork();
2073 if (pid == 0) {
2074 /*
3bd1e081 2075 * Exec consumerd.
8c0faa1d 2076 */
daee5345 2077 if (opt_verbose_consumer) {
53086306
DG
2078 verbosity = "--verbose";
2079 } else {
2080 verbosity = "--quiet";
2081 }
3bd1e081
MD
2082 switch (consumer_data->type) {
2083 case LTTNG_CONSUMER_KERNEL:
94c55f17 2084 /*
c7704d57
DG
2085 * Find out which consumerd to execute. We will first try the
2086 * 64-bit path, then the sessiond's installation directory, and
2087 * fallback on the 32-bit one,
94c55f17 2088 */
63a799e8
AM
2089 DBG3("Looking for a kernel consumer at these locations:");
2090 DBG3(" 1) %s", consumerd64_bin);
2091 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2092 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2093 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2094 DBG3("Found location #1");
94c55f17 2095 consumer_to_use = consumerd64_bin;
94c55f17 2096 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2097 DBG3("Found location #2");
94c55f17 2098 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2099 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2100 DBG3("Found location #3");
eb1e0bd4 2101 consumer_to_use = consumerd32_bin;
94c55f17 2102 } else {
63a799e8 2103 DBG("Could not find any valid consumerd executable");
94c55f17
AM
2104 break;
2105 }
2106 DBG("Using kernel consumer at: %s", consumer_to_use);
2107 execl(consumer_to_use,
2108 "lttng-consumerd", verbosity, "-k",
2109 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2110 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2111 NULL);
3bd1e081 2112 break;
7753dea8
MD
2113 case LTTNG_CONSUMER64_UST:
2114 {
b1e0b6b6 2115 char *tmpnew = NULL;
8f4905da
MD
2116
2117 if (consumerd64_libdir[0] != '\0') {
2118 char *tmp;
2119 size_t tmplen;
2120
2121 tmp = getenv("LD_LIBRARY_PATH");
2122 if (!tmp) {
2123 tmp = "";
2124 }
2125 tmplen = strlen("LD_LIBRARY_PATH=")
2126 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2127 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2128 if (!tmpnew) {
2129 ret = -ENOMEM;
2130 goto error;
2131 }
2132 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2133 strcat(tmpnew, consumerd64_libdir);
2134 if (tmp[0] != '\0') {
2135 strcat(tmpnew, ":");
2136 strcat(tmpnew, tmp);
2137 }
2138 ret = putenv(tmpnew);
2139 if (ret) {
2140 ret = -errno;
c6f76da9 2141 free(tmpnew);
8f4905da
MD
2142 goto error;
2143 }
2144 }
94c55f17 2145 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2146 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2147 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2148 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2149 NULL);
8f4905da
MD
2150 if (consumerd64_libdir[0] != '\0') {
2151 free(tmpnew);
2152 }
2153 if (ret) {
2154 goto error;
2155 }
3bd1e081 2156 break;
7753dea8
MD
2157 }
2158 case LTTNG_CONSUMER32_UST:
2159 {
937dde8e 2160 char *tmpnew = NULL;
8f4905da
MD
2161
2162 if (consumerd32_libdir[0] != '\0') {
2163 char *tmp;
2164 size_t tmplen;
2165
2166 tmp = getenv("LD_LIBRARY_PATH");
2167 if (!tmp) {
2168 tmp = "";
2169 }
2170 tmplen = strlen("LD_LIBRARY_PATH=")
2171 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2172 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2173 if (!tmpnew) {
2174 ret = -ENOMEM;
2175 goto error;
2176 }
2177 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2178 strcat(tmpnew, consumerd32_libdir);
2179 if (tmp[0] != '\0') {
2180 strcat(tmpnew, ":");
2181 strcat(tmpnew, tmp);
2182 }
2183 ret = putenv(tmpnew);
2184 if (ret) {
2185 ret = -errno;
c6f76da9 2186 free(tmpnew);
8f4905da
MD
2187 goto error;
2188 }
2189 }
94c55f17 2190 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2191 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2192 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2193 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2194 NULL);
8f4905da
MD
2195 if (consumerd32_libdir[0] != '\0') {
2196 free(tmpnew);
2197 }
2198 if (ret) {
2199 goto error;
2200 }
7753dea8
MD
2201 break;
2202 }
3bd1e081 2203 default:
76d7553f 2204 PERROR("unknown consumer type");
3bd1e081
MD
2205 exit(EXIT_FAILURE);
2206 }
8c0faa1d 2207 if (errno != 0) {
76d7553f 2208 PERROR("kernel start consumer exec");
8c0faa1d
DG
2209 }
2210 exit(EXIT_FAILURE);
2211 } else if (pid > 0) {
2212 ret = pid;
8c0faa1d 2213 } else {
76d7553f 2214 PERROR("start consumer fork");
8c0faa1d 2215 ret = -errno;
8c0faa1d 2216 }
8f4905da 2217error:
8c0faa1d
DG
2218 return ret;
2219}
2220
693bd40b 2221/*
3bd1e081 2222 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2223 */
3bd1e081 2224static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2225{
c617c0c6 2226 int ret;
edb8b045
DG
2227
2228 /*
2229 * Set the listen() state on the socket since there is a possible race
2230 * between the exec() of the consumer daemon and this call if place in the
2231 * consumer thread. See bug #366 for more details.
2232 */
2233 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2234 if (ret < 0) {
2235 goto error;
2236 }
693bd40b 2237
3bd1e081
MD
2238 pthread_mutex_lock(&consumer_data->pid_mutex);
2239 if (consumer_data->pid != 0) {
2240 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2241 goto end;
2242 }
693bd40b 2243
3bd1e081 2244 ret = spawn_consumerd(consumer_data);
c49dc785 2245 if (ret < 0) {
3bd1e081
MD
2246 ERR("Spawning consumerd failed");
2247 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2248 goto error;
693bd40b 2249 }
c49dc785 2250
3bd1e081
MD
2251 /* Setting up the consumer_data pid */
2252 consumer_data->pid = ret;
48842b30 2253 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2254 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2255
3bd1e081
MD
2256 DBG2("Spawning consumer control thread");
2257 ret = spawn_consumer_thread(consumer_data);
693bd40b 2258 if (ret < 0) {
3bd1e081 2259 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2260 goto error;
2261 }
2262
c49dc785 2263end:
693bd40b
DG
2264 return 0;
2265
2266error:
331744e3 2267 /* Cleanup already created sockets on error. */
edb8b045 2268 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2269 int err;
2270
edb8b045
DG
2271 err = close(consumer_data->err_sock);
2272 if (err < 0) {
2273 PERROR("close consumer data error socket");
2274 }
2275 }
693bd40b
DG
2276 return ret;
2277}
2278
44a5e5eb 2279/*
139ac872
MD
2280 * Compute health status of each consumer. If one of them is zero (bad
2281 * state), we return 0.
44a5e5eb
DG
2282 */
2283static int check_consumer_health(void)
2284{
2285 int ret;
2286
927ca06a 2287 ret = health_check_state(HEALTH_TYPE_CONSUMER);
44a5e5eb
DG
2288
2289 DBG3("Health consumer check %d", ret);
2290
2291 return ret;
2292}
2293
b73401da 2294/*
096102bd 2295 * Setup necessary data for kernel tracer action.
b73401da 2296 */
096102bd 2297static int init_kernel_tracer(void)
b73401da
DG
2298{
2299 int ret;
b73401da 2300
096102bd
DG
2301 /* Modprobe lttng kernel modules */
2302 ret = modprobe_lttng_control();
b73401da 2303 if (ret < 0) {
b73401da
DG
2304 goto error;
2305 }
2306
096102bd
DG
2307 /* Open debugfs lttng */
2308 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2309 if (kernel_tracer_fd < 0) {
2310 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2311 ret = -1;
2312 goto error_open;
54d01ffb
DG
2313 }
2314
2f77fc4b
DG
2315 /* Validate kernel version */
2316 ret = kernel_validate_version(kernel_tracer_fd);
2317 if (ret < 0) {
2318 goto error_version;
b551a063 2319 }
54d01ffb 2320
2f77fc4b
DG
2321 ret = modprobe_lttng_data();
2322 if (ret < 0) {
2323 goto error_modules;
54d01ffb
DG
2324 }
2325
2f77fc4b
DG
2326 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2327 return 0;
2328
2329error_version:
2330 modprobe_remove_lttng_control();
2331 ret = close(kernel_tracer_fd);
2332 if (ret) {
2333 PERROR("close");
b551a063 2334 }
2f77fc4b 2335 kernel_tracer_fd = -1;
f73fabfd 2336 return LTTNG_ERR_KERN_VERSION;
b551a063 2337
2f77fc4b
DG
2338error_modules:
2339 ret = close(kernel_tracer_fd);
2340 if (ret) {
2341 PERROR("close");
b551a063 2342 }
54d01ffb 2343
2f77fc4b
DG
2344error_open:
2345 modprobe_remove_lttng_control();
54d01ffb
DG
2346
2347error:
2f77fc4b
DG
2348 WARN("No kernel tracer available");
2349 kernel_tracer_fd = -1;
2350 if (!is_root) {
f73fabfd 2351 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2352 } else {
f73fabfd 2353 return LTTNG_ERR_KERN_NA;
2f77fc4b 2354 }
54d01ffb
DG
2355}
2356
2f77fc4b 2357
54d01ffb 2358/*
2f77fc4b
DG
2359 * Copy consumer output from the tracing session to the domain session. The
2360 * function also applies the right modification on a per domain basis for the
2361 * trace files destination directory.
36b588ed
MD
2362 *
2363 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2364 */
2f77fc4b 2365static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2366{
2367 int ret;
2f77fc4b
DG
2368 const char *dir_name;
2369 struct consumer_output *consumer;
2370
2371 assert(session);
2372 assert(session->consumer);
54d01ffb 2373
b551a063
DG
2374 switch (domain) {
2375 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2376 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2377 /*
2378 * XXX: We should audit the session creation and what this function
2379 * does "extra" in order to avoid a destroy since this function is used
2380 * in the domain session creation (kernel and ust) only. Same for UST
2381 * domain.
2382 */
2383 if (session->kernel_session->consumer) {
2384 consumer_destroy_output(session->kernel_session->consumer);
2385 }
2f77fc4b
DG
2386 session->kernel_session->consumer =
2387 consumer_copy_output(session->consumer);
2388 /* Ease our life a bit for the next part */
2389 consumer = session->kernel_session->consumer;
2390 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063
DG
2391 break;
2392 case LTTNG_DOMAIN_UST:
2f77fc4b 2393 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2394 if (session->ust_session->consumer) {
2395 consumer_destroy_output(session->ust_session->consumer);
2396 }
2f77fc4b
DG
2397 session->ust_session->consumer =
2398 consumer_copy_output(session->consumer);
2399 /* Ease our life a bit for the next part */
2400 consumer = session->ust_session->consumer;
2401 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2402 break;
2403 default:
f73fabfd 2404 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2405 goto error;
2406 }
2407
2f77fc4b 2408 /* Append correct directory to subdir */
c30ce0b3
CB
2409 strncat(consumer->subdir, dir_name,
2410 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2411 DBG3("Copy session consumer subdir %s", consumer->subdir);
2412
f73fabfd 2413 ret = LTTNG_OK;
54d01ffb
DG
2414
2415error:
2416 return ret;
2417}
2418
00e2e675 2419/*
2f77fc4b 2420 * Create an UST session and add it to the session ust list.
