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


651d2aaa45cec437cabb7cdb4147d66330800065
[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 #define _GNU_SOURCE
24 #include <sys/types.h>
25 #include <sys/socket.h>
26 #include <sys/mman.h>
27 #include <sys/stat.h>
28 #include <sys/types.h>
29 #include <sys/wait.h>
30 #include <fcntl.h>
31 #include <unistd.h>
32 #include <errno.h>
33 #include <pthread.h>
34 #include <semaphore.h>
35 #include <time.h>
36 #include <assert.h>
37 #include <signal.h>
38 #include <limits.h>
39 #include <urcu/uatomic.h>
40 #include <urcu/futex.h>
41 #include <urcu/compiler.h>
42
43 #include <lttng/ust-events.h>
44 #include <lttng/ust-abi.h>
45 #include <lttng/ust.h>
46 #include <lttng/ust-error.h>
47 #include <lttng/ust-ctl.h>
48 #include <urcu/tls-compat.h>
49 #include <ust-comm.h>
50 #include <ust-fd.h>
51 #include <usterr-signal-safe.h>
52 #include <helper.h>
53 #include "tracepoint-internal.h"
54 #include "lttng-tracer-core.h"
55 #include "compat.h"
56 #include "../libringbuffer/rb-init.h"
57 #include "lttng-ust-statedump.h"
58 #include "clock.h"
59 #include "../libringbuffer/getcpu.h"
60 #include "getenv.h"
61
62 /*
63 * Has lttng ust comm constructor been called ?
64 */
65 static int initialized;
66
67 /*
68 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
69 * Held when handling a command, also held by fork() to deal with
70 * removal of threads, and by exit path.
71 *
72 * The UST lock is the centralized mutex across UST tracing control and
73 * probe registration.
74 *
75 * ust_exit_mutex must never nest in ust_mutex.
76 *
77 * ust_fork_mutex must never nest in ust_mutex.
78 *
79 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
80 * counter lazy initialization called by events within the statedump,
81 * which traces while the ust_mutex is held.
82 *
83 * ust_lock nests within the dynamic loader lock (within glibc) because
84 * it is taken within the library constructor.
85 */
86 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
87
88 /* Allow nesting the ust_mutex within the same thread. */
89 static DEFINE_URCU_TLS(int, ust_mutex_nest);
90
91 /*
92 * ust_exit_mutex protects thread_active variable wrt thread exit. It
93 * cannot be done by ust_mutex because pthread_cancel(), which takes an
94 * internal libc lock, cannot nest within ust_mutex.
95 *
96 * It never nests within a ust_mutex.
97 */
98 static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
99
100 /*
101 * ust_fork_mutex protects base address statedump tracing against forks. It
102 * prevents the dynamic loader lock to be taken (by base address statedump
103 * tracing) while a fork is happening, thus preventing deadlock issues with
104 * the dynamic loader lock.
105 */
106 static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
107
108 /* Should the ust comm thread quit ? */
109 static int lttng_ust_comm_should_quit;
110
111 /*
112 * This variable can be tested by applications to check whether
113 * lttng-ust is loaded. They simply have to define their own
114 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
115 * library constructor.
116 */
117 int lttng_ust_loaded __attribute__((weak));
118
119 /*
120 * Return 0 on success, -1 if should quit.
121 * The lock is taken in both cases.
122 * Signal-safe.
123 */
124 int ust_lock(void)
125 {
126 sigset_t sig_all_blocked, orig_mask;
127 int ret, oldstate;
128
129 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
130 if (ret) {
131 ERR("pthread_setcancelstate: %s", strerror(ret));
132 }
133 if (oldstate != PTHREAD_CANCEL_ENABLE) {
134 ERR("pthread_setcancelstate: unexpected oldstate");
135 }
136 sigfillset(&sig_all_blocked);
137 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
138 if (ret) {
139 ERR("pthread_sigmask: %s", strerror(ret));
140 }
141 if (!URCU_TLS(ust_mutex_nest)++)
142 pthread_mutex_lock(&ust_mutex);
143 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
144 if (ret) {
145 ERR("pthread_sigmask: %s", strerror(ret));
146 }
147 if (lttng_ust_comm_should_quit) {
148 return -1;
149 } else {
150 return 0;
151 }
152 }
153
154 /*
155 * ust_lock_nocheck() can be used in constructors/destructors, because
156 * they are already nested within the dynamic loader lock, and therefore
157 * have exclusive access against execution of liblttng-ust destructor.
158 * Signal-safe.
159 */
160 void ust_lock_nocheck(void)
161 {
162 sigset_t sig_all_blocked, orig_mask;
163 int ret, oldstate;
164
165 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
166 if (ret) {
167 ERR("pthread_setcancelstate: %s", strerror(ret));
168 }
169 if (oldstate != PTHREAD_CANCEL_ENABLE) {
170 ERR("pthread_setcancelstate: unexpected oldstate");
171 }
172 sigfillset(&sig_all_blocked);
173 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
174 if (ret) {
175 ERR("pthread_sigmask: %s", strerror(ret));
176 }
177 if (!URCU_TLS(ust_mutex_nest)++)
178 pthread_mutex_lock(&ust_mutex);
179 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
180 if (ret) {
181 ERR("pthread_sigmask: %s", strerror(ret));
182 }
183 }
184
185 /*
186 * Signal-safe.
187 */
188 void ust_unlock(void)
189 {
190 sigset_t sig_all_blocked, orig_mask;
191 int ret, oldstate;
192
193 sigfillset(&sig_all_blocked);
194 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
195 if (ret) {
196 ERR("pthread_sigmask: %s", strerror(ret));
197 }
198 if (!--URCU_TLS(ust_mutex_nest))
199 pthread_mutex_unlock(&ust_mutex);
200 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
201 if (ret) {
202 ERR("pthread_sigmask: %s", strerror(ret));
203 }
204 ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
205 if (ret) {
206 ERR("pthread_setcancelstate: %s", strerror(ret));
207 }
208 if (oldstate != PTHREAD_CANCEL_DISABLE) {
209 ERR("pthread_setcancelstate: unexpected oldstate");
210 }
211 }
212
213 /*
214 * Wait for either of these before continuing to the main
215 * program:
216 * - the register_done message from sessiond daemon
217 * (will let the sessiond daemon enable sessions before main
218 * starts.)
219 * - sessiond daemon is not reachable.
220 * - timeout (ensuring applications are resilient to session
221 * daemon problems).
222 */
223 static sem_t constructor_wait;
224 /*
225 * Doing this for both the global and local sessiond.
226 */
227 static int sem_count = { 2 };
228
229 /*
230 * Counting nesting within lttng-ust. Used to ensure that calling fork()
231 * from liblttng-ust does not execute the pre/post fork handlers.
232 */
233 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
234
235 /*
236 * Info about socket and associated listener thread.
