8d90e2c367dafae1a88b8f077208a8fb0596e82d
[lttng-ust.git] / liblttng-ust / lttng-ust-comm.c
1 /*
2 * lttng-ust-comm.c
3 *
4 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
5 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 *
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; only
10 * version 2.1 of the License.
11 *
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with this library; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #define _LGPL_SOURCE
23 #include <stddef.h>
24 #include <stdint.h>
25 #include <sys/types.h>
26 #include <sys/socket.h>
27 #include <sys/mman.h>
28 #include <sys/stat.h>
29 #include <sys/types.h>
30 #include <sys/wait.h>
31 #include <dlfcn.h>
32 #include <fcntl.h>
33 #include <unistd.h>
34 #include <errno.h>
35 #include <pthread.h>
36 #include <semaphore.h>
37 #include <time.h>
38 #include <assert.h>
39 #include <signal.h>
40 #include <limits.h>
41 #include <urcu/uatomic.h>
42 #include "futex.h"
43 #include <urcu/compiler.h>
44 #include <lttng/urcu/urcu-ust.h>
45
46 #include <lttng/align.h>
47 #include <lttng/ust-events.h>
48 #include <lttng/ust-abi.h>
49 #include <lttng/ust.h>
50 #include <lttng/ust-error.h>
51 #include <lttng/ust-ctl.h>
52 #include <urcu/tls-compat.h>
53 #include <ust-comm.h>
54 #include <ust-fd.h>
55 #include <usterr-signal-safe.h>
56 #include <helper.h>
57 #include "tracepoint-internal.h"
58 #include "lttng-tracer-core.h"
59 #include "compat.h"
60 #include "../libringbuffer/rb-init.h"
61 #include "lttng-ust-statedump.h"
62 #include "clock.h"
63 #include "../libringbuffer/getcpu.h"
64 #include "getenv.h"
65 #include "ust-events-internal.h"
66
67 /* Concatenate lttng ust shared library name with its major version number. */
68 #define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." __ust_stringify(CONFIG_LTTNG_UST_LIBRARY_VERSION_MAJOR)
69
70 /*
71 * Has lttng ust comm constructor been called ?
72 */
73 static int initialized;
74
75 /*
76 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
77 * Held when handling a command, also held by fork() to deal with
78 * removal of threads, and by exit path.
79 *
80 * The UST lock is the centralized mutex across UST tracing control and
81 * probe registration.
82 *
83 * ust_exit_mutex must never nest in ust_mutex.
84 *
85 * ust_fork_mutex must never nest in ust_mutex.
86 *
87 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
88 * counter lazy initialization called by events within the statedump,
89 * which traces while the ust_mutex is held.
90 *
91 * ust_lock nests within the dynamic loader lock (within glibc) because
92 * it is taken within the library constructor.
93 *
94 * The ust fd tracker lock nests within the ust_mutex.
95 */
96 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
97
98 /* Allow nesting the ust_mutex within the same thread. */
99 static DEFINE_URCU_TLS(int, ust_mutex_nest);
100
101 /*
102 * ust_exit_mutex protects thread_active variable wrt thread exit. It
103 * cannot be done by ust_mutex because pthread_cancel(), which takes an
104 * internal libc lock, cannot nest within ust_mutex.
105 *
106 * It never nests within a ust_mutex.
107 */
108 static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
109
110 /*
111 * ust_fork_mutex protects base address statedump tracing against forks. It
112 * prevents the dynamic loader lock to be taken (by base address statedump
113 * tracing) while a fork is happening, thus preventing deadlock issues with
114 * the dynamic loader lock.
115 */
116 static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
117
118 /* Should the ust comm thread quit ? */
119 static int lttng_ust_comm_should_quit;
120
121 /*
122 * This variable can be tested by applications to check whether
123 * lttng-ust is loaded. They simply have to define their own
124 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
125 * library constructor.
126 */
127 int lttng_ust_loaded __attribute__((weak));
128
129 /*
130 * Return 0 on success, -1 if should quit.
131 * The lock is taken in both cases.
132 * Signal-safe.
133 */
134 int ust_lock(void)
135 {
136 sigset_t sig_all_blocked, orig_mask;
137 int ret, oldstate;
138
139 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
140 if (ret) {
141 ERR("pthread_setcancelstate: %s", strerror(ret));
142 }
143 if (oldstate != PTHREAD_CANCEL_ENABLE) {
144 ERR("pthread_setcancelstate: unexpected oldstate");
145 }
146 sigfillset(&sig_all_blocked);
147 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
148 if (ret) {
149 ERR("pthread_sigmask: %s", strerror(ret));
150 }
151 if (!URCU_TLS(ust_mutex_nest)++)
152 pthread_mutex_lock(&ust_mutex);
153 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
154 if (ret) {
155 ERR("pthread_sigmask: %s", strerror(ret));
156 }
157 if (lttng_ust_comm_should_quit) {
158 return -1;
159 } else {
160 return 0;
161 }
162 }
163
164 /*
165 * ust_lock_nocheck() can be used in constructors/destructors, because
166 * they are already nested within the dynamic loader lock, and therefore
167 * have exclusive access against execution of liblttng-ust destructor.
168 * Signal-safe.
169 */
170 void ust_lock_nocheck(void)
171 {
172 sigset_t sig_all_blocked, orig_mask;
173 int ret, oldstate;
174
175 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
176 if (ret) {
177 ERR("pthread_setcancelstate: %s", strerror(ret));
178 }
179 if (oldstate != PTHREAD_CANCEL_ENABLE) {
180 ERR("pthread_setcancelstate: unexpected oldstate");
181 }
182 sigfillset(&sig_all_blocked);
183 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
184 if (ret) {
185 ERR("pthread_sigmask: %s", strerror(ret));
186 }
187 if (!URCU_TLS(ust_mutex_nest)++)
188 pthread_mutex_lock(&ust_mutex);
189 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
190 if (ret) {
191 ERR("pthread_sigmask: %s", strerror(ret));
192 }
193 }
194
195 /*
196 * Signal-safe.
197 */
198 void ust_unlock(void)
199 {
200 sigset_t sig_all_blocked, orig_mask;
201 int ret, oldstate;
202
203 sigfillset(&sig_all_blocked);
204 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
205 if (ret) {
206 ERR("pthread_sigmask: %s", strerror(ret));
207 }
208 if (!--URCU_TLS(ust_mutex_nest))
209 pthread_mutex_unlock(&ust_mutex);
210 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
211 if (ret) {
212 ERR("pthread_sigmask: %s", strerror(ret));
213 }
214 ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
215 if (ret) {
216 ERR("pthread_setcancelstate: %s", strerror(ret));
217 }
218 if (oldstate != PTHREAD_CANCEL_DISABLE) {
219 ERR("pthread_setcancelstate: unexpected oldstate");
220 }
221 }
222
223 /*
224 * Wait for either of these before continuing to the main
225 * program:
226 * - the register_done message from sessiond daemon
227 * (will let the sessiond daemon enable sessions before main
228 * starts.)
229 * - sessiond daemon is not reachable.
230 * - timeout (ensuring applications are resilient to session
231 * daemon problems).
232 */
233 static sem_t constructor_wait;
234 /*
235 * Doing this for both the global and local sessiond.
236 */
237 enum {
238 sem_count_initial_value = 4,
239 };
240
241 static int sem_count = sem_count_initial_value;
242
243 /*
244 * Counting nesting within lttng-ust. Used to ensure that calling fork()
245 * from liblttng-ust does not execute the pre/post fork handlers.
246 */
247 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
248
249 /*
250 * Info about socket and associated listener thread.