36b588ed
MD
2421 *
2422 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2423 */
2f77fc4b
DG
2424static int create_ust_session(struct ltt_session *session,
2425 struct lttng_domain *domain)
00e2e675
DG
2426{
2427 int ret;
2f77fc4b 2428 struct ltt_ust_session *lus = NULL;
00e2e675 2429
a4b92340 2430 assert(session);
2f77fc4b
DG
2431 assert(domain);
2432 assert(session->consumer);
a4b92340 2433
2f77fc4b
DG
2434 switch (domain->type) {
2435 case LTTNG_DOMAIN_UST:
2436 break;
2437 default:
2438 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2439 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2440 goto error;
2441 }
2442
2f77fc4b
DG
2443 DBG("Creating UST session");
2444
dec56f6c 2445 lus = trace_ust_create_session(session->id);
2f77fc4b 2446 if (lus == NULL) {
f73fabfd 2447 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2448 goto error;
2449 }
2450
2f77fc4b
DG
2451 lus->uid = session->uid;
2452 lus->gid = session->gid;
2453 session->ust_session = lus;
00e2e675 2454
2f77fc4b
DG
2455 /* Copy session output to the newly created UST session */
2456 ret = copy_session_consumer(domain->type, session);
f73fabfd 2457 if (ret != LTTNG_OK) {
00e2e675
DG
2458 goto error;
2459 }
2460
f73fabfd 2461 return LTTNG_OK;
00e2e675
DG
2462
2463error:
2f77fc4b
DG
2464 free(lus);
2465 session->ust_session = NULL;
00e2e675
DG
2466 return ret;
2467}
2468
2469/*
2f77fc4b 2470 * Create a kernel tracer session then create the default channel.
00e2e675 2471 */
2f77fc4b 2472static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2473{
2474 int ret;
a4b92340 2475
2f77fc4b 2476 DBG("Creating kernel session");
00e2e675 2477
2f77fc4b
DG
2478 ret = kernel_create_session(session, kernel_tracer_fd);
2479 if (ret < 0) {
f73fabfd 2480 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2481 goto error;
2482 }
2483
2f77fc4b
DG
2484 /* Code flow safety */
2485 assert(session->kernel_session);
2486
2487 /* Copy session output to the newly created Kernel session */
2488 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2489 if (ret != LTTNG_OK) {
a4b92340
DG
2490 goto error;
2491 }
2492
2f77fc4b
DG
2493 /* Create directory(ies) on local filesystem. */
2494 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2495 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2496 ret = run_as_mkdir_recursive(
2497 session->kernel_session->consumer->dst.trace_path,
2498 S_IRWXU | S_IRWXG, session->uid, session->gid);
2499 if (ret < 0) {
2500 if (ret != -EEXIST) {
2501 ERR("Trace directory creation error");
00e2e675
DG
2502 goto error;
2503 }
00e2e675 2504 }
2f77fc4b 2505 }
00e2e675 2506
2f77fc4b
DG
2507 session->kernel_session->uid = session->uid;
2508 session->kernel_session->gid = session->gid;
00e2e675 2509
f73fabfd 2510 return LTTNG_OK;
00e2e675 2511
2f77fc4b
DG
2512error:
2513 trace_kernel_destroy_session(session->kernel_session);
2514 session->kernel_session = NULL;
2515 return ret;
2516}
00e2e675 2517
2f77fc4b
DG
2518/*
2519 * Count number of session permitted by uid/gid.
2520 */
2521static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2522{
2523 unsigned int i = 0;
2524 struct ltt_session *session;
07424f16 2525
2f77fc4b
DG
2526 DBG("Counting number of available session for UID %d GID %d",
2527 uid, gid);
2528 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2529 /*
2f77fc4b 2530 * Only list the sessions the user can control.
00e2e675 2531 */
2f77fc4b
DG
2532 if (!session_access_ok(session, uid, gid)) {
2533 continue;
2534 }
2535 i++;
a4b92340 2536 }
2f77fc4b 2537 return i;
00e2e675
DG
2538}
2539
54d01ffb
DG
2540/*
2541 * Process the command requested by the lttng client within the command
2542 * context structure. This function make sure that the return structure (llm)
2543 * is set and ready for transmission before returning.
2544 *
2545 * Return any error encountered or 0 for success.
53a80697
MD
2546 *
2547 * "sock" is only used for special-case var. len data.
36b588ed
MD
2548 *
2549 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2550 */
53a80697
MD
2551static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2552 int *sock_error)
54d01ffb 2553{
f73fabfd 2554 int ret = LTTNG_OK;
44d3bd01 2555 int need_tracing_session = 1;
2e09ba09 2556 int need_domain;
54d01ffb
DG
2557
2558 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2559
53a80697
MD
2560 *sock_error = 0;
2561
2e09ba09
MD
2562 switch (cmd_ctx->lsm->cmd_type) {
2563 case LTTNG_CREATE_SESSION:
2564 case LTTNG_DESTROY_SESSION:
2565 case LTTNG_LIST_SESSIONS:
2566 case LTTNG_LIST_DOMAINS:
2567 case LTTNG_START_TRACE:
2568 case LTTNG_STOP_TRACE:
6d805429 2569 case LTTNG_DATA_PENDING:
2e09ba09 2570 need_domain = 0;
3aace903 2571 break;
2e09ba09
MD
2572 default:
2573 need_domain = 1;
2574 }
2575
2576 if (opt_no_kernel && need_domain
2577 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2578 if (!is_root) {
f73fabfd 2579 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2580 } else {
f73fabfd 2581 ret = LTTNG_ERR_KERN_NA;
531d29f9 2582 }
4fba7219
DG
2583 goto error;
2584 }
2585
8d3113b2
DG
2586 /* Deny register consumer if we already have a spawned consumer. */
2587 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2588 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2589 if (kconsumer_data.pid > 0) {
f73fabfd 2590 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2591 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2592 goto error;
2593 }
2594 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2595 }
2596
54d01ffb
DG
2597 /*
2598 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2599 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2600 * command.
2601 */
2602 switch(cmd_ctx->lsm->cmd_type) {
2603 case LTTNG_LIST_SESSIONS:
2604 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2605 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2606 case LTTNG_LIST_DOMAINS:
2607 case LTTNG_LIST_CHANNELS:
2608 case LTTNG_LIST_EVENTS:
2609 break;
2610 default:
2611 /* Setup lttng message with no payload */
2612 ret = setup_lttng_msg(cmd_ctx, 0);
2613 if (ret < 0) {
2614 /* This label does not try to unlock the session */
2615 goto init_setup_error;
2616 }
2617 }
2618
2619 /* Commands that DO NOT need a session. */
2620 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2621 case LTTNG_CREATE_SESSION:
2e09ba09 2622 case LTTNG_CALIBRATE:
54d01ffb
DG
2623 case LTTNG_LIST_SESSIONS:
2624 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2625 case LTTNG_LIST_TRACEPOINT_FIELDS:
44d3bd01 2626 need_tracing_session = 0;
54d01ffb
DG
2627 break;
2628 default:
2629 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2630 /*
2631 * We keep the session list lock across _all_ commands
2632 * for now, because the per-session lock does not
2633 * handle teardown properly.
2634 */
74babd95 2635 session_lock_list();
54d01ffb
DG
2636 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2637 if (cmd_ctx->session == NULL) {
bba2d65f 2638 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2639 goto error;
2640 } else {
2641 /* Acquire lock for the session */
2642 session_lock(cmd_ctx->session);
2643 }
2644 break;
2645 }
b389abbe 2646
2e09ba09
MD
2647 if (!need_domain) {
2648 goto skip_domain;
2649 }
a4b92340 2650
54d01ffb
DG
2651 /*
2652 * Check domain type for specific "pre-action".
2653 */
2654 switch (cmd_ctx->lsm->domain.type) {
2655 case LTTNG_DOMAIN_KERNEL:
d1f1c568 2656 if (!is_root) {
f73fabfd 2657 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
2658 goto error;
2659 }
2660
54d01ffb 2661 /* Kernel tracer check */
a4b35e07 2662 if (kernel_tracer_fd == -1) {
54d01ffb 2663 /* Basically, load kernel tracer modules */
096102bd
DG
2664 ret = init_kernel_tracer();
2665 if (ret != 0) {
54d01ffb
DG
2666 goto error;
2667 }
2668 }
5eb91c98 2669
5c827ce0
DG
2670 /* Consumer is in an ERROR state. Report back to client */
2671 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2672 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2673 goto error;
2674 }
2675
54d01ffb 2676 /* Need a session for kernel command */
44d3bd01 2677 if (need_tracing_session) {
54d01ffb 2678 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 2679 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 2680 if (ret < 0) {
f73fabfd 2681 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
2682 goto error;
2683 }
b389abbe 2684 }
7d29a247 2685
54d01ffb 2686 /* Start the kernel consumer daemon */
3bd1e081
MD
2687 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2688 if (kconsumer_data.pid == 0 &&
785d2d0d 2689 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3bd1e081
MD
2690 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2691 ret = start_consumerd(&kconsumer_data);
7d29a247 2692 if (ret < 0) {
f73fabfd 2693 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 2694 goto error;
950131af 2695 }
5c827ce0 2696 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
2697 } else {
2698 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 2699 }
173af62f 2700
a4b92340
DG
2701 /*
2702 * The consumer was just spawned so we need to add the socket to
2703 * the consumer output of the session if exist.
2704 */
2705 ret = consumer_create_socket(&kconsumer_data,
2706 cmd_ctx->session->kernel_session->consumer);
2707 if (ret < 0) {
2708 goto error;
173af62f 2709 }
0d0c377a 2710 }
5c827ce0 2711
54d01ffb 2712 break;
2bdd86d4 2713 case LTTNG_DOMAIN_UST:
44d3bd01 2714 {
2f0790d0
MD
2715 if (!ust_app_supported()) {
2716 ret = LTTNG_ERR_NO_UST;
2717 goto error;
2718 }
5c827ce0
DG
2719 /* Consumer is in an ERROR state. Report back to client */
2720 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2721 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2722 goto error;
2723 }
2724
44d3bd01 2725 if (need_tracing_session) {
a4b92340 2726 /* Create UST session if none exist. */
f6a9efaa 2727 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 2728 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 2729 &cmd_ctx->lsm->domain);
f73fabfd 2730 if (ret != LTTNG_OK) {
44d3bd01
DG
2731 goto error;
2732 }
2733 }
00e2e675 2734
7753dea8
MD
2735 /* Start the UST consumer daemons */
2736 /* 64-bit */
2737 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 2738 if (consumerd64_bin[0] != '\0' &&
7753dea8 2739 ustconsumer64_data.pid == 0 &&
785d2d0d 2740 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2741 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2742 ret = start_consumerd(&ustconsumer64_data);
2bdd86d4 2743 if (ret < 0) {
f73fabfd 2744 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
173af62f 2745 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2bdd86d4
MD
2746 goto error;
2747 }
48842b30 2748
173af62f 2749 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
5c827ce0 2750 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3ff2ecac 2751 } else {
7753dea8
MD
2752 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2753 }
173af62f
DG
2754
2755 /*
2756 * Setup socket for consumer 64 bit. No need for atomic access
2757 * since it was set above and can ONLY be set in this thread.