237 */
238 struct sock_info {
239 const char *name;
240 pthread_t ust_listener; /* listener thread */
241 int root_handle;
242 int constructor_sem_posted;
243 int allowed;
244 int global;
245 int thread_active;
246
247 char sock_path[PATH_MAX];
248 int socket;
249 int notify_socket;
250
251 char wait_shm_path[PATH_MAX];
252 char *wait_shm_mmap;
253 /* Keep track of lazy state dump not performed yet. */
254 int statedump_pending;
255 };
256
257 /* Socket from app (connect) to session daemon (listen) for communication */
258 struct sock_info global_apps = {
259 .name = "global",
260 .global = 1,
261
262 .root_handle = -1,
263 .allowed = 1,
264 .thread_active = 0,
265
266 .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
267 .socket = -1,
268 .notify_socket = -1,
269
270 .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
271
272 .statedump_pending = 0,
273 };
274
275 /* TODO: allow global_apps_sock_path override */
276
277 struct sock_info local_apps = {
278 .name = "local",
279 .global = 0,
280 .root_handle = -1,
281 .allowed = 0, /* Check setuid bit first */
282 .thread_active = 0,
283
284 .socket = -1,
285 .notify_socket = -1,
286
287 .statedump_pending = 0,
288 };
289
290 static int wait_poll_fallback;
291
292 static const char *cmd_name_mapping[] = {
293 [ LTTNG_UST_RELEASE ] = "Release",
294 [ LTTNG_UST_SESSION ] = "Create Session",
295 [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
296
297 [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
298 [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
299 [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
300 [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
301
302 /* Session FD commands */
303 [ LTTNG_UST_CHANNEL ] = "Create Channel",
304 [ LTTNG_UST_SESSION_START ] = "Start Session",
305 [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
306
307 /* Channel FD commands */
308 [ LTTNG_UST_STREAM ] = "Create Stream",
309 [ LTTNG_UST_EVENT ] = "Create Event",
310
311 /* Event and Channel FD commands */
312 [ LTTNG_UST_CONTEXT ] = "Create Context",
313 [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
314
315 /* Event, Channel and Session commands */
316 [ LTTNG_UST_ENABLE ] = "Enable",
317 [ LTTNG_UST_DISABLE ] = "Disable",
318
319 /* Tracepoint list commands */
320 [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
321 [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
322
323 /* Event FD commands */
324 [ LTTNG_UST_FILTER ] = "Create Filter",
325 [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
326 };
327
328 static const char *str_timeout;
329 static int got_timeout_env;
330
331 extern void lttng_ring_buffer_client_overwrite_init(void);
332 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
333 extern void lttng_ring_buffer_client_discard_init(void);
334 extern void lttng_ring_buffer_client_discard_rt_init(void);
335 extern void lttng_ring_buffer_metadata_client_init(void);
336 extern void lttng_ring_buffer_client_overwrite_exit(void);
337 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
338 extern void lttng_ring_buffer_client_discard_exit(void);
339 extern void lttng_ring_buffer_client_discard_rt_exit(void);
340 extern void lttng_ring_buffer_metadata_client_exit(void);
341
342 ssize_t lttng_ust_read(int fd, void *buf, size_t len)
343 {
344 ssize_t ret;
345 size_t copied = 0, to_copy = len;
346
347 do {
348 ret = read(fd, buf + copied, to_copy);
349 if (ret > 0) {
350 copied += ret;
351 to_copy -= ret;
352 }
353 } while ((ret > 0 && to_copy > 0)
354 || (ret < 0 && errno == EINTR));
355 if (ret > 0) {
356 ret = copied;
357 }
358 return ret;
359 }
360 /*
361 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
362 * pointer.
363 */
364 static
365 const char *get_lttng_home_dir(void)
366 {
367 const char *val;
368
369 val = (const char *) lttng_secure_getenv("LTTNG_HOME");
370 if (val != NULL) {
371 return val;
372 }
373 return (const char *) lttng_secure_getenv("HOME");
374 }
375
376 /*
377 * Force a read (imply TLS fixup for dlopen) of TLS variables.
378 */
379 static
380 void lttng_fixup_nest_count_tls(void)
381 {
382 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
383 }
384
385 static
386 void lttng_fixup_ust_mutex_nest_tls(void)
387 {
388 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
389 }
390
391 /*
392 * Fixup urcu bp TLS.
393 */
394 static
395 void lttng_fixup_urcu_bp_tls(void)
396 {
397 rcu_read_lock();
398 rcu_read_unlock();
399 }
400
401 void lttng_ust_fixup_tls(void)
402 {
403 lttng_fixup_urcu_bp_tls();
404 lttng_fixup_ringbuffer_tls();
405 lttng_fixup_vtid_tls();
406 lttng_fixup_nest_count_tls();
407 lttng_fixup_procname_tls();
408 lttng_fixup_ust_mutex_nest_tls();
409 lttng_ust_fixup_fd_tracker_tls();
410 }
411
412 int lttng_get_notify_socket(void *owner)
413 {
414 struct sock_info *info = owner;
415
416 return info->notify_socket;
417 }
418
419 static
420 void print_cmd(int cmd, int handle)
421 {
422 const char *cmd_name = "Unknown";
423
424 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
425 && cmd_name_mapping[cmd]) {
426 cmd_name = cmd_name_mapping[cmd];
427 }
428 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
429 cmd_name, cmd,
430 lttng_ust_obj_get_name(handle), handle);
431 }
432
433 static
434 int setup_local_apps(void)
435 {
436 const char *home_dir;
437 uid_t uid;
438
439 uid = getuid();
440 /*
441 * Disallow per-user tracing for setuid binaries.
442 */
443 if (uid != geteuid()) {
444 assert(local_apps.allowed == 0);
445 return 0;
446 }
447 home_dir = get_lttng_home_dir();
448 if (!home_dir) {
449 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
450 assert(local_apps.allowed == 0);
451 return -ENOENT;
452 }
453 local_apps.allowed = 1;
454 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
455 home_dir,
456 LTTNG_DEFAULT_HOME_RUNDIR,
457 LTTNG_UST_SOCK_FILENAME);
458 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
459 LTTNG_UST_WAIT_FILENAME,
460 uid);
461 return 0;
462 }
463
464 /*
465 * Get socket timeout, in ms.
466 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
467 */
468 static
469 long get_timeout(void)
470 {
471 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
472
473 if (!got_timeout_env) {
474 str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT");
475 got_timeout_env = 1;
476 }
477 if (str_timeout)
478 constructor_delay_ms = strtol(str_timeout, NULL, 10);
479 /* All negative values are considered as "-1". */
480 if (constructor_delay_ms < -1)
481 constructor_delay_ms = -1;
482 return constructor_delay_ms;
483 }
484
485 /* Timeout for notify socket send and recv. */
486 static
487 long get_notify_sock_timeout(void)
488 {
489 return get_timeout();
490 }
491
492 /* Timeout for connecting to cmd and notify sockets. */
493 static
494 long get_connect_sock_timeout(void)
495 {
496 return get_timeout();
497 }
498
499 /*
500 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
501 */
502 static
503 int get_constructor_timeout(struct timespec *constructor_timeout)
504 {
505 long constructor_delay_ms;
506 int ret;
507
508 constructor_delay_ms = get_timeout();
509
510 switch (constructor_delay_ms) {
511 case -1:/* fall-through */
512 case 0:
513 return constructor_delay_ms;
514 default:
515 break;
516 }
517
518 /*
519 * If we are unable to find the current time, don't wait.
