Fix: poll: show the correct number of fds
[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>
0bb7724a 23#include <paths.h>
fac6795d
DG
24#include <pthread.h>
25#include <signal.h>
26#include <stdio.h>
27#include <stdlib.h>
28#include <string.h>
331744e3 29#include <inttypes.h>
0fdd1e2c 30#include <sys/mman.h>
b73401da 31#include <sys/mount.h>
1e307fab 32#include <sys/resource.h>
fac6795d
DG
33#include <sys/socket.h>
34#include <sys/stat.h>
35#include <sys/types.h>
0fdd1e2c 36#include <sys/wait.h>
5c827ce0 37#include <urcu/uatomic.h>
fac6795d 38#include <unistd.h>
3bd1e081 39#include <config.h>
fac6795d 40
990570ed 41#include <common/common.h>
d27c42b8 42#include <common/compat/socket.h>
db758600
DG
43#include <common/defaults.h>
44#include <common/kernel-consumer/kernel-consumer.h>
50c8f484 45#include <common/futex.h>
00e2e675 46#include <common/relayd/relayd.h>
81b86775 47#include <common/utils.h>
449935a1 48#include <common/daemonize.h>
fac6795d 49
10a8a223 50#include "lttng-sessiond.h"
7972aab2 51#include "buffer-registry.h"
54d01ffb 52#include "channel.h"
2f77fc4b 53#include "cmd.h"
00e2e675 54#include "consumer.h"
099e26bd 55#include "context.h"
54d01ffb 56#include "event.h"
4771f025 57#include "kernel.h"
f1e16794 58#include "kernel-consumer.h"
096102bd 59#include "modprobe.h"
0fdd1e2c 60#include "shm.h"
1e307fab 61#include "ust-ctl.h"
00e2e675 62#include "ust-consumer.h"
8e68d1c8 63#include "utils.h"
4063050c 64#include "fd-limit.h"
8782cc74 65#include "health-sessiond.h"
8ac94142 66#include "testpoint.h"
d0b96690 67#include "ust-thread.h"
4d076222 68#include "jul-thread.h"
fac6795d 69
ebaeda94
MD
70#define CONSUMERD_FILE "lttng-consumerd"
71
fac6795d 72const char *progname;
6c71277b 73static const char *tracing_group_name = DEFAULT_TRACING_GROUP;
35f90c40 74static const char *opt_pidfile;
5b8719f5 75static int opt_sig_parent;
97e19046 76static int opt_verbose_consumer;
929de64e 77static int opt_daemon, opt_background;
4fba7219 78static int opt_no_kernel;
1d4b027a 79static pid_t ppid; /* Parent PID for --sig-parent option */
0bb7724a 80static pid_t child_ppid; /* Internal parent PID use with daemonize. */
67e40797 81static char *rundir;
52f740b8 82static int lockfile_fd = -1;
3bd1e081 83
0bb7724a
DG
84/* Set to 1 when a SIGUSR1 signal is received. */
85static int recv_child_signal;
86
a23ec3a7
DG
87/*
88 * Consumer daemon specific control data. Every value not initialized here is
89 * set to 0 by the static definition.
90 */
3bd1e081
MD
91static struct consumer_data kconsumer_data = {
92 .type = LTTNG_CONSUMER_KERNEL,
60922cb0
DG
93 .err_unix_sock_path = DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
94 .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
03550b58
MD
95 .err_sock = -1,
96 .cmd_sock = -1,
173af62f
DG
97 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
98 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
99 .cond = PTHREAD_COND_INITIALIZER,
100 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081 101};
7753dea8
MD
102static struct consumer_data ustconsumer64_data = {
103 .type = LTTNG_CONSUMER64_UST,
60922cb0
DG
104 .err_unix_sock_path = DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
105 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
03550b58
MD
106 .err_sock = -1,
107 .cmd_sock = -1,
173af62f
DG
108 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
109 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
110 .cond = PTHREAD_COND_INITIALIZER,
111 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
7753dea8
MD
112};
113static struct consumer_data ustconsumer32_data = {
114 .type = LTTNG_CONSUMER32_UST,
60922cb0
DG
115 .err_unix_sock_path = DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
116 .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
03550b58
MD
117 .err_sock = -1,
118 .cmd_sock = -1,
173af62f
DG
119 .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
120 .lock = PTHREAD_MUTEX_INITIALIZER,
a23ec3a7
DG
121 .cond = PTHREAD_COND_INITIALIZER,
122 .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
3bd1e081
MD
123};
124
26c9d55e 125/* Shared between threads */
099e26bd 126static int dispatch_thread_exit;
fac6795d 127
54d01ffb
DG
128/* Global application Unix socket path */
129static char apps_unix_sock_path[PATH_MAX];
130/* Global client Unix socket path */
131static char client_unix_sock_path[PATH_MAX];
54d01ffb
DG
132/* global wait shm path for UST */
133static char wait_shm_path[PATH_MAX];
44a5e5eb
DG
134/* Global health check unix path */
135static char health_unix_sock_path[PATH_MAX];
fac6795d 136
1d4b027a 137/* Sockets and FDs */
a4b35e07
MD
138static int client_sock = -1;
139static int apps_sock = -1;
2f77fc4b 140int kernel_tracer_fd = -1;
76d7553f 141static int kernel_poll_pipe[2] = { -1, -1 };
1d4b027a 142
273ea72c
DG
143/*
144 * Quit pipe for all threads. This permits a single cancellation point
145 * for all threads when receiving an event on the pipe.
146 */
76d7553f 147static int thread_quit_pipe[2] = { -1, -1 };
273ea72c 148
099e26bd
DG
149/*
150 * This pipe is used to inform the thread managing application communication
151 * that a command is queued and ready to be processed.
152 */
76d7553f 153static int apps_cmd_pipe[2] = { -1, -1 };
099e26bd 154
d0b96690
DG
155int apps_cmd_notify_pipe[2] = { -1, -1 };
156
1d4b027a 157/* Pthread, Mutexes and Semaphores */
1d4b027a 158static pthread_t apps_thread;
d0b96690 159static pthread_t apps_notify_thread;
099e26bd 160static pthread_t reg_apps_thread;
1d4b027a 161static pthread_t client_thread;
7a485870 162static pthread_t kernel_thread;
099e26bd 163static pthread_t dispatch_thread;
44a5e5eb 164static pthread_t health_thread;
0b2dc8df 165static pthread_t ht_cleanup_thread;
4d076222 166static pthread_t jul_reg_thread;
5eb91c98 167
099e26bd
DG
168/*
169 * UST registration command queue. This queue is tied with a futex and uses a N
170 * wakers / 1 waiter implemented and detailed in futex.c/.h
171 *
172 * The thread_manage_apps and thread_dispatch_ust_registration interact with
173 * this queue and the wait/wake scheme.
174 */
175static struct ust_cmd_queue ust_cmd_queue;
176
b5541356
DG
177/*
178 * Pointer initialized before thread creation.
179 *
180 * This points to the tracing session list containing the session count and a
181 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
182 * MUST NOT be taken if you call a public function in session.c.
04ea676f 183 *
d063d709 184 * The lock is nested inside the structure: session_list_ptr->lock. Please use
54d01ffb 185 * session_lock_list and session_unlock_list for lock acquisition.
b5541356
DG
186 */
187static struct ltt_session_list *session_list_ptr;
188
7753dea8
MD
189int ust_consumerd64_fd = -1;
190int ust_consumerd32_fd = -1;
191
fb6f1fa2
YB
192static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN;
193static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN;
194static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR;
195static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR;
fb09408a 196
2f77fc4b
DG
197static const char *module_proc_lttng = "/proc/lttng";
198
5c827ce0
DG
199/*
200 * Consumer daemon state which is changed when spawning it, killing it or in
201 * case of a fatal error.
202 */
203enum consumerd_state {
204 CONSUMER_STARTED = 1,
205 CONSUMER_STOPPED = 2,
206 CONSUMER_ERROR = 3,
207};
208
209/*
210 * This consumer daemon state is used to validate if a client command will be
211 * able to reach the consumer. If not, the client is informed. For instance,
212 * doing a "lttng start" when the consumer state is set to ERROR will return an
213 * error to the client.
214 *
215 * The following example shows a possible race condition of this scheme:
216 *
217 * consumer thread error happens
218 * client cmd arrives
219 * client cmd checks state -> still OK
220 * consumer thread exit, sets error
221 * client cmd try to talk to consumer
222 * ...
223 *
224 * However, since the consumer is a different daemon, we have no way of making
225 * sure the command will reach it safely even with this state flag. This is why
226 * we consider that up to the state validation during command processing, the
227 * command is safe. After that, we can not guarantee the correctness of the
228 * client request vis-a-vis the consumer.
229 */
230static enum consumerd_state ust_consumerd_state;
231static enum consumerd_state kernel_consumerd_state;
232
ae9e45b3
DG
233/*
234 * Socket timeout for receiving and sending in seconds.
235 */
236static int app_socket_timeout;
237
12744796
DG
238/* Set in main() with the current page size. */
239long page_size;
240
8782cc74
MD
241/* Application health monitoring */
242struct health_app *health_sessiond;
243
4d076222
DG
244/* JUL TCP port for registration. Used by the JUL thread. */
245unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT;
246
59cec096
DG
247/* Am I root or not. */
248int is_root; /* Set to 1 if the daemon is running as root */
249
56045742
MD
250/*
251 * Whether sessiond is ready for commands/health check requests.
252 * NR_LTTNG_SESSIOND_READY must match the number of calls to
253 * lttng_sessiond_notify_ready().
254 */
255#define NR_LTTNG_SESSIOND_READY 2
256int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
257
258/* Notify parents that we are ready for cmd and health check */
259static
260void lttng_sessiond_notify_ready(void)
261{
262 if (uatomic_sub_return(&lttng_sessiond_ready, 1) == 0) {
263 /*
264 * Notify parent pid that we are ready to accept command
265 * for client side. This ppid is the one from the
266 * external process that spawned us.
267 */
268 if (opt_sig_parent) {
269 kill(ppid, SIGUSR1);
270 }
271
272 /*
273 * Notify the parent of the fork() process that we are
274 * ready.
275 */
929de64e 276 if (opt_daemon || opt_background) {
56045742
MD
277 kill(child_ppid, SIGUSR1);
278 }
279 }
280}
281
fb09408a 282static
7753dea8 283void setup_consumerd_path(void)
fb09408a 284{
fc7a59ce 285 const char *bin, *libdir;
fb09408a 286
7753dea8
MD
287 /*
288 * Allow INSTALL_BIN_PATH to be used as a target path for the
ebaeda94
MD
289 * native architecture size consumer if CONFIG_CONSUMER*_PATH
290 * has not been defined.
7753dea8 291 */
ebaeda94 292#if (CAA_BITS_PER_LONG == 32)
fc7a59ce
AM
293 if (!consumerd32_bin[0]) {
294 consumerd32_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
ebaeda94
MD
295 }
296 if (!consumerd32_libdir[0]) {
297 consumerd32_libdir = INSTALL_LIB_PATH;
298 }
299#elif (CAA_BITS_PER_LONG == 64)
fc7a59ce
AM
300 if (!consumerd64_bin[0]) {
301 consumerd64_bin = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
7753dea8 302 }
ebaeda94
MD
303 if (!consumerd64_libdir[0]) {
304 consumerd64_libdir = INSTALL_LIB_PATH;
7753dea8
MD
305 }
306#else
307#error "Unknown bitness"
308#endif
309
fb09408a
MD
310 /*
311 * runtime env. var. overrides the build default.
312 */
fc7a59ce
AM
313 bin = getenv("LTTNG_CONSUMERD32_BIN");
314 if (bin) {
315 consumerd32_bin = bin;
7753dea8 316 }
fc7a59ce
AM
317 bin = getenv("LTTNG_CONSUMERD64_BIN");
318 if (bin) {
319 consumerd64_bin = bin;
ebaeda94 320 }
72f579ee 321 libdir = getenv("LTTNG_CONSUMERD32_LIBDIR");
ebaeda94
MD
322 if (libdir) {
323 consumerd32_libdir = libdir;
324 }
72f579ee 325 libdir = getenv("LTTNG_CONSUMERD64_LIBDIR");
ebaeda94
MD
326 if (libdir) {
327 consumerd64_libdir = libdir;
fb09408a
MD
328 }
329}
330
5eb91c98
DG
331/*
332 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
333 */
d0b96690 334int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
5eb91c98
DG
335{
336 int ret;
337
d0b96690 338 assert(events);
5eb91c98
DG
339
340 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
341 if (ret < 0) {
342 goto error;
343 }
344
345 /* Add quit pipe */
d0b96690 346 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
5eb91c98
DG
347 if (ret < 0) {
348 goto error;
349 }
350
351 return 0;
352
353error:
354 return ret;
355}
356
357/*
358 * Check if the thread quit pipe was triggered.
359 *
360 * Return 1 if it was triggered else 0;
361 */
d0b96690 362int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
5eb91c98
DG
363{
364 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
365 return 1;
366 }
367
368 return 0;
369}
370
273ea72c 371/*
5eb91c98 372 * Init thread quit pipe.
273ea72c
DG
373 *
374 * Return -1 on error or 0 if all pipes are created.
375 */
376static int init_thread_quit_pipe(void)
377{
730389d9 378 int ret, i;
273ea72c 379
730389d9 380 ret = pipe(thread_quit_pipe);
273ea72c 381 if (ret < 0) {
730389d9 382 PERROR("thread quit pipe");
273ea72c
DG
383 goto error;
384 }
385
730389d9
DG
386 for (i = 0; i < 2; i++) {
387 ret = fcntl(thread_quit_pipe[i], F_SETFD, FD_CLOEXEC);
388 if (ret < 0) {
389 PERROR("fcntl");
390 goto error;
391 }
392 }
393
273ea72c
DG
394error:
395 return ret;
396}
397
099e26bd
DG
398/*
399 * Stop all threads by closing the thread quit pipe.
400 */
cf3af59e
MD
401static void stop_threads(void)
402{
5eb91c98
DG
403 int ret;
404
cf3af59e
MD
405 /* Stopping all threads */
406 DBG("Terminating all threads");
54d01ffb 407 ret = notify_thread_pipe(thread_quit_pipe[1]);
5eb91c98
DG
408 if (ret < 0) {
409 ERR("write error on thread quit pipe");
410 }
411
099e26bd 412 /* Dispatch thread */
26c9d55e 413 CMM_STORE_SHARED(dispatch_thread_exit, 1);
099e26bd 414 futex_nto1_wake(&ust_cmd_queue.futex);
cf3af59e
MD
415}
416
e975f9f8
DG
417/*
418 * Close every consumer sockets.
419 */
420static void close_consumer_sockets(void)
421{
422 int ret;
423
424 if (kconsumer_data.err_sock >= 0) {
425 ret = close(kconsumer_data.err_sock);
426 if (ret < 0) {
427 PERROR("kernel consumer err_sock close");
428 }
429 }
430 if (ustconsumer32_data.err_sock >= 0) {
431 ret = close(ustconsumer32_data.err_sock);
432 if (ret < 0) {
a76cbd9f 433 PERROR("UST consumerd32 err_sock close");
e975f9f8
DG
434 }
435 }
436 if (ustconsumer64_data.err_sock >= 0) {
437 ret = close(ustconsumer64_data.err_sock);
438 if (ret < 0) {
a76cbd9f 439 PERROR("UST consumerd64 err_sock close");
e975f9f8
DG
440 }
441 }
442 if (kconsumer_data.cmd_sock >= 0) {
443 ret = close(kconsumer_data.cmd_sock);
444 if (ret < 0) {
445 PERROR("kernel consumer cmd_sock close");
446 }
447 }
448 if (ustconsumer32_data.cmd_sock >= 0) {
449 ret = close(ustconsumer32_data.cmd_sock);
450 if (ret < 0) {
a76cbd9f 451 PERROR("UST consumerd32 cmd_sock close");
e975f9f8
DG
452 }
453 }
454 if (ustconsumer64_data.cmd_sock >= 0) {
455 ret = close(ustconsumer64_data.cmd_sock);
456 if (ret < 0) {
a76cbd9f 457 PERROR("UST consumerd64 cmd_sock close");
e975f9f8
DG
458 }
459 }
460}
461
52f740b8
JG
462/*
463 * Generate the full lock file path using the rundir.
464 *
465 * Return the snprintf() return value thus a negative value is an error.
466 */
467static int generate_lock_file_path(char *path, size_t len)
468{
469 int ret;
470
471 assert(path);
472 assert(rundir);
473
474 /* Build lockfile path from rundir. */
475 ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir);
476 if (ret < 0) {
477 PERROR("snprintf lockfile path");
478 }
479
480 return ret;
481}
482
fac6795d 483/*
d063d709 484 * Cleanup the daemon
fac6795d 485 */
cf3af59e 486static void cleanup(void)
fac6795d 487{
ef599319 488 int ret;
af9737e9 489 struct ltt_session *sess, *stmp;
8c6c56c2 490 char path[PATH_MAX];
fac6795d 491
1d4b027a 492 DBG("Cleaning up");
e07ae692 493
4e449f3f
MD
494 /*
495 * Close the thread quit pipe. It has already done its job,
496 * since we are now called.
497 */
2f77fc4b
DG
498 utils_close_pipe(thread_quit_pipe);
499
35f90c40
DG
500 /*
501 * If opt_pidfile is undefined, the default file will be wiped when
502 * removing the rundir.
503 */
504 if (opt_pidfile) {
505 ret = remove(opt_pidfile);
506 if (ret < 0) {
507 PERROR("remove pidfile %s", opt_pidfile);
508 }
509 }
510
8c6c56c2
MD
511 DBG("Removing sessiond and consumerd content of directory %s", rundir);
512
513 /* sessiond */
514 snprintf(path, PATH_MAX,
515 "%s/%s",
516 rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE);
517 DBG("Removing %s", path);
518 (void) unlink(path);
519
cd9290dd
DG
520 snprintf(path, PATH_MAX, "%s/%s", rundir,
521 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE);
522 DBG("Removing %s", path);
523 (void) unlink(path);
524
8c6c56c2
MD
525 /* kconsumerd */
526 snprintf(path, PATH_MAX,
527 DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
528 rundir);
529 DBG("Removing %s", path);
530 (void) unlink(path);
531
532 snprintf(path, PATH_MAX,
533 DEFAULT_KCONSUMERD_PATH,
534 rundir);
535 DBG("Removing directory %s", path);
536 (void) rmdir(path);
537
538 /* ust consumerd 32 */
539 snprintf(path, PATH_MAX,
540 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
541 rundir);
542 DBG("Removing %s", path);
543 (void) unlink(path);
544
545 snprintf(path, PATH_MAX,
546 DEFAULT_USTCONSUMERD32_PATH,
547 rundir);
548 DBG("Removing directory %s", path);
549 (void) rmdir(path);
550
551 /* ust consumerd 64 */
552 snprintf(path, PATH_MAX,
553 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
554 rundir);
555 DBG("Removing %s", path);
556 (void) unlink(path);
557
558 snprintf(path, PATH_MAX,
559 DEFAULT_USTCONSUMERD64_PATH,
560 rundir);
561 DBG("Removing directory %s", path);
562 (void) rmdir(path);
5461b305 563
99bab54f 564 DBG("Cleaning up all sessions");
fac6795d 565
b5541356 566 /* Destroy session list mutex */
273ea72c
DG
567 if (session_list_ptr != NULL) {
568 pthread_mutex_destroy(&session_list_ptr->lock);
569
570 /* Cleanup ALL session */
54d01ffb
DG
571 cds_list_for_each_entry_safe(sess, stmp,
572 &session_list_ptr->head, list) {
2f77fc4b 573 cmd_destroy_session(sess, kernel_poll_pipe[1]);
273ea72c
DG
574 }
575 }
576
099e26bd 577 DBG("Closing all UST sockets");
56fff090 578 ust_app_clean_list();
7972aab2 579 buffer_reg_destroy_registries();
099e26bd 580
4fba7219
DG
581 if (is_root && !opt_no_kernel) {
582 DBG2("Closing kernel fd");
a4b35e07 583 if (kernel_tracer_fd >= 0) {
76d7553f
MD
584 ret = close(kernel_tracer_fd);
585 if (ret) {
586 PERROR("close");
587 }
a4b35e07 588 }
2f50c8a3 589 DBG("Unloading kernel modules");
096102bd 590 modprobe_remove_lttng_all();
2f50c8a3 591 }
2f77fc4b 592
e975f9f8
DG
593 close_consumer_sockets();
594
52f740b8
JG
595
596 /*
597 * Cleanup lock file by deleting it and finaly closing it which will
598 * release the file system lock.
599 */
600 if (lockfile_fd >= 0) {
601 char lockfile_path[PATH_MAX];
602
603 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
604 if (ret > 0) {
605 ret = remove(lockfile_path);
606 if (ret < 0) {
607 PERROR("remove lock file");
608 }
609 ret = close(lockfile_fd);
610 if (ret < 0) {
611 PERROR("close lock file");
612 }
613 }
614 }
615
616 /*
617 * We do NOT rmdir rundir because there are other processes
618 * using it, for instance lttng-relayd, which can start in
619 * parallel with this teardown.
620 */
621
622 free(rundir);
623
421cb601 624 /* <fun> */
f56a39af 625 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
421cb601
DG
626 "Matthew, BEET driven development works!%c[%dm",
627 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
628 /* </fun> */
fac6795d
DG
629}
630
e065084a 631/*
d063d709 632 * Send data on a unix socket using the liblttsessiondcomm API.
e065084a 633 *
d063d709 634 * Return lttcomm error code.
e065084a
DG
635 */
636static int send_unix_sock(int sock, void *buf, size_t len)
637{
638 /* Check valid length */
c617c0c6 639 if (len == 0) {
e065084a
DG
640 return -1;
641 }
642
643 return lttcomm_send_unix_sock(sock, buf, len);
644}
645
5461b305 646/*
d063d709 647 * Free memory of a command context structure.
5461b305 648 */
a2fb29a5 649static void clean_command_ctx(struct command_ctx **cmd_ctx)
5461b305 650{
a2fb29a5
DG
651 DBG("Clean command context structure");
652 if (*cmd_ctx) {
653 if ((*cmd_ctx)->llm) {
654 free((*cmd_ctx)->llm);
5461b305 655 }
a2fb29a5
DG
656 if ((*cmd_ctx)->lsm) {
657 free((*cmd_ctx)->lsm);
5461b305 658 }
a2fb29a5
DG
659 free(*cmd_ctx);
660 *cmd_ctx = NULL;
5461b305
DG
661 }
662}
663
fac6795d 664/*
0fdd1e2c 665 * Notify UST applications using the shm mmap futex.
fac6795d 666 */
0fdd1e2c 667static int notify_ust_apps(int active)
fac6795d 668{
0fdd1e2c 669 char *wait_shm_mmap;
fac6795d 670
0fdd1e2c 671 DBG("Notifying applications of session daemon state: %d", active);
e07ae692 672
0fdd1e2c
DG
673 /* See shm.c for this call implying mmap, shm and futex calls */
674 wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root);
675 if (wait_shm_mmap == NULL) {
fac6795d
DG
676 goto error;
677 }
678
0fdd1e2c
DG
679 /* Wake waiting process */
680 futex_wait_update((int32_t *) wait_shm_mmap, active);
681
682 /* Apps notified successfully */
683 return 0;
fac6795d
DG
684
685error:
0fdd1e2c 686 return -1;
fac6795d
DG
687}
688
e065084a 689/*
d063d709
DG
690 * Setup the outgoing data buffer for the response (llm) by allocating the
691 * right amount of memory and copying the original information from the lsm
692 * structure.
ca95a216 693 *
d063d709 694 * Return total size of the buffer pointed by buf.
ca95a216 695 */
5461b305 696static int setup_lttng_msg(struct command_ctx *cmd_ctx, size_t size)
ca95a216 697{
f3ed775e 698 int ret, buf_size;
ca95a216 699
f3ed775e 700 buf_size = size;
5461b305 701
ba7f0ae5 702 cmd_ctx->llm = zmalloc(sizeof(struct lttcomm_lttng_msg) + buf_size);
5461b305 703 if (cmd_ctx->llm == NULL) {
76d7553f 704 PERROR("zmalloc");
5461b305 705 ret = -ENOMEM;
ca95a216
DG
706 goto error;
707 }
708
5461b305
DG
709 /* Copy common data */
710 cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
9f19cc17 711 cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
5461b305 712
5461b305
DG
713 cmd_ctx->llm->data_size = size;
714 cmd_ctx->lttng_msg_size = sizeof(struct lttcomm_lttng_msg) + buf_size;
715
ca95a216
DG
716 return buf_size;
717
718error:
719 return ret;
720}
721
7a485870 722/*
5eb91c98 723 * Update the kernel poll set of all channel fd available over all tracing
d063d709 724 * session. Add the wakeup pipe at the end of the set.
