Namespace lttng_context_procname_reset public symbol
[lttng-ust.git] / liblttng-ust / lttng-ust-comm.c
1 /*
2 * SPDX-License-Identifier: LGPL-2.1-only
3 *
4 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
5 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 */
7
8 #define _LGPL_SOURCE
9 #include <stddef.h>
10 #include <stdint.h>
11 #include <sys/types.h>
12 #include <sys/socket.h>
13 #include <sys/mman.h>
14 #include <sys/stat.h>
15 #include <sys/types.h>
16 #include <sys/wait.h>
17 #include <dlfcn.h>
18 #include <fcntl.h>
19 #include <unistd.h>
20 #include <errno.h>
21 #include <pthread.h>
22 #include <semaphore.h>
23 #include <time.h>
24 #include <assert.h>
25 #include <signal.h>
26 #include <limits.h>
27 #include <urcu/uatomic.h>
28 #include "futex.h"
29 #include <urcu/compiler.h>
30 #include <lttng/urcu/urcu-ust.h>
31
32 #include <lttng/ust-events.h>
33 #include <lttng/ust-abi.h>
34 #include <lttng/ust.h>
35 #include <lttng/ust-error.h>
36 #include <lttng/ust-ctl.h>
37 #include <lttng/ust-libc-wrapper.h>
38 #include <urcu/tls-compat.h>
39 #include <ust-comm.h>
40 #include <ust-fd.h>
41 #include <usterr-signal-safe.h>
42 #include <ust-helper.h>
43 #include "tracepoint-internal.h"
44 #include "lttng-tracer-core.h"
45 #include "compat.h"
46 #include "../libringbuffer/rb-init.h"
47 #include "lttng-ust-statedump.h"
48 #include "clock.h"
49 #include "../libringbuffer/getcpu.h"
50 #include "getenv.h"
51 #include "ust-events-internal.h"
52 #include "context-internal.h"
53 #include "ust-compat.h"
54
55 /* Concatenate lttng ust shared library name with its major version number. */
56 #define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." __ust_stringify(CONFIG_LTTNG_UST_LIBRARY_VERSION_MAJOR)
57
58 /*
59 * Has lttng ust comm constructor been called ?
60 */
61 static int initialized;
62
63 /*
64 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
65 * Held when handling a command, also held by fork() to deal with
66 * removal of threads, and by exit path.
67 *
68 * The UST lock is the centralized mutex across UST tracing control and
69 * probe registration.
70 *
71 * ust_exit_mutex must never nest in ust_mutex.
72 *
73 * ust_fork_mutex must never nest in ust_mutex.
74 *
75 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
76 * counter lazy initialization called by events within the statedump,
77 * which traces while the ust_mutex is held.
78 *
79 * ust_lock nests within the dynamic loader lock (within glibc) because
80 * it is taken within the library constructor.
81 *
82 * The ust fd tracker lock nests within the ust_mutex.
83 */
84 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
85
86 /* Allow nesting the ust_mutex within the same thread. */
87 static DEFINE_URCU_TLS(int, ust_mutex_nest);
88
89 /*
90 * ust_exit_mutex protects thread_active variable wrt thread exit. It
91 * cannot be done by ust_mutex because pthread_cancel(), which takes an
92 * internal libc lock, cannot nest within ust_mutex.
93 *
94 * It never nests within a ust_mutex.
95 */
96 static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
97
98 /*
99 * ust_fork_mutex protects base address statedump tracing against forks. It
100 * prevents the dynamic loader lock to be taken (by base address statedump
101 * tracing) while a fork is happening, thus preventing deadlock issues with
102 * the dynamic loader lock.
103 */
104 static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
105
106 /* Should the ust comm thread quit ? */
107 static int lttng_ust_comm_should_quit;
108
109 /*
110 * This variable can be tested by applications to check whether
111 * lttng-ust is loaded. They simply have to define their own
112 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
113 * library constructor.
114 */
115 int lttng_ust_loaded __attribute__((weak));
116
117 /*
118 * Return 0 on success, -1 if should quit.
119 * The lock is taken in both cases.
120 * Signal-safe.
121 */
122 int ust_lock(void)
123 {
124 sigset_t sig_all_blocked, orig_mask;
125 int ret, oldstate;
126
127 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
128 if (ret) {
129 ERR("pthread_setcancelstate: %s", strerror(ret));
130 }
131 if (oldstate != PTHREAD_CANCEL_ENABLE) {
132 ERR("pthread_setcancelstate: unexpected oldstate");
133 }
134 sigfillset(&sig_all_blocked);
135 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
136 if (ret) {
137 ERR("pthread_sigmask: %s", strerror(ret));
138 }
139 if (!URCU_TLS(ust_mutex_nest)++)
140 pthread_mutex_lock(&ust_mutex);
141 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
142 if (ret) {
143 ERR("pthread_sigmask: %s", strerror(ret));
144 }
145 if (lttng_ust_comm_should_quit) {
146 return -1;
147 } else {
148 return 0;
149 }
150 }
151
152 /*
153 * ust_lock_nocheck() can be used in constructors/destructors, because
154 * they are already nested within the dynamic loader lock, and therefore
155 * have exclusive access against execution of liblttng-ust destructor.
156 * Signal-safe.
157 */
158 void ust_lock_nocheck(void)
159 {
160 sigset_t sig_all_blocked, orig_mask;
161 int ret, oldstate;
162
163 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
164 if (ret) {
165 ERR("pthread_setcancelstate: %s", strerror(ret));
166 }
167 if (oldstate != PTHREAD_CANCEL_ENABLE) {
168 ERR("pthread_setcancelstate: unexpected oldstate");
169 }
170 sigfillset(&sig_all_blocked);
171 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
172 if (ret) {
173 ERR("pthread_sigmask: %s", strerror(ret));
174 }
175 if (!URCU_TLS(ust_mutex_nest)++)
176 pthread_mutex_lock(&ust_mutex);
177 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
178 if (ret) {
179 ERR("pthread_sigmask: %s", strerror(ret));
180 }
181 }
182
183 /*
184 * Signal-safe.
185 */
186 void ust_unlock(void)
187 {
188 sigset_t sig_all_blocked, orig_mask;
189 int ret, oldstate;
190
191 sigfillset(&sig_all_blocked);
192 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
193 if (ret) {
194 ERR("pthread_sigmask: %s", strerror(ret));
195 }
196 if (!--URCU_TLS(ust_mutex_nest))
197 pthread_mutex_unlock(&ust_mutex);
198 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
199 if (ret) {
200 ERR("pthread_sigmask: %s", strerror(ret));
201 }
202 ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
203 if (ret) {
204 ERR("pthread_setcancelstate: %s", strerror(ret));
205 }
206 if (oldstate != PTHREAD_CANCEL_DISABLE) {
207 ERR("pthread_setcancelstate: unexpected oldstate");
208 }
209 }
210
211 /*
212 * Wait for either of these before continuing to the main
213 * program:
214 * - the register_done message from sessiond daemon
215 * (will let the sessiond daemon enable sessions before main
216 * starts.)
217 * - sessiond daemon is not reachable.
218 * - timeout (ensuring applications are resilient to session
219 * daemon problems).
220 */
221 static sem_t constructor_wait;
222 /*
223 * Doing this for both the global and local sessiond.
224 */
225 enum {
226 sem_count_initial_value = 4,
227 };
228
229 static int sem_count = sem_count_initial_value;
230
231 /*
232 * Counting nesting within lttng-ust. Used to ensure that calling fork()
233 * from liblttng-ust does not execute the pre/post fork handlers.
234 */
235 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
236
237 /*
238 * Info about socket and associated listener thread.