251 */
252 struct sock_info {
253 const char *name;
254 pthread_t ust_listener; /* listener thread */
255 int root_handle;
256 int registration_done;
257 int allowed;
258 int global;
259 int thread_active;
260
261 char sock_path[PATH_MAX];
262 int socket;
263 int notify_socket;
264
265 char wait_shm_path[PATH_MAX];
266 char *wait_shm_mmap;
267 /* Keep track of lazy state dump not performed yet. */
268 int statedump_pending;
269 int initial_statedump_done;
270 /* Keep procname for statedump */
271 char procname[LTTNG_UST_ABI_PROCNAME_LEN];
272 };
273
274 /* Socket from app (connect) to session daemon (listen) for communication */
275 struct sock_info global_apps = {
276 .name = "global",
277 .global = 1,
278
279 .root_handle = -1,
280 .registration_done = 0,
281 .allowed = 0,
282 .thread_active = 0,
283
284 .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
285 .socket = -1,
286 .notify_socket = -1,
287
288 .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
289
290 .statedump_pending = 0,
291 .initial_statedump_done = 0,
292 .procname[0] = '\0'
293 };
294
295 /* TODO: allow global_apps_sock_path override */
296
297 struct sock_info local_apps = {
298 .name = "local",
299 .global = 0,
300 .root_handle = -1,
301 .registration_done = 0,
302 .allowed = 0, /* Check setuid bit first */
303 .thread_active = 0,
304
305 .socket = -1,
306 .notify_socket = -1,
307
308 .statedump_pending = 0,
309 .initial_statedump_done = 0,
310 .procname[0] = '\0'
311 };
312
313 static int wait_poll_fallback;
314
315 static const char *cmd_name_mapping[] = {
316 [ LTTNG_UST_RELEASE ] = "Release",
317 [ LTTNG_UST_SESSION ] = "Create Session",
318 [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
319
320 [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
321 [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
322 [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
323 [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
324
325 [ LTTNG_UST_EVENT_NOTIFIER_GROUP_CREATE ] = "Create event notifier group",
326
327 /* Session FD commands */
328 [ LTTNG_UST_CHANNEL ] = "Create Channel",
329 [ LTTNG_UST_SESSION_START ] = "Start Session",
330 [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
331
332 /* Channel FD commands */
333 [ LTTNG_UST_STREAM ] = "Create Stream",
334 [ LTTNG_UST_EVENT ] = "Create Event",
335
336 /* Event and Channel FD commands */
337 [ LTTNG_UST_CONTEXT ] = "Create Context",
338 [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
339
340 /* Event, Channel and Session commands */
341 [ LTTNG_UST_ENABLE ] = "Enable",
342 [ LTTNG_UST_DISABLE ] = "Disable",
343
344 /* Tracepoint list commands */
345 [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
346 [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
347
348 /* Event FD commands */
349 [ LTTNG_UST_FILTER ] = "Create Filter",
350 [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
351
352 /* Event notifier group commands */
353 [ LTTNG_UST_EVENT_NOTIFIER_CREATE ] = "Create event notifier",
354
355 /* Session and event notifier group commands */
356 [ LTTNG_UST_COUNTER ] = "Create Counter",
357
358 /* Counter commands */
359 [ LTTNG_UST_COUNTER_GLOBAL ] = "Create Counter Global",
360 [ LTTNG_UST_COUNTER_CPU ] = "Create Counter CPU",
361 };
362
363 static const char *str_timeout;
364 static int got_timeout_env;
365
366 extern void lttng_ring_buffer_client_overwrite_init(void);
367 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
368 extern void lttng_ring_buffer_client_discard_init(void);
369 extern void lttng_ring_buffer_client_discard_rt_init(void);
370 extern void lttng_ring_buffer_metadata_client_init(void);
371 extern void lttng_ring_buffer_client_overwrite_exit(void);
372 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
373 extern void lttng_ring_buffer_client_discard_exit(void);
374 extern void lttng_ring_buffer_client_discard_rt_exit(void);
375 extern void lttng_ring_buffer_metadata_client_exit(void);
376 extern void lttng_counter_client_percpu_32_modular_init(void);
377 extern void lttng_counter_client_percpu_32_modular_exit(void);
378 extern void lttng_counter_client_percpu_64_modular_init(void);
379 extern void lttng_counter_client_percpu_64_modular_exit(void);
380
381 static char *get_map_shm(struct sock_info *sock_info);
382
383 ssize_t lttng_ust_read(int fd, void *buf, size_t len)
384 {
385 ssize_t ret;
386 size_t copied = 0, to_copy = len;
387
388 do {
389 ret = read(fd, buf + copied, to_copy);
390 if (ret > 0) {
391 copied += ret;
392 to_copy -= ret;
393 }
394 } while ((ret > 0 && to_copy > 0)
395 || (ret < 0 && errno == EINTR));
396 if (ret > 0) {
397 ret = copied;
398 }
399 return ret;
400 }
401 /*
402 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
403 * pointer.
404 */
405 static
406 const char *get_lttng_home_dir(void)
407 {
408 const char *val;
409
410 val = (const char *) lttng_getenv("LTTNG_HOME");
411 if (val != NULL) {
412 return val;
413 }
414 return (const char *) lttng_getenv("HOME");
415 }
416
417 /*
418 * Force a read (imply TLS fixup for dlopen) of TLS variables.
419 */
420 static
421 void lttng_fixup_nest_count_tls(void)
422 {
423 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
424 }
425
426 static
427 void lttng_fixup_ust_mutex_nest_tls(void)
428 {
429 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
430 }
431
432 /*
433 * Fixup urcu bp TLS.
434 */
435 static
436 void lttng_fixup_urcu_bp_tls(void)
437 {
438 (void) lttng_ust_urcu_read_ongoing();
439 }
440
441 void lttng_ust_fixup_tls(void)
442 {
443 lttng_fixup_urcu_bp_tls();
444 lttng_fixup_ringbuffer_tls();
445 lttng_fixup_vtid_tls();
446 lttng_fixup_nest_count_tls();
447 lttng_fixup_procname_tls();
448 lttng_fixup_ust_mutex_nest_tls();
449 lttng_ust_fixup_perf_counter_tls();
450 lttng_ust_fixup_fd_tracker_tls();
451 lttng_fixup_cgroup_ns_tls();
452 lttng_fixup_ipc_ns_tls();
453 lttng_fixup_net_ns_tls();
454 lttng_fixup_time_ns_tls();
455 lttng_fixup_uts_ns_tls();
456 }
457
458 int lttng_get_notify_socket(void *owner)
459 {
460 struct sock_info *info = owner;
461
462 return info->notify_socket;
463 }
464
465
466 LTTNG_HIDDEN
467 char* lttng_ust_sockinfo_get_procname(void *owner)
468 {
469 struct sock_info *info = owner;
470
471 return info->procname;
472 }
473
474 static
475 void print_cmd(int cmd, int handle)
476 {
477 const char *cmd_name = "Unknown";
478
479 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
480 && cmd_name_mapping[cmd]) {
481 cmd_name = cmd_name_mapping[cmd];
482 }
483 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
484 cmd_name, cmd,
485 lttng_ust_obj_get_name(handle), handle);
486 }
487
488 static
489 int setup_global_apps(void)
490 {
491 int ret = 0;
492 assert(!global_apps.wait_shm_mmap);
493
494 global_apps.wait_shm_mmap = get_map_shm(&global_apps);
495 if (!global_apps.wait_shm_mmap) {
496 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
497 global_apps.allowed = 0;
498 ret = -EIO;
499 goto error;
500 }
501
502 global_apps.allowed = 1;
503 lttng_pthread_getname_np(global_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
504 error:
505 return ret;
506 }
507 static
508 int setup_local_apps(void)
509 {
510 int ret = 0;
511 const char *home_dir;
512 uid_t uid;
513
514 assert(!local_apps.wait_shm_mmap);
515
516 uid = getuid();
517 /*
518 * Disallow per-user tracing for setuid binaries.
519 */
520 if (uid != geteuid()) {
521 assert(local_apps.allowed == 0);
522 ret = 0;
523 goto end;
524 }
525 home_dir = get_lttng_home_dir();
526 if (!home_dir) {
527 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
528 assert(local_apps.allowed == 0);
529 ret = -ENOENT;
530 goto end;
531 }
532 local_apps.allowed = 1;
533 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
534 home_dir,
535 LTTNG_DEFAULT_HOME_RUNDIR,
536 LTTNG_UST_SOCK_FILENAME);
537 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
538 LTTNG_UST_WAIT_FILENAME,
539 uid);
540
541 local_apps.wait_shm_mmap = get_map_shm(&local_apps);
542 if (!local_apps.wait_shm_mmap) {
543 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
544 local_apps.allowed = 0;
545 ret = -EIO;
546 goto end;
547 }
548
549 lttng_pthread_getname_np(local_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
550 end:
551 return ret;
552 }
553
554 /*
555 * Get socket timeout, in ms.
556 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
557 */
558 static
559 long get_timeout(void)
560 {
561 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
562
563 if (!got_timeout_env) {
564 str_timeout = lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
565 got_timeout_env = 1;
566 }
567 if (str_timeout)
568 constructor_delay_ms = strtol(str_timeout, NULL, 10);
569 /* All negative values are considered as "-1". */
570 if (constructor_delay_ms < -1)
571 constructor_delay_ms = -1;
572 return constructor_delay_ms;
573 }
574
575 /* Timeout for notify socket send and recv. */
576 static
577 long get_notify_sock_timeout(void)
578 {
579 return get_timeout();
580 }
581
582 /* Timeout for connecting to cmd and notify sockets. */
583 static
584 long get_connect_sock_timeout(void)
585 {
586 return get_timeout();
587 }
588
589 /*
590 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
591 */
592 static
593 int get_constructor_timeout(struct timespec *constructor_timeout)
594 {
595 long constructor_delay_ms;
596 int ret;
597
598 constructor_delay_ms = get_timeout();
599
600 switch (constructor_delay_ms) {
601 case -1:/* fall-through */
602 case 0:
603 return constructor_delay_ms;
604 default:
605 break;
606 }
607
608 /*
609 * If we are unable to find the current time, don't wait.