7a485870 725 */
5eb91c98 726static int update_kernel_poll(struct lttng_poll_event *events)
7a485870 727{
5eb91c98 728 int ret;
7a485870
DG
729 struct ltt_session *session;
730 struct ltt_kernel_channel *channel;
731
5eb91c98 732 DBG("Updating kernel poll set");
7a485870 733
54d01ffb 734 session_lock_list();
b5541356 735 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 736 session_lock(session);
7a485870 737 if (session->kernel_session == NULL) {
54d01ffb 738 session_unlock(session);
7a485870
DG
739 continue;
740 }
7a485870 741
54d01ffb
DG
742 cds_list_for_each_entry(channel,
743 &session->kernel_session->channel_list.head, list) {
5eb91c98
DG
744 /* Add channel fd to the kernel poll set */
745 ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
746 if (ret < 0) {
54d01ffb 747 session_unlock(session);
5eb91c98
DG
748 goto error;
749 }
750 DBG("Channel fd %d added to kernel set", channel->fd);
7a485870 751 }
54d01ffb 752 session_unlock(session);
7a485870 753 }
54d01ffb 754 session_unlock_list();
7a485870 755
5eb91c98 756 return 0;
7a485870
DG
757
758error:
54d01ffb 759 session_unlock_list();
7a485870
DG
760 return -1;
761}
762
763/*
54d01ffb 764 * Find the channel fd from 'fd' over all tracing session. When found, check
d063d709 765 * for new channel stream and send those stream fds to the kernel consumer.
7a485870 766 *
d063d709 767 * Useful for CPU hotplug feature.
7a485870 768 */
2bdd86d4 769static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
7a485870
DG
770{
771 int ret = 0;
772 struct ltt_session *session;
173af62f 773 struct ltt_kernel_session *ksess;
7a485870
DG
774 struct ltt_kernel_channel *channel;
775
776 DBG("Updating kernel streams for channel fd %d", fd);
777
54d01ffb 778 session_lock_list();
b5541356 779 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
54d01ffb 780 session_lock(session);
7a485870 781 if (session->kernel_session == NULL) {
54d01ffb 782 session_unlock(session);
7a485870
DG
783 continue;
784 }
173af62f 785 ksess = session->kernel_session;
d9800920 786
173af62f 787 cds_list_for_each_entry(channel, &ksess->channel_list.head, list) {
7a485870
DG
788 if (channel->fd == fd) {
789 DBG("Channel found, updating kernel streams");
790 ret = kernel_open_channel_stream(channel);
791 if (ret < 0) {
b3c750d2 792 goto error;
7a485870 793 }
5c786ded
JD
794 /* Update the stream global counter */
795 ksess->stream_count_global += ret;
d9800920 796
7a485870 797 /*
5eb91c98
DG
798 * Have we already sent fds to the consumer? If yes, it means
799 * that tracing is started so it is safe to send our updated
800 * stream fds.
7a485870 801 */
173af62f
DG
802 if (ksess->consumer_fds_sent == 1 && ksess->consumer != NULL) {
803 struct lttng_ht_iter iter;
804 struct consumer_socket *socket;
805
e7fe706f 806 rcu_read_lock();
173af62f
DG
807 cds_lfht_for_each_entry(ksess->consumer->socks->ht,
808 &iter.iter, socket, node.node) {
173af62f 809 pthread_mutex_lock(socket->lock);
f50f23d9 810 ret = kernel_consumer_send_channel_stream(socket,
2bba9e53
DG
811 channel, ksess,
812 session->output_traces ? 1 : 0);
173af62f
DG
813 pthread_mutex_unlock(socket->lock);
814 if (ret < 0) {
e7fe706f 815 rcu_read_unlock();
173af62f
DG
816 goto error;
817 }
7a485870 818 }
e7fe706f 819 rcu_read_unlock();
7a485870 820 }
b3c750d2 821 goto error;
7a485870
DG
822 }
823 }
54d01ffb 824 session_unlock(session);
7a485870 825 }
54d01ffb 826 session_unlock_list();
b3c750d2 827 return ret;
7a485870 828
b3c750d2 829error:
54d01ffb
DG
830 session_unlock(session);
831 session_unlock_list();
7a485870
DG
832 return ret;
833}
834
487cf67c 835/*
ffe60014
DG
836 * For each tracing session, update newly registered apps. The session list
837 * lock MUST be acquired before calling this.
487cf67c
DG
838 */
839static void update_ust_app(int app_sock)
840{
841 struct ltt_session *sess, *stmp;
842
fdadac08
DG
843 /* Consumer is in an ERROR state. Stop any application update. */
844 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
845 /* Stop the update process since the consumer is dead. */
846 return;
847 }
848
487cf67c
DG
849 /* For all tracing session(s) */
850 cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
4ee14516 851 session_lock(sess);
421cb601
DG
852 if (sess->ust_session) {
853 ust_app_global_update(sess->ust_session, app_sock);
854 }
4ee14516 855 session_unlock(sess);
487cf67c
DG
856 }
857}
858
7a485870 859/*
d063d709 860 * This thread manage event coming from the kernel.
7a485870 861 *
d063d709
DG
862 * Features supported in this thread:
863 * -) CPU Hotplug
7a485870
DG
864 */
865static void *thread_manage_kernel(void *data)
866{
139ac872 867 int ret, i, pollfd, update_poll_flag = 1, err = -1;
5eb91c98 868 uint32_t revents, nb_fd;
7a485870 869 char tmp;
5eb91c98 870 struct lttng_poll_event events;
7a485870 871
6993eeb3 872 DBG("[thread] Thread manage kernel started");
7a485870 873
6c71277b 874 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
927ca06a 875
d5d63bf1
DG
876 /*
877 * This first step of the while is to clean this structure which could free
6d737ce4 878 * non NULL pointers so initialize it before the loop.
d5d63bf1 879 */
6d737ce4 880 lttng_poll_init(&events);
d5d63bf1 881
71ddb4ed 882 if (testpoint(sessiond_thread_manage_kernel)) {
6993eeb3
CB
883 goto error_testpoint;
884 }
8ac94142 885
840cb59c 886 health_code_update();
44a5e5eb 887
71ddb4ed 888 if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
d21b0d71 889 goto error_testpoint;
6993eeb3
CB
890 }
891
7a485870 892 while (1) {
840cb59c 893 health_code_update();
44a5e5eb 894
7a485870 895 if (update_poll_flag == 1) {
d21b0d71
DG
896 /* Clean events object. We are about to populate it again. */
897 lttng_poll_clean(&events);
898
d0b96690 899 ret = sessiond_set_thread_pollset(&events, 2);
d21b0d71
DG
900 if (ret < 0) {
901 goto error_poll_create;
902 }
903
904 ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN);
905 if (ret < 0) {
906 goto error;
907 }
5f822d0a 908
d21b0d71 909 /* This will add the available kernel channel if any. */
5eb91c98
DG
910 ret = update_kernel_poll(&events);
911 if (ret < 0) {
7a485870
DG
912 goto error;
913 }
914 update_poll_flag = 0;
915 }
916
ec4c7666 917 DBG("Thread kernel polling");
7a485870
DG
918
919 /* Poll infinite value of time */
88f2b785 920 restart:
a78af745 921 health_poll_entry();
5eb91c98 922 ret = lttng_poll_wait(&events, -1);
ec4c7666
MD
923 DBG("Thread kernel return from poll on %d fds",
924 LTTNG_POLL_GETNB(&events));
a78af745 925 health_poll_exit();
7a485870 926 if (ret < 0) {
88f2b785
MD
927 /*
928 * Restart interrupted system call.
929 */
930 if (errno == EINTR) {
931 goto restart;
932 }
7a485870
DG
933 goto error;
934 } else if (ret == 0) {
935 /* Should not happen since timeout is infinite */
85611738
DG
936 ERR("Return value of poll is 0 with an infinite timeout.\n"
937 "This should not have happened! Continuing...");
7a485870
DG
938 continue;
939 }
940
0d9c5d77
DG
941 nb_fd = ret;
942
5eb91c98
DG
943 for (i = 0; i < nb_fd; i++) {
944 /* Fetch once the poll data */
945 revents = LTTNG_POLL_GETEV(&events, i);
946 pollfd = LTTNG_POLL_GETFD(&events, i);
7a485870 947
840cb59c 948 health_code_update();
44a5e5eb 949
f0567343
MD
950 if (!revents) {
951 /* No activity for this FD (poll implementation). */
952 continue;
953 }
954
5eb91c98 955 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 956 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 957 if (ret) {
139ac872
MD
958 err = 0;
959 goto exit;
5eb91c98 960 }
7a485870 961
5eb91c98
DG
962 /* Check for data on kernel pipe */
963 if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) {
6cd525e8
MD
964 (void) lttng_read(kernel_poll_pipe[0],
965 &tmp, 1);
f921c78f
DG
966 /*
967 * Ret value is useless here, if this pipe gets any actions an
968 * update is required anyway.
969 */
5eb91c98
DG
970 update_poll_flag = 1;
971 continue;
972 } else {
973 /*
974 * New CPU detected by the kernel. Adding kernel stream to
975 * kernel session and updating the kernel consumer
976 */
977 if (revents & LPOLLIN) {
2bdd86d4 978 ret = update_kernel_stream(&kconsumer_data, pollfd);
5eb91c98
DG
979 if (ret < 0) {
980 continue;
981 }
982 break;
983 /*
984 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
985 * and unregister kernel stream at this point.
986 */
7a485870 987 }
7a485870
DG
988 }
989 }
990 }
991
139ac872 992exit:
7a485870 993error:
5eb91c98 994 lttng_poll_clean(&events);
76d7553f 995error_poll_create:
6993eeb3 996error_testpoint:
6620da75
DG
997 utils_close_pipe(kernel_poll_pipe);
998 kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1;
139ac872 999 if (err) {
840cb59c 1000 health_error();
139ac872 1001 ERR("Health error occurred in %s", __func__);
6620da75
DG
1002 WARN("Kernel thread died unexpectedly. "
1003 "Kernel tracing can continue but CPU hotplug is disabled.");
139ac872 1004 }
8782cc74 1005 health_unregister(health_sessiond);
76d7553f 1006 DBG("Kernel thread dying");
7a485870
DG
1007 return NULL;
1008}
1009
a23ec3a7
DG
1010/*
1011 * Signal pthread condition of the consumer data that the thread.
1012 */
1013static void signal_consumer_condition(struct consumer_data *data, int state)
1014{
1015 pthread_mutex_lock(&data->cond_mutex);
1016
1017 /*
1018 * The state is set before signaling. It can be any value, it's the waiter
1019 * job to correctly interpret this condition variable associated to the
1020 * consumer pthread_cond.
1021 *
1022 * A value of 0 means that the corresponding thread of the consumer data
1023 * was not started. 1 indicates that the thread has started and is ready
1024 * for action. A negative value means that there was an error during the
1025 * thread bootstrap.
1026 */
1027 data->consumer_thread_is_ready = state;
1028 (void) pthread_cond_signal(&data->cond);
1029
1030 pthread_mutex_unlock(&data->cond_mutex);
1031}
1032
1d4b027a 1033/*
3bd1e081 1034 * This thread manage the consumer error sent back to the session daemon.
1d4b027a 1035 */
3bd1e081 1036static void *thread_manage_consumer(void *data)
1d4b027a 1037{
ef1917cf 1038 int sock = -1, i, ret, pollfd, err = -1, should_quit = 0;
5eb91c98 1039 uint32_t revents, nb_fd;
1d4b027a 1040 enum lttcomm_return_code code;
5eb91c98 1041 struct lttng_poll_event events;
3bd1e081 1042 struct consumer_data *consumer_data = data;
1d4b027a 1043
3bd1e081 1044 DBG("[thread] Manage consumer started");
1d4b027a 1045
6c71277b 1046 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER);
927ca06a 1047
855060f8 1048 health_code_update();
9449cc75 1049
5eb91c98 1050 /*
331744e3
JD
1051 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1052 * metadata_sock. Nothing more will be added to this poll set.
5eb91c98 1053 */
331744e3 1054 ret = sessiond_set_thread_pollset(&events, 3);
5eb91c98 1055 if (ret < 0) {
76d7553f 1056 goto error_poll;
5eb91c98 1057 }
273ea72c 1058
edb8b045
DG
1059 /*
1060 * The error socket here is already in a listening state which was done
1061 * just before spawning this thread to avoid a race between the consumer
1062 * daemon exec trying to connect and the listen() call.
1063 */
3bd1e081 1064 ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP);
5eb91c98
DG
1065 if (ret < 0) {
1066 goto error;
1067 }
1068
840cb59c 1069 health_code_update();
44a5e5eb 1070
331744e3 1071 /* Infinite blocking call, waiting for transmission */
88f2b785 1072restart:
a78af745 1073 health_poll_entry();
8ac94142 1074
71ddb4ed 1075 if (testpoint(sessiond_thread_manage_consumer)) {
6993eeb3
CB
1076 goto error;
1077 }
8ac94142 1078
5eb91c98 1079 ret = lttng_poll_wait(&events, -1);
a78af745 1080 health_poll_exit();
273ea72c 1081 if (ret < 0) {
88f2b785
MD
1082 /*
1083 * Restart interrupted system call.
1084 */
1085 if (errno == EINTR) {
1086 goto restart;
1087 }
273ea72c
DG
1088 goto error;
1089 }
1090
0d9c5d77
DG
1091 nb_fd = ret;
1092
5eb91c98
DG
1093 for (i = 0; i < nb_fd; i++) {
1094 /* Fetch once the poll data */
1095 revents = LTTNG_POLL_GETEV(&events, i);
1096 pollfd = LTTNG_POLL_GETFD(&events, i);
1097
840cb59c 1098 health_code_update();
44a5e5eb 1099
f0567343
MD
1100 if (!revents) {
1101 /* No activity for this FD (poll implementation). */
1102 continue;
1103 }
1104
5eb91c98 1105 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1106 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1107 if (ret) {
139ac872
MD
1108 err = 0;
1109 goto exit;
5eb91c98
DG
1110 }
1111
1112 /* Event on the registration socket */
3bd1e081 1113 if (pollfd == consumer_data->err_sock) {
5eb91c98 1114 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3bd1e081 1115 ERR("consumer err socket poll error");
5eb91c98
DG
1116 goto error;
1117 }
1118 }
273ea72c
DG
1119 }
1120
3bd1e081 1121 sock = lttcomm_accept_unix_sock(consumer_data->err_sock);
1d4b027a
DG
1122 if (sock < 0) {
1123 goto error;
1124 }
1125
b662582b
DG
1126 /*
1127 * Set the CLOEXEC flag. Return code is useless because either way, the
1128 * show must go on.
1129 */
1130 (void) utils_set_fd_cloexec(sock);
1131
840cb59c 1132 health_code_update();
44a5e5eb 1133
3bd1e081 1134 DBG2("Receiving code from consumer err_sock");
ee0b0061 1135
712ea556 1136 /* Getting status code from kconsumerd */
54d01ffb
DG
1137 ret = lttcomm_recv_unix_sock(sock, &code,
1138 sizeof(enum lttcomm_return_code));
1d4b027a
DG
1139 if (ret <= 0) {
1140 goto error;
1141 }
1142
840cb59c 1143 health_code_update();
f73fabfd 1144 if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) {
331744e3 1145 /* Connect both socket, command and metadata. */
3bd1e081
MD
1146 consumer_data->cmd_sock =
1147 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
4ce514c4 1148 consumer_data->metadata_fd =
331744e3 1149 lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
92db7cdc
DG
1150 if (consumer_data->cmd_sock < 0
1151 || consumer_data->metadata_fd < 0) {
331744e3 1152 PERROR("consumer connect cmd socket");
a23ec3a7
DG
1153 /* On error, signal condition and quit. */
1154 signal_consumer_condition(consumer_data, -1);
1d4b027a
DG
1155 goto error;
1156 }
9363801e 1157 consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd;
331744e3
JD
1158 /* Create metadata socket lock. */
1159 consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t));
1160 if (consumer_data->metadata_sock.lock == NULL) {
1161 PERROR("zmalloc pthread mutex");
1162 ret = -1;
1163 goto error;
1164 }
1165 pthread_mutex_init(consumer_data->metadata_sock.lock, NULL);
1166
a23ec3a7 1167 signal_consumer_condition(consumer_data, 1);
331744e3
JD
1168 DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock);
1169 DBG("Consumer metadata socket ready (fd: %d)",
4ce514c4 1170 consumer_data->metadata_fd);
1d4b027a 1171 } else {
3bd1e081 1172 ERR("consumer error when waiting for SOCK_READY : %s",
1d4b027a
DG
1173 lttcomm_get_readable_code(-code));
1174 goto error;
1175 }
1176
331744e3 1177 /* Remove the consumerd error sock since we've established a connexion */
3bd1e081 1178 ret = lttng_poll_del(&events, consumer_data->err_sock);
72079cae 1179 if (ret < 0) {
72079cae
DG
1180 goto error;
1181 }
1182
331744e3 1183 /* Add new accepted error socket. */
5eb91c98
DG
1184 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP);
1185 if (ret < 0) {
72079cae 1186 goto error;
5eb91c98
DG
1187 }
1188
331744e3 1189 /* Add metadata socket that is successfully connected. */
4ce514c4 1190 ret = lttng_poll_add(&events, consumer_data->metadata_fd,
331744e3
JD
1191 LPOLLIN | LPOLLRDHUP);
1192 if (ret < 0) {
1193 goto error;
1194 }
1195
840cb59c 1196 health_code_update();
44a5e5eb 1197
331744e3 1198 /* Infinite blocking call, waiting for transmission */
88f2b785 1199restart_poll:
331744e3 1200 while (1) {
ef1917cf
DG
1201 health_code_update();
1202
1203 /* Exit the thread because the thread quit pipe has been triggered. */
1204 if (should_quit) {
1205 /* Not a health error. */
1206 err = 0;
1207 goto exit;
1208 }
1209
331744e3
JD
1210 health_poll_entry();
1211 ret = lttng_poll_wait(&events, -1);
1212 health_poll_exit();
1213 if (ret < 0) {
1214 /*
1215 * Restart interrupted system call.
1216 */
1217 if (errno == EINTR) {
1218 goto restart_poll;
1219 }
1220 goto error;
88f2b785 1221 }
72079cae 1222
331744e3 1223 nb_fd = ret;
0d9c5d77 1224
331744e3
JD
1225 for (i = 0; i < nb_fd; i++) {
1226 /* Fetch once the poll data */
1227 revents = LTTNG_POLL_GETEV(&events, i);
1228 pollfd = LTTNG_POLL_GETFD(&events, i);
5eb91c98 1229
331744e3 1230 health_code_update();
44a5e5eb 1231
f0567343
MD
1232 if (!revents) {
1233 /* No activity for this FD (poll implementation). */
1234 continue;
1235 }
1236
ef1917cf
DG
1237 /*
1238 * Thread quit pipe has been triggered, flag that we should stop
1239 * but continue the current loop to handle potential data from
1240 * consumer.
1241 */
1242 should_quit = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1243
331744e3
JD
1244 if (pollfd == sock) {
1245 /* Event on the consumerd socket */
1246 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1247 ERR("consumer err socket second poll error");
1248 goto error;
1249 }
1250 health_code_update();
1251 /* Wait for any kconsumerd error */
1252 ret = lttcomm_recv_unix_sock(sock, &code,
1253 sizeof(enum lttcomm_return_code));
1254 if (ret <= 0) {
1255 ERR("consumer closed the command socket");
1256 goto error;
1257 }
1258
1259 ERR("consumer return code : %s",
1260 lttcomm_get_readable_code(-code));
1261
1262 goto exit;
4ce514c4 1263 } else if (pollfd == consumer_data->metadata_fd) {
331744e3
JD
1264 /* UST metadata requests */
1265 ret = ust_consumer_metadata_request(
1266 &consumer_data->metadata_sock);
1267 if (ret < 0) {
1268 ERR("Handling metadata request");
1269 goto error;
1270 }
5eb91c98 1271 }
ef1917cf 1272 /* No need for an else branch all FDs are tested prior. */
5eb91c98 1273 }
331744e3 1274 health_code_update();
5eb91c98
DG
1275 }
1276
139ac872 1277exit:
1d4b027a 1278error:
fdadac08
DG
1279 /*
1280 * We lock here because we are about to close the sockets and some other
92db7cdc
DG
1281 * thread might be using them so get exclusive access which will abort all
1282 * other consumer command by other threads.
fdadac08
DG
1283 */
1284 pthread_mutex_lock(&consumer_data->lock);
1285
5c827ce0
DG
1286 /* Immediately set the consumerd state to stopped */
1287 if (consumer_data->type == LTTNG_CONSUMER_KERNEL) {
1288 uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR);
1289 } else if (consumer_data->type == LTTNG_CONSUMER64_UST ||
1290 consumer_data->type == LTTNG_CONSUMER32_UST) {
1291 uatomic_set(&ust_consumerd_state, CONSUMER_ERROR);
1292 } else {
1293 /* Code flow error... */
1294 assert(0);
1295 }
1296
76d7553f
MD
1297 if (consumer_data->err_sock >= 0) {
1298 ret = close(consumer_data->err_sock);
1299 if (ret) {
1300 PERROR("close");
1301 }
a76cbd9f 1302 consumer_data->err_sock = -1;
76d7553f
MD
1303 }
1304 if (consumer_data->cmd_sock >= 0) {
1305 ret = close(consumer_data->cmd_sock);
1306 if (ret) {
1307 PERROR("close");
1308 }
a76cbd9f 1309 consumer_data->cmd_sock = -1;
76d7553f 1310 }
24a54a9b
SS
1311 if (consumer_data->metadata_sock.fd_ptr &&
1312 *consumer_data->metadata_sock.fd_ptr >= 0) {
9363801e 1313 ret = close(*consumer_data->metadata_sock.fd_ptr);
331744e3
JD
1314 if (ret) {
1315 PERROR("close");
1316 }
1317 }
76d7553f
MD
1318 if (sock >= 0) {
1319 ret = close(sock);
1320 if (ret) {
1321 PERROR("close");
1322 }
1323 }
273ea72c 1324
3bd1e081
MD
1325 unlink(consumer_data->err_unix_sock_path);
1326 unlink(consumer_data->cmd_unix_sock_path);
1327 consumer_data->pid = 0;
fdadac08 1328 pthread_mutex_unlock(&consumer_data->lock);
92db7cdc 1329
fdadac08 1330 /* Cleanup metadata socket mutex. */
24a54a9b
SS
1331 if (consumer_data->metadata_sock.lock) {
1332 pthread_mutex_destroy(consumer_data->metadata_sock.lock);
1333 free(consumer_data->metadata_sock.lock);
1334 }
5eb91c98 1335 lttng_poll_clean(&events);
76d7553f 1336error_poll:
139ac872 1337 if (err) {
840cb59c 1338 health_error();
139ac872
MD
1339 ERR("Health error occurred in %s", __func__);
1340 }
8782cc74 1341 health_unregister(health_sessiond);
76d7553f 1342 DBG("consumer thread cleanup completed");
0177d773 1343
5eb91c98 1344 return NULL;
099e26bd
DG
1345}
1346
099e26bd
DG
1347/*
1348 * This thread manage application communication.
