projects / lttng-ust.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Fix: stream fd leaks on error
[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;
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 error_free_bytecode;
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 error_free_bytecode;
846 }
847 ret = len;
848 goto error_free_bytecode;
849 } else {
850 DBG("Incorrect %s bytecode data message size: %zd",
851 bytecode_type_str(lum->cmd), len);
852 ret = -EINVAL;
853 goto error_free_bytecode;
854 }
855 }
856
857 ops = objd_ops(lum->handle);
858 if (!ops) {
859 ret = -ENOENT;
860 goto error_free_bytecode;
861 }
862
863 if (ops->cmd) {
864 ret = ops->cmd(lum->handle, lum->cmd,
865 (unsigned long) bytecode,
866 NULL, sock_info);
867 if (ret)
868 goto error_free_bytecode;
869 /* don't free bytecode if everything went fine. */
870 } else {
871 ret = -ENOSYS;
872 goto error_free_bytecode;
873 }
874
875 goto end;
876
877 error_free_bytecode:
878 free(bytecode);
879 end:
880 return ret;
881 }
882
883 static
884 int handle_message(struct sock_info *sock_info,
885 int sock, struct ustcomm_ust_msg *lum)
886 {
887 int ret = 0;
888 const struct lttng_ust_objd_ops *ops;
889 struct ustcomm_ust_reply lur;
890 union ust_args args;
891 char ctxstr[LTTNG_UST_SYM_NAME_LEN]; /* App context string. */
892 ssize_t len;
893
894 memset(&lur, 0, sizeof(lur));
895
896 if (ust_lock()) {
897 ret = -LTTNG_UST_ERR_EXITING;
898 goto error;
899 }
900
901 ops = objd_ops(lum->handle);
902 if (!ops) {
903 ret = -ENOENT;
904 goto error;
905 }
906
907 switch (lum->cmd) {
908 case LTTNG_UST_REGISTER_DONE:
909 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
910 ret = handle_register_done(sock_info);
911 else
912 ret = -EINVAL;
913 break;
914 case LTTNG_UST_RELEASE:
915 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
916 ret = -EPERM;
917 else
918 ret = lttng_ust_objd_unref(lum->handle, 1);
919 break;
920 case LTTNG_UST_CAPTURE:
921 case LTTNG_UST_FILTER:
922 ret = handle_bytecode_recv(sock_info, sock, lum);
923 if (ret)
924 goto error;
925 break;
926 case LTTNG_UST_EXCLUSION:
927 {
928 /* Receive exclusion names */
929 struct lttng_ust_excluder_node *node;
930 unsigned int count;
931
932 count = lum->u.exclusion.count;
933 if (count == 0) {
934 /* There are no names to read */
935 ret = 0;
936 goto error;
937 }
938 node = zmalloc(sizeof(*node) +
939 count * LTTNG_UST_SYM_NAME_LEN);
940 if (!node) {
941 ret = -ENOMEM;
942 goto error;
943 }
944 node->excluder.count = count;
945 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
946 count * LTTNG_UST_SYM_NAME_LEN);
947 switch (len) {
948 case 0: /* orderly shutdown */
949 ret = 0;
950 free(node);
951 goto error;
952 default:
953 if (len == count * LTTNG_UST_SYM_NAME_LEN) {
954 DBG("Exclusion data received");
955 break;
956 } else if (len < 0) {
957 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
958 if (len == -ECONNRESET) {
959 ERR("%s remote end closed connection", sock_info->name);
960 ret = len;
961 free(node);
962 goto error;
963 }
964 ret = len;
965 free(node);
966 goto error;
967 } else {
968 DBG("Incorrect exclusion data message size: %zd", len);
969 ret = -EINVAL;
970 free(node);
971 goto error;
972 }
973 }
974 if (ops->cmd) {
975 ret = ops->cmd(lum->handle, lum->cmd,
976 (unsigned long) node,
977 &args, sock_info);
978 if (ret) {
979 free(node);
980 }
981 /* Don't free exclusion data if everything went fine. */
982 } else {
983 ret = -ENOSYS;
984 free(node);
985 }
986 break;
987 }
988 case LTTNG_UST_EVENT_NOTIFIER_GROUP_CREATE:
989 {
990 int event_notifier_notif_fd;
991
992 len = ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock,
993 &event_notifier_notif_fd);
994 switch (len) {
995 case 0: /* orderly shutdown */
996 ret = 0;
997 goto error;
998 case 1:
999 break;
1000 default:
1001 if (len < 0) {
1002 DBG("Receive failed from lttng-sessiond with errno %d",
1003 (int) -len);
1004 if (len == -ECONNRESET) {
1005 ERR("%s remote end closed connection",
1006 sock_info->name);
1007 ret = len;
1008 goto error;
1009 }
1010 ret = len;
1011 goto error;
1012 } else {
1013 DBG("Incorrect event notifier fd message size: %zd",
1014 len);
1015 ret = -EINVAL;
1016 goto error;
1017 }
1018 }
1019 args.event_notifier_handle.event_notifier_notif_fd =
1020 event_notifier_notif_fd;
1021 if (ops->cmd)
1022 ret = ops->cmd(lum->handle, lum->cmd,
1023 (unsigned long) &lum->u,
1024 &args, sock_info);
1025 else
1026 ret = -ENOSYS;
1027 break;
1028 }
1029 case LTTNG_UST_CHANNEL:
1030 {
1031 void *chan_data;
1032 int wakeup_fd;
1033
1034 len = ustcomm_recv_channel_from_sessiond(sock,
1035 &chan_data, lum->u.channel.len,
1036 &wakeup_fd);
1037 switch (len) {
1038 case 0: /* orderly shutdown */
1039 ret = 0;
1040 goto error;
1041 default:
1042 if (len == lum->u.channel.len) {
1043 DBG("channel data received");
1044 break;
1045 } else if (len < 0) {
1046 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1047 if (len == -ECONNRESET) {
1048 ERR("%s remote end closed connection", sock_info->name);
1049 ret = len;
1050 goto error;
1051 }
1052 ret = len;
1053 goto error;
1054 } else {
1055 DBG("incorrect channel data message size: %zd", len);
1056 ret = -EINVAL;
1057 goto error;
1058 }
1059 }
1060 args.channel.chan_data = chan_data;
1061 args.channel.wakeup_fd = wakeup_fd;
1062 if (ops->cmd)
1063 ret = ops->cmd(lum->handle, lum->cmd,
1064 (unsigned long) &lum->u,
1065 &args, sock_info);
1066 else
1067 ret = -ENOSYS;
1068 if (args.channel.wakeup_fd >= 0) {
1069 int close_ret;
1070
1071 lttng_ust_lock_fd_tracker();
1072 close_ret = close(args.channel.wakeup_fd);
1073 lttng_ust_unlock_fd_tracker();
1074 args.channel.wakeup_fd = -1;
1075 if (close_ret)
1076 PERROR("close");
1077 }
1078 free(args.channel.chan_data);
1079 break;
1080 }
1081 case LTTNG_UST_STREAM:
1082 {