2758 */
a4b92340
DG
2759 ret = consumer_create_socket(&ustconsumer64_data,
2760 cmd_ctx->session->ust_session->consumer);
2761 if (ret < 0) {
2762 goto error;
173af62f
DG
2763 }
2764
7753dea8 2765 /* 32-bit */
fc7a59ce 2766 if (consumerd32_bin[0] != '\0' &&
7753dea8 2767 ustconsumer32_data.pid == 0 &&
785d2d0d 2768 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2769 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2770 ret = start_consumerd(&ustconsumer32_data);
2771 if (ret < 0) {
f73fabfd 2772 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
173af62f 2773 uatomic_set(&ust_consumerd32_fd, -EINVAL);
7753dea8
MD
2774 goto error;
2775 }
5c827ce0 2776
173af62f 2777 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
5c827ce0 2778 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
7753dea8
MD
2779 } else {
2780 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2bdd86d4 2781 }
173af62f
DG
2782
2783 /*
2784 * Setup socket for consumer 64 bit. No need for atomic access
2785 * since it was set above and can ONLY be set in this thread.
2786 */
a4b92340
DG
2787 ret = consumer_create_socket(&ustconsumer32_data,
2788 cmd_ctx->session->ust_session->consumer);
2789 if (ret < 0) {
2790 goto error;
173af62f 2791 }
44d3bd01
DG
2792 }
2793 break;
48842b30 2794 }
54d01ffb 2795 default:
54d01ffb
DG
2796 break;
2797 }
2e09ba09 2798skip_domain:
33a2b854 2799
5c827ce0
DG
2800 /* Validate consumer daemon state when start/stop trace command */
2801 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2802 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2803 switch (cmd_ctx->lsm->domain.type) {
2804 case LTTNG_DOMAIN_UST:
2805 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2806 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2807 goto error;
2808 }
2809 break;
2810 case LTTNG_DOMAIN_KERNEL:
2811 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2812 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2813 goto error;
2814 }
2815 break;
2816 }
2817 }
2818
8e0af1b4
MD
2819 /*
2820 * Check that the UID or GID match that of the tracing session.
2821 * The root user can interact with all sessions.
2822 */
2823 if (need_tracing_session) {
2824 if (!session_access_ok(cmd_ctx->session,
730389d9
DG
2825 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2826 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
f73fabfd 2827 ret = LTTNG_ERR_EPERM;
8e0af1b4
MD
2828 goto error;
2829 }
2830 }
2831
ffe60014
DG
2832 /*
2833 * Send relayd information to consumer as soon as we have a domain and a
2834 * session defined.
2835 */
2836 if (cmd_ctx->session && need_domain) {
2837 /*
2838 * Setup relayd if not done yet. If the relayd information was already
2839 * sent to the consumer, this call will gracefully return.
2840 */
2841 ret = cmd_setup_relayd(cmd_ctx->session);
2842 if (ret != LTTNG_OK) {
2843 goto error;
2844 }
2845 }
2846
54d01ffb
DG
2847 /* Process by command type */
2848 switch (cmd_ctx->lsm->cmd_type) {
2849 case LTTNG_ADD_CONTEXT:
2850 {
2851 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2852 cmd_ctx->lsm->u.context.channel_name,
979e618e 2853 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
54d01ffb
DG
2854 break;
2855 }
2856 case LTTNG_DISABLE_CHANNEL:
2857 {
2858 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2859 cmd_ctx->lsm->u.disable.channel_name);
2860 break;
2861 }
2862 case LTTNG_DISABLE_EVENT:
2863 {
2864 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2865 cmd_ctx->lsm->u.disable.channel_name,
2866 cmd_ctx->lsm->u.disable.name);
33a2b854
DG
2867 break;
2868 }
54d01ffb
DG
2869 case LTTNG_DISABLE_ALL_EVENT:
2870 {
2bdd86d4 2871 DBG("Disabling all events");
54d01ffb
DG
2872
2873 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2874 cmd_ctx->lsm->u.disable.channel_name);
2875 break;
2876 }
2877 case LTTNG_ENABLE_CHANNEL:
2878 {
7972aab2 2879 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2f77fc4b 2880 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
54d01ffb
DG
2881 break;
2882 }
2883 case LTTNG_ENABLE_EVENT:
2884 {
7972aab2 2885 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
54d01ffb 2886 cmd_ctx->lsm->u.enable.channel_name,
025faf73 2887 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
2888 break;
2889 }
2890 case LTTNG_ENABLE_ALL_EVENT:
2891 {
2bdd86d4 2892 DBG("Enabling all events");
54d01ffb 2893
7972aab2 2894 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
8c9ae521 2895 cmd_ctx->lsm->u.enable.channel_name,
025faf73 2896 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
2897 break;
2898 }
052da939 2899 case LTTNG_LIST_TRACEPOINTS:
2ef84c95 2900 {
9f19cc17 2901 struct lttng_event *events;
54d01ffb 2902 ssize_t nb_events;
052da939 2903
54d01ffb
DG
2904 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2905 if (nb_events < 0) {
f73fabfd 2906 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2907 ret = -nb_events;
2908 goto error;
2ef84c95
DG
2909 }
2910
2911 /*
2912 * Setup lttng message with payload size set to the event list size in
2913 * bytes and then copy list into the llm payload.
2914 */
052da939 2915 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2ef84c95 2916 if (ret < 0) {
052da939 2917 free(events);
2ef84c95
DG
2918 goto setup_error;
2919 }
2920
2921 /* Copy event list into message payload */
9f19cc17 2922 memcpy(cmd_ctx->llm->payload, events,
052da939 2923 sizeof(struct lttng_event) * nb_events);
2ef84c95 2924
9f19cc17 2925 free(events);
2ef84c95 2926
f73fabfd 2927 ret = LTTNG_OK;
2ef84c95
DG
2928 break;
2929 }
f37d259d
MD
2930 case LTTNG_LIST_TRACEPOINT_FIELDS:
2931 {
2932 struct lttng_event_field *fields;
2933 ssize_t nb_fields;
2934
a4b92340
DG
2935 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2936 &fields);
f37d259d 2937 if (nb_fields < 0) {
f73fabfd 2938 /* Return value is a negative lttng_error_code. */
f37d259d
MD
2939 ret = -nb_fields;
2940 goto error;
2941 }
2942
2943 /*
2944 * Setup lttng message with payload size set to the event list size in
2945 * bytes and then copy list into the llm payload.
2946 */
a4b92340
DG
2947 ret = setup_lttng_msg(cmd_ctx,
2948 sizeof(struct lttng_event_field) * nb_fields);
f37d259d
MD
2949 if (ret < 0) {
2950 free(fields);
2951 goto setup_error;
2952 }
2953
2954 /* Copy event list into message payload */
2955 memcpy(cmd_ctx->llm->payload, fields,
2956 sizeof(struct lttng_event_field) * nb_fields);
2957
2958 free(fields);
2959
f73fabfd 2960 ret = LTTNG_OK;
f37d259d
MD
2961 break;
2962 }
00e2e675
DG
2963 case LTTNG_SET_CONSUMER_URI:
2964 {
a4b92340
DG
2965 size_t nb_uri, len;
2966 struct lttng_uri *uris;
2967
2968 nb_uri = cmd_ctx->lsm->u.uri.size;
2969 len = nb_uri * sizeof(struct lttng_uri);
2970
2971 if (nb_uri == 0) {
f73fabfd 2972 ret = LTTNG_ERR_INVALID;
a4b92340
DG
2973 goto error;
2974 }
2975
2976 uris = zmalloc(len);
2977 if (uris == NULL) {
f73fabfd 2978 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2979 goto error;
2980 }
2981
2982 /* Receive variable len data */
77c7c900 2983 DBG("Receiving %zu URI(s) from client ...", nb_uri);
a4b92340
DG
2984 ret = lttcomm_recv_unix_sock(sock, uris, len);
2985 if (ret <= 0) {
2986 DBG("No URIs received from client... continuing");
2987 *sock_error = 1;
f73fabfd 2988 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2989 free(uris);
a4b92340
DG
2990 goto error;
2991 }
2992
00e2e675 2993 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
a4b92340 2994 nb_uri, uris);
f73fabfd 2995 if (ret != LTTNG_OK) {
b1d41407 2996 free(uris);
a4b92340
DG
2997 goto error;
2998 }
2999
3000 /*
3001 * XXX: 0 means that this URI should be applied on the session. Should
3002 * be a DOMAIN enuam.