239 */
240 struct sock_info {
241 const char *name;
242 pthread_t ust_listener; /* listener thread */
243 int root_handle;
244 int registration_done;
245 int allowed;
246 int global;
247 int thread_active;
248
249 char sock_path[PATH_MAX];
250 int socket;
251 int notify_socket;
252
253 char wait_shm_path[PATH_MAX];
254 char *wait_shm_mmap;
255 /* Keep track of lazy state dump not performed yet. */
256 int statedump_pending;
257 int initial_statedump_done;
258 /* Keep procname for statedump */
259 char procname[LTTNG_UST_ABI_PROCNAME_LEN];
260 };
261
262 /* Socket from app (connect) to session daemon (listen) for communication */
263 struct sock_info global_apps = {
264 .name = "global",
265 .global = 1,
266
267 .root_handle = -1,
268 .registration_done = 0,
269 .allowed = 0,
270 .thread_active = 0,
271
272 .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
273 .socket = -1,
274 .notify_socket = -1,
275
276 .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
277
278 .statedump_pending = 0,
279 .initial_statedump_done = 0,
280 .procname[0] = '\0'
281 };
282
283 /* TODO: allow global_apps_sock_path override */
284
285 struct sock_info local_apps = {
286 .name = "local",
287 .global = 0,
288 .root_handle = -1,
289 .registration_done = 0,
290 .allowed = 0, /* Check setuid bit first */
291 .thread_active = 0,
292
293 .socket = -1,
294 .notify_socket = -1,
295
296 .statedump_pending = 0,
297 .initial_statedump_done = 0,
298 .procname[0] = '\0'
299 };
300
301 static int wait_poll_fallback;
302
303 static const char *cmd_name_mapping[] = {
304 [ LTTNG_UST_ABI_RELEASE ] = "Release",
305 [ LTTNG_UST_ABI_SESSION ] = "Create Session",
306 [ LTTNG_UST_ABI_TRACER_VERSION ] = "Get Tracer Version",
307
308 [ LTTNG_UST_ABI_TRACEPOINT_LIST ] = "Create Tracepoint List",
309 [ LTTNG_UST_ABI_WAIT_QUIESCENT ] = "Wait for Quiescent State",
310 [ LTTNG_UST_ABI_REGISTER_DONE ] = "Registration Done",
311 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
312
313 [ LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE ] = "Create event notifier group",
314
315 /* Session FD commands */
316 [ LTTNG_UST_ABI_CHANNEL ] = "Create Channel",
317 [ LTTNG_UST_ABI_SESSION_START ] = "Start Session",
318 [ LTTNG_UST_ABI_SESSION_STOP ] = "Stop Session",
319
320 /* Channel FD commands */
321 [ LTTNG_UST_ABI_STREAM ] = "Create Stream",
322 [ LTTNG_UST_ABI_EVENT ] = "Create Event",
323
324 /* Event and Channel FD commands */
325 [ LTTNG_UST_ABI_CONTEXT ] = "Create Context",
326 [ LTTNG_UST_ABI_FLUSH_BUFFER ] = "Flush Buffer",
327
328 /* Event, Channel and Session commands */
329 [ LTTNG_UST_ABI_ENABLE ] = "Enable",
330 [ LTTNG_UST_ABI_DISABLE ] = "Disable",
331
332 /* Tracepoint list commands */
333 [ LTTNG_UST_ABI_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
334 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
335
336 /* Event FD commands */
337 [ LTTNG_UST_ABI_FILTER ] = "Create Filter",
338 [ LTTNG_UST_ABI_EXCLUSION ] = "Add exclusions to event",
339
340 /* Event notifier group commands */
341 [ LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE ] = "Create event notifier",
342
343 /* Session and event notifier group commands */
344 [ LTTNG_UST_ABI_COUNTER ] = "Create Counter",
345
346 /* Counter commands */
347 [ LTTNG_UST_ABI_COUNTER_GLOBAL ] = "Create Counter Global",
348 [ LTTNG_UST_ABI_COUNTER_CPU ] = "Create Counter CPU",
349 };
350
351 static const char *str_timeout;
352 static int got_timeout_env;
353
354 extern void lttng_ring_buffer_client_overwrite_init(void);
355 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
356 extern void lttng_ring_buffer_client_discard_init(void);
357 extern void lttng_ring_buffer_client_discard_rt_init(void);
358 extern void lttng_ring_buffer_metadata_client_init(void);
359 extern void lttng_ring_buffer_client_overwrite_exit(void);
360 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
361 extern void lttng_ring_buffer_client_discard_exit(void);
362 extern void lttng_ring_buffer_client_discard_rt_exit(void);
363 extern void lttng_ring_buffer_metadata_client_exit(void);
364
365 __attribute__((visibility("hidden")))
366 extern void lttng_counter_client_percpu_32_modular_init(void);
367
368 __attribute__((visibility("hidden")))
369 extern void lttng_counter_client_percpu_32_modular_exit(void);
370
371 __attribute__((visibility("hidden")))
372 extern void lttng_counter_client_percpu_64_modular_init(void);
373
374 __attribute__((visibility("hidden")))
375 extern void lttng_counter_client_percpu_64_modular_exit(void);
376
377 static char *get_map_shm(struct sock_info *sock_info);
378
379 ssize_t lttng_ust_read(int fd, void *buf, size_t len)
380 {
381 ssize_t ret;
382 size_t copied = 0, to_copy = len;
383
384 do {
385 ret = read(fd, buf + copied, to_copy);
386 if (ret > 0) {
387 copied += ret;
388 to_copy -= ret;
389 }
390 } while ((ret > 0 && to_copy > 0)
391 || (ret < 0 && errno == EINTR));
392 if (ret > 0) {
393 ret = copied;
394 }
395 return ret;
396 }
397 /*
398 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
399 * pointer.
400 */
401 static
402 const char *get_lttng_home_dir(void)
403 {
404 const char *val;
405
406 val = (const char *) lttng_ust_getenv("LTTNG_HOME");
407 if (val != NULL) {
408 return val;
409 }
410 return (const char *) lttng_ust_getenv("HOME");
411 }
412
413 /*
414 * Force a read (imply TLS fixup for dlopen) of TLS variables.
415 */
416 static
417 void lttng_fixup_nest_count_tls(void)
418 {
419 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
420 }
421
422 static
423 void lttng_fixup_ust_mutex_nest_tls(void)
424 {
425 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
426 }
427
428 /*
429 * Fixup lttng-ust urcu TLS.
430 */
431 static
432 void lttng_fixup_lttng_ust_urcu_tls(void)
433 {
434 (void) lttng_ust_urcu_read_ongoing();
435 }
436
437 void lttng_ust_fixup_tls(void)
438 {
439 lttng_fixup_lttng_ust_urcu_tls();
440 lttng_fixup_ringbuffer_tls();
441 lttng_fixup_vtid_tls();
442 lttng_fixup_nest_count_tls();
443 lttng_fixup_procname_tls();
444 lttng_fixup_ust_mutex_nest_tls();
445 lttng_ust_fixup_perf_counter_tls();
446 lttng_ust_fixup_fd_tracker_tls();
447 lttng_fixup_cgroup_ns_tls();
448 lttng_fixup_ipc_ns_tls();
449 lttng_fixup_net_ns_tls();
450 lttng_fixup_time_ns_tls();
451 lttng_fixup_uts_ns_tls();
452 }
453
454 int lttng_get_notify_socket(void *owner)
455 {
456 struct sock_info *info = owner;
457
458 return info->notify_socket;
459 }
460
461
462 char* lttng_ust_sockinfo_get_procname(void *owner)
463 {
464 struct sock_info *info = owner;
465
466 return info->procname;
467 }
468
469 static
470 void print_cmd(int cmd, int handle)
471 {
472 const char *cmd_name = "Unknown";
473
474 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
475 && cmd_name_mapping[cmd]) {
476 cmd_name = cmd_name_mapping[cmd];
477 }
478 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
479 cmd_name, cmd,
480 lttng_ust_obj_get_name(handle), handle);
481 }
482
483 static
484 int setup_global_apps(void)
485 {
486 int ret = 0;
487 assert(!global_apps.wait_shm_mmap);
488
489 global_apps.wait_shm_mmap = get_map_shm(&global_apps);
490 if (!global_apps.wait_shm_mmap) {
491 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
492 global_apps.allowed = 0;
493 ret = -EIO;
494 goto error;
495 }
496
497 global_apps.allowed = 1;
498 lttng_pthread_getname_np(global_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
499 error:
500 return ret;
501 }
502 static
503 int setup_local_apps(void)
504 {
505 int ret = 0;
506 const char *home_dir;
507 uid_t uid;
508
509 assert(!local_apps.wait_shm_mmap);
510
511 uid = getuid();
512 /*
513 * Disallow per-user tracing for setuid binaries.
514 */
515 if (uid != geteuid()) {
516 assert(local_apps.allowed == 0);
517 ret = 0;
518 goto end;
519 }
520 home_dir = get_lttng_home_dir();
521 if (!home_dir) {
522 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
523 assert(local_apps.allowed == 0);
524 ret = -ENOENT;
525 goto end;
526 }
527 local_apps.allowed = 1;
528 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
529 home_dir,
530 LTTNG_DEFAULT_HOME_RUNDIR,
531 LTTNG_UST_SOCK_FILENAME);
532 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
533 LTTNG_UST_WAIT_FILENAME,
534 uid);
535
536 local_apps.wait_shm_mmap = get_map_shm(&local_apps);
537 if (!local_apps.wait_shm_mmap) {
538 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
539 local_apps.allowed = 0;
540 ret = -EIO;
541 goto end;
542 }
543
544 lttng_pthread_getname_np(local_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
545 end:
546 return ret;
547 }
548
549 /*
550 * Get socket timeout, in ms.
551 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
552 */
553 static
554 long get_timeout(void)
555 {
556 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
557
558 if (!got_timeout_env) {
559 str_timeout = lttng_ust_getenv("LTTNG_UST_REGISTER_TIMEOUT");
560 got_timeout_env = 1;
561 }
562 if (str_timeout)
563 constructor_delay_ms = strtol(str_timeout, NULL, 10);
564 /* All negative values are considered as "-1". */
565 if (constructor_delay_ms < -1)
566 constructor_delay_ms = -1;
567 return constructor_delay_ms;
568 }
569
570 /* Timeout for notify socket send and recv. */
571 static
572 long get_notify_sock_timeout(void)
573 {
574 return get_timeout();
575 }
576
577 /* Timeout for connecting to cmd and notify sockets. */
578 static
579 long get_connect_sock_timeout(void)
580 {
581 return get_timeout();
582 }
583
584 /*
585 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
586 */
587 static
588 int get_constructor_timeout(struct timespec *constructor_timeout)
589 {
590 long constructor_delay_ms;
591 int ret;
592
593 constructor_delay_ms = get_timeout();
594
595 switch (constructor_delay_ms) {
596 case -1:/* fall-through */
597 case 0:
598 return constructor_delay_ms;
599 default:
600 break;
601 }
602
603 /*
604 * If we are unable to find the current time, don't wait.
