Fix: possible consumer sockets double close on cleanup
[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
DG
1295 /* Set socket timeout for both receiving and ending */
1296 (void) lttcomm_setsockopt_rcv_timeout(sock,
1297 app_socket_timeout);
1298 (void) lttcomm_setsockopt_snd_timeout(sock,
1299 app_socket_timeout);
ae9e45b3 1300
d0b96690 1301 DBG("Apps with sock %d added to poll set", sock);
487cf67c 1302
840cb59c 1303 health_code_update();
44a5e5eb 1304
5eb91c98 1305 break;
0177d773 1306 }
5eb91c98
DG
1307 } else {
1308 /*
54d01ffb
DG
1309 * At this point, we know that a registered application made
1310 * the event at poll_wait.
5eb91c98
DG
1311 */
1312 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1313 /* Removing from the poll set */
1314 ret = lttng_poll_del(&events, pollfd);
1315 if (ret < 0) {
1316 goto error;
1317 }
099e26bd 1318
b9d9b220 1319 /* Socket closed on remote end. */
56fff090 1320 ust_app_unregister(pollfd);
5eb91c98
DG
1321 break;
1322 }
099e26bd 1323 }
44a5e5eb 1324
840cb59c 1325 health_code_update();
099e26bd 1326 }
099e26bd
DG
1327 }
1328
139ac872 1329exit:
099e26bd 1330error:
5eb91c98 1331 lttng_poll_clean(&events);
76d7553f 1332error_poll_create:
6993eeb3 1333error_testpoint:
6620da75
DG
1334 utils_close_pipe(apps_cmd_pipe);
1335 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1336
1337 /*
1338 * We don't clean the UST app hash table here since already registered
1339 * applications can still be controlled so let them be until the session
1340 * daemon dies or the applications stop.
1341 */
1342
139ac872 1343 if (err) {
840cb59c 1344 health_error();
139ac872
MD
1345 ERR("Health error occurred in %s", __func__);
1346 }
927ca06a 1347 health_unregister();
76d7553f 1348 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1349 rcu_thread_offline();
1350 rcu_unregister_thread();
099e26bd
DG
1351 return NULL;
1352}
1353
d0b96690 1354/*
d88aee68
DG
1355 * Send a socket to a thread This is called from the dispatch UST registration
1356 * thread once all sockets are set for the application.
d0b96690
DG
1357 *
1358 * On success, return 0 else a negative value being the errno message of the
1359 * write().
1360 */
d88aee68 1361static int send_socket_to_thread(int fd, int sock)
d0b96690
DG
1362{
1363 int ret;
1364
d0b96690 1365 /* Sockets MUST be set or else this should not have been called. */
d88aee68
DG
1366 assert(fd >= 0);
1367 assert(sock >= 0);
d0b96690
DG
1368
1369 do {
d88aee68 1370 ret = write(fd, &sock, sizeof(sock));
d0b96690 1371 } while (ret < 0 && errno == EINTR);
d88aee68
DG
1372 if (ret < 0 || ret != sizeof(sock)) {
1373 PERROR("write apps pipe %d", fd);
d0b96690
DG
1374 if (ret < 0) {
1375 ret = -errno;
1376 }
1377 goto error;
1378 }
1379
1380 /* All good. Don't send back the write positive ret value. */
1381 ret = 0;
1382error:
1383 return ret;
1384}
1385
f45e313d
DG
1386/*
1387 * Sanitize the wait queue of the dispatch registration thread meaning removing
1388 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1389 * notify socket is never received.
1390 */
1391static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1392{
1393 int ret, nb_fd = 0, i;
1394 unsigned int fd_added = 0;
1395 struct lttng_poll_event events;
1396 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1397
1398 assert(wait_queue);
1399
1400 lttng_poll_init(&events);
1401
1402 /* Just skip everything for an empty queue. */
1403 if (!wait_queue->count) {
1404 goto end;
1405 }
1406
1407 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1408 if (ret < 0) {
1409 goto error_create;
1410 }
1411
1412 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1413 &wait_queue->head, head) {
1414 assert(wait_node->app);
1415 ret = lttng_poll_add(&events, wait_node->app->sock,
1416 LPOLLHUP | LPOLLERR);
1417 if (ret < 0) {
1418 goto error;
1419 }
1420
1421 fd_added = 1;
1422 }
1423
1424 if (!fd_added) {
1425 goto end;
1426 }
1427
1428 /*
1429 * Poll but don't block so we can quickly identify the faulty events and
1430 * clean them afterwards from the wait queue.
1431 */
1432 ret = lttng_poll_wait(&events, 0);
1433 if (ret < 0) {
1434 goto error;
1435 }
1436 nb_fd = ret;
1437
1438 for (i = 0; i < nb_fd; i++) {
1439 /* Get faulty FD. */
1440 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1441 int pollfd = LTTNG_POLL_GETFD(&events, i);
1442
1443 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1444 &wait_queue->head, head) {
1445 if (pollfd == wait_node->app->sock &&
1446 (revents & (LPOLLHUP | LPOLLERR))) {
1447 cds_list_del(&wait_node->head);
1448 wait_queue->count--;
1449 ust_app_destroy(wait_node->app);
1450 free(wait_node);
1451 break;
1452 }
1453 }
1454 }
1455
1456 if (nb_fd > 0) {
1457 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1458 }
1459
1460end:
1461 lttng_poll_clean(&events);
1462 return;
1463
1464error:
1465 lttng_poll_clean(&events);
1466error_create:
1467 ERR("Unable to sanitize wait queue");
1468 return;
1469}
1470
099e26bd
DG
1471/*
1472 * Dispatch request from the registration threads to the application
1473 * communication thread.
1474 */
1475static void *thread_dispatch_ust_registration(void *data)
1476{
12e2b881 1477 int ret, err = -1;
099e26bd
DG
1478 struct cds_wfq_node *node;
1479 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1480 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1481 struct ust_reg_wait_queue wait_queue = {
1482 .count = 0,
1483 };
d0b96690 1484
12e2b881
MD
1485 health_register(HEALTH_TYPE_APP_REG_DISPATCH);
1486
1487 health_code_update();
1488
f45e313d 1489 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1490
1491 DBG("[thread] Dispatch UST command started");
1492
26c9d55e 1493 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1494 health_code_update();
1495
099e26bd
DG
1496 /* Atomically prepare the queue futex */
1497 futex_nto1_prepare(&ust_cmd_queue.futex);
1498
1499 do {
d0b96690 1500 struct ust_app *app = NULL;
7972aab2 1501 ust_cmd = NULL;
d0b96690 1502
f45e313d
DG
1503 /*
1504 * Make sure we don't have node(s) that have hung up before receiving
1505 * the notify socket. This is to clean the list in order to avoid
1506 * memory leaks from notify socket that are never seen.
1507 */
1508 sanitize_wait_queue(&wait_queue);
1509
12e2b881 1510 health_code_update();
099e26bd
DG
1511 /* Dequeue command for registration */
1512 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1513 if (node == NULL) {
00a17c97 1514 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1515 /* Continue thread execution */
1516 break;
1517 }
1518
1519 ust_cmd = caa_container_of(node, struct ust_command, node);
1520
2f50c8a3
DG
1521 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1522 " gid:%d sock:%d name:%s (version %d.%d)",
1523 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1524 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1525 ust_cmd->sock, ust_cmd->reg_msg.name,
1526 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1527
1528 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1529 wait_node = zmalloc(sizeof(*wait_node));
1530 if (!wait_node) {
1531 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1532 ret = close(ust_cmd->sock);
1533 if (ret < 0) {
1534 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1535 }
1536 lttng_fd_put(1, LTTNG_FD_APPS);
7972aab2 1537 free(ust_cmd);
d0b96690
DG
1538 goto error;
1539 }
1540 CDS_INIT_LIST_HEAD(&wait_node->head);
1541
1542 /* Create application object if socket is CMD. */
1543 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1544 ust_cmd->sock);
1545 if (!wait_node->app) {
1546 ret = close(ust_cmd->sock);
1547 if (ret < 0) {
1548 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1549 }
d88aee68
DG
1550 lttng_fd_put(1, LTTNG_FD_APPS);
1551 free(wait_node);
7972aab2 1552 free(ust_cmd);
d0b96690
DG
1553 continue;
1554 }
1555 /*
1556 * Add application to the wait queue so we can set the notify
1557 * socket before putting this object in the global ht.
1558 */
f45e313d
DG
1559 cds_list_add(&wait_node->head, &wait_queue.head);
1560 wait_queue.count++;
d0b96690 1561
7972aab2 1562 free(ust_cmd);
d0b96690
DG
1563 /*
1564 * We have to continue here since we don't have the notify
1565 * socket and the application MUST be added to the hash table
1566 * only at that moment.
1567 */
1568 continue;
1569 } else {
1570 /*
1571 * Look for the application in the local wait queue and set the
1572 * notify socket if found.
1573 */
d88aee68 1574 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1575 &wait_queue.head, head) {
12e2b881 1576 health_code_update();
d0b96690
DG
1577 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1578 wait_node->app->notify_sock = ust_cmd->sock;
1579 cds_list_del(&wait_node->head);
f45e313d 1580 wait_queue.count--;
d0b96690
DG
1581 app = wait_node->app;
1582 free(wait_node);
1583 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1584 break;
1585 }
1586 }
020d7f60
DG
1587
1588 /*
1589 * With no application at this stage the received socket is
1590 * basically useless so close it before we free the cmd data
1591 * structure for good.
1592 */
1593 if (!app) {
1594 ret = close(ust_cmd->sock);
1595 if (ret < 0) {
1596 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1597 }
1598 lttng_fd_put(1, LTTNG_FD_APPS);
1599 }
7972aab2 1600 free(ust_cmd);
d0b96690
DG
1601 }
1602
1603 if (app) {
d0b96690
DG
1604 /*
1605 * @session_lock_list
1606 *
1607 * Lock the global session list so from the register up to the
1608 * registration done message, no thread can see the application
1609 * and change its state.
1610 */
1611 session_lock_list();
1612 rcu_read_lock();
d88aee68 1613
d0b96690
DG
1614 /*
1615 * Add application to the global hash table. This needs to be
1616 * done before the update to the UST registry can locate the
1617 * application.
1618 */
1619 ust_app_add(app);
d88aee68
DG
1620
1621 /* Set app version. This call will print an error if needed. */
1622 (void) ust_app_version(app);
1623
1624 /* Send notify socket through the notify pipe. */
1625 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1626 app->notify_sock);
1627 if (ret < 0) {
1628 rcu_read_unlock();
1629 session_unlock_list();
1630 /* No notify thread, stop the UST tracing. */
1631 goto error;
6620da75 1632 }
d88aee68 1633
d0b96690
DG
1634 /*
1635 * Update newly registered application with the tracing
1636 * registry info already enabled information.
1637 */
1638 update_ust_app(app->sock);
d88aee68
DG
1639
1640 /*
1641 * Don't care about return value. Let the manage apps threads
1642 * handle app unregistration upon socket close.
1643 */
1644 (void) ust_app_register_done(app->sock);
1645
1646 /*
1647 * Even if the application socket has been closed, send the app
1648 * to the thread and unregistration will take place at that
1649 * place.
1650 */
1651 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 1652 if (ret < 0) {
d88aee68
DG
1653 rcu_read_unlock();
1654 session_unlock_list();
1655 /* No apps. thread, stop the UST tracing. */
1656 goto error;
d0b96690 1657 }
d88aee68 1658
d0b96690
DG
1659 rcu_read_unlock();
1660 session_unlock_list();
099e26bd 1661 }
099e26bd
DG
1662 } while (node != NULL);
1663
12e2b881 1664 health_poll_entry();
099e26bd
DG
1665 /* Futex wait on queue. Blocking call on futex() */
1666 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 1667 health_poll_exit();
099e26bd 1668 }
12e2b881
MD
1669 /* Normal exit, no error */
1670 err = 0;
099e26bd
DG
1671
1672error:
d88aee68
DG
1673 /* Clean up wait queue. */
1674 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1675 &wait_queue.head, head) {
d88aee68 1676 cds_list_del(&wait_node->head);
f45e313d 1677 wait_queue.count--;
d88aee68
DG
1678 free(wait_node);
1679 }
1680
099e26bd 1681 DBG("Dispatch thread dying");
12e2b881
MD
1682 if (err) {
1683 health_error();
1684 ERR("Health error occurred in %s", __func__);
1685 }
1686 health_unregister();
099e26bd
DG
1687 return NULL;
1688}
1689
1690/*
1691 * This thread manage application registration.
1692 */
1693static void *thread_registration_apps(void *data)
1d4b027a 1694{
139ac872 1695 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
1696 uint32_t revents, nb_fd;
1697 struct lttng_poll_event events;
099e26bd
DG
1698 /*
1699 * Get allocated in this thread, enqueued to a global queue, dequeued and
1700 * freed in the manage apps thread.
1701 */
1702 struct ust_command *ust_cmd = NULL;
1d4b027a 1703
099e26bd 1704 DBG("[thread] Manage application registration started");
1d4b027a 1705
927ca06a
DG
1706 health_register(HEALTH_TYPE_APP_REG);
1707
6993eeb3
CB
1708 if (testpoint(thread_registration_apps)) {
1709 goto error_testpoint;
1710 }
8ac94142 1711
1d4b027a
DG
1712 ret = lttcomm_listen_unix_sock(apps_sock);
1713 if (ret < 0) {
76d7553f 1714 goto error_listen;
1d4b027a
DG
1715 }
1716
5eb91c98
DG
1717 /*
1718 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1719 * more will be added to this poll set.
1720 */
d0b96690 1721 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1722 if (ret < 0) {
76d7553f 1723 goto error_create_poll;
5eb91c98 1724 }
273ea72c 1725
5eb91c98
DG
1726 /* Add the application registration socket */
1727 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1728 if (ret < 0) {
76d7553f 1729 goto error_poll_add;
5eb91c98 1730 }
273ea72c 1731
1d4b027a 1732 /* Notify all applications to register */
0fdd1e2c
DG
1733 ret = notify_ust_apps(1);
1734 if (ret < 0) {
1735 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
1736 "Execution continues but there might be problem for already\n"
1737 "running applications that wishes to register.");
0fdd1e2c 1738 }
1d4b027a
DG
1739
1740 while (1) {
1741 DBG("Accepting application registration");
273ea72c
DG
1742
1743 /* Inifinite blocking call, waiting for transmission */
88f2b785 1744 restart:
a78af745 1745 health_poll_entry();
5eb91c98 1746 ret = lttng_poll_wait(&events, -1);
a78af745 1747 health_poll_exit();
273ea72c 1748 if (ret < 0) {
88f2b785
MD
1749 /*
1750 * Restart interrupted system call.
