New API: lttng_ust_init_thread() for async-signal tracing
[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-utils.h>
33 #include <lttng/ust-events.h>
34 #include <lttng/ust-abi.h>
35 #include <lttng/ust-fork.h>
36 #include <lttng/ust-error.h>
37 #include <lttng/ust-ctl.h>
38 #include <lttng/ust-libc-wrapper.h>
39 #include <lttng/ust-tracer.h>
40 #include <urcu/tls-compat.h>
41 #include <ust-comm.h>
42 #include <ust-fd.h>
43 #include <usterr-signal-safe.h>
44 #include <ust-helper.h>
45 #include "tracepoint-internal.h"
46 #include "lttng-tracer-core.h"
47 #include "compat.h"
48 #include "../libringbuffer/rb-init.h"
49 #include "lttng-ust-statedump.h"
50 #include "clock.h"
51 #include "../libringbuffer/getcpu.h"
52 #include "getenv.h"
53 #include "ust-events-internal.h"
54 #include "context-internal.h"
55 #include "ust-compat.h"
56 #include "lttng-counter-client.h"
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 static char *get_map_shm(struct sock_info *sock_info);
355
356 ssize_t lttng_ust_read(int fd, void *buf, size_t len)
357 {
358 ssize_t ret;
359 size_t copied = 0, to_copy = len;
360
361 do {
362 ret = read(fd, buf + copied, to_copy);
363 if (ret > 0) {
364 copied += ret;
365 to_copy -= ret;
366 }
367 } while ((ret > 0 && to_copy > 0)
368 || (ret < 0 && errno == EINTR));
369 if (ret > 0) {
370 ret = copied;
371 }
372 return ret;
373 }
374 /*
375 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
376 * pointer.
377 */
378 static
379 const char *get_lttng_home_dir(void)
380 {
381 const char *val;
382
383 val = (const char *) lttng_ust_getenv("LTTNG_HOME");
384 if (val != NULL) {
385 return val;
386 }
387 return (const char *) lttng_ust_getenv("HOME");
388 }
389
390 /*
391 * Force a read (imply TLS fixup for dlopen) of TLS variables.
392 */
393 static
394 void lttng_fixup_nest_count_tls(void)
395 {
396 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
397 }
398
399 static
400 void lttng_fixup_ust_mutex_nest_tls(void)
401 {
402 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
403 }
404
405 /*
406 * Fixup lttng-ust urcu TLS.
407 */
408 static
409 void lttng_fixup_lttng_ust_urcu_tls(void)
410 {
411 (void) lttng_ust_urcu_read_ongoing();
412 }
413
414 void lttng_ust_fixup_tls(void)
415 {
416 lttng_fixup_lttng_ust_urcu_tls();
417 lttng_fixup_ringbuffer_tls();
418 lttng_fixup_vtid_tls();
419 lttng_fixup_nest_count_tls();
420 lttng_fixup_procname_tls();
421 lttng_fixup_ust_mutex_nest_tls();
422 lttng_ust_fixup_perf_counter_tls();
423 lttng_ust_fixup_fd_tracker_tls();
424 lttng_fixup_cgroup_ns_tls();
425 lttng_fixup_ipc_ns_tls();
426 lttng_fixup_net_ns_tls();
427 lttng_fixup_time_ns_tls();
428 lttng_fixup_uts_ns_tls();
429 lttng_ust_fixup_ring_buffer_client_discard_tls();
430 lttng_ust_fixup_ring_buffer_client_discard_rt_tls();
431 lttng_ust_fixup_ring_buffer_client_overwrite_tls();
432 lttng_ust_fixup_ring_buffer_client_overwrite_rt_tls();
433 }
434
435 /*
436 * LTTng-UST uses Global Dynamic model TLS variables rather than IE
437 * model because many versions of glibc don't preallocate a pool large
438 * enough for TLS variables IE model defined in other shared libraries,
439 * and causes issues when using LTTng-UST for Java tracing.
440 *
441 * Because of this use of Global Dynamic TLS variables, users wishing to
442 * trace from signal handlers need to explicitly trigger the lazy
443 * allocation of those variables for each thread before using them.
444 * This can be triggered by calling lttng_ust_init_thread().
445 */
446 void lttng_ust_init_thread(void)
447 {
448 /*
449 * Because those TLS variables are global dynamic, we need to
450 * ensure those are initialized before a signal handler nesting over
451 * this thread attempts to use them.
452 */
453 lttng_ust_fixup_tls();
454 }
455
456 int lttng_get_notify_socket(void *owner)
457 {
458 struct sock_info *info = owner;
459
460 return info->notify_socket;
461 }
462
463
464 char* lttng_ust_sockinfo_get_procname(void *owner)
465 {
466 struct sock_info *info = owner;
467
468 return info->procname;
469 }
470
471 static
472 void print_cmd(int cmd, int handle)
473 {
474 const char *cmd_name = "Unknown";
475
476 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
477 && cmd_name_mapping[cmd]) {
478 cmd_name = cmd_name_mapping[cmd];
479 }
480 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
481 cmd_name, cmd,
482 lttng_ust_obj_get_name(handle), handle);
483 }
484
485 static
486 int setup_global_apps(void)
487 {
488 int ret = 0;
489 assert(!global_apps.wait_shm_mmap);
490
491 global_apps.wait_shm_mmap = get_map_shm(&global_apps);
492 if (!global_apps.wait_shm_mmap) {
493 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
494 global_apps.allowed = 0;
495 ret = -EIO;
496 goto error;
497 }
498
499 global_apps.allowed = 1;
500 lttng_pthread_getname_np(global_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
501 error:
502 return ret;
503 }
504 static
505 int setup_local_apps(void)
506 {
507 int ret = 0;
508 const char *home_dir;
509 uid_t uid;
510
511 assert(!local_apps.wait_shm_mmap);
512
513 uid = getuid();
514 /*
515 * Disallow per-user tracing for setuid binaries.
516 */
517 if (uid != geteuid()) {
518 assert(local_apps.allowed == 0);
519 ret = 0;
520 goto end;
521 }
522 home_dir = get_lttng_home_dir();
523 if (!home_dir) {
524 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
525 assert(local_apps.allowed == 0);
526 ret = -ENOENT;
527 goto end;
528 }
529 local_apps.allowed = 1;
530 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
531 home_dir,
532 LTTNG_DEFAULT_HOME_RUNDIR,
533 LTTNG_UST_SOCK_FILENAME);
534 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
535 LTTNG_UST_WAIT_FILENAME,
536 uid);
537
538 local_apps.wait_shm_mmap = get_map_shm(&local_apps);
539 if (!local_apps.wait_shm_mmap) {
540 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
541 local_apps.allowed = 0;
542 ret = -EIO;
543 goto end;
544 }
545
546 lttng_pthread_getname_np(local_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
547 end:
548 return ret;
549 }
550
551 /*
552 * Get socket timeout, in ms.
553 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
554 */
555 static
556 long get_timeout(void)
557 {
558 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
559
560 if (!got_timeout_env) {
561 str_timeout = lttng_ust_getenv("LTTNG_UST_REGISTER_TIMEOUT");
562 got_timeout_env = 1;
563 }
564 if (str_timeout)
565 constructor_delay_ms = strtol(str_timeout, NULL, 10);
566 /* All negative values are considered as "-1". */
567 if (constructor_delay_ms < -1)
568 constructor_delay_ms = -1;
569 return constructor_delay_ms;
570 }
571
572 /* Timeout for notify socket send and recv. */
573 static
574 long get_notify_sock_timeout(void)
575 {
576 return get_timeout();
577 }
578
579 /* Timeout for connecting to cmd and notify sockets. */
580 static
581 long get_connect_sock_timeout(void)
582 {
583 return get_timeout();
584 }
585
586 /*
587 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
588 */
589 static
590 int get_constructor_timeout(struct timespec *constructor_timeout)
591 {
592 long constructor_delay_ms;
593 int ret;
594
595 constructor_delay_ms = get_timeout();
596
597 switch (constructor_delay_ms) {
598 case -1:/* fall-through */
599 case 0:
600 return constructor_delay_ms;
601 default:
602 break;
603 }
604
605 /*
606 * If we are unable to find the current time, don't wait.