605 */
606 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
607 if (ret) {
608 /* Don't wait. */
609 return 0;
610 }
611 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
612 constructor_timeout->tv_nsec +=
613 (constructor_delay_ms % 1000UL) * 1000000UL;
614 if (constructor_timeout->tv_nsec >= 1000000000UL) {
615 constructor_timeout->tv_sec++;
616 constructor_timeout->tv_nsec -= 1000000000UL;
617 }
618 /* Timeout wait (constructor_delay_ms). */
619 return 1;
620 }
621
622 static
623 void get_allow_blocking(void)
624 {
625 const char *str_allow_blocking =
626 lttng_ust_getenv("LTTNG_UST_ALLOW_BLOCKING");
627
628 if (str_allow_blocking) {
629 DBG("%s environment variable is set",
630 "LTTNG_UST_ALLOW_BLOCKING");
631 lttng_ust_ringbuffer_set_allow_blocking();
632 }
633 }
634
635 static
636 int register_to_sessiond(int socket, enum ustctl_socket_type type)
637 {
638 return ustcomm_send_reg_msg(socket,
639 type,
640 CAA_BITS_PER_LONG,
641 lttng_alignof(uint8_t) * CHAR_BIT,
642 lttng_alignof(uint16_t) * CHAR_BIT,
643 lttng_alignof(uint32_t) * CHAR_BIT,
644 lttng_alignof(uint64_t) * CHAR_BIT,
645 lttng_alignof(unsigned long) * CHAR_BIT);
646 }
647
648 static
649 int send_reply(int sock, struct ustcomm_ust_reply *lur)
650 {
651 ssize_t len;
652
653 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
654 switch (len) {
655 case sizeof(*lur):
656 DBG("message successfully sent");
657 return 0;
658 default:
659 if (len == -ECONNRESET) {
660 DBG("remote end closed connection");
661 return 0;
662 }
663 if (len < 0)
664 return len;
665 DBG("incorrect message size: %zd", len);
666 return -EINVAL;
667 }
668 }
669
670 static
671 void decrement_sem_count(unsigned int count)
672 {
673 int ret;
674
675 assert(uatomic_read(&sem_count) >= count);
676
677 if (uatomic_read(&sem_count) <= 0) {
678 return;
679 }
680
681 ret = uatomic_add_return(&sem_count, -count);
682 if (ret == 0) {
683 ret = sem_post(&constructor_wait);
684 assert(!ret);
685 }
686 }
687
688 static
689 int handle_register_done(struct sock_info *sock_info)
690 {
691 if (sock_info->registration_done)
692 return 0;
693 sock_info->registration_done = 1;
694
695 decrement_sem_count(1);
696 if (!sock_info->statedump_pending) {
697 sock_info->initial_statedump_done = 1;
698 decrement_sem_count(1);
699 }
700
701 return 0;
702 }
703
704 static
705 int handle_register_failed(struct sock_info *sock_info)
706 {
707 if (sock_info->registration_done)
708 return 0;
709 sock_info->registration_done = 1;
710 sock_info->initial_statedump_done = 1;
711
712 decrement_sem_count(2);
713
714 return 0;
715 }
716
717 /*
718 * Only execute pending statedump after the constructor semaphore has
719 * been posted by the current listener thread. This means statedump will
720 * only be performed after the "registration done" command is received
721 * from this thread's session daemon.
722 *
723 * This ensures we don't run into deadlock issues with the dynamic
724 * loader mutex, which is held while the constructor is called and
725 * waiting on the constructor semaphore. All operations requiring this
726 * dynamic loader lock need to be postponed using this mechanism.
727 *
728 * In a scenario with two session daemons connected to the application,
729 * it is possible that the first listener thread which receives the
730 * registration done command issues its statedump while the dynamic
731 * loader lock is still held by the application constructor waiting on
732 * the semaphore. It will however be allowed to proceed when the
733 * second session daemon sends the registration done command to the
734 * second listener thread. This situation therefore does not produce
735 * a deadlock.
736 */
737 static
738 void handle_pending_statedump(struct sock_info *sock_info)
739 {
740 if (sock_info->registration_done && sock_info->statedump_pending) {
741 sock_info->statedump_pending = 0;
742 pthread_mutex_lock(&ust_fork_mutex);
743 lttng_handle_pending_statedump(sock_info);
744 pthread_mutex_unlock(&ust_fork_mutex);
745
746 if (!sock_info->initial_statedump_done) {
747 sock_info->initial_statedump_done = 1;
748 decrement_sem_count(1);
749 }
750 }
751 }
752
753 static inline
754 const char *bytecode_type_str(uint32_t cmd)
755 {
756 switch (cmd) {
757 case LTTNG_UST_ABI_CAPTURE:
758 return "capture";
759 case LTTNG_UST_ABI_FILTER:
760 return "filter";
761 default:
762 abort();
763 }
764 }
765
766 static
767 int handle_bytecode_recv(struct sock_info *sock_info,
768 int sock, struct ustcomm_ust_msg *lum)
769 {
770 struct lttng_ust_bytecode_node *bytecode = NULL;
771 enum lttng_ust_bytecode_node_type type;
772 const struct lttng_ust_abi_objd_ops *ops;
773 uint32_t data_size, data_size_max, reloc_offset;
774 uint64_t seqnum;
775 ssize_t len;
776 int ret = 0;
777
778 switch (lum->cmd) {
779 case LTTNG_UST_ABI_FILTER:
780 type = LTTNG_UST_BYTECODE_NODE_TYPE_FILTER;
781 data_size = lum->u.filter.data_size;
782 data_size_max = LTTNG_UST_ABI_FILTER_BYTECODE_MAX_LEN;
783 reloc_offset = lum->u.filter.reloc_offset;
784 seqnum = lum->u.filter.seqnum;
785 break;
786 case LTTNG_UST_ABI_CAPTURE:
787 type = LTTNG_UST_BYTECODE_NODE_TYPE_CAPTURE;
788 data_size = lum->u.capture.data_size;
789 data_size_max = LTTNG_UST_ABI_CAPTURE_BYTECODE_MAX_LEN;
790 reloc_offset = lum->u.capture.reloc_offset;
791 seqnum = lum->u.capture.seqnum;
792 break;
793 default:
794 abort();
795 }
796
797 if (data_size > data_size_max) {
798 ERR("Bytecode %s data size is too large: %u bytes",
799 bytecode_type_str(lum->cmd), data_size);
800 ret = -EINVAL;
801 goto end;
802 }
803
804 if (reloc_offset > data_size) {
805 ERR("Bytecode %s reloc offset %u is not within data",
806 bytecode_type_str(lum->cmd), reloc_offset);
807 ret = -EINVAL;
808 goto end;
809 }
810
811 /* Allocate the structure AND the `data[]` field. */
812 bytecode = zmalloc(sizeof(*bytecode) + data_size);
813 if (!bytecode) {
814 ret = -ENOMEM;
815 goto end;
816 }
817
818 bytecode->bc.len = data_size;
819 bytecode->bc.reloc_offset = reloc_offset;
820 bytecode->bc.seqnum = seqnum;
821 bytecode->type = type;
822
823 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, bytecode->bc.len);
824 switch (len) {
825 case 0: /* orderly shutdown */
826 ret = 0;
827 goto end;
828 default:
829 if (len == bytecode->bc.len) {
830 DBG("Bytecode %s data received",
831 bytecode_type_str(lum->cmd));
832 break;
833 } else if (len < 0) {
834 DBG("Receive failed from lttng-sessiond with errno %d",
835 (int) -len);
836 if (len == -ECONNRESET) {
837 ERR("%s remote end closed connection",
838 sock_info->name);
839 ret = len;
840 goto end;
841 }
842 ret = len;
843 goto end;
844 } else {
845 DBG("Incorrect %s bytecode data message size: %zd",
846 bytecode_type_str(lum->cmd), len);
847 ret = -EINVAL;
848 goto end;
849 }
850 }
851
852 ops = lttng_ust_abi_objd_ops(lum->handle);
853 if (!ops) {
854 ret = -ENOENT;
855 goto end;
856 }
857
858 if (ops->cmd)
859 ret = ops->cmd(lum->handle, lum->cmd,
860 (unsigned long) &bytecode,
861 NULL, sock_info);
862 else
863 ret = -ENOSYS;
864
865 end:
866 free(bytecode);
867 return ret;
868 }
869
870 static
871 int handle_message(struct sock_info *sock_info,
872 int sock, struct ustcomm_ust_msg *lum)
873 {
874 int ret = 0;
875 const struct lttng_ust_abi_objd_ops *ops;
876 struct ustcomm_ust_reply lur;
877 union lttng_ust_abi_args args;
878 char ctxstr[LTTNG_UST_ABI_SYM_NAME_LEN]; /* App context string. */
879 ssize_t len;
880
881 memset(&lur, 0, sizeof(lur));
882
883 if (ust_lock()) {
884 ret = -LTTNG_UST_ERR_EXITING;
885 goto error;
886 }
887
888 ops = lttng_ust_abi_objd_ops(lum->handle);
889 if (!ops) {
890 ret = -ENOENT;
891 goto error;
892 }
893
894 switch (lum->cmd) {
895 case LTTNG_UST_ABI_REGISTER_DONE:
896 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
897 ret = handle_register_done(sock_info);
898 else
899 ret = -EINVAL;
900 break;
901 case LTTNG_UST_ABI_RELEASE:
902 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
903 ret = -EPERM;
904 else
905 ret = lttng_ust_abi_objd_unref(lum->handle, 1);
906 break;
907 case LTTNG_UST_ABI_CAPTURE:
908 case LTTNG_UST_ABI_FILTER:
909 ret = handle_bytecode_recv(sock_info, sock, lum);
910 if (ret)
911 goto error;
912 break;
913 case LTTNG_UST_ABI_EXCLUSION:
914 {
915 /* Receive exclusion names */
916 struct lttng_ust_excluder_node *node;
917 unsigned int count;
918
919 count = lum->u.exclusion.count;
920 if (count == 0) {
921 /* There are no names to read */
922 ret = 0;
923 goto error;
924 }
925 node = zmalloc(sizeof(*node) +
926 count * LTTNG_UST_ABI_SYM_NAME_LEN);
927 if (!node) {
928 ret = -ENOMEM;
929 goto error;
930 }
931 node->excluder.count = count;
932 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
933 count * LTTNG_UST_ABI_SYM_NAME_LEN);
934 switch (len) {
935 case 0: /* orderly shutdown */
936 ret = 0;
937 free(node);
938 goto error;
939 default:
940 if (len == count * LTTNG_UST_ABI_SYM_NAME_LEN) {
941 DBG("Exclusion data received");
942 break;
943 } else if (len < 0) {
944 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
945 if (len == -ECONNRESET) {
946 ERR("%s remote end closed connection", sock_info->name);
947 ret = len;
948 free(node);
949 goto error;
950 }
951 ret = len;
952 free(node);
953 goto error;
954 } else {
955 DBG("Incorrect exclusion data message size: %zd", len);