1751 */
1752 if (errno == EINTR) {
1753 goto restart;
1754 }
273ea72c
DG
1755 goto error;
1756 }
1757
0d9c5d77
DG
1758 nb_fd = ret;
1759
5eb91c98 1760 for (i = 0; i < nb_fd; i++) {
840cb59c 1761 health_code_update();
139ac872 1762
5eb91c98
DG
1763 /* Fetch once the poll data */
1764 revents = LTTNG_POLL_GETEV(&events, i);
1765 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 1766
5eb91c98 1767 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1768 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1769 if (ret) {
139ac872
MD
1770 err = 0;
1771 goto exit;
90014c57 1772 }
1d4b027a 1773
5eb91c98
DG
1774 /* Event on the registration socket */
1775 if (pollfd == apps_sock) {
1776 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1777 ERR("Register apps socket poll error");
1778 goto error;
1779 } else if (revents & LPOLLIN) {
1780 sock = lttcomm_accept_unix_sock(apps_sock);
1781 if (sock < 0) {
1782 goto error;
1783 }
099e26bd 1784
b662582b
DG
1785 /*
1786 * Set the CLOEXEC flag. Return code is useless because
1787 * either way, the show must go on.
1788 */
1789 (void) utils_set_fd_cloexec(sock);
1790
5eb91c98 1791 /* Create UST registration command for enqueuing */
ba7f0ae5 1792 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 1793 if (ust_cmd == NULL) {
76d7553f 1794 PERROR("ust command zmalloc");
5eb91c98
DG
1795 goto error;
1796 }
1d4b027a 1797
5eb91c98
DG
1798 /*
1799 * Using message-based transmissions to ensure we don't
1800 * have to deal with partially received messages.
1801 */
4063050c
MD
1802 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1803 if (ret < 0) {
1804 ERR("Exhausted file descriptors allowed for applications.");
1805 free(ust_cmd);
1806 ret = close(sock);
1807 if (ret) {
1808 PERROR("close");
1809 }
1810 sock = -1;
1811 continue;
1812 }
d88aee68 1813
840cb59c 1814 health_code_update();
d0b96690
DG
1815 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
1816 if (ret < 0) {
5eb91c98 1817 free(ust_cmd);
d0b96690 1818 /* Close socket of the application. */
76d7553f
MD
1819 ret = close(sock);
1820 if (ret) {
1821 PERROR("close");
1822 }
4063050c 1823 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 1824 sock = -1;
5eb91c98
DG
1825 continue;
1826 }
840cb59c 1827 health_code_update();
099e26bd 1828
5eb91c98 1829 ust_cmd->sock = sock;
34a2494f 1830 sock = -1;
099e26bd 1831
5eb91c98
DG
1832 DBG("UST registration received with pid:%d ppid:%d uid:%d"
1833 " gid:%d sock:%d name:%s (version %d.%d)",
1834 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1835 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1836 ust_cmd->sock, ust_cmd->reg_msg.name,
1837 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 1838
5eb91c98
DG
1839 /*
1840 * Lock free enqueue the registration request. The red pill
54d01ffb 1841 * has been taken! This apps will be part of the *system*.
5eb91c98
DG
1842 */
1843 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
1844
1845 /*
1846 * Wake the registration queue futex. Implicit memory
1847 * barrier with the exchange in cds_wfq_enqueue.
1848 */
1849 futex_nto1_wake(&ust_cmd_queue.futex);
1850 }
1851 }
90014c57 1852 }
1d4b027a
DG
1853 }
1854
139ac872 1855exit:
1d4b027a 1856error:
139ac872 1857 if (err) {
840cb59c 1858 health_error();
139ac872
MD
1859 ERR("Health error occurred in %s", __func__);
1860 }
139ac872 1861
0fdd1e2c
DG
1862 /* Notify that the registration thread is gone */
1863 notify_ust_apps(0);
1864
a4b35e07 1865 if (apps_sock >= 0) {
76d7553f
MD
1866 ret = close(apps_sock);
1867 if (ret) {
1868 PERROR("close");
1869 }
a4b35e07 1870 }
46c3f085 1871 if (sock >= 0) {
76d7553f
MD
1872 ret = close(sock);
1873 if (ret) {
1874 PERROR("close");
1875 }
4063050c 1876 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 1877 }
273ea72c 1878 unlink(apps_unix_sock_path);
0fdd1e2c 1879
76d7553f 1880error_poll_add:
5eb91c98 1881 lttng_poll_clean(&events);
76d7553f
MD
1882error_listen:
1883error_create_poll:
6993eeb3 1884error_testpoint:
76d7553f 1885 DBG("UST Registration thread cleanup complete");
927ca06a 1886 health_unregister();
5eb91c98 1887
1d4b027a
DG
1888 return NULL;
1889}
1890
8c0faa1d 1891/*
3bd1e081 1892 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 1893 * exec or it will fails.
8c0faa1d 1894 */
3bd1e081 1895static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 1896{
a23ec3a7 1897 int ret, clock_ret;
ee0b0061
DG
1898 struct timespec timeout;
1899
a23ec3a7
DG
1900 /* Make sure we set the readiness flag to 0 because we are NOT ready */
1901 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 1902
a23ec3a7
DG
1903 /* Setup pthread condition */
1904 ret = pthread_condattr_init(&consumer_data->condattr);
1905 if (ret != 0) {
1906 errno = ret;
1907 PERROR("pthread_condattr_init consumer data");
1908 goto error;
1909 }
1910
1911 /*
1912 * Set the monotonic clock in order to make sure we DO NOT jump in time
1913 * between the clock_gettime() call and the timedwait call. See bug #324
1914 * for a more details and how we noticed it.
1915 */
1916 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
1917 if (ret != 0) {
1918 errno = ret;
1919 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
1920 goto error;
1921 }
8c0faa1d 1922
a23ec3a7
DG
1923 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
1924 if (ret != 0) {
1925 errno = ret;
1926 PERROR("pthread_cond_init consumer data");
1927 goto error;
1928 }
1929
1930 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
1931 consumer_data);
8c0faa1d 1932 if (ret != 0) {
3bd1e081 1933 PERROR("pthread_create consumer");
ee0b0061 1934 ret = -1;
8c0faa1d
DG
1935 goto error;
1936 }
1937
a23ec3a7
DG
1938 /* We are about to wait on a pthread condition */
1939 pthread_mutex_lock(&consumer_data->cond_mutex);
1940
ee0b0061 1941 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
1942 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
1943 /*
1944 * Set the timeout for the condition timed wait even if the clock gettime
1945 * call fails since we might loop on that call and we want to avoid to
1946 * increment the timeout too many times.
1947 */
1948 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
1949
1950 /*
1951 * The following loop COULD be skipped in some conditions so this is why we
1952 * set ret to 0 in order to make sure at least one round of the loop is
1953 * done.
1954 */
1955 ret = 0;
1956
1957 /*
1958 * Loop until the condition is reached or when a timeout is reached. Note
1959 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
1960 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
1961 * possible. This loop does not take any chances and works with both of
1962 * them.
1963 */
1964 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
1965 if (clock_ret < 0) {
1966 PERROR("clock_gettime spawn consumer");
1967 /* Infinite wait for the consumerd thread to be ready */
1968 ret = pthread_cond_wait(&consumer_data->cond,
1969 &consumer_data->cond_mutex);
1970 } else {
1971 ret = pthread_cond_timedwait(&consumer_data->cond,
1972 &consumer_data->cond_mutex, &timeout);
1973 }
ee0b0061 1974 }
8c0faa1d 1975
a23ec3a7
DG
1976 /* Release the pthread condition */
1977 pthread_mutex_unlock(&consumer_data->cond_mutex);
1978
1979 if (ret != 0) {
1980 errno = ret;
1981 if (ret == ETIMEDOUT) {
ee0b0061
DG
1982 /*
1983 * Call has timed out so we kill the kconsumerd_thread and return
1984 * an error.
1985 */
a23ec3a7
DG
1986 ERR("Condition timed out. The consumer thread was never ready."
1987 " Killing it");
3bd1e081 1988 ret = pthread_cancel(consumer_data->thread);
ee0b0061 1989 if (ret < 0) {
3bd1e081 1990 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
1991 }
1992 } else {
a23ec3a7 1993 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061
DG
1994 }
1995 goto error;
1996 }
1997
3bd1e081
MD
1998 pthread_mutex_lock(&consumer_data->pid_mutex);
1999 if (consumer_data->pid == 0) {
a23ec3a7 2000 ERR("Consumerd did not start");
3bd1e081 2001 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2002 goto error;
2003 }
3bd1e081 2004 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2005
8c0faa1d
DG
2006 return 0;
2007
2008error:
2009 return ret;
2010}
2011
d9800920 2012/*
3bd1e081 2013 * Join consumer thread
d9800920 2014 */
3bd1e081 2015static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2016{
2017 void *status;
cf3af59e 2018
e8209f6b
DG
2019 /* Consumer pid must be a real one. */
2020 if (consumer_data->pid > 0) {
c617c0c6 2021 int ret;
3bd1e081 2022 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2023 if (ret) {
3bd1e081 2024 ERR("Error killing consumer daemon");
cf3af59e
MD
2025 return ret;
2026 }
3bd1e081 2027 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2028 } else {
2029 return 0;
2030 }
2031}
2032
8c0faa1d 2033/*
3bd1e081 2034 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2035 *
d063d709 2036 * Return pid if successful else -1.
8c0faa1d 2037 */
3bd1e081 2038static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2039{
2040 int ret;
2041 pid_t pid;
94c55f17 2042 const char *consumer_to_use;
53086306 2043 const char *verbosity;
94c55f17 2044 struct stat st;
8c0faa1d 2045
3bd1e081 2046 DBG("Spawning consumerd");
c49dc785 2047
8c0faa1d
DG
2048 pid = fork();
2049 if (pid == 0) {
2050 /*
3bd1e081 2051 * Exec consumerd.
8c0faa1d 2052 */
daee5345 2053 if (opt_verbose_consumer) {
53086306
DG
2054 verbosity = "--verbose";
2055 } else {
2056 verbosity = "--quiet";
2057 }
3bd1e081
MD
2058 switch (consumer_data->type) {
2059 case LTTNG_CONSUMER_KERNEL:
94c55f17 2060 /*
c7704d57
DG
2061 * Find out which consumerd to execute. We will first try the
2062 * 64-bit path, then the sessiond's installation directory, and
2063 * fallback on the 32-bit one,
94c55f17 2064 */
63a799e8
AM
2065 DBG3("Looking for a kernel consumer at these locations:");
2066 DBG3(" 1) %s", consumerd64_bin);
2067 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2068 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2069 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2070 DBG3("Found location #1");
94c55f17 2071 consumer_to_use = consumerd64_bin;
94c55f17 2072 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2073 DBG3("Found location #2");
94c55f17 2074 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2075 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2076 DBG3("Found location #3");
eb1e0bd4 2077 consumer_to_use = consumerd32_bin;
94c55f17 2078 } else {
63a799e8 2079 DBG("Could not find any valid consumerd executable");
94c55f17
AM
2080 break;
2081 }
2082 DBG("Using kernel consumer at: %s", consumer_to_use);
2083 execl(consumer_to_use,
2084 "lttng-consumerd", verbosity, "-k",
2085 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2086 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2087 NULL);
3bd1e081 2088 break;
7753dea8
MD
2089 case LTTNG_CONSUMER64_UST:
2090 {
b1e0b6b6 2091 char *tmpnew = NULL;
8f4905da
MD
2092
2093 if (consumerd64_libdir[0] != '\0') {
2094 char *tmp;
2095 size_t tmplen;
2096
2097 tmp = getenv("LD_LIBRARY_PATH");
2098 if (!tmp) {
2099 tmp = "";
2100 }
2101 tmplen = strlen("LD_LIBRARY_PATH=")
2102 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2103 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2104 if (!tmpnew) {
2105 ret = -ENOMEM;
2106 goto error;
2107 }
2108 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2109 strcat(tmpnew, consumerd64_libdir);
2110 if (tmp[0] != '\0') {
2111 strcat(tmpnew, ":");
2112 strcat(tmpnew, tmp);
2113 }
2114 ret = putenv(tmpnew);
2115 if (ret) {
2116 ret = -errno;
c6f76da9 2117 free(tmpnew);
8f4905da
MD
2118 goto error;
2119 }
2120 }
94c55f17 2121 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2122 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2123 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2124 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2125 NULL);
8f4905da
MD
2126 if (consumerd64_libdir[0] != '\0') {
2127 free(tmpnew);
2128 }
2129 if (ret) {
2130 goto error;
2131 }
3bd1e081 2132 break;
7753dea8
MD
2133 }
2134 case LTTNG_CONSUMER32_UST:
2135 {
937dde8e 2136 char *tmpnew = NULL;
8f4905da
MD
2137
2138 if (consumerd32_libdir[0] != '\0') {
2139 char *tmp;
2140 size_t tmplen;
2141
2142 tmp = getenv("LD_LIBRARY_PATH");
2143 if (!tmp) {
2144 tmp = "";
2145 }
2146 tmplen = strlen("LD_LIBRARY_PATH=")
2147 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2148 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2149 if (!tmpnew) {
2150 ret = -ENOMEM;
2151 goto error;
2152 }
2153 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2154 strcat(tmpnew, consumerd32_libdir);
2155 if (tmp[0] != '\0') {
2156 strcat(tmpnew, ":");
2157 strcat(tmpnew, tmp);
2158 }
2159 ret = putenv(tmpnew);
2160 if (ret) {
2161 ret = -errno;
c6f76da9 2162 free(tmpnew);
8f4905da
MD
2163 goto error;
2164 }
2165 }
94c55f17 2166 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2167 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2168 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2169 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
2170 NULL);
8f4905da
MD
2171 if (consumerd32_libdir[0] != '\0') {
2172 free(tmpnew);
2173 }
2174 if (ret) {
2175 goto error;
2176 }
7753dea8
MD
2177 break;
2178 }
3bd1e081 2179 default:
76d7553f 2180 PERROR("unknown consumer type");
3bd1e081
MD
2181 exit(EXIT_FAILURE);
2182 }
8c0faa1d 2183 if (errno != 0) {
76d7553f 2184 PERROR("kernel start consumer exec");
8c0faa1d
DG
2185 }
2186 exit(EXIT_FAILURE);
2187 } else if (pid > 0) {
2188 ret = pid;
8c0faa1d 2189 } else {
76d7553f 2190 PERROR("start consumer fork");
8c0faa1d 2191 ret = -errno;
8c0faa1d 2192 }
8f4905da 2193error:
8c0faa1d
DG
2194 return ret;
2195}
2196
693bd40b 2197/*
3bd1e081 2198 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2199 */
3bd1e081 2200static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2201{
c617c0c6 2202 int ret;
edb8b045
DG
2203
2204 /*
2205 * Set the listen() state on the socket since there is a possible race
2206 * between the exec() of the consumer daemon and this call if place in the
2207 * consumer thread. See bug #366 for more details.