610 */
611 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
612 if (ret) {
613 /* Don't wait. */
614 return 0;
615 }
616 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
617 constructor_timeout->tv_nsec +=
618 (constructor_delay_ms % 1000UL) * 1000000UL;
619 if (constructor_timeout->tv_nsec >= 1000000000UL) {
620 constructor_timeout->tv_sec++;
621 constructor_timeout->tv_nsec -= 1000000000UL;
622 }
623 /* Timeout wait (constructor_delay_ms). */
624 return 1;
625 }
626
627 static
628 void get_allow_blocking(void)
629 {
630 const char *str_allow_blocking =
631 lttng_getenv("LTTNG_UST_ALLOW_BLOCKING");
632
633 if (str_allow_blocking) {
634 DBG("%s environment variable is set",
635 "LTTNG_UST_ALLOW_BLOCKING");
636 lttng_ust_ringbuffer_set_allow_blocking();
637 }
638 }
639
640 static
641 int register_to_sessiond(int socket, enum ustctl_socket_type type)
642 {
643 return ustcomm_send_reg_msg(socket,
644 type,
645 CAA_BITS_PER_LONG,
646 lttng_alignof(uint8_t) * CHAR_BIT,
647 lttng_alignof(uint16_t) * CHAR_BIT,
648 lttng_alignof(uint32_t) * CHAR_BIT,
649 lttng_alignof(uint64_t) * CHAR_BIT,
650 lttng_alignof(unsigned long) * CHAR_BIT);
651 }
652
653 static
654 int send_reply(int sock, struct ustcomm_ust_reply *lur)
655 {
656 ssize_t len;
657
658 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
659 switch (len) {
660 case sizeof(*lur):
661 DBG("message successfully sent");
662 return 0;
663 default:
664 if (len == -ECONNRESET) {
665 DBG("remote end closed connection");
666 return 0;
667 }
668 if (len < 0)
669 return len;
670 DBG("incorrect message size: %zd", len);
671 return -EINVAL;
672 }
673 }
674
675 static
676 void decrement_sem_count(unsigned int count)
677 {
678 int ret;
679
680 assert(uatomic_read(&sem_count) >= count);
681
682 if (uatomic_read(&sem_count) <= 0) {
683 return;
684 }
685
686 ret = uatomic_add_return(&sem_count, -count);
687 if (ret == 0) {
688 ret = sem_post(&constructor_wait);
689 assert(!ret);
690 }
691 }
692
693 static
694 int handle_register_done(struct sock_info *sock_info)
695 {
696 if (sock_info->registration_done)
697 return 0;
698 sock_info->registration_done = 1;
699
700 decrement_sem_count(1);
701 if (!sock_info->statedump_pending) {
702 sock_info->initial_statedump_done = 1;
703 decrement_sem_count(1);
704 }
705
706 return 0;
707 }
708
709 static
710 int handle_register_failed(struct sock_info *sock_info)
711 {
712 if (sock_info->registration_done)
713 return 0;
714 sock_info->registration_done = 1;
715 sock_info->initial_statedump_done = 1;
716
717 decrement_sem_count(2);
718
719 return 0;
720 }
721
722 /*
723 * Only execute pending statedump after the constructor semaphore has
724 * been posted by the current listener thread. This means statedump will
725 * only be performed after the "registration done" command is received
726 * from this thread's session daemon.
727 *
728 * This ensures we don't run into deadlock issues with the dynamic
729 * loader mutex, which is held while the constructor is called and
730 * waiting on the constructor semaphore. All operations requiring this
731 * dynamic loader lock need to be postponed using this mechanism.
732 *
733 * In a scenario with two session daemons connected to the application,
734 * it is possible that the first listener thread which receives the
735 * registration done command issues its statedump while the dynamic
736 * loader lock is still held by the application constructor waiting on
737 * the semaphore. It will however be allowed to proceed when the
738 * second session daemon sends the registration done command to the
739 * second listener thread. This situation therefore does not produce
740 * a deadlock.
741 */
742 static
743 void handle_pending_statedump(struct sock_info *sock_info)
744 {
745 if (sock_info->registration_done && sock_info->statedump_pending) {
746 sock_info->statedump_pending = 0;
747 pthread_mutex_lock(&ust_fork_mutex);
748 lttng_handle_pending_statedump(sock_info);
749 pthread_mutex_unlock(&ust_fork_mutex);
750
751 if (!sock_info->initial_statedump_done) {
752 sock_info->initial_statedump_done = 1;
753 decrement_sem_count(1);
754 }
755 }
756 }
757
758 static inline
759 const char *bytecode_type_str(uint32_t cmd)
760 {
761 switch (cmd) {
762 case LTTNG_UST_CAPTURE:
763 return "capture";
764 case LTTNG_UST_FILTER:
765 return "filter";
766 default:
767 abort();
768 }
769 }
770
771 static
772 int handle_bytecode_recv(struct sock_info *sock_info,
773 int sock, struct ustcomm_ust_msg *lum)
774 {
775 struct lttng_ust_bytecode_node *bytecode = NULL;
776 enum lttng_ust_bytecode_node_type type;
777 const struct lttng_ust_objd_ops *ops;
778 uint32_t data_size, data_size_max, reloc_offset;
779 uint64_t seqnum;
780 ssize_t len;
781 int ret = 0;
782
783 switch (lum->cmd) {
784 case LTTNG_UST_FILTER:
785 type = LTTNG_UST_BYTECODE_NODE_TYPE_FILTER;
786 data_size = lum->u.filter.data_size;
787 data_size_max = FILTER_BYTECODE_MAX_LEN;
788 reloc_offset = lum->u.filter.reloc_offset;
789 seqnum = lum->u.filter.seqnum;
790 break;
791 case LTTNG_UST_CAPTURE:
792 type = LTTNG_UST_BYTECODE_NODE_TYPE_CAPTURE;
793 data_size = lum->u.capture.data_size;
794 data_size_max = CAPTURE_BYTECODE_MAX_LEN;
795 reloc_offset = lum->u.capture.reloc_offset;
796 seqnum = lum->u.capture.seqnum;
797 break;
798 default:
799 abort();
800 }
801
802 if (data_size > data_size_max) {
803 ERR("Bytecode %s data size is too large: %u bytes",
804 bytecode_type_str(lum->cmd), data_size);
805 ret = -EINVAL;
806 goto end;
807 }
808
809 if (reloc_offset > data_size) {
810 ERR("Bytecode %s reloc offset %u is not within data",
811 bytecode_type_str(lum->cmd), reloc_offset);
812 ret = -EINVAL;
813 goto end;
814 }
815
816 /* Allocate the structure AND the `data[]` field. */
817 bytecode = zmalloc(sizeof(*bytecode) + data_size);
818 if (!bytecode) {
819 ret = -ENOMEM;
820 goto end;
821 }
822
823 bytecode->bc.len = data_size;
824 bytecode->bc.reloc_offset = reloc_offset;
825 bytecode->bc.seqnum = seqnum;
826 bytecode->type = type;
827
828 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, bytecode->bc.len);
829 switch (len) {
830 case 0: /* orderly shutdown */
831 ret = 0;
832 goto end;
833 default:
834 if (len == bytecode->bc.len) {
835 DBG("Bytecode %s data received",
836 bytecode_type_str(lum->cmd));
837 break;
838 } else if (len < 0) {
839 DBG("Receive failed from lttng-sessiond with errno %d",
840 (int) -len);
841 if (len == -ECONNRESET) {
842 ERR("%s remote end closed connection",
843 sock_info->name);
844 ret = len;
845 goto end;
846 }
847 ret = len;
848 goto end;
849 } else {
850 DBG("Incorrect %s bytecode data message size: %zd",
851 bytecode_type_str(lum->cmd), len);
852 ret = -EINVAL;
853 goto end;
854 }
855 }
856
857 ops = objd_ops(lum->handle);
858 if (!ops) {
859 ret = -ENOENT;
860 goto end;
861 }
862
863 if (ops->cmd)
864 ret = ops->cmd(lum->handle, lum->cmd,
865 (unsigned long) &bytecode,
866 NULL, sock_info);
867 else
868 ret = -ENOSYS;
869
870 end:
871 free(bytecode);
872 return ret;
873 }
874
875 static
876 int handle_message(struct sock_info *sock_info,
877 int sock, struct ustcomm_ust_msg *lum)
878 {
879 int ret = 0;
880 const struct lttng_ust_objd_ops *ops;
881 struct ustcomm_ust_reply lur;
882 union ust_args args;
883 char ctxstr[LTTNG_UST_SYM_NAME_LEN]; /* App context string. */
884 ssize_t len;
885
886 memset(&lur, 0, sizeof(lur));
887
888 if (ust_lock()) {
889 ret = -LTTNG_UST_ERR_EXITING;
890 goto error;
891 }
892
893 ops = objd_ops(lum->handle);
894 if (!ops) {
895 ret = -ENOENT;
896 goto error;
897 }
898
899 switch (lum->cmd) {
900 case LTTNG_UST_REGISTER_DONE:
901 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
902 ret = handle_register_done(sock_info);
903 else
904 ret = -EINVAL;
905 break;
906 case LTTNG_UST_RELEASE:
907 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
908 ret = -EPERM;
909 else
910 ret = lttng_ust_objd_unref(lum->handle, 1);
911 break;
912 case LTTNG_UST_CAPTURE:
913 case LTTNG_UST_FILTER:
914 ret = handle_bytecode_recv(sock_info, sock, lum);
915 if (ret)
916 goto error;
917 break;
918 case LTTNG_UST_EXCLUSION:
919 {
920 /* Receive exclusion names */
921 struct lttng_ust_excluder_node *node;
922 unsigned int count;
923
924 count = lum->u.exclusion.count;
925 if (count == 0) {
926 /* There are no names to read */
927 ret = 0;
928 goto error;
929 }
930 node = zmalloc(sizeof(*node) +
931 count * LTTNG_UST_SYM_NAME_LEN);
932 if (!node) {
933 ret = -ENOMEM;
934 goto error;
935 }
936 node->excluder.count = count;
937 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
938 count * LTTNG_UST_SYM_NAME_LEN);
939 switch (len) {
940 case 0: /* orderly shutdown */
941 ret = 0;
942 free(node);
943 goto error;
944 default:
945 if (len == count * LTTNG_UST_SYM_NAME_LEN) {
946 DBG("Exclusion data received");
947 break;
948 } else if (len < 0) {
949 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
950 if (len == -ECONNRESET) {
951 ERR("%s remote end closed connection", sock_info->name);
952 ret = len;
953 free(node);
954 goto error;
955 }
956 ret = len;
957 free(node);
958 goto error;
959 } else {
960 DBG("Incorrect exclusion data message size: %zd", len);
961 ret = -EINVAL;