956 ret = -EINVAL;
957 free(node);
958 goto error;
959 }
960 }
961 if (ops->cmd)
962 ret = ops->cmd(lum->handle, lum->cmd,
963 (unsigned long) &node,
964 &args, sock_info);
965 else
966 ret = -ENOSYS;
967 free(node);
968 break;
969 }
970 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE:
971 {
972 int event_notifier_notif_fd, close_ret;
973
974 len = ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock,
975 &event_notifier_notif_fd);
976 switch (len) {
977 case 0: /* orderly shutdown */
978 ret = 0;
979 goto error;
980 case 1:
981 break;
982 default:
983 if (len < 0) {
984 DBG("Receive failed from lttng-sessiond with errno %d",
985 (int) -len);
986 if (len == -ECONNRESET) {
987 ERR("%s remote end closed connection",
988 sock_info->name);
989 ret = len;
990 goto error;
991 }
992 ret = len;
993 goto error;
994 } else {
995 DBG("Incorrect event notifier fd message size: %zd",
996 len);
997 ret = -EINVAL;
998 goto error;
999 }
1000 }
1001 args.event_notifier_handle.event_notifier_notif_fd =
1002 event_notifier_notif_fd;
1003 if (ops->cmd)
1004 ret = ops->cmd(lum->handle, lum->cmd,
1005 (unsigned long) &lum->u,
1006 &args, sock_info);
1007 else
1008 ret = -ENOSYS;
1009 if (args.event_notifier_handle.event_notifier_notif_fd >= 0) {
1010 lttng_ust_lock_fd_tracker();
1011 close_ret = close(args.event_notifier_handle.event_notifier_notif_fd);
1012 lttng_ust_unlock_fd_tracker();
1013 if (close_ret)
1014 PERROR("close");
1015 }
1016 break;
1017 }
1018 case LTTNG_UST_ABI_CHANNEL:
1019 {
1020 void *chan_data;
1021 int wakeup_fd;
1022
1023 len = ustcomm_recv_channel_from_sessiond(sock,
1024 &chan_data, lum->u.channel.len,
1025 &wakeup_fd);
1026 switch (len) {
1027 case 0: /* orderly shutdown */
1028 ret = 0;
1029 goto error;
1030 default:
1031 if (len == lum->u.channel.len) {
1032 DBG("channel data received");
1033 break;
1034 } else if (len < 0) {
1035 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1036 if (len == -ECONNRESET) {
1037 ERR("%s remote end closed connection", sock_info->name);
1038 ret = len;
1039 goto error;
1040 }
1041 ret = len;
1042 goto error;
1043 } else {
1044 DBG("incorrect channel data message size: %zd", len);
1045 ret = -EINVAL;
1046 goto error;
1047 }
1048 }
1049 args.channel.chan_data = chan_data;
1050 args.channel.wakeup_fd = wakeup_fd;
1051 if (ops->cmd)
1052 ret = ops->cmd(lum->handle, lum->cmd,
1053 (unsigned long) &lum->u,
1054 &args, sock_info);
1055 else
1056 ret = -ENOSYS;
1057 if (args.channel.wakeup_fd >= 0) {
1058 int close_ret;
1059
1060 lttng_ust_lock_fd_tracker();
1061 close_ret = close(args.channel.wakeup_fd);
1062 lttng_ust_unlock_fd_tracker();
1063 args.channel.wakeup_fd = -1;
1064 if (close_ret)
1065 PERROR("close");
1066 }
1067 free(args.channel.chan_data);
1068 break;
1069 }
1070 case LTTNG_UST_ABI_STREAM:
1071 {
1072 int close_ret;
1073
1074 /* Receive shm_fd, wakeup_fd */
1075 ret = ustcomm_recv_stream_from_sessiond(sock,
1076 NULL,
1077 &args.stream.shm_fd,
1078 &args.stream.wakeup_fd);
1079 if (ret) {
1080 goto error;
1081 }
1082
1083 if (ops->cmd)
1084 ret = ops->cmd(lum->handle, lum->cmd,
1085 (unsigned long) &lum->u,
1086 &args, sock_info);
1087 else
1088 ret = -ENOSYS;
1089 if (args.stream.shm_fd >= 0) {
1090 lttng_ust_lock_fd_tracker();
1091 close_ret = close(args.stream.shm_fd);
1092 lttng_ust_unlock_fd_tracker();
1093 args.stream.shm_fd = -1;
1094 if (close_ret)
1095 PERROR("close");
1096 }
1097 if (args.stream.wakeup_fd >= 0) {
1098 lttng_ust_lock_fd_tracker();
1099 close_ret = close(args.stream.wakeup_fd);
1100 lttng_ust_unlock_fd_tracker();
1101 args.stream.wakeup_fd = -1;
1102 if (close_ret)
1103 PERROR("close");
1104 }
1105 break;
1106 }
1107 case LTTNG_UST_ABI_CONTEXT:
1108 switch (lum->u.context.ctx) {
1109 case LTTNG_UST_ABI_CONTEXT_APP_CONTEXT:
1110 {
1111 char *p;
1112 size_t ctxlen, recvlen;
1113
1114 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
1115 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
1116 if (ctxlen >= LTTNG_UST_ABI_SYM_NAME_LEN) {
1117 ERR("Application context string length size is too large: %zu bytes",
1118 ctxlen);
1119 ret = -EINVAL;
1120 goto error;
1121 }
1122 strcpy(ctxstr, "$app.");
1123 p = &ctxstr[strlen("$app.")];
1124 recvlen = ctxlen - strlen("$app.");
1125 len = ustcomm_recv_unix_sock(sock, p, recvlen);
1126 switch (len) {
1127 case 0: /* orderly shutdown */
1128 ret = 0;
1129 goto error;
1130 default:
1131 if (len == recvlen) {
1132 DBG("app context data received");
1133 break;
1134 } else if (len < 0) {
1135 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1136 if (len == -ECONNRESET) {
1137 ERR("%s remote end closed connection", sock_info->name);
1138 ret = len;
1139 goto error;
1140 }
1141 ret = len;
1142 goto error;
1143 } else {
1144 DBG("incorrect app context data message size: %zd", len);
1145 ret = -EINVAL;
1146 goto error;
1147 }
1148 }
1149 /* Put : between provider and ctxname. */
1150 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1151 args.app_context.ctxname = ctxstr;
1152 break;
1153 }
1154 default:
1155 break;
1156 }
1157 if (ops->cmd) {
1158 ret = ops->cmd(lum->handle, lum->cmd,
1159 (unsigned long) &lum->u,
1160 &args, sock_info);
1161 } else {
1162 ret = -ENOSYS;
1163 }
1164 break;
1165 case LTTNG_UST_ABI_COUNTER:
1166 {
1167 void *counter_data;
1168
1169 len = ustcomm_recv_counter_from_sessiond(sock,
1170 &counter_data, lum->u.counter.len);
1171 switch (len) {
1172 case 0: /* orderly shutdown */
1173 ret = 0;
1174 goto error;
1175 default:
1176 if (len == lum->u.counter.len) {
1177 DBG("counter data received");
1178 break;
1179 } else if (len < 0) {
1180 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1181 if (len == -ECONNRESET) {
1182 ERR("%s remote end closed connection", sock_info->name);
1183 ret = len;
1184 goto error;
1185 }
1186 ret = len;
1187 goto error;
1188 } else {
1189 DBG("incorrect counter data message size: %zd", len);
1190 ret = -EINVAL;
1191 goto error;
1192 }
1193 }
1194 args.counter.counter_data = counter_data;
1195 if (ops->cmd)
1196 ret = ops->cmd(lum->handle, lum->cmd,
1197 (unsigned long) &lum->u,
1198 &args, sock_info);
1199 else
1200 ret = -ENOSYS;
1201 free(args.counter.counter_data);
1202 break;
1203 }
1204 case LTTNG_UST_ABI_COUNTER_GLOBAL:
1205 {
1206 /* Receive shm_fd */
1207 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1208 &args.counter_shm.shm_fd);
1209 if (ret) {
1210 goto error;
1211 }
1212
1213 if (ops->cmd)
1214 ret = ops->cmd(lum->handle, lum->cmd,
1215 (unsigned long) &lum->u,
1216 &args, sock_info);
1217 else
1218 ret = -ENOSYS;
1219 if (args.counter_shm.shm_fd >= 0) {
1220 int close_ret;
1221
1222 lttng_ust_lock_fd_tracker();
1223 close_ret = close(args.counter_shm.shm_fd);
1224 lttng_ust_unlock_fd_tracker();
1225 args.counter_shm.shm_fd = -1;
1226 if (close_ret)
1227 PERROR("close");
1228 }
1229 break;
1230 }
1231 case LTTNG_UST_ABI_COUNTER_CPU:
1232 {
1233 /* Receive shm_fd */
1234 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1235 &args.counter_shm.shm_fd);
1236 if (ret) {
1237 goto error;
1238 }
1239
1240 if (ops->cmd)
1241 ret = ops->cmd(lum->handle, lum->cmd,
1242 (unsigned long) &lum->u,
1243 &args, sock_info);
1244 else
1245 ret = -ENOSYS;
1246 if (args.counter_shm.shm_fd >= 0) {
1247 int close_ret;
1248
1249 lttng_ust_lock_fd_tracker();
1250 close_ret = close(args.counter_shm.shm_fd);
1251 lttng_ust_unlock_fd_tracker();
1252 args.counter_shm.shm_fd = -1;
1253 if (close_ret)
1254 PERROR("close");
1255 }
1256 break;
1257 }
1258 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE:
1259 {
1260 /* Receive struct lttng_ust_event_notifier */
1261 struct lttng_ust_abi_event_notifier event_notifier;
1262
1263 if (sizeof(event_notifier) != lum->u.event_notifier.len) {
1264 DBG("incorrect event notifier data message size: %u", lum->u.event_notifier.len);
1265 ret = -EINVAL;
1266 goto error;
1267 }
1268 len = ustcomm_recv_unix_sock(sock, &event_notifier, sizeof(event_notifier));
1269 switch (len) {
1270 case 0: /* orderly shutdown */
1271 ret = 0;
1272 goto error;
1273 default:
1274 if (len == sizeof(event_notifier)) {
1275 DBG("event notifier data received");
1276 break;
1277 } else if (len < 0) {
1278 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1279 if (len == -ECONNRESET) {
1280 ERR("%s remote end closed connection", sock_info->name);
1281 ret = len;
1282 goto error;
1283 }
1284 ret = len;
1285 goto error;
1286 } else {
1287 DBG("incorrect event notifier data message size: %zd", len);
1288 ret = -EINVAL;
1289 goto error;
1290 }
1291 }
1292 if (ops->cmd)
1293 ret = ops->cmd(lum->handle, lum->cmd,
1294 (unsigned long) &event_notifier,
1295 &args, sock_info);
1296 else
1297 ret = -ENOSYS;
1298 break;
1299 }
1300
1301 default:
1302 if (ops->cmd)
1303 ret = ops->cmd(lum->handle, lum->cmd,
1304 (unsigned long) &lum->u,
1305 &args, sock_info);
1306 else
1307 ret = -ENOSYS;
1308 break;
1309 }
1310
1311 lur.handle = lum->handle;
1312 lur.cmd = lum->cmd;
1313 lur.ret_val = ret;
1314 if (ret >= 0) {
1315 lur.ret_code = LTTNG_UST_OK;
1316 } else {
1317 /*
1318 * Use -LTTNG_UST_ERR as wildcard for UST internal
1319 * error that are not caused by the transport, except if
1320 * we already have a more precise error message to
1321 * report.
