projects / lttng-ust.git / blob
? search:


59f14050f1b5dfdd4d53dc73d3730d9387a96142
[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 break;
1069 }
1070 case LTTNG_UST_STREAM:
1071 {
1072 /* Receive shm_fd, wakeup_fd */
1073 ret = ustcomm_recv_stream_from_sessiond(sock,
1074 NULL,
1075 &args.stream.shm_fd,
1076 &args.stream.wakeup_fd);
1077 if (ret) {
1078 goto error;
1079 }
1080
1081 if (ops->cmd)
1082 ret = ops->cmd(lum->handle, lum->cmd,
1083 (unsigned long) &lum->u,
1084 &args, sock_info);
1085 else
1086 ret = -ENOSYS;
1087 break;
1088 }
1089 case LTTNG_UST_CONTEXT:
1090 switch (lum->u.context.ctx) {
1091 case LTTNG_UST_CONTEXT_APP_CONTEXT:
1092 {
1093 char *p;
1094 size_t ctxlen, recvlen;
1095
1096 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
1097 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
1098 if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
1099 ERR("Application context string length size is too large: %zu bytes",
1100 ctxlen);
1101 ret = -EINVAL;
1102 goto error;
1103 }
1104 strcpy(ctxstr, "$app.");
1105 p = &ctxstr[strlen("$app.")];
1106 recvlen = ctxlen - strlen("$app.");
1107 len = ustcomm_recv_unix_sock(sock, p, recvlen);
1108 switch (len) {
1109 case 0: /* orderly shutdown */
1110 ret = 0;
1111 goto error;
1112 default:
1113 if (len == recvlen) {
1114 DBG("app context data received");
1115 break;
1116 } else if (len < 0) {
1117 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1118 if (len == -ECONNRESET) {
1119 ERR("%s remote end closed connection", sock_info->name);
1120 ret = len;
1121 goto error;
1122 }
1123 ret = len;
1124 goto error;
1125 } else {
1126 DBG("incorrect app context data message size: %zd", len);
1127 ret = -EINVAL;
1128 goto error;
1129 }
1130 }
1131 /* Put : between provider and ctxname. */
1132 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1133 args.app_context.ctxname = ctxstr;
1134 break;
1135 }
1136 default:
1137 break;
1138 }
1139 if (ops->cmd) {
1140 ret = ops->cmd(lum->handle, lum->cmd,
1141 (unsigned long) &lum->u,
1142 &args, sock_info);
1143 } else {
1144 ret = -ENOSYS;
1145 }
1146 break;
1147 case LTTNG_UST_COUNTER:
1148 {
1149 void *counter_data;
1150
1151 len = ustcomm_recv_counter_from_sessiond(sock,
1152 &counter_data, lum->u.counter.len);
1153 switch (len) {
1154 case 0: /* orderly shutdown */
1155 ret = 0;
1156 goto error;
1157 default:
1158 if (len == lum->u.counter.len) {
1159 DBG("counter data received");
1160 break;
1161 } else if (len < 0) {
1162 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1163 if (len == -ECONNRESET) {
1164 ERR("%s remote end closed connection", sock_info->name);
1165 ret = len;
1166 goto error;
1167 }
1168 ret = len;
1169 goto error;
1170 } else {
1171 DBG("incorrect counter data message size: %zd", len);
1172 ret = -EINVAL;
1173 goto error;
1174 }
1175 }
1176 args.counter.counter_data = counter_data;
1177 if (ops->cmd)
1178 ret = ops->cmd(lum->handle, lum->cmd,
1179 (unsigned long) &lum->u,
1180 &args, sock_info);
1181 else
1182 ret = -ENOSYS;
1183 break;
1184 }
1185 case LTTNG_UST_COUNTER_GLOBAL:
1186 {
1187 /* Receive shm_fd */
1188 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1189 &args.counter_shm.shm_fd);
1190 if (ret) {
1191 goto error;
1192 }
1193
1194 if (ops->cmd)
1195 ret = ops->cmd(lum->handle, lum->cmd,
1196 (unsigned long) &lum->u,
1197 &args, sock_info);
1198 else
1199 ret = -ENOSYS;
1200 break;
1201 }
1202 case LTTNG_UST_COUNTER_CPU:
1203 {
1204 /* Receive shm_fd */
1205 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1206 &args.counter_shm.shm_fd);
1207 if (ret) {
1208 goto error;
1209 }
1210
1211 if (ops->cmd)
1212 ret = ops->cmd(lum->handle, lum->cmd,
1213 (unsigned long) &lum->u,
1214 &args, sock_info);
1215 else
1216 ret = -ENOSYS;
1217 break;
1218 }
1219 case LTTNG_UST_EVENT_NOTIFIER_CREATE:
1220 {
1221 /* Receive struct lttng_ust_event_notifier */
1222 struct lttng_ust_event_notifier event_notifier;
1223
1224 if (sizeof(event_notifier) != lum->u.event_notifier.len) {
1225 DBG("incorrect event notifier data message size: %u", lum->u.event_notifier.len);
1226 ret = -EINVAL;
1227 goto error;
1228 }
1229 len = ustcomm_recv_unix_sock(sock, &event_notifier, sizeof(event_notifier));
1230 switch (len) {
1231 case 0: /* orderly shutdown */
1232 ret = 0;
1233 goto error;
1234 default:
1235 if (len == sizeof(event_notifier)) {
1236 DBG("event notifier data received");
1237 break;
1238 } else if (len < 0) {
1239 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1240 if (len == -ECONNRESET) {
1241 ERR("%s remote end closed connection", sock_info->name);
1242 ret = len;
1243 goto error;
1244 }
1245 ret = len;
1246 goto error;
1247 } else {
1248 DBG("incorrect event notifier data message size: %zd", len);
1249 ret = -EINVAL;
1250 goto error;
1251 }
1252 }
1253 if (ops->cmd)
1254 ret = ops->cmd(lum->handle, lum->cmd,
1255 (unsigned long) &event_notifier,
1256 &args, sock_info);
1257 else
1258 ret = -ENOSYS;
1259 break;
1260 }
1261
1262 default:
1263 if (ops->cmd)
1264 ret = ops->cmd(lum->handle, lum->cmd,
1265 (unsigned long) &lum->u,
1266 &args, sock_info);
1267 else
1268 ret = -ENOSYS;
1269 break;
1270 }
1271
1272 lur.handle = lum->handle;
1273 lur.cmd = lum->cmd;
1274 lur.ret_val = ret;
1275 if (ret >= 0) {
1276 lur.ret_code = LTTNG_UST_OK;
1277 } else {
1278 /*
1279 * Use -LTTNG_UST_ERR as wildcard for UST internal
1280 * error that are not caused by the transport, except if
1281 * we already have a more precise error message to
1282 * report.