962 free(node);
963 goto error;
964 }
965 }
966 if (ops->cmd)
967 ret = ops->cmd(lum->handle, lum->cmd,
968 (unsigned long) &node,
969 &args, sock_info);
970 else
971 ret = -ENOSYS;
972 free(node);
973 break;
974 }
975 case LTTNG_UST_EVENT_NOTIFIER_GROUP_CREATE:
976 {
977 int event_notifier_notif_fd, close_ret;
978
979 len = ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock,
980 &event_notifier_notif_fd);
981 switch (len) {
982 case 0: /* orderly shutdown */
983 ret = 0;
984 goto error;
985 case 1:
986 break;
987 default:
988 if (len < 0) {
989 DBG("Receive failed from lttng-sessiond with errno %d",
990 (int) -len);
991 if (len == -ECONNRESET) {
992 ERR("%s remote end closed connection",
993 sock_info->name);
994 ret = len;
995 goto error;
996 }
997 ret = len;
998 goto error;
999 } else {
1000 DBG("Incorrect event notifier fd message size: %zd",
1001 len);
1002 ret = -EINVAL;
1003 goto error;
1004 }
1005 }
1006 args.event_notifier_handle.event_notifier_notif_fd =
1007 event_notifier_notif_fd;
1008 if (ops->cmd)
1009 ret = ops->cmd(lum->handle, lum->cmd,
1010 (unsigned long) &lum->u,
1011 &args, sock_info);
1012 else
1013 ret = -ENOSYS;
1014 if (args.event_notifier_handle.event_notifier_notif_fd >= 0) {
1015 lttng_ust_lock_fd_tracker();
1016 close_ret = close(args.event_notifier_handle.event_notifier_notif_fd);
1017 lttng_ust_unlock_fd_tracker();
1018 if (close_ret)
1019 PERROR("close");
1020 }
1021 break;
1022 }
1023 case LTTNG_UST_CHANNEL:
1024 {
1025 void *chan_data;
1026 int wakeup_fd;
1027
1028 len = ustcomm_recv_channel_from_sessiond(sock,
1029 &chan_data, lum->u.channel.len,
1030 &wakeup_fd);
1031 switch (len) {
1032 case 0: /* orderly shutdown */
1033 ret = 0;
1034 goto error;
1035 default:
1036 if (len == lum->u.channel.len) {
1037 DBG("channel data received");
1038 break;
1039 } else if (len < 0) {
1040 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1041 if (len == -ECONNRESET) {
1042 ERR("%s remote end closed connection", sock_info->name);
1043 ret = len;
1044 goto error;
1045 }
1046 ret = len;
1047 goto error;
1048 } else {
1049 DBG("incorrect channel data message size: %zd", len);
1050 ret = -EINVAL;
1051 goto error;
1052 }
1053 }
1054 args.channel.chan_data = chan_data;
1055 args.channel.wakeup_fd = wakeup_fd;
1056 if (ops->cmd)
1057 ret = ops->cmd(lum->handle, lum->cmd,
1058 (unsigned long) &lum->u,
1059 &args, sock_info);
1060 else
1061 ret = -ENOSYS;
1062 if (args.channel.wakeup_fd >= 0) {
1063 int close_ret;
1064
1065 lttng_ust_lock_fd_tracker();
1066 close_ret = close(args.channel.wakeup_fd);
1067 lttng_ust_unlock_fd_tracker();
1068 args.channel.wakeup_fd = -1;
1069 if (close_ret)
1070 PERROR("close");
1071 }
1072 free(args.channel.chan_data);
1073 break;
1074 }
1075 case LTTNG_UST_STREAM:
1076 {
1077 int close_ret;
1078
1079 /* Receive shm_fd, wakeup_fd */
1080 ret = ustcomm_recv_stream_from_sessiond(sock,
1081 NULL,
1082 &args.stream.shm_fd,
1083 &args.stream.wakeup_fd);
1084 if (ret) {
1085 goto error;
1086 }
1087
1088 if (ops->cmd)
1089 ret = ops->cmd(lum->handle, lum->cmd,
1090 (unsigned long) &lum->u,
1091 &args, sock_info);
1092 else
1093 ret = -ENOSYS;
1094 if (args.stream.shm_fd >= 0) {
1095 lttng_ust_lock_fd_tracker();
1096 close_ret = close(args.stream.shm_fd);
1097 lttng_ust_unlock_fd_tracker();
1098 args.stream.shm_fd = -1;
1099 if (close_ret)
1100 PERROR("close");
1101 }
1102 if (args.stream.wakeup_fd >= 0) {
1103 lttng_ust_lock_fd_tracker();
1104 close_ret = close(args.stream.wakeup_fd);
1105 lttng_ust_unlock_fd_tracker();
1106 args.stream.wakeup_fd = -1;
1107 if (close_ret)
1108 PERROR("close");
1109 }
1110 break;
1111 }
1112 case LTTNG_UST_CONTEXT:
1113 switch (lum->u.context.ctx) {
1114 case LTTNG_UST_CONTEXT_APP_CONTEXT:
1115 {
1116 char *p;
1117 size_t ctxlen, recvlen;
1118
1119 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
1120 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
1121 if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
1122 ERR("Application context string length size is too large: %zu bytes",
1123 ctxlen);
1124 ret = -EINVAL;
1125 goto error;
1126 }
1127 strcpy(ctxstr, "$app.");
1128 p = &ctxstr[strlen("$app.")];
1129 recvlen = ctxlen - strlen("$app.");
1130 len = ustcomm_recv_unix_sock(sock, p, recvlen);
1131 switch (len) {
1132 case 0: /* orderly shutdown */
1133 ret = 0;
1134 goto error;
1135 default:
1136 if (len == recvlen) {
1137 DBG("app context data received");
1138 break;
1139 } else if (len < 0) {
1140 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1141 if (len == -ECONNRESET) {
1142 ERR("%s remote end closed connection", sock_info->name);
1143 ret = len;
1144 goto error;
1145 }
1146 ret = len;
1147 goto error;
1148 } else {
1149 DBG("incorrect app context data message size: %zd", len);
1150 ret = -EINVAL;
1151 goto error;
1152 }
1153 }
1154 /* Put : between provider and ctxname. */
1155 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1156 args.app_context.ctxname = ctxstr;
1157 break;
1158 }
1159 default:
1160 break;
1161 }
1162 if (ops->cmd) {
1163 ret = ops->cmd(lum->handle, lum->cmd,
1164 (unsigned long) &lum->u,
1165 &args, sock_info);
1166 } else {
1167 ret = -ENOSYS;
1168 }
1169 break;
1170 case LTTNG_UST_COUNTER:
1171 {
1172 void *counter_data;
1173
1174 len = ustcomm_recv_counter_from_sessiond(sock,
1175 &counter_data, lum->u.counter.len);
1176 switch (len) {
1177 case 0: /* orderly shutdown */
1178 ret = 0;
1179 goto error;
1180 default:
1181 if (len == lum->u.counter.len) {
1182 DBG("counter data received");
1183 break;
1184 } else if (len < 0) {
1185 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1186 if (len == -ECONNRESET) {
1187 ERR("%s remote end closed connection", sock_info->name);
1188 ret = len;
1189 goto error;
1190 }
1191 ret = len;
1192 goto error;
1193 } else {
1194 DBG("incorrect counter data message size: %zd", len);
1195 ret = -EINVAL;
1196 goto error;
1197 }
1198 }
1199 args.counter.counter_data = counter_data;
1200 if (ops->cmd)
1201 ret = ops->cmd(lum->handle, lum->cmd,
1202 (unsigned long) &lum->u,
1203 &args, sock_info);
1204 else
1205 ret = -ENOSYS;
1206 free(args.counter.counter_data);
1207 break;
1208 }
1209 case LTTNG_UST_COUNTER_GLOBAL:
1210 {
1211 /* Receive shm_fd */
1212 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1213 &args.counter_shm.shm_fd);
1214 if (ret) {
1215 goto error;
1216 }
1217
1218 if (ops->cmd)
1219 ret = ops->cmd(lum->handle, lum->cmd,
1220 (unsigned long) &lum->u,
1221 &args, sock_info);
1222 else
1223 ret = -ENOSYS;
1224 if (args.counter_shm.shm_fd >= 0) {
1225 int close_ret;
1226
1227 lttng_ust_lock_fd_tracker();
1228 close_ret = close(args.counter_shm.shm_fd);
1229 lttng_ust_unlock_fd_tracker();
1230 args.counter_shm.shm_fd = -1;
1231 if (close_ret)
1232 PERROR("close");
1233 }
1234 break;
1235 }
1236 case LTTNG_UST_COUNTER_CPU:
1237 {
1238 /* Receive shm_fd */
1239 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1240 &args.counter_shm.shm_fd);
1241 if (ret) {
1242 goto error;
1243 }
1244
1245 if (ops->cmd)
1246 ret = ops->cmd(lum->handle, lum->cmd,
1247 (unsigned long) &lum->u,
1248 &args, sock_info);
1249 else
1250 ret = -ENOSYS;
1251 if (args.counter_shm.shm_fd >= 0) {
1252 int close_ret;
1253
1254 lttng_ust_lock_fd_tracker();
1255 close_ret = close(args.counter_shm.shm_fd);
1256 lttng_ust_unlock_fd_tracker();
1257 args.counter_shm.shm_fd = -1;
1258 if (close_ret)
1259 PERROR("close");
1260 }
1261 break;
1262 }
1263 case LTTNG_UST_EVENT_NOTIFIER_CREATE:
1264 {
1265 /* Receive struct lttng_ust_event_notifier */
1266 struct lttng_ust_event_notifier event_notifier;
1267
1268 if (sizeof(event_notifier) != lum->u.event_notifier.len) {
1269 DBG("incorrect event notifier data message size: %u", lum->u.event_notifier.len);
1270 ret = -EINVAL;
1271 goto error;
1272 }
1273 len = ustcomm_recv_unix_sock(sock, &event_notifier, sizeof(event_notifier));
1274 switch (len) {
1275 case 0: /* orderly shutdown */
1276 ret = 0;
1277 goto error;
1278 default:
1279 if (len == sizeof(event_notifier)) {
1280 DBG("event notifier data received");
1281 break;
1282 } else if (len < 0) {
1283 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1284 if (len == -ECONNRESET) {
1285 ERR("%s remote end closed connection", sock_info->name);
1286 ret = len;
1287 goto error;
1288 }
1289 ret = len;
1290 goto error;
1291 } else {
1292 DBG("incorrect event notifier data message size: %zd", len);
1293 ret = -EINVAL;
1294 goto error;
1295 }
1296 }
1297 if (ops->cmd)
1298 ret = ops->cmd(lum->handle, lum->cmd,
1299 (unsigned long) &event_notifier,
1300 &args, sock_info);
1301 else
1302 ret = -ENOSYS;
1303 break;
1304 }
1305
1306 default:
1307 if (ops->cmd)
1308 ret = ops->cmd(lum->handle, lum->cmd,
1309 (unsigned long) &lum->u,
1310 &args, sock_info);
1311 else
1312 ret = -ENOSYS;
1313 break;
1314 }
1315
1316 lur.handle = lum->handle;
1317 lur.cmd = lum->cmd;
1318 lur.ret_val = ret;
1319 if (ret >= 0) {
1320 lur.ret_code = LTTNG_UST_OK;
1321 } else {
1322 /*
1323 * Use -LTTNG_UST_ERR as wildcard for UST internal
1324 * error that are not caused by the transport, except if
1325 * we already have a more precise error message to
1326 * report.