2208 */
2209 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2210 if (ret < 0) {
2211 goto error;
2212 }
693bd40b 2213
3bd1e081
MD
2214 pthread_mutex_lock(&consumer_data->pid_mutex);
2215 if (consumer_data->pid != 0) {
2216 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2217 goto end;
2218 }
693bd40b 2219
3bd1e081 2220 ret = spawn_consumerd(consumer_data);
c49dc785 2221 if (ret < 0) {
3bd1e081
MD
2222 ERR("Spawning consumerd failed");
2223 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2224 goto error;
693bd40b 2225 }
c49dc785 2226
3bd1e081
MD
2227 /* Setting up the consumer_data pid */
2228 consumer_data->pid = ret;
48842b30 2229 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2230 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2231
3bd1e081
MD
2232 DBG2("Spawning consumer control thread");
2233 ret = spawn_consumer_thread(consumer_data);
693bd40b 2234 if (ret < 0) {
3bd1e081 2235 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2236 goto error;
2237 }
2238
c49dc785 2239end:
693bd40b
DG
2240 return 0;
2241
2242error:
331744e3 2243 /* Cleanup already created sockets on error. */
edb8b045 2244 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2245 int err;
2246
edb8b045
DG
2247 err = close(consumer_data->err_sock);
2248 if (err < 0) {
2249 PERROR("close consumer data error socket");
2250 }
2251 }
693bd40b
DG
2252 return ret;
2253}
2254
44a5e5eb 2255/*
139ac872
MD
2256 * Compute health status of each consumer. If one of them is zero (bad
2257 * state), we return 0.
44a5e5eb
DG
2258 */
2259static int check_consumer_health(void)
2260{
2261 int ret;
2262
927ca06a 2263 ret = health_check_state(HEALTH_TYPE_CONSUMER);
44a5e5eb
DG
2264
2265 DBG3("Health consumer check %d", ret);
2266
2267 return ret;
2268}
2269
b73401da 2270/*
096102bd 2271 * Setup necessary data for kernel tracer action.
b73401da 2272 */
096102bd 2273static int init_kernel_tracer(void)
b73401da
DG
2274{
2275 int ret;
b73401da 2276
096102bd
DG
2277 /* Modprobe lttng kernel modules */
2278 ret = modprobe_lttng_control();
b73401da 2279 if (ret < 0) {
b73401da
DG
2280 goto error;
2281 }
2282
096102bd
DG
2283 /* Open debugfs lttng */
2284 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2285 if (kernel_tracer_fd < 0) {
2286 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2287 ret = -1;
2288 goto error_open;
54d01ffb
DG
2289 }
2290
2f77fc4b
DG
2291 /* Validate kernel version */
2292 ret = kernel_validate_version(kernel_tracer_fd);
2293 if (ret < 0) {
2294 goto error_version;
b551a063 2295 }
54d01ffb 2296
2f77fc4b
DG
2297 ret = modprobe_lttng_data();
2298 if (ret < 0) {
2299 goto error_modules;
54d01ffb
DG
2300 }
2301
2f77fc4b
DG
2302 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2303 return 0;
2304
2305error_version:
2306 modprobe_remove_lttng_control();
2307 ret = close(kernel_tracer_fd);
2308 if (ret) {
2309 PERROR("close");
b551a063 2310 }
2f77fc4b 2311 kernel_tracer_fd = -1;
f73fabfd 2312 return LTTNG_ERR_KERN_VERSION;
b551a063 2313
2f77fc4b
DG
2314error_modules:
2315 ret = close(kernel_tracer_fd);
2316 if (ret) {
2317 PERROR("close");
b551a063 2318 }
54d01ffb 2319
2f77fc4b
DG
2320error_open:
2321 modprobe_remove_lttng_control();
54d01ffb
DG
2322
2323error:
2f77fc4b
DG
2324 WARN("No kernel tracer available");
2325 kernel_tracer_fd = -1;
2326 if (!is_root) {
f73fabfd 2327 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2328 } else {
f73fabfd 2329 return LTTNG_ERR_KERN_NA;
2f77fc4b 2330 }
54d01ffb
DG
2331}
2332
2f77fc4b 2333
54d01ffb 2334/*
2f77fc4b
DG
2335 * Copy consumer output from the tracing session to the domain session. The
2336 * function also applies the right modification on a per domain basis for the
2337 * trace files destination directory.
36b588ed
MD
2338 *
2339 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2340 */
2f77fc4b 2341static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2342{
2343 int ret;
2f77fc4b
DG
2344 const char *dir_name;
2345 struct consumer_output *consumer;
2346
2347 assert(session);
2348 assert(session->consumer);
54d01ffb 2349
b551a063
DG
2350 switch (domain) {
2351 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2352 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2353 /*
2354 * XXX: We should audit the session creation and what this function
2355 * does "extra" in order to avoid a destroy since this function is used
2356 * in the domain session creation (kernel and ust) only. Same for UST
2357 * domain.
2358 */
2359 if (session->kernel_session->consumer) {
2360 consumer_destroy_output(session->kernel_session->consumer);
2361 }
2f77fc4b
DG
2362 session->kernel_session->consumer =
2363 consumer_copy_output(session->consumer);
2364 /* Ease our life a bit for the next part */
2365 consumer = session->kernel_session->consumer;
2366 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063
DG
2367 break;
2368 case LTTNG_DOMAIN_UST:
2f77fc4b 2369 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2370 if (session->ust_session->consumer) {
2371 consumer_destroy_output(session->ust_session->consumer);
2372 }
2f77fc4b
DG
2373 session->ust_session->consumer =
2374 consumer_copy_output(session->consumer);
2375 /* Ease our life a bit for the next part */
2376 consumer = session->ust_session->consumer;
2377 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2378 break;
2379 default:
f73fabfd 2380 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2381 goto error;
2382 }
2383
2f77fc4b 2384 /* Append correct directory to subdir */
c30ce0b3
CB
2385 strncat(consumer->subdir, dir_name,
2386 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2387 DBG3("Copy session consumer subdir %s", consumer->subdir);
2388
f73fabfd 2389 ret = LTTNG_OK;
54d01ffb
DG
2390
2391error:
2392 return ret;
2393}
2394
00e2e675 2395/*
2f77fc4b 2396 * Create an UST session and add it to the session ust list.
36b588ed
MD
2397 *
2398 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2399 */
2f77fc4b
DG
2400static int create_ust_session(struct ltt_session *session,
2401 struct lttng_domain *domain)
00e2e675
DG
2402{
2403 int ret;
2f77fc4b 2404 struct ltt_ust_session *lus = NULL;
00e2e675 2405
a4b92340 2406 assert(session);
2f77fc4b
DG
2407 assert(domain);
2408 assert(session->consumer);
a4b92340 2409
2f77fc4b
DG
2410 switch (domain->type) {
2411 case LTTNG_DOMAIN_UST:
2412 break;
2413 default:
2414 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2415 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2416 goto error;
2417 }
2418
2f77fc4b
DG
2419 DBG("Creating UST session");
2420
dec56f6c 2421 lus = trace_ust_create_session(session->id);
2f77fc4b 2422 if (lus == NULL) {
f73fabfd 2423 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2424 goto error;
2425 }
2426
2f77fc4b
DG
2427 lus->uid = session->uid;
2428 lus->gid = session->gid;
2429 session->ust_session = lus;
00e2e675 2430
2f77fc4b
DG
2431 /* Copy session output to the newly created UST session */
2432 ret = copy_session_consumer(domain->type, session);
f73fabfd 2433 if (ret != LTTNG_OK) {
00e2e675
DG
2434 goto error;
2435 }
2436
f73fabfd 2437 return LTTNG_OK;
00e2e675
DG
2438
2439error:
2f77fc4b
DG
2440 free(lus);
2441 session->ust_session = NULL;
00e2e675
DG
2442 return ret;
2443}
2444
2445/*
2f77fc4b 2446 * Create a kernel tracer session then create the default channel.
00e2e675 2447 */
2f77fc4b 2448static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2449{
2450 int ret;
a4b92340 2451
2f77fc4b 2452 DBG("Creating kernel session");
00e2e675 2453
2f77fc4b
DG
2454 ret = kernel_create_session(session, kernel_tracer_fd);
2455 if (ret < 0) {
f73fabfd 2456 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2457 goto error;
2458 }
2459
2f77fc4b
DG
2460 /* Code flow safety */
2461 assert(session->kernel_session);
2462
2463 /* Copy session output to the newly created Kernel session */
2464 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2465 if (ret != LTTNG_OK) {
a4b92340
DG
2466 goto error;
2467 }
2468
2f77fc4b
DG
2469 /* Create directory(ies) on local filesystem. */
2470 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2471 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2472 ret = run_as_mkdir_recursive(
2473 session->kernel_session->consumer->dst.trace_path,
2474 S_IRWXU | S_IRWXG, session->uid, session->gid);
2475 if (ret < 0) {
2476 if (ret != -EEXIST) {
2477 ERR("Trace directory creation error");
00e2e675
DG
2478 goto error;
2479 }
00e2e675 2480 }
2f77fc4b 2481 }
00e2e675 2482
2f77fc4b
DG
2483 session->kernel_session->uid = session->uid;
2484 session->kernel_session->gid = session->gid;
00e2e675 2485
f73fabfd 2486 return LTTNG_OK;
00e2e675 2487
2f77fc4b
DG
2488error:
2489 trace_kernel_destroy_session(session->kernel_session);
2490 session->kernel_session = NULL;
2491 return ret;
2492}
00e2e675 2493
2f77fc4b
DG
2494/*
2495 * Count number of session permitted by uid/gid.
2496 */
2497static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2498{
2499 unsigned int i = 0;
2500 struct ltt_session *session;
07424f16 2501
2f77fc4b
DG
2502 DBG("Counting number of available session for UID %d GID %d",
2503 uid, gid);
2504 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2505 /*
2f77fc4b 2506 * Only list the sessions the user can control.
00e2e675 2507 */
2f77fc4b
DG
2508 if (!session_access_ok(session, uid, gid)) {
2509 continue;
2510 }
2511 i++;
a4b92340 2512 }
2f77fc4b 2513 return i;
00e2e675
DG
2514}
2515
54d01ffb
DG
2516/*
2517 * Process the command requested by the lttng client within the command
2518 * context structure. This function make sure that the return structure (llm)
2519 * is set and ready for transmission before returning.
2520 *
2521 * Return any error encountered or 0 for success.
53a80697
MD
2522 *
2523 * "sock" is only used for special-case var. len data.
36b588ed
MD
2524 *
2525 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2526 */
53a80697
MD
2527static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2528 int *sock_error)
54d01ffb 2529{
f73fabfd 2530 int ret = LTTNG_OK;
44d3bd01 2531 int need_tracing_session = 1;
2e09ba09 2532 int need_domain;
54d01ffb
DG
2533
2534 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2535
53a80697
MD
2536 *sock_error = 0;
2537
2e09ba09
MD
2538 switch (cmd_ctx->lsm->cmd_type) {
2539 case LTTNG_CREATE_SESSION:
2540 case LTTNG_DESTROY_SESSION:
2541 case LTTNG_LIST_SESSIONS:
2542 case LTTNG_LIST_DOMAINS:
2543 case LTTNG_START_TRACE:
2544 case LTTNG_STOP_TRACE:
6d805429 2545 case LTTNG_DATA_PENDING:
2e09ba09 2546 need_domain = 0;
3aace903 2547 break;
2e09ba09
MD
2548 default:
2549 need_domain = 1;
2550 }
2551
2552 if (opt_no_kernel && need_domain
2553 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2554 if (!is_root) {
f73fabfd 2555 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2556 } else {
f73fabfd 2557 ret = LTTNG_ERR_KERN_NA;
531d29f9 2558 }
4fba7219
DG
2559 goto error;
2560 }
2561
8d3113b2
DG
2562 /* Deny register consumer if we already have a spawned consumer. */
2563 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2564 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2565 if (kconsumer_data.pid > 0) {
f73fabfd 2566 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2567 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2568 goto error;
2569 }
2570 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2571 }
2572
54d01ffb
DG
2573 /*
2574 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2575 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2576 * command.
2577 */
2578 switch(cmd_ctx->lsm->cmd_type) {
2579 case LTTNG_LIST_SESSIONS:
2580 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2581 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2582 case LTTNG_LIST_DOMAINS:
2583 case LTTNG_LIST_CHANNELS:
2584 case LTTNG_LIST_EVENTS:
2585 break;
2586 default:
2587 /* Setup lttng message with no payload */
2588 ret = setup_lttng_msg(cmd_ctx, 0);
2589 if (ret < 0) {
2590 /* This label does not try to unlock the session */
2591 goto init_setup_error;
2592 }
2593 }
2594
2595 /* Commands that DO NOT need a session. */
2596 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2597 case LTTNG_CREATE_SESSION:
2e09ba09 2598 case LTTNG_CALIBRATE:
54d01ffb
DG
2599 case LTTNG_LIST_SESSIONS:
2600 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2601 case LTTNG_LIST_TRACEPOINT_FIELDS:
44d3bd01 2602 need_tracing_session = 0;
54d01ffb
DG
2603 break;
2604 default:
2605 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2606 /*
2607 * We keep the session list lock across _all_ commands
2608 * for now, because the per-session lock does not
2609 * handle teardown properly.
2610 */
74babd95 2611 session_lock_list();
54d01ffb
DG
2612 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2613 if (cmd_ctx->session == NULL) {
bba2d65f 2614 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2615 goto error;
2616 } else {
2617 /* Acquire lock for the session */
2618 session_lock(cmd_ctx->session);
2619 }
2620 break;
2621 }
b389abbe 2622
2e09ba09
MD
2623 if (!need_domain) {
2624 goto skip_domain;
2625 }
a4b92340 2626
54d01ffb
DG
2627 /*
2628 * Check domain type for specific "pre-action".