520 */
521 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
522 if (ret) {
523 /* Don't wait. */
524 return 0;
525 }
526 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
527 constructor_timeout->tv_nsec +=
528 (constructor_delay_ms % 1000UL) * 1000000UL;
529 if (constructor_timeout->tv_nsec >= 1000000000UL) {
530 constructor_timeout->tv_sec++;
531 constructor_timeout->tv_nsec -= 1000000000UL;
532 }
533 /* Timeout wait (constructor_delay_ms). */
534 return 1;
535 }
536
537 static
538 void get_blocking_retry_timeout(void)
539 {
540 const char *str_blocking_retry_timeout =
541 lttng_secure_getenv("LTTNG_UST_BLOCKING_RETRY_TIMEOUT");
542
543 if (str_blocking_retry_timeout) {
544 long timeout = strtol(str_blocking_retry_timeout, NULL, 10);
545
546 if (timeout < 0)
547 timeout = -1;
548 if (timeout > INT_MAX) {
549 WARN("Saturating %s value from %ld to %d\n",
550 "LTTNG_UST_BLOCKING_RETRY_TIMEOUT",
551 timeout, INT_MAX);
552 timeout = INT_MAX;
553 }
554 DBG("%s environment variable value is %ld",
555 "LTTNG_UST_BLOCKING_RETRY_TIMEOUT",
556 timeout);
557 lttng_ust_ringbuffer_set_retry_timeout(timeout);
558 }
559 }
560
561 static
562 int register_to_sessiond(int socket, enum ustctl_socket_type type)
563 {
564 return ustcomm_send_reg_msg(socket,
565 type,
566 CAA_BITS_PER_LONG,
567 lttng_alignof(uint8_t) * CHAR_BIT,
568 lttng_alignof(uint16_t) * CHAR_BIT,
569 lttng_alignof(uint32_t) * CHAR_BIT,
570 lttng_alignof(uint64_t) * CHAR_BIT,
571 lttng_alignof(unsigned long) * CHAR_BIT);
572 }
573
574 static
575 int send_reply(int sock, struct ustcomm_ust_reply *lur)
576 {
577 ssize_t len;
578
579 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
580 switch (len) {
581 case sizeof(*lur):
582 DBG("message successfully sent");
583 return 0;
584 default:
585 if (len == -ECONNRESET) {
586 DBG("remote end closed connection");
587 return 0;
588 }
589 if (len < 0)
590 return len;
591 DBG("incorrect message size: %zd", len);
592 return -EINVAL;
593 }
594 }
595
596 static
597 int handle_register_done(struct sock_info *sock_info)
598 {
599 int ret;
600
601 if (sock_info->constructor_sem_posted)
602 return 0;
603 sock_info->constructor_sem_posted = 1;
604 if (uatomic_read(&sem_count) <= 0) {
605 return 0;
606 }
607 ret = uatomic_add_return(&sem_count, -1);
608 if (ret == 0) {
609 ret = sem_post(&constructor_wait);
610 assert(!ret);
611 }
612 return 0;
613 }
614
615 /*
616 * Only execute pending statedump after the constructor semaphore has
617 * been posted by each listener thread. This means statedump will only
618 * be performed after the "registration done" command is received from
619 * each session daemon the application is connected to.
620 *
621 * This ensures we don't run into deadlock issues with the dynamic
622 * loader mutex, which is held while the constructor is called and
623 * waiting on the constructor semaphore. All operations requiring this
624 * dynamic loader lock need to be postponed using this mechanism.
625 */
626 static
627 void handle_pending_statedump(struct sock_info *sock_info)
628 {
629 int ctor_passed = sock_info->constructor_sem_posted;
630
631 if (ctor_passed && sock_info->statedump_pending) {
632 sock_info->statedump_pending = 0;
633 pthread_mutex_lock(&ust_fork_mutex);
634 lttng_handle_pending_statedump(sock_info);
635 pthread_mutex_unlock(&ust_fork_mutex);
636 }
637 }
638
639 static
640 int handle_message(struct sock_info *sock_info,
641 int sock, struct ustcomm_ust_msg *lum)
642 {
643 int ret = 0;
644 const struct lttng_ust_objd_ops *ops;
645 struct ustcomm_ust_reply lur;
646 union ust_args args;
647 char ctxstr[LTTNG_UST_SYM_NAME_LEN]; /* App context string. */
648 ssize_t len;
649
650 memset(&lur, 0, sizeof(lur));
651
652 if (ust_lock()) {
653 ret = -LTTNG_UST_ERR_EXITING;
654 goto error;
655 }
656
657 ops = objd_ops(lum->handle);
658 if (!ops) {
659 ret = -ENOENT;
660 goto error;
661 }
662
663 switch (lum->cmd) {
664 case LTTNG_UST_REGISTER_DONE:
665 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
666 ret = handle_register_done(sock_info);
667 else
668 ret = -EINVAL;
669 break;
670 case LTTNG_UST_RELEASE:
671 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
672 ret = -EPERM;
673 else
674 ret = lttng_ust_objd_unref(lum->handle, 1);
675 break;
676 case LTTNG_UST_FILTER:
677 {
678 /* Receive filter data */
679 struct lttng_ust_filter_bytecode_node *bytecode;
680
681 if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
682 ERR("Filter data size is too large: %u bytes",
683 lum->u.filter.data_size);
684 ret = -EINVAL;
685 goto error;
686 }
687
688 if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
689 ERR("Filter reloc offset %u is not within data",
690 lum->u.filter.reloc_offset);
691 ret = -EINVAL;
692 goto error;
693 }
694
695 bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
696 if (!bytecode) {
697 ret = -ENOMEM;
698 goto error;
699 }
700 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
701 lum->u.filter.data_size);
702 switch (len) {
703 case 0: /* orderly shutdown */
704 ret = 0;
705 free(bytecode);
706 goto error;
707 default:
708 if (len == lum->u.filter.data_size) {
709 DBG("filter data received");
710 break;
711 } else if (len < 0) {
712 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
713 if (len == -ECONNRESET) {
714 ERR("%s remote end closed connection", sock_info->name);
715 ret = len;
716 free(bytecode);
717 goto error;
718 }
719 ret = len;
720 free(bytecode);
721 goto error;
722 } else {
723 DBG("incorrect filter data message size: %zd", len);
724 ret = -EINVAL;
725 free(bytecode);
726 goto error;
727 }
728 }
729 bytecode->bc.len = lum->u.filter.data_size;
730 bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
731 bytecode->bc.seqnum = lum->u.filter.seqnum;
732 if (ops->cmd) {
733 ret = ops->cmd(lum->handle, lum->cmd,
734 (unsigned long) bytecode,
735 &args, sock_info);
736 if (ret) {
737 free(bytecode);
738 }
739 /* don't free bytecode if everything went fine. */
740 } else {
741 ret = -ENOSYS;
742 free(bytecode);
743 }
744 break;
745 }
746 case LTTNG_UST_EXCLUSION:
747 {
748 /* Receive exclusion names */
749 struct lttng_ust_excluder_node *node;
750 unsigned int count;
751
752 count = lum->u.exclusion.count;
753 if (count == 0) {
754 /* There are no names to read */
755 ret = 0;
756 goto error;
757 }
758 node = zmalloc(sizeof(*node) +
759 count * LTTNG_UST_SYM_NAME_LEN);
760 if (!node) {
761 ret = -ENOMEM;
762 goto error;
763 }
764 node->excluder.count = count;
765 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
766 count * LTTNG_UST_SYM_NAME_LEN);
767 switch (len) {
768 case 0: /* orderly shutdown */
769 ret = 0;
770 free(node);
771 goto error;
772 default:
773 if (len == count * LTTNG_UST_SYM_NAME_LEN) {
774 DBG("Exclusion data received");
775 break;
776 } else if (len < 0) {
777 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
778 if (len == -ECONNRESET) {
779 ERR("%s remote end closed connection", sock_info->name);
780 ret = len;
781 free(node);
782 goto error;
783 }
784 ret = len;
785 free(node);
786 goto error;
787 } else {
788 DBG("Incorrect exclusion data message size: %zd", len);
789 ret = -EINVAL;
790 free(node);
791 goto error;
792 }
793 }
794 if (ops->cmd) {
795 ret = ops->cmd(lum->handle, lum->cmd,
796 (unsigned long) node,
797 &args, sock_info);
798 if (ret) {
799 free(node);
800 }
801 /* Don't free exclusion data if everything went fine. */
802 } else {
803 ret = -ENOSYS;
804 free(node);
805 }
806 break;
807 }
808 case LTTNG_UST_CHANNEL:
809 {
810 void *chan_data;
811 int wakeup_fd;
812
813 len = ustcomm_recv_channel_from_sessiond(sock,
814 &chan_data, lum->u.channel.len,
815 &wakeup_fd);
816 switch (len) {
817 case 0: /* orderly shutdown */
818 ret = 0;
819 goto error;
820 default:
821 if (len == lum->u.channel.len) {
822 DBG("channel data received");
823 break;
824 } else if (len < 0) {
825 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
826 if (len == -ECONNRESET) {
827 ERR("%s remote end closed connection", sock_info->name);
828 ret = len;
829 goto error;
830 }
831 ret = len;
832 goto error;
833 } else {
834 DBG("incorrect channel data message size: %zd", len);
835 ret = -EINVAL;
836 goto error;
837 }
838 }
839 args.channel.chan_data = chan_data;
840 args.channel.wakeup_fd = wakeup_fd;
841 if (ops->cmd)
842 ret = ops->cmd(lum->handle, lum->cmd,
843 (unsigned long) &lum->u,
844 &args, sock_info);
845 else
846 ret = -ENOSYS;
847 break;
848 }
849 case LTTNG_UST_STREAM:
850 {
851 /* Receive shm_fd, wakeup_fd */
852 ret = ustcomm_recv_stream_from_sessiond(sock,
853 &lum->u.stream.len,
854 &args.stream.shm_fd,
855 &args.stream.wakeup_fd);
856 if (ret) {
857 goto error;
858 }
859 if (ops->cmd)
860 ret = ops->cmd(lum->handle, lum->cmd,
861 (unsigned long) &lum->u,
862 &args, sock_info);
863 else
864 ret = -ENOSYS;
865 break;
866 }
867 case LTTNG_UST_CONTEXT:
868 switch (lum->u.context.ctx) {
869 case LTTNG_UST_CONTEXT_APP_CONTEXT:
870 {
871 char *p;
872 size_t ctxlen, recvlen;
873
874 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
875 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
876 if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
877 ERR("Application context string length size is too large: %zu bytes",
878 ctxlen);
879 ret = -EINVAL;
880 goto error;
881 }
882 strcpy(ctxstr, "$app.");
883 p = &ctxstr[strlen("$app.")];
884 recvlen = ctxlen - strlen("$app.");
885 len = ustcomm_recv_unix_sock(sock, p, recvlen);
886 switch (len) {
887 case 0: /* orderly shutdown */
888 ret = 0;
889 goto error;
890 default:
891 if (len == recvlen) {
892 DBG("app context data received");
893 break;
894 } else if (len < 0) {
895 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
896 if (len == -ECONNRESET) {
897 ERR("%s remote end closed connection", sock_info->name);
898 ret = len;
899 goto error;
900 }
901 ret = len;
902 goto error;
903 } else {
904 DBG("incorrect app context data message size: %zd", len);
905 ret = -EINVAL;
906 goto error;
907 }
908 }
909 /* Put : between provider and ctxname. */
910 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
911 args.app_context.ctxname = ctxstr;
912 break;
913 }
914 default:
915 break;
916 }
917 if (ops->cmd) {
918 ret = ops->cmd(lum->handle, lum->cmd,
919 (unsigned long) &lum->u,
920 &args, sock_info);
921 } else {
922 ret = -ENOSYS;
923 }
924 break;
925 default:
926 if (ops->cmd)
927 ret = ops->cmd(lum->handle, lum->cmd,
928 (unsigned long) &lum->u,
929 &args, sock_info);
930 else
931 ret = -ENOSYS;
932 break;
933 }
934
935 lur.handle = lum->handle;
936 lur.cmd = lum->cmd;
937 lur.ret_val = ret;
938 if (ret >= 0) {
939 lur.ret_code = LTTNG_UST_OK;
940 } else {
941 /*
942 * Use -LTTNG_UST_ERR as wildcard for UST internal
943 * error that are not caused by the transport, except if
944 * we already have a more precise error message to
945 * report.
946 */
947 if (ret > -LTTNG_UST_ERR) {
948 /* Translate code to UST error. */
949 switch (ret) {
950 case -EEXIST:
951 lur.ret_code = -LTTNG_UST_ERR_EXIST;
952 break;
953 case -EINVAL:
954 lur.ret_code = -LTTNG_UST_ERR_INVAL;
955 break;
956 case -ENOENT:
957 lur.ret_code = -LTTNG_UST_ERR_NOENT;
958 break;
959 case -EPERM:
960 lur.ret_code = -LTTNG_UST_ERR_PERM;
961 break;
962 case -ENOSYS:
963 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
964 break;
965 default:
966 lur.ret_code = -LTTNG_UST_ERR;
967 break;
968 }
969 } else {
970 lur.ret_code = ret;
971 }
972 }
973 if (ret >= 0) {
974 switch (lum->cmd) {
975 case LTTNG_UST_TRACER_VERSION:
976 lur.u.version = lum->u.version;
977 break;
978 case LTTNG_UST_TRACEPOINT_LIST_GET:
979 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
980 break;
981 }
982 }
983 DBG("Return value: %d", lur.ret_val);
984
985 ust_unlock();
986
987 /*
988 * Performed delayed statedump operations outside of the UST
989 * lock. We need to take the dynamic loader lock before we take
990 * the UST lock internally within handle_pending_statedump().
991 */
992 handle_pending_statedump(sock_info);
993
994 if (ust_lock()) {
995 ret = -LTTNG_UST_ERR_EXITING;
996 goto error;
997 }
998
999 ret = send_reply(sock, &lur);
1000 if (ret < 0) {
1001 DBG("error sending reply");
1002 goto error;
1003 }
1004
1005 /*
1006 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1007 * after the reply.
1008 */
1009 if (lur.ret_code == LTTNG_UST_OK) {
1010 switch (lum->cmd) {
1011 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
1012 len = ustcomm_send_unix_sock(sock,
1013 &args.field_list.entry,
1014 sizeof(args.field_list.entry));
1015 if (len < 0) {
1016 ret = len;
1017 goto error;
1018 }
1019 if (len != sizeof(args.field_list.entry)) {
1020 ret = -EINVAL;
1021 goto error;
1022 }
1023 }
1024 }
1025
1026 error:
1027 ust_unlock();
1028
1029 return ret;
1030 }
1031
1032 static
1033 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1034 {
1035 int ret;
1036
1037 if (sock_info->root_handle != -1) {
1038 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
1039 if (ret) {
1040 ERR("Error unref root handle");
1041 }
1042 sock_info->root_handle = -1;
1043 }
1044 sock_info->constructor_sem_posted = 0;
1045
1046 /*
1047 * wait_shm_mmap, socket and notify socket are used by listener
1048 * threads outside of the ust lock, so we cannot tear them down
1049 * ourselves, because we cannot join on these threads. Leave
1050 * responsibility of cleaning up these resources to the OS
1051 * process exit.