1327 */
1328 if (ret > -LTTNG_UST_ERR) {
1329 /* Translate code to UST error. */
1330 switch (ret) {
1331 case -EEXIST:
1332 lur.ret_code = -LTTNG_UST_ERR_EXIST;
1333 break;
1334 case -EINVAL:
1335 lur.ret_code = -LTTNG_UST_ERR_INVAL;
1336 break;
1337 case -ENOENT:
1338 lur.ret_code = -LTTNG_UST_ERR_NOENT;
1339 break;
1340 case -EPERM:
1341 lur.ret_code = -LTTNG_UST_ERR_PERM;
1342 break;
1343 case -ENOSYS:
1344 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
1345 break;
1346 default:
1347 lur.ret_code = -LTTNG_UST_ERR;
1348 break;
1349 }
1350 } else {
1351 lur.ret_code = ret;
1352 }
1353 }
1354 if (ret >= 0) {
1355 switch (lum->cmd) {
1356 case LTTNG_UST_TRACER_VERSION:
1357 lur.u.version = lum->u.version;
1358 break;
1359 case LTTNG_UST_TRACEPOINT_LIST_GET:
1360 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1361 break;
1362 }
1363 }
1364 DBG("Return value: %d", lur.ret_val);
1365
1366 ust_unlock();
1367
1368 /*
1369 * Performed delayed statedump operations outside of the UST
1370 * lock. We need to take the dynamic loader lock before we take
1371 * the UST lock internally within handle_pending_statedump().
1372 */
1373 handle_pending_statedump(sock_info);
1374
1375 if (ust_lock()) {
1376 ret = -LTTNG_UST_ERR_EXITING;
1377 goto error;
1378 }
1379
1380 ret = send_reply(sock, &lur);
1381 if (ret < 0) {
1382 DBG("error sending reply");
1383 goto error;
1384 }
1385
1386 /*
1387 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1388 * after the reply.
1389 */
1390 if (lur.ret_code == LTTNG_UST_OK) {
1391 switch (lum->cmd) {
1392 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
1393 len = ustcomm_send_unix_sock(sock,
1394 &args.field_list.entry,
1395 sizeof(args.field_list.entry));
1396 if (len < 0) {
1397 ret = len;
1398 goto error;
1399 }
1400 if (len != sizeof(args.field_list.entry)) {
1401 ret = -EINVAL;
1402 goto error;
1403 }
1404 }
1405 }
1406
1407 error:
1408 ust_unlock();
1409
1410 return ret;
1411 }
1412
1413 static
1414 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1415 {
1416 int ret;
1417
1418 if (sock_info->root_handle != -1) {
1419 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
1420 if (ret) {
1421 ERR("Error unref root handle");
1422 }
1423 sock_info->root_handle = -1;
1424 }
1425 sock_info->registration_done = 0;
1426 sock_info->initial_statedump_done = 0;
1427
1428 /*
1429 * wait_shm_mmap, socket and notify socket are used by listener
1430 * threads outside of the ust lock, so we cannot tear them down
1431 * ourselves, because we cannot join on these threads. Leave
1432 * responsibility of cleaning up these resources to the OS
1433 * process exit.
1434 */
1435 if (exiting)
1436 return;
1437
1438 if (sock_info->socket != -1) {
1439 ret = ustcomm_close_unix_sock(sock_info->socket);
1440 if (ret) {
1441 ERR("Error closing ust cmd socket");
1442 }
1443 sock_info->socket = -1;
1444 }
1445 if (sock_info->notify_socket != -1) {
1446 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1447 if (ret) {
1448 ERR("Error closing ust notify socket");
1449 }
1450 sock_info->notify_socket = -1;
1451 }
1452 if (sock_info->wait_shm_mmap) {
1453 long page_size;
1454
1455 page_size = LTTNG_UST_PAGE_SIZE;
1456 if (page_size <= 0) {
1457 if (!page_size) {
1458 errno = EINVAL;
1459 }
1460 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1461 } else {
1462 ret = munmap(sock_info->wait_shm_mmap, page_size);
1463 if (ret) {
1464 ERR("Error unmapping wait shm");
1465 }
1466 }
1467 sock_info->wait_shm_mmap = NULL;
1468 }
1469 }
1470
1471 /*
1472 * Using fork to set umask in the child process (not multi-thread safe).
1473 * We deal with the shm_open vs ftruncate race (happening when the
1474 * sessiond owns the shm and does not let everybody modify it, to ensure
1475 * safety against shm_unlink) by simply letting the mmap fail and
1476 * retrying after a few seconds.
1477 * For global shm, everybody has rw access to it until the sessiond
1478 * starts.
1479 */
1480 static
1481 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1482 {
1483 int wait_shm_fd, ret;
1484 pid_t pid;
1485
1486 /*
1487 * Try to open read-only.
1488 */
1489 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1490 if (wait_shm_fd >= 0) {
1491 int32_t tmp_read;
1492 ssize_t len;
1493 size_t bytes_read = 0;
1494
1495 /*
1496 * Try to read the fd. If unable to do so, try opening
1497 * it in write mode.
1498 */
1499 do {
1500 len = read(wait_shm_fd,
1501 &((char *) &tmp_read)[bytes_read],
1502 sizeof(tmp_read) - bytes_read);
1503 if (len > 0) {
1504 bytes_read += len;
1505 }
1506 } while ((len < 0 && errno == EINTR)
1507 || (len > 0 && bytes_read < sizeof(tmp_read)));
1508 if (bytes_read != sizeof(tmp_read)) {
1509 ret = close(wait_shm_fd);
1510 if (ret) {
1511 ERR("close wait_shm_fd");
1512 }
1513 goto open_write;
1514 }
1515 goto end;
1516 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1517 /*
1518 * Real-only open did not work, and it's not because the
1519 * entry was not present. It's a failure that prohibits
1520 * using shm.
1521 */
1522 ERR("Error opening shm %s", sock_info->wait_shm_path);
1523 goto end;
1524 }
1525
1526 open_write:
1527 /*
1528 * If the open failed because the file did not exist, or because
1529 * the file was not truncated yet, try creating it ourself.
1530 */
1531 URCU_TLS(lttng_ust_nest_count)++;
1532 pid = fork();
1533 URCU_TLS(lttng_ust_nest_count)--;
1534 if (pid > 0) {
1535 int status;
1536
1537 /*
1538 * Parent: wait for child to return, in which case the
1539 * shared memory map will have been created.
1540 */
1541 pid = wait(&status);
1542 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1543 wait_shm_fd = -1;
1544 goto end;
1545 }
1546 /*
1547 * Try to open read-only again after creation.
1548 */
1549 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1550 if (wait_shm_fd < 0) {
1551 /*
1552 * Real-only open did not work. It's a failure
1553 * that prohibits using shm.
1554 */
1555 ERR("Error opening shm %s", sock_info->wait_shm_path);
1556 goto end;
1557 }
1558 goto end;
1559 } else if (pid == 0) {
1560 int create_mode;
1561
1562 /* Child */
1563 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1564 if (sock_info->global)
1565 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1566 /*
1567 * We're alone in a child process, so we can modify the
1568 * process-wide umask.
1569 */
1570 umask(~create_mode);
1571 /*
1572 * Try creating shm (or get rw access).
1573 * We don't do an exclusive open, because we allow other
1574 * processes to create+ftruncate it concurrently.
1575 */
1576 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1577 O_RDWR | O_CREAT, create_mode);
1578 if (wait_shm_fd >= 0) {
1579 ret = ftruncate(wait_shm_fd, mmap_size);
1580 if (ret) {
1581 PERROR("ftruncate");
1582 _exit(EXIT_FAILURE);
1583 }
1584 _exit(EXIT_SUCCESS);
1585 }
1586 /*
1587 * For local shm, we need to have rw access to accept
1588 * opening it: this means the local sessiond will be
1589 * able to wake us up. For global shm, we open it even
1590 * if rw access is not granted, because the root.root
1591 * sessiond will be able to override all rights and wake
1592 * us up.
1593 */
1594 if (!sock_info->global && errno != EACCES) {
1595 ERR("Error opening shm %s", sock_info->wait_shm_path);
1596 _exit(EXIT_FAILURE);
1597 }
1598 /*
1599 * The shm exists, but we cannot open it RW. Report
1600 * success.