2629 */
2630 switch (cmd_ctx->lsm->domain.type) {
2631 case LTTNG_DOMAIN_KERNEL:
d1f1c568 2632 if (!is_root) {
f73fabfd 2633 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
2634 goto error;
2635 }
2636
54d01ffb 2637 /* Kernel tracer check */
a4b35e07 2638 if (kernel_tracer_fd == -1) {
54d01ffb 2639 /* Basically, load kernel tracer modules */
096102bd
DG
2640 ret = init_kernel_tracer();
2641 if (ret != 0) {
54d01ffb
DG
2642 goto error;
2643 }
2644 }
5eb91c98 2645
5c827ce0
DG
2646 /* Consumer is in an ERROR state. Report back to client */
2647 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2648 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2649 goto error;
2650 }
2651
54d01ffb 2652 /* Need a session for kernel command */
44d3bd01 2653 if (need_tracing_session) {
54d01ffb 2654 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 2655 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 2656 if (ret < 0) {
f73fabfd 2657 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
2658 goto error;
2659 }
b389abbe 2660 }
7d29a247 2661
54d01ffb 2662 /* Start the kernel consumer daemon */
3bd1e081
MD
2663 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2664 if (kconsumer_data.pid == 0 &&
785d2d0d 2665 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3bd1e081
MD
2666 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2667 ret = start_consumerd(&kconsumer_data);
7d29a247 2668 if (ret < 0) {
f73fabfd 2669 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 2670 goto error;
950131af 2671 }
5c827ce0 2672 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
2673 } else {
2674 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 2675 }
173af62f 2676
a4b92340
DG
2677 /*
2678 * The consumer was just spawned so we need to add the socket to
2679 * the consumer output of the session if exist.
2680 */
2681 ret = consumer_create_socket(&kconsumer_data,
2682 cmd_ctx->session->kernel_session->consumer);
2683 if (ret < 0) {
2684 goto error;
173af62f 2685 }
0d0c377a 2686 }
5c827ce0 2687
54d01ffb 2688 break;
2bdd86d4 2689 case LTTNG_DOMAIN_UST:
44d3bd01 2690 {
5c827ce0
DG
2691 /* Consumer is in an ERROR state. Report back to client */
2692 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2693 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2694 goto error;
2695 }
2696
44d3bd01 2697 if (need_tracing_session) {
a4b92340 2698 /* Create UST session if none exist. */
f6a9efaa 2699 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 2700 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 2701 &cmd_ctx->lsm->domain);
f73fabfd 2702 if (ret != LTTNG_OK) {
44d3bd01
DG
2703 goto error;
2704 }
2705 }
00e2e675 2706
7753dea8
MD
2707 /* Start the UST consumer daemons */
2708 /* 64-bit */
2709 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 2710 if (consumerd64_bin[0] != '\0' &&
7753dea8 2711 ustconsumer64_data.pid == 0 &&
785d2d0d 2712 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2713 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2714 ret = start_consumerd(&ustconsumer64_data);
2bdd86d4 2715 if (ret < 0) {
f73fabfd 2716 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
173af62f 2717 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2bdd86d4
MD
2718 goto error;
2719 }
48842b30 2720
173af62f 2721 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
5c827ce0 2722 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3ff2ecac 2723 } else {
7753dea8
MD
2724 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2725 }
173af62f
DG
2726
2727 /*
2728 * Setup socket for consumer 64 bit. No need for atomic access
2729 * since it was set above and can ONLY be set in this thread.
2730 */
a4b92340
DG
2731 ret = consumer_create_socket(&ustconsumer64_data,
2732 cmd_ctx->session->ust_session->consumer);
2733 if (ret < 0) {
2734 goto error;
173af62f
DG
2735 }
2736
7753dea8 2737 /* 32-bit */
fc7a59ce 2738 if (consumerd32_bin[0] != '\0' &&
7753dea8 2739 ustconsumer32_data.pid == 0 &&
785d2d0d 2740 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2741 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2742 ret = start_consumerd(&ustconsumer32_data);
2743 if (ret < 0) {
f73fabfd 2744 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
173af62f 2745 uatomic_set(&ust_consumerd32_fd, -EINVAL);
7753dea8
MD
2746 goto error;
2747 }
5c827ce0 2748
173af62f 2749 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
5c827ce0 2750 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
7753dea8
MD
2751 } else {
2752 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2bdd86d4 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(&ustconsumer32_data,
2760 cmd_ctx->session->ust_session->consumer);
2761 if (ret < 0) {
2762 goto error;
173af62f 2763 }
44d3bd01
DG
2764 }
2765 break;
48842b30 2766 }
54d01ffb 2767 default:
54d01ffb
DG
2768 break;
2769 }
2e09ba09 2770skip_domain:
33a2b854 2771
5c827ce0
DG
2772 /* Validate consumer daemon state when start/stop trace command */
2773 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2774 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2775 switch (cmd_ctx->lsm->domain.type) {
2776 case LTTNG_DOMAIN_UST:
2777 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2778 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2779 goto error;
2780 }
2781 break;
2782 case LTTNG_DOMAIN_KERNEL:
2783 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2784 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2785 goto error;
2786 }
2787 break;
2788 }
2789 }
2790
8e0af1b4
MD
2791 /*
2792 * Check that the UID or GID match that of the tracing session.
2793 * The root user can interact with all sessions.
2794 */
2795 if (need_tracing_session) {
2796 if (!session_access_ok(cmd_ctx->session,
730389d9
DG
2797 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2798 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
f73fabfd 2799 ret = LTTNG_ERR_EPERM;
8e0af1b4
MD
2800 goto error;
2801 }
2802 }
2803
ffe60014
DG
2804 /*
2805 * Send relayd information to consumer as soon as we have a domain and a
2806 * session defined.
2807 */
2808 if (cmd_ctx->session && need_domain) {
2809 /*
2810 * Setup relayd if not done yet. If the relayd information was already
2811 * sent to the consumer, this call will gracefully return.
2812 */
2813 ret = cmd_setup_relayd(cmd_ctx->session);
2814 if (ret != LTTNG_OK) {
2815 goto error;
2816 }
2817 }
2818
54d01ffb
DG
2819 /* Process by command type */
2820 switch (cmd_ctx->lsm->cmd_type) {
2821 case LTTNG_ADD_CONTEXT:
2822 {
2823 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2824 cmd_ctx->lsm->u.context.channel_name,
979e618e 2825 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
54d01ffb
DG
2826 break;
2827 }
2828 case LTTNG_DISABLE_CHANNEL:
2829 {
2830 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2831 cmd_ctx->lsm->u.disable.channel_name);
2832 break;
2833 }
2834 case LTTNG_DISABLE_EVENT:
2835 {
2836 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2837 cmd_ctx->lsm->u.disable.channel_name,
2838 cmd_ctx->lsm->u.disable.name);
33a2b854
DG
2839 break;
2840 }
54d01ffb
DG
2841 case LTTNG_DISABLE_ALL_EVENT:
2842 {
2bdd86d4 2843 DBG("Disabling all events");
54d01ffb
DG
2844
2845 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2846 cmd_ctx->lsm->u.disable.channel_name);
2847 break;
2848 }
2849 case LTTNG_ENABLE_CHANNEL:
2850 {
7972aab2 2851 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2f77fc4b 2852 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
54d01ffb
DG
2853 break;
2854 }
2855 case LTTNG_ENABLE_EVENT:
2856 {
7972aab2 2857 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
54d01ffb 2858 cmd_ctx->lsm->u.enable.channel_name,
025faf73 2859 &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
2860 break;
2861 }
2862 case LTTNG_ENABLE_ALL_EVENT:
2863 {
2bdd86d4 2864 DBG("Enabling all events");
54d01ffb 2865
7972aab2 2866 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
8c9ae521 2867 cmd_ctx->lsm->u.enable.channel_name,
025faf73 2868 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
2869 break;
2870 }
052da939 2871 case LTTNG_LIST_TRACEPOINTS:
2ef84c95 2872 {
9f19cc17 2873 struct lttng_event *events;
54d01ffb 2874 ssize_t nb_events;
052da939 2875
54d01ffb
DG
2876 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
2877 if (nb_events < 0) {
f73fabfd 2878 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
2879 ret = -nb_events;
2880 goto error;
2ef84c95
DG
2881 }
2882
2883 /*
2884 * Setup lttng message with payload size set to the event list size in
2885 * bytes and then copy list into the llm payload.
2886 */
052da939 2887 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2ef84c95 2888 if (ret < 0) {
052da939 2889 free(events);
2ef84c95
DG
2890 goto setup_error;
2891 }
2892
2893 /* Copy event list into message payload */
9f19cc17 2894 memcpy(cmd_ctx->llm->payload, events,
052da939 2895 sizeof(struct lttng_event) * nb_events);
2ef84c95 2896
9f19cc17 2897 free(events);
2ef84c95 2898
f73fabfd 2899 ret = LTTNG_OK;
2ef84c95
DG
2900 break;
2901 }
f37d259d
MD
2902 case LTTNG_LIST_TRACEPOINT_FIELDS:
2903 {
2904 struct lttng_event_field *fields;
2905 ssize_t nb_fields;
2906
a4b92340
DG
2907 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
2908 &fields);
f37d259d 2909 if (nb_fields < 0) {
f73fabfd 2910 /* Return value is a negative lttng_error_code. */
f37d259d
MD
2911 ret = -nb_fields;
2912 goto error;
2913 }
2914
2915 /*
2916 * Setup lttng message with payload size set to the event list size in
2917 * bytes and then copy list into the llm payload.
2918 */
a4b92340
DG
2919 ret = setup_lttng_msg(cmd_ctx,
2920 sizeof(struct lttng_event_field) * nb_fields);
f37d259d
MD
2921 if (ret < 0) {
2922 free(fields);
2923 goto setup_error;
2924 }
2925
2926 /* Copy event list into message payload */
2927 memcpy(cmd_ctx->llm->payload, fields,
2928 sizeof(struct lttng_event_field) * nb_fields);
2929
2930 free(fields);
2931
f73fabfd 2932 ret = LTTNG_OK;
f37d259d
MD
2933 break;
2934 }
00e2e675
DG
2935 case LTTNG_SET_CONSUMER_URI:
2936 {
a4b92340
DG
2937 size_t nb_uri, len;
2938 struct lttng_uri *uris;
2939
2940 nb_uri = cmd_ctx->lsm->u.uri.size;
2941 len = nb_uri * sizeof(struct lttng_uri);
2942
2943 if (nb_uri == 0) {
f73fabfd 2944 ret = LTTNG_ERR_INVALID;
a4b92340
DG
2945 goto error;
2946 }
2947
2948 uris = zmalloc(len);
2949 if (uris == NULL) {
f73fabfd 2950 ret = LTTNG_ERR_FATAL;
a4b92340
DG
2951 goto error;
2952 }
2953
2954 /* Receive variable len data */
77c7c900 2955 DBG("Receiving %zu URI(s) from client ...", nb_uri);
a4b92340
DG
2956 ret = lttcomm_recv_unix_sock(sock, uris, len);
2957 if (ret <= 0) {
2958 DBG("No URIs received from client... continuing");
2959 *sock_error = 1;
f73fabfd 2960 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 2961 free(uris);
a4b92340
DG
2962 goto error;
2963 }
2964
00e2e675 2965 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
a4b92340 2966 nb_uri, uris);
f73fabfd 2967 if (ret != LTTNG_OK) {
b1d41407 2968 free(uris);
a4b92340
DG
2969 goto error;
2970 }
2971
2972 /*
2973 * XXX: 0 means that this URI should be applied on the session. Should
2974 * be a DOMAIN enuam.