1052 */
1053 if (exiting)
1054 return;
1055
1056 if (sock_info->socket != -1) {
1057 ret = ustcomm_close_unix_sock(sock_info->socket);
1058 if (ret) {
1059 ERR("Error closing ust cmd socket");
1060 }
1061 sock_info->socket = -1;
1062 }
1063 if (sock_info->notify_socket != -1) {
1064 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1065 if (ret) {
1066 ERR("Error closing ust notify socket");
1067 }
1068 sock_info->notify_socket = -1;
1069 }
1070 if (sock_info->wait_shm_mmap) {
1071 long page_size;
1072
1073 page_size = sysconf(_SC_PAGE_SIZE);
1074 if (page_size <= 0) {
1075 if (!page_size) {
1076 errno = EINVAL;
1077 }
1078 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1079 } else {
1080 ret = munmap(sock_info->wait_shm_mmap, page_size);
1081 if (ret) {
1082 ERR("Error unmapping wait shm");
1083 }
1084 }
1085 sock_info->wait_shm_mmap = NULL;
1086 }
1087 }
1088
1089 /*
1090 * Using fork to set umask in the child process (not multi-thread safe).
1091 * We deal with the shm_open vs ftruncate race (happening when the
1092 * sessiond owns the shm and does not let everybody modify it, to ensure
1093 * safety against shm_unlink) by simply letting the mmap fail and
1094 * retrying after a few seconds.
1095 * For global shm, everybody has rw access to it until the sessiond
1096 * starts.
1097 */
1098 static
1099 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1100 {
1101 int wait_shm_fd, ret;
1102 pid_t pid;
1103
1104 /*
1105 * Try to open read-only.
1106 */
1107 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1108 if (wait_shm_fd >= 0) {
1109 int32_t tmp_read;
1110 ssize_t len;
1111 size_t bytes_read = 0;
1112
1113 /*
1114 * Try to read the fd. If unable to do so, try opening
1115 * it in write mode.
1116 */
1117 do {
1118 len = read(wait_shm_fd,
1119 &((char *) &tmp_read)[bytes_read],
1120 sizeof(tmp_read) - bytes_read);
1121 if (len > 0) {
1122 bytes_read += len;
1123 }
1124 } while ((len < 0 && errno == EINTR)
1125 || (len > 0 && bytes_read < sizeof(tmp_read)));
1126 if (bytes_read != sizeof(tmp_read)) {
1127 ret = close(wait_shm_fd);
1128 if (ret) {
1129 ERR("close wait_shm_fd");
1130 }
1131 goto open_write;
1132 }
1133 goto end;
1134 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1135 /*
1136 * Real-only open did not work, and it's not because the
1137 * entry was not present. It's a failure that prohibits
1138 * using shm.
1139 */
1140 ERR("Error opening shm %s", sock_info->wait_shm_path);
1141 goto end;
1142 }
1143
1144 open_write:
1145 /*
1146 * If the open failed because the file did not exist, or because
1147 * the file was not truncated yet, try creating it ourself.
1148 */
1149 URCU_TLS(lttng_ust_nest_count)++;
1150 pid = fork();
1151 URCU_TLS(lttng_ust_nest_count)--;
1152 if (pid > 0) {
1153 int status;
1154
1155 /*
1156 * Parent: wait for child to return, in which case the
1157 * shared memory map will have been created.
1158 */
1159 pid = wait(&status);
1160 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1161 wait_shm_fd = -1;
1162 goto end;
1163 }
1164 /*
1165 * Try to open read-only again after creation.
1166 */
1167 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1168 if (wait_shm_fd < 0) {
1169 /*
1170 * Real-only open did not work. It's a failure
1171 * that prohibits using shm.
1172 */
1173 ERR("Error opening shm %s", sock_info->wait_shm_path);
1174 goto end;
1175 }
1176 goto end;
1177 } else if (pid == 0) {
1178 int create_mode;
1179
1180 /* Child */
1181 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1182 if (sock_info->global)
1183 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1184 /*
1185 * We're alone in a child process, so we can modify the
1186 * process-wide umask.
1187 */
1188 umask(~create_mode);
1189 /*
1190 * Try creating shm (or get rw access).
1191 * We don't do an exclusive open, because we allow other
1192 * processes to create+ftruncate it concurrently.
1193 */
1194 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1195 O_RDWR | O_CREAT, create_mode);
1196 if (wait_shm_fd >= 0) {
1197 ret = ftruncate(wait_shm_fd, mmap_size);
1198 if (ret) {
1199 PERROR("ftruncate");
1200 _exit(EXIT_FAILURE);
1201 }
1202 _exit(EXIT_SUCCESS);
1203 }
1204 /*
1205 * For local shm, we need to have rw access to accept
1206 * opening it: this means the local sessiond will be
1207 * able to wake us up. For global shm, we open it even
1208 * if rw access is not granted, because the root.root
1209 * sessiond will be able to override all rights and wake
1210 * us up.
1211 */
1212 if (!sock_info->global && errno != EACCES) {
1213 ERR("Error opening shm %s", sock_info->wait_shm_path);
1214 _exit(EXIT_FAILURE);
1215 }
1216 /*
1217 * The shm exists, but we cannot open it RW. Report
1218 * success.
1219 */
1220 _exit(EXIT_SUCCESS);
1221 } else {
1222 return -1;
1223 }
1224 end:
1225 if (wait_shm_fd >= 0 && !sock_info->global) {
1226 struct stat statbuf;
1227
1228 /*
1229 * Ensure that our user is the owner of the shm file for
1230 * local shm. If we do not own the file, it means our
1231 * sessiond will not have access to wake us up (there is
1232 * probably a rogue process trying to fake our
1233 * sessiond). Fallback to polling method in this case.