1601 */
1602 _exit(EXIT_SUCCESS);
1603 } else {
1604 return -1;
1605 }
1606 end:
1607 if (wait_shm_fd >= 0 && !sock_info->global) {
1608 struct stat statbuf;
1609
1610 /*
1611 * Ensure that our user is the owner of the shm file for
1612 * local shm. If we do not own the file, it means our
1613 * sessiond will not have access to wake us up (there is
1614 * probably a rogue process trying to fake our
1615 * sessiond). Fallback to polling method in this case.
1616 */
1617 ret = fstat(wait_shm_fd, &statbuf);
1618 if (ret) {
1619 PERROR("fstat");
1620 goto error_close;
1621 }
1622 if (statbuf.st_uid != getuid())
1623 goto error_close;
1624 }
1625 return wait_shm_fd;
1626
1627 error_close:
1628 ret = close(wait_shm_fd);
1629 if (ret) {
1630 PERROR("Error closing fd");
1631 }
1632 return -1;
1633 }
1634
1635 static
1636 char *get_map_shm(struct sock_info *sock_info)
1637 {
1638 long page_size;
1639 int wait_shm_fd, ret;
1640 char *wait_shm_mmap;
1641
1642 page_size = sysconf(_SC_PAGE_SIZE);
1643 if (page_size <= 0) {
1644 if (!page_size) {
1645 errno = EINVAL;
1646 }
1647 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1648 goto error;
1649 }
1650
1651 lttng_ust_lock_fd_tracker();
1652 wait_shm_fd = get_wait_shm(sock_info, page_size);
1653 if (wait_shm_fd < 0) {
1654 lttng_ust_unlock_fd_tracker();
1655 goto error;
1656 }
1657
1658 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1659 if (ret < 0) {
1660 ret = close(wait_shm_fd);
1661 if (!ret) {
1662 PERROR("Error closing fd");
1663 }
1664 lttng_ust_unlock_fd_tracker();
1665 goto error;
1666 }
1667
1668 wait_shm_fd = ret;
1669 lttng_ust_unlock_fd_tracker();
1670
1671 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1672 MAP_SHARED, wait_shm_fd, 0);
1673
1674 /* close shm fd immediately after taking the mmap reference */
1675 lttng_ust_lock_fd_tracker();
1676 ret = close(wait_shm_fd);
1677 if (!ret) {
1678 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1679 } else {
1680 PERROR("Error closing fd");
1681 }
1682 lttng_ust_unlock_fd_tracker();
1683
1684 if (wait_shm_mmap == MAP_FAILED) {
1685 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1686 goto error;
1687 }
1688 return wait_shm_mmap;
1689
1690 error:
1691 return NULL;
1692 }
1693
1694 static
1695 void wait_for_sessiond(struct sock_info *sock_info)
1696 {
1697 /* Use ust_lock to check if we should quit. */
1698 if (ust_lock()) {
1699 goto quit;
1700 }
1701 if (wait_poll_fallback) {
1702 goto error;
1703 }
1704 ust_unlock();
1705
1706 assert(sock_info->wait_shm_mmap);
1707
1708 DBG("Waiting for %s apps sessiond", sock_info->name);
1709 /* Wait for futex wakeup */
1710 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1711 goto end_wait;
1712
1713 while (lttng_ust_futex_async((int32_t *) sock_info->wait_shm_mmap,
1714 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1715 switch (errno) {
1716 case EWOULDBLOCK:
1717 /* Value already changed. */
1718 goto end_wait;
1719 case EINTR:
1720 /* Retry if interrupted by signal. */
1721 break; /* Get out of switch. */
1722 case EFAULT:
1723 wait_poll_fallback = 1;
1724 DBG(
1725 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1726 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1727 "Please upgrade your kernel "
1728 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1729 "mainline). LTTng-UST will use polling mode fallback.");
1730 if (ust_debug())
1731 PERROR("futex");
1732 goto end_wait;
1733 }
1734 }
1735 end_wait:
1736 return;
1737
1738 quit:
1739 ust_unlock();
1740 return;
1741
1742 error:
1743 ust_unlock();
1744 return;
1745 }
1746
1747 /*
1748 * This thread does not allocate any resource, except within
1749 * handle_message, within mutex protection. This mutex protects against
1750 * fork and exit.
1751 * The other moment it allocates resources is at socket connection, which
1752 * is also protected by the mutex.
1753 */
1754 static
1755 void *ust_listener_thread(void *arg)
1756 {
1757 struct sock_info *sock_info = arg;
1758 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1759 long timeout;
1760
1761 lttng_ust_fixup_tls();
1762 /*
1763 * If available, add '-ust' to the end of this thread's
1764 * process name
1765 */
1766 ret = lttng_ust_setustprocname();
1767 if (ret) {
1768 ERR("Unable to set UST process name");
1769 }
1770
1771 /* Restart trying to connect to the session daemon */
1772 restart:
1773 if (prev_connect_failed) {
1774 /* Wait for sessiond availability with pipe */
1775 wait_for_sessiond(sock_info);
1776 if (has_waited) {
1777 has_waited = 0;
1778 /*
1779 * Sleep for 5 seconds before retrying after a
1780 * sequence of failure / wait / failure. This
1781 * deals with a killed or broken session daemon.
1782 */
1783 sleep(5);
1784 } else {
1785 has_waited = 1;
1786 }
1787 prev_connect_failed = 0;
1788 }
1789
1790 if (ust_lock()) {
1791 goto quit;
1792 }
1793
1794 if (sock_info->socket != -1) {
1795 /* FD tracker is updated by ustcomm_close_unix_sock() */
1796 ret = ustcomm_close_unix_sock(sock_info->socket);
1797 if (ret) {
1798 ERR("Error closing %s ust cmd socket",
1799 sock_info->name);
1800 }
1801 sock_info->socket = -1;
1802 }
1803 if (sock_info->notify_socket != -1) {
1804 /* FD tracker is updated by ustcomm_close_unix_sock() */
1805 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1806 if (ret) {
1807 ERR("Error closing %s ust notify socket",
1808 sock_info->name);
1809 }
1810 sock_info->notify_socket = -1;
1811 }
1812
1813
1814 /*
1815 * Register. We need to perform both connect and sending
1816 * registration message before doing the next connect otherwise
1817 * we may reach unix socket connect queue max limits and block
1818 * on the 2nd connect while the session daemon is awaiting the
1819 * first connect registration message.
1820 */
1821 /* Connect cmd socket */
1822 lttng_ust_lock_fd_tracker();
1823 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1824 get_connect_sock_timeout());
1825 if (ret < 0) {
1826 lttng_ust_unlock_fd_tracker();
1827 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1828 prev_connect_failed = 1;
1829
1830 /*
1831 * If we cannot find the sessiond daemon, don't delay
1832 * constructor execution.
1833 */
1834 ret = handle_register_failed(sock_info);
1835 assert(!ret);
1836 ust_unlock();
1837 goto restart;
1838 }
1839 fd = ret;
1840 ret = lttng_ust_add_fd_to_tracker(fd);
1841 if (ret < 0) {
1842 ret = close(fd);
1843 if (ret) {
1844 PERROR("close on sock_info->socket");
1845 }
1846 ret = -1;
1847 lttng_ust_unlock_fd_tracker();
1848 ust_unlock();
1849 goto quit;
1850 }
1851
1852 sock_info->socket = ret;
1853 lttng_ust_unlock_fd_tracker();
1854
1855 ust_unlock();
1856 /*
1857 * Unlock/relock ust lock because connect is blocking (with
1858 * timeout). Don't delay constructors on the ust lock for too
1859 * long.
1860 */
1861 if (ust_lock()) {
1862 goto quit;
1863 }
1864
1865 /*
1866 * Create only one root handle per listener thread for the whole
1867 * process lifetime, so we ensure we get ID which is statically
1868 * assigned to the root handle.
1869 */
1870 if (sock_info->root_handle == -1) {
1871 ret = lttng_abi_create_root_handle();
1872 if (ret < 0) {
1873 ERR("Error creating root handle");
1874 goto quit;
1875 }
1876 sock_info->root_handle = ret;
1877 }
1878
1879 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1880 if (ret < 0) {
1881 ERR("Error registering to %s ust cmd socket",
1882 sock_info->name);
1883 prev_connect_failed = 1;
1884 /*
1885 * If we cannot register to the sessiond daemon, don't
1886 * delay constructor execution.
1887 */
1888 ret = handle_register_failed(sock_info);
1889 assert(!ret);
1890 ust_unlock();
1891 goto restart;
1892 }
1893
1894 ust_unlock();
1895 /*
1896 * Unlock/relock ust lock because connect is blocking (with
1897 * timeout). Don't delay constructors on the ust lock for too
1898 * long.
1899 */
1900 if (ust_lock()) {
1901 goto quit;
1902 }
1903
1904 /* Connect notify socket */
1905 lttng_ust_lock_fd_tracker();
1906 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1907 get_connect_sock_timeout());
1908 if (ret < 0) {
1909 lttng_ust_unlock_fd_tracker();
1910 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1911 prev_connect_failed = 1;
1912
1913 /*
1914 * If we cannot find the sessiond daemon, don't delay
1915 * constructor execution.
1916 */
1917 ret = handle_register_failed(sock_info);
1918 assert(!ret);
1919 ust_unlock();
1920 goto restart;
1921 }
1922
1923 fd = ret;
1924 ret = lttng_ust_add_fd_to_tracker(fd);
1925 if (ret < 0) {
1926 ret = close(fd);
1927 if (ret) {
1928 PERROR("close on sock_info->notify_socket");
1929 }
1930 ret = -1;
1931 lttng_ust_unlock_fd_tracker();
1932 ust_unlock();
1933 goto quit;
1934 }
1935
1936 sock_info->notify_socket = ret;
1937 lttng_ust_unlock_fd_tracker();
1938
1939 ust_unlock();
1940 /*
1941 * Unlock/relock ust lock because connect is blocking (with
1942 * timeout). Don't delay constructors on the ust lock for too
1943 * long.