1d4b027a
DG
1349 */
1350static void *thread_manage_apps(void *data)
099e26bd 1351{
139ac872 1352 int i, ret, pollfd, err = -1;
6cd525e8 1353 ssize_t size_ret;
5eb91c98 1354 uint32_t revents, nb_fd;
5eb91c98 1355 struct lttng_poll_event events;
099e26bd
DG
1356
1357 DBG("[thread] Manage application started");
1358
f6a9efaa
DG
1359 rcu_register_thread();
1360 rcu_thread_online();
1361
6c71277b 1362 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE);
927ca06a 1363
71ddb4ed 1364 if (testpoint(sessiond_thread_manage_apps)) {
6993eeb3
CB
1365 goto error_testpoint;
1366 }
1367
840cb59c 1368 health_code_update();
44a5e5eb 1369
d0b96690 1370 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1371 if (ret < 0) {
76d7553f 1372 goto error_poll_create;
5eb91c98 1373 }
099e26bd 1374
5eb91c98
DG
1375 ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1376 if (ret < 0) {
1377 goto error;
1378 }
099e26bd 1379
71ddb4ed 1380 if (testpoint(sessiond_thread_manage_apps_before_loop)) {
6993eeb3
CB
1381 goto error;
1382 }
8ac94142 1383
840cb59c 1384 health_code_update();
44a5e5eb 1385
5eb91c98 1386 while (1) {
ec4c7666 1387 DBG("Apps thread polling");
099e26bd
DG
1388
1389 /* Inifinite blocking call, waiting for transmission */
88f2b785 1390 restart:
a78af745 1391 health_poll_entry();
5eb91c98 1392 ret = lttng_poll_wait(&events, -1);
ec4c7666
MD
1393 DBG("Apps thread return from poll on %d fds",
1394 LTTNG_POLL_GETNB(&events));
a78af745 1395 health_poll_exit();
099e26bd 1396 if (ret < 0) {
88f2b785
MD
1397 /*
1398 * Restart interrupted system call.
1399 */
1400 if (errno == EINTR) {
1401 goto restart;
1402 }
099e26bd
DG
1403 goto error;
1404 }
1405
0d9c5d77
DG
1406 nb_fd = ret;
1407
5eb91c98
DG
1408 for (i = 0; i < nb_fd; i++) {
1409 /* Fetch once the poll data */
1410 revents = LTTNG_POLL_GETEV(&events, i);
1411 pollfd = LTTNG_POLL_GETFD(&events, i);
1412
840cb59c 1413 health_code_update();
44a5e5eb 1414
f0567343
MD
1415 if (!revents) {
1416 /* No activity for this FD (poll implementation). */
1417 continue;
1418 }
1419
5eb91c98 1420 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1421 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1422 if (ret) {
139ac872
MD
1423 err = 0;
1424 goto exit;
5eb91c98 1425 }
099e26bd 1426
5eb91c98
DG
1427 /* Inspect the apps cmd pipe */
1428 if (pollfd == apps_cmd_pipe[0]) {
1429 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1430 ERR("Apps command pipe error");
0177d773 1431 goto error;
5eb91c98 1432 } else if (revents & LPOLLIN) {
d0b96690
DG
1433 int sock;
1434
5eb91c98 1435 /* Empty pipe */
6cd525e8
MD
1436 size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock));
1437 if (size_ret < sizeof(sock)) {
76d7553f 1438 PERROR("read apps cmd pipe");
5eb91c98
DG
1439 goto error;
1440 }
099e26bd 1441
840cb59c 1442 health_code_update();
44a5e5eb 1443
ffe60014 1444 /*
d0b96690
DG
1445 * We only monitor the error events of the socket. This
1446 * thread does not handle any incoming data from UST
1447 * (POLLIN).
ffe60014 1448 */
d0b96690
DG
1449 ret = lttng_poll_add(&events, sock,
1450 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
1451 if (ret < 0) {
5eb91c98 1452 goto error;
e0c7ec2b 1453 }
acc7b41b 1454
d0b96690 1455 DBG("Apps with sock %d added to poll set", sock);
0177d773 1456 }
5eb91c98
DG
1457 } else {
1458 /*
54d01ffb
DG
1459 * At this point, we know that a registered application made
1460 * the event at poll_wait.
5eb91c98
DG
1461 */
1462 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1463 /* Removing from the poll set */
1464 ret = lttng_poll_del(&events, pollfd);
1465 if (ret < 0) {
1466 goto error;
1467 }
099e26bd 1468
b9d9b220 1469 /* Socket closed on remote end. */
56fff090 1470 ust_app_unregister(pollfd);
5eb91c98 1471 }
099e26bd 1472 }
44a5e5eb 1473
840cb59c 1474 health_code_update();
099e26bd 1475 }
099e26bd
DG
1476 }
1477
139ac872 1478exit:
099e26bd 1479error:
5eb91c98 1480 lttng_poll_clean(&events);
76d7553f 1481error_poll_create:
6993eeb3 1482error_testpoint:
6620da75
DG
1483 utils_close_pipe(apps_cmd_pipe);
1484 apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1;
1485
1486 /*
1487 * We don't clean the UST app hash table here since already registered
1488 * applications can still be controlled so let them be until the session
1489 * daemon dies or the applications stop.
1490 */
1491
139ac872 1492 if (err) {
840cb59c 1493 health_error();
139ac872
MD
1494 ERR("Health error occurred in %s", __func__);
1495 }
8782cc74 1496 health_unregister(health_sessiond);
76d7553f 1497 DBG("Application communication apps thread cleanup complete");
f6a9efaa
DG
1498 rcu_thread_offline();
1499 rcu_unregister_thread();
099e26bd
DG
1500 return NULL;
1501}
1502
d0b96690 1503/*
d88aee68
DG
1504 * Send a socket to a thread This is called from the dispatch UST registration
1505 * thread once all sockets are set for the application.
d0b96690 1506 *
b85dc84c
DG
1507 * The sock value can be invalid, we don't really care, the thread will handle
1508 * it and make the necessary cleanup if so.
1509 *
d0b96690
DG
1510 * On success, return 0 else a negative value being the errno message of the
1511 * write().
1512 */
d88aee68 1513static int send_socket_to_thread(int fd, int sock)
d0b96690 1514{
6cd525e8 1515 ssize_t ret;
d0b96690 1516
b85dc84c
DG
1517 /*
1518 * It's possible that the FD is set as invalid with -1 concurrently just
1519 * before calling this function being a shutdown state of the thread.
1520 */
1521 if (fd < 0) {
1522 ret = -EBADF;
1523 goto error;
1524 }
d0b96690 1525
6cd525e8
MD
1526 ret = lttng_write(fd, &sock, sizeof(sock));
1527 if (ret < sizeof(sock)) {
d88aee68 1528 PERROR("write apps pipe %d", fd);
d0b96690
DG
1529 if (ret < 0) {
1530 ret = -errno;
1531 }
1532 goto error;
1533 }
1534
1535 /* All good. Don't send back the write positive ret value. */
1536 ret = 0;
1537error:
6cd525e8 1538 return (int) ret;
d0b96690
DG
1539}
1540
f45e313d
DG
1541/*
1542 * Sanitize the wait queue of the dispatch registration thread meaning removing
1543 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1544 * notify socket is never received.
1545 */
1546static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
1547{
1548 int ret, nb_fd = 0, i;
1549 unsigned int fd_added = 0;
1550 struct lttng_poll_event events;
1551 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1552
1553 assert(wait_queue);
1554
1555 lttng_poll_init(&events);
1556
1557 /* Just skip everything for an empty queue. */
1558 if (!wait_queue->count) {
1559 goto end;
1560 }
1561
1562 ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
1563 if (ret < 0) {
1564 goto error_create;
1565 }
1566
1567 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1568 &wait_queue->head, head) {
1569 assert(wait_node->app);
1570 ret = lttng_poll_add(&events, wait_node->app->sock,
1571 LPOLLHUP | LPOLLERR);
1572 if (ret < 0) {
1573 goto error;
1574 }
1575
1576 fd_added = 1;
1577 }
1578
1579 if (!fd_added) {
1580 goto end;
1581 }
1582
1583 /*
1584 * Poll but don't block so we can quickly identify the faulty events and
1585 * clean them afterwards from the wait queue.
1586 */
1587 ret = lttng_poll_wait(&events, 0);
1588 if (ret < 0) {
1589 goto error;
1590 }
1591 nb_fd = ret;
1592
1593 for (i = 0; i < nb_fd; i++) {
1594 /* Get faulty FD. */
1595 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1596 int pollfd = LTTNG_POLL_GETFD(&events, i);
1597
f0567343
MD
1598 if (!revents) {
1599 /* No activity for this FD (poll implementation). */
1600 continue;
1601 }
1602
f45e313d
DG
1603 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
1604 &wait_queue->head, head) {
1605 if (pollfd == wait_node->app->sock &&
1606 (revents & (LPOLLHUP | LPOLLERR))) {
1607 cds_list_del(&wait_node->head);
1608 wait_queue->count--;
1609 ust_app_destroy(wait_node->app);
1610 free(wait_node);
1611 break;
1612 }
1613 }
1614 }
1615
1616 if (nb_fd > 0) {
1617 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
1618 }
1619
1620end:
1621 lttng_poll_clean(&events);
1622 return;
1623
1624error:
1625 lttng_poll_clean(&events);
1626error_create:
1627 ERR("Unable to sanitize wait queue");
1628 return;
1629}
1630
099e26bd
DG
1631/*
1632 * Dispatch request from the registration threads to the application
1633 * communication thread.
1634 */
1635static void *thread_dispatch_ust_registration(void *data)
1636{
12e2b881 1637 int ret, err = -1;
099e26bd
DG
1638 struct cds_wfq_node *node;
1639 struct ust_command *ust_cmd = NULL;
f45e313d
DG
1640 struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
1641 struct ust_reg_wait_queue wait_queue = {
1642 .count = 0,
1643 };
d0b96690 1644
6c71277b 1645 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);
12e2b881 1646
a09bf708
MD
1647 if (testpoint(sessiond_thread_app_reg_dispatch)) {
1648 goto error_testpoint;
1649 }
1650
12e2b881
MD
1651 health_code_update();
1652
f45e313d 1653 CDS_INIT_LIST_HEAD(&wait_queue.head);
099e26bd
DG
1654
1655 DBG("[thread] Dispatch UST command started");
1656
26c9d55e 1657 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
12e2b881
MD
1658 health_code_update();
1659
099e26bd
DG
1660 /* Atomically prepare the queue futex */
1661 futex_nto1_prepare(&ust_cmd_queue.futex);
1662
1663 do {
d0b96690 1664 struct ust_app *app = NULL;
7972aab2 1665 ust_cmd = NULL;
d0b96690 1666
f45e313d
DG
1667 /*
1668 * Make sure we don't have node(s) that have hung up before receiving
1669 * the notify socket. This is to clean the list in order to avoid
1670 * memory leaks from notify socket that are never seen.
1671 */
1672 sanitize_wait_queue(&wait_queue);
1673
12e2b881 1674 health_code_update();
099e26bd
DG
1675 /* Dequeue command for registration */
1676 node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
1677 if (node == NULL) {
00a17c97 1678 DBG("Woken up but nothing in the UST command queue");
099e26bd
DG
1679 /* Continue thread execution */
1680 break;
1681 }
1682
1683 ust_cmd = caa_container_of(node, struct ust_command, node);
1684
2f50c8a3
DG
1685 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1686 " gid:%d sock:%d name:%s (version %d.%d)",
1687 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
1688 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
1689 ust_cmd->sock, ust_cmd->reg_msg.name,
1690 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
d0b96690
DG
1691
1692 if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
1693 wait_node = zmalloc(sizeof(*wait_node));
1694 if (!wait_node) {
1695 PERROR("zmalloc wait_node dispatch");
020d7f60
DG
1696 ret = close(ust_cmd->sock);
1697 if (ret < 0) {
1698 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1699 }
51dec90d 1700 lttng_fd_put(LTTNG_FD_APPS, 1);
7972aab2 1701 free(ust_cmd);
d0b96690
DG
1702 goto error;
1703 }
1704 CDS_INIT_LIST_HEAD(&wait_node->head);
1705
1706 /* Create application object if socket is CMD. */
1707 wait_node->app = ust_app_create(&ust_cmd->reg_msg,
1708 ust_cmd->sock);
1709 if (!wait_node->app) {
1710 ret = close(ust_cmd->sock);
1711 if (ret < 0) {
1712 PERROR("close ust sock dispatch %d", ust_cmd->sock);
6620da75 1713 }
51dec90d 1714 lttng_fd_put(LTTNG_FD_APPS, 1);
d88aee68 1715 free(wait_node);
7972aab2 1716 free(ust_cmd);
d0b96690
DG
1717 continue;
1718 }
1719 /*
1720 * Add application to the wait queue so we can set the notify
1721 * socket before putting this object in the global ht.
1722 */
f45e313d
DG
1723 cds_list_add(&wait_node->head, &wait_queue.head);
1724 wait_queue.count++;
d0b96690 1725
7972aab2 1726 free(ust_cmd);
d0b96690
DG
1727 /*
1728 * We have to continue here since we don't have the notify
1729 * socket and the application MUST be added to the hash table
1730 * only at that moment.
1731 */
1732 continue;
1733 } else {
1734 /*
1735 * Look for the application in the local wait queue and set the
1736 * notify socket if found.
1737 */
d88aee68 1738 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1739 &wait_queue.head, head) {
12e2b881 1740 health_code_update();
d0b96690
DG
1741 if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
1742 wait_node->app->notify_sock = ust_cmd->sock;
1743 cds_list_del(&wait_node->head);
f45e313d 1744 wait_queue.count--;
d0b96690
DG
1745 app = wait_node->app;
1746 free(wait_node);
1747 DBG3("UST app notify socket %d is set", ust_cmd->sock);
1748 break;
1749 }
1750 }
020d7f60
DG
1751
1752 /*
1753 * With no application at this stage the received socket is
1754 * basically useless so close it before we free the cmd data
1755 * structure for good.
1756 */
1757 if (!app) {
1758 ret = close(ust_cmd->sock);
1759 if (ret < 0) {
1760 PERROR("close ust sock dispatch %d", ust_cmd->sock);
1761 }
51dec90d 1762 lttng_fd_put(LTTNG_FD_APPS, 1);
020d7f60 1763 }
7972aab2 1764 free(ust_cmd);
d0b96690
DG
1765 }
1766
1767 if (app) {
d0b96690
DG
1768 /*
1769 * @session_lock_list
1770 *
1771 * Lock the global session list so from the register up to the
1772 * registration done message, no thread can see the application
1773 * and change its state.
1774 */
1775 session_lock_list();
1776 rcu_read_lock();
d88aee68 1777
d0b96690
DG
1778 /*
1779 * Add application to the global hash table. This needs to be
1780 * done before the update to the UST registry can locate the
1781 * application.
1782 */
1783 ust_app_add(app);
d88aee68
DG
1784
1785 /* Set app version. This call will print an error if needed. */
1786 (void) ust_app_version(app);
1787
1788 /* Send notify socket through the notify pipe. */
1789 ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
1790 app->notify_sock);
1791 if (ret < 0) {
1792 rcu_read_unlock();
1793 session_unlock_list();
b85dc84c
DG
1794 /*
1795 * No notify thread, stop the UST tracing. However, this is
1796 * not an internal error of the this thread thus setting
1797 * the health error code to a normal exit.
1798 */
1799 err = 0;
d88aee68 1800 goto error;
6620da75 1801 }
d88aee68 1802
d0b96690
DG
1803 /*
1804 * Update newly registered application with the tracing
1805 * registry info already enabled information.
1806 */
1807 update_ust_app(app->sock);
d88aee68
DG
1808
1809 /*
1810 * Don't care about return value. Let the manage apps threads
1811 * handle app unregistration upon socket close.
1812 */
1813 (void) ust_app_register_done(app->sock);
1814
1815 /*
1816 * Even if the application socket has been closed, send the app
1817 * to the thread and unregistration will take place at that
1818 * place.
1819 */
1820 ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
d0b96690 1821 if (ret < 0) {
d88aee68
DG
1822 rcu_read_unlock();
1823 session_unlock_list();
b85dc84c
DG
1824 /*
1825 * No apps. thread, stop the UST tracing. However, this is
1826 * not an internal error of the this thread thus setting
1827 * the health error code to a normal exit.
1828 */
1829 err = 0;
d88aee68 1830 goto error;
d0b96690 1831 }
d88aee68 1832
d0b96690
DG
1833 rcu_read_unlock();
1834 session_unlock_list();
099e26bd 1835 }
099e26bd
DG
1836 } while (node != NULL);
1837
12e2b881 1838 health_poll_entry();
099e26bd
DG
1839 /* Futex wait on queue. Blocking call on futex() */
1840 futex_nto1_wait(&ust_cmd_queue.futex);
12e2b881 1841 health_poll_exit();
099e26bd 1842 }
12e2b881
MD
1843 /* Normal exit, no error */
1844 err = 0;
099e26bd
DG
1845
1846error:
d88aee68
DG
1847 /* Clean up wait queue. */
1848 cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
f45e313d 1849 &wait_queue.head, head) {
d88aee68 1850 cds_list_del(&wait_node->head);
f45e313d 1851 wait_queue.count--;
d88aee68
DG
1852 free(wait_node);
1853 }
1854
a09bf708 1855error_testpoint:
099e26bd 1856 DBG("Dispatch thread dying");
12e2b881
MD
1857 if (err) {
1858 health_error();
1859 ERR("Health error occurred in %s", __func__);
1860 }
8782cc74 1861 health_unregister(health_sessiond);
099e26bd
DG
1862 return NULL;
1863}
1864
1865/*
1866 * This thread manage application registration.
1867 */
1868static void *thread_registration_apps(void *data)
1d4b027a 1869{
139ac872 1870 int sock = -1, i, ret, pollfd, err = -1;
5eb91c98
DG
1871 uint32_t revents, nb_fd;
1872 struct lttng_poll_event events;
099e26bd
DG
1873 /*
1874 * Get allocated in this thread, enqueued to a global queue, dequeued and
1875 * freed in the manage apps thread.
1876 */
1877 struct ust_command *ust_cmd = NULL;
1d4b027a 1878
099e26bd 1879 DBG("[thread] Manage application registration started");
1d4b027a 1880
6c71277b 1881 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);
927ca06a 1882
71ddb4ed 1883 if (testpoint(sessiond_thread_registration_apps)) {
6993eeb3
CB
1884 goto error_testpoint;
1885 }
8ac94142 1886
1d4b027a
DG
1887 ret = lttcomm_listen_unix_sock(apps_sock);
1888 if (ret < 0) {
76d7553f 1889 goto error_listen;
1d4b027a
DG
1890 }
1891
5eb91c98
DG
1892 /*
1893 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1894 * more will be added to this poll set.
1895 */
d0b96690 1896 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 1897 if (ret < 0) {
76d7553f 1898 goto error_create_poll;
5eb91c98 1899 }
273ea72c 1900
5eb91c98
DG
1901 /* Add the application registration socket */
1902 ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
1903 if (ret < 0) {
76d7553f 1904 goto error_poll_add;
5eb91c98 1905 }
273ea72c 1906
1d4b027a 1907 /* Notify all applications to register */
0fdd1e2c
DG
1908 ret = notify_ust_apps(1);
1909 if (ret < 0) {
1910 ERR("Failed to notify applications or create the wait shared memory.\n"
54d01ffb
DG
1911 "Execution continues but there might be problem for already\n"
1912 "running applications that wishes to register.");
0fdd1e2c 1913 }
1d4b027a
DG
1914
1915 while (1) {
1916 DBG("Accepting application registration");
273ea72c
DG
1917
1918 /* Inifinite blocking call, waiting for transmission */
88f2b785 1919 restart:
a78af745 1920 health_poll_entry();
5eb91c98 1921 ret = lttng_poll_wait(&events, -1);
a78af745 1922 health_poll_exit();
273ea72c 1923 if (ret < 0) {
88f2b785
MD
1924 /*
1925 * Restart interrupted system call.
1926 */
1927 if (errno == EINTR) {
1928 goto restart;
1929 }
273ea72c
DG
1930 goto error;
1931 }
1932
0d9c5d77
DG
1933 nb_fd = ret;
1934
5eb91c98 1935 for (i = 0; i < nb_fd; i++) {
840cb59c 1936 health_code_update();
139ac872 1937
5eb91c98
DG
1938 /* Fetch once the poll data */
1939 revents = LTTNG_POLL_GETEV(&events, i);
1940 pollfd = LTTNG_POLL_GETFD(&events, i);
273ea72c 1941
f0567343
MD
1942 if (!revents) {
1943 /* No activity for this FD (poll implementation). */
1944 continue;
1945 }
1946
5eb91c98 1947 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 1948 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 1949 if (ret) {
139ac872
MD
1950 err = 0;
1951 goto exit;
90014c57 1952 }
1d4b027a 1953
5eb91c98
DG
1954 /* Event on the registration socket */
1955 if (pollfd == apps_sock) {
1956 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1957 ERR("Register apps socket poll error");
1958 goto error;
1959 } else if (revents & LPOLLIN) {
1960 sock = lttcomm_accept_unix_sock(apps_sock);
1961 if (sock < 0) {
1962 goto error;
1963 }
099e26bd 1964
16c5c8fa
DG
1965 /*
1966 * Set socket timeout for both receiving and ending.
1967 * app_socket_timeout is in seconds, whereas
1968 * lttcomm_setsockopt_rcv_timeout and
1969 * lttcomm_setsockopt_snd_timeout expect msec as
1970 * parameter.
1971 */
1972 (void) lttcomm_setsockopt_rcv_timeout(sock,
1973 app_socket_timeout * 1000);
1974 (void) lttcomm_setsockopt_snd_timeout(sock,
1975 app_socket_timeout * 1000);
1976
b662582b
DG
1977 /*
1978 * Set the CLOEXEC flag. Return code is useless because
1979 * either way, the show must go on.
1980 */
1981 (void) utils_set_fd_cloexec(sock);
1982
5eb91c98 1983 /* Create UST registration command for enqueuing */
ba7f0ae5 1984 ust_cmd = zmalloc(sizeof(struct ust_command));
5eb91c98 1985 if (ust_cmd == NULL) {
76d7553f 1986 PERROR("ust command zmalloc");
5eb91c98
DG
1987 goto error;
1988 }
1d4b027a 1989
5eb91c98
DG
1990 /*
1991 * Using message-based transmissions to ensure we don't
1992 * have to deal with partially received messages.
1993 */
4063050c
MD
1994 ret = lttng_fd_get(LTTNG_FD_APPS, 1);
1995 if (ret < 0) {
1996 ERR("Exhausted file descriptors allowed for applications.");
1997 free(ust_cmd);
1998 ret = close(sock);
1999 if (ret) {
2000 PERROR("close");
2001 }
2002 sock = -1;
2003 continue;
2004 }
d88aee68 2005
840cb59c 2006 health_code_update();
d0b96690
DG
2007 ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
2008 if (ret < 0) {
5eb91c98 2009 free(ust_cmd);
d0b96690 2010 /* Close socket of the application. */
76d7553f
MD
2011 ret = close(sock);
2012 if (ret) {
2013 PERROR("close");
2014 }
4063050c 2015 lttng_fd_put(LTTNG_FD_APPS, 1);
76d7553f 2016 sock = -1;
5eb91c98
DG
2017 continue;
2018 }
840cb59c 2019 health_code_update();
099e26bd 2020
5eb91c98 2021 ust_cmd->sock = sock;
34a2494f 2022 sock = -1;
099e26bd 2023
5eb91c98
DG
2024 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2025 " gid:%d sock:%d name:%s (version %d.%d)",
2026 ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
2027 ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
2028 ust_cmd->sock, ust_cmd->reg_msg.name,
2029 ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);
54d01ffb 2030
5eb91c98
DG
2031 /*
2032 * Lock free enqueue the registration request. The red pill
54d01ffb 2033 * has been taken! This apps will be part of the *system*.