1283 */
1284 if (ret > -LTTNG_UST_ERR) {
1285 /* Translate code to UST error. */
1286 switch (ret) {
1287 case -EEXIST:
1288 lur.ret_code = -LTTNG_UST_ERR_EXIST;
1289 break;
1290 case -EINVAL:
1291 lur.ret_code = -LTTNG_UST_ERR_INVAL;
1292 break;
1293 case -ENOENT:
1294 lur.ret_code = -LTTNG_UST_ERR_NOENT;
1295 break;
1296 case -EPERM:
1297 lur.ret_code = -LTTNG_UST_ERR_PERM;
1298 break;
1299 case -ENOSYS:
1300 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
1301 break;
1302 default:
1303 lur.ret_code = -LTTNG_UST_ERR;
1304 break;
1305 }
1306 } else {
1307 lur.ret_code = ret;
1308 }
1309 }
1310 if (ret >= 0) {
1311 switch (lum->cmd) {
1312 case LTTNG_UST_TRACER_VERSION:
1313 lur.u.version = lum->u.version;
1314 break;
1315 case LTTNG_UST_TRACEPOINT_LIST_GET:
1316 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1317 break;
1318 }
1319 }
1320 DBG("Return value: %d", lur.ret_val);
1321
1322 ust_unlock();
1323
1324 /*
1325 * Performed delayed statedump operations outside of the UST
1326 * lock. We need to take the dynamic loader lock before we take
1327 * the UST lock internally within handle_pending_statedump().
1328 */
1329 handle_pending_statedump(sock_info);
1330
1331 if (ust_lock()) {
1332 ret = -LTTNG_UST_ERR_EXITING;
1333 goto error;
1334 }
1335
1336 ret = send_reply(sock, &lur);
1337 if (ret < 0) {
1338 DBG("error sending reply");
1339 goto error;
1340 }
1341
1342 /*
1343 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1344 * after the reply.
1345 */
1346 if (lur.ret_code == LTTNG_UST_OK) {
1347 switch (lum->cmd) {
1348 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
1349 len = ustcomm_send_unix_sock(sock,
1350 &args.field_list.entry,
1351 sizeof(args.field_list.entry));
1352 if (len < 0) {
1353 ret = len;
1354 goto error;
1355 }
1356 if (len != sizeof(args.field_list.entry)) {
1357 ret = -EINVAL;
1358 goto error;
1359 }
1360 }
1361 }
1362
1363 error:
1364 ust_unlock();
1365
1366 return ret;
1367 }
1368
1369 static
1370 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1371 {
1372 int ret;
1373
1374 if (sock_info->root_handle != -1) {
1375 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
1376 if (ret) {
1377 ERR("Error unref root handle");
1378 }
1379 sock_info->root_handle = -1;
1380 }
1381 sock_info->registration_done = 0;
1382 sock_info->initial_statedump_done = 0;
1383
1384 /*
1385 * wait_shm_mmap, socket and notify socket are used by listener
1386 * threads outside of the ust lock, so we cannot tear them down
1387 * ourselves, because we cannot join on these threads. Leave
1388 * responsibility of cleaning up these resources to the OS
1389 * process exit.
1390 */
1391 if (exiting)
1392 return;
1393
1394 if (sock_info->socket != -1) {
1395 ret = ustcomm_close_unix_sock(sock_info->socket);
1396 if (ret) {
1397 ERR("Error closing ust cmd socket");
1398 }
1399 sock_info->socket = -1;
1400 }
1401 if (sock_info->notify_socket != -1) {
1402 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1403 if (ret) {
1404 ERR("Error closing ust notify socket");
1405 }
1406 sock_info->notify_socket = -1;
1407 }
1408 if (sock_info->wait_shm_mmap) {
1409 long page_size;
1410
1411 page_size = LTTNG_UST_PAGE_SIZE;
1412 if (page_size <= 0) {
1413 if (!page_size) {
1414 errno = EINVAL;
1415 }
1416 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1417 } else {
1418 ret = munmap(sock_info->wait_shm_mmap, page_size);
1419 if (ret) {
1420 ERR("Error unmapping wait shm");
1421 }
1422 }
1423 sock_info->wait_shm_mmap = NULL;
1424 }
1425 }
1426
1427 /*
1428 * Using fork to set umask in the child process (not multi-thread safe).