3003 */
3004 if (cmd_ctx->lsm->domain.type == 0) {
3005 /* Add the URI for the UST session if a consumer is present. */
3006 if (cmd_ctx->session->ust_session &&
3007 cmd_ctx->session->ust_session->consumer) {
3008 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3009 nb_uri, uris);
3010 } else if (cmd_ctx->session->kernel_session &&
3011 cmd_ctx->session->kernel_session->consumer) {
3012 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3013 cmd_ctx->session, nb_uri, uris);
3014 }
3015 }
3016
b1d41407
DG
3017 free(uris);
3018
00e2e675
DG
3019 break;
3020 }
f3ed775e 3021 case LTTNG_START_TRACE:
8c0faa1d 3022 {
54d01ffb 3023 ret = cmd_start_trace(cmd_ctx->session);
8c0faa1d
DG
3024 break;
3025 }
f3ed775e 3026 case LTTNG_STOP_TRACE:
8c0faa1d 3027 {
54d01ffb 3028 ret = cmd_stop_trace(cmd_ctx->session);
8c0faa1d
DG
3029 break;
3030 }
5e16da05
MD
3031 case LTTNG_CREATE_SESSION:
3032 {
a4b92340
DG
3033 size_t nb_uri, len;
3034 struct lttng_uri *uris = NULL;
3035
3036 nb_uri = cmd_ctx->lsm->u.uri.size;
3037 len = nb_uri * sizeof(struct lttng_uri);
3038
3039 if (nb_uri > 0) {
3040 uris = zmalloc(len);
3041 if (uris == NULL) {
f73fabfd 3042 ret = LTTNG_ERR_FATAL;
a4b92340
DG
3043 goto error;
3044 }
3045
3046 /* Receive variable len data */
77c7c900 3047 DBG("Waiting for %zu URIs from client ...", nb_uri);
a4b92340
DG
3048 ret = lttcomm_recv_unix_sock(sock, uris, len);
3049 if (ret <= 0) {
3050 DBG("No URIs received from client... continuing");
3051 *sock_error = 1;
f73fabfd 3052 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3053 free(uris);
a4b92340
DG
3054 goto error;
3055 }
3056
3057 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3058 DBG("Creating session with ONE network URI is a bad call");
f73fabfd 3059 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3060 free(uris);
a4b92340
DG
3061 goto error;
3062 }
3063 }
3064
3065 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3066 &cmd_ctx->creds);
3067
b1d41407
DG
3068 free(uris);
3069
00e2e675
DG
3070 break;
3071 }
5e16da05
MD
3072 case LTTNG_DESTROY_SESSION:
3073 {
2f77fc4b 3074 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
a4b92340
DG
3075
3076 /* Set session to NULL so we do not unlock it after free. */
256a5576 3077 cmd_ctx->session = NULL;
5461b305 3078 break;
5e16da05 3079 }
9f19cc17 3080 case LTTNG_LIST_DOMAINS:
5e16da05 3081 {
54d01ffb
DG
3082 ssize_t nb_dom;
3083 struct lttng_domain *domains;
5461b305 3084
54d01ffb
DG
3085 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3086 if (nb_dom < 0) {
f73fabfd 3087 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3088 ret = -nb_dom;
3089 goto error;
ce3d728c 3090 }
520ff687 3091
54d01ffb 3092 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
5461b305 3093 if (ret < 0) {
ae9c20bc 3094 free(domains);
5461b305 3095 goto setup_error;
520ff687 3096 }
57167058 3097
54d01ffb
DG
3098 /* Copy event list into message payload */
3099 memcpy(cmd_ctx->llm->payload, domains,
3100 nb_dom * sizeof(struct lttng_domain));
3101
3102 free(domains);
5e16da05 3103
f73fabfd 3104 ret = LTTNG_OK;
5e16da05
MD
3105 break;
3106 }
9f19cc17 3107 case LTTNG_LIST_CHANNELS:
5e16da05 3108 {
6775595e 3109 int nb_chan;
54d01ffb
DG
3110 struct lttng_channel *channels;
3111
b551a063
DG
3112 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3113 cmd_ctx->session, &channels);
54d01ffb 3114 if (nb_chan < 0) {
f73fabfd 3115 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3116 ret = -nb_chan;
3117 goto error;
5461b305 3118 }
ca95a216 3119
54d01ffb 3120 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
5461b305 3121 if (ret < 0) {
ae9c20bc 3122 free(channels);
5461b305
DG
3123 goto setup_error;
3124 }
9f19cc17 3125
54d01ffb
DG
3126 /* Copy event list into message payload */
3127 memcpy(cmd_ctx->llm->payload, channels,
3128 nb_chan * sizeof(struct lttng_channel));
3129
3130 free(channels);
9f19cc17 3131
f73fabfd 3132 ret = LTTNG_OK;
5461b305 3133 break;
5e16da05 3134 }
9f19cc17 3135 case LTTNG_LIST_EVENTS:
5e16da05 3136 {
b551a063 3137 ssize_t nb_event;
684d34d2 3138 struct lttng_event *events = NULL;
9f19cc17 3139
b551a063 3140 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
54d01ffb
DG
3141 cmd_ctx->lsm->u.list.channel_name, &events);
3142 if (nb_event < 0) {
f73fabfd 3143 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3144 ret = -nb_event;
3145 goto error;
5461b305 3146 }
ca95a216 3147
54d01ffb 3148 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
5461b305 3149 if (ret < 0) {
ae9c20bc 3150 free(events);
5461b305
DG
3151 goto setup_error;
3152 }
9f19cc17 3153
54d01ffb
DG
3154 /* Copy event list into message payload */
3155 memcpy(cmd_ctx->llm->payload, events,
3156 nb_event * sizeof(struct lttng_event));
9f19cc17 3157
54d01ffb 3158 free(events);
9f19cc17 3159
f73fabfd 3160 ret = LTTNG_OK;
5461b305 3161 break;
5e16da05
MD
3162 }
3163 case LTTNG_LIST_SESSIONS:
3164 {
8e0af1b4 3165 unsigned int nr_sessions;
5461b305 3166
8e0af1b4 3167 session_lock_list();
730389d9
DG
3168 nr_sessions = lttng_sessions_count(
3169 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3170 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
d32fb093 3171
8e0af1b4 3172 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
5461b305 3173 if (ret < 0) {
54d01ffb 3174 session_unlock_list();
5461b305 3175 goto setup_error;
e065084a 3176 }
5e16da05 3177
6c9cc2ab 3178 /* Filled the session array */
2f77fc4b 3179 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
730389d9
DG
3180 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3181 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
6c9cc2ab 3182
54d01ffb 3183 session_unlock_list();
5e16da05 3184
f73fabfd 3185 ret = LTTNG_OK;
5e16da05
MD
3186 break;
3187 }
d0254c7c
MD
3188 case LTTNG_CALIBRATE:
3189 {
54d01ffb
DG
3190 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3191 &cmd_ctx->lsm->u.calibrate);
d0254c7c
MD
3192 break;
3193 }
d9800920
DG
3194 case LTTNG_REGISTER_CONSUMER:
3195 {
2f77fc4b
DG
3196 struct consumer_data *cdata;
3197
3198 switch (cmd_ctx->lsm->domain.type) {
3199 case LTTNG_DOMAIN_KERNEL:
3200 cdata = &kconsumer_data;
3201 break;
3202 default:
f73fabfd 3203 ret = LTTNG_ERR_UND;
2f77fc4b
DG
3204 goto error;
3205 }
3206
54d01ffb 3207 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 3208 cmd_ctx->lsm->u.reg.path, cdata);
d9800920
DG
3209 break;
3210 }
025faf73 3211 case LTTNG_ENABLE_EVENT_WITH_FILTER:
53a80697
MD
3212 {
3213 struct lttng_filter_bytecode *bytecode;
3214
025faf73 3215 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
f73fabfd 3216 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
3217 goto error;
3218 }
52df2401
MD
3219 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3220 ret = LTTNG_ERR_FILTER_INVAL;
3221 goto error;
3222 }
025faf73 3223 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
53a80697 3224 if (!bytecode) {
f73fabfd 3225 ret = LTTNG_ERR_FILTER_NOMEM;
53a80697
MD
3226 goto error;
3227 }
3228 /* Receive var. len. data */
3229 DBG("Receiving var len data from client ...");
3230 ret = lttcomm_recv_unix_sock(sock, bytecode,
025faf73 3231 cmd_ctx->lsm->u.enable.bytecode_len);
53a80697
MD
3232 if (ret <= 0) {
3233 DBG("Nothing recv() from client var len data... continuing");
3234 *sock_error = 1;
f73fabfd 3235 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
3236 goto error;
3237 }
3238
3239 if (bytecode->len + sizeof(*bytecode)
025faf73 3240 != cmd_ctx->lsm->u.enable.bytecode_len) {
53a80697 3241 free(bytecode);
f73fabfd 3242 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
3243 goto error;
3244 }
3245
7972aab2 3246 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
025faf73
DG
3247 cmd_ctx->lsm->u.enable.channel_name,
3248 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
53a80697
MD
3249 break;
3250 }
6d805429 3251 case LTTNG_DATA_PENDING:
806e2684 3252 {
6d805429 3253 ret = cmd_data_pending(cmd_ctx->session);
806e2684
DG
3254 break;
3255 }
5e16da05 3256 default:
f73fabfd 3257 ret = LTTNG_ERR_UND;
5461b305 3258 break;
fac6795d
DG
3259 }
3260
5461b305 3261error:
5461b305
DG
3262 if (cmd_ctx->llm == NULL) {
3263 DBG("Missing llm structure. Allocating one.");
894be886 3264 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
5461b305
DG
3265 goto setup_error;
3266 }
3267 }
54d01ffb 3268 /* Set return code */
5461b305 3269 cmd_ctx->llm->ret_code = ret;
5461b305 3270setup_error:
b5541356 3271 if (cmd_ctx->session) {
54d01ffb 3272 session_unlock(cmd_ctx->session);
b5541356 3273 }
256a5576
MD
3274 if (need_tracing_session) {
3275 session_unlock_list();
3276 }
54d01ffb 3277init_setup_error:
8028d920 3278 return ret;
fac6795d
DG
3279}
3280
44a5e5eb
DG
3281/*
3282 * Thread managing health check socket.
3283 */
3284static void *thread_manage_health(void *data)
3285{
eb4a2943 3286 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
44a5e5eb
DG
3287 uint32_t revents, nb_fd;
3288 struct lttng_poll_event events;
3289 struct lttcomm_health_msg msg;
3290 struct lttcomm_health_data reply;
3291
3292 DBG("[thread] Manage health check started");
3293
3294 rcu_register_thread();
3295
6d737ce4
DG
3296 /* We might hit an error path before this is created. */
3297 lttng_poll_init(&events);
3cc04881 3298
44a5e5eb
DG
3299 /* Create unix socket */
3300 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3301 if (sock < 0) {
3302 ERR("Unable to create health check Unix socket");
3303 ret = -1;
3304 goto error;
3305 }
3306
b662582b
DG
3307 /*
3308 * Set the CLOEXEC flag. Return code is useless because either way, the
3309 * show must go on.
3310 */
3311 (void) utils_set_fd_cloexec(sock);
3312
44a5e5eb
DG
3313 ret = lttcomm_listen_unix_sock(sock);
3314 if (ret < 0) {
3315 goto error;
3316 }
3317
3318 /*
3319 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3320 * more will be added to this poll set.
3321 */
d0b96690 3322 ret = sessiond_set_thread_pollset(&events, 2);
44a5e5eb
DG
3323 if (ret < 0) {
3324 goto error;
3325 }
3326
3327 /* Add the application registration socket */
3328 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3329 if (ret < 0) {
3330 goto error;
3331 }
3332
3333 while (1) {
3334 DBG("Health check ready");
3335
44a5e5eb
DG
3336 /* Inifinite blocking call, waiting for transmission */
3337restart:
3338 ret = lttng_poll_wait(&events, -1);
3339 if (ret < 0) {
3340 /*
3341 * Restart interrupted system call.
3342 */
3343 if (errno == EINTR) {
3344 goto restart;
3345 }
3346 goto error;
3347 }
3348
0d9c5d77
DG
3349 nb_fd = ret;
3350
44a5e5eb
DG
3351 for (i = 0; i < nb_fd; i++) {
3352 /* Fetch once the poll data */
3353 revents = LTTNG_POLL_GETEV(&events, i);
3354 pollfd = LTTNG_POLL_GETFD(&events, i);
3355
3356 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 3357 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
44a5e5eb 3358 if (ret) {
139ac872
MD
3359 err = 0;
3360 goto exit;
44a5e5eb
DG
3361 }
3362
3363 /* Event on the registration socket */
3364 if (pollfd == sock) {
3365 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3366 ERR("Health socket poll error");
3367 goto error;
3368 }
3369 }
3370 }
3371
3372 new_sock = lttcomm_accept_unix_sock(sock);
3373 if (new_sock < 0) {
3374 goto error;
3375 }
3376
b662582b
DG
3377 /*
3378 * Set the CLOEXEC flag. Return code is useless because either way, the
3379 * show must go on.
3380 */
3381 (void) utils_set_fd_cloexec(new_sock);
3382
44a5e5eb
DG
3383 DBG("Receiving data from client for health...");
3384 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3385 if (ret <= 0) {
3386 DBG("Nothing recv() from client... continuing");
3387 ret = close(new_sock);
3388 if (ret) {
3389 PERROR("close");
3390 }
3391 new_sock = -1;
3392 continue;
3393 }
3394
3395 rcu_thread_online();
3396
3397 switch (msg.component) {
3398 case LTTNG_HEALTH_CMD:
927ca06a 3399 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
44a5e5eb 3400 break;
139ac872 3401 case LTTNG_HEALTH_APP_MANAGE:
927ca06a 3402 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
139ac872 3403 break;
44a5e5eb 3404 case LTTNG_HEALTH_APP_REG:
927ca06a 3405 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
44a5e5eb
DG
3406 break;
3407 case LTTNG_HEALTH_KERNEL:
927ca06a 3408 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
44a5e5eb
DG
3409 break;
3410 case LTTNG_HEALTH_CONSUMER:
3411 reply.ret_code = check_consumer_health();
3412 break;
0b2dc8df
MD
3413 case LTTNG_HEALTH_HT_CLEANUP:
3414 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3415 break;
380e8d6f
MD
3416 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3417 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3418 break;
12e2b881
MD
3419 case LTTNG_HEALTH_APP_REG_DISPATCH:
3420 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3421 break;
44a5e5eb 3422 case LTTNG_HEALTH_ALL:
44a5e5eb 3423 reply.ret_code =
927ca06a
DG
3424 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3425 health_check_state(HEALTH_TYPE_APP_REG) &&
3426 health_check_state(HEALTH_TYPE_CMD) &&
3427 health_check_state(HEALTH_TYPE_KERNEL) &&
0b2dc8df 3428 check_consumer_health() &&
380e8d6f 3429 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
12e2b881
MD
3430 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3431 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
44a5e5eb
DG
3432 break;
3433 default:
f73fabfd 3434 reply.ret_code = LTTNG_ERR_UND;
44a5e5eb
DG
3435 break;
3436 }
3437
3438 /*
3439 * Flip ret value since 0 is a success and 1 indicates a bad health for
3440 * the client where in the sessiond it is the opposite. Again, this is
3441 * just to make things easier for us poor developer which enjoy a lot
3442 * lazyness.