1083 int close_ret;
1084
1085 /* Receive shm_fd, wakeup_fd */
1086 ret = ustcomm_recv_stream_from_sessiond(sock,
1087 NULL,
1088 &args.stream.shm_fd,
1089 &args.stream.wakeup_fd);
1090 if (ret) {
1091 goto error;
1092 }
1093
1094 if (ops->cmd)
1095 ret = ops->cmd(lum->handle, lum->cmd,
1096 (unsigned long) &lum->u,
1097 &args, sock_info);
1098 else
1099 ret = -ENOSYS;
1100 if (args.stream.shm_fd >= 0) {
1101 lttng_ust_lock_fd_tracker();
1102 close_ret = close(args.stream.shm_fd);
1103 lttng_ust_unlock_fd_tracker();
1104 args.stream.shm_fd = -1;
1105 if (close_ret)
1106 PERROR("close");
1107 }
1108 if (args.stream.wakeup_fd >= 0) {
1109 lttng_ust_lock_fd_tracker();
1110 close_ret = close(args.stream.wakeup_fd);
1111 lttng_ust_unlock_fd_tracker();
1112 args.stream.wakeup_fd = -1;
1113 if (close_ret)
1114 PERROR("close");
1115 }
1116 break;
1117 }
1118 case LTTNG_UST_CONTEXT:
1119 switch (lum->u.context.ctx) {
1120 case LTTNG_UST_CONTEXT_APP_CONTEXT:
1121 {
1122 char *p;
1123 size_t ctxlen, recvlen;
1124
1125 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
1126 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
1127 if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
1128 ERR("Application context string length size is too large: %zu bytes",
1129 ctxlen);
1130 ret = -EINVAL;
1131 goto error;
1132 }
1133 strcpy(ctxstr, "$app.");
1134 p = &ctxstr[strlen("$app.")];
1135 recvlen = ctxlen - strlen("$app.");
1136 len = ustcomm_recv_unix_sock(sock, p, recvlen);
1137 switch (len) {
1138 case 0: /* orderly shutdown */
1139 ret = 0;
1140 goto error;
1141 default:
1142 if (len == recvlen) {
1143 DBG("app context data received");
1144 break;
1145 } else if (len < 0) {
1146 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1147 if (len == -ECONNRESET) {
1148 ERR("%s remote end closed connection", sock_info->name);
1149 ret = len;
1150 goto error;
1151 }
1152 ret = len;
1153 goto error;
1154 } else {
1155 DBG("incorrect app context data message size: %zd", len);
1156 ret = -EINVAL;
1157 goto error;
1158 }
1159 }
1160 /* Put : between provider and ctxname. */
1161 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1162 args.app_context.ctxname = ctxstr;
1163 break;
1164 }
1165 default:
1166 break;
1167 }
1168 if (ops->cmd) {
1169 ret = ops->cmd(lum->handle, lum->cmd,
1170 (unsigned long) &lum->u,
1171 &args, sock_info);
1172 } else {
1173 ret = -ENOSYS;
1174 }
1175 break;
1176 case LTTNG_UST_COUNTER:
1177 {
1178 void *counter_data;
1179
1180 len = ustcomm_recv_counter_from_sessiond(sock,
1181 &counter_data, lum->u.counter.len);
1182 switch (len) {
1183 case 0: /* orderly shutdown */
1184 ret = 0;
1185 goto error;
1186 default:
1187 if (len == lum->u.counter.len) {
1188 DBG("counter data received");
1189 break;
1190 } else if (len < 0) {
1191 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1192 if (len == -ECONNRESET) {
1193 ERR("%s remote end closed connection", sock_info->name);
1194 ret = len;
1195 goto error;
1196 }
1197 ret = len;
1198 goto error;
1199 } else {
1200 DBG("incorrect counter data message size: %zd", len);
1201 ret = -EINVAL;
1202 goto error;
1203 }
1204 }
1205 args.counter.counter_data = counter_data;
1206 if (ops->cmd)
1207 ret = ops->cmd(lum->handle, lum->cmd,
1208 (unsigned long) &lum->u,
1209 &args, sock_info);
1210 else
1211 ret = -ENOSYS;
1212 break;
1213 }
1214 case LTTNG_UST_COUNTER_GLOBAL:
1215 {
1216 /* Receive shm_fd */
1217 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1218 &args.counter_shm.shm_fd);
1219 if (ret) {
1220 goto error;
1221 }
1222
1223 if (ops->cmd)
1224 ret = ops->cmd(lum->handle, lum->cmd,
1225 (unsigned long) &lum->u,
1226 &args, sock_info);
1227 else
1228 ret = -ENOSYS;
1229 break;
1230 }
1231 case LTTNG_UST_COUNTER_CPU:
1232 {
1233 /* Receive shm_fd */
1234 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1235 &args.counter_shm.shm_fd);
1236 if (ret) {
1237 goto error;
1238 }
1239
1240 if (ops->cmd)
1241 ret = ops->cmd(lum->handle, lum->cmd,
1242 (unsigned long) &lum->u,
1243 &args, sock_info);
1244 else
1245 ret = -ENOSYS;
1246 break;
1247 }
1248 case LTTNG_UST_EVENT_NOTIFIER_CREATE:
1249 {
1250 /* Receive struct lttng_ust_event_notifier */
1251 struct lttng_ust_event_notifier event_notifier;
1252
1253 if (sizeof(event_notifier) != lum->u.event_notifier.len) {
1254 DBG("incorrect event notifier data message size: %u", lum->u.event_notifier.len);
1255 ret = -EINVAL;
1256 goto error;
1257 }
1258 len = ustcomm_recv_unix_sock(sock, &event_notifier, sizeof(event_notifier));
1259 switch (len) {
1260 case 0: /* orderly shutdown */
1261 ret = 0;
1262 goto error;
1263 default:
1264 if (len == sizeof(event_notifier)) {
1265 DBG("event notifier data received");
1266 break;
1267 } else if (len < 0) {
1268 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1269 if (len == -ECONNRESET) {
1270 ERR("%s remote end closed connection", sock_info->name);
1271 ret = len;
1272 goto error;
1273 }
1274 ret = len;
1275 goto error;
1276 } else {
1277 DBG("incorrect event notifier data message size: %zd", len);
1278 ret = -EINVAL;
1279 goto error;
1280 }
1281 }
1282 if (ops->cmd)
1283 ret = ops->cmd(lum->handle, lum->cmd,
1284 (unsigned long) &event_notifier,
1285 &args, sock_info);
1286 else
1287 ret = -ENOSYS;
1288 break;
1289 }
1290
1291 default:
1292 if (ops->cmd)
1293 ret = ops->cmd(lum->handle, lum->cmd,
1294 (unsigned long) &lum->u,
1295 &args, sock_info);
1296 else
1297 ret = -ENOSYS;
1298 break;
1299 }
1300
1301 lur.handle = lum->handle;
1302 lur.cmd = lum->cmd;
1303 lur.ret_val = ret;
1304 if (ret >= 0) {
1305 lur.ret_code = LTTNG_UST_OK;
1306 } else {
1307 /*
1308 * Use -LTTNG_UST_ERR as wildcard for UST internal
1309 * error that are not caused by the transport, except if
1310 * we already have a more precise error message to
1311 * report.
