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


fba236864e5a3aff7db36a2f43bc634dcb8e1288
[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 <urcu/uatomic.h>
39 #include <urcu/futex.h>
40 #include <urcu/compiler.h>
41
42 #include <lttng/ust-events.h>
43 #include <lttng/ust-abi.h>
44 #include <lttng/ust.h>
45 #include <lttng/ust-error.h>
46 #include <lttng/ust-ctl.h>
47 #include <urcu/tls-compat.h>
48 #include <ust-comm.h>
49 #include <ust-fd.h>
50 #include <usterr-signal-safe.h>
51 #include <helper.h>
52 #include "tracepoint-internal.h"
53 #include "lttng-tracer-core.h"
54 #include "compat.h"
55 #include "../libringbuffer/tlsfixup.h"
56 #include "lttng-ust-statedump.h"
57 #include "clock.h"
58 #include "../libringbuffer/getcpu.h"
59 #include "getenv.h"
60
61 /*
62 * Has lttng ust comm constructor been called ?
63 */
64 static int initialized;
65
66 /*
67 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
68 * Held when handling a command, also held by fork() to deal with
69 * removal of threads, and by exit path.
70 *
71 * The UST lock is the centralized mutex across UST tracing control and
72 * probe registration.
73 *
74 * ust_exit_mutex must never nest in ust_mutex.
75 *
76 * ust_fork_mutex must never nest in ust_mutex.
77 *
78 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
79 * counter lazy initialization called by events within the statedump,
80 * which traces while the ust_mutex is held.
81 *
82 * ust_lock nests within the dynamic loader lock (within glibc) because
83 * it is taken within the library constructor.
84 */
85 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
86
87 /* Allow nesting the ust_mutex within the same thread. */
88 static DEFINE_URCU_TLS(int, ust_mutex_nest);
89
90 /*
91 * ust_exit_mutex protects thread_active variable wrt thread exit. It
92 * cannot be done by ust_mutex because pthread_cancel(), which takes an
93 * internal libc lock, cannot nest within ust_mutex.
94 *
95 * It never nests within a ust_mutex.
96 */
97 static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
98
99 /*
100 * ust_fork_mutex protects base address statedump tracing against forks. It
101 * prevents the dynamic loader lock to be taken (by base address statedump
102 * tracing) while a fork is happening, thus preventing deadlock issues with
103 * the dynamic loader lock.
104 */
105 static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
106
107 /* Should the ust comm thread quit ? */
108 static int lttng_ust_comm_should_quit;
109
110 /*
111 * This variable can be tested by applications to check whether
112 * lttng-ust is loaded. They simply have to define their own
113 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
114 * library constructor.
115 */
116 int lttng_ust_loaded __attribute__((weak));
117
118 /*
119 * Return 0 on success, -1 if should quit.
120 * The lock is taken in both cases.
121 * Signal-safe.
122 */
123 int ust_lock(void)
124 {
125 sigset_t sig_all_blocked, orig_mask;
126 int ret, oldstate;
127
128 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
129 if (ret) {
130 ERR("pthread_setcancelstate: %s", strerror(ret));
131 }
132 if (oldstate != PTHREAD_CANCEL_ENABLE) {
133 ERR("pthread_setcancelstate: unexpected oldstate");
134 }
135 sigfillset(&sig_all_blocked);
136 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
137 if (ret) {
138 ERR("pthread_sigmask: %s", strerror(ret));
139 }
140 if (!URCU_TLS(ust_mutex_nest)++)
141 pthread_mutex_lock(&ust_mutex);
142 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
143 if (ret) {
144 ERR("pthread_sigmask: %s", strerror(ret));
145 }
146 if (lttng_ust_comm_should_quit) {
147 return -1;
148 } else {
149 return 0;
150 }
151 }
152
153 /*
154 * ust_lock_nocheck() can be used in constructors/destructors, because
155 * they are already nested within the dynamic loader lock, and therefore
156 * have exclusive access against execution of liblttng-ust destructor.
157 * Signal-safe.
158 */
159 void ust_lock_nocheck(void)
160 {
161 sigset_t sig_all_blocked, orig_mask;
162 int ret, oldstate;
163
164 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
165 if (ret) {
166 ERR("pthread_setcancelstate: %s", strerror(ret));
167 }
168 if (oldstate != PTHREAD_CANCEL_ENABLE) {
169 ERR("pthread_setcancelstate: unexpected oldstate");
170 }
171 sigfillset(&sig_all_blocked);
172 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
173 if (ret) {
174 ERR("pthread_sigmask: %s", strerror(ret));
175 }
176 if (!URCU_TLS(ust_mutex_nest)++)
177 pthread_mutex_lock(&ust_mutex);
178 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
179 if (ret) {
180 ERR("pthread_sigmask: %s", strerror(ret));
181 }
182 }
183
184 /*
185 * Signal-safe.
186 */
187 void ust_unlock(void)
188 {
189 sigset_t sig_all_blocked, orig_mask;
190 int ret, oldstate;
191
192 sigfillset(&sig_all_blocked);
193 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
194 if (ret) {
195 ERR("pthread_sigmask: %s", strerror(ret));
196 }
197 if (!--URCU_TLS(ust_mutex_nest))
198 pthread_mutex_unlock(&ust_mutex);
199 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
200 if (ret) {
201 ERR("pthread_sigmask: %s", strerror(ret));
202 }
203 ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
204 if (ret) {
205 ERR("pthread_setcancelstate: %s", strerror(ret));
206 }
207 if (oldstate != PTHREAD_CANCEL_DISABLE) {
208 ERR("pthread_setcancelstate: unexpected oldstate");
209 }
210 }
211
212 /*
213 * Wait for either of these before continuing to the main
214 * program:
215 * - the register_done message from sessiond daemon
216 * (will let the sessiond daemon enable sessions before main
217 * starts.)
218 * - sessiond daemon is not reachable.
219 * - timeout (ensuring applications are resilient to session
220 * daemon problems).
221 */
222 static sem_t constructor_wait;
223 /*
224 * Doing this for both the global and local sessiond.
225 */
226 static int sem_count = { 2 };
227
228 /*
229 * Counting nesting within lttng-ust. Used to ensure that calling fork()
230 * from liblttng-ust does not execute the pre/post fork handlers.
231 */
232 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
233
234 /*
235 * Info about socket and associated listener thread.