3443 */
3444 if (reply.ret_code == 0 || reply.ret_code == 1) {
3445 reply.ret_code = !reply.ret_code;
3446 }
3447
3448 DBG2("Health check return value %d", reply.ret_code);
3449
3450 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3451 if (ret < 0) {
3452 ERR("Failed to send health data back to client");
3453 }
3454
3455 /* End of transmission */
3456 ret = close(new_sock);
3457 if (ret) {
3458 PERROR("close");
3459 }
3460 new_sock = -1;
3461 }
3462
139ac872 3463exit:
44a5e5eb 3464error:
139ac872
MD
3465 if (err) {
3466 ERR("Health error occurred in %s", __func__);
3467 }
44a5e5eb
DG
3468 DBG("Health check thread dying");
3469 unlink(health_unix_sock_path);
3470 if (sock >= 0) {
3471 ret = close(sock);
3472 if (ret) {
3473 PERROR("close");
3474 }
3475 }
44a5e5eb
DG
3476
3477 lttng_poll_clean(&events);
3478
3479 rcu_unregister_thread();
3480 return NULL;
3481}
3482
1d4b027a 3483/*
d063d709
DG
3484 * This thread manage all clients request using the unix client socket for
3485 * communication.
1d4b027a
DG
3486 */
3487static void *thread_manage_clients(void *data)
3488{
139ac872 3489 int sock = -1, ret, i, pollfd, err = -1;
53a80697 3490 int sock_error;
5eb91c98 3491 uint32_t revents, nb_fd;
273ea72c 3492 struct command_ctx *cmd_ctx = NULL;
5eb91c98 3493 struct lttng_poll_event events;
1d4b027a
DG
3494
3495 DBG("[thread] Manage client started");
3496
f6a9efaa
DG
3497 rcu_register_thread();
3498
927ca06a
DG
3499 health_register(HEALTH_TYPE_CMD);
3500
6993eeb3
CB
3501 if (testpoint(thread_manage_clients)) {
3502 goto error_testpoint;
3503 }
3504
840cb59c 3505 health_code_update();
44a5e5eb 3506
1d4b027a
DG
3507 ret = lttcomm_listen_unix_sock(client_sock);
3508 if (ret < 0) {
35df2755 3509 goto error_listen;
1d4b027a
DG
3510 }
3511
5eb91c98
DG
3512 /*
3513 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3514 * more will be added to this poll set.
3515 */
d0b96690 3516 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 3517 if (ret < 0) {
35df2755 3518 goto error_create_poll;
5eb91c98 3519 }
273ea72c 3520
5eb91c98
DG
3521 /* Add the application registration socket */
3522 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3523 if (ret < 0) {
3524 goto error;
3525 }
273ea72c 3526
bbd973c2
DG
3527 /*
3528 * Notify parent pid that we are ready to accept command for client side.
1d4b027a
DG
3529 */
3530 if (opt_sig_parent) {
8db8d1dc 3531 kill(ppid, SIGUSR1);
1d4b027a
DG
3532 }
3533
6993eeb3
CB
3534 if (testpoint(thread_manage_clients_before_loop)) {
3535 goto error;
3536 }
8ac94142 3537
840cb59c 3538 health_code_update();
44a5e5eb 3539
1d4b027a 3540 while (1) {
1d4b027a 3541 DBG("Accepting client command ...");
273ea72c
DG
3542
3543 /* Inifinite blocking call, waiting for transmission */
88f2b785 3544 restart:
a78af745 3545 health_poll_entry();
5eb91c98 3546 ret = lttng_poll_wait(&events, -1);
a78af745 3547 health_poll_exit();
273ea72c 3548 if (ret < 0) {
88f2b785
MD
3549 /*
3550 * Restart interrupted system call.
3551 */
3552 if (errno == EINTR) {
3553 goto restart;
3554 }
273ea72c
DG
3555 goto error;
3556 }
3557
0d9c5d77
DG
3558 nb_fd = ret;
3559
5eb91c98
DG
3560 for (i = 0; i < nb_fd; i++) {
3561 /* Fetch once the poll data */
3562 revents = LTTNG_POLL_GETEV(&events, i);
3563 pollfd = LTTNG_POLL_GETFD(&events, i);
3564
840cb59c 3565 health_code_update();
44a5e5eb 3566
5eb91c98 3567 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 3568 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 3569 if (ret) {
139ac872
MD
3570 err = 0;
3571 goto exit;
5eb91c98
DG
3572 }
3573
3574 /* Event on the registration socket */
3575 if (pollfd == client_sock) {
3576 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3577 ERR("Client socket poll error");
3578 goto error;
3579 }
3580 }
3581 }
3582
3583 DBG("Wait for client response");
3584
840cb59c 3585 health_code_update();
44a5e5eb 3586
5eb91c98
DG
3587 sock = lttcomm_accept_unix_sock(client_sock);
3588 if (sock < 0) {
273ea72c 3589 goto error;
273ea72c
DG
3590 }
3591
b662582b
DG
3592 /*
3593 * Set the CLOEXEC flag. Return code is useless because either way, the
3594 * show must go on.
3595 */
3596 (void) utils_set_fd_cloexec(sock);
3597
be040666
DG
3598 /* Set socket option for credentials retrieval */
3599 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3600 if (ret < 0) {
3601 goto error;
3602 }
3603
5eb91c98 3604 /* Allocate context command to process the client request */
ba7f0ae5 3605 cmd_ctx = zmalloc(sizeof(struct command_ctx));
5eb91c98 3606 if (cmd_ctx == NULL) {
76d7553f 3607 PERROR("zmalloc cmd_ctx");
1d4b027a 3608 goto error;
5eb91c98 3609 }
1d4b027a 3610
5eb91c98 3611 /* Allocate data buffer for reception */
ba7f0ae5 3612 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
5eb91c98 3613 if (cmd_ctx->lsm == NULL) {
76d7553f 3614 PERROR("zmalloc cmd_ctx->lsm");
5eb91c98
DG
3615 goto error;
3616 }
1d4b027a 3617
5eb91c98
DG
3618 cmd_ctx->llm = NULL;
3619 cmd_ctx->session = NULL;
1d4b027a 3620
840cb59c 3621 health_code_update();
44a5e5eb 3622
5eb91c98
DG
3623 /*
3624 * Data is received from the lttng client. The struct
3625 * lttcomm_session_msg (lsm) contains the command and data request of
3626 * the client.
3627 */
3628 DBG("Receiving data from client ...");
be040666
DG
3629 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3630 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
5eb91c98
DG
3631 if (ret <= 0) {
3632 DBG("Nothing recv() from client... continuing");
76d7553f
MD
3633 ret = close(sock);
3634 if (ret) {
3635 PERROR("close");
3636 }
3637 sock = -1;
a2c0da86 3638 clean_command_ctx(&cmd_ctx);
5eb91c98
DG
3639 continue;
3640 }
1d4b027a 3641
840cb59c 3642 health_code_update();
44a5e5eb 3643
5eb91c98
DG
3644 // TODO: Validate cmd_ctx including sanity check for
3645 // security purpose.
f7776ea7 3646
f6a9efaa 3647 rcu_thread_online();
5eb91c98
DG
3648 /*
3649 * This function dispatch the work to the kernel or userspace tracer
3650 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3651 * informations for the client. The command context struct contains
3652 * everything this function may needs.
3653 */
53a80697 3654 ret = process_client_msg(cmd_ctx, sock, &sock_error);
f6a9efaa 3655 rcu_thread_offline();
5eb91c98 3656 if (ret < 0) {
36134aa1
DG
3657 ret = close(sock);
3658 if (ret) {
3659 PERROR("close");
53a80697 3660 }
36134aa1 3661 sock = -1;
bbd973c2 3662 /*
5eb91c98 3663 * TODO: Inform client somehow of the fatal error. At
ba7f0ae5 3664 * this point, ret < 0 means that a zmalloc failed
53a80697
MD
3665 * (ENOMEM). Error detected but still accept
3666 * command, unless a socket error has been
3667 * detected.