607 */
608 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
609 if (ret) {
610 /* Don't wait. */
611 return 0;
612 }
613 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
614 constructor_timeout->tv_nsec +=
615 (constructor_delay_ms % 1000UL) * 1000000UL;
616 if (constructor_timeout->tv_nsec >= 1000000000UL) {
617 constructor_timeout->tv_sec++;
618 constructor_timeout->tv_nsec -= 1000000000UL;
619 }
620 /* Timeout wait (constructor_delay_ms). */
621 return 1;
622 }
623
624 static
625 void get_allow_blocking(void)
626 {
627 const char *str_allow_blocking =
628 lttng_ust_getenv("LTTNG_UST_ALLOW_BLOCKING");
629
630 if (str_allow_blocking) {
631 DBG("%s environment variable is set",
632 "LTTNG_UST_ALLOW_BLOCKING");
633 lttng_ust_ringbuffer_set_allow_blocking();
634 }
635 }
636
637 static
638 int register_to_sessiond(int socket, enum ustctl_socket_type type)
639 {
640 return ustcomm_send_reg_msg(socket,
641 type,
642 CAA_BITS_PER_LONG,
643 lttng_ust_rb_alignof(uint8_t) * CHAR_BIT,
644 lttng_ust_rb_alignof(uint16_t) * CHAR_BIT,
645 lttng_ust_rb_alignof(uint32_t) * CHAR_BIT,
646 lttng_ust_rb_alignof(uint64_t) * CHAR_BIT,
647 lttng_ust_rb_alignof(unsigned long) * CHAR_BIT);
648 }
649
650 static
651 int send_reply(int sock, struct ustcomm_ust_reply *lur)
652 {
653 ssize_t len;
654
655 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
656 switch (len) {
657 case sizeof(*lur):
658 DBG("message successfully sent");
659 return 0;
660 default:
661 if (len == -ECONNRESET) {
662 DBG("remote end closed connection");
663 return 0;
664 }
665 if (len < 0)
666 return len;
667 DBG("incorrect message size: %zd", len);
668 return -EINVAL;
669 }
670 }
671
672 static
673 void decrement_sem_count(unsigned int count)
674 {
675 int ret;
676
677 assert(uatomic_read(&sem_count) >= count);
678
679 if (uatomic_read(&sem_count) <= 0) {
680 return;
681 }
682
683 ret = uatomic_add_return(&sem_count, -count);
684 if (ret == 0) {
685 ret = sem_post(&constructor_wait);
686 assert(!ret);
687 }
688 }
689
690 static
691 int handle_register_done(struct sock_info *sock_info)
692 {
693 if (sock_info->registration_done)
694 return 0;
695 sock_info->registration_done = 1;
696
697 decrement_sem_count(1);
698 if (!sock_info->statedump_pending) {
699 sock_info->initial_statedump_done = 1;
700 decrement_sem_count(1);
701 }
702
703 return 0;
704 }
705
706 static
707 int handle_register_failed(struct sock_info *sock_info)
708 {
709 if (sock_info->registration_done)
710 return 0;
711 sock_info->registration_done = 1;
712 sock_info->initial_statedump_done = 1;
713
714 decrement_sem_count(2);
715
716 return 0;
717 }
718
719 /*
720 * Only execute pending statedump after the constructor semaphore has
721 * been posted by the current listener thread. This means statedump will
722 * only be performed after the "registration done" command is received
723 * from this thread's session daemon.
724 *
725 * This ensures we don't run into deadlock issues with the dynamic
726 * loader mutex, which is held while the constructor is called and
727 * waiting on the constructor semaphore. All operations requiring this
728 * dynamic loader lock need to be postponed using this mechanism.
729 *
730 * In a scenario with two session daemons connected to the application,
731 * it is possible that the first listener thread which receives the
732 * registration done command issues its statedump while the dynamic
733 * loader lock is still held by the application constructor waiting on
734 * the semaphore. It will however be allowed to proceed when the
735 * second session daemon sends the registration done command to the
736 * second listener thread. This situation therefore does not produce
737 * a deadlock.
738 */
739 static
740 void handle_pending_statedump(struct sock_info *sock_info)
741 {
742 if (sock_info->registration_done && sock_info->statedump_pending) {
743 sock_info->statedump_pending = 0;
744 pthread_mutex_lock(&ust_fork_mutex);
745 lttng_handle_pending_statedump(sock_info);
746 pthread_mutex_unlock(&ust_fork_mutex);
747
748 if (!sock_info->initial_statedump_done) {
749 sock_info->initial_statedump_done = 1;
750 decrement_sem_count(1);
751 }
752 }
753 }
754
755 static inline
756 const char *bytecode_type_str(uint32_t cmd)
757 {
758 switch (cmd) {
759 case LTTNG_UST_ABI_CAPTURE:
760 return "capture";
761 case LTTNG_UST_ABI_FILTER:
762 return "filter";
763 default:
764 abort();
765 }
766 }
767
768 static
769 int handle_bytecode_recv(struct sock_info *sock_info,
770 int sock, struct ustcomm_ust_msg *lum)
771 {
772 struct lttng_ust_bytecode_node *bytecode = NULL;
773 enum lttng_ust_bytecode_type type;
774 const struct lttng_ust_abi_objd_ops *ops;
775 uint32_t data_size, data_size_max, reloc_offset;
776 uint64_t seqnum;
777 ssize_t len;
778 int ret = 0;
779
780 switch (lum->cmd) {
781 case LTTNG_UST_ABI_FILTER:
782 type = LTTNG_UST_BYTECODE_TYPE_FILTER;
783 data_size = lum->u.filter.data_size;
784 data_size_max = LTTNG_UST_ABI_FILTER_BYTECODE_MAX_LEN;
785 reloc_offset = lum->u.filter.reloc_offset;
786 seqnum = lum->u.filter.seqnum;
787 break;
788 case LTTNG_UST_ABI_CAPTURE:
789 type = LTTNG_UST_BYTECODE_TYPE_CAPTURE;
790 data_size = lum->u.capture.data_size;
791 data_size_max = LTTNG_UST_ABI_CAPTURE_BYTECODE_MAX_LEN;
792 reloc_offset = lum->u.capture.reloc_offset;
793 seqnum = lum->u.capture.seqnum;
794 break;
795 default:
796 abort();
797 }
798
799 if (data_size > data_size_max) {
800 ERR("Bytecode %s data size is too large: %u bytes",
801 bytecode_type_str(lum->cmd), data_size);
802 ret = -EINVAL;
803 goto end;
804 }
805
806 if (reloc_offset > data_size) {
807 ERR("Bytecode %s reloc offset %u is not within data",
808 bytecode_type_str(lum->cmd), reloc_offset);
809 ret = -EINVAL;
810 goto end;
811 }
812
813 /* Allocate the structure AND the `data[]` field. */
814 bytecode = zmalloc(sizeof(*bytecode) + data_size);
815 if (!bytecode) {
816 ret = -ENOMEM;
817 goto end;
818 }
819
820 bytecode->bc.len = data_size;
821 bytecode->bc.reloc_offset = reloc_offset;
822 bytecode->bc.seqnum = seqnum;
823 bytecode->type = type;
824
825 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, bytecode->bc.len);
826 switch (len) {
827 case 0: /* orderly shutdown */
828 ret = 0;
829 goto end;
830 default:
831 if (len == bytecode->bc.len) {
832 DBG("Bytecode %s data received",
833 bytecode_type_str(lum->cmd));
834 break;
835 } else if (len < 0) {
836 DBG("Receive failed from lttng-sessiond with errno %d",
837 (int) -len);
838 if (len == -ECONNRESET) {
839 ERR("%s remote end closed connection",
840 sock_info->name);
841 ret = len;
842 goto end;
843 }
844 ret = len;
845 goto end;
846 } else {
847 DBG("Incorrect %s bytecode data message size: %zd",
848 bytecode_type_str(lum->cmd), len);
849 ret = -EINVAL;
850 goto end;
851 }
852 }
853
854 ops = lttng_ust_abi_objd_ops(lum->handle);
855 if (!ops) {
856 ret = -ENOENT;
857 goto end;
858 }
859
860 if (ops->cmd)
861 ret = ops->cmd(lum->handle, lum->cmd,
862 (unsigned long) &bytecode,
863 NULL, sock_info);
864 else
865 ret = -ENOSYS;
866
867 end:
868 free(bytecode);