1312 */
1313 if (ret > -LTTNG_UST_ERR) {
1314 /* Translate code to UST error. */
1315 switch (ret) {
1316 case -EEXIST:
1317 lur.ret_code = -LTTNG_UST_ERR_EXIST;
1318 break;
1319 case -EINVAL:
1320 lur.ret_code = -LTTNG_UST_ERR_INVAL;
1321 break;
1322 case -ENOENT:
1323 lur.ret_code = -LTTNG_UST_ERR_NOENT;
1324 break;
1325 case -EPERM:
1326 lur.ret_code = -LTTNG_UST_ERR_PERM;
1327 break;
1328 case -ENOSYS:
1329 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
1330 break;
1331 default:
1332 lur.ret_code = -LTTNG_UST_ERR;
1333 break;
1334 }
1335 } else {
1336 lur.ret_code = ret;
1337 }
1338 }
1339 if (ret >= 0) {
1340 switch (lum->cmd) {
1341 case LTTNG_UST_TRACER_VERSION:
1342 lur.u.version = lum->u.version;
1343 break;
1344 case LTTNG_UST_TRACEPOINT_LIST_GET:
1345 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1346 break;
1347 }
1348 }
1349 DBG("Return value: %d", lur.ret_val);
1350
1351 ust_unlock();
1352
1353 /*
1354 * Performed delayed statedump operations outside of the UST
1355 * lock. We need to take the dynamic loader lock before we take
1356 * the UST lock internally within handle_pending_statedump().
1357 */
1358 handle_pending_statedump(sock_info);
1359
1360 if (ust_lock()) {
1361 ret = -LTTNG_UST_ERR_EXITING;
1362 goto error;
1363 }
1364
1365 ret = send_reply(sock, &lur);
1366 if (ret < 0) {
1367 DBG("error sending reply");
1368 goto error;
1369 }
1370
1371 /*
1372 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1373 * after the reply.
1374 */
1375 if (lur.ret_code == LTTNG_UST_OK) {
1376 switch (lum->cmd) {
1377 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
1378 len = ustcomm_send_unix_sock(sock,
1379 &args.field_list.entry,
1380 sizeof(args.field_list.entry));
1381 if (len < 0) {
1382 ret = len;
1383 goto error;
1384 }
1385 if (len != sizeof(args.field_list.entry)) {
1386 ret = -EINVAL;
1387 goto error;
1388 }
1389 }
1390 }
1391
1392 error:
1393 ust_unlock();
1394
1395 return ret;
1396 }
1397
1398 static
1399 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1400 {
1401 int ret;
1402
1403 if (sock_info->root_handle != -1) {
1404 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
1405 if (ret) {
1406 ERR("Error unref root handle");
1407 }
1408 sock_info->root_handle = -1;
1409 }
1410 sock_info->registration_done = 0;
1411 sock_info->initial_statedump_done = 0;
1412
1413 /*
1414 * wait_shm_mmap, socket and notify socket are used by listener
1415 * threads outside of the ust lock, so we cannot tear them down
1416 * ourselves, because we cannot join on these threads. Leave
1417 * responsibility of cleaning up these resources to the OS
1418 * process exit.
1419 */
1420 if (exiting)
1421 return;
1422
1423 if (sock_info->socket != -1) {
1424 ret = ustcomm_close_unix_sock(sock_info->socket);
1425 if (ret) {
1426 ERR("Error closing ust cmd socket");
1427 }
1428 sock_info->socket = -1;
1429 }
1430 if (sock_info->notify_socket != -1) {
1431 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1432 if (ret) {
1433 ERR("Error closing ust notify socket");
1434 }
1435 sock_info->notify_socket = -1;
1436 }
1437 if (sock_info->wait_shm_mmap) {
1438 long page_size;
1439
1440 page_size = LTTNG_UST_PAGE_SIZE;
1441 if (page_size <= 0) {
1442 if (!page_size) {
1443 errno = EINVAL;
1444 }
1445 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1446 } else {
1447 ret = munmap(sock_info->wait_shm_mmap, page_size);
1448 if (ret) {
1449 ERR("Error unmapping wait shm");
1450 }
1451 }
1452 sock_info->wait_shm_mmap = NULL;
1453 }
1454 }
1455
1456 /*
1457 * Using fork to set umask in the child process (not multi-thread safe).
1458 * We deal with the shm_open vs ftruncate race (happening when the
1459 * sessiond owns the shm and does not let everybody modify it, to ensure
1460 * safety against shm_unlink) by simply letting the mmap fail and
1461 * retrying after a few seconds.
1462 * For global shm, everybody has rw access to it until the sessiond
1463 * starts.
1464 */
1465 static
1466 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1467 {
1468 int wait_shm_fd, ret;
1469 pid_t pid;
1470
1471 /*
1472 * Try to open read-only.
1473 */
1474 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1475 if (wait_shm_fd >= 0) {
1476 int32_t tmp_read;
1477 ssize_t len;
1478 size_t bytes_read = 0;
1479
1480 /*
1481 * Try to read the fd. If unable to do so, try opening
1482 * it in write mode.