1322 */
1323 if (ret > -LTTNG_UST_ERR) {
1324 /* Translate code to UST error. */
1325 switch (ret) {
1326 case -EEXIST:
1327 lur.ret_code = -LTTNG_UST_ERR_EXIST;
1328 break;
1329 case -EINVAL:
1330 lur.ret_code = -LTTNG_UST_ERR_INVAL;
1331 break;
1332 case -ENOENT:
1333 lur.ret_code = -LTTNG_UST_ERR_NOENT;
1334 break;
1335 case -EPERM:
1336 lur.ret_code = -LTTNG_UST_ERR_PERM;
1337 break;
1338 case -ENOSYS:
1339 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
1340 break;
1341 default:
1342 lur.ret_code = -LTTNG_UST_ERR;
1343 break;
1344 }
1345 } else {
1346 lur.ret_code = ret;
1347 }
1348 }
1349 if (ret >= 0) {
1350 switch (lum->cmd) {
1351 case LTTNG_UST_ABI_TRACER_VERSION:
1352 lur.u.version = lum->u.version;
1353 break;
1354 case LTTNG_UST_ABI_TRACEPOINT_LIST_GET:
1355 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1356 break;
1357 }
1358 }
1359 DBG("Return value: %d", lur.ret_val);
1360
1361 ust_unlock();
1362
1363 /*
1364 * Performed delayed statedump operations outside of the UST
1365 * lock. We need to take the dynamic loader lock before we take
1366 * the UST lock internally within handle_pending_statedump().
1367 */
1368 handle_pending_statedump(sock_info);
1369
1370 if (ust_lock()) {
1371 ret = -LTTNG_UST_ERR_EXITING;
1372 goto error;
1373 }
1374
1375 ret = send_reply(sock, &lur);
1376 if (ret < 0) {
1377 DBG("error sending reply");
1378 goto error;
1379 }
1380
1381 /*
1382 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1383 * after the reply.
1384 */
1385 if (lur.ret_code == LTTNG_UST_OK) {
1386 switch (lum->cmd) {
1387 case LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET:
1388 len = ustcomm_send_unix_sock(sock,
1389 &args.field_list.entry,
1390 sizeof(args.field_list.entry));
1391 if (len < 0) {
1392 ret = len;
1393 goto error;
1394 }
1395 if (len != sizeof(args.field_list.entry)) {
1396 ret = -EINVAL;
1397 goto error;
1398 }
1399 }
1400 }
1401
1402 error:
1403 ust_unlock();
1404
1405 return ret;
1406 }
1407
1408 static
1409 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1410 {
1411 int ret;
1412
1413 if (sock_info->root_handle != -1) {
1414 ret = lttng_ust_abi_objd_unref(sock_info->root_handle, 1);
1415 if (ret) {
1416 ERR("Error unref root handle");
1417 }
1418 sock_info->root_handle = -1;
1419 }
1420 sock_info->registration_done = 0;
1421 sock_info->initial_statedump_done = 0;
1422
1423 /*
1424 * wait_shm_mmap, socket and notify socket are used by listener
1425 * threads outside of the ust lock, so we cannot tear them down
1426 * ourselves, because we cannot join on these threads. Leave
1427 * responsibility of cleaning up these resources to the OS
1428 * process exit.
1429 */
1430 if (exiting)
1431 return;
1432
1433 if (sock_info->socket != -1) {
1434 ret = ustcomm_close_unix_sock(sock_info->socket);
1435 if (ret) {
1436 ERR("Error closing ust cmd socket");
1437 }
1438 sock_info->socket = -1;
1439 }
1440 if (sock_info->notify_socket != -1) {
1441 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1442 if (ret) {
1443 ERR("Error closing ust notify socket");
1444 }
1445 sock_info->notify_socket = -1;
1446 }
1447 if (sock_info->wait_shm_mmap) {
1448 long page_size;
1449
1450 page_size = LTTNG_UST_PAGE_SIZE;
1451 if (page_size <= 0) {
1452 if (!page_size) {
1453 errno = EINVAL;
1454 }
1455 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1456 } else {
1457 ret = munmap(sock_info->wait_shm_mmap, page_size);
1458 if (ret) {
1459 ERR("Error unmapping wait shm");
1460 }
1461 }
1462 sock_info->wait_shm_mmap = NULL;
1463 }
1464 }
1465
1466 /*
1467 * Using fork to set umask in the child process (not multi-thread safe).
1468 * We deal with the shm_open vs ftruncate race (happening when the
1469 * sessiond owns the shm and does not let everybody modify it, to ensure
1470 * safety against shm_unlink) by simply letting the mmap fail and
1471 * retrying after a few seconds.
1472 * For global shm, everybody has rw access to it until the sessiond
1473 * starts.
1474 */
1475 static
1476 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1477 {
1478 int wait_shm_fd, ret;
1479 pid_t pid;
1480
1481 /*
1482 * Try to open read-only.
1483 */
1484 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1485 if (wait_shm_fd >= 0) {
1486 int32_t tmp_read;
1487 ssize_t len;
1488 size_t bytes_read = 0;
1489
1490 /*
1491 * Try to read the fd. If unable to do so, try opening
1492 * it in write mode.
1493 */
1494 do {
1495 len = read(wait_shm_fd,
1496 &((char *) &tmp_read)[bytes_read],
1497 sizeof(tmp_read) - bytes_read);
1498 if (len > 0) {
1499 bytes_read += len;
1500 }
1501 } while ((len < 0 && errno == EINTR)
1502 || (len > 0 && bytes_read < sizeof(tmp_read)));
1503 if (bytes_read != sizeof(tmp_read)) {
1504 ret = close(wait_shm_fd);
1505 if (ret) {
1506 ERR("close wait_shm_fd");
1507 }
1508 goto open_write;
1509 }
1510 goto end;
1511 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1512 /*
1513 * Real-only open did not work, and it's not because the
1514 * entry was not present. It's a failure that prohibits
1515 * using shm.
1516 */
1517 ERR("Error opening shm %s", sock_info->wait_shm_path);
1518 goto end;
1519 }
1520
1521 open_write:
1522 /*
1523 * If the open failed because the file did not exist, or because
1524 * the file was not truncated yet, try creating it ourself.
1525 */
1526 URCU_TLS(lttng_ust_nest_count)++;
1527 pid = fork();
1528 URCU_TLS(lttng_ust_nest_count)--;
1529 if (pid > 0) {
1530 int status;
1531
1532 /*
1533 * Parent: wait for child to return, in which case the
1534 * shared memory map will have been created.
1535 */
1536 pid = wait(&status);
1537 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1538 wait_shm_fd = -1;
1539 goto end;
1540 }
1541 /*
1542 * Try to open read-only again after creation.
1543 */
1544 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1545 if (wait_shm_fd < 0) {
1546 /*
1547 * Real-only open did not work. It's a failure
1548 * that prohibits using shm.
1549 */
1550 ERR("Error opening shm %s", sock_info->wait_shm_path);
1551 goto end;
1552 }
1553 goto end;
1554 } else if (pid == 0) {
1555 int create_mode;
1556
1557 /* Child */
1558 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1559 if (sock_info->global)
1560 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1561 /*
1562 * We're alone in a child process, so we can modify the
1563 * process-wide umask.
1564 */
1565 umask(~create_mode);
1566 /*
1567 * Try creating shm (or get rw access).
1568 * We don't do an exclusive open, because we allow other
1569 * processes to create+ftruncate it concurrently.
1570 */
1571 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1572 O_RDWR | O_CREAT, create_mode);
1573 if (wait_shm_fd >= 0) {
1574 ret = ftruncate(wait_shm_fd, mmap_size);
1575 if (ret) {
1576 PERROR("ftruncate");
1577 _exit(EXIT_FAILURE);
1578 }
1579 _exit(EXIT_SUCCESS);
1580 }
1581 /*
1582 * For local shm, we need to have rw access to accept
1583 * opening it: this means the local sessiond will be
1584 * able to wake us up. For global shm, we open it even
1585 * if rw access is not granted, because the root.root
1586 * sessiond will be able to override all rights and wake
1587 * us up.