236 */
237 struct sock_info {
238 const char *name;
239 pthread_t ust_listener; /* listener thread */
240 int root_handle;
241 int constructor_sem_posted;
242 int allowed;
243 int global;
244 int thread_active;
245
246 char sock_path[PATH_MAX];
247 int socket;
248 int notify_socket;
249
250 char wait_shm_path[PATH_MAX];
251 char *wait_shm_mmap;
252 /* Keep track of lazy state dump not performed yet. */
253 int statedump_pending;
254 };
255
256 /* Socket from app (connect) to session daemon (listen) for communication */
257 struct sock_info global_apps = {
258 .name = "global",
259 .global = 1,
260
261 .root_handle = -1,
262 .allowed = 0,
263 .thread_active = 0,
264
265 .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
266 .socket = -1,
267 .notify_socket = -1,
268
269 .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
270
271 .statedump_pending = 0,
272 };
273
274 /* TODO: allow global_apps_sock_path override */
275
276 struct sock_info local_apps = {
277 .name = "local",
278 .global = 0,
279 .root_handle = -1,
280 .allowed = 0, /* Check setuid bit first */
281 .thread_active = 0,
282
283 .socket = -1,
284 .notify_socket = -1,
285
286 .statedump_pending = 0,
287 };
288
289 static int wait_poll_fallback;
290
291 static const char *cmd_name_mapping[] = {
292 [ LTTNG_UST_RELEASE ] = "Release",
293 [ LTTNG_UST_SESSION ] = "Create Session",
294 [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
295
296 [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
297 [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
298 [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
299 [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
300
301 /* Session FD commands */
302 [ LTTNG_UST_CHANNEL ] = "Create Channel",
303 [ LTTNG_UST_SESSION_START ] = "Start Session",
304 [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
305
306 /* Channel FD commands */
307 [ LTTNG_UST_STREAM ] = "Create Stream",
308 [ LTTNG_UST_EVENT ] = "Create Event",
309
310 /* Event and Channel FD commands */
311 [ LTTNG_UST_CONTEXT ] = "Create Context",
312 [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
313
314 /* Event, Channel and Session commands */
315 [ LTTNG_UST_ENABLE ] = "Enable",
316 [ LTTNG_UST_DISABLE ] = "Disable",
317
318 /* Tracepoint list commands */
319 [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
320 [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
321
322 /* Event FD commands */
323 [ LTTNG_UST_FILTER ] = "Create Filter",
324 [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
325 };
326
327 static const char *str_timeout;
328 static int got_timeout_env;
329
330 extern void lttng_ring_buffer_client_overwrite_init(void);
331 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
332 extern void lttng_ring_buffer_client_discard_init(void);
333 extern void lttng_ring_buffer_client_discard_rt_init(void);
334 extern void lttng_ring_buffer_metadata_client_init(void);
335 extern void lttng_ring_buffer_client_overwrite_exit(void);
336 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
337 extern void lttng_ring_buffer_client_discard_exit(void);
338 extern void lttng_ring_buffer_client_discard_rt_exit(void);
339 extern void lttng_ring_buffer_metadata_client_exit(void);
340
341 static char *get_map_shm(struct sock_info *sock_info);
342
343 ssize_t lttng_ust_read(int fd, void *buf, size_t len)
344 {
345 ssize_t ret;
346 size_t copied = 0, to_copy = len;
347
348 do {
349 ret = read(fd, buf + copied, to_copy);
350 if (ret > 0) {
351 copied += ret;
352 to_copy -= ret;
353 }
354 } while ((ret > 0 && to_copy > 0)
355 || (ret < 0 && errno == EINTR));
356 if (ret > 0) {
357 ret = copied;
358 }
359 return ret;
360 }
361 /*
362 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
363 * pointer.
364 */
365 static
366 const char *get_lttng_home_dir(void)
367 {
368 const char *val;
369
370 val = (const char *) lttng_getenv("LTTNG_HOME");
371 if (val != NULL) {
372 return val;
373 }
374 return (const char *) lttng_getenv("HOME");
375 }
376
377 /*
378 * Force a read (imply TLS fixup for dlopen) of TLS variables.
379 */
380 static
381 void lttng_fixup_nest_count_tls(void)
382 {
383 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
384 }
385
386 static
387 void lttng_fixup_ust_mutex_nest_tls(void)
388 {
389 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
390 }
391
392 /*
393 * Fixup urcu bp TLS.
394 */
395 static
396 void lttng_fixup_urcu_bp_tls(void)
397 {
398 rcu_read_lock();
399 rcu_read_unlock();
400 }
401
402 void lttng_ust_fixup_tls(void)
403 {
404 lttng_fixup_urcu_bp_tls();
405 lttng_fixup_ringbuffer_tls();
406 lttng_fixup_vtid_tls();
407 lttng_fixup_nest_count_tls();
408 lttng_fixup_procname_tls();
409 lttng_fixup_ust_mutex_nest_tls();
410 lttng_ust_fixup_fd_tracker_tls();
411 }
412
413 int lttng_get_notify_socket(void *owner)
414 {
415 struct sock_info *info = owner;
416
417 return info->notify_socket;
418 }
419
420 static
421 void print_cmd(int cmd, int handle)
422 {
423 const char *cmd_name = "Unknown";
424
425 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
426 && cmd_name_mapping[cmd]) {
427 cmd_name = cmd_name_mapping[cmd];
428 }
429 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
430 cmd_name, cmd,
431 lttng_ust_obj_get_name(handle), handle);
432 }
433
434 static
435 int setup_global_apps(void)
436 {
437 int ret = 0;
438 assert(!global_apps.wait_shm_mmap);
439
440 global_apps.wait_shm_mmap = get_map_shm(&global_apps);
441 if (!global_apps.wait_shm_mmap) {
442 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
443 global_apps.allowed = 0;
444 ret = -EIO;
445 goto error;
446 }
447
448 global_apps.allowed = 1;
449 error:
450 return ret;
451 }
452 static
453 int setup_local_apps(void)
454 {
455 int ret = 0;
456 const char *home_dir;
457 uid_t uid;
458
459 assert(!local_apps.wait_shm_mmap);
460
461 uid = getuid();
462 /*
463 * Disallow per-user tracing for setuid binaries.
464 */
465 if (uid != geteuid()) {
466 assert(local_apps.allowed == 0);
467 ret = 0;
468 goto end;
469 }
470 home_dir = get_lttng_home_dir();
471 if (!home_dir) {
472 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
473 assert(local_apps.allowed == 0);
474 ret = -ENOENT;
475 goto end;
476 }
477 local_apps.allowed = 1;
478 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
479 home_dir,
480 LTTNG_DEFAULT_HOME_RUNDIR,
481 LTTNG_UST_SOCK_FILENAME);
482 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
483 LTTNG_UST_WAIT_FILENAME,
484 uid);
485
486 local_apps.wait_shm_mmap = get_map_shm(&local_apps);
487 if (!local_apps.wait_shm_mmap) {
488 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
489 local_apps.allowed = 0;
490 ret = -EIO;
491 goto end;
492 }
493 end:
494 return ret;
495 }
496
497 /*
498 * Get socket timeout, in ms.
499 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
500 */
501 static
502 long get_timeout(void)
503 {
504 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
505
506 if (!got_timeout_env) {
507 str_timeout = lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
508 got_timeout_env = 1;
509 }
510 if (str_timeout)
511 constructor_delay_ms = strtol(str_timeout, NULL, 10);
512 /* All negative values are considered as "-1". */
513 if (constructor_delay_ms < -1)
514 constructor_delay_ms = -1;
515 return constructor_delay_ms;
516 }
517
518 /* Timeout for notify socket send and recv. */
519 static
520 long get_notify_sock_timeout(void)
521 {
522 return get_timeout();
523 }
524
525 /* Timeout for connecting to cmd and notify sockets. */
526 static
527 long get_connect_sock_timeout(void)
528 {
529 return get_timeout();
530 }
531
532 /*
533 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
534 */
535 static
536 int get_constructor_timeout(struct timespec *constructor_timeout)
537 {
538 long constructor_delay_ms;
539 int ret;
540
541 constructor_delay_ms = get_timeout();
542
543 switch (constructor_delay_ms) {
544 case -1:/* fall-through */
545 case 0:
546 return constructor_delay_ms;
547 default:
548 break;
549 }
550
551 /*
552 * If we are unable to find the current time, don't wait.