1429 * We deal with the shm_open vs ftruncate race (happening when the
1430 * sessiond owns the shm and does not let everybody modify it, to ensure
1431 * safety against shm_unlink) by simply letting the mmap fail and
1432 * retrying after a few seconds.
1433 * For global shm, everybody has rw access to it until the sessiond
1434 * starts.
1435 */
1436 static
1437 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1438 {
1439 int wait_shm_fd, ret;
1440 pid_t pid;
1441
1442 /*
1443 * Try to open read-only.
1444 */
1445 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1446 if (wait_shm_fd >= 0) {
1447 int32_t tmp_read;
1448 ssize_t len;
1449 size_t bytes_read = 0;
1450
1451 /*
1452 * Try to read the fd. If unable to do so, try opening
1453 * it in write mode.
1454 */
1455 do {
1456 len = read(wait_shm_fd,
1457 &((char *) &tmp_read)[bytes_read],
1458 sizeof(tmp_read) - bytes_read);
1459 if (len > 0) {
1460 bytes_read += len;
1461 }
1462 } while ((len < 0 && errno == EINTR)
1463 || (len > 0 && bytes_read < sizeof(tmp_read)));
1464 if (bytes_read != sizeof(tmp_read)) {
1465 ret = close(wait_shm_fd);
1466 if (ret) {
1467 ERR("close wait_shm_fd");
1468 }
1469 goto open_write;
1470 }
1471 goto end;
1472 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1473 /*
1474 * Real-only open did not work, and it's not because the
1475 * entry was not present. It's a failure that prohibits
1476 * using shm.
1477 */
1478 ERR("Error opening shm %s", sock_info->wait_shm_path);
1479 goto end;
1480 }
1481
1482 open_write:
1483 /*
1484 * If the open failed because the file did not exist, or because
1485 * the file was not truncated yet, try creating it ourself.
1486 */
1487 URCU_TLS(lttng_ust_nest_count)++;
1488 pid = fork();
1489 URCU_TLS(lttng_ust_nest_count)--;
1490 if (pid > 0) {
1491 int status;
1492
1493 /*
1494 * Parent: wait for child to return, in which case the
1495 * shared memory map will have been created.
1496 */
1497 pid = wait(&status);
1498 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1499 wait_shm_fd = -1;
1500 goto end;
1501 }
1502 /*
1503 * Try to open read-only again after creation.
1504 */
1505 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1506 if (wait_shm_fd < 0) {
1507 /*
1508 * Real-only open did not work. It's a failure
1509 * that prohibits using shm.
1510 */
1511 ERR("Error opening shm %s", sock_info->wait_shm_path);
1512 goto end;
1513 }
1514 goto end;
1515 } else if (pid == 0) {
1516 int create_mode;
1517
1518 /* Child */
1519 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1520 if (sock_info->global)
1521 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1522 /*
1523 * We're alone in a child process, so we can modify the
1524 * process-wide umask.
1525 */
1526 umask(~create_mode);
1527 /*
1528 * Try creating shm (or get rw access).
1529 * We don't do an exclusive open, because we allow other
1530 * processes to create+ftruncate it concurrently.
1531 */
1532 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1533 O_RDWR | O_CREAT, create_mode);
1534 if (wait_shm_fd >= 0) {
1535 ret = ftruncate(wait_shm_fd, mmap_size);
1536 if (ret) {
1537 PERROR("ftruncate");
1538 _exit(EXIT_FAILURE);
1539 }
1540 _exit(EXIT_SUCCESS);
1541 }
1542 /*
1543 * For local shm, we need to have rw access to accept
1544 * opening it: this means the local sessiond will be
1545 * able to wake us up. For global shm, we open it even
1546 * if rw access is not granted, because the root.root
1547 * sessiond will be able to override all rights and wake
1548 * us up.
1549 */
1550 if (!sock_info->global && errno != EACCES) {
1551 ERR("Error opening shm %s", sock_info->wait_shm_path);
1552 _exit(EXIT_FAILURE);
1553 }
1554 /*
1555 * The shm exists, but we cannot open it RW. Report
1556 * success.
1557 */
1558 _exit(EXIT_SUCCESS);
1559 } else {
1560 return -1;
1561 }
1562 end:
1563 if (wait_shm_fd >= 0 && !sock_info->global) {
1564 struct stat statbuf;
1565
1566 /*
1567 * Ensure that our user is the owner of the shm file for
1568 * local shm. If we do not own the file, it means our
1569 * sessiond will not have access to wake us up (there is
1570 * probably a rogue process trying to fake our
1571 * sessiond). Fallback to polling method in this case.