869 return ret;
870 }
871
872 static
873 int handle_message(struct sock_info *sock_info,
874 int sock, struct ustcomm_ust_msg *lum)
875 {
876 int ret = 0;
877 const struct lttng_ust_abi_objd_ops *ops;
878 struct ustcomm_ust_reply lur;
879 union lttng_ust_abi_args args;
880 char ctxstr[LTTNG_UST_ABI_SYM_NAME_LEN]; /* App context string. */
881 ssize_t len;
882
883 memset(&lur, 0, sizeof(lur));
884
885 if (ust_lock()) {
886 ret = -LTTNG_UST_ERR_EXITING;
887 goto error;
888 }
889
890 ops = lttng_ust_abi_objd_ops(lum->handle);
891 if (!ops) {
892 ret = -ENOENT;
893 goto error;
894 }
895
896 switch (lum->cmd) {
897 case LTTNG_UST_ABI_REGISTER_DONE:
898 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
899 ret = handle_register_done(sock_info);
900 else
901 ret = -EINVAL;
902 break;
903 case LTTNG_UST_ABI_RELEASE:
904 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
905 ret = -EPERM;
906 else
907 ret = lttng_ust_abi_objd_unref(lum->handle, 1);
908 break;
909 case LTTNG_UST_ABI_CAPTURE:
910 case LTTNG_UST_ABI_FILTER:
911 ret = handle_bytecode_recv(sock_info, sock, lum);
912 if (ret)
913 goto error;
914 break;
915 case LTTNG_UST_ABI_EXCLUSION:
916 {
917 /* Receive exclusion names */
918 struct lttng_ust_excluder_node *node;
919 unsigned int count;
920
921 count = lum->u.exclusion.count;
922 if (count == 0) {
923 /* There are no names to read */
924 ret = 0;
925 goto error;
926 }
927 node = zmalloc(sizeof(*node) +
928 count * LTTNG_UST_ABI_SYM_NAME_LEN);
929 if (!node) {
930 ret = -ENOMEM;
931 goto error;
932 }
933 node->excluder.count = count;
934 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
935 count * LTTNG_UST_ABI_SYM_NAME_LEN);
936 switch (len) {
937 case 0: /* orderly shutdown */
938 ret = 0;
939 free(node);
940 goto error;
941 default:
942 if (len == count * LTTNG_UST_ABI_SYM_NAME_LEN) {
943 DBG("Exclusion data received");
944 break;
945 } else if (len < 0) {
946 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
947 if (len == -ECONNRESET) {
948 ERR("%s remote end closed connection", sock_info->name);
949 ret = len;
950 free(node);
951 goto error;
952 }
953 ret = len;
954 free(node);
955 goto error;
956 } else {
957 DBG("Incorrect exclusion data message size: %zd", len);
958 ret = -EINVAL;
959 free(node);
960 goto error;
961 }
962 }
963 if (ops->cmd)
964 ret = ops->cmd(lum->handle, lum->cmd,
965 (unsigned long) &node,
966 &args, sock_info);
967 else
968 ret = -ENOSYS;
969 free(node);
970 break;
971 }
972 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE:
973 {
974 int event_notifier_notif_fd, close_ret;
975
976 len = ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock,
977 &event_notifier_notif_fd);
978 switch (len) {
979 case 0: /* orderly shutdown */
980 ret = 0;
981 goto error;
982 case 1:
983 break;
984 default:
985 if (len < 0) {
986 DBG("Receive failed from lttng-sessiond with errno %d",
987 (int) -len);
988 if (len == -ECONNRESET) {
989 ERR("%s remote end closed connection",
990 sock_info->name);
991 ret = len;
992 goto error;
993 }
994 ret = len;
995 goto error;
996 } else {
997 DBG("Incorrect event notifier fd message size: %zd",
998 len);
999 ret = -EINVAL;
1000 goto error;
1001 }
1002 }
1003 args.event_notifier_handle.event_notifier_notif_fd =
1004 event_notifier_notif_fd;
1005 if (ops->cmd)
1006 ret = ops->cmd(lum->handle, lum->cmd,
1007 (unsigned long) &lum->u,
1008 &args, sock_info);
1009 else
1010 ret = -ENOSYS;
1011 if (args.event_notifier_handle.event_notifier_notif_fd >= 0) {
1012 lttng_ust_lock_fd_tracker();
1013 close_ret = close(args.event_notifier_handle.event_notifier_notif_fd);
1014 lttng_ust_unlock_fd_tracker();
1015 if (close_ret)
1016 PERROR("close");
1017 }
1018 break;
1019 }
1020 case LTTNG_UST_ABI_CHANNEL:
1021 {
1022 void *chan_data;
1023 int wakeup_fd;
1024
1025 len = ustcomm_recv_channel_from_sessiond(sock,
1026 &chan_data, lum->u.channel.len,
1027 &wakeup_fd);
1028 switch (len) {
1029 case 0: /* orderly shutdown */
1030 ret = 0;
1031 goto error;
1032 default:
1033 if (len == lum->u.channel.len) {
1034 DBG("channel data received");
1035 break;
1036 } else if (len < 0) {
1037 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1038 if (len == -ECONNRESET) {
1039 ERR("%s remote end closed connection", sock_info->name);
1040 ret = len;
1041 goto error;
1042 }
1043 ret = len;
1044 goto error;
1045 } else {
1046 DBG("incorrect channel data message size: %zd", len);
1047 ret = -EINVAL;
1048 goto error;
1049 }
1050 }
1051 args.channel.chan_data = chan_data;
1052 args.channel.wakeup_fd = wakeup_fd;
1053 if (ops->cmd)
1054 ret = ops->cmd(lum->handle, lum->cmd,
1055 (unsigned long) &lum->u,
1056 &args, sock_info);
1057 else
1058 ret = -ENOSYS;
1059 if (args.channel.wakeup_fd >= 0) {
1060 int close_ret;
1061
1062 lttng_ust_lock_fd_tracker();
1063 close_ret = close(args.channel.wakeup_fd);
1064 lttng_ust_unlock_fd_tracker();
1065 args.channel.wakeup_fd = -1;
1066 if (close_ret)
1067 PERROR("close");
1068 }
1069 free(args.channel.chan_data);
1070 break;
1071 }
1072 case LTTNG_UST_ABI_STREAM:
1073 {
1074 int close_ret;
1075
1076 /* Receive shm_fd, wakeup_fd */
1077 ret = ustcomm_recv_stream_from_sessiond(sock,
1078 NULL,
1079 &args.stream.shm_fd,
1080 &args.stream.wakeup_fd);
1081 if (ret) {
1082 goto error;
1083 }
1084
1085 if (ops->cmd)
1086 ret = ops->cmd(lum->handle, lum->cmd,
1087 (unsigned long) &lum->u,
1088 &args, sock_info);
1089 else
1090 ret = -ENOSYS;
1091 if (args.stream.shm_fd >= 0) {
1092 lttng_ust_lock_fd_tracker();
1093 close_ret = close(args.stream.shm_fd);
1094 lttng_ust_unlock_fd_tracker();
1095 args.stream.shm_fd = -1;
1096 if (close_ret)
1097 PERROR("close");
1098 }
1099 if (args.stream.wakeup_fd >= 0) {
1100 lttng_ust_lock_fd_tracker();
1101 close_ret = close(args.stream.wakeup_fd);
1102 lttng_ust_unlock_fd_tracker();
1103 args.stream.wakeup_fd = -1;
1104 if (close_ret)
1105 PERROR("close");
1106 }
1107 break;
1108 }
1109 case LTTNG_UST_ABI_CONTEXT:
1110 switch (lum->u.context.ctx) {
1111 case LTTNG_UST_ABI_CONTEXT_APP_CONTEXT:
1112 {
1113 char *p;
1114 size_t ctxlen, recvlen;
1115
1116 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
1117 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
1118 if (ctxlen >= LTTNG_UST_ABI_SYM_NAME_LEN) {
1119 ERR("Application context string length size is too large: %zu bytes",
1120 ctxlen);
1121 ret = -EINVAL;
1122 goto error;
1123 }
1124 strcpy(ctxstr, "$app.");
1125 p = &ctxstr[strlen("$app.")];
1126 recvlen = ctxlen - strlen("$app.");
1127 len = ustcomm_recv_unix_sock(sock, p, recvlen);
1128 switch (len) {
1129 case 0: /* orderly shutdown */
1130 ret = 0;
1131 goto error;
1132 default:
1133 if (len == recvlen) {
1134 DBG("app context data received");
1135 break;
1136 } else if (len < 0) {
1137 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1138 if (len == -ECONNRESET) {
1139 ERR("%s remote end closed connection", sock_info->name);
1140 ret = len;
1141 goto error;
1142 }
1143 ret = len;
1144 goto error;
1145 } else {
1146 DBG("incorrect app context data message size: %zd", len);
1147 ret = -EINVAL;