553 */
554 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
555 if (ret) {
556 /* Don't wait. */
557 return 0;
558 }
559 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
560 constructor_timeout->tv_nsec +=
561 (constructor_delay_ms % 1000UL) * 1000000UL;
562 if (constructor_timeout->tv_nsec >= 1000000000UL) {
563 constructor_timeout->tv_sec++;
564 constructor_timeout->tv_nsec -= 1000000000UL;
565 }
566 /* Timeout wait (constructor_delay_ms). */
567 return 1;
568 }
569
570 static
571 int register_to_sessiond(int socket, enum ustctl_socket_type type)
572 {
573 return ustcomm_send_reg_msg(socket,
574 type,
575 CAA_BITS_PER_LONG,
576 lttng_alignof(uint8_t) * CHAR_BIT,
577 lttng_alignof(uint16_t) * CHAR_BIT,
578 lttng_alignof(uint32_t) * CHAR_BIT,
579 lttng_alignof(uint64_t) * CHAR_BIT,
580 lttng_alignof(unsigned long) * CHAR_BIT);
581 }
582
583 static
584 int send_reply(int sock, struct ustcomm_ust_reply *lur)
585 {
586 ssize_t len;
587
588 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
589 switch (len) {
590 case sizeof(*lur):
591 DBG("message successfully sent");
592 return 0;
593 default:
594 if (len == -ECONNRESET) {
595 DBG("remote end closed connection");
596 return 0;
597 }
598 if (len < 0)
599 return len;
600 DBG("incorrect message size: %zd", len);
601 return -EINVAL;
602 }
603 }
604
605 static
606 int handle_register_done(struct sock_info *sock_info)
607 {
608 int ret;
609
610 if (sock_info->constructor_sem_posted)
611 return 0;
612 sock_info->constructor_sem_posted = 1;
613 if (uatomic_read(&sem_count) <= 0) {
614 return 0;
615 }
616 ret = uatomic_add_return(&sem_count, -1);
617 if (ret == 0) {
618 ret = sem_post(&constructor_wait);
619 assert(!ret);
620 }
621 return 0;
622 }
623
624 /*
625 * Only execute pending statedump after the constructor semaphore has
626 * been posted by each listener thread. This means statedump will only
627 * be performed after the "registration done" command is received from
628 * each session daemon the application is connected to.
629 *
630 * This ensures we don't run into deadlock issues with the dynamic
631 * loader mutex, which is held while the constructor is called and
632 * waiting on the constructor semaphore. All operations requiring this
633 * dynamic loader lock need to be postponed using this mechanism.
634 */
635 static
636 void handle_pending_statedump(struct sock_info *sock_info)
637 {
638 int ctor_passed = sock_info->constructor_sem_posted;
639
640 if (ctor_passed && sock_info->statedump_pending) {
641 sock_info->statedump_pending = 0;
642 pthread_mutex_lock(&ust_fork_mutex);
643 lttng_handle_pending_statedump(sock_info);
644 pthread_mutex_unlock(&ust_fork_mutex);
645 }
646 }
647
648 static
649 int handle_message(struct sock_info *sock_info,
650 int sock, struct ustcomm_ust_msg *lum)
651 {
652 int ret = 0;
653 const struct lttng_ust_objd_ops *ops;
654 struct ustcomm_ust_reply lur;
655 union ust_args args;
656 char ctxstr[LTTNG_UST_SYM_NAME_LEN]; /* App context string. */
657 ssize_t len;
658
659 memset(&lur, 0, sizeof(lur));
660
661 if (ust_lock()) {
662 ret = -LTTNG_UST_ERR_EXITING;
663 goto error;
664 }
665
666 ops = objd_ops(lum->handle);
667 if (!ops) {
668 ret = -ENOENT;
669 goto error;
670 }
671
672 switch (lum->cmd) {
673 case LTTNG_UST_REGISTER_DONE:
674 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
675 ret = handle_register_done(sock_info);
676 else
677 ret = -EINVAL;
678 break;
679 case LTTNG_UST_RELEASE:
680 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
681 ret = -EPERM;
682 else
683 ret = lttng_ust_objd_unref(lum->handle, 1);
684 break;
685 case LTTNG_UST_FILTER:
686 {
687 /* Receive filter data */
688 struct lttng_ust_filter_bytecode_node *bytecode;
689
690 if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
691 ERR("Filter data size is too large: %u bytes",
692 lum->u.filter.data_size);
693 ret = -EINVAL;
694 goto error;
695 }
696
697 if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
698 ERR("Filter reloc offset %u is not within data",
699 lum->u.filter.reloc_offset);
700 ret = -EINVAL;
701 goto error;
702 }
703
704 bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
705 if (!bytecode) {
706 ret = -ENOMEM;
707 goto error;
708 }
709 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
710 lum->u.filter.data_size);
711 switch (len) {
712 case 0: /* orderly shutdown */
713 ret = 0;
714 free(bytecode);
715 goto error;
716 default:
717 if (len == lum->u.filter.data_size) {
718 DBG("filter data received");
719 break;
720 } else if (len < 0) {
721 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
722 if (len == -ECONNRESET) {
723 ERR("%s remote end closed connection", sock_info->name);
724 ret = len;
725 free(bytecode);
726 goto error;
727 }
728 ret = len;
729 free(bytecode);
730 goto error;
731 } else {
732 DBG("incorrect filter data message size: %zd", len);
733 ret = -EINVAL;
734 free(bytecode);
735 goto error;
736 }
737 }
738 bytecode->bc.len = lum->u.filter.data_size;
739 bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
740 bytecode->bc.seqnum = lum->u.filter.seqnum;
741 if (ops->cmd) {
742 ret = ops->cmd(lum->handle, lum->cmd,
743 (unsigned long) bytecode,
744 &args, sock_info);
745 if (ret) {
746 free(bytecode);
747 }
748 /* don't free bytecode if everything went fine. */
749 } else {
750 ret = -ENOSYS;
751 free(bytecode);
752 }
753 break;
754 }
755 case LTTNG_UST_EXCLUSION:
756 {
757 /* Receive exclusion names */
758 struct lttng_ust_excluder_node *node;
759 unsigned int count;
760
761 count = lum->u.exclusion.count;
762 if (count == 0) {
763 /* There are no names to read */
764 ret = 0;
765 goto error;
766 }
767 node = zmalloc(sizeof(*node) +
768 count * LTTNG_UST_SYM_NAME_LEN);
769 if (!node) {
770 ret = -ENOMEM;
771 goto error;
772 }
773 node->excluder.count = count;
774 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
775 count * LTTNG_UST_SYM_NAME_LEN);
776 switch (len) {
777 case 0: /* orderly shutdown */
778 ret = 0;
779 free(node);
780 goto error;
781 default:
782 if (len == count * LTTNG_UST_SYM_NAME_LEN) {
783 DBG("Exclusion data received");
784 break;
785 } else if (len < 0) {
786 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
787 if (len == -ECONNRESET) {
788 ERR("%s remote end closed connection", sock_info->name);
789 ret = len;
790 free(node);
791 goto error;
792 }
793 ret = len;