5eb91c98
DG
2034 */
2035 cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node);
2036
2037 /*
2038 * Wake the registration queue futex. Implicit memory
2039 * barrier with the exchange in cds_wfq_enqueue.
2040 */
2041 futex_nto1_wake(&ust_cmd_queue.futex);
2042 }
2043 }
90014c57 2044 }
1d4b027a
DG
2045 }
2046
139ac872 2047exit:
1d4b027a 2048error:
0fdd1e2c
DG
2049 /* Notify that the registration thread is gone */
2050 notify_ust_apps(0);
2051
a4b35e07 2052 if (apps_sock >= 0) {
76d7553f
MD
2053 ret = close(apps_sock);
2054 if (ret) {
2055 PERROR("close");
2056 }
a4b35e07 2057 }
46c3f085 2058 if (sock >= 0) {
76d7553f
MD
2059 ret = close(sock);
2060 if (ret) {
2061 PERROR("close");
2062 }
4063050c 2063 lttng_fd_put(LTTNG_FD_APPS, 1);
a4b35e07 2064 }
273ea72c 2065 unlink(apps_unix_sock_path);
0fdd1e2c 2066
76d7553f 2067error_poll_add:
5eb91c98 2068 lttng_poll_clean(&events);
76d7553f
MD
2069error_listen:
2070error_create_poll:
6993eeb3 2071error_testpoint:
76d7553f 2072 DBG("UST Registration thread cleanup complete");
a09bf708
MD
2073 if (err) {
2074 health_error();
2075 ERR("Health error occurred in %s", __func__);
2076 }
8782cc74 2077 health_unregister(health_sessiond);
5eb91c98 2078
1d4b027a
DG
2079 return NULL;
2080}
2081
8c0faa1d 2082/*
3bd1e081 2083 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
d063d709 2084 * exec or it will fails.
8c0faa1d 2085 */
3bd1e081 2086static int spawn_consumer_thread(struct consumer_data *consumer_data)
8c0faa1d 2087{
a23ec3a7 2088 int ret, clock_ret;
ee0b0061
DG
2089 struct timespec timeout;
2090
a23ec3a7
DG
2091 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2092 consumer_data->consumer_thread_is_ready = 0;
8c0faa1d 2093
a23ec3a7
DG
2094 /* Setup pthread condition */
2095 ret = pthread_condattr_init(&consumer_data->condattr);
2096 if (ret != 0) {
2097 errno = ret;
2098 PERROR("pthread_condattr_init consumer data");
2099 goto error;
2100 }
2101
2102 /*
2103 * Set the monotonic clock in order to make sure we DO NOT jump in time
2104 * between the clock_gettime() call and the timedwait call. See bug #324
2105 * for a more details and how we noticed it.
2106 */
2107 ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
2108 if (ret != 0) {
2109 errno = ret;
2110 PERROR("pthread_condattr_setclock consumer data");
ee0b0061
DG
2111 goto error;
2112 }
8c0faa1d 2113
a23ec3a7
DG
2114 ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
2115 if (ret != 0) {
2116 errno = ret;
2117 PERROR("pthread_cond_init consumer data");
2118 goto error;
2119 }
2120
2121 ret = pthread_create(&consumer_data->thread, NULL, thread_manage_consumer,
2122 consumer_data);
8c0faa1d 2123 if (ret != 0) {
3bd1e081 2124 PERROR("pthread_create consumer");
ee0b0061 2125 ret = -1;
8c0faa1d
DG
2126 goto error;
2127 }
2128
a23ec3a7
DG
2129 /* We are about to wait on a pthread condition */
2130 pthread_mutex_lock(&consumer_data->cond_mutex);
2131
ee0b0061 2132 /* Get time for sem_timedwait absolute timeout */
a23ec3a7
DG
2133 clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
2134 /*
2135 * Set the timeout for the condition timed wait even if the clock gettime
2136 * call fails since we might loop on that call and we want to avoid to
2137 * increment the timeout too many times.
2138 */
2139 timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
2140
2141 /*
2142 * The following loop COULD be skipped in some conditions so this is why we
2143 * set ret to 0 in order to make sure at least one round of the loop is
2144 * done.
2145 */
2146 ret = 0;
2147
2148 /*
2149 * Loop until the condition is reached or when a timeout is reached. Note
2150 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2151 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2152 * possible. This loop does not take any chances and works with both of
2153 * them.
2154 */
2155 while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
2156 if (clock_ret < 0) {
2157 PERROR("clock_gettime spawn consumer");
2158 /* Infinite wait for the consumerd thread to be ready */
2159 ret = pthread_cond_wait(&consumer_data->cond,
2160 &consumer_data->cond_mutex);
2161 } else {
2162 ret = pthread_cond_timedwait(&consumer_data->cond,
2163 &consumer_data->cond_mutex, &timeout);
2164 }
ee0b0061 2165 }
8c0faa1d 2166
a23ec3a7
DG
2167 /* Release the pthread condition */
2168 pthread_mutex_unlock(&consumer_data->cond_mutex);
2169
2170 if (ret != 0) {
2171 errno = ret;
2172 if (ret == ETIMEDOUT) {
6565552a
DG
2173 int pth_ret;
2174
ee0b0061
DG
2175 /*
2176 * Call has timed out so we kill the kconsumerd_thread and return
2177 * an error.
2178 */
a23ec3a7
DG
2179 ERR("Condition timed out. The consumer thread was never ready."
2180 " Killing it");
6565552a
DG
2181 pth_ret = pthread_cancel(consumer_data->thread);
2182 if (pth_ret < 0) {
3bd1e081 2183 PERROR("pthread_cancel consumer thread");
ee0b0061
DG
2184 }
2185 } else {
a23ec3a7 2186 PERROR("pthread_cond_wait failed consumer thread");
ee0b0061 2187 }
6565552a
DG
2188 /* Caller is expecting a negative value on failure. */
2189 ret = -1;
ee0b0061
DG
2190 goto error;
2191 }
2192
3bd1e081
MD
2193 pthread_mutex_lock(&consumer_data->pid_mutex);
2194 if (consumer_data->pid == 0) {
a23ec3a7 2195 ERR("Consumerd did not start");
3bd1e081 2196 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556
DG
2197 goto error;
2198 }
3bd1e081 2199 pthread_mutex_unlock(&consumer_data->pid_mutex);
712ea556 2200
8c0faa1d
DG
2201 return 0;
2202
2203error:
2204 return ret;
2205}
2206
d9800920 2207/*
3bd1e081 2208 * Join consumer thread
d9800920 2209 */
3bd1e081 2210static int join_consumer_thread(struct consumer_data *consumer_data)
cf3af59e
MD
2211{
2212 void *status;
cf3af59e 2213
e8209f6b
DG
2214 /* Consumer pid must be a real one. */
2215 if (consumer_data->pid > 0) {
c617c0c6 2216 int ret;
3bd1e081 2217 ret = kill(consumer_data->pid, SIGTERM);
cf3af59e 2218 if (ret) {
3bd1e081 2219 ERR("Error killing consumer daemon");
cf3af59e
MD
2220 return ret;
2221 }
3bd1e081 2222 return pthread_join(consumer_data->thread, &status);
cf3af59e
MD
2223 } else {
2224 return 0;
2225 }
2226}
2227
8c0faa1d 2228/*
3bd1e081 2229 * Fork and exec a consumer daemon (consumerd).
8c0faa1d 2230 *
d063d709 2231 * Return pid if successful else -1.
8c0faa1d 2232 */
3bd1e081 2233static pid_t spawn_consumerd(struct consumer_data *consumer_data)
8c0faa1d
DG
2234{
2235 int ret;
2236 pid_t pid;
94c55f17 2237 const char *consumer_to_use;
53086306 2238 const char *verbosity;
94c55f17 2239 struct stat st;
8c0faa1d 2240
3bd1e081 2241 DBG("Spawning consumerd");
c49dc785 2242
8c0faa1d
DG
2243 pid = fork();
2244 if (pid == 0) {
2245 /*
3bd1e081 2246 * Exec consumerd.
8c0faa1d 2247 */
daee5345 2248 if (opt_verbose_consumer) {
53086306
DG
2249 verbosity = "--verbose";
2250 } else {
2251 verbosity = "--quiet";
2252 }
3bd1e081
MD
2253 switch (consumer_data->type) {
2254 case LTTNG_CONSUMER_KERNEL:
94c55f17 2255 /*
c7704d57
DG
2256 * Find out which consumerd to execute. We will first try the
2257 * 64-bit path, then the sessiond's installation directory, and
2258 * fallback on the 32-bit one,
94c55f17 2259 */
63a799e8
AM
2260 DBG3("Looking for a kernel consumer at these locations:");
2261 DBG3(" 1) %s", consumerd64_bin);
2262 DBG3(" 2) %s/%s", INSTALL_BIN_PATH, CONSUMERD_FILE);
2263 DBG3(" 3) %s", consumerd32_bin);
94c55f17 2264 if (stat(consumerd64_bin, &st) == 0) {
63a799e8 2265 DBG3("Found location #1");
94c55f17 2266 consumer_to_use = consumerd64_bin;
94c55f17 2267 } else if (stat(INSTALL_BIN_PATH "/" CONSUMERD_FILE, &st) == 0) {
63a799e8 2268 DBG3("Found location #2");
94c55f17 2269 consumer_to_use = INSTALL_BIN_PATH "/" CONSUMERD_FILE;
eb1e0bd4 2270 } else if (stat(consumerd32_bin, &st) == 0) {
63a799e8 2271 DBG3("Found location #3");
eb1e0bd4 2272 consumer_to_use = consumerd32_bin;
94c55f17 2273 } else {
63a799e8 2274 DBG("Could not find any valid consumerd executable");
6565552a 2275 ret = -EINVAL;
94c55f17
AM
2276 break;
2277 }
2278 DBG("Using kernel consumer at: %s", consumer_to_use);
6565552a 2279 ret = execl(consumer_to_use,
94c55f17
AM
2280 "lttng-consumerd", verbosity, "-k",
2281 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2282 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2283 "--group", tracing_group_name,
94c55f17 2284 NULL);
3bd1e081 2285 break;
7753dea8
MD
2286 case LTTNG_CONSUMER64_UST:
2287 {
b1e0b6b6 2288 char *tmpnew = NULL;
8f4905da
MD
2289
2290 if (consumerd64_libdir[0] != '\0') {
2291 char *tmp;
2292 size_t tmplen;
2293
2294 tmp = getenv("LD_LIBRARY_PATH");
2295 if (!tmp) {
2296 tmp = "";
2297 }
2298 tmplen = strlen("LD_LIBRARY_PATH=")
2299 + strlen(consumerd64_libdir) + 1 /* : */ + strlen(tmp);
2300 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2301 if (!tmpnew) {
2302 ret = -ENOMEM;
2303 goto error;
2304 }
2305 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2306 strcat(tmpnew, consumerd64_libdir);
2307 if (tmp[0] != '\0') {
2308 strcat(tmpnew, ":");
2309 strcat(tmpnew, tmp);
2310 }
2311 ret = putenv(tmpnew);
2312 if (ret) {
2313 ret = -errno;
c6f76da9 2314 free(tmpnew);
8f4905da
MD
2315 goto error;
2316 }
2317 }
94c55f17 2318 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin);
a5a6aff3 2319 ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2320 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2321 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2322 "--group", tracing_group_name,
7753dea8 2323 NULL);
8f4905da
MD
2324 if (consumerd64_libdir[0] != '\0') {
2325 free(tmpnew);
2326 }
3bd1e081 2327 break;
7753dea8
MD
2328 }
2329 case LTTNG_CONSUMER32_UST:
2330 {
937dde8e 2331 char *tmpnew = NULL;
8f4905da
MD
2332
2333 if (consumerd32_libdir[0] != '\0') {
2334 char *tmp;
2335 size_t tmplen;
2336
2337 tmp = getenv("LD_LIBRARY_PATH");
2338 if (!tmp) {
2339 tmp = "";
2340 }
2341 tmplen = strlen("LD_LIBRARY_PATH=")
2342 + strlen(consumerd32_libdir) + 1 /* : */ + strlen(tmp);
2343 tmpnew = zmalloc(tmplen + 1 /* \0 */);
2344 if (!tmpnew) {
2345 ret = -ENOMEM;
2346 goto error;
2347 }
2348 strcpy(tmpnew, "LD_LIBRARY_PATH=");
2349 strcat(tmpnew, consumerd32_libdir);
2350 if (tmp[0] != '\0') {
2351 strcat(tmpnew, ":");
2352 strcat(tmpnew, tmp);
2353 }
2354 ret = putenv(tmpnew);
2355 if (ret) {
2356 ret = -errno;
c6f76da9 2357 free(tmpnew);
8f4905da
MD
2358 goto error;
2359 }
2360 }
94c55f17 2361 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin);
a5a6aff3 2362 ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u",
7753dea8
MD
2363 "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
2364 "--consumerd-err-sock", consumer_data->err_unix_sock_path,
6c71277b 2365 "--group", tracing_group_name,
7753dea8 2366 NULL);
8f4905da
MD
2367 if (consumerd32_libdir[0] != '\0') {
2368 free(tmpnew);
2369 }
7753dea8
MD
2370 break;
2371 }
3bd1e081 2372 default:
76d7553f 2373 PERROR("unknown consumer type");
3bd1e081
MD
2374 exit(EXIT_FAILURE);
2375 }
8c0faa1d 2376 if (errno != 0) {
6565552a 2377 PERROR("Consumer execl()");
8c0faa1d 2378 }
6565552a 2379 /* Reaching this point, we got a failure on our execl(). */
8c0faa1d
DG
2380 exit(EXIT_FAILURE);
2381 } else if (pid > 0) {
2382 ret = pid;
8c0faa1d 2383 } else {
76d7553f 2384 PERROR("start consumer fork");
8c0faa1d 2385 ret = -errno;
8c0faa1d 2386 }
8f4905da 2387error:
8c0faa1d
DG
2388 return ret;
2389}
2390
693bd40b 2391/*
3bd1e081 2392 * Spawn the consumerd daemon and session daemon thread.
693bd40b 2393 */
3bd1e081 2394static int start_consumerd(struct consumer_data *consumer_data)
693bd40b 2395{
c617c0c6 2396 int ret;
edb8b045
DG
2397
2398 /*
2399 * Set the listen() state on the socket since there is a possible race
2400 * between the exec() of the consumer daemon and this call if place in the
2401 * consumer thread. See bug #366 for more details.
2402 */
2403 ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
2404 if (ret < 0) {
2405 goto error;
2406 }
693bd40b 2407
3bd1e081
MD
2408 pthread_mutex_lock(&consumer_data->pid_mutex);
2409 if (consumer_data->pid != 0) {
2410 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785
DG
2411 goto end;
2412 }
693bd40b 2413
3bd1e081 2414 ret = spawn_consumerd(consumer_data);
c49dc785 2415 if (ret < 0) {
3bd1e081
MD
2416 ERR("Spawning consumerd failed");
2417 pthread_mutex_unlock(&consumer_data->pid_mutex);
c49dc785 2418 goto error;
693bd40b 2419 }
c49dc785 2420
3bd1e081
MD
2421 /* Setting up the consumer_data pid */
2422 consumer_data->pid = ret;
48842b30 2423 DBG2("Consumer pid %d", consumer_data->pid);
3bd1e081 2424 pthread_mutex_unlock(&consumer_data->pid_mutex);
693bd40b 2425
3bd1e081
MD
2426 DBG2("Spawning consumer control thread");
2427 ret = spawn_consumer_thread(consumer_data);
693bd40b 2428 if (ret < 0) {
3bd1e081 2429 ERR("Fatal error spawning consumer control thread");
693bd40b
DG
2430 goto error;
2431 }
2432
c49dc785 2433end:
693bd40b
DG
2434 return 0;
2435
2436error:
331744e3 2437 /* Cleanup already created sockets on error. */
edb8b045 2438 if (consumer_data->err_sock >= 0) {
c617c0c6
MD
2439 int err;
2440
edb8b045
DG
2441 err = close(consumer_data->err_sock);
2442 if (err < 0) {
2443 PERROR("close consumer data error socket");
2444 }
2445 }
693bd40b
DG
2446 return ret;
2447}
2448
b73401da 2449/*
096102bd 2450 * Setup necessary data for kernel tracer action.
b73401da 2451 */
096102bd 2452static int init_kernel_tracer(void)
b73401da
DG
2453{
2454 int ret;
b73401da 2455
096102bd
DG
2456 /* Modprobe lttng kernel modules */
2457 ret = modprobe_lttng_control();
b73401da 2458 if (ret < 0) {
b73401da
DG
2459 goto error;
2460 }
2461
096102bd
DG
2462 /* Open debugfs lttng */
2463 kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
2464 if (kernel_tracer_fd < 0) {
2465 DBG("Failed to open %s", module_proc_lttng);
2f77fc4b
DG
2466 ret = -1;
2467 goto error_open;
54d01ffb
DG
2468 }
2469
2f77fc4b
DG
2470 /* Validate kernel version */
2471 ret = kernel_validate_version(kernel_tracer_fd);
2472 if (ret < 0) {
2473 goto error_version;
b551a063 2474 }
54d01ffb 2475
2f77fc4b
DG
2476 ret = modprobe_lttng_data();
2477 if (ret < 0) {
2478 goto error_modules;
54d01ffb
DG
2479 }
2480
2f77fc4b
DG
2481 DBG("Kernel tracer fd %d", kernel_tracer_fd);
2482 return 0;
2483
2484error_version:
2485 modprobe_remove_lttng_control();
2486 ret = close(kernel_tracer_fd);
2487 if (ret) {
2488 PERROR("close");
b551a063 2489 }
2f77fc4b 2490 kernel_tracer_fd = -1;
f73fabfd 2491 return LTTNG_ERR_KERN_VERSION;
b551a063 2492
2f77fc4b
DG
2493error_modules:
2494 ret = close(kernel_tracer_fd);
2495 if (ret) {
2496 PERROR("close");
b551a063 2497 }
54d01ffb 2498
2f77fc4b
DG
2499error_open:
2500 modprobe_remove_lttng_control();
54d01ffb
DG
2501
2502error:
2f77fc4b
DG
2503 WARN("No kernel tracer available");
2504 kernel_tracer_fd = -1;
2505 if (!is_root) {
f73fabfd 2506 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2f77fc4b 2507 } else {
f73fabfd 2508 return LTTNG_ERR_KERN_NA;
2f77fc4b 2509 }
54d01ffb
DG
2510}
2511
2f77fc4b 2512
54d01ffb 2513/*
2f77fc4b
DG
2514 * Copy consumer output from the tracing session to the domain session. The
2515 * function also applies the right modification on a per domain basis for the
2516 * trace files destination directory.
36b588ed
MD
2517 *
2518 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2519 */
2f77fc4b 2520static int copy_session_consumer(int domain, struct ltt_session *session)
54d01ffb
DG
2521{
2522 int ret;
2f77fc4b
DG
2523 const char *dir_name;
2524 struct consumer_output *consumer;
2525
2526 assert(session);
2527 assert(session->consumer);
54d01ffb 2528
b551a063
DG
2529 switch (domain) {
2530 case LTTNG_DOMAIN_KERNEL:
2f77fc4b 2531 DBG3("Copying tracing session consumer output in kernel session");
09a90bcd
DG
2532 /*
2533 * XXX: We should audit the session creation and what this function
2534 * does "extra" in order to avoid a destroy since this function is used
2535 * in the domain session creation (kernel and ust) only. Same for UST
2536 * domain.
2537 */
2538 if (session->kernel_session->consumer) {
2539 consumer_destroy_output(session->kernel_session->consumer);
2540 }
2f77fc4b
DG
2541 session->kernel_session->consumer =
2542 consumer_copy_output(session->consumer);
2543 /* Ease our life a bit for the next part */
2544 consumer = session->kernel_session->consumer;
2545 dir_name = DEFAULT_KERNEL_TRACE_DIR;
b551a063 2546 break;
f20baf8e 2547 case LTTNG_DOMAIN_JUL:
b551a063 2548 case LTTNG_DOMAIN_UST:
2f77fc4b 2549 DBG3("Copying tracing session consumer output in UST session");
09a90bcd
DG
2550 if (session->ust_session->consumer) {
2551 consumer_destroy_output(session->ust_session->consumer);
2552 }
2f77fc4b
DG
2553 session->ust_session->consumer =
2554 consumer_copy_output(session->consumer);
2555 /* Ease our life a bit for the next part */
2556 consumer = session->ust_session->consumer;
2557 dir_name = DEFAULT_UST_TRACE_DIR;
b551a063
DG
2558 break;
2559 default:
f73fabfd 2560 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
54d01ffb
DG
2561 goto error;
2562 }
2563
2f77fc4b 2564 /* Append correct directory to subdir */
c30ce0b3
CB
2565 strncat(consumer->subdir, dir_name,
2566 sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
2f77fc4b
DG
2567 DBG3("Copy session consumer subdir %s", consumer->subdir);
2568
f73fabfd 2569 ret = LTTNG_OK;
54d01ffb
DG
2570
2571error:
2572 return ret;
2573}
2574
00e2e675 2575/*
2f77fc4b 2576 * Create an UST session and add it to the session ust list.
36b588ed
MD
2577 *
2578 * Should *NOT* be called with RCU read-side lock held.
00e2e675 2579 */
2f77fc4b
DG
2580static int create_ust_session(struct ltt_session *session,
2581 struct lttng_domain *domain)
00e2e675
DG
2582{
2583 int ret;
2f77fc4b 2584 struct ltt_ust_session *lus = NULL;
00e2e675 2585
a4b92340 2586 assert(session);
2f77fc4b
DG
2587 assert(domain);
2588 assert(session->consumer);
a4b92340 2589
2f77fc4b 2590 switch (domain->type) {
f20baf8e 2591 case LTTNG_DOMAIN_JUL:
2f77fc4b
DG
2592 case LTTNG_DOMAIN_UST:
2593 break;
2594 default:
2595 ERR("Unknown UST domain on create session %d", domain->type);
f73fabfd 2596 ret = LTTNG_ERR_UNKNOWN_DOMAIN;
00e2e675
DG
2597 goto error;
2598 }
2599
2f77fc4b
DG
2600 DBG("Creating UST session");
2601
dec56f6c 2602 lus = trace_ust_create_session(session->id);
2f77fc4b 2603 if (lus == NULL) {
f73fabfd 2604 ret = LTTNG_ERR_UST_SESS_FAIL;
a4b92340
DG
2605 goto error;
2606 }
2607
2f77fc4b
DG
2608 lus->uid = session->uid;
2609 lus->gid = session->gid;
2bba9e53 2610 lus->output_traces = session->output_traces;
27babd3a 2611 lus->snapshot_mode = session->snapshot_mode;
ecc48a90 2612 lus->live_timer_interval = session->live_timer;
2f77fc4b 2613 session->ust_session = lus;
00e2e675 2614
2f77fc4b
DG
2615 /* Copy session output to the newly created UST session */
2616 ret = copy_session_consumer(domain->type, session);
f73fabfd 2617 if (ret != LTTNG_OK) {
00e2e675
DG
2618 goto error;
2619 }
2620
f73fabfd 2621 return LTTNG_OK;
00e2e675
DG
2622
2623error:
2f77fc4b
DG
2624 free(lus);
2625 session->ust_session = NULL;
00e2e675
DG
2626 return ret;
2627}
2628
2629/*
2f77fc4b 2630 * Create a kernel tracer session then create the default channel.