1148 goto error;
1149 }
1150 }
1151 /* Put : between provider and ctxname. */
1152 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1153 args.app_context.ctxname = ctxstr;
1154 break;
1155 }
1156 default:
1157 break;
1158 }
1159 if (ops->cmd) {
1160 ret = ops->cmd(lum->handle, lum->cmd,
1161 (unsigned long) &lum->u,
1162 &args, sock_info);
1163 } else {
1164 ret = -ENOSYS;
1165 }
1166 break;
1167 case LTTNG_UST_ABI_COUNTER:
1168 {
1169 void *counter_data;
1170
1171 len = ustcomm_recv_counter_from_sessiond(sock,
1172 &counter_data, lum->u.counter.len);
1173 switch (len) {
1174 case 0: /* orderly shutdown */
1175 ret = 0;
1176 goto error;
1177 default:
1178 if (len == lum->u.counter.len) {
1179 DBG("counter data received");
1180 break;
1181 } else if (len < 0) {
1182 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1183 if (len == -ECONNRESET) {
1184 ERR("%s remote end closed connection", sock_info->name);
1185 ret = len;
1186 goto error;
1187 }
1188 ret = len;
1189 goto error;
1190 } else {
1191 DBG("incorrect counter data message size: %zd", len);
1192 ret = -EINVAL;
1193 goto error;
1194 }
1195 }
1196 args.counter.counter_data = counter_data;
1197 if (ops->cmd)
1198 ret = ops->cmd(lum->handle, lum->cmd,
1199 (unsigned long) &lum->u,
1200 &args, sock_info);
1201 else
1202 ret = -ENOSYS;
1203 free(args.counter.counter_data);
1204 break;
1205 }
1206 case LTTNG_UST_ABI_COUNTER_GLOBAL:
1207 {
1208 /* Receive shm_fd */
1209 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1210 &args.counter_shm.shm_fd);
1211 if (ret) {
1212 goto error;
1213 }
1214
1215 if (ops->cmd)
1216 ret = ops->cmd(lum->handle, lum->cmd,
1217 (unsigned long) &lum->u,
1218 &args, sock_info);
1219 else
1220 ret = -ENOSYS;
1221 if (args.counter_shm.shm_fd >= 0) {
1222 int close_ret;
1223
1224 lttng_ust_lock_fd_tracker();
1225 close_ret = close(args.counter_shm.shm_fd);
1226 lttng_ust_unlock_fd_tracker();
1227 args.counter_shm.shm_fd = -1;
1228 if (close_ret)
1229 PERROR("close");
1230 }
1231 break;
1232 }
1233 case LTTNG_UST_ABI_COUNTER_CPU:
1234 {
1235 /* Receive shm_fd */
1236 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1237 &args.counter_shm.shm_fd);
1238 if (ret) {
1239 goto error;
1240 }
1241
1242 if (ops->cmd)
1243 ret = ops->cmd(lum->handle, lum->cmd,
1244 (unsigned long) &lum->u,
1245 &args, sock_info);
1246 else
1247 ret = -ENOSYS;
1248 if (args.counter_shm.shm_fd >= 0) {
1249 int close_ret;
1250
1251 lttng_ust_lock_fd_tracker();
1252 close_ret = close(args.counter_shm.shm_fd);
1253 lttng_ust_unlock_fd_tracker();
1254 args.counter_shm.shm_fd = -1;
1255 if (close_ret)
1256 PERROR("close");
1257 }
1258 break;
1259 }
1260 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE:
1261 {
1262 /* Receive struct lttng_ust_event_notifier */
1263 struct lttng_ust_abi_event_notifier event_notifier;
1264
1265 if (sizeof(event_notifier) != lum->u.event_notifier.len) {
1266 DBG("incorrect event notifier data message size: %u", lum->u.event_notifier.len);
1267 ret = -EINVAL;
1268 goto error;
1269 }
1270 len = ustcomm_recv_unix_sock(sock, &event_notifier, sizeof(event_notifier));
1271 switch (len) {
1272 case 0: /* orderly shutdown */
1273 ret = 0;
1274 goto error;
1275 default:
1276 if (len == sizeof(event_notifier)) {
1277 DBG("event notifier data received");
1278 break;
1279 } else if (len < 0) {
1280 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1281 if (len == -ECONNRESET) {
1282 ERR("%s remote end closed connection", sock_info->name);
1283 ret = len;
1284 goto error;
1285 }
1286 ret = len;
1287 goto error;
1288 } else {
1289 DBG("incorrect event notifier data message size: %zd", len);
1290 ret = -EINVAL;
1291 goto error;
1292 }
1293 }
1294 if (ops->cmd)
1295 ret = ops->cmd(lum->handle, lum->cmd,
1296 (unsigned long) &event_notifier,
1297 &args, sock_info);
1298 else
1299 ret = -ENOSYS;
1300 break;
1301 }
1302
1303 default:
1304 if (ops->cmd)
1305 ret = ops->cmd(lum->handle, lum->cmd,
1306 (unsigned long) &lum->u,
1307 &args, sock_info);
1308 else
1309 ret = -ENOSYS;
1310 break;
1311 }
1312
1313 lur.handle = lum->handle;
1314 lur.cmd = lum->cmd;
1315 lur.ret_val = ret;
1316 if (ret >= 0) {
1317 lur.ret_code = LTTNG_UST_OK;
1318 } else {
1319 /*
1320 * Use -LTTNG_UST_ERR as wildcard for UST internal
1321 * error that are not caused by the transport, except if
1322 * we already have a more precise error message to
1323 * report.
1324 */
1325 if (ret > -LTTNG_UST_ERR) {
1326 /* Translate code to UST error. */
1327 switch (ret) {
1328 case -EEXIST:
1329 lur.ret_code = -LTTNG_UST_ERR_EXIST;
1330 break;
1331 case -EINVAL:
1332 lur.ret_code = -LTTNG_UST_ERR_INVAL;
1333 break;
1334 case -ENOENT:
1335 lur.ret_code = -LTTNG_UST_ERR_NOENT;
1336 break;
1337 case -EPERM:
1338 lur.ret_code = -LTTNG_UST_ERR_PERM;
1339 break;
1340 case -ENOSYS:
1341 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
1342 break;
1343 default:
1344 lur.ret_code = -LTTNG_UST_ERR;
1345 break;
1346 }
1347 } else {
1348 lur.ret_code = ret;
1349 }
1350 }
1351 if (ret >= 0) {
1352 switch (lum->cmd) {
1353 case LTTNG_UST_ABI_TRACER_VERSION:
1354 lur.u.version = lum->u.version;
1355 break;
1356 case LTTNG_UST_ABI_TRACEPOINT_LIST_GET:
1357 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1358 break;
1359 }
1360 }
1361 DBG("Return value: %d", lur.ret_val);
1362
1363 ust_unlock();
1364
1365 /*
1366 * Performed delayed statedump operations outside of the UST
1367 * lock. We need to take the dynamic loader lock before we take
1368 * the UST lock internally within handle_pending_statedump().
1369 */
1370 handle_pending_statedump(sock_info);
1371
1372 if (ust_lock()) {
1373 ret = -LTTNG_UST_ERR_EXITING;
1374 goto error;
1375 }
1376
1377 ret = send_reply(sock, &lur);
1378 if (ret < 0) {
1379 DBG("error sending reply");
1380 goto error;
1381 }
1382
1383 /*
1384 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1385 * after the reply.
1386 */
1387 if (lur.ret_code == LTTNG_UST_OK) {
1388 switch (lum->cmd) {
1389 case LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET:
1390 len = ustcomm_send_unix_sock(sock,
1391 &args.field_list.entry,
1392 sizeof(args.field_list.entry));
1393 if (len < 0) {
1394 ret = len;
1395 goto error;
1396 }
1397 if (len != sizeof(args.field_list.entry)) {
1398 ret = -EINVAL;
1399 goto error;
1400 }
1401 }
1402 }
1403
1404 error:
1405 ust_unlock();
1406
1407 return ret;
1408 }
1409
1410 static
1411 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1412 {
1413 int ret;
1414
1415 if (sock_info->root_handle != -1) {
1416 ret = lttng_ust_abi_objd_unref(sock_info->root_handle, 1);
1417 if (ret) {
1418 ERR("Error unref root handle");
1419 }
1420 sock_info->root_handle = -1;
1421 }
1422 sock_info->registration_done = 0;
1423 sock_info->initial_statedump_done = 0;
1424
1425 /*
1426 * wait_shm_mmap, socket and notify socket are used by listener
1427 * threads outside of the ust lock, so we cannot tear them down
1428 * ourselves, because we cannot join on these threads. Leave
1429 * responsibility of cleaning up these resources to the OS
1430 * process exit.