1483 */
1484 do {
1485 len = read(wait_shm_fd,
1486 &((char *) &tmp_read)[bytes_read],
1487 sizeof(tmp_read) - bytes_read);
1488 if (len > 0) {
1489 bytes_read += len;
1490 }
1491 } while ((len < 0 && errno == EINTR)
1492 || (len > 0 && bytes_read < sizeof(tmp_read)));
1493 if (bytes_read != sizeof(tmp_read)) {
1494 ret = close(wait_shm_fd);
1495 if (ret) {
1496 ERR("close wait_shm_fd");
1497 }
1498 goto open_write;
1499 }
1500 goto end;
1501 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1502 /*
1503 * Real-only open did not work, and it's not because the
1504 * entry was not present. It's a failure that prohibits
1505 * using shm.
1506 */
1507 ERR("Error opening shm %s", sock_info->wait_shm_path);
1508 goto end;
1509 }
1510
1511 open_write:
1512 /*
1513 * If the open failed because the file did not exist, or because
1514 * the file was not truncated yet, try creating it ourself.
1515 */
1516 URCU_TLS(lttng_ust_nest_count)++;
1517 pid = fork();
1518 URCU_TLS(lttng_ust_nest_count)--;
1519 if (pid > 0) {
1520 int status;
1521
1522 /*
1523 * Parent: wait for child to return, in which case the
1524 * shared memory map will have been created.
1525 */
1526 pid = wait(&status);
1527 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1528 wait_shm_fd = -1;
1529 goto end;
1530 }
1531 /*
1532 * Try to open read-only again after creation.
1533 */
1534 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1535 if (wait_shm_fd < 0) {
1536 /*
1537 * Real-only open did not work. It's a failure
1538 * that prohibits using shm.
1539 */
1540 ERR("Error opening shm %s", sock_info->wait_shm_path);
1541 goto end;
1542 }
1543 goto end;
1544 } else if (pid == 0) {
1545 int create_mode;
1546
1547 /* Child */
1548 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1549 if (sock_info->global)
1550 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1551 /*
1552 * We're alone in a child process, so we can modify the
1553 * process-wide umask.
1554 */
1555 umask(~create_mode);
1556 /*
1557 * Try creating shm (or get rw access).
1558 * We don't do an exclusive open, because we allow other
1559 * processes to create+ftruncate it concurrently.
1560 */
1561 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1562 O_RDWR | O_CREAT, create_mode);
1563 if (wait_shm_fd >= 0) {
1564 ret = ftruncate(wait_shm_fd, mmap_size);
1565 if (ret) {
1566 PERROR("ftruncate");
1567 _exit(EXIT_FAILURE);
1568 }
1569 _exit(EXIT_SUCCESS);
1570 }
1571 /*
1572 * For local shm, we need to have rw access to accept
1573 * opening it: this means the local sessiond will be
1574 * able to wake us up. For global shm, we open it even
1575 * if rw access is not granted, because the root.root
1576 * sessiond will be able to override all rights and wake
1577 * us up.
1578 */
1579 if (!sock_info->global && errno != EACCES) {
1580 ERR("Error opening shm %s", sock_info->wait_shm_path);
1581 _exit(EXIT_FAILURE);
1582 }
1583 /*
1584 * The shm exists, but we cannot open it RW. Report
1585 * success.
1586 */
1587 _exit(EXIT_SUCCESS);
1588 } else {
1589 return -1;
1590 }
1591 end:
1592 if (wait_shm_fd >= 0 && !sock_info->global) {
1593 struct stat statbuf;
1594
1595 /*
1596 * Ensure that our user is the owner of the shm file for
1597 * local shm. If we do not own the file, it means our
1598 * sessiond will not have access to wake us up (there is
1599 * probably a rogue process trying to fake our
1600 * sessiond). Fallback to polling method in this case.
1601 */
1602 ret = fstat(wait_shm_fd, &statbuf);
1603 if (ret) {
1604 PERROR("fstat");
1605 goto error_close;
1606 }
1607 if (statbuf.st_uid != getuid())
1608 goto error_close;
1609 }
1610 return wait_shm_fd;
1611
1612 error_close:
1613 ret = close(wait_shm_fd);
1614 if (ret) {
1615 PERROR("Error closing fd");
1616 }
1617 return -1;
1618 }
1619
1620 static
1621 char *get_map_shm(struct sock_info *sock_info)
1622 {
1623 long page_size;
1624 int wait_shm_fd, ret;
1625 char *wait_shm_mmap;
1626
1627 page_size = sysconf(_SC_PAGE_SIZE);
1628 if (page_size <= 0) {
1629 if (!page_size) {
1630 errno = EINVAL;
1631 }
1632 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1633 goto error;
1634 }
1635
1636 lttng_ust_lock_fd_tracker();
1637 wait_shm_fd = get_wait_shm(sock_info, page_size);
1638 if (wait_shm_fd < 0) {
1639 lttng_ust_unlock_fd_tracker();
1640 goto error;
1641 }
1642
1643 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1644 if (ret < 0) {
1645 ret = close(wait_shm_fd);
1646 if (!ret) {
1647 PERROR("Error closing fd");
1648 }
1649 lttng_ust_unlock_fd_tracker();
1650 goto error;
1651 }
1652
1653 wait_shm_fd = ret;
1654 lttng_ust_unlock_fd_tracker();
1655
1656 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1657 MAP_SHARED, wait_shm_fd, 0);
1658
1659 /* close shm fd immediately after taking the mmap reference */
1660 lttng_ust_lock_fd_tracker();
1661 ret = close(wait_shm_fd);
1662 if (!ret) {
1663 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1664 } else {
1665 PERROR("Error closing fd");
1666 }
1667 lttng_ust_unlock_fd_tracker();
1668
1669 if (wait_shm_mmap == MAP_FAILED) {
1670 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1671 goto error;
1672 }
1673 return wait_shm_mmap;
1674
1675 error:
1676 return NULL;
1677 }
1678
1679 static
1680 void wait_for_sessiond(struct sock_info *sock_info)
1681 {
1682 /* Use ust_lock to check if we should quit. */
1683 if (ust_lock()) {
1684 goto quit;
1685 }
1686 if (wait_poll_fallback) {
1687 goto error;
1688 }
1689 ust_unlock();
1690
1691 assert(sock_info->wait_shm_mmap);
1692
1693 DBG("Waiting for %s apps sessiond", sock_info->name);
1694 /* Wait for futex wakeup */
1695 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1696 goto end_wait;
1697
1698 while (lttng_ust_futex_async((int32_t *) sock_info->wait_shm_mmap,
1699 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1700 switch (errno) {
1701 case EWOULDBLOCK:
1702 /* Value already changed. */
1703 goto end_wait;
1704 case EINTR:
1705 /* Retry if interrupted by signal. */
1706 break; /* Get out of switch. */
1707 case EFAULT:
1708 wait_poll_fallback = 1;
1709 DBG(
1710 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1711 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1712 "Please upgrade your kernel "
1713 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1714 "mainline). LTTng-UST will use polling mode fallback.");
1715 if (ust_debug())
1716 PERROR("futex");
1717 goto end_wait;
1718 }
1719 }
1720 end_wait:
1721 return;
1722
1723 quit:
1724 ust_unlock();
1725 return;
1726
1727 error:
1728 ust_unlock();
1729 return;
1730 }
1731
1732 /*
1733 * This thread does not allocate any resource, except within
1734 * handle_message, within mutex protection. This mutex protects against
1735 * fork and exit.