bbd973c2 3668 */
bbd973c2 3669 clean_command_ctx(&cmd_ctx);
5eb91c98
DG
3670 continue;
3671 }
d6e4fca4 3672
840cb59c 3673 health_code_update();
44a5e5eb 3674
54d01ffb
DG
3675 DBG("Sending response (size: %d, retcode: %s)",
3676 cmd_ctx->lttng_msg_size,
9a745bc7 3677 lttng_strerror(-cmd_ctx->llm->ret_code));
54d01ffb 3678 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
5eb91c98
DG
3679 if (ret < 0) {
3680 ERR("Failed to send data back to client");
bbd973c2 3681 }
1d4b027a 3682
5eb91c98 3683 /* End of transmission */
76d7553f
MD
3684 ret = close(sock);
3685 if (ret) {
3686 PERROR("close");
3687 }
3688 sock = -1;
be040666
DG
3689
3690 clean_command_ctx(&cmd_ctx);
44a5e5eb 3691
840cb59c 3692 health_code_update();
273ea72c
DG
3693 }
3694
139ac872 3695exit:
5eb91c98 3696error:
35df2755
CB
3697 if (sock >= 0) {
3698 ret = close(sock);
3699 if (ret) {
3700 PERROR("close");
3701 }
139ac872 3702 }
44a5e5eb 3703
35df2755
CB
3704 lttng_poll_clean(&events);
3705 clean_command_ctx(&cmd_ctx);
3706
3707error_listen:
3708error_create_poll:
6993eeb3 3709error_testpoint:
273ea72c 3710 unlink(client_unix_sock_path);
76d7553f
MD
3711 if (client_sock >= 0) {
3712 ret = close(client_sock);
3713 if (ret) {
3714 PERROR("close");
3715 }
a4b35e07 3716 }
35df2755
CB
3717
3718 if (err) {
840cb59c 3719 health_error();
35df2755 3720 ERR("Health error occurred in %s", __func__);
a4b35e07 3721 }
273ea72c 3722
927ca06a 3723 health_unregister();
35df2755
CB
3724
3725 DBG("Client thread dying");
f6a9efaa
DG
3726
3727 rcu_unregister_thread();
1d4b027a
DG
3728 return NULL;
3729}
3730
3731
fac6795d
DG
3732/*
3733 * usage function on stderr
3734 */
3735static void usage(void)
3736{
b716ce68 3737 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
d6f42150
DG
3738 fprintf(stderr, " -h, --help Display this usage.\n");
3739 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3740 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3741 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3742 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
7753dea8
MD
3743 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3744 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3745 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3746 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
ebaeda94
MD
3747 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3748 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3749 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3750 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
d6f42150
DG
3751 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3752 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3753 fprintf(stderr, " -V, --version Show version number.\n");
3754 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3755 fprintf(stderr, " -q, --quiet No output at all.\n");
3756 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
35f90c40 3757 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3bd1e081 3758 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4fba7219 3759 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
fac6795d
DG
3760}
3761
3762/*
3763 * daemon argument parsing
3764 */
3765static int parse_args(int argc, char **argv)
3766{
3767 int c;
3768
3769 static struct option long_options[] = {
3770 { "client-sock", 1, 0, 'c' },
3771 { "apps-sock", 1, 0, 'a' },
3bd1e081
MD
3772 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3773 { "kconsumerd-err-sock", 1, 0, 'E' },
7753dea8
MD
3774 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3775 { "ustconsumerd32-err-sock", 1, 0, 'H' },
ebaeda94
MD
3776 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3777 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3778 { "consumerd32-path", 1, 0, 'u' },
3779 { "consumerd32-libdir", 1, 0, 'U' },
3780 { "consumerd64-path", 1, 0, 't' },
3781 { "consumerd64-libdir", 1, 0, 'T' },
fac6795d 3782 { "daemonize", 0, 0, 'd' },
5b8719f5 3783 { "sig-parent", 0, 0, 'S' },
fac6795d
DG
3784 { "help", 0, 0, 'h' },
3785 { "group", 1, 0, 'g' },
3786 { "version", 0, 0, 'V' },
75462a81 3787 { "quiet", 0, 0, 'q' },
3f9947db 3788 { "verbose", 0, 0, 'v' },
3bd1e081 3789 { "verbose-consumer", 0, 0, 'Z' },
4fba7219 3790 { "no-kernel", 0, 0, 'N' },
35f90c40 3791 { "pidfile", 1, 0, 'p' },
fac6795d
DG
3792 { NULL, 0, 0, 0 }
3793 };
3794
3795 while (1) {
3796 int option_index = 0;
35f90c40 3797 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
54d01ffb 3798 long_options, &option_index);
fac6795d
DG
3799 if (c == -1) {
3800 break;
3801 }
3802
3803 switch (c) {
3804 case 0:
3805 fprintf(stderr, "option %s", long_options[option_index].name);
3806 if (optarg) {
3807 fprintf(stderr, " with arg %s\n", optarg);
3808 }
3809 break;
b716ce68 3810 case 'c':
fac6795d
DG
3811 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3812 break;
3813 case 'a':
3814 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3815 break;
3816 case 'd':
3817 opt_daemon = 1;
3818 break;
3819 case 'g':
fb09408a 3820 opt_tracing_group = optarg;
fac6795d
DG
3821 break;
3822 case 'h':
3823 usage();
3824 exit(EXIT_FAILURE);
3825 case 'V':
3826 fprintf(stdout, "%s\n", VERSION);
3827 exit(EXIT_SUCCESS);
5b8719f5
DG
3828 case 'S':
3829 opt_sig_parent = 1;
3830 break;
d6f42150 3831 case 'E':
3bd1e081 3832 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
d6f42150
DG
3833 break;
3834 case 'C':
3bd1e081
MD
3835 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3836 break;
3837 case 'F':
7753dea8 3838 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3bd1e081
MD
3839 break;
3840 case 'D':
7753dea8
MD
3841 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3842 break;
3843 case 'H':
3844 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3845 break;
3846 case 'G':
3847 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
d6f42150 3848 break;
4fba7219
DG
3849 case 'N':
3850 opt_no_kernel = 1;
3851 break;
75462a81 3852 case 'q':
97e19046 3853 lttng_opt_quiet = 1;
75462a81 3854 break;
3f9947db 3855 case 'v':
53086306 3856 /* Verbose level can increase using multiple -v */
97e19046 3857 lttng_opt_verbose += 1;
3f9947db 3858 break;
31f73cc9 3859 case 'Z':
3bd1e081 3860 opt_verbose_consumer += 1;
31f73cc9 3861 break;
fb09408a 3862 case 'u':
fc7a59ce 3863 consumerd32_bin= optarg;
ebaeda94
MD
3864 break;
3865 case 'U':
3866 consumerd32_libdir = optarg;
7753dea8
MD
3867 break;
3868 case 't':
fc7a59ce 3869 consumerd64_bin = optarg;
ebaeda94
MD
3870 break;
3871 case 'T':
3872 consumerd64_libdir = optarg;
fb09408a 3873 break;
35f90c40
DG
3874 case 'p':
3875 opt_pidfile = optarg;
3876 break;
fac6795d
DG
3877 default:
3878 /* Unknown option or other error.
3879 * Error is printed by getopt, just return */
3880 return -1;
3881 }
3882 }
3883
3884 return 0;
3885}
3886
3887/*
d063d709 3888 * Creates the two needed socket by the daemon.
d6f42150
DG
3889 * apps_sock - The communication socket for all UST apps.
3890 * client_sock - The communication of the cli tool (lttng).
fac6795d 3891 */
cf3af59e 3892static int init_daemon_socket(void)
fac6795d
DG
3893{
3894 int ret = 0;
3895 mode_t old_umask;
3896
3897 old_umask = umask(0);
3898
3899 /* Create client tool unix socket */
d6f42150
DG
3900 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3901 if (client_sock < 0) {
3902 ERR("Create unix sock failed: %s", client_unix_sock_path);
fac6795d
DG
3903 ret = -1;
3904 goto end;
3905 }
3906
b662582b
DG
3907 /* Set the cloexec flag */
3908 ret = utils_set_fd_cloexec(client_sock);
3909 if (ret < 0) {
3910 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3911 "Continuing but note that the consumer daemon will have a "
3912 "reference to this socket on exec()", client_sock);
3913 }
3914
fac6795d
DG
3915 /* File permission MUST be 660 */
3916 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3917 if (ret < 0) {
d6f42150 3918 ERR("Set file permissions failed: %s", client_unix_sock_path);
76d7553f 3919 PERROR("chmod");
fac6795d
DG
3920 goto end;
3921 }
3922
3923 /* Create the application unix socket */
d6f42150
DG
3924 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3925 if (apps_sock < 0) {
3926 ERR("Create unix sock failed: %s", apps_unix_sock_path);
fac6795d
DG
3927 ret = -1;
3928 goto end;
3929 }
3930
b662582b
DG
3931 /* Set the cloexec flag */
3932 ret = utils_set_fd_cloexec(apps_sock);
3933 if (ret < 0) {
3934 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3935 "Continuing but note that the consumer daemon will have a "
3936 "reference to this socket on exec()", apps_sock);
3937 }
3938
d6f42150 3939 /* File permission MUST be 666 */
54d01ffb
DG
3940 ret = chmod(apps_unix_sock_path,
3941 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
fac6795d 3942 if (ret < 0) {
d6f42150 3943 ERR("Set file permissions failed: %s", apps_unix_sock_path);
76d7553f 3944 PERROR("chmod");
fac6795d
DG
3945 goto end;
3946 }
3947
b662582b
DG
3948 DBG3("Session daemon client socket %d and application socket %d created",
3949 client_sock, apps_sock);
3950
fac6795d
DG
3951end:
3952 umask(old_umask);
3953 return ret;
3954}
3955
3956/*
54d01ffb
DG
3957 * Check if the global socket is available, and if a daemon is answering at the
3958 * other side. If yes, error is returned.
fac6795d 3959 */
cf3af59e 3960static int check_existing_daemon(void)
fac6795d 3961{
7d8234d9 3962 /* Is there anybody out there ? */
099e26bd 3963 if (lttng_session_daemon_alive()) {
7d8234d9 3964 return -EEXIST;
099e26bd 3965 }
b09c7c76
DG
3966
3967 return 0;
fac6795d
DG
3968}
3969
fac6795d 3970/*
d063d709 3971 * Set the tracing group gid onto the client socket.
5e16da05 3972 *
d063d709 3973 * Race window between mkdir and chown is OK because we are going from more
d1613cf5 3974 * permissive (root.root) to less permissive (root.tracing).
fac6795d 3975 */
be040666 3976static int set_permissions(char *rundir)
fac6795d
DG
3977{
3978 int ret;
996b65c8 3979 gid_t gid;
fac6795d 3980
6775595e
DG
3981 ret = allowed_group();
3982 if (ret < 0) {
be040666 3983 WARN("No tracing group detected");
abec5852
MD
3984 /* Setting gid to 0 if no tracing group is found */
3985 gid = 0;
3986 } else {
3987 gid = ret;
fac6795d
DG
3988 }
3989
d6f42150 3990 /* Set lttng run dir */
be040666 3991 ret = chown(rundir, 0, gid);
d6f42150 3992 if (ret < 0) {
be040666 3993 ERR("Unable to set group on %s", rundir);
76d7553f 3994 PERROR("chown");
d6f42150
DG
3995 }
3996
abec5852 3997 /* Ensure all applications and tracing group can search the run dir */
61076f74 3998 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
d1613cf5
JN
3999 if (ret < 0) {
4000 ERR("Unable to set permissions on %s", rundir);
76d7553f 4001 PERROR("chmod");
d1613cf5
JN
4002 }
4003
d6f42150 4004 /* lttng client socket path */
996b65c8 4005 ret = chown(client_unix_sock_path, 0, gid);
fac6795d 4006 if (ret < 0) {
d6f42150 4007 ERR("Unable to set group on %s", client_unix_sock_path);
76d7553f 4008 PERROR("chown");
d6f42150
DG
4009 }
4010
3bd1e081
MD
4011 /* kconsumer error socket path */
4012 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
d6f42150 4013 if (ret < 0) {
3bd1e081 4014 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
76d7553f 4015 PERROR("chown");
3bd1e081
MD
4016 }
4017
7753dea8
MD
4018 /* 64-bit ustconsumer error socket path */
4019 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
4020 if (ret < 0) {
4021 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
76d7553f 4022 PERROR("chown");
7753dea8
MD
4023 }
4024
4025 /* 32-bit ustconsumer compat32 error socket path */
4026 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3bd1e081 4027 if (ret < 0) {
7753dea8 4028 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
76d7553f 4029 PERROR("chown");
fac6795d
DG
4030 }
4031
d6f42150 4032 DBG("All permissions are set");
e07ae692 4033
fac6795d
DG
4034 return ret;
4035}
4036
d6f42150 4037/*
d063d709 4038 * Create the lttng run directory needed for all global sockets and pipe.
d6f42150 4039 */
67e40797 4040static int create_lttng_rundir(const char *rundir)
d6f42150
DG
4041{
4042 int ret;
4043
67e40797
DG
4044 DBG3("Creating LTTng run directory: %s", rundir);
4045
d1613cf5 4046 ret = mkdir(rundir, S_IRWXU);
d6f42150 4047 if (ret < 0) {
b1f11e69 4048 if (errno != EEXIST) {
67e40797 4049 ERR("Unable to create %s", rundir);
b1f11e69
DG
4050 goto error;
4051 } else {
4052 ret = 0;
4053 }
d6f42150
DG
4054 }
4055
4056error:
4057 return ret;
4058}
4059
4060/*
d063d709
DG
4061 * Setup sockets and directory needed by the kconsumerd communication with the
4062 * session daemon.
d6f42150 4063 */
67e40797
DG
4064static int set_consumer_sockets(struct consumer_data *consumer_data,
4065 const char *rundir)
d6f42150
DG
4066{
4067 int ret;
67e40797 4068 char path[PATH_MAX];
d6f42150 4069
67e40797 4070 switch (consumer_data->type) {
7753dea8 4071 case LTTNG_CONSUMER_KERNEL:
60922cb0 4072 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
7753dea8
MD
4073 break;
4074 case LTTNG_CONSUMER64_UST:
60922cb0 4075 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
7753dea8
MD
4076 break;
4077 case LTTNG_CONSUMER32_UST:
60922cb0 4078 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
7753dea8
MD
4079 break;
4080 default:
4081 ERR("Consumer type unknown");
4082 ret = -EINVAL;
4083 goto error;
d6f42150
DG
4084 }
4085
67e40797
DG
4086 DBG2("Creating consumer directory: %s", path);
4087
d1613cf5 4088 ret = mkdir(path, S_IRWXU);
d6f42150 4089 if (ret < 0) {
6beb2242 4090 if (errno != EEXIST) {
f11e84c2 4091 PERROR("mkdir");
3bd1e081 4092 ERR("Failed to create %s", path);
6beb2242
DG
4093 goto error;
4094 }
f11e84c2 4095 ret = -1;
d6f42150
DG
4096 }
4097
4098 /* Create the kconsumerd error unix socket */
3bd1e081
MD
4099 consumer_data->err_sock =
4100 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4101 if (consumer_data->err_sock < 0) {
4102 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
d6f42150
DG
4103 ret = -1;
4104 goto error;
4105 }
4106
12c30c08
MD
4107 /*
4108 * Set the CLOEXEC flag. Return code is useless because either way, the
4109 * show must go on.