1431 */
1432 if (exiting)
1433 return;
1434
1435 if (sock_info->socket != -1) {
1436 ret = ustcomm_close_unix_sock(sock_info->socket);
1437 if (ret) {
1438 ERR("Error closing ust cmd socket");
1439 }
1440 sock_info->socket = -1;
1441 }
1442 if (sock_info->notify_socket != -1) {
1443 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1444 if (ret) {
1445 ERR("Error closing ust notify socket");
1446 }
1447 sock_info->notify_socket = -1;
1448 }
1449 if (sock_info->wait_shm_mmap) {
1450 long page_size;
1451
1452 page_size = LTTNG_UST_PAGE_SIZE;
1453 if (page_size <= 0) {
1454 if (!page_size) {
1455 errno = EINVAL;
1456 }
1457 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1458 } else {
1459 ret = munmap(sock_info->wait_shm_mmap, page_size);
1460 if (ret) {
1461 ERR("Error unmapping wait shm");
1462 }
1463 }
1464 sock_info->wait_shm_mmap = NULL;
1465 }
1466 }
1467
1468 /*
1469 * Using fork to set umask in the child process (not multi-thread safe).
1470 * We deal with the shm_open vs ftruncate race (happening when the
1471 * sessiond owns the shm and does not let everybody modify it, to ensure
1472 * safety against shm_unlink) by simply letting the mmap fail and
1473 * retrying after a few seconds.
1474 * For global shm, everybody has rw access to it until the sessiond
1475 * starts.
1476 */
1477 static
1478 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1479 {
1480 int wait_shm_fd, ret;
1481 pid_t pid;
1482
1483 /*
1484 * Try to open read-only.
1485 */
1486 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1487 if (wait_shm_fd >= 0) {
1488 int32_t tmp_read;
1489 ssize_t len;
1490 size_t bytes_read = 0;
1491
1492 /*
1493 * Try to read the fd. If unable to do so, try opening
1494 * it in write mode.
1495 */
1496 do {
1497 len = read(wait_shm_fd,
1498 &((char *) &tmp_read)[bytes_read],
1499 sizeof(tmp_read) - bytes_read);
1500 if (len > 0) {
1501 bytes_read += len;
1502 }
1503 } while ((len < 0 && errno == EINTR)
1504 || (len > 0 && bytes_read < sizeof(tmp_read)));
1505 if (bytes_read != sizeof(tmp_read)) {
1506 ret = close(wait_shm_fd);
1507 if (ret) {
1508 ERR("close wait_shm_fd");
1509 }
1510 goto open_write;
1511 }
1512 goto end;
1513 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1514 /*
1515 * Real-only open did not work, and it's not because the
1516 * entry was not present. It's a failure that prohibits
1517 * using shm.
1518 */
1519 ERR("Error opening shm %s", sock_info->wait_shm_path);
1520 goto end;
1521 }
1522
1523 open_write:
1524 /*
1525 * If the open failed because the file did not exist, or because
1526 * the file was not truncated yet, try creating it ourself.
1527 */
1528 URCU_TLS(lttng_ust_nest_count)++;
1529 pid = fork();
1530 URCU_TLS(lttng_ust_nest_count)--;
1531 if (pid > 0) {
1532 int status;
1533
1534 /*
1535 * Parent: wait for child to return, in which case the
1536 * shared memory map will have been created.
1537 */
1538 pid = wait(&status);
1539 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1540 wait_shm_fd = -1;
1541 goto end;
1542 }
1543 /*
1544 * Try to open read-only again after creation.
1545 */
1546 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1547 if (wait_shm_fd < 0) {
1548 /*
1549 * Real-only open did not work. It's a failure
1550 * that prohibits using shm.
1551 */
1552 ERR("Error opening shm %s", sock_info->wait_shm_path);
1553 goto end;
1554 }
1555 goto end;
1556 } else if (pid == 0) {
1557 int create_mode;
1558
1559 /* Child */
1560 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1561 if (sock_info->global)
1562 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1563 /*
1564 * We're alone in a child process, so we can modify the
1565 * process-wide umask.
1566 */
1567 umask(~create_mode);
1568 /*
1569 * Try creating shm (or get rw access).
1570 * We don't do an exclusive open, because we allow other
1571 * processes to create+ftruncate it concurrently.
1572 */
1573 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1574 O_RDWR | O_CREAT, create_mode);
1575 if (wait_shm_fd >= 0) {
1576 ret = ftruncate(wait_shm_fd, mmap_size);
1577 if (ret) {
1578 PERROR("ftruncate");
1579 _exit(EXIT_FAILURE);
1580 }
1581 _exit(EXIT_SUCCESS);
1582 }
1583 /*
1584 * For local shm, we need to have rw access to accept
1585 * opening it: this means the local sessiond will be
1586 * able to wake us up. For global shm, we open it even
1587 * if rw access is not granted, because the root.root
1588 * sessiond will be able to override all rights and wake
1589 * us up.
1590 */
1591 if (!sock_info->global && errno != EACCES) {
1592 ERR("Error opening shm %s", sock_info->wait_shm_path);
1593 _exit(EXIT_FAILURE);
1594 }
1595 /*
1596 * The shm exists, but we cannot open it RW. Report
1597 * success.
1598 */
1599 _exit(EXIT_SUCCESS);
1600 } else {
1601 return -1;
1602 }
1603 end:
1604 if (wait_shm_fd >= 0 && !sock_info->global) {
1605 struct stat statbuf;
1606
1607 /*
1608 * Ensure that our user is the owner of the shm file for
1609 * local shm. If we do not own the file, it means our
1610 * sessiond will not have access to wake us up (there is
1611 * probably a rogue process trying to fake our
1612 * sessiond). Fallback to polling method in this case.
1613 */
1614 ret = fstat(wait_shm_fd, &statbuf);
1615 if (ret) {
1616 PERROR("fstat");
1617 goto error_close;
1618 }
1619 if (statbuf.st_uid != getuid())
1620 goto error_close;
1621 }
1622 return wait_shm_fd;
1623
1624 error_close:
1625 ret = close(wait_shm_fd);
1626 if (ret) {
1627 PERROR("Error closing fd");
1628 }
1629 return -1;
1630 }
1631
1632 static
1633 char *get_map_shm(struct sock_info *sock_info)
1634 {
1635 long page_size;
1636 int wait_shm_fd, ret;
1637 char *wait_shm_mmap;
1638
1639 page_size = sysconf(_SC_PAGE_SIZE);
1640 if (page_size <= 0) {
1641 if (!page_size) {
1642 errno = EINVAL;
1643 }
1644 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1645 goto error;
1646 }
1647
1648 lttng_ust_lock_fd_tracker();
1649 wait_shm_fd = get_wait_shm(sock_info, page_size);
1650 if (wait_shm_fd < 0) {
1651 lttng_ust_unlock_fd_tracker();
1652 goto error;
1653 }
1654
1655 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1656 if (ret < 0) {
1657 ret = close(wait_shm_fd);
1658 if (!ret) {
1659 PERROR("Error closing fd");
1660 }
1661 lttng_ust_unlock_fd_tracker();
1662 goto error;
1663 }
1664
1665 wait_shm_fd = ret;
1666 lttng_ust_unlock_fd_tracker();
1667
1668 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1669 MAP_SHARED, wait_shm_fd, 0);
1670
1671 /* close shm fd immediately after taking the mmap reference */
1672 lttng_ust_lock_fd_tracker();
1673 ret = close(wait_shm_fd);
1674 if (!ret) {
1675 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1676 } else {
1677 PERROR("Error closing fd");
1678 }
1679 lttng_ust_unlock_fd_tracker();
1680
1681 if (wait_shm_mmap == MAP_FAILED) {
1682 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1683 goto error;
1684 }
1685 return wait_shm_mmap;
1686
1687 error:
1688 return NULL;
1689 }
1690
1691 static
1692 void wait_for_sessiond(struct sock_info *sock_info)
1693 {
1694 /* Use ust_lock to check if we should quit. */
1695 if (ust_lock()) {
1696 goto quit;
1697 }
1698 if (wait_poll_fallback) {
1699 goto error;
1700 }
1701 ust_unlock();
1702
1703 assert(sock_info->wait_shm_mmap);
1704
1705 DBG("Waiting for %s apps sessiond", sock_info->name);
1706 /* Wait for futex wakeup */
1707 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1708 goto end_wait;
1709
1710 while (lttng_ust_futex_async((int32_t *) sock_info->wait_shm_mmap,
1711 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1712 switch (errno) {
1713 case EWOULDBLOCK:
1714 /* Value already changed. */
1715 goto end_wait;
1716 case EINTR:
1717 /* Retry if interrupted by signal. */
1718 break; /* Get out of switch. */
1719 case EFAULT:
1720 wait_poll_fallback = 1;
1721 DBG(
1722 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1723 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1724 "Please upgrade your kernel "
1725 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1726 "mainline). LTTng-UST will use polling mode fallback.");
1727 if (ust_err_debug_enabled())
1728 PERROR("futex");
1729 goto end_wait;
1730 }
1731 }
1732 end_wait:
1733 return;
1734
1735 quit:
1736 ust_unlock();
1737 return;
1738
1739 error:
1740 ust_unlock();
1741 return;
1742 }
1743
1744 /*
1745 * This thread does not allocate any resource, except within
1746 * handle_message, within mutex protection. This mutex protects against
1747 * fork and exit.