1234 */
1235 ret = fstat(wait_shm_fd, &statbuf);
1236 if (ret) {
1237 PERROR("fstat");
1238 goto error_close;
1239 }
1240 if (statbuf.st_uid != getuid())
1241 goto error_close;
1242 }
1243 return wait_shm_fd;
1244
1245 error_close:
1246 ret = close(wait_shm_fd);
1247 if (ret) {
1248 PERROR("Error closing fd");
1249 }
1250 return -1;
1251 }
1252
1253 static
1254 char *get_map_shm(struct sock_info *sock_info)
1255 {
1256 long page_size;
1257 int wait_shm_fd, ret;
1258 char *wait_shm_mmap;
1259
1260 page_size = sysconf(_SC_PAGE_SIZE);
1261 if (page_size <= 0) {
1262 if (!page_size) {
1263 errno = EINVAL;
1264 }
1265 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1266 goto error;
1267 }
1268
1269 lttng_ust_lock_fd_tracker();
1270 wait_shm_fd = get_wait_shm(sock_info, page_size);
1271 if (wait_shm_fd < 0) {
1272 lttng_ust_unlock_fd_tracker();
1273 goto error;
1274 }
1275 lttng_ust_add_fd_to_tracker(wait_shm_fd);
1276 lttng_ust_unlock_fd_tracker();
1277
1278 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1279 MAP_SHARED, wait_shm_fd, 0);
1280
1281 /* close shm fd immediately after taking the mmap reference */
1282 lttng_ust_lock_fd_tracker();
1283 ret = close(wait_shm_fd);
1284 if (!ret) {
1285 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1286 } else {
1287 PERROR("Error closing fd");
1288 }
1289 lttng_ust_unlock_fd_tracker();
1290
1291 if (wait_shm_mmap == MAP_FAILED) {
1292 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1293 goto error;
1294 }
1295 return wait_shm_mmap;
1296
1297 error:
1298 return NULL;
1299 }
1300
1301 static
1302 void wait_for_sessiond(struct sock_info *sock_info)
1303 {
1304 if (ust_lock()) {
1305 goto quit;
1306 }
1307 if (wait_poll_fallback) {
1308 goto error;
1309 }
1310 if (!sock_info->wait_shm_mmap) {
1311 sock_info->wait_shm_mmap = get_map_shm(sock_info);
1312 if (!sock_info->wait_shm_mmap)
1313 goto error;
1314 }
1315 ust_unlock();
1316
1317 DBG("Waiting for %s apps sessiond", sock_info->name);
1318 /* Wait for futex wakeup */
1319 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1320 goto end_wait;
1321
1322 while (futex_async((int32_t *) sock_info->wait_shm_mmap,
1323 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1324 switch (errno) {
1325 case EWOULDBLOCK:
1326 /* Value already changed. */
1327 goto end_wait;
1328 case EINTR:
1329 /* Retry if interrupted by signal. */
1330 break; /* Get out of switch. */
1331 case EFAULT:
1332 wait_poll_fallback = 1;
1333 DBG(
1334 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1335 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1336 "Please upgrade your kernel "
1337 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1338 "mainline). LTTng-UST will use polling mode fallback.");
1339 if (ust_debug())
1340 PERROR("futex");
1341 goto end_wait;
1342 }
1343 }
1344 end_wait:
1345 return;
1346
1347 quit:
1348 ust_unlock();
1349 return;
1350
1351 error:
1352 ust_unlock();
1353 return;
1354 }
1355
1356 /*
1357 * This thread does not allocate any resource, except within
1358 * handle_message, within mutex protection. This mutex protects against
1359 * fork and exit.
1360 * The other moment it allocates resources is at socket connection, which
1361 * is also protected by the mutex.
1362 */
1363 static
1364 void *ust_listener_thread(void *arg)
1365 {
1366 struct sock_info *sock_info = arg;
1367 int sock, ret, prev_connect_failed = 0, has_waited = 0;
1368 long timeout;
1369
1370 lttng_ust_fixup_tls();
1371 /*
1372 * If available, add '-ust' to the end of this thread's
1373 * process name
1374 */
1375 ret = lttng_ust_setustprocname();
1376 if (ret) {
1377 ERR("Unable to set UST process name");
1378 }
1379
1380 /* Restart trying to connect to the session daemon */
1381 restart:
1382 if (prev_connect_failed) {
1383 /* Wait for sessiond availability with pipe */
1384 wait_for_sessiond(sock_info);
1385 if (has_waited) {
1386 has_waited = 0;
1387 /*
1388 * Sleep for 5 seconds before retrying after a
1389 * sequence of failure / wait / failure. This
1390 * deals with a killed or broken session daemon.
1391 */
1392 sleep(5);
1393 } else {
1394 has_waited = 1;
1395 }
1396 prev_connect_failed = 0;
1397 }
1398
1399 if (sock_info->socket != -1) {
1400 /* FD tracker is updated by ustcomm_close_unix_sock() */
1401 ret = ustcomm_close_unix_sock(sock_info->socket);
1402 if (ret) {
1403 ERR("Error closing %s ust cmd socket",
1404 sock_info->name);
1405 }
1406 sock_info->socket = -1;
1407 }
1408 if (sock_info->notify_socket != -1) {
1409 /* FD tracker is updated by ustcomm_close_unix_sock() */
1410 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1411 if (ret) {
1412 ERR("Error closing %s ust notify socket",
1413 sock_info->name);
1414 }
1415 sock_info->notify_socket = -1;
1416 }
1417
1418 if (ust_lock()) {
1419 goto quit;
1420 }
1421
1422 /*
1423 * Register. We need to perform both connect and sending
1424 * registration message before doing the next connect otherwise
1425 * we may reach unix socket connect queue max limits and block
1426 * on the 2nd connect while the session daemon is awaiting the
1427 * first connect registration message.
1428 */
1429 /* Connect cmd socket */
1430 lttng_ust_lock_fd_tracker();
1431 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1432 get_connect_sock_timeout());
1433 if (ret < 0) {
1434 lttng_ust_unlock_fd_tracker();
1435 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1436 prev_connect_failed = 1;
1437
1438 /*
1439 * If we cannot find the sessiond daemon, don't delay
1440 * constructor execution.
1441 */
1442 ret = handle_register_done(sock_info);
1443 assert(!ret);
1444 ust_unlock();
1445 goto restart;
1446 }
1447 lttng_ust_add_fd_to_tracker(ret);
1448 lttng_ust_unlock_fd_tracker();
1449 sock_info->socket = ret;
1450
1451 ust_unlock();
1452 /*
1453 * Unlock/relock ust lock because connect is blocking (with
1454 * timeout). Don't delay constructors on the ust lock for too
1455 * long.
1456 */
1457 if (ust_lock()) {
1458 goto quit;
1459 }
1460
1461 /*
1462 * Create only one root handle per listener thread for the whole
1463 * process lifetime, so we ensure we get ID which is statically
1464 * assigned to the root handle.
1465 */
1466 if (sock_info->root_handle == -1) {
1467 ret = lttng_abi_create_root_handle();
1468 if (ret < 0) {
1469 ERR("Error creating root handle");
1470 goto quit;
1471 }
1472 sock_info->root_handle = ret;
1473 }
1474
1475 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1476 if (ret < 0) {
1477 ERR("Error registering to %s ust cmd socket",
1478 sock_info->name);
1479 prev_connect_failed = 1;
1480 /*
1481 * If we cannot register to the sessiond daemon, don't
1482 * delay constructor execution.
1483 */
1484 ret = handle_register_done(sock_info);
1485 assert(!ret);
1486 ust_unlock();
1487 goto restart;
1488 }
1489
1490 ust_unlock();
1491 /*
1492 * Unlock/relock ust lock because connect is blocking (with
1493 * timeout). Don't delay constructors on the ust lock for too
1494 * long.
1495 */
1496 if (ust_lock()) {
1497 goto quit;
1498 }
1499
1500 /* Connect notify socket */
1501 lttng_ust_lock_fd_tracker();
1502 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1503 get_connect_sock_timeout());
1504 if (ret < 0) {
1505 lttng_ust_unlock_fd_tracker();
1506 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1507 prev_connect_failed = 1;
1508
1509 /*
1510 * If we cannot find the sessiond daemon, don't delay
1511 * constructor execution.
1512 */
1513 ret = handle_register_done(sock_info);
1514 assert(!ret);
1515 ust_unlock();
1516 goto restart;
1517 }
1518 lttng_ust_add_fd_to_tracker(ret);
1519 lttng_ust_unlock_fd_tracker();
1520 sock_info->notify_socket = ret;
1521
1522 ust_unlock();
1523 /*
1524 * Unlock/relock ust lock because connect is blocking (with
1525 * timeout). Don't delay constructors on the ust lock for too
1526 * long.