1572 */
1573 ret = fstat(wait_shm_fd, &statbuf);
1574 if (ret) {
1575 PERROR("fstat");
1576 goto error_close;
1577 }
1578 if (statbuf.st_uid != getuid())
1579 goto error_close;
1580 }
1581 return wait_shm_fd;
1582
1583 error_close:
1584 ret = close(wait_shm_fd);
1585 if (ret) {
1586 PERROR("Error closing fd");
1587 }
1588 return -1;
1589 }
1590
1591 static
1592 char *get_map_shm(struct sock_info *sock_info)
1593 {
1594 long page_size;
1595 int wait_shm_fd, ret;
1596 char *wait_shm_mmap;
1597
1598 page_size = sysconf(_SC_PAGE_SIZE);
1599 if (page_size <= 0) {
1600 if (!page_size) {
1601 errno = EINVAL;
1602 }
1603 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1604 goto error;
1605 }
1606
1607 lttng_ust_lock_fd_tracker();
1608 wait_shm_fd = get_wait_shm(sock_info, page_size);
1609 if (wait_shm_fd < 0) {
1610 lttng_ust_unlock_fd_tracker();
1611 goto error;
1612 }
1613
1614 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1615 if (ret < 0) {
1616 ret = close(wait_shm_fd);
1617 if (!ret) {
1618 PERROR("Error closing fd");
1619 }
1620 lttng_ust_unlock_fd_tracker();
1621 goto error;
1622 }
1623
1624 wait_shm_fd = ret;
1625 lttng_ust_unlock_fd_tracker();
1626
1627 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1628 MAP_SHARED, wait_shm_fd, 0);
1629
1630 /* close shm fd immediately after taking the mmap reference */
1631 lttng_ust_lock_fd_tracker();
1632 ret = close(wait_shm_fd);
1633 if (!ret) {
1634 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1635 } else {
1636 PERROR("Error closing fd");
1637 }
1638 lttng_ust_unlock_fd_tracker();
1639
1640 if (wait_shm_mmap == MAP_FAILED) {
1641 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1642 goto error;
1643 }
1644 return wait_shm_mmap;
1645
1646 error:
1647 return NULL;
1648 }
1649
1650 static
1651 void wait_for_sessiond(struct sock_info *sock_info)
1652 {
1653 /* Use ust_lock to check if we should quit. */
1654 if (ust_lock()) {
1655 goto quit;
1656 }
1657 if (wait_poll_fallback) {
1658 goto error;
1659 }
1660 ust_unlock();
1661
1662 assert(sock_info->wait_shm_mmap);
1663
1664 DBG("Waiting for %s apps sessiond", sock_info->name);
1665 /* Wait for futex wakeup */
1666 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1667 goto end_wait;
1668
1669 while (lttng_ust_futex_async((int32_t *) sock_info->wait_shm_mmap,
1670 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1671 switch (errno) {
1672 case EWOULDBLOCK:
1673 /* Value already changed. */
1674 goto end_wait;
1675 case EINTR:
1676 /* Retry if interrupted by signal. */
1677 break; /* Get out of switch. */
1678 case EFAULT:
1679 wait_poll_fallback = 1;
1680 DBG(
1681 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1682 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1683 "Please upgrade your kernel "
1684 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1685 "mainline). LTTng-UST will use polling mode fallback.");
1686 if (ust_debug())
1687 PERROR("futex");
1688 goto end_wait;
1689 }
1690 }
1691 end_wait:
1692 return;
1693
1694 quit:
1695 ust_unlock();
1696 return;
1697
1698 error:
1699 ust_unlock();
1700 return;
1701 }
1702
1703 /*
1704 * This thread does not allocate any resource, except within
1705 * handle_message, within mutex protection. This mutex protects against
1706 * fork and exit.
1707 * The other moment it allocates resources is at socket connection, which
1708 * is also protected by the mutex.
1709 */
1710 static
1711 void *ust_listener_thread(void *arg)
1712 {
1713 struct sock_info *sock_info = arg;
1714 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1715 long timeout;
1716
1717 lttng_ust_fixup_tls();
1718 /*
1719 * If available, add '-ust' to the end of this thread's
1720 * process name
1721 */
1722 ret = lttng_ust_setustprocname();
1723 if (ret) {
1724 ERR("Unable to set UST process name");
1725 }
1726
1727 /* Restart trying to connect to the session daemon */
1728 restart:
1729 if (prev_connect_failed) {
1730 /* Wait for sessiond availability with pipe */
1731 wait_for_sessiond(sock_info);
1732 if (has_waited) {
1733 has_waited = 0;
1734 /*
1735 * Sleep for 5 seconds before retrying after a
1736 * sequence of failure / wait / failure. This
1737 * deals with a killed or broken session daemon.
1738 */
1739 sleep(5);
1740 } else {
1741 has_waited = 1;
1742 }
1743 prev_connect_failed = 0;
1744 }
1745
1746 if (ust_lock()) {
1747 goto quit;
1748 }
1749
1750 if (sock_info->socket != -1) {
1751 /* FD tracker is updated by ustcomm_close_unix_sock() */
1752 ret = ustcomm_close_unix_sock(sock_info->socket);
1753 if (ret) {
1754 ERR("Error closing %s ust cmd socket",
1755 sock_info->name);
1756 }
1757 sock_info->socket = -1;
1758 }
1759 if (sock_info->notify_socket != -1) {
1760 /* FD tracker is updated by ustcomm_close_unix_sock() */
1761 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1762 if (ret) {
1763 ERR("Error closing %s ust notify socket",
1764 sock_info->name);
1765 }
1766 sock_info->notify_socket = -1;
1767 }
1768
1769
1770 /*
1771 * Register. We need to perform both connect and sending
1772 * registration message before doing the next connect otherwise
1773 * we may reach unix socket connect queue max limits and block
1774 * on the 2nd connect while the session daemon is awaiting the
1775 * first connect registration message.
1776 */
1777 /* Connect cmd socket */
1778 lttng_ust_lock_fd_tracker();
1779 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1780 get_connect_sock_timeout());
1781 if (ret < 0) {
1782 lttng_ust_unlock_fd_tracker();
1783 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1784 prev_connect_failed = 1;
1785
1786 /*
1787 * If we cannot find the sessiond daemon, don't delay
1788 * constructor execution.