1748 * The other moment it allocates resources is at socket connection, which
1749 * is also protected by the mutex.
1750 */
1751 static
1752 void *ust_listener_thread(void *arg)
1753 {
1754 struct sock_info *sock_info = arg;
1755 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1756 long timeout;
1757
1758 lttng_ust_fixup_tls();
1759 /*
1760 * If available, add '-ust' to the end of this thread's
1761 * process name
1762 */
1763 ret = lttng_ust_setustprocname();
1764 if (ret) {
1765 ERR("Unable to set UST process name");
1766 }
1767
1768 /* Restart trying to connect to the session daemon */
1769 restart:
1770 if (prev_connect_failed) {
1771 /* Wait for sessiond availability with pipe */
1772 wait_for_sessiond(sock_info);
1773 if (has_waited) {
1774 has_waited = 0;
1775 /*
1776 * Sleep for 5 seconds before retrying after a
1777 * sequence of failure / wait / failure. This
1778 * deals with a killed or broken session daemon.
1779 */
1780 sleep(5);
1781 } else {
1782 has_waited = 1;
1783 }
1784 prev_connect_failed = 0;
1785 }
1786
1787 if (ust_lock()) {
1788 goto quit;
1789 }
1790
1791 if (sock_info->socket != -1) {
1792 /* FD tracker is updated by ustcomm_close_unix_sock() */
1793 ret = ustcomm_close_unix_sock(sock_info->socket);
1794 if (ret) {
1795 ERR("Error closing %s ust cmd socket",
1796 sock_info->name);
1797 }
1798 sock_info->socket = -1;
1799 }
1800 if (sock_info->notify_socket != -1) {
1801 /* FD tracker is updated by ustcomm_close_unix_sock() */
1802 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1803 if (ret) {
1804 ERR("Error closing %s ust notify socket",
1805 sock_info->name);
1806 }
1807 sock_info->notify_socket = -1;
1808 }
1809
1810
1811 /*
1812 * Register. We need to perform both connect and sending
1813 * registration message before doing the next connect otherwise
1814 * we may reach unix socket connect queue max limits and block
1815 * on the 2nd connect while the session daemon is awaiting the
1816 * first connect registration message.
1817 */
1818 /* Connect cmd socket */
1819 lttng_ust_lock_fd_tracker();
1820 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1821 get_connect_sock_timeout());
1822 if (ret < 0) {
1823 lttng_ust_unlock_fd_tracker();
1824 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1825 prev_connect_failed = 1;
1826
1827 /*
1828 * If we cannot find the sessiond daemon, don't delay
1829 * constructor execution.
1830 */
1831 ret = handle_register_failed(sock_info);
1832 assert(!ret);
1833 ust_unlock();
1834 goto restart;
1835 }
1836 fd = ret;
1837 ret = lttng_ust_add_fd_to_tracker(fd);
1838 if (ret < 0) {
1839 ret = close(fd);
1840 if (ret) {
1841 PERROR("close on sock_info->socket");
1842 }
1843 ret = -1;
1844 lttng_ust_unlock_fd_tracker();
1845 ust_unlock();
1846 goto quit;
1847 }
1848
1849 sock_info->socket = ret;
1850 lttng_ust_unlock_fd_tracker();
1851
1852 ust_unlock();
1853 /*
1854 * Unlock/relock ust lock because connect is blocking (with
1855 * timeout). Don't delay constructors on the ust lock for too
1856 * long.
1857 */
1858 if (ust_lock()) {
1859 goto quit;
1860 }
1861
1862 /*
1863 * Create only one root handle per listener thread for the whole
1864 * process lifetime, so we ensure we get ID which is statically
1865 * assigned to the root handle.
1866 */
1867 if (sock_info->root_handle == -1) {
1868 ret = lttng_abi_create_root_handle();
1869 if (ret < 0) {
1870 ERR("Error creating root handle");
1871 goto quit;
1872 }
1873 sock_info->root_handle = ret;
1874 }
1875
1876 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1877 if (ret < 0) {
1878 ERR("Error registering to %s ust cmd socket",
1879 sock_info->name);
1880 prev_connect_failed = 1;
1881 /*
1882 * If we cannot register to the sessiond daemon, don't
1883 * delay constructor execution.
1884 */
1885 ret = handle_register_failed(sock_info);
1886 assert(!ret);
1887 ust_unlock();
1888 goto restart;
1889 }
1890
1891 ust_unlock();
1892 /*
1893 * Unlock/relock ust lock because connect is blocking (with
1894 * timeout). Don't delay constructors on the ust lock for too
1895 * long.
1896 */
1897 if (ust_lock()) {
1898 goto quit;
1899 }
1900
1901 /* Connect notify socket */
1902 lttng_ust_lock_fd_tracker();
1903 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1904 get_connect_sock_timeout());
1905 if (ret < 0) {
1906 lttng_ust_unlock_fd_tracker();
1907 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1908 prev_connect_failed = 1;
1909
1910 /*
1911 * If we cannot find the sessiond daemon, don't delay
1912 * constructor execution.
1913 */
1914 ret = handle_register_failed(sock_info);
1915 assert(!ret);
1916 ust_unlock();
1917 goto restart;
1918 }
1919
1920 fd = ret;
1921 ret = lttng_ust_add_fd_to_tracker(fd);
1922 if (ret < 0) {
1923 ret = close(fd);
1924 if (ret) {
1925 PERROR("close on sock_info->notify_socket");
1926 }
1927 ret = -1;
1928 lttng_ust_unlock_fd_tracker();
1929 ust_unlock();
1930 goto quit;
1931 }
1932
1933 sock_info->notify_socket = ret;
1934 lttng_ust_unlock_fd_tracker();
1935
1936 ust_unlock();
1937 /*
1938 * Unlock/relock ust lock because connect is blocking (with
1939 * timeout). Don't delay constructors on the ust lock for too
1940 * long.
1941 */
1942 if (ust_lock()) {
1943 goto quit;
1944 }
1945
1946 timeout = get_notify_sock_timeout();
1947 if (timeout >= 0) {
1948 /*
1949 * Give at least 10ms to sessiond to reply to
1950 * notifications.
1951 */
1952 if (timeout < 10)
1953 timeout = 10;
1954 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1955 timeout);
1956 if (ret < 0) {
1957 WARN("Error setting socket receive timeout");
1958 }
1959 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1960 timeout);
1961 if (ret < 0) {
1962 WARN("Error setting socket send timeout");
1963 }
1964 } else if (timeout < -1) {
1965 WARN("Unsupported timeout value %ld", timeout);
1966 }
1967
1968 ret = register_to_sessiond(sock_info->notify_socket,
1969 USTCTL_SOCKET_NOTIFY);
1970 if (ret < 0) {
1971 ERR("Error registering to %s ust notify socket",
1972 sock_info->name);
1973 prev_connect_failed = 1;
1974 /*
1975 * If we cannot register to the sessiond daemon, don't
1976 * delay constructor execution.
1977 */
1978 ret = handle_register_failed(sock_info);
1979 assert(!ret);
1980 ust_unlock();
1981 goto restart;
1982 }
1983 sock = sock_info->socket;
1984
1985 ust_unlock();
1986
1987 for (;;) {
1988 ssize_t len;
1989 struct ustcomm_ust_msg lum;
1990
1991 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1992 switch (len) {
1993 case 0: /* orderly shutdown */
1994 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1995 if (ust_lock()) {
1996 goto quit;
1997 }
1998 /*
1999 * Either sessiond has shutdown or refused us by closing the socket.
2000 * In either case, we don't want to delay construction execution,
2001 * and we need to wait before retry.
2002 */
2003 prev_connect_failed = 1;
2004 /*
2005 * If we cannot register to the sessiond daemon, don't
2006 * delay constructor execution.
2007 */
2008 ret = handle_register_failed(sock_info);
2009 assert(!ret);
2010 ust_unlock();
2011 goto end;
2012 case sizeof(lum):
2013 print_cmd(lum.cmd, lum.handle);
2014 ret = handle_message(sock_info, sock, &lum);
2015 if (ret) {
2016 ERR("Error handling message for %s socket",
2017 sock_info->name);
2018 /*
2019 * Close socket if protocol error is
2020 * detected.