1736 * The other moment it allocates resources is at socket connection, which
1737 * is also protected by the mutex.
1738 */
1739 static
1740 void *ust_listener_thread(void *arg)
1741 {
1742 struct sock_info *sock_info = arg;
1743 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1744 long timeout;
1745
1746 lttng_ust_fixup_tls();
1747 /*
1748 * If available, add '-ust' to the end of this thread's
1749 * process name
1750 */
1751 ret = lttng_ust_setustprocname();
1752 if (ret) {
1753 ERR("Unable to set UST process name");
1754 }
1755
1756 /* Restart trying to connect to the session daemon */
1757 restart:
1758 if (prev_connect_failed) {
1759 /* Wait for sessiond availability with pipe */
1760 wait_for_sessiond(sock_info);
1761 if (has_waited) {
1762 has_waited = 0;
1763 /*
1764 * Sleep for 5 seconds before retrying after a
1765 * sequence of failure / wait / failure. This
1766 * deals with a killed or broken session daemon.
1767 */
1768 sleep(5);
1769 } else {
1770 has_waited = 1;
1771 }
1772 prev_connect_failed = 0;
1773 }
1774
1775 if (ust_lock()) {
1776 goto quit;
1777 }
1778
1779 if (sock_info->socket != -1) {
1780 /* FD tracker is updated by ustcomm_close_unix_sock() */
1781 ret = ustcomm_close_unix_sock(sock_info->socket);
1782 if (ret) {
1783 ERR("Error closing %s ust cmd socket",
1784 sock_info->name);
1785 }
1786 sock_info->socket = -1;
1787 }
1788 if (sock_info->notify_socket != -1) {
1789 /* FD tracker is updated by ustcomm_close_unix_sock() */
1790 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1791 if (ret) {
1792 ERR("Error closing %s ust notify socket",
1793 sock_info->name);
1794 }
1795 sock_info->notify_socket = -1;
1796 }
1797
1798
1799 /*
1800 * Register. We need to perform both connect and sending
1801 * registration message before doing the next connect otherwise
1802 * we may reach unix socket connect queue max limits and block
1803 * on the 2nd connect while the session daemon is awaiting the
1804 * first connect registration message.
1805 */
1806 /* Connect cmd socket */
1807 lttng_ust_lock_fd_tracker();
1808 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1809 get_connect_sock_timeout());
1810 if (ret < 0) {
1811 lttng_ust_unlock_fd_tracker();
1812 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1813 prev_connect_failed = 1;
1814
1815 /*
1816 * If we cannot find the sessiond daemon, don't delay
1817 * constructor execution.
1818 */
1819 ret = handle_register_failed(sock_info);
1820 assert(!ret);
1821 ust_unlock();
1822 goto restart;
1823 }
1824 fd = ret;
1825 ret = lttng_ust_add_fd_to_tracker(fd);
1826 if (ret < 0) {
1827 ret = close(fd);
1828 if (ret) {
1829 PERROR("close on sock_info->socket");
1830 }
1831 ret = -1;
1832 lttng_ust_unlock_fd_tracker();
1833 ust_unlock();
1834 goto quit;
1835 }
1836
1837 sock_info->socket = ret;
1838 lttng_ust_unlock_fd_tracker();
1839
1840 ust_unlock();
1841 /*
1842 * Unlock/relock ust lock because connect is blocking (with
1843 * timeout). Don't delay constructors on the ust lock for too
1844 * long.
1845 */
1846 if (ust_lock()) {
1847 goto quit;
1848 }
1849
1850 /*
1851 * Create only one root handle per listener thread for the whole
1852 * process lifetime, so we ensure we get ID which is statically
1853 * assigned to the root handle.
1854 */
1855 if (sock_info->root_handle == -1) {
1856 ret = lttng_abi_create_root_handle();
1857 if (ret < 0) {
1858 ERR("Error creating root handle");
1859 goto quit;
1860 }
1861 sock_info->root_handle = ret;
1862 }
1863
1864 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1865 if (ret < 0) {
1866 ERR("Error registering to %s ust cmd socket",
1867 sock_info->name);
1868 prev_connect_failed = 1;
1869 /*
1870 * If we cannot register to the sessiond daemon, don't
1871 * delay constructor execution.
1872 */
1873 ret = handle_register_failed(sock_info);
1874 assert(!ret);
1875 ust_unlock();
1876 goto restart;
1877 }
1878
1879 ust_unlock();
1880 /*
1881 * Unlock/relock ust lock because connect is blocking (with
1882 * timeout). Don't delay constructors on the ust lock for too
1883 * long.
1884 */
1885 if (ust_lock()) {
1886 goto quit;
1887 }
1888
1889 /* Connect notify socket */
1890 lttng_ust_lock_fd_tracker();
1891 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1892 get_connect_sock_timeout());
1893 if (ret < 0) {
1894 lttng_ust_unlock_fd_tracker();
1895 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1896 prev_connect_failed = 1;
1897
1898 /*
1899 * If we cannot find the sessiond daemon, don't delay
1900 * constructor execution.
1901 */
1902 ret = handle_register_failed(sock_info);
1903 assert(!ret);
1904 ust_unlock();
1905 goto restart;
1906 }
1907
1908 fd = ret;
1909 ret = lttng_ust_add_fd_to_tracker(fd);
1910 if (ret < 0) {
1911 ret = close(fd);
1912 if (ret) {
1913 PERROR("close on sock_info->notify_socket");
1914 }
1915 ret = -1;
1916 lttng_ust_unlock_fd_tracker();
1917 ust_unlock();
1918 goto quit;
1919 }
1920
1921 sock_info->notify_socket = ret;
1922 lttng_ust_unlock_fd_tracker();
1923
1924 ust_unlock();
1925 /*
1926 * Unlock/relock ust lock because connect is blocking (with
1927 * timeout). Don't delay constructors on the ust lock for too
1928 * long.