794 free(node);
795 goto error;
796 } else {
797 DBG("Incorrect exclusion data message size: %zd", len);
798 ret = -EINVAL;
799 free(node);
800 goto error;
801 }
802 }
803 if (ops->cmd) {
804 ret = ops->cmd(lum->handle, lum->cmd,
805 (unsigned long) node,
806 &args, sock_info);
807 if (ret) {
808 free(node);
809 }
810 /* Don't free exclusion data if everything went fine. */
811 } else {
812 ret = -ENOSYS;
813 free(node);
814 }
815 break;
816 }
817 case LTTNG_UST_CHANNEL:
818 {
819 void *chan_data;
820 int wakeup_fd;
821
822 len = ustcomm_recv_channel_from_sessiond(sock,
823 &chan_data, lum->u.channel.len,
824 &wakeup_fd);
825 switch (len) {
826 case 0: /* orderly shutdown */
827 ret = 0;
828 goto error;
829 default:
830 if (len == lum->u.channel.len) {
831 DBG("channel data received");
832 break;
833 } else if (len < 0) {
834 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
835 if (len == -ECONNRESET) {
836 ERR("%s remote end closed connection", sock_info->name);
837 ret = len;
838 goto error;
839 }
840 ret = len;
841 goto error;
842 } else {
843 DBG("incorrect channel data message size: %zd", len);
844 ret = -EINVAL;
845 goto error;
846 }
847 }
848 args.channel.chan_data = chan_data;
849 args.channel.wakeup_fd = wakeup_fd;
850 if (ops->cmd)
851 ret = ops->cmd(lum->handle, lum->cmd,
852 (unsigned long) &lum->u,
853 &args, sock_info);
854 else
855 ret = -ENOSYS;
856 break;
857 }
858 case LTTNG_UST_STREAM:
859 {
860 /* Receive shm_fd, wakeup_fd */
861 ret = ustcomm_recv_stream_from_sessiond(sock,
862 NULL,
863 &args.stream.shm_fd,
864 &args.stream.wakeup_fd);
865 if (ret) {
866 goto error;
867 }
868
869 if (ops->cmd)
870 ret = ops->cmd(lum->handle, lum->cmd,
871 (unsigned long) &lum->u,
872 &args, sock_info);
873 else
874 ret = -ENOSYS;
875 break;
876 }
877 case LTTNG_UST_CONTEXT:
878 switch (lum->u.context.ctx) {
879 case LTTNG_UST_CONTEXT_APP_CONTEXT:
880 {
881 char *p;
882 size_t ctxlen, recvlen;
883
884 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
885 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
886 if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
887 ERR("Application context string length size is too large: %zu bytes",
888 ctxlen);
889 ret = -EINVAL;
890 goto error;
891 }
892 strcpy(ctxstr, "$app.");
893 p = &ctxstr[strlen("$app.")];
894 recvlen = ctxlen - strlen("$app.");
895 len = ustcomm_recv_unix_sock(sock, p, recvlen);
896 switch (len) {
897 case 0: /* orderly shutdown */
898 ret = 0;
899 goto error;
900 default:
901 if (len == recvlen) {
902 DBG("app context data received");
903 break;
904 } else if (len < 0) {
905 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
906 if (len == -ECONNRESET) {
907 ERR("%s remote end closed connection", sock_info->name);
908 ret = len;
909 goto error;
910 }
911 ret = len;
912 goto error;
913 } else {
914 DBG("incorrect app context data message size: %zd", len);
915 ret = -EINVAL;
916 goto error;
917 }
918 }
919 /* Put : between provider and ctxname. */
920 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
921 args.app_context.ctxname = ctxstr;
922 break;
923 }
924 default:
925 break;
926 }
927 if (ops->cmd) {
928 ret = ops->cmd(lum->handle, lum->cmd,
929 (unsigned long) &lum->u,
930 &args, sock_info);
931 } else {
932 ret = -ENOSYS;
933 }
934 break;
935 default:
936 if (ops->cmd)
937 ret = ops->cmd(lum->handle, lum->cmd,
938 (unsigned long) &lum->u,
939 &args, sock_info);
940 else
941 ret = -ENOSYS;
942 break;
943 }
944
945 lur.handle = lum->handle;
946 lur.cmd = lum->cmd;
947 lur.ret_val = ret;
948 if (ret >= 0) {
949 lur.ret_code = LTTNG_UST_OK;
950 } else {
951 /*
952 * Use -LTTNG_UST_ERR as wildcard for UST internal
953 * error that are not caused by the transport, except if
954 * we already have a more precise error message to
955 * report.
956 */
957 if (ret > -LTTNG_UST_ERR) {
958 /* Translate code to UST error. */
959 switch (ret) {
960 case -EEXIST:
961 lur.ret_code = -LTTNG_UST_ERR_EXIST;
962 break;
963 case -EINVAL:
964 lur.ret_code = -LTTNG_UST_ERR_INVAL;
965 break;
966 case -ENOENT:
967 lur.ret_code = -LTTNG_UST_ERR_NOENT;
968 break;
969 case -EPERM:
970 lur.ret_code = -LTTNG_UST_ERR_PERM;
971 break;
972 case -ENOSYS:
973 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
974 break;
975 default:
976 lur.ret_code = -LTTNG_UST_ERR;
977 break;
978 }
979 } else {
980 lur.ret_code = ret;
981 }
982 }
983 if (ret >= 0) {
984 switch (lum->cmd) {
985 case LTTNG_UST_TRACER_VERSION:
986 lur.u.version = lum->u.version;
987 break;
988 case LTTNG_UST_TRACEPOINT_LIST_GET:
989 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
990 break;
991 }
992 }
993 DBG("Return value: %d", lur.ret_val);
994
995 ust_unlock();
996
997 /*
998 * Performed delayed statedump operations outside of the UST
999 * lock. We need to take the dynamic loader lock before we take
1000 * the UST lock internally within handle_pending_statedump().
1001 */
1002 handle_pending_statedump(sock_info);
1003
1004 if (ust_lock()) {
1005 ret = -LTTNG_UST_ERR_EXITING;
1006 goto error;
1007 }
1008
1009 ret = send_reply(sock, &lur);
1010 if (ret < 0) {
1011 DBG("error sending reply");
1012 goto error;
1013 }
1014
1015 /*
1016 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1017 * after the reply.
1018 */
1019 if (lur.ret_code == LTTNG_UST_OK) {
1020 switch (lum->cmd) {
1021 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
1022 len = ustcomm_send_unix_sock(sock,
1023 &args.field_list.entry,
1024 sizeof(args.field_list.entry));
1025 if (len < 0) {
1026 ret = len;
1027 goto error;
1028 }
1029 if (len != sizeof(args.field_list.entry)) {
1030 ret = -EINVAL;
1031 goto error;
1032 }
1033 }
1034 }
1035
1036 error:
1037 ust_unlock();
1038
1039 return ret;
1040 }
1041
1042 static
1043 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1044 {
1045 int ret;
1046
1047 if (sock_info->root_handle != -1) {
1048 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
1049 if (ret) {
1050 ERR("Error unref root handle");
1051 }
1052 sock_info->root_handle = -1;
1053 }
1054 sock_info->constructor_sem_posted = 0;
1055
1056 /*
1057 * wait_shm_mmap, socket and notify socket are used by listener
1058 * threads outside of the ust lock, so we cannot tear them down
1059 * ourselves, because we cannot join on these threads. Leave
1060 * responsibility of cleaning up these resources to the OS
1061 * process exit.