1944 */
1945 if (ust_lock()) {
1946 goto quit;
1947 }
1948
1949 timeout = get_notify_sock_timeout();
1950 if (timeout >= 0) {
1951 /*
1952 * Give at least 10ms to sessiond to reply to
1953 * notifications.
1954 */
1955 if (timeout < 10)
1956 timeout = 10;
1957 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1958 timeout);
1959 if (ret < 0) {
1960 WARN("Error setting socket receive timeout");
1961 }
1962 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1963 timeout);
1964 if (ret < 0) {
1965 WARN("Error setting socket send timeout");
1966 }
1967 } else if (timeout < -1) {
1968 WARN("Unsupported timeout value %ld", timeout);
1969 }
1970
1971 ret = register_to_sessiond(sock_info->notify_socket,
1972 USTCTL_SOCKET_NOTIFY);
1973 if (ret < 0) {
1974 ERR("Error registering to %s ust notify socket",
1975 sock_info->name);
1976 prev_connect_failed = 1;
1977 /*
1978 * If we cannot register to the sessiond daemon, don't
1979 * delay constructor execution.
1980 */
1981 ret = handle_register_failed(sock_info);
1982 assert(!ret);
1983 ust_unlock();
1984 goto restart;
1985 }
1986 sock = sock_info->socket;
1987
1988 ust_unlock();
1989
1990 for (;;) {
1991 ssize_t len;
1992 struct ustcomm_ust_msg lum;
1993
1994 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1995 switch (len) {
1996 case 0: /* orderly shutdown */
1997 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1998 if (ust_lock()) {
1999 goto quit;
2000 }
2001 /*
2002 * Either sessiond has shutdown or refused us by closing the socket.
2003 * In either case, we don't want to delay construction execution,
2004 * and we need to wait before retry.
2005 */
2006 prev_connect_failed = 1;
2007 /*
2008 * If we cannot register to the sessiond daemon, don't
2009 * delay constructor execution.
2010 */
2011 ret = handle_register_failed(sock_info);
2012 assert(!ret);
2013 ust_unlock();
2014 goto end;
2015 case sizeof(lum):
2016 print_cmd(lum.cmd, lum.handle);
2017 ret = handle_message(sock_info, sock, &lum);
2018 if (ret) {
2019 ERR("Error handling message for %s socket",
2020 sock_info->name);
2021 /*
2022 * Close socket if protocol error is
2023 * detected.
2024 */
2025 goto end;
2026 }
2027 continue;
2028 default:
2029 if (len < 0) {
2030 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
2031 } else {
2032 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
2033 }
2034 if (len == -ECONNRESET) {
2035 DBG("%s remote end closed connection", sock_info->name);
2036 goto end;
2037 }
2038 goto end;
2039 }
2040
2041 }
2042 end:
2043 if (ust_lock()) {
2044 goto quit;
2045 }
2046 /* Cleanup socket handles before trying to reconnect */
2047 lttng_ust_objd_table_owner_cleanup(sock_info);
2048 ust_unlock();
2049 goto restart; /* try to reconnect */
2050
2051 quit:
2052 ust_unlock();
2053
2054 pthread_mutex_lock(&ust_exit_mutex);
2055 sock_info->thread_active = 0;
2056 pthread_mutex_unlock(&ust_exit_mutex);
2057 return NULL;
2058 }
2059
2060 /*
2061 * Weak symbol to call when the ust malloc wrapper is not loaded.
2062 */
2063 __attribute__((weak))
2064 void lttng_ust_malloc_wrapper_init(void)
2065 {
2066 }
2067
2068 /*
2069 * sessiond monitoring thread: monitor presence of global and per-user
2070 * sessiond by polling the application common named pipe.
2071 */
2072 void __attribute__((constructor)) lttng_ust_init(void)
2073 {
2074 struct timespec constructor_timeout;
2075 sigset_t sig_all_blocked, orig_parent_mask;
2076 pthread_attr_t thread_attr;
2077 int timeout_mode;
2078 int ret;
2079 void *handle;
2080
2081 if (uatomic_xchg(&initialized, 1) == 1)
2082 return;
2083
2084 /*
2085 * Fixup interdependency between TLS fixup mutex (which happens
2086 * to be the dynamic linker mutex) and ust_lock, taken within
2087 * the ust lock.
2088 */
2089 lttng_ust_fixup_tls();
2090
2091 lttng_ust_loaded = 1;
2092
2093 /*
2094 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2095 * the unloading of code used by the ust_listener_threads as we can not
2096 * reliably know when they exited. To do that, manually load
2097 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2098 * this library so it never becomes zero, thus never gets unloaded from the
2099 * address space of the process. Since we are already running in the
2100 * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
2101 * simply increment the refcount and no additionnal work is needed by the
2102 * dynamic loader as the shared library is already loaded in the address
2103 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2104 * unloading of the UST library if its refcount becomes zero (which should
2105 * never happen). Do the return value check but discard the handle at the
2106 * end of the function as it's not needed.
2107 */
2108 handle = dlopen(LTTNG_UST_LIB_SO_NAME, RTLD_LAZY | RTLD_NODELETE);
2109 if (!handle) {
2110 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME);
2111 }
2112
2113 /*
2114 * We want precise control over the order in which we construct
2115 * our sub-libraries vs starting to receive commands from
2116 * sessiond (otherwise leading to errors when trying to create
2117 * sessiond before the init functions are completed).
2118 */
2119 init_usterr();
2120 lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */
2121 init_tracepoint();
2122 lttng_ust_init_fd_tracker();
2123 lttng_ust_clock_init();
2124 lttng_ust_getcpu_init();
2125 lttng_ust_statedump_init();
2126 lttng_ring_buffer_metadata_client_init();
2127 lttng_ring_buffer_client_overwrite_init();
2128 lttng_ring_buffer_client_overwrite_rt_init();
2129 lttng_ring_buffer_client_discard_init();
2130 lttng_ring_buffer_client_discard_rt_init();
2131 lttng_counter_client_percpu_32_modular_init();
2132 lttng_counter_client_percpu_64_modular_init();
2133 lttng_perf_counter_init();
2134 /*
2135 * Invoke ust malloc wrapper init before starting other threads.
2136 */
2137 lttng_ust_malloc_wrapper_init();
2138
2139 timeout_mode = get_constructor_timeout(&constructor_timeout);
2140
2141 get_allow_blocking();
2142
2143 ret = sem_init(&constructor_wait, 0, 0);
2144 if (ret) {
2145 PERROR("sem_init");
2146 }
2147
2148 ret = setup_global_apps();
2149 if (ret) {
2150 assert(global_apps.allowed == 0);
2151 DBG("global apps setup returned %d", ret);
2152 }
2153
2154 ret = setup_local_apps();
2155 if (ret) {
2156 assert(local_apps.allowed == 0);
2157 DBG("local apps setup returned %d", ret);
2158 }
2159
2160 /* A new thread created by pthread_create inherits the signal mask
2161 * from the parent. To avoid any signal being received by the
2162 * listener thread, we block all signals temporarily in the parent,
2163 * while we create the listener thread.
2164 */
2165 sigfillset(&sig_all_blocked);
2166 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
2167 if (ret) {
2168 ERR("pthread_sigmask: %s", strerror(ret));
2169 }
2170
2171 ret = pthread_attr_init(&thread_attr);
2172 if (ret) {
2173 ERR("pthread_attr_init: %s", strerror(ret));
2174 }
2175 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
2176 if (ret) {
2177 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
2178 }
2179
2180 if (global_apps.allowed) {
2181 pthread_mutex_lock(&ust_exit_mutex);
2182 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
2183 ust_listener_thread, &global_apps);
2184 if (ret) {
2185 ERR("pthread_create global: %s", strerror(ret));
2186 }
2187 global_apps.thread_active = 1;
2188 pthread_mutex_unlock(&ust_exit_mutex);
2189 } else {
2190 handle_register_done(&global_apps);
2191 }
2192
2193 if (local_apps.allowed) {
2194 pthread_mutex_lock(&ust_exit_mutex);
2195 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
2196 ust_listener_thread, &local_apps);
2197 if (ret) {
2198 ERR("pthread_create local: %s", strerror(ret));
2199 }
2200 local_apps.thread_active = 1;
2201 pthread_mutex_unlock(&ust_exit_mutex);
2202 } else {
2203 handle_register_done(&local_apps);
2204 }
2205 ret = pthread_attr_destroy(&thread_attr);
2206 if (ret) {
2207 ERR("pthread_attr_destroy: %s", strerror(ret));
2208 }
2209
2210 /* Restore original signal mask in parent */
2211 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
2212 if (ret) {
2213 ERR("pthread_sigmask: %s", strerror(ret));
2214 }
2215
2216 switch (timeout_mode) {
2217 case 1: /* timeout wait */
2218 do {
2219 ret = sem_timedwait(&constructor_wait,
2220 &constructor_timeout);
2221 } while (ret < 0 && errno == EINTR);
2222 if (ret < 0) {
2223 switch (errno) {
2224 case ETIMEDOUT:
2225 ERR("Timed out waiting for lttng-sessiond");
2226 break;
2227 case EINVAL:
2228 PERROR("sem_timedwait");
2229 break;
2230 default:
2231 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2232 strerror(errno));
2233 }
2234 }
2235 break;
2236 case -1:/* wait forever */
2237 do {
2238 ret = sem_wait(&constructor_wait);
2239 } while (ret < 0 && errno == EINTR);
2240 if (ret < 0) {
2241 switch (errno) {
2242 case EINVAL:
2243 PERROR("sem_wait");
2244 break;
2245 default:
2246 ERR("Unexpected error \"%s\" returned by sem_wait",
2247 strerror(errno));
2248 }
2249 }
2250 break;
2251 case 0: /* no timeout */
2252 break;
2253 }
2254 }
2255
2256 static
2257 void lttng_ust_cleanup(int exiting)
2258 {
2259 cleanup_sock_info(&global_apps, exiting);
2260 cleanup_sock_info(&local_apps, exiting);
2261 local_apps.allowed = 0;
2262 global_apps.allowed = 0;
2263 /*
2264 * The teardown in this function all affect data structures
2265 * accessed under the UST lock by the listener thread. This
2266 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2267 * that none of these threads are accessing this data at this
2268 * point.