2975 */
2976 if (cmd_ctx->lsm->domain.type == 0) {
2977 /* Add the URI for the UST session if a consumer is present. */
2978 if (cmd_ctx->session->ust_session &&
2979 cmd_ctx->session->ust_session->consumer) {
2980 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
2981 nb_uri, uris);
2982 } else if (cmd_ctx->session->kernel_session &&
2983 cmd_ctx->session->kernel_session->consumer) {
2984 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
2985 cmd_ctx->session, nb_uri, uris);
2986 }
2987 }
2988
b1d41407
DG
2989 free(uris);
2990
00e2e675
DG
2991 break;
2992 }
f3ed775e 2993 case LTTNG_START_TRACE:
8c0faa1d 2994 {
54d01ffb 2995 ret = cmd_start_trace(cmd_ctx->session);
8c0faa1d
DG
2996 break;
2997 }
f3ed775e 2998 case LTTNG_STOP_TRACE:
8c0faa1d 2999 {
54d01ffb 3000 ret = cmd_stop_trace(cmd_ctx->session);
8c0faa1d
DG
3001 break;
3002 }
5e16da05
MD
3003 case LTTNG_CREATE_SESSION:
3004 {
a4b92340
DG
3005 size_t nb_uri, len;
3006 struct lttng_uri *uris = NULL;
3007
3008 nb_uri = cmd_ctx->lsm->u.uri.size;
3009 len = nb_uri * sizeof(struct lttng_uri);
3010
3011 if (nb_uri > 0) {
3012 uris = zmalloc(len);
3013 if (uris == NULL) {
f73fabfd 3014 ret = LTTNG_ERR_FATAL;
a4b92340
DG
3015 goto error;
3016 }
3017
3018 /* Receive variable len data */
77c7c900 3019 DBG("Waiting for %zu URIs from client ...", nb_uri);
a4b92340
DG
3020 ret = lttcomm_recv_unix_sock(sock, uris, len);
3021 if (ret <= 0) {
3022 DBG("No URIs received from client... continuing");
3023 *sock_error = 1;
f73fabfd 3024 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3025 free(uris);
a4b92340
DG
3026 goto error;
3027 }
3028
3029 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3030 DBG("Creating session with ONE network URI is a bad call");
f73fabfd 3031 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3032 free(uris);
a4b92340
DG
3033 goto error;
3034 }
3035 }
3036
3037 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
3038 &cmd_ctx->creds);
3039
b1d41407
DG
3040 free(uris);
3041
00e2e675
DG
3042 break;
3043 }
5e16da05
MD
3044 case LTTNG_DESTROY_SESSION:
3045 {
2f77fc4b 3046 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
a4b92340
DG
3047
3048 /* Set session to NULL so we do not unlock it after free. */
256a5576 3049 cmd_ctx->session = NULL;
5461b305 3050 break;
5e16da05 3051 }
9f19cc17 3052 case LTTNG_LIST_DOMAINS:
5e16da05 3053 {
54d01ffb
DG
3054 ssize_t nb_dom;
3055 struct lttng_domain *domains;
5461b305 3056
54d01ffb
DG
3057 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3058 if (nb_dom < 0) {
f73fabfd 3059 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3060 ret = -nb_dom;
3061 goto error;
ce3d728c 3062 }
520ff687 3063
54d01ffb 3064 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
5461b305 3065 if (ret < 0) {
ae9c20bc 3066 free(domains);
5461b305 3067 goto setup_error;
520ff687 3068 }
57167058 3069
54d01ffb
DG
3070 /* Copy event list into message payload */
3071 memcpy(cmd_ctx->llm->payload, domains,
3072 nb_dom * sizeof(struct lttng_domain));
3073
3074 free(domains);
5e16da05 3075
f73fabfd 3076 ret = LTTNG_OK;
5e16da05
MD
3077 break;
3078 }
9f19cc17 3079 case LTTNG_LIST_CHANNELS:
5e16da05 3080 {
6775595e 3081 int nb_chan;
54d01ffb
DG
3082 struct lttng_channel *channels;
3083
b551a063
DG
3084 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3085 cmd_ctx->session, &channels);
54d01ffb 3086 if (nb_chan < 0) {
f73fabfd 3087 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3088 ret = -nb_chan;
3089 goto error;
5461b305 3090 }
ca95a216 3091
54d01ffb 3092 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
5461b305 3093 if (ret < 0) {
ae9c20bc 3094 free(channels);
5461b305
DG
3095 goto setup_error;
3096 }
9f19cc17 3097
54d01ffb
DG
3098 /* Copy event list into message payload */
3099 memcpy(cmd_ctx->llm->payload, channels,
3100 nb_chan * sizeof(struct lttng_channel));
3101
3102 free(channels);
9f19cc17 3103
f73fabfd 3104 ret = LTTNG_OK;
5461b305 3105 break;
5e16da05 3106 }
9f19cc17 3107 case LTTNG_LIST_EVENTS:
5e16da05 3108 {
b551a063 3109 ssize_t nb_event;
684d34d2 3110 struct lttng_event *events = NULL;
9f19cc17 3111
b551a063 3112 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
54d01ffb
DG
3113 cmd_ctx->lsm->u.list.channel_name, &events);
3114 if (nb_event < 0) {
f73fabfd 3115 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3116 ret = -nb_event;
3117 goto error;
5461b305 3118 }
ca95a216 3119
54d01ffb 3120 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
5461b305 3121 if (ret < 0) {
ae9c20bc 3122 free(events);
5461b305
DG
3123 goto setup_error;
3124 }
9f19cc17 3125
54d01ffb
DG
3126 /* Copy event list into message payload */
3127 memcpy(cmd_ctx->llm->payload, events,
3128 nb_event * sizeof(struct lttng_event));
9f19cc17 3129
54d01ffb 3130 free(events);
9f19cc17 3131
f73fabfd 3132 ret = LTTNG_OK;
5461b305 3133 break;
5e16da05
MD
3134 }
3135 case LTTNG_LIST_SESSIONS:
3136 {
8e0af1b4 3137 unsigned int nr_sessions;
5461b305 3138
8e0af1b4 3139 session_lock_list();
730389d9
DG
3140 nr_sessions = lttng_sessions_count(
3141 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3142 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
d32fb093 3143
8e0af1b4 3144 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
5461b305 3145 if (ret < 0) {
54d01ffb 3146 session_unlock_list();
5461b305 3147 goto setup_error;
e065084a 3148 }
5e16da05 3149
6c9cc2ab 3150 /* Filled the session array */
2f77fc4b 3151 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
730389d9
DG
3152 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3153 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
6c9cc2ab 3154
54d01ffb 3155 session_unlock_list();
5e16da05 3156
f73fabfd 3157 ret = LTTNG_OK;
5e16da05
MD
3158 break;
3159 }
d0254c7c
MD
3160 case LTTNG_CALIBRATE:
3161 {
54d01ffb
DG
3162 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3163 &cmd_ctx->lsm->u.calibrate);
d0254c7c
MD
3164 break;
3165 }
d9800920
DG
3166 case LTTNG_REGISTER_CONSUMER:
3167 {
2f77fc4b
DG
3168 struct consumer_data *cdata;
3169
3170 switch (cmd_ctx->lsm->domain.type) {
3171 case LTTNG_DOMAIN_KERNEL:
3172 cdata = &kconsumer_data;
3173 break;
3174 default:
f73fabfd 3175 ret = LTTNG_ERR_UND;
2f77fc4b
DG
3176 goto error;
3177 }
3178
54d01ffb 3179 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 3180 cmd_ctx->lsm->u.reg.path, cdata);
d9800920
DG
3181 break;
3182 }
025faf73 3183 case LTTNG_ENABLE_EVENT_WITH_FILTER:
53a80697
MD
3184 {
3185 struct lttng_filter_bytecode *bytecode;
3186
025faf73 3187 if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) {
f73fabfd 3188 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
3189 goto error;
3190 }
52df2401
MD
3191 if (cmd_ctx->lsm->u.enable.bytecode_len == 0) {
3192 ret = LTTNG_ERR_FILTER_INVAL;
3193 goto error;
3194 }
025faf73 3195 bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len);
53a80697 3196 if (!bytecode) {
f73fabfd 3197 ret = LTTNG_ERR_FILTER_NOMEM;
53a80697
MD
3198 goto error;
3199 }
3200 /* Receive var. len. data */
3201 DBG("Receiving var len data from client ...");
3202 ret = lttcomm_recv_unix_sock(sock, bytecode,
025faf73 3203 cmd_ctx->lsm->u.enable.bytecode_len);
53a80697
MD
3204 if (ret <= 0) {
3205 DBG("Nothing recv() from client var len data... continuing");
3206 *sock_error = 1;
f73fabfd 3207 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
3208 goto error;
3209 }
3210
3211 if (bytecode->len + sizeof(*bytecode)
025faf73 3212 != cmd_ctx->lsm->u.enable.bytecode_len) {
53a80697 3213 free(bytecode);
f73fabfd 3214 ret = LTTNG_ERR_FILTER_INVAL;
53a80697
MD
3215 goto error;
3216 }
3217
7972aab2 3218 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
025faf73
DG
3219 cmd_ctx->lsm->u.enable.channel_name,
3220 &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]);
53a80697
MD
3221 break;
3222 }
6d805429 3223 case LTTNG_DATA_PENDING:
806e2684 3224 {
6d805429 3225 ret = cmd_data_pending(cmd_ctx->session);
806e2684
DG
3226 break;
3227 }
5e16da05 3228 default:
f73fabfd 3229 ret = LTTNG_ERR_UND;
5461b305 3230 break;
fac6795d
DG
3231 }
3232
5461b305 3233error:
5461b305
DG
3234 if (cmd_ctx->llm == NULL) {
3235 DBG("Missing llm structure. Allocating one.");
894be886 3236 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
5461b305
DG
3237 goto setup_error;
3238 }
3239 }
54d01ffb 3240 /* Set return code */
5461b305 3241 cmd_ctx->llm->ret_code = ret;
5461b305 3242setup_error:
b5541356 3243 if (cmd_ctx->session) {
54d01ffb 3244 session_unlock(cmd_ctx->session);
b5541356 3245 }
256a5576
MD
3246 if (need_tracing_session) {
3247 session_unlock_list();
3248 }
54d01ffb 3249init_setup_error:
8028d920 3250 return ret;
fac6795d
DG
3251}
3252
44a5e5eb
DG
3253/*
3254 * Thread managing health check socket.
3255 */
3256static void *thread_manage_health(void *data)
3257{
eb4a2943 3258 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
44a5e5eb
DG
3259 uint32_t revents, nb_fd;
3260 struct lttng_poll_event events;
3261 struct lttcomm_health_msg msg;
3262 struct lttcomm_health_data reply;
3263
3264 DBG("[thread] Manage health check started");
3265
3266 rcu_register_thread();
3267
6d737ce4
DG
3268 /* We might hit an error path before this is created. */
3269 lttng_poll_init(&events);
3cc04881 3270
44a5e5eb
DG
3271 /* Create unix socket */
3272 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3273 if (sock < 0) {
3274 ERR("Unable to create health check Unix socket");
3275 ret = -1;
3276 goto error;
3277 }
3278
b662582b
DG
3279 /*
3280 * Set the CLOEXEC flag. Return code is useless because either way, the
3281 * show must go on.
3282 */
3283 (void) utils_set_fd_cloexec(sock);
3284
44a5e5eb
DG
3285 ret = lttcomm_listen_unix_sock(sock);
3286 if (ret < 0) {
3287 goto error;
3288 }
3289
3290 /*
3291 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3292 * more will be added to this poll set.
3293 */
d0b96690 3294 ret = sessiond_set_thread_pollset(&events, 2);
44a5e5eb
DG
3295 if (ret < 0) {
3296 goto error;
3297 }
3298
3299 /* Add the application registration socket */
3300 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3301 if (ret < 0) {
3302 goto error;
3303 }
3304
3305 while (1) {
3306 DBG("Health check ready");
3307
44a5e5eb
DG
3308 /* Inifinite blocking call, waiting for transmission */
3309restart:
3310 ret = lttng_poll_wait(&events, -1);
3311 if (ret < 0) {
3312 /*
3313 * Restart interrupted system call.
3314 */
3315 if (errno == EINTR) {
3316 goto restart;
3317 }
3318 goto error;
3319 }
3320
0d9c5d77
DG
3321 nb_fd = ret;
3322
44a5e5eb
DG
3323 for (i = 0; i < nb_fd; i++) {
3324 /* Fetch once the poll data */
3325 revents = LTTNG_POLL_GETEV(&events, i);
3326 pollfd = LTTNG_POLL_GETFD(&events, i);
3327
3328 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 3329 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
44a5e5eb 3330 if (ret) {
139ac872
MD
3331 err = 0;
3332 goto exit;
44a5e5eb
DG
3333 }
3334
3335 /* Event on the registration socket */
3336 if (pollfd == sock) {
3337 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3338 ERR("Health socket poll error");
3339 goto error;
3340 }
3341 }
3342 }
3343
3344 new_sock = lttcomm_accept_unix_sock(sock);
3345 if (new_sock < 0) {
3346 goto error;
3347 }
3348
b662582b
DG
3349 /*
3350 * Set the CLOEXEC flag. Return code is useless because either way, the
3351 * show must go on.
3352 */
3353 (void) utils_set_fd_cloexec(new_sock);
3354
44a5e5eb
DG
3355 DBG("Receiving data from client for health...");
3356 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3357 if (ret <= 0) {
3358 DBG("Nothing recv() from client... continuing");
3359 ret = close(new_sock);
3360 if (ret) {
3361 PERROR("close");
3362 }
3363 new_sock = -1;
3364 continue;
3365 }
3366
3367 rcu_thread_online();
3368
3369 switch (msg.component) {
3370 case LTTNG_HEALTH_CMD:
927ca06a 3371 reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
44a5e5eb 3372 break;
139ac872 3373 case LTTNG_HEALTH_APP_MANAGE:
927ca06a 3374 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
139ac872 3375 break;
44a5e5eb 3376 case LTTNG_HEALTH_APP_REG:
927ca06a 3377 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
44a5e5eb
DG
3378 break;
3379 case LTTNG_HEALTH_KERNEL:
927ca06a 3380 reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
44a5e5eb
DG
3381 break;
3382 case LTTNG_HEALTH_CONSUMER:
3383 reply.ret_code = check_consumer_health();
3384 break;
0b2dc8df
MD
3385 case LTTNG_HEALTH_HT_CLEANUP:
3386 reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
3387 break;
380e8d6f
MD
3388 case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
3389 reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
3390 break;
12e2b881
MD
3391 case LTTNG_HEALTH_APP_REG_DISPATCH:
3392 reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
3393 break;
44a5e5eb 3394 case LTTNG_HEALTH_ALL:
44a5e5eb 3395 reply.ret_code =
927ca06a
DG
3396 health_check_state(HEALTH_TYPE_APP_MANAGE) &&
3397 health_check_state(HEALTH_TYPE_APP_REG) &&
3398 health_check_state(HEALTH_TYPE_CMD) &&
3399 health_check_state(HEALTH_TYPE_KERNEL) &&
0b2dc8df 3400 check_consumer_health() &&
380e8d6f 3401 health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
12e2b881
MD
3402 health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
3403 health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
44a5e5eb
DG
3404 break;
3405 default:
f73fabfd 3406 reply.ret_code = LTTNG_ERR_UND;
44a5e5eb
DG
3407 break;
3408 }
3409
3410 /*
3411 * Flip ret value since 0 is a success and 1 indicates a bad health for
3412 * the client where in the sessiond it is the opposite. Again, this is
3413 * just to make things easier for us poor developer which enjoy a lot
3414 * lazyness.
3415 */
3416 if (reply.ret_code == 0 || reply.ret_code == 1) {
3417 reply.ret_code = !reply.ret_code;
3418 }
3419
3420 DBG2("Health check return value %d", reply.ret_code);
3421
3422 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3423 if (ret < 0) {
3424 ERR("Failed to send health data back to client");
3425 }
3426
3427 /* End of transmission */
3428 ret = close(new_sock);
3429 if (ret) {
3430 PERROR("close");
3431 }
3432 new_sock = -1;
3433 }
3434
139ac872 3435exit:
44a5e5eb 3436error:
139ac872
MD
3437 if (err) {
3438 ERR("Health error occurred in %s", __func__);
3439 }
44a5e5eb
DG
3440 DBG("Health check thread dying");
3441 unlink(health_unix_sock_path);
3442 if (sock >= 0) {
3443 ret = close(sock);
3444 if (ret) {
3445 PERROR("close");
3446 }
3447 }
44a5e5eb
DG
3448
3449 lttng_poll_clean(&events);
3450
3451 rcu_unregister_thread();
3452 return NULL;
3453}
3454
1d4b027a 3455/*
d063d709
DG
3456 * This thread manage all clients request using the unix client socket for
3457 * communication.