00e2e675 2631 */
2f77fc4b 2632static int create_kernel_session(struct ltt_session *session)
00e2e675
DG
2633{
2634 int ret;
a4b92340 2635
2f77fc4b 2636 DBG("Creating kernel session");
00e2e675 2637
2f77fc4b
DG
2638 ret = kernel_create_session(session, kernel_tracer_fd);
2639 if (ret < 0) {
f73fabfd 2640 ret = LTTNG_ERR_KERN_SESS_FAIL;
00e2e675
DG
2641 goto error;
2642 }
2643
2f77fc4b
DG
2644 /* Code flow safety */
2645 assert(session->kernel_session);
2646
2647 /* Copy session output to the newly created Kernel session */
2648 ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
f73fabfd 2649 if (ret != LTTNG_OK) {
a4b92340
DG
2650 goto error;
2651 }
2652
2f77fc4b
DG
2653 /* Create directory(ies) on local filesystem. */
2654 if (session->kernel_session->consumer->type == CONSUMER_DST_LOCAL &&
2655 strlen(session->kernel_session->consumer->dst.trace_path) > 0) {
2656 ret = run_as_mkdir_recursive(
2657 session->kernel_session->consumer->dst.trace_path,
2658 S_IRWXU | S_IRWXG, session->uid, session->gid);
2659 if (ret < 0) {
2660 if (ret != -EEXIST) {
2661 ERR("Trace directory creation error");
00e2e675
DG
2662 goto error;
2663 }
00e2e675 2664 }
2f77fc4b 2665 }
00e2e675 2666
2f77fc4b
DG
2667 session->kernel_session->uid = session->uid;
2668 session->kernel_session->gid = session->gid;
2bba9e53 2669 session->kernel_session->output_traces = session->output_traces;
27babd3a 2670 session->kernel_session->snapshot_mode = session->snapshot_mode;
00e2e675 2671
f73fabfd 2672 return LTTNG_OK;
00e2e675 2673
2f77fc4b
DG
2674error:
2675 trace_kernel_destroy_session(session->kernel_session);
2676 session->kernel_session = NULL;
2677 return ret;
2678}
00e2e675 2679
2f77fc4b
DG
2680/*
2681 * Count number of session permitted by uid/gid.
2682 */
2683static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
2684{
2685 unsigned int i = 0;
2686 struct ltt_session *session;
07424f16 2687
2f77fc4b
DG
2688 DBG("Counting number of available session for UID %d GID %d",
2689 uid, gid);
2690 cds_list_for_each_entry(session, &session_list_ptr->head, list) {
00e2e675 2691 /*
2f77fc4b 2692 * Only list the sessions the user can control.
00e2e675 2693 */
2f77fc4b
DG
2694 if (!session_access_ok(session, uid, gid)) {
2695 continue;
2696 }
2697 i++;
a4b92340 2698 }
2f77fc4b 2699 return i;
00e2e675
DG
2700}
2701
54d01ffb
DG
2702/*
2703 * Process the command requested by the lttng client within the command
2704 * context structure. This function make sure that the return structure (llm)
2705 * is set and ready for transmission before returning.
2706 *
2707 * Return any error encountered or 0 for success.
53a80697
MD
2708 *
2709 * "sock" is only used for special-case var. len data.
36b588ed
MD
2710 *
2711 * Should *NOT* be called with RCU read-side lock held.
54d01ffb 2712 */
53a80697
MD
2713static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
2714 int *sock_error)
54d01ffb 2715{
f73fabfd 2716 int ret = LTTNG_OK;
44d3bd01 2717 int need_tracing_session = 1;
2e09ba09 2718 int need_domain;
54d01ffb
DG
2719
2720 DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);
2721
53a80697
MD
2722 *sock_error = 0;
2723
2e09ba09
MD
2724 switch (cmd_ctx->lsm->cmd_type) {
2725 case LTTNG_CREATE_SESSION:
27babd3a 2726 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2727 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09
MD
2728 case LTTNG_DESTROY_SESSION:
2729 case LTTNG_LIST_SESSIONS:
2730 case LTTNG_LIST_DOMAINS:
2731 case LTTNG_START_TRACE:
2732 case LTTNG_STOP_TRACE:
6d805429 2733 case LTTNG_DATA_PENDING:
da3c9ec1
DG
2734 case LTTNG_SNAPSHOT_ADD_OUTPUT:
2735 case LTTNG_SNAPSHOT_DEL_OUTPUT:
2736 case LTTNG_SNAPSHOT_LIST_OUTPUT:
2737 case LTTNG_SNAPSHOT_RECORD:
2e09ba09 2738 need_domain = 0;
3aace903 2739 break;
2e09ba09
MD
2740 default:
2741 need_domain = 1;
2742 }
2743
2744 if (opt_no_kernel && need_domain
2745 && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
531d29f9 2746 if (!is_root) {
f73fabfd 2747 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
531d29f9 2748 } else {
f73fabfd 2749 ret = LTTNG_ERR_KERN_NA;
531d29f9 2750 }
4fba7219
DG
2751 goto error;
2752 }
2753
8d3113b2
DG
2754 /* Deny register consumer if we already have a spawned consumer. */
2755 if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
2756 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2757 if (kconsumer_data.pid > 0) {
f73fabfd 2758 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
fa317f24 2759 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
8d3113b2
DG
2760 goto error;
2761 }
2762 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2763 }
2764
54d01ffb
DG
2765 /*
2766 * Check for command that don't needs to allocate a returned payload. We do
44d3bd01 2767 * this here so we don't have to make the call for no payload at each
54d01ffb
DG
2768 * command.
2769 */
2770 switch(cmd_ctx->lsm->cmd_type) {
2771 case LTTNG_LIST_SESSIONS:
2772 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2773 case LTTNG_LIST_TRACEPOINT_FIELDS:
54d01ffb
DG
2774 case LTTNG_LIST_DOMAINS:
2775 case LTTNG_LIST_CHANNELS:
2776 case LTTNG_LIST_EVENTS:
2777 break;
2778 default:
2779 /* Setup lttng message with no payload */
2780 ret = setup_lttng_msg(cmd_ctx, 0);
2781 if (ret < 0) {
2782 /* This label does not try to unlock the session */
2783 goto init_setup_error;
2784 }
2785 }
2786
2787 /* Commands that DO NOT need a session. */
2788 switch (cmd_ctx->lsm->cmd_type) {
54d01ffb 2789 case LTTNG_CREATE_SESSION:
27babd3a 2790 case LTTNG_CREATE_SESSION_SNAPSHOT:
ecc48a90 2791 case LTTNG_CREATE_SESSION_LIVE:
2e09ba09 2792 case LTTNG_CALIBRATE:
54d01ffb
DG
2793 case LTTNG_LIST_SESSIONS:
2794 case LTTNG_LIST_TRACEPOINTS:
f37d259d 2795 case LTTNG_LIST_TRACEPOINT_FIELDS:
44d3bd01 2796 need_tracing_session = 0;
54d01ffb
DG
2797 break;
2798 default:
2799 DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
256a5576
MD
2800 /*
2801 * We keep the session list lock across _all_ commands
2802 * for now, because the per-session lock does not
2803 * handle teardown properly.
2804 */
74babd95 2805 session_lock_list();
54d01ffb
DG
2806 cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
2807 if (cmd_ctx->session == NULL) {
bba2d65f 2808 ret = LTTNG_ERR_SESS_NOT_FOUND;
54d01ffb
DG
2809 goto error;
2810 } else {
2811 /* Acquire lock for the session */
2812 session_lock(cmd_ctx->session);
2813 }
2814 break;
2815 }
b389abbe 2816
2e09ba09
MD
2817 if (!need_domain) {
2818 goto skip_domain;
2819 }
a4b92340 2820
54d01ffb
DG
2821 /*
2822 * Check domain type for specific "pre-action".
2823 */
2824 switch (cmd_ctx->lsm->domain.type) {
2825 case LTTNG_DOMAIN_KERNEL:
d1f1c568 2826 if (!is_root) {
f73fabfd 2827 ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
d1f1c568
DG
2828 goto error;
2829 }
2830
54d01ffb 2831 /* Kernel tracer check */
a4b35e07 2832 if (kernel_tracer_fd == -1) {
54d01ffb 2833 /* Basically, load kernel tracer modules */
096102bd
DG
2834 ret = init_kernel_tracer();
2835 if (ret != 0) {
54d01ffb
DG
2836 goto error;
2837 }
2838 }
5eb91c98 2839
5c827ce0
DG
2840 /* Consumer is in an ERROR state. Report back to client */
2841 if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2842 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2843 goto error;
2844 }
2845
54d01ffb 2846 /* Need a session for kernel command */
44d3bd01 2847 if (need_tracing_session) {
54d01ffb 2848 if (cmd_ctx->session->kernel_session == NULL) {
6df2e2c9 2849 ret = create_kernel_session(cmd_ctx->session);
5eb91c98 2850 if (ret < 0) {
f73fabfd 2851 ret = LTTNG_ERR_KERN_SESS_FAIL;
5eb91c98
DG
2852 goto error;
2853 }
b389abbe 2854 }
7d29a247 2855
54d01ffb 2856 /* Start the kernel consumer daemon */
3bd1e081
MD
2857 pthread_mutex_lock(&kconsumer_data.pid_mutex);
2858 if (kconsumer_data.pid == 0 &&
785d2d0d 2859 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
3bd1e081
MD
2860 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
2861 ret = start_consumerd(&kconsumer_data);
7d29a247 2862 if (ret < 0) {
f73fabfd 2863 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
54d01ffb 2864 goto error;
950131af 2865 }
5c827ce0 2866 uatomic_set(&kernel_consumerd_state, CONSUMER_STARTED);
3ff2ecac
MD
2867 } else {
2868 pthread_mutex_unlock(&kconsumer_data.pid_mutex);
33a2b854 2869 }
173af62f 2870
a4b92340
DG
2871 /*
2872 * The consumer was just spawned so we need to add the socket to
2873 * the consumer output of the session if exist.
2874 */
2875 ret = consumer_create_socket(&kconsumer_data,
2876 cmd_ctx->session->kernel_session->consumer);
2877 if (ret < 0) {
2878 goto error;
173af62f 2879 }
0d0c377a 2880 }
5c827ce0 2881
54d01ffb 2882 break;
b9dfb167 2883 case LTTNG_DOMAIN_JUL:
2bdd86d4 2884 case LTTNG_DOMAIN_UST:
44d3bd01 2885 {
b51ec5b4
MD
2886 if (!ust_app_supported()) {
2887 ret = LTTNG_ERR_NO_UST;
2888 goto error;
2889 }
5c827ce0
DG
2890 /* Consumer is in an ERROR state. Report back to client */
2891 if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
f73fabfd 2892 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2893 goto error;
2894 }
2895
44d3bd01 2896 if (need_tracing_session) {
a4b92340 2897 /* Create UST session if none exist. */
f6a9efaa 2898 if (cmd_ctx->session->ust_session == NULL) {
44d3bd01 2899 ret = create_ust_session(cmd_ctx->session,
6df2e2c9 2900 &cmd_ctx->lsm->domain);
f73fabfd 2901 if (ret != LTTNG_OK) {
44d3bd01
DG
2902 goto error;
2903 }
2904 }
00e2e675 2905
7753dea8
MD
2906 /* Start the UST consumer daemons */
2907 /* 64-bit */
2908 pthread_mutex_lock(&ustconsumer64_data.pid_mutex);
fc7a59ce 2909 if (consumerd64_bin[0] != '\0' &&
7753dea8 2910 ustconsumer64_data.pid == 0 &&
785d2d0d 2911 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2912 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2913 ret = start_consumerd(&ustconsumer64_data);
2bdd86d4 2914 if (ret < 0) {
f73fabfd 2915 ret = LTTNG_ERR_UST_CONSUMER64_FAIL;
173af62f 2916 uatomic_set(&ust_consumerd64_fd, -EINVAL);
2bdd86d4
MD
2917 goto error;
2918 }
48842b30 2919
173af62f 2920 uatomic_set(&ust_consumerd64_fd, ustconsumer64_data.cmd_sock);
5c827ce0 2921 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
3ff2ecac 2922 } else {
7753dea8
MD
2923 pthread_mutex_unlock(&ustconsumer64_data.pid_mutex);
2924 }
173af62f
DG
2925
2926 /*
2927 * Setup socket for consumer 64 bit. No need for atomic access
2928 * since it was set above and can ONLY be set in this thread.
2929 */
a4b92340
DG
2930 ret = consumer_create_socket(&ustconsumer64_data,
2931 cmd_ctx->session->ust_session->consumer);
2932 if (ret < 0) {
2933 goto error;
173af62f
DG
2934 }
2935
7753dea8 2936 /* 32-bit */
f0601a08 2937 pthread_mutex_lock(&ustconsumer32_data.pid_mutex);
fc7a59ce 2938 if (consumerd32_bin[0] != '\0' &&
7753dea8 2939 ustconsumer32_data.pid == 0 &&
785d2d0d 2940 cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) {
7753dea8
MD
2941 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2942 ret = start_consumerd(&ustconsumer32_data);
2943 if (ret < 0) {
f73fabfd 2944 ret = LTTNG_ERR_UST_CONSUMER32_FAIL;
173af62f 2945 uatomic_set(&ust_consumerd32_fd, -EINVAL);
7753dea8
MD
2946 goto error;
2947 }
5c827ce0 2948
173af62f 2949 uatomic_set(&ust_consumerd32_fd, ustconsumer32_data.cmd_sock);
5c827ce0 2950 uatomic_set(&ust_consumerd_state, CONSUMER_STARTED);
7753dea8
MD
2951 } else {
2952 pthread_mutex_unlock(&ustconsumer32_data.pid_mutex);
2bdd86d4 2953 }
173af62f
DG
2954
2955 /*
2956 * Setup socket for consumer 64 bit. No need for atomic access
2957 * since it was set above and can ONLY be set in this thread.
2958 */
a4b92340
DG
2959 ret = consumer_create_socket(&ustconsumer32_data,
2960 cmd_ctx->session->ust_session->consumer);
2961 if (ret < 0) {
2962 goto error;
173af62f 2963 }
44d3bd01
DG
2964 }
2965 break;
48842b30 2966 }
54d01ffb 2967 default:
54d01ffb
DG
2968 break;
2969 }
2e09ba09 2970skip_domain:
33a2b854 2971
5c827ce0
DG
2972 /* Validate consumer daemon state when start/stop trace command */
2973 if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE ||
2974 cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) {
2975 switch (cmd_ctx->lsm->domain.type) {
b9dfb167 2976 case LTTNG_DOMAIN_JUL:
5c827ce0
DG
2977 case LTTNG_DOMAIN_UST:
2978 if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2979 ret = LTTNG_ERR_NO_USTCONSUMERD;
5c827ce0
DG
2980 goto error;
2981 }
2982 break;
2983 case LTTNG_DOMAIN_KERNEL:
2984 if (uatomic_read(&kernel_consumerd_state) != CONSUMER_STARTED) {
f73fabfd 2985 ret = LTTNG_ERR_NO_KERNCONSUMERD;
5c827ce0
DG
2986 goto error;
2987 }
2988 break;
2989 }
2990 }
2991
8e0af1b4
MD
2992 /*
2993 * Check that the UID or GID match that of the tracing session.
2994 * The root user can interact with all sessions.
2995 */
2996 if (need_tracing_session) {
2997 if (!session_access_ok(cmd_ctx->session,
730389d9
DG
2998 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
2999 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds))) {
f73fabfd 3000 ret = LTTNG_ERR_EPERM;
8e0af1b4
MD
3001 goto error;
3002 }
3003 }
3004
ffe60014
DG
3005 /*
3006 * Send relayd information to consumer as soon as we have a domain and a
3007 * session defined.
3008 */
3009 if (cmd_ctx->session && need_domain) {
3010 /*
3011 * Setup relayd if not done yet. If the relayd information was already
3012 * sent to the consumer, this call will gracefully return.
3013 */
3014 ret = cmd_setup_relayd(cmd_ctx->session);
3015 if (ret != LTTNG_OK) {
3016 goto error;
3017 }
3018 }
3019
54d01ffb
DG
3020 /* Process by command type */
3021 switch (cmd_ctx->lsm->cmd_type) {
3022 case LTTNG_ADD_CONTEXT:
3023 {
3024 ret = cmd_add_context(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3025 cmd_ctx->lsm->u.context.channel_name,
979e618e 3026 &cmd_ctx->lsm->u.context.ctx, kernel_poll_pipe[1]);
54d01ffb
DG
3027 break;
3028 }
3029 case LTTNG_DISABLE_CHANNEL:
3030 {
3031 ret = cmd_disable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3032 cmd_ctx->lsm->u.disable.channel_name);
3033 break;
3034 }
3035 case LTTNG_DISABLE_EVENT:
3036 {
3037 ret = cmd_disable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3038 cmd_ctx->lsm->u.disable.channel_name,
3039 cmd_ctx->lsm->u.disable.name);
33a2b854
DG
3040 break;
3041 }
54d01ffb
DG
3042 case LTTNG_DISABLE_ALL_EVENT:
3043 {
2bdd86d4 3044 DBG("Disabling all events");
54d01ffb
DG
3045
3046 ret = cmd_disable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type,
3047 cmd_ctx->lsm->u.disable.channel_name);
3048 break;
3049 }
3050 case LTTNG_ENABLE_CHANNEL:
3051 {
7972aab2 3052 ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain,
2f77fc4b 3053 &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]);
54d01ffb
DG
3054 break;
3055 }
3056 case LTTNG_ENABLE_EVENT:
3057 {
882ef835
JI
3058 struct lttng_event_exclusion *exclusion = NULL;
3059 struct lttng_filter_bytecode *bytecode = NULL;
3060
67676bd8
DG
3061 /* Handle exclusion events and receive it from the client. */
3062 if (cmd_ctx->lsm->u.enable.exclusion_count > 0) {
3063 size_t count = cmd_ctx->lsm->u.enable.exclusion_count;
3064
3065 exclusion = zmalloc(sizeof(struct lttng_event_exclusion) +
3066 (count * LTTNG_SYMBOL_NAME_LEN));
3067 if (!exclusion) {
3068 ret = LTTNG_ERR_EXCLUSION_NOMEM;
3069 goto error;
882ef835 3070 }
882ef835 3071
67676bd8
DG
3072 DBG("Receiving var len exclusion event list from client ...");
3073 exclusion->count = count;
3074 ret = lttcomm_recv_unix_sock(sock, exclusion->names,
3075 count * LTTNG_SYMBOL_NAME_LEN);
3076 if (ret <= 0) {
3077 DBG("Nothing recv() from client var len data... continuing");
3078 *sock_error = 1;
3079 free(exclusion);
3080 ret = LTTNG_ERR_EXCLUSION_INVAL;
3081 goto error;
3082 }
3083 }
3084
3085 /* Handle filter and get bytecode from client. */
3086 if (cmd_ctx->lsm->u.enable.bytecode_len > 0) {
3087 size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len;
3088
3089 if (bytecode_len > LTTNG_FILTER_MAX_LEN) {
3090 ret = LTTNG_ERR_FILTER_INVAL;
3091 free(exclusion);
3092 goto error;
3093 }
3094
3095 bytecode = zmalloc(bytecode_len);
3096 if (!bytecode) {
3097 free(exclusion);
3098 ret = LTTNG_ERR_FILTER_NOMEM;
3099 goto error;
882ef835
JI
3100 }
3101
67676bd8
DG
3102 /* Receive var. len. data */
3103 DBG("Receiving var len filter's bytecode from client ...");
3104 ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len);
3105 if (ret <= 0) {
3106 DBG("Nothing recv() from client car len data... continuing");
3107 *sock_error = 1;
3108 free(bytecode);
3109 free(exclusion);
3110 ret = LTTNG_ERR_FILTER_INVAL;
3111 goto error;
3112 }
3113
3114 if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) {
3115 free(bytecode);
3116 free(exclusion);
3117 ret = LTTNG_ERR_FILTER_INVAL;
3118 goto error;
3119 }
882ef835 3120 }
67676bd8
DG
3121
3122 ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain,
3123 cmd_ctx->lsm->u.enable.channel_name,
3124 &cmd_ctx->lsm->u.enable.event, bytecode, exclusion,
3125 kernel_poll_pipe[1]);
54d01ffb
DG
3126 break;
3127 }
3128 case LTTNG_ENABLE_ALL_EVENT:
3129 {
2bdd86d4 3130 DBG("Enabling all events");
54d01ffb 3131
7972aab2 3132 ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain,
8c9ae521 3133 cmd_ctx->lsm->u.enable.channel_name,
025faf73 3134 cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]);
54d01ffb
DG
3135 break;
3136 }
052da939 3137 case LTTNG_LIST_TRACEPOINTS:
2ef84c95 3138 {
9f19cc17 3139 struct lttng_event *events;
54d01ffb 3140 ssize_t nb_events;
052da939 3141
a5b55f38 3142 session_lock_list();
54d01ffb 3143 nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events);
a5b55f38 3144 session_unlock_list();
54d01ffb 3145 if (nb_events < 0) {
f73fabfd 3146 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3147 ret = -nb_events;
3148 goto error;
2ef84c95
DG
3149 }
3150
3151 /*
3152 * Setup lttng message with payload size set to the event list size in
3153 * bytes and then copy list into the llm payload.
3154 */
052da939 3155 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_event) * nb_events);
2ef84c95 3156 if (ret < 0) {
052da939 3157 free(events);
2ef84c95
DG
3158 goto setup_error;
3159 }
3160
3161 /* Copy event list into message payload */
9f19cc17 3162 memcpy(cmd_ctx->llm->payload, events,
052da939 3163 sizeof(struct lttng_event) * nb_events);
2ef84c95 3164
9f19cc17 3165 free(events);
2ef84c95 3166
f73fabfd 3167 ret = LTTNG_OK;
2ef84c95
DG
3168 break;
3169 }
f37d259d
MD
3170 case LTTNG_LIST_TRACEPOINT_FIELDS:
3171 {
3172 struct lttng_event_field *fields;
3173 ssize_t nb_fields;
3174
a5b55f38 3175 session_lock_list();
a4b92340
DG
3176 nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type,
3177 &fields);
a5b55f38 3178 session_unlock_list();
f37d259d 3179 if (nb_fields < 0) {
f73fabfd 3180 /* Return value is a negative lttng_error_code. */
f37d259d
MD
3181 ret = -nb_fields;
3182 goto error;
3183 }
3184
3185 /*
3186 * Setup lttng message with payload size set to the event list size in
3187 * bytes and then copy list into the llm payload.
3188 */
a4b92340
DG
3189 ret = setup_lttng_msg(cmd_ctx,
3190 sizeof(struct lttng_event_field) * nb_fields);
f37d259d
MD
3191 if (ret < 0) {
3192 free(fields);
3193 goto setup_error;
3194 }
3195
3196 /* Copy event list into message payload */
3197 memcpy(cmd_ctx->llm->payload, fields,
3198 sizeof(struct lttng_event_field) * nb_fields);
3199
3200 free(fields);
3201
f73fabfd 3202 ret = LTTNG_OK;
f37d259d
MD
3203 break;
3204 }
00e2e675
DG
3205 case LTTNG_SET_CONSUMER_URI:
3206 {
a4b92340
DG
3207 size_t nb_uri, len;
3208 struct lttng_uri *uris;
3209
3210 nb_uri = cmd_ctx->lsm->u.uri.size;
3211 len = nb_uri * sizeof(struct lttng_uri);
3212
3213 if (nb_uri == 0) {
f73fabfd 3214 ret = LTTNG_ERR_INVALID;
a4b92340
DG
3215 goto error;
3216 }
3217
3218 uris = zmalloc(len);
3219 if (uris == NULL) {
f73fabfd 3220 ret = LTTNG_ERR_FATAL;
a4b92340
DG
3221 goto error;
3222 }
3223
3224 /* Receive variable len data */
77c7c900 3225 DBG("Receiving %zu URI(s) from client ...", nb_uri);
a4b92340
DG
3226 ret = lttcomm_recv_unix_sock(sock, uris, len);
3227 if (ret <= 0) {
3228 DBG("No URIs received from client... continuing");
3229 *sock_error = 1;
f73fabfd 3230 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3231 free(uris);
a4b92340
DG
3232 goto error;
3233 }
3234
00e2e675 3235 ret = cmd_set_consumer_uri(cmd_ctx->lsm->domain.type, cmd_ctx->session,
a4b92340 3236 nb_uri, uris);
f73fabfd 3237 if (ret != LTTNG_OK) {
b1d41407 3238 free(uris);
a4b92340
DG
3239 goto error;
3240 }
3241
3242 /*
3243 * XXX: 0 means that this URI should be applied on the session. Should
3244 * be a DOMAIN enuam.