1789 */
1790 ret = handle_register_failed(sock_info);
1791 assert(!ret);
1792 ust_unlock();
1793 goto restart;
1794 }
1795 fd = ret;
1796 ret = lttng_ust_add_fd_to_tracker(fd);
1797 if (ret < 0) {
1798 ret = close(fd);
1799 if (ret) {
1800 PERROR("close on sock_info->socket");
1801 }
1802 ret = -1;
1803 lttng_ust_unlock_fd_tracker();
1804 ust_unlock();
1805 goto quit;
1806 }
1807
1808 sock_info->socket = ret;
1809 lttng_ust_unlock_fd_tracker();
1810
1811 ust_unlock();
1812 /*
1813 * Unlock/relock ust lock because connect is blocking (with
1814 * timeout). Don't delay constructors on the ust lock for too
1815 * long.
1816 */
1817 if (ust_lock()) {
1818 goto quit;
1819 }
1820
1821 /*
1822 * Create only one root handle per listener thread for the whole
1823 * process lifetime, so we ensure we get ID which is statically
1824 * assigned to the root handle.
1825 */
1826 if (sock_info->root_handle == -1) {
1827 ret = lttng_abi_create_root_handle();
1828 if (ret < 0) {
1829 ERR("Error creating root handle");
1830 goto quit;
1831 }
1832 sock_info->root_handle = ret;
1833 }
1834
1835 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1836 if (ret < 0) {
1837 ERR("Error registering to %s ust cmd socket",
1838 sock_info->name);
1839 prev_connect_failed = 1;
1840 /*
1841 * If we cannot register to the sessiond daemon, don't
1842 * delay constructor execution.
1843 */
1844 ret = handle_register_failed(sock_info);
1845 assert(!ret);
1846 ust_unlock();
1847 goto restart;
1848 }
1849
1850 ust_unlock();
1851 /*
1852 * Unlock/relock ust lock because connect is blocking (with
1853 * timeout). Don't delay constructors on the ust lock for too
1854 * long.
1855 */
1856 if (ust_lock()) {
1857 goto quit;
1858 }
1859
1860 /* Connect notify socket */
1861 lttng_ust_lock_fd_tracker();
1862 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1863 get_connect_sock_timeout());
1864 if (ret < 0) {
1865 lttng_ust_unlock_fd_tracker();
1866 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1867 prev_connect_failed = 1;
1868
1869 /*
1870 * If we cannot find the sessiond daemon, don't delay
1871 * constructor execution.
1872 */
1873 ret = handle_register_failed(sock_info);
1874 assert(!ret);
1875 ust_unlock();
1876 goto restart;
1877 }
1878
1879 fd = ret;
1880 ret = lttng_ust_add_fd_to_tracker(fd);
1881 if (ret < 0) {
1882 ret = close(fd);
1883 if (ret) {
1884 PERROR("close on sock_info->notify_socket");
1885 }
1886 ret = -1;
1887 lttng_ust_unlock_fd_tracker();
1888 ust_unlock();
1889 goto quit;
1890 }
1891
1892 sock_info->notify_socket = ret;
1893 lttng_ust_unlock_fd_tracker();
1894
1895 ust_unlock();
1896 /*
1897 * Unlock/relock ust lock because connect is blocking (with
1898 * timeout). Don't delay constructors on the ust lock for too
1899 * long.
1900 */
1901 if (ust_lock()) {
1902 goto quit;
1903 }
1904
1905 timeout = get_notify_sock_timeout();
1906 if (timeout >= 0) {
1907 /*
1908 * Give at least 10ms to sessiond to reply to
1909 * notifications.
1910 */
1911 if (timeout < 10)
1912 timeout = 10;
1913 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1914 timeout);
1915 if (ret < 0) {
1916 WARN("Error setting socket receive timeout");
1917 }
1918 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1919 timeout);
1920 if (ret < 0) {
1921 WARN("Error setting socket send timeout");
1922 }
1923 } else if (timeout < -1) {
1924 WARN("Unsupported timeout value %ld", timeout);
1925 }
1926
1927 ret = register_to_sessiond(sock_info->notify_socket,
1928 USTCTL_SOCKET_NOTIFY);
1929 if (ret < 0) {
1930 ERR("Error registering to %s ust notify socket",
1931 sock_info->name);
1932 prev_connect_failed = 1;
1933 /*
1934 * If we cannot register to the sessiond daemon, don't
1935 * delay constructor execution.
1936 */
1937 ret = handle_register_failed(sock_info);
1938 assert(!ret);
1939 ust_unlock();
1940 goto restart;
1941 }
1942 sock = sock_info->socket;
1943
1944 ust_unlock();
1945
1946 for (;;) {
1947 ssize_t len;
1948 struct ustcomm_ust_msg lum;
1949
1950 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1951 switch (len) {
1952 case 0: /* orderly shutdown */
1953 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1954 if (ust_lock()) {
1955 goto quit;
1956 }
1957 /*
1958 * Either sessiond has shutdown or refused us by closing the socket.
1959 * In either case, we don't want to delay construction execution,
1960 * and we need to wait before retry.
1961 */
1962 prev_connect_failed = 1;
1963 /*
1964 * If we cannot register to the sessiond daemon, don't
1965 * delay constructor execution.
1966 */
1967 ret = handle_register_failed(sock_info);
1968 assert(!ret);
1969 ust_unlock();
1970 goto end;
1971 case sizeof(lum):
1972 print_cmd(lum.cmd, lum.handle);
1973 ret = handle_message(sock_info, sock, &lum);
1974 if (ret) {
1975 ERR("Error handling message for %s socket",
1976 sock_info->name);
1977 /*
1978 * Close socket if protocol error is
1979 * detected.