4110 */
4111 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4112 if (ret < 0) {
4113 PERROR("utils_set_fd_cloexec");
4114 /* continue anyway */
4115 }
4116
d6f42150 4117 /* File permission MUST be 660 */
3bd1e081 4118 ret = chmod(consumer_data->err_unix_sock_path,
54d01ffb 4119 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
d6f42150 4120 if (ret < 0) {
3bd1e081 4121 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
67e40797 4122 PERROR("chmod");
d6f42150
DG
4123 goto error;
4124 }
4125
4126error:
4127 return ret;
4128}
4129
fac6795d 4130/*
d063d709 4131 * Signal handler for the daemon
cf3af59e 4132 *
54d01ffb
DG
4133 * Simply stop all worker threads, leaving main() return gracefully after
4134 * joining all threads and calling cleanup().
fac6795d
DG
4135 */
4136static void sighandler(int sig)
4137{
4138 switch (sig) {
cf3af59e 4139 case SIGPIPE:
af87c45a 4140 DBG("SIGPIPE caught");
cf3af59e
MD
4141 return;
4142 case SIGINT:
af87c45a 4143 DBG("SIGINT caught");
cf3af59e
MD
4144 stop_threads();
4145 break;
4146 case SIGTERM:
af87c45a 4147 DBG("SIGTERM caught");
cf3af59e
MD
4148 stop_threads();
4149 break;
4150 default:
4151 break;
fac6795d 4152 }
fac6795d
DG
4153}
4154
4155/*
d063d709 4156 * Setup signal handler for :
1d4b027a 4157 * SIGINT, SIGTERM, SIGPIPE
fac6795d 4158 */
1d4b027a 4159static int set_signal_handler(void)
fac6795d 4160{
1d4b027a
DG
4161 int ret = 0;
4162 struct sigaction sa;
4163 sigset_t sigset;
fac6795d 4164
1d4b027a 4165 if ((ret = sigemptyset(&sigset)) < 0) {
76d7553f 4166 PERROR("sigemptyset");
1d4b027a
DG
4167 return ret;
4168 }
d6f42150 4169
1d4b027a
DG
4170 sa.sa_handler = sighandler;
4171 sa.sa_mask = sigset;
4172 sa.sa_flags = 0;
4173 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
76d7553f 4174 PERROR("sigaction");
1d4b027a 4175 return ret;
d6f42150
DG
4176 }
4177
1d4b027a 4178 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
76d7553f 4179 PERROR("sigaction");
1d4b027a 4180 return ret;
d6f42150 4181 }
aaf26714 4182
1d4b027a 4183 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
76d7553f 4184 PERROR("sigaction");
1d4b027a 4185 return ret;
8c0faa1d
DG
4186 }
4187
1d4b027a
DG
4188 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4189
4190 return ret;
fac6795d
DG
4191}
4192
f3ed775e 4193/*
d063d709
DG
4194 * Set open files limit to unlimited. This daemon can open a large number of
4195 * file descriptors in order to consumer multiple kernel traces.
f3ed775e
DG
4196 */
4197static void set_ulimit(void)
4198{
4199 int ret;
4200 struct rlimit lim;
4201
a88df331 4202 /* The kernel does not allowed an infinite limit for open files */
f3ed775e
DG
4203 lim.rlim_cur = 65535;
4204 lim.rlim_max = 65535;
4205
4206 ret = setrlimit(RLIMIT_NOFILE, &lim);
4207 if (ret < 0) {
76d7553f 4208 PERROR("failed to set open files limit");
f3ed775e
DG
4209 }
4210}
4211
35f90c40
DG
4212/*
4213 * Write pidfile using the rundir and opt_pidfile.
4214 */
4215static void write_pidfile(void)
4216{
4217 int ret;
4218 char pidfile_path[PATH_MAX];
4219
4220 assert(rundir);
4221
4222 if (opt_pidfile) {
4223 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4224 } else {
4225 /* Build pidfile path from rundir and opt_pidfile. */
4226 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4227 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4228 if (ret < 0) {
4229 PERROR("snprintf pidfile path");
4230 goto error;
4231 }
4232 }
4233
4234 /*
4235 * Create pid file in rundir. Return value is of no importance. The
4236 * execution will continue even though we are not able to write the file.
4237 */
4238 (void) utils_create_pid_file(getpid(), pidfile_path);
4239
4240error:
4241 return;
4242}
4243
fac6795d
DG
4244/*
4245 * main
4246 */
4247int main(int argc, char **argv)
4248{
fac6795d
DG
4249 int ret = 0;
4250 void *status;
ae9e45b3 4251 const char *home_path, *env_app_timeout;
fac6795d 4252
335a95b7
MD
4253 init_kernel_workarounds();
4254
f6a9efaa
DG
4255 rcu_register_thread();
4256
7753dea8 4257 setup_consumerd_path();
fb09408a 4258
12744796
DG
4259 page_size = sysconf(_SC_PAGESIZE);
4260 if (page_size < 0) {
4261 PERROR("sysconf _SC_PAGESIZE");
4262 page_size = LONG_MAX;
4263 WARN("Fallback page size to %ld", page_size);
4264 }
4265
fac6795d
DG
4266 /* Parse arguments */
4267 progname = argv[0];
c617c0c6 4268 if ((ret = parse_args(argc, argv)) < 0) {
cf3af59e 4269 goto error;
fac6795d
DG
4270 }
4271
4272 /* Daemonize */
4273 if (opt_daemon) {
ceed52b5
MD
4274 int i;
4275
4276 /*
4277 * fork
4278 * child: setsid, close FD 0, 1, 2, chdir /
4279 * parent: exit (if fork is successful)
4280 */
53094c05
DG
4281 ret = daemon(0, 0);
4282 if (ret < 0) {
76d7553f 4283 PERROR("daemon");
cf3af59e 4284 goto error;
53094c05 4285 }
ceed52b5
MD
4286 /*
4287 * We are in the child. Make sure all other file
4288 * descriptors are closed, in case we are called with
4289 * more opened file descriptors than the standard ones.
4290 */
4291 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4292 (void) close(i);
4293 }
4294 }
4295
4296 /* Create thread quit pipe */
4297 if ((ret = init_thread_quit_pipe()) < 0) {
4298 goto error;
fac6795d
DG
4299 }
4300
4301 /* Check if daemon is UID = 0 */
4302 is_root = !getuid();
4303
fac6795d 4304 if (is_root) {
990570ed 4305 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
67e40797
DG
4306
4307 /* Create global run dir with root access */
4308 ret = create_lttng_rundir(rundir);
d6f42150 4309 if (ret < 0) {
cf3af59e 4310 goto error;
d6f42150
DG
4311 }
4312
fac6795d 4313 if (strlen(apps_unix_sock_path) == 0) {
d6f42150
DG
4314 snprintf(apps_unix_sock_path, PATH_MAX,
4315 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
fac6795d
DG
4316 }
4317
4318 if (strlen(client_unix_sock_path) == 0) {
d6f42150
DG
4319 snprintf(client_unix_sock_path, PATH_MAX,
4320 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4321 }
0fdd1e2c
DG
4322
4323 /* Set global SHM for ust */
4324 if (strlen(wait_shm_path) == 0) {
4325 snprintf(wait_shm_path, PATH_MAX,
4326 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4327 }
67e40797 4328
44a5e5eb
DG
4329 if (strlen(health_unix_sock_path) == 0) {
4330 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4331 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4332 }
4333
67e40797
DG
4334 /* Setup kernel consumerd path */
4335 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
60922cb0 4336 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
67e40797 4337 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4338 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
67e40797
DG
4339
4340 DBG2("Kernel consumer err path: %s",
4341 kconsumer_data.err_unix_sock_path);
4342 DBG2("Kernel consumer cmd path: %s",
4343 kconsumer_data.cmd_unix_sock_path);
fac6795d 4344 } else {
feb0f3e5 4345 home_path = utils_get_home_dir();
b082db07 4346 if (home_path == NULL) {
273ea72c
DG
4347 /* TODO: Add --socket PATH option */
4348 ERR("Can't get HOME directory for sockets creation.");
cf3af59e
MD
4349 ret = -EPERM;
4350 goto error;
b082db07
DG
4351 }
4352
67e40797
DG
4353 /*
4354 * Create rundir from home path. This will create something like
4355 * $HOME/.lttng
4356 */
990570ed 4357 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
67e40797
DG
4358 if (ret < 0) {
4359 ret = -ENOMEM;
4360 goto error;
4361 }
4362
4363 ret = create_lttng_rundir(rundir);
4364 if (ret < 0) {
4365 goto error;
4366 }
4367
fac6795d 4368 if (strlen(apps_unix_sock_path) == 0) {
d6f42150 4369 snprintf(apps_unix_sock_path, PATH_MAX,
b082db07 4370 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
fac6795d
DG
4371 }
4372
4373 /* Set the cli tool unix socket path */
4374 if (strlen(client_unix_sock_path) == 0) {
d6f42150 4375 snprintf(client_unix_sock_path, PATH_MAX,
b082db07 4376 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
fac6795d 4377 }
0fdd1e2c
DG
4378
4379 /* Set global SHM for ust */
4380 if (strlen(wait_shm_path) == 0) {
4381 snprintf(wait_shm_path, PATH_MAX,
d0b96690 4382 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
0fdd1e2c 4383 }
44a5e5eb
DG
4384
4385 /* Set health check Unix path */
4386 if (strlen(health_unix_sock_path) == 0) {
4387 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4388 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4389 }
fac6795d
DG
4390 }
4391
5c827ce0
DG
4392 /* Set consumer initial state */
4393 kernel_consumerd_state = CONSUMER_STOPPED;
4394 ust_consumerd_state = CONSUMER_STOPPED;
4395
847177cd
DG
4396 DBG("Client socket path %s", client_unix_sock_path);
4397 DBG("Application socket path %s", apps_unix_sock_path);
d0b96690 4398 DBG("Application wait path %s", wait_shm_path);
67e40797
DG
4399 DBG("LTTng run directory path: %s", rundir);
4400
4401 /* 32 bits consumerd path setup */
4402 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
60922cb0 4403 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
67e40797 4404 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4405 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
67e40797
DG
4406
4407 DBG2("UST consumer 32 bits err path: %s",
4408 ustconsumer32_data.err_unix_sock_path);
4409 DBG2("UST consumer 32 bits cmd path: %s",
4410 ustconsumer32_data.cmd_unix_sock_path);
4411
4412 /* 64 bits consumerd path setup */
4413 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
60922cb0 4414 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
67e40797 4415 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4416 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
67e40797
DG
4417
4418 DBG2("UST consumer 64 bits err path: %s",
4419 ustconsumer64_data.err_unix_sock_path);
4420 DBG2("UST consumer 64 bits cmd path: %s",
4421 ustconsumer64_data.cmd_unix_sock_path);
847177cd 4422
273ea72c 4423 /*
7d8234d9 4424 * See if daemon already exist.
fac6795d 4425 */
7d8234d9 4426 if ((ret = check_existing_daemon()) < 0) {
75462a81 4427 ERR("Already running daemon.\n");
273ea72c 4428 /*
cf3af59e
MD
4429 * We do not goto exit because we must not cleanup()
4430 * because a daemon is already running.
ab118b20 4431 */
cf3af59e 4432 goto error;
a88df331
DG
4433 }
4434
1427f9b2
DG
4435 /*
4436 * Init UST app hash table. Alloc hash table before this point since
4437 * cleanup() can get called after that point.