1929 */
1930 if (ust_lock()) {
1931 goto quit;
1932 }
1933
1934 timeout = get_notify_sock_timeout();
1935 if (timeout >= 0) {
1936 /*
1937 * Give at least 10ms to sessiond to reply to
1938 * notifications.
1939 */
1940 if (timeout < 10)
1941 timeout = 10;
1942 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1943 timeout);
1944 if (ret < 0) {
1945 WARN("Error setting socket receive timeout");
1946 }
1947 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1948 timeout);
1949 if (ret < 0) {
1950 WARN("Error setting socket send timeout");
1951 }
1952 } else if (timeout < -1) {
1953 WARN("Unsupported timeout value %ld", timeout);
1954 }
1955
1956 ret = register_to_sessiond(sock_info->notify_socket,
1957 USTCTL_SOCKET_NOTIFY);
1958 if (ret < 0) {
1959 ERR("Error registering to %s ust notify socket",
1960 sock_info->name);
1961 prev_connect_failed = 1;
1962 /*
1963 * If we cannot register to the sessiond daemon, don't
1964 * delay constructor execution.
1965 */
1966 ret = handle_register_failed(sock_info);
1967 assert(!ret);
1968 ust_unlock();
1969 goto restart;
1970 }
1971 sock = sock_info->socket;
1972
1973 ust_unlock();
1974
1975 for (;;) {
1976 ssize_t len;
1977 struct ustcomm_ust_msg lum;
1978
1979 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1980 switch (len) {
1981 case 0: /* orderly shutdown */
1982 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1983 if (ust_lock()) {
1984 goto quit;
1985 }
1986 /*
1987 * Either sessiond has shutdown or refused us by closing the socket.
1988 * In either case, we don't want to delay construction execution,
1989 * and we need to wait before retry.
1990 */
1991 prev_connect_failed = 1;
1992 /*
1993 * If we cannot register to the sessiond daemon, don't
1994 * delay constructor execution.
1995 */
1996 ret = handle_register_failed(sock_info);
1997 assert(!ret);
1998 ust_unlock();
1999 goto end;
2000 case sizeof(lum):
2001 print_cmd(lum.cmd, lum.handle);
2002 ret = handle_message(sock_info, sock, &lum);
2003 if (ret) {
2004 ERR("Error handling message for %s socket",
2005 sock_info->name);
2006 /*
2007 * Close socket if protocol error is
2008 * detected.
2009 */
2010 goto end;
2011 }
2012 continue;
2013 default:
2014 if (len < 0) {
2015 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
2016 } else {
2017 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
2018 }
2019 if (len == -ECONNRESET) {
2020 DBG("%s remote end closed connection", sock_info->name);
2021 goto end;
2022 }
2023 goto end;
2024 }
2025
2026 }
2027 end:
2028 if (ust_lock()) {
2029 goto quit;
2030 }
2031 /* Cleanup socket handles before trying to reconnect */
2032 lttng_ust_objd_table_owner_cleanup(sock_info);
2033 ust_unlock();
2034 goto restart; /* try to reconnect */
2035
2036 quit:
2037 ust_unlock();
2038
2039 pthread_mutex_lock(&ust_exit_mutex);
2040 sock_info->thread_active = 0;
2041 pthread_mutex_unlock(&ust_exit_mutex);
2042 return NULL;
2043 }
2044
2045 /*
2046 * Weak symbol to call when the ust malloc wrapper is not loaded.
2047 */
2048 __attribute__((weak))
2049 void lttng_ust_malloc_wrapper_init(void)
2050 {
2051 }
2052
2053 /*
2054 * sessiond monitoring thread: monitor presence of global and per-user
2055 * sessiond by polling the application common named pipe.
2056 */
2057 void __attribute__((constructor)) lttng_ust_init(void)
2058 {
2059 struct timespec constructor_timeout;
2060 sigset_t sig_all_blocked, orig_parent_mask;
2061 pthread_attr_t thread_attr;
2062 int timeout_mode;
2063 int ret;
2064 void *handle;
2065
2066 if (uatomic_xchg(&initialized, 1) == 1)
2067 return;
2068
2069 /*
2070 * Fixup interdependency between TLS fixup mutex (which happens
2071 * to be the dynamic linker mutex) and ust_lock, taken within
2072 * the ust lock.
2073 */
2074 lttng_ust_fixup_tls();
2075
2076 lttng_ust_loaded = 1;
2077
2078 /*
2079 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2080 * the unloading of code used by the ust_listener_threads as we can not
2081 * reliably know when they exited. To do that, manually load
2082 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2083 * this library so it never becomes zero, thus never gets unloaded from the
2084 * address space of the process. Since we are already running in the
2085 * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
2086 * simply increment the refcount and no additionnal work is needed by the
2087 * dynamic loader as the shared library is already loaded in the address
2088 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2089 * unloading of the UST library if its refcount becomes zero (which should
2090 * never happen). Do the return value check but discard the handle at the
2091 * end of the function as it's not needed.
2092 */
2093 handle = dlopen(LTTNG_UST_LIB_SO_NAME, RTLD_LAZY | RTLD_NODELETE);
2094 if (!handle) {
2095 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME);
2096 }
2097
2098 /*
2099 * We want precise control over the order in which we construct
2100 * our sub-libraries vs starting to receive commands from
2101 * sessiond (otherwise leading to errors when trying to create
2102 * sessiond before the init functions are completed).
2103 */
2104 init_usterr();
2105 lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */
2106 init_tracepoint();
2107 lttng_ust_init_fd_tracker();
2108 lttng_ust_clock_init();
2109 lttng_ust_getcpu_init();
2110 lttng_ust_statedump_init();
2111 lttng_ring_buffer_metadata_client_init();
2112 lttng_ring_buffer_client_overwrite_init();
2113 lttng_ring_buffer_client_overwrite_rt_init();
2114 lttng_ring_buffer_client_discard_init();
2115 lttng_ring_buffer_client_discard_rt_init();
2116 lttng_counter_client_percpu_32_modular_init();
2117 lttng_counter_client_percpu_64_modular_init();
2118 lttng_perf_counter_init();
2119 /*
2120 * Invoke ust malloc wrapper init before starting other threads.