1980 */
1981 goto end;
1982 }
1983 continue;
1984 default:
1985 if (len < 0) {
1986 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1987 } else {
1988 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1989 }
1990 if (len == -ECONNRESET) {
1991 DBG("%s remote end closed connection", sock_info->name);
1992 goto end;
1993 }
1994 goto end;
1995 }
1996
1997 }
1998 end:
1999 if (ust_lock()) {
2000 goto quit;
2001 }
2002 /* Cleanup socket handles before trying to reconnect */
2003 lttng_ust_objd_table_owner_cleanup(sock_info);
2004 ust_unlock();
2005 goto restart; /* try to reconnect */
2006
2007 quit:
2008 ust_unlock();
2009
2010 pthread_mutex_lock(&ust_exit_mutex);
2011 sock_info->thread_active = 0;
2012 pthread_mutex_unlock(&ust_exit_mutex);
2013 return NULL;
2014 }
2015
2016 /*
2017 * Weak symbol to call when the ust malloc wrapper is not loaded.
2018 */
2019 __attribute__((weak))
2020 void lttng_ust_malloc_wrapper_init(void)
2021 {
2022 }
2023
2024 /*
2025 * sessiond monitoring thread: monitor presence of global and per-user
2026 * sessiond by polling the application common named pipe.
2027 */
2028 void __attribute__((constructor)) lttng_ust_init(void)
2029 {
2030 struct timespec constructor_timeout;
2031 sigset_t sig_all_blocked, orig_parent_mask;
2032 pthread_attr_t thread_attr;
2033 int timeout_mode;
2034 int ret;
2035 void *handle;
2036
2037 if (uatomic_xchg(&initialized, 1) == 1)
2038 return;
2039
2040 /*
2041 * Fixup interdependency between TLS fixup mutex (which happens
2042 * to be the dynamic linker mutex) and ust_lock, taken within
2043 * the ust lock.
2044 */
2045 lttng_ust_fixup_tls();
2046
2047 lttng_ust_loaded = 1;
2048
2049 /*
2050 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2051 * the unloading of code used by the ust_listener_threads as we can not
2052 * reliably know when they exited. To do that, manually load
2053 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2054 * this library so it never becomes zero, thus never gets unloaded from the
2055 * address space of the process. Since we are already running in the
2056 * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
2057 * simply increment the refcount and no additionnal work is needed by the
2058 * dynamic loader as the shared library is already loaded in the address
2059 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2060 * unloading of the UST library if its refcount becomes zero (which should
2061 * never happen). Do the return value check but discard the handle at the
2062 * end of the function as it's not needed.
2063 */
2064 handle = dlopen(LTTNG_UST_LIB_SO_NAME, RTLD_LAZY | RTLD_NODELETE);
2065 if (!handle) {
2066 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME);
2067 }
2068
2069 /*
2070 * We want precise control over the order in which we construct
2071 * our sub-libraries vs starting to receive commands from
2072 * sessiond (otherwise leading to errors when trying to create
2073 * sessiond before the init functions are completed).
2074 */
2075 init_usterr();
2076 lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */
2077 init_tracepoint();
2078 lttng_ust_init_fd_tracker();
2079 lttng_ust_clock_init();
2080 lttng_ust_getcpu_init();
2081 lttng_ust_statedump_init();
2082 lttng_ring_buffer_metadata_client_init();
2083 lttng_ring_buffer_client_overwrite_init();
2084 lttng_ring_buffer_client_overwrite_rt_init();
2085 lttng_ring_buffer_client_discard_init();
2086 lttng_ring_buffer_client_discard_rt_init();
2087 lttng_counter_client_percpu_32_modular_init();
2088 lttng_counter_client_percpu_64_modular_init();
2089 lttng_perf_counter_init();
2090 /*
2091 * Invoke ust malloc wrapper init before starting other threads.
2092 */
2093 lttng_ust_malloc_wrapper_init();
2094
2095 timeout_mode = get_constructor_timeout(&constructor_timeout);
2096
2097 get_allow_blocking();
2098
2099 ret = sem_init(&constructor_wait, 0, 0);
2100 if (ret) {
2101 PERROR("sem_init");
2102 }
2103
2104 ret = setup_global_apps();
2105 if (ret) {
2106 assert(global_apps.allowed == 0);
2107 DBG("global apps setup returned %d", ret);
2108 }
2109
2110 ret = setup_local_apps();
2111 if (ret) {
2112 assert(local_apps.allowed == 0);
2113 DBG("local apps setup returned %d", ret);
2114 }
2115
2116 /* A new thread created by pthread_create inherits the signal mask
2117 * from the parent. To avoid any signal being received by the
2118 * listener thread, we block all signals temporarily in the parent,
2119 * while we create the listener thread.