1062 */
1063 if (exiting)
1064 return;
1065
1066 if (sock_info->socket != -1) {
1067 ret = ustcomm_close_unix_sock(sock_info->socket);
1068 if (ret) {
1069 ERR("Error closing ust cmd socket");
1070 }
1071 sock_info->socket = -1;
1072 }
1073 if (sock_info->notify_socket != -1) {
1074 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1075 if (ret) {
1076 ERR("Error closing ust notify socket");
1077 }
1078 sock_info->notify_socket = -1;
1079 }
1080 if (sock_info->wait_shm_mmap) {
1081 long page_size;
1082
1083 page_size = sysconf(_SC_PAGE_SIZE);
1084 if (page_size <= 0) {
1085 if (!page_size) {
1086 errno = EINVAL;
1087 }
1088 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1089 } else {
1090 ret = munmap(sock_info->wait_shm_mmap, page_size);
1091 if (ret) {
1092 ERR("Error unmapping wait shm");
1093 }
1094 }
1095 sock_info->wait_shm_mmap = NULL;
1096 }
1097 }
1098
1099 /*
1100 * Using fork to set umask in the child process (not multi-thread safe).
1101 * We deal with the shm_open vs ftruncate race (happening when the
1102 * sessiond owns the shm and does not let everybody modify it, to ensure
1103 * safety against shm_unlink) by simply letting the mmap fail and
1104 * retrying after a few seconds.
1105 * For global shm, everybody has rw access to it until the sessiond
1106 * starts.
1107 */
1108 static
1109 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1110 {
1111 int wait_shm_fd, ret;
1112 pid_t pid;
1113
1114 /*
1115 * Try to open read-only.
1116 */
1117 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1118 if (wait_shm_fd >= 0) {
1119 int32_t tmp_read;
1120 ssize_t len;
1121 size_t bytes_read = 0;
1122
1123 /*
1124 * Try to read the fd. If unable to do so, try opening
1125 * it in write mode.
1126 */
1127 do {
1128 len = read(wait_shm_fd,
1129 &((char *) &tmp_read)[bytes_read],
1130 sizeof(tmp_read) - bytes_read);
1131 if (len > 0) {
1132 bytes_read += len;
1133 }
1134 } while ((len < 0 && errno == EINTR)
1135 || (len > 0 && bytes_read < sizeof(tmp_read)));
1136 if (bytes_read != sizeof(tmp_read)) {
1137 ret = close(wait_shm_fd);
1138 if (ret) {
1139 ERR("close wait_shm_fd");
1140 }
1141 goto open_write;
1142 }
1143 goto end;
1144 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1145 /*
1146 * Real-only open did not work, and it's not because the
1147 * entry was not present. It's a failure that prohibits
1148 * using shm.
1149 */
1150 ERR("Error opening shm %s", sock_info->wait_shm_path);
1151 goto end;
1152 }
1153
1154 open_write:
1155 /*
1156 * If the open failed because the file did not exist, or because
1157 * the file was not truncated yet, try creating it ourself.
1158 */
1159 URCU_TLS(lttng_ust_nest_count)++;
1160 pid = fork();
1161 URCU_TLS(lttng_ust_nest_count)--;
1162 if (pid > 0) {
1163 int status;
1164
1165 /*
1166 * Parent: wait for child to return, in which case the
1167 * shared memory map will have been created.
1168 */
1169 pid = wait(&status);
1170 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1171 wait_shm_fd = -1;
1172 goto end;
1173 }
1174 /*
1175 * Try to open read-only again after creation.
1176 */
1177 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1178 if (wait_shm_fd < 0) {
1179 /*
1180 * Real-only open did not work. It's a failure
1181 * that prohibits using shm.
1182 */
1183 ERR("Error opening shm %s", sock_info->wait_shm_path);
1184 goto end;
1185 }
1186 goto end;
1187 } else if (pid == 0) {
1188 int create_mode;
1189
1190 /* Child */
1191 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1192 if (sock_info->global)
1193 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1194 /*
1195 * We're alone in a child process, so we can modify the
1196 * process-wide umask.
1197 */
1198 umask(~create_mode);
1199 /*
1200 * Try creating shm (or get rw access).
1201 * We don't do an exclusive open, because we allow other
1202 * processes to create+ftruncate it concurrently.
1203 */
1204 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1205 O_RDWR | O_CREAT, create_mode);
1206 if (wait_shm_fd >= 0) {
1207 ret = ftruncate(wait_shm_fd, mmap_size);
1208 if (ret) {
1209 PERROR("ftruncate");
1210 _exit(EXIT_FAILURE);
1211 }
1212 _exit(EXIT_SUCCESS);
1213 }
1214 /*
1215 * For local shm, we need to have rw access to accept
1216 * opening it: this means the local sessiond will be
1217 * able to wake us up. For global shm, we open it even
1218 * if rw access is not granted, because the root.root
1219 * sessiond will be able to override all rights and wake
1220 * us up.
1221 */
1222 if (!sock_info->global && errno != EACCES) {
1223 ERR("Error opening shm %s", sock_info->wait_shm_path);
1224 _exit(EXIT_FAILURE);
1225 }
1226 /*
1227 * The shm exists, but we cannot open it RW. Report
1228 * success.
1229 */
1230 _exit(EXIT_SUCCESS);
1231 } else {
1232 return -1;
1233 }
1234 end:
1235 if (wait_shm_fd >= 0 && !sock_info->global) {
1236 struct stat statbuf;
1237
1238 /*
1239 * Ensure that our user is the owner of the shm file for
1240 * local shm. If we do not own the file, it means our
1241 * sessiond will not have access to wake us up (there is
1242 * probably a rogue process trying to fake our
1243 * sessiond). Fallback to polling method in this case.
1244 */
1245 ret = fstat(wait_shm_fd, &statbuf);
1246 if (ret) {
1247 PERROR("fstat");
1248 goto error_close;
1249 }
1250 if (statbuf.st_uid != getuid())
1251 goto error_close;
1252 }
1253 return wait_shm_fd;
1254
1255 error_close:
1256 ret = close(wait_shm_fd);
1257 if (ret) {
1258 PERROR("Error closing fd");
1259 }
1260 return -1;
1261 }
1262
1263 static
1264 char *get_map_shm(struct sock_info *sock_info)
1265 {
1266 long page_size;
1267 int wait_shm_fd, ret;
1268 char *wait_shm_mmap;
1269
1270 page_size = sysconf(_SC_PAGE_SIZE);
1271 if (page_size <= 0) {
1272 if (!page_size) {
1273 errno = EINVAL;
1274 }
1275 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1276 goto error;
1277 }
1278
1279 lttng_ust_lock_fd_tracker();
1280 wait_shm_fd = get_wait_shm(sock_info, page_size);
1281 if (wait_shm_fd < 0) {
1282 lttng_ust_unlock_fd_tracker();
1283 goto error;
1284 }
1285
1286 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1287 if (ret < 0) {
1288 ret = close(wait_shm_fd);
1289 if (!ret) {
1290 PERROR("Error closing fd");
1291 }
1292 lttng_ust_unlock_fd_tracker();
1293 goto error;
1294 }
1295
1296 wait_shm_fd = ret;
1297 lttng_ust_unlock_fd_tracker();
1298
1299 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1300 MAP_SHARED, wait_shm_fd, 0);
1301
1302 /* close shm fd immediately after taking the mmap reference */
1303 lttng_ust_lock_fd_tracker();
1304 ret = close(wait_shm_fd);
1305 if (!ret) {
1306 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1307 } else {
1308 PERROR("Error closing fd");
1309 }
1310 lttng_ust_unlock_fd_tracker();
1311
1312 if (wait_shm_mmap == MAP_FAILED) {
1313 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1314 goto error;
1315 }
1316 return wait_shm_mmap;
1317
1318 error:
1319 return NULL;
1320 }
1321
1322 static
1323 void wait_for_sessiond(struct sock_info *sock_info)
1324 {
1325 /* Use ust_lock to check if we should quit. */
1326 if (ust_lock()) {
1327 goto quit;
1328 }
1329 if (wait_poll_fallback) {
1330 goto error;
1331 }
1332 ust_unlock();
1333
1334 assert(sock_info->wait_shm_mmap);
1335
1336 DBG("Waiting for %s apps sessiond", sock_info->name);
1337 /* Wait for futex wakeup */
1338 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1339 goto end_wait;
1340
1341 while (futex_async((int32_t *) sock_info->wait_shm_mmap,
1342 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1343 switch (errno) {
1344 case EWOULDBLOCK:
1345 /* Value already changed. */
1346 goto end_wait;
1347 case EINTR:
1348 /* Retry if interrupted by signal. */
1349 break; /* Get out of switch. */
1350 case EFAULT:
1351 wait_poll_fallback = 1;
1352 DBG(
1353 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1354 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1355 "Please upgrade your kernel "
1356 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1357 "mainline). LTTng-UST will use polling mode fallback.");
1358 if (ust_debug())
1359 PERROR("futex");
1360 goto end_wait;
1361 }
1362 }
1363 end_wait:
1364 return;
1365
1366 quit:
1367 ust_unlock();
1368 return;
1369
1370 error:
1371 ust_unlock();
1372 return;
1373 }
1374
1375 /*
1376 * This thread does not allocate any resource, except within
1377 * handle_message, within mutex protection. This mutex protects against
1378 * fork and exit.