3245 */
3246 if (cmd_ctx->lsm->domain.type == 0) {
3247 /* Add the URI for the UST session if a consumer is present. */
3248 if (cmd_ctx->session->ust_session &&
3249 cmd_ctx->session->ust_session->consumer) {
3250 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_UST, cmd_ctx->session,
3251 nb_uri, uris);
3252 } else if (cmd_ctx->session->kernel_session &&
3253 cmd_ctx->session->kernel_session->consumer) {
3254 ret = cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL,
3255 cmd_ctx->session, nb_uri, uris);
3256 }
3257 }
3258
b1d41407
DG
3259 free(uris);
3260
00e2e675
DG
3261 break;
3262 }
f3ed775e 3263 case LTTNG_START_TRACE:
8c0faa1d 3264 {
54d01ffb 3265 ret = cmd_start_trace(cmd_ctx->session);
8c0faa1d
DG
3266 break;
3267 }
f3ed775e 3268 case LTTNG_STOP_TRACE:
8c0faa1d 3269 {
54d01ffb 3270 ret = cmd_stop_trace(cmd_ctx->session);
8c0faa1d
DG
3271 break;
3272 }
5e16da05
MD
3273 case LTTNG_CREATE_SESSION:
3274 {
a4b92340
DG
3275 size_t nb_uri, len;
3276 struct lttng_uri *uris = NULL;
3277
3278 nb_uri = cmd_ctx->lsm->u.uri.size;
3279 len = nb_uri * sizeof(struct lttng_uri);
3280
3281 if (nb_uri > 0) {
3282 uris = zmalloc(len);
3283 if (uris == NULL) {
f73fabfd 3284 ret = LTTNG_ERR_FATAL;
a4b92340
DG
3285 goto error;
3286 }
3287
3288 /* Receive variable len data */
77c7c900 3289 DBG("Waiting for %zu URIs from client ...", nb_uri);
a4b92340
DG
3290 ret = lttcomm_recv_unix_sock(sock, uris, len);
3291 if (ret <= 0) {
3292 DBG("No URIs received from client... continuing");
3293 *sock_error = 1;
f73fabfd 3294 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3295 free(uris);
a4b92340
DG
3296 goto error;
3297 }
3298
3299 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3300 DBG("Creating session with ONE network URI is a bad call");
f73fabfd 3301 ret = LTTNG_ERR_SESSION_FAIL;
b1d41407 3302 free(uris);
a4b92340
DG
3303 goto error;
3304 }
3305 }
3306
3307 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
ecc48a90 3308 &cmd_ctx->creds, 0);
a4b92340 3309
b1d41407
DG
3310 free(uris);
3311
00e2e675
DG
3312 break;
3313 }
5e16da05
MD
3314 case LTTNG_DESTROY_SESSION:
3315 {
2f77fc4b 3316 ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
a4b92340
DG
3317
3318 /* Set session to NULL so we do not unlock it after free. */
256a5576 3319 cmd_ctx->session = NULL;
5461b305 3320 break;
5e16da05 3321 }
9f19cc17 3322 case LTTNG_LIST_DOMAINS:
5e16da05 3323 {
54d01ffb
DG
3324 ssize_t nb_dom;
3325 struct lttng_domain *domains;
5461b305 3326
54d01ffb
DG
3327 nb_dom = cmd_list_domains(cmd_ctx->session, &domains);
3328 if (nb_dom < 0) {
f73fabfd 3329 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3330 ret = -nb_dom;
3331 goto error;
ce3d728c 3332 }
520ff687 3333
54d01ffb 3334 ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain));
5461b305 3335 if (ret < 0) {
ae9c20bc 3336 free(domains);
5461b305 3337 goto setup_error;
520ff687 3338 }
57167058 3339
54d01ffb
DG
3340 /* Copy event list into message payload */
3341 memcpy(cmd_ctx->llm->payload, domains,
3342 nb_dom * sizeof(struct lttng_domain));
3343
3344 free(domains);
5e16da05 3345
f73fabfd 3346 ret = LTTNG_OK;
5e16da05
MD
3347 break;
3348 }
9f19cc17 3349 case LTTNG_LIST_CHANNELS:
5e16da05 3350 {
6775595e 3351 int nb_chan;
22aca9f6 3352 struct lttng_channel *channels = NULL;
54d01ffb 3353
b551a063
DG
3354 nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type,
3355 cmd_ctx->session, &channels);
54d01ffb 3356 if (nb_chan < 0) {
f73fabfd 3357 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3358 ret = -nb_chan;
3359 goto error;
5461b305 3360 }
ca95a216 3361
54d01ffb 3362 ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel));
5461b305 3363 if (ret < 0) {
ae9c20bc 3364 free(channels);
5461b305
DG
3365 goto setup_error;
3366 }
9f19cc17 3367
54d01ffb
DG
3368 /* Copy event list into message payload */
3369 memcpy(cmd_ctx->llm->payload, channels,
3370 nb_chan * sizeof(struct lttng_channel));
3371
3372 free(channels);
9f19cc17 3373
f73fabfd 3374 ret = LTTNG_OK;
5461b305 3375 break;
5e16da05 3376 }
9f19cc17 3377 case LTTNG_LIST_EVENTS:
5e16da05 3378 {
b551a063 3379 ssize_t nb_event;
684d34d2 3380 struct lttng_event *events = NULL;
9f19cc17 3381
b551a063 3382 nb_event = cmd_list_events(cmd_ctx->lsm->domain.type, cmd_ctx->session,
54d01ffb
DG
3383 cmd_ctx->lsm->u.list.channel_name, &events);
3384 if (nb_event < 0) {
f73fabfd 3385 /* Return value is a negative lttng_error_code. */
54d01ffb
DG
3386 ret = -nb_event;
3387 goto error;
5461b305 3388 }
ca95a216 3389
54d01ffb 3390 ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event));
5461b305 3391 if (ret < 0) {
ae9c20bc 3392 free(events);
5461b305
DG
3393 goto setup_error;
3394 }
9f19cc17 3395
54d01ffb
DG
3396 /* Copy event list into message payload */
3397 memcpy(cmd_ctx->llm->payload, events,
3398 nb_event * sizeof(struct lttng_event));
9f19cc17 3399
54d01ffb 3400 free(events);
9f19cc17 3401
f73fabfd 3402 ret = LTTNG_OK;
5461b305 3403 break;
5e16da05
MD
3404 }
3405 case LTTNG_LIST_SESSIONS:
3406 {
8e0af1b4 3407 unsigned int nr_sessions;
5461b305 3408
8e0af1b4 3409 session_lock_list();
730389d9
DG
3410 nr_sessions = lttng_sessions_count(
3411 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3412 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
d32fb093 3413
8e0af1b4 3414 ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_session) * nr_sessions);
5461b305 3415 if (ret < 0) {
54d01ffb 3416 session_unlock_list();
5461b305 3417 goto setup_error;
e065084a 3418 }
5e16da05 3419
6c9cc2ab 3420 /* Filled the session array */
2f77fc4b 3421 cmd_list_lttng_sessions((struct lttng_session *)(cmd_ctx->llm->payload),
730389d9
DG
3422 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds),
3423 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds));
6c9cc2ab 3424
54d01ffb 3425 session_unlock_list();
5e16da05 3426
f73fabfd 3427 ret = LTTNG_OK;
5e16da05
MD
3428 break;
3429 }
d0254c7c
MD
3430 case LTTNG_CALIBRATE:
3431 {
54d01ffb
DG
3432 ret = cmd_calibrate(cmd_ctx->lsm->domain.type,
3433 &cmd_ctx->lsm->u.calibrate);
d0254c7c
MD
3434 break;
3435 }
d9800920
DG
3436 case LTTNG_REGISTER_CONSUMER:
3437 {
2f77fc4b
DG
3438 struct consumer_data *cdata;
3439
3440 switch (cmd_ctx->lsm->domain.type) {
3441 case LTTNG_DOMAIN_KERNEL:
3442 cdata = &kconsumer_data;
3443 break;
3444 default:
f73fabfd 3445 ret = LTTNG_ERR_UND;
2f77fc4b
DG
3446 goto error;
3447 }
3448
54d01ffb 3449 ret = cmd_register_consumer(cmd_ctx->session, cmd_ctx->lsm->domain.type,
2f77fc4b 3450 cmd_ctx->lsm->u.reg.path, cdata);
d9800920
DG
3451 break;
3452 }
6d805429 3453 case LTTNG_DATA_PENDING:
806e2684 3454 {
6d805429 3455 ret = cmd_data_pending(cmd_ctx->session);
806e2684
DG
3456 break;
3457 }
da3c9ec1
DG
3458 case LTTNG_SNAPSHOT_ADD_OUTPUT:
3459 {
6dc3064a
DG
3460 struct lttcomm_lttng_output_id reply;
3461
3462 ret = cmd_snapshot_add_output(cmd_ctx->session,
3463 &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
3464 if (ret != LTTNG_OK) {
3465 goto error;
3466 }
3467
3468 ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
3469 if (ret < 0) {
3470 goto setup_error;
3471 }
3472
3473 /* Copy output list into message payload */
3474 memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
3475 ret = LTTNG_OK;
da3c9ec1
DG
3476 break;
3477 }
3478 case LTTNG_SNAPSHOT_DEL_OUTPUT:
3479 {
6dc3064a
DG
3480 ret = cmd_snapshot_del_output(cmd_ctx->session,
3481 &cmd_ctx->lsm->u.snapshot_output.output);
da3c9ec1
DG
3482 break;
3483 }
3484 case LTTNG_SNAPSHOT_LIST_OUTPUT:
3485 {
6dc3064a
DG
3486 ssize_t nb_output;
3487 struct lttng_snapshot_output *outputs = NULL;
3488
3489 nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
3490 if (nb_output < 0) {
3491 ret = -nb_output;
3492 goto error;
3493 }
3494
3495 ret = setup_lttng_msg(cmd_ctx,
3496 nb_output * sizeof(struct lttng_snapshot_output));
3497 if (ret < 0) {
3498 free(outputs);
3499 goto setup_error;
3500 }
3501
3502 if (outputs) {
3503 /* Copy output list into message payload */
3504 memcpy(cmd_ctx->llm->payload, outputs,
3505 nb_output * sizeof(struct lttng_snapshot_output));
3506 free(outputs);
3507 }
3508
3509 ret = LTTNG_OK;
da3c9ec1
DG
3510 break;
3511 }
3512 case LTTNG_SNAPSHOT_RECORD:
3513 {
6dc3064a
DG
3514 ret = cmd_snapshot_record(cmd_ctx->session,
3515 &cmd_ctx->lsm->u.snapshot_record.output,
3516 cmd_ctx->lsm->u.snapshot_record.wait);
da3c9ec1
DG
3517 break;
3518 }
27babd3a
DG
3519 case LTTNG_CREATE_SESSION_SNAPSHOT:
3520 {
3521 size_t nb_uri, len;
3522 struct lttng_uri *uris = NULL;
3523
3524 nb_uri = cmd_ctx->lsm->u.uri.size;
3525 len = nb_uri * sizeof(struct lttng_uri);
3526
3527 if (nb_uri > 0) {
3528 uris = zmalloc(len);
3529 if (uris == NULL) {
3530 ret = LTTNG_ERR_FATAL;
3531 goto error;
3532 }
3533
3534 /* Receive variable len data */
3535 DBG("Waiting for %zu URIs from client ...", nb_uri);
3536 ret = lttcomm_recv_unix_sock(sock, uris, len);
3537 if (ret <= 0) {
3538 DBG("No URIs received from client... continuing");
3539 *sock_error = 1;
3540 ret = LTTNG_ERR_SESSION_FAIL;
3541 free(uris);
3542 goto error;
3543 }
3544
3545 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3546 DBG("Creating session with ONE network URI is a bad call");
3547 ret = LTTNG_ERR_SESSION_FAIL;
3548 free(uris);
3549 goto error;
3550 }
3551 }
3552
3553 ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris,
3554 nb_uri, &cmd_ctx->creds);
3555 free(uris);
3556 break;
3557 }
ecc48a90
JD
3558 case LTTNG_CREATE_SESSION_LIVE:
3559 {
3560 size_t nb_uri, len;
3561 struct lttng_uri *uris = NULL;
3562
3563 nb_uri = cmd_ctx->lsm->u.uri.size;
3564 len = nb_uri * sizeof(struct lttng_uri);
3565
3566 if (nb_uri > 0) {
3567 uris = zmalloc(len);
3568 if (uris == NULL) {
3569 ret = LTTNG_ERR_FATAL;
3570 goto error;
3571 }
3572
3573 /* Receive variable len data */
3574 DBG("Waiting for %zu URIs from client ...", nb_uri);
3575 ret = lttcomm_recv_unix_sock(sock, uris, len);
3576 if (ret <= 0) {
3577 DBG("No URIs received from client... continuing");
3578 *sock_error = 1;
3579 ret = LTTNG_ERR_SESSION_FAIL;
3580 free(uris);
3581 goto error;
3582 }
3583
3584 if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
3585 DBG("Creating session with ONE network URI is a bad call");
3586 ret = LTTNG_ERR_SESSION_FAIL;
3587 free(uris);
3588 goto error;
3589 }
3590 }
3591
3592 ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
3593 nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
3594 free(uris);
3595 break;
3596 }
5e16da05 3597 default:
f73fabfd 3598 ret = LTTNG_ERR_UND;
5461b305 3599 break;
fac6795d
DG
3600 }
3601
5461b305 3602error:
5461b305
DG
3603 if (cmd_ctx->llm == NULL) {
3604 DBG("Missing llm structure. Allocating one.");
894be886 3605 if (setup_lttng_msg(cmd_ctx, 0) < 0) {
5461b305
DG
3606 goto setup_error;
3607 }
3608 }
54d01ffb 3609 /* Set return code */
5461b305 3610 cmd_ctx->llm->ret_code = ret;
5461b305 3611setup_error:
b5541356 3612 if (cmd_ctx->session) {
54d01ffb 3613 session_unlock(cmd_ctx->session);
b5541356 3614 }
256a5576
MD
3615 if (need_tracing_session) {
3616 session_unlock_list();
3617 }
54d01ffb 3618init_setup_error:
8028d920 3619 return ret;
fac6795d
DG
3620}
3621
44a5e5eb
DG
3622/*
3623 * Thread managing health check socket.
3624 */
3625static void *thread_manage_health(void *data)
3626{
eb4a2943 3627 int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
44a5e5eb
DG
3628 uint32_t revents, nb_fd;
3629 struct lttng_poll_event events;
0c89d795
MD
3630 struct health_comm_msg msg;
3631 struct health_comm_reply reply;
44a5e5eb
DG
3632
3633 DBG("[thread] Manage health check started");
3634
3635 rcu_register_thread();
3636
6d737ce4
DG
3637 /* We might hit an error path before this is created. */
3638 lttng_poll_init(&events);
3cc04881 3639
44a5e5eb
DG
3640 /* Create unix socket */
3641 sock = lttcomm_create_unix_sock(health_unix_sock_path);
3642 if (sock < 0) {
3643 ERR("Unable to create health check Unix socket");
3644 ret = -1;
3645 goto error;
3646 }
3647
6c71277b
MD
3648 if (is_root) {
3649 /* lttng health client socket path permissions */
3650 ret = chown(health_unix_sock_path, 0,
3651 utils_get_group_id(tracing_group_name));
3652 if (ret < 0) {
3653 ERR("Unable to set group on %s", health_unix_sock_path);
3654 PERROR("chown");
3655 ret = -1;
3656 goto error;
3657 }
3658
3659 ret = chmod(health_unix_sock_path,
3660 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
3661 if (ret < 0) {
3662 ERR("Unable to set permissions on %s", health_unix_sock_path);
3663 PERROR("chmod");
3664 ret = -1;
3665 goto error;
3666 }
3667 }
3668
b662582b
DG
3669 /*
3670 * Set the CLOEXEC flag. Return code is useless because either way, the
3671 * show must go on.
3672 */
3673 (void) utils_set_fd_cloexec(sock);
3674
44a5e5eb
DG
3675 ret = lttcomm_listen_unix_sock(sock);
3676 if (ret < 0) {
3677 goto error;
3678 }
3679
3680 /*
3681 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3682 * more will be added to this poll set.
3683 */
d0b96690 3684 ret = sessiond_set_thread_pollset(&events, 2);
44a5e5eb
DG
3685 if (ret < 0) {
3686 goto error;
3687 }
3688
3689 /* Add the application registration socket */
3690 ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
3691 if (ret < 0) {
3692 goto error;
3693 }
3694
56045742
MD
3695 lttng_sessiond_notify_ready();
3696
44a5e5eb
DG
3697 while (1) {
3698 DBG("Health check ready");
3699
44a5e5eb
DG
3700 /* Inifinite blocking call, waiting for transmission */
3701restart:
3702 ret = lttng_poll_wait(&events, -1);
3703 if (ret < 0) {
3704 /*
3705 * Restart interrupted system call.
3706 */
3707 if (errno == EINTR) {
3708 goto restart;
3709 }
3710 goto error;
3711 }
3712
0d9c5d77
DG
3713 nb_fd = ret;
3714
44a5e5eb
DG
3715 for (i = 0; i < nb_fd; i++) {
3716 /* Fetch once the poll data */
3717 revents = LTTNG_POLL_GETEV(&events, i);
3718 pollfd = LTTNG_POLL_GETFD(&events, i);
3719
f0567343
MD
3720 if (!revents) {
3721 /* No activity for this FD (poll implementation). */
3722 continue;
3723 }
3724
44a5e5eb 3725 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 3726 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
44a5e5eb 3727 if (ret) {
139ac872
MD
3728 err = 0;
3729 goto exit;
44a5e5eb
DG
3730 }
3731
3732 /* Event on the registration socket */
3733 if (pollfd == sock) {
3734 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3735 ERR("Health socket poll error");
3736 goto error;
3737 }
3738 }
3739 }
3740
3741 new_sock = lttcomm_accept_unix_sock(sock);
3742 if (new_sock < 0) {
3743 goto error;
3744 }
3745
b662582b
DG
3746 /*
3747 * Set the CLOEXEC flag. Return code is useless because either way, the
3748 * show must go on.
3749 */
3750 (void) utils_set_fd_cloexec(new_sock);
3751
44a5e5eb
DG
3752 DBG("Receiving data from client for health...");
3753 ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
3754 if (ret <= 0) {
3755 DBG("Nothing recv() from client... continuing");
3756 ret = close(new_sock);
3757 if (ret) {
3758 PERROR("close");
3759 }
3760 new_sock = -1;
3761 continue;
3762 }
3763
3764 rcu_thread_online();
3765
239f3aec 3766 memset(&reply, 0, sizeof(reply));
6c71277b
MD
3767 for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) {
3768 /*
3769 * health_check_state returns 0 if health is
3770 * bad.
3771 */
3772 if (!health_check_state(health_sessiond, i)) {
3773 reply.ret_code |= 1ULL << i;
3774 }
44a5e5eb
DG
3775 }
3776
6c71277b 3777 DBG2("Health check return value %" PRIx64, reply.ret_code);
44a5e5eb
DG
3778
3779 ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
3780 if (ret < 0) {
3781 ERR("Failed to send health data back to client");
3782 }
3783
3784 /* End of transmission */
3785 ret = close(new_sock);
3786 if (ret) {
3787 PERROR("close");
3788 }
3789 new_sock = -1;
3790 }
3791
139ac872 3792exit:
44a5e5eb 3793error:
139ac872
MD
3794 if (err) {
3795 ERR("Health error occurred in %s", __func__);
3796 }
44a5e5eb
DG
3797 DBG("Health check thread dying");
3798 unlink(health_unix_sock_path);
3799 if (sock >= 0) {
3800 ret = close(sock);
3801 if (ret) {
3802 PERROR("close");
3803 }
3804 }
44a5e5eb
DG
3805
3806 lttng_poll_clean(&events);
3807
3808 rcu_unregister_thread();
3809 return NULL;
3810}
3811
1d4b027a 3812/*
d063d709
DG
3813 * This thread manage all clients request using the unix client socket for
3814 * communication.
1d4b027a
DG
3815 */
3816static void *thread_manage_clients(void *data)
3817{
139ac872 3818 int sock = -1, ret, i, pollfd, err = -1;
53a80697 3819 int sock_error;
5eb91c98 3820 uint32_t revents, nb_fd;
273ea72c 3821 struct command_ctx *cmd_ctx = NULL;
5eb91c98 3822 struct lttng_poll_event events;
1d4b027a
DG
3823
3824 DBG("[thread] Manage client started");
3825
f6a9efaa
DG
3826 rcu_register_thread();
3827
6c71277b 3828 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD);
927ca06a 3829
840cb59c 3830 health_code_update();
44a5e5eb 3831
1d4b027a
DG
3832 ret = lttcomm_listen_unix_sock(client_sock);
3833 if (ret < 0) {
35df2755 3834 goto error_listen;
1d4b027a
DG
3835 }
3836
5eb91c98
DG
3837 /*
3838 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3839 * more will be added to this poll set.
3840 */
d0b96690 3841 ret = sessiond_set_thread_pollset(&events, 2);
5eb91c98 3842 if (ret < 0) {
35df2755 3843 goto error_create_poll;
5eb91c98 3844 }
273ea72c 3845
5eb91c98
DG
3846 /* Add the application registration socket */
3847 ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI);
3848 if (ret < 0) {
3849 goto error;
3850 }
273ea72c 3851
56045742 3852 lttng_sessiond_notify_ready();
0bb7724a 3853
62c131b7
MD
3854 /* This testpoint is after we signal readiness to the parent. */
3855 if (testpoint(sessiond_thread_manage_clients)) {
3856 goto error;
3857 }
3858
71ddb4ed 3859 if (testpoint(sessiond_thread_manage_clients_before_loop)) {
6993eeb3
CB
3860 goto error;
3861 }
8ac94142 3862
840cb59c 3863 health_code_update();
44a5e5eb 3864
1d4b027a 3865 while (1) {
1d4b027a 3866 DBG("Accepting client command ...");
273ea72c
DG
3867
3868 /* Inifinite blocking call, waiting for transmission */
88f2b785 3869 restart:
a78af745 3870 health_poll_entry();
5eb91c98 3871 ret = lttng_poll_wait(&events, -1);
a78af745 3872 health_poll_exit();
273ea72c 3873 if (ret < 0) {
88f2b785
MD
3874 /*
3875 * Restart interrupted system call.