1527 */
1528 if (ust_lock()) {
1529 goto quit;
1530 }
1531
1532 timeout = get_notify_sock_timeout();
1533 if (timeout >= 0) {
1534 /*
1535 * Give at least 10ms to sessiond to reply to
1536 * notifications.
1537 */
1538 if (timeout < 10)
1539 timeout = 10;
1540 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1541 timeout);
1542 if (ret < 0) {
1543 WARN("Error setting socket receive timeout");
1544 }
1545 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1546 timeout);
1547 if (ret < 0) {
1548 WARN("Error setting socket send timeout");
1549 }
1550 } else if (timeout < -1) {
1551 WARN("Unsupported timeout value %ld", timeout);
1552 }
1553
1554 ret = register_to_sessiond(sock_info->notify_socket,
1555 USTCTL_SOCKET_NOTIFY);
1556 if (ret < 0) {
1557 ERR("Error registering to %s ust notify socket",
1558 sock_info->name);
1559 prev_connect_failed = 1;
1560 /*
1561 * If we cannot register to the sessiond daemon, don't
1562 * delay constructor execution.
1563 */
1564 ret = handle_register_done(sock_info);
1565 assert(!ret);
1566 ust_unlock();
1567 goto restart;
1568 }
1569 sock = sock_info->socket;
1570
1571 ust_unlock();
1572
1573 for (;;) {
1574 ssize_t len;
1575 struct ustcomm_ust_msg lum;
1576
1577 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1578 switch (len) {
1579 case 0: /* orderly shutdown */
1580 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1581 if (ust_lock()) {
1582 goto quit;
1583 }
1584 /*
1585 * Either sessiond has shutdown or refused us by closing the socket.
1586 * In either case, we don't want to delay construction execution,
1587 * and we need to wait before retry.
1588 */
1589 prev_connect_failed = 1;
1590 /*
1591 * If we cannot register to the sessiond daemon, don't
1592 * delay constructor execution.
1593 */
1594 ret = handle_register_done(sock_info);
1595 assert(!ret);
1596 ust_unlock();
1597 goto end;
1598 case sizeof(lum):
1599 print_cmd(lum.cmd, lum.handle);
1600 ret = handle_message(sock_info, sock, &lum);
1601 if (ret) {
1602 ERR("Error handling message for %s socket",
1603 sock_info->name);
1604 /*
1605 * Close socket if protocol error is
1606 * detected.
1607 */
1608 goto end;
1609 }
1610 continue;
1611 default:
1612 if (len < 0) {
1613 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1614 } else {
1615 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1616 }
1617 if (len == -ECONNRESET) {
1618 DBG("%s remote end closed connection", sock_info->name);
1619 goto end;
1620 }
1621 goto end;
1622 }
1623
1624 }
1625 end:
1626 if (ust_lock()) {
1627 goto quit;
1628 }
1629 /* Cleanup socket handles before trying to reconnect */
1630 lttng_ust_objd_table_owner_cleanup(sock_info);
1631 ust_unlock();
1632 goto restart; /* try to reconnect */
1633
1634 quit:
1635 ust_unlock();
1636
1637 pthread_mutex_lock(&ust_exit_mutex);
1638 sock_info->thread_active = 0;
1639 pthread_mutex_unlock(&ust_exit_mutex);
1640 return NULL;
1641 }
1642
1643 /*
1644 * Weak symbol to call when the ust malloc wrapper is not loaded.
1645 */
1646 __attribute__((weak))
1647 void lttng_ust_malloc_wrapper_init(void)
1648 {
1649 }
1650
1651 /*
1652 * sessiond monitoring thread: monitor presence of global and per-user
1653 * sessiond by polling the application common named pipe.
1654 */
1655 void __attribute__((constructor)) lttng_ust_init(void)
1656 {
1657 struct timespec constructor_timeout;
1658 sigset_t sig_all_blocked, orig_parent_mask;
1659 pthread_attr_t thread_attr;
1660 int timeout_mode;
1661 int ret;
1662
1663 if (uatomic_xchg(&initialized, 1) == 1)
1664 return;
1665
1666 /*
1667 * Fixup interdependency between TLS fixup mutex (which happens
1668 * to be the dynamic linker mutex) and ust_lock, taken within
1669 * the ust lock.
1670 */
1671 lttng_ust_fixup_tls();
1672
1673 lttng_ust_loaded = 1;
1674
1675 /*
1676 * We want precise control over the order in which we construct
1677 * our sub-libraries vs starting to receive commands from
1678 * sessiond (otherwise leading to errors when trying to create
1679 * sessiond before the init functions are completed).
1680 */
1681 init_usterr();
1682 init_tracepoint();
1683 lttng_ust_init_fd_tracker();
1684 lttng_ust_clock_init();
1685 lttng_ust_getcpu_init();
1686 lttng_ust_statedump_init();
1687 lttng_ring_buffer_metadata_client_init();
1688 lttng_ring_buffer_client_overwrite_init();
1689 lttng_ring_buffer_client_overwrite_rt_init();
1690 lttng_ring_buffer_client_discard_init();
1691 lttng_ring_buffer_client_discard_rt_init();
1692 lttng_perf_counter_init();
1693 /*
1694 * Invoke ust malloc wrapper init before starting other threads.
1695 */
1696 lttng_ust_malloc_wrapper_init();
1697
1698 timeout_mode = get_constructor_timeout(&constructor_timeout);
1699
1700 get_blocking_retry_timeout();
1701
1702 ret = sem_init(&constructor_wait, 0, 0);
1703 if (ret) {
1704 PERROR("sem_init");
1705 }
1706
1707 ret = setup_local_apps();
1708 if (ret) {
1709 DBG("local apps setup returned %d", ret);
1710 }
1711
1712 /* A new thread created by pthread_create inherits the signal mask
1713 * from the parent. To avoid any signal being received by the
1714 * listener thread, we block all signals temporarily in the parent,
1715 * while we create the listener thread.