2121 */
2122 lttng_ust_malloc_wrapper_init();
2123
2124 timeout_mode = get_constructor_timeout(&constructor_timeout);
2125
2126 get_allow_blocking();
2127
2128 ret = sem_init(&constructor_wait, 0, 0);
2129 if (ret) {
2130 PERROR("sem_init");
2131 }
2132
2133 ret = setup_global_apps();
2134 if (ret) {
2135 assert(global_apps.allowed == 0);
2136 DBG("global apps setup returned %d", ret);
2137 }
2138
2139 ret = setup_local_apps();
2140 if (ret) {
2141 assert(local_apps.allowed == 0);
2142 DBG("local apps setup returned %d", ret);
2143 }
2144
2145 /* A new thread created by pthread_create inherits the signal mask
2146 * from the parent. To avoid any signal being received by the
2147 * listener thread, we block all signals temporarily in the parent,
2148 * while we create the listener thread.
2149 */
2150 sigfillset(&sig_all_blocked);
2151 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
2152 if (ret) {
2153 ERR("pthread_sigmask: %s", strerror(ret));
2154 }
2155
2156 ret = pthread_attr_init(&thread_attr);
2157 if (ret) {
2158 ERR("pthread_attr_init: %s", strerror(ret));
2159 }
2160 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
2161 if (ret) {
2162 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
2163 }
2164
2165 if (global_apps.allowed) {
2166 pthread_mutex_lock(&ust_exit_mutex);
2167 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
2168 ust_listener_thread, &global_apps);
2169 if (ret) {
2170 ERR("pthread_create global: %s", strerror(ret));
2171 }
2172 global_apps.thread_active = 1;
2173 pthread_mutex_unlock(&ust_exit_mutex);
2174 } else {
2175 handle_register_done(&global_apps);
2176 }
2177
2178 if (local_apps.allowed) {
2179 pthread_mutex_lock(&ust_exit_mutex);
2180 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
2181 ust_listener_thread, &local_apps);
2182 if (ret) {
2183 ERR("pthread_create local: %s", strerror(ret));
2184 }
2185 local_apps.thread_active = 1;
2186 pthread_mutex_unlock(&ust_exit_mutex);
2187 } else {
2188 handle_register_done(&local_apps);
2189 }
2190 ret = pthread_attr_destroy(&thread_attr);
2191 if (ret) {
2192 ERR("pthread_attr_destroy: %s", strerror(ret));
2193 }
2194
2195 /* Restore original signal mask in parent */
2196 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
2197 if (ret) {
2198 ERR("pthread_sigmask: %s", strerror(ret));
2199 }
2200
2201 switch (timeout_mode) {
2202 case 1: /* timeout wait */
2203 do {
2204 ret = sem_timedwait(&constructor_wait,
2205 &constructor_timeout);
2206 } while (ret < 0 && errno == EINTR);
2207 if (ret < 0) {
2208 switch (errno) {
2209 case ETIMEDOUT:
2210 ERR("Timed out waiting for lttng-sessiond");
2211 break;
2212 case EINVAL:
2213 PERROR("sem_timedwait");
2214 break;
2215 default:
2216 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2217 strerror(errno));
2218 }
2219 }
2220 break;
2221 case -1:/* wait forever */
2222 do {
2223 ret = sem_wait(&constructor_wait);
2224 } while (ret < 0 && errno == EINTR);
2225 if (ret < 0) {
2226 switch (errno) {
2227 case EINVAL:
2228 PERROR("sem_wait");
2229 break;
2230 default:
2231 ERR("Unexpected error \"%s\" returned by sem_wait",
2232 strerror(errno));
2233 }
2234 }
2235 break;
2236 case 0: /* no timeout */
2237 break;
2238 }
2239 }
2240
2241 static
2242 void lttng_ust_cleanup(int exiting)
2243 {
2244 cleanup_sock_info(&global_apps, exiting);
2245 cleanup_sock_info(&local_apps, exiting);
2246 local_apps.allowed = 0;
2247 global_apps.allowed = 0;
2248 /*
2249 * The teardown in this function all affect data structures
2250 * accessed under the UST lock by the listener thread. This
2251 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2252 * that none of these threads are accessing this data at this
2253 * point.
2254 */
2255 lttng_ust_abi_exit();
2256 lttng_ust_events_exit();
2257 lttng_perf_counter_exit();
2258 lttng_ring_buffer_client_discard_rt_exit();
2259 lttng_ring_buffer_client_discard_exit();
2260 lttng_ring_buffer_client_overwrite_rt_exit();
2261 lttng_ring_buffer_client_overwrite_exit();
2262 lttng_ring_buffer_metadata_client_exit();
2263 lttng_counter_client_percpu_32_modular_exit();
2264 lttng_counter_client_percpu_64_modular_exit();
2265 lttng_ust_statedump_destroy();
2266 exit_tracepoint();
2267 if (!exiting) {
2268 /* Reinitialize values for fork */
2269 sem_count = sem_count_initial_value;
2270 lttng_ust_comm_should_quit = 0;
2271 initialized = 0;
2272 }
2273 }
2274
2275 void __attribute__((destructor)) lttng_ust_exit(void)
2276 {
2277 int ret;
2278
2279 /*
2280 * Using pthread_cancel here because:
2281 * A) we don't want to hang application teardown.
2282 * B) the thread is not allocating any resource.
2283 */
2284
2285 /*
2286 * Require the communication thread to quit. Synchronize with
2287 * mutexes to ensure it is not in a mutex critical section when
2288 * pthread_cancel is later called.
2289 */
2290 ust_lock_nocheck();
2291 lttng_ust_comm_should_quit = 1;
2292 ust_unlock();
2293
2294 pthread_mutex_lock(&ust_exit_mutex);
2295 /* cancel threads */
2296 if (global_apps.thread_active) {
2297 ret = pthread_cancel(global_apps.ust_listener);
2298 if (ret) {
2299 ERR("Error cancelling global ust listener thread: %s",
2300 strerror(ret));
2301 } else {
2302 global_apps.thread_active = 0;
2303 }
2304 }
2305 if (local_apps.thread_active) {
2306 ret = pthread_cancel(local_apps.ust_listener);
2307 if (ret) {
2308 ERR("Error cancelling local ust listener thread: %s",
2309 strerror(ret));
2310 } else {
2311 local_apps.thread_active = 0;
2312 }
2313 }
2314 pthread_mutex_unlock(&ust_exit_mutex);
2315
2316 /*
2317 * Do NOT join threads: use of sys_futex makes it impossible to
2318 * join the threads without using async-cancel, but async-cancel
2319 * is delivered by a signal, which could hit the target thread
2320 * anywhere in its code path, including while the ust_lock() is
2321 * held, causing a deadlock for the other thread. Let the OS
2322 * cleanup the threads if there are stalled in a syscall.