1d4b027a
DG
3458 */
3459static void *thread_manage_clients(void *data)
3460{
139ac872 3461 int sock = -1, ret, i, pollfd, err = -1;
53a80697 3462 int sock_error;
5eb91c98 3463 uint32_t revents, nb_fd;
273ea72c 3464 struct command_ctx *cmd_ctx = NULL;
5eb91c98 3465 struct lttng_poll_event events;
1d4b027a
DG
3466
3467 DBG("[thread] Manage client started");
3468
f6a9efaa
DG
3469 rcu_register_thread();
3470
927ca06a
DG
3471 health_register(HEALTH_TYPE_CMD);
3472
6993eeb3
CB
3473 if (testpoint(thread_manage_clients)) {
3474 goto error_testpoint;
3475 }
3476
840cb59c 3477 health_code_update();
44a5e5eb 3478
1d4b027a
DG
3479 ret = lttcomm_listen_unix_sock(client_sock);
3480 if (ret < 0) {
35df2755 3481 goto error_listen;
1d4b027a
DG
3482 }
3483
5eb91c98
DG
3484 /*
3485 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3486 * more will be added to this poll set.
3487 */
d0b96690 3488 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 3489 if (ret < 0) {
35df2755 3490 goto error_create_poll;
5eb91c98 3491 }
273ea72c 3492
5eb91c98
DG
3493 /* Add the application registration socket */
3494 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3495 if (ret < 0) {
3496 goto error;
3497 }
273ea72c 3498
bbd973c2
DG
3499 /*
3500 * Notify parent pid that we are ready to accept command for client side.
1d4b027a
DG
3501 */
3502 if (opt_sig_parent) {
8db8d1dc 3503 kill(ppid, SIGUSR1);
1d4b027a
DG
3504 }
3505
6993eeb3
CB
3506 if (testpoint(thread_manage_clients_before_loop)) {
3507 goto error;
3508 }
8ac94142 3509
840cb59c 3510 health_code_update();
44a5e5eb 3511
1d4b027a 3512 while (1) {
1d4b027a 3513 DBG("Accepting client command ...");
273ea72c
DG
3514
3515 /* Inifinite blocking call, waiting for transmission */
88f2b785 3516 restart:
a78af745 3517 health_poll_entry();
5eb91c98 3518 ret = lttng_poll_wait(&events, -1);
a78af745 3519 health_poll_exit();
273ea72c 3520 if (ret < 0) {
88f2b785
MD
3521 /*
3522 * Restart interrupted system call.
3523 */
3524 if (errno == EINTR) {
3525 goto restart;
3526 }
273ea72c
DG
3527 goto error;
3528 }
3529
0d9c5d77
DG
3530 nb_fd = ret;
3531
5eb91c98
DG
3532 for (i = 0; i < nb_fd; i++) {
3533 /* Fetch once the poll data */
3534 revents = LTTNG_POLL_GETEV(&events, i);
3535 pollfd = LTTNG_POLL_GETFD(&events, i);
3536
840cb59c 3537 health_code_update();
44a5e5eb 3538
5eb91c98 3539 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 3540 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 3541 if (ret) {
139ac872
MD
3542 err = 0;
3543 goto exit;
5eb91c98
DG
3544 }
3545
3546 /* Event on the registration socket */
3547 if (pollfd == client_sock) {
3548 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3549 ERR("Client socket poll error");
3550 goto error;
3551 }
3552 }
3553 }
3554
3555 DBG("Wait for client response");
3556
840cb59c 3557 health_code_update();
44a5e5eb 3558
5eb91c98
DG
3559 sock = lttcomm_accept_unix_sock(client_sock);
3560 if (sock < 0) {
273ea72c 3561 goto error;
273ea72c
DG
3562 }
3563
b662582b
DG
3564 /*
3565 * Set the CLOEXEC flag. Return code is useless because either way, the
3566 * show must go on.
3567 */
3568 (void) utils_set_fd_cloexec(sock);
3569
be040666
DG
3570 /* Set socket option for credentials retrieval */
3571 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3572 if (ret < 0) {
3573 goto error;
3574 }
3575
5eb91c98 3576 /* Allocate context command to process the client request */
ba7f0ae5 3577 cmd_ctx = zmalloc(sizeof(struct command_ctx));
5eb91c98 3578 if (cmd_ctx == NULL) {
76d7553f 3579 PERROR("zmalloc cmd_ctx");
1d4b027a 3580 goto error;
5eb91c98 3581 }
1d4b027a 3582
5eb91c98 3583 /* Allocate data buffer for reception */
ba7f0ae5 3584 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
5eb91c98 3585 if (cmd_ctx->lsm == NULL) {
76d7553f 3586 PERROR("zmalloc cmd_ctx->lsm");
5eb91c98
DG
3587 goto error;
3588 }
1d4b027a 3589
5eb91c98
DG
3590 cmd_ctx->llm = NULL;
3591 cmd_ctx->session = NULL;
1d4b027a 3592
840cb59c 3593 health_code_update();
44a5e5eb 3594
5eb91c98
DG
3595 /*
3596 * Data is received from the lttng client. The struct
3597 * lttcomm_session_msg (lsm) contains the command and data request of
3598 * the client.
3599 */
3600 DBG("Receiving data from client ...");
be040666
DG
3601 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3602 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
5eb91c98
DG
3603 if (ret <= 0) {
3604 DBG("Nothing recv() from client... continuing");
76d7553f
MD
3605 ret = close(sock);
3606 if (ret) {
3607 PERROR("close");
3608 }
3609 sock = -1;
a2c0da86 3610 clean_command_ctx(&cmd_ctx);
5eb91c98
DG
3611 continue;
3612 }
1d4b027a 3613
840cb59c 3614 health_code_update();
44a5e5eb 3615
5eb91c98
DG
3616 // TODO: Validate cmd_ctx including sanity check for
3617 // security purpose.
f7776ea7 3618
f6a9efaa 3619 rcu_thread_online();
5eb91c98
DG
3620 /*
3621 * This function dispatch the work to the kernel or userspace tracer
3622 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3623 * informations for the client. The command context struct contains
3624 * everything this function may needs.
3625 */
53a80697 3626 ret = process_client_msg(cmd_ctx, sock, &sock_error);
f6a9efaa 3627 rcu_thread_offline();
5eb91c98 3628 if (ret < 0) {
36134aa1
DG
3629 ret = close(sock);
3630 if (ret) {
3631 PERROR("close");
53a80697 3632 }
36134aa1 3633 sock = -1;
bbd973c2 3634 /*
5eb91c98 3635 * TODO: Inform client somehow of the fatal error. At
ba7f0ae5 3636 * this point, ret < 0 means that a zmalloc failed
53a80697
MD
3637 * (ENOMEM). Error detected but still accept
3638 * command, unless a socket error has been
3639 * detected.
bbd973c2 3640 */
bbd973c2 3641 clean_command_ctx(&cmd_ctx);
5eb91c98
DG
3642 continue;
3643 }
d6e4fca4 3644
840cb59c 3645 health_code_update();
44a5e5eb 3646
54d01ffb
DG
3647 DBG("Sending response (size: %d, retcode: %s)",
3648 cmd_ctx->lttng_msg_size,
9a745bc7 3649 lttng_strerror(-cmd_ctx->llm->ret_code));
54d01ffb 3650 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
5eb91c98
DG
3651 if (ret < 0) {
3652 ERR("Failed to send data back to client");
bbd973c2 3653 }
1d4b027a 3654
5eb91c98 3655 /* End of transmission */
76d7553f
MD
3656 ret = close(sock);
3657 if (ret) {
3658 PERROR("close");
3659 }
3660 sock = -1;
be040666
DG
3661
3662 clean_command_ctx(&cmd_ctx);
44a5e5eb 3663
840cb59c 3664 health_code_update();
273ea72c
DG
3665 }
3666
139ac872 3667exit:
5eb91c98 3668error:
35df2755
CB
3669 if (sock >= 0) {
3670 ret = close(sock);
3671 if (ret) {
3672 PERROR("close");
3673 }
139ac872 3674 }
44a5e5eb 3675
35df2755
CB
3676 lttng_poll_clean(&events);
3677 clean_command_ctx(&cmd_ctx);
3678
3679error_listen:
3680error_create_poll:
6993eeb3 3681error_testpoint:
273ea72c 3682 unlink(client_unix_sock_path);
76d7553f
MD
3683 if (client_sock >= 0) {
3684 ret = close(client_sock);
3685 if (ret) {
3686 PERROR("close");
3687 }
a4b35e07 3688 }
35df2755
CB
3689
3690 if (err) {
840cb59c 3691 health_error();
35df2755 3692 ERR("Health error occurred in %s", __func__);
a4b35e07 3693 }
273ea72c 3694
927ca06a 3695 health_unregister();
35df2755
CB
3696
3697 DBG("Client thread dying");
f6a9efaa
DG
3698
3699 rcu_unregister_thread();
1d4b027a
DG
3700 return NULL;
3701}
3702
3703
fac6795d
DG
3704/*
3705 * usage function on stderr
3706 */
3707static void usage(void)
3708{
b716ce68 3709 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
d6f42150
DG
3710 fprintf(stderr, " -h, --help Display this usage.\n");
3711 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
3712 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
3713 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
3714 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
7753dea8
MD
3715 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
3716 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
3717 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
3718 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
ebaeda94
MD
3719 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
3720 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
3721 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
3722 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
d6f42150
DG
3723 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
3724 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
3725 fprintf(stderr, " -V, --version Show version number.\n");
3726 fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n");
3727 fprintf(stderr, " -q, --quiet No output at all.\n");
3728 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
35f90c40 3729 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3bd1e081 3730 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4fba7219 3731 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
fac6795d
DG
3732}
3733
3734/*
3735 * daemon argument parsing
3736 */
3737static int parse_args(int argc, char **argv)
3738{
3739 int c;
3740
3741 static struct option long_options[] = {
3742 { "client-sock", 1, 0, 'c' },
3743 { "apps-sock", 1, 0, 'a' },
3bd1e081
MD
3744 { "kconsumerd-cmd-sock", 1, 0, 'C' },
3745 { "kconsumerd-err-sock", 1, 0, 'E' },
7753dea8
MD
3746 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
3747 { "ustconsumerd32-err-sock", 1, 0, 'H' },
ebaeda94
MD
3748 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
3749 { "ustconsumerd64-err-sock", 1, 0, 'F' },
3750 { "consumerd32-path", 1, 0, 'u' },
3751 { "consumerd32-libdir", 1, 0, 'U' },
3752 { "consumerd64-path", 1, 0, 't' },
3753 { "consumerd64-libdir", 1, 0, 'T' },
fac6795d 3754 { "daemonize", 0, 0, 'd' },
5b8719f5 3755 { "sig-parent", 0, 0, 'S' },
fac6795d
DG
3756 { "help", 0, 0, 'h' },
3757 { "group", 1, 0, 'g' },
3758 { "version", 0, 0, 'V' },
75462a81 3759 { "quiet", 0, 0, 'q' },
3f9947db 3760 { "verbose", 0, 0, 'v' },
3bd1e081 3761 { "verbose-consumer", 0, 0, 'Z' },
4fba7219 3762 { "no-kernel", 0, 0, 'N' },
35f90c40 3763 { "pidfile", 1, 0, 'p' },
fac6795d
DG
3764 { NULL, 0, 0, 0 }
3765 };
3766
3767 while (1) {
3768 int option_index = 0;
35f90c40 3769 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:",
54d01ffb 3770 long_options, &option_index);
fac6795d
DG
3771 if (c == -1) {
3772 break;
3773 }
3774
3775 switch (c) {
3776 case 0:
3777 fprintf(stderr, "option %s", long_options[option_index].name);
3778 if (optarg) {
3779 fprintf(stderr, " with arg %s\n", optarg);
3780 }
3781 break;
b716ce68 3782 case 'c':
fac6795d
DG
3783 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
3784 break;
3785 case 'a':
3786 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
3787 break;
3788 case 'd':
3789 opt_daemon = 1;
3790 break;
3791 case 'g':
fb09408a 3792 opt_tracing_group = optarg;
fac6795d
DG
3793 break;
3794 case 'h':
3795 usage();
3796 exit(EXIT_FAILURE);
3797 case 'V':
3798 fprintf(stdout, "%s\n", VERSION);
3799 exit(EXIT_SUCCESS);
5b8719f5
DG
3800 case 'S':
3801 opt_sig_parent = 1;
3802 break;
d6f42150 3803 case 'E':
3bd1e081 3804 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
d6f42150
DG
3805 break;
3806 case 'C':
3bd1e081
MD
3807 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3808 break;
3809 case 'F':
7753dea8 3810 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3bd1e081
MD
3811 break;
3812 case 'D':
7753dea8
MD
3813 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
3814 break;
3815 case 'H':
3816 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3817 break;
3818 case 'G':
3819 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
d6f42150 3820 break;
4fba7219
DG
3821 case 'N':
3822 opt_no_kernel = 1;
3823 break;
75462a81 3824 case 'q':
97e19046 3825 lttng_opt_quiet = 1;
75462a81 3826 break;
3f9947db 3827 case 'v':
53086306 3828 /* Verbose level can increase using multiple -v */
97e19046 3829 lttng_opt_verbose += 1;
3f9947db 3830 break;
31f73cc9 3831 case 'Z':
3bd1e081 3832 opt_verbose_consumer += 1;
31f73cc9 3833 break;
fb09408a 3834 case 'u':
fc7a59ce 3835 consumerd32_bin= optarg;
ebaeda94
MD
3836 break;
3837 case 'U':
3838 consumerd32_libdir = optarg;
7753dea8
MD
3839 break;
3840 case 't':
fc7a59ce 3841 consumerd64_bin = optarg;
ebaeda94
MD
3842 break;
3843 case 'T':
3844 consumerd64_libdir = optarg;
fb09408a 3845 break;
35f90c40
DG
3846 case 'p':
3847 opt_pidfile = optarg;
3848 break;
fac6795d
DG
3849 default:
3850 /* Unknown option or other error.
3851 * Error is printed by getopt, just return */
3852 return -1;
3853 }
3854 }
3855
3856 return 0;
3857}
3858
3859/*
d063d709 3860 * Creates the two needed socket by the daemon.
d6f42150
DG
3861 * apps_sock - The communication socket for all UST apps.