1379 * The other moment it allocates resources is at socket connection, which
1380 * is also protected by the mutex.
1381 */
1382 static
1383 void *ust_listener_thread(void *arg)
1384 {
1385 struct sock_info *sock_info = arg;
1386 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1387 long timeout;
1388
1389 lttng_ust_fixup_tls();
1390 /*
1391 * If available, add '-ust' to the end of this thread's
1392 * process name
1393 */
1394 ret = lttng_ust_setustprocname();
1395 if (ret) {
1396 ERR("Unable to set UST process name");
1397 }
1398
1399 /* Restart trying to connect to the session daemon */
1400 restart:
1401 if (prev_connect_failed) {
1402 /* Wait for sessiond availability with pipe */
1403 wait_for_sessiond(sock_info);
1404 if (has_waited) {
1405 has_waited = 0;
1406 /*
1407 * Sleep for 5 seconds before retrying after a
1408 * sequence of failure / wait / failure. This
1409 * deals with a killed or broken session daemon.
1410 */
1411 sleep(5);
1412 } else {
1413 has_waited = 1;
1414 }
1415 prev_connect_failed = 0;
1416 }
1417
1418 if (ust_lock()) {
1419 goto quit;
1420 }
1421
1422 if (sock_info->socket != -1) {
1423 /* FD tracker is updated by ustcomm_close_unix_sock() */
1424 ret = ustcomm_close_unix_sock(sock_info->socket);
1425 if (ret) {
1426 ERR("Error closing %s ust cmd socket",
1427 sock_info->name);
1428 }
1429 sock_info->socket = -1;
1430 }
1431 if (sock_info->notify_socket != -1) {
1432 /* FD tracker is updated by ustcomm_close_unix_sock() */
1433 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1434 if (ret) {
1435 ERR("Error closing %s ust notify socket",
1436 sock_info->name);
1437 }
1438 sock_info->notify_socket = -1;
1439 }
1440
1441
1442 /*
1443 * Register. We need to perform both connect and sending
1444 * registration message before doing the next connect otherwise
1445 * we may reach unix socket connect queue max limits and block
1446 * on the 2nd connect while the session daemon is awaiting the
1447 * first connect registration message.
1448 */
1449 /* Connect cmd socket */
1450 lttng_ust_lock_fd_tracker();
1451 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1452 get_connect_sock_timeout());
1453 if (ret < 0) {
1454 lttng_ust_unlock_fd_tracker();
1455 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1456 prev_connect_failed = 1;
1457
1458 /*
1459 * If we cannot find the sessiond daemon, don't delay
1460 * constructor execution.
1461 */
1462 ret = handle_register_done(sock_info);
1463 assert(!ret);
1464 ust_unlock();
1465 goto restart;
1466 }
1467 fd = ret;
1468 ret = lttng_ust_add_fd_to_tracker(fd);
1469 if (ret < 0) {
1470 ret = close(fd);
1471 if (ret) {
1472 PERROR("close on sock_info->socket");
1473 }
1474 ret = -1;
1475 lttng_ust_unlock_fd_tracker();
1476 ust_unlock();
1477 goto quit;
1478 }
1479
1480 sock_info->socket = ret;
1481 lttng_ust_unlock_fd_tracker();
1482
1483 ust_unlock();
1484 /*
1485 * Unlock/relock ust lock because connect is blocking (with
1486 * timeout). Don't delay constructors on the ust lock for too
1487 * long.
1488 */
1489 if (ust_lock()) {
1490 goto quit;
1491 }
1492
1493 /*
1494 * Create only one root handle per listener thread for the whole
1495 * process lifetime, so we ensure we get ID which is statically
1496 * assigned to the root handle.
1497 */
1498 if (sock_info->root_handle == -1) {
1499 ret = lttng_abi_create_root_handle();
1500 if (ret < 0) {
1501 ERR("Error creating root handle");
1502 goto quit;
1503 }
1504 sock_info->root_handle = ret;
1505 }
1506
1507 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1508 if (ret < 0) {
1509 ERR("Error registering to %s ust cmd socket",
1510 sock_info->name);
1511 prev_connect_failed = 1;
1512 /*
1513 * If we cannot register to the sessiond daemon, don't
1514 * delay constructor execution.
1515 */
1516 ret = handle_register_done(sock_info);
1517 assert(!ret);
1518 ust_unlock();
1519 goto restart;
1520 }
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 /* Connect notify socket */
1533 lttng_ust_lock_fd_tracker();
1534 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1535 get_connect_sock_timeout());
1536 if (ret < 0) {
1537 lttng_ust_unlock_fd_tracker();
1538 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1539 prev_connect_failed = 1;
1540
1541 /*
1542 * If we cannot find the sessiond daemon, don't delay
1543 * constructor execution.
1544 */
1545 ret = handle_register_done(sock_info);
1546 assert(!ret);
1547 ust_unlock();
1548 goto restart;
1549 }
1550
1551 fd = ret;
1552 ret = lttng_ust_add_fd_to_tracker(fd);
1553 if (ret < 0) {
1554 ret = close(fd);
1555 if (ret) {
1556 PERROR("close on sock_info->notify_socket");
1557 }
1558 ret = -1;
1559 lttng_ust_unlock_fd_tracker();
1560 ust_unlock();
1561 goto quit;
1562 }
1563
1564 sock_info->notify_socket = ret;
1565 lttng_ust_unlock_fd_tracker();
1566
1567 ust_unlock();
1568 /*
1569 * Unlock/relock ust lock because connect is blocking (with
1570 * timeout). Don't delay constructors on the ust lock for too
1571 * long.
1572 */
1573 if (ust_lock()) {
1574 goto quit;
1575 }
1576
1577 timeout = get_notify_sock_timeout();
1578 if (timeout >= 0) {
1579 /*
1580 * Give at least 10ms to sessiond to reply to
1581 * notifications.