1716 */
1717 sigfillset(&sig_all_blocked);
1718 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
1719 if (ret) {
1720 ERR("pthread_sigmask: %s", strerror(ret));
1721 }
1722
1723 ret = pthread_attr_init(&thread_attr);
1724 if (ret) {
1725 ERR("pthread_attr_init: %s", strerror(ret));
1726 }
1727 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
1728 if (ret) {
1729 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
1730 }
1731
1732 pthread_mutex_lock(&ust_exit_mutex);
1733 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
1734 ust_listener_thread, &global_apps);
1735 if (ret) {
1736 ERR("pthread_create global: %s", strerror(ret));
1737 }
1738 global_apps.thread_active = 1;
1739 pthread_mutex_unlock(&ust_exit_mutex);
1740
1741 if (local_apps.allowed) {
1742 pthread_mutex_lock(&ust_exit_mutex);
1743 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
1744 ust_listener_thread, &local_apps);
1745 if (ret) {
1746 ERR("pthread_create local: %s", strerror(ret));
1747 }
1748 local_apps.thread_active = 1;
1749 pthread_mutex_unlock(&ust_exit_mutex);
1750 } else {
1751 handle_register_done(&local_apps);
1752 }
1753 ret = pthread_attr_destroy(&thread_attr);
1754 if (ret) {
1755 ERR("pthread_attr_destroy: %s", strerror(ret));
1756 }
1757
1758 /* Restore original signal mask in parent */
1759 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
1760 if (ret) {
1761 ERR("pthread_sigmask: %s", strerror(ret));
1762 }
1763
1764 switch (timeout_mode) {
1765 case 1: /* timeout wait */
1766 do {
1767 ret = sem_timedwait(&constructor_wait,
1768 &constructor_timeout);
1769 } while (ret < 0 && errno == EINTR);
1770 if (ret < 0) {
1771 switch (errno) {
1772 case ETIMEDOUT:
1773 ERR("Timed out waiting for lttng-sessiond");
1774 break;
1775 case EINVAL:
1776 PERROR("sem_timedwait");
1777 break;
1778 default:
1779 ERR("Unexpected error \"%s\" returned by sem_timedwait",
1780 strerror(errno));
1781 }
1782 }
1783 break;
1784 case -1:/* wait forever */
1785 do {
1786 ret = sem_wait(&constructor_wait);
1787 } while (ret < 0 && errno == EINTR);
1788 if (ret < 0) {
1789 switch (errno) {
1790 case EINVAL:
1791 PERROR("sem_wait");
1792 break;
1793 default:
1794 ERR("Unexpected error \"%s\" returned by sem_wait",
1795 strerror(errno));
1796 }
1797 }
1798 break;
1799 case 0: /* no timeout */
1800 break;
1801 }
1802 }
1803
1804 static
1805 void lttng_ust_cleanup(int exiting)
1806 {
1807 cleanup_sock_info(&global_apps, exiting);
1808 cleanup_sock_info(&local_apps, exiting);
1809 local_apps.allowed = 0;
1810 /*
1811 * The teardown in this function all affect data structures
1812 * accessed under the UST lock by the listener thread. This
1813 * lock, along with the lttng_ust_comm_should_quit flag, ensure
1814 * that none of these threads are accessing this data at this
1815 * point.
1816 */
1817 lttng_ust_abi_exit();
1818 lttng_ust_events_exit();
1819 lttng_perf_counter_exit();
1820 lttng_ring_buffer_client_discard_rt_exit();
1821 lttng_ring_buffer_client_discard_exit();
1822 lttng_ring_buffer_client_overwrite_rt_exit();
1823 lttng_ring_buffer_client_overwrite_exit();
1824 lttng_ring_buffer_metadata_client_exit();
1825 lttng_ust_statedump_destroy();
1826 exit_tracepoint();
1827 if (!exiting) {
1828 /* Reinitialize values for fork */
1829 sem_count = 2;
1830 lttng_ust_comm_should_quit = 0;
1831 initialized = 0;
1832 }
1833 }
1834
1835 void __attribute__((destructor)) lttng_ust_exit(void)
1836 {
1837 int ret;
1838
1839 /*
1840 * Using pthread_cancel here because:
1841 * A) we don't want to hang application teardown.
1842 * B) the thread is not allocating any resource.
1843 */
1844
1845 /*
1846 * Require the communication thread to quit. Synchronize with
1847 * mutexes to ensure it is not in a mutex critical section when
1848 * pthread_cancel is later called.
1849 */
1850 ust_lock_nocheck();
1851 lttng_ust_comm_should_quit = 1;
1852 ust_unlock();
1853
1854 pthread_mutex_lock(&ust_exit_mutex);
1855 /* cancel threads */
1856 if (global_apps.thread_active) {
1857 ret = pthread_cancel(global_apps.ust_listener);
1858 if (ret) {
1859 ERR("Error cancelling global ust listener thread: %s",
1860 strerror(ret));
1861 } else {
1862 global_apps.thread_active = 0;
1863 }
1864 }
1865 if (local_apps.thread_active) {
1866 ret = pthread_cancel(local_apps.ust_listener);
1867 if (ret) {
1868 ERR("Error cancelling local ust listener thread: %s",
1869 strerror(ret));
1870 } else {
1871 local_apps.thread_active = 0;
1872 }
1873 }
1874 pthread_mutex_unlock(&ust_exit_mutex);
1875
1876 /*
1877 * Do NOT join threads: use of sys_futex makes it impossible to
1878 * join the threads without using async-cancel, but async-cancel
1879 * is delivered by a signal, which could hit the target thread
1880 * anywhere in its code path, including while the ust_lock() is
1881 * held, causing a deadlock for the other thread. Let the OS
1882 * cleanup the threads if there are stalled in a syscall.
1883 */
1884 lttng_ust_cleanup(1);
1885 }
1886
1887 /*
1888 * We exclude the worker threads across fork and clone (except
1889 * CLONE_VM), because these system calls only keep the forking thread
1890 * running in the child. Therefore, we don't want to call fork or clone
1891 * in the middle of an tracepoint or ust tracing state modification.
1892 * Holding this mutex protects these structures across fork and clone.
1893 */
1894 void ust_before_fork(sigset_t *save_sigset)
1895 {
1896 /*
1897 * Disable signals. This is to avoid that the child intervenes
1898 * before it is properly setup for tracing. It is safer to
1899 * disable all signals, because then we know we are not breaking
1900 * anything by restoring the original mask.
1901 */
1902 sigset_t all_sigs;
1903 int ret;
1904
1905 /* Fixup lttng-ust TLS. */
1906 lttng_ust_fixup_tls();
1907
1908 if (URCU_TLS(lttng_ust_nest_count))
1909 return;
1910 /* Disable signals */
1911 sigfillset(&all_sigs);
1912 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
1913 if (ret == -1) {
1914 PERROR("sigprocmask");
1915 }
1916
1917 pthread_mutex_lock(&ust_fork_mutex);
1918
1919 ust_lock_nocheck();
1920 rcu_bp_before_fork();
1921 }
1922
1923 static void ust_after_fork_common(sigset_t *restore_sigset)
1924 {
1925 int ret;
1926
1927 DBG("process %d", getpid());
1928 ust_unlock();
1929
1930 pthread_mutex_unlock(&ust_fork_mutex);
1931
1932 /* Restore signals */
1933 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
1934 if (ret == -1) {
1935 PERROR("sigprocmask");
1936 }
1937 }
1938
1939 void ust_after_fork_parent(sigset_t *restore_sigset)
1940 {
1941 if (URCU_TLS(lttng_ust_nest_count))
1942 return;
1943 DBG("process %d", getpid());
1944 rcu_bp_after_fork_parent();
1945 /* Release mutexes and reenable signals */
1946 ust_after_fork_common(restore_sigset);
1947 }
1948
1949 /*
1950 * After fork, in the child, we need to cleanup all the leftover state,
1951 * except the worker thread which already magically disappeared thanks
1952 * to the weird Linux fork semantics. After tyding up, we call
1953 * lttng_ust_init() again to start over as a new PID.
1954 *
1955 * This is meant for forks() that have tracing in the child between the
1956 * fork and following exec call (if there is any).
1957 */
1958 void ust_after_fork_child(sigset_t *restore_sigset)
1959 {
1960 if (URCU_TLS(lttng_ust_nest_count))
1961 return;
1962 lttng_context_vtid_reset();
1963 DBG("process %d", getpid());
1964 /* Release urcu mutexes */
1965 rcu_bp_after_fork_child();
1966 lttng_ust_cleanup(0);
1967 /* Release mutexes and reenable signals */
1968 ust_after_fork_common(restore_sigset);
1969 lttng_ust_init();
1970 }
1971
1972 void lttng_ust_sockinfo_session_enabled(void *owner)
1973 {
1974 struct sock_info *sock_info = owner;
1975 sock_info->statedump_pending = 1;
1976 }
LTTng 2.0 Userspace Tracer
This page took 0.119983 seconds and 3 git commands to generate.