3862 * client_sock - The communication of the cli tool (lttng).
fac6795d 3863 */
cf3af59e 3864static int init_daemon_socket(void)
fac6795d
DG
3865{
3866 int ret = 0;
3867 mode_t old_umask;
3868
3869 old_umask = umask(0);
3870
3871 /* Create client tool unix socket */
d6f42150
DG
3872 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
3873 if (client_sock < 0) {
3874 ERR("Create unix sock failed: %s", client_unix_sock_path);
fac6795d
DG
3875 ret = -1;
3876 goto end;
3877 }
3878
b662582b
DG
3879 /* Set the cloexec flag */
3880 ret = utils_set_fd_cloexec(client_sock);
3881 if (ret < 0) {
3882 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
3883 "Continuing but note that the consumer daemon will have a "
3884 "reference to this socket on exec()", client_sock);
3885 }
3886
fac6795d
DG
3887 /* File permission MUST be 660 */
3888 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3889 if (ret < 0) {
d6f42150 3890 ERR("Set file permissions failed: %s", client_unix_sock_path);
76d7553f 3891 PERROR("chmod");
fac6795d
DG
3892 goto end;
3893 }
3894
3895 /* Create the application unix socket */
d6f42150
DG
3896 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
3897 if (apps_sock < 0) {
3898 ERR("Create unix sock failed: %s", apps_unix_sock_path);
fac6795d
DG
3899 ret = -1;
3900 goto end;
3901 }
3902
b662582b
DG
3903 /* Set the cloexec flag */
3904 ret = utils_set_fd_cloexec(apps_sock);
3905 if (ret < 0) {
3906 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
3907 "Continuing but note that the consumer daemon will have a "
3908 "reference to this socket on exec()", apps_sock);
3909 }
3910
d6f42150 3911 /* File permission MUST be 666 */
54d01ffb
DG
3912 ret = chmod(apps_unix_sock_path,
3913 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
fac6795d 3914 if (ret < 0) {
d6f42150 3915 ERR("Set file permissions failed: %s", apps_unix_sock_path);
76d7553f 3916 PERROR("chmod");
fac6795d
DG
3917 goto end;
3918 }
3919
b662582b
DG
3920 DBG3("Session daemon client socket %d and application socket %d created",
3921 client_sock, apps_sock);
3922
fac6795d
DG
3923end:
3924 umask(old_umask);
3925 return ret;
3926}
3927
3928/*
54d01ffb
DG
3929 * Check if the global socket is available, and if a daemon is answering at the
3930 * other side. If yes, error is returned.
fac6795d 3931 */
cf3af59e 3932static int check_existing_daemon(void)
fac6795d 3933{
7d8234d9 3934 /* Is there anybody out there ? */
099e26bd 3935 if (lttng_session_daemon_alive()) {
7d8234d9 3936 return -EEXIST;
099e26bd 3937 }
b09c7c76
DG
3938
3939 return 0;
fac6795d
DG
3940}
3941
fac6795d 3942/*
d063d709 3943 * Set the tracing group gid onto the client socket.
5e16da05 3944 *
d063d709 3945 * Race window between mkdir and chown is OK because we are going from more
d1613cf5 3946 * permissive (root.root) to less permissive (root.tracing).
fac6795d 3947 */
be040666 3948static int set_permissions(char *rundir)
fac6795d
DG
3949{
3950 int ret;
996b65c8 3951 gid_t gid;
fac6795d 3952
6775595e
DG
3953 ret = allowed_group();
3954 if (ret < 0) {
be040666
DG
3955 WARN("No tracing group detected");
3956 ret = 0;
fac6795d
DG
3957 goto end;
3958 }
3959
6775595e
DG
3960 gid = ret;
3961
d6f42150 3962 /* Set lttng run dir */
be040666 3963 ret = chown(rundir, 0, gid);
d6f42150 3964 if (ret < 0) {
be040666 3965 ERR("Unable to set group on %s", rundir);
76d7553f 3966 PERROR("chown");
d6f42150
DG
3967 }
3968
d1613cf5 3969 /* Ensure tracing group can search the run dir */
61076f74 3970 ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH);
d1613cf5
JN
3971 if (ret < 0) {
3972 ERR("Unable to set permissions on %s", rundir);
76d7553f 3973 PERROR("chmod");
d1613cf5
JN
3974 }
3975
d6f42150 3976 /* lttng client socket path */
996b65c8 3977 ret = chown(client_unix_sock_path, 0, gid);
fac6795d 3978 if (ret < 0) {
d6f42150 3979 ERR("Unable to set group on %s", client_unix_sock_path);
76d7553f 3980 PERROR("chown");
d6f42150
DG
3981 }
3982
3bd1e081
MD
3983 /* kconsumer error socket path */
3984 ret = chown(kconsumer_data.err_unix_sock_path, 0, gid);
d6f42150 3985 if (ret < 0) {
3bd1e081 3986 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
76d7553f 3987 PERROR("chown");
3bd1e081
MD
3988 }
3989
7753dea8
MD
3990 /* 64-bit ustconsumer error socket path */
3991 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid);
3992 if (ret < 0) {
3993 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
76d7553f 3994 PERROR("chown");
7753dea8
MD
3995 }
3996
3997 /* 32-bit ustconsumer compat32 error socket path */
3998 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid);
3bd1e081 3999 if (ret < 0) {
7753dea8 4000 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
76d7553f 4001 PERROR("chown");
fac6795d
DG
4002 }
4003
d6f42150 4004 DBG("All permissions are set");
e07ae692 4005
fac6795d
DG
4006end:
4007 return ret;
4008}
4009
d6f42150 4010/*
d063d709 4011 * Create the lttng run directory needed for all global sockets and pipe.
d6f42150 4012 */
67e40797 4013static int create_lttng_rundir(const char *rundir)
d6f42150
DG
4014{
4015 int ret;
4016
67e40797
DG
4017 DBG3("Creating LTTng run directory: %s", rundir);
4018
d1613cf5 4019 ret = mkdir(rundir, S_IRWXU);
d6f42150 4020 if (ret < 0) {
b1f11e69 4021 if (errno != EEXIST) {
67e40797 4022 ERR("Unable to create %s", rundir);
b1f11e69
DG
4023 goto error;
4024 } else {
4025 ret = 0;
4026 }
d6f42150
DG
4027 }
4028
4029error:
4030 return ret;
4031}
4032
4033/*
d063d709
DG
4034 * Setup sockets and directory needed by the kconsumerd communication with the
4035 * session daemon.
d6f42150 4036 */
67e40797
DG
4037static int set_consumer_sockets(struct consumer_data *consumer_data,
4038 const char *rundir)
d6f42150
DG
4039{
4040 int ret;
67e40797 4041 char path[PATH_MAX];
d6f42150 4042
67e40797 4043 switch (consumer_data->type) {
7753dea8 4044 case LTTNG_CONSUMER_KERNEL:
60922cb0 4045 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
7753dea8
MD
4046 break;
4047 case LTTNG_CONSUMER64_UST:
60922cb0 4048 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
7753dea8
MD
4049 break;
4050 case LTTNG_CONSUMER32_UST:
60922cb0 4051 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
7753dea8
MD
4052 break;
4053 default:
4054 ERR("Consumer type unknown");
4055 ret = -EINVAL;
4056 goto error;
d6f42150
DG
4057 }
4058
67e40797
DG
4059 DBG2("Creating consumer directory: %s", path);
4060
d1613cf5 4061 ret = mkdir(path, S_IRWXU);
d6f42150 4062 if (ret < 0) {
6beb2242 4063 if (errno != EEXIST) {
f11e84c2 4064 PERROR("mkdir");
3bd1e081 4065 ERR("Failed to create %s", path);
6beb2242
DG
4066 goto error;
4067 }
f11e84c2 4068 ret = -1;
d6f42150
DG
4069 }
4070
4071 /* Create the kconsumerd error unix socket */
3bd1e081
MD
4072 consumer_data->err_sock =
4073 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4074 if (consumer_data->err_sock < 0) {
4075 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
d6f42150
DG
4076 ret = -1;
4077 goto error;
4078 }
4079
4080 /* File permission MUST be 660 */
3bd1e081 4081 ret = chmod(consumer_data->err_unix_sock_path,
54d01ffb 4082 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
d6f42150 4083 if (ret < 0) {
3bd1e081 4084 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
67e40797 4085 PERROR("chmod");
d6f42150
DG
4086 goto error;
4087 }
4088
4089error:
4090 return ret;
4091}
4092
fac6795d 4093/*
d063d709 4094 * Signal handler for the daemon
cf3af59e 4095 *
54d01ffb
DG
4096 * Simply stop all worker threads, leaving main() return gracefully after
4097 * joining all threads and calling cleanup().
fac6795d
DG
4098 */
4099static void sighandler(int sig)
4100{
4101 switch (sig) {
cf3af59e 4102 case SIGPIPE:
af87c45a 4103 DBG("SIGPIPE caught");
cf3af59e
MD
4104 return;
4105 case SIGINT:
af87c45a 4106 DBG("SIGINT caught");
cf3af59e
MD
4107 stop_threads();
4108 break;
4109 case SIGTERM:
af87c45a 4110 DBG("SIGTERM caught");
cf3af59e
MD
4111 stop_threads();
4112 break;
4113 default:
4114 break;
fac6795d 4115 }
fac6795d
DG
4116}
4117
4118/*
d063d709 4119 * Setup signal handler for :
1d4b027a 4120 * SIGINT, SIGTERM, SIGPIPE
fac6795d 4121 */
1d4b027a 4122static int set_signal_handler(void)
fac6795d 4123{
1d4b027a
DG
4124 int ret = 0;
4125 struct sigaction sa;
4126 sigset_t sigset;
fac6795d 4127
1d4b027a 4128 if ((ret = sigemptyset(&sigset)) < 0) {
76d7553f 4129 PERROR("sigemptyset");
1d4b027a
DG
4130 return ret;
4131 }
d6f42150 4132
1d4b027a
DG
4133 sa.sa_handler = sighandler;
4134 sa.sa_mask = sigset;
4135 sa.sa_flags = 0;
4136 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
76d7553f 4137 PERROR("sigaction");
1d4b027a 4138 return ret;
d6f42150
DG
4139 }
4140
1d4b027a 4141 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
76d7553f 4142 PERROR("sigaction");
1d4b027a 4143 return ret;
d6f42150 4144 }
aaf26714 4145
1d4b027a 4146 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
76d7553f 4147 PERROR("sigaction");
1d4b027a 4148 return ret;
8c0faa1d
DG
4149 }
4150
1d4b027a
DG
4151 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
4152
4153 return ret;
fac6795d
DG
4154}
4155
f3ed775e 4156/*
d063d709
DG
4157 * Set open files limit to unlimited. This daemon can open a large number of
4158 * file descriptors in order to consumer multiple kernel traces.
f3ed775e
DG
4159 */
4160static void set_ulimit(void)
4161{
4162 int ret;
4163 struct rlimit lim;
4164
a88df331 4165 /* The kernel does not allowed an infinite limit for open files */
f3ed775e
DG
4166 lim.rlim_cur = 65535;
4167 lim.rlim_max = 65535;
4168
4169 ret = setrlimit(RLIMIT_NOFILE, &lim);
4170 if (ret < 0) {
76d7553f 4171 PERROR("failed to set open files limit");
f3ed775e
DG
4172 }
4173}
4174
35f90c40
DG
4175/*
4176 * Write pidfile using the rundir and opt_pidfile.
4177 */
4178static void write_pidfile(void)
4179{
4180 int ret;
4181 char pidfile_path[PATH_MAX];
4182
4183 assert(rundir);
4184
4185 if (opt_pidfile) {
4186 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4187 } else {
4188 /* Build pidfile path from rundir and opt_pidfile. */
4189 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4190 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4191 if (ret < 0) {
4192 PERROR("snprintf pidfile path");
4193 goto error;
4194 }
4195 }
4196
4197 /*
4198 * Create pid file in rundir. Return value is of no importance. The
4199 * execution will continue even though we are not able to write the file.
4200 */
4201 (void) utils_create_pid_file(getpid(), pidfile_path);
4202
4203error:
4204 return;
4205}
4206
fac6795d
DG
4207/*
4208 * main
4209 */
4210int main(int argc, char **argv)
4211{
fac6795d
DG
4212 int ret = 0;
4213 void *status;
ae9e45b3 4214 const char *home_path, *env_app_timeout;
fac6795d 4215
335a95b7
MD
4216 init_kernel_workarounds();
4217
f6a9efaa
DG
4218 rcu_register_thread();
4219
7753dea8 4220 setup_consumerd_path();
fb09408a 4221
12744796
DG
4222 page_size = sysconf(_SC_PAGESIZE);
4223 if (page_size < 0) {
4224 PERROR("sysconf _SC_PAGESIZE");
4225 page_size = LONG_MAX;
4226 WARN("Fallback page size to %ld", page_size);
4227 }
4228
fac6795d
DG
4229 /* Parse arguments */
4230 progname = argv[0];
c617c0c6 4231 if ((ret = parse_args(argc, argv)) < 0) {
cf3af59e 4232 goto error;
fac6795d
DG
4233 }
4234
4235 /* Daemonize */
4236 if (opt_daemon) {
ceed52b5
MD
4237 int i;
4238
4239 /*
4240 * fork
4241 * child: setsid, close FD 0, 1, 2, chdir /
4242 * parent: exit (if fork is successful)
4243 */
53094c05
DG
4244 ret = daemon(0, 0);
4245 if (ret < 0) {
76d7553f 4246 PERROR("daemon");
cf3af59e 4247 goto error;
53094c05 4248 }
ceed52b5
MD
4249 /*
4250 * We are in the child. Make sure all other file
4251 * descriptors are closed, in case we are called with
4252 * more opened file descriptors than the standard ones.