2120 */
2121 sigfillset(&sig_all_blocked);
2122 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
2123 if (ret) {
2124 ERR("pthread_sigmask: %s", strerror(ret));
2125 }
2126
2127 ret = pthread_attr_init(&thread_attr);
2128 if (ret) {
2129 ERR("pthread_attr_init: %s", strerror(ret));
2130 }
2131 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
2132 if (ret) {
2133 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
2134 }
2135
2136 if (global_apps.allowed) {
2137 pthread_mutex_lock(&ust_exit_mutex);
2138 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
2139 ust_listener_thread, &global_apps);
2140 if (ret) {
2141 ERR("pthread_create global: %s", strerror(ret));
2142 }
2143 global_apps.thread_active = 1;
2144 pthread_mutex_unlock(&ust_exit_mutex);
2145 } else {
2146 handle_register_done(&global_apps);
2147 }
2148
2149 if (local_apps.allowed) {
2150 pthread_mutex_lock(&ust_exit_mutex);
2151 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
2152 ust_listener_thread, &local_apps);
2153 if (ret) {
2154 ERR("pthread_create local: %s", strerror(ret));
2155 }
2156 local_apps.thread_active = 1;
2157 pthread_mutex_unlock(&ust_exit_mutex);
2158 } else {
2159 handle_register_done(&local_apps);
2160 }
2161 ret = pthread_attr_destroy(&thread_attr);
2162 if (ret) {
2163 ERR("pthread_attr_destroy: %s", strerror(ret));
2164 }
2165
2166 /* Restore original signal mask in parent */
2167 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
2168 if (ret) {
2169 ERR("pthread_sigmask: %s", strerror(ret));
2170 }
2171
2172 switch (timeout_mode) {
2173 case 1: /* timeout wait */
2174 do {
2175 ret = sem_timedwait(&constructor_wait,
2176 &constructor_timeout);
2177 } while (ret < 0 && errno == EINTR);
2178 if (ret < 0) {
2179 switch (errno) {
2180 case ETIMEDOUT:
2181 ERR("Timed out waiting for lttng-sessiond");
2182 break;
2183 case EINVAL:
2184 PERROR("sem_timedwait");
2185 break;
2186 default:
2187 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2188 strerror(errno));
2189 }
2190 }
2191 break;
2192 case -1:/* wait forever */
2193 do {
2194 ret = sem_wait(&constructor_wait);
2195 } while (ret < 0 && errno == EINTR);
2196 if (ret < 0) {
2197 switch (errno) {
2198 case EINVAL:
2199 PERROR("sem_wait");
2200 break;
2201 default:
2202 ERR("Unexpected error \"%s\" returned by sem_wait",
2203 strerror(errno));
2204 }
2205 }
2206 break;
2207 case 0: /* no timeout */
2208 break;
2209 }
2210 }
2211
2212 static
2213 void lttng_ust_cleanup(int exiting)
2214 {
2215 cleanup_sock_info(&global_apps, exiting);
2216 cleanup_sock_info(&local_apps, exiting);
2217 local_apps.allowed = 0;
2218 global_apps.allowed = 0;
2219 /*
2220 * The teardown in this function all affect data structures
2221 * accessed under the UST lock by the listener thread. This
2222 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2223 * that none of these threads are accessing this data at this
2224 * point.
2225 */
2226 lttng_ust_abi_exit();
2227 lttng_ust_events_exit();
2228 lttng_perf_counter_exit();
2229 lttng_ring_buffer_client_discard_rt_exit();
2230 lttng_ring_buffer_client_discard_exit();
2231 lttng_ring_buffer_client_overwrite_rt_exit();
2232 lttng_ring_buffer_client_overwrite_exit();
2233 lttng_ring_buffer_metadata_client_exit();
2234 lttng_counter_client_percpu_32_modular_exit();
2235 lttng_counter_client_percpu_64_modular_exit();
2236 lttng_ust_statedump_destroy();
2237 exit_tracepoint();
2238 if (!exiting) {
2239 /* Reinitialize values for fork */
2240 sem_count = sem_count_initial_value;
2241 lttng_ust_comm_should_quit = 0;
2242 initialized = 0;
2243 }
2244 }
2245
2246 void __attribute__((destructor)) lttng_ust_exit(void)
2247 {
2248 int ret;
2249
2250 /*
2251 * Using pthread_cancel here because:
2252 * A) we don't want to hang application teardown.
2253 * B) the thread is not allocating any resource.
2254 */
2255
2256 /*
2257 * Require the communication thread to quit. Synchronize with
2258 * mutexes to ensure it is not in a mutex critical section when
2259 * pthread_cancel is later called.
2260 */
2261 ust_lock_nocheck();
2262 lttng_ust_comm_should_quit = 1;
2263 ust_unlock();
2264
2265 pthread_mutex_lock(&ust_exit_mutex);
2266 /* cancel threads */
2267 if (global_apps.thread_active) {
2268 ret = pthread_cancel(global_apps.ust_listener);
2269 if (ret) {
2270 ERR("Error cancelling global ust listener thread: %s",
2271 strerror(ret));
2272 } else {
2273 global_apps.thread_active = 0;
2274 }
2275 }
2276 if (local_apps.thread_active) {
2277 ret = pthread_cancel(local_apps.ust_listener);
2278 if (ret) {
2279 ERR("Error cancelling local ust listener thread: %s",
2280 strerror(ret));
2281 } else {
2282 local_apps.thread_active = 0;
2283 }
2284 }
2285 pthread_mutex_unlock(&ust_exit_mutex);
2286
2287 /*
2288 * Do NOT join threads: use of sys_futex makes it impossible to
2289 * join the threads without using async-cancel, but async-cancel
2290 * is delivered by a signal, which could hit the target thread
2291 * anywhere in its code path, including while the ust_lock() is
2292 * held, causing a deadlock for the other thread. Let the OS
2293 * cleanup the threads if there are stalled in a syscall.
2294 */
2295 lttng_ust_cleanup(1);
2296 }
2297
2298 static
2299 void ust_context_ns_reset(void)
2300 {
2301 lttng_context_pid_ns_reset();
2302 lttng_context_cgroup_ns_reset();
2303 lttng_context_ipc_ns_reset();
2304 lttng_context_mnt_ns_reset();
2305 lttng_context_net_ns_reset();
2306 lttng_context_user_ns_reset();
2307 lttng_context_time_ns_reset();
2308 lttng_context_uts_ns_reset();
2309 }
2310
2311 static
2312 void ust_context_vuids_reset(void)
2313 {
2314 lttng_context_vuid_reset();
2315 lttng_context_veuid_reset();
2316 lttng_context_vsuid_reset();
2317 }
2318
2319 static
2320 void ust_context_vgids_reset(void)
2321 {
2322 lttng_context_vgid_reset();
2323 lttng_context_vegid_reset();
2324 lttng_context_vsgid_reset();
2325 }
2326
2327 /*
2328 * We exclude the worker threads across fork and clone (except
2329 * CLONE_VM), because these system calls only keep the forking thread
2330 * running in the child. Therefore, we don't want to call fork or clone
2331 * in the middle of an tracepoint or ust tracing state modification.