1588 */
1589 if (!sock_info->global && errno != EACCES) {
1590 ERR("Error opening shm %s", sock_info->wait_shm_path);
1591 _exit(EXIT_FAILURE);
1592 }
1593 /*
1594 * The shm exists, but we cannot open it RW. Report
1595 * success.
1596 */
1597 _exit(EXIT_SUCCESS);
1598 } else {
1599 return -1;
1600 }
1601 end:
1602 if (wait_shm_fd >= 0 && !sock_info->global) {
1603 struct stat statbuf;
1604
1605 /*
1606 * Ensure that our user is the owner of the shm file for
1607 * local shm. If we do not own the file, it means our
1608 * sessiond will not have access to wake us up (there is
1609 * probably a rogue process trying to fake our
1610 * sessiond). Fallback to polling method in this case.
1611 */
1612 ret = fstat(wait_shm_fd, &statbuf);
1613 if (ret) {
1614 PERROR("fstat");
1615 goto error_close;
1616 }
1617 if (statbuf.st_uid != getuid())
1618 goto error_close;
1619 }
1620 return wait_shm_fd;
1621
1622 error_close:
1623 ret = close(wait_shm_fd);
1624 if (ret) {
1625 PERROR("Error closing fd");
1626 }
1627 return -1;
1628 }
1629
1630 static
1631 char *get_map_shm(struct sock_info *sock_info)
1632 {
1633 long page_size;
1634 int wait_shm_fd, ret;
1635 char *wait_shm_mmap;
1636
1637 page_size = sysconf(_SC_PAGE_SIZE);
1638 if (page_size <= 0) {
1639 if (!page_size) {
1640 errno = EINVAL;
1641 }
1642 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1643 goto error;
1644 }
1645
1646 lttng_ust_lock_fd_tracker();
1647 wait_shm_fd = get_wait_shm(sock_info, page_size);
1648 if (wait_shm_fd < 0) {
1649 lttng_ust_unlock_fd_tracker();
1650 goto error;
1651 }
1652
1653 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1654 if (ret < 0) {
1655 ret = close(wait_shm_fd);
1656 if (!ret) {
1657 PERROR("Error closing fd");
1658 }
1659 lttng_ust_unlock_fd_tracker();
1660 goto error;
1661 }
1662
1663 wait_shm_fd = ret;
1664 lttng_ust_unlock_fd_tracker();
1665
1666 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1667 MAP_SHARED, wait_shm_fd, 0);
1668
1669 /* close shm fd immediately after taking the mmap reference */
1670 lttng_ust_lock_fd_tracker();
1671 ret = close(wait_shm_fd);
1672 if (!ret) {
1673 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1674 } else {
1675 PERROR("Error closing fd");
1676 }
1677 lttng_ust_unlock_fd_tracker();
1678
1679 if (wait_shm_mmap == MAP_FAILED) {
1680 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1681 goto error;
1682 }
1683 return wait_shm_mmap;
1684
1685 error:
1686 return NULL;
1687 }
1688
1689 static
1690 void wait_for_sessiond(struct sock_info *sock_info)
1691 {
1692 /* Use ust_lock to check if we should quit. */
1693 if (ust_lock()) {
1694 goto quit;
1695 }
1696 if (wait_poll_fallback) {
1697 goto error;
1698 }
1699 ust_unlock();
1700
1701 assert(sock_info->wait_shm_mmap);
1702
1703 DBG("Waiting for %s apps sessiond", sock_info->name);
1704 /* Wait for futex wakeup */
1705 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1706 goto end_wait;
1707
1708 while (lttng_ust_futex_async((int32_t *) sock_info->wait_shm_mmap,
1709 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1710 switch (errno) {
1711 case EWOULDBLOCK:
1712 /* Value already changed. */
1713 goto end_wait;
1714 case EINTR:
1715 /* Retry if interrupted by signal. */
1716 break; /* Get out of switch. */
1717 case EFAULT:
1718 wait_poll_fallback = 1;
1719 DBG(
1720 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1721 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1722 "Please upgrade your kernel "
1723 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1724 "mainline). LTTng-UST will use polling mode fallback.");
1725 if (ust_err_debug_enabled())
1726 PERROR("futex");
1727 goto end_wait;
1728 }
1729 }
1730 end_wait:
1731 return;
1732
1733 quit:
1734 ust_unlock();
1735 return;
1736
1737 error:
1738 ust_unlock();
1739 return;
1740 }
1741
1742 /*
1743 * This thread does not allocate any resource, except within
1744 * handle_message, within mutex protection. This mutex protects against
1745 * fork and exit.
1746 * The other moment it allocates resources is at socket connection, which
1747 * is also protected by the mutex.
1748 */
1749 static
1750 void *ust_listener_thread(void *arg)
1751 {
1752 struct sock_info *sock_info = arg;
1753 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1754 long timeout;
1755
1756 lttng_ust_fixup_tls();
1757 /*
1758 * If available, add '-ust' to the end of this thread's
1759 * process name
1760 */
1761 ret = lttng_ust_setustprocname();
1762 if (ret) {
1763 ERR("Unable to set UST process name");
1764 }
1765
1766 /* Restart trying to connect to the session daemon */
1767 restart:
1768 if (prev_connect_failed) {
1769 /* Wait for sessiond availability with pipe */
1770 wait_for_sessiond(sock_info);
1771 if (has_waited) {
1772 has_waited = 0;
1773 /*
1774 * Sleep for 5 seconds before retrying after a
1775 * sequence of failure / wait / failure. This
1776 * deals with a killed or broken session daemon.
1777 */
1778 sleep(5);
1779 } else {
1780 has_waited = 1;
1781 }
1782 prev_connect_failed = 0;
1783 }
1784
1785 if (ust_lock()) {
1786 goto quit;
1787 }
1788
1789 if (sock_info->socket != -1) {
1790 /* FD tracker is updated by ustcomm_close_unix_sock() */
1791 ret = ustcomm_close_unix_sock(sock_info->socket);
1792 if (ret) {
1793 ERR("Error closing %s ust cmd socket",
1794 sock_info->name);
1795 }
1796 sock_info->socket = -1;
1797 }
1798 if (sock_info->notify_socket != -1) {
1799 /* FD tracker is updated by ustcomm_close_unix_sock() */
1800 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1801 if (ret) {
1802 ERR("Error closing %s ust notify socket",
1803 sock_info->name);
1804 }
1805 sock_info->notify_socket = -1;
1806 }
1807
1808
1809 /*
1810 * Register. We need to perform both connect and sending
1811 * registration message before doing the next connect otherwise
1812 * we may reach unix socket connect queue max limits and block
1813 * on the 2nd connect while the session daemon is awaiting the
1814 * first connect registration message.
1815 */
1816 /* Connect cmd socket */
1817 lttng_ust_lock_fd_tracker();
1818 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1819 get_connect_sock_timeout());
1820 if (ret < 0) {
1821 lttng_ust_unlock_fd_tracker();
1822 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1823 prev_connect_failed = 1;
1824
1825 /*
1826 * If we cannot find the sessiond daemon, don't delay
1827 * constructor execution.
1828 */
1829 ret = handle_register_failed(sock_info);
1830 assert(!ret);
1831 ust_unlock();
1832 goto restart;
1833 }
1834 fd = ret;
1835 ret = lttng_ust_add_fd_to_tracker(fd);
1836 if (ret < 0) {
1837 ret = close(fd);
1838 if (ret) {
1839 PERROR("close on sock_info->socket");
1840 }
1841 ret = -1;
1842 lttng_ust_unlock_fd_tracker();
1843 ust_unlock();
1844 goto quit;
1845 }
1846
1847 sock_info->socket = ret;
1848 lttng_ust_unlock_fd_tracker();
1849
1850 ust_unlock();
1851 /*
1852 * Unlock/relock ust lock because connect is blocking (with
1853 * timeout). Don't delay constructors on the ust lock for too
1854 * long.
1855 */
1856 if (ust_lock()) {
1857 goto quit;
1858 }
1859
1860 /*
1861 * Create only one root handle per listener thread for the whole
1862 * process lifetime, so we ensure we get ID which is statically
1863 * assigned to the root handle.
1864 */
1865 if (sock_info->root_handle == -1) {
1866 ret = lttng_abi_create_root_handle();
1867 if (ret < 0) {
1868 ERR("Error creating root handle");
1869 goto quit;
1870 }
1871 sock_info->root_handle = ret;
1872 }
1873
1874 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1875 if (ret < 0) {
1876 ERR("Error registering to %s ust cmd socket",
1877 sock_info->name);
1878 prev_connect_failed = 1;
1879 /*
1880 * If we cannot register to the sessiond daemon, don't
1881 * delay constructor execution.
1882 */
1883 ret = handle_register_failed(sock_info);
1884 assert(!ret);
1885 ust_unlock();
1886 goto restart;
1887 }
1888
1889 ust_unlock();
1890 /*
1891 * Unlock/relock ust lock because connect is blocking (with
1892 * timeout). Don't delay constructors on the ust lock for too
1893 * long.
1894 */
1895 if (ust_lock()) {
1896 goto quit;
1897 }
1898
1899 /* Connect notify socket */
1900 lttng_ust_lock_fd_tracker();
1901 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1902 get_connect_sock_timeout());
1903 if (ret < 0) {
1904 lttng_ust_unlock_fd_tracker();
1905 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1906 prev_connect_failed = 1;
1907
1908 /*
1909 * If we cannot find the sessiond daemon, don't delay
1910 * constructor execution.