2269 */
2270 lttng_ust_abi_exit();
2271 lttng_ust_events_exit();
2272 lttng_perf_counter_exit();
2273 lttng_ring_buffer_client_discard_rt_exit();
2274 lttng_ring_buffer_client_discard_exit();
2275 lttng_ring_buffer_client_overwrite_rt_exit();
2276 lttng_ring_buffer_client_overwrite_exit();
2277 lttng_ring_buffer_metadata_client_exit();
2278 lttng_counter_client_percpu_32_modular_exit();
2279 lttng_counter_client_percpu_64_modular_exit();
2280 lttng_ust_statedump_destroy();
2281 exit_tracepoint();
2282 if (!exiting) {
2283 /* Reinitialize values for fork */
2284 sem_count = sem_count_initial_value;
2285 lttng_ust_comm_should_quit = 0;
2286 initialized = 0;
2287 }
2288 }
2289
2290 void __attribute__((destructor)) lttng_ust_exit(void)
2291 {
2292 int ret;
2293
2294 /*
2295 * Using pthread_cancel here because:
2296 * A) we don't want to hang application teardown.
2297 * B) the thread is not allocating any resource.
2298 */
2299
2300 /*
2301 * Require the communication thread to quit. Synchronize with
2302 * mutexes to ensure it is not in a mutex critical section when
2303 * pthread_cancel is later called.
2304 */
2305 ust_lock_nocheck();
2306 lttng_ust_comm_should_quit = 1;
2307 ust_unlock();
2308
2309 pthread_mutex_lock(&ust_exit_mutex);
2310 /* cancel threads */
2311 if (global_apps.thread_active) {
2312 ret = pthread_cancel(global_apps.ust_listener);
2313 if (ret) {
2314 ERR("Error cancelling global ust listener thread: %s",
2315 strerror(ret));
2316 } else {
2317 global_apps.thread_active = 0;
2318 }
2319 }
2320 if (local_apps.thread_active) {
2321 ret = pthread_cancel(local_apps.ust_listener);
2322 if (ret) {
2323 ERR("Error cancelling local ust listener thread: %s",
2324 strerror(ret));
2325 } else {
2326 local_apps.thread_active = 0;
2327 }
2328 }
2329 pthread_mutex_unlock(&ust_exit_mutex);
2330
2331 /*
2332 * Do NOT join threads: use of sys_futex makes it impossible to
2333 * join the threads without using async-cancel, but async-cancel
2334 * is delivered by a signal, which could hit the target thread
2335 * anywhere in its code path, including while the ust_lock() is
2336 * held, causing a deadlock for the other thread. Let the OS
2337 * cleanup the threads if there are stalled in a syscall.
2338 */
2339 lttng_ust_cleanup(1);
2340 }
2341
2342 static
2343 void ust_context_ns_reset(void)
2344 {
2345 lttng_context_pid_ns_reset();
2346 lttng_context_cgroup_ns_reset();
2347 lttng_context_ipc_ns_reset();
2348 lttng_context_mnt_ns_reset();
2349 lttng_context_net_ns_reset();
2350 lttng_context_user_ns_reset();
2351 lttng_context_time_ns_reset();
2352 lttng_context_uts_ns_reset();
2353 }
2354
2355 static
2356 void ust_context_vuids_reset(void)
2357 {
2358 lttng_context_vuid_reset();
2359 lttng_context_veuid_reset();
2360 lttng_context_vsuid_reset();
2361 }
2362
2363 static
2364 void ust_context_vgids_reset(void)
2365 {
2366 lttng_context_vgid_reset();
2367 lttng_context_vegid_reset();
2368 lttng_context_vsgid_reset();
2369 }
2370
2371 /*
2372 * We exclude the worker threads across fork and clone (except
2373 * CLONE_VM), because these system calls only keep the forking thread
2374 * running in the child. Therefore, we don't want to call fork or clone
2375 * in the middle of an tracepoint or ust tracing state modification.
2376 * Holding this mutex protects these structures across fork and clone.
2377 */
2378 void ust_before_fork(sigset_t *save_sigset)
2379 {
2380 /*
2381 * Disable signals. This is to avoid that the child intervenes
2382 * before it is properly setup for tracing. It is safer to
2383 * disable all signals, because then we know we are not breaking
2384 * anything by restoring the original mask.
2385 */
2386 sigset_t all_sigs;
2387 int ret;
2388
2389 /* Fixup lttng-ust TLS. */
2390 lttng_ust_fixup_tls();
2391
2392 if (URCU_TLS(lttng_ust_nest_count))
2393 return;
2394 /* Disable signals */
2395 sigfillset(&all_sigs);
2396 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2397 if (ret == -1) {
2398 PERROR("sigprocmask");
2399 }
2400
2401 pthread_mutex_lock(&ust_fork_mutex);
2402
2403 ust_lock_nocheck();
2404 lttng_ust_urcu_before_fork();
2405 if (lttng_ust_liburcu_bp_before_fork)
2406 lttng_ust_liburcu_bp_before_fork();
2407 lttng_ust_lock_fd_tracker();
2408 lttng_perf_lock();
2409 }
2410
2411 static void ust_after_fork_common(sigset_t *restore_sigset)
2412 {
2413 int ret;
2414
2415 DBG("process %d", getpid());
2416 lttng_perf_unlock();
2417 lttng_ust_unlock_fd_tracker();
2418 ust_unlock();
2419
2420 pthread_mutex_unlock(&ust_fork_mutex);
2421
2422 /* Restore signals */
2423 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2424 if (ret == -1) {
2425 PERROR("sigprocmask");
2426 }
2427 }
2428
2429 void ust_after_fork_parent(sigset_t *restore_sigset)
2430 {
2431 if (URCU_TLS(lttng_ust_nest_count))
2432 return;
2433 DBG("process %d", getpid());
2434 lttng_ust_urcu_after_fork_parent();
2435 if (lttng_ust_liburcu_bp_after_fork_parent)
2436 lttng_ust_liburcu_bp_after_fork_parent();
2437 /* Release mutexes and reenable signals */
2438 ust_after_fork_common(restore_sigset);
2439 }
2440
2441 /*
2442 * After fork, in the child, we need to cleanup all the leftover state,
2443 * except the worker thread which already magically disappeared thanks
2444 * to the weird Linux fork semantics. After tyding up, we call
2445 * lttng_ust_init() again to start over as a new PID.
2446 *
2447 * This is meant for forks() that have tracing in the child between the
2448 * fork and following exec call (if there is any).
2449 */
2450 void ust_after_fork_child(sigset_t *restore_sigset)
2451 {
2452 if (URCU_TLS(lttng_ust_nest_count))
2453 return;
2454 lttng_context_vpid_reset();
2455 lttng_context_vtid_reset();
2456 lttng_context_procname_reset();
2457 ust_context_ns_reset();
2458 ust_context_vuids_reset();
2459 ust_context_vgids_reset();
2460 DBG("process %d", getpid());
2461 /* Release urcu mutexes */
2462 lttng_ust_urcu_after_fork_child();
2463 if (lttng_ust_liburcu_bp_after_fork_child)
2464 lttng_ust_liburcu_bp_after_fork_child();
2465 lttng_ust_cleanup(0);
2466 /* Release mutexes and reenable signals */
2467 ust_after_fork_common(restore_sigset);
2468 lttng_ust_init();
2469 }
2470
2471 void ust_after_setns(void)
2472 {
2473 ust_context_ns_reset();
2474 ust_context_vuids_reset();
2475 ust_context_vgids_reset();
2476 }
2477
2478 void ust_after_unshare(void)
2479 {
2480 ust_context_ns_reset();
2481 ust_context_vuids_reset();
2482 ust_context_vgids_reset();
2483 }
2484
2485 void ust_after_setuid(void)
2486 {
2487 ust_context_vuids_reset();
2488 }
2489
2490 void ust_after_seteuid(void)
2491 {
2492 ust_context_vuids_reset();
2493 }
2494
2495 void ust_after_setreuid(void)
2496 {
2497 ust_context_vuids_reset();
2498 }
2499
2500 void ust_after_setresuid(void)
2501 {
2502 ust_context_vuids_reset();
2503 }
2504
2505 void ust_after_setgid(void)
2506 {
2507 ust_context_vgids_reset();
2508 }
2509
2510 void ust_after_setegid(void)
2511 {
2512 ust_context_vgids_reset();
2513 }
2514
2515 void ust_after_setregid(void)
2516 {
2517 ust_context_vgids_reset();
2518 }
2519
2520 void ust_after_setresgid(void)
2521 {
2522 ust_context_vgids_reset();
2523 }
2524
2525 void lttng_ust_sockinfo_session_enabled(void *owner)
2526 {
2527 struct sock_info *sock_info = owner;
2528 sock_info->statedump_pending = 1;
2529 }
This page took 0.11876 seconds and 3 git commands to generate.