4253 */
4254 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4255 (void) close(i);
4256 }
4257 }
4258
4259 /* Create thread quit pipe */
4260 if ((ret = init_thread_quit_pipe()) < 0) {
4261 goto error;
fac6795d
DG
4262 }
4263
4264 /* Check if daemon is UID = 0 */
4265 is_root = !getuid();
4266
fac6795d 4267 if (is_root) {
990570ed 4268 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
67e40797
DG
4269
4270 /* Create global run dir with root access */
4271 ret = create_lttng_rundir(rundir);
d6f42150 4272 if (ret < 0) {
cf3af59e 4273 goto error;
d6f42150
DG
4274 }
4275
fac6795d 4276 if (strlen(apps_unix_sock_path) == 0) {
d6f42150
DG
4277 snprintf(apps_unix_sock_path, PATH_MAX,
4278 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
fac6795d
DG
4279 }
4280
4281 if (strlen(client_unix_sock_path) == 0) {
d6f42150
DG
4282 snprintf(client_unix_sock_path, PATH_MAX,
4283 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4284 }
0fdd1e2c
DG
4285
4286 /* Set global SHM for ust */
4287 if (strlen(wait_shm_path) == 0) {
4288 snprintf(wait_shm_path, PATH_MAX,
4289 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4290 }
67e40797 4291
44a5e5eb
DG
4292 if (strlen(health_unix_sock_path) == 0) {
4293 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4294 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4295 }
4296
67e40797
DG
4297 /* Setup kernel consumerd path */
4298 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
60922cb0 4299 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
67e40797 4300 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4301 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
67e40797
DG
4302
4303 DBG2("Kernel consumer err path: %s",
4304 kconsumer_data.err_unix_sock_path);
4305 DBG2("Kernel consumer cmd path: %s",
4306 kconsumer_data.cmd_unix_sock_path);
fac6795d 4307 } else {
feb0f3e5 4308 home_path = utils_get_home_dir();
b082db07 4309 if (home_path == NULL) {
273ea72c
DG
4310 /* TODO: Add --socket PATH option */
4311 ERR("Can't get HOME directory for sockets creation.");
cf3af59e
MD
4312 ret = -EPERM;
4313 goto error;
b082db07
DG
4314 }
4315
67e40797
DG
4316 /*
4317 * Create rundir from home path. This will create something like
4318 * $HOME/.lttng
4319 */
990570ed 4320 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
67e40797
DG
4321 if (ret < 0) {
4322 ret = -ENOMEM;
4323 goto error;
4324 }
4325
4326 ret = create_lttng_rundir(rundir);
4327 if (ret < 0) {
4328 goto error;
4329 }
4330
fac6795d 4331 if (strlen(apps_unix_sock_path) == 0) {
d6f42150 4332 snprintf(apps_unix_sock_path, PATH_MAX,
b082db07 4333 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
fac6795d
DG
4334 }
4335
4336 /* Set the cli tool unix socket path */
4337 if (strlen(client_unix_sock_path) == 0) {
d6f42150 4338 snprintf(client_unix_sock_path, PATH_MAX,
b082db07 4339 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
fac6795d 4340 }
0fdd1e2c
DG
4341
4342 /* Set global SHM for ust */
4343 if (strlen(wait_shm_path) == 0) {
4344 snprintf(wait_shm_path, PATH_MAX,
d0b96690 4345 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
0fdd1e2c 4346 }
44a5e5eb
DG
4347
4348 /* Set health check Unix path */
4349 if (strlen(health_unix_sock_path) == 0) {
4350 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4351 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4352 }
fac6795d
DG
4353 }
4354
5c827ce0
DG
4355 /* Set consumer initial state */
4356 kernel_consumerd_state = CONSUMER_STOPPED;
4357 ust_consumerd_state = CONSUMER_STOPPED;
4358
847177cd
DG
4359 DBG("Client socket path %s", client_unix_sock_path);
4360 DBG("Application socket path %s", apps_unix_sock_path);
d0b96690 4361 DBG("Application wait path %s", wait_shm_path);
67e40797
DG
4362 DBG("LTTng run directory path: %s", rundir);
4363
4364 /* 32 bits consumerd path setup */
4365 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
60922cb0 4366 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
67e40797 4367 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4368 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
67e40797
DG
4369
4370 DBG2("UST consumer 32 bits err path: %s",
4371 ustconsumer32_data.err_unix_sock_path);
4372 DBG2("UST consumer 32 bits cmd path: %s",
4373 ustconsumer32_data.cmd_unix_sock_path);
4374
4375 /* 64 bits consumerd path setup */
4376 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
60922cb0 4377 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
67e40797 4378 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4379 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
67e40797
DG
4380
4381 DBG2("UST consumer 64 bits err path: %s",
4382 ustconsumer64_data.err_unix_sock_path);
4383 DBG2("UST consumer 64 bits cmd path: %s",
4384 ustconsumer64_data.cmd_unix_sock_path);
847177cd 4385
273ea72c 4386 /*
7d8234d9 4387 * See if daemon already exist.
fac6795d 4388 */
7d8234d9 4389 if ((ret = check_existing_daemon()) < 0) {
75462a81 4390 ERR("Already running daemon.\n");
273ea72c 4391 /*
cf3af59e
MD
4392 * We do not goto exit because we must not cleanup()
4393 * because a daemon is already running.
ab118b20 4394 */
cf3af59e 4395 goto error;
a88df331
DG
4396 }
4397
1427f9b2
DG
4398 /*
4399 * Init UST app hash table. Alloc hash table before this point since
4400 * cleanup() can get called after that point.
4401 */
4402 ust_app_ht_alloc();
4403
54d01ffb 4404 /* After this point, we can safely call cleanup() with "goto exit" */
a88df331
DG
4405
4406 /*
4407 * These actions must be executed as root. We do that *after* setting up
4408 * the sockets path because we MUST make the check for another daemon using
4409 * those paths *before* trying to set the kernel consumer sockets and init
4410 * kernel tracer.
4411 */
4412 if (is_root) {
67e40797 4413 ret = set_consumer_sockets(&kconsumer_data, rundir);
7753dea8
MD
4414 if (ret < 0) {
4415 goto exit;
4416 }
4417
a88df331 4418 /* Setup kernel tracer */
4fba7219
DG
4419 if (!opt_no_kernel) {
4420 init_kernel_tracer();
4421 }
a88df331
DG
4422
4423 /* Set ulimit for open files */
4424 set_ulimit();
fac6795d 4425 }
4063050c
MD
4426 /* init lttng_fd tracking must be done after set_ulimit. */
4427 lttng_fd_init();
fac6795d 4428
67e40797
DG
4429 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4430 if (ret < 0) {
4431 goto exit;
4432 }
4433
4434 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4435 if (ret < 0) {
4436 goto exit;
4437 }
4438
cf3af59e
MD
4439 if ((ret = set_signal_handler()) < 0) {
4440 goto exit;
fac6795d
DG
4441 }
4442
d6f42150 4443 /* Setup the needed unix socket */
cf3af59e
MD
4444 if ((ret = init_daemon_socket()) < 0) {
4445 goto exit;
fac6795d
DG
4446 }
4447
4448 /* Set credentials to socket */
be040666 4449 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
cf3af59e 4450 goto exit;
fac6795d
DG
4451 }
4452
5b8719f5
DG
4453 /* Get parent pid if -S, --sig-parent is specified. */
4454 if (opt_sig_parent) {
4455 ppid = getppid();
4456 }
4457
7a485870 4458 /* Setup the kernel pipe for waking up the kernel thread */
6620da75
DG
4459 if (is_root && !opt_no_kernel) {
4460 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4461 goto exit;
4462 }
7a485870
DG
4463 }
4464
0b2dc8df
MD
4465 /* Setup the thread ht_cleanup communication pipe. */
4466 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4467 goto exit;
4468 }
4469
099e26bd 4470 /* Setup the thread apps communication pipe. */
ef599319 4471 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
099e26bd
DG
4472 goto exit;
4473 }
4474
d0b96690
DG
4475 /* Setup the thread apps notify communication pipe. */
4476 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4477 goto exit;
4478 }
4479
7972aab2
DG
4480 /* Initialize global buffer per UID and PID registry. */
4481 buffer_reg_init_uid_registry();
4482 buffer_reg_init_pid_registry();
4483
099e26bd
DG
4484 /* Init UST command queue. */
4485 cds_wfq_init(&ust_cmd_queue.queue);
4486
273ea72c 4487 /*
54d01ffb
DG
4488 * Get session list pointer. This pointer MUST NOT be free(). This list is
4489 * statically declared in session.c
273ea72c 4490 */
54d01ffb 4491 session_list_ptr = session_get_list();
b5541356 4492
5eb91c98
DG
4493 /* Set up max poll set size */
4494 lttng_poll_set_max_size();
4495
2f77fc4b 4496 cmd_init();
00e2e675 4497
ae9e45b3
DG
4498 /* Check for the application socket timeout env variable. */
4499 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4500 if (env_app_timeout) {
4501 app_socket_timeout = atoi(env_app_timeout);
4502 } else {
4503 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4504 }
4505
35f90c40
DG
4506 write_pidfile();
4507
0b2dc8df
MD
4508 /* Create thread to manage the client socket */
4509 ret = pthread_create(&ht_cleanup_thread, NULL,
4510 thread_ht_cleanup, (void *) NULL);
4511 if (ret != 0) {
4512 PERROR("pthread_create ht_cleanup");
4513 goto exit_ht_cleanup;
4514 }
4515
44a5e5eb
DG
4516 /* Create thread to manage the client socket */
4517 ret = pthread_create(&health_thread, NULL,
4518 thread_manage_health, (void *) NULL);
4519 if (ret != 0) {
4520 PERROR("pthread_create health");
4521 goto exit_health;
4522 }
4523
cf3af59e 4524 /* Create thread to manage the client socket */
099e26bd
DG
4525 ret = pthread_create(&client_thread, NULL,
4526 thread_manage_clients, (void *) NULL);
cf3af59e 4527 if (ret != 0) {
76d7553f 4528 PERROR("pthread_create clients");
cf3af59e
MD
4529 goto exit_client;
4530 }
fac6795d 4531
099e26bd
DG
4532 /* Create thread to dispatch registration */
4533 ret = pthread_create(&dispatch_thread, NULL,
4534 thread_dispatch_ust_registration, (void *) NULL);
4535 if (ret != 0) {
76d7553f 4536 PERROR("pthread_create dispatch");
099e26bd
DG
4537 goto exit_dispatch;
4538 }
4539
4540 /* Create thread to manage application registration. */
4541 ret = pthread_create(&reg_apps_thread, NULL,
4542 thread_registration_apps, (void *) NULL);
4543 if (ret != 0) {
76d7553f 4544 PERROR("pthread_create registration");
099e26bd
DG
4545 goto exit_reg_apps;
4546 }
4547
cf3af59e 4548 /* Create thread to manage application socket */
54d01ffb
DG
4549 ret = pthread_create(&apps_thread, NULL,
4550 thread_manage_apps, (void *) NULL);
cf3af59e 4551 if (ret != 0) {
d0b96690
DG
4552 PERROR("pthread_create apps");
4553 goto exit_apps;
4554 }
4555
4556 /* Create thread to manage application notify socket */
4557 ret = pthread_create(&apps_notify_thread, NULL,
4558 ust_thread_manage_notify, (void *) NULL);
4559 if (ret != 0) {
76d7553f 4560 PERROR("pthread_create apps");
cf3af59e
MD
4561 goto exit_apps;
4562 }
fac6795d 4563
6620da75
DG
4564 /* Don't start this thread if kernel tracing is not requested nor root */
4565 if (is_root && !opt_no_kernel) {
4566 /* Create kernel thread to manage kernel event */
4567 ret = pthread_create(&kernel_thread, NULL,
4568 thread_manage_kernel, (void *) NULL);
4569 if (ret != 0) {
4570 PERROR("pthread_create kernel");
4571 goto exit_kernel;
4572 }
7a485870 4573
6620da75
DG
4574 ret = pthread_join(kernel_thread, &status);
4575 if (ret != 0) {
4576 PERROR("pthread_join");
4577 goto error; /* join error, exit without cleanup */
4578 }
fac6795d
DG
4579 }
4580
cf3af59e
MD
4581exit_kernel:
4582 ret = pthread_join(apps_thread, &status);
4583 if (ret != 0) {
76d7553f 4584 PERROR("pthread_join");
cf3af59e
MD
4585 goto error; /* join error, exit without cleanup */
4586 }
fac6795d 4587
cf3af59e 4588exit_apps:
099e26bd
DG
4589 ret = pthread_join(reg_apps_thread, &status);
4590 if (ret != 0) {
76d7553f 4591 PERROR("pthread_join");
099e26bd
DG
4592 goto error; /* join error, exit without cleanup */
4593 }
4594
4595exit_reg_apps:
4596 ret = pthread_join(dispatch_thread, &status);
4597 if (ret != 0) {
76d7553f 4598 PERROR("pthread_join");
099e26bd
DG
4599 goto error; /* join error, exit without cleanup */
4600 }
4601
4602exit_dispatch:
cf3af59e
MD
4603 ret = pthread_join(client_thread, &status);
4604 if (ret != 0) {
76d7553f 4605 PERROR("pthread_join");
cf3af59e
MD
4606 goto error; /* join error, exit without cleanup */
4607 }
4608
3bd1e081 4609 ret = join_consumer_thread(&kconsumer_data);
cf3af59e 4610 if (ret != 0) {
76d7553f 4611 PERROR("join_consumer");
cf3af59e
MD
4612 goto error; /* join error, exit without cleanup */
4613 }
a88df331 4614
06f525de
DG
4615 ret = join_consumer_thread(&ustconsumer32_data);
4616 if (ret != 0) {
4617 PERROR("join_consumer ust32");
4618 goto error; /* join error, exit without cleanup */
4619 }
4620
4621 ret = join_consumer_thread(&ustconsumer64_data);
4622 if (ret != 0) {
4623 PERROR("join_consumer ust64");
4624 goto error; /* join error, exit without cleanup */
4625 }
4626
cf3af59e 4627exit_client:
06f525de
DG
4628 ret = pthread_join(health_thread, &status);
4629 if (ret != 0) {
4630 PERROR("pthread_join health thread");
4631 goto error; /* join error, exit without cleanup */
4632 }
4633
44a5e5eb 4634exit_health:
0b2dc8df
MD
4635 ret = pthread_join(ht_cleanup_thread, &status);
4636 if (ret != 0) {
4637 PERROR("pthread_join ht cleanup thread");
4638 goto error; /* join error, exit without cleanup */
4639 }
4640exit_ht_cleanup:
a88df331 4641exit:
cf3af59e
MD
4642 /*
4643 * cleanup() is called when no other thread is running.
4644 */
f6a9efaa 4645 rcu_thread_online();
cf3af59e 4646 cleanup();
f6a9efaa
DG
4647 rcu_thread_offline();
4648 rcu_unregister_thread();
67e40797 4649 if (!ret) {
cf3af59e 4650 exit(EXIT_SUCCESS);
67e40797 4651 }
cf3af59e 4652error:
5e16da05 4653 exit(EXIT_FAILURE);
fac6795d 4654}
This page took 0.368051 seconds and 4 git commands to generate.