1582 */
1583 if (timeout < 10)
1584 timeout = 10;
1585 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1586 timeout);
1587 if (ret < 0) {
1588 WARN("Error setting socket receive timeout");
1589 }
1590 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1591 timeout);
1592 if (ret < 0) {
1593 WARN("Error setting socket send timeout");
1594 }
1595 } else if (timeout < -1) {
1596 WARN("Unsupported timeout value %ld", timeout);
1597 }
1598
1599 ret = register_to_sessiond(sock_info->notify_socket,
1600 USTCTL_SOCKET_NOTIFY);
1601 if (ret < 0) {
1602 ERR("Error registering to %s ust notify socket",
1603 sock_info->name);
1604 prev_connect_failed = 1;
1605 /*
1606 * If we cannot register to the sessiond daemon, don't
1607 * delay constructor execution.
1608 */
1609 ret = handle_register_done(sock_info);
1610 assert(!ret);
1611 ust_unlock();
1612 goto restart;
1613 }
1614 sock = sock_info->socket;
1615
1616 ust_unlock();
1617
1618 for (;;) {
1619 ssize_t len;
1620 struct ustcomm_ust_msg lum;
1621
1622 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1623 switch (len) {
1624 case 0: /* orderly shutdown */
1625 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1626 if (ust_lock()) {
1627 goto quit;
1628 }
1629 /*
1630 * Either sessiond has shutdown or refused us by closing the socket.
1631 * In either case, we don't want to delay construction execution,
1632 * and we need to wait before retry.
1633 */
1634 prev_connect_failed = 1;
1635 /*
1636 * If we cannot register to the sessiond daemon, don't
1637 * delay constructor execution.
1638 */
1639 ret = handle_register_done(sock_info);
1640 assert(!ret);
1641 ust_unlock();
1642 goto end;
1643 case sizeof(lum):
1644 print_cmd(lum.cmd, lum.handle);
1645 ret = handle_message(sock_info, sock, &lum);
1646 if (ret) {
1647 ERR("Error handling message for %s socket",
1648 sock_info->name);
1649 /*
1650 * Close socket if protocol error is
1651 * detected.
1652 */
1653 goto end;
1654 }
1655 continue;
1656 default:
1657 if (len < 0) {
1658 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1659 } else {
1660 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1661 }
1662 if (len == -ECONNRESET) {
1663 DBG("%s remote end closed connection", sock_info->name);
1664 goto end;
1665 }
1666 goto end;
1667 }
1668
1669 }
1670 end:
1671 if (ust_lock()) {
1672 goto quit;
1673 }
1674 /* Cleanup socket handles before trying to reconnect */
1675 lttng_ust_objd_table_owner_cleanup(sock_info);
1676 ust_unlock();
1677 goto restart; /* try to reconnect */
1678
1679 quit:
1680 ust_unlock();
1681
1682 pthread_mutex_lock(&ust_exit_mutex);
1683 sock_info->thread_active = 0;
1684 pthread_mutex_unlock(&ust_exit_mutex);
1685 return NULL;
1686 }
1687
1688 /*
1689 * Weak symbol to call when the ust malloc wrapper is not loaded.
1690 */
1691 __attribute__((weak))
1692 void lttng_ust_malloc_wrapper_init(void)
1693 {
1694 }
1695
1696 /*
1697 * sessiond monitoring thread: monitor presence of global and per-user
1698 * sessiond by polling the application common named pipe.
1699 */
1700 void __attribute__((constructor)) lttng_ust_init(void)
1701 {
1702 struct timespec constructor_timeout;
1703 sigset_t sig_all_blocked, orig_parent_mask;
1704 pthread_attr_t thread_attr;
1705 int timeout_mode;
1706 int ret;
1707
1708 if (uatomic_xchg(&initialized, 1) == 1)
1709 return;
1710
1711 /*
1712 * Fixup interdependency between TLS fixup mutex (which happens
1713 * to be the dynamic linker mutex) and ust_lock, taken within
1714 * the ust lock.
1715 */
1716 lttng_ust_fixup_tls();
1717
1718 lttng_ust_loaded = 1;
1719
1720 /*
1721 * We want precise control over the order in which we construct
1722 * our sub-libraries vs starting to receive commands from
1723 * sessiond (otherwise leading to errors when trying to create
1724 * sessiond before the init functions are completed).
1725 */
1726 init_usterr();
1727 lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */
1728 init_tracepoint();
1729 lttng_ust_init_fd_tracker();
1730 lttng_ust_clock_init();
1731 lttng_ust_getcpu_init();
1732 lttng_ust_statedump_init();
1733 lttng_ring_buffer_metadata_client_init();
1734 lttng_ring_buffer_client_overwrite_init();
1735 lttng_ring_buffer_client_overwrite_rt_init();
1736 lttng_ring_buffer_client_discard_init();
1737 lttng_ring_buffer_client_discard_rt_init();
1738 lttng_perf_counter_init();
1739 /*
1740 * Invoke ust malloc wrapper init before starting other threads.
1741 */
1742 lttng_ust_malloc_wrapper_init();
1743
1744 timeout_mode = get_constructor_timeout(&constructor_timeout);
1745
1746 ret = sem_init(&constructor_wait, 0, 0);
1747 if (ret) {
1748 PERROR("sem_init");
1749 }
1750
1751 ret = setup_global_apps();
1752 if (ret) {
1753 assert(global_apps.allowed == 0);
1754 DBG("global apps setup returned %d", ret);
1755 }
1756
1757 ret = setup_local_apps();
1758 if (ret) {
1759 assert(local_apps.allowed == 0);
1760 DBG("local apps setup returned %d", ret);
1761 }
1762
1763 /* A new thread created by pthread_create inherits the signal mask
1764 * from the parent. To avoid any signal being received by the
1765 * listener thread, we block all signals temporarily in the parent,
1766 * while we create the listener thread.
1767 */
1768 sigfillset(&sig_all_blocked);
1769 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
1770 if (ret) {
1771 ERR("pthread_sigmask: %s", strerror(ret));
1772 }
1773
1774 ret = pthread_attr_init(&thread_attr);
1775 if (ret) {
1776 ERR("pthread_attr_init: %s", strerror(ret));
1777 }
1778 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
1779 if (ret) {
1780 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
1781 }
1782
1783 if (global_apps.allowed) {
1784 pthread_mutex_lock(&ust_exit_mutex);
1785 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
1786 ust_listener_thread, &global_apps);
1787 if (ret) {
1788 ERR("pthread_create global: %s", strerror(ret));
1789 }
1790 global_apps.thread_active = 1;
1791 pthread_mutex_unlock(&ust_exit_mutex);
1792 } else {
1793 handle_register_done(&global_apps);
1794 }
1795
1796 if (local_apps.allowed) {
1797 pthread_mutex_lock(&ust_exit_mutex);
1798 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
1799 ust_listener_thread, &local_apps);
1800 if (ret) {
1801 ERR("pthread_create local: %s", strerror(ret));
1802 }
1803 local_apps.thread_active = 1;
1804 pthread_mutex_unlock(&ust_exit_mutex);
1805 } else {
1806 handle_register_done(&local_apps);
1807 }
1808 ret = pthread_attr_destroy(&thread_attr);
1809 if (ret) {
1810 ERR("pthread_attr_destroy: %s", strerror(ret));
1811 }
1812
1813 /* Restore original signal mask in parent */
1814 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
1815 if (ret) {
1816 ERR("pthread_sigmask: %s", strerror(ret));
1817 }
1818
1819 switch (timeout_mode) {
1820 case 1: /* timeout wait */
1821 do {
1822 ret = sem_timedwait(&constructor_wait,
1823 &constructor_timeout);
1824 } while (ret < 0 && errno == EINTR);
1825 if (ret < 0) {
1826 switch (errno) {
1827 case ETIMEDOUT:
1828 ERR("Timed out waiting for lttng-sessiond");
1829 break;
1830 case EINVAL:
1831 PERROR("sem_timedwait");
1832 break;
1833 default:
1834 ERR("Unexpected error \"%s\" returned by sem_timedwait",
1835 strerror(errno));
1836 }
1837 }
1838 break;
1839 case -1:/* wait forever */
1840 do {
1841 ret = sem_wait(&constructor_wait);
1842 } while (ret < 0 && errno == EINTR);
1843 if (ret < 0) {
1844 switch (errno) {
1845 case EINVAL:
1846 PERROR("sem_wait");
1847 break;
1848 default:
1849 ERR("Unexpected error \"%s\" returned by sem_wait",
1850 strerror(errno));
1851 }
1852 }
1853 break;
1854 case 0: /* no timeout */
1855 break;
1856 }
1857 }
1858
1859 static
1860 void lttng_ust_cleanup(int exiting)
1861 {
1862 cleanup_sock_info(&global_apps, exiting);
1863 cleanup_sock_info(&local_apps, exiting);
1864 local_apps.allowed = 0;
1865 global_apps.allowed = 0;
1866 /*
1867 * The teardown in this function all affect data structures
1868 * accessed under the UST lock by the listener thread. This
1869 * lock, along with the lttng_ust_comm_should_quit flag, ensure
1870 * that none of these threads are accessing this data at this
1871 * point.