1911 */
1912 ret = handle_register_failed(sock_info);
1913 assert(!ret);
1914 ust_unlock();
1915 goto restart;
1916 }
1917
1918 fd = ret;
1919 ret = lttng_ust_add_fd_to_tracker(fd);
1920 if (ret < 0) {
1921 ret = close(fd);
1922 if (ret) {
1923 PERROR("close on sock_info->notify_socket");
1924 }
1925 ret = -1;
1926 lttng_ust_unlock_fd_tracker();
1927 ust_unlock();
1928 goto quit;
1929 }
1930
1931 sock_info->notify_socket = ret;
1932 lttng_ust_unlock_fd_tracker();
1933
1934 ust_unlock();
1935 /*
1936 * Unlock/relock ust lock because connect is blocking (with
1937 * timeout). Don't delay constructors on the ust lock for too
1938 * long.
1939 */
1940 if (ust_lock()) {
1941 goto quit;
1942 }
1943
1944 timeout = get_notify_sock_timeout();
1945 if (timeout >= 0) {
1946 /*
1947 * Give at least 10ms to sessiond to reply to
1948 * notifications.
1949 */
1950 if (timeout < 10)
1951 timeout = 10;
1952 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1953 timeout);
1954 if (ret < 0) {
1955 WARN("Error setting socket receive timeout");
1956 }
1957 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1958 timeout);
1959 if (ret < 0) {
1960 WARN("Error setting socket send timeout");
1961 }
1962 } else if (timeout < -1) {
1963 WARN("Unsupported timeout value %ld", timeout);
1964 }
1965
1966 ret = register_to_sessiond(sock_info->notify_socket,
1967 USTCTL_SOCKET_NOTIFY);
1968 if (ret < 0) {
1969 ERR("Error registering to %s ust notify socket",
1970 sock_info->name);
1971 prev_connect_failed = 1;
1972 /*
1973 * If we cannot register to the sessiond daemon, don't
1974 * delay constructor execution.
1975 */
1976 ret = handle_register_failed(sock_info);
1977 assert(!ret);
1978 ust_unlock();
1979 goto restart;
1980 }
1981 sock = sock_info->socket;
1982
1983 ust_unlock();
1984
1985 for (;;) {
1986 ssize_t len;
1987 struct ustcomm_ust_msg lum;
1988
1989 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1990 switch (len) {
1991 case 0: /* orderly shutdown */
1992 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1993 if (ust_lock()) {
1994 goto quit;
1995 }
1996 /*
1997 * Either sessiond has shutdown or refused us by closing the socket.
1998 * In either case, we don't want to delay construction execution,
1999 * and we need to wait before retry.
2000 */
2001 prev_connect_failed = 1;
2002 /*
2003 * If we cannot register to the sessiond daemon, don't
2004 * delay constructor execution.
2005 */
2006 ret = handle_register_failed(sock_info);
2007 assert(!ret);
2008 ust_unlock();
2009 goto end;
2010 case sizeof(lum):
2011 print_cmd(lum.cmd, lum.handle);
2012 ret = handle_message(sock_info, sock, &lum);
2013 if (ret) {
2014 ERR("Error handling message for %s socket",
2015 sock_info->name);
2016 /*
2017 * Close socket if protocol error is
2018 * detected.
2019 */
2020 goto end;
2021 }
2022 continue;
2023 default:
2024 if (len < 0) {
2025 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
2026 } else {
2027 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
2028 }
2029 if (len == -ECONNRESET) {
2030 DBG("%s remote end closed connection", sock_info->name);
2031 goto end;
2032 }
2033 goto end;
2034 }
2035
2036 }
2037 end:
2038 if (ust_lock()) {
2039 goto quit;
2040 }
2041 /* Cleanup socket handles before trying to reconnect */
2042 lttng_ust_abi_objd_table_owner_cleanup(sock_info);
2043 ust_unlock();
2044 goto restart; /* try to reconnect */
2045
2046 quit:
2047 ust_unlock();
2048
2049 pthread_mutex_lock(&ust_exit_mutex);
2050 sock_info->thread_active = 0;
2051 pthread_mutex_unlock(&ust_exit_mutex);
2052 return NULL;
2053 }
2054
2055 /*
2056 * Weak symbol to call when the ust malloc wrapper is not loaded.
2057 */
2058 __attribute__((weak))
2059 void lttng_ust_libc_wrapper_malloc_init(void)
2060 {
2061 }
2062
2063 /*
2064 * sessiond monitoring thread: monitor presence of global and per-user
2065 * sessiond by polling the application common named pipe.
2066 */
2067 void __attribute__((constructor)) lttng_ust_init(void)
2068 {
2069 struct timespec constructor_timeout;
2070 sigset_t sig_all_blocked, orig_parent_mask;
2071 pthread_attr_t thread_attr;
2072 int timeout_mode;
2073 int ret;
2074 void *handle;
2075
2076 if (uatomic_xchg(&initialized, 1) == 1)
2077 return;
2078
2079 /*
2080 * Fixup interdependency between TLS fixup mutex (which happens
2081 * to be the dynamic linker mutex) and ust_lock, taken within
2082 * the ust lock.
2083 */
2084 lttng_ust_fixup_tls();
2085
2086 lttng_ust_loaded = 1;
2087
2088 /*
2089 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2090 * the unloading of code used by the ust_listener_threads as we can not
2091 * reliably know when they exited. To do that, manually load
2092 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2093 * this library so it never becomes zero, thus never gets unloaded from the
2094 * address space of the process. Since we are already running in the
2095 * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
2096 * simply increment the refcount and no additionnal work is needed by the
2097 * dynamic loader as the shared library is already loaded in the address
2098 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2099 * unloading of the UST library if its refcount becomes zero (which should
2100 * never happen). Do the return value check but discard the handle at the
2101 * end of the function as it's not needed.
2102 */
2103 handle = dlopen(LTTNG_UST_LIB_SO_NAME, RTLD_LAZY | RTLD_NODELETE);
2104 if (!handle) {
2105 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME);
2106 }
2107
2108 /*
2109 * We want precise control over the order in which we construct
2110 * our sub-libraries vs starting to receive commands from
2111 * sessiond (otherwise leading to errors when trying to create
2112 * sessiond before the init functions are completed).
2113 */
2114 ust_err_init();
2115 lttng_ust_getenv_init(); /* Needs ust_err_init() to be completed. */
2116 lttng_ust_tp_init();
2117 lttng_ust_init_fd_tracker();
2118 lttng_ust_clock_init();
2119 lttng_ust_getcpu_init();
2120 lttng_ust_statedump_init();
2121 lttng_ring_buffer_metadata_client_init();
2122 lttng_ring_buffer_client_overwrite_init();
2123 lttng_ring_buffer_client_overwrite_rt_init();
2124 lttng_ring_buffer_client_discard_init();
2125 lttng_ring_buffer_client_discard_rt_init();
2126 lttng_counter_client_percpu_32_modular_init();
2127 lttng_counter_client_percpu_64_modular_init();
2128 lttng_perf_counter_init();
2129 /*
2130 * Invoke ust malloc wrapper init before starting other threads.
2131 */
2132 lttng_ust_libc_wrapper_malloc_init();
2133
2134 timeout_mode = get_constructor_timeout(&constructor_timeout);
2135
2136 get_allow_blocking();
2137
2138 ret = sem_init(&constructor_wait, 0, 0);
2139 if (ret) {
2140 PERROR("sem_init");
2141 }
2142
2143 ret = setup_global_apps();
2144 if (ret) {
2145 assert(global_apps.allowed == 0);
2146 DBG("global apps setup returned %d", ret);
2147 }
2148
2149 ret = setup_local_apps();
2150 if (ret) {
2151 assert(local_apps.allowed == 0);
2152 DBG("local apps setup returned %d", ret);
2153 }
2154
2155 /* A new thread created by pthread_create inherits the signal mask
2156 * from the parent. To avoid any signal being received by the
2157 * listener thread, we block all signals temporarily in the parent,
2158 * while we create the listener thread.
2159 */
2160 sigfillset(&sig_all_blocked);
2161 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
2162 if (ret) {
2163 ERR("pthread_sigmask: %s", strerror(ret));
2164 }
2165
2166 ret = pthread_attr_init(&thread_attr);
2167 if (ret) {
2168 ERR("pthread_attr_init: %s", strerror(ret));
2169 }
2170 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
2171 if (ret) {
2172 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
2173 }
2174
2175 if (global_apps.allowed) {
2176 pthread_mutex_lock(&ust_exit_mutex);
2177 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
2178 ust_listener_thread, &global_apps);
2179 if (ret) {
2180 ERR("pthread_create global: %s", strerror(ret));
2181 }
2182 global_apps.thread_active = 1;
2183 pthread_mutex_unlock(&ust_exit_mutex);
2184 } else {
2185 handle_register_done(&global_apps);
2186 }
2187
2188 if (local_apps.allowed) {
2189 pthread_mutex_lock(&ust_exit_mutex);
2190 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
2191 ust_listener_thread, &local_apps);
2192 if (ret) {
2193 ERR("pthread_create local: %s", strerror(ret));
2194 }
2195 local_apps.thread_active = 1;
2196 pthread_mutex_unlock(&ust_exit_mutex);
2197 } else {
2198 handle_register_done(&local_apps);
2199 }
2200 ret = pthread_attr_destroy(&thread_attr);
2201 if (ret) {
2202 ERR("pthread_attr_destroy: %s", strerror(ret));
2203 }
2204
2205 /* Restore original signal mask in parent */
2206 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
2207 if (ret) {
2208 ERR("pthread_sigmask: %s", strerror(ret));
2209 }
2210
2211 switch (timeout_mode) {
2212 case 1: /* timeout wait */
2213 do {
2214 ret = sem_timedwait(&constructor_wait,
2215 &constructor_timeout);
2216 } while (ret < 0 && errno == EINTR);
2217 if (ret < 0) {
2218 switch (errno) {
2219 case ETIMEDOUT:
2220 ERR("Timed out waiting for lttng-sessiond");
2221 break;
2222 case EINVAL:
2223 PERROR("sem_timedwait");
2224 break;
2225 default:
2226 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2227 strerror(errno));
2228 }
2229 }
2230 break;
2231 case -1:/* wait forever */
2232 do {
2233 ret = sem_wait(&constructor_wait);
2234 } while (ret < 0 && errno == EINTR);
2235 if (ret < 0) {
2236 switch (errno) {
2237 case EINVAL:
2238 PERROR("sem_wait");
2239 break;
2240 default:
2241 ERR("Unexpected error \"%s\" returned by sem_wait",
2242 strerror(errno));
2243 }
2244 }
2245 break;
2246 case 0: /* no timeout */
2247 break;
2248 }
2249 }
2250
2251 static
2252 void lttng_ust_cleanup(int exiting)
2253 {
2254 cleanup_sock_info(&global_apps, exiting);
2255 cleanup_sock_info(&local_apps, exiting);
2256 local_apps.allowed = 0;
2257 global_apps.allowed = 0;
2258 /*
2259 * The teardown in this function all affect data structures
2260 * accessed under the UST lock by the listener thread. This
2261 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2262 * that none of these threads are accessing this data at this
2263 * point.