2323 */
2324 lttng_ust_cleanup(1);
2325 }
2326
2327 static
2328 void ust_context_ns_reset(void)
2329 {
2330 lttng_context_pid_ns_reset();
2331 lttng_context_cgroup_ns_reset();
2332 lttng_context_ipc_ns_reset();
2333 lttng_context_mnt_ns_reset();
2334 lttng_context_net_ns_reset();
2335 lttng_context_user_ns_reset();
2336 lttng_context_time_ns_reset();
2337 lttng_context_uts_ns_reset();
2338 }
2339
2340 static
2341 void ust_context_vuids_reset(void)
2342 {
2343 lttng_context_vuid_reset();
2344 lttng_context_veuid_reset();
2345 lttng_context_vsuid_reset();
2346 }
2347
2348 static
2349 void ust_context_vgids_reset(void)
2350 {
2351 lttng_context_vgid_reset();
2352 lttng_context_vegid_reset();
2353 lttng_context_vsgid_reset();
2354 }
2355
2356 /*
2357 * We exclude the worker threads across fork and clone (except
2358 * CLONE_VM), because these system calls only keep the forking thread
2359 * running in the child. Therefore, we don't want to call fork or clone
2360 * in the middle of an tracepoint or ust tracing state modification.
2361 * Holding this mutex protects these structures across fork and clone.
2362 */
2363 void ust_before_fork(sigset_t *save_sigset)
2364 {
2365 /*
2366 * Disable signals. This is to avoid that the child intervenes
2367 * before it is properly setup for tracing. It is safer to
2368 * disable all signals, because then we know we are not breaking
2369 * anything by restoring the original mask.
2370 */
2371 sigset_t all_sigs;
2372 int ret;
2373
2374 /* Fixup lttng-ust TLS. */
2375 lttng_ust_fixup_tls();
2376
2377 if (URCU_TLS(lttng_ust_nest_count))
2378 return;
2379 /* Disable signals */
2380 sigfillset(&all_sigs);
2381 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2382 if (ret == -1) {
2383 PERROR("sigprocmask");
2384 }
2385
2386 pthread_mutex_lock(&ust_fork_mutex);
2387
2388 ust_lock_nocheck();
2389 lttng_ust_urcu_before_fork();
2390 if (lttng_ust_liburcu_bp_before_fork)
2391 lttng_ust_liburcu_bp_before_fork();
2392 lttng_ust_lock_fd_tracker();
2393 lttng_perf_lock();
2394 }
2395
2396 static void ust_after_fork_common(sigset_t *restore_sigset)
2397 {
2398 int ret;
2399
2400 DBG("process %d", getpid());
2401 lttng_perf_unlock();
2402 lttng_ust_unlock_fd_tracker();
2403 ust_unlock();
2404
2405 pthread_mutex_unlock(&ust_fork_mutex);
2406
2407 /* Restore signals */
2408 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2409 if (ret == -1) {
2410 PERROR("sigprocmask");
2411 }
2412 }
2413
2414 void ust_after_fork_parent(sigset_t *restore_sigset)
2415 {
2416 if (URCU_TLS(lttng_ust_nest_count))
2417 return;
2418 DBG("process %d", getpid());
2419 lttng_ust_urcu_after_fork_parent();
2420 if (lttng_ust_liburcu_bp_after_fork_parent)
2421 lttng_ust_liburcu_bp_after_fork_parent();
2422 /* Release mutexes and reenable signals */
2423 ust_after_fork_common(restore_sigset);
2424 }
2425
2426 /*
2427 * After fork, in the child, we need to cleanup all the leftover state,
2428 * except the worker thread which already magically disappeared thanks
2429 * to the weird Linux fork semantics. After tyding up, we call
2430 * lttng_ust_init() again to start over as a new PID.
2431 *
2432 * This is meant for forks() that have tracing in the child between the
2433 * fork and following exec call (if there is any).
2434 */
2435 void ust_after_fork_child(sigset_t *restore_sigset)
2436 {
2437 if (URCU_TLS(lttng_ust_nest_count))
2438 return;
2439 lttng_context_vpid_reset();
2440 lttng_context_vtid_reset();
2441 lttng_context_procname_reset();
2442 ust_context_ns_reset();
2443 ust_context_vuids_reset();
2444 ust_context_vgids_reset();
2445 DBG("process %d", getpid());
2446 /* Release urcu mutexes */
2447 lttng_ust_urcu_after_fork_child();
2448 if (lttng_ust_liburcu_bp_after_fork_child)
2449 lttng_ust_liburcu_bp_after_fork_child();
2450 lttng_ust_cleanup(0);
2451 /* Release mutexes and reenable signals */
2452 ust_after_fork_common(restore_sigset);
2453 lttng_ust_init();
2454 }
2455
2456 void ust_after_setns(void)
2457 {
2458 ust_context_ns_reset();
2459 ust_context_vuids_reset();
2460 ust_context_vgids_reset();
2461 }
2462
2463 void ust_after_unshare(void)
2464 {
2465 ust_context_ns_reset();
2466 ust_context_vuids_reset();
2467 ust_context_vgids_reset();
2468 }
2469
2470 void ust_after_setuid(void)
2471 {
2472 ust_context_vuids_reset();
2473 }
2474
2475 void ust_after_seteuid(void)
2476 {
2477 ust_context_vuids_reset();
2478 }
2479
2480 void ust_after_setreuid(void)
2481 {
2482 ust_context_vuids_reset();
2483 }
2484
2485 void ust_after_setresuid(void)
2486 {
2487 ust_context_vuids_reset();
2488 }
2489
2490 void ust_after_setgid(void)
2491 {
2492 ust_context_vgids_reset();
2493 }
2494
2495 void ust_after_setegid(void)
2496 {
2497 ust_context_vgids_reset();
2498 }
2499
2500 void ust_after_setregid(void)
2501 {
2502 ust_context_vgids_reset();
2503 }
2504
2505 void ust_after_setresgid(void)
2506 {
2507 ust_context_vgids_reset();
2508 }
2509
2510 void lttng_ust_sockinfo_session_enabled(void *owner)
2511 {
2512 struct sock_info *sock_info = owner;
2513 sock_info->statedump_pending = 1;
2514 }
LTTng 2.0 Userspace Tracer
This page took 0.128433 seconds and 4 git commands to generate.