1872 */
1873 lttng_ust_abi_exit();
1874 lttng_ust_events_exit();
1875 lttng_perf_counter_exit();
1876 lttng_ring_buffer_client_discard_rt_exit();
1877 lttng_ring_buffer_client_discard_exit();
1878 lttng_ring_buffer_client_overwrite_rt_exit();
1879 lttng_ring_buffer_client_overwrite_exit();
1880 lttng_ring_buffer_metadata_client_exit();
1881 lttng_ust_statedump_destroy();
1882 exit_tracepoint();
1883 if (!exiting) {
1884 /* Reinitialize values for fork */
1885 sem_count = 2;
1886 lttng_ust_comm_should_quit = 0;
1887 initialized = 0;
1888 }
1889 }
1890
1891 void __attribute__((destructor)) lttng_ust_exit(void)
1892 {
1893 int ret;
1894
1895 /*
1896 * Using pthread_cancel here because:
1897 * A) we don't want to hang application teardown.
1898 * B) the thread is not allocating any resource.
1899 */
1900
1901 /*
1902 * Require the communication thread to quit. Synchronize with
1903 * mutexes to ensure it is not in a mutex critical section when
1904 * pthread_cancel is later called.
1905 */
1906 ust_lock_nocheck();
1907 lttng_ust_comm_should_quit = 1;
1908 ust_unlock();
1909
1910 pthread_mutex_lock(&ust_exit_mutex);
1911 /* cancel threads */
1912 if (global_apps.thread_active) {
1913 ret = pthread_cancel(global_apps.ust_listener);
1914 if (ret) {
1915 ERR("Error cancelling global ust listener thread: %s",
1916 strerror(ret));
1917 } else {
1918 global_apps.thread_active = 0;
1919 }
1920 }
1921 if (local_apps.thread_active) {
1922 ret = pthread_cancel(local_apps.ust_listener);
1923 if (ret) {
1924 ERR("Error cancelling local ust listener thread: %s",
1925 strerror(ret));
1926 } else {
1927 local_apps.thread_active = 0;
1928 }
1929 }
1930 pthread_mutex_unlock(&ust_exit_mutex);
1931
1932 /*
1933 * Do NOT join threads: use of sys_futex makes it impossible to
1934 * join the threads without using async-cancel, but async-cancel
1935 * is delivered by a signal, which could hit the target thread
1936 * anywhere in its code path, including while the ust_lock() is
1937 * held, causing a deadlock for the other thread. Let the OS
1938 * cleanup the threads if there are stalled in a syscall.
1939 */
1940 lttng_ust_cleanup(1);
1941 }
1942
1943 /*
1944 * We exclude the worker threads across fork and clone (except
1945 * CLONE_VM), because these system calls only keep the forking thread
1946 * running in the child. Therefore, we don't want to call fork or clone
1947 * in the middle of an tracepoint or ust tracing state modification.
1948 * Holding this mutex protects these structures across fork and clone.
1949 */
1950 void ust_before_fork(sigset_t *save_sigset)
1951 {
1952 /*
1953 * Disable signals. This is to avoid that the child intervenes
1954 * before it is properly setup for tracing. It is safer to
1955 * disable all signals, because then we know we are not breaking
1956 * anything by restoring the original mask.
1957 */
1958 sigset_t all_sigs;
1959 int ret;
1960
1961 /* Fixup lttng-ust TLS. */
1962 lttng_ust_fixup_tls();
1963
1964 if (URCU_TLS(lttng_ust_nest_count))
1965 return;
1966 /* Disable signals */
1967 sigfillset(&all_sigs);
1968 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
1969 if (ret == -1) {
1970 PERROR("sigprocmask");
1971 }
1972
1973 pthread_mutex_lock(&ust_fork_mutex);
1974
1975 ust_lock_nocheck();
1976 rcu_bp_before_fork();
1977 }
1978
1979 static void ust_after_fork_common(sigset_t *restore_sigset)
1980 {
1981 int ret;
1982
1983 DBG("process %d", getpid());
1984 ust_unlock();
1985
1986 pthread_mutex_unlock(&ust_fork_mutex);
1987
1988 /* Restore signals */
1989 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
1990 if (ret == -1) {
1991 PERROR("sigprocmask");
1992 }
1993 }
1994
1995 void ust_after_fork_parent(sigset_t *restore_sigset)
1996 {
1997 if (URCU_TLS(lttng_ust_nest_count))
1998 return;
1999 DBG("process %d", getpid());
2000 rcu_bp_after_fork_parent();
2001 /* Release mutexes and reenable signals */
2002 ust_after_fork_common(restore_sigset);
2003 }
2004
2005 /*
2006 * After fork, in the child, we need to cleanup all the leftover state,
2007 * except the worker thread which already magically disappeared thanks
2008 * to the weird Linux fork semantics. After tyding up, we call
2009 * lttng_ust_init() again to start over as a new PID.
2010 *
2011 * This is meant for forks() that have tracing in the child between the
2012 * fork and following exec call (if there is any).
2013 */
2014 void ust_after_fork_child(sigset_t *restore_sigset)
2015 {
2016 if (URCU_TLS(lttng_ust_nest_count))
2017 return;
2018 lttng_context_vpid_reset();
2019 lttng_context_vtid_reset();
2020 lttng_context_procname_reset();
2021 DBG("process %d", getpid());
2022 /* Release urcu mutexes */
2023 rcu_bp_after_fork_child();
2024 lttng_ust_cleanup(0);
2025 /* Release mutexes and reenable signals */
2026 ust_after_fork_common(restore_sigset);
2027 lttng_ust_init();
2028 }
2029
2030 void lttng_ust_sockinfo_session_enabled(void *owner)
2031 {
2032 struct sock_info *sock_info = owner;
2033 sock_info->statedump_pending = 1;
2034 }
LTTng 2.0 Userspace Tracer
This page took 0.105225 seconds and 3 git commands to generate.