2332 * Holding this mutex protects these structures across fork and clone.
2333 */
2334 void ust_before_fork(sigset_t *save_sigset)
2335 {
2336 /*
2337 * Disable signals. This is to avoid that the child intervenes
2338 * before it is properly setup for tracing. It is safer to
2339 * disable all signals, because then we know we are not breaking
2340 * anything by restoring the original mask.
2341 */
2342 sigset_t all_sigs;
2343 int ret;
2344
2345 /* Fixup lttng-ust TLS. */
2346 lttng_ust_fixup_tls();
2347
2348 if (URCU_TLS(lttng_ust_nest_count))
2349 return;
2350 /* Disable signals */
2351 sigfillset(&all_sigs);
2352 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2353 if (ret == -1) {
2354 PERROR("sigprocmask");
2355 }
2356
2357 pthread_mutex_lock(&ust_fork_mutex);
2358
2359 ust_lock_nocheck();
2360 lttng_ust_urcu_before_fork();
2361 if (lttng_ust_liburcu_bp_before_fork)
2362 lttng_ust_liburcu_bp_before_fork();
2363 lttng_ust_lock_fd_tracker();
2364 lttng_perf_lock();
2365 }
2366
2367 static void ust_after_fork_common(sigset_t *restore_sigset)
2368 {
2369 int ret;
2370
2371 DBG("process %d", getpid());
2372 lttng_perf_unlock();
2373 lttng_ust_unlock_fd_tracker();
2374 ust_unlock();
2375
2376 pthread_mutex_unlock(&ust_fork_mutex);
2377
2378 /* Restore signals */
2379 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2380 if (ret == -1) {
2381 PERROR("sigprocmask");
2382 }
2383 }
2384
2385 void ust_after_fork_parent(sigset_t *restore_sigset)
2386 {
2387 if (URCU_TLS(lttng_ust_nest_count))
2388 return;
2389 DBG("process %d", getpid());
2390 lttng_ust_urcu_after_fork_parent();
2391 if (lttng_ust_liburcu_bp_after_fork_parent)
2392 lttng_ust_liburcu_bp_after_fork_parent();
2393 /* Release mutexes and reenable signals */
2394 ust_after_fork_common(restore_sigset);
2395 }
2396
2397 /*
2398 * After fork, in the child, we need to cleanup all the leftover state,
2399 * except the worker thread which already magically disappeared thanks
2400 * to the weird Linux fork semantics. After tyding up, we call
2401 * lttng_ust_init() again to start over as a new PID.
2402 *
2403 * This is meant for forks() that have tracing in the child between the
2404 * fork and following exec call (if there is any).
2405 */
2406 void ust_after_fork_child(sigset_t *restore_sigset)
2407 {
2408 if (URCU_TLS(lttng_ust_nest_count))
2409 return;
2410 lttng_context_vpid_reset();
2411 lttng_context_vtid_reset();
2412 lttng_context_procname_reset();
2413 ust_context_ns_reset();
2414 ust_context_vuids_reset();
2415 ust_context_vgids_reset();
2416 DBG("process %d", getpid());
2417 /* Release urcu mutexes */
2418 lttng_ust_urcu_after_fork_child();
2419 if (lttng_ust_liburcu_bp_after_fork_child)
2420 lttng_ust_liburcu_bp_after_fork_child();
2421 lttng_ust_cleanup(0);
2422 /* Release mutexes and reenable signals */
2423 ust_after_fork_common(restore_sigset);
2424 lttng_ust_init();
2425 }
2426
2427 void ust_after_setns(void)
2428 {
2429 ust_context_ns_reset();
2430 ust_context_vuids_reset();
2431 ust_context_vgids_reset();
2432 }
2433
2434 void ust_after_unshare(void)
2435 {
2436 ust_context_ns_reset();
2437 ust_context_vuids_reset();
2438 ust_context_vgids_reset();
2439 }
2440
2441 void ust_after_setuid(void)
2442 {
2443 ust_context_vuids_reset();
2444 }
2445
2446 void ust_after_seteuid(void)
2447 {
2448 ust_context_vuids_reset();
2449 }
2450
2451 void ust_after_setreuid(void)
2452 {
2453 ust_context_vuids_reset();
2454 }
2455
2456 void ust_after_setresuid(void)
2457 {
2458 ust_context_vuids_reset();
2459 }
2460
2461 void ust_after_setgid(void)
2462 {
2463 ust_context_vgids_reset();
2464 }
2465
2466 void ust_after_setegid(void)
2467 {
2468 ust_context_vgids_reset();
2469 }
2470
2471 void ust_after_setregid(void)
2472 {
2473 ust_context_vgids_reset();
2474 }
2475
2476 void ust_after_setresgid(void)
2477 {
2478 ust_context_vgids_reset();
2479 }
2480
2481 void lttng_ust_sockinfo_session_enabled(void *owner)
2482 {
2483 struct sock_info *sock_info = owner;
2484 sock_info->statedump_pending = 1;
2485 }
LTTng 2.0 Userspace Tracer
This page took 0.120768 seconds and 3 git commands to generate.