3876 */
3877 if (errno == EINTR) {
3878 goto restart;
3879 }
273ea72c
DG
3880 goto error;
3881 }
3882
0d9c5d77
DG
3883 nb_fd = ret;
3884
5eb91c98
DG
3885 for (i = 0; i < nb_fd; i++) {
3886 /* Fetch once the poll data */
3887 revents = LTTNG_POLL_GETEV(&events, i);
3888 pollfd = LTTNG_POLL_GETFD(&events, i);
3889
840cb59c 3890 health_code_update();
44a5e5eb 3891
f0567343
MD
3892 if (!revents) {
3893 /* No activity for this FD (poll implementation). */
3894 continue;
3895 }
3896
5eb91c98 3897 /* Thread quit pipe has been closed. Killing thread. */
d0b96690 3898 ret = sessiond_check_thread_quit_pipe(pollfd, revents);
5eb91c98 3899 if (ret) {
139ac872
MD
3900 err = 0;
3901 goto exit;
5eb91c98
DG
3902 }
3903
3904 /* Event on the registration socket */
3905 if (pollfd == client_sock) {
3906 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3907 ERR("Client socket poll error");
3908 goto error;
3909 }
3910 }
3911 }
3912
3913 DBG("Wait for client response");
3914
840cb59c 3915 health_code_update();
44a5e5eb 3916
5eb91c98
DG
3917 sock = lttcomm_accept_unix_sock(client_sock);
3918 if (sock < 0) {
273ea72c 3919 goto error;
273ea72c
DG
3920 }
3921
b662582b
DG
3922 /*
3923 * Set the CLOEXEC flag. Return code is useless because either way, the
3924 * show must go on.
3925 */
3926 (void) utils_set_fd_cloexec(sock);
3927
be040666
DG
3928 /* Set socket option for credentials retrieval */
3929 ret = lttcomm_setsockopt_creds_unix_sock(sock);
3930 if (ret < 0) {
3931 goto error;
3932 }
3933
5eb91c98 3934 /* Allocate context command to process the client request */
ba7f0ae5 3935 cmd_ctx = zmalloc(sizeof(struct command_ctx));
5eb91c98 3936 if (cmd_ctx == NULL) {
76d7553f 3937 PERROR("zmalloc cmd_ctx");
1d4b027a 3938 goto error;
5eb91c98 3939 }
1d4b027a 3940
5eb91c98 3941 /* Allocate data buffer for reception */
ba7f0ae5 3942 cmd_ctx->lsm = zmalloc(sizeof(struct lttcomm_session_msg));
5eb91c98 3943 if (cmd_ctx->lsm == NULL) {
76d7553f 3944 PERROR("zmalloc cmd_ctx->lsm");
5eb91c98
DG
3945 goto error;
3946 }
1d4b027a 3947
5eb91c98
DG
3948 cmd_ctx->llm = NULL;
3949 cmd_ctx->session = NULL;
1d4b027a 3950
840cb59c 3951 health_code_update();
44a5e5eb 3952
5eb91c98
DG
3953 /*
3954 * Data is received from the lttng client. The struct
3955 * lttcomm_session_msg (lsm) contains the command and data request of
3956 * the client.
3957 */
3958 DBG("Receiving data from client ...");
be040666
DG
3959 ret = lttcomm_recv_creds_unix_sock(sock, cmd_ctx->lsm,
3960 sizeof(struct lttcomm_session_msg), &cmd_ctx->creds);
5eb91c98
DG
3961 if (ret <= 0) {
3962 DBG("Nothing recv() from client... continuing");
76d7553f
MD
3963 ret = close(sock);
3964 if (ret) {
3965 PERROR("close");
3966 }
3967 sock = -1;
a2c0da86 3968 clean_command_ctx(&cmd_ctx);
5eb91c98
DG
3969 continue;
3970 }
1d4b027a 3971
840cb59c 3972 health_code_update();
44a5e5eb 3973
5eb91c98
DG
3974 // TODO: Validate cmd_ctx including sanity check for
3975 // security purpose.
f7776ea7 3976
f6a9efaa 3977 rcu_thread_online();
5eb91c98
DG
3978 /*
3979 * This function dispatch the work to the kernel or userspace tracer
3980 * libs and fill the lttcomm_lttng_msg data structure of all the needed
3981 * informations for the client. The command context struct contains
3982 * everything this function may needs.
3983 */
53a80697 3984 ret = process_client_msg(cmd_ctx, sock, &sock_error);
f6a9efaa 3985 rcu_thread_offline();
5eb91c98 3986 if (ret < 0) {
36134aa1
DG
3987 ret = close(sock);
3988 if (ret) {
3989 PERROR("close");
53a80697 3990 }
36134aa1 3991 sock = -1;
bbd973c2 3992 /*
5eb91c98 3993 * TODO: Inform client somehow of the fatal error. At
ba7f0ae5 3994 * this point, ret < 0 means that a zmalloc failed
53a80697
MD
3995 * (ENOMEM). Error detected but still accept
3996 * command, unless a socket error has been
3997 * detected.
bbd973c2 3998 */
bbd973c2 3999 clean_command_ctx(&cmd_ctx);
5eb91c98
DG
4000 continue;
4001 }
d6e4fca4 4002
840cb59c 4003 health_code_update();
44a5e5eb 4004
54d01ffb
DG
4005 DBG("Sending response (size: %d, retcode: %s)",
4006 cmd_ctx->lttng_msg_size,
9a745bc7 4007 lttng_strerror(-cmd_ctx->llm->ret_code));
54d01ffb 4008 ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size);
5eb91c98
DG
4009 if (ret < 0) {
4010 ERR("Failed to send data back to client");
bbd973c2 4011 }
1d4b027a 4012
5eb91c98 4013 /* End of transmission */
76d7553f
MD
4014 ret = close(sock);
4015 if (ret) {
4016 PERROR("close");
4017 }
4018 sock = -1;
be040666
DG
4019
4020 clean_command_ctx(&cmd_ctx);
44a5e5eb 4021
840cb59c 4022 health_code_update();
273ea72c
DG
4023 }
4024
139ac872 4025exit:
5eb91c98 4026error:
35df2755
CB
4027 if (sock >= 0) {
4028 ret = close(sock);
4029 if (ret) {
4030 PERROR("close");
4031 }
139ac872 4032 }
44a5e5eb 4033
35df2755
CB
4034 lttng_poll_clean(&events);
4035 clean_command_ctx(&cmd_ctx);
4036
4037error_listen:
4038error_create_poll:
273ea72c 4039 unlink(client_unix_sock_path);
76d7553f
MD
4040 if (client_sock >= 0) {
4041 ret = close(client_sock);
4042 if (ret) {
4043 PERROR("close");
4044 }
a4b35e07 4045 }
35df2755
CB
4046
4047 if (err) {
840cb59c 4048 health_error();
35df2755 4049 ERR("Health error occurred in %s", __func__);
a4b35e07 4050 }
273ea72c 4051
8782cc74 4052 health_unregister(health_sessiond);
35df2755
CB
4053
4054 DBG("Client thread dying");
f6a9efaa
DG
4055
4056 rcu_unregister_thread();
1d4b027a
DG
4057 return NULL;
4058}
4059
4060
fac6795d
DG
4061/*
4062 * usage function on stderr
4063 */
4064static void usage(void)
4065{
b716ce68 4066 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
d6f42150
DG
4067 fprintf(stderr, " -h, --help Display this usage.\n");
4068 fprintf(stderr, " -c, --client-sock PATH Specify path for the client unix socket\n");
4069 fprintf(stderr, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4070 fprintf(stderr, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4071 fprintf(stderr, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
7753dea8
MD
4072 fprintf(stderr, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4073 fprintf(stderr, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4074 fprintf(stderr, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4075 fprintf(stderr, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
ebaeda94
MD
4076 fprintf(stderr, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4077 fprintf(stderr, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4078 fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4079 fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
d6f42150 4080 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
929de64e 4081 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
d6f42150
DG
4082 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4083 fprintf(stderr, " -V, --version Show version number.\n");
7fe6d2c5 4084 fprintf(stderr, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
d6f42150
DG
4085 fprintf(stderr, " -q, --quiet No output at all.\n");
4086 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
35f90c40 4087 fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
3bd1e081 4088 fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4fba7219 4089 fprintf(stderr, " --no-kernel Disable kernel tracer\n");
4d076222 4090 fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n");
fac6795d
DG
4091}
4092
4093/*
4094 * daemon argument parsing
4095 */
4096static int parse_args(int argc, char **argv)
4097{
4098 int c;
4099
4100 static struct option long_options[] = {
4101 { "client-sock", 1, 0, 'c' },
4102 { "apps-sock", 1, 0, 'a' },
3bd1e081
MD
4103 { "kconsumerd-cmd-sock", 1, 0, 'C' },
4104 { "kconsumerd-err-sock", 1, 0, 'E' },
7753dea8
MD
4105 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
4106 { "ustconsumerd32-err-sock", 1, 0, 'H' },
ebaeda94
MD
4107 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
4108 { "ustconsumerd64-err-sock", 1, 0, 'F' },
4109 { "consumerd32-path", 1, 0, 'u' },
4110 { "consumerd32-libdir", 1, 0, 'U' },
4111 { "consumerd64-path", 1, 0, 't' },
4112 { "consumerd64-libdir", 1, 0, 'T' },
fac6795d 4113 { "daemonize", 0, 0, 'd' },
5b8719f5 4114 { "sig-parent", 0, 0, 'S' },
fac6795d
DG
4115 { "help", 0, 0, 'h' },
4116 { "group", 1, 0, 'g' },
4117 { "version", 0, 0, 'V' },
75462a81 4118 { "quiet", 0, 0, 'q' },
3f9947db 4119 { "verbose", 0, 0, 'v' },
3bd1e081 4120 { "verbose-consumer", 0, 0, 'Z' },
4fba7219 4121 { "no-kernel", 0, 0, 'N' },
35f90c40 4122 { "pidfile", 1, 0, 'p' },
4d076222 4123 { "jul-tcp-port", 1, 0, 'J' },
06b55dbb 4124 { "background", 0, 0, 'b' },
fac6795d
DG
4125 { NULL, 0, 0, 0 }
4126 };
4127
4128 while (1) {
4129 int option_index = 0;
06b55dbb 4130 c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:b",
54d01ffb 4131 long_options, &option_index);
fac6795d
DG
4132 if (c == -1) {
4133 break;
4134 }
4135
4136 switch (c) {
4137 case 0:
4138 fprintf(stderr, "option %s", long_options[option_index].name);
4139 if (optarg) {
4140 fprintf(stderr, " with arg %s\n", optarg);
4141 }
4142 break;
b716ce68 4143 case 'c':
fac6795d
DG
4144 snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg);
4145 break;
4146 case 'a':
4147 snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg);
4148 break;
4149 case 'd':
4150 opt_daemon = 1;
4151 break;
06b55dbb
DG
4152 case 'b':
4153 opt_background = 1;
4154 break;
fac6795d 4155 case 'g':
6c71277b 4156 tracing_group_name = optarg;
fac6795d
DG
4157 break;
4158 case 'h':
4159 usage();
4160 exit(EXIT_FAILURE);
4161 case 'V':
4162 fprintf(stdout, "%s\n", VERSION);
4163 exit(EXIT_SUCCESS);
5b8719f5
DG
4164 case 'S':
4165 opt_sig_parent = 1;
4166 break;
d6f42150 4167 case 'E':
3bd1e081 4168 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
d6f42150
DG
4169 break;
4170 case 'C':
3bd1e081
MD
4171 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4172 break;
4173 case 'F':
7753dea8 4174 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
3bd1e081
MD
4175 break;
4176 case 'D':
7753dea8
MD
4177 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
4178 break;
4179 case 'H':
4180 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg);
4181 break;
4182 case 'G':
4183 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg);
d6f42150 4184 break;
4fba7219
DG
4185 case 'N':
4186 opt_no_kernel = 1;
4187 break;
75462a81 4188 case 'q':
97e19046 4189 lttng_opt_quiet = 1;
75462a81 4190 break;
3f9947db 4191 case 'v':
53086306 4192 /* Verbose level can increase using multiple -v */
97e19046 4193 lttng_opt_verbose += 1;
3f9947db 4194 break;
31f73cc9 4195 case 'Z':
3bd1e081 4196 opt_verbose_consumer += 1;
31f73cc9 4197 break;
fb09408a 4198 case 'u':
fc7a59ce 4199 consumerd32_bin= optarg;
ebaeda94
MD
4200 break;
4201 case 'U':
4202 consumerd32_libdir = optarg;
7753dea8
MD
4203 break;
4204 case 't':
fc7a59ce 4205 consumerd64_bin = optarg;
ebaeda94
MD
4206 break;
4207 case 'T':
4208 consumerd64_libdir = optarg;
fb09408a 4209 break;
35f90c40
DG
4210 case 'p':
4211 opt_pidfile = optarg;
4212 break;
4d076222
DG
4213 case 'J': /* JUL TCP port. */
4214 {
4215 unsigned long v;
4216
4217 errno = 0;
4218 v = strtoul(optarg, NULL, 0);
4219 if (errno != 0 || !isdigit(optarg[0])) {
4220 ERR("Wrong value in --jul-tcp-port parameter: %s", optarg);
4221 return -1;
4222 }
4223 if (v == 0 || v >= 65535) {
4224 ERR("Port overflow in --jul-tcp-port parameter: %s", optarg);
4225 return -1;
4226 }
4227 jul_tcp_port = (uint32_t) v;
4228 DBG3("JUL TCP port set to non default: %u", jul_tcp_port);
4229 break;
4230 }
fac6795d
DG
4231 default:
4232 /* Unknown option or other error.
4233 * Error is printed by getopt, just return */
4234 return -1;
4235 }
4236 }
4237
4238 return 0;
4239}
4240
4241/*
d063d709 4242 * Creates the two needed socket by the daemon.
d6f42150
DG
4243 * apps_sock - The communication socket for all UST apps.
4244 * client_sock - The communication of the cli tool (lttng).
fac6795d 4245 */
cf3af59e 4246static int init_daemon_socket(void)
fac6795d
DG
4247{
4248 int ret = 0;
4249 mode_t old_umask;
4250
4251 old_umask = umask(0);
4252
4253 /* Create client tool unix socket */
d6f42150
DG
4254 client_sock = lttcomm_create_unix_sock(client_unix_sock_path);
4255 if (client_sock < 0) {
4256 ERR("Create unix sock failed: %s", client_unix_sock_path);
fac6795d
DG
4257 ret = -1;
4258 goto end;
4259 }
4260
b662582b
DG
4261 /* Set the cloexec flag */
4262 ret = utils_set_fd_cloexec(client_sock);
4263 if (ret < 0) {
4264 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4265 "Continuing but note that the consumer daemon will have a "
4266 "reference to this socket on exec()", client_sock);
4267 }
4268
fac6795d
DG
4269 /* File permission MUST be 660 */
4270 ret = chmod(client_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
4271 if (ret < 0) {
d6f42150 4272 ERR("Set file permissions failed: %s", client_unix_sock_path);
76d7553f 4273 PERROR("chmod");
fac6795d
DG
4274 goto end;
4275 }
4276
4277 /* Create the application unix socket */
d6f42150
DG
4278 apps_sock = lttcomm_create_unix_sock(apps_unix_sock_path);
4279 if (apps_sock < 0) {
4280 ERR("Create unix sock failed: %s", apps_unix_sock_path);
fac6795d
DG
4281 ret = -1;
4282 goto end;
4283 }
4284
b662582b
DG
4285 /* Set the cloexec flag */
4286 ret = utils_set_fd_cloexec(apps_sock);
4287 if (ret < 0) {
4288 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4289 "Continuing but note that the consumer daemon will have a "
4290 "reference to this socket on exec()", apps_sock);
4291 }
4292
d6f42150 4293 /* File permission MUST be 666 */
54d01ffb
DG
4294 ret = chmod(apps_unix_sock_path,
4295 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
fac6795d 4296 if (ret < 0) {
d6f42150 4297 ERR("Set file permissions failed: %s", apps_unix_sock_path);
76d7553f 4298 PERROR("chmod");
fac6795d
DG
4299 goto end;
4300 }
4301
b662582b
DG
4302 DBG3("Session daemon client socket %d and application socket %d created",
4303 client_sock, apps_sock);
4304
fac6795d
DG
4305end:
4306 umask(old_umask);
4307 return ret;
4308}
4309
4310/*
54d01ffb
DG
4311 * Check if the global socket is available, and if a daemon is answering at the
4312 * other side. If yes, error is returned.
fac6795d 4313 */
cf3af59e 4314static int check_existing_daemon(void)
fac6795d 4315{
7d8234d9 4316 /* Is there anybody out there ? */
099e26bd 4317 if (lttng_session_daemon_alive()) {
7d8234d9 4318 return -EEXIST;
099e26bd 4319 }
b09c7c76
DG
4320
4321 return 0;
fac6795d
DG
4322}
4323
fac6795d 4324/*
d063d709 4325 * Set the tracing group gid onto the client socket.
5e16da05 4326 *
d063d709 4327 * Race window between mkdir and chown is OK because we are going from more
d1613cf5 4328 * permissive (root.root) to less permissive (root.tracing).
fac6795d 4329 */
be040666 4330static int set_permissions(char *rundir)
fac6795d
DG
4331{
4332 int ret;
996b65c8 4333 gid_t gid;
fac6795d 4334
6c71277b 4335 gid = utils_get_group_id(tracing_group_name);
fac6795d 4336
d6f42150 4337 /* Set lttng run dir */
be040666 4338 ret = chown(rundir, 0, gid);
d6f42150 4339 if (ret < 0) {
be040666 4340 ERR("Unable to set group on %s", rundir);
76d7553f 4341 PERROR("chown");
d6f42150
DG
4342 }
4343
6c71277b
MD
4344 /*
4345 * Ensure all applications and tracing group can search the run
4346 * dir. Allow everyone to read the directory, since it does not
4347 * buy us anything to hide its content.
4348 */
4349 ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
d1613cf5
JN
4350 if (ret < 0) {
4351 ERR("Unable to set permissions on %s", rundir);
76d7553f 4352 PERROR("chmod");
d1613cf5
JN
4353 }
4354
d6f42150 4355 /* lttng client socket path */
996b65c8 4356 ret = chown(client_unix_sock_path, 0, gid);
fac6795d 4357 if (ret < 0) {
d6f42150 4358 ERR("Unable to set group on %s", client_unix_sock_path);
76d7553f 4359 PERROR("chown");
d6f42150
DG
4360 }
4361
3bd1e081 4362 /* kconsumer error socket path */
6c71277b 4363 ret = chown(kconsumer_data.err_unix_sock_path, 0, 0);
d6f42150 4364 if (ret < 0) {
3bd1e081 4365 ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path);
76d7553f 4366 PERROR("chown");
3bd1e081
MD
4367 }
4368
7753dea8 4369 /* 64-bit ustconsumer error socket path */
6c71277b 4370 ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0);
7753dea8
MD
4371 if (ret < 0) {
4372 ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path);
76d7553f 4373 PERROR("chown");
7753dea8
MD
4374 }
4375
4376 /* 32-bit ustconsumer compat32 error socket path */
6c71277b 4377 ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0);
3bd1e081 4378 if (ret < 0) {
7753dea8 4379 ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path);
76d7553f 4380 PERROR("chown");
fac6795d
DG
4381 }
4382
d6f42150 4383 DBG("All permissions are set");
e07ae692 4384
fac6795d
DG
4385 return ret;
4386}
4387
d6f42150 4388/*
d063d709 4389 * Create the lttng run directory needed for all global sockets and pipe.
d6f42150 4390 */
67e40797 4391static int create_lttng_rundir(const char *rundir)
d6f42150
DG
4392{
4393 int ret;
4394
67e40797
DG
4395 DBG3("Creating LTTng run directory: %s", rundir);
4396
d1613cf5 4397 ret = mkdir(rundir, S_IRWXU);
d6f42150 4398 if (ret < 0) {
b1f11e69 4399 if (errno != EEXIST) {
67e40797 4400 ERR("Unable to create %s", rundir);
b1f11e69
DG
4401 goto error;
4402 } else {
4403 ret = 0;
4404 }
d6f42150
DG
4405 }
4406
4407error:
4408 return ret;
4409}
4410
4411/*
d063d709
DG
4412 * Setup sockets and directory needed by the kconsumerd communication with the
4413 * session daemon.
d6f42150 4414 */
67e40797
DG
4415static int set_consumer_sockets(struct consumer_data *consumer_data,
4416 const char *rundir)
d6f42150
DG
4417{
4418 int ret;
67e40797 4419 char path[PATH_MAX];
d6f42150 4420
6c71277b 4421 switch (consumer_data->type) {
7753dea8 4422 case LTTNG_CONSUMER_KERNEL:
60922cb0 4423 snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir);
7753dea8
MD
4424 break;
4425 case LTTNG_CONSUMER64_UST:
60922cb0 4426 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD64_PATH, rundir);
7753dea8
MD
4427 break;
4428 case LTTNG_CONSUMER32_UST:
60922cb0 4429 snprintf(path, PATH_MAX, DEFAULT_USTCONSUMERD32_PATH, rundir);
7753dea8
MD
4430 break;
4431 default:
4432 ERR("Consumer type unknown");
4433 ret = -EINVAL;
4434 goto error;
d6f42150
DG
4435 }
4436
67e40797
DG
4437 DBG2("Creating consumer directory: %s", path);
4438
6c71277b 4439 ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP);
d6f42150 4440 if (ret < 0) {
6beb2242 4441 if (errno != EEXIST) {
f11e84c2 4442 PERROR("mkdir");
3bd1e081 4443 ERR("Failed to create %s", path);
6beb2242
DG
4444 goto error;
4445 }
f11e84c2 4446 ret = -1;
d6f42150 4447 }
6c71277b
MD
4448 if (is_root) {
4449 ret = chown(path, 0, utils_get_group_id(tracing_group_name));
4450 if (ret < 0) {
4451 ERR("Unable to set group on %s", path);
4452 PERROR("chown");
4453 goto error;
4454 }
4455 }
d6f42150
DG
4456
4457 /* Create the kconsumerd error unix socket */
3bd1e081
MD
4458 consumer_data->err_sock =
4459 lttcomm_create_unix_sock(consumer_data->err_unix_sock_path);
4460 if (consumer_data->err_sock < 0) {
4461 ERR("Create unix sock failed: %s", consumer_data->err_unix_sock_path);
d6f42150
DG
4462 ret = -1;
4463 goto error;
4464 }
4465
a24f05ab
MD
4466 /*
4467 * Set the CLOEXEC flag. Return code is useless because either way, the
4468 * show must go on.