2021 */
2022 goto end;
2023 }
2024 continue;
2025 default:
2026 if (len < 0) {
2027 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
2028 } else {
2029 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
2030 }
2031 if (len == -ECONNRESET) {
2032 DBG("%s remote end closed connection", sock_info->name);
2033 goto end;
2034 }
2035 goto end;
2036 }
2037
2038 }
2039 end:
2040 if (ust_lock()) {
2041 goto quit;
2042 }
2043 /* Cleanup socket handles before trying to reconnect */
2044 lttng_ust_abi_objd_table_owner_cleanup(sock_info);
2045 ust_unlock();
2046 goto restart; /* try to reconnect */
2047
2048 quit:
2049 ust_unlock();
2050
2051 pthread_mutex_lock(&ust_exit_mutex);
2052 sock_info->thread_active = 0;
2053 pthread_mutex_unlock(&ust_exit_mutex);
2054 return NULL;
2055 }
2056
2057 /*
2058 * Weak symbol to call when the ust malloc wrapper is not loaded.
2059 */
2060 __attribute__((weak))
2061 void lttng_ust_libc_wrapper_malloc_init(void)
2062 {
2063 }
2064
2065 void lttng_ust_ring_buffer_clients_init(void)
2066 {
2067 lttng_ring_buffer_metadata_client_init();
2068 lttng_ring_buffer_client_overwrite_init();
2069 lttng_ring_buffer_client_overwrite_rt_init();
2070 lttng_ring_buffer_client_discard_init();
2071 lttng_ring_buffer_client_discard_rt_init();
2072 }
2073
2074 void lttng_ust_ring_buffer_clients_exit(void)
2075 {
2076 lttng_ring_buffer_client_discard_rt_exit();
2077 lttng_ring_buffer_client_discard_exit();
2078 lttng_ring_buffer_client_overwrite_rt_exit();
2079 lttng_ring_buffer_client_overwrite_exit();
2080 lttng_ring_buffer_metadata_client_exit();
2081 }
2082
2083 void lttng_ust_counter_clients_init(void)
2084 {
2085 lttng_counter_client_percpu_64_modular_init();
2086 lttng_counter_client_percpu_32_modular_init();
2087 }
2088
2089 void lttng_ust_counter_clients_exit(void)
2090 {
2091 lttng_counter_client_percpu_32_modular_exit();
2092 lttng_counter_client_percpu_64_modular_exit();
2093 }
2094
2095 /*
2096 * sessiond monitoring thread: monitor presence of global and per-user
2097 * sessiond by polling the application common named pipe.
2098 */
2099 void __attribute__((constructor)) lttng_ust_init(void)
2100 {
2101 struct timespec constructor_timeout;
2102 sigset_t sig_all_blocked, orig_parent_mask;
2103 pthread_attr_t thread_attr;
2104 int timeout_mode;
2105 int ret;
2106 void *handle;
2107
2108 if (uatomic_xchg(&initialized, 1) == 1)
2109 return;
2110
2111 /*
2112 * Fixup interdependency between TLS fixup mutex (which happens
2113 * to be the dynamic linker mutex) and ust_lock, taken within
2114 * the ust lock.
2115 */
2116 lttng_ust_fixup_tls();
2117
2118 lttng_ust_loaded = 1;
2119
2120 /*
2121 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2122 * the unloading of code used by the ust_listener_threads as we can not
2123 * reliably know when they exited. To do that, manually load
2124 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2125 * this library so it never becomes zero, thus never gets unloaded from the
2126 * address space of the process. Since we are already running in the
2127 * constructor of the LTTNG_UST_LIB_SONAME library, calling dlopen will
2128 * simply increment the refcount and no additionnal work is needed by the
2129 * dynamic loader as the shared library is already loaded in the address
2130 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2131 * unloading of the UST library if its refcount becomes zero (which should
2132 * never happen). Do the return value check but discard the handle at the
2133 * end of the function as it's not needed.
2134 */
2135 handle = dlopen(LTTNG_UST_LIB_SONAME, RTLD_LAZY | RTLD_NODELETE);
2136 if (!handle) {
2137 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SONAME);
2138 }
2139
2140 /*
2141 * We want precise control over the order in which we construct
2142 * our sub-libraries vs starting to receive commands from
2143 * sessiond (otherwise leading to errors when trying to create
2144 * sessiond before the init functions are completed).
2145 */
2146 ust_err_init();
2147 lttng_ust_getenv_init(); /* Needs ust_err_init() to be completed. */
2148 lttng_ust_tp_init();
2149 lttng_ust_init_fd_tracker();
2150 lttng_ust_clock_init();
2151 lttng_ust_getcpu_init();
2152 lttng_ust_statedump_init();
2153 lttng_ust_ring_buffer_clients_init();
2154 lttng_ust_counter_clients_init();
2155 lttng_perf_counter_init();
2156 /*
2157 * Invoke ust malloc wrapper init before starting other threads.
2158 */
2159 lttng_ust_libc_wrapper_malloc_init();
2160
2161 timeout_mode = get_constructor_timeout(&constructor_timeout);
2162
2163 get_allow_blocking();
2164
2165 ret = sem_init(&constructor_wait, 0, 0);
2166 if (ret) {
2167 PERROR("sem_init");
2168 }
2169
2170 ret = setup_global_apps();
2171 if (ret) {
2172 assert(global_apps.allowed == 0);
2173 DBG("global apps setup returned %d", ret);
2174 }
2175
2176 ret = setup_local_apps();
2177 if (ret) {
2178 assert(local_apps.allowed == 0);
2179 DBG("local apps setup returned %d", ret);
2180 }
2181
2182 /* A new thread created by pthread_create inherits the signal mask
2183 * from the parent. To avoid any signal being received by the
2184 * listener thread, we block all signals temporarily in the parent,
2185 * while we create the listener thread.
2186 */
2187 sigfillset(&sig_all_blocked);
2188 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
2189 if (ret) {
2190 ERR("pthread_sigmask: %s", strerror(ret));
2191 }
2192
2193 ret = pthread_attr_init(&thread_attr);
2194 if (ret) {
2195 ERR("pthread_attr_init: %s", strerror(ret));
2196 }
2197 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
2198 if (ret) {
2199 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
2200 }
2201
2202 if (global_apps.allowed) {
2203 pthread_mutex_lock(&ust_exit_mutex);
2204 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
2205 ust_listener_thread, &global_apps);
2206 if (ret) {
2207 ERR("pthread_create global: %s", strerror(ret));
2208 }
2209 global_apps.thread_active = 1;
2210 pthread_mutex_unlock(&ust_exit_mutex);
2211 } else {
2212 handle_register_done(&global_apps);
2213 }
2214
2215 if (local_apps.allowed) {
2216 pthread_mutex_lock(&ust_exit_mutex);
2217 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
2218 ust_listener_thread, &local_apps);
2219 if (ret) {
2220 ERR("pthread_create local: %s", strerror(ret));
2221 }
2222 local_apps.thread_active = 1;
2223 pthread_mutex_unlock(&ust_exit_mutex);
2224 } else {
2225 handle_register_done(&local_apps);
2226 }
2227 ret = pthread_attr_destroy(&thread_attr);
2228 if (ret) {
2229 ERR("pthread_attr_destroy: %s", strerror(ret));
2230 }
2231
2232 /* Restore original signal mask in parent */
2233 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
2234 if (ret) {
2235 ERR("pthread_sigmask: %s", strerror(ret));
2236 }
2237
2238 switch (timeout_mode) {
2239 case 1: /* timeout wait */
2240 do {
2241 ret = sem_timedwait(&constructor_wait,
2242 &constructor_timeout);
2243 } while (ret < 0 && errno == EINTR);
2244 if (ret < 0) {
2245 switch (errno) {
2246 case ETIMEDOUT:
2247 ERR("Timed out waiting for lttng-sessiond");
2248 break;
2249 case EINVAL:
2250 PERROR("sem_timedwait");
2251 break;
2252 default:
2253 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2254 strerror(errno));
2255 }
2256 }
2257 break;
2258 case -1:/* wait forever */
2259 do {
2260 ret = sem_wait(&constructor_wait);
2261 } while (ret < 0 && errno == EINTR);
2262 if (ret < 0) {
2263 switch (errno) {
2264 case EINVAL:
2265 PERROR("sem_wait");
2266 break;
2267 default:
2268 ERR("Unexpected error \"%s\" returned by sem_wait",
2269 strerror(errno));
2270 }
2271 }
2272 break;
2273 case 0: /* no timeout */
2274 break;
2275 }
2276 }
2277
2278 static
2279 void lttng_ust_cleanup(int exiting)
2280 {
2281 cleanup_sock_info(&global_apps, exiting);
2282 cleanup_sock_info(&local_apps, exiting);
2283 local_apps.allowed = 0;
2284 global_apps.allowed = 0;
2285 /*
2286 * The teardown in this function all affect data structures
2287 * accessed under the UST lock by the listener thread. This
2288 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2289 * that none of these threads are accessing this data at this
2290 * point.