4438 */
4439 ust_app_ht_alloc();
4440
54d01ffb 4441 /* After this point, we can safely call cleanup() with "goto exit" */
a88df331
DG
4442
4443 /*
4444 * These actions must be executed as root. We do that *after* setting up
4445 * the sockets path because we MUST make the check for another daemon using
4446 * those paths *before* trying to set the kernel consumer sockets and init
4447 * kernel tracer.
4448 */
4449 if (is_root) {
67e40797 4450 ret = set_consumer_sockets(&kconsumer_data, rundir);
7753dea8
MD
4451 if (ret < 0) {
4452 goto exit;
4453 }
4454
a88df331 4455 /* Setup kernel tracer */
4fba7219
DG
4456 if (!opt_no_kernel) {
4457 init_kernel_tracer();
4458 }
a88df331
DG
4459
4460 /* Set ulimit for open files */
4461 set_ulimit();
fac6795d 4462 }
4063050c
MD
4463 /* init lttng_fd tracking must be done after set_ulimit. */
4464 lttng_fd_init();
fac6795d 4465
67e40797
DG
4466 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4467 if (ret < 0) {
4468 goto exit;
4469 }
4470
4471 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4472 if (ret < 0) {
4473 goto exit;
4474 }
4475
cf3af59e
MD
4476 if ((ret = set_signal_handler()) < 0) {
4477 goto exit;
fac6795d
DG
4478 }
4479
d6f42150 4480 /* Setup the needed unix socket */
cf3af59e
MD
4481 if ((ret = init_daemon_socket()) < 0) {
4482 goto exit;
fac6795d
DG
4483 }
4484
4485 /* Set credentials to socket */
be040666 4486 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
cf3af59e 4487 goto exit;
fac6795d
DG
4488 }
4489
5b8719f5
DG
4490 /* Get parent pid if -S, --sig-parent is specified. */
4491 if (opt_sig_parent) {
4492 ppid = getppid();
4493 }
4494
7a485870 4495 /* Setup the kernel pipe for waking up the kernel thread */
6620da75
DG
4496 if (is_root && !opt_no_kernel) {
4497 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4498 goto exit;
4499 }
7a485870
DG
4500 }
4501
0b2dc8df
MD
4502 /* Setup the thread ht_cleanup communication pipe. */
4503 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4504 goto exit;
4505 }
4506
099e26bd 4507 /* Setup the thread apps communication pipe. */
ef599319 4508 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
099e26bd
DG
4509 goto exit;
4510 }
4511
d0b96690
DG
4512 /* Setup the thread apps notify communication pipe. */
4513 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4514 goto exit;
4515 }
4516
7972aab2
DG
4517 /* Initialize global buffer per UID and PID registry. */
4518 buffer_reg_init_uid_registry();
4519 buffer_reg_init_pid_registry();
4520
099e26bd
DG
4521 /* Init UST command queue. */
4522 cds_wfq_init(&ust_cmd_queue.queue);
4523
273ea72c 4524 /*
54d01ffb
DG
4525 * Get session list pointer. This pointer MUST NOT be free(). This list is
4526 * statically declared in session.c
273ea72c 4527 */
54d01ffb 4528 session_list_ptr = session_get_list();
b5541356 4529
5eb91c98
DG
4530 /* Set up max poll set size */
4531 lttng_poll_set_max_size();
4532
2f77fc4b 4533 cmd_init();
00e2e675 4534
ae9e45b3
DG
4535 /* Check for the application socket timeout env variable. */
4536 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4537 if (env_app_timeout) {
4538 app_socket_timeout = atoi(env_app_timeout);
4539 } else {
4540 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4541 }
4542
35f90c40
DG
4543 write_pidfile();
4544
cb5f6f14
DG
4545 /* This is to get the TCP timeout value. */
4546 lttcomm_inet_init();
4547
845e9e6b
DG
4548 /*
4549 * Initialize the health check subsystem. This call should set the
4550 * appropriate time values.
4551 */
4552 health_init();
4553
0b2dc8df
MD
4554 /* Create thread to manage the client socket */
4555 ret = pthread_create(&ht_cleanup_thread, NULL,
4556 thread_ht_cleanup, (void *) NULL);
4557 if (ret != 0) {
4558 PERROR("pthread_create ht_cleanup");
4559 goto exit_ht_cleanup;
4560 }
4561
44a5e5eb
DG
4562 /* Create thread to manage the client socket */
4563 ret = pthread_create(&health_thread, NULL,
4564 thread_manage_health, (void *) NULL);
4565 if (ret != 0) {
4566 PERROR("pthread_create health");
4567 goto exit_health;
4568 }
4569
cf3af59e 4570 /* Create thread to manage the client socket */
099e26bd
DG
4571 ret = pthread_create(&client_thread, NULL,
4572 thread_manage_clients, (void *) NULL);
cf3af59e 4573 if (ret != 0) {
76d7553f 4574 PERROR("pthread_create clients");
cf3af59e
MD
4575 goto exit_client;
4576 }
fac6795d 4577
099e26bd
DG
4578 /* Create thread to dispatch registration */
4579 ret = pthread_create(&dispatch_thread, NULL,
4580 thread_dispatch_ust_registration, (void *) NULL);
4581 if (ret != 0) {
76d7553f 4582 PERROR("pthread_create dispatch");
099e26bd
DG
4583 goto exit_dispatch;
4584 }
4585
4586 /* Create thread to manage application registration. */
4587 ret = pthread_create(&reg_apps_thread, NULL,
4588 thread_registration_apps, (void *) NULL);
4589 if (ret != 0) {
76d7553f 4590 PERROR("pthread_create registration");
099e26bd
DG
4591 goto exit_reg_apps;
4592 }
4593
cf3af59e 4594 /* Create thread to manage application socket */
54d01ffb
DG
4595 ret = pthread_create(&apps_thread, NULL,
4596 thread_manage_apps, (void *) NULL);
cf3af59e 4597 if (ret != 0) {
d0b96690
DG
4598 PERROR("pthread_create apps");
4599 goto exit_apps;
4600 }
4601
4602 /* Create thread to manage application notify socket */
4603 ret = pthread_create(&apps_notify_thread, NULL,
4604 ust_thread_manage_notify, (void *) NULL);
4605 if (ret != 0) {
76d7553f 4606 PERROR("pthread_create apps");
cf3af59e
MD
4607 goto exit_apps;
4608 }
fac6795d 4609
6620da75
DG
4610 /* Don't start this thread if kernel tracing is not requested nor root */
4611 if (is_root && !opt_no_kernel) {
4612 /* Create kernel thread to manage kernel event */
4613 ret = pthread_create(&kernel_thread, NULL,
4614 thread_manage_kernel, (void *) NULL);
4615 if (ret != 0) {
4616 PERROR("pthread_create kernel");
4617 goto exit_kernel;
4618 }
7a485870 4619
6620da75
DG
4620 ret = pthread_join(kernel_thread, &status);
4621 if (ret != 0) {
4622 PERROR("pthread_join");
4623 goto error; /* join error, exit without cleanup */
4624 }
fac6795d
DG
4625 }
4626
cf3af59e
MD
4627exit_kernel:
4628 ret = pthread_join(apps_thread, &status);
4629 if (ret != 0) {
76d7553f 4630 PERROR("pthread_join");
cf3af59e
MD
4631 goto error; /* join error, exit without cleanup */
4632 }
fac6795d 4633
cf3af59e 4634exit_apps:
099e26bd
DG
4635 ret = pthread_join(reg_apps_thread, &status);
4636 if (ret != 0) {
76d7553f 4637 PERROR("pthread_join");
099e26bd
DG
4638 goto error; /* join error, exit without cleanup */
4639 }
4640
4641exit_reg_apps:
4642 ret = pthread_join(dispatch_thread, &status);
4643 if (ret != 0) {
76d7553f 4644 PERROR("pthread_join");
099e26bd
DG
4645 goto error; /* join error, exit without cleanup */
4646 }
4647
4648exit_dispatch:
cf3af59e
MD
4649 ret = pthread_join(client_thread, &status);
4650 if (ret != 0) {
76d7553f 4651 PERROR("pthread_join");
cf3af59e
MD
4652 goto error; /* join error, exit without cleanup */
4653 }
4654
3bd1e081 4655 ret = join_consumer_thread(&kconsumer_data);
cf3af59e 4656 if (ret != 0) {
76d7553f 4657 PERROR("join_consumer");
cf3af59e
MD
4658 goto error; /* join error, exit without cleanup */
4659 }
a88df331 4660
06f525de
DG
4661 ret = join_consumer_thread(&ustconsumer32_data);
4662 if (ret != 0) {
4663 PERROR("join_consumer ust32");
4664 goto error; /* join error, exit without cleanup */
4665 }
4666
4667 ret = join_consumer_thread(&ustconsumer64_data);
4668 if (ret != 0) {
4669 PERROR("join_consumer ust64");
4670 goto error; /* join error, exit without cleanup */
4671 }
4672
cf3af59e 4673exit_client:
06f525de
DG
4674 ret = pthread_join(health_thread, &status);
4675 if (ret != 0) {
4676 PERROR("pthread_join health thread");
4677 goto error; /* join error, exit without cleanup */
4678 }
4679
44a5e5eb 4680exit_health:
0b2dc8df
MD
4681 ret = pthread_join(ht_cleanup_thread, &status);
4682 if (ret != 0) {
4683 PERROR("pthread_join ht cleanup thread");
4684 goto error; /* join error, exit without cleanup */
4685 }
4686exit_ht_cleanup:
a88df331 4687exit:
cf3af59e
MD
4688 /*
4689 * cleanup() is called when no other thread is running.
4690 */
f6a9efaa 4691 rcu_thread_online();
cf3af59e 4692 cleanup();
f6a9efaa
DG
4693 rcu_thread_offline();
4694 rcu_unregister_thread();
67e40797 4695 if (!ret) {
cf3af59e 4696 exit(EXIT_SUCCESS);
67e40797 4697 }
cf3af59e 4698error:
5e16da05 4699 exit(EXIT_FAILURE);
fac6795d 4700}
This page took 0.386148 seconds and 4 git commands to generate.