4469 */
4470 ret = utils_set_fd_cloexec(consumer_data->err_sock);
4471 if (ret < 0) {
4472 PERROR("utils_set_fd_cloexec");
4473 /* continue anyway */
4474 }
4475
d6f42150 4476 /* File permission MUST be 660 */
3bd1e081 4477 ret = chmod(consumer_data->err_unix_sock_path,
54d01ffb 4478 S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
d6f42150 4479 if (ret < 0) {
3bd1e081 4480 ERR("Set file permissions failed: %s", consumer_data->err_unix_sock_path);
67e40797 4481 PERROR("chmod");
d6f42150
DG
4482 goto error;
4483 }
4484
4485error:
4486 return ret;
4487}
4488
fac6795d 4489/*
d063d709 4490 * Signal handler for the daemon
cf3af59e 4491 *
54d01ffb
DG
4492 * Simply stop all worker threads, leaving main() return gracefully after
4493 * joining all threads and calling cleanup().
fac6795d
DG
4494 */
4495static void sighandler(int sig)
4496{
4497 switch (sig) {
cf3af59e 4498 case SIGPIPE:
af87c45a 4499 DBG("SIGPIPE caught");
cf3af59e
MD
4500 return;
4501 case SIGINT:
af87c45a 4502 DBG("SIGINT caught");
cf3af59e
MD
4503 stop_threads();
4504 break;
4505 case SIGTERM:
af87c45a 4506 DBG("SIGTERM caught");
cf3af59e
MD
4507 stop_threads();
4508 break;
0bb7724a
DG
4509 case SIGUSR1:
4510 CMM_STORE_SHARED(recv_child_signal, 1);
4511 break;
cf3af59e
MD
4512 default:
4513 break;
fac6795d 4514 }
fac6795d
DG
4515}
4516
4517/*
d063d709 4518 * Setup signal handler for :
1d4b027a 4519 * SIGINT, SIGTERM, SIGPIPE
fac6795d 4520 */
1d4b027a 4521static int set_signal_handler(void)
fac6795d 4522{
1d4b027a
DG
4523 int ret = 0;
4524 struct sigaction sa;
4525 sigset_t sigset;
fac6795d 4526
1d4b027a 4527 if ((ret = sigemptyset(&sigset)) < 0) {
76d7553f 4528 PERROR("sigemptyset");
1d4b027a
DG
4529 return ret;
4530 }
d6f42150 4531
1d4b027a
DG
4532 sa.sa_handler = sighandler;
4533 sa.sa_mask = sigset;
4534 sa.sa_flags = 0;
4535 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
76d7553f 4536 PERROR("sigaction");
1d4b027a 4537 return ret;
d6f42150
DG
4538 }
4539
1d4b027a 4540 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
76d7553f 4541 PERROR("sigaction");
1d4b027a 4542 return ret;
d6f42150 4543 }
aaf26714 4544
1d4b027a 4545 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
76d7553f 4546 PERROR("sigaction");
1d4b027a 4547 return ret;
8c0faa1d
DG
4548 }
4549
0bb7724a
DG
4550 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
4551 PERROR("sigaction");
4552 return ret;
4553 }
4554
4555 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
1d4b027a
DG
4556
4557 return ret;
fac6795d
DG
4558}
4559
f3ed775e 4560/*
d063d709
DG
4561 * Set open files limit to unlimited. This daemon can open a large number of
4562 * file descriptors in order to consumer multiple kernel traces.
f3ed775e
DG
4563 */
4564static void set_ulimit(void)
4565{
4566 int ret;
4567 struct rlimit lim;
4568
a88df331 4569 /* The kernel does not allowed an infinite limit for open files */
f3ed775e
DG
4570 lim.rlim_cur = 65535;
4571 lim.rlim_max = 65535;
4572
4573 ret = setrlimit(RLIMIT_NOFILE, &lim);
4574 if (ret < 0) {
76d7553f 4575 PERROR("failed to set open files limit");
f3ed775e
DG
4576 }
4577}
4578
35f90c40
DG
4579/*
4580 * Write pidfile using the rundir and opt_pidfile.
4581 */
4582static void write_pidfile(void)
4583{
4584 int ret;
4585 char pidfile_path[PATH_MAX];
4586
4587 assert(rundir);
4588
4589 if (opt_pidfile) {
4590 strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path));
4591 } else {
4592 /* Build pidfile path from rundir and opt_pidfile. */
4593 ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/"
4594 DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir);
4595 if (ret < 0) {
4596 PERROR("snprintf pidfile path");
4597 goto error;
4598 }
4599 }
4600
4601 /*
4602 * Create pid file in rundir. Return value is of no importance. The
4603 * execution will continue even though we are not able to write the file.
4604 */
4605 (void) utils_create_pid_file(getpid(), pidfile_path);
4606
4607error:
4608 return;
4609}
4610
52f740b8
JG
4611/*
4612 * Create lockfile using the rundir and return its fd.
4613 */
4614static int create_lockfile(void)
4615{
4616 int ret;
4617 char lockfile_path[PATH_MAX];
4618
4619 ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path));
4620 if (ret < 0) {
4621 goto error;
4622 }
4623
4624 ret = utils_create_lock_file(lockfile_path);
4625error:
4626 return ret;
4627}
4628
cd9290dd
DG
4629/*
4630 * Write JUL TCP port using the rundir.
4631 */
4632static void write_julport(void)
4633{
4634 int ret;
4635 char path[PATH_MAX];
4636
4637 assert(rundir);
4638
4639 ret = snprintf(path, sizeof(path), "%s/"
4640 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir);
4641 if (ret < 0) {
4642 PERROR("snprintf julport path");
4643 goto error;
4644 }
4645
4646 /*
4647 * Create TCP JUL port file in rundir. Return value is of no importance.
4648 * The execution will continue even though we are not able to write the
4649 * file.
4650 */
4651 (void) utils_create_pid_file(jul_tcp_port, path);
4652
4653error:
4654 return;
4655}
4656
fac6795d
DG
4657/*
4658 * main
4659 */
4660int main(int argc, char **argv)
4661{
fac6795d
DG
4662 int ret = 0;
4663 void *status;
ae9e45b3 4664 const char *home_path, *env_app_timeout;
fac6795d 4665
335a95b7
MD
4666 init_kernel_workarounds();
4667
f6a9efaa
DG
4668 rcu_register_thread();
4669
0bb7724a
DG
4670 if ((ret = set_signal_handler()) < 0) {
4671 goto error;
4672 }
4673
7753dea8 4674 setup_consumerd_path();
fb09408a 4675
12744796
DG
4676 page_size = sysconf(_SC_PAGESIZE);
4677 if (page_size < 0) {
4678 PERROR("sysconf _SC_PAGESIZE");
4679 page_size = LONG_MAX;
4680 WARN("Fallback page size to %ld", page_size);
4681 }
4682
fac6795d
DG
4683 /* Parse arguments */
4684 progname = argv[0];
c617c0c6 4685 if ((ret = parse_args(argc, argv)) < 0) {
cf3af59e 4686 goto error;
fac6795d
DG
4687 }
4688
4689 /* Daemonize */
929de64e 4690 if (opt_daemon || opt_background) {
ceed52b5
MD
4691 int i;
4692
929de64e
MD
4693 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
4694 !opt_background);
53094c05 4695 if (ret < 0) {
cf3af59e 4696 goto error;
53094c05 4697 }
0bb7724a 4698
ceed52b5 4699 /*
0bb7724a
DG
4700 * We are in the child. Make sure all other file descriptors are
4701 * closed, in case we are called with more opened file descriptors than
4702 * the standard ones.
ceed52b5
MD
4703 */
4704 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
4705 (void) close(i);
4706 }
4707 }
4708
4709 /* Create thread quit pipe */
4710 if ((ret = init_thread_quit_pipe()) < 0) {
4711 goto error;
fac6795d
DG
4712 }
4713
4714 /* Check if daemon is UID = 0 */
4715 is_root = !getuid();
4716
fac6795d 4717 if (is_root) {
990570ed 4718 rundir = strdup(DEFAULT_LTTNG_RUNDIR);
d3494c13
MD
4719 if (!rundir) {
4720 ret = -ENOMEM;
4721 goto error;
4722 }
67e40797
DG
4723
4724 /* Create global run dir with root access */
4725 ret = create_lttng_rundir(rundir);
d6f42150 4726 if (ret < 0) {
cf3af59e 4727 goto error;
d6f42150
DG
4728 }
4729
fac6795d 4730 if (strlen(apps_unix_sock_path) == 0) {
d6f42150
DG
4731 snprintf(apps_unix_sock_path, PATH_MAX,
4732 DEFAULT_GLOBAL_APPS_UNIX_SOCK);
fac6795d
DG
4733 }
4734
4735 if (strlen(client_unix_sock_path) == 0) {
d6f42150
DG
4736 snprintf(client_unix_sock_path, PATH_MAX,
4737 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK);
4738 }
0fdd1e2c
DG
4739
4740 /* Set global SHM for ust */
4741 if (strlen(wait_shm_path) == 0) {
4742 snprintf(wait_shm_path, PATH_MAX,
4743 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH);
4744 }
67e40797 4745
44a5e5eb
DG
4746 if (strlen(health_unix_sock_path) == 0) {
4747 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4748 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK);
4749 }
4750
67e40797
DG
4751 /* Setup kernel consumerd path */
4752 snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX,
60922cb0 4753 DEFAULT_KCONSUMERD_ERR_SOCK_PATH, rundir);
67e40797 4754 snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4755 DEFAULT_KCONSUMERD_CMD_SOCK_PATH, rundir);
67e40797
DG
4756
4757 DBG2("Kernel consumer err path: %s",
4758 kconsumer_data.err_unix_sock_path);
4759 DBG2("Kernel consumer cmd path: %s",
4760 kconsumer_data.cmd_unix_sock_path);
fac6795d 4761 } else {
feb0f3e5 4762 home_path = utils_get_home_dir();
b082db07 4763 if (home_path == NULL) {
273ea72c
DG
4764 /* TODO: Add --socket PATH option */
4765 ERR("Can't get HOME directory for sockets creation.");
cf3af59e
MD
4766 ret = -EPERM;
4767 goto error;
b082db07
DG
4768 }
4769
67e40797
DG
4770 /*
4771 * Create rundir from home path. This will create something like
4772 * $HOME/.lttng
4773 */
990570ed 4774 ret = asprintf(&rundir, DEFAULT_LTTNG_HOME_RUNDIR, home_path);
67e40797
DG
4775 if (ret < 0) {
4776 ret = -ENOMEM;
4777 goto error;
4778 }
4779
4780 ret = create_lttng_rundir(rundir);
4781 if (ret < 0) {
4782 goto error;
4783 }
4784
fac6795d 4785 if (strlen(apps_unix_sock_path) == 0) {
d6f42150 4786 snprintf(apps_unix_sock_path, PATH_MAX,
b082db07 4787 DEFAULT_HOME_APPS_UNIX_SOCK, home_path);
fac6795d
DG
4788 }
4789
4790 /* Set the cli tool unix socket path */
4791 if (strlen(client_unix_sock_path) == 0) {
d6f42150 4792 snprintf(client_unix_sock_path, PATH_MAX,
b082db07 4793 DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path);
fac6795d 4794 }
0fdd1e2c
DG
4795
4796 /* Set global SHM for ust */
4797 if (strlen(wait_shm_path) == 0) {
4798 snprintf(wait_shm_path, PATH_MAX,
d0b96690 4799 DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid());
0fdd1e2c 4800 }
44a5e5eb
DG
4801
4802 /* Set health check Unix path */
4803 if (strlen(health_unix_sock_path) == 0) {
4804 snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
4805 DEFAULT_HOME_HEALTH_UNIX_SOCK, home_path);
4806 }
fac6795d
DG
4807 }
4808
52f740b8
JG
4809 lockfile_fd = create_lockfile();
4810 if (lockfile_fd < 0) {
4811 goto error;
4812 }
4813
5c827ce0
DG
4814 /* Set consumer initial state */
4815 kernel_consumerd_state = CONSUMER_STOPPED;
4816 ust_consumerd_state = CONSUMER_STOPPED;
4817
847177cd
DG
4818 DBG("Client socket path %s", client_unix_sock_path);
4819 DBG("Application socket path %s", apps_unix_sock_path);
d0b96690 4820 DBG("Application wait path %s", wait_shm_path);
67e40797
DG
4821 DBG("LTTng run directory path: %s", rundir);
4822
4823 /* 32 bits consumerd path setup */
4824 snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX,
60922cb0 4825 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, rundir);
67e40797 4826 snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4827 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, rundir);
67e40797
DG
4828
4829 DBG2("UST consumer 32 bits err path: %s",
4830 ustconsumer32_data.err_unix_sock_path);
4831 DBG2("UST consumer 32 bits cmd path: %s",
4832 ustconsumer32_data.cmd_unix_sock_path);
4833
4834 /* 64 bits consumerd path setup */
4835 snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX,
60922cb0 4836 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, rundir);
67e40797 4837 snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX,
60922cb0 4838 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, rundir);
67e40797
DG
4839
4840 DBG2("UST consumer 64 bits err path: %s",
4841 ustconsumer64_data.err_unix_sock_path);
4842 DBG2("UST consumer 64 bits cmd path: %s",
4843 ustconsumer64_data.cmd_unix_sock_path);
847177cd 4844
273ea72c 4845 /*
7d8234d9 4846 * See if daemon already exist.
fac6795d 4847 */
7d8234d9 4848 if ((ret = check_existing_daemon()) < 0) {
75462a81 4849 ERR("Already running daemon.\n");
273ea72c 4850 /*
cf3af59e
MD
4851 * We do not goto exit because we must not cleanup()
4852 * because a daemon is already running.
ab118b20 4853 */
cf3af59e 4854 goto error;
a88df331
DG
4855 }
4856
1427f9b2
DG
4857 /*
4858 * Init UST app hash table. Alloc hash table before this point since
4859 * cleanup() can get called after that point.
4860 */
4861 ust_app_ht_alloc();
4862
f20baf8e
DG
4863 /* Initialize JUL domain subsystem. */
4864 if ((ret = jul_init()) < 0) {
4865 /* ENOMEM at this point. */
4866 goto error;
4867 }
4868
54d01ffb 4869 /* After this point, we can safely call cleanup() with "goto exit" */
a88df331
DG
4870
4871 /*
4872 * These actions must be executed as root. We do that *after* setting up
4873 * the sockets path because we MUST make the check for another daemon using
4874 * those paths *before* trying to set the kernel consumer sockets and init
4875 * kernel tracer.
4876 */
4877 if (is_root) {
67e40797 4878 ret = set_consumer_sockets(&kconsumer_data, rundir);
7753dea8
MD
4879 if (ret < 0) {
4880 goto exit;
4881 }
4882
a88df331 4883 /* Setup kernel tracer */
4fba7219
DG
4884 if (!opt_no_kernel) {
4885 init_kernel_tracer();
4886 }
a88df331
DG
4887
4888 /* Set ulimit for open files */
4889 set_ulimit();
fac6795d 4890 }
4063050c
MD
4891 /* init lttng_fd tracking must be done after set_ulimit. */
4892 lttng_fd_init();
fac6795d 4893
67e40797
DG
4894 ret = set_consumer_sockets(&ustconsumer64_data, rundir);
4895 if (ret < 0) {
4896 goto exit;
4897 }
4898
4899 ret = set_consumer_sockets(&ustconsumer32_data, rundir);
4900 if (ret < 0) {
4901 goto exit;
4902 }
4903
d6f42150 4904 /* Setup the needed unix socket */
cf3af59e
MD
4905 if ((ret = init_daemon_socket()) < 0) {
4906 goto exit;
fac6795d
DG
4907 }
4908
4909 /* Set credentials to socket */
be040666 4910 if (is_root && ((ret = set_permissions(rundir)) < 0)) {
cf3af59e 4911 goto exit;
fac6795d
DG
4912 }
4913
5b8719f5
DG
4914 /* Get parent pid if -S, --sig-parent is specified. */
4915 if (opt_sig_parent) {
4916 ppid = getppid();
4917 }
4918
7a485870 4919 /* Setup the kernel pipe for waking up the kernel thread */
6620da75
DG
4920 if (is_root && !opt_no_kernel) {
4921 if ((ret = utils_create_pipe_cloexec(kernel_poll_pipe)) < 0) {
4922 goto exit;
4923 }
7a485870
DG
4924 }
4925
0b2dc8df
MD
4926 /* Setup the thread ht_cleanup communication pipe. */
4927 if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
4928 goto exit;
4929 }
4930
099e26bd 4931 /* Setup the thread apps communication pipe. */
ef599319 4932 if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
099e26bd
DG
4933 goto exit;
4934 }
4935
d0b96690
DG
4936 /* Setup the thread apps notify communication pipe. */
4937 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) {
4938 goto exit;
4939 }
4940
7972aab2
DG
4941 /* Initialize global buffer per UID and PID registry. */
4942 buffer_reg_init_uid_registry();
4943 buffer_reg_init_pid_registry();
4944
099e26bd
DG
4945 /* Init UST command queue. */
4946 cds_wfq_init(&ust_cmd_queue.queue);
4947
273ea72c 4948 /*
54d01ffb
DG
4949 * Get session list pointer. This pointer MUST NOT be free(). This list is
4950 * statically declared in session.c
273ea72c 4951 */
54d01ffb 4952 session_list_ptr = session_get_list();
b5541356 4953
5eb91c98
DG
4954 /* Set up max poll set size */
4955 lttng_poll_set_max_size();
4956
2f77fc4b 4957 cmd_init();
00e2e675 4958
ae9e45b3
DG
4959 /* Check for the application socket timeout env variable. */
4960 env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV);
4961 if (env_app_timeout) {
4962 app_socket_timeout = atoi(env_app_timeout);
4963 } else {
4964 app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT;
4965 }
4966
35f90c40 4967 write_pidfile();
cd9290dd 4968 write_julport();
35f90c40 4969
554831e7
MD
4970 /* Initialize communication library */
4971 lttcomm_init();
d831c249
DG
4972 /* This is to get the TCP timeout value. */
4973 lttcomm_inet_init();
554831e7 4974
67e05644
DG
4975 /*
4976 * Initialize the health check subsystem. This call should set the
4977 * appropriate time values.
4978 */
6c71277b 4979 health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES);
8782cc74
MD
4980 if (!health_sessiond) {
4981 PERROR("health_app_create error");
4982 goto exit_health_sessiond_cleanup;
4983 }
67e05644 4984
d5b9fd2f 4985 /* Create thread to clean up RCU hash tables */
0b2dc8df
MD
4986 ret = pthread_create(&ht_cleanup_thread, NULL,
4987 thread_ht_cleanup, (void *) NULL);
4988 if (ret != 0) {
4989 PERROR("pthread_create ht_cleanup");
4990 goto exit_ht_cleanup;
4991 }
4992
d5b9fd2f 4993 /* Create health-check thread */
44a5e5eb
DG
4994 ret = pthread_create(&health_thread, NULL,
4995 thread_manage_health, (void *) NULL);
4996 if (ret != 0) {
4997 PERROR("pthread_create health");
4998 goto exit_health;
4999 }
5000
cf3af59e 5001 /* Create thread to manage the client socket */
099e26bd
DG
5002 ret = pthread_create(&client_thread, NULL,
5003 thread_manage_clients, (void *) NULL);
cf3af59e 5004 if (ret != 0) {
76d7553f 5005 PERROR("pthread_create clients");
cf3af59e
MD
5006 goto exit_client;
5007 }
fac6795d 5008
099e26bd
DG
5009 /* Create thread to dispatch registration */
5010 ret = pthread_create(&dispatch_thread, NULL,
5011 thread_dispatch_ust_registration, (void *) NULL);
5012 if (ret != 0) {
76d7553f 5013 PERROR("pthread_create dispatch");
099e26bd
DG
5014 goto exit_dispatch;
5015 }
5016
5017 /* Create thread to manage application registration. */
5018 ret = pthread_create(&reg_apps_thread, NULL,
5019 thread_registration_apps, (void *) NULL);
5020 if (ret != 0) {
76d7553f 5021 PERROR("pthread_create registration");
099e26bd
DG
5022 goto exit_reg_apps;
5023 }
5024
cf3af59e 5025 /* Create thread to manage application socket */
54d01ffb
DG
5026 ret = pthread_create(&apps_thread, NULL,
5027 thread_manage_apps, (void *) NULL);
cf3af59e 5028 if (ret != 0) {
d0b96690
DG
5029 PERROR("pthread_create apps");
5030 goto exit_apps;
5031 }
5032
5033 /* Create thread to manage application notify socket */
5034 ret = pthread_create(&apps_notify_thread, NULL,
5035 ust_thread_manage_notify, (void *) NULL);
5036 if (ret != 0) {
4d14e26d 5037 PERROR("pthread_create notify");
9563b0ad 5038 goto exit_apps_notify;
cf3af59e 5039 }
fac6795d 5040
4d076222
DG
5041 /* Create JUL registration thread. */
5042 ret = pthread_create(&jul_reg_thread, NULL,
5043 jul_thread_manage_registration, (void *) NULL);
5044 if (ret != 0) {
4d14e26d 5045 PERROR("pthread_create JUL");
4d076222
DG
5046 goto exit_jul_reg;
5047 }
5048
6620da75
DG
5049 /* Don't start this thread if kernel tracing is not requested nor root */
5050 if (is_root && !opt_no_kernel) {
5051 /* Create kernel thread to manage kernel event */
5052 ret = pthread_create(&kernel_thread, NULL,
5053 thread_manage_kernel, (void *) NULL);
5054 if (ret != 0) {
5055 PERROR("pthread_create kernel");
5056 goto exit_kernel;
5057 }
7a485870 5058
6620da75
DG
5059 ret = pthread_join(kernel_thread, &status);
5060 if (ret != 0) {
5061 PERROR("pthread_join");
5062 goto error; /* join error, exit without cleanup */
5063 }
fac6795d
DG
5064 }
5065
cf3af59e 5066exit_kernel:
4d076222
DG
5067 ret = pthread_join(jul_reg_thread, &status);
5068 if (ret != 0) {
5069 PERROR("pthread_join JUL");
5070 goto error; /* join error, exit without cleanup */
5071 }
5072
5073exit_jul_reg:
9563b0ad
DG
5074 ret = pthread_join(apps_notify_thread, &status);
5075 if (ret != 0) {
5076 PERROR("pthread_join apps notify");
5077 goto error; /* join error, exit without cleanup */
5078 }
5079
5080exit_apps_notify:
cf3af59e
MD
5081 ret = pthread_join(apps_thread, &status);
5082 if (ret != 0) {
9563b0ad 5083 PERROR("pthread_join apps");
cf3af59e
MD
5084 goto error; /* join error, exit without cleanup */
5085 }
fac6795d 5086
4ccd8c8f 5087
cf3af59e 5088exit_apps:
099e26bd
DG
5089 ret = pthread_join(reg_apps_thread, &status);
5090 if (ret != 0) {
76d7553f 5091 PERROR("pthread_join");
099e26bd
DG
5092 goto error; /* join error, exit without cleanup */
5093 }
5094
5095exit_reg_apps:
5096 ret = pthread_join(dispatch_thread, &status);
5097 if (ret != 0) {
76d7553f 5098 PERROR("pthread_join");
099e26bd
DG
5099 goto error; /* join error, exit without cleanup */
5100 }
5101
5102exit_dispatch:
cf3af59e
MD
5103 ret = pthread_join(client_thread, &status);
5104 if (ret != 0) {
76d7553f 5105 PERROR("pthread_join");
cf3af59e
MD
5106 goto error; /* join error, exit without cleanup */
5107 }
5108
3bd1e081 5109 ret = join_consumer_thread(&kconsumer_data);
cf3af59e 5110 if (ret != 0) {
76d7553f 5111 PERROR("join_consumer");
cf3af59e
MD
5112 goto error; /* join error, exit without cleanup */
5113 }
a88df331 5114
06f525de
DG
5115 ret = join_consumer_thread(&ustconsumer32_data);
5116 if (ret != 0) {
5117 PERROR("join_consumer ust32");
5118 goto error; /* join error, exit without cleanup */
5119 }
5120
5121 ret = join_consumer_thread(&ustconsumer64_data);
5122 if (ret != 0) {
5123 PERROR("join_consumer ust64");
5124 goto error; /* join error, exit without cleanup */
5125 }
5126
cf3af59e 5127exit_client:
06f525de
DG
5128 ret = pthread_join(health_thread, &status);
5129 if (ret != 0) {
5130 PERROR("pthread_join health thread");
5131 goto error; /* join error, exit without cleanup */
5132 }
5133
44a5e5eb 5134exit_health:
0b2dc8df
MD
5135 ret = pthread_join(ht_cleanup_thread, &status);
5136 if (ret != 0) {
5137 PERROR("pthread_join ht cleanup thread");
5138 goto error; /* join error, exit without cleanup */
5139 }
5140exit_ht_cleanup:
8782cc74
MD
5141 health_app_destroy(health_sessiond);
5142exit_health_sessiond_cleanup:
a88df331 5143exit:
cf3af59e
MD
5144 /*
5145 * cleanup() is called when no other thread is running.
5146 */
f6a9efaa 5147 rcu_thread_online();
cf3af59e 5148 cleanup();
f6a9efaa
DG
5149 rcu_thread_offline();
5150 rcu_unregister_thread();
67e40797 5151 if (!ret) {
cf3af59e 5152 exit(EXIT_SUCCESS);
67e40797 5153 }
cf3af59e 5154error:
5e16da05 5155 exit(EXIT_FAILURE);
fac6795d 5156}
This page took 0.420565 seconds and 4 git commands to generate.