2291 */
2292 lttng_ust_abi_exit();
2293 lttng_ust_abi_events_exit();
2294 lttng_perf_counter_exit();
2295 lttng_ust_ring_buffer_clients_exit();
2296 lttng_ust_counter_clients_exit();
2297 lttng_ust_statedump_destroy();
2298 lttng_ust_tp_exit();
2299 if (!exiting) {
2300 /* Reinitialize values for fork */
2301 sem_count = sem_count_initial_value;
2302 lttng_ust_comm_should_quit = 0;
2303 initialized = 0;
2304 }
2305 }
2306
2307 void __attribute__((destructor)) lttng_ust_exit(void)
2308 {
2309 int ret;
2310
2311 /*
2312 * Using pthread_cancel here because:
2313 * A) we don't want to hang application teardown.
2314 * B) the thread is not allocating any resource.
2315 */
2316
2317 /*
2318 * Require the communication thread to quit. Synchronize with
2319 * mutexes to ensure it is not in a mutex critical section when
2320 * pthread_cancel is later called.
2321 */
2322 ust_lock_nocheck();
2323 lttng_ust_comm_should_quit = 1;
2324 ust_unlock();
2325
2326 pthread_mutex_lock(&ust_exit_mutex);
2327 /* cancel threads */
2328 if (global_apps.thread_active) {
2329 ret = pthread_cancel(global_apps.ust_listener);
2330 if (ret) {
2331 ERR("Error cancelling global ust listener thread: %s",
2332 strerror(ret));
2333 } else {
2334 global_apps.thread_active = 0;
2335 }
2336 }
2337 if (local_apps.thread_active) {
2338 ret = pthread_cancel(local_apps.ust_listener);
2339 if (ret) {
2340 ERR("Error cancelling local ust listener thread: %s",
2341 strerror(ret));
2342 } else {
2343 local_apps.thread_active = 0;
2344 }
2345 }
2346 pthread_mutex_unlock(&ust_exit_mutex);
2347
2348 /*
2349 * Do NOT join threads: use of sys_futex makes it impossible to
2350 * join the threads without using async-cancel, but async-cancel
2351 * is delivered by a signal, which could hit the target thread
2352 * anywhere in its code path, including while the ust_lock() is
2353 * held, causing a deadlock for the other thread. Let the OS
2354 * cleanup the threads if there are stalled in a syscall.
2355 */
2356 lttng_ust_cleanup(1);
2357 }
2358
2359 static
2360 void ust_context_ns_reset(void)
2361 {
2362 lttng_context_pid_ns_reset();
2363 lttng_context_cgroup_ns_reset();
2364 lttng_context_ipc_ns_reset();
2365 lttng_context_mnt_ns_reset();
2366 lttng_context_net_ns_reset();
2367 lttng_context_user_ns_reset();
2368 lttng_context_time_ns_reset();
2369 lttng_context_uts_ns_reset();
2370 }
2371
2372 static
2373 void ust_context_vuids_reset(void)
2374 {
2375 lttng_context_vuid_reset();
2376 lttng_context_veuid_reset();
2377 lttng_context_vsuid_reset();
2378 }
2379
2380 static
2381 void ust_context_vgids_reset(void)
2382 {
2383 lttng_context_vgid_reset();
2384 lttng_context_vegid_reset();
2385 lttng_context_vsgid_reset();
2386 }
2387
2388 /*
2389 * We exclude the worker threads across fork and clone (except
2390 * CLONE_VM), because these system calls only keep the forking thread
2391 * running in the child. Therefore, we don't want to call fork or clone
2392 * in the middle of an tracepoint or ust tracing state modification.
2393 * Holding this mutex protects these structures across fork and clone.
2394 */
2395 void lttng_ust_before_fork(sigset_t *save_sigset)
2396 {
2397 /*
2398 * Disable signals. This is to avoid that the child intervenes
2399 * before it is properly setup for tracing. It is safer to
2400 * disable all signals, because then we know we are not breaking
2401 * anything by restoring the original mask.
2402 */
2403 sigset_t all_sigs;
2404 int ret;
2405
2406 /* Fixup lttng-ust TLS. */
2407 lttng_ust_fixup_tls();
2408
2409 if (URCU_TLS(lttng_ust_nest_count))
2410 return;
2411 /* Disable signals */
2412 sigfillset(&all_sigs);
2413 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2414 if (ret == -1) {
2415 PERROR("sigprocmask");
2416 }
2417
2418 pthread_mutex_lock(&ust_fork_mutex);
2419
2420 ust_lock_nocheck();
2421 lttng_ust_urcu_before_fork();
2422 lttng_ust_lock_fd_tracker();
2423 lttng_perf_lock();
2424 }
2425
2426 static void ust_after_fork_common(sigset_t *restore_sigset)
2427 {
2428 int ret;
2429
2430 DBG("process %d", getpid());
2431 lttng_perf_unlock();
2432 lttng_ust_unlock_fd_tracker();
2433 ust_unlock();
2434
2435 pthread_mutex_unlock(&ust_fork_mutex);
2436
2437 /* Restore signals */
2438 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2439 if (ret == -1) {
2440 PERROR("sigprocmask");
2441 }
2442 }
2443
2444 void lttng_ust_after_fork_parent(sigset_t *restore_sigset)
2445 {
2446 if (URCU_TLS(lttng_ust_nest_count))
2447 return;
2448 DBG("process %d", getpid());
2449 lttng_ust_urcu_after_fork_parent();
2450 /* Release mutexes and reenable signals */
2451 ust_after_fork_common(restore_sigset);
2452 }
2453
2454 /*
2455 * After fork, in the child, we need to cleanup all the leftover state,
2456 * except the worker thread which already magically disappeared thanks
2457 * to the weird Linux fork semantics. After tyding up, we call
2458 * lttng_ust_init() again to start over as a new PID.
2459 *
2460 * This is meant for forks() that have tracing in the child between the
2461 * fork and following exec call (if there is any).
2462 */
2463 void lttng_ust_after_fork_child(sigset_t *restore_sigset)
2464 {
2465 if (URCU_TLS(lttng_ust_nest_count))
2466 return;
2467 lttng_context_vpid_reset();
2468 lttng_context_vtid_reset();
2469 lttng_ust_context_procname_reset();
2470 ust_context_ns_reset();
2471 ust_context_vuids_reset();
2472 ust_context_vgids_reset();
2473 DBG("process %d", getpid());
2474 /* Release urcu mutexes */
2475 lttng_ust_urcu_after_fork_child();
2476 lttng_ust_cleanup(0);
2477 /* Release mutexes and reenable signals */
2478 ust_after_fork_common(restore_sigset);
2479 lttng_ust_init();
2480 }
2481
2482 void lttng_ust_after_setns(void)
2483 {
2484 ust_context_ns_reset();
2485 ust_context_vuids_reset();
2486 ust_context_vgids_reset();
2487 }
2488
2489 void lttng_ust_after_unshare(void)
2490 {
2491 ust_context_ns_reset();
2492 ust_context_vuids_reset();
2493 ust_context_vgids_reset();
2494 }
2495
2496 void lttng_ust_after_setuid(void)
2497 {
2498 ust_context_vuids_reset();
2499 }
2500
2501 void lttng_ust_after_seteuid(void)
2502 {
2503 ust_context_vuids_reset();
2504 }
2505
2506 void lttng_ust_after_setreuid(void)
2507 {
2508 ust_context_vuids_reset();
2509 }
2510
2511 void lttng_ust_after_setresuid(void)
2512 {
2513 ust_context_vuids_reset();
2514 }
2515
2516 void lttng_ust_after_setgid(void)
2517 {
2518 ust_context_vgids_reset();
2519 }
2520
2521 void lttng_ust_after_setegid(void)
2522 {
2523 ust_context_vgids_reset();
2524 }
2525
2526 void lttng_ust_after_setregid(void)
2527 {
2528 ust_context_vgids_reset();
2529 }
2530
2531 void lttng_ust_after_setresgid(void)
2532 {
2533 ust_context_vgids_reset();
2534 }
2535
2536 void lttng_ust_sockinfo_session_enabled(void *owner)
2537 {
2538 struct sock_info *sock_info = owner;
2539 sock_info->statedump_pending = 1;
2540 }
This page took 0.087427 seconds and 5 git commands to generate.