2264 */
2265 lttng_ust_abi_exit();
2266 lttng_ust_abi_events_exit();
2267 lttng_perf_counter_exit();
2268 lttng_ring_buffer_client_discard_rt_exit();
2269 lttng_ring_buffer_client_discard_exit();
2270 lttng_ring_buffer_client_overwrite_rt_exit();
2271 lttng_ring_buffer_client_overwrite_exit();
2272 lttng_ring_buffer_metadata_client_exit();
2273 lttng_counter_client_percpu_32_modular_exit();
2274 lttng_counter_client_percpu_64_modular_exit();
2275 lttng_ust_statedump_destroy();
2276 lttng_ust_tp_exit();
2277 if (!exiting) {
2278 /* Reinitialize values for fork */
2279 sem_count = sem_count_initial_value;
2280 lttng_ust_comm_should_quit = 0;
2281 initialized = 0;
2282 }
2283 }
2284
2285 void __attribute__((destructor)) lttng_ust_exit(void)
2286 {
2287 int ret;
2288
2289 /*
2290 * Using pthread_cancel here because:
2291 * A) we don't want to hang application teardown.
2292 * B) the thread is not allocating any resource.
2293 */
2294
2295 /*
2296 * Require the communication thread to quit. Synchronize with
2297 * mutexes to ensure it is not in a mutex critical section when
2298 * pthread_cancel is later called.
2299 */
2300 ust_lock_nocheck();
2301 lttng_ust_comm_should_quit = 1;
2302 ust_unlock();
2303
2304 pthread_mutex_lock(&ust_exit_mutex);
2305 /* cancel threads */
2306 if (global_apps.thread_active) {
2307 ret = pthread_cancel(global_apps.ust_listener);
2308 if (ret) {
2309 ERR("Error cancelling global ust listener thread: %s",
2310 strerror(ret));
2311 } else {
2312 global_apps.thread_active = 0;
2313 }
2314 }
2315 if (local_apps.thread_active) {
2316 ret = pthread_cancel(local_apps.ust_listener);
2317 if (ret) {
2318 ERR("Error cancelling local ust listener thread: %s",
2319 strerror(ret));
2320 } else {
2321 local_apps.thread_active = 0;
2322 }
2323 }
2324 pthread_mutex_unlock(&ust_exit_mutex);
2325
2326 /*
2327 * Do NOT join threads: use of sys_futex makes it impossible to
2328 * join the threads without using async-cancel, but async-cancel
2329 * is delivered by a signal, which could hit the target thread
2330 * anywhere in its code path, including while the ust_lock() is
2331 * held, causing a deadlock for the other thread. Let the OS
2332 * cleanup the threads if there are stalled in a syscall.
2333 */
2334 lttng_ust_cleanup(1);
2335 }
2336
2337 static
2338 void ust_context_ns_reset(void)
2339 {
2340 lttng_context_pid_ns_reset();
2341 lttng_context_cgroup_ns_reset();
2342 lttng_context_ipc_ns_reset();
2343 lttng_context_mnt_ns_reset();
2344 lttng_context_net_ns_reset();
2345 lttng_context_user_ns_reset();
2346 lttng_context_time_ns_reset();
2347 lttng_context_uts_ns_reset();
2348 }
2349
2350 static
2351 void ust_context_vuids_reset(void)
2352 {
2353 lttng_context_vuid_reset();
2354 lttng_context_veuid_reset();
2355 lttng_context_vsuid_reset();
2356 }
2357
2358 static
2359 void ust_context_vgids_reset(void)
2360 {
2361 lttng_context_vgid_reset();
2362 lttng_context_vegid_reset();
2363 lttng_context_vsgid_reset();
2364 }
2365
2366 /*
2367 * We exclude the worker threads across fork and clone (except
2368 * CLONE_VM), because these system calls only keep the forking thread
2369 * running in the child. Therefore, we don't want to call fork or clone
2370 * in the middle of an tracepoint or ust tracing state modification.
2371 * Holding this mutex protects these structures across fork and clone.
2372 */
2373 void ust_before_fork(sigset_t *save_sigset)
2374 {
2375 /*
2376 * Disable signals. This is to avoid that the child intervenes
2377 * before it is properly setup for tracing. It is safer to
2378 * disable all signals, because then we know we are not breaking
2379 * anything by restoring the original mask.
2380 */
2381 sigset_t all_sigs;
2382 int ret;
2383
2384 /* Fixup lttng-ust TLS. */
2385 lttng_ust_fixup_tls();
2386
2387 if (URCU_TLS(lttng_ust_nest_count))
2388 return;
2389 /* Disable signals */
2390 sigfillset(&all_sigs);
2391 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2392 if (ret == -1) {
2393 PERROR("sigprocmask");
2394 }
2395
2396 pthread_mutex_lock(&ust_fork_mutex);
2397
2398 ust_lock_nocheck();
2399 lttng_ust_urcu_before_fork();
2400 lttng_ust_lock_fd_tracker();
2401 lttng_perf_lock();
2402 }
2403
2404 static void ust_after_fork_common(sigset_t *restore_sigset)
2405 {
2406 int ret;
2407
2408 DBG("process %d", getpid());
2409 lttng_perf_unlock();
2410 lttng_ust_unlock_fd_tracker();
2411 ust_unlock();
2412
2413 pthread_mutex_unlock(&ust_fork_mutex);
2414
2415 /* Restore signals */
2416 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2417 if (ret == -1) {
2418 PERROR("sigprocmask");
2419 }
2420 }
2421
2422 void ust_after_fork_parent(sigset_t *restore_sigset)
2423 {
2424 if (URCU_TLS(lttng_ust_nest_count))
2425 return;
2426 DBG("process %d", getpid());
2427 lttng_ust_urcu_after_fork_parent();
2428 /* Release mutexes and reenable signals */
2429 ust_after_fork_common(restore_sigset);
2430 }
2431
2432 /*
2433 * After fork, in the child, we need to cleanup all the leftover state,
2434 * except the worker thread which already magically disappeared thanks
2435 * to the weird Linux fork semantics. After tyding up, we call
2436 * lttng_ust_init() again to start over as a new PID.
2437 *
2438 * This is meant for forks() that have tracing in the child between the
2439 * fork and following exec call (if there is any).
2440 */
2441 void ust_after_fork_child(sigset_t *restore_sigset)
2442 {
2443 if (URCU_TLS(lttng_ust_nest_count))
2444 return;
2445 lttng_context_vpid_reset();
2446 lttng_context_vtid_reset();
2447 lttng_ust_context_procname_reset();
2448 ust_context_ns_reset();
2449 ust_context_vuids_reset();
2450 ust_context_vgids_reset();
2451 DBG("process %d", getpid());
2452 /* Release urcu mutexes */
2453 lttng_ust_urcu_after_fork_child();
2454 lttng_ust_cleanup(0);
2455 /* Release mutexes and reenable signals */
2456 ust_after_fork_common(restore_sigset);
2457 lttng_ust_init();
2458 }
2459
2460 void ust_after_setns(void)
2461 {
2462 ust_context_ns_reset();
2463 ust_context_vuids_reset();
2464 ust_context_vgids_reset();
2465 }
2466
2467 void ust_after_unshare(void)
2468 {
2469 ust_context_ns_reset();
2470 ust_context_vuids_reset();
2471 ust_context_vgids_reset();
2472 }
2473
2474 void ust_after_setuid(void)
2475 {
2476 ust_context_vuids_reset();
2477 }
2478
2479 void ust_after_seteuid(void)
2480 {
2481 ust_context_vuids_reset();
2482 }
2483
2484 void ust_after_setreuid(void)
2485 {
2486 ust_context_vuids_reset();
2487 }
2488
2489 void ust_after_setresuid(void)
2490 {
2491 ust_context_vuids_reset();
2492 }
2493
2494 void ust_after_setgid(void)
2495 {
2496 ust_context_vgids_reset();
2497 }
2498
2499 void ust_after_setegid(void)
2500 {
2501 ust_context_vgids_reset();
2502 }
2503
2504 void ust_after_setregid(void)
2505 {
2506 ust_context_vgids_reset();
2507 }
2508
2509 void ust_after_setresgid(void)
2510 {
2511 ust_context_vgids_reset();
2512 }
2513
2514 void lttng_ust_sockinfo_session_enabled(void *owner)
2515 {
2516 struct sock_info *sock_info = owner;
2517 sock_info->statedump_pending = 1;
2518 }
This page took 0.076532 seconds and 5 git commands to generate.